Pome

8.7 Continuous products, envelopes and domination5B2332FC78E2

5B2332FC78E2

Uniform admissible approximation allows us to define the mixed Hessian products. Their local bounds retain the derivatives of the Hermitian background; closedness of the background is not required.

Proposition 8.7.1.D2A0F255F625 For continuous admissible $u_1,\ldots ,u_ k$, $1\leq k\leq m$, the current

\[ (\omega +\mathrm{dd}^{\mathrm{c}}u_1)\wedge \cdots \wedge (\omega +\mathrm{dd}^{\mathrm{c}}u_ k) \wedge \omega ^{n-m} \]

is well defined as the limit of uniform smooth admissible approximations. It is positive, local, and continuous when its potentials converge uniformly. Its local masses are bounded in terms of uniform bounds on the potentials. In particular $H_ m(u)$ is a positive Radon measure.

Proof. 8.6.1, 8.6.2 provide the smooth approximations without using any continuous Hessian product or domination input. For a smooth admissible $u$ put $\sigma =\omega +\mathrm{dd}^{\mathrm{c}}u$. Since $\mathrm d\sigma =\mathrm d\omega $, the coefficients of $\mathrm{dd}^{\mathrm{c}}(\chi \sigma ^{r-1}\wedge \omega ^{n-r})$ are sums of smooth test factors paired with powers $\sigma ^ q$, $q\leq r-1$, where $\chi $ is a fixed cutoff. The absolute-form lemma and integration by parts give the Chern–Levine–Nirenberg bound by induction on $r\leq m$:

\[ \int _ K\sigma ^ r\wedge \omega ^{n-r} \leq C_{K,U,\omega }(1+\| u\| _{L^\infty (U)})^ r. \]

At the induction step write $\sigma ^ r=\omega \wedge \sigma ^{r-1} +\mathrm{dd}^{\mathrm{c}}u\wedge \sigma ^{r-1}$, integrate the second term against $\chi $, and move $\mathrm{dd}^{\mathrm{c}}$ to $\chi \sigma ^{r-1}\omega ^{n-r}$. The resulting powers have degrees $r-1,r-2,r-3$, with negative powers absent, and the induction applies on nested slightly larger compact sets. Mixed masses reduce to the single potential average, by positivity of the expansion of its power. Positive mixed currents have total variation controlled by these trace masses.

To prove independence, telescope the difference between two smooth mixed products whose potentials have a common uniform bound. Each term paired with a fixed smooth complementary test is

\[ \int \mathrm{dd}^{\mathrm{c}}(u_ a-v_ a)\wedge T\wedge \eta =\int (u_ a-v_ a)\mathrm{dd}^{\mathrm{c}}(T\wedge \eta ), \]

where $T$ is a product of the remaining admissible factors and $\omega ^{n-m}$. The identity $\mathrm d(\omega +\mathrm{dd}^{\mathrm{c}}u)=\mathrm d\omega $ and its $\mathrm d^{\mathrm c}$ analogue show that derivatives of $T$ introduce only fixed metric derivatives; every term has at most $k-1<m$ admissible factors. The absolute mixed-form bound and the mass estimate just proved bound the variation of $\mathrm{dd}^{\mathrm{c}}(T\wedge \eta )$ uniformly. Consequently the pairing difference is at most $C_\eta \max _ a\| u_ a-v_ a\| _\infty $. The same construction with an arbitrary complementary smooth test also defines the signed products of fewer than $m$ factors without the factor $\omega ^{n-m}$: their variation is bounded by the absolute-form lemma. These products justify the metric-derivative terms above. This proves that all approximations have the same limit and that the products are continuous in the uniform topology. Positivity passes from the smooth mixed products. Using a test supported in an open set on which two limiting potentials coincide proves locality by the same estimate on a smaller neighborhood of its support. $\square$

Write $H_ m(u)$ for the measure with every potential equal to $u$. Locality and uniform continuity of these products give comparison on a contact set.

Lemma 8.7.2.21F540C3F7D1 If $u\leq v$ are continuous admissible functions and the remaining $m-1$ factors are fixed continuous admissible functions, then on $\{ u=v\} $ the mixed measure with first factor $\omega +\mathrm{dd}^{\mathrm{c}}u$ is at most the one with first factor $\omega +\mathrm{dd}^{\mathrm{c}}v$. In particular $\mathbf1_{\{ u=v\} }H_ m(u)\leq \mathbf1_{\{ u=v\} }H_ m(v)$.

Proof. Write $M(w)$ for the mixed measure with first potential $w$. Positivity and locality give, for $t>0$,

\[ M(\max (v,u+t))\geq \mathbf1_{\{ v>u+t\} }M(v)+\mathbf1_{\{ u\geq v\} }M(u). \]

The first indicators increase to $\mathbf1_{\{ v>u\} }$ as $t\downarrow 0$, and the potentials on the left converge uniformly to $v$. Integrate against nonnegative continuous tests and pass to the limit. Subtract the first restricted measure to obtain the claimed contact inequality. Replacing one factor at a time proves the pure-power claim and the analogous contact inequalities for all mixed powers. $\square$

For a continuous real obstacle $h$, define $P(h)$ to be the supremum of continuous admissible functions bounded above by $h$. Constants make the family nonempty. The penalty construction also proves that this supremum is continuous and its Hessian measure lies on the contact set.

Lemma 8.7.3.C39ADEAC74E3 The function $P(h)$ is continuous and admissible, and $H_ m(P(h))$ is supported on $\{ P(h)=h\} $. Moreover $\| P(h_1)-P(h_2)\| _\infty \leq \| h_1-h_2\| _\infty $.

Proof. For smooth $h$, use the penalty solutions $w_ N$ in 8.6.2. Its maximum argument applies to every continuous admissible competitor $v\leq h$, with an error depending on $h$ and $N$, and not on $v$. Taking their supremum gives

\[ P(h)-E_ N-2M_ N/N\leq w_ N\leq P(h), \]

where the error tends to zero. Thus $P(h)$ is their uniform limit and is continuous and admissible. On every compact subset of $\{ P(h)<h\} $ the penalty densities $b e^{N(w_ N-h)}$ tend uniformly to zero. Product continuity implies the contact support assertion. Adding constants to competitors gives the stated Lipschitz bound. Approximate a continuous $h$ uniformly by smooth obstacles. Their envelopes converge uniformly to $P(h)$, proving continuity and admissibility. Any compact subset of $\{ P(h)<h\} $ also lies outside the contact sets of all sufficiently close smooth obstacles; product continuity gives the support assertion for $h$. $\square$

The obstacle also bounds discontinuous competitors. Their upper-test admissibility suffices for the penalty estimate, so allowing them does not change the continuous envelope of a smooth obstacle.

Proposition 8.7.4.26DD802A141E For every smooth real obstacle $h$, the envelope $P(h)$ equals the pointwise supremum of all bounded admissible functions below $h$.

Proof. The solutions $w_ N$ of 8.6.5 are continuous admissible competitors, so $w_ N\leq P(h)$. For each bounded admissible $u\leq h$, the same lemma gives

\[ u\leq w_ N+E_ N+2M_ N/N\leq P(h)+E_ N+2M_ N/N. \]

Let $N\to \infty $ for this fixed obstacle and fixed $b$. Thus $P(h)$ dominates every bounded competitor. Since $P(h)$ itself is a bounded competitor, it equals their supremum. $\square$

The trace test in (8.2.1) gives $L_\omega u\geq -n$, where $L_\omega u=n\mathrm{dd}^{\mathrm{c}}u\wedge \omega ^{n-1}/\omega ^ n$. We use the classical normalized Green estimate

5451B220B0B1 8.7.5
\begin{equation} \int _ X(\sup _ Xu-u)\omega ^ n\leq C_\omega \end{equation}

for continuous admissible $u$. The stationary probability of $L_\omega $ has a smooth positive density equivalent to $\omega ^ n$. The positive measure $L_\omega u+n$ has stationary mass $n$, so the Green-kernel estimate gives this bound. It is a linear elliptic input, and does not assert compactness for a nonlinear Hessian measure.

To use envelopes in domination, we must first exclude a potential with zero total Hessian mass.

Lemma 8.7.6.E7EA08EFFDE7 Every continuous admissible $u$ satisfies $\int _ X H_ m(u)>0$.

Proof. Suppose this integral is zero. For $s>0$ put $w=\max (u,\min _ Xu+s)$ and $U=\{ u<\min _ Xu+s\} $. The open set $U$ is nonempty, $0\leq w-u\leq s$, and $w=u$ outside it. Set $\sigma _ u=\omega +\mathrm{dd}^{\mathrm{c}}u$, $\sigma _ w=\omega +\mathrm{dd}^{\mathrm{c}}w$, and $\mu _ q=\sigma _ u^ q\wedge \sigma _ w^{m-q}\wedge \omega ^{n-m}$. The contact lemma gives $\mathbf1_{X\setminus U}\mu _ q\geq \mathbf1_{X\setminus U}\mu _ m$, for $0\leq q\leq m$. For $q<m$ the telescoping identity is

\[ \mu _ q-\mu _ m=\mathrm{dd}^{\mathrm{c}}(w-u)\wedge T_ q,\qquad T_ q=\sigma _ u^ q\wedge \sum _{r=0}^{m-q-1}\sigma _ w^ r\wedge \sigma _ u^{m-q-1-r} \wedge \omega ^{n-m}. \]

Integration by parts on $X$ therefore gives

\[ \int _ U(\mu _ q-\mu _ m)\leq \int _ X(\mu _ q-\mu _ m) =\int _ X(w-u)\mathrm{dd}^{\mathrm{c}}T_ q. \]

This identity follows first for smooth approximants and then by uniform product continuity. The derivatives of $T_ q$ contain only fixed metric derivatives and products of at most $m-1$ admissible factors, so the absolute-form lemma makes $\mathrm{dd}^{\mathrm{c}}T_ q$ a measure of uniformly controlled variation. Uniform convergence also justifies its continuous weight. On $U$, locality gives $\sigma _ w=\omega $ and thus

\[ |\mathrm{dd}^{\mathrm{c}}T_ q|\leq C\sum _{r=0}^{m-1}\mu _ r. \]

The constant depends only on the metric and $m,n$, and not on $u$ or $s$. Consequently

\[ \sum _{q=0}^ m\int _ U\mu _ q \leq (m+1)\int _ U H_ m(u)+Cms\sum _{r=0}^{m-1}\int _ U\mu _ r. \]

For sufficiently small $s$, absorption and $H_ m(u)=0$ force all these nonnegative integrals to vanish. But $\mu _0=\omega ^ n$ on the nonempty open set $U$, a contradiction. $\square$

The envelope of a stretched difference now converts vanishing of a restricted measure into an order relation. The linear bound (8.7.5) prevents these envelopes from tending to $-\infty $ on an open set.

Proposition 8.7.7.7E8FF9A60609 If continuous admissible $u,v$ satisfy $\mathbf1_{\{ u<v\} }H_ m(u)=0$, then $u\geq v$.

Proof. Replace $v$ by $\max (u,v)$, so $u\leq v$. For $b>1$ put $p_ b=P(bu-(b-1)v)$ and $z_ b=b^{-1}p_ b+(1-b^{-1})v\leq u$. On the contact set $C_ b=\{ p_ b=bu-(b-1)v\} $ one has $z_ b=u$; the contact inequality and polarized positivity give

\[ \mathbf1_{C_ b}b^{-m}H_ m(p_ b) \leq \mathbf1_{C_ b}H_ m(z_ b) \leq \mathbf1_{C_ b}H_ m(u). \]

The envelope measure is supported on $C_ b$ and has positive mass. It vanishes on $C_ b\cap \{ u<v\} $ by the hypothesis, so some point of $C_ b\cap \{ u=v\} $ exists. Hence $\sup _ Xp_ b\geq \inf _ Xv$, while $p_ b\leq u$. The estimate (8.7.5) is consequently uniform for $p_ b$. On the open set $E_ a=\{ u<v-a\} $, for $a>0$, one has $p_ b\leq v-ba$, and therefore

\[ C_\omega \geq \int _ X(\sup _ Xp_ b-p_ b)\omega ^ n \geq (ba-\operatorname {osc}_ Xv)\int _{E_ a}\omega ^ n. \]

Letting $b\to \infty $ gives zero volume for $E_ a$. Since it is open it is empty, and $a\downarrow 0$ gives $u\geq v$. $\square$

Proposition 8.7.8.103D0E7E46B0 If continuous admissible $u,v$ satisfy

\[ \mathbf1_{\{ u<v\} }H_ m(u)\leq c\mathbf1_{\{ u<v\} }H_ m(v),\qquad 0\leq c<1, \]

then $u\geq v$.

Proof. Choose $b>1$ so that $(1-b^{-1})^ m>c$, and put $p=P(bu-(b-1)v)$ and $z=b^{-1}p+(1-b^{-1})v\leq u$. On $C=\{ p=bu-(b-1)v\} $ the contact comparison gives

\[ \mathbf1_ C\bigl (b^{-m}H_ m(p)+(1-b^{-1})^ mH_ m(v)\bigr ) \leq \mathbf1_ C H_ m(u). \]

Restricting to $\{ u<v\} $ and using the hypothesis shows that $H_ m(p)$ vanishes on $C\cap \{ u<v\} $. Its support lies in $C$, where $p<v$ is equivalent to $u<v$. Thus $H_ m(p)$ vanishes on $\{ p<v\} $. The preceding domination proposition gives $p\geq v$. Since $p\leq bu-(b-1)v$, it follows that $u\geq v$. $\square$

Strict domination yields the comparison needed for the proper equations.

Lemma 8.7.9.F28CED8C4FAF For continuous admissible $u,v$ and $a>0$, if

\[ e^{-av}H_ m(v)\geq e^{-au}H_ m(u), \]

then $u\geq v$.

Proof. Fix $\delta >0$. On $\{ u<v-\delta \} $, the assumed inequality gives

\[ H_ m(u)\leq e^{a(u-v)}H_ m(v) \leq e^{-a\delta }H_ m(v-\delta ). \]

The constant $e^{-a\delta }$ is strictly smaller than one, so 8.7.8 applied to $u$ and $v-\delta $ gives $u\geq v-\delta $. Letting $\delta \downarrow 0$ proves the assertion. $\square$

Cite this page

Comments

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