Bishop Constructive Analysis
Bishop Constructive Analysis
Bishop Constructive Analysis
BOOK REVIEWS
i97o]
Foundations of constructive analysis by Errett Bishop. McGrawHill, New York, 1967. xiii + 370 pp. $12.00.
For, compared with the immense expanse of modern mathematics, what would the wretched remnants mean, the few isolated
results, incomplete and unrelated, t h a t the intuitionists have
obtained. . .
(Hilbert, 1927J1
While in a few cases one has succeeded in replacing certain
intuitionistically void proofs by constructive ones, for the majority
this has not been achieved nor is there a prospect of achieving it. . .
(Fraenkel & Bar-Hillel, 1958) 2
L'cole intuitionniste, dont le souvenir n'est sans doute destin
a subsister qu' titre de curiosit historique. . . (Bourbaki, I960) 3
Almost every conceivable type of resistance has been offered to
a straightforward realistic treatment of mathematics. . . . I t is
time to make the attempt.
(Bishop, 1967) 4
302
BOOK REVIEWS
[March
i97o]
BOOK REVIEWS
303
to be equal if their termwise differences are at most 2/n. A real number is positive if, for some n, xn>l/n
and nonnegative if, for all n,
xn 1/w. One could also make a definition of real numbers in terms
of Dedekind cuts or arbitrary Cauchy sequences of rationals. Classically this is more elegant, but constructively it is less elegant. (Constructively, a Cauchy sequence of rationals is a sequence (xn) of
rationals and a sequence (Nk) of integers such that \xn xm\ ^1/fe
for n,
m^Nk.)
Of the remaining chapters, one is a fairly standard treatment of
Lebesgue integration (abstract measure spaces, convergence theorems), and this provides a natural setting for the study of the classical discontinuous functions. Three chapters are somewhat more
specialized: locally compact Abelian groupsthis is quite elegant
(Haar integral, convolution, Fourier inversion, and Pontryagin
duality), commutative Banach algebras (according to the author, "the
only instance in this book of a classical theory whose constructive
version seems forced and unnatural"), and limit operations in measure
theory (containing a new general ergodic theorem, in terms of upcrossing inequalities, which yields constructive versions of Doob's
martingale theorem and Lebesgue's theorem that a function of
bounded variation has a derivative almost everywhere).
Finally, there are two appendices. One is a philosophical addendum
on the role of contradiction and on the computational meaning of
the mathematics of the book. The latter subject is pursued much
further in Bishop's recent essay Mathematics as a numerical language.6
The other appendix is a brief defense of the author's nearly exclusive restriction to metric spaces and his free use of separability
hypotheses. His conclusion is that, at least for the parts of analysis
treated in his book, this is the right setting. There are no constructively defined metric spaces which are known to be nonseparable and,
indeed, all the spaces which arise naturally are separable. (The
metric induced classically by the supnorm on l^ is not constructively
everywhere well defined!) Although the concept of a uniform space,
defined by a set of pseudometrics, would appear promising, Bishop
finds that even those uniform structures naturally associated with
important locally convex spaces are not too significant constructively. For the dual of a separable Banach space a "double-norm"
works better; for spaces of distributions there are other considerations.
6
Proceedings of a Symposium on Intuitionism and Proof Theory, North-Holland,
Amsterdam (to appear).
304
BOOK REVIEWS
[March
i97o]
BOOK REVIEWS
305
tions in my sense and require the logical c-axiom7 for their development.
(Hubert, 1927)1
The classical foundations of calculus, all the more the modern
theory of real functions, including the Lebesgue integral, clearly
become meaningless in this light. (Fraenkel & Bar-Hillel, 1958)2
306
BOOK REVIEWS
[March
i97o]
BOOK REVIEWS
307
308
BOOK REVIEWS
[March
i97o]
BOOK REVIEWS
309
310
BOOK REVIEWS
[March
BOOK REVIEWS
197)
311
sets and more functions, we get the basic structures of mathematics: the rational number system, the real number system, . . . ,
the algebraic number fields, Hubert space,. . . , and so forth. , . .
Everything attaches itself to number, and every mathematical
statement ultimately expresses the fact t h a t if we perform certain
computations within the set of positive integers, we shall get certain results.
Mathematics takes another leap, from the entity which is
constructed in fact to the entity whose construction is hypothetical.
To some extent hypothetical entities are present from the start:
whenever we assert t h a t every positive integer has a certain
property, in essense, we are considering a positive integer whose
construction is hypothetical. But now we become bolder and consider a hypothetical set, endowed with hypothetical operations
subject to certain axioms. In this way we introduce such structures
as topological spaces, groups, and manifolds. The motivation for
doing this comes from the study of concretely constructed examples, and the justification comes from the possibility of applying
the theory of the hypothetical structure to the study of more than
one specific example . . . even the most abstract mathematical
statement has a computational basis.
This is the basis for most write-offs of constructivism, and some of Brouwer's
earlier remarks may have unfortunately contributed to this belief. For instance, in
Intuitionism and formalism, this Bulletin, 1913, he wrote, "the formalist introduces
various concepts, entirely meaningless to the intuitionist, such as for instance 'the
set whose elements are the points of space,' 'the set whose elements are the continuous
functions of a variable' . . . and so forth."
312
BOOK REVIEWS
[March
this does not define an integer. But it does define a subset 16 of {5, 7}
containing at most one integer.
Of course, classically, we are in the same kind of situation when
we consider the set of solutions y y{x) of a specific functional equat i o n / ^ , y) = 0 without having any particular solution at hand, and
even without knowing if there are any. 17 Likewise, to define a set
constructively means only to state precisely what must be accomplished in order to construct an element, and what else must be
accomplished in order that two given elements be equal. We are not
required, in defining a set, to have any way of constructing elements
or any way of deciding if two elements are equal. By a way we mean
here a finitely performable procedure. Sometimes they are at hand;
but in other cases finding them will constitute a major mathematical
problem.
By contrast with the ease in defining sets constructively, it is hard
to construct functions. The definition of a function from a set A to a
set B must provide an explicit way of converting the construction of
any element of A into the construction of a definite element of B.
(This much describes the concept of an operation. To define a function we must also provide a proof that equal members of A are converted into equal members of B.) We can say more. To be complete
the definition of the function must also come with a verification that
it is finitely performable. This verification may make use of other
functions that have already been shown to be well defined, starting
off, when all is spelled out, with the finiteness of the specific integers.
As an illustration, let us consider what must be done to define
constructively a real-valued function on the closed interval, [0, l ] ,
in terms of whatever must be done to define a function from the positive integers to the integers, i.e. in terms of sequences of integers.
For the sake of the present discussion, let us try to be very explicit.
It is no restriction to consider only real numbers of the form
x==(x(w)/2 n ), where (x(n)) is a constructively defined sequence of
integers and, for each n, x(n + l) differs from 2x(n) by at most 1. The
relation of equality xy between reals is then expressed by the condition that, for all n, | x(n) y{n) | rg 2. The requirement that x belong
to [0, l ] means that, for all n, 1 Sx(n) ^ 2 n + l . Therefore, we see
that the definition of any function : [0,1 ]>R must supply, firstly, a
16
1970]
BOOK REVIEWS
313
finitely performable method for converting each definition of a sequence of integers (x(n)) that satisfies (a) for each nf | x(n+l) 2x(n) \
rgl, and (b) 1 ^x(n) ^ 2 n + l, into the definition of another specific
sequence of integers (ux(n)) that satisfies \ux(n + l) 2ux(n)\ ^ 1 ,
and, secondlyy a proof that if, for all n, \x{n)y(n)\ ^ 2 then, for all
n, \ux(n)Uy{n)\ ^ 2 .
To go further and define a uniformly continuous function we would
have to provide a modulus of continuity. This amounts to defining a
sequence of positive integers (k(n)) and giving a proof that, for all n,
if \x(k{n)) y(k(n))\ ^ 2 then \ux(ri)uy(n)\ g 2 . We note in passing
that for each specific n this will be a direct check.
I have taken such exceptional pains to spell everything out
though I could have gone even furtherto counter the common
belief that constructive mathematics must be vague and imprecise,
because itin particular, the concept of a constructionis not
formalized in any way. This belief is shared by the Russian school of
constructivists, and others, who feel it is necessary to base the concept
of a construction on that of a recursive function. (One effect of doing
this is a radical change in the way the mathematics looks.) Sanin, in
On the constructive interpretation of mathematical judgments1* writes
that constructive mathematics "began to be developed successfully
only in the middle of the 1930's after the precise mathematical concept of arithmetic algorithm {computable arithmetic f unction) had been
worked out. Only the introduction into mathematics of the precise
notion of arithmetic algorithm created a satisfactory basis for the
treatment of the constructive interpretation of mathematical propositions and fundamental notions of constructive mathematical
analysis." I believe that Bishop's work effectively refutes the underlying assumption here. The undefined concept of a construction
actually admits a usage no less precise and clear than the undefined
concept of an integer. This remarkable fact can only be obscured by
bringing in recursive functions at this level.
Extramathematical observations. In light of the above discussion,
it should not be a t all surprising that Brouwer, after carefully considering how the definition of a function/: [O, l]>R could possibly
be given, became convinced that one would always be able to extract from it a modulus of continuity. Indeed, the status of this
observation is rather like that of Church's thesis. On the one hand,
18
In Russian (1958). Translated in Amer. Math. Soc. Transi., (2), 23 (1963),
109-189.
314
BOOK REVIEWS
[March
i97o]
BOOK REVIEWS
315
316
BOOK REVIEWS
[March
1970]
BOOK REVIEWS
317
318
BOOK REVIEWS
[March
Nevertheless, constructive mathematics starts off with the Kantian, indeed, ancient, observation that mathematics has content and
meaning, independent of logical considerations. For instance, the
principle of mathematical induction is correct simply by virtue of its
meaning. For the same reason, the assertion that 0 = 1 is incorrect.
In fact, every constructively proved theorem (i.e. every complete
theorem) is correct precisely on account of its meaning.
Moreover, as we have seen, the way constructive mathematics is
directed toward predictive and descriptive assertions suggests that
any incomplete statement should be interpreted as meaning, first of
all, that it has a completion (gotten by actually constructing the
kind of object that is asserted to exist). From this standpoint, a
formally derived assertion of existence, even together with its formal
proof, but without a construction, cannot even be regarded as answering "yes" to the question of whether such an object exists. This
is not at all to ignore the fact that such a formal proof of existence
will, nearly always, provide a verification (i.e. a completion) of some
other incomplete statement which is logically equivalent with the
original one: perhaps its double negation or, in the case of an implication, its contrapositive. But, as one can see quite vividly from
simple examples, this is a different assertion, usually with a very
different meaning. For instance, if we regard the uniform boundedness
principle as asserting the existence of a certain real number, an upper
bound for the set of norms of some family of operators on a Banach
space, then its contrapositive asserts the existence of a point in that
space, having certain properties.
Of course, classically, even if we recognize these differences in
meaning, we must admit the truth of any incomplete assertion which
is shown to be logically equivalent with some other complete (i.e.
constructively proved) theorem. In fact, this is generally regarded as
a very powerful nonconstructive way of correctly guessing the truth
of existence statements, a way that enables us to concentrate on the
pure form and structure of mathematics by separating out the fact
of existence from the business of, somehow, actually effecting a
construction.
Already Kant 2 4 (1781) had, in another context, called attention to
23
i97o]
BOOK REVIEWS
319
320
BOOK REVIEWS
[March
Brouwers observations influenced Hubert's program of formalization in more ways than one. The original concern of Hubert was to
provide a secure framework for Cantorian mathematics, especially
the theory of transfinite numbers, free from the danger of paradoxes.
His scheme, derived from his earlier work on the relative consistency
of geometry, was to construct, from the most primitive objects of
thought, an accurate model of classical mathematics: one whose
consistency would be determined by the most intuitive considerations, though not necessarily while standing on one leg. But Brouwer's
critique, of conceded validity, posed for the formalists a new question
far more serious than that of consistency. Namely, what is the point
On the principle of excluded middle.1
On the infinite.1
Comments on Hubert's second lecture on the foundations
of mathematics}
1970]
BOOK REVIEWS
321
Plus its radical point of view and aspects of the way it was presented.
Intuitionism, an introduction, North-Holland, Amsterdam.
322
BOOK REVIEWS
[March
The moral of this story is not the relatively boring fact that the
classical system can now be "filled out," but that it is time to turn
back to a systematic and realistic consideration of the meaning of
33
There are also metamathematical results along these lines going back to
Kolmogorov, 28 1925, based on comparing a formal model of a portion of constructive
mathematics with the model of classical mathematics gotten by adding to it t h e
principle of excluded middle.
34
This raises the question of exploring the true domain of validity of the principle
of excluded middle. There is a conjecture that by means of a metaconstruction one
can constructivize any classical proof of an existence statement of the form VxETy,
A(x, y) (with each A(x, y) verifiable) so long as that proof can be formalized in
Gdel's formal constructive system with excluded middle adjoined. This is based on
empirical observation t h a t such classical theorems tend to admit completions and on
Spector's important work, "Provably recursive functional of analysis," in Recursive
function theory, Proc. Sympos. in Pure Math., vol. 5, Amer. Math. S o c , Providence,
R. I., 1962.
i97o]
323
BOOK REVIEWS
our mathematics, and on this basis finally begin to realize the rich
promise of a truly Kroneckerian development. By insisting 36 on
admitting the principle of excluded middle, regardless of its meaning,
and thereby abandoning all those natural explanations and concepts
based directly on meaning, classical mathematics took a step from
reality, and not into paradise. 36
T h e successful formalization of mathematics helped keep
mathematics on a wrong course. The fact t h a t space has been
arithmetized loses much of its significance if space, number, and
everything else are fitted into a matrix of idealism where even the
positive integers have an ambiguous computational existence. . . . it
took the full flowering of formalism to kill the insight into the
nature of mathematics which its arithmetization could have given.34
Really, the only way the classical mathematician can judge for
himself about the truth of what is said here is by stepping outside his
systemthis is not easy to do!and then comparing what classical
and constructive mathematics have to say about the phenomena and
structure of the one underlying mathematics. I believe that by doing
this he will discover for himself Bishop's "secret still on the point of
being blabbed". 37
G A B R I E L STOLZENBERG
86
According to Hubert 1 "No one, though he speak with tongues of angels, will
keep people from negating arbitrary assertions, forming partial judgments, or using
the principle of excluded middle."
88
Thus difficulties stemming from the fictitious existence and identity classically
attributed to the elements of any set soon forced abandoning the real Cantorian set
theory in favor of much less natural, and still unsatisfactory, axiomatics and formalistics. Yet, constructively, a sharpened version of the set concept is given the freest
play. It is quite revealing, though beyond the scope of this exposition, to pursue
further the contrasting classical and constructive standpoints on such topics as
countability, uncountability, decidability, formal undecidability, consistency proofs,
the role of formal systems, the structure of the line, Cantor's theory of ordinals and
cardinals (constructively, the line and the plane are not equipotentbecause they
are not homeomorphic), the continuum hypothesis (constructively not valid), and
the axiom of choice (a choice operation is always available constructively, though not
always a choice j'unction).
87
This is the epigraph of Bishop's book, taken from Lascelles Abercrombie's
Emblems of love, the Bodley Head, Ltd. (John Lane), London.