See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/265681632

Problems and results on 3-chromatic Hypergraphs and

some related questions

Article · January 1974

CITATIONS READS
846 2,692

2 authors, including:

Lovász László
Alfréd Rényi Institute of Mathematics
496 PUBLICATIONS 46,221 CITATIONS

SEE PROFILE

All content following this page was uploaded by László Lovász on 29 October 2015.

The user has requested enhancement of the downloaded file.

COLLOQUIA MATHEMATICA SOCIETATIS JAN05 BOLYAI
10 . INFINITE AND FINITE SETS, KESZTHELY (HUNGARY), 1973 .

PROBLEMS AND RESULTS ON 3-CHROMATIC HYPERGRAPHS

AND SOME RELATED QUESTIONS

P . ERDOS - L . LOVÁSZ

A hypergraphi is a collection of sets . This paper deals with finite hy-

pergraphs only . The sets in the hypergraph are called edges, the elements
of these edges are points. The degree of a point is the number of edges
containing it . The hypergraph is r-uniform if every edge has r points .

A hypergraph is simple if any two edges have at most one common

point, and it is called a clique if any two edges have at least one common
point .

The chromatic number of a hypergraph is the least number k such

that the points can be k-colored so that no edge is monochromatic . As
far as we know families of sets with chromatic number 2 were first inves-
tigated systematically by M i 11 e r (who used the term property B) in the
case of infinite edges . There now is a large literature of this subject both
for finite and infinite sets.

The main idea behind our investigations is that being simple or being
a clique imposes surprisingly strict properties on 3-chromatic hypergraphs .

- 609 -
The reason why we relate these two properties with chromatic number is
the following trivial observation :

If a hypergraph has chromatic number > 3 then it has two edges

with exactly one common point.

Let Mk (r) be the minimum number of edges of a (k + 1)-chromat-

ic r-uniform hypergraph . It is known [5], [9]

r-1 < m2(r) G r22r .

r+ 2 2
Perhaps r2r is the correct order of magnitude of m 2 (r), it seems likely
that
m (r)
2r -> Co

A stronger conjecture would be : Let {Ek )k=1 be a 3-chromatic (not nec-

essarily uniform) hypergraph . Let
m
f(r) = min 1 - ,
k=1 2 1Ek1

where the minimum is extended over all hypergraphs with min I Ek I = r.

We conjecture that f(r) -> - as r - - .

Let nk (r), mk (r) denote the minimum number of points and edges
in a (k + 1)-chromatic r-uniform simple hypergraph . We shall prove

Theorem 1 .

lim }~n~*(r) = k ,
r >~
r
lim ~ mk(r) = k2 .
r->

Thus in particular,
4r
c1 < m2(r) < c 2 r4 4' ,
r
i.e. m2(r) is much larger then m 2 (r) .

- 61 0 -
 In fact, we will prove a more general theorem (Theorem 1') which
constructs small hypergraphs with large chromatic number and girth ; see
[4] .

L o v á s z[ 6] and W o o d a 11 [7] proved that in every 3-chromatic

r-uniform hypergraph there is a vertex of degree > r. We improve this
result showing

Theorem 2. A (k + 1)-chromatic r-uniform hypergraph contains an

edge which is intersected by at least kr -1 14 other edges. Thus, the valen-
-1
cy of at least one vertex is > kr /4r .

S t r a w s formulated the following problem : Is there a function f(k)

such that if S in any set of integers with I SI = f(k) then the integers
can be k-colored so that each color meets every translated copy of S
(i.e. every set of form S + a = {x + a : x E S} ) .

A stronger form of this problem asserts that if f(k) is large then

each color occurs at least (1 - e) - k) times in each and similar statement
hold for the lattice points of the n-dimensional space . This problem will
follow from the method of the proof of Theorem 2 . In fact, a general the-
orem on hypergraph coloration can be obtained :

Theorem 3. If each edge of an r-uniform hypergraph H meets at

most kr -1 /4(k - 1)' other edges then the vertices of H can be k-colored
in such a way that each color meets each edge . We also prove the stronger
version of S t r a u s s' conjecture (Theorem 4.)

For simple hypergraphs, we will prove the following sharpening of

Theorem 2 :

Theorem 5 . If H is a simple (k + 1)-chromatic r-uniform hyper-

graph then it contains at least ks-2 f4 (r- 1) points with degree
~kr-2 /4(r- 1) .

This theorem will be needed to prove Theorem 1 . E r d ő s and S h e l a h

[3] observed that in every simple 3-chromatic r-uniform hypergraph there
are two disjoint edges if r is large enpugh . Theorem 5 will imply
 Corollary 2 to Theorem 5 . A simple (k + 1)-chromatic r-uniform
kr-2
hypergraph contains independent edges .
4r (r - 1)

The previously mentioned result of L o v á s z and W o o d a I 1 states

that, if H is a hypergraph such that, for each H' C H,

(1) I E UEH' E I > IH'i + 1

then H is 2-chromatic . W o o d a 11 made the surprising observation that

(1) is best possible in the sense that there is an r-uniform 3 chromatic hy-
pergraph H such that (1) holds for each H' C H (but, of course, not for
H'= H) . In Wood all's example I H I - r! and we suspect that I H I
cannot be much smaller . We also conjecture that for simple hypergraphs
(1) can be replaced by a much weeker a assumption . Perhaps

U EI > I H'1/2`(1
`E
(b'H' C H)

EEH'

will imply that H. is 2-chromatic, provided H is simple.

Consider now r-uniform cliques . Obviously, a clique can have chromat-

ic number 2 or 3 only ; we are interested in those with chromatic number
3 . Let m**(r) denote the minimum number of edges in such a hyper-
graph ; we prove
r- 1
Theorem 6 . m (r) < 7 2 for infinitely many r .

We do not know if }gym**(r) is greater than 2 ; we cannot even show

m**(r) > m(r) .

Somewhat surprisingly, there are only finitely many 3-chromatic r-

uniform cliques for a given r, so we may ask for the maximum number
M(r) of edges in them . We have the inequalities

Theorem 7. r! (e - 1) < M(r) C rr .

To obtain the upper bound we only use the fact that the edges of a
3-chromatic r-uniform hypergraph cannot be represented by r - 1 points .

- 612 -
 Theorem 8 . Let N(r) denote the maximum number of points in a
3-chromatic r-uniform clique. Then
2(2r
2)
.
Zf rr - 11)
S h e l a h and the authors observed that if H is a 3-chromatic r-uni-
form clique then there are two edges E, F with
r
IEnF1>-
log r
Perhaps the right hand side can be replaced by c • r or even r - c, since
the worst example we have is an r-uniform 3-chromatic clique with

IEnFI<r-2

(for infinitely many values of r), and we have no single example with

IEnF1<r-3 .

Theorem 9 . If r is large enough and H is an r-uniform 3-chromat-

ic clique then the cardinalities I E n F1, E, F E H take at least 3 distinct
values.

We make some further remarks on the distribution of I E n Ff, where

E, F are edges in 3-chromatic cliques, but we know here very little .

Finally, we consider the following problem . Denote by q(r) the

smallest integer for which there is an r-uniform clique which cannot be
covered by less than r points (r points, obviously, always cover an r-
unifonn clique ; e .g. the r points of an edge) . We prove
8 r-
Theorem 10 . 3 < q(r) < c • r3 r 2 log r.

It is a challenging problem to prove or disprove q(r) < c • r. We feel

sure that q(r) < c • r • log r holds .

- 61 3 -
 We prove the following statement, which yields the upper bounds
(í.e. lim Ynt*r < k, lim r < k 2 ) in Theorem 1 . The lower bounds
will be proved later (Corollary 2 to Theorem 2 and Corollary 3 to Theo-
rem 5) .

Theorem 1' . Let s>,2, r > 2, k l 2 ; n=4- 205- 1 r3s- 2

0 - 1)("+ 1), m = 4 • 20s , r 3s- 2 , ks(r+ 1), d = 20r2 , V- 1 .

Then there exists an r-uniform hypergraph H on k • n points with

at most m edges and with degrees < d which does not contain any cir-
cuits of length < s and in which each set of n points contains an edge .

This hypergraph is, obviously, at least (k + 1)-chromatic .

Proof. S be any set of k • n points . We construct our hypergraph

H= {Et : i = 1, . . . , t} inductively . Suppose E 1 , . . . , P have already
been chosen so that

(a) E 1 , . . . , EP from no circuit of length < s,

(j3) no point is contained in more than d of them .

Let S1 , . . . , SX be those n-element sets containing no one of E l , . . . ,

p
.
P
If there is no such S, we are finished . Suppose x P > 1 . Choose now
EP + 1 in such a way that E 1 , . . . , EP + 1 satisfy (a) and (0) and EP + 1
is contained in as many SI , ( 1 < i < xP ) as possible . We will show that
this is possible and that EP + 1 will be contained in at least 20 xP /k' sets
as long as p < m . This will imply

(2) XP+ 1 < xp (1 - 20k,: )

Suppose we know that if p < m then (2) holds . Then

-m kn- m
m kn , e2okr < e 2okr
xm <X01 - = 1
201kr 1 < 2

- 61 4 -
thus our procedure stops before the m-th step, i .e. we get a hypergraph
satisfying the requirements with < m edges .

We still have to show (2) . Suppose s = 2s' is even ; the odd case can
be treated similarly . Let 1 < j < xp ; we estimate how many r-tuples of
S could be chosen for Ep+ 1 without violating (a) and (0) .
Let N be the number of those points of S with degree d. Then

d •N'<r •p <r •m , N<

r• m
d = rn .
Therefore, the number of those points in Sj with degree < d is

n -N> n(l - r) .
Any r-tuple chosen from these points will satisfy (g) . Let us see, how
many r-tuples are excluded by (a) . We can describe these r-tuples as those
not containing any pair of points which is at distance < 2s' - 1 in
{E1 , . . . . Ep }; or which are both at distance < s' - 1 from a certain edge
Et , 1 < i < p . Now there are at most rr • ds - 1 points at distance
< s' - 1 from Ei ; therefore, Et excludes at most
ds' -1
< r 2s' , d2s' - 2
i 2
pairs and so, there are at most
r 2 s' d2 s'- 2 < r2 s' d2 s'- 2
p, m

excluded pairs . One excluded pair forbids at most

( r-
2)
r-tuples of
Si ; thus, the total number of r-tuples of Si forbidden by (0) is
< (n - 2 1 , r 2s'd2s'-2
m
r-2
and so, the number of r-tuples of S which are candidates for Ek+ 1 is

~ n~ l n- 2
r
m • r2s' , d2s'- 2
r -( r-2) •

- 615 -
 eínÍ r
_ m• 2s' +2 • d 2s' - 2 ~n)
.
[n) = ( e 4 ) r, > 1

Thus, there are altogether

xP (n)
20 r

r-tuples of S 1 , . . . , Sx which can be chosen .

P

Since the total number of r-tuples is there must be an r-tuple

tknl1r
which is counted in at least

xp - xp
(r)
~ 20k r
20(kn
lr
n-tuples . This proves (2) .

2.

Lemma . Let G be a (finite) graph with maximum degree d and

vertices v l , . . . , v n . Let us associate an event A j with vi (i = 1, . . . , n)
and suppose that A 1 is independent of the set

{Aj : (vi , v ) E E(G) } .

Also suppose

(3) P(A I ) I
4d

Then

(4) P(A 1 . . . A n ) > 0 .

Proof. We prove more, namely that

(5) P(A 1 IA 2 . . .A n )- 2d .

This formula makes sense because we may assume by induction

- 616 -
 P(A 2 . . . A n ) > 0 .

Then (5) obviously implies (4) .

We prove (5) by induction on n . For n = 1 it is trivial . Let

v 2 , . . . , vq be the points adjacent to v l , (q < d + 1) . Then we have

_ P(A I A 2 . . .A q IA q+I . . .A n )
P(A I I A 2 . . . A n )
P(A 2 . . . A q IA q + I . . . A n )

Here, by (3)

P(A I A 2 . . . A g 1A q+I A n ) <,

P(A I 1A q+I . . .A n ) =P(A1)<

1 '

and on the other hand

P(A 2 . . .A q IA q+I . . .A n )=

= 1 - P(A 2 + . . . + Aq JA q + I . . .A n )>

4
>1- Y-, P(A i 1A q+1 . . .A n )>1-( q -1)
i= 2 2d > 2

by the induction hypothesis . Thus

1 1 _ 1
P(A I 1A2 . . .An)~
4d 1 2 2d

This proves the lemma .

Proof of Theorem 2 . Let us color each point of H with colors

1, . . . , k at random, independently of each other and with equal proba-
bility . Let E(H) _ {E 1 , . . . , Em } and let A i denote the event that E i
is monochromatic . Then

P(A i ) I .
= kr1
Let G be the line-graph of H i .e . a graph with points v l , . . . , v m
where v i is adjacent to i iff Ei n Ej * 0 . Then the events A t are as-
sociated with the points of G and obviously, A i is independent of the

- 617 -
set of all A I-'s such that E- n E i.e . (v i , vi ) q E(G) . Moreover, the
maximum degree of G is, obviously d < kr -1 /4 and thus

P(A, ) = k,. l 1 c 4d
í.e. (3) is satisfied . Thus, the lemma gives

P(Á1 . . .Am)>0 .

But in any case when A, . . . A m occurs we get a k-coloration of H .

Thus H is k-colorable .

Corollary 1 . If each point of an r-uniform hypergraph H has degree

kr -1 /4r then the chromatic number of H is < k .

Corollary 2. If H is a simple (k + 1)-chromatic r-uniform hyper-

graph then I V(H)I > c • kr -1

Proof of Theorem 3. Let H = {E 1 . . . Em 1 . Color the points of H

with colors 1, . . . , k at random, independently of each other . Let A 1
denote the event that E- does not get all colors . Obviously,

P(A i ) < k( 1 - k~ r
Considering the line-graph of H again, we get that the maximum degree
is

d S k` 1 14(k - ly

by the assumption, thus

P(A I ) <, 1
4d
holds, and the lemma implies that P(A 1 . . . A m ) > 0 ; this means there
exists a desired coloration .

Theorem 3 immediately implies the weak form of S t r a u s s' con-

jecture ; in fact, f(k) = c • k • log k will be appropriate . The stronger ver-
sion would have a similar generalization to hypergraphs, but it would be

- 618 -
lengthy to formulate it, so we leave it to the reader and only prove

Theorem 4. Let e > 0, k > 2, n > 1 . Then there is an r o = r o (k, e)

such that if S is any set of lattice points in the n-dimensional space with
I SI = r > r p then the lattice points can be k-colored so that each set
S + a obtained by translating S with an integer vector a contains at
least (1 .- e) k points of any given color .
Proof . By compactness, it suffices to show this for a finite collection
H of translated copies of S. Let us color the vertices of H with one of
k given colors at random, independently of each other . The probability
of the event AÍ that the i-th translated copy contains < (1 - e) rk of
a given color is

P(A1)<(I-5)r (5>0)

where S depends on k and e but not on r (this follows from the

central limit theorem) . On the other hand, each copy of S meets less than
r2 other copies (since if S + a meets S + b then b - a must be one
of the vectors joining two points of S) . Thus if

(1 -S)r< 1
4r 2
then we can conclude as in the two previous cases .

Proof of Theorem 5 . Let, for each edge E E H, Z(E) be a point of

E with maximum degree and set E' _ E - {t(E)}, H' _ {E' : E E H} .
Obviously, H' cannot be k-colorable (any k-coloration of H' would
yield one of H) thus by Theorem 2, H' contains a vertex of degree
> k r- 2 14(r - 1). Let E l , . . . , E' be those edges of H' containing x,
t > e -2 14(r -1) . Then t(E l ), . . . , t(Et ) must have degree > t in H
by definition, which proves the assertion .

Corollary 1 . A (k + I)-chromatic r-uniform simple hypergraph can-

not be covered by less than kr- 2 /4(r - 1) points.

Proof. Suppose T covers all edges where I T I < kr - 2 /4(r - 1) . By

- 619 -
Theorem 5, there is a point x with degree > I TI not belonging to T.
But then T cannot cover all edges adjacent to x as H is simple, a con-
tradiction .

This assertion immediately implies

Corollary 2. A (k + I)-chromatic r-uniform simple hypergraph con-

tains at least kr -2 /4r(r - 1) disjoint edges.

Corollary 3. A (k + 1)-chromatic r-uniform simple hypergraphs has

2) 11 br(r - 1) 2 edges.
at least k 2(r-

3.

This paragraph contains constructions of 3-chromatic r-uniform cliques,

and proves some simple properties in general .

(a) All r-tuples from 2r - 1 points form a 3-chromatic r-uniform

clique.

(b) Let S be a set, I SI = 2r - 2 . For each partition P = {S P S2 1

of S with I S l 1 = 1 S21 = r - 1 take a new point x p . Define H to
consist of all r-tuples from S plus all r-tuples of the form S l u {xp }
where P = {Sl , S 2 } is a partition as above . Then H is a 3-chromatic r-
uniform clique .

(c) Let H be a 3-chromatic r-uniform clique . Let T n V(H) _ 0,

I TI = r + 1 and define H' to consist of T and all (r + 1)-tuples of the
form E u {t}, E E H, t E T. Then H' is an (r + I)-uniform 3-chromatic
clique.

(d) Let H be a 3-chromatic r-uniform clique, V(H) _ { 1, . . . , n}.

Let H 1 , . . . , Hn be 3-chromatic p-uniform cliques, V(H) n V(H) _ ~ .
Define

H * = {Etl U . . .UE1r : Ei EH ., {i,, . . . , i r } E H} .

Then H* is a (pr)-uniform 3-chromatic clique .

The proof is straightforward in all cases . Obviously, (c) and (d) yield

-620-
several families of 3-chromatic cliques when applied with different initial
3-chromatic cliques . We will use two initial hypergraphs, the triangle and
the Fano plane on seven points . Let us collect the consequences of the
above constructions .

Proof of Theorem 6 . Apply (d) inductively, with H(1) the Fano

plane, H = H (k) and H l , . . . , H, V(H)i Fano planes to get H(k + 1) Then
H(k) is 3k-uniform and

1 H(k+1)1 = 73k . 1H(k) 1 ,

3k-1
whence IH(k)I= 71+3+ . . .+3 k-1 _ 7 2

Proof of Theorem 7.

I . Starting with the triangle, apply (c) repeatedly . It is easy to see that
the obtained r-uniform 3-chromatic cliques have (e - 1)r! edges .

II . Suppose there is a 3-chromatic r-uniform clique with more than

r' edges . In fact, we will not use that H is 3-chromatic only that H is
an r-uniform clique which cannot be covered by less than r points .

Let E E H . Since there are - r' other edges, there will be a point
x l E E with degree > r r-1

Let us define x 1 , . . . , x r inductively as follows . Suppose x 1 , . . . , xi

are defined in such a way that more than r r- i edges contain all of them .
Since x 1 , . . . , xi do not cover all the edges, there is an edge El not
containing any of them . All the edges containing x 1 , . . . , x i meet Ei ;
therefore, Ei contains a point xi+, such that more than r' -i- 1 edges
contain x, , . . . , x i and x i+, '

Now more than one edge contains x l , . . . . xT , a contradiction .

Proof of Theorem 8 .

l . The lower bound immediately follows from construction (b) .

Il. For every x E V(H), there are two edges E, F E H with E n F =

_ {x} ; for let E be any edge adjacent to x, and consider E - x . This

- 621 -
set does not cover all edges, therefore there is an edge F avoiding E - x.
Since E n F 0 0 we must have E n F= {x} .

Thus the assertion will be implied by the following

Lemma. If H is an r-uniform clique such that, for each point x,

there are two edges with E n F= {x} then

IV(H)1< 2' 1 2 r - 1)'

This is a sharpening of a theorem of C a l c z y n s x a- K a r l o w i c z
The proof uses a method due to L u b e 11 [2 ] .

Proof. Let (x 1 , . . . , xn ) be a permutation of V(H) . There is at

most one point xj such that both {x 1 , . . . , x,) and {x i, . . . , x n } con-
tain an edge, since H is a clique. If we count this for each permutation
of the points each point x is counted; in fact, if E, F E H with
E n F = x then order E U F so that the points of E be on the first r
plances or on the last r places. This can be done in 2(r - 1)!2 ways ;
then choose the 2r - 1 places of E U F, this can be done in n
(2r - 1
ways ; finally, place the n - 2r + 1 remaining points on the remaining
places, this can be done in (n - 2r + 1)! ways. Thus the number of times
x is counted is
2nr.
2[(r - 1)!] 2 n (n - 2r+ 1)! =
í 2r-1 r (2r-11
J
r

Thus we count at most n! points, each point at least

~2r r 1 .
times . Hence I V(G) I c
2
Set, for a hypergraph H,

A(H) = {IEn FI : E,FeH,E F} .

Let H be a 3-chromatic r-uniform clique . The same proof as that of The-

orem 7 yields

- 6 22 -
 max A(H) > r
log r
We don't know how sharp this estimation is ; the construction in the proof
of Theorem 6 yields a 3-chromatic 3 k -uniform clique with

A(H)={1,á, . . .,3k-2} .

But we do not know any example with

maxA(H)<r-3 .

Also note the interesting property of the preceding example that

A(H) consists of odd numbers only . How "lacunary" can A(H) be? We
cannot even prove

IA(H)I - - as r -+ 00

for r-uniform 3-chromatic cliques . The best we can show is I A(H) I > 3
for large enough r.

Proof of Theorem 9 . Suppose I A(H) I < 2 . We know 1 E A(H) . As

known, IHI > 2r -1 . Similarly as in the proof of Theorem 4, we find two
points x, y contained in at least 2r -1 Ir 2 edges F 1 , . . . , Ft . Any two
of these have at least two points in common . Hence I A(H) I = 2, say
A(H) _ {1, k} . Any two of the edges containing x and y must have
exactly k points in common . As
1
t> 2`2 >r 2 -r+ 1
r
if r is large enough, a theorem of D e z a 18] implies F n F= M= const .
for any i, j. Now any edge not covered by M has at least t > r points,
a contradiction .

4.

Modifying slightly the definition of 3-chromatic r-uniform cliques,

let us consider now r-uniform cliques which cannot be covered by less
then r points . As pointed out, the proof of Theorem 4 works, so for the
maximum number of edges in such a clique we have the same bounds as
for M(r) .
- 623 -
 The question of üAe minimum number q(r) of edges is more con-
fusing (or more interesting), as Theorem 10 shows .

Proof of Theorem 10 . 1 . Suppose there are < 3

r - 3 edges in an
r-uniform clique H, we show it can be covered by r - 1 points . Let x l
be a point of with maximum degree, let x be a point of H - x l
H 2

with maximum degree, etc ., let xi+ 1 be a point of H - x - . . . - x l i

with maximum degree (H - x, - . . . - x denotes the hypergraph ob- p

tained by removing all edges which meet {x, , . . . , x,}) . Observe that the
degree of x in H - x - . . . - x is > 4 if I H - x , - . . . -x,!>
i+ 1 l t

>2r+ 1 ; it is _>3 if I H - x, - . . . - x I > r + l, and it is > 2 if i

I H - x, - . . . - x I > 1 . (This immediately follows from the assumption

t

8
that H is a clique) . Hence, if there are r edges to begin with, in

step we get down to < 2r + 3 edges, in another - steps we will

6 3
only have < r + 1 edges, which can be covered by - r points. These
5

are r points altogether . The accurate calculation with integral parts

yields that if < 8 r - 3 then, in fact, we use only r - 1 points to
I HI

cover all edges .

11. For sake of simplicity let r = p" + 1 and our edges will be lines
r + 1
of a finite plane . Set t = 4r312 log r . We can choose í lines
I r2 - t )
2 ~
o ~r -r+
ways ; we will show that all but choices of the lines cannot
r 11
be represented by fewer than r points .
To prove this we make a few simple known remarks about lines in a
finite geometry . Let v 1 , . . . , y _ be vertices and 1P . . .
r
1
be the Ilk

lines determined by them . Let e be the number of v 's on l. . Clearly

t l

(12) = 6r 1l
tZ l2 2

Let el > e ? . . . _>

2
and let B be the number of lines disjoint from
ek

{v 1 , . . . .
vk Simple computation shows
} .

- 62 4 -
 k
B= ~, (ei - 1) + 1
i= 1

and so from (12)

k k
(13)l (r-1)(r-2)=iZei(ei-1)<e, (ei-1)= e,(B-1) .

Another simple argument shows

(14) B>el (r-el ) .

What we need is the following

Lemma.

~_r (r- ~_
r) if e l <r-C,
B>
r - 1 otherwise .

This immediately follows from (13) if e l < C or e l > r - C,

and from (14) if < el < r - hr_ .

Now we are ready to prove out theorem . Assume first e, >r-Y7

r.
The number of ways of choosing such a system of points is
-r+1 <r3C
<(r2-r+
1)lr)lr2 ~-r
Cr
Thus the number of choices of t lines which can be represented by a sys-
tem of r - 1 points with more than r - }fir on a line is
- 2r + 2 l
< r3 C . J
l r2 t
and so the percentage of such choices of t lines among all choices is

- 2r+ 21
,~- rJ r
<r3 yr lr2 t <r3Cr' l 1 r-1~ =0(1))
lr2 -r+ 1)
t

Suppose now el < r - {fir . We can only say that the number of ways of
choosing r - 1 such points is

- 625 -
 < r2 - r + 1 < .
(er)r
l r-1
The number of choices of t lines covered by such systems of r - 1 points
is
< _
-r+ 1 -~ r (r-1r)1
J
lr2 t

Hence the precentage of such choices among all choices is

(er) (r r 2 -r+ 1 -~(r-C)

r
t
- r+ 1)
< (er)r 1 - rr(r r + 1) = 0(1) .

lr2 t

We remark that we feel the natural boundary of the method is r log r .

The first part of the proof, i .e. the case e l > r - > r could be improved
easily to yield this . However, if e l is small then sometimes - very rarely -
B can be close to r 3J2 (let, say, v l , . . . , y r _ 1 be points of a subplane
of order yr - 1). We have no good estimation how often can this happen .

Added in proof. Recently J . Beck (Budapest) proved that m(r)/2 r

-~ (oral communication) .

REFERENCES

[1] M . Cal c z y n s k a- K a r l o v i c z, Theorem on families of finite sets,

Bull. Acad. Polon. Sci. ser. math. astr. phys., 12 (1964), 87-89 .

[2] D . L u b e 11, A short proof of Sperner's Lemma, J. Comb. Th ., 1

(1966), 1299 .

[3] P . Erdős - S . She 1 a h , Separability properties of almost disjoint

families of sets, Israel J. of Math., 12 (1972), 207-214 .

[4] P . E r d ő s- A . H a j n a 1, On chromatic number of graphs and

finite set-systems, Acta Math . Acad . Sci. Hung ., 17 (1966), 61-99 .

- 626 -
[5] P . E r d ő s, On a combinatorial problem II . Acta Math . Hung., 15
(1969), 445-447 .

[6] L . L o v á s z , Graphs and hypergraphs, Beitrüge zur Graphentheorie,

Leipzig, 1967, 99-106 .

[7] D . R . W o o d a 11, Property B and the four colour problem, to ap-

pear.

[8] M . D e z a, Solution d'un probléme de Erdős - Lovász, J. Comb.

Theory, 16 (1974), 166-167 .

[9] W . M . S c h m i d t, Ein kombinatorisches Problem von P. Erdős and

A. Hajnal, ibid, 373-374 .

View publication stats