The name QBGenerator derives from its intended use as a (monomial)
quotient basis generator, that is a way of generating a factor closed
(vector space) basis of power products for the quotient of a
polynomial ring by a zero-dimensional ideal. It is used in the
implementation of the FGLM and the Buchberger-Moeller algorithms -- in
fact these are really the same algorithm (for computing a Groebner
basis of an intersection of one or more zero-dimensional ideals).
Let P denote a polynomial ring (with coefficients in a field
k), and let I be a zero-dimensional ideal in P. Then
mathematically the quotient P/I is a finite dimensional vector space
over k. We seek a basis QB for P/I which is a factor closed
set of power products; i.e. if the power product t is in QB then any
factor of t is in QB too. Groebner basis theory guarantees that such
bases exist; actually it was first proved by Macaulay (a person, not a
computer algebra system).
The elements of QB are determined one at a time, obviously
starting with the trivial power product, 1. Moreover, at every stage
the set of elements in the partially formed QB is factor closed,
and this implies that only certain PPs are candidates for being
adjoined to the QB (we call these corners). When a new
element is adjoined to the QB new elements may appear in the
corner set, these newly adjoined elements form the new corner set
(this is always a subset of the corner set, and may be empty).
During the determination of the QB, some power products will be
discovered which cannot be in the QB (usually based on the failure of a linear
independence criterion). Such PPs form the avoid set: the
QBGenerator will exclude all multiples of all elements of the
avoid set from subsequent consideration.
QBGenerator(PPM) where PPM is the PPMonoid in which we shall
calculate; initially the quotient basis is empty, and the corner set contains
just 1.
There are 3 accessor functions, and 2 true operations:
QBG.myQB() gives the current elements of the quotient basis (as a
vector) in the order they were added;
QBG.myCorners() gives the current elements of the corner set (as a list);
QBG.myNewCorners() gives the newly added elements to the corner set
(as a list);
QBG.myCornerPPIntoQB(pp) move the element pp of the corner set
into the quotient basis (this updates both the corner set and the new corner set);
QBG.myCornerPPIntoAvoidSet(pp) move the element pp of the corner set
into the avoid set (all multiples of pp will skipped hereafter).
The tricky part was designing a good interface. The implementations themselves are relatively straightforward (and actually contain some useful comments!)
The function QBGenerator::myCornerPPIntoQB makes local copies of some
fields to permit full exception safety. This may adversely affect execution
speed, but I believe that in the context of FGLM & Buchberger-Moeller the
slow-down will be negligible (but I have not actually tested my guess).
Class QBGenerator could offer a ctor which accepts a (good)
estimate of the dimension of the quotient, i.e. final number of
elements in the QB. It could use this value to reserve space for
myQBList.