Error estimation for a class of differential eigenproblems (Q1096356)

An error estimate is derived which may be used to control the computational steps in a scalar first order nonlinear eigenvalue problem. The system considered has the form \((i)\quad \phi '=F(x,\phi,\lambda),\) \(a\leq x\leq b\), (ii) \(\phi (a)=\alpha\), \(\phi (b)=\beta\). The computational method is to choose \(\lambda\) to minimize \(f(\lambda)=| \phi (c-,\lambda)-\phi (c+,\lambda)|\) where \(\phi (c\pm,\lambda)\) are obtained by approximate integration to the right from \(x=a\) and to the left from \(x=b\), \(a<c<b\). The difference between the true and approximate eigenvalues \(| \lambda -{\bar \lambda}|\) is shown to have an error bound of the form \(\sum_{1}| \bar M(x_ i)| | e_ i|\) where \(x_ i\) are the mesh points, \(e_ i\) a local error at each step of the computation and \(\bar M(x)\) is an integrating factor associated with a linearized form of the equation for the variation \(\phi\) (x,\(\lambda)\)-\(\Phi\) (x,\({\bar \lambda}\)) where \(\phi\) (x,\(\lambda)\) is the exact solution and \(\Phi\) (x,\({\bar \lambda}\)) the computed solution. The method is applied to the Sturm-Liouville problem after reduction to a first order system of the form (i), (ii) using either the Riccati or Prüfer transformation. The author has written and tested routines which use the above error bound on Sturm-Liouville problems [IMA J. Numer. Anal. 6, 103-123 (1986; Zbl 0591.65057)].