This action might not be possible to undo. Are you sure you want to continue?

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

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

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

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

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

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

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

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

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

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

There is a corresponding theorem for existential sentences. we have k A ∨ B. however. i. The result is a correct Kripke model. and hence k A or k B. see 15 . As the reader has seen. the most important being arithmetic. the result has not been established constructively. But that contradicts the fact that A and B are not forced in K1 and K2 . you can indeed point to a speciﬁc instance. you know that you can established one of the disjuncts. Since i A ∨ B.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). Therefore i A or i B. There is a snag in the conclusions drawn above. The disjunction and existence property hold for a number of prominent theories. Suppose i A and i B. DP and EP are often considered the hallmark of constructive logic. Similarly for existence: if you have shown the existence of something. not overestimate the signiﬁcance of a technical result like this. In this new node no proposition is forced. 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.e. one should. Fortunately there are proof theoretical devices that provide exactly this kind of information. similarly a tree model K2 where B is not forced. the proof made use of reductio ad absurdum. The two models are glued together as follows: put the two models side by side and place a new node k below both. 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. What one would like is a method that extracts from a proof of ∃xA(x) a proof of A(d) for some d. then there is a tree model K1 where A is not forced.

They allow for a very natural interpretations of choice sequences.3 Super Semantics. The use of Beth models for second-order systems can be found in [Dalen 1978]. but by no means all of them. see [Troelstra–van Dalen 1988]. [Veldman 1976]. and on the face of it. It is easily seen that these sets form a topology. Ch. Troelstra have given alternative versions for Beth semantics. There is. Beth models happen to be better adapted to second-order arithmentic with function variables.]] is a valuation as deﬁned in section 2. Veldman was in 1976 the ﬁrst to consider a modiﬁed Kripke semantics. Friedman. [Dalen 1984]. so in each node A ∨ ¬A holds. In particular. This is based on algorithms. [Smorynski 1982]. The combinatorics of Beth models is just more complicated than that of Kripke models. k ∈ U and k ≤ ⇒ ∈ U . One also sees that {k|k A} is an open set. Kripke showed in 1963 that Kripke and Beth models can be converted into each other [Kripke 1965]. [Troelstra–van Dalen 1988]. One might wonder if these semantics are totally unrelated.e.28. Dk = D for all k. a totally diﬀerent interpretation of intuitionistic logic (and arithmetic) in Kleene’s realizability interpretation. i. let us therefore put [[A]] = {k|k A}. Smorynski has shown that in a fairly large number of cases ‘semantic’ proofs can be made constructive. or whether there is some common ground. it is useful for rendering completeness proofs in an intuitionistic metamathematics. u Beth models are easily convertible into topological models. On a tree one can deﬁne a topology as follows: open sets are those subsets U of the tree that are closed under successors. [Fitting 1969]. i.e. the logic is classical. Beth’s original models are slightly more special. one could not ﬁnd a similarity with the above semantics. he considered constant domains. Dummett. i. For convenience we consider Beth models over tree. This is a certain drawback when compared to Kripke models. The obvious 16 .e. . Now it is a matter of routine to show that the function [[. 1. 13. Beth semantics also turn out to be a convenient tool in completeness proofs. for example. the so-called intuitionistic analysis. We have seen a number of interpretations of intuitionistic theories. Finite Beth models are not interesting for intuitionistic logic. For more information on Kripke model theory see [Dalen 1997]. [Dummett 1977]. For in each leaf (maximal node) A or ¬A holds. see also [Sch¨tte 1968].1. Since then De Swart. for which he could give an intuitionistically correct completeness proof.

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

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

Fourman. [Hoeven-Moerdijk 1984]. they carry a natural topology. E. [McLarty 1995]. see [Scott 1970]. Sheaf models were further presented by Scott. [Dummett 1977] Ch. 13 §8. Certain categories are powerful enough to interprete higher-order intuitionistic logic. 6. Scott extended his model to a higher-order theory of reals. 5. [Fourman-Hyland 1979]. Grayson used the same approach in his version of the permutation model of Krol. [Friedman 1977]. 13. [Fitting 1969]. Moerdijk and Van der Hoeven used a special sheaf model to interprete the theory of lawless sequences. Categorical models also turned out to be important for typed lambda calculus. Ch. Ch. [Grayson 1979. 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. Hyland and Grayson. In that sense topos models can be viewed as intuitionistic universes. Joan Moschovakis applied the methods of Scott’s semantics to intuitionistic analysis in the style of Kleene’s FIM. Grayson 1984]. [Dalen 1978].15. they are natural examples of a topos.g. in which he could show that Brouwer’s continuity theorem holds. The model also interprets the theory of the creating subject. Hyland showed that the realizability interpretation ﬁtted into his eﬀective topos. Her objects were continuous mappings from Baire space to Baire space (a model with total elements). Suggested further reading: [Dragalin 1988] Part3. [Fourman-Scott 1979]. The next generalisation after sheaf models is that of categorical model. [Fourman-Scott 1979]. Sheaf models allow an interpretation of higher-order logic.One may add also the ordering to the model of R: [[a < b]] := {t|a(t) < b(t)}. Since Kripke models and Beth models are built over trees. [Moschovakis 1973]. [Rasiowa-Sikorski 1963]. There is no special foundational bias towards ﬁrst-order theories. [MacLane–Moerdijk 1992]. [Gabbay 1982]. [Hyland 1982]. It can be translated straightforwardly into a topological model. [Lambek–Scott 1986]. [Smorynski 1973a]. Grayson 1981.13. E. [Troelstra–van Dalen 1988] Ch. a model for lawless sequences is constructed by means of forcing. but its formulations in a Beth model allows for some extra ﬁne tuning so that techniques from set theory can be applied. as a result they can be viewed as sheaf models.g. Grayson 1982. [Grayson 1981]. §5. but it is an observed fact that large parts of mathematics lend themselves to formu19 . Van Dalen’s model for analysis is based on Beth’s original semantics. [Troelstra–van Dalen 1988]. 2.

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

Hence k0 A ∨ B and so k0 A or k0 B. Then for each i there is a model Ki with Ki A(ci ). only the trivial identities ci = ci ). For EP it is convenient to assume that the theory has a number of constants. Hence AP A or AP B. say {ci |i ∈ I}. Then. we glue the models Ki by means of a bottom world K∗ with a domain consisting of just the elements ci . Let AP A ∨ B and assume AP A and 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. 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. Both are impossible on the grounds of the choice of K1 and K2 . The identiﬁcation of the cj with elements in the models Ki is obvious. Contradition. by the strong completeness theorem. Now let AP ∃xA(x) and AP A(ci ) for all i. it is easy to check that 21 . Again. As we did above. The new model obviously satisﬁes the axiom of AP. No non-trivial atoms are forced in k0 (i. we designate points a and b in K1 and K2 which are identiﬁed with the point 0.e. That is to say.

The result was ﬁrst established by proof theoretic means in [Dalen 1978]. so x = y ↔ ¬¬¬x#y ↔ ¬¬x = y. [Gabbay 1972]. on the other hand. i. It is important to note that the atoms. Hence k0 ∃xA(x). In particular we get a stable equality: ¬¬x = y → x = y. To be precise: AP is conservative over EQ + {Sn |n ≥ 0}. contradiction.g. it has to be to right of a or to the left of b. [Smorynski 1973a]. The apartness axioms have consequences for the equality relations. 22 . There are quite a number of surprising results on ﬁrst-order theories. Hence AP A(ci ) for some i. In fact the equality fragment is axiomatized by an inﬁnite set of quasistability axioms. This shows that even a relatively simple theory like equality is incomparably richer than the classical theory. But then also Ki A(ci ). but the theory is not classical. e. [Gabbay 1973]. 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). [Gabbay 1982]. so to speak.the new model satisﬁes AP. cf. In a way this gives the best possible result since LO+AP is conservative over LO ([Smorynski 1973c]. or in general the quantiﬁer free part of a theory does not yet determine whether it is classical or not. 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.e. 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. [Dalen 1997]). that a < b means that a is “far” to the left of b. the theory of algebraically closed ﬁelds. Apartness and linear order are closely connected. ¬x#y ↔ x = y. and subsequently an elegant model theoretic proof was given in [Smorynski 1973c]. 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). k0 A(ci ) for some i. The Sn axiomatize the equality fragment of AP. For. There are cases where a decidable equality results in the fact that the theory is classical. in the sense that if we pick an arbitrary third point.

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

o clearly. [Beeson 1985].129. 397. [Beeson 1976]. and hence A(n) is true in the standard model for some n. Given a statement of the form A → B. the model refutes M P . who showed that PA can be translated into HA. Markov’s rule. and hence would yield a conﬂict with the independence of the G¨del sentence from PA. hence HA A(n). [Troelstra–van Dalen 1988]. G¨del deﬁned o a translation. For closed ∃xA(x). is true. has a stronger position. In the early days of intuitionism people wondered to what extent HA was safer that PA. Note that A(x) may contain more free variables. p. This was basically done by a judicious distribution of negations. but its companion.e. then PA ∃xA(x). p. Markov’s principle may be unprovable in HA. The heuristic argument is that we have more information in this case. we actually have a proof of ¬¬∃xA(x). where k0 < k. it is no surprise that a great deal of research has been devoted to the subject. Here is the formal deﬁnition: 24 . k0 ∀x(A(x) ∨ ¬A(x)). The letter is impossible. Consider a model with two nodes k0 . Γ A ⇒ Γ B. from this extra evidence we may hope to draw a stronger conclusion. p. the proper sentence of the form ∃xA(x). Now HA A(n) ∨ ¬A(n) and by DP HA A(n) or HA ¬A(n). 366. This was settled in 1933 by G¨del (and independently by o Gentzen). 507. 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”.A simple Kripke model shows that HA M P . the proof of closure under Markov’s rule is particularly simple: if HA ¬¬∃xA(x). cf. Markov’s rule is indeed correct. which is in general weaker than Γ A → B. So k1 ∃xA(x) and hence k0 ¬¬∃xA(x). In the bottom node we put the standard model of natural numbers. But k0 ∃xA(x) would ask for an instance A(n) to be true in the standard model. i. one may formulate a corresponding rule. Since. and a fortiori HA ∃xA(x) ([Smorynski 1973a]. 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. k1 . which from the intuitionistic viewpoint weakened statements. p. Since the theory of natural numbers is at the very heart of mathematics.

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

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

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

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

The elimination runs as follows: A→B B A t:A→B s:A t(s) : B Observe the analogy to the proof interpretation. Since A is an assumption. we introduce a λx in front of the (given) term t for B. it has a hypothetical proof term. Note that the informal argument of page 3 is faithfully reﬂected. By binding x. On discharGing the hypotheses. containing only the connective ‘→’. [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. the proof term for A → B no longer depends on the hypothetical proof x of A. Here is a list: 29 . 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.x. this is Curry combinator K. 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. Let us consider a particular derivation. Let us now consider a cut elimination conversion.We will ﬁrst demonstrate this for a small fragment. 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. say x.

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

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

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

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

which go beyond the already existing Brouwerian counterexamples. where F is a continuous operation. Let α be the Kripke sequence for r ∈ Q. e. [Dalen 1999A]). β) → ∃F ∀αA(α. we show here that.locally uniformly continuous. it conﬂicts with ∀α∃β-continuity. Subsequently Kripke simpliﬁed the presentation by introducing a choice sequence α that “witnesses” a particular statement A. As a corollary the continuum cannot be decomposed into two inhabited parts. was formulated by Kreisel in terms of a tensed modal operator. F (α). 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 . Let f be discontinuous in 0. and when A is proved. Another well-know consequence is the Fan theorem (basically the compactness of the Cantor space). it produces zeros as long as the subject has not established A. So. Proof. 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. and β for r ∈ Q. Using KS. and f (0) = 0. α keeps track of the succes of the subject in establishing A. or experienced. the set of not-not-rationals is indecomposable (cf. In order to demonstrate the technique of the creating subject. The new idea. [Brouwer 1924]. So ∃k∀n∃x(|f (x)| > 2−k ). ([Myhill 1966]). Hence there are xn with |f (xn )| > 2−k . the details of which were published only after 1946. He showed for example that (cf. Brouwer introduced in the twenties an extra strengthening of analysis. α produces a single ‘one’ and goes on with zeros. for example. [Brouwer 1949]. 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.g. 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). known by the name “the creating subject”. [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. that is ∀α∃βA(α.

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

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

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

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

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

Kripke. Berlin. Formal systems for some branches of intuitionistic analysis. 6:918–920. Unsolvability in the constructive predicate calculus of certain classes of formulas containing only monadic predicate variables. Crossley.. Martin-L¨f. Oxford. Rogers (eds. [Mathias–Rogers 1973] Mathias. pages 73–118. S. J.. of the eighth Logic Colloquium. On Constructive Mathematics. Napoli. Proc.A. 1:229–387.A.[Kreisel 1958] G. AMS Transl. Institute Steklov. 40 . Cambridge University press. Ann Math Log. Orevkov Maslov. [MacLane–Moerdijk 1992] MacLane.E. [McCarty 1984] ———– Realizability and Recursive Mathematics. 4:29–31. An Intuitionistic Theory of Types. Bibliopolis. [Martin-L¨f 1975] P. o [Maslov 1965] V. editor. [Lifschitz 1967] V. and H.. Sheaves in Geometry and Logic. pages 92–130. Lifschitz.N. Ann Pure Appl Logic. Cambridge Summer School in Mathematical Logic. Moerdijk.) . Berlin. Scott. 98:1–9. 23:317–330.67 (1962). Springer Verlag. [McCarty 1986] C. Dummett J. A ﬁrst Introduction to Topos Theory.E..P.C. Formal Systems and Recursive Functions. Dokl. and P. S. PhD thesis. North-Holland. Realizability and Recursive Set Theory. The decision problem for some constructive theories of equality. Troelstra. [Lambek–Scott 1986] Lambek. Amsterdam. Amsterdam. McCarty. Elementary completeness properties of intuitionistic logic with a note on negations of prenex formulae. 32:153–183. Seminar Math. Shepherdson. and I. Inst. [Martin-L¨f 1984] ———– Intuitionistic Type Theory. editors. Markov.S.J. Rose and J. G. Steklov. unpublished. Logic Colloquium ’73.Yu. Introductioin to higher order categorical logic. G. and A.. [Kreisel–Troelstra 1970] Kreisel. Soviet Mat. Mints. Kreisel. In M. Cambridge. [Kripke 1965] S. [Kripke 1968] ———– Semantical Analysis of Intuitionistic Logic II. [Markov 1971] A. In o o H.North–Holland. Springer. J Symb Logic. Semantical Analysis of Intuitionistic Logic. Oxford 1963.A. Original Trudy Mat.A.

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

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

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd