Pome

5.8 The line definition and upper semicontinuity55393AEE99C8

55393AEE99C8

We identify $\mathbb {C}^ n$ with $\mathbb R^{2n}$ through $z_ j=x_ j+\mathrm{i}y_ j$, and we call a function on an open subset of $\mathbb {C}^ n$ subharmonic if it is subharmonic in the sense of definition 3.4.2 on the corresponding open subset of $\mathbb R^{2n}$. Recall from subsection 2.9.20 that an affine complex line is a set $L=\{ a+\zeta b:\zeta \in \mathbb {C}\} $ with $a\in \mathbb {C}^ n$ and $b\in \mathbb {C}^ n\setminus \{ 0\} $. For an open set $\Omega \subset \mathbb {C}^ n$, the section $L\cap \Omega $ is identified with the open set $\{ \zeta \in \mathbb {C}:a+\zeta b\in \Omega \} $, and a function on a connected component of $L\cap \Omega $ is called subharmonic if its composite with $\zeta \mapsto a+\zeta b$ is subharmonic; this does not depend on the parametrization of $L$. Accordingly, the restriction of a function $u$ on $\Omega $ to $L\cap \Omega $ is the function $\zeta \mapsto u(a+\zeta b)$.

Definition 5.8.1.6070D2D83EC6 Let $\Omega \subset \mathbb {C}^ n$ be a domain. If $n\ge 1$, a function $u\colon \Omega \to [-\infty ,\infty )$ is plurisubharmonic if $u\not\equiv -\infty $, $u$ is upper semicontinuous, and for every affine complex line $L$ the restriction of $u$ to each connected component of $L\cap \Omega $ is subharmonic or identically $-\infty $. The set of plurisubharmonic functions on $\Omega $ is denoted by $\mathrm{PSH}(\Omega )$. In dimension zero a domain is a single point, and the plurisubharmonic functions on it are the real-valued functions.

We often abbreviate plurisubharmonic to psh. Since the constant $-\infty $ is subharmonic by definition 3.4.2, the alternative “identically $-\infty $” in the line condition is stated only for emphasis; it occurs, for instance, for $\log |z_1|^2$ on the line $\{ z_1=0\} $. In dimension one the only affine complex line is $\mathbb {C}$ itself, so $\mathrm{PSH}(\Omega )$ is the set of nontrivial subharmonic functions on $\Omega $, that is, of those not identically $-\infty $. Whenever a limit of plurisubharmonic functions may degenerate to the function identically $-\infty $, we state this alternative explicitly.

We keep upper semicontinuity in definition 5.8.1, but it is redundant: theorem 5.8.4 below shows that it follows from the other two conditions. The proof of that theorem uses the following slicing lemma, which we apply again in theorem 5.9.1, theorem 5.11.3: for a locally integrable function, subharmonicity on the complex lines parallel to a fixed direction makes the second derivative in that direction a positive distribution.

Lemma 5.8.2.3C70A4ABB14A Let $\Omega \subset \mathbb {C}^ n$ be open, $n\ge 1$, let $u\colon \Omega \to [-\infty ,\infty )$ be a locally integrable function, let $a\in \mathbb {C}^ n$ be a unit vector, and let $\chi \in C^\infty _ c(\Omega )$ be nonnegative. Write the points of $\mathbb {C}^ n$ as $v+ta$ with $v\in (\mathbb {C}a)^\perp $ and $t\in \mathbb {C}$, and put $K_ v=\{ t\in \mathbb {C}:v+ta\in \operatorname{Supp}\chi \} $. Assume that for almost every $v\in (\mathbb {C}a)^\perp $ there is an open set $G_ v\subset \{ t\in \mathbb {C}:v+ta\in \Omega \} $ containing $K_ v$ such that the restriction of $t\mapsto u(v+ta)$ to each connected component of $G_ v$ is subharmonic, possibly identically $-\infty $. Then

\[ \Bigl \langle \sum _{j,k}a_ j\bar a_ ku_{j\bar k},\chi \Bigr \rangle \geq 0. \]

Consequently, if this hypothesis holds for every unit vector $a$ and every nonnegative $\chi \in C^\infty _ c(\Omega )$, then $\mathrm{dd}^{\mathrm{c}}u\geq 0$. This is the case, in particular, if the restriction of $u$ to each connected component of each affine complex-line section of $\Omega $ is subharmonic.

Proof. The map $(v,t)\mapsto v+ta$ is a unitary isomorphism of $(\mathbb {C}a)^\perp \times \mathbb {C}$ onto $\mathbb {C}^ n$, so it preserves Lebesgue measure, and by the chain rule $\partial _ t\partial _{\bar t}[\chi (v+ta)]=\sum _{j,k}a_ j\bar a_ k\chi _{j\bar k}(v+ta)$. We extend $u$ and $\chi $ by zero outside $\Omega $. Since $u$ is locally integrable and the derivatives of $\chi $ are bounded and supported in $\operatorname{Supp}\chi $, Fubini's theorem gives

\[ \Bigl \langle \sum _{j,k}a_ j\bar a_ ku_{j\bar k},\chi \Bigr \rangle =\int _{(\mathbb {C}a)^\perp }I(v)\, \mathrm{d}V_{2n-2}(v), \]

where

\[ I(v)=\int _{\mathbb {C}}u(v+ta)\, \partial _ t\partial _{\bar t}[\chi (v+ta)]\, \mathrm{d}V_2(t). \]

Applied to $|u|\mathbf1_{\operatorname{Supp}\chi }$, Fubini's theorem also shows that for almost every $v$ the function $t\mapsto u(v+ta)$ is integrable on $K_ v$. It therefore suffices to show that $I(v)\geq 0$ for every $v$ for which this integrability holds and an open set $G_ v$ as in the hypothesis exists.

Fix such a $v$. The compact set $K_ v$ is covered by the disjoint open connected components of $G_ v$, so it meets only finitely many of them, say $U_1,\ldots ,U_ N$, and each set $K_ v\cap U_ i$ is compact, being the complement in $K_ v$ of the open set $K_ v\cap \bigcup _{l\neq i}U_ l$. Hence the function $\chi _ i(t)=\chi (v+ta)$, $t\in U_ i$, is a nonnegative test function on $U_ i$, and

\[ I(v)=\sum _{i=1}^ N\int _{U_ i}u(v+ta)\, \partial _ t\partial _{\bar t}\chi _ i(t)\, \mathrm{d}V_2(t). \]

If $\chi _ i\equiv 0$, the $i$-th term vanishes. Otherwise $\chi _ i>0$ on a nonempty open subset of $U_ i$, contained in $K_ v$, on which $t\mapsto u(v+ta)$ is integrable and hence finite almost everywhere. Thus the subharmonic function $w_ i(t)=u(v+ta)$ on the domain $U_ i$ is not identically $-\infty $, and it is locally integrable by proposition 3.4.8. Since $\partial _ t\partial _{\bar t}=\frac14\Delta $, the $i$-th term equals $\frac14\langle \Delta w_ i,\chi _ i\rangle $, which is nonnegative by theorem 3.4.12(1). Hence $I(v)\geq 0$.

If the hypothesis holds for every unit vector $a$ and every $\chi $, then $\sum _{j,k}a_ j\bar a_ ku_{j\bar k}$ is a positive distribution for every unit vector $a$, and by homogeneity for every $a\in \mathbb {C}^ n$. Since $\mathrm{dd}^{\mathrm{c}}u$ is a real $(1,1)$-current, the matrix criterion for positivity in section 5.4 shows that $\mathrm{dd}^{\mathrm{c}}u\geq 0$. For the last assertion, take $G_ v=\{ t\in \mathbb {C}:v+ta\in \Omega \} $ for every $v$; its connected components are the connected components of the affine complex-line sections of $\Omega $ in the direction $a$. $\square$

The proof of the next theorem also uses polar coordinates adapted to the complex lines through a point. Let $n\geq 1$, let $\tau $ be the normalized surface measure on the unit sphere $S^{2n-1}\subset \mathbb {C}^ n$, and let $s_{2n-1}$ be the area of this sphere. Let $x\in \mathbb {C}^ n$ and $r>0$. For every Borel function $F\colon B(x,r)\to [0,\infty ]$, integration in polar coordinates (3.1.2), in which $\mathrm dS=s_{2n-1}\, \mathrm d\tau $, gives the first equality in

A577EADC63E4 5.8.3
\begin{equation} \begin{split} \int _{B(x,r)}F\, \mathrm{d}V_{2n} & =s_{2n-1}\int _{S^{2n-1}}\int _0^ rF(x+\rho a)\rho ^{2n-1}\, \mathrm{d}\rho \, \mathrm{d}\tau (a)\\ & =\frac{s_{2n-1}}{2\pi }\int _{S^{2n-1}}\int _{|t|<r} F(x+ta)|t|^{2n-2}\, \mathrm{d}V_2(t)\, \mathrm{d}\tau (a). \end{split} \end{equation}

For the second, we replace $a$ by $\mathrm{e}^{\mathrm{i}\theta }a$, which does not change $\tau $, average over $\theta \in [0,2\pi ]$, and use polar coordinates $t=\rho \mathrm{e}^{\mathrm{i}\theta }$ in $\mathbb {C}$; Tonelli's theorem justifies the exchanges of integrals.

Theorem 5.8.4.1A391017E17A Let $\Omega \subset \mathbb {C}^ n$ be a domain, $n\ge 1$, and let $u\colon \Omega \to [-\infty ,\infty )$ be a function that is not identically $-\infty $. Assume that for every affine complex line $L$ the restriction of $u$ to each connected component of $L\cap \Omega $ is subharmonic or identically $-\infty $. Then $u$ is upper semicontinuous, and hence $u\in \mathrm{PSH}(\Omega )$.

Proof. We first treat finite nonnegative functions, in Steps 1–3, and then reduce the general case to this one by truncation in Step 4. For the first part we prove by induction on $n\geq 1$ the following assertion: if $\Omega \subset \mathbb {C}^ n$ is a domain and $u\colon \Omega \to [0,\infty )$ is a finite function whose restriction to each connected component of each affine complex-line section of $\Omega $ is subharmonic, then $u$ is upper semicontinuous. For $n=1$ the only affine complex line is $\mathbb {C}$, so $u$ is subharmonic on $\Omega $ and in particular upper semicontinuous.

Let $n\geq 2$, assume the assertion in dimension $n-1$, and let $u$ be as in the assertion. Write $z=(z',w)$ with $z'\in \mathbb {C}^{n-1}$ and $w\in \mathbb {C}$. For fixed $w$, the restriction of $u(\, \cdot \, ,w)$ to each connected component of the open set $\{ z':(z',w)\in \Omega \} $ satisfies the hypothesis in dimension $n-1$. Indeed, affine complex lines in $\mathbb {C}^{n-1}\times \{ w\} $ are affine complex lines in $\mathbb {C}^ n$, a connected component of a line section of the slice lies in a connected component of the corresponding line section of $\Omega $, and subharmonicity is inherited by open subsets. By the induction hypothesis, $u(\, \cdot \, ,w)$ is therefore upper semicontinuous on this open set. For fixed $z'$, the function $u(z',\, \cdot \, )$ is subharmonic on each connected component of $\{ w:(z',w)\in \Omega \} $.

Step 1: $u$ is Borel measurable. We need joint measurability of $u$ before we can integrate it. Fix a ball $B'\subset \mathbb {C}^{n-1}$, a point $w_0\in \mathbb {C}$ and $\rho >0$ with $B'\times B(w_0,2\rho )\subset \Omega $. For integers $N\ge 1$ put $u_ N=\min \{ u,N\} $, and for $0<r<\rho $ and $(z',w)\in B'\times B(w_0,\rho )$ define

\[ F_ r^ N(z',w)=\frac{1}{\pi r^2}\int _{B(w,r)}u_ N(z',\eta )\, \mathrm{d}V_2(\eta ). \]

The integral exists, since $u(z',\, \cdot \, )$ is subharmonic, hence upper semicontinuous, on $B(w_0,2\rho )\supset B(w,r)$, so the integrand is a Borel function with values in $[0,N]$. For fixed $w$, the function $F_ r^ N(\, \cdot \, ,w)$ is upper semicontinuous: if $z'_ l\to z'$ in $B'$, the reverse Fatou lemma, applicable because $0\le u_ N\le N$, gives

\[ \varlimsup _{l\to \infty }F_ r^ N(z'_ l,w) \le \frac{1}{\pi r^2}\int _{B(w,r)}\varlimsup _{l\to \infty }u_ N(z'_ l,\eta )\, \mathrm{d}V_2(\eta ) \le F_ r^ N(z',w), \]

where the last inequality holds because each $u_ N(\, \cdot \, ,\eta )$ is upper semicontinuous on $B'$. For fixed $z'$, the function $F_ r^ N(z',\, \cdot \, )$ is continuous: for $|w-v|\leq r$ the symmetric difference $B(w,r)\mathbin \triangle B(v,r)$ lies in the annulus $\{ r-|w-v|\leq |\eta -w|<r+|w-v|\} $, whose area is $4\pi r|w-v|$, so that $|F_ r^ N(z',w)-F_ r^ N(z',v)|\le 4N|w-v|/r$.

A function on $B'\times B(w_0,\rho )$ that is Borel in $z'$ and continuous in $w$ is Borel; we check this for $F_ r^ N$. For $k\geq 1$ partition $B(w_0,\rho )$ into countably many Borel sets $E_{k,i}$ of diameter less than $1/k$, choose a point $w_{k,i}$ in each nonempty $E_{k,i}$, and put $F_ k(z',w)=F_ r^ N(z',w_{k,i})$ for $w\in E_{k,i}$. Each $F_ k$ is Borel, because each $F_ r^ N(\, \cdot \, ,w_{k,i})$ is upper semicontinuous, hence Borel, and $F_ k(z',w)\to F_ r^ N(z',w)$ for every $(z',w)$, because $F_ r^ N(z',\, \cdot \, )$ is continuous and the chosen points $w_{k,i}$ with $w\in E_{k,i}$ tend to $w$. Thus $F_ r^ N$ is Borel, and so is $G_ N=\varlimsup _{k\to \infty }F^ N_{\rho /(k+1)}$.

Now fix $(z',w)\in B'\times B(w_0,\rho )$. The upper semicontinuous function $u(z',\, \cdot \, )$ is bounded above on the closed disc $\overline{B(w,\rho /2)}$ by some $M$. For $N\geq M$ the truncation does not change $u(z',\, \cdot \, )$ on this disc, so $F^ N_{\rho /(k+1)}(z',w)$ is the mean of $u(z',\, \cdot \, )$ over $B(w,\rho /(k+1))$. The function $u(z',\, \cdot \, )$ is subharmonic and finite on $B(w_0,2\rho )$, hence locally integrable there by proposition 3.4.8, and by (3.4.13) its means over the discs $B(w,\rho /(k+1))$ converge to $u(z',w)$ as $k\to \infty $. Hence $G_ N(z',w)=u(z',w)$ for all $N\geq M$, and $u=\lim _ NG_ N$ is Borel on $B'\times B(w_0,\rho )$. Countably many such product sets cover $\Omega $, so $u$ is Borel on $\Omega $.

Step 2: the canonical representative. By corollary 2.9.28, the finite Borel function $u$ is locally bounded above. Since $u\geq 0$, it follows that $u\in L^1_{\mathrm{loc}}(\Omega )$. Since the restriction of $u$ to each connected component of each affine complex-line section of $\Omega $ is subharmonic, lemma 5.8.2 gives $\mathrm{dd}^{\mathrm{c}}u\geq 0$. In particular $\Delta u=4\sum _ ju_{j\bar j}\geq 0$, so by theorem 3.4.12 $u$ has a canonical subharmonic representative $\widetilde u$, a locally integrable subharmonic function on $\Omega $ equal to $u$ almost everywhere. Its complex Hessian is the same positive current $\mathrm{dd}^{\mathrm{c}}u$. Applying proposition 5.4.4 to the parametrizations of affine complex lines, we see that the restriction of $\widetilde u$ to each connected component of each affine complex-line section of $\Omega $ is subharmonic or identically $-\infty $. Since $\widetilde u$ is upper semicontinuous and agrees almost everywhere with the finite function $u$, we conclude that $\widetilde u\in \mathrm{PSH}(\Omega )$.

Step 3: $u=\widetilde u$ everywhere. Fix $x\in \Omega $ and $r>0$ with $\overline{B(x,r)}\subset \Omega $. The set $Z=\{ u\ne \widetilde u\} $ is Borel, by Step 1, and Lebesgue null, so the left-hand side of (5.8.3) vanishes for $F=\mathbf1_ Z$. Since the weight $|t|^{2n-2}$ is positive for $t\neq 0$, for $\tau $-almost every $a$ the set $\{ t:|t|<r,\ x+ta\in Z\} $ has planar measure zero. Fix such an $a$. On the disc $\{ |t|<r\} $, the function $t\mapsto u(x+ta)$ is finite and subharmonic, while $t\mapsto \widetilde u(x+ta)$ is subharmonic or identically $-\infty $; the latter is impossible, because the two functions agree almost everywhere on the disc. By proposition 3.4.8 both are locally integrable subharmonic functions on the disc that agree almost everywhere, so they agree everywhere by theorem 3.4.12. At $t=0$ this gives $u(x)=\widetilde u(x)$. As $x$ was arbitrary, $u=\widetilde u$ is upper semicontinuous. This completes the induction.

Step 4: truncation. Let $u$ be as in the theorem, and for integers $m\geq 1$ put $u_ m=\max \{ u,-m\} +m$. This is a finite nonnegative function. Its restriction to each connected component of each affine complex-line section of $\Omega $ is subharmonic: if the restriction of $u$ is subharmonic, so is its maximum with the constant $-m$, and if the restriction of $u$ is identically $-\infty $, the restriction of $u_ m$ is the constant $0$. By the case already proved, each $u_ m$, and hence each $\max \{ u,-m\} $, is upper semicontinuous. For every real $c$ we have $\{ u<c\} =\bigcup _ m\{ \max \{ u,-m\} <c\} $, which is open. Thus $u$ is upper semicontinuous, and together with the hypotheses this gives $u\in \mathrm{PSH}(\Omega )$. $\square$

Cite this page

Comments

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