What are Set Functions?

Author(s): Arthur Rosenthal

Source: The American Mathematical Monthly, Vol. 55, No. 1 (Jan., 1948), pp. 14-20
Published by: Taylor & Francis, Ltd. on behalf of the Mathematical Association of America
Stable URL: https://www.jstor.org/stable/2305320
Accessed: 17-10-2018 08:28 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Mathematical Association of America, Taylor & Francis, Ltd. are collaborating with
JSTOR to digitize, preserve and extend access to The American Mathematical Monthly

This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms
This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms
 1948] WHAT ARE SET FUNCTIONS? 15

discussion of set functions will be found in the forthcoming book of H. Hahn

and A. Rosenthal [3].
Further generalizations of basic notions are possible, as for the range of val-
ues of set functions. Instead of numbers one can take points of a space (for in-
stance, of Banach spaces) or, even more generally, sets as values of the function.
In the case of this last generalization, where sets correspond to sets, one arrives
at the notion of mapping one space into another space. We shall not go so far
here, but shall consider only single-valued set functions whose range of values
consists of real numbers (including + oo and - oo).
The notion of set functions becomes fruitful by the introduction of some spe-
cific properties, such as additivity.

2. Additive set functions. Let 9J1 be a system of sets in a quite general space,
and let the set function +(X) be defined for each set X that is a member of 9N,
denoted symbolically by XEd. From the first we shall exclude the case that
(X) = + oo for all XEd or that 4(X) = - o for all XE9. Then 4 is called addi-
tive in 9 if for any two disjoint sets AdN and BEd, for which also their sum
(A +B)EW, we have

(1) 0 (A + B) = (A) + 0 (B).

Here O(A) and 4(B) cannot have infinite val
wise the right side of (1) would be meaningless.
EXAMPLES: Let 9J be the system of all subsets of a given infinite set E.
(a) For every AdN let 4(A) be the number of elements of A. (In particular,
let 'k(A) = + oo if A is an infinite set.)
(b) For every A-EW let 4(A) be one plus the number of elements of A. (In par-
ticular, let c(A) = + oo again if A is an infinite set.)
(c) Let c (A) = 0 if A is finite, and 4 (A) = + oo if A is infinite.
Then 4 is additive in the Examples (a) and (c), while 4 is non-additive in the
Example (b).
Let 4 be additive in 9, and let all sets discussed belong to ). In studying 4
it will be suitable to assume that 9 is a field; that is, with any two sets AeS9)
and Be9, their sum A +B and their difference A -B also belong to 9. It can
easily be seen also that their intersection A *;B belongs to V.
Of course, by induction we obtain immediately from (1):

o(A-1 + A2 + * + An) = 4(A1) + O(A2) + + (An)

provided that the sets Ai, (i = 1, 2, * * * , n), are disjoint.
For two sets which are not necessarily disjoint one obtains:
(2) +(A) + +(B) = 4(A + B) + 4(A * B).

The demonstration of this follows from the fact that A = A B + (A -B) and
B=A.B+(B-A); thus +5(A)=O(A B)+O(A-B) and +(B) =J=(A JB)
+4 (;B -A). Hence, by addition, +)(A) ++(B) =4)(A ;B) + [4(A *B) +4(A -B) +
4)(B -A) ] =O(A * B) +k(A +B).

This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms
 16 WHAT ARE SET FUNCTIONS? [January,

3. Totally additive set functions. The set function 4, additive in 9), is called
totally additive (or completely additive) in 9) if for every sequence of disjoint sets
A,cE9)(v = 1, 2, ***), for which also their sum SA,e9, we have

(3) 4(SA,)= 2(A,).

V

In Example (a) of ?2, 4 is a totally additive set function. But, in Example

(c), 4) is an additive set function which is not totally additive; for if B =Sb, is
any denumerable subset of E, we have 4)(b,) = 0, but 4 (B) + oo .
More significant examples are the Lebesgue measure, which is totally addi-
tive in the system of measurable sets, and the Jordan- content which is additive,
but not totally additive, in the system of the sets measurable in the sense of
Jordan.
For the study of totally additive set functions, it is natural to assume 9) to
be a a-field, that is, a field which along with each sequence of sets { A, } also con-
tains their sum SA,. Now let 4) be totally additive in the a-field 9), and let all
sets discussed belong to 9). Then the following theorems can easily be proved:

THEOREM (a). Let {A,} be a monotone increasing sequence of sets (that is

A,1(AP2 for V1_V2), then 4)(SA,) =lim 4)(A,).
THEOREM (b). Let {A,} be a monotone decreasing sequence of sets (that is,
A 1;A,2 for V1 v2). If not all 4)(A,) are infinite, then 4)(DA,) =lim 4)(A,), where
DA, designates the intersection of the sequence {A,}.

THEOREM (C). If 4) is monotone increasing (that is, A CB implies 4 (A) ? +,(B)),

then we have for every sequence {A, } of sets

(4) 4(SA,) ? E2'(A,).

p
The following important result is due to H. Hahn [41:

THEOREM (d). The set function 4) attains the absolute maximum and the ab-
solute minimum of its values.
From (d) it follows immediately that 4) is bounded either above or below
(since 4) cannot attain both the values + oo and - oo); and, moreover, if 4) is
finite, then 4) is also bounded.
We can reduce 4) to monotone increasing functions by defining the positive-
function 4+ and the negative-function 4- of 4 in the following way.
For AE9), we set

(5) ++(A) = sup 4(X) and 0)(A) =-inf 4(X),

where X designates those subsets of A which belong to 91. Then we have

(6) , = 0+ g,_
Besides, the absolute-function f is defined by

(7) = 0+ + 0-,

This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms
 19481 WHAT ARE SET FUNCTIONS? 17

The functions 0+, 4), s3 are monot

tally additive. The reduction to monotone increasing functions, given by (6), is
very useful for the proofs of many theorems.

4. Zero-sets for 4. A set AeWI is called a zero-set for 4) if q5(X) = 0 for all
subsets X of A which belong to W9. If 4) is totally additive in the a--field 9), then
the system of all zero-sets for 4 is also a c-field.
WI is called complete for 4 if every subset of each zero-set for 4) is contained in
WI-
A given field (or a-field) can be extended into a complete field (or ar-field).
More specifically, if 4 is an additive set function in the field WI, then there is a
field 9N'?D and an extension of 4 into a set function 4)0 which is additive in WI0
such that W9? is complete for 40. In this statement the words "field" and "addi-
tive" can be replaced by the words "cr-field" and "totally additive."

5. 4)-continuous set functions. Let 4 and 4) be two totally additive set

functions in the a-field W9. Then 4 is called 4)-continuous (or totally continuous
with respect to 4)) if for every zero-set X for VI we have also +)(X) = 0; or, what
amounts to the same thing, if every zero-set X for VI is also a zero-set for 4). On
the other hand, 4 is called purely VI-discontinuous if every set MEM, with +(M)
$0, contains a zero-set X for 4) with 4 (X) $0.
Example of a 4,-continuous function: Let ,u be the one-dimensional Lebesgue
measure. Then the Lebesgue integral (Ml)ff (x)dx (defined for the one-dimensional
measurable sets M) is ,-continuous.
Example of a purely P-6discontinuous function: The function 4 defined in
Example (a) of ?2 is purely IA-discontinuous, if in the example E designates
linear infinite set and ,b is again the one-dimensional Lebesgue measure.
It has been proved that a-totally additive set function 4 can be represented
as the sum of a 4)-continuous and a purely 4)-discontinuous set function (both
of which are also totally additive in WI). If 4 is finite, then there is only one such
representation.
For the notion of the 4)-continuous set functions, the zero-sets for 4 play a
distinguished role. A more general theory, also containing other interesting
particular cases, can be developed if the sets S of any c-field SC: are distin-
guished as "singular sets," provided that every XeWI which is a subset of S be-
longs also to S.

6. Measure. The general theory of measure, which contains the Lebesgue

measure as a particular case, was founded by C. Caratheodory [5]. Let E be a
given set, and consider the system e of all subsets A of E. A set function +)(A),
defined in (Y, is called a measure function if the following conditions are satis-
fied.:
(a) For the empty set A we have: 0(A) = 0.
(b) 4(A) is monotone increasing (cf. ?3, (c)).
(c) O(SA.) _<E0(At,).

This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms
 18 WHAT ARE SET FUNCTIONS? [January,

Because of (a) and (b), we always have: +k(A) 20. The essential inequality (c) is
the same as used above in (4).
A set McE is called +-measurable if forM, together with every set A e, we have

(8) +(A) = 4(A M) + 4(A - M).

The +-measurable sets M form a a-field 91, and c is totally additive in 9)1. This
gives the relation between the measure functions (which, in general, are not
additive) and the totally additive set functions.

7. Integration. The integral of a point function f(x) can be defined with

respect to any totally additive set function 4. Of course, the model for a general
theory of integration is the Lebesgue integral. H. Lebesgue [1 ] performed the
integration with respect to the Lebesgue measure in an n-dimensional Euclidean
space Rn. Generalizing the Lebesgue integral and, simultaneously, also the
$tieltjes integral, J. Radon [6] replaced the Lebesgue measure by an arbitrary
totally additive set function. He still took the space Rn as the basis; but then
M. Frechet [7], referring to J. Radon, easily generalized his definition for ab-
stract spaces.
Subsequently, such definitions and theories of integration have been devel-
oped in many ways by quite a few mathematicians. Perhaps the most simple
and most fascinating definition of general integration is due to H. Hahn [8 ], and
it has been extensively discussed in the above-mentioned book [3].
Let 9) be a a-field and let 4 be a totally additive set function in 9W. According
to ?4 it is no restriction to assume 9N to be complete for 4, since 9) may be re-
placed by the extended a-field 9)1, if it is necessary to do so. Let AMJ; then the
subsets Me9 of A form also a a-field, say W. Moreover, let f(x) be a point func-
tion defined on A.
The measurability of f(x) on A can be defined in the same way as in Le-
besgue's theory; that is, for every number y, the set of those points xEA at which
f(x) >y has to belong to 9).
Now let f(x) be measurable on A and let +(A) be finite. The function f is
called 0-integrable on A if there is a set function (MM), which afterwards is said
to be the k-integral of f, defined in W and satisfying-the following two conditions:
(1) (M) is totally additive in W.
(2) )(M) has the following mean-value property:
Let MA and c' <f(x) < c" for all xeM; then c'+(M) <X(M) < c"+ (M) if 4)-(M) = 0,
(that is, if o is monotone increasing on M). c"+ (M) <X(M) < c'+b(M) if 0++(M) = 0,
(that is, if 4 is monotone decreasing on M). [If herein c = ? X and + (M) = 0, one
has to set co(M) =0.]
Iff is +-integrable on A, then it can be proved that the value (A) is uniquely
determined by (1) and (2), and hence it can be designated by

(9) X(A) = (A) fdo.

This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms
 19481 WHAT ARE SET FUNCTIONS? 19

On the basis of this definition the whole theory of integration can be devel-
oped, including the representation of the integral by means of Lebesgue sums
(which H. Lebesgue [1 ] had used as the definition of his integral in the particu-
lar case that 4 designates the Lebesgue measure in Rn). Moreover, the Radon-
Nikodym theorem [6, 9] can be obtained, which gives the following characteri-
zation of the set functions that are +-integrals: In order that a set function M(M)
be a 4-integral, it is necessary and sufficient that (M) be totally additive and
4-continuous.

8. Differentiat.on. The theory of differentiation of set functions, founded

also by H. Lebesgue [2], encounters more difficulties.
Let E be a metric space, and let a be any point of E; we designate by Sap the
sphere with center a and radius p. Let 4 and VI be totally additive set functions in
a a-field 9) of subsets of E. Moreover, let 9) satisfy the following condition: to
every p>0 there is a (non-empty) set of 9), contained in the sphere Sap. We say
the sequence {MP } of sets of 9) converges to a if there is a sequence of positive
numbers {pp } with p0-+O, such that MpCSapp.
If there is a sequence {M, } converging to a, such that the sequence
{ (M )
4) (MP)fJ

converges to the value d, then d is called a derivate of 4 with respect to 4) at the

point a on 9)1. The greatest and smallest of such values give the upper and lower
derivate, respectively, and, if both coincide, we have the derivative of 4) with re-
spect to VI at the point a on 9W.
But these definitions are still too general and would not furnish a satisfac-
tory theory of differentiation without further restrictions on the systems of sets
applied. This was first observed by H. Lebesgue [2] who, therefore, operating
in the Euclidean space R. and using the n-dimensional Lebesgue measure u for
4), introduced the notion of regularity of sets. A family 9)1 of measurable sets M
was called regular by him if for every point a there exists a number P(a) > 0 with
the following property: Let MAE9, (v = 1, 2, * . ), be any sequence of sets con-
verging to a and let S,, be the smallest n-dimensional sphere containing M,.
Then the following condition must be satisfied:

(M(Sp) =

In the theory of differentiation, developed by H. Lebesgue with the aid of

the notion of regularity, the Vitali covering theorem played an essential role.
Now in different ways, conditions formed after the pattern of the Vitali covering
theorem can be stated, such that a general and satisfactory theory of differ-
entiation of 4) with respect to 4 is possible. It includes theorems of the following
type. The derivative of 4 with respect to 4) exists almost everywhere. The 4)-in-
tegral of this derivative, or of any derivate, furnishes the 4)-continuous part of

This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms
 20 MATHEMATICAL NOTES [January,

the primitive function k. Since the theory of differentiation of set functions is

rather complicated, we do not go into further details here. Those interested in
this subject may be referred to the discussion in Chapter V of the above-men-
tioi'ed forthcoming book [3].
References

1. H. Lebesgue, Int6grale, Longueur, Aire (Paris Th&se) 1902 =Annali di mat. (3) 7 (1902),
pp. 231-359.
2. H. Lebesgue, Ann. tc. Norm. (3) 27 (1910), pp. 361-450.
3. H. Hahn and A. Rosenthal, Set Functions, Albuquerque, N. M. (University of New
Mexico Press).
4. H. Hahn, Theorie der reellen Funktionen, vol. I, Berlin 1921, pp. 401-403; Anzeiger Akad.
d. Wiss. Wien 1928, No. 8.
5. C. Carath6odory, Nachrichten Gesellsch d. Wiss. G6ttingen 1914, pp. 404-420; Vorle-
sungen uiber reelle Funktionen, Leipzig-Berlin 1918, Chapter V.
6. J. Radon, Sitzungsberichte Akad. d. Wiss. Wien 122 IHa (1913), pp. 1295-1438.
7. M. Fr6chet, C. R. Acad. Sci. Paris 160 (1915), pp, 839-840; Bull. Soc. math. France 43
(1915), pp. 248-265.
8. H. Hahn, Anzeiger Akad. d. Wiss. Wien 1929, No. 2; Festschrift der 57. Versammlung
Deutscher Philologen u. Schulmiinner in Salzburg 1929, pp. 193 ff.
9. 0. Nikodym, Fundamenta Math. 15 (1936), pp. 168-179.

MATHEMATICAL NOTES

EDITED BY E. F. BECKENBACH, University of California

Material for this department should be sent directly to E. F. Beckenbach, University of
California, Los Angeles 24, California.
SOME INTERESTING SQUARES

P. A. PIzA, San Juan, Puerto Rico

I was about to send to the MONTHLY as a problem the following innocent

looking proposal:

PROBLEM: Find squares of four or more digits (of which the last one is not
zero), so that their last three digits, when reversed, form a cube.

The solution of this problem has led me to find a family of squares which
have some remarkable properties that, in my opinion, deserve communication
in this note.
Let A2l= . . abc be the required squares and let cba =d3. As c 0, d3 can
only take the five values

125, 216, 343, 512, 729.

Then A2 is one of

... 521, . 612, * * 343, ... 215, ... 927,

This content downloaded from 195.43.22.134 on Wed, 17 Oct 2018 08:28:17 UTC
All use subject to https://about.jstor.org/terms