\(L^2\) estimates and existence theorems for the \(\overline{\partial}\) operators in infinite dimensions. I (Q2145843)

The paper develops modifications with significant improvements of \textit{L. Lempert}'s breakthrough work in [J. Am. Math. Soc. 11, No. 3, 485--520 (1998; Zbl 0904.32014)] concerning the solvability and solution methods for the inhomogeneous Cauchy-Riemann equation \(\overline{\partial} u = f\) in the infinite dimensional setting. It was remarked in [loc. cit.] that ``up to now not a single Banach space and an open subset therein have been proposed where this equation could be proved to be solvable under reasonably general conditions on \(f\).'' Actually, the present work provides a satisfactory approach in the case of the \((s,t)\) forms over an infinite dimensional \(\ell^p\) \((1\le p<\infty)\) space equipped with a Hilbert norm defined for functions on \(L^2(\ell^p,P)\) in the weak sense where \(P\) is a Gaussian probability measure constructed as follows. Fixing a sequence \(\{ a_i\}_{i=1}^\infty\) of positive numbers with finite sum, \(P\) is the Kolmogorov extension of the measures \(\mathcal{N}_{a_i}(B) := (2\pi a_i^2)^{-1} \int_B e^{-(x^2+y^2)/(2a_i)} dx\, dy\) on the Borel subsets of \(\mathbb{C}\equiv \mathbb{R}^2\) with \(P =\prod_{i=1}^\infty\mathcal{N}_{a_i}\) from \(\mathbb{C}^\infty := \prod_{i=1}^\infty\mathbb{C}\). We have \(P(\ell^p) =1\) for \(1\le p<\infty\) and one can embed \(L^2(\mathbb{C}^n, \prod_{i=1}^n\mathcal{N}_{a_i})\) into \(L^2(\ell^p,P)\) as a closed subspace by identifying the monomials \(\prod_{k=1}^n z_k^{i_k} :\mathbb{C}^n\!\to\!\mathbb{C}\) with the cylindric functions \(\mathbf{z}^I : \ell^p \!\ni\! (\zeta_1,\zeta_2,\ldots) \mapsto \prod_{k=1}^n \zeta_k^{i_k}\). The authors introduce the \(\overline{\partial}\) operator by means of weak derivatives: Given \(f\in L^2(\ell^p,P)\) such that there exists a sequence \(\{ f_j \}_{j=1}^\infty \subset L^2(\ell^p, P)\) with \(\int_{\ell^p} f_j \cdot \overline{\varphi} \, dP = -\int_{\ell^p} f_j \big[ \partial\varphi/\partial z_j - \varphi\cdot \overline{z_j}(2 a_j^2) \big]\, dP\) for all \(\varphi\in \mathcal{C}^\infty_c\) and \(\sum_{j=1}^\infty a_j^2 \int_{\ell^p} \vert f_j \vert^2 dP <\infty\), for \(\mathbf{z},\mathbf{z}^1 \in \ell^p\) we set \(\overline{\partial} f(\mathbf{z},\mathbf{z}^1) := \sum_{j=1}^\infty f_j(\mathbf{z}) \overline{z^1_j}\). In this manner, we obtain a densely defined closed linear operator \(L^2(\ell^p,P)\to L^2(\ell^p\times\ell^p,P\times P)\) and, with the usual determinant techniques, we define the weak extension \(f(\mathbf{z},\mathbf{z}^1,\ldots,\mathbf{z}^{s+t}) \mapsto \overline{\partial} f(\mathbf{z},\mathbf{z}^1,\ldots,\mathbf{z}^{s+t+1})\) for \(\overline{\partial}\) from the smooth \((s,t)\)-forms as a closed linear operator \(L^2_{(s,t)}(\ell^p) \to L^2_{(s,t+1)}(\ell^p)\) between the Hilbert spaces of the natural \(L^2\)-spaces of \((s,t)\)- resp. \((s,t+1)\)-forms on \(\ell^p\) such that \(\overline{\partial}\overline{\partial} = 0\) and the following Hörmander type estimate holds: For the restrictions \(T := \overline{\partial} \vert L^2_{s,t}(\ell^p)\), \(S := \overline{\partial} \vert L^2_{s,t+1}(\ell^p)\) of the \(\overline{\partial}\) operator, we have \(\mathrm{range}(T) \subset\mathrm{kernel}(S)\), the latter being a closed subspace in \(L^2_{(s,t+1)}(\ell^p)\), and on the dense linear submanifold \[ \mathcal{D}_{(s,t)} := \bigcup_{n=1}^\infty \Big\{ \sum_{\vert I\vert=s,\ \vert J\vert=t,\ \max\{ \vert I\vert,\vert J\vert\} \le n}^{\text{increasing\ indices}} f_{I,J} \, d z^I \wedge d\overline{z}^J : f_{I,J}\in\mathcal{C}^\infty_c\Big\} \] of \(L^2_{(s,t)}(\ell^p)\), we have \[ \Vert f\Vert^2_{{L^2(s,t+1)}(\ell^p)} \le \Vert T^* f\Vert^2_{{L^2(s,t)}(\ell^p)} + \Vert Sf\Vert^2_{{L^2(s,t+2)}(\ell^p)}. \tag{\(*\)} \] Solving the equation \(\overline{\partial} u =f\) on \(\ell^p\) is equivalent to prove the coincidence \(\mathrm{range}(T) = \mathrm{kernel}(N)\) and to extend \((*)\) to all functions \(f\in\mathrm{domain}(S)\cap\mathrm{domain}(T^*)\). In the case of finite dimensions, this extension was achieved by Hörmander with the classical Friedrichs mollifier technique which relies heavily upon an argument not available in infinite dimension, namely, the translation invariance of the Lebesgue measure. The authors overcome this difficulty by means of approximating the forms \[ f = \sum_{\vert I\vert=s,\ \vert J\vert=t}^{\text{incr.\ indices}} f_{I,J} \, d z^I \wedge d\overline{z}^J \in \mathrm{kernel}(S) \] by sequences \[ M_n f := \sum_{\vert I\vert=s,\ \vert J\vert=t,\ \max\{ \vert I\vert,\vert J\vert\} \le n}^{\text{incr.\ indices}} \Pi_n f_{I,J} \, d z^I \wedge d\overline{z}^J, \] where \(\Pi_n\) denotes the orthogonal projection of \ \(L^2(\ell^p,P)\) onto \(L^2(\mathbb{C}^n,\prod_{i=1}^n \mathcal{N}_{a_i})\) as a closed embedded subspace. Due to the cylindric construction of \(P\), we can identify \(M_n f\) as an element of \(L^2_{(s,t+1)}(\mathbb{C}^n)\) belonging to the kernel of the restriction \(S_n: L^2_{(s,t+1)}(\mathbb{C}^n) \to L^2_{(s,t+2)}(\mathbb{C}^n)\) of the \(\overline{\partial}\) operator in the setting where \(P\) is replaced with \(\mathcal{N}\), \(\mathcal{C}^\infty_c\) with \(\mathcal{C}^\infty_c(\mathbb{C}^n)\), respectively. In terms of the analogous restriction \(T_n: L^2_{(s,t)}(\mathbb{C}^n) \to L^2_{(s,t+1)}(\mathbb{C}^n)\) of \(\overline{\partial}\), an analog of the estimate \((*)\) is established and from the proof it is discovered that \(\mathrm{range}(T_n) = \mathrm{kernel}(S_n)\), moreover for each \(g \in\mathrm{kernel}(S_n)\) there exists \(u\in\mathrm{domain}(T_n)\) with \(T_n u = g\) and \(\Vert u\Vert_{L^2_{(s,t+1)}(\mathbb{C}^n)} \le \Vert T^*_n g\Vert_{L^2_{(s,t)}(\mathbb{C}^n)}\). In particular, we can write \(M_n f = T_n u_n\) with suitable \(u_n \in\mathrm{dom}(T_n)\) such that \(\Vert u\Vert_{L^2_{(s,t)}(\mathbb{C}^n)} \le \Vert M_n f\Vert_{L^2_{(s,t+1)}(\mathbb{C}^n)} = \Vert M_f \Vert_{L^2_{(s,t+1)}(\ell^p)} \le \Vert f \Vert_{L^2_{(s,t+1)}(\ell^p)}\), and as a consequence of the separability of \(L^2_{(s,t)}(\ell^p)\) and the Banach-Alaoglu theorem, we can find a weakly convergent subsequence \(u_{n_k}\) in \(L^2_{(s,t)}(\ell^p)\). The main result (Theorem 5.1) states that given \(f\in \mathrm{kernel}(S)\), any sequential weak accumulation point \(u \in L^2_{(s,t)}(\ell^p)\) of the sequence \(u_n\) with \(T_n u_n = M_n f\) satisfies the relations \(u\in \mathrm{domain}(T)\), \(Tu =f\) with \(\Vert u\Vert_{L^2_{(s,t)}(\ell^p)} \le \Vert f\Vert_{L^2_{(s,t+1)}(\ell^p)}\). The paper finishes with revisiting Lempert's counterexample in [\textit{L. Lempert}, J. Am. Math. Soc. 11, No. 3, 485--520 (1998; Zbl 0904.32014)] for a \((0,1)\) form \(f_0\) of class \(C^{p-1}\) on \(\ell^p\) with \(p\in\mathbb{N}\) for which the \(\overline{\partial}\) equation admits no continuous solution on any non-empty open subset of \(\ell^p\). Even in this case \(f_0 \in\mathrm{kernel}(\overline{\partial})\) and, according to Theorem 5.1, there exists \(u_0\in \mathrm{domain}(\overline{\partial})\) such that \(\overline{\partial} u_0 = f_0\).