In this paper, we revisit the old problem of compact finite difference approximations of the homogeneous Dirichlet problem in dimension 1. We design a large and natural set of schemes of arbitrary high order, and we equip this set with an algebraic structure. We give some general criteria of convergence and we apply them to obtain two new results. On the one hand, we use Padé approximant theory to construct, for each given order of consistency, the most efficient schemes and we prove their convergence. On the other hand, we use diophantine approximation theory to prove that almost all of these schemes are convergent at the same rate as the consistency order, up to some logarithmic correction.
IntroductionMany decades ago, compact finite differences methods were widely studied. Nowadays, we can find a huge literature about these methods that are widely applied and used for the approximation of partial differential equations (see, for example, [3] or [10]). In particular, we can find a lot of examples of accurate schemes for elliptic problems and many classical mathematical arguments are proposed to prove their convergence (monotonicity, energy, green functions, ...). However, it seems that there is not general and algebraic study of compact finite difference schemes for elliptic problems, equivalent to what we can find, for example, for the Runge Kutta methods applied to Cauchy problems (general stability criteria, algebraic order conditions using Hopf algebras and trees as we can see in [6] or [5]).As the field of elliptic problems is clearly too wide, we propose, in this paper, a general study of a large and natural class of compact finite difference schemes of high order for the homogeneous Dirichlet problem in dimension 1. In this context, a compact finite difference scheme is a linear system of the form