Optimal error bound of restricted Monte Carlo integration on anisotropic Sobolev classes (Q2455840)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Optimal error bound of restricted Monte Carlo integration on anisotropic Sobolev classes |
scientific article |
Statements
Optimal error bound of restricted Monte Carlo integration on anisotropic Sobolev classes (English)
0 references
26 October 2007
0 references
The calculation of high-dimensional integrals is applied in different fields of science. Recently, \textit{N. S. Bakhvalov} [On the rate of convergence of indeterministic integration processes within the functional classes \(W^\ell_p\). Theory Probab. Appl. 7, 227 (1962)], \textit{S. Heinrich, E. Novak}, and \textit{H. Pfeiffer} [Monte Carlo and quasi-Monte Carlo methods 2002. 27--49 (2004; Zbl 1041.65005)] have proved that using \({\mathcal O}(\log_2n)\) random bits one can achieve the same error bound of the integration problem for classical isotropic Sobolev class \(W_pr([0,1]^d)\) \((1\leq p\leq\infty)\). In the paper some notions as a solution operator mapping \(S\) on a functional space \(F\), generalized Monte Carlo method, a randomized \(n\)-th minimal error are considered. The class of restricted randomized algorithms with \(n\) function evaluations and \(m\) random bits is considered and the randomized \((n,m)\)-th minimal error on \(F\) \(e_{n,m}^{\text{coin}}(S,F)\) is reminded. The definition of the so-called anisotropic Sobolev class \(W_p^r(D)\) is given. Theorem 2 gives the exact order \(n^{-g(r)- {1 \over 2} + \max \{ {1 \over p}- {1 \over 2}, 0 \}}\) of the randomized \((n, \lfloor(2 + {1 \over g(r)})\log_{2}n\rfloor)\)-th minimal error of the integration on the anisotropic class \(W_p^r(D)\). The study is essentially based on the discretization technique developed for the study of classical Sobolev classes.
0 references
anisotropic Sobolev classes
0 references
restricted Monte Carlo integration
0 references
random bits
0 references
randomized \((n,m)\)-th minimal error
0 references
discretization technique
0 references
high-dimensional integrals
0 references
restricted randomized algorithms
0 references