Contemporary Mathematics

Volume 690, 2017

http://dx.doi.org/10.1090/conm/690/13872

Set-theoretic foundations

Penelope Maddy

Contents
1. Foundational uses of set theory
2. Foundational uses of category theory
3. The multiverse
4. Inconclusive conclusion
References

It’s more or less standard orthodoxy these days that set theory – ZFC, extended
by large cardinals – provides a foundation for classical mathematics. Oddly enough,
it’s less clear what ‘providing a foundation’ comes to. Still, there are those who
argue strenuously that category theory would do this job better than set theory
does, or even that set theory can’t do it at all, and that category theory can. There
are also those who insist that set theory should be understood, not as the study
of a single universe, V, purportedly described by ZFC + LCs, but as the study of
a so-called ‘multiverse’ of set-theoretic universes – while retaining its foundational
role. I won’t pretend to sort out all these complex and contentious matters, but I
do hope to compile a few relevant observations that might help bring illumination
somewhat closer to hand.

1. Foundational uses of set theory

The most common characterization of set theory’s foundational role, the char-
acterization found in textbooks, is illustrated in the opening sentences of Kunen’s
classic book on forcing:
Set theory is the foundation of mathematics. All mathematical
concepts are defined in terms of the primitive notions of set and
membership. In axiomatic set theory we formulate . . . axioms

2010 Mathematics Subject Classification. Primary 03A05; Secondary 00A30, 03Exx, 03B30,
18A15.
Key words and phrases. Foundations, ZFC, category theory, multiverse.
It’s an honor to be included in this 60th birthday tribute to Hugh Woodin, who’s done so
much to further, and often enough to re-orient, research on the fundamentals of contemporary set
theory. I’m grateful to the organizers for this opportunity, and especially, to Professor Woodin
for his many contributions.

c
2017 American Mathematical Society

289

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 290 PENELOPE MADDY

about these primitive notions . . . From such axioms, all known

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

1 See[Ebbinghaus, 2007], pp. 42–43.

2 [Ebbinghaus, 2007], pp. 76–79, highlights the role of Hilbert’s program among the moti-
vations for Zermelo’s axiomatization. Apparently Zermelo was keen to include a consistency proof
in his [Zermelo, 1908], but Hilbert encouraged him to publish it as it stood, in part, Ebbinghaus
reports, because Hilbert knew ‘that Zermelo needed publications to promote his career’ (p. 78).

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 291

theorem, that the project didn’t turn out as Hilbert and his followers had hoped,
but the fact remains that the most sweeping moral of Gödel’s work – that classical
mathematics (if consistent) can’t prove its own consistency – is only possible given
the set theory’s codification of the entire subject into a neat set of simple axioms.
And subsequently, the set-theoretic codification made it possible to settle other
important questions of provability and unprovability, sparing the profession from
sadly doomed efforts. So the ‘embedding’ of mathematics in set theory has this
clear use. Presumably we’d all agree that the goal of proving something general
about classical mathematics counts as ‘foundational’ in some sense, so here we do
find set theory playing a foundational role.
Of course, this Meta-mathematical Corral is hardly the only foundational
role set theory has been thought to play, but I think two of these can be dismissed
as spurious. The first sees the ‘embedding’ of classical mathematics in set theory,
often called a ‘reduction’ of classical mathematics to set theory, as the immedi-
ate descendant of Frege’s Logicism. This line of thought takes Frege’s project to
be roughly epistemological: if mathematics can be reduced to logic, then knowing
a mathematical fact is reduced to knowing a logical fact; assuming that we have
an account of logical knowledge, or at least that finding such an account is a less
daunting prospect than finding an account of mathematical knowledge had previ-
ously appeared to be, this reduction is a clear epistemological gain.3 Of course
Frege’s logic turned out to be inconsistent and set theory has taken its place, but
the epistemological analysis is supposed to carry over: we know the theorems of
mathematics because we know the axioms of set theory and prove those theorems
from them. Thus the problem of mathematical knowledge reduces to the problem
of knowing the set-theoretic axioms.
The trouble with this picture is that it’s obviously false: our greatest math-
ematicians know (and knew!) many theorems without deriving them from the
axioms. The observation that our knowledge of mathematics doesn’t flow from the
fundamental axioms to the theorems goes back at least to Russell – who emphasized
that the logical order isn’t the same as the epistemological order, that the axioms
might gain support from the familiar theorems they generate, not vice versa4 –
and is prominent in Zermelo – who defended the Axiom of Choice on the basis of
its consequences.5 As is often noted, a well-known fact of arithmetic or analysis
or geometry may be considerably more certain than the axioms of set theory from
which it derived. For that matter, to make a Wittgensteinian point,6 in most cases
we only believe that there is a proof of a certain mathematical theorem from ZFC
because we believe the theorem on the basis of its ordinary proof and we believe
that all theorems of mathematics are ultimately provable from ZFC! So this pur-
ported foundational use of set theory, as the Epistemic Source of all mathemat-
ical knowledge, is a failure. But this casts no doubt on the Meta-mathematical
Corral.

3 It isn’t obvious that Frege himself saw the matter quite this way: much of his interest is in

objective grounding relations between propositions, which are independent of our human ways of
finding things out. See, e.g., [Burge, 1998].
4 See [Russell, 1907].
5 See [Zermelo, 1908a].
6 [Wittgenstein, 1978], Part III, is thinking of the relations between ordinary decimal cal-

culations and the corresponding inferences in, say, the notation of Principia Mathematica, but
the upshot is the same.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 292 PENELOPE MADDY

The other purported foundational role for set theory that seems to me spurious
is what might be called the Metaphysical Insight. The thought here is that the
set-theoretic reduction of a given mathematical object to a given set actually reveals
the true metaphysical identity that object enjoyed all along. Benacerraf famously
argued that this can’t be right, because, for example, Zermelo took the natural num-
bers to be ∅, {∅}, {{∅}}, . . . , von Neumann took them to be ∅, {∅}, {∅, {∅}}, . . . , and
there’s no principled reason to choose one over the other.7 There are practical rea-
sons to prefer the von Neumann ordinals – they generalize easily to the transfinite,
for example – but this sort of thing isn’t an indicator of ‘what the numbers really
are’. Or so the argument goes.
Of course the practice of set theory is filled with even more arbitrary choices,
like the conventional preference for the Kuratowski ordered pair. It’s worth noticing
that Kunen speaks of ‘defining mathematical concepts’, not identifying mathemat-
ical objects, and Enderton, who does refer to ‘mathematical objects’, only speaks
of how they ‘can be defined’ and how theorems about them ‘can be viewed’. In
yet another textbook, Moschovakis makes the thought behind this circumspection
explicit:
A typical example of the method we will adopt is the ‘identifica-
tion’ of [the geometric line] with the set . . . of real numbers. . . .
What is the precise meaning of this ‘identification’ ? Certainly
not that points are real numbers. . . . What we mean by the ‘iden-
tification’ of [the line] with [the reals] is that the correspondence
. . . gives a faithful representation . . . which allows us to give
arithmetic definitions for all the useful geometric notions and to
study the mathematical properties of [the line] as if points were
real numbers. . . . In the same way, we will discover within the
universe of sets faithful representations of all the mathematical
objects we need, and we will study set theory . . . [8] as if all
mathematical objects were sets. ([Moschovakis, 1994],
pp. 33–34, emphasis in the original)
The trick, in each case, is to identify the conditions that a ‘faithful representation’
must satisfy. For the case of ordered pairs, this is easy: two of them should be equal
iff their first elements are equal and their second elements are equal. The case of the
natural numbers is more demanding: a set of sets with its operations should satisfy
the (full second-order) Peano Postulates. For our purposes, the simple point is
these set-theoretic reductions don’t give any sort of deep metaphysical information
about the nature of the line or of ordered pairs or of natural numbers, nor are they
so intended. Metaphysical Insight, like Epistemic Source, is spurious, leaving
only Meta-mathematical Corral.
But this is hardly the end of the story. The impressive mathematical innova-
tions present in the earliest invocations of sets – Cantor’s in his work on trigonomet-
ric series and Dedekind’s on ideals in the early 1870s – aren’t actually foundational
in character, but around the same time, Dedekind also undertook a project that
was explicitly foundational: to ‘find a purely arithmetical and perfectly rigorous
foundation for the principles of infinitesimal calculus’ ([Dedekind, 1872], p. 767).
7 See
[Benacerraf, 1965].
8 In
the ellipsis, Moschovakis writes, ‘on the basis of the lean axiom system of Zermelo’,
perhaps gesturing toward the Meta-mathematical Corral.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 293

Charged to teach the subject, Dedekind laments ‘the lack of a truly scientific foun-
dation’, finds himself forced to take ‘refuge in geometric evidence’ which ‘can make
no claim to being scientific’ (ibid.). Comparing the line with the rational numbers,
we see in the rationals ‘a gappiness, incompleteness, discontinuity’ and in the line
an ‘absence of gaps, completeness, continuity’, so the key question emerges:
In what then does this continuity consist? Everything must de-
pend on the answer to this question, and only through it shall
we obtain a scientific basis for the investigation of all continuous
domains. ([Dedekind, 1872], p. 771)
The then-current understanding rested only on those geometric intuitions, and on
vague remarks upon the unbroken connection in the smallest
parts [by which] obviously nothing is gained. (Ibid.)
So, the challenge was to replace these unscientific vagaries with
a precise characteristic of continuity that can serve as the basis
for valid deductions. (Ibid.)
And this, of course, is what Dedekind goes on to provide, in his theory of cuts in
the rationals – using set-theoretic machinery.
At first glance, this may look like just one more instance of the set-theoretic
reduction that underlies the Meta-mathematical Corral, but in fact there’s
something more going on. It isn’t that we have an explicit mathematical item –
the ordered pair, or the numbers as described by Peano – which we then ‘identify’
with a set that can play the same role, do the same jobs. Instead, in this case, we
have a vague picture of continuity that’s served us well enough in many respects,
well enough to generate and develop the calculus, but now isn’t precise enough to
do what it’s being called upon to do: allow for rigorous proofs of the fundamental
theorems. For that we need something more exact, more precise, which Dedekind
supplies. This isn’t just a set-theoretic surrogate, designed to reflect the features of
the pre-theoretic item; it’s a set-theoretic improvement, a set-theoretic replacement
of an imprecise notion with a precise one. So here’s another foundational use of set
theory: Elucidation. The replacement of the imprecise notion of function with
the set-theoretic version is another well-known example.9
To isolate another productive foundational use of set theory, we need to trace
some of the profound shifts in the theory and practice of mathematics that took
place over the course of the 19th century, starting with geometry. Since the ancients,
geometry had been closely associated with ordinary diagrams, and late in the 1700s,
Kant developed his elaborate theory of spatiotemporal intuition to undergird this
approach. This comforting view of the matter was challenged early in the 19th
century, as it became clear how much there was to be gained by viewing ordinary
geometry from the richer, projective point of view – with its ‘imaginary points’
(points with complex numbers as coordinates) and ‘points at infinity’ (points where
parallel lines meet). Of course these new points can’t be pictured, so the work was
roundly resisted at first on the grounds that ‘it keeps itself too much aloof from
all intuition, which is the essential trait of mathematical knowledge’.10 The very
notion of invisible, unvisualizable ‘points’ where two disjoint circles or two parallel
lines in fact ‘intersect’ was not only contrary to the intuitive ground of geometry,
9 For a quick overview of the history, see [Maddy, 1997], pp. 118–126.
10 The remark comes from Möbius, he of the strip. Quoted in [Nagel, 1979], p. 219.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 294 PENELOPE MADDY

but to plain common sense. Still, as Nagel remarks in his historical survey, on the
closely related subject of negative and complex numbers:
Scandal or not, there were few mathematicians who did not rec-
ognize their value or whose logical consciences were so tender
than they would not use them. ([Nagel, 1979], p. 202)
The same could be said of the new projective geometry.
I bring up this episode because the ‘scandal’ was resolved in mid-century by von
Staudt, using proto-set-theoretic techniques, in particular a precursor of the method
of equivalence classes: for example, the point at infinity where two horizontal lines
meet is identified with what we’d now see as the set of lines parallel to these two,
and a given point at infinity is on a particular line if that line is in the set with
which that point at infinity has been identified. In this and related ways, von
Staudt managed to build surrogates for heretofore suspicious, possibly dangerous
new items (like points at infinity) out of uncontroversial, unproblematic materials
(ordinary lines), and to redefine the relevant relations so as to validate the existing,
successful theory. His goal in all this is to remove any queasiness we might have
about the legitimacy or coherence of the new, un-intuited items.
As time went by, it became clear that the construction tools needed for this
‘building’ process – tools von Staudt regarded as ‘logical’ – were actually set-
theoretic in character. Speaking of the operations codified in Zermelo’s axioms,
Burgess writes:
A crucial fact . . . is that these are essentially the existence as-
sumptions needed to get new spaces or number systems or what-
ever from old ones . . . in the manner of nineteenth-century in-
troduction of auxiliaries for the study of traditional spaces or
number systems. Indeed, the constructions of the auxiliaries in
question can be, and now in retrospect are, viewed as essentially
‘set-theoretic’ constructions, though some of them actually ante-
date Cantor. ([Burgess, 2015], p. 76, emphasis in the original)
This striking fact – that the methods of von Staudt and others all fall within the
few closure principles used by the early set theorists and codified by Zermelo –
this fact is what eventually made possible what we now know as the set-theoretic
reduction of classical mathematics.
But the story of queasiness-removal doesn’t end there. In the late 1800s, pure
mathematics was on the rise, and with it, the axiomatic method; in place of von
Staudt-like constructions, new fields were introduced instead by an explicit set of
axioms. In his comprehensive history, Kline describes the situation this way:
Mathematics, from a logical standpoint, was by the end of the
nineteenth century a collection of structures each built on its
own system of axioms. . . . As long as mathematics was regarded
as the truth about nature, the possibility of contradictory theo-
rems . . . would have been regarded as absurd. ([Kline, 1972],
p. 1038)
– but without that worldly backing, the question of which of these axiom systems
could be trusted became acute. This new queasiness could best be removed by a
proof of consistency, and set theory again presented itself, now as the source for
such proofs. We return in a moment to the difference between von Staudt-like

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 295

surrogates and axiomatic consistency proofs – roughly, between proving from ZFC
that there’s a structure of a certain sort, and proving from ZFC that there’s a model
that thinks there’s a structure of a certain sort – but for now what’s important is
that neither of these effectively removes queasiness unless ZFC itself is known to be
trustworthy.
Zermelo felt the force of these considerations, remarking that
I have not yet . . . been able to prove rigorously that my axioms
are consistent, though this is certainly very essential; instead I
have had to confine myself to pointing out now and then that the
antinomies discovered so far vanish one and all if the principles
here proposed are adopted as a basis. But I hope to have done at
least some useful spadework hereby for subsequent investigations
of such deeper problems. ([Zermelo, 1908], pp. 200–201)
We now smile, perhaps a bit wistfully, at this optimism.11 In the decades since
Gödel dashed these fond hopes, the hierarchy of large cardinals has arisen as a
measure of consistency strength, and the early foundational goal of conclusive
queasiness-removal has given way to a more nuanced matter of Risk Assessment.
So, for example, in the abstract for a recent ASL talk, Voevodsky speaks of the role
of set theory in his program of ‘univalent foundations’:12
Univalent foundations provide a new approach to the formal rea-
soning about mathematical objects. The languages which arise
in this approach are much more convenient for doing serious
mathematics than ZFC at the cost of being much more complex.
In particular, the consistency issues for these languages are not
intuitively clear. Thus ZFC retains its key role as the theory
which is used [to] ensure that the more and more complex lan-
guages of the univalent approach are consistent. ([Voevodsky, 2014],
p. 108)
Or as he puts it, more carefully, in the slides for this talk:
Set theory will remain the most important benchmark of consis-
tency. . . . each new addition to the . . . language will require for-
mal ‘certification’ by showing, through formally constructed in-
terpretation, that it is at least as consistent as ZFC. ([Voevodsky, 2013],
slide 21)
Obviously this generalizes to ‘at least as consistent as ZFC + one or another large
cardinal’.
Notice that Risk Assessment in either form isn’t the same as Meta-math-
ematical Corral: the point isn’t to round up all classical mathematical items into
one simple package, so as to prove something about all of it all at once, but to
assess a particular new, somehow dangerous or suspicious item to determine just
how risky it is. And it differs from Elucidation as well: von Staudt had before him
a perfectly functional geometric practice with ideal points; his task wasn’t to make
that practice more precise, and thus more functional, but to reproduce it chapter-
and-verse in a way that was less worrisome; conversely, Dedekind’s concern wasn’t
11 Recall footnote 2, above.
12 For the foundational role ‘univalent foundations’ might itself be intended to play, see foot-
note 37.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 296 PENELOPE MADDY

that the real numbers were somehow worrisome, but that they weren’t sufficiently
precise to support the practice, weren’t sufficiently functional. So we have at this
point these three distinct foundational uses for set theory, along with a pair of
spurious ones – Metaphysical Insight, and Epistemic Source.
There remains one more, quite familiar line of foundational thought, namely the
idea that set theory provides decisive answers to questions of ontology13 and proof:
if you want to know whether or not a so-and-so exists, see whether one can be found
in V; if you want to know whether or not such-and-such is provable, see whether it
can be derived from the axioms of set theory. (In fact, both of these are provability
conditions: a so-and-so ‘can be found in V’ iff the existence of a so-and-so can be
proved from the axioms.) This is sometimes expressed with the rhetorical flourish
that set theory is the ‘final court of appeal’ on matters of proof and ontology. I
should confess that I’ve indulged in this flourish myself ([Maddy, 1997], p. 26),
making it sound as if classical mathematics must bow to the dictates of set theory,
but in practice I’ve taken this foundational role to place methodological constraints
on set theory, the founding theory, not on classical mathematics, the theory to be
founded. In particular, I argued that set theory, if it was to play this role, should
be as generous as possible – so as not to curtail pure mathematics – and should
be given by one unified theory that’s as decisive as possible – so as to provide
unequivocal answers to questions of ontology and proof.14
Fearing, then, that the ‘final court of appeal’ is something of an exaggeration,
let’s look a bit more closely to discern what foundational uses are actually in play.
At the very least, there’s the plain sociological fact that derivation from ZFC is
generally regarded as standard of proof in mathematics: in practice, the availability
of the axioms of ZFC goes without saying; if stronger assumptions are in play, this
is explicitly acknowledged;15 if only weaker assumptions are needed, this is noted
to give a more nuanced picture of the dependencies involved. Burgess observes:
There are . . . no official censors preventing a group of dissi-
dents from founding a journal of their own, in which as a mat-
ter of editorial policy results must be proved according to the
group’s restrictive standards (or as the case may be, results may
be proved making use of the group’s preferred additional hy-
potheses), rather than presented as they would be in a journal
enforcing the orthodox standard . . . No one dissident school of
thought, however, produces work of sufficient volume at a suffi-
cient pace to keep a journal of high standards following such a
policy coming out regularly. ([Burgess, 2015], p. 118)
In this foundational role, then, formal derivation in set theory serves as a Shared
Standard of what counts as a proof.

13 This term isn’t intended in any philosophically loaded way: I just mean what the practice

asserts to exist, leaving the semantic or metaphysical issues open. [Mac Lane, 1981], p. 468,
and [Feferman, 1977], p. 151, both see set theory as inseparable from a kind of Platonistic
metaphysics, but I’ve argued that employing a rich set-theoretic ontology is consistent with being
an Arealist (as in [Maddy, 2011]).
14 These are the methodological maxims maximize and unify from [Maddy, 1997].
15 [Burgess, 2015], p. 177, footnote 11, notes an exception: proofs appealing to
Grothendieck’s work sometimes omit what comes to an appeal to inaccessibles. See Section 2
below.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 297

But what lies behind this sociological fact? Partly there’s the recognition that
formal derivation turns out to be a good mathematical model for the scope of
human proving activity, but why from these particular axioms? To shed light on
this question, recall the ‘ontological’ component of the ‘final court of appeal’: there
is a so-and-so if one can be found in V, if the existence of a so-and-so can be
proved from the axioms. A few pages back, in connection with Risk Assessment,
we noted a contrast between queasiness-removal by outright existence proof and
queasiness-removal by consistency proof. Now we find the purported standard of
existence apparently promoting von-Staudt-like construction over the more lenient
Hilbert-like idea that consistency of the appropriate set of axioms is enough. To
illustrate with an example, suppose a mathematician wants to know: is there a
definable (projective) well-ordering of the reals? In ZFC alone, the question can’t be
answered, but assuming, as many set theorists do, that ZFC+LCs is the appropriate
measure, the answer is no. Still, the opposite answer can be had in L. In light of
that fact, would we really want to shut the door on this mathematician? For that
matter, why shouldn’t we follow Hilbert and open that door to the existence claims
of any consistent set of axioms? Why does ‘final court’ insist that we restrict
ourselves to exactly what happens in V?
The source of this more stringent ‘final court’ criterion is simple: the branches of
modern mathematics are intricately and productively intertwined, from coordinate
geometry, to analytic number theory, to algebraic geometry, to topology, to modern
descriptive set theory (a confluence of point-set topology and recursion theory), to
the kind of far-flung interconnections recently revealed in the proof of Fermat’s Last
Theorem. What’s needed is a single arena where all the various structures studied
in all the various branches can co-exist side-by-side, where their interrelations can
be studied, shared fundamentals isolated and exploited, effective methods exported
and imported from one to another, and so on. Burgess puts the point forcefully:
Interconnectedness implies that it will no longer be sufficient to
put each individual branch of mathematics separately on a rigor-
ous basis . . . To guarantee that rigor is not compromised in the
process of transferring material from one branch of mathematics
to another, it is essential that the starting points of the branches
being connected should . . . be compatible. . . . The only obvious
way to ensure compatibility of the starting points . . . is ulti-
mately to derive all branches from a common, unified starting
point. ([Burgess, 2015], pp. 60–62)
Set theory’s universe, V, provides the Generous Arena in which all this takes
place, and that’s why the ‘final court’ condition takes the form it does: to be a
full participant in mathematical interaction, a so-and-so must appear along-side
the full range of its fellows, with all the tools of construction and interaction fully
available.16
16 Of course Shared Standard and Generous Arena depend on the same facts of set-

theoretic reduction as Meta-mathematical Corral: that formal proof is a good model of

provability by humans and that the axioms of set theory codify the fundamental assumptions
of classical mathematics. What separates them are the uses to which these facts are being put:
in Meta-mathematical Corral, ‘derivable in ZFC’ functions as model for ‘provable in classical
mathematics’; in Shared Standard, it’s used as a benchmark for what counts as a legitimate in-
formal proof; in Generous Arena, V brings all the objects and methods of classical mathematics
together for fruitful interaction. As foundational uses, these are distinct.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 298 PENELOPE MADDY

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

2. Foundational uses of category theory

Category theory was introduced in the 1940s, by which time the notion of
set-theoretic foundations had become mainstream orthodoxy. Like Cantor’s initial
appeals to sets, Eilenberg and Mac Lane’s categories emerged in the pursuit of
straightforwardly mathematical goals. Foundational concerns first came into the
picture in the form of criticisms of set theory’s foundational aspirations. Mac Lane
grants the effectiveness to date of set-theoretic foundations:
The prior situation in the foundations of Mathematics had in
one respect a very simple structure. One could produce one for-
mal system, say Zermelo-Fraenkel set theory, with the property
17 For some purposes, it’s useful to consider arenas not quite as generous as full ZFC, e.g.,

L(R), where Choice is false (assuming again that there are large cardinals in V), but all sets of
reals are Lebesgue measurable.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 299

that all ordinary operations of practising Mathematicians could

be carried out within this one system and on objects of this sys-
tem. . . . ‘every’ Mathematical object was or could be defined to
be a set, and . . . all of the arguments about these objects could
be reduced to the axioms of ZF set theory. . . . This one-formal-
system ‘monolithic’ approach has . . . been convenient for spe-
cialists on foundations ever since Frege and Whitehead-Russell.
On the one hand, all the classical nineteenth century problems of
foundations (the construction of integers, real numbers, analysis
. . . ) could be stated in this one system. On the other hand,
alternative formal system[s] could . . . be tested by comparison
(as to strength or relative consistency) with this one system.
([Mac Lane, 1971], pp. 235–236)
Here we find gestures toward Generous Arena, Risk Assessment, Elucidation
(for the case of the reals)18 , Meta-mathematical Corral (in the usefulness to
‘specialists in foundations’) and Shared Standard19 . Mac Lane’s only complaint
is that ‘this happy situation no longer applies to the practice of category theory’
([Mac Lane, 1971], p. 236).
What is it about category theory that purportedly ends the reign of set-
theoretic foundations?
Categorical algebra has developed in recent years as an effective
method of organizing parts of mathematics. Typically, this sort
of organization uses notions such as that of the category G of
all groups. This category consists of two collections: The col-
lection of all groups G and the collection of all homomorphisms
. . . of one group G into another one; the basic operation in this
category is the composition of two such homomorphisms. To re-
alize the intent of this construction it is vital that this collection
G contain all groups; however, if ‘collection’ is to mean ‘set’
. . . this intent cannot be directly realized. ([Mac Lane, 1971],
p. 231)
Because there is no set of all groups, set theory can’t properly found category
theory.20 So the argument goes.
To overcome this problem, Grothendieck devised a system of ‘universes’ essen-
tially equivalent to Zermelo’s hierarchy of Vκ ’s for κ inaccessible,21 and all parties
agree that surrogates for any actual use of category theory can be found within
some Vκ .22 Burgess writes:

18 Cf. [Mac Lane, 1986], p. 362: ‘this approach to Mathematics has the advantage that
every concept can be made absolutely clear and explicit’.
19 Cf. [Mac Lane, 1986], p. 377: ‘we have now stated an absolute standard of rigor: A

Mathematical proof is rigorous when it is (or could be) written out in the first order predicate
language [with membership as the only non-logical symbol] as a sequence of inferences from the
axioms ZFC, each inference made according to one of the stated rules’.
20 A move to NGB would introduce the category of all groups as a proper class, but it would

still leave out important functor categories. See [Mac Lane, 1971].
21 See [Zermelo, 1930].
22 See, e.g., [Feferman, 1977], p. 155. Grothendieck’s accomplishment here runs parallel to

von Staudt’s: show that the job of a worrisome item (point at infinity, category of all groups) can
be carried out with an uncontroversial item (a set of lines, a set of ‘enough’ groups).

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 300 PENELOPE MADDY

For applications . . . one doesn’t need a category of literally all

groups . . . It is always enough to have a category of ‘enough’
groups, though how many is enough may vary from application
to application. . . . Grothendieck’s hypothesis is that every set,
however large, belongs to some local universe [that is, some Vκ ]:
The ‘global universe’ . . . is simply the union of increasingly large
local universes. ([Burgess, 2015], p. 174)
Why doesn’t this settle the matter? Assuming that Risk Assessment is the foun-
dational use of set theory in question here, as Mac Lane sometimes suggests,23
then we have our answer: category theory is no worse off than ‘ZFC+ many Inac-
cessibles’, which doesn’t take us far beyond ZFC itself.24 With this mild extension
of ZFC, Meta-mathematical Corral would also be achieved, nor is there any
obvious difficulty for any of the other foundational uses of set theory.
But Mac Lane knows all this and clearly isn’t satisfied:
Given any universe U ′ , one can always form the category of
all categories within U ′ . This is still not that will-of-the-wisp,
the category of all categories überhaupt. ([Mac Lane, 1971],
p. 234)
What exactly is the imagined deficit? For any particular use of the category of
‘all’ groups or the category of ‘all’ categories, there’s a set-theoretic surrogate that
does the job. The objection here appears to be that the surrogate isn’t the real
thing. But why should it be? In this form, the objection begins to sound analogous
to the complaint, in Benacerraf’s context, that the von Neumann ordinals aren’t
acceptable set-theoretic surrogates for the natural numbers because the actual 2
isn’t an element of the actual 3! In other words, Mac Lane appears to be drifting
into the demand that a foundation provide Metaphysical Insight – a demand
that set theory never properly took on in the first place.
However that may be, the problem of ‘founding’ unlimited categories was taken
up by Feferman in the late 1960s and revisited most recently in [Feferman, 2013].25
Feferman explicitly identifies the issue as one of Risk Assessment, and the theory
whose risk he’s interested in assessing is one that allows the formation of
the category of all structures of a given kind, e.g., the category . . .
of all groups, . . . of all topological space[s], . . . of all categories.
([Feferman, 2013], p. 9)
In other words, not just, for example, the category of ‘enough groups’ (as Burgess
puts it) for some practical purpose, but the category of all groups. These

23 See [Mac Lane, 1986], p. 406: ‘in one sense a foundation is a security blanket: If you
meticulously follow the rules laid down, no paradoxes or contradictions will arise’. He goes on
to point out that a risk of zero can’t be achieved in this way, but the quotation in the text from
[Mac Lane, 1971], p. 236, explicitly recognizes the worth of Risk Assessment: ‘alternative
formal system[s] could . . . be tested by comparison (as to strength or relative consistency) with
this one system.’
24 In fact, [Burgess, 2015], p. 176, points out that the inaccessibles aren’t strictly necessary,

that they can be removed by careful use of reflection. But, he observes, ‘in Grothendieck’s kind
of work the intellectual faculties are being strained to their uttermost limit, and one doesn’t want
the distraction of any sort of bookkeeping requirements’.
25 [Feferman, 1977], p. 155, characterizes the deficit of the Grothendieck-style reduction as

‘aesthetic’, but it’s hard to see why a foundational use has to be beautiful in some way or other.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 301

unrestricted notions . . . are mathematically reasonable on the

face of it and do not obviously lead to paradoxical conclusions.26
(Ibid.)
The challenge, then, is to combine them with all the usual objects and operations
that category theorists call upon in the course of their mathematical work, and to
prove the resulting theory consistent relative to ‘a currently accepted system of set
theory’ (ibid.).
Despite incremental partial successes from Feferman, McLarty and others over
the years, this problem remained open until just recently. In his [Ernst, 2015],
Ernst shows that any theory that allows the formation of the category of all graphs
and that includes the required mathematical staples is in fact inconsistent.27 So
regardless of whether Mac Lane’s demand that set theory provide for unlimited
categories is based on Metaphysical Insight or on another motive, it’s a demand
that can’t be met – not because of any shortcoming on set theory’s part, but
because the objects he demands are themselves inconsistent, beyond the reach of
any consistent foundation.
Nevertheless, as a historical matter, many calls to replace set-theoretic founda-
tions with category-theoretic foundations arose from the ill-fated hope that category
theory, unlike set theory, could meet this impossible demand:
One might hope for some . . . new foundational system (the cat-
egory of categories?) within which all the desired objects live.
([Mac Lane, 1971], p. 236)
Though Lawvere (in [Lawvere, 1966]) takes exactly this thought as his title –
‘The category of categories as a foundation for mathematics’ – his primary focus is
actually on a foundational use distinct from those considered so far. But before we
turn to this new sense of ‘foundation’, we need to ask how this theory of categories,
or really an improved descendant due to [McLarty, 1991],28 fares on the various,
more familiar foundational uses that set theory has been seen to underwrite.
The Category of Categories as a Foundation (CCAF) begins from a simple
axiomatic theory that codifies the usual category-theoretic machinery to yield
the usual results of general category theory . . . as hypotheticals
of the form ‘If A is a cartesian closed category, then . . . ’ or ‘If
A is a non-trivial topos, then . . . ’. . . . On the other hand, the

26 It might appear that a paradox is ready to hand: the category of categories is a category;
the category of groups isn’t a group; what about the category of all non-self-membered categories?
The trouble with this tempting line of thought is that ‘membership’ isn’t native to the world of
naı̈ve category theory. We could define what it is for the category of categories to be ‘self-
membered’: it’s for the category of categories to be a category. Likewise the category of groups
is ‘non-self-membered’ because it’s not a group. But categories in general aren’t given in the
form ‘the category of all X’s’, so a general membership relation can’t be defined in this way.
As [Feferman, 2013], p. 9) remarks, ‘There is no sensible way . . . to form a category of all
categories which do not belong to themselves’.
27 The proof is non-trivial, structured roughly along the lines of a proof that there is no set of

all sets via Cantor’s theorem: Ernst assumes there is a category of all graphs R (actually reflexive
graphs, but the result generalizes to all graphs), concocts a certain exponential, shows there can’t
be a map of R onto this exponential, then uses the fact that the exponential is a substructure of
R to show that there is a map of R onto the exponential.
28 The system of [McLarty, 1991] avoids the shortcomings identified in the review

[Isbell, 1967] of [Lawvere, 1966].

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 302 PENELOPE MADDY

axioms prove the existence of few specific cases of these general

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

this augmentation is often resisted by advocates of category-theoretic foundations, because the

mathematics it’s designed to accommodate is regarded as expendable. This would be a new use
of a foundation – as a way of pruning mathematics! – directly opposed to the thought behind
maximize (see footnote 14).
30 Indeed they have so argued. See, e.g., [Hellman, 2003].
31 See [Von Neumann, 1925].
32 The literature on this topic is copious and tangled. See [Ernst, to appear] for a recent

overview.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 303

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

33 [Bernays, 1934]. See also [Maddy, 1997], pp. 127–128.

34 It might be of interest to investigate more carefully which intuitive notions each theory
appeals to and which it explicates, and where these lists differ, to examine how successfully each
explicates the other’s primitives. For example: ZFC-style set theory explicates ‘function’ with the
much-maligned set of ordered pairs; Mathias derides the category-theoretic treatment of ‘iteration’
as ‘clumsy’ ([Mathias, 2001], p. 227).
35 See [Maddy, 1997], [Maddy, 2011].
36 One reason, perhaps, to resist the pruning in footnote 29.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 304 PENELOPE MADDY

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

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 305

in which classes and membership in classes do not play any role.

([Lawvere, 1966], p. 1)
Mac Lane stresses the replacement of static elements with dynamic functions:
There are other possibilities. For example, the membership rela-
tion for sets can often be replaced by the composition operation
for functions. This leads to an alternative foundation for Math-
ematics upon categories . . . much of Mathematics is dynamic,
in that it deals with morphisms of an object L into another ob-
ject of the same kind. Such morphisms . . . form categories, and
so the approach via categories fits well with the objective of or-
ganizing and understanding Mathematics. ([Mac Lane, 1986],
p. 359)
McLarty emphasizes that this particular foundational use should reflect what math-
ematicians actually do in their mathematical lives:
Mac Lane had a different idea of foundations . . . He took a
foundation of mathematics to be a body of truths which organize
mathematics as do I here. More specifically, that is truths which
actually serve in practice to define the concepts of mathematics
and prove the theorems. I do not mean merely truths which
could in some principled sense possibly organize the practice
but truths actually used in textbooks and journal articles, and
discussed in seminar rooms and over beer, so their notions do
occur in practice. ([McLarty, 2013], p. 80)
The goal here is a foundation for mathematics that will capture the fundamental
character of mathematics as it’s actually done, that will guide mathematicians
toward the truly important concepts and structures, without getting bogged down
in irrelevant details. I propose Essential Guidance as an awkward label for this
foundational goal, hoping to highlight its two aspects: such a foundation is to
reveal the fundamental features – the essence, in practice – of the mathematics
being founded, without irrelevant distractions; and it’s to guide the progress of
mathematics along the lines of those fundamental features and away from false
alleyways.37
Of course, Lawvere, Mac Lane and McLarty are entirely correct when they
point out that set theory does not provide Essential Guidance.38 For the record,
though, we should note that this fact in no way compromises its other foundational
uses: the lack of guidance and the presence of extraneous details don’t under-
cut Meta-mathematical Corral, Elucidation, Risk Assessment, Generous
Arena, or Shared Standard.39 They do, however, keep set theory from providing
Metaphysical Insight; that’s the moral of Benacerraf and the rest. As floated
in passing earlier, could it be that something like this is part of what’s at issue for

37 It may be that this is the use Voevodsky has in mind for ‘univalent foundations’: ‘The

languages which arise in this approach are much more convenient for doing serious mathematics
than ZFC’ ([Voevodsky, 2014], p. 108).
38 Presumably, neither does ETCS. I take it ETCS and its extensions are to be called upon

for some foundational jobs, like Elucidation and Risk Assessment, while something broader,
like Augmented-CCAF, is to provide Essential Guidance.
39 Cf. [Mathias, 1992], p. 115: ‘to reject a claim that set theory supplies a universal mode

of mathematical thought . . . need not compel one to declare set theory is entirely valueless’.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 306 PENELOPE MADDY

some of these category-theoretic thinkers, some hope of uncovering the true nature
of the mathematical structures? However that may be, it does seem likely that
a version of Epistemic Source is implicit in the kind of guidance they have in
mind: our ‘understanding’ of the mathematics is presumably based in our grasp of
its fundamental concepts and techniques. In any case, I don’t see that anything like
Essential Guidance was among the ambitions of set-theoretic foundations in the
first place, so to count them as ‘failures’ of set theory is to fault a cat for not being
a dog. But if a category-theoretic foundation does deliver on this desideratum, it
would enjoy a dramatic advantage over set theory.
Alas, it isn’t clear that category theory does deliver on Essential Guidance.
I think it’s agreed on all sides that a category-theoretic conceptual framework is
a remarkably effective way of thinking in fields like algebraic topology and alge-
braic geometry – no one would suggest that specialists in these areas would do
better to think more like set theorists. The contested claim – if category-theoretic
foundations are to capture what’s most fundamental, to guide us to the mathemat-
ically significant concepts – is that all mathematicians would do better to think like
category theorists. But even Mac Lane, in his sober moments, doesn’t believe this:
Categories and functors are everywhere in topology and in parts
of algebra, but they do not yet relate very well to most of anal-
ysis. ([Mac Lane, 1986], p. 407)
Analysis is where set theory first arose, much as category theory arose in algebra,
so it’s no surprise that set-theoretic thinking is more suitable there.
Mathias, who has done much to bring out the difficulties of category theory in
analysis,40 gives this reading of Mac Lane’s resistance to set theory:
I would guess that his reason is not so much that he objects to the
ontology of set theory but that he finds the set-theoretic cast of
mind oppressive and feels that other models of thought are more
appropriate to the mathematics he wishes to do. ([Mathias, 1992],
p. 115)
But the analyst or set theorist might well feel the same way about the category-
theoretic cast of mind:
The CAT camp [who believe category theory has ‘the one true
view of pure mathematics’] may with justice claim that category
theory brings out subtleties in geometry to which set theory is
blind. . . . The SET camp [who believe set theory has ‘the one
true view of pure mathematics’] may with equal justice claim
that set-theoretic analysis brings out subtleties to which the CAT
camp is blind. ([Mathias, 2001], pp. 226–227)
It seems that both these camps are chasing a false goal: a foundation that delivers
Essential Guidance, a single understanding of what mathematics is, a single
recommendation on how mathematicians should think.41
40 See [Mathias, 2000], [Mathias, 2001]. Beyond analysis, [Ernst, 2014] explores some

potentially problematic examples from graph theory.

41 Mathias himself advocates ‘unity’, akin to Generous Arena (see below), but not ‘uni-

formity’, or Essential Guidance: ‘Is it desirable to press mathematicians all to think in the
same way? I say not: if you take someone who wishes to become a set theorist and force him to
do (say) algebraic topology, what you get is not a topologist but a neurotic’ ([Mathias, 1992],
p. 113).

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 307

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

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 308 PENELOPE MADDY

the objective of organizing and understanding Mathematics’, Mac Lane contin-

ues, ‘That, in truth, should be the goal of a proper philosophy of Mathematics’
([Mac Lane, 1986], p. 359, emphasis added). Perhaps the unremarked shift from
‘foundation of Mathematics’ to ‘philosophy of Mathematics’ is telling; perhaps what
Mac Lane is really after isn’t so much a replacement for set theory as a broader
appreciation for, and more philosophical/methodological attention to, the organi-
zational and expressive powers of category theory. I think we can all agree that
this would be a good thing!

3. The multiverse
The most recent challenge to straightforward set-theoretic foundations in V
comes, oddly enough, from among set theorists themselves. To quote yet another
textbook:
Should we suppose that the continuum hypothesis, for example,
has a definite truth value in a well-defined canonical model? Or
is there a range of models in which the truth value of the con-
tinuum hypothesis varies, none of which has any special onto-
logical priority? Forcing tends to push us in the latter direction.
([Weaver, 2014], p. 118)
Hamkins, whose position Weaver endorses, describes the situation this way:
Our most powerful set-theoretic tools, such as forcing, ultrapow-
ers, and canonical inner models, are most naturally and directly
understood as methods of constructing alternative set-theoretic
universes. A large part of set theory over the past half-century
has been about constructing as many different models of set the-
ory as possible . . . As a result, the fundamental objects of study
in set theory have become the models of set theory, and set theo-
rists move with agility from one model to another. . . . Set theory
appears to have discovered an entire cosmos of set-theoretic uni-
verses. ([Hamkins, 2012], p. 418)
This has come to be called a ‘multiverse’ view. Woodin observes of one such position
that
the refinements of Cohen’s method of forcing in the decades since
his initial discovery of the method and the resulting plethora
of problems shown to be unsolvable, have in a practical sense
almost compelled one to adopt the generic-multiverse position.
([Woodin, 2011], pp. 16–17)
If this is right, our fundamental theory shouldn’t be a theory of sets, but a theory
of set-theoretic universes.
There are actually several different multiverse proposals currently on offer.42
The most generous is Hamkins version:
The background idea of the multiverse . . . is that there should
be a large collection of universes, each a model of (some kind
of) set theory. There seems to be no reason to restrict inclusion
42 [Väänänen, 2014] provides a framework that encompasses the positions of Hamkins, Steel

and Woodin. S. Friedman sometimes uses multiverse language, but it isn’t clear (at least to me)
that his is a true multiverse view. See, e.g., [Arrigoni and Friedman, 2013].

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 309

only to ZFC models, as we can include models of weaker theories

ZF, ZF− , KP, and so on, and perhaps even down to second-order
number theory, as this is set-theoretic in a sense. . . . We want
to consider that the multiverse is as big as we can imagine. At
any time, we are living inside one of the universes, referred to as
V and thought of as the current universe, but by various means,
sometimes metamathematical, we may be able to move around
in the multiverse. ([Hamkins, 2012], pp. 436–437)
Hamkins is out to explain mathematical experience:
This abundance of set-theoretic possibilities poses a serious dif-
ficulty for the universe view . . . one must explain or explain
away as imaginary all of the alternative universes that set the-
orists seem to have constructed. This seems a difficult task, for
we have a robust experience in those worlds . . . The multiverse
view . . . explains this experience by embracing them as real.
([Hamkins, 2012], p. 418)
Steel and Woodin, in contrast, are concerned with the claim, suggested by Weaver
above, that CH has no determinate truth value.43
Steel observes that
we have good evidence that the consistency hierarchy is not a
mirage, that the theories in it we have identified are indeed con-
sistent. This is especially true at the lower levels, where we
already have canonical inner models, and equiconsistencies with
fragments of definable determinacy. This argues for developing
the theories in this hierarchy. ([Steel, 2014], p. 164)
He then defends the resulting ZFC + LCs on the grounds that it allows the set-
theoretic reduction:
The central role of the theories axiomatized by large cardinal
hypotheses argues for adding such hypotheses to our framework.
The goal of our framework theory is to maximize interpretive
power, to provide a language and theory in which all mathemat-
ics, of today and of the future so far as we can anticipate it today,
can be developed. (Ibid., p. 165)
The trouble, of course, is that this strong and attractive theory – ZFC + LCs – can’t
settle statements like CH, as forcing so dramatically demonstrates. Does it follow
that CH has no determinate truth value?44

43 Cf. [Woodin, 2011], p. 16: ‘The generic-multiverse position . . . declares that the Con-

tinuum Hypothesis is neither true nor false’. [Steel, 2014], p. 168, takes his approach to open
the possibility that CH and the like are ‘meaningless . . . pseudo-questions’.
44 Notice the sharp contrast between Hamkins’s goal – embrace the widest possible range of

universes – and Woodin’s and Steel’s shared goal – given what we take ourselves to know about
sets (roughly ZFC + LCs), figure out whether there is a fact of the matter about CH. Though both
can reasonably be called ‘multiverse’ views, they are quite different undertakings, and can’t be
evaluated by the same standards. E.g., though Hamkins’s all-inclusive project naturally prompts
him to include universes in which V = L, it would be inappropriate to criticize Woodin and Steel
for ruling them out.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 310 PENELOPE MADDY

Woodin addresses this question by formulating what he calls the generic mul-
tiverse, a collection of models of ZFC + LCs:45
It is generated from each universe of the collection by closing
under generic extensions (enlargements) and under generic re-
finements (inner models of a universe which the given universe
is a generic extension of). ([Woodin, 2011], p. 14)
In a slogan: if N is a forcing extension of M , and one of them is in the generic
multiverse, then so is the other. Obviously, he and Hamkins differ on the range of
the universes in their multiverses – he and Steel both begin from our current best
theory of sets46 – but they also disagree sharply on the relevant notion of truth:
Woodin takes a claim to be true in the multiverse iff it holds in every universe of
the multiverse; Hamkins rejects this notion47 and holds that his many universes
‘exhibit diverse set-theoretic truths’ ([Hamkins, 2012], p. 416).
Woodin goes on to argue (assuming the Ω conjecture) that generic multiverse
truth amounts to ‘a brand of formalism that denies the transfinite’, and hence that
‘the generic-multiverse position . . . is not reasonable’ ([Woodin, 2011], p. 17).
Steel reports that
Woodin’s paper makes some arguments against the generic mul-
tiverse position, based on the logical complexity of certain truth
predicates, but those arguments do not seem valid to me. ([Steel, 2014],
p. 170; for details, see his footnote 24)
I can only leave this call to the experts.
Steel’s own multiverse approach is more syntactic than either Hamkins’s or
Woodin’s, but before we get to that, let’s pause to consider what becomes of set
theory’s foundational uses on a model-theoretic multiverse conception like these.
As noted in Section 1, I did once believe there was a strong prima facia case against
a multiverse of any kind, based on the assumptions that one of set theory’s leading
goals is to provide a foundation and that one function of a foundation is to serve
as a ‘final court of appeal’. The idea is simple: if set theory is to settle questions
of proof and existence, then set theorists should strive for a single preferred theory
of sets that’s as decisive as possible – so as to produce unequivocal judgments
when called upon. Of course this very same set-theoretic goal also counsels that
set theory be as generous as possible, so as not to limit mathematics,48 but at
least in the most conspicuous cases of potential bifurcation – like ZFC + LCs vs.
ZFC + V = L – it’s possible to avoid an exclusive choice – because ZFC + V = L
can be viewed as the theory of L even in a universe with large cardinals. But this
innocent observation actually paves the way for the less doctrinaire descendants of
the ‘final court’ identified in Section 1, where Shared Standard and Generous
Arena make room for stronger theories, as long as added hypotheses are flagged
so that import/export restrictions can be observed – and this begins to sound a bit

45 Actually both Steel and Woodin begin with models of ZFC, with the addition of large

cardinals as an option, but context (e.g., the above motivation from Steel) suggests they lean
toward taking that option. This is irrelevant for the case of CH, which is their primary focus.
46 See footnote 44.
47 Cf. [Hamkins, 2012], p. 445: ‘Woodin introduced [the generic multiverse] in order to

criticize a certain multiverse view of truth, namely, truth as true in every model of the generic
multiverse. . . . I do not hold such a view of truth’.
48 See footnote 14.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 311

like a multiverse view. So what I once took for an objection becomes a question: is
the multiverse intended to continue to play the foundational roles that the universe
once did, and if so, how does it do this?49
Hamkins addresses these questions directly:
Set theorists commonly take their subject as constituting an on-
tological foundation for the rest of mathematics, in the sense
that abstract mathematical objects can be construed . . . as sets.
([Hamkins, 2012], p. 416)
The multiverse view does not abandon the goal of using set the-
ory as an epistemological and ontological foundation for mathe-
matics, for we expect to find all our familiar mathematical ob-
jects . . . inside any one of the universes of the multiverse. (Ibid.,
p. 419)
The idea is that when we do set theory, ‘we jump inside and explore the nature of
set theory offered by that universe’ (ibid., p. 417), and whenever we do this, all the
usual resources of classical mathematics will be available.50
So what happens when our mathematician asks about that definable well-ord-
ering of the reals?
When a mathematical issue is revealed to have a set-theoretic
dependence . . . then the multiverse response is a careful explana-
tion that the mathematical fact of the matter depends on which
[universe] is used, and this is almost always a very interesting
situation, in which one may weigh the desirability of various
set-theoretic hypotheses with their mathematical consequences.
([Hamkins, 2012], p. 419)
This sounds very like what I said in the universe voice in Section 1: we explain
to the mathematician that there is no such well-ordering in V, but that there is
one in L; we lay out the mathematically attractive and unattractive features of L;
and we caution that if he elects to take this route, he will have to be careful about
exporting and importing resources to and from V.
Hamkins admits that avenues like this are available to the universe advocate,
for outer models as well as inner:
We do have a measure of access into the forcing extensions via
names and the forcing relation, allowing us to understand the
objects and truths of the forcing extension while remaining in
the ground model. ([Hamkins, 2012], p. 419)
So what makes the multiverse approach preferable?
The multiverse view explains our mathematical experience with
these models by positing that, indeed, these alternative universes
exist, just as they seem to exist, with a full mathematical ex-
istence, fully as real as the universe under the universe view.
(Ibid., p. 419)

49 This is to entertain the possibility that unify, in its original form, might not be required

for set theory’s foundational uses.

50 Here, presumably, the multiverse doesn’t include models of theories too weak to do the

job, a stipulation in apparent tension with the generosity cited a few pages back.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 312 PENELOPE MADDY

The multiverse view . . . takes . . . forcing at face-value, as ev-

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

51 Again assuming a more restricted range of models than Hamkins sometimes suggests (see

footnote 50).
52 For that matter, we might wonder whether, e.g., the Generous Arena should include the

entire multiverse, just as we wondered in the case of Augmented-ETCS vs. Augmented-CCAF.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 313

Steel feels the force of this concern. As we’ve seen, his initial presentation
(quoted above) puts the foundational goal front and center:
The goal of our framework theory is to maximize interpretive
power, to provide a language and theory in which all mathemat-
ics, of today and of the future so far as we can anticipate it today,
can be developed. ([Steel, 2014], p. 165)
He particularly recognizes the importance of Shared Standard and Generous
Arena, and of straightforward import/export:
Why not simply develop all the natural theories . . . ? Let 1000
flowers bloom! . . . The problem with this . . . is that we do not
want everyone to have his own private mathematics. We want
one framework theory, to be used by all, so that we can use each
other’s work. It is better for all our flowers to bloom in the same
garden. (Ibid., p. 164)
For this to work, Steel recognizes the need for an explicit, free-standing theory of
the multiverse. In a ‘historical note’, he remarks:
Neither Hamkins nor Woodin presented a language and a first-
order theory in that language, both of which seem necessary for
a true foundation. ([Steel, 2014], p. 170)
He provides this by introducing a multiverse language (ML) that speaks of both sets
and universes, and a list of axioms in that language (MV) that includes assertions
like these:
• For any axiom ϕ of ZFC + LCs, and every world W , ϕW .
• If W is a world and P ∈ W is a poset, then there is a world of the form
W [G], where G is P-generic over W .
• If U is a world, and U = W [G], where G is P-generic over W , then W is
a world. (See [Steel, 2014], p. 165)53
With the help of some serious mathematics, the theory MV can be successfully
formalized.54
Steel then constructs, from any transitive model M of ZFC+LCs, a set of worlds
M G that form a natural model of MV.55 At this point, ‘truth in the multiverse
M G ’ is a perfectly straightforward notion: a statement in ML is true or false in M G
in a standard model-theoretic sense. But this isn’t the notion Steel is after. His
leading question, like Woodin’s, is whether CH, a statement in the pure language of
set theory (LST), has a determinate truth value, and he approaches this question
by asking whether CH is or isn’t meaningful. (If it’s not meaningful, it’s not even
a candidate for truth or falsity.) So the multiverse position Steel considers isn’t
the Woodin-like proposal that a set-theoretic claim is true in the multiverse iff it’s
true in every universe of the multiverse, but something he calls the Weak Relativist
Thesis (WRT).

53 Woodin and Steel differ, e.g., over Steel’s axiom of Amalgamation ([Steel, 2014], p. 165):

‘If U and W are worlds, then there are G, H sets generic over them such that W [G] = U [H]’.
54 The key step for the last of these is a theorem of Laver and Woodin proved in an appendix

to [Woodin, 2011].
55 Let G be M -generic for Col(ω, <ORM ). Then M G is the collection of all ‘W such that

W [H] = M [G ↾ α], for some H set generic over W , and some α ∈ ORM ’ ([Steel, 2014], p. 166).

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 314 PENELOPE MADDY

To sneak up on the WRT, first notice that since the statement of concern,
CH, is a statement of LST, and Steel’s (perfectly ordinary) multiverse truths are
statements of ML, a link between the two languages is crucial. Fortunately, a
theorem of Laver and Woodin implies that there’s a recursive translation function
t from ML to LST such that for all ϕ in ML, M G thinks ϕ iff M thinks t(ϕ).56
The multiverse idea behind the WRT is that the only meaningful statements of set
theory are those expressible in ML, so the Thesis says that a statement ψ of LST
is meaningful iff there’s a ϕ in ML such that ψ is t(ϕ) – or more succinctly, iff ψ is
in the range of t.
So, is CH meaningful? As a warm-up, consider another statement in LST:
‘there is a measurable cardinal’. It’s an axiom of MV that this holds in every world
of the multiverse, but that’s not enough to show that it appears in the range of
t. In fact, it doesn’t. But Steel suggests that it’s reasonable to assume that t(for
all W , ‘there is a measurable cardinal’W ) and ‘there is a measurable cardinal’ are
synonymous.57 And since ‘for all W , ‘there is a measurable cardinal’W ’ is true
in M G , the synonymous set-theoretic statement ‘there is a measurable cardinal’ is
both meaningful and true. So far so good, but the same trick won’t work for CH,
because ‘for all W , CHW ’ won’t translate to anything like the LST statement that
CH. If this were the end of the story, CH would be meaningless and WRT would
yield the same conclusion as in Woodin’s multiverse: CH has no determinate truth
value; it would be, as Steel puts it, a ‘pseudo-question’ ([Steel, 2014], p. 154).
But this isn’t the end of the story, because, Steel notes, there may be ‘traces of CH’
elsewhere in ML, that is, there may be statements ϕ of ML other than ‘for all W ,
CHW ’ such that t(ϕ) is synonymous with CH.
How could this happen? Suppose that the multiverse language is expressive
enough to single out one among its worlds with an explicit definition. Woodin has
observed
that if the multiverse has a definable world, then it has a unique
definable world, and this world is included in all the others.
([Steel, 2014], p. 168)
This minimal world, if there is one, is called the core of the multiverse. If the
multiverse has a core (C) – the Weak Absolutist Thesis (WAT) – then it might be
reasonable to regard t(CHC ) as synonymous with CH.58 In that way, CH could turn
out to be meaningful even on the assumption of WRT, and its truth value would

56 In fact, Steel suggests that we only understand ML via the translation t. If so, this raises

a question of ‘psychological autonomy’ for MV that runs parallel to the case of category theory:
can we understand ML without first understanding LST? Steel recognizes the problem and offers
a rebuttal: ‘This is not a very strong objection, as one could think of what we are doing as
isolating the meaningful part of the standard language, the range of t, while trimming away the
meaningless, in order to avoid pseudo-questions. After climbing our ladder, we throw it away, and
from now on, MV can serve as our foundation’ ([Steel, 2014], p. 168). This move deserves some
scrutiny, but that’s a topic for another occasion.
57 This isn’t explicit in the relevant passage on p. 167 of [Steel, 2014], but Steel has con-

firmed it in correspondence (cited with permission). He also notes a residual concern that the
MV statement may involve a refinement of the original meaning, but I leave this aside (except for
footnote 58).
58 The residual concern in footnote 57 would arise here as well, perhaps even more acutely:

t(CHC ) involves a great deal of mathematical machinery unknown to many people who presumably
do understand CH (like Cantor, for instance).

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 315

depend on what happens in C. But this is speculative for now; much remains to
be explored.
In sum, then, the multiverse position Steel considers (WRT) would take the
axiom system MV as fundamental, in the sense that it serves to circumscribe our
official theory of sets (the range of t). From this perspective, ZFC + LCs (at least)
would be both meaningful and true, and thus available to play the usual founda-
tional roles in the usual ways. Whether CH has a determinate truth value would
remain, for now, an open question.
So, should we prefer a multiverse foundation to the familiar universe founda-
tion? I think we have to allow that the study of multiverse conceptions is in its
infancy, so firm conclusions aren’t possible at this stage (even if I had the wit to
draw them), but I would like to register some discomfort over the terms in which
the debate is often couched. We witnessed Hamkins’ ontological flight above; he
proposes that
each . . . universe exists independently in the same Platonic sense
that proponents of the universe view regard their universe to
exist. ([Hamkins, 2012], pp. 416–417)
Of course, a ‘proponent of the universe view’ might just hold that we should pursue
one preferred theory of sets, without metaphysical addenda,59 but Hamkins, qua
proponent of the multiverse view, takes a stronger line: ‘The multiverse view is one
of higher-order realism – Platonism about universes’ (ibid.).
In contrast, Steel’s multiverse position involves no such ontology. Instead, he
appeals to meaning: the multiverse language is used (via WRT) as an indicator of
which ordinary set-theoretic claims are meaningful, capable of truth or falsity, and
which are not. As we’ve seen, this raises the possibility that
the truth value of CH is not determined by the meaning we cur-
rently assign to the syntax of LST [the language of set theory].
([Steel, 2014], p. 154)
. . . and the suggestion (quoted above) that we should
trim back the current syntax, so that we can stop asking pseudo-
questions. (Ibid.)
Talk of indeterminacy in ‘the concept’, of diverse ‘concepts of set’, is also common in
multiverse thinking. Judging from these discussions, it appears that the overarching
goal of set-theoretic practice is to get these things right, to determine the true
Platonic ontology, the true contours of the meaning of the word ‘set’, or the true
nature of ‘the concept of set’.
But beneath the rhetoric, it emerges that this way of framing the question can’t
be quite right. For example, Hamkins admits that
we may prefer some of the universes in the multiverse to others,
and there is no obligation to consider them all as somehow equal
. . . we may simply be more interested in parts of the multiverse
consisting of universes satisfying very strong theories, such as
ZFC plus large cardinals. ([Hamkins, 2012], pp. 417, 436)
For that matter, it’s even possible that we might have good mathematical reasons
to seek out just one of the universes, just one extension of ZFC, as our unique
preferred theory. As far as meanings and concepts go, I personally doubt there’s
59 See footnote 13.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 316 PENELOPE MADDY

a fact of the matter about what is or isn’t part of the current meaning/concept of
set, but even if there were, and even if it didn’t settle CH, we’d be perfectly free to
decide that there’s good reason to move on to an enhanced meaning/concept that
does. Much like Hamkins, Steel admits this possibility:
Certainly we do not want to employ a syntax which encourages
us to ask pseudo-questions, and the problem then becomes how
to flesh out our current meaning, or trim back the current syntax,
so that we can stop asking pseudo-questions. ([Steel, 2014],
p. 154, emphasis added)
So the metaphysics of abstracta or meanings or concepts are all really beside the
point. The fundamental challenge these multiverse positions raise for the universe
advocate is this: are there good reasons to pursue a single, preferred theory of sets
that’s as decisive as possible, or are there not?
Now Steel laments that with matters of justification or ‘good reasons’, the ‘gen-
eral philosophical questions concerning the nature of . . . evidence . . . rear their
ugly heads’ ([Steel, 2014], p. 154), but I don’t think matters are so dire. I would
argue60 that the relevant reasons are all of a type Steel knows well: straightfor-
wardly mathematical reasons. What mathematical jobs do we want our theory of
sets to do? One answer is that we want it to serve the various foundational roles of
Section 1, but there are many others: Cantor was after a theory of trigonometric
series; Dedekind sought representation-free definitions; contemporary set theorists
hope for a rich theory of sets of reals; and so on. The choice between a universe
approach and a multiverse approach is justified to the extent that it facilitates our
set-theoretic goals. The universe advocate finds good reasons for his view in the
many jobs it does so well, at which point the challenge is turned back to the multi-
verse advocate: given that we could work with inner models and forcing extensions
from within the simple confines of V, as described by our best universe theory, what
mathematical motivation is there to move to a more complex multiverse theory?
Hamkins gestures toward this perspective on the question in his appendix:
The mathematician’s measure of a philosophical position may be
the value of the mathematics to which it leads. ([Hamkins, 2012],
p. 440)
He goes on to describe a pair of research projects inspired by the multiverse per-
spective. I’m in no position to evaluate the mathematics; my question is whether
multiverse thinking is playing more than a heuristic role, whether there’s anything
here that couldn’t be carried out in our single official theory of sets. If not, then
it’s not clear these examples give us good reason to incur the added burden of
devising and adopting an official multiverse theory as our preferred foundational
framework. Presumably this same measure could be applied to Woodin’s generic
multiverse: unless there’s at least some hint that it enjoys mathematical advantages
over the universe approach, we needn’t even concern ourselves about the complexity
of its truth predicate.
For the supporter of Steel’s WRT, the purported advantage of a multiverse
foundation appears to be that it saves us from the misguided pursuit of pseudo-
questions, from doomed efforts to settle matters that are indeterminate.61 If the
60 Have
argued, in [Maddy, 1997] and [Maddy, 2011].
61 A
defender of Woodin’s multiverse might also argue that it saves us from doomed efforts,
though Woodin himself – certainly no defender! – doesn’t say this.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 317

central goal of set theory is to identify the features of ‘the meaning we currently
assign to the syntax’ of set-theoretic language, and if the range of t marks the outer
limit of that meaning, and if no ‘trace of CH’ is hidden in ML, then efforts to settle
CH would be hopeless, inevitably sterile, and the multiverse rejection of CH as a
legitimate question might be a valuable outcome. Of course a universe advocate,
convinced, say, that ZFC + LCs marks the outer limit of what we’ll ever know about
V, could reach the same goal of deterring work on CH by epistemological means,
without recourse to multiverse thinking. But more to the point, there remains the
live possibility that the pursuit of CH isn’t in fact doomed. There might even be
what Hamkins calls a ‘dream solution’: a single new axiom that settles CH.62 Of
an even more ambitious dream, Woodin writes:
I am an optimist . . . There is in my view no reason at all, be-
yond a lack of faith, for believing that there is no extension of
the axioms of ZFC, by one axiom, a posteriori true, which set-
tles all the instances of the Generalized Continuum Hypothesis.
([Woodin, 2011], p. 30)
Indeed, Steel himself sketches a scenario that could lead even a tentative propo-
nent of WRT back to a strengthened universe theory that ‘fleshes out’ the current
meaning to something that settles CH in the positive.63 So I don’t see that there’s
call to throw over our universe foundation in favor of a multiverse quite yet!

4. Inconclusive conclusion
We’ve seen that set theory, largely via the well-known set-theoretic reduction,
serves a number of valuable mathematical ends that ought to qualify as ‘founda-
tional’: in the current form of ZFC+LCs, it provides a simple first-order theory that
interprets all of classical mathematics, so as to allow for meta-mathematical con-
sideration of the whole expanse of that vast subject at once (Meta-mathematical
Corral); it provides the conceptual resources and construction techniques to clar-
ify old mathematical notions in order to take on new demands (Elucidation); in
the hierarchy of large cardinals, it provides a flexible scale of consistency strength
(Risk Assessment); it serves as a benchmark of mathematical proof (Shared
Standard) and a framework in which the various branches of mathematics ap-
pear side-by-side, so that results, methods and resources can be pooled (Generous
Arena). On the other hand, it doesn’t tell us anything about the underlying nature
of mathematical objects (Metaphysical Insight) or of mathematical knowledge
(Epistemic Source).
Criticisms of this set-theoretic foundation from category theorists may falsely
assume that it aspires to Metaphysical Insight, but however that may be, they
explicitly demand that it found a theory that can’t in fact be founded (unlimited
categories). Advocates of a category-theoretic replacement for set-theoretic foun-
dations appeal to the category of sets for Elucidation and presumably for Risk

62 [Hamkins, 2012], pp. 429–230, argues that a dream solution to CH is impossible, but he

requires that the axiom in question be ‘obviously true’ in place of Woodin’s ‘a posteriori true’ in
the following quotation. (I take ‘a posteriori’ here to mean that the justification is extrinsic rather
than intrinsic.)
63 See [Steel, 2014], section 7. In this way, WRT – the assumption that our current meaning

is limited to ML – could ultimately point the way to a mathematically attractive extension of the
current meaning.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 318 PENELOPE MADDY

Assessment, but the intended execution of the other traditional foundational roles
is less clear, and the attractions of Generous Arena may even be rejected. Their
central hope is for a foundation that provides a range of concepts and methods that
capture and guide the productive ways of thinking that mathematicians actually do
and should employ (Essential Guidance) – a role set theory was never designed
to play. Proponents are surprisingly unconcerned that category theory doesn’t ap-
pear equipped to play this role for all areas of mathematics, another indication
that Generous Arena may have fallen away; different foundations might serve
for different branches of the subject. My suggestion is that we do best to retain
set theory in the foundational roles it plays so well, retain Generous Arena in
particular, but also pursue a serious philosophical/methodological investigation of
the various ‘ways of thinking’ in mathematics, beginning with the contrast between
those whose ‘essence’ is well-captured by category theory or by set theory.
Another challenge to ZFC-style set theory in its familiar foundational role comes
from the advocates of a multiverse conception of the subject. Steel in particular
offers an explicit, free-standing multiverse theory, MV, as fundamental – in place of
a theory in the language of set theory. With the multiverse assumption WRT, MV
ratifies a sublanguage of LST that includes ZFC + LCs (more or less by fiat), though
perhaps not the meaningfulness of CH. On this picture, the theory ZFC + LCs
continues in its usual foundational uses, but only at the behest of MV, and the
prospects for extending it are limited. Faced with the charge that this approach
shuts down free inquiry into extensions of ZFC+LCs prematurely, an advocate might
respond that the multiverse investigation has revealed this inquiry to be misguided,
because the meaning or concept of ‘set’ is simply indeterminate beyond certain
limits. Still, even if it’s true that our current meaning or concept is indeterminate
in this way, there remains the possibility that it might be more mathematically
productive, not to give up the quest for an answer to CH, but to seek out a fruitful
successor to our current meaning or concept – a possibility that Steel himself clearly
acknowledges and pursues. So far at least, the grounds for replacing the universe
with a multiverse are inconclusive.
In sum, then, it seems to me that the familiar set-theoretic foundations, rough
and ready as they are, remain the best tool we have for the various important
foundational jobs we want done.
4.1. Acknowledgements. I’m grateful to Michael Ernst, both for his
[Ernst, 2014], [Ernst, 2015], [Ernst, to appear], and for many enjoyable con-
versations on the aspirations of category-theoretic foundations, and to John Steel,
for his patience with my questions and confusions about his views. My admiration
for John Burgess’s [Burgess, 2015] should be obvious. Thanks also to Ernst, Steel,
David Malament, Colin McLarty, and an anonymous referee for helpful comments
on earlier drafts.

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 319

References
[Arrigoni and Friedman, 2013] Tatiana Arrigoni and Sy-David Friedman, The hyperuniverse
program, Bull. Symbolic Logic 19 (2013), no. 1, 77–96, DOI
10.2178/bsl.1901030. MR3087402
[Benacerraf, 1965] Paul Benacerraf, What numbers could not be, Philos. Rev. 74
(1965), 47–73. reprinted in Benacerraf, P., and Putnam, H. (eds),
Philosophy of Mathematics, second edition, Cambridge University
Press, Cambridge, 1983, pp. 272-294. MR0171702

[Bernays, 1934] On platonism in mathematics. reprinted in Benacerraf, P., and

Putnam, H. (eds), Philosophy of Mathematics, second edition,
Cambridge University Press, Cambridge, 1983, pp. 258–271.
MR742474

[Burge, 1998] Tyler Burge, Frege on knowing the foundation, Mind 107 (1998),
no. 426, 305–347, DOI 10.1093/mind/107.426.305. reprinted in
his Truth, Thought, Reason: Essays on Frege, Oxford University
Press, Oxford 2005, pp. 317–355. MR1630330
[Burgess, 2015] John P. Burgess, Rigor and structure, Oxford University Press,
Oxford, 2015. MR3329581
[Dedekind, 1872] Richard Dedekind, Continuity and irrational numbers. W. Beman
and W. Ewald, trans., reprinted in [Ewald, 1996], pp. 766–779.

[Ebbinghaus, 2007] Heinz-Dieter Ebbinghaus, Ernst Zermelo: An approach to his life

and work, Springer, Berlin, 2007. In cooperation with Volker Peck-
haus. MR2301067
[Enderton, 1977] Herbert B. Enderton, Elements of set theory, Academic Press
[Harcourt Brace Jovanovich, Publishers], New York-London, 1977.
MR0439636
[Ernst, 2014] Michael Ernst, The Category-Theoretical Imperative, ProQuest
LLC, Ann Arbor, MI, 2014. Thesis (Ph.D.)–University of Cali-
fornia, Irvine. MR3279213
[Ernst, 2015] Michael Ernst, The prospects of unlimited category theory: doing
what remains to be done, Rev. Symb. Log. 8 (2015), no. 2, 306–327,
DOI 10.1017/S1755020314000495. MR3348001
[Ernst, to appear] Michael Ernst, Category theory and foundations. To appear in
Landry, E., editor, Categories for the Working Philosopher.
[Ewald, 1996] William Ewald (ed.), From Kant to Hilbert: a source book in the
foundations of mathematics. Vol. I, II, Oxford Science Publica-
tions, The Clarendon Press, Oxford University Press, New York,
1996. Compiled, edited and with introductions by William Ewald.
MR1465678
[Feferman, 1977] Solomon Feferman, Categorical foundations and foundations of
category theory, Logic, foundations of mathematics and com-
putability theory (Proc. Fifth Internat. Congr. Logic, Methodology
and Philos. of Sci., Univ. Western Ontario, London, Ont., 1975),
Reidel, Dordrecht, 1977, pp. 149–169. Univ. Western Ontario Ser.
Philos. Sci., Vol. 9. MR0485370
[Feferman, 2013] Solomon Feferman, Foundations of unlimited category theory:
what remains to be done, Rev. Symb. Log. 6 (2013), no. 1, 6–15,
DOI 10.1017/S1755020312000111. MR3033209

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 320 PENELOPE MADDY

[Hamkins, 2012] Joel David Hamkins, The set-theoretic multiverse, Rev. Symb.
Log. 5 (2012), no. 3, 416–449, DOI 10.1017/S1755020311000359.
MR2970696
[Hellman, 2003] Geoffrey Hellman, Does category theory provide a framework for
mathematical structuralism?, Philos. Math. (3) 11 (2003), no. 2,
129–157, DOI 10.1093/philmat/11.2.129. MR1980559
[Hilbert, 1900] David Hilbert, Mathematical problems, Bull. Amer. Math. Soc.
(N.S.) 37 (2000), no. 4, 407–436 (electronic), DOI 10.1090/S0273-
0979-00-00881-8. Reprinted from Bull. Amer. Math. Soc. 8 (1902),
437–479. MR1779412
[Isbell, 1967] John Isbell, Review of [Lawvere, 1996] Mathematical Reviews,
MR0207517 (34 #7332), pp. 1354–1355.
[Kennedy, 2014] Juliette Kennedy (ed.), Interpreting Gödel, Cambridge University
Press, Cambridge, 2014. Critical essays. MR3381588
[Kline, 1972] Morris Kline, Mathematical thought from ancient to modern times,
Oxford University Press, New York, 1972. MR0472307
[Kunen, 1980] Kenneth Kunen, Set theory. An introduction to independence
proofs, Studies in Logic and the Foundations of Mathematics,
vol. 102, North-Holland Publishing Co., Amsterdam-New York,
1980. MR597342
[Lawvere, 1964] F. William Lawvere, An elementary theory of the category of sets,
Proc. Nat. Acad. Sci. U.S.A. 52 (1964), 1506–1511. MR0172807
[Lawvere, 1966] F. William Lawvere, The category of categories as a foundation
for mathematics, Proc. Conf. Categorical Algebra (La Jolla, Calif.,
1965), Springer, New York, 1966, pp. 1–20. MR0207517
[Linnebo and Pettigrew, 2011] Øystein Linnebo and Richard Pettigrew, Category theory as an
autonomous foundation, Philos. Math. (3) 19 (2011), no. 3, 227–
254, DOI 10.1093/philmat/nkr024. MR2842919
[Mac Lane, 1971] Saunders Mac Lane, Categorical algebra and set-theoretic foun-
dations, Axiomatic Set Theory (Proc. Sympos. Pure Math., Vol.
XIII, Part I, Univ. California, Los Angeles, Calif., 1967), Amer.
Math. Soc., Providence, R.I., 1971, pp. 231–240. MR0282791
[Mac Lane, 1981] Saunders Mac Lane, Mathematical models: a sketch for the phi-
losophy of mathematics, Amer. Math. Monthly 88 (1981), no. 7,
462–472, DOI 10.2307/2321751. MR628015
[Mac Lane, 1986] Saunders Mac Lane, Mathematics, form and function, Springer-
Verlag, New York, 1986. MR816347
[Maddy, 1997] Penelope Maddy, Naturalism in mathematics, The Clarendon
Press, Oxford University Press, New York, 1997. MR1699270
[Maddy, 2011] Penelope Maddy, Defending the axioms: on the philosophical foun-
dations of set theory, Oxford University Press, Oxford, 2011.
MR2779203
[Mathias, 1992] Adrian R. D. Mathias, What is Mac Lane missing?, Set theory of
the continuum (Berkeley, CA, 1989), Math. Sci. Res. Inst. Publ.,
vol. 26, Springer, New York, 1992, pp. 113–118, DOI 10.1007/978-
1-4613-9754-0 9. MR1233815
[Mathias, 2000] Adrian R. D. Mathias, Strong statements of analysis, Bull.
London Math. Soc. 32 (2000), no. 5, 513–526, DOI
10.1112/S0024609300217037. MR1767703

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 SET-THEORETIC FOUNDATIONS 321

[Mathias, 2001] Adrian R. D. Mathias, The strength of Mac Lane set the-
ory, Ann. Pure Appl. Logic 110 (2001), no. 1-3, 107–234, DOI
10.1016/S0168-0072(00)00031-2. MR1846761
[McLarty, 1991] Colin McLarty, Axiomatizing a category of categories, J. Sym-
bolic Logic 56 (1991), no. 4, 1243–1260, DOI 10.2307/2275472.
MR1136454
[McLarty, 2004] Colin McLarty, Exploring categorical structuralism, Philos. Math.
(3) 12 (2004), no. 1, 37–53, DOI 10.1093/philmat/12.1.37.
MR2033719
[McLarty, 2013] Colin McLarty, Foundations as truths which organize math-
ematics, Rev. Symb. Log. 6 (2013), no. 1, 76–86, DOI
10.1017/S1755020312000159. MR3033213
[Moschovakis, 1994] Yiannis N. Moschovakis, Notes on set theory, Undergraduate Texts
in Mathematics, Springer-Verlag, New York, 1994. MR1260432
[Nagel, 1979] Ernest Nagel, The formation of modern conceptions of formal
logic in the development of geometry, in his Teleology Revis-
ited and Other Essays in the Philosophy and History of Science,
Columbia University Press, New York, 1979, pp. 195–259.
[Russell, 1907] Bertrand Russell, The regressive method of discovering the
premises of mathematics. Reprinted in his Essays in Analysis,
Lackey, D., ed., George Braziller, New York, 1973, pp. 272–283.
[Steel, 2014] John R. Steel, Gödel’s program, Kennedy [2014], Interpreting
Gödel, Cambridge Univ. Press, Cambridge, 2014, pp. 153–179.
MR3468186
[Väänänen, 2014] Jouko Väänänen, Multiverse set theory and absolutely undecid-
able propositions, Kennedy [2014], Interpreting Gödel, Cambridge
Univ. Press, Cambridge, 2014, pp. 180–208. MR3468187
[Van Heijenoort, 1967] Jean Van Heijenoort (ed.), From Frege to Gödel, Harvard Univer-
sity Press, Cambridge, 1967.
[Voevodsky, 2013] Vladimir Voevodsky, Slides for his [Voevodsky, 2014]
Available at https://www.math.ias.edu/vladimir/lectures (#10).
[Voevodsky, 2014] Vladimir Voevodsky, Univalent foundations and set theory (ab-
stract). Bull. Symbolic Logic, 20(1):108.
[Von Neumann, 1925] John Von Neumann, An axiomatization of set theory. Reprinted
in [Van Heijenoort, 1967], pp. 394–413.
[Weaver, 2014] Nik Weaver, Forcing for mathematicians, World Scientific Pub-
lishing Co. Pte. Ltd., Hackensack, NJ, 2014. MR3184751
[Wittgenstein, 1978] Ludwig Wittgenstein, Remarks on the foundations of mathemat-
ics, Revised edition, MIT Press, Cambridge, Mass.-London, 1978.
Edited by G. H. von Wright, R. Rhees and G. E. M. Anscombe;
Translated from the German by G. E. M. Anscombe. MR519516

[Woodin, 2011] W. Hugh Woodin, The continuum hypothesis, the generic-

multiverse of sets, and the Ω conjecture, Set theory, arithmetic,
and foundations of mathematics: theorems, philosophies, Lect.
Notes Log., vol. 36, Assoc. Symbol. Logic, La Jolla, CA, 2011,
pp. 13–42. MR2882650

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.
 322 PENELOPE MADDY

[Zermelo, 1908] Ernst Zermelo, Investigations in the foundations of set theory. S.

Bauer-Mengelberg, trans., reprinted in [Van Heijenoort, 1967]
pp. 200–215. Also in his [Zermelo, 2010], pp. 188–229.
[Zermelo, 1908a] Ernst Zermelo, A new proof of the possibility of a well-ordering. S.
Bauer-Mengelberg, trans., reprinted in [Van Heijenoort, 1967],
pp. 183–198. Also in his [Zermelo, 2010], pp. 120–159.
[Zermelo, 1930] Ernst Zermelo, On boundary numbers and domains of sets: new
investigations in the foundations of set theory. M. Hallett, trans.,
in [Ewald, 1996], pp. 1219–1233. Also in his [Zermelo, 2010],
E. de Pellegrin, trans., pp. 400–431.
[Zermelo, 2010] Ernst Zermelo, Ernst Zermelo: collected works/gesammelte
Werke. Vol. I/Band I, Schriften der Mathematisch-
Naturwissenschaftlichen Klasse der Heidelberger Akademie
der Wissenschaften [Publications of the Mathematics and
Natural Sciences Section of Heidelberg Academy of Sciences],
vol. 21, Springer-Verlag, Berlin, 2010. Set theory, miscel-
lanea/Mengenlehre, Varia; Edited by Heinz-Dieter Ebbinghaus,
Craig G. Fraser and Akihiro Kanamori. MR2640544

Logic and Philosophy of Science, 3151 Social Science Plaza A, University of Cali-
fornia, Irvine , California 92697-5100
E-mail address: [email protected]
URL: http://faculty.sites.uci.edu/pjmaddy/

This is a free offprint provided to the author by the publisher. Copyright restrictions may apply.