Four Functions Theorem

Four Functions Theorem
E. Roventa
Computer Science Dept.
Glendon College, York University
2275 Bayview Ave.
Toronto (Ont.), Canada M4N 3M6
[email protected]
T. Spircu
Dept. of Medical Informatics
"C. Davila" Univ. of Medicine and Pharmacy
8 Eroii Sanitari Blvd.
Bucharest, Romania
[email protected]
Abstract – Isotone real functions defined over a finite Boolean algebra are studied. Weak maxima and weak
minima are defined. A four functions theorem is proved.
Index terms: Boolean algebra, isotonic function.
1. WEAK MAXIMA AND MINIMA
It is well known [1] that any finite Boolean algebra  = (X, 0, 1, ¯, , ) is isomorphic with
the Boolean algebra (2N, , N, , , ) of the subsets of N = {e1, e2, …, en}, the set of atoms
of .
The inclusion between subsets of N transforms 2N into a partial ordered set.
In what follows we will consider real functions defined on 2N that are compatible with this
partial order and we will explore this structure.
Definition. A real function  : 2 N  R  is called isotonic if it satisfies the following
condition:
(ISO) if M  P  N , then ( M )  ( P) .
Denote by Iso(N) the set of isotonic functions  : 2 N  R  .
Of course, the algebraic structure of R  (additive semigroup, multiplicative group) is
transferred on Iso(N) in an obvious manner.
In what follows we will explore the transfer of the natural order structure of R  , which
seems slightly more interesting.
Given ,   Iso( N ) ,    means ( M )  ( M ) for any M  N . Of course, we could
define   max( , ) and   min( , ) by
 ( M )  max( ( M ), ( M ))
 ( M )  min( ( M ), ( M )) for M  N .
But, because our aim is to extend the Ahlswede-Daykin theorem (i.e. the four functions
theorem), the weak-max  and the weak-min  operations on Iso(N) will not be defined in
the obvious manner.
Definition. Given ,  : 2 N  R  , the function  : 2 N  R  is defined recursively as
follows:
(1) ()()  max{ (), ()} ,
(2) for M  N , M  
{
J M
()( M )  max ()( J )  max
{( K )  ( L)}} .
K , L N
K L J
K LM
( J ) ( J )
In a dual manner, the function  : 2 N  R  is defined recursively as follows:
(1’) ()()  min{ (), ()} ,
(2’) for M  N , M  
{
J M
( )( M )  max ()( J ) 
min
{(M )  (M )}}.
K , L N
K L J
K LM
( K )
 ( L)
In particular, for N = {1} we have
( N ) ( N )
()( N )  max {(), ()}  max{
,
},
() ()
( N ) ( N )
()( N )  min {(), ()}  min {
,
},
() ()
Proposition 1. If ,   Iso( N ) , then ,   Iso( N ) .
Proof. Let us consider M  P  N . Taking into account the recursive definition of  ,
we have
( K ) ( L)
()( P)  ()( M )  max {

}.
K , L N ( M )  ( M )
K LM
K L P
Because M  K , L  P , we have ( M )  ( K ) and ( M )  ( L) , thus the factor max is
 1 . Hence ()( M )  ()( P) .
In a dual manner, taking into account the recursive definition of  , we have
( P) ( P)
( )( P)  ()( M )  min {

}.
K , L N ( K )  ( L)
K LM
K L P
This time the factor min is  1 , because ( P )  ( K ) and  ( P )   ( L) , hence
( )( M )  ( )( P) .
Proposition 2. If ,   Iso( N ) , then:
(1) for any M , P  N we have
( M )  ( P)  ()( M  P)  ()( M  P)
and in particular
(2) for any M  N we have
( M )  ( M )  ()( M )  ( )( M ) .
Proof. Let us begin with (2). For M   we have, by definition,
()()  ()()  ()  () .
Suppose M   . Then there exists J *  M and K *, L * such that K * L*  J * ,
K *  L*  M and
( M ) ( M )
.
()( M )  ()( J *) 

( K *) ( L*)
By induction we know that ()( J *)  ()( J *)  ( J *)  ( J *) . From here, taking into
account the recursive definition of  , we have
()( M )  ( )( M ) =
( M ) ( M )
( K ) ( L)
()( J *) 

 max ()( J )  max {

} 
( K , L ) ( J ) ( J )
( K *) ( L*) J  M
( M ) ( M )
( K *) ( L*)
()( J *) 

 ()( J *) 

 ( M )  ( M ) .
( K *) ( L*)
( J *) ( J *)
{
}
Let us prove now (1). Consider M , P  N .
If M  P   , then obviously we have the equality
( M )  ( P)  ()( M  P)  ()( M  P)
because M  N  M  P   .
If M  P   , then M  P  M  P and we have
()( M  P)  ()( M  P) 
 ()( M  P) 
max
( K )
 ( L)

}
( M  P) ( M  P)
{
K , L N
K LM P
K LM P
( M )
( P)
.

( M  P) ( M  P)
Now we could exploit (recursively) the inequality
()( M  P)  ()( M  P)  ( M  P)   ( M  P) .
and the proposition is proved.
The operations weak-max  and the weak-min  are not genuine max/min operations. For
example, in general      , all what can be established is that
   ,    .
But the properties of these operations are well fitted for an extension of the AhlswedeDaykin theorem.
2. THE FOUR FUNCTIONS THEOREM
Restate the classical 4-functions theorem as follows.
Theorem (Ahlswede-Daykin). If  = (L, , ) is a finite distributive lattice and if
 ,  ,  ,  : L  [0, ) are such that
 ( x)  ( y )   ( x  y ) ( x  y )
for every x, y  L , then
 ( X ) (Y )   ( X  Y ) ( X  Y )
for every X , Y  L .
Here the extension  of a function  : L  [0, ) is defined by
 (Z ) 
( z) Z ( z) for Z  L
zL
where  Z is the characteristic function of Z (as a subset of L). Moreover, X  Y and X  Y
are defined as follows:
a) if X   or Y   , then X  Y  X  Y   ;
b) if X , Y   , then X  Y  {x  y | x  X , y  Y } şi X  Y  {x  y | x  X , y  Y } .
The proof can be found in [2] or [3].
Consider now the Boolean algebra  = (2N, , N, , , ) and suppose  : 2 N  [0, ) is
~
an ordinary function. Denote  : Iso( N )  [0, ) the extension of  to Iso(N), defined by
~
 ( )    ( M )  ( M ) .
M N
Proposition 3. Suppose  ,  ,  ,  : 2 N  [0, ) are four ordinary functions satisfying the
condition
 ( M )   ( P)   ( M  P)   ( M  P)
for all M , P  N . Then
~
~
~()   ()  ~()   () for any ,   Iso ( N )
where  is the weak minimum and  is the weak maximum of isotonic functions.
Proof. Suppose  ,  ,  ,  satisfy the condition
 ( M )   ( P)   ( M  P)   ( M  P) for M , P  N .
and ,  are isotonic functions defined on 2 N . Then, from Proposition 2 (1) above one has
( M )  ( P)  ()( M  P)  ()( M  P)
If we denote
 ' ( X )   ( X )  ( X ) ,  ' ( X )   ( X )  ( X ) ,  ' ( X )   ( X )  ()( X ) ,
 ' ( X )   ( X )  ()( X ) for X  N ,
then it is obvious that
 ' ( M )   ' ( P)   ' ( M  P)   ' ( M  P) for M , P  N .
By applying Theorem 1 above (where L is the canonical Boolean algebra over 2 N . one
obtains
 ' (2 N )   ' (2 N )   ' ( 2 N )   ' (2 N ) .
Now,
 ' (2 N ) 
 ' ( z)   ' (M )   (M )  (M )  ~() ,
z2 N
M N
M N
~
~
similarly  ' (2 N )   () ,  ' (2 N )  ~() ,  ' (2 N )   () , and Proposition 3 is
proved.
REFERENCES
[1]
[2]
[3]
[4]
G. Birkhoff, Lattice Theory. Amer. Math. Colloq. Publ. Vol. 25. New York, 1948.
B. Bollobás, Combinatorics. Cambridge Univ. Press, Cambridge UK, 1986.
P. C. Fishburn, Correlations in p. o. sets. Discrete Appl. Math. 39(1992), 173-191.
P. C. Fishburn and L. Shepp, The Ahlswede-Daykin theorem. In: I. Althöfer, N. Cai, G. Dueck, L.
Khachatrian, M. S. Pinsker, A. Sárközi, I. Wegener, Z. Zhang (Editors), Numbers, Information
and Complexity. Kluwer Acad. Publ., Boston-Dordrecht-London, 2000.