Convergence analysis of a conforming adaptive finite element method for an obstacle problem (Q2454717)

The authors develop an adaptive finite element algorithm for the elliptic obstacle problem: Find \(u \in K\) such that \(a(u,v-u) \geq b(v-u)\) for all \(v \in K\), where \(V= H^1_0(\Omega), K= \{ v \in V| v \geq \chi \text{ a.e. in } \Omega\}, a(v,w):=(\nabla v, \nabla w)_{0,\Omega}, b(v):=(f,v)_{0,\Omega}, f\in H^1(\Omega)\). An obstacle is defined in \(\Omega\) by an affine function \(\chi\) on \(\bar \Omega\) with \(\chi \leq 0\) on \(\Gamma\). The analysis of the adaptive method shows that in general we can expect an energy reduction property, but not necessarily a guaranteed reduction in the energy norm. The proposed analysis uses the equivalence of two well-known error estimates for the obstacle problem which depend on the jumps on the edges of the mesh. The adaptive loop focusing on a residual-type a posteriori error estimator and a bulk criterion selecting edges for refinement is presented. The discrete local efficiency for the reliability of the estimator is shown. The R-linear convergence of the part of the energy above its minimal value is established, if there is appropriate control of the data oscillations. The main result of the paper is the following: The energy reduction error estimate for the discrete variational problem on the level \(l\: \lim_{l\rightarrow \infty} \varepsilon _l^{(1/l)} <1\) is shown. Here \(\varepsilon_l=[\Pi(u_l)-\Pi(u)]\) with the energy function: \(\Pi(v):=\frac{1}{2}a (v,v)-b(v)\). The precise proof of this assertion is presented. Numerical results illustrating the performance of the error estimator are also proposed.