The Blackwell Guide to

Philosophica Logic. Ed. L.
Gobble. Blackwell, Oxford.
2001, 224–257.
Intuitionistic Logic
Dirk van Dalen
1 Basic principles
There are basically two ways to view intuitionistic logic: (i) as a philosoph
ical–foundational issue in mathematics, (ii) as a technical discipline within
mathematical logic.
Let us deal with the philosophical aspects first, they will provide the
motivation for the subject. In doing so, we will follow L.E.J. Brouwer, the
founding father of intuitionism. Brouwer did contribute little to intuitionis-
tic logic as we know it from text books and papers, but he pointed the way
for his successors.
Logic in Brouwer’s intuitionism takes a secondary place, the first place is
reserved for mathematics. Mathematics has to be understood in the widest
possible sense, it is the constructional mental activity of the individual (for
a survey of the Brouwerian global philosophy of man and his mathematical
enterprise, see [Dalen 1999]). The role of logic is to note and systematically
study certain regularities in the mathematical constructional process. Con-
trary to traditional views, logic thus is dependent on mathematics and not
vice versa.
Mathematical practice has taught us that relatively few logical connec-
tives suffice for an efficient treatment of arguments. In the case of intu-
itionism, the meaning of the connectives has to be explained in terms of the
basic mathematical notion: construction. A fact A is established by means
of a construction. An easy example is 3 + 2 = 5, this is established by the
following construction: construct 3, construct 2 and compare the outcome
with the result of the construction of 5. The outcome is a confirmation of
the above equation.
The construction criteria for ‘truth’ also yields an interpretation of the
connectives. Let us write “a : A” for “a is a construction that establishes
A”, we call this a a proof of A.
1
A proof of A∧B is simply a pair of proofs a and b of A and B. For conve-
nience we introduce a a notation for the pairing of constructions, and for the
inverses (projections); (a, b) denotes the pairing of a and b, and (c)
0
,(c)
1
, are
the first and second projection of c. Now, the proof of a disjunction A ∨ B
is a pair (p, q) such that p carries the information, which disjunct is correct,
and q is the proof of it. We stipulate that p ∈ ¦0, 1¦. So p = 0 and q : A or
p = 1 and q : B. Note that this disjunction is effective, in the sense that the
disjunct is specified, this in contrast with classical logic, where one does not
have to know which disjunct holds.
Negation is also defined by means of proofs: p : A says that each proof
a of A can be converted by the construction p into a proof of an absurdity,
say 0 = 1. A proof of A thus tells us that A has no proof!
The most interesting propositional connective is the implication. The
classical solution, i.e. A → B is true if A is false or if B is true, cannot
be used because here the classical disjunction is used; moreover, it assumes
that the truth values of A and B are known before one can settle the status
of A → B.
Heyting showed that this is asking too much. Consider A = “there occur
twenty consecutive 7’s in the decimal expansion of π”, and B = “there
occur nineteen consecutive 7’s in the decimal expansion of π”. Then A∨B
does not hold constructively, but on the interpretation following here, the
implication, A → B is obviously correct.
The intuitionistic approach, based on the notion of proof, demands a
definition of a proof a of the implication A → B in terms of (possible) proofs
of A and B. The idea is quite natural: A → B is correct if we can show the
correctness of B as soon as the correctness of A has been established.
In terms of proofs: p : A → B if p transforms each proof q : A into a proof
p(q) : B. The meaning of the quantifiers is specified along the same lines.
Let us assume that we are dealing with a given domain D of mathematical
objects. A proof p of ∀xA(x) is a construction which yields for every object
d ∈ D a proof p(d) : A(d). A proof p of ∃xA(x) is a pair (p
0
, p
1
) such that
p
1
: A(p
0
). Thus the proof of an existential statement requires an instance
plus a proof of this instance.
2
a : A conditions
a :⊥ false
a : A∧ B a = (a
1
, a
2
), where a
1
: A and a
2
: B
a : A∨ B a = (a
1
, a
2
), where a
2
: A if a
1
= 0 and a
2
: B if a
1
= 1
a : A → B for all p with p : A a(p) : B
a : ∃xA(x) a = (a
1
, a
2
) and a
2
: A(a
1
)
a : ∀xA(x) for all d ∈ D a(d) : A(d) where D is a given domain
a : A for all p : A a(p) :⊥
Observe that an equivalent characterization of the disjunction can be
given: a = (a
1
, a
2
), where a
1
= 0 and a
2
: A or a
1
= 1 and a
2
: B. The above
formulation has the technical advantage that only ∧ and → are required,
and they belong to the ‘negative fragment’. The interpretation of the con-
nectives in terms of proofs was made explicit by Heyting in [Heyting 1934].
Kolmogorov had given a similar interpretation in terms of problems and
solutions, [Kolmogorov 1932]. The formulations are, up to terminology, vir-
tually identical.
We will demonstrate the proof interpretation for a few statements.
1. A → (B → A).
We want an operation p that turns a proof a : A into a proof of
B → A. But if we already have a proof a : A, then there is a simple
transformation that turns a proof q : B into a proof of A, i.e. the
constant mapping q → a, which is denoted by λq a. And so the
construction that takes a into λq a is λa (λq a), or in an abbreviated
notation λaλq a.
Hence λaλq.a : A → (B → A).
2. (A∨ B) → (B ∨ A).
Let p : A ∨ B, then p
0
= 0 and (p)
1
: A or (p)
0
= 1 and (p)
1
: B. By
interchanging A and B we get, looking for q : B ∨ A: (q)
0
= 1 and
(q)
1
: B or (q)
1
= 0 and (q)
1
: A, which comes to sg((p)
0
) = (q)
0
and
(p
1
) : B or sg((p)
0
) = (q)
0
and (q)
1
: A that is (sg((p
0
)), (p)
1
) : B∨A.
And so λp (sg((p
0
), (p)
1
) : A∨ B → B ∨ A.
3. A → A
A proof q of A is a proof of A →⊥. Assume p : A, and q : A.
Then q(p) :⊥, so λq q(p) : A. Hence λpλq q(p) : A → A.
4. A → A
This turns out to be unprovable. For assume that a : A, then we
3
know that A →⊥ has no proof, and that is all. No more information
can be extracted. Hence we cannot claim that there is a proof of A.
5. A∨ A
p : A∨ A ⇔ (p)
0
= 0 and (p)
1
: A or (p)
0
= 1 and (p)
1
: A.
However, for an arbitrary proposition A we do not know whether A
or A has a proof, and hence (p)
0
cannot be computed. So, in general
there is no proof of A∨ A.
6. ∃xA(x) → ∀xA(x)
p : ∃xA(x) ⇔ p(a) :⊥ for a proof a : ∃xA(x). We have to find
q : ∀xA(x), i.e. q(d) : A(d) →⊥ for any d ∈ D.
So pick an element d and let r : A(d), then (d, r) : ∃xA(x) and so
p((d, r)) :⊥. Therefore we put (q(d))(r) = p((d, r)), so q = λrλd
p((d, r)) and hence λpλrλd p((d, r)) : ∃xA(x) → ∀xA(x).
Brouwer handled logic in an informal way, he showed the untenability
of certain classical principles by a reduction to unproven statements. His
technique is goes by the name of ‘Brouwerian counterexample’. A few ex-
amples are shown here. Brouwer considered certain open problems, usually
formulated in terms of the decimal expansion of π. In spite of consider-
able computational power, there are enough open questions concerning the
occurrence of specific sequences of decimals.
1
Let us illustrate the technique: we compute simultaneously the decimals
of π and the members of a Cauchy sequence. We use N(k) as an abbrevia-
tion for “the decimals p
k−89
, . . . , p
k
of π are all 9”. Now we define:
a
n
=

(−2)
−n
if ∀k ≤ nN(k)
(−2)
−k
if k ≤ n and N(k)
a
n
is an oscillating sequence of negative powers of −2 until a sequence of 90
nines in π occurs, from then onwards the sequence is constant:
1, −
1
2
,
1
4
, −
1
8
,
1
16
, −
1
32
, . . . (−2)
−k
, (−2)
−k
, (−2)
−k
, (−2)
−k
, . . ..
The sequence determines a real number a, in the sense that it satisfies the
Cauchy condition. The sequence is well-defined, and N(n) can be checked
for each n (at least in principle).
For this particular a we cannot say that it is possitive, negative or zero.
a > 0 ⇔ N(k) holds for the first time for an even number
a < 0 ⇔ N(k) holds for the first time for an odd number
1
Brouwer considered various sequences, for one then the occurrence has been estab-
lished: 01234567890 does indeed occur among the decimals of π, cf. [Borwein 1998].
4
a = 0 ⇔ N(k) holds for no k.
Since we have no effective information on the status of the occurence of
N(k)’s, we cannot affirm the trichotomy law, ∀x ∈ R(x < 0∨x = 0∨x > 0),
cannot be said to have a proof.
The above number a cannot be irrational, for then N(k) would never
apply, and hence a = 0. Contradiction. Hence we have shown (a is ra-
tional). But there is no proof that a is rational. So A → A fails. One
also easily sees that a = 0 ∨ a ,= 0 fails to have proof.
Brouwer used his counter examples to show that some classical mathe-
matical theorems are not intuitionistically valid, cf.[Brouwer 1923].
The Brouwerian counterexamples are weak in the sense that they show
that some proposition has as yet no proof, but it is not excluded that eventu-
ally a proof may be found. In formal logic there is a similar distinction: ,¬ A
and ¬ A. The Brouwerian counter examples are similar to the first case,
strong counterexamples cannot always be expected. For example, although
there are instances of the Principle of the Excluded Third (PEM) where no
proof has been provided, the negation cannot be proved. For (A ∨ A)
is equivalent to A ∧ A, which is a contradiction! We will get to some
strong refutations of classical principles in the following sections.
Brouwer did not develop logic for its own sake, but he was the first to
establish a non-trivial result: A ↔ A ([Brouwer 1919b]).
Suggested further reading: [Heyting 1934], [Heyting 1956], [?], [Dummett 1977]
Ch. 1, 7, [Bridges-Richman 1987] Ch. 1, Philosophia Mathematica, vol 6 –
Special Issue: Perspectives on Intuitionism (ed. R. Tieszen), p. 129 – 226.
In 1928 the Dutch Mathematics Society posed in its traditional prize con-
test a problem asking for a formalization of Brouwer’s logic. Arend Heyting
sent in an essay, with the motto ‘stones for bread’, in which he provided a
formal system for intuitionistic predicate logic. The system was presented
in ‘Hilbert style’ (as it is called now). That is to say, a system with only
two derivation rules and a large number of axioms. Heyting had patiently
checked the system of Russell’s Principia Mathematica, and isolated a set of
intuitionistically acceptable axioms. His system was chosen in such a way
that the additon of PEM yields full classical logic. We give here a list of
axioms taken from [Troelstra–van Dalen 1988], p. 86.
A∧ B → A, A∧ B → B,
A → (B → (A∧ B),
5
A → A∨ B, B → A∨ B,
(A → (B → C)) → ((A∨ B → C)),
A → (B → A),
(A → (B → C) → ((A → B) → (A → C)),
⊥→ A,
A(t) → ∃xA(x),
∃x(A(x) → B) → (∃xA(x) → B),
∀xA(x) → A(t),
∀x(B → A(x)) → (B → ∀xA(x)).
There are two derivation rules: the well-known modus ponens (i.e. → E, see
below) and the rule of generalization: ¬
i
A(x) ⇒ ¬
i
∀xA(x), where ¬
i
stands
for intuitionistic derivability. When no confusion arises, we will simply write
¬. The ∀I-rule (see below) would have done just as well.
Heyting’s formalization was published in 1930, Glivenko (1929) and Kol-
mogorov (1925) had already published similar formalizations, which, how-
ever, did not cover full intuitionistic logic, cf. [Troelstra 1978].
In 1934 Gerhardt Gentzen introduced two new kinds of formalization
of logic. Both of these are eminently suited for the investigation of formal
deriations as objects in their own right.
In the system of Natural Deduction there are introduction and elimina-
tion rules for the logical connectives that reflect their meanings. The rules
are given below in an abbreviated notation.
6
∧I A B
A∧ B
∧E A∧ B
A
A∧ B
B
∨I A
A∨ B
B
A∨ B
[A] [B]
T T

∨E A∨ B C C
C
[A]
T
→ I B
A → B
→ E A A → B
B
⊥ E ⊥
A
T
∀I A(x)
∀xA(x)
∀E ∀xA(x)
A(t)
∃I A(t)
∃xA(x)
[A(x)]
T
∃E ∃xA(x) C
C
There are a few conventions for the formulation of the natural deduc-
tion system. (i) A hypothesis of a derivation between square brackets is
cancelled, that is to say, it no longer counts as a hypothesis. Usually all
of the hypotheses are cancelled simultaneously. This is not really required,
and it may even be be necessary to allow for ‘selective’ cancellation (cf.
[Troelstra–van Dalen 1988] p. 559, 568). For most practical purposes it is
however a convenient convention. (ii) For the quantifier rules there are some
natural conditions on the free variables of the rule. In ∀I the variable x may
not occur free in the hypotheses of T. In ∀E and ∃I the term t must be
free for x in A. Finally in ∃E the variable x may not occur in C or in the
remaining hypotheses of T. For an explanation see [Dalen 1997].
The rules are instructive for more than one reason. In the first place,
they illustrate the idea of the proof-interpretation. Recall that p : A → B
stands for “for every a : A (p(a) : B)”.
Now → E tells us that if we have a derivation T of A and a derivation
7
T

of A → B, then the combination
→ E
T T

A A → B
B
=
T

B
is a derivation of B. So we have an automatic procedure that, given T

,
converts T into a derivation T

of B.
The introduction rule also illustrates the transformation character of the
implication.
Let
A
T
B
be given, then → I yields
[A]
T
B
A → B
.
The first derivation shows that if we add a proof T

of A, we automatically
get a proof of B. So the rules tell us that there is a particular construction,
converting proofs of A into proofs of B. This is exactly the justification for
the derivation of A → B.
For, say, the conjunction rules the analogy is even more striking. We will
see below that the analogy can be made even more explicit in the Curry-
Howard isomorphism.
Martin-L¨of has designed in the seventies systems of type theory that
embodies many of the features of Gentzen’s natural deduction. He remarked
that the introduction and elimination rules determine the meaning of the
connectives ‘in use’. E.g. ∧I tells us that if we know (have evidence for) A
and B, then we also know A ∧ B. So ∧I specifies what one has to require
for A∧ B. The elimination rule tells us what we may claim on the grounds
of evidence for A∧ B. If we know A∧ B then we also know A and likewise
B.
The introduction and elimination rules have been chosen so that they
are ‘in harmony’. The evidence required in ∧I is exactly the evidence one
can derive in ∧E.
Michael Dummett used the same feature of Gentzen’s natural deduction
to support his claim that intuitionistic logic fits the requirement of ‘meaning
is use’, that is to say, the meaning of the logical operations can be concluded
from their use in introduction and elimination rules. Dummett insists on
demonstrative knowledge of mathematical notions, ‘the grasp of the mean-
ing of a mathematical statement must, in general, consist of a capacity to
use that statement in a certain way, or to respond in a certain way to its use
by others’([Dummett 1975]). As a consequence the traditonal, Platonistic,
8
notion of truth has to be replaced by something more palpable; the notion
of proof is exactly what will fill the need for communicability and observ-
ability. Hence the slogan “a grasp of the meaning of a statement consists in
a capacity to recognize a proof of it when one is presented to us”.
This, of course, is in complete accord with intuitionistic practice. The
rejection of the Platonistic notion of truth is indeed an aspect of Dummett’s
anti-realism. Brouwer had always denied the realistic thesis, that there is
an outer world independent of us.
The second derivation system of Gentzen is his sequent calculus. It differs
from the natural deduction system in various ways. The most striking is
the extra memory the derivation rules carry along. In natural deduction
the rules are (more or less) local, the rules handle the formulas involved in
the inference, but the assumptions (hypotheses) largely remain unnoticed
(of course, in rules with discharge, some hypotheses that play a role in
the derivation step play a role). Furthermore the introduction-elimination
feature is replaced by left-right technique.
We give an example:
Γ, A ¬
i
´
Γ, A∧ B ¬
i
´
Γ
1
¬
i
´
1
, A Γ
2
¬
i
´
2
, B
Γ
1
∪ Γ
2
¬
i
´
1
∪ ´
2
A∧ B
Γ ¬
i
´ should be thought of as ‘the conjunction of all propositions in Γ
yields the disjunction of the proposition in ´’ (where Γ and ´ are finite sets
of propositions). The ‘right’-rule tells us what we have to know in order to
conclude A∧B, the ‘left-rule tells us that in order to draw a conclusion from
A∧ B, it suffices to draw the conclusion from A (or B, for that matter).
For a treatment of the sequent calculus we refer the reader to [Kleene 1952],
[Girard 1987], [Girard 1989], [Troelstra–van Dalen 1988].
Suggested further reading: [?], [Dalen 1997] Ch. 5, [Dragalin 1988] Part 1,
[Dummett 1977] Ch. 4, [Kleene 1952] Part II, [Negri–von Plato], [Troelstra-Schwichtenberg 1996]
Ch. 1,2,3, [Troelstra–van Dalen 1988] Ch. 2.
Formal logic has two faces, the syntactic, combinatorial part, and the
denotational part. The syntactic part may be viewed as a pure game with
symbols, according to given rules. It goes without saying that this position
is usually softened by an informal meaning concept. After all, one has to
9
know why to play this game and not that. Proof theory, traditionally, is
placed in the combinatorial part of logic.
The study of interpretations is called semantics, from this viewpoint
formulas denote certain things. Frege already pointed out that propositions
denote truth values, namely ‘true’ and ‘false’, conveniently denoted by 1 and
0.
The interpretations of logic under this two-valued semantics is handled
by the well-known truth tables (cf. [Dalen 1997]).
The method has a serious drawback: all propositions are supposed to be
true or false, and PEM automatically holds (some might perhaps see this as
an advantage rather than a drawback). It certainly is too much of a good
thing for intuitionistic logic. By our choice of axioms (or rules) intuitionistic
logic is a subsystem of classical logic, so the two-valued semantics obliterates
the distinction between the two logics: too many propositions become true!
One might say that we should not worry about the problem of semantics,
since we already have the intended proof interpretation. This certainly is
the case, however, the proof-interpretation is too little specific to yield sharp
decisive results, as one would like in model theory. One would need more
assumptions about ‘constructions’, before technical problems can be settled.
All of the formal semantics discussed below are strongly complete for
intuitionistic predicate and propositional logic, in the sense that Γ ¬
i
A ⇔
Γ [= A, where [= is the semantical consequence relation in the particular
semantics.
1.1 The topological interpretation
In the mid-thirties a number of systematic semantics were introduced that
promised to do for intuitionistic logic what the ordinary truth tables did for
classical logic.
Heyting had already introduced many valued truth tables in his formal-
ization papers, e.g. to establish non-definability of the connectives. Jaskowski
presented in 1936 a truth table family that characterized intuitionistic propo-
sitional logic, [Jaskowski 1936]. G¨odel had dispelled the expectation that
intuitionistic logic was the logic of some specific finite truthvalue system,
[G¨odel 1932].
An elegant interpretation was introduced by Tarski, [Tarski 1938]. It
was, in fact, a generalization of the Boolean valued interpretation of classical
logic. Ever since Boole it was known that the laws of logic correspond exactly
to those of Boolean algebra (think of the powerset of a given set with ∩, ∪
10
and
c
as operations, corresponding to ∧, ∨ and ). Now we want a similar
algebra with the property that U
cc
,= U(for A and A are not equivalent).
By Brouwer’s theorem (A ↔ A), we expect U
ccc
= U
c
. The remaining
laws of logic demand that (U ∩V )
cc
= U
cc
∩V
cc
and (U ∪V )
cc
⊆ U
cc
∪V
cc
.
This suggests that the operator
cc
behaves as a closure operator in topology.
It turns out that it is a good choice to let the open sets in a topological
space X play the role of arbitrary sets in the power set of X. So the family
O(X) plays the role of T(X).
Here is the notation: [[A]] is the open set of X assigned to A. The valuation
[[]] : PROP → O(X) is defined inductively for all propositions; let [[A]] be
given for all atomic A, where [[⊥]] = ∅, then
[[A∧ B]] = [[A]] ∩ [[B]]
[[A∨ B]] = [[A]] ∪ [[B]]
[[A → B]] = Int([[A]]
c
∪ [[B]])
[[A]] = Int([[A]]
c
)
Here Int(K) is the interior of the set K (i.e. the largest open subset of
K). Note that this looks very much like the traditional Venn diagrams,
with the extra requirement that negation is interpreted by the interior of
the complement. This is necessary if we want to get open sets all the way.
A simple calculation shows that ¬
i
A ⇒ [[A]] = X (we use ¬
i
, resp. ¬
c
, for
derivability in intuitionistic, resp. classical, logic). Since X is the largest
open set in O(X), it is plausible to call A true in O(X) if [[A]] = X.
This interpretation can be used to show underivability of propositions.
Consider, for example, O(R), the open sets of real numbers. Assign to the
atom A the set R−¦0¦. Then [[A]] = R−¦0¦, [[A]] = Int(¦0¦) = ∅, [[A]] =
R. So [[A → A]] = Int[[A]]
c
∪ A = ∅ ∪ A = A ,= R.
Hence ,¬
i
A → A. Similarly ,¬
i
A∨ A, for [[A∨ A]] = A ,= R.
The topological interpretation can be extended to predicate logic. Let a
domain D be given, then
[[∃xA(x)]] =

¦[[A(d)]][d ∈ D¦
[[∀xA(x)]] = Int(

¦[[A(d)]][d ∈ D¦)
The topological interpretation is indeed complete for intuitionistic logic.
Let us say that A is true in O(X) if [[A]] = X for all assignments of open
sets to atoms. A is true if A is true under all topological interpretations.
Completeness can now be formulated as usual: ¬
i
A ↔ A is true. The
theory of topological interpretations is treated in [Rasiowa-Sikorski 1963].
Classical logic appears as a special case when we provide a set X with the
trivial topology: O(X) = T(X).
11
The algebra of open subsets of a topological space is a special case of a
Heyting algebra. Heyting algebra’s are defined, much like Boolean algebra’s,
by axioms for the various operations.
A Heyting algebra has binary operations ∧, ∨, →, a unary operation
and two constants, 0, 1.
The laws are listed below:
a ∧ b = b ∧ a
a ∨ b = b ∨ a
a ∧ (b ∧ c) = (a ∧ b) ∧ c
a ∨ (b ∨ c) = (a ∨ b) ∨ c
a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c)
a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c)
1 ∧ a = a
1 ∨ a = 1
0 ∧ a = 0
0 ∨ a = a
a → a = 1
a ∧ (a → b) = a ∧ b
b ∧ (a → b) = b
a → (b ∧ c) = (a → b) ∧ (a → c)
a = a → 0
Since it is necessary for the interpretation of predicate logic to allow
infima and supprema of collections of elements, we often consider complete
Heyting algebra’s. That is, algebra’s with the property that for a collection
¦a
i
[i ∈ I¦ of elements there is a supprenum

i∈I
a
i
(the unique least element
majorizing all a
i
), and an infimum

i∈I
a
i
For a complete Heyting algebra (i.e. an algebra in which all sups and infs
exist), the laws can be simplified somewhat, we adopt the standard axioms
for a latice and add
a ∧

S =

¦a ∧ s[s ∈ S¦
1.2 Beth–Kripke semantics
Elegant as the topological interpretation may be, it is not as flexibled as two
later interpretations introduced by E.W. Beth and Saul Kripke. Both se-
mantics have excellent heuristics. We will consider here a hybrid semantics,
Beth–Kripke models, introduced in [Dalen 1984].
The basic idea is to mimick the mental activity of Brouwer’s individual,
who creates all of mathematics by himself. This idealized mathematician,
also called creating subject by Brouwer, is involved in the construction of
mathematical objects, and in the construction of proofs of statements. This
process takes place in time. So at each moment he may create new elements,
and at the same time he observes the basic facts that hold for his universe
so far. In passing from one moment in time to the next, he is free how
12
to continue his activity, so the picture of his possible activity looks like a
partially ordered set (even like a tree). At each moment there is a number
of possible next stages. These stages have become known as possible worlds.
Let us just consider, for the moment, the first-order case, that is to say
we consider elements of one and the same sort, furthermore there is a finite
number of relations and functions (as in a standard first-order language).
The stages for the individual form a partially ordered set ¸K, ≤). We view
k ≤ as “k is before or coincides with . For each k ∈ K there is a local
domain of elements created so far, denoted by D(k). It is reasonable to
assume that no elements are destroyed later, so k ≤ ⇒ D(k) ⊆ D().
A path in the poset K is a maximal ordered subset. For a node k in the
poset K a bar B is a subset with the property that very path through k
intersects B. Now we will stipulate how the individual arrives at the atomic
facts. He does not necessarily establish an atomic fact A ‘at the spot’, but
he will, no matter how he pursues his research, establish A eventually. This
means that there is a bar B for k such that for all nodes in B the statement
A holds at . Note that the individual does not (have to) observe composite
states of affairs.
The next step is to interprete the connectives. Let us write “k ¬ A”
for “A holds at k”. The technical terminology is “k forces A”. For atomic
A k ¬ A is already given, ⊥ is never forced.
k ¬ A∧ B k ¬ A and k ¬ B
k ¬ A∨ B ∃B∀ ∈ B ¬ A or ¬ B
k ¬ A → B ∀ ≥ k( ¬ A ⇒ ¬ B)
k ¬ A k ¬ A →⊥
∀ ≥ k( ¬ A ⇒ ¬⊥
∀ ≥ k( ,¬ A)
k ¬ ∃xA(x) ∃B∀ ∈ B∃a ∈ D() ¬ A(a)
k ¬ ∀xA(x) ∀ ≥ k∀a ∈ D() ¬ A(a)
Observe that the ‘truth’ at a node k essentially depends on the future.
This is an important feature in intuitionism (and in constructive mathemat-
ics, in general). The dynamic character of the universe demands that the
future is taken into account. This is particularly clear for ∀. If we claim that
“all dogs are friendly”, then one unfriendly dog in the future may destroy
the claim.
13
A Beth–Kripke model with the property that the bar Bin all the defining
clauses is precisely the node k itself, is called a Kripke model. And if all the
local domains D(k) are identical, we have a Beth model.
An individual model over a poset K is denoted by /.
We say that “A is true in a Beth–Kripke model /” if for all k ∈ K k ¬ A.
“A is true” if A is true in all Beth–Kripke models. ‘Semantical consequence’
is defined as Γ ¬ A iff for all Beth–Kripke models / and all k ∈ K k ¬ C
for all C ∈ Γ → k ¬ A
Beth–Kripke- , Kripke- and Beth- semantics are strongly complete for
intuitionistic logic: Γ ¬
i
A ⇔ Γ ¬ A (special case: ¬
i
A ⇔ A is true).
There are a number of simple propertiwes that can easily be shown, e.g.:
forcing is monotone, i.e. k ¬ A, k ≤ ⇒ ¬ A,
in Beth–Kripke and Beth models k ¬ A ⇔ ∃B∀ ∈ B ¬ A.
In practice, Kripke models are the more flexible tools for metamathe-
matical purposes. A large number of applications have been given since
their introduction. We will discuss them first.
Since Kripke models are given by partially ordered sets, we can indicate
them by means of diagrams. By labelling the set with propositions, it is
easy to see what the logical properties are. We give some examples.
Note that k ¬ A if ∀ ≥ k( ,¬ A), i.e. A is not forced after k. k ,¬ A
iff for some ≥ k ¬ A
So k ¬ A if for each ≥ k there is an m ≥ with m ¬ A.
(a) Consider an atomic A and let k
1
> k
0
, k
1
¬ A, but k
0
,¬ A. By the
above remark, we see that k
0
¬ A. Hence k
0
,¬ A → A.
Furthermore k
0
,¬ A, so k
0
,¬ A∨ A.
(b) k
1
> k
0
, k
2
> k
0
, k
1
¬ A and k
2
¬ B. We see that k
0
,¬ A, k
0
,¬ B, so
k
0
,¬ A∨ B. k
1
,¬ A∧ B, k
2
,¬ A∧ B, so k
0
¬ (A∧ B).
Hence k
0
,¬ (A∧ B) → (A∨ B) (De Morgan’s law fails).
(c) k
1
> k
0
, k
0
,¬ A(0), k
1
,¬ A(0), k
1
¬ A(1). Clearly k
1
,¬ ∀xA(x), so
k
0
¬ ∀xA(x). If k
0
¬ ∃xA(x) then k
0
¬ A(1). But this contradicts
k
1
¬ A(1). Hence k
0
,¬ ∃xA(x). So k
0
,¬ ∀xA(x) → ∃xA(x).
There is an extensive modeltheory for Kripke semantics. In particular
there are a large number of results on the structure of the partially ordered
sets. For example, ipredicate logic is complete for Kripke models over trees,
14
,
,
k
0
k
1
A
(a)
,
, ,
¡
¡
¡
¡
¡
e
e
e
e
e
k
0
k
1
k
2
A B
(b)
`

`

,
, ,
0
0 1 k
1
A(1)
k
0 (c)
and for propositional logic this can even be strengthened to finite trees (the
so-called finite model property: ,¬ A ⇒ A is false in a Kripke model over a
finite tree).
The completeness over tree models can be used to prove the disjunction
property
(DP) ¬
i
A∨ B ⇒¬
i
A or ¬
i
B
The proof uses reductio ad absurdum. Suppose ,¬
i
A and ,¬
i
B, then there
is a tree model /
1
where A is not forced, similarly a tree model /
2
where
B is not forced. The two models are glued together as follows: put the two
models side by side and place a new node k below both. In this new node no
proposition is forced. The result is a correct Kripke model. Since ¬
i
A∨B,
we have k ¬ A ∨ B, and hence k ¬ A or k ¬ B. But that contradicts the
fact that A and B are not forced in /
1
and /
2
. Therefore ¬
i
A or ¬
i
B.
There is a corresponding theorem for existential sentences, the existence
property
(EP) ¬
i
∃xA(x) ⇒¬
i
A(t) for a closed term t
These theorems lend a pleasing support to the intuitionistic intended mean-
ing of ‘existence’: if you have established a disjunction, you know that you
can established one of the disjuncts. Similarly for existence: if you have
shown the existence of something, you can indeed point to a specific in-
stance.
The disjunction and existence property hold for a number of promi-
nent theories, the most important being arithmetic. DP and EP are often
considered the hallmark of constructive logic; one should, however, not over-
estimate the significance of a technical result like this.
There is a snag in the conclusions drawn above. As the reader has seen,
the proof made use of reductio ad absurdum, i.e. the result has not been
established constructively. What one would like is a method that extracts
from a proof of ∃xA(x) a proof of A(d) for some d. Fortunately there are
proof theoretical devices that provide exactly this kind of information, see
15
28. Smorynski has shown that in a fairly large number of cases ‘semantic’
proofs can be made constructive, [Smorynski 1982].
For more information on Kripke model theory see [Dalen 1997], [Fitting 1969],
[Troelstra–van Dalen 1988].
Finite Beth models are not interesting for intuitionistic logic. For in each leaf
(maximal node) A or A holds, so in each node A∨A holds, i.e. the logic
is classical. Beth’s original models are slightly more special, he considered
constant domains, i.e. D
k
= D

for all k, . This is a certain drawback when
compared to Kripke models. The combinatorics of Beth models is just more
complicated than that of Kripke models.
Kripke showed in 1963 that Kripke and Beth models can be converted
into each other [Kripke 1965], see also [Sch¨ utte 1968].
Beth models are easily convertible into topological models. For con-
venience we consider Beth models over tree. On a tree one can define a
topology as follows: open sets are those subsets U of the tree that are closed
under successors, i.e. k ∈ U and k ≤ ⇒ ∈ U. It is easily seen that these
sets form a topology. One also sees that ¦k[k ¬ A¦ is an open set, let us
therefore put [[A]] = ¦k[k ¬ A¦. Now it is a matter of routine to show that
the function [[.]] is a valuation as defined in section 2.1.
Beth semantics also turn out to be a convenient tool in completeness
proofs, see [Troelstra–van Dalen 1988], Ch. 13, [Dummett 1977]. In partic-
ular, it is useful for rendering completeness proofs in an intuitionistic meta-
mathematics, Veldman was in 1976 the first to consider a modified Kripke
semantics, for which he could give an intuitionistically correct completeness
proof, [Veldman 1976]. Since then De Swart, Friedman, Dummett, Troelstra
have given alternative versions for Beth semantics.
Beth models happen to be better adapted to second-order arithmentic
with function variables, the so-called intuitionistic analysis. They allow for
a very natural interpretations of choice sequences. The use of Beth models
for second-order systems can be found in [Dalen 1978], [Dalen 1984].
1.3 Super Semantics.
We have seen a number of interpretations of intuitionistic theories, but by no
means all of them. There is, for example, a totally different interpretation of
intuitionistic logic (and arithmetic) in Kleene’s realizability interpretation.
This is based on algorithms, and on the face of it, one could not find a
similarity with the above semantics. One might wonder if these semantics
are totally unrelated, or whether there is some common ground. The obvious
16
common ground is the logic that they are modelling, but that would not be
sufficient to link, say Kripke models and realizability.
Fortunately there is a general kind of semantics based on category theory.
The pioneer in this area was Lawvere, who saw the possibilities for treating
logical notions in a categorical setting. He, in particular, discovered the
significance of adjointness for logic.
Historically these newer interpretations grew out of existing semantics,
e.g. Dana Scott showed in his extension of the topological interpretation to
second-order systems, how one could capture strikingly intuitionistic fea-
tures, [Scott 1968, Scott 1970]. Subsequently Joan Moschovakis and Van
Dalen used the same technique for interpreting the theory of choice sequences
over natural numbers, and second order arithmetic, HAS, [Moschovakis 1973],
[Dalen 1974].
The proper formulation of this kind of semantics is given in the so-called
sheaf-semantics.
In order to avoid complications we will look at Scott’s original model,
adapted to the sheaf semantics.
Consider the topology of the real line, R. The open sets will be our truth
values. We will interprete the intuitionistic theory of reals. The objects are
partial continuous real-valued functions with an open domain. This domain
‘measures’ the existence of the object: Ea ∈ O(R).
DRAWING
The interpretation forces us to consider the notion of partial object and
the notions of ‘equivalence’ and ‘strict equality’. All objects are partial, and
we say that a is total if Ea = R. Equality has to be reconsidered in this
light: [[a = b]] = Int¦t ∈ R[a(t) = b(t)¦
DRAWING
Observe that [[a = a]] = Ea, and in general [[a = b]] ⊆ E(a) ∩ E(b).
In addition to the notion of equality there is a convenient notion of
equivalence: “a and b coincide where they exist”, in symbols a · b :=
Ea ∨ Eb → a = b. According to the above definition we have [[a · a]] = R.
The operation on the partial elements are defined pointwise:
(a +b)(t) = a(t) +b(t).
(a b)(t) = a(t) b(t)
The definition of the inverse is, however, problematic. One can define
it pointwise for values distinct from 0, but then there seem to be problems
with the invertibility of non-zero elements.
17
DRAWING
The fact is that ‘non-zero’ is not good enough to have an inverse. Let
a(t) = t, then [[a = 0]] = Int¦t[a(t) = t = 0¦ = Int¦0¦ = ∅.
So [[a ,= 0]] = [[a = 0]] = Int[[a = 0]]
c
= R
This a is non-zero in the model. There is however no b such that [[a.b =
1]] = R. This is where apartness comes in. It is well-known that a real
number has an inverse iff it is apart from 0. So let us interprete # in the
model:
[[a#b]] = ¦t[a(t) ,= b(t)¦.
The condition for an inverse now becomes
a#0 → ∃b(a b = 1).
In the model we now get an inverse for a. Take b(t) = t
−1
for t ,= 0, then
[[a#0]] = R −¦0¦ =

c[[a c = 1]] ⊆ [[a b = 1]] = R −¦0¦.
The introduction of an existence predicate, and partial equality, of course,
carries obligations for our quantifiers. In particular
∃xA(x) ↔ ∃x(Ex ∧ A(x)) and
∀xA(x) ↔ ∀x(Ex → A(x)).
The logical axioms, or rules, those have to be revised. It suffices to
consider the quantifier rules and the equality rules:
[Ex]
T
A(x) ∀xA(x) Et
∀xA(x) A(t)
[A(x)][Ex]
T
A(t)Et ∃xA(x) C
∃xA(x) C
The axiomatization of the equality aspects is simple if we use the equiv-
alence relation instead of identity itself. We add the axioms:
x · y ∧ A(x) → A(y)
∀z(x · z ↔ y · z) → x · y.
That is, we take ‘·

as a primitive, ‘=’ is then regained by defining t = s :=
Et ∧ Es ∧ t · s.
For the details of the logic can be found in [Troelstra–van Dalen 1988]
and Scott’s original [Scott 1979].
18
One may add also the ordering to the model of R:
[[a < b]] := ¦t[a(t) < b(t)¦.
Scott extended his model to a higher-order theory of reals, in which he
could show that Brouwer’s continuity theorem holds, see [Scott 1970].
Joan Moschovakis applied the methods of Scott’s semantics to intuition-
istic analysis in the style of Kleene’s FIM, [Moschovakis 1973]. Her objects
were continuous mappings from Baire space to Baire space (a model with
total elements). Van Dalen’s model for analysis is based on Beth’s original
semantics. It can be translated straightforwardly into a topological model,
but its formulations in a Beth model allows for some extra fine tuning so
that techniques from set theory can be applied. E.g. a model for lawless
sequences is constructed by means of forcing. The model also interprets
the theory of the creating subject, [Dalen 1978]. Grayson used the same
approach in his version of the permutation model of Krol, [Grayson 1981].
Sheaf models were further presented by Scott, Fourman, Hyland and
Grayson, [Fourman-Hyland 1979], [Fourman-Scott 1979], [Grayson 1979, Grayson 1981,
Grayson 1982, Grayson 1984].
Sheaf models allow an interpretation of higher-order logic, they are nat-
ural examples of a topos. Moerdijk and Van der Hoeven used a special sheaf
model to interprete the theory of lawless sequences, [Hoeven-Moerdijk 1984].
Since Kripke models and Beth models are built over trees, they carry a nat-
ural topology; as a result they can be viewed as sheaf models.
The next generalisation after sheaf models is that of categorical model.
Certain categories are powerful enough to interprete higher-order intuition-
istic logic. In that sense topos models can be viewed as intuitionistic uni-
verses. Categorical models also turned out to be important for typed lambda
calculus. The reader is referred to the literature for details
It should be mentioned that the topos semantics managed to capture
most of the known semantics. E.g. Hyland showed that the realizability in-
terpretation fitted into his effective topos, [Hyland 1982], [Troelstra–van Dalen 1988],
Ch. 13 ¸8.
Suggested further reading: [Dragalin 1988] Part3, [Dummett 1977] Ch. 5,
[Fitting 1969], [Fourman-Scott 1979], [Friedman 1977], [Smorynski 1973a],
[Gabbay 1982], [Lambek–Scott 1986], [MacLane–Moerdijk 1992], [McLarty 1995],
[Rasiowa-Sikorski 1963], [Troelstra–van Dalen 1988] Ch. 2, ¸5, 6, Ch. 13,13,15.
There is no special foundational bias towards first-order theories, but it is
an observed fact that large parts of mathematics lend themselves to formu-
19
lation in a first-order language. We will look in this section at a number of
first-order theories and point out some salient properties.
Predicate logic itself is, just like its classical counterpart, undecidable.
Fragments are, however, decidable. E.g. the class of prenex formulae is
decidable (and as a corollary, not every formula has a prenex normal form
in IQC), cf. [Kreisel 1958], on the other hand, monadic predicate logic is
decidable in classical, but undecidable in intuitionistic logic, [Kripke 1968],
[Maslov 1965] Lifschitz showed that the theory of equality is undecidable,
[Lifschitz 1967]. An extensive treatment of first-order theories is given in
papers of Gabbay and the dissertation of Smorynski, see [Gabbay 1982],
[Smorynski 1973b], [Smorynski 1973a].
In propositional logic one may impose restrictions on the underlying
trees of Kripke models, in predicate logic one may also put restrictions on
the domains of the models. A well-known instance is the constant domain
theory of S. G¨orneman (Koppelberg). She showed that predicate logic plus
the axiom ∀x(A(x) ∨B) → ∀xA(x) ∨B is complete for Kripke models with
constante domains, [Goernemann 1971].
In general Kripke semantics is a powerful tool for meta-mathematical
purposes. Many examples can be found in the cited works of Smorynski and
Gabbay. We have demonstrated the usefulness in section 3. in the case of
the disjunction property. The key method in the proof was the joining of
a number of disjoint Kripke models by placing one extra node (the root of
the tree) below the models. This technique has become known as gluing.
Following Smorynski we will apply it to a few theories.
First we point out that, although a number of intuitionistic first-order
theories share their axioms with their classical counterparts, in general the-
ories are sensitive to the formulation of the axioms, notably in the absence
of PEM. We consider a few basic theories below.
The theory of equality has the usual axioms – reflexivity, symmetry,
transitivity.
One can strengthen the theory in many ways, for example, the theory of
stable equality is given by
EQ
st
= EQ+∀xy(x = y → x = y)
And the decidable theory of equality is axiomatized by
EQ
dec
= EQ+∀xy(x = y ∨x ,= y), where x ,= y abbreviates x = y.
In intuitionistic mathematics there is a strong notion of inequality: apart-
ness. Introduced by Brouwer in [Brouwer 1919a] and axiomatized by
Heyting in [Heyting 1925]. The symbol for apartness is #.
20
The axioms of AP are given by EQ and the following list









∀xyx

y

(x#y ∧ x = x

∧ y = y

→ x

#y

)
∀xy(x#y → y#x)
∀xy(x#y ↔ x = y)
∀xyz(x#y → x#z ∨ y#z)
We will now use the gluing technique to show that AP has the disjunc-
tion and existence property.
Let AP ¬ A ∨ B and assume AP ,¬ A and AP ,¬ B. Then, by the
strong completeness theorem, there are models /
1
and /
2
of AP such that
/
1
,¬ A and /
2
,¬ B


k
0
K
1
d
d
d
d
d
d
d

We consider the disjoint union of /
1
and /
2
and place the one-point
world below it. That is to say, we designate points a and b in /
1
and /
2
which are identified with the point 0. The new model obviously satisfies
the axiom of AP. Hence k
0
¬ A ∨ B and so k
0
¬ A or k
0
¬ B. Both are
impossible on the grounds of the choice of /
1
and /
2
. Contradition. Hence
AP ¬ A or AP ¬ B.
For EP it is convenient to assume that the theory has a number of
constants, say ¦c
i
[i ∈ I¦.
Now let AP ¬ ∃xA(x) and AP ,¬ A(c
i
) for all i. Then for each i there
is a model /
i
with /
i
,¬ A(c
i
).
As we did above, we glue the models /
i
by means of a bottom world /

with a domain consisting of just the elements c
i
. No non-trivial atoms are
forced in k
0
(i.e. only the trivial identities c
i
= c
i
). The identification of the
c
j
with elements in the models /
i
is obvious. Again, it is easy to check that
21
the new model satisfies AP. Hence k
0
¬ ∃xA(x), i.e. k
0
¬ A(c
i
) for some i.
But then also /
i
¬ A(c
i
), contradiction. Hence AP ¬ A(c
i
) for some i.
The gluing operation demonstrates that there are interesting operations
in Kripke model theory that make no sense in traditional model theory.
The apartness axioms have consequences for the equality relations. In
particular we get a stable equality: x = y → x = y. For, x#y ↔ x = y,
so x = y ↔ x#y ↔ x = y.
In fact the equality fragment is axiomatized by an infinite set of quasi-
stability axioms.
Put x ,=
0
y := x = y
x ,=
n+1
y := ∀z(z ,=
n
x ∨ z ,=
x
y)
For these “approximations to apartness” we formulate quasi-stability
axioms: Sn := ∀xy(x ,=
n
y → x = y). The S
n
axiomatize the equality
fragment of AP. To be precise: AP is conservative over EQ+¦S
n
[n ≥ 0¦.
The result was first established by proof theoretic means in [Dalen 1978],
and subsequently an elegant model theoretic proof was given in [Smorynski 1973c].
This shows that even a relatively simple theory like equality is incomparably
richer than the classical theory.
Apartness and linear order are closely connected. The theory LO of
linear order has axioms:
∀xyz(x < y ∧ y < z → x < z)
∀xyz(x < y → z < y ∨ x < z)
∀xyz(x = y ↔ x < y ∧ y < z)
The second axiom is interesting because it tells, so to speak, that a < b
means that a is “far” to the left of b, in the sense that if we pick an arbi-
trary third point, it has to be to right of a or to the left of b. The relation
with apartness is given by LO+AP ¬ x < y ∨ y < x ↔ x#y
One can also use x < y ∨ y < x to define apartness. In a way this gives the
best possible result since LO+APis conservative over LO([Smorynski 1973c],
[Dalen 1997]).
It is important to note that the atoms, or in general the quantifier free
part of a theory does not yet determine whether it is classical or not. There
are cases where a decidable equality results in the fact that the theory is
classical, e.g. the theory of algebraically closed fields; on the other hand,
for arithmetic we can prove ∀xy(x = y ∨ x ,= y), but the theory is not
classical. There are quite a number of surprising results on first-order the-
ories, cf. [Gabbay 1972], [Gabbay 1973], [Gabbay 1982], [Smorynski 1973a],
22
[Smorynski 1978].
The best known first-order theory is, of course, arithmetic. Intuitionistic
arithmetic, HA, called after Heyting, [Heyting 1930b, Heyting 1930a], is ax-
iomatized by exactly the same axioms as Peano’s arithmetic. The difference
is the underlying logic: PA = HA+PEM.
The gluing technique also works for intuitionistic arithmetic, HA. In
this case one has to do some extra work to check the induction axiom. The
result is that for HA we have the existence property for numerals:
HA ¬ ∃xA(x) ⇒HA ¬ A(n) for some n.
The disjunction property is an obvious consequence of the existence prop-
erty, as ∨ is definable in terms of ∃:
HA ¬ (A∨ B) ↔ ∃x((x = 0 → A) ∧ (x ,= 0 → B))
It would seem rather unlikely that the existence property is a conse-
quence of the disjunction property; Harvey Friedman proved however, to
the surprise of the insiders, that in the case of HA (and a number of related
systems) this is indeed the case, [Friedman 1975].
Intuitionistic arithmetic is, of course, even more incomplete than classical
arithmetic because it is a subsystem of PA. In fact PA is an unbounded
extension of HA (Smorynski).
There are a number of interesting extensions of intuitionistic arithmetic,
one may add principles that have a certain constructive motivation. One of
such principles is, Markov’s Principle:
MP ∀x(A(x) ∨ A(x)) ∧ ∃xA(x) → ∃xA(x)
The principle is a generalization of the original formulation of Markov,
[Markov 1971]; Markov considered the halting of a Turing machine, see also
[Dragalin 1988]. Such a machine is an abstract computing device that oper-
ates on a potentyially infinite tape. The key question for Turing machines
is, does the machine, when presented with an input on the tape, eventu-
ally halt (and hence produce an output)? Suppose now that someone tells
us that it is impossible that the machine never halts, do we know that it
indeed halts? Markov argued ‘yes’. The decision procedure for the halting
in this case consists of turning on the machine and waiting for it to halt.
An intuitionist would not buy the argument. When somebody claims that
a Turing machine will stop, the natural question is ‘when?’ We want an
actual bound on the computation time. Reading ‘the machine halts at time
x’ for A(x), the above formulation exactly covers Markov’s argument. In
fact, in Markov’s case A(x) is primitive recursive.
23
A simple Kripke model shows that HA ,¬ MP. Consider a model with
two nodes k
0
, k
1
, where k
0
< k. In the bottom node we put the standard
model of natural numbers, in the top node we put a (classical) non-standard
model / in which the negation of G¨odel’s sentence “I am not provable in
PA”, i.e. the proper sentence of the form ∃xA(x), is true.
So k
1
¬ ∃xA(x) and hence k
0
¬ ∃xA(x). But k
0
¬ ∃xA(x) would
ask for an instance A(n) to be true in the standard model, and hence would
yield a conflict with the independence of the G¨odel sentence fromPA. Since,
clearly, k
0
¬ ∀x(A(x) ∨ A(x)), the model refutes MP.
Markov’s principle may be unprovable in HA, but its companion, Markov’s
rule, has a stronger position.
Given a statement of the form A → B, one may formulate a correspond-
ing rule, Γ ¬ A ⇒ Γ ¬ B, which is in general weaker than Γ ¬ A → B.
For example: Markov’s rule MR with respect to HA says:
HA ¬ ∀x(A(x) ∨ A(x) ∧ ∃xA(x) ⇒HA ¬ ∃A(x)
We say that HA is closed under Markov’s rule. The heuristic argument
is that we have more information in this case, we actually have a proof of
∃xA(x), from this extra evidence we may hope to draw a stronger conclu-
sion. Markov’s rule is indeed correct, cf. [Troelstra–van Dalen 1988], p.129,
p. 507, [Beeson 1976], [Beeson 1985], p. 397. Note that A(x) may contain
more free variables. For closed ∃xA(x), the proof of closure under Markov’s
rule is particularly simple: if HA ¬ ∃xA(x), then PA ¬ ∃xA(x), and
hence A(n) is true in the standard model for some n. Now HA ¬ A(n) ∨
A(n) and by DP HA ¬ A(n) or HA ¬ A(n). The letter is impossible,
hence HA ¬ A(n), and a fortiori HA ¬ ∃xA(x) ([Smorynski 1973a], p. 366.
Since the theory of natural numbers is at the very heart of mathematics,
it is no surprise that a great deal of research has been devoted to the subject.
In the early days of intuitionism people wondered to what extent HA was
safer that PA. This was settled in 1933 by G¨odel (and independently by
Gentzen), who showed that PA can be translated into HA. G¨odel defined
a translation, which from the intuitionistic viewpoint weakened statements.
This was basically done by a judicious distribution of negations. Here is the
formal definition:
24
A

= A for atomic A
(A∧ B)

= A

∧ B

(A∨ B)

= (A

∧ B

)
(A → B)

= A

→ B

(∀xA(x))

= ∀xA

(x)
(∃xA(x))

= ∀xA

(x)
For this G¨odel translation we get PA ¬
c
A ⇔HA ¬
i
A

.
Hence PA ¬
c
0 = 1 ⇔ HA ¬
i
0 = 1 ⇔ HA ¬
i
0 = 1
So PA is consistent if and only if HA is so. In other words, no deep
philosophical insight can be expected here.
It is an easy consequence of the G¨odel translation theorem that the
universal fragment of PA is conservative over HA.
The G¨odel translation, of course, also works for predicate logic:
Γ ¬
c
A ⇔ Γ

¬
i
A

,
where Γ

= ¦B

[B ∈ Γ¦ (cf. [Dalen 1997], p. 164). Kolmogorov had already
in 1925 proposed a similar translation (unknown to G¨odel in 1933), he added
a double negation in front of every subformula (cf. [Troelstra–van Dalen 1988],
p. 59). Kolmogorov had in fact axiomatized a fragment of intuitionistic pred-
icate logic, but the translation and the proof of the corresponding theorem
immediately extends to full predicate logic.
The result of the G¨odel translation may be improved for certain simple
formulas. Kreisel showed that PA and HA prove the same Π
0
2
sentences,
PA ¬
c
∀x∃yA(x, y) ⇔HA ¬
i
∀x∃yA(x, y), where A(x, y) is quantifier free.
The proof has some quite interesting features. First we note that a
quantifier free formula A(x, y) is equivalent in HAto an equation t(x, y) = 0
(where we assume that HAhas defining equations for the primitive recursive
functions).
Next we introduce the Friedman translation: for a given formula F we
obtain A
F
from A by replacing all atoms P by P ∨ F.
It is a routine exercise to show that
Γ ¬
i
A ⇒ Γ
F
¬
i
A
F
A ¬
i
A
F
HA ¬
i
A ⇒HA ¬
i
A
F
.
Now consider a term t and let HA ¬ ∃x(t(x, y) = 0). We apply the
Friedman translation with respect to F := ∃x(t(x, y) = 0).
25
((∃x(t(x, y) = 0) →⊥) →⊥))
∃x(t(x,y)=0
= ((∃x(t(x, y) = 0) ∨∃x(t(x, y) =
0)) → (⊥ ∨∃x(t(x, y) = 0)) → (⊥ ∨∃x(t(x, n) = 0))).
This formula is equivalent to ∃x(t(x, y) = 0). Hence HA ¬
i
∃x(t(x, y) = 0)
Observe that we now have closure under Markov’s rule with numerical
parameters.
It is just one step now to get to Kreisel’s theorem. We know from (for-
malized) recursion theory that a function f is provably recursive in HAwith
index e if HA ¬ ∀x∃yT(e, x, y), where T is Kleene’s T-predicate, which for-
malizes the notion y is a (halting) computation on input x (strictly speaking
‘y is the code of a computation . . . ’).
Now PA ¬ ∀x∃yT(e, x, y) ⇔PA ¬ ∃yT(e, x, y) ⇔ (G¨odel translation)
HA ¬ ∃yT(e, x, y) ⇔ (closure under Markov’s Rule) HA ¬ ∃yT(e, x, y)
⇔HA ¬ ∀x∃yT(e, x, y).
Friedman showed that this fact also holds in intuitionistic set theory,
IZF, ([Friedman 1977])
Finally we mention one more principle that has a metamathematical use
in G¨odel’s Dialectica translation (cf. [Troelstra 1973]): the independence of
premiss principle, IP.
IP (A → ∃xB(x)) → ∃x(A → B(x))
IP is not provable in HA. One can see that it strengthens the intended
meaning as follows: in the left hand formula, the existence of x is concluded
on the basis of a proof of A, so this proof may enter into the computation
of x. Whereas in the right hand formula, x must be computed straightaway.
However, HA is closed under the Independence of Premiss Rule:
HA ¬ (A → ∃xB(x)) ⇒HA ¬ ∃x(A → B(x)) (cf. [?], p. 296).
1.4 Set Theories
Following the tradition in set theory, the various intuitionistic modifications
of ZF are also first-order. For most classical theories one can consider one
or more corresponding intuitionistic theories. In some cases it suffices to
omit PEM from the logical axioms, one has to be careful however, when
some non-logical axioms are by themselves strong enough to imply PEM.
Here is an example: the full axiom of choice implies PEM. Let A be a
statement, define
P = ¦n ∈ N[n = 0 ∨ (n = 1 ∧ A)¦
Q = ¦n ∈ N[n = 1 ∨ (n = 0 ∧ A)¦,
26
S = ¦X, Y ¦.
Obviously ∀X ∈ S∃x ∈ N(x ∈ X) AC would yield a choice function F
such that ∀X ∈ S(F(x) ∈ X). Observe that F(P), F(Q) ∈ N, so F(P) =
F(Q) ∨F(P) ,= F(Q). If F(P) = F(Q), then A holds. If, on the other hand
F(P) ,= F(Q), then A. So we get A∨ A.
The above fact was discovered by Diaconescu ([Diaconescu 1975]), the
proof given here is Goodman and Myhill’s ([Goodman-Myhill 1978]).
Harvey Friedman studied IZF, an intuitionistic version of ZF. For a
formulation see ([Beeson 1985], Ch. 8, ¸1). The axioms for sets are mod-
ifications of the traditional ZF ones. The theory is very sensitive to the
formulation, a wrong choice of axiom may introduce unwanted logical prin-
ciples. E.g. ∈-induction is used rather than Foundation, and similarly the
Axiom of Collection takes the place of the replacement axiom.
Friedman has shown that the G¨odel translation theorem works for a
suitable formulation of set theory ([Friedman 1973], [Beeson 1985], p. 174).
For more information on the topic the reader is referred to the literature.
(see [Beeson 1985]).
Peter Aczel considered another version of constructive set theory, CZF.
This set theory has the attractive feature that it is interpretable in a par-
ticular type theory of Martin-L¨of, cf. [Troelstra–van Dalen 1988], p. 624.
Set theory under the assumption of Church’s Thesis was extensively stud-
ied by David McCarty, [McCarty 1984], [McCarty 1986], [McCarty 1991].
He built a model of cumulative constructive set theory in which a number
of interesting phenomena can be observed, e.g. in Kleene’s realizability uni-
verse sets with apartness are subcountable (i.e. the range of a function on
N).
Suggested further reading: [Kleene 1952], [Troelstra–van Dalen 1988], [Dummett 1977],
[Gabbay 1982].
Hilbert designed his proof theory for the purpose of consistency proofs,
Gentzen on the other hand considered proofs as an object for study. In his
system the structure of derivations is the object of investigation.
The fundamental theorem of the natural deduction system is the so-
called normalization theorem. Gentzen noted that there are obvious super-
fluous steps in proofs that should be avoided. Here is one:
T T’
A B
A∧ B
A
.
We first introduced A ∧ B and subsequently eliminated it. The derivation
27
T
A
would have accomplished the same.
Introductions, immediately followed by eliminations are called cuts. Gentzen
showed how to eliminate cuts from natural deduction derivations (in intu-
itionistic or classical prediate logic). A cut free derivation is also called a
normal derivation, and the process of eliminating cuts is called normaliza-
tion.
The normalization theorem says that each derivation can be converted
by cut elimination steps to a normal derivation. The theorem has a strong
and a weak form. The weak form states that a derivation can be converted
into a normal form. The strong form says that each sequence of conversions
will lead to a normal form (and actually one and the same normal form).
There is a structure theorem for normal derivations that tells us that in
normal proofs one can run in a systematic way through derivation (starting
at the top land going from major assumption to conclusion) so that all the
elimination steps precede the introduction steps (with possible ⊥-steps in
between).
From the normal form theorem many corrollaries can be drawn. For
example:
(i) Intuitionistic proposition calculus is decidable, i.e. there is a decision
method for ¬
i
A.
(ii) The disjunction property for IQC. If ∨ does not occur positively in any
formula in Γ ¬
i
A∨ B, the Γ ¬
i
A or Γ ¬
i
B.
(iii) The existence property for IQC. If ∃ and ∨ do not occur positively in
any formula in Γ and Γ ¬
i
∃xA(x), then Γ ¬
i
A(t) for some closed t.
(iv) ¬
i
A is decidable for prenex formulas A.
Since full predicate logic is not decidable (Church’s theorem for intuition-
istic predicate calculus), this result shows that in IQC not every formula is
equivalent to a prenex formula.
For details about normalization see [Gentzen 1935], [Girard 1987],
[Girard 1989], [Dalen 1997], [Troelstra–van Dalen 1988], [Troelstra-Schwichtenberg 1996].
The revival of Gentzen’s ideas is due to Prawitz’s beautiful work, [Prawitz 1965],
[Prawitz 1970], [Prawitz 1977].
Since natural deduction is so close in nature to the proof interpretation,
it is perhaps not surprising that a formal correspondence between a term
calculus and natural deduction can be established.
28
We will first demonstrate this for a small fragment, containing only the
connective ‘→’. Consider an → introduction:
[A]
T
B
A → B
[x : A]
T
t : B
λx t : A → B
We assign in a systematic way λ-term to formulas in the derivation. Since
A is an assumption, it has a hypothetical proof term, say x. On discharGing
the hypotheses; we introduce a λx in front of the (given) term t for B. By
binding x, the proof term for A → B no longer depends on the hypothetical
proof x of A.
The elimination runs as follows:
A → B A t : A → B s : A
B t(s) : B
Observe the analogy to the proof interpretation. Let us consider a par-
ticular derivation.
[A] [x : A]
B → A λy x : B → A
A → (B → A) λx λy x : A → (B → A)
Thus the proof term of A → (B → A) is λxy.x, this is Curry combinator
K. Note that the informal argument of page 3 is faithfully reflected.
Let us now consider a cut elimination conversion.
x : B
T
t : A T

λx t : B → A s : B
(λx t)(s) : A
reduces to
T

s : B
T[∫/¸]
t[s/x]A
The proof theoretic conversion corresponds to the β-reduction of the*
λ-calculus.
In order to deal with full predicate logic we have to introduce specific
operations in order to render the meaning of the connectives and their deriva-
tion rules. Here is a list:
29

p −− pairing
p
0
, p
1
−− projections

D −− discriminator (“case dependency”)
k −− case obliteration
E – witness extractor
⊥ – ex falso operator
∧ I t
0
: A
0
t
1
: A
1
∧ E t : A
0
∧ A
1
(i ∈ ¦0, 1¦)
p(t
0
, t
1
) : A
0
∧ A
1
p
i
(t) : A
i
∨ I t : A
i
(i ∈ ¦0, 1¦) ∨ E t : A∨ B t
0
[x
A
] : C t
1
[x
B
] : C
k
i
; A
0
∨ A
1
D
u,v
(t, t
0
[u], t
1
[v]) : C
→ I t[x
A
] : B → E t : A → B t

: A
λy
A
t[y
A
] : A → B t(t

) : B
∀ I t[x] : A(x) ∀ E t : ∀xA(x)
λy t[y] : ∀yA(y) t(t

) : A(t

)
∃ I t
1
: A(t
0
) ∃ E t : ∃xA(x) t
1
[y, z
A(y)
] : C
p(t
0
, t
1
) : ∃xA(x) E
u,v
(t, t
1
[u, v]) : C
There are a number of details that we have to mention.
(i) In → I the dependency on the hypothesis has to be made explicit in
the term. We do this by assigning to each hypothesis its own variable.
E.g. x
A
: A.
(ii) In ∨E (and similarly ∃E) the dependency on the particular (auxilliary)
hypotheses A and B disappears. This is done by a variable binding
technique. In D
u
, v the variables u and v are bound.
(iii) In the falsum rule the result, of course, depends on the conclusion A.
So A has its own ex falso operator ⊥
A
.
Now the conversion rules for the derivation automatically suggest the con-
version for the term.
Normalization and cut-elimination for second-order logic is far more com-
plicated because of the presence of the comprehension rule, which introduces
a impredicativity. The proof of a normalization theorem was a spectacular
30
breakthrough, the names to mention here are Girard, Prawitz and Martin-
L¨of, see their papers in [Fenstad 1971].
Suggested further reading: [Troelstra-Schwichtenberg 1996], [Troelstra–van Dalen 1988]
Ch. 10, [Dummett 1977] Ch. 4, [Girard 1989], [Buss 1998], [Beeson 1985],
[Kleene 1952] Ch. XV, [Troelstra 1973].
We have seen in the previous section that the term calculus corresponds
with the natural deduction system. This suggests a correspondence between
proofs and propositions on the one hand and elements (given by the terms)
and types (the spaces where these terms are to be found). This correspon-
dence was first observed for a simple case (the implication fragment) by
Haskell Curry, [Curry-Feys 1958], ch. 9, ¸ E, and extended to full intuition-
istic logic by W. Howard, [Howard 1980]. Let us first look at a simple case,
the one considered by Curry.
Since the meaning of proposition is expressed in terms of possible proofs
— we know the meaning of A if we know what things qualify as proofs —
one may take an abstract view and consider a proposition as its collection of
proofs. From this viewpoint there is a striking analogy between propositions
and sets. A set has elements, and a proposition has proofs. As we have seen,
proofs are actually a special kind of constructions, and they operate on each
other. E.g. if we have a proof p : A → B and a proof q : A then p(q) : B.
So proofs are naturally typed objects.
Similarly one may consider sets as being typed in a specific way. If A
and B are typed sets then the set of all mappings from A to B is of a higher
type, denoted by A → B or B
A
. Starting from certain basic sets with types,
one can construct higher types by iterating this ‘function space’-operation.
Let us denote ‘a is in type A’ by a ∈ A.
Now there is this striking parallel.
Propositions Types
a : A a ∈ A
p : A → B, q : A p ∈ A → B, q ∈ A
⇒ p(q) : B ⇒ p(q) ∈ B
x : A ⇒ t(x) : B x : A ⇒ t(x) ∈ B
then λx t : A → B then λx t ∈ A → B
It now is a matter of finding the right types corresponding to the remain-
ing connectives. For ∧ and ∨ we introduce a product type and a disjoint
sum type. For the quantifiers generalizations are available. The reader is
referred to the literature, cf. [Howard 1980], [Gallier 1995].
31
The main aspect of the Curry-Howard isomorphism, (also knowns as
“proofs as types”), is the faithful correspondence:
proofs
propositions
=
elements
types
with their conversion and normalization properties.
Per Martin-L¨of was the first logician to see the full importance of the
connection between intuitionistic logic and type theory. Indeed, in his ap-
proach the two are so closely interwoven, that they actually merge into one
master system. His type systems are no mere technical innovations, but
they intend to catch the foundational meaning of intuitionistic logic and
the corresponding mathematical universe. Expositions can be found in e.g.
[Martin-L¨of 1975], [Martin-L¨of 1984].
Suggested further reading: [Girard 1989], [Gallier 1995], [Troelstra–van Dalen 1988].
Whereas first-order intuitionistic logic is a subsystem of classical logic, higher-
order logic may contain rules or axioms that are constructively justified, but
which contradict classical logic.
In practice there are two ways to formulate second-order logic: with set
variables and with function variables.
1.5 IQC
2
Second-order logic with set variables is a straightforward adaptation of the
classical formulation; cf. [Dalen 1997], [Troelstra-Schwichtenberg 1996], CH.
11. It is a surprising fact that in IQC
2
the connectives are not independent
as in first-order logic. Prawitz showed that one can define the connectives
in ∀
1
, ∀
2
and →, [Prawitz 1965].
⊥ ↔ ∀
2
X.X
A∧ B ↔ ∀
2
X((A → (B → X)) → X)
A∨ B ↔ ∀
2
X((A → X) ∧ (B → X) → X)

1
xA ↔ ∀
2
X(∀
1
y(A → X) → X)

2
XA ↔ ∀
2
X(∀Y ((A → X) → X)
(where X is a 0-ary predicate)
Classically, sets and functions are interdefinable as each set has a charac-
teristic function. But intuitionistically S has a characteristic function of its
membership is decidable:

a ∈ S ⇔ k
S
(a) = 1
a ,∈ S ⇔ k
S
(a) = 0
32
Since k
S
(a) = 1∨k
S
(a) = 0, we get a ∈ S∨a ,∈ S. The moral is that there are
lots of sets without characteristic functions. We see that the “set”-approach
and the “function”-approach to second-order logic (arithmetic) yield diverg-
ing theories. Generally speaking, total functions, with their ‘input–output’
behaviour are more tractable than sets. That is a good reason to study
second-order arithmetic with function variables. Another reason is that this
formulation is a natural framework for treating Brouwer’s choice sequences.
For the proof theory of IQC
2
, see [Prawitz 1970], [Troelstra-Schwichtenberg 1996],
[Troelstra 1973].
1.6 The theory of choice sequences
For the practice of intuitionistic mathematics, second-order arithmetic with
function variables is even more significant than the version with set vari-
ables. The reason is that this theory allows us to capture the properties of
Brouwer’s choice sequences. Since this survey is about logic, the topic is not
quite within its scope. Let us therefore just briefly note the main points, for
more information the reader is referred to the literature.
Brouwer’s chief contribution to this part of intuitionistic logic is that
he realized that particular quantifier combinations get a specific reading.
Choice sequences of, say, natural numbers are infinite sequences, α, of natu-
ral numbers, chosen more or less arbitrarily. That is to say, in general there
is no law that determines future choices. Suppose now that we have shown
∀α∃nA(α, n) for some formula A, this means that when we generate a choice
sequence α, we will eventually be able to compute the number n, such that
A(α, n) holds. Roughly speaking, this tells us that at some stage in the
generation of α, we have all the information we need for the computation
of n, but then we need no further information, and any β that coincides so
far with α will yield the same n. This sketch is somewhat simplified, for
an extensive analysis cf. [van Atten–van Dalen ??]. This, in a nutshell, is
Brouwer’s Continuity Principle:
∀α∃xA(α, n) → ∀α∃xy∀β(α(y) = β(y) → A(β.x)
Here γ(y) stands for the initial segment of length y of a sequence γ
On the basis of this principle, formulated for the ∀α∃!x in [Brouwer 1918],
already a number of intuitionistic facts, conflicting with classical mathemat-
ics, can be derived. It allows, for example, a simple rejection of PEM.
In later papers Brouwer further investigated the continuity phenomenon.
He added a powerful induction principle, which enabled him to show his fa-
mous Continuity Theorem: all total real functions on the continuum are
33
locally uniformly continuous, [Brouwer 1924]. As a corollary the contin-
uum cannot be decomposed into two inhabited parts. Another well-know
consequence is the Fan theorem (basically the compactness of the Cantor
space).
Brouwer introduced in the twenties an extra strengthening of analysis,
the details of which were published only after 1946. The new idea, known
by the name “the creating subject”, was formulated by Kreisel in terms of
a tensed modal operator. Subsequently Kripke simplified the presentation
by introducing a choice sequence α that “witnesses” a particular statement
A. α keeps track of the succes of the subject in establishing A; it produces
zeros as long as the subject has not established A, and when A is proved, or
experienced, α produces a single ‘one’ and goes on with zeros. The existence
of such a α is the content of Kripke’s schema:
KS ∃α(A ↔ ∃xα(x) = 1)
Brouwer used the creating subject (and implicitly Kripke’s schema) to es-
tablish strong refutations, which go beyond the already existing Brouwerian
counterexamples.
He showed for example that (cf. [Brouwer 1949], [Hull 1969])
∀x ∈ R(x > 0 → x > 0) and
∀x ∈ R(x ,= 0 → x#0)
Kripke’s schema has indeed consequences for the nature of the mathe-
matical universe, e.g. it conflicts with ∀α∃β-continuity, ([Myhill 1966]), that
is ∀α∃βA(α, β) → ∃F∀αA(α, F(α), where F is a continuous operation.
Using KS, van Dalen showed that Brouwer’s indecomposability theorem
for R can be extended to all dense negative subsets of R(A ⊆ R is negative
if ∀x(x ∈ A ↔ x ∈ A). So, for example, the set of not-not-rationals is
indecomposable (cf. [Dalen 1999A]).
In order to demonstrate the technique of the creating subject, we show
here that, under the assumption of Kripke’s schema one can show a converse
of Brouwer’s indecomposability theorem: KS + R is indecomposable ⇒there
are no discontinuous real functions.
Proof. Let f be discontinuous in 0, and f(0) = 0. So ∃k∀n∃x([f(x)[ > 2
−k
).
Hence there are x
n
with [f(x
n
)[ > 2
−k
.
Let α be the Kripke sequence for r ∈ Q, and β for r ,∈ Q. Put

γ(2n) = α(n)
γ(2n + 1) = β(n)
c
n
=

x
n
if ∀p ≤ n γ(k) = 0
x
p
if p ≤ n and γ(p) = 1
and c = lim(c
n
)
34
Now, f(c) < 2
−k
∨ f(c) > 0. If f(c) < 2
−k
then f(0) = 0, so ∀p(γ(p) = 0).
Contradiction. If f(c) > 0, then r ∈ Q ∨ r ,∈ Q. Hence ∀r(r ∈ Q ∨ r ,∈ Q).
This contradicts the indecomposability of R.
The theory of choice sequences has received a great deal of attention
after Kleene and Kreisel formulated suitable formalizations. Some notions,
such as “lawless sequence”, have found important applications in metamath-
ematics. C.f. [Brouwer 1981], [Brouwer 1992], [?], [Kleene–Vesley 1965],
[Kreisel–Troelstra 1970], [Troelstra 1977], [Troelstra–van Dalen 1988] Ch. 12,
[Beeson 1985], [Dummett 1977] Ch. 3.
There is an extensive literature on the semantics of second-order arith-
metic, c.f. [Scott 1968], [Scott 1970] [Moschovakis 1973], [Dalen 1978], [Fourman-Hyland 1979],
[Fourman-Scott 1979], [Hoeven-Moerdijk 1984].
References
[van Atten–van Dalen ??] M.van Atten, D. van Dalen. Arguments for
Brouwer’s continuity principle. Forthcoming.
[Beeson 1976] M. Beeson. The unprovability in intuitionistic formal systems
of the continuity of effective operations on the reals. J Symb Logic, 41:
18–24.
[Beeson 1985] ———– Foundations of Constructive Mathematics. Springer,
Berlin.
[Beth 1956] E.W. Beth. Semantic Construction of Intuitionistic Logic. Kon.
Nederlandse Ac. Wetenschappen afd. Letteren:Mededelingen, 19/11:
357–388.
[Borwein 1998] J.M. Borwein. Brouwer-Heyting Sequences Converge. Math.
Intelligencer, 20:14–15.
[Bridges-Richman 1987] F. Richman, Bridges, D. Varieties of Constructive
Mathematics. Cambridge University Press, Cambridge.
[Brouwer 1918] L.E.J. Brouwer. Begr¨ undung der Mengenlehre unabh¨angig
vom logischen Satz vom ausgeschlossenen Dritten. Erster Teil, Allge-
meine Mengenlehre. Kon Ned Ak Wet Verhandelingen, 5:1–43.
[Brouwer 1919a] ———– Begr¨ undung der Mengenlehre unabh¨angig vom
logischen Satz vom ausgeschlossenen Dritten. Zweiter Teil, Theorie der
Punktmengen. Kon Ned Ak Wet Verhandelingen, 7:1–33.
35
[Brouwer 1919b] ———– Intuitionistische Mengenlehre . Jahresber. Dtsch.
Math.Ver., 28:203–208. Appeared in 1920.
[Brouwer 1923] ———–
¨
Uber die Bedeutung des Satzes vom ausgeschlosse-
nen Dritten in der Mathematik insbesondere in der Funktionentheorie.
J Reine Angew Math, 154:1–8.
[Brouwer 1924] ———– Beweis dass jede volle Funktion gleichm¨assig stetig
ist. Nederl Ak Wetensch Proc, 27:189–193.
[Brouwer 1949] ———– Essentieel negatieve eigenschappen. Ind. Math.,
10:322-323.
[Brouwer 1981] ———– Brouwer’s Cambridge Lectures on Intuitionism.
Cambridge University Press, Cambridge.
[Brouwer 1992] ———– (Ed. D. van Dalen) Intuitionismus. Bibliographis-
ches Institut, Wissenschaftsverlag, Mannheim.
[Buss 1998] S. Buss. handproof. Elsevier, Amsterdam.
[Curry-Feys 1958] R. Feys Curry, H.B. Combinatory Logic. Vol. I. North–
Holland, Amsterdam.
[Dalen 1974] ———– A model for HAS. A topological interpretation of
second-order intuitionistic arithmetic with species variables. Fund.
Math., 82:167–174.
[Dalen 1978] ———– An interpretation of intuitionistic analysis. Ann Math
Log, 13:1–43.
[Dalen 1984] ———– How to glue analysis models. J Symb Logic, 49:1339–
1349.
[Dalen 1997] ———– Logic and Structure (3rd edition, revised 2008).
Springer Verlag, Berlin.
[Dalen 1999] ———– Mystic, Geometer, and Intuitionist: The Life of
L.E.J. Brouwer Vol. 1. The dawning Revolution. Oxford University
Press, Oxford.
[Dalen 2005] ———– Mystic, Geometer, and Intuitionist: The Life of
L.E.J. Brouwer Vol. 2. Hope and Disillusion. Oxford University Press,
Oxford.
36
[Dalen 1999A] ———– From Brouwerian Counter Examples to the Creating
Subject. Studia Logica, 62:305–314.
[Dalen 2002] ———– Intuitionistic Logic, In Handbook of Philosophical
Logic, 2nd edition, vol. 5, eds. D. Gabbay, F, G¨ unthner. pages 1–114.
Kluwer, Dordrecht.
[Dalen 1978] R. Statman Dalen. Equality in the presence of apartness. In
Hintikka et al(eds.) Essays on Mathematical and Philosophical Logic,
pages 95–116,Reidel. Dordrecht.
[Diaconescu 1975] R. Diaconescu. Axiom of choice and complementation.
Proc. American Math. Soc., 51:176–178.
[Dragalin 1988] A. Dragalin. Mathematical Intuitionism. American Mathe-
matical Society, Providence.
[Dummett 1975] M. Dummett. The philosophical basis of intiutionistic
logic. In J. Shepherdson H.E. Rose, editor, Logic Colloquium ’73,, pages
5–40, North-Holland. Amsterdam.
[Dummett 1977] ———– Elements of Intuitionism. Oxford University
Press, Oxford.
[Fenstad 1971] J.E. Fenstad (ed.) Proceedings of the Second Scandinavian
Logic Symposium, 405 pp. North–Holland, Amsterdam.
[Fitting 1969] M. Fitting. Intuitionistic model theory, and forcing. North-
Holland Publ. Co., Amsterdam.
[Fourman-Scott 1979] Fourman, M.P. and D.S. Scott. Sheaves and Logic,
pages 280–301. Springer, Berlin.
[Fourman-Hyland 1979] Fourman, M.P. and J.M.E. Hyland. Sheaf models
for analysis, pages 280–301. Springer, Berlin.
[Friedman 1973] H. Friedman. Some applications of Kleene’s methods for
intuitionistic systems. In [Mathias–Rogers 1973], 113–170.
[Friedman 1975] ———– The disjunction property implies the numerical
existence property. Proc. Nat. Acad. Science USA, 72:2877–2878.
[Friedman 1977] ———– Set theoretic foundations for constructive analy-
sis. Ann Math, 105:1–28.
37
[Gabbay 1972] D.M. Gabbay. Decidability of some intuitionistic theories. J
Symb Logic, 37:579–587.
[Gabbay 1973] ———– The undecidability of intuitionistic theories of alge-
braically closed and real closed fields. J Symb Logic, 38:86–92.
[Gabbay 1982] ———– Semantical Investigations in Heyting’s intuitionistic
logic. Reidel, Dordrecht.
[Gallier 1995] J. Gallier. On the Correspondence between proofs and λ-
terms. In Ph. de Groote, editor, The Curry–Howard Isomorphism,
pages 55–138, Academia. Louvain-la-Neuve.
[Gentzen 1935] G. Gentzen. Untersuchungen ¨ uber das logische Schliessen I,
II. Math. Zeitschr, 39:176–210, 405–431.
[Girard 1987] J.-Y. Girard. Proof Theory and Logical Complexity. Vol. I.
Bibliopolis, Napoli.
[Girard 1989] P. Taylor Girard, J.-Y., Y. Lafont. Proofs and Types. Cam-
bridge University Press, Cambridge.
[G¨odel 1932] K. G¨odel. Zum intuitionistischen Aussagenkalk¨ u l. Anzeiger
der Akademie der Wissenschaften in Wien, 69:65–66, . (Coll.Works I,
p. 222–225).
[Goernemann 1971] S. Goernemann. A logic stronger than intuitionism. J
Symb Logic, 36:249–261.
[Goodman-Myhill 1978] Goodman, N.D. and J. Myhill. Choice implies ex-
cluded middle. Z.Math.Logic Grundlagen Math., 24:461.
[Grayson 1979] R.J. Grayson. Heyting-valued models for intuitionistic set
theory. In (Eds. M.P. Fourman, C.J. Mulvey, D.S. Scott, editor,
durham, pages 402–414,Springer. Berlin.
[Grayson 1981] ———– Concepts of general topolog in constructive math-
ematics and in sheaves. Ann Math Log, 20:1–41.
[Grayson 1982] ———– Concepts of general topolog in constructive math-
ematics and in sheaves. II. Ann Math Log, 23:55–98.
[Grayson 1984] ———– Heyting–valued Semantics. In Lolli, G., G. Longo,
A. Marcia, editor, Proceedings of the Logic Colloquium ’82, pages 181–
208, Amsterdam. Elsevier.
38
[Heyting 1925] A. Heyting. Intu¨ıtionistische axiomatiek der projectieve
meetkunde. PhD thesis, Amsterdam.
[Heyting 1930a] ———– Die formalen Regeln der intuitionistischen
Logik. Die Preussische Akademie der Wissenschaften. Sitzungsberichte.
Physikalische-Mathematische Klasse., 42–56.
[Heyting 1930b] ———– Die formalen Regeln der intuitionistischen Mathe-
matik II, III. Die Preussische Akademie der Wissenschaften. Sitzungs-
berichte. Physikalische-Mathematische Klasse., 57–71, 158–169.
[Heyting 1934] ———– Mathematische Grundlagenforschung. Intuitionis-
mus, Beweistheorie. Springer Verlag, Berlin.
[Heyting 1956] ———– Intuitionism, an Introduction . North–Holland,
Amsterdam,.
[Hoeven-Moerdijk 1984] Hoeven, G..F. and Moerdijk, I. Sheaf models for
choice sequences. Ann Pure Appl Logic, 27:63–107.
[Howard 1980] W.A. Howard. The Formulas-as-Types Notion of Construc-
tion. In J.R. Hindley J.P. Seldin, editor, To H.B. Curry: Essays on
Combinatory Logic, Lambda calculus and Formalism., pages 479–490,
Academic Press. New York.
[Hull 1969] R.G. Hull Counterexamples in intuitionistic analysis using
Kripke’s schema. Zeitschr. math. Logik, 15:241–246.
[Hyland 1982] J.M.E. Hyland. The Effective Topos. In D. van Dalen A.S.
Troelstra, editor, The L.E.J. Brouwer Centenary Symposium, pages
165–216,Elsevier. Amsterdam.
[Jaskowski 1936] Recherches sur le syst`eme de la logique intuitioniste,
Congr. intern. Phil. des Sciences, 6: 58–61, Hermann. Paris.
[Kleene 1952] S.C. Kleene. Introduction to Metamathematics. North-
Holland, Amsterdam..
[Kleene–Vesley 1965] Kleene, S.C. and R.E. Vesley. The Foundations of
Intuitionistic Mathematics especially in relation to Recursive Functions.
North–Holland, Amsterdam..
[Kolmogorov 1932] A.N. Kolmogorov. Zur Deutung der intuitionistischen
Logik. Math. Zeitschr, 35:58–65..
39
[Kreisel 1958] G. Kreisel. Elementary completeness properties of intuition-
istic logic with a note on negations of prenex formulae. J Symb Logic,
23:317–330..
[Kreisel–Troelstra 1970] Kreisel, G. and A.S. Troelstra. Formal systems for
some branches of intuitionistic analysis. Ann Math Log, 1:229–387..
[Kripke 1965] S.A. Kripke. Semantical Analysis of Intuitionistic Logic. In
M.A.E. Dummett J.N. Crossley, editor, Formal Systems and Recursive
Functions. Proc. of the eighth Logic Colloquium, Oxford 1963, pages
92–130, North-Holland. Amsterdam.
[Kripke 1968] ———– Semantical Analysis of Intuitionistic Logic II. un-
published,
[Lambek–Scott 1986] Lambek, J. and P.J. Scott. Introductioin to higher
order categorical logic. Cambridge University press, Cambridge.
[Lifschitz 1967] V. Lifschitz. The decision problem for some constructive
theories of equality. Seminar Math. Institute Steklov, 4:29–31.
[MacLane–Moerdijk 1992] MacLane, S. and I. Moerdijk. Sheaves in Geom-
etry and Logic. A first Introduction to Topos Theory. Springer, Berlin.
[Markov 1971] A.A. Markov. On Constructive Mathematics. AMS Transl.,
98:1–9. Original Trudy Mat. Inst. Steklov.67 (1962).
[Martin-L¨of 1975] P. Martin-L¨of. An Intuitionistic Theory of Types. In
H.E. Rose and J.C. Shepherdson, editors, Logic Colloquium ’73, pages
73–118,North–Holland. Amsterdam.
[Martin-L¨of 1984] ———– Intuitionistic Type Theory. Bibliopolis, Napoli.
[Maslov 1965] V.P. Orevkov Maslov, S.Yu., G. Mints. Unsolvability in the
constructive predicate calculus of certain classes of formulas containing
only monadic predicate variables. Soviet Mat. Dokl., 6:918–920.
[Mathias–Rogers 1973] Mathias,A. and H. Rogers (eds.) . Cambridge Sum-
mer School in Mathematical Logic. Springer Verlag, Berlin.
[McCarty 1986] C. McCarty. Realizability and Recursive Set Theory. Ann
Pure Appl Logic, 32:153–183.
[McCarty 1984] ———– Realizability and Recursive Mathematics. PhD
thesis, Oxford.
40
[McCarty 1991] ———– Polymorphism and Apartness. Notre Dame Jour-
nal of Formal Logic, 32:513–532.
[McLarty 1995] C. McLarty. Elementary Categories, Elementary Toposes.
Clarendon Press, Oxford.
[Moschovakis 1973] J. Moschovakis. A Topological Interpretation of Intu-
itionistic Analysis. Comp Math, 26:261–275.
[Myhill 1966] J. Myhill, Notes towards an axiomatization of intuitionistic
analysis, Logique et Analyse 9, 280-297.
[Negri–von Plato] S. Negri, J. von Plato. Structured Proof Theory. Forth-
coming.
[Prawitz 1965] D. Prawitz. Natural Deduction. A proof-theoretical study.
Almqvist & Wiksell, Stockholm.
[Prawitz 1970] ———– Some results for intuitionistic logic with second
order quantification. In Myhill, J and A. Kino and R. Vesley, editor,
buffalo, pages 259–269, North-Holland. Amsterdam.
[Prawitz 1977] ———– Meaning and Proofs: On the Conflict between Clas-
sical and Intuitionistic Logic. Theoria, 43:2–40.
[Rasiowa-Sikorski 1963] Rasiowa, H. and R. Sikorski. The Mathematics of
Metamathematics. Panstowe Wydawnictwo Naukowe, Warsaw.
[Sch¨ utte 1968] K. Sch¨ utte. Vollst¨andige Systeme modaler und intuitionis-
tischer Logik. Springer, Berlin.
[Troelstra-Schwichtenberg 1996] A.S. Troelstra Schwichtenberg, H. Basic
proof theory. Cambridge University Press, Cambridge.
[Scott 1968] D.S. Scott. Extending the topological interpretation to intu-
itionistic analysis. Comp. Math., 20:194–210.
[Scott 1970] ———– Extending the topological interpretation to intuition-
istic analysis II. In Myhill, J and A. Kino and R. Vesley, editor, buffalo,
pages 235–255,North-Holland. Amsterdam.
[Scott 1979] ———– Identity and Existence in Intuitionistic logic, pages
660–696. Springer, Berlin.
41
[Smorynski 1973a] C.S. Smorynski. Applications of Kripke models. In
A.S. Troelstra, editor, Metamathematical Investigation of Intuitionistic
Arithmetic and Analysis, pages 324–391, Springer. Berlin.
[Smorynski 1973b] ———– Investigation of Intuitionistic Formal Systems
by means of Kripke Models. PhD thesis, Chicago.
[Smorynski 1973c] ———– On axiomatizing fragments. J Symb Logic, 42:
530–544.
[Smorynski 1978] ———– The axiomatizing problem for fragments. Ann
Math Log, 14:193–221.
[Smorynski 1982] ———– Nonstandard models and constructivity. In D.
van Dalen A.S. Troelstra, editor, The L.E.J. Brouwer Centenary Sym-
posium, pages 459–464, Elsevier. Amsterdam.
[Tarski 1938] A. Tarski. Der Aussagenkalk¨ ul und die Topologie. Fund.
Math., 31:103–134.
[Troelstra 1973] A.S. Troelstra. Metamathematical Investigation of Intu-
itionistic Arithmetic and Analysis. Springer Verlag, Berlin.
[Troelstra 1977] ———– Choice Sequences. A chapter of intuitionistic
mathematics. Clarendon Press, Oxford.
[Troelstra 1978] ———– Heyting, “Die formale Regeln der intuitionistis-
chen Logik” (1930). Commentary. In “Two decades of mathematics in
the Netherlands”, Eds. Bertin, E.M.J., H.J.M. Bos, A.W. Grootendorst,
1. Mathematical Centr. Amsterdam. 163–175.
[Troelstra–van Dalen 1988] Troelstra, A.S., D. van Dalen. Constructivism
in Mathematics, I, II. North-Holland Publ. Co., Amsterdam.
[Veldman 1976] W. Veldman. An intuitionistic completeness theorem for
intuitionistic predicate calculus. J. Symb. Logic, 41:159–166.
42

A proof of A∧B is simply a pair of proofs a and b of A and B. For convenience we introduce a a notation for the pairing of constructions, and for the inverses (projections); (a, b) denotes the pairing of a and b, and (c)0 ,(c)1 , are the first and second projection of c. Now, the proof of a disjunction A ∨ B is a pair (p, q) such that p carries the information, which disjunct is correct, and q is the proof of it. We stipulate that p ∈ {0, 1}. So p = 0 and q : A or p = 1 and q : B. Note that this disjunction is effective, in the sense that the disjunct is specified, this in contrast with classical logic, where one does not have to know which disjunct holds. Negation is also defined by means of proofs: p : ¬A says that each proof a of A can be converted by the construction p into a proof of an absurdity, say 0 = 1. A proof of ¬A thus tells us that A has no proof! The most interesting propositional connective is the implication. The classical solution, i.e. A → B is true if A is false or if B is true, cannot be used because here the classical disjunction is used; moreover, it assumes that the truth values of A and B are known before one can settle the status of A → B. Heyting showed that this is asking too much. Consider A = “there occur twenty consecutive 7’s in the decimal expansion of π”, and B = “there occur nineteen consecutive 7’s in the decimal expansion of π”. Then ¬A ∨ B does not hold constructively, but on the interpretation following here, the implication, A → B is obviously correct. The intuitionistic approach, based on the notion of proof, demands a definition of a proof a of the implication A → B in terms of (possible) proofs of A and B. The idea is quite natural: A → B is correct if we can show the correctness of B as soon as the correctness of A has been established. In terms of proofs: p : A → B if p transforms each proof q : A into a proof p(q) : B. The meaning of the quantifiers is specified along the same lines. Let us assume that we are dealing with a given domain D of mathematical objects. A proof p of ∀xA(x) is a construction which yields for every object d ∈ D a proof p(d) : A(d). A proof p of ∃xA(x) is a pair (p0 , p1 ) such that p1 : A(p0 ). Thus the proof of an existential statement requires an instance plus a proof of this instance.

2

a:A a :⊥ a:A∧B a:A∨B a:A→B a : ∃xA(x) a : ∀xA(x) a : ¬A

conditions false a = (a1 , a2 ), where a1 : A and a2 : B a = (a1 , a2 ), where a2 : A if a1 = 0 and a2 : B if a1 = 1 for all p with p : A a(p) : B a = (a1 , a2 ) and a2 : A(a1 ) for all d ∈ D a(d) : A(d) where D is a given domain for all p : A a(p) :⊥

Observe that an equivalent characterization of the disjunction can be given: a = (a1 , a2 ), where a1 = 0 and a2 : A or a1 = 1 and a2 : B. The above formulation has the technical advantage that only ∧ and → are required, and they belong to the ‘negative fragment’. The interpretation of the connectives in terms of proofs was made explicit by Heyting in [Heyting 1934]. Kolmogorov had given a similar interpretation in terms of problems and solutions, [Kolmogorov 1932]. The formulations are, up to terminology, virtually identical. We will demonstrate the proof interpretation for a few statements. 1. A → (B → A). We want an operation p that turns a proof a : A into a proof of B → A. But if we already have a proof a : A, then there is a simple transformation that turns a proof q : B into a proof of A, i.e. the constant mapping q → a, which is denoted by λq · a. And so the construction that takes a into λq · a is λa · (λq · a), or in an abbreviated notation λaλq · a. Hence λaλq.a : A → (B → A). 2. (A ∨ B) → (B ∨ A). Let p : A ∨ B, then p0 = 0 and (p)1 : A or (p)0 = 1 and (p)1 : B. By interchanging A and B we get, looking for q : B ∨ A: (q)0 = 1 and (q)1 : B or (q)1 = 0 and (q)1 : A, which comes to sg((p)0 ) = (q)0 and (p1 ) : B or sg((p)0 ) = (q)0 and (q)1 : A that is (sg((p0 )), (p)1 ) : B ∨ A. And so λp · (sg((p0 ), (p)1 ) : A ∨ B → B ∨ A. 3. A → ¬¬A A proof q of ¬¬A is a proof of ¬A →⊥. Assume p : A, and q : ¬A. Then q(p) :⊥, so λq · q(p) : ¬¬A. Hence λpλq · q(p) : A → ¬¬A. 4. ¬¬A → A This turns out to be unprovable. For assume that a : ¬¬A, then we

3

know that A →⊥ has no proof, and that is all. No more information can be extracted. Hence we cannot claim that there is a proof of A. 5. A ∨ ¬A p : A ∨ ¬A ⇔ (p)0 = 0 and (p)1 : A or (p)0 = 1 and (p)1 : ¬A. However, for an arbitrary proposition A we do not know whether A or ¬A has a proof, and hence (p)0 cannot be computed. So, in general there is no proof of A ∨ ¬A. 6. ¬∃xA(x) → ∀x¬A(x) p : ¬∃xA(x) ⇔ p(a) :⊥ for a proof a : ∃xA(x). We have to find q : ∀x¬A(x), i.e. q(d) : A(d) →⊥ for any d ∈ D. So pick an element d and let r : A(d), then (d, r) : ∃xA(x) and so p((d, r)) :⊥. Therefore we put (q(d))(r) = p((d, r)), so q = λrλd · p((d, r)) and hence λpλrλd · p((d, r)) : ¬∃xA(x) → ∀x¬A(x). Brouwer handled logic in an informal way, he showed the untenability of certain classical principles by a reduction to unproven statements. His technique is goes by the name of ‘Brouwerian counterexample’. A few examples are shown here. Brouwer considered certain open problems, usually formulated in terms of the decimal expansion of π. In spite of considerable computational power, there are enough open questions concerning the occurrence of specific sequences of decimals.1 Let us illustrate the technique: we compute simultaneously the decimals of π and the members of a Cauchy sequence. We use N (k) as an abbreviation for “the decimals pk−89 , . . . , pk of π are all 9”. Now we define: an = (−2)−n if ∀k ≤ n¬N (k) (−2)−k if k ≤ n and N (k)

an is an oscillating sequence of negative powers of −2 until a sequence of 90 nines in π occurs, from then onwards the sequence is constant: 1 1 1 1, − 2 , 1 , − 1 , 16 , − 32 , . . . (−2)−k , (−2)−k , (−2)−k , (−2)−k , . . .. 4 8 The sequence determines a real number a, in the sense that it satisfies the Cauchy condition. The sequence is well-defined, and N (n) can be checked for each n (at least in principle). For this particular a we cannot say that it is possitive, negative or zero. a > 0 ⇔ N (k) holds for the first time for an even number a < 0 ⇔ N (k) holds for the first time for an odd number
Brouwer considered various sequences, for one then the occurrence has been established: 01234567890 does indeed occur among the decimals of π, cf. [Borwein 1998].
1

4

But there is no proof that a is rational. with the motto ‘stones for bread’. Contradiction. strong counterexamples cannot always be expected. but he was the first to establish a non-trivial result: ¬A ↔ ¬¬¬A ([Brouwer 1919b]).[Brouwer 1923]. but it is not excluded that eventually a proof may be found. [Dummett 1977] Ch. One also easily sees that a = 0 ∨ a = 0 fails to have proof. 1. for then N (k) would never apply. vol 6 – Special Issue: Perspectives on Intuitionism (ed. Brouwer did not develop logic for its own sake. We give here a list of axioms taken from [Troelstra–van Dalen 1988]. cannot be said to have a proof. So ¬¬A → A fails. [Heyting 1956]. For ¬(A ∨ ¬A) is equivalent to ¬A ∧ ¬¬A. Hence we have shown (¬¬a is rational). The above number a cannot be irrational. which is a contradiction! We will get to some strong refutations of classical principles in the following sections. That is to say. Tieszen). In formal logic there is a similar distinction: A and ¬A. The Brouwerian counterexamples are weak in the sense that they show that some proposition has as yet no proof. the negation cannot be proved. For example. p. Arend Heyting sent in an essay. p. [Bridges-Richman 1987] Ch. 86. a system with only two derivation rules and a large number of axioms. we cannot affirm the trichotomy law. Philosophia Mathematica. R. in which he provided a formal system for intuitionistic predicate logic. 1. ∀x ∈ R(x < 0 ∨ x = 0 ∨ x > 0). Heyting had patiently checked the system of Russell’s Principia Mathematica. although there are instances of the Principle of the Excluded Third (PEM) where no proof has been provided. 7. [?]. A ∧ B → A. 5 . A ∧ B → B.a = 0 ⇔ N (k) holds for no k. 129 – 226. Brouwer used his counter examples to show that some classical mathematical theorems are not intuitionistically valid. The system was presented in ‘Hilbert style’ (as it is called now). and hence a = 0. Since we have no effective information on the status of the occurence of N (k)’s. cf. The Brouwerian counter examples are similar to the first case. In 1928 the Dutch Mathematics Society posed in its traditional prize contest a problem asking for a formalization of Brouwer’s logic. and isolated a set of intuitionistically acceptable axioms. A → (B → (A ∧ B). His system was chosen in such a way that the additon of PEM yields full classical logic. Suggested further reading: [Heyting 1934].

e. In 1934 Gerhardt Gentzen introduced two new kinds of formalization of logic. ∀x(B → A(x)) → (B → ∀xA(x)). → E. [Troelstra 1978]. cf. see below) and the rule of generalization: i A(x) ⇒ i ∀xA(x). There are two derivation rules: the well-known modus ponens (i. ⊥→ A. Both of these are eminently suited for the investigation of formal deriations as objects in their own right. did not cover full intuitionistic logic. however. When no confusion arises. (A → (B → C) → ((A → B) → (A → C)). 6 . which. ∀xA(x) → A(t). Heyting’s formalization was published in 1930. Glivenko (1929) and Kolmogorov (1925) had already published similar formalizations. where i stands for intuitionistic derivability. In the system of Natural Deduction there are introduction and elimination rules for the logical connectives that reflect their meanings. The ∀I-rule (see below) would have done just as well. (A → (B → C)) → ((A ∨ B → C)). A → (B → A). we will simply write . B → A ∨ B. ∃x(A(x) → B) → (∃xA(x) → B).A → A ∨ B. The rules are given below in an abbreviated notation. A(t) → ∃xA(x).

Finally in ∃E the variable x may not occur in C or in the remaining hypotheses of D. The rules are instructive for more than one reason. For most practical purposes it is however a convenient convention. [Troelstra–van Dalen 1988] p. For an explanation see [Dalen 1997]. Recall that p : A → B stands for “for every a : A (p(a) : B)”. In ∀I the variable x may not occur free in the hypotheses of D. Now → E tells us that if we have a derivation D of A and a derivation 7 . In ∀E and ∃I the term t must be free for x in A. 568).∧I A B A∧B ∧E A∧B A [A] D C C A∧B B [B] D C ∨I A A∨B [A] D B A→B B A∨B ∨E A∨B →I →E A A→B B ⊥ A ⊥E ∀I D A(x) ∀xA(x) ∀E ∀xA(x) A(t) [A(x)] D C C ∃I A(t) ∃xA(x) ∃E ∃xA(x) There are a few conventions for the formulation of the natural deduction system. In the first place. they illustrate the idea of the proof-interpretation. Usually all of the hypotheses are cancelled simultaneously. and it may even be be necessary to allow for ‘selective’ cancellation (cf. (i) A hypothesis of a derivation between square brackets is cancelled. that is to say. 559. This is not really required. (ii) For the quantifier rules there are some natural conditions on the free variables of the rule. it no longer counts as a hypothesis.

g. the meaning of the logical operations can be concluded from their use in introduction and elimination rules.D of A → B. The elimination rule tells us what we may claim on the grounds of evidence for A ∧ B. that is to say. E. So ∧I specifies what one has to require for A ∧ B. then → I yields . This is exactly the justification for the derivation of A → B. the conjunction rules the analogy is even more striking. then the combination D D D A A→B = B B →E is a derivation of B. [A] A D Let D be given. If we know A ∧ B then we also know A and likewise B. we automatically get a proof of B. given D . 8 . B B A→B The first derivation shows that if we add a proof D of A. ∧I tells us that if we know (have evidence for) A and B. Platonistic. or to respond in a certain way to its use by others’([Dummett 1975]). converts D into a derivation D of B. converting proofs of A into proofs of B. Michael Dummett used the same feature of Gentzen’s natural deduction to support his claim that intuitionistic logic fits the requirement of ‘meaning is use’. He remarked that the introduction and elimination rules determine the meaning of the connectives ‘in use’. in general. say. We will see below that the analogy can be made even more explicit in the CurryHoward isomorphism. consist of a capacity to use that statement in a certain way. For. So we have an automatic procedure that. Martin-L¨f has designed in the seventies systems of type theory that o embodies many of the features of Gentzen’s natural deduction. Dummett insists on demonstrative knowledge of mathematical notions. So the rules tell us that there is a particular construction. then we also know A ∧ B. As a consequence the traditonal. The introduction and elimination rules have been chosen so that they are ‘in harmony’. ‘the grasp of the meaning of a mathematical statement must. The evidence required in ∧I is exactly the evidence one can derive in ∧E. The introduction rule also illustrates the transformation character of the implication.

[Troelstra-Schwichtenberg 1996] Ch. combinatorial part. In natural deduction the rules are (more or less) local. and the denotational part. of course. that there is an outer world independent of us. [Girard 1987]. The rejection of the Platonistic notion of truth is indeed an aspect of Dummett’s anti-realism. the ‘left-rule tells us that in order to draw a conclusion from A ∧ B. A ∧ B i Γ1 i 1 . 1. [Dalen 1997] Ch. 2.2. the rules handle the formulas involved in the inference. B Γ1 ∪ Γ2 i 1 ∪ 2 A ∧ B Γ i should be thought of as ‘the conjunction of all propositions in Γ yields the disjunction of the proposition in ’ (where Γ and are finite sets of propositions). For a treatment of the sequent calculus we refer the reader to [Kleene 1952]. [Dummett 1977] Ch. [Kleene 1952] Part II. but the assumptions (hypotheses) largely remain unnoticed (of course. Suggested further reading: [?]. [Girard 1989]. [Troelstra–van Dalen 1988].3. for that matter). After all. Formal logic has two faces. [Dragalin 1988] Part 1. Hence the slogan “a grasp of the meaning of a statement consists in a capacity to recognize a proof of it when one is presented to us”. according to given rules. [Troelstra–van Dalen 1988] Ch. This. Brouwer had always denied the realistic thesis. The most striking is the extra memory the derivation rules carry along. It differs from the natural deduction system in various ways. one has to 9 . It goes without saying that this position is usually softened by an informal meaning concept. is in complete accord with intuitionistic practice. The syntactic part may be viewed as a pure game with symbols. The second derivation system of Gentzen is his sequent calculus. 4. some hypotheses that play a role in the derivation step play a role). the syntactic. We give an example: Γ. The ‘right’-rule tells us what we have to know in order to conclude A ∧ B.notion of truth has to be replaced by something more palpable. 5. A i Γ. Furthermore the introduction-elimination feature is replaced by left-right technique. [Negri–von Plato]. the notion of proof is exactly what will fill the need for communicability and observability. in rules with discharge. it suffices to draw the conclusion from A (or B. A Γ2 i 2 .

G¨del had dispelled the expectation that o intuitionistic logic was the logic of some specific finite truthvalue system. It certainly is too much of a good thing for intuitionistic logic.know why to play this game and not that. [Jaskowski 1936]. a generalization of the Boolean valued interpretation of classical logic. since we already have the intended proof interpretation. from this viewpoint formulas denote certain things. before technical problems can be settled. so the two-valued semantics obliterates the distinction between the two logics: too many propositions become true! One might say that we should not worry about the problem of semantics. One would need more assumptions about ‘constructions’. Jaskowski presented in 1936 a truth table family that characterized intuitionistic propositional logic. in the sense that Γ i A ⇔ Γ |= A. is placed in the combinatorial part of logic. [Dalen 1997]). Proof theory. Ever since Boole it was known that the laws of logic correspond exactly to those of Boolean algebra (think of the powerset of a given set with ∩. By our choice of axioms (or rules) intuitionistic logic is a subsystem of classical logic.g. The interpretations of logic under this two-valued semantics is handled by the well-known truth tables (cf. traditionally. however. The method has a serious drawback: all propositions are supposed to be true or false.1 The topological interpretation In the mid-thirties a number of systematic semantics were introduced that promised to do for intuitionistic logic what the ordinary truth tables did for classical logic. This certainly is the case. The study of interpretations is called semantics. conveniently denoted by 1 and 0. and PEM automatically holds (some might perhaps see this as an advantage rather than a drawback). [Tarski 1938]. [G¨del 1932]. namely ‘true’ and ‘false’. Heyting had already introduced many valued truth tables in his formalization papers. the proof-interpretation is too little specific to yield sharp decisive results. o An elegant interpretation was introduced by Tarski. in fact. 1. to establish non-definability of the connectives. All of the formal semantics discussed below are strongly complete for intuitionistic predicate and propositional logic. Frege already pointed out that propositions denote truth values. where |= is the semantical consequence relation in the particular semantics. e. ∪ 10 . It was. as one would like in model theory.

So the family O(X) plays the role of P(X). A simple calculation shows that i A ⇒ [[A]] = X (we use i . Completeness can now be formulated as usual: i A ↔ A is true. O(R). let [[A]] be given for all atomic A. Consider. Note that this looks very much like the traditional Venn diagrams. Let a domain D be given. where [[⊥]] = ∅. Now we want a similar algebra with the property that U cc = U (for ¬¬A and A are not equivalent). [[¬A]] = Int({0}) = ∅. A is true if A is true under all topological interpretations. Hence i ¬¬A → A. c . It turns out that it is a good choice to let the open sets in a topological space X play the role of arbitrary sets in the power set of X. ∨ and ¬). Similarly i A ∨ ¬A. the largest open subset of K). corresponding to ∧. then [[∃xA(x)]] = {[[A(d)]]|d ∈ D} [[∀xA(x)]] = Int( {[[A(d)]]|d ∈ D}) The topological interpretation is indeed complete for intuitionistic logic. for derivability in intuitionistic. This suggests that the operator cc behaves as a closure operator in topology. The valuation [[·]] : P ROP → O(X) is defined inductively for all propositions. we expect U ccc = U c .e. The remaining laws of logic demand that (U ∩ V )cc = U cc ∩ V cc and (U ∪ V )cc ⊆ U cc ∪ V cc . Let us say that A is true in O(X) if [[A]] = X for all assignments of open sets to atoms. Since X is the largest open set in O(X). for [[A ∨ ¬A]] = A = R. This interpretation can be used to show underivability of propositions. resp. [[¬¬A]] = R. By Brouwer’s theorem (¬¬¬A ↔ ¬A). This is necessary if we want to get open sets all the way. the open sets of real numbers. logic). 11 . Then [[A]] = R −{0}. The theory of topological interpretations is treated in [Rasiowa-Sikorski 1963]. resp. The topological interpretation can be extended to predicate logic. with the extra requirement that negation is interpreted by the interior of the complement. Here is the notation: [[A]] is the open set of X assigned to A. Assign to the atom A the set R −{0}. for example. it is plausible to call A true in O(X) if [[A]] = X. So [[¬¬A → A]] = Int[[¬¬A]]c ∪ A = ∅ ∪ A = A = R. then [[A ∧ B]] [[A ∨ B]] [[A → B]] [[¬A]] = = = = [[A]] ∩ [[B]] [[A]] ∪ [[B]] Int([[A]]c ∪ [[B]]) Int([[A]]c ) Here Int(K) is the interior of the set K (i.and c as operations. Classical logic appears as a special case when we provide a set X with the trivial topology: O(X) = P(X). classical.

e. This process takes place in time. 0. Beth–Kripke models. and at the same time he observes the basic facts that hold for his universe so far. In passing from one moment in time to the next. the laws can be simplified somewhat. A Heyting algebra has binary operations ∧. also called creating subject by Brouwer. So at each moment he may create new elements. is involved in the construction of mathematical objects. →. This idealized mathematician. introduced in [Dalen 1984]. and in the construction of proofs of statements. it is not as flexibled as two later interpretations introduced by E. Beth and Saul Kripke.2 Beth–Kripke semantics Elegant as the topological interpretation may be. and an infimum i∈I ai For a complete Heyting algebra (i. who creates all of mathematics by himself. We will consider here a hybrid semantics. we adopt the standard axioms for a latice and add a ∧ S = {a ∧ s|s ∈ S} 1. Both semantics have excellent heuristics. a unary operation ¬ and two constants. much like Boolean algebra’s. That is.The algebra of open subsets of a topological space is a special case of a Heyting algebra. The basic idea is to mimick the mental activity of Brouwer’s individual. he is free how 12 . an algebra in which all sups and infs exist). ∨. The laws are listed below: a∧b a∨b a ∧ (b ∧ c) a ∨ (b ∨ c) a ∧ (b ∨ c) a ∨ (b ∧ c) 1∧a 1∨a = = = = = = = = b∧a b∨a (a ∧ b) ∧ c (a ∨ b) ∨ c (a ∧ b) ∨ (a ∧ c) (a ∨ b) ∧ (a ∨ c) a 1 0∧a 0∨a a→a a ∧ (a → b) b ∧ (a → b) a → (b ∧ c) ¬a = = = = = = = 0 a 1 a∧b b (a → b) ∧ (a → c) a→0 Since it is necessary for the interpretation of predicate logic to allow infima and supprema of collections of elements. 1.W. we often consider complete Heyting algebra’s. by axioms for the various operations. algebra’s with the property that for a collection ai (the unique least element {ai |i ∈ I} of elements there is a supprenum i∈I majorizing all ai ). Heyting algebra’s are defined.

establish A eventually. The stages for the individual form a partially ordered set K. This is an important feature in intuitionism (and in constructive mathematics. Let us just consider. Let us write “k A” for “A holds at k”. Note that the individual does not (have to) observe composite states of affairs. so k ≤ ⇒ D(k) ⊆ D( ). This is particularly clear for ∀. the first-order case. He does not necessarily establish an atomic fact A ‘at the spot’. It is reasonable to assume that no elements are destroyed later. We view k ≤ as “k is before or coincides with . For a node k in the poset K a bar B is a subset with the property that very path through k intersects B. For atomic A k A is already given. for the moment. ⊥ is never forced. in general). The dynamic character of the universe demands that the future is taken into account. If we claim that “all dogs are friendly”. A path in the poset K is a maximal ordered subset. that is to say we consider elements of one and the same sort. but he will. At each moment there is a number of possible next stages. furthermore there is a finite number of relations and functions (as in a standard first-order language).to continue his activity. The technical terminology is “k forces A”. This means that there is a bar B for k such that for all nodes in B the statement A holds at . no matter how he pursues his research. The next step is to interprete the connectives. ≤ . denoted by D(k). so the picture of his possible activity looks like a partially ordered set (even like a tree). k k k k A∧B A∨B A→B ¬A k A and k B ∃B∀ ∈ B A or B ∀ ≥ k( A⇒ B) k A →⊥ ∀ ≥ k( A⇒ ⊥ ∀ ≥ k( A) ∃B∀ ∈ B∃a ∈ D( ) A(a) ∀ ≥ k∀a ∈ D( ) A(a) k k ∃xA(x) ∀xA(x) Observe that the ‘truth’ at a node k essentially depends on the future. For each k ∈ K there is a local domain of elements created so far. then one unfriendly dog in the future may destroy the claim. Now we will stipulate how the individual arrives at the atomic facts. 13 . These stages have become known as possible worlds.

There is an extensive modeltheory for Kripke semantics.and Beth. ¬B. We say that “A is true in a Beth–Kripke model K” if for all k ∈ K k A. Kripke. i. We will discuss them first. so k0 A ∨ ¬A. but k0 A. But this contradicts k1 A(1). A. so ¬(A ∧ B) → (¬A ∨ ¬B) (De Morgan’s law fails).e. we see that k0 ¬¬A. If k0 ∃x¬A(x) then k0 ¬A(1). k1 above remark.g. Clearly k1 ∀xA(x). “A is true” if A is true in all Beth–Kripke models. k1 A(0). forcing is monotone. By labelling the set with propositions. So k0 ¬∀xA(x) → ∃x¬A(x). And if all the local domains D(k) are identical.semantics are strongly complete for intuitionistic logic: Γ i A ⇔ Γ A (special case: i A ⇔ A is true). A is not forced after k. k2 > k0 . k1 Hence k0 A and k2 A ∧ B. There are a number of simple propertiwes that can easily be shown. 14 . k ¬A A iff for some ≥ k So k ¬¬A if for each ≥ k there is an m ≥ with m A. it is easy to see what the logical properties are. (b) k1 > k0 . We see that k0 ¬A. e. k A.. (c) k1 > k0 . so k0 ¬∀xA(x). We give some examples. In practice. ‘Semantical consequence’ is defined as Γ A iff for all Beth–Kripke models K and all k ∈ K k C for all C ∈ Γ → k A Beth–Kripke. For example. we can indicate them by means of diagrams. Since Kripke models are given by partially ordered sets. k2 B. k0 A(0). Kripke models are the more flexible tools for metamathematical purposes. An individual model over a poset K is denoted by K. Hence k0 ∃x¬A(x). Furthermore k0 ¬A. Note that k ¬A if ∀ ≥ k( A). In particular there are a large number of results on the structure of the partially ordered sets.A Beth–Kripke model with the property that the bar B in all the defining clauses is precisely the node k itself.: A.e. k0 A ∧ B. we have a Beth model. By the (a) Consider an atomic A and let k1 > k0 . k1 k0 ¬A ∨ ¬B. Hence k0 ¬¬A → A. k1 A(1). is called a Kripke model. A large number of applications have been given since their introduction. so k0 ¬(A ∧ B). k ≤ ⇒ in Beth–Kripke and Beth models k A ⇔ ∃B∀ ∈ B A. i. ipredicate logic is complete for Kripke models over trees.

DP and EP are often considered the hallmark of constructive logic.e. The disjunction and existence property hold for a number of prominent theories. In this new node no proposition is forced. then there is a tree model K1 where A is not forced. But that contradicts the fact that A and B are not forced in K1 and K2 . There is a corresponding theorem for existential sentences. There is a snag in the conclusions drawn above. see 15 . however.A r k1 rA k e1 e e B r ¡ k2 ¡ ¡ e ¡ ¡ er  A(1) q q 0 1 k1   q k 0 (c) 0  (a) k0 r (b) k0 and for propositional logic this can even be strengthened to finite trees (the so-called finite model property: A ⇒ A is false in a Kripke model over a finite tree). The two models are glued together as follows: put the two models side by side and place a new node k below both. Similarly for existence: if you have shown the existence of something. the existence property (EP ) i ∃xA(x) ⇒ i A(t) for a closed term t These theorems lend a pleasing support to the intuitionistic intended meaning of ‘existence’: if you have established a disjunction. The result is a correct Kripke model. similarly a tree model K2 where B is not forced. Since i A ∨ B. and hence k A or k B. you can indeed point to a specific instance. The completeness over tree models can be used to prove the disjunction property (DP ) i A ∨ B ⇒ i A or i B The proof uses reductio ad absurdum. the proof made use of reductio ad absurdum. What one would like is a method that extracts from a proof of ∃xA(x) a proof of A(d) for some d. we have k A ∨ B. As the reader has seen. Fortunately there are proof theoretical devices that provide exactly this kind of information. i. the most important being arithmetic. not overestimate the significance of a technical result like this. you know that you can established one of the disjuncts. Suppose i A and i B. Therefore i A or i B. the result has not been established constructively. one should.

Veldman was in 1976 the first to consider a modified Kripke semantics. the logic is classical. i.e. This is based on algorithms. We have seen a number of interpretations of intuitionistic theories. the so-called intuitionistic analysis. In particular. and on the face of it. They allow for a very natural interpretations of choice sequences. a totally different interpretation of intuitionistic logic (and arithmetic) in Kleene’s realizability interpretation. Friedman. Since then De Swart. One also sees that {k|k A} is an open set. Beth semantics also turn out to be a convenient tool in completeness proofs. so in each node A ∨ ¬A holds. for example. For in each leaf (maximal node) A or ¬A holds. i. but by no means all of them. [Dummett 1977]. Finite Beth models are not interesting for intuitionistic logic. Beth models happen to be better adapted to second-order arithmentic with function variables. Kripke showed in 1963 that Kripke and Beth models can be converted into each other [Kripke 1965]. The obvious 16 . i. 1. . One might wonder if these semantics are totally unrelated. he considered constant domains.e. Smorynski has shown that in a fairly large number of cases ‘semantic’ proofs can be made constructive. [Veldman 1976]. For more information on Kripke model theory see [Dalen 1997].e. 13.28. For convenience we consider Beth models over tree. for which he could give an intuitionistically correct completeness proof. or whether there is some common ground. Now it is a matter of routine to show that the function [[. [Dalen 1984]. Beth’s original models are slightly more special. The use of Beth models for second-order systems can be found in [Dalen 1978]. The combinatorics of Beth models is just more complicated than that of Kripke models. it is useful for rendering completeness proofs in an intuitionistic metamathematics. It is easily seen that these sets form a topology. Dummett. one could not find a similarity with the above semantics. [Fitting 1969]. see [Troelstra–van Dalen 1988]. This is a certain drawback when compared to Kripke models. [Troelstra–van Dalen 1988]. let us therefore put [[A]] = {k|k A}.3 Super Semantics. Dk = D for all k.1. u Beth models are easily convertible into topological models.]] is a valuation as defined in section 2. Troelstra have given alternative versions for Beth semantics. On a tree one can define a topology as follows: open sets are those subsets U of the tree that are closed under successors. k ∈ U and k ≤ ⇒ ∈ U . There is. Ch. see also [Sch¨tte 1968]. [Smorynski 1982].

g. We will interprete the intuitionistic theory of reals. Subsequently Joan Moschovakis and Van Dalen used the same technique for interpreting the theory of choice sequences over natural numbers. (a · b)(t) = a(t) · b(t) The definition of the inverse is. According to the above definition we have [[a a]] = R. The objects are partial continuous real-valued functions with an open domain. say Kripke models and realizability. In addition to the notion of equality there is a convenient notion of equivalence: “a and b coincide where they exist”. and in general [[a = b]] ⊆ E(a) ∩ E(b). who saw the possibilities for treating logical notions in a categorical setting. Scott 1970]. [Dalen 1974]. 17 . however. in symbols a b := Ea ∨ Eb → a = b. One can define it pointwise for values distinct from 0. e.common ground is the logic that they are modelling. Consider the topology of the real line. R. The open sets will be our truth values. All objects are partial. This domain ‘measures’ the existence of the object: Ea ∈ O(R). Fortunately there is a general kind of semantics based on category theory. DRAW IN G The interpretation forces us to consider the notion of partial object and the notions of ‘equivalence’ and ‘strict equality’. and we say that a is total if Ea = R. discovered the significance of adjointness for logic. In order to avoid complications we will look at Scott’s original model. Historically these newer interpretations grew out of existing semantics. The operation on the partial elements are defined pointwise: (a + b)(t) = a(t) + b(t). in particular. He. Dana Scott showed in his extension of the topological interpretation to second-order systems. The proper formulation of this kind of semantics is given in the so-called sheaf-semantics. [Moschovakis 1973]. [Scott 1968. Equality has to be reconsidered in this light: [[a = b]] = Int{t ∈ R|a(t) = b(t)} DRAW IN G Observe that [[a = a]] = Ea. and second order arithmetic. HAS. but then there seem to be problems with the invertibility of non-zero elements. how one could capture strikingly intuitionistic features. The pioneer in this area was Lawvere. but that would not be sufficient to link. adapted to the sheaf semantics. problematic.

and partial equality. ‘=’ is then regained by defining t = s := Et ∧ Es ∧ t s. So let us interprete # in the model: [[a#b]] = {t|a(t) = b(t)}. 18 . of course. There is however no b such that [[a. those have to be revised.DRAW IN G The fact is that ‘non-zero’ is not good enough to have an inverse. This is where apartness comes in. carries obligations for our quantifiers. we take ‘ as a primitive. So [[a = 0]] = [[¬a = 0]] = Int[[a = 0]]c = R This a is non-zero in the model. Let a(t) = t. then [[a#0]] = R − {0} = c[[a · c = 1]] ⊆ [[a · b = 1]] = R − {0}.b = 1]] = R. The logical axioms. That is. We add the axioms: x y ∧ A(x) → A(y) ∀z(x z ↔ y z) → x y. In the model we now get an inverse for a. It suffices to consider the quantifier rules and the equality rules: [Ex] D A(x) ∀xA(x) ∀xA(x) Et A(t) [A(x)][Ex] D C C A(t)Et ∃xA(x) ∃xA(x) The axiomatization of the equality aspects is simple if we use the equivalence relation instead of identity itself. The introduction of an existence predicate. then [[a = 0]] = Int{t|a(t) = t = 0} = Int{0} = ∅. For the details of the logic can be found in [Troelstra–van Dalen 1988] and Scott’s original [Scott 1979]. The condition for an inverse now becomes a#0 → ∃b(a · b = 1). In particular ∃xA(x) ↔ ∃x(Ex ∧ A(x)) and ∀xA(x) ↔ ∀x(Ex → A(x)). or rules. Take b(t) = t−1 for t = 0. It is well-known that a real number has an inverse iff it is apart from 0.

Hyland showed that the realizability interpretation fitted into his effective topos. [Rasiowa-Sikorski 1963]. but it is an observed fact that large parts of mathematics lend themselves to formu19 . Grayson used the same approach in his version of the permutation model of Krol. 13. 13 §8. [Smorynski 1973a]. [Lambek–Scott 1986]. [Troelstra–van Dalen 1988]. [Moschovakis 1973]. Hyland and Grayson. Her objects were continuous mappings from Baire space to Baire space (a model with total elements). Since Kripke models and Beth models are built over trees. E. in which he could show that Brouwer’s continuity theorem holds. Ch. [Dummett 1977] Ch. The next generalisation after sheaf models is that of categorical model. as a result they can be viewed as sheaf models.g. [Grayson 1979. Grayson 1982. [Dalen 1978]. they are natural examples of a topos. [Fourman-Scott 1979]. [Hyland 1982]. 6. Categorical models also turned out to be important for typed lambda calculus.g. Suggested further reading: [Dragalin 1988] Part3. [Fitting 1969]. they carry a natural topology. Sheaf models were further presented by Scott. Scott extended his model to a higher-order theory of reals. [Troelstra–van Dalen 1988] Ch. [Friedman 1977].15. but its formulations in a Beth model allows for some extra fine tuning so that techniques from set theory can be applied. a model for lawless sequences is constructed by means of forcing. 5. It can be translated straightforwardly into a topological model. Sheaf models allow an interpretation of higher-order logic. [Fourman-Hyland 1979]. The model also interprets the theory of the creating subject. [MacLane–Moerdijk 1992]. There is no special foundational bias towards first-order theories. Grayson 1984]. In that sense topos models can be viewed as intuitionistic universes. Ch. Certain categories are powerful enough to interprete higher-order intuitionistic logic. [Hoeven-Moerdijk 1984]. [McLarty 1995]. see [Scott 1970].13. 2. Fourman. Joan Moschovakis applied the methods of Scott’s semantics to intuitionistic analysis in the style of Kleene’s FIM. Moerdijk and Van der Hoeven used a special sheaf model to interprete the theory of lawless sequences. §5. Van Dalen’s model for analysis is based on Beth’s original semantics. [Fourman-Scott 1979]. [Gabbay 1982]. E. [Grayson 1981]. The reader is referred to the literature for details It should be mentioned that the topos semantics managed to capture most of the known semantics.One may add also the ordering to the model of R: [[a < b]] := {t|a(t) < b(t)}. Grayson 1981.

We have demonstrated the usefulness in section 3. A well-known instance is the constant domain theory of S. We will look in this section at a number of first-order theories and point out some salient properties. not every formula has a prenex normal form in IQC). The theory of equality has the usual axioms – reflexivity. although a number of intuitionistic first-order theories share their axioms with their classical counterparts. just like its classical counterpart. [Kreisel 1958]. Fragments are. symmetry. notably in the absence of PEM.g. 20 . transitivity.lation in a first-order language. The symbol for apartness is #. in the case of the disjunction property. An extensive treatment of first-order theories is given in papers of Gabbay and the dissertation of Smorynski. where x = y abbreviates ¬x = y. see [Gabbay 1982]. First we point out that. Following Smorynski we will apply it to a few theories. We consider a few basic theories below. [Goernemann 1971]. however. the theory of stable equality is given by EQst = EQ + ∀xy(¬¬x = y → x = y) And the decidable theory of equality is axiomatized by EQdec = EQ + ∀xy(x = y ∨ x = y). but undecidable in intuitionistic logic. [Kripke 1968]. cf. The key method in the proof was the joining of a number of disjoint Kripke models by placing one extra node (the root of the tree) below the models. for example. G¨rneman (Koppelberg). This technique has become known as gluing. In general Kripke semantics is a powerful tool for meta-mathematical purposes. in general theories are sensitive to the formulation of the axioms. monadic predicate logic is decidable in classical. on the other hand. One can strengthen the theory in many ways. undecidable. Many examples can be found in the cited works of Smorynski and Gabbay. [Smorynski 1973b]. decidable. [Lifschitz 1967]. [Maslov 1965] Lifschitz showed that the theory of equality is undecidable. E. [Smorynski 1973a]. In propositional logic one may impose restrictions on the underlying trees of Kripke models. She showed that predicate logic plus o the axiom ∀x(A(x) ∨ B) → ∀xA(x) ∨ B is complete for Kripke models with constante domains. In intuitionistic mathematics there is a strong notion of inequality: apartness. in predicate logic one may also put restrictions on the domains of the models. Introduced by Brouwer in [Brouwer 1919a] and axiomatized by Heyting in [Heyting 1925]. Predicate logic itself is. the class of prenex formulae is decidable (and as a corollary.

Let AP A ∨ B and assume AP A and AP B. No non-trivial atoms are forced in k0 (i.e. we designate points a and b in K1 and K2 which are identified with the point 0. Hence k0 A ∨ B and so k0 A or k0 B. say {ci |i ∈ I}. there are models K1 and K2 of AP such that A and K2 B K1 # # "!           #   d   K1 "! d d d d d d k0 "! We consider the disjoint union of K1 and K2 and place the one-point world below it. For EP it is convenient to assume that the theory has a number of constants. Then for each i there is a model Ki with Ki A(ci ). Now let AP ∃xA(x) and AP A(ci ) for all i. The identification of the cj with elements in the models Ki is obvious. we glue the models Ki by means of a bottom world K∗ with a domain consisting of just the elements ci . Hence AP A or AP B.The axioms of AP are given by EQ and the following list   ∀xyx y (x#y ∧ x = x ∧ y = y → x #y )    ∀xy(x#y → y#x)  ∀xy(¬x#y ↔ x = y)    ∀xyz(x#y → x#z ∨ y#z) We will now use the gluing technique to show that AP has the disjunction and existence property. That is to say. Both are impossible on the grounds of the choice of K1 and K2 . it is easy to check that 21 . only the trivial identities ci = ci ). Again. by the strong completeness theorem. Contradition. Then. As we did above. The new model obviously satisfies the axiom of AP.

In fact the equality fragment is axiomatized by an infinite set of quasistability axioms. The Sn axiomatize the equality fragment of AP. [Dalen 1997]). [Gabbay 1982]. This shows that even a relatively simple theory like equality is incomparably richer than the classical theory. i.g. [Gabbay 1972]. The gluing operation demonstrates that there are interesting operations in Kripke model theory that make no sense in traditional model theory. for arithmetic we can prove ∀xy(x = y ∨ x = y). and subsequently an elegant model theoretic proof was given in [Smorynski 1973c]. it has to be to right of a or to the left of b. For. ¬x#y ↔ x = y. so x = y ↔ ¬¬¬x#y ↔ ¬¬x = y. k0 A(ci ) for some i. [Gabbay 1973]. There are cases where a decidable equality results in the fact that the theory is classical. cf. In particular we get a stable equality: ¬¬x = y → x = y. e. It is important to note that the atoms. There are quite a number of surprising results on first-order theories. The theory LO of linear order has axioms: ∀xyz(x < y ∧ y < z → x < z) ∀xyz(x < y → z < y ∨ x < z) ∀xyz(x = y ↔ ¬x < y ∧ ¬y < z) The second axiom is interesting because it tells. The apartness axioms have consequences for the equality relations.the new model satisfies AP.e. or in general the quantifier free part of a theory does not yet determine whether it is classical or not. the theory of algebraically closed fields. but the theory is not classical. But then also Ki A(ci ). Apartness and linear order are closely connected. contradiction. The relation with apartness is given by LO + AP x < y ∨ y < x ↔ x#y One can also use x < y ∨ y < x to define apartness. The result was first established by proof theoretic means in [Dalen 1978]. 22 . in the sense that if we pick an arbitrary third point. that a < b means that a is “far” to the left of b. To be precise: AP is conservative over EQ + {Sn |n ≥ 0}. In a way this gives the best possible result since LO+AP is conservative over LO ([Smorynski 1973c]. Hence k0 ∃xA(x). Put x =0 y := ¬x = y x =n+1 y := ∀z(z =n x ∨ z =x y) For these “approximations to apartness” we formulate quasi-stability axioms: Sn := ∀xy(¬x =n y → x = y). Hence AP A(ci ) for some i. so to speak. [Smorynski 1973a]. on the other hand.

The gluing technique also works for intuitionistic arithmetic. one may add principles that have a certain constructive motivation. Such a machine is an abstract computing device that operates on a potentyially infinite tape. Reading ‘the machine halts at time x’ for A(x). One of such principles is. In fact PA is an unbounded extension of HA (Smorynski). [Friedman 1975]. see also [Dragalin 1988]. does the machine. HA. eventually halt (and hence produce an output)? Suppose now that someone tells us that it is impossible that the machine never halts. In this case one has to do some extra work to check the induction axiom. Intuitionistic arithmetic is. is axiomatized by exactly the same axioms as Peano’s arithmetic. There are a number of interesting extensions of intuitionistic arithmetic. even more incomplete than classical arithmetic because it is a subsystem of PA. to the surprise of the insiders. In fact. called after Heyting. Intuitionistic arithmetic. The decision procedure for the halting in this case consists of turning on the machine and waiting for it to halt. The disjunction property is an obvious consequence of the existence property. The result is that for HA we have the existence property for numerals: HA ∃xA(x) ⇒ HA A(n) for some n. The difference is the underlying logic: PA = HA + P EM. of course. of course. An intuitionist would not buy the argument. as ∨ is definable in terms of ∃: HA (A ∨ B) ↔ ∃x((x = 0 → A) ∧ (x = 0 → B)) It would seem rather unlikely that the existence property is a consequence of the disjunction property. when presented with an input on the tape. [Markov 1971]. The best known first-order theory is. that in the case of HA (and a number of related systems) this is indeed the case. The key question for Turing machines is.[Smorynski 1978]. the above formulation exactly covers Markov’s argument. Heyting 1930a]. HA. 23 . do we know that it indeed halts? Markov argued ‘yes’. Markov’s Principle: MP ∀x(A(x) ∨ ¬A(x)) ∧ ¬¬∃xA(x) → ∃xA(x) The principle is a generalization of the original formulation of Markov. in Markov’s case A(x) is primitive recursive. When somebody claims that a Turing machine will stop. the natural question is ‘when?’ We want an actual bound on the computation time. Harvey Friedman proved however. [Heyting 1930b. Markov considered the halting of a Turing machine. arithmetic.

397. Here is the formal definition: 24 .129. and hence would yield a conflict with the independence of the G¨del sentence from PA. Markov’s rule. o clearly. p. Given a statement of the form A → B. p. 507. Note that A(x) may contain more free variables. The letter is impossible. In the early days of intuitionism people wondered to what extent HA was safer that PA. Since. k0 ∀x(A(x) ∨ ¬A(x)). p. For example: Markov’s rule M R with respect to HA says: HA ∀x(A(x) ∨ ¬A(x) ∧ ¬¬∃xA(x) ⇒ HA ∃A(x) We say that HA is closed under Markov’s rule. Consider a model with two nodes k0 . Markov’s principle may be unprovable in HA.e. has a stronger position. cf. it is no surprise that a great deal of research has been devoted to the subject. hence HA A(n). the proper sentence of the form ∃xA(x). Now HA A(n) ∨ ¬A(n) and by DP HA A(n) or HA ¬A(n). and a fortiori HA ∃xA(x) ([Smorynski 1973a]. k1 . i. from this extra evidence we may hope to draw a stronger conclusion. p. This was settled in 1933 by G¨del (and independently by o Gentzen). 366. [Beeson 1976]. [Troelstra–van Dalen 1988]. But k0 ∃xA(x) would ask for an instance A(n) to be true in the standard model. the proof of closure under Markov’s rule is particularly simple: if HA ¬¬∃xA(x). For closed ∃xA(x). but its companion. where k0 < k. Γ A ⇒ Γ B. then PA ∃xA(x). So k1 ∃xA(x) and hence k0 ¬¬∃xA(x). which from the intuitionistic viewpoint weakened statements. This was basically done by a judicious distribution of negations. The heuristic argument is that we have more information in this case. In the bottom node we put the standard model of natural numbers.A simple Kripke model shows that HA M P . [Beeson 1985]. who showed that PA can be translated into HA. Since the theory of natural numbers is at the very heart of mathematics. in the top node we put a (classical) non-standard model M in which the negation of G¨del’s sentence “I am not provable in o PA”. which is in general weaker than Γ A → B. G¨del defined o a translation. and hence A(n) is true in the standard model for some n. we actually have a proof of ¬¬∃xA(x). Markov’s rule is indeed correct. is true. the model refutes M P . one may formulate a corresponding rule.

no deep philosophical insight can be expected here. y) = 0). It is a routine exercise to show that Γ i A ⇒ ΓF i AF A i AF HA i A ⇒ HA i AF . p. y) = 0 (where we assume that HA has defining equations for the primitive recursive functions). of course. y) = 0). In other words. he added o a double negation in front of every subformula (cf. Kolmogorov had already in 1925 proposed a similar translation (unknown to G¨del in 1933). 2 PA c ∀x∃yA(x. We apply the Friedman translation with respect to F := ∃x(t(x. but the translation and the proof of the corresponding theorem immediately extends to full predicate logic. also works for predicate logic: o Γ c A ⇔ Γ ◦ i A◦ . First we note that a quantifier free formula A(x. Next we introduce the Friedman translation: for a given formula F we obtain AF from A by replacing all atoms P by P ∨ F . 164). The result of the G¨del translation may be improved for certain simple o formulas. 25 . Now consider a term t and let HA ¬¬∃x(t(x. The G¨del translation. p. 59). Kreisel showed that PA and HA prove the same Π0 sentences. where A(x. y) ⇔ HA i ∀x∃yA(x. Kolmogorov had in fact axiomatized a fragment of intuitionistic predicate logic. It is an easy consequence of the G¨del translation theorem that the o universal fragment of PA is conservative over HA. y) is quantifier free.A◦ (A ∧ B)◦ (A ∨ B)◦ (A → B)◦ (∀xA(x))◦ (∃xA(x))◦ = = = = = = ¬¬A for atomic A A◦ ∧ B ◦ ¬(¬A◦ ∧ ¬B ◦ ) A◦ → B ◦ ∀xA◦ (x) ¬∀x¬A◦ (x) For this G¨del translation we get o PA c A ⇔ HA i A◦ . y). [Troelstra–van Dalen 1988]. y) is equivalent in HA to an equation t(x. where Γ◦ = {B ◦ |B ∈ Γ} (cf. Hence PA c 0 = 1 ⇔ HA i ¬¬0 = 1 ⇔ HA i 0 = 1 So PA is consistent if and only if HA is so. The proof has some quite interesting features. [Dalen 1997].

One can see that it strengthens the intended meaning as follows: in the left hand formula. which formalizes the notion y is a (halting) computation on input x (strictly speaking ‘y is the code of a computation . [?]. [Troelstra 1973]): the independence of o premiss principle. Let A be a statement. Whereas in the right hand formula. where T is Kleene’s T -predicate. x. HA is closed under the Independence of Premiss Rule: HA (¬A → ∃xB(x)) ⇒ HA ∃x(¬A → B(x)) (cf. y). We know from (formalized) recursion theory that a function f is provably recursive in HA with index e if HA ∀x∃yT (e. y). y) = 0) Observe that we now have closure under Markov’s rule with numerical parameters. so this proof may enter into the computation of x. when some non-logical axioms are by themselves strong enough to imply PEM. . y) ⇔ (G¨del translation) o HA ¬¬∃yT (e. ’). IP (¬A → ∃xB(x)) → ∃x(¬A → B(x)) IP is not provable in HA. Here is an example: the full axiom of choice implies PEM. However.y)=0 = ((∃x(t(x. 1. x. y) ⇔ (closure under Markov’s Rule) HA ∃yT (e. Hence HA i ∃x(t(x. 296). y) = 0)) → (⊥ ∨∃x(t(x. . ([Friedman 1977]) Finally we mention one more principle that has a metamathematical use in G¨del’s Dialectica translation (cf. This formula is equivalent to ∃x(t(x. y) ⇔ PA ∃yT (e. x. 26 . IP . one has to be careful however.((∃x(t(x. x must be computed straightaway. define P = {n ∈ N|n = 0 ∨ (n = 1 ∧ A)} Q = {n ∈ N|n = 1 ∨ (n = 0 ∧ A)}. y) = 0). n) = 0))). In some cases it suffices to omit PEM from the logical axioms. Friedman showed that this fact also holds in intuitionistic set theory.4 Set Theories Following the tradition in set theory. the various intuitionistic modifications of ZF are also first-order. y) = 0) →⊥) →⊥))∃x(t(x. For most classical theories one can consider one or more corresponding intuitionistic theories. It is just one step now to get to Kreisel’s theorem. y) ⇔ HA ∀x∃yT (e. x. IZF. p. y) = 0)) → (⊥ ∨∃x(t(x. Now PA ∀x∃yT (e. x. x. y) = 0) ∨ ∃x(t(x. the existence of x is concluded on the basis of a proof of ¬A.

624. A∧B A We first introduced A ∧ B and subsequently eliminated it. the proof given here is Goodman and Myhill’s ([Goodman-Myhill 1978]). p. [Troelstra–van Dalen 1988]. The above fact was discovered by Diaconescu ([Diaconescu 1975]). CZF. The fundamental theorem of the natural deduction system is the socalled normalization theorem. so F (P ) = F (Q) ∨ F (P ) = F (Q). Gentzen on the other hand considered proofs as an object for study. Y }.e. [Dummett 1977]. [Beeson 1985]. [McCarty 1986]. [Gabbay 1982]. in Kleene’s realizability universe sets with apartness are subcountable (i. §1). In his system the structure of derivations is the object of investigation. Here is one: . For a formulation see ([Beeson 1985]. p. [McCarty 1984]. Gentzen noted that there are obvious superD D’ A B fluous steps in proofs that should be avoided. then A holds. Harvey Friedman studied IZF. E. The derivation 27 .g. Peter Aczel considered another version of constructive set theory. Obviously ∀X ∈ S∃x ∈ N(x ∈ X) AC would yield a choice function F such that ∀X ∈ S(F (x) ∈ X). The theory is very sensitive to the formulation. cf. an intuitionistic version of ZF. F (Q) ∈ N. e. Friedman has shown that the G¨del translation theorem works for a o suitable formulation of set theory ([Friedman 1973]. If F (P ) = F (Q). Hilbert designed his proof theory for the purpose of consistency proofs. Suggested further reading: [Kleene 1952]. the range of a function on N). He built a model of cumulative constructive set theory in which a number of interesting phenomena can be observed. on the other hand F (P ) = F (Q). The axioms for sets are modifications of the traditional ZF ones. For more information on the topic the reader is referred to the literature. (see [Beeson 1985]). [McCarty 1991]. ∈-induction is used rather than Foundation. Observe that F (P ). Ch. If. [Troelstra–van Dalen 1988].S = {X. o Set theory under the assumption of Church’s Thesis was extensively studied by David McCarty.g. This set theory has the attractive feature that it is interpretable in a particular type theory of Martin-L¨f. So we get A ∨ ¬A. and similarly the Axiom of Collection takes the place of the replacement axiom. a wrong choice of axiom may introduce unwanted logical principles. then ¬A. 174). 8.

i. The theorem has a strong and a weak form. A Introductions. immediately followed by eliminations are called cuts. (iii) The existence property for IQC. [Prawitz 1970]. it is perhaps not surprising that a formal correspondence between a term calculus and natural deduction can be established.e. From the normal form theorem many corrollaries can be drawn. [Troelstra–van Dalen 1988]. (iv) i A is decidable for prenex formulas A. (ii) The disjunction property for IQC. 28 . A cut free derivation is also called a normal derivation. this result shows that in IQC not every formula is equivalent to a prenex formula. Since full predicate logic is not decidable (Church’s theorem for intuitionistic predicate calculus). If ∃ and ∨ do not occur positively in any formula in Γ and Γ i ∃xA(x). the Γ i A or Γ i B. The normalization theorem says that each derivation can be converted by cut elimination steps to a normal derivation. and the process of eliminating cuts is called normalization. Gentzen showed how to eliminate cuts from natural deduction derivations (in intuitionistic or classical prediate logic). [Girard 1987]. [Girard 1989]. The weak form states that a derivation can be converted into a normal form. there is a decision method for i A. For example: (i) Intuitionistic proposition calculus is decidable. There is a structure theorem for normal derivations that tells us that in normal proofs one can run in a systematic way through derivation (starting at the top land going from major assumption to conclusion) so that all the elimination steps precede the introduction steps (with possible ⊥-steps in between).D would have accomplished the same. [Prawitz 1965]. Since natural deduction is so close in nature to the proof interpretation. [Prawitz 1977]. The revival of Gentzen’s ideas is due to Prawitz’s beautiful work. [Dalen 1997]. The strong form says that each sequence of conversions will lead to a normal form (and actually one and the same normal form). If ∨ does not occur positively in any formula in Γ i A ∨ B. then Γ i A(t) for some closed t. For details about normalization see [Gentzen 1935]. [Troelstra-Schwichtenberg 1996].

x:B D t:A D reduces to λx · t : B → A s:B (λx · t)(s) : A D s:B D[∫ /§] t[s/x]A The proof theoretic conversion corresponds to the β-reduction of the* λ-calculus. Note that the informal argument of page 3 is faithfully reflected. Here is a list: 29 . In order to deal with full predicate logic we have to introduce specific operations in order to render the meaning of the connectives and their derivation rules. On discharGing the hypotheses. we introduce a λx in front of the (given) term t for B.We will first demonstrate this for a small fragment. it has a hypothetical proof term. say x. [A] B→A A → (B → A) [x : A] λy · x : B → A λx · λy · x : A → (B → A) Thus the proof term of A → (B → A) is λxy. By binding x.x. this is Curry combinator K. containing only the connective ‘→’. Let us consider a particular derivation. Since A is an assumption. The elimination runs as follows: A→B B A t:A→B s:A t(s) : B Observe the analogy to the proof interpretation. Let us now consider a cut elimination conversion. Consider an → introduction: [A] D B A→B [x : A] D t:B λx · t : A → B We assign in a systematic way λ-term to formulas in the derivation. the proof term for A → B no longer depends on the hypothetical proof x of A.

which introduces a impredicativity. So A has its own ex falso operator ⊥A . t1 [u. t1 ) : A0 ∧ A1 t : Ai (i ∈ {0. v the variables u and v are bound. We do this by assigning to each hypothesis its own variable.v (t. v]) : C ∨I ∨E →I λy A ∀I →E t[x] : A(x) λy · t[y] : ∀yA(y) t1 : A(t0 ) p(t0 . z A(y) ] : C Eu. 1}) pi (t) : Ai t : A ∨ B t0 [xA ] : C t1 [xB ] : C Du. of course. t1 ) : ∃xA(x) ∀E ∃I ∃E There are a number of details that we have to mention. (iii) In the falsum rule the result. E. Now the conversion rules for the derivation automatically suggest the conversion for the term.p −− pairing p0 . (ii) In ∨E (and similarly ∃E) the dependency on the particular (auxilliary) hypotheses A and B disappears. t0 [u].g.v (t. The proof of a normalization theorem was a spectacular 30 . A0 ∨ A1 t[xA ] : B · t[y A ] : A → B ∧E t : A0 ∧ A1 (i ∈ {0. This is done by a variable binding technique. Normalization and cut-elimination for second-order logic is far more complicated because of the presence of the comprehension rule. xA : A. 1}) ki . t1 [v]) : C t:A→B t :A t(t ) : B t : ∀xA(x) t(t ) : A(t ) t : ∃xA(x) t1 [y. depends on the conclusion A. In Du . (i) In → I the dependency on the hypothesis has to be made explicit in the term. p1 −− projections D −− discriminator (“case dependency”) k −− case obliteration E – witness extractor ⊥ – ex falso operator ∧I t 0 : A0 t 1 : A1 p(t0 .

This suggests a correspondence between proofs and propositions on the one hand and elements (given by the terms) and types (the spaces where these terms are to be found). [Gallier 1995].breakthrough. The reader is referred to the literature. This correspondence was first observed for a simple case (the implication fragment) by Haskell Curry. We have seen in the previous section that the term calculus corresponds with the natural deduction system. Let us denote ‘a is in type A’ by a ∈ A. XV. So proofs are naturally typed objects. ch. the one considered by Curry.g. E. the names to mention here are Girard. Since the meaning of proposition is expressed in terms of possible proofs — we know the meaning of A if we know what things qualify as proofs — one may take an abstract view and consider a proposition as its collection of proofs. If A and B are typed sets then the set of all mappings from A to B is of a higher type. [Troelstra 1973]. From this viewpoint there is a striking analogy between propositions and sets. For the quantifiers generalizations are available. A set has elements. 9. one can construct higher types by iterating this ‘function space’-operation. cf. [Howard 1980]. [Howard 1980]. o Suggested further reading: [Troelstra-Schwichtenberg 1996]. Let us first look at a simple case. if we have a proof p : A → B and a proof q : A then p(q) : B. q : A ⇒ p(q) : B x : A ⇒ t(x) : B then λx · t : A → B Types a∈A p ∈ A → B. Now there is this striking parallel. proofs are actually a special kind of constructions. see their papers in [Fenstad 1971]. § E. and a proposition has proofs. [Curry-Feys 1958]. and extended to full intuitionistic logic by W. [Kleene 1952] Ch. [Girard 1989]. Propositions a:A p : A → B. Starting from certain basic sets with types. q ∈ A ⇒ p(q) ∈ B x : A ⇒ t(x) ∈ B then λx · t ∈ A → B It now is a matter of finding the right types corresponding to the remaining connectives. denoted by A → B or B A . Howard. [Dummett 1977] Ch. 10. Prawitz and MartinL¨f. and they operate on each other. [Buss 1998]. [Beeson 1985]. 31 . 4. As we have seen. For ∧ and ∨ we introduce a product type and a disjoint sum type. Similarly one may consider sets as being typed in a specific way. [Troelstra–van Dalen 1988] Ch.

But intuitionistically S has a characteristic function of its a ∈ S ⇔ kS (a) = 1 membership is decidable: a ∈ S ⇔ kS (a) = 0 32 . ∀2 and →. higherorder logic may contain rules or axioms that are constructively justified. [Troelstra-Schwichtenberg 1996].The main aspect of the Curry-Howard isomorphism. [Troelstra–van Dalen 1988]. sets and functions are interdefinable as each set has a characteristic function. CH. 11. cf. In practice there are two ways to formulate second-order logic: with set variables and with function variables.g. in his approach the two are so closely interwoven. Per Martin-L¨f was the first logician to see the full importance of the o connection between intuitionistic logic and type theory. Prawitz showed that one can define the connectives in ∀1 . His type systems are no mere technical innovations. [Martin-L¨f 1975]. (also knowns as “proofs as types”). but which contradict classical logic. that they actually merge into one master system. is the faithful correspondence: proofs elements = propositions types with their conversion and normalization properties. Whereas first-order intuitionistic logic is a subsystem of classical logic. It is a surprising fact that in IQC2 the connectives are not independent as in first-order logic. [Martin-L¨f 1984]. [Gallier 1995]. [Prawitz 1965]. but they intend to catch the foundational meaning of intuitionistic logic and the corresponding mathematical universe. 1.X ∀2 X((A → (B → X)) → X) ∀2 X((A → X) ∧ (B → X) → X) ∀2 X(∀1 y(A → X) → X) ∀2 X(∀Y ((A → X) → X) (where X is a 0-ary predicate) Classically. ⊥ A∧B A∨B ∃1 xA ∃2 XA ↔ ↔ ↔ ↔ ↔ ∀2 X.5 IQC2 Second-order logic with set variables is a straightforward adaptation of the classical formulation. [Dalen 1997]. Expositions can be found in e. o o Suggested further reading: [Girard 1989]. Indeed.

x) Here γ(y) stands for the initial segment of length y of a sequence γ On the basis of this principle. Brouwer’s chief contribution to this part of intuitionistic logic is that he realized that particular quantifier combinations get a specific reading. with their ‘input–output’ behaviour are more tractable than sets. of natural numbers. such that A(α. in a nutshell. and any β that coincides so far with α will yield the same n. n) for some formula A. Choice sequences of. for example.6 The theory of choice sequences For the practice of intuitionistic mathematics. For the proof theory of IQC2 .Since kS (a) = 1∨kS (a) = 0. We see that the “set”-approach and the “function”-approach to second-order logic (arithmetic) yield diverging theories. [van Atten–van Dalen ??]. but then we need no further information. In later papers Brouwer further investigated the continuity phenomenon. for an extensive analysis cf. we will eventually be able to compute the number n. we have all the information we need for the computation of n. That is a good reason to study second-order arithmetic with function variables. chosen more or less arbitrarily. This sketch is somewhat simplified. already a number of intuitionistic facts. total functions. for more information the reader is referred to the literature. see [Prawitz 1970]. That is to say. Generally speaking. we get a ∈ S∨a ∈ S. conflicting with classical mathematics. is Brouwer’s Continuity Principle: ∀α∃xA(α. n) → ∀α∃xy∀β(α(y) = β(y) → A(β. α. [Troelstra-Schwichtenberg 1996]. Since this survey is about logic. second-order arithmetic with function variables is even more significant than the version with set variables. Let us therefore just briefly note the main points. in general there is no law that determines future choices. The moral is that there are lots of sets without characteristic functions. [Troelstra 1973]. the topic is not quite within its scope. Suppose now that we have shown ∀α∃nA(α. this tells us that at some stage in the generation of α. say. which enabled him to show his famous Continuity Theorem: all total real functions on the continuum are 33 . It allows. can be derived. n) holds. natural numbers are infinite sequences. This. Another reason is that this formulation is a natural framework for treating Brouwer’s choice sequences. The reason is that this theory allows us to capture the properties of Brouwer’s choice sequences. formulated for the ∀α∃!x in [Brouwer 1918]. a simple rejection of PEM. 1. Roughly speaking. He added a powerful induction principle. this means that when we generate a choice sequence α.

α keeps track of the succes of the subject in establishing A. β) → ∃F ∀αA(α. known by the name “the creating subject”. The new idea. As a corollary the continuum cannot be decomposed into two inhabited parts. was formulated by Kreisel in terms of a tensed modal operator. [Dalen 1999A]). Brouwer introduced in the twenties an extra strengthening of analysis. Subsequently Kripke simplified the presentation by introducing a choice sequence α that “witnesses” a particular statement A. In order to demonstrate the technique of the creating subject. [Brouwer 1949]. So ∃k∀n∃x(|f (x)| > 2−k ). van Dalen showed that Brouwer’s indecomposability theorem for R can be extended to all dense negative subsets of R(A ⊆ R is negative if ∀x(x ∈ A ↔ ¬¬x ∈ A). we show here that. He showed for example that (cf. where F is a continuous operation. Put γ(2n) = α(n) xn if ∀p ≤ n γ(k) = 0 cn = γ(2n + 1) = β(n) xp if p ≤ n and γ(p) = 1 and c = lim(cn ) 34 . and β for r ∈ Q. Let α be the Kripke sequence for r ∈ Q. and when A is proved. [Brouwer 1924]. Hence there are xn with |f (xn )| > 2−k . [Hull 1969]) ¬∀x ∈ R(¬¬x > 0 → x > 0) and ¬∀x ∈ R(x = 0 → x#0) Kripke’s schema has indeed consequences for the nature of the mathematical universe. Another well-know consequence is the Fan theorem (basically the compactness of the Cantor space). ([Myhill 1966]). that is ∀α∃βA(α.g. Let f be discontinuous in 0. α produces a single ‘one’ and goes on with zeros. So. for example. it produces zeros as long as the subject has not established A. e. the set of not-not-rationals is indecomposable (cf. or experienced. Proof.locally uniformly continuous. under the assumption of Kripke’s schema one can show a converse of Brouwer’s indecomposability theorem: KS + R is indecomposable ⇒ there are no discontinuous real functions. and f (0) = 0. which go beyond the already existing Brouwerian counterexamples. the details of which were published only after 1946. F (α). it conflicts with ∀α∃β-continuity. Using KS. The existence of such a α is the content of Kripke’s schema: KS ∃α(A ↔ ∃xα(x) = 1) Brouwer used the creating subject (and implicitly Kripke’s schema) to establish strong refutations.

Begr¨ndung der Mengenlehre unabh¨ngig u a vom logischen Satz vom ausgeschlossenen Dritten. Semantic Construction of Intuitionistic Logic. Brouwer-Heyting Sequences Converge. Brouwer’s continuity principle. [Scott 1970] [Moschovakis 1973].van Atten. Wetenschappen afd. [Beth 1956] E. D. Beth. Cambridge. so ∀p(γ(p) = 0). Kon Ned Ak Wet Verhandelingen. 5:1–43. [?]. [Brouwer 1992]. If f (c) < 2−k then f (0) = 0. Brouwer.E. Varieties of Constructive Mathematics. This contradicts the indecomposability of R. [Troelstra 1977].M. Cambridge University Press.Now. 20:14–15.J. Nederlandse Ac. Arguments for [Beeson 1976] M. then r ∈ Q ∨ r ∈ Q. [Brouwer 1919a] ———– Begr¨ndung der Mengenlehre unabh¨ngig vom u a logischen Satz vom ausgeschlossenen Dritten. Math. The unprovability in intuitionistic formal systems of the continuity of effective operations on the reals. Kon Ned Ak Wet Verhandelingen. 12. There is an extensive literature on the semantics of second-order arithmetic. Berlin. [Scott 1968]. Zweiter Teil. Letteren:Mededelingen. If f (c) > 0. Erster Teil. The theory of choice sequences has received a great deal of attention after Kleene and Kreisel formulated suitable formalizations. Kon.f. Springer. such as “lawless sequence”. 7:1–33. Borwein. References [van Atten–van Dalen ??] M. [Dalen 1978].W. Contradiction. c. D. Beeson. [Bridges-Richman 1987] F. [Hoeven-Moerdijk 1984]. [Fourman-Scott 1979]. [Beeson 1985]. [Kleene–Vesley 1965]. 41: 18–24. Allgemeine Mengenlehre. f (c) < 2−k ∨ f (c) > 0. [Dummett 1977] Ch.f. Bridges. have found important applications in metamathematics. C. Intelligencer. Richman. [Fourman-Hyland 1979]. Some notions. [Borwein 1998] J. Theorie der Punktmengen. van Dalen. 3. Hence ∀r(r ∈ Q ∨ r ∈ Q). 35 . 19/11: 357–388. J Symb Logic. [Troelstra–van Dalen 1988] Ch. [Kreisel–Troelstra 1970]. [Beeson 1985] ———– Foundations of Constructive Mathematics. Forthcoming. [Brouwer 1981]. [Brouwer 1918] L.

[Brouwer 1949] ———– Essentieel negatieve eigenschappen. Cambridge University Press. Geometer. and Intuitionist: The Life of L. Math. Brouwer Vol. Combinatory Logic. Jahresber. handproof. 10:322-323. [Buss 1998] S. Geometer.J.. Elsevier. [Brouwer 1924] ———– Beweis dass jede volle Funktion gleichm¨ssig stetig a ist. 13:1–43.J.. Amsterdam. 36 . Appeared in 1920.. [Dalen 1999] ———– Mystic. Math. Amsterdam. [Curry-Feys 1958] R. 1. Dtsch. ¨ [Brouwer 1923] ———– Uber die Bedeutung des Satzes vom ausgeschlossenen Dritten in der Mathematik insbesondere in der Funktionentheorie.[Brouwer 1919b] ———– Intuitionistische Mengenlehre . Oxford. [Brouwer 1981] ———– Brouwer’s Cambridge Lectures on Intuitionism. The dawning Revolution. Buss. Wissenschaftsverlag. 28:203–208. revised 2008). van Dalen) Intuitionismus. Vol. Ind. Oxford University Press.B. Math. Berlin. 2. 154:1–8. Bibliographisches Institut. Oxford University Press. Springer Verlag.E. [Dalen 2005] ———– Mystic. [Dalen 1974] ———– A model for HAS. Nederl Ak Wetensch Proc. Oxford. Mannheim. Hope and Disillusion. [Brouwer 1992] ———– (Ed. [Dalen 1978] ———– An interpretation of intuitionistic analysis.E. Cambridge. [Dalen 1997] ———– Logic and Structure (3rd edition. 82:167–174. Ann Math Log. J Symb Logic. [Dalen 1984] ———– How to glue analysis models. Fund. H. A topological interpretation of second-order intuitionistic arithmetic with species variables. North– Holland. Feys Curry. D.Ver. J Reine Angew Math. 27:189–193. I. Brouwer Vol. 49:1339– 1349. and Intuitionist: The Life of L.

[Dalen 1978] R. In J. Oxford University [Fenstad 1971] J. vol. North-Holland. Ann Math. Scott. American Mathematical Society. [Dragalin 1988] A. 2nd edition. Statman Dalen. D. and J. pages 280–301. G¨nthner. In Handbook of Philosophical Logic. Rose. Fitting. Diaconescu. F. Gabbay. [Dummett 1977] ———– Press. [Dalen 2002] ———– Intuitionistic Logic. Providence. Sheaf models for analysis.P. Proc. Acad. Shepherdson H. [Fourman-Scott 1979] Fourman. Friedman. Amsterdam. 72:2877–2878.E.. 113–170. NorthHolland Publ.E. In [Mathias–Rogers 1973]. u Kluwer. The philosophical basis of intiutionistic logic.) Essays on Mathematical and Philosophical Logic. Co. Mathematical Intuitionism. editor.S. Springer. 5. pages 5–40. 105:1–28. Oxford. Soc. Dummett. Studia Logica. 405 pp. Sheaves and Logic. [Fourman-Hyland 1979] Fourman. Dordrecht. Intuitionistic model theory. 51:176–178.) Proceedings of the Second Scandinavian Logic Symposium. In Hintikka et al(eds. Axiom of choice and complementation. pages 95–116. Hyland. Equality in the presence of apartness.P. 37 .Reidel. Dordrecht. Amsterdam. Springer. M. Logic Colloquium ’73. [Dummett 1975] M. Berlin. Proc. pages 280–301. M. North–Holland. Fenstad (ed. Science USA. Nat.. [Friedman 1973] H. and forcing. Berlin. Amsterdam. and D. Some applications of Kleene’s methods for intuitionistic systems.M. pages 1–114.. [Fitting 1969] M.E. [Diaconescu 1975] R. American Math. [Friedman 1977] ———– Set theoretic foundations for constructive analysis. eds. Elements of Intuitionism. 62:305–314.[Dalen 1999A] ———– From Brouwerian Counter Examples to the Creating Subject. [Friedman 1975] ———– The disjunction property implies the numerical existence property. Dragalin.

38:86–92. de Groote. [Grayson 1984] ———– Heyting–valued Semantics. M. Gabbay.J. Marcia. 405–431. Berlin. and J. editor. 36:249–261. [Gabbay 1973] ———– The undecidability of intuitionistic theories of algebraically closed and real closed fields. J Symb Logic. Y. Anzeiger o o u der Akademie der Wissenschaften in Wien. Fourman. pages 55–138. Amsterdam. pages 402–414. Decidability of some intuitionistic theories. J Symb Logic. Goernemann. 39:176–210. [Grayson 1981] ———– Concepts of general topolog in constructive mathematics and in sheaves. Grayson. [Girard 1987] J. A. Ann Math Log. Louvain-la-Neuve. durham. G. A logic stronger than intuitionism. (Coll. Napoli. In (Eds. Proof Theory and Logical Complexity. Zum intuitionistischen Aussagenkalk¨ l.P. 222–225). Untersuchungen uber das logische Schliessen I. Cambridge University Press. Girard. 20:1–41. Bibliopolis. Math. 38 . Gallier. [Gallier 1995] J.-Y. J. C. Proceedings of the Logic Colloquium ’82. J Symb Logic. Scott.D. [Goernemann 1971] S. Gentzen. 24:461. Reidel. Zeitschr. [Girard 1989] P. Proofs and Types. I. D.. Longo. [G¨del 1932] K.Springer. Ann Math Log. Mulvey. editor. [Gentzen 1935] G. 23:55–98.J. [Gabbay 1982] ———– Semantical Investigations in Heyting’s intuitionistic logic. In Ph. Lafont. The Curry–Howard Isomorphism. 69:65–66. G. Elsevier.. Z. Heyting-valued models for intuitionistic set theory. Vol. Cambridge.[Gabbay 1972] D. pages 181– 208.Logic Grundlagen Math. 37:579–587. II. G¨del. [Grayson 1979] R. . ¨ II. In Lolli.Math. On the Correspondence between proofs and λterms.-Y.. N. Academia. Myhill. p.S. [Goodman-Myhill 1978] Goodman. Taylor Girard. [Grayson 1982] ———– Concepts of general topolog in constructive mathematics and in sheaves. Dordrecht. Choice implies excluded middle.M.Works I. editor.

Amsterdam. [Heyting 1934] ———– Mathematische Grundlagenforschung. 42–56. Amsterdam.. Academic Press. Logik. Zeitschr. 15:241–246. In J. North- [Kleene–Vesley 1965] Kleene. 158–169.E. [Hyland 1982] J. Zur Deutung der intuitionistischen Logik.P. [Jaskowski 1936] Recherches sur le syst`me de la logique intuitioniste. Paris. Ann Pure Appl Logic. Physikalische-Mathematische Klasse.F. Hindley J. [Kolmogorov 1932] A. Springer Verlag. Phil. Howard. editor. Kleene. Amsterdam..S. Introduction to Metamathematics.E. [Howard 1980] W. Sheaf models for choice sequences. des Sciences.G. [Heyting 1930b] ———– Die formalen Regeln der intuitionistischen Mathematik II.. 57–71. The Formulas-as-Types Notion of Construction. Amsterdam. [Hull 1969] R. [Kleene 1952] S. e Congr.. Sitzungsberichte. 39 . math.E.R. 6: 58–61. Hermann. Seldin. van Dalen A. [Heyting 1956] ———– Intuitionism. To H. Math.B. pages 165–216. Intuitionismus. 27:63–107.C. North–Holland. Holland.Elsevier. Curry: Essays on Combinatory Logic. an Introduction . Vesley. [Hoeven-Moerdijk 1984] Hoeven. Die Preussische Akademie der Wissenschaften.. The Effective Topos.M. and R. In D. Sitzungsberichte.. Heyting. Zeitschr. The Foundations of Intuitionistic Mathematics especially in relation to Recursive Functions.J. S. Die Preussische Akademie der Wissenschaften.C. Beweistheorie. and Moerdijk. Lambda calculus and Formalism. G. Troelstra. Kolmogorov.N.[Heyting 1925] A.. Physikalische-Mathematische Klasse. 35:58–65. Intu¨ ıtionistische axiomatiek der projectieve meetkunde. Hyland. Hull Counterexamples in intuitionistic analysis using Kripke’s schema. Amsterdam. [Heyting 1930a] ———– Die formalen Regeln der intuitionistischen Logik. The L. pages 479–490. PhD thesis. Berlin. North–Holland. intern.. editor. New York. Brouwer Centenary Symposium.A. I. III.

4:29–31. Ann Pure Appl Logic.A.. [MacLane–Moerdijk 1992] MacLane. S. Rogers (eds. [Mathias–Rogers 1973] Mathias. and I.. Semantical Analysis of Intuitionistic Logic. [McCarty 1984] ———– Realizability and Recursive Mathematics. Institute Steklov. AMS Transl. 1:229–387. Markov. Rose and J. and H. Lifschitz. 98:1–9. J. and P. Springer. Elementary completeness properties of intuitionistic logic with a note on negations of prenex formulae. Orevkov Maslov. pages 92–130. Cambridge. [Lambek–Scott 1986] Lambek. [Martin-L¨f 1975] P.P. Ann Math Log. McCarty. Realizability and Recursive Set Theory. Steklov. [Kripke 1965] S. Inst. Original Trudy Mat..E. Scott. Cambridge University press. J Symb Logic.North–Holland. 40 . On Constructive Mathematics. editor.J. Amsterdam.. Soviet Mat. o [Maslov 1965] V. Proc.A. 6:918–920. editors.C. pages 73–118.A. Troelstra. Oxford 1963. Dummett J. In M. Cambridge Summer School in Mathematical Logic. Amsterdam. Moerdijk. North-Holland. Formal systems for some branches of intuitionistic analysis.A.) . Springer Verlag.N. [Martin-L¨f 1984] ———– Intuitionistic Type Theory.E. The decision problem for some constructive theories of equality.. Formal Systems and Recursive Functions.[Kreisel 1958] G. Seminar Math. of the eighth Logic Colloquium. Dokl. [Markov 1971] A. Unsolvability in the constructive predicate calculus of certain classes of formulas containing only monadic predicate variables. Berlin.Yu. Mints. A first Introduction to Topos Theory. An Intuitionistic Theory of Types. [Kreisel–Troelstra 1970] Kreisel. Oxford. Bibliopolis. Kreisel.S. G. G. Introductioin to higher order categorical logic. [Lifschitz 1967] V. and A. Berlin. 32:153–183. Shepherdson. [Kripke 1968] ———– Semantical Analysis of Intuitionistic Logic II. S. Napoli. Crossley. 23:317–330. Kripke. PhD thesis. In o o H. Sheaves in Geometry and Logic. unpublished. Logic Colloquium ’73. [McCarty 1986] C.67 (1962). Martin-L¨f.

Comp Math. editor. Berlin. [Scott 1968] D.North-Holland.S. Myhill. 32:513–532. 280-297. Almqvist & Wiksell. Troelstra Schwichtenberg. von Plato. Vesley. editor. Basic proof theory. Amsterdam. A proof-theoretical study. Springer. Sikorski. Stockholm. Natural Deduction. pages 660–696. In Myhill. [Moschovakis 1973] J. Scott. In Myhill. Panstowe Wydawnictwo Naukowe. Forthcoming. Berlin. H. Sch¨tte. Warsaw. buffalo. Logique et Analyse 9. 20:194–210. Vesley. Moschovakis. Oxford.. [McLarty 1995] C. Structured Proof Theory. [Troelstra-Schwichtenberg 1996] A. [Rasiowa-Sikorski 1963] Rasiowa. Cambridge University Press. 43:2–40. pages 235–255. buffalo. [Sch¨tte 1968] K. J and A. [Prawitz 1970] ———– Some results for intuitionistic logic with second order quantification. The Mathematics of Metamathematics. Comp. Notre Dame Journal of Formal Logic.[McCarty 1991] ———– Polymorphism and Apartness. H. Clarendon Press. Kino and R. Theoria. and R. [Scott 1979] ———– Identity and Existence in Intuitionistic logic. Notes towards an axiomatization of intuitionistic analysis. 41 . Vollst¨ndige Systeme modaler und intuitionisu u a tischer Logik. pages 259–269. [Scott 1970] ———– Extending the topological interpretation to intuitionistic analysis II. Elementary Toposes.S. [Prawitz 1965] D. [Negri–von Plato] S. Kino and R. Elementary Categories. [Myhill 1966] J. Negri. Cambridge. J. Extending the topological interpretation to intuitionistic analysis. A Topological Interpretation of Intuitionistic Analysis. [Prawitz 1977] ———– Meaning and Proofs: On the Conflict between Classical and Intuitionistic Logic. North-Holland. J and A. 26:261–275. Prawitz. McLarty. Math. Amsterdam. Springer.

Oxford.J. 41:159–166. Der Aussagenkalk¨l und die Topologie. Eds. [Smorynski 1982] ———– Nonstandard models and constructivity.M. Bertin.[Smorynski 1973a] C. II. The L. [Veldman 1976] W.S. Bos. D. Commentary. [Troelstra–van Dalen 1988] Troelstra. Tarski.S. Berlin. Metamathematical Investigation of Intuitionistic Arithmetic and Analysis. Berlin. Applications of Kripke models. In A. In “Two decades of mathematics in the Netherlands”.. Amsterdam. J Symb Logic. Clarendon Press. In D. A.S. Constructivism in Mathematics. Veldman.M. Chicago. A chapter of intuitionistic mathematics. [Smorynski 1973c] ———– On axiomatizing fragments. Mathematical Centr. H. van Dalen A. [Troelstra 1978] ———– Heyting. “Die formale Regeln der intuitionistischen Logik” (1930). u Math. [Smorynski 1973b] ———– Investigation of Intuitionistic Formal Systems by means of Kripke Models.. Troelstra. [Troelstra 1977] ———– Choice Sequences.. Springer.J. PhD thesis.S. Metamathematical Investigation of Intuitionistic Arithmetic and Analysis. Springer Verlag. Amsterdam. Amsterdam. 31:103–134. 42 . editor. Troelstra. Elsevier. [Troelstra 1973] A.S. Ann Math Log. pages 459–464. 1. Co. 14:193–221. Logic.. Fund.E. Grootendorst. North-Holland Publ. J. A.W. [Tarski 1938] A. An intuitionistic completeness theorem for intuitionistic predicate calculus.J. pages 324–391. I. Brouwer Centenary Symposium. van Dalen. [Smorynski 1978] ———– The axiomatizing problem for fragments. 163–175. Troelstra. Smorynski. 42: 530–544. E. editor. Symb.

Sign up to vote on this title
UsefulNot useful