Gradient estimate for solutions of second-order elliptic equations (Q2668099)

It is known since the works of De Giorgi and Nash that weak solutions \(u \in H^{1,2}\) for equations \(\partial_{j} (a_{ij}(x) \partial_{i } u)=0\) are Hölder continuous when the coefficient matrix \(A=(a_{ij})\) is symmetric with real, bounded, measurable entries. Stronger conditions on the coefficients are known to force solutions to be of class \(C^{1}\). In the intermediate region, it is interesting to investigate when solutions are guaranteed to be Lipschitz or to estimate the rate of blow-up of their gradient. In this article the authors extend on one of their previous results [\textit{V. Maz'ya} and \textit{R. McOwen}, J. Differ. Equations 250, No. 2, 1137--1168 (2011; Zbl 1225.35070)] to provide sharp estimates on the possible blow-up rate of the gradient of the solution at a point. This is done under the additional hypothesis of {\em square-Dini continuous} coefficients; that is, \[ \sup_{|x| = r} |a_{ij}(x) -\delta_{ij}| \le \omega(r) \ \ \mbox{ as } \ \ r\to 0, \] where the modulus of continuity \(\omega(r) \) is a nondecreasing, positive, continuous function for \(r > 0\) satisfying \(\int_{0}^{1} \frac{ \omega^{2}(r)}{r} \ dr < \infty.\) These estimates are controlled by the behavior near the origin of the smallest eigenvalue (as a function of \(r\)) of the matrix \((R(r)+R^{*}(r))\), where for each \(r\) fixed, \(R(r)\) is the average on the unit sphere of the matrix-valued function \[\theta \in \mathbb{S}^{n-1} \to A(r\theta) -nA(r\theta)\theta \otimes \theta.\]