Also published inthis series Theories of Ethics edited by Philippa Foot Knowledge and Belief edited by A. Phillips Griffiths The Philosophy of Sclence edited by P. H. Nidditch ‘The Theory of Meaning edited by G. H. R. Parkinson Political Philosophy edited by Anthony Quinton ‘Philosophical Logi edited by P. F. Strawson The Philosophy of Perception edited by G. J. Warnock The Philosophy of Action edited by Alan R. White Other volumes are in preparation THE PHILOSOPHY OF MATHEMATICS Eéited by JAAKKO HINTIKKA OXFORD UNIVERSITY PRESS 1969Oxford University Press, Ely House, London W.1 © oxronp university Press 1969 CONTENTS m. v. ML vu. van. a x x ‘Sewanee ENTAILMENT AND FORMAL DERIVABILITY byE. W. Beth “Tur Cowrveressss oF Tue Finst-Onpen FUNCTIONAL ‘CaLcutus by Leon Henkin ‘Covruzreness 1 THe Tatory oF Tyres by Leon Henkin [LANGUAGES IN wc SELF REFERENCE 8 PossteLe by Raymond M. Smullyan TwFoRMAL RIOOUR AND COMPLETENESS PROOFS by Georg Kreisel SysTEas oF PREDICATIVE ANALYSIS by Solomon Feferman ‘AN InTERPRETATION OF THE INTUTIONISTIC SENTENTAL, ‘Looie by Kurt Gédet ‘Tue Prasenr Tisony oF Tunino Macuine Comrurantry by Hartley Rogers, Jr ‘Matueuarica Lootc: What HAs It Dove FoR THE Punvosopay oF Manisarics? (Excerpt) by Georg Kreis! ‘Tue Meramnsics or tie Catcutus by Abraham Robinson ‘Waar 1s Eusnenrany Growerny? by Alfred Tarski [Nomis on rte ConTmipuroRs oes oF Nass 2 st " 95 ur 153, 164 176 178 1sINTRODUCTION ‘ray seem strange o reprint the papers collected hereunder th itl "The Philosophy of Mathematics’. In the majority of ther, n0 specially philo- sophicl problems sem to be discused ‘Although this impression is not unfounded, in the presen circumstances the papers appearing here nevertheless exemplify the kind of reading which judge to be by far the most wseful for 8 student of the plilosophy of ‘mathematics. In these day, the worthwhile articles and books devoted to the philosophy of mathematics inthe narrow sense ofthe wordpilsophy” are few, and thee quality frequently leaves a great deal tobe desired. Often the reader soon discovers that such a book or article isnot based on any- thing ike the adequate acquaintance with the large and important body of ‘materials which s indispensabe for any future philosophy of mathematics, ‘namely the literature of symbolic logic and foundational stdies. I is not likely that any substantial progress can be made in the genuinely philo- sophical study of mathematics without using the concepts and results ‘developed in this literature to a much greater extent than has happened so far. However true it may be that no philosophical problems are solved by results in logic, not 10 use the wealth of material which is available here simply makes the discussion as unrealistic asa discussion of legal philo= Sophy would bei it were not based on any knowledge of the actual oper ‘ton of the la. Thus the most useful contibuton that a book of readings can make inthis area isto ensure that some ofthe recent result in logic tnd foundational studies are accessible to an interested reader. To do this Js the fist im ofthe present volume. ‘What, theo, are these philosophically relevant insights that a philos- ‘opher of mathematics ought to be interested in? This question cannot be answered adequately within te confines ofthis introduction, for ananswer ‘would amount o a survey of an extensive, rapidly growin, and complex field. Furthermore, about as good a survey as anyone ean hope to find already exists in Andree) Mostowski's volume Thirty Years of Foundational ‘Studies, listed in the bibliography at the end ofthis volume. (All other ‘references in this introduction are also to works listed inthe same bibio- raphy, unless otherwise specified) ‘What can be done here is 10 try to supply enough background for the ‘selections reprinted i this volume to enable a reader to make use of ther ‘The following informal sketch cannot help being loose and inaccurate. I have tried mainly fo indicate some ofthe most interesting interconnections2 INTRODUCTION between the diferent types of recent work represented here, and aio 10 mention a few recent developments which ate not represented in our selections Probably the mest important large-scale recent development in logic and foundational studies isthe growth ofthe theory which used to be known a8 logical semantics but whichis now usualy referred to as mode! theory. It ‘an be characterized, very roughly, as a study of the interrelations of Tanguage—usually a! precise logical symbolism—and the “reality” this language represents. The basic questions concer the conditions in which a formula can be suid to be true ina structure or ‘mode’ Simple a this ‘question may seem, it soon leads 1 interesting problems. One of them concerns the class of sentences true in all models—the class of logical truths, as they are usually called. Can this class be represented athe the- ‘orems of some axiomatic system? This completeness question is answered afirmaively for firstorder languages (quantifcatonal languages). The first proof was given by Gade in the early thirties, (See Kurt G&del, ‘Die Vollstindigkeit der Axiome des ogschen Funktionenkalkil', Monats- ef far Mathematik und Physik, Vol 37 (1930), pp. 349-80.) One of the ‘mos important subsequent proofs i given in Henkin’ fit paper reprinted in the preent volume. Henkins proofs interesting for several reasons, one ‘of which is his strategy of forming e sequence of certain maximal consistent sts of sentences whose union can thea be sen to be usable as a mode! in ‘hich al its members are true. Certain es of formulas thus exemplify the very structure which these formulae impose on the models in which they are true, It is known that no similar proof exists in higher-order logics. Logical ‘truths abou arbitrary sets, unlike logical ruth about arbitrary individuals, ‘cannot be axiomatize. This fllows from G8del's incompleteness proof for arithmetic, forall the requisite arithmetical assumptions are eatily ex- pressed in terms of higher-order logics. Most ofthe purportedly popular discussions of Géde!'s famous result are somewhat tnstiefactory, how lever. One of the best seifcontained expositions seems to be Raymond, ‘Smullyan's paper reprinted here. The whole complex of problems concer ing Gode's proof of the impossibility ofa consistency proof ofa system containing arithmetic within the same system is discussed in a wider per- spective in Feferman’s salient paper “Arthmetization of Metamathe- matics However, Henkin’s other paper reprinted here shows that completeness can be reached even in higher-order logis if we ae willing to admit ‘non- Standard models in which quantification over “all sets’ no longer refers to all arbitrary subsets of our domain of individuals but only to all members INTRODUCTION a ‘of some' ‘arbitrary’ subset ofthe power set ofthe set of individuals; and similarly for pairs, triples, et. These subsets cannot be completly abiteary, however, fr they must be rch enough tallow the interpretation of al out formulas (second-order formulas). This sa rather strong requirement, for it means intr alaclosure wih espect to the ‘projective’ operations that correspond to different kinds of quantiation and no just with respect fo the Boolean operations. If this fact is overiooked, Henkin's completeness proof for type theory may give the impression that higher-order logics are ‘much simpler than they really are. Tie actual dic is illustrated, for example, by the complexity of the problem of extending the Gentzen- Herbrand techniques (10 be mentioned later) from first-order logic 10 higher-order logics. Takeut’s problem which is discussed in Schate, “Syntatial and Semantcal Properties’ and which ha only recently been partially solved (by Tait, Prawitz and others) sa cas in pont. ‘An intriguing aspect ofthe completeness and incompleteness results i that one of their starting-point (viz. our conoept of what constitutes com- pleteness) is inevitably an idea which ean perhaps be formulated in nave Set-theoretical terms bat which either is not formulated axiomatcally 10 ‘begin with or which (in the case of incompleteness) cannot even possibly be 0 formulated, Yet concepts of this kind are most intersting. We seem to hhave many clear intuitions concerning them, and itis important to develop ‘waysof handling them. Someaspectsofthissituationaredscusedin our first selection from Kreis, entitled ‘Informal Rigour and Completeness Proof ‘One might describe the phenomenon of incompleteness by saying that ‘one cannot axiomatcally rule out all non-standard models and catch oaly the intended ‘standard’ ones. The inevitable presence of non-standard ‘models can occasionally be turned into a blessing, however, 2% demon strated by Abraham Robinson's non-standard model for analysis. These models serve to vindicate some ofthe locations and idets ofthe old ‘metar physics ofthe calculus’ which meanvile have been relegated to the status ‘of hopelessly loose heuristic ideas. Even infnitesimals, the bugbear of every introductory calculus course, can thus be made perfecly respectable. Here is one of several recent developments that demonstrate scikingly the ‘elevance of current work t0 the tational issues in the philosophy of mathematics, ‘Completeness and incompleteness results are only samples, however, of the variety of results that have been obtained in model theory concerning ‘what can or cannot be expressed in diferent kinds of language and con- czming the variety of models that a theory can or cannot have. These insights into the limits of what can be expresed in diferent kinds of languages ae among the most important recent developments in the whole‘ INTRODUCTION tea of logic and foundations, and deserve mote attention than has been ven to them by philosophers. The only result that has provoked much Dhilosophial discussion i the Léwenheim-Skolem result that every axiomatizable first-order theory has a countable model. (For ie variants and generalizations, see Vaught’s 1964 paper.) The fact that this includes, systems of axiomatic set theory in which the existence of uncountable ses ‘an be proved is but a further instance of the dificult of ruling out n= desirable ‘non-standard’ models axiomatically, Other results are not much les interesting. A good example of them is Vaught’s survey of models of ‘compete theories, ‘Asillustated by the Lowenheim-Skolem theorem and by Gde!’s com- pleteness and incompleteness results, many ofthe earlier basic results of mode logic belong essentially to model theory. However, much of the ‘systematic development of this atea i relatively recent, often deriving its inspiration from Tarski Contributions tothe Theory of Models, etc.) and ‘Abraham Robinson. ‘Thedevelopment of model theory hasnot remained without repercussions {or the deetve techniques of first-order logic. The basic insights go back to Herbrand and Gentzen. Ther techniques were gradually adopted in the {orm of ‘natural deduction methods in many standard expositions, but an Interest in the underlying semantial ideas was not revived on a larget scale until 1955. Around that time, Hintkka, Beth, and Seite (inter alia) published treatments of firstorder logic in which the proof of logical truth of a first-order formula can be thought of as an unsuosesful attempt to describe a counter-example to it, These counter-example de- scriptions, reminiscent of Henkin’s completeness proof, could even be ‘thought of as actusl models in which the formula in question would be fase ‘The essential simplification lies inthe use of simpler descriptions of the ‘possible worlds" which would serve as counter-modele~descriptons simpler than Henkin’s very extensive maximal consistent sts. (This simplification is present also in Hintikka’s method, contrary to the i= pression which the concluding remarks of Beth's paper in the present, ‘olume might give) It is clear, however, that similar ideas were already inspiing Herbrand, although they were not developed systematically a the ‘Beth's paper included ere is the fst and perhaps the freshest, descrip- tion ofthe approach which he cals the method of semantical tableaux. A ‘completeness proof is sketched fr this method i his paper, and a number ‘of suggestive comparisons with other issues in logic and the philosophy of ‘mathematics are presented. “The basic idea of thinking of proof and disproofs as attempted mode! INTRODUCTION 5 constructions seems to me of great philosophical interest and importance. ‘The Herbrand-ype pont of vew hasin any cae led o fruitful and interest- Ing consequences especially inthe dretion of Craig's interpolation lemma {proved in is 1957 paper) From the stme point of view, arestment of intuitionistic logic was given in 1956 by Beth. Ie was commeated on by Kreisel and Dyson and carried further by Kripke, Kripke’ starting-point in his 1964 paper was bis earlier ‘madel-theoreticl treatment of modal logics along the same ins. Like so ‘many ideas underlying the work of the last few decades, the ist ideas in this diration go back to G8dal's early papers. In the paper entitled "Ene Interpretation des intltionstichen Aussagenkalkl’ and transite into English here, the connection between intuitionistic and modal logis i pointed out forthe frst time. Some ofthe other suggestions made in Godel's Paper, notably his remarks on what happens i's interpreted as express- ing provabiity in some particular logieal system, are developed further by Richard Montague in his discussion of “Syntatial Treatments of Modality’, Acta Philesophica Fennice, Vol. 16 (1963), pp 183-67 ‘The model-heoretical discussions by Beth and Kripke of intuitionistic logic do not decide the interpretation of intuitionistic mathematics, how ever this remains an interesting and lively area fr further work and even {or esentally new ideas. An interested reader wl ind a rich lode of sup- ‘estions in those writings (or parts or writings) of Georg Kresl which are not printed bere. “The idea that disproofs can be thought of as attempted model construc tions calls one's attention tothe broader problem of the elation of formulas and ther models. These relations turn largely on the concepts of truth and satisfaction, These conceps are normally explsined in fairly strong set- theoretical terms, Can thexe concepts be vated in suitable ways? Can they ‘beexplained so as to rule out already at thi stage some ofthe nonstandard models? No satisfactory answers to these questions ae kel tobe found inthe lteratore. Can formulae (Setences) and models be related to each ‘other some wiys which ae closer to our actual methods of eating them “im practie” than are the st-theoretical methods usually employed? In- teresting suggestions, couched In game-theortical terms, have been made ‘sentially inthis direction by Lorenzen. (In adtion tothe paper listed in fur bibliography, see W. Stegmiller, ‘Remarks on the Completeness of Logical Systems Relative to the Valiity-Concepts of P. Lorenzen and K. Lorenz’, Notre Dame Joural of Formal Logie, Vol. (1964), pp. 81-112) ‘A closely similar suggestion is put forward by Hintikka, who aso argues for the importance ofthese ideas in understanding te actual wse of quanti+ fication concepts inthe applications of language. (See ‘Language-Games‘ INTRODUCTION for Quant’, American Philosophical Quarterly, Supplementary Mono= ‘raph no. 2, 1968.) Thee suggestions seem to have at lest the mecit that they make it extremely natural to consider cetai types of deviations from classical logic. Lorenzen stresses the naturalness of intuitionistic loge from this point of view. Other deviations ean be obtained by requiring suitable kinds of recursvity (computability) of the strategies to be used in these ‘games. (What could it mean actually to se a non-recursve strategy In a ‘game? There doesnot seem to be any unique way of doing tis, however, land we soon run into the complex of problems connected with such more or less consirutivistic ideas at Géde's extension ofthe fnitstic point of ‘iew, the ‘no counterexample interpretation’ of logic and arithmetic, te. ‘The work represented inthe rest of our selections requires fewer com> ‘ments, The first pat of Feferman's paper i a lucid survey of the develop ‘ment of one important older idea inthe foundations of mathematics, vie the idea of predicativity. The second part presets important results con. ‘cerning this concept. Predicative reasoning turns out to be characterized, roughly, by transnite induction up to a definite ordinal of the second ‘number clas. (This isan interesting example of a general phenomenon: frequently the strength of one's assumptions turns out tobe measurable by the ordinals one needs in one's work) The details ofthis part of Fferman's paper may be rough going for most readers of this volume, but the main ‘and ther importance ought to be possible to grasp without all the ‘etal, Feferman’s new paper ‘Autonomous Transfiite Progressions te. forces and partly replaces the older one. (It contains an improved form lation of the technical results and extends them further: it also presents sharper appraisal of what has been accomplished and what remains to be ‘done in their area.) The reader i referred to ths new pape fora discussion ‘of recent developments in the study of predicativty. ‘The philosophical relevance of the theory of recursive functions ought to ‘be obvious, but nevertheless litle use hasbeen made of ts potentialities by philsophers. The fact thatthe only major conceptual idealization involved inthe theory of Turing machines, as compared with ral digital computers, {sinfnite memory aso illustrates the importance of thi theory Little more can be sti here, however, than to refer the reader not only to Hatley Rogers riliant expository paper, printed here, but to his long-awaited book. The theory of automata withou this idelizing assumption of infinite ‘memory is discussed inthe already classical paper by Rabin and Scot ‘Although we seem to have ia the notion of a Turing machine = good ‘explication of the idea of effective computability, the station is not a8 simple as it may appear. From G8del's note on “Eine bisher noch nih bendtate Erweiterung ds riten Standpunkts’ one can see that there are f INTRODUCTION 7 ways of extending the finite point of view by resorting to constructive ents of a higher type which are not initlly given any “mechanical ‘counterpart. From thi, a distinction between what is inistic and what ‘constructive seems to enue, Futher distinctions are due tothe fact that a proof ofthe termination ofa combinatorial procedure may not itself be combinatorial—a standard objection by the intuitionsts to the use of ‘ecursive function theory to explicate thei intentions. ‘Even more obvious than the relevance of recursive function theory isthe ‘connection between recent work and traditional problems in the case ofthe oundatons of geometry. The lines indicated in Tarsi's paper have been since followed by Schwabhiuser and by a numberof other logicians and ‘mathematicians ‘So far, have said nothing about what soften thought of asthe central area of foundational studies, viz. set theory. Nor is any selection on set ‘theory included inthe present volume. There exists, however, an excellent survey ofthe most important recent results in Paul Cohen's ite volume. te that is definitive can be suid ofthe subject mater of axiomatic set theory in the present situation, it seems to me. It has been found that some ‘of the mos important open questions (especially the status ofthe Con= tinuum Hypothesis) cannot be solved on the bass ofthe customary axioms ‘of set theory. Uhope that itis fir to say that most ofthe diferent ways of ‘extending these axioms are ether unconvincing or too weak to be of much interest. Although many interesting suggestions are being discussed, itis hard to tell where they will lead ‘The only point that T want to make here concerns the dretion in which urter axioms have been sought, (This, € seems 10 me, migh tun out 10 be the crucial philosophical problem in this area.) These new axioms have ‘often been formulated as postlations of suitable “very large’ cardinal numbers. Theres certain inner logic at work here, but one cannot help ‘wondering whether some completely diferent way oflooking a them might be philosophically defensible, After ll, these assumptions have in general defite purely arithmetical consequences, although we donot seem to have ‘strong enough intuitions about these arithmetical consequences to say ruc aboot them. But are some entirely different intuitions or heuristic ideas perhaps possible here? To ilustrat the implications ofthis question, itmay be mentioned that there exists a small and possibly frutese attempt to strike out in 2 completely diferent direction, to which the wider per- spectives just mentioned nevertheless lend some interest. Mycielski has ‘experimented with completely unorthodox assumptions formulated in fame-theoretical terms Which have certain very pleasant mathematical consequences. (See J. Mycielski, “On the Axiom of Determinatenest,5 INTRODUCTION Fandamenta Mathematicae, Vol. $3 (1964), and J. Mycielski and H. ‘Steinhaus,‘A Mathematical Axiom Contraditing the Axiom of Choice’, Bulletin de I Academie Polonaise des Sciences, Ser. 1M, Vol. 10 (1962)) ‘Whether they wil remain mere curiosities, nobody seems to know, and the nicely ordered infinite numbers have so far been too seductive for the ‘majority of logicians to trade them for Myeiesk's axiom, with its much ‘more disorderly consequences forthe interrelations of infnite numbers, Some philosophical suggestions of the recent work in logic and in foundational studies are made in Krese’s simulating paper, “Mathe- ‘matical Loge: What Has It Done forthe Philosophy of Mathematics, of hich an excerpt is reprinted her. (Some new information concerning the topics discused theres found inthe appendix to this pape, 28 wel a in Kreise’s subsequent papers) Some of the issues touched upon earlier in the preset introduction are also discussed by Kreisel inthis long essay. Other philosophical suggestions have been made elsewhere. For instance, in ‘Are Logical Truths Analytic and in other papers Hintkka argues that certain distinctions one can make in frst-order logic constitute the best modern ‘explication of Kant's distinction between analytic and synthetic peo- Pottions, at least in so far as this distinction applies to loge and mathe= matics. Some ofthe historical basi of this claim also appears from Beth's brief discussion (in Section II of his paper in the present volume) ofthe Aristotelian concept of ecthesir and its later history SEMANTIC ENTAILMENT AND FORMAL, DERIVABILITY EW, Berit “The aggrnate of al applications of loge wll not compare wih tte of he pre ery ta For when one hs Surveyed the whole subject one wil see that the theory ol logis instar at we atin to it the vision and the ata tment ofthat Ressorabeness ‘for the sake of which the ‘Heavens andthe Earh have ben created 8. Pance 1, Introduction, If Uy, Un aNd Var sentences (which may be ether true or fle) then we often say that is (oF n00) @ LOGICAL CONSEQUENCE Of Uy. Use vm oF that the conclusion V LOGICALLY FOLLOWS from the premisses Uj, Us, fr instance, we might say thatthe conclusion: ‘Some Panthers are not Swans, {oGteaLL roLLoWs from the premises ‘Some Panthers are not Mammels, ‘Some Mammals are not Swans. ‘isthe tsk of logic (an, in fac, its main task) to clarify his notion and to point out the conditions forts application. Now if we try to carry out ‘this task, we find thatthe term ‘logical consequence’ covers two distinct notions which, approximately, can be characterized as follows (@ Formal Decay. There are certain formal rales of inference, each of which, if applied to appropriate premises, yields a certain immediate fonclsion, Well-known rules of inference ate, for instance, the modus ponens: UP then Q: cetera 2 and the conerio simplex: Some A's are B; ‘Some B's are A From Madudsnge onde Kounkithe Nederland Akademie van Wetescharpen, fain Laverton. (Arteta 1939, 9p. 309. Reseed ‘ypermintion of te Royal Neibetands Academy of Scene0 5. W. BETH [Now Vis called a Locicat consequence of Uj, Us.» if it is formally Aerieable from U;, ay. thats i sartng from the premisses U, Uy tnd applying agin and again the fules of inference, we can finally obtain ‘the conclusion V. ‘Te rules of inference ae called ‘formal’ on account of the fact that they ‘an be stated in purely “‘ypegraphial terms, without any reference to the meaning ofthe sentences to which they are applied.) (Gi) Semantie Encaiiment. is known from Aristotle's logistics tha the conclusion: Some Panthers are not Swans does nt LOOICALLY FOLLOW from the premisses: Some Panthers are not Mammal, and: Some Mammals are ‘ot Swans. If we wish to show tis, we usually argue as follows. Let us replace the terms ‘Panther “Swar’, and “Mammal” by “Pi, ‘Swine’, and ‘Mammoth, respectively. Then we get new premises, namely’ ‘Some Pigs are not Mammoths, ‘Some Mammoths are not Swine, anda new conclusion, namely Some Pigs are not Swine [Now the new premises are true, whereas the new conclusion is false. Hence the new conclusion cannot Locicatty FOLLOW from the new remisses and in view ofthe simalarity as to logical form the old conclusion {doesnot LOGICALLY FOLLOW from the old premises, Inthis discussion, the “ath vale (that i, the truth oF falsehood) of the {oGCAL CONSEQUENCE whichis relevant inthe context may be denoted es semantic enalment, It can be defined as follows: Vissid tobe semantically entailed by Us, Uy if we cannot replace the terms in Uy Use va ind bby new terms in such & manner thatthe new premises UY, Uf... are ue ‘whereas the new conclusion V* is ale. If, however, a inthe above example, such new terms canbe found (in which case Vis not semantically entailed by U;, Us.) then we say that these terms provide us witha suitable courtererample for proving that V oes no! LOGICALLY FOLLOW from Ui, Use wil be clear that, on the other hand, the meaning of the old terms ‘Panther’ Swan’ nd Mammal’ is completely irrelevant in fac, the above counter-example shows that no conclusion ofthe Form: Some P's are not S, SEMANTIC ENTAILMENT " {is @ LOGICAL Consequence of the premises: ‘Some P's are not My Some M'sare no. Such expressions, containing indeterminates' 15, hich stand for arbitrary terms, were already frequently usd in traditional logic. Modern Topic instead uses formal Pe) aE oiR@ & MC) @IMOVESO ‘Though the meaning of such formulas will be pretty clear from the present content, twill be beter to Say afew words about the manner in ‘hich thay are fo be interpreted! ‘We select some (non-empty) set of individual, which will be called the universe of discourse (may, for instance, contain ) all human beings, oF @ all natural numbers). The letters °P' “Mf, ‘S", ... stand for tems (properties or predicates) which can be applied to some individuals inthe universe (for instance, () healt, mal, grown-up, o Gl) od, prime, square) ‘The letters", af used as earables ‘ranging’ over the universe, “The atomic formala “P(x 8 Wed t0 state the condition fora certain in- dividual x to have the propectyP. The quantifier “(xis used to express the fact (oF the supposition) that al individuals inthe universe full the con- ition which follows the quantiSer the quant “(Ex)” is sed to express the fat (or the supposition) that some individual in the universe ful the condition by which itis followed. The quantifiers") '(2vu (Es “(Ea), are to be interpreted in the same manner. Te symbols: ed nd "if. then’. Later on, we also se the symbols we which are read: or" and “and only I ‘A counter-example (which yf may not be suitable in some context) Is obtained by selecting a universe and by electing properties P, M, S, for all letters PMS? on? are read: not, and’, 1 A mor eed exposition ound, fr nance, in ny Podement lope des et pba tae conoiay owe he em modelo "errant ‘a bce sled Souseresample” Unaly, adel eran crpretion i 08)2 ew. BETH 2. The Completeness Theorem. The two notions of LOGICAL coNseQuENce hich have Been described in Section 1 are both wsed in applied loge this fact creates situation which, from a methodological point of view, i far from satisfactory. I wish fist to explain the situation and then to discuss its implications; my explanation is based upon the consideration that in some contexts the notion of formal dervability x more elpful, whereas in ‘other contexts we naturally tend to apply the notion of semantic entailment. @ Suppose that we wish to show that Vis a Lodicat CONSEQUENCE of Up, Uys then ts natural to resort to the notion of formal derivablity. ‘We ty to establish a formal deviation of V from Us, Us oon tha i, 8 sequence of applications of the rules of inference which, starting from the premises Ui, U,.-inallyyeldsthe conclusion Vis obviously sufficient {point out one such derivation If instead we should wish 10 apply the notion of semantic enailment, ‘then it would be necesary to examine a posible counter-examples and 10 show thet among them no suitable counter-example for proving that ¥ is ‘ota Locteat CONSEQUENCH of U;, Us, can be found, @ Suppose, onthe other hand, that we wish to show that ¥ does not {oateaLLy FOLLOW from Uj, Uy ..; then we would prefer to apply the notion of semantic enailment. For ii sufcent fo point out one suitable counter-example for proving that V does not LOGICALLY oLLOW from Uj, Uz... But if we should wish to apply the notion of formal derivailiy, then it would be necessary to examine all formal derivations which start, fromthe premises Uj, U0 pol ot that mane of tha the However, the practical advantages which can be obtained by using simultaneously two different notions of Locieat CONSEQUENCE cant, of cours, justify such a procedure. Specifically, by accepting it we incur the risk of obtaining conficting results @ ifthe notion of formal derivabilitis wider than the notion of semantic ‘nialimen, then it may happen tht a coaclusion Vis formally derivable from certain premisses U;, Us». without being semantically entaled by them; if Vis not semantically entailed by Uj, Us. then there is some suitable counterexample for proving ths fact. This counter-example enables us fo repace Uj, Ua. V by new sentences U3, US, ¥* such that His agin formally derivable from U3, US,» whereas U4, US, ‘sppenigh Bat fe iat fete pon ea thts cease eat fitmtie, The tational mame: G's counererample ie “nance Coraer Intec Arstote, Pry Anais Behe) ot nce here Se ‘Berutherobtiou soonest the pea SEMANTIC ENTAILMENT ry are rue and ¥* is fale. Now in the formal derivation of V* from Ut, U3, ve there must occu some rue of inference which is applied to certain true remiss 2, X°,.. nd yields an immediate cooctusion Y which is fale. ‘G) Suppose the notion of semantic entailment to be wider than the ‘notion of formal dervabiity. Then it may happen that a conclusion Vis ‘Semantially entailed by the premises U,, U,... but not formally derivable from them, In other words: Vis not formally drivable from Ui, Us but there is no suitable counter-example for proving that it does not oaIeALLy FoLLow from them. Tn both eases we throw the blame forthe discrepancy upon the notion of formal derivabilty (which shows, by the way, that the notion of semantic entailment isthe more fundamental one ofthe two) and we try toregularize the situation by revising the rues of inference. I this atempt i succesful, ‘then the formal rules of inference () do not yield a false conclusion if applied to tue premisses, and (jj) enable sto derive the conclusion ¥ from the premisses U;, Us... whenever itis semantically entailed by them. Now the two notion! of LOGICAL CONSEQUENCE are equpollent (ey have the same extension) and hence they may be used indiferently. In the systematic construction ofa logical theory, one usually proceeds 18 follows (0) The rules of inference ae stated andthe notion of formal derivability introduced. (AD) The notion of truth (or related notion) is introduced, and it ‘shown that from tru premises no false conclusion canbe formally derived. ‘lD) The notion of semantic entailment i introduced, and its shown ‘that, whenever Vis semantically entailed by Ui, Us »-» it cam aso be formally derived from them, This part ofthe construction, which is known ts the proof of the Completeness Theorem? for the logical theory under consideration, usually presents considerable ficulties. (On the bass of investigations which I have made during the lst few yearn Tam now able to construct logical theories 4: to avoid the above dificulis. The main idea ‘4. Heuristic Considerations: Transformation of Semantic Tableaux. Let us rearrange our tableau for problem (i) inthe following manner: we do ‘not change the left colums, but we extend it by adding below al formulas {nthe right column in the reverse order; the result is a8 follows. 1, M, Bosh, Ament Formal Lol Amsterdam, 1951) p.47; J. aasewen Arai Spepie Oxford 1950, 39 SEMANTIC ENTAILMENT n © eRe) =" (orem @ @SOeMON — Grm © S@aNG) ene) S30 ‘ Me ® © r= o 10) Fay _ ty 2) oD MG mse) © Pa rn) @ soe yan 0) © ESO&PA — hye) ‘This ought to be a pleasant surprise! For it provides us with formal Yin ale column or) Xe+ Yina right column, then the tableau must be spit up, and we insert (@) Xin one left subcolumn and Yin the other (©) Yinone left cokamn and Xin the right coluran nor conjugate with it. (Xin one right subcolumn and Yin the other. SEMANTIC ENTAILMENT a (@ Both Xand Yin one left column and once more both X and Yin the right column nor conjugate with (© Xin one lft subcolumn, Yin the right subcolumn conjugate witht, in the ote et subsolam, and Xie the remaining ght sub (iD) IF (Ex) XC) appeats ina let column oF (x)X(s) in right colon, then we introduce a new individual p and weinsertX(p) in the same column (tis convenient not o apply thisrule untill posibilties of aplying other ‘les have been exhausted; the par of the construction which stats with the introduction ofthe A" individual and which terminates atthe introduction ofthe (k +1)" ones denoted a it & sage), (Wil) If one and the same formula appears in two conjugate columns, then the corresponding tableau (r subtablea) is closed. (G0) Ill subtableaux subordinate to some tablea (or subtableay) are loved, then that tableau (or subtabesu) itself is also closed, 0 If all possiblities of applying rules ()-(vi) have been exhausted, ‘whereisthetableauis not closed, then the tableau provides us witha suitable counter-example, and is termined 1 wil be clear that the rules ()-(ei) exactly reflect the meaning of the words ‘all ome’ not ‘and’ "if. ther or If and only If, bof the symbols by which they have Been replaced. For instance, part of rule (iv) ‘corresponds tothe following rule of semantics: The sentence, "X and Ys true and the formla:'& ¥"is valid), ifand only both Xend Yare true (ce valid). However, these rules assume a completely formal character, if Instead of speaking about an individual p we speak about a symbol. Perhaps it will be useful to give an example ofthe application ofthe ules, ‘Let us construct a semantic tableau, taking as inital formulas (0) LAC V BON in the left column, and (40 V OB) inthe right column (se tableau on p. 2) ‘We have now taken care of point () on the programme outlined ia Section 5. So we tur to points (I) and (ID; these points can be handled ‘very quickly, by statin the following definitions @ A formal proof inthe System of Natural Deduction F of the sequent i, Use ove Va Pa {na closed semantic tableau in which the inital formulas are: Uj, Us. in the left column, and Yj, Fin the right column.2 BW. BETH Valid | Ira @ (Aw) V OBL) @ @ae @ GX) 4) V BON) OM W)A@V BON @ A@V BC) @ Ae) | © wa @ © B®) © 4@) | G0) Ba) GD 0) (40) V BON (12) A@)V BO) ww (3) 4@) « 14 2) (G) A formal derivation in F of the conclusion V from the premises U,, Uys formal proof in F ofthe sequent: Uy Ua (i A formal derivation in F ofthe formula U as a loical contradiction {4 formal proof in F ofthe vequnt ure (the symbol “Oi used to express the fact that no formula appears atthe right side ofthe sequen} iv) A formal derivation in F ofthe formula V a a logical identity is & formal proof inF of the sequent orr. “The main content ofthese defnitions could aso be stated as follows: A formal derivation in F ofthe conciusion Vfrom the premises Uj, Us, 18 ‘closed semantic tableau in which the intl formulas are: U,, Ui, the leftcolumn, and Vin the right column. twill be ler that on account of this treatment of point (IID we avoid ‘the ncesity of dealing with point (1V). It can, however, be argued that itis preferable to give the derivations in Fa more familiar shape. ‘We sball not yield to this argument, but nevrthees i is interesting to show that, i necessary, we could easily give the formal derivations in F a ‘more normal appearance. Specifically, we coud either borrow the general SEMANTIC ENTAILMENT a structure of Gentzen's Calculus of Natural Deduction or adapt the prin- °T Giples underlying his Callus of Sequents “Though the connections of our semantic tableau to some System of [Natural Deduction wee already discussed in Section 4, it wll be useful also to give our above tableau the familiar shape of a derivation in a system of this kin. UV BON (erem) OD WA@V BO 2) AY BO) (itemma) © Fo Galt) © WR) © 4 (ant2) @ We : @ @ADVOB) — Calt,2) we try to transform the closed table into a proofin some Calculus of Sequents, the result is as follows. We obtain a version of this Caleulus, Bs hich is closely related to Hasenjaeger's Symmetric Calculus Aca) AC) Bi) +B) ‘Aa)+ Aca) Be) BU) + Ala), BOD) Ala) VB) F AC), BOO) 0) A) V BOF AC), BO) (90) Ae) V BO] F(@)AC, (0) BO) CO) (A) V BON FAY OBO) ‘As these formal derivations have been constructed, so to speak, in & purely mechanical manner (we have indeed come alarmingly near the Tealization of the “ideal” of a calculus raiocinator, a “logical machine; 1 shal return to this point later on), one might expect them to be rather clumsy and cumbersome. But this anticipation is not corroborated; tothe ‘contrary, the above derivations are remarkably concise, andthe derivations ‘obtained by our procedure ean even be proved tobe, in a sense the shortest ‘ones which are posibe. The frst derivation, for instance, uses rather ‘daring dilemma, which I would certainly have avoided (by means of & ‘reductio), bat which sin view of the tableau, completely justified. I we are submited a derivation belonging to For to some other sytem, such a thse in Hilbert-Ackermann’s Grandsige or in Quine's Methods of Logic) which seoms 10 be incorrect or needesly involved, then, by con structing a suitable tableau, we obtain a correct ora simplied derivation, Drovided ruch a derivation can be found, In particular, an indirect proofu E,W. BETH willbe replaced by a direct one, whenever such a proof is avaliable this is tn interesting result in connection with a paper on ths subject by Liwen- beim? 1, Infinite Tableau. shall now show that in certain cases the construc tion of a semantic tableau involves infinitely many steps, and discuss the ‘posibility of pointing out the existence of a suitable counter-example if uch a situation arses. (0) The frst and rather obvious, case is that ofa sequent: Up, Upson aint FY, Which involves infritely many premisses Uj, Us In this ease, its con- Yenint slighty to change the division ofthe construction into sucessve age, o5 follows. We begin the fst stage by taking into account the initial formulas U and V and by introducing one individual . We pass on t & ‘ew stage either by takinginto count anew formula U, or by introducing 4 new individual p; if the A'™ stage stated with the introdvetion of anew {ndividualp then the (k= 1) stage muse star with taking into acount a ‘ew inital formula U,; and if the k stage started with taking into account ‘ew intial formala U, then the (+1) stage will mormally start with the introduction of « new individual p; however, it may happen that atthe end ofthe stage we have no reason to introduce a new individual, andin that fase the (E-1 stage wil tart with taking into account still another new {nial formula U,,,-—Itwil be clear that th above sequent willbe provable in F, if and only if for some the sequent Ui, Uren UG EY is provable in F. Accordingly, the conclusion 17s derivable in F from the premisses Uy, Uz, df if and only if for some jit is derivable from the premisses Uy, Uz. Uy (2) Even one single inital formula U may give rise to an infinite tableau, provided it contains some binary predicate (or relation). Let us consier as fan example the sequent: ure, “where Visa certain formal stdied by K. Schutt (1934): 2) ERE) CE) (REx) (2) IRE) > REED 2, ymin On aig ne Ph Data Ql spa Ma a is sor fetes wo eater woth by Boe and Sr hectis Sebts nt by encod arta, Mah ese Sons Gat ¥ SEMANTIC ENTAILMENT alia (© @) RED &ENREy) &() Fist sage » QDR RED GH. NSEILD 2 sage OMIT ORD © Gite ke.a1 8 RUD = Aa) 1 a) RUD (2 ROD RO2 T 4) RQ,2) 19 RED.) 06) R22, (18) (EYRE REDD ‘Third stage (19) RO,3) RENE} 20) 20,3) QD G)IRG.2)—- RE.) (2) R223) 2H RU,3) 25) RG,2) R23) (28) RG,2)>RG,3) GD RG.) RG.2 £0) of) RU R238) RED RE) (..) (8 stage Rk e+) BENE RuE+D, RU, K) = RO, +, Rk, R18) nei RG k) > RO ks Re b> REL kD) REREAD oh) R11) Rt 1) FETED REN.) 2) stage ROH KS) &EVL I Trwalid a7 R@,.2) 9) RG,2) 6 RGD 25) RG, Rb kD. Rk Des RU KD), RUB)6 BW, BETH “The corresponding semantic tableau is (schematically) represented on 1. 25 It is convenient, in such a cas, to denote the individuals by the ‘umerals 2, 3. [wih lay some stress onthe following points, () There is no counter-example with a fine universe. For in te first, ‘plac the universe must not be empy. So we tke some individual and give it the name 1; then we have R(T, 1) and, furthermore, there must be some individual which falls the conditions: RUJ,») and (2)(R(,1) > RG) a8 1 does not full the first condition, the individual we need now must be Giffereat from 1. If its piven the ame’ then we clearly have R(,2) and ()RG.1) > Rl,2) in adlition, we have RCD, and there must be some individual which ffs te conditions RO,y) and ()1R,2)—» Ry); we find that this individual must be diferent from I and 2 and we give it the name's" Ee Gi) In accordance with the discussion under (), the construction of our tableaw tums out tobe fatally determined by the data of the problem. On ‘account of rule (vib) a section 6, the tableau has to be split up again and fzain, but most ofthe subtableaux ths obtained are quickly closed whereas those which are not yield exactly the same truth values forthe atomic Formulas R(1), RCs2), ROB), on RON, RQ,2), oy RGD oe His ‘easy to ste thatthe tableau indeed provides us witha counter-example, ‘whichcan be simply desribed a follows: the universe consists of all natural fhumbers I 2 3, and forthe binary predicate R we take the relation Smaller han Between natural numbers. ‘Gi) One might suspect that the emergence of infinite tableaux (and, hence, of eountr-examples involving finitely many individual) points to some deficiency in our approach. However, sucha view would be mistaken. ‘We wish to establish logical theory which is adapted to such situations as may present themselves in sietic argument. For instance, it ought to provide the framework fra discussion on the (unsolved) problem whether ‘not Fermat's Last Theorem is drivable from the axioms of arithmetic. ‘The corresponding semantic tableau would look as follows Inealid Fermat's Last Theorem Now if counter-example isto satiny the conditions stated in the left, column, it must certainly favole infinitely many individuals So iin some SEMANTIC ENTAILMENT a manner we exclude infinite tableau, we cannot hope ever to tablish an appropriate framework for dealing with problems of the above kind {@) Te wil be clear that, generally speaking, with respect to the problem whether or not a czrain conclusion V LOGICALLY FOLLOWS from given promises Uy, Us. (the general problem whether or nota certain sequent OLDS TRUE is treated in the same manner), we have 10 anticipate three Fpossbiies, “@) The tableau provides us with a suitable ite counter-example for © proving that Vis nor semantically entailed by U, Usy- this ase, Vis ‘ot derivable in F from Uj, Us (@) The construction ofthe tableau breaks down: inthis case Vis both semantically entailed by, and derivable in F from Ujy Us (i) The construction ofthe tableau involves infinitely many steps; in his case the tableau isnot close, hence Vis not derivable in Ffrom Uj, Us, ‘In cases () and (i), the completeness theorem for the system F clearly holds tue. Ie will also old true incase (i) ifthe construction ofthe tableau provides us with asitabe counter-example. ‘In the particular situation which has just been examined, we were indeed able to point out the existence ofa suitable counter-example; however, this situation was exceptionally simple: the counter-example corresponded toa familiar mathematical structure, and soit was recognized rater than dis: covered. In general we must be prepared to mest with chaotic succession of, tableaux being split up and of subtableaux being closed. We cannot expect, that we will always be able to discover some regularity in the process and then to “ead” some counter-example from the tableau "Now inorder to have at least a guide in this labyrinth, we shall represent ‘our tablea by a certain configuration whch still more concise and there- fore gives an even clearer insight into the progres of our systematic attempt at constructing a suitable counter-example. This configuration consists of points and line segments, and is called a (binary) fre. For each formal which weinsertin our tableau, one points added tothe corresponding tee. The points corresponding tothe formulas in one and the same subuableau ar stringwise connected in accordance with the order ‘in which the formulas appear inthe tableau, the distinction between the two column a subtableau being discegarded, Ifthe tableau (or subtableu) is split up, thea the tee contains a fork; this shows the relations of subordi- nation which may exist between various tableaux. By way oflustration, we have constructed in fig 2 part of the tee which ‘corresponds tothe tableau on p. 25. To save space, we have omited the part (a, (0), and (2) which are exatly lke (02s BW. BETH willbe clear that the tre canbe considered as composed of branches, which star at the Yop” and stretch downward as far as possible. Such & branch corresponds toa Sequence of ‘ested’ subtableau, each subableau Jn the sequence being subordinate to all preceding ones. This sequence ‘obviously must belong to one of three ' types. (@) Tt may terminate on account ofthe fact that all possibilities of applying rules (oi) In Section 6 have been ex haustd, (i) Ht may break off on account of the closure ofthe last subtableau which iteconiins. {i) Tt goes on indefinitely; twill be ear that this situation will presen itelf ‘and only ifthe corresponding branch in the ree goes on indefinitely. ‘Type (i) i of no importance for us, ast does not provide s with a suitable ‘counter-example; however, both type () tnd type (i) yield suitable counter cexamples and hence will be discussed. ‘Actually, we shall only deal with type (ip, as itis very easy to adapt the argue seat to ype Let us suppose that Ui, Us, and ¥ ‘have been the initial formulas in the tableau and that they contain a property letter‘ anda relation leer Re the ‘universe consist of the numbers 1, 2, 3, ‘We consider the atomic formulas [A(p) 408 RR ET, 2,35 Wich fnpear in the tableau and subtableaux belonging to our sequence. AS none of these tables is closed, no formula can ‘appear both left and rig. Th order to obtain a counter-example, ‘we mow must select predicates for the leters'A"and (Section 1). We take ‘the property A and the relation R which ‘can be defined a follows SEMANTIC ENTAILMENT » (A) We sy that the natural number & has the property A, if and only if ‘the atomic formula 4(k) appearsin a left column (or subcolumn belonging to the above sequence (R) Wessay that the natural number jis in te relation Ro the natural ‘number k, if and ony ifthe atomic formula R(J,K) appears aleftcokuma {or subcoluma) belonging othe above sequence. Finally, we have to show thatthe counter-example thus obtained is a suitable counter-example for proving thatthe formula Vis not semantically entailed by U; and Uz. This means that we have to point out tha, for the fbove property A and relation R and with respect to the universe which consists ofall natural numbers, the formulas U; and Us are valid, whereas the formula Vis invalid. We do so by showing that all forms in a left column (f our sequence re valid and that al formulas in right column fe invalid; let us observe that, forall atomic formulas, this assertion is justified on acount of definitions (A) and (R). We now consider non-atomic formulas; it wl be sufcient to discuss few particular cases, in connection withthe rules fo the construction of semantic tableau. (Gi) Suppose that (2) (2) appears in a left column, but i invalid if (4).X()is invalid then, for some natural number k, X(8) must be invalids ‘on the other hand, al formulas X(K) appear ina left column. I follows that some formula X(K) appears in a let column, but is invali (Gia) Suppose that XV Y appear ina left column, buts invalid in all, continuations ofthis left column, either X or Y must appear; on the other hhand, if XV is invalid, then both ¥ and. ¥ must be Invalid It fllows that cher ¥ or ¥ appears ina left column, but is invalid. ‘On account of this discussion, ny discrepancy between the tableau and the counter-example is transferred to ever shorter formulas, and inal to the atomic formulas. But we know already that, fr atomic formula, there «am be no such discrepancy. So we have proved our assertion; itfllows that twe have indeed obtained a suitable counter-example (in fact, the above discussion merely shows thatthe rues ()4vi) in Setion 6 were stably chosen). So, even the construction of semantic tableau involves infinitely many steps, it will nevertheless enable ws to point out suitable counterexample, provided the corresponding tre contains a branch which goes on indef- Bitely. And hence the proof of the completeness theorem for our System F ‘can be concluded by showing that every tee which contains infinitely many Points has a branch ofthis kind. This can indeed be shown, but it demands A digression on the general theory of trees. 8. A Theorem on Tres. The configurations, which ae known 35 "ee, ‘an be generally characterized follows. There is one point0 BW. BETH ‘origin’, to which we assign the rank 1. Furthermore, there are a finite numberof point of rank 2 finite numberof points of rank 3,8 fini ‘umber of points of rank k, and 20 on; we neither exclude nor require that, ‘Starting witha certain rank , the numberof points iO; itis convenient intoduce, in addition, a zero-re, which does not contain any point. "A point of rank k can be an endpoint; otherwise, it is connected with a least one pint ofranik + I-A pointofrankk + Tis connected withexactly fone point of rank k, Two points whose ranks are not consecutive can never be connected, "A sequence of points of ranks 1,2 3, such that any two points of consecutive ranks are connected snd which, without violating this con- mak Fie} ition, cannot be extended, is called a Branch of the tee; such a branch Imus sart atthe oni, and it must either break off tan endpoint or goon indefaitely. Each point P ofa tee B clearly isthe origin ofa subiree BIP of B. If ‘BY contains only Snitely many points, then P is clled a point of the frat kind; otherwise, Pi sai to bea point ofthe second kind. Let B® be the configuration which remainsif, fom tree B, we cancel all points of the fst kin. “) Let P and Q be connected points of ranks k and k + 1 ina tree B; ‘then we have: @) if Piso the fist kind, then Q i also ofthe fist kinds and {Dif Q iso the second kind, then Ps also of the second kind {@) Lat P be of rank kin a tee B; then we have: () if Ps ofthe second kind, then its coonectd with at least oe point Q of rank k +1 and ofthe second kind; and (i fall points Q of rank k +1 and connected with Pare Of the first kind, then Pis also ofthe ist kind. SEMANTIC ENTAILMENT a (Tre cnfewation Bs abo ae, Foor By heaven (0 sb any pot @ of ak k +1 in BY econ sexed whet es on ot Penk in (6) No pit in 5 cab eno. Tl fallows rom 8 (5 Fora we tere onto potion namely) coins «only ty many pn a ey franc in Breaks of ad) 8 Contains ity any pin Sd henna east one trench ch foo on inde, Fie.4 roo. Lets consder te tee BY Now ther ae oly vo posi namely: (0 B® isthe zeroes and) B" contin teat oe poi 140) follows thatthe erin tf Bea po ofthe fit kind so BO = B ont onyEiely many point Bence very ranch nes {24i) 1 B* contine npn then by theorem, stmt contain branch ‘ich gr on nd. On, ths ranch sa ape in hence 8 must contain iniely many pot - "ew te ear tat theorem () ile oe ast gap in ou poo of the £ complcenes theorem for be Sy F |" Notwithstanding the eomentary sharacer ofthe above roo ofthe ice | theorem canal be ase rm a none plat of vio, at deans an ape othe pipe of he exloed ti. ee at an inuloicaly vad proof an be bigs rom Broun prot of he socalled Fundamental Theor on Flory Spreat, ba hs rues a 5 fecal intrpretion of the theorem by which ts apatin is seveey | Rte. In fat, my attention was drawn othe te there and tthe2 5 W. BETH problems connected with is proof by a discussion with L. Henkin, A. Heyting, and J. de Tongh during seminar on intuitionistic mathematics, eld iat November 1954: this discussion was followed by @ conversation ‘wth P. Eds and L, Henkin on related subjects. "Besides mathematics and botany, many other scenoes deal from time to ‘ene with res In traditional logic, we ind the arbor Porphyriana, whichis tuo Known asthe“ Ramean tree. According to Jevons("“Ieremy Bentham peaks (uly of "the matchless beauty of the Ramean tree’ In recent Philosophical literature, the configuration appears, for instance, in con ection with the interpretation of Plato's theory of ideal numbers, 28 ‘efended by J. Stenel and O. Becker, and in discussions on the intuitionistic theory of continuum." 9, Range of Applicaton In order to indicate the domain of aplication of te above methods, I shall use te following (summary) division ofthe ‘Somain of modern logic. Lets first make a horizontal distinction beween {) classical (or tvo-valved) logic, (I) intuitionistic and modal loi and {Ah the systems of many-raued logic. Within each ofthese domains, there {Sa vertical gradation into (a) the sentenil calculus, (elementary lic, lind (logic of higher order. Classical logic isthe system which s curently appli in sientiie argument, and hence it has been studied rather horoughly. It dsplas the fll hierarchy (a)-(0. The systems under (1) are fo sometimes applied, on account of certain objections fo classical lei. ‘The systems under (ID are experimental constructions, they have hardly ‘ver been applied. For the systems under (H) and II} we have as yet only (the sentential calculus, and () elementary log "The above treatment covered only the classical vententsl calculus and assical elementary log Ttean, however, be extended tothe classical logic “higher order and so virtuallycovers the entire domain of clasical ogc. "* ‘Ofcourse, it remains to be seen, wheter a similar treatments possible for the systems under (TD and (D- 10, Logical Machines. The sttongly mechanical character of the pro- cedures which we have described suggests the possiblity of constructing & “Torcal machine’, which would automaticaly solve any lgial problems of the types which have been discussed. The outer part of such a machine ‘Mould include: a dial for feeding’ the inital formulas to the machine, and 10, say Svat Th Price of Sec int, Landon, 13)». ; "ar pa ram, AH. We above Bd) ls SEMANTIC ENTAILMENT o repel, aa ren, nef which woul star barsig machine had turned during a certain time. ioral (@) The reds s mento annonce tht te emantc tbe forthe robles coved then te machine produc sa paper showing ‘corresponding formal derivation; Heeeeeecarae () The yellow lights meant to announce that the une thatthe construction of the tableau is terminated under rule (0); then the machine produces a stip of paper showing a suitable nite counterexample; (@) The green light is meant to announce thatthe construction of the tableau involves infinitely many steps; then the machine prodvcesasrp of per showing some pont of B, "inr prt th mchne wou em ie hie Would hve perm he lowing (A) To consti hema the machine; He {@) Toconirat he corespodig te (© Toco ee mrt ponte o const a mache whieh pro ‘the task under (C). For it will be clear that it would have heats ‘Be pottvie,org the ontion ofthe bea an ofS ogee Now ef te prt of Bwhch connie he pom oan ka toe oy beth pat of Bier knowe asso an Bsa Tei osome other tee CwstaneC = Baton tat Do Cy satiny ons that Can be conn in uch aan hat aco Sepang point wnkt= Thea C*s esos, nto abt rant Ray Hence snout const on bebe peae woecontcion st Tealows that ost const gl machin fhe above ind th coluion apes nih 2 fnew and comple eee) theorem ich vas pueda IN6ty A Cth Hower poe to comtt opal mache with fe ads yl gc and sash Iachie woud already be exten bp Friant wold crab Ute chk an gen dra, ad ml ie pone “he aboveccumef heconcpttn opal mtn so afi o serine th pro ch hs bsn en fete hae, Ti brot presen spun noon lor we tec et Wei-efoed en ths eas tht we sume te posi of oun Brotony pitt to ana, Hee toni, ‘uch ot tons cnet asp te ove oo On eee Ra, ist poste to prove the eonpicenen hear o te bss weet ‘sumo for conven, te conpcencs theorem Innes te ts til formals ‘ed! toPy fw. BETH theorem. It is possible, however, to give an intuitioniscally valid proof of fa weeker version ofthe completeness theorem (J. Herbrand, 1990)? "The logical machine with red and a yellow light displays a striking analogy tothe reasoning power ofthe human intellect and infact, itis able to perform a considerable numberof its operations and even to perform them witha higher degre of eciency and of accuracy. However, the human Fnttlociseauipped with aditional operations which go beyond the power fof such a logical machine. For instance, its not dificult to construct the toe BY if isthe ree of fig 2; S01 seems that our intellect isequipped wth 1 Find of green light, Nevertheless, we may not asume that our intellect ‘operates this green ight ina systematicand hence perfectly reliable manner; for this would imply thatthe human intellect incorporated, sto speak, 2 Jogical machine with a red, a yellow, and a green light, and we know that such a machine cannot exist 11, Historical Remarks—Tradional Logie and Symbolic Loge. ‘The above results may have given some impression of the progress made by Symbolic loge since Land's communication on this subject. At the seme time, the gap between symbolic and traditional logic has considerably nar- rowed: this i not due to the influence of regressive tendencies in con- femporary loge, but rather tothe development ofthe systems of natural ‘Seduction, which ave enabled symbolic logic to incorporate certain notions ‘which are alteady found in traditional loge but which remained foreign to ‘Symbolic logic in its ealer stages. In fact, none ofthe procedures which fhe been applied in the present communication: the introduction of penal symbols, and the construction of counterexamples, of semantic tableau, of formal derivations, and of tres, were entirely unknowa in 1211 sy be etl to ve a brie atateint concesig the fet of he adore cacti of our tf te completeness tne whit, 09 account of the I rr rong is here the aa of meaker ssp Pp root a Sram doesnot fet th al rot of he converse theorem, acoding 10. neh treat dacs saps tee etaent Section 2, We 2a ‘eu toon ar obvious the coment of ou eaten "Graf ee tb ols ofthe completes bere, tt doc no omg 0 ei ieee conglcnes tore to be tu, But wot prowl a sot ly Beet om ace tsampone se mre aly made athe above expo ‘Septbcythe cmitenen hoe other mnt be 8 ale eon rae fo feng pte here i vai (ii interpretation ss te {Eile hom he "norma eo” compen areas ppv Yo Fe 85 JPEN. Lanth"On s Ssppose Improvemect in Formal Look’ Vor. & Med 2e ® iT" Chnk, Convention Lope and Modes Lari (Woodstock, Ma, 195 : SEMANTIC ENTAILMENT 3s teaditionallogc; but novadayssuch devies are more systematically applied ‘and more thoroughly analysed, This factor, besides other ones (in particular, the interaction of logical research with investigations ito the foundations ‘of mathematics), explains the superiority of modern symbolic logic over {raion oi which appears, for nn, rom the following covert @ Symbolic logic provides a more detailed analysis for forms of aegu- Gi Mt shows that certain problems which were vividly discussed in tra- ‘tional logic derive from inaccurate notions and that, if these notions are corrected, these problems simply vanish, (i). Certain other problems, rightly asked by traditional lope but not solved in a satisfactory manner are solved by modern logic. (Gv) Symbolic logic has discovered certain fundamental problems which traditional logic overioaked on account oft failure to make a clea di tinction between the two notions of LoCICAL CONSEQUENCE as discussed i Section 1. ‘ad (i) AS an example T mention the above analysis of the syllogism in FisTNo, Wehave dissolved this type of argument nto nine successive steps. ‘As all formulas involved were dissolved into atomic formulas, ts prety ‘lear that a more detailed analysis i impossible, Now none ofthe steps (oF rules of inference) which are found in our analysis were unknown in tr tional logic; furthermore the genera idea which underlies our analysis fully agrees with the principe of Aristotle's method of proof by éxdeoss ‘Nevertheless, traditional logic never succeeded in ving’ « complete analysis ofthe diferent mod ofthe yogis, ‘ad () A problem which is frequently discussed in traditional logic is con- czmed with the possibility of reducing the so-aled hypothetical syllogia {othecategoricalsllogism.'* Now tis problem cannot arise the situation is correctly understood. Both te categorical andthe hypothetical syllogism ‘an be dissolved ino (and hence reduced to) a certain numberof rues of inference of more elementary character. The posit of such an analysis, was already clearly seen by Aristotle himself. The rules of inference which play roein the analysis ofthe hypothetical syllogism belong encusively 0 ‘the domain of the sentenial caleulus;!” an analysis of the categorical "pals ra 3 et eo ‘modus socanbo, which is closely related t0 the modus FesriNo. a ra Wi a suites osha aumento ete Sb Sire ean inneeeieeae sae Logic (Berkeley-Los Angses, 1983)36 5 Ww. BETH syllogism involves n adtion, certain rules of inference which belont to the theory of quantifiers. Hence the hypothetical syllogism is certainly of more elementary character, bat of couse it doesnot follow that it can be echoed to the categorical logis. (ad (ij) Let us consier a situation which frequently arse in mathematical ‘rgumenttion. Suppose, for instance, that we wish to establish a certain theorem on chordal quadrangles. Then we might stat as fllows: Let I'be 1 piven cil, and let PORS be a given quadrangle inscribed in it; then, tiding these supposition tothe ‘given premises’ (namely, the axioms of fometty), we might find that the quadrangle PORS has some peculiar (property E. And then we would argu: the special choice of the circle I’ does fot matter a all; the only pot i that forthe quadrangle PORS, sich a Grete T'can be found. Accordingly, we would state our concitsion 28 follows: Any quadrangle PORS, for which there isa circle J passing ‘through the four vertices P, Q,R, and S, bas the property vArstotle'* was presumably the fist to discuss this type of argument: “Therefore it has also been said that it is necessary to suppose some- thing fase just as even the geometers suppose to be afoot long what {nota foot ong. But this cannot be the case. For the geometers do not Suppose anything whichis false (for in the formal argument such & remiss does not cut) “The above example gives an illustration of what is meant. In a certain part of the proof it may appear asi the quadrangle PORS ought to be [scribed in the particular circle I, and this ofcourse isnot cores. So one right suspect thatthe proof of the theorem involves an appeal 1 a false ssumption. A similar situation arises inthe formal derivation which was ‘Sscuned in Section 4, after hypothesis 1 has been introduced. In al these tases, however, the (possibly false) assumption, which ata certain moment tna ben introduced scliminsted later on;s0it doesnot count asa premiss, and this is exacly what Aristotle observes. However, if we wish exactly t0 Know what is ging on, then we ought to consult the semantic tableau, In the formal derivation of Section 4, we know by premiss (2), that some individual fue the condition S()) & M(), and we agree o give this individual the name‘a’. Likewise, ifor PORS there i circle i which itis inscribed, thea we may agree t0 ve tis circle the name °I"; and, in ‘Arboles example if we are given a certain line segment, why should not tre take it ae a unity of length? 8 auto, Metophytes 2 108% 21-25 thi text connected wth Metpher AOR Bae ety, wih Pairir Sais A, 730 ¥ ee SEMANTIC ENTAILMENT ” CConsiderabe attention hasbeen given to this matter by Locke, Berkeley, Hume, and Kant” The views defended by Locke, Berkley, and Hume are reasonable enough, even though, naturally, a detailed logical analysis ofthe situation was beyond ther power, According to Kant te introduction of ‘an individual (Tor a) is entailed by the necessity of connecting the formal argument with the construction of such an individual hence this type of argument involvesanappealto spatial intuition, i ischaractersticof mathe ‘matics and cannot be applied in other Hels. Curiously enough, & sini remarkis made by Aleander in connection with Aristotle'smethod of root by éndcous: he asserts that this method involves an appeal to perceptual evidence?” Such conceptions, however, are clearly inconsistent withthe Insight which we easily derive from an inspection of the semantic tableau. "Another problem which ha been discussed in traditional logic but which ‘only symbolic logic hasbeen abet reat nan efficient manner isconcerned ‘with the analysis of arguments which involve relations. We have seen in Section 7, thatthe analysis of arguments of this kd leas o considerable complictions. Strangely enough, itis sill maintained by some authors that itis easy to reduce such arguments to categorical syllogism ‘ad (iv) The two nations of Locrcat coNstQUENCE were (implicitly) used Long before they were clearly distinguished and correctly defined. In logical theory, people always tended to use the notion of formal deivability- The ‘method of the counter-example and hence the notion ofsemanticentalnent ‘was sed in philosophical discussion” The development of symbolic logic resulted, of couse inthe fist place ina clarification of the notion of formal B). (i) If Aisa wif and x an individual variable then (x) isa wf. Method for forming wisiscaled quantification with respect othe variable x. Any fecurrence of the variable x in the formula (s)A is called Bound. Any ‘ocurrence of «symbol which is not a bound occurence ofan individual ‘variable according to this rules called fre. Inauiton to formal manipulation of the formulas ofthis system we stall be concerned with their meaning according tothe following interpretation. ‘The propositional constants are to denote one of the truth values, Tor F, the symbol" denoting F, and the propositional variables are to have the set of these truth values as thee range. Let so arbitrary se, I, be specified ‘domain of individuals, and let each individual constant denote a par~ ‘ticular element of this domain while che individual variables have a8 theit ‘ange, The funcional constants (variable) of degree nar to denote (range ‘ver) subsets of the set of all ordered mtuples of Gtx, %) st have ‘the value T of F according as the ntuple xy.) of individuals is or is rot inthe st (4 > B) isto havethe value Fit is T and Bis otherwise ‘Tyand (x) is to bave the value T justin case has the value T for every clement xin 14 Aisa wi, Fa domain, and if there is some asignment of denotations to the constants of 4 and of values of the appropriate kind tothe variables tith free occurrences in 4, such that fortis assignment takes onthe value TT according to the above interpretation, we say that is savsfabe with respect fo. every such assignment yes the value T for A we sy that A Istelid with respect to I. Ais aid iis valid with respect to every domain, ‘We shall give ase of axioms and formal rules of inference adequate to permit formal proofs of every valid formula ‘Before giving the axioms, however, we describe certain rules of “4A more prc tact acount of tht det can be formated alone ties of ral aed Tash ‘ber Waster en oman Speech Su Phbnphce: Vat (1930), ps 24l-408) Bu tas seraata verwon wil {ce for our purpose“ LEON HENKIN abbreviation which we use to simplify the appearance of wis and formula Schemata. If 4s any wif and x any individual variable we write A for (ADP), (DA for ~@)~A. From the rules of interpretation it is seen that ~A hes the valve Tot F according a8 has the value F of T, while 3x)A denotes T justin case ‘thee is some individual xi J for which has the value T. Furthermore we may omit outermost parentheses, replace a let parea- ‘thesis by a dot omitting its mate at the same time if its mate comes at the end ofthe formula (except possibly for other right parentheses), and put Sequence of wis separated by occurences of" when association to the lets intended. For example, A>B24C>D>E for (A> B)2(C>D)>B), where 4, B, C, D, may be wis. oF abbreviations of ws IFA, B, Care any ws, the following are called axiom 1 €3.B5¢ 2 42 B>-4>(B20)>-42€ 3 A>fofoa 4 (A> B)> +> 098, where x is any individual variable with no {ree eecurencein 5. ()A> B, where «is any individual variable, y any individual symbol, and B is obtained by substituting y foreach fee occurrence of ¥ in A, Drovided that no free occurrence of x in Ais in a well-formed pat of A of the form (0). ‘There are two formal rules of inference: 1 (Modus Ponen). To infer B fom any pair of formulas 4, A> B. M Generalization) To infer (2) from 4, wheres any individual variable. A finite sequence of wisi called a formal proof from assumptions I where I's set of wis ifevery formula ofthe sequence is either an axiom, ‘an element of FoF elie arses from one or two previous formulas of the Sequence by madusponens or peeralization, except that no variable with 3 free occurrence in some formula of "may be generalized upon. If Ais he lastformulaofsucha sequence we write’ F A Instead of (Fy) + BUL.A) denoting the set formed from I’ by adjoining the wi 4), we shall write 7, Ab BLIET is the empty set we call he Sequence simply & formal proof and ‘rite Fin this ease 4 is called forma theorem, Our object is to show that every valid formula isa formal eorem, and hence that our eytem of| saxioms and rules i complete ¥ pe FIRST-ORDER FUNCTIONAL CALCULUS 45 ‘The following theorems about the fist-orde functional cleus are all cither wellknown and contained in standard works, of else very simply ‘erivable from such results, We shall use them without proof here, refering ‘the reader to Church for fuller account. (The Deduction Theorem). IFT, A t Bthen T'} A> B for any we A, Band any st of wi) 6 FB>2sf2.B>C 1. FBDeC>f>4B>C>f 8 Faxd>f)>.Gn4>F 9. BD />6GxXB2/) IV. If Pisa set of ws no one of which contains fee occurrence of the individual symbol wf isa wif and B i obtained from it by replacing ‘each fee occurrence of by the individual symbol x (none ofthese occur- ences of x being bound in B), then if I 4, also "FB. ‘This completes ou description of the formal sytem; or, more sccurately, ofa class of formal systems, a certain degree of arbitrariness having been Jeft wth espct tothe nature and numberof primitive symbols, Let $) be # particular system determined by some definite choice of primitive symbols. A set of wis of Sy will be called inconsistent if A Ff, ‘otherwise consistent. Aset A of wis of S willbe sid tobe simultaneously satisfable in some domain I of individuals if there is some assignment of eoctations (values) of the appropriate type to the constants (variables) ‘with fee occurences in formulas of, for which each ofthese formulas, ‘has the value T under the interpretation previously described. Tuonsne [fA is @ set of formulas of So in which no member has any occurrence of a free invidual variable and if A is consistent, then As ‘simultaneously satsfabl ina domain of ndaiuals having the same cardinal ‘number asthe se of primitive symbols of Sp. "Weshall cary out the ptoo forthe caze where Shas only a denumerable infinity of symbols, and indicate afterward the simple modifications needed in the peeral case Let (j= 1,23...) be symbols not occurring among the symbols of ‘Sq For each (71, 2,3, ..) let $; be the frstorder functional calculus ‘whose primitive symbols are obtained from those of S,.. by adding the symbols u, (J~ 1,2, 3 --) a5 individual constants. Let 8, be the system ‘whose symbols are those appearing in any one ofthe systems SIs ea5y ‘to 00 that the ws ofS, are denumerable,and we shall suppose that some 5 Alonso Chute, Iradacton to Mathmatal Lope, Part I, Anna of Matbe- sms sod Frinton Unverty Prey 54 LEON HENKIN particular enumeration i fixed on so that we may speak ofthe fist, second, th, formula ofS, inthe standard ordering. ‘We can use this ordering to construct in Sy a maximal consistent set of ‘nfs, T', which contain the given set A. (We use ew” to mean closed wf {2 which contains no free occurence ofan) individual variable.) Ty is ‘narimal consistent in the sense that i Ais any ew of Sy whichis notin Ty, then I, AF fs but not Ty Ff "To construct Ty let Toy be A and let B, be the first (in the standard ordering) cw 4 of S,such that (Ta) is consistent. Form Py, by adding 1B, to Ty, Continue this proces as follows. Assuming that I, and B have ben found let By, be the fist nf 4 (of S) after By such that (Py. A) is consistent; then form Poy; By adding B,,y #0 Py Finally let Ty becom posed of those formulas appearing in any To, (70, 1, Clanly Ty ‘contaas A. Tis consistent, fori Ty +f then the formal proof off from ‘assumptions I, would be a formal proof of from some finite subset of Fy 4s assumptions, and hence for some i (= 1,0, ...) Toy Ff contrary to onstruction of the sets of I, Finally, yi maximal consistent because if Ais owfl of Sy such that (Ty, A) is edosistent then surely (Fay) is ‘consistent foreach f hence will appear in some I, and so in Ty. Having obtained Ip we proceed tothe system S, and form a set of its cfs as follows, Select the first inthe standard ordering) cw of Ty which thas the form (3x) (unabbreviated: (2X) >), and let A” be the result of substituting the symbol uy, of S forall free oocurrences of the variable inthe WHT A. The set (Ta4’) must be a consistent st of ews Of Si. For suppose that Ip, A” Ff Then by II (the Deduction Theorem), Ty A'> f heave by IV, Tp F A> fi by Il Pp (3A); and so by 8 land i, Pe FCa)4>f. But by assumption I, (3x)A. Hence modus ones gives yf contrary tothe construction of I as a consistent st. "We proceed in tur to each ew of I, having the form (3x) 4, and forthe {ofthese we add to Ty the cwff of; obtained by substituting the ‘Constant wy foreach free occurence ofthe variable in the wi. Each of ‘here adjunctions leaves us with a consistent set of ews ofS, by the argu- ‘ment above. Finally, fer all such formulas 4’ have been added, we enlarge the resulting set of formlas to a maximal consistent set of ews ofS, in the same way that wat obtained from A in ,.Thisset of ews wecall “After the set has been formed in the system S, we construt Py in ‘Sia bythe same method used in geting I from Te but using the constants ust) G =I, 23s) in place of i, Finally we let, be the st of ews of ‘Se consisting of all those formulas which are in any I. It is easy to see ‘that I, possesses the following properties (@ Tg isa maximal consistent st of cus of Su FIRST-ORDER FUNCTIONAL CALCULUS a (i) If a formula ofthe form (3x)A is in I, then T, aso contains a formula 4’ obtained from the wil 4 by subsisting some constant wy for ‘each ree occurrence ofthe variable x. (Our entre construction has been forthe purpose of obtaining a set of, formulas with these two properties; they ae the only properties we shall ‘se now in showing thatthe elements of, are simultaneously satisiablein ‘ denumerable domain of individuals. Infact we tke as our domain I simply the st of individual constants of ‘Soy nd we assign to each such constant (considered as ¢ symbol in an interpreted system) itself (considered as an individual) a8 denotation. It remains to assign velues in the form of truthvalves to propositional symbols, and sets of ordered »-tuple of individuals to functional symbols of degree nn sch a way a to lead toa value T foreach ew of I, very propositional symbol, 4, of Sys a cwff ofS; we assign to tthe value Tor F according as I, F 4 or aot. Let G'be any functional symbol ‘of degre n. We assign tot the case of those m-tuples ay, Of ind ‘idual constants such tha FF Oy» ‘This assignment determines a unique truthvalue for each ew of 5. under the fundamental interpretation prescribed for quantification and" (We may note that the symbol ‘/" is assigned F in agreement with that {nterpretation since I's consistent.) We now go on to show the spesccesigreanepemmwamNmr Lesata, For cack ow A of S, the associated value fs T or F according at Tagh Aor not. “The proofs by induction oo the length of 4. We may notice, frst, that itwedonothave [, F Afor some ci 4 ofS, then wedahave I, Ff. For by property @)of I, we would have Py, A Ffand so, °4>/by m1. sre Tncase dian elementary cf the lemma i clearly true from the natureof the assignment Suppose Ais B> C. If hasthevalueT, by induction hypothesis P, # C; thea I, FBC by 1 and L. This agrees with the lemma since B>"C has ‘the vale Tin thiscase. Similarly, iB has the value F wedonothave I, +B by induction hypothesis, Hence I, # B>f, and I’, + B> C by 6 and | Again we have agreement with the lemma since B C has the value T in ‘his caze also, Finally if Band C have the values Tand F respectively, 0 that (induction hypothesis) I, FB while I, FC f, we must show that T., + B> C does not hold (sine B> C has the value Fin this case). But by 7 and two applications of we conclude that I, F B> C> f. Now we tee that if P, FB> C then, +f by I contrary to the fact that Tis fconsisent (property.“ LEON HENKIN Suppose A is (x) B. Ty F(2)B then (by $ and 1), +BY, where Bis ‘oluained by replacing all free occurences of x in B by some (arbitrary) individual constant. That i Gnduction hypothesis), has the value T for ‘every individual x of 1; therefore 4 has the value T and the lemma i ‘tablished in this ese. If, onthe other hand, we do not have T, F(A), then I, (3) B> f whence (by 9, DI F(28)(B>/). Hee, by property iWof Ps, for some individual constant m, we have I’, + B”> f, where 8° Is obtained from B by replacing each fre occurence of x by u,, Hence for this my we cannot have PF.’ else I, Ff by I contrary to the fact that Tis Consistent (property i). That is, by induction hypothesis, has the value F fr at Jest the one individual u, of I and so (2) B has the value F asserted by the lemma for this case. This concludes the inductive proof of the lemma. In particular the formulas of I, all have the value T for our assignment and so are simul- ‘ancously satisiable in the denumerable domain 7 Since the formulas of ‘A are included among those of I, our theorem is proved forthe case of a ‘system Sg whose primitive symbols are denumerable ‘To modify the proot in the cae of an arbitrary sytem Sp it is only necessary to replace the set of symbols wy by Symbols we, Where ranges ‘over the postive integers as before but «anges over a set with the same cardinal umber a5 the set of primitive symbols of 5,; and to fix on some particular well-ordring of the formulas of the new ‘in place of the Standard enumeration employed ebove. (Of course the axiom of choke must be used in this connection ) “The completeness ofthe system 5) is an immediate consequence of our theorem. (ConoaRy 1 If is aval wif of Sp then FA. Fst consider the case where 4 i a cw. Since A is valid A value F for any assignment with respect to any domain; ie, A >/'s not Satisiable. By our theorem iti therefore inconsistent: 4. Hence FA>f> fy land F A by 3 and ‘The case of wif 4 which contain some free oocurrence ofan individual variablemay be reduced tothe case ofthe ew A (Ihe clue of 4) obtained by prefxing to 4” universal quuntiers with respect to cach individual variable with free occurencesin (in the order in which they appeat). For itis clear from the definition of validity that if 4” is valid soi 4. But then 4. From which we may infer 4” by suscesive applications of Sand Conousany 2, Let Sy be a functional calculus of first order and m the cardinal number of the st of its primitive symbol. If A is a set of ets 7 sateen FIRST-ORDER FUNCTIONAL CALCULUS which is simultaneously satighable then tn partiealar A is simultaneously tisfale in some domain of cardinal m, "Thisisan immediate consequence of our theorem and the fact that if is simultaneovslysatisable it must also be consistent (ince ules of inference preserve the property of having the value fo any particular assignment in fry domain, and so could not lead tothe formula f). For the special case ‘where m is Ny this corollary is the well-known Skolem-Lwenkeim rest should be noticed, for this case, thatthe asetion ofa set of ens A can ‘no more compel a domain tobe finite than on-denumerably infinite: there js alvays a denumerably infinite domain available. There are also aways domains of any cardinality greater than Xin which a consistent set is simultaneously saisable, and sometimes finite domains. However, for certain A no finite domain wil do. ‘Along with the (rath functions of propositional calculus and quanti cation with espect to individual variables the fist-ordr functional caleulis sometimes formulated so ast include the notion of equality as between Individuals. Formally this may be accomplished by singling out some fune- tional constant of degree 2 8y Q, abbrevating Q(x») as x= (fr indie vidual symbols x, 9), and adding the axiom schemata EL. x=. EZ, x= 94> B, where Bis obtained from 4 by replacing some fee occurrence of x by a fre occurence of y Fora system Sof this kind our theorem holds if we replace “the same cardinal number &s° by ‘a cardinal number not greater tha’, where the ‘efnition of ‘simultaneously satisiable' must be supplemented by the pro- vision that the symbol ‘= shall denote the relation of equality between individuals, To prove this we notice that a set of ewfls in the system Sj say be regarded asa set of ews (1,E),E:) in the system S,, where Eis the set of closures of axioms El((~ 1,2) Since Ej, E Fx=p> y= and Ej, Ej bx=y>ay~z>x=2 we se that the assignment which gives & value to each formula of 4, Fj, Ems assign some equivalence elation ‘the functional symbol Q. If we take the domain I’ of equivalence clasce ‘determined by this relation over the original domain J of constants, and ‘assign to each individual constant (as denotation) the lass determined by itself, we are led to a new assignment which i easy sen to satisfy A (Gimutaneously) in 8, ‘A set of wis may be thought of 25a set of axioms determining certain, domains as models; namely, domains in which the wis are simultaneously ‘stisable, Fora frstorder calculus containing the notion of equality we ‘+7, sto, “Doers Grndapragi dec Mathena’, Sirfer ito ber Norse Vdshape-talen Or 808 a0 can find axiom ets which restit modesto be finite, unlike the situation for calcul without equality. More spcisclly, given any finite set of finite umber there exist axiom sets whose models are precisely those domains invwhich the numberof individual is one of the elements of the givens (or example, ifthe set of numbers isthe pair (1,3) the single axiom (OKE= 1) V4 NENG!) ~ Cem) A~CE= 2 (=2)A@aterviayvens will sufice, where 4 AB, Av B abbreviate ~(A > ~ B), A> B> B re: Spesively.) However, an axiom Set which has models of arbitrarily large finite cardinality must also poses an infinite model as one sees by con- sidering the formulas Ct Gxihe) -- Oxde~ ly ma) A~ EH =) Since by hypothesis any finite number ofthe C, are simultaneously sti LEON HENKIN Aa Guia fable they are consistent. Hence all the C; af consistent and so simul- ianeously satisiable—which can happen oaly in an infinite domain of individuals “There ate axiom sets with no finite models—namely, the set ofall for. ‘mulas €; defined above. Evey axiom set with an infinite model has models ‘wth arbitrary infinite cardinality. For if, B range over any set whatever the set of all formulas ~(xq ~ xp for distinct, B wil be consistent (since the assumption of an infnte model guacantes consistency for any finite Set ofthese formulas) and 50 can be simultaneously satis. ‘tm simplified form the proof of our theorem and corollary 1 may be caried out forthe propositional elculus. For this stem the symbols yy Sand the construction ofS, may be omitted, an assignment of values being fade direily from I. Wile such a proof of the completenes of the propositional alulus is short compared with other proofs the Iieratre the later are to be prefered since they furnish a constructive method for Finding & formal proof of any given tautology, rather than merely demon- strate its existence.” 1 [abou it des sot inefre wth th sebtnc of enki’ aie i shou be oa te don of forma rot fom tsumptone (pis somewat SEES Hors seu te dle and of method forcorretng ee Richard Miomgor ted Lave Henen “On the Deion 0 Formal Dedbcton”, Jere! of ‘Shooto Lop, Vol 20939), pp. 19. E4 piece eee MR nn COMPLETENESS IN THE THEORY OF TYPES! Leon Henkin? “Tus fst order funcional calelus was proved complete by Gd!" in 1930 Roughly speaking, this proof demonstrates that exch formula ofthe ealeulus is formal theorem which becomes a true sentence under every one of & ‘cetain intended cass of interpretations of the formal system. For the functional ealculus of second order, in which predicate variables may be bound, a very different kind of result is known: no matter what (ocursive) set of axioms are chosen, the system will contain a formula which is valid but not a formal theorem, This follows from results of Gide concerning systems containing a theory of natural numbers, because a finite categorical set of axioms forthe positive integers canbe formulated within a second order calculus to which a functional constant has been added, By a valid formula of the scond order calculus is meant one which ex- press a true proposition whenever the individual variable are interpreted 48 ranging over an (arbitrary) domain of elements while the functional ‘Variables of degree ange over al sets of ordered ntuples of individuals. ‘Under this dfition of validity, we must conclude from Gaels results that the calelus is essentially incomplete ‘Ithappens, however, that there isa wider clas of models which furnish an interpretation forthe symbolism ofthe callus consistent withthe usual fom lof ye in V8 9 HN Cnr 8 lle pt ee ce a emma ai aed Te cone ma ne ‘Doctor of Pauosophy and sxe in Qube 19 The reas were announeed at SEER ae en con men ot "TRE a SOht a? yrs ah ea Ty EN vata dros da ohn Rn, sm upset a mt RL i a St ieee nt See Ra ee we2 LEON HENKIN axioms and formal rules of inference. Roughly, these models consist ofan arbitrary domain of individuals, 25 before, but now an arbitra else of Sets of ordered m-tpes of individuals asthe range for functional variables of degree, If We redefine the notion of vali formula to mean one which expresses a true proposition with respect to every one ofthese models, we an then prove that the usual axiom system forth second order calculus is ‘complete a formulas valid if and ony if itis a formal theorem.* ‘similar result hold forte calcul of higher order. nts paper, we will sive the details fora system of order w embodying a simple theory of inte) types. We shall employ the rater eleznt formulation of Church,” the ets of which are summarized below: Type symbols (0 be used as subscript): 1. ond care type symbols 2 Ia, Bare type symbols so is (26). Primitve symbols (where a may be any type symbol) Improper: A,(,) Hae ‘Welt,formed formulas (fs) and thee ype: 1. A variable or constant alone i wif and has the typeof its subscript. 2, M Agg and By ar ws of type (eB) and fi respectively, then (4p) isa wff of pea 3. If 4, isa wif of te « and ap variable of type B then (ap 4a) is @ ‘wf of type (af. An oceurrenoe of variable isBoundiitis ina wi ofthe form (Aaya); others the occurence i free 2h cy aot ely te en nana arbiray manner ery {onus eo ave an inerpreation, For api, the formal FA) erp ‘eaning that wnat Cesta FU) mea thal inte comes ot ‘hence the tage fr functional vara sacha F shoud be ced ater come entation, Siar tf rer to tof ordre pin soee mde hese retina fant he mui nt) hs ion fhe ‘eth conpounding formes ofthe eur ha soca wih same Sere fon onthe drains of mol wth tesco whch te demas mut fe cael ‘The tauren of somplenens an be gen Peel Hod proved Yor Mode sng ‘hese dovreconsiions B) for (~A)¥ Be) ( for (Moa Ba) Ga)B, for (~(eX~AD) G0gB) f01 (iea(AteBe) Qrenyny £08 OOPS aMUon a) > Sone) A,= BD for (ion 4X8) In writing wis and subscrips, we shall practise the omision of pren- theses and their spplatation by dos on occasion, the principal rules of restoration being Bt thatthe formula shall be well formed; second, that association isto the lft; and thirdly, that dot sto be replaced by let parenthesis having is mate as far tothe right as posible. (For detailed Statement of usage, refer to Church’) “Axioms and Axiom Schemata 1 Gav) 2x2 GVI) 3 Gv 7) > 009 a) £57) 22Gev AIP EVI) $6 Tota Sos > forte 6. (90% ¥fonke) 474 Hojo fou 10. x58 28, %o UM Saaty —FuBXD > Sop =8ap MF fouta>Sealteon fo) Rates of Inference: 1. To replace any part Ay ofa formula by tho result of substituting ay for by throughout Ay, provide that bp is not a re variable of a 8nd ep docs not occur in TL. To replace any part (la, Ag), ofa wi bythe result of substituting £8, fora, throughout 4p, provided that the bound variables of 4p are ‘ine oth rom a, ad ie fre variables of B, TH, To infer A, from B, if B, may be iafered from Ay by a single pplication of Re TV. From Asya 10 infer 4.x Bs if the variable ¥. From ay Band 4, toate By, 10 infer yay dug Brovided tat the variable ag not ot fre in Aa* LEON HENKIN A Gite sequence of wis each of which is an axiom or obtained from preceding ements of the sequence by a single aplication of one ofthe ‘ules I-VI calle a formal proof. IA is an element of some formal prof, ‘wevwrite FA and sty that dis formal theorem. “This complete our description ofthe formal system. In order to discuss the question of ts completeness, we must now give a presse account of the ‘manner in which this formalism is tobe iterpreted. 'By a standard model, we mean fry of domains, ove for each type symbol, as follows Dis an arbitrary se of elements called indus, Dy fs the set consisting of two truth vals, T and F, and Dag isthe set of al functions dened over Dp with values nD. ‘By an assignment with respect 10 & standard model (D,), we mean 2 ‘mapping ¢ ofthe variables ofthe formal system into the domains ofthe ‘model such that fora variable ay of type «a argument the valve $a) OF isan element of D, ‘We shall associate with etch assignment # a mapping V4 of all the formulas of th formal sytem soch that ¥4(4,)isan element of, for cach ‘if Ay of ype a. We shal deine the values ¥4(4,) simultaneously forall 4 by induction on the length ofthe wi 4, (0 Agi avaiable set YA) = #49). Let V4(No) be the function whose vals are sven by the tele ] YYMot8) tl F Fl or Let V4(4oqs) be the function whose value for arguments T, F are the functions given by the tables 1,2 respectively. 2 | Milde XDA) 2x | VylAeesXPHD T Tr T T F T Fl oF 1) be the function which has the value T just for te single ‘argument which isthe function mapping D, int the constant value T Let V4lseoa) be Some fixed function whose valu for any argument fof Dye ‘one ofthe elements of D mapped into T by ff thet is wach an element). (i) IfAy as the form Bap Cp define Vs(BqgCp) 10 be the value of the function V4(Bzg) for the argument V4(Cy>- THEORY OF TYPES ss an bose ec eadainn eam sauces Fo Nosed hr neo waco ocr on neem gaara ae Me eas ae amauemneeeet ToL “T for every assignment ¢ with respect to every standard model (D,).* es mer orotcin aces arama niece Eeruitednots Cmca tetany Cuassoenoatanet Peet eect pacer hone ge cnet Sea ae eer deblae ti acta eaeetuean rane cicememeat aman on artes crete anne nevus cetcbes vinci So cerca rn teeeeenen cei pracesine oan mente tual cleat poe ca ol aleret geass irtaca tn eorssoae anton ferent one eee Leena cetneseramaamam a conte errr nes Se oe em pet teeta oar Hitcemen mame cece casio Ce ee eee ee nea eceaar ares So necuanieemienrteess colt cmtael macs Scencenans cones enema saoneencseeraas Sat cneay co nesa moan rinse arn mee oer ae ere ye Fr teen ant onteper iter Sone cn cctnantees eatiora eee tne Meawataeommevernits tart are er eae a icraeteevarcenomeame es * I aon to the noon of yy, the mappings Yay be wel to define tbe goo othe detain of «wit cain no ee xcerrence of an) vara Beeston yey Sand dae eens ae ee ‘hen te deta of Ze simpy Fs for any We tho ee the nto ot Seba tg tor ery ty tene nyt Pepe kp fo she model (By) I Fis susie with rape to some modal, Westy sy hat ‘irs% LEON HENKIN the axioms 1-6, 10%, 11% a formal proof can be obtained for some wi Ay, wwe ite A FAs 16 FA, for every wt, we say that is inconsistent, otherwise consent. ‘Tueonsn 1, IF ts any consistent ser of ew, there ia general model (in which each domain Dis denumerable with respect to which A stifle ‘We shall make use of the following derived results about the formal cakulus which we quote without proof: Vil. The deduction theorem holds: If A, 4, # By, then Ab A> By ‘where Ais any set of ews, 4, is any cw, and B, i any wl. (A proof is ‘iven in Church.) 12, bAgDe~ Ay >Be 1h bAL> Boe FEA, PCD eM 8G dy re C 1 year? Aaa Cu Anal) ~ An) TlajonyAos F AeaCa® Avalon Ce) FB, 8)> 5, BL FGA AL ‘The first step in our proof of Theorem 1 i to construct a maximal ‘consistent set of ews such that contains 4, where by maxima is met {hati 4, is any owt not in then the enlarged set (74, is inconsistent. Socha set P'may be obtained in many ways. If weenumerateal ofthe cw in some standard order, we may tet them one at aime, adding them to and previously added formulas whenever this does not result in an incon- Sistent set. The union ofthis ineeasing sequence of sets is then easy seen to be maximal consistent, ' THEORY OF TYPES 7 ‘has certain simple properit which we shall se. I 4 is any ows clea that we cannot have both IF and I" F~ 4, for hen by 12and ¥, ‘we would obtain /F Bs for any By contrary tothe consistency of On the ‘other hand, a least one ofthe tls Ay = gmt bein For otherwise, ‘singthemaximal property ofPwewouldhaveT’ 4, } Banal’ Act Dy for any By. By Vil then follows that I? 4, > 2, and + A> Be whence by 13 and V FB contary tothe consistency of Twoowils A, of ype ill beclleequtaletif 7+ Ay ~ By. Using 14, land V, ees Se tat this a genie congruence relatos so that the set ofall ews of type is partitioned into doin equivalent clases (4) 18.) -suchthat(4,)004 [2,]areequalifand only i 4, iseqivalet |" We now dtine by induction on 8 fame of domains (D,), and simul taneously one-one mapping ® of equivalence cases ont the domains © De such hat (4, isin Dy ‘Dye the et of two truth values, T and F, and for any cw 4, of ype 0 4 a((a,pis Tor Faccordiog a A, ox ~Ay isin 7 We must show that Ds ‘function of equivalence classes and doesnot realy depend onthe pats ticular representative 4 chosen. But by 17 andV, wesee thatifF VA. and 2B, 8 equivalent t0 4 ie, IF y= By) then Pt By: and siaaly it Te Asad By is equivalen to Ay then T t= B, by 8. To see that Bis ‘one-obe, we use 19 to show that if O{L4}] and O4(B,) are both T (ie, Tray and PFBQ, then Pb dy By 50 that (Ao (Be Siaely 29 shows that (4,] x (B]in case Lp and BEB at both F ', issimply the et of equivalence lasses [ofall ens of ype «And HUA.D is (4,0 that @ is certainly one ove "Now suppose that, and Ds have been defined, aswel as the value of © forall equivalence clases of formuls of type = and of tpe Pad that ery clement of Dy of Dy isthe value of for some (4.1, or (8) reper tively Define 4p) 10 be the function whooe value, forthe ment ‘®( B,D of Dp is MA,ppD. Ths deinon i jsied by the fat that if Aug tnd By are equivalent to. yp sd By respectively, then sg is atvalent to dag Bg, one ses by 21, To ace that sone, Soe {hat Od Ap) and BA have the sme valve for every ®B4) of Dy Hence OL 4p 8g) ~DCAepBl) and so, by the induction fypotkers {that is one-one for equivalence case of formulas of type a, AopB is 5 eauivalen to Azp8y foreach ew By In particulary, ite ake By be (Crp) ~ agny = Aegis) we sx by 2 tha dug and pave cauvalent so that (4291 (Aagh Theone-onetunstion @ having beet thas completely © efne, oe deine By tobe te st of valves (tay) for all ews dag | Nowlt#beany assignment mapping ach variable, ito some eneat= LEON HENKIN KAD of De where A sa of. Given ay wi Bp fet BS bea of tae rom By ty relacing al ee ocareneesin By of an able By some cts that dy) ~ UA. Lene For very and By we hae V4) ~ OUBfD- oot by induction on the nth of Bp Tibi aac ad Up ibe sent AAD € Dy then by anton wae ene Yo Hy a 1489 489 = iid UBD. ee By Nee Hs No (AD 8, thn by eon Pide whenie by 2 Pt~ Noo dy 9 that (Noy is. That SAUD maps T ito F. Convey, i (44) is F hen by defon PE das at NAD Hs ey MMe mas Fn. Hence V4 Bp ~ Bap i is case ee sey ha ens Bis day H ACCD 6 Te by Seton TF Cy wee by 381 F Ase fF a7 4a 20th AAs AD 1 To muter wheter (Aap Fo F. Sma, wing 28 and 26 we se that DAs Coed neo Fi BUC) Fe and LARD Tot F tapectvey Comparing tis with he dfiton of (Ane We se tha thetema holds inthis cae a0. Suppose Bp yey whence Bi Tuy Ibe Yalu of nF tenon MUA pia Ts then Poe dyg were by 277 FhosCu for crayon Cy soba (4a) mapsevery element of D, to T- On te tier and ae marsesery (CD int then web ig te foricaar case whe Cy 8 Ors) dagta FF Aa) ~ Aone) once y 28 Pr TTgyy ae ThA BUDD maps OAD oT “Th era hols in sas Suppone 2p tery whence Bf is ony Let Ayg bea cw soc that xa) abe sete BUC.D into T 30 tak PP AayCy. Then by 2 Pi igacerte) so that the webs of Alga forthe agumet Sica supped into by theater, Therefore we may ake lan tara) ; iy Spe that By haste form Bp Cy, We assume Gndtionhypo- ans) dat we have say shown OUD) ~ M(B) and ICSD VC). valve of B,D forthe argument (CRD, which $U88,C$D. Bat nc his simply BE, C2. Hence 4B) ~ H/C). ‘ay Soppe tat Bhs the foray, and ou incon hypothe is that Ge) = FCC) fr every assignment # Let (4) be any ee nmap nina asecanim piers THEORY OF TYPES 2 aentof Dy, Then he alc of (Ga, CD forthe argument (4, Disby Geiniion Sq) ‘But by applying I tothe right member of the instance F (Aa, C,)¥ Ay (22,6) of 16, we find F(a, Cy) y= CBwhere i the aslomeat ‘which has sae vlc sf forevery argument except the variable ad Hop MMl4yD. That i. (0a,C,)4,1~ (CE) 20 tha the vale of P44, C.D forthe argument H((4,) fs BCH) —oF ¥4{C,) by ind tion hypothesis. Since for every arguinent $(((Aa, C,)*D and ¥40a, C,) fave te same val they rust sequal “Tis concise the proof of ola, “Theorem I now follows die from the lemma, ta the fest plac, the frame of domains (D,) ia general model singe V4.8 s an lemon of Dp for every wit Bp and ssigment¢ Beaune the cements of any Da ae io ‘one-one coneapondence with equivalence cases of fy each domain is enumerble. Since for every ow 48 ~ du being an abirary asi. ten sinc therfore for every onl 4, of Pwohave MLA.) =, asnce ‘ise ssc of flows that V4) iT fr any clemant Avo Aes ‘Ais saisfiable wi spect othe todal (Dy) ‘Tutones 2 For any wif Ay, we have + Ay and only if Ay ald the general sms. From the definition of validity, we easly se that 4, is valid if and only ithe ew eg). ap Ay i8 Vali, Where yu AF the Vatables with free occurrences in ds; and hence Ais valid i and ony i V4(~(x,) (caa)Ao) is F for every assignment with respect t0 every general modal (D). By Theorem 1, this condition implies thatthe set 4 whose only ele- ‘ment isthe ew ~(r,).. (£a,)4y is inconsistent and hence, in particular, Un) oo» anda Flt) <2” od dee Now applying VII, 30, and 31 (several times), we ee that i Ai valid, then F 4. The converse can be veri directly by checking the Valdty ofthe axioms and noticing thatthe rules of inference operating on valid formulas lead only to valid formulas, ‘Taonta 3. A ser T' of ews i satgfiable with reapect ro some model of ‘demumerable domains Dyifand only if every ite subset Aof is suisiable By Theorem 1, if "isnot satisiable with respect to some model of denumerable domains, then I” is inconsistent so that, in particular, TF Gi) Since the formal proof of (x) contains only finite number ‘of formulas, there must be some finite subset = (dy, Ay) of T such tha jy ny Ay FC) whence by repeated applications of VHF Ay > « sss 2 dy (i) But then by Theorem ?, thew Ay > Ay > Chote is valid so that we must have Some ¥4(4) = F, = I for any 6 with respect to any model; ie, isnot salisSable, Thus, every fit subset@ LEON HENKIN of Pissatisfable, then Pi satisable with respect to a model of denumer- able domains. The converse is immediate. I Fis satsfable, then so are its rite subsets, and hence P's satistable swith respect t some mode! of denumerable domains, This may be taken {sa generalization ofthe Skolent-Lowenheim theorem forthe first order functional calculus a ete ies at a ra mune areuate steamy Sete Sennen iat paragon ranean omens Eee erence ete cc eh ate ee Tem ome psceearaatea pera ee ie ae arent) me eran aah a ace may Sere te emp ere reo a an rt neice Peers ote era ea Re eae rooeneeeteems Se er et na otha Bp cee are an phon oepeeerepi Sa aia oad ecteateres SS eer teatro ‘Constants es) must be replaced bya primitive notion of quantifiers. “Theorem 3 can be applied to throw ight on formalized systems of mamber theory. “The concept of elementary number theory may be introduced into the pre funcional calculus of order w by definition, a form particulary suited to the present formulation being given in Chureh? Under this approach, the haturl numbers ar identified with certain functions. Alternatively we may Choose 1 identify the natural numbers with the individuals making up the domain of type «In such system, its convenient to construct an applied aculas by introducing the constants 0, and S,, and adding the folowing {ormal equivalents of Peano's postulates: PL G)e~ Sux.=0, 00» Su = SH. PS Gada for0, > «EDL foa%s > flSuX1> (8) Socks THEORY OF TYPES a ‘The Peano axioms are generally thought to characterize the number- sequence fully inthe sense that they form a eategorcal axiom set any two ‘mods for which are isomorphic. As Skolem? points Out, however, this ‘concition obtains only if ‘se'—as it appears in the axiom of complete induction (our P3}—is interpreted with ts standard meaning. Since, how. lever, the scope Call sets of inividul’) ofthe quantifier (fa) may very | from one general model to another, it follows that we may expect non. standard models for he Peano anions “This argument may be somewhat ere if we consider in dtl the sual woof ofthe category of Peano’ postulates One easly shows that ay ode fr the snioms must conan sequence ofthe orde-type ofthe tatural nner by considering the ind 0,50, Sc(S,00e- ad UsingPl and P2toshow them dint and without ota cece Then the proof contnses flows ‘Suppose that the domain of individuals contained ements other than those of this sequence (which we may ae well identy with the natural umber themselves). Then conse the clas of individuals consisting of Just th natural numters Sine it contin eo (0. ands closed under the successor faction (S,), we infer fom the axiom of compete indocton (€3) tat it conan al individuals, contrary tothe hypothe that some ‘nds were not number By examining this proof wesc tat wecan conclude only that ia general sods! stsies Pean's anions anda the stme time possess domain of individuals ot isomorphic the nara numbers then the domain De of ses of individuals ato contain het consiting of ust hones ‘hich are number ‘Although Skolm indies thatthe meaning of ‘natural number rela tive tothe variable meaning of ee doesnot give any example of sonstandard number system satsying al of Peano sons In wo tet bapers! however, be proves thats impossible to characterize the nar umber sequence by any denumerble sytem of axioms formulated within the ist order functional eae (to wich may be aed any Seto ane: tooal onstans denoting numeri functions anda), the ini Thor Stole, “Obe eine Grandgefiagen der Mathematt’ keer it a are Vike ad on 609), om in ‘De whi ps every ment of Kino and every oe kena Due 3 Fer ae" Utah ne oldie Carte ‘er Zatlers ees ence Masonic No rei crs eno. 10 (995) py T-ah And Ober de Newtons ‘7 Canes music Wr abrilie wendleh Sace Ronse ae ‘Sine Zaenrarben, Fundaments matemacae, Vol 2) (S30 ge a@ LEON HENKIN ‘variables ranging over the ‘numbers themselves. Skolem makes ingenious _useofa theorem on sequences of functions (which he had previously proved) to construct, fo each set of axioms forthe number sequence (of the type ‘described above) ast of numerical functions which satisfy the axioms, but havea diferent order ype than te natural numbers. Tis result, for axiom. _ysters which do not involve class variables, cannot be regarded as being at all paradorical since the claim had never been made tat such systems were categorical. By appealing to Theorem 3, however, it becomes simple matter {0 construct a model containing @ non-standard number system which will tify all of the Pea postulates a well ab any preassigned set of further fxioms (which may include constants fr special functions as well s con- ‘tants and variables of higher type). We have oly to adjoin anew primitive constant u, and add to the given st of axioms the infinite list of formulas 105 # Sip ty # (SO), = Since any finite subset of the en- larged system of formula is clearysatisable, it follows from theorem that some denumerable mode! stisis the fl set of formulas, and such a model bas the properties sought. By adding a non-dequmerable numberof peimi- tive constants rf together with al formulas of of for €,#f, we may ‘even build models for which the Peano axioms are valid and which contain ‘number system having any given cardinal." ‘These same remarks suffice to show more generally that no mathematical, ‘axiom system can be genuinely categorical (determine its models to within isomorphism) unless it constrains its domain of elements to have some definite finite cardinal number—provided that the logical notions of set and function are axiomatized along with the specific mathematical notions. "The existence of non-standard models satisfying axiomesyatems for num ter theory throws new light on the phenomenon of avinconsstncy, fst investigated by Tarski and GBdel. A formal system is arinconsistent if for ome formula Ay, the Formas AyD Ao( S00) Ao(S(S.0)s > (edo, a all provable. Tarski, and later Gédel, showed th existence ‘of consistent systems which were w-inconisten, We can now se that such ‘stems can and must be interpreted as refering toa non-standard number ‘stem whose clements include the natural numbers asa proper subset {is generally recognized that all theorems of number theory now in the Iiterature can be formalized and proved within the funtional calculus of 12 imlar sea for formulations of athe within the fs ode untonl ala war eased by A Mase, "Unersuggen as dem Cebit e mathe Since Logi Resi mumarign, na Wal (936, pp 33-36. Maer ‘ud of pot beats cranes to the mend wd bate ay nde {o Protea Church or bringing tar Paper tomy tenon (Added Feb 1 0) THEORY OF TYPES « ‘order w with axioms PI-PS added. (Infact, much weaker systems suc.) ‘On the one hand, it follows from Theorem 1 thit thew systems can be reinterpreted as tue asetions about & great variety of numberaysems ‘thet than the natural numbers. On the other hand, i follows from the results of Gédel that there are true theorems about the natural numbers ‘which cannot be proved by extant methods (consistency assumed). "Now Gade’ proof furnishes ertain special formulas which are shown to be tue but unprovable, but tere is no general method inicated for estab lishing tht a given theorem cannot be proved from given axioms. From ‘Theorem 1, we see that such a method is supplied by the procedure of constructing non-standard models for number theory in which ‘set and “Tunetion’ are reinterpreted It therefore, becomes of practical interest to ‘number theorists to study the structure of such models. ‘A detailed investigation ofthese numerical structures is beyond the ope ofthe present paper. Asan example, however, we quote one simple result: Every nonstandard denumerable model forthe Peat axioms has the order type w + (a* + u)y, where 8 the type of the rational,Iv LANGUAGES IN WHICH SELF REFERENCE IS POSSIBLE! Raywonp M. SMULLYAN 1. Introduction. This paper treats of semanticl systems S of suient strength so that for any st W definable in ina sense which wil be made reise), there must exist a sentence X which is true in $ifand only itis fn element of H.? We call such an X a Tarsk sentence for W. It isthe Sentence which (in a purely extensional ens) says of itself that itis in 77 If IVs the set ofall expressions not provable in some syntactical system C, then X isthe Gidel sentence which i tue Gin $) if and only if is not provable (in C). We provide # novel method forthe construction of these Sentences, which yields sentences particularly simple in structure. The ‘method is applicable toa variety of systems, including a form of elementary arithmetic, and some systems of protosyntsk self applied In aplication to the former, we obtain an extremely imple and direct proof of theorem, Iwhich i essentially Tarski theorem that the uth et of elementary aithe Inetc snot arithmetical definable. “The erux of our method i in the use of a certain function, the ‘norm’ function, which replaces the classical use of the digeonal function. To givea 5547. Copii © 1957 From the our of Symbol Li, Vol 2 195), shor ihe poche ve Amen Mathemaal Sooty and Reprinted ty pein of te pols he Amer sty of Calforna at Los Anges, and to Profesor Jobn Kemeny and Dr. Edward J. ano Burcu lle fre vata arto Ya wi fk (By semuntia system 5 we mean ae of expressions tring of sg), togcher swith’, nln 8 of eles elle semen of 5 (Setermane by set of le of eon; oosber ia tbe ff mens eal re sete of $ (Gatrmind by ato aes of ut for 3 STA err msn a nina pen he oy a me can Find Sot cos tus fe teureme for ¥ tobe 2 Tas setece foe Mi ‘sce weak any sence ohc athe bth eu dino fae Se seve! However ihe wb mc awe ocd of 2 Tash seotence or ‘Susie rem we pte cng smc em, the ne mold ‘the former wil be care ou a this ape tnd the liter na forthcoming paper, ‘ptm Proonya Sa Appl ' SELF REFERENCE IN LANGUAGES heuristic des of the norm function, let us define the norm ofan expression (of informal English) as E followed by its own quotation. Now, given set W (of expressions), to construct a sentence X which sys of tel that ie in W, we do $0 88 follow: FW contains the norm of "W contains the norm of This sentence X says ‘that the norm ofthe expression “W contains the norm of"isin W. However, thenorm of thisexpresionisX itself Hence X's trueifand only if X= W ‘This construction is much like one dve to Quine® We cary it out for some formalized languages. In Section 2, whichis esenilly expository, we construct avery precise, though uit rival semantical system 5 which takes quotation and the norm fonction as primitive. The study ofthis system will havea good deal of heuristic value, inasmuch a8 $, despite ts embodies the crucial ideas behind undcidabiity results for deeper systems. We then consider, in Section 3, the general use ofthe ‘norm funtion, and we finaly apply the results in Section 4, toasystem Sy, Which isa formal variant of elementary arithmetic. This variant consists of taking the lower functional calculus with class abstractors, rater than auantifers, as primitive. Ths alteration, though in no way affecting the strength of the system, nevertheless makes possible the particulary simple roof of Tarski or Gels theorem, sine the aithmetization of subst- {ution can thereby be circumvented quite simply. By the norm ofan expression E (of S,) we mean E followed by its own Gidet numeral x, the numeral designating its Godel number). Now, even any set 17 of expressions whose set of Gidel numbers is arithmeticlly ‘definable, we show quite easily the existence of an expression H of lass sbstraction, such that for any expression E, H followed by the Gédet ‘numeral of Bisa true sentence if and only ithe norm of Eis in W. Then, it ‘we follow H by its own Gede! numeral h the resulting sentence Hh which 1s the norm of H) is rue i and only if tis in WV. This i a rough sketch of ‘our procedure. 2. The Preliminary System S and the Semantcal System Sp Ya this section, we formalize the ideas behind the preceding heuristic account of ‘thenorm function. For convenience, we fist construct preliminary system tn contat wih thie contraction, let dine the dagnaation of E a he seul of substan the quotation of forall curt te sera in ‘Then ie lowing fara seme for when formalin) te cal cme: tin’ W contain the agomaaion of conta the Sgoruieaon sf, Tos ines kann ove ain Gon goa) hes the ‘ies Tr eser to ferme Scion 3) se eeeseertee nde ero cn yl ahod when apne« RAYMOND M, SMULLYAN ‘So, whose expressions wl be bul rom the three signs‘ Second sign wil seve as our formal quotation mark, since we reserve ‘ordinary quotation marks for metalinguistic use. The sign “N’ will be endowed withthe same meaning athe norm of’ The sign 4" willbe an tndeined predicate constant. For any property (se) Pof expressions ofS, ‘we then define the semantcal system by giving rule of uth for Sp. For Any Pf wil be interpreted in Spas designating P. Signs of Sy: $y *.. Preliminary Definivions, (1) By an expression (of Sq) we mean any string ‘ull from the three signs ofS. (2) By the formal) quotation of an ex- pression, we mean the expresion surrounded by stars. 3) By the ormof an txpresion, we mean the expression fllowed by its ovn (Formal oration Formation Rules fr (Individual) Designaor: () The quotation of any expresion ia designator. (@) If Bisa designator, sos "NE" (ie, °N' followed by E). Alternative Definition: (yA designator isan expression which is either a quotation (of some other expression) of a quotation preceded by one or more "N's. ues of Designation in Sy RI, The quotation ofan expression E designates E. RQ. IFE, designates E, then “NE, "designates the norm of Ey. Definition of « Sentence ofS (1) A sentence of 5) is an expression consisting of" followed by a designator. The Semaniical System Sy Forany property P, we define the semantical system S, 25 follows: () The rules for designators, designation and sentence formation in Sp are the same a in Sy {@) The rule of trith for Si he following 3. Forany designator, "gE istrucin Sp the expression designated by EGin Sp) has the property P. ‘Thionem 21. There exes an expression of Sy which designates itself Poor. #N" designates 'N’ (by Rule 1). Hence ‘N¢N* designates the norm of ‘N" (by Rule 2) which is "NPN" ‘Thus ‘NEN designates tse i SELF REFERENCE IN LANGUAGES @ ‘Tuomun 2.2. There exists a sentence Gof Sy such shat for any property P, ts tre in Sp = G has the property P. oor. “N*#N* designates SN GN" (by RI and R2). ‘Thus G, viz, GN PN" is our desired sentence. Raaank. Gis, ofcourse, the formalized version of W contains the orm of W contain the norm of "fis but an abbreviation of W contains, and SN’ abbreviates the norm of" (Conotsany 2.3, P cannot be coextensive withthe set ofall fle (nontrae) sentences ofS nor i Peoextensive with he set ofall expressions of Sp which (re not true sentences of Sp. 2.4. Astnonorn ona oF minokin 2.2. By predicate we mean ether‘ (of followed by one or more'N. ‘We say that an expression F sais a predicate H (in S;) if H followed. by the quotation '*E* of, is truc in Sp. Lastly, we say that a set W of expressions of Si definable Gin) i thar exists a predicate H which is ‘atistied by all and only those expressions which are in W. tis worth noting at this point, that if, designates E, then Esatsies H if and only if "HE," is tru. This fllows from R3 by induction on the number of N's occurring in H. For any set W, we et (37) St ofall expressions whose norm iin W. Lenoua 25. IW ie definable in Sp, then 30 4) Proor. Let H be the predicate which defines WC, which is satsfed by Just thos elements which are in W), Then H followed by N' will bests by precisely those elements which ae in (W). Thus y() is definable Gn 5). "We can now state the following theorem, of which Theorem 22 is @ special case ‘Twsomsn 2.6, For any ser definable In Sp there Isa sentence X which i true, and only f X=. 21 we wished wo cons, a minature stem Lr which formalizes the dagosal ‘goctig athe same mayne S doe the or funciny me take four ti, wee tnd the rat By Came as Shy Ht Ey depen, ten BEY en a steel fu oc thdle tien Lek and only Ey asthe propery P Thes the expen of Finer, SWAG a Ba eT te sen 35a RAYMOND M. SMULLYAN Poor, Assume Wis definable, Then sois 1(W/) (by Lemma). Hence there ‘exists a predicate Hsuch that for any expression E, "HE Vis true in ,) = Een) ew. = TEE Taking EH, HH” is true = HOH” e. Thus X, viz, "HSH", i our desired sentence, Rewank, Theorem 2.6 says (in view of the truth functionality of the ‘conditional no more nor les than this: each set definable in Sp either contains some truths of Sor lacks some flschoods. Conottany 27. The se of false sentences of Sp isnt definable in Spy nor Is the complement (relative tothe set of allexpresions ofS) ofthe sof re sentences of Sp definable in Sp Conouiany 28, Suppase we extend Sto she enlarged semantical system ‘Sj by adding the mew sem’, and adn the folowing evo rales: 4, IE X's a sentence, 50 is "=H. RS, [His te in Sj Xis not tue in Sp. ‘Then in ths sytem Sp, the truth set of Sp is not definable Poor. For Sj has the property tha the complement of any set definable {nS is agai definable in S, since if H defines W, then °~H1" defies the complement of W. Hence the truth set fnot definable, ine its complement {isnt definable by Corollary 2.7. Raman Spi about as simple a system as can be constructed which has the interesting property that the ruth set of the system is not definable within the system and that, moreover, any possible extension of Swill retain this feature. By an extension, we mean any system constructed fom ‘S$; by possibly adding ational sign, and rules, bu retaining the old rues in which, however, the word ‘expression’ reinterpreted to mean an ex: pression of the enlarged system. Likewise if we take any extension of Sp, then although we may greatly enlarge the collection of definable sets, none ‘of them can possibly be coextensive withthe set of false sentences ofthe 2.8, EXTENSION OFS, 10 seuanrico-syeTACTICAL svSTEM Sf. Suppose row that we select an afbitrary set of sentences of Sand call them axioms, land select set of rues for inferring Sentences from other sentences (oF finite sets of sentences), The axioms, together withthe rules of inference, orm a so-called syntactical system, of calculus C. Let $ be the ordered SELF REFERENCE IN LANGUAGES 6 pair (Sp,C). Thus $$ isa mathematical sytem, or nterpretedcaleulas, We Tet Tbe the set of tue Sentences ofS (lb called tras sentences of $$) nd Th, these of sentences provable in C (also called provable sentences, 0 theorems, of 59). Wealready know tha the complement T of T (relative to the set af expressions isnot semantically definable in S$ (i.e, not definable in Sp); however Th may wel happen tobe. Its, however, thea we have, san immediate corollary of26, the fellowing miniature version of Gaels theove: ‘Tasonsa29. these Tis semantically definable in S$, then ether some sentence trae in S§ camot be proved in Sor some false sentence can be proved. ‘This situation is sometimes described by saying that Sis either semmanti- cally incomplete oF semantically inconsistent. 2.10 We can easily construct a system Sf obeying the hypothesis of theorem 2.98 fellons: Before we choose a property P, Weis construct & completely arbitrary calculus C. Then we simply define Po be the set ofall expressions not provable in C. Then “4 itself wll be the predicate which semantically defines TH in Sand the sentence G, viz, "GN"IN®, of ‘Theorem 22 willbe our Gide sentence for Sf, which struc if and only it ‘not provable in the system. Infact, for purposes of ilustation, let us considera calculus C with enya finite umber of exioms, and no rules of inference. Thus the theorems of C are the axioms of C. Now, if G was included as one of the axioms, itis automaticaly fase (in ths system), ‘whereasifGwas left out, thenit steve, by very veweof being left out. Thus, tis sytem i, with dramatic larity, obviously inconsistent or incomplete, [Restanx. Suppose that we take P 10 be the set of sentences which are provable in C. Then G becomes the Henkin sentence for the system S$, ‘whic tre in this system, if and only if is provable in $5. Is G tre in ‘87 This obviously depends on C: If, for example, we take C such that its Set of axioms is null, then G is certainly both false and non-provable, An ‘example of a choie of C (ether than an obvious one in which G itself san axiom) for which G is trues the following: We take for our single axiom 14, the expresion *SN*AN**" We take a single rule R: If two desige ‘iors and E, have the same designatum in Sp then "AE i directly HE is true = FCE is tre (by Rule 9) = E satisfies F. Thus the seis respectively defined by H and F are the same, ‘A formula Fy, will be called normalize of forma Fi Fis satis by Just those expressions E whose norm satisfies F. Inthe light ofthe preceding Deragraph, the statement that Sis semantically normal equivalent tothe SMatement tht every formula F (vith oe fre variable) has «normalize Fy (ince Fy defines (1), when F defies 1). ‘Tutonen 4,5, 5 semantically normal, relative to roo. We must show that every Fhas a normalizer Fy. Wel tke Fy to be the result of replacing the free variable x of F by = 10" fr rather, bythe ‘unabbreviated form “(a) (HITT) 5 Had we used gy, rater than # or our Ge corspondece ten, if «were the gy oF Bt bn Ofte norm oC E hou tae bat DIO” RAYMOND M. SMULLYAN ‘Then, for any number x, Fy) ad F(n() have the sme trth-vaues. “Thos, for any expresion E, Essatisfis Fy Fy) is true = Fon) is troe “& thenorm of E satisfies F (since mE) isthe gn. of, norm of E!), Hence Fis satisfied by those E whose nem satisfies F and is ‘hus a normalizes of ConoLLany 4.2 (1) For every definable set ofS there isa Tarski sentence. (2) The complement ofthe truth set of snot definable nS (elatoe 8). (DT tse not definable nS, (elas 0). (@) Any proposed axiomari- ation ofS such thatthe ge fits theorems definable nS (relative fo) 8 semantically incomplete or inconsistent. (and (2) immediately fllow from the preceding theorem, together with the results of Section 3. In particular, in (1), to construct a Tarski sentence for a set W defined by formula F, we st construct the normalize Fy of F by the method ofthe preceding theorem, then take the abstraction H of Fy. and then fllow H by is own Godel number. Thus the Tarski sentence for Wis the norm of the abstraction of the normalize ofthe formula which defines W. 3) and (8) follow, since S, contains negation (definable fom , Resank. (4) of Corollary 4.2 can be thought of as one form of Gaels ‘theorem. Defnability in Si actually equivalent to definabiity in proto “syntax (nthe sense of Quine) Thus any formal system for S whose set of {theorems is protosyntactclly definable will be semantically incomplete or Inconsistent. This is esseatialy similar to Quine's result that protosyntax ives not protosyntacticallycompletable, “43, We have just shown a method for constructing normalizers which works for the particular Godel correspondence g, which we employed ‘Actually, i ill work for any G8del correspondence relative to which the ‘norm function (.e, the function which assigns to each expression its norm) is sre definable, nthe following sense ‘A function F (rom expressions to expressions) will be said tobe srcly , Fy is the Gidel numeral of Ei and only if GCE, Bis wus (which we express by saying that g itself is definable relative to a), then the formula F(ayiGty,). &-C(B.y.4))” defines the norm fonction and S, is normal Hence “Tneones 47, sufciencondiion for St be normal, relatice 1, i that concatenation ang isef bor be definable, relative tog [Restank, Gadel correspondences satisfying the hypothesis of Theorem 47 include all those that are effective (ie, include all those such thatthe unetion h, which asigns to each mumber x the Gédel number of (he numeral designating) x, ia recursive function.” Ths, in conjunction with Drevious results, yields the proposition tha, relative to any effective Godel ‘correspondence g, the truth set of Sis not definable. This, in essence, is Taski's Theorem 5. Concluding Remarks: Diagonlization ws. Normalzaion. We should like, in conclusion, to compare the norm function, used throughout this paper, wth the more familiar diagonal function, used forsystemsinstandard formalization, Firstly, to sketch a general account of diagonalization,’ analogous to Section 3 Yor normalization, wecoasider now an arbitrary language L which (like $ of Section 2) contain expressions, sentences, true sentences, and individual constants. Instead of predicates, however, we now have certain ‘expresionscallodTormlas' and others called variables, andcertain ocur ences of variables in formulas termed “tee occurences, subject to the condition thatthe substitution of individual constants fr all fee occur- rences of variables in a formula always yields a sentence. We again have a Gide correspondence mapping each expression F onto an individual constant. For any formula F with one free variable 2 and any expression Ewe define F(E) asthe result of substituting E forall fee occurrences of 2 in F. Theexpression F(F) is defined 0 be the diagonalization of F. The st swetiy dena, mr ‘must eat a ore Tt i ‘ay or any Magis nena) eves we Pate eh We he ee ednon 2 alm: Aad (Bh 0a uber we ake 4) uber Then we die 8) (3) IA.) & Add. 109] Wh contraction can be simp by intoduing decors). Thus the icy eorspondene s+ |, etd ny for Pres of thy. sae Say SELF REFERENCE IN LANGUAGES 7 ‘ofall E such that FE) is tr, i called the sot defined by F. For any set ‘we define D(W) a the et fall F whose diagonalization sin HW. Then the analogue of Theorem 3.1 18° sufcient condition forthe existence of a Tarski sentence for His that D(W) be definable’ Hence also, DOP) is not definable. We would then define normality for such a language L, by the ‘condition that whenever IY is definable iL, 0 is DCP). Then al other theorems in Section 3 have thee obvious analogues. "Toapply these general notions to systems in standard formalization, €. elementary arithmetic, ve would have, in analogy with the notion “nor- ‘malig’, that of ‘diagonalize’, where a dagonalizer Fp ofa formula F ‘ould be formula satisfied by just those expressions whose diagonalization tstisiedF, Then, i Ws defined by F, and if there exists a diagonalizer Fp for F, then the diagonalization of Fy (wbich is Fy(F,)) is the Tarski sentence for W. "This is essentially the classical construction, The construction of the iagonalizer Fs considerably more involved than the construction ofthe rormalizer Fy. Aga, we might say, this is du tothe fact that concate- nation is easier to arthmetize than substttion, We can profitably ato repetition of analogous arguments for S and by senading tha speach fore eral srctute Ths appronh vl te ‘wnt na oricoming peer “Atseat Sica of Uasiued There In hic we toy, In enuderaie generally, the Spe proper of Undesasbe ‘ema sncoveredby Goel tnd oer,v INFORMAL RIGOUR AND COMPLETENESS PROOFS Gtona Kreiste Iisa commonplace that formal rigour consists in setting out formal rues ‘and checking that a given derivation follows these rules; one of the more Jmportant achievements of mathematica loge is Turing’s analysis of what «formal rleis. Formal rigour doesnt apply tothe discovery or choice of formal rules no of notions; nether of basic notions such as se in so-called Classical mathematic, nor of technical notions such as group oF tensor (product (echncal, because formulated in terms of an alretdy existing basic framework), “The ‘old fashioned? ides is that one obtains rules and defnitions by analysing intuitive notions and puting down their properties. Ths is cet- tainly what mathematicians thought they were doing when defining length for ea of, for that mater, logiians when finding rule of inference or fxioms (properties) of mathematical structures such as the continuum. The teneral idea applis equally tothe so-called realist conception of mathe- ‘atics which supposes that these intuitive notions are related tothe external ‘world more of les asthe number 4 enters into configurations consisting of ‘elements, and to the idealist conception which denies this or, atleast, Considers tis relation as inessenal to. mathematics. What the ‘old fashioned’ ides assumes is quite simply that the intuitive notions are sienficant, be it in the external word or in thought (and a precise forme lation of whats signfcant in a subject is the result, nota starting point of research into that subject). Tigour wants () to make this analysisas precise as posible (vith lable, in particular to eliminate doubtful properties of the intuitive notions when drawing conclusions about them; and (i) to extend this analysis; in particular not to leave undecided questions which can be decided by full use of evident properties ofthese intuitive notions. Blow the principal emphasis ion intitive notions which do net occur in ordinary ‘mathematical practice (co-alled new primitive noions, but lead to new om Problem inthe Phivopty of Market, Ime Lakatos (Norh-Holn ing Copy. re. Reo h oy Seeman ofthe plover and thar {hela 4 peper of aesoy. pp. 38 iy rated teres topes ah some selene fragt hem Ea) RT OE, Seesneeenias INFORMAL RIGOUR » axioms for curent notions. We give three applications, mort following the ‘old fashioned” ide of pushing abit farther than before the analysis of the intuitive notions considered. Section I concerns the diference between familiar independence results, eg. ofthe axiom of paral from the other axioms of geometry, on the one hand and the independence of the con tinuum hypothesis on the other; the diference is formulated in terms of higher-order consequence. Section 2 deals withthe relation between in- twitve logical consequence on the one hand and so-called semantic resp. symtactic consequence on the other. Section 3 [i omited here—E8 Sections 2 and 3 affect completeness questions fr classical and intutinist predicate logic, which accounts for the title ofthis talk; quite generally, ‘roblems of completeness (of rues involve informal rigour, atleast when ‘one is trying to decide completeness with respect to an intuitive notion of| consequence, (©) The case against informal rigour (or: antiphilosophic doctrines). The present conference showed beyond a shadow of doubt that several recent results in logic, particulary the independence results for st theory, have leftlogicians bewildered about wha to do next: in other words, these results 4o not "speak for themselves’ (0 these logician). I believe the reasons underlying their reaction, necessarily also make them suspicious of informal ‘igour. I shall ry to analyse these reasons here (@) Doctrinaire objections (pragmatism, psitcism). Two falar objec. tions to informal rigour are these: (@) Why should one pay s0 much stenton to intuitive notions? What we ‘want are definitions and rules that are full they doa’t have o be faith ‘to notions that we have already, One might perhaps add: these notions are {ormed without highly developed experience; so why should they be exe pected tobe fruitful? ‘Besides this (pragmatic) objection we have a more theoretical (positivist) ‘objection. ‘These intuitive notions in particular the abstract) notions of vat, ‘se, natural number or, so ano! to eave out intitonism, intuitively con ‘incing proof, are illusions. When one examines them one finds that theit solid content fies in what we do, in how we act; and, in mathematic, this is ‘contained inthe formal operations we perform. ‘A certain supericial plausibility cannot be denied to these objections. First, when some abstract intuitive notion turns out 1 be equivalent, at least ina certain context, to a postvistc relation, Le, one definable in particulary restricted terms, this has always important consequences. For instance (ordeal se Section 2) logical consequence applied to frst order» GEORG KREISEL formulae, is equivalent to formal drivability; and fist order axiom systems permit a more general theory than higher order systems. Consequently, st ' particular stage, the (pragmatically) most rewarding work inthe subject ‘nay consist uit simply in exploiting the discovery ofsuch an equivalence, ‘Second, one may be impressed by the slow progres of work on sme of the intuitive notions, particularly thse associated wit traditional philosophic ‘questions: pragmatism discourages such work, and positivism tries to give theoretical reasons forthe slow progress. Now, objectively suc a negative attitude snot supported bythe acs because progress was also slow in cases ‘where decisions were evenally obtained. (About 30 years between Hilber’s fist formulation of his fintist programme, cf), p. 84, and Ges incompleteness theorems; nearly afrther 30 yeas lla precise analysis of finitst proof was attempted.) But, subjectively, if a particular person is discouraged by the slow progress he had surely better find himsef another ‘ocupation- Certainly, sentially speaking on is in a wholly futile posi- tion fone finds oneself tuck both wth philosophy asa profession and with antiphilosophical views such as pragmatism or positivism (perhaps, after Ihaving been attractod by traditional questions in one's youth). For, having repudiated specifcally philosophical otions one i eft with those tht are tbo familiar to specialists in other feds: what jobs can one hope to do as ‘well as these specialist? including the jobs of clarification or explication Gif tey are to be done in current tra)? T thik this futility i elt quite consciously by many of the people involved. Having granted all tht: whats wrong with () and i)? Quite simply this ‘Though the raise perfectly legitimate doubts or possiblities, they just do not respect the fect, atleast he facts of actual nellecrual experience. Tis is particulary ivtating because pragmatism pays so much ip service to ex- perience, and positivism claims tobe empirically minded. ‘Ad\i, Let us even take fr granted that we know roughly what is fitful afer al, here again (ie a8 the case of signitiance) a precise formulation ‘may only be possible after a good deal of experience, Perhaps we do not [know a general reason why intuitive nations should have stood the tests of ‘experience well; a8 one sometimes sas! they might not have done so. But the factremains tha hey or, at east, many of them have. Reflection shows that we certainly couldn't have what we understand by ‘scence if they had Instead of trying {0 find reasons for, of limitations of, this ‘eually remarkable situation, () denies ts existence! (a most unpragmatic pragmatism). Two related %o-alled pragmatic principles are to be men- tioned. One says that one must eat each problem ‘on its merits; one might yhave to; but taken iteraly this would leave litle room for general theory ot oo nate INFORMAL RIGOUR a for the distinction between whats Fundamentl and what i secondary, Or again (in_mathematcs), one sometimes criticizes complacently “old fashioned? disputes on the right definition of measure o the Fight topology, ‘because there are several definitions. The mos striking fact hee how few Sem 0 be useful: these haven't dropped from heaven; they, obviously, ‘were formulated before this applications were made, and they were not generally obtained by rial and error. If hey ad been So obtained, mathe- ‘maticians shouldn't be as contemptuous as they are about the stud of title ‘variants in definitions. Similar remarks apply tothe choice of axioms; but Since this is of direct logical importance the subject will be taken up in its proper place in (8). Ad i). Creal if, consciously or unconsciously, one insists on analysing the’ sold content in positivist, n particular, formal, terms thsis what one will ind. Though more specific points about formalism and formalization are taken up throughout this paper in particular (€) and Section 2 below, ‘some matters of principle are in order hee. It might have turned out that {he notions which present the most Serious dificult in practice are indeed abetract ones. But, quite nalvey, this isnot so: knowing whether two inscriptions mean the same soften no harder than knowing i they read the same! Equally, as was mentioned on p. 79 above, sometimes it does turn fut that some notion are fully represented in formalitic terms: but thishas {o be verfed and Section I shows limitations, Pethaps one should die: ‘inguish between formalism (and positivism), whichis merely a negative antiphilosophical doctrine, and a mechanisfe conception of reasoning (Cmechanism’ in the sense of Turin), which would lead one o expect full, formalistc analysis of actual reasoning Iti to be remarked that, sofa, the ‘most that has ever been shown in support ef this conception is that in curtain areas (@, elementary logic, Section 2) reasoning could be mech- ‘nial inthe sense that a mechanism would get the same esults; not that it that would follow the same route. It may be thatthe mechanistic conception isthe only moderately clear idea of reasoning that we have at present. But a good positivist should not conclude from this that therefore this idea ie correct. () Unrliablty of some inttive notions; the role of formalization in thelr ‘analysis. A much more serious pont than the portmanteau objections () land (i) under (a) concerns specie abstract notions, for intance—to take ‘the mest famous example—the notion of set. Have not the paradoxes shown ‘the comple unveil of our Intutve convictions at least about ths par- ticular notion? First of historically speaking, this couldn't be farther fom the truth!2 GEORG KREISEL Wasn't Cantor @ misunderstood martyr in the face of widespread re- actionry preludice against employing the notion of set (or, as it was then ‘called: clas) in mathematics? If 0, the paradoxes supported the intuitive ‘convictions of those reactionaries. Tt is probably tre to say thatthe reactionary caution was due to this class preseated itself as a vague notion, or, specifically, mixture of nations including () finite sets of individuals (.. objets without members), oi) sets of somethin (a in mathematics, sets of numbers, sets of points), but ‘so (i) properties or intensins where one has no a priori bound on the textension (which are very common in ordinary thought but nt in mathe- Inatics). If we are thinking of sets of something, eg, of objects belonging to fa then the comprehension axiom isto be resticted to read (for any pro- erty P) ax Vy exes [yea &POID: ‘ut if we are thinking of properties, given in intension, whose range of Por + DD) > Wx cone replaces the second-order quantierP by alist of those P which are xplicily defined in ordinary first-order frm (from + and x, for instance) ‘Rmemen’s reflection shows that the evidence ofthe firs-order axiom “theme detves from the second-order schema: the diference is that when ‘one puts down the fis-orde schema one is supposed to have convinced ‘nese that the specie formulae used in particular, the logical operations) re well defined in any structure that one considers this wil be taken up in (below. Warning The choos offrstorder schema isnot uniquely deter- (ploed by the sosond-order axioms! Thus Pean's own axioms mention tapi only the constant O andthe sucessr function S, not addition nor ‘multiplication, The first-order schema bailt up from O and S isa weak, Incidentally decidable, subsystem of classical first-order arithmetic above, land quite inadequate for formulating curtent informal arithmetic. Informal Figour requires a much more detailed justification Fr the choice of + and. tani usualy supplied.) “An interesting example of concept that needs a third-order definitions that of measurable cardinal? Such concepts are rare; for, whenever theorticlly a whole hierarchy resents itself in practice one only uses the first few levels or a notion ouside that hierarchy altogether; two coneets may bementioned here which re not definable by any formulain the whole Perarchy of languages of predate loge. Evident, neither the cumulative type structure itself not the structure consisting of the ordinals with the ordering relation is definable by any formula of finite (or transite) order: for (at least, usually) one requires, thatthe universe of the structures considered bea st, and no set iso ‘morphic Yo the totality of al ordinals, let alone of al ses. 'NB. The (somewhat crude cassfcationin terms of order of the language considered has recently been refined by the wse of infinite formulae, for Instance infinite first-order Formule. These or, a least, important classes ‘@ of them are intermediate between fist-order and second-order formulae since any structure definable by a formula in is also definable by means fof a (nite) second-order formula, but not conversely. The ordinal i an example. 2% tnt and Sct, “Casing nace cardial’, Noes Amer. Math soe Wve pa INFORMAL RIGOUR " iis ear hat wat i achive ne case of i tena of eta st of neers by means ft an of integer whch ed Seal i the tery of inint omals)—petap ss om ede fen to tee and bli es about, ine numbers The aon ite st aac nth "Whatirmach mor intresting han ts bv remake fst hata ‘ih theory on forma an be econ pty, any ato {herons tout nie neon onl ean eens oa onet (Canin chose at oto to nie cond oe forme {@) A redhtn of ertons aout igh cde cocci cxGer sateen inte lneguge oct hay Since henton elation a oral (of moe) of any gen re loratadine Cascmoton ote omy epee ey ty Be, ik aconanence ‘thc econo formula ened ty 2 ort formal oft ther. Moe precy expe by such fom ve he Quan tried trang: vel ct of teu pe str Ooe ‘Sos ply btn transi iat hs oder neve fanquat forall aera; so eet adequate for preing ‘omdorer comsqnc somebody nol ave node ‘sil cclaon eesti; soe te atone xa sane tse alos ta of franc see ony ead Ot ‘iar Var (oe al se) flowed by ora wove Gani ae ‘Rrced tLe (vic) one har frm onan se gas (VbEa). act As ool any contusion tat we ay formulates fon exter canaguce canals forme tes afore asrion Shout ie clave type tus, However, hia tat fer faving the fron nero, my be very afl oink nem of ‘ont ner serio ‘ample Let be Zermc's xo wth the tom of inf, nd et Cit sete (anni fomslasn o he conn hypotensin he flowing fom ye colecion of ereday fee witht indi, Cass Co WEah Cun ~ Corn 2 HC CH sett XE Cay > SEV =F, 4.06 canna be 100 percent sur: for instance, comer the sale th efi tio, Weave hee st Fo marl acer nay Gadel numbers af corde ‘ol fst hoary, sch tat ne Fee Stearn) sty Fi dead (= 4, &yVinw 5s DY ‘As Tar emphasind, 1 nol defable by ans of «fester formal i the rece bens bow)