Local zero estimates and effective division in rings of algebraic power series (Q1707509): Difference between revisions

This article studies the algebraicity of modules over formal power series rings $k[[x]]$, i.e., the meanwhile classical question under which conditions a module over $k[[x]]$ is the completion of a module over $k\langle x \rangle $. Here $k$ denotes a field, $x=(x_1, \dots ,x_n)$ variables, $k \langle x\rangle $ the subring of algebraic power series. \par The following notations are used: The height of $f\in k \langle x\rangle$ is the maximum $H(f)$ of the degrees of the coefficients of its minimal polynomial. Furthermore, for a $k[[x]]$-module $M$ the order of a non-zero element is defined as $\mathrm{ord}_M(m):= \sup \{ c \in \mathbb{N} \mapsto m\in (x)^c M\}$. If $p\in k[x]^s$ (or $p\in k\langle x\rangle ^s$ respectively), degree (or height respectively) are defined as the corresponding maximum for their components. \par The main result of the paper is Theorem 1.1 which states that for $M=k[[x]]^s / N$ with a submodule $N$ generated by a $k\langle x \rangle $-submodule of $k\langle x \rangle ^s$ there exists a map $C: \mathbb{N} \to \mathbb{R}_{>0}$ such that $\mathrm{ord}_M(f) \leq C(\mathrm{Deg}(f))\cdot\mathrm{H}(f)$ for all $f\in k\langle x \rangle ^s \backslash N$, where $\mathrm{Deg}(f)$ denotes the degree of the field extension $k(x) \to k(x,f)$. Moreover, $C$ depends polynomially on $\mathrm{Deg}(f)$ if $\mathrm{char} (k)=0$. \par An immediate corollary can be deduced, stating that in the above situation there exists a positive constant $C'$ such that for $p\in k[x]^s \backslash N$ we have $\mathrm{ord}_M(p) \leq C' \cdot\deg(p)$ for all $p\in k [ x ] ^s \backslash N$. The condition involved here is also refereed to as \textit{local zero estimate} in a sense influenced by combinatorics. \par As a partial converse of the corollary, the following remarkable result is shown (theorem 1.3): Let $I \subseteq k[[x]] $ be an ideal which is intersection of powers of prime ideals of equal height. If there exists a positive constant $C$ such that $\mathrm{ord}_{k[[x]] / I} (p) \leq C\cdot\deg (p)$ for all $p\in k[[x]] \backslash I $, then $I$ is generated by algebraic power series. Combined with the previously mentioned corollary, this gives a criterion for a principal ideal to admit an algebraic power series as a generator. \par The corollary and the following theorem generalize results of \textit{S. Izumi} [Proc. Japan Acad., Ser. A 68, No. 10, 307--309 (1992; Zbl 0781.32012); J. Math. Kyoto Univ. 32, No. 1, 245--258 (1992; Zbl 0805.32005); in: Real analytic and algebraic singularities. Harlow: Longman. 189--206 (1998; Zbl 0904.32004)], where the assumptions $\mathrm{char} (k)=0$, $s=1$ have been made and $N$ (or $I$, respectively) is a prime ideal. \par The proof of the main theorem is given using an inductive procedure, based on effective Weierstraß-division: When dividing an algebraic power series, the resulting quotient and the remainder are algebraic as well. What is done here, is bounding the complexity of quotient and remainder depending on the complexity invariants of the input. Furthermore, if an algebraic power series $f$ is an element of an ideal generated by algebraic power series $g_1,\dots ,g_p$, bounds are studied for the complexity of algebraic power series $a_i$ such that $f=\sum a_ig_i$. These results -- effective division and the discussion of the effective ideal membership problem for algebraic power series -- are the main ingredients of the proof. They are apparently of independent interest, as well as the included investigation of the ideal membership problem in localizations of polynomial rings. \par The author applies his results to partially answer a long standing question of Hironaka related to the Grauert-Hironaka-Galligo division of a power series $f$ by series $g_1, \dots ,g_s$: Let $f= a_1g_1 + \dots + a_sg_s + r$, where no term in the support of $r$ is divisible by an initial term of any $g_i$ (the fixed term order being induced by a linear form with only positive coefficients via the lex order). By \textit{H. Grauert}'s result from his celebrated paper [Invent. Math. 15, 171--198 (1972; Zbl 0237.32011)], convergence of $f$ and all $g_i$ implies convergence of $r$. The corresponding case for algebraic power series $f, g_1, \dots ,g_s$ is more subtle. In fact, $r$ and the $a_i$ are not algebraic in general. By an example of Kashiwara-Gabber and its generic form discussed here, $r\in k\langle x \rangle $ does not only fail in special cases but seems to be rather an exception. \par The paper concludes with the following contribution to Hironaka's question how to characterize $r$ in the algebraic case, proving a gap theorem for the remainder of the division of algebraic power series: \par By a classical result of \textit{F. K. Schmidt} [Math. Ann. 108, 1--25 (1933; JFM 59.0155.01)], an algebraic power series cannot have too large gaps in its expansion. In the above context, let $r$ be the remainder of the division of an algebraic power series by given ones with respect to a monomial order induced by a linear form. Then for the quotients of degrees $n(k)$ of consecutive non-zero homogeneous parts of $r$, the upper limit $\limsup_{k\to \infty} \frac{n(k+1)}{n(k)} $ is finite. This was known from F. K. Schmidt's result in case $r$ is algebraic. Now we know -- even if the remainder is not algebraic -- the above gap-property is still satisfied.