Pome

5.11 Extension across closed exceptional setsE197EAFECEC0

E197EAFECEC0

Let $S$ be a relatively closed subset of a domain $\Omega \subset \mathbb {C}^ n$. We give two conditions under which plurisubharmonic functions on $\Omega \setminus S$ extend across $S$. In the first, $S$ is locally contained in the zero set of a holomorphic function, and the function is assumed to be locally bounded above near $S$; a logarithmic barrier built from the holomorphic function then gives the extension. In the second, $S$ is small in the sense of Hausdorff measure, and the local upper bound is not assumed but follows from the smallness of $S$ by an argument with complex lines.

Theorem 5.11.1.C910F1109424 Let $\Omega \subset \mathbb {C}^ n$ be a domain, $n\ge 1$, and let $S\subset \Omega $ be relatively closed. Assume that every point of $S$ has a connected neighborhood $W\subset \Omega $ with a holomorphic function $f\not\equiv 0$ on $W$ such that $S\cap W\subset \{ f=0\} $. Let $u$ be a plurisubharmonic function on $\Omega \setminus S$ that is locally bounded above near $S$, that is, every point of $S$ has a neighborhood $W$ such that $u$ is bounded above on $W\setminus S$. Then $u$ has a unique plurisubharmonic extension $\widetilde u$ to $\Omega $, and its value at $x\in S$ is

\[ \widetilde u(x)=\varlimsup _{\Omega \setminus S\ni y\to x}u(y). \]

Proof. Define $\widetilde u$ by $u$ on $\Omega \setminus S$ and by the displayed formula on $S$. Since plurisubharmonicity is a local property, it suffices to show that every point $x_0\in S$ has a neighborhood on which $\widetilde u$ is plurisubharmonic; uniqueness is proved at the end. By hypothesis there are a connected neighborhood $W_0$ of $x_0$ with a holomorphic function $f\not\equiv 0$ on $W_0$ such that $S\cap W_0\subset \{ f=0\} $, and a neighborhood $W_1$ of $x_0$ such that $u$ is bounded above on $W_1\setminus S$. Choose a ball $W$ about $x_0$ with $\overline W\subset W_0\cap W_1$. By the identity theorem, $f$ does not vanish identically on $W$, and $u$ is bounded above on $W\setminus S$. Multiplying $f$ by a nonzero constant, we may assume that $|f|\leq 1$, that is, $\log |f|^2\le 0$, on $W$. For $\varepsilon >0$ define $u_\varepsilon $ on $W$ by

\[ u_\varepsilon =u+\varepsilon \log |f|^2\quad \text{on }W\setminus S,\qquad u_\varepsilon =-\infty \quad \text{on }S\cap W. \]

Step 1: $u_\varepsilon \in \mathrm{PSH}(W)$. At a point $x\in S\cap W$ we have $f(x)=0$, so $\log |f(y)|^2\to -\infty $ as $y\to x$, while $u$ is bounded above near $x$; hence $u_\varepsilon (y)\to -\infty $ as $y\to x$, and $u_\varepsilon $ is upper semicontinuous at $x$. On $W\setminus S$, $u_\varepsilon $ is upper semicontinuous as a sum of upper semicontinuous functions.

Now let $\Lambda $ be a connected component of $L\cap W$ for an affine complex line $L$. If $f$ vanishes identically on $\Lambda $, then $u_\varepsilon \equiv -\infty $ on $\Lambda $. Otherwise the zeros of $f$ on the connected set $\Lambda $ are isolated, by the identity theorem in one variable. Since $S$ is relatively closed, $A=S\cap \Lambda $ is therefore a closed discrete subset of $\Lambda $, hence countable, because each of the countably many compact sets exhausting $\Lambda $ contains only finitely many of its points. It is polar, because points are polar in the plane and countable unions of polar sets are polar by proposition 4.1.11.

Each connected component of $\Lambda \setminus A$ is a connected subset of $L\cap (\Omega \setminus S)$ and therefore lies in a connected component of this line section, on which the restriction of $u$ is subharmonic or identically $-\infty $. Hence the restriction of $u$ to $\Lambda \setminus A$ is subharmonic, the constant $-\infty $ being subharmonic. The function $\log |f|^2$ is subharmonic on $\Lambda $ by example 3.4.15. Thus $u_\varepsilon $ is subharmonic on $\Lambda \setminus A$, and it is bounded above there by $\sup _{W\setminus S}u$, because $\log |f|^2\leq 0$. By theorem 3.4.27, it has a subharmonic extension to $\Lambda $ whose value at $x\in A$ is $\varlimsup _{\Lambda \setminus A\ni y\to x}u_\varepsilon (y)=-\infty $, which is the value of $u_\varepsilon $ at $x$. Hence the restriction of $u_\varepsilon $ to $\Lambda $ is subharmonic. Finally, $u_\varepsilon $ is not identically $-\infty $, because $u$ is finite almost everywhere on $W\setminus S$ and $\{ f=0\} $ is Lebesgue null by lemma 5.5.1. Thus $u_\varepsilon \in \mathrm{PSH}(W)$.

Step 2: the extension. As $\varepsilon \downarrow 0$ the functions $u_\varepsilon $ increase, because $\log |f|^2\leq 0$, and they are bounded above on $W$ by $\sup _{W\setminus S}u$. Let $F=\sup _{\varepsilon >0}u_\varepsilon $. By theorem 5.10.2, $F^*\in \mathrm{PSH}(W)$ and $F^*=F$ almost everywhere. Since $F=u$ outside the null set $\{ f=0\} $, we get $F^*=u$ almost everywhere on $W\setminus S$; both functions are locally integrable and subharmonic there, so $F^*=u$ everywhere on $W\setminus S$ by theorem 3.4.12.

It remains to show that $F^*=\widetilde u$ on $S\cap W$, which is Lebesgue null as a subset of $\{ f=0\} $. Let $x\in S\cap W$. Since $F^*=u$ on $W\setminus S$, the upper semicontinuity of $F^*$ gives $\widetilde u(x)=\varlimsup _{W\setminus S\ni y\to x}u(y)\leq F^*(x)$. Conversely, for small $r>0$ the ball submean inequality for $F^*$ gives $F^*(x)\leq |B(x,r)|^{-1}\int _{B(x,r)\setminus S}F^*\, \mathrm dV\leq \sup _{B(x,r)\setminus S}u$, because removing the null set $S$ does not change the integral, and letting $r\to 0$ gives $F^*(x)\leq \widetilde u(x)$. Hence $\widetilde u=F^*$ on $W$, and $\widetilde u$ is plurisubharmonic on $W$.

Uniqueness. The set $S$ is Lebesgue null, since it is covered by countably many neighborhoods $W$ as in the hypothesis and $S\cap W\subset \{ f=0\} $ is null. Two plurisubharmonic extensions of $u$ therefore agree almost everywhere, and hence everywhere by theorem 5.9.1. $\square$

For the second criterion we measure the size of $S$ by Hausdorff measure. Let $q>0$. A subset $E$ of a metric space is $\mathcal H^ q$-null if for all $\delta ,\eta >0$ it can be covered by countably many balls of radii $r_ l<\delta $ with $\sum _ l r_ l^ q<\eta $. This means that the $q$-dimensional Hausdorff measure of $E$, which is defined by coverings with sets of small diameter, vanishes, whatever its normalization: a set of diameter $\ell $ lies in the ball of radius $\ell '$ about any of its points for every $\ell '>\ell $, and a ball of radius $r$ has diameter at most $2r$. Every subset of an $\mathcal H^ q$-null set is $\mathcal H^ q$-null, and so is every countable union of $\mathcal H^ q$-null sets $E_1,E_2,\ldots $: cover $E_ l$ by balls of radii less than $\delta $ with sum of the $q$-th powers of the radii less than $2^{-l}\eta $.

We use two further elementary properties. First, if $E$ is $\mathcal H^ q$-null and $\Phi $ is a Lipschitz map with constant $L>0$ from $E$ to a metric space, then $\Phi (E)$ is $\mathcal H^ q$-null. To see this, discard the covering balls that do not meet $E$, and replace each remaining ball $B(c,r)$ by $B(c',2r)$ with $c'\in E\cap B(c,r)$; then $\Phi (E\cap B(c',2r))\subset B(\Phi (c'),2Lr)$, so $\Phi (E)$ is covered by balls of radii $2Lr_ l<2L\delta $ with $\sum _ l(2Lr_ l)^ q<(2L)^ q\eta $. Second, if $q\leq b$, every $\mathcal H^ q$-null subset of $\mathbb R^ b$ is Lebesgue null, because the total volume of the covering balls is at most a constant times $\sum _ lr_ l^ b\leq \delta ^{b-q}\sum _ l r_ l^ q$.

For $n\geq 2$, let $\sigma $ be the unitarily invariant probability measure on $\mathbb P^{n-1}(\mathbb C)$ and $d$ the metric on $\mathbb P^{n-1}(\mathbb C)$ introduced in subsection 2.9.17. For $v\in \mathbb {C}^ n\setminus \{ 0\} $ we write $[v]\in \mathbb P^{n-1}(\mathbb C)$ for the complex line spanned by $v$; thus the restriction of $v\mapsto [v]$ to the unit sphere is the Hopf map of subsection 2.9.17. For $y\neq x$ in $\mathbb {C}^ n$, the point $[y-x]$ is the direction of $y$ from $x$.

Lemma 5.11.2.B88B826B1EDC Let $n\ge 2$ and let $K\subset \mathbb {C}^ n$ be a compact $\mathcal H^{2n-3}$-null set. Then:

  1. $K$ is real polar;

  2. for every $x\in \mathbb {C}^ n$, the set of directions from $x$ of the points of $K\setminus \{ x\} $ has $\sigma $-measure zero;

  3. the orthogonal projection of $K$ onto any complex hyperplane is Lebesgue null.

Proof. (1) Suppose that $K$ is not polar. By theorem 4.1.17, $K$ carries a nonzero positive measure of finite energy; dividing it by its mass, we obtain a probability measure $\mu $ on $K$ of finite energy $I(\mu )$ for the Newtonian kernel $k(x,y)=|x-y|^{2-2n}/((2n-2)s_{2n-1})$ of subsection 4.1.1 in $\mathbb R^{2n}$. Its potential $U^\mu $ is lower semicontinuous, and it is finite $\mu $-almost everywhere because $\int U^\mu \, \mathrm d\mu =I(\mu )<\infty $. The sets $E_ M=K\cap \{ U^\mu \le M\} $, $M\in \mathbb N$, are compact, because $U^\mu $ is lower semicontinuous, and they increase to a set of full $\mu $-measure; hence $\mu (E_ M)>0$ for $M$ large enough. Let $\nu $ be the restriction of $\mu $ to $E_ M$. For $a\in E_ M$ and $r>0$,

\[ \frac{r^{2-2n}\, \nu (B(a,r))}{(2n-2)s_{2n-1}}\leq \int _{B(a,r)}k(a,y)\, \mathrm d\nu (y)\leq U^\mu (a)\leq M, \]

so $\nu (B(a,r))\le C_ nMr^{2n-2}$ with $C_ n=(2n-2)s_{2n-1}$. Now cover $E_ M$ by balls $B(c_ l,r_ l)$ with $r_ l<1$ and $\sum _ lr_ l^{2n-3}<\eta $, and discard those not meeting $E_ M$. Each remaining ball lies in $B(a_ l,2r_ l)$ for some $a_ l\in E_ M$, so

\[ \nu (E_ M)\leq \sum _ l\nu (B(a_ l,2r_ l))\leq 2^{2n-2}C_ nM\sum _ lr_ l^{2n-2}\leq 2^{2n-2}C_ nM\eta , \]

where we used $r_ l<1$. Since $\eta >0$ is arbitrary, $\nu (E_ M)=0$, a contradiction. Hence $K$ is polar.

(2) Fix $x$. For $m\geq 1$ the set $K_ m=K\cap \{ |y-x|\geq 1/m\} $ is compact and $\mathcal H^{2n-3}$-null, and $K\setminus \{ x\} =\bigcup _ mK_ m$. The direction map $\Phi (y)=[y-x]$ is Lipschitz on $K_ m$. Indeed, $\Phi (y)=[(y-x)/|y-x|]$. The normalization $y\mapsto (y-x)/|y-x|$ has Lipschitz constant at most $2m$ on $\{ |y-x|\geq 1/m\} $: for nonzero vectors $a,b$,

\[ \Bigl |\frac a{|a|}-\frac b{|b|}\Bigr |\leq \frac{|a-b|}{|a|}+|b|\, \Bigl |\frac1{|a|}-\frac1{|b|}\Bigr |=\frac{|a-b|+\bigl ||b|-|a|\bigr |}{|a|}\leq \frac{2|a-b|}{|a|}, \]

and we apply this with $a=y-x$, $b=y'-x$ and $|a|\geq 1/m$. The Hopf map is Lipschitz with constant $\pi /2$ for the metric $d$. Indeed, let $e,e'$ be unit vectors and $\delta =\arccos \operatorname{Re}\langle e,e'\rangle \in [0,\pi ]$ their spherical distance. Since $d$ is the minimum of the spherical distance over the unit representatives (subsection 2.9.17), we have $d([e],[e'])\leq \delta $. Moreover $|e-e'|^2=2-2\cos \delta =4\sin ^2(\delta /2)$, and $\sin s\geq 2s/\pi $ for $0\leq s\leq \pi /2$, so $|e-e'|\geq 2\delta /\pi $. Hence

\[ d([e],[e'])\leq \delta \leq \tfrac \pi 2|e-e'|. \]

Since Lipschitz images of $\mathcal H^{2n-3}$-null sets are $\mathcal H^{2n-3}$-null, the compact set $\Phi (K_ m)$ is $\mathcal H^{2n-3}$-null in $(\mathbb P^{n-1}(\mathbb C),d)$. Cover it by balls of radii $r_ l<1$ with $\sum _ lr_ l^{2n-3}<\eta $. By lemma 2.9.19, and since $r_ l<1$, we have $\sigma (\Phi (K_ m))\leq \sum _ lr_ l^{2n-2}<\eta $, so $\sigma (\Phi (K_ m))=0$. Hence the set of directions $\Phi (K\setminus \{ x\} )=\bigcup _ m\Phi (K_ m)$ is a Borel set of $\sigma $-measure zero.

(3) An orthogonal projection onto a complex hyperplane $H\cong \mathbb R^{2n-2}$ is $1$-Lipschitz. Hence the projection of $K$ is $\mathcal H^{2n-3}$-null in $H$, and therefore Lebesgue null because $2n-3<2n-2$. $\square$

Theorem 5.11.3.76D63A606619 Let $n\ge 2$, let $\Omega \subset \mathbb {C}^ n$ be a domain, and let $S\subset \Omega $ be relatively closed and locally $\mathcal H^{2n-3}$-null, that is, every point of $S$ has a neighborhood $W$ such that $S\cap W$ is $\mathcal H^{2n-3}$-null. Then every plurisubharmonic function on $\Omega \setminus S$ extends uniquely to a plurisubharmonic function on $\Omega $, given on $S$ by the same upper-limit formula as in theorem 5.11.1.

Proof. Let $u$ be a plurisubharmonic function on $\Omega \setminus S$. For every closed ball $\overline B\subset \Omega $, the compact set $S\cap \overline B$ is covered by finitely many neighborhoods $W$ as in the hypothesis, and is therefore $\mathcal H^{2n-3}$-null.

Step 1: $u$ is locally bounded above near $S$. Fix $p\in S$ and $r>0$ with $\overline B=\overline{B(p,r)}\subset \Omega $, and write $B=B(p,r)$. By lemma 5.11.2(2) applied with $x=p$, almost every complex line through $p$ meets $S\cap \overline B$ only at $p$; fix such a line $L=\{ p+te:t\in \mathbb {C}\} $, with $|e|=1$. The circle $\Gamma =L\cap \partial B$ is a compact subset of $\Omega \setminus S$, so the upper semicontinuous function $u$ is bounded above on $\Gamma $ by a finite constant $C$, and the open set $O=\{ y\in \Omega \setminus S:u(y)<C+1\} $ contains $\Gamma $.

Suppose that $u$ is not bounded above near $p$. Then there are points $x_ j\in B\setminus S$ with $x_ j\to p$ and $u(x_ j)\to +\infty $. For each $j$, lemma 5.11.2(2) applied with $x=x_ j$ shows that the directions of the lines through $x_ j$ meeting $S\cap \overline B$ form a $\sigma $-null set. Every nonempty open ball in $\mathbb P^{n-1}(\mathbb C)$ has positive $\sigma $-measure by lemma 2.9.19, so the complement of this null set is dense. We can therefore choose directions $\xi _ j$ outside this null set with $d(\xi _ j,[e])<1/j$. Since $d$ is the minimum of the spherical distance over the unit representatives, $\xi _ j$ has a unit representative $e_ j$ whose spherical distance to $e$ is less than $1/j$, and then $|e_ j-e|<1/j$, because a chord is shorter than the arc. Thus $e_ j\to e$, and the lines $L_ j=\{ x_ j+te_ j:t\in \mathbb {C}\} $ do not meet $S\cap \overline B$.

We claim that $L_ j\cap \partial B\subset O$ for all large $j$. Let $c_ j$ be the point of $L_ j$ closest to $p$, that is, the orthogonal projection of $p$ to $L_ j$; then $|c_ j-p|\leq |x_ j-p|\to 0$. Since $L_ j$ is a complex line, $c_ j=x_ j+\langle p-x_ j,e_ j\rangle e_ j$ and $\langle c_ j-p,e_ j\rangle =0$ for the Hermitian inner product, so $\operatorname{Re}\langle c_ j-p,se_ j\rangle =0$ and $|c_ j+se_ j-p|^2=|c_ j-p|^2+|s|^2$ for $s\in \mathbb {C}$. Hence $L_ j\cap \partial B$ is the circle $\{ c_ j+\rho _ j\mathrm{e}^{\mathrm{i}\theta }e_ j\} $ with $\rho _ j=(r^2-|c_ j-p|^2)^{1/2}\to r$. Since $|c_ j+\rho _ j\mathrm{e}^{\mathrm{i}\theta }e_ j-p-r\mathrm{e}^{\mathrm{i}\theta }e|\leq |c_ j-p|+|\rho _ j-r|+r|e_ j-e|$, these circles converge uniformly in $\theta $ to $\Gamma =\{ p+r\mathrm{e}^{\mathrm{i}\theta }e\} $. The distance from the compact set $\Gamma $ to the closed complement of the open set $O$ is positive, so the circles lie in $O$ for large $j$.

For such $j$, the closed disc $\Delta _ j=L_ j\cap \overline B$ lies in $\Omega \setminus S$, and the restriction of $u$ to the connected component of $L_ j\cap (\Omega \setminus S)$ containing $\Delta _ j$ is subharmonic or identically $-\infty $. Its values on the boundary circle of $\Delta _ j$ are less than $C+1$, so the maximum principle, that is, theorem 3.4.5 with the constant harmonic function $C+1$, gives $u\leq C+1$ on $\Delta _ j$. Since $x_ j\in B\cap L_ j\subset \Delta _ j$, this gives $u(x_ j)\leq C+1$, contradicting $u(x_ j)\to +\infty $. Hence $u$ is bounded above near every point of $S$.

Step 2: the subharmonic extension. By lemma 5.11.2(1), every compact set $S\cap \overline B$ as above is real polar. Countably many such sets cover $S$, so $S$ is real polar by proposition 4.1.11. By Step 1 and the upper semicontinuity of $u$ on $\Omega \setminus S$, the function $u$ is bounded above on $K\setminus S$ for every compact $K\subset \Omega $. Moreover $u$ is subharmonic on $\Omega \setminus S$ by theorem 5.9.1. By theorem 3.4.27, $u$ has a subharmonic extension $\widetilde u$ to $\Omega $, given on $S$ by the upper-limit formula. Since $S$ is Lebesgue null and $u$ is finite almost everywhere on $\Omega \setminus S$, the function $\widetilde u$ is not identically $-\infty $ on the domain $\Omega $, so it is locally integrable.

Step 3: $\widetilde u$ is plurisubharmonic. We verify the hypothesis of lemma 5.8.2 for the locally integrable function $\widetilde u$. Fix a unit vector $a\in \mathbb {C}^ n$ and a nonnegative $\chi \in C^\infty _ c(\Omega )$, and write $z=v+ta$ with $v\in (\mathbb {C}a)^\perp $ and $t\in \mathbb {C}$, as in that lemma. The compact set $S\cap \operatorname{Supp}\chi $ is $\mathcal H^{2n-3}$-null, since it is covered by finitely many of the sets $S\cap \overline B$ of the first paragraph, so by lemma 5.11.2(3) its orthogonal projection onto $(\mathbb {C}a)^\perp $ is Lebesgue null. For $v$ outside this projection, the line $\{ v+ta:t\in \mathbb {C}\} $ meets $\operatorname{Supp}\chi $ only in $\Omega \setminus S$. Thus the open set $G_ v=\{ t:v+ta\in \Omega \setminus S\} $ contains $K_ v=\{ t:v+ta\in \operatorname{Supp}\chi \} $, and its connected components are the connected components of an affine complex-line section of $\Omega \setminus S$, on each of which the restriction of $\widetilde u=u$ is subharmonic or identically $-\infty $. As $a$ and $\chi $ are arbitrary, lemma 5.8.2 gives $\mathrm{dd}^{\mathrm{c}}\widetilde u\geq 0$. Since $\widetilde u$ is a locally integrable subharmonic function, it is the canonical representative of its class, and it is plurisubharmonic by theorem 5.9.1. Uniqueness follows from the same theorem, since two plurisubharmonic extensions agree outside the Lebesgue null set $S$. $\square$

Cite this page

Comments

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