You are on page 1of 17

May 20, 2010

HUME’S PRINCIPLE, BEGINNINGS
ALBERT VISSER
Abstract. In this note we derive Robinson’s Arithmetic from Hume’s Prin-
ciple in the context of very weak theories of classes and relations.
1. Introduction
Frege’s derivation of arithmetic from Hume’s Principle has great beauty. In this
derivation, numbers are created by an autocatalytic process starting from nothing.
Moreover, in one fell swoop, we also define the arithmetical operations of addition
and multiplication. In this paper we try to obtain a closer understanding of the
fine structure of the employment of Hume’s Principle to obtain basic arithmetical
principles.
1.1. Contents of The Paper. In the present work, I attempt a detailed and slow
study to address the following questions:
a. How much comprehension is needed to develop a weak system of arithmetic?
b. Can we get a more modular picture of how arithmetic follows from Hume’s Prin-
ciple?
John Burgess shows, in his book [Bur05], that a predicative system of second order
logic, with variables for binary relations, expanded with Hume’s Principle interprets
Robinson’s Arithmetic Q. This result suggests the following picture. We can divide
the proof of Frege’s Theorem into two steps. First, in the context of a weak theory
of binary relations over a domain we can develop a weak arithmetic with the help
of Hume’s principle and, then, by increasing the strength of the theory of relations
we raise the strength of the resulting arithmetical system.
Zooming in on the development of arithmetic in a weak second order theory
using Hume’s principle, we see that there are, prima facie, three ingredients.
a. Hume’s principle gives us an infinity of objects.
b. Hume’s principle allows us to treat numbers as objects and, hence, enables us
to speak about classes of numbers in a second order context.
c. Hume’s principle helps us to define the arithmetical operations.
In the present paper we will concentrate on the development of first order arithmetic
from Hume’s principle, so on (a) and (c). Item (b) will only occur in so far as it is
used for (c). We will show that if we start with adjunctive relation theory, a very
weak second order system, then the only contribution needed of Hume’s principle
to develop an interpretation of Q is (a). We can define the arithmetical operations
as operations on classes modulo equinumerosity without projecting these classes to
2000 Mathematics Subject Classification. 03A05, 03B30, 03F25,
Key words and phrases. interpretations, weak arithmetical theories, Hume’s Principle.
I am grateful to Panu Raatikainen. This contribution is a spin-off of my thinking about his
questions. Aldo Antonelli gave detailed comments on an earlier version of this paper. I thank him
for that.
1
2 ALBERT VISSER
objects. In other words, Hume’s principle is not needed for (c) after all. Here, we
use the presence of relations (i) to define the relation of equinumerosity and (ii) to
implement the arithmetical operations on classes.
We can get a somewhat more refined and more interesting picture, if we do not
work with relations but only with classes. We show that, if we start with adjunctive
class theory with a primitively given equinumerosity relation plus the principle that
says that there is no universal class, then we can develop Q
add
, the analogue of
Q for addition, without multiplication, but that we cannot develop Q.
1
To define
multiplication, we need Hume’s principle, so, in a class context, Hume’s principle
is essential for (c).
Thus, in the context of a theory of classes with equinumerosity, Hume’s principle
is connected to what one could call the reverse mathematics of addition and mul-
tiplication. Such a program involves matching ‘natural’ pairs of class theories with
pairs of a theory of addition and a theory of addition and multiplication. We do
not develop the full picture here. We present one clear result: the adjunctive theory
of classes with equinumerosity and Hume’s principle is mutually interpretable with
Q.
2
The precise relationship between adjunctive class theory with equinumerosity
and no-universe and, on the other hand, a suitable extension of Q
add
still has to be
analyzed. We show that the first theory interprets the second. It seems plausible
that, in the other direction, there is an informative reduction relation, along the
lines sketched by Feferman and Vaught in their paper [FV59] for the case of true
theories. We show, in Section 6, that the second theory does not interpret the first.
Hence, we really do need the less restrictive Feferman-Orey relation between the
theories.
1.2. Almost Philosophical Considerations. In the kind of foundational re-
search exemplified by this paper, there are, very roughly, two colors, say blue and
yellow.
3
Blue is the study of the fine structure of reasoning. Here we have reverse
mathematics, theory reduction, interpretability and the like. Yellow has a tighter
link with philosophy: we are looking for a serious justification of numerical reason-
ing, of sets, etcetera. A lot of contemporary work on Frege’s Grundgesetze falls
under yellow. Of course, the separation is not strict. E.g., Burgess’ book [Bur05]
exhibits both colors. Of course, work falling under blue can provide useful data for
research of color yellow.
The primary focus of the present paper is the fine structure of reasoning, making
it blue. Thus, the fact that we work in very weak theories does not necessarily reflect
a conviction that these theories can be justified, or that we cannot justify much
more.
One important aspect of both Burgess’ development of Q from a predicative
theory of binary relations expanded with Hume’s Principle and the development in
our paper is that we grant ourselves the freedom of gaining more ‘good’ properties
of our numbers by contracting the number domain. This methodology was initi-
ated by Edward Nelson in his book [Nel86], thus transforming Solovay’s method of
shortening cuts (introduced in [Sol76]) into a foundational tool.
1
The system Q
add
is a subsystem of Presburger Arithmetic PresA. We show in an appendix
that Q
add
locally interprets PresA.
2
It would be interesting to explore the existence of tighter relations. For certain extensions of
both theories we do indeed have tighter connections.
3
The choice of colors is inspired by a famous scene from Monty Python and the Holy Grail.
HUME’S PRINCIPLE, BEGINNINGS 3
In the context of a Frege-style development of number theory (of color yellow)
this strategy is not unproblematic. In a Frege-style program, we are interested in
defining the natural numbers. Not every class of objects having some desired prop-
erties usually associated with the natural numbers qualifies. After all, we want to
say what the natural numbers are. Thus, it could be our aim to see what certain
principles allow us to prove about the numbers as given by some fixed definition,
e.g., the precise definitions employed by Frege, rather than to fiddle around with
the definition to get better properties.
4
Similarly, we could be interested in a recon-
struction that justifies this specific reasoning and not in justifying quite different
reasoning with the same conclusion.
We note that the preferred definitions view is not without its own problems. The
claim that such-and-such a definition is the intended one, is in need of informally
rigorous argumentation. For example, Frege’s definition of the natural numbers
using the ancestral seems to me to be an ordinal-style definition, where Burgess’
definition of proto-natural numbers (p156 of [Bur05]) has much more a cardinal
flavour: it is a strengthened version of Dedekind finiteness. Given that Frege’s
development is intended to build the cardinals, which is the better one?
Let me stress that the kind of blue project we are involved in is not necessarily
antithetical to a yellow view involving intended definitions. Suppose, e.g., we have
a preferred definition N
0
of, say, the natural numbers, and suppose we want them
to have the good property P. Suppose further we have proved P in a very weak
context for a ‘contracted’ virtual class of natural numbers N
1
. Then a good theory
in which to get P for N
0
could be a theory in which N
1
and N
0
coincide. So we
immediately know that a principle that implies the equality of N
0
and N
1
will have
numbers with the good property P.
On the other hand, it would be interesting to see whether there is a coherent
foundational idea corresponding to the methodology of shortening cuts. This would
involve rejecting the claim that induction is almost analytical for the natural num-
bers (made by Linnebo in his paper [Lin04]).
5
When working with this method
one develops the feeling that such a philosophy is just around the corner. Nelson’s
program as developed in [Nel86] seems to be founded on the conviction that, yes,
there is a foundational idea that justifies building up the numbers by ‘number sys-
tem hopping’. Unfortunately, Nelson gives us no hint what such an idea could look
like.
2. Preliminaries
This note will presuppose familiarity with the notions and notations introduced
in [Vis09]. In this section, we provide a list of principles and theories. The best
reading strategy is to glance through this section and return to it as soon as a
principle is added.
We will use the following convention. Theories will be specified as e.g. [a3, b2]
where a3 and b2 are already specified principles. The signature of the theory will
4
For a paper that studies preferred definitions for a Frege-style development of Arithmetic,
see Øystein Linnebo’s paper [Lin04]. Linnebo provides an interesting discussion of Burgess’ proof
that predicative relational Frege theory interprets Q.
5
Induction in the case discussed by Linnebo is not really induction anyway, in the light of the
restriction on comprehension.
4 ALBERT VISSER
be the minimal signature that is involved in all the axioms of the theory. If a theory
U is given U[a3, b2] will be the result of adding a3 and b2 to U.
2.1. Class Theories. We start with class theories. We have a language with two
sorts: o, the sort of objects and c, the sort of classes. We let x, y, z, . . . range over
objects and X, Y, Z, . . . over classes. We have identity for each sort and one binary
predicate ∈ of sort oc. We have the following list of principles.
c1 ¬ ∃X∀x x ,∈ X (empty class axiom)
c2 ¬ ∃Y ∀y (y ∈ Y ↔ (y ∈ X ∨ y = x)) (adjunction axiom),
c3 ¬ ∀z (z ∈ X ↔ z ∈ Y ) → X = Y (extensionality)
c4 ¬ ∃Z ∀z (z ∈ Z ↔ (z ∈ X ∧ z ∈ Y )) (closure under intersection)
c5 ¬ ∃Y ∀y (y ∈ Y ↔ (y ∈ X ∧ y ,= x)) (subtraction axiom)
c6 ¬ ∃Z ∀z (z ∈ Z ↔ (z ∈ X ∧ z ,∈ Y )) (closure under class-subtraction)
c7 ¬ ∃Z ∀z (z ∈ Z ↔ (z ∈ X ∨ z ∈ Y )) (closure under union).
c8 ¬ ∃x x ,∈ X (no-universe axiom)
The theory ac is [c1,c2,c3]. We note that it has a one-sorted version, with just
classes and the relation ⊆, where we interpret the objects as the atoms of the ⊆-
ordering.
6
We also note that axiom c3 is really superfluous, since we can always
interpret identity on the classes as extensional equivalence.
We note that if we have the theory with all our principles, say, ac
+
, then c2 and
c5 can be replaced by an axiom stating the existence of singletons. The theory ac
+
can easily be shown to be complete via quantifier elimination by a minor adaptation
of the proof of the L¨owenheim-Behman theorem. See [Bur05] for an exposition of
the original theorem.
2.2. Adding Equinumerosity. We extend the language of ac with a binary re-
lation symbol ≡ of type cc. We have the following principles for the extended
language. In the statement of some of these principles we use operations on classes.
We will only use the principles in the context of axioms stating that these operations
are total. We write X # Y for ∀z (z ,∈ X ∨ z ,∈ Y ).
e1 ¬ (X ≡ ∅ ∨ ∅ ≡ X) ↔ X = ∅
e2 ¬ (x ,∈ X ∧ y ,∈ Y ) → ((X ∪ ¦x¦) ≡ (Y ∪ ¦y¦) ↔ X ≡ Y )
e3 ¬ X ≡ X
e4 ¬ X ≡ Y → Y ≡ X
e5 ¬ (X ≡ Y ∧ Y ≡ Z) → X ≡ Z
e6 ¬ Y ⊂ X → Y ,≡ X (Dedekind finiteness axiom)
e7 ¬ (X ⊆ Y ≡ Y

) → ∃X

(X ≡ X

⊆ Y

)
e8 ¬ (X ≡ X

∧ Y ≡ Y

∧ X ⊆ Y ∧ X

⊆ Y

) → (Y ¸ X) ≡ (Y

¸ X

)
e9 ¬ (X # Y ∧ X

# Y

∧ X ≡ X

∧ Y ≡ Y

) → (X ∪ Y ) ≡ (X

∪ Y

)
e10 ¬ ∃Y

(Y ≡ Y

∧ X # Y

)
The theory eqnum is ac[e1,e2]. Feferman and Vaught in their classical paper [FV59],
show that, for any cardinality κ, the theory of all classes of elements from a domain
of size κ, with the subset ordering and with the relation of equinumerosity, is
decidable. We can interpret eqnum into these theories by identifying the objects
6
If we set it up correctly this one-sorted version is bi-interpretable with the original theory. This
means that there are interpretations between the two theories whose compositions are definably
and provably isomorphic with the identity interpretation.
HUME’S PRINCIPLE, BEGINNINGS 5
with the atoms of ⊆. It follows that eqnum has a decidable extension and, hence,
does not interpret Q.
2.3. Adding the Hume Relation. To formulate Hume’s principe, we add the
Hume relation to the language. This relation is of type co. We choose to work
with a Hume relation as opposed to a Hume function because the functionality
never seems to play any role in the arguments. Moreover, functionality is a rather
strong demand. For example, in the context of ZF without choice we need Scott’s
trick to produce a Hume function, which depends on a lot of details of the theory.
So by dropping functionality we have a substantial gain in generality. Here is a list
of principles.
h1 ¬ ∃x X x
h2 ¬ (X z ∧ Y z) → X ≡ Y
h3 ¬ X ≡ Y z → X z
The theory Hume Light or HL is eqnum[h1,h2].
We note that we could eliminate ≡ from the language by redefining X ≡ Y as
∃z (X z ∧ Y z). This would ask for a careful choice of axioms. Moreover,
we would get some properties of the equinumerosity relation like reflexivity and
symmetry for free.
2.4. Relational Theories. We introduce adjunctive relation theory ar. Adjunc-
tive relation theory is a two-sorted theory with sorts o, the sort of objects, and r,
the sort of relations. We use x, y, z, . . . to range over objects and R, S, . . . to range
over relations. Our theory has identity for each sort and a ternary relation app of
type roo. We write Rxy for app(R, x, y). Here is a list of principles.
r1 ¬ ∃R ∀x, y Rxy
r2 ¬ ∃S ∀u, v (Suv ↔ (Ruv ∨ (x = u ∧ y = v)))
r3 ¬ ∃S ∀u, v (Suv ↔ (Ruv ∧ (x ,= u ∨ y ,= v)))
The theory ar is given by [r1,r2]. In ar we can define classes as diagonal relations, i.e.,
as relations such that, for all x, y, if Rxy, then x = y. Clearly this interpretation
does yield ac for our classes. We will add principles concerning classes to ar as a
matter of course under this reading. So, e.g. ar[c8] is ar plus ¬ ∃x Rxx.
We remind the reader of a result of [Vis09]. There it is shown that ar plus the no-
universe axiom interprets Q as a theory of cardinals over the given object domain.
2.5. Arithmetical Theories. We consider the following list of principles in the
inclusive signature 0, S, +, , ≤. We often write x y or xy for x y.
q1 ¬ Sx ,= 0
q2 ¬ Sx = Sy → x = y
q3 ¬ x + 0 = x
q4 ¬ x + Sy = S(x +y)
q5 ¬ x 0 = 0
q6 ¬ x Sy = (x y) +x
q7 ¬ x = 0 ∨ ∃y x = y + 1
q8 ¬ x ≤ x
q9 ¬ (x ≤ y ∧ y ≤ z) → x ≤ z
q10 ¬ (x ≤ y ∧ y ≤ x) → x = y
q11 ¬ x ≤ y → (x = y ∨ Sx ≤ y)
q12 ¬ x ≤ y ∨ y ≤ x
6 ALBERT VISSER
q13 ¬ x ≤ y ↔ ∃z (z +x = y)
q14 ¬ Sx ,= x
q15 ¬ (x +y) +z = x + (y +z)
q16 ¬ x +y = y +x
q17 ¬ x +z = y +z → x = y
q18 ¬ x (y +z) = (x y) + (x z)
q19 ¬ x (y z) = (x y) z
The theory Q
add
is the theory [q1, q2, q3, q4, q7], in the signature with 0, S, +. The
theory Q is Q
add
[q5, q6] in the signature with 0, S, +, .
3. No-Universe
We first show that a version of Hume’s Principle allows us to construct a totality
of classes that is adjunctive, but does not contain the universe of objects. To be
precise, we show that eqnum directly interprets the no-universe axiom ac8.
3.1. Bootstrap in ac. We start to work in ac. As a first step, define A
0
as the
virtual class of classes such that X is in A
0
iff, for all Y , X ∩ Y exists. We easily
see that A
0
is closed under empty class, adjunction, and intersection. We note
that this argument uses the associativity of intersection in the strong sense that if
(X ∩ Y ) ∩ Z is defined, then so is X ∩ (Y ∩ Z) and both are equal, and vice versa.
The equality relies on extensionality, which we use here as a matter of course. We
relativize our classes to A
0
, thereby gaining the extra axiom c4: that classes are
closed under intersection.
We work in ac[c4]. We consider A
1
, where X is in A
1
if, for all x, X ¸ ¦x¦ exists.
It is easy to see that A
1
is closed under empty class, adjunction, intersection and
subtraction. (We need the presence of intersection for closure under subtraction!)
We relativize to A
1
, thereby gaining the axioms c4 and c5. Here c5 tells us that
classes are closed under subtraction. Thus, we have directly interpreted ac[c4, c5].
3.2. Bootstrap in eqnum. By the result of the previous subsection, we can directly
interpret eqnum[c5] in eqnum. We work in eqnum[c5]. We define A
2
as the totality
of those classes X such that:
X ≡ X and ∀Y (X ≡ Y → Y ≡ X) and ∀Y, Z ((X ≡ Y ∧ Y ≡ Z) → X ≡ Z).
It is easy to see that ∅ ∈ A
2
. We show that A
2
is closed under adjunction. We
treat the case of transitivity. Suppose (X ∪ ¦x¦) ≡ Y and Y ≡ Z. If x is in X, we
are immediately done. Suppose x ,∈ X. We note that Y and Z cannot be empty.
Let y ∈ Y and z ∈ Z. We find that X ≡ (Y ¸ ¦y¦) and (Y ¸ ¦y¦) ≡ (Z ¸ ¦z¦).
Ergo X ≡ (Z ¸ ¦z¦). Hence, (X ∪¦x¦) ≡ Z. Contracting our classes to A
2
we gain
the axioms stating that ≡ is an equivalence relation. We note that this property is
preserved under any further contraction of our classes. Relativizing to A
2
gives us
eqnum[e3, e4, e5].
We work in eqnum[e3, e4, e5]. We note that e3,e4,e5 are universal and, hence, pre-
served under contraction of the totality of classes. We repeat the procedure of the
previous subsection to regain the subtraction axiom: we get eqnum[c5, e3, e4, e5].
7
7
We note that in in eqnum[c5] plus the no-universe principle, i.e. eqnum[c5, c8], we can directly
prove that X
2
is closed under subtraction.
HUME’S PRINCIPLE, BEGINNINGS 7
We work in eqnum[c5, e3, e4, e5]. We say that X is Dedekind finite iff, for every
Y ⊂ X, we have X ,≡ Y . Let A
3
consist of the Dedekind finite classes. Clearly, the
empty class is in A
3
.
We show that A
3
are closed under adjunction. Suppose X is Dedekind finite.
We show that X ∪ ¦x¦ is Dedekind finite. We may assume that x ,∈ X. Suppose
X
0
⊂ (X ∪¦x¦) and X
0
≡ (X ∪¦x¦). It follows that X
0
is not empty. Let x
0
be x
in case x ∈ X
0
and let x
0
be some arbitrary element of X
0
otherwise. We find that
(X
0
¸ ¦x
0
¦) ⊂ X and (X
0
¸ ¦x
0
¦) ≡ X. A contradiction with the assumption that
X is Dedekind finite.
It is easy to see that A
3
is closed under subtraction. We relativize to A
3
, thus
gaining the axiom e6, that all classes are Dedekind finite. So we have directly
interpreted eqnum[c5, e3, e4, e5, e6].
We note a useful fact.
Lemma 3.1 (in eqnum). Suppose Y is Dedekind finite, X ≡ Y and x ,∈ X. Then,
for some y, y ,∈ Y . In another formulation: suppose the universe is a Dedekind
finite class, then any class equinumerous to it is again the universe.
Proof. Suppose Y is Dedekind finite, X ≡ Y and x ,∈ X. Suppose Y would be
the universe, then we have (X ∪ ¦x¦) ⊆ Y , and hence X ⊂ Y . But X ≡ Y ,
contradicting the fact that Y is Dedekind finite. 2
3.3. Bootstrap in HL. At this point we switch to the theory HL. Our develop-
ment in this subsection is a straightforward adaptation of the treatment predicative
second order logic with the Hume function in [Bur05].
By the preceding development, we can directly interpret eqnum[c5, e3, e4, e5, e6],
so we can work in HL[c5, e3, e4, e5, e6]. By replacing X y by ∃Z X ≡ Z y, we
gain the axiom h3. So we may work in HL
+
:= HL[c5, e3, e4, e5, e6, h3].
Let A
4
consist of all X, such that there is a class Y such that X ≡ Y and, for
every y ∈ Y , there is an X
0
⊂ X such that X
0
y. Clearly, ∅ ∈ A
4
.
We show that A
4
is closed under adjunction. Suppose X is in A
4
. Let Y witness
that X is in A
4
. Consider X ∪ ¦x¦. We may assume that x ,∈ X. Suppose X y.
We show that Y ∪ ¦y¦ has the desired property.
First, we note that y ,∈ Y . If it were, we would have X y and X
0
y, for
some X
0
⊂ X. So X
0
≡ X, contradicting the fact that X is Dedekind finite.
Ergo, (X ∪ ¦x¦) ≡ (Y ∪ ¦y¦). Consider y

∈ Y ∪ ¦y¦. If y ∈ Y , we can find
X
0
⊂ X ⊂ X ∪ ¦x¦ with X
0
y. If y

= y, we have X y and X ⊂ X ∪ ¦x¦.
We prove that no element of A
4
is the universe. Suppose X is in A
4
. Let Y be the
promised witness of this fact. Let Y y. We find y ,∈ Y . It follows, by Lemma 3.1,
that, for some x, we have x ,∈ X. Finally, we contract our classes to A
4
. This gives
us HP[c8].
3.4. Bootstrap in adjunctive-subtractive relation theory. We consider ad-
junctive-subtractive relation theory ar[r3]. In this theory we treat classes as diagonal
relations. We define the equinumerosity of X and Y as the existence of a bijection
between X and Y . It is immediate that ar[r3] verifies the axioms e1 and e2. Thus,
when we add Hume’s principle in the form of h1 and h2 for equinumerosity-as-
defined, we obtain a direct interpretation of HL. Thus, by the result of the previous
subsection, we may conclude that ar[r3,h1,h2] directly interprets ar plus the no-
universe axiom c8.
8 ALBERT VISSER
By the results of [Vis09], we find that ar[r3,h1,h2] interprets Q as a theory of
cardinals. In the converse direction, we know that Q interprets I∆
0
+ Ω
1
. Using
familiar methods one may interpret ar[r3,h1,h2] in this last theory. So, we find that
ar[r3,h1,h2] is mutually interpretable with Q.
We note that ar[c8] does not directly interpret ar[r3,h1,h2]. Suppose it did. Con-
sider a standard model / of ar[c8] with an infinite domain and all finite relations
over the domain. We fix a finite set of parameters of the interpretation in /.
Since, any relation in / is definable from finitely many objects, we may assume
that the parameters are objects. The interpretation provides us with an internal
model ^ of ar[r3,h1,h2]. Without loss of generality, we may assume that we also
have h3. Since every permutation of the objects in / extends to an automorphism
of /, and since the objects of ^ coincide with the objects of /, any permutation
of the objects of ^ that fixes the parameters extends to an automorphism of ^.
Clearly, we can find a standardly finite set X in ^ and an element n in ^ such
that X
N
n and the elements of X (according to ^) and n are disjoint from the
parameters.
8
Take any permutation σ of the domain that fixes the parameters. We
have X ≡
N
σX
N
σn. It follows that, for any non-parameter n

, X
N
n

. Quod
impossibile.
Open Question 3.1. We need the subtraction axiom r3 in our development since
the definition of equinumerosity in this context contains an existential quantifier
over relations. Thus, equinumerosity is not absolute for contraction of relations.
For this reason, we might loose axiom h2 when we try to obtain subtraction of
relations via contraction. Can we eliminate the use of the subtraction axiom r3 in
our result?
4. Development of the Theory of Addition in eqnum[c8]
In this section we show how to interpret Q
add
in eqnum[c8]. We will provide a
bit more of the theory of addition since this flows naturally from the development.
In Appendix A, we show that we can go on to locally interpret full Presburger
Arithmetic PresA in Q
add
.
We begin with a lemma that works already in ac[c4].
Lemma 4.1 (ac[c4]). Suppose } is closed under empty class and adjunction. Let
D(}) consist of those classes Y such that:
∀Y

⊆Y ∀Z∈} (Y

∪ Z) ∈ }.
9
Then, D(}) is closed under empty class, adjunction and union and downwards
closed under ⊆. As the consequence of the last fact, closure under intersection,
class subtraction, etc., are preserved from } to D(}).
Proof. Closure under empty class, singletons, and downward closure under ⊆ are
trivial. We verify closure under ∪. Suppose Y
0
and Y
1
are in D(}). Let Y


(Y
0
∪ Y
1
). Then (Y

∩ Y
1
) ⊆ Y
1
. Consider any Z in }. We find that (Y

∩ Y
1
) ∪ Z
is in }. Since, (Y

∩ Y
0
) ⊆ Y
0
, we find that
(Y

∪ Z) = ((Y

∩ Y
0
) ∪ ((Y

∩ Y
1
) ∪ Z)) ∈ }.
We may conclude that Y
0
∪ Y
1
is in D(}). 2
8
Note that X in N may be represented as

Y , m in M. However, the precise representation of
X is immaterial for our argument.
9
Here ‘(Y

∪ Z) ∈ Y’ reads: (Y

∪ Z) exists and is in Y.
HUME’S PRINCIPLE, BEGINNINGS 9
We start with eqnum[c8]. We first follow the bootstrap of the previous section to
gain eqnum[c5, c8, e3, e4, e5, e6]. Note that c8 is universal in the class quantifiers
and, hence, always preserved under contractions of the totality of classes.
To work conveniently we contract our classes to add closure under X ∩ Y , X ¸ Y ,
X ∪ Y . We leave this as an easy exercise to the reader. So, we may work in
eqnum
+
:= eqnum[c4, c6, c7, c8, e3, e4, e5, e6]. (Note that c5 is derivable.)
We work in eqnum
+
. We consider the collection of classes A
5
, such that X is in A
5
iff, for all Y, Y

, if X ⊆ Y ≡ Y

, then there is an X

, such that X ≡ X

⊆ Y

.
Clearly, the empty class is in A
5
. We show that A
5
is closed under adjunction.
Consider X ∈ A
5
and any x. Without loss of generality we may assume that x ,∈ X.
Suppose X ∈ A
5
and (X ∪ ¦x¦) ⊆ Y ≡ Y

. Then, for some X

, X ≡ X

⊆ Y

. If
X

= Y

, then X ≡ Y , and hence, by Dedekind finiteness, X = Y , contradicting
the fact that x ∈ Y and x ,∈ X. So, for some x

, x

∈ Y

and x

,∈ X

. We have
(X ∪ ¦x¦) ≡ (X

∪ ¦x

¦) ⊆ Y

.
Consider the class A
6
of all X such that for all X

, Y, Y

with X ≡ X

, Y ≡ Y

,
X ⊆ Y , X

⊆ Y

, we have (Y ¸ X) ≡ (Y

¸ X

).
It is clear that A
6
contains the empty class. We prove closure under adjunction.
Consider X ∈ A
6
and an arbitrary x. We may assume that x ,∈ X. Consider
X

, Y, Y

with (X ∪ ¦x¦) ≡ X

, Y ≡ Y

, (X ∪ ¦x¦) ⊆ Y , X

⊆ Y

. We find
that X

is not empty. Suppose x

∈ X

. We find that X ≡ (X

¸ ¦x

¦), X ⊆ Y ,
(X

¸ ¦x

¦) ⊆ Y

. Hence (Y ¸ X) ≡ (Y

¸ (X

¸ ¦x

¦). It follows that:
(Y ¸ (X ∪ ¦x¦)) = ((Y ¸ X) ¸ ¦x¦) ≡ ((Y

¸ (X

¸ ¦x

¦)) ¸ ¦x

¦) = (Y

¸ X

).
Let A
7
:= D(A
5
∩ A
6
). We find that A
7
is closed under empty class, adjunction,
union and downwards closed under ⊆. Since the existential quantifier in the defi-
nition of A
5
is ⊆-bounded and subtraction of classes in the definition of A
6
brings
us to subclasses , relativization to A
7
preserves eqnum
+
and gains us e7 and e8. So
we may work in eqnum

:= eqnum
+
[e7, e8].
We are now ready to develop the theory of the ordering of the cardinal numbers.
We work in eqnum

. We define X ≤ Y by ∃X

, Y

(X ≡ X

∧ Y ≡ Y

∧ X

⊆ Y

).
Clearly, ≡ is a congruence for ≤. It is immediate that X ≤ X and that, if X ⊆ Y ,
then X ≤ Y .
We prove that X ≤ Y iff, for some X

, we have X ≡ X

⊆ Y . Suppose
X ≡ X

⊆ Y

≡ Y . By e8, we may find an X

such that X

≡ X

⊆ Y . Hence,
X ≡ X

⊆ Y . The converse direction is trivial. It follows that:
X ≤ Y ≤ Z → ∃X

, Y

(X ≡ X

⊆ Y ≡ Y

⊆ Z)
→ ∃X

(X ≡ X

≡ X

⊆ Y

⊆ Z)
→ X ≤ Z
Suppose that X ≤ Y and Y ≤ X. It follows that X ≡ X

⊆ Y ≡ Y

⊆ X. We find
that, for some X

, X ≡ X

⊆ Y

⊆ X. Since X is Dedekind finite, it follows that
X

= X and, hence, that Y

= X. We may conclude that X ≡ Y .
Thus we see that we have derived the principles q8, q9, q10, under the interpretation
of numbers as classes modulo equinumerosity.
Here is a useful further insight. Suppose X ≤ (Y
0
∪ Y
1
). We show that there
are X
0
, X
1
, such that X = (X
0
∪ X
1
) and X
0
≤ Y
0
and X
1
≤ Y
1
. We have
10 ALBERT VISSER
X ≡ X

⊆ (Y
0
∪ Y
1
). It follows that (Y
0
∩ X

) ⊆ X

≡ X. Hence, for some X
0
, we
have (Y
0
∩ X

) ≡ X
0
⊆ X. It follows that:
X
1
:= (X ¸ X
0
) ≡ (X

¸ (Y
0
∩ X

)) = (X

¸ Y
0
) ⊆ Y
1
.
It follows that X
1
≤ Y
1
.
We prove a lemma.
Lemma 4.2 (eqnum

). Suppose } is closed under empty class and adjunction. Let
D
+
(}) consist of those classes Y such that:
∀Y

≤Y ∀Z∈} (Y

∪ Z) ∈ }.
Then, D
+
(}) is closed under empty class, adjunction and union and downwards
closed under ≤ and hence a fortiori under ⊆. As the consequence of this last fact,
D
+
(}) is closed under subtraction, intersection, etc.
Proof. Closure under empty class and downward closure under ≤ are trivial. It
is also easy to see that all singletons are in D
+
(}). We verify closure under ∪.
Suppose Y
0
and Y
1
are in D
+
(}). Let Y

≤ (Y
0
∪ Y
1
). We can find Y

0
, Y

1
, such
that (Y

0
∪ Y

1
) = Y

and Y

0
≤ Y
0
and Y

1
≤ Y
1
. Consider any Z ∈ }. Since Y
1
is
in D
+
(}), we find that (Y

1
∪ Z) ∈ }. Since Y
0
∈ D
+
(}), we obtain (Y

∪ Z) =
(Y

0
∪ (Y

1
∪ Z)) ∈ }. We may conclude that Y
0
∪ Y
1
is in D
+
(}). 2
We proceed with the development of the arithmetic of 0, successor and addition.
We define 0 := ∅. Clearly 0 is the minimal element of ≤.
We define SXY by ∃x (x ,∈ X ∧ Y ≡ (X ∪ ¦x¦)). It is easy to see that S is a total
injective function modulo ≡ and that ∅ is not in the range of S. We also have that
each non-empty class has a predecessor. By Dedekind finiteness, we find that, if
SXY , then X ,≡ Y . Thus we have, via the obvious interpretation, axioms q1, q2,
q7, q14.
We show that, if X ≤ Y , then either X ≡ Y or, for some Z, we have SXZ and
Z ≤ Y . Suppose X ≤ Y . We have X ≡ X

⊆ Y , for some X

. In case X

= Y ,
we have X ≡ Y and we are done. Otherwise, there is an x

, such that x

∈ Y and
x

,∈ X

. Pick any x ,∈ X. We have (X ∪ ¦x¦) ≡ (X

∪ ¦x

¦) ⊆ Y , and hence
(X ∪ ¦x¦) ≤ Y . Clearly SX(X ∪ ¦x¦). Thus, we have proved the interpretation of
q11.
Suppose SXZ. Then, clearly X ≤ Z. Suppose X ≤ Y ≤ Z. It follows that
either X ≡ Y or Z ≡ SX ≤ Y ≤ Z, and, hence Z ≡ Y . So X is an immediate
≤-predecessor of Z.
To obtain the linearity of ≤ modulo ≡, we must again contract our classes. Let A
8
be the totality of all classes X such that, for all Y , X ≤ Y or Y ≤ X. Clearly, ∅
is in A
8
. We show that A
8
is closed under successor and, hence, under adjunction.
Suppose X ∈ A
8
and SXZ. Consider any Y . In case Y ≤ X, we immediately
have Y ≤ Z. Suppose X ≤ Y . Then, either X ≡ Y , in which case Y ≤ Z, or
Z ≤ Y . We note that S and ≤ are absolute under contraction to ≤-downwards
closed totalities of classes. We contract to D
+
(A
8
). We preserve eqnum

and gain
linearity of ≤. We proceed to work in eqnum

[q12], where of course we read q12
modulo the interpretation we developed.
Before defining addition, we first contract our classes to gain two desirable proper-
ties. We consider the collection of classes A
9
of all X such that for all X

, Y, Y

, if
HUME’S PRINCIPLE, BEGINNINGS 11
X # Y , X

# Y

, X ≡ X

, Y ≡ Y

, then (X ∪ Y ) ≡ (X

∪ Y

). Clearly, ∅ ∈ A
9
.
We prove that A
9
is closed under adjunction. Suppose X ∈ A
9
, (X ∪ ¦x¦) # Y ,
X

# Y

, (X ∪¦x¦) ≡ X

, Y ≡ Y

. If x ∈ X, we are immediately done, so suppose
x ,∈ X. Clearly, X

is not empty. Let x

∈ X

. We have (X # Y ), (X

¸¦x

¦) # Y

,
X ≡ (X

¸ ¦x

¦), Y ≡ Y

. So (X ∪ Y ) ≡ ((X

¸ ¦x

¦) ∪ Y

). Since x ,∈ (X ∪ Y ) and
x

,∈ ((X

¸ ¦x

¦) ∪ Y

), we find ((X ∪ ¦x¦) ∪ Y ) ≡ (X

∪ Y

).
We define A
10
as the totality of all classes X such that:
∀Y ∃Y

(Y ≡ Y

∧ (X ∩ Y

) = ∅).
Clearly, ∅ is in A
10
. We show that A
10
is closed under adjunction. Let X ∈ A
10
and
x ,∈ X. Consider any Y . Suppose Y

≡ Y and (X ∩ Y

) = ∅. By the no-universe
principe, there is a y not in X ∪ ¦x¦ ∪ Y

and there is a y

not in X ∪ ¦x, y¦ ∪ Y

.
We take Y

:= (Y

¸ ¦x¦) ∪ ¦y, y

¦.
We contract our classes to D
+
(A
9
∩ A
10
). Note that the existential quantifier
in the definition of A
10
can be ≤-bounded. It follows that we directly interpret
eqnum

[q12, e9, e10]. We proceed in this theory.
We define:
• AXY Z :↔ ∃X

, Y

((X

∩ Y

) = ∅ ∧ X ≡ X

∧ Y ≡ Y

∧ (X

∪ Y

) ≡ Z).
We note that ≡ is a congruence relation for A. Using e9 and e10, one can easily
show that:
AXY Z ↔ ∃Y

((X ∩ Y

) = ∅ ∧ Y ≡ Y

∧ (X ∪ Y

) ≡ Z).
Moreover, by e9 and e10, A defines a total function modulo ≡.
Trivially AX∅Z iff X ≡ Z. So we have the interpretation of q3.
We show that SXZ iff AX¦y¦Z. We have AX¦y¦Z iff, for some Y

, ¦y¦ ≡ Y

and
(X∩Y

) = ∅. It follows that Y = ¦y

¦, for some y

with y

,∈ X. So Z ≡ (X∪¦y

¦).
The converse is even easier.
It is immediate that addition is commutative. We show that it is associative. Sup-
pose e.g. AXY Z and AZUV . We may find X

, Y

, Z

, U

, V

such that X

, Y

, U

are pairwise disjoint and X

∪ Y

= Z

and Z

∪ U

= V

and X ≡ X

, Y ≡ Y

,
U ≡ U

, V ≡ V

. Let W

:= Y

∪ U

. Then we have AY

U

W

and AX

W

V

.
Ergo, AY UW

and AXW

V . The converse is similar.
We note that the previous two results imply the interpretation of q4. Moreover, we
have the interpretations of q15 and q16.
We show X ≤ Y iff, for some Z, AXZY . Suppose X ≤ Y . Then, X ≡ X

⊆ Y .
It is easily seen that AX(Y ¸ X

)Y . Conversely, suppose AXZY . Then, for some
disjoint X

, Z

, we have X ≡ X

, Z ≡ Z

and (X

∪ Z

) ≡ Y . We can find an X

,
with X

≡ X

⊆ Y . So X ≡ X

⊆ Y , and hence X ≤ Y . Thus, by commutativity,
we have interpreted q13.
We leave the easy verification of the interpretation of q17 to the reader.
In summary: we have directly interpreted in eqnum[c8] the principles q1-4, and
q7-17.
12 ALBERT VISSER
5. Development of Multiplication in HL
The basic insight of this section is something very familiar to anyone who worked
out his or her own arithmetization. To develop recursion for a function F such
that x < Fx, we do not need computation sequences, since computation sets are
sufficient. In the present context the idea is not just convenient but essential.
We start with the theory HL. By the development in the previous section, we can
contract our classes to gain the principles of eqnum

[q12, e9, e10]. Thus, we also
have q1-4, and q7-17 available for the interpretations of the arithmetical operations
and relations specified in the previous section. Without loss of generality, we may
also add h3. We call the resulting theory HL

. We work in HL

.
We define ^ as the virtual class of x such that ∃X Xx. We define, for x, y in ^,
x · y iff ∃Z (Z x ∧Z y). Clearly, · is an equivalence relation with domain ^.
We find that gives us a bijection between the classes modulo ≡ and ^ modulo
·. This means that we can induce Z and and ≤ and S and A on ^ modulo ·. We
call the resulting relations Z

, ≤

, S

, and A

.
We turn to the definition of multiplication. We will first give the definition and
then add some motivating remarks. We define MXY Z iff, either X = 0 and Z = 0,
or X ,= 0 and, there is an Y

≡ Y , and a z ,∈ Y

with Z z, and an x with X x,
such that:
i. Y

⊆ ^,
ii. For all y, y

∈ (Y

∪ ¦z¦), if y · y

, then y = y

.
iii. For all y ∈ (Y

∪ ¦z¦), Z

y or, for some y

∈ (Y

∪ ¦z¦), A

y

xy and, for all w,
if y

<

w <

y, then w ,∈ Y

.
iv. For all y ∈ Y

, z >

y.
The idea is as follows. First, the case that X is empty is special. So, we set
that aside. If X is not empty, X times Y is Z (modulo equinumerosity) if there
is a computation that witnesses this fact. The computation is a set of numbers
Y

∪ ¦z¦. We demand that the computation starts with a zero and ends with a
number z representing the cardinality of Z. We have a normalizing condition (ii)
that says that every number has at most one representative in the computation.
This is needed since a number could have many ‘copies’ in our set up where we only
have a Hume relation. Item (iii) tells us that our computation proceeds by adding
a number of size X in each ‘step’. The bit in (iii) that there are no non-intended
intervening elements between two elements of a step is just there because in the
weak context we can not prove that. We build it in just in order not to have to
prove it. Clause (iv) is another case of ‘no intervening elements’. It is added for
the same reason as the extra bit in (iii).
We note that ≡ is a congruence for M.
We first prove that MX0Z iff Z = 0. In case X = 0, we are immediately done, so
suppose X ,= 0. We have MX0Z iff, for some z with Z z and x with X x, we
have Z

z or A

zxz. The second case is excluded by q17. So Z

z, and hence, Z = 0.
Suppose MXY Z, SY U, AZXV . We show that MXUV . This is immediate when
X = 0. So, suppose X ,= 0. Let Y

, z, x be the promised witnesses for MXY Z.
Suppose V v and A

zxv. We consider U

:= Y

∪ ¦z¦. Clearly, U ≡ U

. Since
HUME’S PRINCIPLE, BEGINNINGS 13
v >

z >

y

, for any y

in Y

, we find that v ,· u

, for any u

in U

. It follows that
U

, v, x, V satisfy conditions (i,ii,iii,iv) for MXUV .
Suppose MXUV and SY U. We show that, for some Z, MXY Z and AZXV . In
case X = 0 this is immediate. Suppose X ,= 0. It is easily seen that, since U ,= 0,
we also have V ,= 0. Let U

, v, x, V satisfy the conditions for MXUV . By (iii) and
the fact that not Z

(v), there is a z ∈ U

such that A

zxv. Let Y

:= U

¸ ¦z¦.
Clearly Y ≡ Y

. Let Z z. Then, Y

, z, x, Z witness MXY Z. Moreover AZXV .
We now consider A
11
. The class of all X such that, for all U, there is, modulo
≡, a unique Z, such that MUXZ. By the previous results, it is clear that A
11
is
closed under 0 and successor. We contract our classes to D
+
(A
11
), thus gaining the
uniqueness clause for multiplication. The relation M is absolute for this contraction
(and similar contractions). We note a subtlety here: the totality ^ is not absolute
but contracts to an initial segment that is closed under addition. However, for the
question whether we have MXY Z relativized to A
11
, we only need to worry about
elements below z. For elements below z the old and the new numbers coincide.
Remark 5.1. At this point we could take a shortcut and obtain the desired inter-
pretation of Q by employing the result of [
ˇ
Sve07]. However, since we are so close
to the the end of our development, we will finish it. Moreover, we get a little bit
more in this way, since we will also preserve full HL

.
We let A
12
be the totality of classes X, such that:
∀Y ≤X∀U, V (MUXV → ∃W ≤ V MUY W).
Clearly, 0 is in A
12
. Suppose X is in A
12
. We show that any X

with SXX

is in
A
12
. Suppose that MUX

V . In case U = 0, we are immediately done, so we may
assume that U ,= 0. Suppose Y ≤ X

. We can easily prove that Y ≤ X or Y = X

.
In the second case we are immediately done. In the first case, we can find a Z, such
that MUXZ and AZUV . It follows that, for some W ≤ Z, we have MUY W, since
Z ≤ V , we find W ≤ V .
We relativize to D
+
(A
12
), thus obtaining the principle
(p1) ¬ ∀Y ≤X∀U, V (MUXV → ∃W ≤ V MUY W).
Let A
13
consist of those X such that, for all U, X

, X

, Z, Z

, Z

, if AXX

X

,
AZZ

Z

, MUXZ and MUX

Z

, then MUX

Z

. It is easy to see that 0 is in A
13
.
We prove that A
13
is closed under successor. Suppose X ∈ A
13
, SXY , AY Y

Y

,
AZZ

Z

, MUY Z and MUY

Z

. Let AXY

X

. We find that, for some W, we
have MUXW and AUWZ. Let AWZ

W

. We may conclude that MUX

W

. We
note that SX

Y

, since SXY and AXY

X

and AY Y

Y

. Also AW

UZ

, since
AWZ

W

and AWUZ and AZZ

Z

. We may conclude that MUY

Z

.
We relativize to D
+
(A
13
). Thus obtaining the principle:
(p2) ¬ (AXX

X

∧ AZZ

Z

∧ MUXZ ∧ MUX

Z

) → MUX

Z

.
Let A
14
be the totality of all X, such that, MWY U and MUXV implies that there
is a Z, with MYXZ and MWZV .
It is easily seen that 0 is in A
15
. Let X ∈ A
14
. Suppose SXX

. MWY U and
MUX

V

. Then, for some V , MUXV and AV UV

. It follows that for some Z,
MY XZ and MWZV . We find that, for some Z

, MY X

Z

and AZY Z

. We have
14 ALBERT VISSER
AZY Z

, AV UV

, MWZV , MWY U. Hence, MWZ

V

. If we contract our classes
to D
+
(A
14
), we gain the principle:
(p3) ¬ (MWY U ∧ MUXV ) ∧ MY XZ) → MWZV ).
Finally, let A
15
consist of all X, such that, for all U, there is a Z with MUXZ. We
can easily see that A
14
is downwards closed w.r.t. ≤ and is closed under 0, singletons
and addition. It follows that A
14
is closed under union, since if A(X ¸ Y )Y Z, then
(X ∪ Y ) ≡ Z. We show that multiplication is defined on A
14
and that it is closed
under multiplication. Suppose that X
0
, X
1
∈ A
14
. We have MX
0
X
1
X, for some
X. We also have, for some Z
0
and Z, MUX
0
Z
0
and MZ
0
X
1
Z. By p3, we have:
MUXZ.
When we contract our classes to X
15
, we preserve HP

and gain q5, q6, q18 and
q19. Thus we end with HP

[q5, q6, q18, q19], which contains Q.
Conversely, since Q interprets I∆
0
+ Ω
1
, we can show that Q interprets HL.
6. An Application
We show that the presence of a certain preorder on our domain is equivalent (in
the sense of mutual direct interpretability) to Hume’s principle.
We expand the signature of eqnum with new symbols ^ of type o, Z

of type
o, and ≤

of type oo. We add axioms stating that ≤

is a linear preorder on ^
with Z

the non-empty virtual class of its initial elements. We demand that every
element of ^ has strict ≤

-successors. We call the resulting theory eqnum
ord
. Let
the induced equivalence relation of ≤

be ·.
We show that HL is directly interpretable in eqnum
ord
. We work in eqnum
ord
. We
first contract our classes so that we gain subtraction of elements plus the fact that
≡ is an equivalence relation. This enables us to work in eqnum
ord
[c5, e3, e4, e5].
We say that X is an x-class iff, x ∈ ^, and, for all x

∈ X, x

<

x, and, for all
x

<

x, there is precisely one x

· x

, such that x

∈ X.
We consider the totality ^
0
of all x in ^, such that for all x-classes X

and X

,
we have X

≡ X

. Clearly, all elements of Z

are in ^
0
.
Suppose that x is in ^
0
and that y is a direct successor of x. We show that y is
in ^
0
. Suppose Y

and Y

are y-classes. Since x is is a direct predecessor of y, we
will have direct predecessors of y in Y

and Y

, say these are x

, respectively x

.
We note that x · x

and x · x

. Clearly, Y

¸ ¦x

¦ and Y

¸ ¦x

¦ are x-classes.
Hence, (Y

¸ ¦x

¦) ≡ (Y

¸ ¦x

¦). But then also Y

≡ Y

.
Let A

be the totality of classes X for which there is a y in ^
0
and a Y , such
that Y is a y-class and X ≡ Y . It is easy to see that A

is closed under empty
class and adjunction. Moreover, a Y that witnesses that X is in A

, will be itself
in A

.
We contract our classes to A

. We gain the following principle: for every X
there is a y ∈ ^
0
, and a y-class Y , such that X ≡ Y .
We define X y iff, y ∈ ^
0
and, for some y-class Y , we have X ≡ Y . We have
already seen that we must have h1. Suppose X y and X

y. Let Y and Y

witness this fact. Since, y ∈ ^
0
, we have Y ≡ Y

. Hence, X ≡ X

. This gives us
h2.
We apply the above insight to show that Presburger Arithmetic PresA does not
interpret eqnum[c8]. Our argument is is quick and easy. Probably inspection of
the quantifier elimination for PresA could yield much stronger results. To exclude
HUME’S PRINCIPLE, BEGINNINGS 15
any possible interpretation we must take into account that our interpretations could
have special features like the presence of parameters or being piecewise. Fortunately,
we can work in a structure in which every element is definable and that has infinitely
definable elements. This means that interpretations with parameters and pieces can
always be replaced by interpretations without parameters and pieces.
Suppose PresA does interpret eqnum[c8], say via K. We may assume that K
is a parameter-free, one-piece interpretation. However, K may still be more-
dimensional. We consider the internal model / given by K in the standard model
N
add
:= ¸ω, 0, S, +, ≤).
The object domain of / will consist of a definable infinite totality of tuples
(n
0
, . . . , n
k−1
). We can define, in N
add
, the following ordering on these tuples:
(n
0
, . . . , n
k−1
) _ (m
0
, . . . , m
k−1
) iff (n
0
+ . . . + n
k−1
) < (m
0
+ . . . + m
k−1
) or
((n
0
+. . . +n
k−1
) = (m
0
+. . . +m
k−1
) and (n
0
< m
0
or (n
0
= m
0
and n
1
< m
1
) or
. . . (n
0
= m
0
and . . . n
k−1
≤ m
k−1
)). This gives our domain order type ω. Thus
we can extend our interpretation of eqnum[c8] to an interpretation of eqnum
ord
.
This gives us an interpretation of HL and, hence, of Q in PresA. Quod impossibile,
since Q is essentially undecidable and PresA is decidable.
References
[Bur05] John Burgess. Fixing Frege. Princeton Monographs in Philosophy. Princeton University
Press, Princeton, 2005.
[FV59] S. Feferman and R.L. Vaught. The first order properties of products of algebraic systems.
Fundamenta Mathematicae, 47:57–103, 1959.
[Lin04] Øystein Linnebo. Predicative fragments of frege arithmetic. Bulletin of Symbolic Logic,
10(2):153–174, 2004.
[Nel86] E. Nelson. Predicative arithmetic. Princeton University Press, Princeton, 1986.
[Smo91] C. Smory´ nski. Logical Number Theory I, an Introduction. Universitext. Springer, New
York, 1991.
[Sol76] R. M. Solovay. Interpretability in set theories. Unpublished letter to P. H´ ajek,
http://www.cs.cas.cz/~hajek/RSolovayZFGB.pdf, 1976.
[
ˇ
Sve07] V.
ˇ
Svejdar. An interpretation of Robinson’s Arithmetic in Grzegorczyk’s weaker variant.
Fundamenta Informaticae, 81:347–354, 2007.
[Vis09] A. Visser. Cardinal arithmetic in the style of baron von M¨ unchhausen. Review of Symbolic
Logic, 2(3):570–589, 2009. doi: 10.1017/S1755020309090261.
Appendix A. Presburger Arithmetic
Presburger Arithmetic, PresA, for the natural numbers is axiomatized by the
following principles.
10
The signature consists of 0, 1 and +.
PresA1 ¬ x + 1 ,= 0,
PresA2 ¬ x +z = y +z → x = y,
PresA3 ¬ x + 0 = x,
PresA4 ¬ x + (y +z) = (x +y) +z,
PresA5 ¬ x = 0 ∨ ∃y x = y + 1,
PresA6 ¬ x +y = y +x,
PresA7 ¬ ∃z (x +z = y ∨ x = y +z),
PresA8 for any n ≥ 2, we have ¬ ∃y (x = ny ∨ x = ny + 1 ∨ . . . x = ny + (n−1)).
10
I took this axiomatization principles from Clemens Grabmayer’s Master’s Thesis. See
Clemens’ site http://www.phil.uu.nl/~clemens/. I only changed the order a bit and omitted
a superfluous axiom.
16 ALBERT VISSER
We call the instances of PresA8: PresA8n. We easily show that, for n, m ≥ 2,
PresA8n and PresA8m are equivalent, over the other axioms, to PresA8(nm). It
follows that any finite set of instances of PresA8 is equivalent to a single instance.
We call the theory PresA1,2,3,4,5,6,7,8n: PresAn. One can show that, for n > 1,
PresA8(n + 1) is not derivable from PresA8n. See e.g. [Smo91].
We show how to interpret PresAn (for n ≥ 2) in Q
add
. We work in Q
add
. First let
¸
0
be the virtual class of all x such that for all y, z, (y +z) +x = y + (z +x) and
y +x = z +x → y = z and 0 +x = x and 1 +x = x + 1. We easily prove that ¸
0
contains 0, 1 and is closed under addition and predecessor. Relativizing to ¸
0
gives
us axioms PresA1, 2, 3, 4, 5, ¬ 0 + x = x and ¬ x + 1 = 1 + x. This theory clearly
extends Q
add
. We work in the extended theory. We will follow the usual convention
of omitting brackets, which is justified by associativity.
Let ¸
1
be the class of all x such that, for all y, y + x = x + y. We easily see that
¸
1
is closed under 0, 1, + and predecessor. Relativizing to ¸
1
, gives us axioms
PresA1, 2, 3, 4, 5, 6. We proceed to work in this system.
We define x ≤ y by: ∃z x+z = y. It is easy to verify that ≤ is a partial order with
minimum 0 and that addition is monotonic w.r.t. ≤.
Let ¸
2
be the class of all x such that, for all z ≤ x and for all y, z ≤ y or y ≤ z.
Clearly, ¸
2
is downwards closed under ≤ and 0 and 1 are in ¸
2
. Suppose x
0
and
x
1
are in ¸
2
. We show that (x
0
+ x
1
) ∈ ¸
2
. Consider any z ≤ (x
0
+ x
1
) and any
y. In case z ≤ x
0
we are immediately done. Suppose x
0
≤ z, say z = x
0
+ u.
It follows that u ≤ x
1
. In case y ≤ u, we have y ≤ z. Suppose u ≤ y, say
u + v = y. If v ≤ x
0
, we have y ≤ z; if v ≥ x
0
, we have y ≥ z. Relativizing
to ¸
2
, gives us axioms PresA1,2,3,4,5,6,7. For PresA7, note that it is equivalent to
∀x, y (x ≤ y ∨ y ≤ x) and that ≤ is absolute under relativization to downwards
closed sets, since x ≤ y ↔ ∃z≤y x +z = y.
Finally, let ¸
3
be the class of all x, such that, for all z ≤ x, we have:
∃y (z = ny ∨ z = ny + 1 ∨ . . . z = ny + (n−1)).
It is easy to see that ¸
3
is downward closed under ≤ and that 0, 1 are in J
3
. We
show that ¸
3
is closed under addition. Suppose x
0
, x
1
∈ J
3
and z ≤ (x
0
+x
1
). In
case z ≤ x
0
, we are immediately done. Otherwise, z = x
0
+u, for some u. It follows
that u ≤ x
1
. We have, for some y
0
, y
1
and for some standard k
0
, k
1
, x
0
= ny
0
+k
0
,
u = ny
1
+k
1
. Hence,
z = x
0
+u = ny
0
+k
0
+ny
1
+k
0
= n(y
0
+y
1
) +k
0
+k
1
.
In case (k
0
+k
1
) < n, we are done. Otherwise 0 ≤ (k
0
+k
1
−n) < n, and we find
z = n(y
0
+y
1
+ 1) +k
0
+k
1
−n.
We note that the quantifier ∃y in axiom PresA8n can be bounded by x. It follows
that relativization to ¸
3
, gives us PresAn.
Appendix B. Questions
(1) Is ac[c8] interpretable in PresA?
(2) Is the complexity of deciding the theory of finite classes with equinumerosity
over an infinite domain higher than the complexity of deciding PresA?
(3) Is PresA interpretable in Q
add
? I conjecture: no.
HUME’S PRINCIPLE, BEGINNINGS 17
(4) Is an extension of Q
add
interpretable in Q
add
that contains a faster-than-
linear function? I conjecture: no.
Department of Philosophy, Utrecht University, Heidelberglaan 8, 3584 CS Utrecht,
The Netherlands
E-mail address: Albert.Visser@phil.uu.nl

there is an informative reduction relation. For certain extensions of both theories we do indeed have tighter connections. a suitable extension of Qadd still has to be analyzed. there are. but that we cannot develop Q.2 ALBERT VISSER objects. in the context of a theory of classes with equinumerosity. we use the presence of relations (i) to define the relation of equinumerosity and (ii) to implement the arithmetical operations on classes.2. Here. This methodology was initiated by Edward Nelson in his book [Nel86]. Of course. The primary focus of the present paper is the fine structure of reasoning. In other words. work falling under blue can provide useful data for research of color yellow. Almost Philosophical Considerations. 1The system Q add is a subsystem of Presburger Arithmetic PresA. making it blue. We show in an appendix that Qadd locally interprets PresA. the analogue of Q for addition. Thus. of sets. . the fact that we work in very weak theories does not necessarily reflect a conviction that these theories can be justified. without multiplication.3 Blue is the study of the fine structure of reasoning. say blue and yellow. Hume’s principle is connected to what one could call the reverse mathematics of addition and multiplication. One important aspect of both Burgess’ development of Q from a predicative theory of binary relations expanded with Hume’s Principle and the development in our paper is that we grant ourselves the freedom of gaining more ‘good’ properties of our numbers by contracting the number domain. theory reduction. then we can develop Qadd . etcetera. very roughly. We show that the first theory interprets the second. so. if we start with adjunctive class theory with a primitively given equinumerosity relation plus the principle that says that there is no universal class. in Section 6. Here we have reverse mathematics. we really do need the less restrictive Feferman-Orey relation between the theories. Burgess’ book [Bur05] exhibits both colors.1 To define multiplication. Thus. two colors. 2 It would be interesting to explore the existence of tighter relations. in a class context.. We do not develop the full picture here.2 The precise relationship between adjunctive class theory with equinumerosity and no-universe and. We can get a somewhat more refined and more interesting picture. Hume’s principle is not needed for (c) after all. that the second theory does not interpret the first. Of course. Such a program involves matching ‘natural’ pairs of class theories with pairs of a theory of addition and a theory of addition and multiplication. the separation is not strict. 1. In the kind of foundational research exemplified by this paper. on the other hand. E. Hence. if we do not work with relations but only with classes. We show. in the other direction. or that we cannot justify much more. Hume’s principle is essential for (c). along the lines sketched by Feferman and Vaught in their paper [FV59] for the case of true theories. A lot of contemporary work on Frege’s Grundgesetze falls under yellow. Yellow has a tighter link with philosophy: we are looking for a serious justification of numerical reasoning. we need Hume’s principle. We present one clear result: the adjunctive theory of classes with equinumerosity and Hume’s principle is mutually interpretable with Q. interpretability and the like. It seems plausible that. 3The choice of colors is inspired by a famous scene from Monty Python and the Holy Grail. thus transforming Solovay’s method of shortening cuts (introduced in [Sol76]) into a foundational tool. We show that.g.

say. it would be interesting to see whether there is a coherent foundational idea corresponding to the methodology of shortening cuts.g. Theories will be specified as e. In a Frege-style program. After all.5 When working with this method one develops the feeling that such a philosophy is just around the corner. the precise definitions employed by Frege. Not every class of objects having some desired properties usually associated with the natural numbers qualifies. and suppose we want them to have the good property P . we provide a list of principles and theories. Linnebo provides an interesting discussion of Burgess’ proof that predicative relational Frege theory interprets Q. Nelson’s program as developed in [Nel86] seems to be founded on the conviction that. there is a foundational idea that justifies building up the numbers by ‘number system hopping’. . rather than to fiddle around with the definition to get better properties. The claim that such-and-such a definition is the intended one. We note that the preferred definitions view is not without its own problems. [a3. In this section. it could be our aim to see what certain principles allow us to prove about the numbers as given by some fixed definition. BEGINNINGS 3 In the context of a Frege-style development of number theory (of color yellow) this strategy is not unproblematic. b2] where a3 and b2 are already specified principles. The signature of the theory will 4For a paper that studies preferred definitions for a Frege-style development of Arithmetic. Suppose further we have proved P in a very weak context for a ‘contracted’ virtual class of natural numbers N1 . where Burgess’ definition of proto-natural numbers (p156 of [Bur05]) has much more a cardinal flavour: it is a strengthened version of Dedekind finiteness. see Øystein Linnebo’s paper [Lin04]. we want to say what the natural numbers are. we could be interested in a reconstruction that justifies this specific reasoning and not in justifying quite different reasoning with the same conclusion. Unfortunately. Suppose. Given that Frege’s development is intended to build the cardinals. For example. On the other hand. Then a good theory in which to get P for N0 could be a theory in which N1 and N0 coincide. This would involve rejecting the claim that induction is almost analytical for the natural numbers (made by Linnebo in his paper [Lin04]). which is the better one? Let me stress that the kind of blue project we are involved in is not necessarily antithetical to a yellow view involving intended definitions. 5Induction in the case discussed by Linnebo is not really induction anyway. 2. Preliminaries This note will presuppose familiarity with the notions and notations introduced in [Vis09]. we have a preferred definition N0 of.. Nelson gives us no hint what such an idea could look like. is in need of informally rigorous argumentation. Frege’s definition of the natural numbers using the ancestral seems to me to be an ordinal-style definition. e. e.. So we immediately know that a principle that implies the equality of N0 and N1 will have numbers with the good property P . We will use the following convention. in the light of the restriction on comprehension. yes.g.g. we are interested in defining the natural numbers.4 Similarly. the natural numbers.HUME’S PRINCIPLE. The best reading strategy is to glance through this section and return to it as soon as a principle is added. Thus.

with the subset ordering and with the relation of equinumerosity. 2. We write X # Y for ∀z (z ∈ X ∨ z ∈ Y ). If a theory U is given U [a3. e1 (X ≡ ∅ ∨ ∅ ≡ X) ↔ X = ∅ e2 (x ∈ X ∧ y ∈ Y ) → ((X ∪ {x}) ≡ (Y ∪ {y}) ↔ X ≡ Y ) e3 X ≡ X e4 X ≡ Y → Y ≡ X e5 (X ≡ Y ∧ Y ≡ Z) → X ≡ Z e6 Y ⊂ X → Y ≡ X (Dedekind finiteness axiom) e7 (X ⊆ Y ≡ Y ) → ∃X (X ≡ X ⊆ Y ) e8 (X ≡ X ∧ Y ≡ Y ∧ X ⊆ Y ∧ X ⊆ Y ) → (Y \ X) ≡ (Y \ X ) e9 (X # Y ∧ X # Y ∧ X ≡ X ∧ Y ≡ Y ) → (X ∪ Y ) ≡ (X ∪ Y ) e10 ∃Y (Y ≡ Y ∧ X # Y ) The theory eqnum is ac[e1. We start with class theories. since we can always interpret identity on the classes as extensional equivalence. Adding Equinumerosity. 2.e2]. . the sort of classes. for any cardinality κ. See [Bur05] for an exposition of o the original theorem. ac+ . where we interpret the objects as the atoms of the ⊆ordering. The theory ac+ can easily be shown to be complete via quantifier elimination by a minor adaptation of the proof of the L¨wenheim-Behman theorem. show that. . We have identity for each sort and one binary predicate ∈ of sort oc. the sort of objects and c. is decidable. over classes.1. b2] will be the result of adding a3 and b2 to U . y.6 We also note that axiom c3 is really superfluous.2. range over objects and X. In the statement of some of these principles we use operations on classes. We have the following list of principles. This means that there are interpretations between the two theories whose compositions are definably and provably isomorphic with the identity interpretation. with just classes and the relation ⊆. Feferman and Vaught in their classical paper [FV59]. We note that it has a one-sorted version. c3 ∀z (z ∈ X ↔ z ∈ Y ) → X = Y (extensionality) c4 ∃Z ∀z (z ∈ Z ↔ (z ∈ X ∧ z ∈ Y )) (closure under intersection) c5 ∃Y ∀y (y ∈ Y ↔ (y ∈ X ∧ y = x)) (subtraction axiom) c6 ∃Z ∀z (z ∈ Z ↔ (z ∈ X ∧ z ∈ Y )) (closure under class-subtraction) c7 ∃Z ∀z (z ∈ Z ↔ (z ∈ X ∨ z ∈ Y )) (closure under union). .c3]. z. say. We will only use the principles in the context of axioms stating that these operations are total. We let x. We have a language with two sorts: o. c1 ∃X ∀x x ∈ X (empty class axiom) c2 ∃Y ∀y (y ∈ Y ↔ (y ∈ X ∨ y = x)) (adjunction axiom). . We can interpret eqnum into these theories by identifying the objects 6If we set it up correctly this one-sorted version is bi-interpretable with the original theory. . c8 ∃x x ∈ X (no-universe axiom) The theory ac is [c1. the theory of all classes of elements from a domain of size κ. then c2 and c5 can be replaced by an axiom stating the existence of singletons.4 ALBERT VISSER be the minimal signature that is involved in all the axioms of the theory. Class Theories. We note that if we have the theory with all our principles. We extend the language of ac with a binary relation symbol ≡ of type cc. Y. We have the following principles for the extended language.c2. . . Z.

Here is a list of principles. We introduce adjunctive relation theory ar. for all x. So by dropping functionality we have a substantial gain in generality. We consider the following list of principles in the inclusive signature 0. Adjunctive relation theory is a two-sorted theory with sorts o. So. Moreover.3. We write Rxy for app(R. we would get some properties of the equinumerosity relation like reflexivity and symmetry for free. . Moreover. if Rxy. ×. Our theory has identity for each sort and a ternary relation app of type roo. q1 Sx = 0 q2 Sx = Sy → x = y q3 x + 0 = x q4 x + Sy = S(x + y) q5 x × 0 = 0 q6 x × Sy = (x × y) + x q7 x = 0 ∨ ∃y x = y + 1 q8 x ≤ x q9 (x ≤ y ∧ y ≤ z) → x ≤ z q10 (x ≤ y ∧ y ≤ x) → x = y q11 x ≤ y → (x = y ∨ Sx ≤ y) q12 x ≤ y ∨ y ≤ x . We remind the reader of a result of [Vis09]. We note that we could eliminate ≡ from the language by redefining X ≡ Y as ∃z (X z ∧ Y z). functionality is a rather strong demand. . ar[c8] is ar plus ∃x ¬ Rxx. h1 ∃x X x h2 (X z ∧ Y z) → X ≡ Y h3 X ≡ Y z → X z The theory Hume Light or HL is eqnum[h1. in the context of ZF without choice we need Scott’s trick to produce a Hume function. then x = y. We often write x · y or xy for x × y. +. to range over objects and R. y. S. y). . We will add principles concerning classes to ar as a matter of course under this reading. ≤. BEGINNINGS 5 with the atoms of ⊆. Arithmetical Theories.HUME’S PRINCIPLE. We use x. Relational Theories. We choose to work with a Hume relation as opposed to a Hume function because the functionality never seems to play any role in the arguments. Clearly this interpretation does yield ac for our classes. . as relations such that. v (Suv ↔ (Ruv ∨ (x = u ∧ y = v))) r3 ∃S ∀u. to range over relations. z. There it is shown that ar plus the nouniverse axiom interprets Q as a theory of cardinals over the given object domain. and r. Here is a list of principles. It follows that eqnum has a decidable extension and.5. y. 2. 2. 2. i.4. x. This would ask for a careful choice of axioms. This relation is of type co. we add the Hume relation to the language. hence. Adding the Hume Relation.h2].e. y ¬Rxy r2 ∃S ∀u. the sort of objects. does not interpret Q. .r2]. e. S. which depends on a lot of details of the theory. r1 ∃R ∀x. the sort of relations. To formulate Hume’s principe. In ar we can define classes as diagonal relations. v (Suv ↔ (Ruv ∧ (x = u ∨ y = v))) The theory ar is given by [r1..g. For example. .

S. q4. Thus. Relativizing to X2 gives us eqnum[e3. We work in ac[c4]. Ergo X ≡ (Z \ {z}). preserved under contraction of the totality of classes.7 7We note that in in eqnum[c5] plus the no-universe principle. c5]. We work in eqnum[e3. We start to work in ac. We relativize our classes to X0 . We repeat the procedure of the previous subsection to regain the subtraction axiom: we get eqnum[c5. Let y ∈ Y and z ∈ Z. define X0 as the virtual class of classes such that X is in X0 iff. We find that X ≡ (Y \ {y}) and (Y \ {y}) ≡ (Z \ {z}). To be precise. thereby gaining the axioms c4 and c5. we can directly interpret eqnum[c5] in eqnum. q6] in the signature with 0. e5].e4. We note that this argument uses the associativity of intersection in the strong sense that if (X ∩ Y ) ∩ Z is defined. It is easy to see that X1 is closed under empty class. We define X2 as the totality of those classes X such that: X ≡ X and ∀Y (X ≡ Y → Y ≡ X) and ∀Y. where X is in X1 if. we are immediately done. in the signature with 0. we can directly prove that X2 is closed under subtraction. 3. We note that this property is preserved under any further contraction of our classes. Contracting our classes to X2 we gain the axioms stating that ≡ is an equivalence relation. which we use here as a matter of course. +. and intersection.1. ×. and vice versa. We show that X2 is closed under adjunction. We easily see that X0 is closed under empty class. It is easy to see that ∅ ∈ X2 . but does not contain the universe of objects.e5 are universal and.6 ALBERT VISSER q13 q14 q15 q16 q17 q18 q19 x ≤ y ↔ ∃z (z + x = y) Sx = x (x + y) + z = x + (y + z) x+y =y+x x+z =y+z →x=y x × (y + z) = (x × y) + (x × z) x × (y × z) = (x × y) × z The theory Qadd is the theory [q1. e4. then so is X ∩ (Y ∩ Z) and both are equal. We note that Y and Z cannot be empty. If x is in X. Bootstrap in eqnum. (X ∪ {x}) ≡ Z. q3. e5]. Here c5 tells us that classes are closed under subtraction. +. eqnum[c5. The theory Q is Qadd [q5. for all x. Suppose x ∈ X. 3. We treat the case of transitivity. Z ((X ≡ Y ∧ Y ≡ Z) → X ≡ Z). X \ {x} exists. e5]. By the result of the previous subsection. We work in eqnum[c5]. As a first step. hence. i. No-Universe We first show that a version of Hume’s Principle allows us to construct a totality of classes that is adjunctive. thereby gaining the extra axiom c4: that classes are closed under intersection. e3.2. adjunction. e4. we have directly interpreted ac[c4. for all Y . Suppose (X ∪ {x}) ≡ Y and Y ≡ Z. We note that e3. q2. (We need the presence of intersection for closure under subtraction!) We relativize to X1 . . e4. X ∩ Y exists. 3. S. Hence. we show that eqnum directly interprets the no-universe axiom ac8. q7]. We consider X1 . Bootstrap in ac. intersection and subtraction. adjunction. The equality relies on extensionality. c8].e.

e3. Consider X ∪ {x}. we have x ∈ X. Suppose Y would be the universe. we may conclude that ar[r3. we have X y and X ⊂ X ∪ {x}. We say that X is Dedekind finite iff. so we can work in HL[c5. e5. Let X3 consist of the Dedekind finite classes. In another formulation: suppose the universe is a Dedekind finite class. that all classes are Dedekind finite. Let Y be the promised witness of this fact. and hence X ⊂ Y .1. Bootstrap in HL. If y ∈ Y .1 (in eqnum). by Lemma 3. By replacing X y by ∃Z X ≡ Z y. We find that (X0 \ {x0 }) ⊂ X and (X0 \ {x0 }) ≡ X.HUME’S PRINCIPLE. We may assume that x ∈ X. Proof. we gain the axiom h3. thus gaining the axiom e6. We show that X ∪ {x} is Dedekind finite. BEGINNINGS 7 We work in eqnum[c5. Suppose X y. 3. there is an X0 ⊂ X such that X0 y. We find y ∈ Y . e4. If it were. for some y. e3. Consider y ∈ Y ∪ {y}. for every y ∈ Y . 2 3. then any class equinumerous to it is again the universe. e4. So X0 ≡ X.h1. we have X ≡ Y . when we add Hume’s principle in the form of h1 and h2 for equinumerosity-asdefined. e3. Thus. Clearly. X ≡ Y and x ∈ X.3. e6]. Suppose X is in X4 . we contract our classes to X4 . ∅ ∈ X4 . we can find X0 ⊂ X ⊂ X ∪ {x} with X0 y. X ≡ Y and x ∈ X. Suppose Y is Dedekind finite. e5. It is immediate that ar[r3] verifies the axioms e1 and e2. Suppose X0 ⊂ (X ∪ {x}) and X0 ≡ (X ∪ {x}). for some X0 ⊂ X. y ∈ Y . e4. that. By the preceding development. Suppose X is in X4 . for every Y ⊂ X. First. e5. h3]. for some x. e5. Let x0 be x in case x ∈ X0 and let x0 be some arbitrary element of X0 otherwise. . So we may work in HL+ := HL[c5. We consider adjunctive-subtractive relation theory ar[r3]. A contradiction with the assumption that X is Dedekind finite. But X ≡ Y . we note that y ∈ Y . So we have directly interpreted eqnum[c5. such that there is a class Y such that X ≡ Y and. e4. contradicting the fact that X is Dedekind finite. It is easy to see that X3 is closed under subtraction. the empty class is in X3 . At this point we switch to the theory HL. We define the equinumerosity of X and Y as the existence of a bijection between X and Y . This gives us HP[c8]. Bootstrap in adjunctive-subtractive relation theory. Let Y y. We show that X3 are closed under adjunction. We may assume that x ∈ X. then we have (X ∪ {x}) ⊆ Y . e3. Lemma 3. e6. It follows. Clearly. Let X4 consist of all X. Ergo. We prove that no element of X4 is the universe. e3. Suppose X is Dedekind finite. e4.4. Let Y witness that X is in X4 . Our development in this subsection is a straightforward adaptation of the treatment predicative second order logic with the Hume function in [Bur05]. We relativize to X3 . If y = y. contradicting the fact that Y is Dedekind finite.h2] directly interprets ar plus the nouniverse axiom c8. by the result of the previous subsection. e6]. We show that X4 is closed under adjunction. We note a useful fact. Then. We show that Y ∪ {y} has the desired property. we obtain a direct interpretation of HL. Finally. Thus. we can directly interpret eqnum[c5. Suppose Y is Dedekind finite. we would have X y and X0 y. e6]. It follows that X0 is not empty. (X ∪ {x}) ≡ (Y ∪ {y}). e5]. In this theory we treat classes as diagonal relations.

We fix a finite set of parameters of the interpretation in M.h2]. we show that we can go on to locally interpret full Presburger Arithmetic PresA in Qadd . We find that (Y ∩ Y1 ) ∪ Z is in Y. Proof. etc. . Quod impossibile. We need the subtraction axiom r3 in our development since the definition of equinumerosity in this context contains an existential quantifier over relations. Can we eliminate the use of the subtraction axiom r3 in our result? 4. We will provide a bit more of the theory of addition since this flows naturally from the development.h2] is mutually interpretable with Q.8 Take any permutation σ of the domain that fixes the parameters. Suppose Y is closed under empty class and adjunction. we know that Q interprets I∆0 + Ω1 .h1. However.h1. are preserved from Y to D(Y). and downward closure under ⊆ are trivial.. In Appendix A. Then (Y ∩ Y1 ) ⊆ Y1 . Thus.h2]. Without loss of generality. Let D(Y) consist of those classes Y such that: ∀Y ⊆Y ∀Z∈Y (Y ∪ Z) ∈ Y. X N n .h1. we may assume that the parameters are objects.h1.h2] interprets Q as a theory of cardinals. any permutation of the objects of N that fixes the parameters extends to an automorphism of N . Consider a standard model M of ar[c8] with an infinite domain and all finite relations over the domain. singletons. closure under intersection. We note that ar[c8] does not directly interpret ar[r3. we can find a standardly finite set X in N and an element n in N such that X N n and the elements of X (according to N ) and n are disjoint from the parameters. Lemma 4. Using familiar methods one may interpret ar[r3.1 (ac[c4]). class subtraction. we may assume that we also have h3.1. any relation in M is definable from finitely many objects. In the converse direction. Development of the Theory of Addition in eqnum[c8] In this section we show how to interpret Qadd in eqnum[c8]. for any non-parameter n . Suppose it did.h2] in this last theory. Since every permutation of the objects in M extends to an automorphism of M. we find that ar[r3. we find that (Y ∪ Z) = ((Y ∩ Y0 ) ∪ ((Y ∩ Y1 ) ∪ Z)) ∈ Y. So. It follows that. adjunction and union and downwards closed under ⊆. We begin with a lemma that works already in ac[c4]. Let Y ⊆ (Y0 ∪ Y1 ). We may conclude that Y0 ∪ Y1 is in D(Y). Consider any Z in Y. We have X ≡N σX N σn. Open Question 3. (Y ∩ Y0 ) ⊆ Y0 . Since. the precise representation of X is immaterial for our argument. m in M. Closure under empty class. We verify closure under ∪. and since the objects of N coincide with the objects of M. Suppose Y0 and Y1 are in D(Y). we find that ar[r3.h1. we might loose axiom h2 when we try to obtain subtraction of relations via contraction.9 Then. D(Y) is closed under empty class. As the consequence of the last fact.8 ALBERT VISSER By the results of [Vis09]. equinumerosity is not absolute for contraction of relations. 2 8Note that X in N may be represented as Y . For this reason. Clearly. 9Here ‘(Y ∪ Z) ∈ Y’ reads: (Y ∪ Z) exists and is in Y. Since. The interpretation provides us with an internal model N of ar[r3.

Y (X ≡ X ∧ Y ≡ Y ∧ X ⊆ Y ). Consider X ∈ X5 and any x. We have . X ≡ X ⊆ Y ⊆ X. It follows that: (Y \ (X ∪ {x})) = ((Y \ X) \ {x}) ≡ ((Y \ (X \ {x })) \ {x }) = (Y \ X ). c7. We consider the collection of classes X5 . Y ≡ Y .HUME’S PRINCIPLE. The converse direction is trivial. e8]. e4. hence. for some X . union and downwards closed under ⊆. if X ⊆ Y ≡ Y . X1 . Y ≡ Y . we may find an X such that X ≡ X ⊆ Y . e5. X ∪ Y . then X ≤ Y . Y . for all Y. Y (X ≡ X ⊆ Y ≡ Y ⊆ Z) ∃X (X ≡ X ≡ X ⊆ Y ⊆ Z) → X≤Z Suppose that X ≤ Y and Y ≤ X. We have (X ∪ {x}) ≡ (X ∪ {x }) ⊆ Y . we have (Y \ X) ≡ (Y \ X ). e4. Then. Suppose X ∈ X5 and (X ∪ {x}) ⊆ Y ≡ Y .) We work in eqnum+ . X ⊆ Y . We may assume that x ∈ X. ≡ is a congruence for ≤. Y with (X ∪ {x}) ≡ X . the empty class is in X5 . We prove that X ≤ Y iff. (X ∪ {x}) ⊆ Y . under the interpretation of numbers as classes modulo equinumerosity. Here is a useful further insight. such that X = (X0 ∪ X1 ) and X0 ≤ Y0 and X1 ≤ Y1 . X = Y . X ≡ X ⊆ Y . We are now ready to develop the theory of the ordering of the cardinal numbers. It follows that: X≤Y ≤Z → → ∃X . Thus we see that we have derived the principles q8. Y with X ≡ X . We find that X7 is closed under empty class. Let X7 := D(X5 ∩ X6 ). e3. By e8. Clearly. It follows that X ≡ X ⊆ Y ≡ Y ⊆ X. We first follow the bootstrap of the previous section to gain eqnum[c5. Note that c8 is universal in the class quantifiers and. Without loss of generality we may assume that x ∈ X. for some X . BEGINNINGS 9 We start with eqnum[c8]. We leave this as an easy exercise to the reader. So. So we may work in eqnum := eqnum+ [e7. then X ≡ Y . We find that X ≡ (X \ {x }). and hence. e5. for some X . X ⊆ Y . it follows that X = X and. always preserved under contractions of the totality of classes. x ∈ Y and x ∈ X . e6]. for some x . X ⊆ Y . contradicting the fact that x ∈ Y and x ∈ X. We find that. Consider X ∈ X6 and an arbitrary x. Since X is Dedekind finite. (X \ {x }) ⊆ Y . X \ Y . We define X ≤ Y by ∃X . We work in eqnum . Suppose x ∈ X . q10. Consider X . we have X ≡ X ⊆ Y . c8. It is clear that X6 contains the empty class. hence. we may work in eqnum+ := eqnum[c4. Suppose X ≡ X ⊆ Y ≡ Y . X ⊆ Y . It is immediate that X ≤ X and that. We show that there are X0 . X ≡ X ⊆ Y . So. then there is an X . such that X is in X5 iff. (Note that c5 is derivable. if X ⊆ Y . We may conclude that X ≡ Y . Y. c6. We show that X5 is closed under adjunction. that Y = X. adjunction. If X = Y . relativization to X7 preserves eqnum+ and gains us e7 and e8. Since the existential quantifier in the definition of X5 is ⊆-bounded and subtraction of classes in the definition of X6 brings us to subclasses . Hence. such that X ≡ X ⊆ Y . c8. Suppose X ≤ (Y0 ∪ Y1 ). q9. Consider the class X6 of all X such that for all X . by Dedekind finiteness. Hence (Y \ X) ≡ (Y \ (X \ {x }). e6]. To work conveniently we contract our classes to add closure under X ∩ Y . Clearly. e3. We find that X is not empty. We prove closure under adjunction. Y.

We may conclude that Y0 ∪ Y1 is in D+ (Y). Suppose Y is closed under empty class and adjunction. we first contract our classes to gain two desirable properties. Thus we have. It is also easy to see that all singletons are in D+ (Y). We define 0 := ∅. if + . By Dedekind finiteness. then either X ≡ Y or. Suppose Y0 and Y1 are in D+ (Y). we obtain (Y ∪ Z) = (Y0 ∪ (Y1 ∪ Z)) ∈ Y. if SXY . Clearly. Suppose X ≤ Y ≤ Z. X ≤ Y or Y ≤ X. clearly X ≤ Z. We contract to D+ (X8 ). It follows that (Y0 ∩ X ) ⊆ X ≡ X. we have proved the interpretation of q11. Suppose X ≤ Y . hence Z ≡ Y . Clearly 0 is the minimal element of ≤. for some Z. Lemma 4. where of course we read q12 modulo the interpretation we developed. In case X = Y . such that x ∈ Y and x ∈ X . we have X ≡ Y and we are done. We show that X8 is closed under successor and. We show that. q2. Then. Let Y ≤ (Y0 ∪ Y1 ). in which case Y ≤ Z. successor and addition. we have (Y0 ∩ X ) ≡ X0 ⊆ X. Then. we have SXZ and Z ≤ Y . axioms q1. adjunction and union and downwards closed under ≤ and hence a fortiori under ⊆. and. We can find Y0 . and hence (X ∪ {x}) ≤ Y . Otherwise. It follows that: X1 := (X \ X0 ) ≡ (X \ (Y0 ∩ X )) = (X \ Y0 ) ⊆ Y1 . Consider any Z ∈ Y. We note that S and ≤ are absolute under contraction to ≤-downwards closed totalities of classes. we find that (Y1 ∪ Z) ∈ Y. Since Y0 ∈ D+ (Y). q7. for all Y . there is an x . Closure under empty class and downward closure under ≤ are trivial. We preserve eqnum and gain linearity of ≤. We proceed to work in eqnum [q12]. Consider any Y . To obtain the linearity of ≤ modulo ≡. We consider the collection of classes X9 of all X such that for all X . via the obvious interpretation. Clearly SX(X ∪ {x}). hence. We have (X ∪ {x}) ≡ (X ∪ {x }) ⊆ Y . D+ (Y) is closed under subtraction. Suppose X ∈ X8 and SXZ. So X is an immediate ≤-predecessor of Z. It follows that X1 ≤ Y1 . Y. then X ≡ Y . Y . or Z ≤ Y . etc. Proof.10 ALBERT VISSER X ≡ X ⊆ (Y0 ∪ Y1 ). under adjunction. we find that. Let D+ (Y) consist of those classes Y such that: ∀Y ≤Y ∀Z∈Y (Y ∪ Z) ∈ Y. Suppose SXZ. we must again contract our classes.2 (eqnum ). We have X ≡ X ⊆ Y . We define SXY by ∃x (x ∈ X ∧ Y ≡ (X ∪ {x})). Pick any x ∈ X. As the consequence of this last fact. We prove a lemma. Thus. D (Y) is closed under empty class. Let X8 be the totality of all classes X such that. We verify closure under ∪. It is easy to see that S is a total injective function modulo ≡ and that ∅ is not in the range of S. q14. we immediately have Y ≤ Z. It follows that either X ≡ Y or Z ≡ SX ≤ Y ≤ Z. Hence. Before defining addition. intersection. for some X0 . either X ≡ Y . if X ≤ Y . ∅ is in X8 . In case Y ≤ X. Since Y1 is in D+ (Y). 2 We proceed with the development of the arithmetic of 0. Then. such that (Y0 ∪ Y1 ) = Y and Y0 ≤ Y0 and Y1 ≤ Y1 . Suppose X ≤ Y . Y1 . for some X . We also have that each non-empty class has a predecessor.

suppose AXZY . Y ≡ Y . We define: • AXY Z :↔ ∃X . By the no-universe principe. We define X10 as the totality of all classes X such that: ∀Y ∃Y (Y ≡ Y ∧ (X ∩ Y ) = ∅). Y . we have interpreted q13. Y . So (X ∪ Y ) ≡ ((X \ {x }) ∪ Y ). Using e9 and e10. Suppose Y ∗ ≡ Y and (X ∩ Y ∗ ) = ∅. e9. for some Y .HUME’S PRINCIPLE. So X ≡ X ⊆ Y . Y ≡ Y . y }. We note that ≡ is a congruence relation for A. for some Z. X is not empty. X ≡ (X \ {x }). and q7-17. We have AX{y}Z iff. AY U W and AXW V . Let x ∈ X . Y ≡ Y . In summary: we have directly interpreted in eqnum[c8] the principles q1-4. So Z ≡ (X ∪ {y }). The converse is even easier. we have X ≡ X . Then. there is a y not in X ∪ {x} ∪ Y ∗ and there is a y not in X ∪ {x. and hence X ≤ Y . A defines a total function modulo ≡. Y ((X ∩ Y ) = ∅ ∧ X ≡ X ∧ Y ≡ Y ∧ (X ∪ Y ) ≡ Z). Moreover. so suppose x ∈ X. If x ∈ X. Let W := Y ∪ U . We can find an X . for some disjoint X . It is easily seen that AX(Y \ X )Y . So we have the interpretation of q3. (X ∪ {x}) ≡ X . Moreover. X ≡ X . It follows that Y = {y }. We note that the previous two results imply the interpretation of q4. X # Y . Then we have AY U W and AX W V . V such that X . Suppose X ≤ Y . Conversely. Suppose X ∈ X9 . We show that SXZ iff AX{y}Z. U . We show that it is associative. y} ∪ Y ∗ . We contract our classes to D+ (X9 ∩ X10 ). Clearly. Let X ∈ X10 and x ∈ X. for some y with y ∈ X. Trivially AX∅Z iff X ≡ Z. It is immediate that addition is commutative. We leave the easy verification of the interpretation of q17 to the reader. Clearly. X # Y . Clearly. V ≡ V . Z ≡ Z and (X ∪ Z ) ≡ Y . Y ≡ Y . Ergo.g. Consider any Y . we have the interpretations of q15 and q16. AXY Z and AZU V . BEGINNINGS 11 X # Y . Since x ∈ (X ∪ Y ) and x ∈ ((X \ {x }) ∪ Y ). Thus. one can easily show that: AXY Z ↔ ∃Y ((X ∩ Y ) = ∅ ∧ Y ≡ Y ∧ (X ∪ Y ) ≡ Z). U are pairwise disjoint and X ∪ Y = Z and Z ∪ U = V and X ≡ X . with X ≡ X ⊆ Y . We may find X . (X \ {x }) # Y . we find ((X ∪ {x}) ∪ Y ) ≡ (X ∪ Y ). (X ∪ {x}) # Y . Z . U ≡ U . The converse is similar. then (X ∪ Y ) ≡ (X ∪ Y ). Z . We have (X # Y ). Note that the existential quantifier in the definition of X10 can be ≤-bounded. ∅ is in X10 . X ≡ X ⊆ Y . We prove that X9 is closed under adjunction. we are immediately done. AXZY . We show X ≤ Y iff. ∅ ∈ X9 . Then. by e9 and e10. It follows that we directly interpret eqnum [q12. Suppose e. We take Y := (Y ∗ \ {x}) ∪ {y. e10]. We show that X10 is closed under adjunction. . We proceed in this theory. by commutativity. {y} ≡ Y and (X ∩ Y ) = ∅.

we do not need computation sequences. Y ⊆ N . The idea is as follows. we set that aside. Development of Multiplication in HL The basic insight of this section is something very familiar to anyone who worked out his or her own arithmetization. First. e10]. and A . and hence. we are immediately done. the case that X is empty is special. z > y. iv. for x. since computation sets are sufficient. Suppose MXY Z. ii. X times Y is Z (modulo equinumerosity) if there is a computation that witnesses this fact. S . Without loss of generality. is an equivalence relation with domain N . then w ∈ Y . So. Clause (iv) is another case of ‘no intervening elements’. or X = 0 and. We define MXY Z iff. ≤ . We show that MXU V . we have Z z or A zxz. Clearly. We define. We note that ≡ is a congruence for M. for some y ∈ (Y ∪ {z}). z. we can contract our classes to gain the principles of eqnum [q12. iii. We find that gives us a bijection between the classes modulo ≡ and N modulo . We start with the theory HL. U ≡ U . Since . It is added for the same reason as the extra bit in (iii). SY U . Suppose V v and A zxv. then y = y . We define N as the virtual class of x such that ∃X X x.12 ALBERT VISSER 5. and an x with X x. For all y. This is needed since a number could have many ‘copies’ in our set up where we only have a Hume relation. for some z with Z z and x with X x. suppose X = 0. if y y . there is an Y ≡ Y . We will first give the definition and then add some motivating remarks. Z y or. Z = 0. e9. We work in HL . and a z ∈ Y with Z z. for all w. y in N . AZXV . We have MX0Z iff. The second case is excluded by q17. We demand that the computation starts with a zero and ends with a number z representing the cardinality of Z. The computation is a set of numbers Y ∪ {z}. Clearly. To develop recursion for a function F such that x < F x. So Z z. The bit in (iii) that there are no non-intended intervening elements between two elements of a step is just there because in the weak context we can not prove that. Thus. So. We turn to the definition of multiplication. Item (iii) tells us that our computation proceeds by adding a number of size X in each ‘step’. we also have q1-4. if y < w < y. we may also add h3. such that: i. In the present context the idea is not just convenient but essential. In case X = 0. x y iff ∃Z (Z x ∧ Z y). This is immediate when X = 0. We call the resulting theory HL . We first prove that MX0Z iff Z = 0. x be the promised witnesses for MXY Z. We call the resulting relations Z . This means that we can induce Z and and ≤ and S and A on N modulo . By the development in the previous section. either X = 0 and Z = 0. We consider U := Y ∪ {z}. For all y ∈ (Y ∪ {z}). If X is not empty. We have a normalizing condition (ii) that says that every number has at most one representative in the computation. Let Y . y ∈ (Y ∪ {z}). For all y ∈ Y . so suppose X = 0. A y xy and. and q7-17 available for the interpretations of the arithmetical operations and relations specified in the previous section. We build it in just in order not to have to prove it.

Then. BEGINNINGS 13 v > z > y . since U = 0. thus gaining the uniqueness clause for multiplication. We find that. Let X14 be the totality of all X. Suppose Y ≤ X . x. we have MU XW and AU W Z. thus obtaining the principle (p1) ∀Y ≤X ∀U. Suppose SXX . for the question whether we have MXY Z relativized to X11 . We note that SX Y . Y . AZZ Z . V satisfy conditions (i. We relativize to D+ (X12 ). Clearly Y ≡ Y . modulo ≡. v. The class of all X such that. for all U . for some Z . Let AXY X . It follows that for some Z. It follows that. We may conclude that MU X W . Suppose X is in X12 . since Z ≤ V . Let Y := U \ {z}. we find W ≤ V . z. such that: ∀Y ≤X ∀U. Let Z z. We now consider X11 . if AXX X . such that MU XZ. such that MU XZ and AZU V . for all U . By (iii) and the fact that not Z (v). Then. We relativize to D+ (X13 ). Remark 5.1. Moreover AZXV . At this point we could take a shortcut and obtain the desired interˇ pretation of Q by employing the result of [Sve07]. MW Y U and MU X V . we find that v u . we are immediately done. It follows that U . V (MU XV → ∃W ≤ V MU Y W ). Thus obtaining the principle: (p2) (AXX X ∧ AZZ Z ∧ MU XZ ∧ MU X Z ) → MU X Z . We prove that X13 is closed under successor. We show that any X with SXX is in X12 . there is a z ∈ U such that A zxv. Z . since AW Z W and AW U Z and AZZ Z . Also AW U Z . MXY Z and AZXV . Let U . then MU X Z . it is clear that X11 is closed under 0 and successor. x. In the first case. Z witness MXY Z. such that. It is easy to see that 0 is in X13 . MW Y U and MU XV implies that there is a Z. Moreover.HUME’S PRINCIPLE.iii. we also have V = 0. It is easily seen that 0 is in X15 . we have MU Y W . Suppose X ∈ X13 . Let X13 consist of those X such that. In case U = 0. a unique Z. X . AZZ Z . We contract our classes to D+ (X11 ). SXY . Z . for some Z. for some W ≤ Z. so we may assume that U = 0. MY X Z and AZY Z . MU Y Z and MU Y Z . Z. V satisfy the conditions for MXU V . MU XV and AV U V . However. Suppose MXU V and SY U . we can find a Z. x. We can easily prove that Y ≤ X or Y = X . for any u in U . It is easily seen that. Clearly. v. Let AW Z W . We find that. we only need to worry about elements below z. We may conclude that MU Y Z . However. Suppose X = 0. since we are so close to the the end of our development. We show that. there is. AY Y Y . We have . We note a subtlety here: the totality N is not absolute but contracts to an initial segment that is closed under addition. for any y in Y . with MYXZ and MW ZV . By the previous results. Let X ∈ X14 . we get a little bit more in this way. Suppose that MU X V . 0 is in X12 .iv) for MXU V . for some V . For elements below z the old and the new numbers coincide. X . MU XZ and MU X Z . since SXY and AXY X and AY Y Y . We let X12 be the totality of classes X. MY XZ and MW ZV . The relation M is absolute for this contraction (and similar contractions). for some W .ii. V (MU XV → ∃W ≤ V MU Y W ). In case X = 0 this is immediate. In the second case we are immediately done. since we will also preserve full HL . we will finish it.

t. we have X ≡ Y . Since x is is a direct predecessor of y. we have Y ≡ Y . We define X y iff. Suppose Y and Y are y-classes. Hence. We have MX0 X1 X. such that for all x-classes X and X . 6. and a y-class Y . We add axioms stating that ≤ is a linear preorder on N with Z the non-empty virtual class of its initial elements. for some Z0 and Z. We apply the above insight to show that Presburger Arithmetic PresA does not interpret eqnum[c8]. We note that x x and x x . x ∈ N . MW ZV . We show that HL is directly interpretable in eqnumord . since if A(X \ Y )Y Z. all elements of Z are in N0 . we gain the principle: (p3) (MW Y U ∧ MU XV ) ∧ MY XZ) → MW ZV ). X ≡ X . q6. X1 ∈ X14 . ≤ and is closed under 0. By p3. We call the resulting theory eqnumord . q18 and q19. e5]. say these are x . we will have direct predecessors of y in Y and Y . We say that X is an x-class iff. We contract our classes to X .14 ALBERT VISSER AZY Z . MU X0 Z0 and MZ0 X1 Z. will be itself in X . which contains Q. Thus we end with HP [q5. Probably inspection of the quantifier elimination for PresA could yield much stronger results. We gain the following principle: for every X there is a y ∈ N0 . q18. q19]. It follows that X14 is closed under union. Finally. We demand that every element of N has strict ≤ -successors. We can easily see that X14 is downwards closed w. respectively x . Hence. We consider the totality N0 of all x in N . for some X. e4. This gives us h2. for some y-class Y . Suppose that X0 . and. Let the induced equivalence relation of ≤ be . y ∈ N0 and. such that x ∈ X. We first contract our classes so that we gain subtraction of elements plus the fact that ≡ is an equivalence relation. We have already seen that we must have h1. MW Y U . such that. x < x. Suppose X y and X y. Let Y and Y witness this fact. such that X ≡ Y . This enables us to work in eqnumord [c5. Clearly. MW Z V . If we contract our classes to D+ (X14 ). We expand the signature of eqnum with new symbols N of type o. AV U V . y ∈ N0 . Our argument is is quick and easy. and. Hence. we preserve HP and gain q5. for all U . since Q interprets I∆0 + Ω1 . Y \ {x } and Y \ {x } are x-classes. It is easy to see that X is closed under empty class and adjunction. e3. we have: MU XZ. a Y that witnesses that X is in X . We show that multiplication is defined on X14 and that it is closed under multiplication. let X15 consist of all X. Since. we can show that Q interprets HL. such that Y is a y-class and X ≡ Y . We work in eqnumord . Conversely. then (X ∪ Y ) ≡ Z. Moreover. q6. Z of type o. there is precisely one x x . for all x < x.r. To exclude . Suppose that x is in N0 and that y is a direct successor of x. (Y \ {x }) ≡ (Y \ {x }). Clearly. But then also Y ≡ Y . there is a Z with MU XZ. for all x ∈ X. we have X ≡ X . singletons and addition. An Application We show that the presence of a certain preorder on our domain is equivalent (in the sense of mutual direct interpretability) to Hume’s principle. We show that y is in N0 . We also have. When we contract our classes to X15 . Let X be the totality of classes X for which there is a y in N0 and a Y . and ≤ of type oo.

Quod impossibile. Interpretability in set theories. K may still be moredimensional. PresA1 x + 1 = 0. . . . . . + mk−1 ) or ((n0 + . . . The object domain of K will consist of a definable infinite totality of tuples (n0 . We consider the internal model K given by K in the standard model Nadd := ω. Princeton University Press. Smory´ ski. Springer. nk−1 ). I only changed the order a bit and omitted a superfluous axiom. 2004.nl/~clemens/. We can define. nk−1 ≤ mk−1 )). Solovay. 10(2):153–174. . Suppose PresA does interpret eqnum[c8]. See Clemens’ site http://www. of Q in PresA. one-piece interpretation. x = n y + (n−1)). Svejdar. 1976. This means that interpretations with parameters and pieces can always be replaced by interpretations without parameters and pieces. . 2007. . Unpublished letter to P. 10 I took this axiomatization principles from Clemens Grabmayer’s Master’s Thesis. in Nadd . . This gives us an interpretation of HL and.phil. References [Bur05] John Burgess. M. for the natural numbers is axiomatized by the following principles. [Vis09] A. New n York. PresA8 for any n ≥ 2. [Smo91] C. + nk−1 ) < (m0 + . say via K. Feferman and R. Cardinal arithmetic in the style of baron von M¨nchhausen.cas. 1991. BEGINNINGS 15 any possible interpretation we must take into account that our interpretations could have special features like the presence of parameters or being piecewise. 81:347–354. doi: 10. . PresA4 x + (y + z) = (x + y) + z. Predicative fragments of frege arithmetic.HUME’S PRINCIPLE. . However. PresA3 x + 0 = x. S.10 The signature consists of 0. Fundamenta Mathematicae. +. This gives our domain order type ω. . ˇ ˇ [Sve07] V. [FV59] S. . Presburger Arithmetic Presburger Arithmetic. we can work in a structure in which every element is definable and that has infinitely definable elements. 2(3):570–589. . 1959. Thus we can extend our interpretation of eqnum[c8] to an interpretation of eqnumord . an Introduction. since Q is essentially undecidable and PresA is decidable. mk−1 ) iff (n0 + . [Sol76] R. + nk−1 ) = (m0 + . Fixing Frege. Fortunately. PresA. + mk−1 ) and (n0 < m0 or (n0 = m0 and n1 < m1 ) or . . Logical Number Theory I. Bulletin of Symbolic Logic. . [Lin04] Øystein Linnebo. . [Nel86] E. hence. Princeton. . The first order properties of products of algebraic systems. . Appendix A. a http://www. PresA2 x + z = y + z → x = y. Princeton Monographs in Philosophy. An interpretation of Robinson’s Arithmetic in Grzegorczyk’s weaker variant. ≤ . 47:57–103. 2005.cs. Fundamenta Informaticae. Universitext. H´jek. the following ordering on these tuples: (n0 . (n0 = m0 and . PresA5 x = 0 ∨ ∃y x = y + 1. .uu. . Review of Symbolic u Logic. Vaught.cz/~hajek/RSolovayZFGB. we have ∃y (x = n y ∨ x = n y + 1 ∨ . nk−1 ) (m0 .pdf . Princeton. Visser. 1986.1017/S1755020309090261. . Princeton University Press. . . 0. PresA6 x + y = y + x.L. 2009. PresA7 ∃z (x + z = y ∨ x = y + z). Predicative arithmetic. Nelson. We may assume that K is a parameter-free. 1 and +. .

7. z = n y + (n−1)). PresA8n and PresA8m are equivalent.6. z ≤ y or y ≤ z. such that.2. Relativizing to J0 gives us axioms PresA1. We define x ≤ y by: ∃z x + z = y. for some y0 .t. 1 and is closed under addition and predecessor. We work in Qadd . m ≥ 2.4. we have: ∃y (z = n y ∨ z = n y + 1 ∨ . 2. ≤. x0 = n y0 + k 0 .16 ALBERT VISSER We call the instances of PresA8: PresA8n. x1 ∈ J3 and z ≤ (x0 + x1 ). 4. Appendix B. z = x0 + u = n y0 + k 0 + n y1 + k 0 = n (y0 + y1 ) + k0 + k1 . say u + v = y. z = x0 + u. . note that it is equivalent to ∀x. For PresA7. gives us axioms PresA1. If v ≤ x0 . Otherwise. It follows that any finite set of instances of PresA8 is equivalent to a single instance. we have y ≤ z. Clearly. let J3 be the class of all x. Relativizing to J1 . we are immediately done. We note that the quantifier ∃y in axiom PresA8n can be bounded by x. It follows that u ≤ x1 . 3. It follows that u ≤ x1 .2. One can show that. Suppose x0 and x1 are in J2 . It is easy to see that J3 is downward closed under ≤ and that 0.4. This theory clearly extends Qadd . Suppose x0 . Otherwise 0 ≤ (k0 + k1 − n) < n. . 6. which is justified by associativity. Let J2 be the class of all x such that. We call the theory PresA1.8n: PresAn. In case z ≤ x0 . . We easily show that. for all z ≤ x and for all y.g. we are done. Hence. Relativizing to J2 . to PresA8(nm). (y + z) + x = y + (z + x) and y + x = z + x → y = z and 0 + x = x and 1 + x = x + 1.5. gives us PresAn. We show how to interpret PresAn (for n ≥ 2) in Qadd . We have. 1. over the other axioms. 3. 2. y (x ≤ y ∨ y ≤ x) and that ≤ is absolute under relativization to downwards closed sets. We will follow the usual convention of omitting brackets. Questions (1) Is ac[c8] interpretable in PresA? (2) Is the complexity of deciding the theory of finite classes with equinumerosity over an infinite domain higher than the complexity of deciding PresA? (3) Is PresA interpretable in Qadd ? I conjecture: no. for all y. y1 and for some standard k0 .r. and we find z = n (y0 + y1 + 1) + k0 + k1 −n. In case y ≤ u. We easily prove that J0 contains 0. We proceed to work in this system. we have y ≤ z. We show that J3 is closed under addition. u = n y1 + k 1 . Let J1 be the class of all x such that. Suppose x0 ≤ z. Consider any z ≤ (x0 + x1 ) and any y. we have y ≥ z. since x ≤ y ↔ ∃z≤y x + z = y.6. [Smo91]. PresA8(n + 1) is not derivable from PresA8n.3. First let J0 be the virtual class of all x such that for all y. We work in the extended theory. It follows that relativization to J3 . In case z ≤ x0 we are immediately done. 4. In case (k0 + k1 ) < n. 5. 5. It is easy to verify that ≤ is a partial order with minimum 0 and that addition is monotonic w. for some u. 0 + x = x and x + 1 = 1 + x. say z = x0 + u. Finally.3. See e.5. for all z ≤ x. for n. We show that (x0 + x1 ) ∈ J2 . if v ≥ x0 . + and predecessor. y + x = x + y. J2 is downwards closed under ≤ and 0 and 1 are in J2 . gives us axioms PresA1. k1 . for n > 1.7. z. Suppose u ≤ y. 1 are in J3 . We easily see that J1 is closed under 0.

The Netherlands E-mail address: Albert. Utrecht University. 3584 CS Utrecht.HUME’S PRINCIPLE.uu. Heidelberglaan 8. BEGINNINGS 17 (4) Is an extension of Qadd interpretable in Qadd that contains a faster-thanlinear function? I conjecture: no.Visser@phil. Department of Philosophy.nl .