Numerical integration in log-Korobov and log-cosine spaces (Q907580): Difference between revisions

Quasi-Monte Carlo rules are an effective tool for numerical approximation of integrals over the \(s\)-dimensional unite cube \([0,1]^s\) for functions of large classes. The first purpose of the present paper is the error of the integration of functions in the unit ball of the so-called log-Korobov and log-cosine spaces to be studied. The second purpose is to study different kinds of the tractability of the numerical integration for weighted Korobov spaces. In Section 1, the concept of the classical Korobov spaces is presented. The main tool for investigation of the error of the integration is using the so-called rank-1 lattice point rules. Also, the details of the reproducing kernel Hilbert spaces as a reproducing kernel, a form of the inner product and the formula for the worst-case error of the integration in these spaces are given. In Section 2, a log scale in the definition of the Korobov spaces is done. In Subsection 2.1, the details of the log-Korobov spaces are presented. In Subsection 2.2, this is made for the log-cosine spaces. In Subsection 2.3, the so-called tent-transformation is reminded. A formula which relates the worst-case error of the integration in log-cosine and log-Korobov spaces is obtained. In Section 3, a component-by-component construction of lattice rule is presented. In Subsection 3.1, an algorithm for constructing the good generating vector \(\mathbf{g} = (g_1, \dots, g_s)\) which minimizes the worst-case error of the \((d+1)\)- dimensional lattice rules is given. In Subsection 3.2, a formula for the squared worst-case error which depends on the generating vector \(\mathbf{g}\) is obtained. In Section 4, the component-by-component constructions of the lattice rule is used to give bounds of the integration error in log-Korobov spaces. In Theorem 1, an order \({\mathcal O}\left(N^{-{1 \over 2}}(\log N)^{-{\mu(1 - \lambda) \over 2}}\right)\), for any \({1 \over \mu} < \lambda \leq 1\) and \(\mu > 1\), of the worst-case error of the integration in the introduced log-Korobov space by using a lattice point set generated by the good vector, constructed in Algorithm 1, is obtained. In Theorem 2, a relation between the worst-case error of the integration in the Korobov space and the log-Korobov space is obtained. As a consequence of Theorems 1 and 2, an order \({\mathcal O}\left(N^{-{1 \over 2}}(\log N)^{-{\mu(1 - \lambda^\prime) \over 2}}\right)\) of the worst-case error of the integration in the introduced Korobov space by using a lattice point set generated by the good vector, constructed in Algorithm 1, is obtained. In Section 5,the tractability of the integration in log-Korobov spaces is considered. The concepts of the weak tractability, polynomial tractability and strong tractability of the integration in Hilbert spaces are reminded. In Theorem 3, the necessary and sufficient conditions for strong polynomial tractability, polynomial tractability and weak tractability of the integration in Korobov and log-Korobov spaces are presented. In Theorem 4, a more refined result for the strong polynomial tractability is presented. In Section 6, problems for a possible extension of the results of the present paper are discussed.