1 Introduction

Let \(D\subset \mathbb {C}^n\). Using a local convolution, for any plurisubharmonic function u on D, one can find a sequence \(u_k\) of smooth plurisubharmonic functions, which decreases to u on compact subsets of D. The purpose of this note is to show that (for regular enough D) it is possible to choose such a sequence near any point on the boundary.

A domain \(D\subset \mathbb {C}^n\) will be called an \(\mathcal {S}\)-domain if for any plurisubharmonic function u on D, one can find a sequence \(u_k\) of smooth plurisubharmonic functions on D, which decreases to u.Footnote 1 Note that pseudoconvex domains, tube domains and Riendhard domains are \(\mathcal {S}\)-domains (see [3]).

Theorem 1

Let \(D\subset \mathbb {C}^n\) be a domain with Lipschitz boundary. Then for any boundary point P there is a neighbourhood U of P such that \(D\cap U\) is an \(\mathcal {S}\)-domain with Lipschitz boundary.

In Sect. 3, by a slight modification of an example from [2], we show the necessity of the assumption on the boundary.

Because of the fact that not every smooth domain is an \(\mathcal {S}\)-domain (see [1]) we have the following (surprising for the author) corollary:

Corollary 2

Being an \(\mathcal {S}\)-domain is not a local property of the boundary.

2 Proof

We need the following lemma:

Lemma 3

Let \(D\subset \mathbb {R}^m\) be an open set. Let u be a subharmonic function on D and let \(P\in D\). Let \(a,b,R,C>0\), \(B=\{x\in \mathbb {R}^m:|x-P|<R\}\), \(K=\{(x',x_m)\in \mathbb {R}^{m-1}\times \mathbb {R}=\mathbb {R}^m:-a< x_m<-b|x'|\}\) and \(B+K=\{x+y:x\in B, y\in K\}\subset D\). Assume that for any \(x\in B\) and \(y\in K\)

$$\begin{aligned} u(x+y)\le u(x)+\delta (|y|), \end{aligned}$$

where \(\delta :(0,+\infty )\rightarrow (0,+\infty )\) is increasing and such that \(\lim _{t\rightarrow 0^+}\delta (t)=0\). Then u is continuous at P.

Proof

Let \((x_n)\) be any sequence in D which converges to P. Let \(S_n=\{x\in \mathbb {R}:|x-P|=2|x_n-P|\}\), \(A_n=S_n\cap (\{x_n\}+K)\) and \(B_n=S_n{\setminus } A_n\). For n large enough \(x_n\in B\) and \(|x_n-P|\le \frac{a}{3} \). Hence there is a constant \(\alpha >0\) (which depends only on b) such that \(\alpha _n=\frac{\sigma (A_n)}{\sigma (S_n)}\ge \alpha \) where \(\sigma \) is the standard measure on a sphere. Let \(M_n=sup_{S_n}u\). We can estimate using the assumptions:

$$\begin{aligned}&u(P)\le \sigma (S_n)^{-1}\int _{A_n}ud\sigma +\sigma (S_n)^{-1}\int _{B_n}ud\sigma \\&\quad \le \alpha _n(u(x_n)+\delta (3|x_n-P|))+(1-\alpha _n)M_n, \end{aligned}$$

hence

$$\begin{aligned} u(x_n)\ge u(P)+\frac{1-\alpha _n}{\alpha _n}(u(P)-M_n)-\delta (3|x_n-P|). \end{aligned}$$

Letting n to \(\infty \) we get

$$\begin{aligned} \varliminf _{n\rightarrow \infty }u(x_n)\ge u(P). \end{aligned}$$

Since u is upper semicontinuous the proof is completed. \(\square \)

The function \(P_Df:=\sup \{u\in \mathcal {PSH}(D):u\le f\}\), where \(D\subset \mathbb {C}^n\) and f is a (real) function on D, is called a plurisubharmonic envelope of f.

Proof of Theorem 1

We use the following notation \(\mathbb {C}^n\ni z=(a,x)\in (\mathbb {C}^{n-1}\times \mathbb {R})\times \mathbb {R}\). We put \(B=\{a\in \mathbb {C}^{n-1}\times \mathbb {R}:|a|<1\}\). After an affine change of coordinates we can assume that there is a constant \(C>0\) and a function \(F:B\rightarrow (3C,4C)\) such that:

  1. (i)

    \(F(a)-F(b)\le C|a-b|\) for \(a,b\in B\),

  2. (ii)

    \(\partial D\cap B\times [-5C,5C]=\{(a,F(a)):a\in B\}\) and \(P=(0,F(0))\),

  3. (iii)

    \(0\in D\).

Let \(U=\{(a,x)\in B\times \mathbb {R}:|a|^2+\left( \frac{x}{5C}\right) ^2<1\}\). We will show that \(\Omega =D\cap U\) is an \(\mathcal {S}\)-domain. Observe that

$$\begin{aligned} \partial \Omega \cap \left( B\times [0,5C]\right) =\{(a,\hat{F}(a)):a\in B\}, \end{aligned}$$

where \(\hat{F}(a)=\min \{F(a),5C\sqrt{1-|a|^2}\}\). By elementary calculations we get

$$\begin{aligned} \{\hat{F}=F\}\subset B'=\left\{ a\in \mathbb {C}^{n-1}\times \mathbb {R}:|a|<\frac{4}{5}\right\} \end{aligned}$$
(1)

and

$$\begin{aligned} \hat{F}(a)-\hat{F}(b)\le C'|a-b| \quad \hbox {for } a,b\in B', \end{aligned}$$
(2)

where \(C'=\frac{20}{3}C<7C\). For \(0<\varepsilon <\sup _{B'}F-3C\) let

$$\begin{aligned} K=K(\varepsilon )=\{(a,x):-\varepsilon <x<-7C|a|\} \end{aligned}$$

and let

$$\begin{aligned} \Omega _k=\Omega _k(\varepsilon )=\{z\in \Omega :z+kw\in \Omega \hbox { for any } w\in \bar{K}\},\hbox { for }k=1,2. \end{aligned}$$

By (1) and (2), we have \(\partial \Omega _2\cap \partial D=\partial \Omega \cap \partial D\). This gives us

$$\begin{aligned} \partial \Omega _2=\left( \partial \Omega _2\cap \partial D\right) \cup \left( \partial \Omega _2\cap \partial U\right) \cup \left( \partial \Omega _2\cap \Omega _1\right) . \end{aligned}$$

Therefore by (2) it is clear that

$$\begin{aligned} z+w\in \Omega \end{aligned}$$

for \(z\in \partial \Omega _2\cap \partial D\) and \(w\in \bar{K}\). The same inequality holds on \(\partial \Omega _2\cap \partial U\) because of the convexity of U. Thus we get

$$\begin{aligned} L=L(\varepsilon ):=\{z\in \bar{\Omega }_2:\mathrm{dist}(z,\partial \Omega )\ge \mathrm{dist}(z+w,\partial \Omega )\hbox { for some } w\in \bar{K}\}\Subset \Omega _1. \end{aligned}$$

Let \(d=-\log (\mathrm{dist}(\cdot ,\partial \Omega ))\) and \(d'=-\log (\mathrm{dist}(\cdot ,\partial U))\). We only know that the second function is plurisubharmonic, but by the construction of \(\Omega \) and \(\Omega _2\) (decreasing \(\varepsilon \) if necessary) we have \(d=d'\) on \(\Omega {\setminus }\Omega _2\), hence on this set, the function d is plurisubharmonic too.

Let \(u\in \mathcal {PSH}(\Omega )\) and let \(\phi _k\) be a sequence of continuous functions on \(\Omega \) which decreases to u. We can choose an increasing convex function \(p:\mathbb {R}\rightarrow \mathbb {R}\) such that for a function \(\rho =p\circ d\) we have \(\lim _{z\rightarrow \partial \Omega }\rho -\phi _1=+\infty \) (see claim 3.5 in [5]). Put \(\tilde{\phi }_k=\max \{\phi _k,\rho -k\}\). Observe that functions \(\hat{\phi }_k=P_\Omega \tilde{\phi }_k\) are plurisubharmonic and they decrease to u.

Fix k. Because \(\tilde{\phi }_k=\rho -k\) outside of a compact set, for \(\varepsilon >0\) small enough, we have

$$\begin{aligned} cl_\Omega (\Omega {\setminus }\Omega _2)\subset S, \end{aligned}$$

where \(S=S(\varepsilon )=int\{\tilde{\phi }_k=\rho -k\}\).

The function \(\rho ':=p\circ d'\) is plurisubharmonic and \(\rho '\le \tilde{\phi }_k\). Thus on \(\Omega _2\) we have

$$\begin{aligned} \tilde{\phi }_k\ge P_{\Omega _2}\tilde{\phi }_k\ge \hat{\phi }_k\ge \rho '-k \end{aligned}$$

and therefore the function v given by

$$\begin{aligned} v(z)=\left\{ \begin{array}{ll} P_{\Omega _2}\tilde{\phi }_k(z) &{} \quad \hbox {for} \quad z\in \Omega _2\\ \rho (z)'-k(=\tilde{\phi }_k) &{} \quad \hbox {on} \quad \Omega {\setminus }\Omega _2, \end{array}\right. \end{aligned}$$

is plurisubharmonic on \(\Omega \) and smaller than \(\tilde{\phi _k}\). Thus \( P_{\Omega _2}\tilde{\phi }_k=\hat{\phi }_k|_{\Omega _2}\). Let

$$\begin{aligned} L'=L'(\varepsilon ):=\{z\in \bar{\Omega }_2:\rho (z)\ge \sup _{\Omega \setminus S}\phi _k\}\Subset \Omega _1, \end{aligned}$$

then we have

$$\begin{aligned} N=N(\varepsilon ):=(L\cup L')+\bar{K}\Subset \Omega . \end{aligned}$$

Observe that for \(z\in \Omega _2\) and \(w\in K\) we have \(\tilde{\phi }_k(z)\ge \tilde{\phi }_k(z+w)-\omega (|w|)\), where \(\omega \) is the modulus of continuity of the function \(\tilde{\phi }_k|_N\). Therefore, \(\hat{\phi }_k(z)\ge \hat{\phi }_k(z+w)-\omega (|w|)\). By Lemma 3 the function \(\hat{\phi }_k|_{\Omega _2}\) is continuous.

Because any \(z\in \Omega \) is in \(\Omega _2(\varepsilon )\) for some \(\varepsilon \) as above, we obtain that the function \(\hat{\phi }_k\) is continuous on \(\Omega \). Using the Richberg theorem we can modify the sequence \(\hat{\phi }_k\) to a sequence \(u_k\) of smooth plurisubharmonic functions which decreases to u. \(\square \)

The approximation by continuous functions, can be proved in the same way in a much more general situation.

Theorem 4

Let \(\mathbf F \) be a constant coefficient subequation such that all \(\mathbf F \)-subharmonic functions are subharmonic and all convex functions are \(\mathbf F \)-subharmonic. Let \(D\subset \mathbb {R}^n\) be a domain with Lipschitz boundary. Then for any point \(P\in \bar{D}\) there is a neighbourhood U of P such that for any function \(u\in \mathbf F (D\cap U)\) there is a sequence \((u_k)\subset \mathbf F (D\cap U)\) of continuous functions decreasing to u.

Here we use terminology from [4].Footnote 2

3 Example

Similarly as in Lecture 14 in [2] we construct a domain \(\Omega \subset \mathbb {C}^n\) and a plurisubharmonic function u on \(\Omega \) which can not be regularized. Let \(A=\{\frac{1}{k}:k\in \mathbb {N}\}\) and a sequence \((x_k)\subset (0,1){\setminus } A\) is such that its limit set is equal \(\bar{A}\). Put

$$\begin{aligned} \lambda (z)=\sum _{k=1}^\infty c_k\log |z-x_k| \quad \hbox {for}\,\, z\in \mathbb {C}, \end{aligned}$$

where \(c_k\) is a such sequence of numbers rapidly decreasing to 0, such that

  1. (i)

    \(\lambda \) is a subharmonic function on \(\mathbb {C}\) and

  2. (ii)

    \(\lambda |_A\ge -\frac{1}{2}\).

For \(k\in \mathbb {N}\) let \(D_k\) be a disc with center \(x_k\) such that \(\lambda |_{D_k}<-1\). Now, we can define

$$\begin{aligned} \Omega =\{(z',z_n)\in \mathbb {C}^{n-1}\times \mathbb {C}:|z'|^2+|z_n-1|^2<1\}{\setminus } K, \end{aligned}$$

where

$$\begin{aligned} K=\{(z',z_n)\in \mathbb {C}^{n-1}\times \mathbb {C}:|z'|=|z_n| \hbox { and } z_n\notin \cup _{k=1}^\infty D_k\}, \end{aligned}$$

and

$$\begin{aligned} u(z',z_n)=\left\{ \begin{array}{ll} -1 &{} \quad \hbox {for} \quad z\in \Omega \cap D\\ \max \{\lambda {(z_n)},-1\} &{} \quad \hbox {on} \quad \Omega {\setminus } D, \end{array}\right. \end{aligned}$$

where \(D=\{(z',z_n)\in \mathbb {C}^{n-1}\times \mathbb {C}:|z'|>|z_n| \}\).

Let U be any neighbourhood of 0. We can choose numbers \(k\in \mathbb {N}\), \(0<r<\frac{1}{k}\) such that

$$\begin{aligned} \Omega _U:=\left\{ (z',z_n)\in \mathbb {C}^{n-1}\times \mathbb {C}:|z'|\le \frac{2}{k},|z_n-\frac{1}{k}|\le r\right\} {\setminus } K\subset \Omega \cap U. \end{aligned}$$

Let \(y_p\) be a subsequence of \(x_p\) such that \(\lim _{p\rightarrow \infty }y_p=\frac{1}{k}\) and for all p we have \(|y_p-\frac{1}{k}|<r\). Now, we can repeat the argument from [2]. If \(u_q\) is a sequence of smooth plurisubharmonic functions decreasing to u on \(\Omega _U\), then for q sufficiently large \(u_q\le -\frac{3}{4}\) on the set

$$\begin{aligned} \left\{ (z',z_n)\in \mathbb {C}^{n-1}\times \mathbb {C}:|z'|=\frac{2}{k},|z_n-\frac{1}{k}|\le r\right\} \subset \partial \Omega _U. \end{aligned}$$

By the maximum principle (on sets \(\{(z',z_n)\in \mathbb {C}^{n-1}\times \mathbb {C}:|z'|\le \frac{2}{k},z_n=y_p\}\subset \mathbb {C}^{n-1}\times \{y_p\}\)) we also have \(u_q(0,y_p)\le -\frac{3}{4}\) and by continuity of \(u_q\) we get \(u_q(0,\frac{1}{k})\le -\frac{3}{4}<u(0,\frac{1}{k})\). This is a contradiction.

Note that we have not only just proved that for the above \(\Omega \) Theorem 1 does not hold but we also have the following stronger result:

Proposition 5

Let \(\Omega \) and u be as above. Then the function u can not be smoothed on \(U\cap \Omega \) for any neighbourhood U of \(0\in \partial \Omega \).

4 Questions

In this last section we state some open questions related to the content of the note.

  1. 1.

    Is it possible to characterize \(\mathcal {S}\)-domains? In view of Corollary 2, even in the class of smooth domains, it seems to be a challenging problem.

  2. 2.

    Let D be an \(\mathcal {S}\)-domain and let f be a continuous function which is bounded from below. Is the plurisubharmonic envelope of f continuous? Assume in addition that f is smooth. What is the optimal regularity of \(P_Df\)? The author does not know answers even in the case of the ball in \(\mathbb {C}^n\). Note that if D is not an \(\mathcal {S}\)-domain, then there is a smooth function f bounded from below such that \(P_Df\) is discontinuous.

  3. 3.

    What is the optimal assumption about the regularity of the boundary of D in the Theorem 1? Is it enough to assume that \(D=int\bar{D}\)?

  4. 4.

    Let M be a real (smooth or Lipschitz) hypersurface in \(\mathbb {C}^n\). Is it true that for any \(P\in M\) there exists a smooth pseudoconvex neighbourhood \(U\subset B\) such that M divides U into two \(\mathcal {S}\)-domains?