Pome

12.3 The equality case of concavity and the first variation484879715C58

484879715C58

Since $\lambda _\mu $ is an infimum of linear functions, each minimizing Gauduchon metric at $\theta $ defines a linear functional that supports $\lambda _\mu $ at $\theta $. If equality holds in the concavity inequality along a segment, a metric minimizing at an interior point must minimize at both endpoints, and theorem 12.2.3 then makes the two endpoint solution forms proportional. Differentiating the smooth equation shows that the supporting functional is the derivative of $\lambda _\mu $. For $\theta \in \mathscr P$ we write $\gamma _\theta $ for a minimizing Gauduchon metric, which exists by theorem 12.2.3; variation directions are smooth real $(1,1)$-forms.

Theorem 12.3.1.2220E0C47510 Let $\theta _0,\theta _1\in \mathscr P$ and $0<s<1$. Then

\[ \lambda _\mu ((1-s)\theta _0+s\theta _1) \geq (1-s)\lambda _\mu (\theta _0)+s\lambda _\mu (\theta _1). \]

Equality holds if and only if, for some $r>0$ and some smooth real $f$,

\[ \theta _1=r\theta _0+\mathrm{dd}^{\mathrm{c}}f. \]

Proof. Concavity was proved in proposition 12.2.2. Suppose first that $n\geq 2$ and that equality holds. Take a minimizing Gauduchon metric for the middle background $(1-s)\theta _0+s\theta _1$. Its quotient at the middle background is the same convex combination of its quotients at $\theta _0$ and $\theta _1$, and each of these is at least the corresponding infimum. Equality therefore forces this metric to minimize both endpoint quotients.

By theorem 12.2.3, the solution forms $\Omega _ i=\Omega _{\theta _ i}$, with constants $c_ i=c_{\theta _ i}(\mu )$ for $i=0,1$, are then both conformal to this metric. Their volume forms are $c_0\mu $ and $c_1\mu $, so their pointwise proportionality factor is the constant $r=(c_1/c_0)^{1/n}$. Hence $\Omega _1=r\Omega _0$, which gives the stated smooth $\mathrm{dd}^{\mathrm{c}}$ relation. Conversely, if that relation holds, invariance under smooth $\mathrm{dd}^{\mathrm{c}}$ changes and homogeneity give equality.

When $n=1$, the function is $\theta \mapsto \int _ X\theta $. Take $r=\int _ X\theta _1/\int _ X\theta _0$. The difference $\theta _1-r\theta _0$ has integral zero, so the Green operator of the compact curve writes it as $\mathrm{dd}^{\mathrm{c}}f$ with $f$ smooth. Thus the same equivalence holds. $\square$

Proposition 12.3.2.94AFD9E43DE0 The function $\lambda _\mu $ is smooth along smooth finite-dimensional families in $\mathscr P$. For a smooth real $(1,1)$-form $\chi $ and $n\geq 2$, its derivative at $\theta $ is

\[ (\mathrm{d}\lambda _\mu )_\theta (\chi ) =\frac{\displaystyle \int _ X\chi \wedge \gamma _\theta ^{n-1}}{D_\mu (\gamma _\theta )}, \]

where $\gamma _\theta $ is any minimizing Gauduchon metric. In dimension one the derivative is $\int _ X\chi $.

Proof. Choose a smooth potential $u_*$ with $\theta +\mathrm{dd}^{\mathrm{c}}u_*>0$. Backgrounds $\theta _ t$ near $\theta $ then have the positive representatives $\theta _ t+\mathrm{dd}^{\mathrm{c}}u_*$. The implicit-function argument in the proof of proposition 11.40.4 applies to this family and shows that the solution and its constant depend smoothly on the parameters. Adding $u_*$ to these solution potentials and fixing the mean gives the normalized solutions for $\theta _ t$, so $\lambda _\mu (\theta _ t)=c_{\theta _ t}(\mu )^{1/n}$ is smooth.

To compute the derivative, write $\lambda =\lambda _\mu (\theta )$ and $\Omega =\Omega _\theta $, and write the minimizing metric as $\gamma _\theta =\mathrm{e}^ g\Omega $ with $g$ smooth and real, as theorem 12.2.3 permits. Along $\theta _ t=\theta +t\chi $ for small real $t$, put $\Omega _ t=\theta _ t+\mathrm{dd}^{\mathrm{c}}u_ t$ and $\lambda _ t=\lambda _\mu (\theta _ t)$, and let dots denote derivatives at $t=0$. Differentiating $\Omega _ t^ n=\lambda _ t^ n\mu $ gives

\[ (\chi +\mathrm{dd}^{\mathrm{c}}\dot u)\wedge \Omega ^{n-1} =\lambda ^{n-1}\dot\lambda \, \mu . \]

We multiply by $\mathrm{e}^{(n-1)g}$, so that $\Omega ^{n-1}$ becomes $\gamma _\theta ^{n-1}$, and integrate. The term with $\mathrm{dd}^{\mathrm{c}}\dot u$ vanishes by the Gauduchon condition, and $\int _ X\chi \wedge \gamma _\theta ^{n-1}=\lambda ^{n-1}\dot\lambda \int _ X\mathrm{e}^{(n-1)g}\mu $. On the other hand, $\gamma _\theta ^ n=\mathrm{e}^{ng}\lambda ^ n\mu $, so that

\[ D_\mu (\gamma _\theta ) =\lambda ^{n-1}\int _ X \mathrm{e}^{(n-1)g}\mu . \]

Dividing the two identities proves the derivative formula. In dimension one the formula follows directly from the definition. $\square$

Cite this page

Comments

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