Anda di halaman 1dari 24

Paradoxes in Transfinite Arithmetic and the Perceived

Inconsistency of ZF Set Theory


B. M. Smith

Innovative Nuclear Space Power and Propulsion Institute


University of Florida
PO Box 116502, Gainesville, FL 32611-6502, USA
Email: smithbm@inspi.ufl.edu

April 2003

Abstract
Perhaps no other area of contemporary mathematics invites quite as many doomsayers
as does set theory and related foundational mathematical theories dealing with the concept
of the infinite. This paper looks at recent articles posted on the World Wide Web (WWW)
that claim that Zermelo-Fraenkel set theory is inconsistent. Although other purported
proofs have been posted on the Web, this paper considers a particular class of proof that
centers around the operation of exponentiation of infinite numbers. The common result
cited is that 2ℵ0 = c Ÿ ℵ0 , which seems to contradict another theorem of transfinite set
theory that |2ω | 4 |ω ω | = |ω| = ℵ0 . The arguments that claim such results yield internal
contradictions within ZF set theory are examined and shown to be false and based upon a
complete misunderstanding of set theoretic principles. A brief outline of possible reasons
why many people perceive set theory to be inconsistent is given. The paper is written for
a student readership for mainly pedagogical purposes. No original theorems are derived.
Derivations are informal yet tight enough in logical structure to enable students moderately
familiar with set theory to check the derivations in more rigorous formal language at their
leisure.

In a series of sweeping studies on the foundations of mathematics the logician Kurt Gödel
succeeded in proving that any axiomatization of the self-evident (intuitive) laws of arithmetic
in a formal system, with sufficient power to prove theorems, is doomed to be either inconsistent

1
2

or yield true theorems that cannot be proven to be true by application of the laws of the formal
system. One axiom schema that attempts to formalize the most basic human intuitions about
mathematical logic is Zermelo Fraenkel set theory (ZFT). From the very beginning foundations
of ZF set theory there have been attempts to prove that the theory is inconsistent and that it
yields paradoxes that are too difficult to reconcile with human intuition. Some paradoxes are
merely counter-intuitive, but other paradoxes have been claimed from time to time as proof of
the inconsistency, and hence utter uselessness, of the theory. This paper examines one type of
argument that claims ZFT is inconsistent. The type of argument referred to is not original,
but it has appeared in new guises in articles posted on the World Wide Web. For this reason,
an attempt has been made in this paper to alert students embarking on studies of set theory
and foundational issues in mathematical logic to the errors of logic and reason found in some of
these articles.
A very basic level of introduction to relevant concepts in ZFT is provided. Where results are
required an informal style of proof is adopted and technical definitions are kept to a minimum.
The intent is to keep the subject matter easy to read with an emphasis more on recognition of
errors in reasoning that are often made when just such informal proofs are attempted. The pur-
pose of this paper is therefore largely for a student audience. After providing a bare modicum
of mathematical background and history, a semi-formal presentation of the claimed proofs of
inconsistency of ZF set theory is given in Section 2. Section 3 gives a cursory introduction to
transfinite arithmetic and gives the necessary background for unravelling the purported para-
doxes and understanding why they are not even close to inconsistencies in ZFT. Section 4 shows
why the purported proofs of inconsistency are wrong and analyzes possible reasons why vari-
ous authors have made such errors (particularly pertinent considering that some authors have
claimed to be defending reason, logic and intellectual integrity, making it ironic that their own
work lacks such integrity). Section 5 branches into a wider discussion on more philosophical
objections to ZF set theory and transfinite sets in particular.

1 Preliminary Definitions and Remarks


This section presents some minimum mathematical background for students who are relative
newcomers to the study of set theory. It is assumed that the reader has some familiarity with
the basics of axiomatic formal system development and mathematical style theorem and proof
procedures.

1.1 ZF Set Theory—Basics


ZFT does not actually define what a set is, but rather provides a scheme for inductively estab-
lishing sethood by recursion starting with the concept of one set called the empty set or ∅, and a
primitive undefined binary predicate ∈ that is interpreted as indicating “membership of” a set.
For all sets A and B either A ∈ B or A ∈ / B (the former in English would be “A is an element
(or member) of B”, the latter would be “A is not an element (not a member) of B”.
3

The empty set is a member of all sets. Other sets are defined or proven to be sets inductively.
The symbol ω is the set of natural numbers {0, 1, 2, . . .}. That ω is actually a set (and not just
a collection of numbers) is an axiom of ZF theory (the Axiom of Infinity), so it is a postulate in
the same sense that ∅ is simply assumed to be a set. It is not trivial that an infinite collection of
sets (each natural number can be proven to be a set) is itself a set. However, the reasonableness
of the postulated sethood of ω was established by Dedekind long before ZFT was thought of,
indeed, Dedekind succeeded in defining a totally new class of mathematical objects (sets) that
could support a one-to-one mapping into a proper subset of the original set, and proved that
no finite set could support such a mapping. Thus the abstract existence of infinite sets was
known before ZF set theory was formulated. The Axiom of Infinity was therefore not at all
unreasonable or unwarranted. To some readers it may seem strange that anyone should even
question the right of mathematicians to call the collection of all natural numbers a set. While
one can sympathize with the simple intuition that all the natural numbers can be conceived of
as a unity, it would be nicer not to have to make this an axiom, one would rather like to prove
it from simpler intuitions. ZFT takes the position that infinite sets can be motivated abstractly
without any reference to finite objects, and thus that the Axiom of Infinity is simply thereby
held to be non-controversial.
Yet the question can be legitimately asked, “can an inconsistency be derived from this
axiom?” A formal inconsistency would be, at a minimum, a pair of theorems, one showing that
some statement “S” is a theorem, and one showing that “not S” is a theorem. This is also called
an internal contradiction. Note that such a contradiction is entirely distinct from the type of
contradiction generated in proofs using the method of reductio ad absurdem (RAA). In proofs
employing RAA one never assumes the truth of statement S, even though teachers often use
sloppy English and in a proof will write “now let’s assume S. . . ” and proceed to generate the
inconsistency, this is bad form, because in RAA what one really does is prove a more compact
statement like (S ⇒ ¬(T )) where T is some previously established theorem of the theory, and
“¬T ” is the negation of this theorem. In other words, the author of the proof never actually
assumes that statement “S” is true of the current axiomatized theory, but rather, the author
is in effect imagining a logically possible universe in which all the axioms of the theory hold
in addition to the statement S as a new axiom and then the proof proceeds to work out the
purely logical consequences of this counterfactual world. If a logical inconsistency ensues then
we can know that in any consistent mathematical universe the statement S must therefore be
false (where consistency is relative only to the prior postulates and axioms of the theory).
Any such inconsistency in an axiomatic scheme endowed with a suitable formal language. In
set theory the language of predicate calculus1 is used in formal proofs. If a logically inconsistent
set of statements could be derived as theorems it would be catastrophic because it would render
the axiom scheme useless, inasmuch as any statement whatsoever could then be deduced as a
1
Devised by Peano this language consists of all well-formed (grammatical) statements that can be made from
some simple atomic sentences (axioms or assumed true postulates and simple objects of the universe of discourse)
using the seven basic symbols ¬ (negation), ∨ (or), ∧ (and), ⇒ (implies), ⇔ (if and only if), ∀ (for all), ∃ (there
exists, or ‘for some’). A comma or colon is sometimes added for “such that”, and parentheses are used to form
compound well-formed sentences.
4

valid theorem by including in a derivation the conjunction (S ∧ (¬S)) or “S and (not S)”. In
the so-called ℵ0 paradox the inconsistency can be pithily summarized in two statements, one is
that “2ℵ0 = c > ℵ0 ”, and the other is “ℵ0ℵ0 = ℵ0 < c”. The aim of the first part of this paper is
to show that there is no such inconsistency derivable in ZF set theory, at least not in any form
given by the claimants [1, 2].
The axioms and basic theorems of ZF set theory are assumed as established as valid deduc-
tions, but the consistency of the ZF axioms is not assumed. So this paper does not assume that
ZF theory is consistent. Indeed, the purpose is to examine this as a distinct possibility. A proof
of the inconsistency of ZF set theory would indeed be highly welcome, because it would force
mathematicians to look for perhaps better foundations for modern axiomatic mathematics.

2 Description of the ℵ0 Paradox


This section presents the main purported proofs of inconsistencies in ZFT. In the following
ω should be strictly interpreted as the ordinal number, the least ordinal greater than all the
natural numbers. But without cause of error one can also interpret ω loosely as the set ω = N
the set of all natural numbers.

Conjecture 1. 1. By Cantor’s diagonal argument theorem 2ℵ0 Ÿ ℵ0 is a theorem of ZFT.

2. It is also a theorem of ZFT that |ω ω | = |ω| = ℵ0 , so in particular |ω ω | ≺ 2ℵ0 .

3. A third theorem of ZFT is that ℵ0ℵ0 = c = 2ℵ0 .


The conjecture is therefore that ZFT` (|ω||ω| ≺ ℵ0 )∧¬(|ω||ω| ≺ ℵ0 ), yielding an inconsistency
in ZFT.

This conjecture is actually false, so a proof is not given. Instead the following attempted
proof is taken almost verbatim from the references as an illustration of how thinking about
transfinite set theory can go astray.
Proof. (Attempted.) The idea is that for exponentiation of cardinals a and b one finds |ab | =
|a||b| . Then by results 2 and 3 we have |ω||ω| = ℵ0 ≺ c. So ZFT` ¬(|ω||ω| ≺ ℵ0 ).
But by result 3 we also have ℵℵ0 0 ≺ ℵ0 , or in other terms, |ω||ω| ≺ ℵ0 . So ZFT` (|ω||ω| ≺ ℵ0 ).
Therefore ZFT` (|ω||ω| ≺ ℵ0 ) ∧ ¬(|ω||ω| ≺ ℵ0 ), as required.
This type of argument is given for example in [1] and [2]. The error in this type of argument is
obvious to anyone with some formal background in transfinite set theory. To provide sufficient
background for understanding the error the following sections present a cursory summary of
transfinite arithmetic. Section 4 on page 16 will explain the simple error for any reader who
cannot see it immediately.
5

3 Cardinal and Ordinal Numbers


Dispensing with a lot of technical formal development, the reader is now invited on a whirl-wind
tour of the arithmetic of finite and infinite sets. For background the reader is referred to either of
the references [3, 4, 5]—these references are ordered in decreasing level of technicality—reference
[5] is the easiest to read though still fairly technical in parts.

3.1 Ordinal Arithmetic


Having established the existence of infinite sets in the abstract, and then having found a good
candidate for an infinite set, namely ω, the natural question is to ask whether or not more
infinite sets can be constructed. The first way of creating new infinite sets is to just continue
the process that was used to build up the entire set of natural numbers, namely by continuing
to make new sets by appending the entire collection of previous sets to itself. This is the process
of formation of ordinal numbers.
This is a very important notion. One might think that adding a small number like 1 to an
infinity like ω cannot make a larger number. But that is a mistaken interpretation of what
ordinal counting is doing. In ordinal counting one merely asks what order or arrangement of
abstract sets can be built up, so adding “1” to ω is only shorthand for saying that there is a
successor set to the set ω called ω + . Mathematicians label this set ω + 1 in analogy with the
way successors of natural numbers are named. Thus strictly in ZF set theory 3={0, 1, 2}, but
for shorthand it is often written as “2+1”. If this idea of ordering numbers is kept in mind, and
“adding +1” to get the next ordinal number is understood as merely a convenient shorthand
expression, not an actual arithmetical operation, then the idea of ordinal numbers should be
much clearer. The confusing thing is that most people still refer to this as ordinal addition, so
in a sense we are stuck with this unfortunate name for the process. It’s not all that bad though
because it works out to be the same as true addition for natural numbers. This is an important
point because, as we will soon see, it is not the same for infinite numbers, indeed 1 + ω 6= ω + 1
is a fact that follows from the strict definition of this “ordinal addition”.
Abstractly speaking, all ordinals correspond to ordered sets, such that an ordinal number
a would be specified by giving an example of a set S such that if one could count S in the
correct order then one would count up to the ordinal a. Then the ordinal corresponding to some
set A is found by ignoring the actual appearance of the individual members of A and instead
concentrating on their order or arrangement. Then a + b is obtained by first counting to a and
then counting b steps further. The ordinal a · b is obtained by counting to a and then doing
this again b many times. The exponential ab is obtained by (i) counting a, then counting again
another a steps repeating this a many times, then (ii) repeating (i) b many times.
However, these three types of ordinal operation (addition, multiplication and exponentiation
of ordinals) are defined with the order of doing things strictly in mind, so that b·a is conceptually
distinct from a·b and likewise for a+b and b+a. Counting to a first then counting b steps further
is clearly different to counting to b first and then counting a steps further. The end result is
the same for natural numbers (and for all finite numbers), but conceptually one is obtaining the
6

same result but in two different ways (you can run the 800m by going twice around an Olympic
track either clockwise or anticlockwise). Some concrete examples are given in the following
pages.
Thus one starts with 0 = ∅, then append this to itself to make a new set called “1”, so
1 = {∅, {∅}}, then append this to itself to make the set called “2”, so 2 = {∅, {∅}, {∅, {∅}}},
this process is continued indefinitely to create all the naturals numbers, which by the Axiom of
Infinity is itself a set, ω = {0, 1, 2, 3, . . .}. So the next ordinal number after ω is simply written
as,

ω + 1 = {0, 1, 2, 3, . . .} ∪ {ω} = {0, 1, 2, 3, . . . , ω} (1)


Three things are notable here. First, ω + 1 is just a name, it does not mean “add +1 to ω”, and
as one can clearly see by the righthand side of this definition the concept of this named set is
completely sensible. But it is very closely analogous to adding “1”, by the interpretation of this
set as the successor to ω. The second thing is that instead of the boldface symbol ω here the
plain symbol ω is used. This is the convention chosen in this paper to refer to the set of natural
numbers as an ordinal number, rather than specifically as a set. This is purely a matter of style
and taste for the present, but there are good reasons why the set ω and the ordinal number ω
formed from this set should be conceptually distinguished, even though it makes little difference
if one uses the same symbol for both. The third notable thing is that the last comma in Eq.(1)
is put in to indicate that all the natural numbers go before ω in this set, whereas ω appearing in
this set is a completely new addition, it is a single entity and a subset of the new entity “ω + 1”.
So now that this new entity ω + 1 is established the process can continue and we get,

ω + 2 = {0, 1, 2, 3, . . . , ω, ω + 1}
ω + 3 = {0, 1, 2, 3, . . . , ω, ω + 1, ω + 2}
.. (2)
.
ω + ω = ω · 2.

The two ideas expressed here are (i) that when you have an ordinal number a you can find a next
ordinal a + 1, and (ii) when there is a sequence of increasing ordinals a + 1, a + 2, a + 3, . . . then
there is a last ordinal that is greater than all these that could be expressed as lim(a + n) = a+ ω.
Strictly speaking the “+1”, “+2”, and “+3” here do not have to be necessarily the natural
numbers, they could be any other sequence of ordinals. So continuing the list begun in Eq.(2)
we could have, ω · 2 + 1, ω · 2 + 2, . . . ω · 2 + ω = ω · 3, where after the dots we indicate reaching
yet another limit ordinal. Continuing on we get: ω · 3 + 1, ω · 3 + 2, . . . , ω · 3 + ω = ω · 4, . . . , ω · 5,
. . . , ω · 6, . . . , ω · ω = ω 2 . In this last stretch a lot of intermediate ordinals have been skipped
over. To speed things up the next infinite sequence of limit ordinals can be conceptually jumped
through to yield another limit ordinal as follows, first we get the sequence ω 2 , ω 3 ,. . . ω ω . After
this limit we can skip through the sequences ω ω+1 , ω ω+2 ,. . . ω ω+ω = ω ω·2 .
2
Next jump through the sequences ω ω·2 , ω ω·3 , . . . ω ω·ω = ω ω , another limit ordinal. Now jump
2 2 2 2 2 2 2 3 ω
through these as follows, ω ω , ω ω +1 , . . . ω ω +ω , . . . ω ω +ω·2 , . . . ω ω +ω = ω ω ·2 , . . . ω ω , . . . ω ω .
7

At each limit here one imagines speeding up the progression to the rate of the previous limit,
so that sequences of lower limit ordinals are entirely skipped over. Eventually we can jump
through another sequence as follows,
.
2 ω ω ..
ωω ω ω +1 ω ω·2 ωω ωω ωω
ω ,ω ...,ω ,...,ω ,...,ω ,...ω = ω ω, ω ω + 1, . . .
where the last limit ordinal here is a really big one with ω many exponentiations, which is called
“ω tetrated to the power of ω”, and after the next ordinal ω ω + 1 the “. . .” indicates that we
still haven’t stopped. Indeed, this way of progressing through the ordinal numbers never reaches
an exhaustion point. Unlike each limit ordinal there is no actual limit of all the ordinals. This
can be seen by considering the collection Ord of all ordinal numbers. Is this collection itself an
ordinal number? If it is then Ord + 1 is an ordinal number not in the collection of all ordinal
numbers, a contradiction. So by reductio ad absurdem there is no set of all ordinal numbers.
The hierarchy of ordinal numbers is fascinating, but the main concern of this paper is the
more mundane seeming topic of how to add and multiply these ordinal numbers, since this is
where the initial inclings of paradox emerge. The first sign of something funny is that ordinal
addition turns out to be non-commmutative for the infinite ordinals, even though ordinal addi-
tion is commutative for finite numbers. This is easy to see in the simplest case. Consider the
following results,

1 + ω = {1} ∪ {0, 1, 2, . . .} = ω
ω + 1 = {0, 1, 2, . . .} ∪ {ω} = {0, 1, 2, . . . , ω} 6= 1 + ω

Both of these results can be obtained rigorously from set theory, but the short answer to why
1 + ω 6= ω + 1 is evident in the way ordinal numbers were defined by the two rules (i) and (ii)
above. We obtain the next ordinal always by “adding 1” (technically, by forming a new set
from the previous ordinal by joining that set with the previous ordinal itself), but in the other
order of operation the set {1} is joined with an ordinal, but since {1} is already a member of
the set of all successor ordinals greater than 1 we gain nothing by taking the union of ω many
natural numbers with {1}. On the other hand, ω ∈ / ω and ω is a limit ordinal ω = lim(1 + n),
so adding “1” to ω gives a next ordinal. It’s all a matter of definition. One does not have to like
the outcome of this process for obtaining higher and higher ordinals, but one can’t escape the
mathematical fact that the procedure is valid and gives some sort of rigorous type of transfinite
arithmetic—a conceptually distinct mathematical structure to finite arithmetic that a priori
need not be assumed to follow the same rules as finite arithmetic.
Intuitively though it doesn’t really seem all that strange. By loosening the formalism a little,
just imagine 1 + ω as conceptually adding ω many 1’s to 1 where “1” could stand for anything,
apples, circles, planets, steinlagers, or whatever. So in, 1 + ω=1+(1+1+1+. . . ), clearly this is
no different than simply adding together ω many 1’s, so the result is just ω · 1 = ω. For the
reverse order operation just imagine ω + 1 as adding 1 to an already reached limit of ω many
1’s, so ω + 1=(1+1+1+. . . ) +1, and this is conceptually, if not intuitively, different from the
former because in this case one has to imagine having first summed up to ω, so there is no
8

next natural number that can be obtained by continuing to add 1’s, thus a new ordinal must
be obtained, ω + 1 6= ω. Again, the difficulty some students have with this results from the
misinterpretation of ordinal addition as literally adding up numbers, which is wrong, what we’re
doing is examining arrangements of numbers in ordered sets. We’re not even trying to perform
addition in the traditional sense. If you like to think of it as addition then that’s fine, it is a
generalized sort of addition. Here are some more general assorted results.

1. ω + ω = ω · 2 6= 2 · ω = ω, so two lots of ω is quite different to ω many 2’s.

2. ω + ω 2 = {ω, ω + 1, ω + 2, . . . , ω · 2, . . . , ω · 3, . . . , ω · n . . .} = lim(ω · n) = ω 2 . So there


is no point in having ω apples if you are just going to add ω 2 more of them!

3. ω 2 + ω 2 = ω 2 · 2 6= 2 · ω 2 = ω 2 . So two lots of ω 2 gives a lot more steinlagers than ω 2


many two-packs of steinies!

4. ω 2 + ω 3 = ω 3 =
6 ω3 + ω2 = ω3 + ω2.

5. ω 6 + ω 3 · 3 + ω 6 = ω 6 · 2.

6. ω 10 + 1010 + ω ω + 5 = ω ω + 5.

Further, as an exercise the reader might like to try adding these three ordinals in all six
possible orders, α = ω, β = ω + 5, γ = ω 2 . There are five different results that should be
obtained: ω 2 , ω 2 + ω + 5, ω 2 + ω · 2 + 5, ω 2 + ω, and ω 2 + ω · 2.
These results follow from the definitions of the natural numbers on the one hand and the
definition of ordinal addition on the other. Now what about ordinal multiplication?
Ordinal multiplication is also non-commutative. But this is parasitic on the non-commutativity
of ordinal addition, so nothing essentially new is discovered. Multiplication has already been
employed in many of the results above. However, grouping is important. Here are some more
assorted results, displayed conceptually with the set theoretic derivations implicit.

1. 2 · ω = “ω lots of 2’s” = 2+2+2+. . . = 1+1+1+1+1+1+. . . =ω.

2. ω · 2 = “2 lots of ω’s” = (1+1+1+. . . ) +(1+1+1+. . . ) = ω · 2 > ω.

3. ω · ω = ω + ω + ω + . . . = lim(ω · n) = ω 2 .

4. (ω + 3) · ω = ω 2 .

5. ω · (ω + 3) = ω 2 + ω · 3 6= (ω + 3) · ω.

6. (ω + 3) · (ω + 2) = ω 2 + (ω + 3) + ω + 3) = ω 2 + ω · 2 + 3.

7. (ω + 1) · 3 = ω + 1 + ω + 1 + ω + 1 = ω · 3 + 1.

8. 3 · (ω + 1) = (3+3+3+. . . )+3 = ω + 3.
9

For all these results one gets the right answers with the knowledge that ω is a transfinite
ordinal. If ω were just a symbol then none of the above results could be derived unambiguously.
Normally one would just assume that any undefined symbols can be treated as finite numbers, so
symbolic variables then revert to obeying the good old rules of commutative arithmetic. When
the use of transfinite ordinals is clear from context then addition and multiplication no longer
can be dealt with as trivially as before. If the symbols are undefined then one also wouldn’t know
whether to treat them as complex numbers, Grassmann variables, multivectors, octonions, or
whatever. The distinction between finite and infinite sets is a categorical one, entirely analogous
to the distinction between real numbers and complex numbers.
Subtraction and division of ordinals are defined using addition and multiplication, but as
with the natural numbers, subtraction is only defined for β − α when β > α. In that case it is
possible to prove that there is a unique ordinal number γ that satisfies the equation β = α + γ.
For example, (ω + 3) − 2 = ω + 3, but (3 + ω) − 2 = ω, uniquely. And (ω + 1) − ω = 1 uniquely,
and ω · 2 − ω = ω because ω + ω = ω · 2.
Similarly, as long as α > 0 one can prove that for any other ordinal β there exists unique
ordinal numbers γ and ρ such that β = α · γ + ρ, where 0 ≤ ρ < β. When ρ = 0 it is said that
α is a left divisor of β, and γ is a right divisor of β.

Cardinal Exponentiation
Finally ordinal exponentiation should be considered (logarithms of ordinals will be left for the
experts). First note that a result used above was that ω · ω = ω 2 . The ordinal ω 2 here can
also be thought of as a limit ordinal, namely the first ordinal a such that ω + a = a. To see
this, note that ω · ω=ω + ω + ω + . . ., so it’s easy to see that putting an extra “ω+” in front
of this makes no difference to the arrangement. Now in a similar way ω ω = ω · ω · ω · . . ., so
another way of thinking of ω ω is as the first ordinal a such that ω · a = a, since putting “ω·”
in front of a = ω · ω · ω · . . . will make no difference to the order. The general definition of
ordinal exponentiation is a bit like this, it starts inductively by (i) defining 00 = f0 (0) = 0 and
a0 = fa (0) = 1 for any ordinal a > 0. Then (ii) to get at ab first consider all the successor
ordinals in b, call them o+ , then fa (o+ ) = fa (o) · a, and finally (iii) for every limit ordinal β in
b, define fa (β) = ∪{fa (γ) : γ < β}, then rules (i), (ii), and (iii) define ab as follows,

ab = fa (β + ). (3)
The notation ∪{f (x) : R(x, y)} used here means the union of all elements of sets defined by
the function f (x), over all sets x that have a relation R(x, y) with the set y. In rule (iii) the
relation is “ordinally less than”, and the union is of all elements of the function fa () on x.
Thus ordinal exponentiation is defined recursively from the bottom up. To evaluate ordinal
exponentials you go backwards from the top down, seeing which rules apply. To see that this
works out as you would expect from the previous rules for ordinal multiplication, consider the
following examples.
10

1. ω 3 = fω (2) · ω, by rule (ii), since 3+ = 4 is not a limit ordinal


= fω (1) · ω · ω, rule (ii) again
= fω (0) · ω · ω · ω = 1 · ω · ω · ω, by rule (i)
= ω · ω · ω.

2. 3ω = f3 (ω + ), by rule (iii), since ω is a limit ordinal


= ∪{f3 (n) : n < ω}, still by rule (iii)
= ∪{f3 (0), f3 (1), f3 (2), . . .}, by rule (ii) successively applied
= ∪{1, 3, 9, 27, . . .} = ω.

3. ω ω = fω (ω + ), by rule (iii), since ω is a limit ordinal


= ∪{fω (n) : n < ω}, still by rule (iii)
= ∪{fω (0), fω (1), fω (2), . . .}, by rule (ii) successively applied
= ∪{1, ω, ω 2 , ω 3 , . . .}, which itself is a limit ordinal as previously
deduced heuristically.
practise for
1. 5ω+1 = f5 (ω + 1) = f5 (ω) · 5 = ∪{h5 (0), f5 (1), f5 (2), . . .} · 5 = {1, 5, 25, . . .} = ω · 5.
2. (ω +1)3 = fω+1 (3) = fω+1 (2)·(ω +1) = fω+1 (1)·(ω +1)·(ω +1) = 1·((ω +1)·(ω +1))·(ω +1)
= (ω 2 + ω + 1) · (ω + 1) = (ω 2 + ω + 1) · ω + (ω 2 + ω + 1) · 1 = ω 3 + (ω 2 + ω + 1) =
ω 3 + ω 2 + ω + 1.
3. (ω +2)ω+1 = fω+2 (ω +1) = fω+2 (ω)·(ω +2) = ∪{1, ω +2, (ω +2)2 , . . .}·(ω +2) = ω ω ·(ω +2)
= ω ω+1 + ω ω · 2.
4. (ω 2 + ω)ω = . . . = ∪{1, (ω 2 + ω), (ω 2 + ω)2 , . . .} = (ω 2 )ω = ω 2·ω = ω ω .
5. 3ω·3+5 = f3 (ω · 3) · 35 = f3 (ω + ω + ω) · 243 = f3 (ω)f3 (ω)f3 (ω) · 243 = (ω)(ω)(ω) · 234 =
ω 3 · 243.
The last few of these examples were done rather heuristically.
A good exercise for the student is to prove that the following laws hold (order still important)
(1) ab+c = ab · ac , and (2) (ab )c = ab·c . Other provable rules like these, such as nω = ω for all
natural numbers n ≥ 2, help to ease the calculation of ordinal exponentials.
The main thing is that for finite ordinals the results of ordinal addition and multiplication are
commutative and obey all the familiar rules of arithmetic for natural numbers, and for transfinite
ordinals addition, multiplication, subtraction, division and exponentiation are non-commutative
yet entirely well-defined, no contradictions arise.
11

3.2 Cardinal Arithmetic


Now, the finite ordinals are just (isomorphic to) all the natural numbers. But a set like α =
{0, 1, 3, 5, 7, } is not a natural number. One would still like to perform arithmetic with such
finite sets. In order to define a generalized addition for finite sets that are not natural numbers
a new notion is required. This will be the concept of the cardinality of a set. It will enable a
generalized concept of size to be defined for all sets, and cardinal arithmetic will turn out to be
definable on all finite sets, not just the natural numbers.
In order to do this the rank of a set is defined recursively by a function r(α) as follows. Let
a be any fixed ordinal number. Let “y = f (x)” be a formula. The for any ordinal number b and
successor ordinal b+ , define the function ra () as follows

df
ra (0) = 0.
df
ra (b+ ) = f (ra (b)), whenever b+ < a.
df
ra (b) = ∪{ra (x) : x < b}, whenever b is a limit ordinal in a.

Then define the rank of a set X to be α if X ∈ r(α+ )+ (α+ ) \ rα+ (α) = r(α+ ) \ r(α). Here the
symbol \ denotes set subtraction or relative complementation. If such an ordinal α exists then
X is said to have rank and has rank α.
In more down to earth language, the rank of a set can be viewed as a function from sets to
ordinal numbers. The rank of a set is then the least ordinal number greater than the rank of
any member of the set. A few important results about rank can be stated (without proof here,
but see for example [4, chapter 6] for proofs).

1. Every ordinal number has itself as its rank.

2. If α < β then rα ⊂ rβ.

3. If a set has a rank then it is well-founded.

4. The Axiom of Regularity is equivalent to the statement that “every set has rank”.

5. The Axiom of Regularity guarantees that every ordered set corresponds to a unique car-
dinal number.

Some examples: The empty set has rank 0 (since it has no members and 0 is the least ordinal
number). The set 1={{}} has rank 1 (since its only member {}, has rank 0). The set {{{}}},
which is not a natural number, has rank 2 because 2 is the least ordinal greater than the rank
of this sets member element 1. The set {{}, {{}}, . . .} has rank ω.
Now the reason why this concept of rank is important is because it turns out to be very useful
for defining the cardinality of sets without resorting to a controversial axiom called the Axiom of
Choice. Instead a much weaker axiom can be adopted: the Axiom of Regularity, which asserts
simply that “every set is well-founded”. A set A is well-founded if every nonempty subset B
12

of A contains an element b for which b ∩ B = ∅. Unfortunately this axiom restricts set theory
somewhat because collections of objects like “all the ordinal numbers” cannot be considered
as sets, because they would violate the Axiom of Regularity. But such collections are already
known to be paradoxical, so not much power is lost by adopting the axiom of regularity. The
natural numbers are all well-founded, and so is ω. These sets are sufficient for development of
a huge load of powerful mathematics and applications.
So far, the result of most interest for transfinite arithmetic is item 5 on this list. Yet the
concept of a cardinal number has not yet been defined. This is now remedied.
First, let X ∼ = Y or X ↔ Y indicate that there is a bijective map from the set X into the
set Y .
For any set A, there is a set called the cardinal of A written |A| defined to be the collection
of all sets X of lowest rank for which X ∼ = A. This is the technical definition of the cardinal
of a set without reference to the Axiom of Choice. A less formal definition would be that the
cardinal of a set A is the least ordinal a such that a ∼= A, where a is viewed as the set {o : o < a}
of ordinals o that are less than a.
In practise the result that is most often used is the establishment of a bijective correspondence
between sets X ↔ Y that therefore proves that |X| = |Y |. Alternatively one may write X ∼ = Y.
All these three statements can be taken as equivalent.
To compare cardinals a special type of order symbol is used to unambiguously distinguish the
ordering from the totally distinct ordering relations <, ≤, >, and ≥ used for ordinal numbers.
So for any two cardinals κ1 and κ2 on writes “κ1 4 κ2 whenever there are sets A1 and A2 with
|A1 | = κ1 , |A2 | = κ2 such that there is an injective mapping of A1 into A2 . The converse also
turns out to be true. Furthermore, if κ1 6= κ2 then one can write κ1 ≺ κ2 . It is obvious that in
these cases respectively an alternative “cardinally greater than” is to write κ2 < κ1 iff there is
an injection A1 → A2 , and when κ1 = 6 κ2 one could write κ2 Ÿ κ1 .
Having defined what a cardinal is it is then possible to talk about arithmetic of cardinals,
specifically cardinal addition, multiplication and exponentiation. Since it is primarily cardinal
exponentiation that is of interest later on, the following discussion of cardinal arithmetic will be
brief and even more informal than the previous discussion of ordinal arithmetic.
Cardinal addition is fairly intuitive. If κ1 and κ2 are two cardinals, then to find the cardinal
κ1 + κ2 one simply finds two sets K1 and K2 such that |K1 | = κ1 and |K2 | = κ2 . Then define
κ1 + κ2 = |K1 + K2 |.
The conceptual difference between ordinal and cardinal addition is this. To find the cardinal
2+3 find a set X with 2 elements and find a set Y with 3 elements, then find the smallest ordinal
a such that the set {o : o < a} can be put in bijective correspondence with X ∪ Y . Whereas to
find the ordinal 2+3 by ordinal addition first count to two then count a further three numbers.
To define cardinal multiplication, suppose again that κ1 and κ2 are two cardinals, the we de-
fine κ1 ·κ2 to be the cardinal of the Cartesian product κ1 ×κ2 . Note that all cardinals are first and
foremost sets, so their cartesian product is always defined. For reference, the Cartesian product
of two sets X and Y is the set of pairs {ha, bi : a ∈ X, and b ∈ Y }. For example the Cartesian
product of the ordinals 2 and 3 is the set 2×3={0, 1}×{0, 1, 2}={h0, 0i, h0, 1i, h0, 2i, h1, 0i, h1, 2i}.
But note that the Cartesian product of natural numbers will not always be a natural number.
13

This is why cardinal arithmetic is useful, it can be employed on any sets, not just ordinals,
because every set will have a corresponding unique cardinal (or cardinality) according to the
Axiom of Regularity. (Note of course that all cardinals are limit ordinals by definition.)
Now the raw results of both cardinal multiplication and cardinal addition turn out to be
the same as ordinal addition for finite numbers. However, for transfinite numbers cardinal and
ordinal addition are completely different and so it is advisable to use a different symbol for
the cardinal form of arithmetical operations. Accordingly in the rest of this paper when two
cardinals are added this will be indicated by the bold symbolic operator + , and likewise when
two cardinals are multiplied this will be indicated by the symbolic operator × , thus the following
symbolic notation will be used

κ + λ, tells us implicitly to add these as cardinal number.


κ + λ, tells us implicitly to use ordinal addition.
(4)
κ × λ, tells us implicitly to use cardinal multiplication.
κ · λ, tells us implicitly to use ordinal multiplication.
The key difference for transfinite sets is that when one of κ or λ or both are infinite then
κ + λ = max(κ, λ). This is a very mundane result and it makes cardinal multiplication of
transfinite numbers a fairly easy job. It is quite different to ordinal multiplication. For example,
ordinally ω + ω = ω · 2 6= ω, but cardinally one gets |ω|+ + |ω| = |ω| = ℵ0 . The law for transfinite
cardinal addition follows because all infinite sets support a noninjective surjection, so the smaller
cardinal can always be mapped to the larger cardinal with enough elements left-over in the larger
set to still map bijectively this subset of the larger cardinal to itself.
A similar result holds for transfinite cardinal multiplication. The Cartesian product can be
imagined as the square or rectangular array with the first row having all the elements of one of
the cardinals κ, and the first column having all of the elements of the second cardinal λ. The
Cartesian product defining κ× × λ is then the array formed by pairing the ith first row entry with
th
the j first column entry, forming an infinite array whenever one or both of κ or λ are transfinite
sets. A bijection between the largest of these cardinals and the whole Cartesian product array
is obtained by first noting that κ × κ = κ, because any transfinite cardinal can be bijected to
it’s own Cartesian self-product by simply filling in the Cartesian product array with ordinals
chosen from κ (every cardinal is a limit ordinal, and hence is an ordered set). This is done by
filling in the array by mapping the ordinals in κ in order starting at the h0, 0i top left corner
entry of the Cartesian product array and then successively traversing the next column to the
right going down to the diagonal entry and then traversing that entry’s row to the left until
hitting the leftmost entry (as in the proof of the denumerability of the rationals from the map
N × N −→ Q). So a bijection is established κ ↔ κ × κ, thus κ × κ = κ as required.
Now because κ × κ = κ the result κ × λ = max(κ, λ) must follow as a corollary. To see
why, just note that the Cartesian product array for κ × λ can just as easily be traversed by
successively mapping elements from the larger of these cardinals (say λ for arguments sake) in
the same way as in the proof of the result for κ ↔ κ × κ. The asymmetry of the array in this
case is of absolutely no concern.
14

Cardinal Exponentiation
Unlike cardinal addition and multiplication, cardinal exponentiation poses a challenge. Expo-
nentiation of ordinals was fairly-straight forward recall. In the case of ordinals the extension
of multiplication to exponentiation was natural and similar to the way finite exponentiation is
defined in terms of multiplication. For cardinals one has to generalize from the rather different
idea of multiplication as the formation of a Cartesian product of sets. How is this accomplished?
Intuitively λκ could be thought of as adding up λ many objects, then adding another λ
objects to get twice λ, and repeating this in total κ many times to get λκ objects. In other
words, we are multiplying κ many products of λ’s. In terms of proper cardinal multiplication
we’d translate this as λκ = λ× × λ× × λ× × . . . (κ many × ’s here). This will be an array of dimension
κ. To simplify the conception of this infinite dimensional array the preferred thing to do is to
recognize that such an array can be considered as isomorphic to all sequences of length κ that can
be formed from any elements of λ. Call such sequences Seqκ (λ), thus these are all the sequences
that are possible by choosing elements of λ over and over again until one has a sequence of κ
terms. This can be viewed as a set isomorphic to λκ . Thus for example, all members of Seq3 (2)
can be written as,{h0, 0, 0i, h0, 1, 0i, h0, 0, 1i, h0, 1, 1i, h1, 0, 0i, h1, 1, 0i, h1, 0, 1i, h1, 1, 1i}. There
are eight members of this class of sequence, just as one expects for a representation of |23 |.
Using this definition the following results can be proven (in the following, κ, λ, µ and ρ are
all assumed to be cardinals).

1. For all cardinals κ, κ|1| = κ. Also, κ|0| = |1|, and |0||0| = |1| also. Also |0|κ = |0|, because
in this case the domain for the κ-sequence is empty.

2. Since the set 1 = {0} only has one member then |1|κ = |1|, because there is only one
unique κ-sequence that can be formed, i.e. |Seqκ (1)| = |1|.

3. κλ+ = κλ × κρ .

4. (κ × λ)ρ = κρ × λρ .

5. (κλ )ρ = κλ+ .

6. If κ1 4 κ3 and κ2 4 κ4 , then κκ1 2 4 κκ3 4 .

7. If κ2 ≺ κ3 then it may be possible that a κ1 can be found such that κκ1 2 = κ1κ3 . For
example, for all natural numbers n ∈ ω, |ω||n| = |ω| = ℵ0 .

8. There are examples known of cardinals κ1 , κ2 , κ3 , κ4 , such that κ1 ≺ κ3 & κ2 ≺ κ4 , such


that κ1κ2 = κκ3 4 .

9. There can be cardinals found for which κ1 ≺ κ3 such that κλ1 = κ3λ . For example, a well
known result is |2||!| = |R||!| = |R|. (|R| is the cardinality of the real numbers, also
denoted c.)
15

10. For all natural numbers n such that 2 ≤ n ≤ ω then |n||!| = |2||!| . Here ω can be
considered as the ordinal ω or the set ω of natural numbers.

11. For every set X, |℘(X)| = |2||X| . So in particular |℘(ω)| = |2||!| = c, by result 9.

12. If ℵ0 is the first transfinite cardinal. Formally one gets ℵ0 = |ω|, which is sensible since ω
is the first transfinite ordinal.

13. A special result is that |2ω | = |ω| = ℵ0 , where within the || cardinal operator one performs
ordinal exponentiation.

14. It can also be proven that, c Ÿ ℵ0 , which is Cantor’s diagonal argument theorem, in an
even fuller account this means, c = |2|0ℵ = |2||omega| Ÿ ℵ0 .

So the solution for defining cardinal exponentiation is quite natural. The problem is that
none of the more interesting results yield cardinals of known cardinal order. It is known that the
set of all cardinals is well-orderable, so any two distinct cardinals κ and λ are related by either
κ ≺ λ or κ Ÿ λ. But when asking simple questions about how the exponentials of cardinals are
related it is not always the case that an unambiguous answer can be given. For a start we have
a well-ordered set of finite cardinals, the natural numbers. Then all the transfinite cardinals are
denoted (in increasing order) ℵ0 , ℵ1 , ℵ2 ,. . . ℵω , ℵω+1 , . . . and so on. The ordinals can be used to
index all these “alephs” which are the ordered cardinals.
The problem lies in interpreting the exponentials of cardinals as cardinals. Naı̈vely one
would expect the exponentiation of cardinals results in another cardinal. But one of the very
simplest nontrivial transfinite cardinal exponentiations, namely |℘(ω)| = |2||!| = c is of unknown
cardinality. This is the so-called Continuum Problem. Cantor originally conjectured that c =
ℵ1 , which is one version of the Continuum Hypothesis (CH). But this later turned out to be
unprovable within ZFC. In 1940 Gödel showed that there is a model of ZFC in which |℘(ω)| = ℵ1 ,
this meant that CH is consistent with ZFC, so that one can never prove that c 6= ℵ0 . But that
only means that Cantor was not provably wrong within the axioms of ZFC. In 1963 Paul Cohen
showed that the negation of the CH is also consistent with ZFC, he did this by showing that
there is a model of ZFC in which ℘(ω) = ℵ2 , and there is also a model in which ℘(ω) = ℵ3 , and
indeed there are a whole lot of models that together mean that ℘(ω) can be almost any aleph
within reason. Together Gödel’s and Cohen’s results say that ZFC is not powerful enough to
determine uniquely the cardinality of ℘(ω) = c.
Another more philosophical version of a continuum hypothesis, that perhaps far more aptly
bears the same name, is that, “c = |R| is the cardinality of the continuous line”. It is very hard
to know whether the real numbers can fill in a line completely. It is known that the reals form
a metrically complete set, but no one knows if this can be used to infer that the infinity of the
reals is so big that this many point sets can actually fill in a continuous line. In fact, due to
the result that c × c = c this means that c (or equivalently |2||ω| or |℘(ω)|) many points must
also be numerous enough to fill in any other continua like the 2D plane, or a 3D volume and
so on. The flip side of this is the question of whether a point (a single number) is actually of
absolutely zero width or extent, or is it some infinitesimal extent, smaller than any real number,
16

but not absolutely zero? It’s even possible that for most applications it does not matter what
the cardinality of points in a continuum is, it may even be consistent to have absolutely infinitely
many point in any given continua. As long as a chosen set of points like the real numbers is
dense and complete on the line, or R × R plane, etc., then probably most useful mathematics
can be done by assuming the points fill in the continua, but without having to explicitly assume
so. In this light, the question really is, “what abstract property of a continuum is so vital that
denseness and completeness of point set topologies on the continuum will not suffice to impart
this property?” This is the question for which no one has so far provided a good intuitive answer.
The theory of hyperreal numbers and nonstandard analysis may have answers to these ques-
tions. However, this subject is not really related to the more mundane Continuum Hypothesis
(CH) of cardinals, which is a more mathematically well-defined question. So the interesting
topic of the cardinality of points forming a line is usually not explored by standard set theory.
?
The mathematical CH, that 2ℵ0 = ℵ1 , however is actively pursued by set theorists, and
extensions of ZFC are sought for which the CH may be provably true of false based upon some
extra simple intuitions. So far none have been found that are universally agreed upon. So
the CH remains an open question in mathematics. Some people are even of the opinion that
mathematics need not decide on only one consistent universe, so that both CH and ¬CH may
simply define alternative mathematical frameworks of equal validity. Again, this is a question
that is not universally agreed upon.

Transfinite Arithmetic
If ordinal and cardinal arithmetic was not enough, it turns out thatPinfinite summandsQand
infinite products can also be defined, but not just boring old sums like ∞i=0 and products

i=0 ,
but rather real transfinite cumulative sums and products, in other words, sums or products of a
sequence where terms in the sequence can be functionally dependent on any transfinite ordinal,
such as ω 2 + ω · 3 or whatever. So if β is any transfinite ordinal we can get a transfinite sum

or product, written as say α<β Γα where Γα is some sequence defined on the set of ordinals

{a : a < β}. A similar type of transfinite product can be defined. Not surprisingly this
type of generalized transfinite summand and product over higher transfinite ordinals is non-
commutative, so the order of the terms Γα of the sequence must be maintained in the correct
order.
But now this is taking us too far afield (see [4] for an exposition of transfinite summands
and products). We want to get to the claimed proofs of inconsistency in ZF set theory upon
which all this strange generalized transfinite arithmetic is based.

4 The answer to the “ℵ0” Paradox and Why it is Not an


Inconsistency in Set Theory
After the lengthy exposition of transfinite arithmetic above, one is now in a position to fully
appreciate why the Conjectures presented in the first section are not contradictions. It should
17

be patently clear to the attentive reader why the attempted proof of Conjecture 1 is invalid.
The short answer is that ordinal and cardinal arithmetic have been thoroughly confused in the
attempted proof. In fact results of ZFT have actually been used inconsistently in the purported
proofs. To some extent this is an inexcusable error, but allowance for possible excuses for these
errors are discussed later in Section 5.
So the short answer why the paradox is no a paradox, and certainly not an inconsistency is
really in the first line of the attempted proof where it is state that |ab | = |a||b| for any cardinals
a and B. This is then used invalidly to infer that |ω ω | = |ω||ω| . This is obviously invalid because
on the left ordinal exponentiation is implicitly adopted, whereas on the right it is clearly cardinal
exponentiation that is implied.
It is also obvious that countless paradoxes and inconsistencies would result if one were
to similarly equate ordinal arithmetical operations with cardinal arithmetical operations. To
attempt to say that there is only one valid generalization of finite arithmetic for transfinite
numbers is a rank amateur sort of attempt to consider transfinite sets on an equal footing with
the completely distinct mathematical structures of finite sets. There is no a priori philosophical
basis for such attempts. This is likely to be the case for any generalization of arithmetic to
transfinite sets given the characteristic distinctness between the finite and the infinite.
Equally obvious is the seeming fact that this explanation is not satisfying to some who
continue to see a paradox in the totally confused mix-up between ordinal and cardinal expo-
nentiation. The conceptual difference between these operations is simply not factored in to any
philosophical thinking by such critics of set theory. Some possible reasons why they might insist
that a paradox and inconsistency remains are discussed in Section 5 below.
Notice that on page 16 above a number of alternative statements of the CH were given.

c = ℵ1 ,
|ω|
|2| = ℵ1 ,
ℵ0
|2| = ℵ1 , (5)
2ℵ0 = ℵ1 ,
|℘(ω)| = ℵ1 ,

are all equivalent statements of the CH. The 3rd and 4th statements use the fact that “2” is a set
as well as a cardinal number. So it should be unambiguous to write 2ℵ0 as implying the operation
of cardinal exponentiation rather than ordinal exponentiation. The reason being that ℵ0 is a
cardinal, so when used in the exponent one naturally assumes that cardinal exponentiation is
implied. However, writing 2ω is not the same as writing |2||ω| because in the former case the
ordinal exponentiation operation is implied, while in the latter case an unambiguous cardinal
exponentiation operation is implied. And indeed the results of these respective operations are
completely different, thu8s 2ω = ω, thus |2ω | = |ω| = ℵ0 which does not equal |2||ω| Ÿ ℵ1 , or if
the CH is adopted as an axiom, then in ZFT+CH one would have |2||ω| = ℵ1 .
So it is entirely consistent in ZFT to have |2||ω| = ℵ1 and |2ω | 6= ℵ1 . In these and other
so-called “aleph paradoxes” one is never in the position of deriving statements “S and “not(S)”
18

both as theorems. To suppose that “|2ω | 6= ℵ1 ” is equivalent to “not(|2||ω| = ℵ1 ) is a gross


amateur mistake and oversight of mathematical language and logic.
This in essence is the source of the first ℵ paradox. If one inconsistently (falsely in fact)
assumes that ordinal and cardinal exponentiation should yield the same results then the paradox
is ensured. This is of course totally misguided and a violence against logic and reason to
suppose that cardinal and ordinal exponentiation should be the same. If it were the case then
one wouldn’t need exponentiation of alephs and ω’s to arrive at a paradox, simple transfinite
multiplication and addition would suffice.
This completes the refutation of the conjectures given in Section 2.

4.1 Why are These Results Thought of as Paradoxes?


OK, so all should be clear now that no paradoxes have been proven in ZFT. There is no incon-
sistency in the formal system of ZF set theory, at least not as claimed by the cited references.
So what was the basic problem or fundamental motivation that may have driven people awry?
The real question is, “why would anyone who understands set theory be so contrarily disposed
to the theory to reach at such desperate straws to vainly hope for some imagined death blow
contradiction, or so acerbic in regard for what was originally a beautifully rebellious branch of
mathematics—a quintessence of novelty and brilliant expanding of the human mind—to actually
want so badly to find an inconsistency where none exists?”2
There seem to be two identifiable objections to ZF set theory that are alluded to in the
aborted attempts to prove inconsistency.

1. The results just don’t look right and should be rejected for reasons of aesthetic taste or
philosophical presumption.

2. There is some deeper mathematical intuition telling us that even transfinite numbers
should not behave this way, indicating a serious error in the very conception of the infinite
(starting with the Axiom of Infinity perhaps).

Philosophical objections of type 1 are highly dubious and amount to straight out prejudice
and an ad hoc attitude to mathematics whereby one accepts a bunch of primitive axioms only
as long as they imply consequences that one’s delicate philosophical intuitions can swallow,
at which point one simply chooses to abandon the axioms because they are “obviously wrong”
with no justification other than the apparent ugliness of the theorems to the eye of the beholder.
Nevertheless, some of the saner philosophical objections along these lines will be discussed later
in Section 5 below. To put things in a starker light, the only aspects of a mathematical framework
that should be questioned are it’s axioms, not it’s valid theorems. If the theorems look strange
and counterintuitive then this may be reason to call the axioms into question, but unless a true
2
It should be stated again here that this paper does not claim that ZF set theory is consistent, and the
brilliance of the theory is not taken in any way as a reason why mathematicians should not be looking for better
foundations. For even if the ZF axioms are consistent there may still be a more satisfying and pleasing foundation
for mathematics that just needs inventing.
19

inconsistency is found one cannot exactly object to the axioms as long as they are in themselves
reasonable and intuitive.
Such is the case with ZFT, most mathematicians find the axioms fairly intuitive and self-
evident. In this case the supposedly unpalatable theorems about transfinite sets are either true
inconsistencies or they have to be accepted and absorbed into the consciousness for meditation
in case they are indeed revealing profound aspects of the formalization. No one says that ZFT
is consistent, but until actually proven inconsistent there is no reason to believe that ZFT is
leading the mathematical community down a proverbial garden path.
Type 2 seems to be the more valid objection. One could ask for instance, “suppose ZF set
theory is consistent, that doesn’t mean it has to be the only formalization of human mathematical
intuition, so maybe there is another consistent formal system that captures human intuitions
better and yields a less counterintuitive form of transfinite arithmetic and algebra?”. Section 5
also makes an attempt to touch upon this type of objection to transfinite set theory.
Yet in this case, if transfinite arithmetic turned out to obey the same rules as finite arithmetic,
and if ordinal and cardinal arithmetic gave equivalent isomorphic results then this would be in
all likelihood a better reason to prefer ZFT to the fictional alternative. Why should one prefer
the more counterintuitive theory? The short answer is that if ordinal and cardinal operations on
transfinite sets were no different to the same operations on finite sets then there would in fact
be little difference at all between finite and infinite. This would be far more counter-intuitive
than having the purported paradoxes of ZFT. It’s not that one prefers paradoxes, but rather
that in the case of finite versus infinite one fully expects the patterns and rules of the numbers
to be totally different. It is simply outrageous to expect that a theory of the infinite can be
based in a theory of the finite. Yet those who seek a form of set theory that is grounded in the
idea that patterns and rules that hold for finite sets as limits are approached should therefore
extend to infinite sets are essentially asking for just such a counter-intuitive set theory.
So again, what is the deeper intuition that tells some people that transfinite arithmetic
should obey similar rules that hold for finite sets in the limit that the infinite is approached?
Well, there does seem to be some merit in supposing that if relations between finite numbers
hold right up until the infinite is actually reached then the same rules shouldn’t just magically
change when the infinite is actually attained. This is the exact position adopted in [2]. There
are two comments that could be offered in response to such intuitions. First, one would prefer
to base such an intuition in simpler axioms, and not have to take this sort of behaviour as an
axiom. Secondly, as was the case for the Type 1 objection, there does not really seem to be any
philosophical supremacy for holding such a view in the first place.
Consider that in actually imagining abstractly attaining the infinite ω one is in some sense
going beyond all finite realms of number patterns. One could ask, “Why should there not be
some radical change in behaviour of the generalized numbers that are reached after ω?” In
this light ω is a definite dividing line between two classes of ordinals, and ℵ0 is a clear division
between two classes of cardinals. Why should they not define a sudden catastrophic change
in behaviour? Such catastrophe’s are littered throughout science, why not in mathematics
too? Indeed, much of the current research in chaos theory, complexity and catastrophe theory
employs mathematical models that demonstrate such jumps in characteristic behaviour and
20

sudden “magical” changes in pattern that one sees described by the ZFT version of change in
going from the finite to the infinite.

5 Philosophical Objections to Set Theory


It might be argued, by some, that the results of following through the logic of the axioms of ZF
set theory yields such strange and counterintuitive results that we (the collective community of
mathematicians) should just dump set theory and start again. Why can’t we have a set theory
where transfinite addition and subtraction are defined and commutative, and where exponen-
tiation is sensible and yields the same results on ordinals and cardinal numbers alike? These
requests do not seem unreasonable. But are they really all that intuitive and philosophically
appealing that we’d want to hold on to them at all costs?
Well, to answer the critics, it turns out that commutative addition and multiplication oper-
ations can be defined on all the ordinals if that’s what you like. The operations are called the
Hessenberg natural sum of ordinals and the Hessenberg natural product of ordinals.
the idea is simply to use expansions of ordinals in terms of powers of ω. This is “natural” in
the same sense that natural numbers are added on an abacus or table in terms of powers of
(usually) 10.
Think of what perhaps is one of the most suspicious ideas in transfinite set theory, that one
can remove an object from one set S1 that stands in bijective correspondence with a second set
S2 , to give a smaller set S10 and yet still be able to find a bijection between the smaller set S10
and the unchanged former S2 . One can even more counter-intuitively do this when removing
infinitely many objects from S1 . The bijection D : ω ↔ E defined by D : k → 2k defined for all
natural numbers k where Even is the set of all even natural numbers, is a classic example. We
remove all infinite odd numbers from ω and yet end up with a smaller set (the even’s) that can
still be mapped bijectively to ω. What’s going on here?
There is a double-barrelled answer to this question. The first is that we are talking about
finite sets all the time up until someone suddenly introduces the infinite set ω. Then all hell
seems to break loose. A smart philosopher might however ask the question in reverse: namely,
suppose a newborn babe had grown up having been taught only transfinite number theory, then
all of a sudden the child is exposed to finite number theory and quickly matches up their innate
intuitions about finite collections of objects (apples, crayons, lego blocks, etc) with this new
concept of finite numbers. The child is perplexed that all the beautiful theorems of transfinite
number theory seem to be totally out of kilter and jumbled when translated into finite sets. Of
course no human could practically ever experience such a paradigm shock, but essentially every
young mathematician experiences an exactly analogous shock when all the years of instruction
on finite number arithmetic is suddenly seemingly overturned by the introduction of transfinite
numbers in their first year university course. The shock is of course illusory, because no results
whatsoever of finite number theory are abandoned! One is merely being exposed to a completely
new class of sets. This is the proper way to teach the theory. For indeed, transfinite sets are
absolutely distinguised from finite sets. Indeed, Dedekind first established the result in 1888
that a set that could be put into a non-surjective injection with itself is fundamentally distinct
21

from a finite set (for provably no finite set can be mapped to itself in this way) and that this
characterizes all infinite sets as different from finite sets.
Moreover Dedekind’s characterization of infinite sets in this fashion was completely inde-
pendent of any reference to finite sets. So one cannot object the he was in any way placing
artificial man-made restrictions on the mathematization of the infinite, or “finitizing the infi-
nite” as some author’s have put it[2]. Quite the contrary in fact. By objecting to ZFT on the
basis of merely counter-intuitive theorems the philosophers who want to abandon ZF set theory
as a foundation for mathematics are really the ones who are finitizing the infinite. They are in
essence demanding that patterns and rules that hold for finite sets should also hold for infinite
sets. It has been argued in this paper that it is immature to expect all theorems, even in a
consistent mathematical universe of discourse, to conform to intuition. The only results that
the community of mathematicians could reasonably be asked to get in conformity with intuition
are the results that humans can compute and construct.
Thus there is a fundamental dichotomy by definition, the finite and the infinite. It is in fact
closely analogous to the fundamental dichotomy between the concepts of Zero and One. How
do you get something from nothing? Start with the empty set. How do we get the idea of “1”
without assuming that “1” or “One” is a coherent notion? The answer provided by set theory
is to form the set {∅} whose only member is the empty set, and call this set “1”. The shock of
this should be roughly the same as the shock in realizing that ω is fundamentally different from
any finite number, and hence one should not expect in the least that any theorems about finite
sets will be at all applicable to infinite sets like ω.
Note that this method of asking questions in reverse is always a good rubric or method
of philosophy, indeed such methods have been used effectively in physics to resolve apparent
paradoxes about time reversibility of the laws of physics and the thermodynamic arrow of time.
See for example [9].
In this paper the claim that ZF set theory is free of paradoxes when transfinite sets are
introduced is not being made. The aim is simply to warn students that the counterintuitive
results of transfinite set theory may in fact turn out to be tremendously profound and far
reaching consequences of not just ZF theory but perhaps of any possible formal axiomatization
of human mathematical intuitions! At least of any formalization powerful enough to capture
the very minimal most idea of the infinite.

5.1 Philosophical Issues Related to Cardinality Paradoxes


Similar philosophical statements can be made concerning the Cardinality paradoxes of transfinite
set theory that were made concerning the Golfer’s paradox. But in the previous exposition of the
Cardinality parados, there is a more serious issue related to mathematical pedagogy that can be
addressed. This is the issue of lack of rigor in internet postings and other informal expositions
and essays. The intent here is not to criticize the lack of rigor of amateur mathematicians
and essay writer’s who feel they have something to contribute. But the increasing reliance by
students upon the easily accessible World Wide Web for readily digestible information threatens
to “un-educate” unwary students. Yet there is a fantastic freedom offered by publishing on
22

the Web that shouldn’t be withheld, and any serious budding mathematicians will presumably
always check the facts before being sucked down pits of non-rigorous diatribe. But there must
be a better golden mean between easy information and factually checked information.
Concerning the Cardinality paradox, the problem clearly consisted of the inability to distin-
guish between different interpretations of the same symbol used in the mathematics literature.
Following the style of some authors, this paper has attempted to remove this ambiguity by using
distinct symbols for the ordering relations. Clearly Web writers need to take some responsibility
for this as well. I suggest a simple remedy. The following rules or “nettiquette” are suggested,
primarily or author’s who are writing about science and technical subjects:

• Authors who want to write technical articles and publish them should obey the rule (netti-
quette) that all quoted results of technical content and all references to technical literature
should cite peer reviewed literature.

• No assumptions should be made based upon other material published on the World Wide
Web unless that material has been previously published in peer reviewed literature or has
otherwise been similarly checked for veracity by the community fo scholars in some process
equivalent to peer review.

• An equivalent to peer review could for example be a technical monograph or textbook


published by any reputable technical publishing house, or any online encylopedia (such as
Wikipedia) in which the relevant results are quoted and directly referenced.

If these or stronger rules were adhered to, then the present author feels that the mistakes
made in [1] and [6] would not so easily have been made. Mistakes cannot of course be eliminated,
but every effort should be made as part of good nettiquette to reduce speculation and flaky logic
in technical literature published on the Web under the guise or claim of serious science. The
breadth of information and discussion available on the WWW is a treasure and should not be
seen as undermining the ability of serious students to search out the facts for themselves and
to actively become intellectual police for the Web itself. The fact that mistaken articles are
published on the WWW concerning fundamental challenges to science and mathematics should
be welcomed as an intellectual challenge to set right, but certainly erroneous articles should not
encouraged for the mere sake of the challenge!

5.2 A Note on the Sociology of Mathematics


Breifly, the position that most closely sympathizes with the discussions in this paper is the idea
advanced by Chaitin [7, 8] that mathematics is more like an empirical science than a pure search
for absolute truths. The old idea that mathematics discovers eternal truths about numbers and
other objects is still tenable. But in the light of the 20th century mathematics of Gödel, Church,
Turing, and Chaitin, it seems clear that the role of human mathematicians is somewhat less
absolute. The sum of the wisdom that this paper musters on this topic consists in the simple
realization that given a consistent axiom scheme then all the theorems one can derive therefrom
23

are indeed eternal absolute truths but only in that circumscribed domain that is traced out by the
given axioms. It has to be realized that all of mathematics cannot be encapsulated in any finite
set of axioms, and every consistent set of axioms defines a different universe for mathematics.
Thus mathematics is blessed with an embarrassingly rich ontology. One can almost choose
what universe to do one’s maths in. Humans are however constrained to the exploration of the
necessarily limited realms of mathematics where intuition can be transformed into axiom.
The consequences of these realizations are nevertheless liberating and powerful for the so-
ciology of mathematics (how humans interact and function as mathematical explorers). First,
it means that young mathematicians need not feel compelled to follow the lore handed down
to them by their professors. Secondly, it means that there is truly no end to the depths and
wonders that mathematicians have open for exploration. Unlike the physical ocean, the mathe-
matical “ocean” is not only infinite in expanse, but is also infinite in wonder and multi-valued
in it’s ontology. New theorems are virtually guaranteed to be found that are non-trivial.

6 Conclusions
Hopefully this paper has shown to the beginning student of set theory the importance of for-
malizing heuristic arguments, and the role of intuition that goes in concert with formalization.
The exploration of completely new foundations for set theory should be encouraged, particularly
in the area of transfinite number theory. More than anything, it is hoped that students will
gain insight into the sociological nature of the human mathematical enterprize, and understand
that mathematics is as much about experimentation and hypothesis forming as the branches of
physical science. Each inequivalent axiomatization of mathematics yields a different abstract
universe worth exploring. Cantorian, or Zermelo-Frankel theory is just one (possibly consistent)
such universe. Unlike physical science, mathematics is blessed with virtually (and surely literally
as well) infinitely many distinct and wondrous universes in the entire cosmos of mathematics.

References
[1] M. H. Knowles. 2003. The Good Shepherd’s Paradox: A New Paradox of Infinity in Set
Theory. PAIAS, Palo Alto Institute for Advanced Study. http://www.paias.com/paias/-
home/Mathematics/. . . /GSP.htm

[2] D. R. Garcia. Counting the Reals: A New Look at Infinities. Private communication, March
11-April 7, 2003.

[3] W. S. Hatcher. Foundations of Mathematics, W. B. Saunders, Philadelphia, 1968.

[4] Martin M. Zuckerman. Sets and Transfinite Numbers, Macmillan, New York, 1974.

[5] R. Rucker. Infinity and the Mind: the science and philosophy of the infinite, Bantam Books,
New York, 1983.
24

[6] K. Delaney, 2002. A Critique of the Diagonal Method, http://descmath.com/diag/-


index.html

[7] G. Chaitin. The Limits of Mathematics: A course on information theory and the limits of
formal reasoning, Springer-Verlag, 1998.

[8] G. Chaitin. The Unknowable, Springer Series in Discrete Mathematics and Computer Sci-
ence, Springer-Verlag, 1999.

[9] H. Price. Time’s Arrow and Archimedes’ Point: New directions for the physics of time,
Oxford University Press, 1997.

Anda mungkin juga menyukai