Non-intrusive uncertainty quantification using reduced cubature rules

L. M.M. van den Bos, B. Koren, R. P. Dwight

Research output: Contribution to journalArticleScientificpeer-review

9 Citations (Scopus)

Abstract

For the purpose of uncertainty quantification with collocation, a method is proposed for generating families of one-dimensional nested quadrature rules with positive weights and symmetric nodes. This is achieved through a reduction procedure: we start with a high-degree quadrature rule with positive weights and remove nodes while preserving symmetry and positivity. This is shown to be always possible, by a lemma depending primarily on Carathéodory's theorem. The resulting one-dimensional rules can be used within a Smolyak procedure to produce sparse multi-dimensional rules, but weight positivity is lost then. As a remedy, the reduction procedure is directly applied to multi-dimensional tensor-product cubature rules. This allows to produce a family of sparse cubature rules with positive weights, competitive with Smolyak rules. Finally the positivity constraint is relaxed to allow more flexibility in the removal of nodes. This gives a second family of sparse cubature rules, in which iteratively as many nodes as possible are removed. The new quadrature and cubature rules are applied to test problems from mathematics and fluid dynamics. Their performance is compared with that of the tensor-product and standard Clenshaw–Curtis Smolyak cubature rule.

Original languageEnglish
Pages (from-to)418-445
Number of pages28
JournalJournal of Computational Physics
Volume332
DOIs
Publication statusPublished - 1 Mar 2017

Keywords

  • Cubature rules
  • Numerical integration
  • Uncertainty quantification

Fingerprint Dive into the research topics of 'Non-intrusive uncertainty quantification using reduced cubature rules'. Together they form a unique fingerprint.

Cite this