A7364A53065E
Multiplier ideals give another way of comparing singularities. By the valuative criterion of theorem 6.12.16, inclusions between multiplier ideals are governed by generic Lelong numbers along divisors over $X$. We identify the resulting order and show that it is preserved by upper envelopes, which requires pulling envelopes back to smooth models.
In this section $\{ \theta \} $ may be any pseudo-effective class. A prime divisor over $X$ is a prime divisor $E$ on a smooth model of $X$, as in definition 6.12.3, and $\nu (u,E)$ denotes its generic Lelong number (proposition 6.9.7). Since multiplier ideals are local, the valuative criterion uses prime divisors over small neighborhoods of a point; accordingly, in this section a prime divisor over $X$ may also be a prime divisor over an open subset $U\subset X$, that is, a prime divisor on a smooth model of $U$, and $\nu (u,E)$ then refers to the restriction of $u$ to $U$. Multiplier ideals and Lelong numbers of quasi-plurisubharmonic functions are computed from local plurisubharmonic functions that differ from them by smooth functions (see the definitions following lemma 6.14.4), so the results of section 6.6 apply to them.
Proposition 16.24.1.585A4246348B For quasi-plurisubharmonic functions $u,v$ on $X$, the following are equivalent:
$\mathcal I(tv)\subseteq \mathcal I(tu)$ for every real $t>0$;
$\mathcal I(tv)\subseteq \mathcal I(tu)$ for every positive integer $t$;
$\nu (v,E)\geq \nu (u,E)$ for every prime divisor $E$ over $X$.
We write $v\preceq _{\mathcal I}u$ when these conditions hold, and $v\sim _{\mathcal I}u$ when $v\preceq _{\mathcal I}u$ and $u\preceq _{\mathcal I}v$. In particular, equality of the multiplier ideals of all positive integer multiples is equivalent to equality for all positive real multiples, and to equality of all divisorial Lelong numbers.
Proof.
Condition (2) is a special case of (1). Assume (2) and let $E$ be a prime divisor over an open subset $U\subset X$ (possibly $U=X$). The inclusions (2) restrict to $U$, and a smaller ideal has a larger order along $E$, so $\operatorname {ord}_ E\mathcal I(tv)\geq \operatorname {ord}_ E\mathcal I(tu)$ for every positive integer $t$. Dividing by $t$ and letting $t\to \infty $, corollary 6.12.14, applied on $U$, gives $\nu (v,E)\geq \nu (u,E)$; this is (3). Conversely, assume (3), let $t>0$, and let $f$ be a germ of $\mathcal I(tv)$ at a point $x$. The valuative criterion, theorem 6.12.16, gives $\varepsilon >0$ and a neighborhood $V$ of $x$ such that
\[ \operatorname {ord}_ E f+A_ X(E) \geq (1+\varepsilon )t\nu (v,E) \geq (1+\varepsilon )t\nu (u,E) \]
for every prime divisor $E$ over an open neighborhood of $x$ contained in $V$ whose center contains $x$; the second inequality is (3) for these divisors. Since $t\nu (u,E)=\nu (tu,E)$ by proposition 6.7.9, the same criterion, applied to $tu$, shows that $f\in \mathcal I(tu)$; this is (1). Applying the equivalence in both directions proves the assertions about equalities.
$\square$
Lemma 16.24.2.FFB5A7083041 Let $(v_ j)_{j\in J}$ be a nonempty family in $\mathrm{PSH}(X,\theta )$ that is uniformly bounded above, and let $v=(\sup _ jv_ j)^*$. Then
\[ \nu (v,E)=\inf _ j\nu (v_ j,E) \]
for every prime divisor $E$ over $X$. Consequently, maxima, sums and positive multiples preserve the relation $\sim _{\mathcal I}$; and if two such families are termwise equivalent for $\sim _{\mathcal I}$, their upper envelopes are equivalent as well.
Proof.
Let $\pi \colon Y\to U$ be a smooth model of an open subset $U\subset X$ (possibly $U=X$) on which $E$ is a divisor, and let $Z\subset Y$ be a nowhere dense analytic subset outside which $\pi $ is a biholomorphism onto an open subset of $U$; such a set exists by the definition of a smooth model, and it has Lebesgue measure zero. We first show that $v\circ \pi $ is the regularized supremum of the family $(v_ j\circ \pi )$. By theorem 5.10.2, a countable subfamily has the same regularized supremum $v$, and its increasing finite maxima $w_ k$ satisfy $\sup _ kw_ k=v$ almost everywhere. The functions $w_ k\circ \pi $ are $\pi ^*\theta $-plurisubharmonic, increase with $k$, and are locally uniformly bounded above; let $W$ be their regularized supremum, so that $W=\sup _ kw_ k\circ \pi $ almost everywhere by theorem 5.10.2. If $N\subset U$ is a Lebesgue null set outside which $\sup _ kw_ k=v$, then $\pi ^{-1}(N)\setminus Z$ is a null set, because $\pi $ restricts to a biholomorphism of $Y\setminus Z$ onto an open set; hence $\sup _ kw_ k\circ \pi =v\circ \pi $ outside the null set $(\pi ^{-1}(N)\setminus Z)\cup Z$. Thus the quasi-plurisubharmonic functions $W$ and $v\circ \pi $ agree almost everywhere, and therefore everywhere. Since every $v_ j$ is at most $v$, we get $W\leq (\sup _ jv_ j\circ \pi )^*\leq v\circ \pi =W$, so $v\circ \pi $ is the regularized supremum of the whole family $(v_ j\circ \pi )$.
It remains to compare Lelong numbers. Near a point $y\in E$, adding to all these functions the pullback of a smooth local potential of $\theta $ near $\pi (y)$ makes them plurisubharmonic, so proposition 6.7.9 gives $\nu (v\circ \pi ,y)=\inf _ j\nu (v_ j\circ \pi ,y)$ at every point $y\in E$. Taking the infimum over $y\in E$, which defines the generic Lelong number (proposition 6.9.7) and commutes with the infimum over $j$, we obtain the formula.
For the remaining assertions, let $\varphi $ and $\psi $ be quasi-plurisubharmonic functions on $X$, and recall from proposition 6.7.9 that Lelong numbers at a point are additive and positively homogeneous, and that the Lelong number of a maximum is the minimum of the Lelong numbers. Taking infima over $y\in E$, we obtain $\nu (c\varphi ,E)=c\nu (\varphi ,E)$ for $c>0$ and $\nu (\max \{ \varphi ,\psi \} ,E)=\min \{ \nu (\varphi ,E),\nu (\psi ,E)\} $. For sums we use that, by proposition 6.9.7, the infima defining $\nu (\varphi ,E)$, $\nu (\psi ,E)$ and $\nu (\varphi +\psi ,E)$ are attained at every point of $E$ outside a countable union of proper analytic subsets of $E$. Such a point exists, since by the same proposition the complement in $E$ of such a union is dense in $E$, and in particular nonempty. At a point of $E$ where the three infima are attained, the additivity at a point gives $\nu (\varphi +\psi ,E)=\nu (\varphi ,E)+\nu (\psi ,E)$. By these three rules and proposition 16.24.1, maxima, sums and positive multiples preserve $\sim _{\mathcal I}$, and by the formula just proved, so do upper envelopes of termwise equivalent families.
$\square$
Comments
Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.