... ..... .. - ,!!!! 1 -111 - M- ! , - , - -- ----- -6 - - I- - I ~~~~1 THE LEBESGIJE DENSITY THEOREM IN ABSTRACT MEASURE SPACES by MIRIAM AMALIE LIPSCHUTZ YEVICK B.A., New York University, M.S., 1943 Massachusetts Institute of Technology, 1945 Submitted in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY from the MASSACHUSETTS INSTITUTE OF TECHNOLOGY 1947 Signature of Author.., Department of Mathematics, September 5, 1947 Signature of Professor in Charge of Research.... Signature of Chairman of Department' Committee on Graduate Studies.... USQ U TABLE OF CONTENTS Page Acknowledgements I Introduction and Summary of Results Chapter I: II The Density Theorem in Euclidean Space Chapter II: Chapter III: Chapter IV: Appendix: I: The Density Theorem in Abstract Measure Spaces The Density Theorem in Finite Product Measure Spaces The Density Theorem in Infinite Product Measure Spaces Proof that m(S(Xn)) O J- II: Procedure used in the construction of the example of Chapter IV Biographical Note 289911 1 6 14 18 28 30 _ ACKNOWLEDGEILETS The author wishes to express her gratitude to Professor Witold Hurewics for suggesting the problem of this thesis and for his kind help and encouragement during the time that this work was in progress. INTRODUCTION AND SUMMARY OF RESULTS Let us consider a measurable set define: a) The average density of C-4 1); b) The density of (24 lim -; -4 O A A A on the line. in an interval at a point P We I as as where the sequence of intervals In C.,,] 'i In = d(In) -,O, converges to P. for all The Lebesgue density theorem asserts that the n.) (i.e. diameter A = 1 density of any measurable- set and equals nero for almost all In>P for almost all PF CA. P e A This theorem can be interpreted as stating that generally a set is not uniformly distributed over the line byt rather that its points are very concentrated in certain regions and very rare in others; the exact measure of the concentration is specified by the inequality: The set of intervals 6c (A) in which the average den.sity of A 0) exceeds a., 0 a 1, has measure .-2 ~ ) )) 6z CO /1 m If we are given in an abstract measure space CA) E, a family of measurable sets (which we shall call elementary sets) and a notion of convergence, and point p of E S. E a concept of density with respect We now assume in inequality similar to (1). existence of a function qf6 L is such that to each converges at least one sequence of sets of S, we can define in to the family S E for this family an Specifically we assume the = l.u.b. (/2(p),such that the S liii union6,: fA) of elementqry sets in which density of A exceeds satisfies: a A 9! C4) (2) the average ( 4 Busemann and Feller (B3) hage shown that in Euclidean spaces the validity of (2) for a family of sets S implies that S. the density theorem holds with respect to the family Theorem I of Chapter II states that this sufficiency condition can be extended to abstract spaces provided we impose upon the sets of S the restriction that given an arbitrary system of elementary sets S' C S any measurable set can be approximated in measure to withing any E> 0, a sequence of sets Bn, union of sets of S'. by From the validity of the density theorem with respect to a family of sets follows, that the Lebesgue integral of any bounded function can be differentiated with respect to this family and that the derivative of the integral equals the integrand. Theorem I of Chapter III states that if (2) holds in a finite number of abstract spaces functions - with 96(d} then for any measurable set the direct product space when'W'r/ El, E2,.---. *n A in E =7T Ei is the union of all elementary sets in El (i.e. all sets which are the product of elementary sets in the fadtor spaces) in which the density of A It follows in particular that since (2) holds in the density theorem holds in a. exceeds R1, Rn for the family of all inter- vals and that differentiation with respect to this family is allowed. Proceeding to the product of an infinite number of abstract spaces, Theorem I of Chapter IV asserts that condition (2) breaks down; i.e. for a bounded set 6 , (A) A the set may be arbitrarily large. Moreover an example is constructed in Chapter IV leading to Theorem II which states that: The density theorem is not valid with respect to the set of all intervals in the space interval. E =7T Ei whose factors are the unit The same can be expected to be true for the direct product of an infinite number of arbitrary finite abstract measure spaces unless the elementary sets and the functionsc4(V)are greatly restricted.* From this follows theorem III of Chapter IV stating that in differentiating an integral in an infinite number of dimensions as defined for instance by Jessen, we are not entitled to performing the limiting process involved consecutively with respect to each variable separately. Thus the main results of this thesis are contained in 1, i.e. the only *For instance if 4t() is always equal to elementary sets converging to points of itself. A are the set A Theorems I of Chapters II and IIt, and Theorems I, II, and III of Chapter IV. Finally we note that Theorem I, Chapter II, can also be considered as a theorem on conditional probabilities. if the spaces m(A) Ei For consist of a finite number of points and is ~defined as the ratio of the number of points in to the total number of points in conditional probability of A E, then '4 with respect to 4T) is A the I. We believe that in this form Theorem I of Chapters III and IV should possibly allow of some interesting applications in probability theory and statistics. I I. THE DENSITY THEOREM IN EUCLIDEAN SPACE In the extension of the Lebesgue density theorem to higher dimensions a difference arises between differentiation with respect to the system of all cubes or any system regular with respect to the set of cubes and differentiation with respect to systems that do not satisfy the regularity requirement. An absolutely continuous set function, i.e. an F (e) indefinite Lebesgue integral 41P can be differentiated with respect to any system regular with respect to the set of all cubes and the derivative of F(p) = f(p). A family of sets with respect to the family P R T is said to be regular of cubes, if of the space converges a sequence ( is contained in a cube each __P) o ;__ of to each point T R such that with for all k -where o(P) is independent of the particular sequence chosen. Letting E f(p) be the characteristic function of a set it follows that the density theorem holds with respect to any such system. A simple geometrical proof of this fact in which the &are by Sierpinski (S2). n dimensional spheres was given The same result was obtained by de la Vallee Poussin (V2) using the notion of differentiation on a net.* *This notion becomes very useful when considering infinite Euclidean product Spaces. SIM These proofs and the proof usually given (see Caratheodory (Cl) ) are esaentially based on the following lemma. Lemma I. A Let A be a set such that in each point for any sequence 4 . (~P /~ > converging to A O( p of p, ?~and the same with the inequalities reversed. This lemma can then be combined with the following (see R. de Possel (P2) ). Lemma II. If g(p) are two functions defined and f(p) in the points of a space R, such that whenever in all the points of a set A, with a point all p p of of A with f(p)-7 a g(p) > a m(A)2> 0, there exists then f(p)-P. g(p) for R( and identically with the inequality signs reversed) and therefore RiP) >o< , 3 JF 1(y) -,)(P)hs > and ') reversing the inequalities C.e. d ()>o(implies We note that sinceJ2 = The proof of Lemma I is based directly on Vitali's covering theorem. Namely, we cover the set number of disjoint sets A 7 9 i with an enumerable in each of which From the complete additivity and absolute ,. continuity of A m and F follows for any ; OPIUM b, v XI -c<%)I< - t m(A). F(A) > and finally If we eliminate the condition of regularity, Vitali's theorem is no longer valid. For the system of all intervals this was shown in an example of H. Bohr (B2) and another of S. Banach (Bl). It also follows from the work of Busemann and Feller (B3). A number of proofs exist to show that nevertheless the density theorem is valid with respect to the system of all intervals and therefore that for any bounded function DF(p) = f(p). f(p), The simplest of these is due to F. Riesz (RI). Other proofs can be found in S. Saks (Sl). The general problem of finding whether the density theorem holds in Euclidean spaces with respect to an arbitrary family R, has been treated by Busemann and Feller. They assume that the family R is such that to each point p of the space tends a sequence of sets JPn of R (where e by tend we mean = .>lfor all Cke)o). For any measurable set C)<$o1<j with n and the diameter of C (Ca) , X with m(X)"; 0, and any 0, letG,- g-(x)be the union of all sets f I ? <. R > 0< The following conditions are both necessary and suffi- cient for the density theorem to hold with respect to the system R. (1) //76 null sets. zX*IV (/)j*A where N1 and N2 are Given a monotonically decreasing sequence of bounded measurable sets > , - - -? , , with -- ( approaching zero, and a monotonically decreasing sequence of positive numbers S, > it dI// - - > 0-> . g -- with -*0- 4/ The following condition is always sufficient for the density theorem to hold, and is necessary and sufficient if the family R contains with each all sets that are similar and similarily situated to (3) and R ''7" There exists a functionqe2 depending upon < only, such that for every set C6';w_) fC/l 'O' X with m(X) 0 (X/ Busemann and Feller have shown (and we shall show more directly) that (3) in Rn. is satisfied by the set of all intervals However the set of all rectangles in Rn (i.e. with sides not necessarily parallel to the axis) violates (2); thus the density theorem does not hold with respect to this system.* From the validity of the density theorem with respect to the set of all intervals in Rn we deduce this important fact: *See also N. Nikodym (Nl) who proves this fact in a different way. -..- ,W..IIMILE.Hanomolliill .L .I 11 III111111 -Im. .II. --,. Given the function Let 4 = hXk -- F(x,y) = (gqd ( d AK 2 in the differential quotient C§E(Ai}> ) P(, F~x4 y Af-r 4)) =~ < C ye' /u( )/4 YRxy#,(/ +i'F(Ayj xe4yj The existence of this limit followed from the fact that the set of rectangles with I hk)< M is regular. Since the density theorem holds for intervals we can replace by lim h lim k, i.e. ( /,),, W~K, i)) - <f~orf~ Ay~frj and similarly for higher dimensions. 6o 4+~ o Jim (hk) £ f,+ - d *For f(f q)unbounded wb cannot conclude what follwws from the validity of the density theorem with respect to the set of intervals, but a stronger criterion must be satisfied. . The details are in Busemann and Feller. See also B. Jessen, J. Marcinkiewicz, and A. Zygmund (Jl). 3y A F(-#4p)J II. THE DENSITY THEDREM IN ABSTRACT MEASURE SPACES In this chapter an attempt is made to generalize the results obtained for Euclidean spaces to abstract measure spaces. Consider an abstract set points p. 1 Let E whose elements are the be a Borel Field of sets of as well as each point of E contains the space E, which Let E. be a completely additive, non.-negative set function, m(A) defined for the sets A . of are The sets of called measurable sets. Assume that we are given in point a family &) p p. the point The E corresponding to each of measurable sets k'i/each containing (g/ will be called elementary sets. (/ To eachl/w let there correspond a positive parameter A sequence P/(,)of elementary sets is said to converge to (a) p, whenever lim/Z 0/,* 0. If the setSw9 contains converging to at least one sequence /kw9 system of elementary sets for the point of a set p point sets J A o we call -S99 p. a If to each corresponds a system of elementary , the totality of all S.f/ for all points of A is called a system of elementary sets for the set denoted by A and S(A). A subset SI(E)C S(E) is itself a system of elementary sets for the space E, provided it still contains for each point now p of E, a sequence of J19;) A be a measurable set and the average density of ,~ =' A p converging to a point of A. We define in the elementary set C Anp p. Let VW9 v) We define the upper and lower densities of the set a fixed point sets S(E) ,AV 0 A at p, with respect to the system of elementary as: respecti4ey = as: C If D =1 for almost all for almost all pFCA, we say that the density S(E) theorem holds with respect to the system speak of the density p F A, or and we D = D = D. The following conditions have been established as necessary and sufficient for the density theorem to hold in an abstract measure space, with respect to the system S(E) of elementary sets.* Given a number a, 0<ac1, a set A widai*a and an arbitrary sub-system of elementary sets A. Ther exists always a point corresponding to p p and contained in (A) > 0 S'(E)C S(E), A and a set S'(E) such that of V/() B. There exists always an enumerable sequence of elem*See R. de Possel (Pl). Condition B was established independently under a slightly different form by B. Younovitch, (Yl). C I p entary sets (/O)CS(E corresponding to points pn of A such that: Letting a = 1 -ii , condition A can be expressed as: A'. Any measurable set of positive measure contains an elementary set of S'(E) within S measure. Condition B can be shown to be a consequence of A. The validity of the density theorem follows from B by replacing the l4".) by disjoint sets VL- ) and deriving from this an equivalent of Lebesgue's theorem for abstract spaces. The proof of the density theorem then proceeds as before (see Chapter I). Given an arbitrar. any measurable set I, m(X) 0, for a., 0< a <1, denote by 6w- ,4K)the union of all elem- entary sets ViC S(X)with (a) (b) K/ )->0 'C"- / Observe that: and if (2) Theorem I(a): > ,,X, (kr) ) A necessary and sufficient condition for the density theorem to hold with respect to the system S(E) of **These conditions are identical with those of Busemann and Feller for Euclidean spades except for the addition of requirement (C) in part (b) of the theorem. The proofs are also the same with the exception of the step in the sufficiency proof of (2') which was condition (C). E elementary sets, in an abstract space measurable set X, with is that: for any m(X)- 0, I o A neceszary condition for the density theorem to hold with respect to the system S(E) is that, given an arbitrary sequence of measurable sets X - - - ,- X, > and an arbitrary sequence of positive numbers . (2') lim4i (G -- p - -- > E,> -- E L CWoi Proof: lTe' 'Condition (1'): Necessity: for fixed a, W contains J>O all elementary sets of arbitrary small parameter, of the space E holds for in which D & a. TlIWa if the density theorem S(E), (1') follows. 7F, Sufficiency: if (1') is true, OC. x< I J>* ( +Nj,; m( 3) =4A/,)= and the density theorem holds. Condition (2'): Necessity: for/> CX n:??(c' but the density theorem gives: and therefore (K,)'since lim C = xn lim'ICGd (AjPL 0. Assume now that the system S(E) of elementary sets satisfies the following condition: (C) Given an arbitrary sub-system always possible for any measurable set t o construct a sequence of sets of the form: S'(E)C S(E), it is X, with m(X>- 0, Bn, each containing X, (x)#+q3 (b) S'(E) for all k and all a, co0 such that7T Bn = X + N where N is a set of measure zero, --*I =/ i.e. fi (X) = lim m(Bn). Theorem I(b): under the assumption (C) for the system S(E) (2') is sufficient for the density theorem to hold with respect to the system S(E) of elementary sets, and, (3') if there exists a function Aik) depending only on a (O<a<l) and the system S(E), such that for any measurable set (, ' C'6 ;; X, ( < ()) the density theorem holds with respect to 4)&v S(E). Proof: Condition (2') Sufficiency: letA set. Given a, 04 a<l, let A CA be the set of points in which the lower density of A to each point (a) -27 be a bounded measurable p of (Q)/, Ctl Denote for the points (1 - a); i.e. is less than converges a sequence of Vft)with: 7 4/-- 2 ';: - p of j I /} the system of the / Oj C ,< satisfying (a) by S'(A ). By condition (C) we can construct from the elementary sets of S'( ) a sequence: /= (where we let:1Zd") <-I with 7 and 2-0(k,) en(V') ,,-<r, for all a, such that +O; 0, let X ,,-) - o . .') Now since +~G7K m(A) zE then,2x C '6- ,,- (, ---)X,,.>., r combining this with (a) since ,) is contained in the union of sets A ,1 cn()j ) satisfying (CA)but:h7(Q~, (K-,))yand therefore, 0'0 0. = Condition (3'): Sufficiency: if (3') holds so does (2'), V* now show: thus the density theorem holds. Theorem II; Conditions (C) and (2') (or (3') ) combined imply A p(7 ). X, with m(X)'>0, there Proof: Assume that given a set exists an such that for all elementary sets S(E) where Bn V of S'(X) (VA) en (W satisfies (C), we can construct a sequence (a) Since S'(E)cs(E) a, and a system of elementary sets CS 2with z say, and Consider the sequence: 25 we have A, :> Now since xC thus: (b) V (d4&, we have del L -n C , = 19 Bn but is the union of V, satisfying (b), = 0. lim( Therefore B thus: C 61, lim m(Bn) = 0, which contradicts our hypothesis. Since A implies B, and B implies C, Theorem II establishes the complete connection between the results of de Possel and the results of Busemann and Feller when generalized to abstract spaces. Remark: Condition is always fulfilled if the space C is a compact metric space and the V with the parameter V S'(E)<:ZS(E), each point ( sets S'(E) S(E) defined as the diameter /Z the elementary set of E are open 4() of For, given any sub-system . p of E is contained in a ) open, the system of elementary , forms an open covering of E. Since E is separable this covering contains an enumerable covering. Thus given a measurable set X, with m(X)>0, and a sequence f, --- 0, we can cover X with an enumerable ,, such that number of sets B17, Condition B is always fulfilled if for any sub-system St(E)e S(E) the elementary sets of for the Borel field $of measurable set with S'(E) For, given a measurable sets. m(X)>0, we have ow for all coverings of the set X form a basis (Par by elementary sets Q( C.S(E). Given a sequence m(X) - m(Bn))< -> 0, let E, , Bn =2i, Bn-> X, with Bn. this gives the desired sequence Finally we state:* Theorem II: a necessary and sufficient condition for the DF(P) derivative Jf(P) dP of the indefinite integral of any bounded measurable function exist with respect to the system and to equal f(P) S(E) F(P) = f(P) to of elementary sets is that the density theorem holds with respect to this system. *See proof in R. de Possel ~I III. THE DENSITY THEOREM IN FINITE PRODUCT MEASURE SPACES El, Given a finite number of abstract measure spaces E 2, . . . En: let the Borel field 70 =TEi E be their product. spanned by the sets A in Consider E which are the direct product of measurable sets in the component spaces. (A =TAi, Ai V.Ei, Ai measurable.) It is possible to define a completely additive, non-negative measure for the sets of . such that m(A) =27Fm(Ai). The proof can be found in S. Saks (Sl') -or a more elementary proof in Lomnicki and Ulam (Li). Theorem I: Let I and Y be two abstract measure spaces S1 , S2 in each of which ther is defined a family subsets l. Given a number arbitrary measurable sets existence of functions (yk where(, such that 6~ X Z o9 Y. Define the family arbitrary measurable in ein))< )being C S2 .. > ( as the collection of all sets{5 = We assert: for '/ such that o( at,, Consider the product space R = Si X S2 (a), assume the the union of all sets{C Sl, (.> 17 a, O< a< 1, and X' C X, Y' C Y, 9(), ( of X X Y, r the union of all sets X 0l o < /hZi << 6i for which Ai-7 2/ (C-l Proof: Let a be a set&64) Q For any set x C x . Let Xo > Rr >& - - < (a/2-7(Ko) 2= ct')A'e for: =: Z -(x) (ek 6 Now let§ We have for any Y and therefore C i.e. of for which X) O' r<a. We prove - C 9A(A' (e) for all , (ex) = 2 5 'C L' let: be the set of all 0" Ar6v where X We can write We have i.e. _ C27 C G 1 E but X (A giving the desired result: Similarly for the product of n with functions ' c) - - spaces 4 (2)) <"r ' S1, S2 C 7, /h-- S2 Let the families of subsets x - ,4 be two given systems of elementary sets in X and Y. 6< a <ae Assume that condition (C) of Chapter II is satisfied for the systems s(X), S(Y), S(X) XS(Y). and asserts that if in both X and Y Theorem I then inequalities of the , ) - - form (3') hold and therefore the density theorem holds with respect to S(X) and S(Y), then the density theorem holds with respect to the system of elementary sets in .. XX Y. Let now X x on the line; points in A S(X)? S) x1 , x2 , be a finite set of points let m(A), AC I be the number of divided by the total number of points in i.e. the probability of finding at random a point in With each point xi associated all sub-sets Yi A. x of and which are such that with two points xi and they contain all intermediate points (i.e. all xr with sets V- k.< r<j). Let S(X) for all points of Take A arbitrary in ding each point xi of points to each side of A xi, con- xj taining xk X, be the totality of all such X. X. C 'l)is obtained by inclu-. V in a extending 1/a (3 i.e. of total length This result is also true if X is infinite and is replaced by the appropriate measure defined on S(R1) -) m(A) We X. being the now apply this result to the space R1, family of all open intervals on the line. Replacing points by intervals in the preceding reasoning, we have for an arbitrary measurable set X: Pi (6$ < __ -, combined with the first part of the remark in Chapter II, this gives: *Note that S(X)x S(Y) determines a system of elementary sets in the space Xx Y, if convergence in X X Y is defined in terms of convergence in the factor spaces X and Y. Theorem II A: the density theorem holds in of all intervals. Rix R1 = R2 : ZcRn for the system Applying Theorem I to the product space s(R2 ) is the set of all intervals in the plane. For R- For any Z C R2 X , i.e. all and measurable, and measurable i.e. Theorem II B. The density theorem holds in 2. Rn. z IV THE DENSITY THEOREH IN INFINITE MEASURE PRODUCT SPACES Let E E = be the direct product number of abstract measure spaces Consider the Borel fiel A Ei of an infinite of total measure 2. Ei panned by the sets A in E which are cylinders over a finite number of dimensions (i.e. sets of the form TE;, S =Tix Ameasurable. It is possible to define a completely additive, non-negative measure for the sets of - such that ti(A) The proof of this fact has been established under various restrictions by the following people (among others): Steinhaus(S3): Ej is the unit interval (0,l). The equiv- alence of the measure in the infinite dimensional space to the Lebesgue measure on the line is shown by extending to an infinite number of dimensions the usual 1:1 mapping (within a null set) of the points inside the n dimensional unit cube onto the unit interval.* Kolmogoroff (K2): space Ei Use is made of the fact that each factor is compact and that consequently E compact if convergence in E is also is defined in terms of conver- gence in each component space. Doob (Dl): Ej is any abstract measure space, but this case *That is, to , , ,Cwhere j9 0o, 1,, Z-6,-f, corresponds the point o1 0 19,,, 9 t9 on the line, and vice versa. 11 3i,& 9,3- on the oaI is reduced by a set isomorphism to that treated by Kolmo. goroff.* Kakutani (Kl): The Ei are any abstract measure spaces. The proof given by this author is the most straightforward of all and does not use any concepts of topology. Jessen (Jl): The are the unit circles, Ej E is called De la Vallee Poussin's the infinite ,dimensional torus space. notion of dissections and nets is extended to this space in the following fashion. D, A dissection is defined by removing a finite number of points from each of a finite number of circles E1 ....B, and considering all the inter- vals that can be formed by taking direct products of arcs in the Ei, so obtained, by TEi. A subdissection D2 removes additional points and contains all possible resulting intervals, including those of dissections D1 , D2,.*.*..Dn*** such that for any fixed the maximal length of each arc of Dn ->o0 as A sequence of sub-. Dl. n -- '0o Ek k, which corresponds to generates a net in E. With the aid of so-called corresponding nets in E and the unit circle** there exists a 1:1 mapping (within a *The complete additivity of the measure in E .(i.e. A = (Ai) m(Ai) )'is A rE, Ai 1lAJ = 0, i $ J to show m(A) = proven by performing a measure preserving mapping in which the enumerable number of sets Ai involved become determined each by a set of inequalities imposed upon a real valued function. To the Ai thus correspond an enumerable num1W of disjoint sets A of equal measure in the space E =TT Ei whose factors are the unit intervals Ei. See L null set) of El, which is measure preserving back onto E E, into meshes of the corres- and maps meshes of the net in ponding net in El. From the existence of this mapping is concluded: E (a) The equivalence of the measures in the torus space and the circle El. (b) Given a net, denote byA Q(for any value of stepfunction which in any interval In n, the n-th dissec- of the = is equal to the corresponding quotient tion Dn with F(A) = f(x)dx AC E then almost everywhere. limL " ,(x) = f(x) 0O We are however not so much interested in knowing whether the density theorem holds with respect to a particular family of intervals in E, such as a net, but rather whether it holds with respect to Wh(family of all intervals in interval in E, i.e. all sets of the form Eij, where only a finite number of the different from the whole space by TAi, Aj x = (xlx2****xn*...); E1. Ai an are Denote a point of E Definition: a sequence of points in E, x, x 2 , .... xn,.... converges to x, if and Halmos and von Neumann (Hl) for a clearer exposition of this mapping. dn "That is, such that for each In c Dn, there is an in (dissection in the unit circle) and vice-versa with m(In) = m(in) and In+l C In implies in+IC- in' **MSteinhaus' method of proof is merely a particlar case of this one. only if the corresponding sequence of coordinates x xi, i. for all We can, but do not need here, give an explicit expresE, in terms of distance in each of its sion for distance in we merely assume that the diameter of an component spaces;* interval in E tends to zero if and only if the diameter of El, El;X yE each of its projections on the spaces 2 ,**..-etc. 0. -+E Lemma: , can be written asX If X C E h contains an interval of arbitrarily small EZ, and diameter, then xCv,X- +~() G)-.- V 4 and -IV k for all X - where V = for let - then ) /,(< It is easy to see that: Theorem I: Condition (3) product space E =1 is never fulfilled in the infinite E(o,). For consider a set of the form where the Xi follows that , X = k - are intervals in Ei; from the lemma it M 6( (C C A)) *See for instance Frechet (Fl). --- ) ' (,-)- ~ - thus limdo# ,--;7 - lim : The same is true if =V. i E is the 07 -;14h infinite product of an abstract measure space by itself in which the function-A)of theorem I is ,> 1. Elementary sets in E are defined as sets of the form elementary in Ei. c h X2 . =..-X.= for any set / LetI(vX)> X1 = X ,rK. X = Now take AC E be given by :( 'wher . 2 and > Xn, then. e/, it is always possible to again, given any function find a set 77 V withn (/2) > ffCW))/. Condition (21) is also not fulfilled in the space E: the following is an example of a sequence of measureable -- ' sets: ,)-o,-and a sequence rX S>A S4,j;&z- IS712r - >. > -with 07 for all k and + 0 lim where is the unioh of all intervals in E with 0<x<a x0 ;gx, X, = s 63 Jr C44) IX 2XE 42.£I~ 'an r (dand Denote the interval +f C, such that: ~ by (0,a). Let 4 s n 5 K E Z I~ x x - 4(>~a )~ ;I K(,L)(o±) I -( wl10';,,, dv -;p K 4k (-4L/ ,z 3 -~~g CXe 2 . k1e fi7 A . x ~k~ / (rc-1 , 3 Are"S "'-o 37 -7'. ey It x~Ar'6 (aI)JoIx k q -. IJ~-~-4AXL~KO, +C~k +Ex~z. x~se (oL,01J)[ -C2 4qx4r Es x ( .. / x S X 3 X§ xX Ex3 k ' E a x ( ,x( X8 X r4O 9 J-4 ) X (0, K sz .i)>g$ )C4 )x( -L x("! g$Oj) .,x - ,C-k~ -xQ V / ,x ) x(", Ef-- 'V({() Z C' E -t. (o Er ecja r, -- E p etc. This construction can be explained as follows: consists (2n)2n = p(2 11) terms, each term being the of the sum of product of Xn factors equal to the interval 2n = k(2n) by the remaining spaces in their entirety. In generalA' is obtained from fashion: with each term ol terms of first 2 . A X, in the followi ng 2.. we associate By "associate" we mean that we take the factors of each term of contained inside 74'r I the 2n X factors of the term of with which it is asso- "ii4 ciated; the other 2 being taken each time inside an "utilized" before. (2- we utilize El and 4 8 X g 2#2 7,6 in the fourth term. = 64 terms associated with each term of £ and utilizes in addition up to, which has not been E3 Thus in forming X E2 in the2.- first term, up to has , different from (0,l) factors of 6 up to A . -Xutilizes , etc. We have for this sequence:,, (S (S,,)) whereY(X,)is the sum of all overlapping parts of the terms of ; i.e. -, 6(KXis the measure of the sets that are added up more than once in obtaining the first term: . 2. A somewhat lengthy calculation** shows that m is of a higher order in thus limm(X)= n than( (S(x4) VNI (m ( X = 0. N1 -->to Now using the lemma on page 2.1 and (4) on page & we have: 0(j -j) 0C>C 0 -f - - C E X Ey -- Cetc. **See appendix, where we shall also explain how the above construction was arrived at. (X) and in general] G contains for each term the sum of the products of< of each factor by all the other factors. (See Now certainly: P-- 0 is the set of all parts of gc where X )that are added up more than once due to overlapping of the terms of O( -. . 2 h (SC It is easily seen* that Finally take~>- 5) where - --- The sequence - 0 for lim is such that i; all Taking now for the set ) X J1 for a term of the form, say: ,where-- K; t ~'A 4 ~i~i~~ ~ e+c. the sum of a sufficient number of elementary sets of the form e-3 CX C~ ~x -fi. .- Cwx-. (' c' for all with cJ()) so that their sum equals e; ()( another set of with first: xX will sum ) i, then (?( X X'"k--- etc. xSee appendix. **Note that by the lemma of page 21 this is all that is required of the components of C in C, (X. for e to be contained -~ We find that in and lim i1 ;(6 Theorem II: () 060) A.E.D. E = In the space are the unit interval (0,l) Ei, whose factors Ei the density theorem does not hold with respect to the set of all intervals in E. The definition of a Lebesgue integral in a space of an infinite number of dimensions has been given by Daniell (Dl) or by Jessen among others. S~~ -, of Consider such an integral f {t ~1 . X(. a we".. - - '., --,-t,; -- d4 61-,L and a 7 = sequence of intervals ), whose diameter tends to zero as n -0: (where/ I,,V It is not generally true that Theorem III: lim .. - X Ap For otherwise letting function of a set true in If E , r - -. ,,--) be the characteristic A, the density theorem would always be for the set of all intervals. E = Ei with the Ei finite abstract measure spaces which are such that: (a) Sets of any measure between zero and one exist in each of the Ei; (b) The functionffi, defined byn, (ejI)) =f4 ( Xfor a measurable set X in j40is greater than one; Ei, ' then we can construct an example similar to the one above to show that the density theorem does not hold with respect to the family of all elementary sets in E. In general we must know something more about the elementary sets (i.e., about-qoc}) and the measure space itself before the analogue of Theorem II can be established. Theorem IV: E =TFE± There exists a measurable set A in -the space (Ei the unit interval) and a family St(E) of intervals in E, containing for each point sequence of intervals converging to p of p, such that A a S (A) Vn, contains no enumerable collection of elementary sets covering A, with) /r A)- 4P" ( V for E positive and arbitrarily small. This follows from condition B of R. de Possel (p. 7 ). --JI APPENDIX I. Proof that m S() =O( As a first, and too large approximation of m(S(Xn)) we find the sum of the measures of the parts in common to the terms of when combined two by two. Xn The manner in which the measure of the part common to Xn two terms of is calculated is most easily illustrated by an example: (14x A = Say x( - - - x' ( 44&Y 'Y B=( then AB = L} m(AB) =(o i.e. it is the product of those factors different from (0,l) common to the two sets by the remaining factors different from (0,1) of each of the two sets. Now, in the set (1)LI2") Xn we have: in common and there are in T(2 (2) factors terms which have two by two Xn the sum of collections of such terms. groups of .? K p2 ^ ) terms such that each term from one of these groups has .I' tors in common with each term of another of the groups. fac- 1 12i)such There are (3) r(2'') collections of groups. terms such that each term groups of of one of these groups has 27 factors in common with each There are PcL.'~) such collections of term of another group. groups. * (n) * 0 0 pr.'%) * 4P *0 00 0 groups of * 0 0 0 0 0 0 0 0 0* (tt)terms such that each of one of these groups has no factors in common with any term of the other group. We now recall the formula: Given k groups of n objects each, the total number of different combinations of objects, not belonging to the same group, is given by: -_) __ We thus get: 1 m(S(Xn)) __2_ 2-- PC 2.O 4 .2.ell% ) 2_tt~ [ kZO A, a" Ic Le --- -ti 3 IV ChapVe so that m(X )c each term of th diec while LbA) h is replaced by Xn prdutofk ( 2 11) (A2 - esA CXh for - terms giving: ckwz h II. Procedure Used in the Construction of the Example of Chapter IV We know by Theorem I, p.2-1 that for a set A which is the direct product of C^ (a C k sets, Al XA2 To utilize this fact for the L) , A) construction of a sequence of sets ?(, m(Xn) -* 0 with particular if X we would have: the latter expression more slowly than - - x m(Xn). In xE "X X lim A.( If the factors demand - ),, since *CC' some of the n '-x,- we had: X.%X and LX:)>0, lim m(6' must tend to zero with XAk3 we have . of Qare )is ; e -->a at least for all for some intervals, the diameter of i i while . m(Xn) =O(-) at least, and thus t(X But we implies If the factors of Xn are the sum of intervals whose length tends to zero( in which case we can satisfy C whose sum is ,and form , satisfy - Y , ' (( ,)-e say, intervals of the it can be shown to be impossible to ')*- . Xn As a next step we then take as a sum of oo,)dis- joint terms each of -ko') factors whose product equals where we require (3) The measure of Xn ~f tends to zero: Now by (2) each factor of each term 1/ C. -c this can be achieved by taking ko%) = n, L.- ) z in Chapter IV. 1 cmt) = nn. Thus which leads us to the example we displayed (We use instead of nn.) BIBLIOGRAPHY (Bl): Banach, S. : Sur le theoreme de M. Vitali, Fundamenta Mathematicae i p. 130 (1924) (B2): Bohr, H.: see Caratheodory, Vorlesungen uber Reelle Funktionen, Druck and Teubner, p. 689 (1927) (B3): Busemann, H. and Feller, W.: Differentiation der Lebesgue Integrale, Fundamenta Mathematicae 12 p. 226 (Cl): (1934) Caratheodory, C.: Vorlesungen uber Reelle Funktionen p. 492 (1927) (Dl): Daniell, P. J. : Integrals in an infinite number of dimensions, Ann. of Math. (2) 20 p. 281 (1919) (D2): Doob, J. L.: Stochastric processes with an integral valued parameter. Trans. Amer. Math. Soc. 44 p. 87 (1938) (D3): Dunford, N. and Tamarkin, J.: A principle of Jessen and general Fubini theorems. Duke Math. J. 8 p. 743 (1941) (Fl): Frechet, M.: Les Espaces abstraits, Gauthier Villars, p. 164 (1928) Sur quelque points du Calcul Fonctionnel, Rend. del Cire. Math. di Palerno, 22 p. 1 (1906) (Hl): Halmos, P. and von Neumann, J.: Operator methods in p. 332 classical mechanics II, Ann. of Math. 4 (1941) (Jl): Jessen, B.: Theory of integration in an infinite number of dimensions, Acta Math. _1 p. 249 (1934) (J2): Jessen, B., Marcinkiewicz, J. and Zygmund, A., Fundam. Math. 21 p. 217 (1935) (Kl): Kakutani, S: Notes on infinite product measure spaces I Proc. Imp. Acad, Tokyo _9 p. 143 (1943) (K2): Kolmoporoff, A.: Grundbegriffe der Wahrscheinlichkeits rechnung, Ergebnisse der Math; 2 p. 27 (1933) (Li): Lomnicki,Z. and Ulam S.:Sur la theorie de la mesure dans les espaces combinatoires et son application au calcul des probabilites I: Variables independantes.Fund. Math. 23 p 2 3 7 (1934) $Ni): Nikodym, N.: Sur les ensembles accessibles,Fund. Math. IQ p i67 (1927) (Pl): De Possel, R: Derivation abstraite des fonebions d'ensemble. J.de Math. pures et appliquees, 9ieme s, 12 p 391 (1936) (Ri): Riesz, F.: Sur les points de densite au sens fort, Fund. Math. gg p. 220 (L934) (Sl): Saks S: Theory of the integralVitali's theorem, p.106, Proof of the density theorem p.129 Theory de l'integraleproof of the densiw .ty theorem p. 231, measure in product spaces p.259 $2): Sierpinsky ,W: Demonstwation elementaire du theoreme sur la densite des ensembles, Fund.Math.4 £. p.167 (1923) (S3): Steinhaus ,H: Sur la probabilite de la. convergence des series. Stud.Math. 2 p.21 (1930) (Vi): DE la V llee Poussin:Integrales de Lebesgue, Gauthier Villars, p.57 (1916) (YI): Younovitch, B: Sur la derivation des fonctions absolument additives d'ensemble. Comptes Rendus de l'Acc. des sSiences URSS, 30 no. 2 p.112 (1941) BIOGRAPHICAL NOTE The author of this thesis was born on August 28, 1924 in Scheveningen, Holland. She attended the Montessori school in the Hague. Her secondary education was received at the College Marie Jose in Antwerp, Belgium and at the Lycee Francais de New York (from 1940-.41) where she passed her baccalaureat examination. From 1941 to 1943 she attended New York University and obtained a B.A. degree. She entered M. I. T. as a graduate student and research assistant in physics in November, 1943, and obtained a M. S. degree in Physics in February, 1945.