Proof.
The assertion for $s=0$ follows by bounded convergence for volume. Induct on $s$, assuming it for $s-1$, and put $q=s-1$ and $\theta =\omega ^{n-s}$. For $j\geq k$ let $h=v_ k-v_ j\geq 0$. The support of $h$ lies in one fixed compact set $K_0\Subset U$. Smooth capacity domination and the induction hypothesis give, uniformly over all bounded local smooth mixed potentials,
021C38133493
8.11.7
\begin{equation} \int h\, \mathrm{dd}^{\mathrm{c}}g_1\cdots \mathrm{dd}^{\mathrm{c}}g_ q\omega ^{n-q} \longrightarrow 0\quad (k\to \infty ). \end{equation}
To see this, split at $h=\delta $. On $\{ h\leq \delta \} $ the mass bound gives $C\delta $. On the remaining set use $h\leq v_ k-v$, the capacity bound and the induction hypothesis. Let $k$ tend to infinity before $\delta \downarrow 0$. For $q=0$ this is the volume estimate. Constants are uniform in $j\geq k$ and all potentials with the fixed common bound.
Consider the real smooth energy
\[ E_ s(w)=\int _ U w(\mathrm{dd}^{\mathrm{c}}w)^ s\theta , \qquad w_ t=v_ j+t h,\quad 0\leq t\leq 1. \]
The energies $E_ s(v_ j)$ are bounded: on the fixed collar the integrand is fixed, and the remaining compact set has a uniform smooth mass bound. Since $h$ vanishes near the boundary, integration by parts and closedness of $\mathrm{dd}^{\mathrm{c}}w_ t$ give the exact identity
CA5C66E53CFA
8.11.8
\begin{equation} \begin{split} \frac{\mathrm d}{\mathrm dt}E_ s(w_ t) & =(s+1)\int h(\mathrm{dd}^{\mathrm{c}}w_ t)^ s\theta \\ & \quad +s\int h(\mathrm{dd}^{\mathrm{c}}w_ t)^ q \bigl (\mathrm dw_ t\wedge \mathrm d^{\mathrm c}\theta -\mathrm d^{\mathrm c}w_ t\wedge \mathrm d\theta +w_ t\mathrm{dd}^{\mathrm{c}}\theta \bigr ). \end{split} \end{equation}
Here
\[ \begin{aligned} \mathrm d\theta & =(n-s)\mathrm d\omega \wedge \omega ^{n-s-1},\\ \mathrm{dd}^{\mathrm{c}}\theta & =(n-s)\mathrm{dd}^{\mathrm{c}}\omega \wedge \omega ^{n-s-1}\\ & \quad +(n-s)(n-s-1)\mathrm d\omega \wedge \mathrm d^{\mathrm c}\omega \wedge \omega ^{n-s-2}, \end{aligned} \]
and the analogous $\mathrm d^{\mathrm c}$ formula holds. A zero scalar coefficient means that its term is absent. The two first-order terms in (8.11.8) tend uniformly to zero by 8.11.4, the uniform local gradient bound, and (8.11.7). The last term has the same property by the absolute-form estimate at degree $q<m$. More explicitly, the first torsion integral is bounded by
\[ C\left(\int h\, \mathrm dw_ t\wedge \mathrm d^{\mathrm c}w_ t \wedge (\mathrm{dd}^{\mathrm{c}}w_ t)^ q\omega ^{n-q-1}\right)^{1/2} \left(\int h(\mathrm{dd}^{\mathrm{c}}w_ t)^ q\omega ^{n-q}\right)^{1/2}. \]
The first factor is bounded since $h$ has a common bound and 8.11.1 applies; the second tends to zero by (8.11.7). Conjugation handles the other first-order term. Thus, with $e_ k\downarrow 0$ after taking a tail supremum,
\[ E_ s(v_ k)-E_ s(v_ j)=(s+1)\Delta _{kj}+r_{kj}, \quad \Delta _{kj}=\int _0^1\int h\Lambda _ s(w_ t)\, \mathrm dt\geq 0, \quad |r_{kj}|\leq e_ k. \]
It follows that $E_ s(v_ j)\leq E_ s(v_ k)+e_ k$ for $j\geq k$. Taking a subsequence $k$ approaching the liminf shows that the bounded energy sequence converges. Hence $\Delta _{kj}\to 0$ uniformly for $j\geq k$. Positivity of the binomial expansion $\Lambda _ s(w_ t)\geq (1-t)^ s \Lambda _ s(v_ j)$ then gives
A133A0104951
8.11.9
\begin{equation} \int h\Lambda _ s(v_ j)\longrightarrow 0 \quad \text{uniformly for }j\geq k. \end{equation}
It remains to replace each Hessian factor by a capacity competitor. For any smooth $-1\leq a\leq 0$ in the local cone, put
\[ I_\ell =\int h(\mathrm{dd}^{\mathrm{c}}v_ j)^\ell (\mathrm{dd}^{\mathrm{c}}a)^{s-\ell }\theta , \qquad \eta _\ell =(\mathrm{dd}^{\mathrm{c}}v_ j)^\ell (\mathrm{dd}^{\mathrm{c}}a)^{s-\ell -1}, \quad 0\leq \ell <s. \]
First-order Stokes, with every metric derivative retained, gives
\[ I_\ell =-\int \mathrm dh\wedge \mathrm d^{\mathrm c}a\wedge \eta _\ell \theta +\int h\, \mathrm d^{\mathrm c}a\wedge \eta _\ell \mathrm d\theta . \]
The second term is $o(1)$ uniformly in $a$ and $j\geq k$. Use finite polarization in 8.11.4; each subset sum has a uniform local gradient bound and its lower trace is controlled by (8.11.7).
The form $\eta _\ell \theta $ is positive of bidegree $(n-1,n-1)$: a complementary nonnegative rank-one form completes a polarized product of $s\leq m$ cone forms. Its associated Hermitian gradient pairing is nonnegative, so Cauchy–Schwarz and the uniform gradient bound for $a$ give
\[ I_\ell \leq C F_\ell (h)^{1/2}+o(1), \qquad F_\ell (h)=\int \mathrm dh\wedge \mathrm d^{\mathrm c}h\wedge \eta _\ell \theta . \]
Another exact first-order Stokes identity gives
\[ F_\ell (h)= I_{\ell +1} -\int h\mathrm{dd}^{\mathrm{c}}v_ k\wedge \eta _\ell \theta +\int h\, \mathrm d^{\mathrm c}h\wedge \eta _\ell \mathrm d\theta \leq I_{\ell +1}+o(1). \]
The dropped term is nonnegative. For the last term the same polarized torsion estimate applies: its first gradient factor is uniformly bounded because $h=v_ k-v_ j$, and its second factor tends to zero by (8.11.7); here $h^2\leq C h$ and the gradient energy of the difference is at most twice the sum of those of $v_ k$ and $v_ j$ for every positive single-power gradient pairing. Starting with (8.11.9) at $I_ s$, finite backward induction gives $I_0\to 0$ uniformly in $a$ and $j\geq k$.
Chebyshev's inequality now yields $c_ s\{ v_ k-v_ j>\delta \} \leq \delta ^{-1}\sup _ a I_0$. These sets increase to $\{ v_ k-v>\delta \} $ as $j\to \infty $. Continuity of $c_ s$ under increasing unions proves the induction.
For any one $0\leq s\leq m$, choose $k_ N$ so that the open sets $G_ N=\{ v_{k_ N}-v>2^{-N}\} $ satisfy $c_ s(G_ N)<\epsilon 2^{-N}$, $N\geq 1$. They are open because $v$ is upper semicontinuous and $v_{k_ N}$ is smooth. On $U\setminus \bigcup _ NG_ N$ these smooth functions converge uniformly to the pointwise limit $v$. Hence $v$ is continuous there. The union has capacity less than $\epsilon $. Boundary localization transfers this conclusion to $f$ on each smaller compact set. Open sets have the same outer capacity, which also proves the assertion for $c_ s^*$.
$\square$
Comments
Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.