Pome

12.22 Comparison under an arbitrary modification9BED3D691F82

9BED3D691F82

The orders of degeneration are defined on a fixed manifold. We now show that pulling a nef background back by a modification can only increase them. Recall that a modification $\pi :Y\to X$ is a proper surjective holomorphic map that restricts to a biholomorphism between dense Zariski open subsets. The proof compares Gauduchon tests on $X$ with tests on $Y$. The pullback of a Gauduchon test is degenerate along the exceptional locus, and adding a small positive test on $Y$ repairs this. To control the effect of the added test uniformly, we first bound the pairing of pulled-back tests with a Hermitian form on $Y$ by their pairing with a Hermitian form on $X$, uniformly over the whole cone of tests.

In this subsection $\pi :Y\to X$ is a modification of connected compact complex manifolds of dimension $n\geq 2$. We use Hironaka's analytic elimination of indeterminacies in the following form: there are a finite composition $p:Z\to X$ of blowups along smooth centers and a modification $F:Z\to Y$ with $p=\pi \circ F$. It is obtained by resolving the indeterminacies of the inverse meromorphic map $\pi ^{-1}\circ p$; the identity $p=\pi \circ F$ holds on the dense set where the maps are biholomorphic, hence everywhere. This classical input is discussed in section 25.17. For an effective real divisor $D$, we write $\eta _ D$ for the curvature form of a smooth Hermitian metric on the corresponding real line bundle, as in section 12.35.

Lemma 12.22.1.0A017BFF9468 Let $p:Z\to X$ be a finite composition of blowups along smooth centers of codimension at least two, and let $h_ X$ be a Hermitian form on $X$. There are a constant $C_0>0$, an effective real divisor $D$ on $Z$ whose components are exceptional for $p$, and a smooth curvature form $\eta _ D$ such that

1427E28847CB 12.22.2
\begin{equation} h_ Z=C_0p^*h_ X-\eta _ D>0. \end{equation}

Every component of $D$ is mapped by $p$ into a set of complex codimension at least two in $X$.

Proof. First let $q:M'\to M$ be one blowup along a smooth center, with exceptional divisor $E$, and let $h_ M$ be a Hermitian form on $M$. The restriction of $\mathcal O(-E)$ to each projective fiber of $E$ is $\mathcal O(1)$. A Hermitian metric on the normal bundle of the center therefore induces a metric on $\mathcal O(-E)|_ E$ whose curvature is positive on the tangent spaces of the fibers. Extend this metric to $\mathcal O(-E)$ on $M'$ by extending the smooth logarithm of its ratio to a fixed reference metric from the closed submanifold $E$, and let $\beta $ be its curvature. The semipositive form $q^*h_ M$ is positive off $E$, and at points of $E$ its kernel is exactly the tangent space of the fiber, on which $\beta $ is positive. Hence $Cq^*h_ M+\beta >0$ for every sufficiently large $C$: otherwise, unit tangent vectors violating the inequality for $C\to \infty $ would converge, by compactness, to a vector in the kernel of $q^*h_ M$ on which $\beta $ is nonpositive. With $\eta _ E=-\beta $, the curvature of the induced metric on $\mathcal O(E)$, this reads $Cq^*h_ M-\eta _ E>0$.

For a composition of blowups, apply this step successively, with $h_ M$ the positive form already obtained on the previous blowup. Each step multiplies the previous divisor by a positive constant, pulls it back and adds the new exceptional divisor, so the divisor stays effective and exceptional. This proves (12.22.2). Each exceptional divisor maps into the image of its center, which has dimension at most $n-2$. $\square$

Lemma 12.22.3.51797AE0DE1F Fix Hermitian forms $h_ X,h_ Y$ for a modification $\pi :Y\to X$. There is a constant $K>0$ such that

3145A1C49A8A 12.22.4
\begin{equation} \int _ Y h_ Y\wedge \pi ^*\Psi \leq K\int _ X h_ X\wedge \Psi \end{equation}

for every smooth strictly positive $(n-1,n-1)$-form $\Psi $ on $X$ with $\mathrm{dd}^{\mathrm{c}}\Psi =0$.

Proof. Use the factorization $p=\pi \circ F$ above and lemma 12.22.1. By compactness there is $a>0$ with $h_ Z\geq aF^*h_ Y$. Pulling back the semipositive form $\Psi $ and integrating, we obtain

\[ a\int _ Yh_ Y\wedge \pi ^*\Psi \leq \int _ Zh_ Z\wedge p^*\Psi =C_0\int _ Xh_ X\wedge \Psi -\int _ Dp^*\Psi . \]

Here the changes of variables use that $F$ and $p$ have degree one, and the Poincaré–Lelong formula, together with $\mathrm{dd}^{\mathrm{c}}p^*\Psi =0$, replaces the pairing with $\eta _ D$ by the integral over $D$. This last integral vanishes: on the regular locus of each component of $D$, the restriction of $p$ has rank at most $n-2$, so the restriction of the $(n-1,n-1)$-form $p^*\Psi $ is zero. Taking $K=C_0/a$ proves the assertion. $\square$

Proposition 12.22.5.8AB4DFD2719F Let $\pi :Y\to X$ be a modification of connected compact complex manifolds of dimension $n\geq 2$, and let $\theta $ be a smooth nef background on $X$. Then $\pi ^*\theta $ is nef. For fixed Hermitian forms $h_ X,h_ Y$ and smooth positive probability forms $\mu _ X,\mu _ Y$, there are constants $C,K>0$, independent of $\epsilon $, such that

2D24229F8C67 12.22.6
\begin{equation} \mathcal V_{\mu _ Y}(\pi ^*\theta +\epsilon h_ Y) \leq C\mathcal V_{\mu _ X}(\theta +K\epsilon h_ X) \qquad (\epsilon >0). \end{equation}

Consequently

\[ \underline a(\pi ^*\theta )\geq \underline a(\theta ),\qquad \overline a(\pi ^*\theta )\geq \overline a(\theta ). \]

In dimension one a modification is an isomorphism and preserves both orders.

Proof. Step 1: nefness. Choose $C'$ with $\pi ^*h_ X\leq C'h_ Y$. For $\epsilon >0$, pull back a smooth representative $\theta +\mathrm{dd}^{\mathrm{c}}u_\tau \geq -\tau h_ X$ with $\tau <\epsilon /C'$; this gives a positive representative of $\pi ^*\theta +\epsilon h_ Y$. Thus $\pi ^*\theta $ is nef, and both volumes in (12.22.6) are positive.

Step 2: regularized tests. The ratio $\pi ^*\mu _ X/\mu _ Y$ is smooth, nonnegative and bounded on the compact manifold $Y$, so there is $c>0$ with $\mu _ Y\geq c\pi ^*\mu _ X$. Fix a smooth Gauduchon metric $\gamma _0$ on $Y$, given by theorem 11.4.2. For an arbitrary Gauduchon metric $\gamma $ on $X$, write $\Psi =\gamma ^{n-1}$, and for $\delta >0$ define $\Gamma _\delta $ by

\[ \Gamma _\delta ^{n-1}=\pi ^*\Psi +\delta \gamma _0^{n-1}. \]

The right side is a strictly positive $(n-1,n-1)$-form, and such a form has a unique smooth positive $(n-1)$st root, obtained by solving for the Hermitian matrix of the root from the positive cofactor matrix. Since the right side is $\mathrm{dd}^{\mathrm{c}}$-closed, $\Gamma _\delta $ is a smooth Gauduchon metric on $Y$.

Step 3: the numerator. For $\epsilon >0$, lemma 12.22.3 gives

C7A393B9C4B8 12.22.7
\begin{equation} \lim _{\delta \downarrow 0}\int _ Y(\pi ^*\theta +\epsilon h_ Y) \wedge \Gamma _\delta ^{n-1} =\int _ X\theta \wedge \Psi +\epsilon \int _ Yh_ Y\wedge \pi ^*\Psi \leq \int _ X(\theta +K\epsilon h_ X)\wedge \Psi . \end{equation}

The first term in the middle is nonnegative because $\theta $ is nef, and the second is positive because $\pi ^*\Psi $ is positive on the set where $\pi $ is biholomorphic, so the limit is positive.

Step 4: the denominator. We claim that

1419E4EEEE39 12.22.8
\begin{equation} \lim _{\delta \downarrow 0}D_{\mu _ Y}(\Gamma _\delta ) =\int _ Y(\pi ^*\gamma ^ n)^{(n-1)/n}\mu _ Y^{1/n} \geq c^{1/n}D_{\mu _ X}(\gamma )>0. \end{equation}

To justify the limit on the exceptional locus, represent $\Gamma _\delta ^{n-1}$ by its positive cofactor matrix. The density of $(\Gamma _\delta ^ n)^{(n-1)/n}$ is a fixed dimensional constant times the $1/n$th power of the determinant of that matrix, and this expression extends continuously to positive semidefinite matrices. Since $Y$ is compact, dominated convergence applies. The inequality follows from $\mu _ Y\geq c\pi ^*\mu _ X$ and the change of variables for the degree-one map $\pi $.

Step 5: conclusion. Test the infimum defining $\lambda _{\mu _ Y}(\pi ^*\theta +\epsilon h_ Y)$ with $\Gamma _\delta $, use (12.22.7) and (12.22.8), and let $\delta \downarrow 0$. For every Gauduchon metric $\gamma $ on $X$ this gives

\[ \lambda _{\mu _ Y}(\pi ^*\theta +\epsilon h_ Y) \leq c^{-1/n} \frac{\int _ X(\theta +K\epsilon h_ X)\wedge \gamma ^{n-1}}{D_{\mu _ X}(\gamma )}. \]

Taking the infimum over $\gamma $ and then the $n$th power proves (12.22.6) with $C=c^{-1}$.

Dividing the logarithm of (12.22.6) by $\log \epsilon <0$ reverses the inequality. The bounded error $\log C$ and the rescaling $\epsilon \mapsto K\epsilon $ do not affect lower and upper limits, as in the proof of proposition 12.20.1. This gives $\underline a(\pi ^*\theta )\geq \underline a(\theta )$ and $\overline a(\pi ^*\theta )\geq \overline a(\theta )$. Finally, a proper holomorphic map of degree one between compact Riemann surfaces is an isomorphism, which settles the case $n=1$. $\square$

Cite this page

Comments

Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.