A finiteness property of graded sequences of ideals (Q442441)

In recent years, the asymptotic properties of graded sequences of ideals \(\mathfrak{a}_{\bullet}= (\mathfrak{a}_{m})_{m\geq 1}\) on algebraic varieties have been studied systematically (see for example [\textit{L. Ein} et al., Pure Appl. Math. Q. 1, No. 2, 379--403 (2005; Zbl 1139.14008)] and [Ann. Inst. Fourier 56, No. 6, 1701--1734 (2006; Zbl 1127.14010)]). The main motivating example is the graded sequence \(\mathfrak{a}_{\bullet}^{L}\) associated to a line bundle \(L\), where \(\mathfrak{a}_{k}^{L}\) is the multiplier ideal sheaf associated to the linear system \(|L^{k}|\). Another more analytic example is the graded sequence \(\mathfrak{a}_{\bullet}\) associated to a quasi-psh function \(\varphi\), where \(\mathfrak{a}_{k}\) is the multiplier ideal sheaf associated to the quasi-psh function \(k\cdot\varphi\). It has become more and more evident that in order to study a single linear series, we should study it in an asymptotic way. In this article, to answer a question raised by M.Păun (Păun's question was motivated by the infinite blow-up process in Siu's analytic approach to the problem of the finite generation of the canonical ring, cf. Section 6.3 of [``Techniques for the analytic proof of the finite generation of the canonical ring'', \url{arXiv:0811.1211}]), the authors study the log canonical threshold of graded sequence. Recall first that for a single ideal sheaf \(\mathfrak{a}\), the log canonical threshold \(\mathrm{lct}(\mathfrak{a})\), which measures the singularity of the multiplier ideal sheaf, is defined as \[ \mathrm{lct}(\mathfrak{a})=\inf\limits_{E}\frac{A( \mathrm{ord}_{E} ) }{ \mathrm{ord}_{E} ( \mathfrak{a} ) }, \] where the infimum is taken over all divisors \(E\) over \(X\) (i.e., all prime divisors on normal varieties \(Y\) that have a proper birational morphism to \(X\)), and \(A( \mathrm{ord}_{E})\) is the log discrepancy of \(E\) over \(X\). For a graded sequence \(\mathfrak{a}_{\bullet}\), one can define the log canonical threshold as \[ \mathrm{lct}(\mathfrak{a}_{\bullet})=\lim\limits_{m\rightarrow \infty}m\cdot \mathrm{lct}(\mathfrak{a}_{m}) . \] One of the most remarkable properties of the above asymptotic definition is that, if \(L\) is a big line bundle on a smooth projective variety, then \(\mathrm{lct} (\mathfrak{a}_{\bullet}^{L})=\infty\) if and only if \(L\) is numerically effective (see [``Valuations and asymptotic invariants for sequences of ideals'' \url{arXiv:1011.3699}] by the same authors). The main result of the article can be now state as follows. Let \(\mathfrak{a}_{\bullet}\) be a graded sequence of ideals on a smooth variety \(X\). Let \(I\subseteq \mathbb{Z}_{> 0}\) be a subset such that for all \(m\in I\) we have a divisor \(E_{m}\) over \(X\) that computes \(\mathrm{lct}(\mathfrak{a}_{m})\). If \(\{A ( \mathrm{ord}_{E_{m}})\}_{m\in I}\) is bounded, then the set \(\{E_{m}\}_{I}\) is finite. As a direct consequence, the infinite blow-up process in Siu's paper can be ``stopped'' if \(\{A ( \mathrm{ord}_{E_{m}})\}\) is bounded. This paper looks very nice and interesting.