Actes, Congds intem Math, 1970. Tome 3, p. 201 a 210.

ON THE APPLICATION OF COMBINATORIAL

ANALYSIS TO MUMBER THEORY
GEOMETRY AND ANALYSIS
by P. ERDijS

In this lecture I will discussthe application of some well known and lesswell
known theorems in combinatorial analysis to various other branches of mathe-
matics. In other words I will not mention combinatorial types of reasoning
the useof which is of coursevery wide-spread(e.g. the classicalproof of Carleson
on the almost everywhere convergenceof Fourier seriesof functions in L, is
full of combinatorial reasoning), but will restrict myself to caseswhere definite
quotable theorems are used.My paper in no ways claim to give a complete survey
of all the applications of combinatorial theoremsand is certainly heavily biased
towards my own work. Though combinatorics has been successfullyapplied to
many branchesof mathematicsthese can not be comparedneither in importance
nor in depth to the applications of analysisin number theory or algebrato topo-
logy, but I hope that time and the ingenuity of the younger generationwill change
this.
First we discusssome applications of Ramsey’stheorem. The classicaltheorem
of Ramseystates as follows : Let S be an infinite set. Split the k-tuples of S into
r classes.Then there is an infinite subsetS, of S all whose k-tuples are in the
same class. The finite form of Ramsey’stheorem states that to every k and
u,, . * a, U, there is a smallestinteger Rf) (u,, . . . , u,) so that if we split the
k-tuples of a set 131= R, (ul, . . . , u,) into r classesthen for at leastone i there
is dn Sj C S, ISi1> tli all whose k-tuples are in the i-th class.The exact determi-
nation, or even good estimation of Rt) (u,, . . . , u,) is a difficult problem which
is very far from being solved and we do not discussit here.
Ramsey’stheorem wasoften rediscovered.Szekeres[ 1] rediscoveredit in connec-
tion with the problem of Miss Klein, Miss Klein observedthat if there are 5 points
in the plane no three of them on a straight line then there are always 4 of them
which determine a convex quadrilateral. She then asked : Is there a smallestf(n)
so that if there are f(n) points in the plane no three on a line then there are
always n of them which determine the vertices of a convex n-gon. Szekeres
observed that
(1) f(n) G Ri2) (5, n)
since an n-gon all whosequadrilaterals are convex is itself convex. Thus Ramseys
theorem immediately gives a positive answerto Miss Klein’s question.
(1) gives a very poor upper bound for f(n). Szekeres in fact conjectured
f(n) = 2”-2 + 1. This is Miss Klein’s result for n = 4 and for n = 5 it was
202 P. ERD6S ES

proved by Turin and E. Makai by methods of elementary geometry. n > 5 is

not settled so far. Szekeres and I proved f(n) > 2”-’ + 1 [2] (there are some
minor inaccuracies in our proof which were corrected by Kalbfleisch), and I [ 1]
proved f(n) < (21124).

Ramsey originally discovered his theorem for the purpose of some logical appli-
cations. Hajnal, Rado and I in our partition calculus [3] systematically studied
the generalisations of Ramseys theorem to nigher cardinal numbers, our results
have applications to logic and model theory, also Hajnal and Juhlsz applied
our results to set theoretic topology, but I do not discuss these transfinite appli-
cations here. Also Ramsey’s theorem has many generalisations and extensions but
I can not discuss them here. (Erdbs-Rado London Journal 1950, Nash-Williams. . .
Cambridge Phil. SOL).
It is obvious that if there are n f 2 points in n dimensional space then not all
the distances can be equal. Schoenberg and Seidel in fact determined the minimum
of the ratio of the maximal distance divided by the minimal distance. Several
years ago Coxeter asked me to determine or estimate the smallest integer f(n)
so that if there are f(n) points in n dimensional space then they determine at
least three different distances. It immediately follows from Ramseys theorem that
(2) f(n) Q R, (n + 2, n + 2)
(2) in fact is a very poor estimate, probably f(n) < c, ncz and perhaps
f(n) =(++o(l)) fi2. By fairly complicatedargumentsI can prove f(n) < exp (n* -“).

Let A (k , n) be the smallestinteger so that if there are given any A (k , n)

points in k-dimensionalspace one can always find n of them so that all their
distances are distinct. It seemsquite difficdt to determine A (k , n) even for
k = 1, only crude upper bounds are known for the general case.A (2 ,3) = 7
and Croft proved A (3 ,3) = 9 [4].
Ramsey’stheorem easily implies that to every e > 0 and n there is a B (E , n),
SO that if there are B (e , n) points in the plane then there are always n of them
Xl,..., x, which determine a convex polygon for which the angle (xi , xi ,x,)
is greater than f - e (for every 1 Q i < j < r < n).
A well known theorem of Schur statesthat if we split the integersnot exceeding
e n ! into n classesthen the equation x + y = z is satisfiedin at least one of the
classes.V.T. S6s communicated to me the following simple proof of Schur’s
theorem : Consider the partition of pairs (i , j) so that the pair (i , j), i < j,
belongsto the r-th classif j - i belongsto the r-th 8ass. By Ramsey’s theo-
rem at least one of the new classescontains a triangle (i , j , I), but then
(j -i) + (I - j) = 1 - i,
or x + y = z is solvabiein the original class.It is known that
 APPLICATION OF COMBINATORIAL ANALYSIS 203

which completes the proof of Schur’s theorem.

The determination of the exact bound in Schur’s theorem is a very difficult
problem, probably en ! can be replaced by c”, The value of fr) (3, . . . ,3)
is not known for r > 3. [5].
It seems that the following theorem of J. Sanders can not be proved so simply :
To every r and n there is an f,(n) so that if we split the integersfrom 1 to&.(n)
into r classesthere are n distinct integers o1 < . . < a, so that all the 2” - 1
SLlJllS

;r?:ei ai, cI = 0 or 1, not all ei = 0

i=1
are in the sameclass.Radosresults[lo] imply the theorem of Sanders.
Graham and Rotschild [6] have a very generaltheorem from which this follows
as a special case. They have the following very interesting problem : Split the
integersinto two (or more generally into r) classes. Is it true that there is an infinite
sequences a 1 < . . . sothat all the sums

(3) Iz Ejai ei=Oor l,notallci=O

are in the sameclass? (in (3) of courseonly a finite number of e’s are 1). It is
not even known if there is an infinite sequencea, < . , . for which a1 < a,<. . .
and al + aj, 1 < i < j < 00all belong to the sameclass.
On the other hand it immediately follows from Ramsey’stheorem that for
every 1 there is an infinite subsequence
a(lr) < . . . so that all distinct sumstaken 2
at a time belong to the sameclass.I do not know if there is an infinite subsequence
al, <... so that for every t, (t = 1, 2, . . .).a11distinct sumstaken t at a time
belong to the sameclass,the classmay depend on t.
It is easy to see that every infinite sequenceof integers contains an infinite
subsequenceso that either no two members of the subsequencedivide each
other or each term of the subsequencedivides the subsequentone. This follows
immediately from Ramsey’stheorem but perhapsis not a good example of its
use since the direct proof is easier.
Cdsdr [7] proved the following theorem which arosein his joint work with
Czipszer : Let g,(x) . . . . g,(x) be n bounded real functions and f(x) another
real function. Assume that there are two real numbers E > 0 , 6 > 0 so that
whenever f(x) -f(y) > E there is an i, 1 < i < n so that gj(x) -g,(v) > 8.
Then f(x) is also bounded. Csaszargave a direct proof of this theorem and
V.T. S6s observed that it immediately follows from Ramsey’stheorem.
A theorem of Van der Waerden states that if we split the integersinto two
classesat least one of them contains an arbitrarily long arithmetic progression,
The finite form of Van der Waerden’stheorem statesthat there is a smallestf(n)
so that if we split the integersfrom 1 to f(n) into two classesat least one of them
contains an arithmetic progressionof n terms. No satisfactory upper bound is
know for f(n), the best lower bound is due to Berlekamp [8].
204 P. ERDijS E5

Van der Waerden’s theorem also has many applications e.g. A Brauet [!?I proved
that if p > me is a sufficiently large prime then there are k consecutive qua-
dratic residues and non-residues mod p. Rado [lo] gives many interesting genera-
lisations and applications to new number theoretic and combinatorial problems.
The theorem of Graham and Rotschild [6] can be considered as a generalisation of
Van der Waerdens theorem. Finally I would like to drow your attention to a
beautiful conjecture of Rota [6] which seems very deep. Added in proof : Rotas
conjecture has been proved by Graham, Leeb and Rotschild.
Turin and I [ 1l] raised the following problem in combinatorial number theory :
Denote by r,(n) the maximum value of 1 for which there exists a sequence of inte-
gersa, <... < a, < n which do not contain an arithmetic progression of k terms.
Determine or estimate r,(n). If we could prove that for every k there is an n,(k)
so that for n > n,(k) r,(n) < n/2, then Van der Warden’stheorem would imme-
diately follow. Unfortunately this hasnever been proved.
It is known that
n 1-cl’d’w n < r3 (n) < c* n/log logn
The lower bound is due to Behrend [ 121 and the upper to Roth [ 131.Szemeredi.
[14] proved r,(n) = o(n).
I would like to mention one more old conjecture of mine from combinatorial
number theory : Let g(n) = +- 1 be an arbitrary function. Then to every E there
is a d and m so that

2 dkd)l > c
k=l
Now we give someapplications of combinatoral inequalities and extremal pro-
blems. Let u, <... < ok < n be a sequenceof integers no a divides any other.
Then it is easy to seethat max k = [(n + 1)/2]. On the other hand if we assume
that no a divides the product of two others then [ 151
(4) s(n) + cl n213< max k < n(n) + c2 n2’3/(log n)’

The proof of both the upper and the lower bound in (4) usescombinatorial
results. The lower bound usesSteiner triplets and the upper bound the trivial
result that a graph of n vertices and n edgescontains a circuit. Assume now
that all the products ai ui are distinct. Then [ 161
(5) s(n) + cj n”/4,(log n) 3/2< max k < a(n) + cq n314/(logn)3/2
Here the lower bound usesthe existenceof finite geometriesfor n = p2 + p + 1
and the upper bound usesthe following result : Let $J be a graph of t, vertices
which contains no rectangle, further assumethat there are t, vertices so that
every edge of our graph is incident to one of these vertices. Then the number
R(s) of edgesof 9 is lessthan

t, ++i] +t: (1+[$I >-’

 APPLICATION OF COMBINATORIAL ANALYSIS 205

Thus, in particular, if $$has n vertices and contains no rectangle then R (9) <cn3j2.
W. Brown and Renyi, V.T. S6s and I proved that [ 171 if 9 contains no rectangle
then

max c(g) =(++ o(1)) n3j2.

Assume now that the number of solutions of aiaj = m is bounded. Then we

have the following result : Let a, < . . . <a, <nnn>nn,(e,Z).Assume
(6) k > (1 + c) n (loglog ?&l/(1 - 1) ! log n
Then for somem the number of solutions of m = ai aj is > 2’. (6) is bestpossible
in the sensethat it fails if 1 + e is replacedby 1 - E [ 181.
(6) implies that if aI < . . . is an infinite sequenceof integers so that every
large integer can be written in the form aiaj then the number of solutions of
n = aiaj is unbounded. Now I state an old conjecture of Tudn and myself which
is an additive analogueof this result (and which in fact lead me to this result) :
Let a, <... be an infinite sequenceof integers, denote by f(n) the number of
solutions of n = a, + aj. Assume that f(n) > 0 for all sufficiently large n. Then
lim supf(n) = 00. This conjecture if true is probably very deep.
The combinatorial theorem needed for the proof of (6) statesas follows. Let r
and t be given, E = e(r , t) small and n > n,(e, r , t). Let
I3l=n and A,, . . .,A,,,u>n’-’
are r-tuples contained in 8. Then there are rt distinct elements~7) , i = 1, . . . , t ;
j= l,..., r of s so that all the r’ sets (x(il,, , ~$2,. , , x$‘!) occur among the
A’s. For r = 2 this is a theorem of KovAri and the Turzlns [ 191.
A well known theorem of Tur&n [20] statesthat in a graph of n vertices which
hasmore than
r-2
(7) f(n,r)=z(rz’ --h2)+($),n=h(mod r-l),O<hhr-1

edges there is a complete rgon. This theorem and its extensions have many
applications in geometry and potential theory. A. Meir, the Turdns and I are
publishing a seriesof joint papers on this subject. Here I state only one such
application. Let 8 be a set of diameter 1 and x1, . . , , x,,, n points in 8. Let
the packing constants of 8 be p2 = 1 >, p3 > I , i . Then at most f(n ,r) of
the distancesd(x r ,xj) are greater than p,. Let C, be the r-th covering constant
of 8, usinga theorem of Moser and myself (Australien 3. Math. XI (1970), 92-97),
V.T. Sos obtained a lower bound for the number of distancesd(x, , x,) > C,.
This will also appear in the quadruple paper.
Another application of (7) is due to Katona [21]. Let fr (x), . . . , fn(x) be
n functions which satisfy J fi(m* dx > 1. Then there are at most [n2/4] pairs
(i,i), i<j for which /(fi(x)+fi(X))2dx< 1.
I investigated the following question : Let X1, . . , , xk be n distinct points
in k-dimensionalEuclidean space. For how many pairs i, j (i <j) can we have
206 P. ERDdS E5

d(xi ,xi) = 1 ? Denote this maximum by f(k , n). For k = 2 and k = 3 I have
no good estimationsfor f(k , n), e.g. for k = 2 I only know that
(8) nl+ cm3logn < f(2 , n) < C’n3/2

It seemsthat in (8) the lower bound is close to being best possible.

For k > 4 one knows very much more. Lenz and I proved [23]

lim f(k , n)/n* = t (1 - h)

n=o
and if k = 4, n E 0 (mod 8) I proved [23]
n1
(9) f(k) n) = T+ n.

(9) follows from the following result of Simonovits and myself with will
soon appear in Acta Hungarica : Denote by $$(n; 1) a graph of n vertices and
1 edges.k (u,, . . . , ur) denotes the complete r-chromatic graph where there are
ui vertices of the i-th colour and every two vertices of different colour are joined.
We proved that every ~(n ; [$I + n + 1) contains a k (1,3, 3). This result

is best possible,there is a $n [$I + n) which doesnot contain a k( 1, 3, 3).

A well known theorem of Sperner [24] statesthat if 131= n, A, C 3, 1 G i G k

is a family of subsets,no one of which contains any other, then

max k = c-1
&y2] ,
This result and its generalisationsand extensionshas many applications. Using
(10) Behrend [25] proved that if a, < . . . < ak < n is a primitive sequence(a
sequenceof integersis called primitive if no a divides any other) then

(11) i d < c log n/(log log n)li2.

i= 1

and Pillai proved that (11) is best possible.

Using a more complicated refinement of (10) Sarkiizi, Szemeredi and I [26]
proved that if a, < . . . is an infinite primitive sequencethen
1 (log log X)1/2-’ = o
lim z -
Oi<X ai ( logx 1

We also proved that

1
max x -= f1 + O(l)) (2r l~fgw)‘,2
#i<X ai

where the maximum is taken over all primitive sequences[27].

 APPLICATION OF COMBINATORIAL ANALYSIS 207

I made strong use of Sperners theorem in my papers on the distribution func-

tion of additive arithmetic functions [28].
Let 191 = n, Ai C g 1 <i <k. Assume that the union of two A’s never
equals a third. I conjectured that then k <c Kleitman [29] proved this
(,;2,>.
conjecture as well as several other related conjectures, all of which have number
theoretic applications (301.
Sharpening a result of Littlewood and Offord [31] I immediately deduced from
(lO)thatifx>l,i= l,...,~tthenthenumberofsums

i El xi, Ei = + 1

which fall inside an interval of length 2 is at most I conjectured that

(A )*
the same holds if the x, are vectors in a Banach space (the interval of length
2 has to be replaced by a sphere of radius 1). This was first proved for the plane
independently by Katona and Kleitman [32] using an ingenious extension of (10).
Very recently Kleitman proved my general conjecture in a surprisingly simple
way without using Sperners theorem. Kleitman’s proof is not yet published.
Rado and I proved the. following theorem. Let II > 2 and b > 1 be integers.
Then there is a smallest integer f(a ,b) so that if we have f(o , b) + 1 sets each
having at most b elements there are always a + 1 of them which have pairwise
the same intersection [33]. We proved
b-l
(12) f(a,b)<b!ab=’ 1-J----L- 3!g ..--bT
2 !a
(12) is far from being best possibleand very likely
(13) f(a ,b) < cb+’ d’+’
We are very far from being able to prove (13), even for a = 2 (12) has not
even been proved with o(b!).
Sauer determinedf(u, 2) for every a. For b > 2 there are only relatively crude
upper and lower bounds for f(a , b).
(12) has many applications which could be significantly strengthenedif (13)
would be proved.
Denote by f,(n) the smallestinteger so that if
1 <a, <..‘ < aI Q n ,I = f,(n)

is an arbitrary sequenceof integers, one can always find r a’s which have pair-
wise the same greatest common divisor. First I proved by number theoretical
methods that

n
f,(n) < exp ((log n)‘12+ )
208 P. ERD6S E5

Later I observed that (12) implies that for every t and e > 0 there is an n,
so that for all n > n, (t , E)
exp (ct log n/log log n) < f,(n) < n3/4+E
(13) would imply that the lower bound gives the right order of magnitude,
Using (12) I proved 1351 that for every k there are squarefreeintegers satis-
fying (V(n) denotes the number of distinct prime factors of n)
(Ui ) Ui) = 1, cg(u,)= y?(u& V(u,) = V(Uj>, 1 Q i <j Q k
If (13) would hold we could add a&) = a@).
(12) has been improved by Abbot and others but as far as I know nobody
came close to ( 13).
Dodson [36] investigated the following problem : Denote by r*(k) p”) the
smallest value of s for which for every choice of the integers a,, . . . , Q,
s
x %Xi k s 0 (mod p”) has a non trivial solution in integersxi, i = 1, . _. , s)
i=l
(i.e not all the xi are multiples of p). In one of the cases(12) was needed.
(12) has also many applicationsto combinatorial problemsand set theory (see
Engelking and others).
Before completing the paper I want to state a few miscellaneouscombina-
torial resultswhich have applicationsin various branchesof mathematics.
Let ]s]=n,A, C;S, 1 GSiQr,r +masn+M, ]Ai]>cn, O<c< 1 for
l<i<r,r-*m as n+-
Then there are two indices i and j for which jAi n Ai I > (cz + (I( 1))n. This
statement can be proved easily by using the characteristic functions of the sets
~4, and it is easy to state various generalisationsfor the intersection of more
than two sets, one can also reformulate the result for measurablesets[J7]. This
theorem has many applications to combinatorial analysis, number theory and
analysis.
A theorem of Szekeresand myself states that if there are given 2” points in
the plane they always determine an angle > n 1 -$ . This result is best pos-
( >
sible since Szekeresshowed previously that to every E > 0 one can give 2” points
1
in the plane so that all the anglesare < s 1 + f ; see [2].
( -;;>

The fact that 2” points determine an angle > n 1 - 1 follows from the
( ,>
fact that the complete graph of 2” + 1 points is not the union of n bipartite
graphs. The sharper result that one of the anglesis > A 1 -i) follows from
(
a more careful study of the decompositionsof the complete graph of 2” vertices
into n bipartite graphs.
 APPLICATION OF COMBINATORIAL ANALYSIS 209

Probability methods have often been applied successfully to solve combinatorial

problems which seemed intractable by more direct methods and conversely com-
binatorial results often imply unexpectedly beautiful results in probability. e.g.
the arc sine law of Andersen [38], see also my paper with Hunt [39]. Finally I
want to mention that Davies and Rogers [40] rediscovered arts used a little
known theorem of Hajnal and myself on chromatic graphs in the study of Haus-
dorff dimension of sets.

REFERENCES

[l] ERDGS P and SZEKERES G. - A combinatorialproblemin geometry,Composirio

Math. 2, 1935, p. 463-470.
[Z] ERD& P. and SZEKERES G. - On some extremum problems in elementary geome-
try, Annales Univ. Sci. Budapest, 3-4, 1960, p. 53-62.
[3] ERD&S P., HAJNAL A. and RADO R. ~ Partition relations for cardinal numbers,
16, 1965, p. 93-196.
[4] CROFT H.T. - 9-point and 7 point configurations in three space, Proc. London
Math. Sci. XII, 1962, p. 400-424.
[5] GREEMVOOD R.E. and GLEASON A.M. - Combinatoxial theorems and chromatic
graphs, Canad. I. Math. 7, 1955, p. 1-7.
[6] GRAHAM R.L. and ROTSCHILD B.L. - Ramsey’s theorem for n-parameter sets,
Colloquium, Comb. theory and applications Balatonfiired, Hungary 1969.
[7] CSASZAR A. - Sur une critere d’approximation uniforme, Publ. Mat. Inst. Hung.
Acad. 8, 1963, p. 413-416.
[S] BERLEIUMP E.R. - A construction for partitions which avoid long arithmetic
progressions. Canad. Bull. Math. 11, 1968, p. 409-414.
[9] BRAUER A. - tirber Sequenzen von Potenzresten I und II, Sitrungsber. Preuss.
Aknd. Wiss. Phys. Math. Kiass 19, 1931, p. 329-341.
[lo] RADO R. - Studien zur Kombinatorik, Mufh. Zeifschriff 1933, p. 424-480.
[ll] ERD& P. and TURAN P. - On some sequences of integers. 1. London Math. Sot.,
11, 1936, p. 261-264.
[12] BEHREND F.A. - On sets of integers which contain no three terms in arithmetical
progression, Proc. Mat. Acad. Sci., U.S.A., 32, 1946, p. 331-332.
[13] ROTH K.F. - On certain sets of integers, J. London Mafh. Sot., 28, 1953, p.
104-109.
1141 SZEMER~DI E. - On sets of integers containing no four elements in arithmetic
progressions, Acfa Math. Acad. Sci. Hung., 20, 1969, p. 89-104.
[15] ERD~S P. - On sequences of integers no one of which divides the product of
two others and on some related problems, Tomsk. Gos. Univ. Vcen Zap., 2,
1938, p. 74-82.
[16] ERD~S P. - On some applications of graph theory to number theoretic problems,
Publ. Ramanujan Inst. No. 1, p. 132-136.
[17] BROWN W.G. - On graphs that do not contain a Thomsen graph, Canad. Mufh.
Bull., 9, 1966, p. 281-285, ERDOS P., RBNYI A. and SOS V.T., On a problem
of graph theory, Sfudia Sci. Math. Hung, 1, 1966, p. 215-235.
[I81 ERD& P. - On the multiplicative representation of integers, Israel J. Math., 2,
1964, p. 251-261.
210 P. ERDijS ES

[19] ERD~S P. - On extremal problems of graphs and generalised graphs, Israel J.

Math., 2, 1964, p. 183-190. KOVARI T., SOS V.T. and TURAN P. - On a
problem of K. Zarankievicz, Colloquium M&h., 3, 1950, p. 50-57.
[20] TURAN P. - Eine Extremalaufgabe aus der Graphentheorie (in Hungarian), Mat.
is Fiz. Lapok, 48, 1941, p, 436-462.
[21] KATONA Gy. - Graphs, vectors and inequalities in probability theory (in Hun-
garian), Mat. Lapok, 20, 1969, p. 123-127,
[22] ERD~S P. - On sets of distances of n points, Amer. Math. Monthly, 53, 1946.
p. 258-260.
[23] ER& P. - On sets of distances of n points in an Euclidean space, Pubf. Math.
Inst. Hung. Acad., 5, 1960, p. 165-169; ERDOS P., On some applications of
graph theory to geometry, Canadian J. Math., 19, 1967, p. 968-971.
[24] SPERNER E. - Ein Satz iiber Untermengen einer endlichen Menge, Mat/z. Zeit-
schrift, 27, 1928, p. 544-548.
[25] BEHREND F. - On sequences of numbers not divisible one by another, 1. London
Math. Sot., 10, 1935, p. 42-49, see also chapter V of the book Sequences of
H. Halberstam and K.F. Roth.
1261 Ear& P., SARKO~ A. and SZEMERBDI E. - On a theorem of Behrend, J. Austra-
lian Math. Sot. 7, 1967, p. 9-16.
[27] ERD& P., SARKOZI A. and SZEW~R&DDIE. - On an extremal problem concerning
primitive sequences, J. London Math. Sot., 42, 1967, p. 484-488.
[28] ERAS P. - On the density of some sequences of numbers I, II and III. J. London
Math. Sot., 11, 1935, p. 120-125; 13, 1937, p. 7-11; 14, 1938, p. 185-192.
[29] KLXITMAN D. - On a combinatorial problem of Erdos, Proc. Amer. Math. Sot.,
17, 1966, p. 139-141.
[30] ERD& P., SARK~ZI A. and SZEME&DI E. - On the solvability of the equations
[a,, al] = a, and (al.aj) = a, in sequences of positive density, I. Math.
Anal. and Applications, 15, 1966, p. 60-64.
[31] Ea& P. - On a Lemma of Littlewood and Offord, Bull. Amer. Math. Sot., 31,
1945, p. 898-902.
[32] KATONA G. - On a conjoncture of Erdos and a stronger form of Spencer’s
theorem, Studiu Sci. Math. Hungar., 1, 1966, p. 59-63, KLEITMAN D. - On
a lemma of Littlewood and Offord on the distribution of certain sums, Math.
Z., 90, 1965, p. 251-259.
[33] ERD& P. and RADO R. - Intersection theorems for systems of sets, 1. London
Math. Sot., 35, 1960, p. 85-90.
1341 ERD& P. - On a problem in elementary number theory and a combinatorial
problem, Math. of Computation, 18, 1964, p. 644-646.
[35] Ea& P. - Some remarks on the functions cp and G, Bull. Acad. Polon. Sci., X,
10, 1962, p. 617-619.
[36] DODSON M. - Homogeneous additive congruences, Philos. Trans. Roy. Sot.
London, Ser. A, 261, 1967.
[37] Eanas P., NEVEU I. and RBNYI A. - An elementary inequality between the
probabilities of events, Math. Stand., 13, 1963, p. 99-104. See also papers by
GILLIS, LORENT~, SUCHESTON, SHAPIRO H.S.
1381 ANDERSEN ES. - On the number of positive sums of random variables, Skandina-
visk Aktuarietidsskrift, 32, 1949, p. 27-36.
[39] ER& P. and HUNT G.A. - Changes of sign of sums of random variables,
Pacific I. Math.. 3, 1953, p. 673-687.

Mathematical Institute Hungarian Academy of Sciences

Realtanoda U. 13-I 5
Budapest V (Hongrie)