Pome

15.33 Power capacity bounds and threshold equations119B0BE69A7D

119B0BE69A7D

In this section, $X$ is a connected compact complex manifold of dimension $n\geq 3$, with a Hermitian form $h$ and a smooth positive probability volume $\mathrm dV$. The root-tail condition of proposition 15.32.9 controls the scalar threshold. A power bound for one density's measure supplies the missing control of its approximating potentials: the capacity estimates give a common bound, and stability against the solution measure gives uniform convergence.

Theorem 15.33.1.35816840F731 Let $0\leq f\in L^1(\mathrm dV)$ have positive integral. Suppose that $A,\alpha >0$ satisfy

7DAF690D3482 15.33.2
\begin{equation} \int _ Ef\, \mathrm dV\leq A\operatorname {Cap}_ h(E)^{1+\alpha } \qquad (E\subset X\text{ Borel}). \end{equation}

There is $C>0$, depending only on $(X,h),A,\alpha $, such that every measurable $0\leq g\leq f$ with positive integral satisfies

E5CF15D0251A 15.33.3
\begin{equation} b_ h(g)\geq C\left(\int _ Xg\, \mathrm dV\right)^{-\alpha /(1+\alpha )}, \qquad \rho _ h(g^{1/n})\leq C^{-1/n} \left(\int _ Xg\, \mathrm dV\right)^{\alpha /(n(1+\alpha ))}. \end{equation}

In particular, $f^{1/n}\in \mathcal D_ h^0$. There is $\phi \in \mathrm{PSH}(X,h)\cap C(X)$, with $\sup _ X\phi =0$, such that

1A3F6C7B811B 15.33.4
\begin{equation} \mathrm{MA}_ h(\phi )=b_ h(f)f\, \mathrm dV. \end{equation}

The coefficient $b_ h(f)$ and the mass of this measure are finite and strictly positive.

Proof. Step 1: the coefficient and root tails. Put $g_ j=\min (g,j)$ and let $\phi _ j$ be normalized continuous solutions of

\[ \mathrm{MA}_ h(\phi _ j)=c_ jg_ j\, \mathrm dV, \qquad c_ j=c_ h(g_ j),\qquad \sup _ X\phi _ j=0, \]

given by theorem 14.19.2. The definition (15.28.1) gives $c_ j\downarrow b_ h(g)$. Their measures satisfy a power capacity bound with constant $c_ jA$.

Take competitor zero, $H=1$ and $\varepsilon =1/2$ in lemma 14.19.16, lemma 14.19.14. Choose $D>0$ fixed with $2D<\delta _{1/2}$ and put $U_ D=\{ \phi _ j<\inf _ X\phi _ j+D\} $. These estimates give

\[ D\leq C_\alpha (C_1c_ jA)^{1/n} \operatorname {Cap}_ h(U_ D)^{\alpha /n}, \qquad D^ n\operatorname {Cap}_ h(U_ D) \leq C_1c_ j\int _ Xg_ j\, \mathrm dV. \]

Combining them yields

\[ c_ j^{1+\alpha }\left(\int _ Xg_ j\, \mathrm dV\right)^\alpha \geq c>0. \]

Taking the limit proves the lower bound for $b_ h(g)$. The identity (15.29.2), with its full family of Gauduchon tests, gives the root estimate. Apply it to $g=f\mathbf1_{\{ f>K^ n\} }$ whenever this density is nonzero. Its mass tends to zero; the zero density has root norm zero. Thus the tails of $f^{1/n}$ tend to zero in both $L^ n$ and $\rho _ h$, and proposition 15.32.9 gives $f^{1/n}\in \mathcal D_ h^0$.

Step 2: uniform convergence of a subsequence. Now use $f_ j=\min (f,j)$, $c_ j=c_ h(f_ j)$ and normalized continuous solutions $\phi _ j$. Step 1 gives $b_ h(f)>0$, while $c_ j\leq c_1<\infty $. Hence

\[ \mathrm{MA}_ h(\phi _ j)(E) \leq Ac_1\operatorname {Cap}_ h(E)^{1+\alpha }. \]

The first part of the proof of proposition 14.19.23, using the capacity tail estimate, gives one $M\geq 1$ with $-M\leq \phi _ j\leq 0$ for all $j$.

By theorem 5.25.2, extract a subsequence converging in $L^1(\mathrm dV)$ and almost everywhere. Since $f\, \mathrm dV$ is a fixed finite absolutely continuous measure and the potentials have a common bound, dominated convergence gives

\[ \int _ X|\phi _ j-\phi _ k|\, \mathrm{MA}_ h(\phi _ j) \leq c_1\int _ X|\phi _ j-\phi _ k|f\, \mathrm dV\longrightarrow 0 \]

along this subsequence as both indices tend to infinity. Apply proposition 14.19.23 in both orders, with $H=M$. The subsequence is uniformly Cauchy. Its limit $\phi $ is continuous, $h$-plurisubharmonic by distributional positivity, and has supremum zero.

Step 3: passage of the products. On a coordinate ball choose a fixed smooth $q$ with $\mathrm{dd}^{\mathrm{c}}q\geq h$, and put $S_ j=\mathrm{dd}^{\mathrm{c}}(\phi _ j+q)$ and $S=\mathrm{dd}^{\mathrm{c}}(\phi +q)$. For every $\ell \geq 1$, the calculus of bounded products gives

\[ S_ j^\ell -S^\ell =\mathrm{dd}^{\mathrm{c}}\left[(\phi _ j-\phi ) \sum _{a=0}^{\ell -1}S_ j^ a\wedge S^{\ell -1-a}\right]. \]

The local Chern–Levine–Nirenberg estimate bounds its pairing with a compactly supported smooth test form by $C\| \phi _ j-\phi \| _\infty $. Expand the Hermitian products as in proposition 5.28.2:

\[ \mathrm{MA}_ h(\phi _ j) =\sum _{\ell =0}^ n\binom n\ell S_ j^\ell \wedge (h-\mathrm{dd}^{\mathrm{c}}q)^{n-\ell }. \]

Every term converges, with its smooth correction factor, to the corresponding term for $\phi $. On the other hand, $c_ jf_ j\to b_ h(f)f$ in $L^1(\mathrm dV)$. The two limits prove (15.33.4); its prescribed mass is $b_ h(f)\int _ Xf\, \mathrm dV$. $\square$

The capacity assumption concerns the fixed datum and implies the root-tail condition. Whether root tails alone suffice for the exact equation when $v_-(h)=0$ remains part of 24.2.2; whether every positive-threshold root has these tails is the question following proposition 15.32.9.

15.33.5 A logarithmic density with critical coefficient oneB8C5C5469618

The power capacity condition can hold for a density outside every $L^ p$ class with $p>1$. A bounded continuous potential with logarithmic modulus at one point gives such a datum on each fixed Hermitian background.

Theorem 15.33.6.B37DD22EFF75 In the setting of theorem 15.33.1, fix $\alpha >0$. There is $u\in \mathrm{PSH}(X,h)\cap C(X)$ with $h+\mathrm{dd}^{\mathrm{c}}u\geq h/2$ whose measure $\mathrm{MA}_ h(u)=f\, \mathrm dV$ satisfies

\[ f\in L^1(\mathrm dV),\qquad f\notin L^ p(\mathrm dV)\quad \text{for every }p>1. \]

For some $A>0$ it satisfies (15.33.2), and $b_ h(f)=1$. For all sufficiently large $K$,

E4AFA559F42C 15.33.7
\begin{equation} \rho _ h\bigl (f^{1/n}\mathbf1_{\{ f>K\} }\bigr ) \leq C(\log K)^{-\alpha }. \end{equation}

Proof. Choose a holomorphic ball with coordinates centred at $x_0$ and $R=|z|<1$, and a smooth cutoff $0\leq \chi \leq 1$ supported in it and equal to one on a smaller ball. Put

\[ q(s)=(1-s)^{-\alpha },\qquad s=\log R^2, \qquad \psi =\chi q(\log R^2)+1-\chi . \]

Set $\psi (x_0)=0$ and extend $\psi =1$ outside the ball. In the smaller ball, $q'=\alpha (1-s)^{-\alpha -1}>0$ and $q''=\alpha (\alpha +1)(1-s)^{-\alpha -2}>0$, so $\psi $ is plurisubharmonic there, including its continuous extension at $x_0$. The transition is smooth and $\mathrm{dd}^{\mathrm{c}}\psi \geq -K_0h$ globally for a fixed $K_0$. For sufficiently small $\delta >0$, the normalized potential $u=\delta \psi -\sup _ X\delta \psi $ satisfies $h+\mathrm{dd}^{\mathrm{c}}u\geq h/2$.

The bounded product gives zero mass to $x_0$. Indeed, after a smooth local correction, its expansion is dominated in total variation by a constant times the top product of a bounded plurisubharmonic function plus a strictly plurisubharmonic quadratic function. Rescaling this sum into $[0,1]$ bounds that top product by local capacity, and a pluripolar point has capacity zero. Off $x_0$ the potential is smooth. Thus the measure is absolutely continuous; write it as $f\, \mathrm dV$. The strict lower bound implies $f\geq c>0$ almost everywhere.

In the smaller ball the tangential and radial eigenvalues of the Hessian of $q(\log R^2)$ are $q'/R^2$ and $q''/R^2$. In the operator convention of the book this gives

F1C776CF9796 15.33.8
\begin{equation} (\mathrm{dd}^{\mathrm{c}}u)^ n =\frac{\delta ^ nn!\alpha ^ n(\alpha +1)}{\pi ^ n} R^{-2n}(1-\log R^2)^{-n(\alpha +1)-1}\, \mathrm d\lambda , \end{equation}

where $\mathrm d\lambda $ is Euclidean volume.

Integration over the ball gives

\[ \int _{B_ r}(\mathrm{dd}^{\mathrm{c}}u)^ n =\delta ^ n\alpha ^ n(1-\log r^2)^{-n(\alpha +1)}. \]

For $k<n$, the degree-$k$ mixed terms with the smooth positive form $h$ have density bounded by $CR^{-2k}(1-\log R^2)^{-k(\alpha +1)}$ and mass $O(r^2)$ on $B_ r$. The expansion has positive terms in this ball. Hence $f\in L^1$. For $p>1$, the pure top term gives a positive multiple of the divergent integral

\[ \int _0^ r R^{2n(1-p)-1} (1-\log R^2)^{-p(n(\alpha +1)+1)}\, \mathrm dR \]

as a lower bound for $\int _{B_ r}f^ p\, \mathrm dV$, since smooth volume densities are comparable. Thus $f\notin L^ p$. For all sufficiently small $r$,

F69DF10A862E 15.33.9
\begin{equation} \int _{B_ r}f\, \mathrm dV \leq C\bigl [(1-\log r^2)^{-n(\alpha +1)}+r^2\bigr ], \qquad \mathop{\rm ess\, sup}_{X\setminus B_ r}f\leq Cr^{-2n}. \end{equation}

By (5.29.3),

\[ \mathrm dV(E)\leq C\exp \bigl (-a\operatorname {Cap}_ h(E)^{-1/n}\bigr ). \]

Let $s=\operatorname {Cap}_ h(E)>0$ be small and choose $r=\exp [-as^{-1/n}/(4n)]$. Splitting $E$ inside and outside $B_ r$ and applying (15.33.9), we obtain

\[ \int _ Ef\, \mathrm dV \leq Cs^{1+\alpha } +C\exp [-as^{-1/n}/(2n)] +C\exp [-as^{-1/n}/2] \leq C's^{1+\alpha }. \]

For larger $s$ use the finite total mass; for $s=0$ the volume estimate gives $\mathrm dV(E)=0$. This proves (15.33.2) for every Borel set.

By theorem 15.33.1, there is a normalized continuous $\phi $ with $\mathrm{MA}_ h(\phi )=b_ h(f)f\, \mathrm dV$. The explicit potential $u$ is a subsolution with coefficient one, so theorem 15.28.4 gives $b_ h(f)\geq 1$. Suppose $b_ h(f)>1$, and choose $0<\varepsilon <1$ with $(1-\varepsilon )^ nb_ h(f)>1$. Positive multilinearity gives

\[ \mathrm{MA}_ h((1-\varepsilon )\phi ) \geq (1-\varepsilon )^ nb_ h(f)f\, \mathrm dV. \]

Apply theorem 14.19.6 to $u,\phi $ on

\[ U_ s=\{ u<(1-\varepsilon )\phi + \inf _ X(u-(1-\varepsilon )\phi )+s\} . \]

For sufficiently small $s>0$ its reverse factor $Q_\varepsilon (s)$ is defined, tends to one as $s\downarrow 0$, and satisfies

\[ (1-\varepsilon )^ nb_ h(f)\int _{U_ s}f\, \mathrm dV \leq Q_\varepsilon (s)\int _{U_ s}f\, \mathrm dV. \]

Choose $s$ so that $Q_\varepsilon (s)<(1-\varepsilon )^ nb_ h(f)$. The set $U_ s$ is open and nonempty, and $f\geq c>0$ almost everywhere, so its $f\, \mathrm dV$ measure is positive. This contradiction proves $b_ h(f)=1$.

Finally take $r_ K=K^{-1/(4n)}$. The outside bound in (15.33.9) gives $f\leq CK^{1/2}<K$ almost everywhere on $X\setminus B_{r_ K}$ for large $K$. Therefore

\[ \int _{\{ f>K\} }f\, \mathrm dV \leq C(\log K)^{-n(\alpha +1)}+CK^{-1/(2n)} \leq C'(\log K)^{-n(\alpha +1)}. \]

Apply (15.33.3) to $g=f\mathbf1_{\{ f>K\} }$ to get (15.33.7). $\square$

The potential has logarithmic modulus at $x_0$ and is not Hölder there. Its root-tail estimate follows from the capacity bound despite the failure of higher $L^ p$ integrability. The construction applies to each fixed background, including one with $v_-(h)=0$.

Cite this page

Comments

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