The regularity problem for second order elliptic operators with complex-valued bounded measurable coefficients (Q2339325)

The paper is an important step in the authors' impressive research program on non-smooth elliptic boundary value problems. It deals with the problem \(Lu = 0\) in the upper half space \(\mathbb{R}^{n} \times (0,\infty)\), where \(L = \text{div} A \nabla\) is a uniformly elliptic operator with \(t\)-independent complex-valued coefficients. Given a smooth compactly supported datum \(f\) on the boundary, this problem is known to have a unique weak solution that admits \(f\) has an (appropriately defined) boundary trace (see [\textit{C. E. Kenig} and \textit{D. J. Rule}, Trans. Am. Math. Soc. 361, No. 1, 125--160 (2009; Zbl 1178.35155)] and Section 2 of the paper under review). The Dirichlet problem is then said to be solvable in \(L^{q}\) if the weak solution satisfies a non-tangential maximal function estimate \(\|N_{*}(u)\|_{q} \lesssim \|f\|_{q}\). The regularity problem, on the other hand, is said to be solvable in \(L^{p}\) if the gradient of the weak solution satisfies a (modified) non-tangential maximal function estimate \(\|\tilde{N}(\nabla u)\|_{p} \lesssim \|\nabla_{\parallel} f\|_{p}\). The paper under review confirms the commonly held belief that the two problems are dual to each other, by showing that, given \(p\in(1,2]\), the regularity problem is solvable in \(L^p\) if and only if the Dirichlet problem for \(L^{*}\) is solvable in \(L^{p'}\) (and the weak solution satisfies a square function estimate). This had been previously established only in special cases involving real coefficients. In another paper [\textit{S. Hofmann} et al., J. Am. Math. Soc. 28, No. 2, 483--529 (2015; Zbl 1326.42028)] the authors have proved that, for real coefficients, there always exists a \(p'\in (2,\infty)\) such that the Dirichlet problem is solvable in \(L^{p'}\). The paper under review thus includes, as a striking corollary, the fact that the regularity problem is always solvable for some \(p \in (1,2)\) when the coefficients are small complex perturbations of real coefficients. The authors prove much more than these solvability results, and consider the related well-posedness issues, as well as layer potential representations. The relevant subtleties, as well as some key aspects of the history of the problem, are presented very nicely in the introduction. At the technical level, one can notice the following aspects. First of all, the existence of De Giorgi-Nash-Moser bounds on the weak solution is a standing assumption. This holds for small perturbations of real coefficients, but not for general complex coefficients. Secondly, the authors establish, and fundamentally use, a partial replacement of Rellich inequalities in \(L^p\): if the Dirichlet problem is solvable in \(L^{p'}\) (and the weak solution satisfies a square function estimate) then \[ \|\partial _{\nu_{A}} u\|_{p} \lesssim \|\nabla_{\parallel} f\|_{p}. \] This is very much a result of independent interest. Its proof is based on an \(L^p\) version of the solution of the Kato problem (see [\textit{P. Auscher} et al., Ann. Math. (2) 156, No. 2, 633--654 (2002; Zbl 1128.35316)] for the case \(p=2\), and e.g. [\textit{P. Auscher} and \textit{P. Tchamitchian}, Square root problem for divergence operators and related topics. Astérisque 249. Paris: Société Mathématique de France (1998; Zbl 0909.35001)] for the extension to other values of \(p\) in the presence of pointwise Gaussian bounds, which hold here thanks to the De Giorgi-Nash-Moser assumption) along with tent space estimates. This Rellich inequality is combined with layer potentials estimates that prove that the solution of the Dirichlet problem satisfies \(\|\tilde{N}(\nabla u)\|_{p} \lesssim \|\nabla_{\parallel} f\|_{p}+\|\partial _{\nu_{A}} u\|_{p}\).