An evaluation of the gradient-weighted moving-finite-element method in one space dimension (Q1206588): Difference between revisions

For several kinds of parabolic and hyperbolic partial differential equations involving fine-scale structures the moving-grid methods are becoming increasingly popular. The moving-finite-element method, originally proposed by \textit{K. Miller} and \textit{R. N. Miller} [SIAM J. Numer. Anal. 18, 1019-1032 (1981; Zbl 0518.65082) and ibid. 18, 1033-1057 (1981; Zbl 0518.65083)], belongs to these methods. The main difficulty of application of moving-grid methods is the threat of grid distorsion which can only be avoided by using penalty terms which are artificial and invariably involve parameter tuning. The parameter tuning is very important not only to provide for a safe automatic grid-point selection but also for efficiency in the time-stepping process. The gradient weighted moving-finite-element method has some promising properties to reduce the need of tuning. The gradient-weighting amounts to the use of weighting functions in the finite-element formulation that depend on the gradient of the solution. This treatment results in a more robust process in which the parameter tuning becomes easier and also critical. In this paper the gradient-weighted moving-finite-element method is tested on a set of five difficult example problems with steep moving fronts from different areas of application. The robustness of the preconditioned gradient-weighted moving-finite-element method compared with the moving-finite-element method has been observed. The authors do not advocate to use the gradient-weighted moving-finite- element method as a general purpose method for all kinds of evolutionary problems. The disadvantage is not only the much more complex nonlinear system resulting from the addition of the strongly nonlinear grid equation, but also the fact that this method does not get on with the method-of-lines approach. The authors advise to use the gradient-weighted moving-finite-element method mainly when the solution is known to have steep moving fronts over the whole time-integration interval.