On the necessity of low-effective dimension (Q2576277)

This paper deals with the problem of necessity condition for the low-effective dimension, which is essentially involved in the process of understanding why quasi-Monte Carlo (QMC), the deterministic version of Monte Carlo (MC) method, is proved to outperform MC by a wide margin for certain very high-dimensional integrals. \textit{I. M. Sobol'} [Mat. Model. 2, No.~1, 112--118 (1990; Zbl 0974.00506)] introduced in 1990 the notion of global sensitivity for the analysis of nonlinear mathematical problems. \textit{R. E. Caflisch, W. Morokoff} and \textit{A. B. Owen} [Valuation of mortgage backed securities using Brownian bridges to reduce effective dimension, J. Comput. Finance 1, 27--46 (1997)] were the first who formally defined the notion of effective dimension using the ANOVA (ANalysis Of VAriance) decomposition method, and attempted to show empirically that low-effective dimension may provide the key to understanding the relationship between QMC and MC methods. \textit{A. B. Owen} [Necessity of low effective dimension, manuscript (2002)] pointed out that low-effective dimension is not a sufficient condition that QMC beats MC, and conjectured that low-effective dimension could be a necessary condition that explains why QMC behaves better than MC. Now, the main result of this article is to prove that Owen's conjecture is not true, i.e. low-effective dimension is also not a necessary condition that QMC is better than MC. This result is obtained as follows: the author introduces a class of functions in high-dimensions, which have the maximum-effective dimension. Then the generalized Sobol' sequences are proved to provide the \(O(N^{-1})\) convergence rate for the integration of the introduced class of functions, with \(N\) being the natural number rank of the integration error.