Galerkin and Runge-Kutta methods: unified formulation, a posteriori error estimates and nodal superconvergence (Q634613)

The authors consider a unified formulation of several standard implicit one-step time-stepping numerical integration methods, in particular collocation and perturbed collocation implicit Runge-Kutta schemes, as well as continuous and discontinuous Galerkin methods. All of them are formulated as a unique abstract method applied to the linear differential equation \( u'(t) + A u(t) = f(t)\), \(u(0) = u^0\), \(0 < t < T\), where \(A\) is a positive definite, self-adjoint, linear operator defined in a Hilbert space. The purpose then is to derive a posteriori error estimates at the nodes of the time partition. In the process of constructing a unified formulation for all methods, an important role is played by the concept of reconstruction, particularly for the discontinuous Galerkin method. Once a unified abstract method has been formulated, a representation formula for the error at the nodes is proposed taking profit of the advantages the unified formulation presents. This formula is subsequently used to establish a posteriori error estimates at the nodes for the different methods considered. Since these estimates usually require quite demanding discrete compatibility conditions, it is then natural to examine them for each class of methods. This is done in detail in particular for the Galerkin and collocation methods.