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

net/publication/225927716

What we know and what we do not know about Turán numbers

Article  in  Graphs and Combinatorics · June 1995

DOI: 10.1007/BF01929486

CITATIONS READS

111 185

1 author:

Alexander Sidorenko

68 PUBLICATIONS   1,059 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Turan numbers View project

Spectral Estimation View project

All content following this page was uploaded by Alexander Sidorenko on 25 August 2019.

The user has requested enhancement of the downloaded file.

 What we know and what we do not know
about Turán numbers ∗
Alexander Sidorenko

Courant Institute of Mathematical Sciences

New York University
251 Mercer St., New York, N.Y. 10012

Abstract
The numbers which are traditionally named in honor of Paul Turán
were introduced by him as a generalization of a problem he solved
in 1941. The general problem of Turán having an extremely simple
formulation but being extremely hard to solve, has become one of the
most fascinating extremal problems in combinatorics. We describe the
present situation and list conjectures which are not so hopeless.

1 The definition and equivalent formulations

A system of r-element subsets (blocks) of an n-element set Xn is called a
Turán (n, k, r)-system if every k-element subset of Xn contains at least one
of the blocks. The Turán number T (n, k, r) is the minimum size of such a
system. The problem of determining T (n, k, r) was posed by Paul Turán [57].
The Turán numbers sometimes appear in different notation. For instance,
the covering number C(n, m, p) is defined as the minimum number of m-
element subsets of Xn needed to cover all p-element subsets (n ≥ m ≥ p).
Obviously,
C(n, m, p) = T (n, n − p, n − m) .
∗
to appear in Graphs and Combinatorics 11 (1995)

1
Let U (n, q, r) be the minimum number of subsets of Xn whose sizes are at
least r and for which the size of any transversal (i.e. a subset intersecting
each of them) is at least q. It is easy to see that
U (n, q, r) = T (n, n − q + 1, r) . (1)
It was shown in [45] that T (n, k, r) is equal to the minimal length of a dis-
junctive normal form with n Boolean variables, which takes value 1 if at least
k variables are equal to 1, and takes value 0 if less than r variables are equal
to 1.

2 Recursive inequalities
Schönheim [39], and independently, Katona, Nemetz and Simonovits [19]
showed that  
n
T (n, k, r) ≥ T (n − 1, k, r) . (2)
n−r
Indeed, if we omit one element from a Turán (n, k, r)-system together with all
blocks containing this element, we get a Turán (n − 1, k, r)-system. Omitting
an element in n possible ways, we get n such subsystems; each of them has
at least T (n − 1, k, r) blocks. Every block of the (n, k, r)-system appears in
n − r of the considered subsystems. Hence (n − r)T (n, k, r) ≥ nT (n − 1, k, r).
Inequality (2) implies that the ratio T (n, k, r)/(nr ) is non-decreasing. Thus
the limit
T (n, k, r)
t(k, r) = lim n


n→∞
r
exists and  
n
T (n, k, r) ≤ t(k, r) . (3)
r
The values t(k, r) were obtained only for r = 2 (except the trivial case
k = r). Erdős [9] offered a reward for determining t(k, r) for a single pair
(k, r) with k > r > 2.
A stronger version of inequality (2) was given in [43] : if the ratio
T (m, k, r)/(mr ) is constant for m = n − l, . . . , n − 1 (which means that
the extremal (n − 1, k, r)-system is an exact l-scheme) and T (n, k, r)/(nr ) 6=
T (n − 1, k, r)/(n−1
r ), then
 
nT (n − 1, k, r) + l
T (n, k, r) ≥ . (4)
n−r

2
 Given a Turán (n − 1, k, r)-system A and a Turán (n − 1, k − 1, r)-system
B, a Turán (n, k, r)-system can be produced in a simple way ([41]). We
assume that A and B share the same element set. By adding a new element
v to every block of B we get a system B + v whose blocks are r-element
subsets. The union of A and B + v is a Turán (n, k, r)-system. This gives
the inequality

T (n, k, r) ≤ T (n − 1, k, r) + T (n − 1, k − 1, r − 1) . (5)

Applying (2), we have

r n−r
T (n, k, r) = T (n, k, r) − T (n, k, r)
n n
≤ T (n, k, r) − T (n − 1, k, r) ≤ T (n − 1, k − 1, r) .

Thus,
T (n, k, r) T (n − 1, k − 1, r − 1)
n

≤ n−1


r r−1
which results in
t(k, r) ≤ t(k − 1, r − 1) . (6)
Obviously, the union of disjoint Turán (n0 , k 0 , r)- and (n00 , k 00 , r)-systems
is a Turán (n0 + n00 , k 0 + k 00 − 1, r)-system. Moreover, the union of l disjoint
Turán systems with parameters (ni , αi + 1, r) where i = 1, 2, . . . , l , is a Turán
system with parameters (n1 + n2 + . . . + nl , α1 + α2 + . . . + αl + 1, r). This
implies the inequality
l l
! l
X X X
T ni , αi + 1, r ≤ T (ni , αi + 1, r) . (7)
i=1 i=1 i=1

3 Lower bounds
The best known general lower bound is
   
n−k+1 n k−1
T (n, k, r) ≥ . (8)
n−r+1 r r−1
A particular case of (8) was proven for k = 5, r = 4 by Giraud [15], and
then generalized for k = r + 1 by Sidorenko [42]. In the most general case,
inequality (8) was proven by de Caen [3].

3
 γ
It is sufficient to use (7) and (8) to get the correct magnitude of T (k
 , k, r)
k−1
with a fixed γ and k → ∞ . Namely, inequality (7) with l = r−1 , α1 =
α2 = . . . = αl = r − 1 yields
 r
1 kγ
γ
T (k , k, r) ≤ l · ≤ c0 k (γ−1)r+1 .
r! l
On the other hand, inequality (8) gives
k γ−1
r
T (k γ , k, r) ≥ (1 + o(1)) k ≥ c00 k (γ−1)r+1 .
r
It also follows from (8) that
1
t(k, r) ≥ k−1

, (9)
r−1

1
t(r + 1, r) ≥. (10)
r
For odd r, the last inequality was slightly improved by Giraud [17]:
2
t(r + 1, r) ≥  p , r ≡ 1 mod 2 . (11)
r 1 + r/(r + 4)

The corresponding upper bound (see the next section) is

( 21 + o(1)) ln r
t(r + 1, r) ≤ .
r

Conjecture 1.
lim r · t(r + 1, r) = ∞
r→∞

This was posed by de Caen [5] and he offered a reward for the first proof or
disproof.

4 Upper bounds
According to (3), any upper bound for t(k, r) yields an upper bound for
T (n, k, r). Thus we are concentrating on the bounds for t(k, r). We start
with k = r + 1 and then proceed to the general case.

4
4.1 The case k = r + 1
It was found first that t(r + 1, r) = O(r−1/2 ) (see [41]). This fact follows from
the inequality  
2s −2s
t(2s + 1, 2s) ≤ 2 (12)
s
and inequality (6). A major improvement came when Kim and Roush [20]
proved
1 + 2 ln r
t(r + 1, r) ≤ .
r
Their result was consecutively improved by Frankl an Rödl [12] :
ln r + O(1)
t(r + 1, r) ≤ , (13)
r
and by Sidorenko ([46]) :
ln r
t(r + 1, r) ≤ (1 + o(1)) . (14)
2r
For small r, a stronger result was obtained in [7] :
1
t(2s + 1, 2s) ≤ + 2−2s . (15)
4
This gives the best known bound for t(5, 4) and t(7, 6). Starting s = 4,
inequality (15) can be improved.
Bounds (12), (13), and (15) are based on Constructions 1-3.
Construction 1 [41]. Most of the known constructions of Turán systems
have a relatively small number of classes of equivalent elements. In a typical
system, the set of elements is partitioned into a fixed number of groups, and
whether r elements form a block depends only on the groups they belong to.
In contrast, our construction is based on the ordering of the elements. We
denote them 1, 2, . . . , n . Elements i1 < i2 < . . . < i2s form a block of the
system if the Boolean vector

((i1 + 1) mod 2, (i2 + 2) mod 2, . . . , (i2s + 2s) mod 2)

contains exactly s zeros and s ones. It is easy to prove by induction on

s that

the resulting

n
system is a Turán (n, 2s + 1, 2s)-system. Its size is
2s −2s
s
2 + o(1) 2s which gives (12).

5
Construction 2 [7].  We 
identify the n elements with the lines (rows and
n n
columns) of an 2 × 2 -matrix M whose entries are zeros and ones. We
say that a submatrix of M is even if the number of its rows, the number of
its columns and the sum of its entries are even numbers. Now let 2s lines of
M form a block of the system if either all of them are rows, or all of them
are columns, or they induce an even submatrix. Obviously, the resulting
system is a Turán (n, 2s + 1, 2s)-system. Its size depends on the matrix M .
Let the entries of M be independent random variables taking values 0 and 1
with probability 21 each. Then any 2i rows and any 2(s − i) columns (with
1 ≤ i ≤ s − 1) induce an even submatrix with probability 12 . Thus the
expected number of blocks is
 n   n  s−1    n  
2 2
1 X n2 2
+ +
2s 2s 2 i=1 2i 2(s − i)
 n   n  s  n   n  
1 2 2
1 X
2 2
= + +
2 2s 2s 2 i=0 2i 2(s − i)
    
1 1 1 1 n
= + · + O
22s 2 2 n 2s
which yields (15).
Construction 3 [12]. We assume that n ≡ 0 mod l and divide the n
elements into l equal groups A0 , A1 , . . . , Al−1 . For a subset B ⊆ A0 ∪ A1 ∪
. . . ∪ Al−1 , we denote by d(B) the number of indices i ∈ {0, 1, . . . , l − 1}
satisfying Ai ∩ B = ∅ , and set
l−1
X
w(B) = i|Ai ∩ B| .
i=0

Let Ak denote the family of all k-element subsets of A0 ∪ A1 ∪ . . . ∪ Al−1 .

Let a subfamily Bj consist of those B ∈ Ar which satisfy
w(B) + j ≡ 0, 1, . . . or d(B) (mod l) . (16)
We claim that Bj is a Turán (n, r + 1, r)-system. Indeed, for any C ∈ Ar+1 ,
there are n − d(C) indices i such that B ∩ Ai 6= ∅ and thus at least one such
index can be found among
(w(C) + j) mod l , (w(C) + j − 1) mod l , . . . , (w(C) + j − d(C)) mod l .

6
For any x ∈ (B ∩ Ai ), the subset B = C\x satisfies (16), since w(B) =
w(C) − i (mod l) and d(B) ≥ d(C).
0
We denote Ai = {B ∈ Ar : B ∩ Ai = ∅}. Now, to estimate the minimum
of |B0 |, |B1 |, . . . , |Bl−1 |, we use the fact that any B ∈ Ar belongs to exactly
d(B) + 1 families among B0 , B1 , . . . , Bl−1 . Thus
l−1
0 0 0
X X
|Bj | = (d(B) + 1) = |Ar | + |A0 | + |A1 | + . . . + |Al−1 |
j=0 B∈Ar
   l−1    r   
n l
n 1 n
= + l· ≤ 1+l· 1− (17)
r r l r

and
l−1   r   
1X 1 1 n
min{|B0 |, |B1 |, . . . , |Bl−1 |} ≤ |Bj | ≤ + 1− .
l j=0 l l r

It gives  r
1 1
t(r + 1, r) ≤ + 1 − .
l l
r
The substitution l = ln r
(1 + o(1)) produces (13).
To obtain (14), we combine the main ideas of Constructions 1-3. Consider
a system of r-element blocks. Its automorphism is a permutation of the ele-
ments which preserves the set of blocks. The automorphism group generates
an equivalence relation on the elements as well as on the blocks. We denote
the classes of equivalent elements by A1 , A2 , . . . , Al . Then any equivalence
class B of blocks corresponds to an integer partition r = b1 + b2 + . . . + bl
such that

B = B(b1 , b2 , . . . , bl ) = {B : |B ∩A1 | = b1 , |B ∩A2 | = b2 , . . . , |B ∩Al | = bl }.

Suppose that the considered system is a Turán (n, r + 1, r)-system. We

claim that one may omit at least half of the blocks from every present class
B(b1 , b2 , . . . , bl ) whenever b1 ≥ 2, b2 ≥ 2, . . . , bl ≥ 2. As a result, we will
have a Turán (n, r + 1, r)-system of a smaller size. Let us first suppose that
r = 2l and b1 = b2 = . . . = bl = 2. We pick up an l-dimensional matrix M
whose ith dimension corresponds to the equivalence class Ai . A block B ∈
B(2, 2, . . . , 2) corresponds to a (2×2×. . .×2)-submatrix of M . We say that B

7
is even or odd if the sum of the entries of the corresponding submatrix is such.
Similarly to Construction 2, we may omit all odd blocks from B(2, 2, . . . , 2).
Now let r ≥ 2l, and b1 , b2 , . . . , bl satisfy b1 + b2 + . . . + bl = r and bi ≥ 2 with
i = 1, 2, . . . l. Similarly to Construction 1, we linearly order the elements
within each class Ai . For a block B ∈ B(b1 , b2 , . . . , bl ), we name its 2-
projection the (2l)-element subset which includes the two maximal elements
from each intersection B ∩ Ai . We omit all blocks B ∈ B(b1 , b2 , . . . , bl ) which
have odd 2-projections (with respect to the l-dimensional matrix M ). If the
entries of M are independent random variables equal to zero or one with
probability 12 , the expected number of omitted blocks is 21 |B(b1 , b2 , . . . , bl )|.
Applying this to Construction 3 and choosing l = r/(ln r+ln ln r), we get (14)
(see [46] for details).
If we are able to strengthen one of the intermediate assertions, we would
improve inequality (14). Namely, having n elements divided into l equal
groups, we need to find a family of (ml)-element subsets, where (a) every
subset contains m elements from each group, and (b) any (ml + 1)-element
subset, that contains at least m elements from each group, has to contain at
least one of the chosen (ml)-element subsets. If we find such a family of size
 n l
(Cm + o(1)) l as n, l → ∞ ,
m
we will get by similar arguments the upper bound
ln r
t(r + 1, r) ≤ (Cm + o(1)) .
r
1
To obtain (14), we used m = 2 and C2 = 2
.

4.2 The general case

The simplest general upper bound
 r−1
r−1
t(k, r) ≤ (18)
k−1
was obtained in [41]. The construction is based on the elementary fact:
Lemma. Let b0 , b1 , . . . , bl−1 be (cyclically ordered) reals, and b = (b0 + b1 +
. . . + bl−1 )/l. Then there exists an index m such that
∀s = 1, 2, . . . , l : bm + bm+1 + . . . + bm+s−1 ≥ sb

8
(we denote bl = b0 , bl+1 = b1 and so on).
Proof. Choose m, q ∈ {0, 1, . . . , l − 1} to minimize the value bm−q + bm−q+1 +
. . . + bm−1 − qb. Then

bm−q + bm−q+1 + . . . + bm−1 + bm + . . . + bm+s−1 − (q + s)b ≥

≥ bm−q + bm−q+1 + . . . + bm−1 − qb

which implies the statement of the Lemma.

Construction 4 [41]. We select integers l and d such that r = ddk/le and
divide the n elements into l approximately equal groups A0 , A1 , . . . , Al−1 .
An r-element block B is included into the system if the sizes of intersections
bi = |B ∩ Ai | (i = 0, 1, . . . , l − 1) satisfy the condition: there exists an index
m such that s
X k
∀s = 1, 2, . . . , d : bm+i−1 ≥ s
i=1
l
(we again denote bl = b0 , bl+1 = b1 and so on). It follows from Lemma that
this is a Turán (n, k, r)-system. Counting the blocks, we get
     X
1 n r 1
T (n, k, r) ≤ 1 + O l (19)
n l a1 !a2 ! . . . ad !
where the sum is taken over all sequences of non-negative integers (a1 , a2 , . . . , ad )
satisfying a1 + a2 + . . . + ad = r and a1 + a2 + . . . as ≥ s kl with s = 1, 2, . . . , d.
If we choose l = k − 1 and d = r − 1, the value of the sum is equal to
1
r!
(r − 1)r−1 which gives (18). For some k and r, there is a better choice. In
particular, (19) with k = 7, r = 5, l = 5, d = 3 yields t(7, 5) ≤ 121 625
.
Frankl an Rödl [12] used probabilistic arguments to prove the bound
a(a + 4 + o(1)) ln r
t(r + a, r) ≤ r

as r → ∞ , a = const . (20)
a

Their arguments were extended by Sidorenko [46] to prove the inequalities

(1 + o(1))(k − r + 1) ln kr


t(k, r) ≤ k

as r → ∞, k ≥ r +r/ log2 r , (21)
r

and
(cγ + o(1))r2
t(k, r) ≤ k

as r → ∞, k = (γ + o(1))r (22)
r

9
with cγ = (γ − 1)[γ ln γ − (γ − 1) ln(γ − 1) + o(1)], γ > 1. It is interesting
to compare (21) or (22) with the lower bound (9).
If r = 2 (see Section 6), the equality in (18) is attained. The conjec-
tured values of the Turán numbers with r = 3 (see Section 7) also satisfy the
equality in (18). Turán [58] conjectured that the equality is attained when-
ever k − 1 is a multiple of r − 1. A counterexample k = 13, r = 4 was found
in [41]. In general, by using (7) with l = 2s − 1, α1 = α2 = . . . αl = 2s and
applying (12), we get the counterexample k = 4s2 −2s+1, r = 2s, k−1 r−1
= 2s
for any s ≥ 2. Inequality (22) shows that the conjecture fails for any ratio
k−1
r−1
when r is sufficiently large.

n
5 The case of small k−1
In the case qr ≤ n, one may easily prove by induction that the system of q
disjoint r-element subsets is the optimal Turán (n, n − q + 1, r)-system. Thus
n r
T (n, α + 1, r) = n − α if 1 ≤ ≤ . (23)
α r−1
The next zone was found in [43] (the proof was also published in [45]) :
 3r−2 r n 3r
 d r ne − 3α if r ≡ 0 mod 2, r−1 ≤ α ≤ 3r−4 ;
T (n, α + 1, r) = (24)
3r−1 r n 3r+1
3n − b r−1 αc if r ≡ 1 mod 2, r−1 ≤ α ≤ 3r−3 .


The right hand side of (24) gives a lower bound for any larger n.

6 The case r = 2
Mantel [26] in 1907 determined T (n, 3, 2), and Turán [56] in 1941 determined
T (n, k, 2) for any k. We write the result of Turán in the following form:

m(m + 1) n
T (n, α + 1, 2) = mn − α if m ≤ ≤m+1. (25)
2 α
Turán also proved that the extremal system is unique. Namely, the n ele-
ments are divided into α = k − 1 nearly equal groups; a pair of elements
{x, y} is a block of the system if x and y belong to the same group.

10
7 The case r = 3
Two old conjectures of Turán concern the numbers T (n, 4, 3) and T (n, 5, 3).
In order to get a Turán (n, 4, 3)-system, we divide the n elements into 3
almost equal groups A0 , A1 , A2 , and take all triples {x, y, z} with x, y ∈
Ai , z ∈ (Ai ∪ Ai+1 ), i = 0, 1, 2, where A3 = A0 . This gives

 m(m − 1)(2m − 1) if n = 3m ;
T (n, 4, 3) ≤ m2 (2m − 1) if n = 3m + 1 ; (26)
 2
m (2m + 1) if n = 3m + 1 .

One of the famous conjectures of Turán is

Conjecture 2. The equality in (26) holds.
Katona, Nemetz and Simonovits [19] established the equality for n ≤
9, Stanton and Bate [47] for n = 10, Radziszowski and Zou [37] for n =
11, 12, 13.
In contrast to the case r = 2, this construction, if optimal, is not unique.
Other variants of (n, 4, 3)-systems with the same number of blocks were found
by Brown [2]. Later Kostochka [21] found 2m−2 non-isomorphic constructions
of Turán (3m, 4, 3)-systems of size m(m − 1)(2m − 1). A nice representation
of those constructions in terms of directed graphs was given by Fon-Der-
Flaass [10]. The Kostochka’s systems are indexed by ordered partitions m =
l1 + l2 + . . . + ls where l1 ≥ 1, . . . , ls−1 ≥ 1, ls ≥ 2. There are 2m−2 such
partitions. Given a partition, we divide 3m elements into 3s subsets Aij such
that i ∈ Z3 , j ∈ {1, 2, . . . s}, and |Aij | = lj . Now we take triples T which
satisfy
1. T ⊆ sj=1 Aij with some i ∈ Z3 , or
S
S  S 
s+1−j s
2. T = {x, y, z} where x ∈ Aij and y, z ∈ t=1 Ai+1,t ∪ t=s+2−j Ai−1,t .
They form a Turán (3m, 4, 3)-system of the required size.
For m ≤ 4, all (3m, 4, 3)-systems of size m(m−1)(2m−1) are Kostochka’s
systems (it was shown in [44] for m = 3 and in [37] for m = 4).
We say that a subset of elements of a system is a clique if any r-tuple
within this subset is a block of the system. In each of the Kostochka’s
systems, the elements are divided into 3 cliques of size m.
Conjecture 3. The elements of any Turán (3m, 4, 3)-system of size m(m −
1)(2m − 1) are divided into 3 cliques of size m.

11
 By dividing n elements into two almost equal groups and taking all triples
within each group, we get a Turán (n, 5, 3)-system. Hence,
   
dn/2e bn/2c
T (n, 5, 3) ≤ + . (27)
3 3

Turán conjectured (see, for instance, [58]) that this construction is the best
possible and the equality in (27) holds. However, it was disproved for all odd
n ≥ 9 by Sidorenko and Kostochka [44] (for n = 9 it was done earlier by
Surányi [49]).
Construction 5. We divide the n elements into 9 groups A1 , A2 , . . . , A9
which correspond to the points a1 , a2 , . . . , a9 of the finite affine plane of order
3. The 12 lines of the plane are denoted as follows:

L1 = {a1 , a6 , a8 }, L9 = {a2 , a4 , a9 }, M1 = M9 = {a1 , a5 , a9 },

L2 = {a1 , a2 , a3 }, L8 = {a7 , a8 , a9 }, M2 = M8 = {a2 , a5 , a8 },

L3 = {a3 , a4 , a8 }, L7 = {a2 , a6 , a7 }, M3 = M7 = {a3 , a5 , a7 },
L4 = {a1 , a4 , a7 }, L6 = {a3 , a6 , a9 }, M4 = M6 = {a4 , a5 , a6 }.
We note that ai ∈ Li and ai , a5 ∈ Mi for i 6= 5. Now we take triples x ∈ Ai ,
y ∈ Aj , z ∈ Ak that satisfy

1. i = j = k, or

2. i, j, k are pairwise distinct, and ai , aj , ak are collinear, or

3. i = j, k 6= i and ak ∈ Li ∪ Mi .

If we also require |A5 | = 1, then the resulting triples form a Turán (n, 5, 3)-
system. Making the sizes of the remaining 8 groups nearly equal,

|A1 | ≥ |A9 | ≥ |A3 | ≥ |A7 | ≥ |A2 | ≥ |A8 | ≥ |A6 | ≥ |A4 | ≥ |A1 | − 1,

we get    
m+1 m
T (2m + 1, 5, 3) ≤ + − f (m) (28)
3 3

12
where  m

 2
if m ≡ 0 mod 4 ;



 m−3
if m ≡ 1 mod 4 ;


 2
f (m) =
m−4
if m ≡ 2 mod 4 ;




 2


 m−3
if m ≡ 3 mod 4 .

2

For m = 7, a better bound T (15, 5, 3) ≤ 89 is known (cf [44, 1]).

The values T (n, 5, 3) and all the optimal systems for small values of n
were determined in [49, 47, 1] :

T (6, 5, 3) = 2, T (7, 5, 3) = 5, T (8, 5, 3) = 8, T (9, 5, 3) = 12,

T (10, 5, 3) = 20, T (11, 5, 3) = 29, T (12, 5, 3) = 40, T (13, 5, 3) = 52.

The unique (13,5,3)-system of size 52 is formed by the collinear triples of
points of the projective plane of order 3. For even values of n, no counterex-
ample is known to the Turán’s conjecture on the equality in (27).
n/2


In contrast to (n, 4, 3)-systems, one may find (n,
√ 5, 3)-systems with 2 3
blocks where the size of the largest clique is only n.
Construction 6. Let p be a permutation of order m. We construct a system
whose elements are divided into 2m groups A1 , A2 , . . . , Am , B1 , B2 , . . . , Bm ,.
A triple {x, y, z} is a block of the system if
1. x ∈ Ai , y ∈ Aj , z ∈ Ak where i, j, k are distinct, or
2. x ∈ Bi , y ∈ Bj , z ∈ Bk where i, j, k are distinct, or
3. x, y ∈ Ai , z ∈ Bj where j 6= i , or
4. x, y ∈ Bj , z ∈ Ai where i 6= p(j) , or
5. x, y, z ∈ Ai , or
6. x, y, z ∈ Bj .
It is easy to check that any 5-element subset contains at least one block of
the system. We denote ai = |Ai |, bi = |Bi |, n = a1 + a2 + . . . + am , and set
the additional condition:
7. ai = bi = ap(i) for i = 1, 2, . . . , m .

n−1
Now any element belongs to exactly 2
blocks. Therefore, we have a
n


Turán (2n, 5, 3)-system of size 2 3 . If indices i1 , i2 , . . . , ik form a cycle C of
the permutation p, all numbers ai1 , ai2 , . . . , aik , bi1 , bi2 , . . . , bik must be equal

13
to the same value d = d(C). We denote the cycles of the permutation by
C1 , C2 , . . . , Ct . Let ci be the length of Ci , and di = d(Ci ). Then

c1 + c2 + . . . + ct = m ,
c1 · d1 + c2 · d2 + . . . + ct · dt = n . (29)

We call a set of parameters (c1 , c2 , . . . , ct , d1 , d2 , . . . , dt ) that satisfies (29)

a representation. Every representation produces a (2n, 5, 3)-system of size
2 n3 . Sometimes, different representations produce isomorphic systems. We

name a representation (c1 , c2 , . . . , ct , d1 , d2 , . . . , dt ) canonical if one satisfies

the following conditions:
(i) d1 ≥ d2 ≥ . . . ≥ dt ;
(ii) di = di+1 implies ci ≥ ci+1 ;
(iii) if t = 1 then d1 > 1 ;
(iv) if t > 1 then dt−1 > 1 ;
(v) if t = 2 then c1 + c2 > 2 ;
(vi) if t = 2, c1 = 1, d2 = 1 then d1 6= c2 .
Different canonical representations produce non-isomorphic Turán systems.
Construction 7. Consider a pair of representations

c01 · d01 + c02 · d02 + . . . + c0t0 · d0t0 = n0 , (30)

c001 · d001 + c002 · d002 + . . . + c00t00 · d00t00 = n00 . (31)

Let A01 , A02 , . . . , B10 , B20 , . . . and A001 , A002 , . . . , B100 , B200 , . . . denote the equivalence
classes of elements in the systems produced by these representations. Take
all blocks of both systems and add blocks {x, y, z} which satisfy
8. x, y ∈ A0i , z ∈ Bj00 , or
9. x, y ∈ A00i , z ∈ A0j , or
10. x, y ∈ Bi0 , z ∈ A00j , or
11. x, y ∈ Bi00 , z ∈ Bj0 , or
12. x ∈ A0i , y ∈ A0j , i 6= j, z ∈ A00k , or
13. x ∈ A00i , y ∈ A00j , i 6= j, z ∈ Bk0 , or
14. x ∈ Bi0 , y ∈ Bj0 , i 6= j, z ∈ Bk00 , or
15. x ∈ Bi00 , y ∈ Bj00 , i 6= j, z ∈ A0k .
0 00

It gives a Turán (2(n0 + n00 ), 5, 3)-system of size 2 n +n 3
. We call it the
system produced by the pair of representations (30), (31). Different pairs of
canonical representations with t0 > 1 and t00 > 1 produce non-isomorphic

14
systems, and these systems are distinct from the systems produced by single
representations.
The systems in Constructions 6 and 7 do not have to contain large cliques.
In particular, the representation 2k · k = 2k 2 , as well as the pair of repre-
sentations k · k = k 2 , k · k = k 2 , produces a Turán (4k 2 , 5, 3)-system with
2

2 2k3 blocks, where the size of the largest clique is 2k.
The best known (n, k, 3)-systems with k ≥ 6 can be constructed as unions
of (n0 , 5, 3)-, (n00 , 4, 3)- and (n000 , 3, 3)-systems. In particular, using (7), (26)
and (27), we get
α−1  2n+i 
1X α
T (n, α + 1, 3) ≤ , (32)
2 i=0 3

or equivalently,
   
m m+1 2n
T (n, α + 1, 3) ≤ n− α for m ≤ ≤m+1.
2 3 α

Applying (28), one may improve (32) for those values of n where 2n α
is not
2n
integral. In the case when α is an integer, we expect that the old conjecture
of Turán [58] is correct:
Conjecture 4.  
 αm  α m
T , α + 1, 3 = .
2 2 3

Thus we believe that the asymptotic behavior of the numbers T (n, k, 3)

is what Turán and other researchers (e.g. see [4]) conjectured:
Conjecture 5.  2
2
t(α + 1, 3) = .
α

The best known lower bound for t(4, 3) was found by Giraud [17] (see
formula (11) in Section 3):
√
7 − 21
t(4, 3) ≥ = 0.4029 . . . .
6

15
 n
For small ratios α
, the values T (n, α + 1, 3) were found in [43, 44]:


 n − α if 1 ≤ αn ≤ 23 ;




 3n − 4α if 23 ≤ αn ≤ 2 ;


T (n, α + 1, 3) = (33)
4n − 6α if 2 ≤ αn ≤ 49 , 4n − 9α 6= −1 ;








4n − 6α + 2 if 4n − 9α = −1, 1, 2 .


We expect the next zone to be as follows.

Conjecture 6.
 
15 9 n 5
T (n, α + 1, 3) = 8n − 2 α if ≤ ≤ .
2 4 α 2

The corresponding upper bound follows from formulae (7), (28) and (32).

8 The case r = 4
Formulae (23) and (24) imply

 n − α if α ≤ n ≤ 43 α ;
T (n, α + 1, 4) = (34)
 5  4 3
2
n − 3α if 3
α ≤n≤ 2
α .

It was proven in [43] that

T (6s + 1, 4s + 1, 4) = 3s + 4 ,

T (6s + 2, 4s + 2, 4) = 3s + 3 .
Feng-Chu Lai and Gerard J. Chang [23] proved U (n, q, 4) + n ≥ 29 q, and
this can be casted as an upper bound for Turán numbers (see Section 1,
formula (1)) with n ≥ 32 α :

7 9
T (n, α + 1, 4) ≥ n− α.
2 2

16
 Note that Turán numbers with r = 2 form convex sequence:

T (n, k, 2) − T (n − 1, k, 2) ≤ T (n + 1, k, 2) − T (n, k, 2) .

The same is true for all determined values of Turán numbers with r = 3.
As (24) and (34) show, there is no convexity starting r = 4. For instance, if
2
3
α ≤ m ≤ 34 α, then

T (2m, α+1, 4) = 5m−3α, T (2m+1, α+1, 4) = 5m−3α+3, T (2m+2, α+1, 4) = 5m−3α+5,

T (2m+1, α+1, 4)−T (2m, α+1, 4) > T (2m+2, α+1, 4)−T (2m+1, α+1, 4) .
The only case with r > 3, when there exists a plausible conjecture on
the asymptotic behavior of Turán numbers, is k = 5, r = 4. In this case,
a beautiful construction was found by Giraud [16] and then generalized by
de Caen, Kreher and Wiseman [7]. We have already described it in Section 4
(see Construction 2). Let h(l, m) denote the minimal number of even (2 × 2)-
submatrices in an (l × m)-matrix. We also set h(2l) = h(l, l) and h(2l + 1) =
h(l, l + 1). Construction 2 yields the upper bound
 n   n+1 
T (n, 5, 4) ≤ 2 + 2 + h(n) . (35)
4 4

Minimization of the number of even (2 × 2)-submatrices is equivalent to the

minimization of the sum of square entries of AA∗ where A is an (l × m)-
matrix with entries equal to 1 or -1. Naturally, this problem is related to the
existence of Hadamard matrices. We are going to show that

b2c b 4 c + b n+34 c
 n
 n+2



 2 2 2
if 8k − 3 ≤ n ≤ 8k + 1,





b n2 c
b n+2
4 c + b 4 c

n+3


h(n) = (36)
 2 2 2




+ b 4 c + b 4 c if 8k − 5 ≤ n ≤ 8k − 4,

 n
n+2

2 2

provided that there exists an Hadamard matrix of order 4k. Indeed, the
lower bound in (36) for 8k − 3 ≤ n ≤ 8k + 1 follows from the inequalities
 m   m+1 
h(2, m) ≥ 2 + 2
2 2

17
and  
l
h(l, m) ≥ h(2, m) .
2
In any
(3 × (4k − 2))-matrix, there are two rows which induce more than
2 2k−1
2
even (2 × 2)-submatrices. Thus
   
l 2k − 1
h(l, 4k − 2) ≥ ·2 + T (l, 3, 2)
2 2
which is the lower bound in (36) for n = 8k − 4, 8k − 5. Now assume that
there exists an Hadamard matrix of order 4k. We replace its negative
 n  entries
with zeros. The resulting matrix and its submatrices of size 2 × n+1


2
yield the upper bound in (36) for 8k − 5 ≤ n ≤ 8k. To get the upper bound
for n = 8k + 1, we extend the matrix by an arbitrary column.
The identity matrix of size 5 yields h(10) = 40. The case n = 8k+2, k > 1
remains unresolved.
The exact values of T (n, 5, 4) were determined for n ≤ 10 in [40] and [6].
In the former work, the uniqueness of the Turán (10, 5, 4)-system of size 50
was also proven. It is interesting that the equality in (35) is attained (and
thus the described construction is optimal) for every n ≤ 10. We expect that
this construction is asymptotically optimal.
Conjecture 7.
5
t(5, 4) = .
16
For k = 6, 7, we describe the best asymptotical constructions that we
know.
Construction 8. We take two matrices with the same number of rows: an
(m × m0 )-matrix M 0 and an (m × m00 )-matrix M 00 ; their elements are zeros
and ones. The elements of the system are divided into 3 groups of sizes
m, m0 and m00 ; they correspond to the set of rows (which is common for both
matrices), the set of columns of M 0 , and the set of columns of M 00 . We take all
quadruples within each group as well as those quadruples which correspond
to even (2 × 2)-submatrices in either matrix. We also take every quadruple
which corresponds to a pair of columns of M 0 and a pair of columns of M 00 .
It gives us a Turán (m + m0 + m00 , 6, 4)-system. Therefore,
   0   00   0  00 
0 00 m m m 0 00 m m
T (m+m +m , 6, 4) ≤ + + +h(m, m )+h(m, m )+ .
4 4 4 2 2
(37)

18
If m0 = m00 = bnpc, m = n − m0 − m00 , inequality (37) and the simple
probabilistic bound h(l, m) ≤ 12 (l2 )(m
2 ) yield
 
4 2 2 4 n
T (n, 6, 4) ≤ (8p + 6p (1 − 2p) + (1 − 2p) ) + O(n3 ) .
4

The polynomial in the right hand side attains the minimum value 0.176614234 . . .
at p = 0.29419 . . . . Thus we get t(6, 4) < 0.17662.
The exact values of T (n, 6, 4) were determined in [8] for n ≤ 10.
Construction 9. In order to construct a Turán (n, 7, 4)-system, we divide
the n elements into 8 groups Aijk where i, j, k ∈ {0, 1} and

|A000 | ≤ |A001 | ≤ |A010 | ≤ |A011 | ≤ |A100 | ≤ |A101 | ≤ |A110 | ≤ |A111 | ≤ |A000 |+1 .

The system includes quadruples Q which satisfy

1. Q ⊆ Aijk , i, j, k ∈ {0, 1}, or
2. |Q ∩ Aijk | = 3, |Q ∩ Al,1−j,k | = 1, i, j, k, l ∈ {0, 1}, or
3. |Q ∩ Aijk | = |Q ∩ Almk | = 2, i, j, k, l, m ∈ {0, 1}, (i, j) 6= (l, m),
or
4. |Q ∩ Aijk | = 2, |Q ∩ A0,m,1−k | = |Q ∩ A1,m,1−k | = 1, i, j, k, m ∈
{0, 1}, or
5. |Q ∩ A00k | = |Q ∩ A01k | = |Q ∩ A10k | = |Q ∩ A11k | = 1, k ∈ {0, 1}.
3
This construction yields the asymptotical upper bound t(7, 4) ≤ 32 and the
best known bounds for small n: T (11, 7, 4) ≤ 17, T (12, 7, 4) ≤ 26, T (13, 7, 4) ≤
40, T (14, 7, 4) ≤ 58, T (15, 7, 4) ≤ 81, T (16, 7, 4) ≤ 108. The construction is
optimal for n ≤ 10: T (8, 7, 4) = 2, T (9, 7, 4) = 5, T (10, 7, 4) = 10.

9 The case of small n − k (covering numbers)

While the behavior of the Turán numbers T (n, k, r) is studied with a fixed
r (or fixed r and k), the covering numbers C(n, m, p) = T (n, n − p, n − m)
are usually investigated with fixed m and p. The titles “Turán problem” and
“covering problem” represent two different viewpoints in the 3-dimensional
parameter space: they state along which subspaces the problem will be pri-
marily studied. A detailed survey of results on the covering numbers can be
found in [34]. Tables of the best known upper bounds for C(n, m, p) with
n ≤ 32, m ≤ 16, p ≤ 8 are published in [18].

19
 Inequality (2) yields the Schönheim bound:
C(n, m, p) ≥ L(n, m, p) ,
where      
n n−1 n−m+1
L(n, m, p) = ... ... .
m m−1 1
Rödl [38] proved that
C(n, m, p) = (1 + o(1))L(n, m, p) (38)
where m, p are fixed and n → ∞ . Kuzyurin [22] showed that (38) can be
n
extended to the case when p is fixed and m →∞.
The case m = 3, p = 2 was solved by Fort and Hedlund [11] :
C(n, 3, 2) = L(n, 3, 2) ,
and the case m = 4, p = 2 by Mills [27, 28] :

 L(n, 4, 2) + 2 if n = 19 ;
C(n, 4, 2) = L(n, 4, 2) + 1 if n = 7, 9, 10 ;
L(n, 4, 2) if n 6= 7, 9, 10, 19 .


The case m = 5, p = 2 was not resolved completely (cf. [14, 31, 24, 35, 36,
33, 48]). For instance, the subcases n ≡ 0 mod 4 and n ≡ 13 mod 20 are still
open.
For m = 4, p = 3, it was proven in [29, 32] that C(n, 4, 3) = L(n, 4, 3) for
any n except a finite number of values n ≡ 7 mod 12 in the range n ≤ 54211.
If a system of m-element blocks covers all p-element subsets of an n-
element set, one may replace each element by l distinct elements and get
a system of (lm)-element blocks covering all p-element subsets of an (ln)-
element set. Thus
C(ln, lm, p) ≤ C(n, m, p) . (39)
Because of (39), one may hope to classify all values n, m such that C(n, m, p) =
n
t (with fixed p, t) in terms of m . In the cases p = 2 and p = 3 such a clas-
sification was obtained for t ≤ 13 (see [30, 53]) and t ≤ 8 (see [50, 54]),
respectively.
Todorov [51] proved
l l−1
!
X X pl+1 − 1
C ai pl−i , ai pl−1−i , l =
i=0 i=0
p−1

20
where p is a prime power and a0 ≥ a1 ≥ . . . ≥ al > 0. The construction is
based on finite projective geometries, and the lower bound follows from the
Schönheim bound. Some related results are given in [13, 25, 52, 55].
In most of the cited results, C(n, m, p) is equal or very close to L(n, m, p).
On the other hand, formulae (23) and (24) imply

(p + 1)m
C(n, m, p) = p + 1 if n ≤ , (40)
p

if n − m ≡ 0 mod 2, (p+1)m ≤ n ≤ 3(p+1)m

  2n 
 3(p + 1) − n−m
 p 3p−1
,
C(n, m, p) = j k
 3(p + 1) − 2(n−p−1)

if n − m ≡ 1 mod 2, (p+1)m ≤ n ≤ (p+1)(3m−1) .
n−m−1 p 3p−1
(41)
The right hand side of (40) is equal to the Schönheim bound, L(n, m, p), but
the right hand side of (41) is far from L(n, m, p).
Inequalities (4) and (8) sometimes are better than the Schönheim bound.

21
References
[1] Boyer, E. D., Kreher, D. L., Radziszowski, S. P., Zou, J., and
Sidorenko, A.: On (n, 5, 3)-Turán systems. Ars Combinatoria 37,
1-19 (1994).
[2] Brown, W. G.: On an open problem of Paul Turán concerning 3-
graphs. In: Studies in pure mathematics, pp. 91-93. Basel-Boston,
Mass.: Birkhäuser 1983. MR87e:05083
[3] Caen, D. de : Extension of a theorem of Moon and Moser
on complete subgraphs. Ars Combinatoria 16, 5-10 (1983).
MR85d:05132
[4] Caen, D. de : A note on the probabilistic approach to Turán’s
problem. J. Comb. Theory B 34, 340-349 (1983)
[5] Caen, D. de : The current status of Turán’s problem on hyper-
graphs. In: Extremal Problems for Finite Sets, Visegrád (Hun-
gary), 1991, Bolyai Society Mathematical Studies, Vol. 3, pp.
187-197
[6] Caen, D. de, Kreher, D. L., Radziszowski, S. P., and Mills, W.
H.: On the coverings of t-sets with (t + 1)-sets: C(9, 5, 4) and
C(10, 6, 5). Discrete Mathematics 92, 65-77 (1991)
[7] Caen, D. de, Kreher, D. L., and Wiseman, J.: On constructive
upper bounds for the Turán numbers T (n, 2r + 1, r). Congressus
Numerantium 65, 277-280 (1988). MR90c:05003
[8] Droesbeke, F., Lorea, M.: Détermination de valeurs du nombre
de Turán T (n, 4, 6). Cahiers du Cent. Étud. Rech. Oper. 24, 185-
191 (1982). MR84g:05081
[9] Erdős, P.: On the combinatorial problems I would most like to
see solved. Combinatorica 1, 25-42 (1981). MR82k:05001
[10] Fon-Der-Flaass, D. G.: A method for constructing (3,4)-graphs.
Math. Notes 44 (1988). MR90a:05105
[11] Fort, M. K., Hedlund, G. A.: Minimal coverings of pairs by
triples. Pacific J. of Mathematics 8, 709-719 (1958). MR21#2595

22
[12] Frankl, P., Rödl, V.: Lower bounds for Turán’s problem. Graphs
and Combinatorics 1, 213-216 (1985)

[13] Füredi, Z.: Covering pairs by q 2 + q + 1 sets. J. Comb. Theory A

54, 248-271 (1990)

[14] Gardner, B.: Results on coverings of pairs with special reference

to coverings by quintuples. Congressus Numerantium 23, 169-178
(1971). MR49#4816

[15] Giraud, G.: Majoration du nombre de Ramsey ternaires-bicolore

en (4,4). C. R. Acad. Sci. Paris AB269, A620-A622 (1969).
MR41#1591

[16] Giraud, G.: Remarques sur deus problèmes estrémaux. Discrete

Mathematics 84, 319-321 (1990)

[17] Giraud, G.: Une minoration de la premiere fonction ternaire de

Turán, manuscript

[18] Gordon, D. M., Kuperberg, G., and Patashnik, O.: New con-
structions for covering designs. J. Comb. Design, to appear

[19] Katona, G., Nemetz, T., and Simonovits, M.: On a graph prob-
lem of Turán, (in Hungarian). Mat. Lapok 15, 228-238 (1964).
MR30#2483

[20] Kim, K. H., Roush, F. W.: On a problem of Turán. In: Studies in

pure mathematics, pp. 423-425. Basel-Boston, Mass.: Birkhäuser
1983. MR86m:05007

[21] Kostochka, A. V.: A class of constructions for Turán’s (3,4)-

problem. Combinatorica 2, 187-192 (1982). MR84k:05053

[22] Kuzyurin, N. N.: Asymptotic investigation of the covering prob-

lem. Cybernetic Problems 37, 19-56 (1980). MR82i:05029

[23] Feng-Chu Lai, Chang, G. J.: An upper bound for the transversal
numbers of 4-uniform hypergraphs. J. Comb. Theory B 50, 129-
133 (1990)

23
[24] Lamken, E., Mills, W. H., Mullin, R. C., and Vanstone, S. A.:
Covering of pairs by quintuples. J. Comb. Theory A 44, 49-68
(1987). MR87m:05064
[25] Lorea, M., On Turán hypergraphs. Discrete Mathematics 22, 281-
285 (1978). MR80f:05053
[26] Mantel, W.: Vraagstuk XXVIII. Wiskundige Opgaven met de
Oplossingen 10, 60-61 (1907)
[27] Mills, W. H.: On the covering of pairs by quadruples I. J. Comb.
Theory A 13, 55-78 (1972). MR45#8534
[28] Mills, W. H.: On the covering of pairs by quadruples II. J. Comb.
Theory A 13, 138-166 (1973). MR45#8546
[29] Mills, W. H.: On the covering of triples by quadruples. In: Pro-
ceedings of the Fifth Southeastern Conference on Combinatorics,
Graph Theory and Computing, Boca Raton, Fla., 1974, pp. 573-
581. Winnipeg: Utilitas 1974. MR50#9638.
[30] Mills, W. H.: Covering designs I : covering by a small number of
subsets. Ars Combinatoria 8, 199-315 (1979). MR81e:05058
[31] Mills, W. H.: A covering of pairs by quintuples. Ars Combinatoria
18, 21-31 (1984). MR87c:05042
[32] Mills, W. H., Mullin, R. C.: Covering triples by quadruples:
an asymptotic solution. J. Comb. Theory A 41, 117-138 (1986).
MR87e:05047
[33] Mills, W. H., Mullin, R. C.: Covering pairs by quintuples: the
case v ≡ 3 modulo 4. J. Comb. Theory A 49, 308-322 (1988).
MR89i:05092
[34] Mills, W. H., Mullin, R. C.: Coverings and packings. In: Con-
temporary Design Theory: A Collection of Surveys, pp. 371-399.
Wiley 1992
[35] Mullin, R. C.: On covering pairs by quintuples: the cases
v ≡ 3 or 11 modulo 20, J. of Combinatorial Mathematics and
Combinatorial Computing 2, 133-146 (1987). MR89c:05028

24
[36] Mullin, R. C.: On the determination of the covering numbers
C(2, 5, v). J. of Combinatorial Mathematics and Combinatorial
Computing 4, 123-132 (1988). MR89m:05039

[37] Radziszowski, S. P., Zou, J.: Personal communication

[38] Rödl, V.: On a packing and covering problem Europ. J. Comb.

5, 69-78 (1985). MR86k:05033

[39] Schönheim, J.: On coverings. Pacific J. of Mathematics 14, 1405-

1411 (1964). MR30#1954

[40] Sidorenko, A. F.: On Turán numbers T (n, 5, 4) and numbers of

monochromatic 4-cliques in 2-colored 3-graphs (in Russian). Vo-
prosi Kibernetiki no.64, 117-124 (1980)

[41] Sidorenko, A. F.: Systems of sets that have the T -property.

Moscow University Mathematics Bulletin 36, no.5, 22-26 (1981).
MR83f:05054

[42] Sidorenko, A. F.: The method of quadratic forms and Turán’s

combinatorial problem. Moscow University Mathematics Bulletin
37, no.1, 1-5 (1982). MR83g:05040

[43] Sidorenko, A. F. Extremal constants and inequalities for distri-

butions of sums of random vectors (in Russian). Ph.D. Thesis,
Moscow State University, 1982

[44] Sidorenko, A. F.: The Turán problem for 3-graphs (in Russian).
Combinatorial Analysis (Russian), no.6, pp. 51-57. Moscow State
University, 1983. MR87h:05065

[45] Sidorenko, A. F..: Exact values of Turán numbers. Math. Notes

42, 913-918 (1987). MR90a:05008

[46] Sidorenko, A.: Upper bounds on Turán numbers. Submitted to J.

Comb. Theory, ser. A

[47] Stanton, R. G., Bate, J. A.: A computer search for B-coverings.

Lect. Notes in Math., no. 829, 37-50 (1980). MR82h:05015

25
 [48] Stanton, R. G., Buskens, R. W., and Allston, J. L.: Computation
of the covering number N (2, 5, 24). Utilitas Mathematica 37, 127-
144 (1990). MR91g:05030

[49] Surányi, J.: Some combinatorial problems of geometry (in Hun-

garian). Mat. Lapok 22, 215-230 (1971). MR49#143

[50] Todorov D. T., Tonchev, V. D.: On some coverings of triples. C.

R. Bulg. Acad. Sci. 35, 1209-1211 (1982). MR84c:05033

[51] Todorov, D. T.: A method of constructing coverings. Math. Notes

35, 869-876 (1984). MR85h:05035

[52] Todorov D. T.: Some coverings derived from finite planes. In:
Colloq. Math. Soc. János Bolyai, Vol. 37, Finite and Infinite
Sets, Eger, 1981, pp. 697-710. Budapest: Akad. Kiadó 1985.
MR87b:05044

[53] Todorov, D. T.: On the covering of pairs by 13 blocks. C. R.

Bulg. Acad. Sci. 38, 691-694 (1985). MR86m:05022

[54] Todorov, D. T.: On the covering of triples by eight blocks. Serdica

12, 20-29 (1986). MR87h:05066

[55] Todorov, D. T.: Lower bounds for coverings of pairs by large

blocks. Combinatorica 9, 217-225 (1989). MR90k:05055

[56] Turán, P.: Egy gráfelméleti szélsöértékfeladatról, Mat. és Fiz.

Lapok 48, 436-453 (1941). MR8-284

[57] Turán, P.: Research problems. Maguar Tud. Akad. Mat. Kutato
Int. Közl. 6, 417-423 (1961). MR31#2107

[58] Turán, P.: Applications of graph theory to geometry and poten-

tial theory. In: Combinatorial Structures and Their Applications,
pp. 423-434. New York: Gordon and Breach 1970. MR42#7551

26

View publication stats