Maximum-norm error analysis of a non-monotone FEM for a singularly perturbed reaction-diffusion problem (Q996800)

This paper is concerned with the numerical solution of singularly perturbed one-dimensional boundary value problems of the type \[ - \varepsilon^2 u'' + r u = f \text{ in } (0,1), u(0)= u(1)=0, \] where \( r,f \in {\mathcal C}^2[0,1]\), \( 0 << \varepsilon < 1\), \( r(x) > \rho^2 > 0 \) for \( x \in [0,1]\), whose solution possess two layers of width \({\mathcal O} ( \varepsilon \log \varepsilon^{-1} )\) at the endpoints of the domain. The author proposes an FE discretization in an arbitrary mesh \( ( x_j )_{j=0}^N \) in \([0,1]\), where the test function \( \varphi_i\) is the standard hat function associated to the \(i\)-th mesh node. It is found that this discretization leads to a finite difference scheme that -- unlike the standard central difference schemes -- is not inverse monotone. In this context the author proves that it is stable in the maximum norm, and for the layer adapted meshes proposed by Shishkin and Bakhvalov some convergence results are established. The paper ends with the numerical results obtained for two test problems with the above meshes.