[proofplan] The proof has two halves: an *a priori estimate* and a *duality argument*. For the a priori estimate we expand the weighted graph norm $\|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt\eta\,\bar\partial^*\alpha\|^2$ of a compactly supported test form using the definition of the formal adjoint and the Bochner–Kodaira–Nakano identity; the first-order term carrying $\partial\eta$ is then absorbed by a pointwise Cauchy–Schwarz inequality with parameter $\lambda$, which manufactures exactly the term $-\lambda^{-1}i\partial\eta\wedge\bar\partial\eta$ and upgrades the weight on $\bar\partial^*$ from $\eta$ to $\eta+\lambda$. This yields $\langle\!\langle B_{\eta,\lambda}\alpha,\alpha\rangle\!\rangle \le \|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\|^2$, which completeness of $\omega$ extends from compactly supported forms to the operator domain. Positivity of $B_{\eta,\lambda}$ then converts the finiteness hypothesis into a bound on the linear functional $\alpha\mapsto\langle\!\langle f,\alpha\rangle\!\rangle$ restricted to $\ker\bar\partial$, and the Hahn–Banach and [Riesz representation](/theorems/67) theorems produce the solution $u$ together with its weighted norm bound; the [orthogonal decomposition](/theorems/436) of $L^2$ relative to $\ker\bar\partial$ promotes the identity $\langle\!\langle \bar\partial^*\alpha,u\rangle\!\rangle = \langle\!\langle f,\alpha\rangle\!\rangle$ to all test forms, giving $\bar\partial u = f$. [/proofplan] [step:Fix the Hilbert space framework and the operators] Throughout, $\mathcal{D}^{n,q}(X,L)$ denotes the space of smooth, compactly supported $L$-valued $(n,q)$-forms, and $L^2_{n,q}(X,L)$ the [Hilbert space](/page/Hilbert%20Space) of square-integrable $L$-valued $(n,q)$-forms under $\langle\!\langle\cdot,\cdot\rangle\!\rangle$ (linear in the first argument, conjugate-linear in the second). Since $X$ is connected and $\omega$, $h$ are smooth, $\mathcal{D}^{n,q}(X,L)$ is dense in $L^2_{n,q}(X,L)$. We regard \begin{align*} \bar\partial : L^2_{n,q-1}(X,L) &\to L^2_{n,q}(X,L), & \bar\partial : L^2_{n,q}(X,L) &\to L^2_{n,q+1}(X,L) \end{align*} as closed, densely defined operators (the maximal distributional extensions), and write $\bar\partial^*$ for the Hilbert-space adjoint of the first of these. On $\mathcal{D}^{n,q}(X,L)$ the adjoint $\bar\partial^*$ coincides with the formal adjoint $\vartheta$ ([integration by parts](/theorems/2098) has no boundary contribution). Because $\bar\partial^2 = 0$, we have the standing inclusion \begin{align*} \operatorname{Range}(\bar\partial) \subseteq \ker\bar\partial , \end{align*} and, $\bar\partial$ being a closed operator, $\ker\bar\partial \subseteq L^2_{n,q}(X,L)$ is a closed subspace and $(\bar\partial^*)^* = \bar\partial$. Let $D'$ denote the $(1,0)$-part of the Chern connection of $(L,h)$ acting on $L$-valued forms, and $D'^*$ its formal adjoint. For a smooth $1$-form $\gamma$ we write $e_\gamma$ for the operator $\beta \mapsto \gamma \wedge \beta$ and $e_\gamma^*$ for its pointwise adjoint (contraction). Since $B_{\eta,\lambda}$ is a positive-definite Hermitian endomorphism of a finite-rank bundle, its pointwise square root $B_{\eta,\lambda}^{1/2}$ and inverse $B_{\eta,\lambda}^{-1}$ are well-defined smooth positive-definite endomorphisms; we abbreviate $B := B_{\eta,\lambda}$. Our goal is to produce $u\in L^2_{n,q-1}(X,L)$ with $\bar\partial u = f$ distributionally and $\int_X(\eta+\lambda)^{-1}|u|^2_{\omega,h}\,dV_\omega \le \int_X(B^{-1}f,f)_{\omega,h}\,dV_\omega$. [/step] [step:Expand the weighted graph norm through the Bochner–Kodaira–Nakano identity] Let $\alpha \in \mathcal{D}^{n,q}(X,L)$. Because $\eta$ is a real-valued function, the Leibniz rule gives, for any smooth form $\beta$, \begin{align*} \bar\partial(\eta\beta) &= \eta\,\bar\partial\beta + e_{\bar\partial\eta}\beta, & \bar\partial^*(\eta\beta) &= \eta\,\bar\partial^*\beta - e_{\bar\partial\eta}^*\beta, \end{align*} the second identity being the adjoint of the first. Using these and the definition of the adjoint, \begin{align*} \|\sqrt\eta\,\bar\partial\alpha\|^2 &= \langle\!\langle \eta\,\bar\partial\alpha,\bar\partial\alpha\rangle\!\rangle = \langle\!\langle \eta\,\bar\partial^*\bar\partial\alpha,\alpha\rangle\!\rangle - \langle\!\langle e_{\bar\partial\eta}^*\bar\partial\alpha,\alpha\rangle\!\rangle ,\\ \|\sqrt\eta\,\bar\partial^*\alpha\|^2 &= \langle\!\langle \eta\,\bar\partial^*\alpha,\bar\partial^*\alpha\rangle\!\rangle = \langle\!\langle \eta\,\bar\partial\bar\partial^*\alpha,\alpha\rangle\!\rangle + \langle\!\langle e_{\bar\partial\eta}\bar\partial^*\alpha,\alpha\rangle\!\rangle . \end{align*} Adding, with $\Box_{\bar\partial} := \bar\partial^*\bar\partial + \bar\partial\bar\partial^*$, \begin{align*} \|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt\eta\,\bar\partial^*\alpha\|^2 = \langle\!\langle \eta\,\Box_{\bar\partial}\alpha,\alpha\rangle\!\rangle + \langle\!\langle \bar\partial^*\alpha, e_{\bar\partial\eta}^*\alpha\rangle\!\rangle - \langle\!\langle \bar\partial\alpha, e_{\bar\partial\eta}\alpha\rangle\!\rangle . \tag{$\mathrm{A}$} \end{align*} We now invoke (citing a result not yet in the wiki: the *Bochner–Kodaira–Nakano identity*) on $L$-valued $(n,q)$-forms, which states $\Box_{\bar\partial} = D'^*D' + D'D'^* + [i\Theta_h(L),\Lambda]$. Since the $(1,0)$-degree of an $(n,q)$-form is already maximal, $D'\alpha = 0$ identically, so $\Box_{\bar\partial}\alpha = D'D'^*\alpha + [i\Theta_h(L),\Lambda]\alpha$. Therefore \begin{align*} \langle\!\langle \eta\,\Box_{\bar\partial}\alpha,\alpha\rangle\!\rangle = \langle\!\langle \eta\,[i\Theta_h(L),\Lambda]\alpha,\alpha\rangle\!\rangle + \langle\!\langle \eta\,D'D'^*\alpha,\alpha\rangle\!\rangle, \end{align*} and, using $D'^*(\eta\alpha) = \eta\,D'^*\alpha - e_{\partial\eta}^*\alpha$, \begin{align*} \langle\!\langle \eta\,D'D'^*\alpha,\alpha\rangle\!\rangle = \langle\!\langle D'^*\alpha, D'^*(\eta\alpha)\rangle\!\rangle = \|\sqrt\eta\,D'^*\alpha\|^2 - \langle\!\langle D'^*\alpha, e_{\partial\eta}^*\alpha\rangle\!\rangle . \end{align*} Substituting into $(\mathrm{A})$ gives \begin{align*} \|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt\eta\,\bar\partial^*\alpha\|^2 = \langle\!\langle \eta\,[i\Theta_h(L),\Lambda]\alpha,\alpha\rangle\!\rangle + \|\sqrt\eta\,D'^*\alpha\|^2 + P + Q, \tag{$\mathrm{B}$} \end{align*} where $P := -\langle\!\langle D'^*\alpha, e_{\partial\eta}^*\alpha\rangle\!\rangle - \langle\!\langle \bar\partial\alpha, e_{\bar\partial\eta}\alpha\rangle\!\rangle$ and $Q := \langle\!\langle \bar\partial^*\alpha, e_{\bar\partial\eta}^*\alpha\rangle\!\rangle$. Finally we collect the curvature contributions. We use the Kähler commutation identities for the Chern connection on a Kähler manifold, in particular \begin{align*} \bar\partial^* &= i[\Lambda,D'], & D'^* &= -i[\Lambda,\bar\partial], \end{align*} with the corresponding commutator rule for wedge multiplication by a scalar form. These identities apply here because $\omega$ is Kähler, hence the Chern connection has no torsion contribution in the commutation formulas. Applying them to the scalar function $\eta$ gives \begin{align*} -\langle\!\langle D'^*\alpha, e_{\partial\eta}^*\alpha\rangle\!\rangle - \langle\!\langle \bar\partial\alpha, e_{\bar\partial\eta}\alpha\rangle\!\rangle = -\langle\!\langle [i\partial\bar\partial\eta,\Lambda]\alpha,\alpha\rangle\!\rangle + \overline{\langle\!\langle \bar\partial^*\alpha,e_{\bar\partial\eta}^*\alpha\rangle\!\rangle}. \end{align*} Thus the first-derivative terms in $P$ reassemble as \begin{align*} P = -\langle\!\langle [i\partial\bar\partial\eta,\Lambda]\alpha,\alpha\rangle\!\rangle + \overline{Q}. \end{align*} Since the left-hand side of $(\mathrm{B})$ and the terms $\langle\!\langle \eta[i\Theta_h(L),\Lambda]\alpha,\alpha\rangle\!\rangle$, $\|\sqrt\eta\,D'^*\alpha\|^2$ are real, the quantity $P+Q = -\langle\!\langle [i\partial\bar\partial\eta,\Lambda]\alpha,\alpha\rangle\!\rangle + (Q + \overline Q)$ is real, and we obtain the **twisted Bochner–Kodaira–Nakano identity** \begin{align*} \|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt\eta\,\bar\partial^*\alpha\|^2 = \big\langle\!\big\langle [\eta\, i\Theta_h(L) - i\partial\bar\partial\eta,\;\Lambda]\,\alpha,\alpha\big\rangle\!\big\rangle + \|\sqrt\eta\,D'^*\alpha\|^2 + 2\,\mathrm{Re}\,\langle\!\langle \bar\partial^*\alpha, e_{\bar\partial\eta}^*\alpha\rangle\!\rangle. \tag{$\mathrm{C}$} \end{align*} [guided] The aim of this step is to compute the weighted graph norm $\|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt\eta\,\bar\partial^*\alpha\|^2$ of a test form and express it in terms of curvature. Why a *weighted* norm? Because the eventual estimate must see the curvature of the *twisted* bundle, and the twisting function $\eta$ enters precisely as a weight. We first move the weight $\eta$ off one factor and onto the operator using the formal adjoint. Since $\eta$ is a real function ($0$-form), the Leibniz rule reads $\bar\partial(\eta\beta) = \eta\,\bar\partial\beta + \bar\partial\eta\wedge\beta = \eta\,\bar\partial\beta + e_{\bar\partial\eta}\beta$. Taking the formal adjoint of multiplication-then-$\bar\partial$ yields $\bar\partial^*(\eta\beta) = \eta\,\bar\partial^*\beta - e_{\bar\partial\eta}^*\beta$, where $e_{\bar\partial\eta}^*$ is contraction against $\overline{\partial\eta}$ (the sign flips because adjoining $e_{\bar\partial\eta}$ produces its contraction with a sign). Concretely: \begin{align*} \|\sqrt\eta\,\bar\partial\alpha\|^2 = \langle\!\langle \bar\partial^*(\eta\,\bar\partial\alpha),\alpha\rangle\!\rangle = \langle\!\langle \eta\,\bar\partial^*\bar\partial\alpha - e_{\bar\partial\eta}^*\bar\partial\alpha,\alpha\rangle\!\rangle, \end{align*} and likewise $\|\sqrt\eta\,\bar\partial^*\alpha\|^2 = \langle\!\langle \bar\partial(\eta\,\bar\partial^*\alpha),\alpha\rangle\!\rangle = \langle\!\langle \eta\,\bar\partial\bar\partial^*\alpha + e_{\bar\partial\eta}\bar\partial^*\alpha,\alpha\rangle\!\rangle$. Adding gives $(\mathrm{A})$: the principal term $\langle\!\langle\eta\,\Box_{\bar\partial}\alpha,\alpha\rangle\!\rangle$ plus two first-order error terms that carry one derivative of $\eta$. Now we must turn $\Box_{\bar\partial}$ into curvature. This is exactly what the Bochner–Kodaira–Nakano identity does: $\Box_{\bar\partial} - \Box_{D'} = [i\Theta_h(L),\Lambda]$. The decisive simplification in form-degree $(n,q)$ is that $D'\alpha = 0$ — there is no room to raise the holomorphic degree past $n$ — so the $D'$-Laplacian collapses to $\Box_{D'}\alpha = D'D'^*\alpha$. This is why the theorem is naturally stated for $(n,q)$-forms: only the manifestly nonnegative term $\|\sqrt\eta\,D'^*\alpha\|^2$ survives from the $\partial$-side. Pulling the weight through $D'^*$ exactly as before ($D'^*(\eta\alpha) = \eta\,D'^*\alpha - e_{\partial\eta}^*\alpha$) produces $\|\sqrt\eta\,D'^*\alpha\|^2$ and a third first-order error term, giving $(\mathrm{B})$. What remains is bookkeeping of the three first-order terms. Two of them ($P$) involve $\partial\eta$ paired with $D'^*\alpha$ and $\bar\partial\alpha$. The Kähler commutation identities used here are $\bar\partial^* = i[\Lambda,D']$ and $D'^* = -i[\Lambda,\bar\partial]$, together with the same commutator rule applied to wedge multiplication by $\partial\eta$ and $\bar\partial\eta$. These identities are available because $\omega$ is Kähler, so the torsion terms that would appear on a general Hermitian manifold vanish. Applying them to the scalar function $\eta$ reorganizes $P$ into the *second*-order curvature term $-\langle\!\langle[i\partial\bar\partial\eta,\Lambda]\alpha,\alpha\rangle\!\rangle$ — this is where the Hessian $i\partial\bar\partial\eta$ of the twisting function enters — together with a copy of $\overline Q$. Reality of every other term in $(\mathrm{B})$ forces $P+Q$ to equal $-\langle\!\langle[i\partial\bar\partial\eta,\Lambda]\alpha,\alpha\rangle\!\rangle + 2\mathrm{Re}\,Q$. The single surviving first-order term $2\mathrm{Re}\,Q = 2\mathrm{Re}\,\langle\!\langle\bar\partial^*\alpha,e_{\bar\partial\eta}^*\alpha\rangle\!\rangle$ is deliberately left intact: it is the term we will absorb in the next step, and the parameter introduced there is exactly the second twisting function $\lambda$. [/guided] [/step] [step:Absorb the first-order term by Cauchy–Schwarz to obtain the twisted a priori inequality] Discard the nonnegative term $\|\sqrt\eta\,D'^*\alpha\|^2 \ge 0$ in $(\mathrm{C})$ and rearrange: \begin{align*} \|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt\eta\,\bar\partial^*\alpha\|^2 \;\ge\; \big\langle\!\big\langle [\eta\, i\Theta_h(L) - i\partial\bar\partial\eta,\;\Lambda]\,\alpha,\alpha\big\rangle\!\big\rangle + 2\,\mathrm{Re}\,\langle\!\langle \bar\partial^*\alpha, e_{\bar\partial\eta}^*\alpha\rangle\!\rangle . \end{align*} Apply the pointwise [Cauchy–Schwarz inequality](/theorems/432) to the integrand of the last term, with weight split by the positive function $\lambda$: for every $x \in X$, \begin{align*} 2\,\mathrm{Re}\,(\bar\partial^*\alpha, e_{\bar\partial\eta}^*\alpha)_{\omega,h} \;\ge\; -\,\lambda\,|\bar\partial^*\alpha|^2_{\omega,h} \;-\; \lambda^{-1}\,|e_{\bar\partial\eta}^*\alpha|^2_{\omega,h}. \end{align*} By the pointwise Kähler commutation identity for wedge and contraction on $(n,q)$-forms, \begin{align*} e_{\bar\partial\eta}e_{\bar\partial\eta}^* = [i\,\partial\eta\wedge\bar\partial\eta,\Lambda] \end{align*} on the relevant form degree, and therefore $|e_{\bar\partial\eta}^*\alpha|^2_{\omega,h} = ([i\,\partial\eta\wedge\bar\partial\eta,\Lambda]\alpha,\alpha)_{\omega,h}$. Integrating the inequality over $X$ and substituting, \begin{align*} \|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt\eta\,\bar\partial^*\alpha\|^2 + \lambda\,\|\bar\partial^*\alpha\|^2_{\lambda} \;\ge\; \big\langle\!\big\langle [\eta\, i\Theta_h(L) - i\partial\bar\partial\eta - \lambda^{-1} i\partial\eta\wedge\bar\partial\eta,\;\Lambda]\,\alpha,\alpha\big\rangle\!\big\rangle, \end{align*} where $\lambda\,\|\bar\partial^*\alpha\|^2_\lambda := \int_X \lambda\,|\bar\partial^*\alpha|^2_{\omega,h}\,dV_\omega$. Since $\int_X \eta\,|\bar\partial^*\alpha|^2_{\omega,h}\,dV_\omega + \int_X \lambda\,|\bar\partial^*\alpha|^2_{\omega,h}\,dV_\omega = \|\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\|^2$, and recognizing $\Theta_{\eta,\lambda}$ and $B = [\Theta_{\eta,\lambda},\Lambda]$ on the right, we arrive at the **twisted a priori inequality**: for all $\alpha \in \mathcal{D}^{n,q}(X,L)$, \begin{align*} \langle\!\langle B\alpha,\alpha\rangle\!\rangle \;\le\; \|\sqrt\eta\,\bar\partial\alpha\|^2 + \|\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\|^2 . \tag{$\star$} \end{align*} [guided] We have an identity, $(\mathrm{C})$, and we want a clean inequality bounding a curvature quadratic form by the weighted graph norm. Two adjustments are needed. First, the term $\|\sqrt\eta\,D'^*\alpha\|^2$ is a genuine nonnegative quantity we have no use for, so we simply drop it; this is the only place an inequality (rather than equality) enters from the $\partial$-side, and it costs us nothing because we only ever want a *lower* bound on the graph norm. Second, the surviving first-order term $2\mathrm{Re}\,\langle\!\langle\bar\partial^*\alpha,e_{\bar\partial\eta}^*\alpha\rangle\!\rangle$ has an indefinite sign and must be controlled. The standard device is the elementary inequality $2\mathrm{Re}\,(a,b) \ge -\lambda|a|^2 - \lambda^{-1}|b|^2$, valid pointwise for any $\lambda > 0$ — this is Cauchy–Schwarz combined with $2st \le \lambda s^2 + \lambda^{-1}t^2$. Why introduce a *function* $\lambda$ rather than a constant? Because the freedom to let $\lambda$ vary over $X$ is exactly what makes the resulting estimate flexible enough to be useful; $\lambda$ is the second piece of twisting data. Track what the two pieces of the bound do. The term $-\lambda|\bar\partial^*\alpha|^2$ is paired with the existing $\eta|\bar\partial^*\alpha|^2$ from the left side: moving it across, the weight on $\bar\partial^*\alpha$ is promoted from $\eta$ to $\eta + \lambda$. This is the precise origin of the factor $\eta+\lambda$ that will reappear in the norm of the solution. The term $-\lambda^{-1}|e_{\bar\partial\eta}^*\alpha|^2$ is a new curvature-type contribution; the pointwise wedge-contraction identity $e_{\bar\partial\eta}e_{\bar\partial\eta}^* = [i\partial\eta\wedge\bar\partial\eta,\Lambda]$ on $(n,q)$-forms gives $|e_{\bar\partial\eta}^*\alpha|^2 = ([i\partial\eta\wedge\bar\partial\eta,\Lambda]\alpha,\alpha)$ and identifies it as the action of the rank-one nonnegative $(1,1)$-form $i\partial\eta\wedge\bar\partial\eta$. Subtracting $\lambda^{-1}$ times it from the curvature is exactly the definition of $\Theta_{\eta,\lambda}$. Assembling, the right-hand side becomes $\langle\!\langle[\Theta_{\eta,\lambda},\Lambda]\alpha,\alpha\rangle\!\rangle = \langle\!\langle B\alpha,\alpha\rangle\!\rangle$, and we obtain $(\star)$. Notice that the entire content of the positivity hypothesis on $B$ is now visible: $(\star)$ asserts that the positive quadratic form $\langle\!\langle B\alpha,\alpha\rangle\!\rangle$ is dominated by the weighted graph norm. [/guided] [/step] [step:Extend the a priori inequality to the operator domain by completeness] Inequality $(\star)$ holds for compactly supported smooth forms. We extend it to all $\alpha$ in the relevant domain. Because $(X,\omega)$ is a complete Kähler manifold, smooth compactly supported forms are dense, in the graph norm $\alpha \mapsto \big(\|\alpha\|^2 + \|\bar\partial\alpha\|^2 + \|\bar\partial^*\alpha\|^2\big)^{1/2}$, in $\operatorname{Dom}(\bar\partial)\cap\operatorname{Dom}(\bar\partial^*)$ (citing a result not yet in the wiki: the *Andreotti–Vesentini density lemma* for complete Hermitian manifolds; one may take a sequence of Lipschitz cutoffs $\chi_\nu$ with $|\bar\partial\chi_\nu| \to 0$ uniformly, available precisely because $\omega$ is complete). Since $\eta$ and $\eta+\lambda$ are smooth, hence locally bounded, both sides of $(\star)$ are well-defined for any $\alpha \in \operatorname{Dom}(\bar\partial)\cap\operatorname{Dom}(\bar\partial^*)$ (allowing the value $+\infty$ on the right), and the inequality passes to such $\alpha$ by the density just stated. In particular, $(\star)$ holds for every $\alpha \in \ker\bar\partial \cap \operatorname{Dom}(\bar\partial^*)$. For such $\alpha$ we have $\bar\partial\alpha = 0$, so $(\star)$ reduces to \begin{align*} \langle\!\langle B\alpha,\alpha\rangle\!\rangle \;\le\; \|\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\|^2 . \tag{$\star\star$} \end{align*} [/step] [step:Convert curvature positivity into a bound on the functional $\alpha \mapsto \langle\!\langle f,\alpha\rangle\!\rangle$] Set the finite constant $C_0 := \int_X (B^{-1}f,f)_{\omega,h}\,dV_\omega < \infty$ (finiteness is the hypothesis). Fix $\alpha \in \ker\bar\partial \cap \operatorname{Dom}(\bar\partial^*)$. Since $B$ is pointwise self-adjoint and positive definite, $B^{1/2}$ is self-adjoint, and pointwise \begin{align*} (f,\alpha)_{\omega,h} = (B^{-1/2}f,\;B^{1/2}\alpha)_{\omega,h}. \end{align*} Applying the pointwise [Cauchy–Schwarz inequality](/theorems/432) and then the Cauchy–Schwarz inequality for the integral, \begin{align*} |\langle\!\langle f,\alpha\rangle\!\rangle| \;\le\; \int_X |B^{-1/2}f|_{\omega,h}\,|B^{1/2}\alpha|_{\omega,h}\,dV_\omega \;\le\; \Big(\int_X (B^{-1}f,f)_{\omega,h}\,dV_\omega\Big)^{1/2}\Big(\int_X (B\alpha,\alpha)_{\omega,h}\,dV_\omega\Big)^{1/2}, \end{align*} that is, $|\langle\!\langle f,\alpha\rangle\!\rangle|^2 \le C_0\,\langle\!\langle B\alpha,\alpha\rangle\!\rangle$. Combining with $(\star\star)$, \begin{align*} |\langle\!\langle f,\alpha\rangle\!\rangle|^2 \;\le\; C_0\,\|\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\|^2 \qquad \text{for all } \alpha \in \ker\bar\partial \cap \operatorname{Dom}(\bar\partial^*). \tag{$\dagger$} \end{align*} [/step] [step:Solve $\bar\partial u = f$ by Hahn–Banach and Riesz representation] Consider the linear subspace \begin{align*} E_0 := \big\{\, \sqrt{\eta+\lambda}\,\bar\partial^*\alpha \;:\; \alpha \in \ker\bar\partial \cap \operatorname{Dom}(\bar\partial^*),\ \sqrt{\eta+\lambda}\,\bar\partial^*\alpha \in L^2_{n,q-1}(X,L) \,\big\} \;\subseteq\; L^2_{n,q-1}(X,L), \end{align*} and define $\ell_0 : E_0 \to \mathbb{C}$ by $\ell_0\big(\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\big) := \langle\!\langle \alpha,f\rangle\!\rangle$. *$\ell_0$ is well-defined and bounded.* If $\alpha_1,\alpha_2 \in \ker\bar\partial \cap \operatorname{Dom}(\bar\partial^*)$ satisfy $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha_1 = \sqrt{\eta+\lambda}\,\bar\partial^*\alpha_2$, then $\bar\partial^*(\alpha_1-\alpha_2) = 0$ almost everywhere (as $\eta+\lambda > 0$), and $\alpha_1-\alpha_2 \in \ker\bar\partial \cap \operatorname{Dom}(\bar\partial^*)$; estimate $(\dagger)$ gives $|\langle\!\langle \alpha_1-\alpha_2,f\rangle\!\rangle|^2 = |\langle\!\langle f,\alpha_1-\alpha_2\rangle\!\rangle|^2 \le C_0\,\|\sqrt{\eta+\lambda}\,\bar\partial^*(\alpha_1-\alpha_2)\|^2 = 0$, so $\ell_0$ is well-defined. Boundedness is $(\dagger)$ itself: $|\ell_0(\beta)| \le C_0^{1/2}\,\|\beta\|$ for $\beta \in E_0$. By the [Hahn–Banach Theorem](/theorems/880), $\ell_0$ extends to a bounded linear functional $\ell$ on all of $L^2_{n,q-1}(X,L)$ with $\|\ell\| \le C_0^{1/2}$. By the [Riesz Representation Theorem](/theorems/221) in the [Hilbert space](/page/Hilbert%20Space) $L^2_{n,q-1}(X,L)$, there is a unique $g \in L^2_{n,q-1}(X,L)$ with \begin{align*} \ell(\beta) = \langle\!\langle \beta, g\rangle\!\rangle \quad \text{for all } \beta, \qquad \|g\| = \|\ell\| \le C_0^{1/2}. \end{align*} Define $u := \sqrt{\eta+\lambda}\,g$. Then $u \in L^2_{n,q-1}(X,L)$ and, since $(\eta+\lambda)^{-1}|u|^2_{\omega,h} = |g|^2_{\omega,h}$, \begin{align*} \int_X (\eta+\lambda)^{-1}\,|u|^2_{\omega,h}\,dV_\omega = \|g\|^2 \le C_0 = \int_X (B^{-1}f,f)_{\omega,h}\,dV_\omega, \end{align*} which is the asserted norm bound. *Verification that $\bar\partial u = f$ distributionally.* Let $\alpha \in \mathcal{D}^{n,q}(X,L)$ be an arbitrary test form. By the [Orthogonal Decomposition Theorem](/theorems/241) applied to the closed subspace $\ker\bar\partial \subseteq L^2_{n,q}(X,L)$, write $\alpha = \alpha' + \alpha''$ with $\alpha' \in \ker\bar\partial$ and $\alpha'' \in (\ker\bar\partial)^\perp$. For any $\gamma \in \operatorname{Dom}(\bar\partial)$ we have $\bar\partial\gamma \in \ker\bar\partial$ (as $\bar\partial^2=0$), so $\langle\!\langle \alpha'', \bar\partial\gamma\rangle\!\rangle = 0$; by definition of the adjoint this means $\alpha'' \in \operatorname{Dom}(\bar\partial^*)$ with $\bar\partial^*\alpha'' = 0$. Consequently $\alpha' = \alpha - \alpha'' \in \operatorname{Dom}(\bar\partial^*)$ and \begin{align*} \bar\partial^*\alpha' = \bar\partial^*\alpha = \vartheta\alpha, \end{align*} which, $\alpha$ being smooth and compactly supported, is itself smooth and compactly supported; hence $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha' \in L^2_{n,q-1}(X,L)$, i.e. $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha' \in E_0$ with $\alpha' \in \ker\bar\partial \cap \operatorname{Dom}(\bar\partial^*)$. Moreover $f \in \ker\bar\partial$ gives $\langle\!\langle \alpha'',f\rangle\!\rangle = 0$, so $\langle\!\langle \alpha,f\rangle\!\rangle = \langle\!\langle \alpha',f\rangle\!\rangle$. Therefore \begin{align*} \langle\!\langle \vartheta\alpha, u\rangle\!\rangle = \langle\!\langle \bar\partial^*\alpha', \sqrt{\eta+\lambda}\,g\rangle\!\rangle = \langle\!\langle \sqrt{\eta+\lambda}\,\bar\partial^*\alpha', g\rangle\!\rangle = \ell\big(\sqrt{\eta+\lambda}\,\bar\partial^*\alpha'\big) = \langle\!\langle \alpha',f\rangle\!\rangle = \langle\!\langle \alpha,f\rangle\!\rangle . \end{align*} Thus $\langle\!\langle \vartheta\alpha, u\rangle\!\rangle = \langle\!\langle \alpha,f\rangle\!\rangle$ for every $\alpha \in \mathcal{D}^{n,q}(X,L)$, which is exactly the statement that $\bar\partial u = f$ in the sense of distributions. Together with the norm bound established above, this completes the proof. [guided] We hold two facts: the analytic estimate $(\dagger)$, controlling $\langle\!\langle f,\alpha\rangle\!\rangle$ by $\|\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\|$ on closed forms, and the desired conclusion, the *existence* of $u$ solving $\bar\partial u = f$ with a weighted bound. The bridge is the duality between solving an equation and bounding a functional — the Hörmander method. Why phrase the functional on $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha$ rather than on $\bar\partial^*\alpha$? Because the target norm of the solution is the *weighted* norm $\int(\eta+\lambda)^{-1}|u|^2$. Writing $u = \sqrt{\eta+\lambda}\,g$ converts this weighted norm into the plain $L^2$ norm $\|g\|^2$, and the natural pairing $\langle\!\langle\bar\partial^*\alpha,u\rangle\!\rangle = \langle\!\langle\sqrt{\eta+\lambda}\,\bar\partial^*\alpha,g\rangle\!\rangle$ then lands precisely the weighted quantity $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha$ appearing in $(\dagger)$. This is why $\eta+\lambda$ — and not $\eta$ — governs the solution: it is the weight that $(\dagger)$ delivers. The functional $\ell_0$ on $E_0 = \{\sqrt{\eta+\lambda}\,\bar\partial^*\alpha\}$ is forced to send $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha \mapsto \langle\!\langle \alpha,f\rangle\!\rangle$. This placement of $f$ in the second slot is necessary because $\langle\!\langle\cdot,\cdot\rangle\!\rangle$ is linear in the first argument, so $\alpha \mapsto \langle\!\langle\alpha,f\rangle\!\rangle$ is complex-linear. Is this even consistent? Two forms $\alpha_1,\alpha_2$ might produce the same $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha$; we must check $\langle\!\langle \alpha_1,f\rangle\!\rangle = \langle\!\langle \alpha_2,f\rangle\!\rangle$. This is precisely where $(\dagger)$ is used as a *consistency* statement: it bounds $\langle\!\langle f,\cdot\rangle\!\rangle$ by the very quantity that vanishes, forcing well-definedness. The same inequality is the operator-norm bound $\|\ell_0\| \le C_0^{1/2}$. Hahn–Banach extends $\ell_0$ to the whole [Hilbert space](/page/Hilbert%20Space) without increasing its norm; Riesz then represents it by a vector $g$ with $\|g\| \le C_0^{1/2}$. Setting $u = \sqrt{\eta+\lambda}\,g$ immediately gives the weighted bound $\int(\eta+\lambda)^{-1}|u|^2 = \|g\|^2 \le C_0$. The final task is to confirm we actually solved the equation, not merely on closed forms but as a distributional identity. We have $\langle\!\langle\bar\partial^*\alpha,u\rangle\!\rangle = \langle\!\langle f,\alpha\rangle\!\rangle$ for $\alpha \in \ker\bar\partial$; we want it for *all* test forms $\alpha$. Here the [orthogonal decomposition](/theorems/436) $\alpha = \alpha' + \alpha''$ relative to the closed subspace $\ker\bar\partial$ does the work. The component $\alpha''$ is orthogonal to $\ker\bar\partial$, hence orthogonal to $\operatorname{Range}(\bar\partial) \subseteq \ker\bar\partial$; orthogonality to the range of $\bar\partial$ is *by definition* membership in $\operatorname{Dom}(\bar\partial^*)$ with $\bar\partial^*\alpha'' = 0$. What would go wrong without this? If $\alpha''$ contributed to $\bar\partial^*\alpha$, the closed-form estimate could not reach general test forms. But it does not: $\bar\partial^*\alpha' = \bar\partial^*\alpha$, and crucially this equals $\vartheta\alpha$, which is *compactly supported* because $\alpha$ is — so $\sqrt{\eta+\lambda}\,\bar\partial^*\alpha'$ is genuinely in $E_0$ even when $\eta+\lambda$ is unbounded. (This is the one subtlety with unbounded twisting data, and testing against compactly supported $\alpha$ resolves it cleanly.) Meanwhile $f \in \ker\bar\partial$ kills the pairing with $\alpha''$. The two facts combine to give $\langle\!\langle\vartheta\alpha,u\rangle\!\rangle = \langle\!\langle \alpha,f\rangle\!\rangle$ for every $\alpha \in \mathcal{D}^{n,q}$, which is the definition of $\bar\partial u = f$ in the distributional sense under the convention that the Hilbert inner product is linear in the first argument. [/guided] [/step]