Matematicky cosopis 21 (1971), No .

DECOMPOSITIONS OF COMPLETE GRAPHS

INTO FACTORS WITH DIAMETER TWO

JUR AJ BOSÁK, Bratislava, P kL ERDÖS, Budapest (Hungary)

and ALEXANDER ROSA, Hamilton (Canada)

In the present paper the question is studied from three points of vieAv
whether to any natural number k _> 2 there exists a complete graph decom-
posable into k factors with diameters two . The affirmative answer to this
question is given and some estimations for the minimal possible number
of vertices of such a complete graph are deduced . As a corollary it follows that
given k diameters di, d2 , . . ., dk (where k 3 3 and di > 2 for i = 1, 2, 3, . . . , k) ,.
there always exists a finite complete graph decomposable into k factors with
diameters di , d2, . . . , dk . Thus Problem I from [I] is solved .

In this paper we deal only with nonoriented graphs . By a factor of a graph G

,ve mean any subgraph of G containing all the vertices of G . By a diameter-
of G we understand the supremum of the set of all distances between the
pairs of vertices of G (e . g . a disconnected graph has the diameter or,) . Tile
symbol <W', denotes the complete graph with n vertices .
Let to be a natural number . By a decomposition of a graph G into k factors,
we mean a finite system {T1, (f,2, . . . , Tkj of factors of G such that every edge
of G belongs to exactly one of the factors ~z , (P2 , . . . , TA, The symbol
Fk (d i , d2, . . . . dk) denotes the smallest natural number n, such that the complete
,graph <n.> can be decomposed into k factors with diameters d l , d2 , . . ., dk
if such an n does not exists, we put Fk(dl, d2, . . ., dk) _ oo . Further, put
f,(d) = Fk (d, d, . . ., d) . The main aim of the present paper is to find estimations
for fk (2) . From [I] it follows that f2(2) = á, 12 < f3 (2) < 13 .
Theorem 1 . For any integer k > 3 we have :

6k - 7
4k - I < fk(2)
9j, - 9

14


Proof. To prove the upper estimation it suffices to decompose the graph

6k 7
G-
2k - 2

into k factors with diameters two . The vertices of G can be represented by

(2k - 2)-tuples formed from elements 1, 2, 3, . . ., 6k - 7. The íth factor
(i. = 1, 2, . . . . k) consists of all edges joining (2k - 2)-tuples with just z - 1
common elements . The remaining edges can be added to any factor . It is
easy to prove that all the factors have diameter two .
Suppose that for some k > 4 we have fk(2) < 41 , - 2 . Then, according
to Theorem I of [1], <4k - 2> is decomposable into k factors rl?r , d-2, . .- 9~'k
with diameter two . Put n = 4k, - 2 . None of the factors (Fi (i = 1, 2, . . . , k°)
inay- have a vertex of degree n - 1 (otherwise the other factors are not con-
nected), therefore, by [4], Tr has at least 2n - 5 edges . The number of all
edges of < n' is
(n)
> k(2n - 5),
Z
whence it follows that
(1) ». { l0k > 4kn + n. .
But
n. 2 I Ok - 16k 2 - 6k -}- 4,

4kn+n=16k' -4k- 2,
thus for k > 4 we have )2 2 10k < 4kn + n, which contradicts (1) . For
k = 3 our assertion follows from [1], Theorem 7 .
Remark. The upper estimation given in Theorem 1 is too high . Therefore
we later present some methods enabling to improve it, namely for a „small" 4 .
in the second part of this article, and for a „great" k- in the third part .
Lemma 1 . Let k > 2, 2 = rlr <- d2 <- d3 ~< . . . - d k < 30- 11 'e leave
Fk(dr, d2, . . ., dk ) < fk(2) -(- d r T c1,2 T . . . + dk - 2k .
Proof. From Theorem 1 it follows that fk(2) is a natural number . If dl =
d2 = . . . = (h = 2, the assertion of the lemma is evident . Thus we can
suppose that there exists an integer is (1 5 i < k - 1) such that r1r = d 2 _
_ . . . = d 2 = 2 < d ,_ r S . . . dk . Let us construct a decomposition of the
graph

into k factors with diameters d1,, d2 , . . ., dk .

The vertex set of G consists (as we may suppose) of vertices ui, u2,
213, . . ., u),{'_) and of vertices 2'j ,r, wj , 2, Vj,g, . . ., vjdj-2 (i + 1 j <1 lt) . Ob-
viously, the total number of vertices is fk(2) + d i = d2 + . . . = dk - 2k .
The complete subgraph of G generated by the vertices U1, V2, UP3, . . . I u,x(2)
according to the definition of fk (2) can be decomposed into k factors T1, ff 2 , . . . ,
uk with diameter two . Define a decomposition of G into factors (f,~,, (in =
= 1, 2, . . ., k) thus : Into (', there belong (i) all the edges of (P „ ; (ü) all the
edges a,v ; .t (1. < s < fk(2), i + 1 5 j < k, 1 <, t < dj - 2) such that the
edge it, ul belongs to .,n. and j * m: ; (iii) all the edges of the path uiv.m,jv,,,,2 . .v
z~,,; ,,,z ., (if nz > i + 1) . 311 the remaining edges are placed into T,, .
It is easy to show that T,,, has diameter d,-„ (m = 1, 2, . . . , k) . The lemma
follows .
Lemma 2 . Let k > 3, 2 < d i ~< d2 < . . . dk < co . Then we have
Gk - 7
Fk(di,d2, . . .,d1) <_ +di+d2~ - . . . +dk- 2k .
2k - `?

Proof . Distinguish two cases :

L di = 2 . Then the assertion follows from Lemma 1 and Theorem 1
11 . di > 2 . By [11, Theorem 4, 11-e have :

Fk(di,d2, . . .,dk) < d_=d 2 =, . . .+d k -k .

-Since for any k > 2 we have

6k - 7 ~
k
2k - 2

the lemma follows .

Corollary. Let, k > 3, 2 < di < d., .. . dk oo . Then. Fk(di, (7z, . . ., dk)
is a natural )number
.

Proof. If dk < ac, our assertion follows from Lemma 2 . If d2 = x, , the

assertion follows from [1], Theorem 3 . Therefore we may suppose that d 2 < ~ .
d k = co, i . e . there is an integer i (2 i k - 1) such that 2 S di S d2

If i > 3, according to Lemma 2, Fz (di , d 2 , . . ., dr) is a natural number .

Therefore the finite complete graph
G = <F Z (d i , d2 ; . . . ; d2)

is decomposable into i factors with diameters d i . d2, . . . , d 2 . If we add k - i

null factors (i . e., factors without edges), we obtain a, decomposition of G
into k factors with diameters dl , d2 , . . . , dt , d, + i, . . . , dk .

1 (i


If i = 2, then according to Theorem 8 of [I] F 3 (di , d2 , d3 = oo) is a natural

number. Since
F k (d i , d 2 , d3 = oo, . . . ; dk - oo) S F3(di, d2, d3 = ca),

then Fk(dl, d 2 , . . ., dk:) is also a natural number . The corollary follows .

Remark . As the supposition d i < d2 . . . dk is not essential, the
preceding corollary completely soles Problem I from [1], p . 53 .

Let a natural number n and a set A c t1, 2, . . ., n} be given . A is called

an S,,,-set if each x c- {1, 2, . . . , n}, a 0 A can be written in at least one of the
following forms
x = a b,
x=a:-b,

x=2n+I-(a .+b),
where a, b E A .
Let k be a natural number . Denote by g(k) the least natural number I such
that the set {1, 2, . . . . I} can be partitioned into k disjoint SI-sets . (If such
<i natural number I does not exist, put g(k) = oo .)
Lemma 3 . fk(2) < 2g(k) + 1 for any integer k _> 2 .
Proof . Let natural numbers na and n be given . We shall call a finite graph
(without loops or multiple edges) with m labelled vertices V1, V2, . . . , v„ cyclic .
2

if it contains with each edge vzv; (i, j e {1, 2, . . ., na}) the edge vi- iv, ._i (the
indices taken modulo nz .) as well. By the length of an edge a vj we mean the
number
min {Ii-j',7n- la-j1J .
Evidently, a cyclic graph contains either every or no edge of length i for
each i c- 11, 2, . . . , [ nal2]f .
Assign to a given S,,,-set A a cyclic graph with 212 1 vertices containing
edges of length i if and only if i E A (i = 1, 2, . . ., n) . It is clear that thus
a one-to-one correspondence between cyclic graphs with 2n + 1 labelled
vertices with diameter two and S,,-sets is defined . Further, it is obvious that
to different [disjoint] S,,-sets different [edge-disjoint, respectively] cyclic
factors with diameter two of <2n + 1 ; are assigned . Therefore the assertion
of the lemma follows immediately from the definitions of fk(2) and g(k) .
Let natural numbers n, i, integers c, d and a set A c {1, 2, . . ., n.} be given .
Denote by red„c the (uniquely determined) integer r such that

17


r - e (mod 22n --IL- 1),

rj < -n .
Further, put
rM = jred,,rij,
c - d = Ired,,cdl,
c-A=le-d dEA} .
Evidently, we always have
( :`) 0<c~d<n,
c A {0, 1, 2, . . ., n} .

Lemma 4 . If it and r are such natural nunz .bers that the greatest cornrrzon• dh?isor
(2íz. + l, r) = 1 and A is an S,,-set, then r o A is an S n -set ccs aoell .
Proof. Choose x c {1, 2, . . ., n} . It suffices to prove thatt either :r c r
or there exist a, b E A such that one of the equalities

x=r-a+r b,
x= .r-a-r~b,
x=(2n.+1.)-(roa+r2b)

holds .
It is easy, to see that there is a y c {1, 2, . . ., n} such that r In fact,
as (r, 2n + 1) = 1, the congruence
rz - .x (mod 2n 1)

has a solution z c t1, 2, . . ., 2n} . If 1 < z < zz, we put y - z, and if n + 1 <
< z < 2n., we put y = 2n + 1 - z .
Since A is an S .-set, either y c- A or there exist a, b c- A such that one of
the following eases occurs
y =a.-b,
y=a+b,
y=2n!, 1-(a+b) .

If y e A, then evidently x - r - y c r - A . Let us analyze the other cases

(all the following congruences arc related to the modul 2n + 1) .
(1) y -= a - b . Obviously + r - y =_ ry = ra - rb, where ra - J- í • a,
rb-+. r-b .

is


By examining all 8 possibilities for choice of signs we find that one of the
following 4 cases occurs (we use inequality () )

x=ray- roa+rob, hence x=raa+rob,

x=ro y- raa-rab, hence x=roa-rab,
x=ray- -roa=, rab, hence x=rab-roa .,

x=ray--raa-rob-(2n 1)-roa-rob,

so x=2n .+1- (roa+rab) .

(II) y = a + b . Evidently
±ko y- ky=ka-kb- -koa4_kob,
where the again have 8 possibilities for choice of the signs . Further procedure
is the sauce as in case (1) .
(11I) y = 2n + 1 - (a + b) . We have : -j-k- o y - ky = k(2n 1) - ka -
- kb - -ka - kb - =k a a --k o b . Further we proceed as in case (1) .
The lemma follows .

Lemma 5 . Let r •, n and k be such natural numbers that

(l) 2n + 1 is a pri-rrte munber,

(2) k divides n-,
( .3) r is a prinr.itive root of 2n + 1, ( 1 )
(4) A = { q(k) , -r(zk) , y(3k), . . ., r(n) is an S .-get .
Then g(k) < n .
Proof. From (1) and (3) it follows that (r, 2n + 1) = 1 and that the numbers
r, r' r'?? , . . . , r •2 1 ? represent all non-zero residue classes modulo 21? 1.
From this fact it can be easily deduced that {r('), r°( 2), . . ., r( 701 - {1, 2, . . . . nI .
From (2) and (4) it follows that the sets A, r o A, r2 o A, . . . , rk - i o A are
mutually disjoint . They are 5n-sets, as it follows from (4) and Lemma 4 .
Therefore the set {L 2, . . ., n} can be decomposed into k; disjoint S,z -sets .
consequently g(k) < n .

Lemma 6 . li'e have : g(1) < 1, g(2) < 2 ; g(3) < 6, g(4) < 20 . g(5) < 35,
g(6) < 78, g(7) < 98, g(8) < 96, g(9) < 189, g(10) < 260 .
Proof. We use the method from Lemma 5 : we look for such a multiple 11
of k that (1) is valid and the least primitive root r of 2n - 1 satisfies (4) .
With the help of tables of the least primitive roots of primes (see, e . g . [5])
we can construct the following S,t -sets A

(i) 3 natural number r is called a primitive root of a prime number p if the numbers
,r, r 2 , r3 , . . , r-1 = 1 represent all'non-zero residue classes modulo p .

19


4= 1, >a 1, 2' = 2, A = { 11 .
1'=2, ~z= 2, r=2, A={1} .
k = 3, ~a = 6, r = 2, A = {1, 5} .
k = 4, n = 20, r = 3, A = {l, 4, 10, 16 . 18} .
1' = 5, n = 35, r = 7 . A = 11, 20, 23, 26, 30, 32, 34Í .
k = 6, n = 78 . ' = 5, A = {1, 4, 14, 16 . 27, 39, 46 . 49,556 . 58, 64, 67, 751 .
1• = 7, .rz = 98 . 7 = 2, A = {l, 6, 14, 19, 20 . 33, 36, 68, 69, 77, 83, 84,
87 . 93{ .
k = 8, n = 96, r = 5, A = 11, 7, 9, 12, 16, 43, 49, 55, 63, 81, 84 . 85` .
k = 9, n = 189, r = 2, A - {1, 5, 25, 39 . 51, 52 . 57, 68 . 76, 86, 91, 93 . 94,
119, 124, 125, 133, 138, 162 . 163, 1841 .
7, = 10, n = 260, r = 3, A = {1, 10, 18, 29, 32, 42, 52 . 55, 62, 74, 98, 99,
100, 101, 106, 114, 176, 180, 197, 201, 219, 226 . 231, 235, 237, 2551 .

To check that they- are S .,,-sets is a matter of routine . The rest of the proof
follows from Lemma 5 .
Remark . It can be easily- found that even g(1) = 1, g(2) = 2, g(3) = 6 .
By a systematic examination we can also establish that g(4) = 20, but, on
the other hand, g(5) = 30 . (The inequality g(5) < 30 follows from the fact
that A = { 1, 5, 6, 11, 14, 29}, 3 - A, 3 2 c A, 33 - A and 3 4 A are disjoint
~~ sets
.)
Theorem 2 . We have : f2(2) < 5,f3 (2) < 13,f á (2) -< 41,fá (2) 61J6(2) -< 157,
f, (2) 5 193, f8 (2) <, 193, f9 (2) 379, fio(2) <, 521 .
Pr oo f. For k -+ 5, k -+ 7 the upper estimation of fA (2) follows from Leinnias 3
and 6 . For k = 5 it suffices to apply Lemma and the preceding remark .
For L- = 7 «e proceed thus : Evidently f,(2) S f8(2), because from a decom-
position of a complete graph into 8 factors with diameter two we obtain
a decomposition into 7 factors with diameter two by unifying edges of any
two of the 8 given factors leaving the other 6 factors without any change .
Since f3(2) S 193, we have f,(2) 5 193 as well .

Lemma 7 . There exists a natural number Lh such that for all natu ats n > 1"
ure_ have : The number A, of all factors of <n y with t = [ 1- 30 log n] edges and
u.-ith a diameter greater than tuvo is less than

1 ~ 7z }
2
n t
Proof uses methods similar to those used in [2] .

20


(.
(I) Pick a vertex x of ;n. , . Let i be an integer for which

are valid . Now, we have :

?L-1

X ~~ J , ~~ ,
t_`l
t T 21 - 1
a2a
ai

«
(t-2l=1)(t-21+2) . . .(t-

(i = 1)(i 1 2) . . . 21

(i - 1)(i -1 2) . . . 21
2 21-2
n-1

n
t
(
n

YL

21-
-

21

i
0

i
1

I
}

)
i

2 -
+\
t
holds . Denote by a t the number of factors of <n) with t edges, in which the
degree of x is i . Evidently, we have :

(II) Put 1 = [ I/ 3n log n] . Prove that there is a number X, such that for
i = 0, 1, 2, . . . .1 and for every natural n > , we have

It is easy- to see that for any natural n the inequalities

nl S t,
21 -< t

n - 1
2 )
t-i
ln - 1,

t - 21
//2

2) . . . 21
2) . . . (,1. - 21)

(t-21+ 1)(t-2l
n-1

n2

n21 -i
(n - i - 1)(n-i-2) . . .(n-21.)
2

)
21 -i
t
+
2

2) . . .(t-i)

2 1


t= (i 1)(i - 2) . . . 21
X -- -- X
( -21 1)(t-21+2) . . .(t-i) (21)21-i

n, 21 -i t 21-i 1 1 1 -- 2 21 n, 21
X , -- X
(n-21) ~t-21 l 21 21 >1 n-21

4
t 21 3 )1-1 21 2i E) 15 i-1
X 5
~t - 2l + 1 4 4 ~16

5 15 bri 1
< <
4 16 »3

for every natural n > N1, if ?4'1 is a sufficiently large constant .

(III) Let us prove that the number B.,&) of the factors of 62' with t edges,
in which the degree of x does not exceed 1, is less than

(2} \
1 t
2 n2

for every sufficiently large n. .

Obviously, according to (II) for n > AT, we have :

n2 B n.(x) ao + a1 -]- . . . + ai
n2 1<

«n»
« 1~~

ao -{- a1 -{- . . . + ai ao a1 al
n 2 = n 2 1 <
a21 a21 a21 a,i

1 r [1/3n log r1 ] = 1
< n2(1 + 1} -
n3 n

Evidently, the last expression tends to zero for r1. oo . Therefore

[~1 3n log n ] + l 1
71 2


for is > ái'2, where 11:2 is a sufficiently large constant so that

n 2Bn(x) 1
<
2

1
B'.(x) <
2 n. 2

for n > max {Z1-i, 2} .

(IV) j 'e prove now that the number B,,, of the factors of <n with t edges
containing a vertex of degree < l, is less than

1 ((n)) 2Z
2n t

for it, > inax {Xj X,

Evidently . we have
B n < >7 B,z (x) ,
z
where x runs through the vertex set of <n> . Therefore, using (111) we obtain

B .n, B,,(x) < n 1 1 2} = 1 ~2}

2 n2 t 2n t

for iz > mix { - i ., X2} .

(V) Fix now two different vertices x and y of <n /, and two integers i and j
satisfying the relations l < i < n . I < j < n . .
Denote by D,,(x, y,, i, j) the number of factors of <n , with t edges in which x
has degree i, y has degree j, and x is not joined with y by an edge . We have :
n - 21
(n-2l 99
D . (x, y. i, .7) _
a / i t-i-j

Further; denote by E,,(x, y, i, j) the number of factors of <n. , with t edges

in which x has degree i, y has degree j, and the distance of x and y is greater

than two . Evidently,

( zt-?
( it -2~ln - ,2 -i,
E. i j) )
t-t J
We shall find a natural number l"3 such that for every n > 1 3 «-e have
En, (x, s, i, j) 1
&(x, y, i, j) n3
Obviously, 'we have
E,(,y,z,J) zz-i-2 n - i - 3 ?é.-i-j-1
...
D,, ( .r, ?/, i, j) v, - 2 n - 3 n - j - 1

it, 2 n-3-1 z-i

1<
it -2 ( n-2

It is easy to see that there exists a • natural number L%3 such thatt for all
n. > 1"3 Ave have
at-?->1 .

Evidently, it suffices to prove that for every n > X1'3 we have

> zz, 3 .
n.- 3-l

But for zt > 13 we have

1 n-2

í
zZ - .,
> e.
- 1
l=1

It follows that
1 ti-2 (1+1)'
IT 1 n-? >
n-3-1 n -2
l -}- 1

Q+1)' (13n log n ),

> en-2 > e '6 = n3 .

24


(VI) Let C, z be the number of factors of <n> with t edges in which all the
vertices have degrees greater than t and with diameters greater than two .
From (V) it follows that for every n . > N 3 we have :

C,, <- Fn(x, y, 2, j)

D,,(x, y, i,j) 1
D, (x, y, i, j) <
)2 3 -

7t
I!'-'I I'''I
n3- < _~7L

where (x, y) runs through the set of all unordered hairs of different : vertices
of <n) (-i, j ) runs through the set of all ordered pairs of integers such that
1<i<n, l<jGn .
(VII) Put ti = rna.x {l'i , l'2, Ll'3} . Their, according to (IV) and (VI) for
every natural number ra > -Y we have

'»
n
«2 » «2» (2
t
A n <, B12 =, C 7z <
?n ?n z.

The lemma follows .

)Lemma 8 . A v-atural n.2cm,ber lI exists such that for every irzteyer n . > ,lI
zce have : <n ;~ eoz2tains

n-?
I ]~/
L 12 log n

edge-disjoirat factors with diameter t2c~o .

Proof . According to Lemma 7 there exists a positive integer l% such that

for every integer n > V we have
721
1 22
An < (
11 t


Put

n
2
U =
t

Evidently there is a natural number 1Y4 such that for every n > A' we have 4
it < it . Put 117 = max f Y, N4} . Obviously for n > 2 we have :

n(n - 1) n(n - 1}
2[V3n3 log n] [_2 V3n 3 log n

n2 2n--, 1 n2-2n. 1 [
/[J 12n log n
>
1211, log n 12 log n '

Therefore it suffices to prove that for n > X the graph <n ; contains u. edge-
disjoint factors with diameter two .
If we assume the contrary, then each of the

Y4-1

~.-o
1 (n) t
i,)

p= nt

systems S consisting of it edge-disjoint factors of <n>, each with t edges,

contains at least one factor with diameter greater than two . Any such factor
with t edges and with diameter greater than two occurs just in

1 (2) _
4-i

it

-i t
q
(~~-1)!

systems S . Therefore the number of factors of <n ) with t edges and with a dia-
meter greater than two is at least

n. rt
p - 1 > 1 2
q ec t n t

which contradicts Lemma 7 . Thus Lemma 8 follo-vvs .

Theorem 3 . There exists a positive integer K such that for any integer k- > K
u±e have

26


49 2
fk (2) ~ 10 ) k2 log k .

Proof . Pick a natural number K, such that for every k - Ki «,e have

--) k2 log kl > -ll ,

u

10 '
where -11 is the constant from Lemma 8 .
Pick a natural number K 2 in such a wav that for any k > K 2
1,2 log k > 750,

and, consequently .
1
- 3 > - - k:2 log k .
250
Further, pick a natural number K3 such that for every integer k > Ií3
we have
~491 2
I log k k2000 .
10f

Put K = max {K 1 , K2 , K3} . Pick an integer k > K . Put

n = k .
\19l2Vlog
Then we have

k2 log k - 1 - 2 k2 log k - 3
2 Z 10)
,

log n 4 09 2
k-2 l og kl 21og k + log 410
log 9) log k
1 1

49 2 1
k2 log k - - k 2 1og k-
250
(10) = 12k 2 .
2 log k + log (k2000 )

It follows that
n-2
k S
12 log n:

`7


where n > 11 . From Lemma á it follows that <n, can be decomposed into
k edge-disjoint factors with diameter two (the remaining edges may be added
to any factor) . Consequently,
49
ft.(2) -n k 2 log k .
10)

The theorem follows .

Remark . It can be proved that there exist positive constants Ci and Cz
such that
Ci k2 < g(k) < Cz k 2 log k
for every sufficiently large 1- : the left inequality is obvious ; the right one cann
be obtained using similar methods as in our Theorem 3 and in [3] ; this re-
mains true even if we do not allow representations of the form 2n --- 1 -
- (a -}- b) . Now, using Lemma 3 we can again obtain that f2(k) < Ck 2 log k
for certain constant C and all sufficiently large k .
Problem 1 . Is g(k)11,2 bounded?
fk(2)
Problem 2. Determine lim
z"~ le

REFEREXCES

[1] Bosdk J ., Rosa A ., Znám S ., Orz decompositiorrs of complete graphs áWo factors with
ginenn diameters, Theory of Graphs, Proc . Colloq . Tihany 1966, Akadémiai Kiadó,
Budapest 1963, 37-56 .
[2] Erdös P ., Rényi A ., Orz the evolution of raardom- graphs, Ma .yar tud . akad . Ma.t
kutató int . közt. 5 (1960) . 17-61 .
[3] Erdös P ., Rényi A., Additive properties of r •a mlo-rn, sequences of positire integer'
.
Acta Arithm . 6 (1960), 83-110 .
[4] Erdös P ., Rényi A ., Egg gráfelméleti problénGról, Magyar tud . akad . Mat . kut<
int . közt . i (1962), 623-641 .
[5] Brrxorpajo13 11 . M ., OCHOQbb meopau iuee_;i, rllTT ;1, 11ocHBa- 1eHvHrpaj 1952 .

A magyar tu.domárzyos akadémi,

wernatikai kutató intézet, B-udape

llatematickg rístaz~
S1oz, enskej akadémie vied, Bratislr(xa•

J1c .Master University

Departm.e)tt of í uth.emati,cs, H(onilt%

28