scientific article; zbMATH DE number 3858319

From MaRDI portal
Publication:3326751





[For the entire collection see Zbl 0499.00015.] Motivated by a parameter-dependent nonlinear boundary value problem from the theory of thermal combustion, the authors consider the problem of computing turning points of solution branches of nonlinear equations \(g(x,\lambda)=0\) (x in some Banach space, \(\lambda\in {\mathbb{R}})\). First, they review the basics of the theory of turning points and derive the conditions for (simple) quadratic and cubic turning points. For computing turning points, they consider (as usual) an extended system that has the turning point \((x_ 0,\lambda_ 0)\) as an isolated solution. For quadratic turning points, the extended system considered in this paper is given by \(g(x,\lambda)=0\), \(\psi g_ x(x,\lambda)\phi =0,\) where \(\phi\) and \(\psi\) are right- and left-eigenvectors for \(g_ x(x,\lambda)\), respectively, corresponding to a simple eigenvalue \(\sigma =\sigma(x,\lambda)\) with \(\sigma(x_ 0,\lambda_ 0)=0.\) For the finite-dimensional case, i.e., \(x\in {\mathbb{R}}^ n\), the authors set up Newton's method for their extended system (which of course converges locally quadratically). They claim that each Newton step in their procedure is (slightly) less time-consuming than one Newton step for other extended systems like the one due to Seydel. The key point in their analysis seems to be a method for obtaining the eigenvectors \(\phi\) and \(\psi\) that have to be calculated in each Newton step approximately without actually having to solve an eigenvalue problem. However, there are no remarks about the effect of the use of these approximate eigenvectors instead of the exact ones on the whole procedure. Using essentially the same ideas, the authors treat the problem of computing cubic turning points in a similar way.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3326751)