[Contemporary Mathematics 690] Andres Eduardo Caicedo, James Cummings, Peter Koellner, Paul B. Larson - Foundations of Mathematics_ Logic at Harvard Essays in Honor of W. Hugh Woodin’s 60th Birthday March 27-29, 2015 Harvard Unive

690
Foundations of Mathematics
Logic at Harvard
Essays in Honor of W. Hugh Woodin’s 60th Birthday
March 27–29, 2015
Harvard University, Cambridge, MA
Andrés Eduardo Caicedo

James Cummings
Peter Koellner
Paul B. Larson
Editors
American Mathematical Society

Logic at Harvard
March 27–29, 2015

James Cummings
Peter Koellner
Paul B. Larson
Editors
Photo courtesy of D. Schrittesser
690
Logic at Harvard
March 27–29, 2015

James Cummings
Peter Koellner
Paul B. Larson
Editors
American Mathematical Society

Providence, Rhode Island
EDITORIAL COMMITTEE
Dennis DeTurck, Managing Editor
Michael Loss Kailash Misra Catherine Yan
2010 Mathematics Subject Classification. Primary 03E55; Secondary 03E60, 03E57,

03E45, 03E35, 03E15, 00A30, 03D03.
Library of Congress Cataloging-in-Publication Data

Names: Woodin, W. H. (W. Hugh) — Caicedo, Andrés Eduardo, 1974–
Title: Foundations of mathematics : logic at Harvard : essays in honor of Hugh Woodin’s 60th
birthday, March 27–29, 2015, Harvard University, Cambridge, MA / Andrés Eduardo Caicedo
[and three others], editors.
Description: Providence, Rhode Island : American Mathematical Society, [2017] | Series: Con-
temporary mathematics ; volume 690 | Includes bibliographical references.
Identifiers: LCCN 2016043599 | ISBN 9781470422561 (alk. paper)
Subjects: LCSH: Set theory | Continuum hypothesis | AMS: Mathematical logic and founda-
tions – Set theory – Large cardinals. msc | Mathematical logic and foundations – Set theory –
Determinacy principles. msc | Mathematical logic and foundations – Set theory – Generic ab-
soluteness and forcing axioms. msc | Mathematical logic and foundations – Set theory – Inner
models, including constructibility, ordinal definability, and core models. msc | Mathematical logic
and foundations – Set theory – Consistency and independence results. msc | Mathematical logic
and foundations – Set theory – Descriptive set theory. msc | General – General and miscella-
neous specific topics – Philosophy of mathematics. msc | Mathematical logic and foundations –
Computability and recursion theory – Other degrees and reducibilities. msc
Classification: LCC QA248 .F6945 2017 | DDC 511.3/22–dc23 LC record available at
https://lccn.loc.gov/2016043599
Contemporary Mathematics ISSN: 0271-4132 (print); ISSN: 1098-3627 (online)
DOI: http://dx.doi.org/10.1090/conm/690
Color graphic policy. Any graphics created in color will be rendered in grayscale for the
printed version unless color printing is authorized by the Publisher. In general, color graphics will
appear in color in the online version.
Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting
for them, are permitted to make fair use of the material, such as to copy select pages for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Permissions to reuse
portions of AMS publication content are handled by Copyright Clearance Center’s RightsLink
service. For more information, please visit: http://www.ams.org/rightslink.
Send requests for translation rights and licensed reprints to reprint-permission@ams.org.
Excluded from these provisions is material for which the author holds copyright. In such cases,
requests for permission to reuse or reprint material should be addressed directly to the author(s).
Copyright ownership is indicated on the copyright page, or on the lower right-hand corner of the
first page of each article within proceedings volumes.

c 2017 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 22 21 20 19 18 17
To Hugh
Contents
Preface ix
Bibliography of W. Hugh Woodin xiii
List of Participants xix
Norming infinitesimals of large fields
H. G. Dales 1
The enumeration degrees: Local and global structural interactions
Theodore A. Slaman and Mariya I. Soskova 31
Ramsey properties of finite measure algebras and topological dynamics
of the group of measure preserving automorphisms: Some results
and an open problem
A. S. Kechris, M. Sokić, and S. Todorcevic 69
Topological Ramsey numbers and countable ordinals
Andrés Eduardo Caicedo and Jacob Hilton 87
Open determinacy for class games
Victoria Gitman and Joel David Hamkins 121
Open problems on ultrafilters and some connections to the continuum
M. Malliaris and S. Shelah 145
Obtaining Woodin’s cardinals
P. D. Welch 161
Woodin’s axiom (∗), or Martin’s maximum, or both?
Ralf Schindler 177
Translation procedures in descriptive inner model theory
Grigor Sargsyan 205
Implications of very large cardinals
Scott Cramer 225
What makes the continuum ℵ2
Justin Tatch Moore 259
Set-theoretic foundations
Penelope Maddy 289
vii
Preface
This volume presents a collection of papers related to the work of W. Hugh

Woodin, who has been one of the leading figures in set theory since the early 1980s.
Woodin’s earliest work, when he was still an undergraduate at Caltech, was on
automatic continuity in Banach spaces. It resulted in his earliest (unpublished) pa-
per, which won the Eric Temple Bell Undergraduate Mathematics Research Prize
in 1976 (and is mentioned in Garth Dales’s contribution to this volume). Later,
but still early in his career, he studied possible behaviors for the continuum func-
tion, showing with Matthew Foreman that consistently (relative to the existence of
suitable large cardinals) the Generalized Continuum Hypothesis can fail at every
infinite cardinal; their work in this area and early insights by Woodin on Radin
forcing led to the development of Radin- and Prikry-like forcings, and also pro-
vided some of the key insights that resulted in Moti Gitik’s theorems on the precise
consistency strength of the Singular Cardinal Hypothesis. With Ted Slaman he has
studied degree structures in recursion theory, with a focus on what are now called
the biinterpretability conjectures.
His central work, however, has been in the theory of large cardinals and de-
terminacy. Among his many results in these areas, the most striking might be
his identification of the existence of infinitely many Woodin cardinals as the exact
consistency strength of the Axiom of Deteminacy. The proof of this theorem in-
volves two bodies of work which Woodin developed at essentially the same time.
For the upper bound, Woodin developed the general notion of stationary tower em-
beddings, building on work of Foreman, Menachem Magidor and Saharon Shelah.
For the lower bound, Woodin developed an analysis of the inner model HOD in de-
terminacy models, building on work of Robert Solovay, Howard Becker and others.
Among Woodin’s other important contributions to set theory are AD+ , arguably
the right version of the axiom of determinacy and central to modern developments
in the area; the theory of Pmax forcing, a method for forcing over models of de-
terminacy to produce models of the Axiom of Choice which are maximal for the
structure P(ω1 ); the extender algebra, which can be used to show that for cer-
tain models of set theory, every set is generic over an elementary extension; the
theory of universally Baire sets, first defined in joint work with Qi Feng and Magi-
dor and which has proved to be essential in the study of generic absoluteness; and
the core-model induction, a method for deriving large-cardinal consistency strength
from combinatorial principles. Each of these contributions has given rise to entire
research areas.
In recent years Woodin has been advancing a philosophical program that makes
the case for new axioms that settle statements at the level of the Continuum Hy-
pothesis and far beyond. With his work on the search for an ultimate inner model
ix
x PREFACE
for large cardinals (the so-called Ultimate-L model) as well as with the associated
V = Ultimate-L Conjecture, Woodin is in fact making a case for new axioms that
would arguably resolve (with the help of large cardinal axioms) all statements left
undecided by the standard Zermelo-Fraenkel axioms with the Axiom of Choice
(ZFC). This program also includes a careful analysis of extender models for very
large cardinals.
A meeting to celebrate Hugh’s 60th birthday was held during March 27–29,
2015, in the Fong Auditorium at Harvard University. The meeting featured nine
one-hour lectures on topics reflecting the various stages of Hugh’s career. The
speakers included some of his closest collaborators (Dales, Foreman, Alekos Kechris,
Magidor, Tony Martin, Slaman) and others (Ronald Jensen, Grigor Sargsyan, John
Steel) who have worked in inner model theory and determinacy, in areas close to the
central concerns in Hugh’s research throughout his career. Feng, who has written
two important papers with Hugh, was scheduled to speak but could not attend.
The meeting was attended by Hugh’s advisor Solovay, several of his students (Joan
Bagaria, Scott Cramer, Vincenzo Dimonte, Joel David Hamkins, George Kafkoulis,
Paul Larson, Xianghui Shi, Liuzhen Wu) and a large number of people working in
areas related to Woodin’s work.
After this preface we include Hugh’s bibliography, followed by a list of the
participants of the conference, with their academic affiliations at the time. We
apologize for any omissions.
Below is a short description of the papers in this volume; they include some by
the speakers (Dales, Kechris, Sargsyan, Slaman), some by Hugh’s former students
(Hamkins, Cramer, Andrés Caicedo), and others by participants of the meeting
(Maryanthe Malliaris, Philip Welch, Ralf Schindler). The volume’s final two pa-
pers are by Justin Moore, whose work in combinatorial set theory has often com-
plemented Woodin’s work, and Penelope Maddy, who has written extensively on
philosophical issues connected with large cardinals and determinacy.
The paper by Dales is a survey on norming the infinitesimals of large fields and
on discontinuous homomorphisms from Banach algebras of continuous functions,
and includes some charming recollections of his collaboration with Hugh. The paper
by Slaman and Soskova is an investigation of the structure of the automorphism
group of the enumeration degrees and related structures. The paper by Kechris,
Miodrag Sokić and Stevo Todorcevic studies the class of ordered finite measure
algebras in the context of structural Ramsey theory and topological dynamics. The
paper by Caicedo and Jacob Hilton studies the topological partition calculus in
the realm of countable ordinals, in particular obtaining a topological version of
the Erdős-Milner theorem. The paper by Victoria Gitman and Hamkins studies
the determinacy of open games on proper classes, in particular showing that in
consistency strength this version of determinacy transcends ZFC. The paper by
Malliaris and Shelah is a survey of problems on the theory of ultrafilters in set
theory, model theory and topology, related to their program to study Keisler’s
order. Welch’s paper considers a reflection principle that implies the existence
of a proper class of measurable Woodin cardinals. Schindler’s paper considers
a principle that extends simultaneously Martin’s maximum and Woodin’s axiom
(∗). Sargsyan’s paper presents a procedure to translate hod mice into regular
mice that does not require global determinacy assumptions. Cramer’s paper is a
survey of large cardinals at the level of I0 , the statement that there is an elementary
PARTICIPANTS IN THE GROUP PICTURE xi
embedding from some L(Vλ+1 ) into itself with critical point below λ. Moore’s paper
is a survey of results on the size of the continuum from reflection principles and
other natural combinatorial statements. Maddy’s paper considers three different
approaches to the foundations of mathematics, focusing on set theory and large
cardinals, on the recent multiverse approach, and on category-theoretic alternatives.
The meeting was supported by Harvard University and the National Science
Foundation (partially through the Mid-Atlantic Mathematical Logic Seminar). We
want to thank all the speakers and participants at the conference, the authors and
referees who helped make this volume a reality, the editors of the Contemporary
Mathematics series, and especially Christine Thivierge, for all her help and support
throughout the process.
This volume is a small token of our appreciation and admiration for Hugh.
The editors
Andrés Eduardo Caicedo
James Cummings
Peter Koellner
Paul B. Larson
Participants in the group picture

A group picture of most of the participants of the meeting appears as a fron-
tispiece to this volume. We thank David Schrittesser for allowing us to include
it.
There are 53 people in the picture. They are listed below, roughly from left to
right.
Douglas Blue, Scott Cramer, Liuzhen Wu, Nam Trang, Daisuke
Ikegami, Xianghui Shi, Vincenzo Dimonte, Joseph Van Name,
Tony Martin, Alexander Kechris, Joan Bagaria, Laura
Fontanella, Paul McKenney, Kaethe Minden, Kameryn
Williams, Paul Larson, Sheila Miller, Ronald Jensen, Steve
Homer, Juliette Kennedy, David Schrittesser, W. Hugh
Woodin, Gunter Fuchs, Arthur Apter, Menachem Magidor,
Charles Parsons, Jouko Väänänen, Ralf Schindler, Rehana
Patel, Nate Ackerman, John Steel, George Kafkoulis, Ilijas
Farah, Martin Zeman, Assaf Peretz, Grigor Sargsyan, Akihiro
Kanamori, Trevor Wilson, Maryanthe Malliaris, Hossein Lamei
Ramandi, Philip D. Welch, H. Garth Dales, Derrick DuBose,
Gabriel Goldberg, Joel David Hamkins, Ted Slaman, Jacob
Davis, Doug Hoffman, Joshua Reagan, Matthew Foreman,
Zeynep Soysal, Daniel Rodrı́guez, and Peter Koellner.
Thanks to Derrick DuBose, Miha Habič, Joel David Hamkins, Asaf Karagila,
Benedikt Löwe, Iian Smythe, and Yizheng Zhu for their help with the list.
Bibliography of W. Hugh Woodin
[BW97] Joan Bagaria and W. Hugh Woodin, Δ1 sets of reals, J. Symbolic Logic 62 (1997),
∼n
no. 4, 1379–1428, DOI 10.2307/2275649. MR1618004
[CFS08a] Chitat Chong, Qi Feng, Theodore A. Slaman, W. Hugh Woodin, and Yue Yang
(eds.), Computational prospects of infinity. Part I. Tutorials, Lecture Notes Series.
Institute for Mathematical Sciences. National University of Singapore, vol. 14, World
Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2008. Lectures from the workshop
held at the National University of Singapore, Singapore, June 20–August 15, 2005.
MR2449476
[CFS08b] Chitat Chong, Qi Feng, Theodore A. Slaman, W. Hugh Woodin, and Yue Yang
(eds.), Computational prospects of infinity. Part II. Presented talks, Lecture Notes
Series. Institute for Mathematical Sciences. National University of Singapore, vol. 15,
World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2008, Lectures from the
workshop held at the National University of Singapore, Singapore, June 20–August
15, 2005. MR2449456
[CFS14] Chitat Chong, Qi Feng, Theodore A. Slaman, W. Hugh Woodin, and Yue Yang (eds.),
E-recursion, forcing and C ∗ -algebras, Lecture Notes Series. Institute for Mathemat-
ical Sciences. National University of Singapore, vol. 27, World Scientific Publishing
Co. Pte. Ltd., Hackensack, NJ, 2014. Lecture notes from the 2012 Asian Initiative
for Infinity (AII) Logic Summer School held at the National University of Singapore,
Singapore. MR3287593
[CFS16] Chitat Chong, Qi Feng, Theodore A. Slaman, W. Hugh Woodin, and Yue Yang (eds.),
Forcing, iterated ultrapowers, and Turing degrees, Lecture Notes Series. Institute for
Mathematical Sciences. National University of Singapore, vol. 29, World Scientific
Publishing Co. Pte. Ltd., Hackensack, NJ, 2016. Lecture notes from the 2010 and
2011 Asian Initiative for Infinity (AII) Graduate Logic Summer Schools held at the
Institute for Mathematical Sciences (IMS), National University of Singapore, Singa-
pore. MR3409336
[CFSW14] Chitat Chong, Qi Feng, Theodore A. Slaman, and W. Hugh Woodin (eds.), Infinity
and truth, Lecture Notes Series. Institute for Mathematical Sciences. National Uni-
versity of Singapore, vol. 25, World Scientific Publishing Co. Pte. Ltd., Hackensack,
NJ, 2014. Including lectures from the workshop held at the National University of
Singapore, Singapore, July 25–29, 2011. MR3185586
[DW87] H. G. Dales and W. H. Woodin, An introduction to independence for analysts, London
Mathematical Society Lecture Note Series, vol. 115, Cambridge University Press,
Cambridge, 1987. MR942216
[DW96] H. Garth Dales and W. Hugh Woodin, Super-real fields, London Mathematical Society
Monographs. New Series, vol. 14, The Clarendon Press, Oxford University Press,
New York, 1996. Totally ordered fields with additional structure; Oxford Science
Publications. MR1420859
[FMW92] Qi Feng, Menachem Magidor, and Hugh Woodin, Universally Baire sets of reals,
Set theory of the continuum (Berkeley, CA, 1989), Math. Sci. Res. Inst. Publ.,
vol. 26, Springer, New York, 1992, pp. 203–242, DOI 10.1007/978-1-4613-9754-0 15.
MR1233821
[FW91] Matthew Foreman and W. Hugh Woodin, The generalized continuum hypothesis can
fail everywhere, Ann. of Math. (2) 133 (1991), no. 1, 1–35, DOI 10.2307/2944324.
MR1087344
xiii
xiv BIBLIOGRAPHY OF W. HUGH WOODIN
[FW96] Sy D. Friedman and W. Hugh Woodin, δ 12 without sharps, Proc. Amer. Math. Soc.
∼
124 (1996), no. 7, 2211–2213, DOI 10.1090/S0002-9939-96-03297-2. MR1322923
[FW03] Q. Feng and W. H. Woodin, P -points in Qmax models, Ann. Pure Appl. Logic 119
(2003), no. 1-3, 121–190, DOI 10.1016/S0168-0072(02)00036-2. MR1937848
[FWW08] Sy-David Friedman, Philip Welch, and W. Hugh Woodin, On the consistency strength
of the inner model hypothesis, J. Symbolic Logic 73 (2008), no. 2, 391–400, DOI
10.2178/jsl/1208359050. MR2414455
[GMW85] M. Gitik, M. Magidor, and H. Woodin, Two weak consequences of 0 , J. Symbolic
Logic 50 (1985), no. 3, 597–603, DOI 10.2307/2274315. MR805670
[Hir15] Denis R. Hirschfeldt, Slicing the truth, Lecture Notes Series. Institute for Mathemat-
ical Sciences. National University of Singapore, vol. 28, World Scientific Publishing
Co. Pte. Ltd., Hackensack, NJ, 2015. On the computable and reverse mathematics
of combinatorial principles; Edited and with a foreword by Chitat Chong, Qi Feng,
Theodore A. Slaman, W. Hugh Woodin and Yue Yang. MR3244278
[HMW85] J. M. Henle, A. R. D. Mathias, and W. Hugh Woodin, A barren extension, Methods
in mathematical logic (Caracas, 1983), Lecture Notes in Math., vol. 1130, Springer,
Berlin, 1985, pp. 195–207, DOI 10.1007/BFb0075312. MR799042
[HRW08] Joel David Hamkins, Jonas Reitz, and W. Hugh Woodin, The ground axiom is con-
sistent with V = HOD, Proc. Amer. Math. Soc. 136 (2008), no. 8, 2943–2949, DOI
10.1090/S0002-9939-08-09285-X. MR2399062
[HW99] Kai Hauser and W. Hugh Woodin, Π13 sets and Π13 singletons, J. Symbolic Logic 64
(1999), no. 2, 590–616, DOI 10.2307/2586486. MR1777772
[HW00] Joel David Hamkins and W. Hugh Woodin, Small forcing creates neither strong
nor Woodin cardinals, Proc. Amer. Math. Soc. 128 (2000), no. 10, 3025–3029, DOI
10.1090/S0002-9939-00-05347-8. MR1664390
[HW05] Joel D. Hamkins and W. Hugh Woodin, The necessary maximality principle for
c.c.c. forcing is equiconsistent with a weakly compact cardinal, MLQ Math. Log. Q.
51 (2005), no. 5, 493–498, DOI 10.1002/malq.200410045. MR2163760
[HW11] Michael Heller and W. Hugh Woodin (eds.), Infinity, Cambridge University Press,
Cambridge, 2011. New research frontiers. MR2850464
[JJW92] H. Judah, W. Just, and H. Woodin (eds.), Set theory of the continuum, Mathemat-
ical Sciences Research Institute Publications, vol. 26, Springer-Verlag, New York,
1992. Papers from the workshop held in Berkeley, California, October 16–20, 1989.
MR1233806
[JKSW14] S. Jackson, R. Ketchersid, F. Schlutzenberg, and W. H. Woodin, Determinacy
and Jónsson cardinals in L(R), J. Symb. Log. 79 (2014), no. 4, 1184–1198, DOI
10.1017/jsl.2014.49. MR3343535
[JSW90] Haim Judah, Saharon Shelah, and W. H. Woodin, The Borel conjecture, Ann.
Pure Appl. Logic 50 (1990), no. 3, 255–269, DOI 10.1016/0168-0072(90)90058-A.
MR1086456
[JW85] Thomas J. Jech and W. Hugh Woodin, Saturation of the closed unbounded filter on
the set of regular cardinals, Trans. Amer. Math. Soc. 292 (1985), no. 1, 345–356,
DOI 10.2307/2000184. MR805967
[KKMW81] Alexander S. Kechris, Eugene M. Kleinberg, Yiannis N. Moschovakis, and W. Hugh
Woodin, The axiom of determinacy, strong partition properties and nonsingular mea-
sures, Cabal Seminar 77–79 (Proc. Caltech-UCLA Logic Sem., 1977), Lecture Notes
in Math., vol. 839, Springer, Berlin-New York, 1981, pp. 75–99. MR611168
[KLW87] A. S. Kechris, A. Louveau, and W. H. Woodin, The structure of σ-ideals of compact
sets, Trans. Amer. Math. Soc. 301 (1987), no. 1, 263–288, DOI 10.2307/2000338.
MR879573
[KW83] Alexander S. Kechris and W. Hugh Woodin, Equivalence of partition properties and
determinacy, Proc. Nat. Acad. Sci. U.S.A. 80 (1983), no. 6 i., 1783–1786. MR699440
[KW86] Alexander S. Kechris and W. Hugh Woodin, Ranks of differentiable functions,
Mathematika 33 (1986), no. 2, 252–278 (1987), DOI 10.1112/S0025579300011244.
MR882498
BIBLIOGRAPHY OF W. HUGH WOODIN xv
[KW91] A. S. Kechris and W. H. Woodin, A strong boundedness theorem for dilators, Ann.
Pure Appl. Logic 52 (1991), no. 1-2, 93–97, DOI 10.1016/0168-0072(91)90041-J. In-
ternational Symposium on Mathematical Logic and its Applications (Nagoya, 1988).
MR1104056
[KW08a] Alexander S. Kechris and W. Hugh Woodin, The equivalence of partition properties
and determinacy, Games, scales, and Suslin cardinals. The Cabal Seminar. Vol. I,
Lect. Notes Log., vol. 31, Assoc. Symbol. Logic, Chicago, IL, 2008, pp. 355–378, DOI
10.1017/CBO9780511546488.018. MR2463618
[KW08b] Alexander S. Kechris and W. Hugh Woodin, Generic codes for uncountable ordinals,
partition properties, and elementary embeddings, Games, scales, and Suslin cardinals.
The Cabal Seminar. Vol. I, Lect. Notes Log., vol. 31, Assoc. Symbol. Logic, Chicago,
IL, 2008, pp. 379–397, DOI 10.1017/CBO9780511546488.019. MR2463619
[KW09] Peter Koellner and W. Hugh Woodin, Incompatible Ω-complete theories, J. Symbolic
Logic 74 (2009), no. 4, 1155–1170, DOI 10.2178/jsl/1254748685. MR2583814
[KW10] Peter Koellner and W. Hugh Woodin, Large cardinals from determinacy, Handbook of
set theory. Vols. 1, 2, 3, Springer, Dordrecht, 2010, pp. 1951–2119, DOI 10.1007/978-
1-4020-5764-9 24. MR2768702
[Lar04] Paul B. Larson, The stationary tower, University Lecture Series, vol. 32, American
Mathematical Society, Providence, RI, 2004. Notes on a course by W. Hugh Woodin.
MR2069032
[MR205] Set theory, Oberwolfach Rep. 2 (2005), no. 4, 3121–3174, DOI 10.4171/OWR/2005/5.
Abstracts from the meeting held December 4–10, 2005, Organized by Sy-David Fried-
man, Menachem Magidor and Hugh Woodin, Oberwolfach Reports. Vol. 2, no. 4.
MR2238849
Abstracts from the workshop held January 13–19, 2008; Organized by Sy-David Fried-
man, Menachem Magidor and W. Hugh Woodin; Oberwolfach Reports. Vol. 5, no. 1.
MR2492484
Abstracts from the workshop held January 9–15, 2011; Organized by Sy-David Fried-
man, Menachem Magidor and W. Hugh Woodin; Oberwolfach Reports. Vol. 8, no. 1.
MR2849500
Abstracts from the workshop held January 12–18, 2014; Organized by Ilijas Farah,
Sy-David Friedman, Menachem Magidor and W. Hugh Woodin. MR3309126
[MW08] Donald A. Martin and W. Hugh Woodin, Weakly homogeneous trees, Games, scales,
and Suslin cardinals. The Cabal Seminar. Vol. I, Lect. Notes Log., vol. 31, Assoc. Sym-
bol. Logic, Chicago, IL, 2008, pp. 421–438, DOI 10.1017/CBO9780511546488.022.
MR2463621
[SW84] Saharon Shelah and Hugh Woodin, Forcing the failure of CH by adding a real, J.
Symbolic Logic 49 (1984), no. 4, 1185–1189, DOI 10.2307/2274270. MR771786
[SW86] Theodore A. Slaman and W. Hugh Woodin, Definability in the Turing degrees, Illinois
J. Math. 30 (1986), no. 2, 320–334. MR840131
[SW89] Theodore A. Slaman and W. Hugh Woodin, Σ1 -collection and the finite injury pri-
ority method, Mathematical logic and applications (Kyoto, 1987), Lecture Notes in
Math., vol. 1388, Springer, Berlin, 1989, pp. 178–188, DOI 10.1007/BFb0083670.
MR1015729
[SW90] Saharon Shelah and Hugh Woodin, Large cardinals imply that every reasonably de-
finable set of reals is Lebesgue measurable, Israel J. Math. 70 (1990), no. 3, 381–394,
DOI 10.1007/BF02801471. MR1074499
[SW97] Theodore A. Slaman and W. Hugh Woodin, Definability in the enumeration degrees,
Arch. Math. Logic 36 (1997), no. 4-5, 255–267, DOI 10.1007/s001530050064. Sacks
Symposium (Cambridge, MA, 1993). MR1473024
[SW98] Theodore A. Slaman and W. Hugh Woodin, Extending partial orders to dense
linear orders, Ann. Pure Appl. Logic 94 (1998), no. 1-3, 253–261, DOI
10.1016/S0168-0072(97)00075-4. Conference on Computability Theory (Oberwolfach,
1996). MR1640271
xvi BIBLIOGRAPHY OF W. HUGH WOODIN
[SW01] E. Schimmerling and W. H. Woodin, The Jensen covering property, J. Symbolic Logic
66 (2001), no. 4, 1505–1523, DOI 10.2307/2694959. MR1877007
[Tod14] Stevo Todorcevic, Notes on forcing axioms, Lecture Notes Series. Institute for Math-
ematical Sciences. National University of Singapore, vol. 26, World Scientific Pub-
lishing Co. Pte. Ltd., Hackensack, NJ, 2014. Edited and with a foreword by Chitat
Chong, Qi Feng, Yue Yang, Theodore A. Slaman and W. Hugh Woodin. MR3184691
[VW98] Boban Velickovic and W. Hugh Woodin, Complexity of reals in inner models of
set theory, Ann. Pure Appl. Logic 92 (1998), no. 3, 283–295, DOI 10.1016/S0168-
0072(98)00010-4. MR1640916
[Woo76] W. Hugh Woodin, Discontinuous homomorphisms from C(Ω) and the partially or-
dered set ω ω , Unpublished. Winner of the E. T. Bell undergraduate mathematics
research prize, Caltech, 1976.
[Woo82] W. Hugh Woodin, On the consistency strength of projective uniformization, Pro-
ceedings of the Herbrand symposium (Marseilles, 1981), Stud. Logic Found. Math.,
vol. 107, North-Holland, Amsterdam, 1982, pp. 365–384, DOI 10.1016/S0049-
237X(08)71895-0. MR757040
[Woo83a] W. Hugh Woodin, AD and the uniqueness of the supercompact measures on P ω1 (λ),
Cabal seminar 79–81, Lecture Notes in Math., vol. 1019, Springer, Berlin, 1983,
pp. 67–71, DOI 10.1007/BFb0071694. MR730587
[Woo83b] W. Hugh Woodin, Some consistency results in ZFC using AD, Cabal seminar 79–
81, Lecture Notes in Math., vol. 1019, Springer, Berlin, 1983, pp. 172–198, DOI
10.1007/BFb0071701. MR730594
[Woo84] W. Hugh Woodin, DISCONTINUOUS HOMOMORPHISMS OF C(OMEGA) AND
SET THEORY, ProQuest LLC, Ann Arbor, MI, 1984. Thesis (Ph.D.)–University of
California, Berkeley. MR2634119
[Woo86] Hugh Woodin, Aspects of determinacy, Logic, methodology and philosophy of science,
VII (Salzburg, 1983), Stud. Logic Found. Math., vol. 114, North-Holland, Amsterdam,
1986, pp. 171–181, DOI 10.1016/S0049-237X(09)70692-5. MR874787
[Woo88] W. Hugh Woodin, Supercompact cardinals, sets of reals, and weakly homoge-
neous trees, Proc. Nat. Acad. Sci. U.S.A. 85 (1988), no. 18, 6587–6591, DOI
10.1073/pnas.85.18.6587. MR959110
[Woo93] W. Hugh Woodin, A discontinuous homomorphism from C(X) without CH, J.
London Math. Soc. (2) 48 (1993), no. 2, 299–315, DOI 10.1112/jlms/s2-48.2.299.
MR1231717
[Woo94] W. Hugh Woodin, Large cardinal axioms and independence: the continuum prob-
lem revisited, Math. Intelligencer 16 (1994), no. 3, 31–35, DOI 10.1007/BF03024354.
MR1281752
[Woo96] W. Hugh Woodin, The universe constructed from a sequence of ordinals, Arch. Math.
Logic 35 (1996), no. 5-6, 371–383, DOI 10.1007/s001530050051. MR1420264
[Woo98] W. Hugh Woodin, The Tower of Hanoi, Truth in mathematics (Mussomeli, 1995),
Oxford Univ. Press, New York, 1998, pp. 329–351. MR1688350
[Woo99] W. Hugh Woodin, The axiom of determinacy, forcing axioms, and the nonstationary
ideal, de Gruyter Series in Logic and its Applications, vol. 1, Walter de Gruyter &
Co., Berlin, 1999. MR1713438
[Woo01a] W. Hugh Woodin, The continuum hypothesis. I, Notices Amer. Math. Soc. 48 (2001),
no. 6, 567–576. MR1834351
[Woo01b] W. Hugh Woodin, The continuum hypothesis. II, Notices Amer. Math. Soc. 48 (2001),
no. 7, 681–690. MR1842471
[Woo01c] W. Hugh Woodin, The Ω conjecture, Aspects of complexity (Kaikoura, 2000), de
Gruyter Ser. Log. Appl., vol. 4, de Gruyter, Berlin, 2001, pp. 155–169. MR1884265
[Woo02a] W. Hugh Woodin, Beyond Σ2 absoluteness, Proceedings of the International Congress
∼1
of Mathematicians, Vol. I (Beijing, 2002), Higher Ed. Press, Beijing, 2002, pp. 515–
524. MR1989202
[Woo02b] W. Hugh Woodin, Correction: “The continuum hypothesis. II”, Notices Amer. Math.
Soc. 49 (2002), no. 1, 46. MR1869995
[Woo04] W. Hugh Woodin, Set theory after Russell: the journey back to Eden, One hundred
years of Russell’s paradox, de Gruyter Ser. Log. Appl., vol. 6, de Gruyter, Berlin,
2004, pp. 29–47. MR2104736
BIBLIOGRAPHY OF W. HUGH WOODIN xvii
[Woo05] W. Hugh Woodin, The continuum hypothesis, Logic Colloquium 2000, Lect. Notes
Log., vol. 19, Assoc. Symbol. Logic, Urbana, IL, 2005, pp. 143–197. MR2143878
[Woo06] W. Hugh Woodin, The cardinals below |[ω1 ]<ω1 |, Ann. Pure Appl. Logic 140 (2006),
no. 1-3, 161–232, DOI 10.1016/j.apal.2005.09.014. MR2224057
[Woo08a] W. Hugh Woodin, A tt version of the Posner-Robinson theorem, Computational
prospects of infinity. Part II. Presented talks, Lect. Notes Ser. Inst. Math. Sci.
Natl. Univ. Singap., vol. 15, World Sci. Publ., Hackensack, NJ, 2008, pp. 355–392.
MR2449474
[Woo08b] W. Hugh Woodin, Tutorial outline: suitable extender sequences, Computational
prospects of infinity. Part I. Tutorials, Lect. Notes Ser. Inst. Math. Sci. Natl.
Univ. Singap., vol. 14, World Sci. Publ., Hackensack, NJ, 2008, pp. 195–253, DOI
10.1142/9789812794055 0004. MR2449480
[Woo10a] W. Hugh Woodin, The axiom of determinacy, forcing axioms, and the nonstationary
ideal, Second revised edition, de Gruyter Series in Logic and its Applications, vol. 1,
Walter de Gruyter GmbH & Co. KG, Berlin, 2010. MR2723878
[Woo10b] W. Hugh Woodin, Strong axioms of infinity and the search for V , Proceedings of the
International Congress of Mathematicians. Volume I, Hindustan Book Agency, New
Delhi, 2010, pp. 504–528. MR2827903
[Woo10c] W. Hugh Woodin, Suitable extender models I, J. Math. Log. 10 (2010), no. 1-2,
101–339, DOI 10.1142/S021906131000095X. MR2802084
[Woo11a] W. Hugh Woodin, The continuum hypothesis, the generic-multiverse of sets, and
the Ω conjecture, Set theory, arithmetic, and foundations of mathematics: theorems,
philosophies, Lect. Notes Log., vol. 36, Assoc. Symbol. Logic, La Jolla, CA, 2011,
pp. 13–42. MR2882650
[Woo11b] W. Hugh Woodin, A potential subtlety concerning the distinction between determin-
ism and nondeterminism, Infinity, Cambridge Univ. Press, Cambridge, 2011, pp. 119–
129. MR2767236
[Woo11c] W. Hugh Woodin, The realm of the infinite, Infinity, Cambridge Univ. Press, Cam-
bridge, 2011, pp. 89–118, DOI 10.1017/CBO9780511976889.006. MR2767235
[Woo11d] W. Hugh Woodin, Suitable extender models II: beyond ω-huge, J. Math. Log. 11
(2011), no. 2, 115–436, DOI 10.1142/S021906131100102X. MR2914848
[Woo11e] W. Hugh Woodin, The transfinite universe, Kurt Gödel and the foundations of math-
ematics, Cambridge Univ. Press, Cambridge, 2011, pp. 449–471. MR2866873
[Woo14] W. Hugh Woodin, The weak ultimate L conjecture, Infinity, computability, and meta-
mathematics, Tributes, vol. 23, Coll. Publ., London, 2014, pp. 309–329. MR3307892
[Woo16] W. Hugh Woodin, The weak ultrafilter axiom, Arch. Math. Logic 55 (2016), no. 1-2,
319–351, DOI 10.1007/s00153-015-0467-2. MR3453588
[Woo99] W. Hugh Woodin, The continuum problem. I [MR1834351] (Hungarian, with English
and Hungarian summaries), Mat. Lapok (N.S.) 8/9 (1998/99), no. 3-4, 54–73 (2003).
MR2023677
[WS05] W. Hugh Woodin and Theodore A. Slaman, Definability in degree structures,
Preprint, 2005.
[WS06] W. Hugh Woodin and Theodore A. Slaman, Mathematical logic (The Berkeley un-
dergraduate course), 2006.
List of Participants
Nate Ackerman Joel David Hamkins

Harvard University The City University of New York
Arthur Apter Daniel Hathaway
The City University of New York University of Michigan
Joan Bagaria Douglas J. Hoffman

Universitat de Barcelona Northwestern Connecticut Community
College
Douglas Blue
Harvard University Steve Homer
Boston University
Scott Cramer
Rutgers University Daisuke Ikegami
Kobe University
H. Garth Dales
Lancaster University Ronald B. Jensen
Humboldt-Universität zu Berlin
Jacob Davis
George Kafkoulis
Carnegie Mellon University
Florida International University
Vincenzo Dimonte
Akihiro Kanamori
Technische Universität Wien
Boston University
Derrick DuBose
Alexander S. Kechris
University of Nevada, Las Vegas
California Institute of Technology
Ilijas Farah Juliette Kennedy
York University University of Helsinki
Laura Fontanella Peter Koellner
The Hebrew University of Jerusalem Harvard University
Matthew D. Foreman Paul B. Larson
University of California, Irvine Miami University
Gunter Fuchs Menachem Magidor
The City University of New York The Hebrew University of Jerusalem
Gabriel Goldberg Maryanthe Malliaris
Harvard University University of Chicago
Miha Habič Donald A. Martin
The City University of New York University of California, Los Angeles
xix
xx PARTICIPANTS
Paul McKenney Nam Trang

Miami University University of California, Irvine
Sheila Miller Jouko Väänänen
The City University of New York University of Helsinki
Kaethe Minden Joseph Van Name

The City University of New York The City University of New York
Charles Parsons Philip D. Welch

Harvard University University of Bristol
Rehana Patel Kameryn Williams

Olin College The City University of New York
Trevor Wilson
Assaf Peretz
University of California, Irvine
THINQon
W. Hugh Woodin
Hossein Lamei Ramandi Harvard University
Cornell University
Liuzhen Wu
Joshua Reagan Chinese Academy of Sciences
University of Nevada, Las Vegas
Martin Zeman
Daniel Rodrı́guez University of California, Irvine
Carnegie Mellon University
Gerald Sacks
Harvard University
Grigor Sargsyan
Rutgers University
Ralf Schindler
Universität Münster
David Schrittesser
University of Copenhagen
Xianghui Shi
Beijing Normal University
Richard Shore
Cornell University
Theodore A. Slaman
University of California, Berkeley
Robert M. Solovay
Zeynep Soysal
Harvard University
John R. Steel
Contemporary Mathematics
Volume 690, 2017
http://dx.doi.org/10.1090/conm/690/13861
Norming infinitesimals of large fields
H. G. Dales
Abstract. We give a survey of results on norming the infinitesimals of large
fields and on constructing discontinuous homomorphisms from Banach alge-
bras of continuous functions; we raise questions that remain from earlier work.
The article is expanded from a talk given in Harvard on 27 March 2015 at the
conference in honour of the 60th birthday of W. Hugh Woodin. It includes
some historical remarks about the first written mathematics of Hugh.
Contents
1. Introduction
2. Ordered groups and fields
3. Structure of ordered sets, groups, and fields
4. Related questions with CH
5. The independence of NDH
6. Norming larger fields and super-real fields
References
1. Introduction
It is a pleasure and an honour to be invited to lecture on the theory of discontinuous
homomorphisms from Banach algebras of continuous functions, and the closely
related question of norming the algebra of infinitesimals of various large fields.
The question whether all algebra norms on the algebra C(Ω) are equivalent to
the uniform norm on Ω, for each compact space Ω, was first raised by Kaplansky
long ago [21], and led to much work on the structure of commutative Banach
algebras, especially by Bade and Curtis [3]. It was shown independently by myself
[5] and by Jean Esterle [14] in two papers that were both submitted for publication
in May 1976 that, in the theory ZFC + CH, there are algebra norms that are not
equivalent to the uniform norm on C(Ω) for each infinite, compact space Ω. (The
results were summarized in [7].) We both regarded it, with little thought, as obvious
that the use of CH in the proof was a blemish, and that a little more ingenuity would
remove the need to appeal to this axiom.
2010 Mathematics Subject Classification. Primary 46H40, Secondary 03E35, 03E50.

Key words and phrases. Kaplansky’s problem, continuous functions, discontinuous homo-
morphisms, Banach algebras, norming infinitesimals, super-real fields.
2017
c American Mathematical Society
1
2 H. G. DALES
We were stunned to discover that it was a result of W. H. Woodin that this was
not the case; there are models of ZFC in which all algebra norms on each algebra
C(Ω) are equivalent to the uniform norm. We were even more amazed to find that,
at the time of this proof, Hugh was an undergraduate at Caltech; there are further
comments on this history later.
It is perhaps the case that Kaplansky’s question was the first one that appeared
to be totally within functional analysis, and yet was shown, to general surprise, to
be independent of ZFC.
I first met Hugh in the year 1983–84 when he was one of my TAs for a calculus
class that I was teaching at Berkeley.1 Subsequently we worked together on two
books [9, 10], and this meant that Hugh came to our house in England quite a few
times. This was great pleasure for us. It was something of a shock to my wife and
myself to receive an invitation to come to Harvard for a conference to mark Hugh’s
60th birthday: how could this charming young man have reached so mature a level?
And now Hugh is a Professor at Harvard: our best congratulations!
Work after 1983 clarified and extended several of the original results, and proofs
were presented from a different aspect, which we shall describe here; the main
exposition of these results is given in [6, §5.7]. Nevertheless, there remain a number
of apparently challenging, related open questions; some of these will be mentioned
below.
1.1. Basic definitions. We first recall some basic definitions.

The cardinality of a set S is written as |S|; the continuum has cardinality c,
and so c = 2ℵ0 ; CH is the continuum hypothesis, i.e., the statement that c = ℵ1 ;
GCH is the generalized continuum hypothesis, which we shall use just to see that
2ℵ0 = ℵ1 and 2ℵ1 = ℵ2 .
Let A be an associative algebra over a field F, always the real field R or the
complex field C. The algebra formed by adjoining an identity to a non-unital
algebra A is denoted by A , with A = A when A already has an identity; the
identity of A is denoted by eA . The Jacobson radical of A is denoted by rad A;2
the algebra A is semi-simple if rad A = {0} and radical if rad A = A.
An element a in an algebra A is nilpotent if an = 0 for some n ∈ N; the set
of these elements is denoted by nil A. An element a in a complex algebra A is
quasi-nilpotent if zeA − a is invertible in the unital algebra A for each non-zero
complex number z; the set of these elements is denoted by Q(A). It is clear that
nil A ⊂ Q(A) and that rad A ⊂ Q(A); further, A is radical if and only if Q(A) = A.
An ideal P in an algebra A is a prime ideal if P is a proper ideal and if either
a ∈ P or b ∈ P whenever {axb : x ∈ A} is a subset of P . In the case where A
is commutative, a proper ideal P is prime if and only if either a ∈ P or b ∈ P
whenever ab ∈ P , and then A is an integral domain if {0} is a prime ideal; in this
latter case, A has a quotient field , or field of fractions, that consists of the fractions
a/b, where a, b ∈ A with b = 0, taken with the obvious algebraic operations.
1 As a TA, Hugh was initially quite puzzled that there were students at Berkeley who did not
understand concepts that were very obvious to him, but he quickly learned about reality.
2 The general definition of rad A for non-commutative algebras A is not important for us.
However we recall that: a left ideal I in an algebra A is modular if there exists u ∈ A such that
a − au ∈ I (a ∈ A) ; rad A is defined to be the intersection of all the maximal modular left ideals
in A, with A taken to be radical when there are no maximal modular left ideals; rad A is always
a two-sided ideal in A. See [6, §1.3].
NORMING INFINITESIMALS OF LARGE FIELDS 3
A character on a complex algebra A is a homomorphism from A onto C. The

set of all characters on A is denoted by ΦA ; this is the character space of A.
Definition 1.1. An algebra semi-norm on an algebra A is a map · : A → R+
such that:
a + b ≤ a + b , αa = |α| a , ab ≤ a b
for a, b ∈ A and α ∈ F; the semi-norm is an algebra norm on A if, further, a = 0
for a = 0. An algebra A is semi-normable, respectively, normable, if there is a
non-zero algebra semi-norm, respectively, an algebra norm, on A.
A pair (A, · ), where A is a complex algebra, is a Banach algebra if · is an
algebra norm on A and (A, · ) is a Banach space.
Let A be a Banach algebra. We recall that rad A is always a closed, two-sided
ideal in A and that A/rad A is a semi-simple Banach algebra. All characters ϕ on
A are continuous linear functionals, with ϕ ≤ 1 [2, Theorem 4.43], and so ΦA can
be regarded as a subset of the closed unit ball (A )[1] of the dual space A of A; of
course, A is the Banach space consisting of all continuous linear functionals on the
Banach space A, and, by the classical Banach–Alaoglu theorem, (A )[1] is compact
with respect to the weak∗ -topology on A . Indeed, ΦA is a locally compact space
with respect to this topology.
By the ‘spectral radius formula’ (given in [2, Theorem 4.23] and [6, Theorem
2.3.8(iii)]), an element a in a Banach algebra A is quasi-nilpotent if and only if
1/n
limn→∞ an = 0. In the case where A is a commutative Banach algebra, an
element a ∈ A is quasi-nilpotent if and only if ϕ(a) = 0 for each character ϕ, and
so

1/n
rad A = Q(A) = a ∈ A : lim an = 0 = {a ∈ A : ϕ(a) = 0 (ϕ ∈ ΦA )} .
n→∞
Further, the maximal modular ideals in A are exactly the kernels of the characters
on A, and so all have codimension 1 in A. Thus, in this case, the theory simplifies
considerably.
A foundation stone of the theory of automatic continuity is the following the-
orem of Šilov from 1947 [6, Theorem 2.3.3].
Theorem 1.2. Let A and B be Banach algebras, and let θ : A → B be a
homomorphism. Suppose that B is commutative and semi-simple. Then θ is auto-
matically continuous.
There is an introductory account of Banach spaces and Banach algebras that
(more than) covers the above theory in [2], and there is a comprehensive account
in the monograph [6]. In particular, [6] gives many more automatic continuity
theorems, some of which will be mentioned below.
Throughout this article, X will often denote a non-empty, completely regular
topological space, and then C(X) is the (real) algebra of all continuous, real-valued
functions on X; the corresponding complex algebra is C(X, C).3 Of course, the
algebraic operations in these algebras are defined pointwise. The subalgebra of
C(X, C) consisting of the bounded functions is denoted by C b (X, C), with ∞ (C)
3 A Hausdorff topological space X is completely regular if, for each x ∈ X and each open
neighbourhood U of x, there exists f ∈ C(X) with f (x) = 1 and f (y) = 0 (y ∈ X \ U ). Every

locally compact space is completely regular.
4 H. G. DALES
for C b (N, C), the standard Banach space of all bounded sequences on N. The
algebra C b (X, C) is a commutative Banach algebra with respect to the uniform
norm | · |X on X, where
|f |X = sup{|f (x)| : x ∈ X} (f ∈ C b (X, C)) ,
and C b (X, C) is semi-simple because it is clear that there are no non-zero quasi-
nilpotent elements in C b (X, C). It is standard that the algebra C b (X, C) can be
identified by an isometric algebra isomorphism with C(βX, C), where βX is the
Stone–Čech compactification of the space X;4 in particular, we identify ∞ (C)
with the space C(βN, C). We shall write X ∗ for the growth of the space X in βX,
so that X ∗ = βX \ X; in particular, we have the notation N∗ = βN \ N.
In the case where X is locally compact (always Hausdorff), we take C0 (X, C)
to be the closed ideal in C b (X, C) consisting of the functions that vanish at infinity,
and set c 0 (C) = C0 (N, C), so that c 0 (C) is the space of all null-sequences; C(Ω, C)
is a complex, commutative, unital Banach algebra for the uniform norm whenever
Ω is a non-empty, compact space.
The basic examples to have in mind are the following: C(I, C), where I = [0, 1]
is the closed unit interval in C; ∞ (C); and c 0 (C). Perhaps surprisingly, the most
complicated of these is C(I, C).
Let Ω be a non-empty, compact space, and take x ∈ Ω. Then
Mx = {f ∈ C(Ω, C) : f (x) = 0} ,
and Jx is the set of functions in C(Ω, C) that vanish on a neighbourhood of x, the
neighbourhood depending on the function. Clearly Jx = Mx whenever x is isolated
in Ω, and Jx is always a dense ideal in Mx for each x ∈ Ω. It is standard and easy
to see (see [2, Example 4.49] or [6, Theorem 4.2.1(i)]) that {Mx : x ∈ Ω} is the
family of all maximal ideals in C(Ω, C), equivalently, every character on C(Ω, C)
has the form
εx : f → f (x) , C(Ω, C) → C ,
for some x ∈ Ω.
Definition 1.3. A radical homomorphism from an algebra A is a non-zero

homomorphism from A into a commutative, radical Banach algebra.
Let Ω be a compact space, and take x ∈ Ω.

Suppose that θ : Mx → B is a non-zero homomorphism into a commutative
Banach algebra B such that θ | Jx = 0. For each ψ ∈ ΦB , we have (ψ ◦ θ) | Jx = 0,
and so (ψ ◦ θ) | Mx = 0 because ψ ◦ θ is continuous and Jx = Mx . Thus
θ(Mx ) ⊂ {ker ψ : ψ ∈ ΦB } = rad B, and so θ is a radical homomorphism.
Conversely, suppose that θ : Mx → B is a homomorphism into a commutative,
radical Banach algebra B, and take f ∈ Jx . Then there exists g ∈ Mx with f g = f ,
and so θ(f )(eB − θ(g)) = 0. Since B is a radical algebra, eB − θ(g) is invertible in
B , and so θ(f ) = 0. Thus θ | Jx = 0.
It follows that a non-zero homomorphism θ from an algebra Mx into a commu-
tative Banach algebra is a radical homomorphism if and only if θ | Jx = 0. Thus
there is a non-zero homomorphism θ from Mx into a commutative, radical Banach
algebra B if and only if there is an ideal I = ker θ in C(Ω, C) with Jx ⊂ I Mx
4 This is a special case of the Gel’fand representation theorem, given in [6, Theorem 2.3.25],
for example. It is also immediate from the definition of the Stone–Čech compactification, βX.
such that the induced homomorphism from Mx /I into B is an injection, and hence
such that Mx /I is normable, i.e., if and only if Mx /Jx is semi-normable.
1.2. Kaplansky’s problem. Let Ω be a non-empty, locally compact space,

and take · to be an algebra norm on C0 (Ω, C). Then Kaplansky proved in [21]
that
|f |Ω ≤ f (f ∈ C0 (Ω, C)) ;
see [6, Theorem 4.2.3(ii)]. He then asked whether the two norms are always equiv-
alent, in the sense that there is a constant C > 0 such that
f ≤ C |f |Ω (f ∈ C0 (Ω, C)) .
Of course, it is an immediate consequence of Banach’s isomorphism theorem [2,
Corollary 3.41] that this is true whenever the algebra norm · is complete. Also,
by Theorem 1.2, it is true whenever the completion of (C0 (Ω, C), · ) is semi-
simple.
Suppose that · is a norm on C0 (Ω, C) that is not equivalent to the uniform
norm, and take B to be the commutative Banach algebra that is the completion of
(C0 (Ω, C), · ). Then the embedding of (C0 (Ω, C), | · |Ω ) into B is discontinuous.
Conversely, suppose that θ : C0 (Ω, C) → B is a discontinuous homomorphism into
a Banach algebra B. Then the formula
f = max{|f |Ω , θ(f )} (f ∈ C0 (Ω, C))
defines an algebra norm · on C0 (Ω, C) that is not equivalent to the uniform norm.
Thus the following theorem [6, Theorem 4.2.3(iii)] is a re-wording of Kaplansky’s
result.
Theorem 1.4. Let Ω be a non-empty, locally compact space. Then the following
are equivalent:
(a) there is an algebra norm on C0 (Ω, C) that is not equivalent to the uniform
norm;
(b) there is a discontinuous homomorphism from C0 (Ω, C) into some Banach
algebra.
In fact, Kaplansky’s problem has usually been considered in the setting of clause
(b), above.
To ease future discussion, it is convenient to state an axiom called NDH (‘no
discontinuous homomorphisms’); this was first specifically formulated by Solovay.
NDH: For each compact space Ω, each homomorphism from C(Ω, C) into a
Banach algebra is continuous.
Thus Kaplansky’s problem asks whether NDH is true in ZFC. We note that
there are many (non-commutative) Banach algebras A such that all homomor-
phisms from A into every Banach algebra are automatically continuous; see §4.3
and [6].
Several claims of a positive solution of Kaplansky’s problem were made, and
they were investigated by Bade and Curtis around 1960 in a seminar at Yale; all
the alleged proofs turned out to be erroneous. However, the request of Charles
Rickart to these younger colleagues to investigate the claims led to their interest in
Kaplansky’s question, and hence to the seminal paper [3], so the false claims did
play an important role in the story.
6 H. G. DALES
The main theorem of the paper of Bade and Curtis describes in detail the
structure of a homomorphism from a general algebra of the form C0 (Ω, C) (and,
in fact, from more general ‘strong Ditkin algebras’) into a commutative Banach
algebra. We give an abbreviated and informal form of this theorem here; a full
version is in [6, Theorem 5.4.22] and [9, Theorem 1.6].
Theorem 1.5. Let Ω be a non-empty, compact space, and suppose that θ is a

homomorphism from C(Ω, C) into a Banach algebra A. Then either θ is continuous
or there is a non-empty, finite subset {x1 , . . . , xn } of Ω such that θ | B is continuous
for a certain dense, unital subalgebra B of C(Ω, C) such that B contains the ideal
Jx1 ∩ · · · ∩ Jxn . Further, in the latter case, there is a radical homomorphism from
Mx for some x ∈ Ω.
Corollary 1.6. Let Ω be a non-empty, compact space. Then there is an

algebra norm on C(Ω, C) that is not equivalent to the uniform norm if and only
if there exists x ∈ Ω such that Jx Mx and the quotient space Mx /Jx is semi-
normable.
Thus the study of Kaplansky’s problem for C(Ω, C) reduces to the question of
the existence of x ∈ Ω and an ideal I in C(Ω, C) such that Jx ⊂ I Mx and Mx /I
is normable.
We slightly reformulate and extend the above in the following convenient set-
ting: now Ω is a non-empty, locally compact, non-compact space, and θ is an
assumed discontinuous homomorphism from C0 (Ω) into a Banach algebra B. The
continuity ideal , denoted by I(θ), of θ is defined to be the largest ideal I of C0 (Ω)
such that θ | I is continuous.5 It follows from the theorem of Bade and Curtis
and extensions thereof by Esterle [13] and by Sinclair [26] that the ideals ker θ and
I(θ) are always intersections of prime ideals and that I(θ) = ker θ whenever B is
a radical Banach algebra. Thus we obtain a further reformulation of Kaplansky’s
problem, as follows; this is the form in which it will be discussed.
Theorem 1.7. Let Ω be a non-empty, compact space. Then the following are
equivalent:
(a) there is an algebra norm on C(Ω, C) that is not equivalent to the uniform
norm;
(b) there is a maximal ideal M and a prime ideal P in C(Ω, C) with P M
such that the algebra M/P is normable.
(c) there is a maximal ideal M and a prime ideal P in C(Ω, C) with P M
and an embedding of M/P into a commutative, radical Banach algebra.
We note that, for each infinite, compact space Ω, the algebra C(Ω, C) contains
a maximal ideal Mx and a prime ideal P with Jx ⊂ P Mx such that |Mx /P | = c
for some x ∈ Ω.
The main theorem proved in [5] and [13] (in different ways and by long calcu-
lations) was the following.
5 There are more general definitions in [6].

Theorem 1.8. (CH) Let Ω be an infinite compact space. Then there is a

maximal ideal M , a prime ideal P in C(Ω, C) with P M , and a homomorphism
with kernel P from M into a certain radical Banach algebra, and hence the algebra
M/P is normable.
Thus CH implies ¬NDH.

The above suggests a formulation of our question in the language of ordered
fields, and we shall consider such a version in §3.2.
2. Ordered groups and fields

We first recall some background in the theory of ordered groups and fields. All the
results of this section are taken from [10].
2.1. Ordered sets. Let S be a non-empty set. A strict partial order on S is

a binary relation < on S such that:
(i) if a < b and b < c in S, then a < c;
(ii) a < a for each a ∈ S.
The order is a total order if, for each a, b ∈ S, either a < b or a = b or b < a. A
partially ordered set is a pair (P, <), where S is a non-empty set and < is a strict
partial order on S. We set a ≤ b if a < b or a = b in S. A totally ordered set S is
well-ordered if each non-empty subset of S has a minimum element.
We shall require the following examples later.
Examples 2.1. (i) Let R be a commutative, radical algebra, and consider the
set S = R \ {0}. For a, b ∈ S, set a b if a ∈ bR. Then is a strict partial order
on S, called the divisibility order .
(ii) Let f, g ∈ NN , and then define f <F g if there exists n0 ∈ N such that
f (n) < g(n) (n ≥ n0 ) and f F g if g(n) − f (n) → ∞ as n → ∞. We see that <F
and F are strict partial orders on NN , called the Fréchet order and the strong
Fréchet order , respectively.
Let (S, <) and (T, <) be partially ordered sets. A map π : S → T is isotonic,
respectively, anti-isotonic, if π(a) < π(b), respectively, π(b) < π(a), whenever a < b
in S; the map π is an order-isomorphism if it is a bijection such that π and π −1
are isotonic.
Let (S, ≤) be a totally ordered set. A subset T of S is: order-dense if, for each
s1 , s2 ∈ S with s1 < s2 there exists t ∈ T with s1 ≤ t ≤ s2 ; cofinal in S if, for each
s ∈ S, there exists t ∈ T with s ≤ t; coinital in S if, for each s ∈ S, there exists
t ∈ T with t ≤ s. The weight, w(S), of S is the minimum cardinality of an order-
dense subset; the cofinality and coinitiality of S are the minimum cardinalities κ
and λ of a strictly increasing and strictly decreasing, respectively, sequence whose
range is cofinal and coinitial, respectively, in S; we then write
cof S = κ and coi S = λ .
Take S1 and S2 to be subsets of S. Then S1 S2 if s1 < s2 whenever s1 ∈ S1 and
s 2 ∈ S2 .
Of course (R, ≤) is a totally ordered set of cardinality |R| = c and weight
w(R) = ℵ0 ; the subset Q of rational numbers is an order-dense subset.
8 H. G. DALES
2.2. Ordered groups. Let G = (G, +) be an abelian group with identity 0.

Then G is divisible if, for each x ∈ G and each n ∈ N, there exists y ∈ G with
ny = x A map ψ : G → H between two groups G and H is a group morphism if
ψ(x + y) = ψ(x) + ψ(y) (x, y ∈ G) .
Now suppose that G is a group with a partial order ≤. Then (G, +, ≤) is an
ordered group if x + z < y + z whenever x < y in G and z ∈ G. We then write
G+ = {x ∈ G : x ≥ 0} for the positive elements of G. Suppose further that (G, ≤)
is a lattice; for example, RS is an ordered group that is a lattice with respect to
the usual operations for each non-empty set S. Then x+ = x ∨ 0 and x− = x ∧ 0
for x ∈ G, so that x = x+ + x− (x ∈ G), and we define |x| = x+ − x− (x ∈ G).
A subset S of G is absolutely convex if x ∈ S whenever |x| ≤ |y| for some y ∈ S.
Further, an ordered group (G, +, ≤) is a totally ordered group if ≤ is a total order
on G; in this case (G, ≤) is a lattice. The base number of such a group is
δ(G) = coi(G+ \ {0}) .
Let G be a totally ordered group. For x, y ∈ G, set x = O(y) if |x| ≤ n |y| for
some n ∈ N, and x ∼ y if x = O(y) and y = O(x). The set
ΓG = (G \ {0})/ ∼
is the value set of G, and the quotient map v is the archimedean valuation; set
v(x) ≤ v(y) if x = O(y). Then (ΓG , ≤) is a totally ordered set.
Two totally ordered groups G and H are isomorphic if there is a group mor-
phism from G onto H that is also an order-isomorphism.
2.3. Ordered fields. A field is taken to be a field over R, and all isomorphisms
between our fields must be R-linear; the identity of a field is usually denoted by 1.
Let K be a field, and suppose that (K, +, ≤) is a totally ordered group. Then
(K, +, · , ≤) is an ordered field if ab = a · b > 0 whenever a, b > 0 in K and if
αa > 0 whenever a > 0 in K and α > 0 in R.
A convex subgroup of K is a subgroup I of (K, +) such that 1 ∈ I and I is an
absolutely convex set.
An ordered field K is real-closed if every positive element is a square and every
polynomial over K of odd degree has a root; equivalently, K is real-closed if the
complexification K(i) of K is algebraically closed. The Artin–Schreier theorem
says that every ordered field has an algebraic extension to a real-closed field of the
same cardinality, and so, when seeking to show that all ordered fields of a given
cardinality are normable, it is sufficient to consider real-closed fields. Of course the
real line R is the proto-typical real-closed field.
Definition 2.2. Let K be an ordered field with identity 1, and let a ∈ K.
Then: a is an infinitesimal if |a| ≤ (1/n)1 for all n ∈ N; a is finite if |a| ≤ n1 for
some n ∈ N; and a is infinitely large if |a| ≥ n1 for all n ∈ N.
Thus 0 is an infinitesimal in each ordered field; in a large field there are non-zero
infinitesimals. Suppose that a > 0 and a is an infinitesimal. Then 1/a is infinitely
large.
We write K and K ◦ for the algebras of finite elements and of infinitesimals,
respectively, in an ordered field K. Thus K is formed by adjoining an identity to
the algebra K ◦ , and so is (K ◦ ) in the previous notation.
Let K be an ordered field with value set ΓK . Define + on ΓK by

v(a) + v(b) = v(ab) (a, b ∈ K \ {0}) .
Then the operation + is well-defined, and (ΓK , +, ≤) is a totally ordered group,
called the value group of K. Clearly
K ◦ = {a ∈ K : v(a) > 0} and K = {a ∈ K : v(a) ≥ 0} .
The value group ΓK is divisible whenever K is real-closed.
For example, let X be a non-empty, completely regular topological space, and
let M be a maximal ideal in the real algebra C(X). Then it is easily checked that
the quotient algebra C(X)/M is a real-closed ordered field; we shall prove a more
general result below. These are the hyper-real fields; see the seminal text of Gillman
and Jerison [18, §§5.6, 13.4] and [10, Definition 4.17].
In particular, suppose that X is discrete; it is just a set. Then we obtain real-
closed ordered fields that are isomorphic to fields of the form Rκ /M , where κ is
a cardinal and M is a maximal ideal in the real algebra Rκ . By the Gel’fand–
Kolmogorov theorem,6 maximal ideals in Rκ correspond to ultrafilters on κ, and
so these fields are called ultrapowers. Thus every ultrapower is a hyper-real field.
We recall that a filter F on a non-empty set S is a non-empty family of subsets of
S with the following properties: ∅ ∈ F; F ∩ G ∈ F whenever F, G ∈ F; G ∈ F
whenever G is a subset of S and G ⊃ F for some F ∈ F. An ultrafilter on S is a
maximal filter when the family of filters on S is ordered by inclusion; equivalently,
a filter U on S is an ultrafilter if F ∈ U or G ∈ U whenever F and G are subsets of
S and F ∪ G ∈ U. These ultrapowers Rκ /M are usually written as
Rκ /U or (Rκ /U, <U ) ,
where U is the ultrafilter on κ corresponding to the maximal ideal M . Here the
equivalence class containing f ∈ Rκ is
[f ]U = {g ∈ Rκ : {σ < κ : g(σ) = f (σ)} ∈ U} ;
we set f <U g in Rκ if {σ < κ : f (σ) < g(σ)} ∈ U, and [f ]U <U [g]U when f <U g.
Thus <U is a total order on Rκ /U.
The classic text on ultrafilters and ultrapowers is Comfort and Negrepontis [4].
There is a study of a generalization of ultrafilters to z-ultrafilters on a completely
regular space X in [18] and [6], and then the z-ultrafilters on X correspond to the
points of βX.
2.4. Hahn groups and fields. We now give a construction of some groups
F(S) and F(1) (S) and some fields F(G) and F(1) (G); for more general constructions,
see [10, Chapters 1 and 2].
Let S be a set, and take f ∈ RS . Then supp f = {s ∈ S : f (s) = 0}.
Let S be a totally ordered set. Then
F(S) = {f ∈ RS : supp f is well-ordered}
and
F(1) (S) = {f ∈ F(S) : supp f is countable} .
6 For more general versions of this theorem, see [6, Theorem 4.2.12] or [18, §7.3].
10 H. G. DALES
It is clear that both F(S) and F(1) (S) are real-linear subspaces of R S , and so
they are divisible groups with respect to the pointwise addition of functions. For
f ∈ F(S) with f = 0, set
v(f ) = inf supp f ,
and set f > 0 in F(S) if v(f ) > 0. Then v is the Hahn valuation on F(S), and
(F(S), +, ≤) is a totally ordered group, called the Hahn group of S. The value set of
(F(S), +, ≤) and of (F(1) (S), +, ≤) is S itself, and the Hahn valuation agrees with
the previous archimedean valuation.
The Hahn embedding theorem [10, Theorem 1.35] gives a certain ‘universal
property’ of the Hahn groups among totally ordered, divisible groups; we shall not
use this theorem.
Now suppose that G is a totally ordered group. Then we can define a product
on the linear space F(G): for f, g ∈ F(G), set

(f g)(t) = {f (r)g(s) : r, s ∈ G, r + s = t} (t ∈ G) .
In fact, the above sum is always a finite sum,7 and f g ∈ F(G); in the case where
f, g ∈ F(1) (G), then also f g ∈ F(1) (G). It is now easy to see that (F(G), ) is
an algebra over R and that F(1) (G) is a subalgebra of F(G). The algebra (F(G), )
is called the Hahn algebra over G; the valuation v recovers the value group G of
(F(G), ).
The main theorem in this setting is that (F(G), ) is in fact an ordered field,
and that F(1) (G) is a subfield of F(G); a theorem of Saunders MacLane shows that
F(G) and F(1) (G) are real-closed fields whenever G is a divisible group. For these
theorems, see [10, Theorem 2.15].
The Kaplansky embedding theorem [10, Theorem 2.17] gives a certain ‘universal
property’ of the Hahn fields (F(G), ) among real-closed ordered fields; a closely
related result is Kaplansky’s isomorphism theorem [6, Corollary 1.7.36], which is
expressed in the language of ‘maximal, complex valuation algebras’.
It was Jean Esterle who recognized the importance of the Hahn algebra in this
area, and also gave much shorter proofs of some very classical theorems; see [17],
for example.
2.5. Formal power series and Allan’s theorem. Here is the most elemen-
tary, non-trivial example of a Hahn field. For m ∈ Z, we denote by δm the function
on Z such that δm (m) = 1 and δm (n) = 0 for n ∈ Z with n = m.
Consider the totally ordered group (Z, +). Then F(Z) consist of the formal
Laurent series
∞

a= αn δn ,
n=n0
where n0 ∈ Z (with n0 possibly negative) and αn ∈ R (n ≥ n0 ), and multiplication
is defined by requiring that δm δn = δm+n (m, n ∈ Z).8 The Hahn valuation v is
given by
v(a) = min{n ∈ Z : αn = 0} ,
7 More precisely, the sum has only finitely many non-zero terms. This is because there is no
strictly decreasing, infinite sequence in the well-ordered set that is supp f .

8 Thus these formal series correspond exactly to the usual Laurent series of elementary com-
plex analysis.
and F(Z) = {a : v(a) ≥ 0} is clearly identified with

∞

F(Z+ ) = a = αn X n : αn ∈ R (n ∈ Z+ ) ,
n=0
the algebra of formal power series in one variable. Here Z+ = {n ∈ Z : n ≥ 0}.

This algebra is often denoted by R[[X]]; its complex version is C[[X]], and we shall
denote this latter algebra here just by F, writing F◦ for its unique maximal ideal.
The algebra F is a so-called Fréchet algebra with respect to the topology of
coordinatewise convergence τc , and τc is the unique topology with respect to which
F is a Fréchet algebra [6, Theorem 4.6.1]. It is easy to see that F is not a Banach
algebra with respect to any norm, but it is certainly not obvious whether or not F
is normable. But, in a significant break-through, this was shown to be the case by
Graham Allan in 1972 [1]. This theorem is given as [6, Theorem 5.7.1]; it uses the
algebraic extension methods of [6, §1.7] and an obvious ‘framework map’, namely
n → an , N → A, for a suitable element a in a Banach algebra A, and was an avatar
for later embedding theorems.
Definition 2.3. Let A be a commutative Banach algebra, and let a ∈ A. Then
a has finite closed descent if there exists n ∈ N such that
an ∈ an+1 A \ an+1 A ;
a has finite closed descent k if the minimum such n ∈ N is k.
The main theorem of Allan is the following.
Theorem 2.4. Let A be a commutative, unital Banach algebra, and let a ∈ A.

Then there is a unital embedding θ : F → A with θ(X) = a if and only if a ∈ rad A
and a has finite closed descent in A.
There are many commutative, radical Banach algebras that contain elements
of finite closed descent. For example, let V be the Banach space L1 (I), and set
s
(f g)(s) = f (s − t)g(t) dt (s ∈ I)
0
for f, g ∈ V. Then it is easy to check [6, Theorem 4.7.40] that V is a commutative,
radical Banach algebra; it is called the Volterra algebra. Allan noted in [1] that
there are elements of finite closed descent in V , and hence that there is a unital
embedding of F into V . It follows that F is normable.
2.6. Prime ideals in algebras C(X) and super-real fields. Let X be

a non-empty, completely regular topological space. There are substantial, classic
studies of the family of prime ideals in the algebras C(X). The starting point is
[18, Chapter 14]; this study is taken up in [10].
The following notion is surprisingly important in the study of ideals in the
algebras C(X); see [18], for example. The zero set of f ∈ C(X) is
Z(f ) = {x ∈ X : f (x) = 0} = f −1 ({0}) .
An ideal I in C(X) is a z-ideal if f ∈ I whenever Z(f ) = Z(g) for some g ∈ I.
Let P be a prime ideal in C(X). Then there are two basic facts. First, P is
contained in a unique maximal ideal of C(X), and, second, the set of primes in
C(X) containing P forms a totally ordered family with respect to inclusion. Each
12 H. G. DALES
prime ideal contains a minimal prime ideal. A prime z-ideal is a prime ideal that
is also a z-ideal. There are prime ideals that are not prime z-ideals, but every
minimal prime ideal is a prime z-ideal.
Let Ω be a non-empty, compact space, and take a prime ideal P in C(Ω, C).
Then there exists x ∈ Ω such that Jx ⊂ P ⊂ Mx .
Sometimes the ideal Jx in C(Ω, C) is itself a prime ideal; this is the case for each
x ∈ βN, and now Jx is the unique minimum prime ideal contained in Mx , and the
prime ideals in Mx form one totally ordered family. The point x ∈ βN corresponds
to an ultrafilter U on N, and the quotient algebra Mx /Jx is naturally identified with
the algebra (R ω /U)◦ of infinitesimals in the field that is the ultrapower R ω /U; we
shall consider free ultrafilters, corresponding to points of N∗ .9
However, sometimes Jx is not a prime ideal in C(Ω, C); for example, this is the
case for C(I, C) and each x ∈ I, and then the family of prime ideals contained in
Mx is a (very complicated) tree.
Definition 2.5. Let X be a non-empty, completely regular topological space,

and let P be a prime ideal in C(X). Then AP is the quotient algebra C(X)/P ,
and KP is its quotient field.
Certainly AP is an integral domain, and so AP does have a quotient field, KP .

It is a basic, if elementary, fact that AP is a totally ordered, commutative, unital
algebra [10, Theorem 4.14]. Indeed, take a ∈ AP , say a = πP (f ) for f ∈ C(X),
where πP denotes the quotient map. Then f = f + + f − , and clearly f + f − = 0, so
that either f + ∈ P or f − ∈ P . Hence either a ≤ 0 or a ≥ 0 in the quotient order
≤ on AP . Thus (AP , ≤) is a totally ordered set. Clearly a + c < b + c whenever
a, b, c ∈ AP with a < b, and ab > 0 whenever a, b ∈ AP with a, b > 0. Finally,
αa > 0 in AP whenever α > 0 in R and a > 0 in AP . It follows easily that KP is
an ordered field in the above sense, and that AP ⊂ KP .
Definition 2.6. Let K be an ordered field. Then K is a super-real field if K is

not isomorphic to R and K is isomorphic to an ordered field KP , where P is prime
ideal in C(X) for some completely regular space X.
Super-real fields are the topic of the monograph [10].

Clearly every hyper-real field is a super-real field. The key fact that super-
real fields are real-closed is proved in [10, Theorem 4.27], somewhat extending the
classical fact that hyper-real fields are real-closed [18, Theorem 13.4].10 In fact,
every super-real field is isomorphic to a field KP where P is a non-maximal prime
in C(Ω) for some compact space Ω.
There is a substantial classification of prime ideals in algebras of continuous
functions in [10]. We have noted that always AP ⊂ KP . In fact, AP = KP if and
only if AP is a valuation prime, in the sense that AP is a valuation algebra; see
[10, Proposition 4.32]. This is not always the case. Here is an easy preliminary
remark: there are valuation primes that are not z-ideals, and there are prime z-
ideals that are not valuation primes.
9 In fact, each ideal J in an algebra C(Ω, C) is a prime ideal if and only if Ω is an F -space,
x
which means that U ∩ V = ∅ whenever U and V are open Fσ -sets with U ∩ V = ∅.
10 These proofs use some elementary real analysis, and Rouché’s theorem from complex
analysis.
We now see that there is a further normability condition that would give a
solution to Kaplansky’s problem (in one direction).
Theorem 2.7. Let Ω be a compact space. Suppose that there is a non-maximal,

prime ideal P in C(Ω) such that the algebra KP of finite elements in the super-
real field KP is normable. Then there is an algebra norm on C(Ω, C) that is not
equivalent to the uniform norm.
I do not know whether KP is normable whenever AP is normable.
This leads us to the question, which we shall consider below:
Let K be an ordered field. When is the subalgebra K of finite elements of K
normable?
We remark that, by the Gel’fand–Mazur theorem in the form of [6, Theorem
2.2.42], the only real fields that are themselves normable are R and C.
2.7. Equivalence of compact spaces. It is natural to wonder if all infinite,

compact spaces Ω are equivalent for Kaplansky’s problem, in the sense that there is
a discontinuous homomorphism from each C(Ω, C), where Ω is an infinite, compact
space, whenever this is true for one particular compact space. The following theo-
rem, from [9, Theorems 1.8 and 1.13] reduces this question to the consideration of
the two spaces Ω = βN and Ω = N ∪ {∞} and the corresponding algebras ∞ (C)
and c0 (C) .
In the following, Jx is regarded as an ideal in C(βN, C) for x ∈ βN.
Theorem 2.8. (i) Assume that there is a discontinuous homomorphism from

C(Ω, C) into a Banach algebra for some compact space Ω. Then there exists x ∈ N∗
and a radical homomorphism θ from c0 (C) with ker θ ⊃ Jx ∩ c0 (C), and so there is
a discontinuous homomorphism from c0 (C) into a Banach algebra.
(ii) Assume that there is a discontinuous homomorphism from ∞ (C) into a
Banach algebra. Then there is a discontinuous homomorphism from C(Ω, C) into
a Banach algebra for each infinite, compact space Ω.
We shall reformulate the above theorem in the language of ultrafilters. Here U

and V are free ultrafilters on N.
The above theorem tells us the following: Assume that there exists U such that
(R ω /U)◦ is semi-normable. Then there exists V such that c0 /V is semi-normable.
This suggests the following questions:
(1) Assume that there exists U such that c0 /U is semi-normable. Does there
exist V such that (R ω /V)◦ is semi-normable?
(2) Assume that c0 /U is semi-normable. Is (R ω /U)◦ semi-normable?
(3) Assume that (R ω /U)◦ is semi-normable. Is c0 /U semi-normable?
(4) Assume that (R ω /U)◦ is semi-normable. Is (R ω /U)◦ normable?
A positive answer to (1) would show that all infinite, compact spaces Ω are
equivalent for Kaplansky’s problem. Questions (2) and (3) are variants of Question
(1) that seem to be of interest in their own right. A positive answer to Question (4)
would cleanse the theory. We seem not to have an answer to any of these questions.
We shall see that, in the theory ZFC + CH, the four algebras (R ω /U)◦ ,
(R /V)◦ , c0 /U, and c0 /V are all pairwise isomorphic (as real algebras) for any
ω
14 H. G. DALES
two free ultrafilters U and V; this was first proved by Barry Johnson in [20]. How-
ever, it follows from a theorem of Dow [11, Corollary 2.3] that, if CH fails, there are
two free ultrafilters U and V such that (R ω /U)◦ and (R ω /V)◦ are not isomorphic
even as rings.
3. Structure of ordered sets, groups, and fields

We shall now recall the notions of α1 –, β1 –, and η1 –structures on totally ordered
sets, totally ordered groups, and ordered fields. These notions go back to Hausdorff
around 1907.11 See [10, Chapter 1].
3.1. Definitions and examples. Let (S, ≤) be a totally ordered set. Then
S is an α1 –set if each non-empty subset of S has a countable cofinal and coinitial
subset, and an η1 –set if, for each countable subsets S1 and S2 of S (including the
case where either S1 or S2 is empty) with S1 S2 , there exists an element s ∈ S
with S1 {s} S2 . For example, (R, ≤) is an α1 –set, but not an η1 –set. Thus
we think of S as ‘small’ if it is an α1 –set and ‘large’ if it is an η1 –set; clearly no
totally ordered set is both an α1 –set and an η1 –set. But an η1 –set can be ‘nearly’
an α1 –set if it is a β1 –set: S is a β1 –set if it is the union of a chain of α1 –subsets
of S.
In fact there is ‘exactly one’ ordered set that is both a β1 –set and an η1 –set.
We denote by Q the set of all sequences (ατ : τ < ω1 ) of length ω1 , where
ατ ∈ {0, 1} (τ < ω1 ), for which the set {τ < ω1 : ατ = 1} is non-empty and has
a largest member; this set Q is taken with the the lexicographic order . The
following theorem is part of [10, Proposition 1.9 and Theorem 1.15].
Theorem 3.1. The pair (Q, ) is a β1 –η1 –set. Further, |Q| = w(Q) = c
and cof Q = ℵ1 , and (Q, ) is unique, in the sense that each β1 –η1 –set is order-
isomorphic to (Q, ).
The ordered set Q is called Sierpiński’s set. This set is universal in the class
of β1 –sets: every β1 –set is order-isomorphic to a subset of Q.
Now take (G, +, ≤) to be a totally ordered group. Then G is an α1 –group if
(G, ≤) is an α1 –set and an η1 –group if (G, ≤) is an η1 –set. Further, G is a β1 –group
if it is the union of a chain of α1 –subgroups.12 In fact there is ‘exactly one’ divisible
group that is both a β1 –group and an η1 –set.
We define G to be F(1) (Q), in the notation of §2.4. The following theorem is
part of [10, Propositions 1.43 and 1.44 and Theorem 1.48].
Theorem 3.2. The triple (G, +, ≤) is a divisible β1 –η1 –group. Further,

|G| = w(G) = c and cof G = δ(G) = ℵ1 ,
and (G, +, ≤) is unique, in the sense that each divisible β1 –η1 –group is isomorphic
to (G, +, ≤).
In particular, any two divisible η1 –groups of cardinality ℵ1 are isomorphic. The

group G is universal in the class of divisible β1 –groups: every divisible β1 –group is
isomorphic to a subgroup of G [10, Theorem 1.46].
11 The subscript ‘1’ is associated with the cardinal ℵ ; there are similar notions for larger
1
cardinals.
12 Note that it is apparently not sufficient that G be a union of a chain of α –subsets.
1
We finally step up further to consider fields.

Let K = (K, +, · , ≤) be an ordered field (still over R). Then K is an α1 –field
if (K, ≤) is an α1 –set and an η1 –field if (K, ≤) is an η1 –set. Further, K is a β1 –field
if it is the union of a chain of α1 –subfields. The subfields in the definition of a
β1 –field must be fields over R; maybe apparently weaker hypotheses are already
sufficient to imply this.
Example 3.3. Each hyper-real field is an η1 –field; this is a very classical fact
from Gillman and Jerison [18, §13.8].
Now suppose that Ω is a compact space and that P is a non-maximal prime
ideal in C(Ω). Then the super-real field KP is an η1 –field if and only if δ(KP ) ≥ ℵ1
[10, Theorem 4.28(ii)]. Thus super-real fields are ‘usually’ real-closed η1 –fields.
Let us consider the question whether the class, called super-real η1 in [10], of
super-real fields that are η1 -fields is actually the same as the class, hyper-real, of
hyper-real fields, in the sense that each member of super-real η1 is isomorphic as
a field to a member of hyper-real. This seems to be a substantial question that
we shall remark on below.
There is an example of a super-real field KP such that P is a prime z-ideal in
some C(Ω), but KP is not an η1 -field [10, Theorem 4.47]. This suggest consideration
of the class z-ideal η1 of super-real η1 -fields that are isomorphic to a field KP ,
where P is a prime z-ideal in some C(Ω). We do not know whether this class is the
same as the class hyper-real. There are intermediate possibilities; see Question
12, p. 342, of [10].
In fact there is ‘exactly one’ real-closed ordered field that is both a β1 –field and
an η1 –set. We now define this field, which is called R, to be
R = F(1) (G) = F(1) (F(1) (Q)) .
The following theorem is part of [10, Theorem 2.23 and Corollary 2.33].
Theorem 3.4. The field (R, +, · , ≤) is a real-closed β1 –η1 –field, with value
group G. Further,
|R| = w(R) = c and cof R = δ(R) = ℵ1 ,
and (R, +, · , ≤) is unique, in the sense that each real-closed β1 –η1 –field is isomor-
phic to (R, +, · , ≤) as an ordered field.
In particular, any two real-closed η1 –fields of cardinality ℵ1 are isomorphic.

The field R is universal in the class of β1 –fields: every β1 –field is isomorphic to a
subfield of R [10, Theorem 2.30].
3.2. Positive results on normability. The original proofs in [5] and [13]
evolved into the following theorem, taken from [6, Theorem 5.7.18]. The proof de-
pends heavily on the construction of a ‘framework map’, described in [6, §1.7], and is
still quite substantial, involving transfinite induction and algebraic manipulations.
The theorem holds in the theory ZFC.
Theorem 3.5. The algebra R of finite elements in the real-closed β1 –η1 –field
R is normable.
We immediately obtain the following consequence from the above remarks.

16 H. G. DALES
Theorem 3.6. Let K be a β1 –field. Then K is normable. In particular, let Ω

be a compact space, and suppose that P is a non-maximal, prime ideal in C(Ω) for
which KP is a β1 –field. Then there is an embedding of C(Ω)/P into a commutative
Banach algebra.
Recall that, for each infinite compact space Ω, the algebra C(Ω) contains a
maximal ideal Mx and a prime ideal P with Jx ⊂ P Mx and |Mx /P | = c. With
CH, |Mx /P | = ℵ1 , and so KP is a β1 –field. This recovers Theorem 1.8.
In fact, using remarks in §2.6, we see that the following slightly stronger theorem
holds [6, Theorem 5.7.13(ii)].
Theorem 3.7. Assume that there is a free ultrafilter U on N such that the
ultrapower R ω /U is a β1 –field. Then there is a discontinuous homomorphism from
C(Ω, C) into a Banach algebra for each infinite compact space Ω.
Clearly one would obtain a result without appealing to CH if one could find a
free ultrafilter U on N such that the ultrapower R ω /U is a β1 –field. However this is
not easy. First, whenever CH fails, there is always some free ultrafilter U on N such
that the ultrapower R ω /U is not a β1 –field. Further, we have the following result
from [10, Corollary 6.28]; here, MA is ‘Martin’s axiom’. Of course, CH implies
MA.
Theorem 3.8. (MA + ¬CH) For each free ultrafilter U on N, the ordered set
(R ω /U, ≤) is not a β1 –set.
Thus, to follow the above route, one must work in a theory where MA does not
hold. This has been carried out by Woodin in [31] in a special case.
Theorem 3.9. Assume that ZFC is consistent. Then there is a model of the
theory ZFC + ¬MA in which there is a free ultrafilter U on N such that the ultra-
power R ω /U is a β1 –field, and hence such that there is a discontinuous homomor-
phism from C(Ω, C) into a Banach algebra for each infinite compact space Ω.
This shows that ¬NDH is not equivalent to CH in ZFC.

Hugh obtained his model by adding ℵ2 Cohen reals to a model of ZFC + CH.
In particular, in this model c = ℵ2 . It seems to remain open whether one can obtain
such a model with c > ℵ2 . It also remains open whether, given a model of ZFC,
there is a model of ZFC + MA + ¬NDH; for a remark on this involving the proper
forcing axiom, see below.
3.3. Other algebras. We have shown that, with CH, the integral domains of
the form C(Ω, C)/P are normable when they have cardinality c. This result can be
generalized; the following theorem is [10, Theorem 5.25], extending [6, Theorem
5.7.10].
Theorem 3.10. Let A be a non-unital, complex integral domain with |A| = c.

Then there is a free ultrafilter U on N such that there is an embedding of A into
c0 (C)/U.
Corollary 3.11. (CH) Let A be a non-unital, complex integral domain with

|A| = c. Then A is normable.
The above result suggests that, with CH, there is a discontinuous homomor-
phism into a Banach algebra from each infinite-dimensional, commutative Banach
algebra. Towards this, we have the following theorem [6, Theorem 5.7.32].
Theorem 3.12. (CH) Let A be an infinite-dimensional, commutative Banach
algebra satisfying at least one of the following conditions:
(i) there is a non-maximal, prime ideal P in A with |A/P | = c;
(ii) there are infinitely many distinct characters on A;
(iii) for each a ∈ A, the set {an : n ∈ N} is linearly dependent;
(iv) the linear subspace spanned by {ab : a, b ∈ A} has infinite codimension in
the algebra A.
Then there is discontinuous homomorphism from A into some Banach algebra.
I have not been able to find an infinite-dimensional, commutative Banach al-
gebra that does not satisfy at least one of the above four conditions, and it is
conceivable that every infinite-dimensional, commutative Banach algebra satisfies
at least one of them, in which case our suggestion would hold. However, so far, the
question whether there is a discontinuous homomorphism into a Banach algebra
from each infinite-dimensional, commutative Banach algebra remains open.
4. Related questions with CH

There is a variety of related questions when we do assume CH, and we mention
some of these here.
4.1. The range of radical homomorphisms. In Theorem 3.5, we stated

that the algebra R is normable. This says that there is an embedding of R◦ into
a commutative, radical Banach algebra. One could ask: Into which commutative,
radical Banach algebras can we embed R◦ ?
We already know from Allan’s theorem, Theorem 2.4, that we can embed the
maximal ideal F◦ of F into a commutative, radical Banach algebra R if and only
if R contains an element of finite closed descent. It is remarkable that the latter
class of commutative, radical Banach algebras is exactly the class into which we
can embed the ‘much bigger’ algebra R◦ .
There are further characterizations of these algebras. We use the notation
R+• = {x ∈ R : x > 0} and Q+• = R+• ∩ Q, and take A to be an algebra. A real
semigroup, respectively, rational semigroup, in A is a morphism ψ : R+• → (A, · ),
respectively, ψ : Q+• → (A, · ), and these semigroups are non-zero if their range
does not contain 0.
The following theorem is due to Esterle [16]; see [6, Theorem 5.7.28].
Theorem 4.1. Let R be a commutative, radical Banach algebra. Then the
following conditions on R are equivalent:
(a) there is an embedding of F◦ into R;
(b) there is an embedding of R◦ into R;
(c) R contains an element of finite closed descent;
(d) R contains a real semigroup;
(e) R contains a rational semigroup;
(f) there is a sequence (an ) in R \ {0} such that an ∈ a2n+1 R (n ∈ N).
18 H. G. DALES
There are many further examples of commutative, radical Banach algebras

satisfying the above conditions in [6]. Here is a particularly easy one. Let A(D)
be the disc algebra of all continuous functions on the the closed unit disc D that
are analytic on the open unit disc D. Then A(D) is a uniformly closed subalgebra
of C(D). Let M = {f ∈ A(D) : f (1) = 0}, so that M is a maximal ideal in A(D).
Define
ζ +1
f0 (ζ) = exp (ζ ∈ D) ,
ζ −1
with f0 (1) = 0. Then f0 M is a closed ideal in M , and M/f0 M is a commutative,
radical Banach algebra that satisfies the equivalent conditions of Theorem 4.1.
4.2. The normability of quotients. We have shown that, with CH, for
each non-maximal, prime ideal P in an algebra C(Ω) such that |C(Ω)/P | = c, the
quotient algebra C(Ω)/P is normable. The natural question that arises is: Let I
be a non-maximal ideal in C(Ω). When is C(Ω)/I normable?
An easy extension of the main theorem shows that, with CH, a quotient C(Ω)/I
is normable whenever |C(Ω)/I| = c and the ideal I is a finite intersection of prime
ideals. In [6], it was said to be a ‘natural conjecture’ that the continuity ideal I(θ)
of a discontinuous homomorphism θ from an algebra C(Ω) into a Banach algebra is
always a finite intersection of prime ideals. This natural conjecture was proved by
Esterle in [13] for various compact spaces Ω, including βN and N∗ ; see [6, Theorem
5.4.35]. However this conjecture turned out to be false, as shown by work of Hung
Le Pham.
Here is a weaker result than the conjecture. First we give a definition from
[22, Definition 3.1] and [24, Definition 3.2].
Definition 4.2. Let A be a commutative algebra. A family {Pi : i ∈ S} of

prime ideals in A is pseudo-finite if a ∈ Pi for all save finitely many i ∈ S whenever
a ∈ Pi for some i ∈ S. A family P of prime ideals in A is relatively compact if every
sequence of prime ideals in P contains a pseudo-finite subsequence.
There is a sense in which there is a topology on the set of prime ideals in A

that makes a relatively compact family according to the above definition into a
relatively compact set with respect to this topology. Pham proved the following
theorem in [24].
Theorem 4.3. Let Ω be a locally compact, non-compact space, and let θ be

a homomorphism from C0 (Ω) into a commutative, radical Banach algebra. Then
ker θ = I(θ) is the intersection of a relatively compact family of non-modular prime
ideals in C0 (Ω).
The most interesting aspect of the above theorem is that the converse is ‘almost
true’, and so we are close to characterizing the ideals I in C0 (Ω) that are the
kernels of radical homomorphisms from C0 (Ω). The next theorem is taken from
[24, Theorems 6.7 and 6.8].
Theorem 4.4. (CH) Let Ω be a locally compact, non-compact space, and let
I be an ideal in C0 (Ω) with |C0 (Ω)/I| = c. Suppose that I is the intersection of
a relatively compact family P of non-modular prime ideals in C0 (Ω) satisfying the
additional condition that every chain in the closure of P is countable. Then there
is a homomorphism with kernel I from C0 (Ω) into a commutative, radical Banach

algebra, and so C0 (Ω)/I is normable.
The closure of P in the above theorem is the set of all the unions of pseudo-
finite sequences in P; the latter are necessarily non-modular, prime ideals in C0 (Ω).
Every chain in the closure of P is well-ordered by inclusion; if P is countable, then
every chain in its closure is indeed countable. It is possible that the ‘additional
condition’ in the above theorem is redundant.
The point of Theorem 4.4 is that there are many uncountable, locally compact,
metrizable spaces Ω, including the real line R itself, and corresponding ideals I that
satisfy the conditions in that theorem, without being finite intersections of primes,
and so there are ideals I in C0 (Ω) such that C0 (Ω)/I is normable, but I is not a
finite intersection of prime ideals. Indeed, Pham proves in [23] that, with CH, there
exists a radical homomorphism from C0 (R) whose kernel is not the intersection of
even countably many prime ideals.
4.3. C ∗ -algebras. As we remarked there are classical positive results about

the automatic continuity of arbitrary homomorphisms from (non-commutative) Ba-
nach algebras into Banach algebras [6]. The case when the domain is a C ∗ -algebra
is particularly interesting. For example, let H be a Hilbert space. Then each homo-
morphism from B(H), the C ∗ -algebra of all bounded operators on H, into a Banach
algebra is automatically continuous [6, Corollary 5.4.13]; more general results are
due to Johnson in [19].
The analogue of the theorem of Bade and Curtis giving the structure of an
arbitrary homomorphism θ from a C ∗ -algebra A into a Banach algebra B such that
θ(A) = B is mainly due to Sinclair [25]; see [6, Theorem 5.4.40].
Discontinuous homomorphisms from general C ∗ -algebras are constructed in [6,
Theorem 5.7.3] (based on [8]), as follows.
Theorem 4.5. (CH) Let A be a Banach algebra such that A has infinitely many
non-equivalent simple representations of dimension k13 for some k ∈ N. Then there
is a discontinuous homomorphism from A into some Banach algebra.
This suggested the following question, raised as Question 5.4.A in [6]: Let A be
a C ∗ -algebra, such that, for each k ∈ N, there are only finitely many non-equivalent
simple representations of A of dimension k. Does it follow that each homomorphism
from A into a Banach algebra is automatically continuous? If true, such a result
would characterize the C ∗ -algebras such that all homomorphisms from the algebra
are continuous and generalize all known theorems on the automatic continuity of
homomorphisms from C ∗ -algebras.
The following pleasing result was proved by Esterle in [15]; see [6, Theorem
5.4.27].
Theorem 4.6. Let Ω be a non-empty, compact space. Then every epimorphism

from C(Ω) onto a Banach algebra is automatically continuous.
13 A representation of dimension k of an algebra A is a homomorphism from A into the
algebra of k × k matrices acting on Ck ; the representation is simple if the only subspaces of Ck

that are invariant under all images of elements of A are trivial.
20 H. G. DALES
Of course, this suggested the following question, raised as Question 5.4.E in

[6]: Is every epimorphism from a C ∗ -algebra onto a Banach algebra automatically
continuous?
I regret to report that it seems that no conclusive progress has been made on
either of the above two questions. For the best partial results, see [12].
5. The independence of NDH

We now come to consideration of the fact that NDH and the solution to Kaplansky’s
problem is independent of ZFC; we shall discuss the early work of Woodin. It may
be of interest to record some history.
5.1. Historical remarks. The paper [5] was mainly worked out in the year
1973–74, whilst I was at UCLA; I thank Phil. Curtis for inviting me to UCLA and
for much support. I gave lectures on this at the inaugural conference on Banach
algebras at UCLA in July, 1974. The work was mostly written in the year 1974–75.
Let P be a non-maximal, prime ideal in C(Ω) with |AP | = c. For a long time,
I had a construction of a non-zero homomorphism from a ‘large’ subalgebra, B, of
AP into a commutative, radical Banach algebra, but I could not see that B was the
whole of AP . Suddenly, I realised that, whilst maybe B = AP , it was the case that,
with CH, B was isomorphic to AP , and that this was sufficient to obtain Theorem
1.8. The paper [5] was submitted in May, 1976.
I then thought how one could remove ‘(CH)’ from the theorem, but could not
do this; with some trepidation I wrote in June 1976 to Professor Robert Solovay at
Berkeley, and asked for his help in this. (Young colleagues might like to know that
one wrote by hand on paper in those distant days; but we did have airmail.) This
letter eventually reached Solovay at Caltech, where he was on leave.
The next part of the story is based on information from Frederick Dashiell,
then a Bateman Research Instructor at Caltech; I am grateful to Fred for this.
Hugh Woodin, then a junior at Caltech, approached Fred in January 1976 for
a topic for his senior thesis. Fred suggested that Hugh organize what was known
about Kaplansky’s problem up to that point, and explain the heart of the open
question on homomorphisms from C(βN, C). Hugh ignored the survey part of the
suggestion, and immediately started trying to construct a model of NDH; he was
learning about forcing at that time, and discussed the question with Solovay, the
leading expert on forcing. Hugh produced a type-written document [29] that I have
in ‘Fall 1976’; by that time Hugh had seen the preprint of [5]. It seems that this
document was not submitted as a senior thesis, and that it has not been published,
but results from it are contained in Hugh’s thesis [30] and in [9]. The paper [29]
seems to be Hugh’s first contribution to mathematics.
In fact, Hugh proved in [29, §5] that it is consistent with ZFC that there is
an ultrafilter U such that ∞ /U does not admit a non-zero algebra semi-norm,
and, in [29, §6], he gave a set-theoretic condition, now ‘Woodin’s condition’, which,
if satisfied, implies NDH. Subsequently Solovay showed that this condition was
consistent with ZFC, and lectured on this on 26 October 1976; this argument was
never published because Hugh himself soon gave a shorter proof of the same result. I
am impressed that, as an undergraduate, Hugh could confidently use such sentences
as: ‘Let M be a countable standard model of ZFC + CH, and let N be the extension
obtained by adding ω2 Cohen reals.’
In the work [29], Hugh also discusses ‘weak semi-normability’; some of these
remarks reappear in [9].
After some time Solovay kindly replied to my letter, saying that ‘Woodin’ had
shown that there are models of set theory in which NDH is true. I wrote to the
person that I took to be ‘Professor Woodin’ at Caltech to ask about this, and
received [29] in response.
5.2. Woodin’s condition. I wish to recall the proof of Woodin’s condition;

it is taken from [9, Chapter 3].
Recall first that the divisibility order and the strong Fréchet order F were
defined in Examples 2.1.
Let g ∈ NN with 1 F g, where 1 now denotes the sequence that is constantly
equal to 1. Then
g = {f ∈ NN : 1 F f F g} ,
so that (g, F ) is a partially ordered set. We regard (gU , <U ) as a totally
ordered subset of the ultrapower (Rω /U, <U ).
Theorem 5.1. Assume that there is a free ultrafilter U on N such that c 0 /U

is semi-normable. Then there is an unbounded, monotonically increasing function
g ∈ NN and an isotonic map from (gU , <U ) into (NN , <F ).
Proof. There is a commutative, radical Banach algebra (R, · ) and a non-
zero homomorphism θ : c 0 /U → R. Choose a ∈ c 0 /U with θ(a) = 0; we may
suppose that a > 0. Then choose a sequence α ∈ c 0 such that [α]U = a; we may
suppose that 0 < αn < 1 (n ∈ N). Finally, choose an unbounded, monotonically
increasing function g ∈ NN such that

1 1
(5.1) log → ∞ as n → ∞ .
g(n)2 αn
We shall now construct an isotonic map π : (gU , <U ) → (NN , <F ) as a com-
position
(gU , <U ) → (c 0 /U, ) → (R \ nil R, ) → (NN , <F ) .
ν θ τ
The map θ : c 0 /U → R is given, and clearly this map is isotonic; the maps ν and
τ will be anti-isotonic.
Take f ∈ g, and set
2

2

αf /g = αnf (n)/g(n) .
2
It follows from (5.1) that αf /g ∈ c 0 \ {0}. For [f ] ∈ gU , set
2
ν([f ]) = [αf /g ] ∈ c 0 /U ;
the map ν : gU → c 0 /U is well-defined. Suppose that [f1 ] <U [f2 ] in gU ; we
may suppose that 1 ≤ f1 (n) < f2 (n) ≤ g(n) (n ∈ N). Set

2

β = αn(f2 (n)−f1 (n))/g(n) .
Again by (5.1), we have β ∈ c 0 \ {0}, and clearly ν([f1 ]) [β] = ν([f2 ]), so that
ν([f2 ]) ν([f1 ]) in c 0 /U. Thus ν : (gU , <U ) → (c 0 /U, ) is anti-isotonic.
22 H. G. DALES
We next claim that the range of θ ◦ ν is contained in R \ nil R. Indeed, take

f ∈ g and k ∈ N, and set

2

γ = αnkf (n)/g(n) .
Then γ ∈ c 0 and ν([f ])k [γ] = a, and so (θ ◦ ν)([f ])k = 0, giving the claim.
For x ∈ R \ nil R, define
τ (x)(n) = min{k ∈ N : k ≥ 1/ xn } (n ∈ N) ,
so that τ (x) ∈ NN . Take x, y ∈ R \ nil R with x y, say x = yz, where z ∈ R.
Then z ∈ nil R, and
xn −1 ≥ y n −1 z n −1 (n ∈ N) .
Since R is a radical algebra, z n −1 → ∞ as n → ∞, and so τ (x)(n) > τ (y)(n)
eventually. Thus τ (y) <F τ (x), and so τ : (R\nil R, ) → (NN , <F ) is anti-isotonic.
Set π = τ ◦ θ ◦ ν : (gU , <U ) → (NN , <F ). Then π is an isotonic map, as
required.
We write Z : n → n, so that Z ∈ NN . The conclusion of the following theorem

is Woodin’s condition from [29].
Theorem 5.2. Assume that there is a non-empty, compact space Ω and a

discontinuous homomorphism from C(Ω, C) into a Banach algebra. Then there is
a free ultrafilter V on N and an isotonic map from (ZV , <V ) into (NN , <F ).
Proof. Let g and U be as specified in Theorem 5.1, and then define the set σn
to be g −1 ({n}) for n ∈ N, so that each σn is finite and {σn : n ∈ N} is a partition
of N. Define
V = {σ ⊂ N : {σn : n ∈ σ} ∈ U} .
Certainly V is a filter on N. If σ ⊂ N, then either σ ∈ V or N \ σ ∈ N, and so V is
an ultrafilter on N. Indeed V is a free ultrafilter on N.
Take f ∈ Z. Then (f ◦ g)(n) → ∞ and g(n) − (f ◦ g)(n) as n → ∞, and so
f ◦ g ∈ g. The map
[f ]V → [f ◦ g]U , (ZV , <V ) → (gU , <U ) ,
is well-defined and isotonic. Thus the result follows from Theorem 2.8(i) and The-
orem 5.1
The force of the above theorem is, of course, that in the case where (given a
model of ZFC) there is a model of ZFC in which there is no isotonic map from
(ZV , <V ) into (NN , <F ), it will follow that ZFC + NDH is relatively consistent
with ZFC. The construction of such a model was an achievement of Woodin; the
construction is given in Chapter 6 of [9], where the theory of gaps in totally ordered
sets is developed, and in Chapter 8 of [9], where we explain ‘iterated forcing’,
following the seminal work of Solovay and Tennenbaum [27]. Explicitly we have
the following theorem of Woodin [9, Theorem 8.25].
Theorem 5.3. Assume that M is a model such that M |= ZFC. Then there is
a model N extending M such that N |= ZFC + MA + NDH.
The proper forcing axiom PFA is a widely-used strengthening of Martin’s ax-

iom, MA. Stevo Todorčević proved in his book [28] (see Theorem 8.8 and some
remarks on page 87) that Woodin’s condition follows from PFA. This means, first
of all, that NDH is a theorem of ZFC + PFA. It also explains why it may be difficult
to obtain models of ZFC + MA + ¬NDH since these cannot be models of PFA.
We conclude this section with the following theorem.
Theorem 5.4. Assume that ZFC is consistent. Then NDH is independent from
ZFC.
6. Norming larger fields and super-real fields

In §2.6, we raised the question: Let K be an ordered field. When is the subalgebra
K of finite elements of K normable? In Theorem 3.5, we stated that this is true
for the specific ordered field R, and this implies that, with CH, K is normable
for each ordered field K with |K| = c. Now we investigate whether this is true for
some larger fields than R.
6.1. A theorem of Esterle. First we give a theorem of Jean Esterle [13]

that gives an upper bound to the cardinality of fields K such that K is normable;
see [10, Theorem 5.3].
Theorem 6.1. Let K be an ordered field such that K is normable. Then

|ΓK | ≤ c, and so |K| ≤ 2c .
Proof. The norm on K is denoted by · .
First take a, b ∈ K ◦ \ {0} with v(a) > v(b), where v is the archimedean valua-
tion, so that a = bc for some c ∈ K ◦ . Then 0 < an ≤ bn cn (n ∈ N). Since
1/n
limn→∞ cn = 0, it follows that an < bn eventually. The map
ψ : a → (an ) , K ◦ \ {0} → RN ,

induces an injection from ΓK into RN . Thus |ΓK | ≤ RN = c.
Now assume GCH. Then we know whether K is normable for every ordered
field K except for fields that satisfy the following combination of cardinalities:
|K| = ℵ2 and |ΓK | = ℵ1 . (∗)
The first question is whether there are any ordered fields satisfying (∗), and, if so,
whether any fall into ‘nice’ classes such as the classes of ultrapowers, of hyper-
real fields, or of super-real fields. Note that, with CH, the ordered field R is an
ultrapower, but that, in the theory MA + ¬CH, by a result related to Theorem
3.8, R is not even order-isomorphic to a super-real field [10, Corollary 6.26].
It is shown fairly easily in [10, Theorem 7.2] that, with GCH, there is indeed
a super-real field KP , with P a prime z-ideal, that satisfies (∗), and we can find
a hyper-real field that satisfies (∗) if we are allowed the combinatorial principle
‘diamond’ (3), a principle that is consistent with ZFC. Is there an ultrapower that
satisfies (∗)? In [10, Theorem 7.18], it is shown that it cannot be proved in ZFC
that there is not an ultrapower satisfying (∗).
The second question is whether K is normable for fields in this class. The first
obvious conjecture is the following.
24 H. G. DALES
Conjecture 6.2. Let K be an ordered field such that ΓK is a β1 -group (so

that |ΓK | = ℵ1 with CH). Then K is normable.
A proof of the following weaker conjecture would still amount to an attractive
strengthening of Theorem 3.6.
Conjecture 6.3. Let K be an ordered field such that w(K) = ℵ1 . Then K

is normable.
6.2. The field R. Recall that, in §3.1, we defined R to be F(1) (G). We now

define R to be the following ‘much bigger’ field than R:
= F(G) = F(F(1) (Q)) .
R
The following theorem is part of [10, Theorem 2.23].
+, · , ≤) is a real-closed η1 –field with value group

Theorem 6.4. The field (R,
G. Further,
= 2ℵ1 and cof R = δ(R) = ℵ1 .
R = w(R)

The field R is universal in the class of fields K such that ΓK is a β1 -group:

every such
field is isomorphic to a subfield of R [10, Theorem 2.24]. Since |R| = c
ℵ1
and R = 2 , it is possible that |R| = R , but the two fields are different, in the
sense that they are not even order-isomorphic [10, Theorem 2.23(vi)]. In particular,
R is not a β1 -field.

We see that, with GCH, R = ℵ2 and ΓR
= ℵ1 , and so R satisfies (*), above.
In fact, the field R is a test case for Conjecture 6.2: the algebra K is normable
for each ordered field K such that ΓK is a β1 -group if and only if R is normable.
We do not know whether R is normable.

We now consider whether the field R belongs to one of our ‘nice’ classes. How-
ever, it is shown in [10, Chapter 6] that this R is not even a super-real field, and

so maybe R is not too interesting for our normability question.
To show that a specific field K does not belong to a particular class of fields,
one can seek an invariant of the class that K does not possess. The invariant that
is used in [10, Chapter 6] is that of ‘having an exponentiation’; this is a special case
of the theory of an operational calculus for super-real fields, and this latter theory
has proved to be important.
The following is [10, Definition 6.15].
Definition 6.5. Let K be an ordered field, and let I be a convex subgroup of

K. Then an exponentiation on I is an order-isomorphism
exp : I → K + \ {0}
such that:
(i) exp(a + b) = (exp a)(exp b) (a, b ∈ I);
(ii) exp 0 = 1 and exp 1 = e1.
In the discussion of prime ideals in algebras C(Ω), we have mentioned some

subclasses of the class of all prime ideals. Here is another such class [10, Definition
4.45].
Let Ω be a compact space. Then a prime ideal P in C(Ω) is an exponential
prime if the function exp(−1/f ) belongs to C(Ω) \ P whenever f ∈ C(Ω)+ \ P .
Here exp(−1/f ) is defined pointwise. Clearly each prime z-ideal is an exponential
prime; there are exponential primes that are not z-ideals and valuation primes that
are not exponential primes. The following theorem combines Theorems 6.20, 6.22,
and 6.23 of [10].
Theorem 6.6. Let Ω be a compact space, and let P be a prime ideal in C(Ω).
Then there is an exponentiation on a convex subgroup of the super-real field KP , and
there is an exponentiation on KP itself when P is an exponential prime. However
there is no exponentiation on a convex subgroup of the field R, and so R is not
isomorphic as a field to any super-real field.
In particular, the above result shows that there is a real-closed η1 -field that is
not a super-real field. A stronger result will be mentioned below.
6.3. Complete ordered fields. As a preliminary to a discussion of the final

field that we wish to introduce, we define complete ordered fields; see [10, Chap-
ter 3]. The next definition is an obvious generalization of the basic concept of
completeness for a metric space.
Let G be a totally ordered group, and take an ordinal σ. Then a sequence
(xτ : τ < σ) is a Cauchy sequence if, for each ε > 0 in G, there exists σ0 < σ such
that |xτ − xρ | < ε whenever σ0 < τ < ρ < σ, and the sequence is convergent to
x ∈ G if, for each ε > 0 in G, there exists σ0 < σ such that |xτ − x| < ε whenever
σ0 < τ < σ. The group G is Cauchy complete if each Cauchy sequence of length
δ(G) is convergent to some element of G. A completion of G is a totally ordered
group H such that H is complete and G is isomorphic to an order-dense subgroup
of H.
Suppose further that δ(G) is infinite. Then it is easy to see that G has a unique
completion H, called the Cauchy completion, and that H is divisble whenever G is
divisible. Further ΓH = ΓG , and the archimedean valuation on H, when restricted
to G, agrees with the archimedean valuation on G.
The completion of G is denoted by G. We note that G ≤ 2w(G) .
Here is an example of a completion. Let S be a totally ordered set. Then the
divisible group F(S) is itself Cauchy complete. Now consider a subgroup G of F(S).
A function f ∈ F(S) belongs locally to G if, for each s ∈ S, there exists fs ∈ G such
that fs (t) = f (t) for all t ∈ S with t ≤ s. In particular, consider G = F(1) (S). Then
G is naturally identified with the space of functions in F(S) that belong locally to
G.
Now suppose that K is an ordered field. Then K is Cauchy complete if the
totally ordered group (K, +) is Cauchy complete. It is not hard to see that the
totally ordered completion K of the group (K, +) can be made into an ordered
field in such a way that K is a subfield of K; in the case that K is real-closed, the
field K is also real-closed. The field K is the Cauchy completion of K.
26 H. G. DALES
6.4. The field R. We finally come to my favourite real-closed field. We define

to be the Cauchy completion of R, so that R
R is the space of functions f in F(G)
that belong locally to R = F(1) (G).
The following theorem is part of [10, Theorem 3.21].
Theorem 6.7. The field (R, +, · , ≤) is a real-closed, Cauchy complete η1 –field,

with value group G. Further,

= c , and cof R = δ(R)
= ℵ1 .
R = 2ℵ1 , w((R)
With (CH), (R, +, · , ≤) is unique, in the sense that each Cauchy complete η1 –field
of weight ℵ1 is isomorphic to (R, +, · , ≤) as an ordered field.
The field R is universal in the class of completions of β1 –fields: every completion

of a β1 –field is isomorphic to a subfield of R [10, Theorem 3.22]. Thus R contains
every real-closed field of weight ℵ1 .
Recall that the real line R is the unique Cauchy complete (equivalently, Dede-
kind complete) real-closed η0 -field of weight ℵ1 , and that |R| = ℵ1 with CH. Thus
R is the natural analogue of R, one cardinality higher, and hence R is a ‘grown-up’
version of the real line.
We remark that R is different from both R and R. This is immediate on
cardinality grounds if we assume CH. However, in general, there is no isotonic map
from R into R and no isotonic map from R into R [10, Theorem 3.23], and so R
is not order-isomorphic to either R or R.
I would like to show that R is normable, but have not succeeded in this. The
field R is a test case for Conjecture 6.3, above: the algebra K is normable for each
ordered field K with w(K) = ℵ1 if and only if R is normable.

We now consider whether the field R belongs to one of our ‘nice’ classes. This
seems to be much harder to decide than the corresponding question for R.
We showed that R is not a super-real field by showing that it did not have an
exponentiation on any convex subgroup. However this route is not open to us for
R because there is an exponentiation on R itself [10, Theorem 6.24].
The first result that we mention is [10, Theorem 9.6(ii)]: the ordered field R
is not isomorphic as a field to any ultrapower. Indeed, a somewhat stronger result
is given in [10, Theorem 9.9].
Let U be a free ultrafilter on N, with corresponding ultrapower Rω /U. Then
R /U is not just a real-closed η1 -field of cardinality c, but it is also a ‘non-standard
ω
model of analysis’: much of the structure of R can be transfered to Rω /U. It is of

great importance in [10] that a version of this additional structure exists for each
super-real field of the form KP . This additional structure is rather straightforward
to construct when the prime ideal P is a z-ideal because we can work ‘pointwise’,
but it seems to be much more subtle when P is a general prime ideal.
A first fruit of this additional structure is a ‘gap theorem’ [10, Theorem 8.61].
This leads to a deep result [10, Theorem 8.63]: there is a real-closed η1 -field that
is not even order-isomorphic to any super-real field.
Study of this additional structure leads to [10, Theorem 9.26], the main theorem
of Chapter 9. It strengthens the wish to determine whether or not R is normable.
is a hyper-real field.
Theorem 6.8. (GCH) The ordered field R
6.5. Conclusions. The final two chapters of [10] are devoted to showing that
the additional structure that has been associated with a super-real field is sometimes
an invariant of the field. Thus any classification of super-real fields must involve
a prior classification of these structures. In some sense, it is shown that all the
structure of these super-real fields is determined just by the order structure of the
field. One consequence [10, Theorem 11.49] is that it is relatively consistent with
ZFC + GCH that there is a super-real field KP that is an η1 -field of weight ℵ1
such that KP is not even order-isomorphic to any super-real field KQ , where Q is
a prime z-ideal in some algebra C(Ω), and hence it cannot be proved in ZFC that
all super-real fields are order-isomorphic to a hyper-real field.
A final diagram in [10] considers twelve classes of real-closed η1 -fields, including
some classes not mentioned in this article, and eleven classes of super-real fields that
are η1 -fields. An aim was to determine, for any two classes, whether one is contained
in the other, and whether they are equal; ‘contained in’ could mean as fields or as
totally ordered groups or as ordered sets. We worked in the theory ZFC + GCH.
Rather a lot of inclusions are obtained; some inclusions are ‘relative consistency’
results obtained by forcing. We suspect that there are no more inclusions than
the ones that we established; we also suspect that results obtained by forcing are
not theorems of ZFC + GCH. There is a list of open questions in Chapter 12
of [10]; we are not aware of any resolution of any of these open questions, and we
commend contemplation of them, in particular to those who attended the conference
in Harvard.
It will be apparent that the original insights into the theorems of the second
half of [10], including all theorems that involve forcing, were due to Hugh Woodin.
My comment on his method of proof is the following: first, there were discussions on
the likelihood that one class was contained in another class, and possible counter-
examples were considered; the questions were clarified and the undergrowth cleared
away; then Hugh thought about the matter for some time, and drew a large number
of squiggly lines on a sheet of white paper; then he opined, using remarkable insight
on what would be true in certain models of set theory, that a particular containment
would be relatively consistent with ZFC + GCH. These insights were almost always
completely correct in the end; but it took many days and many pages to craft a
careful proof of these insights.
It was a privilege and a pleasure to work with Hugh on these books; I trust that
his extraordinary mathematical powers and insights will continue to inspire and to
produce beautiful and significant results in the higher reaches of set theory, logic,
and the philosophy of mathematics, and maybe even in Banach algebra theory.
References
[1] G. R. Allan, Embedding the algebra of formal power series in a Banach algebra, Proc. London
Math. Soc. (3) 25 (1972), 329–340. MR0305071
[2] G. R. Allan, Introduction to Banach spaces and algebras, Oxford Graduate Texts in Mathe-
matics, vol. 20, Oxford University Press, Oxford, 2011. MR2761146
[3] W. G. Bade and P. C. Curtis Jr., Homomorphisms of commutative Banach algebras, Amer.
J. Math. 82 (1960), 589–608. MR0117577
[4] W. W. Comfort and S. Negrepontis, The theory of ultrafilters, Springer-Verlag, New York-
Heidelberg, 1974. MR0396267
[5] H. G. Dales, A discontinuous homomorphism from C(X), Amer. J. Math. 101 (1979), no. 3,
647–734, DOI 10.2307/2373803. MR533196
28 H. G. DALES
[6] H. G. Dales, Banach algebras and automatic continuity, London Mathematical Society Mono-
graphs. New Series, vol. 24, The Clarendon Press, Oxford University Press, New York, 2000.
Oxford Science Publications. MR1816726
[7] H. G. Dales and J. Esterle, Discontinuous homomorphisms from C(X), Bull. Amer. Math.
Soc. 83 (1977), no. 2, 257–259. MR0430786
[8] H. G. Dales and V. Runde, Discontinuous homomorphisms from non-commutative Banach al-
gebras, Bull. London Math. Soc. 29 (1997), no. 4, 475–479, DOI 10.1112/S0024609397002981.
MR1446567
[9] H. G. Dales and W. H. Woodin, An introduction to independence for analysts, London Math-
ematical Society Lecture Note Series, vol. 115, Cambridge University Press, Cambridge, 1987.
MR942216
[10] H. G. Dales and W. H. Woodin, Super-real fields, London Mathematical Society Mono-
graphs. New Series, vol. 14, The Clarendon Press, Oxford University Press, New York, 1996.
MR1420859
[11] A. Dow, On ultrapowers of Boolean algebras, Proceedings of the 1984 topology conference
(Auburn, Ala., 1984), Topology Proc. 9 (1984), no. 2, 269–291. MR828984
[12] O. Ermert, Continuity of homomorphisms from AF-C ∗ -algebras and other inductive limit
C ∗ -algebras, J. London Math. Soc. (2) 54 (1996), no. 2, 369–386, DOI 10.1112/jlms/54.2.369.
MR1405062
[13] J. Esterle, Semi-normes sur C(K) (French), Proc. London Math. Soc. (3) 36 (1978), no. 1,
27–45. MR0482215
[14] J. Esterle, Injection de semi-groupes divisibles dans des algèbres de convolution et construc-
tion d’homomorphismes discontinus de C(K) (French), Proc. London Math. Soc. (3) 36
(1978), no. 1, 59–85. MR0482218
[15] J. Esterle, Theorems of Gelfand-Mazur type and continuity of epimorphisms from C(K), J.
Funct. Anal. 36 (1980), no. 3, 273–286, DOI 10.1016/0022-1236(80)90091-9. MR571409
[16] J. Esterle, Universal properties of some commutative radical Banach algebras, J. Reine
Angew. Math. 321 (1981), 1–24, DOI 10.1515/crll.1981.321.1. MR597976
[17] J. Esterle, Remarques sur les théorèmes d’immersion de Hahn et Hausdorff et sur les corps
de séries formelles (French, with English summary), Quarterly J. Math. 51 (2000), no. 2,
211–219, DOI 10.1093/qjmath/51.2.211. MR1765791
[18] L. Gillman and M. Jerison, Rings of continuous functions, The University Series in Higher
Mathematics, D. Van Nostrand Co., Inc., Princeton, N.J.-Toronto-London-New York, 1960.
MR0116199
[19] B. E. Johnson, Continuity of homomorphisms of algebras of operators, J. London Math. Soc.
42 (1967), 537–541. MR0215110
[20] B. E. Johnson, Norming C(U ) and related algebras, Trans. Amer. Math. Soc. 220 (1976),
37–58. MR0415326
[21] I. Kaplansky, Normed algebras, Duke Math. J. 16 (1949), 399–418. MR0031193
[22] H. L. Pham, The kernels of radical homomorphisms and intersections of prime ideals,
Trans. Amer. Math. Soc. 360 (2008), no. 2, 1057–1088, DOI 10.1090/S0002-9947-07-04325-5.
MR2346483
[23] H. L. Pham, Uncountable families of prime z-ideals in C0 (R), Bull. Lond. Math. Soc. 41
(2009), no. 2, 354–366, DOI 10.1112/blms/bdp009. MR2496511
[24] H. L. Pham, The kernels and continuity ideals of homomorphisms from C0 (Ω), J. Aust. Math.
Soc. 88 (2010), no. 1, 103–130, DOI 10.1017/S1446788709000329. MR2770930
[25] A. M. Sinclair, Homomorphisms from C ∗ -algebras, Proc. London Math. Soc. (3) 29 (1974),
435–452. MR0358368
[26] A. M. Sinclair, Homomorphisms from C0 (R), J. London Math. Soc. (2) 11 (1975), no. 2,
165–174. MR0377517
[27] R. M. Solovay and S. Tennenbaum, Iterated Cohen extensions and Souslin’s problem, Ann.
of Math. (2) 94 (1971), 201–245. MR0294139
[28] S. Todorčević, Partition problems in topology, Contemporary Mathematics, vol. 84, American
Mathematical Society, Providence, RI, 1989. MR980949
[29] W. H. Woodin, Discontinuous homomorphisms from C(Ω) and the partially ordered set ω ω ,
unpublished notes, pp. 24, Caltech, Fall 1976.
[30] W. H. Woodin, Discontinuous homomorphisms of C(Ω) and set theory, Thesis, University of
California, Berkeley, 1984.
[31] W. H. Woodin, A discontinuous homomorphism from C(X) without CH, J. London Math.
Soc. (2) 48 (1993), no. 2, 299–315, DOI 10.1112/jlms/s2-48.2.299. MR1231717
Department of Mathematics and Statistics, University of Lancaster, Lancaster

LA1 4YF, United Kingdom
E-mail address: g.dales@lancaster.ac.uk
Volume 690, 2017
The enumeration degrees: Local and global

structural interactions
Theodore A. Slaman and Mariya I. Soskova

With congratulations to Professor W. Hugh Woodin on the occasion of his sixtieth birthday
Abstract. We show that the structure of the enumeration degrees De has a

finite automorphism base consisting of finitely many total elements below the
first jump of its least element. As a consequence we obtain that the rigidity
of the structure of the enumeration degrees is implied by the rigidity of the
local structures of the Σ02 enumeration degrees, the Δ02 Turing degrees and the
computably enumerable Turing degrees.
Contents
1. Introduction
2. Preliminaries
3. Step 1: An indexing of the total Δ02 degrees
4. Step 2: An indexing of the total IT (0e ) degrees
6. Applications
7. Determining the low enumeration degrees from the low co-d.c.e. and the
co-c.e. degrees
8. Determining the low co-d.c.e. enumeration degrees
from the co-c.e. degrees
References
1. Introduction
We continue our investigations of the notion of relative definability between
sets of natural numbers, presented as a degree structure. There is a spectrum of
notions giving a more precise meaning to the term “relative definability”. At the
two endpoints, the many-one degrees and the hyperarithmetical degrees, we have
2010 Mathematics Subject Classification. Primary 03D30.

Key words and phrases. Enumeration degrees, automorphism base, rigidity.
The first author was partially supported by National Science Foundation grant number DMS-
1301659. The second author was partially supported by an FP7-MC-IOF grant STRIDE (298471)
and by Sofia University Science Fund. We thank an anonymous referee for carefully reading an
earlier draft of this paper and for suggesting improvements both in content and presentation.
2017
31
32 T. A. SLAMAN AND M. I. SOSKOVA
complete descriptions of the degree structures and these descriptions are completely
different.
The partial ordering of the many-one degrees Dm was characterized algebraic-
ally by Ershov [Ers75] and Paliutin [Pal75]. It is the unique partial order P with
the following properties: it has cardinality the continuum, it is a distributive upper-
semi-lattice with least element, it has the countable predecessor property and given
any other distributive upper-semi-lattice L with least element, with cardinality
less than the continuum, and with the countable predecessor property and given an
isomorphism π between an ideal I in L and an ideal π(I) in P , there is an extension
π ∗ of π to an isomorphism between L and π ∗ (L), such that π ∗ (L) is an ideal in P .
That is to say that Dm is a universal object in a natural way. Along the same line,
ω
the automorphism group of Dm has cardinality 22 and every element of Dm other
than its least one, 0m , has a nontrivial orbit. Consequently, 0m is the only element
of Dm which can be defined by its order-theoretic properties.
The partial ordering of the hyperarithmetical degrees Dh was characterized
model theoretically by Slaman and Woodin [SW05]. It is biinterpretable with the
structure of second-order arithmetic. There is a way within the ordering Dh to
represent the standard model of arithmetic N, +, ∗, <, 0, 1 and each set of natural
numbers X so that the relation “ p represents the set X ⊂ N and x is the hyper-
arithmetical degree of X” can be defined in Dh as a property of p and x. It follows
that there is no nontrivial automorphism of Dh , it is rigid. Further, a relation
on Dh is order-theoretically definable within Dh if and only if it corresponds to a
degree-invariant relation on sets which is definable in second-order arithmetic.
We would like to know where in the spectrum of relative definability rigidity
breaks down. However, for the structures that form the middle of this spectrum
we have only a partial understanding. Simpson [Sim77] proved that the first order
theory of the Turing degrees is computably isomorphic to the theory of second-
order arithmetic. Slaman and Woodin conjectured that the relationship between
the structure of the Turing degrees and second-order arithmetic is much stronger.
Their biinterpretability conjecture is that DT can be characterized in the same
way as the hyperarithmetical degrees. Slaman and Woodin [SW05] established
that the structure of the Turing degrees has a finite automorphism base and as a
consequence obtained that the biinterpretability conjecture is true modulo the use
of finitely many parameters. They showed that the automorphism group of the
Turing degrees is countable and that every member has an arithmetically definable
presentation. Furthermore, they proved that every relation in DT , induced by a
degree invariant relation that is definable in second-order arithmetic, is first order
definable with parameters in DT . Finally they showed that biinterpretability and
rigidity are equivalent for DT .
We will focus on an extension of the Turing degrees, the structure of the enumer-
ation degrees, De . This structure arises from the relation enumeration reducibility,
introduced by Friedberg and Rogers [FR59] in 1959. Enumeration reducibility is
close to Turing reducibility in the spectrum of relative definability. Instead of an
effective transformation of computations, it is based on effective transformation of
enumerations. Turing reducibility can be expressed in these terms as well: a set
A can be computed from a set B if and only if every enumeration of B ⊕ B can
THE ENUMERATION DEGREES: LOCAL AND GLOBAL STRUCTURES 33
be effectively transformed into an enumeration of A ⊕ A, and so the Turing de-

grees have a natural isomorphic copy in De , the total enumeration degrees. Selman
[Sel71] showed a reverse connection: every enumeration degree is completely deter-
mined by the set of total enumeration degrees above it. Thus the total enumeration
degrees are an automorphism base for the enumeration degrees.
The first order theory of this extended structure was characterized by Slaman
and Woodin [SW97] also as computably isomorphic to second-order arithmetic. In
contrast to the Turing degrees, where we do not have many examples of natural
first order definitions of classes of degrees, in the enumeration degrees we have
quite a few. Kalimullin [Kal03] showed that the jump operation is first order
definable by a simple formula in the language of partial orders. Papers of Ganchev
and Soskova [GS12] and [GS15] extended Kalimullin’s method to obtain other
definability results. This line of investigation culminated in the paper by Cai,
Ganchev, Lempp, Miller and Soskova [CGL16] which established the first order
definability of the total enumeration degrees. The total enumeration degrees are
thus a definable automorphism base for the Turing degrees. This allows us to
transfer all results from Slaman and Woodin’s automorphism analysis of DT to
De : the enumeration degrees have a finite automorphism base, their automorphism
group is countable and every member has an arithmetically definable presentation,
relations on De , induced by degree invariant relations that are definable in second-
order arithmetic, are definable with parameters in De , and the rigidity of De is
equivalent to its biinterpretability with second-order arithmetic. Some of these
conclusions were obtained earlier by Soskova [Sos16], but the definability of the
total degrees provides a more systematic treatment.
We now know that the automorphism problems for the two structures are con-
nected: a nontrivial automorphism of the enumeration degrees induces a nontrivial
automorphism of the Turing degrees. Thus the biinterpretability conjecture for the
Turing degrees implies biinterpretability of the enumeration degrees and second-
order arithmetic.
In this article we study the complexity of automorphism bases for the enu-
meration degrees. We combine methods developed in our earlier paper [SS] with
previously known results to obtain a finite automorphism base of total enumeration
degrees below 0e . In [SS] we investigated the automorphism group of the local sub-
structure of the Turing degrees DT (≤ 0T ), consisting of the Δ02 Turing degrees. We
established that the local structure of the Turing degrees relates in the same way
to first order arithmetic as the Turing degrees and the enumeration degrees relate
to second-order arithmetic: it is biinterpretable with first order arithmetic modulo
the use of finitely many parameters and that full biinterpretability is equivalent to
rigidity for the structure DT (≤ 0T ). The first order definability of 0e and the total
enumeration degrees below 0e now allows us to tie the automorphism problem for
the Δ02 Turing degrees, or equivalently their biinterpretability with first-order arith-
metic, to the automorphism problem for the enumeration degrees, or equivalently
their biinterpretability with second-order arithmetic: a nontrivial automorphism of
De yields a nontrivial automorphism of DT (≤ 0T ). Finally, we connect the rigidity
of the enumeration degrees to the rigidity of an even finer structure: the computably
enumerable Turing degrees.
2. Preliminaries
We start with an overview of the main notions and results that are used in this
article. We also give a brief outline of the proof of the main result.
2.1. The enumeration degrees. In this subsection we will introduce enu-
meration reducibility and its associated degree structure together with certain prop-
erties that will be used in this article. For other results and background inforamtion
about the structure of the enumeration degrees we refer to Cooper [Coo90].
Definition 1 (Friedberg and Rogers [FR59]). A set A is enumeration re-
ducible to a set B (denoted by A ≤e B) if there is a c.e. set Φ, such that
A = Φ(B) = {n : ∃u(n, u ∈ Φ & Du ⊆ B)},
where Du denotes the finite set with code (canonical index) u under the standard
coding of finite sets.
A set A is enumeration equivalent to a set B (denoted by A ≡e B) if A ≤e B
and B ≤e A. The equivalence class of A under the relation ≡e is the enumeration
degree de (A) of A. The enumeration degrees are ordered by de (A) ≤ de (B) if and
only if A ≤e B. The least element in this ordering is 0e = de (∅), the set of all
c.e. sets. We define a least upper bound operation by setting de (A) ∨ de (B) =
de (A ⊕ B). The enumeration jump of a set A was defined by Cooper [Coo84], as
KA ⊕ KA , where KA = {e, x| x ∈ Φe (A)}. The enumeration jump of the degree
of a A is de (A) = de (KA ⊕ KA ). Thus the structure of the enumeration degrees
De , ≤, ∨, , 0e is a an upper semi-lattice with least element and jump operation.
We can already state the first of many definability results that will be applied
in this article.
Theorem 1 (Kalimullin [Kal03]). The enumeration jump operator is first
order definable in De .
It is not difficult to see that a set A is Turing equivalent to a set B if and only
if A ⊕ A is enumeration equivalent to B ⊕ B. Thus, the map ι, defined by
ι(dT (A)) = de (A ⊕ A),
is an embedding of DT in De , which preserves the order, the least element and
the least upper bound. McEvoy [McE85] showed that it also preserves the jump
operation. The image of the Turing degrees under this embedding is the set of total
enumeration degrees.
Definition 2. An automorphism base for a structure A with domain A is a
set B ⊆ A, such that for very pair of automorphism π1 and π2 of A if π1 and π2
agree on all elements in B then π1 = π2 .
The total enumeration degrees are an important substructure of the enumera-
tion degrees. The following result shows that every enumeration degree is uniquely
determined by the set of total enumeration degrees above it.
Theorem 2 (Selman[Sel71]). Let a and b be enumeration degrees. Then a ≤ b
if and only if {x ≥ b | x is total} ⊆ {x ≥ a | x is total}.
Selman’s theorem shows that the total enumeration degrees form an automor-
phism basis for the structure of the enumeration degrees. This theorem was further
strengthen by Rozinas [Roz78], who proved that every enumeration degree is the
greatest lower bound of two total enumeration degrees.
Many properties of the Turing degrees can be transferred to the enumeration
degrees via the following result.
Theorem 3 (Cai, Ganchev, Lempp, Miller and Soskova[CGL16]). The set of
total enumeration degrees is first order definable in De .
It follows, in particular, that any automorphism base of the Turing degrees is
an automorphism base of the enumeration degrees. Slaman and Woodin [SW05]
show that the set of Turing degrees below 0(5) is an automorphism base for DT .
Thus we have the following theorem:
(5)
Theorem 4. The set of total enumeration degrees below 0e is an automor-
phism base for the structure of the enumeration degrees.
A further definability result that we will use in this article concerns the relation
“c.e. in” on Turing degrees.
Definition 3. Let a and b be Turing degrees. We say that a is c.e. in b if
there are sets A ∈ a and B ∈ b, such that A is c.e. in B.
Theorem 5 (Cai, Ganchev, Lempp, Miller and Soskova [CGL16]). The image
of the relation “c.e. in” under the standard embedding of DT in De is first-order
definable in the structure of the enumeration degrees.
2.2. The coding method. The structure De (≤ 0e ), also called the local
structure of the enumeration degrees, consists of the enumeration degrees of Σ02
sets. The proof of Theorem 3 extends an earlier result of Ganchev and Soskova
[GS15]: the total enumeration degrees are definable in De (≤ 0e ). In [SS] we stud-
ied the properties of the local structure of the Turing degrees DT (≤ 0T ), which we
can view as a definable substructure of De (≤ 0e ). We described a method for repre-
senting the standard model of first-order arithmetic as a partial order and discussed
various ways to define such a structure in DT (≤ 0T ) with finitely many parameters.
In particular we introduced the notion of indexing, which we now transfer to the
setting of the enumeration degrees.
Definition 4. Let Z be a class of degrees, represented by a sequence {Zi }i<ω ,
i.e. Z = {de (Zi ) | i ∈ ω}. We say that p define an indexing of Z with respect
to the sequence {Zi }i<ω , if they define a model of arithmetic M and a function
ϕ : NM → De , such that ϕ(iM ) = de (Zi ).
Slaman and Woodin [SW86] showed that in DT (≤ 0T ) there is an indexing of
the c.e. degrees (with respect to the standard enumeration of all c.e. sets {We }e<ω )
which is definable from parameters. Note that the image of the Turing degree
dT (We ) under the standard embedding is the enumeration degree ι(dT (We )) =
de (W e ). In other words, the c.e. Turing degrees are mapped to the Π01 enumeration
degrees. Transferring Slaman and Woodin’s result to the enumeration degrees we
have the following:
Theorem 6 (Slaman and Woodin [SW86]). There are finitely many total
parameters in De (≤ 0e ) that define an indexing of the Π01 enumeration degrees with
respect to the sequence {W e }e<ω .
A basic indexing is an indexing of the Π01 enumeration degrees with respect to

the sequence {W e }e<ω . Parameters that define a basic indexing will be called basic
indexing parameters.
The main result in [SS] shows the existence of finitely many parameters in
DT (≤ 0T ) that define an indexing of the Δ02 degrees with respect to the sequence
{Xe }e<ω , where

• If {e}∅ is a total {0, 1}-valued function then Xe is the set with character-

istic function {e}∅ .
• Xe = ∅ otherwise.
Again, we can transfer this result to show that there is indexing definable from
finitely many parameters of the total enumeration degrees with respect to the
sequence{Xe ⊕ Xe }e<ω . We will say that this is an indexing of the total Δ02 degrees.
In the next section we improve this result to show that every set of basic indexing
parameters defines an indexing of the total Δ02 degrees.
Our next goal will be to extend the indexing power of any set of basic indexing
parameters outside the local structure. Let us denote by IT (0e ) the union of all
intervals [x, x ], where x is a total enumeration degree below 0e . The definability
of the image of the relation “c.e. in”, stated as Theorem 5, will be one of the main
tools used in Section 4, where we show that any set of basic indexing parameters
can be used to identify the total enumeration degrees in the set IT (0e ). This is
done in two steps.
We first define an indexing of the class of all enumeration degrees that are
images of a Turing degree y that is c.e. in and above some Turing degree x ≤T 0 .
This will be called a 0e -basic indexing and will be with respect to the sequence
X
{Ye ⊕ Ye }e<ω , where e = e0 , e1 and Ye = Xe1 ⊕ We0 e1 .
Then relativizing the results from Section 3, we show that every set of basic
indexing parameters defines an indexing of the total degrees in IT (0e ) with respect
to the sequence {Ze ⊕ Ze }e<ω , where e = e0 , e1 and:

• If {e0 }Xe1 is a total {0, 1}-valued function then Ze = Xe1 ⊕ {e0 }Xe1 .
• Ze = ∅ otherwise.
We will call this an indexing of the total IT (0e ) degrees.
Finally, in Section 5 we extend the power of any set of basic indexing parameters
to cover all total enumeration degrees in the interval [0e , 0e ]: we show that they
define an indexing of the total Δ03 degrees. We can iterate the results from Section
4 and Section 5 in relativized form to cover the total enumeration degrees in the
(n)
interval [0e , 0e ] for any natural number n. In particular, when n = 5 we see
that any set of basic indexing parameters determines the total enumeration degrees
(5)
below 0e . By Theorem 4 this is an automorphism base for the enumeration
degrees.

The goal of this section is to prove the following theorem:
Theorem 7. Every set of basic indexing parameters p ∈ De (≤ 0e ) defines an

indexing of the total Δ02 degrees.
This result is significantly stronger than that from [SS], where we showed
merely the existence of a finite set of parameters p ∈ DT (≤ 0T ), defining an index-
ing of the Δ02 degrees. We will outline the main steps that comprised the proof of
the result in [SS] and explain how they can be implemented and improved in the
context of the enumeration degrees. Using ideas that go back to works of Jockusch
and Posner we first showed that every Δ02 Turing degree can be determined using
four low Δ02 Turing degrees. Here a degree (Turing or enumeration) is low if its
jump is equal to the jump of the least degree. In the enumeration degrees we have
an even simpler way of determining the total Δ02 enumeration degrees from the low
enumeration degrees.
Theorem 8 (Ganchev, Soskova [GS15]). The set of low enumeration degrees
is first order definable in De (≤ 0e ). Every total enumeration degree y ≤ 0e is the
least upper bound of two low enumeration degrees.
Next we showed that every low Turing degree has a unique position with respect
to the c.e. Turing degrees and an additional set of Δ02 Turing degrees Z. The set Z
was constructed in a way that allowed us to argue that there are parameters that
index the c.e. sets and a sequence representing Z. Working in the richer world of
the enumeration degrees, we show that the image of the c.e. Turing degrees under
the standard embedding, the set of Π01 enumeration degrees, is sufficient to uniquely
determine the position of every low enumeration degree and hence of every total
enumeration degree below 0e . We prove this in the following two theorems. Recall
that a set is d.c.e. if it is the difference of two c.e. sets. The co-d.c.e. sets are
the complements of d.c.e. sets and the co-d.c.e. enumeration degrees are those that
contain co-d.c.e. sets. The first theorem shows that we can use the low co-d.c.e.
enumeration degrees in place of Z. This is where we step outside the realm of the
Turing degrees: the class of co-d.c.e. enumeration degrees is the first class in the
difference hierarchy that contains non-total elements. It is easy to see that every
d.c.e. set is enumeration equivalent to a co-c.e. set and hence is of total degree. The
degrees of co-d.c.e. sets (or equivalently of 3-c.e. sets) however need not be total.
These basic facts can be found in Cooper [Coo90].
Theorem 9. Let x, y be Δ02 enumeration degrees, such that x = 0e and y x.
There are Δ02 enumeration degrees gi , Π01 enumeration degrees ai and low co-d.c.e.
enumeration degrees ci , bi for i = 1, 2, such that:
(1) gi is the least element below ai which joins bi above ci ;
(2) x ≤ g1 ∨ g2 ;
(3) y g1 ∨ g2 .
The second theorem determines the position of every low co-d.c.e. enumeration
degree among all Δ02 enumeration degrees using the Π01 enumeration degrees.
Theorem 10. Suppose x is a co-d.c.e. enumeration degree, x = 0e and y is
a Δ02 enumeration degree, such that y x. There are Δ02 enumeration degrees gi
and Π01 enumeration degrees ai , ci , bi for i = 1, 2, such that:
(2) x ≤ g1 ∨ g2 ;
(3) y g1 ∨ g2 .
The proofs of these two theorems are lengthy and technical. We will leave them
for Section 7 and Section 8 respectively.
Before we proceed with the proof of the main result for this section we observe
the following consequence of Theorem 8, Theorem 9 and Theorem 10. We say that
a function f fixes an element a if f (a) = a.
Corollary 1. If π : De (≤ 0e ) → De (≤ 0e ) is an automorphism which fixes
all Π01 enumeration degrees, then π fixes:
(1) all low co-d.c.e. enumeration degrees;
(2) all low Δ02 enumeration degrees;
(3) all total Δ02 enumeration degrees.
Proof. Let π be an automorphism of De (≤ 0e ), which fixes all Π01 enumeration
degrees. Consider first a low co-d.c.e. enumeration degree x and let y = π(x). From
the definability of the low enumeration degrees it follows that y is low, and hence
Δ02 . By Theorem 10 it must be that y ≤ x. Now consider z = π −1 (x). This too
must be a low enumeration degree. Noting that π −1 is also an automorphism which
fixes all Π01 enumeration degrees we have in this case as well by Theorem 10 that
z ≤ x. So we have that z ≤ x and hence π(z) ≤ π(x), i.e. x ≤ y. But we already
know y ≤ x, thus x = π(x).
The other two statements follow in a similar way from Theorem 9 and Theorem
8 respectively, using the fact that the low and the total enumeration degrees are
definable subclasses of the Δ02 enumeration degrees.
Proof of Theorem 7. Let p be basic indexing parameters, M be the stan-
dard model of arithmetic defined by p and ϕ be the basic indexing defined by p. Let
{Ue }e<ω be an enumeration of all low co-d.c.e. sets, defined in the following way:
let e = e1 , e2 then Ue = We1 \ We2 if We1 \ We2 is of low enumeration degree
and Ue = ∅ otherwise. Using parameters p we first define an indexing θ of the low
co-d.c.e degrees with respect to the sequence {Ue }e<ω .
Let e be an enumeration degree. The property e ∈ NM is definable from p, so
we may assume that e = eM , for some natural number e. Recall that we have an
effective way to translate an arithmetic statement about e, say ψ(e), into a degree
theoretic statement ψ̂(e, p) so that:
N |= ψ(e) if and only if De (≤ 0e ) |= ψ̂(e, p).
We will say that ψ(e) is true in M to mean that ψ̂(e, p) is true in De (≤ 0e ). Using
Theorem 10 and the definability of the low enumeration degrees we define θ(e) as
the largest low enumeration degree in the set of all x with the following property:
for every ai , bi , ci and gi where i = 1, 2 for which there are elements eM M
ai , ebi
M M M M
and eci , such that ϕ(eai ) = ai , ϕ(ebi ) = bi and ϕ(eci ) = ci , and gi is the least
element below ai which joins bi above ci for i = 1, 2 and in M the statement:
“de (Ue ) is bounded by the join of the least enumeration degree
below de (Wea1 ) which joins de (Web1 ) above de (Wec1 ); and the
least enumeration degree below de (Wea2 ) which joins de (Web2 )
above de (Wec2 ).”
is true, then x ≤ g1 ∨ g2 .
Next we use Theorem 9 to define an indexing ψ of the low enumeration degrees
with respect to the sequence {Ve }e<ω , where Ve = We (K∅ ⊕ K∅ ), if this is a low
set, and Ve = ∅ otherwise. McEvoy [McE85] showed that low enumeration degrees
contain only Δ02 sets, so the sequence {Ve }e<ω consists entirely of Δ02 sets.
Assume that e = eM . Then ψ(e) is the largest enumeration degree out of all
low enumeration degrees x with the following property: for every ai , bi , ci and gi
where i = 1, 2 for which there are elements eM M M M
ai , ebi and eci , such that ϕ(eai ) = ai ,
θ(eM M
bi ) = bi and θ(eci ) = ci , and gi is the least element below ai which joins bi
above ci for i = 1, 2 and in M the statement:
“de (Ve ) is bounded by the join of: the least e-degree below
de (Wea1 ) which joins de (Ueb1 ) above de (Uec1 ); and the least e-
degree below de (Wea2 ) which joins de (Ueb2 ) above de (Uec2 ).”
is true then x ≤ g1 ∨ g2 .
Finally, we define an indexing δ of the total enumeration degrees with respect
to the sequence {Xe ⊕ Xe }e<ω , i.e. an indexing of the total Δ02 degrees. Recall,

that Xe is the set with characteristic function {e}∅ , if this is a total {0, 1}-valued
M
function and ∅ otherwise. Let e = e for some natural number e. We let δ(e) =
ψ(iM ) ∨ ψ(j M ), where iM and j M are enumeration degrees satisfying the following
arithmetic statement in M:
“ The pair (i, j) is lexicographically the least pair of indices, such
that Vi ⊕ Vj ≡e Xe ⊕ Xe ”.
That such a pair exists follows from Theorem 8.
4. Step 2: An indexing of the total IT (0e ) degrees

Recall that IT (0e ) consists of all intervals [x, x ], where x ≤ 0e is a total
enumeration degree. The main goal of this section is to prove the following theorem:
Theorem 11. Let p be basic indexing parameters. Then p define:
(1) a 0e -basic indexing.
(2) an indexing of the total IT (0e ) degrees.
The second statement of this theorem will follow easily from the first statement
using Theorem 7 relativized above any total enumeration degree. The challenge is
to prove the first statement. We need to show that we can determine the image of
the Turing degrees (under the standard embedding ι) of sets that are c.e. in and
above some Δ02 set. Fix a total b ≤ 0e and let a be the image of a Turing degree
that is c.e. in and above ι−1 (b). If a ≥e 0e then we can use Shoenfield’s jump
inversion theorem, the definability of the enumeration jump and the indexing of
the total Δ02 degrees to determine a.
Theorem 12 (Shoenfield [Sho59]). If a is c.e. in and above 0 then a = x
for some total x ≤ 0 .
So let us assume that a is not above 0e . We can reduce our task further by
using a relativized form of Sack’s splitting theorem with cone avoidance:
Theorem 13 (Sacks [Sac63]). Let x and y be nonzero c.e. Turing degrees.
There are low c.e. Turing degrees a and b, such that x = a ∨ b and a y and
b y.
What is left is to determine a, assuming it is low relative to b and not above
0e . The definability of the jump operation and the image of the relation “c.e. in”
will let us pick out the images of all degrees y which are low relative to b, not
above 0 and c.e. in b. The following theorem relativized to b shows that a can be
distinguished among all images of such y using only elements below 0e .
Theorem 14. If A and W are c.e. sets and Y is a low c.e. set then one of the
following is true:
(1) W ≤T Y ;
(2) A ≤T Y ;
(3) There are sets U and V computable from W , such that A joins U above
V , whereas Y does not join U above V .
Proof. Let A and W be c.e. sets and let Y be a low c.e. set. Suppose that
W T Y and A T Y . We will build a c.e. approximation {U [s]}s<ω to a set U
and a Δ02 approximation{V [s]}s<ω to a set V . We will only enumerate a number
n in U or change the value of V (n) at a given stage if a number less than n enters
W at the same stage. This ensures that U and V are both computable from W .
To ensure that A⊕U ≥T V we use a movable marker strategy. To every natural
number x we will attach a marker γ(x), initially γ(x) = x + 1. The value of the
marker can be raised during the construction if we observe a change in A γ(x)
or we make a change in U γ(x). Similarly, we will only change the value of V (x)
provided that we can raise the value of γ(x). The construction will ensure that
every marker reaches a finite limit value. Then V (x) = V [s](x), where s is the least
stage in the construction at which A[s] and U [s] are correct up to γ(x)[s].
Let {Θe }e<ω be a listing of all Turing functionals. For every e we have to also
satisfy the requirement:
Re : ΘYe ⊕U = V.
The strategy for Re is to search for an opportunity to diagonalize ΘYe ⊕U against
V . If at stage s we find a number x, such that ΘYe ⊕U (x) is defined at stage s and
the computation has use θe (x) < γ(x), then we will call x an opportunity for e at
stage s. The opportunity is realized once a number y < x enters W . We can then
change U γ(x) by enumerating γ(x) in U , while preserving U θe (x), and set
V (x) = ΘeY ⊕U .
If Y [s] θe (x) = Y θe (x) then we say that x is a correct opportunity. In
this case the diagonalization will remain true at all further stages. If Y is incorrect
infinitely often, we risk making γ(x) unbounded. We use the fact that Y is low
and hence we have Δ02 approximations to sets computable from Y , to avoid this.
We approximate whether or not x is a correct opportunity as follows. Using the
recursion theorem we can assume we know an index of this construction and so we
can find a Δ02 approximation to the set
Qe (w) = {x | ∃t > w[θe (x)[t] < γ(x)[t] & (Y θe (x))[t] Y ]}.

Here w is meant to be the first stage since Re started working with x, recorded
as we (x) in the construction. Once Re starts working with x at stage we (x) at
every further stage it checks if x ∈ Qe (we (x)). While this is true, the strategy tries
to turn x into an opportunity by using every possibility to raise the value of γ(x)
above θe (x). Changes in A give us such possibilities.
If x ∈ e (x) ↓ then there is an inconsistency. We can speed
/ Qe (we (x)), but ΘY,U
up the approximations to Θe , Y and Qe , until we either observe that the computa-
tion was incorrect or x enters Qe (we (x)). If x ∈/ Qe (we (x)) permanently then the
strategy will give up on x as a diagonalization witness.
This process will be repeated in turn for all x > e until a permanent diago-
nalization is achieved. If there are infinitely many Y -correct opportunities, none of
which are allowed to be realized by W then we will argue that W ≤T Y . If there

are only finitely many Y -correct opportunities for e and ΘYe ⊕U is total then we will
show that A ≤T Y .
Before we proceed to describe a formal construction, we need to consider how
all of the strategies work together. This complicates matters as now U can change
due to actions by other strategies. On the other hand if ΘY,U e is not total, there
is a risk that infinite activity of one strategy interferes with the work of lower
priority requirements. To solve the first problem an Re -strategy will set up a
restraint R(e + 1), and require that lower priority strategies do not modify U or V
on numbers less than R(e + 1). To solve the second problem, we use that Y is low
once again to approximate the set
Te (w) = {x | ∃s > w[ΘY,U

e (x)[s] ↓ & (Y θe (x))[s] Y ]}.
The Re strategy will act only for numbers x ∈ Te (we (x)). We will assume that
Turing functionals are defined so that if y < x and ΘZ (x)[s] ↓ then ΘZ (y)[s] ↓ and
e (x) ↑ for some number x then eventually Re will stop
θ(y) < θ(x). Thus, if ΘY,U
acting.
Construction
At stage 0, U0 = V0 = ∅, γ(x)[0] = x + 1 for all x, the restraint R(e)[0] = 0
for every e, the stage we (x) at which a number x starts being examined by Re is
initially undefined. At stage s > 0 we examine the requirements R0 . . . Rs−1 in turn
until one of them terminates the stage. Unless otherwise specified all parameters
inherit their values from the previous stage. If a requirement Re is cancelled at
stage s then we (x) is set to s + 1 for all x.
Requirement Re : Examine each x, such that R(e) < x < s, in turn until an
instruction forces us to stop.
First we ensure that the approximation to Y and to the sets Te and Qe agree
about x. We will say that they agree if
• ΘY,U
e (x)[s] ↑ and x ∈ / Te (we (x))[s] or
• ΘY,U
e (x)[s] ↓, x ∈ T e (we (x))[s] and x ∈ Qe (we (x))[s].
• ΘY,U
e (x)[s] ↓, x ∈ T e (we (x))[s], x ∈
/ Qe (we (x))[s] and θe (x)[s] ≥ γ(x)[s].
If none of the cases above hold then we speed up the approximations to Y , Θe ,
Te (we ) and Qe (we ) until they do. Next we pick the first of the following cases
which applies to x.
(1) If x ∈/ Te (we (x))[s] then we cancel all we (y) for y > x and move on to the
next requirement.
(2) If R(e + 1) ≤ θe (x)[s] or R(e + 1) < R(e) then we redefine R(e + 1) as
max(R(e), θe (x)[s] + 1). We cancel all lower priority requirements and all
we (y) for y > x and end this stage.
(3) If θe (x)[s] ≥ γ(x) and A(z)[s − 1] = A(z)[s] for some z < γ(x) then we
redefine γ(y) for all y ≥ x to be numbers larger than any seen in the
construction so far. We cancel all lower priority requirements and all
we (y) for y > x and end this stage.
(4) If x ∈/ Qe (we (x))[s] then we move on to x + 1.
(5) If V (x)[s] = ΘY,U
e (x)[s] then we cancel all we (y) for y > x and move on
to the next requirement.
(6) If θe (x)[s] < γ(x) and W (z)[s − 1] = W (z)[s] for some z < x then we set
V (x)[s] = 1 − ΘY,U
e (x)[s]. We set U (γ(x))[s] = 1 and reset γ(y) for all
y ≥ x to be fresh numbers. Finally we set R(e + 1) = γ(x) + 1. Cancel all
lower priority requirements and all we (y) for y > x and end this stage.
(7) If none of the above cases applies to x then move on to x + 1.
If all x are scanned without ending the stage then move on to the next requirement.
End of Construction
We will show by induction on e that Re is satisfied. We will furthermore show

that there is a stage se+1 , such that at every t > se+1 the Re strategy does not
initialize lower priority strategies or change the value of any parameters. Assume
inductively that this is true for i < e. We can find a stage se , such that at all stages
t > se , Re is examined, not initialized, the value of R(e) is fixed and the γ-markers
for x ≤ R(e) do not change.
Fix x > R(e) and assume that after stage sx ≥ se , the strategy Re does not
change any parameters when examining y < x. Then after stage sx the value
of we (x) does not change and x is examined by Re at every stage t > sx . If
x ∈/ Te (we (x)) then ΘY,Ue is undefined at x. Let se+1 be the stage at which the
approximation to Te (we (x)) stabilizes on x. Then at every stage t > se+1 the
strategy Re moves on to the next requirement under Case (1) for x, without further
actions.
If x ∈ Te (we (x)) then as x is examined at every stage t after the last time we (x)
was changed, let s1 > we (x) be the least stage, such that ΘY,U e (x)[s1 ] is defined
and Y -correct. Then after stage s1 Case(1) will not apply for x anymore. We
pass through Case 2 at stage s1 if R(e + 1)[s1 ] ≤ θe (x)[s1 ]. Hence the Y -correct
computation is preserved at all stages t > s1 and Case (1) or (2) will never again
apply for x. As θe (x) does not change after stage s1 it follows that there can be at
most one stage t > s1 , such that Case (3) applies to x. If there is such a stage then
we move the value of γ(x) above θe (x) permanently.
Let s2 ≥ s1 be the least stage such the approximation Qe (we (x)) stabilizes
on x at stage s2 and Cases (1) (2) and (3) do not apply for x. If x ∈ / Qe (we (x))
then at every stage t > s2 Case (4) applies for x, the strategy does not change any
parameters when examining x and x + 1 is always accessible. We say that x is not
an opportunity.
Suppose that x ∈ Qe (we (x)). Then there is a stage t ≥ we (x), such that
θe (x)[t] < γ(x)[t] and Y [t] θe (x) Y . As s1 is the least stage at which the com-
putation ΘY,U is Y -correct and we preserved it at all further stages, it follows that
θe (x) < γ(x)[t] at all t > s2 . If there is a stage s3 ≥ s2 such V (x)[s3 ] = ΘY,U
e (x)[s3 ]
then this inequality is preserved at all further stages thus Re is permanently satis-
fied. In particular if at any stage t ≥ s2 Case (6) applies then at stage t + 1 Case 5
will apply for x. At every stage t > s3 Re moves on to the next requirement under
Case (5) for x. It does not change R(e + 1) or change any Γ-marker.
The only other option for x is that at every stage t > s2 Case(7) applies for x.
In this case as well the Γ-markers for all y ≤ x do not change after stage s2 and x
does not cancel the values of we (y) for y > x. We will say that x is an unrealized
opportunity.
It follows from the analysis above that for every x the value of γ(x) reaches a
limit and so V ≤T A ⊕ U .
If ΘU,Y
e is not total then eventually Re will reach a number x, such that x ∈ /
Te (we (x)) and so Re will always move onto Re+1 under Case (1) when examining
x, without further actions. Suppose that ΘU,Y e is total. We claim that there is a
largest x, examined by Re at all but finitely many stages, such that x is a realized
correct opportunity and ΘU,Y e (x) = V (x). In this case as well, Re will always move
onto Re+1 under Case (5) without further action, thus completing the induction
step.
Assume towards a contradiction that every x > R(e) is either not an oppor-
tunity or an unrealized opportunity. Suppose that there is a number x0 , such
that every x > x0 is not an opportunity. We show that A ≤T Y , contradicting
our assumptions. Fix a number x > x0 . To compute A(x) we find a stage s in
the construction, such that ΘU,Ye (x)[s] ↓ and the computation is Y -correct. Then
A(x) = A(x)[s]. Indeed if A(x) changes after stage s then we will move the marker
γ(x) above θ(x) and x will become an opportunity.
If there are infinitely many unrealized opportunities then we show that W ≤T
Y , contradicting our assumptions: to compute W (z), search for an element x > z
and a stage s such that ΘY,U e (x)[s] ↓ and is Y -correct and θe (x) < γ(x). As there
are infinitely many unrealized opportunities we will eventually find such a pair of
elements s and x. Then W (z) = W (z)[s], otherwise x would become realized.
Proof of Theorem 11. Let p be basic indexing parameters and let M and
δ be the definable from p model of arithmetic and indexing of the total Δ02 degrees.

We will use p to define a 0e -basic indexing ϕ0e , i.e. an indexing with respect to
X
the sequence {Ye ⊕ Ye }e<ω , where Ye = Xe1 ⊕ We0 e1 and Xe1 is the e1 -th Δ02 set,
whenever e = e0 , e1 . So fix e and let e be a natural number, such that eM = e.

We determine ϕ0e (e) according to the following cases:
Case 1: If in M we have that ∅ ≤T Ye , then we use Theorem 12 to find iM , such
that in M the following statement is true:
“Xi is Turing equivalent to Ye .”

We then set ϕ0e (e) = δ(iM ) using that the enumeration jump is definable by
Theorem 1 and that the standard embedding ι preserves the jump operation.
Case 2: If in M we have that Ye ≡T Xe 1 and ∅ T Ye then by Theorem 14
relativized to Xe1 we have that if Y is c.e. in and above Xe1 and does not compute
∅ or Ye then there are Δ02 sets U and V , such that Ye joins U above V , whereas
Y does not join U above V . Using that the image of the relation “c.e. in” is

definable by Theorem 5 we can determine ϕ∅ (e) as the least element in the set of
all x ≥ δ(eM M
1 ) that satisfy the following statement: x = δ(e1 ) , 0e x, ι
−1
(x) is
−1 M M M
c.e. in ι (δ(e1 )) and δ(j ) ≤ x ∨ δ(i ) for every i and j, such that in M the
following statement is true
“Ye joins Xi above Xj .”
Case 3: If none of the case above are true then we use Theorem 13 to find i and j,
such that in M:
“Yi and Yj are c.e. in and above Xe1 , they are low over Xe1 , they
do not compute ∅ and Ye ≡T Yi ⊕ Yj .”

We use Case 2 to define ϕ0e (iM ) and ϕ0e (iM ) and set ϕ0e (e) = ϕ0e (iM ) ∨
0e M
ϕ (j ).

Now that we have defined ϕ0e we use it together with Theorem 7, relativized

above any total Δ02 enumeration degree, to define an indexing ξ 0e of the total
degrees in IT (0e ) with respect to the sequence {Ze ⊕ Ze }e<ω , where e = e0 , e1

and Ze = {e0 }Xe1 if this is a total {0, 1} valued function and Ze = ∅ otherwise.
The proof follows the definition of δ from Theorem 7.

The final step in our proof gives a method to obtain an indexing of the total
Δ03 degrees from an indexing of the total IT (0e ) degrees. Let x be a Turing degree
below 0 . If x ≥ 0 we can use Theorem 11 to determine the image of x under
the standard embedding, as it would be a total degree in IT (0e ). So suppose that
x 0 . We will use genericity to determine x in three steps. Consult [SW05] for
basic properties of generic sets.
Proposition 1. Let n be a natural number and x be a Turing degree. If
x ≤ 0(n+1) then x = (y1 ∨ y2 ) ∧ (y3 ∨ y4 ), where for each i = 1, 2, 3, 4 we have that
yi ≤ 0(n+1) .
Proof. If x = 0(n+1) then we use Sacks’ splitting theorem (Theorem 13)
relativized to 0(n) to split x into two low relative to 0(n) degrees y1 and y2 . Then
x = (y1 ∨ y2 ) ∧ (y1 ∨ y2 ). So suppose x < 0(n+1) and let i + 1 be the least number,
such that x 0(i+1) .
Fix X ∈ x. Recall that a set Y is 1-generic relative to X if Y meets or avoids
every Σ01 (X) set of finite binary strings. Every incomputable c.e. set computes a
1-generic set. Hence relativizing X ⊕ ∅(i+1) computes a 1-generic relative to X set
Y . We split Y into odd and evens numbers Y = U ⊕ V . The sets U and V are
mutually 1-generic relative to X, hence U ⊕ X and V ⊕ X form a minimal pair
relative to X. We further transform U ⊕ X and V ⊕ X using the functional C.
The functional C maps a pair of sets X and G to a set C(X, G) as follows. Let
G = Geven ⊕ Godd :

Geven (n − m) if Godd (n) = 0 and |Godd ∩ [0 . . . n]| = m
C(X, G)(n) =
X(m) if Godd (n) = 1 and n is the m-th element of Godd .
Note that G ⊕ X ≡T Godd ⊕ C(X, G). Let y1 = dT (Uodd ), y2 = dT (C(X, U )),
y3 = dT (Vodd ) and y4 = dT (C(X, V )). Then x = (y1 ∨ y2 ) ∧ (y3 ∨ y4 ).
In [SS] we showed that if G is 1-generic relative to X then so is C(X, G). If G is
1-generic then G ≤T G ⊕ ∅ . Hence for every i = 1, 2, 3, 4 we have that yi ≤ 0(n+1) .

We have reduced our task to finding a way to determine only the total degrees
x with x ≤ 0e . We will reduce it further to determining only the images of the
2-generic degrees. Recall, that a set is (n + 1)-generic, if it is generic relative to
∅(n) and a degree is (n + 1)-generic, if it contains an (n + 1)-generic set.
Proposition 2. If x ≤ 0(n+1) then there are (n + 1)-generic degrees gi ≤
0(n+1)
, i = 1, 2, 3, 4, such that x = (g1 ∨ g2 ) ∧ (g3 ∨ g4 ).
Proof. Fix X ∈ x. As in the proof of Proposition 1 if G is 1-generic relative to
X, then we can split it into even and odd bits L ⊕ R, and then use the functional C
to define g1 = dT (Lodd ), g2 = dT (C(X, L)), g3 = dT (Rodd ) and g4 = dT (C(X, R)).
We would like to ensure that all of the elements in this definition of x are also
(n + 1)-generic. It will be sufficient to ensure that G, C(X, L) and C(X, R) are
(n + 1)-generic.
We extend the definition of C so that it can have a finite binary string of
even length as a second parameter. Given a finite binary string τ , we can break
it up into two finite binary strings τeven and τodd , so that τeven (n) = τ (2n) and
τodd (n) = τ (2n + 1). We will write τ = τeven ⊕ τodd .
⎧
⎪
⎨τeven (n − m) if τodd (n) = 0 and |{x < n | τodd (x) = 1}| = m;
C(Y, τ )(n) = Y (m) if τodd (n) = 1 and n is the m-th element of τodd ;
⎪
⎩
↑ if τodd (n) ↑.
For every Z let De (Z) denote the dense set De (Z) = {σ | σ ∈ WeZ } ∪ {σ | ∀τ !
σ(τ ∈/ WeZ )}. Note that for every e, De (Z) is computable from Z . We need to
ensure that G meets De (X) and that G, C(X, L) and C(X, R) meet De (∅(n) ) for
every e. AsX ≤T ∅(n+1) it follows that these sets are computable from ∅(n+1) . We
build G as s σ[s] by the finite extension method using oracle ∅(n+1) . Let σ0 = ∅.
Suppose we have constructed σ[s].
(1) If s = 0, e then σ[s + 1] is the least extension of σ[s] in De (X).
(2) If s = 1, e then σ[s + 1] is the least extension of σ[s] in De (∅(n) ).
(3) If s = 2, e then let σ[s] = σL ⊕ σR . We can assume that |σL | is even. Let
ρ = C(X, σL ). We would like to extend σ[s] so that ρ has an extension
in De (∅(n) ). Find a finite binary string τ , such that ρˆτ in De (∅(n) ). Let
k = |τ |. We can represent ρˆτ as C(X, σLˆτ ∗ ) where τ ∗ is of length 2k and
is defined as τ (0)ˆ0ˆτ (1) . . .ˆ0ˆτ (k)ˆ0, i.e. the even bits correspond to τ and
the odd bits are all 0. We set σ[s + 1] to be the least string which extends
σ[s] and such that there is a string σR [s + 1] with σL [s]ˆτ ∗ ⊕ σR [s + 1] =
σ[s + 1].
(4) If s = 3, e then let σ[s] = σL ⊕ σR [s]. Similarly to the previous case, we
find a finite binary string τ , such that C(X, σR [s])ˆτ is in De (∅(n) ). Let
k = |τ |. Set σR [s + 1] = σRˆτ (0)ˆ0ˆτ (1) . . .ˆ0ˆτ (k)ˆ0 and let σ[s + 1] be the
least string which extends σ[s] and such that there is a string σL [s + 1]
with σL [s + 1] ⊕ σR [s + 1] = σ[s + 1].

Finally we show that 2-generic degrees can be determined from degrees with
images under ι in the set IT (0e ). Recall that a Δ02 set is high if its Turing jump
computes 0 and a Turing degree is high if it contains high set.
Theorem 15. There are high Turing degrees h1 < 0 and h2 < 0 , such that
if x is 2-generic then (h1 ∨ x) ∧ (h2 ∨ x) ≡T x.
Proof. Intuitively H1 and H2 will be constructed by combining Shoenfield’s
Jump Inversion Theorem with the attempt to make for every pair of Turing-
operators Φ, Ψ the following set dense:
{τ | ∃x(Φτ ⊕H1 (x) = Ψτ ⊕H2 (x)}).
Note that for a particular pair of Turing functionals (Φ, Ψ) this set is Σ01 (∅ ). Let
G be a 2-generic, such that ΦG⊕H1 and ΨG⊕H2 are total. If our attempt succeeds
then ΦG⊕H1 = ΨG⊕H2 . It will follow from the construction that if our attempt fails
and ΦG⊕H1 = ΨG⊕H2 = X then X ≤T G.

Using ∅ as oracle, we build sequences μs and λs so that H1 = s μs and H2 =
s λs . Let {Cs }s<ω be a computable in ∅ c.e. approximation of ∅ . For every s let
cs be the least element in C[s] \ C[s − 1]. Stages s, such that Cs cs + 1 = C cs + 1
will be called true stages. There are infinitely many true stages.
To every stage s we associate the set of possible extensions of a string σ at
stage s to be
Ps (σ) = {τ ! σ | ∀c, x(c < cs & c, x ≥ |σ| ⇒ τ (c, x) = Cs (c))}.
Note that for every s and every σ the set Ps (σ) is computable, although not
uniformly computable. If at every stage s we respect that μs+1 ∈ Ps (μs ) and
λs+1 ∈ Ps (λs ) then both sets H1 and H2 will be high, by the limit lemma:
C(c) = lim x Hi (c, x). To see this in detail, fix c and let s be larger than c
and the first stage s0 , such that Cs0 c + 1 = C c + 1. Then for every x, such
that c, x ≥ |μs |, we will have that H1 (c, x) = C(c). Similarly for every x, such
that c, x ≥ |λs |, we will have that H2 (c, x) = C(c). We also note that if s is a
true stage then for every t > s we have that μt ∈ Ps (μs ) and similarly λt ∈ Ps (λs ).
Now let Φe , Ψe , σe be a computable listing of all triples of Turing operators Φe
and Ψe and finite binary string σe , called requirements. Initially all requirements
are unsatisfied. Set μ0 = λ0 = ∅. At stage s + 1 we have constructed μs and λs .
For every unsatisfied requirement with index e < s in turn we ask the question
“Does there exists a string τ ! σe and strings μ ∈ Ps (μs ), λ ∈ Ps (λs ) and a natural
number x, such that:
Φeτ ⊕μ (x) ↓= Ψτe ⊕λ (x) ↓?
This question is computable in ∅ . Let e be the index of the least unsatisfied
requirement for which the answer is “yes”. Find such μ and λ of length larger than
the length of μs and λs respectively and set μs+1 = μ and λs+1 = λ. Declare the
requirement with index e satisfied. If the answer for all requirements is “no” then
we extend μs to μs+1 ∈ Ps (μs ) of length larger than |μs | and λs to λs+1 ∈ Ps (λs )
of length larger than |λs |.
Now let Ψ and Φ be, such that the set S = {τ | ∃x(Φτ ⊕H1 (x) = Ψτ ⊕H2 (x))}
is not dense. Let σ be a finite binary string with no extension in S. Let e be the
index of Φ, Ψ, σ. Let s + 1 > e be a stage, such that all requirements with indices
less that e that ever get satisfied are satisfied and such that s + 1 is a true stage.
At stage s + 1 the answer to the question for the requirement e is “no”. We claim
that for any set G ! σ, if ΦG⊕H1 and ΨG⊕H2 are total and equal to X then X
is computable from G. Indeed, to compute X(x), find the least μ ∈ Ps (μs ) and
λ ∈ Ps (λs ), such that ΦG⊕μ (x) ↓= ΨG⊕λ (x) ↓. Such a pair exists, because s + 1 is
a true stage and hence H1 and H2 are infinite paths in the sets Ps (μs ) and Ps (λs )
respectively, and because ΦG⊕H1 = ΨG⊕H2 . If X(x) = ΦG⊕λ (x), there would be
a natural number n, such that ΦG⊕H1 n (x) ↓= ΦG⊕λ (x). But then H1 n and μs
would have a common extension μ∗ in Ps (μs ) and so the answer of the question
would be “yes”. So, X(x) = ΦG⊕λ (x).
Definition 5. An indexing of the total Δ0n+1 degrees will be any indexing
with respect to the sequence {Xe ⊕ Xe }e<ω , where Xe = {e}∅
(n)
if this is a total
{0, 1}-valued functions and Xe = ∅ otherwise.
Theorem 16. Every set of basic indexing parameters defines an indexing of
the total Δ03 degrees.
Proof. Let p be basic indexing parameters with a model of true arithmetic

M. By Theorem 11 let ψ 0e be a definable from p indexing of the total IT (0e )

degrees. We use p to define an indexing δ 0e of the total Δ03 degrees with respect

to the sequence {Xe ⊕ Xe }e<ω , where Xe = {e}∅ if this is a total {0, 1}-valued
functions and Xe = ∅ otherwise.
Let e be an enumeration degree and e be a natural number, such that e = eM .

We determine δ 0e (e) according to the following cases:
Case 1: In M the statement “Xe belongs to a 2-generic degree.” is true. We
use Theorem 15 to find high sets H1 and H2 , such that dT (Xe ) is the greatest
lower bound of the degrees dT (Xe ⊕ H1 ) and dT (Xe ⊕ H2 ). Next we note that the
set Xe ⊕ H1 is computable from ∅ and hence its degree belongs to the interval
[dT (H1 ), dT (H1 ) ]. Thus, in M we can search for the least natural number i, such
that H1 ⊕ Xe ≡T Zi , where {Zi ⊕ Zi }i<ω is the sequence of representing all total

IT (0e ) degrees indexed by ψ 0e . Similarly in M we can search for the least natural

number j, such that H2 ⊕ Xe ≡T Zj . We define δ 0e (e) = ψ 0e (iM ) ∧ ψ 0e (j M ).
Case 2: In M the statement “Xe ≤T ∅ ” is true. Then we use Proposition 2 to find
in M the Δ3 indices i1 , i2 , i3 and i4 of 2-generic sets G1 , G2 , G3 and G4 , such that

dT (Xe ) = (dT G1 ∨dT (G2 ))∧(dT G3 ∨dT (G4 )). We use Case 1 to determine δ 0e (iM j )

0e M 0e M 0e M
for j = 1, 2, 3, 4 and set δ 0e (e) = (δ 0e (iM
1 ) ∨ δ (i2 )) ∧ (δ (i3 ) ∨ δ (i4 )).
Case 3: Otherwise, we use Proposition 1 to find in M the Δ3 indices i1 , i2 , i3
and i4 of sets Y1 , Y2 , Y3 and Y4 whose jumps are computable in ∅ and such that

dT (Xe ) = (dT Y1 ∨ dT (Y2 )) ∧ (dT Y3 ∨ dT (Y4 )). We use Case 2 to determine δ 0e (iM j )

0e M 0e M 0e M
for j = 1, 2, 3, 4 and set δ 0e (e) = (δ 0e (iM
1 ) ∨ δ (i2 )) ∧ (δ (i3 ) ∨ δ (i4 )).
6. Applications
Let p be basic indexing parameters. In Section 3 we showed that p define an
indexing δ of the total Δ02 degrees. In Section 4 we extended this to show that p

defines first a 0e -basic indexing ϕ0e and then an indexing ψ 0e of the total IT (0e )
degrees. Finally in Section 5 we extended this result one step further to show that

p define an indexing δ 0e of the total Δ03 degrees. We can continue in this fashion:

from p and δ 0e we can define a 0e -basic indexing ϕ0e and then an indexing ψ 0e of
the total IT (0e ) degrees. Then we use the methods of Section 5 with Theorem 15

relativized to 0 to define an indexing δ 0e of the total Δ04 degrees. Iterating further
we obtain our main result.
Theorem 17. Let p be basic indexing parameters. For every natural number
n, p define an indexing of the total Δ0n degrees.
Combining this result with Theorem 4 we obtain the following result:
Theorem 18. Every set of parameters p that define a basic indexing is an
automorphism base of De .
Proof. Let p be parameters that define a basic indexing and let π : De → De
be an automorphism of the enumeration degrees which fixes all element of p. Let M
be the standard model of arithmetic associated with p. For every natural number
n, nM is definable from p, hence π(nM ) = nM . By Theorem 17 using p we can
(4) (5)
define an indexing δ 0e of total Δ06 degrees. So if x ≤ 0e is a total enumeration
degree then there is an e, such that x = δ 0e (eM ) and so π(x) = π(δ 0e (eM )) =
(4) (4)
δ 0e (π(eM )) = δ 0e (eM ) = x. In other words π fixes all total degrees below 0e ,

(4) (4) (5)
which by Theorem 4 is an automorphism base for De . It follows that π is the

identity.
Finally we give a connection between local structures and the global structure
of the enumeration degrees.
Theorem 19. The rigidity of the global structure of the enumeration degrees
is implied by any of the following:
(1) The rigidity of De (≤ 0e );
(2) The rigidity of DT (≤ 0T );
(3) The rigidity of the computably enumerable Turing degrees.
Proof. Every automorphism of De induces an automorphism of De (≤ 0e ),
DT (≤ 0T ) and the computably enumerable degrees, because the enumeration jump,
the total enumeration degrees below 0e and the image of the computably enumer-
able degrees under the standard embedding of DT into De are each first order
definable in De by Theorem 1, Theorem 3 and Theorem 5 respectively. If we as-
sume that any of these structures is rigid then any automorphism of De would fix
all total enumeration degrees below 0e . This is obvious for De (≤ 0e ) and DT (≤ 0T )
and follows from Corollary 1 for the computably enumerable Turing degrees. From
Theorem 6 it follows that there are total enumeration degrees in De (≤ 0e ) that
define a basic indexing. By the previous theorem they are an automorphism base
for De . Thus it would follow that the only possible automorphism of De is the
identity.
Our investigations naturally lead to the following open questions:
Question 1. Does rigidity for any one of the local structures imply rigidity for
either of the other local structures? In particular, is every enumeration degree in
De (≤ 0e ) definable in De (≤ 0e ) from total degrees?
Question 2. Does rigidity for any of the local structures or De imply rigidity
for the structure of the Turing degrees?
7. Determining the low enumeration degrees from the low co-d.c.e. and
the co-c.e. degrees
This section is devoted to the proof of Theorem 9. We wish to prove that for
every pair of Δ02 enumeration degrees x, y, such that x = 0e and y x there are Δ02
enumeration degrees gi , Π01 enumeration degrees ai and low co-d.c.e. enumeration
degrees ci , bi for i = 1, 2, such that:
(2) x ≤ g1 ∨ g2 ;
(3) y g1 ∨ g2 .
In the previous sections we used the notation Φ(A) to denote the image of a
set A under the enumeration operator Φ, as it was useful to have different ways
of representing the relations “c.e. in” and “≤e ”. From now one we will adopt the
notation ΦA for Φ(A), the result of applying the enumeration operator Φ to A.
We will also use ΦA,B to denote Φ(A ⊕ B) and more generally ΦA1 ,...,An to denote
Φ(A1 ⊕ · · · ⊕ An ).
We will use a characterization of the low sets due to McEvoy and Cooper
[MC85].
Proposition 3. A set X is low if an only if X has a Δ02 approximation
{X[s]}s<ω with the following property: for every enumeration operator W , approx-
imated by its standard c.e. approximation {W [s]}s<ω , and every natural number x
if x ∈ W [s]X[s] at infinitely many stages s then x ∈ W X .
We will call an approximation with the property described above, a low Δ02
approximation.
Let X be a low Δ02 set with a low Δ02 approximation {X[s]}s<ω . Fix Y e X to
be a Δ02 set with a Δ02 approximation {Y [s]}s<ω . We will construct approximations
to sets Ai , Bi , Ci and Gi for i = 1, 2 to witness the statement of the theorem for X
and Y . To ensure that Ai is Π01 , the approximation {Ai [s]}s<ω will have Ai [0] = N
and for all s, Ai [s] ⊇ Ai [s + 1]. To ensure that any E ∈ {C1 , C2 , B1 , B2 } is co-d.c.e,
its approximation will have the property E[0] = N and for every natural number n
there will be at most one stage s, such that n ∈ E[s] \ E[s + 1]. In other words if
the number n leaves the approximation to E then it can return to it, but once it
returns, it can never leave again. Finally, the sets Gi , i = 1, 2 will be Δ02 as their
approximations will satisfy: Gi [0] = N and for every n, lim s Gi [s](n) exists.
Our construction must be designed so that the following requirements are met.
(1) Λi : Gi = ΛA i , ensuring that Gi is enumeration reducible to Ai .
i
Bi ,Gi
(2) Γi : Ci = Γi , ensuring that Bi joins Gi above Ci .
(3) Ω: X = ΩG1 ,G2 , ensuring that X is enumeration reducible to G1 ⊕ G2 .
(4) Next we ensure that, out of all sets with the first two properties above,
Gi has the least enumeration degree. Let (Θe , Φe )e<ω be a listing of all
pairs of enumeration operators. For every e we have the requirement:
Ai
Rie : If ΘΦ e
e ,Bi
= Ci then there is an enumeration operator Δe , such that
Ai
Φe
Gi = Δe .
(5) To ensure that Y is not enumeration reducible to G1 ⊕ G2 , we have the
following list of requirements. Let (Ψe )e<ω be a listing of all enumeration
operators. For every e we have the requirement:
Qe : ΨG e
1 ,G2
= Y .
(6) Finally to ensure that Bi and Ci are low we use Proposition 3. Let {Υi }i<ω
be a listing of all enumeration operators. For every e = i, x, we have the
lowness requirement:
Le : If x ∈ ΥB i
1 ,B2 ,C1 ,C2
holds at infinitely many stages, then x ∈
B1 ,B2 ,C1 ,C2
Υi .
7.1. Description of strategies. To every requirement we attach a strategy,

designed to work towards its satisfaction. The construction will then run in stages
and every stage will consist of activating different strategies. The choice of which
strategy is activated will depend on the actions of higher priority strategies that
are activated at the current stage and the approximations to the given sets. For
any given operator W that we approximate we will assume that if x ∈ W Z [s] then
x < s and x ∈ W Zs . We start with the description of the strategies Λi , Γi and Ω,
which are called global strategies and will be activated at the beginning of every
stage. The goal of each is to construct an enumeration operator, the name of which
coincides with theirs.
The Λ-strategies. The strategy Λi uses markers to build its operator. If n ∈

Gi \ ΛAi , then any axiom for n enumerated previously (if any) has become invalid.
i
As Ai is constructed as a Π01 set these axioms will not become valid ever again
and so the strategy abandons any old markers that it has previously assigned to
n. It then assigns to n a new marker, λi (n), chosen fresh (larger than twice the
largest number mentioned so far), and enumerates a new axiom n, Ai λi (n) in
the operator Λi . Here A x is meant to be all numbers that are in A and are less
than or equal to x. During the construction, whenever we extract n from Gi , we
will automatically also extract the marker λi (n) from Ai , so that we always have
Gi (n) = ΛA i (n). Note that this will be the only way in which we modify the set
i
Ai , so assuming that Gi is Δ02 this strategy will in fact produce an enumeration

operator that reduces Gi to Ai .
The Γ-strategies. The Γi -strategy also assigns to every natural number c a
marker γ(c). However, c will also have an additional marker gi (c), controlled by
the Ri -strategies. When activated, the strategy ensures that if currently c ∈ Ci
then there is a valid axiom for c in Γi of the form c, Bi γi (c), Gi gi (c). During
the construction we will permanently extract a number c from Ci if we can be sure
that all axioms for c enumerated so far in Γi can never be valid again.
The Ω-strategy. For every element x the strategy defines an Ω-marker ω(x).
We will distinguish between the current Ω-marker ω(x), but also keep track of
previously defined Ω-markers. We will denote the set of all Ω-markers ever defined
by the strategy for x by O(x). The marker ω(x) will always be chosen as a fresh
number, so that ω(x) ∈ G1 ∩G2 . Every axiom enumerated in Ω for x, say x, D1 , D2
will include some marker o ∈ O(x) in both D1 and D2 . Thus extracting o from
either oracle set will invalidate the axiom. The Ω-strategy makes sure that at every
stage s we have ΩG1 ,G2 [s] = X[s] by enumerating axioms for numbers in X[s] and
extracting numbers from G1 or G2 to invalidate axioms for numbers not in X[s].
Now we turn to the two more difficult requirements, Rie and Qe . We will handle
them using a tree of strategies. We will denote Rie strategies by α, and Qe strategies
by β. We order the requirements by priority linearly as follows:
R10 < · · · < R1e < R2e < Qe < R1e+1 . . .

Strategies of level n in the tree of strategies are assigned the n-th requirement in
the priority ordering. The branching in the tree is determined by the outcomes of
the strategies. The lexicographical ordering of nodes in the tree induces a priority
ordering of the strategies.
The R-strategies. Suppose that α is working towards satisfying the requirement
R1e . We will drop the subscript e in the discussion below. The strategy will have
three possible outcomes: f <L i <L w. The outcome f will signify that the strategy
A
was successful in diagonalizing ΘΦ1 ,B1 against C1 : a number c was found, such that
A
c ∈ ΘΦ1 ,B1 \C1 . We explain below how this outcome can be reached, however let us
first think about how such diagonalization can be preserved at further stages. The
strategy would like to prevent certain numbers from leaving the approximations
A
to A1 and B1 to keep c ∈ ΘΦ1 ,B1 . Restraints on B1 will be achieved through
initialization of lower priority strategies and the general rule that every time a
strategy restarts its work, it deals with fresh numbers, not used so far. Restraints
on A1 are more difficult, as changes in X might require changes in G1 and therefore

changes in A1 . In order to preserve A1 , the strategy will have to set things up as
follows.
The strategy α keeps a parameter s0 (α), where it records the first stage at
which it is activated since the last time it was initialized. The strategy α first
leaves room for the Ω-strategy to successfully define its operator on all numbers
less than s0 (α). Every time X s0 (α) changes, α is restarted. Thus α can assume
in its later work that X s0 (α) does not change and no number less than ω(s0 (α))
will be forced out of either Gi or Ai . We denote by s1 (α) the least stage, such
that α is not restarted after it. At stage s1 (α) the strategy selects a threshold dα
as the least number which is greater than ω(s0 (α)) and in G2 . The design of the
Ω-strategy will then ensure that α can from now on at any time preserve G1 (and
so A1 ) on elements greater than or equal to dα by extracting dα from G2 . Once all
of this is set up the strategy proceeds to its actual work.
A1
First α looks at the largest number l such that C1 l ⊆ ΘΦ ,B1 . We will call
this the length of inclusion. Note that this notion is different from the usual length
of agreement, but it plays a similar role: a bounded length of inclusion guarantees
that the two sets are different. If at the current stage this length is greater than
it was at previous stages, we say that the stage is expansionary. The strategy will
have outcome w at stages that are not expansionary. If from some stage onwards
A1
the strategy always has outcome w, then ΘΦ ,B1 = C1 as there is some number
A1
y ∈ C1 , such that y ∈ / ΘΦ ,B1 .
At an expansionary stage s, α starts working on its operator Δα . It scans
numbers one by one, trying to enumerate axioms for n ∈ G1 and checking that there
are no valid axioms for numbers n ∈ / G1 . If all numbers less than s are scanned
then the outcome will be i. If this is the true outcome at infinitely many stages
A1
then we will argue that ΔΦ α = G1 . If on the other hand α discovers a number
A1
n ∈ ΔΦ α \ G 1 , then this number will be used for a successful diagonalization and
α will have outcome f . To explain the mechanism behind this, we need to go into
the details of how Δα is constructed.
The strategy α has a list of chits Cα containing only numbers larger than
s1 (α). Every natural number n ∈ G1 is assigned a chit c(n) ∈ Cα . When this chit
is assigned it is a fresh number, the strategy Γ1 has not enumerated axioms for it.
We set its second marker g1 (c(n)) to n. Thus every Γ1 -axiom for c(n) that will be
enumerated from now on will use G1 n. The number c(n) will remain the chit
for n until (if ever) we see a change in G1 n. If we see such a change, we will
pick a fresh chit for n. The chits are used to define axioms for n in Δα . If n ∈ G1
A1
has been assigned the chit c(n) and c(n) ∈ ΘΦ1 ,B1 via an axiom c(n), DΦ , DB ,
then the strategy enumerates the axiom n, DΦ in Δα . We say that this axiom is
defined using the triple c(n), DΦ , DB .
ΦA1
Suppose that later on n ∈ Δα e \ G1 and the valid axiom for n in Δα is defined
via the triple c(n), DΦ , DB . We enumerate DB back in B1 , as lower priority
A1
strategies might have extracted numbers out of B1 , ensuring that c(n) ∈ ΘΦ ,B1 .
We then extract c(n) from C1 and achieve the desired diagonalization. To preserve
it at further stages we initialize all lower priority strategies and extract dα from
G2 , as discussed above.
Finally we must check that this change in C1 will not injure the strategy Γ1 .
Note that as n ∈/ G1 , any axiom defined for c(n) until this change occurred will be
invalid. To ensure that Γ1 is correct on c(n) we must invalidate all axioms defined
after n left the approximation to G1 . We do so by extracting the corresponding
γ1 (c(n)) markers. It follows that they are larger than max DB and so will not
interfere with the restraint on B1 . As long as α is not initialized or restarted this
will keep Γ1 correct. If α does get restarted, we will enumerate the number c(n)
back in C1 .
The Q-strategies. Let β be a Q-strategy working on the requirement Qe . We
drop the index e in the discussion below. The strategy β will have one outcome
f , but it will cancel all lower priority strategies whenever it acts. The goal of the
strategy is to construct an operator Ξ, such that ΞX = ΨG1 ,G2 on all numbers up
to and including the first difference between the sets ΨG1 ,G2 and Y . As Y e X,
it will follow that this strategy will eventually find a permanent finite difference
between ΨG1 ,G2 and Y .
Just like the R-strategies, β first leaves space for the Ω-strategy to define
its operator correctly by keeping a parameter s0 (β), the first activation after the
previous initialization. Whenever G1 s0 (β) or G2 s0 (β) change the strategy is
restarted. The last stage when β is restarted is recorded in s1 (β). From this stage
onwards it assumes that G1 s0 (β) and G2 s0 (β) are fixed.
At further stages the strategy enumerates axioms in Ξ for all elements x, such
that x ∈ ΨG1 ,G2 and x ≤ l, where l is the least difference between Y and ΨG1 ,G2 .
It finds the least valid axiom x, D1 , D2 ∈ Ψ. Let DX be the set of all n ∈ X, such
that O(n) ∩ (D1 ∪ D2 ) = ∅. Notice that as long as DX ⊆ X, we will also have that
D1 ⊕ D2 ⊆ G1 ⊕ G2 and hence x ∈ ΨG1 ,G2 . The strategy enumerates the axiom
x, DX in Ξ.
The Δ02 nature of X allows for the following possibility. A number n ∈ DX exits
the approximation to X, which in turn forces the Ω strategy to extract a number
from D1 or D2 out of G1 or G2 respectively. If this invalidates all axioms for x in
Ψ, at the next visit to β we will have x ∈ / ΨG1 ,G2 . If meanwhile the number n has
re-entered X, we are left with a mistake in Ξ. We can fix this by re-enumerating D1
and D2 back in G1 and G2 respectively. We implement this by attaching a list P
of promises to β. A new promise is made for every axiom enumerated in Ξ. It has
the form DX , DG1 , DG2 and means that we promise: if DX ⊆ X then DGi ⊆ Gi
for i = 1, 2. At every stage after s1 (α) the strategy ensures that it is keeping all its
promises. It does that by changing the approximation to G1 ⊕ G2 (only on numbers
greater than s0 (β)) according to the promise list and the current approximation to
the set X.
The actions of β ensure that at every stage s we have that ΞX agrees with
Y on the largest initial segment such that ΨG1 ,G2 agrees with Y . If this initial
segment is unbounded then for every x there would be infinitely many s, such
that ΞX (x)[s] = Y (x)[s]. By our choice of a low approximation for X and a Δ02
approximation to Y , it would follow that ΞX (x) = Y (x). As Y e X, the initial
segment on which Y and ΨG1 ,G2 agree must be bounded by a least number, say
l. So there must be some number x ≤ l such that at infinitely many stages s
we have that Y [s](x) = ΨG1 ,G2 [s](x). Now, if x ∈ Y it follows immediately that
Y (x) = ΨG1 ,G2 (x). If x ∈
/ Y then by construction x ∈ ΞX at infinitely many stages,
and now again by the lowness of X, x ∈ ΞX . By the way we design our promise
list it will follow that x ∈ ΨG1 ,G2 .
The L-strategies. Let e = i, x. The Le -strategy will be activated at all stages
s ≥ e. It’s goal is to ensure that if x ∈ ΥB i
1 ,B2 ,C1 ,C2
[s] at infinitely many stages
B1 ,B2 ,C1 ,C2
s then x ∈ Υi . Whenever it sees a valid axiom for x in Υi , it tries to
restrain elements in B1 , B2 , C1 and C2 in order to keep this axiom valid at all
further stages. The R strategies are the ones that can interfere with this plan.
We know, however that an R-strategy α only extracts numbers from B1 , B2 , C1 or
C2 that are larger than s1 (α) and it does this only once: when it implements a
diagonalization. So the Le -strategy operates as follows: if at stage s it sees that
x ∈ ΥB i
1 ,B2 ,C1 ,C2
[s] \ ΥBi
1 ,B2 ,C1 ,C2
[s − 1] then it restarts all R-strategies α with
current value of the parameter s0 (α) > e. By restarting an R-strategy α we change
the value of its parameter s1 (α), but do not change the value of s0 (α), hence Le
will have finite impact on the R-strategies.
7.2. Construction. We combine the ideas described above in a formal con-

struction. We have global strategies : Λ0 , Λ1 , Γ0 , Γ1 and Ω, the lowness strategies
Le for e < ω and a tree T ⊆ {f, i, w}<ω of R and Q strategies. Every strategy α
in the tree T is given higher priority than all strategies that extend it or that are
to the right of it in T .
To every strategy we attach parameters as follows:
(1) Λi has as parameter the operator that it constructs. It dynamically assigns

markers λi (n) to every number n.
(2) Γi has as parameter the operator that it constructs. It dynamically assigns
markers γi (c) to every number c.
(3) Ω also has it operator and assigns markers ω(x) to all natural numbers
x. For every x we collect all markers assigned by Ω to x until the current
moment in a set denoted by O(x),
(4) An Le -strategy has a flag we , which is initially 0 and can be turned to 1.
(5) An Rie -strategy α has the following parameters: s0 (α), the stage when
α started work after last being initialized; s1 (α) the stage after the last
time α is restarted; a threshold dα , used to ensure that Gi and Ai can
be preserved; the number lα , measuring the length of inclusion at every
stage; an operator Δα ; a list of chits Cα . In addition the strategy will
dynamically assign chits c(n) to natural numbers n, define their gi -markers
and record the stage at which this happened in sc(n) . When α is initialized
all parameters become undefined. When α is restarted all parameters
except s0 (α) become undefined. In both cases any chit that α extracts
from Ci is enumerated back in Ci .
(6) A Qe -strategy β has the following parameters: s0 (β), the stage when β
started work after its last initialization; s1 (β), the β-true stage after the
last restart (see Step V below); a number lβ , recording the first differ-
ence between Y and ΨG e
1 ,G2
; a list of promises Pβ ; a functional Ξβ ; If β
is initialized all parameters become undefined. When β is restarted all
parameters except s0 (β) become undefined.
Construction:
At stage 0 we initialize all strategies and set Ai = Bi = Ci = Gi = N. At stage
s ≥ 0 all parameters inherit their values from the previous stage unless they are
explicitly modified during stage s. We start stage s by visiting the global strategies
as follows:
Step I: Ω: Scan all n < s.
(1) If x ∈ X \ ΩG1 ,G2 then redefine ω(x) to be a fresh number and enumerate
in Ω the axiom x, G1 ω(x), G2 ω(x) (the axiom is defined with marker
ω(x)).
(2) If x ∈
/ X then scan all axioms for x in Ω. An axiom is potentially valid if it
is valid, or if all thresholds for Ri -strategies that it includes are curently
in Gi respectively. For every potentially valid axiom for x defined with
marker o ∈ O(x), extract o from G1 and λ1 (o) from A1 if o is even; extract
o from G2 and λ2 (o) from A2 if o is odd.
Step II: Λi : Scan all elements n < s. If n ∈ Gi \ ΛA

i
i
then redefine λi (n) to be a
fresh number and enumerate the axiom n, Ai λi (n) in Λi .
Step III: Γi : Scan all n < s. If n ∈ Ci \ ΓB i
i ,Gi
then redefine γi (n) to be a fresh
number. If gi (s) ↑ then set gi (s) = 0. Enumerate in Γi the axiom n, Bi γi (n), Gi
gi (n).
Step IV: L: For all e = i, x < s do the following:
(1) If we = 1 but x ∈/ ΥB
i
1 ,B2 ,C1 ,C2
[s] then set we = 0.
B1 ,B2 ,C1 ,C2
(2) If we = 0 and x ∈ Υi [s] then set we to 1 and restart all R-
strategies α with e ≤ s0 (α)[s].
Step V: Construction of δ: We construct a finite path δ[s] in T . The path δ[s] is

defined inductively. Set δ[s] 0 = ∅. Suppose that we have constructed δ[s] k. If
k = s, then set δ[s] = δ[s] k. Otherwise, we activate the strategy δ[s] k. We say
that s is a δ[s] k-true stage. If the strategy ends the stage then δ[s] = δ[s] k.
Otherwise the strategy produces an outcome o, and δ[s] k + 1 = (δ[s] k)ô. We
have three cases depending on the type of the strategy δ[s] k:
Case Rie Suppose that δ[s] k is an R1e -strategy α. Let s− be the previous α-true
stage and o− be the outcome that α had at stage s− . (If α has never been visited
then s− = 0 and o− = w). Pick the first case which applies:
(1) If s0 (α) is not defined then set s0 (α) = s. End this stage.
(2) If s1 (α) is not defined then set s1 (α) = s and dα = ω(s0 (α))+1. Set Cα =
{α̂, n | n > s1 (α)} where α̂ is the code of α in some fixed computable
coding of all finite binary strings. End this stage.
(3) If s1 (α) is defined, but X s0 (α) changed at a stage t, such that s1 (α) <
t ≤ s then restart α. End this stage.
(4) If o− = f then let the outcome be f .
ΦA1 ,B
(5) Let lα be the maximal number l such that C1 l ⊆ Θα α 1 . If lα is
bounded by the largest number in the set {lα[t] | t < s} (the set of all
previous values of this parameter) then let the outcome be w.
ΦA1
(6) If there is an element n, such that n ∈ Δα e \ G1 then pick the least
such n. Suppose the least valid axiom for n in Δα is defined via the
triple c, DΦ , DB . Enumerate DB back in B1 . Extract the chit c from

C1 . Extract from B1 all Γ1 -markers for c that are defined at stages t ≤ s
when n ∈ / G1 . Extract dα from G2 and λ2 (dα ) from A2 . Initialize all
lower priority strategies and let the outcome be f .
(7) Scan all n ≤ s and perform the following actions for each n:
If c(n) ↑ or if G1 [t] n = G1 [sc(n) ] n at some stage t ∈ [sc(n) , s] then
let c(n) be a fresh number in Cα and set sc(n) = s and g1 (c(n)) = n. If
n ∈ G1 \ΔΦ α and c(n) < lα then find the least valid axiom c(n), DΦ , DB
α
in Θα and enumerate the axiom n, DΦ in Δα .

Once all elements are scanned end with outcome i.
If δ[s] k is an R2e -strategy, the instructions are the same as above with G1 , A1 , B1 ,
and C1 swapped with G2 , A2 , B2 , and C2 .
Case Qe Suppose that δ[s] k is a Qe -strategy β. Pick the first case which applies:
(1) If s0 (β) is not defined then set s0 (β) = s. End this stage.
(2) If s1 (β) is not defined then set s1 (β) = s. End this stage.
(3) If G1 s0 (β) or G2 s0 (β) changed at a stage t, such that s0 (β) < t ≤ s,
then restart β. End this stage.
(4) If there is an entry DX , DG1 , DG2 in the list of promises Pβ , such that
DX ⊆ X, but DG1 ⊕ DG2 G1 ⊕ G2 , (in other words if there is a promise
that is currently not being kept) then enumerate DG1 in G1 and DG2 in
G2 . End this stage.
(5) Let lβ be the least number such that ΨG β
1 ,G2
(lβ ) = Y (lβ ). If there is a
number x ≤ lβ such that x ∈ Ψ G1 ,G2
\ Ξ then find the least valid axiom
X
x, D1 , D2 ∈ Ψ. Let DX = {n ∈ X | O(n) ∩ (D1 ∪ D2 ) = ∅}. Enumerate

x, DX in Ξ and DX , D1 , D2 in Pβ . Repeat this for each such number
and end this stage.
(6) If none of the above cases hold, β has outcome f .
We end stage s by canceling all strategies of lower priority than δ[s] and proceed
to stage s + 1.
End of construction
7.3. Verification. We note that, as no numbers are ever enumerated in A1

and A2 , it follows that both of these sets are in fact Π01 . We start the verification
with a technical lemma, which concerns promises. This lemma will then allow us to
show that the constructed sets G1 and G2 are Δ02 and that the Λ-strategy succeeds.
Lemma 1. Let β be a Qe -strategy not initialized in the nonempty interval of
stages [s, t]. Suppose β makes a promise DX , DG1 , DG2 at stage s. If
DG1 ⊕ DG2 (G1 ⊕ G2 )[t]
then there is a stage t ∈ (s, t], such that DX X[t ].
Proof. Suppose for concreteness that o ∈ DG1 \ G1 [t]. Note that by our
convention for approximating given operators we have that o < s. First note that
o cannot be a threshold of an R-strategy as higher priority strategies do not act
at stages in the interval [s, t], or else β would be initialized, and lower priority
strategies are initialized at stage s, hence have thresholds larger than s. Thus
o ∈ O(n) for some number n. As o ∈ / G1 [t] there must be a stage t ≤ t, such that

n∈ / X[t ] and o is the marker of a potentially valid axiom for n in Ω.
This axiom must have been potentially valid at stage s as well. To see this note
that if a threshold of an Rie strategy d is ever extracted from the respective set Gi
then it can never be re-enumerated in Gi , so if an axiom is not potentially valid
at stage s then it stays this way at all further stages. This is because Q-strategies
always initialize lower priority strategies when the make a new entry in the list
of promises, so if an entry includes a threshold of a lower priority strategy then
this threshold will never be extracted, and because higher priority strategies Rie
strategies initialize all lower priority strategies when they extract their thresholds,
thus this threshold can never enter a new entry of a list of promises and hence can
never be re-enumerated back in Gi . It follows that n ∈ X[s] or else the Ω-strategy
would ensure o ∈ / G1 [s], contradicting DG1 ⊆ G1 [s]. Thus t > s and n ∈ DX , thus

DX X[t ].
Lemma 2. Gi is Δ02 and ΛA

i = Gi .
i
Proof. Fix n. If n is a threshold then by Lemma 1 it follows that n can be

extracted from Gi at most once, as Qe strategies never need to enumerate thresholds
back in Gi . If n is an ω-marker n ∈ O(x) then again by Lemma 1 any extraction
of n from Gi corresponds to an extraction of x from X. Hence once X(x) stops
changing, so will Gi (n). As X is Δ02 , it follows that Gi is Δ02 .
Assume inductively that ΛA i (m) = Gi (m) for m < n. First note that if n ∈ / Gi
i
then by construction when we extract n from Gi , we also extract its λi -marker and
hence there is no valid axiom for n in Λi . Suppose that n ∈ Gi then let sn be a
stage, such that Gi [sn ] n = Gi [t] n at all t ≥ sn . Then the axiom enumerated
for n at stage sn will never be invalidated again.
The goal of the proof is to show that there is a true path on the tree, consisting
of strategies that are visited infinitely often and initialized finitely often. Strategies
along this path will be shown to be successful. This will be done inductively and
broken up into a couple of lemmas. The first one deals with Q-strategies.
Lemma 3. Let β be a Qe -strategy which is not initialized after stage s0 and
visited infinitely often. Then Qe is satisfied and there is a stage sβ after which β
does not end stages at which it is visited.
Proof. We may assume that strategies of higher priority than β do not change
the approximations to Gi at stages t > s0 , otherwise β would be initialized. After
stage s0 the strategy β has a fixed parameter s0 (β). As G1 and G2 are Δ02 by
Lemma 2 there will be a least stage s1 ≥ s0 after which G1 s0 (β) and G2 s0 (β)
do not change. At the next β-true stage after s1 the parameter s1 (β) attains its
final value. After stage s1 (β) Cases (1),(2) and (3) do not apply for β.
Suppose that the sequence {lβ [t]}t>s1 (β) is unbounded. We show that ΞX = Y ,
contradicting our choice of X and Y . Fix a natural number x. And suppose that
x ∈ ΞX . Let sx > s1 (β) be the β-true stage when the least valid axiom for x, say
x, Dx , is enumerated in Ξ. By construction there is an axiom x, DG1 , DG2 ∈ Ψ
and a promise DX , DG1 , DG2 ∈ P . Fix a stage s, such that DX ⊆ X[t] at all
t > s. Then by Lemma 1 at all β-true t > s DG1 ⊆ G1 [t] and DG2 ⊆ G2 [t]. Thus
x ∈ ΨG1 ,G2 [t]. As the sequence {lβ [t]}t>s1 (β) is unbounded and Y is Δ02 it follows
that x ∈ Y . Suppose now that x ∈ Y . Then there are infinitely many stages s such
that x ∈ Y [s] and lβ [s] > x. By construction at all such stages s, x ∈ ΞX [s]. By
our choice of low approximation for X, it follows that x ∈ ΞX .
We have shown that there is a least number lβ such that lβ [t] ≤ lβ at all β-true
stages t ≥ s1 (β). It follows that there is a least number x < lβ such that Y [t](x) =
ΨG1 ,G2 [t](x) at infinitely many stages t. If x ∈ Y then x ∈ Y \ ΨG1 ,G2 . Otherwise
x ∈ ΨG1 ,G2 [t] at infinitely many stages t at which x < lβ [t]. By construction it
follows that x ∈ ΞX [t] at infinitely many stages t and hence by the lowness of X,
x ∈ ΞX . There is a least valid axiom for x in Ξ for which the promise is permanently
kept. It follows that x ∈ ΨG1 ,G2 \ Y .
Furthermore for all numbers x < lβ that are in ΨG1 ,G2 at infinitely many stages
there will eventually be permanent valid axioms in the operator Ξ. So there will be
a stage se after which no new axioms are enumerated in Ξ and no new promises are
made. Then for every stage t > se Case (5) does not apply and the set of promises
does not change: Pβ [t] = Pβ [se ]. Denote this final value by Pβ . Let n be larger
than the maximal element of DX in any promise DX , DG1 , DG2 ∈ Pβ . Let sX
be the least stage after se , such that X n does not change after stage sX . There
are finitely many promises DX , DG1 , DG2 ∈ Pβ with DX ⊆ X. By Lemma 1 if at
stage s ≥ sX the strategy β keeps a promise DX , DG1 , DG2 by passing through
Case (4) then this promise is kept at all further stages. Thus after finitely many
passes through Case (4) all of these promises are kept forever and there is a stage
sβ , such that after stage sβ the strategy β always ends in Case (6). In this case too
the strategy does not end stages prematurely.
In order to prove that the leftmost path of nodes that are visited infinitely often
is the true path, we must also show that strategies along it are initialized finitely
often. We we will need to consider also the actions of the L- and R-strategies.
Lemma 4. The strategy Le satisfies its requirement and restarts R-strategies
finitely often.
Proof. Assume inductively that the statement is true for Le , where e < e.
Let e = i, x. There are finitely many pairs (s0 (α), α), such that α is an R-strategy
with s0 (α) < e at some stage in the construction. For each such pair either there is
a stage t > e at which α is cancelled or there is a stage s1 (α) after which α is not
restarted as X s0 (α) reaches a limit and L-strategies do not act by the inductive
hypothesis. Note that the number of times α can change the approximation to Bi
and Ci is bounded by the number of times it is restarted. Suppose that s is a stage
after all of these finitely many strategies have stopped changing the approximation
to Bi and Ci . If at all t ≥ s we have x ∈ / ΥB i
1 ,B2 ,C1 ,C2
[t] then Le is satisfied and
does not restart R-strategies after stage s.
If at stage t ≥ s we see that x ∈ ΥB i
1 ,B2 ,C1 ,C2
[t] then we restart all strategies
with s0 (α) ≥ e and change we to 1. These strategies are activated again after stage
t and can only change the approximation to B1 , B2 , C1 , C2 in relation to chits that
are chosen fresh after stage t. Thus x ∈ ΥB i
1 ,B2 ,C1 ,C2
is preserved at all further
stages and Le does not restart R-strategies after stage t.
Lemma 5. Bi and Ci are co-d.c.e
Proof. A chit c ∈ Cα can only be extracted once from Ci by α. This is
because different strategies have disjoint lists of chits and when an R-strategy α
is initialized, Cα is restarted and becomes defined to contain only fresh numbers.
Similarly, a Γi -marker γi (c) can only be extracted once from Bi by the strategy α,
such that c ∈ Cα .
Lemma 6. Let α be an Rie -strategy which is not initialized after stage s0 and
visited infinitely often. Then Rie is satisfied and there is a stage sα after which α
does not end stages at which it is visited. If α extracts a chit c from Ci after stage
s0 then c ∈/ ΓB
i
i ,Gi
.
Proof. Suppose for concreteness i = 1. After stage s0 the parameter s0 (α)

does not change and higher priority R- and Q-strategies do not make any further
changes to any of the sets: Ai , Ci , Bi , and Gi . As X is Δ02 and by Lemma 4 L-
strategies with index i < s0 (α) act finitely many times, there will be a least stage
s1 ≥ s0 after which α does not get restarted. At the next α-true stage after s1 the
parameters s1 (α), dα and Cα will attain their final value. The final value of dα is
ω(s0 (α))[s1 ] + 1. As X s0 does not change at any further stage it follows that
G1 dα does not change at any further stage. After stage s1 (α) Cases (1), (2) and
(3) do not apply for α.
Suppose that there is a stage s at which Case (6) applies for α. There is a
ΦA1
number n ∈ (Δα e \ G1 )[s]. At stage s the threshold dα is extracted from G2 and
all lower priority strategies are initialized. It follows from the proof of Lemma 1
that dα remains out of G2 [t] at all further stages t, hence G1 s does not change at
any further stage, in particular n ∈ / G1 . From this we get that A1 [s] s ⊆ A1 and
hence ΦA e
1
[s] s ⊆ Φ A1
e . Suppose that the valid axiom for n in Δα was defined via
the triple c, DΦ , DB at stage t < s. Then c was assigned to n at stage sc < t. At
stage sc the chit c was selected as a fresh number, i.e. no axiom for c enumerated
in Γ1 before stage sc + 1. The marker g1 (c) was set to equal n. Thus all axioms for
Γ1 defined at stages at which n ∈ G1 are invalid. At stage s the strategy extracts
also all Γ1 -markers defined at stages at which n ∈ / G1 . These markers are defined
after stage t and hence are all larger than max DB .
At stage s all lower priority strategies are initialized so that B1 [s] s ⊆ B1 at all
ΦA1 ,B
further stages. Thus c ∈ Θe e 1 \C1 and the requirement is satisfied. Furthermore
we have shown that c ∈ / ΓB
1
1 ,G1
.
If Case (6) never applies for α after stage s1 (α) then neither does Case (4).
Hence α does not end any true stage t > s1 (α) and does not extract any numbers
from B1 or C1 .
Suppose that there is a stage s5 , such that at all α-true stages t > s5 Case
(5) applies for α. It follows that there is a number x such that at infinitely many
ΦA1 ,B1
stages t > s5 we have that x ≤ lα [t] and x ∈ C1 [t] \ Θe e [t]. We know that A1
ΦA1 ,B
is Π01 and by Lemma 5 the sets B1 and C1 are co-d.c.e, hence Θe e 1 = C1 and
R1e is satisfied.
Finally suppose that at infinitely many α-true stages Case (7) applies. Suppose
ΦA1 ,B ΦA1
that Θe e 1 = C1 . We will show that Δα e = G1 . Fix a natural number n. If
n ∈ G1 then by Lemma 2 we know that G1 n will eventually stop changing and
ΦA1 ,B
n will be assigned a final chit c(n) ∈ C1 . As Θe e 1 = C1 there will be a least
valid axiom for c(n) in Θe that is valid at all large enough stages. So eventually a
valid axiom for n will be enumerated in Δα . If n ∈ / G1 [s] at an α-true stage s then
ΦA1
n∈
/ Δα e [s] or else Case (6) would apply for n.
From Lemma 3 and Lemma 6 we get immediately the true path lemma.
Corollary 2. There is an infinite path f in the tree, such that for every n,
f n is visited infinitely often and initialized only finitely often. We call it the true
path. All R and Q requirements are satisfied.
We complete the verification with two lemmas, showing that the strategies Γ1 ,
Γ2 and Ω succeed.
Lemma 7. ΓB
i
i ,Gi
= Ci
Proof. Suppose that ΓB i
i ,Gi
(m) = C(m) for m < c and let sc be a stage, such
that Ci c does not change after stage sc . If c ∈ Ci then no number x ≤ γi (c)
is extracted from Bi after stage sc . This is true, as only numbers e < c can have
γi -markers less than γi (c) and these markers are the only numbers ever extracted
from Bi , always in response to these numbers leaving the approximation to Ci .
Hence once the approximation to Gi gi (c) has also settled, there will be an axiom
for c in Γi which will never be invalidated.
Suppose that c ∈/ Ci . Then c is a chit for an element n used by an Ri -strategy
α in an attack at stage s and α is not initialized after stage s. It follows from
Lemma 6 that c ∈ / ΓB
i
i ,Gi
.
Lemma 8. ΩG1 ,G2 (n) = X(n).
Proof. At every stage s the Ω-strategy ensures that for every x < s,
ΩG1 ,G2 (x)[s] = X(x)[s].
Thus if x ∈ / X then x ∈/ ΩG1 ,G2 . If x ∈ X then let sx > x be a stage, such that
X x does not change at stages larger than sx . Consider the first R-strategy along
the true path with s0 = s0 (α) > sx . Let s1 be the final value of the parameter
s1 (α). It follows from Lemma 6 that (G1 ⊕ G2 )[s1 ] ω(s0 ) does not change after
stage s1 . Hence the axiom enumerated at stage s1 for x in Ω is valid at all further
stages.
8. Determining the low co-d.c.e. enumeration degrees

from the co-c.e. degrees
In this section we explain how to modify the construction from the previous
section to show that Theorem 10 is true. Let X be a low co-d.c.e. set and Y be a Δ02
set, such that Y e X. We wish to construct Π01 sets Ai , Bi , Ci , co-d.c.e. sets Gi for
i = 1, 2 to satisfy the same list of requirements, excluding the lowness requirements.
Bi ,Gi
For the requirements Λi , stating that Gi = ΛA i , and Γi , stating that Ci = Γi
i
,
we use the exact same strategies as in the previous section. Similarly we treat the
list of requirements {Qe }e<ω , where Qe states that ΨG e
1 ,G2
= Y , exactly the same.
We will only modify the strategies for the Ω-requirement, stating that X = ΩG1 ,G2
Ai
and for the list of R-requirements, where Rie states that if ΘΦ e
e ,Bi
= Ci then there
Ai
Φe
is an enumeration operator Δe , such that Gi = Δe .
8.1. Description of the modified strategies and the construction.
The Ω-strategy. In this case as well, the strategy Ω will define an Ω-marker ω(n)
for every natural number n. We will collect all ω-markers defined for a particular
x in a set denoted by O(x). The current Ω-marker ω(x) will always be chosen as
a fresh number, so that ω(x) ∈ G1 ∩ G2 . Every axiom enumerated in Ω for x, say

x, D1 , D2 will include some marker o ∈ O(x) in both D1 and D2 . Thus extracting
o from either oracle set will invalidate the axiom. The modification is that now
Re -strategies will make the choice whether o will be extracted from G1 or from G2 .
For this reason o will come with a flag i(o) which has values 1 or 2. Initially i(o) = 1
for all numbers o. At stage s the strategy makes sure that ΩG1 ,G2 [s] = X[s] by
enumerating axioms for numbers in X[s] and extracting numbers from G1 or G2 to
invalidate axioms for numbers not in X[s]. If o needs to be extracted from G1 or
G2 then it will be extracted from Gi(o) .
We replace Step I in the construction by the following:
Step I: Ω: Scan all n < s.
(1) If x ∈ X \ ΩG1 ,G2 then let ω(x) be a fresh number and enumerate in Ω the
axiom x, G1 ω(x), G2 ω(x) (the axiom is defined with marker ω(x)).
Set i(ω(x)) = 1.
(2) If x ∈
/ X then for potentially valid axiom for x in Ω, i.e. an axiom which
is not already invalidated due to the extraction of an Rie threshold d from
the respective set Gi , let o ∈ O(x) be the marker that was used to define
it. Extract o from Gi(o) and λi(o) (o) from Ai(o) .
Steps II and III remain the same and Step IV is deleted. To complete the
construction we first explain how the R-strategies must be modified.
The R-strategies. We will not be able to use the same design for the R-strategies
as in the proof of Theorem 9. The reason is that the sets Bi and Ci have to be
Π01 . Thus once a number is extracted from either set, it must remain extracted
at all further stages. This creates a potential conflict between R-strategies of the
same kind. A lower priority Ri -strategy α might need to diagonalize and extract
some number b from Bi to keep Γi correct. A higher priority Ri -strategy α might
be counting on b ∈ Bi . It has used a triple c, DΦ , DB with b ∈ Bi to define an
axiom for a number n in Δα . If n leaves the approximation to Gi and Db Bi ,
there is no way to correct the operator Δα . To resolve this conflict we will design
the R-strategies so that they are more careful about preserving the work of higher
priority strategies. We introduce three modifications.
Suppose that α is working towards satisfying the requirement R1e . It will as
in the previous proof first define its parameter s0 (α) and wait until X s0 (α)
stops changing. It will be restarted every time a change is observed. Then at
stage s1 (α) it will define its threshold dα . But it will not yet select its list of
chits. Before the strategy goes on to satisfying its own requirement - diagonalizing
A1 A1
ΘΦ ,B1 against C1 or constructing an operator Δα , such that ΔΦ α = G1 , it first
works towards ensuring that its attempts at diagonalization will not injure higher
priority R1 strategies that are also constructing their own operators. The strategy
α must ensure that all higher priority strategies have defined their final axioms
for numbers n ≤ dα . To do this the strategy will wait until all of these strategies
have enumerated in their operators axioms for every m ≤ dα , such that m ∈ G1 .
For every higher priority R1 -strategy α , such that α believes that α is building
an enumeration operator Δα , i.e. such that αî α, the strategy α will set the
flag i(n) to 2 for all numbers n > dα necessary to preserve some axiom that α has
defined. Note this is a finite process as G1 dα does not change at any further stage
and once an axiom is preserved it will remain valid at all further stages, as lower
priority R2 -strategies will be initialized at these stages. We denote by s2 (α) the

stage when α has completed this task. At this stage the strategy α finally selects
its list of chits Cα , containing numbers larger than s2 (α).
A1
Next α examines the largest number lα such that C1 lα ⊆ ΘΦ ,B1 , to
determine whether or not the stage is expansionary and has outcome w if it is
not. If the stage is expansionary α proceeds to check if Δα is correct and to add
more axioms for it if necessary.
Here as well we use chits to define axioms in Δα . Every natural number n ∈ G1
is assigned a chit c(n) ∈ Cα . When this chit is assigned it is a fresh number, the
strategy Γ1 has not yet defined its g1 -marker. We set g1 (c(n)) to n. We change
the chit for n every time G1 n changes, so that all axioms for c(n) in Γi assume
that n ∈ G1 . If there is no valid axiom for n in Δα and there is a valid axiom
c(n), DΦ , DB for c(n) in Θ we define a new axiom for n in Δα . Here is where we
introduce the second modification: the axiom for n will include all currently valid
axioms for numbers m < n, such that m ∈ G1 . This will ensure that if α later on
needs to attack with n then all numbers m < n that were in G1 when we defined
the axioms for n are also in G1 at the stage of the attack.
A1
Finally if α sees a number n, such that n ∈ ΔΦ α \ G1 (n) and the valid axiom
for n is defined using the triple c(n), DΦ , DB , then we attack : we extract c(n)
from C1 and extract from B1 the least markers γ(c(n)) used in an axiom which
assumes n ∈ / G1 . We extract dα from G2 and dump all m > n in G1 . We say that
this attack is caused by the chit c(n). This is the third modification.
Let us reevaluate the possible conflict described above with the newly designed
strategy. It will follow from the construction that G1 is co-d.c.e. Thus a number
can be extracted only once from G1 . If α is a lower priority strategy which attacks
with n at stage s then by the first modification this attack will be caused by a chit
that is defined after all axioms for m < dα in Δα are already defined, and hence
the change in B1 is not on a number used in any axiom for m < dα . By the third
modification all m > n are dumped in G1 at the stage of the attack and will never
again leave G1 , so α will never have to attack with them. Similarly, any number in
the interval [dα , n] that is in G1 at the stage of the attack will remain in G1 at all
further stages due to the extraction of the threshold dα by α . Any number in the
interval [dα , n) that is not in G1 at the stage of the attack is already out when α
defined the axiom of attack by the second modification, and α extracts numbers
defined after this stage. This leaves us with only one problematic number: the
number n and the possibility that α extracts a number out of B1 , that invalidates
the axiom that is used by α for n in the definition of Δα . But as n is extracted
only once from G1 it follows that α preserves any number seen in any axiom while
n ∈ G1 . Thus the conflict is resolved.
We modify Case Rie from Step V of the construction as follows:
Case Rie Suppose that δ[s] k is an R1e -strategy α. The strategy has three ad-
ditional parameters: s2 (α), the stage when α can start its work; mα , the largest
number scanned during a visit and Rα , the largest number from the set A1 used in
an axiom for Δα . Let s− be the previous stage at which α was visited and o− be
the outcome that α had at stage s− . (If α has never been or has been initialized
since stage s− then let and o− = w). Pick the first case which applies to α:
(1) If s0 (α) is not defined then set s0 (α) = s. End this stage.
(2) If s1 (α) is not defined then set s1 (α) = s and dα = ω(s0 (α)) + 1. End this
stage.
(3) If s1 (α) is defined, but X s0 (α) changed at a stage t, such that s1 (α) <
t ≤ s, then restart α. End this stage.
(4) If αî α and mα ≤ dα + 1 or a new axiom for an element m < dα was
enumerated in Δα since stage s− then for every n, such that dα < n and
λ1 (n) ≤ Rα , set i(n) = 2. Make s2 (α), Cα and Δα undefined. End this
stage.
(5) If s2 (α) is not defined then set s2 (α) = s. Set Cα = {α̂, n | n > s2 (α)}
where α̂ is the code of α in some fixed computable coding of all finite
binary strings. End this stage.
(6) If the outcome at stage s− was f then let the outcome be f .
(7) If the current stage is not expansionary then let the outcome be w.
ΦA1
(8) If there is an element n, such that n ∈ Δα e \ G1 , then pick the least
such n. Suppose the least valid axiom for n in Δα is defined via the triple
c, DΦ , DB and this is a valid axiom in Θe . Extract the chit c from C1 ,
extract from B1 the least marker γ1 (c) which is used in an axiom that
assumes n ∈ / G1 . Extract dα from G2 and λ2 (dα ) from A2 . Enumerate all
numbers m ∈ (n, s] in G1 . Initialize all lower priority strategies and let
the outcome be f .
(9) Scan all n ≤ s, such that n ∈ G1 , and perform the following actions:
If c(n) ↑ or if G1 n[t] = G1 n[sc(n) ] at some stage t ≥ sc(n) then
let c(n) be a fresh number in Cα and set sc(n) = s and g1 (c(n)) = n.
ΦA1 ,B
If c(n) is greater than the length of agreement between Θe e 1 and C1
then stop the scan. Otherwise if n ∈ / ΔΦα then find the least valid axiom
α
c(n), DΦ , DB in Θe . Let Dn be the union of the currently valid axioms in

Δα for m < n, such that m ∈ G1 , and enumerate the axiom n, Dn ∪ DΦ
in Δα .
Once the scan has been completed (or stopped) define mα as the largest
scanned element for which there is a valid axiom in Δα . We set Rα to be
the largest element that is in a Φe -axiom for an element in a Δα axiom
and end with outcome i.
If δ[s] k is an R2e -strategy, the instructions are the same as above with G1 , A1 , B1 ,
and C1 swapped with G2 , A2 , B2 , and C2 .
8.2. Verification. We start with a technical lemma that deals with the mod-
ified R-strategies and serves as an important tool for unravelling the construction.
Lemma 9. Let α be an R1e -strategy which extracts its threshold dα from G2
at stage s to perform a diagonalization relative to a number n. Suppose the valid
axiom for n in Δα is defined via the triple c(n), DΦ , DB at stage s < s. Then at
all stages t > s:
(1) There is threshold dα ≤ dα for a strategy α ≤ α, such that
dα ∈ (G1 ∩ G2 )[s − 1] and dα ∈
/ (G1 ∪ G2 )[t].
(2) If G1 n[s ] ⊆ G1 [t] then (B1 γ1 (c(n)))[s ] B1 [t]
(3) All numbers m in the interval [dα , s], such that m ∈ G1 [s] are in G1 [t].
The symmetric statement with subscripts 1 and 2 swapped is also true.
Proof. Suppose that the statement is true for all α of higher priority than
α. The strategy α selects its threshold dα at stage s1 (α) < s as ω(s0 (α)) + 1 and
α is not initialized between stages s0 (α) and s. At stage s it extracts dα from G2 .
First we show that a Q-strategy β does not enumerate dα back in G2 . Towards
a contradiction suppose that a Q-strategy β has a promise DX , DG1 , DG2 with
dα ∈ DG2 . This promise must have been made before stage s by β and β must
have higher priority than α. Lower priority strategies are initialized at stage s
and when a Q-strategy is initialized it moves the value of its parameter s0 (β) and
restarts every time a change in Gi s0 (β) for i = 1, 2 is observed. But in this
case, as α is not initialized between stages s0 (α) and s, it follows that β made this
promise before stage s0 (α), hence max DG2 < s0 (α) < dα . So if dα ∈ G2 [t] for
t > s then an R2 -strategy α must be responsible for this. Again strategies that
select their thresholds after stage s have thresholds of value greater than dα and
cannot enumerate dα in G2 . This rules out lower priority strategies. It follows that
α is of higher priority than α and at stage t it extracts from G1 its own threshold
dα < dα . Thus dα ∈ G1 [s − 1] \ G1 [t]. By induction the statement follows.
Now similarly suppose that n is enumerated back in G1 at some stage t ≥ s.
The same argument as the one above shows that n is not enumerated back in G1
by a Q-strategy β, or by a lower priority R-strategy. Thus this must be done by a
higher priority R1 -strategy α which attacked at stage t with a number m < n. Pick
the least number m, such that some R-strategy α of higher priority attacks at some
stage t > s with m. Then m ∈ / G1 [t ] at all t ≥ t. If m ∈ G1 [s ] then this proves
the statement. If m ∈ / G1 [s ] then the chit c (m) for m at α was selected while

m ∈ G1 hence before stage s , and even before the chit c(n) was assigned to n at α.
At stage s we have that m ∈ / G1 [s ] hence the least marker for c (m) that assumes
m∈ / G1 is already defined, it is in B1 [s ] and it is smaller than γ1 (c(n))[s ]. As α
extracts this marker from B1 at stage t it follows that (B1 γ1 (c(n)))[s ] B1 [t].
Finally let m ∈ [dα , s] and m ∈ G1 [s]. As higher priority strategies have
thresholds smaller than m at stage s and lower priority strategies are initialized at
stage s it follows that m cannot be extracted by an R-strategy. If m is an ω-marker
used in some axiom x, D1 , D2 in Ω then this axiom is defined in the time period
(s0 (α), s] and hence must have (G1 dα ⊕ G2 dα )[s − 1] ⊆ D1 ⊕ D2 . By the first
statement in this lemma it follows that this axiom is invalid at all further stages
and so the strategy Ω will never extract it from G1 .
Now we can establish that G1 and G2 are co-d.c.e. sets. As in the previous
section this lets us conclude that the Λ-strategies are successful.
Lemma 10. G1 and G2 are co-d.c.e. and ΛA
i = Gi .
i
Proof. We show that G1 is co-d.c.e. Let n be a natural number. Then n is

extracted from G1 for two reasons: either it is a threshold for some R2 -strategy or
else it is an Ω-marker for some natural number x. If n is a threshold then it is the
threshold of a unique strategy α, it is not an Ω-marker and can only be extracted
from G1 once by α. This is ensured by the way we select thresholds: relative to the
Ω-marker of the first stage after initialization of α. At this stage α initializes lower
priority strategies and reserves this stage for its own use.
If n is an ω-marker, i.e n ∈ O(x) for some x, then it can be extracted only
by the strategy Ω at stages s, such that x ∈ / X[s]. Once it is extracted, it can
be enumerated back by a Q-strategy at a further stage t ≥ s, only if there is a
promise DX , DG1 , DG2 , such that x ∈ DX and n ∈ D1 and Dx ⊆ X[t]. But as
X is co-d.c.e. it follows that x ∈ X[t ] at all further stages t ≥ t and hence Ω will
not extract n a second time. It can also be enumerated back by an R1 -strategy α
only when α attacks with a number m < n at stage t > n. But then by Lemma 9
we have that n ∈ G1 at all further stages. It follows that G1 is co-d.c.e.
A similar proof shows that G2 is co-d.c.e. The proof that ΛA i = Gi is can now
i
be implemented as in Lemma 2.
We next show that the Γi strategies succeed.
Lemma 11. ΓB
i
i ,Gi
= Ci
Proof. Suppose that ΓB i

i ,Gi
(m) = Ci (m) for m < c and let sn be a stage,
such that Ci c does not change after stage sc . If c ∈ Ci then Bi γi (c) does not
change after stage sc . Hence once the approximation to Gi gi (c) has also settled,
there will be an axiom for c in Γi which will never be invalidated. Suppose that
c∈/ Ci . Then c is a chit for an element n used by an Ri -strategy α in an attack at
stage s. Then gi (c) = n and at stage s α extracted from Bi the least marker γi (c)
used in an axiom that assumes n ∈ / Gi . The structure of the axioms in Γi ensures
that this will invalidate all later axioms for c defined until stage s. By the second
clause of Lemma 9 the axioms that assume n ∈ Gi are also invalid at all further
stages.
Lemma 12. Let α be an Rie -strategy which is not initialized after stage s0 and
visited infinitely often. Then Rie is satisfied and there is a stage sα after which α
Proof. For concreteness let i = 1. After stage s0 the parameter s0 (α) does not
change and higher priority R- and Q-strategies do not make any further changes
to any of the sets: Ai , Ci , Bi , and Gi . As X is co-d.c.e. there will be a least stage
s1 ≥ s0 after which X s0 (α) does not change. At the next α-true stage after
s1 the parameters s1 (α) and dα attain their final value. The final value of dα is
ω(s0 (α))[s1 ] + 1. As X s0 does not change at any further stage it follows that
G1 dα does not change at any further stage. After stage s1 (α) Cases (1), (2) and
(3) do not apply for α.
Suppose that α is a higher priority R1e -strategy, such that αî ⊆ α. As α is
visited at infinitely many stages, it follows that α has infinitely many expansionary
stages. Fix m ≤ dα . As G1 m does not change after stage s1 (α), the α -chit for
m, c (m), does not change after stage s1 (α). Hence once the length of inclusion
examined by α exceeds c (n), the strategy α enumerates a valid axiom for n in
Δα . At the first stage when α is visited after that it sets the flag i(x) = 2 for every
number x ≥ dα , such that a change in G1 (x) can cause a change in A1 Rα . As
higher priority strategies do not act and lower priority strategies are initialized at
this stage, no one can change the flag back to 1. Finally as by assumption X s0
and hence G1 dα do not change after stage s1 (α), it follows that A1 Rα is
preserved at all further stages and hence the axiom for m in Δα remains valid at
all further stages. It follows from this analysis that there is a stage s2 , such that
at all stages t ≥ s2 all higher priority strategies α have mα > dα and do not
enumerate more axioms for elements m < dα . At the first α-true stage after stage
s2 the final values of the parameters s2 (α) and Cα are defined. Cases (4) and (5)
do not apply to α at any further stage.
Suppose that Case (6) applies for α at a least stage s6 . Then at all further
stages t ≥ s6 the strategy α ends with outcome f and does not initialize lower
priority strategies. Consider the α-true stage before s6 , call it s. At stage s Case
(8) applies for α and dα was extracted from G2 and all lower priority strategies are
initialized. By Lemma 9 as higher priority strategies do not act we know that G1 s
and hence A1 s do not change at further stages and hence no number can leave
the approximation to ΦA e s at further stages. There is a a chit c ∈ Cα [s], such
1
ΦA1 ,B
that c ∈ Θe e 1 (c)[s] via the valid axiom c, DΦ , DB . We extract c from C1 and
extract from B1 only γ1 (c)-markers that assume that n ∈ / G1 , hence elements larger
than max DB . Then we initialize all lower priority strategies so that DB ⊂ B1 [t] at
ΦA1 ,B
all further stages t ≥ s. It follows that c ∈ Θe e 1 \ C1 and R1e is satisfied.
If Case (6) never applies for α after stage s1 (α) then neither does Cases (8).
Hence α does not end any true stage t > s2 (α). Suppose that there is a stage
s7 , such that at all α-true stages t > s7 Case (7) applies for α. It follows that
ΦA1 ,B1
Θe e C1 and R1e is satisfied.
=
Finally suppose that at infinitely many α-true stages Case (9) applies. Suppose
ΦA1 ,B ΦA1
that Θe e 1 = C1 . We will show that Δα e = G1 . Fix a natural number n. If
n∈/ G1 [t] at any stage t > s2 (α) then no axiom is ever enumerated in Δα for n. If
ΦA1 ,B
n ∈ G1 then n will eventually be assigned final chit c(n) ∈ C1 . As Θe e 1 = C1
there will be a least axiom for c(n) in Θe that is valid at all large enough stages.
So eventually a valid axiom for n will be enumerated in Δα .
Suppose that n ∈ / G1 but an axiom n, Dn ∪ DΦ is enumerated in Δα at stage
s. By Lemma 10 it follows that there is a stage sn > s, such that n ∈ G1 [t] at all
t < sn and n ∈ / G1 [t] at all t ≥ sn . We will show that the axiom n, Dn ∪ DΦ is
invalid. The axiom was defined via a chit c(n) and an axiom c(n), DΦ , DB valid
at stage s and includes in Dn all valid axioms at stage s for elements n < n, such
that n ∈ G1 [s].
Let t be an α-true stage, such that n ∈ / G1 [t]. If B1 [s] s ⊆ B1 [t] then
DΦ ΦA e
1
[t] or else Case (8) would apply to α. To complete the proof, we need to
show that B1 [s] s ⊆ B1 [t]. Towards a contradiction suppose that B1 [s] s B1 .
Let α be the R1 -strategy that changed B1 and suppose that this happened at a
least stage t0 : s ≤ t0 < t. Then at stage t0 the strategy α acted under Case (8)
and attacked because of some m ∈ / G1 . The axiom for m in Δα was enumerated
at stage s , and α extracted a number larger than s defined after m left G1 . It
follows that s < s and m ∈ / G1 [s]. Note that if n ≤ dα then as we argued above α
will define its list Cα after stage s, and can extract from B1 only γ-markers defined
after stage s, in particular no number from DB . So it follows that dα < n < s. By
Part (3) of Lemma 9 it follows that if n ∈ G1 [t0 ] then n ∈ G1 , hence n ∈ / G1 [t0 ]
and in particular n ≤ m. The axiom for m in Δα is defined before stage s and
hence includes an axiom for n which is valid at stage t0 . As α attacks with the
least possible element it follows that m = n. But this cannot be, as m ∈ / G1 [s] and
n ∈ G1 [s].
To deal with Q strategies we can use the same approach as in the proof of
Theorem 9.
Lemma 13. Let β be a Qe -strategy which is not initialized after stage s0 and
visited infinitely often. Then Qe is satisfied and there is a stage sβ after which β
Proof. We first establish the technical fact about how promises are treated.
We show that if β is a Qe -strategy not initialized in the interval of stages [s, t] and β
makes a promise DX , DG1 , DG2 at stage s, so that at stage t > s, DG1 ⊕ DG2
G1 ⊕ G2 then there is a stage t ∈ (s, t], such that DX X[t ]. Suppose for
concreteness that o ∈ DG1 \ G1 [t]. It follows that o < s. First note that o cannot
be a threshold of an R-strategy as R-strategies of higher priority do not act and
R-strategies of lower priority are initialized at stage s. Thus o is an Ω-marker for
some number n ∈ / X[t ] at some stage t ≥ s. It follows that i(o)[s] = i(o)[t] = 1
as higher priority strategies do not act in this interval of stages and lower priority
strategies are initialized at stage s. If n ∈/ X[s] then by the definition of the Ω-
/ G1 [s]. But DG1 ⊆ G1 [s], hence our assumptions lead to n ∈ X[s]. By
strategy o ∈
the definition of DX it follows that n ∈ DX , thus DX X[t].
The rest of the proof can now proceed is that of Lemma 3.
To complete the proof we use Lemma 12 and Lemma 13 to show that the true
path exists and use it to prove that the Ω-strategy is successful:
Corollary 3. There is an infinite path f in the tree, such that for every n,
f n is visited infinitely often and initialized only finitely often. All R and Q
requirements are satisfied.
Lemma 14. ΩG1 ,G2 (n) = X(n).
Proof. The proof is the same as that of Lemma 8.
References
[CGL16] Mingzhong Cai, Hristo A. Ganchev, Steffen Lempp, Joseph S. Miller, and Mariya I.
Soskova, Defining totality in the enumeration degrees, J. Amer. Math. Soc. 29 (2016),
no. 4, 1051–1067, DOI 10.1090/jams/848. MR3522609
[Coo84] S. B. Cooper, Partial degrees and the density problem. II. The enumeration degrees of the
Σ2 sets are dense, J. Symbolic Logic 49 (1984), no. 2, 503–513, DOI 10.2307/2274181.
MR745377
[Coo90] S. Barry Cooper, Enumeration reducibility, nondeterministic computations and relative
computability of partial functions, Recursion theory week (Oberwolfach, 1989), Lecture
Notes in Math., vol. 1432, Springer, Berlin, 1990, pp. 57–110, DOI 10.1007/BFb0086114.
MR1071514
[Ers75] Yu. Ershov. The uppersemilattice of enumerations of a finite set. Alg. Log., 14:258–284,
1975.
[FR59] Richard M. Friedberg and Hartley Rogers Jr., Reducibility and completeness for sets of
integers, Z. Math. Logik Grundlagen Math. 5 (1959), 117–125. MR0112831
[GS12] Hristo Ganchev and Mariya I. Soskova, Cupping and definability in the local struc-
ture of the enumeration degrees, J. Symbolic Logic 77 (2012), no. 1, 133–158, DOI
10.2178/jsl/1327068696. MR2951634
[GS15] Hristo A. Ganchev and Mariya I. Soskova, Definability via Kalimullin pairs in the struc-
ture of the enumeration degrees, Trans. Amer. Math. Soc. 367 (2015), no. 7, 4873–4893,
DOI 10.1090/S0002-9947-2014-06157-6. MR3335403
[Kal03] I. Sh. Kalimullin, Definability of the jump operator in the enumeration degrees, J. Math.
Log. 3 (2003), no. 2, 257–267, DOI 10.1142/S0219061303000285. MR2030087
[MC85] Kevin McEvoy and S. Barry Cooper, On minimal pairs of enumeration degrees, J. Sym-
bolic Logic 50 (1985), no. 4, 983–1001 (1986), DOI 10.2307/2273985. MR820127
[McE85] Kevin McEvoy, Jumps of quasiminimal enumeration degrees, J. Symbolic Logic 50
(1985), no. 3, 839–848, DOI 10.2307/2274335. MR805690
[Pal75] E. Paliutin. Addendum to the paper of Ershov [1975]. Alg. Log., 14:284–287, 1975.
[Roz78] Mikhael G. Rozinas. The semi-lattice of e-degrees. In Recursive functions, pages 71–84,
Ivanovo, 1978. Ivano. Gos. Univ. Russian.
[Sac63] Gerald E. Sacks, Degrees of unsolvability, Princeton University Press, Princeton, N.J.,
1963. MR0186554
[Sel71] Alan L. Selman, Arithmetical reducibilities. I, Z. Math. Logik Grundlagen Math. 17
(1971), 335–350. MR0304150
[Sho59] J. R. Shoenfield, On degrees of unsolvability, Ann. of Math. (2) 69 (1959), 644–653.
MR0105355
[Sim77] Stephen G. Simpson, First-order theory of the degrees of recursive unsolvability, Ann.
of Math. (2) 105 (1977), no. 1, 121–139. MR0432435
[Sos16] Mariya I. Soskova, The automorphism group of the enumeration degrees, Ann. Pure
Appl. Logic 167 (2016), no. 10, 982–999, DOI 10.1016/j.apal.2014.01.007. MR3522651
[SS] Theodore A. Slaman and Mariya I. Soskova. The Δ02 Turing degrees: Automorphisms
and definability. submitted.
[SW86] Theodore A. Slaman and W. Hugh Woodin, Definability in the Turing degrees, Illinois
J. Math. 30 (1986), no. 2, 320–334. MR840131
[SW97] Theodore A. Slaman and W. Hugh Woodin, Definability in the enumeration degrees,
Arch. Math. Logic 36 (1997), no. 4-5, 255–267, DOI 10.1007/s001530050064. Sacks Sym-
posium (Cambridge, MA, 1993). MR1473024
[SW05] Theodore A. Slaman and W. Hugh Woodin. Definability in degree structures. Preprint,
2005.
Department of Mathematics, University of California, Berkeley, Berkeley, Cali-

fornia 94720-3840
E-mail address, Slaman: slaman@math.berkeley.edu
Faculty of Mathematics and Informatics, Sofia University, 5 James Bourchier

Blvd., 1164 Sofia, Bulgaria
E-mail address: msoskova@fmi.uni-sofia.bg
Volume 690, 2017
Ramsey properties of finite measure algebras and topological

dynamics of the group of measure preserving
automorphisms: Some results and an open problem
A. S. Kechris, M. Sokić, and S. Todorcevic

Dedicated to Hugh Woodin on his 60th birthday
Abstract. We study in this paper ordered finite measure algebras from the
point of view of Fraı̈ssé and Ramsey theory. We also propose an open problem,
which is a homogeneous version of the Dual Ramsey Theorem of Graham-
Rothschild, and derive consequences of a positive answer to the study of the
topological dynamics of the automorphism group of a standard probability
space and also the group of measure preserving homeomorphisms of the Cantor
space.
Contents
1. Introduction
2. Preliminaries
3. Finite ordered measure algebras
4. A dense subgroup of the group of measure
preserving automorphisms
5. Homogeneous measure algebras and the Ramsey Property
6. Applications to extreme amenability and calculation
of the universal minimal flow
7. Reformulation of the Ramsey Property
for homogeneous measure algebras
References
2010 Mathematics Subject Classification. Primary 03C13; Secondary 03C15, 05D10, 37B05,
37A15, 54H20.
Key words and phrases. Structural Ramsey theory, topological dynamics.
The first author was partially supported by NSF grants DMS-0455285, DMS-0968710 and
DMS-1464475.
The third author was partially supported by grants from NSERC and CNRS.
2017
69
70 A. S. KECHRIS, M. SOKIĆ, AND S. TODORCEVIC
1. Introduction
In this paper we continue the theme developed in the paper Kechris-Pestov-
Todorcevic [KPT] of exploring the connections between structural Ramsey the-
ory and topological dynamics of Polish groups, especially the study of extreme
amenability and the calculation of universal minimal flows.
A topological group G is called extremely amenable (or said to have the fixed
point on compacta property) if every continuous action of G on a (non-empty, Haus-
dorff) compact space has a fixed point. There is now a plethora of Polish groups
known to have this very strong fixed point property, e.g., the unitary group of
a separable infinite-dimensional Hilbert space (Gromov-Milman [GM]), the auto-
morphism group of the ordered rationals (Pestov [P1]), the isometry group of the
Urysohn space (Pestov [P2]), the automorphism group of a standard measure space
(Giordano-Pestov [GP]), etc. An excellent exposition of the theory of such groups
and its connections with other areas, such as asymptotic geometric analysis, in par-
ticular concentration of measure phenomena, and Ramsey theory, can be found in
the recent book Pestov [P3]. As pointed out in this monograph there are primar-
ily two main known general techniques for establishing extreme amenability: one
analytical, via the concept of a Lévy group and concentration of measure, and the
other combinatorial, via finite Ramsey theory.
In [KPT] structural Ramsey theory has been used to prove extreme amenabil-
ity of many automorphism groups of countable structures, where concentration of
measure techniques are not directly applicable. It is interesting to investigate to
what extent Ramsey methods can be used to establish extreme amenability for
other types of groups of a more analytic or topological nature and a first step in
that direction was taken in [KPT], where a new proof of the extreme amenabil-
ity of the isometry group of the Urysohn space, Iso(U), was established, a result
originally proved in Pestov [P2] using concentration of measure (see also Pestov
[P3]). An important dividend of this approach is that it motivates the discovery
of new classes that satisfy the Ramsey Property. For example, the Ramsey theory
approach to the extreme amenability of Iso(U) led [KPT] to conjecture that the
class of finite metric spaces has the Ramsey Property, which was indeed established
in Neśetřil [N].
Until now this example is the only one in which combinatorial Ramsey tech-
niques are used to replace analytical arguments in establishing extreme amenability.
We propose in this paper to apply such techniques to another important example
of an extremely amenable group, namely the automorphism group Aut(X, μ) of a
standard probability measure space (X, μ). This result was originally established
in Giordano-Pestov [GP] by concentration of measure arguments. Again our ap-
proach leads us to propose a new Ramsey theorem, a homogeneous version of the
Dual Ramsey Theorem. A proof of this Ramsey theorem would also allow us to
compute the universal minimal flow of another interesting group, namely the group
of measure preserving homeomorphisms of the Cantor space (with the topology it
inherits as a closed subgroup of the homeomorphism group).
We now proceed to describe more precisely the contents of this paper. First we
recall some basic notions that are discussed in detail in Section 2.
A class of finite structures in a given countable language is said to be a Fraı̈ssé
class if it is unbounded (i.e., contains structures of arbitrarily large finite cardinal-
ity), has only countably many members, up to isomorphism, is hereditary (under
RAMSEY PROPERTIES OF FINITE MEASURE ALGEBRAS 71
embeddability) and satisfies the joint embedding and amalgamation properties.

For such a class K there is a unique countably infinite structure, its Fraı̈ssé limit,
Flim(K), which is locally finite (finitely generated substructures are finite), ultraho-
mogeneous (isomorphisms between finite substructures extend to automorphisms)
and whose class of finite substructures, up to isomorphism, coincides with K.
In this paper we study K = OMBAQ2 , the class of naturally ordered finite mea-
sure algebras with measure taking values in the dyadic rationals. These are struc-
tures of the form A = A, ∧, ∨, 0, 1, μ, <, where A, ∧, ∨, 0, 1 is a finite Boolean
algebra, μ : A → [0, 1] a positive measure on A with values in the set Q2 of dyadic
rationals, and < an ordering induced antilexicographically by an ordering of the
atoms. We first show the following result.
Theorem 1.1. The class K = OMBAQ2 is a Fraı̈ssé class.
It is not hard then to compute that the Fraı̈ssé limit of this class is
B ∞ , λ, ≺,
where B ∞ is the algebra of clopen subsets of the Cantor space 2N (i.e., the countable
atomless Boolean algebra), λ is the usual product measure on 2N and ≺ the so-called
canonical ordering on B ∞ , defined by
B ∞ , ≺ ∼
= Flim(OBA),
where OBA is the class of naturally ordered finite Boolean algebras (see [KPT],
6,(D)).
Next recall that a class K has the Ramsey Property (RP) if for any two struc-
tures A ≤ B in K (where A ≤ B means that A can be embedded in B) and
N = 1, 2, . . . , there is a C ∈ K, with B ≤ C, such that for any coloring with N
colors of the set of isomorphic copies of A in C, there is a copy B of B in C so
that all the copies of A contained in B have the same color. We now propose the
following problem:
Problem 1.2. Is it true that the class K = OMBAQ2 has the Ramsey Prop-
erty?
In fact this statement can be viewed as a homogeneous version of the Dual
Ramsey Theorem of Graham-Rothschild [GR] which asserts the following: Given
k < and N there is m > so that for any coloring with N colors of the set of
equivalence relations on {1, . . . , m} with k many classes, there is an equivalence
relation F with many classes, such that any coarser than F equivalence relation
with k many classes has the same color.
It is easy to show that the Dual Ramsey Theorem is equivalent to the assertion
that the class of finite Boolean algebras BA has the Ramsey Property. One can
now see that the class K = OMBAQ2 has the Ramsey Property iff the class of
homogeneous finite measure Boolean algebras with measure taking values in Q2
has the Ramsey Property. Here homogeneous means that all atoms have the same
measure. This in turn translates to the following homogeneous version of the Dual
Ramsey Theorem. Call an equivalence relation homogeneous if all its classes have
the same cardinality. Then we have that for each k < , N , there is m > such that
for any coloring with N colors of the set of homogeneous equivalence relations on
{1, . . . , 2m } with 2k many classes, there is a homogeneous equivalence relation F on
{1, 2, . . . , 2m } with 2 many classes such that all coarser homogeneous equivalence
relations with 2k many classes have the same color.
In fact there is no reason to restrict ourselves to numbers that are powers of
two, except for some obvious restrictions: Let below k mean that k < and k
divides . Then we have the following problem.
Problem 1.3 (Homogeneous Dual Ramsey). Is it true that given N and k ,
there is m % such that for any coloring with N colors of the homogeneous
equivalence relations on {1, . . . , m} with k many classes, there is a homogeneous
equivalence relation F with many classes, so that all homogeneous equivalence
relations coarser than F with k many classes have the same color?
Assuming now a positive answer to Problem 1.2, we can apply the general
theory of [KPT] to give a combinatorial proof of the extreme amenability for
the automorphism group Aut(X, μ) of a standard measure space (X, μ). Clearly
we can take X = 2N , μ = λ. By Theorem 1.1, a positive answer to Problem
1.2 and the results in [KPT], Aut(B ∞ , λ, ≺) is extremely amenable. Now (the
Polish group) Aut(B ∞ , λ, ≺), with the pointwise convergence topology, can be
continuously embedded in a natural way in Aut(2N , λ), where, as usual, Aut(2N , λ)
has the standard (weak) topology.
Theorem 1.4. The image of the natural embedding of Aut(B ∞ , λ, ≺) in the
group Aut(2N , λ) is dense in Aut(2N , λ).
It is clear that if an extremely amenable group G embeds densely in a group H,
then H is also extremely amenable. Thus, assuming a positive answer to Problem
2, one obtains a new, combinatorial proof of the Giordano-Pestov [GP] result that
the automorphism group of a standard measure space, Aut(X, μ), is extremely
amenable.
One can also use Theorem 1.1, a positive answer to Problem 1.2 (plus a bit
more), as well as the results in [KPT], to calculate the universal minimal flow of
another interesting group. Recall that the universal minimal flow of a topological
group G is the (unique up to isomorphism) compact G-flow X (i.e., compact space
with a continuous action of G) which is minimal (orbits are dense) and has the
property that for any minimal G-flow Y there is a continuous epimorphism from
X to Y preserving the actions.
Theorem 1.5. Assuming a positive answer to Problem 1.2, the universal min-
imal flow of the group of measure preserving homeomorphisms of the Cantor space
2N (with the usual product measure λ) is the (canonical action of this group on the)
space of all orderings on the Boolean algebra B ∞ of clopen sets which have the
property that their restrictions to finite subalgebras are natural.
We note here that in [KPT] it is shown that the universal minimal flow of
the group of homeomorphisms of 2N is (its canonical action on) the same space of
orderings.
2. Preliminaries
(A) We will first review some standard concepts concerning Fraı̈ssé classes.
Fix a countable signature L. A Fraı̈ssé class in L is a class of finite structures
in L which contains structures of arbitrarily large (finite) cardinality, contains only
countably many structures, up to isomorphism, and satisfies the following proper-

ties:
(i) (Hereditary Property – HP) If B ∈ K and A ≤ B, then A ∈ K. (Here
A ≤ B means that A can be embedded in B.)
(ii) (Joint Embedding Property – JEP) If A, B ∈ K, then there is C ∈ K with
A ≤ C, B ≤ C.
(iii) (Amalgamation Property – AP) If A, B, C ∈ K and f : A → B, g : A → C
are embeddings, then there is D ∈ K and embeddings r : B → D, s : C → D such
that r ◦ f = s ◦ g.
A structure F in L is called a Fraı̈ssé structure if it is countably infinite, lo-
cally finite (i.e., finite generated substructures are finite) and ultrahomogeneous,
i.e., every isomorphism between finite substructures of F can be extended to an
automorphism of F .
For a Fraı̈ssé structure F , we denote by Age(F ) the class of all finite structures
that can be embedded in F .
A basic theorem of Fraı̈ssé associates to each Fraı̈ssé class K a unique (up to
isomorphism) Fraı̈ssé structure F such that K = Age(F ). This is denoted by
Flim(K)
and called the Fraı̈ssé limit of K. The map K → Flim(K) is a bijection be-
tween Fraı̈ssé classes and Fraı̈ssé structures (up to isomorphism) with inverse F →
Age(F ).
An order Fraı̈ssé class in a signature L ⊇ {<} is a Fraı̈ssé class K such that for
each A ∈ K, <A is a linear ordering.
Let L ⊇ L0 be two signatures. For a structure A in L, A|L0 is its reduct to
L0 . We will also call A an expansion of A|L0 . If K is a class of structures in L we
denote by
K|L0 = {A|L0 : A ∈ K}
the class of reducts to L0 of structures in K. We also call K|L0 the reduct of K to

L0 and call K an expansion of K|L0 to L.
We say that K is reasonable (with respect to (L, L0 )) if for every A0 , B 0 ∈ K|L0 ,
embedding π : A0 → B 0 and expansion A of A0 such that A ∈ K there is an
expansion B of B 0 such that B ∈ K, so that π : A → B is also an embedding.
The proof of Proposition 5.2 in [KPT] can be trivially adapted to show that if
L ⊇ L0 , K is a Fraı̈ssé class in L, K0 = K|L0 and F = Flim(K), F 0 = Flim(K0 ),
then the following are equivalent:
(i) K0 is a Fraı̈ssé class and F 0 = Flim(K0 )
(ii) K is reasonable.
Given a signature L ⊇ {<}, let L0 = L \ {<}. If K is a class of structures in L,
let K0 = K|L0 . We say that K satisfies the ordering property if for every A0 ∈ K0 ,
there is B 0 ∈ K0 such that for every linear ordering ≺ on A0 (the universe of A0 )
and linear ordering ≺ on B0 , if A = A0 , ≺ ∈ K and B = B, ≺ ∈ K, then
A ≤ B.
Finally, for any countable structure A we denote by Aut(A) its automorphism
group. It is a closed subgroup of the group of all permutations of A under the
pointwise convergence topology, thus a Polish group.
(B) We next recall some basic concepts of topological dynamics. A topological

group is always assumed to be Hausdorff and a compact space non-empty and
Hausdorff.
Given a topological group G a G-flow is a continuous action of G on a compact
space X. If no confusion arises, we simply identify X with the flow. A G-flow is
minimal if all orbits are dense. A minimal G-flow X is universal if for every minimal
G-flow Y , there is a continuous surjection π : X → Y preserving the actions. A
basic fact of topological dynamics asserts the existence of a universal minimal flow
for each G, which is unique up to isomorphism (i.e., homeomorphism preserving
the actions). It is denoted by M (G).
We call G extremely amenable if M (G) is a single point. Since every G-flow
contains a minimal subflow, this is equivalent to the statement that every G-flow
has a fixed point.
(C) We now review some concepts of Ramsey theory.
B
Given structures A, B in a signature L with A ≤ B we denote by the
A
set of all substructures of B which are isomorphic to A, i.e.,

B
= {A ⊆ B : A ∼= A},
A
where A ⊆ B means that A is a substructure of B. Given N = 1, 2, . . . and
A ≤ B ≤ C (all structures in L) the notation
C → (B)A N

C C
signifies that for every coloring c : → {1, . . . , N }, there is B ∈ such
A B
B
that c is constant on . We say that a class K of structures in L, closed under
A
isomorphism, satisfies the Ramsey Property (RP) is for every N ≥ 1 and A ≤ B in
K there is C ∈ K, C ≥ B such that C → (B)A N.
3. Finite ordered measure algebras

We denote by BA the class of all finite Boolean algebras (viewed as first order
structures in the language {∧, ∨, 0, 1}). A natural ordering on such a Boolean
algebra is one induced antilexicographically by an ordering of the atoms (see [KPT,
6, (D)]). We denote by OBA the class of all finite Boolean algebras with natural
orderings (in the language {∧, ∨, 0, 1, <}). If Q2 is the ring of dyadic rationals, let
MBAQ2 be the class of all finite measure algebras taking values in Q2 . These can
be viewed as first-order structures in the language {∧, ∨, 0, 1, Mr }r∈Q2 , where Mr
are unary relations with the intended meaning: Mr (a) ⇔ μ(a) = r, where μ is
the measure in the Boolean algebra. Recall that measures in Boolean algebras are
always positive: μ(a) > 0, if a = 0. Finally let
K = OMBAQ2
be the class of all finite naturally ordered Boolean algebras with measure taking
values in Q2 . Thus every A ∈ K is of the form
A = A, ∧, ∨, 0, 1, <, μ,
where A, ∧, ∨, 0, 1, < is a naturally ordered Boolean algebra and μ : A → [0, 1] is

a measure on A with μ(a) ∈ Q2 . (Again these can be viewed as structures in the
language L = {∧, ∨, 0, 1, <, Mr }r∈Q2 .) Let L0 = L \ {<}.
Theorem 3.1. The class K = OMBAQ2 is a reasonable, with respect to

(L, L0 ), Fraı̈ssé order class.
Proof. The proof that K is reasonable is as in [KPT, 6.13] and we will not
repeat it here.
Since MBAQ2 and OBA are hereditary (for the second, see [KPT, 6.13]), so
is K.
JEP follows from AP by considering the 2-element Boolean algebra. Finally
we verify that K satisfies AP.
Fix B ∈ K, say with atoms
b1 <B · · · <B bn
(where <B is the ordering in B−the case n = 1 corresponds to the 2-element
Boolean algebra). Fix also embeddings f : B → C ∈ K, g : B → D ∈ K. We will
find E ∈ K and embeddings r : C → E, s : D → E which satisfy r ◦ f = s ◦ g.
The embedding f sends bi to (the join of) a set of atoms Ci of C (i.e., f (bi ) =
Ci ) and C1 , . . . , Cn is a partition of the atoms of C. Similarly g sends bi to Di
and D1 , . . . , Dn is a partition of the atoms of D. Take E to be the Boolean algebra
with set of atoms:

n
(Ci × Di ).
i=1
Below we will write c ⊗ d instead of (c, d), where c is an atom of C and d is an
atom of D. We define r : C → E by
r(c) = c ⊗ Di , if c ∈ Ci ,

(where we literally mean r(c) = c ⊗ Di = {c ⊗ d : d ∈ Di }) and s : D → E by
s(d) = Ci ⊗ d, if d ∈ Di .
Clearly

r(f (bi )) = r( Ci ) = (Ci ⊗ Di ),
s(g(bi )) = s( Di ) = (Ci ⊗ Di ),
so r ◦ f = s ◦ g.
We now need to define the measure and the ordering of E.
(1) Measure
If for c ⊗ d ∈ Ci × Di we put
μC (c)μD (d)
μ (c ⊗ d) = ,
μB (bi )
this would work, except that μ may not take values in Q2 . We therefore need to
be more careful.
Fix 1 ≤ k ≤ n and put b = bk , P = Ck , Q = Dk . Say P has atoms c1 , . . . , cp and
Q has atoms d1 , . . . , dq . We need to define μE (ci ⊗ dj ) = xij (1 ≤ i ≤ p, 1 ≤ j ≤ q).
They have to satisfy:
0 < xij , xij ∈ Q2 ,
def
q
xij = ai = μC (ci ), 1 ≤ i ≤ p,
(∗) j=1
p def
i=1 xij = bj = μD (dj ), 1 ≤ j ≤ q.
p q
Note here that i=1 ai = j=1 bj = μB (b).
First assume p, q ≥ 2. Then (∗) can be rewritten as
⎧
⎪
⎪xi1 = ai − qj=2 xij , 1 < i ≤ p,
⎪
⎨x = b − p x , 1 < j ≤ q,
1j j ij
(∗∗) i=2
⎪
⎪
q
x11 = a1 − j=2 x1j ,
⎪
⎩
x11 = b1 − pi=2 xi1 ,
and the last equation is redundant as
q q p
a1 − j=2 xij = a1 − j=2 1(bj − i=2 xij )
q q p
= a1 − j=2 bj + j=2 i=2 xij ,
p p q
b1 − i=2 xi1 = b1 − i=2 (ai − j=2 xij )
p p q
= b1 − i=2 ai + i=2 j=2 xij

= a1 − qj=2 bj + qj=2 pi=2 xij ,
so the system is equivalent to the system (∗ ∗ ∗) consisting of the first 3 rows of
(∗∗), which express xk1 , x1l , for 1 ≤ k ≤ p, 1 ≤ l ≤ q, in terms of xij , i ≥ 2, j ≥ 2.
Since ai , bj ∈ Q2 , if all xij , i ≥ 2, j ≥ 2, are in Q2 , so are all xij satisfying (∗ ∗ ∗).
This system has a solution in positive numbers in Q, namely
x∗ij = μ (ci ⊗ dj ).
It is clear then that if we choose positive xij ∈ Q2 , i ≥ 2, j ≥ 2, very close to x∗ij
and determine xi1 , x1j using (∗ ∗ ∗), then all xij will be positive and in Q2 . So the
system (∗) admits a positive solution in Q2 , say xij , and we define
μE (ci ⊗ dj ) = xij .
This clearly works, i.e., r, s are also measure preserving.
If now p = q = 1, we can simply take μE (ci ⊗ dj ) = μE (c1 ⊗ d1 ) = μ (c1 ⊗ d1 ) =
μC (c1 )μD (d1 )
μB (b) ∈ Q2 , as μB (b) = μC (c1 ) = μD (d1 ). This definition also works if say
μC (c1 )μD (dj )
p = 1, q = 2, since then μC (c1 ) = μB (b), so μE (c1 ⊗ dj ) = μB (b) ∈ Q2 .
(2) Order
Let the atoms of B be b1 <B · · · <B bn , the atoms of C, γ1 <C · · · <C γm ,
and of D, δ1 <D · · · <D δk . We will define the order <E of the atoms of E. Denote
by γi,1 the <C -largest element of Ci and similarly define δi,1 for Di . We have
C1 < C2 · · · <C Cn

(which abbreviates C1 <C C2 < · · · <C Cn ) and thus
γi,1 <C · · · <C γn,1 ,
and similarly
D1 <D · · · <D Dn ,
therefore
δi,1 <D · · · <D δn,1 .
(In particular, γn,1 = γm , δn,1 = δk .)
Our goal is to find an order ≺ for all the atoms of E of the form
γ ⊗ δi,1 , γ ∈ Ci ,
γi,1 ⊗ δ, δ ∈ Di ,
so that
(i) γ ⊗ δi,1 ≺ γ ⊗ δi ,1 ⇔ γ <C γ ,
(ii) γi,1 ⊗ δ ≺ γi ,1 ⊗ δ ⇔ δ <D δ .
Then we let <E agree with ≺ on these atoms and let all the other atoms of
E be smaller in <E than these atoms and otherwise ordered arbitrarily. It is clear
then that r, s are also order preserving and completes the proof.
To find ≺ consider
X = {γ1,1 ⊗ δ1,1 , γ2,1 ⊗ δ2,1 , . . . , γn,1 ⊗ δn,1 }
and put
γi,1 ⊗ δi,1 <X γi ,1 ⊗ δi ,1 ⇔ i < i
(⇔ γi,1 <C γi ,1 ⇔ δi,1 <D δi ,1 ).
Let also
Y = {γ ⊗ δi,1 : γ ∈ Ci , 1 ≤ i ≤ n}
and put
γ ⊗ δi,1 <Y γ ⊗ δi ,1 ⇔ γ <C γ .
Finally, let
Z = {γi,1 ⊗ δ : δ ∈ Di , 1 ≤ i ≤ n},
and put
γi,1 ⊗ δ <Z γi ,1 ⊗ δ ⇔ δ <D δ .
Then X = Y ∩ Z, <X =<Y |X =<Z |X. So by the amalgamation property for
linear orderings, there is an order ≺ on Y ∪ Z such that ≺ |Y =<Y , ≺ |Z =<Z and
we are done.
If K0 = MBAQ2 , then, as verified in Kechris-Rosendal [KR], MBAQ2 is a
Fraı̈ssé class with Fraı̈ssé limit
Flim(K0 ) ∼
= B ∞ , λ,
where B ∞ = the algebra of clopen sets in 2N and λ the usual product measure.
Since K is a reasonable expansion of K0 , we have, by [KPT, 5.2],
Flim(K) = B ∞ , λ, ≺
for some appropriate ordering ≺. We will next identify ≺.
If K = OBA, then K is a reasonable order Fraı̈ssé class, with respect to
(L , L0 ), where L = {∧, ∨, 0, 1, <}, L0 = L \ {<}) (see [KPT, 6, (D)]) and

Flim(K ) = B ∞ , ≺ ,
with ≺ the canonical ordering on B ∞ (see [KPT], 6,(D)). We will verify that
≺=≺ . To see this note that K = OMBAQ2 is reasonable with respect to (L, L ),
where L = L \ {Mr }r∈Q2 . This means that for any A0 , B 0 ∈ K , embedding
π : A0 → B 0 and measure μ on A0 , there is a measure μ on B 0 so that π :
A0 , μ) → B 0 , μ is also an embedding. (Measures here take values in Q2 .) To
see this, let a1 , . . . , ak be the atoms of A0 and let the set of atoms in π(ai ) be
Bi , so that B1 , . . . , Bk is a partition of the atoms of B 0 . Say μ(ai ) = xi , so that
k
x1 + · · · + xk = 1. Since xi ∈ Q2 , write xi = m 2n , so that
i n
i=1 mi = 2 . Say Bi has
i
ki atoms, bi1 , . . . , biki . We need to define μ (bij ) so that kj=1 μ (bij ) = xi . We can
clearly assume that mi ≥ ki . Take then μ (bi1 ) = · · · = μ (bi(ki −1) ) = 21n , μ (biki ) =
mi −(ki −1)
2n > 0. Thus K is a reasonable expansion of K , so we have
Theorem 3.2. The Fraı̈ssé limit of K = OMBAQ2 is
Flim(K) = B ∞ , λ, ≺,
where B ∞ is the algebra of clopen sets of 2N , λ the usual product measure, and ≺
the canonical ordering on B ∞ defined by:
Flim(OBA) ∼ = B ∞ , ≺.
4. A dense subgroup of the group of measure

preserving automorphisms
Clearly we can view an element of Aut(B ∞ , λ, ≺) as a measure preserving
homeomorphism of 2N , which has the additional property that when viewed as
an automorphism of B ∞ it also preserves the canonical ordering ≺. As usual we
denote by Aut(2N , λ) the Polish group of measure-preserving automorphisms of
(2N , λ) with the weak topology. Thus there is a canonical continuous embedding of
the group Aut(B ∞ , λ, ≺) into Aut(2N , λ).
Theorem 4.1. The image of the canonical embedding of Aut(B ∞ , λ, ≺) is

dense in Aut(2N , λ).
Proof. Since (see Halmos [H]) the dyadic permutations, i.e., the maps of the
form
sˆx → α(s)ˆx,
where α ∈ S2n = the symmetric group of 2n (viewed as the set of all binary
sequences of length n), are dense in Aut(2N , λ), it is enough to show that for
every such dyadic permutation (also denoted by) α and every nbhd V of α in
Aut(2N , λ), V ∩Aut(B ∞ , λ, ≺) = ∅ (where we identify Aut(B ∞ , λ, ≺) with its image
here). We can assume that V has the following form:
For some m > n, > 0, V consists of all T ∈ Aut(2N , λ) such that
λ(T (Nt )Δα(Nt )) < , ∀t ∈ 2m ,
where
Nt = {x ∈ 2N : t ⊆ x}.
Note that α(Nt ) = Nα(t) , for some α ∈ S2m . So it is enough to show that for any
given m > 0, > 0, β ∈ S2m , there is T ∈ Aut(B ∞ , λ, ≺) such that
λ(T (Nt )ΔNβ(t) ) < , ∀t ∈ 2m .
m
First choose M > 2m, with 2 2M+1 < . Next let A be the substructure of
B ∞ , λ, ≺, whose underlying Boolean algebra A0 has atoms {Nt : t ∈ 2m }. Con-
sider also the substructure C of B ∞ , λ, ≺, whose underlying Boolean algebra C 0
has atoms {NS : S ∈ 2M }.
For t ∈ 2m , let
Nt∗ = {NS : S ∈ 2M , S ⊇ t}.
∗
Clearly Nt = Nt and {Nt∗ }t∈2m is a partition of the atoms of C 0 . We will find
a new partition
t }t∈2 of the atoms of C 0 such that
{N m
(i) λ( Nt∗ Δ Nt ) < , ∀t ∈ 2m ,


(ii) λ( Nt) = λ( Nt∗ ) = 2−m , ∀t ∈ 2m ,
(iii) Nt → Nβ(t) , t ∈ 2m , preserves ≺ i.e.,

Nt ≺ Nt ⇒ Nβ(t) ≺ Nβ(t ) .

Let thenC 0 be the Boolean algebra with atoms { Nt }t∈2m . Consider the
map Nt → Nβ(t) from the atoms of A0 to the atoms of C 0 . Then (ii), (iii)
above imply that this map extends to an isomorphism of A with C 0 , λ, ≺. By
the ultrahomogeneity
of B ∞ , λ, ≺ this extends to T ∈ Aut(B ∞ , λ, ≺). Then
T (Nt ) = Nβ(t) , t ∈ 2m , so

∗
λ(T (Nt )ΔNβ(t) ) = λ( Nβ(t) Δ Nβ(t) ) < ,
and we are done.
Construction of {Nt }t∈2m :
Say Nt1 ≺ Nt2 ≺ · · · ≺ Nt2m . Let a1 ≺ a2 ≺ · · · ≺ a2m be the last 2m atoms of

C 0 in the order ≺. Let
ki = |Nt∗i ∩ {a1 , . . . , a2m }|,
so that 0 ≤ ki ≤ 2m . Put
Nt = Nt∗ \ {a1 , . . . , a2m },
so that |Nt | = |Nt∗ | − ki . Note that |Nt∗ | = 2M −m > 2m ≥ ki , thus |Nt | > 0. Let

Nβ(ti)
= Nβ(ti)
∪ {ai },

so ai is the ≺-last element of Nβ(ti)
, therefore

i<j⇒ Nβ(ti)
≺ Nβ(tj)
,
i.e.,

Nt ≺ Nt ⇒ Nβ(t)
Nβ(t≺).

However |Nt | may not equal 2M −m , thus λ( Nt ) may not equal λ( Nt∗ ).
Note though that the {Nt }t∈2m are pairwise disjoint and

m
2
|Nti \ {aj : j ≤ 2m }| = 2M − 2m ,
i=1
so we can redistribute the 2M − 2m atoms of C 0 , which are not in {a1 , . . . , a2m },

among the Nti to define Nti so that |Nti | = 2M −m , ai is still the ≺-last element of
Nβ(ti ) , and in this redistribution we do not move more than 2m elements in or out
of each Nti . Thus
(2m + 1)
λ( Nt Δ Nt∗ ) ≤ < ,
2M
and we are done.
5. Homogeneous measure algebras and the Ramsey Property

We call A ∈ OMBAQ2 homogeneous if every atom has the same measure. Thus
a homogeneous A has 2n atoms, for some n, and each atom has measure 2−n . We
denote by H the subclass of OMBAQ2 consisting of the homogeneous structures.
Proposition 5.1. The class H is cofinal under embeddability in OMBAQ2 .
Proof. Let B ∈ OMBAQ2 , with atoms b1 < · · · < bn and say μ(Bi ) =
mi n N N
2N (so that i=1 mi = 2 ). Consider the Boolean algebra A0 with 2 atoms
bi,1 , . . . , bi,m1 ; b2,1 , . . . , b2,m2 ; . . . ; bn,1 , . . . , bn,mn . Assign to each measure equal to
−N
2 . Also order them as follows:
bi,1 < · · · < b1,m1 < b2,1 < · · · < b2,m2 < . . .
< bn,1 < · · · < bn,mn .
Call A ∈ OMBAQ2 the resulting structure.
Clearly A ∈ H. The map

mi
bi → bi,j
j=1
is an embedding of B into A.
Call also A ∈ MBAQ2 homogeneous if every atom has the same measure and
denote by H0 the class of such measure algebras. If A, B ∈ H and A0 , B 0 ∈ H0
are their reducts (in which we drop the ordering), then
= B ⇔ A0 ∼
A∼ = B0.
(Note that this fails for the class OMBAQ2 itself.) Using this it is easy to check
that the following holds:
H has the Ramsey Property ⇔ H0 has the Ramsey Property.
We now propose the following problem:
Problem 5.2. Is it true that the class H0 of finite homogeneous measure alge-
bras with measure taking values in the dyadic rationals has the Ramsey Property?
Next we use a positive answer to Problem 5.2 to derive the Ramsey Property
for OMBAQ2 . For that we need the following general result.
Proposition 5.3. Let K be an order Fraı̈ssé class and K∗ ⊆ K a cofinal (under
embeddability) subclass closed under isomorphism. If K∗ has the Ramsey Property,
so does K.
Proof. It should be kept in mind below that, since K is an order Fraı̈ssé class,
if C ∼
= D are in K, then there is a unique isomorphism between C, D.
Fix A ≤ B in K. Let A∗ ∈ K∗ be such that A ≤ A∗ . For each A∗0 ∼ = A∗ , let
∗ ∗
(A0 ) be the isomorphic copy of A in A0 , whose universe is lexicographically least
in the ordering of A∗0 .
Lemma 5.4. We can find B ∗ ∈ K∗ with A∗ ≤ B ∗ such that there is a copy
B0 ∼= B with B 0 ⊆ B ∗ with the property that for any copy A0 ∼
= A, A0 ⊆ B 0 ,
there is a copy A∗0 ∼
= A∗ , A∗0 ⊆ B ∗ with (A∗0 ) = A0 .
Proof. Let e1 : A → B, . . . , en : A → B be all the embeddings of A into

B. Using amalgamation, we will construct recursively B 1 , . . . , B n ∈ K containing,
resp., copies B 01 , . . . , B 0n of B such that for 1 ≤ i ≤ n, if A01 , . . . , A0i are the copies
of A in B 0i corresponding to the embeddings e1 , . . . , ei , then there are A∗1 , . . . , A∗i ,
copies of A∗ in B i , for which (A∗j ) = A0j , ∀j ≤ i. Then B n has the property that
it contains a copy B 0n = B 0 of B such that for any copy A0 of A in B 0 , there is a
copy A∗0 of A∗ in B n with (A∗0 ) = A0 . Extend then B n to B ∗ ∈ K∗ to complete
the proof.
We first construct B 1 by amalgamating e1 : A → B and f1 : A → A∗ ,
whose range f1 (A) is such that (A∗ ) = f1 (A). Let B 01 be the image of B in this
amalgamation. Assume now that B i has been constructed. Let ēi+1 : A → B 0i
be the embedding of A into B 0i that corresponds to ei+1 . Since B 0i ⊆ B i , view
ēi+1 : A → B i as an embedding of A into B i . Let B i+1 be constructed by
amalgamating ēi+1 and f1 : A → A∗ and let B 0i+1 be the image of B 0i under this
amalgamation.
Using this lemma and the Ramsey Property for K∗ , given any k, we can find
∗
C ∗ ∈ K∗ such that A∗ ≤ B ∗ ≤ C ∗ and C ∗ → (B ∗ )A k . We will show that
∗
C → (B)A k . Indeed, fix a coloring
∗
C
c: → {1, . . . , k}.
A
Define then the coloring ∗
∗ C
c : → {1, . . . , k}
A∗
by
c∗ (A∗0 ) = c((A∗0 )).
Let B ∗0∼ ∗ ∗ ∗
= B , B 0 ⊆ C be homogeneous for c∗ . By the lemma there is a copy B 0
of B, B 0 ⊆ B ∗0 such that for any copy A0 of A in B 0 there is a copy A∗0 of A∗ in
B ∗0 with (A∗0 ) = A0 . Then c(A0 ) = c∗ (A∗0 ), so that B 0 is homogeneous for c.
Theorem 5.5. If Problem 5.2 has a positive answer, then the class K =
OMBAQ2 has the Ramsey Property.
Proof. By the positive answer to Problem 5.2 and the remarks preceding it,
H has the Ramsey Property. The result then follows from Propositions 5.1 and
5.3.
We note here that MBAQ2 itself does not have the Ramsey Property. (Since
H0 is cofinal in MBAQ2 and has the Ramsey Property, it follows that the use of
orderings is crucial in Proposition 5.3.) To see this, let A0 be the measure Boolean
algebra with two atoms a, b such that μ(a) = 14 , μ(b) = 34 . Let B 0 be the measure
Boolean algebra with 4 atoms of measure 14 . Suppose, towards a contradiction, that
C 0 is a measure algebra with
C 0 → (B 0 )A
2 .
0
Take a natural order ≺ on C 0 and color all copies A0 of A0 in C 0 as follows:

If a , b are the two atoms of A0 (corresponding to a, b), put

0, if a ≺ b ,
c(A0 ) =
1, if a ( b .
We claim that there is no copy of B 0 that verifies C 0 → (B 0 )A 2 . Indeed assume

0
B 0 was one, say with atoms a1 ≺ a2 ≺ a3 ≺ a4 . Let A0 be the copy of A0 in B 0

with atoms a = a1 , b = a2 ∨ a3 ∨ a4 . Then, as a ≺ b , we have c(A0 ) = 0. Next let

A0 be the copy of A0 in B 0 with atoms a = a4 , b = a1 ∨ a2 ∨ a3 . Then a ( b ,
so c(A0 ) = 1.
Finally, using the general result in [KPT, 10.3], a positive answer to Problem
5.2 and Proposition 6.3 below, one can compute the Ramsey degree (see [KPT, 10]
for the definition) of each A0 ∈ MBAQ2 in the class MBAQ2 : It is equal to
|Aut(A0 )| , where n is the number of atoms in A0 . If 0 < λ1 < · · · < λm < 1, where
n!
1 ≤ m ≤ n, are the values of the measure of A0 at its atoms, and the value λi is
obtained by ni atoms, so that m i=1 ni = n, clearly |Aut(A0 )| = n1 !n2 ! . . . nm !, so
the Ramsey degree is (n 1 +···+nm )!
n1 !n2 !...nm ! .
6. Applications to extreme amenability and calculation

of the universal minimal flow
If Problem 5.2 has a positive answer, then K = OMBAQ2 has the Ramsey
Property and its Fraı̈ssé limit is B ∞ , λ, ≺, so we immediately obtain, using [KPT,
6.1], the following result.
Theorem 6.1. If Problem 5.2 has a positive answer, the group Aut(B ∞ , λ, ≺)
is extremely amenable.
Using Theorem 6.1 and Theorem 4.1, this would then give a new proof of the
following result of Giordano-Pestov [GP], a proof that has the feature that it uses
Ramsey theory as opposed to the original proof that used concentration of measure
techniques.
Theorem 6.2 (Giordano-Pestov [GP]). The automorphism group Aut(X, μ)
of a standard probability measure space is extremely amenable.
Our next application conditionally computes the universal minimal flow of the
group of measure preserving homeomorphisms of 2N . First we note the following
fact.
Proposition 6.3. The class K = OMBAQ2 has the ordering property.
Proof. Let A0 ∈ MBAQ2 with corresponding measure μ. Let a1 , . . . , an be
the atoms of A0 with μ(ai ) = 2mNi , 1 ≤ mi ≤ 2N , so that ni=1 mi = 2N . Let
B 0 ∈ MBAQ2 be the homogeneous algebra with 2N atoms. We claim that for
any natural orderings ≺, ≺ of A0 , B 0 , resp., there is an embedding of A0 , ≺ into
B 0 , ≺ . Say aj1 ≺ · · · ≺ ajn . Since n ≤ 2N , let b1 ≺ · · · ≺ bn be the last n atoms
of B 0 in the ordering ≺ . Let X1 , . . . , Xn be a partition of the atoms of B 0 such
that bi ∈ Xi and |Xi | = mji . Then aji → Xi is an embedding of A0 , ≺ into
B 0 , ≺ .
We view the group of measure preserving homeomorphisms of 2N as a closed

subgroup of the Polish group of homeomorphisms of 2N . It can therefore be identi-
fied with the automorphism group Aut(B ∞ , λ). Recall that B ∞ , λ is the Fraı̈ssé
limit of K = MBAQ2 . It follows then from [KPT, 7.5] and the preceding results
that we can compute the universal minimal flow of Aut(B ∞ , λ) as follows:
Theorem 6.4. If Problem 5.2 has a positive answer, then the universal minimal
flow of the group of measure preserving homeomorphisms of 2N , Aut(B ∞ , λ), is (its
canonical action on) the space of orderings on B ∞ , whose restrictions to finite
subalgebras of B ∞ are natural.
From this and [KPT, 8.2], we see that if Problem 5.2 has a positive answer,
then the groups of homeomorphisms of 2N and measure preserving homeomorphisms
of 2N have the same underlying compact metric space as their universal minimal
flow (and the action of the second group is the restriction of the action of the first
group).
7. Reformulation of the Ramsey Property

for homogeneous measure algebras
We start with a few remarks. Recall first the Dual Ramsey Theorem.
Theorem 7.1 (The Dual Ramsey Theorem, Graham-Rothschild [GR]). For
any positive integers N, k < , there is a positive integer m > with the following
property: For every coloring c : EQk → {1, . . . , N } of the set of all equivalence
relations on {1, . . . , m} which have exactly k classes, there is an equivalence relation
E on {1, . . . , m} with exactly classes such that c is constant on the set of all
equivalence relations F ⊇ E with exactly k classes.
Identifying a finite set of size m with the Boolean algebra with m atoms, B m ,
we note that an equivalence relation E of {1, . . . , m} with k classes corresponds
exactly to an isomorphic copy of B k in B m (the copy having as atoms the (joins
of the) equivalence classes of E). Thus the Dual Ramsey Theorem is exactly the
Ramsey Property for the class BA of finite Boolean algebras.
Theorem 7.2 (The Dual Ramsey Theorem). The class BA of finite Boolean
algebras has the Ramsey Property.
Keeping this interpretation in mind, one can reformulate (a somewhat stronger
version) of Problem 5.2 as a Homogeneous Dual Ramsey Theorem. Recall that an
equivalence relation is homogeneous if all its classes have the same size. Below we
let
k << ⇔ k < and k divides
Problem 7.3 (Homogeneous Dual Ramsey). Is it true that for any positive
integers N, k << , there is a positive integer m >> such that for every coloring
c : EQhom k → {1, . . . , N } of the set of all homogeneous equivalence relations on
{1, . . . m} which have exactly k classes, there is a homogeneous equivalence rela-
tion E on {1, . . . , m} with exactly classes so that c is constant on the set of all
homogeneous equivalence relations F ⊇ E with exactly k classes?
Indeed, translated to Boolean algebras, this is simply the Ramsey Property for
the class of all homogeneous measure Boolean algebras (with arbitrary measure -
not necessarily in Q2 ).
We finally state another reformulation of Problem 7.3. For convenience we will
use below the following notation and terminology:
We use k, , m, . . . for positive integers. A k-algebra is a finite Boolean algebra
with exactly k atoms. A homogeneous k-subalgebra of a finite Boolean algebra B
is a Boolean subalgebra A of B with k atoms, so that each atom of A contains
exactly the same number of atoms of B. Given a finite Boolean algebra B, we

denote by

B
k hom
the set of all homogeneous k-subalgebras of B. We then have the following refor-
mulation of Problem 7.3.
Problem 7.4 (Homogeneous Dual Ramsey). Is it true that for every positive
integers
N, k << , there is m >> suchthat for every m-algebra C andcoloring

C C B
c: → {1, . . . , N }, there is B ∈ so that c is constant on
k hom hom k hom
(and moreover, if k, are powers of 2, m can be chosen to be a power of 2)?
Remark 7.5. Tony Wang has verified, using Mathematica, that Problem 7.3
has a positive answer for N = 2, k = 2, l = 4 if we take m = 8.
References
[GP] T. Giordano and V. Pestov, Some extremely amenable groups related to operator al-
gebras and ergodic theory, J. Inst. Math. Jussieu 6 (2007), no. 2, 279–315, DOI
10.1017/S1474748006000090. MR2311665
[GR] R. L. Graham and B. L. Rothschild, Ramsey’s theorem for n-parameter sets, Trans. Amer.
Math. Soc. 159 (1971), 257–292. MR0284352
[GRS] R. L. Graham, B. L. Rothschild, and J. H. Spencer, Ramsey theory, 2nd ed., Wiley-
Interscience Series in Discrete Mathematics and Optimization, John Wiley & Sons, Inc.,
New York, 1990. A Wiley-Interscience Publication. MR1044995
[GM] M. Gromov and V. D. Milman, A topological application of the isoperimetric inequality,
Amer. J. Math. 105 (1983), no. 4, 843–854, DOI 10.2307/2374298. MR708367
[H] P. R. Halmos, Lectures on ergodic theory, Chelsea Publishing Co., New York, 1960.
MR0111817
[KPT] A. S. Kechris, V. G. Pestov, and S. Todorcevic, Fraı̈ssé limits, Ramsey theory, and topo-
logical dynamics of automorphism groups, Geom. Funct. Anal. 15 (2005), no. 1, 106–189,
DOI 10.1007/s00039-005-0503-1. MR2140630
[KR] A. S. Kechris and C. Rosendal, Turbulence, amalgamation, and generic automorphisms
of homogeneous structures, Proc. Lond. Math. Soc. (3) 94 (2007), no. 2, 302–350, DOI
10.1112/plms/pdl007. MR2308230
[N] J. Nešetřil, Metric spaces are Ramsey, European J. Combin. 28 (2007), no. 1, 457–468,
DOI 10.1016/j.ejc.2004.11.003. MR2261831
[P1] V. G. Pestov, On free actions, minimal flows, and a problem by Ellis, Trans. Amer. Math.
Soc. 350 (1998), no. 10, 4149–4165, DOI 10.1090/S0002-9947-98-02329-0. MR1608494
[P2] V. Pestov, Ramsey-Milman phenomenon, Urysohn metric spaces, and extremely amenable
groups, Israel J. Math. 127 (2002), 317–357, DOI 10.1007/BF02784537. MR1900705
[P3] V. Pestov, Dynamics of infinite-dimensional groups, University Lecture Series, vol. 40,
American Mathematical Society, Providence, RI, 2006. The Ramsey-Dvoretzky-Milman
phenomenon; Revised edition of Dynamics of infinite-dimensional groups and Ramsey-type
phenomena [Inst. Mat. Pura. Apl. (IMPA), Rio de Janeiro, 2005; MR2164572]. MR2277969
Department of Mathematics, California Institute of Technology, Mail Code 253-

37, Pasadena, California 91125
E-mail address: kechris@caltech.edu
Department of Mathematics and Statistics, York University, N520 Ross, 4700 Keele
Street, Toronto, Ontario M3J 1P3, Canada
E-mail address: miosokic@gmail.com
Department of Mathematics, University of Toronto, Toronto, Ontario M5S 2E4,

Canada and Institut de Mathématiques de Jussieu, CNRS UMR 7586, case 247, 4, Place
Jussieu, 75252 Paris Cedex, France
E-mail address: stevo@math.toronto.edu
E-mail address: stevo@math.jussieu.fr
Volume 690, 2017
Topological Ramsey numbers and countable ordinals
Andrés Eduardo Caicedo and Jacob Hilton

To W. Hugh Woodin, on the occasion of his birthday
Abstract. We study the topological version of the partition calculus in the

setting of countable ordinals. Let α and β be ordinals and let k be a positive
integer. We write β →top (α, k)2 to mean that, for every red-blue coloring
of the collection of 2-sized subsets of β, there is either a red-homogeneous set
homeomorphic to α or a blue-homogeneous set of size k. The least such β is
the topological Ramsey number Rtop (α, k).
We prove a topological version of the Erdős-Milner theorem, namely that
Rtop (α, k) is countable whenever α is countable. More precisely, we prove that
β β·k
Rtop (ω ω , k + 1) ≤ ω ω for all countable ordinals β and finite k. Our proof
is modeled on a new easy proof of a weak version of the Erdős-Milner theorem
that may be of independent interest.
We also provide more careful upper bounds for certain small values of α,
proving among other results that Rtop (ω + 1, k + 1) = ω k + 1, Rtop (α, k) < ω ω
whenever α < ω 2 , Rtop (ω 2 , k) ≤ ω ω and Rtop (ω 2 + 1, k + 2) ≤ ω ω·k + 1 for all
finite k.
Our computations use a variety of techniques, including a topological
pigeonhole principle for ordinals, considerations of a tree ordering based on
the Cantor normal form of ordinals, and some ultrafilter arguments.
Contents
1. Introduction
2. Preliminaries
3. The closed pigeonhole principle for ordinals
4. The ordinal ω + 1
5. Stepping up by one
6. Ordinals less than ω 2
7. The ordinal ω 2
8. The anti-tree partial ordering on ordinals
9. The ordinal ω 2 + 1
10. The weak topological Erdős-Milner theorem
11. Questions
References
2010 Mathematics Subject Classification. Primary 03E02; Secondary 03E10, 54A25.

Key words and phrases. Partition calculus, countable ordinals.
The second author’s research was conducted under the supervision of John K. Truss with the
support of an EPSRC Doctoral Training Grant Studentship.
2017
87
88 ANDRÉS EDUARDO CAICEDO AND JACOB HILTON
1. Introduction
Motivated by Ramsey’s theorem, the partition calculus for cardinals and Rado’s
arrow notation were introduced by Erdős and Rado in [ER53]. The version for
ordinals, where the homogeneous set must have the correct order type, first appears
in their seminal paper [ER56]. In the following definition, [X]n denotes the set of
subsets of X of size n.
Definition 1.1. Let κ be a cardinal, let n be a positive integer, and let β and
all αi be ordinals for i ∈ κ. We write
β → (αi )ni∈κ
to mean that for every function c : [β]n → κ (a coloring) there exists some subset
X ⊆ β and some i ∈ κ such that X is an i-homogeneous copy of αi , i.e., [X]n ⊆
c−1 ({i}) and X is order-isomorphic to αi .
This relation has been extensively studied, particularly the case n = 2, where
we define the (classical ) ordinal Ramsey number R(αi )i∈κ to be the least ordinal
β such that β → (αi )2i∈κ . Ramsey’s theorem can be stated using this notation as
R(ω, ω) = ω.
A simple argument that goes back to Sierpiński shows that β → (ω + 1, ω)2
for every countable ordinal β [Sie33], which means that if α > ω and R(α, γ)
is countable, then γ must be finite (see also [ER56, Theorem 19] and [Spe56,
Theorem 4]). On the other hand, Erdős and Milner showed that indeed R(α, k)
is countable whenever α is countable and k is finite [EM72]. Much work has
been done to compute these countable ordinal Ramsey numbers. In particular, as
announced without proof by Haddad and Sabbagh [HS69a, HS69b, HS69c], there
are algorithms for computing R(α, k) for several classes of ordinals α < ω ω and all
finite k; details are given in [Cai15] for the case α < ω 2 and in [Mil71] for the case
α = ω m for finite m. See also [Wil77, Chapter 7], [HL10], [Sch10] and [Wei14].
With the development of structural Ramsey theory, many variants of the basic
definition above have been introduced. In this article we are concerned with a
“topological” version and a closely-related “closed” version. (Jean Larson suggested
to call the second version “limit closed”.) Both of these make use of the topological
structure of an ordinal, which we endow with the order topology. Although the
second version may appear more natural, it is the first that has been considered
historically, since it can be defined for arbitrary topological spaces; the second
version additionally requires an order structure.
In the following definition, ∼ = denotes the homeomorphism relation. Follow-
ing Baumgartner [Bau86], we also say that a subspace X of an ordinal is order-
homeomorphic to an ordinal α to mean that there is a bijection X → α that
is both an order-isomorphism and a homeomorphism, or equivalently, X is both
order-isomorphic to α and closed in its supremum.
Definition 1.2. Let κ be a cardinal, let n be a positive integer, and let β and
all αi be ordinals for i ∈ κ.
We write
β →top (αi )ni∈κ
to mean that for every function c : [β]n → κ there exists some subspace X ⊆ β
and some i ∈ κ such that X is an i-homogeneous topological copy of αi , i.e., [X]n ⊆
c−1 ({i}) and X ∼= αi .
TOPOLOGICAL RAMSEY NUMBERS 89
We write
β →cl (αi )ni∈κ
to mean that for every function c : [β]n → κ there exists some subset X ⊆ β and
some i ∈ κ such that X is an i-homogeneous closed copy of αi , i.e., [X]n ⊆ c−1 ({i})
and X is order-homeomorphic to αi .
Note that in both cases, c is arbitrary (no continuity or definability is required).
The two versions coincide in many cases, in particular for ordinals of the form
ω γ or ω γ · m + 1 with m ∈ ω [Hil16, Theorem 2.16]. But in general they may
differ; for example, ω + 2 is homeomorphic but not order-homeomorphic to ω + 1,
and thus ω + 1 →top (ω + 2)11 while ω + 1 →cl (ω + 2)11 .
We focus primarily on the case n = 2, where we define the topological ordinal
Ramsey number Rtop (αi )i∈κ to be the least ordinal β (when one exists) such that
β →top (αi )i∈κ , and the closed ordinal Ramsey number Rcl (αi )i∈κ likewise. In
particular, we study Rtop (α, k) and Rcl (α, k) when α is a countable ordinal and k
a positive integer, which have not previously been explored.
An important ingredient in our work is the case n = 1, which was first studied
by Baumgartner and Weiss [Bau86]. Analogously to Ramsey numbers, we define
the pigeonhole number P (αi )i∈κ to be the least ordinal β such that β → (αi )1i∈κ , and
we define the topological pigeonhole number P top (αi )i∈κ and the closed pigeonhole
number P cl (αi )i∈κ in a similar fashion. An algorithm for finding the topological
pigeonhole numbers is given in [Hil16], and we will describe the modifications
required to obtain the closed pigeonhole numbers (see Section 3). Repeatedly we
make use of the classical fact that the indecomposable ordinals, the powers of ω,
satisfy P (ω α )k = ω α for all finite k.
Returning to the case n = 2, previous work has tackled uncountable Ramsey
numbers. Erdős and Rado [ER56] showed that ω1 → (ω1 , ω + 1)2 . Laver noted in
[Lav75] (and a proof, using a pressing down argument, is described for example in
[Sch12]) that one actually has ω1 → (Stationary, top ω + 1)2 , meaning that one can
ensure either a 0-homogeneous stationary subset or a 1-homogeneous topological
copy of ω + 1. Since every stationary subset of ω1 contains topological copies of
every α ∈ ω1 [Fri74], this therefore shows that ω1 →top (α, ω + 1)2 for all countable
α. In turn, this result was later extended by Schipperus using elementary submodel
techniques to show that ω1 →top (α)2k for all α ∈ ω1 and all finite k [Sch12] (the
topological Baumgartner-Hajnal theorem). Meanwhile, both ω1 → (ω1 , α)2 for all
α ∈ ω1 [Tod83] and ω1 → (ω1 , ω + 2)2 [Haj60] are consistent with ZFC, though
the topological version of the former remains unchecked. Finally, it is also known
that β →top (ω + 1)2ℵ0 for all ordinals β [Wei90].
When n > 2, not much more can be said in the setting of countable ordinals,
since Kruse showed that if n ≥ 3 and β is a countable ordinal, then β → (ω + 1, n +
1)n [Kru65]. As for the ordinal ω1 , [ER56, Theorem 39 (ii)] shows that ω1 → (ω +
1)nk for all finite n, k and, in fact, it seems to be a folklore result that ω1 →top (ω+1)nk
for all finite n, k; a proof can be found in [HJW90]. On the other hand, we have the
negative relations ω1 → (n + 1)nω for n ≥ 2 [ER56], ω1 → (ω + 2, n + 1)n for n ≥ 4
[Kru65], ω1 → (ω + 2, ω)3 [Jon00] and ω1 → (ω1 , 4)3 [Haj64]. All that remains
to be settled is the conjecture that ω1 → (α, k)3 for all α ∈ ω1 and all finite k, with
the strongest result to date being that ω1 → (ω · 2 + 1, k)3 for all finite k [Jon17],
building on the techniques of [Tod83]. The topological version of this result remains
unexplored. In a different direction, Rosenberg [Ros15] has recently verified that
X →top (ω + 1)nk for all finite n, k whenever X is an uncountable separable metric

space. Apparently, this last result has been discovered a few times. In particular,
Stevo Todorcevic has informed us that in unpublished work from around 1990, he
and Weiss characterized those metric spaces X for which X →top (ω + 1)rk for all
finite r, k. He also provided us with a copy of his short unpublished note [Tod96]
from 1996, where he shows that a regular space X with a point-countable basis is
left-separated iff X →top (ω + 1)22 iff for some integers r, k ≥ 2, X →top (ω + 1)rk .
For a general introduction to the partition calculus of countable ordinals, see
[Cai14], which also provides some context for the work we present here. For a gen-
eral introduction to topological Ramsey theory, see [Wei90]. In a sense, [Bau86]
is the direct precursor of this paper and [Hil16]. Both authors were working inde-
pendently on this topic when the talk on which [Cai14] is based was given. Upon
learning of each other’s research, we decided to combine and extend our notes.
Among other extensions, this resulted in the weak topological Erdős-Milner theo-
rem.
1.1. Our results. Our main result is the weak topological Erdős-Milner the-
orem:
Theorem. Let α and β be countable nonzero ordinals, and let k > 1 be a
positive integer. If
α
ω ω →top (ω β , k)2 ,
then
ω ω ·β →top (ω β , k + 1)2 .
α
α α
This is Theorem 10.2. Since trivially ω ω →top (ω ω , 2)2 , it follows by in-
α α·k
duction on k that Rtop (ω ω , k + 1) ≤ ω ω . Hence the ordinals Rtop (α, k) (and
cl
R (α, k), since the two versions coincide when α is a power of ω) are countable for
all countable α and all finite k. (The reason why we use here the adjective weak is
discussed in Section 10.)
The paper is organized as follows.
Section 2 completes the preliminaries, including further details of the relation-
ship between the topological and closed partition relations.
In Section 3 we introduce the case n = 1 and solve the closed pigeonhole
principle for ordinals: given a cardinal κ and an ordinal αi ≥ 2 for each i ∈ κ,
we provide an algorithm to compute P cl (αi )i∈κ . This is presented in Theorem 3.2.
(The corresponding result for P top , building on [Bau86, Theorem 2.3], is the main
theorem of [Hil16].) We also prove a simple lower bound for Ramsey numbers
in terms of pigeonhole numbers, which remains our best lower bound with the
exception of a couple of special cases.
In Section 4 we look at ω + 1 and prove Theorem 4.1:
Theorem. If k is a positive integer, then Rtop (ω+1, k+1) = Rcl (ω+1, k+1) =
k
ω + 1.
In Section 5 we describe several “stepping up” techniques, resulting in Propo-
sitions 5.1 and 5.5:
Theorem. Let α be a successor ordinal, and let k, m and n be positive integers
with k ≥ 2.
(1) Rcl (α + 1, k + 1) ≤ P cl (Rcl (α, k + 1), Rcl (α + 1, k)) + 1.
(2) Rcl (α + 1, k + 1) ≤ P cl (Rcl (α, k + 1))k + Rcl (α + 1, k).

(3) Rcl (ω · m + n + 1, k + 1) ≤ Rcl (ω · m + 1, k + 1) + P cl ((Rcl (ω · m + n +
1, k))2m , R(n, k + 1)).
From these one may deduce upper bounds for Rcl (ω + n, k) for all finite n, k.
We also indicate a lower bound in Lemma 5.3, yielding Rcl (ω + 2, 3) = ω 2 · 2 + ω + 2.
In Section 6 we use the last of these techniques, which in turn utilizes an
ultrafilter argument, to prove Theorem 6.1:
Theorem. If k and m are positive integers, then Rtop (ω · m + 1, k + 1) =
R (ω · m + 1, k + 1) < ω ω .
cl
In fact, in each case the proof gives an explicit upper bound below ω ω .
The ultrafilter approach is then further refined in Section 7 to prove Theorem
7.1:
Theorem. If k is a positive integer, then Rtop (ω 2 , k) = Rcl (ω 2 , k) ≤ ω ω .
Despite the use of a non-principal ultrafilter on ω, our arguments are formaliz-
able without any appeal to the axiom of choice (see Remark 6.8).
In Section 8 we introduce a different approach in terms of the anti-tree partial
ordering on ordinals, which was independently considered by Piña in [Pn15] for
her work on extending some results of [Bau86]. This approach is not used directly
in later sections, but does provide a helpful perspective. We use this approach to
provide a second proof of Theorem 4.1, and to prove Theorem 8.1 (but see Remark
8.10):
Theorem. ω 2 · 3 ≤ Rtop (ω · 2, 3) ≤ ω 3 · 100.
In Section 9 we climb beyond ω 2 and prove Corollary 9.2:
Theorem. If k is a positive integer, then Rtop (ω 2 + 1, k + 2) = Rcl (ω 2 + 1, k +
2) ≤ ω ω·k + 1.
We deduce this from a more general result, which can also be seen as a gener-
alization of Theorem 4.1:
Theorem. Let α and β be countable ordinals with β > 0, let k be a positive
integer, and suppose they satisfy a “cofinal version” of
α
ω ω →cl (ω β , k + 2)2 .
Then
ω ω ·(k+1) + 1 →cl (ω β + 1, k + 2)2 .
α
α
Moreover, if ω ω > ω β , then in fact
α
·k
ωω + 1 →cl (ω β + 1, k + 2)2 .
This is Theorem 9.1. (We will give the precise meaning of this cofinal partition
relation in the full statement of the theorem.)
Our main result, the weak topological Erdős-Milner theorem, is finally proved
in Section 10. The proof is somewhat technical, but does not directly appeal to any
earlier results. From this we obtain Corollary 10.5:
Theorem. Let α be a countable nonzero ordinal, and let k, m and n be positive
integers.
α α α·k
(1) Rtop (ω ω , k + 1) = Rcl (ω ω , k + 1) ≤ ωω .
ω α·k
α α ω + 1, if α is infinite
(2) Rtop (ω ω +1, k+1) = Rcl (ω ω +1, k+1) ≤ (n+1)·k−1
ωω + 1, if α = n is finite.
k
·n
(3) Rtop (ω n · m + 1, k + 2) = Rcl (ω n · m + 1, k + 2) ≤ ω ω · R(m, k + 2) + 1.
We close in Section 11 with some questions.
1.2. Acknowledgements. The first author wants to thank Jean Larson for
her enthusiasm and encouragement when this project was just getting started. We
also want to thank Omer Mermelstein and Stevo Todorcevic, for their feedback on
an earlier version of this paper and their permission to quote some of their results.
2. Preliminaries
In this paper, all arithmetic operations on ordinals are in the ordinal sense.
We use interval notation in the usual fashion, so that for example if α and β are
ordinals then [α, β) = {x | α ≤ x < β}. We denote the cardinality of a set X by
|X|, and use the symbol ∼ = to denote the homeomorphism relation.
The classical, topological and closed partition relations are defined in the in-
troduction. When n = 1 we may work with β rather than [β]1 for simplicity, and
we write β → (α)nκ for β → (αi )ni∈κ when αi = α for all i ∈ κ, and similarly for the
topological and closed relations. We refer to the function c in these definitions as
a coloring, and when the range of c is 2 we identify 0 with the color red and 1 with
the color blue.
The classical, topological, and closed ordinal pigeonhole and Ramsey num-
bers are also defined in the introduction. Note that R((αi )i∈κ , (2)λ ) = R(αi )i∈κ
for any cardinal λ, and that for fixed κ, R(αi )i∈κ is a monotonically increasing
function of (αi )i∈κ (pointwise), and similarly for the topological and closed rela-
tions. Note also that the closed partition relation implies the other two, and hence
R(αi )i∈κ ≤ Rcl (αi )i∈κ and Rtop (αi )i∈κ ≤ Rcl (αi )i∈κ (and similarly for the pigeon-
hole numbers).
The topological and closed partition relations are closely related, thanks to the
following notion.
Definition 2.1. An ordinal α is said to be order-reinforcing if and only if,
whenever X is a subspace of an ordinal with X ∼= α, then there is a subspace
Y ⊆ X such that Y is order-homeomorphic to α.
It is clear from this definition that, when applied only to order-reinforcing
ordinals, the topological and closed pigeonhole and Ramsey numbers coincide. In
particular, if α is order-reinforcing, then Rtop (α, k) = Rcl (α, k) for all finite k.
Building on work of Baumgartner [Bau86, Theorem 0.2], the order-reinforcing
ordinals were classified in [Hil16, Corollary 2.17].
Theorem 2.2. An ordinal α is order-reinforcing if and only if either α is finite,
or α = ω γ ·m+1 for some nonzero ordinal γ and some positive integer m, or α = ω γ
for some nonzero ordinal γ.
Finally we require the notions of Cantor-Bendixson derivative and rank, which
are central to the study of ordinal topologies.
Definition 2.3. Let X be a topological space. The Cantor-Bendixson deriv-

ative X of X is the set of limit points of X, i.e.,
X = X \ {x ∈ X | x is isolated}.
The iterated derivatives of X are defined recursively for γ an ordinal by
(1) X (0) = X,

(2) X (γ+1) = X (γ) , and

(3) X (γ) = δ<γ X (δ) when γ is a nonzero limit.
Definition 2.4. If x is a nonzero ordinal, then there is a unique sequence of or-
dinals γ1 > γ2 > · · · > γn and a unique sequence of positive integers m1 , m2 , . . . , mn
such that
x = ω γ1 · m1 + ω γ2 · m2 + · · · + ω γn · mn .
This representation is the well-known Cantor normal form of x. The Cantor-
Bendixson rank of x is defined by

γn , if x > 0
CB(x) =
0, if x = 0.
The relationship between these two notions is that if α is an ordinal endowed

with the order topology and x ∈ α, then the Cantor-Bendixson rank of x is the
greatest ordinal γ such that x ∈ α(γ) . Some basic proofs that make use of these
notions can be found in the preliminaries of [Hil16].
Though we do not need it here, we recall the classification of ordinals up to
homeomorphism, due to Flum and Martı́nez [FM88, 2.5 Remark 3]. Here we state
the result using the terminology of Kieftenbeld and Löwe [KL06]. In order to do
this, for a nonzero ordinal x with Cantor normal form
x = ω γ1 · m1 + ω γ2 · m2 + · · · + ω γn · mn ,
write γ1 (x) for γ1 and m1 (x) for m1 , and let p(x), the purity of x, be 0 if x =
ω γ1 (x) · m1 (x) and γ1 (x) > 0, and ω CB(x) otherwise.
Theorem 2.5 (Flum-Martı́nez). Two nonzero ordinals α and β are homeo-
morphic if and only if γ1 (α) = γ1 (β), m1 (α) = m1 (β), and p(α) = p(β).
More relevant to this paper is perhaps the classification of ordinals up to biem-
beddability (being homeomorphic to a subspace of one another) rather than home-
omorphism. See [Hil16] for further details.
3. The closed pigeonhole principle for ordinals

The pigeonhole numbers are an important prerequisite to the study of Ramsey
numbers. For example, we have the following lower bound for a Ramsey number
given by considering a k-partite graph.
Proposition 3.1. If α ≥ 2 is an ordinal and k is a positive integer, then
R(α, k + 1) ≥ P (α)k ,
and similarly for the topological and closed relations: Rtop (α, k + 1) ≥ P top (α, k)
and Rcl (α, k + 1) ≥ P cl (α, k).
Proof. Suppose β < P (α)k . By definition of the pigeonhole number, it follows

that there exists a coloring c : β → k such that for all i ∈ k, no subset of c−1 ({i})
is order-isomorphic to α. To see that β → (α, k + 1)2 , simply consider the coloring
d : [β]2 → {red,blue} given by

red, if c(x) = c(y)
d({x, y}) =
blue, if c(x) = c(y).
It is straightforward to verify that d does indeed witness β → (α, k + 1)2 .

For the topological and closed relations, simply replace “order-isomorphic” with
“homeomorphic” or “order-homeomorphic” as necessary.
The values of P top are given in [Hil16], and the key result of this section
extends this to P cl .
Theorem 3.2 (The closed pigeonhole principle for ordinals). Given a cardinal
κ and an ordinal αi ≥ 2 for each i ∈ κ, there is an algorithm to compute P cl (αi )i∈κ .
In particular, suppose that k < ω and that α1 , . . . , αk are countable and bigger than
one. For any nonzero countable ordinals β1 , . . . , βk we have:
(1) P cl (β1 , . . . , βk , 1) = P cl (β1 , . . . , βk ), P cl (β1 ) = β1 , P cl (β1 , . . . , βk ) is in-
variant under permutations of the βi .
(2) P cl (ω β1 + 1, . . . , ω βk + 1) = ω β1 #···#βk + 1, where # denotes the natural
(Hessenberg) sum.
(3) P cl (ω β1 , . . . , ω βk ) = ω β1 ···βk , where ) denotes the Milner-Rado sum.
(4) If for all i ≤ k, βi is least such that αi ≤ ω βi , and if equality holds for at
least one i, then P cl (α1 , . . . , αk ) = P cl (ω β1 , . . . , ω βk ).
(5) If no αi is a power of ω, write αi = 1 + γi if αi is finite and αi =
ω βi + 1 + γi otherwise, where γi < ω βi +1 . Write ti = 1 or ti = ω βi + 1
depending on whether αi is finite. If, for each i ≤ k, we let Qi =
P cl (α1 , . . . , αi−1 , γi , αi+1 , . . . , αk ), then P cl (α1 , . . . , αk ) = P cl (t1 , . . . , tk )+
max{Q1 , . . . , Qk }.
The cases listed cover all uses of the theorem in this paper. To avoid clutter-
ing the paper with an off-topic detour, the reader can take our use of the word
“algorithm” here merely as a shorthand for the full statement of the result, which
divides into several additional cases (depending on whether we use infinitely many
colors rather than a finite number k, or if some of the αi are uncountable). Rather
than listing all these cases explicitly, we refer the reader to [Hil16] for the full
statement for P top , and explain within the proof the modifications required to ob-
tain P cl . We should remark, in particular, that P cl restricted to countable ordinals
is a primitive recursive set function and that, given reasonable representations of
the Cantor normal forms of the countable ordinals α1 , . . . , αk , and of the expo-
nents present in these expressions, the reader should have no difficulties providing
an actual algorithm that returns (a representation of) the Cantor normal form of
P cl (α1 , . . . , αk ).
For completeness, we recall that if α = ω γ1 · m1 + · · · + ω γk · mk and β =
ω · n1 + · · · + ω γk · nk , where γ1 > · · · > γk and the mi and ni are natural numbers,
γ1
then α#β = ω γ1 · (m1 + n1 ) + · · · + ω γk · (mk + nk ), and set 0#α = α#0 = α for all
α. Also, α ) β is the least γ such that if α̃ < α and β̃ < β, then α̃#β̃ = γ. Both of
these operations are commutative and associative, allowing brackets to be omitted.
Before providing the proof of the theorem, let us first look at a few special
cases. Some of these will be used later, while others are merely illustrative. The
reader may attempt to verify them directly.
Fact 3.3. Let k and m be positive integers.
(1) P top (m, k) = P cl (m, k) = P (m, k) = m + k − 1.
(2) P top (ω)2 = P cl (ω)2 = P (ω)2 = ω.
(3) P top (ω + 1, k) = P cl (ω + 1, k) = ω · k + 1.
(4) P top (ω + 1, ω) = P cl (ω + 1, ω) = ω 2 .
(5) P top (ω + m, ω + k) = P top (ω + 1)2 = ω 2 + 1. On the other hand, if m ≥ k
then P cl (ω + m, ω + k) = ω 2 + ω · (m − 1) + k.
(6) P top (ω·2)2 = P cl (ω·2)2 = ω 2 ·2. If m > 2, then P cl (ω·m)2 = ω 2 ·(2m−2),
whereas P top (ω · m)2 = ω 2 · (2m − 3) + 1.
(7) P top (ω·m+k)2 = P top (ω·m+1)2 = ω·(2m−1)+1, while P cl (ω·m+k)2 =
ω 2 · (2m − 1) + ω · (k − 1) + k.
Here is one final special case, for which we provide a short proof.
Proposition 3.4. If k is a positive integer, then
P top (ω + 1)k = P cl (ω + 1)k = ω k + 1.
(By contrast, P (ω + 1)k = ω · k + 1.)
Proof. The first equality is immediate from the fact that ω + 1 is order-
reinforcing.
To see that ω k →top (ω + 1)1k , simply color each x ∈ ω k with color CB(x), and
observe that each color class is discrete.
We prove that ω k +1 →top (ω +1)1k by induction on k. The case k = 1 is trivial.
k−1suppose k ≥
For the inductive step, 2 and let c : ω k + 1 → k be a coloring. For each
m ∈ ω, let Xm = ω · m + 1, ω k−1
· (m + 1) ∼ = ω k−1 + 1. If c−1 ({i}) ∩ Xm = ∅
for some i ∈ k and some m ∈ ω, then the restriction of c to Xm uses at most
k − 1 colors, and we are done by the inductive hypothesis. Otherwise there exists
xm ∈ c−1 ({j}) ∩ Xm for each m ∈ ω where j = c(ω k ), whence {xm | m ∈ ω} ∪ {ω k }
is a topological copy of ω + 1 in color j.
Here is the proof in full of the closed pigeonhole principle for ordinals.
Proof of Theorem 3.2. The argument assumes familiarity with [Hil16],
where P top (αi )i∈κ is computed. The same proof shows that P cl (αi )i∈κ = P top (αi )i∈κ
except in two cases, which we now examine.
The first case is when κ is finite and greater than 1, αr ≥ ω1 + 1 for some r ∈ κ,
αi is finite for all i ∈ κ \ {r}, and αr is not a power of ω, say αr = ω β · m + 1 + γ
for some ordinal β, some positive integer m and some ordinal γ ≤ ω β (case 2(c)(ii)
in [Hil16]). Then
⎛ ⎞

P top (αi )i∈κ = ω β · ⎝ (αi − 1) + m⎠ + 1,
i∈κ\{r}
whereas a very similar argument shows that

⎛ ⎞

P cl (αi )i∈κ = ω β · ⎝ (αi − 1) + m⎠ + 1 + γ.
i∈κ\{r}
The second case is when (αi )i∈κ is a finite sequence of countable ordinals (case
6). In this case the following results carry across for β1 , β2 , . . . , βk ∈ ω1 \ {0}.
• P cl (ω β1 + 1, ω β2 + 1, . . . , ω βk + 1) = ω β1 #β2 #···#βk + 1, where # denotes
the natural sum [Hil16, Theorem 4.5].
• P cl (ω β1 , ω β2 , . . . , ω βk ) = ω β1 β2 ···βk , where ) denotes the Milner-Rado
sum [Hil16, Theorem 4.6].
• Let α1 , α2 , . . . , αk ∈ ω1 with αr = ω βr for some r ∈ {1, 2, . . . , k}. Sup-
pose βi is minimal subject to the condition that αi ≤ ω βi for all i ∈
{1, 2, . . . , k}. We then have that P cl (α1 , α2 , . . . , αk ) = P cl (ω β1 , ω β2 , . . . , ω βk )
by [Hil16, Theorem 4.7].
Thus it remains to compute P cl (α1 , α2 , . . . , αk ) when αi ∈ ω1 is not a power
of ω for any i ∈ {1, 2, . . . , k}. In that case, for all i ∈ {1, 2, . . . , k} we may write
αi = ω βi + 1 + γi for some βi ∈ ω1 \ {0} and some ordinal γi < ω βi +1 (or αi = 1 + γi
for some ordinal γi < ω if αi is finite).
Let Qi = P cl (α1 , . . . , αi−1 , γi , αi+1 , . . . , αk ) for each i ∈ {1, 2, . . . , k}. We claim
that
P cl (α1 , α2 , . . . , αk ) = P cl (ω β1 + 1, ω β2 + 1, . . . , ω βk + 1) + max{Q1 , Q2 , . . . , Qk }
(replacing ω βi + 1 with 1 if αi is finite). That this is large enough is clear. That no
smaller ordinal is large enough follows from the existence for each r ∈ {1, 2, . . . , k}
of a coloring cr : ω β1 #β2 #···#βk + 1 → {1, 2, . . . , k} with the property that for each
i ∈ {1, 2, . . . , k}, there is a closed copy of ω βi + 1 in color i if and only if i = r,
and no closed copy of any ordinal larger than ω βr + 1 in color r. To obtain such
a coloring, simply extend the coloring given in [Hil16, Theorem 4.5] by setting
cr (ω β1 #β2 #···#βk ) = r. This equation allows P cl (α1 , α2 , . . . , αk ) to be computed
recursively, since it will be reduced to the three cases above after a finite number
of steps.
The main idea ultimately leading to Theorem 3.2 in the setting of countable
ordinals is a result of Baumgartner and Weiss [Bau86, Theorem 2.3]. Some of
the details of that argument reappear in several of the proofs below. A particular
special case of Theorem 3.2 of interest is the following easy consequence of that
result [Bau86, Corollary 2.4].
Corollary 3.5 (Baumgartner-Weiss). If β is a nonzero countable ordinal and
β β
k is finite, then P cl (ω ω )k = ω ω .
Although this result follows directly from [Bau86, Theorem 2.3], we briefly
illustrate how to obtain it as a special case of our more general result.
Proof. If αi < ω β for all i ∈ {1, . . . , k}, then α1 # · · · #αk < ω β , so that
ω ) · · · ) ω β = ω β . Now use clause ( 3) of Theorem 3.2.
β

4. The ordinal ω + 1
In this section we prove the following result.
Theorem 4.1. If k is a positive integer, then
Rtop (ω + 1, k + 1) = Rcl (ω + 1, k + 1) = ω k + 1.
Our proof is quite similar to the proof of Proposition 3.4. Later, we will provide
a second proof (see Section 8).
Proof. The first equality is immediate from the fact that ω + 1 is order-
reinforcing, and the fact that Rtop (ω + 1, k + 1) ≥ ω k + 1 follows from Propositions
3.1 and 3.4.
It remains to prove that ω k + 1 →top (ω + 1, k + 1)2 , which we do by induction
on k. The case k = 1 is trivial. For the inductive step, suppose k ≥ 2 and let
c : [ω k + 1]2 → {red, blue}
be a coloring. As in the proof of Proposition 3.4, for
each m ∈ ω, let Xm = ω k−1 · m + 1, ω k−1 · (m + 1) ∼ = ω k−1 + 1. For each m ∈ ω,
we may assume by the inductive hypothesis that Xm has a blue-homogeneous set
Bm of size k, or else it has a red-homogeneous closed copy of ω + 1 and we are
done. If for any m ∈ ω it is the case that c({x, ω k }) = blue for all x ∈ Bm , then
Bm ∪ {ω k } is a blue-homogeneous set of size k + 1 and we are done. Otherwise for
each m ∈ ω we may choose xm ∈ Bm with c({xm , ω k }) = red. Finally, by Ramsey’s
theorem, within the set {xm | m ∈ ω} there is either a blue-homogeneous set of
size k + 1, in which case we are done, or an infinite red-homogeneous set H, in
which case H ∪ {ω k } is a red-homogeneous closed copy of ω + 1, and we are done
as well.
It is somewhat surprising that we are able to obtain an exact equality here
using the lower bound from Proposition 3.1. As we shall see, for ordinals larger
than ω + 1 there is typically a large gap between our upper and lower bounds, and
we expect that improvement will usually be possible on both sides.
5. Stepping up by one
In this section we consider how to obtain an upper bound for Rcl (α + 1, k + 1),
given upper bounds for Rcl (α, k + 1) and Rcl (α + 1, k) for a successor ordinal α and
a positive integer k. Note that if α is an infinite successor ordinal, then α + 1 ∼= α,
so trivially Rtop (α + 1, k + 1) = Rtop (α, k + 1), which is why we study Rcl instead.
First we give two simple upper bounds. The first comes from considering the
edges incident to the largest point, as in a standard proof of the existence of the
finite Ramsey numbers. The second typically gives worse bounds, but is more
similar to the technique we will consider next.
Proposition 5.1. Let α be a successor ordinal, and let k ≥ 2 be a positive
integer.
(1) Rcl (α + 1, k + 1) ≤ P cl (Rcl (α, k + 1), Rcl (α + 1, k)) + 1.
(2) Rcl (α + 1, k + 1) ≤ P cl (Rcl (α, k + 1))k + Rcl (α + 1, k).
Proof. First note that since α is a successor ordinal, a closed copy of α + 1 is
obtained from any closed copy of α together with any larger point.
(1) Let β = P cl (Rcl (α, k + 1), Rcl (α + 1, k)) and let c : [β + 1]2 → {red, blue}
be a coloring. This induces a coloring d : β → {red, blue} given by d(x) =
c({x, β}). By definition of P cl , there exists X ⊆ β such that either X ⊆
d−1 ({red}) and X is a closed copy of Rcl (α, k + 1), or X ⊆ d−1 ({blue})
and X is a closed copy of Rcl (α + 1, k). In the first case, by definition
of Rcl , we are either done immediately, or we obtain a red-homogeneous
closed copy Y of α, in which case Y ∪ {β} is a red-homogeneous closed
copy of α + 1. The second case is similar.
(2) Let β = P cl (Rcl (α, k + 1))k and let c : [β + Rcl (α + 1, k)]2 → {red, blue}
be a coloring. By definition of Rcl , either we are done or there is a blue-
homogeneous set of k points, say {x1 , x2 , . . . , xk }, with xi ≥ β for all
i ∈ {1, . . . , k}. If for any y < β we have c({y, xi }) = blue for all i, then
we are done. Otherwise define a coloring d : β → k by taking d(y) to
be some i such that c({y, xi }) = red. Then by definition of P cl , there
exists i and X ⊆ β such that X ⊆ d−1 ({i}) and X is a closed copy of
Rcl (α, k + 1). But then X either contains a blue-homogeneous set of k + 1
points, or a red-homogeneous closed copy Y of α, in which case Y ∪ {xi }
is a red-homogeneous closed copy of α + 1, as required.
Note that Rcl (α, 2) = α for any ordinal α. Hence if α is a successor ordinal
and we have an upper bound on Rcl (α, k) for every positive integer k, then we may
easily apply either of these two inequalities recursively to obtain upper bounds on
Rcl (α + n, k) for all positive integers n and k.
Unfortunately, neither of these techniques appears to generalize well to limit
ordinals, since they are “backward-looking” in some sense. Our “forward-looking”
technique is a little more complicated and, in the form presented here, it only works
below ω 2 . Before we state the general result, here is an illustrative special case.
Lemma 5.2. Rcl (ω + 2, 3) ≤ Rcl (ω + 1, 3) + P cl (ω + 2)2 = ω 2 · 2 + ω + 2.
Proof. First note that Rcl (ω + 1, 3) + P cl (ω + 2)2 = (ω 2 + 1) + (ω 2 + ω + 2) =

ω · 2 + ω + 2 by Theorem 4.1 and Fact 3.3. It remains to prove that Rcl (ω + 1, 3) +
2
P cl (ω + 2)2 →cl (ω + 2, 3)2 .

Let β = Rcl (ω + 1, 3), let c : [β + P cl (ω + 2)2 ]2 → {red, blue} be a coloring, and
suppose for contradiction that there is no red-homogeneous closed copy of ω + 2
and no blue triangle.
By definition of Rcl and our assumption that the coloring admits no blue tri-
angles, there exists a red-homogeneous closed copy X of ω + 1 contained in β. Let
x be the largest point in X and let H = X \ {x}.
Let A1 = {y ≥ β | c({x, y}) = blue} and let A2 = {y ≥ β | c({h, y}) = blue for
all but finitely many h ∈ H}.
First of all, we claim that A1 is red-homogeneous. This is because if y, z ∈ A1
and c({y, z}) = blue, then {x, y, z} is a blue triangle.
Next, we claim that A2 is red-homogeneous. This is because if y, z ∈ A2 and
c({y, z}) = blue, then c({h, y}) = c({h, z}) = blue for all but finitely many h ∈ H,
and so {h, y, z} is a blue triangle for any such h.
Finally, we claim that if y ≥ β, then y ∈ A1 ∪ A2 . For otherwise we would have
y ≥ β with c({x, y}) = red and c({h, y}) = red for all h in some infinite subset
K ⊆ H, whence K ∪ {x, y} is a red-homogeneous closed copy of ω + 2. Hence by
definition of P cl , either A1 or A2 contains a closed copy of ω + 2, which by the
above claims must be red-homogeneous, and we are done.
Digressing briefly, we show that in this particular case the upper bound is
optimal, and thus Rcl (ω + 2, 3) = ω 2 · 2 + ω + 2. The coloring we present was found
by Omer Mermelstein.
Lemma 5.3. Rcl (ω + 2, 3) ≥ ω 2 · 2 + ω + 2.
Proof. We provide a coloring witnessing ω 2 · 2 + ω + 1 →cl (ω + 2, 3)2 . In

order to define it, let G be the graph represented by the following diagram.
2 2
ω ω ·2
2 2
{ω · (n + 1) | n ∈ ω} ω + ω · (n + 1) | n ∈ ω ω ·2+ω
2 2
{0} ∪ x + 1 | x ∈ ω 2 ω + x + 1 | x ∈ ω2 ω ·2+x+1|x∈ω
Define a coloring c : [ω 2 · 2 + ω + 1]2 → {red, blue} by setting c({x, y}) = blue if and
only if x and y lie in distinct, adjacent vertices of G. First note that there is no blue
triangle since G is triangle-free. Now suppose X is a closed copy of ω + 2, and write
X = Z ∪ {x} ∪ {y} with z < x < y for all z ∈ Z. If Z ∪ {x} is red-homogeneous,
then either x = ω 2 and Z ⊆ {0} ∪ {x + 1 | x ∈ ω 2 }, or x = ω 2 · 2 and (discarding a
finite initial segment of Z if necessary) we may assume Z ⊆ {ω 2 + x + 1 | x ∈ ω 2 }.
In each case either c({x, y}) = blue or c({z, y}) = blue for all z ∈ Z. Hence X
cannot be red-homogeneous, and we are done.
Here is the general formulation of our “forward-looking” technique.
Proposition 5.4. Let k, m and n be positive integers with k ≥ 2. Then
Rcl (ω · m + n + 1, k + 1) ≤ Rcl (ω · m + n, k + 1) + P cl (Rcl (ω · m + n + 1, k))2m+n−1 .
Proof. Let β = Rcl (ω · m + n, k + 1) and let c : [β + P cl (Rcl (ω · m + n +
1, k))2m+n−1 ]2 → {red, blue} be a coloring.
By definition of Rcl , either there is a blue-homogeneous set of k + 1 points, in
which case we are done, or there exists X ⊆ β such that X is a red-homogeneous
closed copy of ω · m + n. In that case, write
X = H1 ∪ {x1 } ∪ H2 ∪ {x2 } ∪ · · · ∪ {xm−1 } ∪ Hm ∪ {y1 , y2 , . . . , yn }
with H1 , H2 , . . . , Hm each of order type ω and h1 < x1 < h2 < x2 < · · · < xm−1 <
hm < y1 < y2 < · · · < yn whenever hi ∈ Hi for all i ∈ {1, 2, . . . m}.
For each z ≥ β, if every single one of the following 2m + n − 1 conditions holds,
then X ∪ {z} contains a closed copy of ω · m + n + 1, and we are done.
• c({h, z}) = red for infinitely many h ∈ Hi (one condition for each i ∈
{1, 2, . . . m})
• c({xi , z}) = red (one condition for each i ∈ {1, 2, . . . , m − 1})
• c({yi , z}) = red (one condition for each i ∈ {1, 2, . . . , n})
Thus we may assume that one of these conditions fails for each z ≥ β. This
induces a 2m+n−1-coloring of these points, so by definition of P cl there is a closed
copy X of Rcl (ω · m + n + 1, k) among these points such that the same condition
fails for each z ∈ X.
Finally, by definition of Rcl , X contains either a red-homogeneous closed copy
of ω · m + n + 1, in which case we are done, or a blue-homogeneous set of k points,
in which case using the failed condition we can find a final point with which to
construct a blue-homogeneous set of k + 1 points.
We now present a refinement of Proposition 5.4 that introduces ideas we will
be elaborating on in the following sections. This refinement allows us to step up
from ω · m + 1 to ω · m + n + 1 in one step, which in turn translates into improved
bounds.
Proposition 5.5. For all positive integers k, m, n with k ≥ 2, we have

Rcl (ω·m+n+1, k+1) ≤ Rcl (ω·m+1, k+1)+P cl ((Rcl (ω·m+n+1, k))2m , R(n, k+1)),
where the notation ((x)2m , y) is shorthand for (x, x, . . . , x, y) with x appearing pre-
cisely 2m times.
Proposition 5.5 formally supersedes Proposition 5.4. We have decided to include
both versions since the combinatorics are somewhat simpler in the earlier argument.
The new proof uses a non-principal ultrafilter on ω. It turns out that this use is
superfluous, but makes the idea more transparent, since in some sense the ultrafilter
makes many choices for us. The result is actually provable in (a weak sub-theory
of) ZF (see Remark 6.8).
Proof. Fix a non-principal ultrafilter U on ω, let
γ = Rcl (ω · m + 1, k + 1) and β = P cl ((Rcl (ω · m + n + 1, k))2m , R(n, k + 1)),
and consider a coloring c : [γ +β]2 → {red,blue}. By definition of γ, we may assume
that there is a set H = H1 ∪ {x1 } ∪ · · · ∪ Hm ∪ {xm } ⊆ γ that is a red-homogenous
closed copy of ω ·m+1 (else, there is a blue-homogeneous subset of γ of k +1 points,
and we are done). Here, each Hi has type ω, and a < xi < b < xi+1 whenever
i < m, a ∈ Hi and b ∈ Hi+1 . List the elements of each Hi in increasing order as
hi,0 < hi,1 < . . . .
Now every ordinal in γ + β larger than xm lies in at least one of the following
2m + 1 sets:
• Ai = {a > xm | c(xi , a) = blue}, for each i ≤ m,
• Bi = {a > xm | {n | c(hi,n , a) = blue} ∈ U}, for each i ≤ m, and
• C = {a > xm | c(xj , a) = red and {n | c(hj,n , a) = red} ∈ U for all j ≤
m}.
Hence, by definition of β, one of the following must hold.
(1) Some Ai contains a closed copy of Rcl (ω · m + n + 1, k). In turn, this copy
either contains a red-homogeneous closed copy of ω · m + n + 1, and we are
done, or a blue-homogeneous set D of k points, in which case {xi } ∪ D is
blue-homogeneous of size k + 1, and we are done as well.
(2) Some Bi contains a closed copy of Rcl (ω · m + n + 1, k). This copy either
contains a red-homogeneous closed copy of ω · m + n + 1, and we are done,
or a blue-homogeneous set D of k points, in which case {x} ∪ D is blue-
homogeneous of size k + 1 for any x ∈ {h ∈ Hi | c(h, a) = blue for all
a ∈ D}, and we are done as well. Here we are using that the intersection
of finitely many sets in U is non-empty.
(3) C contains a set of size R(n, k + 1). This set either contains a red-
homogeneous set D of size n, in which case for each i there is an infinite
subset Hi of Hi such that Hi ∪ D is red-homogeneous, and therefore
H0 ∪ {x0 } ∪ · · · ∪ Hm

∪ {xm } ∪ D
is a red-homogeneous closed copy of ω · m + n + 1, and we are done, or
else it contains a blue-homogeneous set E of size k + 1, in which case we
are done as well. Here we are using that the intersection of finitely many
sets in U is actually infinite.
The result follows.
Because of Theorem 4.1, any one of the inequalities from Propositions 5.1, 5.4
and 5.5 is enough for us to obtain upper bounds for Rcl (ω + n, k) for all finite n, k.
We conclude this section with an explicit statement of a few of these upper bounds.
Curiously, in the second part, we require both Propositions 5.1 and 5.5 in order to
obtain the best bound.
Corollary 5.6. (1) Rcl (ω + 2, 4) ≤ ω 4 · 3 + ω 3 + ω 2 + ω + 2.
(2) If k ≥ 4 is a positive integer, then
Rcl (ω+2, k+1) ≤ ω r ·3+ω r−1 +ω r−2 +ω r−3 +ω r−4 +ω r−8 +ω r−13 +ω r−19 +· · ·+ω k +2,
2
where r = k +k−4
2 .
(3) If n is a positive integer, then Rcl (ω + n, 3) < ω 3 . In particular,
Rcl (ω + 3, 3) ≤ ω 2 · 4 + ω · 2 + 3.
Proof. (1) By Proposition 5.5 and Lemma 5.2, Rcl (ω + 2, 4) ≤ Rcl (ω +
1, 4) + P ((Rcl (ω + 2, 3))2 , R(1, 4)) = ω 3 + 1 + P cl (ω 2 · 2 + ω + 2)2 =
cl
ω 4 · 3 + ω 3 + ω 2 + ω + 2.
(2) This can be obtained from part 1 by recursively applying the first inequal-
ity from Proposition 5.1.
(3) By Proposition 5.5, Rcl (ω + n + 1, 3) ≤ ω 2 + 1 + P cl ((ω + n + 1)2 , R(n, 3)).
Using Lemma 5.2 for the base case, it is easy to see by induction that the
second term here is below ω 3 , but then so is the full term. The particular
case can be verified by direct computation.
For comparison, the corresponding lower bounds given by Proposition 3.1 are
as follows. If k, n ≥ 3 are positive integers, then
Rcl (ω + 2, k + 1) ≥ P cl (ω + 2)k = ω k + ω k−1 + · · · + ω + 2
and
Rcl (ω + n, 3) ≥ P cl (ω + n)2 = ω 2 + ω · (n − 1) + n,
but note the latter is already far behind Rcl (ω +n, 3) ≥ Rcl (ω +2, 3) = ω 2 ·2+ω +2.
6. Ordinals less than ω 2

So far we only have upper bounds on Rcl (α, k) for α < ω · 2. In this section we
extend this to α < ω 2 with the following result.
Theorem 6.1. If k and m are positive integers, then
Rcl (ω · m + 1, k + 1) < ω ω .
In fact, in each case the proof gives an explicit upper bound below ω ω .
Our proof builds upon the stepping up technique from Proposition 5.5. We also
make use of some classical ordinal Ramsey theory, namely, the following result.
Theorem 6.2 (Erdős-Rado). If k and m are positive integers, then R(ω ·
m, k) < ω 2 .
In fact, Erdős and Rado computed the exact values of these Ramsey numbers
in terms of a combinatorial property of finite digraphs. More precisely, we consider
digraphs for which loops are not allowed, but edges between two vertices pointing
in both directions are allowed. The complete digraph on m vertices is denoted
∗
by Km . Recall that a tournament of order k is a digraph obtained by assigning
directions to the edges of the complete (undirected) graph on k vertices, and that a
tournament is transitive if and only if these assignments are compatible, that is, if
and only if whenever x, y and z are distinct vertices with an edge from x to y and
an edge from y to z, then there is also an edge from x to z. The class of transitive
tournaments of order k is denoted by Lk .
Using this terminology, Theorem 6.2 can be deduced from the following two
results of Erdős and Rado (who stated them in a slightly different manner), see
[ER56, Theorem 25] and [ER67].
Lemma 6.3 (Erdős-Rado). If k and m are positive integers, then there is a
positive integer p such that any digraph on p or more vertices admits either an
independent set of size m, or a transitive tournament of order k. We denote the
∗
least such p by R(Km , Lk ).
Theorem 6.4 (Erdős-Rado). If k, m > 1 are positive integers, then R(ω ·
∗
m, k) = ω · R(Km , Lk ).
Before indicating how to prove Theorem 6.1 in general, we first illustrate the
key ideas with the following special case. We use the special case of Theorem 6.4
that R(ω · 2, 3) = ω · 4.
Lemma 6.5. Rcl (ω · 2 + 1, 3) ≤ ω 8 · 7 + 1.
As with Proposition 5.5, the proof uses a non-principal ultrafilter on ω, but see
Remark 6.8.
Proof. To make explicit the reason for using the ordinal ω 8 · 7 + 1, let
• β1 = P cl (ω · 2 + 1, ω · 2 + 1, ω 2 + 1) = ω 4 · 3 + 1,
• β2 = P cl (ω · 2 + 1, ω · 2 + 1, ω 2 + 1 + β1 ) = ω 6 · 5 + 1,
• β3 = P cl (ω · 2 + 1, ω · 2 + 1, ω 2 + 1 + β2 ) = ω 8 · 7 + 1 and
• β = ω 2 + 1 + β3 = ω 8 · 7 + 1.
Fix a non-principal ultrafilter U on ω and let c : [β]2 → {red, blue} be a coloring.
Among the first ω 2 + 1 elements of β, we may assume that we have a red-
homogeneous closed copy of ω + 1. Let x be its largest point and let H be its subset
of order type ω. Write H = {hn | n ∈ ω} with h0 < h1 < . . . .
Now the points in β larger than x form a disjoint union A1 ∪ A2 ∪ A3 , where
• A1 = {a > x | c({x, a}) = blue},
• A2 = {a > x | c({x, a}) = red but {n ∈ ω | c({hn , a}) = blue} ∈ U} and
• A3 = {a > x | c({x, a}) = red and {n ∈ ω | c({hn , a}) = red} ∈ U}.
If either A1 or A2 contains a closed copy of ω · 2 + 1, then we are done. (For
A2 , we use the fact that if U, V ∈ U then U ∩ V ∈ U.) So by definition of β3 , we
may assume that A3 contains a closed copy X of ω 2 + 1 + β2 .
Now we repeat the argument within X. Among the first ω 2 + 1 members of X,
we may assume we that have a red-homogeneous closed copy of ω + 1. Let y be its
largest point, and let I be its subset of order type ω. Write I = {in | n ∈ ω . . . }
with i0 < i1 < . . . . (Note that at this stage, {n ∈ ω | c({hn , i}) = red} ∈ U for all
i ∈ I, yet we cannot conclude from this that there are infinite subsets H ⊆ H and
I ⊆ I such that H ∪ I is red-homogeneous.)
Just as before, write the subset of X lying above y as a disjoint union B1 ∪ B2 ∪
B3 , where b ∈ B1 if and only if c({y, b}) = blue, b ∈ B2 if and only if c({y, b}) = red
but {n ∈ ω | c({in , b}) = blue} ∈ U, and b ∈ B3 if and only if c({y, b}) = red and
{n ∈ ω | c({in , b}) = red} ∈ U. Again we may conclude from the definition of β2

that B3 contains a closed copy Y of ω 2 + 1 + β1 .
Repeat this argument once again within Y , and then pass to a final red-
homogeneous closed copy of ω + 1. We obtain a closed set
H ∪ {x} ∪ I ∪ {y} ∪ J ∪ {z} ∪ K ∪ {w}
of order type ω · 4 + 1 (with J = {jn | n ∈ ω}, j0 < j1 < . . . and K = {kn | n ∈ ω},
k0 < k1 < . . . ) such that H ∪{x}, I ∪{y}, J ∪{z} and K ∪{w} are red-homogeneous,
{x, y, z, w} is red-homogeneous, c({x, in }) = c({x, jn }) = c({x, kn }) = c({y, jn }) =
c({y, kn }) = c({z, kn }) = red for all n ∈ ω, and finally for any a > x, b > y and c > z
in this set, we have {n ∈ ω | c({hn , a}) = red} ∈ U, {n ∈ ω | c({in , b}) = red} ∈ U
and {n ∈ ω | c({jn , c}) = red} ∈ U.
At last we use the ultrafilter, in which is the crucial step of the argument. Let
• H = {h ∈ H | c({h, y}) = c({h, z}) = c({h, w}) = red},
• I = {i ∈ I | c({i, z}) = c({i, w}) = red} and
• J = {j ∈ J | c({j, w}) = red},
each of which corresponds to some U ∈ U and is therefore infinite. (Note that
if we had tried to argue directly, without using the ultrafilter or modifying the
construction in a substantial way, then we would have been able to deduce that J
is infinite, but it would not have been apparent that H or I are.) This ensures
that c({a, b}) = red whenever a ∈ {x, y, z, w} and b ∈ H ∪ {x} ∪ I ∪ {y} ∪ J ∪
{z} ∪ K ∪ {w}.
To complete the proof, recall that that ω · 4 → (ω · 2, 3)2 . It follows that there
is either a blue triangle, in which case we are done, or a red-homogeneous subset
M ⊆ H ∪ I ∪ J ∪ K of order type ω · 2. Let S be the initial segment of M
of order type ω and T = M \ S, and let s = sup(S) and t = sup(T ) (so that
s, t ∈ {x, y, z, w}). Then S ∪ {s} ∪ T ∪ {t} is a red-homogeneous closed copy of
ω · 2 + 1, and we are done.
This argument easily adapts to give the following.
Proposition 6.6. Rcl (ω · 2, 3) ≤ ω 4 · 2.
Proof. The proof follows the argument for Lemma 6.5, but is simpler. The
point is that at each step of the process we only need to split into two rather than
three sets, and there is no need to use ultrafilters. In more detail, let
• β1 = P cl (ω · 2, ω) = ω 2 ,
• β2 = P cl (ω · 2, ω 2 + 1 + β1 ) = ω 3 · 2,
• β3 = P cl (ω · 2, ω 2 + 1 + β2 ) = ω 4 · 2, and
• β = ω 2 + 1 + β3 = ω 4 · 2.
Fix a coloring c : [β]2 → {red,blue}. Among the first ω 2 + 1 elements of β we may
assume we have a red-homogeneous closed copy H1 ∪ {x1 } of ω + 1, with h < x for
all h ∈ H1 . Split the ordinals in β above x into two sets:
• A1 = {a > x | c(x, a) = blue}, and
• A2 = {a > x | c(x, a) = red}.
As usual, we may assume A1 is red-homogeneous and does not contain a closed
copy of ω · 2, so that A2 contains a closed copy X of ω 2 + 1 + β2 . As in the proof
of Lemma 6.5, iterating the argument eventually produces a closed copy
H = H1 ∪ {x1 } ∪ H2 ∪ {x2 } ∪ H3 ∪ {x3 } ∪ H4
of ω · 4 with each Hi of type ω and h1 < x1 < h2 < x2 < h3 < x3 < h4 for all
hi ∈ Hi , i ∈ {1, 2, 3, 4}, such that Hi ∪ {xi } is red-homogeneous for all i ≤ 3, and
c(xi , h) = red for all i ≤ 3 and all h > xi . Now use that ω · 4 → (ω · 2, 3)2 to find
either a blue-homogeneous triangle in H, or some i < j in {1, 2, 3, 4}, and infinite
sets Hi ⊆ Hi , Hj ⊆ Hj such that Hi ∪ Hj is a red-homogeneous copy of ω · 2, in
which case Hi ∪ {xi } ∪ Hj is a red-homogeneous closed copy of ω · 2.
We now indicate the modifications to the argument of Lemma 6.5 that are
required to obtain the general result.
Proof of Theorem 6.1. The proof is by induction on k. The case k = 1 is
trivial. For the inductive step, suppose k ≥ 2. We can now use the argument of
Lemma 6.5 with just a couple of changes.
Firstly, we require ω k + 1 points in order to be able to assume that we have a
red-homogeneous closed copy of ω + 1.
Secondly, it is no longer enough for A1 or A2 to contain a closed copy of ω·m+1,
but it is enough for one of them to contain a closed copy of Rcl (ω · m + 1, k), which
we have an upper bound on by the inductive hypothesis (and likewise for B1 and
B2 , and so on).
Finally, in order to complete the proof using Theorem 6.4, we must iterate the
∗
argument R(Km , Lk+1 ) times.
This argument demonstrates that Rcl (ω ·m+1, k +1) ≤ ω k +1+βR(Km ∗ ,L
k+1 )−1
,
where β0 = 0 and βi = P (R (ω · m + 1, k), R (ω · m + 1, k), ω + 1 + βi−1 ) for
cl cl cl k
∗
i ∈ {1, 2, . . . , R(Km , Lk+1 ) − 1}.
To obtain upper bounds for Rcl (ω · m + n, k) for all finite k, m and n, one can
again use any of the three inequalities from Propositions 5.1 and 5.5, which may
give better bounds than simply using the bound on Rcl (ω · (m + 1) + 1, k) given by
Theorem 6.1.
Remark 6.7. It is perhaps worth pointing out that the classical version of this
problem, the precise computation of the numbers R(ω · m + n, k) for finite k, m
and n, was solved more than 40 years ago. It proceeds by reducing the problem
to a question about finite graphs that can be effectively, albeit unfeasibly, solved
with a computer. This was announced without proof in [HS69a] and [HS69b].
See [Cai15] for further details.
Remark 6.8. At the cost of a somewhat more cumbersome approach, we may
eliminate the use of the non-principal ultrafilter and any appeal to the axiom of
choice throughout the paper. Rather than presenting these versions of the proofs,
we mention a simple and well-known absoluteness argument ensuring that choice
is indeed not needed.
We present the argument in the context of Lemma 6.5; the same approach re-
moves in all proofs the need to use choice to get access to a non-principal ultrafilter.
Work in ZF. With β as in the proof of Lemma 6.5, consider a coloring c : [β]2 → 2,
and note that L[c] is a model of choice, and that the definitions of β and of ho-
mogeneous closed copies of ω · 2 + 1 and 3 are absolute between the universe of
sets and this inner model. Since L[c] is a model of choice, the argument of Lemma
6.5 gives us a homogeneous set as required, with the additional information that it
belongs to L[c]. Similar arguments verify that no use of choice is needed for (1)–(5)
in Theorem 3.2 or in the relevant portions of [Hil16]).
7. The ordinal ω 2
In this section we adapt the argument from the previous section to prove the
following result.
Theorem 7.1. If k is a positive integer, then ω ω →cl (ω 2 , k).
Since ω 2 is order-reinforcing, it follows that Rtop (ω 2 , k) = Rcl (ω 2 , k) ≤ ω ω .
The ordinal ω ω appears essentially because P cl (ω ω )m = ω ω for all finite m,
allowing us to iterate the argument of Lemma 6.5 infinitely many times.
Again we require a classical ordinal Ramsey result. This one is due to Specker
[Spe56] (see also [HS69b]).
Theorem 7.2 (Specker). If k is a positive integer, then ω 2 → (ω 2 , k)2 .
Proof of Theorem 7.1. The proof is by induction on k. The cases k ≤ 2
are trivial. For the inductive step, suppose k ≥ 3. Fix a non-principal ultrafilter U
on ω and let c : [ω ω ]2 → {red, blue} be a coloring.
We argue in much the same way as in the proof of Lemma 6.5. Among the first
ω k−1 + 1 elements of ω ω , we may assume that we have a red-homogeneous closed
copy of ω + 1. Let x0 be its largest point and let H0 be its subset of order type
ω. Write the set of points ≥ ω k−1 + 1 as a disjoint union A1 ∪ A2 ∪ A3 as in the
proof of Lemma 6.5. If either A1 or A2 contains a closed copy of ω ω , then by the
inductive hypothesis we may assume it contains a blue-homogeneous set of k − 1
points, and we are done by the definitions of A1 and A2 . But P cl (ω ω )3 = ω ω , so
we may assume that A3 contains a closed copy X1 of ω ω .
We can now work within X1 and iterate this argument infinitely many times
to obtain a closed set
H0 ∪ {x0 } ∪ H1 ∪ {x1 } ∪ . . .
of order type ω 2 . For each i ∈ ω, write Hi = {hi,n | n ∈ ω} with hi,0 < hi,1 < . . . .
By construction, for all i, j ∈ ω with i < j,
(1) c({xi , xj }) = red,
(2) c({xi , hj,n }) = red for all n ∈ ω and
(3) {n ∈ ω | c({hi,n , xj }) = red} ∈ U.
We would like to be able to assume that condition 3 can be strengthened to
c({hi,n , xj }) = red for all n ∈ ω by using the ultrafilter to pass to a subset. However,
for each i there are infinitely many j > i, and we can only use the ultrafilter to deal
with finitely many of these.
In order to overcome this difficulty, we extend the previous argument in two
ways. The first new idea is to modify our construction so as to ensure that for all
i, j ∈ ω with i < j, we also have
(4) c({hi,n , xj }) = red for all n < j.
We can achieve this by modifying the construction of Xj (the closed copy of ω ω
from which we extracted Hj ∪ {xj }). Explicitly, we now include in our disjoint
union one additional set for each pair (i, n) with i, n < j, which contains the points
y that remain with c({hi,n , y}) = blue. We then extract Xj using the fact that
P cl (ω ω )j 2 +3 = ω ω .
This extra condition is enough for us to continue. The second new idea is to
pass to a subset of the form
H = {hi,n | i ∈ I, n ∈ N } ∪ {xi | i ∈ I}
for some infinite I, N ⊆ ω, and to build up I and N using a back-and-forth argu-

ment. To do this, start with I = N = ∅ and add an element to I and an element to
N alternately in such a way that c({hi,n , xj }) = red whenever n ∈ N and i, j ∈ I
with i < j. Condition 4 ensures that we can always add a new element to I simply
by taking it to be larger than all other members of I and all members of N so far.
Meanwhile, condition 3 ensures there is always some U ∈ U from which we may
choose any member to add to N : at each stage, there are only finitely many new
conditions and so our ultrafilter is enough.
Then c({a, b}) = red whenever a ∈ H and b ∈ H ∩ {xi | i ∈ ω}. Finally,
by Theorem 7.2 we may assume that there is a red-homogeneous subset M ⊆
H \ {xi | i ∈ ω} of order type ω 2 , and then the topological closure of M in H is a
red-homogeneous closed copy of ω 2 .
Remark 7.3. We have organized this argument in such a way that the reader
may readily verify the following. For any positive integer k and any coloring c :
[ω ω ]2 → {red,blue}, there is either a blue-homogeneous set of k points, or a red-
homogeneous closed copy of an ordinal larger than ω 2 , or a red-homogeneous closed
copy of ω 2 that is moreover cofinal in ω ω . This strengthening of Theorem 7.1 will
be useful in Section 9.
8. The anti-tree partial ordering on ordinals

The techniques from the last few sections enable us to reach ω 2 , but do not seem
to get us any further without cumbersome machinery. In this section we introduce
a new approach, which provides a helpful perspective and ultimately suggests not
just how to get past ω 2 , but even how to reach our main theorem. Here, we use
this approach to prove the following result.
Theorem 8.1. ω 2 · 3 ≤ Rtop (ω · 2, 3) ≤ ω 3 · 100.
It is more transparent to describe this new approach in terms of a new partial
ordering on ordinals. A variant of this ordering was independently considered by
Piña in [Pn15], who identified countable ordinals with families of finite sets. Read-
ers who are familiar with that work may find it helpful to note that for ordinals
less than ω ω , our new relation ≤∗ coincides with the superset relation ⊇ under that
identification. (Note that none of the results we prove here are used outside of this
section.)
Definition 8.2. Let α and β be ordinals. If β > 0, then write β = η + ω γ
with η a multiple of ω γ . Then we write α <∗ β to mean that β > 0 and α = η + ζ
for some 0 < ζ < ω γ . We write α ∗ β to mean that α <∗ β and there is no ordinal
δ with α <∗ δ <∗ β.
Equivalently, α <∗ β if and only if β = α + ω γ for some γ > CB(α), and α ∗ β
if and only if β = α+ω CB(α)+1 . For example, ω 3 +ω <∗ ω 3 ·2 and ω 3 ·2+1 <∗ ω 3 ·3,
but ω 3 + ω <∗ ω 3 · 3 and ω 3 · 2 + 1 <∗ ω 3 · 2.
Here are some simple properties of these relations.
(1) <∗ is a strict partial ordering.
(2) If α <∗ β then α < β and CB(α) < CB(β).
(3) If α ∗ β then CB(β) = CB(α) + 1.
(4) The class of all ordinals forms an “anti-tree” under the relation <∗ in
the sense that for any ordinal α, the class of ordinals β with α <∗ β is
well-ordered by <∗ .
By property 4, if k is a positive integer, then ω k + 1 forms a tree under the
relation >∗ . It fact it is what we will call a perfect ℵ0 -tree of height k.
Definition 8.3. Let k be a positive integer and let X be a single-rooted tree.
We say that x ∈ X has height k to mean that x has exactly k predecessors. We say
that x ∈ X is a leaf of X to mean that x has no immediate successors, and denote
the set of leaves of X by (X). We say that X is a perfect ℵ0 -tree of height k to
mean that every non-leaf of X has ℵ0 immediate successors and every leaf of X has
height k.
Let X be a perfect ℵ0 -tree of height k. We say that a subset Y ⊆ X is a full
subtree of X to mean that Y is a perfect ℵ0 -tree of height k under the induced
relation.
Note that if X is a full subtree of ω k + 1, then X ∼
= ω k + 1. Note also that full
subtrees are determined by their leaves.
Figure 1. A perfect ℵ0 -tree of height 2, corresponding to the

ordinal ω 2 + 1
Here is a simple result about colorings of perfect ℵ0 -trees of height k. The

proof essentially amounts to k applications of the infinite pigeonhole principle.
Lemma 8.4. Let k be a positive integer, let X be a perfect ℵ0 -tree of height k
and let c : (X) → {red, blue} be a coloring. Then there exists a full subtree Y of
X such that (Y ) is monochromatic.
Proof. The proof is by induction on k. The case k = 1 is simply the infinite
pigeonhole principle, so assume k > 1. Let Z be the set of elements of X of height
at most k − 1, so Z is a perfect ℵ0 -tree of height k − 1. Then for each z ∈ (Z), by
the infinite pigeonhole principle again there exists d(z) ∈ {red, blue} and an infinite
subset Yz of the successors of z such that c(x) = d(z) for all x ∈ Yz . This defines
a coloring d : (Z) → {red, blue}, so by the inductive hypothesis there exists a full
subtree W of Z such that (W ) is monochromatic for d. Finally let

Y =W ∪ Yw .
w∈(W )
Then Y is as required.
Recall Theorem 4.1, which states that if k is a positive integer, then Rtop (ω +
1, k + 1) = Rcl (ω + 1, k + 1) ≥ ω k + 1. To illustrate the relevance of these notions,
we now provide a second proof of this result. This is also the proof that we will
mirror when the result is generalized in Theorem 9.1.
The crux of the proof is the following result, which says that any {red, blue}-
coloring of ω k + 1 avoiding both a red-homogeneous topological copy of ω + 1 and a
blue-homogeneous topological copy of ω is in some sense similar to the k + 1-partite
{red, blue}-coloring that falls out of the proofs of Propositions 3.1 and 3.4.
Lemma 8.5. Let k be a positive integer and let c : [ω k + 1]2 → {red, blue} be a
coloring. Suppose that
(a) there is no red-homogeneous topological copy of ω + 1, and
(b) there is no blue-homogeneous topological copy of ω.
Under these assumptions, there is a full subtree X of ω k + 1 such that for all
x, y, z ∈ X:
(1) if x ∗ z and y ∗ z then c({x, y}) = red; and
(2) if x <∗ y then c({x, y}) = blue.
The proof makes use of Lemma 8.4.
Proof. The proof is by induction on k.
For the base case, suppose k = 1. By the infinite Ramsey theorem there
exists an infinite homogeneous subset Y ⊆ ω. By condition ( b) this must be
red-homogeneous. Now by the infinite pigeonhole principle there must exist i ∈
{red, blue} and an infinite subset Z ⊆ Y such that c({x, ω}) = i for all x ∈ Z. By
condition ( a) we must have i = blue. Then Z ∪ {ω} is a full subtree of ω + 1 with
the required properties.
For the inductive step, suppose
k > 1. First apply
thek−1
inductive hypothesis to
obtain a full subtree Ym of ω k · m + 1, ω k · (m + 1) ∼= ω + 1 for each m ∈ ω,

and let Y = m∈ω Ym ∪ {ω }. Then use the inductive hypothesis again to obtain
k
a full subtree Z of Y \ (Y ) ∼

= ω k−1 + 1, and let
W = Z ∪ {y ∈ (Y ) | y ∗ z for some z ∈ Z}.
By our uses of the inductive hypothesis, conditions ( 1) and ( 2) hold whenever
x, y, z ∈ Ym for some m ∈ ω or x, y, z ∈ Z. Thus it is sufficient to find a full subtree
X of W such that c({x, ω k }) = blue for all x ∈ (X). To do this, define a coloring

c : (W ) → {red, blue} by c(x) = c({x, ω k }). Then apply Lemma 8.4 to obtain
i ∈ {red, blue} and a full subtree X of W such that c({x, ω k }) = i for all x ∈ (X).
Now let V be a cofinal subset of (X) of order type ω. By the infinite Ramsey
theorem there exists an infinite homogeneous subset U ⊆ V , which by condition
(b) must be red-homogeneous. But then U ∪ {ω k } is a topological copy of ω + 1,
so by condition (a) we must have i = blue, and we are done.
Theorem 4.1 now follows easily.
Second proof of Theorem 4.1. As in the first proof, Rtop (ω + 1, k + 1) =
R (ω + 1, k + 1) ≥ ω k + 1.
cl
To see that ω k + 1 →top (ω + 1, k + 1)2 , let c : [ω k + 1]2 → {red, blue} be

a coloring. If there is a red-homogeneous topological copy of ω + 1 or a blue-
homogeneous topological copy of ω, then we are done. Otherwise, choose X ⊆ ω k +1
as in Lemma 8.5. Then any branch (i.e., maximal chain under >∗ ) of X forms a
blue-homogeneous set of k + 1 points.
We conclude this section by proving our bounds on Rtop (ω · 2, 3). Before doing
this, we remark that it is crucial that we consider here the topological rather than
the closed Ramsey number. Since ω + n ∼ = ω + 1 for every positive integer n,
from a topological perspective, ω · 2 is the simplest ordinal space larger than ω + 1.
Moreover, there are sets of ordinals containing a topological copy of ω · 2 but not
even a closed copy of ω + 2, such as (ω · 2 + 1) \ {ω}. Accordingly, we have only been
able to apply the technique we present here to this simplest of cases. Nonetheless,
it may still be possible to adapt this technique to obtain upper bounds on closed
(as well as topological) Ramsey numbers.
We begin by proving the lower bound. Recall from Fact 3.3 that P top (ω · 2)2 =
ω · 2. Thus we have indeed improved upon the lower bound given by Proposition
2
3.1. As with the lower bound of Lemma 5.3, we provide a simple coloring based on
a small finite graph.
Lemma 8.6. Rtop (ω · 2, 3) ≥ ω 2 · 3.
Proof. Since any ordinal less than ω 2 · 3 is homeomorphic to a subspace of
ω · 2 + 1, it is sufficient to prove that ω 2 · 2 + 1 →top (ω · 2, 3)2 . To see this, let G
2
be the graph represented by the following diagram.
{ω 2 } {ω 2 · 2}
{ω · (n + 1) | n ∈ ω} {ω 2 + ω · (n + 1) | n ∈ ω}
{0} ∪ {x + 1 | x ∈ ω 2 } {ω 2 + x + 1 | x ∈ ω 2 }
Now define a coloring c : [ω 2 ·2+1]2 → {red, blue} by setting c({x, y}) = blue if and
only if x and y lie in distinct, adjacent vertices of G. First note that there is no blue
triangle since G is triangle-free. To see that there is no red-homogeneous topological
copy of ω · 2, first note that every vertex of G is discrete, and moreover the union of
any vertex from the left half of G and any vertex the right half of G is also discrete.
Therefore the only maximal red-homogeneous subspaces that are not discrete are
{ω · (n + 1) | n ∈ ω} ∪ {ω 2 } ∪ {ω 2 · 2} and {ω 2 + ω · (n + 1) | n ∈ ω} ∪ {ω 2 } ∪ {ω 2 · 2},
neither of which contains a topological copy of ω · 2.
Our upper bound makes use of several classical ordinal Ramsey results.
Firstly, we use the special case of Theorem 7.2 that ω 2 → (ω 2 , 3)2 .
Secondly, we use Theorem 6.4. In fact, we essentially prove that Rtop (ω · 2, 3) ≤
∗ ∗
ω 3 · R(K10 , L3 ). The best known upper bound on R(K10 , L3 ) is due to Larson and
Mitchell [LM97].
Theorem 8.7 (Larson-Mitchell). If n > 1 is a positive integer, then R(Kn∗ , L3 ) ≤
2
n .
∗
In particular, R(K10 , L3 ) ≤ 100 and hence ω · 100 → (ω · 10, 3)2 by Theorem
6.4.
Finally, we use the following result, which was claimed without proof by Had-
dad and Sabbagh [HS69b] and has since been proved independently by Weinert
[Wei14].
Theorem 8.8 (Haddad-Sabbagh; Weinert). R(ω 2 · 2, 3) = ω 2 · 10.

We are now ready to prove our upper bound. The first part of the proof is the
following analogue of Lemma 8.5. The proof uses in an essential manner that we
are looking for a topological rather than a closed copy of ω · 2.
Lemma 8.9. Let c : [ω 2 + 1]2 → {red, blue} be a coloring. Suppose that
(a) there is no red-homogeneous topological copy of ω · 2, and
(b) there is no blue triangle.
Under these assumptions, there is a full subtree X of ω 2 + 1 such that for all
x, y ∈ X:
(1) if CB(x) = CB(y) then c({x, y}) = red;
(2) if CB(x) = 0 then c({x, ω 2 }) = blue; and
(3) if CB(x) = 1 then c({x, ω 2 }) = red.
Proof. First note that (ω 2 + 1) \ (ω 2 + 1) has order type ω 2 , so by condition
( b) it has a red-homogeneous subset W of order type ω 2 , since ω 2 → (ω 2 , 3)2 . Let
Y0 be a full subtree of ω 2 + 1 with (Y0 ) ⊆ W . By applying the infinite Ramsey
theorem to Y0 \ Y0 , we may similarly pass to a full subtree Y1 of Y0 such that
(2)
Y1 \ Y1 is red-homogeneous. Thus Y1 satisfies condition ( 1).

(2)
Next apply Lemma 8.4 to obtain i ∈ {red, blue} and a full subtree Z of Y1
such that c({x, ω 2 }) = i for all x ∈ (Z). If i = red, then (Z) ∪ {ω 2 } would be
red-homogeneous, so by condition ( a), i = blue and Z satisfies condition ( 2).
Finally apply the infinite pigeonhole principle to Z \Z (2) to obtain j ∈ {red, blue}
and a full subtree X of Z such that c({x, ω 2 }) = j for all x ∈ X with CB(x) = 1. If
j = blue, then by condition ( b) we would have c({x, y}) = red for all x, y ∈ X with
CB(x) = 0 and CB(y) = 1, whence X \ {ω 2 } would be red-homogeneous. Hence by
condition ( a), j = red and X is as required.
We can now complete the proof of our upper bound and hence of Theorem 8.1.
Proof of Theorem 8.1. By Lemma 8.6, it remains only to prove that

Rtop (ω · 2, 3) ≤ ω 3 · 100.
Let X = ω 3 · 100, let c : [X]2 → {red, blue} be a coloring and suppose by
contradiction that there is no red-homogeneous topological copy of ω · 2 and no
blue triangle.
First note that X (2) \ X (3) has order type ω · 100, so it has a red-homogeneous
subset U of order type ω · 10, since ω · 100 → (ω · 10, 3)2 . Next let V = {x ∈
X | x ∗ y for some y ∈ U }. Note that V has order type ω 2 · 10, and so V has
a red-homogeneous subset W of order type ω 2 · 2, since ω 2 · 10 → (ω 2 · 2, 3)2 by
Theorem 8.8. Finally, let
Y = cl(W ) ∪ {x ∈ X | x ∗ y for some y ∈ W },
where cl denotes the topological closure operation. Replacing Y with Y \{max Y } if
necessary, we may then assume that Y ∼ = ω 3 ·2, and by construction both Y (2) \Y (3)

and Y \ Y (2)
are red-homogeneous.
that Y = ω · 2. By applying Lemma 8.9 to
3
Assume for notational convenience
the interval ω · α + 1, ω · (α + 1) for each α ∈ ω · 2, we may assume that
2 2
c({ω 2 · α + ω · (n + 1), ω 2 · (α + 1)}) = red

for all n ∈ ω. By applying Lemma 8.9 to (ω 3 + 1) , we may then assume that
c({ω 2 · (α + 1), ω 3 }) = red and c({ω 2 · α + ω · (n + 1), ω 3 }) = blue for all α, n ∈ ω.
Finally by applying the infinite pigeonhole principle to {ω 2 · (α + 1) | α ∈
[ω, ω · 2)}, we may assume that
c({ω 3 , ω 2 · (α + 1)}) = i
for all α ∈ [ω, ω · 2), where i ∈ {red, blue}, and then by applying the infinite
pigeonhole principle to {ω · (n + 1) | n ∈ ω}, we may assume that
c({ω · (n + 1), ω 3 + ω 2 }) = j,
where j ∈ {red, blue}. Now if i = red, then (ω 3 · 2)(2) would be a red-homogeneous
topological copy of ω · 2, and if j = red, then
{ω · (n + 1) | n ∈ ω} ∪ {ω 3 + ω · (n + 1) | n ∈ ω} ∪ {ω 3 + ω 2 }
would be a red-homogeneous topological copy of ω · 2. So i = j = blue. But then
{ω, ω 3 , ω 3 + ω 2 } is a blue triangle.
Remark 8.10. Contrast this upper bound with Rcl (ω · 2, 3) ≤ ω 4 · 2 (see

Proposition 6.6). Very recently, Omer Mermelstein has produced a draft [Mer17]
where, using a careful topological analysis building in part on the ideas from this
section, he obtains that Rcl (ω · 2, 3) = ω 3 · 2, in particular improving our upper
bounds in both the topological and the closed case.
9. The ordinal ω 2 + 1
We now use our earlier result on ω 2 together with some of the ideas from the
previous section to obtain upper bounds for ω 2 +1. We deduce these from Theorem
7.1 and the following general result.
Theorem 9.1. Let α and β be countable ordinals with β > 0, let k be a positive
integer, and suppose they satisfy a “cofinal version” of
α
ω ω →cl (ω β , k + 2)2 .
Then
α
·(k+1)
ωω + 1 →cl (ω β + 1, k + 2)2 .
α
Moreover, if ω ω > ω β , then in fact
α
·k
ωω + 1 →cl (ω β + 1, k + 2)2 .
The cofinal version of the partition relation requires that for every coloring c :
ω α 2
ω → {red, blue},
• there is a blue-homogeneous set of k + 2 points, or
α
• there is a red-homogeneous closed copy of ω β that is cofinal in ω ω , or
• there is already a red-homogeneous closed copy of ω β + 1.
Before providing the proof, we first deduce our upper bounds for ω 2 + 1. Since
ω 2 + 1 is order-reinforcing, it follows that Rtop (ω 2 + 1, k + 2) = Rcl (ω 2 + 1, k + 2) ≤
ω ω·k + 1.
Corollary 9.2. If k is a positive integer, then ω ω·k + 1 →cl (ω 2 + 1, k + 2)2 .
Proof. By Theorem 9.1, since ω ω > ω 2 it is enough to prove the cofinal

version of ω ω →cl (ω 2 , k + 2)2 . The usual version is precisely Theorem 7.1, and
the cofinal version is easily obtained from the same proof, as indicated in Remark
7.3.
Observe that by applying Ramsey’s theorem instead of Theorem 7.1, one ob-
tains yet another proof of Theorem 4.1 from the case α = 0. Indeed, our proof
of Theorem 9.1 is similar to our second proof of that result, though we do not
explicitly use any of our results on the anti-tree partial ordering.
The bulk of the proof of Theorem 9.1 is in the following result, which is our
analogue of Lemma 8.5. The proof makes detailed use of the topological structure
of countable ordinals. In particular, we use two arguments due to Weiss from the
proof of [Bau86, Theorem 2.3]. It may be helpful for the reader to first study that
proof.
Lemma 9.3. Let α, β and k be as in Theorem 9.1. Let l be a positive integer
α 2
and let c : ω ω ·l + 1 → {red, blue} be a coloring. Suppose that
(1) there is no red-homogeneous closed copy of ω β + 1, and
(2) there is no blue-homogeneous set of k + 2 points.
Under these assumptions, there exists a cofinal subset X ⊆ ω ω ·l such that X is a
α
closed copy of ω ω ·l and c({x, ω ω ·l }) = blue for all x ∈ X.

α α
Proof. The proof is by induction on l.

α α α
For the case l = 1, since ω ω →cl (ω ω )12 , there exists X ⊆ ω ω and i ∈
α α
{red, blue} such that X is a closed copy of ω ω (and therefore X is cofinal in ω ω )
α
and c({x, ω ω }) = i for all x ∈ X. Suppose for contradiction that i = red. By
our assumptions together with the definition of the cofinal version of the partition
relation, there exists a cofinal subset Y ⊆ X such that Y is a closed copy of ω β
and [Y ]2 ⊆ c−1 ({red}). But then Y ∪ {ω ω } is a red-homogeneous closed copy of
α
ω β + 1, contrary to assumption 1. Hence i = blue and we are done.

For the inductive step, suppose l > 1. Let
α
Z = ω ω · γ | γ ∈ ω ω ·(l−1) \ {0} ,
α
so Z is a closed copy of ω ω ·(l−1) . By the inductive hypothesis, there exists a cofinal

α
subset Y ⊆ Z such that Y is a closed copy of ω ω ·(l−1) and c({x, ω ω ·l }) = blue

α α
for all x ∈ Y . Write Y = {yδ | δ ∈ ω ω ·(l−1) } in increasing order. Then by Weiss’s

α
lemma [Bau86, Lemma 2.6], for each δ ∈ ω ω ·(l−1) there exists a cofinal subset
α
α
Zδ ⊆ (yδ , yδ+1 ) such that Zδ is a closed copy of ω ω .
Now since ω ω →cl (ω ω )12 , for each δ ∈ ω ω ·(l−1) there exists Xδ ⊆ Zδ and
α α α
α
iδ ∈ {red, blue} such that Xδ is a closed copy of ω ω (and therefore Xδ is cofinal in
Zδ ) and c({x, ω ω ·l }) = iδ for all x ∈ Xδ . Recall now that ω ω ·(l−1) → (ω ω ·(l−1) )12
α α α
since ω ω ·(l−1) is a power of ω. It follows that there exists S ⊆ ω ω ·(l−1) of order

α α
type ω ω ·(l−1) and i ∈ {red, blue} such that iδ = i for all δ ∈ S.

α
Suppose for contradiction that i = red. We now use an argument from the proof
of [Bau86, Theorem 2.3]. Let (δm )m∈ω be a strictly increasing cofinal sequence
from S, and let (ηm )m∈ω be a strictly increasing cofinal sequence from ω α (or let
ηm = 0 for all m ∈ ω if α = 0). For each m∈ ω, pick Wm ⊆ Xδm such that
Wm is a closed copy of ω ηm + 1, and let W = m∈ω Wm . Note that W is cofinal
in ω ω ·l , W is a closed copy of ω ω and c({x, ω ω ·l }) = red for all x ∈ W . By
α α α
our assumptions together with the definition of the cofinal version of the partition
relation, there exists a cofinal subset V ⊆ W such that V is a closed copy of ω β
and [V ]2 ⊆ c−1 ({red}). But then V ∪ {ω ω ·l } is a red-homogeneous closed copy of
α
β
ω + 1, contrary to assumption 1.
Therefore i = blue. Finally, let

X= Xδ ∪ cl({yδ+1 | δ ∈ S}),
δ∈S
where cl denotes the topological closure operation. Then the set X is as required.

We may now deduce Theorem 9.1 in the much same way that we deduced
Theorem 4.1 from Lemma 8.5.
α
Proof of Theorem 9.1. First assume that ω ω > ω β . We prove by induc-
tion on l that for all l ∈ {1, 2, . . . , k},
α
·l
ωω + 1 →cl (ω β + 1, l + 2)2 .
In every case, if either of the assumptions in Lemma 9.3 does not hold, then
we are done since l ≤ k. We may therefore choose X as in Lemma 9.3.
For the base case l = 1, to avoid a blue triangle, X must be red-homogeneous.
α
But X contains a closed copy of ω β + 1 since ω ω ≥ ω β + 1, and so we are done.
For the inductive step, suppose l ≥ 2. Then X has a closed copy Y of ω ω ·(l−1) +
α
1. By the inductive hypothesis, either Y contains a red-homogeneous closed copy

of ω β + 1, in which case we are done, or Y contains a blue-homogeneous set Z of
l + 1 points. But in that case Z ∪ {ω ω ·l } is a blue-homogeneous set of l + 2 points,
α
and we are done.

α
Finally, if we cannot assume that ω ω > ω β , then the base case breaks down.
α
However, we may instead use the base case ω ω + 1 →cl (ω β + 1, 2)2 , which follows
α
from the fact that ω ω ≥ ω β . The inductive step is then identical.
10. The weak topological Erdős-Milner theorem

Finally we reach our main result, which demonstrates that Rtop (α, k) and
cl
R (α, k) are countable for all countable α and all finite k.
This is a topological version of a classical result due to Erdős and Milner
[EM72]. Before stating it, we first provide a simplified proof of the classical version.
Theorem 10.1 (Weak Erdős-Milner). Let α and β be countable nonzero ordi-
nals, and let k > 1 be a positive integer. If
ω α → (ω 1+β , k)2 ,
then
ω α+β → (ω 1+β , k + 1)2 .
Since trivially ω 1+α → (ω 1+α , 2)2 , it follows by induction on k that R(ω 1+α , k+
1) ≤ ω 1+α·k for all countable α and finite k. In fact, Erdős and Milner proved a
stronger version of the above theorem, in which k + 1 is replaced by 2k, implying
that R(ω 1+α , 2k ) ≤ ω 1+α·k . This is why we use the adjective weak here.
Our proof is essentially a simplified version of the original, which can be found
in [Wil77, Theorem 7.2.10]. The basic idea is to write ω α+β as a sequence of ω β
intervals, each of order type ω α , to list these intervals in type ω, and to recursively
build up a red-homogeneous copy of ω β consisting of one element from each interval.

This would achieve the above theorem with 1 + β weakened to β. To obtain a copy
of ω 1+β , we simply choose infinitely many elements from each interval instead of
just one. In the proof we use the fact that ω α → (ω α )1m for all finite m.
2
Proof. Let c : ω α+β → {red, blue} be a coloring.
For each x ∈ ω β , let
Ix = [ω α · x, ω α · (x + 1)) ,
so Ix has order type ω α . Let (xn )n∈ω be a sequence of points from ω β in which
every member of ω β appears infinitely many times. We attempt to inductively
build a red-homogeneous set A = {an | n ∈ ω} of order-type ω 1+β with an ∈ Ixn
for every n ∈ ω.
Suppose that we have chosen a1 , a2 , . . . , am−1 for some m ∈ ω. Let
P = {an | n ∈ {1, 2, . . . , m − 1}, an ∈ Ixm },
and let
Ixm , if P = ∅
J=
Ixm \ [0, max P ], if P = ∅,
so J has order type ω α . For each n ∈ {1, 2, . . . , m − 1} let
Jn = {a ∈ J | c({an , a}) = blue}.
m−1
If n=1 Jn = J, then since ω α → (ω α )1m−1 , Jn has order type ω α for some n ∈
{1, 2, . . . , m − 1}. Then since ω α → (ω 1+β , k)2 , Jn either has a red-homogeneous
subset of order type ω 1+β , in which case we are done, or a blue-homogeneous subset
B of k points, in which case B ∪ {an } is a blue-homogeneous set of k + 1 points,
m−1
and we are done. Thus we may assume that J \ n=1 Jn = ∅, and choose am to
be any member of this set.
Clearly the resulting set A is red-homogeneous. To see that it has order type
ω 1+β , simply observe that by choice of (xn )n∈ω , A ∩ Ix is infinite for all x ∈ ω β ,
and that by choice of J, A ∩ Ix in fact has order type ω for all x ∈ ω β .
Here is our topological version of the weak Erdős-Milner theorem.
Theorem 10.2 (The weak topological Erdős-Milner theorem). Let α and β be
countable nonzero ordinals, and let k > 1 be a positive integer. If
α
ω ω →top (ω β , k)2 ,
then
α
·β
ωω →top (ω β , k + 1)2 .
In particular, as we shall deduce in Corollary 10.5, Rtop (α, k) and Rcl (α, k) are
countable whenever α is a countable ordinal and k is a positive integer.
Our proof follows the same outline as our proof of the classical version, except
that we use intervals in the sense of the anti-tree partial ordering rather than in the
usual sense. Furthermore, rather than constructing a closed copy of ω β directly,
we instead construct a larger set and then thin it out. As in the previous section,
we make detailed use of the structure of countable ordinals, including an argument
from the proof of [Bau86, Lemma 2.6]. Note that the proof does not directly use
any of our previous results.
α 2
Proof. Let c : ω ω ·β → {red, blue} be a coloring.
α
First of all, fix a strictly increasing cofinal sequence (γn )n∈ω from ω ω .
Define an indexing set of pairs

S = (x, y) | x ∈ β, ω ω ·(x+1) · y ∈ ω ω ·β ,
α α
and for each (x, y) ∈ S, let

α
X(x,y) = ω ω ·(x+1) · y + ω ω ·x · z | z ∈ ω ω \ {0} ,
α α
α
so X(x,y) is a closed copy of ω ω . Let (xn , yn )n∈ω be a sequence of pairs from S in
which every member of S appears infinitely many times. We attempt to inductively
build a red-homogeneous set A = {an | n ∈ ω}, which will contain a closed copy of
ω β , with an ∈ X(xn ,yn ) for every n ∈ ω.
Suppose that we have chosen a1 , a2 , . . . , am−1 for some m ∈ ω. Let
! "
P = an | n ∈ {1, 2, . . . , m − 1}, an ∈ X(xm ,ym ) ,
let α
Q = P ∪ ω ω ·(xm +1) · ym + ω ω ·xm · γ|P | ,
α
and let
Y = X(xm ,ym ) \ [0, max Q],
ωα
so Y is a closed copy of ω . For each n ∈ {1, 2, . . . , m − 1} let
Yn = {a ∈ Y | c({an , a}) = 1}.
m−1 α α α
If n=1 Yn = Y , then since ω ω →cl (ω ω )1m−1 , Yn contains a closed copy of ω ω for
α
some n ∈ {1, 2, . . . , m − 1}. Then since ω ω →cl (ω β , k)2 , Yn either contains a red-
β
homogeneous closed copy of ω , in which case we are done, or a blue-homogeneous
set B of k points, in which case B ∪ {an } is a blue-homogeneous set of k + 1 points,

and we are done. Thus we may assume that Y \ m−1 n=1 Yn = ∅, and choose am to
be any member of this set.
Clearly the resulting set A is red-homogeneous. To complete the proof, observe
that by choice of (xn , yn )n∈ω , A ∩ X(x,y) is infinite for all (x, y) ∈ S, and that by
choice of Q, A ∩ X(x,y) is a cofinal subset of X(x,y) of order type ω for all (x, y) ∈ S.
We claim that this property is enough to ensure that A contains a closed copy of
ωβ .
To prove the claim, for each δ ∈ [1, β] and each ordinal y with ω ω ·δ · y ∈ ω ω ·β
α α
we find a cofinal subset Cδ,y ⊆ ω ω ·δ · y + 1, ω ω ·δ · (y + 1) such that Cδ,y ⊆ A

α α
and Cδ,y is a closed copy of ω δ ; then Cβ,0 ⊆ A is a closed copy of ω β , as required.

We do this by induction on δ.
First suppose δ ∈ [1, β] is a successor ordinal, say δ = x + 1. Fix an ordinal
y with ω ω ·δ · y ∈ ω ω ·β , and observe that (x, y) ∈ S and that X(x,y) is a cofinal
α α
subset of the interval # α

ω ω ·δ · y + 1, ω ω ·δ · (y + 1) .
α
Recall now that A ∩ X(x,y) is a cofinal subset of X(x,y) of order type ω. Thus if
δ = 1 then we may simply take C1,y = A ∩ X(0,y) , so assume δ > 1, and write
A ∩ X(x,y) = {bn | n ∈ ω} in increasing order. For each n ∈ ω \ {0} we find
a cofinal subset Dn ⊆ (bn−1 , bn ) such that Dn ⊆ A and Dn is a closed copy of
ω x ; then we may take Cδ,y = n∈ω\{0} Dn ∪ {bn }. We do this using an argument
essentially taken from the proof of [Bau86, Lemma 2.6]. Fix n ∈ ω \ {0} and write
bn = ω ω ·(x+1) · y + ω ω ·x · z with z ∈ ω ω \ {0}. Let v = ω ω · y + z so that

α α α α
bn = ω ω ·x · v. If v is a successor ordinal, say v = u + 1, then by the inductive

α
hypothesis we may take Dn = Cx,u . If v is a limit ordinal, then let (um )m∈ω be a
strictly increasing cofinal sequence from v with ω ω ·x · u0 ≥ bn−1 and let (ηm )m∈ω
α
be a strictly increasing cofinal sequence from ω x . By the inductive hypothesis, for

each m ∈ ω we may choose a subset Em ⊆ Cx,um such that Em is a closed copy of
ηm + 1. Then take Dn = m∈ω Em .
Suppose instead δ ∈ [1, β] is a limit ordinal. Fix an ordinal y with ω ω ·δ ·
α
y ∈ ω ω ·β . Let (xn )n∈ω be a strictly increasing cofinal sequence from δ. For

α
each n ∈ ω, let ζn be the ordinal such that δ = xn + 1 + ζn and let yn =

ω α ·ζn
ω ω ·(xn +1) · yn = ω ω ·δ · y + ω ω ·(xn +1) and so Cxn +1,yn ⊆
α α α
ωα ·δ · y + 1.ωα Then
ω
· y + ω ·(xn +1) + 1, ω ω ·δ · y + ω ω ·(xn +1) · 2 . By the inductive hypothesis,
α α
ω
for each n ∈ ω we may choose a subset Dn ⊆ Cxn +1,yn such that Dn is a closed
copy of ω xn + 1. Then take Cδ,y = n∈ω Dn .
Remark 10.3. We expect that as in the original Erdős-Milner theorem, it

should be possible to improve k + 1 to 2k in Theorem 10.2. Indeed, the basic
argument from [Wil77, Theorem 7.2.10] works in the topological setting. The
key technical difficulty appears to be in formulating and proving an appropriate
topological version of statement (1) in that account.
The weak topological Erdős-Milner theorem allows us to obtain upper bounds
for countable indecomposable ordinals. Before describing some of these, we first
observe that by very slightly adapting our argument, we may obtain improved
bounds for other ordinals. Here is a version for ordinals of the form ω β · m + 1.
Theorem 10.4. Let α and β be countable nonzero ordinals, and let k > 1 be a
positive integer. If
α
ω ω →cl (ω β · m + 1, k)2 ,
then
α
·β
ωω · R(m, k + 1) + 1 →top (ω β · m + 1, k + 1)2 .
Proof. Write ω ω ·β · R(m, k + 1) + 1 as a disjoint union M ∪ N , where
α
α
N = ω ω ·β · (y + 1) | y ∈ {0, 1, . . . , R(m, k + 1) − 1} ,
and let c : [M ∪ N ]2 → {red, blue} be a coloring.

First of all, we may assume that N contains a red-homogeneous set of m
points, say a0 , a1 , . . . , am−1 . Now continue as in the proof of Theorem 10.2 and
attempt to build a red-homogeneous set A = {an | n ∈ ω}, only start by including
a0 , a1 , . . . , am−1 , and then work entirely within M .
If we succeed, then the same proof as the one in Theorem 10.2 shows that for
each y ∈ {0, 1, . . . , R(m, k +1)−1}, A\{a0 , a1 , . . . , am−1 } contains a closed
copy Cy
of ω β that is a cofinal subset of the interval ω ω ·β · y + 1, ω ω ·β · (y + 1) . Writing
α α
ai = ω ω ·β · (yi + 1) for each i ∈ {0, 1, . . . , m − 1}, we see that

α

m−1
Cyi ∪ {ai }
i=0
is a red-homogeneous closed copy of ω β · m + 1, as required.

We conclude this section with some explicit upper bounds implied by our re-
sults. It is easy to verify similar results for ordinals of other forms.
Corollary 10.5. Let α be a countable nonzero ordinal and let k, m and n be
positive integers.
α α α·k
(1) Rtop (ω ω , k + 1) = Rcl (ω ω , k + 1) ≤ ωω .
ω α·k
α α ω + 1, if α is infinite
(2) Rtop (ω ω +1, k+1) = Rcl (ω ω +1, k+1) ≤ (n+1)·k−1
ωω + 1, if α = n is finite.
k
·n
(3) Rtop (ω n · m + 1, k + 2) = Rcl (ω n · m + 1, k + 2) ≤ ω ω · R(m, k + 2) + 1.
Proof. First note that all three equalities are immediate since all ordinals
considered are order-reinforcing. It remains to prove the inequalities.
(1) This follows immediately from Theorem 10.2 by induction on k, the case
k = 1 being trivial.
(α+1)·(k−1) α+1
(2) It follows from part 1 that ω ω →cl (ω ω , k)2 (the case k = 1
(α+1)·(k−1) α
again being trivial) and hence ω ω →cl (ω ω + 1, k)2 . Therefore
(α+1)·(k−1)+α α
ωω + 1 →cl (ω ω + 1, k + 1)2 by Theorem 10.4. If α is infinite,
then (α+1)·(k−1)+α = α·k, and if α = n is finite, then (α+1)·(k−1)+α =
(n + 1) · k − 1, as required.
k k
(3) It follows from part 1 that ω ω →cl (ω ω , k + 1)2 and hence ω ω →cl
(ω · m + 1, k + 1) . The result then follows from Theorem 10.4.
n 2

Remark 10.6. For the case in which α = n is a positive integer, if a cofi-

nal version of part 1 holds, then we could use Theorem 9.1 to improve part 2 to
n n·(k+1)
·k
Rcl (ω ω + 1, k + 2) ≤ ω ω + 1.
11. Questions
We close with a few questions. Firstly, there is typically a large gap between
our lower and upper bounds, leaving plenty of room for improvement. In particular,
our general lower bound in Proposition 3.1 is very simple and yet is still our best
bound with the exception of a couple of special cases.
Some further exact equalities could be informative. One of the key reasons why
the classical results detailed in [HS69a, HS69b, HS69c, Cai15, Mil71] are more
precise than our topological results in Sections 5–9 is that for various α < ω ω , the
computation of R(α, k) reduces to a problem in finite combinatorics. Some hint of
a topological version of this appears in the proof of the lower bounds of Lemmas
5.3 and 8.6, though it is far from clear how to obtain an exact equality.
Question 11.1. Is it possible to reduce the computation of Rtop (α, k) or
R (α, k) to finite combinatorial problems, even for α < ω 2 ?
cl
A partition ordinal is an ordinal α satisfying α → (α, 3)2 . As we have seen, ω

and ω 2 are partition ordinals. Other than these, every countable partition ordinal
β β
has the form ω ω , and in the other direction, ω ω is a partition ordinal if β has the
γ γ δ
form ω or ω + ω [Sch10].
Question 11.2. Are there any countable “topological partition” ordinals, sat-
isfying α →top (α, 3)2 , other than ω?
β
By our lower bound (Proposition 3.1), these must all have the form ω ω . Since
every power of ω is order-reinforcing, this question is equivalent to the version
resulting from replacing the topological partition relation by the closed partition
relation.
We expect that a strong version of the topological Erdős-Milner theorem should
hold, that is, we expect that in Theorem 10.2 it should be possible to improve k + 1
to 2k.
Question 11.3. Let α and β be countable nonzero ordinals, and let k > 1 be
a positive integer. Is it the case that if
α
ω ω →top (ω 1+β , k)2 ,
then
α
·β
ωω →top (ω 1+β , 2k)2 ?
See Remark 10.3 for further details.
Finally, it would be nice to determine whether the existing conjecture ω1 →
(α, k)3 holds, and also to discover something about the topological or closed version
of this relation.
Question 11.4. Is it the case that for all countable ordinals α and all finite k,
ω1 →top (α, k)3 ?
Again, since every power of ω is order-reinforcing, this question is equivalent
if we replace the topological partition relation by the closed partition relation. If
the answer to this question is yes, then the classical and closed relations are in fact
equivalent whenever the ordinal on the left-hand side is ω1 and the ordinals on the
right-hand side are countable.
References
[Bau86] J. E. Baumgartner, Partition relations for countable topological spaces, J. Combin. The-
ory Ser. A 43 (1986), no. 2, 178–195, DOI 10.1016/0097-3165(86)90059-2. MR867644
[Cai14] A. E. Caicedo. Teorı́a de Ramsey de ordinales contables muy pequeños. Submitted.
Available at https://andrescaicedo.wordpress.com/papers/, 2014.
[Cai15] A. E. Caicedo. Partition calculus of small countable ordinals. A survey of results of
Haddad and Sabbagh. Preprint, 2015.
[EM72] P. Erdős and E. C. Milner, A theorem in the partition calculus, Canad. Math. Bull. 15
(1972), 501–505, DOI 10.4153/CMB-1972-088-1. MR0332492
[ER53] P. Erdős and R. Rado, A problem on ordered sets, J. London Math. Soc. 28 (1953),
426–438, DOI 10.1112/jlms/s1-28.4.426. MR0058687
[ER56] P. Erdős and R. Rado, A partition calculus in set theory, Bull. Amer. Math. Soc. 62
(1956), 427–489, DOI 10.1090/S0002-9904-1956-10036-0. MR0081864
[ER67] P. Erdős and R. Rado, Partition relations and transitivity domains of binary relations,
J. London Math. Soc. 42 (1967), 624–633, DOI 10.1112/jlms/s1-42.1.624. MR0218248
[FM88] J. Flum and J. C. Martı́nez, On topological spaces equivalent to ordinals, J. Symbolic
Logic 53 (1988), no. 3, 785–795, DOI 10.2307/2274571. MR960998
[Fri74] H. Friedman, On closed sets of ordinals, Proc. Amer. Math. Soc. 43 (1974), 190–192.
MR0327521
[Haj60] A. Hajnal, Some results and problems on set theory (English, with Russian summary),
Acta Math. Acad. Sci. Hungar. 11 (1960), 277–298. MR0150044
[Haj64] A. Hajnal, Remarks on the theorem of W. P. Hanf, Fund. Math. 54 (1964), 109–113.
MR0160734
[Hil16] J. Hilton, The topological pigeonhole principle for ordinals, J. Symb. Log. 81 (2016),
no. 2, 662–686, DOI 10.1017/jsl.2015.45. MR3519451
[HJW90] A. Hajnal, I. Juhász, and W. Weiss, Partitioning the pairs and triples of topological
spaces, Topology Appl. 35 (1990), no. 2-3, 177–184, DOI 10.1016/0166-8641(90)90103-
9. MR1058798
[HL10] A. Hajnal and J. A. Larson, Partition relations, Handbook of set theory. Vols. 1, 2, 3,
Springer, Dordrecht, 2010, pp. 129–213, DOI 10.1007/978-1-4020-5764-9 3. MR2768681
[HS69a] L. Haddad and G. Sabbagh, Sur une extension des nombres de Ramsey aux ordinaux
(French), C. R. Acad. Sci. Paris Sér. A-B 268 (1969), A1165–A1167. MR0246778
[HS69b] L. Haddad and G. Sabbagh, Calcul de certains nombres de Ramsey généralisés (French),
C. R. Acad. Sci. Paris Sér. A-B 268 (1969), A1233–A1234. MR0248030
[HS69c] L. Haddad and G. Sabbagh, Nouveaux résultats sur les nombres de Ramsey généralisés
(French), C. R. Acad. Sci. Paris Sér. A-B 268 (1969), A1516–A1518. MR0248031
[Jon00] A. L. Jones, A short proof of a partition relation for triples, Electron. J. Combin. 7
(2000), Research Paper 24, 9. MR1755613
[Jon17] Albin L. Jones. Even more on partitioning triples of countable ordinals. Proc. Amer.
Math. Soc., to appear.
[KL06] V. Kieftenbeld and B. Löwe. A classification of ordinal topologies. Unpublished. Avail-
able at https://www.illc.uva.nl/Research/Publications/PP-2006-57.text.pdf, 2006.
[Kru65] A. H. Kruse, A note on the partition calculus of P. Erdős and R. Rado, J. London
Math. Soc. 40 (1965), 137–148. MR0214476
[Lav75] R. Laver, Partition relations for uncountable cardinals ≤ 2ℵ0 , Infinite and finite sets
(Colloq., Keszthely, 1973; dedicated to P. Erdős on his 60th birthday), Vol. II, North-
Holland, Amsterdam, 1975, pp. 1029–1042. Colloq. Math. Soc. Janós Bolyai, Vol. 10.
MR0371652
[LM97] J. A. Larson and W. J. Mitchell, On a problem of Erdős and Rado, Ann. Comb. 1
(1997), no. 3, 245–252, DOI 10.1007/BF02558478. MR1630775
[Mer17] O. Mermelstein. Calculating the closed ordinal Ramsey number Rcl (ω · 2, 3)2 . Preprint.
Available at https://arxiv.org/abs/1702.03878, 2017.
[Mil71] E. C. Milner, A finite algorithm for the partition calculus, Proceedings of the Twenty-
Fifth Summer Meeting of the Canadian Mathematical Congress (Lakehead Univ.,
Thunder Bay, Ont., 1971), Lakehead Univ., Thunder Bay, Ont., 1971, pp. 117–128.
MR0332507
[Pn15] C. Piña, A topological Ramsey classification of countable ordinals, Acta Math. Hungar.
147 (2015), no. 2, 477–509, DOI 10.1007/s10474-014-0413-5. MR3420590
[Ros15] S. Rosenberg. Infinite closed monochromatic subsets of a metric space. Preprint. Avail-
able at https://arxiv.org/abs/1508.02366, 2015.
[Sch10] R. Schipperus, Countable partition ordinals, Ann. Pure Appl. Logic 161 (2010), no. 10,
1195–1215, DOI 10.1016/j.apal.2009.12.007. MR2652192
[Sch12] R. Schipperus, The topological Baumgartner-Hajnal theorem, Trans. Amer. Math. Soc.
364 (2012), no. 8, 3903–3914, DOI 10.1090/S0002-9947-2012-04990-7. MR2912439
[Sie33] W. Sierpiński, Sur un problème de la théorie des relations (French), Ann. Scuola Norm.
Sup. Pisa Cl. Sci. (2) 2 (1933), no. 3, 285–287. MR1556708
[Spe56] E. Specker, Teilmengen von Mengen mit Relationen (German), Comment. Math. Helv.
31 (1957), 302–314. MR0088454
[Tod83] S. Todorčević, Forcing positive partition relations, Trans. Amer. Math. Soc. 280 (1983),
no. 2, 703–720, DOI 10.2307/1999642. MR716846
[Tod96] S. Todorčević. A partition property of spaces with point-countable bases. Unpublished.
1 pg., June 1996.
[Wei90] W. Weiss, Partitioning topological spaces, Mathematics of Ramsey theory, Algorithms
Combin., vol. 5, Springer, Berlin, 1990, pp. 154–171, DOI 10.1007/978-3-642-72905-8 11.
MR1083599
[Wei14] T. V. Weinert, Idiosynchromatic poetry, Combinatorica 34 (2014), no. 6, 707–742, DOI
10.1007/s00493-011-2980-1. MR3296182
[Wil77] N. H. Williams, Combinatorial set theory, Studies in Logic and the Foundations of
Mathematics, vol. 91, North-Holland Publishing Co., Amsterdam, 1977. MR3075383
Department of Mathematics, Boise State University, 1910 University Drive, Boise,

Idaho 83725-1555
Current address: Mathematical Reviews, 416 Fourth Street, Ann Arbor, Michigan 48103-
4820
E-mail address: aec@ams.org
URL: http://www-personal.umich.edu/∼caicedo/
School of Mathematics, University of Leeds, Leeds LS2 9JT, United Kingdom

E-mail address: mmjhh@leeds.ac.uk
Volume 690, 2017
Open determinacy for class games
Victoria Gitman and Joel David Hamkins
Abstract. The principle of open determinacy for class games—two-player

games of perfect information with plays of length ω, where the moves are cho-
sen from a possibly proper class, such as games on the ordinals—is not prov-
able in Zermelo-Fraenkel set theory ZFC or Gödel-Bernays set theory GBC, if
these theories are consistent, because provably in ZFC there is a definable open
proper class game with no definable winning strategy. In fact, the principle of
open determinacy and even merely clopen determinacy for class games implies
Con(ZFC) and iterated instances Conα (ZFC) and more, because it implies
that there is a satisfaction class for first-order truth, and indeed a transfinite
tower of truth predicates Trα for iterated truth-about-truth, relative to any
class parameter. This is perhaps explained, in light of the Tarskian recur-
sive definition of truth, by the more general fact that the principle of clopen
determinacy is exactly equivalent over GBC to the principle of elementary
transfinite recursion ETR over well-founded class relations. Meanwhile, the
principle of open determinacy for class games is provable in the stronger the-
ory GBC + Π11 -comprehension, a proper fragment of Kelley-Morse set theory
KM.
Contents
1. Introduction
2. A broader context
3. The truth-telling game
4. Clopen determinacy is equivalent over GBC to ETR
5. Proving open determinacy in strong theories
6. Questions
References
2010 Mathematics Subject Classification. Primary 03E60; Secondary 03E30, 03C62.

Key words and phrases. Open determinacy, GBC, KM.
We are especially pleased to be a part of this volume celebrating the 60th birthday of W. Hugh
Woodin, in light of the fact that Woodin served (years ago) as the PhD supervisor of the second
author, who himself served (more recently) as the PhD supervisor of the first author. Thus, with
Woodin, we span three mathematical generations.
Commentary concerning this paper can be made at http://jdh.hamkins.org/open-
determinacy-for-class-games.
2017
c by the authors
121
122 VICTORIA GITMAN AND JOEL DAVID HAMKINS
1. Introduction
The past half-century of set theory has revealed a robust connection between infini-
tary game theory and fundamental set-theoretic principles, such as the existence
of certain large cardinals. The existence of strategies in infinite games has often
turned out to have an unexpected set-theoretic power. In this article, we should
like to exhibit another such connection in the case of games of proper class size, by
proving that the principle of clopen determinacy for class games is exactly equiv-
alent to the principle of elementary transfinite recursion ETR along well-founded
class relations. Since this principle implies Con(ZFC) and iterated instances of
Conα (ZFC) and more, the principles of open determinacy and clopen determinacy
both transcend ZFC in consistency strength.
We consider two-player games of perfect information, where two players alter-
nately play elements from an allowed space X of possible moves, which in our case
may be a proper class, such as the class of all ordinals X = Ord. Together, the
players build an infinite sequence α = α0 , α1 , α2 , . . . in X ω , which is the play
resulting from this particular instance of the game. The winner is determined by
consulting a fixed class of plays A ⊆ X ω , possibly a proper class: if α ∈ A, then
the first player has won this play of the game, and otherwise the second player has
won. A strategy for a player is a (class) function σ : X <ω → X, which tells a player
how to move next, given a finite position in the game. Such a strategy is winning
for that player, if playing in accordance with the strategy leads to a winning play of
the game, regardless of how the other player has moved. The game is determined,
if one of the players has a winning strategy. We may formalize all talk of classes
here in Gödel-Bernays GBC set theory, or in ZFC if one prefers to regard classes
as definable from parameters.
The case of open games, generalizing the finite games, is an attractive special
case, which for set-sized games has been useful in many arguments. Specifically, a
game is open for a particular player, if for every winning play of the game for that
player, there occurred during the course of play a finite position where the winning
outcome was already ensured, in the sense that all plays extending that position are
winning for that player. This is equivalent to saying that the winning condition set
for that player is open in the product topology on X ω , where we put the discrete
topology on X. Similarly, a game is clopen, if it is open for each player; these are
the games for which every play of the game has a finite stage where the outcome is
already known.
It is a remarkable elementary fact, the Gale-Stewart theorem [GS53], that in
the context of set-sized games, every open game is determined. In order to discuss
the problems that arise in the context of proper-class games, let us briefly sketch
two classic proofs of open determinacy for set games. Suppose that we have a game
that is open for one of the players, with an open winning condition A ⊆ X ω for
that player, where X is the set of possible moves.
For the first proof of open determinacy, suppose that the open player does not
have a winning strategy in the game. So the initial position of the game is amongst
the set of positions from which the open player does not have a winning strategy.
The closed player may now simply play so as to stay in that collection, because
if every move from a position p leads to a winning position for the open player,
then the open player can unify those strategies into a single winning strategy from
position p. This way of playing is a winning strategy for the closed player, because
OPEN DETERMINACY FOR CLASS GAMES 123
no such position can be an already-won position for the open player. Therefore,
the game is determined.
For a second proof, we use the elegant theory of ordinal game values. Namely,
define that a position p in the game has value 0, if it is an already-won position
for the open player, in the sense that every play extending p is in A. A position p
with the open player to move has value α + 1, if α is minimal such that the open
player can play to a position p x with value α. A position p with the closed
player to play has a defined value only when every possible subsequent position
p y already has a value, and in this case the value of p is the supremum of those
values. The key observation is that if a position has a value, then the open player
can play so as to decrease the value, and the closed player cannot play so as to
increase it or make it become undefined. Thus, by means of this value-reducing
strategy, the open player can win from any position having a value, because the
decreasing sequence of ordinals must eventually hit 0; and the closed player can win
from any position lacking an ordinal value, by maintaining the play on unvalued
positions. So the game is determined, because the initial position either has a value
or is unvalued. (The topic of ordinal game values appears widely in the literature,
but for a particularly accessible discussion of the concrete meaning of small-ordinal
game values, we refer the reader to [EH14, EHP].)
Both of these proofs of open determinacy become problematic in ZFC and also
GBC when X is a proper class. The problem with the first proof is subtle, but
note that winning strategies are proper class functions σ : X <ω → X, and so for
a position p to have a winning strategy for a particular player is a second-order
property of that position. Thus, in order to pick out the class of positions for which
the open player has a winning strategy, we would seem to need a second-order
comprehension principle, which is not available in ZFC or GBC. The proof can,
however, be carried out in GBC + Π11 -comprehension, as explained in theorem 10.
The problem with the ordinal-game-value proof in the proper class context is a little
more clear, since in the inductive definition of game values, one takes a supremum
of the values of p x for all x ∈ X, but if X is a proper class, this supremum
could exceed every ordinal. It would seem that we would be pushed to consider
meta-ordinal game values larger than Ord. So the proofs just don’t seem to work
in ZFC or GBC.
What we should like to do in this article is to consider more seriously the
case where X is a proper class. In this case, as we mentioned, the strategies
σ : X <ω → X will also be proper classes, and the winning condition A ⊆ X ω may
also be a proper class.
Question 1. Can we prove open determinacy for class games?
For example, does every definable open class game in ZFC admit a definable winning
strategy for one of the players? In GBC, must every open class game have a winning
strategy? We shall prove that the answers to both of these questions is no. Our
main results are the following.
Main Theorems.
(1) In ZFC, there is a first-order definable clopen proper-class game with no
definable winning strategy for either player.
(2) In GBC, the existence of a winning strategy for one of the players in the
game of statement (1) is equivalent to the existence of a satisfaction class
for first-order set-theoretic truth.
(3) Consequently, the principle of clopen determinacy for class games in GBC
implies Con(ZFC) and iterated consistency assertions Conα (ZFC) and
more.
(4) Indeed, the principle of clopen determinacy for class games is equiva-
lent over GBC to the principle ETR of elementary transfinite recursion,
which is a strictly weaker theory (assuming consistency) than GBC + Π11 -
comprehension, which is strictly weaker than Kelley-Morse KM set theory.
(5) Meanwhile, open determinacy for class games is provable in GBC + Π11 -
comprehension.
These claims will be proved in Theorems 2, 3, 9 and 10. Note that because
GBC includes the global choice principle, every proper class X is bijective with the
class of all ordinals Ord, and so in GBC one may view every class game as a game
on the ordinals. We shall also prove other theorems that place ETR and hence
clopen determinacy into their setting in second-order set theory beyond GBC.
1.1. Acknowledgements. The authors would like to acknowledge helpful ex-

changes with Thomas A. Johnstone, Stephen G. Simpson, Philip Welch, Kentaro
Fujimoto, and Kentaro Sato concerning the work of this paper. The second author
is grateful for the support of Simons Foundation grant 209252 and for his Visiting
Professorship in 2015 in the Philosophy Department of New York University, where
the initial main result was obtained. Both authors are thankful for the support pro-
vided in Summer and Fall of 2015 by the Isaac Newton Institute for Mathematical
Sciences in Cambridge, U.K., where the work was finalized while they were Visiting
Fellows for the program on the Mathematical, Computational and Foundational
Aspects of the Higher Infinite.
2. A broader context
Let us place the results of this article in a broader context. The need to con-
sider clopen determinacy for class games has arisen in recent developments in large
cardinal set theory and forcing axioms, which sparked our interest. Specifically, Au-
drito and Viale [AV, Aud16], generalizing the uplifting cardinals and resurrection
axioms of [HJ14, HJ], introduced the (α)-uplifting cardinals, the HJ(α)-uplifting
cardinals, and the iterated resurrection axioms RAα (Γ), all of which are defined
in terms of winning strategies in certain proper-class-sized clopen games. Audrito
and Viale had at first formalized their concepts in Kelley-Morse set theory, which
is able to prove the requisite determinacy principles. The main results of this paper
identify the minimal extensions of Gödel-Bernays set theory able satisfactorily to
treat these new large cardinals and forcing axioms.
After having proved our theorems, however, we noticed the connection with
analogous results in second-order arithmetic, where there has been a vigorous in-
vestigation of the strength of determinacy over very weak theories. Let us briefly
survey some of that work. First, there is a natural affinity between our theorem 9,
which shows that clopen determinacy is exactly equivalent over GBC to the princi-
ple of elementary transfinite recursion ETR over well-founded class relations, with
the 1977 dissertation result of Steel (see [Sim09, Thm V.8.7]), showing that clopen
determinacy for games on the natural numbers is exactly equivalent in reverse math-
ematics to the theory of arithmetical transfinite recursion ATR0 . Simpson reported
in conversation with the second author that Steel’s theorem had had a strong in-
fluence on the beginnings of the reverse mathematics program. In both the set and
class contexts, we have equivalence of clopen determinacy with a principle of first-
order transfinite recursion. In the case of games on the natural numbers, however,
Steel proved that ATR0 is also equivalent with open determinacy, and not merely
clopen determinacy, whereas the corresponding situation of open determinacy for
class games is not yet completely settled; the best current upper bound provided
by theorem 10.
Following Steel’s work, the reverse mathematics program proceeded to con-
sider the strength of determinacy for games having higher levels of complexity.
Tanaka [Tan90] established the equivalence of Π11 -comprehension and Σ01 ∧ Π01 -
determinacy, as well as the equivalence of Π11 -transfinite recursion and Δ02 -determin-
acy, both over RCA0 . The subsequent paper [Tan91] showed that Σ02 -determinacy
is equivalent over RCA0 to a less familiar second-order axiom Σ11 -MI, known as the
axiom of Σ11 -monotone inductive definition 1 . MedSalem and Tanaka [MT07] con-
sidered Δ03 -determinacy, proving it in Δ13 -comprehension plus Σ13 -induction, and
showing that it does not follow from Δ13 -comprehension alone. MedSalem and
Tanaka [MT08] settled the exact strength of Δ03 -determinacy over the theory
RCA0 + Π13 -transfinite induction by introducing a new axiom for iterating Σ11 -
inductive definitions. Philip Welch [Wel11] characterized the ordinal stage by
which the strategies for Σ03 games appear in the constructible hierarchy, continuing
the program initiated by Blass [Bla72], who showed that every computable game
has its strategy appearing before the next admissible ordinal. He also shows that
Π13 -comprehension proves not just Π03 -determinacy, but that there is a β-model of
Π03 -determinacy. Montalbán and Shore [MS12] established a precise bound for the
amount of determinacy provable in full second-order arithmetic Z2 . They showed
that for each fixed n, Π1n+2 -comprehension proves determinacy for n-length Boolean
combinations of Π04 -formulas, but Z2 cannot prove Δ04 -determinacy.
Although there is a clear analogy between our theorems concerning clopen
determinacy for proper-class games and the analysis of clopen determinacy on the
natural numbers, nevertheless, one should not naively expect a tight connection
between the determinacy of Σ0n definable games in second-order arithmetic, say,
with that of Σ0n definable class games in the Lévy hierarchy. The reason is that
determinacy for first-order definable Σ0n sets of reals in the arithmetic hierarchy
is provable in ZFC, and with sufficient large cardinals, determinacy runs through
the second-order projective hierarchy Σ1n as well, but determinacy for first-order
definable games in set theory is simply refutable in ZF already at the level of Δ02 in
the Lévy hierarchy, in light of theorem 5. Rather, one should expect a connection
between the analysis of Σ0n determinacy in arithmetic and the corresponding level
of the proper-class analogue of the Borel hierarchy for subclasses of Ordω , which
we discuss in section 6. There are several fundamental disanalogies for determinacy
1 A function Γ : P (ω) → P (ω) is called a monotone operator (over ω) if whenever X ⊆ Y ,
then Γ(X) ⊆ Γ(Y ). The axiom of Σ11 -monotone inductive definition asserts that for every Σ11 -
monotone operator Γ (meaning {(x, X) | x ∈ Γ(X)} is Σ11 ), there exists a sequence Γα | α ≤ σ

for some ordinal σ such that Γα = Γ( β<α Γβ ) for all α ≤ σ and such that Γσ = α<σ Γα , so
that Γσ is a fixed point of the operator Γ.
in second-order arithmetic in comparison with second-order set theory that lead us

to expect differences in the resulting theory, among them the facts that (i) Ordω
is not separable in the product topology whereas Baire space ω ω is separable; (ii)
wellfoundedness for class relations is first-order expressible in set theory, whereas
it is Π11 -complete in arithmetic; and finally, (iii) individual plays of a game on Ord
are first-order objects in set theory, making the payoff collection a class, whereas
in arithmetic a play of a game is already a second-order object and the payoff
collection is a third-order object.
3. The truth-telling game

Let us now prove the initial claims of the main theorem.
Theorem 2. In ZFC, there is a particular definable clopen proper-class game,
for which no definition and parameter defines a winning strategy for either player.
The proof therefore provides in ZFC a completely uniform counterexample
to clopen determinacy, with respect to definable strategies, because the particular
game we shall define has no definable winning strategy for either player in any model
of set theory. Theorem 2 is a theorem scheme, ranging over the possible definitions
of the putative winning strategy. We shall prove the theorem as a consequence of
the following stronger and more revealing result.
Theorem 3. There is a particular first-order definable clopen game, whose
determinacy is equivalent in GBC to the existence of a satisfaction class for first-
order set-theoretic truth. Consequently, in GBC the principle of clopen determi-
nacy for class games implies Con(ZFC), as well as iterated consistency assertions
Conα (ZFC) and much more.
Proof. To begin, we introduce the truth-telling game, which will be a definable
open game with no definable winning strategy. The truth-telling game has two
players, the interrogator and the truth-teller, who we may imagine play out the game
in a court of law, with the truth-teller in the witness box answering tricky pointed
questions posed by the opposing counsel, in the style of a similar game described by
Adrian Mathias [Mat15] in the context of extensions of PA in arithmetic. On each
turn, the interrogator puts an inquiry to the the truth-teller concerning the truth
of a particular first-order set-theoretic formula ϕ(a) with parameters. The truth-
teller must reply to the inquiry by making a truth pronouncement either that it is
true or that it is false, not necessarily truthfully, and in the case that the formula
ϕ is an existential assertion ∃x ψ(x, a) declared to be true, then the truth teller
must additionally identify a particular witness b and pronounce also that ψ(b, a)
is true. So a play of the game consists of a sequence of such inquiries and truth
pronouncements.
The truth-teller need not necessarily answer truthfully to win! Rather, the
truth-teller wins a play of the game, provided merely that she does not violate the
recursive Tarskian truth conditions during the course of play. What we mean, first,
is that when faced with an atomic formula, she must pronounce it true or false
in accordance with the actual truth or falsity of that atomic formula; similarly,
she must pronounce that ϕ ∧ ψ is true just in case she pronounces both ϕ and
ψ separately to be true, if those inquiries had been issued by the interrogator
during play; she must pronounce opposite truth values for ϕ and ¬ϕ, if both are
inquired about; and she must pronounce ∃x ϕ(x, a) to be true if and only if she ever
pronounces ϕ(b, a) to be true of any particular b (the forward implication of this is
already ensured by the extra pronouncement in the existential case of the game).
This is an open game for the interrogator, because if the truth-teller ever should
violate the Tarskian conditions, then this violation will be revealed at finite stage
of play, and this is the only way for the interrogator to win.
We remind the reader that a satisfaction class or truth predicate for first-order
truth is a class Tr of pairs ϕ, a consisting of a formula ϕ and a list of parameters
a assigned to the free variables of that formula, which obeys the Tarskian recursive
definition of truth (for simplicity we shall write the pair simply as ϕ(a), suppressing
the variable assignment, but keep in mind that these are mentions of formulas rather
than uses). So in the atomic case, we’ll have (a = b) ∈ Tr if and only if a = b, and
(a ∈ b) ∈ Tr if and only if a ∈ b; for negation, ¬ϕ(a) ∈ Tr if and only if ϕ(a) ∈
/ Tr;
for conjunction, (ϕ ∧ ψ)(a) ∈ Tr if and only if ϕ(a) ∈ Tr and ψ(a) ∈ Tr; and for
quantifiers, ∃x ϕ(x, a) ∈ Tr just in case there is b for which ϕ(b, a) ∈ Tr. Tarski
proved that in any sufficiently strong first-order theory no such truth predicate
for first-order truth is definable in the same language. Meanwhile, in the second-
order Kelley-Morse set theory KM and even in the weaker theory GBC plus the
principle of transfinite recursion over well-founded class relations, we can define a
truth predicate for first-order truth, simply because the Tarskian recursion itself
is a well-founded recursion on the complexity of the formulas, where we define the
truth of ϕ(a) in terms of ψ(b) for simpler formulas ψ.
Lemma 3.1. The truth-teller has a winning strategy in the truth-telling game
if and only if there is a satisfaction class for first-order truth.
Proof. We may understand this lemma as formalized in Gödel-Bernays GBC
set theory, which includes the global choice principle. Clearly, if there is a satisfac-
tion class for first-order truth, then the truth-teller has a winning strategy, which
is simply to answer all questions about truth in accordance with that satisfaction
class, using the global choice principle to pick Skolem witnesses in the existential
case. Since by definition that class obeys the Tarskian conditions, she will win the
game, no matter which challenges are issued by the interrogator.
Conversely, suppose that the truth-teller has a winning strategy τ in the game.
We shall use τ to build a satisfaction class for first-order truth. Specifically, let Tr be
the collection of formulas ϕ(a) that are pronounced true by τ in any play according
to τ , including the supplemental truth pronouncements made in the existential
case about the particular witnesses. We claim that Tr is a satisfaction class. Since
the truth-teller was required to answer truthfully to all inquiries about atomic
formulas, it follows that Tr contains all and only the truthful atomic assertions.
In particular, the answers provided by the strategy τ on inquiries about atomic
formulas are independent of the particular challenges issued by the interrogator
and of the order in which they are issued. Next, we generalize this to all formulas,
arguing by induction on formulas that the truth pronouncements made by τ on a
formula is always independent of the play in which that formula arises. We have
already noticed this for atomic formulas. In the case of negation, if inductively all
plays in which ϕ(a) is issued as a challenge or arises as a witness case come out true,
then all plays in which ¬ϕ(a) arises will result in false, or else we could create a play
in which τ would violate the Tarskian truth conditions, simply by asking about ϕ(a)
after ¬ϕ(a) was answered affirmatively. Similarly, if ϕ and ψ always come out the
same way, then so must ϕ ∧ ψ. We don’t claim that τ must always issue the same
witness b for an existential ∃x ψ(x, a), but if the strategy ever directs the truth-
teller to pronounce this statement to be true, then it will provide some witness b
and pronounce ψ(b, a) to be true, and by induction this truth pronouncement for
ψ(b, a) is independent of the play on which it arises, forcing ∃x ϕ(x, a) to always
be pronounced true. Thus, by induction on formulas, the truth pronouncements
made by the truth-teller strategy τ allow us to define from τ a satisfaction class for
first-order truth.
It follows by Tarski’s theorem on the non-definability of truth that there can
be no definable winning strategy for the truth-teller in this game, because there
can be no definable satisfaction class.
Lemma 3.2. The interrogator has no winning strategy in the truth-telling game.
Proof. Suppose that σ is a strategy for the interrogator. So σ is a proper
class function that directs the interrogator to issue certain challenges, given the
finite sequence of previous challenges and truth-telling answers. By the reflection
theorem, there is a closed unbounded proper class of cardinals θ, such that σVθ ⊆
Vθ . That is, Vθ is closed under σ, in the sense that if all previous challenges and
responses come from Vθ , then the next challenge will also come from Vθ . Since
Vθ , ∈ is a set, we have a truth predicate on it, as well as a Skolem function
selecting existential witnesses. Consider the play, where the truth-teller replies
to all inquiries by consulting truth in Vθ , rather than truth in V , and using the
Skolem function to provide the witnesses in the existential case. The point is that
if the interrogator follows σ, then all the inquiries will involve only parameters a in
Vθ , provided that the truth-teller also always gives witnesses in Vθ , which in this
particular play will be the case. Since the truth predicate on Vθ does satisfy the
Tarskian truth conditions, it follows that the truth-teller will win this instance of
the game, and so σ is not a winning strategy for the interrogator.
Thus, if open determinacy holds for classes, then there is a truth predicate
Tr for first-order truth. But we have not yet quite proved the theorem, because
the truth-telling game is an open game, rather than a clopen game, whereas the
theorem concerns determinacy for clopen games. The truth-teller wins the truth-
telling game only by playing the game out for infinitely many steps, and this is not
an open winning condition for her, since at any point the play could have continued
in such a way so as to produce a loss for the truth-teller, if the players cooperated
in order to achieve that.
So let us describe a modified game, the counting-down truth-telling game, which
will be clopen and which we may use in order to prove the theorem. Specifically, the
counting-down truth-telling game is just like the truth-telling game, except that we
insist that the interrogator must also state on each move a specific ordinal αn , which
descend during play α0 > α1 > · · · > αn . If the interrogator gets to 0, then the
truth-teller is declared the winner. For this modified game, the winner will be known
in finitely many moves, because either the truth-teller will violate the Tarskian
conditions or the interrogator will hit zero. So this is a clopen game. Since the
counting-down version of the game is harder for the interrogator, it follows that the
interrogator still can have no winning strategy. We modify the proof of lemma 3.1
for this game by claiming that if τ is a winning strategy for the truth-teller in
the counting-down truth-telling game, then the truth pronouncements made by τ
in response to all plays with sufficiently large ordinals all agree with one another
independently of the interrogator’s play. The inductive argument of lemma 3.1 still
works under the assumption that the counting-down ordinal is sufficiently large,
because there will be enough time to reduce a problematic case. The ordinal will
depend only on the formula and not on the parameter. For example, if ϕ(a) always
gets the same truth pronouncement for plays in which it arises with sufficiently
large ordinals, then so also does ¬ϕ(a), with a slightly larger ordinal, because in a
play with the wrong value for ¬ϕ(a) we may direct the interrogator to inquire next
about ϕ(a) and get a violation of the Tarskian recursion. Similar reasoning works
in the other cases, and so we may define a satisfaction class from a strategy in the
modified game. Since that game is clopen, we have proved that clopen determinacy
for class games implies the existence of a satisfaction class for first-order truth.
We complete the proof of theorem 3 by explaining how the existence of a sat-
isfaction class implies Con(ZFC) and more. Working in Gödel-Bernays set theory,
we may apply the reflection theorem to the class Tr and thereby find a proper class
club C of cardinals θ for which Vθ , ∈, Tr ∩ Vθ ≺Σ1 V, ∈, Tr. In particular, this
implies that Tr ∩ Vθ is a satisfaction class on Vθ , which therefore agrees with truth
in that structure, and so these models form a continuous elementary chain, whose
union is the entire universe:
Vθ0 ≺ Vθ1 ≺ · · · ≺ Vλ ≺ · · · ≺ V.
There is a subtle point here concerning ω-nonstandard models, namely, in order to
see that all instances of the ZFC axioms are declared true by Tr, it is inadequate
merely to note that we have assumed ZFC to be true in V , because this will give
us only the standard-finite instances of those axioms in Tr, but perhaps we have
nonstandard natural numbers in V , beyond the natural numbers of our metatheory.
Nevertheless, because in GBC we have the collection axiom relative to the truth
predicate itself, we may verify that all instances of the collection axiom (including
nonstandard instances, if any)

∀b ∀z ∀x ∈ b ∃y ϕ(x, y, z) → ∃c ∀x ∈ b ∃y ∈ c ϕ(x, y, z)
must be declared true by Tr, because we may replace the assertion of ϕ(x, y, z)
with the assertion ϕ(x, y, z) ∈ Tr, which reduces the instance of collection for the
(possibly nonstandard) formula ϕ to an instance of standard-finite collection in
the language of Tr, using the Gödel code of ϕ as a parameter, thereby collecting
sufficient witnesses y into a set c. So even the nonstandard instances of the collection
axiom must be declared true by Tr. It follows that each of these models Vθ for θ ∈ C
is a transitive model of ZFC, understood in the object theory of V , and so we may
deduce Con(ZFC) and Con ZFC + Con(ZFC) and numerous iterated consistency
statements of the form Conα (ZFC), which must be true in all such transitive models
for quite a long way. Alternatively, one can make a purely syntactic argument for
Con(ZFC) from a satisfaction class, using the fact that the satisfaction class is
closed under deduction and does not assert contradictions.
Note that in the truth-telling games, we didn’t really need the interrogator to
count down in the ordinals, since it would in fact have sufficed to have him count
down merely in the natural numbers; the amount of time remaining required for
the truth pronouncements to stabilize is essentially related to the syntactic com-
plexity of ϕ. We could have insisted merely that on the first move, the interrogator
announce a natural number n, and then the game ends after n moves, with the
interrogator winning only if the Tarski conditions are violated by the truth-teller
within those moves.
Proof of theorem 2. We use the same game as in the proof of theorem 3.
Any definable winning strategy in the counting-down truth-telling game would pro-
vide a definable truth predicate, but by Tarski’s theorem on the non-definability of
truth, there is no such definable truth predicate. Thus, the counting-down truth-
telling game is a first-order parameter-free definable clopen game in ZFC, which
can have no definable winning strategy (allowing parameters) for either player.
It is interesting to observe that one may easily modify the truth-telling games
by allowing a fixed class parameter B, so that clopen determinacy implies over GBC
that there is a satisfaction class relative to truth in V, ∈, B. For example, we may
get a truth predicate Tr1 for the structure V, ∈, Tr itself, so that Tr1 concerns
truth-about-truth. One may iterate this idea much further, to have predicates Trα
for every ordinal α, which are truth predicates for the structure V, ∈, Trβ β<α .
Somewhat more uniformly, we may prefer a single binary predicate Tr ⊆ Ord × V ,
whose every slice Trα = { x | α, x ∈ Tr } is a truth predicate for the structure
V, ∈, Tr (α × V ), and this is a more expressive treatment than having separate
predicates, since one may now quantify over the earlier stages of truth. Indeed, one
may hope to iterate truth predicates beyond Ord along any class well-order as in
theorem 8.
Using the same ideas as in the proof of theorem 3, one may formulate an
iterated-truth-telling game, where the truth-teller answers inquiries about such it-
erated truth predicates, and then prove from clopen determinacy that there is
indeed such an iterated truth predicate. This conclusion also follows immediately
from theorem 9, however, as the iterated truth predicate can be defined by an el-
ementary transfinite recursion, and furthermore, the iterated-truth-telling game is
fundamentally similar to the iteration game we use to prove theorem 9. We shall
therefore not give a separate proof for the iterated-truth case. Theorems 8 and 9
show that clopen determinacy is equivalent over GBC to the existence of iterated
truth predicates over any class well-order.
Next, we briefly clarify the role of the global choice principle with the following
well-known result.
Theorem 4 (Folklore). In Gödel-Bernays set theory GB, the principle of
clopen determinacy implies the global axiom of choice.
Proof. Consider the game where player I plays a nonempty set b an player
II plays a set a, with player II winning if a ∈ b. This is a clopen game, since it is
over after one move for each player. Clearly, player I can have no winning strategy,
since if b is nonempty, then player II can win by playing any element a ∈ b. But a
winning strategy for player II amounts exactly to a global choice function, selecting
uniformly from each nonempty set an element.
The set analogue of the proof of theorem 4 shows in ZF that clopen determinacy
for set-sized games implies the axiom of choice, and so over ZF the principle of
clopen determinacy for set-sized games is equivalent to the axiom of choice. As a
consequence, we may prove in ZF that the universal axiom of determinacy, which
asserts that every game on every set is determined, is simply false: either there is
some clopen game that is not determined, or the axiom of choice holds and there
is a game on the natural numbers that is not determined. Pushing this idea a bit
further leads to the following:
Theorem 5. There is a Δ02 -definable set-sized game in ZF that is not deter-
mined.
Proof. Let us emphasize that in the statement of the theorem we are referring
to Δ02 in the sense of the Lévy hierarchy of the first-order language of set theory
(and not in the sense of the arithmetic or projective hierarchies of descriptive set
theory).
Consider the game G where player I begins by playing a nonempty set of reals
A ⊆ ω ω , with player II next playing an element of it a ∈ A; after this, player I
plays a non-determined set B ⊆ ω ω , and then play proceeds as in the game GB
determined by B. The first player to violate those requirements loses, and otherwise
the winner is determined by the resulting play in GB .
The winning condition for this game is Δ02 -definable in set theory, because it is
sufficiently local. For example, the assertion that a set B ⊆ ω ω is non-determined
has complexity Δ02 in set theory: (Π02 ) for every strategy for one of the players,
there is a strategy for the other that defeats it; (Σ02 ) there is a set M such that
M = Vω+3 and M satisfies “B is not determined.”
Finally, we argue in ZF that neither player has a winning strategy for this
game. Player I cannot have a winning strategy in G, since this would require
him to play a non-determined set B and then win the play of GB , contradicting
that B is not determined. If player II has a winning strategy in G, then by the
argument of theorem 4, we get the axiom of choice for sets of reals and hence a
well-ordering of the reals. From this, we know that there are non-determined sets
of reals B, which player I can play, and then the winning strategy for player II
would provide a winning strategy for GB , contradicting the assumption that B was
not determined.
A second simple observation about theorem 4 is that this argument answers

a special case of question 1. Namely, since there are models of ZFC where the
global axiom of choice fails for definable classes, there must be some models of ZFC
having definable clopen games with no definable winning strategy. Our theorem 5,
in contrast, establishes the stronger result that every model of ZFC, including
those with global choice, has a definable clopen game with no definable strategy,
and furthermore, the definition of the game is uniform.
4. Clopen determinacy is equivalent over GBC to ETR

We shall now generalize the argument of theorem 3 to prove a stronger result,
which we believe explains the phenomenon of theorem 3. Specifically, in theorem 9
we shall prove that clopen determinacy is exactly equivalent over GBC to the princi-
ple of elementary transfinite recursion ETR over well-founded class relations. This
explains the result of theorem 3 because, as we have mentioned, truth itself is de-
fined by such a recursion, namely, the familiar Tarskian recursive definition of truth
defined by recursion on formulas, and so ETR implies the existence of a satisfaction
class for first-order truth. Following Kentaro Fujimoto [Fuj12, Definition 88], we
introduce the principle of elementary transfinite recursion.
Definition 6. The principle of elementary transfinite recursion over well-

founded class relations, denoted ETR, is the assertion that every first-order re-
cursive definition along any well-founded binary class relation has a solution.
Let us explain in more detail. A binary relation on a class I is well-founded,

if every nonempty subclass B ⊆ I has a -minimal element. This is equivalent in
GBC to the assertion that there is no infinite -descending sequence, and indeed
one can prove this equivalence in GB + DC, meaning the dependent choice principle
for set relations: clearly, if there is an infinite -descending sequence, then the set of
elements on that sequence is a set with no -minimal element; conversely, if there is
a nonempty class B ⊆ I with no -minimal element, then by the reflection principle
relativized to the class B, there is some Vθ for which B ∩ Vθ is nonempty and has
no -minimal element; but using DC for ∩ Vθ we may successively pick xn+1 xn
from B ∩ Vθ , leading to an infinite -descending sequence. We find it interesting
to notice that in this class context, therefore, well-foundedness for class relations
becomes a first-order concept, which is a departure from the analogous situation in
second-order number theory, where of course well-foundedness is Π11 -complete and
definitely not first-order expressible in number theory.
Continuing with our discussion of recursion, suppose that we have a well-
founded binary relation on a class I, and suppose further that ϕ(x, b, F, Z) is
a formula describing the recursion rule we intend to implement, where ϕ involves
only first-order quantifiers, F is a class variable for a partial solution and Z is a fixed
class parameter, henceforth suppressed. The idea is that ϕ(x, b, F ) expresses the
recursive rule to be iterated. Namely, a solution of the recursion is a class F ⊆ I ×V
such that for every b ∈ I, the bth slice of the solution Fb = { x | ϕ(x, b, F b) } is
defined by the recursive rule, where F b = F ∩ ({ c ∈ I | c b } × V ) is the partial
solution on slices prior to b. In this way, each slice of the solution Fb is determined
via the recursive rule from the values on the slices Fc for earlier values c.
The principle of elementary transfinite recursion ETR over well-founded class
relations asserts that for every such well-founded relation I, and any first-order
recursive rule ϕ as above, there is a solution. One may equivalently consider only
well-founded partial-order relations, or well-founded tree orders, or class well-orders.
Lemma 7. The principle of ETR is equivalently formulated over GBC with any
of the following types of well-founded relations:
(1) well-founded class relations.
(2) well-founded class partial orders.
(3) well-founded class tree orders.
(4) well-ordered class relations.
Proof. Since the families of relations are becoming more specialized, it is clear
that each statement implies the next (1 → 2 → 3 → 4). It remains to prove the
converse implications.
(2 → 1) If is any well-founded binary class relation, then let < be the tran-
sitive closure of , which is a well-founded partial order. Any first-order recursion
on is easily transformed to a recursion on <, using as a class parameter if
necessary.
(3 → 2). If < is any well-founded partial order, then let T be the tree of all
finite <-descending sequences, ordered by extension, so that longer is lower. This
tree is well-founded, and any recursion defined on < can be easily transferred to
the tree order.
(4 → 3) Suppose that we have a first-order recursion defined on a well-founded
class tree T of sequences (the root is at top, the tree grows downward). We can
linearize the tree by the Kleene-Brouwer order, by which s is less than t if s extends
t or if when they disagree, then s is lower than t in that coordinate, with respect
to a fixed global well-ordering of the universe. This is a class well-order, and it is
easy to transfer the recursion from T to this linear order.
Note that if GBC is consistent, then it does not prove that all recursions along
set well-orders have a solution, because a recursion of length ω suffices to define a
truth predicate by the Tarskian recursion on formulas, and the existence of such
a predicate implies Con(ZFC) and therefore also Con(GBC). So GBC plus ETR
is strictly stronger than GBC in consistency strength, although it is provable in
Kelley-Morse set theory KM, in essentially the same way that GBC proves the
set-like special case.
Next, let’s explain the tight connection between ETR and the existence of
iterated truth predicates, a result similar to those obtained by Fujimoto [Fuj12]
(for instance, see his corollary 61). If I, is a class well-order, then T is an
iterated truth predicate over I, relative to Z, if for each i ∈ I, the ith slice Ti is
a truth predicate for the structure V, ∈, Z, T i, satisfying the Tarskian recursion
for formulas in the language of set theory augmented with predicates for Z and
T i, where T i = T ∩ ({ j ∈ I | j i } × V ) is the restriction of T to the -earlier
stages of truth.
Theorem 8. The principle ETR of elementary transfinite recursion is equiv-
alent over GBC to the assertion that for every class parameter Z and every class
well-ordering I, there is an iterated truth predicate T along I, over the pa-
rameter Z.
Proof. The forward implication is straightforward, since the truth predicate
itself is defined by a transfinite recursion of length ω · I, . Namely, to get the
truth predicate for the next stage, one simply performs the Tarskian recursion
through the formula complexity hierarchy, which has height ω.
Conversely, suppose that for class parameter Z and class well-ordering I, ,
we have an iterated truth predicate T over relative to Z. Now suppose that we
have an instance of ETR iterating a formula ϕ(x, i, F, Z) along I, . We claim
that from parameter T , we may define a solution F to this recursion. Specifically,
we claim that there is a formula ϕ̄ such that if one extracts from T the class defined
by ϕ̄, namely, F = { i, x | T (i, ϕ̄, x) }, then F is a solution to the recursion of ϕ
along . The formula ϕ̄ should simply be chosen so that V, ∈, Z, T i |= ϕ̄(x, i) if
and only if V, ∈, Z, F i |= ϕ(x, i), where F is defined as just mentioned using ϕ̄.
Such a formula ϕ̄ exists by Gödel’s fixed-point lemma: for any e, let ψ(e, x, i) be
the assertion V, ∈, Z, { i, x | T (i, e, x) } |= ϕ(x, i), and then by the usual fixed-
point trick find a formula ϕ̄(x, i), for which V, ∈, Z, T i |= ψ(ϕ̄, x, i) ↔ ϕ̄(x, i).
It follows that the class F iteratively defined from T by ϕ̄ satisfies ϕ at each step
and therefore is a solution to the recursion of ϕ along , as desired.
An important immediate consequence of theorem 8 is that ETR is expressible as
a single second-order assertion in the language of GBC, and so we needn’t treat it as
a scheme (see also the remarks at the beginning of the proof of [Fuj12, theorem 88]).
Namely, ETR is equivalent to the assertion that for every class well-order I,
and every class parameter Z, there is an iterated truth-predicate T along I,
relative to Z.
We come now to the next main contribution of this article, the equivalence of
clopen determinacy for class games with ETR, and with the existence of iterated
truth predicates over class well-orders.
Theorem 9. In Gödel-Bernays set theory GBC, the following are equivalent.
(1) Clopen determinacy for class games. That is, in any two-player game of
perfect information whose winning condition class is both open and closed,
there is a winning strategy for one of the players.
(2) The principle ETR of elementary transfinite recursion over well-founded
class relations: every such recursion has a solution.
(3) Existence of iterated-truth predicates. That is, for every class parameter
Z and every class well-ordering I, , there is an iterated truth predicate
T along I, over parameter Z.
Proof. (2 ↔ 3) This is established by theorem 8.
(2 → 1) Assume the principle ETR of elementary transfinite recursion, and
suppose we are faced with a clopen game. Consider the game tree, consisting of
positions arising during play, up to the moment that a winner is known, orienting
the tree so that the root is at the top and play proceeds downward. This tree is
well-founded precisely because the game is clopen. Let us label the terminal nodes
of the tree with I or II according to who has won the game in that position, and
more generally, let us label all the nodes of the tree with I or II according to the
following transfinite recursion: if a node has I to play, then it will have label I if
there is a move to a node already labeled I, and otherwise II; similarly, when it
is player II’s turn to play, then if she can play to a node labeled II, we label the
original node with II, and otherwise I. By the principle of elementary transfinite
recursion, there is a labeling of the entire tree that accords with this recursive rule.
It is now easy to see that if the initial node is labeled with I, then player I has a
winning strategy, which is simply to stay on the nodes labeled I. (We use the global
choice principle to choose a particular such node with the right label; this use can
be avoided if the space X of possible moves is already well-ordered, such as in the
case of games on the ordinals X = Ord.) Note that player II cannot play in one
move on her turn from a node labeled I to one labeled II. Similarly, if the initial
node is labeled II, then player II has a winning strategy, which is simply to stay
on the nodes labeled II. And so the game is determined, and we have established
clopen determinacy.
(1 → 2) This implication is the main new content of this theorem. Assume the
principle of clopen determinacy for class games, and suppose that we are faced with
a recursion along a well-founded class partial-order relation on a class I, using a
first-order recursion rule ϕ(x, b, F ), possibly with a fixed class parameter Z, which
we suppress. We shall define a certain clopen game, and prove that any winning
strategy for this game will produce a solution for the recursion.
At first, we consider a simpler open game, the recursion game, which will be
much like the truth-telling game used in theorem 3, except that in this game,
the truth-teller will also provide information about the putative solution of the
recursion in question; later, we shall revise this game to a clopen game. In the
recursion game, we have the same two players again, the interrogator and the
truth-teller, but now the interrogator will make inquiries about truth in a structure
of the form V, ∈, , F , where is the well-founded class relation and F is a binary
class predicate, not yet specified, but which we hope will become a solution of the
recursion, with F ⊆ I × V . Specifically, the interrogator is allowed to ask about
the truth of any first-order formula ϕ(a) in the language of this structure and in
particular to inquire as to whether F (i, x) or not. The truth-teller, as before, will
answer the inquiries by pronouncing either that ϕ(a) is true or that it is false, and
in the case ϕ(a) = ∃x ψ(x, a) and the formula was pronounced true, then the truth-
teller shall also provide as before a witness b for which she also pronounces ψ(b, a)
to be true. The truth-teller loses immediately, if she should ever violate Tarski’s
recursive definition of truth, and she also is required to pronounce any instance of
the recursion rule F (i, x) ↔ ϕ(x, i, F i) to be true, where F i denotes the class
F ∩ ({ j ∈ I | j i } × V ). Specifically, we form the formula ϕ(x, i, F i) in the
language of set theory with a predicate for F by replacing any atomic occurrence
of the predicate F (j, y) in ϕ with F (j, y) ∧ j i. Since violations of any of these
requirements, if they occur at all, do so at a finite stage of play, it follows that the
game is open for the interrogator.
Lemma 9.1. The interrogator has no winning strategy in the recursion game.
Proof. To prove this lemma, we use a modification of the idea of lemma 3.2.
Suppose that σ is a strategy for the interrogator. So σ is a class function that
instructs the interrogator how to play next, given a position of partial play. By the
reflection theorem, there is an ordinal θ such that Vθ is closed under σ, and using the
satisfaction class that comes from clopen determinacy, we may actually also arrange
that Vθ , ∈, ∩ Vθ , σ ∩ Vθ ≺ V, ∈, , σ. Consider the relation ∩ Vθ , which is
a well-founded relation on I ∩ Vθ . Since ZFC and hence GBC proves the existence
of solutions to transfinite recursions for sets, there is a (unique) solution f ⊆ (I ∩
Vθ ) × Vθ such that the ith slice fi = { x ∈ Vθ | Vθ , ∈, ∩ Vθ , f |= ϕ(x, i, f i) } is
defined by the recursion of ϕ, where f i means the restriction of the predicate
f ∩ ({ j ∈ I ∩ Vθ | j i } × Vθ ) to the predecessors of i that are also in Vθ . Consider
now the play of the recursion game in V , where the interrogator uses the strategy σ
and the truth-teller plays in accordance with truth in the structure Vθ , ∈, ∩Vθ , f ,
which is a little sneaky because the function f is a solution of the recursion rule
ϕ only on the relation ∩ Vθ , rather than the full relation . But since Vθ was
closed under σ, the interrogator will never issue challenges outside of Vθ in this
play; and since the function f fulfills the recursive rule f (i, x) ↔ ϕ(x, i, f i) in
this structure, the truth-teller will not be trapped in any violation of the Tarski
conditions or the recursion condition. Thus, the truth-teller will win this instance
of the game, and so σ is not a winning strategy for the interrogator, as desired.
Lemma 9.2. The truth-teller has a winning strategy in the recursion game if
and only if there is a solution of the recursion.
Proof. If there is a solution F of the recursion, then by clopen determinacy
we know there is also a satisfaction class Tr for first order truth in the structure
V, ∈, , F , and the truth-teller can answer all queries of the interrogator in the
recursion game by referring to what Tr asserts is true in this structure. This will
be winning for the truth-teller, since Tr obeys the Tarskian conditions and makes
all instances of the recursive rule true with the predicate F .
Conversely, suppose that τ is a winning strategy for the truth-teller in the re-
cursion game. We may see as before that the truth pronouncements made by τ
about truth in the structure V, ∈, are independent of the play in which they
occur, and they provide a satisfaction class for this structure. This is proved just as
for the truth-telling game by induction on the complexity of the formulas: the strat-
egy must correctly answer all atomic formulas, and the answers to more complex
formulas must be independent of the play since violations of this would lead to vio-
lations of the Tarski conditions by reducing to simpler formulas, as before, and this
would contradict our assumption that τ is a winning strategy for the truth-teller.
Consider next the truth pronouncements made by τ in the language involving
the class predicate symbol F . We shall actually need this property only in the
restricted languages, where for each i ∈ I, we consider formulas asserting truth
in the structure V, ∈, , F i, rather than concerning truth in the full structure
V, ∈, , F . We claim by induction on i, with an embedded induction on formulas,
that for every i ∈ I, the truth pronouncements provided by the strategy τ in this
language are independent of the play in which they are made and furthermore
provide a truth predicate for a structure of the form V, ∈, , F i. The case
where i is -minimal is essentially similar to the case we already handled, where
no reference to F is made, since F i must be asserted to be empty in this case.
Suppose inductively that our claim is true for assertions in the language with F j,
whenever j i, and consider the language with F i. (Note that the claim we
are proving by induction is first-order expressible in the class parameter τ , and
so this induction can be legitimately undertaken in GBC; we haven’t allowed an
instance of Π11 -comprehension to sneak in here.) It is not difficult to see that τ
must pronounce that the various predicates F j cohere with one another on their
common domain, since any violation of this will give rise to a violation of the
Tarskian recursion. So our induction assumption ensures that τ has determined a
well-defined class predicate F i. Furthermore, since τ is required to affirm that
F obeys the recursive rule, it follows that τ asserts that F i obeys the recursive
rule up to i.
We now argue by induction on formulas that the truth pronouncements made
by τ about the structure V, ∈, , F i forms a satisfaction class for this structure.
In the atomic case, the truth pronouncements about this structure are independent
of the play of the game in which they occur, since this is true for atomic formulas
in the language of set theory and for atomic assertions about , by the rules of the
game, and it true for atomic assertions about F i by our induction hypothesis on
i. Continuing the induction, it follows that the truth pronouncements made about
compound formulas in this structure are similarly independent of the play and obey
the Tarskian conditions, since any violation of this can be easily exposed by having
the interrogator inquire about the constituent formulas, just as in the truth-telling
game. So the claim is also true for F i.
Thus, for every i ∈ I, the strategy τ is providing a satisfaction class for the
structure V, ∈, , F i, which furthermore verifies that the resulting class predi-
cate F i determined by this satisfaction class fulfills the desired recursion relation
up to i. Since these restrictions F i also all agree with one another, the union
of these class predicates is a class predicate F ⊆ I × V that for every i obeys the
desired recursive rule Fi = { x | ϕ(x, i, F i) }. So the recursion has a solution, and
this instance of the principle of first-order transfinite recursion along well-founded

class relations is true.
So far, we have established that the principle of open determinacy implies the
principle ETR of elementary transfinite recursion. In order to improve this impli-
cation to use only clopen determinacy rather than open determinacy, we modify
the game as in lemma 3.1 by requiring the interrogator to count-down during play.
Specifically, the count-down recursion game proceeds just like the recursion game,
except that now we also insist that the interrogator announce on the first move a
natural number n, such that the interrogator loses if the truth-teller survives for at
least n moves (we could have had him count down in the ordinals instead, which
would have made things more flexible for him, but the analysis is essentially the
same). This is now a clopen game, since the winner will be known by the time
this clock expires, either because the truth-teller will violate the Tarski conditions
or the recursion condition before that time, in which case the interrogator wins, or
else because she did not and the clock expired, in which case the truth-teller wins.
So this is a clopen game.
Since the modified version of the game is even harder for the interrogator, there
can still be no winning strategy for the interrogator. So by the principle of clopen
determinacy, there is a winning strategy τ for the truth-teller. This strategy is
allowed to make decisions based on the number n announced by the interrogator
on the first move, and it will no longer necessarily be the case that the theory
declared true by τ will be independent of the interrogator’s play, since the truth-
teller can relax as the time is about to expire, knowing that there isn’t time to be
caught in a violation. Nevertheless, it will be the case, we claim, that the theory
pronounced true by τ for all plays with sufficiently many remaining moves will be
independent of the interrogator’s play. One can see this by observing that if an
assertion ψ(a) is independent in this sense, then also ¬ψ(a) will be independent
in this sense, for otherwise there would be plays with a large number of plays
remaining giving different answers for ¬ψ(a) and we could then challenge directly
afterward with ψ(a), which would have to give different answers or else τ would
not win. Similarly, since τ is winning for the truth-teller, one can see that allowing
the interrogator to specify a bound on the total length of play does not prevent the
arguments above showing that τ describes a coherent solution predicate F ⊆ I × V
satisfying the recursion F (i, x) ↔ ϕ(x, i, F i), provided that one looks only at
plays in which there are sufficiently many moves remaining. There cannot be a
-least i where the value of F (i, x) is not determined in this sense, and so on just
as before. So the strategy must give us a class predicate F and a truth predicate
for V, ∈, , F witnessing that it solves the desired recursion, as desired.
In conclusion, the principle of clopen determinacy for class games is equivalent
to the principle ETR of elementary transfinite recursion along well-founded class
relations.
5. Proving open determinacy in strong theories

It follows from theorems 3 and 9 that the principle of open determinacy for class
games cannot be proved in set theories such as ZFC or GBC, if these theories are
consistent, since there are models of those theories that have no satisfaction class
for first-order truth. We should now like to prove, in contrast, that the principle
of open determinacy for class games can be proved in stronger set theories, such
as Kelley-Morse set theory KM, as well as in GBC + Π11 -comprehension, which is a

proper fragment of KM.
In order to undertake this argument, however, it will be convenient to consider
the theory KM+ , a natural strengthening of Kelley-Morse set theory KM that
we consider in [GHJa]. The theory KM+ extends KM by adding the class-choice
scheme, which asserts of any second-order formula ϕ, that for every class parameter
Z, if for every set x there is a class X with property ϕ(x, X, Z), then there is a class
Y ⊆ V × V , such that for every x we have ϕ(x, Yx , Z), where Yx denotes the xth
slice of Y . Thus, the axiom asserts that if every set x has a class X with a certain
property, then we can choose particular such classes and put them together into a
single class Y in the plane, such that the xth slice Yx is a witness for x. In [GHJa],
we prove that this axiom is not provable in KM itself, thereby revealing what may
be considered an unfortunate weakness of KM. The class-choice scheme can also
naturally be viewed as a class collection axiom, for the class Y gathers together
a sufficient collection of classes Yx witnessing the properties ϕ(x, Yx , Z). In this
light, the weakness of KM in comparison with KM+ is precisely analogous to the
weakness of the theory ZFC- in comparison with the theory ZFC− that we identified
in [GHJb]—these are the theories of ZFC without power set, using replacement or
collection + separation, respectively—since in each case the flawed weaker theory
has replacement but not collection, which leads to various unexpected failures for
the respective former theories.
The natural weakening of the class-choice scheme to the case where ϕ is a first-
order assertion, having only set quantifiers, is called the first-order class-choice
principle, and it is expressible as a single assertion, rather than only as a scheme,
in KM and indeed in GBC + ETR, since in these theories we have first-order truth-
predicates available relative to any class. A still weaker axiom makes the assertion
only for choices over a fixed set, such as the first-order class ω-choice principle:

∀Z ∀n ∈ ω ∃X ϕ(n, X, Z) → ∃Y ⊆ ω × V ∀n ∈ ω ϕ(n, Yn , Z) ,
where ϕ has only first-order quantifiers, and this is also finitely expressible in GBC+
ETR. In our paper [GHJa], we separate these axioms from one another and prove
that none of them is provable in KM, assuming the consistency of an inaccessible
cardinal.
The Π11 -comprehension axiom is the assertion that for any Π11 formula ϕ(x, Z),
with class parameter Z, we may form the class { a | ϕ(a, Z) }. By taking comple-
ments, this is equivalent to Σ11 -comprehension.
Theorem 10. Kelley-Morse set theory KM proves the principle of open de-
terminacy for class games. Indeed, this conclusion is provable in the subtheory
consisting of GBC plus Π11 -comprehension.
We are unsure whether the strictly weaker theory of GBC + ETR suffices for this
conclusion; see additional discussion in section 6.
Proof. Assume GBC plus Π11 -comprehension. In order to make our main
argument more transparent, we shall at first undertake it with the additional as-
sumption that the first-order class-choice principle holds (thus, we work initially
in a fragment of KM+ ). Afterwards, we shall explain how to eliminate our need
for the class-choice principle, and thereby arrive at a proof using just GBC plus
Π11 -comprehension.
Consider any open class game A ⊆ X ω , where A is the open winning condition
and X is the class of allowed moves. We shall show the game is determined. To
do so, notice that for any position p ∈ X <ω , the assertion that a particular class
function σ : X <ω → X is a winning strategy for player I in the game proceeding
from position p is an assertion about σ involving only first-order quantifiers; one
must say simply that every play of the game that proceeds from p and follows σ on
player I’s moves after that, is in A. Thus, the assertion that player I has a winning
strategy for the game starting from position p is a Σ11 assertion about p. Using
Π11 -comprehension, therefore, we may form the class
$ %
W = p∈X <ω
| Player I has a winning strategy
game proceeding from position p
in the .
With this class, we may now carry out a class analogue of one of the usual soft
proofs of the Gale-Stewart theorem, which we mentioned in the introduction of this
article. Namely, if the initial (empty) position of the game is in W , then player
I has a winning strategy, and we are done. Otherwise, the initial node is not in
W , and we simply direct player II to avoid the nodes of W during play. If this
is possible, then it is clearly winning for player II, since he will never land on a
node all of whose extensions are in the open class, since such a node is definitely
in W , and so he will win. To see that player II can avoid the nodes of W , observe
simply that at any position p, if it is player II’s turn to play, and player I does
not have a strategy in the game proceeding from p, then we claim that there must
be at least one move that player II can make, to position p x for some x ∈ X,
such that p x ∈ / W . If not, then p x ∈ W for all moves x, and so for each such
x there is a strategy τx that is winning for player I in the game proceeding from
p x. By the first-order class-choice principle (and this is precisely where we use
our extra assumption), we may gather such strategies τx together into a single class
and thereby construct a strategy for player I that proceeds from position p in such
a way that if player II plays x, then player I follows τx , which is winning for player
I. Thus, there is a winning strategy for player I from position p, contradicting our
assumption that p ∈ / W , and thereby establishing our claim. So if p ∈ / W and it
is player II’s turn to play, then there is a play p x that remains outside of W .
Similarly, if p ∈ / W and it is player I’s turn to play, then clearly there can be no
next move p y placing it inside W , for then player I would have also had a strategy
from position p. Thus, if the initial position is not in W , then player II can play so
as to retain that property (using global choice to pick a particular move realizing
that situation), and player I cannot play so as to get inside W , and this is therefore
a winning strategy for player II. So the game is determined.
The argument above took place in the theory GBC + Π11 -comprehension +
the first-order class-choice principle. And although it may appear at first to have
made a fundamental use of the class-choice principle, we shall nevertheless explain
how to eliminate this use. The first observation to make is that Π11 -comprehension
implies the principle ETR of elementary transfinite recursion along any well-founded
relation. To see this, suppose that is any well-founded relation on a class I and
ϕ(x, i, F, Z) is a formula to serve as the recursive rule. The class of i ∈ I that
are in the domain of some partial solution to the recursion is Σ11 -definable. And
furthermore, all such partial solutions must agree on their common domain, by
an easy inductive argument along . It follows that the union of all the partial
solutions is Σ11 -definable and therefore exists as a class, and it is easily seen to obey
the recursion rule on its domain. So it is a maximal partial solution. If it is not

defined on every slice in I, then there must be a -minimal element i whose ith
slice is not defined; but this is impossible, since we could apply the recursive rule
once more to determine the slice Fi = { x | ϕ(x, i, F i, Z) } to place at i, thereby
producing a partial solution that includes i. So the maximal partial solution is
actually a total solution, verifying this instance of the transfinite recursion principle.
Next, we shall explain how to continue the constructibility hierarchy beyond
Ord. This construction has evidently been discovered and rediscovered several
times in set theory, but rarely published; the earliest reference appears to be the
dissertation of Leslie Tharp [Tha65], although Bob Solovay reportedly also under-
took the construction as an undergraduate student, without publishing it. In the
countable realm, of course, the analogous construction is routine, where one uses
reals to code arbitrary countable structures including models of set theory of the
form Lα , ∈. For classes, suppose that Γ = Ord, ≤Γ is a meta-ordinal, which is
to say, a well-ordered class relation ≤Γ on Ord; this relation need not necessarily
be set-like, and the order type can reach beyond Ord. By ETR, we may iterate
the constructible hierarchy up to Γ, and thereby produce a class model LΓ , ∈Γ of
V = L, whose ordinals have order-type Γ. Specifically, we reserve a class of nodes
to be used for representing the new “(meta-)sets” at each level of the LΓ hierar-
chy, and define ∈Γ recursively, so that at each level, we add all and only the sets
that are definable (from parameters) over the previous structure. To be clear, the
domain of the structure LΓ , ∈Γ is a class LΓ ⊆ V , and the relation ∈Γ is not the
actual ∈ relation, but nevertheless ∈Γ is a well-founded extensional relation in our
original model, and the structure LΓ , ∈Γ looks internally like the constructible
universe. Thus, we have what might be termed merely a code for or presentation of
the fragment LΓ of the constructibility hierarchy up to Γ, which someone outside
the universe might prefer to think of as an actual transitive set.
In order to speak of LΓ in our GBC context, then, we must be aware that
different choices of Γ will lead to different presentations, with sets being represented
differently and by different sets. Nevertheless, we may assume without loss that
the actual sets in L = LOrd are represented in this presentation in some highly
canonical way, so that the ordinals are represented by themselves, for example, and
the other sets are represented by their own singletons (say), and so in particular, all
the various LΓ will agree on their ∈Γ relations for sets constructed before Ord. Also,
using the principle of first-order transfinite recursion, it is easy to see that any two
meta-ordinals Γ and Γ are comparable, in the sense that one of them is (uniquely)
isomorphic to an initial segment of the other, and similarly the structures LΓ and
LΓ admit such coherence as well; in particular, if Γ and Γ are isomorphic, then
so also are the structures LΓ and LΓ . Consider the meta-ordinals Γ for which
there is a larger meta-ordinal Θ, such that LΘ has as an element a well-ordered
structure Γ = Ord, ≤Γ with order-type isomorphic to Γ. In a sense, these are the
meta-ordinals below (Ord+ )L . In this case, there will be an L-least such code Γ in
LΘ . And furthermore, any other meta-ordinal Θ which constructs such a code will
agree on this L-least code. Since we assumed that the ordinals were represented as
themselves in LΘ , we may view Γ as a meta-ordinal in our original model. Thus,
the meta-ordinals Γ realized by a relation in some LΘ have canonical codes, the
meta-ordinals that are L-least with respect to some (and hence all sufficiently large)
LΘ . There is exactly one such code for each meta-ordinal order type that is realized
inside any LΘ .
Now, let L be the collection of classes B ⊆ L that are realized as an element
in some LΓ —these are the classes of the meta-L that are contained in the actual
L—and consider the model L = L, ∈, L. It is not difficult to see that this is a
model of GBC, precisely because the L-hierarchy closes under definability at each
step of the recursion. Furthermore, the existence of canonical codes will allow us
to show that this model satisfies the first-order class-choice principle. Suppose that
L |= ∀b ∃X ϕ(b, X), where ϕ has only first-order quantifiers. For each set b, there
is a class X with property ϕ(b, X), and such a class X exists as a set in some LΓ for
some meta-ordinal Γ. We may consider Γ to be a canonical code for a meta-ordinal
which is minimal with respect to the property of having such an X, and in this
case, the class Γ = Γb is Σ11 -definable (and actually Δ11 -definable) from b. So the
map b → Γb exists as a class in the ground model, and we may therefore form a
meta-ordinal Θ that is larger than all the resulting meta-ordinals Γb . Inside LΘ ,
we may select the L-least Xb witnessing ϕ(b, Xb , Z), and thereby form the class
{ (b, c) | c ∈ Xb }, which fulfills this instance of the first-order class-choice principle
(and the argument easily accommodates class parameters).
A similar idea shows that L satisfies Π11 -comprehension, provided that this was
true in the original model (and indeed L satisfies KM, if this was true in the original
model, and in this case one can also verify the class-choice scheme in L , without
requiring this in V , which shows that Con(KM) → Con(KM+ ); see [GHJa].) It will
be more convenient to establish Σ11 -comprehension, which is equivalent. Consider
a formula of the form ∃X ϕ(b, X), where ϕ has only first-order quantifiers. The
class B = { b ∈ L | ∃X ∈ L ϕ(b, X) } is Σ11 -definable and therefore exists as a class
in our original universe. We need to show it is in L. For each b ∈ B, there is a class
X such that X ∈ L and ϕ(b, X), and such a class X is constructed in some LΓ
at some minimal meta-ordinal stage Γ, which we may assume is a canonical code.
Thus, the map b → Γb is Σ11 -definable, and so it exists as a class. Thus, we may
form a single meta-ordinal Θ larger than all the Γb , and in LΘ , we may define the
set B. So B ∈ L, verifying this instance of Σ11 -comprehension in L , as desired.
One may now check that the construction of the previous paragraph relativizes
to any class Z ⊆ Ord, leading to a model L[Z], ∈, S that satisfies GBC + Π11 -
comprehension + the first-order class-choice principle, in which Z is a class. One
simply carries the class parameter Z through all of the previous arguments. If Z
codes all of V , then the result is a model V, ∈, S, whose first-order part has the
same sets as the original model V .
Using this, we may now prove the theorem. Consider any open class game
A ⊆ X ω , where X is the class of allowed moves. Let Z ⊆ Ord be a class that codes
in some canonical way every set in V and also the classes X and A. The resulting
structure V, ∈, S described in the previous paragraph therefore satisfies GBC+Π11 -
comprehension + the first-order class-choice principle. The game A ⊆ X ω exists in
this structure, and since it is open there, it follows by the first part of the proof of
this theorem that this game is determined in that structure. So there is a strategy
σ ∈ S for the game A that is winning for one of the players. But this is absolute
to our original universe, because the two universes have exactly the same sets and
therefore exactly the same plays of the game. So the game is also determined in our
original universe, and we have thus verified this instance of the principle of open
determinacy for class games.
One might view the previous argument as a proper class analogue of Blass’s
result [Bla72] that computable games have their winning strategies appearing in the
L-hierarchy before the next admissible set, since we found the winning strategies for
the open class game in the meta-L hierarchy on top of the universe. Nevertheless,
we are unsure sure exactly what it takes in the background theory to ensure that
the meta-L structure is actually admissible.
6. Questions
The work of this article suggests numerous questions for further investigation.
Can we weaken the assumption of Π11 -comprehension in theorem 10 to use only
the principle ETR of elementary transfinite recursion over well-founded class rela-
tions? If so, it would follow that open determinacy and clopen determinacy for class
games are both equivalent over GBC to the principle of transfinite recursion, which
would resonate with the corresponding situation in reverse mathematics for games
on the natural numbers, where both open determinacy and clopen determinacy
are equivalent to the principle of transfinite recursion over ACA0 . But perhaps
open determinacy is strictly stronger than clopen determinacy over GBC. Which
strengthening of GBC suffices to prove the meta-L structure we construct in the-
orem 10 is admissible? If this is possible in GBC plus ETR, then the proper class
analogue of the Blass result mentioned earlier might show that open determinacy
and clopen determinacy for classes are equivalent over GBC. Is there a class game
analogue of Martin’s proof [Mar75] of Borel determinacy? What does it take to
prove the class analogue of Borel determinacy for class games? There is a natural
concept of class Borel codes, which in KM+ gives rise to a collection of classes that
is the smallest collection of classes containing the open classes and closed under
countable unions and complements. Are all such class games determined? If κ
is an inaccessible cardinal, then the full second-order structure Vκ , ∈, Vκ+1 is a
model of KM+ that satisfies Borel determinacy for class games. Is there a proper
class analogue of Harvey Friedman’s famous proof [Fri71] that Borel determinacy
requires strength? We have taken up all these questions in current work.
References
[Aud16] G. Audrito. Generic large cardinals and absolutenss. Dottorato in Matematica - XXVII
ciclo, Università degli Studi di Torino, March 2016. Ph.D. dissertation, under supervision
of Matteo Viale.
[AV] G. Audrito and M. Viale. Absoluteness via reflection. arxiv:1404.2111.
[Bla72] A. Blass, Complexity of winning strategies, Discrete Math. 3 (1972), 295–300.
MR0323543
[EH14] C. D. A. Evans and J. D. Hamkins, Transfinite game values in infinite chess, Integers
14 (2014), Paper No. G2, 36. MR3225916
[EHP] C. D. A. Evans, J. D. Hamkins, and N. L. Perlmutter. A position in infinite chess with
game value ω 4 . Manuscript under review, Newton Institute preprint ni15065.
[Fri71] H. M. Friedman, Higher set theory and mathematical practice, Ann. Math. Logic 2
(1970/1971), no. 3, 325–357. MR0284327
[Fuj12] K. Fujimoto, Classes and truths in set theory, Ann. Pure Appl. Logic 163 (2012), no. 11,
1484–1523, DOI 10.1016/j.apal.2011.12.006. MR2959657
[GHJa] V. Gitman, J. D. Hamkins, and T. A. Johnstone. Kelley-Morse set theory and choice
principles for classes. Manuscript in preparation.
[GHJb] V. Gitman, J. D. Hamkins, and T. A. Johnstone, What is the theory ZFC without power
set?, MLQ Math. Log. Q. 62 (2016), no. 4-5, 391–406, DOI 10.1002/malq.201500019.
MR3549557
[GS53] D. Gale and F. M. Stewart, Infinite games with perfect information, Contributions to
the theory of games, vol. 2, Annals of Mathematics Studies, no. 28, Princeton University
Press, Princeton, N. J., 1953, pp. 245–266. MR0054922
[HJ] J. D. Hamkins and T. Johnstone. Strongly uplifting cardinals and the boldface resurrec-
tion axioms. Under review, http://arxiv.org/abs/1403.2788.
[HJ14] J. D. Hamkins and T. A. Johnstone, Resurrection axioms and uplifting cardinals, Arch.
Math. Logic 53 (2014), no. 3-4, 463–485, DOI 10.1007/s00153-014-0374-y. MR3194674
[Mar75] D. A. Martin, Borel determinacy, Ann. of Math. (2) 102 (1975), no. 2, 363–371.
MR0403976
[Mat15] A. R. D. Mathias, In praise of logic: on Bourbaki and (the lack of ) the teaching of logic
(Italian, with English and Italian summaries), Mat. Soc. Cult. Riv. Unione Mat. Ital. (I)
8 (2015), no. 1, 43–74, 165–166. Translated by Alberto Peruzzi. MR3364899
[MS12] A. Montalbán and R. A. Shore, The limits of determinacy in second-order arithmetic,
Proc. Lond. Math. Soc. (3) 104 (2012), no. 2, 223–252, DOI 10.1112/plms/pdr022.
MR2880240
[MT07] M. O. MedSalem and K. Tanaka, Δ03 -determinacy, comprehension and induction, J.
Symbolic Logic 72 (2007), no. 2, 452–462, DOI 10.2178/jsl/1185803618. MR2320285
[MT08] M. O. MedSalem and K. Tanaka, Weak determinacy and iterations of inductive defi-
nitions, Computational prospects of infinity. Part II. Presented talks, Lect. Notes Ser.
Inst. Math. Sci. Natl. Univ. Singap., vol. 15, World Sci. Publ., Hackensack, NJ, 2008,
pp. 333–353, DOI 10.1142/9789812796554 0018. MR2449473
[Sat14] K. Sato, Relative predicativity and dependent recursion in second-order set theory and
higher-order theories, J. Symb. Log. 79 (2014), no. 3, 712–732, DOI 10.1017/jsl.2014.28.
MR3248781
[Sim09] S. G. Simpson, Subsystems of second order arithmetic, 2nd ed., Perspectives in Logic,
Cambridge University Press, Cambridge; Association for Symbolic Logic, Poughkeepsie,
NY, 2009. MR2517689
[Tan90] K. Tanaka, Weak axioms of determinacy and subsystems of analysis. I. Δ02 games, Z.
Math. Logik Grundlag. Math. 36 (1990), no. 6, 481–491, DOI 10.1002/malq.19900360602.
MR1114101
[Tan91] K. Tanaka, Weak axioms of determinacy and subsystems of analysis. II. Σ02 games,
Ann. Pure Appl. Logic 52 (1991), no. 1-2, 181–193, DOI 10.1016/0168-0072(91)90045-N.
International Symposium on Mathematical Logic and its Applications (Nagoya, 1988).
MR1104060
[Tha65] L. H. Tharp, Constructibility in Impredicative Set Theory, ProQuest LLC, Ann Arbor,
MI, 1965. Thesis (Ph.D.)–Massachusetts Institute of Technology. MR2939618
[Wel11] P. D. Welch, Weak systems of determinacy and arithmetical quasi-inductive definitions,
J. Symbolic Logic 76 (2011), no. 2, 418–436, DOI 10.2178/jsl/1305810756. MR2830409
The City University of New York, CUNY Graduate Center, Mathematics Program,
365 Fifth Avenue, New York, New York 10016
E-mail address: vgitman@nylogic.org
URL: http://boolesrings.org/victoriagitman
Philosophy, New York University & Mathematics, Philosophy, Computer Science,

The Graduate Center of The City University of New York, 365 Fifth Avenue, New
York, New York 10016 & Mathematics, College of Staten Island of CUNY
E-mail address: jhamkins@gc.cuny.edu
URL: http://jdh.hamkins.org
Volume 690, 2017
Open problems on ultrafilters and some connections

to the continuum
M. Malliaris and S. Shelah

For Hugh Woodin on the occasion of his birthday.
Abstract. We discuss a range of open problems at the intersection of set

theory, model theory, and general topology, mainly around the construction of
ultrafilters. Along the way we prove a uniqueness theorem for a weak notion
of cut.
Contents
1. The cut spectrum
2. A uniqueness theorem
3. On internal maps
4. Bases for goodness
Appendix
References
Recently we discovered a surprising connection [15] between a model-theoretic

problem from the sixties known as Keisler’s order and a question about cardinal
invariants of the continuum arising from work of Rothberger in the forties [20],
whether p = t. Keisler’s order compares theories via the saturation properties
of their regular ultrapowers. Its study has brought certain connections of model
theory and set theory sharply into focus, which has been very helpful in illuminating
model-theoretic structure. Indications from continuing work on Keisler’s order are
that the connection of the two problems was more than a lucky coincidence, and
that the potential interest of the order and of regular ultrafilters for set theory and
general topology also goes potentially much beyond this one example.
We therefore present here a series of representative open problems in set the-
ory, motivated by these interactions with model theory, and mainly around the
construction of regular ultrafilters. Along the way we prove a uniqueness theo-
rem for a weak notion of cut, Theorem 2.4, and give a new sufficient condition for
2010 Mathematics Subject Classification. Primary 03C20, Secondary 03E05.

Key words and phrases. Regular ultrafilter, Keisler’s order, cut spectrum, good ultrafilter.
The first author was partially supported by NSF grant DMS-1300634 and by a Sloan fellow-
ship. The second author was partially supported by European Research Council grant 338821.
This is manuscript 1069 in Shelah’s list of publications.
2017
145
146 M. MALLIARIS AND S. SHELAH
goodness of an ultrafilter, Lemma 3.2. We note that the emphasis here on regular
ultrafilters reverses a long trend of emphasis in set theory: already by the 1960s,
the model-theoretic and set-theoretic use of ultrafilters had diverged, with model
theorists focusing on regular ultrafilters for their uniform saturation properties and
set theorists often focusing (over uncountable sets) on ultrafilters which were non
regular (e.g. were normal). We hope such a paper will help to centralize and com-
municate intuition and to stimulate work from both sides in this fascinating and
largely unexplored area.
We thank the referee for an attentive and very helpful reading.
1. The cut spectrum

We begin with the problem of determining the possible cut spectrum of a regular
ultrafilter D. (After stating the problem, we will explain all of the terms used.)
General Problem 1.1 (cf. [15] 2.9). For λ ≥ ℵ0 and D a regular ultrafilter
on I, |I| = λ, determine the possible values of
C(D) ={(κ1 , κ2 ) : κ1 , κ2 are infinite and regular,
κ1 + κ2 ≤ |I| and (ω, <)I /D has a (κ1 , κ2 )-cut}.
First, say that a linearly ordered set has a (κ1 , κ2 )-pre-cut if there is a strictly
increasing sequence ā = aα : α < κ1 of cofinality κ1 and a strictly decreasing
sequence b̄ = bβ : β < κ2 of cofinality κ2 such that for α < α < κ1 and
β < β < κ2 we have aα < aα < bβ < bβ . We say the set has a (κ1 , κ2 )-cut if there
are ā, b̄ as above and in addition, there is no c such that for all α < κ1 and β < κ2 ,
aα < c < bβ . (Some authors use “gap” in place of “unfilled cut.” For us a cut is
always unfilled.) We say a filter D is “on I” to mean that D ⊆ P(I) and I ∈ D.
Recall that a filter D on I is called regular if there exist a family of |I| elements
of D, called a regularizing family, such that the intersection of any infinite subfamily
is empty. Equivalently, each t ∈ I belongs to at most finitely many elements of the
regularizing family. Such ultrafilters always exist: given an infinite cardinal λ, let
I = [λ]<ℵ0 and consider the family F = {Xβ : β < λ} where Xβ = {v ∈ I : β ∈ v}.
Then F has the desired regularity property and also the finite intersection property,
so it may be extended to an ultrafilter.1
Regularity of D entails that 1.1 would not change if we used an infinite linear
order other than ω. More precisely, we might have defined:
Definition 1.2. Given an ultrafilter D on I and a linear order X, let
C(D, X) ={(κ1 , κ2 ) : κ1 , κ2 infinite and regular,
κ1 + κ2 ≤ |I| and X I /D has a (κ1 , κ2 )-cut}.
1 Donder [2] proved that in the core model all uniform ultrafilters are regular, and these will
be our main focus here. Still, we note that one outcome of our recent investigations has been a
certain reuniting of work on regular and irregular ultrafilters (e.g. κ-complete ultrafilters where κ
is measurable or compact). Essentially, one can build a good regular filter D0 on I such that the
quotient P(I)/D0 is isomorphic to a certain complete Boolean algebra B. One can then build an
ultrafilter D∗ on B, which need not be regular, to have certain prescribed properties. Combining
D∗ and D0 one obtains a regular ultrafilter on I which inherits some of the given features. This
approach is outlined in [17] §1.3 p. 619 and Definition 2.11 p. 624 through §2.2, p. 628.
OPEN PROBLEMS ON ULTRAFILTERS... 147
Then by regularity we have the following. (Morally speaking, this fact is a

consequence of Keisler’s stronger result on saturation, [8] Theorem 2.1a, discussed
below. A proof is included in the Appendix.)
Fact 1.3. If D is a regular ultrafilter and (N, <N ), (M, <M ) are any two
infinite linearly ordered sets, then C(D, N ) = C(D, M ).
Discussion 1.4. Thus, we are justified in speaking of C(D) and using (ω, <)
for our computations.
In particular, if D is a regular ultrafilter on I, |I| = λ, no new gaps may appear
when replacing ω by other infinite ordinals, and we may use orders such as (Q, <)
to prove facts about C(D, ω). Moreover, we will have (κ, μ) ∈ C(D) if and only if
(μ, κ) ∈ C(D). So we will usually assume κ ≤ μ.
Some useful facts about regular ultrapowers of linear order: the regular ultra-
power (ω, <)I /D will have cofinality > λ, and will contain both strictly increasing
and strictly decreasing sequences of length λ by [8, Theorem 1.5a]. So for every
regular κ ≤ λ there is some θ ≤ 2λ such that X I /D contains a (κ, θ)-cut.2
In what follows we will assume all first-order languages are countable and all
theories complete unless otherwise stated. Given a model M , if M I /D is |I|+ -
saturated let us say that “D saturates M .” The strong incompleteness visible
in the regularizing family gives regular ultrafilters an important model-theoretic
property, namely that if M ≡ N then D saturates M if and only if D saturates
N (see Keisler [8, Corollary 2.1a]). So when D is regular, it is well defined to say
that “D saturates T .” Keisler’s order is precisely the pre-order on complete,
countable theories given by:
Definition 1.5 (Keisler’s order). T1 T2 if for every regular ultrafilter D, if
D saturates T2 then D saturates T1 .
Keisler’s order induces a dual pre-order on regular ultrafilters:
Definition 1.6. D1 uf D2 if for every complete countable theory T , if D1
saturates T then D2 saturates T .
We will keep Keisler’s order and its dual in mind throughout the paper.3
The real interest of C(D) now begins with the fact that there exist ultrafilters
for which C(D) = ∅. This is explained in some detail in [15] pps. 240-242. Briefly,
however, Keisler had defined and proved the existence of so-called good ultrafilters
under GCH [6], an assumption later eliminated by Kunen [10].
Definition 1.7. Say that the ultrafilter D on I is κ+ -good if every monotonic
f : [κ]<ℵ0 → D has a multiplicative refinement, where monotonic means that u ⊆
v ∈ [λ]<ℵ0 implies f (u) ⊇ f (v) and a multiplicative refinement of f is a function
f such that f (u) ⊇ f (u) ∈ D for all finite u and f (u) ∩ f (v) = f (u ∪ v) for all
finite u, v. Say that D is good if it is |I|+ -good.
Keisler then proved that the maximum class in Keisler’s order exists and con-
sists precisely of those theories such that if D is regular and saturates T then D
2 It follows, in model theoretic terminology, that the only failures of |I|+ -saturation in such
an ultrapower come from the cuts visible to C(D).

3 The relation is usually considered as a partial order on the equivalence classes. For further
information on the classification around Keisler’s order, the reader may consult the original paper
of Keisler 1967 [8] or the recent papers [13], [16].
must be good [8]. This implies that uf has a maximum class consisting precisely
of the good regular ultrafilters. Shelah proved in [21] VI.2.6 that the theory of
infinite dense linear order belongs to the maximum Keisler class.4 Together with
Fact 1.3, these theorems imply that if D is a regular ultrafilter on I, then (ω, <)I /D
is |I|+ -saturated if and only if D is good. In other words,
Fact 1.8. If D is a regular ultrafilter, then C(D) = ∅ if and only if D is good.
Now, Fact 1.8 suggests that a first way to observe the strength of an ultrafilter
is to consider how far C(D) is from being empty: “the fewer the cuts in C(D), the
more powerful the ultrafilter.” This idea motivated many of our results in [15].
(A different5 way to measure the strength of an ultrafilter is by the set of theories
which it saturates, as in uf .)
At this point, we remind the reader of the big picture question, 1.1. As stated,
this question is related to one of the most basic questions about regular ultrafilters:
& for regular D, where for D an ultrafilter on I we define
the possible values of μ(D)
μ(D) = min{κ : κ = | i ni /D| for some sequence ni : i ∈ I and κ ≥ ℵ0 }, i.e.
the minimum element in the spectrum of infinite cardinals which are cardinalities
of ultraproducts of finite cardinals mod D. By [8], [21], [9], and [23] this and the
spectrum is fully understood.6
However, the general picture for cuts has proved much more elusive. The main
structure theorem of [15], in the context of ultrapowers, tells us that:
Theorem 1.9 ([15] Theorem 10.25). Let D be a regular ultrafilter on I, |I| ≥ λ.
Then the following are equivalent:
(a) C(D) = ∅.
(b) C(D) has no symmetric cuts, i.e. (κ, κ) ∈
/ C(D) for all κ ≤ λ.
(c) D is good.
Thus, “the first cut is symmetric”, meaning that if κ is minimal such that
(κ1 , κ2 ) ∈ C(D) and κ1 + κ2 = κ then necessarily (κ, κ) ∈ C(D). Several immediate
questions arise. First, what other cuts may appear with that first symmetric cut?
Is every entry in C(D) witnessed by a corresponding symmetric cut?
Problem 1.10. Suppose that D is an ultrafilter on I and C(D) = ∅, so there is
some minimal κ ≤ |I| such that (κ, κ) ∈ C(D). For which, if any, other cardinals
μ < κ may it be the case that also (κ, μ) ∈ C(D)?
Problem 1.11. Suppose (κ1 , κ2 ) ∈ C(D). Is (κ1 + κ2 , κ1 + κ2 ) ∈ C(D)?
The paper [15] gave a new sufficient condition for the cut spectrum to be empty:
existence of upper bounds of paths through trees.
4 This was surprising: after all, Keisler’s proof of the existence of a maximal class essentially
shows that sufficiently rich theories, such as Peano arithmetic, may explicitly code failures of
goodness as failures of saturation. With the theory of linear ordering, there is no such coding.
5 Nonetheless, we do not know how different: for instance, whether every Keisler class cor-
responds to a set of cuts, say, whether C(D) = C(D ) implies that D and D saturate the same
theories. Such questions motivated some recent work separating cut-like properties of ultrafil-
ters from other properties, as in (2) → (3) of [13] Theorem 4.2. This appears to be a question
about how ultrafilters deal with the independence property in model theory, and progress on
understanding Keisler’s order on the simple unstable theories may help to illuminate this.
6 Moreover, μ(D) is an indicator for whether or not M I /D is |I|+ -saturated whenever Th(M)
is countable and stable with the fcp, see for instance [13] §1.1, especially Theorem D(2) there.
Theorem 1.12 ([15] Theorem 10.25). Suppose that D is a regular ultrafilter on

|I| and D has “|I|+ -treetops,” meaning that if (T , ) is any tree, in the ultrapower
N = T I /D any N -strictly increasing sequence of cofinality κ < |I|+ has an upper
bound in N . Then C(D) = ∅.
In particular, we may add to Theorem 1.9 the equivalent condition
(d) D has λ+ -treetops.
In that paper, we had motivated the connection to p and t by defining cardinals
pD = min{κ : (κ1 , κ2 ) ∈ C(D) and κ1 + κ2 = κ} and tD = min{λ : in some D-
ultrapower of a tree there is a strictly increasing λ-indexed sequence which has
no upper bound}. Then Theorem 1.9 essentially follows from showing that for
regular D, the cardinals pD and tD are equal.7 It is natural to ask whether cuts
and unbounded paths have a similar relationship above the size of the index set.
By a tree we will mean a partially ordered set in which the set of predecessors of
any given node is well ordered.
Problem 1.13. Suppose κ1 , κ2 are infinite and regular, λ ≤ κ1 + κ1 < 2λ , and
(ω, <)I /D has a (κ1 , κ2 )-cut. Let κ = κ1 + κ2 . Is it necessarily the case that for
some tree T , in the ultrapower N = T λ /D there is a strictly increasing κ-indexed
sequence with no upper bound in N ?
Some results on the case of (κ1 , κ2 )-cuts in ultrapowers where κ1 + κ2 > 2λ ,
and the index model M is quite saturated, were obtained by Golshani and Shelah
[4].
At this point in our discussion, a very interesting path opens up: the main proof
of [15] is amenable to axiomatic analysis. Namely, that proof proceeded via various
component lemmas analyzing the structure of C(D) under treetops (proving sym-
metry, uniqueness, and so forth). One can consider these component phenomena
in isolation and try to determine their strength.8 In particular, consider:
Definition 1.14 (Uniqueness). Say the regular ultrafilter D on I has unique-
ness for the regular cardinal κ when: if (κ, θ) ∈ C(D) and (κ, θ ) ∈ C(D) then
θ = θ .
Uniqueness was established in the case of regular ultrafilters for κ strictly below
the goodness of D, or equivalently, for D with κ+ -treetops, in [15] Theorem 3.1.
Problem 1.15. Find necessary and sufficient conditions on a regular ultrafilter
D on I so that for all κ = cf(κ) ≤ λ = |I|, D has uniqueness for κ.
7 Connecting these cardinals to the true p and t requires several more steps. First, replace
the pair of a model and its ultrapower by the more general pair of a model and an elementary
extension satisfying certain pseudofiniteness and expansion properties; such a pair is called a
cofinality spectrum problem s. For any such s, there are corresponding cardinals ps and ts , which
specialize to the pD , tD mentioned above in the special case when the c.s.p. arises as a regular
ultrapower. A major theorem of [15] was a proof in ZFC that for any c.s.p. s, ps < ts is impossible.
Assume then for a contradiction that p < t. For a certain well chosen s arising as a model of set
theory and its generic ultrapower in some forcing extension, one can check that ps ≤ p and t ≤ ts ,
so p < t would entail ps < ts , a contradiction.
8 This is part of a broader question of examining that proof, for instance with the goal of
removing the forcing argument from the proof that p = t. A beginning in this direction appears
in our paper [19].
For example, any regular ultrafilter has uniqueness for ℵ0 , but it is also possible
to construct certain failures of uniqueness, as is done in our paper [19]. Remarkably,
it is not known whether uniqueness is strictly weaker than goodness:
Problem 1.16. For which, if any, I does there exist a regular ultrafilter D on
I such that D has uniqueness for all regular κ ≤ |I| but C(D) = ∅?
We shall return to this question. However, it seems likely that Uniqueness is
a much more general phenomenon. Before discussing 1.16 further let us give some
new evidence for this statement.
2. A uniqueness theorem
In this section we prove Theorem 2.4, which leverages Bell’s Theorem (stated
below) to give uniqueness for a weak notion of cut. Recall that if S is a subset of
a partially ordered set, we say S is centered if every finitely many elements of S
have a common lower bound, and σ-centered if it is the union of countably many
centered subsets.
Theorem A (Bell [1]). p is the first cardinal κ for which M Aκ (σ-centered)
fails, i.e., p = min{κ : there are a σ-centered poset P and a collection D of κ
dense subsets of P such that no filter on P meets all members of D}.
Let D be a filter on N, which we will always assume contains all co-finite subsets
of N. By “D + A” we mean the filter generated by D ∪ {A}, which ' isnontrivial
when A ∈ D + , i.e. A is not equivalent to ∅ mod D. We use and to mean
infinite conjunction and disjunction, respectively.
Definition 2.1. Let D be a filter on N and let δ, θ be limit ordinals, usually
regular cardinals. We omit D when D is the filter of cofinite subsets.
(1) A weakly peculiar (δ, θ)-D-cut in ω ω is a pair (fα : α < δ, gβ : β < θ)
of sequences of functions in ω ω such that:
(a) (∀α < α < δ)(fα <D fα )
(b) (∀β < β < θ)(gβ ≤D gβ )
(c) (∀α < δ)(∀β < θ)(fα <D gβ )
(d) for no A ∈ D+ and f ∈ ω ω do we have
(∀α < δ)(∀β < θ)( fα <D+A f <D+A gβ ).
(2) A left peculiar (δ, θ)-D-cut in ω ω is a weakly peculiar such cut which, in
addition, satisfies: '
(e) if A ∈ D+ and f ∈ ω ω and α<δ fα <D+A f then β<θ gβ ≤D+A f .
Definition 2.2. We define
C(D, ω ω) = {(κ1 , κ2 ) : κ1 , κ2 regular and there is a left peculiar (κ1 , κ2 )-D-cut}.
As before, when D is the co-finite filter, we may omit it.
Note that Definition 2.2 differs from the same notation as used in [15, 14.12(2)].
Left peculiar is a nontrivial weakening of peculiar.9
The following cardinals will be useful in describing the uniqueness spectrum.
9 For example, it can be shown that if (f¯, ḡ) is a left-peculiar and right-peculiar (κ , κ )-cut
1 2
and κ1 = κ2 then some c.c.c. forcing notion forces that (f¯, ḡ) is left peculiar but not right peculiar.
Definition 2.3. With C(ω ω) as in 2.2, we define:

(a) specuq = {κ : if (κ, θ ) ∈ C(ω ω) for = 0, 1 then θ1 = θ2 }.
(b) uq = min{specuq \ℵ1 } = {κ : κ ≥ ℵ1 , κ ∈ specuq }.
(c) nuq = min{κ : κ = cf(κ) > ℵ0 , κ ∈ / specuq }.
In the following theorem, it will follow from the remark after equation (3) that
θ ≤ 2ℵ0 for = 0, 1. However, the theorem does not follow from the construction
in [15] Section 14 because there we collapse the continuum to h which is ≥ p but
≤ b.
Theorem 2.4. p ≤ nuq.

Proof. Suppose that (κ, θ ) ∈ C(ω ω) for = 0, 1, exemplified by (f , g ) for
= 0, 1. Assume κ < p and we will prove that θ0 = θ1 . Throughout the proof, let
D denote the co-finite filter.
We may assume there is a sequence kn∗ : n ∈ N of integers such that for
= 0, 1 and α < κ, fα (n) ≤ kn∗ for all but finitely many n and for = 0, 1 and
β < θ , gβ ≤ kn∗ for all but finitely many n. For instance, kn∗ = g00 (n) + g01 (n) + 1
will suffice. We first consider whether, on infinitely many n ∈ N, we may adjust
the spacing of elements so that the increasing sequences of the two cuts (or cofinal
subsets of the same) strictly interlace. The claim asserts that this is sufficient to
prove the decreasing sequences must have the same cofinality.
Claim 2.5 (A sufficient condition). To show that θ0 = θ1 it would suffice to
show that there exist an infinite A ⊆ N, a club E of κ, a sequence k̄ = kn : n ∈
A ∈ A N, and sequences of functions π̄ 0 = πn0 : n ∈ A, π̄ 1 = πn1 : n ∈ A such
that:
(a) for = 0, 1 and β < θ , for all but finitely many n ∈ A, gβ (n) ∈ [0, kn ).
(b) for = 0, 1 and every n ∈ A, the function πn : [0, kn ) → [0, kn ) is
nondecreasing and onto an initial segment.
(c) if α < β are from E then for all but finitely many n ∈ A,
max{πn0 (fα0 (n)), πn1 (fα1 (n))} < min{πn0 (fβ0 (n)), πn1 (fβ1 (n))}.
Proof of Claim 2.5. Suppose such objects exist. Renaming, without loss of
generality E = κ. When n ∈ / A, let πn be the identity and let kn = kn∗ .
First we give a name to each fα after its movement by the maps πn : for each
= 0, 1 and α < κ let fα ∈ ω ω be the function defined (for all but at most finitely
many n) by
(1) fα (n) = πn (fα (n)).
Define gβ analogously for β < θ . By hypothesis (c) of the Claim,
(2) α < β < κ =⇒ fα1 <D+A fβ0 ∧ fα0 <D+A fβ1 .
Recall that fα (n) < fβ (n) for all but finitely many n by definition of C(ω ω), and
that πn is nondecreasing and is the identity for n ∈ N \ A. So for = 0, 1,
(3) α < κ ∧ β0 < β1 < θ =⇒ fα <D fα+1

≤D gβ 1 ≤D gβ 0 .
Note that β < β < θ implies only gβ ≤D gβ , not necessarily <D , however the
sequence cannot be eventually constant (mod D). If it were, then writing g∗ for the
eventually constant value, the function defined by g∗ (n) − 1 for each n such that
g∗ (n) ≥ 1 (and 0 otherwise) would contradict 2.1(1)(d). So for = 0, 1,
(4) for every β < θ there is β with β < β < θ such that gβ <D gβ .
Let us check that for = 0, 1, and every infinite A ⊆ A,
(5) no h ∈ ω ω satisfies: for all α < κ, β < θ we have fα ≤D+A h ≤D+A gβ .
such an h. Let hmax (n) = max{k

Suppose for a contradiction there were ≤ kn :
πn (k) ≤ h(n)} and hmin (n) = min {k ≤ kn : πn (k) ≥ h(n)} ∪ {kn } . For each
α < κ, we have that fα <D fα+1

≤D+A h, so fα <D+A hmax . Likewise, recalling

(4), hmin <D+A gβ . As the πn are onto an initial segment, hmax = hmin =: h.
Then h and A contradict the fact that (f¯ , ḡ ) is weakly peculiar, 2.1(1)(d). This
proves (5). Let us check that moreover for = 0, 1, for any infinite A ⊆ A and any
h ∈ ω ω,
(
(6) if fα <D+A h then gβ ≤D+A h.
α<κ β<θ
Suppose h satisfies the “if” clause. As before we may define h = hmax , and by a
similar argument, for each α < κ, fα <D+A h. As (f¯ , ḡ ) is left peculiar10 , using
A for A in 2.1(2)(e) there is γ < θ such that gγ ≤D+A h. Hence gγ ≤D+A h,
which completes the check of (6).
Now we use the interlacing of the ascending sequences on A. Since θ0 , θ1 are
regular cardinals, to prove the claim it will suffice to show that the descending
sequences also interlace on A, i.e. :
(I) if γ < θ0 then for some β < θ1 we have gβ1 ≤D+A gγ0 , and
(II) if γ < θ1 then for some β < θ0 we have gβ0 ≤D+A gγ1 .
As these are symmetric, it suffices to prove (I). Let γ < θ0 be given. For all α < κ,
by (2) and (3),
(7) fα1 <D+A fα+1
0
≤D gγ0 .
Thus, fα1 <D+A gγ0 for all α < κ. Applying equation (6) with the present A we find
β < θ1 such that gβ1 ≤D+A gγ0 . This proves (I) and the claim. Claim 2.5.
Continuing the proof of Theorem 2.4, we define a forcing notion Q as follows:
A. p ∈ Q iff p = (v, m, u, π 0 , π 1 , k̄) where:
(i) m ∈ N
(ii) v ⊆ N is finite, with max v ≤ m
(iii) u ⊆ κ is finite
(iv) for = 0, 1, fα [m, ω) : α ∈ u is increasing
(v) π = πn : n ∈ v
(vi) k̄ = kn : n ∈ v, with kn ≥ g00 (n) + g01 (n) + 1 for each n ∈ v, and
πn : kn → kn is nondecreasing onto an initial segment.
B. p ≤Q q when:
(i) vp ⊆ vq , mp ≤ mq , up ⊆ uq , vq \ vp ⊆ [mp , mq ),

(ii) πn,p
= πn,q for n ∈ vp , and
10 The
hypothesis ‘left peculiar’ is essential in proving 2.5. If the cut is only assumed to be
weakly peculiar, it is possible to force a counterexample to the claim using κ > p, noting that
κ < p was not used in that proof. In fact, for weakly peculiar cuts, it is possible to force a
counterexample to the theorem.
(iii) if n ∈ vq \ vp and α < β are from up then

0
max{πn,q (fα0 (n)), πn,q
1
(fα1 (n))} < min{πn,q
0
(fβ0 (n)), πn,q
1
(fβ1 (n))}.
Now we record some properties of Q. First, letting only u vary, we have that fixing
an appropriate five-tuple x, the set {p ∈ Q : (vp , mp , π 0p , π 1p , k̄p ) = x} is directed.
This shows that Q is σ-centered. Second, if α < κ then the set
Iα1 = {p ∈ Q : α ∈ up }
is dense open. Third, if m ∈ N the set
Im
2
= {p ∈ Q : mp ≥ m}
is dense open. Fourth, if m ∈ N the set
Im
3
= {p ∈ Q : vp ⊆ {0, . . . m − 1}}
is dense open, since given p, we may choose i > mp such that fα (i) : α ∈ u is
increasing for = 0, 1. Together, letting α < κ, m ∈ N vary, we have a collection
of no more than κ < p dense open subsets of Q, and Q is σ-centered.
By Bell’s theorem, there exists a filter G
on Q which is generic for the collection
of dense open sets just built. Then A = {vp : p ∈ G}, E = κ, and πm
= πp,m
for some p ∈ G satisfy the hypotheses of Claim 2.5. This completes the proof of
Theorem 2.4.
Informally, the connection of p to cuts in linear orders, especially in the context

of [15], may be described by saying: from our present perspective, p expresses some
kind of saturation (which is maximal from that perspective). There are natural
weaker notions corresponding to saturation of some combinatorially natural theo-
ries, such as Tf eq , the theory of an existentially closed model of a parametrized fam-
ily of independent equivalence relations, or the theory of the generic tetrahedron-
free three-hypergraph, called T3,2 in [18] Definition 0.4. Do these weaker notions
also give rise to natural and interesting cardinal invariants? In the remainder of
the paper, we will touch only on a part of this question, having to do with Tf eq and
its relation to the existence of certain internal maps.
3. On internal maps
Returning to the context of regular ultrafilters, it turns out that the phenome-
non of Uniqueness in the p = t proof from [15] in fact hides something stronger. If
N = M I /D is an ultrapower and g : N → N a partial map, we call g internal (or
induced ) if it genuinely comes from the index models, that is, if we could expand
the language of N by a new binary relation symbol R and for each t ∈ I, we could
choose Mt to be M expanded by the new symbol R interpreted as the graph of a
partial map from M to M , in such a way that in the ultraproduct (Mt )I /D the
relation R is the graph of g.
What was shown in [15] was not only uniqueness, but the a priori stronger fact
that uniqueness is witnessed by internal order preserving maps.
Fact 3.1 ([15] Cor. 3.7). If D is a regular ultrafilter on I and D is good (or
equivalently: D has |I|+ -treetops) then for any κ ≤ |I| the following holds:
) )
κ = κ (D):
for any two strictly increasing κ-indexed sequences aα : α < κ,
bα : α < κ of elements of N = (ω, <)I /D, there is an internal

partial order-preserving map f : N → N extending aα → bα .
) )
We might have defined κ,M = κ,M (D) for M an infinite linear order, with
) )
κ = κ,ω . However, by an argument similar to that)
of Fact 1.3, one can use
regularity of D to show that when κ ≤ |I| the property κ,M does not depend on
the choice of linear order.
Curiously, this strengthening turns out to already be equivalent to goodness.
Lemma 3.2. Let D be a regular ultrafilter on I.
)
(a) Let κ ≤ |I| be regular.
) If κ from 3.1 holds, then (κ, κ) ∈
/ C(D).
(b) In particular, if κ holds for all regular κ ≤ |I| then D is good.
Proof of Lemma 3.2. Clause (b) follows from clause (a) by Theorem 1.9.
To prove clause
) (a), recall
) from the remarks before the proof that having fixed
D, we have that κ ⇐⇒ κ,Q and by Fact 1.3, C(D) = C(D, Q). So it will suffice
)
to show that if κ,Q then N = M λ /D contains no (κ, κ)-cut. As ultrapowers
commute with reducts, we may further assume M = (Q; +, <), i.e. our Q has
the usual structure of an ordered additive group. Suppose then that (ai : i <
κ, bi : i < κ) is a (κ, κ)-pre-cut of N . Without loss of generality, a is strictly
increasing and b is strictly decreasing. Using the available addition and subtraction,
let d = b0 + 1 and define a sequence bi : i < κ of elements of N by: bi = d − bi .
Clearly bi : i < κ is a strictly increasing sequence.
By assumption, there is an internal order-preserving partial bijection f on N
with {ai : i < κ} ⊆ dom(f ) and {bi : i < κ} ⊆ range(f ) such that i < κ =⇒
f (ai ) = bi . (Actually, an internal map between cofinal subsequences would suffice.)
In the language with a symbol for f , clearly dom(f ) is a definable set.
Let X = {a ∈ dom(f ) : a < d − f (a)}. Clearly X is an internal set, since it is
definable from the internal functions +, f . First we check that X is nonempty. For
each i < κ, ai ∈ dom(f ) and d−f (ai ) = d−bi = d−(d−bi ) = bi . Since ai < bi by our
original choice of sequences, this shows that ai ∈ X for each i < κ. Second, since f is
order preserving, X is an initial segment of dom(f ). Third, for each i < κ, consider
bi . If bi ∈
/ dom(f ), clearly bi ∈/ X. Otherwise, recalling that ai < bi and f is order
preserving, necessarily f (ai ) < f (bi ). Then d − f (bi ) < d − f (ai ) = d − bi = bi . So
for each i < κ, bi ∈ / X. We have shown that the pre-cut (ai : i < κ, bi : i < κ) is
internally definable, therefore realized (=filled) in the ultrapower N . This completes
the proof.
Thus, if there exist ultrafilters satisfying Problem 1.16, uniqueness for such
ultrafilters will not always be witnessed by internal order preserving maps.
This line of questioning leads us to recall an ultrafilter property, called “good
for equality,” from work of Malliaris. For simplicity, we focus here on the case
λ = |I| though the definition makes sense for any infinite λ ≤ |I|.
Fact 3.3 ([11], [12]). The regular ultrafilter D on I is good for equality if and
only if for any two sequences ai : i < λ and bi : i < λ of distinct elements of
N = ω I /D, there exists an internal bijection h : N → N such that h(ai ) = bi for
all i < λ. Informally, say that D admits internal maps between small sets.
For simplicity, we focus here on the case λ = |I| though the definition makes
sense for any infinite λ ≤ |I|.
The original motivation for this property came from its interesting model-
theoretic content: for λ ≤ |I|, “D is λ-good for equality” is necessary and sufficient
for D to λ+ -saturate the theory Tf eq mentioned in Section 2. A table connecting
the half-dozen known properties of ultrafilters and of the theories which require
these properties for saturation is in [13] Section 4.
Here is a very simply stated question: for a regular ultrafilter, can existence of
internal maps be separated from existence of internal order-preserving maps? This
is probably the most pressing ultrafilter construction problem from the point of view
of Keisler’s order, and is likely to require very new ideas in ultrafilter construction.
Problem 3.4. Let D be a regular ultrafilter on I, |I| ≥ λ. Suppose D admits
internal maps between sets of size λ. Must it be the case that for any two strictly
increasing sequences ai : i < λ, bi : i < λ of elements of N = ω I /D, there exists
an internal order-preserving partial function h with dom(h) ⊆ N and range(H) ⊆
N such that h(ai ) = bi for all i < λ?
To summarize and restate the construction problem:
General Problem 3.5. Let D be a regular ultrafilter.
If D is good for equality, is C(D) = ∅?
Equivalently: if D is good for equality, must D be good?
Equivalently: if D admits internal maps between small sets in the sense captured
by ‘good for equality,’, must D admit internal order-preserving maps between small
sequences in the sense captured by Problem 3.4?
We note here that a negative answer to Problem 3.5, separating good for equal-
ity from good in ZFC, is likely to also solve an old question of Dow from 1985 [3].
We had given a positive answer to this question in [13] Corollary 1.11, assuming a
measurable cardinal. Again, we state the question before defining the terms.
Problem 3.6 (Dow 1985 [3]). Do there exist regular ultrafilters which are λ+ -
OK but not λ+ -good?
The connection of this problem to 3.5 is that any ultrafilter on I which is good
for equality is so-called flexible, therefore |I|-OK.11 The condition“D is λ-OK”
weakens “D is λ+ -good” by asking that only the monotone functions f : [λ]<ℵ0 → D
such that |u| = |v| implies f (u) = f (v) have a multiplicative refinement. (It is
unfortunate that the cardinals are off by a plus sign, but for clarity in these brief
remarks, we retain the notation of the original sources. That is, Problem 3.6 asks
that all ‘nice’ maps from [λ+ ]<ℵ0 to D all have multiplicative refinements, despite
the fact that some maps from [κ]<ℵ0 to D do not, for some κ ≤ λ. One could
presumably solve Problem 3.5 using just κ = λ+ , but we would expect such a
solution to also shed light on the case of κ ≤ λ.)
We may also consider a cardinal invariant analogue of 3.5. Recalling 2.3, define
specD,uqm = { κ : if (κ, θ ) ∈ C(D, ω ω) for = 0, 1 , then θ1 = θ2 and there is
an internal order preserving partial map which witnesses it }. Here “uqm” is for
“unique mapping,” noting that the definition of internal map makes sense also for
reduced products.
11 The original proof goes by showing: there is a minimal T P theory in Keisler’s order, [11]
2
Corollary 6.10; D is good for equality iff it saturates this minimal T P2 theory, [11] Lemma 6.7
plus [12] Theorem 5.21; if D saturates any theory with T P2 , in particular the minimal theory,
then it is flexible, [11] Lemma 8.8; flexible implies OK, [11] Claim 8.4.
Problem 3.7. Is specuq = specuqm ? At the very least, are the derived cardinal
invariants equal?
To conclude the discussion of Problem 3.5, it is worth underlining that there is a
rich model theory related to these problems, and a successful ultrafilter construction
may provide the most direct way of illuminating this. It concerns the structure of
the so-called non-simple theories, the large region on the right in the diagram.
The picture, explained in [18], illustrates the known map of Keisler’s order.
The lines indicate dividing lines and the white regions are not yet mapped. The
arrow indicates the location of a very surprising recent discovery, the existence of
an infinite descending chain, [18] Theorem 6.6. The large shaded region on the
right is the maximum class; we don’t know whether it may encompass all non-
simple theories. This hinges precisely on the question of whether good and good
for equality coincide: this is because on one hand, as mentioned in the preceeding
footnote, there is a minimum Keisler class among the non-simple theories and its
members are precisely those theories T such that if D is regular and good for T then
D is good for equality, and on the other hand, there is a maximum Keisler class
whose members are precisely those theories T such that if D is regular and good
for T then D is good. Thus, progress on the set-theoretic problem of constructing
regular ultrafilters which are good for equality but not good would establish a
nontrivial gap in the model-theoretic complexity of theories in these two classes.12
4. Bases for goodness

In this section we describe one problem illustrative of the fact that model-
theoretic ‘patterns’ arising from realization or omission of types in ultrapowers
may usefully reflect back onto the structure of the underlying ultrafilter. To begin,
we explain a result mentioned above.
Fact 4.1 (Keisler [5]). If D is a regular ultrafilter on λ, then D is good if
and only if for every complete countable theory T and any M |= T , M λ /D is λ+ -
saturated (i.e. if and only if for every complete countable theory T , D is good for
T ).
Why? Consider a complete, countable theory T and a regular ultrafilter D on
λ. Given an ultrapower N = M λ /D, we fix in advance a lifting to M λ so that the
coordinate projections a[t] are well defined for any a ∈ N and t ∈ λ. When ā is a
tuple, write ā[t] for ai [t] : i < lg(ā). The following fact gives a direct connection
12 In model-theoretic language, this would separate the Keisler-class of T
f eq (the minimum
class among theories with the tree property) and the Keisler-class of theories with the strong tree
property SOP2 .
between realization of types and multiplicative refinements for certain monotonic

functions. For a proof, see [13] 1.7-1.8.
Fact 4.2. Let M |= T and N = M λ /D. Let Xα : α < λ be a regularizing
family for D. Let p(x) = {ϕα (x; āα ) : α < λ} enumerate some consistent partial
type in the ultrapower N . Consider the map d on [λ]<ℵ0 given by:
( *
s → {t ∈ λ : M |= ∃x {ϕα (x; āα [t]) : α ∈ s} ∩ Xα .
α∈s
Then d is a monotonic map into D by L os’ theorem, and p is realized if and only
if d has a multiplicative refinement.
Fact 4.1 then follows from showing that in any sufficiently rich theory (set
theory, Peano arithmetic) all relevant monotonic functions may be represented by
appropriate types: one can directly code a failure of multiplicativity in the ultrafilter
as the omission of a type in the ultrapower.
From this discussion, the reader may already guess that Keisler’s order reflects
a certain structure on the family of all monotonic functions from [λ]<ℵ0 → D which
is largely invisible to current tools. For instance, as was discussed above, the theory
T∗ = T h(Q, <) is obviously much less expressive than Peano arithmetic. However,
T∗ and P A are in the same class, the maximum class, in Keisler’s order. This means
that in order that every monotonic function from [λ]<ℵ0 to D have a multiplicative
refinement, it is sufficient to have multiplicative refinements for the functions arising
as codes for types describing cuts in linear order.
General Problem 4.3. Give an internal set-theoretic characterization of a
family F ⊆ [λ]<ℵ0 → D of monotonic functions which is in some sense minimal
for the property that if every f ∈ F has a multiplicative refinement then D is good.
The following rephrasing of Problem 3.5 may be a useful weakening of 4.3.
H(μ) are the sets hereditarily of cardinality less than μ for some sufficiently large
μ.
Problem 4.4. Suppose D is a regular ultrafilter on λ such that the structure
N = (H(μ), )λ /D is “λ+ -saturated for functions,” i.e. if f is a partial function
with domain of cardinality ≤ λ from N to N then there is an internal function
which extends it. Is D good?
Progress on the set-theoretic Problems 4.3 and 4.4 may help with a longstanding
model-theoretic problem: finding a model-theoretic characterization of the maximal
Keisler class. At present we have only the set-theoretic characterization given by
Keisler, i.e. T is -maximum if and only if the regular ultrafilters which are good
for T are precisely the good regular ultrafilters. The most recent advance on this
question, a new sufficient model-theoretic condition for maximality given in [15]
Theorem 1.11, led to many of the results and ideas in the present paper, and surely
there is much more to be said.
Appendix
Proof of Fact 1.3. Suppose (κ, θ) ∈ C(D, N ), witnessed by (aα : α <
κ, bβ : β < θ) in the ultrapower N I /D. Fix a lifting N I /D → N , so that for each
t ∈ I the coordinate projections aα [t], bβ [t] are well defined. Let A = {aα : α < κ}
and B = {bβ : β < θ}. Since D is regular, we may find a regularizing family and
choose κ+θ distinct elements of this family which we index as {Xα : α < κ}∪{Xβ :
β < θ}. Define a map f : A ∪ B → D by aα → Xα for α < κ and bβ → Xβ for
β < θ. For each t ∈ I, let A[t] = {aα [t] : t ∈ f (aα )} and B[t] = {bβ [t] : t ∈ f (bβ )}.
For each t ∈ I, both A[t] and B[t] are finite, and so A[t] ∪ B[t] is a finite, discretely
ordered set in N . Let it : A[t] ∪ B[t] → M be an order preserving injection. (For
greater transparency, we may choose this map so that if x < y are from dom(it )
and N |= (∃z)(x < z < y) then M |= (∃z)(it (x) < z < it (y)); but this isn’t nec-
essary.) Fix an element 0 of M . For α < κ and t ∈ I, define cα,t to be it (aα [t])
if t ∈ Xα , and 0 otherwise. For β < θ and t ∈ I, define dβ,t to be it (bβ [t]) if
t ∈ Xβ , and 0 otherwise. Let cα := cα,t : t ∈ I/D ∈ M I /D, and likewise let
dβ = dβ,t : t ∈ I/D ∈ M I /D. By L os’ theorem, (aα : α < κ, bβ : β < θ) is a
cut in N I /D if and only if (bα : α < κ, dβ : β < θ) is a cut in M I /D, and so
(κ, θ) ∈ C(D, M ).
By using M = N in Fact 1.3 and choosing each it in the proof of that Fact to
be order reversing, (κ, μ) ∈ C(D) if and only if (μ, κ) ∈ C(D).
References
[1] M. G. Bell, On the combinatorial principle P (c), Fund. Math. 114 (1981), no. 2, 149–157.
MR643555
[2] H.-D. Donder, Regularity of ultrafilters and the core model, Israel J. Math. 63 (1988), no. 3,
289–322, DOI 10.1007/BF02778036. MR969944
[3] A. Dow, Good and OK ultrafilters, Trans. Amer. Math. Soc. 290 (1985), no. 1, 145–160, DOI
10.2307/1999788. MR787959
[4] M. Golshani and S. Shelah, “On cuts in ultraproducts of linear orders, I.” [GoSh:1075].
[5] H. J. Keisler, Ultraproducts and saturated models, Nederl. Akad. Wetensch. Proc. Ser. A 67
= Indag. Math. 26 (1964), 178–186. MR0168483
[6] H. J. Keisler, Good ideals in fields of sets, Ann. of Math. (2) 79 (1964), 338–359. MR0166105
[7] H. J. Keisler, Ultraproducts of finite sets, J. Symbolic Logic 32 (1967), 47–57. MR0235998
[8] H. J. Keisler, Ultraproducts which are not saturated, J. Symbolic Logic 32 (1967), 23–46.
MR0218224
[9] S. Koppelberg, Cardinalities of ultraproducts of finite sets, J. Symbolic Logic 45 (1980), no. 3,
574–584, DOI 10.2307/2273424. MR583375
[10] K. Kunen, Ultrafilters and independent sets, Trans. Amer. Math. Soc. 172 (1972), 299–306.
MR0314619
[11] M. E. Malliaris, Hypergraph sequences as a tool for saturation of ultrapowers, J. Symbolic
Logic 77 (2012), no. 1, 195–223, DOI 10.2178/jsl/1327068699. MR2951637
[12] M. E. Malliaris, Independence, order, and the interaction of ultrafilters and theories, Ann.
Pure Appl. Logic 163 (2012), no. 11, 1580–1595, DOI 10.1016/j.apal.2011.12.010. MR2959661
[13] M. Malliaris and S. Shelah, Constructing regular ultrafilters from a model-theoretic point of
view, Trans. Amer. Math. Soc. 367 (2015), no. 11, 8139–8173, DOI 10.1090/S0002-9947-2015-
06303-X. MR3391912
[14] M. Malliaris and S. Shelah, A dividing line within simple unstable theories, Adv. Math. 249
(2013), 250–288, DOI 10.1016/j.aim.2013.08.027. MR3116572
[15] M. Malliaris and S. Shelah, Cofinality spectrum theorems in model theory, set theory, and
general topology, J. Amer. Math. Soc. 29 (2016), no. 1, 237–297, DOI 10.1090/jams830.
MR3402699
[16] M. Malliaris and S. Shelah, General topology meets model theory, on p and t, Proc. Natl. Acad.
Sci. USA 110 (2013), no. 33, 13300–13305, DOI 10.1073/pnas.1306114110. MR3105597
[17] M. Malliaris and S. Shelah, Existence of optimal ultrafilters and the fundamental complex-
ity of simple theories, Adv. Math. 290 (2016), 614–681, DOI 10.1016/j.aim.2015.12.009.
MR3451934
[18] M. Malliaris and S. Shelah, Keisler’s order has infinitely many classes, Israel Journal of
Mathematics, to appear
[19] M. Malliaris and S. Shelah, Cofinality spectrum problems: the axiomatic approach, Topology
Appl. 213 (2016), 50–79, DOI 10.1016/j.topol.2016.08.019. MR3563070
[20] F. Rothberger, On some problems of Hausdorff and of Sierpiński, Fund. Math. 35 (1948),
29–46. MR0029958
[21] S. Shelah, Classification Theory, North-Holland, 1978. MR0513226. Revised edition: 1990.
MR1083551
[22] S. Shelah, A comment on “p < t”, Canad. Math. Bull. 52 (2009), no. 2, 303–314, DOI
10.4153/CMB-2009-033-4. MR2518968
[23] S. Shelah, “The spectrum of ultraproducts of finite cardinals for an ultrafilter.” arXiv:
1312.6780
Department of Mathematics, University of Chicago, 5734 S. University Avenue,

Chicago, Illinois 60637
E-mail address: mem@math.uchicago.edu
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The He-
brew University of Jerusalem, Jerusalem, 91904, Israel, and Department of Mathemat-
ics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110
Frelinghuysen Road, Piscataway, New Jersey 08854-8019
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at
Volume 690, 2017
Obtaining Woodin’s cardinals
P. D. Welch
Abstract. Since the 1980’s work on Projective Determinacy and ADL(Ê)

the concept of Woodin cardinal has become to be seen as central in the the-
ory of large cardinals and inner model theory. The use by Woodin himself
of a background assumption in many arguments that the universe contains
unboundedly many such cardinals again draws attention to the centrality of
this concept.
As is well known the Reflection Principles dating to a more classicial era
only provide large cardinals consistent with V = L, and not the wherewithal
for such theorems on absoluteness under set forcing that Woodin has proven.
We discuss here a reflection principle derived from weak sub-compactness
that implies the existence of a proper class of measurable Woodin cardinals -
thus providing adequate background assumptions for many of Woodin’s abso-
luteness results in his work.
O, there has been much throwing about of brains.

Guildenstern; Hamlet II.2
Contents
1. Introduction
2. The Cantorian versus the Zermelian realms
3. Reflection
4. The Ackermannian realm and reflection
5. Global Reflection
6. Is GRP a reflection principle?
7. Strengthening GRP?
8. Conclusion
References
2010 Mathematics Subject Classification. Primary 03E55; Secondary 03E05.

Key words and phrases. Woodin cardinals, reflection.
2017
c P. D. Welch
161
162 P. D. WELCH
1. Introduction
This article1 is not a history of the origins of Woodin’s notion of large cardinal
now named after him, which was so central in the arguments used to establish
Projective Determinacy by Martin and Steel, and Woodin himself for ADL(Ê) ; this
history is told elsewhere - see for example [9]. The ubiquity of Woodin cardinals is
attested by the literature today not on just determinacy issues, but on so very many
of the consistency results with which we gauge the strength of set theoretic objects.
However it is not due to the power of infinitely many Woodin cardinals to settle the
question of definable determinacy (whether taken in the PD or ADL(Ê) form) or as
Woodin has written [26] to give us as good as complete a theory of the hereditarily
countable sets, HC, as possible, but rather the use of unboundedly many Woodin
cardinals throughout the ordinals as an, again uniquitous, background assumption
for many of his, and others’ work on, for example, establishing the absoluteness of
many properties of our universe, most typically into imagined generic extensions of
the universe V by set-sized forcing notions.
If a hoped for “reduction in incompleteness” over our standard axioms of ZFC
is to be achieved by the adoption of new axioms, and if we are to attempt to fully
justify those axioms, then arguing for an axiom that yields a proper class of Woodin
cardinals is an excellent place to start. Let UW abbreviate this axiom (for ‘there
exist Unboundedly many Woodin cardinals’).
What conception of set or universe of sets can we have that will deliver this for
us? That ZFC could and should be extended was famously pleaded for by Gödel
in [7] which is by now a locus classicus:
“the axioms of set theory by no means form a system closed in itself, but, quite
on the contrary, the very concept of set on which they are based suggests their
extension by new axioms which assert the existence of still further iterations of the
operation ‘set of ’.”
. . . [The ZFC axioms may be] supplemented without arbitrariness by new axioms
which only unfold the content of the concept of set”.
A discussion of the nature of intrinsic necessity in the words of Gödel, or of
intrinsic versus extrinsic justification should probably now be made but I shall
short-circuit this by referring the reader to Koellner’s article [10] and discussion.
It is not my intention to wade in here. The discussion here is about what possible
“conception of set” could lead to UW.
2. The Cantorian versus the Zermelian realms

Cantor’s discoveries and advances were made as a mathematician would work:
in a non-formalised manner (and even that phrase is anachronistic). His viewpoint
concerning the world of order types and cardinalities would be formed in an intuitive
fashion. In the past it was stated that Cantor’s views were that of a ‘naive set
theorist’, a description not as usually used, but with rather too much emphasis on
‘naive’. Now, however we realise that in fact he was quite aware of the pitfalls of
what we would call the set/class distinction. At different stages of his career he
1 This paper was partially prepared whilst the author was a Simons Foundation Visiting
Fellow at the Isaac Newton Institute for Mathematical Sciences in the programme ‘Mathematical,
Foundational and Computational Aspects of the Higher Infinite’ (HIF), supported by EPSRC
grant EP/K032208/1, to both of which the author would like to express his gratitude.
OBTAINING WOODIN’S CARDINALS 163
used the phrase “The Absolute Infinite” or around the time of the publication of
Burali-Forti (1897) - “inconsistent multiplicities”, or later - both.
In a letter to Dedekind (1899)[5]:
A multiplicity can be of such a nature, that the assumption of
the ‘togetherness’ (‘Zusammenseins’) of a multiplicity’s elements
leads to a contradiction, so that it is impossible to conceive the
multiplicity as a unity, as a ‘finished thing’. Such multiplicities I
call absolute infinities or inconsistent multiplicities.
In the passage here he is aiming for the ‘finished things’ or consistent things, to be
sets. The multiplicity of all alephs for example, cannot constitute a finished set and
so cannot be assigned a cardinal number. (See [8] for a discussion of this notion of
Zusammensein.)
In latter day jargon we should call such things ‘proper classes’. It does not show
that Cantor no longer thought of the set theoretic universe as an ‘inaugmentable
totality’, which he had called it earlier. (Another description was that it was an
‘absolute maximum’ ([4], pp.410-411)). We cannot be entirely clear what Cantor
had in mind when discussing this universe of sets in this pre-formalised manner,
but it was clearly different from the view Zermelo was to come to have.2
Zermelo’s maturest picture has come down to us from his [27]. The view is
that there are only sets and that these satisfy, let us say here, first order ZFC
(although Zermelo was concerned to promote a second order view and eschewed
the first order fomulations of Skolem and others). For Zermelo the only collections
are sets. For Zermelo when we do set theory our quantifiers range over a domain of
discourse D say. The ‘paradoxes’ show us that the collection D cannot be a member
of itself. Hence we can enlarge this domain to a larger domain of discourse D in
which D is a set. Hence we have a never ending sequence of, in his words, ‘normal
domains’ which are models of second order ZF, (and hence their ordinal heights are
strongly inaccessible cardinals); the sequence of these domains can be indexed by
Cantor’s ordinal numbers. Zermelo talks of ‘creative advancement’ as one proceeds
through these domains; and that we should talk about such a sequence of domains
in some meta-theory. However this meta-theory is never laid out, much beyond the
indication that the normal domains should be in a (1-1) correspondence with the
Cantorian ordinals. (Readers of [27] will see that I am deliberately ignoring one,
inessential for this discussion, aspect of Zermelo’s picture: the presence of a possible
non-empty collection of urelements over which a normal domain may also be built.
Zermelo’s normal domains encompassed models of ZF 2 with or without urelements,
but for this paper it is inessential: we shall consider only normal domains without
urelements. One should note that also under Zermelo’s second order conception
Choice is a logical, as opposed to a set theoretical, axiom. Hence ZF 2 is to be
regarded as including Choice.)
However these different views give us at least two broad-brush pictures of the
universe of sets: a ‘potentialist’ view - Zermelo’s creative advancement, and an
2 Interestingly from the same letter, recalling that ‘equivalent’ meant bijective, one has a
prefiguration of Replacement: “Two equivalent multiplicities are either both “sets” or are both
inconsistent”.
164 P. D. WELCH
‘actualist’ view - that the universe of sets is an absolute maximum and an inaug-
mentable totality. It is possible to discuss these views unlinked to any kind of
position concerning platonism or realism.
3. Reflection
A potentialist view makes it hard to appeal to most kinds of reflection principle.
Zermelo cannot consider the whole universe, and reflect on that, since for him
there is always the potential to make the universe yet larger. All that can be
said here is that there are unboundedly many normal domains (which following
Mirimanoff/von Neumann we should now recognise as Vκ ’s) and so a proper class of
strongly inaccessible cardinals. But even this can not be obtained by an adherent
of the Zermelian school as a result of reflection: the statement “∀α ∈ On ∃β >
α(ZF2 )Vβ ” expresses quantification over all ordinals, and this is just what cannot
be done in this organic view. The statement can however be readily obtained by
allowing second order reflection of the whole (‘actual’) universe (V, ∈).
By allowing domains of all sets and all classes - as formalised by NBG say - we
now are able to second-order quantify over all such classes and formulate reflection
principles that yield Mahlo cardinals, and second order indescribability. The story
of this is familiar enough, that we don’t repeat this here.
The point remains that all such principles only derive cardinals consistent with
V = L - so we may call them intra-constructible.
Gödel stated that he thought all large cardinals could be obtained through
reflection:
The Universe of sets cannot be uniquely characterized (i.e. dis-
tinguished from all its initial segments) by any internal structural
property of the membership relation in it, which is expressible in
any logic of finite or transfinite type, including infinitary logics
of any cardinal number.
(Wang - [24])
Gödel again:
All the principles for setting up the axioms of set theory should
be reducible to a form of Ackermann’s principle: The Absolute
is unknowable. The strength of the principle increases as we get
stronger and stronger systems of set theory. The other principles
are only heuristic principles. Hence the central principle is the
reflection principle, which presumably will be understood better
as our experience increases. (Wang - ibid.)
This seems rather sweeping and our experience with Reflection Principles seems
to go against it. Both on the one hand because Reinhardt specifically noted that
third order reflection with parameters is inconsistent, and on the other, because our
reflection principles have remain stuck in the intra-constructible.
There have been specific attempts to get around this obstacle by restricting the
syntax: Marshall [15] obtains higher order reflection (and large cardinals) by this
method. Tait [21] uses Relativized Cantorian Principles based on certain Existence
Conditions. As motivating conditions these allow him to define certain syntactically
(m)
characterised higher order classes of formulae Γn for m ≥ 2 (the superscript
indicates that higher order universal quantification must be of at most order m).
(m)
Tait shows that for m = 2, Vκ satisfying Γn reflection implies that κ is n-ineffable
(in the definition of Baumgartner), and that measurability of κ sufficed to show
that Vκ satisfied Γ2n -reflection for all n. This left open the question of whether such
principles were extra-constructible. Koellner answered this negatively by showing
that if κ is κ(ω)-Erdős cardinal then for every n Vκ satisfies Γ2n -reflection. He further
answers negatively what was asked, and left open, by Tait: the Γ(m) principles are
all inconsistent for m ≥ 3.
So, even with this syntactical constraint these classes of reflection principles
are either inconsistent or are still intra-constructible.
Koellner finishes his Section 4 with a heuristic argument that any form of reflec-
tion principle which is consistent relative to large cardinals is consistent relative to
κ(ω). If κ is ω-Erdős, then (Vκ , ∈) has an infinite sequence of indiscernibles I ⊆ κ.
Take the Skolem hull H of I in Vκ . Then any order preserving map j0 : I → I in-
duces a non-identity first order elementary map j : M → M , where the ZFC-model
M is the transitive collapse of (the countable) H. We then have a situation sim-
ilar to the (inconsistent with AC) assertion that there is a non-trivial elementary
j : V → V . Koellner argues that from the point of view of consistency proof “it
would appear that whatever reflection is provable from j : V → V should also be
provable from j : M → M . Since reflection would appear to be an entirely internal
matter , this is a reason for thinking that any conceivable reflection principle must
have consistency strength below that of κ(ω).” (My emphasis) Well, is reflection
an entirely internal matter? The view I shall be putting forward here, is that it
is not. It is, or can be widened to be, a Gesamtauffassung that incorporates the
whole, consisting of both the Cantorian sets and absolute infinities. If so, then it
is not internal, and we have a hope for finding extraconstructible principles.
4. The Ackermannian realm and reflection

Another set theory, due to Ackermann was introduced [1] and studied in the
1950’s and 60’s. Ackermann’s set theory A provided for a universe with extension-
ally determined entities (classes) and a predicate V̇ for set-hood: “x ∈ V ”. Besides
axioms for extensionality, a class construction scheme, and set completeness (“all
classes that are subclasses of sets are sets”), it contained the following crucial prin-
ciple:
• (Ackermann’s Main Principle) If X ⊆ V is definable using only set para-

meters, and not using the predicate V̇ , then X ∈ V . Thus if θ does not contain
V:
x ∈ V ∧ ∀t(θ(x, t)−→t ∈ V ) −→ ∃z ∈ V ∀t(t ∈ z↔θ(x, t))
Ackermann interpreted Cantor’s “By a set we understand any collection of

definite distinct objects . . . into a whole” as saying
“we must require from already defined sets that they are determ-
ined and well-differentiated, thus the [foregoing] conditions for
a totality [to be a set] only turn on that it must be sufficiently
166 P. D. WELCH
sharply delimited what belongs to a totality and what does not be-
long to it. However now the concept of set is thoroughly open.”
(Ackermann [1] p.337)3 .
Indeed Reinhardt, whilst working from the premise that Ackermann considered
the concept of set itself as not sharply delimited ([18], p190-1), surmises that the
intuition behind Ackermann’s Main Principle is that a sharply defined collection of
sets is a set, and that, given the set x, the property ‘t is a set such that θ(x, t)’ is
independent of the (extension of) the concept of set, but gives a sufficient condition
for a collection to be sharply delimited. We therefore see that on the other hand a
collection is not sufficiently well-differentiated if it is defined through its relationship
to the concept of set.
The Ackermann quotation continues (in paraphrase) that in the Cantorian
definition it is intended that a collection should be investigated only on a case by
case basis as to whether it represents a set, and it is not meant that it is determined
all at once for all classes whether they are sets or not.
Levy, Vaught [12] added Foundation to A calling this A∗ . Then A∗ is con-
sistent relative to A, and proves the existence of the classes: {V }, {{V }}, . . . ,
P(V ), PP(V ) . . .
Thus the classes over V in A∗ advance for infinitely many types beyond. The
picture is thus qute different from a first-order, or even second-order ZF . However,
Levy considered models of A∗ of the form Vα , ∈, Vβ . The paper [11] showed that
A∗ is L∈˙ - conservative over ZF: A∗ , σ V =⇒ ZF , σ.
Reinhardt in [18] proved the converse implication of this last result also; hence
putting these together the set-theoretical content of A∗ had always just been that
of ZF. It is thus to be noted that two rather different conceptualisations - the
one leading to the ZF formalisation, the other that of Ackermann’s - have the same
content as far as the strictly considered set part is concerned. Reinhardt considered
in [19] (and [20]) ideas that involved having an ‘imaginary realm’ beyond the
Cantorian universe V which he wrote as VΩ . He imagines having classes, say P ,
which are then ‘projected’ into the imaginary realm as jP . The difference between
classes and sets is that the projection of the latter are themselves, whilst that
of one of the former contains more imaginary sets and ordinals. VΩ itself is an
imaginary set in this projected universe. Going yet further, he imagines a typed
hierarchy of ‘Ω-classes’ up to some λ > Ω, and collecting these together as Vλ he will
project Vλ into some virtual realm Vλ . He formulates an extendability principle
E0 (Ω, λ; Ω , λ ):
(i) Ω < jΩ = Ω < λ .
(ii) ∀x ∈ VΩ jx = x;
(iii) j : (Vλ , ∈) →Σω (Vλ , ∈).
Of course this has come down to us distilled in set-sized form as heading towards
the definition of α-extendible cardinals. In all of these theories, there is formed the
idea of some ‘realm’, ‘universe’ etc. beyond V .
We mention these Reinhardtian views of upwards projection of (V, ∈) by way
of a contrast to the Global Reflection Principle to come.
3 “. . . wir von den schon definierten Mengen verlangen müssen, dass sie bestimmt und wohlun-
terschieden sind, so kann es sich bei der obigen Bedingung für eine Gesamtheit nur darum handeln,
dass genügend scharf abgegrenzt sein muss, was zu der Gesamtheit gehört und was nicht zu ihr
gehört. Nun ist aber der Mengenbegriff durchaus offen.”
5. Global Reflection
If we are contemplating an ab initio conception of the universe of sets, then we
may proceed as follows.
By “conception of the universe of sets” we mean here something like the notion
of “concept of set structure” in one of Martin’s versions of concept of set [16]. He
writes that for him the modern, iterative concept has four important components:
(1) the concept of extensionality
(2) concept of ‘set of x’s’
(3) concept of transfinite iteration
(4) concept of absolute infinity.
He is thinking of the concept of sets as the concept of ‘structuralist’s structure’
and thus does not have to add anything as to what kind of things sets are. We
adopt this view here. (Martin remains silent as to which flavour of structuralism’s
structure might be at play here.) A ‘set structure’ is then what is obtained by
iterating the concept ‘set of x’s’ absolutely infinitely many times.
Still again at some pre-formal stage, he then takes some Informal Axioms encap-
sulating set theoretical principles (Extensionality, Comprehension) and rehearses
the categoricity arguments going back to Zermelo, that for any two V1 = (V1 , ∈1 ),
V2 = (V2 , ∈2 ) obtained by iterating the models’ Vα function throughout all the
absolute infinity of ordinals, we have an isomorphism π : (V1 , ∈1 ) → (V2 , ∈2 ). In
short, whatever view we take of what exact set formation process takes place when
we take the “set of x’s”, we end up with isomorphic universes. We, as set the-
orists, thus shall pay no more attention to the mysteries of what exactly “the set
of x’s” is or what precisely ∈ “is”, than we do every day, and shall simply refer
to the set theoretic universe as (V, ∈). But we do further remark for later that
π OnV1 : OnV1 ∼ = OnV2 where OnVi is the absolute infinity of von Neumann
ordinals in the structure Vi .
We then proceed in his paper as follows. We consider the universe of sets,
V , (as above, unique up to this informal isomorphism argument) as the universe
of the domain of purely mathematical discourse: whatever mathematical objects
the mathematician needs, there are (isomorphic copies of) such in V . Indeed we
regard sets themselves as mathematical objects. As we know, of necessity there are
entities outside of V , where the modality of ‘necessity’ here is ‘logical necessity’:
logic requires that the Russell class, or the class of ordinals, or indeed V itself is
not a set. We swallow the Cantorian pill that there are two kinds of entities: the
mathematical-discourse sets, and the absolute infinities.
However we depart from von Neumann, who seemingly treated both kinds of
entities in an equally ‘mathematical’ spirit (see [22], [23]) when developing his
functions-as-classes theory: his classes were subject to mathematical laws. We
draw a firmer conceptual line, and do not treat the absolute infinities in such a
mathematico-functional manner.
In a paper with Leon Horsten ([25])4 we have recently discussed the possible
interpretations of classes prior to a development of a formal theory of them. We
rule out a theory of classes as plurals: a plural, which in any case is a linguistic
construct, is not supposed to add anything more ontologically to the objects we
4 Readers of which will note how much of the discussion here is indebted to it. I want to
express my thanks to him for his initial suggestion for that paper and our many fruitful, and at
times very entertaining, discussions on it.
168 P. D. WELCH
have, namely sets. However we do have more, V is not just “some sets so that
x = x”. We accept that classes are entities that enter into structural arguments,
which do not have to have any prima facie description as a plurality. However
a mereological description of absolute infinities, as being parts of V , the absolute
infinity of all sets, allows us to give sufficient substance to these entities without
tying them to any language, or syntax. We may take over a theory of mereology,
such as Lewis [13], [14] and apply this to V together with its parts. (We have to
make some adjustments: Lewis is dubious about ∅ and the x {x} map; but we
shall ignore these and treat our theory of parts as one which identifies sets x also
as ‘small’ parts.) Thus we take sets and the set elementhood relation as a given:
we are not trying to alter our conception of sets. Lewis thinks that the parthood
relation goes someway to help us understand the set-elementhood relation, but we
are not committing ourselves to what exactly this latter relation is.
We may thus think of second order quantifiers, if later we come to formalise our
notions, as ranging over the parts of V . It should be not associating the parts of V
with any particular linguistic structure: whereas pluralities could be interpreted in
some minimal NBG model sitting above (V, ∈), a mereological view can sit happily
with a Kelley-Morse formal theory of classes, but on its own is not restricting the
absolute infinities that there are. How could it?
One should also note that there is no hierarchy of “super-parts” of collections
or “collections of parts” or any such that threatens to build a ramified hierarchy of
classes beyond On: the power set operation that collects together all the subsets of
a set into a set is a mathematical operation applied to sets. Our acceptance of a
power set operation does not require us to countenance a “power-absolute-infinity”
operation. To insist that we must consider such an operation, if we posit it for sets,
is similar to insisting that if our (physical, space-time) universe is finite then “there
must be something beyond it”.
We denote by C the collection of parts of V . We identify parts of V that are
parts of sets, as themselves sets. The other parts of V are the absolute infinit-
ies. Then (V, ∈, C) is the realm of ‘Cantorian discourse’. Admittedly C inherits
the ineffability of the notion of absolute infinity. Initially then C would have been
populated with examples of absolute infinities that we are familiar with and had
been defined by the early researchers. But we do not insist on restricting to this
definability. (We cash in the remark we made earlier that two possible notions of
“set of x’s ” led to isomorphic universes V1 = (V1 , ∈1 ), V2 = (V2 , ∈2 ) with an
isomorphism π between them, which in particular restricted to an isomorphic map
between the absolute infinity of V1 ’s ordinals with those of V2 . The same argument
shows that ‘parts of V1 ’ carry over to parts of V2 . So clearly we may extend the
isomorphism to π : (V1 , ∈1 , C1 ) → (V2 , ∈2 , C2 ). In other words two differing notions
of “set of” cannot lead to essentially differing models when their parts are also
included. Moreover we view the content of our set, and class, theoretical ideas to
be captured by this isomorphism type.)
We want to invert the Reinhardtian approach and stand it on its head: instead
of projecting into some ‘virtual realm’, we reflect the structure (V, ∈, C) to some
set-sized part of itself. The approach taken is that we regard (V, ∈, C) as absolutely
indistinguishable from one of its initial segments.
Definition 1. (GRP) The Global Reflection Property holds if the universe

˙ -elementary substructure, (V , ∈, C ) ≺ (V, ∈, C)
(V, ∈, C) admits of a set-sized L +
∈
with V = Vκ for some κ and which is isomorphic to a structure of exactly the same
kind, namely (Vκ , ∈, Vκ+1 ).
We thus intend the principle as saying that there is a set-sized version of
(V, ∈, C) (that is V together with its classes) which is of exactly the same kind,
namely is some (Vκ , ∈, Vκ+1 ) (so Vκ together with its classes) and with some (re-
stricted) elementarity reflected between them. Here L + ∈˙ denotes the usual first order
language L∈˙ , but augmented with second order free variable symbols Ȧ, Ḃ . . .. The
interpretation of the second order variables to range over the collections C, C of
parts of V . The second order variables are thus not quantified over. (We shall
write, for example, ‘Σ0n ’ for formulae at that level of complexity in L + ˙ , to indicate
∈
n-alternating set quantifiers.)
It is also possible to see this indistinguishability as the endpoint of a spectrum
of more limited reflection properties which we outline first. Just for convenience we
phrase these as embedding properties more familiar to set theorists.
Definition 2. (Limited Global Reflection) There is κ ∈ On, there is j = id,
with crit(j) = κ, and a non-empty D ⊆ Vκ+1 so that:
j : (Vκ , ∈, D) −→Σ0ω (V, ∈, C).
Here we stipulate that −→Σ0ω denotes an embedding that preserves truth of
L+
∈
˙ formulae. As crit(j) = κ we have: (i) jVκ = id Vκ ; (ii) if κ ∈ D then
j(κ) = On ∈ C. We thus have:
ϕ(x, X)(Vκ ,∈,D) ↔ ϕ(x, j(X))(V,∈,C)
The strength of such a principle depends entirely on the nature of D, which is
the non-trivial part of the domain of j. If D is extremely thin, then the property
is saying hardly anything at all. If we had allowed D = ∅ then all we should have
had is that (Vκ , ∈) ≺Σ01 (V, ∈).
• If D ⊂ P (κ)L then in general we shall not have a reflection property that is
extra-constructible. Indeed, several ‘classical’ reflection properties can be expressed
in this way.
• However if D ⊇ P (κ)L then we can define in the usual way an L-ultrafilter
on κ:
X ∈ U ↔ (X ∈ L ∧ κ ∈ j(X)) (1)
By standard arguments this is an amenable normal ultrafilter, and we may define
a wellfounded ultrapower Ult((L, ∈), U ) which is isomorphic to L itself. In other
words, we have some non-identity embedding j0 : L−→L, i.e. 0 exists.
• As D is reckoned to be larger and larger, then the strength of the principle
increases: if some other definable inner model M has D ⊇ P (κ)M then again we
shall be able to define an ultrapower of M : Ult((M, ∈), U ) if U is defined in the
same way. Such a model may then also be seen to be non-rigid.
The logical limit, and principle of main interest here, is when D becomes max-
imal at the end of this spectrum, i.e. to become Vκ+1 . Unlike the pitfall that was
waiting for Reinhardt, this principle when extended to the limit can be shown
consistent relative to large cardinals.
We then have the following GRP paraphrase:
170 P. D. WELCH
There is κ ∈ On, there is j = id, with crit(j) = κ, so that:

j : (Vκ , ∈, Vκ+1 ) −→Σ0ω (V, ∈, C).
Some points are then clear: κ is strongly inaccessible; as there is a wellordering
W of Vκ in Vκ+1 , then j(W ) is a wellordering of V . Thus we must have global
choice holding in V .
• GRP is equivalent to the principle obtained by weakening Σ0ω by Σ01 (but
not by the usual self-strengthening argument of Gaifman, because that requires the
range of the map j to be cofinal - which does not apply here).
As soon as we have D = Vκ+1 , we have that the ultrafilter U defined at (1)
by removing the conjunct “X ∈ L” is a normal measure in V . Hence κ is meas-
urable, and by the supposed elementarity, and by simple reflection arguments, we
immediately have a proper class of measurable cardinals. But we easily have more.
Theorem 1. GRP implies there is a proper class of Shelah cardinals.

Proof: Recall that μ is Shelah if ∀f ∈ μ μ∃N, j j : V → N ∧ Vj(f )(μ) ⊆N . We
show that κ is Shelah in the statement of GRP and this easily implies by element-
arity that there is a proper class of such.
Let f ∈ κ κ ⊆ Vκ+1 , be arbitrary. Then j(f ) : On −→ On and j(f )“κ ⊆ κ.
Take λ > κ a sufficiently large inaccessible, so that j(f )(κ) < λ, and consider the
“λ-strong” extender derived from j:
For a ∈ [λ]<ω : Ea =df {z ∈ P([κ]|a| ) : a ∈ j(z)}; E = Ea : a ∈ [λ]<ω .
This has the following properties:
(1) E is a (κ, λ)-extender such that Ult((V, ∈), E) is wellfounded; with kE : V →

Ult((V, ∈), E), and if l : Ult((V, ∈), E) ∼
= N , is the unique transitivisation collapse
map, then setting jE = l ◦ kE , jE : V −→ N and , j(f )(κ) = jE (f )(κ) < λ, and
Vλ = (Vλ )N .
As jE (f )(κ) = kE (f )(κ) < λ, then VjE (f )(κ) ⊆N follows; hence we have the She-
lah property for this f .
Clearly stronger properties can be argued for, but in any case we have U W :
Corollary 1. GRP implies there is a proper class of measurable Woodin
cardinals.
6. Is GRP a reflection principle?

Instead of viewing GRP as a limiting principle as the class domain grows larger
until it contains all of P (κ), one could view it outright as asserting in a strong form
that (V, ∈, C), that is V together with all its parts, is indistinguishable from one of its
initial segments and its parts: (Vκ , ∈, Vκ+1 ). We view V together with its parts, and
its initial segments and their parts as being so rich that we may form a substructure
(V , ∈, C ) with V = Vκ for some κ such that (V, ∈, C) and (Vκ , ∈, Vκ+1 ) can stand
in this relationship. The latter is a simulacrum of the former. The inverse of the
isomorphism j mediates that indistinguishability through being a truth preserving
elementary embedding.
If this viewpoint is viable, then we are taking the whole of (V, ∈, C) and re-
flecting this to a (Vκ , ∈, Vκ+1 ). It is not a syntactic, formula-by-formula reflection,
whether first or second order, or something expressed in some logic. In these senses
it is not a reflection principle such as Gödel may have had in mind. It is not the
viewpoint that says “nothing we can say in whatever logic/language pins down
(V, ∈, C), or is only true in (V, ∈, C)” (such a viewpoint would be too weak for our
purposes). To assert GRP is to assert that there is a substructure of the given kind,
or that there exists an isomorphism j doing the work of linking these collections of
parts together, those of Vκ and those of V . One cannot argue for GRP using the
iterative conception of set alone, that is for a concept of ‘set-structure’ (with the
components (1)-(4) listed at the beginning of the last section), but must be coupled
with the reflective idea involving C.
Of course the assertion of the existence of the substructure (Vκ , ∈, C ) (or equi-
valently the interceding map j), is a second order existential assertion, and by
elementary arguments, it cannot be a definable class of (V, ∈, C). So when we come
to formalise our principle GRP this will require the admission that impredicative
objects such as C or j should be part of the discussion. There is no sense that the j
of GRP has ‘come from somewhere’ or is ‘canonical’ in any way (it certainly cannot
be definable). Friedman and Honzik ([6], Sect. 2) raise this non-canonicity as
somehow a defect of GRP: “However, in our opinion, such strong forms of reflec-
tion seem to be too “uncanonical” to count as true formalization of (Reflection)”.
However their paper is concerned with something quite different from reflection
of an ‘actualised’ universe (V, ∈); it is not a potentialist version either. It seeks to
gain some insight into possible new axioms or hypotheses (such as the “Inner Model
Hypothesis ”) that may then be offered for consideration as true of V , by looking
at countable transitive models, pretending that V is one of them, and thereby pos-
sibly “sharp-generating” this c.t.m. by iteration of an external premouse. Hence
our viewpoint and this procedure are quite different (and at odds with each other).
Perhaps the externally provided iteration maps π̃ when restricted to the c.t.m. V
are deemed more ‘canonical’ ? (Although, if so, one may counter this canonicity, by
remarking that the externally posited ‘premouse’, and so its maps, are not unique
either.) However this is a rather sophisticated approach that applies to countable
transitive models ‘V ’, and does not really touch (in this author’s view) a Cantorian
view of V as an inaugmentable totality of sets together with a pre-formal conception
of reflection.
Indeed Peter Koellner has suggested that perhaps a “resemblance” property is
a better name. Whatever one thinks of the nomenclature, GRP is different from
other properties that are sometimes termed resemblance. We consider some such
forms now for a differential comparison; these are forms which assert that there is
some reflection or accumulation point κ so that anything that occurs above κ in
some sense has an occurrence below. One such is the Vopenka principle:
Definition 3. (Vopenka’s Principle) If Mα | α ∈ On is a proper class of
first order structures each in V of the same signature, then there is α < β and an
elementary embedding j so that j : Mα −→Mβ .
Notice that this is a richness principle of V that claims of any such class of
structures that we have such a triple α, β, j. We can tie it closer to the structure
of V by taking the Mα of the form (Vf (α) , ∈, {α}, Rα ) where α < f (α) for some
increasing function f : On −→ On; and where Rα ⊆ Vf (α) . The presence of {α}
172 P. D. WELCH
makes all the difference as then if we have α, β, j as in Vopenka’s Principle, then

we must have j(α) = β, and thus j is not simple identity. Hence VP is a strong
property which implies that the class of extendible cardinals is Mahlo in On (see
[17]).
A more extensive study of VP like principles where the target structures have
certain elementarity properties in V is given in Joan Bagaria’s paper [2]:
Definition 4. (i) C (n) =df {α | (Vα , ∈) ≺Σn (V, ∈)};
(ii) κ is a C (n) -measurable cardinal if it is the critical point of an elementary
embedding j : V −→M , with M transitive, and j(κ) ∈ C (n) ;
(iii) κ is a C (n) -extendible cardinal if for all λ there exists μ,j with crit(j) = κ;
j : Vλ −→Σω Vμ with j(k) ∈ C (n) .
As [2] analyzes, if κ is measurable then it is C (n) -measurable, and the prefix
in this sense adds nothing; whilst, C (n) -extendibility is a genuine stengthening of
extendibility.
One final definition before we can state Bagaria’s categorization. This is now a
form of resemblance where κ is some kind of ‘reflection point’ or ‘cut point’ in the
universe V :
Definition 5. VP(κ, Σn ) holds iff for every proper class C of structures of
the same type τ such that both τ and the parameters of some Σn -definition of C,
if any, belong to Hκ , then C reflects below κ, i.e.,
∀B ∈ C∃A ∈ C ∩ Hκ (A is elementarily embeddable into B).

Notice this is not so dissimilar structurally to the GRP : to assert the existence
of an elementary embeddable A as above, of course is also to assert the existence of
an elementary substructure of B, namely the range of that elementary embedding,
which is isomorphic to a structure of the same kind: the original A in C. GRP also
asserts the existence of a substructure isomorphic to a small (equals set-sized) one
of the same kind.
Theorem 2. (Bagaria [2], 4.15) The following are equivalent:
(i) VP;
(ii) For all n there is a proper class of κ so that VP(κ, Σn );
(iii) ∀n∃κ(κ is C (n) -extendible).
We thus see that VP can be given an equivalence in terms of a proper class of
reflection or cut points for any kind of definable class whatsoever.
This in fact brings out parallels with a much earlier paper of Magidor’s on
supercompacts:
Theorem 3. (Magidor [17] ) Let κ be the first supercompact cardinal; then κ
is the least cardinal so that VP(κ,Σ2 ).
We have mentioned the details of these definitions to see that the kind of
reflection they represent is some form of internal reflection of the ramified layers
Vα rather than the idea of reflection of the whole of universe (V, ∈) which cannot be
pinned down in ways mentioned by Ackermann and Gödel. So there is a qualitative
difference between GRP and such principles.
However it is easily noted that:
Theorem 4. Con (ZFC +∃κ(κ is 1-extendible))−→ Con(NBG + GRP).

But the arrow is not reversible. GRP thus falls just sort of implying the exist-
ence of those embeddings j that are discontinuous at the successor of the critical
point: j”κ+ is bounded in j(κ+ ). Hence GRP is consistent, or can be made con-
sistent, with Global Square, and λ everywhere, by a class forcing. (For these
methods see [3].) It is thus a reflection principle that marks off this threshold.
7. Strengthening GRP?
Whilst the last theorem indicates what the strength of the basic GRP is, the
motivation for top-down reflection rather than upwards projection came originally
from a weakening of the notion of subcompactness:
Definition 6. κ is subcompact if for any A ⊆ Hκ+ , there are j, μ < κ, with
crit(j) = μ, j(μ) = κ and a B ⊆ Hμ+ with
j : (Hμ+ , B) −→e (Hκ+ , A).
To strengthen GRP we may ask for many j’s and κ’s. Or else, more interest-
ingly, we may increase the elementarity of j to be (partially or fully) second order
reflecting, that is, to preserve for example Σ1n -formulae with now quantification
over the second-order variables of L+ ∈˙ - such an extended language we shall call L∈
2
˙.
Then if j : (Vκ , ∈, Vκ+1 ) −→Σ1ω (V, ∈, C) we shall conclude that all instances of im-
predicative comprehension - which are true in (Vκ , ∈, Vκ+1 ), being a Kelley-Morse
(KM) model, will also hold in (V, ∈, C).
If j is Σ1ω -elementary more can be said about the range of j: for example Σ11 -
elementarity shows that the class of Shelah cardinals is stationary in (V, ∈, C). We
may go further and fomalise second order satisfaction, as follows.
We may for any n ∈ ω define a Σ1n formula Satn (v0 , . . . , v4 , Y0 ) so that, prov-
ably in NBG +Global Choice (the latter holds if GRP does, and is needed to define
Skolem functions):
∀h ∈ ω V ∀X1 , · · · ∀Xm
→
−
[Satn (ϕ, k, m, h0 , . . . , hk−1 , X1 , . . . , Xm )↔ϕ( h,X1 , . . . , Xm )],
for any Σ1n formula ϕ(v1 , . . . , vk , Y1 , . . . , Ym ) with the vi to be interpreted as sets,
and the Yi as classes. Let Sat be the amalgamation of these Satn predicates.
Definition 7. (GRP+ ) :There is κ ∈ On, there is j = id, crit(j) = κ,
j : (Vκ , ∈, Vκ+1 , SatVκ ) −→Σ11 (V, ∈, C, Sat).
We thus require j to be Σ11 -elementary in the full second order language L2∈,
˙ Ṡ
with a predicate Ṡ for Sat. It is easy to argue that subcompactness of some λ
then yields a model of GRP+ . Just to demonstrate the additional strength of this
principle over GRP we show the following.
Proposition 1. Assume GRP+ . Then there is a commuting system
μα , jαβ α≤β∈On
of embeddings jαβ : (Vμα , ∈, Vμα +1 , ) −→Σ01 (Vμβ , ∈, Vμβ +1 ), with each jαβ , α < β,
witnessing the simple GRP at μβ . Thus each jαβ μα = id μα and jαβ (μα ) = μβ .
Moreover for α ∈ On, there are maps jα : (Vμα , ∈, Vμα +1 ) −→Σ01 (V, ∈, C) also
witnessing GRP in the universe.
174 P. D. WELCH
This is obtained in a way very similar to the following from a subcompact.
Proposition 2. Let κ be subcompact. Then there is a commuting system

μα , jαβ α≤β≤κ of embeddings jαβ : (Vμα +1 , ∈) −→e (Vμβ +1 , ∈) with, μα < μβ for
α < β; μκ = κ; each jαβ μα = id μα , and jαβ (μα ) = μβ , and thus with each jαβ ,
α < β, witnessing the 1-extendibility of μα .
Proof: For λ ∈ Card let Satλ be the satisfaction relation for (Hλ , ∈). Then we
view Satλ as a subset of Hλ . Let κ be subcompact as above, and let A = Sat =
Satκ+ Then Sat ⊆Hκ+ , and applying the definition of subcompactness there are
μ < κ and j, and Sat with
j : (Hμ+ , Sat) −→e (Hκ+ , Sat).
(1) Sat = Satμ+ .

Pf: Suppose Hμ+ |= ϕ(x)↔¬Sat(ϕ, x). Apply j to get a contradiction.
(1)
Definition 8. (Hκ+ , Sat) |=“ran(k) ≺e (V, ∈)”⇐⇒df
∀x ∈ ran(k)∀ϕ ≡ ∃zψ(z, v1 )
∃z Sat(ψ, z, k(x)) −→ ∃z ∈ ran(k) Sat(ψ, z, k(x)).
We thus use the standard Tarski-Vaught criterion on elementary substructure

hood to formalise the notion of being an elementary submodel of V . Note that
j ⊆ Hμ+ × Hκ+ and |j| = |Hμ+ | < κ and so j ∈ Hκ+ . Clearly, by the elementarity
of j:
(2) (Hκ+ , Sat) |=“ran(j) ≺e (V, ∈)” indeed:
(Hκ+ , Sat) |=“There are k, κ0 , with k : (Hκ+ , ∈) −→ (V, ∈) and

0
ran(k) ≺e (V, ∈), and thus k is an elementary map.”
But by invoking j we have:
(3) (Hμ+ , Satμ+ ) |=“There are k, κ0 , with k : (Hκ+ , ∈) −→ (V, ∈) and

0
ran(k) ≺e (V, ∈), and thus k is an elementary map.”
This gives us two links in a chain of models we are looking for in the Proposition.
Suppose there are no chains of length κ of the kind sought. Let C = μα , jαβ α≤β≤τ
be a maximal such commuting chain with the properties: (i) there is a final model
(Vμτ +1 , ∈), and (ii) with a final map jτ : (Vμτ +1 , ∈) −→ (Vκ+1 , ∈). Suppose τ < κ
(for otherwise we are done). For each α < τ we have that (Hκ+ , Sat) |=“ran(jτ ◦
jατ ) ≺e (V, ∈)” By elementarity of j we have in fact that, C is similarly a 1-
extendible chain, but now that there is some jτ : (Vμτ +1 , ∈) −→ (Vμ+1 , ∈), i.e.,
with the chain also maximal such, but with target (Vμ+1 , ∈). However this is a
contradiction since (Hκ+ , Sat) nows sees that C can be extended one more link
(namely via jτ ), and this reflects into Hμ+ , so C is not a maximal chain going up
to (Vμ+1 , ∈).
8. Conclusion
The strengthenings of GRP considered in the last section go beyond the kind
of position outlined earlier: the purely mathematical objects reside in V , it is the
parts of V that form the proper classes of C. In GRP we reflect, in as much that
there should be an initial segment Vκ together with its parts, that is a simulacrum
of (V, ∈, C), that is, it is isomorphic to a substructure of (V, ∈, C). We classify Σ0n
statements as mathematical; the second order expressions quantifying over parts
are mereological: these are about the parts of V . However if the substructure of
(V, ∈, C) is supposed to be mereologically elementary, that is, if the elementarity
is such that it is required to reflect structural statements or further commitments
about the parts of V beyond those expressible in L+ ˙ (or in other words, that the
∈
embedding j is more than first order respecting) then this can be construed as a
step beyond the pure Cantorian picture we have argued for.
References
[1] W. Ackermann, Zur Axiomatik der Mengenlehre (German), Math. Ann. 131 (1956), 336–345.
MR0078936
[2] J. Bagaria, C (n) -cardinals, Arch. Math. Logic 51 (2012), no. 3-4, 213–240, DOI
10.1007/s00153-011-0261-8. MR2899689
[3] A. D. Brooke-Taylor and S.-D. Friedman, Subcompact cardinals, squares, and stationary
reflection, Israel J. Math. 197 (2013), no. 1, 453–473, DOI 10.1007/s11856-013-0007-x.
MR3096624
[4] G. Cantor, Gesammelte Abhandlungen mathematischen und philosophischen Inhalts. Mit
erläuternden Anmerkungen sowie mit Ergänzungen aus dem Briefwechsel Cantor-Dedekind
(German), Herausgegeben von Ernst Zermelo. Nebst einem Lebenslauf Cantors von Adolf
Fraenkel, Georg Olms Verlagsbuchhandlung, Hildesheim, 1962. MR0148517
[5] G. Cantor. Letter to Dedekind. In J. van Heijenhoort, editor, From Frege to Gödel, pages
113–117. Harvard University Press, 1967.
[6] S-D. Friedman and R. Honzik. On strong forms of reflection in set theory. Mathematical Logic
Quarterly.
[7] K. Gödel. What is Cantor’s continuum problem? Oxford University Press, 1990.
[8] K. Hauser. Cantor’s Absolute in metaphysics and mathematics. International Philosophical
Quarterly, 53(2):161–188, 2013.
[9] A. Kanamori, The higher infinite, Perspectives in Mathematical Logic, Springer-Verlag, Ber-
lin, 1994. Large cardinals in set theory from their beginnings. MR1321144
[10] P. Koellner, On reflection principles, Ann. Pure Appl. Logic 157 (2009), no. 2-3, 206–219,
DOI 10.1016/j.apal.2008.09.007. MR2499709
[11] A. Lévy, On Ackermann’s set theory, J. Symb. Logic 24 (1959), 154–166. MR0126371
[12] A. Lévy and R. Vaught, Principles of partial reflection in the set theories of Zermelo and
Ackermann, Pacific J. Math. 11 (1961), 1045–1062. MR0139519
[13] D. Lewis. Parts of Classes. Basil Blackwell, 1986.
[14] D. Lewis, Mathematics is megethology, Philos. Math. (3) 1 (1993), no. 1, 3–23, DOI
10.1093/philmat/1.1.3. MR1210471
[15] M. V. Marshall R., Higher order reflection principles, J. Symbolic Logic 54 (1989), no. 2,
474–489, DOI 10.2307/2274862. MR997881
[16] D. A. Martin. Completeness or Incompleteness of Basic Mathematical Concepts. EFI Harvard
Workshop Papers, 2012.
[17] M. Magidor, On the role of supercompact and extendible cardinals in logic, Israel J. Math.
10 (1971), 147–157. MR0295904
[18] W. N. Reinhardt, Ackermann’s set theory equals ZF, Ann. Math. Logic 2 (1970), no. 2,
189–249. MR0274277
176 P. D. WELCH
[19] W. N. Reinhardt, Remarks on reflection principles, large cardinals, and elementary em-
beddings, Axiomatic set theory (Proc. Sympos. Pure Math., Vol. XIII, Part II, Univ. Cali-
fornia, Los Angeles, Calif., 1967), Amer. Math. Soc., Providence, R. I., 1974, pp. 189–205.
MR0401475
[20] W. N. Reinhardt, Set existence principles of Shoenfield, Ackermann, and Powell, Fund. Math.
84 (1974), no. 1, 5–34. MR0351809
[21] W. W. Tait. Constructing cardinals from below. In W. W. Tait, editor, The Provenance of
Pure Reason: essays in the philosophy of mathematics and its history, pages 133–154. Oxford
University Press, Oxford, 2005.
[22] J. von Neumann, Eine Axiomatisierung der Mengenlehre (German), J. Reine Angew. Math.
154 (1925), 219–240, DOI 10.1515/crll.1925.154.219. MR1581062
[23] J. von Neumann, Die Axiomatisierung der Mengenlehre (German), Math. Z. 27 (1928), no. 1,
669–752, DOI 10.1007/BF01171122. MR1544933
[24] H. Wang, A logical journey. From Gödel to philosophy, Representation and Mind, MIT Press,
Cambridge, MA, 1996. Final editing and with an addition to the preface by Palle Yourgrau
and Leigh Cauman. MR1433803
[25] P.D. Welch and L. Horsten. Reflecting on Absolute Infinity. Journal of Philosophy, to appear.
[26] W. H. Woodin, The continuum hypothesis. I, Notices Amer. Math. Soc. 48 (2001), no. 6,
567–576. MR1834351
[27] E. Zermelo. Über Grenzahlen und Mengenbereiche: Neue Untersuchungen über die Grundla-
gen der Mengenlehre. Fundamenta Mathematicae, 16:29–47, 1930.
School of Mathematics, University of Bristol – and – Isaac Newton Institute for

Mathematical Sciences, Cambridge, United Kingdom
E-mail address: p.welch@bristol.ac.uk
URL: http://www.maths.bristol.ac.uk/~mapdw
Volume 690, 2017
Woodin’s axiom (∗), or Martin’s Maximum, or both?
Ralf Schindler
This paper is dedicated to W. Hugh Woodin on the occasion of his 60th birthday
Abstract. We present and discuss a new axiom, Martin’s Maximum∗,++ , cf.

Definition 2.10, which amalgamates Woodin’s Pmax axiom (∗) and Martin’s
Maximum++ .
If a mathematical object can be

imagined in a reasonable way, then
it exists!
Menachem Magidor
Contents
1. Introduction
2. From MM++ and (∗) to MM∗,++
3. A–iterable mice and A–iterable Pmax conditions
4. (∗)Γ∞ is equivalent to Γ∞ –BMM++,∗
5. Appendix: UBH and universally Baire iteration strategies
References
1. Introduction
Building upon earlier work of J. Steel and R. Van Wesep, cf. [StVW82],
W. Hugh Woodin introduced in [Woo83] an axiom which he called ∗, cf. [Woo83,
p. 189]. He shows that ZFC plus ∗ is consistent relative to1 ZF plus AD and that
∗ implies that δ 12 = ℵ2 and NSω1 is saturated, cf. [Woo83, Theorems 3.2 and 3.3].
The axiom ∗ of [Woo83] is a precursor to the axiom (∗) which W. Hugh Woodin
introduced in [Woo99, Definition 5.1]. Recall that the axiom (∗) of [Woo99] says
that
(1) AD holds in L(R), and
(2) there is some G which is Pmax –generic over L(R) such that P(ω1 ) ⊂
L(R)[G].
2010 Mathematics Subject Classification. Primary 03E57; Secondary 03E45, 03E55, 03E60,
03E65.
Key words and phrases. Pmax forcing, Martin’s Maximum.
1 Here, AD is the axiom of determinacy.
177
2017
178 RALF SCHINDLER
As ∗, the Pmax axiom (∗) also implies that δ 12 = ℵ2 and NSω1 is saturated –
the latter under the additional hypothesis that V = L(R)[G], where G witnesses
(∗), cf. [Woo99, Theorems 4.50 and 4.53]. Besides, “δ 12 = ℵ2 ,” (∗) yields many
other interesting statements whose complexity is Π2 over Hω2 , e.g. φAC , [Woo99,
Corollary 5.7], and ψAC , [Woo99, Lemma 5.18], cf. also [Woo99, Theorems 5.74
and 5.76].
The axiom (∗) may in fact be construed as a maximality principle with respect
to truths which are Π2 over Hω2 . E.g., every sentence of that complexity which holds
true in V already holds true in every Pmax extension of L(R), cf. [Woo99, Theorem
4.64], and (∗) implies that every sentence which is Π2 over Hω2 and which is Ω–
consistent holds true in V , cf. [Woo99, Theorem 10.149].
Another way of spelling out the Π2 maximality feature of (∗) is given by
[AspSch14, Theorem 2.7] which states that (∗) is in fact equivalent to a gener-
alized version of Bounded Martin’s Maximum++ . The main theorem of the present
paper, Theorem 4.2, will be an expansion of [AspSch14, Theorem 2.7].
There is a discussion in [Woo99] of the relationship of (∗) with forcing axioms,
but to this date it still remains a mystery.
Martin’s Maximum, MM (cf.[FoMaSh88]), expresses the idea that V is max-
imal in the sense that if certain Σ1 truths may be forced to hold in stationary
set preserving forcing extensions of V , then these truths already hold in V . Cf.
e.g. [ClaSch12, Theorem 1.3] for a precise formulation. Many consequences of (∗)
which are Π2 over Hω2 have been verified to follow also from MM, cf. [Woo99, The-
orems 3.17, 5.9, and 5.14], [ClaSch09], and [DoeSch09].
Recall that Martin’s Maximum++ , MM++ for short, is the statement that for
every stationary set preserving poset P, for every family {Di : i < ω1 } of dense
subsets of P, and for every collection {τi : i < ω1 } of names for stationary subsets
of ω1 there is a filter G such that G ∩ Di = ∅ for all i < ω1 and τiG = {ξ < ω1 : ∃p ∈
G p ξˇ ∈ τi } is stationary in ω1 for every i < ω1 .
It is fair to say that ZFC plus (∗) and ZFC plus MM++ are the two most promi-
nent axiomatizations of set theory which both negatively decide the continuum
problem. However, the following questions are still wide open, cf. [Woo99, pp.
769ff. and p.924 Question (18) a)] and [Lar08, Question 7.2].
(Q1) Assuming ZFC plus the existence of large cardinals, must there be a (semi–
proper) forcing P such that if G is P–generic over V , then V [G] |= (∗)?
(Q2) Is Martin’s Maximum++ consistent with (∗)?
(Q3) Assume Martin’s Maximum++ . Must (∗) hold true?
The reader may consult [Woo99, Theorem 10.14 and 10.70], [Lar00], [Lar08], and
[SchWoo∞] to find out what is known concerning these questions.
Inspired by our work on Jensen’s L–forcing which led to the papers [ClaSch09],
[DoeSch09], and [AspSch14, Definition 2.6], the present paper proposes a new
axiom which I shall call Martin’s Maximum∗,++ , MM∗,++ , and which amalgamates
(∗) and MM++ . Cf. Definition 2.10 below.2 MM∗,++ may be thought of as resulting
from MM++ in the formulation of [ClaSch12, Theorem 1.3] by replacing “may be
forced to hold in stationary set preserving forcing extensions of V ” by “is honestly
consistent” (cf. Definition 2.8 below), where honest consistency in turn states a
form of Ω–consistency which also guarantees that the structure NSω1 be respected.
2 I first presented MM∗,++ in my talk “An axiom” at the Fields Institute, Toronto, on No-
vember 12, 2012.

WOODIN’S AXIOM (∗), OR MARTIN’S MAXIMUM, OR BOTH? 179
We would like to mention that in recent work, J. Steel takes the alternate route
by propagating a determinacy hypothesis, AD2 , cf. [Lar∞], from which MM++ (c+ )
may (probably) be deduced to hold in Pmax extensions of a determinacy model V ,
compare [Woo99, Theorem 9.44]. The ultimate hope might be to design a global
determinacy hypothesis which gives MM++ in Pmax extensions of a determinacy
model V . We do not know how AD2 relates to MM∗,++ .
We construe this paper as proposing a framework for discussing the above
questions (Q1) – (Q3).
Section 2 is elementary and introduces MM∗,++ . In Section 3, we shall use inner
model theory to obtain Pmax conditions which are A–iterable for A ⊂ R. This will
be used in Section 4 to formulate and prove an equivalence of a strong form of (∗)
with a bounded version of Martin’s Maximum∗,++ . In the Appendix, Section 5,
we will include a proof that the Unique Branch Hypothesis gives universally Baire
iterations strategies for collapses of countable substructures of V .
The new results of this paper are Theorems 3.14 and 4.2. No new techniques
had to be developed to prove these results that were not already made available on
the market by D.A. Martin, J.R. Steel, W.H. Woodin, and others, and to make the
paper more self–contained we allowed ourselves to include the presentation of some
tools which are relevant to our questions and which to a large extent play also a
crucial role in the core model induction, a method first explored by W.H. Woodin,
cf. [SchSt∞].
2. From MM++ and (∗) to MM∗,++

We shall write R for ω ω and refer to it as the set of real numbers. We say that
x ∈ R codes a transitive set iff
Ex = {(n, m) : x(n, m) = 0}
is extensional and well–founded.3 If x codes a transitive set, then we shall write
decode(x) for πx (0), where
∼ (Mx ; ∈)
πx : (ω; Ex ) =
is the transitive collapse of (ω; Ex ). That way, every z ∈ HC is coded by a real in
the sense that there is some x ∈ R coding a transitive set such that z = decode(x).
If x, x ∈ R, then we shall write
x∼ = x
to express that fact that both x and x code transitive sets and decode(x) =
decode(x ).
Let us write C for the set of all reals coding a transitive set. Then C is a Π11
set, and ∼ = is a Σ11 equivalence relation on C.
If f : R → R is a function, then we say that f is universally Baire iff the graph
of f is universally Baire as a subset of R × R, cf. [FeMaWo92], i.e., if there are
trees T and U , both on ω × ω × OR, such that
(a) f = p[T ], and
(b) for all posets P, V P |= p[U ] = R2 \ p[T ].
By absoluteness, for all posets P and for all x ∈ R ∩ V P , there is at most one
y ∈ R ∩ V P with (x, y) ∈ p[T ], i.e., p[T ] ∩ V P is a (possibly partial) function in V P .
3 Here, n, m → n, m denotes the Gödel pairing function from ω × ω onto ω.

180 RALF SCHINDLER
Definition 2.1. Let f : R → R be a universally Baire function. We then say

that f is total and code invariant in all set generic extensions iff the following holds
true. Let the trees T and U witness that f is universally Baire, with f = p[T ].
Then for all posets P,
(a) V P |= ∀x ∈ R ∃y ∈ R (x, y) ∈ p[T ], and
(b) V P |= ∀{x, x , y, y } ⊂ R ((x, y) ∈ p[T ]∧(x , y ) ∈ p[T ]∧x ∼
= x −→ y ∼
= y ).
If f is universally Baire as being witnessed by T , U as well as by T , U , with
f = p[T ] = p[T ], then for every poset P, V P |= p[T ] = p[T ], cf. [FeMaWo92].
Hence the truth value of (a) and (b) in Definition 2.1 is not sensitive to the choice
of T , U , so that being total and code invariant in all set generic extensions is really
a property of the function f .
For the record, let us note the following easy criterion.
Lemma 2.2. Let f : R → R be a function such that there is a tree T on ω × ω ×
OR such that
f = p[T ], and
for all posets P, V P |= ∀x ∈ R ∃y ∈ R (x, y) ∈ p[T ].
f is then universally Baire.
Proof. Let the tree U on ω × ω × (ω × OR) search for (x, z, (y, α )) such that
) ∈ T and z = y. T , U then witness that f is universally Baire.
(x, y, α
Let f be a universally Baire function which is total and code invariant in all
set generic extensions. Let P be a poset, and let g be P–generic over V . We may
then define inside V [g] a natural (total) map
(2.1) f P,g : V [g] → V [g]
as follows. Let the trees T and U witness that f is universally Baire, with f = p[T ].
Let X ∈ V [g], let θ ≥ Card(TC({X}),4 let H be Col(ω, θ)–generic over V [g], and let
x ∈ R ∩ V [g][H] code a transitive set such that X = decode(x). By (a) of Definition
2.1 applied to V [g][H], there is then some y ∈ R ∩ V [g][H] with (x, y) ∈ p[T ], and
by (b) of Definition 2.1 applied with x = x and y = y, y codes a transitive set.
Set
f P,g (X) = decode(y).
If H is Col(ω, θ)–generic over V [g][H], if x ∈ R ∩ V [g][H ] codes a transitive set

such that X = decode(x) and if y ∈ R ∩ V [g][H ] is such that (x , y ) ∈ p[T ],

then by (b) of Definition 2.1 applied to V [g][H][H ], y ∼ = y , hence decode(y ) =
P,g
decode(y). Therefore f (X) is sensitive neither to the choice of H nor of x, so
that f P,g (X) ∈ V [g] and the function f P,g is in fact well–defined inside V [g].
Let f : R → R be a total function such that the graph of f is an analytic subset
of R × R and in V ,
∼ x −→ y =
∀{x, x , y, y } ⊂ R (y = f (x) ∧ y = f (x ) ∧ x = ∼ y ).
Then f is universally Baire, and if T , U witness this with f = p[T ], then both (a)
and (b) of Definition 2.1 hold true for any poset P, as those statements may be
rephrased in a Π12 fashion.
More complex examples may be given in the presence of large cardinals, cf.
Theorem 2.5, Corollary 2.6, and Section 3.
4 TC({X}) is the transitive closure of {X}.
Definition 2.3. Let F : HC → HC be a function. We say that F is universally

Baire in the codes iff there is a universally Baire function f : R → R such that if
z ∈ HC and x ∈ R codes a transitive set with z = decode(x), then f (x) codes a
transitive set with F (z) = decode(f (x)).
We also say that F is strongly universally Baire in the codes iff F is universally
Baire in the codes as being witnessed by a universally Baire function f : R → R
which is total and code invariant in all generic extensions.
Let F : HC → HC be strongly universally Baire in the codes as being witnessed
by f . Let P be a poset, and let g be P–generic over V . We may then define inside
V [g] a natural (total) map
(2.2) F P,g : V [g] → V [g]
by setting F P,g (X) = f P,g (X). We claim that F P,g is in fact well–defined in that
the definition given is not sensitive to the choice of f . To see this, let f : R → R
and h : R → R be universally Baire functions which are total and code invariant
in all geeric extensions and which both witness that F : HC → HC is universally
Baire in the codes. Let T , U witness that f is universally Baire, with f = p[T ],
and let T , U witness that h is universally Baire, with h = p[T ]. Suppose that P
is a poset, g is P–generic over V , and f P,g = hP,g , say f P,g (X) = hP,g (X), where
X ∈ V [g]. Let θ ≥ Card(TC({X}), let H be Col(ω, θ)–generic over V [g], and let
x ∈ R ∩ V [g][H] code a transitive set such that X = decode(x). There are then
reals y, y such that (x, y) ∈ p[T ], (x, y ) ∈ p[T ], and
decode(y) = f P,g (X) = hP,g (X) = decode(y ),
so that y y . As C and = ∼ are both universally Baire, cf. p. 179, we may in
V construct a tree S searching for reals x̄, ȳ, ȳ ∈ C such that (x̄, ȳ) ∈ p[T ],
(x̄, ȳ ) ∈ p[T ], and ȳ ȳ . As S is ill–founded in V [g][H], S has to be ill–founded
in V by absoluteness, and if x̄, ȳ, ȳ ∈ C ∩ V are given by an infinite branch
through S, then f (x̄) = ȳ ȳ = h(x̄), so that if X̄ = decode(x̄), Ȳ = decode(ȳ),
and Ȳ = decode(ȳ ), then F (X) = Y = Y = F (X) by the choice of f and h.
Contradiction!
It is worth pointing out that of course
(2.3) F P,g ⊃ F.
If A ⊂ R is universally Baire, if P is a poset, and if g is P–generic over V , then
we will follow [FeMaWo92] and denote by Ag the set p[T ] ∩ V [g], where T is such
that there is some U such that T , U witnesses that A is universally Baire, with
A = p[T ]. Ag is the “new version” of A in V [g] and is not sensitive to the choice of
T, U.
The following is a crude corollary to seminal theorems by D.A. Martin, J.R.
Steel, and W.H. Woodin.
Theorem 2.4. Assume that there is a proper class of Woodin cardinals. Let
A ⊂ R be universally Baire, let ϕ(A) be any statement which is projective in A, let
P be any poset, and let g be P–generic over V . Then
V |= ϕ(A) ⇐⇒ V [g] |= ϕ(Ag ).
Proof. Let us assume throughout this proof that there is a proper class of
Woodin cardinals.
182 RALF SCHINDLER
The reader may find definitions of the concepts of “homogeneously Souslin” and
“weakly homogeneously Souslin” e.g. in [Sch14, p. 322f.].5 By a theorem of W.H.
Woodin, if A ⊂ R is universally Baire, then A is weakly homogeneously Souslin.
For a proof, cf. e.g. [St07, Theorem 1.2]. By a theorem of D.A. Martin and J.R.
Steel, if A ⊂ R2 is weakly homogeneously Souslin, then R \ pA is homogeneously
Souslin, where pA = {x ∈ R : ∃y ∈ R (x, y) ∈ A} is the projection of A. For a proof,
cf. [MaSt89, Theorem 5.11]. If A ⊂ R is weakly homogeneously Souslin, then A is
universally Baire, cf. e.g. [Sch14, Problem 13.4, p. 323].
These results imply that A ⊂ R is universally Baire iff A is (weakly) homoge-
neously Souslin, and an inspection of the proof of [Sch14, Problem 13.4, p. 323],
say, then gives the conclusion of the theorem.
Corollary 2.5. Assume that there is a proper class of Woodin cardinals. Let
f : R → R be universally Baire such that in V ,
∀{x, x , y, y } ⊂ R (y = f (x) ∧ y = f (x ) ∧ x ∼
= x −→ y ∼
= y ).
Then f is total and code invariant in all generic extensions.
Proof. If f is as in the statement of this corollary, then by Theorem 2.4 both
(a) and (b) of Definition 2.1 hold true for any poset P, as those statements may be
rephrased in a Π12 (f ) fashion.
Corollary 2.6. Assume that there is a proper class of Woodin cardinals. If
F : HC → HC is universally Baire in the codes, then F is strongly universally Baire
in the codes. Hence if P is a poset and if g is P–generic over V , then the function
F P,g : V [g] → V [g] as in ( 2.2) is well–defined.
We will discuss examples of functions F : HC → HC which are universally Baire
in the codes and which arise from inner model theory in Section 3.
Definition 2.7. Let F : HC → HC be strongly universally Baire. Let θ be
arbitrary, let g be Col(ω, θ)–generic over V , and let A ∈ V [g] be a transitive set.
We say that A is closed under F , or F –closed, iff
(a) (F Col(ω,θ),g ) ” A ⊂ A and
(b) (A, F Col(ω,θ),g A) is amenable, i.e., F Col(ω,θ) M ∈ A for every M ∈ A.
If F : HC → HC is Σ11 (z) in the codes, z ∈ R, then every transitive model A of
−
ZFC with z ∈ A is closed under F by downward Π12 absoluteness.6
There is an obvious correspondence between the notion of closure in the sense
of Definition 2.7 and the concept of A–closure in the sense of [Woo99, Definition
10.139], cf. also [BaCaLa06]. To discuss this correspondence, let us assume that
there is a proper class of Woodin cardinals to have the arguments from the proof
of Theorem 2.4 at our disposal.
Let F : HC → HC be universally Baire in the codes, and let the universally
Baire function f : R → R be a witness to this fact. Let
(2.4) A = {(x, m, n) : ∃y(f (x) = y ∧ y(m) = n)}.
Then A is universally Baire, and if the transitive model A ∈ V is A–closed in
the sense of [Woo99, Definition 10.139], then by [Woo99, Lemma 10.143], if g is
5 In what follows, by “homogeneously Souslin” and “weakly homogeneously Souslin” we mean
∞–homogeneously Souslin and ∞–weakly homogeneously Souslin, respectively.

6 Here and in what follows, ZFC− denotes ZFC without the power set axiom.
generic over A for some poset P ∈ A, then f ”A[g] ⊂ A[g] and in fact f R ∩ A[g] ∈
A[g] and f R ∩ A[g] is definable in A[g] from parameters in A. This is easily seen
to imply that A is F –closed in the sense of Definition 2.7.
On the other hand, let A ⊂ R be universally Baire, say as being witnessed by
T , U , with A = p[T ]. For a countable poset P, for p ∈ P, and for a “nice” name
τ ∈ V P for a real,
(a) p PV τ ∈ p[T ] iff there is j : N → Vθ+1 , N countable and transitive,
{P, p, τ, T θ} ⊂ ran(j), and p PN τ ∈ p[j −1 (T )], and
(b) p PV τ ∈ p[T ] iff there is q ≤P p, j : N → Vθ+1 , N countable and
transitive, {P, q, τ, U θ} ⊂ ran(j), and q PN τ ∈ p[j −1 (U )],
which is easily seen to imply that the function F : HC → HC with
⎧
⎪
⎪ 0 if P is not a poset, p ∈ / P, or τ is not a “nice” name for a real,
⎪
⎨
F or otherwise:
P, p, τ →
⎪
⎪ 1 if p PV τ ∈ p[T ], and
⎪
⎩
2 if p PV τ ∈ p[T ]
is universally Baire in the codes. But then if A is F –closed in the sense of Definition
2.7, then A is A–closed in the sense of [Woo99, Definition 10.139].
We therefore arrive at an equivalent formulation of the concept of Ω–consistency
if in [Woo99, Definition 10.144] we replace the requirement that the certifying
model be A–closed in the sense of [Woo99, Definition 10.139] by the requirement
that it be F –closed in the sense of Definition 2.7 for some F : HC → HC which is
universally Baire in the codes. We shall be interested in a version of Ω–consistency
which allows parameters in Hω2 or beyond.
In what follows we shall frequently consider the languages
L∈,
˙ I˙NS and L∈,
˙ I˙NS
ω1 ω1 ,Ȧ
which arise from the usual first order language of set theory with the binary relation
symbol ∈˙ for membership by adding a unary relation symbol I˙NSω1 and also, in the
latter case, a unary constant symbol Ȧ. In transitive models A of ZFC− + “ω1
exists,” ∈˙ is always to be interpreted by ∈ A, I˙NSω1 is to be interpreted by what A
thinks is the collection of all nonstationary subsets of ω1 , and Ȧ will be interpreted
by a given universally Baire set of reals.
The following defines a strong form of consistency for a statement in L∈, ˙ I˙NS .
ω1
Definition 2.8. Let Ψ(v0 ) be a formula in the language L∈,

˙ I˙NS , and letω1
M ∈ V . Let θ = ℵ1 + Card(TC({M })). We say that Ψ(M ) is honestly consistent
iff for every F : HC → HC which is strongly universally Baire in the codes, if g is
Col(ω, 2θ )–generic over V , then in V [g] there is some transitive model A such that
(a) A is F –closed,
(b) A |= ZFC− ,
(c) (Hθ+ )V ⊂ A,
(d) if S ∈ V , V |= “S is a stationary subset of ω1 ,” then A |= “S is a stationary
subset of ω1 ,” and
(e) A |= Ψ(M ).
184 RALF SCHINDLER
Notice that A will typically be countable in V [g] and by a straightforward

absoluteness argument, if there is an A with (a) through (d) in some outer model7
W ⊃ V of ZFC, then there is some such A in V [g].
Item (d) in Definition 2.8 may of course also be written as
(2.5) (I˙NSω )A ∩ V = (NSω )V .
1 1
Whereas every true statement is obviously honestly consistent, not every hon-
estly consistent statement can be true: e.g., let M = (ω2 )V and Ψ(v0 ) ≡ “v0 has
size ℵ1 .”
Definition 2.9. Let M = (M ; ∈, R) be a model, where M is transitive and
R = (Ri : i < ω1 ) is a list of ℵ1 relations on M . Let ϕ(v0 ) be a Σ1 formula in the
language L∈, ˙ I˙NS . We then let Ψ(M, ϕ) be the following assertion.
ω1
“There is π : M̄ = (M̄ ; ∈, (R̄i : i < ω1 )) → M such that π is (fully)

elementary (in the language associated with M), and ϕ(M̄) holds true.”
Notice that Ψ(M, ϕ) is again a Σ1 sentence in the language L∈,
˙ I˙NS with
ω1
parameter M. If M is as in Definition 2.9, then ϕ(M) trivially implies Ψ(M, ϕ)
in every transitive ZFC− model A which contains M. If Card(M ) = ℵ1 and e.g. for
every x ∈ M there is some i < ω1 with Ri (y) ⇐⇒ y = x, then Ψ(M, ϕ) is simply
equivalent with ϕ(M) in every transitive ZFC− model A which contains M.
Definition 2.10. By Martin’s Maximum∗,++ , abbreviated by MM∗,++ , we
mean the statement that whenever M = (M ; ∈, R) is a model, where M is transitive

and R is a list of ℵ1 relations on M and whenever ϕ(v0 ) is a Σ1 formula in the
language L∈,
˙ I˙NSω such that ϕ(M) is honestly consistent, then Ψ(M, ϕ) is true in
1
V.
By remarks before Definition 2.10, MM∗,++ could also be phrased as the more
cumbersome statement that for M and ϕ(v0 ) as in Definition 2.10, if Ψ(M, ϕ) is
honestly consistent, then Ψ(M, ϕ) is true in V .
Theorem 2.11. MM∗,++ implies MM++ .
Proof. Trivially, if F : HC → HC is strongly universally Baire in the codes,
then for every poset P, every P–generic extension of V is F –closed.
Now let P be a stationary set preserving poset P, let D = {Di : i < ω1 } be a
family of dense subsets of P, and let τ = {τi : i < ω1 } be a collection of names for
stationary subsets of ω1 . For i < ω, let Ti = {(p, ξ) ∈ P × ω1 : p P ξˇ ∈ τi }. Let
κ = Card(TC({P})) + ℵ1 . Let
(Ti : i < ω1 )).
M = (Hκ+ ; ∈, P, D,
Let g be P–generic over V . Set A = (H(2κ )+ )V [g] . We may pick some g which
is Col(ω, 22 )–generic over V such that g ∈ V [g ], so that A ∈ V [g ].
κ
Let F : HC → HC be strongly universally Baire in the codes. Then A is F –

closed, A |= ZFC− , and (H(2κ )+ )V ⊂ A. As P is stationary set preserving, (2.5) also
holds true. The following assertion, call it ϕ, may be written as a Σ1 statament in
L∈,˙ I˙NS and in the parameter M.
ω1
7 We assume to have made sense of “F –closure” for outer models which are not generic
extensions.
There is some filter h ⊂ P such that h ∩ Di = ∅ for all i < ω and such
that {ξ < ω1 : ∃p ∈ h (p, ξ) ∈ Ti } is stationary for every i < ω1 .
This statement is true with h = g. This gives that ϕ(M) is true in A.
We have verified that ϕ(M) is honestly consistent. By MM∗,++ , Ψ(M, ϕ) is
therefore true. But it is straightforward to verify that if
(Ti : i < ω1 ))
π : M̄ = (M̄ ; ∈, P̄, (D̄i : i < ω1 ), (T̄i : i < ω)) → (Hκ+ ; ∈, D,
witnesses Ψ(M, ϕ) and h is as in ϕ(M), then π”h generates a filter G such that
G ∩ Di = ∅ for every i < ω1 and τiG is stationary for every i < ω1 .
The proof of [ClaSch12, Theorem 1.3] actually gives more information than
Theorem 2.11, and [ClaSch12, Theorem 1.3] as well as [AspSch14, Theorem 2.7]
was the inspiration for formulating MM∗,++ .
By [ClaSch12, Theorem 1.3], which is part of the folklore, MM++ is equivalent
to the statement that for all M and ϕ as in Definition 2.9, if there is some stationary
set preserving poset P and some g which is P–generic over V such that V [g] |=
Ψ(M, ϕ), then Ψ(M, ϕ) holds true in V . MM∗,++ results from this statement
by replacing “may be forced by some stationary set preserving forcing” by the
apparently more liberal “is honestly consistent.”
This leads to an obvious question which we formulate as a conjecture, as it is
an obvious variant of Woodin’s Ω–conjecture, cf. e.g. [Woo99, Question (23)].
Conjecture 2.12. Assume that there is a proper class of Woodin cardinals.
Let Ψ(v0 ) be a Σ1 formula in the language L∈, ˙ I˙NSω , and let M ∈ V . If Ψ(M ) is
1
honestly consistent, then there is a stationary set preserving poset P such that if g
is P–generic over V , then Ψ(M ) holds true in V [g].
If Conjecture 2.12 holds true, then in the presence of a proper class of Woodin
cardinals, MM++ is equivalent with MM∗,++ . It is conceivable that Conjecture 2.12
is provable relative to some natural extra hypothesis, cf. [SchWoo∞].
The new axiom MM∗,++ is an attempt to amalgamate MM++ and (∗). The
following is a version of [AspSch14, Theorem 2.7].
Theorem 2.13. MM∗,++ implies (∗).
[AspSch14, Theorem 2.7] in fact proves an equivalence of (∗) with a bounded
version of MM∗,++ . In Section 4 we shall prove a generalized form of [AspSch14,
Theorem 2.7]. Let us introduce the relevant bounded version of MM∗,++ .
From now on we shall write Γ∞ for the set of all sets of reals which are univer-
sally Baire.
Definition 2.14. Let A ∈ Γ∞ . Let Ψ(v0 ) be a formula in the language
L∈,
˙ I˙NSω ,Ȧ , and let M ∈ V . Let θ = ℵ1 + Card(TC({M })). We say that Ψ(M ) is
1
honestly consistent at A iff for every F : HC → HC which is strongly universally
Baire in the codes, if g is Col(ω, 2θ )–generic over V , then in V [g] there is some
transitive model A such that
(a) A is F –closed,
(b) A |= ZFC− ,
(c) (Hθ+ )V ⊂ A,
(d) if S ∈ V , V |= “S is a stationary subset of ω1 ,” then A |= “S is a stationary
subset of ω1 ,” and
186 RALF SCHINDLER
(e) A |= Ψ(M ) with the understanding that in A, Ȧ is interpreted by Ag , i.e.,

ȦA = Ag ∩ A.
The following definition results from [AspSch14, Definition 2.6] by crossing
out the hypothesis that NSω1 be precipitous.
Definition 2.15. Let A ∈ Γ∞ . By A–Bounded Martin’s Maximum∗,++ , ab-
breviated by A–BMM∗,++ , we mean the statement that if M ∈ Hω2 and whenever
Ψ(v0 ) is a Σ1 formula in the language L∈,
˙ I˙NS ,Ȧ such that Ψ(M ) is honestly con-
ω1
sistent at A, then Ψ(M ) holds true in V with the understanding that in V , Ȧ is

interpreted by A.
If Γ ⊂ Γ∞ , then by Γ–Bounded Martin’s Maximum∗,++ , abbreviated by Γ–
BMM∗,++ , we mean the statement that A–Bounded Martin’s Maximum∗,++ is true
for every A ∈ Γ.
We shall prove below, cf. Corollary 4.6, that MM∗,++ implies Γ∞ –BMM∗,++ .
We don’t know an elementary proof of this fact, though.
3. A–iterable mice and A–iterable Pmax conditions

The new result of this section is Theorem 3.14 which produces the existence of
A–iterable Pmax conditions in a way that it may be used to prove (a) =⇒ (b) of
Theorem 4.2. (Here, A ⊂ R.) Along the way, we shall discuss how the hypotheses
of Theorems 3.14 and 4.2 may be realized, cf. Theorem 3.13.
We start out with the following concept which is due to W.H. Woodin. It
encapsulates, in terms of inner model theory, a form of saying that a given A ⊂ R
is universally Baire, cf. Theorem 3.3.
Definition 3.1. Let A ⊂ R. Let N be a countable premouse,8 let δ ∈ N ,
and assume that N |= “ZFC− plus δ is a Woodin cardinal.” Let Σ be an iteration
strategy for N witnessing that N be ω1 + 1 iterable.9 Let τ ∈ N Col(ω,δ) .
We then say that (N, δ, τ, Σ) captures A provided that the following hold true.
(a) Σ satisfies hull condensation10 and branch condensation11 and is posi-
tional.12
(b) If T is an iteration tree on N of successor length θ + 1 < ω1 which is built
according to Σ such that there is no drop on the main branch [0, θ]T , if
T
π0θ : N → MTθ
is the associated iteration map, and if g ∈ V is Col(ω, i(δ))–generic over
MTθ , then
T
π0θ (τ )g = A ∩ MTθ [g].
8 Here and in what follows, in particular in Conjecture 3.4, we actually allow N to be a hod
premouse, cf. [Sa15, Definition 1.34].

9 The relevant iteration trees here and in what follows are finite stacks of normal trees.
10 A hull of a tree according to Σ is again according to Σ. Cf. [Sa15, Definition 1.31] for a
precise statement.
11 If a non–dropping branch model of a tree according to Σ embeds into a non–dropping limit
model on a tree according to Σ in a commuting way, then the branch is according to Σ. Cf.
[Sa15, Definition 2.14] for a precise statement.
12 The iteration strategy for a Σ–iterate N ∗ of N which is induced by Σ does not depend on
how to get from N to N ∗ . Cf. [Sa15, Definition 2.35 (4)]; the last “positional” in [Sa15, Definition
2.35 (4)] should read ”weakly positional,” though.
We also say that (N, Σ) captures A iff there is {δ, τ } ⊂ N such that (N, δ, τ, Σ)
captures A.
Capturing gives rise to “Suslin capturing,” cf. [SchSt∞, Section 1.4]. Let us
present the relevant constructions.
Suppose that (N, δ, τ, Σ) captures A ⊂ R. For every x ∈ R there are T , θ, and
g as in (b) of Definition 3.1 such that x ∈ MTθ [g]; in fact, we may choose T here
in such a way that T does not have any drops. Therefore,

T
(3.1) A = {π0θ (τ )g : T , θ, and g as in (b) of Definition 3.1}.
Let I(N, Σ) be the collection of all MTθ , where T and θ are as in (b) of Definition

3.1. Let MTθ and MTθ both be in I(N, Σ). As Σ is positional by (a) of Definition
3.1, every MTθ ∈ I(N, Σ) has an iteration strategy ΣMTθ which is induced by Σ and
which only depends on the model MTθ and not the particular tree T . Moreover, as
Σ also satisfies hull condensation by (a) of Definition 3.1, Σ is commuting,13 which
means that every ΣMTθ for MTθ ∈ I(N, Σ) has the Dodd–Jensen property in the
sense of [Sa15, Definition 2.35].
We may therefore let
(3.2) <ω1
N∞ , (πP,∞ : P ∈ I(N, Σ))
be the direct limit of the directed system consisting of I(N, Σ), together with the
(unique) respective iteration maps between any two points in I(N, Σ).
As Σ satisfies branch condensation by (a) of Definition 3.1, the system giving
rise to (3.2) induces tree representations for A and its complement as follows.
We let x ∈ p[T ] iff there is some iteration tree T on N of length θ + 1 < ω1
such that
(i) T has no drops at all,
(ii) there is a system (ψi : i ≤ θ) of elementary embeddings such that ψ0 =
πN,∞ and for all i ≤T j ≤ θ, ψi : MTi → N∞ <ω1 T
and ψj ◦ πij = ψi , and
T T
(iii) there is some g which is Col(ω, π0θ (δ))–generic over Mθ such that x ∈
MTθ [g], and in fact x ∈ π0θT
(τ )g .
We let x ∈ p[U ] be defined in exacly the same way except for that in clause (c), we
T T
replace “x ∈ π0θ (τ )g ” by “x ∈
/ π0θ (τ )g .” It is then easy to see that
(3.3) A = p[T ] and R \ A = p[U ].
Still suppose that (N, Σ) captures A. Let κ ≥ ℵ1 be a cardinal. It is straightfor-
ward to verify that there is at most one Σ̃ ⊃ Σ such that Σ̃ is an iteration strategy
for N witnessing that N be κ+ + 1 iterable and
(c) Σ̃ satisfies hull condensation,
and any such Σ̃ will be positional and satisfy branch condensation.
Let us assume that there is such a Σ̃. We may then define a direct limit
+
(3.4) <κ
N∞ , (πP,∞ : P ∈ I(N, Σ̃, κ+ ))
in much the same way as we defined (3.2), where I(N, Σ̃, κ+ ) is the collection of
all MTθ , where T is an iteration tree on N of successor length θ + 1 < κ+ which
is built according to Σ̃ such that there is no drop on the main branch [0, θ]T . We
may then define trees T κ and U κ exactly as we defined T and U above, except for
13 Cf. [Sa15, Definition 2.35 (9)] and [Sa15, Proposition 2.36].
188 RALF SCHINDLER
that “ψi : MTi → N∞ ” gets replaced by “ψi : MTi → N∞

+
<ω1 <κ
.” Of course, we will
again have that
(3.5) A = p[T κ ] and R \ A = p[U κ ].
We claim that T κ and U κ witness that A is κ–universally Baire. By (3.5), we
just need to see that in V [g], p[T κ ] ∪ p[U κ ] = R.
Fix x ∈ V [g]. Let x = ρg , where ρ ∈ V Col(ω,κ) . Inside V , let us construct an
iteration tree T on N of length ≤ κ+ + 1 such that T has no drops at all and is
according to Σ, as follows.14 Say T i + 1 has been defined, where i < κ. If there
MT
is some Eν i = ∅ which is total on MTi and such that for some p ∈ Col(ω, κ),15
MT MT
(3.6) p Col(ω,κ) ρ violates the axiom of BπT i(δ) associated with Eν i
,
0i
MT
then we let EαT = Eν i for the least such ν. If there is no such ν, then we stop the
construction and set lh(T ) = i + 1. This defines T .
We claim that lh(T ) < κ+ . Otherwise there is some p ∈ Col(ω, κ) and a
stationary set S ⊂ [0, κ+ )T such that (3.6) holds true for all i for which there
is some j ≤ i, j ∈ S, such that i + 1 = min([0, κ+ )T \ (j + 1)). We may then
pick some g with p ∈ g such that g is Col(ω, κ)–generic over V . Then S is still
stationary in V [g ], and the usual hull argument yields some j ∈ S such that setting
MT
i + 1 = min([0, κ+ )T \ (j + 1)), ρg satisfies the axiom of BπT i(δ) associated with
0i
EiT . This is a contradiction.
MT
Setting θ = lh(T ) < κ+ , we now have that x is BπT θ(δ) –generic over MTθ .
0θ
Setting ψi = πMTi ,∞ for i ≤ θ, it is now easy to see that T witnesses that x ∈
p[T κ ] ∪ p[U κ ], as desired.
Definition 3.2. Let A ⊂ R. We say that (N, Σ) strongly capures A provided
that (N, Σ) captures A and for every cardinal κ ≥ ℵ1 there is some iteration strategy
Σ̃ ⊃ Σ witnessing that N is κ+ + 1 iterable such that (c) above holds true.
We have shown:
Theorem 3.3. Let A ⊂ R. If (N, Σ) strongly captures A, then A is universally
Baire.
Theorem 3.3 verifies (C) =⇒ (A) of the following bold conjecture. (A) =⇒ (B)
is true by the above–mentioned results of Martin, Steel, and Woodin, cf. the proof
of Theorem 2.4.
Conjecture 3.4. Assume MM plus the existence of a proper class of Woodin
cardinals. Let A ⊂ R. The following are equivalent.
(A) A is universally Baire.
(B) A is determined.
(C) There is some (N, Σ) such that (N, Σ) strongly captures A.
Now assume that N is a premouse and Σ is a iteration strategy for N . If X is
any self–wellordered transitive set and if N ∈ L1 (X),16 then M is a Σ–premouse
14 Cf. [ScTr∞, Definition 4.13].
T
15 BMi T (δ) in the sense of MT .
T (δ) denotes the extender algebra
π0i
at π0i i
16 I.e., N is simply definable from X.
over X iff M is a J–model of the form Jα [E, S,

X] where E codes a sequence of
(partial and total) extenders satisfying the usual axioms for X–premice with the
necessary adjustments due to the following aditional feature. S codes a partial
17
iteration strategy for N , organized as follows.
Let γ < γ + δ ≤ α be such that Jγ [E, X] |= ZFC− and γ is the largest
S,
S,
cardinal of Jγ+δ [E, X]. Suppose that T ∈ Jγ [E, S, X] is Jγ [E,
S,
X]–
least such that T is an iteration tree on N of limit length, T is according
to S γ, but (S γ)(T ) is undefined. Suppose also that δ = lh(T ),
and δ does not have measurable cofinality in Jγ+δ [E, S,
X]. Then Σ(T )
+ δ) is an amenable code for (T , Σ(T )).18
is defined, and S(γ
It is easy to see that if κ < λ < α and both κ and λ are regular cardinals of
M = Jα [E, S],
then Σ Jκ [E, S]
∈ Jκ+ω [E, S].

If M is a Σ–premouse over X and if Γ is an iteration strategy for M, then
we say that Γ moves Σ correctly iff every iterate M∗ of M which is obtained via
Γ is again a Σ–premouse over X. We call M a Σ–mouse over X iff for every
sufficiently elementary σ : M̄ → M with M̄ being countable and transitive there
is some iteration strategy Γ for M̄ which witnesses that M is ω1 + 1 iterable and
which moves Σ correctly.
Theorem 3.14 will make use of the following concept.
Definition 3.5. Let N be a countable premouse, and let Σ be an iteration
strategy for N . Let X be a self–wellordered transitive set such that N ∈ L1 (X),
and let n < ω. Then we denote by
Mn#,Σ (X)
the unique Σ–mouse M over X, if it exists, such that M is sound above X, M is
not n–small above X, but every proper initial segment of M is n–small above X.
As Mn#,Σ (X) is sound above X, it also projects to X, and it is its own least
initial segment which satisfies “There is a measure above n Woodin cardinals.”
Lemma 3.6. Suppose that there is a proper class of Woodin cardinals. Let N
be a countable premouse, and let Σ be an iteration strategy for N witnessing that
N is < OR iterable. Assume that Σ satisfies hull condensation.
Then for every self–wellordered transitive set X with N ∈ L1 (X) and for ev-
ery n < ω, Mn#,Σ (X) exists, and there is an iteration strategy Γ witnessing that
Mn#,Σ (X) be < OR–iterable which moves Σ correctly.
Proof. (Sketch.) Let Y be any self–wellordered transitive set such that N ∈
L1 (Y ). We perform an (n+1)–small L[E, Σ](Y ) construction, an obvious variant of
the L[E](Y ) construction, where we also keep feeding in Σ. By hull condensation,
all models from the L[E, Σ](Y ) construction will be Σ–premice over Y . Moreover,
if M is the transitive collapse of a countable substructure of an ω–small model
(with no definable Woodin cardinal) from the L[E, Σ](Y ) construction, then the
realization strategy will witness that M be countably iterable and it will move Σ
correctly. In particular, the (n + 1)–small L[E, Σ](X) construction reaches a model
M which is not n–small, but the L[E, Σ](Y ) constructions for Y with X ∈ L1 (Y )
17 Cf. [ScTr∞, Section 3] and the discussion in [ScTr∞, Appendix B].

18 Hence if T on N is according to S,

then T is according to Σ.
190 RALF SCHINDLER
produce the relevant Q–structures to show that M is actually < OR iterable via
an iteration stratagy which moves Σ correctly.
Let N be a countable premouse, and let Σ be an iteration strategy for N
witnessing that N is < OR iterable. Assume that Σ satisfies hull condensation.
Suppose also that Σ HC is strongly universally Baire in the codes and that there
is a proper class of Woodin cardinals. It is straightforward to verify that then for
every poset P and for every g which is P–generic over V , (Σ HC)P,g is an iteration
strategy for N with hull condensation which witnesses that N is < OR iterable in
V [g], and hence Σ = (Σ HC)P,g V .19
Lemma 3.7. Suppose that there is a proper class of Woodin cardinals. Let N
be a countable premouse, and let Σ be an iteration strategy for N witnessing that
N is < OR iterable. Assume that Σ satisfies hull condensation and that Σ HC is
strongly universally Baire in the codes.
Then the function
X → Mn#,Σ (X), where X ∈ HC is self–wellordered and N ∈ L1 (X)
is strongly universally Baire in the codes.
Proof. (Sketch.) We use Lemma 2.2. Let the trees T ∗ , U ∗ witness that
Σ HC is strongly universally Baire in the codes. Let κ be an uncountable cardinal,
and let θ >> κ be a cardinal such that there are at least n + 1 Woodin cardinals
between κ and θ. Say that T ∗ 2θ and U ∗ 2θ witness that p[T ∗ ]∩V is θ–universally
Baire.
It is straightforward to design a tree T searching for x, y, M , Σ̄, P , g, σ such
that
(i) x ∈ R codes some X ∈ HC with N ∈ L1 (X),
(ii) y ∈ R codes M , a Σ̄–premouse over X which is sound above X, is not
n–small above X, but every of its proper initial segments is n–small above
X,20
(iii) σ : P → H(2θ )+ is a fully elementary embedding such that P is countable
and transitive and {N, Hκ+ , θ, T ∗ 2θ , U ∗ 2θ } ⊂ ran(σ),
(iv) g is Col(ω, σ −1 (κ))–generic over P ,
(v) X ∈ σ −1 (Hκ+ )[g],
(vi) p[σ −1 (T ∗ 2θ )] is consulted in P [g] to yield T → Σ̄(T ) à la (2.2), and
(vii) M results from a L[E, Σ̄](X)–construction performed inside P [g] and us-
ing extenders with critical point above σ −1 (κ).
Notice that if h is Col(ω, σ −1 (θ)–generic over P [g], then
p[σ −1 (T ∗ 2θ )] ∩ P [g][h] = p[T ∗ ] ∩ P [g][h],
so that an M as above will actually be a Σ–premouse over X.
We claim that in fact every M as above is a Σ–mouse over X. To see this, let
the strategy Γ for countable trees on M be defined as follows. Suppose that T on M
is of countable limit length and according to Γ. Then T induces a (non–dropping)
tree U(T ) on the background universe P [g] as in [MiSt94, §12]. We may construe
U(T ) as an iteration tree on P . Construed that way, we may let b be a cofinal
19 We here use the notation from (2.2).
20 In
the light of what is to follow, we my and shall in fact arrange that y depends on x, i.e.,
x → y will be a function.
U (T )
branch through U such that the direct limit model Mb may be emdedded back
into H2θ + in a commuting way. Notice that M is (n + 1)–small, so that U(T ) is
“simple” enough so that there is such a branch.21 We may then let Γ(T ) = b,
construed as a branch through T .
Now let T have countable successor length α + 1 and be according to Γ. Let
U(T ) be the induced tree on P [g]. Then MTα embeds into a model Nξ of the
U (T )
L[E, Σ̃](X)–construction of Mα , where the predicate Σ̃ is given by consulting
U (T )
p[π0α (σ −1 (T ∗ 2θ ))]
U (T ) U (T )
inside Mα . However, {σ −1 (T ∗ 2θ ), σ −1 (U ∗ 2θ )} ⊂ P , and Mα , construed
U (T )
as an iterate of P , may be reembedded into H2θ + , say via k : Mα → H2θ + , in a
way such that
U (T )
k ◦ π0α = σ.
U (T )
But then p[π0α (σ −1 (T ∗ 2θ ))] and p[T ∗ ], restricted to the relevant generic exten-
U (T )
sions of Mα , are equal with one another. This implies that Σ̃ conforms with Σ,
i.e., MTα is a Σ–premouse over X. We have shown that Γ moves Σ correctly.
Definition 3.8. Let A ⊂ A. We say that (N, Σ) u.B.–strongly captures A22
provided that (N, Σ) strongly captures A and Σ HC is universally Baire in the
codes.
Let A ⊂ R. How can we construct a pair (N, Σ) which u.B.–strongly captures
A? One key idea, capturing by self–justifying systems, cf. Definition 3.9, is due to
W.H. Woodin. The proof of Lemma 3.10 is as in [St16], cf. [St16, Lemma 3.7], cf.
also [SchSt∞, Section 3.7].
Definition 3.9. Let A ⊂ P(R) be a countable collection of sets of reals. We
say that A is a self justifying system iff the following holds true. Every A ∈ A
admits a scale (≤n : n < ω) such that each individual ≤n belongs to A, too, and
such that if A ∈ A, then R \ A ∈ A.
We refer the reader to [Wi15] for a thorough discussion of how self justifying
systems may be constructed.
Lemma 3.10. (Woodin) (Term Condensation) Let A be a self–justifying system.
Let M be a transitive model of ZFC− , and let δ ∈ M . Let C ⊂ ω δ be a comeager
set of Col(ω, δ)–generics over M and suppose that for each A ∈ A there is a term
τA ∈ M Col(ω,δ) such that whenever G ∈ C, then τAG = A ∩ M [G]. Let π : M̄ → M
be elementary with {δ} ∪ {τA : A ∈ A} ⊂ ran(π). Let π(δ̄) = δ and π(τ̄A ) = τA for
A ∈ A.
Then whenever g is Col(ω, δ̄)–generic over M̄ , then τ̄Ag = A ∩ M̄ [g] for all
A ∈ A.
Proof. Fix any A ∈ A for a while, and let (ψn : n < ω) be a scale on A such
that for every n < ω, if ≤n is the prewellorder on R given by ψn then ≤n ∈ A. Let
τn ∈ M be such that τnG =≤n ∩M [G] for all G ∈ C. Let φn ∈ M Col(ω,δ) be such
that for every G which is Col(ω, δ)–generic over M , φGn is the norm on A ∩ M [G]
21 Cf. [MaSt94] on the existence of such a realizable branch.

22 Of course, “u.B.” here stands for “universally Baire.”
192 RALF SCHINDLER
given by τnG . Let Un ∈ M Col(ω,δ) be a term for the nth level of the tree associated
to these norms, i.e., for all G being Col(ω, δ)–generic over M ,
U̇nG = {(x n, (φG
0 (x), . . . , φn−1 (x))) : x ∈ A ∩ M [G]}.
G
Now let G0 be Col(ω, δ)–generic over M , and let G1 be Col(ω, δ)–generic over
M [G0 ]. Then for any appropriate s, α , and h ∈ {0, 1} we have that (s, α
) ∈ U̇nGh
iff there is some ph ∈ Gh such that
Col(ω,δ)
ph M ˇ ) ∈ U̇n .
(š, α
As C is comeager, we may build G∗0 ∈ C and G∗1 ∈ C such that for some real y,
(p0 y) n ∈ G∗0 and (p1 y) n ∈ G∗1 for all n < ω. In particular, we have
G∗ G∗
M [G∗0 ] = M [G∗1 ], which implies τn 0 =≤n ∩M [G∗0 ] =≤n ∩M [G∗1 ] = τn 1 , and so
∗ ∗ ∗ ∗
G G G G
) ∈ U̇nG0 iff (s, α
U̇n 0 = U̇n 1 . Hence (s, α ) ∈ U̇n 0 iff (s, α
) ∈ U̇n 1 iff (s, α
) ∈ U̇nG1 .
G
This means that U̇n is independent from the particular choice of the Col(ω, δ)–
generic G, and therefore there is Un ∈ M such that Un = U̇nG for all G which are
Col(ω, δ)–generic over M . In fact, Un ∈ ran(π) for every n < ω. Let U be the tree
whose nth level is Un . (Possibly U ∈ / M .)
Claim 3.11. Whenever G is Col(ω, δ)–generic over M , A ∩ M [G] ⊂ p[U ] ⊂ A.
To verify the claim, notice that A ∩ M [G] ⊂ p[U ] is obvious from the definition
of U . Let (x, f ) ∈ [U ]. Let G be Col(ω, δ)–generic over M . Let n < ω. Then the
nth level of U is UnG , and so we can find a real xn ∈ A with xn n = x n and
∀i < n(φGi (xn ) = f (i)). So for any i < ω, φi (xn ) is eventually constant as n → ω.
G
Hence ψi (xn ) is eventually constant as n → ω. But (ψi : i < ω) is a scale on A,

thus x ∈ A. This shows p[U ] ⊂ A. We have shown Claim 3.11.
For any n < ω, we now have that
Col(ω,δ)
M ∀x (x ∈ τA → (x n, (φ0 (x), . . . , φn−1 (x))) ∈ Un ).
The elementarity of π gives that
Col(ω,δ̄) ∀x (x ∈ τ̄A → (x n, (π −1 (φ0 )(x), . . . , π −1 (φn−1 )(x))) ∈ Ūn ],
where Ūn = π −1 (Un ). Let Ū be the tree whose nth level is Ūn . Of course, p[Ū ] ⊂
p[U ]. But now if x ∈ τ̄Ag , where g is Col(ω, δ̄)–generic over M̄ , then x ∈ p[Ū ] ⊂
p[U ] ⊂ A, by Claim 3.11. So τ̄Ag ⊂ A.
However, the same reasoning with R\A ∈ A and τR\A instead of A and τA shows
that τ gR\A ⊂ R\A, and thus in fact τ̄Ag = A ∩ M̄ [g], as τ̄R\A
g
= (R ∩ M̄ [g]) \ τ̄Ag .
The following hypothesis was introduced by D.A. Martin and J.R. Steel, cf.
[MaSt94, p. 47].
Definition 3.12. Let E be a class of (total) V –extenders. We say that the
Unique Branch Hypothesis, UBH for short, holds true for E iff whenever T is an
iteration tree on V of limit length which only uses extenders from E and its images,
T has at most one cofinal well–founded branch.
Let us fix A ⊂ R, A ∈ Γ∞ . Suppose that E is a class of (total) V –extenders
such that UBH holds true for E.
Let N be a premouse, and let Σ0 be an iteration strategy for N witnessing that
N is < OR iterable. Assume that Σ0 satisfies hull condensation.23
23 Σ is going to play a pasive role in what follows, we shall need that its Wadge rank is large
0
enough so as to allow (†) to hold true.
For some self–wellordered transitive set X, let us perform an L[E, Σ0 ](X) con-
struction in much the same way as in the proof of Lemma 3.6, with the following
changes, though.
(i) We don’t impose any smallness restrictions on the premice occurring in
the L[E, Σ0 ](X)–construction, but on the other hand
(ii) we only use extenders from E as background certificates.
By UBH, the construction cannot break down as all the models appearing in that
construction will be fully iterable.
Let us pretend X = ∅, let us write L[E, Σ0 ] for the resulting model, and let us
suppose that E witnesses that δ is a Woodin cardinal. Let H be Col(ω, 2δ )–generic
over V . Suppose that in V [H],
(†) there is a self–justifying system A = {An : n < ω} with AH ∈ A, there
are τn ∈ L[E, Σ0 ]Col(ω,δ) , n < ω, and there is some comeager set C of
Col(ω, δ)–generics over L[E, Σ0 ] such that if G ∈ C, then
τnG = An ∩ L[E, Σ0 ][G]
for every n < ω.
Write λ = δ ++L[E,Σ0 ] . We may assume that τn ∈ Lλ [E, Σ0 ] for every n < ω.
In V , let us pick an elementary embedding
σ : P → Lλ [E, Σ0 ],
where P is countable and transitive and {δ} ∪ {τn : n < ω} ⊂ ran(σ). For n < ω,
write τ̄n = σ −1 (τn ).
UBH for E yields a canonical iteration strategy Σ for P . Namely, if T ∈ V is
on P , of limit length, and according to Σ, then we may first use σ to copy T onto
L[E, Σ], getting σT on L[E, Σ0 ], and we may then lift σT to a (non–dropping) tree
U(σT ) on V as in [MiSt94, §12], cf. the proof of Lemma 3.7. The unique cofinal
well–founded branch through U(σT ) will then give rise to Γ(T ): formally, in fact,
Γ(T ) is the unique cofinal well–founded branch through U(σT ).
For future reference, cf. Theorem 3.13, we shall refer to the strategy Σ thus
defined as the E–induced pullback strategy for P . We will actually have that Σ HC
is universally Baire in the codes, cf. Theorem 5.2. We defer a proof of this result
to the appendix, Section 5, cf. Lemma 5.2.
Let us assume that T ∈ V ∩ HC on P is according to Σ. For i < lh(T ), we may
write
σ i : MTi → MσT
i
for the canonical copying map, and we may write
U (σT )
π i : MσT
i → (Nξi )Mi , some ξ,
U (σT )
for the map obtained by lifting σT to U(σT ); here, (Nξi )Mi is a model from the
U (σT ) U (σT ) 24
L[E, π0i (Σ0 )]–construction of Mi . If [0, i)T does not drop, then, writing
U (σT )
λ∗ = π0i (λ),
U (σT )
U (σT )
π i ◦ σ i : MσT
i → Lλ∗ [E, π0i (Σ0 )]Mi
U(σT ) U(σT )
24 We have π
0i (Σ0 ) = Σ0 Mi , as Σ0 has hull condensation, but we don’t need
that here.
194 RALF SCHINDLER
T U (σT ) U (σT )
is elementary with π i ◦σ i ◦π0i = π0i ◦σ = π0i (σ), and hence by absoluteness,
U (σT ) U (σT )
as P and T are countable in Mi , inside Mi there is some elementary map
U (σT )
ϕi : MTi → Lλ∗ [E, π0i (Σ0 )]Mi
σT
T U (σT ) U (σT )
with ϕi ◦ π0i = π0i (σ). By elementarity of π0i , there is hence an elementary
map
ψ i : MTi → Lλ [E, Σ0 ]
T T
such that ψ ◦ i
= σ. In particular, if n < ω, then ψ i (π0i
π0i (τ̄n )) = τn , so that by
the Term Condensation Lemma 3.10, if g ∈ V is Col(ω, π0i (σ −1 (δ)))–generic over
T
MTi , then
T
(π0i (τ̄0 ))g = AH ∩ MTi [g] = A ∩ MTi [g].
We have shown that (b) of Definition 3.1 holds true for (P, σ −1 (δ), τ̄0 , Σ).
This comes close to having (P, σ −1 (δ), τ̄0 , Σ) capture A, but UBH for E does
not seem to abstractly yield (a) of Definition 3.1, whereas in practice it will, cf. the
discussion after the statement of Theorem 3.13.
In order to verify (†), let us assume the following. Inside V [H],
(††) there is a self–justifying system A with AH ∈ A such that for every n < ω,
CΣ11 (An ) (Lκ [E, Σ0 ]) ⊂ L[E, Σ0 ],
where κ = δ +L[E,Σ0 ] .
Here, Σ11 (D)is the set of all sets of reals which are Σ11 in D, and for Δ ⊂ P(R) and x,
y ∈ R, y ∈ CΔ (x) iff y is Δ(x) in a countable ordinal; if M is a countable transitive
set and Y ⊂ M , then Y ∈ CΔ (M ) iff for comeager many Col(ω, M )–generics G,
{n < ω : (∪G)(n) ∈ M } ∈ CΔ (∪G). (Cf. [St16].)
Assuming (††), let us fix n < ω. We aim to produce τ = τn as in (†) by an
argument as in [SchSt∞, Section 3.7]. Let us write D = An . We let (p, σ) ∈ τ iff
p ∈ Col(ω, δ), σ ∈ L[E, Σ0 ]Col(ω,δ) is a standard term for a real, and for
comeager many g being Col(ω, δ)–generic over L[E, Σ0 ], if p ∈ g, then
σ g ∈ D.
Trivially, τ ⊂ L[E, Σ0 ]Col(ω,δ) , in fact, writing κ = δ +L[E,Σ0 ] , we have that τ ⊂
Lκ [E, Σ0 ].
We claim that
(3.7) τ ∈ L[E, Σ0 ].
To this end, let x ∈ R be Col(ω, Lκ [E, Σ0 ])–generic over L[E, Σ0 ], i.e.,
ϕ
(ω; Ex ) ∼
= (Lκ [E, Σ0 ]; ∈).
Let us write τx = {n < ω : ϕ(n) ∈ τ } for the “real” coding τ relative to x. If
(p, σ)∈ τ , then the comeager set of g witnessing (p, σ) ∈ τ may be taken of the
form n<ω On , where each On is open dense. It is then clear that
(3.8) {τx } ∈ Σ11 (D)({x}).
However, CΣ11 (D) (Lκ [E, Σ0 ]) ⊂ L[E, Σ0 ] and (3.8) imply that τx ∈ L[E, Σ0 ][x] and
hence also τ ∈ L[E, Σ0 ][x]. If x and x are mutually Col(ω, Lκ [E, Σ0 ])–generic over
L[E, Σ0 ], then τ ∈ L[E, Σ0 ][x] ∩ L[E, Σ0 ][x ]. But τ ⊂ L[E, Σ0 ], and therefore (3.7).
We now claim that in V [H], there is a comeager set C of Col(ω, δ)–generics over
L[E, Σ0 ] such that if G ∈ C, then τ G = D ∩ L[E, Σ0 ][G]. For p ∈ Col(ω, δ) and
σ ∈ L[E, Σ0 ]Col(ω,δ) a standard term for a real, let Cp,σ = {G : p ∈ G ∧ σ G ∈ D}

and Cp,σ = {G : p ∈ G ∧ σ G ∈ / D}, with the understanding that G is always
Col(ω, δ)–generic over L[E, Σ0 ]. We have τ = {(p, σ) : Cp,σ is comeager in Up },
where Up = {G : p ∈ G}.

We claim that for all σ, {p ∈ Col(ω, δ) : Cp,σ or Cp,σ is comeager in Up } is dense
in Col(ω, δ). Fix σ. Let q ∈ Col(ω, κ). Suppose that Cq,σ is not comeager. As Cq,σ
has the property of Baire, there is an open set O such that (O \ Cq,σ ) ∪ (Cq,σ \ O)

is meager. If O = ∅, then Cq,σ is comeager. Let us assume that O = ∅. Then there
is some p such that Up \ Cq,σ is meager, where Up = {G : p ∈ G}. We may assume
that lh(p) ≥ lh(q). We must have that p ≤ q, as otherwise Up \ Cq,σ = Up , which is
not meager. But now Cp,σ is comeager in Up , as Up \ Cp,σ ⊂ Uq \ Cp,σ .
∗
If Cp,σ or Cp,σ is comeager, then let Cp,σ denote the comeager one of them.
There are only countably many such p’s and σ so that
*
∗
C= Cp,σ
p,σ
is a comeager set.
Now let G ∈ C. Then σ G ∈ τ G implies that there is some p ∈ G with (p, σ) ∈ τ ,
so that there is some p ∈ G with Cp,σ being comeager in Up , hence σ G ∈ A. As

Cp,σ is not comeager in Up iff Cp,σ is comeager in Up , the same reasoning yields
that σ ∈
G
/ τ implies that σ ∈
G G
/ A. We have verified (†).
Recall that by a theorem of J. Steel, cf. [Lar04, Theorem 3.3.19], hypothesis
(1) of Theorem 3.13 implies that the pointclass Γ∞ admits the scale property. We
have thus shown the following Theorem, via (††) and (†) and also Theorem 5.2.
Theorem 3.13. Assume that E is a class of V –extenders such that
(1) E witnesses that there is a proper class of Woodin cardinals, and
(2) UBH holds true for E.
Let us also assume the following.
(3) Let A ⊂ Γ∞ be countable. There is then some premouse N and some iter-
ation strategy Σ0 for N witnessing that N is < OR–iterable such that Σ0
has hull condensation, and if δ is a Woodin cardinal and if H is Col(ω, 2δ )–
generic over V , then in V [H],
CΣ11 (D) (Lκ [E, Σ0 ]) ⊂ L[E, Σ0 ]
for every D ∈ A.
(4) E–induced pullback strategies for collapses of countable substructures of
sufficiently small initial segments of L[E, Σ0 ] satisfy hull condensation
and branch condensation and are positional.
For every A ∈ Γ∞ there is then some (P, Σ) u.B.–strongly capturing A.
It is conceivable that (3) follows from (1) and (2) (or even just from (1)), and
one may attempt to prove this by an induction on the Wadge rank of the set A ∈ Γ∞
in question. This leads to the core model induction, cf. [SchSt∞] and [Sa15], and
specifically to Strong Mouse Set Capturing, cf. [Sa15, p. 8]. (4) should also follow
from (1) and (2), cf. e.g. [Sa15, Theorem 2.42] on “positional.”
The new theorem of this section is now a generalized version of [AspSch14,
Lemma 2.12].
196 RALF SCHINDLER
Woodin is able to produce iterable “coarse” mice which capture a given set of
reals in a determinacy model by using the HOD of a slightly stronger determinacy
model, cf. [KoeWoo10, Theorem 5.40] and [St16, Lemma 3.12]. Those coarse
mice do not seem to be useful for our purposes, though, as we don’t seem to be
able to make sense of a directed system generated by a coarse mouse N giving rise
<κ+
to a version of N∞ as in (3.4), and also the coarse mice don’t seem to produce
a substitute for the total code invariant universally Baire function derived from
X → M2#,Σ (N, X) and obtained via Theorem 3.14 which plays a crucial role in the
proof of (a) =⇒ (b) of Theorem 4.2, cf. (4.13).
Pmax conditions are defined in [Woo99, Definition 4.33]. A transitive model
(M ; ∈, I) is called a Pmax precondition iff there is some a such that (M ; ∈, I, a)
is a Pmax condition, cf. [Lar10, Definition 2.1]. Cf. [Woo99, Definition 4.3] for
A–iterability, where A ⊂ R.
Theorem 3.14. Let A ∈ Γ∞ , and suppose that (N, δ, τ, Σ) captures A. Let
X ∈ HC, and suppose that
M = M2#,Σ (N, X)
exists. Let δ0 be the bottom Woodin cardinal of M , let g0 ∈ V be (Col(ω1 , < δ0 ))M –
generic over M , and let g1 ∈ V be Q–generic over M [g0 ], where Q ∈ M [g0 ] is the
standard c.c.c. forcing for Martin’s Axiom.
Then
(3.9) p = (M [g0 , g1 ]; ∈, (NSω1 )M [g0 ,g1 ] )
is an A–iterable Pmax precondition.
Proof. As M [g0 ] |= “NSω1 is presaturated,” cf. e.g. [Woo99, Theorem 2.61],
well–known arguments show that p is an iterable Pmax precondition, cf. [Woo99,
Lemma 3.10 and Remark 3.11]. We thus need to see that
(3.10) A ∩ M [g0 , g1 ] ∈ M [g0 , g1 ]
and if
(3.11) i : p → p∗ = (M ∗ ; ∈, I ∗ )
arises from a countable generic iteration of p, then
(3.12) i(A ∩ M [g0 , g1 ]) = A ∩ M ∗ .
To this end, let us first show that A∩M is δ1 –universally Baire inside M , where
δ1 is the top Woodin cardinal of M .
Let
T = (T δ1 )M and U = (U δ1 )M
be the trees as defined on p. 187 for κ = δ1 and running the definition inside M .
This is possible, as Σ M is amenable to M .
Let us fix some G ∈ V which is Col(ω, δ1 )–generic over M . We claim that
(3.13) p[T ] ∩ M [G] = A ∩ M [G].
Let us first assume that x ∈ p[T ] ∩ M [G], as being witessed by T , N ∗ , and
h. As Σ has branch condensation in V , item (b) gives that T is according to Σ.
Therefore, x ∈ A, as (N, δ, τ, Σ) captures A.
Now let x ∈ A ∩ M [G]. The argument on pp. 188f. then shows that x ∈ p[T ].
The very same argument shows that

(3.14) p[U ] ∩ M [G] = (ω ω \ A) ∩ M [G].
Now by (3.13), A ∩ M [g0 , g1 ] = p[T ] ∩ M [g0 , g1 ] = A ∩ M [g0 , g1 ], so that (3.10)
holds true.
We now aim to verify that (3.12) holds true for all i as in (3.11). Suppose
otherwise, and let i : p → p∗ = (M ∗ ; ∈, I ∗ ) be as in (3.11) such that (3.12) is
false, i.e., i(A ∩ M [g0 , g1 ]) = A ∩ M ∗ . Exploiting the fact that δ1 is a Woodin
cardinal in M [g0 , g1 ] as being witnessed by lifts of extenders from the M –sequence,
we may iterate M [g0 , g1 ] using such lifts to get some countable iterate M + [g0 , g1 ]
of M [g0 , g1 ] together with some iteration map
j : M [g0 , g1 ] → M + [g, g1 ]
such that for some G+ which is Col(ω, j(δ1 ))–generic over M + [g0 , g1 ],
i (M [g0 , g1 ]|(δ0+3 )M ) ∈ M + [g0 , g1 ][G+ ].
(Cf. [St10].) As M = M2#,Σ (N, z) is a Σ–mouse, the proof showing (3.13) and
(3.14) also yields that
(3.15) p[j(T )] ∩ M + [g0 , g1 ][G+ ] = A ∩ M + [g0 , g1 ][G+ ]
and
(3.16) p[j(U )] ∩ M + [g0 , g1 ][G+ ] = (ω ω \ A) ∩ M + [g0 , g1 ][G+ ].
This means that M + [g0 , g1 ][G+ ] knows that i (M [g0 , g1 ]|(δ0+3 )M ) is a generic
∗
iteration such that i(A ∩ M [g0 , g1 ]) = p[j(T )] ∩ M ∗ |i(δ0+2 )M . This assertion is
+
forced to be true over M [g0 , g1 ], so that by the elementarity of j we may conclude
that there is some G∗ which is Col(ω, δ1 )–generic over M [g0 , g1 ] such that there is
some generic iteration
i : (M [g0 , g1 ]|(δ0+3 )M ) → M
with i (A ∩ M [g0 , g1 ]) = p[T ] ∩ M . However, i may be lifted to a generic iteration
i ⊃ i with
i : M [g0 , g1 ] → M .
By (3.13) and (3.14), T and U witness that A ∩ M [g0 , g1 ] is δ1 –universally Baire in
M [g0 , g1 ]. Assume that i (A∩M [g0 , g1 ])\p[T ]∩M = ∅, and let x ∈ i (A∩M [g0 , g1 ])\
p[T ] ∩ M . As i (M [g0 , g1 ]|(δ0+3 )M ) ∈ M [g0 , g1 ][G∗ ], x ∈ M [g0 , g1 ][G∗ ], so that
x ∈ / p[T ] implies that x ∈ p[U ] ⊂ p[i (U )]. By elementarity i (A ∩ M [g0 , g1 ]) =
i (A ∩ M [g0 , g1 ]) = p[i (T )] ∩ M , so that x ∈ p[i (T )] ∩ p[i (U )]. By absoluteness,

M |= p[i (T )] ∩ p[i (U )] = ∅, and hence by elementarity M [g0 , g1 ] |= p[T ] ∩

p[U ] = ∅. This contradicts (3.13) and (3.14). A symmetric argument shows that
(p[T ] ∩ M ) \ i (A ∩ M [g0 , g1 ]) = ∅ is impossible.
We reached a contradiction!
4. (∗)Γ∞ is equivalent to Γ∞ –BMM++,∗

We formulate a natural strengthening of the axiom (∗) of [Woo99] and prove
that it is equivalent with Γ∞ –BMM∗,++ . For Γ ⊂ P(R), a filter G ⊂ Pmax is Γ–
generic iff G ∩ D = ∅ for every open D ⊂ Pmax for which there is some D∗ ∈ Γ
with
(4.1) D = {p ∈ Pmax : ∃x ∈ D∗ (x codes a transitive set and p = decode(x))},
198 RALF SCHINDLER
i.e., such that D may be coded by a set of reals in Γ.

Definition 4.1. Let Γ ⊂ P(R). By (∗)Γ we mean the statement that
(1) every A ∈ Γ is determined, and
(2) there is some Γ–generic filter G ⊂ Pmax such that P(ω1 ) ⊂ L(R)[G].
Therefore, (∗) is equal to (∗)P(R)∩L(R) .
Theorem 4.2. Assume that there is a proper class of Woodin cardinals. As-
sume further that for every A ∈ Γ∞ there is some (N, Σ) which u.B.–strongly
captures A.
The following are equivalent.
(a) (∗)Γ∞ .
(b) Γ∞ –BMM∗,++ .
Proof. We first aim to verify (b) =⇒ (a). To this end, let us fix some A0 ⊂ ω1
L[A ]
such that ω1 0 = ω1 . Let G be the set of all p = (M0 ; ∈, J0 , a0 ) ∈ Pmax such that
there is some generic iteration
((Mi , πi,j : i ≤ j ≤ ω1 ), (Gi : i < ω1 ))
of M0 = p such that π0,ω1 (a0 ) = A0 and, writing Mω1 = (Mω1 ; ∈, Jω1 , A0 ), every
set in
Jω+1 = (P(ω1 ) ∩ Mω1 ) \ Jω1
is stationary in V .
We claim that G is a Γ∞ –generic filter and that
(4.2) L(R)[G] = L(R)[A0 ] = L(P(ω1 ))
holds true for G. In order to verify this, we shall need to prove the following three
Claims 4.3, 4.4, and 4.5.
Claim 4.3. G is a filter.
Claim 4.4. If D is an open dense subset of Pmax for which there is some
D∗ ∈ Γ∞ with (4.1), then D ∩ G = ∅.
By a standard Pmax –argument, if p ∈ G, then there is a unique generic iteration
((Mi , πi,j : i ≤ j ≤ ω1 ), (Gi : i < ω1 ))
of M0 = p such that π0,ω1 (a0 ) = A0 . Assuming Claims 4.3 and 4.4 and following
[Woo99], we shall then write P(ω1 )G for the set of all X ⊂ ω1 for which there is
some p ∈ G such that if
((Mi , πi,j : i ≤ j ≤ ω1 ), (Gi : i < ω1 ))
is the generic iteration of M0 = p with π0ω1 (a0 ) = A0 , then X ∈ ran(πi,ω1 ) for
some i < ω1 .
Claim 4.5. P(ω1 ) = P(ω1 )G .
If NSω1 were assumed to be saturated, then Claim 4.3 would be given by
[Woo99, Theorem 4.74] and Claim 4.5 would follow from [Woo99, Lemma 3.12
and Corollary 3.13]. Assuming the existence of a measurable cardinal κ above a
Woodin cardinal δ as well as Bounded Martin’s Maximum++ , though, one can prove
Claims 4.3 and 4.5 by an easy modification of the forcing developed in [ClaSch09],
as follows. Let g be Col(ω1 , < δ)–generic over V . Inside V [g], (NSω1 )V [g] is pre–
saturated and hence precipitous, cf. [Woo99, Theorem 2.61].
From this, [ClaSch09] designs a stationary set preserving forcing which (for a
given regular cardinal θ ≥ (2κ )+ ) adds a generic iteration
(Mi , πi,j : i ≤ j ≤ ω1 )
of a countable model M0 = (M0 ; ∈, I0 ) such that Mω1 = ((Hθ )V [g] ; ∈, (NS)V [g] ).
This immediately gives Claim 4.5 by Bounded Martin’s Maximum++ . Also, if p,
q ∈ G, then we may assume without loss of generality that p, q, A0 ∩ ω1M0 ∈ M0 ,
so that Bounded Martin’s Maximum++ also yields Claim 4.3.
However, Claims 4.3 and 4.5 also follow by a simplified variant of the following
argument which shows Claim 4.4 from our hypothesis (b) together with a proper
class of Woodin cardinals.
Let us fix D ⊂ Pmax , an open dense set in Pmax , such that there is some
D∗ ∈ Γ∞ with (4.1). Let δ be a Woodin cardinal, and let κ > δ be measurable.
Write ρ = 22 . As D∗ is 2ρ –universally Baire, we may pick trees T and U on ω ×22
κ ρ
such that
D∗ = p[T ] and Col(ω, 2ρ ) p[Ǔ ] = ω ω \ p[Ť ].
Let g be Col(ω1 , < δ)–generic over V , and let h be Col(ω, ρ)–generic over V [g].
By [Woo99, Theorem 2.61] and [Woo99, Lemma 3.10 and Remark 3.11],
p0 = ((H2κ )V [g]; ∈, (NSω1 )V [g] , A0 )
is then a Pmax condition in V [g, h]. The statement
(4.3) ∀p ∈ Pmax ∃q ∈ Pmax (q ≤Pmax p ∧ q ∈ D)
which expresses that D is dense in Pmax is Π12 in any set of reals coding Pmax ⊕ D
in a natural way, so that by absoluteness, cf. Theorem 2.4, there is some q = (N0 ; ∈
, J0 , A0 ) ∈ V [g, h] belonging to the set of Pmax –conditions coded by (D∗ )g,h and
such that q <Pmax p0 . Let
(4.4) j0 : p0 → (N̄0 ; ∈, J¯0 , A0 )
be a generic iteration of p0 such that p0 , j0 ∈ N0 , i.e., (4.4) witnesses that q < p0 .
Inside V [g, h], let
(Sξ : ξ < (ρ+ )V ) ∈ V [g, h]
be a partition of (ρ+ )V = (ω1 )V [g,h] into stationary sets. Working inside V [g, h],
we may then build a generic iteration
(Ni , σi,j : i ≤ j ≤ (ρ+ )V ),
of N0 = (N0 ; ∈, J0 , A0 ) = q such that, writing N(ρ+ )V = (N ; ∈, J, A ),
∀S ∈ (P((ρ+ )V ) ∩ N ) \ J ∃ξ < (ρ+ )V ∃β < (ρ+ )V Sξ \ β ⊂ S.
(Cf. e.g. [ClaSch09, proof of Lemma 5].) In particular,
(4.5) J = (NSω1 )V [g,h] ∩ N.
Writing
(4.6) j = σ0,(ρ+ )V (j0 ) : p0 → σ0,(ρ+ )V ((N̄0 ; ∈, J¯0 , A0 )),
we then have that
(4.7) σ0,(ρ+ )V ((N̄0 ; ∈, J¯0 , A0 )) = (σ0,(ρ+ )V (N̄0 ); ∈, J ∩ σ0,(ρ+ )V (N̄0 ), A )
200 RALF SCHINDLER
and
(4.8) J ∩ σ0,(ρ+ )V (N̄0 ) = (NSω1 )V [g,h] ∩ σ0,(ρ+ )V (N̄0 ).
We may lift the generic iteration j of (4.6) to a generic iteration
ĵ : (V [g]; ∈, (NSω1 )V [g] , A0 → (M ; ∈, (NSω1 )V [g,h] ∩ M, A ).
Let us write
M̄ = {ĵ((Vα )V ) : α ∈ OR}
and
ĵ0 : V → M̄ .
Notice that by elementarity, if S ∈ P((ω1 )M̄ ) ∩ M̄ and M̄ |= “S is stationary,” then
S is also stationary in M and hence S is stationary in V [g, h] by (4.8).
Now there is a canonical Σ1 statement ϕ(A0 , Ḋ∗ , I˙NSω1 ) expressing the existence
of a Pmax –condition in G coded by a real in D∗ :
There is some x ∈ Ḋ∗ coding a Pmax –condition p and there is some generic
iteration of p of length ω1 +1 with iteration map j : p → (Q; ∈, J,˜ A0 ) such
that J˜ = I˙NSω1 ∩ Q.
By D∗ –Bounded Martin’s Maximum∗,++ , in order to finish off the proof of Claim
4.4 it will suffice to verify that ϕ(A0 , Ḋ∗ , I˙NSω1 ) is honestly consistent.
Let x ∈ p[T ] ∩ V [g, h] code q, so that also
(4.9) x ∈ p[ĵ0 (T )].
Let us now fix a universally Baire function F : R → R in V , let θ >> κ, say
a fixed point under ĵ, and let T and U a pair of trees on ω × 2θ witnessing the
θ–universal Baireness of F , with F = p[T ]. Let k be Col(ω, θ)–generic over V such
that g, h ∈ V [k]. Write T ∗ = ĵ(T ) and U ∗ = ĵ(U ), so that p[T ] = p[T ∗ ] and
p[U ] = p[U ∗ ] in V [k].
In V [k], there is a p[T ∗ ]–closed model A such that HωM̄2 ⊂ A, every set in
(P(ω1 ) \ NSω1 )M̄ is stationary in A, and such that
(4.10) A |= ZFC− ∧ ϕ(ĵ0 (A0 ), Ḋ∗ , İNSω1 ),
with the understanding that Ḋ∗ be interpreted by p[ĵ0 (T )] ∩ A. To get such an A in
V [k], just let A be some appropriate rank–initial segment of V [g, h], cf. (4.7), (4.8),
and (4.9); also, ĵ(A0 ) = j(A0 ) = A0 . By absoluteness between M̄ [k] and V [k],
there is then inside M̄ [k] some p[T ∗ ]–closed model A such that HωM̄2 ⊂ A, every set
in (P(ω1 ) \ NSω1 )M̄ is stationary in A, and (4.10) holds true, where again Ḋ∗ is
interpreted by p[ĵ0 (T )] ∩ A. By the homogeneity of Col(ω, θ), the existence of such
an A is forced by the trivial condition 1Col(ω,θ) over M̄ , so that by the elementarity
of ĵ0 , there is then in V [k] some p[T ]–closed model A such that HωV2 ⊂ A, every set
in (P(ω1 ) \ NSω1 )V is stationary in A, and
(4.11) A |= ZFC− ∧ ϕ(A0 , Ḋ∗ , İNSω1 ),
with the understanding that Ḋ∗ is interpreted by p[T ] ∩ A.
As F was arbitrary, we verified that ϕ(A0 , Ḋ∗ , I˙NSω1 ) is honestly consistent.
We are now going to prove (a) =⇒ (b).
Let us fix B ∈ Γ∞ and A ∈ Hω2 . Let ϕ(A, Ḃ, I˙NSω1 ) be a Σ1 formula which is
honestly consistent in the sense of Definition 2.8, with the understanding that Ḃ is
to be interpreted by (the version of) B (in the generic extension). We aim to show
that ϕ(A, Ḃ, I˙NSω1 ) holds true in V , now with the understanding that Ḃ is to be
interpreted by B.
Suppose not. We may assume without loss of generality that A ⊂ ω1 and in
fact that A is Pmax –generic over L(R) (cf. [Woo99, Theorem 4.60]). Let Ȧ be the
canonical name for A. Now say that
(4.12) p = (M, ∈, I, a) ¬ϕ(Ȧ, Ḃ, I˙NSω1 ),
where p ∈ GA = {q = (N, ∈, I , a ) ∈ Pmax : a = A ∩ ω1N ∧ there exists some generic

iterate (N ∗ , ∈, I ∗ , A) of q with I ∗ = NSω1 }. We shall derive a contradiction by
finding some q <Pmax p with q ϕ(Ȧ, Ḃ, I˙NSω1 ).
By our hypothesis, we may pick some (N, δ, τ, Σ) which u.B.–strongly captures
B. The function
(4.13) F : X → M2#,Σ (X), where X ∈ HC is self–wellordered and N ∈ L1 (X)
is then well–defined, total, and strongly universally Baire in the codes, cf. Lemmata
3.6 and 3.7.
Let θ ≥ 2ℵ1 , and let g be Col(ω, θ)–generic over V . Let A ∈ V [g] be an F –
closed witness to the fact that ϕ(A, Ḃ, I˙NSω1 ) is honestly consistent. Let X ∈ A be
transitive and such that (P(ω1 ) ∩ A) ∪ {(NSω1 )A } ∈ X. Write M = M2#,Σ (X), and
let δ0 , g0 , g1 , and Q be as in the statement of Theorem 3.14. By the conclusion of
Theorem 3.14, inside V [g] we have that
q = (M [g0 , g1 ]; ∈, NSM [g0 ,g1 ] , A)

is a Bg –iterable Pmax condition with q <Pmax p, and
q |= ϕ(A, Bg ∩ M [g0 , g1 ], (NSω1 )M [g0 ,g1 ] ),

so that q ϕ(Ȧ, Ḃ, I˙NSω1 ).
However, the assertion that there is such a q is absolute between V and V [g],
cf. Theorem 2.4. We obtained a contradiction!
Corollary 4.6. Assume that there is a proper class of Woodin cardinals.

Assume also that for every A ∈ Γ∞ there is some (N, Σ) which u.B.–strongly cap-
tures A.
Then MM∗,++ implies Γ∞ –BMM∗,++ .
5. Appendix: UBH and universally Baire iteration strategies

Let T be a normal iteration tree on V . We say that T is 2ℵ0 –closed iff for every
i < lh(T ), MTi |= “The support of EiT is 2ℵ0 –closed.” Let b be a maximal branch
through an iteration tree T . Then T is said to be continuously ill–founded off b iff
there is some sequence (αi : i < lh(T ) \ b) of ordinals such that for all i, j ∈ lh(T ) \ b
T
with i <T j, αj < πij (αi ).
The following is essentially proven in [MaSt94].
Lemma 5.1. Let T be a normal 2ℵ0 –closed iteration tree on V such that λ =
lh(T ) is a countable limit ordinal. Suppose that T has exactly one cofinal well–
founded branch, b. Then T is continuously ill–founded off b.
202 RALF SCHINDLER
Proof. If π : N → Vθ is an elementary embedding, where N is transitive and

T ∈ ran(π), then T̄ = π −1 (T ) is an iteration tree on N with the same length and
tree order as T . Let us write πi for π MT̄i , where λ.
ℵ0
Let us also assume that 2 Vθ ⊂ Vθ and that Card(N ) = 2ℵ0 . We claim that

there is ((Xin : n < ω) : i < λ) such that for each i < λ, MT̄i = {Xin : n < ω} and
for each n < ω, πn Xin ∈ MTi .
Such (Xin : n < ω) is easily constructed by recursion on i < λ. It is trivial for
i = 0. If j = T –pred(i+1) < λ, then πi Xin ∈ MTi implies πi (Xin ∩supp(EiT )) ∈
MTi+1 , as MTi |= “The support of EiT is 2ℵ0 –closed.” But then if f ∈ Xjm and
a ∈ Xin ∩ supp(EiT ),
T̄ T
π(πji+1 (f )(a)) = (πji+1 (πj Xjm )(f ))((πi Xin )(a)),
so that we may let25
m,n T̄
Xi+1 = {πji+1 (f )(a) : f ∈ Xjm ∧ a ∈ Xin ∩ supp(EiT )}.
If i < λ is a limit ordinal and (in : n < ω) is cofinal in i, then we may simply set
m,n
Xi+1 = πiT̄n i ”Xim
n
.
Let us now fix a strictly increasing sequence (λn : n < ω) which is cofinal in λ
and λn ∈ b for all n < ω. We shall assume in what follows that for all i ∈ (λ \ b)
and for all n, m < ω, Xin ⊂ Xin+1 , Xin ≺ MT̄i , and if j = max([0, i]T ∩ λm ), then
T̄
πji ”Xjn ⊂ Xin .
Let us finally also assume that ω N ⊂ N . There cannot be a cofinal branch
c∈ / b through T̄ together with some elementary σ ∗ : MT̄c → Vθ . This is because
then c ∈ N and MT̄c is well–founded, which by elementarity implies that c = b is a
well–founded branch through T .
We now let the tree U search for a cofinal branch c = b through T̄ together with
some elementary σ ∗ : MT̄c → Vθ . Formally, we let (i, σ) ∈ U iff i ∈
n(i)
/ b, σ : Xi → Vθ
T̄ T̄ −1 n(i)
is elementary, and σ ◦ π0i (x) = π(x) for all x ∈ (π0i ) ”Xi . For (j, σ ), (i, σ) ∈ U
we let (j, σ ) <U (i, σ) iff i <T j, there is some n < ω with i < λn < j, and
T̄
σ ◦ πij T̄ −1
n(i) n(j)
(x) = σ(x) for all x ∈ Xi ∩ (πij ) ”Xj .
For i ∈/ b, we now let
n(i)
αi = ||(i, πi Xi )||π0i
T (U) .
∈ MTi and πi ◦ π0i T̄ = π0i

T T
n(i)
This makes sense, as πi Xi ◦ π = π0i (π), so that
n(i) T T̄
(i, πi Xi ) ∈ π0i (U ). Moreover, if x ∈ Mi , i ≤T k = max([0, j]T ∩ λn(i) ),
T̄ T̄ n(j)
πik (x) ∈ Xkm , and m ≤ n(j), then πij (x) ∈ Xj . This implies that every infinite
branch through U would give rise to a cofinal branch c through T̄ together with
some elementary σ ∗ : MT̄c → Vθ . Therefore, αi < ∞.
T̄ n(i) T
If i <T j and n < ω is such that i < λn < j, then πj ◦ πij Xi = πij ◦ πi
n(i) T n(i) n(j) T n(i)
Xi = πij (πi Xi ), so that (j, πj Xj T (U) (i, π (πi X
) <πij ij i ), i.e.,
T
πij (αi )> αj .
We have shown that (αi : i ∈
/ b) witesses that T is continuously ill–founded off
b.
25 ·, · denotes Gödel pairing
Lemma 5.2. Let κ be such that for all A ⊂ R, A is universally Baire iff A is

κ–universally Baire. Suppose that E is a class of V –extenders with critical point
> κ whose support is 2ℵ0 –closed such that E witnesses that there is a proper class
of Woodin cardinals. Assume that UBH holds true for E. Let Ω > θ > κ, and let
σ : N → Hθ
be an elementary embedding such that N is countable and E ∩ Hθ ∈ ran(σ).
There is then an iteration strategy Σ for countable iteration trees on N which
use extenders from σ −1 (E ∩ Hθ ) and its images such that Σ is universally Baire in
the codes.
Proof. The strategy Σ is of course given by copying a given countable tree T
on N of limit length which uses extenders from σ −1 (E ∩ Hθ ) and its images onto V
via σ and pulling back the unique maximal well–founded branch.
It is easy to design a tree T searching for x, y, (αi : i < ρ) such that if x ∈ R
codes a countable iteration tree T on N of limit length lh(T ) which uses extenders
from σ −1 (E ∩ Hθ ) and its images, then
(i) y ∈ R codes a maximal branch b through T ,26 and
(ii) writing λ = sup(b) ≤ lh(T ), if σT λ is the tree on V obtained by copying
T onto V via σ, then ρ ≥ λ and (αi : i ∈ λ \ b) witnesses that σT λ is
continuously ill–founded off b.
As it is still true in V Col(ω,κ) that UBH is true for E, cf. [SchWoo∞], in the light of
Lemma 2.2 and the choice of κ it is easy to see that T witnesses that Σ is universally
Baire in the codes.
References
[AspSch14] D. Asperó and R. Schindler, Bounded Martin’s maximum with an asterisk, Notre
Dame J. Form. Log. 55 (2014), no. 3, 333–348, DOI 10.1215/00294527-2688051.
MR3263531
[BaCaLa06] J. Bagaria, N. Castells, and P. Larson, An Ω-logic primer, Set theory, Trends Math.,
Birkhäuser, Basel, 2006, pp. 1–28, DOI 10.1007/3-7643-7692-9 1. MR2267144
[ClaSch09] B. Claverie and R. Schindler, Increasing u2 by a stationary set preserving forcing, J.
Symbolic Logic 74 (2009), no. 1, 187–200, DOI 10.2178/jsl/1231082308. MR2499426
[ClaSch12] B. Claverie and R. Schindler, Woodin’s axiom (∗), bounded forcing axioms, and
precipitous ideals on ω1 , J. Symbolic Logic 77 (2012), no. 2, 475–498, DOI
10.2178/jsl/1333566633. MR2963017
[DoeSch09] P. Doebler and R. Schindler, Π2 consequences of BMM + NSω1 is precipitous and
the semiproperness of stationary set preserving forcings, Math. Res. Lett. 16 (2009),
no. 5, 797–815, DOI 10.4310/MRL.2009.v16.n5.a4. MR2576698
[FeMaWo92] Q. Feng, M. Magidor, and H. Woodin, Universally Baire sets of reals, Set theory of
the continuum (Berkeley, CA, 1989), Math. Sci. Res. Inst. Publ., vol. 26, Springer,
New York, 1992, pp. 203–242, DOI 10.1007/978-1-4613-9754-0 15. MR1233821
[FoMaSh88] M. Foreman, M. Magidor, and S. Shelah, Martin’s maximum, saturated ideals,
and nonregular ultrafilters. I, Ann. of Math. (2) 127 (1988), no. 1, 1–47, DOI
10.2307/1971415. MR924672
[KoeWoo10] P. Koellner and W. H. Woodin, Large cardinals from determinacy, Handbook of set
theory. Vols. 1, 2, 3, Springer, Dordrecht, 2010, pp. 1951–2119, DOI 10.1007/978-1-
4020-5764-9 24. MR2768702
[Lar00] P. Larson, Martin’s maximum and the Pmax axiom (∗), Ann. Pure Appl. Logic 106
(2000), no. 1-3, 135–149, DOI 10.1016/S0168-0072(00)00020-8. MR1785758
26 We may arrange that given x, y is unique such that (αi : i < ρ) as in (ii) exists.
204 RALF SCHINDLER
[Lar04] P. B. Larson, The stationary tower, University Lecture Series, vol. 32, American
MR2069032
[Lar08] P. B. Larson, Martin’s maximum and definability in H(ℵ2 ), Ann. Pure Appl. Logic
156 (2008), no. 1, 110–122, DOI 10.1016/j.apal.2008.06.012. MR2474445
[Lar10] P. B. Larson, Forcing over models of determinacy, Handbook of set theory. Vols. 1,
2, 3, Springer, Dordrecht, 2010, pp. 2121–2177, DOI 10.1007/978-1-4020-5764-9 25.
MR2768703
[Lar∞] P. Larson, AD2 , a typed writeup of some notes by J. Steel.
[MaSt89] D. A. Martin and J. R. Steel, A proof of projective determinacy, J. Amer. Math.
Soc. 2 (1989), no. 1, 71–125, DOI 10.2307/1990913. MR955605
[MaSt94] D. A. Martin and J. R. Steel, Iteration trees, J. Amer. Math. Soc. 7 (1994), no. 1,
1–73, DOI 10.2307/2152720. MR1224594
[MiSt94] W. J. Mitchell and J. R. Steel, Fine structure and iteration trees, Lecture Notes in
Logic, vol. 3, Springer-Verlag, Berlin, 1994. MR1300637
[ScTr∞] F. Schlutzenberg, and N. Trang, Scales in hybrid mice over R, available at
http://arxiv.org/pdf/1210.7258v3.pdf
[Sa15] G. Sargsyan, Hod mice and the mouse set conjecture, Mem. Amer. Math. Soc. 236
(2015), no. 1111, viii+172, DOI 10.1090/memo/1111. MR3362806
[Sch14] R. Schindler, Set theory. Exploring independence and truth, Universitext, Springer,
Cham, 2014. MR3243739
[SchWoo∞] R. Schindler and W. Hugh Woodin, The influence of cf(Γ∞ ), in preparation.
[SchSt∞] R. Schindler and J. R. Steel, The core model induction, in preparation.
[StVW82] J. R. Steel and R. Van Wesep, Two consequences of determinacy consistent with
choice, Trans. Amer. Math. Soc. 272 (1982), no. 1, 67–85, DOI 10.2307/1998951.
MR656481
[St10] J. R. Steel, An outline of inner model theory, Handbook of set theory. Vols. 1,
2, 3, Springer, Dordrecht, 2010, pp. 1595–1684, DOI 10.1007/978-1-4020-5764-9 20.
MR2768698
[St16] J.R. Steel, A theorem of Woodin on mouse sets, in Ordinal definability and recursion
theory: The Cabal Seminar, Volume III, 243–256, Cambridge Univ. Press, 2016.
[St07] J. R. Steel, A stationary-tower-free proof of the derived model theorem, Advances in
logic, Contemp. Math., vol. 425, Amer. Math. Soc., Providence, RI, 2007, pp. 1–8,
DOI 10.1090/conm/425/08113. MR2322359
[Wi15] T. M. Wilson, The envelope of a pointclass under a local determinacy
hypothesis, Ann. Pure Appl. Logic 166 (2015), no. 10, 991–1018, DOI
10.1016/j.apal.2015.05.003. MR3356616
[Woo83] W. H. Woodin, Some consistency results in ZFC using AD, Cabal seminar 79–
81, Lecture Notes in Math., vol. 1019, Springer, Berlin, 1983, pp. 172–198, DOI
10.1007/BFb0071701. MR730594
[Woo99] W. H. Woodin, The axiom of determinacy, forcing axioms, and the nonstationary
ideal, de Gruyter Series in Logic and its Applications, vol. 1, Walter de Gruyter &
Co., Berlin, 1999. MR1713438
Institut für Mathematische Logik und Grundlagenforschung, Universität Münster,

Einsteinstr. 62, Münster, FRG
E-mail address: rds@wwu.de
Volume 690, 2017
Translation procedures in descriptive inner model theory
Grigor Sargsyan
Abstract. We develop a basic translation procedure that translates a hod
mouse to an equivalent mouse. Unlike the translation procedure used by Steel
and Zhu, ours works in a coarse setting without assuming AD + . Nevertheless,
the procedure resembles the one developed by Steel. We use the translation
procedures to answer a question of Trevor Wilson. Namely, we show that if
there is a stationary class of λ such that λ is a limit of Woodin cardinals and
the derived model at λ satisfies AD + + θ0 < Θ then there is a transitive model
M such that Ord ⊆ M and M
“there is a proper class of Woodin cardinals
and a strong cardinal”. Using a theorem of Woodin on derived models it is
not hard to see that the reverse of the aforementioned theorem is also true,
thus proving that the two theories are in fact equiconsistent.
Contents
1. Introduction
2. Translatable structures
3. A question of Wilson
References
1. Introduction
In descriptive inner model theory, the role of hod mice is to capture descriptive
set theoretic strength of sets of reals inside a fine structural model. Such capturing
leads to representation of HOD of models of AD+ +V = L(P(R)) as a fine structural
model (see Theorem 4.24 of [Sar15]). It also leads to partial proofs of the Mouse
Set Conjecture (see Theorem 6.19 of [Sar15]). For more on the role of hod mice in
descriptive inner model theory see [Sar13].
While the theory of hod mice has been advanced to the level of the theory
Largest Suslin Axiom (LSA)1 the relationship between hod mice and ordinary mice
is not as well understood. Steel, in [Ste08], initiated the study of the relationship
between hod mice and ordinary mice. Yizheng Zhu, in unpublished work, building
on ideas of Steel, has shown that many models of AD+ + V = L(P(R)) are derived
models of mice. He devised a translation procedure that translates a hod mouse,
2010 Mathematics Subject Classification. Primary 03E45, Secondary 03E15, 03E60.

Key words and phrases. Mouse, inner model theory, descriptive set theory, hod mouse.
1 LSA is the conjunction of the following axioms, AD + , ¬AD , V = L(P(R)) and “the largest
R
Suslin cardinal is a member of the Solovay sequence”, see [Sara] and [Sarb].
2017
205
206 GRIGOR SARGSYAN
which is below the theory ADR + “Θ is regular”, into a mouse while making sure
that the derived models of the original hod mouse and the resulting mouse are the
same. What is being translated in these constructions is the strategy predicate
of the hod mouse in question. It is being translated into an ordinary extender
sequence.
The motivating question that has started the current project is whether such
translation can be done locally. In both Steel’s work and Zhu’s work, the transla-
tion is done using AD+ theory. Their procedure doesn’t work inside an arbitrary
hod mouse but rather a specially chosen one. In this paper, we devise a transla-
tion procedure that can be done locally. It eliminates several complications of the
two aforementioned constructions. We will then use our construction to answer a
question of Wilson (see Question 3.1).
1.1. Acknowledgments. The author would like to thank Trevor Wilson for
asking Question 3.1. He also wishes to thank the referee for a very long list of
improvements.
2. Translatable structures
We start by introducing the hybrid models that we will translate into ordinary
mice. We note once again that the general structure of our construction follows that
of Steel’s construction introduced in Section 15 of [Ste08]. The objects introduced
in this section are quite standard and their prototypes first appeared in Woodin’s
work on the analysis of HOD (see, for instance, Section 3 of [Sar14] or Section 1
of [ST14]).
Before we start, we remark that definition of the premouse in this paper is
somewhat different than in other papers in print. We will use an indexing scheme
in which extenders with critical point κ, which is a limit of Woodin cardinals and is a
cutpoint, are indexed at the successor of the least cutpoint of their own ultrapower.
More precisely, if M is a premouse and Eα ∈ E M is an extender then if crit(Eα )
is a cutpoint in M and M
“crit(Eα ) is a limit of Woodin cardinals” then in
U lt(M, E), α is the successor of ν where ν ≥ κ+ is the least cutpoint of U lt(M, E).
Suppose that V
ZF C+“there is a proper class of Woodin cardinals”. We
say that the transitive set X is self-well-ordered if there is a well-ordering of X in
Jω (X). We then let
Definition 2.1. W(X) = ∪{M : M is an Ord-iterable sound X-mouse such
that ρω (M) = X}.
We then, for α ∈ Ord, define Wα (X) by induction using the following scheme:
(1) W0 (X) = W(X).
(2) Wξ+1 = W(Wξ (X)).
(3) For limit ordinal λ, Wλ (X) = ∪ξ<λ Wξ (X).
Suppose now that P is a premouse. We then say
Definition 2.2 (Suitable Premouse). P is suitable if there is an ordinal δ such
that
(1) P
“δ is the unique Woodin cardinal”,
(2) for every strong P-cutpoint2 η, P|(η + )P = W(P|η).

P such that ξ ∈ (crit(E), α(E)) where α(E) is
2 We say ξ is P-cutpoint if there is no E ∈ E
the index of E. We say η is a strong cutpoint if for any E as above, ξ ∈ [crit(E), α(E)).
TRANSLATION PROCEDURES IN DESCRIPTIVE INNER MODEL THEORY 207
(3) P = Wω (P|δ).
Suppose now that P is a suitable premouse. We let δ P be the Woodin cardinal

of P. Given an iteration strategy Σ for P, we say Σ is fullness preserving if whenever
Q is a Σ-iterate of P via Σ then Q is suitable (see, for instance, Definition 6.2 of
[Sar14]). We say Σ has branch condensation if whenever i : P → Q is an iteration
via Σ, T is a stack according to Σ such that T ∈ dom(Σ), b is a branch of T
such that πbT exists and there is σ : MTb → Q with the property that i = σ ◦ πbT ,

then b = Σ(T ) (see Definition 2.14 of [Sar15]). Both fullness preservation and
branch condensation are important inner model theoretic notions and have been
fully explored in literature (see [Sar15]).
We say (P, Σ) is a hod pair if P is suitable and Σ is an (< Ord, < Ord)-iteration
strategy for P that is fullness preserving and has branch condensation. Notice that
all the notions introduced thus far are relative to M , and to emphasize this we will
sometimes say “M -suitable”, “M -fullness preserving” and etc. Next we introduce
M -super fullness preservation, which also comes from [Sar15] (see Definition 3.33
of [Sar15]). This notion is key in showing that the strategy can be interpreted
on to generic extensions (see, for instance, Lemma 3.35 of [Sar15]). We continue
working in M .
Definition 2.3 (Super fullness preservation). Suppose (P, Σ) is a hod pair

such that Σ has branch condensation and is fullness preserving. Let δ = δ P and for
n ∈ ω, let δn = (δ +n )P . We say Σ is η-super fullness preserving if for every n < ω,
+n P
there is a term τ ∈ P Coll(ω,(δ ) ) such that
(1) P
Coll(ω,δn ) τ ⊆ R2 ,
(2) whenever g ⊆ Coll(ω, δn ) is P-generic,
τg ⊆ {(x, y) : x ∈ R ∧ y codes a sound x-mouse projecting to ω}.
(3) whenever j : P → Q is an iteration according to Σ VηM , X ∈ VηM
is a self-well-ordered set and g ⊆ Coll(ω, j(δn )) is Q-generic such that
X ∈ HC Q[g] , letting x ∈ RQ[g] be such that x codes X and x is generic
over Jω [X],
W(X) = ∪{M : ∃y((x, y) ∈ (j(τ ))g ∧ y codes the S-translate of M to an
x-mouse}.
Recall that using S-constructions, it is possible to translate an X-mouse into

an x-mouse and vice versa. See Section 3.8 of [Sar15] for more on S-translations.
We need one more notion before we can introduce translatable structures.
Suppose Σ is super fullness preserving. We say M is generically closed under
Σ if in all generic extensions of M , Σ has super fullness preserving extension. More
precisely,
Definition 2.4. We say M is generically closed under Σ if for all M -cardinals

η and for all ≤ η-generics g there is a unique Λ ∈ M [g], denoted by Σg,η , such that
Λ is an ((η + )M , (η + )M )-strategy, Λ M is the ((η + )M , (η + )M )-fragment of Σ and
M [g]
“Λ is super fullness preserving”.
Suppose now that M is generically closed under Σ. Given an M -generic g, we

let Σg = ∪η∈Ord Σg,η .
208 GRIGOR SARGSYAN
P
Suppose next that P is suitable. Let δnP = ((δ P )+n )P . Suppose τ ∈ P Coll(ω,δn )
is a term relation for a set of reals. We then let
γτP = sup(Hull1P ({τ }) ∩ δ P ) and
HτP = Hull1P ({τ } ∪ γτP ).
Suppose now that Σ is a super fullness preserving strategy for P such that M
is generically closed under Σ. We then say
Definition 2.5. Σ guides τ correctly if for any M -generic g, Σg -iterates Q
and R of P such that the iteration embeddings i : P → Q and j : P → R exist, and
an integer n < ω, whenever x ∈ RM [g] , k ⊆ Coll(ω, i(δn )) and l ⊆ Coll(ω, j(δn ))
are such that k is Q-generic, l is R-generic and x ∈ Q[k] ∩ R[l] then
x ∈ i(τ )k ↔ x ∈ j(τ )l
We let AgΣ,τ be the set of reals in M [g] determined by Σg and τ . Notice that
for any Q as above i(τ ) term captures AgΣ,τ (see Definition 2.18 of [Sar15]).
Next we identify strategies that are guided by term relations (see Definition
4.14 of [Sar15]).
Definition 2.6. Suppose (P, Σ) is a hod pair such that Σ is super fullness
preserving and M is generically closed under Σ. We say Σ is guided by the sequence
(τi : i < ω) if for any M -generic g
P
(1) for every i there is n such that τi ∈ P Coll(ω,δn ) is a term relation for a
set of reals,
(2) for each i < ω, Σg guides τi correctly,
(3) given any Σg -iterate Q of P such that the iteration embedding π : P → Q
Q
exists, Q = ∪i<ω Hπ(τ i)
.
(4) given any Σ iterate Q of P such that the iteration embedding π : P → Q
g
exists and given any normal tree T ∈ dom(ΣgQ ), letting b = ΣgQ (T ), if

πbT exists then b is the unique branch of T such that for every i < ω,
πbT (π(τi )) locally term captures AgΣ,τi ,
(5) given any η that is a limit of Woodin cardinals of M [g] such that P ∈
M [g]
Vη , if h ⊆ Coll(ω, < η)-generic then for every i < ω, Ag∗h
Σ,τi is an OD
set of reals in the derived model of M [g] as computed by h.
We say Σ is strongly guided by the sequence (τi : i < ω) if whenever π : P → Q
comes from an iteration according to Σg and R is such that there are σ : P → R
and k : R → Q such that π = k ◦ σ then R is suitable and for every i, k−1 (π(τi ))
term captures AgΣ,τi .
We can now introduce the structures that we will translate into ordinary pre-
mice.
Definition 2.7 (Translatable structure). We say K is a translatable structure
if K is a hybrid strategy premouse (see [Sar15, Section 1.2]) satisfying the following
conditions.
(1) K
“there are proper class of Woodin cardinals and there is no measurable
limit of Woodin cardinals”.
(2) Let δ be the least Woodin cardinal of K. In K, let P = Wω (K|δ). Then P
is K-suitable, and ΣK is a K-super fullness preserving iteration strategy
for P such that K is generically closed under ΣK and there is a collection

of term relations A ⊆ P such that whenever g ⊆ Coll(ω, P) is K-generic
and (τi : i < ω) is a generic enumeration of A in K[g], (ΣK )g is strongly
guided by (τi : i < ω).
(3) Suppose κ < λ are two consecutive Woodin cardinals greater than δ. Let
ν ∈ (κ, λ) be a cardinal of K. Then for any η, K|(ν + )K , as a ΣK -premouse,
has an (η, η)-iteration strategy Λ ∈ K for stacks that are based on the
window (κ, ν) such that Λ is also a strategy for K3 .
(4) K
“there is no inner model with a superstrong cardinal”.
(5) Suppose (δi : i < ω) are the first ω many Woodin cardinals of K and
δω is their limit. Suppose g ⊆ Coll(ω, < δω ) is K-generic and M is the
derived model computed in K(R∗ )4 . Let Φ = (ΣK )g HC M .Then Φ ∈ M
and in M , K|(δ1+ω )K , as a Φ-premouse, is suitable and ∅-iterable (for the
definitions of these notions see Section 1 of [ST14])5
The following is our main theorem of this section.
Theorem 2.8. Suppose K is a translatable structure. Then, in K, there is a
proper class premouse M such that M has a proper class of Woodin cardinals and
a strong cardinal.
We spend the rest of this section describing M. Just like in Steel’s construction,
we will introduce an operator in order to compute extenders, much like the way
Steel does in [Ste08].
For the rest of this section, we fix a translatable structure K and work entirely
inside K. Let δ < λ be the first two Woodin cardinals of K. Let P = Wω (K|δ) and

N = (J E [P])K|λ . Thus, N is the output of the J E construction of K|λ done over
P using extenders with critical point > o(P) . 6
Before we go on, we record the following useful lemmas about translatable

structures. Recall from [Sar15] (see Definition 1.26 of [Sar15]) that if S is a
premouse and η < o(S) then
OηS = ∪{S||β : ρω (S||β) ≤ η and η is a cutpoint in S}.
Lemma 2.9. Suppose T ∈ K is a normal tree on K without fatal drops7 based
on the window (δ, λ) such that T has a limit length ≤ λ, T doesn’t have a branch
in K and for every limit α < lh(T ), Q(T α) MTα exists8 and is Ord-iterable
as a ΣK -mouse. Then for a stationary many α < λ, (α+ )M(T ) = (α+ )K .
Proof. Let λ∗ be the supremum of the first ω Woodin cardinals of K and let
g ⊆ Coll(ω, < λ∗ ) be K-generic. Let M be the derived model computed in K(R∗ )
3 Note that we easily get that K is internally (Ord, Ord)-iterable for stacks based on the
window (κ, ν).

4 Thus, M = J (Hom∗ , R∗ ) where R∗ = ∪ K[g∩Coll(ω,ξ) and in K(R∗ ), Hom∗ = {A ⊆
ξ<δω R
R∗ : for some ξ < δω there is a pair of δω -absoluteley complementing trees (S, T ) ∈ K[g∩Coll(ω, ξ)]
∗
such that (p[S])K(R ) = A}.
5 ∅-iterable refers to the following. Suppose P = K|(δ +ω )K and T is a tree on P such that
1
Q(T ) doesn’t exist, i.e., T has no Q-structures. Such trees are called maximal. Then ∅-iterability
says that T has a branch b such that MT b is full, has all the relevant mice projecting to δ(T ).
def sup(X ∩ Ord).
6 In this paper, we let o(X) =
7 Recall from [Sar15] (see Definition 1.27 of [Sar15]) that we say T has a fatal drop if for
MT
some α < lh(T ) and η, the rest of T after stage α is based on Oη α and is above η.
8 Recall that Q(T α) is the Q-structure of T α (see Definition 1.25 of [Sar15]).
210 GRIGOR SARGSYAN
and let Φ = Σg HC M . We let R be the Φ-mouse representation of (VΘHODΦ )M .

More precisely, M is the direct limit of pairs (Q, B) such that Q is a suitable
B-iterable Φ-premouse and B ∈ ODΦ ∩ P(R). For more on this construction see
Section 1.3 of [ST14].
It follows from clause 5 of Definition 2.7 that there is an embedding π :
K|(λ+ω )
Hull1 → Hull1R . Thus, it is enough to prove that in R, whenever T is
a normal tree on R such that lh(T ) ≤ ΘM , T has no fatal drops and T is
correctly guided9 then either T has a branch or for stationary many α < ΘM ,
(α+ )M(T ) = (α+ )R . It follows from Woodin’s computation of HOD relativized to
Φ, that HODM Φ
“R is iterable for trees that are in R” (see, for instance, Theorem
1.3 of [Tra14]).
The proof now follows easily. Fix a normal correctly guided T ∈ R on R such
that lh(T ) ≤ ΘM . Let b ∈ HODM Φ be the branch of T . We have two cases. Suppose
first that πbT [ΘM ] ⊆ ΘM . It then easily follows that Q(T ) exists and Q(T ) MTb .
But because Q(T ) ∈ R, we have that b ∈ R. It follows that we must have that
πbT [ΘM ] ⊆ ΘM . It now follows that the set of α < Θ such that (α+ )R = (α+ )M(T )
contains a club in HODM Φ . Therefore, that set is stationary in R.
Following Steel’s translation procedure, we will describe an operator S : K|λ →

K|λ10 . The domain of our operator S will consist of X ∈ K|λ such that X is a
premouse with the property that P X. To define the S-operator over one such
X, we first need to isolate a large class of universal models over X.
2.1. Universal mouse. Fix a premouse X ∈ K|λ such that P X. The

universal mice over X that we will consider act like the mice that are constructed
via fully backgrounded constructions in K|λ. Suppose η < λ is such that X ∈ K|η.

Let then QX,η be the output of J E [X]-construction of K|λ in which all extenders
used have critical points > η. we have that L[QX,η ]
“λ is a Woodin cardinal”. It
follows from clause 3 of Definition 2.7 that
Lemma 2.10. QX,η is (λ, λ)-iterable.
We let ΛX,η be the (λ, λ)-strategy induced by the background strategy.
Definition 2.11 (Universal mouse). We say Q ⊆ K|λ is a universal mouse
over X if Q is (λ, λ)-iterable, for every η ∈ (o(X), λ), OηQ W(Q|η) and for a
stationary set of ξ < λ, (ξ + )Q = ξ + .
Our first lemma on universal models shows that they compute lower parts
correctly.
Lemma 2.12. Suppose Q is a universal mouse over X. Then the function
Y → W(Y ) is definable over Q.
Proof. We first show that if η > o(X) is a Q-cardinal then OηQ = W(Q|η).
Q be the least extender such that η is a cutpoint in U lt(Q, E).
To see this, let E ∈ E
It is now enough to show that U lt(Q, E)|(η + )Ult(Q,E) = W(Q|η). Since we already
9 This notion can be found in [Sar15] on page 105. It essentially says that T chooses unique
branches identified by iterable Q-structures.

10 S stands for Steel.
have that
U lt(Q, E)|(η + )Ult(Q,E) W(Q|η),
it is enough to show that W(Q|η) (η + )Ult(Q,E) .
Fix then an M W(Q|η) such that ρω (M) = η and M U lt(Q, E). Let
π : H → K|(λ+ )K be such that letting crit(π) = ν, ν is an inaccessible cardinal in
K, π(ν) = λ, E ∈ E Q|ν and U lt(Q, E) ∈ rng(π)11 . Let Λ be the (λ, λ)-strategy of
U lt(Q, E) and let T and U be the trees of length ν + 1 on respectively U lt(Q, E)|ν
and M build using the comparison process. Let R be the last model of T and S
be the last model of U.
Because M must win the comparison with U lt(Q, E)|ν, we have that π T [ν] ⊆
ν. It follows that R computes unboundedly many successors below ν correctly,
implying that M couldn’t out iterate R, contradiction!
Next, fix Y ∈ Q. We want to see that W(Y ) is definable. Let η be a cardinal
such that Y, X ∈ Q|η.
Claim 2.13. M W(Y ) if an only if there is R OηQ such that ρ(R) = η, R
has ω many Woodin cardinals with supremum ν, M ∈ R and R
“M is ω1 -iterable
in the derived model at ν”.
Proof. We prove the forward direction. The reverse direction follows from the
homogeneity of the collapse. Let λ∗ be the supremum of the first ω many Woodin
cardinals of K and let g ⊆ Coll(ω, < λ∗ ) be K generic. Let M be the derived model
computed in K(R∗ ). It follows from clause 2 of Definition 2.7 that OηQ = Lp(Q|η)
(see [Sar15, Theorem 3.10]). We also have that M Lp(Y ).
Let Φ = Σg HC M . Because Φ ∈ M , we have that M
θ0 < Θ. It follows that
in M , there is a pair (S, Λ) such that S is ω-suitable mouse over Q|η, Λ is a fullness
preserving iteration strategy for S and every (Σ21 )M set is in the derived model of
S as computed via Λ (see, for instance, the proof of Lemma 5.13 of [Sar15]).
We then have that M ∈ S and letting ξ be the supremum of the Woodin
cardinals of S, S
“M has an ω1 -iteration strategy in the derived model at ξ”.
We can then obtain such an R by simply taking a hull of S which is η-sound.
Clearly the claim finishes the proof of the lemma.
2.2. Good universal mouse. We continue working with our X ∈ K|λ. Given
any X-premouse Q ⊆ K|λ we let κQ be the least < λ-strong of Q if exists. Suppose
η ∈ (o(X), λ).
Lemma 2.14. κQ
X,η
is a limit of Woodin cardinals in QX,η .
Proof. Let Q = QX,η . Assume then that for some η, Q = QX,η . We show
that λ is a limit of Q-cardinals ξ such that W(Q|ξ)
“ξ is a Woodin cardinal”.
The proof will in fact show that there is a club of such ξ. Suppose not and let

η be the supremum of such ξ. Let S be the output of J E construction of Q in
which all extenders used have critical point > η. Then standard arguments show
that P iterates to S (see Lemma 2.11 of [Sar15])12 . But this implies that S is not
universal contradicting clause 4 of Definition 2.7 (see Lemma 2.13 of [Sar15]).
11 This reflection into H is required because of limited iterability.
12 IfP side comes short then we can use S-constructions to show that there is ξ > η with the
property that W(Q|ξ)
“ξ is a Woodin cardinal”.
212 GRIGOR SARGSYAN
Fix now η ∈ (o(X), λ) and let Q = QX,η . Just like in the proof of the Mouse
Set Conjecture, we would like to recover a tail of (P, Σ) inside Q. The details of
this construction can be found in Section 6.3 of [Sar15]. It has also appeared in
Section 4 of [ST14]. Below we just review the construction.
Suppose g ⊆ Coll(ω, < κQ ) is a Q-generic and M is the derived model com-
puted in Q[g]. Working in M , let F = {(R, A) : R is suitable and A-iterable}. Let
MQ Q
∞ be the direct limit of F. It then follows that M∞ ∈ Q and letting ξ be the
Q
Woodin cardinal of M∞ ,
Wω (MQ Q
∞ |ξ) = M∞ .
We let
+ω MQ
RQ = MQ ∞ |(ξ ) ∞.
S
Notice that we could define R for any S that has the same properties as Q. In
particular it is defined for all iterates of Q. The following lemma can be proved
using the proof of Theorem 6.5 of [Sar15] (see the successor case).
Lemma 2.15. Let Λ be the (λ, λ)-iteration strategy for Q induced by the back-
ground strategy. Let i : Q → M be an iteration according to Λ. Then RM is a
Σ-iterate of P and whenever g is a K-generic for a poset in M,
ΣgRM M[g] ∈ J [M[g]].
The next lemma shows that all iterates of QX,η are universal.
Lemma 2.16. Suppose η ∈ (o(X), λ) and that Λ is the strategy of QX,η induced
by the (λ, λ)-strategy of K|λ. Let M be a Λ-iterate of QX,η such that the iteration
embedding i : QX,η → M exists. Then M is a universal mouse over X.
Proof. It is enough to show that for stationary many α, M computes (α+ )K
correctly (as this implies that M wins coiteration with any other mouse). Let K∗
be the iterate of K obtained by applying P-to-RM iteration to K. Let K∗∗ =
(J E,ΣRM )M . It follows from universality of background constructions (see Lemma

2.13 of [Sar15]) that K∗ and K∗∗ co-iterate to a common K∗∗∗ ⊆ K|λ via normal
trees T and U respectively.
Suppose first T has a branch b ∈ K. It follows from the universality of K∗∗
that πβT [λ] ⊆ λ. This then implies that for stationary many α, K∗∗∗ , and hence
K∗∗ , compute (α+ )K -correctly. If T doesn’t have a branch then it follows from
Lemma 2.9 that for stationary many α, K∗∗∗ and hence K∗∗ , compute (α+ )K -
correctly.
We now define the good universal mice.
Definition 2.17 (Good universal mouse). We say Q is a good universal mouse
if it has a (λ, λ)-iteration strategy Λ such that whenever i : Q → M is an iteration
according to Λ, M is universal, RM is a Σ-iterate of P and whenever g is a K-
generic for a poset in M,
ΣgRM M[g] ∈ J [M[g]].
Lemma 2.15 and Lemma 2.16 imply that there are good universal mice. Let
then Q be a good universal mouse. We then let π Q : P → RQ be the iteration
embedding according to Σ.
From now on we let A ⊆ P be the collection of all terms that Σ guides correctly.
For τ ∈ A we let πτQ = π Q HτP and set
Q+ = (Q|((κQ )+ )Q , ∈, {πτQ }τ ∈A ).
Finally we can define the S-operator.
Definition 2.18. Suppose X ∈ K|λ and Q is a good universal mouse over X.

Then
S Q (X) = Hull1Q (X).
+
In order to prove useful lemmas about the S operator, we use a standard way
of splitting the S operator into smaller pieces. For a finite B ⊆ A, we set
Q + Q Q
Q+
B = (Q|((κ ) ) , ∈, {πτ : τ ∈ B}).
Let then
Q+
SB (X) = Hull1 B (X)
Clearly if B ⊆ C ⊆ A are such that B and C are finite then there is a canonical
X
σB,C : SB (X) → SC (X). It then follows that the direct limit of (SB : B ⊆f in A)
X
under the maps σB,C is just S(X).
Our first lemma shows that the S operator is independent of the choice of Q.
Lemma 2.19. S(X) is independent of the choice of Q.
Proof. Let Q0 and Q1 be two good universal mice over X. Let π : H →

K|(λ+ )K be such that if crit(π) = ν then ν is an inaccessible cardinal in K, X ∈ H,
π(ν) = λ and Q0 , Q1 ∈ rng(π). Moreover, letting for i = 0, 1, Q̄i = π −1 (Qi ), for
each i ∈ 2 there are unboundedly in ν cardinals ξ such that
(ξ + )Q̄i = (ξ + )K (*).
Notice that we have that for i = 0, 1, Q̄i is fully iterable (see clause 3 of Defini-
tion 2.7). Let Λi be an iteration strategy for Q̄i . Comparing Q̄0 with Q̄1 using Λ0
and Λ1 respectively, we obtain (U0 , Q3 ) and (U1 , Q4 ) such that
(1) Ui is a tree on Q̄i according to Λi ,
(2) Q3 is the last model of U0 ,
(3) Q4 is the last model of U1 ,
(4) Q3 Q4 or Q4 Q3 .
It follows from the usual comparison argument and (*) that Q3 = Q4 and o(Q3 ) =
ν.
Suppose now τ ∈ A. Then
π U0 [πτQ0 ] = π U1 [πτQ1 ] =def στ .
It then easily follows that if ξ is the least < ν-strong of Q3 then letting D =
(Q3 |(ξ + )Q3 , ∈, {στ : τ ∈ A}),
S Q0 (X) = S Q1 (X) = Hull1D (X).

214 GRIGOR SARGSYAN
2.3. Basic properties of S(X). The following is an easy lemma and follows
from super fullness preservation of Σ.
Lemma 2.20. W(X) ∈ S(X)
Next we would like to show that S operator is somehow easy to compute relative
to Σ. Our first theorem towards it is that Σ moves S operator correctly. For the
duration of this section let ν be the second Woodin cardinal of N and let Λ be the
iteration strategy of N that acts on stacks that based on the window (δ, ν) (recall
that N = (J E [P])K|λ ).

Lemma 2.21. Suppose i : N → N̄ is an embedding coming from an iteration

according to Λ. Then for any B ⊆f in A,
i(SB (N |ν)) = SB (N̄ |i(ν))
Proof. The claim follows by noticing that N is a universal mouse over N |ν
and N̄ is a universal mouse over N̄ |i(ν). It also uses the following claim.
Claim 2.22. For τ ∈ A, i(πτN ) = πτN̄ .
Proof. We have that there is j : K → K∗ and σ : N̄ → j(N ) such that
j N = σ ◦ i. We then have that for every τ ∈ A, j(πτN ) = πτ . The claim now
j(N )
follows from clause 2 of Definition 2.7.
Our next lemma shows that models obtained via S-constructions from S oper-
ators are themselves S operators.
Lemma 2.23. Suppose i : N → N̄ is an embedding coming from an itera-
tion according to Λ. Suppose X is such that X is generic over Jω [N̄ |i(ν)] and
Jω [N̄ |i(ν)][X] = J [X]. Fix B ⊆f in A and let D be the mouse obtained over X
from SB (N̄ |i(ν))[X] via S-construction. Then
D = SB (X).
Proof. The proof is a straightforward chase thru definitions. The important
fact to note is that if N ∗ is the X-mouse obtained from N̄ [X] via S-construction
∗ ∗
then N ∗ is a good universal mouse over X, RN = RN̄ and π N = π N̄ .
2.4. Certified extenders and premice. Next we show that the S-operator
is captured by M#,Σ
1 operator. Recall that ν is the second Woodin cardinal of N .
Let κ = κN .
Lemma 2.24. Suppose ξ ∈ ((κ+ )N , λ) is an N -cardinal. Then S(N |ξ) ∈
M#,Σ
1 (N |ξ).
Proof. Let N ∗ = U lt(N , E) where E ∈ E N is such that U lt(N , E)

“ξ is
∗
a cutpoint”. Notice that N is a universal mouse over N |ξ. Let i : N → N̄ be
iteration of N according to Λ such that N |ξ is generic over the extender algebra of
∗
N̄ at i(ν). Let Q = N̄ |(i(ν)+ω )N̄ . Notice that N ∗ |(ξ +ω )N is generic over Jω [Q]
and
∗ ∗
Jω [Q][N ∗ |(ξ +ω )N ] = Jω [N ∗ |(ξ +ω )N ].
It follows from the proof of Lemma 2.23 that if D is the model obtained via S-
∗ ∗
construction from S(Q)[N ∗ |(ξ +ω )N ] then D = S(N ∗ |(ξ +ω )N ). Notice also that
∗ ∗
|(ξ +ω )N )
S(N |ξ) = HullS(N (N |ξ).
Finally, notice that Q ∈ M#,Σ

1 (N |ξ), S(N |(ν +ω )N ) ∈ M#,Σ
1 (N |ξ) and
S(Q) = ∪B⊆f in A i(SB (N |(ν +ω ))).
It then follows that S(N |ξ) ∈ M#,Σ

1 (N |ξ).
The next lemma shows that extender can be uniformly identified via the S-
operator.
Lemma 2.25. Suppose ξ ∈ [(κ+ )N , λ) be such that for some E ∈ E N with

crit(E) = κ, E is indexed at ξ. Let ζ be least cutpoint > κ of U lt(N , E). Then
E ∈ M#,Σ
1 (U lt(N , E)|ζ) and has a uniform definition from U lt(N , E)|ζ and Σ.
Proof. Notice that
S(Ult(N ,E)|ζ)
S(N |κ) = Hull1 (N |κ)
π (S(N |κ))
S(U lt(N , E)|ζ) = Hull1 E (U lt(N , E)|ζ)
Let then σ : S(N |κ) → S(U lt(N , E)|ζ) and k : S(U lt(N , E)|ζ) → πE (S(N |κ)) be
the uncollapse maps. We then have that πE S(N |κ) = k ◦ σ. It follows from
Lemma 2.24 that σ ∈ M#,Σ 1 (U lt(N , E)|ζ).
It follows from the construction that B ∈ Ea if and only if a ∈ σ(B). It then
also follows that E ∈ M#,Σ
1 (U lt(N , E)|ζ). Chasing thru the proof of Lemma 2.24
and the current proof, we easily see that the definition of E in M#,Σ
1 (U lt(N , E)|ζ)
is uniform from U lt(N , E)|ζ and Σ.
We now give a more precise description of the procedure that defines the ex-
tender on the sequence of N . Recall that we let κ = κN .
Definition 2.26 (Σ-certified extenders). Suppose Q is a premouse such that
N |κ Q and N |(κ+ )N = Q|(κ+ )Q . We say E ∈ M#,Σ 1 is a (Σ, Q)-certified
extender if
(1) crit(E) = κ,
Q
(2) Q πE (Q|κ),
S(Q)
(3) Hull1 (N |κ) = S(N |κ),
(4) E is the (κ, o(Q))-extender derived from k : S(N |κ) → S(Q) where k is
the uncollapse map.
We say Q is a Σ-certified premouse if N |κ Q, N |(κ+ )N = Q|(κ+ )Q and when-
Q is an extender such that crit(Eα ) = κ, Eα is (Σ, Q|α)-certifed.
ever Eα ∈ E
The next lemma is a condensation lemma.
Lemma 2.27. Suppose Q is a Σ-certified premouse, and R is such that N |κ R
and R|(κ+ )R = N |(κ+ )N . Suppose that there is an embedding i : R → Q. Then
R is Σ-certified.
216 GRIGOR SARGSYAN
Proof. Suppose α ∈ dom(E) is such that R|α is Σ-certified but EαR is not
(Σ, R|α)-certified. Let Eβ = i(Eα ). Let σ : U lt(R, EαR ) → U lt(Q, EβQ ) be the
Q
canonical factor map. Let M0 ∈ U lt(Q, EβQ ) and M1 ∈ U lt(R, EαR ) be the iterates
of Wω (N |ν) (recall that ν is the second Woodin cardinal of N ) according to Λ that
make respectively Q|β and R|α generic. It follows from branch condensation of Λ
(the strategy of N based on the window (δ, ν)) and clause 2 of Definition 2.7 that
if for i ∈ 2, πi : Wω (N |ν) → Mi are the iteration embeddings according to Λ, then
π0 = (σ M1 ) ◦ π1 . It then follows from the proof of Lemma 2.24 that σ R|α
can be lifted to σ + : S(R|α) → S(Q|β) such that letting k : S(N |κ) → S(Q|β) and
l : S(N |κ) → S(R|α) be the uncollapse maps,
k = σ + ◦ l.
But then we have that
(a, B) ∈ EαR ↔ a ∈ πEαR (B)
↔ σ(a) ∈ πE Q (B)
β
↔ σ(a) ∈ k(B)
↔ σ + (a) ∈ σ + (l(B))
↔ a ∈ l(B)

2.5. The construction. In the section we describe a construction that ex-

tends N = (J E )K|λ and converges to the model M as in Theorem 2.8. The
construction is a kind of K c construction extending N . In this construction, there
are two conditions on the background extenders. Suppose E is an extender that we
would like to use as a background extender. Suppose first that crit(E) > λ. In this
case, we require that E ∈ E K and satisfy the background conditions imposed by the

usual fully backgrounded J E -constructions. Suppose next that crit(E) < λ. In this
case, it is required that crit(E) = κN and moreover, letting Q be the stage of the
construction where it is possible to add E, it is required that E is (Σ, Q)-certified.
The following is a more precise description of the construction.
Definition 2.28. Let (Mξ , Nξ : ξ < Ω) be a sequence of premice defined as
follows:
(1) N0 = N .
is a passive premouse. Nξ+1 is obtained from Mξ as
(2) Mξ = (JαE , ∈, E)
follows:
(a) There is an extender F ∗ on the sequence of K with critical point
> λ such that F ∗ coheres the construction and for some ν < α, if
F ν = F ∗ ∩ ([ν]<ω × JαF ) then

E)
Nξ+1 = (JαE , ∈, E,
(b) The condition above fails, and there is a (Σ, Mξ )-cerified extender E
such that
E)
Nξ+1 = (JαE , ∈, E,
is a premouse.
(c) Both (a) and (b) fail and Nξ+1 = J (Mξ ).
In all the above cases, we let Mξ+1 = Cω (Nξ+1 ).
(3) Suppose λ is limit. Then E = EαMλ for some α if and only if there is
M
ξ < λ such that for all β ∈ (ξ, λ), E = Eα ξ .
The construction halts at stage ξ ∈ Ord if either Cω (Nξ ) is not defined or else
ρω (Mξ ) < λ. We say the construction converges if doesn’t halt at any ξ.
Our next goal is to show that the construction converges. A similar argument
then will show that κN is a strong cardinal in the final model. Proofs of both of
these facts are very similar to the proofs given by Steel in [Ste08] (see Section 13
and 15 of that paper). Our first lemma shows that certain comparisons do not
encounter disagreements with extenders that have critical point κN .
Lemma 2.29. Suppose λ < ξ < θ are two ordinals. Suppose further that π :
H → K|θ is such that letting crit(π) = ν, the following holds:
(1) (Mα , Nα : α < ξ) ∈ rng(π) and Nξ ∈ rng(π).
(2) ν ∈ (κN , λ) is an inaccessible cardinal in K, and π(ν) = λ.
(3) For every α ≤ ξ, Nα
“λ is a Woodin cardinal”.
Let N̄ = π −1 (N ), (Sα , Qα : α < ζ) = π −1 ((Mα , Nα : α < ξ)) and Qζ = π −1 (Nξ ).
Let S N be the least such that S
“ν is a Woodin cardinal” and J1 (S)
“ν
is not a Woodin cardinal”. Then the comparison of the construction of H that
produces (Sα , Qα : α < ζ) and Qζ with S doesn’t encounter disagreements involving
extenders with critical point κN .
Proof. Suppose not. Let T and U be the iteration trees on H and S respec-
tively coming from the comparison process that expose a disagreement involving
extenders with critical point κN . Let H ∗ and S ∗ be the last models of T and U
respectively. Notice that both T and U are above ν.
Let Q be the model appearing in the construction of H ∗ before stage π T (ζ) such
that the disagreement between the construction of H ∗ and S ∗ is between Q and
S ∗ . Suppose first that there is an extender with critical point κN on the sequence
of Q that causes a disagreement. Let α be such that EαQ is this extender. Thus we
have that Q|α = S ∗ |α.
∗
Suppose EαS = ∅. It then follows from our construction, Lemma 2.27 and
∗
Lemma 2.25 that both EαS and EαQ are (Σ, Q|α)-certified. Clause 4 of Defini-
∗
tion 2.26 implies that EαS = EαQ .
∗
Suppose next that EαS = ∅. We then have two cases. Suppose first for some
∗ ∗
β > α, EβS = ∅ and crit(EβS ) = κ. Fix minimal such β. Then α is a cutpoint
∗
cardinal of U lt(S ∗ , EβS ) and (κ, α)-extender derived from πE S ∗ is (Σ, Q|α)-certified.
β
Then again Clause 4 of Definition 2.26 and initial segment condition imply that
∗
EαS = EαQ .
Next suppose that α is a cutpoint of S ∗ . Because EαQ is (Σ, Q|α)-certified, we
have that W(Q|α) U lt(Q, EαQ ). It then follows that S ∗ U lt(Q, EαQ ) implying
that Q
“ν is not a Woodin cardinal” of Q, contradiction.
Next, suppose that H ∗ side is not active. Let α be least such that S ∗ |α = Q|α,
S∗ Q and crit(EαS ∗ ) = κN . Arguing as above, we see that α must be a
Eα = E
∗
cutpoint in Q. It then follows that Q U lt(S ∗ , EαS ), implying that Q S ∗ . This
contradicts our assumption that Q participates in a disagreement.
We now prove that
Lemma 2.30. For each α, Mα is defined.
218 GRIGOR SARGSYAN
Proof. Suppose not. Let α be least such that Mα is not defined.

Claim 2.31. Nα
“λ is a Woodin cardinal”.
Proof. Suppose not. Let ξ be least such that Nξ
“λ is a Woodin cardinal”
but J1 (Nξ )
“λ isn’t a Woodin cardinal”. Fix θ such that the construction up to
stage ξ can be done inside K|θ. Let π : H → K|θ be such that letting crit(π) = ν,
the following holds:
“ν is
not a Woodin cardinal”. It follows from Lemma 2.29 that Qζ = S.
Since ν is an inaccessible cardinal of K, it follows that ρω (S) = ν. Since
J1 (Nξ )
“λ is not a Woodin cardinal”, we must have that H
“ν is not a Woodin
cardinal”, contradiction!
Notice now that α must be a successor ordinal. Fix θ such that the construction
up to stage α can be done inside K|θ. Let π : H → K|θ be such that letting
crit(π) = ν, the following holds:
(1) (Mβ , Nβ : β < α) ∈ rng(π) and Nα ∈ rng(π).
Let N̄ = π −1 (N ), (Sβ , Qβ : β < ζ) = π −1 ((Mβ , Nβ : β < α)) and Qζ = π −1 (Nα ).
“ν is
not a Woodin cardinal”.
It follows from the Claim and Lemma 2.29 that the comparison process that
compares the construction of H producing (Sβ , Qβ : β < ζ) and Qζ with S must
halt without ever using extenders with critical point κN . Since H-side must lose,
we have that Cω (Qζ ) must be defined, contradiction.
Finally we show that κN is a strong cardinal in the model produced via the con-
struction of Definition 2.28. The proof is very much like the proof of Lemma 2.30.
Lemma 2.32. Suppose M is the model to which the construction of Defini-
tion 2.28 converges. Then M
“κN is a strong cardinal”.
Proof. Suppose not. Let η > λ be a cutpoint of M. Let ξ be such that
Nξ |(η + )M = M|(η + )M . Fix θ such that the construction up to stage ξ can be
done inside K|θ. Let π : H → K|θ be such that letting crit(π) = ν, the following
holds:
(3) η ∈ rng(π).
“ν is
not a Woodin cardinal”.
We now compare the construction of H producing the sequence (Sα , Qα : α <
ζ) and Qζ with S. Let i : H → H ∗ and j : S → S ∗ be result of this comparison. It
follows from Lemma 2.29 that crit(i), crit(j) > ν. We must have that i(Qζ ) S ∗∗ .
Let γ = i(π −1 (η)).
Suppose next that γ is a cutpoint of S ∗ . It then follows that S ∗ W(i(Qζ )|γ).

Since M#,Σ1 (i(Qζ )) ∈ H ∗ , we have that W(i(Qζ )|γ) ∈ H ∗ . It then follows that
S ∈ H implying that H ∗
“ν is not a Woodin cardinal”, contradiction.
∗ ∗
We must then have that γ is not a cutpoint in S ∗ . Let α be least such that
∗
E =def EαS is the first extender overlapping γ in S ∗ . It follows from Lemma 2.25
that E is (Σ, i(Qζ )|γ)-certified. Again, since M#,Σ 1 (i(Qζ )) ∈ H ∗ , H ∗
“E is
(Σ, i(Qζ )|γ)-certified”. Hence, E should have been on the sequence of i(Qζ ), con-
tradiction.
3. A question of Wilson
In a private conversation Trevor Wilson asked the question.
Question 3.1 (Wilson). Assume there are proper class of Woodin cardinals.
Suppose the class
S = {λ : λ is a limit of Woodin cardinals and the derived model at λ satisfies
AD+ + θ0 < Θ}
is stationary. Is there a transitive model M satisfying ZF C such that Ord ⊆ M
and M has a proper class of Woodin cardinals and a strong cardinal?
We show that the answer is yes.
Theorem 3.2. Suppose there are proper class of Woodin cardinals. Assume
further that the class S above is stationary. Then there is a transitive model M
satisfying ZF C such that Ord ⊆ M and M has a proper class of Woodin cardinals
and a strong cardinal.
Proof. Given λ ∈ S let f (λ) be the least cardinal η < λ such that if g ⊆
Coll(ω, η) then the universal Π21 set of the derived model at λ is witnessed to be
< λ-uB in V [g]. Because S is stationary we have that for some stationary class
S ∗ ⊆ S, f S ∗ is a constant function. We assume, without a loss of generality,
that S ∗ = S. Let g ⊆ Coll(ω, η) be V -generic.
Suppose first that there is λ ∈ S − η such that the derived model at λ satisfies
θ1 < θ. Then it follows from unpublished result of Steel (but see the proof of
Theorem 15.1 of [Ste08]) that there is a countable active mouse M such that if κ
is the critical point of the last extender of M then M
“κ is a limit of Woodin
cardinals and there is a < κ-strong cardinal”. Iterating the last extender of M
through the ordinals we obtain a class size model as desired.
Assume then that for all λ ∈ S − η the derived model at λ satisfies AD+ + θ1 =
Θ. Similarly, using the results of [Sar15], we can conclude that for every λ ∈ S − η
the derived model at λ satisfies MC (mouse capturing). Otherwise the derived
model at such a λ would have an initial segment satisfying ZF + AD+ + θ2 = Θ
and the aforementioned result of Steel would still imply that there is a mouse M
as above. Thus, we assume that
(1) for all λ ∈ S − η the derived model at λ satisfies AD+ +
M C + θ1 = Θ.
Fix λ ∈ S − η. Let h ⊆ Coll(ω, < λ) be V [g]-generic. Let M be the derived
model at λ as computed by h ∩ Coll(ω, < λ). Let (P, Σ) ∈ M be a hod pair such
220 GRIGOR SARGSYAN
that
M
“λP = 0 and Σ is a super fullness preserving iteration strategy with branch
condensation that is strongly guided by A”
where A is some semi-scale on the universal Π21 set of M consisting of ODM

prewellorderings.
∗
We have that in V [g ∗ h], Σ ∈ (Hom∗ )V [R ] . Let again f (λ) < λ be the least
cardinal < λ such that it is forced by Coll(ω, < λ) that a < λ-uB code of Σ appears
after collapsing f (λ). Again we obtain that f is constant on a tail of S. Let ν
be this constant value and let k ⊆ Coll(ω, ν) be V [g]-generic. Then in V [g][k], we
have a pair (P, Σ) such that
(2) λP = 0, Σ is universally Baire iteration strategy for P as
witnessed by some class size trees T, U with p[T ] = Σ such that
for every λ ∈ S − ν, if h ⊆ Coll(ω, < λ) is V [g][k]-generic, M is
∗
the derived model at λ computed by h and Λ = (p[T ])V (R ) then
M
“Λ is a super fullness preserving iteration strategy with
branch condensation that is strongly guided by A, which is a
semi-scale on the universal Π21 -set consisting of OD prewellorder-
ings”.
Given a V [g][k]-generic h we let Σh = (p[T ])V [g][k][h] . We now work in W =
V [g][k]. We will abuse the notation and use Σ instead of Σg∗k .
Suppose now that D ⊆ Ord and f : D → Ord is an increasing function such
that for α ∈ D, f [α] ⊆ α, if α is not the maximum element of D then f (α) ∈ D
and f (α) is a strong limit cardinal in W . We then let Qf be the output of the

fully backgrounded J E,Σ -construction done as follows. Let (ξα : α ∈ γ) be an
enumeration of D in increasing order. We also assume that f (ξα+1 ) = ξα+2 . Here
we allow γ = Ord. We then have the following conditions.

(1) Qf |f (ξ0 ) is the output of J E,Σ construction done inside Wf (ξ0 ) using
extenders with critical point > ξ0 .

(2) Qf |f (ξα+2 ) is the output of the J E,Σ construction done inside Wf (ξα+2 )
over Qf |f (ξα ) using extenders with critical point > f (ξα+1 ) = ξα+2 .
(3) If λ ∈ D is a limit ordinal then Qf |λ = ∪α<λ Qf |ξα .
(4) If for any α < β, some level of Qf |ξβ projects across ξα then we stop the
construction.
Otherwise we let Qf be the final model of the construction.
Notice that if there is a class size inner model with a Woodin cardinal that is
a limit of Woodin cardinals then we are done proving our theorem. Otherwise, it
follows from the results of [Nee02], that for every f , if Ord ⊆ Qf then it is because
for some α < β, some level of Qf |ξβ projects across ξα .
Given a transitive set X, let W Σ (X) be the union of all sound Σ-mice over
X that project to X and are fully iterable. Also, let LpΣ,+ (X) be the union of
all sound Σ-mice over X that project to X and whose countable submodels are
fully iterable in all generic extensions. Recall that LpΣ (X) is the union of all sound
Σ-mice over X that project to X and whose countable submodels are ω1 +1-iterable.
We now prove a claim that shows that the LpΣ operator stabilizes at various
derived model.
Claim 3.3. There is a stationary class S0 such that

(1) if λ ∈ S0 then λ is a limit of Woodin cardinals,
(2) whenever h ⊆ Coll(ω, < η) is W -generic and M is the derived model
computed in W (R∗ ) then letting Φ = Σh HC M , for any X ∈ HC M and
M LpΦ (X) such that ρω (M) = o(X), W [h]
“M has a universally
Baire iteration strategy”.
Proof. Suppose not. Then the set of λ ∈ S for which clause 2 above fails
is stationary. Given such a λ, we let f (λ) < λ be least such that whenever h ⊆
Coll(ω, f (λ)) is W -generic, there is a set of reals A ∈ W [h] such that A is < λ-
universally Baire but not universally Baire. We then have that for a stationary
class E, f (λ) is constant. Let η be this constant value and let h ⊆ Coll(ω, < η)
be W -generic. Then for each λ ∈ E we have a set of reals Aλ ∈ W [h] such that in
W [h], Aλ is < λ-uB but not uB. It follows that for a stationary class F ⊆ E and
for some set of reals A ∈ W [h], A = Aλ for all λ ∈ F .
Notice that if λ ∈ S0 then LpΣ,+ operator on sets in Wλ is just the LpΣ -operator
of the derived model. Our next claim connects ∅-iterability with background con-
structions.
Claim 3.4. Suppose λ ∈ S0 and h ⊆ Coll(ω, < λ) is W -generic. Let M be

the derived model computed in W (R∗ ) and let Φ = Σh HC M . Let Q be a Φ-
suitable premouse that is ∅-iterable. Let η < λ be such that Q ∈ Wη [h ∩ Coll(ω, η)].
Let ν ∈ (η, λ) be the least such that LpΣ,+ (Wν )
“ν is a Woodin cardinal”. Let
A ∈ P(ν) ∩ J (Wν ) code Wν and let R− = (J E,Σ )Wν in which extenders used have

critical point > η and cohere A. Let R = Lpω (R− ). Then R is a correct iterate
Σ,+
of Q, and hence, in M , R is ∅-iterable.
Proof. It follows from clause 2 of Claim 3.3 and the universality of background
constructions (see Lemma 2.11 of [Sar15]) that Q indeed iterates to R− in the sense
there is a correctly guided tree T on Q such that R− = M(T ).
It is then enough to show that R
“ν is a Woodin cardinal”. Suppose not. We
have that A and hence Wν is generic for the extender algebra of R− at ν. It then
follows that LpΣ,+ (Wν ), using S-constructions (see Section 3.8 of [Sar15]), can be
translated into Σ-mouse over R− . It therefore follows that LpΣ,+ (R− )
“ν is a
Woodin cardinal”.
Notice that a standard application of Fodor’s lemma gives as a stationary class

S1 and ordinals η ∗ < ν ∗ such that fixing any set A ∈ P(ν ∗ ) that codes Wν ∗ and

letting R− = (J E,Σ )Wν ∗ in which extenders used have critical point > η ∗ and
−
ω (R ) is ∅-iterable in the derived model at
cohere A, then for every λ ∈ S1 , LpΣ,+
−
λ. Let then P0 = Lpω (R ).
Σ,+
Let now k0 : S1 → Ord be defined by letting k0 (α) be the least W -cardinal

> α such that LpΣ,+ (Wk0 (α) )
“k0 (α) is a Woodin cardinal” and for any X ∈
Wα , letting A ∈ P(k0 (α)) ∩ J (Wk0 (α) ) code Wk0 (α) and R− be the output of

J E,Σ construction of Wk0 (α) done over X using extenders with critical point > α,
−
ω (R ) is an ∅-iterable Σ-suitable premouse in the derived model at any λ ∈
LpΣ,+
S1 − α (here we use the relativised version of Claim 3.4).
222 GRIGOR SARGSYAN
Suppose now that D = (ξα : α < γ) ⊆ S1 is such that for every α < γ,
k0 [ξα ] ⊆ ξα and let fD : D ∪ rng(k0 D) → Ord be given by

k0 (ξ) : ξ ∈ D
fD (ξ) =
η : ξ ∈ D ∧ η = min(D − ξ)
Claim 3.5. Suppose λ ∈ S1 and R ∈ Wλ are such that for some ζ, R is a

Σ-suitable premouse in the derived model at λ and letting δ be the largest Woodin
cardinal of R, δ is a W -cardinal. Then there is η ≥ λ such that for any D ⊆ Ord−η
such that fD is defined, every level of QfD construction build over R preserves the
Woodiness of δ.
Proof. Suppose not. We can then produce class many backgrounded con-
structions (Qα : α ∈ Ord) and strictly increasing ordinals (ηα : α ∈ Ord) such
that for every α, each Qα is done over R and size δ hulls of every model ap-
pearing in the construction producing Qα is ηα -iterable. It then follows that
for some α, for all β ≥ α, Qβ |(δ + )Qβ = Qα |(δ + )Qα . Fix such α, we get that
Qα |(δ + )Qα LpΣ,+ (R|δ), contradiction.
Given λ ∈ S1 , let ηλ be the least ordinal that witnesses the validity of the claim
for any R ∈ Wλ . We now define a set D = (λα : α ∈ Ord) by induction as follows.
(1) λ0 = min(S1 − o(P0 )),
(2) letting λ∗α+1 = min(S1 − k0 (λα )), λα+1 = min(S1 − ηλ∗α+1 ),
(3) for limit ordinals ξ, λξ = supα<ξ λα
It now follows from Claim 3.5 that for any even α, k0 (λα ) is a Woodin cardinal
in QfD which is done over P0 . This can be proven by a simple induction using
Claim 3.5.
Claim 3.6. QfD is a translatable structure.
Proof. Notice that it follows from our construction that for any QfD -cutpoint
ξ > o(P0 ), if η is the least Woodin cardinal of QfD above ξ and λ = min(S1 − ξ)
then QfD |(η +ω )QfD is an ∅-iterable Σ-suitable premouse in the derived model at
λ. This observation takes care of clause 3 of Definition 2.7. Clause 4 is satisfies as
otherwise we already have the model that we are trying to build. The rest of the
clause are consequence of our choice of Σ and can be easily shown by using generic
interpretability and generic comparisons (see Theorem 3.10 and Lemma 3.35 of
[Sar15])
Applying Theorem 2.8 we obtain a model as in the statement of Theorem 3.1.

Our final corollary shows that the two theories considered in Theorem 3.1 are
equiconsistent.
Corollary 3.7. The follows theories are equiconsistent.
(1) (T1 :) ZF C + “there is a stationary class of λ such that λ is a limit of
Woodin cardinals and the derived model at λ satisfies θ0 < Θ”
(2) (T2 :) ZF C + “there is a proper class of Woodin cardinals and a strong
cardinal”
Proof. It follows from Theorem 3.1 that Con(T1 ) → Con(T2 ). The reverse
direction is a theorem of Woodin that can be found in [Ste09]. Woodin showed
that under the hypothesis of clause 2, if λ is a limit of Woodin cardinals bigger
than a strong cardinal then the derived model at λ satisfies θ0 < Θ.
References
[Nee02] Itay Neeman, Inner models in the region of a Woodin limit of Woodin cardinals, Ann.
Pure Appl. Logic 116 (2002), no. 1-3, 67–155, DOI 10.1016/S0168-0072(01)00103-8.
MR1900902
[Sara] Grigor Sargsyan, The analysis of HOD below the theory AD + +“the largest Suslin
cardinal is a member of the Solovay sequence”, available at math.rutgers.edu/∼gs481.
[Sarb] Grigor Sargsyan, The comparison theory of hod pairs below AD + +“the largest Suslin
cardinal is a member of the Solovay sequence”, available at math.rutgers.edu/∼gs481.
[Sar13] Grigor Sargsyan, Descriptive inner model theory, Bull. Symbolic Logic 19 (2013), no. 1,
1–55, DOI 10.2178/bsl.1901010. MR3087400
[Sar14] Grigor Sargsyan, Nontame mouse from the failure of square at a singular strong limit
cardinal, J. Math. Log. 14 (2014), no. 1, 1450003, 47, DOI 10.1142/S0219061314500032.
MR3225589
[Sar15] Grigor Sargsyan, Hod mice and the mouse set conjecture, Mem. Amer. Math. Soc. 236
(2015), no. 1111, viii+172, DOI 10.1090/memo/1111. MR3362806
[ST14] Grigor Sargsyan and Nam Trang, Non-tame mice from tame failures of the unique branch
hypothesis, Canad. J. Math. 66 (2014), no. 4, 903–923, DOI 10.4153/CJM-2013-036-x.
MR3224270
[Ste08] John R. Steel, Derived models associated to mice, Computational prospects of infin-
ity. Part I. Tutorials, Lect. Notes Ser. Inst. Math. Sci. Natl. Univ. Singap., vol. 14,
World Sci. Publ., Hackensack, NJ, 2008, pp. 105–193, DOI 10.1142/9789812794055 0003.
MR2449479
[Ste09] John R. Steel, The derived model theorem, Logic Colloquium 2006, Lecture Notes
Log., Association for Symbolic Logic, Chicago, IL, 2009, pp. 280–327, DOI
10.1017/CBO9780511605321.014. MR2562557
[Tra14] Nam Trang, HOD in natural models of AD+ , Ann. Pure Appl. Logic 165 (2014), no. 10,
1533–1556, DOI 10.1016/j.apal.2014.04.006. MR3226054
Department of Mathematics, Rutgers University, 100 Frelinghuysen Road, Piscat-

away, New Jersey 08854
URL: http://math.rutgers.edu/∼gs481
E-mail address: grigor@math.rutgers.edu
Volume 690, 2017
Implications of very large cardinals
Scott Cramer
Abstract. We trace some of the history of results which follow from the
existence of very large cardinals. In particular we concentrate on the axiom I0
and its implications for the structure of the inner model L(Vλ+1 ).
Contents
1. Introduction
2. Kunen’s Theorem and Rank-into-Rank Embeddings
3. Reflection properties
4. Structure of L(Vλ+1 )
5. AD-like axioms for L(Vλ+1 )
6. Beyond I0
References
1. Introduction
The large cardinal hierarchy is a bedrock of modern set theory as the key mea-
sure against which seemingly every other axiom in mathematics can be compared in
terms of consistency strength.1 In this article we explore some of the large cardinals
axioms which exist at the edge of this hierarchy: what we call ‘very large cardinals’.
Our goals in this study are both to identify consequences of the existence of very
large cardinals and to understand how these consequences help us understand the
structure and extent of the large cardinal hierarchy itself.
The central focus of our study is the axiom I0 and the structure L(Vλ+1 ). This
structure was first defined and studied by H. Woodin in the 1980’s in unpublished
work in order to more fully understand how the structure of L(R) is influenced
by large cardinals. Surprisingly, similarities between the structure of L(Vλ+1 ) and
the structure of L(R) abound under large cardinal assumptions, and much of this
article will be devoted to recounting what is presently understood about similarities
and differences in their structure.
Beyond understanding the structure of very large cardinals by themselves, re-
cent work has also returned to Woodin’s original motivation and found increasingly
2010 Mathematics Subject Classification. Primary 03E55; Secondary 03E60.

Key words and phrases. Axiom I0 , determinacy, square roots of embeddings.
1 See [Kan94] for a general history of large cardinals.
©2017 American Mathematical Society
225
226 SCOTT CRAMER
deep structural connections between L(Vλ+1 ) and models of determinacy. These

results give a new method for obtaining strong models of determinacy axioms,
and could potentially help motivate and analyze very strong determinacy axioms.
They also give a picture of the interaction between large cardinals and determinacy
axioms on a global scale.
We hope to paint a picture of intricate structure at the level of L(Vλ+1 ) which
gives some evidence of the consistency of the axiom I0 . Optimistically, the analysis
of L(Vλ+1 ) extends well beyond this level, and hopefully, if our analysis becomes
detailed enough, where this structure breaks down and inconsistency occurs will
become more clear. At this point, however, this occurrence is a hope rather than a
reality.
The outline of this article is as follows. In Section 2 we discuss Kunen’s In-
consistency Theorem and the various large cardinals which exist just below this
barrier. In Section 3 we look at reflection properties of these cardinals. In Section
4 we look more closely at the axiom I0 and the structural aspects of L(Vλ+1 ) under
I0 . In Section 5 we consider the notion of an AD-like axiom for L(Vλ+1 ) and various
attempts at defining such an axiom. Finally in Section 6 we briefly consider axioms
beyond I0 .
Our proofs are throughout rather sketchy as we focus on key points while trying
to avoid a tidal wave of calculations. Hopefully our brevity will be appreciated by
the reader, and, moreover, the exercise of filling in the appropriate details, in some
cases, would perhaps serve as a fruitful introduction to the subject.
1.1. Set theory conventions. We denote by Vα for α an ordinal the strat-

ification of V according to rank. So V0 = ∅, Vα+1 = P (Vα ) and Vλ = α<λ Vα
for λ a limit. For a transitive set A, L(A) is the constructible hierarchy built on
top of A. So L0 (A) = A, Lα+1 (A) = Def(Lα (A)) and Lλ (A) = α<λ Lα (A) for λ
a limit. By L(X, A) we mean the constructible hierarchy built on top of A, with
X as a predicate.
So L0 (X, A) = A, Lα+1 (X, A) = Def(Lα (A), X ∩ Lα (A)) and
Lλ (X, A) = α<λ Lα (X, A) for λ a limit.
Suppose that M and N are models of a fragment of set theory. Then j : M → N
is an elementary embedding if for all φ[x1 , . . . , xn ] and a1 , . . . , an ∈ M we have that
M |= φ(a1 , . . . , an ) ⇒ N |= φ(j(a1 ), . . . , j(an )).
We use the convention that all elementary embeddings are nontrivial, so j = id.
And crit (j) is the least ordinal α such that j(α) > α.
2. Kunen’s Theorem and Rank-into-Rank Embeddings

By far the most common type of large cardinal axiom is the assertion that
there is a non-trivial elementary embedding j of the universe V into an inner
model M . In general, the more M agrees with V the stronger the large cardinal.
This observation led W. Reinhardt in 1967 to propose the following axiom, called
a Reinhardt cardinal: there exists a non-trivial elementary embedding j : V →
V . The existence of a Reinhardt cardinal is a natural upper bound of this type
of elementary embedding existence axiom, but K. Kunen [Kun71] subsequently
showed that Reinhardt cardinals are inconsistent with the Axiom of Choice.
IMPLICATIONS OF VERY LARGE CARDINALS 227
Theorem 1 (Kunen[Kun71]). (KM)2 Suppose that j : V → M is a non-trivial

elementary embedding. Then M = V .
In fact he showed the following stronger fact.
Theorem 2 (Kunen). (ZFC) There is no λ such that there is a non-trivial
elementary embedding
j : Vλ+2 → Vλ+2 .
This theorem follows immediately from the proof of Theorem 23 below and
Solovay’s Theorem on splitting stationary sets under ZFC. Several additional proofs
can be found in [Kan94].
A key point, and the starting point for the analysis of axioms just below this
inconsistency, is that all known proofs of Kunen’s Theorem require the Axiom of
Choice (AC). So, on the one hand, there is the natural question of whether there
can be a (non-trivial) elementary embedding Vλ+2 → Vλ+2 under ZF, and on the
other hand, in analyzing this possibility, we must necessarily consider models in
which AC fails. This is perhaps the first indication that such an analysis would
have similarities to other non-choice settings, such as the context of the Axiom of
Determinacy (AD).
We now define some of the large cardinals which lie just below this inconsis-
tency. These elementary embeddings are collectively referred to as ‘rank-into-rank
embeddings’. Note that from now on all elementary embeddings are assumed to be
non-trivial.
Definition 3. Let λ be an ordinal. We make the following definitions.
(1) I3 holds at λ if there exists an elementary embedding3 j : Vλ → Vλ and λ
is a limit.
(2) I2 holds at λ if there exists an elementary embedding j : V → M such
that Vλ ⊆ M and crit (j) < λ.
(3) I1 holds at λ if there exists an elementary embedding j : Vλ+1 → Vλ+1 .
(4) I1 (Σ1 ) holds at λ if there exists a Σ1 -elementary embedding j : Vλ+1 →
Vλ+1 .
(5) I1 (iterable) holds at λ if there exists an iterable elementary embedding
j : Vλ+1 → Vλ+1 .
W. H. Woodin defined an axiom called I0 which is even stronger than I1 but still
below the level of Reinhardt cardinals. Before introducing this axiom, we mention
as motivation that the existence of an elementary embedding Vλ+2 → Vλ+2 can be
weakened by restricting to certain sets in Vλ+2 = P (Vλ+1 ). This restriction succeeds
if we avoid the subsets of Vλ+1 which the axiom of choice gives us in the proof of
Theorem 2. This situation is very reminiscent of considering regularity properties,
such as Lebesgue measurability, of sets of real numbers in classical descriptive set
theory. In that situation the axiom of choice implies that there are sets of reals
which do not satisfy, for instance, Lebesgue measurability. However, by restricting
2 Note that Kunen’s theorem in this form is a statement in Kelley-Morse set theory. See
[HKP12] for a discussion of this fact.

3 Note that by our convention j must be nontrivial, and hence crit (j) < λ is the only possi-
bility, so we do not need to specify that this is true of the critical point. On the other hand this
is not immediate for I2 or I0 , so we must specify that crit (j) < λ in those cases.
228 SCOTT CRAMER
λ - - - λ
···
..
j(j) .

1
j
j(κ1 ) = κ2 κ2
1 -

j(κ0 ) = κ1 κ1
1 -

κ0 κ0
-
Figure 1. Rank-into-rank embeddings.
to only certain sets of reals (such as Borel sets or projective sets) we have a chance
of showing that all such sets have the desired regularity properties. We will see in
Section 4 that this analogy holds to a remarkable extent, and that we can actually
achieve remarkably similar descriptive set-theoretic results in this context.
We now define I0 and some variations.
Definition 4. 4 Let λ and α be ordinals. We make the following definitions.
(1) I0 (α) holds at λ if there exists an elementary embedding j : Lα (Vλ+1 ) →
Lα (Vλ+1 ) such that crit (j) < λ and λ is a limit.
(2) I0 (< α) holds at λ if for all β < α, I0 (β) holds at λ.
(3) I0 holds at λ if there exists an elementary embedding j : L(Vλ+1 ) →
L(Vλ+1 ) such that crit (j) < λ.
(4) I0# holds at λ if there exists an elementary embedding5
# #
j : L(Vλ+1 , Vλ+1 ) → L(Vλ+1 , Vλ+1 )
such that crit (j) < λ.
(5) I0# (α) holds at λ if there exists an elementary embedding
# #
j : Lα (Vλ+1 , Vλ+1 ) → Lα (Vλ+1 , Vλ+1 )
(6) Suppose X ⊆ Vλ+1 . We say that X is Icarus or I0 (X) holds if there
exists a non-trivial elementary embedding j : L(X, Vλ+1 ) → L(X, Vλ+1 )
We now collect some basic facts about rank-into-rank embeddings and the
structure L(Vλ+1 ).
Proposition 5. Let λ be an ordinal. Then the following hold.
4 Ournotation above in (1) and (2) is not standard in the literature, although we shall use it
for convenience in this paper. The terminology in (5) first appeared in [Dim11].
5 See [Sol78] for the definition of R# and [Dim11] for the corresponding definition of V # ,
λ+1
which is very similar.
(1) (λ+ )V = (λ+ )L(Vλ+1 ) , so in L(Vλ+1 ), λ+ is regular.

(2) Suppose that j : Vλ → Vλ is an elementary embedding. For κi | i < ω
defined by6 induction by κ0 = crit (j) and κi+1 = j(κi ) for i < ω, we have
that
lim κi = λ
i<ω

where either λ = λ or λ + 1 = λ.
(3) If I3 holds at λ then cof(λ) = ω.
(4) Suppose I0 holds at λ. Then L(Vλ+1 ) does not satisfy the axiom of choice.
(5) Suppose that j : Vλ → Vλ is an elementary embedding. Then j has a
unique extension to a Σ0 -elementary embedding
ĵ : Vλ+1 → Vλ+1 .
(6) If n < ω is odd and j : Vλ+1 → Vλ+1 is Σn -elementary, then j is Σn+1 -
elementary (Martin).
Proof. To see (1), note that for any ordinal α < λ+ , there is an x ∈ Vλ+1
which codes an ordering of λ in ordertype α. And hence (λ+ )V = (λ+ )L(Vλ+1 ) since
Vλ+1 ⊆ L(Vλ+1 ).
To see (2), let limi<ω κi = λ , and assume towards a contradiction that λ + 2 ≤
λ. Note that j(λ ) = λ . And hence j Vλ +2 : Vλ +2 → Vλ +2 . But λ > crit (j),
and hence j Vλ +2 is non-trivial and violates Theorem 2.
(3) immediately follows from (2), and (4) follows by Kunen’s Theorem7 .

To see (5), define ĵ by ĵ(a) = n<ω a ∩ Vλn for λn | n < ω some increasing
cofinal sequence in λ. Then clearly ĵ does not depend on which cofinal sequence
was chosen, and it is a Σ0 -elementary embedding Vλ+1 → Vλ+1 .
For a proof of (6) see [Lav97].
Because of (5), we will use the convention below that ĵ and j are identified. So
if j : Vλ → Vλ is elementary then we will write j(j) for ĵ(j). Note that we have the
following theorem.
Theorem 6 (Laver[Lav97]). If j, k : Vλ+1 → Vλ+1 are Σn -elementary then
j(k) : Vλ+1 → Vλ+1 is Σn -elementary.
Hence, conveniently, the set of I1 -embeddings acts on itself using the extensions
ĵ for j an I1 embedding. This is also true for I3 -embeddings. Such facts allow for
algebras of elementary embeddings. We refer the interested reader to [Lav02] and
[LM13] for more on these algebras.
3. Reflection properties
We will now look at the reflection properties of rank-into-rank embeddings.
These reflection theorems are in a sense harder to achieve than is typically the
case with large cardinal axioms. The reason is that in reflecting a rank-into-rank
embedding we not only wish to decrease the critical point of the embedding, but
we wish to decrease λ, the first fixed point of the embedding. The usual arguments
for reflecting large cardinals in this case give us what we will call square roots of
6 κ | i < ω is referred to as the critical sequence of j.

i
7 Note that the proof of Kunen’s Theorem does not require that j actually be in V.
230 SCOTT CRAMER
our embedding, whereas to reflect the large cardinal by decreasing λ will gener-
ally require more involved arguments, and, in particular, sequences of embeddings.
While this might at first seem like a simple nuisance, the techniques needed for
these reflection results will end up being useful in a wide range of Theorems below.
We concentrate on the following fairly weak form of reflection: we say a large
cardinal axiom LCA1 reflects a large cardinal axiom LCA2 if for any λ if LCA1
holds at λ then LCA2 holds at λ and there is λ̄ < λ such that LCA2 holds at λ̄.
We also write LCA2 < LCA1. The following theorem summarizes some results on
reflection of rank-into-rank embeddings in the order in which they were proved.
These results can be extended to stronger forms of reflection, such as finding an
ω-club of λ̄ below λ at which LCA2 holds. We will discuss other strengthenings of
these results at the end of this section.
Theorem 7.
(1) I3 < I2 (Gaifman[Gai74], Solovay).
(2) I3 < I1 (iterable) < I1 (Σ2 ) = I2 (Martin).
(3) I1 < I0 (Woodin[Woo11]).
(4) I1 (Σ2 ) < I1 (Σ4 ) < I1 (Σ6 ) < · · · < I1 < I0 (1) (Laver[Lav97]).
(5) I0 (1) < I0 (ω + 1) < I0 (λ+ ) < I0 (λ+ + ω + 1). (Laver[Lav01])8
(6) I0 (< λ+ + ω) < I0 (λ+ + ω) and I0 (< α + ω) < I0 (α + ω) for α the least
admissible ordinal (C. [Cra15e]).
(7) For α good9 , I0 (α + ω) holding at λ implies that there is λ̄ < λ and ᾱ such
that for all n < ω there is an elementary embedding
Lᾱ+n (Vλ̄+1 ) → Lα+n (Vλ+1 )
and hence I0 (< ᾱ + ω) holds at λ̄ (C. [Cra15e]).10
(8) I0 < I0# (ω) (C. [Cra15e]).
The proofs for (1) and (2) involved taking the ωth iterate of an embedding
and then considering the tree of attempts to build an I3 embedding. The proofs of
(4)-(7) on the other hand involved inverse limits of rank-into-rank embeddings, a
technique introduced by Laver that we will describe below.
As mentioned above, these reflection results were shown using sequences of em-
beddings (either direct limits or inverse limits) rather than individual embeddings.
The usual method for obtaining reflection from a large cardinal in the case of rank-
into-rank embeddings gives us what we call square roots. The primary method for
obtaining such square roots is the following fact, which we generally refer to as the
Square Root Lemma:
Lemma 8. (Martin, Laver) Let α be good. If j : Lα+1 (Vλ+1 ) → Lα+1 (Vλ+1 ) is
an elementary embedding with crit (j) < λ, a, b ∈ Vλ+1 , then there is an elementary
embedding k : Lα (Vλ+1 ) → Lα (Vλ+1 ) such that
(1) k(k Vλ ) = j Vλ
(2) b ∈ rng k
8 Note that in (5) and (6) we mean that if I (λ+ + ω + 1) holds at λ then there is a λ̄ < λ
0
such that I0 (λ̄+ ) holds at λ̄.
9 α is good if every element of L (V
α λ+1 ) is definable over Lα (Vλ+1 ) from an element of Vλ+1 .
The good ordinals are cofinal in Θλ (see [Lav01])
10 We state (7) in this manner because for larger α there is no simple definition of α as in the
case of α = λ+ for instance. So we cannot state the reflection as we do above.

(3) j(a) = k(a).

Proof sketch. Apply j to the statement of the lemma and then show that j
Lα (Vλ+1 ) witnesses the result. The fact that α is good implies that j Lα (Vλ+1 ) ∈
Lα+1 (Vλ+1 ).
Of course a could be replaced by a finite tuple a1 , . . . , an and similarly for b.

In fact these can replaced by tuples of length < crit (j).
If k(k Vλ ) = j Vλ then we say that k is a square root of j. The Square
Root Lemma has many varitions, as for instance Laver included the fact that for a
given α < crit (j), k has the property that α < crit (k) < crit (j). Notice however,
that we can require that α ∈ rng k. And this implies that α < crit (k) < crit (j)
because k(crit (k)) = crit (j) implies crit (k) < crit (j), and α < crit (k) since for any
β ∈ [crit (k), crit (j)), β ∈
/ rng k. In fact, property (3) in the statement of the Square
Root Lemma is in fact redundant, by property (2) by the following proposition.
Proposition 9. Suppose that j, k : Vλ → Vλ and k is a square root of j. Then
the following hold.
(1) crit (k) < crit (j) and k(crit (k)) = crit (j).
(2) If a ∈ Vλ+1 and a ∈ rng k, then k(a) = j(a).
(3) For all α < λ, k(α) ≥ j(α).
Proof. (1) follows by computing11
k(crit (k)) = crit (k(k)) = crit (j).
Similarly, for (2), if k(b) = a, then, informally,12
k(a) = k(k(b)) = k(k)(k(b)) = j(a).
For (3), suppose not, and let α be least such that k(α) < j(α). Then using the
fact that α is least such, j(α) must be in the range of k, since it is the least ordinal
above k(α) in the range of j. Hence, α ∈ rng k. But then k(α) = j(α) by (2), a
contradiction.
We now come to a key application of the Square Root Lemma: the construction
of inverse limits of rank-into-rank embeddings. These structures are key to proving
reflection results for rank-into-rank embeddings, and they naturally arise from the
Square Root Lemma. In fact, an inverse limit can be constructed simply as the
result of ω-many applications of the Square Root Lemma for a fixed j, where at
each stage we choose ki such that everything in the construction so far is required
to be in the range of ki .
Definition 10 (Laver). We say that (J, ji | i < ω) = (J, j) is an inverse limit
if the following hold.
(1) For all i < ω, ji : Vλ → Vλ is elementary.
(2) crit (j0 ) < crit (j1 ) < · · · < λ and limi<ω crit (ji ) = λ̄J < λ.
11 Note that we are being careful only to use elementarity below λ. That is, we are using
properties of k which are properties of k Vα for large enough α < λ.

12 As before, we should restate this in terms of k V
α+1 (a ∩ Vα ) for arbitrarily large α < λ,
and then use continuity of this property.
232 SCOTT CRAMER
λ - - - λ
*

j0 (j1 (λ̄))
···
* .
..
*

cr(j0 (j1 (j2 )))
-
j0 (λ̄) cr(j0 (j1 (j1 )))
j0 (j1 (j2 ))

*
3 ..
.

cr(j0 (j2 ))

cr(j0 (j1 )) j0 (j1 )
λ̄ -
.. cr(j0 (j0 ))
.
cr(j2 )
3

cr(j1 ) j
0
cr(j0 )
Figure 2. Direct limit decomposition of an inverse limit.
(3) J : Vλ̄J → Vλ is given by

J(a) = lim(j0 ◦ j1 ◦ · · · ◦ ji )(a)
i<ω
for any a ∈ Vλ̄ .
We write J = j0 ◦ j1 ◦ · · · to denote an inverse limit. For n < ω, we write Jn =
jn ◦ jn+1 ◦ · · ·.
Notice in this definition that for any a ∈ Vλ̄ there is an i < ω such that for all
n ≥ i, jn (a) = a, so the limit in the definition of J makes sense. An inverse limit
can also be written as a direct limit as follows (see Figure 2):
J = j0 ◦ j1 ◦ · · · = · · · (j0 ◦ j1 )(j2 ) ◦ j0 (j1 ) ◦ j0 .
Proposition 11 (Laver[Lav97]). Suppose that α is good and j : Lα+1 (Vλ+1 ) →

Lα+1 (Vλ+1 ) is elementary. Then there is an inverse limit (K, k) such that for all
i < ω, ki is a square root of j,
k0 , . . . , ki ∈ rng ki+1 ,
and ki extends to an elementary embedding Lα (Vλ+1 ) → Lα (Vλ+1 ).
Proof. 13 Choose ki for i < ω by induction using the Square Root Lemma,
requiring that for all i < ω, k0 , . . . , ki ∈ rng ki+1 . Then we have by the above
13 The following proof is a slight simplification from Laver’s original as we use the remarks
made above about the Square Root Lemma. Also, Laver did not include the fact that ki is a
square root of j or that k0 , . . . , ki ∈ rng ki+1 .
remarks
crit (k0 ) < crit (k1 ) < · · · < crit (j) < λ.
And hence K = k0 ◦ k1 ◦ · · · is the desired inverse limit.
As is to some extent clear from the definition, inverse limits are especially useful
for proving reflection results on rank-into-rank embeddings. The key question in
performing this reflection is to what extent an inverse limit (J, j), which only gives
an embedding Vλ̄J → Vλ can be extended to Vλ̄J +1 or beyond to L(Vλ̄J +1 ). One key
fact in finding such an extension is that inverse limits also satisfy a certain Square
Root Lemma. In fact, both the extension of inverse limits and the corresponding
Square Root Lemma can be viewed as instances of the same phenomenon: a prop-
erty of the embeddings comprising an inverse limit transfers to the inverse limit
itself. Although some instances of this transference can be viewed as immediate
consequences of the continuity of the given property, in other cases the proof is
much more involved.
We now state the Square Root Lemma for inverse limits.
Proposition 12 (Laver). Suppose that α is good and (J, j) is an inverse limit
such that for all i < ω, ji extends to an elementary embedding Lα+1 (Vλ+1 ) →
Lα+1 (Vλ+1 ). Then for any a ∈ Vλ̄J +1 and b ∈ Vλ+1 there is an inverse limit (K, k)
such that the following hold:
(1) for all i < ω ki is a square root of ji ,
(2) ki extends to an elementary embedding Lα (Vλ+1 ) → Lα (Vλ+1 ),
(3) λ̄J = λ̄K ,
(4) K(a) = J(a) and b ∈ rng K.
Proof sketch. Choose ki by induction on i < ω using the Square Root
Lemma, ensuring that the following hold:
(1) for all n < ω, jn ∈ rng ki ,
(2) for all n < i, kn ∈ rng ki ,
(3) a, b ∈ rng ki .
A computation then shows that (K, k) is as desired.
The square root lemma for inverse limits, Proposition 12, is the key technique
used in Laver’s proof that I0 (1) < I0 (ω+1). The proof proceeds by using the square
root lemma to derive elementarity of the inverse limit. This type of argument
generally proceeds by induction on the elementarity of the embeddings making up
the inverse limit, and the square root lemma is used to ‘capture’ witnesses to a given
statement. Notice that the strength of the inverse limit (K, k) given by the square
root lemma decreases from that of (J, j), a fact which is necessary for this specific
statement of the lemma.14 However, this decrease in strength is what necessitates
induction in this method.
On the other hard, the proof that I0 < I0# (ω) uses an alternative form of
Proposition 12 and avoids the need for induction. This alternative square root
lemma is also important in defining the axiom we will call Inverse Limit Reflection
in Section 5. First we define the notion of a limit root.
14 To see this, take (J,
j) with critical point as small as possible. Then notice that crit (K) <
crit (J).
234 SCOTT CRAMER
Definition 13 (C.). Let (J, j) and (K, k) be inverse limits and n < ω. Then
we say that (K, k) is an (n-close) limit root of (J, j) if the following hold
(1) for all i < n, ki = ji ,
(2) for all i ≥ n, ki is a square root of ji ,
(3) λ̄J = λ̄K .
A key difference between limit roots and square roots is the following: if k is a
square root of j then crit (k) < crit (j) (since k(crit (k)) = crit (j)), but if (K, k) is
a limit root of (J, j) then it is possible to have crit (K) = crit (J) (in particular, iff
K is n-close to+ J for n ≥, 1). Hence this opens up the possibility of a sequence of
inverse limits K i | i < ω such that for all i < ω, K i+1 is a limit root of K i . We
can in fact achieve this by considering an inverse limit (J, j) such that as i → ω,
the strength of ji increases. We illustrate this phenomenon in the following lemma.
Lemma 14 (C.). Assume there is an elementary embedding j : Lω (Vλ+1 ) →
Lω (Vλ+1 ) with crit j < λ. Let E be the set of inverse limits (J, j) such that for
some sequence in | n < ω below ω the following hold:
(1) for all n < ω, 0 < in < in+1 ,
(2) for all n < ω, jn extends to an elementary embedding Lin (Vλ+1 ) →
Lin (Vλ+1 ).
Then E satisfies the following: for any (J, j) ∈ E there is an n < ω such that for
all a ∈ Vλ̄J +1 and b ∈ Vλ+1 , there is (K, k) ∈ E such that the following hold:
(1) K is an n-close limit root of J,
(2) K(a) = J(a) and b ∈ rng Kn .
Notice that we would not be able to have such a fact be true about square roots
of embeddings. The key point is that we only require that b ∈ rng Kn rather than
b ∈ rng K. It is also important in practice that the given n work for any a and b
(although this is not much more difficult to achieve).
Proof. For (J, j) ∈ E, let in | n < ω witness this fact. Then let n be such
that for all n ≥ n, in > 0. We can use the proof of Proposition 12 on Jn , a, b,
and α = in − 1 to get Kn . Let K be the n-close limit root of J defined this way.
We then have that
K(a) = (k0 ◦ k1 ◦ · · · kn−1 )(Kn (a))
= (j0 ◦ j1 ◦ · · · jn−1 )(Kn (a))
= (j0 ◦ j1 ◦ · · · jn−1 )(Jn (a)) = J(a)
and b ∈ rng Kn .
We say that E is saturated if it satisfies the conclusion of Lemma 14. The
following then illustrates the type of reflection result which can be obtained for
inverse limits.
Theorem 15 (C.[Cra15e]). Suppose that α is good and there is an elementary
embedding
Lα+ω (Vλ+1 ) → Lα+ω (Vλ+1 )
with critical point below λ. Then there are ᾱ, λ̄ and a saturated set of inverse limits
E such that for all (J, j) ∈ E, J extends to an elementary embedding
J : Lᾱ (Vλ̄+1 ) → Lα (Vλ+1 ).
The importance of this stronger form of reflection15 comes from its structural
implications for L(Vλ+1 ). This fact allows us to isolate a stronger property of sub-
sets X ⊆ Vλ+1 which appears to be more closer to an AD-like axiom for L(X, Vλ+1 )
than the property that X is Icarus. We will investigate this idea in Section 5.
4. Structure of L(Vλ+1 )
We now look at the structure of L(Vλ+1 ) under the assumption that I0 holds
at λ. The main theme in this study is that the structure of L(Vλ+1 ) under I0 is
very similar to the structure of L(R) assuming ADL(R) . However, there are several
important differences between the two situations which we highlight:
Perhaps the most apparent distinction, but the most easily remedied, is that
I0 is not a first-order statement in L(Vλ+1 ), whereas AD is a first-order statement
in L(R). The statement I0 (< Θλ ) is however a first-order statement in L(Vλ+1 ),
and most of the structural implications of I0 for L(Vλ+1 ) follow from this weaker
axiom.
Woodin showed that assuming a proper class of Woodin cardinals, the theory
of L(R) is generically absolute. Such a fact is not the case for L(Vλ+1 ) however, as
the theory of Vλ is not generically absolute for small forcings (such as changing the
size of the continuum). Therefore in deciding whether a given property holds in
L(Vλ+1 ) assuming I0 , one must also allow for the possibility that it is dependent on
V . One solution to this problem, suggested by Woodin, is to consider the structure
of L(Vλ+1 ) in a canonical inner model satisfying I0 . However, as such an ‘ultimate
L’ does not yet exist, there have been no such results. Nevertheless there are many
important properties of L(Vλ+1 ) which do not depend on V , and we will describe
these below.
Arguably the most important distinction is the following. The structural prop-
erties of L(Vλ+1 ) under I0 do not necessarily translate into properties of L(X, Vλ+1 )
assuming X ⊆ Vλ+1 is Icarus. The axiom I0 therefore does not appear to be an ap-
propriate analog of AD in the sense of being a fundamental regularity property, for
instance. This fact naturally leads to the question of what would be an appropriate
AD-like axiom, and this question is the subject of Section 5.
Some of the results below require that the embedding j be proper. As we
are focusing on the case of L(Vλ+1 ), we refer the reader to [Woo11] Section 3
for the definition of proper elementary embeddings. In the case of L(Vλ+1 ), the
word proper can simply be removed from everything we say (although not all I0
embeddings are proper).
4.1. Measurable cardinals below Θλ . We begin by looking at measurable

cardinals in L(Vλ+1 ). We first recall the situation of L(R), where it has been shown
that ΘL(R) is a limit of of large cardinals.
Theorem 16 (Solovay(see [KW10]), Steel[Ste95]). Assume ADL(R) . Then
in L(R), ω1 is measurable, and in fact the club filter is an ultrafilter. Also every
regular cardinal below Θ is measurable.
Woodin was able to show a number of these facts hold for L(X, Vλ+1 ) assuming
X is Icarus.
15 In fact you need a slightly stronger form of reflection, Strong Inverse Limit Reflection,
which we will define below.

236 SCOTT CRAMER
Theorem 17 (Woodin[Woo11]). Assume that j : L(X, Vλ+1 ) → L(X, Vλ+1 )

is a proper elementary embedding. Then λ+ is measurable in L(X, Vλ+1 ), and in
fact Θ = ΘL(X,Vλ+1 ) is a limit of γ such that the following hold.
(1) γ is weakly inaccessible.
(2) γ = ΘLγ (X,Vλ+1 ) and j(γ) = γ.
(3) for all β < γ, P (β) ∩ L(X, Vλ+1 ) ⊆ Lγ (X, Vλ+1 ).
(4) for cofinally many κ < γ, κ is measurable in L(X, Vλ+1 ) as witnessed by
the club filter on a stationary set,
(5) Lγ (X, Vλ+1 ) ≺ LΘ (X, Vλ+1 ).
Below we will sketch a proof that there is a stationary subset S of λ+ such
that the club filter restricted to S is an ultrafilter in L(Vλ+1 ) (Theorem 22). This
fact shows that λ+ is indeed measurable in L(Vλ+1 ). The main point is that the
existence of the I0 embedding shows that λ+ cannot be partitioned into λ-many
disjoint stationary sets. We mention a few questions left open by this analysis.
Question 18. If I0 holds at λ, in L(Vλ+1 ) are λ++ , λ+++ , etc. regular or
singular?
Question 19. If I0 holds at λ, in L(Vλ+1 ) is every regular cardinal below Θλ
and above λ measurable?
In the course of proving this theorem, Woodin also proved the following analog
of Moschovakis’ Coding Lemma.
Theorem 20 (Woodin). Assume that X ⊆ Vλ+1 is Icarus. Suppose ρ : Vλ+1 →
η is a surjection with ρ ∈ L(X, Vλ+1 ). Then there exists γρ < ΘL(X,Vλ+1 ) such that
for all sets A ⊆ Vλ+1 × Vλ+1 with A ∈ L(X, Vλ+1 ), there exists B ⊆ Vλ+1 × Vλ+1
such that
(1) B ⊆ A,
(2) for all α < η, if there exists (a, b) ∈ A with ρ(a) = α, then there exists
(a, b) ∈ B with ρ(a) = α,
(3) B ∈ Lγρ (X, Vλ+1 ).
This theorem however gives a rather coarse version of the Coding Lemma, and
therefore leaves open such questions as the following.
Question 21 (Woodin). If I0 holds at λ, is
P (λ+ ) ∩ L(Vλ+1 ) = P (λ+ ) ∩ Lλ (H(λ+ ))?
More generally, at what point have all the subsets of λ+ in L(Vλ+1 ) already ap-
peared?
As mentioned above, a key fact in the proof of Theorem 17 is a restriction
on partitioning regular cardinals into stationary sets. In particular we have the
following.
Theorem 22 (Woodin). Assume X ⊆ Vλ+1 is Icarus and γ < Θ is such that
cof(γ) > λ. Then for
S = {ξ < γ| cof(γ) = ω},
there is a partition
Sα | α < η ∈ L(X, Vλ+1 ),
of S such that η < λ and such that in L(X, Vλ+1 ), for all α < η, F Sα is an
ultrafilter, where F is the ω-club filter on γ.
Proof. Suppose towards a contradiction that there is such a γ, and let γ ∗ be

least such. Hence for j an I0 (X)-embedding, we have j(γ ∗ ) = γ ∗ . Let
Sα | α < crit (j)
be a partition of S into crit (j)-many disjoint stationary sets. Let
Tα | α < j(crit (j)) = j(Sα | α < crit (j)).
Then by elementarity Tα | α < j(crit (j)) is a partition of S into disjoint stationary
sets (here we are using that j(S) = S). Note that
C = {δ < γ ∗ | j(δ) = δ}
is ω-club below γ ∗ . Hence there is δ ∈ C ∩ Tcrit (j) . But there is some α < crit (j)
such that δ ∈ Sα . And hence
j(δ) = δ ∈ Tα = j(Sα ).
And this is a contradiction since then Tα and Tcrit (j) are not disjoint.
Note that in Theorem 22 (if j is finitely iterable) we can replace S by S β =

{ξ < γ| cof(ξ) = β} where β < λ is a regular cardinal. We then modify the proof
by considering a large enough finite iterate of j so that C is indeed β-club below
γ∗.
We now look at the specific case of λ+ in L(Vλ+1 ). In that case we have the
following.
Theorem 23 (Woodin). Assume I0 holds at λ. Then in L(Vλ+1 ), λ+ can be
partitioned into stationary sets Sα for α < λ such that for all α < λ, F Sα is an
ultrafilter, where F is the club filter on λ+ .
One question is to what extent this theorem can be improved so that it more
closely resembles the corresponding result under AD that ω1 is measurable as wit-
nessed by the club filter. Since there are λ many regular cardinals below λ+ , this
result cannot be improved as stated, but one might ask whether the club filter
restricted to S κ = {η < λ+ | cof(η) = κ} is an ultrafilter. The next theorem shows
that for κ uncountable this cannot be proven from I0 .
Theorem 24 (Woodin). Suppose I0 holds at λ. Let κ < λ be an uncountable
cardinal and g ⊆ Coll(κ, κ+ ) be V -generic. Then in V [g], I0 holds at λ and in
L(V [g]λ+1 ), S κ can be partitioned into two stationary sets.
Proof. Let S1 = {α < λ+ | cof(α)V = κ} and S2 = {α < λ+ | cof(α)V = κ+ }.
Since g does not add any ω-sequences we have that Vλ+1 ∈ L(V [g]λ+1 ) since Vλ ∈
V [g]λ+1 . So clearly S1 , S2 ∈ L(V [g]λ+1 ), they are stationary, and they partition
S κ.
In fact Woodin has shown a stronger fact using Radin forcing.

Theorem 25 (Woodin). Suppose I0 holds at λ as witnessed by j. Let κ <
crit (j). Then there is P ∈ Vcrit (j) such that for g ⊆ P V -generic, we have in V [g]
that I0 holds at λ, Vκ = V [g]κ , and in L(V [g]λ+1 ), S ω1 can be partitioned into
κ-many stationary sets.
238 SCOTT CRAMER
This theorem shows that I0 alone gives virtually no information as far as to what
extent, for instance, S ω1 can be partitioned. Similar results hold for cofinalities
larger than ω1 . Importantly, however, it leaves open the case of κ = ω, and in fact
Woodin proved the following theorem which shows that the above proof fails in the
κ = ω case. This theorem is a corollary of Theorems 36 and 35.
Theorem 26 (Woodin). Suppose that I0 holds at λ. Suppose that P ∈ Vλ , g is
P-generic over V and
(λω )V = (λω )V [g] .
Then in V [g],
Vλ+1 ∈
/ L(V [g]λ+1 ).
Now, concentrating on the case of S ω , the following theorem gives evidence
that perhaps the club filter is an ultrafilter when restricted to S ω .
Theorem 27 (C., Woodin (for Lλ (Vλ+1 ))). Suppose that I0 holds at λ. Then
there are no disjoint stationary (in V ) sets S1 , S2 ∈ L(Vλ+1 ) such that S1 , S2 ⊆ S ω .
However this theorem does not quite show that in L(Vλ+1 ) there are no disjoint
stationary subsets of S ω , since such sets might not be stationary in V . Hence we
have the following question.
Question 28 (Woodin). Suppose I0 holds at λ. In L(Vλ+1 ) is the club filter
restricted to S ω an ultrafilter?
As for partition properties at λ+ , we have since ω1 -DC holds in L(Vλ+1 ) that
L(Vλ+1 ) |= λ+ → (λ+ )ω
λ .
1
This leads to the following question.

Question 29 (Woodin). If I0 holds at λ, then for all α < ω1 do we have
L(Vλ+1 ) |= λ+ → (λ+ )α
λ?
Woodin showed that for α such that ω·α = α, if this properties fails in V then it
must fail in all generic extensions by a forcing P ∈ Vλ . In the positive direction the
best known partial result is the following, whose proof uses Radin forcing together
with generic absoluteness properties, Theorems 62 and 82 below.
Theorem 30 (Woodin). Suppose I0 holds at λ. Then for all α < β < ω1 ,
L(Vλ+1 ) |= λ+ → (β)α
λ.
4.2. Perfect set property. The perfect set property for a set of reals X is
a strong version of the continuum hypothesis. It states that either X is countable
or it contains a perfect set16 , in which case |X|= |R|= 2ℵ0 . While the Axiom of
Choice implies that there is a set of reals which does not satisfy the perfect set
property, in fact AD implies that every set of reals has the perfect set property.
Theorem 31 (M. Davis[Dav64]). Assume AD holds. Then if X ⊆ R, either
X is countable or X contains a perfect set.
16 A set is perfect if it is closed and has no isolated points.

Using U (j)-representations (see Section 5), Woodin was able to show that a
version of the perfect set theorem in the context of AD holds for Lλ (Vλ+1 ). To
do this, we regard Vλ+1 as a topological space with basic open sets O(a,α) , where
α < λ, a ⊆ Vα and
O(a,α) = {b ∈ Vλ+1 | b ∩ Vα = a}.
Since cof(λ) = ω, this is a metric topology, and it is complete.
Perfect sets in the usual context of the reals are somewhat simpler than perfect
sets in the context of Vλ+1 , since perfect subsets of Vλ+1 can have varying sizes.
We are most interested in perfect sets which have the largest possible size, so that
our perfect set property continues to be a strong form of the generalized continuum
hypothesis. With this in mind we make the following definition.
Definition 32. Suppose P ⊆ Vλ+1 is a perfect set. For κ ≤ λ, we say that P
is κ-splitting 17 if for all a ∈ P , γ < λ, and α < κ, there is a β < λ such that
|{b ∩ Vβ | b ∈ P and b ∩ Vγ = a ∩ Vγ }|≥ α.
We say that X ⊆ Vλ+1 has the λ-splitting perfect set property if either |X|≤ λ
or X contains a λ-splitting perfect set.
Using U (j)-representations (see Section 5.1), Woodin was able to prove the
following.
Theorem 33 (Woodin). Suppose I0 holds at λ. Then if Z ∈ Lλ (Vλ+1 ) ∩ Vλ+2 ,
then either |Z|≤ λ or Z contains a perfect set.
Shi-Woodin later improved this fact using Theorem 62 with diagonal super-
compact Prikry forcing.
Theorem 34 (Shi-Woodin[Shi15]). Suppose I0 holds at λ. Then for all Z ∈
Lλ (Vλ+1 )∩Vλ+2 , either |Z|≤ λ or Z contains a λ-splitting perfect set, and therefore
|Z|= 2λ , so Z has the λ-splitting perfect set property.
Using the tool of inverse limit reflection, we were able to improve this result to
all of L(Vλ+1 ).
Theorem 35 (C.[Cra15e]). Suppose I0 holds at λ. Then if Z ∈ L(Vλ+1 ) ∩
Vλ+2 , then either |Z|≤ λ or Z contains a λ-splitting perfect set, and therefore
|Z|= 2λ .
There are several interesting consequences of this perfect set theorem. First we
obtain some information on L(Vλ+1 ) in forcing extensions. In particular we have
the following.
Theorem 36 (Woodin[Woo11] Theorem 175). Suppose that I0 holds at λ.
Suppose that P ∈ Vλ , g is P-generic over V and
(λω )V = (λω )V [g] .
Then L(Vλ+1 , V [g]λ+1 ) does not satisfy the λ-splitting perfect set theorem.
17 One can also define κ-splitting in terms of the splitting number of the natural tree of initial
segments of elements of P .
240 SCOTT CRAMER
Proof sketch. The point is that in V [g], since |Vλ+1 |> λ, there must be a
λ-splitting perfect set P ⊆ Vλ+1 such that P ∈ L(Vλ+1 , V [g]λ+1 ). But since P is a
small forcing, this shows18 that in fact any ω sequence of ordinals below λ can be
coded into an element of P . And hence (λω )V [g] ⊆ V , which is a contradiction.
Theorem 36 gives an interesting example of an X ⊆ Vλ+1 such that X is Icarus,
but subsets of Vλ+1 in L(X, Vλ+1 ) do not all satisfy a certain regularity property,
namely the λ-splitting perfect set property. This example has a number of inter-
esting consequences for the AD-like axioms which we will consider in Section 5. In
particular it gives a contradiction to the assertion that I0 is in fact the appropriate
analog of AD in the context of Vλ+1 , since I0 (X) holds for this X, and it at the
same time displays the difficulty in propagating the AD-like axioms of Section 5,
as these axioms do not propagate generally throughout all L(X, Vλ+1 ) structures.
On the other hand, results such as Theorem 36 potentially give criteria for evalu-
ating extensions of L(Vλ+1 ), and they highlight the importance of propagating the
potential AD-like axioms beyond I0 (see Section 6).
4.3. Uniformization. The notion of a uniformization of a relation R on R×R
is easily generalized to Vλ+1 .
Definition 37. Suppose R ⊆ Vλ+1 × Vλ+1 is a relation. Then U ⊆ R is a
uniformization of R if for all x ∈ dom(R), there is exactly one y ∈ Vλ+1 such that
(x, y) ∈ U .
However, the question in L(Vλ+1 ) of which relations have a uniformization is
not very well understood at this moment. As in the case of L(R), uniformization
cannot hold in L(Vλ+1 ).
Fact 38. In L(Vλ+1 ) there is a set Z ⊆ Vλ+1 × Vλ+1 such that Z has no
uniformization.
Proof. Work in L(Vλ+1 ). Let Z be defined by
Z = {(x, y)| y is not OD from x}.
Suppose U ⊆ Z is a uniformization. Then U is OD from some x ∈ Vλ+1 . But then
U (x) is OD from x, which is a contradiction.
In the case of L(R), however, uniformizations can be obtained from scales, and
every Σ1 (L(R), {R} ∪ R)-set has a scale (see [MS83]).
In the L(Vλ+1 ) case on the other hand, where the first failure of uniformization
occurs is far from clear.
Question 39 (Woodin). Suppose I0 holds at λ. Let Rsqrt ⊆ Vλ+1 × Vλ+1 be
defined by
Rsqrt = {(j, k)| j, k : Vλ+1 → Vλ+1 are elementary and k is a square root of j}.
Does Rsqrt have a uniformization in L(Vλ+1 )?
Most of the results on uniformization for L(Vλ+1 ) are negative results stem-
ming from Fact 38 and the positive results on AD-like properties for subsets of
Vλ+1 in L(Vλ+1 ). In particular we have Corollaries 68 and 76 which state that
18 See [Woo11] Theorem 175, Page 395.

uniformization does not follow from either U (j)-representability or strong inverse

limit reflection.
We now give some background surrounding Question 39 and show that we can
get somewhat close to such a uniformization. In particular we will restrict the do-
main of the relation Rsqrt , and then show that the restriction has a uniformization.
Of course, this process involves a restriction of the domain, and therefore does not
succeed in uniformizing Rsqrt . However, we can also show that if a uniformization
of Rsqrt exists, it is an extension of of the type that we define below. This fact
gives perhaps a small indication that such a uniformization does not exist.
We first make the following definition for α < Θλ , j : Vλ → Vλ elementary, and
b ∈ Lα (Vλ+1 ):
Eα (b) = {k : Vλ → Vλ | k extends to k̂, k̂ : Lα (Vλ+1 ) → Lα (Vλ+1 ) with b ∈ rng k̂},
Eα = Eα (∅) and Eαj (b) = {k ∈ Eα (b)| k is a square root of j }.
We also write Eα (a, b) for Eα ({a, b}).
Proposition 40. Assume I0 holds at λ. Suppose that R ⊆ Vλ+1 × Vλ+1 and
R ∈ L(Vλ+1 ). Let α < Θ be good and such that R ∈ Lα (Vλ+1 ). Let j ∈ Eα+1
j
and suppose there is k ∈ Eα+1 (R) and a ∈ Vλ+1 such that (k, a) ∈ R. Then
j
R Eα+1 (R, k(a)) is uniformizable in L(Vλ+1 ).
Proof. Let j, k and a be as in the hypothesis. Then by applying k̂ to the fact

that (k, a) ∈ R we have that (j, k(a)) ∈ k̂(R). But k̂(R) = ĵ(R) since k is a square
root of j, α is good, and R ∈ rng k. So (j, k(a)) ∈ ĵ(R).
We define
U = {(, −1 (k(a)))| ∈ Eα+1
j
(R, k(a))}.
Note that we have
(j, k(a)) ∈ ĵ(R) ⇒ (, −1 (k(a))) ∈ ˆ−1 (ĵ(R)) = R
ˆ and ∈ E j (R, k(a)). Hence, U is a uniformization
since we can pull back by , α+1
j
of R Eα+1 (R, k(a)), and clearly U ∈ L(Vλ+1 ).
Corollary 41. Assume I0 holds at λ. Let j ∈ E1 and suppose (j, k) ∈ Rsqrt .
Then Rsqrt E1j (k) is uniformizable in L(Vλ+1 ).
We can also prove the following partial converse of the above proposition.
Proposition 42. Assume I0 holds at λ. Suppose that R ⊆ Vλ+1 × Vλ+1 and
R ∈ L(Vλ+1 ). Also assume that U is a uniformization of R and U ∈ Lα (Vλ+1 )
where α is good. Then for any j ∈ Eα+1 if b ∈ Vλ+1 is such that (j, b) ∈ ĵ(U ), we
have that
j
U Eα+1 (U, b) = {(, −1 (b))| ∈ Eα+1
j
(U, b)}.
Proof. To see this, suppose that (j, b) ∈ ĵ(U ) is as in the hypothesis. Let
j
∈ Eα+1 (U, b). Then we have that
(j, b) ∈ ĵ(U ) ⇒ (, −1 (b)) ∈ ˆ−1 (ĵ(U )) = U.

For the next definition we use the notation that j(n) is the nth iterate of j.
That is, j(0) = j and j(n+1) = j(j(n) ) for n < ω.
242 SCOTT CRAMER
Definition 43. For j ∈ Eκ and a ∈ Lκ (Vλ+1 ), we let Ẽκj (a) be the set of
embeddings k ∈ Eκ such that the following hold:
(1) For some m and n < ω, j(m) = k(n) .
(2) For some (all) m and n as in (1), j0,m (a) ∈ rng k0,n .
−1
If k ∈ Ẽκj (a), we set Ẽκj (k)[a] = k0,n (j0,m (a)) where n and m are as in (2).
We can then improve Proposition 40 to these larger sets in the case R is fixed
by j.
Proposition 44. Suppose that j ∈ Eκ and R ⊆ Vλ+1 × Vλ+1 , R ∈ Lκ (Vλ+1 )
is such that j(R) = R. Suppose that (j, a) ∈ R. Then for Uj,a the set of (k, b) such
that k ∈ Ẽκj ({a, R}) and b = Ẽκj (k)[a], we have that the following:
(1) Uj,a is a uniformization of R Ẽκj ({a, R}).
(2) If κ is good and j extends to an elementary embedding j ∗ : Lκ+1 (Vλ+1 ) →
Lκ+1 (Vλ+1 ) then j ∗ (Uj,a ) = Uj,a .
The proof of this proposition is very similar to the proof of Proposition 40.
And we also have the corresponding result to Proposition 42, though we leave the
statement to the reader. These results lead us to the following conjecture.
Conjecture 45. Assuming I0 holds at λ as witnessed by j, the relations
Rsqrt E0j and Rsqrt Ẽ0j (∅) have no uniformizations in L(Vλ+1 ).
We make this conjecture as a result of the above converse propositions, which
seem to indicate that the only method for creating uniformizations of these types
of relations is through the above ‘pullback’ method, which is necessarily insufficient
for a full uniformization.
5. AD-like axioms for L(Vλ+1 )

We have seen in the previous section that many structural results can be ob-
tained for L(Vλ+1 ) by assuming I0 holds at λ. These facts point to I0 being an
‘AD-like axiom’ in the sense that it gives a structure theory of L(Vλ+1 ). As we noted
above, however, there are various objections which could be raised against this as-
sertion. In this section we introduce several alternative axioms19 which have been
proposed as AD-like axioms. First we will introduce U (j)-representations which
were defined by Woodin, then we will introduce inverse limit reflection which was
defined by the author, and lastly we will define some more recent axioms motivated
by U (j)-representations.
5.1. U (j)-representations. By way of motivation we first define Suslin and

weakly homogeneously Suslin representations. In general, in the context of R, for
a set A we say that X ⊆ R has a tree representation T on ω × A if T is a tree20
on21 (ω × A)<ω and for any x ∈ R,
x ∈ X ⇐⇒ Tx is illfounded.
19 Note that, importantly, we will see that these axioms do follow from I for L(V
0 λ+1 ). It
would therefore make sense to think of I0 as being the large cardinal axiom which is implying the
AD-like axiom for L(Vλ+1 ) much in the same way as large cardinals imply AD L(R) .
20 A tree is a set closed under initial segments.
21 We will confuse (ω × A)<ω with ω <ω × A<ω , as is customary.
Here
Tx = {b| ∃n < ω ((x n, b) ∈ T )}
<ω
is the tree on A associated to x. Saying that Tx is illfounded is equivalent to the
existence of some b ∈ Aω such that for all n < ω, (x n, b n) ∈ T .
Tree representations in general are usually not of interest, as any X ⊆ R has a
trivial tree representation on ω × R, for instance, defined by
(x, b) ∈ T ⇐⇒ x ∈ ω <ω and ∃x∗ ∈ X (b = (x∗ , x∗ , . . . , x∗ )∧|b|= |x|∧x = x∗ |x|).
If however we restrict the set A, we arrive at a much more interesting notion. For
κ an ordinal, we say that X ⊆ R is κ-Suslin if X has a tree representation on
ω × κ. Of course, if AC holds and κ ≥ |X|, this again is a rather uninteresting
notion. However, in the context of determinacy, these Suslin representations are
very important(see [KLS08]).
In the context of AC however, the notion of a Suslin representation can be
augmented with the use of measures. We then arrive at what are called homoge-
neously Suslin and weakly homogeneously Suslin representations (for a complete
introduction to these representations see [Lar04]).
For δ a cardinal and Z a set, we say that T is a δ-weakly homogeneous tree
if there exists a countable set σ of δ-complete measures on Z <ω such that for all
x ∈ p[T ], there exists a countably complete tower of measures22 μn | n < ω such
that for all n < ω, μn (Txn ) = 1. And X ⊆ R is δ-weakly homogeneously Suslin if
there is a δ-weakly homogeneous tree T such that A = p[T ].
We now define U (j)-representations, which are analogous to weakly homoge-
neously Suslin representations, but in the context of L(Vλ+1 ). The main difference
is that we restrict the measures which can be used to witness homogeneity to a
collection of measures called U (j)-measures. In addition we have to work with
the particularities of working with λ instead of ω; in particular there is no natural
cofinal sequence in λ.
We first introduce some terminology. We fix j : L(Vλ+1 ) → L(Vλ+1 ) an iterable
elementary embedding with crit (j) < λ. We say that a sequence a = an | n < ω
is weakly fixed by j if for all n < ω, |an |< λ, an ⊆ an+1 and there exists an m such
that j(m) (an ) = an . Here j(m) is the m-th iterate of j.
For S a set of embeddings we let
Fix(S) = {a| ∀k ∈ S (k(a) = a))},
and we set Fix(k) = Fix({k}).
For k an elementary embedding we denote by k(n) the nth iterate of k, and we
define for κ an ordinal

Fixκ (k) = {a ∈ Lκ (Vλ+1 )| k(a) = a}, Fixω
κ (k) = Fixκ (k(n) ),
n<ω
and let
E k (κ) = {k : Lκ (Vλ+1 ) → Lκ (Vλ+1 )| ∃n, m(k(n)

= k(m) )}
22 A tower of measures is such that each measure projects down to the previous measures, and
nth measure concentrates on n-sequences. So for instance A ∈ μ1 ⇐⇒ {t s | t ∈ A, s ∈ Z} ∈

μ2 . To be countable complete means that each sequence of measure one sets can be threaded.
Equivalently, the direct limit of the ultrapowers given by the measures is wellfounded. See [Lar04]
for more details.
244 SCOTT CRAMER
if k : Lκ (Vλ+1 ) → Lκ (Vλ+1 ) is elementary and iterable. Also for a ∈ Lκ (Vλ+1 ) let

E k (κ, a) = {k ∈ E k (κ)| k (a) = a}.
Note that if j(κ) = κ then j(Fixω κ (j)) = Fixκ (j). We then set F (κ, a) to be the
ω k
filter generated by the sets Fix(S) where S ∈ [E (κ, a)] .

k λ
For a ∈ [Lκ (Vλ+1 )]ω weakly fixed by j, we let Aj (κ, a) be the set of sequences
An | n < ω such that for all n < ω, An ∈ F j (κ, an ). We also set for A ∈ Aj (κ, a),
to be the largest tree T such that any node s of T is such that for all large
T F (A)
enough n, s ∈ An .
We now proceed to define the set of U (j)-measures and U (j)-representations.
Definition 46 (Woodin). Let U (j) be the set of U ∈ L(Vλ+1 ) such that in
L(Vλ+1 ) the following hold:
(1) U is a λ+ -complete ultrafilter.
(2) For some γ < Θ, U ∈ Lγ (Vλ+1 ).
(3) For all sufficiently large n < ω, j(n) (U ) = U and for some A ∈ U ,
{a ∈ A| j(n) (a) = a} ∈ U.
The next lemma gives us a way of generating lots of U (j) measures. The proof
is very much along the lines of the proof of Theorem 22. For each ordinal κ, let
ΘLκ (Vλ+1 ) denote the supremum of the ordinals α such that there is a surjection
ρ : Vλ+1 → α such that {(a, b)| ρ(a) < ρ(b)} ∈ Lκ (Vλ+1 ).
Lemma 47 (Woodin). Suppose κ < Θ, κ ≤ ΘLκ (Vλ+1 ) , a ∈ Lκ (Vλ+1 ) and that
j(κ, a) = (κ, a). Then there is δ < crit (j) and a partition {Sα | α < δ} ∈ L(Vλ+1 )
of Lκ (Vλ+1 ) into F j (κ, a)-positive sets such that for each α < δ,
F j (κ, a) Sα ∈ U (j).
Suppose that κ < Θ and κ ≤ ΘLκ (Vλ+1 ) and ai | i < ω is weakly fixed by j.
Let U (j, κ, ai | i < ω) denote the set of U ∈ U (j) such that there exists n < ω such
that for all k ∈ E j (κ, ai | i ≤ n),
Fix(k) ∈ U.
We can now define U (j)-representations for subsets of Vλ+1 .
Definition 48 (Woodin). Suppose κ < Θ, κ is weakly inaccessible in L(Vλ+1 ),
and ai | i < ω is an ω-sequence of elements of Lκ (Vλ+1 ) such that for all i < ω
there is an n < ω such that j(n) (ai ) = ai .
Suppose that Z ∈ L(Vλ+1 ) ∩ Vλ+2 . Then Z is U (j, κ, ai | i < ω)-representable
if there exists an increasing sequence λi | i < ω, cofinal in λ and a function

π : {Vλi +1 × Vλi +1 × {i}| i < ω} → U (j, κ, ai | i < ω)
such that the following hold:
(1) For all i < ω and (a, b, i) ∈ dom(π) there exists A ⊆ (L(Vλ+1 ))i such that
A ∈ π(a, b, i).
(2) For all i < ω and (a, b, i) ∈ dom(π), π(a, b, i) ∈ U (j, κ, ai ).23
23 This condition is slightly strengthened from the corresponding condition in [Woo11]. The
set of U (j)-representable sets is unchanged however. We make this restriction because of useful
properties exploited in [Cra15a].
(3) For all i < ω and (a, b, i) ∈ dom(π), if m < i then

(a ∩ Vλm , b ∩ Vλm , m) ∈ dom(π)
and π(a, b, i) projects to π(a ∩ Vλm , b ∩ Vλm , m).
(4) For all x ⊆ Vλ , x ∈ Z if and only if there exists y ⊆ Vλ such that
(a) for all m < ω, (x ∩ Vλm , y ∩ Vλm , m) ∈ dom(π),
(b) the tower
π(x ∩ Vλm , y ∩ Vλm , m)| m < ω
is well founded.
For Z ∈ L(Vλ+1 ) ∩ Vλ+2 we say that Z is U (j)-representable if there exists
(κ, ai | i < ω) such that Z is U (j, κ, ai | i < ω)-representable.
5.1.1. Propagation of U (j)-representations. First we give an outline of how
U (j)-representations have been propagated throughout L(Vλ+1 ). Then we give
details of some of the methods involved in this propagation.
It is relatively easy to see that U (j)-representations are closed under unions of
size λ and existential quantification (see [Woo11], Lemmas 114, 115). However,
the case of complements is much more complicated. In spite of this, Woodin was
able to first propagate U (j)-representations up to λ.
Theorem 49 (Woodin[Woo11], Theorem 134). Suppose j witnesses I0 holds
at λ. Then in L(Vλ+1 ) every set X ∈ Lλ (Vλ+1 ) ∩ Vλ+2 is U (j)-representable.
He also showed that a certain continuous ill-foundedness condition called the
Tower Condition (see Definition 50) implies that U (j)-representations are closed
under complements. We then showed [Cra15e] that the Tower Condition holds
in general for L(X, Vλ+1 ) assuming that X is Icarus. This result by a Theorem
of Woodin, pushed the U (j)-representations in L(Vλ+1 ) beyond λ+ . Beyond this
point, Woodin also defined a certain game (see Definition 54) on fixed points of
embeddings, which if one could prove had a high enough rank, then the set of
U (j)-representations could be propagated even further. We gave such an anal-
ysis[Cra15c], which effectively pushed the U (j)-representations beyond the first
Σ1 -gap (Theorem 60). Finally, in order to show that every subset of Vλ+1 in
L(Vλ+1 ) has a U (j)-representation, these representations were propagated along
with j-Suslin representations (see [Cra15a] and Section 5.3).
We now look at some of the details of these theorems and some of the methods
used in their arguments. First we start with the Tower Condition, which basically
says that given a collection of U (j)-measures there is a continuous choice of measure
one sets, such that if there is an illfounded tower, this is witnessed by the chosen
measure one sets.
Definition 50 (Woodin). Suppose A ⊆ U (j), A ∈ L(Vλ+1 ), and |A|≤ λ. The
Tower Condition for A is the following statement: There is a function F : A →
L(Vλ+1 ) such that the following hold:
(1) For all U ∈ A, F (U ) ∈ U .
(2) Suppose Ui | i < ω ∈ L(Vλ+1 ) and for all i < ω, there exists Z ∈ Ui such
that Z ⊆ L(Vλ+1 )i , Ui ∈ A, and Ui+1 projects to Ui . Then the tower
Ui | i < ω is wellfounded in L(Vλ+1 ) if and only if there exists a function
f : ω → L(Vλ+1 ) such that for all i < ω, f i ∈ F (Ui ).
246 SCOTT CRAMER
The Tower Condition for U (j) is the statement that for all A ⊆ U (j) if A ∈
L(Vλ+1 ) and |A|≤ λ then the Tower Condition holds for A.
Theorem 51. Let j : L(Vλ+1 ) → L(Vλ+1 ) be elementary such that crit (j) < λ.
Then the following hold:
(1) Tower Condition for U (j) holds in L(Vλ+1 ) (C. [Cra15e]).
(2) The set of U (j)-representable sets is closed under complements (Woodin
[Woo11]).
Proof sketch. In order to prove (1), for a λ-sized set A of U (j)-measures,
there is no inverse limit J such that for all U ∈ A, U is in the range of the extension
of J. However, we can almost achieve this in that any U ∈ A is in the range of the
extension of (j0 ◦ j1 ◦ · · · ◦ jn−1 )(Jn ) for some n. Having such a J, we pull back24 to
an Ā which has size λ̄J . Using that each measure in U (j) is λ+ complete, we can
show that Ā has the Tower Condition. A tower function for Ā can then by pushed
forward to obtain a tower function for A.
To prove (2), we look at the tree of attempts to continuously illfounded the
tower given by a U (j)-representation π.
In fact the proof holds more generally for X Icarus.
Theorem 52. Let j : L(X, Vλ+1 ) → L(X, Vλ+1 ) be a proper elementary em-
bedding such that crit (j) < λ. Then the following hold:
(1) Tower Condition for U (j) holds in L(X, Vλ+1 ) (C. [Cra15e]).
(2) The set of U (j)-representable sets in L(X, Vλ+1 ) is closed under comple-
ments (Woodin [Woo11]).
As a consequence of the Tower Condition, Woodin showed that U (j)-represen-
tations propagate beyond λ+ .
Theorem 53 (Woodin[Woo11]). Suppose there exists an elementary embed-
ding
j : L(X, Vλ+1 ) → L(X, Vλ+1 ).
Let Y be U (j)-representable in L(X, Vλ+1 ). Let κ = λ+ and set
η = sup{(κ+ )L[A] | A ⊆ λ}.
Then every set
Z ∈ Lη (Y, Vλ+1 ) ∩ Vλ+2
is U (j)-representable in L(X, Vλ+1 ).
To propagate U (j)-representations beyond this point we now look at the fol-
lowing game on fixed points of embeddings, introduced by Woodin.
Definition 54. Suppose γ < ΘL(Vλ+1 ) , j(γ) = γ,
E j (γ) := E jLγ (Vλ+1 ) (γ),
and
ai | i < ω ∈ (Lγ (Vλ+1 ))ω
and we have:
(1) γ ≤ ΘLγ (Vλ+1 ) ,
24 We do not actually need to reflect, which is the reason why we can prove this result for
L(X, Vλ+1 ).
(2) for all i < ω, ai ⊆ ai+1 ⊆ γ and |ai |< λ,

(3) for all i < ω, there exists an n < ω such that j(n) (ai ) = ai .
Then let G(j, γ, ai | i < ω) denote the following game. Player I plays a sequence
+ + , ,
(γi , bim : m < ω ) : i < ω
and player II plays a sequence Ei : i < ω such that the following hold:
(1) Ei ⊆ E j (γi ), |Ei |≤ λ, and for each k ∈ Ei there exists m < ω such that
k(bim ) = bim .
(2) γ0 = γ, γi+1 < γi and there exists m < ω such that
k(bim ) = bim ⇒ k(γi+1 ) = γi+1
for all k ∈ Ei .
+ 0 all i < ω,, γi ≤ Θ i
Lγ (Vλ+1 )
(3) for ,
(4) bm : m < ω = am : m < ω.
(5) for all m < ω, bim ⊆ bim+1 ⊆ γi and |bim |< λ
(6) for all m < ω there exists m∗ < ω such that
k(bim∗ ) = bim∗ ⇒ k(bi+1 i+1
m ) = bm
for all k ∈ Ei .
Of course II always wins this game, but we are interested in the rank of this
game, which we define as follows.
Definition 55. Let
Gδ (j, γ, ai | i < ω)
have the same definition as G(j, γ, ai | i < ω) except that II must also play ordinals
δ0 > δ1 > · · · such that δ0 < δ. Then if δ is least such that II has a quasi-winning
strategy in Gδ (j, γ, ai | i < ω), then we set δ = rank(j, γ, ai | i < ω).
Theorem 58 shows that for any δ < Θ we can find γ and ai | i < ω such that
rank(j, γ, ai | i < ω) ≥ δ.
That is, the rank of this game can be made arbitrarily large by an appropriate
choice of parameters. We give some definitions now so that we can give sufficient
criteria for the rank to be large.
Definition 56. Suppose γ < ΘL(Vλ+1 ) , S ⊆ Lγ (Vλ+1 ), and ai | i < ω ∈
(Lγ (Vλ+1 ))ω and we have:
(1) γ ≤ ΘLγ (Vλ+1 ) ,
(2) for all i < ω, ai ⊆ ai+1 ⊆ γ and |ai |< λ,
i < ω, ai ∈ Fixγ+1 (j).
ω
(3) for all
(4) S = i<ω ai .
Then we say that ai | i < ω is a j-stratification of S.
Suppose that j : L(Vλ+1 ) → L(Vλ+1 ). Note that for all S ⊆ Fixω Θ (j) such that
|S|≤ λ, there is a γ < Θ and a ai | i < ω ∈ (Lγ (Vλ+1 ))ω such that ai | i < ω is
a j-stratification of S. Hence for any γ < λ+ , if ai | i < ω is a j-stratification of
γ, then rank(j, γ, ai | i < ω) = γ. An instructive example then is to show that
rank(j, λ+ , ∅) = λ+ , which we leave to the reader.
Definition 57. Fix κ < Θ good with cof(κ) > λ. Let S ⊆ κ such that |S|≤ λ.
Then we say that S is λ-threaded if the following hold:
248 SCOTT CRAMER
(1) Suppose α < sup S is such that there exists β ∈ S <ω and a ∈ Vλ such
that α is definable over Lκ (Vλ+1 ) from β and a. Then α ∈ S.
(2) Suppose α ∈ S and α is a limit ordinal with cof(α) < λ. Then S ∩ α is
cofinal in α.
Theorem 58. Let j : L(Vλ+1 ) → L(Vλ+1 ) be elementary. Fix κ < Θ good in
L(Vλ+1 ). Suppose that S has a largest element α0 , S is λ-threaded, and ai | i < ω
is a j-stratification of S. Then rank(j, κ + α0 , a) ≥ α0 .
Proof sketch. The proof uses a generalization of the fact that for an I0
embedding j, and Mω the ωth iterate of j LΘ (Vλ+1 ), and j0,ω the natural embed-
ding into Mω , we have that j(j0,ω (a)) = a for any a ∈ LΘ (Vλ+1 ). In particular we
generalize this to λ-sequences of embeddings to find common fixed points among
them. This involves defining an appropriate notion of inverse limit for λ-sequences
of embeddings. See [Cra15c] for the details of this construction.
We immediately have the following corollary, which gives the following result
on U (j)-representations by Theorem 148 of [Woo11].
Corollary 59. Suppose there exists an elementary embedding j : L(Vλ+1 ) →
L(Vλ+1 ) with crit j < λ. Then the supremum of rank(j, κ, a) for all possible κ and
a is Θ.
Theorem 60. Assume there exists an elementary embedding j : L(Vλ+1 ) →
L(Vλ+1 ) with crit j < λ. Let κ be least such that
V ∪{Vλ+1 }
Lκ (Vλ+1 ) ≺1 λ+1 Lκ+1 (Vλ+1 ).
Then for all sets X ⊆ Vλ+1 such that X ∈ Lκ (Vλ+1 ), X is U (j)-representable in
L(Vλ+1 ).
Finally, by propagating U (j)-representations along with j-Suslin representa-
tions (Section 5.3), we showed the following in [Cra15a].
Theorem 61. Assume j witnesses that I0 holds at λ. Then in L(Vλ+1 ) every
set X ∈ Vλ+2 is U (j)-representable.
We will discuss some of the details of this proof in Section 5.3.
5.1.2. Consequences of U (j)-representations. Two of the most important con-
sequences of U (j)-representations are (1) a certain generic absoluteness theorem
for certain forcings between Mω and V , and (2) obtaining weakly homogeneously
Suslin representations after collapsing λ to ω. The generic absoluteness result al-
lows us to obtain structural consequences for L(Vλ+1 ), and the fact that we can
obtain weakly homogeneously Suslin sets allows us to connect the study of L(Vλ+1 )
to models of determinacy.
We start with a generic absoluteness result due to Woodin, and then we show
some of its consequences.
Theorem 62 (Woodin). 25 Suppose j witnesses that I0 holds at λ and j is
iterable. Let j0,ω : L(Vλ+1 ) → Mω be the embedding into the ωth iterate of L(Vλ+1 )
25 Noticehere that VλMω = Vλ . Also, this theorem does not actually follow directly from the
existence of U (j)-representations, but rather it follows from the existence of a uniform version of
a U (j)-representation.
by j. Suppose P ∈ j0,ω (Vλ ), g ∈ V is P-generic over Mω , and (cof(λ))Mω [g] = ω.

Then we have for any α < λ an elementary embedding
Lα (Mω [g] ∩ Vλ+1 ) → Lα (Vλ+1 )
which is the identity below λ.
We now isolate the generic absoluteness property present in this theorem.
Definition 63. Suppose j witnesses that I0 holds at λ and j is iterable. Let
j0,ω : L(Vλ+1 ) → Mω
be the embedding into the ωth iterate of L(Vλ+1 ) by j. We say that generic
absoluteness holds between Mω and Lα (Vλ+1 ) if for some ᾱ we have the following.
Suppose P ∈ j0,ω (Vλ ), g ∈ V is P-generic over Mω , and (cof(λ))Mω [g] = ω. Then
there is an elementary embedding
Lᾱ (Mω [g] ∩ Vλ+1 ) → Lα (Vλ+1 )
which is the identity below λ.
We could then rephrase Theorem 62 as stating that if I0 holds at λ as witnessed
by j, then for any α < λ, generic absoluteness holds between Mω and Lα (Vλ+1 ).
We will see below that there are in fact other representations which imply this
generic absoluteness property.
We mention a couple of consequences of this theorem. Theorem 34 follows from
Theorem 62 and we also obtain the following theorem on the failure of SCH at λ.
Theorem 64 (Dimonte-Friedman[DF14], Woodin independently). Suppose I0
holds at λ. Then it is consistent that I1 holds at λ and the singular cardinal hy-
pothesis fails at λ.
Proof sketch. Let j be an I0 embedding. Force so that GCH fails at crit (j)
and so that the generic g is mapped coherently by j, so that we may extend j to
the extension. Let j ∗ be the natural extension of j. In the extension let Mω by
the ωth iterate of L(V [g]λ+1 ) by j ∗ . Then Mω satisfies that GCH fails at λ. Let
κ be the critical sequence of j ∗ . Then Mω [κ] satisfies that SCH fails at λ, and by
Theorem 62, I1 holds at λ.
Note that in defining an AD-like axiom from U (j)-representations it is the
generic absoluteness property which seems to imply regularity properties for sub-
sets of Vλ+1 . And this generic absoluteness property does not obviously follow
in the same way in the context of L(X, Vλ+1 ), assuming the existence of uniform
U (j)-representations. Therefore the AD-like axiom would seem to be the generic ab-
soluteness result itself, unless a general generic absoluteness result could be proven
from the existence of uniform U (j)-representations.
We now show that U (j)-representations can be used to connect to models of
determinacy when collapsing λ to ω. To see this, let Γ∞ be the set of universally
Baire sets of reals. The determinacy axiom LSA states that the largest Suslin
cardinal exist and is a Θα , that is a member of the Solovay sequence. Corollary
168 of [Woo11] together with Theorem 61 gives us the following theorem.
Theorem 65. Suppose that λ is a limit of supercompact cardinals and there is
a proper class of Woodin cardinals. Suppose that I0 holds at λ. Let G ⊆ Coll(ω, λ)
250 SCOTT CRAMER
be V -generic. Then for

Γ∞ ∞ V [G]
G = (Γ ) ∩ L(Vλ+1 )[G]
we have that L(Γ∞
G ) satisfies LSA.
Although G. Sargsyan recently showed that Con(LSA) follows from a Woodin

limit of Woodins, the above theorem gives an alternative proof of Con(LSA) from
large cardinals. We also obtain the following theorem, which shows a strong rela-
tionship between L(Vλ+1 ) and models of determinacy after collapsing λ to ω.
Theorem 66. Suppose that λ is a limit of supercompact cardinals and there is
Γ∞ ∞ V [G]
G = (Γ ) ∩ L(Vλ+1 )[G]
we have that
∞
ΘL(Vλ+1 ) = ΘL(ΓG ) .
Proof. To see that
∞
ΘL(Vλ+1 ) ≤ ΘL(ΓG )
let α < Θ and let X ⊆ Vλ+1 code a prewellordering of Vλ+1 of ordertype at least α.
We have that X is U (j)-representable in L(Vλ+1 ), which implies that in L(Vλ+1 )[G],
X can be coded as a subset Y ⊆ RL(Vλ+1 )[G] such that Y is weakly homogeneously
Suslin. Hence since there is a proper class of Woodin cardinals, Y is universally
∞
Baire. And hence α < ΘL(ΓG ) .
To see that
∞
ΘL(Vλ+1 ) ≥ ΘL(ΓG )
we show that in L(Vλ+1 )[G] there is no surjection f : Vλ+1 → ΘL(Vλ+1 ) . Suppose
this is not the case, and let τ ∈ L(Vλ+1 ) be a term for f . Then we have that
g : Vλ+1 × Coll(ω, λ) → ΘL(Vλ+1 ) defined by g(x, p) = α iff p τ (x) = α is clearly
a surjection onto ΘL(Vλ+1 ) . But g ∈ L(Vλ+1 ), which is a contradiction. Hence the
theorem follows.
This theorem gives some evidence that the connection between large cardinals
and determinacy axioms holds even up to the level of very large cardinals. However,
the nature of this connection is far from understood at this point. For instance we
have the following open question of Woodin.
Question 67. Suppose that λ is a limit of supercompact cardinals and there is
Γ∞ ∞ V [G]
G = (Γ ) ∩ L(Vλ+1 )[G]
what is the largest Suslin cardinal of L(Γ∞
G )?
Finally, we mention one negative result which is that uniformization does not
follow from U (j)-representability. This is a direct corollary of Theorem 61 and Fact
38.
Theorem 68 (C.). Assume I0 holds at λ. Then in L(Vλ+1 ) there is a relation
R ⊆ Vλ+1 × Vλ+1 which is U (j)-representable, but has no uniformization.
5.2. Inverse limit reflection. In this section we introduce the axioms inverse
limit reflection and strong inverse limit reflection. These axioms arose out of the
reflection results in Section 3 and the structural results in Section 4. In order to
state these axioms we will need to use the terminology introduced in Section 3. In
addition we make the definition for E saturated that
CL(E) = {(J, j) ∈ E| ∀n < ω ∃(K, k) ∈ E (k n = j n)}.
Also let E be the set of inverse limits (J, j).
Definition 69. We define inverse limit reflection at α to mean the following:
There exists λ̄, ᾱ < λ and a saturated set E ⊆ E such that for all (J, j) ∈ E, J
extends to Jˆ : Lᾱ (Vλ̄+1 ) → Lα (Vλ+1 ) which is elementary.
We define strong inverse limit reflection at α to mean the following: There
exists λ̄, ᾱ < λ and a saturated set E ⊆ E such that for all (J, j) ∈ CL(E), J
extends to Jˆ : Lᾱ (Vλ̄+1 ) → Lα (Vλ+1 ) which is elementary.
We can also make this definition relativized to some X ⊆ Vλ+1 . To do this we
let
E(X) = {(J, ji | i < ω)| ∀i (ji : (Vλ+1 , X) → (Vλ+1 , X)) and
J = j0 ◦ j1 ◦ · · · : (Vλ̄+1 , X̄) → (Vλ+1 , X) is Σ0 }.
Here we let X̄ = J −1 [X]. We modify the definition of saturated to X-saturated,
requiring in addition that J −1 [X] = K −1 [X].
Definition 70. Suppose X ⊆ Vλ+1 . We define inverse limit X-reflection at α
to mean the following: There exists λ̄, ᾱ < λ, X̄ ⊆ Vλ̄+1 and an X-saturated set
E ⊆ E(X) such that for all (J, j) ∈ E, J extends to Jˆ : Lᾱ (X̄, Vλ̄+1 ) → Lα (X, Vλ+1 )
which is elementary.
We define strong inverse limit X-reflection at α to mean the following: There
exists λ̄, ᾱ < λ, X̄ ⊆ Vλ̄+1 and an X-saturated set E ⊆ E(X) such that for all
(J, j) ∈ CL(E), J extends to Jˆ : Lᾱ (X̄, Vλ̄+1 ) → Lα (X, Vλ+1 ) which is elementary.
We will say strong inverse limit X-reflection for strong inverse limit X-reflection
at ω, and similarly for the other terminology.
The following result was shown in [Cra15e].
Theorem 71. Suppose that there exists an elementary embedding
j : LΘ (Vλ+1 ) → LΘ (Vλ+1 ).
Then inverse limit reflection holds at α for all α < Θ.
It was subsequently shown in [Cra15d] that:
Theorem 72. For any good α < ΘL(Vλ+1 ) , if there exists an elementary em-
bedding
Lα+ω (Vλ+1 ) → Lα+ω (Vλ+1 )
then strong inverse limit reflection holds at α.
These results in fact extend slightly beyond I0 by the same proof to get the
following.
252 SCOTT CRAMER
#
Theorem 73. For any good α < ΘL(Vλ+1 ,Vλ+1 ) , if there exists an elementary
embedding
# #
Lα+ω (Vλ+1 , Vλ+1 ) → Lα+ω (Vλ+1 , Vλ+1 )
#
then strong inverse limit Vλ+1 -reflection holds at α.
Some of the methods needed for proving these theorems were discussed in
Section 3. Theorem 72, however, requires a more detailed analysis of how inverse
limits behave when passing to limit roots. We refer the reader to [Cra12] for
a detailed analysis of inverse limit reflection results throughout L(Vλ+1 ). A key
question is whether these results can be extended way beyond I0 through the Eα0
hierarchy (see Section 6).
5.2.1. Consequences of inverse limit reflection. The reason for defining strong
inverse limit X-reflection is its consequences for the properties of X.
Theorem 74 (C. [Cra15e]). Let X ⊆ Vλ+1 and suppose that strong inverse
limit X-reflection holds. Then the following hold:
(1) there are no disjoint sets S1 , S2 ∈ Lω (X, H(λ+ )) such that S1 , S2 ⊆ λ+
and both S1 and S2 are stationary (in V );
(2) every set X ∈ Lω (X, Vλ+1 ) ∩ Vλ+2 either has size λ or contains a λ-
splitting perfect set.
Theorem 74 gives very strong consequences of strong inverse limit X-reflection
for sets X ⊆ Vλ+1 , and it is this theorem which seems to give the most compelling
AD-like properties of any axiom defined for the context of Vλ+1 so far.
We also have the following important corollary.
Theorem 75. Suppose I0 is consistent. Then it is consistent that for some
X ⊆ Vλ+1 , X is Icarus, but strong inverse limit X-reflection does not hold.
Proof. This follows immediately from Theorem 36.
By Fact 38 we also have the following.
Corollary 76. Suppose I0 holds at λ. Then in L(Vλ+1 ) there is R ⊆ Vλ+1 ×
Vλ+1 such that strong inverse limit R-reflection holds, but R has no uniformization.
On the other hand, it is not clear that any generic absoluteness results follow
directly from strong inverse limit X-reflection, which is one of the most compelling
consequences of the existence of U (j)-representations. Hence we are left with the
following question.
Question 77. Does generic absoluteness between Mω and Lα (Vλ+1 ) follow
directly from strong inverse limit reflection at α?
5.3. j-Suslin representations. In [Cra15b] we defined tree representations
for subsets of Vλ+1 which also give generic absoluteness between Mω and L(Vλ+1 ).
These tree representations seem rather similar to Suslin representations in the sense
that they are not augmented by measures. On the other hand, they are weaker than
Suslin representations in the sense that instead of a tree on ordinals, they are given
by a tree on fixed points of iterates of a fixed embedding j. This definition is
very much motivated by U (j)-representations, which, through the Tower Condi-
tion, also give such trees. However, we must add an additional requirement on the
tree structure itself in order to obtain a non-trivial representation and our generic
absoluteness result. These representations, which we call j-Suslin representations,

are therefore ostensibly not directly related to U (j)-representations. At the end
of this section we will define what is very naturally a combination of j-Suslin and
U (j)-representations, which we call a weakly homogeneously j-Suslin representa-
tion. This stronger representation in fact does immediately give the two weaker
representations.
The main point of defining these representations was in order to help in the
propagation of U (j)-representations. However, we will see that j-Suslin representa-
tions (and therefore weakly homogeneously j-Suslin representations) give a slightly
stronger generic absoluteness result.
For this section we fix j : L(Vλ+1 ) → L(Vλ+1 ) an elementary embedding with
crit (j) < λ. For k an elementary embedding we denote by k(n) the nth iterate of
k, and as before we define Fixκ (k), E k (κ), and E k (κ, a).
Definition 78. For κ = κi | i < ω increasing and cofinal in λ, we let W κ be

the set of sequences s ∈ Vλω such that
(1) for some n < ω, |s|= n and for all i < n, s(i) ⊆ Vκi ,
(2) if i ≤ m < |s| then s(i) = s(m) ∩ Vκi .
Also let Wnκ = {s ∈ W κ | |s|= n}. In this context if x ∈ Vλ+1 , we set
x̂ = x̂κ = x ∩ Vκn | n < ω ∈ W κ ,
where we use the first notation if the sequence κ is understood.
Suppose that κ < Θ. Let X ⊆ Vλ+1 . We say that T is a (j, κ)-Suslin rep-
resentation for X if for some sequence κi | i < ω increasing and cofinal in λ the
following hold.
(1) T is a (height ω) tree on Vλ ×Fixω κ (j) such that for all (s, a) ∈ T , s ∈ W|s| .

κ
(2) For all s ∈ W κ , Ts := {a| (s, a) ∈ T } ∈ FixΘ (j).

ω
(3) For all x ∈ Vλ+1 , x ∈ X iff Tx̂ := {a| ∃n < ω (a ∈ Tx̂n )} is illfounded.

We say that X is j-Suslin if for some κ, X has a (j, κ)-Suslin representation. If T
satisfies conditions 1 and 3 then we say that T is a weak (j, κ)-Suslin representation
for X.
By definition of the Tower Condition, if π is a U (j)-representation for X and
F is a tower function for rng π, then we immediately obtain a weak j-Suslin rep-
resentation for X from F . However, obtaining weak j-Suslin representations is not
particularly difficult (and apparently not useful). To see this consider the pointwise
image of a set X ⊆ Vλ+1 under the map j0,ω of L(Vλ+1 ) to Mω . Every element of
Mω ∩ LΘ (Vλ+1 ) is fixed by an iterate of j, for j an I0 embedding, and hence we
can obtain from this pointwise image a weak j-Suslin representation for X. We will
have to work considerably harder to obtain j-Suslin representations.
Similarly we say that T is a uniform (j, κ)-Suslin representation for X if the
following hold.
(1) T is a function on [λ]<ω such that for all s ∈ [λ]ω , if T (s) is the tree
whose nth level is given by T (s n), then T (s) is a (height ω) tree on
Vλ × Fixω κ (j).
(2) For all s ∈ [λ]ω such that s is cofinal in λ, T (s) is a (j, κ)-Suslin represen-
tation for X.
254 SCOTT CRAMER
We now state the generic absoluteness results which follow from the existence
of uniform j-Suslin representations. In order to do this, we make the following
definition about the existence of uniform j-Suslin representations.
Definition 79. Suppose that j : L(Y, Vλ+1 ) → L(Y, Vλ+1 ) is an I0 (Y ) elemen-
tary embedding. We say that the uniform j-Suslin conjecture holds in L(Y, Vλ+1 )
if the following holds. Suppose X ⊆ Vλ+1 , X ∈ L(Y, Vλ+1 ) is such that there is an
a ∈ Fixω 0 (j) and a δ such that X is definable over Lδ (Vλ+1 ) from a (so X ∈ FixΘ (j)).
ω
Then for some κ < Θ there exists a uniform (j, κ)-Suslin representation T for X
such that T ∈ Fixω Θ (j).
The following is proved in [Cra15a], which shows that uniform j-Suslin repre-
sentations do exist in L(Vλ+1 ) under I0 .
Theorem 80. Suppose that I0 holds at λ. Then the uniform j-Suslin conjecture
holds in L(Vλ+1 ). In fact if I0# holds at λ then the uniform j-Suslin conjecture holds
#
in L(Vλ+1 ).
Using this result, the following generic absoluteness results are shown in [Cra15b].
Theorem 81. Suppose that I0 holds at λ as witnessed by j. Then for κ the
critical sequence of j, if α < Θ is good then for some ᾱ < λ there is an elementary
embedding
Lᾱ (Mω [κ] ∩ Vλ+1 ) → Lα (Vλ+1 ).
Theorem 82. Suppose that I0 holds at λ as witnessed by j. Suppose g ∈ V
is P-generic over Mω where P ∈ Mω . Also assume that cof(λ)Mω [g] = ω. Then if
α < Θ is good, for some ᾱ < λ there is an elementary embedding
Lᾱ (Mω [g] ∩ Vλ+1 ) → Lα (Vλ+1 ).
Note that this theorem gives a slightly stronger version of generic absoluteness
than the definition of generic absoluteness between Mω and L(Vλ+1 ) which we
made above. In particular we do not require that P ∈ j0,ω (Vλ ). This is potentially
a significant strengthening as for instance we cannot have j ∈ Mω [g] where g is
Mω -generic for a forcing P ∈ j0,ω (Vλ ). This reasoning however does not apply to
forcings in Mω .
Now we wish to define the notion of a weakly homogeneously j-Suslin rep-
resentation. This representation combines properties of a j-Suslin representation
and a U (j)-representation, and is important for [Cra15a] when propagating these
representations throughout L(Vλ+1 ). We use the same terminology as in Section
5.1.
Definition 83. Let X ⊆ Vλ+1 . We say that T is a weakly (δ, a)-homogeneously
(j, κ)-Suslin representation for X if T is a (j, κ)-Suslin representation for X, and
the following hold.
(1) a ∈ [Lκ (Vλ+1 )]ω is weakly fixed by j.
(2) For all x ∈ Vλ+1 and A ∈ Aj (δ, a), if [Tx̂ ] = ∅ then
= ∅.
[Tx̂ ] ∩ [T F (A)]
We say that X is weakly homogeneously j-Suslin if for some δ, a and κ, X is
weakly (δ, a)-homogeneously (j, κ)-Suslin.
Theorem 84 (C. [Cra15a]). Assume that j : L(Vλ+1 ) → L(Vλ+1 ) is an I0

embedding. Then every subset X ⊆ Vλ+1 such that X ∈ L(Vλ+1 ) satisfies the
following in L(Vλ+1 ).
(1) X is U (j)-representable.
(2) X is j-Suslin.
(3) X is weakly homogeneously j-Suslin.
The proof of this theorem has many similarities to the propagation of scales in
L(R). In particular there is a version of a closed game representation (see [MS83]
and [Ste83]) which is sufficiently weaker so that every subset of Vλ+1 in L(Vλ+1 )
has such a representation. This fact together with the results of [Cra15c] are the
key elements in the proof.
One unfortunate aspect of Theorem 84 is that the result is not local, especially
not as we saw of strong inverse limit reflection above (Theorem 72). While obtaining
U (j)-representations locally is basically impossible because the measures do not
occur locally, this is not obviously the case for j-Suslin representations. On the
other hand, the definition of U (j)-representations could be weakened in order to
allow for such a local result. We state these ideas in the following question.
Question 85. (1) Is there a local version of Theorem 84?
(2) Do U (j), j-Suslin, and weakly homogeneously j-Suslin representations fol-
low directly from strong inverse limit reflection?
An affirmative answer to (2) would presumably simplify the propagation of
these representations dramatically, and would perhaps answer (1) at the same time.
6. Beyond I0
In the previous sections we for the most part concentrated on consequences of
I0 and I0# , and so a natural question is to what extent these results extend beyond
this point.26 A natural way of framing this question is through the AD-like axioms
which we considered in Section 5. So we could ask, for which X ⊆ Vλ+1 is it
consistent that L(X, Vλ+1 ) satisfies these AD-like axioms? As we saw, however, we
can find examples of X ⊆ Vλ+1 for which these AD-like axioms fail in L(X, Vλ+1 ),
and so the answer to this question will most likely be rather complicated.
On the other hand, there is a natural hierarchy of axioms beyond I0 , introduced
by Woodin, which do seem like good candidates for the propagation of these AD-like
axioms. This hierarchy, called the Eα0 -hierarchy, was inspired by the hierarchy of
models above L(R) which give the minimal model of ADR . Very roughly speaking,
0
the sequence Eα0 is defined by progressively increasing the size of ΘL(Eα ) , and Eα0
is such that
Eα0 = Vλ+2 ∩ L(Eα0 ).
This is achieved by either adding in the sharp of the previous model, adding in
λ-sequences of the previous model, or adding in a predicate for the ω-club filter on
Θ of the previous model. Alternatively the next model can be defined using the set
of elementary embeddings on the previous model. A key feature of this hierarchy
is that it satisfies absoluteness properties similar to L(Vλ+1 ).
26 Of course another question is how to define such large cardinals.

256 SCOTT CRAMER
The analysis of the Eα0 -hierarchy is beyond the scope of this article, and we refer
the interested reader to [Woo11], [Dim11], and [DF14], and state the following
question mentioned above.
Question 86. Do the AD-like axioms of Section 5 (namely, U (j)-represen-
tability, (strong) inverse limit reflection, j-Suslin representability, weakly homo-
geneously j-Suslin representability, generic absoluteness) propagate throughout the
Eα0 -hierarchy?
We close with one additional question, first posed by Woodin, which seems to
be at the heart of the dynamics of passing to larger cardinals beyond I0 .
Question 87. Suppose that j is an I0 embedding. Then does the following
necessarily hold
L(j LΘλ (Vλ+1 ), Vλ+1 ) ∩ Vλ+2 = L(Vλ+1 ) ∩ Vλ+2 ?
If the answer is yes, and this property holds more generally, there would perhaps
be a straightforward way of building a hierarchy beyond I0 , by building models
which include the predicate for j up to some point. Understanding this question
is therefore seemingly essential to understanding how (and to what extent) the
hierarchy of large cardinals proceeds beyond I0 .
References
[Cra12] Scott Stefan Cramer, Inverse Limit Reflection and the Structure of L(Vλ+1 ), Pro-
Quest LLC, Ann Arbor, MI, 2012. Thesis (Ph.D.)–University of California, Berkeley.
MR3152601
[Cra15a] Scott Cramer. Existence of tree representations in L(Vλ+1 ). Submitted, 2015.
[Cra15b] Scott Cramer. Generic absoluteness from tree representations in L(Vλ+1 ). Submitted,
2015.
[Cra15c] Scott Cramer. Rank of fixed point measures. Submitted, 2015.
[Cra15d] Scott Cramer. Strong inverse limit reflection. Submitted, 2015.
[Cra15e] Scott S. Cramer, Inverse limit reflection and the structure of L(Vλ+1 ), J. Math. Log.
15 (2015), no. 1, 1550001, 38, DOI 10.1142/S0219061315500014. MR3367775
[Dav64] Morton Davis, Infinite games of perfect information, Advances in game theory, Prince-
ton Univ. Press, Princeton, N.J., 1964, pp. 85–101. MR0170727
[DF14] Vincenzo Dimonte and Sy-David Friedman, Rank-into-rank hypotheses and the failure of
GCH, Arch. Math. Logic 53 (2014), no. 3-4, 351–366, DOI 10.1007/s00153-014-0369-8.
MR3194669
[Dim11] Vincenzo Dimonte, Totally non-proper ordinals beyond L(Vλ+1 ), Arch. Math. Logic 50
(2011), no. 5-6, 565–584, DOI 10.1007/s00153-011-0232-0. MR2805297
[Gai74] Haim Gaifman, Elementary embeddings of models of set-theory and certain subtheories,
Axiomatic set theory (Proc. Sympos. Pure Math., Vol. XIII, Part II, Univ. Califor-
nia, Los Angeles, Calif., 1967), Amer. Math. Soc., Providence R.I., 1974, pp. 33–101.
MR0376347
[HKP12] Joel David Hamkins, Greg Kirmayer, and Norman Lewis Perlmutter, Generalizations of
the Kunen inconsistency, Ann. Pure Appl. Logic 163 (2012), no. 12, 1872–1890, DOI
10.1016/j.apal.2012.06.001. MR2964876
[Kan94] Akihiro Kanamori, The higher infinite, Perspectives in Mathematical Logic, Springer-
Verlag, Berlin, 1994. Large cardinals in set theory from their beginnings. MR1321144
[KLS08] Alexander S. Kechris, Benedikt Löwe, and John R. Steel (eds.), Games, scales, and
Suslin cardinals. The Cabal Seminar. Vol. I, Lecture Notes in Logic, vol. 31, 2008.
MR2463612
[Kun71] Kenneth Kunen, Elementary embeddings and infinitary combinatorics, J. Symbolic
Logic 36 (1971), 407–413. MR0311478
[KW10] Peter Koellner and W. Hugh Woodin, Large cardinals from determinacy, Handbook of
set theory. Vols. 1, 2, 3, Springer, Dordrecht, 2010, pp. 1951–2119, DOI 10.1007/978-1-
4020-5764-9 24. MR2768702
[Lar04] Paul B. Larson, The stationary tower, University Lecture Series, vol. 32, American
MR2069032
[Lav97] Richard Laver, Implications between strong large cardinal axioms, Ann. Pure Appl. Logic
90 (1997), no. 1-3, 79–90, DOI 10.1016/S0168-0072(97)00031-6. MR1489305
[Lav01] Richard Laver, Reflection of elementary embedding axioms on the L[Vλ+1 ] hierarchy,
Ann. Pure Appl. Logic 107 (2001), no. 1-3, 227–238, DOI 10.1016/S0168-0072(00)00035-
X. MR1807846
[Lav02] R. Laver, On very large cardinals, Paul Erdős and his mathematics, II (Budapest, 1999),
Bolyai Soc. Math. Stud., vol. 11, János Bolyai Math. Soc., Budapest, 2002, pp. 453–469.
MR1954738
[LM13] Richard Laver and Sheila K. Miller, The free one-generated left distributive algebra:
basics and a simplified proof of the division algorithm, Cent. Eur. J. Math. 11 (2013),
no. 12, 2150–2175, DOI 10.2478/s11533-013-0290-0. MR3111712
[MS83] Donald A. Martin and John R. Steel, The extent of scales in L(R), Cabal semi-
nar 79–81, Lecture Notes in Math., vol. 1019, Springer, Berlin, 1983, pp. 86–96, DOI
10.1007/BFb0071697. MR730590
[Shi15] Xianghui Shi, Axiom I0 and higher degree theory, J. Symb. Log. 80 (2015), no. 3, 970–
1021, DOI 10.1017/jsl.2015.15. MR3395357
[Sol78] Robert M. Solovay, The independence of DC from AD, Cabal Seminar 76–77 (Proc.
Caltech-UCLA Logic Sem., 1976), Lecture Notes in Math., vol. 689, Springer, Berlin,
1978, pp. 171–183. MR526918
[Ste83] John R. Steel, Scales in L(R), Cabal seminar 79–81, Lecture Notes in Math., vol. 1019,
Springer, Berlin, 1983, pp. 107–156, DOI 10.1007/BFb0071699. MR730592
[Ste95] John R. Steel, HODL(R) is a core model below Θ, Bull. Symbolic Logic 1 (1995), no. 1,
75–84, DOI 10.2307/420947. MR1324625
[Woo11] W. Hugh Woodin, Suitable extender models II: beyond ω-huge, J. Math. Log. 11 (2011),
no. 2, 115–436, DOI 10.1142/S021906131100102X. MR2914848
Department of Mathematics, Rutgers University, Hill Center - Busch Campus, 110

Frelinghuysen Road, Piscataway, New Jersey 08854-8019
Current address: Department of Mathematics, California State University San Bernardino,
5500 University Parkway, San Bernardino CA 92407-2318
E-mail address: cramer.scott@gmail.com
Volume 690, 2017
What makes the continuum ℵ2
Justin Tatch Moore

This article is dedicated to Hugh Woodin on the occasion of his 60th birthday.
Contents
1. Introduction
2. The Continuum Hypothesis, the Perfect Set Property,
and the Open Coloring Axiom
3. Consequences of the Continuum Hypothesis
4. Woodin’s Σ21 -Absoluteness Theorem
5. Todorcevic’s Open Coloring Axiom
6. A Ramsey-theoretic proof that the continuum is ℵ2
7. Forcing axioms and generic absoluteness
8. Stationary set reflection and 2ℵ0 ≤ ℵ2
9. The Conjectures of Chang and Rado
10. Woodin’s Pmax -extension of L(R)
11. Simply definable well orderings of R
12. Iterated forcing and the Continuum Hypothesis
13. The Semifilter Trichotomy
14. Open problems
References
1. Introduction
The question of how many real numbers there are traces all the way back to
the beginning of set theory. It has played an important role in the development
and modernization of the subject throughout the last century. Cantor, who was the
first to consider this problem [24], was most interested in the question of whether
or not |R| = ℵ1 , also known as the Continuum Hypothesis (CH). Of course one can
2010 Mathematics Subject Classification. Primary 03E50,; Secondary 03E35, 03E57, 03E65.
Key words and phrases. Chang’s Conjecture, Continuum hypothesis, continuum problem,
forcing axiom, Martin’s Maximum, Near Coherence of Filters, Open Coloring Axiom, P-Ideal
Dichotomy, Rado’s Conjecture, reflection principle, Semi-Filter Trichotomy, Woodin.
While several people read early drafts of this paper and offered suggestions and corrections,
Matthew Foreman and Stevo Todorcevic deserve special thanks for their comments, especially
concerning historical aspects of the discussion. The preparation of this article was supported in
part by NSF grant DMS–1262019.
2017
259
260 JUSTIN TATCH MOORE
more generally ask how the cardinality of the continuum relates to any particular
value of the ℵ function. A remarkable empirical discovery over the last 50 years
has been that the assertion |R| = ℵ2 is quite rich in its consequences, while up to
the present, the same cannot be said for any other ℵξ . It is this phenomenon which
will occupy our attention for the duration of this article.
The starting point and inspiration for this article is [113] and in many ways
the present survey can be regarded as an expansion and update of that article.
The intention of this article is not to give a philosophical argument in favor of
|R| = ℵ2 . Instead it will collect as many proofs of ℵ2 ≤ 2ℵ0 and 2ℵ0 ≤ ℵ2 from
different hypotheses as possible and place them in their mathematical context.
Whether this is “evidence” of anything is left to the reader, although it is worth
pointing out that in some cases the hypotheses used in these different proofs are not
consistent with each other. The article does not consider the arguments in favor
of, for instance, CH or try to weigh the “evidence” for it against the “evidence”
for any other value of the continuum. The reader interested in such a philosophical
discussion is referred to, for instance, [45], Maddy’s [75] [76], Woodin’s [128] [129]
and Foreman’s rebuttal of this argument in [47] (see also [48], [130]).
This article will begin with an examination the relationship between the Con-
tinuum Hypothesis, the Perfect Set Property, and the Open Coloring Axiom. The
classical consequences of CH will be discussed in Section 3, followed by Woodin’s
Σ21 -Absoluteness Theorem in Section 4. Section 5 will present a more detailed dis-
cussion of OCA and Section 6 will outline a Ramsey-theoretic proof of 2ℵ0 = ℵ2 .
Martin’s Maximum and its consequences will be over-viewed in Section 7. Station-
ary reflection and its impact on the continuum will be described in Section 8. In
Section 9, the conjectures of Chang and Rado will be introduced and related to
both the Continuum Problem and stationary reflection. Woodin’s Pmax -extension
will be briefly described in Section 10; Section 11 contains related information con-
cerning well orderings of R which are definable over H(ω2 ). Section 12 will discuss
some more sophisticated consequences of CH which are related to iterated forcing.
The Semifilter Trichotomy and the principle of Near Coherence of Filters will be
over-viewed in Section 13, along with a discussion of how these principles may imply
2ℵ0 = ℵ2 . The conclusion of the article contains a long list of open problems.
One omission worth noting is Gödel’s outline of a proof that 2ℵ0 = ℵ2 in [57].
The ingredients of this proof are both diverse and unconventional: the perfect set
properties of uncountable Boolean combinations of open subsets of R, an analysis of
strong measure 0 sets, and scales for Rω . Gödel’s argument was analyzed in detail
recently in [22] and the interested reader is referred there for more information.
Many of these concepts are still poorly understood in spite of the advances in
modern set theory and the “proof” remains incomplete in the sense that it is not
known that the hypotheses needed to carry it out are consistent (although see
[22, §3] for a discussion of this point). The reference [22] also contains a number
of questions which are interesting but tangential to the present discussion and thus
are not included in the final section.
The notation and terminology in this article is mostly standard. If X is a set
and κ is a (possibly finite) cardinal, then [X]κ will denote the collection of subsets
of X of cardinality κ. If f, g ∈ ω ω , then f <∗ g means that f (n) < g(n) except
for finitely many n ∈ ω (similarly one defines f ≤∗ g). I will use ω ω /fin to denote
the structure (ω ω , <∗ ). Suppose that X is an uncountable set. A subset C ⊆ [X]ω
WHAT MAKES THE CONTINUUM ℵ2 261
is club if it is the set of all f -closed sets for some f : X <ω → X (here M ⊆ X is
f -closed if f maps M <ω into M ). A subset S ⊆ [X]ω is stationary if S intersects
every club. Notice that ω1 ⊆ [ω1 ]ω is club. In particular, the stationary subsets of
[X]ω are a generalization of the stationary subsets of ω1 .
2. The Continuum Hypothesis, the Perfect Set Property,

and the Open Coloring Axiom
A common philosophical justification for CH is that we cannot effectively dem-
onstrate the existence of a subset of R of cardinality strictly between |N| and |R|.
In fact the reason why we cannot effectively produce an example of such an X is
that those X which are Borel or even analytic have what is known as the perfect
set property: they are either at most countable or else contain a homeomorphic
copy of the Cantor set, also known as a perfect set. In fact, in the presence of large
cardinals, every reasonably definable subset of a Polish space has the perfect set
property. Recall that L(R) is the smallest transitive model of Zermelo-Frankel set
theory which contains all of the reals and all of the ordinals.
Theorem 2.1. [101]1 If there is a proper class of Woodin cardinals, then every
set of reals in L(R) has the perfect set property.
Now consider the following graph-theoretic assertion for a given separable met-
ric space X:
OCA∗ (X): If G is a graph with vertex set X and whose incidence relation
is open when regarded as a symmetric subset of X 2 , then G either admits
a countable vertex coloring or else contains a perfect clique.
This principle first appeared in [46] and was derived from Todorcevic’s Open Color-
ing Axiom (OCA) which asserts that every open graph on a separable metric space
is either countably chromatic or else contains an uncountable clique. Notice that
by considering the complete graph on X, OCA∗ (X) trivially implies the perfect set
property. Theorem 2.1 admits the following strengthening.
Theorem 2.2. [46] If there is a proper class of Woodin cardinals, then the
principle OCA∗ (X) is true whenever X is a separable metric space in L(R).
In fact both CH and OCA follow from the Axiom of Determinacy (see [64] for an
introduction to AD). It is therefore quite interesting that in the presence of the
Axiom of Choice, these two principles take on a quite different character from each
other. CH is equivalent to the assertion that |R| = ℵ1 — i.e. that there is a well
ordering of R in which every proper initial segment of the well order is countable.
On the other hand, we will see that not only does OCA refute CH in the presence
of the Axiom of Choice, but also many of its typical consequences. In fact it is
unknown whether OCA implies |R| = ℵ2 .
3. Consequences of the Continuum Hypothesis

There are many uses of CH in mathematics. It is often utilized to build ul-
trafilters with special properties, nontrivial morphisms between structures, and
pathological topological spaces. In many cases the objects constructed have prop-
erties which are, to quote Gödel [56], “highly implausible.” The vast majority of
1 Actually this theorem is proved from the hypothesis of a supercompact cardinal in [101].
The refinement stated here is due to Woodin and can be found in [73].
these arguments involve a recursive construction of length continuum in order to

build some mathematical object of interest. The hypothesis |R| = ℵ1 allows one
to arrange that at each stage of the construction, there have only been countably
many previous stages. This section collects a number of such consequences of CH.
The reader is referred to [102] for additional examples.
The first example is perhaps overly simplistic, but it illustrates the basic idea
at the core of the more complex constructions.
Theorem 3.1. Assume CH. There is an uncountable subset of R in which every
Lebesgue measure 0 subset is countable and there is an uncountable subset of R in
which every first category subset is countable.
For instance, by enumerating all Borel sets of measure 0, it is straightforward to
recursively construct a set X which is uncountable but such that every measure 0
subset of X is countable.
Similarly, one can build the ultrafilter described in the next theorem of Choquet
by enumerating the functions f : ω → ω in a sequence of length ω1 and then building
a base for the ultrafilter by recursion.
Theorem 3.2. [25] [26] Assume CH. There is an ultrafilter U on ω such that
if f : ω → ω then f is either constant or one-to-one on a set in U.
Ultrafilters with the property stated in this theorem are known as selective ultra-
filters and are well studied in the literature [17] [35] [38] [67] [78].
The next result, independently proved by Christensen and Mokobodzki (who
was inspired by Choquet’s construction), asserts that CH implies the existence of
so called medial limits.
Theorem 3.3. (see [27], [79]) Assume CH. Then there exists a universally
measurable μ : P(ω) → [0, 1] which is a finitely additive probability measure such
that μ(F ) = 0 whenever F ⊆ ω is finite.
Actually CH is a much stronger hypothesis than needed for both of these theorems;
Martin’s Axiom (MA) is sufficient.
The next result shows that CH can have an influence on the phenomenon
of automatic continuity and provides a consistent answer to an old problem of
Kaplansky [65].
Theorem 3.4. [31] Assume CH. If X is an infinite compact space then there
is a discontinuous homomorphism of C(X) into a Banach algebra.
Solovay and Woodin later proved that it is consistent that if X is a compact space,
then every homomorphism of C(X) into a Banach algebra is continuous; in fact
this conclusion follows from Martin’s Maximum discussed in Section 7 (see [32]).
See also [40] for a general discussion of these results.
In some cases, consequences of CH can actually be shown to be equivalents of
CH.
Theorem 3.5. [86] CH is equivalent to the existence of a surjection γ : R → R2
with the property that for every t at least one of the coordinate functions γ1 and γ2
is differentiable at t.
Theorem 3.6. [60] CH is equivalent to the assertion that there is an uncount-
able subset of [0, 1]ω with the property that every finite dimensional subset is at most
countable.
Theorem 3.7. [102] CH is equivalent to the existence of an uncountable subset

X ⊆ [0, 1]ω with the property that if Y ⊆ X is uncountable, then Y is projected onto
the interval by all but finitely many of the coordinate projections.
The following are two typical examples of how CH can be used to influence the
existence of morphisms between structures.
Theorem 3.8. [37] Assume CH. If X ⊆ R has cardinality continuum, then
there exists a Y ⊆ X of cardinality continuum such that if f ⊆ Y × Y is a partial
monotone injective function, then
|{x ∈ dom(f ) : x = f (x)}| < 2ℵ0 .
In particular, CH implies that there are |P(R)|-many pairwise nonisomorphic ℵ1 -
dense suborders of R.
Theorem 3.9. [63] Assume CH. If I and J are two proper Fσ -ideals on ω,
then P(ω)/I and P(ω)/J are isomorphic.
These two theorems contrast each other in an interesting way. The first demon-
strates that under CH, there are nonisomorphic suborders of R for which there is
no apparent obstruction to an isomorphism. In fact Baumgartner has shown that
if CH is true then there is a c.c.c. forcing extension in which all ℵ1 -dense suborders
of R are isomorphic [9]. Thus the reason for the failure of the isomorphism to exist
under CH is not absolute; it simply reflects that there are “not enough” monotone
functions from R to R in models of CH.
On the other hand, in the case of the Theorem 3.9 there is no obvious reason why
there should be an isomorphism between the quotients P(ω)/fin and P(ω)/I1/n
(here I1/n is the collection of subsets of ω for which the sum of the reciprocals is
convergent). CH allows the construction of the isomorphism via a back and forth
argument of length continuum — at each stage of the construction one extends
a countable partial isomorphism between the two Boolean algebras. Indeed it is
possible to show that there is no isomorphism which is induced by a function from ω
to ω or even an isomorphism between these quotients which has a Baire-measurable
lifting Φ : P(ω) → P(ω) [61]. Moreover, starting from any model of set theory,
there is a forcing extension in which P(ω)/fin and P(ω)/I1/n are not isomorphic
[61]. In this case, the existence of an isomorphism between these two structures
can be “blamed” on there not being enough subsets of ω to prevent the existence
of an isomorphism.
A relative of Theorem 3.9 is the following result of W. Rudin.
ℵ0
Theorem 3.10. [94] Assume CH. There are 22 autohomeomorphisms of
ℵ0
the Čech-Stone remainder of ω. Equivalently there are 22 automorphisms of
P(ω)/fin. In particular, there is an automorphism which is not induced by a func-
tion from ω to ω.
As in the proof of Theorem 3.9, the elements of P(ω) are enumerated in length ω1
and partial automorphisms are constructed recursively. The construction allows a
subset of ω1 as a parameter, so that each choice of parameter results in a different
automorphism.
Unlike in the case of Theorem 3.8, there is no general means to continue such
an inductive construction following uncountable limit stages. In fact while every
countable partial automorphism of P(ω)/fin can always be extended to an arbi-

trary element not in its domain, it is a theorem of ZFC that there are partial
automorphisms of P(ω)/fin of cardinality ℵ1 which cannot be extended to certain
subsets of ω.
This is perhaps best understood in terms of model theory. Recall that a struc-
ture A is homogeneous if every partial automorphism of cardinality less than the
domain of A can be extended to an automorphism of A (note that this is con-
siderably stronger than what is meant by homogeneity outside of model-theoretic
literature). A structure A is saturated if every partial type of cardinality less than
the domain of A is realized in A. Here a type is a collection of n-ary formulas in
the language of A which is finitely satisfiable. We now have the following theorem.
Theorem 3.11. The following are equivalent:
(1) The Continuum Hypothesis.
(2) P(ω)/fin is a saturated Boolean algebra.
(3) P(ω)/fin is a homogeneous Boolean algebra.
The CH implication in this theorem is closely related to a reformulation of following
topological result of Parovičenko.
Theorem 3.12. [89] Assume CH. The Čech-Stone remainder of ω is the unique
0-dimensional compact Hausdorff space having no isolated points such that if F and
G are two disjoint open Fσ -sets, then there exists a closed and open set U such that
F ⊆ U and U ∩ G = ∅.
What is counter-intuitive about this theorem is that it implies that many compact
spaces are homeomorphic even when there is no apparent homeomorphism: For
instance the Čech-Stone remainders of the infinite countable limit ordinals are all
homeomorphic assuming CH (contrast this with [43]).
CH also produces a wealth of exotic examples of topological spaces. I will
mention only a few; it is impossible to adequately give even a brief survey.
Theorem 3.13. ([54]; see also [107]) If CH is true, then for each n there is
a topological space X such that X n+1 contains an uncountable family of pairwise
disjoint open sets but such that X n does not.
Theorem 3.14. [66] If CH is true, then there is a nonseparable compact Haus-
dorff space K which does not contain an uncountable discrete subspace. This space
moreover supports a Borel probability measure μ for which the following are equiv-
alent for Y ⊆ K:
• Y is nowhere dense;
• Y is first category;
• μ(Y ) = 0;
• Y is second countable;
• Y is separable.
(Similar constructions were carried out independently by Haydon [59] and Tala-
grand [105].)
Theorem 3.15. [44] If CH is true, then for each d > 0 there is an infinite
compact Hausdorff space Kd in which every infinite closed subspace has dimension
d. In particular, Kd contains neither convergent sequences nor a copy of βω.
4. Woodin’s Σ21 -Absoluteness Theorem

CH can be phrased as asserting the existence of a binary relation on P(ω)
such that:
• (P(ω), ) is a linear ordering;
• for each y ∈ R, {x ∈ P(ω) : x y} is countable.
In particular, CH is equivalent to a Σ21 -sentence — a sentence of the form ∃X ⊆
P(ω)φ(X) where φ involves only quantification over P(ω). In fact many of the
consequences of CH can be phrased as Σ21 -sentences.
Moreover, it was noticed that, empirically at least, CH tended to be sufficient to
settle the truth of Σ21 -sentences which would come up in practice. Quite remarkably,
there is a general result due to Woodin which explains this.
Theorem 4.1. (see [73]) Assume there is a proper class of measurable Woodin
cardinals and that CH is true. Any Σ21 -sentence which can be forced is true.
It should be noted that some care is needed in stating this theorem. For instance
the assertion that there is a nonconstructible real is trivially a Σ21 -sentence (in fact
this assertion can be formalized using only quantification over P(ω)). By Cohen’s
work it is consistent and yet it is not true in L, which satisfies CH. Thus we cannot
hope to have a result which says that if a Σ21 -sentence is consistent with ZFC, then
it is a consequence of CH.
5. Todorcevic’s Open Coloring Axiom

We will now return to the Open Coloring Axiom introduced in Section 2 above.
This axiom has a strong influence on structures closely related to separable metric
spaces and often helps provide an alternative mathematical theory to that imposed
by CH. This axiom’s name was based on a family of similar Ramsey-theoretic
principles in Abraham, Rubin, and Shelah’s [2]. The following are examples of
open graphs.
Example 5.1. Suppose that A, B ⊆ R and define a graph on A × B by con-
necting (x0 , y0 ) to (x1 , y1 ) whenever x0 < x1 and y0 < y1 or x1 < x0 and y1 < y0 .
Observe that this graph is open. It is countably chromatic only when either A or B
is countable. On the other hand, a clique is clearly the graph of a partial increasing
function from A to B. Thus OCA implies that if A, B ⊆ R are uncountable, then
they have uncountable isomorphic suborders. In particular, by Theorem 3.8, OCA
refutes CH.
The next examples are more typical of the use of OCA in applications. They
provided Todorcevic’s motivation isolating his formulation of OCA from the similar
forms initially studied in [2].
Example 5.2. Suppose that A ⊆ P(ω) and define a graph on A where {a, b}
is an edge if neither a ⊆ b nor b ⊆ a. In particular, OCA implies that if A is
uncountable, then either A contains an uncountable chain or antichain with respect
to containment (here antichain means pairwise incomparable). This conclusion was
originally deduced by Baumgartner from PFA [11].
Example 5.3. Let X ⊆ ω ω × ω ω consist of all pairs (a, b) for which a ≤ b
coordinate-wise. Define {(a0 , b0 ), (a1 , b1 )} to be an edge of the graph if there is an
i such that
max(a0 (i), a1 (i)) > min(b0 (i), b1 (i)).
Typically one considers induced subgraphs of this graph as follows. Suppose that
A, B ⊆ ω ω have the property that a ≤∗ b whenever a ∈ A and b ∈ B and that A
and B are closed under making finite modifications to their elements. Consider the
subgraph in which the vertices come from A×B. This graph is countably chromatic
exactly when there exist sets cn (n ∈ ω) such that for every a ∈ A and b ∈ B, there
is an n such that a ≤∗ cn ≤∗ b.
This second example readily gives the following classification of gaps in ω ω /fin
in the presence of OCA. Here we recall that a (κ, λ∗ )-gap in ω ω /fin is a pair of
sequences aξ (ξ ∈ κ) and bη (η ∈ λ) such that aξ <∗ aξ <∗ bη <∗ bη for all
ξ < ξ < κ and η < η < λ and yet there is no c such that aξ <∗ c <∗ bη for all ξ
and η.
Theorem 5.4. [109] Assume OCA. If κ and λ are regular cardinals and there
is a (κ, λ∗ )-gap, then either min(κ, λ) = ω or else κ = λ = ω1 .
This complements the following classical results of Hausdorff.
Theorem 5.5. [58] There is an (ω1 , ω1∗ )-gap in ω ω /fin.
Theorem 5.6. [58] The following are equivalent for a regular cardinal κ:
• There is an unbounded chain in ω ω /fin of ordertype κ.
• There is a (κ, ω ∗ )-gap in ω ω /fin.
• There is a (ω, κ∗ )-gap in ω ω /fin.
In particular, there exists a regular cardinal κ for which there exist both (κ, ω ∗ )-gaps
and (ω, κ∗ )-gaps.
We will finish this section by mentioning some consequences of OCA and MA
concerning quotient algebras which contrast the influence of CH detailed in Section
3. Building on work of Just, Shelah, Steprāns, and Veličković, Farah proved the
following general results.
Theorem 5.7. [42] Assume OCA and MA. If I and J are analytic ideals
and at least one is either a P -ideal or countably generated, then every isomorphism
between P(ω)/I and P(ω)/J has a Baire measurable lifting.
Theorem 5.8. [42] If I and J are analytic ideals, at least one is either a
nonpathological P -ideal or countably generated, and P(ω)/I and P(ω)/J are
isomorphic via a map admitting a Baire measurable lifting, then I and J are
isomorphic.
In particular the conjunction of OCA and MA implies that the quotient algebras
P(ω)/fin and P(ω)/I1/n are nonisomorphic, in contrast to Theorem 3.9. (This
instance of Theorems 5.7 and 5.8 was first proved by Just [61] [62].) Farah later
adapted these techniques to the context of C∗ -algebras [41], complementing a pre-
vious CH construction of Phillips and Weaver [90].
Recall that the measure algebra is the quotient of the algebra of Borel subsets
of [0, 1] by the ideal of Lebesgue measure 0 sets. In the presence of CH, P(ω)/fin
is saturated and hence any Boolean algebra of cardinality at most 2ℵ0 can be
embedded inside it. On the other hand, we have the following result of Dow and
Hart.
Theorem 5.9. [36] Assume OCA. The measure algebra does not embed into
P(ω)/fin.
6. A Ramsey-theoretic proof that the continuum is ℵ2

The only precursor of OCA considered in [2] which is not a formal weakening
of OCA is OCA[ARS] :
If X is a separable metric space of cardinality ℵ1 and c : [X]2 →
{0, 1} is a continuous function, then X can be decomposed into
countably many c-homogeneous sets.
Both OCA and OCA[ARS] are consequences of Martin’s Maximum (discussed in
Section 7 below) and both refute CH. Formally they are unrelated — it is consistent
that either hold but not the other. On one hand OCA[ARS] only makes assertions
about graphs which are closed and open and in which the underlying vertex set has
cardinality ℵ1 . On the other hand, unlike with OCA, the conclusion of OCA[ARS]
is necessarily “global.”
While it remains unknown whether either of these assertions decides the value
of the continuum, we have the following result — a purely Ramsey-theoretic proof
that the continuum is ℵ2 .
Theorem 6.1. [82] The conjunction of OCA and OCA[ARS] implies that |R| =
ℵ2 .
We have already seen that OCA implies that all gaps are of one of the following
forms: (κ, ω ∗ ), (ω, κ∗ ), or (ω1 , ω1∗ ). In fact this conclusion already implies that
b ≤ ℵ2 , where b is the minimum cardinality of an unbounded subset of ω ω /fin. To
see this, suppose that b > ℵ2 and fix a bounded chain aξ (ξ < ω2 ) in ω ω /fin. It
is routine to show that this is the lower part of an (ω2 , λ∗ )-gap for some infinite
regular cardinal λ. By Theorem 5.6, λ > ω and thus by Theorem 5.4, OCA fails.
It is also the case that OCA implies that b > ℵ1 and thus that b = ℵ2 . This is
proved via the following ZFC result.
Proposition 6.2. [118] Suppose that X ⊆ ω ω /fin consists of monotone in-
creasing functions, is unbounded and countably directed. For every n, there exists
x, y ∈ X such that osc(x, y) = n (in particular, there exist x = y ∈ X with x ≤ y).
Here osc(x, y) < n exactly when ω can be covered by n intervals I such that x ≤ y
on I or y ≤ x on I. Notice that if b = ℵ1 , then there is an unbounded chain X
in ω ω /fin of length ω1 which moreover consists of monotone functions. If G is the
positive oscillation graph on X, then G is open and is neither countably chromatic
nor has an uncountable clique.
Corollary 6.3. [109] OCA implies that b = ℵ2 .
The remainder of the proof of Theorem 6.1 hinges on the following variation
on Proposition 6.2.
Proposition 6.4. [82] There is a continuous partial function σ : [ω ω ]2 → 2<ω
with an open domain such that:
• if x <∗ y, then σ(x, y) is defined and has the same length as the common
initial part of x and y;
• if X is unbounded and countably directed in ω ω /fin and r is in 2ω , then
there exist x = y in X such that σ(x, y) is an initial part of r.
If r is in 2ω , then define an open graph Gr on ω ω by putting {x, y} ∈ Gr if σ(x, y)

is defined and is an initial part of r. A code for r is an uncountable clique in Gr .
Observe that a given H can be a code for at most one r. It is immediate from the
previous proposition that OCA implies that every unbounded countably directed
subset of ω ω /fin contains codes for all elements of 2ω . The proof is then finished
by Corollary 6.3 and the following proposition.
Proposition 6.5. OCA[ARS] implies that if X ⊆ ω ω /fin is a chain of cardi-
nality ℵ1 , then X contains codes for at most ℵ1 elements of 2ω .
This is achieved by showing that OCA[ARS] implies that for each n, any X ⊆
ω ω /fin of cardinality ℵ1 can be partitioned into sets Xσ for σ a length n binary
sequence so that if H ⊆ X codesr ∈ 2ω , then H is contained in Xrn modulo a
countable set. If we define Xr = n Xrn , then {Xr : r ∈ 2ω } is a disjoint family
and hence only ℵ1 many can be uncountable (or even nonempty). The proposition
then follows from the observation that any code for r is contained in Xr modulo a
countable set.
7. Forcing axioms and generic absoluteness

Both forms of the Open Coloring Axiom follow from a much more general set-
theoretic hypothesis known as Martin’s Maximum. MM was first formulated and
proved consistent relative to the existence of a supercompact cardinal by Foreman,
Magidor, and Shelah [51]. It is the strongest example from an important class of
axioms known as forcing axioms.
Forcing axioms are generalizations of the Baire Category Theorem and grew
out of a line of research which started with Solovay and Tennenbaum’s proof that
Souslin’s Hypothesis is consistent [103]. A stratification of these axioms has been
included below mostly for historical accuracy. The reader who is not interested
in fine details will lose little of the general picture by assuming MM in place of
whatever forcing axiom is needed as a hypothesis (this may, however, mean that
the result was proved earlier and possibly by someone else).
We will now review some of the basic definitions associated to these axioms. If
θ is a cardinal and Q is a class of partially ordered sets, then FAθ (Q) is the assertion
that if Q is in Q and A is a family of maximal antichains in Q with |A| ≤ θ, then
there is a filter G ⊆ Q which meets each element of A — such a filter is said to be
A-generic. (Here antichain means in the sense of forcing — a family of elements of
the poset such that no pair of distinct elements has a common lower bound.) It is
not difficult to show that if Q is the class of all posets, then FAℵ0 (Q) is a theorem
of ZFC. This is closely related to the assertion that no compact Hausdorff space
can be covered by countably many nowhere dense sets, which is itself a variation
of the Baire Category Theorem. On the other hand, it is not difficult to see that if
Q is the poset of all finite partial functions from ω to ω1 , then there is a family A
of ℵ1 maximal antichains in Q such that there is no A-generic filter. In particular
FAℵ1 (Q) is false if Q is the class of all partial orders (or even all partial orders of
cardinality at most ℵ1 ).
If Q is the class of c.c.c. partial orders (those in which all antichains are
countable), then FAθ (Q) is denoted MAθ and known as Martin’s Axiom for θ
antichains. This was historically the first forcing axiom. It was isolated by D. A.
Martin from the proof that Souslin’s Hypothesis is consistent [103] and provided
the template for the stronger forcing axioms which followed later. Unlike Souslin’s
Hypothesis, Martin’s Axiom turned out to be widely applicable in analysis and
point-set topology. The reader is referred to the encyclopedic [53] for a full account
of the consequences of MAθ . While it is trivial that MAθ implies that 2ℵ0 > θ,
MAθ has no other influence on the cardinality of the continuum other than that
2ℵ0 = 2θ and hence that cf(2ℵ0 ) > θ (see Theorem 11.1 below).
Forcing Axioms for broader classes of posets than c.c.c. posets are typically
inconsistent if θ > ℵ1 and it is therefore common to use FA(Q) to denote FAℵ1 (Q).
(Note however that MA typically denotes the assertion that MAθ holds for all
θ < 2ℵ0 .) A poset Q is said to be proper if forcing with it preserves stationary
subsets of [X]ω whenever X is an uncountable set. The Proper Forcing Axiom
(PFA) is FA(proper). Martin’s Maximum is the forcing axiom for posets which
preserve stationary subsets of ω1 . These axioms were proved consistent relative to
the existence of a supercompact cardinal by Baumgartner and by Foreman, Magidor
and Shelah, respectively.
Theorem 7.1. (see [33]) If there is a supercompact cardinal κ, then there is a
forcing extension which satisfies PFA.
Theorem 7.2. [51] If there is a supercompact cardinal κ, then there is a forcing
extension which satisfies MM.
MM is the strongest forcing axiom in the sense that if Q is a poset which does
not preserve some stationary subset of ω1 , then there is a family A of ℵ1 maximal
antichains such that there does not exist an A-generic filter.
The theory of MM has been the subject of extensive study, both for its set-
theoretic consequences and for its applications to other branches of mathematics
— see for instance [53], [83], [116]. The following are some of its most important
early consequences.
Theorem 7.3. [51] MM implies that the nonstationary ideal on ω1 is satu-
rated: if Sξ (ξ < ω2 ) are stationary subsets of ω1 , then there exist ξ = η < ω2 such
that Sξ ∩ Sη is stationary.
Theorem 7.4. [51] Assume MM and let κ > ℵ1 be a regular cardinal. Suppose
that Aξ : ξ ∈ ω1 are pairwise disjoint stationary subsets of ω1 and Bξ : ξ ∈ ω1
are stationary subsets of κ which each consist of ordinals of countable cofinality.
There is a continuous increasing sequence γξ : ξ ∈ ω1 in κ such that for all
ξ ∈ ω1 , if ξ is in Aξ , then γξ is in Bξ .
An immediate consequence of the second result is that, assuming MM, κℵ1 = κ
whenever κ > ℵ1 is a regular cardinal. In particular, MM implies that 2ℵ1 ≤ ℵ2 ℵ1 =
ℵ2 ; this is historically the first use of a forcing axiom to produce a bound on the
cardinality of the continuum. Since it is trivial that MM refutes CH, we also have
that 2ℵ0 = ℵ2 is a consequence of MM.
Woodin showed that Theorem 7.3 leads to a more dramatic failure of CH.
Theorem 7.5. [127, Ch.3] If the ideal of nonstationary subsets of ω1 is satu-
rated and P(ω1 ) exists, then δ 12 = ω2 .
˜
Here δ 12 is the supremum of the lengths of all δ 12 pre-wellorderings of R. The
˜
hypothesis P(ω1 ) exists is a consequence of MM˜ as well; it also follows from the
existence of a measurable cardinal. In particular, MM not only refutes CH, but in
fact gives an effective failure of CH — one at the level of descriptive set theory.
This addressed question raised in [50].
PFA, which is a weakening of MM, plays an important role in classifying struc-
tures of cardinality ℵ1 .
Theorem 7.6. [9] If CH is true and X and Y are ℵ1 -dense suborders of R, then
there is a c.c.c. forcing extension in which X and Y are isomorphic as linear orders.
In particular PFA implies that any two ℵ1 -dense suborders of R are isomorphic.2
Theorem 7.7. [108] Assume PFA. Every directed system of cardinality at most
ℵ1 is cofinally equivalent to one of the following: 1, ω, ω1 , ω × ω1 , [ω1 ]<ω .
This classification was extended to the transitive relations on ω1 in [112].
Next recall that an Aronszajn line is an uncountable linear order in which
all separable and scattered suborders are countable. Such linear orders were first
constructed by Aronszajn and Kurepa (see [68]). A Countryman line is an un-
countable linear order with the property that its Cartesian square is a countable
union of chains. Such orders are necessarily Aronszajn and were first constructed
by Shelah [96] (see also [117]). The following results show that there is a rather
strong analogy between the Aronszajn lines and the countable linear orderings (with
Countryman lines playing the roles of ω and −ω).
Theorem 7.8. Assume PFA. The following are true:
• [1] Every pair of normal Countryman lines are isomorphic or reverse iso-
morphic.
• [80] Every Aronszajn line contains a Countryman suborder.
• [85] If C is a Countryman line, then
dirlim C × (−C) × · · · × (−C)
is an injectively universal Aronszajn line.
• [77] The Aronszajn lines are well quasi-ordered by embeddability.
Much of the theory of MM can be developed through the combinatorial con-
sequences of PFA. We have already noted that both OCA and MAℵ1 follow from
PFA. Another useful consequence of PFA is Todorcevic’s P -Ideal Dichotomy (PID):
If S is any set and I ⊆ [S]ω is a P -ideal, then either:
• there is an uncountable X ⊆ S such that [X]ω ⊆ I or else
• there is a decomposition S = ∞ n=0 Sn such that for each n,
no infinite subset of Sn is in I .
(Here I is a P -ideal on S if it is an ideal which includes all finite subsets of S
and which is countably directed under ⊆∗ .) PID was analyzed in detail for sets
of size ℵ1 in [3]. That paper also briefly mentions the general case and how to
prove its consistency based on methods in [109], although a detailed analysis of its
consequences and a proof of its consistency were first given in [114]. This axiom
has a similar influence on gaps as OCA does:
Theorem 7.9. [114] Assume PID. If κ and λ are regular cardinals and there
is a (κ, λ∗ )-gap in ω ω /fin, then either min(κ, λ) = ω or else κ = λ = ω1 .
2 PFA had not been formulated at the time of [9].

In particular, PID implies b ≤ ℵ2 . On the other hand, unlike OCA, PID is con-
sistent with CH [114]. PID also influences the combinatorics of sets which are far
removed from R.
Theorem 7.10. [3] PID implies that every uncountable tree either contains an
uncountable chain or an uncountable antichain.
Theorem 7.11. [125] PID implies that if κ is a singular strong limit cardinal,
then 2κ = κ+ .
Another general consequence of MM which we will discuss is the Strong Reflec-
tion Principle (SRP):
If ω1 ⊆ X and S ⊆ [X]ω , then there is a continuous ⊆-chain
Nξ : ξ ∈ ω1 such that ξ ⊆ Nξ for all ξ ∈ ω1 and if ν ∈ ω1 is a
limit ordinal and there is an ω1 -end extension of Nν which is in
S, then Nν is in S.
Here N is an ω1 -end extension of M if M ⊆ N and M ∩ ω1 = N ∩ ω1 .
SRP was first formulated by Todorcevic who abstracted it from arguments
in [51]. SRP captures a number of consequence of MM which were originally
directly deduced in [51], including the saturation of the ideal of nonstationary
subsets of ω1 (NSω1 ), Chang’s Conjecture, and the conclusion of Theorem 7.4 [14].
SRP was also the inspiration for the Mapping Reflection Principle (MRP), another
related consequence of PFA [84] which played an important role in the solution of
the basis problem for uncountable linear orders [80]. We will discuss some of the
consequences of SRP in the next section; see [14] for more details on how to derive
its consequences.
We will finish this section with two more variants on the theme of forcing
axioms which will be needed the sections which follow. If membership to Q is
an invariant of forcing equivalence, then BFA(Q) is the weakening of FA(Q) in
which the families A consist only of antichains of cardinality at most ℵ1 . By work
of Bagaria [7], this is equivalent to the assertion that H(ℵ2 ) is a Σ1 -elementary
substructure of any generic extension by a poset from Q. The assertion FA+ (Q)
is the following strengthening of FA(Q): If Q is in Q, A is a family of at most ℵ1
maximal antichains in Q and Ṡ is a Q-name for a stationary set, then there is an
A-generic filter G ⊆ Q such that
{ξ ∈ ω1 : ∃p ∈ G(p ξˇ ∈ Ṡ}
is stationary; FA++ (Q) is the strengthening in which ℵ1 -many names for stationary
sets are allowed.
8. Stationary set reflection and 2ℵ0 ≤ ℵ2

This section contains an outline of why the reflection of stationary sets in [ω2 ]ω
implies that 2ℵ0 ≤ ℵ2 . Let θ ≥ ω2 be a cardinal. A stationary subset S ⊆ [θ]ω
is said to reflect if there exists an X ⊆ θ of cardinality ℵ1 such that ω1 ⊆ X
and S ∩ [X]ω is stationary in [X]ω . This notion was perhaps first considered by
Baumgartner, who proved the following results.
Theorem 8.1. (see [10, §8]) If there is supercompact cardinal, then there is a
forcing extension in which PFA++ holds.
Theorem 8.2. (see [10, §8]) PFA++ implies that every stationary subset of
ω
[X] reflects.
The assertion that stationary subsets of [θ]ω reflect is a natural hypothesis and
one which has been studied extensively in the literature [28], [29], [30], [100],
[104], [111]. We will begin by noting that it follows from MM both directly and
also via the principle SRP.
Theorem 8.3. [51] MM implies that if θ ≥ ℵ2 is a cardinal, then every sta-
tionary subset of [θ]ω reflects.
Theorem 8.4. [14] SRP implies that if θ ≥ ℵ2 is any cardinal, then every
stationary subset of [θ]ω reflects.
In fact if S ⊆ [θ]ω and Nξ (ξ < ω1 ) satisfies the conclusion of SRP for S, then
{M ∈ S : M ∩ ω1 ∈ {ξ ∈ ω1 : Nξ ∈ S}}
is nonstationary. Since the set of all M ∈ [θ]ω which ω1 -end extend an element of
{Nξ : ξ ∈ ω1 } is club, it follows that {ξ ∈ ω1 : Nξ ∈ S} must be stationary if S is
stationary.
Quite remarkably, we have the following result.
Theorem 8.5. [119] If every stationary subset of [ω2 ]ω reflects, then 2ℵ0 ≤ ℵ2 .
The reason for this is twofold. On one hand, it is a result of Baumgartner
and Taylor that every club in [ω2 ]ω has cardinality at least 2ℵ0 [13]. On the
other, Todorcevic proved that there is a subset F of [ω2 ]ω of cardinality ℵ2 whose
complement is either nonstationary or else is equal to a nonreflecting stationary set
modulo the club filter [119].
In fact by appealing to a result of Gitik [55], Todorcevic was able to draw an
even stronger conclusion.
Theorem 8.6. [111] Suppose that every stationary subset of [ω2 ]ω reflects. If
N is an inner model such that ℵ2 N = ℵ2 V , then R ⊆ N . In particular |R| ≤ ℵ2 .
Todorcevic’s set F can be described as follows. Fix for a moment a sequence
e = eβ : β ∈ ω2 such that eβ is an injection from β into ω1 . A subset X of ω2 is
e-closed if X ∩ ω1 is an ordinal and whenever β is in X and α ∈ β, eβ (α) is in X if
and only if α is in X. If we define E to be the collection of all e-closed elements of
[ω2 ]ω , then it is easily checked that E is club. Also, observe that if ω1 ⊆ X ⊆ ω2
and X is e-closed, then X is an ordinal. Define
Fα,β = {ξ ∈ β : eβ (ξ) < α}
F = {Fα,β : α < ω1 ≤ β < ω2 }
and set A = E \ F . Notice that if δ ∈ ω2 , then {e−1
δ ξ : ξ ∈ ω1 } is club in [δ] . In
ω
particular, if ω1 ⊆ X ⊆ ω2 and X has cardinality ℵ1 , then A∩[X] is nonstationary

ω
in [X]ω .
Gitik’s result can be described as follows. Suppose that a, b0 , b1 are countable
subsets of ω2 and ξ ∈ a and define r ∈ 2ω and (ξk )k ∈ ω2 ω recursively as follows:
r(k) = 1 if and only if min(b0 \ ξk ) > min(b1 \ ξk )
ξ0 = ξ and ξk+1 = min(a \ min(br(k) \ ξk ))
Set hξ (a, b0 , b1 ) = r. Gitik proved that for every club E ⊆ [ω2 ]ω and r ∈ 2ω , there
exist a, b0 , b1 ∈ E and ξ ∈ a such that hξ (a, b0 , b1 ) = r. Thus if an inner model
contains a club subset of [ω2 ]ω , that inner model also contains all the reals.
It is worth noting that Todorcevic has a different method for proving that club
subsets of [ω2 ]ω have cardinality at least 2ℵ0 which is particularly elegant. First
recall that, by work of Shelah [95], there is a sequence Cδ : δ ∈ limω (ω2 ) such
that:
• each Cδ is a cofinal ω-sequence in δ and
• if E ⊆ ω2 is closed and unbounded, then there exists a δ such that Cδ ⊆ E.
Fix such a sequence and if x ⊆ θ has no last element, define
pat(x) = {|α ∩ Cδ | : α ∈ x}
where δ = sup(x). For each infinite r ⊆ ω, it can be verified that
Sr = {N ∈ [ω2 ]ω : pat(N ) = r}
is stationary; clearly {Sr : r ⊆ ω} are pairwise disjoint. The proof that Sr is
stationary was generalized considerably by Foreman and Todorcevic in [52].
9. The Conjectures of Chang and Rado

This section will introduce two conjectures which arose outside of set theory and
which relate to the phenomenon of stationary reflection described in the previous
section. Each implies 2ℵ0 ≤ ℵ2 and moreover that the conclusion of Theorem 8.6
holds. The first is a strengthening of a well known model-theoretic transfer principle
known as Chang’s Conjecture. Chang’s Conjecture (CC) is the assertion that if A is
a structure of cardinality ℵ2 and |X A | = ℵ1 for some unary relation X, then there
is an elementary substructure B ≺ A such that |B| = ℵ1 and |X B | = ℵ0 . Chang’s
Conjecture has been well studied in the literature; see [64] for further reading. The
following strengthening allows one to readily build the elementary substructure B
via a recursive process (see Theorem 1.3 of [98]).
CC∗ : If θ ≥ ω2 is a sufficiently large regular cardinal, is a well ordering of
H(θ), and M ≺ (Hθ , ∈, ) is countable, then there is an M ≺ H(θ) such
that M ⊆ M , M ∩ ω1 = M ∩ ω1 , but M ∩ ω2 = M ∩ ω2 .
The relevance to the present discussion is the following result.
Theorem 9.1. [51] If every stationary subset of [2ω1 ]ω reflects, then CC∗ is
true.
The hypothesis CC∗ implies that the set F described above contains a club regard-
less of the choice of the sequence e = eβ : β ∈ ω2 used to define it. In particular,
we have the following result.
Theorem 9.2. [111] CC∗ implies 2ℵ0 ≤ ℵ2 .
Next we turn to a conjecture made by Richard Rado who was motivated by an
analogous theorem which he proved in finite combinatorics. Recall that an interval
graph is the intersection graph of a collection of intervals in a linear order. Rado
proved that if the chromatic number of an interval graph is greater than n, then
there is a (necessarily complete) subgraph on n + 1 vertices which already has
chromatic number greater than n [91]. He then made the following conjecture [92]:
RC: If an interval graph is not countably chromatic, then it has a subgraph

of cardinality ℵ1 which is not countably chromatic.
Todorcevic showed that Rado’s Conjecture is equivalent to the assertion that every
nonspecial tree has a subset of cardinality ℵ1 which is nonspecial. Here we recall
that a tree of height ω1 is special if it is a union of countably many antichains. Notice
that there are many trees of cardinality continuum which have no uncountable
branch and yet are non-special — the well ordered subsets of Q ordered by end
extension and the tree of all closed subsets of a stationary co-stationary subset of
ω1 are two standard examples. Thus while Rado’s Conjecture implies a number of
consequences of Martin’s Maximum, it is inconsistent with MAℵ1 , which implies
that all trees of cardinality ℵ1 without uncountable branches are special [12].
Theorem 9.3. [106] If there is a supercompact cardinal, then there is a forcing
extension in which Rado’s Conjecture is true. Moreover, the forcing extension can
be arranged to satisfy either 2ℵ0 = ℵ1 or 2ℵ0 = ℵ2 .
Theorem 9.4. [111] Rado’s Conjecture implies CC∗ and that θ ℵ0 = θ whenever
θ ≥ ℵ2 is a regular cardinal. In particular, Rado’s Conjecture implies 2ℵ0 ≤ ℵ2 .
10. Woodin’s Pmax -extension of L(R)

The hypothesis Martin’s Maximum discussed in Section 7 above has been highly
successful in proving consistency results and developing a rich mathematical theory
extending that axiomatized by ZFC. It has long been empirically observed that the
theory of MM is quite close to being complete, at least as far as the theory of H(ℵ2 )
is concerned. This is partially explained by the following recent result of Viale.
Theorem 10.1. [126] There is a strengthening MM+++ of MM++ which holds
in the standard models for MM and with the property that any two generic exten-
sions by NSω1 -preserving forcings which satisfy MM+++ have the same theory for
H(ℵ2 ).
It has long been a mystery whether a result such as this can be proved if the
assumption of preserving stationary sets is dropped. There is no known canonical
model of MM and there is no known sense in which its theory is optimal or canonical.
Woodin’s Pmax -extension of the model L(R) is motivated in part by these sorts
of philosophical considerations. The next theorem is the starting point for the
development of Pmax .
Theorem 10.2. [101]3 If there are a proper class of Woodin cardinals, then
L(R)V is elementarily embedded in L(R)V [G] whenever G is V -generic for a set
forcing. In particular, if there is a proper class of Woodin cardinals, then the
theory of L(R) cannot be changed by forcing.
Notice that the inner model L(P(ω1 )) is correct about CH and that many
mathematical statements which are independent of ZFC are actually assertions
about what is true in this model. In light of Theorem 10.2, it is natural to spec-
ulate whether the theory of L(R) might be used to generate a canonical theory of
L(P(ω1 )) which extends ZFC. Woodin has shown that this is indeed the case.
Theorem 10.3. [127] Assume there is a proper class of Woodin cardinals. In
L(R) there is a forcing Pmax with the following properties:
3 The comment in the footnote to Theorem 2.1 applies here as well.
• Pmax is homogeneous;
• Pmax is σ-closed;
• If G ⊆ Pmax is L(R)-generic, then L(R)[G] satisfies AC;
• If G ⊆ Pmax is L(R)-generic and A ⊆ ω1 is in L(R)[G] \ L(R), then
L(R)[A] = L(R)[G].
• If φ is a Π2 -sentence in the language of the structure
(H(ω2 ), ∈, ω1 , NSω1 )
and φ is Ω-consistent, then Pmax forces H(ω2 ) satisfies φ.
I will not define Ω-logic here; see [127] and also [8]. Suffice it to say that if ZFC
proves that φ can be forced, then φ is Ω-consistent. Notice that since Pmax is σ-
closed, forcing with it does not change L(R). Since Pmax is homogeneous, the theory
of the generic extension coincides with the set of sentences which are forced by every
condition. In particular, the theory of the generic extension does not depend on
the generic filter and can be computed in the ground model. Furthermore, in the
presence of a proper class of Woodin cardinals the theory of L(R) is fixed by forcing
and hence the theory of the Pmax -extension of L(R) is fixed as well.
Much of the theory of this generic extension has been worked out and it extends
the known consequences of MM for the structure L(P(ω1 )). In particular, both
theories include the equality 2ℵ0 = ℵ2 ; this is explored in more detail in the next
section. It remains an open problem whether the technical strengthening MM++
of Martin’s Maximum implies that L(P(ω1 )) is a Pmax -extension of L(R); see [4],
[71], [72], [126].
11. Simply definable well orderings of R

A highly nontrivial feature of the Pmax -extension of L(R) is that it satisfies the
Axiom of Choice. By contrast, in the presence of large cardinals, L(R) satisfies
the Axiom of Determinacy which negates AC in a number of essential ways. (For
instance AD implies all subsets of R satisfy the perfect set property, are Lebesgue
measurable, and have the Baire Property.) Since the Pmax -extension is of the form
L(P(ω1 )), the task of verifying the Axiom of Choice reduces to exhibiting a well
ordering of P(ω1 ) which is definable in L(P(ω1 )).
Before proceeding, let us first note that in the presence of MAℵ1 , there is a
bijection between P(ω) and P(ω1 ) which is definable over H(ω2 ). This often
reduces the task of defining a well ordering of P(ω1 ) to the task of defining a well
ordering of P(ω). There are many means of achieving this; one is Solovay’s almost
disjoint coding.
Theorem 11.1. (see [53, 23A]) Assume MAθ . If {Aξ : ξ ∈ θ} is a family of
infinite subsets of ω which have pairwise finite intersection and X ⊆ θ, then there
is a Y ⊆ ω such that
X = {ξ ∈ θ : Aξ ∩ Y is infinite}.
Woodin was the first to show that MM implied that L(P(ω1 )) contained a well
ordering of R. This is done by mimicking the proof of Theorem 7.4 but utilizing a
definable family of stationary subsets of ω2 . Suppose for a moment that S ⊆ ω1 is
stationary. Define S̃ ⊆ ω2 to consist of all ordinals γ such that ω1 ≤ γ < ω2 and if
E is a well ordering of ω1 isomorphic to γ, then
{ν ∈ ω1 : otp(E ν) ∈ S}
contains a club. That is, S̃ is the set of all γ ∈ [ω1 , ω2 ) such that every condition
in P(ω1 )/NSω1 forces that γ ∈ j(S), where j is the induced generic elementary
embedding. Observe that if S and T are disjoint, then so are S̃ and T̃ .
Theorem 11.2. [127, Ch.5] Assume MM. If S ⊆ ω1 is stationary, then {γ ∈
S̃ : cof(γ) = ω} is also stationary.
Woodin’s coding principle can now be described as follows.
φAC : Whenever Si : i ∈ ω and Ti : i ∈ ω are sequences of pairwise disjoint
sets such that each Ti is stationary, there exists a continuous increasing
sequence γξ : ξ ∈ ω1 in ω2 such that if ξ is in Si , then γξ is in T̃i .
An immediate consequence of φAC is that if Ti : i ∈ ω is a sequence of pairwise
disjoint stationary sets, then for every x ⊆ ω, there is a δ < ω2 of cofinality ω1 such
that
x = {i ∈ ω : T̃i ∩ δ is stationary}.
Let δx denote the least such ordinal. If we define x y to mean that δx < δy , then
is a well ordering of P(ω) which is definable from the parameter Ti : i ∈ ω.
Theorem 11.3. [127, Ch.5] MM implies φAC and φAC is true in the Pmax -
extension of L(R).
The following is a variation of φAC which has technical advantages in certain
contexts.
ψAC : Suppose that S, T ⊆ ω1 are stationary and co-stationary. There exists
an uncountable γ < ω2 such that T corresponds to the truth value of the
formula γ ∈ j(S) in the generic extension by the forcing P(ω1 )/NSω1 .
Woodin has also shown that, like φAC , ψAC both is a consequence of MM and holds
in the Pmax -extension of L(R) [127, Ch.5].
A subtle point in the case of both φAC and ψAC is that, even though both
sentences are in the language of (H(ω2 ), ∈, ω1 ) and both follow from MM, it is not
clear if ZFC proves that an instance of these assertions can always be forced with a
NSω1 -preserving forcing (Woodin has showed that this is the case if a measurable
cardinal exists [127, 10.95]). Todorcevic developed the following coding for this
purpose, which relies instead on an ω1 -sequence of reals as a parameter.
θAC : If rξ : ξ ∈ ω1 is a sequence of distinct reals, then for every A ⊆ ω1 ,
there exist α < β < γ < ω2 such that for some continuous ⊆-increasing
sequence Nξ : ξ < ω1 of countable sets which covers γ we have that for
all ξ ∈ ω1 , ξ ∈ A if and only if
Δ(rNξ ∩α , rNξ ∩β ) > Δ(rNξ ∩β , rNξ ∩γ )
Here Δ(r, s) = min{n : r(n) = s(n)} and rN = rotp(N ) if N is a countable set of
ordinals.
Theorem 11.4. [115] Any instance of θAC can be forced with an NSω1 -pre-
serving forcing. In particular, BMM implies θAC holds.
Previously Woodin had proved that BMM together with the existence of a mea-
surable cardinal implies ψAC [127, 10.95]; Asperó previously proved (unpublished)
that BMM implies d = ℵ2 .
The statements φAC , ψAC , and θAC all (seemingly) require the use of forcings
which are improper — they destroy stationary subsets of [θ]ω for some uncountable
cardinal θ. They also a priori require substantial a large cardinal hypothesis for
their consistency. The next combinatorial statement was isolated in [84]. It follows
from the Proper Forcing Axiom and can be forced assuming only the existence
of an inaccessible cardinal. In order to formulate the coding principle, we need
the following specialized notation. For each countable limit ordinal δ, fix a ladder
Cδ ⊆ δ. If M ⊆ N are countable sets of ordinals of limit ordertype and M is
bounded in N , then define w(M, N ) = |Cν ∩ π(M )| where π : N → ν = otp(N ) is
the transitive collapse.
υAC : For every ladder system C on ω1 and every A ⊆ ω1 there is an un-
countable δ < ω2 and a continuous ⊆-chain Nξ : ξ ∈ ω1 cofinal in [δ]ω
such that for all limit ν < ω1 , there is a ν̄ < ν such that if ν̄ < ξ < ν,
then
w(Nξ ∩ ω1 , Nν ∩ ω1 ) < w(Nξ , Nν ) iff Nν ∩ ω1 ∈ A.
This coding principle was the inspiration for the Mapping Reflection Principle
(MRP) [84]. MRP implies υAC and played a central role in the solution to the
basis problem for uncountable linear orders [80]. It also implies κℵ0 = κ for regular
κ > ℵ1 [124] and the failure of (κ) for all regular κ > ℵ1 [84].
A careful examination of the sentences φAC , ψAC , θAC , and υAC reveals that
all yield a well ordering which is Σ1 -definable over the structure (H(ω1 ), ∈, NSω1 ).
These sentences naturally define well orderings of P(ω1 )/NSω1 and the the trans-
ference of this well ordering to one of P(ω1 ) depends on a fixed partition of ω1
into disjoint stationary sets. In [23], Caicedo and Veličković modified the ideas
of [84] and combined them with those from [55] and [123] in order to remove
the reference to the predicate NSω1 . This improvement has important structural
implications concerning models of MM and other forcing axioms.
Theorem 11.5. [23] BPFA implies there is a well ordering of R which is Δ1 -
definable with a parameter which is a ladder system on ω1 .
The details require somewhat more specialized notation and rather than develop
the coding here, we refer the reader to [23]. The Caicedo-Veličković coding also
played an important role in the proof of Theorem 12.2 below.
12. Iterated forcing and the Continuum Hypothesis

It is somewhat rare to encounter consequences of CH which require an elaborate
proof. Typically the argument proceeds by a diagonalization argument of length ω1
in which a continuum of tasks are handled by appropriate book keeping. In some
cases, however, the argument is more subtle and less effective.
The following theorem is a special case of the main result of [34].
Theorem 12.1. Suppose that if Cδ : δ ∈ lim(ω1 ) is a ladder system and
f : ω1 → 2, then there is a function g : ω1 → 2 such that for all limit ordinals δ,
g Cδ ≡∗ f (δ)
(here g ≡∗ n means that {ξ ∈ dom(g) : g(ξ) = n} is finite). Then 2ℵ0 = 2ℵ1 and in
particular CH is false.
To see why this is true, assume the hypothesis of the theorem and fix functions
p, q : ω → ω such that n → (p(n), q(n)) is a surjection onto ω × ω and so that
p(n + 1) ≤ n for all n. Given an f : ω1 → 2, construct a sequence gn : n ∈ ω by

recursion such that g0 = f and gn+1 : ω1 → 2 satisfies:
gn+1 Cδ ≡∗ gp(n) (δ + q(n)).
It is now readily verified that
f → gn ω : n ∈ ω
is a one-to-one function.
It can be shown, however, that for any ladder system and any f , there is a
forcing extension with the same reals (and hence the same ω1 ) in which there is
a g satisfying the conclusion of Theorem 12.1. This theorem therefore represents
an obstruction to being able to iterate within a certain class of forcings while
not introducing new reals. Theorem 7.5, which implies that if NSω1 is saturated
and there is a measurable cardinal then CH is false, is another example of this
phenomenon: the forcing to seal antichains in P(ω1 )/NSω1 does not add new reals.
The theory of iterated forcing in the presence of CH is much more complex
than the theory of iterated forcing which preserves ω1 ; see [99] and also [39]. One
reason for this is that, unlike in the case of preserving ω1 , there is no heuristically
largest class of forcings which we can iterate while preserving that no reals are
added. This was first formally demonstrated in the course of proving the following
result, which is in contrast to Theorem 10.3.
Theorem 12.2. [5] There are two Π2 -sentences ψ0 and ψ1 in the language of
(H(ω2 ), ∈, ω1 ) such that if i is 0 or 1,
(H(ω2 ), ∈, ω1 ) |= ψi ∧ CH
can be forced if there is an inaccessible limit of measurable cardinals but such that
ψ0 ∧ ψ1 implies 2ℵ0 = 2ℵ1 .
Since both ψ0 and ψ1 can be regarded as fragments of Martin’s Maximum, this

result can be interpreted as saying that there is no strongest forcing axiom which
is consistent with CH. The next result gives another consequence of CH which is
also related to the breakdown of a theory of iterated forcing for not adding reals.
Theorem 12.3. [81] Assume CH. There is a tree T of height ω1 with no cofinal
branch such that T is completely proper as a forcing but such that
{(s, t) ∈ T 2 : ht(s) = ht(t) but s = t}
is a countable union of antichains (i.e. T 2 is special off the diagonal).
Here complete properness is a condition which rules out the coding of Devlin and
Shelah described in the beginning of this section. Shelah has shown [99] that this
condition can be supplemented by quite different hypotheses in order to prove that
reals are not introduced in iterated forcing constructions.
Theorem 12.3 shows in particular that CH implies that there is a Baire tree
whose square is special off the diagonal; it is not known whether this is a theorem
of ZFC. (A tree is Baire if the intersection of any countable family of dense open
subsets is dense.)
13. The Semifilter Trichotomy

Next we will examine a combinatorial statement whose origins lie outside of set
theory. While this statement has a wide variety of consequences and may imply
2ℵ0 = ℵ2 , it is incompatible with Martin’s Axiom (and in particular with MM).
Consider the following two questions:
Question 13.1. Does the Čech-Stone remainder of [0, ∞) have only one com-
posant?
Question 13.2. Is it impossible to express the ideal of compact operators on a
separable Hilbert space as the sum of two smaller ideals?
(Recall here that a composant of a point in a continua is the union of all proper
subcontinua which contain the point.) These questions have no apparent relation-
ship to each other and indeed come from different areas of mathematics — continua
theory and functional analysis, respectively. M.E. Rudin proved that under CH,
the remainder of the half-line has more than one component [93]. Similarly, Blass
and Weiss demonstrated that question 13.2 also has a negative answer under CH
[21].
Blass and Weiss [21] isolated the following statement — now known as the Near
Coherence of Filters (NCF) — which is equivalent to each of the above questions
having a positive answer [16]:
NCF: If p and q are two nonprincipal ultrafilters on ω, then there is a finite-
to-one map f : ω → ω such that βf (p) = βf (q).
This statement was soon after proved consistent by Shelah [19]. In fact, NCF holds
in any forcing extension of a model of CH obtained by iterating rational perfect set
forcing ℵ2 times with countable supports [20].
It turns out that NCF is a consequence of an inequality involving cardinal
invariants of the continuum. Recall that a family G of infinite subsets of ω is said
to be groupwise dense if G is closed under modulo finite containment and whenever
f : ω → ω is a finite-to-one function, there is an infinite set X ⊆ ω such that f −1 X
is in G. The cardinal g is the minimum cardinality of a family of groupwise dense
sets which have empty intersection. Recall that the cardinal u is the minimum
cardinality of a base for a nonprincipal ultrafilter on ω.
Theorem 13.3. [18] The inequality u < g implies NCF.
This inequality holds in the models originally used to prove that NCF is consistent
as well as the model obtained by iterating rational perfect set forcing. In fact
u < g is equivalent to a strengthening of NCF known as the Semifilter Trichotomy:
if S ⊆ [ω]ω is closed under taking almost supersets then there is a finite-to-one
function f : ω → ω such that {f S : S ∈ S} is either a family of co-finite sets, an
ultrafilter, or [ω]ω [15] [69].
It is worth noting at this point that Larson has proved the following result,
contrasting Theorem 3.3 above.
Theorem 13.4. [70] The Semifilter Trichotomy implies that medial limits do
not exist.
Quite remarkably, u < g places rather strong restrictions on the other cardinal
invariants of the continuum.
Theorem 13.5. [15] Assume u < g. Then u = b < g = d = c.

In [121], Todorcevic posed the following question in the context of Theorem
13.5.
Question 13.6. Does u < g imply u = ℵ1 ?
Recently Shelah proved the following result:
Theorem 13.7. [97] The inequality g ≤ b+ always holds.
Therefore Todorcevic’s question is equivalent to asking whether u < g implies 2ℵ0 =
ℵ2 .
14. Open problems

I will conclude this paper with a collection of open problems which in some
way connect to the relationship between the continuum and the second uncountable
cardinal. There are many themes represented in these problems:
• What are the simplest hypotheses which imply 2ℵ0 = ℵ2 ?
• Are there foundational principles which entail 2ℵ0 = ℵ2 ?
• Does classification ever necessitate 2ℵ0 = ℵ2 ?
• Are there consequences of 2ℵ0 = ℵ2 which are verifiable?
The bulk of these questions concern the inequality 2ℵ0 < ℵ3 , although I have
included some questions relating to ℵ1 < 2ℵ0 as well.
Problem 14.1. (Todorcevic [109]) Does Todorcevic’s Open Coloring Axiom
imply that the continuum is ℵ2 ?
Problem 14.2. (Todorcevic; see [116]) Does the P-Ideal Dichotomy imply that
the continuum is at most ℵ2 ?
As noted above, both OCA and PID imply that b ≤ ℵ2 because of their influence
on gaps in ω ω /fin.
In Section 6, we saw that it is possible to supplement OCA with another,
related Ramsey-theoretic hypothesis OCA[ARS] in order to prove 2ℵ0 = ℵ2 [82]. It
is similarly unknown if OCA[ARS] implies that the continuum is ℵ2 , although the
difficulties in this problem are of a different nature than in the case of OCA.
Problem 14.3. (Abraham, Rubin, Shelah [2]) Does the OCA[ARS] imply that
the continuum is ℵ2 ?
It is natural to ask whether there is an analog of Theorem 12.2 for 2ℵ0 > ℵ2 .
Problem 14.4. Are there two Π2 -sentences in the language of the structure
(H(ω2 ), ∈, ω1 ) which are each forcibly consistent with 2ℵ0 > ℵ2 but whose conjunc-
tion implies 2ℵ0 ≤ ℵ2 ?
Presumably if this question has a negative answer, then the reason for this is the
existence of a Pmax -like forcing extension of L(R) in which 2ℵ0 is large.
Problem 14.5. (Foreman, Magidor [50]) Assume that L(R) satisfies the Ax-
iom of Determinacy. Is ΘL(R) ≤ ω3 ?
The relationship between the Pmax -extension of L(R) and more conventional
forcing extensions is still not well understood. For instance, the following problem
remains unresolved.
Problem 14.6. (Woodin [127, Ch.11]) Is it possible to force, starting from a

large cardinal hypothesis, that L(P(ω1 )) is a Pmax -extension of L(R)?
Woodin conjectured that the existence of ω 2 Woodin cardinals sufficed for a positive
answer. It is also interesting to ask whether this can be achieved by a semiproper
forcing; this is closely related to the following question (compare to [127, 11.15]).
Problem 14.7. (Woodin [127, Ch.10]) Does M M ++ imply that L(P(ω1 )) is

a Pmax -extension of L(R)?
The reader is referred to [4], [71], [72] for partial results concerning this problem.
Problem 14.8. (Woodin [127, §3.2]) Is it consistent with CH that the nonsta-
tionary ideal on ω1 is saturated?
Problem 14.9. (Woodin [127, 11.7]) Assume that the nonstationary ideal on
ω1 is ℵ1 -dense. Must the continuum be ℵ2 ?
Set-theoretic hypotheses which are needed to prove classification results about

structures of cardinality ℵ1 are closely related to those which imply that 2ℵ0 = ℵ2 .
Problem 14.10. (Moore [83]) Is there a consistent classification of structures

which entails that the continuum is ℵ2 ?
A candidate is the classification of Aronszajn lines. PFA implies that this class
is in many ways analogous to the class of countable linear orderings (see [74]): it
has a two element basis [80], it has a universal element [85], and it is well quasi-
ordered by embeddability [77]. Moreover, this development was precipitated by
the discovery of a new proof that PFA implies that 2ℵ0 = ℵ2 [84].
Problem 14.11. (Moore [83]) Suppose the following are true: (a) every two
ℵ1 -dense nonstationary Countryman lines are isomorphic or reverse isomorphic,
(b) every Aronszajn line can be embedded into ηC , and (c) the class of Aronszajn
lines is well quasi-ordered by embeddability. Must the continuum be ℵ2 ?
The following problem is naturally suggested by the results of [15].
Problem 14.12. Does the Semi-Filter Trichotomy imply that the continuum is
ℵ2 ?
As noted above, this is equivalent to Todorcevic’s question which asks whether

u < g implies u = ℵ1 .
While it is comparatively easy to influence the combinatorics of [ω1 ]ω1 , more
powerful forcings are required to alter the properties of the club filter on ω1 . This
is illustrated, for instance, by the fact that the club filter in V generates the club
filter in any generic extension V [G] by a c.c.c. forcing. By contrast, MAℵ1 has
a substantial impact on the combinatorics of [ω1 ]ω1 and can always be forced by
a c.c.c. forcing. Since methods for producing models in which 2ℵ0 = ℵ2 via iter-
ated forcing are more limited, it may be that 2ℵ0 = ℵ2 impacts the combinatorial
properties of the club filter. This is the motivation behind the following question.
Problem 14.13. (Moore) Suppose that whenever Dα : α ∈ ω1 satisfies that

Dα ⊆ α is closed for each α ∈ ω1 , then there is a club E ⊆ ω1 such that for all
limit α ∈ ω1 , there is a ᾱ < α such that (ᾱ, α) ∩ E is either contained in or disjoint
from Dα . Must the continuum be ℵ2 ? 4
The combinatorial assertion (μ) in this problem is an immediate consequence of

MRP which itself does imply 2ℵ0 = 2ℵ1 = ℵ2 . Whether (μ) is consistent with CH
seems to be a good test question for extending the theory of iterated forcing in the
context of CH.
Next we turn to a pair of questions which concern whether certain consequences
of 2ℵ0 ≤ ℵ2 are verifiable. The first is a topological problem posed by Peter Nyikos.
Problem 14.14. (Nyikos [87]) Does there exist a separable, first countable,
countably compact, noncompact Hausdorff space?
Notice that ω1 , with the order topology, satisfies all of the properties in the problem
except separability. It is known that this question has a positive answer if either
t = ℵ1 or if b = 2ℵ0 . Since t ≤ b, any model in which this problem has a negative
answer must necessarily satisfy 2ℵ0 > ℵ2 . A survey of this question can be found
in [88].
The next question comes from partition calculus and Todorcevic’s and Veličk-
ović’s Ramsey-theoretic analysis of Martin’s Axiom [110] [122].
ω3 ω1,1
Problem 14.15. (Todorcevic [120, 9.3.2]) Is it true that ω3 → ω ω ?
1,1
Here we recall that θθ → ω ω ω means that if f : θ × θ → ω is any function, then f
is constant on the Cartesian product of two infinite sets. The least θ for which this
is true is at most (2ℵ0 )+ . In fact any ZFC bound on this θ which does not involve
1,1
cardinal exponentiation would be of interest. It is know that if κκ → ω ω ω , then
MAκ is equivalent to the assertion that every κ sized subset of a c.c.c. partial order
has a centered subset of cardinality κ [110]. It has also been shown that, in the
2 ω1,1
presence of MAℵ2 , ω ω2 → ω ω is equivalent to Chang’s Conjecture [110].
I will finish the article with the following open-ended question: Is there a natural
hypothesis which implies that the continuum is ℵ3 ? The cardinal ℵ3 is of course
somewhat arbitrary — what is relevant is that it is accessible and greater than ℵ2 . It
would not be of particularly great interest to adapt combinatorial statements which
imply 2ℵ0 = ℵ2 so that they imply 2ℵ0 = ℵ3 . (This may, however, be interesting
at a technical level.) The point is that the proofs that 2ℵ0 = ℵ2 , from various
hypotheses, all employ combinatorial ideas which cannot be modified to produce a
proof of 2ℵ0 = ℵ1 . Typically these ideas concern the combinatorial properties of
ℵ1 . Are there transcendent combinatorial phenomena at ℵ2 which are related to
proofs that 2ℵ0 = ℵ3 ?
4 Added in proof: Asperó and Mota [6] have solved this problem by demonstrating the com-
binatorial statement (μ) is consistent with 2ℵ0 > ℵ2 . It remains open whether (μ) is consistent
with CH.
References
[1] U. Abraham and S. Shelah, Isomorphism types of Aronszajn trees, Israel J. Math. 50 (1985),
no. 1-2, 75–113, DOI 10.1007/BF02761119. MR788070
[2] U. Abraham, M. Rubin, and S. Shelah, On the consistency of some partition theorems for
continuous colorings, and the structure of ℵ1 -dense real order types, Ann. Pure Appl. Logic
29 (1985), no. 2, 123–206, DOI 10.1016/0168-0072(84)90024-1. MR801036
[3] U. Abraham and S. Todorčević, Partition properties of ω1 compatible with CH, Fund. Math.
152 (1997), no. 2, 165–181. MR1441232
[4] D. Asperó, Guessing and non-guessing of canonical functions, Ann. Pure Appl. Logic 146
(2007), no. 2-3, 150–179, DOI 10.1016/j.apal.2007.02.002. MR2320944
[5] D. Asperó, P. Larson, and J. T. Moore, Forcing axioms and the continuum hypothesis, Acta
Math. 210 (2013), no. 1, 1–29, DOI 10.1007/s11511-013-0089-7. MR3037610
[6] D. Asperó and M. Á. Mota, Measuring club-sequences together with the continuum large, J.
Symbolic Logic, to appear.
[7] J. Bagaria, Bounded forcing axioms as principles of generic absoluteness, Arch. Math. Logic
39 (2000), no. 6, 393–401, DOI 10.1007/s001530050154. MR1773776
[8] J. Bagaria, N. Castells, and P. Larson, An Ω-logic primer, Set theory, Trends Math.,
Birkhäuser, Basel, 2006, pp. 1–28, DOI 10.1007/3-7643-7692-9 1. MR2267144
[9] J. E. Baumgartner, All ℵ1 -dense sets of reals can be isomorphic, Fund. Math. 79 (1973),
no. 2, 101–106. MR0317934
[10] J. E. Baumgartner, Applications of the Proper Forcing Axiom. In Handbook of set-theoretic
topology, 913–959. North-Holland, 1981.
[11] J. E. Baumgartner, Chains and antichains in P(ω), J. Symbolic Logic 45 (1980), no. 1,
85–92, DOI 10.2307/2273356. MR560227
[12] J. Baumgartner, J. Malitz, and W. Reinhardt, Embedding trees in the rationals, Proc. Nat.
Acad. Sci. U.S.A. 67 (1970), 1748–1753. MR0314621
[13] J. E. Baumgartner and A. D. Taylor, Saturation properties of ideals in generic extensions.
I, Trans. Amer. Math. Soc. 270 (1982), no. 2, 557–574, DOI 10.2307/1999861. MR645330
[14] M. Bekkali, Topics in set theory: Lebesgue measurability, large cardinals, forcing axioms,
rho-functions, Lecture Notes in Mathematics, vol. 1476, Springer-Verlag, Berlin, 1991. Notes
on lectures by Stevo Todorčević. MR1119303
[15] A. Blass, Groupwise density and related cardinals, Arch. Math. Logic 30 (1990), no. 1, 1–11,
DOI 10.1007/BF01793782. MR1074445
[16] A. Blass, Near coherence of filters. II. Applications to operator ideals, the Stone-Čech re-
mainder of a half-line, order ideals of sequences, and slenderness of groups, Trans. Amer.
Math. Soc. 300 (1987), no. 2, 557–581, DOI 10.2307/2000357. MR876466
[17] A. Blass, Selective ultrafilters and homogeneity, Ann. Pure Appl. Logic 38 (1988), no. 3,
215–255, DOI 10.1016/0168-0072(88)90027-9. MR942525
[18] A. Blass and C. Laflamme, Consistency results about filters and the number of inequivalent
growth types, J. Symbolic Logic 54 (1989), no. 1, 50–56, DOI 10.2307/2275014. MR987321
[19] A. Blass and S. Shelah, There may be simple Pℵ1 - and Pℵ2 -points and the Rudin-Keisler
ordering may be downward directed, Ann. Pure Appl. Logic 33 (1987), no. 3, 213–243, DOI
10.1016/0168-0072(87)90082-0. MR879489
[20] A. Blass and S. Shelah, Near coherence of filters. III. A simplified consistency proof,
Notre Dame J. Formal Logic 30 (1989), no. 4, 530–538, DOI 10.1305/ndjfl/1093635236.
MR1036674
[21] A. Blass and G. Weiss, A characterization and sum decomposition for operator ideals, Trans.
Amer. Math. Soc. 246 (1978), 407–417, DOI 10.2307/1997982. MR515547
[22] J. Brendle, P. Larson, and S. Todorčević, Rectangular axioms, perfect set properties and
decomposition, Bull. Cl. Sci. Math. Nat. Sci. Math. 33 (2008), 91–130. MR2609603
[23] A. E. Caicedo and B. Veličković, The bounded proper forcing axiom and well orderings of
the reals, Math. Res. Lett. 13 (2006), no. 2-3, 393–408, DOI 10.4310/MRL.2006.v13.n3.a5.
MR2231126
[24] G. Cantor. Ein Beitrag zur Mannigfaltigkeitslehre. Journal für die Reine und Angewandte
Mathematik, 37:127–142, 1887.
[25] G. Choquet, Construction d’ultrafiltres sur N (French), Bull. Sci. Math. (2) 92 (1968),
41–48. MR0234405
[26] G. Choquet, Deux classes remarquables d’ultrafiltres sur N. Bull. Sci. Math. (2), 92:143–153,
1968.
[27] J. P. R. Christensen, Topology and Borel structure, North-Holland Publishing Co.,
Amsterdam-London; American Elsevier Publishing Co., Inc., New York, 1974. Descriptive
topology and set theory with applications to functional analysis and measure theory; North-
Holland Mathematics Studies, Vol. 10. (Notas de Matemática, No. 51). MR0348724
[28] J. Cummings, Compactness and incompactness phenomena in set theory, Logic Collo-
quium ’01, Lect. Notes Log., vol. 20, Assoc. Symbol. Logic, Urbana, IL, 2005, pp. 139–150.
MR2143895
[29] J. Cummings, Large cardinal properties of small cardinals, Set theory (Curaçao, 1995;
Barcelona, 1996), Kluwer Acad. Publ., Dordrecht, 1998, pp. 23–39. MR1601964
[30] J. Cummings, M. Foreman, and M. Magidor, Squares, scales and stationary reflection, J.
Math. Log. 1 (2001), no. 1, 35–98, DOI 10.1142/S021906130100003X. MR1838355
[31] H. G. Dales and J. Esterle, Discontinuous homomorphisms from C(X), Bull. Amer. Math.
Soc. 83 (1977), no. 2, 257–259. MR0430786
[32] H. G. Dales and W. H. Woodin, An introduction to independence for analysts, London
Mathematical Society Lecture Note Series, vol. 115, Cambridge University Press, Cambridge,
1987. MR942216
[33] K. J. Devlin, The Yorkshireman’s guide to proper forcing, Surveys in set theory, London
Math. Soc. Lecture Note Ser., vol. 87, Cambridge Univ. Press, Cambridge, 1983, pp. 60–115,
DOI 10.1017/CBO9780511758867.003. MR823776
[34] K. J. Devlin and S. Shelah, A weak version of ♦ which follows from 2ℵ0 < 2ℵ1 , Israel J.
Math. 29 (1978), no. 2-3, 239–247. MR0469756
[35] C. A. Di Prisco and S. Todorcevic, Perfect-set properties in L(R)[U ], Adv. Math. 139
(1998), no. 2, 240–259, DOI 10.1006/aima.1998.1752. MR1654181
[36] A. Dow and K. P. Hart, The measure algebra does not always embed, Fund. Math. 163
(2000), no. 2, 163–176. MR1752102
[37] B. Dushnik and E. W. Miller, Concerning similarity transformations of linearly ordered
sets, Bull. Amer. Math. Soc. 46 (1940), 322-326. MR0001919
[38] T. Eisworth, Selective ultrafilters and ω → (ω)ω , Proc. Amer. Math. Soc. 127 (1999), no. 10,
3067–3071, DOI 10.1090/S0002-9939-99-04835-2. MR1600136
[39] T. Eisworth, J, Tatch Moore. Iterated forcing and the Continuum Hypothesis. In Ap-
palachian set theory 2006-2012, London Math Society Lecture Notes Series, 406:207–244.
Cambridge University Press, 2013.
[40] J. Esterle, Kaplansky’s and Michael’s problems: a survey, Ann. Funct. Anal. 3 (2012), no. 2,
66–88, DOI 10.15352/afa/1399899933. MR2948389
[41] I. Farah, All automorphisms of the Calkin algebra are inner, Ann. of Math. (2) 173 (2011),
no. 2, 619–661, DOI 10.4007/annals.2011.173.2.1. MR2776359
[42] I. Farah, Analytic quotients: theory of liftings for quotients over analytic ideals on the
integers. Mem. Amer. Math. Soc., 148(702), 2000.
[43] I. Farah, Dimension phenomena associated with βN-spaces. Topology Appl., 125(2):279–297,
2002.
[44] V. V. Fedorčuk, A bicompactum whose infinite closed subsets are all n-dimensional
(Russian), Mat. Sb. (N.S.) 96(138) (1975), 41–62, 167. MR0362259
[45] S. Feferman, H. M. Friedman, P. Maddy, and J. R. Steel, Does mathematics need new
axioms?, Bull. Symbolic Logic 6 (2000), no. 4, 401–446, DOI 10.2307/420965. MR1814122
[46] Q. Feng, Homogeneity for open partitions of pairs of reals, Trans. Amer. Math. Soc. 339
(1993), no. 2, 659–684, DOI 10.2307/2154292. MR1113695
[47] M. Foreman, Has the continuum hypothesis been settled?, Logic Colloquium ’03, Lect. Notes
Log., vol. 24, Assoc. Symbol. Logic, La Jolla, CA, 2006, pp. 56–75. MR2207349
[48] M. Foreman, Ideals and generic elementary embeddings, Handbook of set theory. Vols. 1, 2,
3, Springer, Dordrecht, 2010, pp. 885–1147, DOI 10.1007/978-1-4020-5764-9 14. MR2768692
[49] M. Foreman, Potent axioms, Trans. Amer. Math. Soc. 294 (1986), no. 1, 1–28, DOI
10.2307/2000115. MR819932
[50] M. Foreman and M. Magidor, Large cardinals and definable counterexamples to the con-
tinuum hypothesis, Ann. Pure Appl. Logic 76 (1995), no. 1, 47–97, DOI 10.1016/0168-
0072(94)00031-W. MR1359154
[51] M. Foreman, M. Magidor, and S. Shelah, Martin’s maximum, saturated ideals, and non-
regular ultrafilters. I, Ann. of Math. (2) 127 (1988), no. 1, 1–47, DOI 10.2307/1971415.
MR924672
[52] M. Foreman and S. Todorcevic, A new Löwenheim-Skolem theorem, Trans. Amer. Math.
Soc. 357 (2005), no. 5, 1693–1715, DOI 10.1090/S0002-9947-04-03445-2. MR2115072
[53] D. H. Fremlin. Consequences of Martin’s Axiom. Cambridge University Press, 1984.
[54] F. Galvin, Chain conditions and products, Fund. Math. 108 (1980), no. 1, 33–48. MR585558
[55] M. Gitik, Nonsplitting subset of Pκ (κ+ ), J. Symbolic Logic 50 (1985), no. 4, 881–894 (1986),
DOI 10.2307/2273978. MR820120
[56] K. Gödel, What is Cantor’s continuum problem? Amer. Math. Monthly, 54:515–525, 1947.
[57] K. Gödel, Some considerations leading to the probable conclusion that the true power of the
continuum is ℵ2 . In Kurt Gödel: collected works, v. III, 420–422, Oxford University Press,
Oxford, 1995.
[58] F. Hausdorff, Die Graduierung nach dem Endverlauf. Abhandlungen König. Sächsische
Gesellschaft Wissenschaften, Math.-Phys. Kl., 296–334, 1909.
[59] R. Haydon, On dual L1 -spaces and injective bidual Banach spaces, Israel J. Math. 31 (1978),
no. 2, 142–152, DOI 10.1007/BF02760545. MR516250
[60] W. Hurewicz. Une remarque sur l’hypothèse du continu. Fund. Math., 19:8–9, 1932.
[61] W. Just, A modification of Shelah’s oracle-c.c. with applications, Trans. Amer. Math. Soc.
329 (1992), no. 1, 325–356, DOI 10.2307/2154091. MR1022167
[62] W. Just, A weak version of AT from OCA, Set theory of the continuum (Berkeley, CA,
1989), Math. Sci. Res. Inst. Publ., vol. 26, Springer, New York, 1992, pp. 281–291, DOI
10.1007/978-1-4613-9754-0 17. MR1233823
[63] W. Just and A. Krawczyk, On certain Boolean algebras P(ω)/I, Trans. Amer. Math. Soc.
285 (1984), no. 1, 411–429, DOI 10.2307/1999489. MR748847
[64] A. Kanamori, The higher infinite: Large cardinals in set theory from their beginnings, 2nd
ed., Springer Monographs in Mathematics, Springer-Verlag, Berlin, 2003. MR1994835
[65] I. Kaplansky, Normed algebras, Duke Math. J. 16 (1949), 399–418. MR0031193
[66] K. Kunen, A compact L-space under CH, Topology Appl. 12 (1981), no. 3, 283–287, DOI
10.1016/0166-8641(81)90006-7. MR623736
[67] K. Kunen, Some points in βN . Math. Proc. Cambridge Philos. Soc., 80(3):385–398, 1976.
[68] Dj. Kurepa. Ensembles ordonnés et ramifiés. Publ. Math. Univ. Belgrade, 4:1–138, 1935.
[69] C. Laflamme, Equivalence of families of functions on the natural numbers, Trans. Amer.
Math. Soc. 330 (1992), no. 1, 307–319, DOI 10.2307/2154166. MR1028761
[70] P. B. Larson, The filter dichotomy and medial limits, J. Math. Log. 9 (2009), no. 2, 159–165,
DOI 10.1142/S0219061309000872. MR2679437
[71] P. B. Larson, Martin’s maximum and definability in H(ℵ2 ). Ann. Pure Appl. Logic,
156(1):110–122, 2008.
[72] P. B. Larson, Martin’s maximum and the Pmax axiom. Ann. Pure Appl. Logic, 106(1-3):135–
149, 2000.
[73] P. B. Larson, The stationary tower: Notes on a course by W. Hugh Woodin, University
Lecture Series, vol. 32, American Mathematical Society, Providence, RI, 2004. MR2069032
[74] R. Laver, On Fraı̈ssé’s order type conjecture, Ann. of Math. (2) 93 (1971), 89–111.
MR0279005
[75] P. Maddy, Believing the axioms. I, J. Symbolic Logic 53 (1988), no. 2, 481–511, DOI
10.2307/2274520. MR947855
[76] P. Maddy, Believing the axioms. II, J. Symbolic Logic 53 (1988), no. 3, 736–764, DOI
10.2307/2274569. MR960996
[77] C. Martinez-Ranero, Well-quasi-ordering Aronszajn lines, Fund. Math. 213 (2011), no. 3,
197–211, DOI 10.4064/fm213-3-1. MR2822417
[78] A. R. D. Mathias, Happy families, Ann. Math. Logic 12 (1977), no. 1, 59–111. MR0491197
[79] P. A. Meyer, Limites médiales, d’après Mokobodzki, Séminaire de Probabilités, VII (Univ.
Strasbourg, année universitaire 1971–1972), Springer, Berlin, 1973, pp. 198–204. Lecture
Notes in Math., Vol. 321. MR0404564
[80] J. T. Moore, A five element basis for the uncountable linear orders, Ann. of Math. (2) 163
(2006), no. 2, 669–688, DOI 10.4007/annals.2006.163.669. MR2199228
[81] J. T. Moore, Forcing axioms and the continuum hypothesis. Part II: transcending ω1 -
sequences of real numbers, Acta Math. 210 (2013), no. 1, 173–183, DOI 10.1007/s11511-
013-0092-z. MR3037613
[82] J. T. Moore, Open colorings, the continuum and the second uncountable cardinal, Proc.
Amer. Math. Soc. 130 (2002), no. 9, 2753–2759 (electronic), DOI 10.1090/S0002-9939-02-
06376-1. MR1900882
[83] J. T. Moore, The Proper Forcing Axiom. Proceedings of the 2010 ICM, Hyderabad, India,
2:3–29, Hindustan Book Agency, 2010.
[84] J. T. Moore, Set mapping reflection. J. Math. Log., 5(1):87–97, 2005.
[85] J. T. Moore, A universal Aronszajn line. Math. Res. Lett., 16(1):121–131, 2009.
[86] M. Morayne, On differentiability of Peano type functions, Colloq. Math. 48 (1984), no. 2,
261–264. MR758535
[87] P. Nyikos, Progress on countably compact spaces, General topology and its relations to
modern analysis and algebra, VI (Prague, 1986), Res. Exp. Math., vol. 16, Heldermann,
Berlin, 1988, pp. 379–410. MR952624
[88] P. Nyikos, On first countable, countably compact spaces. III. The problem of obtaining
separable noncompact examples. In Open problems in topology, 127–161, North-Holland,
Amsterdam, 1990.
[89] I. I. Parovičenko, On a universal bicompactum of weight ℵ, Dokl. Akad. Nauk SSSR 150
(1963), 36–39. MR0150732
[90] N. C. Phillips and N. Weaver, The Calkin algebra has outer automorphisms, Duke Math. J.
139 (2007), no. 1, 185–202, DOI 10.1215/S0012-7094-07-13915-2. MR2322680
[91] R. Rado, Covering theorems for ordered sets, Proc. London Math. Soc. (2) 50 (1949), 509–
535. MR0029957
[92] R. Rado, Theorems on intervals of ordered sets, Discrete Math. 35 (1981), 199–201, DOI
10.1016/0012-365X(81)90208-9. MR620672
[93] M. E. Rudin, Composants and βN , Proc. Washington State Univ. Conf. on General Topology
(Pullman, Wash., 1970), Pi Mu Epsilon, Dept. of Math., Washington State Univ., Pullman,
Wash., 1970, pp. 117–119. MR0266162
[94] W. Rudin, Homogeneity problems in the theory of Čech compactifications, Duke Math. J.
23 (1956), 409–419. MR0080902
[95] S. Shelah, Cardinal arithmetic, Oxford Logic Guides, vol. 29, The Clarendon Press, Oxford
University Press, New York, 1994. Oxford Science Publications. MR1318912
[96] S. Shelah, Decomposing uncountable squares to countably many chains, J. Combinatorial
Theory Ser. A 21 (1976), no. 1, 110–114. MR0409196
[97] S. Shelah, Groupwise density cannot be much bigger than the unbounded number, MLQ
Math. Log. Q. 54 (2008), no. 4, 340–344, DOI 10.1002/malq.200710032. MR2435897
[98] S. Shelah, Proper forcing. Lecture Notes in Mathematics, 940. Springer-Verlag, Berlin, 1982.
[99] S. Shelah, Proper and Improper Forcing. Springer-Verlag, Berlin, second edition, 1998.
[100] S. Shelah, Reflection implies the SCH, Fund. Math. 198 (2008), no. 2, 95–111, DOI
10.4064/fm198-2-1. MR2369124
[101] S. Shelah and H. Woodin, Large cardinals imply that every reasonably definable set of reals
is Lebesgue measurable, Israel J. Math. 70 (1990), no. 3, 381–394, DOI 10.1007/BF02801471.
MR1074499
[102] W. Sierpiński, Hypothèse du continu (French), Chelsea Publishing Company, New York, N.
Y., 1956. 2nd ed. MR0090558
[103] R. M. Solovay and S. Tennenbaum, Iterated Cohen extensions and Souslin’s problem, Ann.
of Math. (2) 94 (1971), 201–245. MR0294139
[104] J. Steel and S. Zoble, Determinacy from strong reflection, Trans. Amer. Math. Soc. 366
(2014), no. 8, 4443–4490, DOI 10.1090/S0002-9947-2013-06058-8. MR3206466
[105] M. Talagrand, Séparabilité vague dans l’espace des mesures sur un compact (French, with
English summary), Israel J. Math. 37 (1980), no. 1-2, 171–180, DOI 10.1007/BF02762878.
MR599312
[106] S. Todorčević, On a conjecture of R. Rado, J. London Math. Soc. (2) 27 (1983), no. 1, 1–8,
DOI 10.1112/jlms/s2-27.1.1. MR686495
[107] S. Todorčević, Remarks on chain conditions in products. Composito Math., 55(3):295–302,
1985.
[108] S. Todorčević, Directed sets and cofinal types, Trans. Amer. Math. Soc. 290 (1985), no. 2,
711–723, DOI 10.2307/2000309. MR792822
[109] S. Todorčević, Partition Problems In Topology. Amer. Math. Soc., 1989.
[110] S. Todorčević, Remarks on Martin’s axiom and the continuum hypothesis, Canad. J. Math.
43 (1991), no. 4, 832–851, DOI 10.4153/CJM-1991-048-8. MR1127033
[111] S. Todorčević, Conjectures of Chang and Rado and cardinal arithmetic. In Finite and infinite
combinatorics in sets and logic, of NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 411:385–
398, Kluwer Acad. Publ., 1993.
[112] S. Todorcevic, A classification of transitive relations on ω1 , Proc. London Math. Soc. (3)
73 (1996), no. 3, 501–533, DOI 10.1112/plms/s3-73.3.501. MR1407459
[113] S. Todorčević, Comparing the continuum with the first two uncountable cardinals. In Logic
and scientific methods (Florence, 1995), 145–155. Kluwer Acad. Publ., Dordrecht, 1997.
[114] S. Todorčević, A dichotomy for P-ideals of countable sets, Fund. Math. 166 (2000), no. 3,
251–267. MR1809418
[115] S. Todorčević, Generic absoluteness and the continuum, Math. Res. Lett. 9 (2002), no. 4,
465–471, DOI 10.4310/MRL.2002.v9.n4.a6. MR1928866
[116] S. Todorčević, Notes on Forcing Axioms, volume 26 of Lecture Notes Series, Institute for
Mathematical Sciences, National University of Singapore. World Scientific, Singapore, 2014.
[117] S. Todorčević, Partitioning pairs of countable ordinals, Acta Math. 159 (1987), no. 3-4,
261–294, DOI 10.1007/BF02392561. MR908147
[118] S. Todorčević, Oscillations of real numbers, Logic colloquium ’86 (Hull, 1986), Stud. Logic
Found. Math., vol. 124, North-Holland, Amsterdam, 1988, pp. 325–331, DOI 10.1016/S0049-
237X(09)70663-9. MR922115
[119] S. Todorčević, Reflecting stationary sets. handwritten notes, October 1984.
[120] S. Todorčević, Walks on ordinals and their characteristics. Progress in Mathematics, 263,
Birkhäuser, 2007.
[121] S. Todorchevich and I. Farah, Some applications of the method of forcing, Yenisei Series in
Pure and Applied Mathematics, Yenisei, Moscow; Lycée, Troitsk, 1995. MR1486583
[122] S. Todorčević and B. Veličković, Martin’s axiom and partitions, Compositio Math. 63 (1987),
no. 3, 391–408. MR909386
[123] B. Veličković, Forcing axioms and stationary sets, Adv. Math. 94 (1992), no. 2, 256–284,
DOI 10.1016/0001-8708(92)90038-M. MR1174395
[124] M. Viale, The proper forcing axiom and the singular cardinal hypothesis, J. Symbolic Logic
71 (2006), no. 2, 473–479, DOI 10.2178/jsl/1146620153. MR2225888
[125] M. Viale, A family of covering properties. Math. Res. Lett., 15(2):221–238, 2008.
[126] M. Viale, Category forcings, MM+++ , and generic absoluteness for the theory of strong
forcing axioms, J. Amer. Math. Soc. 29 (2016), no. 3, 675–728, DOI 10.1090/jams/844.
MR3486170
[127] W. H. Woodin, The axiom of determinacy, forcing axioms, and the nonstationary ideal, de
Gruyter Series in Logic and its Applications, vol. 1, Walter de Gruyter & Co., Berlin, 1999.
MR1713438
[128] W. H, Woodin, The continuum hypothesis. I. Notices Amer. Math. Soc., 48(6):567–576,
2001.
[129] W. H, Woodin, The continuum hypothesis. II. Notices Amer. Math. Soc., 48(7):681–690,
2001.
[130] W. H. Woodin, Strong axioms of infinity and the search for V , Proceedings of the Inter-
national Congress of Mathematicians. Volume I, Hindustan Book Agency, New Delhi, 2010,
pp. 504–528. MR2827903
Department of Mathematics, Cornell University, Ithaca, New York 14853–4201

E-mail address: justin@math.cornell.edu
Volume 690, 2017
Set-theoretic foundations
Penelope Maddy
Contents
1. Foundational uses of set theory
2. Foundational uses of category theory
3. The multiverse
4. Inconclusive conclusion
References
It’s more or less standard orthodoxy these days that set theory – ZFC, extended
by large cardinals – provides a foundation for classical mathematics. Oddly enough,
it’s less clear what ‘providing a foundation’ comes to. Still, there are those who
argue strenuously that category theory would do this job better than set theory
does, or even that set theory can’t do it at all, and that category theory can. There
are also those who insist that set theory should be understood, not as the study
of a single universe, V, purportedly described by ZFC + LCs, but as the study of
a so-called ‘multiverse’ of set-theoretic universes – while retaining its foundational
role. I won’t pretend to sort out all these complex and contentious matters, but I
do hope to compile a few relevant observations that might help bring illumination
somewhat closer to hand.
1. Foundational uses of set theory

The most common characterization of set theory’s foundational role, the char-
acterization found in textbooks, is illustrated in the opening sentences of Kunen’s
classic book on forcing:
Set theory is the foundation of mathematics. All mathematical
concepts are defined in terms of the primitive notions of set and
membership. In axiomatic set theory we formulate . . . axioms
2010 Mathematics Subject Classification. Primary 03A05; Secondary 00A30, 03Exx, 03B30,
18A15.
Key words and phrases. Foundations, ZFC, category theory, multiverse.
It’s an honor to be included in this 60th birthday tribute to Hugh Woodin, who’s done so
much to further, and often enough to re-orient, research on the fundamentals of contemporary set
theory. I’m grateful to the organizers for this opportunity, and especially, to Professor Woodin
for his many contributions.
2017
289
290 PENELOPE MADDY
about these primitive notions . . . From such axioms, all known

mathematics may be derived. ([Kunen, 1980], p. xi)
These days, familiarity has dulled our sense of just how astounding this fact really
is. In his introductory text, Enderton makes sure that his students appreciate its
scope and power:
It is sometimes said that ‘mathematics can be embedded in set
theory’. This means that mathematical objects (such as numbers
and differentiable functions) can be defined to be certain sets.
And the theorems of mathematics (such as the fundamental the-
orem of calculus) then can be viewed as statements about sets.
Furthermore, these theorems will be provable from our axioms.
Hence our axioms provide a sufficient collection of assumptions
for the development of the whole of mathematics – a remarkable
fact. ([Enderton, 1977], pp. 10–11)
The question for us is: what’s the point of this exercise? What goal, properly
thought of as ‘foundational’, is served by this ‘embedding’ ?
A glance at the history delivers a hint. At the turn of the last century, Hilbert
had just proved the consistency of geometry by relying on analysis, so he set the
task of proving the consistency of analysis as the second on his famous list of
problems in [Hilbert, 1900]. In earlier correspondence with Cantor, Hilbert had
come to understand that there could be no set of all alephs, and to feel the need for
a consistent axiomatization of set theory, as well.1 Zermelo arrived in Göttingen in
1897:
When I was a Privatdozent in Göttingen, I began, under the
influence of D. Hilbert, to whom I owe more than to anybody else
with regard to my scientific development, to occupy myself with
questions concerning the foundations of mathematics, especially
with the basic problems of Cantorian set theory. (Quoted in
[Ebbinghaus, 2007], p. 28)
By 1908, in his famous axiomatization, Zermelo drew the explicit connection be-
tween set theory and analysis:2
Set theory is that branch of mathematics whose task is to investi-
gate mathematically the fundamental notions ‘number’, ‘order’,
and ‘function’, taking them in their pristine, simple form, and to
develop thereby the logical foundations of all of arithmetic and
analysis. ([Zermelo, 1908], p. 200)
Subsequent developments extended set theory’s reach to the whole of classical math-
ematics, as indicated in the quotations from Kunen and Enderton.
Suppose, then, that your goal is to prove something or other about the vast
variety of classical mathematics – for the Hilbert school, its consistency. To do this,
you first need to corral it all into some manageable package, and set theory turned
out to be up to that task. Of course Gödel saw to it, with his second incompleteness
1 See[Ebbinghaus, 2007], pp. 42–43.

2 [Ebbinghaus, 2007], pp. 76–79, highlights the role of Hilbert’s program among the moti-
vations for Zermelo’s axiomatization. Apparently Zermelo was keen to include a consistency proof
in his [Zermelo, 1908], but Hilbert encouraged him to publish it as it stood, in part, Ebbinghaus
reports, because Hilbert knew ‘that Zermelo needed publications to promote his career’ (p. 78).
SET-THEORETIC FOUNDATIONS 291
theorem, that the project didn’t turn out as Hilbert and his followers had hoped,
but the fact remains that the most sweeping moral of Gödel’s work – that classical
mathematics (if consistent) can’t prove its own consistency – is only possible given
the set theory’s codification of the entire subject into a neat set of simple axioms.
And subsequently, the set-theoretic codification made it possible to settle other
important questions of provability and unprovability, sparing the profession from
sadly doomed efforts. So the ‘embedding’ of mathematics in set theory has this
clear use. Presumably we’d all agree that the goal of proving something general
about classical mathematics counts as ‘foundational’ in some sense, so here we do
find set theory playing a foundational role.
Of course, this Meta-mathematical Corral is hardly the only foundational
role set theory has been thought to play, but I think two of these can be dismissed
as spurious. The first sees the ‘embedding’ of classical mathematics in set theory,
often called a ‘reduction’ of classical mathematics to set theory, as the immedi-
ate descendant of Frege’s Logicism. This line of thought takes Frege’s project to
be roughly epistemological: if mathematics can be reduced to logic, then knowing
a mathematical fact is reduced to knowing a logical fact; assuming that we have
an account of logical knowledge, or at least that finding such an account is a less
daunting prospect than finding an account of mathematical knowledge had previ-
ously appeared to be, this reduction is a clear epistemological gain.3 Of course
Frege’s logic turned out to be inconsistent and set theory has taken its place, but
the epistemological analysis is supposed to carry over: we know the theorems of
mathematics because we know the axioms of set theory and prove those theorems
from them. Thus the problem of mathematical knowledge reduces to the problem
of knowing the set-theoretic axioms.
The trouble with this picture is that it’s obviously false: our greatest math-
ematicians know (and knew!) many theorems without deriving them from the
axioms. The observation that our knowledge of mathematics doesn’t flow from the
fundamental axioms to the theorems goes back at least to Russell – who emphasized
that the logical order isn’t the same as the epistemological order, that the axioms
might gain support from the familiar theorems they generate, not vice versa4 –
and is prominent in Zermelo – who defended the Axiom of Choice on the basis of
its consequences.5 As is often noted, a well-known fact of arithmetic or analysis
or geometry may be considerably more certain than the axioms of set theory from
which it derived. For that matter, to make a Wittgensteinian point,6 in most cases
we only believe that there is a proof of a certain mathematical theorem from ZFC
because we believe the theorem on the basis of its ordinary proof and we believe
that all theorems of mathematics are ultimately provable from ZFC! So this pur-
ported foundational use of set theory, as the Epistemic Source of all mathemat-
ical knowledge, is a failure. But this casts no doubt on the Meta-mathematical
Corral.
3 It isn’t obvious that Frege himself saw the matter quite this way: much of his interest is in
objective grounding relations between propositions, which are independent of our human ways of
finding things out. See, e.g., [Burge, 1998].
4 See [Russell, 1907].
5 See [Zermelo, 1908a].
6 [Wittgenstein, 1978], Part III, is thinking of the relations between ordinary decimal cal-
culations and the corresponding inferences in, say, the notation of Principia Mathematica, but
the upshot is the same.
292 PENELOPE MADDY
The other purported foundational role for set theory that seems to me spurious
is what might be called the Metaphysical Insight. The thought here is that the
set-theoretic reduction of a given mathematical object to a given set actually reveals
the true metaphysical identity that object enjoyed all along. Benacerraf famously
argued that this can’t be right, because, for example, Zermelo took the natural num-
bers to be ∅, {∅}, {{∅}}, . . . , von Neumann took them to be ∅, {∅}, {∅, {∅}}, . . . , and
there’s no principled reason to choose one over the other.7 There are practical rea-
sons to prefer the von Neumann ordinals – they generalize easily to the transfinite,
for example – but this sort of thing isn’t an indicator of ‘what the numbers really
are’. Or so the argument goes.
Of course the practice of set theory is filled with even more arbitrary choices,
like the conventional preference for the Kuratowski ordered pair. It’s worth noticing
that Kunen speaks of ‘defining mathematical concepts’, not identifying mathemat-
ical objects, and Enderton, who does refer to ‘mathematical objects’, only speaks
of how they ‘can be defined’ and how theorems about them ‘can be viewed’. In
yet another textbook, Moschovakis makes the thought behind this circumspection
explicit:
A typical example of the method we will adopt is the ‘identifica-
tion’ of [the geometric line] with the set . . . of real numbers. . . .
What is the precise meaning of this ‘identification’ ? Certainly
not that points are real numbers. . . . What we mean by the ‘iden-
tification’ of [the line] with [the reals] is that the correspondence
. . . gives a faithful representation . . . which allows us to give
arithmetic definitions for all the useful geometric notions and to
study the mathematical properties of [the line] as if points were
real numbers. . . . In the same way, we will discover within the
universe of sets faithful representations of all the mathematical
objects we need, and we will study set theory . . . [8] as if all
mathematical objects were sets. ([Moschovakis, 1994],
pp. 33–34, emphasis in the original)
The trick, in each case, is to identify the conditions that a ‘faithful representation’
must satisfy. For the case of ordered pairs, this is easy: two of them should be equal
iff their first elements are equal and their second elements are equal. The case of the
natural numbers is more demanding: a set of sets with its operations should satisfy
the (full second-order) Peano Postulates. For our purposes, the simple point is
these set-theoretic reductions don’t give any sort of deep metaphysical information
about the nature of the line or of ordered pairs or of natural numbers, nor are they
so intended. Metaphysical Insight, like Epistemic Source, is spurious, leaving
only Meta-mathematical Corral.
But this is hardly the end of the story. The impressive mathematical innova-
tions present in the earliest invocations of sets – Cantor’s in his work on trigonomet-
ric series and Dedekind’s on ideals in the early 1870s – aren’t actually foundational
in character, but around the same time, Dedekind also undertook a project that
was explicitly foundational: to ‘find a purely arithmetical and perfectly rigorous
foundation for the principles of infinitesimal calculus’ ([Dedekind, 1872], p. 767).
7 See
[Benacerraf, 1965].
8 In
the ellipsis, Moschovakis writes, ‘on the basis of the lean axiom system of Zermelo’,
perhaps gesturing toward the Meta-mathematical Corral.
Charged to teach the subject, Dedekind laments ‘the lack of a truly scientific foun-
dation’, finds himself forced to take ‘refuge in geometric evidence’ which ‘can make
no claim to being scientific’ (ibid.). Comparing the line with the rational numbers,
we see in the rationals ‘a gappiness, incompleteness, discontinuity’ and in the line
an ‘absence of gaps, completeness, continuity’, so the key question emerges:
In what then does this continuity consist? Everything must de-
pend on the answer to this question, and only through it shall
we obtain a scientific basis for the investigation of all continuous
domains. ([Dedekind, 1872], p. 771)
The then-current understanding rested only on those geometric intuitions, and on
vague remarks upon the unbroken connection in the smallest
parts [by which] obviously nothing is gained. (Ibid.)
So, the challenge was to replace these unscientific vagaries with
a precise characteristic of continuity that can serve as the basis
for valid deductions. (Ibid.)
And this, of course, is what Dedekind goes on to provide, in his theory of cuts in
the rationals – using set-theoretic machinery.
At first glance, this may look like just one more instance of the set-theoretic
reduction that underlies the Meta-mathematical Corral, but in fact there’s
something more going on. It isn’t that we have an explicit mathematical item –
the ordered pair, or the numbers as described by Peano – which we then ‘identify’
with a set that can play the same role, do the same jobs. Instead, in this case, we
have a vague picture of continuity that’s served us well enough in many respects,
well enough to generate and develop the calculus, but now isn’t precise enough to
do what it’s being called upon to do: allow for rigorous proofs of the fundamental
theorems. For that we need something more exact, more precise, which Dedekind
supplies. This isn’t just a set-theoretic surrogate, designed to reflect the features of
the pre-theoretic item; it’s a set-theoretic improvement, a set-theoretic replacement
of an imprecise notion with a precise one. So here’s another foundational use of set
theory: Elucidation. The replacement of the imprecise notion of function with
the set-theoretic version is another well-known example.9
To isolate another productive foundational use of set theory, we need to trace
some of the profound shifts in the theory and practice of mathematics that took
place over the course of the 19th century, starting with geometry. Since the ancients,
geometry had been closely associated with ordinary diagrams, and late in the 1700s,
Kant developed his elaborate theory of spatiotemporal intuition to undergird this
approach. This comforting view of the matter was challenged early in the 19th
century, as it became clear how much there was to be gained by viewing ordinary
geometry from the richer, projective point of view – with its ‘imaginary points’
(points with complex numbers as coordinates) and ‘points at infinity’ (points where
parallel lines meet). Of course these new points can’t be pictured, so the work was
roundly resisted at first on the grounds that ‘it keeps itself too much aloof from
all intuition, which is the essential trait of mathematical knowledge’.10 The very
notion of invisible, unvisualizable ‘points’ where two disjoint circles or two parallel
lines in fact ‘intersect’ was not only contrary to the intuitive ground of geometry,
9 For a quick overview of the history, see [Maddy, 1997], pp. 118–126.
10 The remark comes from Möbius, he of the strip. Quoted in [Nagel, 1979], p. 219.
294 PENELOPE MADDY
but to plain common sense. Still, as Nagel remarks in his historical survey, on the
closely related subject of negative and complex numbers:
Scandal or not, there were few mathematicians who did not rec-
ognize their value or whose logical consciences were so tender
than they would not use them. ([Nagel, 1979], p. 202)
The same could be said of the new projective geometry.
I bring up this episode because the ‘scandal’ was resolved in mid-century by von
Staudt, using proto-set-theoretic techniques, in particular a precursor of the method
of equivalence classes: for example, the point at infinity where two horizontal lines
meet is identified with what we’d now see as the set of lines parallel to these two,
and a given point at infinity is on a particular line if that line is in the set with
which that point at infinity has been identified. In this and related ways, von
Staudt managed to build surrogates for heretofore suspicious, possibly dangerous
new items (like points at infinity) out of uncontroversial, unproblematic materials
(ordinary lines), and to redefine the relevant relations so as to validate the existing,
successful theory. His goal in all this is to remove any queasiness we might have
about the legitimacy or coherence of the new, un-intuited items.
As time went by, it became clear that the construction tools needed for this
‘building’ process – tools von Staudt regarded as ‘logical’ – were actually set-
theoretic in character. Speaking of the operations codified in Zermelo’s axioms,
Burgess writes:
A crucial fact . . . is that these are essentially the existence as-
sumptions needed to get new spaces or number systems or what-
ever from old ones . . . in the manner of nineteenth-century in-
troduction of auxiliaries for the study of traditional spaces or
number systems. Indeed, the constructions of the auxiliaries in
question can be, and now in retrospect are, viewed as essentially
‘set-theoretic’ constructions, though some of them actually ante-
date Cantor. ([Burgess, 2015], p. 76, emphasis in the original)
This striking fact – that the methods of von Staudt and others all fall within the
few closure principles used by the early set theorists and codified by Zermelo –
this fact is what eventually made possible what we now know as the set-theoretic
reduction of classical mathematics.
But the story of queasiness-removal doesn’t end there. In the late 1800s, pure
mathematics was on the rise, and with it, the axiomatic method; in place of von
Staudt-like constructions, new fields were introduced instead by an explicit set of
axioms. In his comprehensive history, Kline describes the situation this way:
Mathematics, from a logical standpoint, was by the end of the
nineteenth century a collection of structures each built on its
own system of axioms. . . . As long as mathematics was regarded
as the truth about nature, the possibility of contradictory theo-
rems . . . would have been regarded as absurd. ([Kline, 1972],
p. 1038)
– but without that worldly backing, the question of which of these axiom systems
could be trusted became acute. This new queasiness could best be removed by a
proof of consistency, and set theory again presented itself, now as the source for
such proofs. We return in a moment to the difference between von Staudt-like
surrogates and axiomatic consistency proofs – roughly, between proving from ZFC
that there’s a structure of a certain sort, and proving from ZFC that there’s a model
that thinks there’s a structure of a certain sort – but for now what’s important is
that neither of these effectively removes queasiness unless ZFC itself is known to be
trustworthy.
Zermelo felt the force of these considerations, remarking that
I have not yet . . . been able to prove rigorously that my axioms
are consistent, though this is certainly very essential; instead I
have had to confine myself to pointing out now and then that the
antinomies discovered so far vanish one and all if the principles
here proposed are adopted as a basis. But I hope to have done at
least some useful spadework hereby for subsequent investigations
of such deeper problems. ([Zermelo, 1908], pp. 200–201)
We now smile, perhaps a bit wistfully, at this optimism.11 In the decades since
Gödel dashed these fond hopes, the hierarchy of large cardinals has arisen as a
measure of consistency strength, and the early foundational goal of conclusive
queasiness-removal has given way to a more nuanced matter of Risk Assessment.
So, for example, in the abstract for a recent ASL talk, Voevodsky speaks of the role
of set theory in his program of ‘univalent foundations’:12
Univalent foundations provide a new approach to the formal rea-
soning about mathematical objects. The languages which arise
in this approach are much more convenient for doing serious
mathematics than ZFC at the cost of being much more complex.
In particular, the consistency issues for these languages are not
intuitively clear. Thus ZFC retains its key role as the theory
which is used [to] ensure that the more and more complex lan-
guages of the univalent approach are consistent. ([Voevodsky, 2014],
p. 108)
Or as he puts it, more carefully, in the slides for this talk:
Set theory will remain the most important benchmark of consis-
tency. . . . each new addition to the . . . language will require for-
mal ‘certification’ by showing, through formally constructed in-
terpretation, that it is at least as consistent as ZFC. ([Voevodsky, 2013],
slide 21)
Obviously this generalizes to ‘at least as consistent as ZFC + one or another large
cardinal’.
Notice that Risk Assessment in either form isn’t the same as Meta-math-
ematical Corral: the point isn’t to round up all classical mathematical items into
one simple package, so as to prove something about all of it all at once, but to
assess a particular new, somehow dangerous or suspicious item to determine just
how risky it is. And it differs from Elucidation as well: von Staudt had before him
a perfectly functional geometric practice with ideal points; his task wasn’t to make
that practice more precise, and thus more functional, but to reproduce it chapter-
and-verse in a way that was less worrisome; conversely, Dedekind’s concern wasn’t
11 Recall footnote 2, above.
12 For the foundational role ‘univalent foundations’ might itself be intended to play, see foot-
note 37.
296 PENELOPE MADDY
that the real numbers were somehow worrisome, but that they weren’t sufficiently
precise to support the practice, weren’t sufficiently functional. So we have at this
point these three distinct foundational uses for set theory, along with a pair of
spurious ones – Metaphysical Insight, and Epistemic Source.
There remains one more, quite familiar line of foundational thought, namely the
idea that set theory provides decisive answers to questions of ontology13 and proof:
if you want to know whether or not a so-and-so exists, see whether one can be found
in V; if you want to know whether or not such-and-such is provable, see whether it
can be derived from the axioms of set theory. (In fact, both of these are provability
conditions: a so-and-so ‘can be found in V’ iff the existence of a so-and-so can be
proved from the axioms.) This is sometimes expressed with the rhetorical flourish
that set theory is the ‘final court of appeal’ on matters of proof and ontology. I
should confess that I’ve indulged in this flourish myself ([Maddy, 1997], p. 26),
making it sound as if classical mathematics must bow to the dictates of set theory,
but in practice I’ve taken this foundational role to place methodological constraints
on set theory, the founding theory, not on classical mathematics, the theory to be
founded. In particular, I argued that set theory, if it was to play this role, should
be as generous as possible – so as not to curtail pure mathematics – and should
be given by one unified theory that’s as decisive as possible – so as to provide
unequivocal answers to questions of ontology and proof.14
Fearing, then, that the ‘final court of appeal’ is something of an exaggeration,
let’s look a bit more closely to discern what foundational uses are actually in play.
At the very least, there’s the plain sociological fact that derivation from ZFC is
generally regarded as standard of proof in mathematics: in practice, the availability
of the axioms of ZFC goes without saying; if stronger assumptions are in play, this
is explicitly acknowledged;15 if only weaker assumptions are needed, this is noted
to give a more nuanced picture of the dependencies involved. Burgess observes:
There are . . . no official censors preventing a group of dissi-
dents from founding a journal of their own, in which as a mat-
ter of editorial policy results must be proved according to the
group’s restrictive standards (or as the case may be, results may
be proved making use of the group’s preferred additional hy-
potheses), rather than presented as they would be in a journal
enforcing the orthodox standard . . . No one dissident school of
thought, however, produces work of sufficient volume at a suffi-
cient pace to keep a journal of high standards following such a
policy coming out regularly. ([Burgess, 2015], p. 118)
In this foundational role, then, formal derivation in set theory serves as a Shared
Standard of what counts as a proof.
13 This term isn’t intended in any philosophically loaded way: I just mean what the practice
asserts to exist, leaving the semantic or metaphysical issues open. [Mac Lane, 1981], p. 468,
and [Feferman, 1977], p. 151, both see set theory as inseparable from a kind of Platonistic
metaphysics, but I’ve argued that employing a rich set-theoretic ontology is consistent with being
an Arealist (as in [Maddy, 2011]).
14 These are the methodological maxims maximize and unify from [Maddy, 1997].
15 [Burgess, 2015], p. 177, footnote 11, notes an exception: proofs appealing to
Grothendieck’s work sometimes omit what comes to an appeal to inaccessibles. See Section 2
below.
But what lies behind this sociological fact? Partly there’s the recognition that
formal derivation turns out to be a good mathematical model for the scope of
human proving activity, but why from these particular axioms? To shed light on
this question, recall the ‘ontological’ component of the ‘final court of appeal’: there
is a so-and-so if one can be found in V, if the existence of a so-and-so can be
proved from the axioms. A few pages back, in connection with Risk Assessment,
we noted a contrast between queasiness-removal by outright existence proof and
queasiness-removal by consistency proof. Now we find the purported standard of
existence apparently promoting von-Staudt-like construction over the more lenient
Hilbert-like idea that consistency of the appropriate set of axioms is enough. To
illustrate with an example, suppose a mathematician wants to know: is there a
definable (projective) well-ordering of the reals? In ZFC alone, the question can’t be
answered, but assuming, as many set theorists do, that ZFC+LCs is the appropriate
measure, the answer is no. Still, the opposite answer can be had in L. In light of
that fact, would we really want to shut the door on this mathematician? For that
matter, why shouldn’t we follow Hilbert and open that door to the existence claims
of any consistent set of axioms? Why does ‘final court’ insist that we restrict
ourselves to exactly what happens in V?
The source of this more stringent ‘final court’ criterion is simple: the branches of
modern mathematics are intricately and productively intertwined, from coordinate
geometry, to analytic number theory, to algebraic geometry, to topology, to modern
descriptive set theory (a confluence of point-set topology and recursion theory), to
the kind of far-flung interconnections recently revealed in the proof of Fermat’s Last
Theorem. What’s needed is a single arena where all the various structures studied
in all the various branches can co-exist side-by-side, where their interrelations can
be studied, shared fundamentals isolated and exploited, effective methods exported
and imported from one to another, and so on. Burgess puts the point forcefully:
Interconnectedness implies that it will no longer be sufficient to
put each individual branch of mathematics separately on a rigor-
ous basis . . . To guarantee that rigor is not compromised in the
process of transferring material from one branch of mathematics
to another, it is essential that the starting points of the branches
being connected should . . . be compatible. . . . The only obvious
way to ensure compatibility of the starting points . . . is ulti-
mately to derive all branches from a common, unified starting
point. ([Burgess, 2015], pp. 60–62)
Set theory’s universe, V, provides the Generous Arena in which all this takes
place, and that’s why the ‘final court’ condition takes the form it does: to be a
full participant in mathematical interaction, a so-and-so must appear along-side
the full range of its fellows, with all the tools of construction and interaction fully
available.16
16 Of course Shared Standard and Generous Arena depend on the same facts of set-
theoretic reduction as Meta-mathematical Corral: that formal proof is a good model of

provability by humans and that the axioms of set theory codify the fundamental assumptions
of classical mathematics. What separates them are the uses to which these facts are being put:
in Meta-mathematical Corral, ‘derivable in ZFC’ functions as model for ‘provable in classical
mathematics’; in Shared Standard, it’s used as a benchmark for what counts as a legitimate in-
formal proof; in Generous Arena, V brings all the objects and methods of classical mathematics
together for fruitful interaction. As foundational uses, these are distinct.
298 PENELOPE MADDY
Viewed in this light, our flat answer to the mathematician’s question deserves a
bit of shading. A definable well-ordering of the reals occurs in L, a well-understood
structure where all the axioms of ZFC are satisfied. This means that L itself is
a fairly generous arena: all the usual constructions of ZFC are available; all the
standard theorems from all the familiar branches of the subject are in place; so
serious mathematical work can be carried out in the presence of a definable well-
ordering. The drawback is that care has to be taken with export and import. But
the pure Hilbert-style case is different: proving that there is a model for a set of
axioms that implies the existence of a so-and-so can provide Risk Assessment,
but it doesn’t by itself install a so-and-so in the Generous Arena where classical
mathematics takes place.
Stripped of its pretensions, then, the ‘final court’ condition comes down to
this: a Shared Standard of proof designed to generate a Generous Arena for
the pursuit and flourishing of pure mathematics. From this point of view, the
requirement that assumptions beyond ZFC be noted explicitly makes perfect sense:
our mathematician is welcome to work with his definable well-ordering in L as long
as he recognizes that his conclusions can’t be freely exported to the more standard
arena with ZFC alone, and that work predicated on the popular assumption of
large cardinals can’t be imported. On the plus side, he gains all the benefits of our
existing understanding of the details of life in L, and his work can be understood
as further illuminating what goes on there. Work in ZFC + V = L has obvious
value, but of course some extensions of ZFC are more mathematically rewarding
than others!17
In sum, then, we’ve collected an array of important uses of set theory that
ought to qualify as ‘foundational’ – Meta-mathematical Corral, Elucidation,
Risk Assessment, Shared Standard and Generous Arena – as well as a pair
that are spurious – Metaphysical Insight and Epistemic Source. The famed
set-theoretic reduction of classical mathematics lies in the background for most of
this, fruitful and spurious alike. Details aside, we see that the remarkable fact
of the reduction doesn’t, by itself, dictate any particular foundational use. For
this reason, it seems to me counter-productive to begin from the question: does a
given theory provide a foundation for classical mathematics? Rather, we should be
asking: what foundational purposes does the given theory serve, and how? With
this in mind, let’s turn to set theory’s famous rival.
2. Foundational uses of category theory

Category theory was introduced in the 1940s, by which time the notion of
set-theoretic foundations had become mainstream orthodoxy. Like Cantor’s initial
appeals to sets, Eilenberg and Mac Lane’s categories emerged in the pursuit of
straightforwardly mathematical goals. Foundational concerns first came into the
picture in the form of criticisms of set theory’s foundational aspirations. Mac Lane
grants the effectiveness to date of set-theoretic foundations:
The prior situation in the foundations of Mathematics had in
one respect a very simple structure. One could produce one for-
mal system, say Zermelo-Fraenkel set theory, with the property
17 For some purposes, it’s useful to consider arenas not quite as generous as full ZFC, e.g.,
L(R), where Choice is false (assuming again that there are large cardinals in V), but all sets of
reals are Lebesgue measurable.
that all ordinary operations of practising Mathematicians could

be carried out within this one system and on objects of this sys-
tem. . . . ‘every’ Mathematical object was or could be defined to
be a set, and . . . all of the arguments about these objects could
be reduced to the axioms of ZF set theory. . . . This one-formal-
system ‘monolithic’ approach has . . . been convenient for spe-
cialists on foundations ever since Frege and Whitehead-Russell.
On the one hand, all the classical nineteenth century problems of
foundations (the construction of integers, real numbers, analysis
. . . ) could be stated in this one system. On the other hand,
alternative formal system[s] could . . . be tested by comparison
(as to strength or relative consistency) with this one system.
([Mac Lane, 1971], pp. 235–236)
Here we find gestures toward Generous Arena, Risk Assessment, Elucidation
(for the case of the reals)18 , Meta-mathematical Corral (in the usefulness to
‘specialists in foundations’) and Shared Standard19 . Mac Lane’s only complaint
is that ‘this happy situation no longer applies to the practice of category theory’
([Mac Lane, 1971], p. 236).
What is it about category theory that purportedly ends the reign of set-
theoretic foundations?
Categorical algebra has developed in recent years as an effective
method of organizing parts of mathematics. Typically, this sort
of organization uses notions such as that of the category G of
all groups. This category consists of two collections: The col-
lection of all groups G and the collection of all homomorphisms
. . . of one group G into another one; the basic operation in this
category is the composition of two such homomorphisms. To re-
alize the intent of this construction it is vital that this collection
G contain all groups; however, if ‘collection’ is to mean ‘set’
. . . this intent cannot be directly realized. ([Mac Lane, 1971],
p. 231)
Because there is no set of all groups, set theory can’t properly found category
theory.20 So the argument goes.
To overcome this problem, Grothendieck devised a system of ‘universes’ essen-
tially equivalent to Zermelo’s hierarchy of Vκ ’s for κ inaccessible,21 and all parties
agree that surrogates for any actual use of category theory can be found within
some Vκ .22 Burgess writes:
18 Cf. [Mac Lane, 1986], p. 362: ‘this approach to Mathematics has the advantage that
every concept can be made absolutely clear and explicit’.
19 Cf. [Mac Lane, 1986], p. 377: ‘we have now stated an absolute standard of rigor: A
Mathematical proof is rigorous when it is (or could be) written out in the first order predicate
language [with membership as the only non-logical symbol] as a sequence of inferences from the
axioms ZFC, each inference made according to one of the stated rules’.
20 A move to NGB would introduce the category of all groups as a proper class, but it would
still leave out important functor categories. See [Mac Lane, 1971].
21 See [Zermelo, 1930].
22 See, e.g., [Feferman, 1977], p. 155. Grothendieck’s accomplishment here runs parallel to
von Staudt’s: show that the job of a worrisome item (point at infinity, category of all groups) can
be carried out with an uncontroversial item (a set of lines, a set of ‘enough’ groups).
300 PENELOPE MADDY
For applications . . . one doesn’t need a category of literally all

groups . . . It is always enough to have a category of ‘enough’
groups, though how many is enough may vary from application
to application. . . . Grothendieck’s hypothesis is that every set,
however large, belongs to some local universe [that is, some Vκ ]:
The ‘global universe’ . . . is simply the union of increasingly large
local universes. ([Burgess, 2015], p. 174)
Why doesn’t this settle the matter? Assuming that Risk Assessment is the foun-
dational use of set theory in question here, as Mac Lane sometimes suggests,23
then we have our answer: category theory is no worse off than ‘ZFC+ many Inac-
cessibles’, which doesn’t take us far beyond ZFC itself.24 With this mild extension
of ZFC, Meta-mathematical Corral would also be achieved, nor is there any
obvious difficulty for any of the other foundational uses of set theory.
But Mac Lane knows all this and clearly isn’t satisfied:
Given any universe U , one can always form the category of
all categories within U . This is still not that will-of-the-wisp,
the category of all categories überhaupt. ([Mac Lane, 1971],
p. 234)
What exactly is the imagined deficit? For any particular use of the category of
‘all’ groups or the category of ‘all’ categories, there’s a set-theoretic surrogate that
does the job. The objection here appears to be that the surrogate isn’t the real
thing. But why should it be? In this form, the objection begins to sound analogous
to the complaint, in Benacerraf’s context, that the von Neumann ordinals aren’t
acceptable set-theoretic surrogates for the natural numbers because the actual 2
isn’t an element of the actual 3! In other words, Mac Lane appears to be drifting
into the demand that a foundation provide Metaphysical Insight – a demand
that set theory never properly took on in the first place.
However that may be, the problem of ‘founding’ unlimited categories was taken
up by Feferman in the late 1960s and revisited most recently in [Feferman, 2013].25
Feferman explicitly identifies the issue as one of Risk Assessment, and the theory
whose risk he’s interested in assessing is one that allows the formation of
the category of all structures of a given kind, e.g., the category . . .
of all groups, . . . of all topological space[s], . . . of all categories.
([Feferman, 2013], p. 9)
In other words, not just, for example, the category of ‘enough groups’ (as Burgess
puts it) for some practical purpose, but the category of all groups. These
23 See [Mac Lane, 1986], p. 406: ‘in one sense a foundation is a security blanket: If you
meticulously follow the rules laid down, no paradoxes or contradictions will arise’. He goes on
to point out that a risk of zero can’t be achieved in this way, but the quotation in the text from
[Mac Lane, 1971], p. 236, explicitly recognizes the worth of Risk Assessment: ‘alternative
formal system[s] could . . . be tested by comparison (as to strength or relative consistency) with
this one system.’
24 In fact, [Burgess, 2015], p. 176, points out that the inaccessibles aren’t strictly necessary,
that they can be removed by careful use of reflection. But, he observes, ‘in Grothendieck’s kind
of work the intellectual faculties are being strained to their uttermost limit, and one doesn’t want
the distraction of any sort of bookkeeping requirements’.
25 [Feferman, 1977], p. 155, characterizes the deficit of the Grothendieck-style reduction as
‘aesthetic’, but it’s hard to see why a foundational use has to be beautiful in some way or other.
unrestricted notions . . . are mathematically reasonable on the

face of it and do not obviously lead to paradoxical conclusions.26
(Ibid.)
The challenge, then, is to combine them with all the usual objects and operations
that category theorists call upon in the course of their mathematical work, and to
prove the resulting theory consistent relative to ‘a currently accepted system of set
theory’ (ibid.).
Despite incremental partial successes from Feferman, McLarty and others over
the years, this problem remained open until just recently. In his [Ernst, 2015],
Ernst shows that any theory that allows the formation of the category of all graphs
and that includes the required mathematical staples is in fact inconsistent.27 So
regardless of whether Mac Lane’s demand that set theory provide for unlimited
categories is based on Metaphysical Insight or on another motive, it’s a demand
that can’t be met – not because of any shortcoming on set theory’s part, but
because the objects he demands are themselves inconsistent, beyond the reach of
any consistent foundation.
Nevertheless, as a historical matter, many calls to replace set-theoretic founda-
tions with category-theoretic foundations arose from the ill-fated hope that category
theory, unlike set theory, could meet this impossible demand:
One might hope for some . . . new foundational system (the cat-
egory of categories?) within which all the desired objects live.
([Mac Lane, 1971], p. 236)
Though Lawvere (in [Lawvere, 1966]) takes exactly this thought as his title –
‘The category of categories as a foundation for mathematics’ – his primary focus is
actually on a foundational use distinct from those considered so far. But before we
turn to this new sense of ‘foundation’, we need to ask how this theory of categories,
or really an improved descendant due to [McLarty, 1991],28 fares on the various,
more familiar foundational uses that set theory has been seen to underwrite.
The Category of Categories as a Foundation (CCAF) begins from a simple
axiomatic theory that codifies the usual category-theoretic machinery to yield
the usual results of general category theory . . . as hypotheticals
of the form ‘If A is a cartesian closed category, then . . . ’ or ‘If
A is a non-trivial topos, then . . . ’. . . . On the other hand, the
26 It might appear that a paradox is ready to hand: the category of categories is a category;
the category of groups isn’t a group; what about the category of all non-self-membered categories?
The trouble with this tempting line of thought is that ‘membership’ isn’t native to the world of
naı̈ve category theory. We could define what it is for the category of categories to be ‘self-
membered’: it’s for the category of categories to be a category. Likewise the category of groups
is ‘non-self-membered’ because it’s not a group. But categories in general aren’t given in the
form ‘the category of all X’s’, so a general membership relation can’t be defined in this way.
As [Feferman, 2013], p. 9) remarks, ‘There is no sensible way . . . to form a category of all
categories which do not belong to themselves’.
27 The proof is non-trivial, structured roughly along the lines of a proof that there is no set of
all sets via Cantor’s theorem: Ernst assumes there is a category of all graphs R (actually reflexive
graphs, but the result generalizes to all graphs), concocts a certain exponential, shows there can’t
be a map of R onto this exponential, then uses the fact that the exponential is a substructure of
R to show that there is a map of R onto the exponential.
28 The system of [McLarty, 1991] avoids the shortcomings identified in the review
[Isbell, 1967] of [Lawvere, 1966].

302 PENELOPE MADDY
axioms prove the existence of few specific cases of these general

results. ([McLarty, 1991], pp. 1258–1259)
This minimal background theory is then supplemented ‘with axioms on particular
categories or functors’ (ibid.), depending on the intended use. For foundational
purposes, one adds the axioms of the Elementary Theory of the Category of Sets
(ETCS) – also introduced by Lawvere (in his [Lawvere, 1964]) – which characterize
a weak set theory (equivalent to ZC with bounded separation). [Mathias, 2001]
describes various shortfalls of ETCS, mostly the failure of iterative constructions,
but these can be achieved by adding a category-theoretic version of Replacement
to generate a system equivalent to ZFC, and large cardinal axioms to duplicate the
full force of ZFC + LCs.29
If the foundational uses are to be recovered in this way, opponents might argue
that ZFC + LCs is doing the true foundational work, and this Augmented-ETCS is
just piggy-backing,30 but both Mac Lane and McLarty argue that ETCS is itself an
independent, thoroughly category-theoretic system:
The standard ‘foundation’ for Mathematics start[s] with sets and
their elements. It is possible to start differently, by axiomatiz-
ing not elements of sets but functions between sets.[31] This can
be done using the language of categories and universal construc-
tions. ([Mac Lane, 1986], p. 398)
ETCS is a set theory. It is not a membership-based set the-
ory like ZF. It is a function-based set theory. Mac Lane generally
uses the phrase ‘set theory’ to mean ZF, a habit of more than
thirty years before ETCS was conceived. But we cannot let his
terminology misdirect us. He is explicit that ETCS is his pre-
ferred account of sets. ([McLarty, 2004], p. 39)
The thought is that ZF-style set theory doesn’t enjoy exclusive rights to the pre-
theoretic notion of ‘collection’. We’re to imagine ourselves starting from scratch,
thinking purely category-theoretically, and devising a theory of collections in top-
down function-based terms, rather than bottom-up element-based terms.
I should note that this particular objection to category-theoretic foundations
is part of a broader concern about category theory’s ‘autonomy’, beginning in
[Feferman, 1977].32 [Linnebo and Pettigrew, 2011] summarize the complaints
under three headings: logical autonomy – can the theory be stated without appeal
to set theory?, conceptual autonomy – can the theory be understood without a
prior understanding of set theory?, and justificatory autonomy – can the theory be
justified without appealing to set theory and/or its justifications? Though I make
no claim to have grasped all the ins and outs of this debate, it seems clear that both
ZFC + LCs and Augmented-CCAF (= CCAF + Augmented-ETCS) are straightfor-
ward first-order theories, each inspired by a range of intuitive, proto-mathematical
29 See [McLarty, 2004] for these extensions of ETCS. As [Ernst, to appear] points out,
this augmentation is often resisted by advocates of category-theoretic foundations, because the

mathematics it’s designed to accommodate is regarded as expendable. This would be a new use
of a foundation – as a way of pruning mathematics! – directly opposed to the thought behind
maximize (see footnote 14).
30 Indeed they have so argued. See, e.g., [Hellman, 2003].
31 See [Von Neumann, 1925].
32 The literature on this topic is copious and tangled. See [Ernst, to appear] for a recent
overview.
notions: for set theory, these include collection, membership, iteration and a com-
binatorial idea of ‘all possible subsets’ as described, for example, by Bernays33 ; for
category theory, perhaps collection, function, composition of functions, . . . .34 As
long as ‘collection’ isn’t awarded exclusively to set theory by some kind of natural
right, the two appear to be equally autonomous.
As for justificatory autonomy, Linnebo and Pettigrew propose that the iter-
ative conception justifies ZFC, and argue that the category theorist is unable to
come up with anything sufficiently comparable. My own view is that the iterative
conception is a brilliant heuristic device, but that the justification for the axioms it
suggests (and even for potential axioms it doesn’t suggest!) rests on their power to
further various mathematical goals of set theory, including its foundational goals.35
This mode of justification (which Linnebo and Pettigrew don’t consider) is open
to category theory as well, and granting the dramatic success of category-theoretic
methods in various areas of mathematics, there can be little doubt that its concepts
and techniques are well-justified. So, for what it’s worth, I see no threat to the au-
tonomy of category theory in these justificatory pathways. What we’ve been asking
here, in this exploration of category theory’s foundational aspirations, is whether
the axioms of Augmented-CCAF are (partly) justified by their effectiveness toward
the foundational goals, but our question is whether there is such a justification at
all, not whether it’s parasitic on set theory.
Returning to our main theme, then: how are the foundational uses of set theory
recovered in this category-theoretic context, in Augmented-CCAF? As an example,
McLarty takes up the construction of reals as Dedekind cuts, concluding that
once you get beyond axiomatic basics, to the level of set theory
that mathematicians normally use, ZF and ETCS [ETCS plus
Replacement?] are not merely intertranslatable. They work just
alike. ([McLarty, 2004], p. 41)
So it seems Elucidation works much as before: a surrogate is found in ETCS rather
than ZF, and the clarificatory benefits are pretty much the same. Presumably Risk
Assessment makes use of the large cardinals of Augmented-ETCS in the familiar
ways.36
I’m less sure how to think about Meta-mathematical Corral, Generous
Arena, and Shared Standard. On the category theorist’s foundational scheme,
what do we say to the mathematician who wants to know whether or not there’s
a definable well-ordering of the reals? What theory do we turn to if we want to
formulate questions of what can or can’t be proved in ‘classical mathematics’, or
to determine conclusively whether or not a purported informal proof is legitimate?
All the reduced items appear side-by-side, and theorems about them are provable,
in the category satisfying Augmented-ETCS, which suggests Augmented-ETCS as a
likely candidate for Generous Arena and the rest, but the fact that CCAF posits
33 [Bernays, 1934]. See also [Maddy, 1997], pp. 127–128.

34 It might be of interest to investigate more carefully which intuitive notions each theory
appeals to and which it explicates, and where these lists differ, to examine how successfully each
explicates the other’s primitives. For example: ZFC-style set theory explicates ‘function’ with the
much-maligned set of ordered pairs; Mathias derides the category-theoretic treatment of ‘iteration’
as ‘clumsy’ ([Mathias, 2001], p. 227).
35 See [Maddy, 1997], [Maddy, 2011].
36 One reason, perhaps, to resist the pruning in footnote 29.
304 PENELOPE MADDY
items outside that category casts doubt on this move. Perhaps Augmented-CCAF
itself, then?
Surprisingly, this line of thought is undercut by passages where Mac Lane ap-
pears to deny the very desirability of Generous Arena in the first place, preferring
a Hilbert-like approach to a von-Staudt-style construction in the case of the reals:
This careful construction of the real numbers was long accepted
as standard in graduate education in mathematics, even though
many mathematicians did not much believe in it. . . . This view-
point can be expressed . . . formally: Do not construct the reals,
but describe them axiomatically as an ordered field, complete
in the sense that every bounded set has a least upper bound.
([Mac Lane, 1981], p. 467)
Let’s leave this unsettling thought aside for the moment and look first at the new
foundational use that Lawvere and others see as deciding the case in favor of cate-
gory theory over set theory.
It first appears as an objection to set-theoretic foundations:
This Grand Set Theoretic Foundation . . . does not adequately
describe which are the relevant mathematical structures to be
built up from the starting point of set theory. A priori from
set theory there could be very many such structures, but in fact
there are a few which are dominant . . . natural numbers, rational
numbers, real numbers . . . group, ring, order and partial order
. . . . The ‘Grand Foundation’ does not provide any way in which
to explain the choice of these concepts. ([Mac Lane, 1981],
p. 468)
While set theory has the wherewithal to build all the mathematically important
structures, its construction techniques are indiscriminate, generating a vast store
of mathematically useless structures along the way and providing no guidance as
to which are which. There’s also some discomfort about the way in which those
structures are built up:
In the mathematical development of recent decades ones sees
clearly the rise of the conviction that the relevant properties of
mathematical objects are those which can be stated in terms
of their abstract structure rather than in terms of the elements
which the objects were thought to be made of. ([Lawvere, 1966],
p. 1)
Set-theoretic constructions introduce a lot of irrelevant structure: for example, a
Dedekind cut is a set of rationals, which are equivalence classes of sets of pairs
of natural numbers, which are ordinals, and so on, but none of this detail has
any direct connection to their intended behavior as surrogates for the reals, as the
availability of alternatives like Cauchy sequences serves to demonstrate.
Complaints like these about set-theoretic foundations led to the suggestion that
category theory might be better suited to the task. Lawvere hopes for a foundation
that will bring ‘abstract structure’ to the forefront:
The question thus naturally arises whether one can give a foun-
dation for mathematics which expresses wholeheartedly this con-
viction concerning what mathematics is about, and in particular,
in which classes and membership in classes do not play any role.

([Lawvere, 1966], p. 1)
Mac Lane stresses the replacement of static elements with dynamic functions:
There are other possibilities. For example, the membership rela-
tion for sets can often be replaced by the composition operation
for functions. This leads to an alternative foundation for Math-
ematics upon categories . . . much of Mathematics is dynamic,
in that it deals with morphisms of an object L into another ob-
ject of the same kind. Such morphisms . . . form categories, and
so the approach via categories fits well with the objective of or-
ganizing and understanding Mathematics. ([Mac Lane, 1986],
p. 359)
McLarty emphasizes that this particular foundational use should reflect what math-
ematicians actually do in their mathematical lives:
Mac Lane had a different idea of foundations . . . He took a
foundation of mathematics to be a body of truths which organize
mathematics as do I here. More specifically, that is truths which
actually serve in practice to define the concepts of mathematics
and prove the theorems. I do not mean merely truths which
could in some principled sense possibly organize the practice
but truths actually used in textbooks and journal articles, and
discussed in seminar rooms and over beer, so their notions do
occur in practice. ([McLarty, 2013], p. 80)
The goal here is a foundation for mathematics that will capture the fundamental
character of mathematics as it’s actually done, that will guide mathematicians
toward the truly important concepts and structures, without getting bogged down
in irrelevant details. I propose Essential Guidance as an awkward label for this
foundational goal, hoping to highlight its two aspects: such a foundation is to
reveal the fundamental features – the essence, in practice – of the mathematics
being founded, without irrelevant distractions; and it’s to guide the progress of
mathematics along the lines of those fundamental features and away from false
alleyways.37
Of course, Lawvere, Mac Lane and McLarty are entirely correct when they
point out that set theory does not provide Essential Guidance.38 For the record,
though, we should note that this fact in no way compromises its other foundational
uses: the lack of guidance and the presence of extraneous details don’t under-
cut Meta-mathematical Corral, Elucidation, Risk Assessment, Generous
Arena, or Shared Standard.39 They do, however, keep set theory from providing
Metaphysical Insight; that’s the moral of Benacerraf and the rest. As floated
in passing earlier, could it be that something like this is part of what’s at issue for
37 It may be that this is the use Voevodsky has in mind for ‘univalent foundations’: ‘The
languages which arise in this approach are much more convenient for doing serious mathematics
than ZFC’ ([Voevodsky, 2014], p. 108).
38 Presumably, neither does ETCS. I take it ETCS and its extensions are to be called upon
for some foundational jobs, like Elucidation and Risk Assessment, while something broader,
like Augmented-CCAF, is to provide Essential Guidance.
39 Cf. [Mathias, 1992], p. 115: ‘to reject a claim that set theory supplies a universal mode
of mathematical thought . . . need not compel one to declare set theory is entirely valueless’.
306 PENELOPE MADDY
some of these category-theoretic thinkers, some hope of uncovering the true nature
of the mathematical structures? However that may be, it does seem likely that
a version of Epistemic Source is implicit in the kind of guidance they have in
mind: our ‘understanding’ of the mathematics is presumably based in our grasp of
its fundamental concepts and techniques. In any case, I don’t see that anything like
Essential Guidance was among the ambitions of set-theoretic foundations in the
first place, so to count them as ‘failures’ of set theory is to fault a cat for not being
a dog. But if a category-theoretic foundation does deliver on this desideratum, it
would enjoy a dramatic advantage over set theory.
Alas, it isn’t clear that category theory does deliver on Essential Guidance.
I think it’s agreed on all sides that a category-theoretic conceptual framework is
a remarkably effective way of thinking in fields like algebraic topology and alge-
braic geometry – no one would suggest that specialists in these areas would do
better to think more like set theorists. The contested claim – if category-theoretic
foundations are to capture what’s most fundamental, to guide us to the mathemat-
ically significant concepts – is that all mathematicians would do better to think like
category theorists. But even Mac Lane, in his sober moments, doesn’t believe this:
Categories and functors are everywhere in topology and in parts
of algebra, but they do not yet relate very well to most of anal-
ysis. ([Mac Lane, 1986], p. 407)
Analysis is where set theory first arose, much as category theory arose in algebra,
so it’s no surprise that set-theoretic thinking is more suitable there.
Mathias, who has done much to bring out the difficulties of category theory in
analysis,40 gives this reading of Mac Lane’s resistance to set theory:
I would guess that his reason is not so much that he objects to the
ontology of set theory but that he finds the set-theoretic cast of
mind oppressive and feels that other models of thought are more
appropriate to the mathematics he wishes to do. ([Mathias, 1992],
p. 115)
But the analyst or set theorist might well feel the same way about the category-
theoretic cast of mind:
The CAT camp [who believe category theory has ‘the one true
view of pure mathematics’] may with justice claim that category
theory brings out subtleties in geometry to which set theory is
blind. . . . The SET camp [who believe set theory has ‘the one
true view of pure mathematics’] may with equal justice claim
that set-theoretic analysis brings out subtleties to which the CAT
camp is blind. ([Mathias, 2001], pp. 226–227)
It seems that both these camps are chasing a false goal: a foundation that delivers
Essential Guidance, a single understanding of what mathematics is, a single
recommendation on how mathematicians should think.41
40 See [Mathias, 2000], [Mathias, 2001]. Beyond analysis, [Ernst, 2014] explores some
potentially problematic examples from graph theory.

41 Mathias himself advocates ‘unity’, akin to Generous Arena (see below), but not ‘uni-
formity’, or Essential Guidance: ‘Is it desirable to press mathematicians all to think in the
same way? I say not: if you take someone who wishes to become a set theorist and force him to
do (say) algebraic topology, what you get is not a topologist but a neurotic’ ([Mathias, 1992],
p. 113).
Now the odd thing is that Mac Lane would apparently agree:
We conclude that there is as yet no simple and adequate way of
conceptually organizing all of Mathematics. ([Mac Lane, 1986],
p. 407)
In other words, no foundation, not even a category-theoretic foundation, has a cor-
ner on Essential Guidance. Yet we’ve seen that he champions category-theoretic
foundations over set-theoretic foundations because its approach ‘fits well with the
objective of organizing and understanding Mathematics’ ([Mac Lane, 1986],
p. 359). Perhaps this apparent conflict actually dovetails with the suggestion
above that Mac Lane might reject some of the foundational desiderata that set
theory successfully satisfies, for example, Generous Arena. If we deny the impor-
tance of bringing (surrogates for) all mathematical structures into a shared context
where they can be compared side-by-side, their interrelations revealed, methods
and results imported and exported, and so on, then the Hilbert-style approach of
leaving the axiom system for each mathematical structure to stand separately, on
its own, might seem preferable to the von-Staudt-style construction. The need for
a single foundational scheme would disappear, and the possibility of a range of
schemes might seem attractive: ‘the variety of proposals for organizations reflects
the diversity and richness of Mathematics’ ([Mac Lane, 1986], p. 407).
If this is right, then Mac Lane’s ultimate point is more radical than the simple
claim that category theory is a better foundation for mathematics than set theory.
His point, rather, is that the most important function of a foundation is Essential
Guidance, and if this conflicts with some of the traditional foundational uses
set theory has been put to, like Generous Arena, then so much the worse for
those traditional uses. I’ve tried to sketch how set theory came to be seen as the
Generous Arena, why this was thought to be important, what work it did, and
so on. [Burgess, 2015] gives a more extended review of the history; Mathias gives
a mathematician’s eye view:
One of the remarkable things about mathematics is that I can
formulate a problem, be unable to solve it, pass it to you; you
solve it; and then I can make use of your solution. There is
a unity here: we benefit from each other’s efforts. . . . But if I
pause to ask why you have succeeded where I have failed . . . I
find myself faced with the baffling fact that you have thought of
the problem in a very different way from me: and if I look around
the whole spectrum of mathematical activity the huge variety of
styles of thought becomes even more evident. . . . The purpose
of foundational work in mathematics is to promote the unity of
mathematics; the larger hope is to establish an ontology within
which all can work in their different ways. ([Mathias, 1992],
pp. 113–114)
I leave it to the reader’s conscience to decide whether Generous Arena, and
possibly other of the traditional foundational uses should be jettisoned.
Perhaps the best course would be to stop quibbling about the word ‘founda-
tion’, leave set theory to the important functions it so ably performs, and turn
serious philosophical and methodological attention to the matter of distinguishing
and exploring the distinctive ‘ways of thinking’ that flourish in different areas of
pure mathematics. Directly after his claim that category theory ‘fits well with
308 PENELOPE MADDY
the objective of organizing and understanding Mathematics’, Mac Lane contin-

ues, ‘That, in truth, should be the goal of a proper philosophy of Mathematics’
([Mac Lane, 1986], p. 359, emphasis added). Perhaps the unremarked shift from
‘foundation of Mathematics’ to ‘philosophy of Mathematics’ is telling; perhaps what
Mac Lane is really after isn’t so much a replacement for set theory as a broader
appreciation for, and more philosophical/methodological attention to, the organi-
zational and expressive powers of category theory. I think we can all agree that
this would be a good thing!
3. The multiverse
The most recent challenge to straightforward set-theoretic foundations in V
comes, oddly enough, from among set theorists themselves. To quote yet another
textbook:
Should we suppose that the continuum hypothesis, for example,
has a definite truth value in a well-defined canonical model? Or
is there a range of models in which the truth value of the con-
tinuum hypothesis varies, none of which has any special onto-
logical priority? Forcing tends to push us in the latter direction.
([Weaver, 2014], p. 118)
Hamkins, whose position Weaver endorses, describes the situation this way:
Our most powerful set-theoretic tools, such as forcing, ultrapow-
ers, and canonical inner models, are most naturally and directly
understood as methods of constructing alternative set-theoretic
universes. A large part of set theory over the past half-century
has been about constructing as many different models of set the-
ory as possible . . . As a result, the fundamental objects of study
in set theory have become the models of set theory, and set theo-
rists move with agility from one model to another. . . . Set theory
appears to have discovered an entire cosmos of set-theoretic uni-
verses. ([Hamkins, 2012], p. 418)
This has come to be called a ‘multiverse’ view. Woodin observes of one such position
that
the refinements of Cohen’s method of forcing in the decades since
his initial discovery of the method and the resulting plethora
of problems shown to be unsolvable, have in a practical sense
almost compelled one to adopt the generic-multiverse position.
([Woodin, 2011], pp. 16–17)
If this is right, our fundamental theory shouldn’t be a theory of sets, but a theory
of set-theoretic universes.
There are actually several different multiverse proposals currently on offer.42
The most generous is Hamkins version:
The background idea of the multiverse . . . is that there should
be a large collection of universes, each a model of (some kind
of) set theory. There seems to be no reason to restrict inclusion
42 [Väänänen, 2014] provides a framework that encompasses the positions of Hamkins, Steel
and Woodin. S. Friedman sometimes uses multiverse language, but it isn’t clear (at least to me)
that his is a true multiverse view. See, e.g., [Arrigoni and Friedman, 2013].
only to ZFC models, as we can include models of weaker theories

ZF, ZF− , KP, and so on, and perhaps even down to second-order
number theory, as this is set-theoretic in a sense. . . . We want
to consider that the multiverse is as big as we can imagine. At
any time, we are living inside one of the universes, referred to as
V and thought of as the current universe, but by various means,
sometimes metamathematical, we may be able to move around
in the multiverse. ([Hamkins, 2012], pp. 436–437)
Hamkins is out to explain mathematical experience:
This abundance of set-theoretic possibilities poses a serious dif-
ficulty for the universe view . . . one must explain or explain
away as imaginary all of the alternative universes that set the-
orists seem to have constructed. This seems a difficult task, for
we have a robust experience in those worlds . . . The multiverse
view . . . explains this experience by embracing them as real.
([Hamkins, 2012], p. 418)
Steel and Woodin, in contrast, are concerned with the claim, suggested by Weaver
above, that CH has no determinate truth value.43
Steel observes that
we have good evidence that the consistency hierarchy is not a
mirage, that the theories in it we have identified are indeed con-
sistent. This is especially true at the lower levels, where we
already have canonical inner models, and equiconsistencies with
fragments of definable determinacy. This argues for developing
the theories in this hierarchy. ([Steel, 2014], p. 164)
He then defends the resulting ZFC + LCs on the grounds that it allows the set-
theoretic reduction:
The central role of the theories axiomatized by large cardinal
hypotheses argues for adding such hypotheses to our framework.
The goal of our framework theory is to maximize interpretive
power, to provide a language and theory in which all mathemat-
ics, of today and of the future so far as we can anticipate it today,
can be developed. (Ibid., p. 165)
The trouble, of course, is that this strong and attractive theory – ZFC + LCs – can’t
settle statements like CH, as forcing so dramatically demonstrates. Does it follow
that CH has no determinate truth value?44
43 Cf. [Woodin, 2011], p. 16: ‘The generic-multiverse position . . . declares that the Con-
tinuum Hypothesis is neither true nor false’. [Steel, 2014], p. 168, takes his approach to open
the possibility that CH and the like are ‘meaningless . . . pseudo-questions’.
44 Notice the sharp contrast between Hamkins’s goal – embrace the widest possible range of
universes – and Woodin’s and Steel’s shared goal – given what we take ourselves to know about
sets (roughly ZFC + LCs), figure out whether there is a fact of the matter about CH. Though both
can reasonably be called ‘multiverse’ views, they are quite different undertakings, and can’t be
evaluated by the same standards. E.g., though Hamkins’s all-inclusive project naturally prompts
him to include universes in which V = L, it would be inappropriate to criticize Woodin and Steel
for ruling them out.
310 PENELOPE MADDY
Woodin addresses this question by formulating what he calls the generic mul-
tiverse, a collection of models of ZFC + LCs:45
It is generated from each universe of the collection by closing
under generic extensions (enlargements) and under generic re-
finements (inner models of a universe which the given universe
is a generic extension of). ([Woodin, 2011], p. 14)
In a slogan: if N is a forcing extension of M , and one of them is in the generic
multiverse, then so is the other. Obviously, he and Hamkins differ on the range of
the universes in their multiverses – he and Steel both begin from our current best
theory of sets46 – but they also disagree sharply on the relevant notion of truth:
Woodin takes a claim to be true in the multiverse iff it holds in every universe of
the multiverse; Hamkins rejects this notion47 and holds that his many universes
‘exhibit diverse set-theoretic truths’ ([Hamkins, 2012], p. 416).
Woodin goes on to argue (assuming the Ω conjecture) that generic multiverse
truth amounts to ‘a brand of formalism that denies the transfinite’, and hence that
‘the generic-multiverse position . . . is not reasonable’ ([Woodin, 2011], p. 17).
Steel reports that
Woodin’s paper makes some arguments against the generic mul-
tiverse position, based on the logical complexity of certain truth
predicates, but those arguments do not seem valid to me. ([Steel, 2014],
p. 170; for details, see his footnote 24)
I can only leave this call to the experts.
Steel’s own multiverse approach is more syntactic than either Hamkins’s or
Woodin’s, but before we get to that, let’s pause to consider what becomes of set
theory’s foundational uses on a model-theoretic multiverse conception like these.
As noted in Section 1, I did once believe there was a strong prima facia case against
a multiverse of any kind, based on the assumptions that one of set theory’s leading
goals is to provide a foundation and that one function of a foundation is to serve
as a ‘final court of appeal’. The idea is simple: if set theory is to settle questions
of proof and existence, then set theorists should strive for a single preferred theory
of sets that’s as decisive as possible – so as to produce unequivocal judgments
when called upon. Of course this very same set-theoretic goal also counsels that
set theory be as generous as possible, so as not to limit mathematics,48 but at
least in the most conspicuous cases of potential bifurcation – like ZFC + LCs vs.
ZFC + V = L – it’s possible to avoid an exclusive choice – because ZFC + V = L
can be viewed as the theory of L even in a universe with large cardinals. But this
innocent observation actually paves the way for the less doctrinaire descendants of
the ‘final court’ identified in Section 1, where Shared Standard and Generous
Arena make room for stronger theories, as long as added hypotheses are flagged
so that import/export restrictions can be observed – and this begins to sound a bit
45 Actually both Steel and Woodin begin with models of ZFC, with the addition of large
cardinals as an option, but context (e.g., the above motivation from Steel) suggests they lean
toward taking that option. This is irrelevant for the case of CH, which is their primary focus.
46 See footnote 44.
47 Cf. [Hamkins, 2012], p. 445: ‘Woodin introduced [the generic multiverse] in order to
criticize a certain multiverse view of truth, namely, truth as true in every model of the generic
multiverse. . . . I do not hold such a view of truth’.
48 See footnote 14.
like a multiverse view. So what I once took for an objection becomes a question: is
the multiverse intended to continue to play the foundational roles that the universe
once did, and if so, how does it do this?49
Hamkins addresses these questions directly:
Set theorists commonly take their subject as constituting an on-
tological foundation for the rest of mathematics, in the sense
that abstract mathematical objects can be construed . . . as sets.
([Hamkins, 2012], p. 416)
The multiverse view does not abandon the goal of using set the-
ory as an epistemological and ontological foundation for mathe-
matics, for we expect to find all our familiar mathematical ob-
jects . . . inside any one of the universes of the multiverse. (Ibid.,
p. 419)
The idea is that when we do set theory, ‘we jump inside and explore the nature of
set theory offered by that universe’ (ibid., p. 417), and whenever we do this, all the
usual resources of classical mathematics will be available.50
So what happens when our mathematician asks about that definable well-ord-
ering of the reals?
When a mathematical issue is revealed to have a set-theoretic
dependence . . . then the multiverse response is a careful explana-
tion that the mathematical fact of the matter depends on which
[universe] is used, and this is almost always a very interesting
situation, in which one may weigh the desirability of various
set-theoretic hypotheses with their mathematical consequences.
([Hamkins, 2012], p. 419)
This sounds very like what I said in the universe voice in Section 1: we explain
to the mathematician that there is no such well-ordering in V, but that there is
one in L; we lay out the mathematically attractive and unattractive features of L;
and we caution that if he elects to take this route, he will have to be careful about
exporting and importing resources to and from V.
Hamkins admits that avenues like this are available to the universe advocate,
for outer models as well as inner:
We do have a measure of access into the forcing extensions via
names and the forcing relation, allowing us to understand the
objects and truths of the forcing extension while remaining in
the ground model. ([Hamkins, 2012], p. 419)
So what makes the multiverse approach preferable?
The multiverse view explains our mathematical experience with
these models by positing that, indeed, these alternative universes
exist, just as they seem to exist, with a full mathematical ex-
istence, fully as real as the universe under the universe view.
(Ibid., p. 419)
49 This is to entertain the possibility that unify, in its original form, might not be required
for set theory’s foundational uses.

50 Here, presumably, the multiverse doesn’t include models of theories too weak to do the
job, a stipulation in apparent tension with the generosity cited a few pages back.
312 PENELOPE MADDY
The multiverse view . . . takes . . . forcing at face-value, as ev-

idence that there actually are V-generic filters and the corre-
sponding universes V[G] to which they give rise, existing outside
the universe. This is a claim we cannot prove within set theory,
of course, but the philosophical position makes sense of our ex-
perience – in a way that the universe view does not – simply by
filling in the gaps, by positing as a philosophical claim the ac-
tual existence of the generic objects . . . With forcing, we seem to
have discovered the existence of other mathematical universes,
outside our own universe, and the multiverse view asserts that
yes, indeed, this is the case. (Ibid., p. 425)
This is high metaphysics! Let’s set it aside for the moment and return to the more
tractable question of how Hamkins’ multiverse carries out its foundational duties.
What we’ve been hearing about so far, over and above the reduction itself, is ap-
parently the ‘final court’ descendants, Shared Standard and Generous Arena;
in practice, these are treated much as the universe advocate would, with ZFC as
the default Shared Standard and V the default Generous Arena, with added
hypotheses noted and import/export carefully regulated. Elucidation seems un-
problematic, as it, too, can be carried out within any universe of the multiverse.51
For Risk Assessment, presumably we call on a universe with large cardinals. For
purposes of Meta-mathematical Corral, given that there are many universes
outside any given universe, perhaps we’re to turn to the theory of the multiverse
itself, much as we might turn to Augmented-CCAF in the case of category-theoretic
foundations.52 But this raises a new question: what is the theory of the multiverse?
What is the counterpart to Augmented-CCAF or ZFC + LCs? What fundamental
theory is to tell us what universes there are, something we need to know before any
of this can get off the ground?
Hamkins cautions that we shouldn’t expect a first-order theory in the language
of set theory,
since the entire point of the multiverse perspective is that there
may be other universes outside a given one. ([Hamkins, 2012],
p. 436)
Still, he does offer a number of multiverse axioms, such as:
For any universe V and any forcing notion P in V, there is
a forcing extension V[G], where G ⊆ P is V-generic. (Ibid.,
p. 437)
Obviously there is appeal here to quite technical set-theoretic notions, so we need
to ask how principles like this one could be formulated without a prior, ordinary
theory of sets in the background. We’ve met with this sort of concern before, in
the ‘logical autonomy’ objection to category-theoretic foundations, but it seems,
if anything, more apt here than it was there: can the theory of the multiverse be
stated without presupposing a theory of the universe? Unless it can, no autonomous
alternative is actually on offer.
51 Again assuming a more restricted range of models than Hamkins sometimes suggests (see
footnote 50).
52 For that matter, we might wonder whether, e.g., the Generous Arena should include the
entire multiverse, just as we wondered in the case of Augmented-ETCS vs. Augmented-CCAF.

Steel feels the force of this concern. As we’ve seen, his initial presentation
(quoted above) puts the foundational goal front and center:
The goal of our framework theory is to maximize interpretive
power, to provide a language and theory in which all mathemat-
ics, of today and of the future so far as we can anticipate it today,
can be developed. ([Steel, 2014], p. 165)
He particularly recognizes the importance of Shared Standard and Generous
Arena, and of straightforward import/export:
Why not simply develop all the natural theories . . . ? Let 1000
flowers bloom! . . . The problem with this . . . is that we do not
want everyone to have his own private mathematics. We want
one framework theory, to be used by all, so that we can use each
other’s work. It is better for all our flowers to bloom in the same
garden. (Ibid., p. 164)
For this to work, Steel recognizes the need for an explicit, free-standing theory of
the multiverse. In a ‘historical note’, he remarks:
Neither Hamkins nor Woodin presented a language and a first-
order theory in that language, both of which seem necessary for
a true foundation. ([Steel, 2014], p. 170)
He provides this by introducing a multiverse language (ML) that speaks of both sets
and universes, and a list of axioms in that language (MV) that includes assertions
like these:
• For any axiom ϕ of ZFC + LCs, and every world W , ϕW .
• If W is a world and P ∈ W is a poset, then there is a world of the form
W [G], where G is P-generic over W .
• If U is a world, and U = W [G], where G is P-generic over W , then W is
a world. (See [Steel, 2014], p. 165)53
With the help of some serious mathematics, the theory MV can be successfully
formalized.54
Steel then constructs, from any transitive model M of ZFC+LCs, a set of worlds
M G that form a natural model of MV.55 At this point, ‘truth in the multiverse
M G ’ is a perfectly straightforward notion: a statement in ML is true or false in M G
in a standard model-theoretic sense. But this isn’t the notion Steel is after. His
leading question, like Woodin’s, is whether CH, a statement in the pure language of
set theory (LST), has a determinate truth value, and he approaches this question
by asking whether CH is or isn’t meaningful. (If it’s not meaningful, it’s not even
a candidate for truth or falsity.) So the multiverse position Steel considers isn’t
the Woodin-like proposal that a set-theoretic claim is true in the multiverse iff it’s
true in every universe of the multiverse, but something he calls the Weak Relativist
Thesis (WRT).
53 Woodin and Steel differ, e.g., over Steel’s axiom of Amalgamation ([Steel, 2014], p. 165):
‘If U and W are worlds, then there are G, H sets generic over them such that W [G] = U [H]’.
54 The key step for the last of these is a theorem of Laver and Woodin proved in an appendix
to [Woodin, 2011].
55 Let G be M -generic for Col(ω, <ORM ). Then M G is the collection of all ‘W such that
W [H] = M [G α], for some H set generic over W , and some α ∈ ORM ’ ([Steel, 2014], p. 166).
314 PENELOPE MADDY
To sneak up on the WRT, first notice that since the statement of concern,
CH, is a statement of LST, and Steel’s (perfectly ordinary) multiverse truths are
statements of ML, a link between the two languages is crucial. Fortunately, a
theorem of Laver and Woodin implies that there’s a recursive translation function
t from ML to LST such that for all ϕ in ML, M G thinks ϕ iff M thinks t(ϕ).56
The multiverse idea behind the WRT is that the only meaningful statements of set
theory are those expressible in ML, so the Thesis says that a statement ψ of LST
is meaningful iff there’s a ϕ in ML such that ψ is t(ϕ) – or more succinctly, iff ψ is
in the range of t.
So, is CH meaningful? As a warm-up, consider another statement in LST:
‘there is a measurable cardinal’. It’s an axiom of MV that this holds in every world
of the multiverse, but that’s not enough to show that it appears in the range of
t. In fact, it doesn’t. But Steel suggests that it’s reasonable to assume that t(for
all W , ‘there is a measurable cardinal’W ) and ‘there is a measurable cardinal’ are
synonymous.57 And since ‘for all W , ‘there is a measurable cardinal’W ’ is true
in M G , the synonymous set-theoretic statement ‘there is a measurable cardinal’ is
both meaningful and true. So far so good, but the same trick won’t work for CH,
because ‘for all W , CHW ’ won’t translate to anything like the LST statement that
CH. If this were the end of the story, CH would be meaningless and WRT would
yield the same conclusion as in Woodin’s multiverse: CH has no determinate truth
value; it would be, as Steel puts it, a ‘pseudo-question’ ([Steel, 2014], p. 154).
But this isn’t the end of the story, because, Steel notes, there may be ‘traces of CH’
elsewhere in ML, that is, there may be statements ϕ of ML other than ‘for all W ,
CHW ’ such that t(ϕ) is synonymous with CH.
How could this happen? Suppose that the multiverse language is expressive
enough to single out one among its worlds with an explicit definition. Woodin has
observed
that if the multiverse has a definable world, then it has a unique
definable world, and this world is included in all the others.
([Steel, 2014], p. 168)
This minimal world, if there is one, is called the core of the multiverse. If the
multiverse has a core (C) – the Weak Absolutist Thesis (WAT) – then it might be
reasonable to regard t(CHC ) as synonymous with CH.58 In that way, CH could turn
out to be meaningful even on the assumption of WRT, and its truth value would
56 In fact, Steel suggests that we only understand ML via the translation t. If so, this raises
a question of ‘psychological autonomy’ for MV that runs parallel to the case of category theory:
can we understand ML without first understanding LST? Steel recognizes the problem and offers
a rebuttal: ‘This is not a very strong objection, as one could think of what we are doing as
isolating the meaningful part of the standard language, the range of t, while trimming away the
meaningless, in order to avoid pseudo-questions. After climbing our ladder, we throw it away, and
from now on, MV can serve as our foundation’ ([Steel, 2014], p. 168). This move deserves some
scrutiny, but that’s a topic for another occasion.
57 This isn’t explicit in the relevant passage on p. 167 of [Steel, 2014], but Steel has con-
firmed it in correspondence (cited with permission). He also notes a residual concern that the
MV statement may involve a refinement of the original meaning, but I leave this aside (except for
footnote 58).
58 The residual concern in footnote 57 would arise here as well, perhaps even more acutely:
t(CHC ) involves a great deal of mathematical machinery unknown to many people who presumably
do understand CH (like Cantor, for instance).
depend on what happens in C. But this is speculative for now; much remains to
be explored.
In sum, then, the multiverse position Steel considers (WRT) would take the
axiom system MV as fundamental, in the sense that it serves to circumscribe our
official theory of sets (the range of t). From this perspective, ZFC + LCs (at least)
would be both meaningful and true, and thus available to play the usual founda-
tional roles in the usual ways. Whether CH has a determinate truth value would
remain, for now, an open question.
So, should we prefer a multiverse foundation to the familiar universe founda-
tion? I think we have to allow that the study of multiverse conceptions is in its
infancy, so firm conclusions aren’t possible at this stage (even if I had the wit to
draw them), but I would like to register some discomfort over the terms in which
the debate is often couched. We witnessed Hamkins’ ontological flight above; he
proposes that
each . . . universe exists independently in the same Platonic sense
that proponents of the universe view regard their universe to
exist. ([Hamkins, 2012], pp. 416–417)
Of course, a ‘proponent of the universe view’ might just hold that we should pursue
one preferred theory of sets, without metaphysical addenda,59 but Hamkins, qua
proponent of the multiverse view, takes a stronger line: ‘The multiverse view is one
of higher-order realism – Platonism about universes’ (ibid.).
In contrast, Steel’s multiverse position involves no such ontology. Instead, he
appeals to meaning: the multiverse language is used (via WRT) as an indicator of
which ordinary set-theoretic claims are meaningful, capable of truth or falsity, and
which are not. As we’ve seen, this raises the possibility that
the truth value of CH is not determined by the meaning we cur-
rently assign to the syntax of LST [the language of set theory].
([Steel, 2014], p. 154)
. . . and the suggestion (quoted above) that we should
trim back the current syntax, so that we can stop asking pseudo-
questions. (Ibid.)
Talk of indeterminacy in ‘the concept’, of diverse ‘concepts of set’, is also common in
multiverse thinking. Judging from these discussions, it appears that the overarching
goal of set-theoretic practice is to get these things right, to determine the true
Platonic ontology, the true contours of the meaning of the word ‘set’, or the true
nature of ‘the concept of set’.
But beneath the rhetoric, it emerges that this way of framing the question can’t
be quite right. For example, Hamkins admits that
we may prefer some of the universes in the multiverse to others,
and there is no obligation to consider them all as somehow equal
. . . we may simply be more interested in parts of the multiverse
consisting of universes satisfying very strong theories, such as
ZFC plus large cardinals. ([Hamkins, 2012], pp. 417, 436)
For that matter, it’s even possible that we might have good mathematical reasons
to seek out just one of the universes, just one extension of ZFC, as our unique
preferred theory. As far as meanings and concepts go, I personally doubt there’s
59 See footnote 13.
316 PENELOPE MADDY
a fact of the matter about what is or isn’t part of the current meaning/concept of
set, but even if there were, and even if it didn’t settle CH, we’d be perfectly free to
decide that there’s good reason to move on to an enhanced meaning/concept that
does. Much like Hamkins, Steel admits this possibility:
Certainly we do not want to employ a syntax which encourages
us to ask pseudo-questions, and the problem then becomes how
to flesh out our current meaning, or trim back the current syntax,
so that we can stop asking pseudo-questions. ([Steel, 2014],
p. 154, emphasis added)
So the metaphysics of abstracta or meanings or concepts are all really beside the
point. The fundamental challenge these multiverse positions raise for the universe
advocate is this: are there good reasons to pursue a single, preferred theory of sets
that’s as decisive as possible, or are there not?
Now Steel laments that with matters of justification or ‘good reasons’, the ‘gen-
eral philosophical questions concerning the nature of . . . evidence . . . rear their
ugly heads’ ([Steel, 2014], p. 154), but I don’t think matters are so dire. I would
argue60 that the relevant reasons are all of a type Steel knows well: straightfor-
wardly mathematical reasons. What mathematical jobs do we want our theory of
sets to do? One answer is that we want it to serve the various foundational roles of
Section 1, but there are many others: Cantor was after a theory of trigonometric
series; Dedekind sought representation-free definitions; contemporary set theorists
hope for a rich theory of sets of reals; and so on. The choice between a universe
approach and a multiverse approach is justified to the extent that it facilitates our
set-theoretic goals. The universe advocate finds good reasons for his view in the
many jobs it does so well, at which point the challenge is turned back to the multi-
verse advocate: given that we could work with inner models and forcing extensions
from within the simple confines of V, as described by our best universe theory, what
mathematical motivation is there to move to a more complex multiverse theory?
Hamkins gestures toward this perspective on the question in his appendix:
The mathematician’s measure of a philosophical position may be
the value of the mathematics to which it leads. ([Hamkins, 2012],
p. 440)
He goes on to describe a pair of research projects inspired by the multiverse per-
spective. I’m in no position to evaluate the mathematics; my question is whether
multiverse thinking is playing more than a heuristic role, whether there’s anything
here that couldn’t be carried out in our single official theory of sets. If not, then
it’s not clear these examples give us good reason to incur the added burden of
devising and adopting an official multiverse theory as our preferred foundational
framework. Presumably this same measure could be applied to Woodin’s generic
multiverse: unless there’s at least some hint that it enjoys mathematical advantages
over the universe approach, we needn’t even concern ourselves about the complexity
of its truth predicate.
For the supporter of Steel’s WRT, the purported advantage of a multiverse
foundation appears to be that it saves us from the misguided pursuit of pseudo-
questions, from doomed efforts to settle matters that are indeterminate.61 If the
60 Have
argued, in [Maddy, 1997] and [Maddy, 2011].
61 A
defender of Woodin’s multiverse might also argue that it saves us from doomed efforts,
though Woodin himself – certainly no defender! – doesn’t say this.
central goal of set theory is to identify the features of ‘the meaning we currently
assign to the syntax’ of set-theoretic language, and if the range of t marks the outer
limit of that meaning, and if no ‘trace of CH’ is hidden in ML, then efforts to settle
CH would be hopeless, inevitably sterile, and the multiverse rejection of CH as a
legitimate question might be a valuable outcome. Of course a universe advocate,
convinced, say, that ZFC + LCs marks the outer limit of what we’ll ever know about
V, could reach the same goal of deterring work on CH by epistemological means,
without recourse to multiverse thinking. But more to the point, there remains the
live possibility that the pursuit of CH isn’t in fact doomed. There might even be
what Hamkins calls a ‘dream solution’: a single new axiom that settles CH.62 Of
an even more ambitious dream, Woodin writes:
I am an optimist . . . There is in my view no reason at all, be-
yond a lack of faith, for believing that there is no extension of
the axioms of ZFC, by one axiom, a posteriori true, which set-
tles all the instances of the Generalized Continuum Hypothesis.
([Woodin, 2011], p. 30)
Indeed, Steel himself sketches a scenario that could lead even a tentative propo-
nent of WRT back to a strengthened universe theory that ‘fleshes out’ the current
meaning to something that settles CH in the positive.63 So I don’t see that there’s
call to throw over our universe foundation in favor of a multiverse quite yet!
4. Inconclusive conclusion
We’ve seen that set theory, largely via the well-known set-theoretic reduction,
serves a number of valuable mathematical ends that ought to qualify as ‘founda-
tional’: in the current form of ZFC+LCs, it provides a simple first-order theory that
interprets all of classical mathematics, so as to allow for meta-mathematical con-
sideration of the whole expanse of that vast subject at once (Meta-mathematical
Corral); it provides the conceptual resources and construction techniques to clar-
ify old mathematical notions in order to take on new demands (Elucidation); in
the hierarchy of large cardinals, it provides a flexible scale of consistency strength
(Risk Assessment); it serves as a benchmark of mathematical proof (Shared
Standard) and a framework in which the various branches of mathematics ap-
pear side-by-side, so that results, methods and resources can be pooled (Generous
Arena). On the other hand, it doesn’t tell us anything about the underlying nature
of mathematical objects (Metaphysical Insight) or of mathematical knowledge
(Epistemic Source).
Criticisms of this set-theoretic foundation from category theorists may falsely
assume that it aspires to Metaphysical Insight, but however that may be, they
explicitly demand that it found a theory that can’t in fact be founded (unlimited
categories). Advocates of a category-theoretic replacement for set-theoretic foun-
dations appeal to the category of sets for Elucidation and presumably for Risk
62 [Hamkins, 2012], pp. 429–230, argues that a dream solution to CH is impossible, but he
requires that the axiom in question be ‘obviously true’ in place of Woodin’s ‘a posteriori true’ in
the following quotation. (I take ‘a posteriori’ here to mean that the justification is extrinsic rather
than intrinsic.)
63 See [Steel, 2014], section 7. In this way, WRT – the assumption that our current meaning
is limited to ML – could ultimately point the way to a mathematically attractive extension of the
current meaning.
318 PENELOPE MADDY
Assessment, but the intended execution of the other traditional foundational roles
is less clear, and the attractions of Generous Arena may even be rejected. Their
central hope is for a foundation that provides a range of concepts and methods that
capture and guide the productive ways of thinking that mathematicians actually do
and should employ (Essential Guidance) – a role set theory was never designed
to play. Proponents are surprisingly unconcerned that category theory doesn’t ap-
pear equipped to play this role for all areas of mathematics, another indication
that Generous Arena may have fallen away; different foundations might serve
for different branches of the subject. My suggestion is that we do best to retain
set theory in the foundational roles it plays so well, retain Generous Arena in
particular, but also pursue a serious philosophical/methodological investigation of
the various ‘ways of thinking’ in mathematics, beginning with the contrast between
those whose ‘essence’ is well-captured by category theory or by set theory.
Another challenge to ZFC-style set theory in its familiar foundational role comes
from the advocates of a multiverse conception of the subject. Steel in particular
offers an explicit, free-standing multiverse theory, MV, as fundamental – in place of
a theory in the language of set theory. With the multiverse assumption WRT, MV
ratifies a sublanguage of LST that includes ZFC + LCs (more or less by fiat), though
perhaps not the meaningfulness of CH. On this picture, the theory ZFC + LCs
continues in its usual foundational uses, but only at the behest of MV, and the
prospects for extending it are limited. Faced with the charge that this approach
shuts down free inquiry into extensions of ZFC+LCs prematurely, an advocate might
respond that the multiverse investigation has revealed this inquiry to be misguided,
because the meaning or concept of ‘set’ is simply indeterminate beyond certain
limits. Still, even if it’s true that our current meaning or concept is indeterminate
in this way, there remains the possibility that it might be more mathematically
productive, not to give up the quest for an answer to CH, but to seek out a fruitful
successor to our current meaning or concept – a possibility that Steel himself clearly
acknowledges and pursues. So far at least, the grounds for replacing the universe
with a multiverse are inconclusive.
In sum, then, it seems to me that the familiar set-theoretic foundations, rough
and ready as they are, remain the best tool we have for the various important
foundational jobs we want done.
4.1. Acknowledgements. I’m grateful to Michael Ernst, both for his
[Ernst, 2014], [Ernst, 2015], [Ernst, to appear], and for many enjoyable con-
versations on the aspirations of category-theoretic foundations, and to John Steel,
for his patience with my questions and confusions about his views. My admiration
for John Burgess’s [Burgess, 2015] should be obvious. Thanks also to Ernst, Steel,
David Malament, Colin McLarty, and an anonymous referee for helpful comments
on earlier drafts.
References
[Arrigoni and Friedman, 2013] Tatiana Arrigoni and Sy-David Friedman, The hyperuniverse
program, Bull. Symbolic Logic 19 (2013), no. 1, 77–96, DOI
10.2178/bsl.1901030. MR3087402
[Benacerraf, 1965] Paul Benacerraf, What numbers could not be, Philos. Rev. 74
(1965), 47–73. reprinted in Benacerraf, P., and Putnam, H. (eds),
Philosophy of Mathematics, second edition, Cambridge University
Press, Cambridge, 1983, pp. 272-294. MR0171702
[Bernays, 1934] On platonism in mathematics. reprinted in Benacerraf, P., and

Putnam, H. (eds), Philosophy of Mathematics, second edition,
Cambridge University Press, Cambridge, 1983, pp. 258–271.
MR742474
[Burge, 1998] Tyler Burge, Frege on knowing the foundation, Mind 107 (1998),
no. 426, 305–347, DOI 10.1093/mind/107.426.305. reprinted in
his Truth, Thought, Reason: Essays on Frege, Oxford University
Press, Oxford 2005, pp. 317–355. MR1630330
[Burgess, 2015] John P. Burgess, Rigor and structure, Oxford University Press,
Oxford, 2015. MR3329581
[Dedekind, 1872] Richard Dedekind, Continuity and irrational numbers. W. Beman
and W. Ewald, trans., reprinted in [Ewald, 1996], pp. 766–779.
[Ebbinghaus, 2007] Heinz-Dieter Ebbinghaus, Ernst Zermelo: An approach to his life

and work, Springer, Berlin, 2007. In cooperation with Volker Peck-
haus. MR2301067
[Enderton, 1977] Herbert B. Enderton, Elements of set theory, Academic Press
[Harcourt Brace Jovanovich, Publishers], New York-London, 1977.
MR0439636
[Ernst, 2014] Michael Ernst, The Category-Theoretical Imperative, ProQuest
LLC, Ann Arbor, MI, 2014. Thesis (Ph.D.)–University of Cali-
fornia, Irvine. MR3279213
[Ernst, 2015] Michael Ernst, The prospects of unlimited category theory: doing
what remains to be done, Rev. Symb. Log. 8 (2015), no. 2, 306–327,
DOI 10.1017/S1755020314000495. MR3348001
[Ernst, to appear] Michael Ernst, Category theory and foundations. To appear in
Landry, E., editor, Categories for the Working Philosopher.
[Ewald, 1996] William Ewald (ed.), From Kant to Hilbert: a source book in the
foundations of mathematics. Vol. I, II, Oxford Science Publica-
tions, The Clarendon Press, Oxford University Press, New York,
1996. Compiled, edited and with introductions by William Ewald.
MR1465678
[Feferman, 1977] Solomon Feferman, Categorical foundations and foundations of
category theory, Logic, foundations of mathematics and com-
putability theory (Proc. Fifth Internat. Congr. Logic, Methodology
and Philos. of Sci., Univ. Western Ontario, London, Ont., 1975),
Reidel, Dordrecht, 1977, pp. 149–169. Univ. Western Ontario Ser.
Philos. Sci., Vol. 9. MR0485370
[Feferman, 2013] Solomon Feferman, Foundations of unlimited category theory:
what remains to be done, Rev. Symb. Log. 6 (2013), no. 1, 6–15,
DOI 10.1017/S1755020312000111. MR3033209
320 PENELOPE MADDY
[Hamkins, 2012] Joel David Hamkins, The set-theoretic multiverse, Rev. Symb.
Log. 5 (2012), no. 3, 416–449, DOI 10.1017/S1755020311000359.
MR2970696
[Hellman, 2003] Geoffrey Hellman, Does category theory provide a framework for
mathematical structuralism?, Philos. Math. (3) 11 (2003), no. 2,
129–157, DOI 10.1093/philmat/11.2.129. MR1980559
[Hilbert, 1900] David Hilbert, Mathematical problems, Bull. Amer. Math. Soc.
(N.S.) 37 (2000), no. 4, 407–436 (electronic), DOI 10.1090/S0273-
0979-00-00881-8. Reprinted from Bull. Amer. Math. Soc. 8 (1902),
437–479. MR1779412
[Isbell, 1967] John Isbell, Review of [Lawvere, 1996] Mathematical Reviews,
MR0207517 (34 #7332), pp. 1354–1355.
[Kennedy, 2014] Juliette Kennedy (ed.), Interpreting Gödel, Cambridge University
Press, Cambridge, 2014. Critical essays. MR3381588
[Kline, 1972] Morris Kline, Mathematical thought from ancient to modern times,
Oxford University Press, New York, 1972. MR0472307
[Kunen, 1980] Kenneth Kunen, Set theory. An introduction to independence
proofs, Studies in Logic and the Foundations of Mathematics,
vol. 102, North-Holland Publishing Co., Amsterdam-New York,
1980. MR597342
[Lawvere, 1964] F. William Lawvere, An elementary theory of the category of sets,
Proc. Nat. Acad. Sci. U.S.A. 52 (1964), 1506–1511. MR0172807
[Lawvere, 1966] F. William Lawvere, The category of categories as a foundation
for mathematics, Proc. Conf. Categorical Algebra (La Jolla, Calif.,
1965), Springer, New York, 1966, pp. 1–20. MR0207517
[Linnebo and Pettigrew, 2011] Øystein Linnebo and Richard Pettigrew, Category theory as an
autonomous foundation, Philos. Math. (3) 19 (2011), no. 3, 227–
254, DOI 10.1093/philmat/nkr024. MR2842919
[Mac Lane, 1971] Saunders Mac Lane, Categorical algebra and set-theoretic foun-
dations, Axiomatic Set Theory (Proc. Sympos. Pure Math., Vol.
XIII, Part I, Univ. California, Los Angeles, Calif., 1967), Amer.
Math. Soc., Providence, R.I., 1971, pp. 231–240. MR0282791
[Mac Lane, 1981] Saunders Mac Lane, Mathematical models: a sketch for the phi-
losophy of mathematics, Amer. Math. Monthly 88 (1981), no. 7,
462–472, DOI 10.2307/2321751. MR628015
[Mac Lane, 1986] Saunders Mac Lane, Mathematics, form and function, Springer-
Verlag, New York, 1986. MR816347
[Maddy, 1997] Penelope Maddy, Naturalism in mathematics, The Clarendon
Press, Oxford University Press, New York, 1997. MR1699270
[Maddy, 2011] Penelope Maddy, Defending the axioms: on the philosophical foun-
dations of set theory, Oxford University Press, Oxford, 2011.
MR2779203
[Mathias, 1992] Adrian R. D. Mathias, What is Mac Lane missing?, Set theory of
the continuum (Berkeley, CA, 1989), Math. Sci. Res. Inst. Publ.,
vol. 26, Springer, New York, 1992, pp. 113–118, DOI 10.1007/978-
1-4613-9754-0 9. MR1233815
[Mathias, 2000] Adrian R. D. Mathias, Strong statements of analysis, Bull.
London Math. Soc. 32 (2000), no. 5, 513–526, DOI
10.1112/S0024609300217037. MR1767703
[Mathias, 2001] Adrian R. D. Mathias, The strength of Mac Lane set the-
ory, Ann. Pure Appl. Logic 110 (2001), no. 1-3, 107–234, DOI
10.1016/S0168-0072(00)00031-2. MR1846761
[McLarty, 1991] Colin McLarty, Axiomatizing a category of categories, J. Sym-
bolic Logic 56 (1991), no. 4, 1243–1260, DOI 10.2307/2275472.
MR1136454
[McLarty, 2004] Colin McLarty, Exploring categorical structuralism, Philos. Math.
(3) 12 (2004), no. 1, 37–53, DOI 10.1093/philmat/12.1.37.
MR2033719
[McLarty, 2013] Colin McLarty, Foundations as truths which organize math-
ematics, Rev. Symb. Log. 6 (2013), no. 1, 76–86, DOI
10.1017/S1755020312000159. MR3033213
[Moschovakis, 1994] Yiannis N. Moschovakis, Notes on set theory, Undergraduate Texts
in Mathematics, Springer-Verlag, New York, 1994. MR1260432
[Nagel, 1979] Ernest Nagel, The formation of modern conceptions of formal
logic in the development of geometry, in his Teleology Revis-
ited and Other Essays in the Philosophy and History of Science,
Columbia University Press, New York, 1979, pp. 195–259.
[Russell, 1907] Bertrand Russell, The regressive method of discovering the
premises of mathematics. Reprinted in his Essays in Analysis,
Lackey, D., ed., George Braziller, New York, 1973, pp. 272–283.
[Steel, 2014] John R. Steel, Gödel’s program, Kennedy [2014], Interpreting
Gödel, Cambridge Univ. Press, Cambridge, 2014, pp. 153–179.
MR3468186
[Väänänen, 2014] Jouko Väänänen, Multiverse set theory and absolutely undecid-
able propositions, Kennedy [2014], Interpreting Gödel, Cambridge
Univ. Press, Cambridge, 2014, pp. 180–208. MR3468187
[Van Heijenoort, 1967] Jean Van Heijenoort (ed.), From Frege to Gödel, Harvard Univer-
sity Press, Cambridge, 1967.
[Voevodsky, 2013] Vladimir Voevodsky, Slides for his [Voevodsky, 2014]
Available at https://www.math.ias.edu/vladimir/lectures (#10).
[Voevodsky, 2014] Vladimir Voevodsky, Univalent foundations and set theory (ab-
stract). Bull. Symbolic Logic, 20(1):108.
[Von Neumann, 1925] John Von Neumann, An axiomatization of set theory. Reprinted
in [Van Heijenoort, 1967], pp. 394–413.
[Weaver, 2014] Nik Weaver, Forcing for mathematicians, World Scientific Pub-
lishing Co. Pte. Ltd., Hackensack, NJ, 2014. MR3184751
[Wittgenstein, 1978] Ludwig Wittgenstein, Remarks on the foundations of mathemat-
ics, Revised edition, MIT Press, Cambridge, Mass.-London, 1978.
Edited by G. H. von Wright, R. Rhees and G. E. M. Anscombe;
Translated from the German by G. E. M. Anscombe. MR519516
[Woodin, 2011] W. Hugh Woodin, The continuum hypothesis, the generic-

multiverse of sets, and the Ω conjecture, Set theory, arithmetic,
and foundations of mathematics: theorems, philosophies, Lect.
Notes Log., vol. 36, Assoc. Symbol. Logic, La Jolla, CA, 2011,
pp. 13–42. MR2882650
322 PENELOPE MADDY
[Zermelo, 1908] Ernst Zermelo, Investigations in the foundations of set theory. S.

Bauer-Mengelberg, trans., reprinted in [Van Heijenoort, 1967]
pp. 200–215. Also in his [Zermelo, 2010], pp. 188–229.
[Zermelo, 1908a] Ernst Zermelo, A new proof of the possibility of a well-ordering. S.
Bauer-Mengelberg, trans., reprinted in [Van Heijenoort, 1967],
pp. 183–198. Also in his [Zermelo, 2010], pp. 120–159.
[Zermelo, 1930] Ernst Zermelo, On boundary numbers and domains of sets: new
investigations in the foundations of set theory. M. Hallett, trans.,
in [Ewald, 1996], pp. 1219–1233. Also in his [Zermelo, 2010],
E. de Pellegrin, trans., pp. 400–431.
[Zermelo, 2010] Ernst Zermelo, Ernst Zermelo: collected works/gesammelte
Werke. Vol. I/Band I, Schriften der Mathematisch-
Naturwissenschaftlichen Klasse der Heidelberger Akademie
der Wissenschaften [Publications of the Mathematics and
Natural Sciences Section of Heidelberg Academy of Sciences],
vol. 21, Springer-Verlag, Berlin, 2010. Set theory, miscel-
lanea/Mengenlehre, Varia; Edited by Heinz-Dieter Ebbinghaus,
Craig G. Fraser and Akihiro Kanamori. MR2640544
Logic and Philosophy of Science, 3151 Social Science Plaza A, University of Cali-
fornia, Irvine , California 92697-5100
E-mail address: pjmaddy@uci.edu
URL: http://faculty.sites.uci.edu/pjmaddy/
Selected Published Titles in This Series
690 Andrés Eduardo Caicedo, James Cummings, Peter Koellner, and Paul B.
Larson, Editors, Foundations of Mathematics, 2017
686 Alp Bassa, Alain Couvreur, and David Kohel, Editors, Arithmetic, Geometry,
Cryptography and Coding Theory, 2017
685 Heather A. Harrington, Mohamed Omar, and Matthew Wright, Editors,
Algebraic and Geometric Methods in Discrete Mathematics, 2017
684 Anna Beliakova and Aaron D. Lauda, Editors, Categorification in Geometry,
Topology, and Physics, 2017
683 Anna Beliakova and Aaron D. Lauda, Editors, Categorification and Higher
Representation Theory, 2017
682 Gregory Arone, Brenda Johnson, Pascal Lambrechts, Brian A. Munson,
and Ismar Volić, Editors, Manifolds and K-Theory, 2017
681 Shiferaw Berhanu, Nordine Mir, and Emil J. Straube, Editors, Analysis and
Geometry in Several Complex Variables, 2017
680 Sergei Gukov, Mikhail Khovanov, and Johannes Walcher, Editors, Physics and
Mathematics of Link Homology, 2016
679 Catherine Bénéteau, Alberto A. Condori, Constanze Liaw, William T. Ross,
and Alan A. Sola, Editors, Recent Progress on Operator Theory and Approximation in
Spaces of Analytic Functions, 2016
678 Joseph Auslander, Aimee Johnson, and Cesar E. Silva, Editors, Ergodic Theory,
Dynamical Systems, and the Continuing Influence of John C. Oxtoby, 2016
677 Delaram Kahrobaei, Bren Cavallo, and David Garber, Editors, Algebra and
Computer Science, 2016
676 Pierre Martinetti and Jean-Christophe Wallet, Editors, Noncommutative
Geometry and Optimal Transport, 2016
675 Ana Claudia Nabarro, Juan J. Nuño-Ballesteros, Raúl Oset Sinha, and Maria
Aparecida Soares Ruas, Editors, Real and Complex Singularities, 2016
674 Bogdan D. Suceavă, Alfonso Carriazo, Yun Myung Oh, and Joeri Van der
Veken, Editors, Recent Advances in the Geometry of Submanifolds, 2016
673 Alex Martsinkovsky, Gordana Todorov, and Kiyoshi Igusa, Editors, Recent
Developments in Representation Theory, 2016
672 Bernard Russo, Asuman Güven Aksoy, Ravshan Ashurov, and Shavkat
Ayupov, Editors, Topics in Functional Analysis and Algebra, 2016
671 Robert S. Doran and Efton Park, Editors, Operator Algebras and Their
Applications, 2016
670 Krishnendu Gongopadhyay and Rama Mishra, Editors, Knot Theory and Its
Applications, 2016
669 Sergiı̌ Kolyada, Martin Möller, Pieter Moree, and Thomas Ward, Editors,
Dynamics and Numbers, 2016
668 Gregory Budzban, Harry Randolph Hughes, and Henri Schurz, Editors,
Probability on Algebraic and Geometric Structures, 2016
667 Mark L. Agranovsky, Matania Ben-Artzi, Greg Galloway, Lavi Karp, Dmitry
Khavinson, Simeon Reich, Gilbert Weinstein, and Lawrence Zalcman, Editors,
Complex Analysis and Dynamical Systems VI: Part 2: Complex Analysis, Quasiconformal
Mappings, Complex Dynamics, 2016
666 Vicenţiu D. Rădulescu, Adélia Sequeira, and Vsevolod A. Solonnikov, Editors,
Recent Advances in Partial Differential Equations and Applications, 2016
665 Helge Glöckner, Alain Escassut, and Khodr Shamseddine, Editors, Advances in
Non-Archimedean Analysis, 2016
For a complete list of titles in this series, visit the

AMS Bookstore at www.ams.org/bookstore/conmseries/.
690 CONM
Foundations of Mathematics • Caicedo et al., Editors
This volume contains the proceedings of the Logic at Harvard conference in honor of
W. Hugh Woodin’s 60th birthday, held March 27–29, 2015, at Harvard University. It
presents a collection of papers related to the work of Woodin, who has been one of the
leading figures in set theory since the early 1980s.
The topics cover many of the areas central to Woodin’s work, including large cardinals,
determinacy, descriptive set theory and the continuum problem, as well as connections
between set theory and Banach spaces, recursion theory, and philosophy, each reflecting a
period of Woodin’s career. Other topics covered are forcing axioms, inner model theory,
the partition calculus, and the theory of ultrafilters.
This volume should make a suitable introduction to Woodin’s work and the concerns
which motivate it. The papers should be of interest to graduate students and researchers in
both mathematics and philosophy of mathematics, particularly in set theory, foundations
and related areas.
ISBN 978-1-4704-2256-1
9 781470 422561
CONM/690
AMS

[Contemporary Mathematics 690] Andres Eduardo Caicedo, James Cummings, Peter Koellner, Paul B. Larson - Foundations of Mathematics_ Logic at Harvard Essays in Honor of W. Hugh Woodin’s 60th Birthday March 27-29, 2015 Harvard Unive

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

[Contemporary Mathematics 690] Andres Eduardo Caicedo, James Cummings, Peter Koellner, Paul B. Larson - Foundations of Mathematics_ Logic at Harvard Essays in Honor of W. Hugh Woodin’s 60th Birthday March 27-29, 2015 Harvard Unive

Diunggah oleh

Hak Cipta:

Format Tersedia

690

Andrés Eduardo Caicedo

American Mathematical Society

Andrés Eduardo Caicedo

Andrés Eduardo Caicedo

American Mathematical Society

2010 Mathematics Subject Classiﬁcation. Primary 03E55; Secondary 03E60, 03E57,

Library of Congress Cataloging-in-Publication Data

This volume presents a collection of papers related to the work of W. Hugh

Participants in the group picture

Nate Ackerman Joel David Hamkins

Joan Bagaria Douglas J. Hoﬀman

Paul McKenney Nam Trang

Kaethe Minden Joseph Van Name

Charles Parsons Philip D. Welch

Rehana Patel Kameryn Williams

Norming inﬁnitesimals of large ﬁelds

2010 Mathematics Subject Classiﬁcation. Primary 46H40, Secondary 03E35, 03E50.

1.1. Basic definitions. We first recall some basic definitions.

A character on a complex algebra A is a homomorphism from A onto C. The

neighbourhood U of x, there exists f ∈ C(X) with f (x) = 1 and f (y) = 0 (y ∈ X \ U ). Every

Definition 1.3. A radical homomorphism from an algebra A is a non-zero

Let Ω be a compact space, and take x ∈ Ω.

1.2. Kaplansky’s problem. Let Ω be a non-empty, locally compact space,

Theorem 1.5. Let Ω be a non-empty, compact space, and suppose that θ is a

Corollary 1.6. Let Ω be a non-empty, compact space. Then there is an

5 There are more general deﬁnitions in [6].

Theorem 1.8. (CH) Let Ω be an inﬁnite compact space. Then there is a

Thus CH implies ¬NDH.

2. Ordered groups and ﬁelds

2.1. Ordered sets. Let S be a non-empty set. A strict partial order on S is

2.2. Ordered groups. Let G = (G, +) be an abelian group with identity 0.

Let K be an ordered ﬁeld with value set ΓK . Deﬁne + on ΓK by

strictly decreasing, inﬁnite sequence in the well-ordered set that is supp f .

and F(Z) = {a : v(a) ≥ 0} is clearly identiﬁed with

the algebra of formal power series in one variable. Here Z+ = {n ∈ Z : n ≥ 0}.

Theorem 2.4. Let A be a commutative, unital Banach algebra, and let a ∈ A.

2.6. Prime ideals in algebras C(X) and super-real ﬁelds. Let X be

Definition 2.5. Let X be a non-empty, completely regular topological space,

Certainly AP is an integral domain, and so AP does have a quotient ﬁeld, KP .

Definition 2.6. Let K be an ordered ﬁeld. Then K is a super-real ﬁeld if K is

Super-real ﬁelds are the topic of the monograph [10].

Theorem 2.7. Let Ω be a compact space. Suppose that there is a non-maximal,

2.7. Equivalence of compact spaces. It is natural to wonder if all inﬁnite,

Theorem 2.8. (i) Assume that there is a discontinuous homomorphism from

We shall reformulate the above theorem in the language of ultraﬁlters. Here U

3. Structure of ordered sets, groups, and ﬁelds

Theorem 3.2. The triple (G, +, ≤) is a divisible β1 –η1 –group. Further,

In particular, any two divisible η1 –groups of cardinality ℵ1 are isomorphic. The

We ﬁnally step up further to consider ﬁelds.

In particular, any two real-closed η1 –ﬁelds of cardinality ℵ1 are isomorphic.

We immediately obtain the following consequence from the above remarks.

Theorem 3.6. Let K be a β1 –ﬁeld. Then K is normable. In particular, let Ω

This shows that ¬NDH is not equivalent to CH in ZFC.

Theorem 3.10. Let A be a non-unital, complex integral domain with |A| = c.

Corollary 3.11. (CH) Let A be a non-unital, complex integral domain with

4. Related questions with CH

4.1. The range of radical homomorphisms. In Theorem 3.5, we stated

There are many further examples of commutative, radical Banach algebras

Definition 4.2. Let A be a commutative algebra. A family {Pi : i ∈ S} of

There is a sense in which there is a topology on the set of prime ideals in A

Theorem 4.3. Let Ω be a locally compact, non-compact space, and let θ be

is a homomorphism with kernel I from C0 (Ω) into a commutative, radical Banach

4.3. C ∗ -algebras. As we remarked there are classical positive results about

+, · , ≤) is a real-closed η1 –ﬁeld with value group

The ﬁeld R is universal in the class of ﬁelds K such that ΓK is a β1 -group:

6.4. The field R. We finally come to my favourite real-closed field. We define

Theorem 6.7. The ﬁeld (R, +, · , ≤) is a real-closed, Cauchy complete η1 –ﬁeld,

The ﬁeld R is universal in the class of completions of β1 –ﬁelds: every completion

Theorem 7. Every set of basic indexing parameters p ∈ De (≤ 0e ) deﬁnes an

Qe (w) = {x | ∃t > w[θe (x)[t] < γ(x)[t] & (Y θe (x))[t] Y ]}.

Proof. Let p be basic indexing parameters with a model of true arithmetic

triple c, DΦ , DB . Enumerate DB back in B1 . Extract the chit c from

in Θα and enumerate the axiom n, DΦ in Δα .

x, D1 , D2 ∈ Ψ. Let DX = {n ∈ X | O(n) ∩ (D1 ∪ D2 ) = ∅}. Enumerate

c(n), DΦ , DB in Θe . Let Dn be the union of the currently valid axioms in