Abstract
We study the density of polynomials in \(H^2(E,\varphi )\), the space of square integrable functions with respect to \(\mathrm{e}^{-\varphi }\mathrm{d}m\) and holomorphic on the interior of E in \({\mathbb {C}}\), where \(\varphi \) is a subharmonic function and dm is a measure on E. We give a result where E is the union of a Lipschitz graph and a Carathéodory domain, which we state as a weighted \(L^2\)-version of the Mergelyan theorem. We also prove a weighted \(L^2\)-version of the Carleman theorem.
Similar content being viewed by others
1 Introduction
Let \(E \subset {\mathbb {C}}\) be a measurable set, m a measure on E and \(\varphi \) a measurable function, locally bounded above on E. Denote by \(L^2(E ,\varphi )\) the space of measurable functions f in E which are square integrable with respect to the measure \(\mathrm{e}^{-\varphi }\mathrm{d}m\) i.e.,
Set
where \({\mathcal {O}}(\mathring{E})\) stands for the space of holomorphic functions on the interior of E.
In this paper, we generalize the classical holomorphic approximation theorems to weighted \(L^2\)-spaces. The theory of holomorphic approximation started in 1885 with two now classical theorems: the Weierstrass theorem and the Runge theorem. The first one states that a continuous function on a bounded interval of \({\mathbb {R}}\) can be approximated arbitrarily well by polynomials for the uniform convergence on the interval. We prove the following weighted \(L^2\)-version of the Weierstrass theorem:
Theorem 1.1
Let \(\gamma \) be a Lipschitz graph over a bounded interval and \(\varphi \) a subharmonic function in a neighborhood of \(\gamma \) in \({\mathbb {C}}\). Then polynomials are dense in \(L^2(\gamma , \varphi )\).
Recall that a Carathéodory domain \(\Omega \) is a simply-connected bounded planar domain whose boundary \(\partial {\Omega }\) is also the boundary of an unbounded domain. Combined with Theorem 1.3 from [2], it leads us to the following weighted \(L^2\)-version of the Mergelyan theorem:
Theorem 1.2
Let \(\Omega \subset {\mathbb {C}}\) be a Carathéodory domain and \(\gamma \subset {\mathbb {C}}\) a Lipschitz graph with one endpoint p in \({\overline{\Omega }}\), the rest of \(\gamma \) be in the unbounded component of the complement of \({\overline{\Omega }}\). Assume that the boundary of \(\Omega \) is \({\mathcal {C}}^2\) near p. Let \(\varphi \) be a subharmonic function in a neighborhood of \({\overline{\Omega }} \cup \gamma \). Then polynomials are dense in \(H^2(\Omega \cup \gamma , \varphi )\).
Here,
where d\(\lambda _z\) is the Lebesgue measure on \(\Omega \) and ds is the arc length element.
We can even generalize to
Theorem 1.3
We use the previous notations and assumptions. If \(K_\ell \), \(\ell =1,\ldots , N\) is either a Lipschitz graph or a bounded Carathéodory domain in \({\mathbb {C}}\) as in Theorem1.2 such that \({\mathbb {C}}\setminus \left( \bigcup _{\ell =1}^N K_\ell \right) \) is connected, \(\overline{K_i}\) and \(\overline{K_j}\) have at most one common point \(\lbrace {p_{ij}}\rbrace \) and \(K_j\) is outside of the relatively compact connected component of \(\overline{K_i}^c\) for each \(i \ne j\). Let \(\varphi \) be a subharmonic function in a neighborhood of \(\bigcup _{\ell =1}^NK_\ell \). Then polynomials are dense in \(H^2(\bigcup _{\ell =1}^NK_\ell , \varphi )\) (Fig. 1).
Let \(\Gamma \) be the graph of a locally Lipschitz function over the real axis in \({\mathbb {C}}\). We may assume \(\Gamma = \{(t, \phi (t)) \}\) with \(\phi : {\mathbb {R}}\rightarrow {\mathbb {R}}\) a locally Lipschitz continuous function.
Thanks to Theorem 1.1, we prove the following weighted \(L^2\)-version of the Carleman theorem:
Theorem 1.4
Let \(\Gamma \) be the graph of a locally Lipschitz function over the real axis in \({\mathbb {C}}\) and \(\varphi \) a subharmonic function in a neighborhood of \(\Gamma \). Denote by \(\Gamma _n\), \(n\in {\mathbb {Z}}\) the part of the graph \(\Gamma \) over the interval \([n,n+1]\). Then for any \(f\in L^2(\gamma , \varphi )\) and for any positive numbers \(\varepsilon _n\), there exists an entire function F, so that for each \(n\in {\mathbb {Z}}\),
The paper is organized as follows: In Sect. 2, we give a necessary and sufficient condition in terms of Lelong number for the exponential \(\mathrm{e}^{-\varphi }\) to be integrable on an arc in \({\mathbb {C}}\). In Sect. 3, we prove Theorem 1.1. In Sect. 4, we prove Theorem 1.2. In Sect. 5, we prove Theorem 1.4. In the last section, we give an example which shows that there are no non-zero polynomials in \(L^2(\gamma , \varphi )\) for some rectifiable non-Lipschitz arc \(\gamma \) and some subharmonic function \(\varphi \).
2 “Exponential Integrability” on Arcs in \({\mathbb {C}}\)
In the following, we assume that \(\varphi \) is subharmonic on \({\mathbb {C}}\), even though it is enough to assume it to be subharmonic in a neighborhood of the given subset. This fact relies on the following lemmas.
Lemma 2.1
Let \(\mu \ge 0\) be a measure on \({\mathbb {C}}\) with finite mass on each compact set. Then there exists a subharmonic function \(\varphi \) on \({\mathbb {C}}\) so that \(\Delta \varphi = \mu \).
Proof
Set
Let \(\varphi _n (z)= \int \log |z-\zeta | d \mu _n(\zeta )\). Then \(\varphi _n(z)\) is harmonic on \(\{z: |z| < n-1\}\). Thus there exists a holomorphic function \(h_n(z)\) on \(\{z: |z| < n-1\}\) so that \(\varphi _n(z) = {\text {Re}}h_n(z)\). By using the classical Mergelyan theorem to \(h_n(z)\) on \(\{z: |z| \le n-2\}\) there exists a polynomial \(P_n(z)\) so that
Then \(\varphi (z)= \sum \limits _{n=1}^{\infty } {\text {Re}}( h_n(z) - P_n(z))\) is subharmonic on \({\mathbb {C}}\) and \(\Delta \varphi = \mu \). \(\square \)
Lemma 2.2
Let \(U \subset \subset V\) be two open sets and \(\varphi \) a subharmonic function on V. Then there exists a subharmonic function \(\psi \) on \({\mathbb {C}}\) so that \(\varphi = \psi + h\) on U, where h is harmonic on U.
Proof
Choose a smooth cut off function \(\chi : {\mathbb {C}}\rightarrow [0,1]\) so that \(\chi \equiv 1\) on a neighborhood of \({\overline{U}}\) and \(\mathrm{supp}\chi \subset V\). Then \(\mu := (\Delta \varphi ) \cdot \chi \) is a positive measure with finite mass on each compact set in \({\mathbb {C}}\). By Lemma 2.1 there is a globally defined subharmonic function \(\psi \) such that \(\Delta \psi =\mu .\) But then \(\varphi =\psi +h\) on U for some harmonic function h. \(\square \)
Since h is uniformly bounded on U, as a direct consequence, we get
Lemma 2.3
Let \(E\subset \subset U\) and \(\varphi , \psi \) be as in Lemma 2.2. Then the Hilbert spaces \(L^2(E, \varphi ) = L^2(E, \psi )\) and the norms are “equivalent”, i.e., there exist positive constants \(C_1, C_2\) so that
We prove similar statements to those in Section 2 of [5] but on an arc \(\gamma \). Those results will allow us to prove at the end of the section the local integrability of the exponential \(\mathrm{e}^{-\varphi }\) at a point \(x\in \gamma \) if and only if the Lelong number \(\nu (\varphi )(x)\) is strictly less than 1.
Definition 2.4
A function \(f: E\rightarrow {\mathbb {R}}\), \(E \subset {\mathbb {R}}\), is said to be \(L-\)Lipschitz, \(L\ge 0\), if
for every pair of points \((t_1,t_2) \in E\times E\). We say that a function is Lipschitz if it is \(L-\)Lipschitz for some L.
Lemma 2.5
(Chapter 5 of [13]) If \(f:[a,b] \rightarrow {\mathbb {R}}\) is a Lipschitz function, then f is differentiable at almost every point in [a, b] and
Let \(\gamma \) be the graph of a Lipschitz function \(y(t): [a,b] \rightarrow {\mathbb {R}}\). In the following, we say that \(\gamma \) is a Lipschitz graph if \(\gamma \) is the graph of an L-Lipschitz function y on [a, b], \(a,b<\infty \). Locally Lipschitz graph means that for each point \(p\in \gamma \), there exists a neighborhood U where the graph is Lipschitz, up to a rotation of \(\gamma \). We denote by \(|\gamma |\) the arc length of \(\gamma \), defined as follows
where \(|\gamma '(t)| = \sqrt{ 1+(y'(t))^2}\).
Let \(z_0 = (t_0, y_0)\in {\mathbb {C}}\) and \(0< \beta < 1\).
Lemma 2.6
Let \(\gamma \) be a graph, then \(|\gamma (t) - z_0| \ge |t-t_0|\) on [a, b].
Lemma 2.7
Let \(\gamma \) be a Lipschitz graph, then
Proof
By Lemma 2.6 we have
If \(t_0>b\), then
If \(t_0 <a\), then
If \(a \le t_0 \le b\), then
Thus no matter whatever the condition on \(t_0\), we have that
\(\square \)
We generalize the previous Lemma to a product
Lemma 2.8
Let \(\gamma \) be a Lipschitz graph. Suppose \(z_i\in {\mathbb {C}}\), \(\beta _i>0\), for \(i=1,\ldots , m\) and \(\sum _{i=1}^{m} \beta _i=\beta <1.\) Then
Proof
According to Corollary 2.3 in [5] we know that
By Lemma 2.7, we finally have
\(\square \)
Let \(\gamma \) be a Lipschitz graph. We take an arc length parametrization of \(\gamma \) that we denote by s. Let h be a function on \(\gamma \), define
From Lemma 2.8, we are able to prove
Theorem 2.9
Let \(\gamma \) be a Lipschitz graph. Let \(\mu \) be any nonnegative measure with total mass \(\beta <1\) on an open set U in \({\mathbb {C}}\) containing \(\gamma \). If \(\varphi (z)=\int \log |z-\zeta | d\mu (\zeta )\), then we have
where \(C_{L,\beta ,a, b} >0\) is a constant depending on \(L,\beta ,a, b.\)
Proof
Define \(\psi _n(z,\zeta )=\max \{\log |z-\zeta |,-n\}\) and
Then \(\varphi _n\) is continuous and \(\varphi _n\searrow \varphi \) pointwise. Hence \(\mathrm{e}^{-\varphi _n(z)}\nearrow \mathrm{e}^{-\varphi (z)}\). Therefore, it is enough to show
We fix n. Let \(\delta >0.\) Since \(\psi _n\) is continuous, by Lemma 2.4 in [5], there exist \(\zeta _i\in U\) such that the measure \(\mu _n=\sum _{i=1}^m \beta _i\delta _{\zeta _i}\) has total mass \(\beta <1\) and \( |{\tilde{\varphi }}_n -\varphi _n|< \delta , \) where
Hence, we get
By Lemma 2.8, we get
Hence, by combining (2.1) and (2.2) and by choosing \(\delta \) small enough, we get
\(\square \)
Corollary 2.10
Theorem 2.9 holds with possibly larger constant for subharmonics function \(\varphi \) on a neighborhood U of \(\gamma \) in \({\mathbb {C}}\) with \(\mu := \dfrac{1}{2\pi }{\Delta \varphi }_{\mid _U}\) being of total mass on U strictly less than 1.
Proof
By Riesz decomposition theorem (see for example Theorem 3.7.9 in [12]), we can decompose any subharmonic function as \(\varphi (z)=\int _U \log \vert z-\zeta \vert d\mu (\zeta )+ h(z)\) where h is harmonic. Because h is bounded on U, Theorem 2.9 gives the result with a constant depending, in addition, on h. \(\square \)
We recall now the definition of the Lelong number \(\nu (\varphi )\) of a subharmonic function \(\varphi \) at a point z in \({\mathbb {C}}\), where \(\varphi \not \equiv -\infty \), that is equal to the mass of the “Riesz” measure \(\mu =\dfrac{1}{2\pi }\Delta \varphi \) at the point z. Below is an equivalent definition given by [8, 9],
From this definition, Kiselman [8], Theorem 4.1 proves that if \(\mathrm{e}^{-\varphi }\) is locally integrable at z in \({\mathbb {C}}^n\) then \(\nu (\varphi )(z)<2n\).
We recall the following result which is a converse of the previous property in complex dimension one:
Theorem 2.11
If \(\varphi \not \equiv -\infty \) is subharmonic and \(\nu (\varphi )(z)<2\) for a point z, then \(\mathrm{e}^{-\varphi }\) is locally integrable in a neighborhood of z.
Proof
We refer to the note at top of p. 99 in Hörmander’s book [7]. This is also a consequence of Theorem 2.5 in [5]. We also refer to Proposition 7.1 in [14] for a generalization to higher dimensionsFootnote 1. \(\square \)
Now we state the previous Theorem for points in a Lipschitz graph.
Theorem 2.12
Let \(\gamma \) be a Lipschitz graph and \(\varphi \not \equiv -\infty \) a subharmonic function on \({\mathbb {C}}\). Then \(\nu (\varphi )(\gamma (t))<1\) for a point \(\gamma (t)\in \gamma \) if and only if \(\mathrm{e}^{-\varphi }\) is locally integrable in a neighborhood of \(\gamma (t)\) on \(\gamma \).
Proof
\(\Rightarrow )\) By hypothesis, \(\nu (\varphi )(\gamma (t))=\dfrac{1}{2\pi }\Delta \varphi (\gamma (t))<1\) so this is a direct consequence of Corollary 2.10.
\(\Leftarrow )\) If there exists \(z_0 =(t_0,y_0)\in \gamma \) so that \(\nu (\varphi )(z_0)\ge 1\), then we have \(\frac{1}{2\pi } \Delta \varphi (z_0) \ge 1\). Therefore we have \(\frac{1}{2\pi } \Delta \varphi - \delta _{z_0} \ge 0\) on \({\mathbb {C}}\). Hence we may find a subharmonic function \(\psi \) on \({\mathbb {C}}\) such that \(\frac{1}{2\pi }\Delta \psi = \frac{1}{2\pi }\Delta \varphi - \delta _{z_0}\). Up to a harmonic function we may write \(\varphi = \log |z-z_0| + \psi \). Thus
where \(\eta \) is a sufficiently small constant. \(\square \)
Let \(\alpha >0\). We will write \(\nu (\varphi )<\alpha \) to mean that \(\nu (\varphi )(z)<\alpha \) for all points z in the given subset. However, \(\nu (\varphi )\ge \alpha \) should be understood as there exist at least one point z such that \(\nu (\varphi )(z)\ge \alpha \).
3 Proof of Theorem 1.1
In this section, we prove Theorem 1.1 which generalizes the Weierstrass theorem to weighted \(L^2\)-spaces on Lipschitz graphs. The particularity here is to allow the weight \(\varphi \) to have singularities on the given set. We will then carefully take the local integrability of \(\mathrm{e}^{-\varphi }\) (Sect. 2) into account.
We first recall some classical results
Theorem 3.1
(Weierstrass 1883 [15]) Suppose f is a continuous function on a closed bounded interval \([a, b] \subset {\mathbb {R}}\). For each \(\varepsilon >0\) there exists a polynomial P such that
Theorem 3.2
(Lavrent’ev, 1936 [10]) Let \(K\subset {\mathbb {C}}\) be compact with \({\mathbb {C}}\setminus K\) connected. Suppose that f is continuous on K. If \(\mathring{K} = \emptyset \), then for each \(\varepsilon >0\) there exists a polynomial P such that
Theorem 3.3
(Mergelyan, 1951 [11]) Let \(K\subset {\mathbb {C}}\) be compact with \({\mathbb {C}}\setminus K\) connected. Suppose that f is continuous on K and holomorphic on \(\mathring{K}\). Then, for each \(\varepsilon >0\) there exists a polynomial P such that
In order to prove Theorem 1.1, we first prove
Theorem 3.4
Let \(\gamma \) be a Lipschitz graph over [a, b]. Suppose that \(\varphi \) is measurable on \(\gamma \) and that
Then polynomials are dense in \(L^2(\gamma , \varphi ).\)
In order to prove this theorem we need the following Lemma.
For each nonnegative real valued \(f \in L^2(\gamma , \varphi )\), define \(f_n=\min \{f,n\}.\)
Lemma 3.5
\(f_n\rightarrow f\) in \(L^2(\gamma , \varphi )\) as \(n\nearrow \infty \).
Proof
We need to prove that \(\int _{\gamma } |f-f_n|^2\mathrm{e}^{-\varphi }\mathrm{d}s\rightarrow 0.\) Now \(|f|^2\mathrm{e}^{-\varphi }\) is an \(L^1\) function and \(|f-f_n|^2\mathrm{e}^{-\varphi }\le |f|^2 \mathrm{e}^{-\varphi }.\) Moreover \(|f-f_n|^2\mathrm{e}^{-\varphi }\rightarrow 0\) a.e.. Hence by the Lebesgue dominated convergence theorem, we have that
\(\square \)
Proof of Theorem 3.4
Let f be such that \(\int _{\gamma } |f|^2\mathrm{e}^{-\varphi } \mathrm{d}s<\infty .\) To approximate f by polynomials, it suffices to consider the case when \(f\ge 0\). If P is any polynomial, then \(\Vert P-f\Vert _{L^2(\gamma , \varphi )}\le \Vert P-f_n\Vert _{L^2(\gamma , \varphi )}+\Vert f_n-f\Vert _{L^2(\gamma , \varphi )}\). So by Lemma 3.5, it suffices to approximate nonnegative bounded measurable functions \(f_n\) by polynomials.
Each bounded measurable function \(f_n\) can be uniformly approximated by simple functions on \(\gamma \). Since \(\int _{\gamma } \mathrm{e}^{-\varphi }\mathrm{d}s<\infty \), this approximation also holds in the weighted norm of \(L^2(\gamma , \varphi )\). Each simple function is the finite linear combination of characteristic functions on measurable sets in \(\gamma \), hence it suffices to approximate a characteristic function by polynomials on measurable sets in \(\gamma \). Since \(\mathrm{e}^{-\varphi }\) is \(L^1\) integrable, there exists for any \(\varepsilon >0\) a constant \(\delta >0\) so that if E is any measurable set in \(\gamma \) with Lebesgue measure \(|E|<\delta ,\) then \(\int _E \mathrm{e}^{-\varphi }\mathrm{d}s<\varepsilon .\) For any measurable set F there exists a finite union of graphs over intervals I so that \(|F\setminus I|,|I\setminus F|<\delta /2.\) Hence \(\Vert \chi _F-\chi _I\Vert _{L^2(F, \varphi )}<\varepsilon .\) Then it suffices to approximate a characteristic function of a graph over an interval in [a, b] in \(\gamma \).
Let I be a graph over an interval in \(\gamma \) and \(\delta >0\). There exists \(f\in {\mathcal {C}}^0(I,{\mathbb {C}})\) so that
So
By the Lavrent’ev theorem, it suffices to approximate a continuous function by polynomials, and then we are done. \(\square \)
Remark 3.6
Theorem 3.4 holds for non-Lipschitz rectifiable graphs.
Theorem 1.1 generalizes Theorem 3.4 by relaxing the assumption on the integrability of \(\mathrm{e}^{-\varphi }\) on \(\gamma \). To prove this generalization, we need the following result.
Theorem 3.7
Let \(\gamma \) be a Lipschitz graph over [a, b] and \(\varphi \) a subharmonic function on \({\mathbb {C}}\). Then polynomials are dense in \(L^2(\gamma , \varphi )\) if and only if the function \(\sqrt{Q}\) can be approximated arbitrarily well by polynomials in \(L^2(\gamma , \varphi )\) where Q is a polynomial vanishing at the points \(\gamma (t)\) to order \([\nu (\varphi )(\gamma (t))]\) with Lelong number \(\nu (\varphi )(\gamma (t)) \ge 1\).
Here \([\nu (\varphi )(\gamma (t))]:=\max \lbrace {m\in {\mathbb {Z}}\mid \nu (\varphi )(\gamma (t)) \ge m}\rbrace \) is also called the floor function of \(\nu (\varphi )\) at \(\gamma (t)\).
Proof
By Theorems 2.12 and 3.4, it suffices to consider \(\nu (\varphi ) \ge 1\). Then there exist finitely many points \(\gamma (t_1), \ldots , \gamma (t_n)\) such that \(\nu (\varphi )(\gamma (t_i))\ge 1\) for \(t_i\in [a,b]\), \(i=1,\dots , n\) and Q which can be expressed as \(Q(z) = \prod _{j=1}^n (z-\gamma (t_i))^{[\nu (\varphi )(\gamma (t_i))]}\). We may then choose a subharmonic function \(\psi \) so that \(\varphi = \psi + \log |Q|\) with \(\nu (\psi )<1\) on \(\gamma \).
\(\Rightarrow )\) Remark that \(\sqrt{Q}\in L^2(\gamma , \varphi )\) by Theorem 2.12:
Then by assumption, there exists a polynomial that approximates arbitrarily well \(\sqrt{Q}\).
\(\Leftarrow )\) Remark that \(f\in L^2(\gamma , \varphi )\) is equivalent to
So \(\dfrac{f}{\sqrt{Q}}\in L^2(\gamma , \psi )\) and by Theorem 3.4, for each \(\varepsilon >0\), there exists a polynomial P so that
Thus if \(\sqrt{Q}\) can be approximated arbitrarily well by polynomials in \(L^2(\gamma , \varphi )\), then f can be approximated by polynomials in \(L^2(\gamma , \varphi )\). \(\square \)
Proof of Theorem 1.1
By Theorem 2.12 again, it suffices to consider \(\nu (\varphi ) \ge 1\). Let Q and \(\psi \) be as in the proof of Theorem 3.7. We prove that for every \(\varepsilon >0\), there is a polynomial P that vanishes at \(\gamma (t_i)\) to order \([\nu (\varphi )(\gamma (t_i))]\) so that
For convenience, we look for P such as
Then (3.1) is equivalent to find some polynomial \({\widetilde{P}}\) so that
Let \(\delta >0\). Set \(g(z)= \dfrac{1}{\sqrt{Q(z)}}\) except on the arcs \(I_i\) on \(\gamma \) with length \(2\delta \) and center at \(\gamma (t_i)\). We can make g continuous and \(\vert \sqrt{Q(z)}g(z)\vert \le 1\) on such arcs of length \(2\delta \). Then
Since \(\bigcup _{i=1}^n I_i\) is a measurable set of measure \(2\delta n\) and \(\mathrm{e}^{-\psi }\in L^1_{\mathrm{loc}}\), we may choose \(\delta \) sufficiently small in order for (3.3) to be \(<\varepsilon \). Since \(g\in L^2(\gamma , \psi )\), by Theorem 3.4, there exists a polynomial A satisfying
Then by Cauchy–Schwarz and the previous estimate,
Combine (3.3) and (3.4), we may choose \({\widetilde{P}} = A\). \(\square \)
Remark 3.8
In fact we show that \({\mathcal {P}} \cap L^2(\gamma , \varphi )\) is dense in \(L^2(\gamma , \varphi )\) where \({\mathcal {P}}\) is the set of all polynomials. If \(\nu (\varphi ) \ge 1\), not all polynomials are in \(L^2(\gamma , \varphi )\), e.g. \(1\notin L^2(\gamma , \varphi )\).
4 Proof of Theorem 1.2
In this section, we give a weighted \(L^2\)-version of the Mergelyan theorem for compact sets in \({\mathbb {C}}\) that are the union of bounded Carathéodory domains and Lipschitz graphs. Recall the following theorem
Theorem 4.1
(Theorem 1.3 in [2]) Let \(\Omega \subset {\mathbb {C}}\) be a Carathéodory domain and \(\varphi \) a subharmonic function in \({\mathbb {C}}\). Then polynomials are dense in \(H^2 (\Omega , \varphi )\).
One generalization of Theorem 4.1 is the following
Theorem 4.2
Let \(\Omega _1 \subset {\mathbb {C}}\) be a Carathéodory domain, let \(\Omega _2\) be another Carathéodory domain which is inside of a bounded component of \(\overline{\Omega _1}^c\). Let \(\varphi \) be a subharmonic function in \({\mathbb {C}}\). Then polynomials are dense in \(H^2(\Omega _1\cup \Omega _2, \varphi )\) (Fig. 2).
In order to prove this theorem we need the following proposition which can be easily deduced from the proof of Proposition 1.2 in [2].
Proposition 4.3
Let \(\Omega \) be a bounded Carathéodory domain and \(\varphi \) a subharmonic function in \({\mathbb {C}}\). Then for each \(f\in H^2(\Omega , \varphi )\) there exist functions \(f_n \in H^2(\Omega _n, \varphi )\) such that \(\Vert f_n -f \Vert _{L^2( \Omega , \varphi )} \rightarrow 0\) and \(\Vert f_n\Vert _{L^2(\Omega _n \setminus \Omega , \varphi )} \rightarrow 0\) as \(n\rightarrow \infty \), where \(\Omega _n\supset {\overline{\Omega }}\) is a sequence of bounded simply-connected domains so that \(\partial \Omega _n\) converges to \(\partial \Omega \) in the sense of the Hausdorff distance.
Proof of Theorem 4.2
First we consider the case
By Theorem 4.1, for each \(\varepsilon >0\) there exists a holomorphic polynomial \(P_1\) so that
Put
Since \(\Omega _i, i =1,2\) is a Carathéodory domain, there exists a sequence \(\{\Omega _{i,n}\}\) of Jordan domains such that \(\overline{\Omega _i}\subset \Omega _{i,n}\) and \({{\overline{\Omega }}_{i,n+1}}\subset \Omega _{i,n}\) and the Hausdorff distance between \(\partial \Omega _{i,n}\) and \(\partial \Omega _i\) tends to zero as \(n\rightarrow \infty \) (see Chapter I, Section 3 of [6]). Let n be sufficiently large so that \(\Omega _1 \cap \Omega _{2,n} = \emptyset \).
Since \(g\in H^2(\Omega _1, \varphi )\), by using Proposition 4.3 on \(\Omega _1\), for each \(\varepsilon >0\) we get for large enough n functions \(g_{1,n} \in H^2{(\Omega _{1,n}, \varphi )}\) so that
and
By Theorem 4.1, for each n there exists a polynomial \(Q_n\) so that
Thus for sufficiently large n we have
and
Next we consider the case \(\nu (\varphi ) \ge 2\). Then there exist finitely many points \(z_j \in {\overline{\Omega }}_1\cup {\overline{\Omega }}_2\), \(1 \le j \le N\) so that \(\nu (\varphi ) \ge 2\) at those points. We may find a polynomial Q with zeros at those points, subharmonic function \(\psi \) satisfying \(\varphi =\psi + 2\log \vert Q \vert \) and \(\nu (\psi ) <2\) on \( {\overline{\Omega }}_1\cup {\overline{\Omega }}_2\). Let \(f\in {\mathcal {O}}(\Omega _1\cup \Omega _2) \cup L^2 (\Omega _1\cup \Omega _2, \varphi )\),
Then by the first case, \(\frac{f}{Q}\) can be approximated by a polynomial P in \(L^2(\Omega _1\cup \Omega _2, \psi )\). Thus f can be approximated by the polynomial \(P \cdot Q\) in \(L^2 (\Omega _1\cup \Omega _2, \varphi )\). \(\square \)
The proof of the main theorem of this section, Theorem 1.2, is divided into 3 cases that correspond to the locus of the zeros of the polynomial Q in the decomposition of the weight function \(\varphi =\psi +\log \vert Q\vert \). We will then need the following result:
Theorem 4.4
Let \(\Omega \) be a Carathéodory domain and \(\varphi \) a subharmonic function in \({\mathbb {C}}\). Suppose P is a polynomial with \(\int _\Omega |P|^2 \mathrm{e}^{-\varphi } \mathrm{d}\lambda < \infty \). Let \(p\in \partial \Omega \) and a disc \(\Delta \subset \Omega ^c\) with \(p\in \partial \Delta \). Then we can approximate P by \({\widetilde{P}}\) which is holomorphic on a neighborhood of \({\overline{\Omega }}\) with \({\widetilde{P}}(p) = 0\) in the norm of \(L^2(\Omega , \varphi )\).
Proof
Set \( M= \int _\Omega |P(z)|^2 \mathrm{e}^{-\varphi (z)} \mathrm{d}\lambda _z\). Then for each \(\varepsilon >0\) there exists a small neighborhood U(p) of p so that
Take \(h_n(z) = \left( \frac{p-q}{z-q} \right) ^n\), choose n sufficiently large so that
Now fix such n. Set \({\widetilde{P}}(z) =(1-h_n(z)) \cdot P(z)\), then \({\widetilde{P}}\) is holomorphic on a neighborhood of \(\Omega \) with \({\widetilde{P}}(p) = 0\) satisfying
\(\square \)
Remark 4.5
The assumption on the existence of such a disc on the boundary of \(\Omega \) is verified when \(\partial \Omega \) is \({\mathcal {C}}^2\).
Proof of Theorem 1.2
First we consider: Case 1:
Let \(f\in {\mathcal {O}}(\Omega ) \cup L^2 (\Omega \cup \gamma , \varphi )\). By Theorem 4.1, for each \(\varepsilon >0\), there exists polynomial \(P_1\) so that
By the Theorem 1.1, there exists polynomial \(P_2\) so that
Since \(\Omega \) is a Carathéodory domain, there exists a sequence \(\{\Omega _j\}\) of bounded simply-connected domains such that \({\overline{\Omega }}\subset \Omega _j\) and \({{\overline{\Omega }}_{j+1}}\subset \Omega _j\) and the Hausdorff distance between \(\partial \Omega _j\) and \(\partial \Omega \) tends to zero as \(j\rightarrow \infty \). By Corollary 2.10 we may choose j sufficiently large so that
Now fix such j. Let \(\chi : {\mathbb {C}} \rightarrow [0,1]\) be a smooth function with \( \chi \equiv 1 \) on \( \Omega _{j+1}\) and \(\chi \equiv 0\) outside of \(\Omega _{j}\). Set
Then h(z) is holomorphic on \(\Omega \), continuous on \({\overline{\Omega }} \cup \gamma \). Set
By Mergelyan approximation theorem, there exists a polynomial P so that
Then
Now we consider Case 2:
Then there exist finitely many points \(\gamma (t_i) \in \gamma \), \(t_i\in [a,b]\), \(1\le i\le N\) such that
a polynomial Q and a subharmonic function \(\psi \) satisfying \(\varphi =\psi +\log \vert Q\vert \) where Q vanishes only at \(\gamma (t_i)\) and \(\nu (\psi )<1\) on \(\gamma \). Since the polynomial Q has no zeros on \({\overline{\Omega }}\setminus \{\gamma (a)\}\), \(\nu (\psi )(z)= \nu (\varphi )(z)\) for each \(z\in {\overline{\Omega }}\setminus \{\gamma (a)\}\).
We need now to distinguish two subcases depending on the nature of \(p:=\gamma (a)\):
Subcase A: If \(t_1 =a\). Let \(f\in {\mathcal {O}}(\Omega ) \cup L^2 (\Omega \cup \gamma , \varphi )\). Then by Theorem 4.1 there exists a polynomial \(P_1\) satisfying
Since the boundary of \(\Omega \) is \({\mathcal {C}}^2\) near p, by Theorem 4.4, we may choose a holomorphic function \({\mathcal {P}}_1\) on \(\Omega _j\) satisfying
By the construction in the proof of Theorem 1.1, there exists a polynomial \(P_2\) so that
Set \(\displaystyle {M=\int _{\Omega } \mathrm{e}^{-\psi (z)}\mathrm{d}\lambda _z + \int _\gamma \mathrm{e}^{-\psi }\mathrm{d}s.}\) Since \(f\in {\mathcal {O}}(\Omega ) \cup L^2 (\Omega \cup \gamma , \varphi )\), by (4.9) we may choose j sufficiently large so that \(\gamma (t_i) \in (\Omega _j)^c\), \(2\le i \le N\) and
Now fix such j. Choose \(\chi \) be as above and
Then h and \(\frac{h}{Q}\) are holomorphic on \(\Omega _j\), continuous on \(\overline{\Omega _j} \cup \gamma \). By Theorem 3.3 there exists a polynomial G so that
Then
Subcase B: If \(\gamma (t_1)\ne p\). We may choose j sufficiently large so that \(\gamma (t_i) \in (\Omega _j)^c \cap \gamma , 1\le i \le N\) and the formula (4.11) also holds. Then the following proof is similar to subcase A.
Finally we consider: Case 3: \(\nu (\varphi )\ge 2\). There exist finitely many points \(z_j \in {\overline{\Omega }}\), \(\gamma (t_j) \in \gamma \) so that \(\nu (\varphi )\ge 2\) at those points. We may find a polynomial \(Q_1\) with zeros at those points, subharmonic function \(\psi \) satisfying \(\varphi =\psi + 2\log \vert Q_1 \vert \) and \(\nu (\psi ) <2\) on \( {\overline{\Omega }} \cup \gamma \). Let \(f\in {\mathcal {O}}(\Omega ) \cup L^2 (\Omega \cup \gamma , \varphi )\),
Then by Case 1 or 2, \(\frac{f}{Q_1}\) can be approximated by a polynomial P in \(L^2(\Omega \cup \gamma , \psi )\). Thus f can be approximated by the polynomial \(P \cdot Q_1\) in \(L^2 (\Omega \cup \gamma , \varphi )\). \(\square \)
5 Proof of Theorem 1.4
The classical Carleman approximation theorem applies to continuous function f on \({\mathbb {R}}.\) Let \(\varepsilon (x) >0\) be a continuous function.
Theorem 5.1
( [3]) There exists an entire function F so that \(|F(x)-f(x)|<\varepsilon (x)\) on \({\mathbb {R}}\).
This theorem is equivalent to the following corollary
Corollary 5.2
Let f be a continuous function on \({\mathbb {R}}\). For any \(\{ \varepsilon _n\}_{n=-\infty }^{\infty } \) with \(\varepsilon _n >0\), there exists an entire function F so that for each n,
It was pointed out by Alexander [1] that Carleman’s proof actually gives
Theorem 5.3
( [1, 3]) If \(\gamma : {\mathbb {R}}\rightarrow {\mathbb {C}}\), \(\gamma \) is a locally rectifiable curve and properly embedded, then for each continuous function f on \(\gamma \) and continuous function \(\varepsilon >0\), there exists an entire function F so that \(|F-f|<\varepsilon \) on \(\gamma \).
We prove below Theorem 1.4 which is a weighted \(L^2\)- version of this generalization for Lipschitz graphs.
Let \(\Gamma \) be the graph of a locally Lipschitz function over the real axis in \({\mathbb {C}}\). We may assume \(\Gamma = \{(t, \phi (t)) \}\) with \(\phi : {\mathbb {R}}\rightarrow {\mathbb {R}}\) a locally Lipschitz continuous function. For each \([n, n+1]\), \(\Gamma _n: =\{ (t, \phi (t))|\; n\le t \le n+1 \}\) is a Lipschitz graph.
Proof of Theorem 1.4
Case 1: \(\nu (\varphi )< 1\) on \(\Gamma \). Let \(f \in L^2(\Gamma , \varphi )\). By Theorem 3.4, there exists a continuous function \(g_n\) on \(J_n: = \{(t,\phi (t)) | \; n-1 \le t \le n+2 \}\) so that
Choose a partition of unity \(\{ \chi _n \}_{n\in {\mathbb {Z}}}\) of \(\Gamma \) so that \(\chi _n \ge 0\) on \(\Gamma \), \(\chi _n =0\) outside of \(J_n\) and \(\sum _n\chi _n =1 \) on \(\Gamma \). Let \(g = \sum _n \chi _n g_n\). Then g is continuous on \(\Gamma \) and for each n,
By using Theorem 5.3 on the continuous function g of \(\Gamma \), we can find an entire function F so that for each n
Thus
Hence we have
Now we consider: Case 2: \(\nu (\varphi ) \ge 1\). We may list the points \(\{\Gamma (t_j)\}_j\) with \(\nu (\varphi )(\Gamma (t_j)) \ge 1\), where \(\Gamma (t_j) = (t_j, \phi (t_j))\). Then there exists an entire function Q which vanishes at each \(\Gamma (t_j)\) to exact order \([\nu (\varphi )(\Gamma (t_j))]\). We may define \(\sqrt{Q}\) to be continuous on \(\Gamma \), without loss of generality we may set \(Q(z)= \prod \limits _j (z-\Gamma (t_j))^{[\nu (\varphi )(\Gamma (t_j))]} \mathrm{e}^{p_j(z)}\), where \(p_j(z)\) are entire functions. Then there exists a subharmonic function \(\psi \) such that \(\varphi = \psi + \log |Q|\) with \(\nu (\psi ) < 1\) on \(\Gamma \). Let \(f \in L^2(\Gamma , \varphi )\). Then \( \frac{f}{\sqrt{Q}} \in L^2( \Gamma , \psi ). \) By Case 1, there exists an entire function F so that for each n
Thus it suffices to find an entire function H vanishing at \(\Gamma (t_j)\) to order \([\nu (\varphi )(\Gamma (t_j))]\) so that for each n
We look for H for convenience as
The estimate (5.3) is then equivalent to find an entire function \({\widetilde{H}}\) so that
Let \(\delta _j>0\). Set \(g(z)= \dfrac{1}{\sqrt{Q(z)}}\) except on arcs \(\Gamma ^j\) of \(\Gamma \) with length \(2\delta _j\) and center at \(\Gamma (t_j)\). We can make g continuous and \(\vert \sqrt{Q} \cdot g\vert \le 1\) on such arcs of length \(2\delta _j\). Then
Since \(\bigcup \limits _{\Gamma (t_j) \in J_n} \Gamma ^j\) is a measurable set and \(\mathrm{e}^{-\psi }\in L^1_{\mathrm{loc}}\), we may choose \(\delta _j\) sufficiently small in order to
Since g is continuous on \(\Gamma \), by the classical Carleman approximation theorem there exists an entire function A satisfying for each n
Then by Cauchy-Schwarz and the previous estimate, for each n,
By taking \({\widetilde{H}} =A\), we get (5.4) and then (5.3). Hence there exist entire functions F, Q, A so that for each n
\(\square \)
6 Rectifiable Non-Lipschitz Arcs
Here we construct a rectifiable non-Lipschitz arc \(\gamma \) and a subharmonic function \(\varphi \) in a neighborhood of \(\gamma \) so that the conclusion of Theorem 1.1 does not hold. To find such an arc, we first look at the vertical arcs \(\gamma _a=\{(a,it),|t|\le a\}\). Let \(0<\alpha <1, z_0=0.\) We then notice that
and
Since
we get that \(\int _{\gamma _a}\frac{\mathrm{d}s}{|z-z_0|^\alpha } \sim a^{1-\alpha }\) uniformly in \(\alpha .\)
Let \(b_n\in [0,1], n=1,2,3,\cdots , \) be a decreasing sequence which tends to 0. We will fix \(b_n\) later. We remark that by setting \(\varphi (z) = \sum \limits _{n=1}^\infty \alpha _n \log \left| \frac{z-b_n}{2} \right| \), where \(\alpha _n=\frac{1}{n^3}\) a rapidly then \(\varphi \) is subharmonic on \(\Delta (0,2-b_1)\): \(\varphi \) is the limit of the decreasing sequence \(\{ \varphi _k\}\) of subharmonic functions, \(\varphi _k = \sum \limits _{n=1} ^k \alpha _n \log \left| \frac{z-b_n}{2} \right| \). Now let’s build a rectifiable non-Lipschitz arc \(\gamma \) such that
-
(1)
\(| \gamma | < \infty \);
-
(2)
\(\int _{\gamma _n} \mathrm{e}^{-\varphi }\mathrm{d}s = \infty \).
Here \(\gamma _n\) is a curve with endpoints \((b_{n+1},0)\) and \((b_n,0)\) and \(\gamma \) consists of the union of the \(\gamma _n\) and the origin. Define \(c_n\) so that
We define \(\{b_n\}\) by the following conditions:
add the requirement that \(b_n \rightarrow 0.\) Then we have that
Now define \(b_n^k\in [ b_{n+1}, b_n]\) satisfying
Then \(b_n^k\rightarrow b_{n+1}\) as \(k\rightarrow \infty \) and \(b_n^1=b_n.\) Finally, we define \(\gamma _n = \left( \cup _{k=1}^\infty \gamma _n^k \right) \cup T_n\cup S_n \), where
\(T_n= \cup _{\ell =0}^\infty \{y_x-b_{n+1}, b_n^{2\ell +2}\le x\le b_n^{2\ell +1}\}\) and \(S_n=\cup _{\ell \ge 1}^\infty \{y=0, b_n^{2\ell +1}\le x\le b_n^{2\ell }\}\) which connect the \(\gamma _n^k\) making \(\gamma _n\) an arc. Then \(|\gamma _n^k| = (b_n^k-b_{n+1})\), \(|T_n| < \sqrt{2}(b_n-b_{n+1})=\sqrt{2}(b_n^1-b_{n+1})\) and \(|S_n|<b_n-b_{n+1}\). Then, we have
Since
we know that
Thus \(\gamma \) is a rectifiable non-Lipschitz arc.
On the other hand, we have
Now we will prove that polynomials are not dense in \(L^2(\gamma , \varphi )\). By contradiction, for each \(f\in L^2(\gamma , \varphi )\), if there exists a sequence of polynomials \(P_N\) so that
then by (6.3) we have \(P_N(b_n) = 0\) for any n if N is sufficiently large. Since \(b_n \rightarrow 0\) by uniqueness property of holomorphic function we know \(P_N\equiv 0\). Thus \(\int _{\gamma }|f|^2\mathrm{e}^{-\varphi }\mathrm{d}s =0\). That is \(f =0\) a.e. on \(\gamma \). Thus \(L^2(\gamma , \varphi ) = \{0\}\). On the other hand \(f(z) := \mathrm{e}^{\frac{\varphi }{2}} =\sqrt{\Pi \left| \frac{z-b_n}{2} \right| ^{\alpha _n}} \in L^2(\gamma , \varphi )\). This is a contradiction.
Remark 6.1
In this example, there are no non-zero polynomials in \(L^2(\gamma , \varphi )\) and polynomials are not dense in \(L^2(\gamma , \varphi )\). The key to this example is that Theorem 2.12 does not hold for rectifiable non-Lipschitz arcs. However, we don’t know if there exists a rectifiable non-Lipschitz arc \(\gamma \) and a subharmonic function \(\varphi \) so that all the polynomials are in \(L^2(\gamma , \varphi )\) but not dense in it.
Notes
Many authors refer to this result as “Skoda’s exponential integrability”
References
Alexander, H.: A Carleman theorem for curves in \({\mathbb{C}}^n\). Math. Scand. 45(1), 70–76 (1980)
Biard, S., Fornæss, J.E., Wu, J.: Weighted \(L^2\) polynomial approximation in \({\mathbb{C}}\). Trans. Am. Math. Soc. 373(2), 919–938 (2020)
Carleman, T.: Sur un théorème de Weierstrass. Ark. Mat. 20B, 1–5 (1927)
Donnelly, H., Fefferman, C.: \(L^2\)-cohomology and index theorem for the Bergman metric. Ann. Math. 118, 593–618 (1983)
Fornæss, J.E., Wu, J.: Weighted approximation in \({\mathbb{C}}\). Math. Z. 294(3), 1051–1064 (2019)
Gaier, D.: Lectures on Complex Approximation. Birkhäuser, Boston, Basel, Stuttgart (1980)
Hörmander, L.: An Introduction to Complex Analysis in Several Variables. North Holland, Amsterdam (1991)
Kiselman, C.O.: Densité des fonctions plurisousharmoniques. Bull. Soc. Math. France 107(3), 295–304 (1979)
Kiselman, C.O.: Attenuating the singularities of plurisubharmonic functions. Ann. Polon. Math. 60(2), 173–197 (1994)
Lavrent’ev, M.A.: Sur les fonctions d’une variable complexe représentables par des séries de polynomes. Bull. Am. Math. 44, 22 (1936)
Mergelyan, S.N.: On a theorem of M.A. Lavrent’ev. Dokl. Akad. Nauk SSSR. 77(4) (1951) (Russian); Am. Math. Soc. Translation, 86, 7 pp, 1953 (Translated)
Ransford, T.: Potential Theory in the Complex Plane. Cambridge University Press, Cambridge (1995)
Royden, H.: Real Analysis, 2nd edn. Macmillan, New York (1968)
Skoda, H.: Sous-ensembles analytiques d’ordre fini ou infini dans \({\mathbb{C}}^n\). Bull. Soc. Math. France 100, 353–408 (1972)
Weierstrass, K.: Über die analytische Darstellbarkeit sogenannter willkü rlicher Functionen einer reellen Veränderlichen. Sitzungsberichte der Königlich Preussischen Akademie der Wissenschaften zu Berlin 2, 633–639 (1885)
Acknowledgements
Open Access funding provided by NTNU Norwegian University of Science and Technology. The first author was supported in part by Rannis-Grant 152572-051. The second author was supported by the Norwegian Research Council Grant 240569. The third author was supported by the NSFC Grant 11601120 and Norwegian Research Council Grant 240569.
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.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Biard, S., Fornæss, J.E. & Wu, J. Weighted \(L^2\) Version of Mergelyan and Carleman Approximation. J Geom Anal 31, 3889–3914 (2021). https://doi.org/10.1007/s12220-020-00418-x
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12220-020-00418-x