Abstract
We discuss some basic properties of the Sibony functions and pseudometrics.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(G\subset {\mathbb {C}}^n\) be a domain. For \(a\in G\) let
where \({\mathbb {D}}\subset {\mathbb {C}}\) stands for the unit disc, \({\mathcal {O}}(G,{\mathbb {D}})\), resp. \(\mathcal {PSH}(G)\) denote the set of all holomorphic functions on G having values in \({\mathbb {D}}\), resp. the set of all plurisubharmonic functions on G, and “\(u\in \mathcal C^p(\{a\})\)” means that u is of class \(\mathcal C^p\) in a neighborhood of a (cf. [1, § 4.2]). Note that \({\mathcal {S}}^{(1)}_G(a)\) is different from \({\mathcal {K}}_G(a)\) (see Remark 2.1(c)). Put
Obviously, \({\mathcal {M}}_G(a)\subset {\mathcal {S}}_G(a)\subset {\mathcal {K}}_G(a)\) and \({\mathcal {S}}^{(p)}_G(a)\subset {\mathcal {K}}_G(a)\), \(p\in {\mathbb {N}}\). If \({\mathcal {F}}\in \{{\mathcal {M}}, {\mathcal {S}}^{(p)}, {\mathcal {K}}\}\), then we define:
For \({\mathcal {F}}\in \{{\mathcal {M}}, {\mathcal {S}}, {\mathcal {K}}\}\) the families \((d^{\mathcal {F}}_G)_G\) and \((\delta ^{\mathcal {F}}_G)_G\) are holomorphically contractible, i.e.
\(\bullet \) \(d^{\mathcal {F}}_{{\mathbb {D}}}(0,t)=t\), \(t\in [0,1)\), \(\delta ^{\mathcal {F}}_{{\mathbb {D}}}(0;1)=1\);
\(\bullet \) for any domains \(G\subset {\mathbb {C}}^n\), \(D\subset {\mathbb {C}}^m\) and for any holomorphic mapping \(F:G\longrightarrow D\) we have
In particular, the families \((d^{\mathcal {F}}_G)_G\) and \((\delta ^{\mathcal {F}}_G)_G\) are invariant under biholomorphic mappings.
If \({\mathcal {F}}={\mathcal {M}}\), then we get the Möbius pseudodistance\(\varvec{m}_G:=d_G^{\mathcal {M}}\) and the Carathéodory–Reiffen pseudometric\({\varvec{\gamma }}_G:=\delta _G^{\mathcal {M}}\). It is known that
If \({\mathcal {F}}={\mathcal {S}}\), then we get the Sibony function\(\varvec{s}_G:=d_G^{\mathcal {S}}\) and the Sibony pseudometric\(\varvec{S}_G:=\delta _G^{\mathcal {S}}\). It is known that
where \({\mathcal {L}}u(a;X):=\sum _{j,k=1}^n\frac{\partial ^2u}{\partial z_j\partial \overline{z}_k}(a)X_j{{\overline{X}}}_k\) is the Levi form (cf. [1, Proposition 4.2.16]). In particular, \(\varvec{S}_G(a;\cdot )\) is a \({\mathbb {C}}\)-seminorm.
If \({\mathcal {F}}={\mathcal {K}}\), then we get the pluricomplex Green function\(\varvec{g}_G:=d_G^{\mathcal {K}}\) and the Azukawa pseudometric\(\varvec{A}_G:=\delta _G^{\mathcal {K}}\). It is known that \(\varvec{g}_G(a,\cdot )\in {\mathcal {K}}_G(a)\), \(\log \varvec{A}_G(a;\cdot )\in \mathcal {PSH}({\mathbb {C}}^n)\), and
If \({\mathcal {F}}={\mathcal {S}}^{(p)}\), \(p\ne 2\), then we get the higher order Sibony function\(\varvec{s}^{(p)}_G:=d^{{\mathcal {S}}^{(p)}}_G\) and the higher order Sibony pseudometric\(\varvec{S}^{(p)}_G:=\delta ^{{\mathcal {S}}^{(p)}}_G\).
Recall that many properties of complex domains/manifolds are encoded in \(\varvec{m}_G\), \({\varvec{\gamma }}_G\), \(\varvec{g}_G\), and \(\varvec{A}_G\). Therefore, these functions have been important tools to investigate various geometric problems in several complex analysis. Their study over the last years has led to the point that their basic properties are now well understood. To extent the family of invariant functions higher order functions were introduced and studied, e.g. the higher Carathéodory-Reiffen pseudometrics. But in contrast to that little is known on properties of \(\varvec{S}_G\) and almost nothing on \(\varvec{s}^{(p)}_G\), \(p\in {\mathbb {N}}\), and \(\varvec{S}^{(p)}_G\), \(p\ne 2\).
The main aim of this note is to show that the basic properties of \(\varvec{s}^{(p)}_G\) and \(\varvec{S}^{(p)}_G\) differ essentially from the corresponding properties of \(\varvec{m}_G\), \(\varvec{g}_G\), \({\varvec{\gamma }}_G\), and \(\varvec{A}_G\). Surprisingly, as we will see, many properties of the so far studied invariant functions fail to hold.
2 Holomorphic contractibility
Remark 2.1
-
(a)
\({\mathcal {S}}^{(p)}_G(a)=\{\root p \of {u}:\;u:G\longrightarrow [0,1): \log u\in \mathcal {PSH}(G),\;u\in \mathcal C^p(\{a\}),\;{\text {ord}}_au\ge p\}\), where \({\text {ord}}_au\) denotes the order of zero of u at a.
-
(b)
In view of the Taylor formula, we have
$$\begin{aligned} \varvec{S}^{(p)}_G(a;X)=\sup \big \{(\tfrac{1}{p!}|u^{(p)}(a)(X)|)^{1/p}: \root p \of {u}\in {\mathcal {S}}^{(p)}_G(a)\big \},\quad a\in G,\;X\in {\mathbb {C}}^n, \end{aligned}$$where \(u^{(p)}(a):{\mathbb {C}}^n\longrightarrow {\mathbb {R}}\) stands for the p-th Fréchet differential of u at a.
-
(c)
In view of (b), we get \(\varvec{S}^{(p)}_G(a;\cdot )\equiv 0\) for p odd. In particular, \(\varvec{S}^{(1)}_{\mathbb {D}}(0;1)=0<1=\varvec{A}_{\mathbb {D}}(0;1)\).
-
(d)
\(\varvec{s}^{(p)}_G\le \varvec{g}_G\), \(\varvec{S}^{(p)}_G\le \varvec{A}_G\). In particular, \(\varvec{s}^{(p)}_{{\mathbb {D}}}(0,\lambda )\le \varvec{g}_{{\mathbb {D}}}(0,\lambda )=|\lambda |\), \(\varvec{S}^{(p)}_{{\mathbb {D}}}(0;1)\le \varvec{A}_{{\mathbb {D}}}(0;1)=1\).
-
(e)
If \(\varvec{g}^{p+\varepsilon }_G(a,\cdot )\in \mathcal C^p(\{a\})\) for \(0<\varepsilon \ll 1\), then \(\varvec{g}^{1+\varepsilon /p}_G(a,\cdot )\in {\mathcal {S}}^{(p)}_G(a)\). Consequently, \(\varvec{s}^{(p)}_G(a,\cdot )=\varvec{g}_G(a,\cdot )\). In particular, \(\varvec{s}^{(p)}_{{\mathbb {D}}}(0,\lambda )=|\lambda |\), \(\lambda \in {\mathbb {D}}\).
-
(f)
If \(\varvec{g}^{2p}_G(a,\cdot )\in \mathcal C^{2p}(\{a\})\), then \(\varvec{S}^{(2p)}_G(a;\cdot )=\varvec{A}_G(a;\cdot )\). In particular, \(\varvec{S}^{(2p)}_{\mathbb {D}}(0;1)=1\).
-
(g)
If \(F:G\longrightarrow D\) is holomorphic, then \(v\circ F\in {\mathcal {S}}^{(p)}_G(a)\) for every \(v\in {\mathcal {S}}^{(p)}_D(F(a))\). Consequently, the family \((\varvec{s}^{(p)}_G)_G\) (resp. \((\varvec{S}^{(p)}_G)_G\)) satisfies (1.1) (resp. (1.2)).
-
(h)
The families \((\varvec{s}^{(p)}_G)_G\) and \((\varvec{S}^{(2p)}_G)_G\) are holomorphically contractible. They will be the main objects of our investigation in the sequel.
-
(i)
\(\varvec{m}_G\le \varvec{s}^{(p)}_G\le \varvec{g}_G\) and \({\varvec{\gamma }}_G\le \varvec{S}^{(2p)}_G\le \varvec{A}_G\).
3 Upper semicontinuity
It is known that for \({\mathcal {F}}\in \{{\mathcal {M}}, {\mathcal {K}}\}\) the functions \(G\times G\ni (z,w)\longmapsto d_G^{\mathcal {F}}(z,w)\) and \(G\times {\mathbb {C}}^n\ni (z,X)\longmapsto \delta _G^{\mathcal {F}}(z;X)\) are upper semicontinuous (cf. [1, Propositions 2.6.1, 2.7.1(c), 4.2.10(g,k)]). We will prove that in general the functions \(\varvec{s}^{(p)}_G(\cdot ,z^0)\) and \(\varvec{S}^{(2p)}_G(\cdot ;X^0)\) are not upper semicontinuous (Examples 3.1, 3.3).
Recall that \(\varvec{S}_G(a;\cdot )\) is a seminorm and therefore it is continuous. We do not know whether the functions \(\varvec{s}_G(a,\cdot )\), \(p\in {\mathbb {N}}\), and \(\varvec{S}^{(2p)}_G(a;\cdot )\), \(p\ge 2\), are upper semicontinuous.
Example 3.1
(cf. [1, Example 4.2.18]) Let
with
where \((a_k)^\infty _{k=1}\subset {\mathbb {D}}{\setminus }\{0\}\) is a dense subset of \({\mathbb {D}}\) and \((\lambda _k)_{k=1}^\infty \subset (0,1]\) are chosen so that \(\varphi (0,0)>-\infty \) and \(\varphi \in \mathcal C^\infty ({\mathbb {C}}\times {\mathbb {C}}_*)\), where \({\mathbb {C}}_*:={\mathbb {C}}{\setminus }\{0\}\). Note that G is a pseudoconvex Hartogs domain.
Let \(c_t:=(0,0,t)\in G\), \(t>0\), \(z^0:=(b,0,0)\in G\) with \(b\ne 0\), and let \(X^0:=(1,0,0)\). We will show that
which shows that the functions \(\varvec{s}^{(p)}_G(\cdot ,z^0)\) and \(\varvec{S}^{(2p)}_G(\cdot ;X^0)\) are not upper semicontinuous at 0.
Indeed, the function \(G\ni (z_1,z_2,z_3)\overset{v}{\longmapsto }(|z_1|e^{\varphi (z_2,z_3)})^{1+\varepsilon /p}\) belongs to \({\mathcal {S}}^{(p)}_G(c_t)\) for all \(\varepsilon >0\) and \(t>0\). Hence, \(\varvec{s}^{(p)}_G(c_t,z^0)\ge |b|e^{\varphi (0,0)}>0\). Analogously, the function \(G\ni (z_1,z_2,z_3)\overset{v}{\longmapsto }|z_1|e^{\varphi (z_2,z_3)}\) belongs to \({\mathcal {S}}^{(2p)}_G(c_t)\) for all \(t>0\). Hence, \(\varvec{S}^{(2p)}_G(c_t;X^0)\ge \limsup _{\lambda \rightarrow 0}\frac{v(c_t+\lambda X^0)}{|\lambda |}=e^{\varphi (0,t)}\ge e^{\varphi (0,0)}>0\).
On the other hand, let \(\root p \of {u}\in {\mathcal {S}}^{(p)}_G(0)\) (resp. \(\root 2p \of {u}\in {\mathcal {S}}^{(2p)}_G(0)\)). Since \({\mathbb {C}}\times \{a_k\}\times \{0\}\subset G\), we get \(u(z_1,a_k,0)={\text {const}}(k)\), \(z_1\in {\mathbb {C}}\), \(k\in {\mathbb {N}}\). Since \(\{0\}\times {\mathbb {C}}\times \{0\}\subset G\), we get \(u(0,z_2,0)={\text {const}}=u(0)=0\), \(z_2\in {\mathbb {C}}\). Thus, \(u(z_1,a_k,0)=0\), \(z_1\in {\mathbb {C}}\), \(k\in {\mathbb {N}}\). Since \(u\in \mathcal C^p(\{0\})\) (resp. \(u\in \mathcal C^{2p}(\{0\})\)), we conclude that \(u=0\) in \(U\times \{0\}\), where U is a neighborhood of (0, 0). Since \(\log u\in \mathcal {PSH}(G)\), we get \(u(z_1,z_2,0)=0\) for all \((z_1,z_2,0)\in G\). Consequently, \(\varvec{s}^{(p)}_G(0,z^0)=0\) (resp. \(\varvec{S}^{(2p)}_G(0;X^0)=0\)).
Example 3.2
In view of Example 3.1, one could expect that perhaps the families \((\varvec{s}^{(p)*}_G)_G\) and/or \((\varvec{S}^{(2p)*}_G)_G\) are holomorphically contractible, where \(\varvec{s}^{(p)*}_G:=(\varvec{s}^{(p)}_G)^*\), \(\varvec{S}^{(2p)*}_G:=(\varvec{S}^{(2p)}_G)^*\), and \({}^*\) denotes the upper semicontinuous regularization. We will prove that unfortunately they are not holomorphically contractible.
Keep the notation from Example 3.1. Let
Then \(\varvec{s}^{(p)*}_G(0,z^0)\ge \limsup _{t\rightarrow 0+}\varvec{s}^{(p)}_G(c_t,z^0)\ge |b|e^{\varphi (0,0)}>0\) and \(\varvec{S}^{(2p)*}_G(0;X^0)\ge \limsup _{t\rightarrow 0+}\varvec{S}^{(2p)}_G(c_t;X^0)\ge e^{\varphi (0,0)}>0\).
On the other hand, let \(w^0\in D\cap ({\mathbb {C}}\times {\mathbb {D}})\) and let \(\root p \of {u}\in {\mathcal {S}}^{(p)}_D(\{w^0\})\) (resp. \(\root 2p \of {u}\in {\mathcal {S}}^{(2p)}_D(\{w^0\})\)). Since \({\mathbb {C}}\times \{a_k\}\subset D\), we get \(u(z_1,a_k)={\text {const}}(k)\), \(z_1\in {\mathbb {C}}\), \(k\in {\mathbb {N}}\). Since \(\{0\}\times {\mathbb {C}}\subset D\), we get \(u(0,z_2)={\text {const}}=u(0,0)\), \(z_2\in {\mathbb {C}}\). Thus, \(u(z_1,a_k)={\text {const}}\), \(z_1\in {\mathbb {C}}\), \(k\in {\mathbb {N}}\). Since \(u\in \mathcal C^p(\{w^0\})\) (resp. \(u\in \mathcal C^{2p}(\{w^0\})\)), we conclude that \(u=0\) in \(U\times \{0\}\), where U is a neighborhood of \(w^0\). Hence, since \(\log u\in \mathcal {PSH}(G)\), we get \(u(z_1,z_2)=0\) for all \((z_1,z_2)\in D\). Consequently, \(\varvec{s}^{(p)}_D=0\) on \((D\cap ({\mathbb {C}}\times {\mathbb {D}}))\times D\) (resp. \(\varvec{S}^{(2p)}_D=0\) on \((D\cap ({\mathbb {C}}\times {\mathbb {D}}))\times {\mathbb {C}}^2\)). In particular, \(\varvec{s}^{(p)*}_D(0,(b,0))=0\) (resp. \(\varvec{S}^{(2p)*}_D(0;(1,0))=0\)) and therefore
Example 3.3
For \(n\ge 2\) and \(\alpha =(\alpha _1,\dots ,\alpha _n)\in {\mathbb {R}}^n{\setminus }\{0\}\) let
where \({\mathbb {C}}^n(\alpha ):=\{(z_1,\dots ,z_n)\in {\mathbb {C}}^n: \forall {j\in \{1,\dots ,n\}}: (\alpha _j<0 \Longrightarrow z_j\ne 0)\}\). Note that \(\varvec{D}_\alpha \) is a pseudoconvex Reinhardt domain. For \(a=(a_1,\dots ,a_n)\in \varvec{D}_\alpha \) define
Note that if \(\sigma (a)=1\), then \(r(a)=\mu (a)\).
The following results are known (cf. [1, §§ 6.2, 6.3], and [2, Theorem 1]).
\(\bullet \) If \(\alpha _1,\dots ,\alpha _n\in {\mathbb {Z}}\) are relatively prime, then
\(\bullet \) If \(\alpha \notin {\mathbb {R}}\cdot {\mathbb {Z}}^n\), then
In particular, if \(n=3\) and \(\alpha =(1,2,2)\), then \(\varvec{s}_{\varvec{D}_\alpha }((0,0,0),z)=|z^\alpha |\) and \(\varvec{s}_{\varvec{D}_\alpha }((1/k,0,0),z)=|z^\alpha |^{1/2}\), \(k\in {\mathbb {N}}\). Thus, the function \(\varvec{s}_{\varvec{D}_\alpha }(\cdot , z^0)\) is not upper semicontinuous at (0, 0, 0) for all \(z^0=(z^0_1,z^0_2,z^0_3)\in \varvec{D}_\alpha \) with \(z^0_1z^0_2z^0_3\ne 0\).
Notice that using the above effective formulas one may easily construct many other counterexamples.
Example 3.4
Keep the notation from Example 3.3. Assume that \(\alpha _1,\dots ,\alpha _n\in {\mathbb {R}}_*:={\mathbb {R}}{\setminus }\{0\}\), \(a_1\cdots a_s\ne 0\), \(a_{s+1}=\dots =a_n=0\), \(s:=n-\sigma (a)\). In particular, \(\alpha _{s+1},\dots ,\alpha _n>0\).
First observe that if \(\sigma (a)\le 1\), then \(\varvec{g}^{p+\varepsilon }_{\varvec{D}_\alpha }(a,\cdot )\in \mathcal C^p(\{a\})\) and consequently \(\varvec{s}^{(p)}_{\varvec{D}_\alpha }(a,\cdot )=\varvec{g}_{\varvec{D}_\alpha }(a,\cdot )\) (Remark 2.1(e)). Similarly, if \(\sigma (a)\le 1\), then \(\varvec{g}^{2p}_{\varvec{D}_\alpha }(a,\cdot )\in \mathcal C^\infty (\{a\})\) and consequently \(\varvec{S}^{(2p)}_{\varvec{D}_\alpha }(a;\cdot )=\varvec{A}_{\varvec{D}_\alpha }(a;\cdot )\) (Remark 2.1(f)). Problems start when \(\sigma (a)\ge 2\). We do not know effective formulas for \(\varvec{s}^{(p)}_{\varvec{D}_\alpha }(a,\cdot )\), \(p\ne 2\), and \(\varvec{S}^{(2p)}_{\varvec{D}_\alpha }(a;\cdot )\), \(p\ge 2\). To illustrate problems we discuss some particular cases.
\(\bullet \) Assume that \(\sigma (a)\ge 1\) and \(k_j:=\frac{p\alpha _j}{r(a)}\in {\mathbb {N}}\), \(j=s+1,\dots ,n\). Then
and consequently \(\varvec{g}^{2p}_{\varvec{D}_\alpha }(a,\cdot )\in \mathcal C^\infty (\{a\})\) which gives \(\varvec{S}^{(2p)}_{\varvec{D}_\alpha }(a;\cdot )=\varvec{A}_{\varvec{D}_\alpha }(a;\cdot )\). For example, if \(n=2\), \(\alpha =(1,1)\), \(a=(0,0)\), then \(\varvec{S}^{(4k)}_{\varvec{D}_\alpha }(a;X)=|X_1X_2|^{1/2}\), \(k\in {\mathbb {N}}\).
\(\bullet \) Assume that \(\sigma (a)\ge 1\) and there exists a \(j_0\in \{s+1,\dots ,n\}\) such that \(\frac{2p\alpha _{j_0}}{r(a)}\notin {\mathbb {N}}\). Then \(\varvec{S}^{(2p)}_{\varvec{D}_\alpha }(a;\cdot )\equiv 0\).
Indeed, we may assume that \(j_0=n\). Let \(r:=r(a)\) and let \(k\in {\mathbb {N}}_0\) be such that \(k<\frac{2p\alpha _n}{r}<k+1\). In view of Remark 2.1(b), we have to prove that \(u^{(2p)}(a)\equiv 0\) for all \(\root 2p \of {u}\in {\mathcal {S}}^{(2p)}_{\varvec{D}_\alpha }(a)\). Fix such a u and suppose that \(u^{(2p)}(a)(X^0)\ne 0\) for some \(X^0\ne 0\). We have
Write \(u^{(2p)}(a)(X_1^0,\dots ,X_{n-1}^0,tX_n^0)=A_dt^d+\dots +A_0\), \(t\in {\mathbb {R}}\), with \(A_d\ne 0\). We have \(|A_dt^d+\dots +A_0|\le {\text {const}}|t|^{2p\alpha _n/r}\), \(t\in {\mathbb {R}}\). Taking \(t\longrightarrow \infty \) we get \(d\le k\). On the other hand, taking \(t\longrightarrow 0\) we get \(A_d=0\); a contradiction.
For example let \(n=2\), \(\alpha =(q,1)\), \(a=(0,0)\), where
Then \(\varvec{S}^{(2p)}_{\varvec{D}_\alpha }(a;\cdot )\equiv 0\).
\(\bullet \) As a consequence, we conclude that for every \(s\in \{0,\dots ,n-2\}\) there exists a set \(C_s\) dense in \({\mathbb {R}}_*^s\times {\mathbb {R}}_{>0}^{n-s}\) (\({\mathbb {R}}_{>0}:=(0,+\infty )\)) such that for any \(\alpha \in C_s\), \(a\in \varvec{D}_\alpha \cap ({\mathbb {C}}_*^s\times \{0\}^{n-s})\), and \(p\in {\mathbb {N}}\) we have \(\varvec{S}^{(2p)}_{\varvec{D}_\alpha }(a;\cdot )\equiv 0\).
Indeed, we may put
Now we turn to discuss a special case where \(G\subset {\mathbb {C}}^n\) is a complete n-circled domain (Example 3.5).
Example 3.5
Let \(G\subset {\mathbb {C}}^n\) be a complete n-circled domain, i.e. for any \(z=(z_1,\dots ,z_n)\in G\) and \(\lambda =(\lambda _1,\dots ,\lambda _n)\in {{\overline{{\mathbb {D}}}}}^n\), the point \(\lambda \cdot z:=(\lambda _1z_1,\dots ,\lambda _nz_n)\) belongs to G.
-
(a)
Since \(\varvec{s}^{(p)}_G(0,\cdot )\le \varvec{g}_G(0,\cdot )\) and the Green function is upper semicontinuous, the function \(\varvec{s}^{(p)}_G(0,\cdot )\) is continuous at 0.
-
(b)
The function \(\varvec{s}^{(p)}_G(0,\cdot )\) is upper semicontinuous in the domain \(G{\setminus }\varvec{V}_0\), where \(\varvec{V}_0:=\{(z_1,\dots ,z_n)\in {\mathbb {C}}^n: z_1\cdots z_n=0\}\).
Indeed, let \(M:=\{a\in G: \varvec{s}^{(p)}_G(0,\cdot ) \text { is not upper semicontinuous at }a\}\). Since \(\varvec{s}^{(p)}_G(0,\cdot )\) is invariant under n-rotations (i.e. under mappings \(G\ni z\longmapsto \lambda \cdot z\in G\), \(\lambda \in {\mathbb {T}}^n\), where \({\mathbb {T}}:=\partial {\mathbb {D}}\)), the set M is also invariant under n-rotations. It is known that M is pluripolar, i.e. there exists a \(v\in \mathcal {PSH}({\mathbb {C}}^n)\), \(v\not \equiv -\infty \), such that \(M\subset v^{-1}(-\infty )\) (cf. [3, Theorem 4.7.6]). Suppose that \(a=(a_1,\dots ,a_n)\in M{\setminus }\varvec{V}_0\). Then \(v(\lambda \cdot a)=-\infty \) for all \(\lambda \in {\mathbb {T}}^n\). Consequently, by the maximum principle for plurisubharmonic functions, \(v(z_1,\dots ,z_n)=-\infty \) for all \(|z_j|\le |a_j|\), \(j=1,\dots ,n\). Hence, \(v\equiv -\infty \); a contradiction.
-
(c)
Let \(a=(0,\dots ,0,a_{s+1},\dots ,a_n)=:(0,b)\in G\cap \varvec{V}_0\), \(1\le s\le n-1\), \(a_{s+1}\cdots a_n\ne 0\). Define \(D:=\{\zeta \in {\mathbb {C}}^{n-s}: (\underset{s\times }{\underbrace{0,\dots ,0}},\zeta )\in G\}\). Note that D is a complete \((n-s)\)-circled domain with \(b\in D\). Let \({\mathfrak {h}}_D\) denote the Minkowski functional of D (\({\mathfrak {h}}_D(\zeta ):=\inf \{1/t: t>0,\;t\zeta \in D\}\), \(\zeta \in {\mathbb {C}}^{n-s}\)). Observe that \({\mathfrak {h}}_D\) is continuous (because D is \((n-s)\)-circled).
Assume that \(\varvec{s}^{(p)}_D(0,b)={\mathfrak {h}}_D(b)\). Then the function \(\varvec{s}^{(p)}_G(0,\cdot )\) is upper semicontinuous at a.
Indeed, let \(0<R<1\) and \(k>0\) be such that \({\mathfrak {h}}_D(b)<R\) and \(\Vert b\Vert <k\). Note that \(\{\zeta \in D: {\mathfrak {h}}_D(\zeta )<R,\;\Vert \zeta \Vert <k\}\subset \subset D\). Consequently, there exists an \(\varepsilon >0\) such that \(U:=\{(z',z'')\in {\mathbb {C}}^n: \Vert z'\Vert<\varepsilon ,\;{\mathfrak {h}}_D(z'')<R,\;\Vert z''\Vert <k\}\subset G\). Then for \(z=(z',z'')\in U\) we have
Hence,
Letting \(R\longrightarrow 1\) and \(k\longrightarrow +\infty \) we get \(\limsup _{z\rightarrow a}\varvec{s}^{(p)}_G(0,z)\le {\mathfrak {h}}_D(b)\).
On the other side, since the projection \({\mathbb {C}}^s\times {\mathbb {C}}^{n-s}\ni (z',z'')\longmapsto z''\in D\) is well-defined, we get \({\mathfrak {h}}_D(b)=\varvec{s}^{(p)}_D(0,b)\le \varvec{s}^{(p)}_G(0,a)\).
-
(d)
Observe that \(\varvec{s}^{(p)}_D(0,b)={\mathfrak {h}}_D(b)\) in the case where D is convex. If \(s=n-1\), then D is either a disc or the whole \({\mathbb {C}}\). Thus, if \(s=n-1\), then the function \(\varvec{s}^{(p)}_G(0,\cdot )\) is upper semicontinuous at each point \(a\in \varvec{V}_0\) of the form \(a=(0,\dots ,0,a_j,0,\dots ,0)\in G\).
-
(e)
Consequently, if \(n=2\), then the function \(\varvec{s}^{(p)}_G(0,\cdot )\) is globally upper semicontinuous.
-
(f)
If G is bounded, then the function \(\varvec{s}^{(p)}_G(0,\cdot )\) is globally upper semicontinuous.
Indeed, we proceed by induction on \(n\ge 2\). The case \(n=2\) is solved in (e). Suppose the result is true for \(n-1\ge 2\). Let \(a=(a_1,\dots ,a_n)\in G\cap \varvec{V}_0\) (see Example 3.5 (b)). We may assume that \(a_{n-1}\ne 0\), \(a_n=0\). Define \(D:=\{z'\in {\mathbb {C}}^{n-1}: (z',0)\in G\}\); D is a bounded complete \((n-1)\)-circled domain. Thus, by the inductive assumption, \(\varvec{s}^{(p)}_D(0,\cdot )\) is upper semicontinuous. Since G is bounded, for every \(0<r<1\) with \(a\in rG\) there exists an \(\varepsilon >0\) such that \((rD)\times {\mathbb {D}}(\varepsilon )\subset \subset G\). Suppose that \(G\subset {\mathbb {D}}^n(R)\) and let \(\eta >0\) be such that \(rR|\frac{z_n}{z_{n-1}}|<\varepsilon \) for \(z\in U:=\{z=(z',z_n)\in a+{\mathbb {D}}^n(\eta ): z'\in rD\}\). For \(z\in U\) consider the holomorphic mapping \(F_z:rD\longrightarrow G\), \(F_z(w):=(w,w_{n-1}\frac{z_n}{z_{n-1}})\). We have \(\varvec{s}^{(p)}_G(0,F_z(w))\le \varvec{s}^{(p)}_{rD}(0,w)=\varvec{s}^{(p)}_D(0,w/r)\). In particular, \(\varvec{s}^{(p)}_G(0,z)=\varvec{s}^{(p)}_G(0,F_z(z'))\le \varvec{s}^{(p)}_D(0,z'/r)\). Thus, \(\limsup _{z\rightarrow a}\varvec{s}^{(p)}_G(0,z)\le \limsup _{z\rightarrow a}\varvec{s}^{(p)}_D(0,z'/r)=\varvec{s}^{(p)}_D(0,a'/r)\). Letting \(r\longrightarrow 1-\) (and using once again the upper semicontinuity of \(\varvec{s}^{(p)}_D(0,\cdot )\) we get \(\limsup _{z\rightarrow a}\varvec{s}^{(p)}_G(0,z)\le \varvec{s}^{(p)}_D(0,a')\le \varvec{s}^{(p)}_G(0,a)\) (cf. Example 3.5 (c)).
Note that if \(n\ge 3\) and D is unbounded, then it is not known whether the function \(\varvec{s}^{(p)}_G(0,\cdot )\) is globally upper semicontinuous.
4 Increasing domains property
Let \((G_k)_{k=1}^\infty \) be sequence of domains in \({\mathbb {C}}^n\) such that \(G_k\nearrow G\), i.e. \(G_k\subset G_{k+1}\), \(k\in {\mathbb {N}}\), and let \(G=\bigcup _{k=1}^\infty G_k\). It is known that if \({\mathcal {F}}\in \{{\mathcal {M}}, {\mathcal {K}}\}\), then \(d_{G_k}^{\mathcal {F}}\searrow d_G^{\mathcal {F}}\) and \(\delta _{G_k}^{\mathcal {F}}\searrow \delta _G^{\mathcal {F}}\) (cf. [1, Propositions 2.7.1(a), 4.2.10(a)]). We will show that this is not true for \({\mathcal {F}}={\mathcal {S}}^{(p)}\).
Example 4.1
Let
Observe that \(\varphi _k\in \mathcal {PSH}\) and \(\varphi _k\searrow \varphi \). Moreover, \(\varphi _k\in \mathcal C^\infty (\tfrac{1}{k}{\mathbb {D}})\). Define
Note that \(G_k\) is a Hartogs domain in \({\mathbb {C}}^2\), \(k\ge 2\), and \(G_k\nearrow G\). For each \(k\ge 2\) the function \(G_k\ni (z_1,z_2)\longmapsto (|z_2|e^{\varphi _k(z_1)})^{1+\varepsilon /p}\) belongs to \({\mathcal {S}}^{(p)}_{G_k}((0,0))\), \(\varepsilon >0\). Hence, \(\varvec{s}^{(p)}_{G_k}((0,0),(0,z_2))\ge |z_2|e^{\varphi _k(0)}\ge |z_2|e^{\varphi (0)}\) for \(|z_2|<e^{-\varphi (0)}\).
Analogously, since the function \(G_k\ni (z_1,z_2)\longmapsto |z_2|e^{\varphi _k(z_1)}\) belongs to \({\mathcal {S}}^{(2p)}_{G_k}((0,0))\), we get \(\varvec{S}^{(2p)}_{G_k}((0,0);(0,X_2))\ge |X_2|e^{\varphi (0)}\) for \(X_2\in {\mathbb {C}}\) and \(k\ge 2\).
Now let \(\root p \of {u}\in {\mathcal {S}}^{(p)}_G((0,0))\). Since \(\{1/s\}\times {\mathbb {C}}\subset G\), the Liouville type theorem for subharmonic functions gives \(u(1/s,z_2)={\text {const}}(s)=:c_s\), \(s\ge 2\), \(z_2\in {\mathbb {C}}\). Since \(u(0,0)=0\), we conclude that \(c_s\longrightarrow 0\). Since u is continuous near (0, 0), we get \(u(0,z_2)=\lim _{s\rightarrow +\infty }u(1/s,z_2)=\lim _{s\rightarrow +\infty }c_s=0\), \(|z_2|\ll 1\). Hence, since \(\log u\in \mathcal {PSH}(G)\), we have \(u(0,z_2)=0\) for all \(|z_2|<e^{\varphi (0)}\). Consequently, \(\varvec{s}^{(p)}_G((0,0),(0,z_2))=0\), \(|z_2|<e^{\varphi (0)}\), and \(\varvec{S}^{(p)}_G((0,0);(0,X_2))=0\), \(X_2\in {\mathbb {C}}\).
5 Relations between \((\varvec{m}_G, \varvec{s}_G, \varvec{g}_G)\) and \(({\varvec{\gamma }}_G, \varvec{S}_G, \varvec{A}_G\))
We will discuss the following two problems. Find a pseudoconvex domain \(G\subset {\mathbb {C}}^n\), \(a\in G\), and \(z^0\in G\) (resp. \(X_0\in {\mathbb {C}}^n\)) such that
Example 5.1
If \(\alpha _1,\dots ,\alpha _n\in {\mathbb {Z}}\) are relatively prime, \(\sigma (a)\ge 2\), and \(\mu (a)\ge 2\), then the domain \(G=\varvec{D}_\alpha \) (cf. Example 3.3) is an example of a pseudoconvex domain (unfortunately, unbounded) such that \(\varvec{m}_G(a,\cdot )<\varvec{s}_G(a,\cdot )<\varvec{g}_G(a,\cdot )\) on \(\varvec{D}_\alpha {\setminus }\varvec{V}_0\). It is not known whether there exists a bounded pseudoconvex domain with this property.
Example 5.2
Let \(G\subset {\mathbb {C}}^n\) be a balanced domain (i.e. \({{\overline{{\mathbb {D}}}}}\cdot G=G\)) and let \({\mathfrak {h}}_G(z)\) be the Minkowski functional of G. It is known that \(G=\{z\in {\mathbb {C}}^n: {\mathfrak {h}}_G(z)<1\}\). Moreover, \(\varvec{g}_G(0,\cdot )={\mathfrak {h}}_G\) in G\(\Longleftrightarrow \)\(\varvec{A}_G(0;\cdot )\equiv {\mathfrak {h}}_G\)\(\Longleftrightarrow \)G is pseudoconvex \(\Longleftrightarrow \)\(\log {\mathfrak {h}}_G\in \mathcal {PSH}({\mathbb {C}}^n)\) (cf. [1, Proposition 4.2.10(b)]).
Let \(\widehat{G}\) be the convex envelope of G. It is known that \(\widehat{G}\) is also balanced and \({\mathfrak {h}}_{\widehat{G}}=\sup \{q: q:{\mathbb {C}}^n\longrightarrow [0,+\infty )\) is a \({\mathbb {C}}\)-seminorm with \(q\le {\mathfrak {h}}_G\}\). Moreover (cf. [1, Proposition 2.3.1(d)]), \({\varvec{\gamma }}_G(0;\cdot )\equiv {\mathfrak {h}}_{\widehat{G}}\). Thus, if G is pseudoconvex, then \({\varvec{\gamma }}_G(0;\cdot )={\mathfrak {h}}_{\widehat{G}}\ge \varvec{S}_G(0;\cdot )\) and hence \({\varvec{\gamma }}_G(0;\cdot )\equiv \varvec{S}_G(0;\cdot )\equiv {\mathfrak {h}}_{\widehat{G}}\le {\mathfrak {h}}_G\equiv \varvec{A}_G(0;\cdot )\). Consequently, we get the following result.
If G is a balanced pseudoconvex non-convex domain, then
In particular, the result solves the problem formulated in Example 4.2.17 from [1].
Example 5.3
Keep the notation from Example 3.1. Then
Indeed, the function \(G\ni (z_1,z_2,z_3)\overset{v}{\longmapsto }|z_1|e^{\varphi (z_2,z_3)}\) is of the class \({\mathcal {S}}^{(2p)}_G(c_t)\), which gives
Observe that the mapping \(e^{-\varphi (0,t)}{\mathbb {D}}\ni \lambda \overset{F}{\longmapsto }(\lambda ,0,t)\in G\) is well-defined. Hence, using the holomorphic contractibility, we get
Thus, \(\varvec{S}^{(2p)}_G(c_t;X^0)=\varvec{A}_G(c_t;X^0)=e^{\varphi (0,t)}\ge e^{\varphi (0,0)}>0\), \(t>0\).
Now, to get the result it suffices to show that \({\varvec{\gamma }}_G((0,0,0);X^0)=0\) and then use the continuity of \({\varvec{\gamma }}_G(\cdot ;X^0)\). For this, let \(f\in {\mathcal {O}}(G,{\mathbb {D}})\) such that \(f(0,0,0)=0\). Since \(\{0\}\times {\mathbb {C}}^2\subset G\), the Liouville theorem implies that \(f(0,\cdot ,\cdot )={\text {const}}\). Since \(f(0,0,0)=0\), we get \(f(0,\cdot ,\cdot )\equiv 0\). Since \({\mathbb {C}}\times \{a_k\}\times \{0\}\subset G\), we get \(f(\cdot ,a_k,0)={\text {const}}(k)\). Thus, \(f(\cdot ,a_k,0)\equiv 0\). Since the sequence \((a_k)_{k=1}^\infty \) is dense in \({\mathbb {D}}\), we conclude that \(f=0\) on \(({\mathbb {C}}\times {\mathbb {D}}\times \{0\})\cap G\). Thus, \(f(z_1,0,0)=0\) provided that \(|z_1|<e^{-\varphi (0,0)}\). Hence, \(f'(0,0,0)(X^0)=0\) and so \({\varvec{\gamma }}_G((0,0,0);X^0)=0\).
6 Derivative
Recall that for \({\mathcal {F}}\in \{{\mathcal {M}}, {\mathcal {K}}\}\) we have \(\delta ^{\mathcal {F}}_G(a;X)=\limsup _{\lambda \rightarrow 0}\frac{d^{\mathcal {F}}_G(a,a+\lambda X)}{|\lambda |},\quad a\in G,\;X\in {\mathbb {C}}^n\) (cf. (1.3) and (1.4)). It is an open problem whether
Observe that
so the problem is trivial if \(\varvec{S}^{(2p)}_G(a;X)=\varvec{A}_G(a;X)\).
References
Jarnicki, M., Pflug, P.: Invariant Distances and Metrics in Complex Analysis, 2nd extended edition, de Gruyter Expositions in Mathematics 9, Walter de Gruyter, xvii+861 pp (2013)
Jarnicki, M., Pflug, P.: A remark on the Sibony function. J. Math. Anal. Appl. 461, 1374–1377 (2018)
Klimek, M.: Pluripotential Theory. Oxford University Press, Oxford (1991)
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The research was partially supported by the OPUS Grant No. 2015/17/B/ST1/00996 that was financed by the National Science Centre, Poland.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Jarnicki, M., Pflug, P. Remarks on the Sibony functions and pseudometrics. Arch. Math. 113, 291–300 (2019). https://doi.org/10.1007/s00013-019-01338-1
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00013-019-01338-1