In this paper we consider an algorithmic technique more general than t
hat proposed by Zharkov and Blinkov for the involutive analysis of pol
ynomial ideals. It is based on a new concept of involutive monomial di
vision which is defined for a monomial set. Such a division provides f
or each monomial the self-consistent separation of the whole set of va
riables into two disjoint subsets. They are called multiplicative and
non-multiplicative. Given an admissible ordering, this separation is a
pplied to polynomials in terms of their leading monomials. As special
cases of the separation we consider those introduced by Janet, Thomas
and Pommaret for the purpose of algebraic analysis of partial differen
tial equations. Given involutive division, we define an involutive red
uction and an involutive normal form. Then we introduce, in terms of t
he latter, the concept of involutivity for polynomial systems. We prov
e that an involutive system is a special, generally redundant, form of
a Grobner basis. An algorithm for construction of involutive bases is
proposed. It is shown that involutive divisions satisfying certain co
nditions, for trample, those of Janet and Thomas, provide an algorithm
ic construction of an involutive basis for any polynomial ideal. Some
optimization in computation of involutive bases is also analyzed. In p
articular, we incorporate Buchberger's chain criterion to avoid unnece
ssary reductions. The implementation for Pommaret division has been do
ne in Reduce. (C) 1998 IMACS/Elsevier Science B.V.