Several Remarks on Index Generation Functions Dan A. Simovici Marius Zimand Univ. of Massachusetts Towson University Boston Dept. of Comp. and Dept. of Comp. Science Information Sciences Boston, Massachusetts Towson, Maryland dsim@cs.umb.edu mzimand@towson.edu Dan Pletea Univ. of Massachusetts Boston Dept. of Comp. Science Boston, Massachusetts dpletea@cs.umb.edu Abstract—We propose a probabilistic greedy algorithm for decomposing partially specified index generation functions. These functions have numerous applications in a variety of circuit design problems. We show that finding an optimal decomposition is an intractable problem, which motivates our approach. Keywords-index generation function; NP-complete; proba- bilistic algorithm; I. I NTRODUCTION Partially specified index generation functions (PSIGFs) play an important role in circuit design due to their multiple applications. As pointed by T. Sasao [4], [5], [6], PSIGFs play are used in implementing address tables for routers, in building terminal access controllers for local area networks and they have other applications in memory patch circuits, electronic dictionaries, password lists, etc. A partial function f between the sets A and B is denoted as f : A B. Its definition domain, which is a subset of A is denoted as Dom f . A PSIGF is an injective partial function whose definition domain Dom(f ) is a subset of {0, 1} k and whose set of values is {1,...,n}, where n = | Dom(f )|. The equality n = | Dom(f )| together with the injectivity of f means that f is a bijection between Dom(f ) and {1,...,n}. The mem- bers of Dom(f ), written as row vectors in {0, 1} k are re- ferred to as registered vectors. Our purpose is to formulate an algorithm that allows us to replace the variables x 1 ,...,x k with a smaller number of linear combinations of these variables y 1 ,...,y (over GF(2)) defined by linear functions of the form y i = g i (x 1 ,...,x n )= x ij1 ⊕···⊕ x ijp i such that we can express (for some suitable F ) f (x 1 ,...,x k )= F (g 1 (x 1 ,...,x k ),...,g (x 1 ,...,x k )) for all (x 1 ,...,x k ) Dom(f ). We mention that linear techniques have been used frequently in circuit design (see, for example [3]). As we shall see in Section III, this is an intractable problem. We propose a greedy algorithm that achieves a satisfactory solution. II. BOOLEAN MATRICES AND COLLECTIONS OF SETS Let GF(2) = ({0, 1}, +, ·, 0, 1) be the two-element Galois field, where the addition is the exclusive-or operation defined by a + b = 0 if a = b, 1 otherwise. It is common to also write ab instead of a+b. Vectors from GF(2) p will be denoted by bold letters; a vector that has only one component equal to 1, located in the i th place will be denoted by e i . The transpose of a matrix A is denoted by A . The components of a vector x GF(2) p will be denoted by x 1 ,...,x p . The one-column vector in GF(2) k whose components are equal to 1 is denoted by 1 k ; the one- column vector in GF(2) k whose components are equal to 0 is denoted by 0 k . The set of n × k-matrices over GF(2) is denoted by GF(2) n×k . The standard product of two matrices P GF(2) m×n and Q GF(2) n×p , where the elements of B are regarded as the real numbers 0 and 1 will be denoted by AB. The inner product of vectors x GF(2) n , y GF(2) n , is denoted xy. Suppose that Dom(f )= {m 1 ,..., m n }, where f is a PSIGF. The problem that we are discussing can be formu- lated in matrix terms as follows. Given an n × k-matrix over GF(2), M f = m 1 . . . m n we look for a minimum number and a matrix A = (a 1 ··· a ) GF(2) k× such that for all i = j we have 2012 IEEE 42nd International Symposium on Multiple-Valued Logic 0195-623X/12 $26.00 © 2012 IEEE DOI 10.1109/ISMVL.2012.17 179