Effective discrepancy and numerical experiments
2012
Abstract Many problems require the computation of a high dimensional integral, typically with a few tens of input factors, with a low number of integrand evaluations. To avoid the curse of dimensionality, we reduce the dimension before applying the Quasi-Monte Carlo method. We will show how to reduce the dimension by computing approximate Sobol indices of the variables with a two-levels fractional factorial design. Then, we will use the Sobol indices to define the effective discrepancy, which turns out to be correlated with the QMC error and thus enables one to choose a good sequence for the integral estimation.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
24
References
4
Citations
NaN
KQI