Pome

15.21 Local density and boundary controlD0D15F32906B

D0D15F32906B

Theorem 15.18.1 requires a smooth positive density on the whole regular locus of a global strict barrier, and the global minimum of the difference between the solution and the barrier drives its proof. If the density is smooth and positive only on a smaller open set, smoothness can fail there. A bounded solution with minimal singularities can be discontinuous on a ball where its density is constant (theorem 15.21.12), and a locally Lipschitz plurisubharmonic function with constant density need not be $C^1$ (example 15.21.15). Boundary data give a different route to regularity: if the solution has smooth boundary values on small balls, uniqueness for the bounded Dirichlet problem identifies it with the smooth Dirichlet solution.

In this section $\theta $ is a smooth real $(1,1)$-form on $X$, and $\rho \in \mathrm{PSH}(X,\theta )$ satisfies $\theta +\mathrm{dd}^{\mathrm{c}}\rho \geq \varepsilon \omega $ for some $\varepsilon >0$; in particular $\theta $ is big. We fix an open set $\Omega \subset X$ on which $\rho $ is smooth and $\mathrm{d}\theta =0$.

15.21.1 Smooth boundary traces93E0127DCAFF

Let $B\Subset \Omega $ be a coordinate ball. A function $u$ on $B$ has boundary trace $g$, a function on $\partial B$, if $u(z)\to g(\xi )$ as $B\ni z\to \xi $, for every $\xi \in \partial B$.

Theorem 15.21.2.9826C9FA5510 Suppose that $u\in \mathrm{PSH}(X,\theta )$ has minimal ordinary singularities and

\[ \mathrm{MA}_\theta (u)=f\, \mathrm{d}V\quad \text{on }\Omega , \qquad f\in C^\infty (\Omega ),\quad f>0. \]

Assume that $\Omega $ is covered by coordinate Euclidean balls $B_ j\Subset \Omega $ such that $\theta $ has a smooth local potential near $\overline{B_ j}$ and $u$ has a smooth boundary trace on $\partial B_ j$. Then $u\in C^\infty (\Omega )$ and $\theta +\mathrm{dd}^{\mathrm{c}}u>0$ there.

Proof. Since $\rho -\sup _ X\rho \leq V_\theta $, minimal singularities give $u\geq \rho -C$ for a constant $C$. Thus $u$ is locally bounded on $\Omega $. Fix one of the balls, write it as $B=B_ R(z_0)$, let $g$ be the boundary trace of $u$ on $\partial B$, and choose a smooth function $h$ near $\overline B$ with $\mathrm{dd}^{\mathrm{c}}h=\theta $. By proposition 5.28.2, the bounded plurisubharmonic function $v=u+h$ has Bedford–Taylor measure $f\, \mathrm{d}V$ and smooth boundary trace $\varphi =g+h|_{\partial B}$.

For a smooth function $V$ and the Euclidean volume $\mathrm{d}\Lambda $,

80E9773F7495 15.21.3
\begin{equation} (\mathrm{dd}^{\mathrm{c}}V)^ n=\frac{n!}{\pi ^ n}\det (V_{j\bar k})\, \mathrm{d}\Lambda . \end{equation}

The smooth Dirichlet theorem recalled in subsection 15.18.7, with zero background and density $f\, \mathrm{d}V$, gives a smooth strictly plurisubharmonic $V$ on $\overline B$ with $(\mathrm{dd}^{\mathrm{c}}V)^ n=f\, \mathrm{d}V$ and boundary value $\varphi $; a smooth extension $\Phi $ of $\varphi $ gives the required strict subsolution $\Phi +A(|z-z_0|^2-R^2)$ for large $A$. In dimension one this is the Dirichlet problem for the Poisson equation.

The difference $v-V$ tends to zero at every boundary point. Compactness of $\partial B$ makes this convergence uniform near the boundary: otherwise a sequence approaching the boundary with difference bounded away from zero would have a subsequence approaching one fixed boundary point. Uniqueness for the bounded Dirichlet problem, that is, the Bedford–Taylor comparison principle, gives $v=V$. Therefore $u=V-h$ is smooth on $B$, and $\theta +\mathrm{dd}^{\mathrm{c}}u=\mathrm{dd}^{\mathrm{c}}V>0$ there. Since the balls cover $\Omega $, the theorem follows. $\square$

The proof uses minimal singularities only to obtain the local lower bound by $\rho $. Consequently the same conclusion holds for any solution already known to be locally bounded on $\Omega $ and satisfying the same trace condition. Closedness of $\theta $ on $\Omega $ provides the local potentials $h$ once the coordinate balls are small enough; the background may be nonclosed elsewhere.

Example 15.21.4.DF2D43B90C47 For every $n\geq 1$, the trace criterion applies to a bounded continuous solution whose global density belongs to $L^1$ but to no $L^ p$ with $p>1$, and which is not Hölder continuous at the points of $X\setminus \Omega $.

Let $E=\mathbb C/(\mathbb Z+\mathrm{i}\mathbb Z)$, $X=E^ n$, and

\[ \omega _ j=\frac{\mathrm{i}}{2\pi }\, \mathrm{d}z_ j\wedge \mathrm{d}\bar z_ j, \qquad \theta =\sum _{j=1}^ n\omega _ j, \qquad D=\{ 0\} \times E^{n-1}. \]

Take $\mathrm{d}V=\theta ^ n$ and $\rho =0$. Choose a nonnegative smooth cutoff $\chi $ supported in a coordinate disk $|z|<r_0<1$ of $E$ and equal to one near zero. For $\beta >0$, define

\[ \psi (z)=\chi (z)(-\log |z|^2)^{-\beta }\quad (z\ne 0), \qquad \psi (0)=0, \]

and extend by zero. On the smaller punctured disk put $t=|z|^2$ and $L=-\log t$. Differentiation gives

334EFF96259F 15.21.5
\begin{equation} \psi _{z\bar z}=\frac{\beta (\beta +1)}{tL^{\beta +2}}, \qquad t\frac{\mathrm{d}}{\mathrm{d}t}L^{-\beta } =\beta L^{-\beta -1}\longrightarrow 0. \end{equation}

The first expression is integrable and the second is the radial flux. Integration by parts on punctured disks, using also $\psi \to 0$, shows that the distributional Hessian has no atom at zero. On the cutoff annulus its negative part is bounded. Hence, for a sufficiently small fixed $\eta >0$, $u=\eta \psi (z_1)$ satisfies $\theta +\mathrm{dd}^{\mathrm{c}}u\geq \theta /2$.

Only one coordinate contributes to $\mathrm{dd}^{\mathrm{c}}u$. Locally approximate the bounded subharmonic potential in that coordinate from above and use theorem 5.27.1; the smooth diagonal determinant identity passes to the limit and yields

\[ \mathrm{MA}_\theta (u)=f\theta ^ n, \qquad f=1+\eta \psi _{z\bar z}(z_1). \]

Give $f$ any positive value on the null set $D$. It is smooth and positive on $\Omega =X\setminus D$, where $u$ is smooth and has smooth traces on all relatively compact coordinate balls. Near $D$ its singular term is comparable to $|z_1|^{-2}(-\log |z_1|^2)^{-\beta -2}$. Its radial $L^1$ integral converges, whereas

\[ \int _0^{r_0}s^{1-2p}(-\log s^2)^{-p(\beta +2)}\, \mathrm{d}s=+\infty \qquad (p>1). \]

Also $(-\log |z|^2)^{-\beta }/|z|^\alpha \to +\infty $ for every $\alpha >0$, so $u$ is not Hölder at $D$. Finally, boundedness gives minimal ordinary singularities, since every global $\theta $-plurisubharmonic competitor is bounded above.

15.21.6 A bounded discontinuous potentialAD0BD024C9A0

We construct a bounded plurisubharmonic function on a bidisk whose Monge–Ampère measure is a constant multiple of Lebesgue measure, but which is discontinuous. The construction places logarithmic poles with summable weights at points accumulating at the origin; the resulting weight $b$ vanishes at the poles but not at their limit. It is the local model for theorem 15.21.12.

Let $U=W=\{ |z|<1/4\} \subset \mathbb C$, with coordinates $z$ and $w$ on the two factors. Write $\mathrm{d}A$ for planar Lebesgue measure and set

\[ a_ k=\alpha _ k=2^{-k-4},\qquad A=\sum _{k\geq 1}\alpha _ k=\frac1{16},\qquad S=\{ 0\} \cup \{ a_ k:k\geq 1\} . \]

Define on $U$

6321AD313CED 15.21.7
\begin{equation} h(z)=\sum _{k\geq 1}\alpha _ k\log |z-a_ k|^2, \qquad b(z)=\mathrm{e}^{h(z)}, \end{equation}

with $b=0$ where $h=-\infty $.

Lemma 15.21.8.5E22269BEE55 The function

\[ a(z)=\frac1\pi \int _ U\log |z-\zeta |^2\, b(\zeta )^{-1}\, \mathrm{d}A(\zeta ) \]

is bounded and continuous on $U$. The bounded function

\[ P(z,w)=a(z)+b(z)|w|^2 \]

is plurisubharmonic on $U\times W$, is smooth off $S\times W$, and satisfies

69692418B3EE 15.21.9
\begin{equation} (\mathrm{dd}^{\mathrm{c}}P)^2=\frac2{\pi ^2}\, \mathrm{d}A(z)\, \mathrm{d}A(w). \end{equation}

It is discontinuous at every $(0,w_0)$ with $w_0\ne 0$.

Proof. The summands defining $h$ are negative on $U$, so their partial sums decrease. Since $h(0)=2\sum _ k\alpha _ k\log a_ k> -\infty $, the limit is subharmonic. On compact subsets of $U\setminus S$, the series and its derivatives converge uniformly, making $h$ harmonic there. Moreover $0\leq b\leq 1$, $b(a_ k)=0$, and $b(0)>0$.

Weighted arithmetic–geometric mean gives, almost everywhere,

331160A2E4B5 15.21.10
\begin{equation} b(z)^{-2}\leq \sum _{k\geq 1}\frac{\alpha _ k}{A}|z-a_ k|^{-4A}. \end{equation}

For the countable inequality, first use the first $N$ weights and assign the remaining weight to the factor $1$; then let $N$ increase. Since $4A=1/4<2$, the integrals of the terms $|z-a_ k|^{-4A}$ on $U$ are bounded uniformly in $k$. Thus $b^{-1}\in L^2(U)$.

The logarithmic kernel has uniformly bounded $L^2$ norm on the translates of $U$ that occur in the definition of $a$. Cauchy–Schwarz bounds $a$ uniformly. Choose a compactly supported $L^2(\mathbb C)$ kernel agreeing with $\log |z|^2$ on $U-U$. Translation continuity in $L^2$, paired with $b^{-1}\mathbf1_ U\in L^2$, proves continuity of $a$. The distribution identity $\partial _ z\partial _{\bar z}\log |z|^2=\pi \delta _0$ gives $a_{z\bar z}=b^{-1}$. In particular $a$ is subharmonic and, by local elliptic regularity, smooth off $S$.

The second summand of $P$ is $\exp (h(z)+\log |w|^2)$, a convex increasing composition of a plurisubharmonic function, extended by zero at its poles. Thus $P$ is bounded and plurisubharmonic. At smooth points the harmonicity of $h=\log b$ gives $bb_{z\bar z}=|b_ z|^2$, and hence

\[ \det (P_{j\bar k}) =ba_{z\bar z}+|w|^2\bigl (bb_{z\bar z}-|b_ z|^2\bigr )=1. \]

The set $S\times W$ is contained in the pole set of $h(z)+\log |z|$, a plurisubharmonic function not identically $-\infty $. It is also Lebesgue-null. Pluripolar nullity in theorem 5.27.1 and (15.21.3) prove (15.21.9) throughout the bidisk. Finally,

\[ P(a_ k,w_0)=a(a_ k)\longrightarrow a(0) <a(0)+b(0)|w_0|^2=P(0,w_0) \]

when $w_0\ne 0$. $\square$

15.21.11 Compact gluingAABABE424D5C

We now glue the local example into a compact torus by a regularized maximum with a quadratic function. The result is a bounded solution with minimal singularities whose density is constant on a ball, where the solution is discontinuous; in the transition region the density acquires singularities that destroy every $L^ p$ bound with $p>1$.

We work on the flat complex torus $X=(\mathbb C/(\mathbb Z+\mathrm{i}\mathbb Z))^2$, identifying a small coordinate ball with a ball in $U\times W$. For a smooth function $v$ on the ball write $H_ v=(v_{j\bar k})$ for its complex Hessian and $I$ for the two-dimensional identity matrix.

Theorem 15.21.12.5FC2412C22E9 There are a positive flat Kähler form $\theta $ on $X$, a bounded $u\in \mathrm{PSH}(X,\theta )$ with minimal ordinary singularities, and a density $F>0$ almost everywhere such that

\[ \mathrm{MA}_\theta (u)=F\theta ^2,\qquad F\in L^1(X,\theta ^2),\qquad F\notin L^ p(X,\theta ^2)\quad (p>1). \]

On an open coordinate ball $\Omega $, $F$ is a positive constant and $u$ is discontinuous. The constant function $\rho =0$ is a smooth strict barrier on $X$.

Proof. Set $R=1/8$ and choose $M$ so that $P_0=P+M$ satisfies $0\leq P_0\leq L$ on $B_ R\Subset U\times W$ for some finite $L$. Choose $C>0$ with

477A59B42B24 15.21.13
\begin{equation} \frac{3CR^2}{16}\geq 2,\qquad \frac{5CR^2}{16}>L+2, \end{equation}

and put $Q(Z)=C(|Z|^2-R^2/4)$ for $Z=(z,w)$. Let $\chi $ be the convolution of $|t|$ with an even nonnegative smooth probability kernel supported in $(-1,1)$. Then $\chi $ is smooth and convex, $|\chi '|\leq 1$, $\chi '(0)=0$, and $\chi (t)=|t|$ for $|t|\geq 1$. The function

\[ m(s,t)=\frac{s+t+\chi (s-t)}2,\qquad v=m(P_0,Q) \]

is bounded and plurisubharmonic, because $m$ is convex and nondecreasing in each variable. The collar inequalities imply

\[ v=P_0\quad \text{on }B_{R/4},\qquad v=Q\quad \text{when }3R/4\leq |Z|<R. \]

Indeed, $Q\leq -2\leq P_0-2$ on the first ball and $Q>L+2\geq P_0+2$ on the collar. Define

\[ \theta =\frac{\mathrm{i}C}{2\pi } (\mathrm{d}z\wedge \mathrm{d}\bar z+\mathrm{d}w\wedge \mathrm{d}\bar w), \qquad u=\begin{cases} v-Q& \text{on }B_ R,\\ 0& \text{outside }B_ R.\end{cases} \]

The collar identity makes $u$ identically zero near $\partial B_ R$ and gives a global bounded potential with $\theta +\mathrm{dd}^{\mathrm{c}}u=\mathrm{dd}^{\mathrm{c}}v$ on the ball and $\theta +\mathrm{dd}^{\mathrm{c}}u=\theta $ outside. Thus $u$ is admissible and has minimal ordinary singularities.

Outside the pluripolar set $(S\times W)\cap B_ R$, set $\lambda =(1+\chi '(P_0-Q))/2$. Differentiating the regularized maximum gives

BAE98470F15D 15.21.14
\begin{equation} H_ v=\lambda H_ P+(1-\lambda )CI +\frac{\chi ''(P_0-Q)}2 \partial (P_0-Q)\otimes \overline{\partial (P_0-Q)}. \end{equation}

Both $H_ P$ and $CI$ are positive definite there; the final term is positive semidefinite. Hence the classical density is positive. Bounded products charge no pluripolar sets, so there is no further measure on the exceptional set. Their mass on compact $X$ is finite by proposition 5.28.2. This proves $\mathrm{MA}_\theta (u)=F\theta ^2$ for an almost-everywhere positive $F\in L^1$. On $\Omega =B_{R/4}$, (15.21.9) gives the representative $F=C^{-2}$. There $u=P+M-Q$ is discontinuous at $(0,w_0)$ for $0<|w_0|<R/4$.

We prove failure of every global $L^ p$ directly in the transition region. For large $k$, set

\[ m_ k=a(a_ k)+M\in [0,L],\qquad w_ k=\bigl (R^2/4+m_ k/C-|a_ k|^2\bigr )^{1/2}, \qquad Z_ k=(a_ k,w_ k). \]

Then $w_ k>0$, $P_0(Z_ k)=Q(Z_ k)=m_ k$, and (15.21.13) gives $|Z_ k|^2<9R^2/16$. Near the isolated pole $a_ k$,

\[ h(z)=\alpha _ k\log |z-a_ k|^2+H_ k(z), \]

where $H_ k$ is smooth and harmonic. Consequently $b$ and $P_0$ are continuous near $Z_ k$. There is a product neighborhood $V_ k\Subset B_ R$ of $Z_ k$, with $|w|$ bounded below by a positive number, on which $1/4\leq \lambda \leq 3/4$.

At smooth points of $V_ k$, positivity and the two-dimensional determinant expansion give

\[ \begin{split} \det H_ v& \geq \det \bigl (\lambda H_ P+(1-\lambda )CI\bigr )\\ & \geq \lambda (1-\lambda )C\operatorname {tr}H_ P \geq c_ k b_{z\bar z}, \end{split} \]

where $c_ k>0$ may depend on $k$ and $C$. Here $P_{z\bar z}=b^{-1}+|w|^2b_{z\bar z}$. With $\zeta =z-a_ k$, shrink the first factor of $V_ k$ so that

\[ b_{z\bar z}=|\zeta |^{2\alpha _ k}\mathrm{e}^{H_ k} \left|\frac{\alpha _ k}{\zeta }+\partial _ zH_ k\right|^2 \geq c'_ k|\zeta |^{2\alpha _ k-2} \]

for some $c'_ k>0$. Since $F=C^{-2}\det H_ v$, this lower bound implies $F\notin L^ p$ whenever $p(1-\alpha _ k)\geq 1$: the $w$-disk has positive area and the radial integral in $z$ contains

\[ \int _0^{\delta _ k}r^{1-2p(1-\alpha _ k)}\, \mathrm{d}r=+\infty . \]

For each $p>1$ such a $k$ exists because $\alpha _ k\downarrow 0$. $\square$

The bounded-product capacity estimate in section 5.27 also holds for the discontinuous potential $P$. Thus a first-order capacity bound, even together with a constant positive density, cannot supply the missing boundary control.

Example 15.21.15.41D1398D77FC On the bidisk in $\mathbb C^2$ with coordinates $(z,w)$, the function

\[ H(z,w)=2|z|(1+|w|^2) \]

is locally Lipschitz, plurisubharmonic and not $C^1$ along $\{ z=0\} $. Plurisubharmonicity follows by writing its two summands as convex exponential compositions of $\log |z|$ and $\log |z|+\log |w|^2$. At $z\ne 0$,

\[ H_{z\bar z}=\frac{1+|w|^2}{2|z|},\qquad H_{w\bar w}=2|z|,\qquad |H_{z\bar w}|^2=|w|^2. \]

Hence $\det (H_{j\bar k})=1$, and pluripolar nullity gives $(\mathrm{dd}^{\mathrm{c}}H)^2=2\pi ^{-2}\, \mathrm{d}\Lambda $ on the entire bidisk.

The last example rules out a purely local implication from Hölder continuity and a smooth positive density to smoothness; being local, it says nothing about the global hypotheses of theorem 15.18.1. Nor does theorem 15.21.12 contradict that theorem: there the solution is not locally Hölder continuous on the ball $\Omega $, and its density is not smooth on the rest of $X$, which is the regular locus of the barrier $\rho =0$. The trace criterion of theorem 15.21.2 avoids both global hypotheses by identifying the bounded solution with a smooth Dirichlet solution on each ball.

Cite this page

Comments

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