Scrambled geometric net integration over general product spaces (Q2397747): Difference between revisions

Quasi-Monte Carlo (QMC) sampling is designed for problems of integration over the unit cube \([0,1]^s\) and it has superior accuracy to Monte Carlo for integrands of bounded variation. Scrambled net quadrature allows replication-based error estimation for QMC with at least the same accuracy and for smooth enough integrands even better accuracy than the plain QMC. In the presented paper the quasi-Monte Carlo integration over product spaces of the form \(\chi^s,\) where \(\chi\) is a bounded set of dimension \(d\) is considered. The construction of the van der Corput sequence from the unit triangle is generalized to some other sets. Also the van der Corput sequence is replaced by digital nets in dimension \(s\) to obtain QMC points in \(\chi^s.\) In the introduction, the QMC integration over product spaces is discussed. The concept of the randomization of digital nets and the advantage of reducing the error is presented. In Section 2, the background of QMC and randomized QMC (RQMC) methods is reminded. In Subsection 2.1, the concept of digital nets and sequences are reminded. In Subsection 2.2, the scrambling of digital nets is shown. The concept of the nested uniform scramble is presented. In Theorem 1, the order \({\mathcal O}(n^{-3 + \varepsilon})\) of the variance of the QMC estimator for functions with continuous derivatives by the use of nested uniform scramble of a digital net is obtained. In Theorems 2 and 3 analogous problems are solved. Here the integrands are from the space \(L^2([0,1)^s)\) and the nodes are nested uniform scramble of a \((t,m,s)\)-nets in base \(b.\) In Section 3, the notion of splitting sets is considered. In Subsections 3.1 and 3.2 the geometrical techniques of splits and recursive splitting, and splittings the disk and spherical triangle respectively are presented. In Subsection 3.3, the geometric van der Corput sequence by recursive splitting in base \(b\) is constructed. In Section 4, digital geometric nets via splitting are defined. In Subsection 4.1, the proposed construction is graphically illustrated. In Theorem 4, the error \(\displaystyle {\mathcal O}\left({1 \over n}\right)\) of nested uniform scramble of a \((t,m,s)\)-net in base \(b\) is obtained. In Section 5, the analysis of the variance (ANOVA) and the multiresolution for \(\chi^{1:s}\) is presented. In Subsection 5.1, the ANOVA of \(\chi^{1:s}\) is considered. In Subsection 5.2, the multiresolution of \(\chi^{1:s}\)4 is given. In Section 6, a Sobol' extension is considered. The notion of a Sobol' extensible regions are illustrated. In Theorem 6, the function \(f\) has a continuous mixed partial derivative \(\partial^{1:m} f\) and \(\tilde{f}\) is its Sobol' extension. The decomposition of the function \(\tilde{f}\) is shown. In Section 7 under smoothness and sphericity conditions it is proved that the variance of averages over scrambled geometric nets is \({\mathcal O}(n^{-1 - {2 \over d}}(\log n)^{s-1})\). In Section 8, discussions about the integration of smooth functions are given.