Annals of Pure and Applied Logic 140 (2006) 233–252

www.elsevier.com/locate/apal

Levy and set theory

Akihiro Kanamori ∗
Department of Mathematics, Boston University, 111 Cummington Street, Boston, MA 02215, United States

Available online 25 October 2005

Abstract

Azriel Levy (1934–) did fundamental work in set theory when it was transmuting into a modern, sophisticated field of
mathematics, a formative period of over a decade straddling Cohen’s 1963 founding of forcing. The terms “Levy collapse”,
“Levy hierarchy”, and “Levy absoluteness” will live on in set theory, and his technique of relative constructibility and connections
established between forcing and definability will continue to be basic to the subject. What follows is a detailed account and analysis
of Levy’s work and contributions to set theory.
c 2005 Elsevier B.V. All rights reserved.

Keywords: Levy collapse; Levy hierarchy; Levy absoluteness; Relative constructibility; Reflection principles; Ackermann’s set theory; Urelements;
Fraenkel–Mostowski method; Forcing; Measurable cardinals; Indescribable cardinals.

Azriel Levy (1934–) did fundamental work in set theory when it was transmuting into a
modern, sophisticated field of mathematics, a formative period of over a decade straddling
Cohen’s 1963 founding of forcing. The terms “Levy collapse”, “Levy hierarchy”, and “Levy
absoluteness” will live on in set theory, and his technique of relative constructibility and
connections established between forcing and definability will continue to be basic to the subject.
Levy came into his prime at what was also a formative time for the State of Israel and has
been a pivotal figure between generations in the flowering of mathematical logic at the Hebrew
University of Jerusalem.1 There was initially Abraham Fraenkel, and then Yehoshua Bar-Hillel,
Abraham Robinson, and Michael Rabin. With Levy subsequently established at the university in
the 1960s there would then be Saharon Shelah, Levy’s student and current university president
Menachem Magidor, and Ehud Hrushovski, all together with a stream of students who would achieve worldwide
prominence in set theory.
What follows is a detailed account and analysis of Levy’s work and contributions to set theory. Levy’s work has
featured several broad, interconnected themes coincident with those in the pioneering work of Fraenkel: Axiom of
Choice independences using urelements, the investigation of axioms and the comparative strengths of set theories,
and the study of formal definability. To these Levy brought in concerted uses of model-theoretic reflection arguments

∗ Tel.: +1 617 353 9538; fax: +1 617 353 8100.

E-mail address: [email protected].
1 See Bentwich [49] for a history of the Hebrew University to 1960.

0168-0072/$ - see front matter 

c 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.apal.2005.09.009
234 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

and the method of forcing, and he thereby played an important role in raising the level of set-theoretic investigation
through metamathematical means to a new height of sophistication.
Already in his M.Sc. work under Fraenkel’s supervision, Levy [2,3] used urelement models to establish
independences for several definitions of finiteness. Levy’s ground-breaking 1958 Hebrew University dissertation,
Contributions to the metamathematics of set theory (in Hebrew with an English summary), under the joint supervision
of Fraenkel and Robinson featured work in three directions: relative constructibility, reflection principles, and
Ackermann’s set theory. Each of these is discussed in turn in the first three sections. Section 4 discusses Levy’s
hierarchy of formulas of set theory and his well-known absoluteness result. This work as well as the further
development of the thesis topics were undertaken when Levy was a Sloan postdoctoral fellow at the Massachusetts
Institute of Technology 1958–1959 and a visiting assistant professor at the University of California at Berkeley 1959–
1961. At Berkeley Levy through shared interests associated particularly with Robert Vaught. From 1961 Levy had
academic positions at the Hebrew University, eventually becoming professor of mathematics there while continuing
to have active visiting positions in the United States.
In the early 1960s Levy focused on independence results, first with Fraenkel’s urelement approach as developed by
Mostowski, and as soon as it appeared, with Cohen’s method of forcing. These are discussed in Sections 5 and 6, and
in Sections 7 and 8 Levy’s work on definability using forcing is presented. The latter section is focused on the Levy
collapse of an inaccessible cardinal, and in Section 9 his further work on large cardinals, measurable and indescribable
cardinals, is described. Levy put capstones to his work in different directions in the late 1960s, and these are taken up
in Section 10. There remains an envoi.
In such an account as this it is perhaps inevitable that the more basic results are accorded more of an airing and
the subsequent developments are summarily sketched, giving the impression that the latter are routine emanations.
This is far from the case, but on the other hand a detailed analysis of basic results becomes a natural undertaking
when discussing Levy’s work, since one sees and wants to bring out how they have become part and parcel of our
understanding and investigation of set theory.

1. Relative constructibility

Set theory was launched on an independent course as a distinctive field of mathematics by Gödel’s work [63,64]
on the inner model L of constructible sets. Not only did this work establish the relative consistency of the Axiom of
Choice (AC) and the Generalized Continuum Hypothesis (GCH), but it promoted a new relativism about the notion
of set as mediated by first-order logic, which beyond its sufficiency as a logical framework for mathematics was seen
to have considerable operational efficacy. Gödel’s work however stood as an isolated monument for quite a number
of years, the world war having a negative impact on mathematical progress. In retrospect, another inhibitory factor
may have been the formal presentation of L in Gödel’s monograph [65]. There he pointedly avoided the use of the
satisfaction predicate and, following John von Neumann’s lead, used a class theory to cast, in effect, definability.
This presentation of L, however, tended to obfuscate the metamathematical ideas, especially the reflection Skolem
closure argument for the GCH. Levy’s work in general would serve to encourage the use and exhibit the efficacy of
metamathematical methods in set theory.
In the next generation, András Hajnal, Joseph Shoenfield and Levy came to generalize Gödel’s construction in order
to establish conditional independence results. Their presentations would be couched in the formalism of Gödel [65],
but the metamathematical ideas would soon become clear and accessible. Hajnal [73,74] in his Hungarian dissertation
essentially formulated for a given set A the constructible closure L(A), the smallest inner model M of ZF such that
A ∈ M. To summarize, for any structure N and subset y of its domain, y is definable over N iff there is a (first-order)
formula ψ(v0 , v1 , v2 , . . . , vn ) in the free variables as displayed and a1 , a2 , . . . , an in the domain of N such that

y = {z | N | ψ[z, a1 , . . . , an ]},

where N | ψ[z, a1 , . . . , an ] asserts that the formula ψ is satisfied in N when v0 is interpreted as z and each vi is
interpreted as ai for i ≥ 1. For any set x now define

def(x) = {y ⊆ x | y is definable over x, ∈}.

 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 235

Finally, with tc(x) denoting the transitive closure of set x, define:

L 0 (A) = tc({A}); L α+1 (A) = def(L α (A)); L δ (A) = L α (A) for limit δ > 0;
α<δ

and

L(A) = α L α (A).

L(∅) is just Gödel’s L, and the general construction starts instead with L 0 (A) as an “urelement” basis. Although
L(A) is indeed an inner model of ZF, unless tc({A}) has a well-ordering in L(A), L(A) does not satisfy the Axiom of
Choice. This general situation was not broached by Hajnal, who used sets A of ordinals to establish the first relative
consistency results about cardinal arithmetic after Gödel. Hajnal showed that in L(A), if A ⊆ κ + , then for every
cardinal λ ≥ κ, 2λ = λ+ , the case λ = κ requiring some refinement of Gödel’s original argument. In particular,
if the Continuum Hypothesis (CH) fails, one can use an A ⊆ ω2 coding ℵ2 sets of natural numbers and injections:
= ω2 and (2ℵ0 = 2ℵ1 = ω2 ) L( A) ) to establish Con(ZFC + ¬CH) implies
L( A)
α → ω1 for every α < ω2 (so that ω2
Con(ZFC + ¬CH + ∀α ≥ 1(2ℵα = ℵα+1 )), or more dramatically, if one can prove 2ℵ0 = ℵ2 or 2ℵ0 = 2ℵ1 , then one
can prove the Continuum Hypothesis.
Levy [1,8] on the other hand developed for a given class A the inner model L[ A] of sets constructible relative to
A, i.e. the smallest inner model M such that for every x ∈ M, A ∩ x ∈ M. While L(A) realizes the algebraic idea
of building up a model starting from a basis of generators, L[ A] realizes the idea of building up a model using A
construed as a predicate. Let
def A (x) = {y ⊆ x | y is definable over x, ∈, A ∩ x},
incorporating A ∩ x as a unary relation for definitions. Now define


L 0 [ A] = ∅; L α+1 [ A] = def A (L α [ A]); L δ [ A] = L α [ A] for limit δ > 0;
α<δ

and


L[ A] = L α [ A].
α

For sets A, unlike for L(A) what remains of A is only A ∩ L[ A] ∈ L[ A], so that for example L[R] = L for the reals
R. L[ A] is more constructive since knowledge of A is incorporated through the hierarchy of definitions, and like L,
L[ A] satisfies the Axiom of Choice for every A. Levy pointed out an important absoluteness, that with A = A ∩ L[ A],
L[ A] = L[ A].
Shoenfield [111,112] had separately sketched a special case of this construction to establish partial results toward
the conditional independence Con(ZFC + V = L) implies Con(ZFC + GCH + V = L). Levy [8] then established
this result by refining Shoenfield’s argument to show that the full GCH can be proved in an appropriate L[ A].
The work of Hajnal and of Levy on constructibility elicited some interest in the correspondence between Paul
Bernays and Gödel [71, 151–5].2 Although differing in their formal presentations, since Hajnal and Levy both
worked with sets A of ordinals so that L[ A] = L(A), distinctions would surface only later. Because of its intrinsic
absoluteness, Levy’s construction L[ A] would become basic for the inner model theory of large cardinals. Hajnal’s
construction L(A) has also become basic, particularly with the constructible closure of the reals L(R) becoming the
focal inner model for the Axiom of Determinacy.

2. Reflection principles

Reflection has been an abiding motif in set theory, with its first appearance in a proof occurring in Gödel’s proof
of the GCH in L. Gödel himself saw roots in Russell’s Axiom of Reducibility and in Zermelo’s Axiom of Separation,

2 In later correspondence, Bernays (Gödel [71, 199]) referred Gödel to a review of Levy [1] by Shepherdson [110], but amusingly this review is
mistitled and is actually a review of another paper drawn from Levy’s dissertation, Levy [4].
236 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

writing [66, 178] that: “. . . since an axiom of reducibility holds for constructible sets it is not surprising that the
axioms of set theory hold for the constructible sets, because the axiom of reducibility or its equivalents, e.g., Zermelo’s
Aussonderungsaxiom is really the only essential axiom of set theory”. A heuristic appearing early on for reflection was
that any particular property attributable to the class of all ordinals, since its extent is not characterizable, should already
be attributable to some cardinal. This heuristic was at work in the early postulations of large cardinals, cardinals that
posit structure in the higher reaches of the cumulative hierarchy and prescribe their own transcendence over smaller
cardinals. The weakest of the now standard large cardinals are the inaccessible cardinals, those uncountable regular
cardinals κ such if α < κ, then 2α < κ, so that in the rank hierarchy Vκ models ZFC.
The formalization of reflection properties was one of the early developments of model-theoretic initiatives in set
theory. With the basic concepts and methods of model theory being developed by Tarski and his students at Berkeley,
Richard Montague [95] in his 1957 Berkeley dissertation had studied reflection properties in set theory and had shown
that Replacement is not finitely axiomatizable over Zermelo set theory in a strong sense. Levy [9,10] then exploited
the model-theoretic methods to establish the broader significance of reflection principles and the close involvement of
the Mahlo hierarchy of large cardinals.
The ZF Reflection Principle, drawn from Montague [95, 99] and Levy [9, 234], asserts that for any (first-order)
formula ϕ(v1 , . . . , vn ) in the free variables as displayed and any ordinal β, there is a limit ordinal α > β such that for
any x 1 , . . . , x n ∈ Vα ,
ϕ[x 1, . . . , x n ] iff ϕ Vα [x 1 , . . . , x n ],
where as usual ϕ M denotes the relativization of the formula ϕ to M. The idea is to carry out a Skolem closure argument
with the collection of subformulas of ϕ. Montague showed that the principle holds in ZF, and Levy showed that it is
actually equivalent to the Replacement Schema together with the Axiom of Infinity in the presence of the other axioms
of ZF. Through this work the ZF Reflection Principle has become well-known as making explicit how reflection is
intrinsic to the ZF system.
Levy [9] took the ZF Reflection Principle as motivation for stronger reflection principles. The first in his hierarchy
asserts that for any formula ϕ(v1 , . . . , vn ), there is an inaccessible cardinal α such that for any x 1 , . . . , x n ∈ Vα ,
ϕ[x 1, . . . , x n ] iff ϕ Vα [x 1 , . . . , x n ].
Levy showed that this principle is equivalent to the assertion that the class of inaccessible cardinals is definably
stationary, i.e. every definable closed unbounded class of ordinals contains an inaccessible cardinal. Paul Mahlo
[90–92] had studied what are now known as the weakly Mahlo cardinals, those regular cardinals κ such that the set of
smaller regular cardinals is stationary in κ, i.e. every closed unbounded subset of κ contains a regular cardinal. Levy’s
work thus established an evident connection between Mahlo’s cardinals and structural principles about sets. Levy
recast Mahlo’s concept by replacing regular cardinals by inaccessible cardinals. On the other hand, whereas Mahlo had
entertained arbitrary closed unbounded subsets, Levy’s principle is restricted to definable closed unbounded classes.
Be that as it may, it would be through Levy’s work that Mahlo’s cardinals would come into use in modern set theory
cast as the strongly Mahlo cardinals, those regular cardinals κ such that the set of smaller inaccessible cardinals is
stationary in κ.3
Levy proceeded to develop a hierarchy of reflection principles, the next principle being the one above with
“inaccessible” replaced by “strongly Mahlo”. Mahlo himself had developed a hierarchy of his cardinals, and Levy’s
work recast it as reflecting reflection: A reflection scheme is first formulated and is then itself reflected. In this way,
Levy showed how the iterative formalization of reflection illuminates Mahlo’s original scheme, formulated a half-
century before.
Levy also substantiated how various reflection principles have proof-theoretic transcendence over each other. He
had formulated the following, drawing on his dissertation work (cf. Levy [4]): For theories T0 ⊆ T in the same
language and subsuming enough of arithmetic to encode formal consistency, T is essentially reflexive over T0 if for
any sentence σ , T  σ → Con(T0 +σ ). This is an elegant formulation of the transcendence of one theory over another;
note that no consistent extension of T is finitely axiomatizable over T0 , since for any σ , if T0 + σ were to extend T ,
we would have T0 + σ  Con(T0 + σ ). Montague [95] had shown in effect that ZF is essentially reflexive over

3 See Kanamori [83, Sections 1, 6].

 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 237

Z, Zermelo set theory. Levy [10] considered “partial” reflection principles weaker than the ZF Reflection Principle
and studied [15] their minimal models of form Vα ; Levy–Vaught [12] showed that these partial principles are also
essentially reflexive over Z. In [11] Levy showed that each of the strong reflection principles in his [9] hierarchy à la
Mahlo is essentially reflexive over the previous. Moreover, Levy [11] showed that between any two of these reflection
principles there is a whole spectrum of theories each essentially reflexive over the previous.
In further ramifications, to a volume dedicated to Fraenkel on the occasion of his 70th birthday Levy contributed
a paper [14] that compared the Axiom of Choice with its global form, i.e. there is a class choice function for all sets,
and showed that the set consequences of the global form follows from his reflection principle down to inaccessible
cardinals. The elder Bernays [50] also contributed a paper to that volume, one in which, inspired by Levy’s work
on reflection, he developed reflection principles based on second-order formulas which were seen to subsume all of
Levy’s principles.
The ZF Reflection Principle was foreshadowed in Gödel’s remarks [67]; he there introduced the ordinal-definable
sets, and to develop their theory requires reflection in some form (cf. Gödel [69, 146]). In his expository article on
Cantor’s Continuum Problem, Gödel [68, 521] mentioned the Mahlo cardinals in connection with the proposal to
search for new large cardinal axioms that would settle the Continuum Hypothesis. Bernays cited the paper Levy
[9] in a letter of 12 October 1961 to Gödel (Gödel [71, 196ff]), and Gödel noted in a letter of 13 August 1965 to
Cohen (Gödel [71, 385ff]), in a discussion about evidence for inaccessible cardinals, that “Levy’s principle might be
considered more convincing than analogy [with the integers]”. What presumably impressed Gödel was how reflection,
a persistent heuristic in his own work, had been newly marshaled to account for Mahlo’s cardinals. Finally, Gödel
wrote in a letter of 7 July 1967 to Robinson (Gödel [72, 195]):

. . . I perhaps stimulated work in set theory by my epistemological attitude toward it, and by giving some
indications as to the further developments, in my opinion, to be expected and to be aimed at. I did not, however,
give any clues as to how these aims were to be attained. This has become possible only due to the entirely new
ideas, primarily of Paul J. Cohen and, in the area of axioms of infinity, of the Tarski school and of Azriel Levy.

3. Ackermann’s set theory

Ackermann [45] formulated a distinctive axiomatic theory of sets and classes, and this theory quickly came under
the scrutiny of Levy whose extended analysis did a great deal to bring it into the fold of the standard ZF axiomatization.
Much of the analysis was already present in Levy’s dissertation and was subsequently extended in Levy [5] and Levy–
Vaught [12]. Gödel wrote to Bernays in a letter of 30 September 1958 (Gödel [71, 155]): “Of the results announced
in the introduction to Levy’s dissertation, the most interesting seems to me to be that on Ackermann’s system of set
theory. That really looks very surprising”.
Ackermann’s theory A is a first-order theory that can be cast as follows: There is one binary relation ∈ for
membership and one constant V ; the objects of the theory are to be referred to as classes, and members of V as
sets. The axioms of A are the universal closures of:
(1) Extensionality: ∀z(z ∈ x ↔ z ∈ y) −→ x = y.
(2) Comprehension: For each formula ψ not involving t, ∃t∀z(z ∈ t ←→ z ∈ V ∧ ψ).
(3) Heredity: x ∈ V ∧ (t ∈ x ∨ t ⊆ x) −→ t ∈ V .
(4) Ackermann’s Schema: For each formula ψ in free variables x 1 , . . . , x n , z and having no occurrence of V ,
x 1 , . . . , x n ∈ V ∧ ∀z(ψ → z ∈ V ) −→ ∃t ∈ V ∀z(z ∈ t ↔ ψ).
This last, a comprehension schema for sets, is characteristic of Ackermann’s system. It forestalls Russell’s Paradox,
and its motivation was to allow set formation through properties independent of the whole extension of the set concept
and thus to be considered sufficiently definite and delimited.
Ackermann [45] himself argued that every axiom of ZF, when relativized to V , can be proved in A. However,
Levy [5] found a mistake in Ackermann’s proof of the Replacement Schema, and whether Replacement can be derived
from Ackermann’s Schema would remain an issue for some time. Toward a closer correlation with ZF, Levy came to
the idea of working with
A∗ : A together with the Axiom of Foundation relativized to V .
238 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

As for ZF, Foundation focuses the sets with a stratification into a cumulative hierarchy. Levy [5] showed that, leaving
aside the question of Replacement, A∗ establishes substantial reflection principles. On the other hand, he also showed
through a sustained axiomatic analysis that for a sentence σ of set theory (so without V ) : If σ relativized to V is
provable in A∗ , then σ is provable in ZF. The thrust of this work was to show that Ackermann’s Schema can be
assimilated into ZF — this is presumably what Gödel found surprising — and that ZF and A∗ have almost the same
theorems for sets.
Levy–Vaught [12] later observed by an inner model argument that, as for ZF and Foundation, if A is consistent,
then so is A∗ . They then went on to confirm that the addition of Foundation to A was substantive; they showed
that Ackermann’s Schema is equivalent to a reflection principle in the presence of the other A∗ axioms, and that A∗
establishes the existence of {V } and the power classes P(V ), P(P(V )), and so forth.
Years later, returning to the original issue about Replacement, William Reinhardt [103] in his 1967 Berkeley
dissertation under the supervision of Vaught built on Levy–Vaught [12] to establish for A∗ what Ackermann could
not establish for A: Every axiom of ZF, when relativized to V , can be proved in A∗ . Thus, A∗ and ZF do have exactly
the same theorems for sets. Reinhardt also developed a theory of natural models of A∗ ; these are connected to the
indescribable cardinals (see Section 9) and led to further large cardinal postulations.4

4. Levy hierarchy and absoluteness

In his first work in a distinctive direction from his dissertation, Levy in [6], and much later in full exposition [28],
formulated the now standard hierarchy of first-order formulas of the language of set theory. He showed that
the hierarchy provides the scaffolding for an efficacious analysis of logical complexity, getting to a substantial
absoluteness result that cast reflection in a new light.
For formulating his hierarchy Levy struck on the key idea of discounting bounded quantifiers, those that can be
rendered as ∀v ∈ w or ∃v ∈ w, an idea perhaps novel at the time in set theory but now subsumed into its modern
sensibilities. There was an antecedent in the discounting of the bounded numerical quantifiers ∀k < n and ∃k < n
in Stephen Kleene’s [86] formulation of the arithmetical hierarchy over the recursive predicates, but the motivations
were rather different, and Levy had to make a conceptual leap because of the arbitrariness of sets.
In brief, a formula of set theory is Σ0 and Π0 in the Levy hierarchy if its only quantifiers are bounded. Recursively,
a formula is Σn+1 if it is of the form ∃v1 . . . ∃vk ϕ where ϕ is Πn , and Πn+1 if it is of the form ∀v1 . . . ∀vk ϕ where
ϕ is Σn . The classification of definable concepts in this hierarchy depends on the governing theory. For a set theory
T , a formula ϕ is ΣnT iff for some Σn formula ϕ  , T  ϕ ↔ ϕ  ; and similarly for ΠnT . ΣnZF and ΠnZF formulas are
equivalent to formulas with blocks of like quantifiers contracted into one through applications of the Pairing Axiom.
Also, bounded quantification does not add to complexity in ZF: If ϕ is ΣnZF (respectively, ΠnZF ), then so is ∃v ∈ wϕ
and ∀v ∈ wϕ. This depends on Replacement to “push” the bounded quantifiers to the right and is a crucial point about
the Levy hierarchy. Finally, that Σ0T formulas are wide-ranging yet absolute for transitive models of weak set theories
T has become a basic feature of the semantic analysis of set theory.
Levy [6] pointed out that his hierarchy is proper in ZF, i.e. there are formulas in ΠnZF − ΣnZF and in ΣnZF − ΠnZF
for every n > 0, and that ZF establishes the consistency, for any particular n, of Zermelo set theory plus Replacement
restricted to Σn formulas. Levy [25,28] worked out for each n > 0 a Σn (respectively, Πn ) satisfaction formula for
the Σn (respectively, Πn ) formulas and thereby got careful hierarchy results. The antecedent was the Σn0 universal
predicate for the Σn0 predicates in the Kleene arithmetical hierarchy, built directly on the normal forms of recursive
predicates. Levy laid out satisfaction sequences à la Tarski level-by-level, once again drawing metamathematical
methods into set theory.
The main advance of Levy [6] was a now well-known and basic absoluteness result. Shoenfield [113] had
established an absoluteness result seminal for modern descriptive set theory; he showed that, as we now say, every

12 set of reals is ω1 -Suslin in a constructible way, and concluded in particular that every (lightface) Σ21 set of natural
numbers is in L. As detailed in [28] Levy wove in the Shoenfield idea to establish in ZF together with the Axiom of
Dependent Choices (DC) that any sentence (without parameters) Σ1 in his hierarchy, if holding in V , also holds in L.
More formally, we have the Shoenfield–Levy Absoluteness Lemma:

4 See Kanamori [83, Section 23].

 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 239

For any Σ1 sentence σ , ZF + DC  σ ↔ σ L .

Levy readily concluded that any Σ1 or Π1 theorem of ZF + V = L is already a theorem of ZF + DC, so that any
uses of e.g. GCH in a proof of such a sentence can be eliminated. Levy [6] actually pointed out that L can be replaced
by a countable L γ fixed for all σ , so that any Σ1 sentence is absolute for every transitive M ⊇ L γ .5 Levy’s proof,
starting with a Σ1 sentence σ , first appealed to DC to reflect down from the universe to a countable transitive model
of σ . He then got a countable L α modeling σ by applying Shoenfield’s main idea of relying on the absoluteness of
well-foundedness, i.e. the equivalence of no infinite descending chains and the existence of a ranking. One can view
the Shoenfield and Levy absoluteness results as two sides of the same coin, one in the context of descriptive set theory
and the other in the context of general set theory, with either one readily leading to the other.
The Shoenfield–Levy Absoluteness Lemma can be seen as an effective refinement of the ZF Reflection Principle
that reflects a restricted sentence down to some countable L γ , and as such it would find wide use in effective contexts
like admissible set theory. Even just Levy’s initial reflection down, in effect into the domain of hereditarily countable
sets, would become basic to admissible set theory as the Levy Absoluteness Principle. In his book on admissible set
theory Barwise [47, 77] wrote: “One of the main features of this book (at least from our point of view) is the systematic
use of the Levy Absoluteness Principle to simplify results by reducing them to the countable case”.

5. Independence with urelements

From the beginning Levy had a steady interest in the independence of choice principles and in the pre-Cohen
era established penetrating results based on the Fraenkel–Mostowski method. To establish the independence of AC,
Fraenkel had come to the fecund idea of starting with urelements, objects without members yet distinct from each
other; building a model of set theory by closing off under set-theoretic operations; and exploiting automorphisms
of the model induced by permutations of the urelements. Fraenkel [59] in one construction started with urelements
A = {an | n ∈ ω} and considered a generated model in which for any set x there is a finite s ⊆ A with the following
property: x is fixed by any automorphism of the model induced by a permutation of A that fixes each member of s
and at most interchanges pairs within the cells {a2n , a2n+1 } for n ∈ ω. There can then be no choice function for the
countable set of pairs {{a2n , a2n+1 } | n ∈ ω} in the model, since for any purported such function one can take some
a2n , a2n+1 not in its support and apply a permutation interchanging them.
Andrzej Mostowski [97] developed Fraenkel’s constructions by imposing algebraic initiatives. First, the set of
urelements can be structured e.g. with an ordering; second, a model is built based on invariance with respect to a
specified group of permutations, group in the algebraic sense with respect to composition; and third, supports of sets
are closely analyzed, a support of a set to be a set of urelements such that if a permutation fixes each, the induced
automorphism also fixes the set. Mostowski in particular established that there is a model in which AC fails but the
Ordering Principle (OP) holds, where:
(OP) Every set can be linearly ordered.
He began with a countable set of urelements ordered in the ordertype of the rationals; built a model based on the
group of order-preserving permutations; established that every set has a ⊆-least finite support; and showed by these
means that the set of urelements cannot be well-ordered yet the model itself has a class linear ordering. Levy’s initial
work [2,3,7] in this direction was directly based on Mostowski’s model for OP.
Coming into his own, Levy [16] constructed a model in which Cn holds for every natural number n yet C<ℵ0 fails,
where
(Cκ ) Every set consisting of sets of cardinality ≤ κ has a choice function
(C<κ ) Every set consisting of sets of cardinality < κ has a choice function
and moreover the Axiom of Multiple Choice (MC) holds, where
(MC) For any set x there is a function f on x such that for any non-empty y ∈ x,
f (y) is a non-empty finite subset of y.

5 In modern terms, γ can be taken to be the least stable ordinal, where δ is stable iff L ≺ L, i.e. L and L satisfy the same Σ formulas with
δ 1 δ 1
parameters from L δ .
240 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

Levy began with a countable set of urelements A partitioned as k∈ω Pk , where Pk = {a1k , . . . , a kpk } with pk the kth
prime. Let πk be that permutation of A fixing every member of A − Pk and such that πk (aik ) = ai+1 k for 1 ≤ i < pk
and π(a pk ) = a1 . Levy then took the group of permutations generated by the πk and generated a model based on finite
k k

supports. As in the Fraenkel model described above, the set {Pk | k ∈ ω} does not have a choice function so C<ℵ0 fails.
That MC holds Levy affirmed with an argument also applicable to the Fraenkel model. Finally, the specifics of Levy’s
model came into play when he showed with algebraic arguments about cyclic permutations that Cn holds for every
natural number n. That OP implies C<ℵ0 is simple to see, so that OP must fail in Levy’s model. Hans Läuchli [88]
built another model in which C<ℵ0 holds yet OP fails.
Notably, it was later observed that in ZF, MC actually implies AC.6 Thus, Levy’s work shows that having
urelements can separate these principles. Levy [17] subsequently applied his [16] model in a considerable analysis
of some graph-theoretic propositions studied by Mycielski. Also, Levy [26] developed transfinite versions of his
algebraic methods and argued e.g. that C<κ does not imply Cκ for limit alephs κ.7 However, this seems to be the
single instance when Levy was proved wrong, but even the error stimulated results.8
Levy [22] established independence results for various choice principles indexed by alephs κ (cf. Jech [81, 119ff]):
(DCκ ) Suppose that x is a set and r a binary relation such that for every α < κ and s: α → x there is a y ∈ x
satisfying s r y. Then there is a function f : κ → x such that f |α r f (α) for every α < κ.
(ACκ ) Every set x with |x| = κ has a choice function.
(Wκ ) Every set x is comparable with κ, i.e. |x| ≤ κ or κ ≤ |x|.
With DCκ Levy generalized the Axiom of Dependent Choices, which is DCℵ0 . Wκ generalizes the proposition that
every infinite set has a countable subset, which is Wℵ0 .9 DCκ implies both ACκ and Wκ . Levy was expanding on work
of Mostowski [98], who showed that there is a model satisfying ¬ACℵ1 + DCℵ0 (and, as noticed later, ¬Wℵ1 ).
After drawing implications among these principles for various κ, Levy established several independences. He
constructed a basic model by starting with λ urelements, considering all permutations, and working with supports
of cardinality < λ. For a singular cardinal, taking λ = ℵω1 as a typicality one gets ¬ACℵ1 + DCℵ0 + ∀κ <
ℵω1 (Wκ ) + ¬Wℵω1 . For a successor cardinal, taking λ = ℵ1 one gets:
∀κ(ACκ ) + DCℵ0 + ¬Wℵ1 ( and so ¬DCℵ1 ).
In particular, Well-ordered Choice ∀κ(ACκ ) does not imply DCℵ1 . Jensen [82] later established the surprising result
∀κ(ACκ ) implies DCℵ0 , that Well-ordered Choice actually implies Dependent Choices.
Levy constructed an interesting, second model, assuming 2ℵ0 = ℵ1 and starting with a set of urelements ordered
in the ordertype of the reals. He then used the group of order-preserving permutations and supports generated by the
“Dedekind cuts” (−∞, r ) to get:
∀κ(ACκ ) + ¬DCℵ1 + Wℵ1 + ¬Wℵ2 .
Thus, DCℵ1 and Wℵ1 were separated.
Already in the work on conditional independence results via relative constructibility there was an air of anticipation
about possible independences from ZF. This became palatable in the work on choice principles by Fraenkel–
Mostowski methods with, e.g. Levy [22, 145] writing: “Even the independence of the axiom of choice itself is still an
open problem for systems of set theory which do not admit urelements or non-founded sets. Thus we can hope, for the
time being, to prove the above mentioned independence results only for a set theory which admits either urelements
or non-founded sets”. Of course, independence with respect to ZF was what was really wanted, and this would come
about in a dramatic turn of events.

6 See Jech [81, 133]; the observation was first made by David Pincus in his 1969 Harvard dissertation. Working in the post-Cohen era, Pincus’s
results there (cf. Pincus [101]) also showed how to “transfer” Levy’s result to get the consistency relative to ZF of Cn holds for every natural
number n and C<ℵ0 fails.
7 Levy [26] was a summary of Fraenkel–Mostowski methods given at a 1963 symposium; in that summary (p. 225) Levy pointed out the open
problem of whether 2 · m = mfor all infinite cardinals mimplies AC, and eventually his student Gershon Sageev [104] established that it does not.
8 Paul Howard [79] later established that in every Fraenkel–Mostowski model, C
<ℵ0 already implies “C∞ ”, i.e. that every set consisting of
well-orderable sets has a choice function. Pincus [102] and Sageev [104] independently established the ZF independence of Cℵ0 from C<ℵ0 .
9 Actually, Levy [22] worked with a more involved proposition H(κ), which has the property that W is equivalent to H(κ) together with
κ
∀λ < κ(Wλ ).
 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 241

6. ZF independence

Paul Cohen [52,53] in 1963 established the independence of AC from ZF and the independence of the CH from
ZFC. That is, complementing Gödel’s relative consistency results with L Cohen established that Con(ZF) implies
Con(ZF + ¬AC) and that Con(ZFC) implies Con(ZFC + ¬CH). These were, of course, the inaugural examples of
forcing, a remarkably general and flexible method, with strong intuitive underpinnings, for extending models of set
theory. If Gödel’s construction of L had launched set theory as a distinctive field of mathematics, then Cohen’s method
of forcing began its transformation into a modern, sophisticated one. Cohen’s particular achievement lies in devising
a concrete procedure for extending well-founded models of set theory in a minimal fashion to well-founded models of
set theory with new properties but without altering the ordinals. Set theory had undergone a sea-change, and beyond
simply how the subject was enriched, it is difficult to convey the strangeness of it.10
Cohen circulated a preprint [51] in April and soon started to give talks on his results. Already evident in the
preprint were two features of forcing that resonated with Levy’s work: Cohen relied on relative constructibility to
extend a model of V = L to one satisfying GCH + V = L, and Cohen used in effect a reflection argument to
show that his beginning with a countable ∈-model of ZF was formally unnecessary for getting consistency results
relative to ZF. Levy first heard the details of Cohen’s results at a July model theory conference at Berkeley, and later
that summer fully assimilated forcing working with Solomon Feferman. Feferman had been extensively consulted by
Cohen at Stanford when he was coming up with forcing, and Feferman [56,57] was the first after Cohen to establish
results by forcing. In quick succession several abstracts appeared in the Notices of the American Mathematical Society:
Levy [18], Feferman–Levy [19], Levy [20], and Levy [21], all received September 3, 1963 and soon after, Halpern–
Levy [23], which was actually noted as received August 29, 1963. With this work Levy became the first after Cohen
himself to exploit forcing in a sustained fashion to establish a series of significant results. We attend to the AC
independence results through the rest of this section; the others, on definability, were seeds for papers which are
discussed in the next section. Throughout, we assume familiarity with forcing and give only scant descriptions of the
models.
Cohen had shown how to “collapse” a cardinal, i.e. adjoin a generic bijection to a smaller ordinal. Feferman–
Levy [19] started with a model of ZFC + GCH and adjoined separate generic bijections between ℵn and ℵ0 for every
natural number n. In the resulting generated model the former ℵω becomes the new ℵ1 so that it is singular, and
moreover the reals are a countable union of countable sets. ACℵ0 thus fails, and it is seen in a drastic way how that
principle is necessary to develop Borel and Lebesgue measure. In Cohen’s own exposition [55, 143] the Feferman–
Levy model was presented in some detail.
James Halpern [75,76] in his 1962 Berkeley dissertation11 had shown that the Boolean Prime Ideal Theorem (BPI)
holds in the Mostowski [97] model for OP, where:
(BPI) Every Boolean algebra has a prime ideal.
The significance here is that BPI had become a focal choice principle, one that implies OP. Levy [17] had dealt with
BPI, observing that it fails in his [16] model. Halpern’s argument required a new sophistication, with a Ramsey-type
partition theorem being brought into play.
When Cohen’s work appeared, Halpern–Levy [23] showed that in the original Cohen model for the independence
of the AC, OP and in fact the Kinna–Wagner Selection Principle (KW) holds, where:
(KW) For any set x there is a function f such that whenever y ∈ x has at least
two elements, f (y) is a non-empty proper subset of y.
Kinna–Wagner [85] had formulated KW, a weak form of MC which they showed equivalent to: For any set x there
is an ordinal α and a bijection between x and a subset of P(α). KW implies OP, and though Halpern–Levy [23]
emphasized the new independence of AC from OP, there are two observations about their getting KW which are
worth mentioning: As noted earlier, the stronger MC implies AC in ZF. Also, KW was effectively unattainable through
Fraenkel–Mostowski methods, as Mostowski [99] had observed that KW fails in every Fraenkel–Mostowski model in
which the urelements are not well-orderable.

10 See Moore [96] on the origins of forcing and for what follows.
11 Halpern [76] wrote that his dissertation was written under the “supervision of” Levy; Levy visited Berkeley in the academic years 1959–1961.
242 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

Cohen’s original model for the independence of AC was the result of adjoining countably many Cohen reals and
the set x consisting of these, so that x has no well-ordering in the resulting model. Halpern–Levy [23] in effect
argued in the Cohen model with the Cohen reals acting like urelements. Cohen [54, 40] moreover acknowledged
the similarities between his AC independence result and the previous Fraenkel–Mostowski models. In any case, the
revelatory Halpern–Levy work initiated the process of “transferring” consistency results with Fraenkel–Mostowski
models to ZF consistency results via forcing by correlating urelements with generic sets.
For the BPI, there was no routine transfer of the Halpern [76] independence of AC from BPI. Levy saw the need for
a strengthened, “tree” Ramsey-type partition theorem to effect a ZF independence result. Halpern–Läuchli [77] then
duly established this result. Finally, Halpern–Levy [35] by 1966 had established that in Cohen’s original model, BPI
holds. This Halpern–Läuchli–Levy collaboration established a new level of sophistication in effecting a transfer from
the Fraenkel–Mostowski context to the Cohen one. Work at this level would soon be pursued by Jech, Sochor, Pincus,
and others, and the Halpern–Läuchli partition theorem would lead to an important extension by Richard Laver [89],
one also applied to forcing.

7. ZF definability

The abstracts Levy [18,20,21] had to do with formal definability, and the papers Levy [27,34] provided extended
accounts in a context of appropriate generality for the proofs. Levy probed the limits of ZFC definability, establishing
consistency results about definable sets of reals and well-orderings and in descriptive set theory.
Heralded by Levy [18], Levy [27] established the relative consistency of ZFC + GCH together with there being a
non-constructible real yet every definable set is constructible. Here, “definable” meant the broad notion of hereditarily
ordinal-definable. A set x is ordinal-definable iff there is a formula ψ(v0 , . . . , vn ) in the free variables as displayed
and ordinals α1 , . . . , αn such that x = {y | ψ[y, α1 , . . . , αn ]}. A set x is hereditarily ordinal definable iff the transitive
closure of {x} is ordinal definable. The ordinal-definable sets were introduced by Gödel [67] as mentioned in Section 2,
and their theory was developed by John Myhill and Dana Scott by 1964 (cf. Myhill–Scott [100]) with explicit appeal
to the ZF Reflection Principle, with which one can replace the informal satisfaction of x = {y | ψ[y, α1 , . . . , αn ]} by:
for some Vα and ordinals α1 , . . . , αn ∈ Vα , x = {y ∈ Vα | ψ Vα [y, α1 , . . . , αn ]}. OD denotes the (thus definable) class
of ordinal-definable sets, and HOD, the class of hereditarily ordinal-definable sets.
Feferman [56,57] had shown that in Cohen’s model which is the result of starting from a model of V = L and
adjoining a Cohen real there is no set-theoretically definable well-ordering of the reals, i.e. no formula in two free
variables that defines such a well-ordering. Levy [27] showed that the model actually satisfies V = L = HOD, so
that in particular there is no definable well-ordering of the reals even if ground model parameters are allowed in the
definition. The crux is that the partial order of conditions for adjoining a Cohen real is homogeneous in the sense that
for any pair p, q of conditions there is an automorphism e of the partial order such that p and e(q) are compatible.
Hence, for a formula ϕ(v1 , . . . , vn ) of the forcing language and sets x 1 , . . . , x n in the ground model, any condition
forces ϕ(x̌ 1 , . . . , x̌ n ) exactly when all conditions do. That HOD = L follows by induction on rank. Levy’s appeal to
homogeneity and automorphism, related to his earlier work with urelements and realigned by the work of Cohen and
Feferman, would become a basic motif that connects forcing with definability.
In an eventual sequel, Levy [34] worked out his main delimitative results. He first considered Cohen’s model which
is the result of starting with a model of V = L and collapsing ω1L , i.e. adjoining a generic bijection between ω and
ω1L . As before one has V = L = HOD, but this now easily implies that every OD well-ordering of reals is at most
countable. This confirmed an announcement in Levy [18].
With his next theorem Levy [34] provided an important delimitation for descriptive set theory, confirming an
announcement in Levy [20]. Classical descriptive set theory, in its probing into the first levels of the projective
hierarchy, had pushed against the limits of axiomatic set theory.12 Levy presumably had assimilated this work in
large part from John Addison at Berkeley (cf. Addison [46]), but in any case Levy quickly saw how to apply forcing
to illuminate the central issue of uniformization. For binary relations A and B,
A is uniformized by B iff
B ⊆ A ∧ ∀x(∃y(x, y ∈ A) −→ ∃!y(x, y ∈ B)).

12 See Kanamori [83, Sections 12–14] for the background and basic concepts of descriptive set theory.
 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 243

∃! abbreviates the formalizable “there exists exactly one”, and so this asserts that A can be refined to a function B.
That every relation can be uniformized is a restatement of the Axiom of Choice. A high point of classical descriptive
set theory was the result of Motokiti Kondô [87] that, in terms of the projective hierarchy, every 11 relation on reals
can uniformized by a 11 relation. This implied via projection that every
12 relation on reals can be uniformized
by a
12 relation, and by looking at complements that not every 12 relation on reals can be uniformized by a 12
relation. Whether every 12 relation on reals can be uniformized by a projective relation had remained open. Bringing
in axiomatics Addison [46] established that assuming V = L, for any n ≥ 2 every
1n relation on reals can be
uniformized by a
1n relation, so that in particular every 12 relation on reals, being
13 , can be uniformized by a
13
relation. In contradistinction, Levy established the relative consistency of there being a 12 on reals that cannot be
uniformized by any projective relation.
Levy considered Cohen’s model which is the result of starting with a model of V = L and adjoining (a sequence
of) ω1L Cohen reals. Taking the cue from ordinal-definability, say that a set x is real-ordinal-definable iff there is
a formula ψ(v0 , . . . , vn , vn+1 ) in the free variables displayed, a Vα , a real r ∈ Vα , and ordinals α1 , . . . , αn ∈ Vα
such that x = {y ∈ Vα | ψ Vα [y, α1 , . . . , αn , r ]}. That is, a real parameter is to be allowed in the definition. Levy
considered the relation A on reals defined by:  f, g ∈ A iff g ∈ / L[ f ]. This relation, formulated in terms of his
notion of relative constructibility, is 12 by an elaboration of an argument in Addison [46]. Suppose now that B is
a real-ordinal-definable relation on reals. By a definability argument the real parameter r in the definition can be
taken to be one coding countably many of the Cohen reals. There is then an s such that r, s ∈ A. If however there
were an s such that r, s ∈ B, then s, like B, would be real-ordinal-definable with real parameter r . But then, a
homogeneity argument shows that s ∈ L[r ]. Consequently, no real-ordinal-definable relation, and consequently no
projective relation, can uniformize A.

8. Levy collapse

The theory of large cardinals was revitalized by pivotal results in the early 1960s, and with Cohen’s forcing large
cardinals would enter the mainstream of set theory by providing hypotheses and methods to analyze strong set-
theoretic propositions. Levy’s earlier work on reflection principles had established a central place for Mahlo cardinals;
in the post-Cohen era Levy made basic contributions to the fast growing theory of large cardinals.
In the last of the 1963 abstracts Levy [21], Levy announced a result that depended on what we now call the Levy
collapse. In general terms, for infinite regular cardinals λ < κ, Col(λ, κ) is the partial order for adjoining a κ-sequence
of surjections λ → α for α < κ. If κ is inaccessible, then Col(λ, κ) has the κ-chain condition13; hence, κ becomes
the successor λ+ of λ in the generic extension. The forcing with Col(λ, κ) is then called a Levy collapse of κ to λ+ .
Already in the first flush of forcing Levy [21,34] used the Levy collapse of an inaccessible cardinal to ω1 to establish
the relative consistency of:
(∗) Every real-ordinal-definable well-ordering of reals is at most countable.
Forcing ω1 to be countable had led to the consistency of every OD well-ordering of reals being at most countable;
forcing every α below an inaccessible to be countable provides enough closure to achieve (∗). Levy also considered
the proposition:
(∗∗) Every real-ordinal definable set of reals is either countable or of cardinality 2ℵ0 .
The Levy collapse of an inaccessible to ω1 entails CH so that (∗∗) is vacuous, but Levy showed that in any further
extension where many Cohen reals adjoined (∗∗) continues to hold, this already inherent in the early abstract [21].
In deploying an inaccessible cardinal Levy was a pioneer in taking the modern approach to large cardinals: They
are not novel hypotheses burdened by ontological commitment but are the repository of means for carrying out
mathematical arguments. Cohen [55, 147] acknowledged this use of an inaccessible cardinal. As pointed out by
Levy [34, 131–2], either (∗) or (∗∗) implies that the ω1 (of the universe) is an inaccessible cardinal in the sense
of L. So, Levy’s work was party to the first instance of an important phenomenon in set theory, the derivation

13 Actually, that Col(λ, κ) has the κ-chain condition only requires, by a so-called delta-system argument, that α < κ implies that α <λ < κ. Full
inaccessibility is typically required in other parts of an argument using the Levy collapse.
244 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

of equiconsistency results based on the complementary methods of forcing and inner models.14 After this heady
introduction the Levy collapse would become standard fare in the theory of large cardinals, the way to render a large
cardinal accessible yet still with substantial properties to establish the relative consistency of strong combinatorial
propositions low in the cumulative hierarchy.
Levy’s model was used by Robert Solovay to establish a now famous relative consistency result. Solovay played
a prominent role in the forging of forcing as a general method, and he above all in this period raised the level
of sophistication of set theory across its breadth from forcing to large cardinals. Already in the spring of 1964
Solovay [116,117] exposed what standard of argument was possible when showing that if an inaccessible cardinal
is Levy collapsed to ω1 , every real-ordinal-definable set of reals is Lebesgue measurable, and proceeding to the
corresponding inner model HROD of the hereditarily real-ordinal-definable sets, that
HROD | DC + “Every set of reals is Lebesgue measurable”.
That this model satisfies Dependent Choices bolsters it as a bona fide one for mathematical analysis. Solovay thus
illuminated the classical measure problem of Henri Lebesgue with the modern technique of forcing. Solovay also
showed that the reals in this HROD model have several other substantial properties, one being the perfect set property:
Every set of reals is countable or else has a perfect subset. This refined Levy’s result with (∗∗) above, any perfect
set of reals having cardinality 2ℵ0 , and established the equiconsistency of the perfect set property with the existence
of an inaccessible cardinal.15 For quite some years, it was speculated that an inaccessible cardinal can be avoided for
getting all sets of reals to be Lebesgue measurable. However, in 1979 Shelah [108] established that DC + “Every set
of reals is Lebesgue measurable” implies that ω1 is inaccessible in L, vindicating Solovay’s use of the Levy collapse
for the Lebesgue measurability result.
Solovay [117, 2] announced a joint result with Levy which eventually appeared in Levy–Solovay [38]. Levy and
Solovay built on the structure uncovered by Solovay in the Levy collapse model to establish a further “regularity”
property about sets of reals, being the union of an ℵ1 -sequence of Borel sets. In the classical investigation of the
projective hierarchy, though the second level
12 seemed complicated, Sierpiński [114] had established that every
12
set of reals is the union of ℵ1 Borel sets. Of course, if CH holds, then every set of reals is the union of ℵ1 Borel
sets, namely the singletons of its members. It came to light with the emergence of Martin’s Axiom that ¬CH together
with the converse of the Sierpiński result, that every union of ℵ1 Borel sets is
12 , is relatively consistent (Martin–
Solovay [93, Section 3]).
Considering arbitrary sets of reals, Levy and Solovay first noted with a simple construction via transfinite recursion
that ZFC + ¬CH implies that there is a set of reals which is not the union of any ℵ1 Borel sets, and then went on
to show that some substantial use of AC is necessary. As with Levy’s (∗∗), Solovay [117] had shown that, with V1
the Levy collapse model and V2 a further extension where many Cohen reals are adjoined, the propositions about
Lebesgue measurability and so forth for real-ordinal-definable sets that hold in V1 continue to hold in V2 , thereby
establishing the further relative consistency of having a large continuum. Levy–Solovay [38] showed that in any such
V2 , every real-ordinal-definable set of reals is the union of ℵ1 Borel sets. Specifically, such a set of reals is the union
of Borel sets coded in V1 , and with CH holding in V1 , there are only ℵ1 such Borel sets. Thus, Levy and Solovay
established the relative consistency of ZFC + ¬CH + “Every real-ordinal-definable sets of reals is the union of ℵ1
Borel sets”. Passing to the inner HROD model Levy and Solovay then established the relative consistency of DC +
“Every well-ordering of reals is at most countable” + “Every set of reals is the union of ℵ1 Borel sets”. In this way
the property of being the union of ℵ1 Borel sets was adjoined to the regularity properties of sets of reals illuminated
by the Levy collapse model.
9. Measurable and indescribable cardinals
In addition to their collaborative work on the Levy collapse model, Levy [24] and Solovay [115] independently es-
tablished a general result about large cardinals that would become much cited in connection with the Continuum Prob-
lem. As is well-known, Gödel [68, 520] had speculated that large cardinal postulations might decide CH. He himself

14 See Kanamori [83, 135ff].

15 Solovay [117, 45ff] wrote: “Our proof that every [real-ordinal-definable] subset of [the reals] is countable or contains a perfect subset is,
essentially, a slight refinement of the following result of Levy [34]: Every uncountable [real-ordinal-definable] subset of [the reals] has power 2ℵ0 ”.
Compare Levy [34, 140ff].
 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 245

took CH to be false and made remarks amounting to the observation that those large cardinals consistent with V = L
cannot disprove CH. Addressing this issue, Levy–Solovay [29] showed that measurable cardinals κ remain measurable
in “mild” forcing extensions, those via partial orders of cardinality < κ. That this betokened what would become a
widely applied observation, that inaccessible large cardinals retain their characteristic properties in mild forcing exten-
sions, is often regarded as springing from Levy–Solovay [29] though they themselves wrote that this was well-known
for many large cardinals. At the time there was a particular point in that Scott [106] had dramatically established
that the existence of a measurable cardinal contradicts V = L, and Levy–Solovay [29] pointed out that measurable
cardinals, though loosened from the moorings of constructibility, cannot decide CH and other issues like Suslin’s
Hypothesis. In years to come, the growing success of the theory of large cardinals led to more and more allusions to
“Gödel’s Program”. The Levy–Solovay result would be consistently cited as a watershed, if only to point to a delimita-
tion to be superseded by other, more subtle invocations of large cardinals in connection with the Continuum Problem.
Levy’s last major contribution to the theory of large cardinals dealt with natural extensions of his earlier reflection
principles (cf. Section 2) set in a higher order context of Πnm and Σnm formulas.16 William Hanf and Scott in their
abstract [78] considered higher-order reflection properties for structures Vκ , ∈, R where κ is a cardinal and R ⊆ Vκ
and thereby provided a hierarchical scheme for large cardinals. For Q either Πnm or Σnm ,
κ is Q-indescribable iff for any R ⊆ Vκ and Q sentence ϕ
such that Vκ , ∈, R | ϕ, there is
an α < κ such that Vα , ∈, R ∩ Vα  | ϕ.
Including R ⊆ Vκ suffices to bolster the concept to accommodate general relational structures; as Vκ is closed under
pairing, the definition is equivalent to one where R is replaced by any finite number of finitary relations.17 Hanf–
Scott [78] observed that, with πnm the least Πnm -indescribable cardinal and σnm the least Σnm -indescribable cardinal, for
m > 0: πnm < πn+1m and if n > 0, then π m is not Σ m -indescribable. They also pointed out that the Π 1 -indescribable
n n 1
cardinals are exactly the weakly compact cardinals and that measurable cardinals are Π12 -indescribable. This provided
probably the earliest proof that below a measurable cardinal there are many weakly compact cardinals. Vaught [119]
subsequently pointed out that below a measurable cardinal there is cardinal Πnm -indescribable for every m, n ∈ ω. The
evident connection between Levy’s earlier reflection principles and the indescribable cardinals had an interconnecting
node in the work of Bernays [50], who had extended Levy’s principles by positing, in effect, the Πn1 -indescribability
for every n ∈ ω of the class of all ordinals.
Levy [36] carried out a systematic study of the sizes of indescribable cardinals, extending aspects of a combinatorial
study of large cardinals in Keisler–Tarski [84]. The starting point of Levy’s approach was that various large cardinal
properties are not only attributable to cardinals, but to their subsets. For X ⊆ κ and Q either Πnm or Σnm ,
X is Q-indescribable in κ iff for any R ⊆ Vκ and Q sentence ϕ
such that Vκ , ∈, R | ϕ, there is
an α ∈ X such that Vα , ∈, R ∩ Vα  | ϕ.
This leads to the consideration of {X ⊆ κ | κ − X is not Q -indescribable in κ}, which when κ is Q-indescribable is
a (proper) filter, the Q-indescribable filter over κ.18
Using universal satisfaction formulas Levy showed that these definable filters have a crucial property: For m, n > 0
and Q either Πnm or Σnm , the Q-indescribable filter over κ is normal. Recall that a filter F over acardinal κ is normal
iff it is closed under diagonal intersections, i.e. whenever {X α | α < κ} ⊆ F, {ξ < κ | ξ ∈ α<ξ X α } ∈ F. The
previously known normal filters were the closed unbounded filters over regular uncountable cardinals and normal

16 Higher-order languages have typed variables of every finite type (or order), quantifications of these, and beyond the atomic formulas specified
by the language, X ∈ Y and X = Y for any typed variables X and Y . In the intended semantics, if D is the domain of a structure, type 1 variables
play the usual role of first-order variables, type 2 variables range over P(D), and generally, type i + 1 variables range over P i (D) where P i
denotes i iterations of the power set operation. A formula is Πnm iff it starts with a block of universal quantifiers of type m + 1 variables, followed
by a block of existential quantifiers of type m + 1 variables, and so forth with at most n blocks in all, followed afterward by a formula containing
variables of type at most m + 1 and quantified variables of type at most m. A formula is Σnm iff it starts instead with existential quantifiers. Of
course, formulas containing only type 1 variables can be construed as the usual first-order formulas. This classification of formulas is cumulative
because of the “at most”: any Πnm or Σnm formula is also Πsr and Σsr for any r > m, or r = m and s > n.
17 Hanf–Scott [78] formulated their concept for inaccessible κ and with κ in place of V , but the difference is inessential as the R can code V .
κ κ
18 Levy himself called the members of this filter weakly Q-enforceable at κ, but we follow the formulation in Kanamori [83, Section 6].
246 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

ultrafilters found over a measurable cardinal. Levy further established that {α < κ | α is P-indescribable in α} is in
the Q-indescribable filter over κ, where: (a) P is Πn1 and Q is Πn+11 ; or (b) m > 1 and n > 0, P is Σ m , and Q is
n
Πn ; or (c) m > 1 and n > 0, P is Πn , and Q is Σn . These various results showed in concert that, just as normal
m m m

ultrafilters over a measurable cardinal provide intrinsic senses to how large measurable cardinals are as had been
shown in Keisler–Tarski [84], e.g. normal ultrafilters are closed under Mahlo’s Operation, so too do indescribable
cardinals have inherent transcendence over smaller cardinals, specifically those in the indescribable hierarchy itself.
The technique of ascribing large cardinal properties of cardinals also to their subsets has become an important part of
large cardinal theory and has been used in particular by James Baumgartner [48] to establish important hierarchical
results about the n-subtle and n-ineffable cardinals.

10. Capstones

In the later 1960s, Levy capped his investigations in various directions with papers reflective of previous themes
and techniques but also distinctive in how they resolve basic issues in axiomatics.
Levy’s last result applying the Fraenkel–Mostowski method concerned Cantor’s very notion of cardinality. The
problem of cardinal definability in set theory is how to assign to every set x a set |x| such that for every x, y we
have |x| = |y| iff x ≈ y, i.e. there is a bijection between x and y. Of course, with AC the initial (von Neumann)
ordinals construed as Cantor’s alephs serve as such |x|. But even without AC, one can use the “trick” of Scott [105] to
formulate |x| as the set of sets of least rank bijective with x. Levy [32], in a 1966 conference proceedings, established
that relative to ZF, ZF − Foundation + “There is no set-theoretic term τ such that for every x, y we have τ (x) = τ (y)
iff x ≈ y” is consistent in several strong senses.19 In this distinctive setting without Foundation the interplay between
urelement constructions and forcing is not pertinent. It had been known for over a decade that the Fraenkel–Mostowski
method with urelements can be recast, following Specker [118] and Mendelson [94], in ZF − Foundation with sets
a = {a} in the role of urelements. Levy ultimately relied on this recasting, but worked directly with urelements
and automorphisms. In one model he used ℵω urelements and generated an inner model in which they become a
proper class; in another, he proceeded similarly but started from Mostowski’s model for OP.20 With the appearance
of the forcing method, one might have thought that Fraenkel–Mostowski methods would be superseded, but in the
years to come, there would be a continuing cottage industry investigating Fraenkel–Mostowski models as intrinsically
interesting constructions in their own right.21
Levy and Georg Kreisel in their [30] provided a detailed exposition that established a central place for proof-
theoretic reflection principles in the comparative investigation of theories. Levy brought together his work on the
transcendence of theories through set-theoretic reflection principles, and the inimitable Kreisel, whose hand is evident
in the sections with the many italicizations, brought to bear his initiatives in the proof theory of arithmetic and analysis.
The main unifying motif was the proof-theoretic Uniform Reflection Principle:
(URP(S)) ∀ p∀n(ProvS ( p, s( ϕ  , n)) −→ ϕ(n)).
Intended for theories sufficient to carry out Gödel numbering, ProvS (x, y) is to assert that x is the Gödel number of a
proof in the theory S of the formula with Gödel number y; s( ϕ  , n) is the Gödel number of the sentence obtained by
substituting the numeral of the natural number n for the one free variable of ϕ; and finally, the uniformity has to do
with having the parametrization with the numerical variable n. URP(S) is an assertion of soundness; instantiating to a
ϕ refutable in S, URP(S) implies that something is not provable and hence the formal consistency of S. As Kreisel–
Levy [30, Section 1] pointed out, URP(S) actually subsumes both the assertion of ω-consistency for S and a general
form of induction.
Kreisel and Levy established a strong, general result about how URP(S) leads to transcendence over S in terms of
quantifier complexity: If U is a theory in the same language as S and URP(S) is a theorem of U, then for no set Σ
consisting of sentences of bounded quantifier complexity are the theorems of U provable in S + Σ . For set theories,

19 This was also done, in a strong sense, by Robert Gauntt [61].

20 Pincus [102] later addressed the issue of cardinal representatives, i.e. having a set-theoretic term τ such that for every x, y we have τ (x) = τ (y)
iff x ≈ y, and moreover |x| ≈ x. Transferring from Mostowski’s model for OP Pincus showed that relative to ZF it is consistent to have ZF +
“There are no cardinal representatives”. Rather surprisingly, he also showed by an iterated forcing argument that relative to ZF it is consistent to
have ZF + ¬AC + “There are cardinal representatives”.
21 See Howard–Rubin [80].
 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 247

Σn in the Levy hierarchy typifies a set of formulas of bounded complexity. Levy [6] had already announced a result
that implied that for any natural number n, no consistent extension of ZF can be obtained by adjoining to Zermelo set
theory any set of Σn sentences. As noted, various results from Levy’s previous work [4,9,11], which had the above
form except that the Σ was a finite set of sentences, could now be strengthened. The general result was moreover
applicable to Peano Arithmetic, Second-Order Arithmetic (Analysis), and the like to show that these theories cannot
be axiomatized over weaker theories using any set of axioms of bounded quantifier complexity.
In the last sections, Kreisel–Levy [30] established direct connections between URP(S) and schemas of transfinite
induction. By the classical work of Gerhard Gentzen [62], Peano Arithmetic (PA) establishes the coded schema
of transfinite induction up to any particular ordinal less than 0 , the least ordinal α such that α ω = α in ordinal
arithmetic, yet PA does not establish the schema of transfinite induction up to 0 itself. Kreisel and Levy showed
that over PA, URP(PA) is equivalent to transfinite induction up to 0 , and that over Second-Order Arithmetic Z1 ,
URP(Z1 ) is equivalent to transfinite induction up to 0 . These results provided elegant characterizations that connect
two formulations of the consistency of well-known theories.
Levy [25,37] investigated the logical complexity, in terms of his hierarchy of formulas, of basic statements of set
theory like AC, GCH, and V = L. His result on AC typifies the articulation and argumentation, made possible by
forcing. The Axiom of Choice is evidently Π2 . Levy showed that AC is not Σ2 in the following strong sense: For any
Σ2 sentence σ , if ZF  σ → AC, then ZF  ¬σ . The following is the argument in brief:
Suppose that σ is Σ2 , say ∃x∀yϕ(x, y) with ϕ being Σ0 , and ZF  σ → AC. The following can then be formalized
to establish ZF  ¬σ : Assume to the contrary that there is a set x 0 such that ∀yϕ(x 0 , y). Let M be a transitive structure
with x 0 ∈ M and modeling enough of set theory to construct a transitive forcing extension N in which AC fails and the
ZF axioms that went into a proof of σ → AC hold. Since ∀yϕ(x 0 , y) holds (in the universe) and ϕ is Σ0 , ∀yϕ(x 0 , y)
and hence σ holds in N. But then, this contradicts having that ZF proof of σ → AC.
This argument not only featured forcing as a model-theoretic method within a proof but also forcing over
uncountable structures, for the x 0 above could be arbitrary. Modern set theory would come to incorporate many
such tailored uses of forcing, and Levy’s application to definability was a remarkably early instance.
Levy [33] provided an analysis of Π2 statements of set theory in a different direction, one that addresses effectivity
in terms of witnessing terms. Suppose that ∀x∃yχ(x, y) is Π2 , with χ(x, y) being Σ0 , and recall that tc(x) denotes
the transitive closure of x. Levy established that if ZF  ∀x∃yχ(x, y), then there is a set-theoretic term τ (u) such that

ZF  ∀x(∃ finite u ⊆ tc(x) ∧ χ(x, τ (u))).

Note that one cannot do much better for the AC assertion ∀x∃y(x = ∅ ∨ y ∈ x). Levy also established that if ZFC
 ∀x∃yχ(x, y), then there is a set-theoretic term τ (u) such that

ZF  ∀x∀r (r is a well-ordering of tc(x) −→ χ(x, τ (r ))).

In his final article [40] Levy came full circle back to the bedrock of the main comprehension schemas of
ZF to investigate their forms anew. He began with the ZF axioms as most often given, as bi-conditionals with
e.g. ∀u∃y∀x(x ∈ y ←→ x ⊆ u) for the Power Set Axiom. He first addressed the issue of the parameters allowed in
the Separation and Replacement schemas. Let S0 denote the Separation schema restricted to formulas ϕ(x) with one
free variable x, ∀u∃y∀x(x ∈ y ←→ x ∈ u ∧ ϕ(x)). By a clever coding argument Levy established the positive result
that, over a weak subtheory (Extensionality, Pairs, Union, and a weak form of Power Set), S0 implies full Separation.
He established an analogous result for Replacement. Hence, those universal quantifications of parameters, distracting
when learning or teaching set theory, are not formally necessary after all.
Levy’s main, negative results addressed another issue of self-refinement in axiomatics and showed that his
aforementioned positive result is reasonably sharp. In the presence of Separation the generative axioms are sometimes
given parsimoniously in a weaker, conditional form, e.g. ∀u∃y∀x(x ⊆ u −→ x ∈ y) for the Power Set Axiom. With
his positive result, the set T consisting of the usual ZF axioms, but with the Separation schema replaced by S0 and
the Replacement schema replaced by the conditional version, is an axiomatization of ZF. Levy established that full
Separation is not a consequence of T if the Power Set axiom is weakened to the conditional form. He also established
the analogous results for the conditional version of Union and the conditional version of Pairing.
Levy established these delimitative results by, in effect, taking Cohen’s original AC independence model and
building appropriate submodels. In this he appealed to the Halpern–Levy [35] work on the Cohen model, the work that
248 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

effected the first substantial transfer from the Fraenkel–Mostowski context to the Cohen context. Levy’s sophisticated
results on the independence of Separation are a fitting coda, one that resonates with the work of Fraenkel [58], who
long ago and far away, steeped in the Hilbertian axiomatic tradition, established the independence of Separation from
Zermelo’s other axioms.

11. Envoi

Looking back over Levy’s researches in set theory, we see a steady and in fact increasing exploitation of model-
theoretic reflection and the method of forcing to establish substantial results about definability and axiomatizations.
Levy often saw and developed potentialities after an initial ground-breaking move made by others, and had a way
of establishing a full context with systematic, magisterial results. With his work set theory reached a new plateau in
the direction of understanding the scope and limits of formal expressibility and derivability. With this assimilated, set
theory would move forward over a broad range from the analysis of fine structure to a wealth of objectifications and
principles provided by large cardinal hypotheses, becoming infused with more and more combinatorial arguments as
well as sophisticated techniques involving forcing and inner models.
Around 1970 Levy turned to the writing of books, works that would establish a broad standard of understanding
about set theory. The classic Foundations of Set Theory by Fraenkel and Bar-Hillel [60] had become outdated because
of the many advances made in the 1960s, and so a “second revised edition” Fraenkel–Bar-Hillel–Levy [39] was
brought out. Fraenkel was by then deceased, and Levy in fact carried out an almost complete rewriting of the second
chapter. One section was published separately as Levy [41]. Throughout the discussion of the axiomatic foundations
one sees how the subject has become more elucidated by Levy’s own work.
Levy’s distinctive book Basic Set Theory [42], largely written when he was a visiting professor at Yale University
1971–1972 and at the University of California at Los Angeles 1976–1977, provided a systematic presentation of
the broad swath of set theory between elementary beginnings and advanced topics. Levy deliberately set out the
extent of set theory before the use of model-theoretic methods and forcing, working out the extensive combinatorial
development in a classical setting as rigorized by axiomatic foundations. In a way, it is quite remarkable that Levy
forestalled the inclusion of most of his own work by insisting on this middle way. On the other hand, the book is a
singular achievement of detailed exposition about what there is in set theory up to the use of the satisfaction predicate.
The account of trees is typical, on the one hand a bit idiosyncratic in dealing with generalities but on the other hand
broaching an interesting concept, that of a thin tree: Trees that have been studied on uncountable cardinals κ > ω1
usually have the property that their αth level has cardinality 2α , but Levy raised the issue about having cardinality at
most |α|. Throughout the book, the specialist is treated to a reckoning of the historical sources for the concepts. And
the student finds full and patient treatment of topics on which other texts might leave one queasy, like the set theorist’s
view of the reals. Descriptive set theorists from early on converted from the traditional construal of the real numbers
as the continuum to the function space ω ω of functions: ω → ω, and Levy explained in extensive detail the topological
and measure-theoretic connections among the various “real” spaces. What is perhaps most notable is the appendix to
the book, where at last Levy’s own rigorous and axiomatic approach to set theory casts a telling light. There, he allayed
another source of queasiness in accounts of set theory by showing, in some of the most thoroughgoing arguments in
the book, that introduced class terms can be formally eliminated in regression back to the primitive language of set
theory.
In the development of a mathematical field, a modest turn of events sometimes has an unforeseen effect and
achieves a folkloric distinction. Shelah had considerably developed his general concept of proper forcing by the early
1980s, a concept amenable to iteration schemes and having remarkably wide applications. Shelah lectured at the
Hebrew University on proper forcing, and Levy took systematic notes. These notes, refined and edited, eventually
became the bulk of the first three chapters of the monograph Shelah [107] and of the subsequent book Shelah [109]
as therein acknowledged. For a generation of set theorists the Levy exposition was the entré into proper forcing; once
drawn in, the return of Shelah’s inimitable hand in subsequent chapters led to new realizations.
In later years, Levy wrote two texts [43,44] in Hebrew on mathematical logic. The writing of books is an important
venture for the advancement of mathematical fields, and as one who has written a book knows well, it is a difficult
and time-consuming undertaking, especially when one aspires to codify mathematical knowledge over a broad range
and to make its many facets available to succeeding generations.
 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 249

Turning at last to his teaching and administrative work at the Hebrew University, there is a remarkable legacy of
renowned students, as listed below. Levy himself has served as the Dean of the Faculty of Science and the Chairman of
Institute of Mathematics and Computer Science, among other positions both in the university and with the academic
union. And, as mentioned before, his student Menachem Magidor is currently president of the university.
Soon after his 60th birthday, a 1996 issue of the Archive for Mathematical Logic (vol. 35, no. 5–6), was dedicated
to Levy, with the following words in the dedication (p. 279):
Azriel, besides being the important mathematician he is, is also a unique human being. His friends know him
as the epitome of wisdom. He can always be approached for good advice, which is given without any personal
interest, but purely out of a desire to help. His contribution to the public are innumerable. In any capacity he has
held—University administrator, in the educational system in Israel, as a member of usual important editorial
boards—you could always rely on his common sense, wisdom and devotion. Azriel is a symbol of intellectual
honesty and integrity. His former students will always remember him as a devoted and inspiring teacher.

To this we add the words of Ecclesiastes 9:7–9 (King James Version):

Go thy way, eat thy bread with joy, and drink thy wine with a merry heart; for God now accepteth thy works.
Let thy garments be always white; and let thy head lack no ointment.
Live joyfully with the wife whom thou lovest all the days . . . which he hath given thee under the sun . . .

Doctoral students of Azriel Levy

Dov Gabbay, Non-classical logics, 1969.

Shlomo Vinner, Some problems in first order predicate calculus with numerical quantifier, 1971. Second
supervisor: Haim Gaifman.
Gadi Moran, Size direction games over the real line, 1972.
Menachem Magidor, On supercompact cardinals, 1972.
Gershon Sageev, An independence result concerning the axiom of choice, 1973.
Moti Gitik, All uncountable cardinals can be singular, 1979. Second supervisor: Menachem Magidor.
Uri Abraham, Isomorphisms of Aronszajn trees and forcing without the generalized continuum hypothesis, 1979.
Second supervisor: Saharon Shelah.
Aharon Beller, Applications of Jensen’s coding technique, 1980.
Ami Litman, Combinatorial characterization of definable properties in the constructible universe, 1981. Second
supervisor: Menachem Magidor.
Hannah Perl, Teaching mathematics in high school with graphing calculators, 2002.

References

Publications of Azriel Levy

[1] Indépendance conditionnelle de V = L et d’axiomes qui se rattachent au système de M. Gödel, Comptes Rendus Hebdomadaires des
Séances de l’Académie des Sciences, Paris 245 (1957) 1582–1583.
[2] A note on definitions of finiteness, The Bulletin of the Research Council of Israel, Section F: Mathematics and Physics 7F (1958) 83–84.
[3] The independence of various definitions of finiteness, Fundamenta Mathematicae 46 (1958) 1–13.
[4] Comparison of subtheories, Proceedings of the American Mathematical Society 9 (1958) 942–945.
[5] On Ackermann’s set theory, The Journal of Symbolic Logic 24 (1959) 154–166.
[6] A hierarchy of formulae of set theory (abstract), Notices of the American Mathematical Society 6 (1959) 826.
[7] On models of set theory with urelements, Bulletin de l’Academie Polonaise des Sciences, Serie des Sciences Mathématiques, Astronomiques
et Physiques 8 (1960) 463–465.
[8] A generalization of Gödel’s notion of constructibility, The Journal of Symbolic Logic 25 (1960) 147–155.
[9] Axiom schemata of strong infinity in axiomatic set theory, Pacific Journal of Mathematics 10 (1960) 223–238. Reprinted in: Mengenlehre,
Wissensschaftliche Buchgesellschaft Darmstadt, 1979, pp. 238–253.
[10] Principles of reflection in axiomatic set theory, Fundamenta Mathematicae 49 (1960) 1–10.
[11] On a spectrum of set theories, Illinois Journal of Mathematics 4 (1960) 413–424.
[12] (with Robert L. Vaught) Principles of partial reflection in the set theories of Zermelo and Ackermann, Pacific Journal of Mathematics 11
(1961) 1045–1062.
250 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

[13] Axiomatization of induced theories, Proceedings of the American Mathematical Society 12 (1961) 251–253.
[14] Comparing the axioms of local and universal choice, in: Yehoshua Bar-Hillel, E.I.J. Poznanski, Michael O. Rabin, Abraham Robinson (Eds.),
Essays on the Foundations of Mathematics, Magnes Press, Jerusalem, 1961, pp. 83–90 (dedicated to Professor A.A. Fraenkel on his 70th
Birthday).
[15] On the principles of reflection in axiomatic set theory, in: Ernest Nagel, Patrick Suppes, Alfred Tarski (Eds.), Logic, Methodology and
Philosophy of Science (Proceedings of the 1960 International Congress), Stanford University Press, Stanford, 1962, pp. 87–93.
[16] Axioms of multiple choice, Fundamenta Mathematicae 50 (1962) 475–483.
[17] Remarks on a paper by J. Mycielski, Acta Mathematica Academiae Scientiarum Hungaricae 14 (1963) 125–130.
[18] Independence results in set theory by Cohen’s method I (abstract), Notices of the American Mathematical Society 10 (1963) 592–593.
[19] (with Solomon Feferman) Independence results in set theory by Cohen’s method II (abstract), Notices of the American Mathematical Society
10 (1963) 593.
[20] Independence results in set theory by Cohen’s method III (abstract), Notices of the American Mathematical Society 10 (1963) 593.
[21] Independence results in set theory by Cohen’s method IV (abstract), Notices of the American Mathematical Society 10 (1963) 593.
[22] The interdependence of certain consequences of the axiom of choice, Fundamenta Mathematicae 54 (1964) 135–157.
[23] (with James D. Halpern) The ordering theorem does not imply the axiom of choice (abstract), Notices of the American Mathematical Society
11 (1964) 56.
[24] Measurable cardinals and the continuum hypothesis, Notices of the American Mathematical Society 11 (1964) 769–770.
[25] A hierarchy of formulas in set theory (abstract), The Journal of Symbolic Logic 29 (1964) 226–227.
[26] The Fraenkel–Mostowski method for independence proofs in set theory, in: John W. Addison Jr., Leon Henkin, Alfred Tarski (Eds.), The
Theory of Models (Proceedings of the 1963 International Symposium at Berkeley), North-Holland, Amsterdam, 1965, pp. 221–228.
[27] Definability in axiomatic set theory I, in: Yehoshua Bar-Hillel (Ed.), Logic, Methodology and Philosophy of Science (Proceedings of the
1964 International Congress at Jerusalem), North-Holland, Amsterdam, 1965, pp. 127–151.
[28] A hierarchy of formulas in set theory, Memoirs of the American Mathematical Society 57 (1965) 76.
[29] (with Robert M. Solovay) Measurable cardinals and the continuum hypothesis, Israel Journal of Mathematics 5 (1967) 234–248.
[30] (with Georg Kreisel) Reflection principles and their use for establishing the complexity of axiomatic systems, Zeitschrift für Mathematische
Logik und Grundlagen der Mathematik 14 (1968) 97–142.
[31] On Von Neumann’s axiom system for set theory, American Mathematical Monthly 75 (1968) 762–763.
[32] The definability of cardinal numbers, in: Jack J. Bulloff, Thomas C. Holyoke, Samuel W. Hahn (Eds.), Foundations of Mathematics
(Symposium Papers Commemorating the Sixtieth Birthday of Kurt Gödel), Springer-Verlag, Berlin, 1969.
[33] The effectivity of existential statements in axiomatic set theory, Information Sciences 1 (1969) 119–130.
[34] Definability in axiomatic set theory II, in: Yehoshua Bar-Hillel (Ed.), Mathematical Logic and Foundations of Set Theory (Proceedings of
an International Colloquium), 11–14 November 1968, Jerusalem, North-Holland, Amsterdam, 1970, pp. 129–145.
[35] (with James D. Halpern) The Boolean prime ideal theorem does not imply the axiom of choice, in: Dana S. Scott (Ed.), Axiomatic Set
Theory, in: Proceedings of Symposia in Pure Mathematics, vol. 13, part I, American Mathematical Society, Providence, 1971, pp. 83–134.
[36] The sizes of the indescribable cardinals, in: Dana S. Scott (Ed.), Axiomatic Set Theory, in: Proceedings of Symposia in Pure Mathematics,
vol. 13, part I, American Mathematical Society, Providence, 1971, pp. 205–218.
[37] On the logical complexity of several axioms of set theory, in: Dana S. Scott (Ed.), Axiomatic Set Theory, in: Proceedings of Symposia in
Pure Mathematics, vol. 13, part I, American Mathematical Society, Providence, 1971, pp. 219–230.
[38] (with Robert M. Solovay) On the decomposition of sets of reals to Borel sets, Annals of Mathematical Logic 5 (1972) 1–19.
[39] (with Abraham A. Fraenkel and Yehoshua Bar-Hillel) Foundations of Set Theory, North-Holland, Amsterdam, 1973, 404 pages.
[40] Parameters in the comprehension axiom schemas of set theory, in: Leon Henkin et al. (Eds.), Proceedings of the Tarski Symposium,
in: Proceedings of Symposia in Pure Mathematics, vol. 25, American Mathematical Society, Providence, 1974, pp. 309–324.
[41] The role of classes in set theory (an excerpt from [39]), in: Gert H. Müller (Ed.), Sets and Classes, North-Holland, Amsterdam, 1976,
pp. 173–215.
[42] Basic Set Theory, Springer-Verlag, Berlin, 1979, 391 pages. Reprinted by Dover Publications, 2003.
[43] Mathematical Logic, A Course of the Open University of Israel, 1985 (in Hebrew).
[44] Mathematical Logic A, Academon Press, 1997 (in Hebrew).

Other references
[45] Wilhelm Ackermann, Zur Axiomatik der Mengenlehre, Mathematische Annalen 131 (1956) 336–345.
[46] John W. Addison Jr., Some consequences of the axiom of constructibility, Fundamenta Mathematicae 46 (1959) 337–357.
[47] Jon Barwise, Admissible Sets and Structures, Sprriwger-Verlag, Berlin, 1975.
[48] James E. Baumgartner, Ineffability properties of cardinals I, in: András Hajnal, Richard Rado, Vera T. Sós (Eds.), Infinite and Finite Sets,
in: Colloquia Mathematica Societatis Janos Bolyai, vol. 10, North-Holland, Amsterdam, 1975.
[49] Norman Bentwich, The Hebrew University of Jerusalem, 1918–1960, Weidenfeld and Nicolson, London, 1961.
[50] Paul Bernays, Zur Frage der Unendlichkeitsschemata in der axiomatischen Mengenlehre, in: Yehoshua Bar-Hillel, E.I.J. Poznanski, Michael
O. Rabin, Abraham Robinson (Eds.), Essays on the Foundations of Mathematics, Magnes Press, Jerusalem, 1961, pp. 3–49 (dedicated to
Professor A.A. Fraenkel on his 70th Birthday).
[51] Paul J. Cohen, The independence of the Axiom of Choice, Stanford University, 1963 (mimeographed preprint).
[52] Paul J. Cohen, The independence of the continuum hypothesis. I, Proceedings of the National Academy of Sciences, USA 50 (1963)
1143–1148.
 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252 251

[53] Paul J. Cohen, The independence of the continuum hypothesis. II, Proceedings of the National Academy of Sciences, USA 51 (1964)
105–110.
[54] Paul J. Cohen, Independence results in set theory, in: John W. Addison Jr., Leon Henkin, Alfred Tarski (Eds.), The Theory of Models
(Proceedings of the 1963 International Symposium at Berkeley), North-Holland, Amsterdam, 1965, pp. 39–54.
[55] Paul J. Cohen, Set Theory and the Continuum Hypothesis, Benjamin, New York, 1966.
[56] Solomon Feferman, Some applications of the notions of forcing and generic sets (summary), in: John W. Addison Jr., Leon Henkin,
Alfred Tarski (Eds.), The Theory of Models (Proceedings of the 1963 International Symposium at Berkeley), North-Holland, Amsterdam,
1965, pp. 89–95.
[57] Solomon Feferman, Some applications of the notions of forcing and generic sets, Fundamenta Mathematicae 56 (1965) 325–345.
[58] Abraham A. Fraenkel, Zu den Grundlagen der Cantor-Zermeloschen Mengenlehre, Mathematische Annalen 86 (1922) 230–237.
[59] Abraham A. Fraenkel, Über den Begriff ‘definit’ und die Unabhängigkeit des Auswahlaxioms, Sitzungsberichte der Preussischen Akademie
der Wissenschaften, Physikalish-mathematische Klass (1922) 253–257. Translated in van Heijenoort (Ed.), From Frege to Gödel: a Source
Book in Mathematical Logic, 1879–1931, Harvard University Press, Cambridge 1967.
[60] Abraham A. Fraenkel, Yehoshua Bar-Hillel, Foundations of Set Theory, North-Holland, Amsterdam, 1958, 415 pages.
[61] Robert J. Gauntt, Undefinability of cardinality, in: Lectures notes prepared in connection with the Summer Institute on Axiomatic Set Theory
held at University of California, Los Angeles, IV-M, 1967.
[62] Gerhard Gentzen, Beweisbarkeit und Unbeweisbarkeit van Anfangsfällen der transfiniten Induktion in der reinen Zahlentheorie,
Mathematische Annalen 119 (1943) 140–161. Translated in: M.E. Szabo (Ed.), The Collected Papers of Gerhard Gentzen, North-Holland,
Amsterdam 1969, pp. 287–308.
[63] Kurt Gödel, The consistency of the axiom of choice and of the generalized continuum-hypothesis, Proceedings of the National Academy of
Sciences, USA 24 (1938) 556–557. Reprinted in [69], pp. 26–27.
[64] Kurt Gödel, Consistency-proof for the generalized continuum-hypothesis, Proceedings of the National Academy of Sciences, USA 25 (1939)
220–224. Reprinted in [69], pp. 28–32.
[65] Kurt Gödel, The Consistency of the Axiom of Choice and of the Generalized Continuum Hypothesis with the Axioms of Set Theory,
in: Annals of Mathematics Studies #3, Princeton University Press, Princeton, 1940. Reprinted in [69], pp. 33–101.
[66] Kurt Gödel, Lecture [on the] consistency [of the] continuum hypothesis, Brown University, in [70], pp. 175–185, and the page references are
to these, 1940.
[67] Kurt Gödel, Remarks before the Princeton bicentennial conference on problems in mathematics, 1–4, in [69, 1946], pp. 150–153.
[68] Kurt Gödel, What is Cantor’s continuum problem? American Mathematical Society 54 (1947) 515–525; American Mathematical Society
55 (1947) 151 (errata). Reprinted in [69], pp. 176–187.
[69] Kurt Gödel, in: Solomon Feferman (Ed.), Collected Works, vol. 2, Oxford University Press, New York, 1990.
[70] Kurt Gödel, in: Solomon Feferman (Ed.), Collected Works, vol. 3, Oxford University Press, New York, 1995.
[71] Kurt Gödel, in: Solomon Feferman, John W. Dawson Jr. (Eds.), Collected Works, vol. 4, Clarendon Press, Oxford, 2003.
[72] Kurt Gödel, in: Solomon Feferman, John W. Dawson Jr. (Eds.), Collected Works, vol. 5, Clarendon Press, Oxford, 2003.
[73] András Hajnal, On a consistency theorem connected with the generalized continuum problem, Zeitschrift für Mathematische Logik und
Grundlagen der Mathematik 2 (1956) 131–136.
[74] András Hajnal, On a consistency theorem connected with the generalized continuum problem, Acta Mathematica Academiae Scientiarum
Hungaricae 12 (1961) 321–376.
[75] James D. Halpern, The independence of the axiom of choice from the Boolean prime ideal theorem (abstract), Notices of the American
Mathematical Society 8 (1961) 279–280.
[76] James D. Halpern, The independence of the axiom of choice from the Boolean prime ideal theorem, Fundamenta Mathmaticae 55 (1964)
57–66.
[77] James D. Halpern, Hans Läuchli, A partition theorem, Transactions of the American Mathematical society 124 (1966) 360–367.
[78] William P. Hanf, Dana S. Scott, Classifying inaccessible cardinals (abstract), Notices of the American Mathematical Society 8 (1961) 445.
[79] Paul E. Howard, Limitations of the Fraenkel–Mostowski method of independence proofs, The Journal of Symbolic Logic 38 (1973) 416–422.
[80] Paul E. Howard, Jean E. Rubin, Consequences of the Axiom of Choice, in: Mathematical Surveys and Monographs, vol. 59, American
Mathematical Society, Providence, 1998.
[81] Thomas J. Jech, The Axiom of Choice, North-Holland, Amsterdam, 1973.
[82] Ronald B. Jensen, Consistency results for ZF, Notices of the American Mathematical Society 14 (1967) 137.
[83] Akihiro Kanamori, The Higher Infinite, corrected second ed., Springer, Berlin, 2005.
[84] H. Jerome Keisler, Alfred Tarski, From accessible to inaccessible cardinals, Fundamenta Mathematicae 53 (1964) 225–308; Fundamenta
Mathematicae 57 (1965) 119 (errata).
[85] W. Kinna, K. Wagner, Über eine Abschwächung des Auswahlpostulates, Fundamenta Mathematicae 42 (1955) 75–82.
[86] Stephen C. Kleene, Recursive predicates and quantifiers, Transactions of the American Mathematical Society 53 (1943) 41–73.
[87] Motokiti Kondô, L’uniformization des complémentaires analytiques, Proceedings of the Imperial Academy of Japan 13 (1937) 287–291.
[88] Hans Läuchli, The independence of the ordering principle from a restricted axiom of choice, Fundamenta Mathematicae 54 (1964) 31–43.
[89] Richard Laver, Products of infinitely many perfect trees, Journal of the London Mathematical Society 29 (1984) 385–396.
[90] Paul Mahlo, Über lineare transfinite Mengen, Berichte über die Verhandlungen der Königlich Sächsischen Gesellschaft der Wissenschaften
zu Leipzig, Mathematische-Physische Klass 63 (1911) 187–225.
[91] Paul Mahlo, Zur Theorie und Anwendung der ρ0 -Zahlen, Berichte über die Verhandlungen der Königlich Sächsischen Gesellschaft der
Wissenschaften zu Leipzig, Mathematische-Physische Klass 64 (1912) 108–112.
252 A. Kanamori / Annals of Pure and Applied Logic 140 (2006) 233–252

[92] Paul Mahlo, Zur Theorie und Anwendung der ρ0 -Zahlen II, Berichte über die Verhandlungen der Königlich Sächsischen Gesellschaft der
Wissenschaften zu Leipzig, Mathematische-Physische Klass 65 (1913) 268–282.
[93] Donald A. Martin, Robert M. Solovay, Internal Cohen extensions, Annals of Mathematical Logic 2 (1970) 143–178.
[94] Elliot Mendelson, The independence of a weak axiom of choice, The Journal of Symbolic Logic 21 (1956) 350–366.
[95] Richard M. Montague, Fraenkel’s addition to the axioms of Zermelo, in: Yehoshua Bar-Hillel, E.I.J. Poznanski, Michael O. Rabin,
Abraham Robinson (Eds.), Essays on the Foundations of Mathematics, Magnes Press, Jerusalem, 1961, pp. 91–114 (dedicated to Professor
A.A. Fraenkel on his 70th Birthday).
[96] Gregory H. Moore, The origins of forcing, in: Frank R. Drake, John K. Truss (Eds.), Logic Colloquium’86, North-Holland, Amsterdam,
1988.
[97] Andrzej M. Mostowski, Über die Unabhängigkeit des Wohlordnungssatzes vom Ordnungsprinzip, Fundamenta Mathematicae 32 (1939)
201–252.
[98] Andrzej M. Mostowski, On the principle of dependent choices, Fundamenta Mathematicae 35 (1948) 127–130.
[99] Andrzej M. Mostowski, On a problem of W. Kinna and K. Wagner, Colloquium Mathematicum 6 (1958) 207–208.
[100] John Myhill, Dana S. Scott, Ordinal definability, in: Dana S. Scott (Ed.), Axiomatic Set Theory, in: Proceedings of Symposia in Pure
Mathematics, vol. 13, part I, American Mathematical Society, Providence, 1971, pp. 271–278.
[101] David Pincus, Zermelo–Fraenkel consistency results by Fraenkel–Mostowski methods, The Journal of Symbolic Logic 37 (1972) 721–743.
[102] David Pincus, Cardinal representatives, Israel Journal of Mathematics 18 (1974) 321–344.
[103] William N. Reinhardt, Ackermann’s set theory equals ZF, Annals of Mathematical Logic 2 (1970) 189–249.
[104] Gershon Sageev, An independence result concerning the axiom of choice, Annals of Mathematical Logic 8 (1975) 1–184.
[105] Dana S. Scott, Definitions by abstraction in axiomatic set theory, Bulletin of the American Mathematical Society 61 (1955) 442.
[106] Dana S. Scott, Measurable cardinals and constructible sets, Bulletin de l’Académie Polonaise des Sciences, Série des Sciences
Mathématiques, Astronomiques et Physiques 9 (1961) 521–524.
[107] Saharon Shelah, Proper Forcing, in: Lecture Notes in Mathematics, vol. 940, Springer-Verlag, Berlin, 1982.
[108] Saharon Shelah, Can you take Solovay’s inaccessible away? Israel Journal of Mathematics 49 (1984) 1–47.
[109] Saharon Shelah, Proper and Improper Forcing, in: Perspectives in Mathematical Logic, Springer, Berlin, 1998.
[110] John C. Shepherdson, Purported review of Levy [1], The Journal of Symbolic Logic 26 (1959) 226.
[111] Joseph R. Shoenfield, Constructible sets (abstract), in: Summaries of talks, Summer Institute for Symbolic Logic, Cornell University, 1957,
Institute for Defense Analysis, Princeton, vol. 2, 1957, p. 214.
[112] Joseph R. Shoenfield, On the independence of the axiom of constructibility, American Journal of Mathematics 81 (1959) 537–540.
[113] Joseph R. Shoenfield, The problem of predicativity, in: Yehoshua Bar-Hillel, E.I.J. Poznanski, Michael O. Rabin, Abraham Robinson (Eds.),
Essays on the Foundations of Mathematics, Magnes Press, Jerusalem, 1961, pp. 132–139 (dedicated to Professor A.A. Fraenkel on his 70th
Birthday).
[114] Wacław Sierpiński, Sur une classe d’ensembles, Fundamenta Mathematicae 7 (1925) 237–243.
[115] Robert M. Solovay, Measurable cardinals and the continuum hypothesis (abstract), Notices of the American Mathematical Society 12 (1965)
132.
[116] Robert M. Solovay, The measure problem (abstract), Notices of the American Mathematical Society 12 (1965) 217.
[117] Robert M. Solovay, A model of set theory in which every set of reals is Lebesgue measurable, Annals of Mathematics 92 (1970) 1–56.
[118] Ernst Specker, Zur Axiomatik der Mengenlehre (Fundierungs- und Auswahlaxiome), Zeitschrift für Mathematische Logik und Grundlagen
der Mathematik 3 (1957) 173–210.
[119] Robert L. Vaught, Indescribable cardinals (abstract), Notices of the American Mathematical Society 10 (1963) 126.