Derivative based global sensitivity measures

Serge Kucherenko (1) , Bertrand Iooss (2, 3, 4)

1 Imperial College London
2 IMT - Institut de Mathématiques de Toulouse UMR5219
3 EDF R&D MRI - Management des Risques Industriels
4 GdR MASCOT-NUM - Méthodes d'Analyse Stochastique des Codes et Traitements Numériques

The method of derivative based global sensitivity measures (DGSM) has recently become popular among practitioners. It has a strong link with the Morris screening method and Sobol' sensitivity indices and has several advantages over them. DGSM are very easy to implement and evaluate numerically. The computational time required for numerical evaluation of DGSM is generally much lower than that for estimation of Sobol' sensitivity indices. This paper presents a survey of recent advances in DGSM concerning lower and upper bounds on the values of Sobol' total sensitivity indices $S_{i}^{tot}$. Using these bounds it is possible in most cases to get a good practical estimation of the values of $S_{i}^{tot} $. Several examples are used to illustrate an application of DGSM.

Mots clés: Computer experiments; Sobol' indices; Sensitivity analysis; Uncertainty; Poincare inequality; Morris method; Model derivatives

Domaines: Statistiques [math.ST]

Serge Kucherenko, Bertrand Iooss. Derivative based global sensitivity measures. R. Ghanem, D. Higdon and H. Owhadi (eds). Handbook of uncertainty quantification, Springer, 2017. ⟨hal-01079358v3⟩