# MATHEI~ATICS

OF INFINITY

Per M a r t i n - L ~ f Department of Mathematics, University of S t o c k h o l m

Box 6701,

113 85 Stockholm,

Sweden

HOCBHmaeTcH AH~peH HHKonaeBMqa

na~t~TM KOnMOFOpOBa

25.4.1903

-- 20.10.1987

t47

Intuitive considerations Through the technique of so called lazy evaluation, it has become possible to compute with infinite objects of various kinds. The most typical and wellknown example is that of a stream

(ao~(al,(a2,..o))),
which at any finite stage of its development looks like an initial segment of an ordinary list

( a o , ( a l , ( a 2 ..... ( a n _ l , n i l ) . . . ) ) ) ,
but differs by proceeding indefinitely. Another example is obtained by conceiving of an infinite binary sequence as an infinite composition of two unary constructors

o(](o(...))),
rather than, as is customary, as a function from the set of natural numbers to the two element set. An even simpler example, the simplest possible, in a way, which will play a central role in

the following, is the infinite natural number

s(s~s(...))),
the successor of the successor of the successor of etc. in infinitum. No other mathematical object, if only we can understand it as such, deserves better to be denoted by the traditional infinity symbol ~. I

Introduced by J. Wallis, De Sectionibus Conicis, Nova Methodo Expositis, Tractatus, Oxford, 1655, in the laconic parenthesis (esto enim ~ nota humeri infiniti;), apparently without worrying about its meaningfulness.

I

t48

All the preceding examples are examples of infinite elements of sets. But we may also let sets be infinite, not in the usual sense of containing infinitely many elements, but in the sense of having infinite depth, or proceeding indefinitely, like

A

X(A X(A X . . . ) ) ,

where the set A may itself be infinitely proceeding, or

((...)

+ (...))

+ ((...)

+ (...)).

The latter set is the disjoint union of two sets, each of which is the disjoint union of two sets, each of which is the disjoint union of two sets, etc. in infinitum. It looks very much like the Cantor set. If we can conceive of infinitely proceeding sets, we can certainly also conceive of infinitely proceeding propositions: because of the correspondence between propositions and sets, there is no substantial difference. A typical example is

NNNO..

= ((... o i )

od.)~_t,

the negation of the negation of the negation of etc. in infinitum. Since such an infinitely proceeding proposition has no bottom that you reach in a finite number of steps, it is not at all immediately clear what it should mean for it to be true. Nor does it seem clear whether this particular one ought to come out true or false. (To anticipate matters, on the interpretation that I

shall adopt, it will come out false.) An infinite object of yet another kind is the iterative set, that is, set in the sense of the cumulative hierarchy,

149

It is the singleton set whose only element is the singleton set whose only element is the singleton set etc. in infinitum. such set exists, of course~ No

in the usual cumulative hierarchy,

but it is as simple as possible an example of a nonwellfounded set in the sense of Aczel. nonmathematical 2 Finally, you may even conceive of like the wellknown

examples of infinite objects,

picture of the artist painting his own portrait. When you start thinking about infinite objects, like the

aforementioned ones, you soon realize that they are maybe not so novel creatures after all. We have also old examples, finite decimal fractions like in-

aooala2a3 . . . .

ao + ~o(al

+ ~o(a2 + ~o(a3 + ...)))

and infinite continued fraction expansions b0 a0 + aI
+

bI b2 a2 + a 3 + ...,

which proceed indefinitely in just the same way as the streams of the computer scientist. ~ud we all know the mathematics that

has been developed in order to deal rigorously with these particular infinite objects, namely, calculus, or analysis, in its

P. Aczel, Non-Well-Founded Sets, CSLI Lecture Notes, Number 14, Stanford University, 1988.

2

150

various forms. Originally, Euler, approximately,

from its inception to the time of calculus was really a calthat is, of infinitely like

infinitesimal

culus of infinites and infinitesimals, large and infinitely small quantities,

and I
'

I
2'

I
" ' "

respectively.

But, when it was put on a rigorous basis by Cauchy the infinites and infinitesimals were gradually

and Weierstrass,

eliminated in favour of the notion of limit, earlier introduced by d'Alembert, our notations, although traces of them still remain in some of like
mo l~o

n=o@

lima n , Z

n=l

an, nTylan ,

. . .

Usually,

though, we take great pains to explain that ~

makes

no sense by itself, that is, is no detachable part of the notation for a limit, an infinite sum, product, would have difficulties or the like, and

in interpreting an expression like

(1 + --~)~ .

Only during the period of the last thirty years has there been a resurgence of interest in infinite and infinitesimal numbers as a result of Abraham Robinson's conception of his nonstandard

Brouwer's much more radically novel idea of choice sequences. Foundations of Constructive Analysis. Broawer. 1918. Eine Erweiterung der Infinitesimalrechnung. InfinitesimalkalkG1. Verhandelingen der Koninklijke Akademie van Wetenschappen te Amsterdam. ones mentioned in the beginning. on the other hand. McGrawHill Book Company. E. lishing Company. New York. Amsterdam. I. Non-standard Analysis. 1-39. 69.151 analysis. 3 There is also the slightly earlier but less wellknown infinitesimal calculus of Schmieden and Laugwitz. for denotational like the are Scott's theory of domains semantics and Aczel's theory of nonwellfounded 3 A. BegrHndung der Mengenlehre unabh~ngig vom logischen Satz vom ausgeschlossenen Dritten. Mathematische Zeitschrift.) The two recent theories that have been contrived precisely for dealing with streams and other infinite objects. Bishop. Sect. 4 Otherwise. 5. Bibliographisches Institut. PP. Laugwitz. (Observe that the theory of choice sequences antedates nonstandard analysis by forty years. 6 It is one of the purposes of the present work to show that the introduction of choice sequences is an intuitionistic version of the formation of reduced products in nonstandard analysis. 1958. pp. 1978. which succeeds containing the infinity symbol all in making sense of expressions in a much more elementary and constructive way. the straightof analysis in the style of Cauchy forward constructivization and Weierstrass carried out by Bishop. and the further references given there. on the one hand. Robinson. 1967. 5 and. Vol. Eine elementare EinfHhrung in die Nichtstandard-Analysis. Vol. Erster Tell: Allgemeiae Mengenlehre. Mannheim. 3-43. 6 L. . we have. See also the book by D. J. 12. Laugwitz. 5 E. No. 1966. these various forms of analysis are classical theories. intuitionistic On the side. North-Holland Pub- 4 C° Schmieden and D.

pp. Berlin. etc.152 sets. op. 1984. Scott. cit. But what happens? We indeed get an interpretation of logic satisfying all the usual laws of intuitionistic logic: the only trouble with it is that it trivializes in the sense that it makes every proposition. Lecture Notes in Computer Science. Schmidt. Hence. as nonemptiness. 7 Of these. Automata. and P. 577-613. the interpretation of logic might seem as simple as it could possibly be: just stick to the interpretation of propositions as sets. namely. 8 p. The principal aim of domain theory is to make proper mathematical sense of the fixed point operator fix(f) : f(f(f(° . if propositions. elements of a set as points of a domain. We have here yet another example of an infinite object to add to the long list in the beginning. Languages and Programming.. Vol. Edited by M° Nielsen and E. M. 8 are validated in the domain interpretation. Now. truth as nonemptiness. Aczel. are interpreted as domains and truth every proposition comes out true. Intuitionistic Type Theory. Domains for denotational semantics. domain theory indeed succeeds in making sense of streams and similar infinite objects9 but it does not succeed in interpreting logic in a nontrivial way which harmonizes with the interpretation of sets as domains. come out true° The reason is that every the bottom element. as formalized in my type theory. since all the set theoretical laws. etc. . written in type theoret- 7 D. Bibliopolis. even absurdity. SpringerVerlag. Now. 1982. Martin-LGf. domain contains an element.))). like sets. functions from one set to another as approximable mappings in Scott's sense. 140. Napoli.

Thus type theory becomes inconsistent when the formal laws for the fixed point operator are adjoined to it. if the formal laws for the fixed point operator are adjoined to standard type theory. the formal laws for the fixed point operator that and succeeds in satisfying.153 ical notation. also to be thought of as a proposition. Hence. (x ~ A) f(x) e A (x e A) f(x) ~ A fix(f) = f(fix(f)) e A fix(f) E A They say that every approximable mapping from a domain into itself has a fixed point. we can derive fix(id) E A for an arbitrary set A. we can certainly derive l id(x) ~ A (x 6 A). id(x) = x ~A (x E A ) by explicit definition in standard type theory. are domain theory seeks to satisfy. fix(id) E N o = S.. The previous laws for the fixed point operator then get modified into . We might try to avoid this inconsistency by only requiring a function from a nonempty set into itself to have a fixed point. for an arbitrary set A. In particular. On the other hand.

That the second rule for the modified fixed point operator becomes satisfied follows from the axiom second recursion equation.(x.a. y ) f ( y ) ) E A. fix(a. and. conversely.s) ~ N. governed by the axioms = s(~) ~ N. although we no longer get the fixed point of the identity function as an element of the empty set alias . given the modified fixed point operator.f)) E A.y)f(y)) = rec(s(~).a.154 (x ~ A) aEA f(x) 6 A fix(a. but. a . fix(a. ( x .(x.(x.f) = rec(~. are readily seen to have the same effect as introducing an infinite natural number. This seems fine.f) (x ~ A) f(x) E A = f(fix(a.a.f) = r e c ( ~ .f)) ~ A These laws. we can define the modified fixed point operator by performing an ordinary recursion up to infinity. Indeed.y)f(y)) = f(rec(~. in turn. we can define infinity by putting = fix(O.y)f(y))) oo = s ( ~ ) E N and the = f(fix(a.f) E A a E A fix(a. where s(x) E N (x E N) is the successor function.

Aczel's approach in his theory of nonwellfounded sets is to avoid contradiction in introducing nonwellfounded sets.. objects should remain valid for the nonstandard.S(~)) true ~ ~ N. which is in contradiction with it. from which N i(N. On the other follows by instantiation once we allow hand. I(N. all the laws for the standard. or infinitely proceeding. namely.155 proof of falsehood. My own intuition has been that. this would mean giving up the principle of mathematical induction. In the case of arithmetic rather than set theory. which is to say that my thoughts have gone rather in the direction of nonstandard arith- .s(x)) t r u e (x E N) is readily proved by mathematical induction in standard type theory. the system is still inconsistent. objects. by relaxing the axioms of standard set theory. or wellfounded. which ~ = s(~) ~ N is inadmis- shows that the circular definition sible.??}.~. Thus we have reached a contradiction. by giving up the foundation axiom and replacing it by his antifoundation axiom. equal- E N. because NI(N. like n= (((.. on the contrary. ity ~ = s(~) or intensional.x.~gs(~)) true follows of course from the definitional. nonwellfounded.

and thereby all sorts of infinite objects. from the single infinite natural number =.. where A i is a nonempty set. s(s(o)).. At the zeroth ~0 except that it belongs to the stage. we know nothing about . Also. How are we then to make proper mathematical sense of the infinite? To get out of the dead end that we have reached. we must turn to the theory of choice sequences. s(o). .. which is greater than all standard natural numbers. 2. .))) is determined process by a noncircular but nonwel!founded definitional ~I = f I ( ~ 2 )' ~2 = f2(~3 ). Here fi is a function from Ai+ I to A i. The intuition is the following.o = s ( s ( s ( .. etc. . a set containing an element a i.156 metic and analysis. I. A choice sequence O( = fo(f](f2(. In particular. that is. that is. the idea of getting the fixed point operator. for i = 0. ) ) ) is very reminiscent of the fundamental property of a nonstandard model of arithmetic that it contain a natural number which is greater than o. A 0 is the set to which the choice sequence = ~0 O( belongs..

. Then. At the second stage. and the references to earlier works given there. whereas has become partially determined: it is neither freely variable nor constant but something midway in between. from which are the choice sequence is obtained by infinite composition. This notion of choice sequence is essentially due to Troelstra. and get to know that ~I = fI(° (2 )' where o( 2 is an ele- ment of A 2. think of = s(s(s(. as defined by the nonwellfounded 9 A.fi_1(ai). The information about o( thereby gets refined from fo(O(1) to f o ( f 1 ( ~ 2 ) ) . 152-160. we can break it off by putting ~ i = ai' where a i The choice is the element which shows the set A i to be nonempty. Thus ~ 0 is simply a variable ranging over A O. 1977.))) as a choice sequence. See particularly Appendix C.157 set A O.. at any finite stage. Clarendon Press.)).. At this first stage. Now. pp. ~0 is. Troelstra. definitional process that is. continuous functions on the Baire space. we ask what element of A I that 0( I is.. In this way. A Chapter of Intuitionistic Mathematics. . sequence ~ then gets approximated by fo(f1(. Oxford. we get to know that ~i when we ask what element of A 0 that 0 = fo(° (1 )' where ~I belongs to A 1. ~0 is freely variable. which is a standard element of A O. The reason why the nonwellfoundedness the definition will not lead to any inconsistency is that. 9 The only difference is that he has been concerned with the case when the functions. S.. the definitional process of continues without end. Choice Sequences..

f) (x ~ A) f(x) E A = f(fixi+1(a.. etc. = s(~i+1) ~ ~ N E N by the explicit defini- for i = 0. Thus the and thereby satisfy the axioms for ~ i ' .f) ~ A Conversely. It was only the accidental fact that the par~ = s(s(s(.t58 = n 0 = s(~ 1 ). I. obeys the rules where a E A and f(x) E A (x E A).. and to define g N. I.f)) E A fixi(a.a~(x. Then fixi(a.. which we have seen to be inconsistent. we can define ~ i ~ N by putting ~i = fixi(0's) ~ N i = 0.f) = rsc(~i. given fixi(a.Y)f(Y)) E A.f) as governed by these two rules. make the explicit definition fixi(a.))) proceeds in the same ticular choice sequence way all the time that seduced us into making the circular defini~ N. ° ° I : s(~2)' ~2 : s(~3)' • ~ 6 We are then lead to extend standard type theory by adjoining the axioms { tion ~ = ~0 tion ~ = s(~) ~i ~i ~ N. etc.f) (x ~ A) a E A f(x) ~ A a E A fixi(a.

f) as the indexed fixed point operator. x E N. and most importantly. etc. is an external = s(c'° i+1) sequence of nonstandard natural numbers satisfying ~ i N. I.159 rules for fixi(a. i = 0. Observe that the presence of ~ ~ N allows us to get a closed notation for the limit of an internal sequence of mathematical objects of whatever kind. since they entail true. if A(x) is a set for x G N. then lim a(x) = a ( ~ ) X=~ if a(x) ~ A(x) for ~ A(~). Likewise. etc.f) and the axioms for ~ i are formally equi- valent. (x) = s(~(s(x))) e N (x ~ N). i = O. lead of course to a contradiction. (Vx E N)(~(s(x)) <~(x)) which is in contradiction with the principle of mathematical induction. Since substitution is defined so as to com- . then lim A(x) = A ( ~ ) X= is a set. It is important that ~ i g N. because the axioms for a corresponding internal sequence. 1.. It is natural to refer to fixi(a. For instance.. Thus the limit operation lim X= is expressed simply as the literal for the variable x which substitution of the infinity symbol ~ tends to infinity. that is.

and can be expressed simply by substituting the infinity symbol for the variable which tends to infinity. a -. Let the predecessor and cut off subtraction functions be defined as usual by the recursion equations I pd(a) pd(O) pd(s(a)) = 0 E N.160 mute with every other operation. we get inversely ~i+I = Pd(~i) ~ N. For instance. This example shows two virtues of the nonnamely.(x. from the axiom E N.0 = a E N. = rec(a.O.a. lim (a(x) + b(x)) = a ( ~ ) + b(~) = lim a(x) + lim b(x) E Q holds by definition. in type theoretical notation.(x.y)pd(y)) Then. where Q is the set of rational numbers.y)x) E N. that is. b(x) ~ Q for x ~ N. i = s ( ~ i + 1 ) E N. . -. that limits always exist standard approach to analysis.b : rec(b.s ( b ) = pd(a b ) E N. = a E N. I ~ . this means that the limit operif a(x) E Q and then ation also automatically does so.

etc. pdi(~) = ~-si(O) E N the single the definitional for i = 0. is an infinite natural number in the sense of nonstandard exceeds arithmetic. by instantiation.si(O)) m N for i = O.161 Iterated use of the latter equation yields ~i = Pd(~i-1) . From the axioms ~ i is readily = s(~i+1) E N for i = O. Thus it is enough to introduce infinite number ~EN and require it to satisfy equalities - si(o) = pdi(~) = s(pdi÷ 1(~)) = s(~to si÷1(0)) e N. although we can manage with a single new constant. I. i and j = 0.. (x E N). I. true i+3 function is for arbitrary monotonic. w h i c h is to say that it we have all standard natural numbers. in ~urn. and hence. . from the axioms ~ i . it we still need infinitely many axioms to characterize proved that each ~ . = s(~i+1) E N. it. etc. I. etc. 1. etc. But the successor so that we can conclude sJ(o) ~ s J ( ~ i + j) true. 0 ~ . The latter equalities. So. are equivalent = si(pdi(~)) = si(~ ... there On the other hand. 0 ~ x true Indeed...

.162 follow the definitional . I.< S S N~ 8 N( 8 ... fi(xi~1) E A i (xi+ I E Ai+ 1). ( fi-1 f0 fl fi fi+1 or. where i = O.. a projective A0 ( extension of type t heor F. Nonstandard Let there be given. ai ~ Ai . etc.. to the picture . ~a i = 0 EN. etc. The case that we shall be especially ested in is when I N< A i set. which shows that each ~ i in the sense of nonstandard arithmetic. inter- I corresponding N( S A i = N. system of nonempty A 1 < . ~ equalities = S j (c~i+ j ) E N. Therefore we have sJ(o)~o0 i true is infinite for all j = O... sets Ai ~ Ai+ I < .. in the system T of standard type theory. . I. fi(xi+1) = s(xi+ I) E N (xi+ I ~ N ) .... in the formal notation.

fini I ~xi+11.. .. fini_1(xi+11. ...163 In the most general case.. .. xi+Ini+ I ) E Aini(fi1(xi+11.. xi+ini÷1 ) E Ail. . each A i need not be just a single set but a whole context Xil E Ail....Xin i is to be interpreted as a sequence of variables... xi+Ini+ I .... fini(Xi+11...... Xini_ I ~ Aini_1(Xil. ....~...xi+ini+1). .. 'xi+Ini+1 ) . namely. which means that fi1(xi+11.o. ... xi+ini÷ I as a sequence of functions of several variables mapping the context at stage i+I into the context at stage i.Xini_1) set (Xil E All. xi+ini+1_1). .9 Xin i E Aini(Xil .Xini_2))Then xi = x i I'''" . Aini(Xil.Xini_1). . . xi+ini+1))~ in all cases for xi+11 E Ai÷11. as the sequence of variables occurring in that context...-.. .... fi(xi+1) = fi1(xi÷11 ..... and Ai+Ini+1(xi+11 ~ .. . which is to say that Ail set.

as an instance of the context at stage i = O. 1. may be interpreted as saying that the projective system of nonempty sets is itself nonempty.. It is natural to speak with Troelstra of the general case as that of a network of interdependent single one... o ain i E Ain i(a il. I given projective ~i E A i.. S. and shall continue to consider in the following. 154..° ))).. . I0 choice sequences instead of just a Now. which makes it natural to refer 10 A. I.aini as a sequence of elements ail E Ail.. i = fi ( ~ i+I ) E A i The new axioms limit of the for i = 0. the general case can be re- duced notationally to the special case that I started by considering. etc.'aini_1 ) that is.164 a i = ail . and call the extension T ~ . etc. ~ = f0(fl(f2(. op. This is a counterpart of the so called countable saturation principle of nonstandard analysis.. cit. .. extend the system T of standard type theory by adjoining the axioms for a single choice sequence that is. p. With this amount of vector notation. Troelstra.

S° Albeverio. I. Lindstrum... O(j_ I : f j _ 1 ( ~ j ) E Aj_ I. It may even have to be strictly greater.fj_1(xj). Fenstad. at some stage j = O. A judgement is provable in T ~ if and only if. J. To prove the necessity because in T ~ of the finiteness of the condition. 11 extension but also as the saturation How is provability to provability in the nonstandard extension T ~ related in the standard lemma. a judgement is provable if and only if it is provable in Tj at some stage j = O.. theory T? This question is an- swered by the following Lemma (proof theoretic). I. New York. Hoegh-Krohn.165 to T ~ not only as the nonstandard of T. etc. etc. observe first that. . 1986. Nonstandard Methods in Stochastic Analysis and Mathematical Physics. p.. where Tj is the finite extension of T obtained by adjoining the 2j+I axioms ~j e A j. and T. E. the judgement which of O~ i by each occurrence is provable in T from the assumption The stage j up to which you need to go must of course be at least as great as the maximum of the indices of the constants ~i that occur in the judgement in question. { 11 o~j_ I E Aj_ I ..)) xj E Aj. of a proof. 46. is obtained from it by replacing fi(fi+1(.. Academic Press. R.

Observe that the only rules of T. in addition to the axioms of T. so that we can first substitute ~ j for xj and then replace fi(fi+1(. Thus Tj is a definitional extension of the theory which is obtained from T by adjoining the single axiom ~ j e A jo This being the only axiom which governs the constant ~j. thereby transforming the axiom ~ j ~ Aj into the assumption xj E Aj. we have = fi(fi+1('''fj-1(~j )'. successively defining ~j-1' Cgj_2. we have access to the axiom ~ j e Aj as well as the definitional equalities ~i = fi(fi+1('''fj-1 (~j)° '')) E A i for 0 ~ i < j. and.. since T ~ Tj. provability in Tj entails provability in T ~ . Moreover. This proves the necessity of the condition.)) by cg i. in Tj. this is nothing but a sequence of explicit definitions. The sufficiency of the condition is clear. . etc.')) E Aj for 0 ~ i < j. is an extension of fj_1(~j). because Tj is an extension of T. and hence of Tj and T ~ . ultimately 6 0 in terms of ~j..166 { ~i ~0 ~0 g AO' = fO(~l ) ~ AO' But for the crucial first of these axioms.° .. Indeed. we may as well replace it by a variable xj.

Let A be a proposition expressed in the language of T. where b(xj) ~ A (xj E Aj) is provable in T already. Transfer principle (proof theoretic). Also. By means of the lemma. the following proof theoretic version of the transfer principle is easily established. a = b ( ~ j ) ~ A. . aj E Aj is provable in T. Then A can be proved to be true in T ~ if and only if A can be proved to be true in T already. It is thus immaterial exactly what the proper axioms of T are. by the lemma. by assumption. but. by substitution. The sufficiency of the condition is clear since T ~ extension of T. Hence. The proof expression a is of course is an in general nonstandard.167 that we have used in the course of the proof are the substitution and equality rules. assume that A true is provable in T ~ . so is b(aj) E A. that aEA is provable in T ~ for some a. To prove the necessity. that is.

or intended interpretation. But. in the theory. every expression is to have its usual meaning. to conclude that T ~ is consistent outright. as compared with when you specify the theory T. Applying it to the proposition ~= N O . when specifying T. whereas. we can conclude that A true is provable in T already. Let M be the standard model of the formal system T of standard type theory. which is certainly expressed in is consistent rela- the language of T. 12 Inductive limit i nterpretatio ~ The preceding proof theoretic treatment of the nonstandard extension has an exact model theoretic counterpart. that is. we can conclude that T ~ tive to T. there is the set N of natural numbers and the particular natural number O. The transfer principle is a consequence of the possibility of approximating a nonstandard proof by a standard one. among the objects in the standard model. we need to combine it with the semantic consistency proof for the standard theory T. whereas.. or point of view. there 12 P. Martin-L~f. 69-70. . op. For example. it is to be interpreted purely formalistically. as standing for itself and not for its meaning. When specifying the model ~. there is no difference in the symbols that you put down: the difference is only one of attitude.168 Suppressing the proof expression. cit. When specifying M. pp. This relative consistency proof is of a very elementary nature.

= A type in the sense of M i is a family of types over Ai. more generally. are thought of as objects rather than exin the language of T. namely. etc. and an object of the type in the sense of M i is a function defined on A i whose value for a certain argument ject of the type in the family corresponding is an ob- to that argument. of sets. fi-] Ai C fi A0 ( f0 AI ~ Ai+ I ~ fi+] the sequence .° ~ fl is that.~.. fi(xi+1) ~ A i (xi+ I E Ai+ I). nonstandard theory. in symbols. Mi M/Ai . let M i Using category theoretic be the comma model M over Ai. . that is. and exprescategories in the other sions of the corresponding case. I. The only difference mappings this time.169 is the set expression N and the numerical ference in attitude the various expression 0. a type which has been made dependent on a variable which varies over Ai. i pressions rA i set. . a i ~ Ai for i = 0. between them.. terminology and notation. and elements contained • in them. as the system of nonempty sets~ or. The nonstandard syntactical model is built over the same projective contexts. The difof is brought about by speaking of objects categories semantical in the one case.

and it becomes a model of T i by interpreting the constant o(. i+I ~ . a set in the sense of M i is a family of sets over Ai. f~ ~M/A i ~ ~ / A i + i-I z ] f.. For instance. then f~(A)(xi+ I) = A(fi(xi+1)) (xi+ I E Ai+1). When passing from A i to M/A i. Here f~ denotes composition with fi" Thus f~ takes an object in i i the sense of M i. and an element of that set in the sense of M i is a function a(x i) E A(x i) (x i e Ai). say A(x i) set (x i E A i).6A 1 simply as x i E A i (x i ~ A i). a function defined on the set A i. that is. an object in the sense of Mi+ I.. and composes it with fi' which yields a function on Ai+1. the given projective system of sets is transformed into the injective system of models 1 ~/A° ~ f o M/A] ~ ] ~ . that is. that is.. .170 For example.. as the identity function on A i. •i is clearly a nonstandard model of T. if A(x i) set (x i E A i) is a set in the sense of M i.

in our case. property of commuting with every operation. modules. then f~(a)(xi+ 1) = a(fi(xi+1)) E A(fi(xi+l)) (xi+ I ~ Ai+1). This is what we proof theoretically as a transto a since. of an inductive limit. The action of f~ on other types of objects is similar. since which is translated and thereby composition. from M i to Mi+ I. to the definition . ~ .171 which is a set in the sense of Mi+1. homomorphism corresponds model theoretically between models° Summing up. has the characteristic like T]-. etc "' f~ is a homomorphism i should expect. According which is a very rich kind of algebraic rings.l = fi ( ~ i+I ) E A i ' Thus fi translates Ti+1. and. = lim M i = lim M/A i --~ limit. we have indeed to do structures. structure Let M~ be its inductive. app. except substitution. as compared with groups. ~i every symbol of T i into the same symbol of into fi(o(i+1). or the like. which is an element of the set f~(A) in the sense of Mi+ I . if a(x i) E A(x i) (x i E A i) is an element of that set in the sense of M i. i theoretically. Now. with an inductive system of algebraic models of type theory. fm is the translation i Viewed proof from the theory T i to the theory Ti+ I given by the eqaation ~. a set or direct. what appears lation between theories.

a fam- A(X i) (xi E Ai). Other types of objects in the sense of M ~ are defined similarly. where i have put fik(Xk) = fi(fi+1(~. Likewise. for some i. provided a(fjl(Xl)) = b(fkl(Xl)) E A(fil(Xi)) (x I E A I) at some stage 1 ~ max(j. whose index k>__i may differ from j.)) E A i (xk E A k) for the sake of brevity. and it is defined to be equal to another such element.. say B(xj) (xj E Aj). provided A(fik(Xk)) = B(fjk(Xk)) (xk E Ak) for some k ~ max(i. that is. whose index j may differ from i.fk_1(Xk). and it is defined to be equal to another such set.k) >~ i. a family of sets over the . an element of the set A(x i) (xi E A i) in the sense of M ~ is a function a(xj) ~ A(fij(xj)) (xj E Aj) for some j ~ i.172 in the sense of M ~ ily of sets is a set in the sense of M i. To take yet another example..j). which we shall need in the following. say b(x k) ~ A(fik(Xk)) (xk ~ Ak)..

x e A(fil(X 1))) for some i ~ max(j.x) is a family of sets of (xj ~ Aj. in general. I have only explained how the various types are interpreted in M ~ . if a(x i) ~ N (x i ~ A i) and b(xj) e N (xj eAj) are two nonstandard natural numbers. x ~ A(fij(xj))) for some j ~ i. natural numbers in the sense of M ~ .X) (x k E A k.x) = C(fkl(X l).x) (x I e A I. with the only extra complication that. First of all.173 set A(x i) (xi E A i) in the sense of M ~ two arguments B(xj. becomes a nonstandard model of the standard theory T by letting every operation of T act pointwise. M ~ I proceed to verify that M R is a model of T ~ . x e A(fik(Xk))). since the operands begin to exist at different stages. they have to be shifted out to a common later stage before the operation can be applied. whose index k ~ i may differ from j. like in Mi. then their sum is (a + b)(x k) = a(fik(Xk)) + b(fjk(Xk)) e N (x k e A k) . that is. say C(Xk. provided B(fjl(X l). So far. and it is defined to be equal to another such family. For example.k) ~ i.

13 A. . arbitrarily contrived model: it is. cit. Hence. Robinson.174 with k = max(i. Thus M ~ = i+I. the axioms I ~ i E A i. op. the function on A 0 which is constantly equal to the standard object in question. 36-37.j).. 13 For example. a nonstandard model of the standard theory. standard theory T on the objects of M ~ mains to verify that the axioms that are proper to T ~ . pp. Clearly. ~i = fi(~i+1 ) E A i become satisfied in M ~ . of the nonstandard theory. It only renamely. is indeed a model of T ~ . The interpretation of the constant ~ i x i E A i (x i E Ai). is the identity function that govern the choice sequence ~ . to be sure. or intended interpretation.i+S) j = i+I. but it is the standard model. in order to satisfy the definitional equality ~i we must see to it that fij(xj) = fi(fi+lj(Xj)) E A i (xj ~ Aj) it suffices to take And it is not an = fi ( ~ i+I ) ~ Ai' for some j ~ max(i. namely. The action of the other operations of the is similar. This is the analogue of the star embedding of classical nonstandard analysis. Every standard object gives rise to a nonstandard object.

A(foi(ai)) On the other hand. by invoking the set A i to be nonempty. by substitution. . we can conclude. and assume that their embeddings A(x o) = A (x o E and A o) B(x o) = B (x o e A o) are equal in the sense of M ~ . and so on for other types of objects. A(foi(Xi)) that is. as an embedding should be. again by substitution. a standard element a of A to the nonstandard (x 0 E A 0) defined by the equation element of A(x 0) a(x 0) = a E A = A(x 0) (x 0 e A0). Then. let A and B be two standard sets.t75 a standard set A gives rise to the nonstandard set A(x o) = A (xo EA0). This is an embedding of the standard model M into the nonstandard model M ~ o To prove that it is injective. = B(foi(ai)). A(foi(ai)) = A and B(foi(ai)) = B. that (x i E A i) the element a i which shows that = B(foi(Xi)) at some stage i ~ O.

The proof of the injectiveness of the star embedding of other types of objects is similar. by symmetry and transitivity. Because of the definition of the standard notion of truth. such a nonstandard proposition is nonstandardly true if it has a nonstandard proof.176 so that. is a propositional or a nonstandard proposition. according to which truth is tantamount to the . The star embedding was the last arrow to be explained in the commutative diagram subtheory embedding T-intended interpretation M ~ ~ M~ >T interpretation = !ira M/A i. A=B. A(x i) prop (x i ~ Ai). a proposition in the sense of M ~ . function on some A i. Because of the identification of propositions and sets. By definition. Observe also the similarity with the proof of the model theoretic transfer principle given below. star embedding which summarizes the structure that we have erected. if there exists an a(xj) E A(fij(xj)) (xj ~ Aj) at some stage j ~ i. that is.

and suppose true in the standard sense. A(x O) true (x 0 ~ AO). Then the embedding of A into the nonstandard model is nonstandardly true if and only if A is true in the standard sense. at some stage j ~ i. let A(x O) : A (x 0 E A O) be the embedding of the standard proposition A into the nonstandard model. Then~ by weakening. To prove the sufficiency of the condition. by the principle that zruth is preserved under definitional equality. this is clearly equivalent to requiring A(fij(xj)) true (xj E Aj). But . or. Let A be a standard proposition.177 existence of a proof. it is easy enough to establish the follow- ing model theoretic version of the transfer principle. (Vxj c Aj)A(fij(xj)) true. A true (x0 E AO). if you prefer. and. Once the notion of nonstandard truth has been duly introduced. Transfer principle (model theoretic).

Hence. we can conclude A true as desired.178 f00(x0) = x 0 E A 0 (x 0 E A0). by preservation of truth under definitional equality. Conversely. that is. On the other hand. you will see that it is its exact model theoretic counterpart. If you compare this proof with the earlier proof of the proof theoretic version of the transfer principle. . again by the same principle. so that. assume that A(x 0) (x 0 E A 0) is nonstandardly true. A(f0i(xi)) true (x i 6 A i) already at stage i = 0. we get A(f0i(ai)) by substitution. true Then. which shows that A(x 0) (x 0 E A 0) is nonstandardly true. since we have a i E A i at every stage i. we get A(f0i(ai)) = A. by substituting f0i(ai) for x 0 in the definition of A(x0). that A(f0i(xi)) true (x i E A i) at some stage i ~ 0 .

which roughly amounts to working with the Fr@chet filter itself instead of an ultrafilter extending it. that a nonstandard model is standard in respect of the logical operations is tantamount to saying that nonstandard truth commutes with the logical operations. . the logical operations receive a nonstandard interpretation. which is to say.179 Commutation of nonstandard truth and the logical operations Classical nonstandard analysis is built on the highly nonconstructive existence of an ultrafilter extending the Fr@chet One filter of all cofinal subsets of the set of natural numbers° may wonder how we have been able to circumvent this in the preceding construction of the inductive limit model of nonstandard type theory. Now. And it is easier to construct a nonstandard model if it is allowed to be nonstandard throughout than if it is to be nonstandard in its interpretation of the notion of natural number and at the same time standard in its interpretation of the logical operations. to what extent. in Kripke semantics. however nonconstructive classical nonstandard analysis. The result is propositional connectives the following. for instance. in our inductive nonstandard truth com- mutes with the logical operations. speaking. The answer seems to be that. whereas. It is thus desirable to investigate limit interpretation. theory. loosely exactly how nonstandard is its interpretation of the and the quantifiers. operations is nevertheless and nonstandard is its interpretation of the logical in nonstandard type standard. as they normally do in intuitionistic model theory.

Hence A & B is nonstandardly true if and only if A is nonstandardly true and B is nonstandardly true. But it can also be checked directly as follows. general. &. The simplest way to prove that nonstandard truth commutes with conjunction is to note that.[. Of all the logical operations. By saying that nonstandard truth commutes with _[.is nonstandardly true if and only if J. they must be validated by that interpretation. Thus i false. that is. since the rule of conjunction introduction. in commutes with ]_. and 3 . A true B true A & B true and the two rules of conjunction elimination s A & B true A true A & B true B true are formally derivable in standard type theory. Let A(x i) (xi ~ A i) and . is true. and the inductive limit interpretation is a nonstandard model of it. Hence it follows directly from the model theoretic transfer principle that . I mean of course that it is absurd that J_ is nonstandardly true.180 Theorem. that J. but not with nonstandard truth V.is nonstandardly false. is indeed nonstandardly which is manifestly not the case. O. and V . But _L is a standard proposition.

181 B(xj) (Xj ~ Aj) be two nonstandard propositions. respectively. On the other hand. A(fil(fln(X n and B(fjm(fmn(Xn j~ ) true (x n ~ An). & B(fjk(Xk)) that (x k E Ak). by preservation of truth under definitional equality and . Then.m) be a common later stage. fil(f!n(Xn)) = f~_(x m )± n and fjm(fmn(Xn)) = fjn(Xn) = fjk(fkn(Xn)) EAj (x n E A n ) . = fik(fkn(Xn )) E A i ~fXn E A n ) ) true (xn E A n ) Hence. Let n = max(l. that is~ by the definition A(fil(X I ) true (x ! E A I) and B(fjm(X m ) true ( x m E Am ) at some stages I ~ i and m ~ j. Then their nonstandard conjunction is A(fik(Xk)) where k = max(i. that the two nonstandard are nonstandardly truth.j). by substitution. conjuncts of nonstandard Suppose true.

To complete the proof of the positive part of the theorem. A(fil(Xl)) true (x I E A l) with 1 ~ i . The simple way to do it is to note that the usual rule of existence introduction.j). and B(fjl(Xl)) true (x I E A l) with 1 ~ j . which shows that the two nonstandard conjuncts are both nonstandardly true. c ~ (~x ~ A)s(x) o E (3x ~ A)S(x) p(c) E A B(p(c)) true . j ) = k. a E A B(a) true ( 3 x E A)B(x) t r u e ' as well as the strong rules of existence elimination. by the definition of nonstandard truth. that A(fik(fkl(Xl))) & B(fjk(fkl(Xl))) true (x I E A l) at some stage I ~ k = max(i.m) ~ m a x ( i .182 conjunction introduction. by conjunction elimination and preservation of truth under definitional equality. that is. A(fik(fkn(Xn))) & B(fjk(fkn(Xn))) true (xn E A n ) for n = max(1. assume that this is the case. we must convince o~rselves that nonstandard truth also commutes with existence. which is to say that the non- standard conjunction of the two nonstandard propositions is nonstandardly true. Conversely. Then.

then the nonstandard existential proposition ( S x E A)B(x) is nonstandardly true. Let A(xil (xiE Ai) be a nonstandard set. x ~ A(fij(xj))) with j \$ i be a nonstandard propositional function over it. Quantifying it existentially in the sense of the nonstandard model. Hence. Conversely~ assume that (3 x E A)B(x) is nonstandardly true. this means that it has a nonstandard proof c. we get the nonstandard proposition ( ~ x E A(fij(xj)))B(xj. and let B(xj.x) Let a(x k) E A(fik(Xk)) (x k E A k) with k ~ i be a nonstandard element of the nonstandard set A. Then B(a) in the sense of the nonstandard model is the nonstandard proposition . if B(a) is nonstandardly true for a nonstandard element a of the nonstandard set A. so that they must be validated by the nonstandard model.x) (xj E Aj. The more laborious direct verification proceeds as follows. (xj E Aj). By the definition of nonstandard truth. Thus nonstandard truth commutes with existence. Then the left projection p(c) of that proof in the sense of the nonstandard model is a nonstandard element of the nonstandard set A such that the nonstandard proposition B(p(c)) is nonstandardly true.183 are all derivable in standard type theory.

Assume it to be true. k ) ~ j. assume B(fjm(Xm). we can conclude from this that P(C(Xk)) E A(fik(Xk)) and B(fjk(Xk). (xk ~ A k) Thus we have found a nonstandard element of the nonstandard set A(xi) (xi ~ Ai) which satisfies the nonstandard propositional function B(xj. that is.X) (x k E A k) By the strong rules of existence elimination.a(fkm(Xm))) true (xm E Am ) at some stage m ~ 1. that is. I shall make use of the projective system This finishes the .a(fkl(Xl))) (x I E AI). where 1 = max(j.X) true ( x m ~ = m a x ( j . To prove the negative part of the theorem. which shows that the nonstandard exisConversely. according to the definition of the notion of nonstandard truth. assume tential proposition is nonstandardly true° that the nonstandard existential proposition is nonstandardly true.184 B(fjl(Xl). (Sx for m ~ l E A(fim(Xm)))B(fjm(Xm). that it has a nonstandard proof c(x k) E ( ~ x with k ~ j ~i.P(C(Xk))) true (x k E Ak).x) (xj ~ Aj. x ~ A(fij(xj)))o proof of the positive part of the theorem. E A(fik(Xk)))B(fjk(Xk). by standard existence introduction. Then.k).

It is nonstandardly true provided ( V x 0 ~ Ao)(A(x O) B(xo)) is true in the standard sense. consider the nonstandard implicative proposition A(?) ~ B(?). ~ id AO 4~---. since the implication in the judgement (Vx 0 E A0)A(x0) V ( ~ x 0 ~ Ao)B(x0) 2:) (Vx 0 ~ A0)(A(x 0) V B(x0)) true cannot be reversed. Hence. let A(x O) and B(x O) be two standard propositional functions of the variable x 0 E A O. Then the nonstandard proposition A(?) is nonstandardly true provided A(x O) is true for x 0 ~ A O. Now. a set containing an element a 0 E A O. in general.. Furthermore. But the outermost im- . that is. ( V x 0 ~ AO)A(x O) is true. It is nonstandardly true provided ( V x 0 E A O) (A(Xo) V B(Xo)) is true in the standard sense. Let ? = ?o = id(?1)' ?1 = id(?2).A 0 id id where A 0 is a fixed nonempty set. nonstandard truth does not commute with disjunction. Another example of the failure of nonstandard truth to commute with disjunction will be given later. Likewise. B(?) is nonstandardly true provided ( V x 0 E Ao)B(x O) is true in the standard sense. consider the nonstandard disjunctive proposition A(?) V B(?).185 A0 < id ]IIA 0 ( id . Next. or. equivalently.. ?2 = id(?3)' be the choice sequence that it defines. in the standard sense.

and B(?. both in the standard sense.app(z.186 plication in the judgement ( ~ x o ~ A0)(A(x o) ~ B(x0)) ( ( ~ x 0 E A0)A(x O) ~ ( V x 0 ~ A0)B(Xo)) true cannot be r e v e r s e d . Hence the nonstandard truth of ( V x g A(?)) B(?. and the proof of the theorem is finished. it is nonstandardly true provided ( V x 0 ~ A o ) ( ~ x E A(Xo)) B(x0. and B(Xo. the nonstandard truth of A(?) entails the nonstandard truth of B(?). By the definition of nonstandard truth.x). Hence. an arbitrary nonstandard element of A(?) is of the form a(?).X) ( ~ z E (]~x 0 ~ A 0 ) A ( x 0 ) ) ( ~ x 0 E A0)B(x0.x) entails the nonstandard truth of B(?. where a(x 0) ~ A(x O) for x 0 E A 0 is a standard function. and consider the nonstandard proposition ( ~ x E A(?))B(?. Thus nonstandard truth fails to commute with universal quantification. On the other hand. Finally.x0)) true cannot be reversed. since the implication in the judgement (Vx o ~ Ao)(Vx ~ A(xol)~(Xo.x) is true in the standard sense.a(?)) for all nonstandard elements a(?) of A(?).a(x0)) is true in the standard sense. . i f A(?) ~ B ( ? ) i s non- standardly true. in general.X) a propositional function of the two variables x 0 E A 0 and x E A(x0).a(?)) is nonst andardly true provided ( V x 0 E Ao)B(x0. that is. i n g e n e r a l . let A(x 0) be a set depending on the variable x 0 E Ao. but not conversely. but not conversely. which shows that nonstandard truth fails to commute with implication. a family of sets over AO.

Make the pair of definitions I Hence. but a few examples may help to give an idea of what can be done with the nonstandard concepts. .y + ]). Then the judgement Even(x) V Odd(x) true (x E N) is easily proved by induction on x in standard type theory. substituting ~ for x. In classical nonstandard analysis.. do we have Even(si(x)) nor do we have 0dd(si(x)) true (x E N). Odd(a) = ( 3 y ~ N)I(N. which is to say that the two nonstandard propositions E v e n ( m ) and O d d ( ~ ) are both nonstandardly false.y). I.a.2. true (x ~ N).a.2. We have here another. Even(w) in the nonstandard extension° Semantically. Example I. V 0dd(~) nonstandard proposition E v e n ( ~ ) On the other hand. at no stage i = 0. etc. where a E N.I87 Miscellaneous examples This is not the place to begin a systematic development of intuitionistic nonstandard analysis. we can derive V Odd(~) true this means that the is nonstandardly true. Even(a) = (7~y E N)I(N. more memorable example of the failure of nonstandard truth to commute with disjunction.

so that we can pass to A(si(O) + x) true (x E N) . whose non- standard truth entails the nonstandard truth of A(a) for all infinite a ~ N. If we define addition of natural numbers by recursion on the first argument. (a) + b = s(a + b) ~ N. Let A(x) (x ~ N) be a standard property of natural numbers. this means that A(si(x)) true (x ~ N) at some stage i = 0. Then A ( ~ ) is a nonstandard proposition. version of nonstandard analysis.t88 one of the two nonstandard propositions Even(~) and 0 d d ( ~ ) is forced to be true. By the definition of nonstandard truth. I~ +b =b~N. arbitrary. This is made possible by the nonstandard intuitionistic interpretation of disjunction in the present. which one of them depends on the choice of the ultrafilter extending the Fr~chet filter of cofinal subsets of the set of natural numbers. I. we h a v e si(x) = si(o) + x ~ N (x ~ N ) . assume A(~) true. etc. Example 2. To see this. and 0 d d ( ~ ) Since this choice is anyway left it seems more natural not to force any one of E v e n ( ~ ) to be true.

y) D A(y)) = ( V y E N)(si(0) ~ y D A(y)) = (~y~si(O))A(y) by d e f i n i t i o n . in particular.si(0) + x. etc.y) D A(y)) true.si(0) by 1-elimination tionistic (~y But and + x. si(0) ~ a true.y) D A(y) true (y E N) D-introduction. equality. Since the standard ard interpretation. a g a i n by p r e s e r v a t i o n of tr~tfl under (~/y~ s±(O))A(y) Now. in the sense of nonstand- infinite ard arithmetic. definitional so t h a t . let a E N be an arbitrary true.. logical laws continue we can then conclude A(a) true to hold in the nonstand- . From this. predicate ~ N)((3x (Vy ~ N)((3x ~ N)I(N.si(0) + x.189 by preservation we get I(N. I. which is to say that sJ(0) ~ a true for all j = 0. logic now yield ~ N)I(N. so that. Standard rules of intuiof truth under definitional equality.

etc. etc. etc. that. Example 4. Thus the nonstandard truth of A ( ~ ) entails that of A(a) for an arbitrary infinite a ~ N.. for all j = 0. in turn.. etc. there exists a stage k = 0. that sJ(o) ~ f ( ~ i ) true for all j = 0. I. By the definition of nonstandard truth. On the other hand. e(s(x)) = ~P(x). where ) ~ N f ( x ) mN (x ~ N ) is a standard number theoretic function. Define a propositional function P(x) of the variable x ~ N by the recursion equations f P(o) = J_. I. Example 3. such that sJ(0) ~ f(sk(x)) true (x E N)o under a Thus an infinite natural number is the image of an ~ i standard number theoretic function which grows beyond all bounds.. This is easily done in standard type theory by means of the . a = f(~i for some i = 0. that a = f ( ~ i ) E N is infinite in the sense of nonstandard arithmetic means. but what does an arbitrary infinite a E N look like? By the definition of the nonstandard model. I. this means. by definition.190 by ~ - and ~-elimination. I. We already know that each ~ i is infinite.

I shall construct a nonstandard element of List(N) which produces the stream of integers (O. Thus. etc. the definition P(x) = T(rec(x. Since we have ~ is a nonstandard proposition. Example 5.(s(O).y)~y)) E N in the nonstandard extension.(s(s(O)). I. etc. this means that p(si(x)) true (x ~ N) and at some stage i = O. Thus P ( ~ ) = P(s(si(o))) and N p ( s i ( O ) ) = Np(si(o)). This being so. which is im- is nonstandardly false.(x. tion is defined by the equation -~A = A ~ J _ .. Is P ( ~ ) appears as an endless nonstandardly true or non- standardly false? Suppose it to be nonstandardly true. I. Indeed.. By definition. nega- and we already know that nonstandard truth fails to commute with implication. P ( ~ i ) is actually nonstandardly false for all i = 0. although P ( ~ i ) = P ( s ( ~ i + 1 ) ) = -vP(~i+1).))) .191 universe axioms. which comes as no surprise. would be true. and P(~) will do. P ( ~ ) sequence of negation signs... There is no contra- diction in this: it only shows that nonstandard truth fails to commute with negation. when it is evaluated lazily. In fact. But p(si(s(O))) so that both p(si(O)) possible..J. This entails that both p(si(O)) p(si(s(O))) are true at that stage.

o) = f(s(~]).O) in the nonstandard extension. it reduces to the primitive recursion g(O) = ( A n ) n i l e N--~List(N).o) = (O. putting E List(N). s ( n ) ) ) ) N-~List(N) in standard type theory. which has the solution g(1) = r e c ( l .S(O))) = (O.s(n))) e N-~List(N). This is a double recursion. define the standard function f(1.s(O))) = (O. a p P ( y .n). To this end.f(~l.n) ~ List(N) (I ~ N. We can now use the axiom ~ derive f(~.(s(O). = (~ n)(n.f(~2.n) = app(g(1). ( ~ n ) n i ! . O ) lazily. f(l.app(g(1). ( x .f(s(~2).192 when it is evaluated lazily.n) = nil @ List(N).n) = (n. we get EN to f(~.f(l. . y ) ( k n ) ( n . . ~ e N ) by the pair of equations I g(s(1)) f(O. . . f(s(1).s(s(o))))) . 6 List(N) Evaluating f ( ~ .s(n))) but.

g(1.x) E A(n) (n E N.s(n))) EA(n). May 1987. 14 Example 6. 14 The idea of dealing with stream computation by introducing an infinite number is independently due to S. I. Honolulu. He proposes to apply Robinson's nonstandard analysis to interpret number theory extended by a new constant for the infinite number. f(n. have such a sequence internally without running into contradiction. suppose in the sense of the nonstandard model. (We cannot. . Goto. US-Japan Workshop. n ~ f f ) I g(O. etc.n) = a(n) E A(n). Nonstandard normalization.) To see this. x ~ A(s(n))). g(s(1). Let a projective system of nonempty sets of the sort that underlies the inductive limit interpretation be given internally. externally indexed sequence of elements ~i which satisfy the equations e A(si(O)) I A(n) set (n ~ N). define first an auxiliary function g(1. that is.n) EA(n) by the equations (l~ N.n) = f(n.193 Thus the stream of integers is produced. in general. a(n) ~ A(n) (n E N) Then we can construct an ~'l = f(si(O)'° (i+ 1) e A(si(O)) for i = O.

s(si(O)))) = f(si(O). This example explains the canonical character of the choice sequence o@ = s(o@1) = s(s(o@2)) = etc° .n) = app(h(1).si(O)) = f(si(O).(x.194 This is a double recursion. because.s(n)))) (Tin We can now put i = g ( ~ i 'si(O)) e A(si(O)) N)A(n). etc.g(~i+1. ~ (ITn ~ N)A(n).app(y.n) E A(n). This is an external sequence which satisfies ~i = g ( ~ i 'si(O)) = g(s(~i+1).(~n)a(n). we can define any other choice sequence as- . ~i+I ) E A(si(O)) as desired. where h(1) e (TEn E N)A(n) is defined by the primitive recursion (i ~ N) l h(0) = ( ~ n ) a ( n ) ~ (]Tn ~ N)A(n).g(~i+1.app(h(1). I.y)(kn)f(n.s(n))) which has the solution h(1) = rec(1. h(s(1)) = (kn)f(n.si+1(O))) = f(si(O). once we have access to it. for i = O. but it is readily solved in standard type theory by putting g(l.

course. . with respect to the usual uniform distribution. integral of a function I shall show how to define the f E B(c. it suffices to put B(n) = T(rec(n.y)(y + y))). provided only that it is given internally. B(s(n)) = B(r. 65. which is as nonstandard as the set to which it belongs. cit.. p. Consider now the nonstandard set B ( ~ ) . Albeverio et al.) + B(n). tions It satisfies the equa(n E N) B(~) = 3(s(~I)) + B ( ~ I) = B(s(oo2)) + B ( s ( ~ 2 ) ) = B(~]) = (B(~ 2) + B(~2)) + ( B ( ~ 2) + B(~2)) Thus B ( ~ ) can be endlessly divided into two equal halves: it is a nonstandard version of the Cantor space. Define by recursion a family of sets B(n) satisfying the equations B(0) = NI. Example 7..n 1. op.195 sociated with a projective system of nonempty sets.~ Q. The universe axioms allow you to do this in standard type theory. Indeed.(x.o)-. 15 Let Q denote the standard set of rational numbers. see S.. of To this 15 For a nonstandard version of the Cantor space in classical nonstandard analysis.

f) by the total number of elements of the set B(n).f) : a p p ( f .196 end. where F(n) E (B(n)--~Q)--~Q (n ~ N) is defined by the primitive recursion I F(O) = ( k f ) a p p ( f . These are e a s i l y s o l v e d i n s t a n d a r d type t h e o r y by p u t t i n g S(n.f) E Q (n E N. F(s(n)) = (kf)(app(F(n). define the sequence of sums S(n.j(y)))).(~y)app(f.(~y)app(f. 0 1 ) . f E B ( n ) . f E B(n)~-~Q) is defined by putting I(n. We can now express the searched for integral of f E B(~)--~Q simply as .f) : s(n~f) 2n ' that is.i(x))) + app(F(n). by dividing S(n.(kx)app(f.i(x))) + S(n.f).f) = S(n.f) = app(F(n). 0 1 ) .~ Q ) by the recursive equations S(0. S(s(n).j(y))). The analogue of the sequence of Riemann sums I(n.(kx)app(f.f) E Q (n E N.

.g(~i)) which is the value at ~ i numbers of the standard l(si(x). At stage j = i.g(sJ-i(0))) Convergence is another matter. exist from stags i approxima- •his means that the rational approximations and onwards. model. I(~. i+I. where g(x) E B(si(x))--~Q is a standard function. the rational the program E q. the meaning of of the nonstandard --~Q is that f = g(~i ) E B(sl(~i))-~Q = B(~)--~Q.f) = s. sequence of rational (x E N) : I(~.f) Hence = I(si(~i). Let us see what the se- approximations is of which it is the limit. .(~f) 2~ This is a nonstandard quence of rational By the definition f E B(~) rational number. tion is obtained by running l(sJ(0).197 I(~. etc.g(~i)).g(x)) E Q (x E N).