Automated Parameters for Troubled-Cell Indicators Using Outlier Detection

More Info
expand_more

Abstract

In Vuik and Ryan [J. Comput. Phys., 270 (2014), pp. 138–160] we studied the
use of troubled-cell indicators for discontinuity detection in nonlinear hyperbolic partial differential equations and introduced a new multiwavelet technique to detect troubled cells. We found that these methods perform well as long as a suitable, problem-dependent parameter is chosen. This parameter is used in a threshold which decides whether or not to detect an element as a troubled cell. Until now, these parameters could not be chosen automatically. The choice of the parameter has an impact on the approximation: it determines the strictness of the troubled-cell indicator. An inappropriate choice of the parameter will result in the detection (and limiting) of too few or too many elements. The optimal parameter is chosen such that the minimal number of troubled cells is detected and the resulting approximation is free of spurious oscillations. In this paper we will see that for each troubled-cell indicator the sudden increase or decrease of the indicator value with respect to the neighboring values is important for detection. Indication basically reduces to detecting the outliers of a vector (one dimension) or matrix (two dimensions). This is done using Tukey’s boxplot approach to detect which coefficients in a vector are straying far beyond the others [J. W. Tukey, Exploratory Data Analysis, Addison-Wesley, 1977]. We provide an algorithm that can be applied to various troubled-cell indication variables. With this technique, the problem-dependent parameter that the
original indicator requires is no longer necessary as the parameter will be chosen automatically.