Minimal keys and-antikeys

By V. D. THI

§ 1. Introduction

The relational model, defined by E. F. Codd [3] is one of the most investigated
data base models of the last years. Many papers have appeared concerning combina-
torial characterization of functional dependencies, systems of minimal keys and anti-
keys. A set of minimal keys and a set of antikeys form Sperner-systems. Sperner-sys-
tems and sets of minimal keys are equivalent in the sense that for an arbitrary Sperner-
system S a family of functional dependencies F can be constructed so that the mini-
mal keys of F a r e exactly the elements of S (cf. [4]).
In the present paper we propose some combinational algorithms to determine
antikeys and minimal keys. In the second part of the paper, we are going to study
connections between minimal keys and antikeys for special Sperner-systems.
We start with some necessary definitions.
Definition 1.1. Let £2 be a finite set, and denote P(Q) its power set. The mapping
F: P(i3)—P(i2) is called a closure operation over Q if, for every A, BQ Q,
(1) AQF(A) (extensivity),
(2) AQB implies F(A)QF(B) (monotonity),
(3) F(A)=F(F(A)) (idempotency).
In few cases Q is represented by the set {1, ..., n} or by the set of columns of an
mXn matrix M. If we use the second representation, a special closure operation F M
can be defined over the set of the columns of M :
The i-th column of M belongs to FM(A) if and only if for any two rows of M
which are identical on A they are equal on the i-th column, too. •
It is easy to see, that FM(A) is a closure operation. It is known (see [1]) that any
closure operation F over a finite set £2 can be represented by an appropriate matrix
M, that is we can choose M and represent Q by the set of the columns of M so that F
coincides with Fm-
Definition 1.2. Let F be a closure operation over Q, and AQQ. We say that
— A is a k e y of F, if F(A)=Q.
— A is a minimal key of F, if A is a key of F and for any BQ A, F(B)= Q
implies B=A, i.e. no proper subset of A is a key of F..
Let us denote by KF the set of all minimal keys of F. It is clear that KF forms a
Sperner-system.
362 V. D. Thi

If K is a Sperner-system over Q, let us define S(K) as 5(A')=min {m: K=KPhf:

M is an mXn matrix representation of i2}. For a Sperner-system K, we can define
the set of antikeys, denoted by K ~ \ as follows:
K~1 = {A<^Q: (B£K)^(B%A) and (AczC) =>(3B£K) (B g C)}.
It is easy to see that K " 1 is the set of subsets of Q, which does not contain the elements
of /Tand which is maximal for this property. They are the maximal non-keys. Clearly,
K~x is also a Sperner-system.
In this paper we assume thatSperner-systems playing the role of the set of mini-
mal keys (antikeys) are not empty (do not contain the full set Q).

§ 2. Connection between minimal keys and antikeys

The following important result was proved in [I], [5]:

Remark 2.1. If K is an arbitrary Sperner-system, then there exists a closure
operation F, for which K=KF and a closure operation F', for which K—Kf,1.
Let us given an arbitrary Sperner-system K={B1, ..., Bm) over Q. We are
now going to construct the set of antikeys K-1. Let us follow the algorithm described
below:
Let JT 1 ={i2\{a}: a^B^. It is easy to see that K1={B1}-1.
Let us suppose that we have constructed Kq= {Bx, ..., Bq}~x for q<m. We
assume that Xl, ..., Xp are the elements of Kq containing Bq+1. So Kq=Fq\J {Xt, ...
...,XP), where Fq={A£Kq: Bq+1%A). For all i (/=1,...,/?), we construct the
antikeys of {Bi+j} on Xi in the analogous way as which are the maximal subsets
of Xi not containing Bq+1. We denote them by A[, ..., A\. (/= 1, ..., p).
Let
1
Kq+1 = {A't: A£Fq => A\ <$: A, 1 s= i ^ p, 1 ^ t ^ t J .

We have to prove, that A9+-1={J?1, ..., Bq+l}~1. For this using the inductive
hypothesis K q ={B!, ...,i? 9 } _1 we show that
a) if AdKq+1 then A is the subset of Q not containing B, ((= 1, ...,q +1)
and being maximal for this property, i.e. A£{Bt, ..., i? 9 + 1 } - 1 ,
b) every AQQ not containing the elements B, (t=1, ...,q+1) and being maxi-
mal for this property is an element of K q + 1 . First we prove the validity of (a). Let
A£Kq+1. If A£Fq then A does not contain the elements Bt (t= 1, ..., q) and A is
maximal for this property and at the same time Bq+1£A. Consequently, A is a
maximal subset of Q not containing B, (<=1, ..., <7+1).
Let AdKq+1\Fq. It is clear that there is an A1, (1 si^p and 1 such
that A—Aj. Our construction shows that B,A\ for all / (/= 1, ..., qr+1). Because
A\ is an antikey of {i?9+i} for Xt we obtain y 4 j = f o r some b£Bq+l. It is
obvious that Bq+1QAi,(j{b}. If a£Q\Xi then, by the inductive hypothesis, for
AitU{a,b}=XiU{a} there exists Bs (s=l,...,q) such that Bs<gA\\J{a,b). X,
does not contain Bu...,Bq by X£Kq. Hence a£Bs. If then
BsQA't\J{a}. For every Bs (1 ^s^q) with ^ U f a } and BS%A} we have
b(LBs. Hence Bs\{a, b}^Alt. Consequently, there exists an A^Fq such that;
 Minimal keys and antikeys 363

AjcA^. This contradicts A£Kq + l\Fq. So there is a Bs (1 Ss^q) such that

Next we turn to the proof of (b). Suppose that A is the maximal subset of £2 not
containing Bt (1 1). By the inductive hypothesis, there is a Y£Kq such that
AQY.
The first case: If Bq+1^Y then Ydoes not contain Bt, ..., Bq+l. Because A is
the maximal subset of Q not containing B, (1 +1) we obtain A=Y. Bq+1%Y
implies A£Fq. Consequently, we have A£Kq+1.
The second case: If Bq+1c Y then Y=Xt holds for some i in {1, ...,p} and
AQA} holds for some t in {1, ...,-r,}. If there exists an A1^Fq such that AitcA1,
then we also have AcAt. By the definition of Fq it is clear that A1 does not contain
Bi, ..., Bq+1. This contradicts the definition of A. Hence A't£Kq+1. It is easy to see
that A\ does not contain Bl, ..., Bq+1. By the definition of A we obtain A = A\,
i.e. Kq+1={Bt, ..., 2? i+1 } -1 .
By the above proof it is clear that Km={Blt ..., Bm}~1. Thus we have
Theorem 2.2. K ^ K ' 1 .
Because AT and are uniquely determined by each other, the determination of
K_1 based on our algorithm does not depend on the order of Bl, ...,Bm.
Now we assume that the elementary step being counted is the comparison of two
attribute names. Consequently, if we assume that subsets of Q are represented as sor-
ted lists of attribute names, then a Boolean operation on two subsets of Q requires
at most |i2| elementary steps.
Let K0 = {O}. According to the construction of our algorithm we have Kq=
^ i ^ U l Z j , ..., X, }, where l^q^m—l. Denote lq the number of elements of Kq.
It is clear that for constructing Kq+1 the worst-case time of algorithm is 0(n2(lq—
-tq)tq) if tq<lq and 0(n2tq) if 1q = tq- Consequently, the total time spent by the
algorithm in the worst cases is

Ш-1
o(n2 2 ',«,)» w h e r e
I= n
-
4=1

It is obvious that, if Fq—Q, then lq=tq.

It can be seen that when there are only a few minimal keys (that is m is small)
our algorithm is very effective, it does not requires exponential time in In cases
for which l q ^l m (Vq: l S g S m - l ) it is obvious that our algorithm requires a
number of elementary operations which is not greater than 0(n 2 |Z||Ar _1 | 2 ). Thus,
in these cases our algorithm finds К i n polynomial time in |i2|, and
After Theorem 2.12 we shall give an example to show that our algorithm requires
exponential time in | Q \. On the other hand Kq in each step of our algorithm is obviously
a Sperner-system. It is known ([4]) that the size of arbitrary Sperner-system over Q
w
can not be greater than ([„/2])' ^ere ([и/2]) asymptotically equal to
364 V. D. Thi

2 " + l/2
. Consequently, the worst-case time of our algorithm can not be more than
(n-nf2
exponential in the number of attributes.
Let K~1={A1, ..., A,} be a set of antikeys. Let R={h0, hx, ..., h,} be a rela-
tion over Î2 given as follows: for all aÇQ, h(a)=0
0 if a£A„
for ; (1 i s I), /I,(a) = { .
if aÇQ\Ar
If we consider R as a matrix, then R represents K (see [5]). Thus, based on our algo-
rithm, for an arbitrarily given Sperner-system K, we can construct a matrix which
represents K.
Example 2.3. Let Q = {1, 2, 3, 4, 5, 6} and K= {(2, 3, 4), (1, 4)}. According to
the above algorithm we have ^ = { ( 1 , 3,4, 5, 6), (1, 2, 4, 5, 6)} U F1, where Fx=
= {(1,2,3,5,6)}, and A, = {(3, 4, 5, 6), (2, 4, 5, 6), (1,2, 3, 5, 6)}. It is obvious
that K~X=K2.
We consider the following matrix:
The attributes:
1 2 3 4 5 6
'0 0 0 0 0 0'
M= 110 0 0 0
2 0 2 0 0 0
LO 0 0 3 0 0.
It is clear that M represents K.
Now we describe the "reverse" algorithm: for given Sperner-system considered
as the set of antikeys we construct its origin. The following definitions are necessary
for us.
Let F be a closure operation over Q. Set
Z(F) = {A g Q: F(A) = A)

and T(JF) = {A a Q: A£Z(F) and AczB=> F(B) = Q.

The elements of Z(F) are called closed sets. It is clear that T(F) is the family of
maximal closed sets (except Q). Now we prove the following lemma.
Lemma 2.4. Let .Fbe a closure operation over Q, and KF the set of minimal keys
of F. Then KF1=T(F).
Proof. Let A be an arbitrary antikey and suppose that AcF(A). Hence
F(F(A))—F(A)=Q. Consequently, A is a key. This contradicts \/B^KF\ B%A.
If there is an A' such that Ac: A' and / f e Z ( F ) \ { i 2 } , then A' is a key. This con-
tradicts A'<zQ.
On the other hand, if A is a maximal closed set and there is a 2? (B£KF) such
that BQA, then F(A)=Q, which conflicts with the fact that AczQ. If AaD(DQ
Q Q), then it can be seen that F(D)= Q (because A is the maximal closed set). Con-
sequently, A is an antikey. The lemma is proved.
 Minimal keys and antikeys 365

Now we construct an algorithm for finding a minimal key.

Let if be a Sperner-system and Q<{H. We take a B (B£H) and an a£Q\B.
We suppose that B={bx, ...,bm}. Let G={B,£H: a$Bt} and T0=B\J{a). define

_ (Tq\{bq+1) if V ^ i i X G : Tq\{bq+1} i B„
TQ+1
I TQ otherwise.

Theorem 2.5. If H i s a set of antikeys, then {r o , 7\, ..., r m } are the keys and
Tm is a minimal key.
Proof. By Remark 2.1 there exists a closure F such that H=Kp1: We prove
the theorem by the induction. It is clear that T0 is a key. If Tq and —T q , then
it is obvious that Tq+1 is a key. If Tq+1=Tq\{bq+1} and F{Tq+1)^Q then, by
Lemma 2.4, there is a B£H such that F(Tq+1)QBt. Hence Tq+1QB„ which
conflicts with the fact MB£H: Tq+1<£Bt. Consequently, Tq+1 is a key.
Now suppose that A is a proper subset of Tm. If a$A, then, clearly, F(A)^Q.
If at: A, then there exists a bqeB such that bq£ Tm\A (1 ^q). By the given algorithm
there exists a B£H\G such that T^^bjQB,. We obtain r m \{fc 9 }g
g r f . 1 { 6 l } g j ( by TmQTq (O^q^m-l). Hence F(A)^Q. Consequently, Tm
is a minimal key. The theorem is proved.
Remark 2.6. Theorem 2.5 is also true if r 0 = { ^ i s is an
arbitrary key.
At this time define
= (Tq\{bq+1} if VBtH: Tq\{bq+1} ^ B,,
q+1
~\Tq otherwise.
— It is clear that the worst-case time of the algorithm is 0(« 2 • where
n=|i2|, \H\ is the number of elements of H.
— It is best to choose B such that |B | is minimal.
— If there is a B such that \/B£H\{B}: B,C\B=® and a£ |J B,
BtiH\{B)
then a\Jb is a minimal key (\/b£B).
— If (£2\ (J B,)^0, then a£Q\ (J B, is a minimal key.
Bt<LH B,£H
— Let Y= U S, If B \ V ^ 0 , then it is best to choose T0=
Bf ff
= ( J e n r ) U { a } U { 4 where b£B\Y.
Remark 2.7. Let H be a Sperner-system (Q$H) and AcQ. We can give an
algorithm (which is analogous to the above one) to decide whether A is a key or not.
If A is a key, then this algorithm finds an A' such that A'Q A and A' is a minimal key.
Remark 2.8. In the paper [5] the equality sets of the relation are defined as
follows: Let R={A l 5 ..., HM} be a relation over Q. For I^J, we denote by EIS
the set {a€i2: hi(a)=hJ(a)}, where l S / S m , l ^ ' s m . Now we define M =
= {ETJ : 3EPQ such that EI} c:E Pi }. Practically, it is possible that there are some ETJ
which are equal to each other. We choose one EU from M. According to Lemma 2.4
it can be seen that M is the set of antikeys of KFR (we consider R as a matrix).
366 V. D. Thi

Example 2.9. Let (2= {1, 2, 3, 4, 5, 6} and R be the following relation:

Ó I 0 0 1 0
10 10 0 1
2 0 0 1 2 2
0 1 2 2 0 3
3 2 1 0 3 0
It can be seen that M = {(1, 2), (3, 4, 5), (4, 6)}, where £ U = { 1 , 2}, £ 1 5 ={4, 6}
and ¿r25= {3, 4, 5}. By Theorem 2.5 and Remark 2.6, it is clear that {1, 3}, {1, 4},
{1, 5}, {1, 6}, {2, 3}, {2, 4}, {2, 5}, {2, 6} are the minimal keys. We use the algorithm
(Theorem 2.5) with J 0 = {3, 4, 6} and T 0 = {4, 5, 6}, then it can be seen that {3, 6}
and {5, 6} are minimal keys. Thus, based on this algorithm for an arbitrarily given
relation R we can find a minimal key of R.
Let K be ah arbitrary Sperner-system. The following theorem has been proved
in [21.
Theorem 2.10.
( ^ j ^ l A r - M s S W - l .

- Denote by the family of all ^-element subsets of Q. Let Fk(n)=max {S(/Q:

Theorem 2.11. ([6])

[n/(2*-2)]
Fk(n)^f2(\k_/J
We define the function f2k-1- N—N for 2k—\ s « by
(2k— \Yiek-»
(i
if « = 0 (mod(2fc-l)),

— iMw/<2fc—i)]—i (2k-U
+P\
lfe-lj X[ fc-1 J if n = p (mod(2fc-l))
and
_ \\lnl(2k-1)]
(t-.'I ]><
(fe-l) if
" = p (mod (2fc-l>)
and k ^ p ^2k-2,
and the function / 2(i _ 2 for 2k—2^ n by
[ f e _jJ if « = 0 (mod(2fc—2)),

x lf n
= < U - l J ( fc-1 J = P (mod(2fc-2))
/¡a-aO) = <
and 1 S p S fc — 1,
f2k-2Yn'<2t-2)1 i f ) , ,
x lf P
t it — I J U - l J " = (mod(2fc-2))
and k^ps2k-3,
 Minimal keys and antikeys 367

where N denotes the set of natural numbers. Let us take a partition Q—Xx U...
...UXmUW, where aw = J and \Xi\ = 2k-l (1 g f e r a ) .
Let

K = {B: \B\ = k,BQ Xt, V/} if \IV\ = 0.

K = {B: |B| = k,B g Xt (1 ^ i s i m - 1 ) a n d B g XmUW} if

K= {B\ = k, B g Xi (1 s? / si m) and B g W} if 2k-2.

It is clear that

K'1 = {A: \AC]Xi\ = Jk —1»Vi> .if \fV\=0.

K~L = {A: \AnXi\ = fe-1 (1 S i S m - 1 ) a n d \A(~)(Xm{JW)\ = /c-1}

if 1 ^ \W\ ^ k—1.

K-1 = {A: 1/4 0*11 = (1 S / S m) and \ADW\ = k-1}

if k^\W\^2k-2.

It can be seen that fik_1(n)=\K~1\. If we take, a partition: Q=X1U... UA^U W;

m anc
where ~\2k—2] ^ l^il =2k~2, in an analogous way we

K={B: \B\ = k,BQXi,\fi} if \W\ = 0.

K= {B-. | 5 | = k,B g Xi (1 ss / m-1) a n d 5 g XmUW} if

K={B: | S | = k,BQX; (1 s i s m ) a n d BQW} if k^\W\s2k-3.

It is clear that \lnl(2fc—2)]

/a-a(") = and / a _ g ( n ) = J

Theorem 2.12. Let £2= {1, ...,«}.

If « = 0 (mod(2A:-2)(2A:-l)), then f 2 k -i(n)^f 2 k -2(n)- For a fixed A:,

/2<fe 1
llim "- ---

Proof. If k=2, then it is easy to prove that V«: / 3 (n)S/ 2 (n). If « = 6 or

n ^ 8 , then / 3 (n)>/»(«). Let

^ _ I fc—l J _ (~T~J

U - l J U - l J
368 V. D. Thi

It is known that n\ = fhin ^jj"xe a " / ( 1 2 n ) , where O<0„<1. So

"~2fcJ p
2"/(a-i). ( c | ( e
|

I ]/7t(fc — 1) J I ]/n(k-2) )

For this E we obtain, that

ln £
= ¿ T (ln i1 - 4 ) +
2 F I 2 ( i l n («<*" 1 » " 2 4 ( F r T ) ) ]
and by
I t . \l

we have
n

2k—1 ¿ 2 (T1"<*<*" ^ " M f i b l ) ) " ¿ I ) '

It can be seen that if k=3, then

and, for every

Hence

2 ^ (T1" ^ " 1
- 24^1) ) - ¿ T >
Consequently, if n = 0 (mod (2k—2)(2k — l)), then /•¡,k-i(n)^'fik_2(n). Now let n
be an arbitrary natural number. It can be seen that, for a fixed k, there exists a number
0 such that
(2k-\+p\ ( p \
—L-k-i i - m V-1* <m

U - i J U - u
 Minimal keys and antikeys 369

Hence In E—°°. Consequently, F-*«>. Thus,

B-^eo ' Tl-*•<&>

fik-ljn)
ftk-M °°'
n-t- 00

(It is easy to see that k=2 is also true.) The theorem is proved.
As a consequence of Theorem 2.12 and Theorem 2.10 we have
Corollary 2.13
FM S i2fik.1(n).

Example 2.14. In Theorem 2.12 let k=2. Then we have n-\r=\K\^n +2

and 3 (B/4) </^(n), where n=\Q\., i.e. 3 (n/4) <|AT _1 |. Thus, we always can construct
an example, in which the number of K (minimal keys) is not greater than n+2, but
the size of (antikeys) is exponential in the number of attributes.

§ 3. Some special Sperner-systems

In this section we investigate connections between the minimal keys and antikeys
for some special Sperner-systems.
The notion of saturated Sperner-system is defined in [7], as follows :
A Sperner-system K over Q is saturated if for any AQQ, ATU {A} is not a
Sperner-system.
An important result in [7] has been proved; if K is a saturated Sperner-system
then K=KF uniquely determines F, where F is a closure operation.
Now we investigate some special Sperner-systems which are strictly connected
with saturated Sperner-systems.
We consider the following example.
Example 3.1. Let Q = {1, 2, 3, 4, 5, 6} and N= {(1, 2), (3, 4), (5, 6)} be a
Sperner-system. It can be seen that i V - 1 = { ( l , 3, 5), (1, 3, 6), (1,4, 5), (1,4, 6),
(2, 3, 5), (2, 3, 6), (2, 4, 5), (2, 4, 6)}. Let K^NUN'1. It is clear that K is saturated.
We use the algorithm which finds a set of antikeys. Then K_1= {(1, 3), (1, 4), (1, 5),
(1, 6), (2, 3), (2, 4), (2, 5), (2, 6), (3, 5), (3, 6), (4, 5), (4, 6)}.
By the fact that K~iU{ 1, 2} is a Sperner-system it is obvious that K-1 is not
saturated. Thus, we have
Corollary 3.2. There is a K so that K is saturated and K-1 is not saturated.
Now we define the following notion.
Definition 3.3. Let K be a Sperner-system over Q. We say that K is embedded,
if for every AÇ.K there is a BÇ.H such that AczB, where H~1=K. We have
Theorem 3.4. Let K be a Sperner-system over Q. K is saturated if and only if
K~l is embedded.
370 V. D. Thi

Proof. Let A' be a saturated Sperner-system. According to the definition of AT-1

it is clear that K ~ l is embedded. Assume that K i s an embedded Sperner-system,
but Kis not saturated. Consequently, for K there exists an AczQ such that K(J{A}
is a Sperner-system. It can be seen that, for every C£K, we have C c Q (because of
Q$K). Hence we can construct B such that AQB, K\J {/?} is a Sperner-system and,
for every B'(B<^B'), there is a C£K with CQB'. It can be seen that B^K'1.
1
This contradicts the fact that K~ is embedded. The proof is complete.
Now we define an. inclusive Sperner-system.
Definition 3.5. Let K be a Sperner-system over £2. We say that K is inclusive,
if for every A£K, there exists a BdK-1 such that BczA. We have
Theorem 3.6. K is an inclusive Sperner-system if and only if K~x is a saturated
one.
Proof. Now, assume that K is an inclusive Sperner-system but A" -1 is not
saturated. By the definition of K~\ there is a ¿ e C ^ - 1 ) - 1 such that A: _1 L!{£}
is a Sperner-system. By Remark 2.1, for AT there is a closure operation F such that
K=KP. If F(B)czQ, then by Lemma 2.4 there exists an A^K^1 with F(B)<^A
(the set of antikeys is family of the maximal closed sets), which conflicts with the
fact that AT-1U {.B} is a Sperner-system. Consequently, B is a key. If we use the
algorithm which finds a minimal key in Theorem 2.5, then it can be seen that there
exists a B'(B'<gB) such that B'£K, and it is clear that K~l\J{B'} is a Sperner-
system. This contradicts the definition of K. Thus, AT-1 is saturated.
On the other hand by the definition of K~x and by the assumption that K"1
is saturated it is clear that K is an inclusive Sperner-system. The theorem is proved.
Now, we have the following corollary by Theorem 3.4 and Theorem 3.6.
Corollary 3.7. Let K be a Sperner-system over Q. Denote H a Sperner-system,
for which H~1 = K. The following facts are equivalent:
(1) AT is saturated,
(2) K - 1 is embedded,
(3) H is inclusive.
Proposition 3.8. There exists a Sperner-system K such that
(1) AT is saturated, but K~1 is not saturated.
(2) K is saturated, but H is not saturated.
(3) K is embedded, but is not embedded.
(4) K is embedded, but H is not embedded.
(5) K is inclusive, but is not inclusive.
(6) K is inclusive, but H is not inclusive,
where H denotes a Sperner-system for which H~1=K.
Proof. From Example 3.1 we have (1). By Theorem 3.4, (AT -1 ) -1 is not embedded
in this example. Hence we have (3). By Theorem 3.6, in Example 3.1 H is inclusive,
where H~1=K. Now, we suppose that, if AT is inclusive, then the set of antikeys of K
is also inclusive. Consequently, in Example 3.1, H is inclusive, and AT is an inclusive
Sperner-system. From Theorem 3.6, K~l is saturated. This constradicts the fact that
AT-1 in Example 3.1 is not a saturated Sperner-system. Hence we have (5). (2) can be
proved as follows: Let AT be a Sperner-system. Let K1=K and, for define
 Minimal keys and antikeys 371

JC by the equality (Kn)~1=Kn~1. We know that the number of the Sperner-systems

over Q is finite (at most 2 2 '"'). On the other hand, AT and K~ x are determined uniquely
by each other. Consequently, there exists a number m ( 2 ^ m s 2 2 ' a l ) such that
Km=K and Km~1=K~1. If we suppose that K is saturated, then H is also saturated,
where H~~l=K. This means that for every p with Kp is also saturated.
This contradicts Corollary 3.2. Thus, there is a Sperner-system A" such that A" is satu-
rated, but H is not saturated. By similar arguments we have also (4) and (6). The
proposition is proved.

Acknowledgement

The author would like to take this opportunity to express deep gratitude to Professor J. De-
metrovics and L. Hannák for their help, valuable comments and suggestions.

COMPUTER A N D AUTOMATION INSTITUTE

H U N G A R I A N A C A D E M Y OF SCIENCES
K E N D E U . 13—17.
BUDAPEST, H U N G A R Y
H—1502

References

[1] ARMSTRONG, W. W., Dependency Structures of Data Base Relationships, Information Processing
74, North-Holland Publ. Co. (1974) 580—583.
[2] BÉKÉSSY, A . , DEMETROVICS J., H A N N Á K L . , KATONA G . O. H . , F R A N K L P . , On the number of maxi-
mal dependencies in data relation of fixed order. Discrete Math., 30 (1980) 83—88.
[3] C O D D , E . F . , Relational model of data for large shared data banks. Communications of the ACM,
13, ( 1 9 7 0 ) 3 7 7 — 3 8 4 .
[ 4 ] DEMETROVICS, J . , On the equivalence of candidate keys with Sperner systems. Acta Cybernetica
4 (1979) 247—252.
[5] DEMETROVICS, J., Relációs adatmodell logikai és strukturális vizsgálata. MTA SZTAKI Tanul-
mányok, Budapest, 114 (1980) 1—97.
[6] DEMETROVICS J., FÜREDI Z., KATONA G. O. H., Minimum matrix representation of closure ope-
rations. Preprint of the mathematical institute of the Hungarian academy of sciences, Budapest,
12 (1983) 1—22.
[7] DEMETROVICS J . , FÜREDI Z . , KATONA G . , A függőségek és az individumok száma közötti kapcsolat
összetett adatrendszerek esetén. Alkalmazott Matematikai Lapok 9 (1983) 13—21.

(Received April 2,1985.)

2 Acta Cybernetica VU/4