Asymptotically correct error estimation for collocation methods applied to singular boundary value problems (Q2486677): Difference between revisions

This paper is concerned with the numerical solution of singular boundary value problems (BVPs) with the form \(z'(t)= t^{-1} M(t) z(t)+ f(t,z(t)), \) \(t \in (0,1]\), \( B_0 z(0) + B_1 z(1) = \beta \) where \( z:[0,1] \to {\mathbb R}^m\) is a real function, \( M(t)\) is an \( n \times n\) matrix that can be written as \( M(t)=M(0)+t C(t)\) with a continuous \( C(t)\) and \(f\) is an \(n\)--dimensional smooth function. Further \(B_0\) and \(B_1\) are constant matrices so that the above BVP is a well posed problem in the sense given in the paper of \textit{F. R. de Hoog} and \textit{R. Weiss} [SIAM J. Numer. Anal. 13, 775-813 (1976; Zbl 0372.65034)]. For the above class of problems the author studies a posteriori error estimations for collocation methods with an even number of collocation points based on the defect correction principle, extending previous results on this subject in which \(M(0)\) have both eigenvalues with negative and positive real parts. A crucial point in the derivation of the desired error estimate is suitable decomposition of this error into two parts that can be studied separately. As a final conclusion the (asymptotically correct) error estimate for collocation with polynomials of degree \(m\) is \(\text{ O}(| \log h| ^{n_0-1} \; h^{m+1} )\) with some positive integer \( n_0\).