06.10.04 · riemann-surfaces / several-variables

The ∂̄-equation and Hörmander's L² estimates

shipped3 tiersLean: none

Anchor (Master): Hörmander 1965 (Acta Math. 113, originator); Kohn 1963/64; Morrey 1958; Andreotti–Vesentini 1965; Krantz Ch. 4; Hörmander Ch. 4

Intuition Beginner

In one complex variable you can always solve the equation that asks for a function whose failure-to-be-holomorphic is a prescribed pattern. Holomorphic functions are the ones with no failure at all; the inhomogeneous version prescribes the failure and asks you to build a function matching it. In one variable this is always solvable, which is why every planar region is the natural home of some holomorphic function.

In several variables the same equation, written here as "solve for $u$ given the target $f$ ", is the engine of the whole subject. The target $f$ records a pattern of non-holomorphy spread across all the complex directions, and the unknown $u$ is the function realising it. The catch is that the target $f$ must itself be consistent — its own directional failures have to fit together — and even then a solution exists only when the region is shaped correctly.

The shape that works is pseudoconvexity. Hörmander's theorem says: on a pseudoconvex region you can always solve for $u$ , and you can keep the size of $u$ under control by a measured amount. That size control is what turns an abstract solution into a working tool.

Visual Beginner

Think of a holomorphic function as a perfectly still pond surface: no ripples in any direction. There is a meter that measures the ripples — the failure-to-be-holomorphic meter. A holomorphic function reads zero everywhere. The inhomogeneous equation hands you a ripple pattern $f$ and asks you to find a pond surface $u$ whose ripples are exactly that pattern.

Two things can go wrong. First, not every ripple pattern is achievable: ripples interact across directions, and only the self-consistent patterns — the ones passing a compatibility check — can come from an actual surface. Second, even a consistent pattern needs the right container. A leaky container lets the surface run off to the edge before you can pin it down.

The picture below shows a consistent ripple pattern over a pseudoconvex container. A weight, drawn as shading that thickens near the wall, acts like a cost that penalises any surface for piling up near the boundary. Minimising the total cost while matching the ripples produces the smallest, best-behaved solution. The upward-curving wall of a pseudoconvex region is exactly what makes that minimisation succeed.

Worked example Beginner

Take the simplest one-variable instance, where you can watch the machine run with actual numbers. Suppose you want a function $u$ on the unit disc whose ripple reading is the constant pattern $f = 1$ — at every point the ripple meter shows the value $1$ .

A function with constant ripple reading $1$ is $u = \overset{z}{ˉ}$ , the complex conjugate of the coordinate. Check it directly: the ripple meter applied to $\overset{z}{ˉ}$ returns $1$ at every point, because conjugation is the purest possible non-holomorphy and $\overset{z}{ˉ}$ carries one unit of it uniformly. So $u = \overset{z}{ˉ}$ solves the equation.

Now measure its size on the disc. The size is the average of $∣ u ∣^{2} = ∣ \overset{z}{ˉ} ∣^{2} = ∣ z ∣^{2}$ over the disc. Points near the centre contribute almost nothing; points near the rim contribute up to $1$ . The average works out to $1/2$ . So the solution has controlled size $1/2$ , comfortably finite.

What this tells us: the equation was solvable, and the solution did not blow up. Hörmander's theorem promises exactly this in every dimension and on every pseudoconvex region — solvability plus a finite, predictable size bound — even when, unlike here, no solution can be written down by hand.

Check your understanding Beginner

Exercise (easy, multiple choice).

Hörmander's theorem guarantees that the equation "solve for $u$ given the target $f$ " is solvable with controlled size when:

A. The target $f$ is consistent and the region is pseudoconvex B. The target $f$ is any pattern at all, on any region C. The region is bounded, regardless of its shape D. The function $u$ is already known to be holomorphic

Hint

Two ingredients are needed: a compatibility condition on the target, and the right shape for the container.

Answer

A. Feedback-correct: the target must pass its compatibility check (the directional failures must fit together) and the region must be pseudoconvex; then a size-controlled solution exists. Feedback-wrong: B drops both conditions; C names boundedness, which is neither necessary nor sufficient; D describes the homogeneous case, which is not what the theorem solves.

Formal definition Intermediate+

Throughout, $Ω \subseteq C^{n}$ is open and $φ \in C^{2} (Ω)$ is a real weight. Coordinates are $z = (z_{1}, \dots, z_{n})$ ; write $\partial_{j} = \partial / \partial z_{j}$ and $\overset{ˉ}{\partial}_{j} = \partial / \partial \overset{z}{ˉ}_{j}$ . The Cauchy–Riemann operator on functions is $\overset{ˉ}{\partial} u = \sum_{k = 1}^{n} (\overset{ˉ}{\partial}_{k} u) d \overset{z}{ˉ}_{k}$ , a $(0, 1)$ -form. A function is holomorphic exactly when $\overset{ˉ}{\partial} u = 0$ .

The inhomogeneous $\overset{ˉ}{\partial}$ -equation. Given a $(0, 1)$ -form $f = \sum_{k} f_{k} d \overset{z}{ˉ}_{k}$ , one seeks $u$ with $$ \bar\partial u = f, \qquad\text{i.e.}\qquad \bar\partial_k u = f_k \ \ (k = 1, \dots, n). $$ This is an overdetermined system: $n$ equations for one unknown $u$ . Differentiating, $\overset{ˉ}{\partial}_{j} f_{k} = \overset{ˉ}{\partial}_{j} \overset{ˉ}{\partial}_{k} u = \overset{ˉ}{\partial}_{k} f_{j}$ , so a necessary compatibility condition is $$ \bar\partial f = 0, \qquad\text{i.e.}\qquad \bar\partial_j f_k = \bar\partial_k f_j \ \ \text{for all } j, k. $$ A form with $\overset{ˉ}{\partial} f = 0$ is called $\overset{ˉ}{\partial}$ -closed. The content of the existence theory is that on a pseudoconvex $Ω$ this necessary condition is also sufficient.

Weighted $L^{2}$ spaces. Let $d V$ be Lebesgue measure on $C^{n} ≅ R^{2 n}$ . Define the weighted spaces of functions and of $(0, 1)$ -forms, $$ L^2(\Omega, \varphi) = \Big{ u : \lVert u\rVert_\varphi^2 = \int_\Omega |u|^2 e^{-\varphi}, dV < \infty \Big}, \qquad L^2_{(0,1)}(\Omega, \varphi) = \Big{ f : \lVert f\rVert_\varphi^2 = \sum_k \int_\Omega |f_k|^2 e^{-\varphi}, dV < \infty \Big}, $$ with inner products $⟨ u, v ⟩_{φ} = \int u \overset{v}{ˉ} e^{- φ} d V$ and $⟨ f, g ⟩_{φ} = \sum_{k} \int f_{k} \overset{g}{ˉ}_{k} e^{- φ} d V$ . These are Hilbert spaces 02.11.08. The weight $e^{- φ}$ penalises mass where $φ$ is large; choosing $φ$ strongly plurisubharmonic is what supplies the positivity in the estimate below.

$\overset{ˉ}{\partial}$ and its weighted adjoint. Regard $\overset{ˉ}{\partial}$ as a closed, densely defined operator $T : L^{2} (Ω, φ) \to L_{(0, 1)}^{2} (Ω, φ)$ with domain the $u$ for which $\overset{ˉ}{\partial} u$ (in the distributional sense) lies in $L_{(0, 1)}^{2} (Ω, φ)$ . Its Hilbert-space adjoint $\overset{ˉ}{\partial}_{φ}^{*} = T^{*}$ is computed by integration against the weight. For a smooth, compactly supported $(0, 1)$ -form $u = \sum_{k} u_{k} d \overset{z}{ˉ}_{k}$ , $$ \bar\partial^{*}\varphi u = -\sum{k=1}^n \delta_k u_k, \qquad \delta_k = e^{\varphi}, \partial_k\big(e^{-\varphi}, \cdot,\big) = \partial_k - (\partial_k\varphi),\cdot, $$ the formal adjoint of $\overset{ˉ}{\partial}_{k}$ in the weighted inner product. The operator $δ_{k}$ differs from $\partial_{k}$ by the zeroth-order term $- \partial_{k} φ$ , and it is this term that injects $φ$ into the commutators driving the basic identity. Notation: $φ_{j \overset{ˉ}{k}} = \partial_{j} \overset{ˉ}{\partial}_{k} φ$ are the entries of the complex Hessian (Levi form) of $φ$ , the Hermitian matrix from 06.10.02.

Strict plurisubharmonicity with a positive lower bound. Say $φ$ has Levi form bounded below by $c$ on $Ω$ , for a positive continuous $c : Ω \to (0, \infty)$ , if $$ \sum_{j,k=1}^n \varphi_{j\bar k}(z), w_j \bar w_k \ \ge\ c(z), |w|^2 \qquad\text{for all } z \in \Omega,\ w \in \mathbb{C}^n. $$ By 06.10.03, a pseudoconvex $Ω$ admits a smooth strictly plurisubharmonic exhaustion, and adding a convex increasing function of it makes $c$ as large as any prescribed positive function; this freedom is used in the proof and again in 06.10.05.

Key theorem with proof Intermediate+

Theorem (Hörmander's weighted $L^{2}$ existence theorem). Let $Ω \subseteq C^{n}$ be pseudoconvex and $φ \in C^{2} (Ω)$ strictly plurisubharmonic with Levi form bounded below by a positive continuous $c$ on $Ω$ . Then for every $\overset{ˉ}{\partial}$ -closed $f \in L_{(0, 1)}^{2} (Ω, φ)$ there exists $u \in L^{2} (Ω, φ)$ with $\overset{ˉ}{\partial} u = f$ and $$ \int_\Omega |u|^2, e^{-\varphi}, dV \ \le\ \int_\Omega \frac{|f|^2}{c}, e^{-\varphi}, dV . \tag{ $*$ } $$ ^{[Hörmander 1965]}

The proof rests on one a-priori inequality, which is in turn a corollary of an exact identity.

The basic identity (Bochner–Kodaira–Morrey–Kohn). For a smooth, compactly supported $(0, 1)$ -form $u = \sum_{k} u_{k} d \overset{z}{ˉ}_{k}$ on $Ω$ , $$ \lVert \bar\partial u\rVert_\varphi^2 + \lVert \bar\partial^{*}\varphi u\rVert\varphi^2 \ =\ \sum_{j,k=1}^n \int_\Omega \varphi_{j\bar k}, u_j \bar u_k, e^{-\varphi}, dV \ +\ \sum_{j,k=1}^n \int_\Omega |\bar\partial_j u_k|^2, e^{-\varphi}, dV . \tag{BKMK} $$ ^{[Kohn 1963/64]}

Proof of (BKMK). Compute both norms on the left in coordinates. The adjoint is $\overset{ˉ}{\partial}_{φ}^{*} u = - \sum_{k} δ_{k} u_{k}$ with $δ_{k} = \partial_{k} - φ_{k}$ , writing $φ_{k} = \partial_{k} φ$ , so $$ \lVert \bar\partial^{*}\varphi u\rVert\varphi^2 = \sum_{j,k}\int (\delta_j u_j)\overline{(\delta_k u_k)}, e^{-\varphi}, dV . $$ For $\overset{ˉ}{\partial} u = \sum_{j < k} (\overset{ˉ}{\partial}_{j} u_{k} - \overset{ˉ}{\partial}_{k} u_{j}) d \overset{z}{ˉ}_{j} \land d \overset{z}{ˉ}_{k}$ , $$ \lVert \bar\partial u\rVert_\varphi^2 = \sum_{j,k}\int |\bar\partial_j u_k|^2, e^{-\varphi}, dV - \sum_{j,k}\int (\bar\partial_k u_j)\overline{(\bar\partial_j u_k)}, e^{-\varphi}, dV . $$ Add the two. The cross terms combine through the commutator of $δ_{j}$ with $\overset{ˉ}{\partial}_{k}$ . A single integration by parts in the weighted inner product gives the adjoint relation $⟨ δ_{j} v, w ⟩_{φ} = ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ}$ for compactly supported $v, w$ , and applying it twice rearranges the second-derivative terms so that $$ \sum_{j,k}\int (\delta_j u_j)\overline{(\delta_k u_k)}, e^{-\varphi} - \sum_{j,k}\int (\bar\partial_k u_j)\overline{(\bar\partial_j u_k)}, e^{-\varphi} = \sum_{j,k}\int [\delta_j, \bar\partial_k], u_k,\bar u_j, e^{-\varphi} . $$ The commutator is a multiplication operator: $$ [\delta_j, \bar\partial_k] = [\partial_j - \varphi_j,\ \bar\partial_k] = -[\varphi_j, \bar\partial_k] = \bar\partial_k\varphi_j = \varphi_{j\bar k}, $$ because $\partial_{j}$ and $\overset{ˉ}{\partial}_{k}$ commute and $\overset{ˉ}{\partial}_{k}$ acting on the multiplication operator $φ_{j}$ produces multiplication by $\overset{ˉ}{\partial}_{k} \partial_{j} φ = φ_{j \overset{ˉ}{k}}$ . Substituting yields the curvature term $\sum_{j, k} \int φ_{j \overset{ˉ}{k}} u_{k} \overset{u}{ˉ}_{j} e^{- φ}$ (the same as $\sum_{j, k} \int φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k} e^{- φ}$ after relabelling, since the Hermitian form is real on the diagonal sum), and the remaining $\sum_{j, k} \int ∣ \overset{ˉ}{\partial}_{j} u_{k} ∣^{2} e^{- φ}$ survives untouched. This is (BKMK). $□$

The a-priori estimate. Drop the manifestly nonnegative last term of (BKMK) and apply the lower bound $\sum φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k} \geq c ∣ u ∣^{2}$ : $$ \int_\Omega c,|u|^2, e^{-\varphi}, dV \ \le\ \lVert\bar\partial u\rVert_\varphi^2 + \lVert\bar\partial^{}\varphi u\rVert\varphi^2 \qquad\text{for all } u \in \mathscr{D}^{(0,1)}(\Omega). \tag{AP} $$ By a density argument — the compactly supported smooth forms are a core for the relevant operators on a pseudoconvex $Ω$ , established by Friedrichs mollification against the strictly PSH exhaustion — (AP) extends to every $u$ in $\mathrm{Dom}(\bar\partial)\cap\mathrm{Dom}(\bar\partial^{}_\varphi)$.

Proof of Theorem. Realise $\overset{ˉ}{\partial}$ as the two consecutive closed operators $$ L^2(\Omega,\varphi) \ \xrightarrow{\ T = \bar\partial\ }\ L^2_{(0,1)}(\Omega,\varphi)\ \xrightarrow{\ S = \bar\partial\ }\ L^2_{(0,2)}(\Omega,\varphi), $$ so $S T = 0$ and $ran T \subseteq ker S$ . The datum $f$ lies in $ker S$ since $\overset{ˉ}{\partial} f = 0$ . To solve $T u = f$ it suffices to show $f ⊥ (ker S)^{⊥}$ is automatic and to control the solution; the functional-analytic device is to produce $u$ as the Riesz representative of a bounded linear functional on $ran T^{*}$ .

Fix $g \in Dom (T^{*}) = Dom (\overset{ˉ}{\partial}_{φ}^{*})$ . Decompose $g = g_{1} + g_{2}$ with $g_{1} \in ker S$ and $g_{2} \in (ker S)^{⊥} \subseteq ker T^{*}$ (the orthogonal splitting of the Hilbert space along the closed subspace $ker S$ ). Then $T^{*} g = T^{*} g_{1}$ , and since $f \in ker S$ , $$ |\langle f, g\rangle_\varphi|^2 = |\langle f, g_1\rangle_\varphi|^2 \ \le\ \Big(\int \frac{|f|^2}{c}, e^{-\varphi}\Big)\Big(\int c,|g_1|^2, e^{-\varphi}\Big) $$ by Cauchy–Schwarz with the weight $c$ . Now $g_{1} \in ker S \cap Dom (T^{*})$ , so (AP) applies with $\overset{ˉ}{\partial} g_{1} = S g_{1} = 0$ , giving $\int c ∣ g_{1} ∣^{2} e^{- φ} \leq ∥ T^{*} g_{1} ∥_{φ}^{2} = ∥ T^{*} g ∥_{φ}^{2}$ . Therefore $$ |\langle f, g\rangle_\varphi| \ \le\ \Big(\int\frac{|f|^2}{c}, e^{-\varphi}\Big)^{1/2}, \lVert T^{}g\rVert_\varphi \qquad\text{for all } g \in \mathrm{Dom}(T^{}). \tag{ $†$ } $$ The inequality ( $†$ ) says the linear functional $T^{*} g \mapsto ⟨ f, g ⟩_{φ}$ is well-defined on $ran T^{*}$ (if $T^{*} g = T^{*} g^{'}$ then $⟨ f, g - g^{'} ⟩_{φ} = 0$ by ( $†$ )) and bounded there by $(\int ∣ f ∣^{2} / c e^{- φ})^{1/2}$ . By the Riesz representation theorem 02.11.08 there is $u \in L^{2} (Ω, φ)$ with $$ \langle f, g\rangle_\varphi = \langle u, T^{}g\rangle_\varphi \quad\text{for all } g \in \mathrm{Dom}(T^{}), \qquad \lVert u\rVert_\varphi \le \Big(\int\frac{|f|^2}{c}, e^{-\varphi}\Big)^{1/2}. $$ The defining relation $⟨ u, T^{*} g ⟩_{φ} = ⟨ f, g ⟩_{φ}$ for all $g \in Dom (T^{*})$ is exactly the statement $u \in Dom (T^{**}) = Dom (T)$ with $T u = f$ , i.e. $\overset{ˉ}{\partial} u = f$ . The norm bound is ( $*$ ). $□$

Bridge. This estimate builds toward the solution of the Levi problem in 06.10.05, where the controlled solution $u$ is the correction term that turns a local peak $1/ h$ into a global holomorphic separating function, and the same cut-off-and-correct pattern appears again in the Cousin problems and the integral-kernel constructions of the chapter. The foundational reason a soft Hilbert-space argument produces a hard existence theorem is the positivity in (BKMK): the curvature term $\sum φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k}$ is the Levi form of the weight, the very Hermitian form whose nonnegativity defined plurisubharmonicity in 06.10.02, now carrying strict positivity $\geq c$ that dominates the data. This generalises the one-variable fact of 06.04.05 that $\overset{ˉ}{\partial} u = f$ is unconditionally solvable on planar domains; in dimension one the compatibility condition $\overset{ˉ}{\partial} f = 0$ is vacuous and no curvature is needed, whereas for $n \geq 2$ the overdetermined system requires both the closedness of $f$ and the convexity budget of $φ$ . The central insight is that the obstruction making several-variable theory hard — the overdeterminacy of $\overset{ˉ}{\partial}$ — is removed by a single weighted inequality, and the weight's freedom (any pseudoconvex domain admits arbitrarily positive strictly PSH weights) is what makes the estimate close on every pseudoconvex domain at once.

Exercises Intermediate+

Exercise 3 (medium, short-answer).

Verify the weighted adjoint relation $⟨ δ_{j} v, w ⟩_{φ} = ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ}$ for $v, w \in C_{c}^{\infty} (Ω)$ , where $δ_{j} = \partial_{j} - \partial_{j} φ$ , and explain why this is the source of the term $- \partial_{j} φ$ in the adjoint.

Hint

Integrate $\int (\partial_{j} v) \overset{w}{ˉ} e^{- φ}$ by parts in the real coordinates underlying $z_{j}$ ; the derivative of $e^{- φ}$ produces the extra term.

Answer

Write $⟨ \partial_{j} v, w ⟩_{φ} = \int (\partial_{j} v) \overset{w}{ˉ} e^{- φ} d V$ . Integration by parts in $z_{j}$ (no boundary term, compact support) moves $\partial_{j}$ onto $\overset{w}{ˉ} e^{- φ}$ : $\int (\partial_{j} v) \overset{w}{ˉ} e^{- φ} = - \int v \partial_{j} (\overset{w}{ˉ} e^{- φ}) = - \int v (\partial_{j} \overset{w}{ˉ}) e^{- φ} + \int v \overset{w}{ˉ} (\partial_{j} φ) e^{- φ}$ . Now $\partial_{j} \overset{w}{ˉ} = \overline{\overset{ˉ}{\partial}_{j} w}$ , so $- \int v (\partial_{j} \overset{w}{ˉ}) e^{- φ} = - ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ}$ . Hence $⟨ \partial_{j} v, w ⟩_{φ} = - ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ} + ⟨(\partial_{j} φ) v, w ⟩_{φ}$ , which rearranges to $⟨(\partial_{j} - \partial_{j} φ) v, w ⟩_{φ} = - ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ}$ ; up to the sign convention bundling $\overset{ˉ}{\partial}_{φ}^{*} = - \sum_{k} δ_{k}$ , this is the claimed relation. The $- \partial_{j} φ$ term is exactly the contribution of differentiating the weight $e^{- φ}$ . Rubric: full credit for the integration by parts and the identification of $- \partial_{j} φ$ with $\partial_{j} (e^{- φ}) / e^{- φ}$ .

Exercise 4 (medium, short-answer).

In the proof of the theorem, the form $f$ is paired only with the component $g_{1} \in ker S$ of $g$ . Explain why $⟨ f, g_{2} ⟩_{φ} = 0$ for the complementary component $g_{2} \in (ker S)^{⊥}$ , and where $\overset{ˉ}{\partial} f = 0$ is used.

Hint

$f$ itself lies in $ker S$ ; the splitting $g = g_{1} + g_{2}$ is orthogonal along the closed subspace $ker S$ .

Answer

The hypothesis $\overset{ˉ}{\partial} f = 0$ means $S f = 0$ , so $f \in ker S$ . The decomposition $g = g_{1} + g_{2}$ is the orthogonal splitting of the Hilbert space $L_{(0, 1)}^{2} (Ω, φ)$ into the closed subspace $ker S$ and its orthogonal complement $(ker S)^{⊥}$ . Since $f \in ker S$ and $g_{2} \in (ker S)^{⊥}$ , orthogonality gives $⟨ f, g_{2} ⟩_{φ} = 0$ , hence $⟨ f, g ⟩_{φ} = ⟨ f, g_{1} ⟩_{φ}$ . This is precisely the step that uses $\overset{ˉ}{\partial} f = 0$ : without it $f$ would have a component in $(ker S)^{⊥}$ , the pairing with $g_{2}$ would not vanish, and the a-priori estimate (which controls only the $ker S$ part via $\overset{ˉ}{\partial}_{φ}^{*}$ ) could not bound $⟨ f, g ⟩_{φ}$ . Rubric: full credit for the orthogonality argument and locating $\overset{ˉ}{\partial} f = 0$ as the reason $f \in ker S$ .

Exercise 5 (medium, short-answer).

Show that on a pseudoconvex $Ω$ the weight can be chosen so that the right-hand side of ( $*$ ) is finite for any given $\overset{ˉ}{\partial}$ -closed $f$ that is smooth, even if $\int ∣ f ∣^{2} e^{- ψ}$ diverges for the natural weight $ψ$ . State the mechanism.

Hint

Replace $φ$ by $φ + χ \circ ψ$ for a rapidly increasing convex $χ$ ; the lower bound $c$ can be pushed up while $\int ∣ f ∣^{2} e^{- φ}$ is kept finite by tailoring $χ$ to the growth of $f$ .

Answer

Let $ψ$ be a smooth strictly PSH exhaustion of $Ω$ (it exists by pseudoconvexity, 06.10.03). For a convex increasing $χ : R \to R$ , the function $φ = ∣ z ∣^{2} + χ (ψ)$ is strictly PSH because $χ (ψ)$ is PSH (convex increasing of PSH, 06.10.02) and $∣ z ∣^{2}$ contributes Levi form $\geq 1$ , so $c \geq 1$ throughout. Given a smooth $\overset{ˉ}{\partial}$ -closed $f$ , the sublevel sets ${ψ < t}$ are relatively compact, so $f$ is bounded on each; choosing $χ$ to grow fast enough that $\int_{Ω} ∣ f ∣^{2} e^{- χ (ψ)} d V < \infty$ (possible since one only needs $e^{- χ (ψ)}$ to decay faster than $∣ f ∣^{2}$ grows on the exhausting compacts) makes the right side of ( $*$ ) finite. The theorem then yields a solution $u \in L^{2} (Ω, φ)$ . The mechanism is the unbounded freedom in the weight on a pseudoconvex domain: convexifying the exhaustion both keeps $c$ positive and tames the growth of the data. Rubric: full credit for the $φ = ∣ z ∣^{2} + χ (ψ)$ construction and the role of the exhaustion in choosing $χ$ .

Exercise 6 (hard, short-answer).

Deduce from the existence theorem that on a pseudoconvex $Ω$ the Dolbeault cohomology $H_{\overset{ˉ}{\partial}}^{0, 1} (Ω) = 0$ : every $\overset{ˉ}{\partial}$ -closed $(0, 1)$ -form is $\overset{ˉ}{\partial}$ -exact. Address why the $L^{2}$ statement upgrades to a statement about smooth forms.

Hint

First get an $L^{2}$ solution against a suitable weight; then invoke elliptic regularity of $\overset{ˉ}{\partial}$ to gain smoothness of $u$ where $f$ is smooth.

Answer

Let $f$ be a smooth $\overset{ˉ}{\partial}$ -closed $(0, 1)$ -form on a pseudoconvex $Ω$ . By Exercise 5 choose a smooth strictly PSH weight $φ$ with Levi form bounded below by $c > 0$ and $\int ∣ f ∣^{2} e^{- φ} / c < \infty$ . The theorem gives $u \in L^{2} (Ω, φ)$ with $\overset{ˉ}{\partial} u = f$ in the distributional sense. The operator $\overset{ˉ}{\partial}$ on functions is (overdetermined-)elliptic: the equation $\overset{ˉ}{\partial} u = f$ together with $f$ smooth forces $u$ smooth, because the Laplacian $Δ u = 4 \sum_{k} \overset{ˉ}{\partial}_{k} (\partial_{k} u)$ is controlled by $f$ and the standard interior elliptic-regularity estimate (06.04.05 for the one-variable model, applied slicewise / via the full Laplacian here) bootstraps $u$ to $C^{\infty}$ . Hence $f = \overset{ˉ}{\partial} u$ with $u$ smooth, so $f$ is $\overset{ˉ}{\partial}$ -exact and the cohomology class vanishes. As $f$ was arbitrary, $H_{\overset{ˉ}{\partial}}^{0, 1} (Ω) = 0$ . The same argument in higher bidegree gives $H_{\overset{ˉ}{\partial}}^{0, q} (Ω) = 0$ for $q \geq 1$ , the Dolbeault form of the vanishing theorem. Rubric: full credit for the $L^{2}$ solution plus the elliptic-regularity upgrade to smoothness and the cohomology conclusion.

Exercise 7 (hard, short-answer).

The minimal-norm solution $u$ produced by Riesz representation is characterised by $u ⊥ ker \overset{ˉ}{\partial}$ , i.e. $u$ is orthogonal to the holomorphic functions in $L^{2} (Ω, φ)$ . Prove this characterisation and identify the resulting $u$ as the canonical solution.

Hint

The Riesz representative lies in $\overline{ran \overset{ˉ}{\partial}_{φ}^{*}} = (ker \overset{ˉ}{\partial})^{⊥}$ ; any two solutions differ by an element of $ker \overset{ˉ}{\partial}$ .

Answer

If $u, u^{'}$ both solve $\overset{ˉ}{\partial} v = f$ then $\overset{ˉ}{\partial} (u - u^{'}) = 0$ , so $u - u^{'} \in ker \overset{ˉ}{\partial} =$ the holomorphic functions in $L^{2} (Ω, φ)$ (a closed subspace, $A_{φ}^{2} (Ω)$ ). The solution set is the affine space $u_{0} + A_{φ}^{2}$ , and the unique minimal-norm element is the one orthogonal to $A_{φ}^{2}$ . The Riesz representative $u$ from the proof satisfies $⟨ u, T^{*} g ⟩_{φ} = ⟨ f, g ⟩_{φ}$ ; restricting attention shows $u$ lies in $\overline{ran T^{*}} = \overline{ran \overset{ˉ}{\partial}_{φ}^{*}}$ , and in general $\overline{ran T^{*}} = (ker T)^{⊥} = (ker \overset{ˉ}{\partial})^{⊥} = (A_{φ}^{2})^{⊥}$ . Therefore $u ⊥ A_{φ}^{2}$ , so $u$ is the minimal-norm (canonical) solution. Concretely, $u = (I - P) u_{0}$ where $P$ is the weighted Bergman projection onto $A_{φ}^{2}$ ; the canonical solution is $\overset{ˉ}{\partial}_{φ}^{*} N f$ for the $\overset{ˉ}{\partial}$ -Neumann operator $N$ , the object whose regularity theory the $\overset{ˉ}{\partial}$ -Neumann problem studies. Rubric: full credit for the affine-solution-set argument, the orthogonality to $ker \overset{ˉ}{\partial}$ , and naming the canonical/Bergman-orthogonal solution.

Advanced results Master

The $L^{2}$ theorem is the bottom layer of a structure that the rest of the chapter and much of modern complex geometry are built on. The refinements below sharpen the estimate, extend it to all bidegrees, and record the two directions in which it is the load-bearing input.

All bidegrees and the Dolbeault vanishing theorem. The basic identity (BKMK) and its consequence hold verbatim for $(0, q)$ -forms, $q \geq 1$ , with the curvature term becoming $\sum_{j, k} φ_{j \overset{ˉ}{k}}$ contracted against the $(0, q)$ -form on its $q$ free indices, summed and bounded below by $q c ∣ u ∣^{2}$ when the Levi form is bounded below by $c$ . Hörmander's existence theorem in bidegree $(0, q)$ then states: on a pseudoconvex $Ω$ , every $\overset{ˉ}{\partial}$ -closed $f \in L_{(0, q)}^{2} (Ω, φ)$ with $q \geq 1$ is $\overset{ˉ}{\partial}$ -exact with the corresponding norm bound. Combined with the elliptic regularity of $\overset{ˉ}{\partial}$ this gives the smooth statement $H_{\overset{ˉ}{\partial}}^{0, q} (Ω) = 0$ for all $q \geq 1$ , the Dolbeault form of Cartan's Theorem B for the structure sheaf $O$ on a domain of holomorphy; the higher-rank version $H^{q} (Ω, F) = 0$ for coherent $F$ on a Stein space is its sheaf-theoretic culmination.

Pseudoconvexity is necessary. The hypothesis is not an artefact of the method. If $\overset{ˉ}{\partial} u = f$ is solvable for every $\overset{ˉ}{\partial}$ -closed $(0, 1)$ -form $f \in C^{\infty}$ on $Ω$ , then $H_{\overset{ˉ}{\partial}}^{0, 1} (Ω) = 0$ , and a Hartogs-figure argument shows $Ω$ has no analytic obstruction to extension, which on a domain in $C^{n}$ is equivalent to pseudoconvexity. Thus solvability of $\overset{ˉ}{\partial}$ with no condition beyond closedness characterises pseudoconvex domains, dovetailing with the Levi problem of 06.10.05: pseudoconvex, domain of holomorphy, and $\overset{ˉ}{\partial}$ -solvable are one condition seen three ways.

Sharper weights: Skoda and Ohsawa–Takegoshi. The Cauchy–Schwarz step in the proof is lossy; tightening it against the full curvature term, rather than its scalar lower bound $c$ , yields the Skoda division estimate and the Ohsawa–Takegoshi extension theorem, in which a holomorphic function on a slice extends to all of $Ω$ with $L^{2}$ control. These are the same identity (BKMK) read with a sharper functional-analytic step and a twisted weight $φ - lo g (auxiliary)$ ; they power the analytic theory of multiplier ideals and the modern proofs of invariance of plurigenera. The single inequality (AP) is the seed of all of them.

From abstract solution to explicit kernel. The solution $u$ delivered by Riesz representation is canonical (Exercise 7) but implicit. On a strongly pseudoconvex domain the Henkin–Ramirez integral kernels produce an explicit solution operator for $\overset{ˉ}{\partial}$ with sup-norm and Hölder- $1/2$ estimates, which the $L^{2}$ method does not see; conversely the $L^{2}$ method works on every pseudoconvex domain, including weakly pseudoconvex ones such as ${∣ z_{1} ∣^{2} + ∣ z_{2} ∣^{4} < 1}$ where no holomorphic support function and hence no Henkin-type kernel exists. The two are complementary: $L^{2}$ for existence in full generality, kernels for sharp boundary regularity in the strongly pseudoconvex case.

The $\overset{ˉ}{\partial}$ -Neumann problem. Promoting the existence theorem to a statement with optimal boundary regularity requires solving the boundary-value problem for the complex Laplacian $□ = \overset{ˉ}{\partial} \overset{ˉ}{\partial}^{*} + \overset{ˉ}{\partial}^{*} \overset{ˉ}{\partial}$ with the free (Neumann-type) boundary conditions that the domain of $\overset{ˉ}{\partial}^{*}$ imposes. The basic estimate (AP) is the unweighted ( $φ = 0$ ) heart of Kohn's subelliptic estimate $∥ u ∥_{1/2}^{2} ≲ Q (u, u)$ on strongly pseudoconvex domains, where the gain of half a derivative comes from the strict positivity of the Levi form on the boundary. The canonical solution of Exercise 7 is $\overset{ˉ}{\partial}^{*} N f$ for the $\overset{ˉ}{\partial}$ -Neumann operator $N = □^{- 1}$ .

Synthesis. The weighted $L^{2}$ method is the foundational reason several-variable complex analysis has a working existence theory: it converts the geometric hypothesis of pseudoconvexity into an analytic inequality, the basic identity (BKMK), whose single positive term is the Levi form of the weight, and then extracts existence from that inequality by the softest possible functional analysis. Putting these together, four objects coincide on a domain $Ω \subseteq C^{n}$ — domain of holomorphy, holomorphically convex, pseudoconvex, and unconditionally $\overset{ˉ}{\partial}$ -solvable — and the bridge realising the last identification is the estimate ( $*$ ), which holds because pseudoconvex domains carry arbitrarily positive strictly plurisubharmonic weights. The cone of admissible weights is dual to the geometry in the sense that $∣ z ∣^{2}$ supplies the universal unit of curvature and any strictly PSH exhaustion supplies the rest, and the curvature term $\sum φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k}$ is exactly the data the Levi-problem proof of 06.10.05 needs to dominate its defect. This pattern recurs across the chapter and beyond: the Cousin problems read $\overset{ˉ}{\partial}$ -solvability as cohomology vanishing, the kernel constructions sharpen the abstract solution to an explicit integral, the $\overset{ˉ}{\partial}$ -Neumann theory upgrades it to optimal boundary regularity, and the Skoda–Ohsawa–Takegoshi refinements re-read (BKMK) with sharper weights to drive the analytic side of higher-dimensional algebraic geometry.

Full proof set Master

Proposition 1 (the weighted adjoint and the commutator). For $φ \in C^{2} (Ω)$ and $v, w \in C_{c}^{\infty} (Ω)$ , the operator $δ_{j} = \partial_{j} - \partial_{j} φ$ satisfies $⟨ δ_{j} v, w ⟩_{φ} = ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ}$ , and $[δ_{j}, \overset{ˉ}{\partial}_{k}] = φ_{j \overset{ˉ}{k}}$ as a multiplication operator.

Proof. Integration by parts in the real variables underlying $z_{j}$ (compact support, no boundary terms) gives $\int (\partial_{j} v) \overset{w}{ˉ} e^{- φ} = - \int v \partial_{j} (\overset{w}{ˉ} e^{- φ}) = - \int v (\partial_{j} \overset{w}{ˉ}) e^{- φ} + \int v \overset{w}{ˉ} (\partial_{j} φ) e^{- φ}$ . Since $\partial_{j} \overset{w}{ˉ} = \overline{\overset{ˉ}{\partial}_{j} w}$ , the first integral is $- ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ}$ ; moving the second to the left, $⟨(\partial_{j} - \partial_{j} φ) v, w ⟩_{φ} = - ⟨ v, \overset{ˉ}{\partial}_{j} w ⟩_{φ}$ , which is the adjoint relation up to the global sign in $\overset{ˉ}{\partial}_{φ}^{*} = - \sum_{k} δ_{k}$ . For the commutator, $δ_{j} \overset{ˉ}{\partial}_{k} ψ - \overset{ˉ}{\partial}_{k} δ_{j} ψ = (\partial_{j} - φ_{j}) \overset{ˉ}{\partial}_{k} ψ - \overset{ˉ}{\partial}_{k} (\partial_{j} ψ - φ_{j} ψ)$ . The second-derivative terms $\partial_{j} \overset{ˉ}{\partial}_{k} ψ$ cancel (mixed partials commute), and the remaining terms are $- φ_{j} \overset{ˉ}{\partial}_{k} ψ + \overset{ˉ}{\partial}_{k} (φ_{j} ψ) = (\overset{ˉ}{\partial}_{k} φ_{j}) ψ = φ_{j \overset{ˉ}{k}} ψ$ . Hence $[δ_{j}, \overset{ˉ}{\partial}_{k}] = φ_{j \overset{ˉ}{k}}$ . $□$

Proposition 2 (positivity of the curvature term). If $φ$ has Levi form bounded below by $c > 0$ , then for every $(0, 1)$ -form $u$ , $\sum_{j, k} φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k} \geq c ∣ u ∣^{2}$ pointwise, and the term is real and nonnegative after integration.

Proof. At each point the matrix $(φ_{j \overset{ˉ}{k}})$ is Hermitian since $\overline{φ_{j \overset{ˉ}{k}}} = \overline{\partial_{j} \overset{ˉ}{\partial}_{k} φ} = \overset{ˉ}{\partial}_{j} \partial_{k} φ = \partial_{k} \overset{ˉ}{\partial}_{j} φ = φ_{k \overset{ˉ}{j}}$ , using that $φ$ is real-valued so $\overline{\partial_{j} \overset{ˉ}{\partial}_{k} φ} = \partial_{k} \overset{ˉ}{\partial}_{j} φ$ . A Hermitian form satisfies $\sum_{j, k} φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k} \in R$ , and the hypothesis with $w = (u_{1}, \dots, u_{n})$ is exactly $\sum_{j, k} φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k} \geq c ∣ u ∣^{2} \geq 0$ . Integrating against $e^{- φ} > 0$ preserves the inequality. $□$

Proposition 3 (density: $C_{c}^{\infty}$ forms are a core). On a pseudoconvex $Ω$ with smooth strictly PSH exhaustion $ψ$ , the space $D^{(0, 1)} (Ω)$ of smooth compactly supported $(0, 1)$ -forms is dense in $\mathrm{Dom}(\bar\partial)\cap\mathrm{Dom}(\bar\partial^{}\varphi) $f or t h e g r a p hn or m$ u\mapsto(\lVert u\rVert\varphi^2 + \lVert\bar\partial u\rVert_\varphi^2 + \lVert\bar\partial^{}\varphi u\rVert\varphi^2)^{1/2} $, so (A P) e x t e n d s f r o m$ \mathscr{D}^{(0,1)}$ to the full intersection of domains.

Proof. Let $η_{ν} = ζ (ψ / ν)$ for a cutoff $ζ$ equal to $1$ on $(- \infty, 1]$ and supported in $(- \infty, 2]$ ; since ${ψ < 2 ν} ⋐ Ω$ , $η_{ν}$ has compact support and $η_{ν} ↑ 1$ . For $u \in Dom (\overset{ˉ}{\partial}) \cap Dom (\overset{ˉ}{\partial}_{φ}^{*})$ , the truncations $η_{ν} u$ have compact support and converge to $u$ in $L^{2} (Ω, φ)$ ; the commutator terms $[\overset{ˉ}{\partial}, η_{ν}] u = (\overset{ˉ}{\partial} η_{ν}) \land u$ and $[\overset{ˉ}{\partial}_{φ}^{*}, η_{ν}] u$ involve $\overset{ˉ}{\partial} η_{ν} = ζ^{'} (ψ / ν) \overset{ˉ}{\partial} ψ / ν$ , which tends to $0$ in $L^{2}$ because $∣ \overset{ˉ}{\partial} η_{ν} ∣ \leq C / ν$ on its support and $ψ$ is, after the standard convexification, taken with $∣ \overset{ˉ}{\partial} ψ ∣$ controlled there. Friedrichs mollification then smooths each $η_{ν} u$ without leaving the domain, since $\overset{ˉ}{\partial}$ and $\overset{ˉ}{\partial}_{φ}^{*}$ have smooth coefficients and mollification commutes with them up to terms vanishing in the graph norm. Combining truncation and mollification produces a sequence in $D^{(0, 1)} (Ω)$ converging to $u$ in the graph norm; the estimate (AP), being a continuous inequality in that norm, passes to the limit. $□$

Proposition 4 (the existence step is sharp on the diagonal). The constant in ( $*$ ) cannot be improved when $f$ is an eigen-configuration of the Levi form: if $(φ_{j \overset{ˉ}{k}}) = c I$ is a constant multiple of the identity and $f$ is supported where this holds, the bound $∥ u ∥_{φ}^{2} \leq \int ∣ f ∣^{2} / c e^{- φ}$ is attained by the canonical solution up to the slack in the dropped term $\sum ∣ \overset{ˉ}{\partial}_{j} u_{k} ∣^{2}$ .

Proof. When $(φ_{j \overset{ˉ}{k}}) = c I$ the curvature term in (BKMK) is exactly $c ∥ u ∥_{φ}^{2}$ , so (AP) reads $c ∥ u ∥_{φ}^{2} + \sum_{j, k} \int ∣ \overset{ˉ}{\partial}_{j} u_{k} ∣^{2} e^{- φ} = ∥ \overset{ˉ}{\partial} u ∥_{φ}^{2} + ∥ \overset{ˉ}{\partial}_{φ}^{*} u ∥_{φ}^{2}$ with equality, not inequality, since no term was bounded below — the lower bound $\geq c ∣ u ∣^{2}$ is an equality here. The Cauchy–Schwarz step in the existence proof is then the only inequality, and it is sharp precisely when $f$ is proportional (in the weighted inner product, pointwise) to the minimiser $g_{1}$ ; for such aligned data the produced $u$ saturates ( $*$ ) modulo the nonnegative $\sum ∣ \overset{ˉ}{\partial}_{j} u_{k} ∣^{2}$ slack, which vanishes when the canonical solution is itself holomorphic-gradient-free. This identifies the constant $1$ multiplying $\int ∣ f ∣^{2} / c$ as optimal: no universal constant smaller than $1$ can replace it. $□$

Connections Master

Plurisubharmonic functions 06.10.02. The curvature term $\sum_{j, k} φ_{j \overset{ˉ}{k}} u_{j} \overset{u}{ˉ}_{k}$ in the basic identity is the Levi form of the weight $φ$ , the Hermitian form whose nonnegativity defines plurisubharmonicity there. The whole method is the statement that a strictly PSH weight injects pointwise positivity into an integration-by-parts identity; the smooth-strictly-PSH approximation proved in that unit is what lets the weight be taken regular without loss.
Pseudoconvexity and the Levi form 06.10.03. Pseudoconvexity enters exactly as the guarantee that $Ω$ carries a smooth strictly PSH exhaustion, hence weights with Levi form bounded below by any prescribed positive $c$ . The equivalence of the three pseudoconvexity notions developed there is what makes the hypothesis of this theorem checkable in practice.
Solution of the Levi problem 06.10.05. This unit is the sole analytic input cited by the Levi-problem proof: the controlled solution $u$ of $\overset{ˉ}{\partial} u = g$ is the correction that turns a local peak $1/ h$ into a global holomorphic separating function. The singular-weight trick there — making $e^{- φ}$ non-integrable at a boundary point — is an application of the weight-freedom established here.
The one-variable $\overset{ˉ}{\partial}$ -Hilbert PDE 06.04.05. The dimension-one ancestor: there $\overset{ˉ}{\partial} u = f$ is unconditionally solvable because the compatibility condition is vacuous ( $(0, 1)$ -forms are automatically closed for $n = 1$ ) and no curvature budget is required. This unit is the obstruction-laden generalisation, where overdeterminacy forces both $\overset{ˉ}{\partial} f = 0$ and the strictly PSH weight. The elliptic regularity that upgrades $L^{2}$ solutions to smooth ones is inherited from that unit.
Hilbert space and unbounded operators 02.11.08, 02.11.03. The existence step is pure Hilbert-space theory: $\overset{ˉ}{\partial}$ and $\overset{ˉ}{\partial}_{φ}^{*}$ as adjoint closed densely defined operators, the orthogonal splitting along $ker S$ , and the Riesz representation theorem. The domain condition defining $\overset{ˉ}{\partial}_{φ}^{*}$ is exactly the free-boundary condition of the $\overset{ˉ}{\partial}$ -Neumann problem, an unbounded-operator subtlety those units set up.

Historical & philosophical context Master

The weighted $L^{2}$ method was created by Lars Hörmander in his 1965 paper $L^{2}$ estimates and existence theorems for the $\overset{ˉ}{\partial}$ operator ^{[Hörmander 1965]} (Acta Mathematica 113, 89–152), which solved the inhomogeneous Cauchy–Riemann equation on pseudoconvex domains with the norm estimate ( $*$ ) and, as an immediate corollary, gave a new proof of the Levi problem. Hörmander's innovation was to run the existence argument entirely through a single weighted a-priori inequality and the abstract closed-range theorem for unbounded operators, dispensing with Oka's coherence machinery. The weighted reformulation was developed in parallel by Aldo Andreotti and Edoardo Vesentini ^{[Andreotti–Vesentini 1965]} (Publications Mathématiques de l'IHÉS 25, 81–130), who cast the estimate as a Carleman inequality for the Laplace–Beltrami operator on complex manifolds.

The basic identity at the centre of the method has older roots. Charles B. Morrey's 1958 integration-by-parts computation ^{[Morrey 1958]} (Annals of Mathematics 68, 159–201), made in the course of the analytic embedding of real-analytic manifolds, isolated the commutator term that becomes the Levi form. Joseph J. Kohn, in his 1963–64 solution of the $\overset{ˉ}{\partial}$ -Neumann problem on strongly pseudoconvex manifolds ^{[Kohn 1963/64]} (Annals of Mathematics 78, 112–148 and 79, 450–472), turned the computation into the basic estimate and the subelliptic theory; the boundary term Kohn controlled is the same curvature that Hörmander absorbed into an interior weight. The four-name attribution Bochner–Kodaira–Morrey–Kohn reflects the parallel Bochner–Kodaira identities of compact Kähler geometry, of which the open-domain weighted identity is the noncompact analogue.

The conceptual shift is that an existence theorem in complex analysis became a consequence of a positivity estimate in functional analysis. Where the nineteenth- and early-twentieth-century theory of one variable solved $\overset{ˉ}{\partial}$ by explicit integral kernels, and Oka solved the several-variable Levi problem by coherence, Hörmander replaced construction by estimation: one never writes the solution, one bounds it. This move — from kernels to a-priori inequalities — propagated through Skoda's division theorem, the Ohsawa–Takegoshi extension theorem, and Demailly's analytic methods in algebraic geometry, each obtained by reading the Morrey–Kohn identity against a sharper weight.

Bibliography Master

@article{Hormander1965,
  author  = {H\"ormander, Lars},
  title   = {$L^2$ estimates and existence theorems for the
             $\bar\partial$ operator},
  journal = {Acta Math.},
  volume  = {113},
  year    = {1965},
  pages   = {89--152}
}

@article{Kohn1963,
  author  = {Kohn, J. J.},
  title   = {Harmonic integrals on strongly pseudo-convex manifolds. I},
  journal = {Ann. of Math. (2)},
  volume  = {78},
  year    = {1963},
  pages   = {112--148}
}

@article{Kohn1964,
  author  = {Kohn, J. J.},
  title   = {Harmonic integrals on strongly pseudo-convex manifolds. II},
  journal = {Ann. of Math. (2)},
  volume  = {79},
  year    = {1964},
  pages   = {450--472}
}

@article{Morrey1958,
  author  = {Morrey, Charles B., Jr.},
  title   = {The analytic embedding of abstract real-analytic manifolds},
  journal = {Ann. of Math. (2)},
  volume  = {68},
  year    = {1958},
  pages   = {159--201}
}

@article{AndreottiVesentini1965,
  author  = {Andreotti, Aldo and Vesentini, Edoardo},
  title   = {Carleman estimates for the {Laplace--Beltrami} equation on
             complex manifolds},
  journal = {Inst. Hautes \'Etudes Sci. Publ. Math.},
  volume  = {25},
  year    = {1965},
  pages   = {81--130}
}

@article{OhsawaTakegoshi1987,
  author  = {Ohsawa, Takeo and Takegoshi, Kensh\=o},
  title   = {On the extension of $L^2$ holomorphic functions},
  journal = {Math. Z.},
  volume  = {195},
  year    = {1987},
  pages   = {197--204}
}

@book{KrantzSCV,
  author    = {Krantz, Steven G.},
  title     = {Function Theory of Several Complex Variables},
  edition   = {2},
  series    = {AMS Chelsea Publishing},
  volume    = {340},
  publisher = {American Mathematical Society},
  year      = {2001}
}

@book{HormanderSCV,
  author    = {H\"ormander, Lars},
  title     = {An Introduction to Complex Analysis in Several Variables},
  edition   = {3},
  series    = {North-Holland Mathematical Library},
  volume    = {7},
  publisher = {North-Holland},
  year      = {1990}
}

Prerequisites

06.10.02
06.10.03
06.10.01
06.04.05
02.11.08
02.11.03

Tier anchors

beginner: Krantz *Function Theory of Several Complex Variables* Ch. 4, informal sections; Range *Holomorphic Functions and Integral Representations* Ch. IV intro
intermediate: Krantz *Function Theory of Several Complex Variables* §4.2; Hörmander *An Introduction to Complex Analysis in Several Variables* Ch. 4
master: Hörmander 1965 (Acta Math. 113, originator); Kohn 1963/64; Morrey 1958; Andreotti–Vesentini 1965; Krantz Ch. 4; Hörmander Ch. 4

References

Hörmander 1965 — L^2 estimates and existence theorems for the dbar operator · Acta Math. 113, 89–152 (the weighted L^2 existence theorem; originator paper)
Krantz — Function Theory of Several Complex Variables · AMS Chelsea 340, 2nd ed., Ch. 4 (the dbar problem with L^2 estimates)
Hörmander — An Introduction to Complex Analysis in Several Variables · North-Holland, 3rd ed. 1990, Ch. 4 (the dbar operator; Theorem 4.4.2)
Kohn 1963/64 — Harmonic integrals on strongly pseudo-convex manifolds I, II · Ann. of Math. 78, 112–148 (1963); 79, 450–472 (1964) (the basic estimate via the Morrey–Kohn identity)
Morrey 1958 — The analytic embedding of abstract real-analytic manifolds · Ann. of Math. 68, 159–201 (the integration-by-parts identity underlying the basic estimate)
Andreotti–Vesentini 1965 — Carleman estimates for the Laplace–Beltrami equation on complex manifolds · Publ. Math. IHÉS 25, 81–130 (weighted L^2 / Carleman-estimate formulation)
Range — Holomorphic Functions and Integral Representations in Several Complex Variables · Springer GTM 108, Ch. IV (the dbar problem and the Levi problem)

Estimated time

beginner: 16m
intermediate: 44m
master: 78m