8aa5e05c5cb420cb.tex
1: \begin{abstract}
2:   We design a two-scale finite element method (FEM) for linear
3:   elliptic PDEs in non-divergence form $A(x) : D^2 u(x) =
4:   f(x)$. The fine scale is given by the meshsize $h$ whereas the
5:     coarse scale $\epsilon$ is dictated by an integro-differential
6:     approximation of the PDE. We show that the FEM satisfies
7:   the discrete maximum principle (DMP) for any uniformly positive
8:   definite matrix $A(x)$ provided that the mesh is weakly
9:   acute. Combining the DMP with weak operator consistency of the
10:     FEM, we establish convergence of the numerical solution
11:     $u_h^\epsilon$ to the viscosity solution $u$ as $\epsilon, h\to0$
12:     and $\epsilon\ge C h|\ln h|$.
13: 
14:  We develop a discrete Alexandroff-Bakelman-Pucci (ABP) estimate
15:   which is suitable for finite element analysis. Its proof relies on
16: a geometric interpretation of Alexandroff estimate and control of the
17: measure of the sub-differential of piecewise linear functions in terms
18: of jumps, and thus of the discrete PDE. The discrete ABP estimate
19: leads, under suitable regularity assumptions on $A$ and $u$,
20: to the estimate
21: %
22: \begin{equation*}
23: \inftynorm{u - u^{\epsilon}_h} \leq \; C \big( h^2 \Abs{\ln h}
24: \big )^{\alpha /(2 + \alpha)} \qquad 0< \alpha \leq 2,
25: \end{equation*}
26: %
27: provided $\epsilon \approx (h^2 |\ln h|)^{1/(2+\alpha)}$.
28: Such a convergence rate is at best of order 
29: $ h \Abs{\ln h}^{1/2}$, which turns out to be quasi-optimal.
30: 
31: \end{abstract}
32: