Finite Element Matrix Sequences: the Case of Rectangular Domains

In the present paper, we consider a preconditioning strategy for Finite Element (FE) matrix sequences {A _n (a)} _n discretizing the elliptic problem $$\left\{ \begin{gathered} A_a u \equiv ( - )^k \nabla ^k a(x,y)\nabla ^k u(x,y)] = f(x,y),{ }(x,y) \in \Omega = (0,1)^2 , \hfill \\ \left. {\left( {\frac{{\partial ^s }}{{\partial v^s }}u(x,y)} \right)} \right|_{\partial \Omega } \equiv 0,{ }s = 0,...,k - 1,{ }^{^{^{^{^{^{(1)} } } } } } \hfill \\ \end{gathered} \right.$$ with a(x,y) being a uniformly positive function and ν denoting the unit outward normal direction. More precisely, in connection with preconditioned conjugate gradient (PCG) like methods, we define the preconditioning sequence: {P _n (a)} _n , P _n (a):= $$\widetilde D$$ _n ^1/2(a)A _n (1) $$\widetilde D$$ _n ^1/2(a), where $$\widetilde D$$ _n (a) is the suitable scaled main diagonal of A _n (a). In fact, under the mild assumption of Lebesgue integrability of a(x), the weak clustering at the unity of the corresponding preconditioned sequence is proved. Moreover, if a(x,y) is regular enough and if a uniform triangulation is considered, then the preconditioned sequence shows a strong clustering at the unity so that the sequence {P _n (a)} _n turns out to be a superlinear preconditioning sequence for {A _n (a)} _n . The computational interest is due to the fact that the computation with A _n (a) is reduced to computations involving diagonals and two-level Toeplitz structures {A _n (1)} _n with banded pattern. Some numerical experimentations confirm the efficiency of the discussed proposal.