Pome

15.3 Bounded equations in a semipositive big background9F0629AA9014

9F0629AA9014

In this section $\theta $ is again semipositive and big, and we solve the exponential and the normalized equations with $L^ p$ densities by bounded potentials. The basic tool is a bounded subsolution. A strict barrier, corrected by the solution of a Hermitian equation, is composed with a bounded convex function; the result is a bounded $\theta $-plurisubharmonic function whose Monge–Ampère measure dominates a prescribed density. The derivative of the convex function decays polynomially near the poles of the barrier, and we compensate for this decay by multiplying the auxiliary density by the reciprocal weight.

Fix a strict barrier $\rho \in \mathrm{PSH}(X,\theta )$ and $0<a\leq 1$ such that

\[ \theta +\mathrm{dd}^{\mathrm{c}}\rho \geq a\omega ,\qquad \rho \leq -1. \]

No property of $\rho $ other than these two inequalities will be used.

Lemma 15.3.1.D716F01D638B Fix $p>1$ and $L>0$. There is $b>0$ such that, for every $f\geq 0$ with $\| f\| _{L^ p(\mathrm{d}V)}\leq L$, one can find $s\in \mathrm{PSH}(X,\theta )$ satisfying

\[ -1\leq s\leq 0,\qquad \mathrm{MA}_\theta (s)\geq b f\, \mathrm{d}V. \]

The constant $b$ depends only on the fixed geometry and on $\rho ,a,p,L$.

Proof. Choose $1<q<p$. Since a fixed quasi-plurisubharmonic function belongs to every finite $L^ r$ space, Hölder's inequality gives a uniform $L^ q$ bound for

\[ g=(-\rho )^{2n}f. \]

By theorem 14.19.2 there are $c>0$ and a continuous $\omega $-plurisubharmonic $v$, normalized by $\sup _ Xv=-1$, such that

\[ \omega _ v^ n=c(1+g)\, \mathrm{d}V. \]

The construction in that theorem gives a uniform upper bound for $c$, because in its Gauduchon trace estimate the root integral of $1+g$ is bounded below by the root integral of $1$. The lower-density estimate of theorem 14.19.19 gives $c\| 1+g\| _ q\geq c_ q>0$, so $c$ is also bounded below. The potential estimate of the same theorem then gives $-C\leq v\leq -1$ with a uniform $C$.

Put $h=\rho +av\leq -1$. Then $\theta _ h\geq a\omega _ v$. On $(-\infty ,-1]$ define

\[ \chi (t)=-t^{-1}-1,\qquad \chi '(t)=t^{-2}\in (0,1],\qquad \chi ''(t)=-2t^{-3}>0, \]

and set $s=\chi (h)$, with value $-1$ at the poles of $h$. To justify both the positivity of $\theta +\mathrm{dd}^{\mathrm{c}}s$ and the computation of its Monge–Ampère measure, we first work with the bounded functions $h_ k=\max (h,-k)$, $k\geq 1$. The chain rule for bounded potentials and $\theta \geq 0$ give

\[ \theta +\mathrm{dd}^{\mathrm{c}}\chi (h_ k) =(1-\chi '(h_ k))\theta +\chi '(h_ k)\theta _{h_ k} +\chi ''(h_ k)\, \mathrm{d}h_ k\wedge \mathrm d^{\mathrm c}h_ k\geq 0. \]

These bounded potentials decrease to $s$, so $s\in \mathrm{PSH}(X,\theta )$ and $-1\leq s\leq 0$. On $\{ h>-k\} $, locality and positivity of mixed products give

\[ \mathrm{MA}_\theta (s) \geq a^ n|h|^{-2n}\omega _ v^ n \geq a^ nc|h|^{-2n}(-\rho )^{2n}f\, \mathrm{d}V. \]

Letting $k\to \infty $, we obtain this inequality on the set where $h$ is finite; its complement is pluripolar and has zero volume. Since $|h|\leq (-\rho )+aC\leq (1+aC)(-\rho )$, the assertion follows with any common lower bound for $a^ nc(1+aC)^{-2n}$. $\square$

The lemma gives a subsolution for each density in an $L^ p$-bounded family. It does not give a lower bound for the masses of all bounded $\theta $-plurisubharmonic functions.

15.3.2 The exponential equation0853ABB454FC

We approximate $\theta $ from above by the Hermitian forms $\theta +\varepsilon \omega $. The bounded subsolution provides a common lower barrier for the corresponding solutions, and exponential comparison shows that they decrease as $\varepsilon $ decreases. Their limit solves the equation in the original background.

Theorem 15.3.3.96A341EAF1D4 Let $0\leq f\in L^ p(\mathrm{d}V)$, $p>1$, be nonzero, and let $\lambda >0$. There is a unique bounded $v\in \mathrm{PSH}(X,\theta )$ such that

\[ \mathrm{MA}_\theta (v)=\mathrm{e}^{\lambda v}f\, \mathrm{d}V. \]

For fixed $\lambda $, these solutions are uniformly bounded when $\| f\| _ p$ is bounded above and $\int _ X f^{1/n}\, \mathrm{d}V$ is bounded away from zero.

Proof. For $0<\varepsilon \leq 1$, set $\theta _\varepsilon =\theta +\varepsilon \omega $. By the Hermitian existence theorem theorem 14.19.3, there is a continuous solution $v_\varepsilon $ of

\[ \mathrm{MA}_{\theta _\varepsilon }(v_\varepsilon ) =\mathrm{e}^{\lambda v_\varepsilon }f\, \mathrm{d}V. \]

Take $s,b$ from lemma 15.3.1, and put $w=s+\lambda ^{-1}\log b$. Positivity of bounded products gives

\[ \mathrm{MA}_{\theta _\varepsilon }(w) \geq \mathrm{MA}_\theta (s)\geq bf\, \mathrm{d}V \geq \mathrm{e}^{\lambda w}f\, \mathrm{d}V, \]

and bounded exponential comparison (corollary 14.19.12) yields $w\leq v_\varepsilon $. If $0<\varepsilon '<\varepsilon $, then $v_{\varepsilon '}$ is also $\theta _\varepsilon $-plurisubharmonic and

\[ \mathrm{MA}_{\theta _\varepsilon }(v_{\varepsilon '}) \geq \mathrm{e}^{\lambda v_{\varepsilon '}}f\, \mathrm{d}V, \]

so the same comparison gives $v_{\varepsilon '}\leq v_\varepsilon $. Hence $v_\varepsilon $ decreases to a bounded $\theta $-plurisubharmonic function $v$ with $w\leq v\leq v_1$.

To pass to the limit in the equation, we use the fixed Hermitian background $\Omega =\theta +\omega $ and the expansion

\[ (\theta _\varepsilon +\mathrm{dd}^{\mathrm{c}}v_\varepsilon )^ n =\sum _{k=0}^ n\binom nk (\Omega +\mathrm{dd}^{\mathrm{c}}v_\varepsilon )^ k \wedge \bigl (-(1-\varepsilon )\omega \bigr )^{n-k}. \]

Continuity of bounded products along monotone sequences applies to every term, and the smooth coefficients converge. Dominated convergence applies to the right-hand sides. This proves the equation. Uniqueness follows from bounded domination in the semipositive big background (theorem 16.28.56, proposition 16.28.55) by the usual exponential comparison argument.

For the uniform assertion, note first that the lower barrier $w$ is uniform by lemma 15.3.1. The upper barrier $v_1$ solves an equation in the fixed Hermitian background $\Omega $. The proof of lemma 14.19.26 bounds its supremum using only an upper bound for the density norm and a positive lower bound for the root integral: these two bounds control, respectively, the Hölder term and the denominator in (14.19.27), and they survive the smooth approximation used there. Thus $v_1$, and hence $v$, has a uniform upper bound. $\square$

Local continuity of this bounded solution will follow from theorem 15.6.1.

15.3.4 The normalized equation727A21D220CE

For the normalized equation the approximating potentials need not be monotone. Upper envelopes and lower envelopes of their tails provide monotone sequences on either side of an $L^1$ cluster limit. Comparison identifies the two limits, and hence the equation, while the constant is controlled by the Gauduchon trace and the bounded subsolution.

Theorem 15.3.5.A972C11565ED For every nonzero $0\leq f\in L^ p(\mathrm{d}V)$, $p>1$, there are a bounded $\varphi \in \mathrm{PSH}(X,\theta )$ and a uniquely determined $c>0$ such that

\[ \mathrm{MA}_\theta (\varphi )=cf\, \mathrm{d}V,\qquad \sup _ X\varphi =0. \]

The construction gives uniform bounds for $\| \varphi \| _\infty $, $c$ and $c^{-1}$ when $\| f\| _ p$ is bounded above and $\int _ Xf^{1/n}\, \mathrm{d}V$ is bounded away from zero.

Only the constant $c$ is asserted to be unique; the normalized potential need not be unique, as theorem 14.26.1 shows already for a Hermitian background.

Proof. Step 1: approximating equations and their constants. For integers $j\geq 1$, put $\omega _ j=\theta +2^{-j}\omega $, and choose continuous normalized solutions

\[ \mathrm{MA}_{\omega _ j}(\varphi _ j)=c_ jf\, \mathrm{d}V, \qquad \sup _ X\varphi _ j=0. \]

We first bound $c_ j$. Fix a Gauduchon metric $\gamma _0$; when $n=1$ any Hermitian form will do, and the calculation below reduces to Stokes' theorem. The smooth approximation in the proof of theorem 14.19.2, together with the Gauduchon trace, gives a common upper bound: the numerator $\int _ X\omega _ j\wedge \gamma _0^{n-1}$ is bounded above by $\int _ X(\theta +\omega )\wedge \gamma _0^{n-1}$, and the root integral of $f$ is positive and fixed. Here the estimate is applied to the smooth approximating equations before passing to their continuous limits.

For a lower bound, take $s,b$ from lemma 15.3.1. If $c_ j<b$, then

\[ \mathrm{MA}_{\omega _ j}(\varphi _ j) \leq (c_ j/b)\mathrm{MA}_{\omega _ j}(s+A) \]

for every constant $A$, and strict domination in the Hermitian background $\omega _ j$ (theorem 14.19.10) would give $\varphi _ j\geq s+A$ for every $A$, which is absurd. Thus $c_ j\geq b$. In fact the constants decrease with $j$: if $k>j$, positivity gives $\mathrm{MA}_{\omega _ j}(\varphi _ k)\geq c_ kf\, \mathrm{d}V$, and the same argument excludes $c_ j<c_ k$. Hence $c_ j\downarrow c>0$.

Step 2: a uniform bound for the potentials. For the Hermitian form $\Omega =\theta +\omega $, all $\varphi _ j$ are normalized $\Omega $-plurisubharmonic functions. Using uniform exponential integrability (lemma 5.26.2) and Hölder's inequality, choose $\gamma >0$ small and $1<q<p$ such that

\[ g_ j=\mathrm{e}^{-\gamma \varphi _ j}f \]

is bounded in $L^ q$. Since $\varphi _ j\leq 0$, we have $g_ j\geq f$. Theorem 15.3.3, applied with integrability exponent $q$ and exponential exponent $\gamma $, therefore gives uniformly bounded $w_ j\in \mathrm{PSH}(X,\theta )$ with

\[ \mathrm{MA}_\theta (w_ j)=\mathrm{e}^{\gamma w_ j}g_ j\, \mathrm{d}V. \]

Choose $C_0\geq \max \{ 1,\sup _ jc_ j\} $ and set $z_ j=w_ j-\gamma ^{-1}\log C_0$. Then

\[ \mathrm{MA}_{\omega _ j}(z_ j) \geq \mathrm{e}^{\gamma z_ j}c_ jg_ j\, \mathrm{d}V, \qquad \mathrm{MA}_{\omega _ j}(\varphi _ j) =\mathrm{e}^{\gamma \varphi _ j}c_ jg_ j\, \mathrm{d}V, \]

and exponential comparison gives $z_ j\leq \varphi _ j\leq 0$. Thus all $\varphi _ j$ take values in a common bounded interval.

Step 3: the upper envelopes. Extract a subsequence converging in $L^1$ and almost everywhere to the canonical $\theta $-plurisubharmonic representative $\varphi $. We keep the indices $j$; the coefficients of $\omega $ in $\omega _ j$ still decrease to zero. Define

\[ \Phi _ j^+=\left(\sup _{\ell \geq j}\varphi _\ell \right)^*, \qquad h_ j=\inf _{\ell \geq j}\varphi _\ell , \qquad p_ j=P_\theta (h_ j). \]

All these functions take values in a common bounded interval. The upper envelopes are $\omega _ j$-plurisubharmonic and decrease to $\varphi $, since the regularized upper envelope agrees almost everywhere with the $L^1$ limit; this also follows from the local Hartogs lemma applied to the bounded tails.

For $\ell \geq j$, positivity gives $\mathrm{MA}_{\omega _ j}(\varphi _\ell )\geq c_\ell f\, \mathrm{d}V\geq cf\, \mathrm{d}V$. By ordered contact (lemma 5.28.3), finite maxima of these functions satisfy the same lower bound. Continuity of bounded products along increasing sequences, followed by the decreasing limit in $j$ and the expansion in the smooth background, gives

9D2B0195702F 15.3.6
\begin{equation} \mathrm{MA}_{\omega _ j}(\Phi _ j^+)\geq cf\, \mathrm{d}V, \qquad \mathrm{MA}_\theta (\varphi )\geq cf\, \mathrm{d}V. \end{equation}

Step 4: the lower envelopes. For the reverse inequality we use lower envelopes, which carry a useful exponential weight. For $N\geq j$ put

\[ p_{j,N}=P_\theta \left(\min _{j\leq \ell \leq N}\varphi _\ell \right). \]

The obstacle is continuous and bounded, so the measure of $p_{j,N}$ is carried by its contact set. Partition this set into disjoint Borel pieces according to the first index $\ell $ at which $p_{j,N}=\varphi _\ell $. On such a piece, positivity and ordered contact in the background $\omega _\ell $ give

\[ \mathrm{MA}_\theta (p_{j,N}) \leq \mathrm{MA}_{\omega _\ell }(p_{j,N}) \leq c_\ell f\, \mathrm{d}V. \]

Moreover $p_{j,N}=\varphi _\ell \geq h_ j$ there, and $c_\ell \leq c_ j$. Summing over the pieces therefore yields the global inequality

\[ \mathrm{MA}_\theta (p_{j,N}) \leq \mathrm{e}^{p_{j,N}-h_ j}c_ jf\, \mathrm{d}V. \]

As $N\to \infty $, the envelopes decrease to $p_ j$. Indeed, their limit lies below each tail obstacle quasi-everywhere and is thus a competitor for $P_\theta (h_ j)$, while the reverse inequality is immediate. Continuity of bounded products along monotone sequences and dominated convergence give

\[ \mathrm{MA}_\theta (p_ j)\leq \mathrm{e}^{p_ j-h_ j}c_ jf\, \mathrm{d}V. \]

The functions $p_ j$ increase. Let $\varphi '=(\lim _ jp_ j)^*$. Since $h_ j\to \varphi $ almost everywhere, we have $\varphi '\leq \varphi $ for the canonical representatives. Passing to the increasing limit and using (15.3.6), we obtain

\[ \mathrm{MA}_\theta (\varphi ') \leq \mathrm{e}^{\varphi '-\varphi }cf\, \mathrm{d}V \leq \mathrm{e}^{\varphi '-\varphi }\mathrm{MA}_\theta (\varphi ). \]

Bounded exponential comparison forces $\varphi '\geq \varphi $. Thus $\varphi '=\varphi $, and the last inequality, together with (15.3.6), proves the equation. If necessary, subtracting $\sup _ X\varphi $ gives the required normalization.

Step 5: uniformity and uniqueness of $c$. All bounds used above depend only on the two stated density bounds and the fixed data, which proves the uniform assertion for the constructed solution. Finally, two different positive constants for the same $f$ would contradict strict domination after an arbitrary translation of one of the bounded solutions, exactly as in the lower bound for $c_ j$. This proves the uniqueness of $c$. $\square$

For $\lambda >0$, let $v_\lambda $ be the bounded solution of $\mathrm{MA}_\theta (v_\lambda )=\mathrm{e}^{\lambda v_\lambda }f\, \mathrm{d}V$, and let $(\varphi ,c)$ be a normalized solution given by theorem 15.3.5. The two exponential barriers used in proposition 14.19.34 give

\[ \| \lambda v_\lambda -\log c\| _\infty \leq \lambda \operatorname {osc}_ X\varphi \]

also in this semipositive background.

Cite this page

Comments

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