Uniform expansions along monomial curves #
Suppose the Taylor coefficients of a polynomial p below an exponent v in lexicographic
order vanish at every point of a set S. An evaluator c separates v from all larger
exponents by weighted degree. There is then a single polynomial remainder, with coefficients
polynomial in the center a, such that on S
p(a + y^c) = y^(weight c v) * (p_{a,v} + y * R(a,y)).
The coefficient ring can itself be a polynomial ring in a further variable. In that case the
identity retains that variable, even when ordinary specialization vanishes identically. The
least removed Lazard exponent on a set supplies the required vanishing of lower coefficients;
for nonzero p, the leading term at every point attaining that minimum is its nonzero Lazard
evaluation.
This uniform identity is used to deform Lazard evaluations into ordinary fibers.
References #
S. McCallum, A. Parusiński, L. Paunescu, Validity proof of Lazard's method for CAD construction, Journal of Symbolic Computation 92 (2019), 52–69. See arXiv:1607.00264v2, Section 5.1, Proposition 5.6, equation (8).
If all Taylor coefficients lexicographically below v vanish on S, restriction to a
monomial curve with evaluator c factors uniformly as X^(weight c v) times the sum of the
coefficient at v and X times a remainder polynomial. The remainder is polynomial in the center
and the curve parameter, not a separately chosen polynomial at each center. No nonvanishing
of the coefficient at v is required.
On any nonempty set of base points, choose an evaluator for all removed Lazard exponents
and a prescribed finite set V of other exponents, and the lexicographic minimum of the removed
exponents. The polynomial admits a single remainder expansion on the whole set with that
minimum as the common factored exponent. For R = A[Z], the remainder retains Z as well as the
center and the monomial-curve parameter. The coefficient at the minimum may vanish at points with
a larger removed exponent. The extra set V allows the same curve to detect the valuations of
leading coefficients, trailing coefficients, and discriminants.