Hörmander's L² estimates on pseudoconvex domains

Written by Claude Opus 5.5 (Anthropic), October 2026. Self-checked by the writing AI. Public domain (CC0). Weighted-shell qualifications and summation: GPT-6.1 Sol (OpenAI), Ultra; author self-check, CC0.

To show that a \(\bar\partial\)-closed form is \(\bar\partial\)-exact on a whole domain, not only near each point, one needs a global method. L. Hörmander's method treats \(\bar\partial\) as a closed unbounded operator between weighted \(L^2\) spaces. An integration by parts shows that the weight \(e^{-\varphi}\) contributes the complex Hessian of \(\varphi\) to an a priori inequality; when \(\varphi\) is sufficiently strictly plurisubharmonic, the inequality gives existence of solutions with estimates, by the Riesz representation theorem. This lesson proves the resulting existence theorem on every open set of \(\mathbf C^n\) that has a smooth strictly plurisubharmonic exhaustion. The next lesson draws the consequences for holomorphic functions and coherent sheaves.

We use Plurisubharmonic functions and Stein manifolds. From analysis we use Lebesgue integration, the space \(L^2\) and its completeness, convolution with mollifiers, and distributions on open subsets of \(\mathbf R^{2n}\), as in the core course Measure and Integration, and the Riesz representation theorem for Hilbert spaces Hilbert spaces and compact operators, Theorem 2.3.

Basic references are [Demailly] and [Hörmander 1965].

The Hilbert-space inputs for Lemma 1.1 are Proposition 1.1 (Cauchy–Schwarz), the completion lemma (norm-preserving extension of bounded functionals), and Theorems 2.2–2.3 (orthogonal projection and Riesz–Fréchet representation). These providers use inner products linear in the first variable and assume no separability. The operators in Lemma 1.1 may be unbounded; Riesz representation is applied to the bounded functional constructed in its proof.

1. An abstract existence lemma

Let \(H_1,H_2,H_3\) be Hilbert spaces. A densely defined operator \(T:H_1\dashrightarrow H_2\) is a linear map defined on a dense subspace \(\operatorname{Dom}T\). Its adjoint \(T^*\) has as domain the set of \(f\in H_2\) for which \(x\mapsto\langle Tx,f\rangle\) is bounded on \(\operatorname{Dom}T\), and \(T^*f\) is the element with \(\langle Tx,f\rangle=\langle x,T^*f\rangle\) for all \(x\in\operatorname{Dom}T\). Directly from the definition, \(f\perp\operatorname{Im}T\) implies \(f\in\operatorname{Dom}T^*\) and \(T^*f=0\).

Lemma 1.1. Let \(T:H_1\dashrightarrow H_2\) and \(S:H_2\dashrightarrow H_3\) be densely defined, with \(\ker S\) closed and \(T(\operatorname{Dom}T)\subset\ker S\). Suppose that for some \(C>0\)

Then for every \(g\in\ker S\) there is \(u\in H_1\) with \(\|u\|^2\leq C^{-1}\|g\|^2\) and \(\langle u,T^*f\rangle=\langle g,f\rangle\) for all \(f\in\operatorname{Dom}T^*\).

Proof. Let \(f\in\operatorname{Dom}T^*\) and write \(f=f_1+f_2\) with \(f_1\in\ker S\) and \(f_2\perp\ker S\). As \(\operatorname{Im}T\subset\ker S\), \(f_2\perp\operatorname{Im}T\), so \(f_2\in\operatorname{Dom}T^*\) and \(T^*f_2=0\). Hence \(f_1\in\operatorname{Dom}T^*\cap\ker S\) and \(T^*f_1=T^*f\). Since \(g\in\ker S\perp f_2\), (1.1) applied to \(f_1\) gives

With inner products linear in the first slot, \(T^*f\mapsto\langle f,g\rangle\) is a well-defined linear functional on \(\operatorname{Im}T^*\) of norm at most \(C^{-1/2}\|g\|\). Extend it by continuity to the closure and by zero on the orthogonal complement; the Riesz representation theorem gives \(u\) with \(\langle T^*f,u\rangle=\langle f,g\rangle\) and \(\|u\|\leq C^{-1/2}\|g\|\). Conjugating gives the claim. \(\square\)

2. Weighted spaces and the operator \(\bar\partial\)

Let \(\Omega\subset\mathbf C^n\) be open and \(\varphi\) a real continuous function on \(\Omega\). Let \(L^2_{(0,q)}(\Omega,\varphi)\) be the Hilbert space of \((0,q)\)-forms \(f=\sum'_{|J|=q}f_J\,d\bar z_J\) with measurable coefficients and

where \(\sum'\) runs over increasing multi-indices and \(\lambda\) is Lebesgue measure. We extend the coefficients to all multi-indices so that they are alternating; then, for \(|K|=q-1\), \(f_{jK}\) is defined, and \(\sum'_J|f_J|^2=\frac1q\sum'_K\sum_j|f_{jK}|^2\).

Fix three continuous weights \(\varphi_1,\varphi_2,\varphi_3\). Let \(T:L^2_{(0,q-1)}(\Omega,\varphi_1)\dashrightarrow L^2_{(0,q)}(\Omega,\varphi_2)\) be \(\bar\partial\) in the sense of distributions, with domain the forms \(u\) for which \(\bar\partial u\) belongs to the target space, and let \(S:L^2_{(0,q)}(\Omega,\varphi_2)\dashrightarrow L^2_{(0,q+1)}(\Omega,\varphi_3)\) be defined in the same way. Both contain the smooth compactly supported forms \(\mathcal D_{(0,\bullet)}(\Omega)\) in their domains, so they are densely defined. \(\ker S\) is closed: if \(f_\nu\to f\) in \(L^2(\varphi_2)\) and \(\bar\partial f_\nu=0\), then \(\bar\partial f_\nu\to\bar\partial f\) as distributions, because \(L^2(\varphi_2)\)-convergence implies \(L^2\)-convergence on compact sets, where the weight is bounded. And \(S\circ T=0\) since \(\bar\partial^2=0\) for distributions.

Let now the weights be smooth. For \(u\in\mathcal D_{(0,q-1)}\) and \(f\in\mathcal D_{(0,q)}\), integration by parts gives \(\langle Tu,f\rangle_{\varphi_2}=\langle u,\vartheta f\rangle_{\varphi_1}\), where

Indeed, \(\langle\bar\partial u,f\rangle_{\varphi_2}=\sum'_K\sum_j\int(\partial u_K/\partial\bar z_j)\,\overline{f_{jK}}\,e^{-\varphi_2}d\lambda\) by expanding the coefficients of \(\bar\partial u\), and \(\int(\partial u_K/\partial\bar z_j)\,\overline F=-\int u_K\,\overline{\partial F/\partial z_j}\) for compactly supported smooth functions. Consequently \(\mathcal D_{(0,q)}\subset\operatorname{Dom}T^*\) with \(T^*f=\vartheta f\); and if \(f\in\operatorname{Dom}T^*\), then \(T^*f=\vartheta f\) in the sense of distributions, \(\vartheta\) being a first-order operator with smooth coefficients.

Measure and distribution inputs, at the weights used here. The programme reading Banach estimates, quotient spaces and compact parameter arguments supplies the completed Lebesgue measure construction, dominated convergence and Tonelli–Fubini, Hölder, Minkowski and Young inequalities, and completeness, smooth density and translation continuity for finite \(p\). Here is their precise weighted application.

For a real continuous weight \(\varphi\), multiplication by \(e^{-\varphi/2}\) is an isometric bijection from \(L^2(\Omega,e^{-\varphi}d\lambda)\) onto \(L^2(\Omega,d\lambda)\), with inverse multiplication by \(e^{\varphi/2}\). Both multipliers are measurable, and the norm identity follows directly from their squares. The unweighted space is complete by the cited proof, so the weighted space is complete. Applying this coefficientwise and taking the finite Hilbert sum over increasing form indices proves completeness of every \(L^2_{(0,q)}(\Omega,\varphi)\) above. This does not require a global upper or lower bound on the weight, or differentiability of \(\varphi\).

For a nonempty compact \(K\Subset\Omega\), continuity gives constants \[ 0<c_K=\min_K e^{-\varphi}\leq e^{-\varphi}\leq C_K=\max_K e^{-\varphi}<\infty. \] Thus \(\|f\|_{L^2(K)}\leq c_K^{-1/2}\|f\|_\varphi\). In particular, weighted convergence implies local unweighted \(L^2\) convergence. If \(\eta\) is a smooth test function supported in \(K\), Cauchy–Schwarz gives \[ \left|\int_K(f_j-f)\eta\,d\lambda\right| \leq c_K^{-1/2}\|f_j-f\|_\varphi\|\eta\|_{L^2(K)}\longrightarrow0. \] This defines the continuous passage to distributions used above, separately for each coefficient and each of the three weights. Distributional derivatives preserve this convergence by their definition on derivatives of test functions; see the weak-derivative limit proof.

For the ordinary weighted \(L^2\) density assertion, take an exhaustion and smooth cutoffs \(\chi_j\) as in the next section, with \(0\leq\chi_j\leq1\) and \(\chi_j\to1\) pointwise. Dominated convergence applied to \(|(1-\chi_j)f|^2e^{-\varphi}\leq |f|^2e^{-\varphi}\) gives \(\chi_jf\to f\) in weighted norm. For fixed \(j\), extend \(\chi_jf\) by zero to Euclidean space and convolve with a smooth kernel supported in the unit ball and of integral one. Restrict its scale to be at most half the distance from \(\operatorname{supp}\chi_j\) to \(\partial\Omega\) (any fixed positive bound if \(\Omega\) is the whole space). The scaled approximate-identity proof gives smooth functions converging in unweighted \(L^2\), all supported in one fixed compact subset of \(\Omega\). On that compact set the upper weight bound converts this to weighted convergence. Choosing one sufficiently small scale for each \(j\) proves density of compactly supported smooth forms. The construction is coefficientwise and uses only finitely many coefficients.

The local distribution calculus, Proposition 4.1, proves the cutoff product rule and commutation of partial derivatives directly on test functions. Consequently the constant-coefficient complex derivatives commute too; in \(\bar\partial^2\), their symmetric products cancel against the antisymmetric wedge factors. The measure reading uses \(D_j=-i\partial_j\), whereas this lesson writes ordinary partial derivatives; multiplying its derivative identities by the corresponding powers of \(i\) gives exactly the convention here. For compactly supported cutoffs, its local zero-extension and convolution argument also justifies differentiation of the mollifications. No zero extension across the boundary of a general Sobolev function is asserted.

These inputs justify the Hilbert spaces, local distributional limits, cutoff tails and ordinary mollification used in Sections 2–3. They do not replace the Friedrichs commutator argument or the simultaneous graph-norm approximation in Lemma 3.2: approximation of a function alone does not establish approximation of both operators. Those proofs, their weight relations and their hypotheses remain necessary below.

Weighted Hilbert spaces, local distribution limits and smooth density: GPT-6 Astra (OpenAI), Ultra; author self-check, CC0.

3. Density of compactly supported forms

Exact cutoff and partition inputs. An open set \(\Omega\subset\mathbf C^n=\mathbf R^{2n}\), with its usual smooth structure, is Hausdorff and second countable: intersections with rational Euclidean balls form a countable basis. Thus Local tools, Lemmas 3.A–3.B and Theorems 3.E–3.1 apply. They prove compact exhaustion, a locally finite subordinate atlas, and smooth partitions whose individual supports are compact; their flat factor is proved in Lemma 0.5 of the same reading. The source attribution in that reading is retained.

For the cutoffs below, apply its Theorem 3.1 to the closed set \(K_\nu\) and the open set \(\operatorname{int}K_{\nu+1}\). It gives \(\eta_\nu=1\) on a whole neighbourhood of \(K_\nu\), not merely pointwise on it. Its support is closed in \(\Omega\) and contained in the compact set \(K_{\nu+1}\), hence is compact. Therefore the support of its derivative misses \(K_\nu\) and lies inside \(\operatorname{int}K_{\nu+1}\). In particular the annular support claim below holds with these choices. Near any point, choose \(N\) with the point in \(\operatorname{int}K_N\). On this neighbourhood all derivatives with \(\nu\geq N\) vanish; the finitely many remaining continuous derivatives show directly that \(M\) is locally bounded. No uniform global bound for the derivatives is needed.

For the majorant, take the compactly supported locally finite partition supplied by the same theorem and discard any identically zero member. Local boundedness makes \(c_\alpha=\sup_{\operatorname{supp}\chi_\alpha}\log M\) finite: a finite subcover of each compact support supplies one bound. The sum \(\psi=\sum_\alpha c_\alpha\chi_\alpha\) is smooth because near every point it is a finite sum. At a point with \(\chi_\alpha>0\), one has \(c_\alpha\geq\log M\); nonnegativity and \(\sum\chi_\alpha=1\) give \(\psi\geq\log M\), exactly as required in (3.1).

The scalar foundation, Section 13.10, also provides an explicit coordinate cutoff equal to one on any prescribed inner closed ball. Its support is the closed outer ball, not just the set where its value is nonzero; choosing that closed ball inside \(\Omega\) gives a valid local bump. This useful explicit alternative does not replace the locally finite partition construction for an arbitrary cover.

Choose compact sets \(K_1\subset K_2\subset\cdots\) exhausting \(\Omega\), with \(K_\nu\) in the interior of \(K_{\nu+1}\), and smooth functions \(\eta_\nu\) with \(0\leq\eta_\nu\leq1\), \(\eta_\nu=1\) on \(K_\nu\) and support in the interior of \(K_{\nu+1}\). The support of \(\bar\partial\eta_\nu\) lies in \(K_{\nu+1}\setminus K_\nu\), and these sets are disjoint for different \(\nu\). Hence \(M=1+\sup_\nu\sum_k|\partial\eta_\nu/\partial\bar z_k|^2\) is locally bounded, and there is \(\psi\in\mathcal C^\infty(\Omega,\mathbf R)\) with \(e^\psi\geq M\): majorize the locally bounded function \(\log M\) by \(\sum_\alpha c_\alpha\chi_\alpha\) for a smooth partition of unity \((\chi_\alpha)\) with compact supports, with \(c_\alpha\) the supremum of \(\log M\) on the support of \(\chi_\alpha\). We fix \(\eta_\nu\) and \(\psi\), so that

From now on, for a smooth real function \(\varphi\), the three weights are

Lemma 3.1 (Friedrichs). Let \(P=\sum_ka_k\,\partial/\partial x_k+b\) be a first-order operator on an open set of \(\mathbf R^m\) with \(a_k\in\mathcal C^1\) and \(b\) continuous, let \(v\in L^2\) have compact support in it, and let \(\rho_\varepsilon(x)=\varepsilon^{-m}\rho(x/\varepsilon)\) be a mollifier. Then \(P(v*\rho_\varepsilon)-(Pv)*\rho_\varepsilon\to0\) in \(L^2\) as \(\varepsilon\to0\), where \(Pv\) is computed in the sense of distributions.

Proof. The term \(b\) causes no difficulty, since \(b(v*\rho_\varepsilon)\) and \((bv)*\rho_\varepsilon\) both tend to \(bv\). For \(P=a\,\partial/\partial x_k\), an integration by parts gives

With \(C\) a bound for \(|da|\) near the support of \(v\), \(|w_\varepsilon(x)|\leq C\int|v(x-\varepsilon y)|\,(|y|\,|\partial_k\rho(y)|+\rho(y))\,dy\), so \(\|w_\varepsilon\|_{L^2}\leq C'\|v\|_{L^2}\) uniformly in \(\varepsilon\). For smooth \(v\), \(w_\varepsilon\to0\) uniformly with supports in a fixed compact set. Since smooth compactly supported functions are dense in \(L^2\) and the operators \(v\mapsto w_\varepsilon\) are uniformly bounded, \(w_\varepsilon\to0\) for every \(v\). \(\square\)

Lemma 3.2. With the weights (3.2), \(\mathcal D_{(0,q)}(\Omega)\) is dense in \(\operatorname{Dom}T^*\cap\operatorname{Dom}S\) for the graph norm \(\|f\|_{\varphi_2}+\|T^*f\|_{\varphi_1}+\|Sf\|_{\varphi_3}\).

Proof. Cut-off. Let \(f\in\operatorname{Dom}T^*\cap\operatorname{Dom}S\). Then \(\eta_\nu f\to f\) in \(L^2(\varphi_2)\) by dominated convergence. In the sense of distributions \(S(\eta_\nu f)=\eta_\nu Sf+\bar\partial\eta_\nu\wedge f\), and by (3.1)

For \(T^*\): for \(u\in\operatorname{Dom}T\), \(\eta_\nu u\in\operatorname{Dom}T\) and \(\langle Tu,\eta_\nu f\rangle_{\varphi_2}=\langle T(\eta_\nu u),f\rangle_{\varphi_2}-\langle\bar\partial\eta_\nu\wedge u,f\rangle_{\varphi_2}\), which is bounded in \(\|u\|_{\varphi_1}\); so \(\eta_\nu f\in\operatorname{Dom}T^*\) with

and the last term has \(L^2(\varphi_1)\)-norm squared at most \(q\int_{\Omega\setminus K_\nu}|\partial\eta_\nu|^2|f|^2e^{-2\psi}e^{-\varphi+2\psi}\leq q\int_{\Omega\setminus K_\nu}|f|^2e^{-\varphi_2}\to0\), since \(|\partial\eta_\nu/\partial z_j|=|\partial\eta_\nu/\partial\bar z_j|\) for real \(\eta_\nu\), \(|\sum_j\alpha_jf_{jK}|^2\leq|\alpha|^2\sum_j|f_{jK}|^2\), and \(\sum'_K\sum_j|f_{jK}|^2=q|f|^2\). So we may assume \(f\) has compact support.

Mollification. For \(f\) with compact support, let \(f_\varepsilon=f*\rho_\varepsilon\), coefficientwise; for small \(\varepsilon\) these forms lie in \(\mathcal D_{(0,q)}\) with supports in a fixed compact set, on which all weights are bounded above and below, so all weighted norms are equivalent to unweighted ones there. Then \(f_\varepsilon\to f\), \(Sf_\varepsilon=(Sf)*\rho_\varepsilon\to Sf\) because \(\bar\partial\) has constant coefficients, and \(T^*f_\varepsilon=\vartheta f_\varepsilon\to\vartheta f=T^*f\) by Lemma 3.1 applied to the first-order operator \(\vartheta\). \(\square\)

4. The basic identity

For a smooth real function \(\varphi\) put \(\delta_jw=e^\varphi\,\partial(e^{-\varphi}w)/\partial z_j=\partial w/\partial z_j-(\partial\varphi/\partial z_j)\,w\), and write \(\varphi_{j\bar k}=\partial^2\varphi/\partial z_j\partial\bar z_k\).

Lemma 4.1 (Morrey–Kohn–Hörmander identity). For every \(f\in\mathcal D_{(0,q)}(\Omega)\),

Proof. Expanding the coefficients of \(\bar\partial f=\sum'_J\sum_j(\partial f_J/\partial\bar z_j)\,d\bar z_j\wedge d\bar z_J\) with the alternating convention, the terms with repeated indices cancel and one obtains the pointwise identity

For compactly supported smooth \(v,w\), \(\int(\partial w/\partial\bar z_k)\,\bar v\,e^{-\varphi}=-\int w\,\overline{\delta_kv}\,e^{-\varphi}\), and the commutator is \(\delta_k(\partial v/\partial\bar z_j)-\partial(\delta_kv)/\partial\bar z_j=\varphi_{k\bar j}v\). Hence

using \(\int g\,\overline{\partial h/\partial\bar z_j}\,e^{-\varphi}=-\int\delta_jg\,\bar h\,e^{-\varphi}\) and \(\overline{\varphi_{k\bar j}}=\varphi_{j\bar k}\). The sum of the left side over all \(j,k\) is unchanged when the names \(j\) and \(k\) are exchanged, so summing over \(j,k\) and \(K\) computes the integral of the last term of (4.2); substituting into the integral of (4.2) gives (4.1). \(\square\)

The first term of (4.1) is \(\|\vartheta_\varphi f\|^2_\varphi\), where \(\vartheta_\varphi\) is (2.1) with \(\varphi_1=\varphi_2=\varphi\). With the weights (3.2), (2.1) becomes \(e^{\psi}T^*f=\vartheta_\varphi f-\sum'_K\sum_j(\partial\psi/\partial z_j)f_{jK}\,d\bar z_K\), so that, with \(|\partial\psi|^2=\sum_j|\partial\psi/\partial z_j|^2\),

5. The existence theorem

Write \(\lambda_\varphi(z)\) for the smallest eigenvalue of the complex Hessian of \(\varphi\) at \(z\).

Theorem 5.1. Let \(\Omega\subset\mathbf C^n\) be open, let \(\psi\) be as in (3.1), and let \(\varphi\in\mathcal C^\infty(\Omega,\mathbf R)\) satisfy

Let \(q\geq1\). For every \(g\in L^2_{(0,q)}(\Omega,\varphi-\psi)\) with \(\bar\partial g=0\) in the sense of distributions there is \(u\in L^2_{(0,q-1)}(\Omega,\varphi-2\psi)\) with \(\bar\partial u=g\) in the sense of distributions and

Proof. For \(f\in\mathcal D_{(0,q)}\), the Hermitian form in the last term of (4.1) satisfies \(\sum'_K\sum_{j,k}\varphi_{j\bar k}f_{jK}\overline{f_{kK}}\geq\lambda_\varphi\sum'_K\sum_j|f_{jK}|^2=q\lambda_\varphi|f|^2\). Dropping the nonnegative first term on the right of (4.1) and using (4.3) and \(\|Sf\|_{\varphi_3}=\|\bar\partial f\|_\varphi\):

by (5.1). Hence \(\|T^*f\|^2_{\varphi_1}+\|Sf\|^2_{\varphi_3}\geq\|f\|^2_{\varphi_2}\) on \(\mathcal D_{(0,q)}\), and by Lemma 3.2 on all of \(\operatorname{Dom}T^*\cap\operatorname{Dom}S\). Lemma 1.1 with \(C=1\) gives \(u\) with \(\|u\|_{\varphi_1}\leq\|g\|_{\varphi_2}\), which is (5.2), and \(\langle u,T^*f\rangle_{\varphi_1}=\langle g,f\rangle_{\varphi_2}\) for all \(f\in\operatorname{Dom}T^*\). Taking \(f\in\mathcal D_{(0,q)}\), where \(T^*f=\vartheta f\) is given by (2.1), this says that \(\bar\partial u=g\) in the sense of distributions. \(\square\)

Corollary 5.2. Let \(\Omega\subset\mathbf C^n\) have a smooth strictly plurisubharmonic exhaustion \(\psi_0\), and let \(\psi\) be as in (3.1).

1. For every continuous function \(m\) on \(\Omega\) there is a convex increasing smooth function \(\chi\) such that \(\varphi=\chi\circ\psi_0\) satisfies (5.1) and \(\varphi\geq m\). If \(\varphi\) satisfies (5.1), so does \(\varphi+\theta\circ\psi_0\) for every convex nondecreasing smooth \(\theta\).
2. (Hörmander) For \(q\geq1\), every \((0,q)\)-form \(g\) with locally square integrable coefficients and \(\bar\partial g=0\) in the sense of distributions is \(\bar\partial u\) for a \((0,q-1)\)-form \(u\) with locally square integrable coefficients.

Proof. (1) Let \(\mu(t)=\min\{\lambda_{\psi_0}(z):\ \psi_0(z)\leq t\}>0\) and \(M(t)=\max\{2|\partial\psi|^2+2e^\psi:\ \psi_0\leq t\}\), using compactness of the sublevel sets. Choose \(\chi\) smooth, convex and increasing with \(\chi'(t)\mu(t)\geq M(t)\) and \(\chi(\psi_0(z))\geq m(z)\); for instance \(\chi\) with \(\chi'\) increasing and larger than \(M/\mu\) and than the slopes needed to dominate \(\max\{m:\ \psi_0\leq t\}\). By Lemma 1.2 of the previous lesson, \(\lambda_{\chi\circ\psi_0}\geq\chi'(\psi_0)\lambda_{\psi_0}\geq M(\psi_0)\), which gives (5.1). Adding \(\theta\circ\psi_0\) adds a positive semidefinite form to the Hessian, by the same lemma.

(2) In (1), choose \(\chi\) so that in addition \(\chi(\nu-1)\geq\nu-1+\log\bigl(1+\int_{\{\psi_0\leq\nu\}}|g|^2e^{\psi}\,d\lambda\bigr)\) for every integer \(\nu\) whose shell \(\{\nu-1\leq\psi_0<\nu\}\) is nonempty. On the measurable shell \(\{\nu-1\leq\psi_0<\nu\}\), which lies in the compact closed sublevel \(\{\psi_0\leq\nu\}\), we have \(e^{-\chi\circ\psi_0}\leq e^{-\chi(\nu-1)}\), so the integral of \(|g|^2e^{\psi-\varphi}\) over the shell is at most \(e^{-(\nu-1)}\). Since \(\psi_0\) is bounded below, the indices of nonempty shells are bounded below. The geometric sum of their bounds is finite, so \(g\in L^2(\varphi-\psi)\). Theorem 5.1 gives \(u\in L^2(\varphi-2\psi)\), which has locally square integrable coefficients. \(\square\)

Reference: For the weighted L² method, see [Hörmander 1965], §§1.1–1.2 and 2.1–2.2; the presentation of Section 1 follows [Demailly]. The three weights (3.2) are used here for the compact-support density argument.

Corollary 5.2: the actual smooth convex weight. The following construction gives one smooth convex weight satisfying all required Hessian and growth bounds at once. Minima and maxima use nonempty sublevels. The half-open shells used for integration need not be compact; they lie in compact closed sublevels. Empty shells impose no condition.

If \(\Omega\) is empty, both assertions are vacuous, and one may choose \(\chi(t)=t\). Otherwise choose \(c\) with \(\{\psi_0<c\}\ne\varnothing\). Its closure \(K\) is compact and contained in \(\Omega\), by the definition of exhaustion in Stein manifolds, Section 2. The compactness proof and the attained-extrema proof, Section 12.7 apply to the nonempty compact set \(K\). Thus \(b=\min_K\psi_0\) exists, and \(b<c\). Outside \(K\) one has \(\psi_0\geq c\), so \(b\) is the attained global minimum. Put \[ \rho=\psi_0-b+1\geq1,\qquad K_k=\{z\in\Omega:\rho(z)\leq k\}\quad(k\geq1). \] Each \(K_k\) is nonempty and compact: it is a closed subset of the compact closure of \(\{\rho<k+1\}\). This is the closed-sublevel argument in Proposition 2.2 of that same Stein reading. The Hessians of \(\rho\) and \(\psi_0\) are identical.

We also make the positive Hessian bound precise. For a Hermitian form \(H\) on \(\mathbf C^n\), write \(H(\xi)=\sum_{j,l}h_{j\bar l}\xi_j\bar\xi_l\), and set \(\lambda(H)=\min_{|\xi|=1}H(\xi)\). The unit sphere is compact in the original real and imaginary coordinates, by the compactness proof just cited. To identify this minimum with the smallest eigenvalue, write \(H(\xi)=\langle L\xi,\xi\rangle\), where the Hermitian operator \(L\) is defined by this form and the inner product is linear in its first argument. At a minimizing unit vector \(v\), differentiating the quotient \(H(v+tw)/|v+tw|^2\) at \(t=0\), first for \(w\) and then for \(iw\), gives \[ \operatorname{Re}\langle Lv-\lambda(H)v,w\rangle=0,\qquad \operatorname{Im}\langle Lv-\lambda(H)v,w\rangle=0. \] Thus \(Lv=\lambda(H)v\). Every other eigenvalue is the value of the same quotient at a unit eigenvector and is at least this minimum. For two Hermitian forms, finite Cauchy–Schwarz gives, for \(|\xi|=1\), \[ |H(\xi)-\widetilde H(\xi)| \leq\max_{j,l}|h_{j\bar l}-\widetilde h_{j\bar l}|\left(\sum_j|\xi_j|\right)^2 \leq n\max_{j,l}|h_{j\bar l}-\widetilde h_{j\bar l}|. \] Taking the minimum in both directions proves the same bound for \(|\lambda(H)-\lambda(\widetilde H)|\). Hence \(z\mapsto\lambda_\rho(z)\) is continuous and strictly positive. On each of the nonempty compact sets above define \[ \mu_k=\min_{K_k}\lambda_\rho>0,\qquad M_k=\max_{K_k}\bigl(2|\partial\psi|^2+2e^\psi\bigr),\qquad B_k=\max\{1,M_k/\mu_k\}. \] All these numbers are finite. For part (1), set \(D_k=B_{k+1}\) and \(V_k=\max\{0,\max_{K_{k+1}}m\}\). We will construct a smooth function \(F:\mathbf R\to\mathbf R\) with \[ F'>0,\qquad F''\geq0,\qquad F'(k)\geq D_k,\qquad F(k)\geq V_k\quad(k\geq1). \] The same construction permits any additional finite lower bounds in \(V_k\); monotonicity of the prescribed sequences is not required.

Use the flat smooth function from Section 13.10, (OC38)–(OC40): \(\eta(t)=0\) for \(t\leq0\) and \(\eta(t)=e^{-1/t}\) for \(t>0\). That proof includes the vanishing of every derivative at zero. Define \[ \sigma(t)=\frac{\eta(t)}{\eta(t)+\eta(1-t)},\qquad J=\int_0^1\sigma(s)\,ds. \] The denominator is positive for every real \(t\). Therefore \(\sigma\) is smooth, equals zero on \(( -\infty,0]\), equals one on \([1,\infty)\), and takes values in \([0,1]\). Since \(\eta'\geq0\), including at zero, \[ \sigma'(t)=\frac{\eta'(t)\eta(1-t)+\eta(t)\eta'(1-t)}{(\eta(t)+\eta(1-t))^2}\geq0. \] Continuity and \(\sigma(1)=1\) give \(0<J\leq1\).

Put \(a_0=\max\{1,D_1\}\) and \(c_0=\max\{0,V_1\}\). Suppose \(a_1,\ldots,a_{k-2}\) have been chosen for some \(k\geq2\), and define the finite sums \[ P_k(t)=a_0+\sum_{j=1}^{k-2}a_j\sigma(t-j),\qquad Q_k(t)=c_0+a_0t+\sum_{j=1}^{k-2}a_j\int_0^t\sigma(s-j)\,ds. \] Choose, explicitly, \[ a_{k-1}=\max\left\{0,\ D_k-P_k(k),\ \frac{V_k-Q_k(k)}{J}\right\}. \] These are finite nonnegative numbers determined recursively. Now put \[ D(t)=a_0+\sum_{j=1}^{\infty}a_j\sigma(t-j),\qquad F(t)=c_0+\int_0^tD(s)\,ds. \] The sum defining \(D\), and all its derivatives, is locally finite: on any bounded interval only finitely many indices have \(j<t\). It follows that \(D\) is smooth, \(D\geq a_0>0\), and \(D'\geq0\). For completeness, the difference quotient of \(\int_0^tD(s)\,ds\) differs from \(D(t)\) by at most \(\sup_{|s-t|\leq|h|}|D(s)-D(t)|\), which tends to zero as \(h\to0\). Thus \(F'=D\), and repeated differentiation proves the asserted smoothness and convexity.

At the integer \(k\), the newly added step has value \(\sigma(1)=1\) and contributes exactly \(J\) to its integral from zero to \(k\). Hence \(F'(k)=P_k(k)+a_{k-1}\geq D_k\) and \(F(k)=Q_k(k)+a_{k-1}J\geq V_k\). Terms with \(j\geq k\) vanish throughout that integration interval and at its endpoint, so future choices do not alter these inequalities. At \(k=1\) they follow from \(a_0\) and \(c_0\). This proves all the countably many value and derivative constraints, not merely each finite subset.

For \(k\leq t<k+1\), monotonicity gives \(F'(t)\geq B_{k+1}\) and \(F(t)\geq\max_{K_{k+1}}m\). If \(\rho(z)=t\), then \(z\in K_{k+1}\). The exact Hessian rule is plurisubharmonic composition, Lemma 1.2; in this one-variable case it reads \[ H_{F\circ\rho}(z)(\xi) =F'(\rho(z))H_\rho(z)(\xi)+F''(\rho(z))\left|\sum_j\rho_j(z)\xi_j\right|^2. \] It follows for every original vector \(\xi\) that \[ H_{F\circ\rho}(z)(\xi)\geq B_{k+1}\mu_{k+1}|\xi|^2 \geq M_{k+1}|\xi|^2 \geq\bigl(2|\partial\psi(z)|^2+2e^{\psi(z)}\bigr)|\xi|^2. \] Thus \(\varphi=F\circ\rho\) satisfies (5.1) and \(\varphi\geq m\). In the original exhaustion variable the required function is \(\chi(t)=F(t-b+1)\), which is smooth, increasing and convex on all of \(\mathbf R\). For any smooth convex nondecreasing \(\theta\), the same Hessian formula with \(\theta'\geq0\) and \(\theta''\geq0\) shows that adding \(\theta\circ\psi_0\) adds a positive semidefinite form. This proves the full additional assertion in part (1).

The weighted shell estimate and the local solution. Keep the same \(\rho\), \(K_k\) and derivative targets. For the form \(g\) in part (2), put \[ I_k=\int_{K_k}|g|^2e^\psi\,d\lambda<\infty. \] The finiteness follows from local square integrability, compactness and the upper bound for the continuous function \(e^\psi\) on \(K_k\). In the preceding construction replace the value targets by \[ V_k=\max\left\{0,\ \max_{K_{k+1}}m,\ k+\log(1+I_{k+1})\right\}\quad(k\geq1). \] One may take \(m=0\) for part (2); allowing the original arbitrary continuous \(m\) shows that both sets of requirements can be met simultaneously. Each target is finite, so the construction gives one weight satisfying all of them as well as (5.1).

The measurable half-open shells \(S_k=\{k\leq\rho<k+1\}\), \(k\geq1\), are disjoint and cover \(\Omega\). Each lies in the compact set \(K_{k+1}\), but the shell itself is not claimed compact. On it \(F(\rho)\geq F(k)\), and therefore \[ \int_{S_k}|g|^2e^{\psi-\varphi}\,d\lambda \leq e^{-F(k)}I_{k+1} \leq e^{-k}\frac{I_{k+1}}{1+I_{k+1}} \leq e^{-k}. \] This includes \(I_{k+1}=0\). Nonnegative countable additivity, equivalently the monotone convergence proof in Section 15.1, now gives \[ \int_\Omega|g|^2e^{\psi-\varphi}\,d\lambda =\sum_{k=1}^{\infty}\int_{S_k}|g|^2e^{\psi-\varphi}\,d\lambda \leq\sum_{k=1}^{\infty}e^{-k}=\frac1{e-1}<\infty. \] Only these nonempty-range shell indices are needed; no minimum on an empty sublevel or summation of \(e^{-k}\) over all negative integers is used. Consequently \(g\in L^2_{0,q}(\Omega,\varphi-\psi)\), with precisely the weight and weak-closedness hypotheses of Theorem 5.1. That theorem supplies \(\bar\partial u=g\) and \(u\in L^2_{0,q-1}(\Omega,\varphi-2\psi)\).

Finally, on any nonempty compact \(E\subset\Omega\), the continuous positive weight has an attained positive minimum \(c_E=\min_E e^{-\varphi+2\psi}>0\). Thus \[ \int_E|u|^2\,d\lambda\leq c_E^{-1}\int_E|u|^2e^{-\varphi+2\psi}\,d\lambda \leq c_E^{-1}\|u\|_{\varphi-2\psi}^2<\infty. \] On the empty compact set this is vacuous. This proves local square integrability without changing the distributional equation or the degree of the form. This proves the local solution assertion of Corollary 5.2.

Smooth convex weights and shell-integrability estimates: GPT-6 Astra (OpenAI), Ultra; author self-check, CC0.

6. Exercises

Exercise 6.1. Let \(T=d/dx\) on \(L^2(0,1)\) with domain the functions with square integrable derivative. Show that \(T^*=-d/dx\) with domain the functions in this space vanishing at \(0\) and \(1\), so that \(T^*\) is not the formal adjoint on its natural maximal domain.

Solution. For \(u\) in the domain of \(T\) and smooth \(f\) on \([0,1]\), \(\langle u',f\rangle=-\langle u,f'\rangle+u(1)\bar f(1)-u(0)\bar f(0)\). The functional \(u\mapsto\langle u',f\rangle\) is bounded in \(\|u\|\) exactly when the boundary terms vanish for all \(u\), that is, \(f(0)=f(1)=0\); by density this describes \(\operatorname{Dom}T^*\), and then \(T^*f=-f'\). The boundary conditions are the reason for the cut-offs of Lemma 3.2.

Endpoint traces and the full adjoint argument for Exercise 6.1. Here the domain means \[ H^1(0,1)=\{u\in L^2(0,1):u'=g\in L^2(0,1)\text{ in the sense of distributions}\}, \] and \(\langle u,f\rangle=\int_0^1u\overline f\) is linear in its first variable. The displayed boundary values in the solution must first be defined for this whole domain, not just for smooth functions.

Let \(g=u'\). Cauchy–Schwarz on the interval gives \(g\in L^1(0,1)\). Define \(F(x)=\int_0^xg(t)\,dt\), including \(x=0,1\). For \(0\leq x\leq y\leq1\), \[ |F(y)-F(x)|\leq\|g\|_2|y-x|^{1/2}, \] so \(F\) is continuous on the closed interval. The Tonelli–Fubini proof already used above applies to \(g(t)\phi'(x)\mathbf1_{t<x}\): its absolute integral is finite for every test \(\phi\in C_c^\infty(0,1)\). Consequently \[ -\int_0^1F(x)\phi'(x)\,dx =-\int_0^1g(t)\left(\int_t^1\phi'(x)\,dx\right)dt =\int_0^1g(t)\phi(t)\,dt. \] Thus \((u-F)'=0\). Weak equations, Theorem 1.1 gives \(u=F+c\) as distributions for a unique constant \(c\). Equality here is also equality almost everywhere: a locally integrable zero distribution, after a compact cutoff, has vanishing small smooth convolutions on each smaller compact set; the \(L^1\) approximate-identity proof makes the original function zero almost everywhere there. Compact exhaustion finishes this identification. We have therefore constructed a continuous representative on \([0,1]\). It is unique, since two continuous representatives equal almost everywhere on \((0,1)\) agree there and then at both endpoints by continuity. This argument uses the zero-derivative theorem, not its corollary requiring a continuous right side.

Define \(u(0),u(1)\) by that representative. Averaging \(u(x)=u(0)+\int_0^xg\), and also \(u(x)=u(1)-\int_x^1g\), yields \[ |u(0)|\leq\|u\|_2+\|u'\|_2,\qquad |u(1)|\leq\|u\|_2+\|u'\|_2. \] The endpoint maps are therefore linear and continuous for the \(H^1\) norm. They are not asserted to be continuous for the \(L^2\) norm alone.

For \(u,f\in H^1(0,1)\), write \(u(x)=u(0)+\int_0^xg\) and \(f(x)=f(0)+\int_0^xh\), with \(g=u'\), \(h=f'\). In \(\int g\overline f+\int u\overline h\), the two double integrals cover respectively \(s<t\) and \(t<s\) in the unit square. Their integrand \(g(t)\overline{h(s)}\) is absolutely integrable, and the diagonal has measure zero. Fubini therefore makes their sum \((\int g)\overline{\int h}\). Adding the constant terms proves the full-domain identity \[ \langle u',f\rangle=-\langle u,f'\rangle +u(1)\overline{f(1)}-u(0)\overline{f(0)}. \] No smooth approximation with uncontrolled endpoint values is used.

Now \(\operatorname{Dom}T=H^1(0,1)\) is dense in \(L^2(0,1)\), since it contains the compactly supported smooth functions whose density was proved in the foundation input above. If \(f\in\operatorname{Dom}T^*\), there is \(a\in L^2(0,1)\) such that \(\langle u',f\rangle=\langle u,a\rangle\) for every \(u\in H^1(0,1)\). Testing first on compactly supported smooth functions and conjugating the distribution identity gives \(f'=-a\); hence \(f\in H^1(0,1)\). The identity just proved now makes the boundary term zero for every \(u\in H^1(0,1)\). Taking \(u(x)=x\) and \(u(x)=1-x\) separately gives \(f(1)=0\) and \(f(0)=0\). Conversely those two zero traces make \(\langle Tu,f\rangle=\langle u,-f'\rangle\), bounded by \(\|u\|_2\|f'\|_2\). Thus \[ \operatorname{Dom}T^*=\{f\in H^1(0,1):f(0)=f(1)=0\},\qquad T^*f=-f'. \] This establishes both directions at the exact weak-derivative scope of the exercise and explains the domain distinction in the original solution.

Cutoff constructions and weak-derivative endpoint traces: GPT-6 Astra (OpenAI), Ultra; author self-check, CC0.

Exercise 6.2. In \(\mathbf C\) (so \(q=1\) is the only case), verify (4.1) for \(f=g\,d\bar z\) with \(g\in\mathcal D(\Omega)\).

Solution. Here \(K=\emptyset\), \(f_1=g\), \(\bar\partial f=0\), and (4.1) reads \(\int|\delta g|^2e^{-\varphi}=\int|\partial g/\partial\bar z|^2e^{-\varphi}+\int\varphi_{z\bar z}|g|^2e^{-\varphi}\) with \(\delta g=\partial g/\partial z-\varphi_zg\). This follows from \(\int(\partial g/\partial\bar z)\overline{(\partial g/\partial\bar z)}e^{-\varphi}=-\int g\,\overline{\delta(\partial g/\partial\bar z)}e^{-\varphi}=-\int g\,\overline{\partial(\delta g)/\partial\bar z+\varphi_{z\bar z}g}\,e^{-\varphi}=\int|\delta g|^2e^{-\varphi}-\int\varphi_{z\bar z}|g|^2e^{-\varphi}\), the computation of Lemma 4.1 with one index.

Exercise 6.3. Show that a weight is needed: there is a smooth compactly supported function \(g\) on \(\mathbf C\) for which \(\partial u/\partial\bar z=g\) has no solution \(u\in L^2(\mathbf C)\).

Solution. Let \(g\geq0\) be smooth with support in \(\{|z|<1\}\) and \(\int g\,d\lambda=1\), and suppose \(u\in L^2(\mathbf C)\) solves the equation. With \(v=-\frac1\pi\int g(\zeta)(\zeta-z)^{-1}d\lambda(\zeta)\), which is smooth with \(\partial v/\partial\bar z=g\) by The Dolbeault complex, Lemma 2.1, \(u-v\) satisfies \(\partial(u-v)/\partial\bar z=0\) in the sense of distributions, so it is holomorphic (next lesson, Lemma 1.2), and \(u\) is smooth. On \(\{|z|>1\}\), \(u\) is holomorphic with a Laurent expansion \(\sum_ka_kz^k\). By orthogonality of the powers on circles, \(\int_{|z|>1}|u|^2d\lambda=\sum_k|a_k|^2\int_{|z|>1}|z|^{2k}d\lambda\), and the integrals are infinite for \(k\geq-1\); so \(a_{-1}=0\). But by Green's formula, for \(R>1\), \(\int_{|z|=R}u\,dz=\int_{|z|<R}\frac{\partial u}{\partial\bar z}\,d\bar z\wedge dz=2i\int g\,d\lambda=2i\), so \(a_{-1}=\frac1{2\pi i}\int_{|z|=R}u\,dz=\frac1\pi\neq0\), a contradiction.

Exact complex-analysis inputs for Exercise 6.3. The Laurent-series solution uses the following proof locations and details for a weak \(L^2\) solution. Write \(dA=dx\,dy\) and \(\partial_{\bar z}=(\partial_x+i\partial_y)/2\). Every circle integral below is counterclockwise unless a boundary orientation is explicitly specified.

The existing boundary-flux proof, Theorem 2.1 and Corollary 2.3, gives \[ \int_{\partial D}F\,dz=2i\int_D\partial_{\bar z}F\,dA. \] It applies to the bounded discs and punctured discs used here: multiply \(F\) by a smooth compact cutoff equal to one near the closed region, so the compact-support hypothesis in that proof is satisfied. On a region with a hole the inner boundary is clockwise. The arclength identification proved at the beginning of its Corollary 2.4 identifies these boundary integrals with the circle parametrizations used below. Also \(d\bar z\wedge dz=2i\,dx\wedge dy\), which fixes the sign in the original solution.

The polar integration proof, Section 15.6, (PC1)–(PC7), includes nonnegative functions with infinite integral and absolutely integrable complex functions. In particular \[ \int_{|\eta|<r}\frac{dA(\eta)}{|\eta|}=2\pi r. \] This estimate controls the singular kernel, not its derivative. If \(F\) is \(C^1\) near the closure of a bounded disc or annulus \(D\), and \(z\in D\), apply the preceding Green identity to \(F(\zeta)/(\zeta-z)\) on \(D\setminus\overline{B(z,\epsilon)}\). Choose its cutoff with support away from \(z\). The counterclockwise small-circle integral is \[ i\int_0^{2\pi}F(z+\epsilon e^{it})\,dt\longrightarrow2\pi iF(z). \] The omitted area integral has modulus at most \(2\pi\epsilon\sup|\partial_{\bar z}F|\). Letting \(\epsilon\downarrow0\), with the inner boundary sign retained, proves \[ F(z)=\frac1{2\pi i}\int_{\partial D}\frac{F(\zeta)}{\zeta-z}\,d\zeta -\frac1\pi\int_D\frac{\partial_{\bar\zeta}F(\zeta)}{\zeta-z}\,dA(\zeta). \] Thus the needed disc and annulus cases of Cauchy–Pompeiu follow from actual programme proofs. For comparison, the separate core source is Jiří Lebl, Guide to Cultivating Complex Analysis, version 1.9 (11 July 2026), Theorem 4.6.3, pp. 115–116. Its Theorem 4.6.1 states Green’s theorem without a proof; the boundary-flux link above supplies that proof here. The core book retains its dual CC BY-SA 4.0 / CC BY-NC-SA 4.0 terms and is not embedded in this reading.

For the compactly supported smooth \(g\) in the exercise, Dolbeault, Lemma 2.1, constructs \[ v(z)=-\frac1\pi\int_{\mathbf C}\frac{g(z+\eta)}{\eta}\,dA(\eta). \] On each compact set of \(z\)’s, all translated supports lie in one bounded \(\eta\)-disc. Any fixed derivative of the numerator is uniformly bounded there. The preceding integrable-kernel bound therefore permits differentiation under the integral, repeatedly, with each derivative placed on \(g\). Applying the proved Cauchy–Pompeiu formula to \(g\) on a disc containing both \(z\) and its support makes the boundary term zero and gives \(\partial_{\bar z}v=g\). This does not differentiate \(1/(\zeta-z)\) through its singularity.

Now \(h=u-v\) is locally integrable: on a compact set, Cauchy–Schwarz turns the assumed \(L^2\) bound on \(u\) into an \(L^1\) bound, and \(v\) is smooth. Also \(\partial_{\bar z}h=0\) weakly. These are precisely the hypotheses of Stein domains, Lemma 1.2. Its proof is elementary and does not use the later applications of Hörmander’s theorem in that reading. It uses the local \(L^1\) approximate identity, the Cauchy averaging formula, and the locally uniform limit theorem. For clarity, choose \(\delta>0\) so the closed \(\delta\)-neighbourhood \(K_\delta\) of a given compact \(K\) lies inside the open domain. For sufficiently small mollifier radii the holomorphic differences satisfy \[ \sup_K|h_\epsilon-h_\eta| \leq\frac1{\pi\delta^2}\|h_\epsilon-h_\eta\|_{L^1(K_\delta)}\longrightarrow0. \] This proves the uniform-on-compact convergence used by Lemma 1.2. That lemma gives a holomorphic representative of \(h\), not a pointwise assertion about an arbitrary measurable representative. Holomorphic functions are smooth by Theorem 2.1 and Proposition 2.2. Replace \(u\) by this representative plus \(v\). It is smooth, has the same \(L^2\) class, and satisfies \(\partial_{\bar z}u=g\) pointwise. Its circle integrals are consequently well-defined.

The Laurent step and the divergent norm. Here is the annular argument at the exact exterior-domain scope of the exercise; compare Lebl’s Theorem 4.4.2, pp. 104–107. Let \(f\) be holomorphic on \(\{|z|>1\}\). For \(1<a<|z|<b\), the formula proved above, now with \(\partial_{\bar z}f=0\), gives \[ f(z)=\frac1{2\pi i}\int_{|\zeta|=b}\frac{f(\zeta)}{\zeta-z}\,d\zeta -\frac1{2\pi i}\int_{|\zeta|=a}\frac{f(\zeta)}{\zeta-z}\,d\zeta. \] In the first integral expand \((\zeta-z)^{-1}=\sum_{j\geq0}z^j\zeta^{-j-1}\); in the second expand \(-(\zeta-z)^{-1}=\sum_{j\geq0}\zeta^jz^{-j-1}\). On each smaller closed annulus the ratios \(|z|/b\) and \(a/|z|\) are uniformly less than one. The bounded circle values of \(f\) and the resulting geometric bounds justify integrating both series term by term and show absolute uniform convergence there. Their coefficients are \[ a_k=\frac1{2\pi i}\int_{|\zeta|=s}f(\zeta)\zeta^{-k-1}\,d\zeta, \qquad k\in\mathbf Z,\quad s>1. \] They do not depend on \(s\): apply Green’s formula to the holomorphic function \(f(\zeta)\zeta^{-k-1}\) between any two such circles. Thus all the expansions obtained with different \(a,b\) agree, giving \(f(z)=\sum_{k\in\mathbf Z}a_kz^k\) locally uniformly and absolutely on the entire exterior domain. Conversely, integrating any such expansion against \(\zeta^{-k-1}d\zeta\) recovers its coefficient by circle parametrization; this proves uniqueness as well as the formula for \(a_{-1}\).

For a fixed \(r>1\), the symmetric finite sums converge uniformly on \(|z|=r\). Their squared moduli therefore converge uniformly there. Direct integration of \(e^{i(k-j)t}\) gives zero for distinct integers \(k,j\), and \(2\pi\) for \(k=j\). Passing from finite sums to their limit yields \[ \int_0^{2\pi}|f(re^{it})|^2\,dt =2\pi\sum_{k\in\mathbf Z}|a_k|^2r^{2k}. \] The polar formula and nonnegative monotone convergence, or Tonelli, now give the exterior norm identity with each summand interpreted as a nonnegative integral: \[ \int_{|z|>1}|f(z)|^2\,dA =2\pi\sum_{k\in\mathbf Z}\int_1^\infty |a_k|^2r^{2k+1}\,dr. \] This is valid even when either side is infinite. If \(a_k\ne0\) and \(k\geq-1\), its integral diverges; at \(k=-1\) the divergence is logarithmic. Consequently exterior \(L^2\) integrability forces every such coefficient, in particular \(a_{-1}\), to vanish. For the smooth representative of \(u\), Green’s formula instead gives \(\int_{|z|=R}u\,dz=2i\int g\,dA=2i\) for every \(R>1\), and therefore \(a_{-1}=1/\pi\). This gives the contradiction; the sum and integral above were interchanged only for nonnegative summands.

There is also a short quantitative check that does not require the Laurent expansion. Put \(ds=R\,dt\) on the counterclockwise radius-\(R\) circle. Cauchy–Schwarz and the same fixed contour integral imply \[ 4=\left|\int_{|z|=R}u\,dz\right|^2 \leq 2\pi R\int_{|z|=R}|u|^2\,ds, \qquad \int_{|z|=R}|u|^2\,ds\geq\frac2{\pi R}. \] Integrating in \(R\) with the proved polar formula gives, for every \(T>1\), \[ \int_{1<|z|<T}|u|^2\,dA\geq\frac2\pi\log T\longrightarrow\infty. \] The contour identity therefore gives an explicit logarithmic lower bound alongside the Laurent-coefficient obstruction.

Cauchy–Pompeiu, Laurent-series and exterior-norm arguments: GPT-6 Astra (OpenAI), Ultra; author self-check, CC0.

References

• [Demailly] J.-P. Demailly, Complex Analytic and Differential Geometry, version of 21 June 2012, freely available from the author with permission to copy, modify and redistribute with credit. https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf
• [Hörmander 1965] L. Hörmander, \(L^2\) estimates and existence theorems for the \(\bar\partial\) operator, Acta Mathematica 113 (1965), 89–152. https://doi.org/10.1007/BF02391775 Freely readable article.