|
Article information
2017 , Volume 22, ¹ 1, p.3-16
Voytishek A.V., Prasol D.A.
On the choice of distribution densities in adaptive mesh nodes for stochastic algorithms of numerical integration and function approximation
In this paper we formulate the analog of the equidistributional principle for stochastic adaptive meshes which are used for solving the multi-dimensional problems of numerical integration. It means that the integral for density of continuous 1D random variable 𝜉 over the interval (𝜉 1 , 𝜉2 ) between two sample values of the variable 𝜉 has the “universal” distribution with the density equal to the so called “hat-function”. The importance sampling technique is treated as a method for constructing an adaptive mesh for the problem of numerical integration. The discrete-stochastic approach with the “modelled” basis of the Strang - Fix approximation for numerical simulation of the corresponding adaptive mesh is proposed. We also consider the problem on the choice of the distribution density of nodes in implementation of self-organization in the T. Kohonen’s algorithm. The research is aimed at construction of adaptive meshes for solving the problem of numerical function approximation. For the smooth approximated functions we propose to use the square root of the second derivative module of the approximated function as density of adaptive nodes. The example of function for which the formulated recommendation leads to improvement of approximating properties of mesh approach is given. The most part of results is formulated in the simplest 1D-version. The survey of possible multi-dimensional versions of these results and directions of further investigations is presented in the conclusion.
[full text]
Keywords: adaptive mesh, equidistributional principle, probabilistic distribution density, Monte Carlo method, importance sampling method, Kohonens scheme, choice of nodes density for function approximation
Author(s):
Voytishek Anton Vaclavovich
Dr. , Professor
Position: Leading research officer
Office: Institute of Numerical Mathematics and Mathematical Geophysics of Siberian Division of RAS
Address: 630090, Russia, Novosibirsk, prospect Akademika Lavrentyeva, 6
Phone Office: (383)3307721
E-mail: vav@osmf.sscc.ru
SPIN-code: 7494-4885Prasol Denis Aleksandrovich
Position: The master of mathematics
Office: Institute of Computational Mathematics and Mathematical Geophysics SB RAS
Address: 630090, Russia, Novosibirsk, Pirogova street, 2
E-mail: duohuc.1993@inbox.ru
References:
[1] Khakimzyanov, G. S., Shokin, Yu. I. Raznostnye skhemy na adaptivnykh setkakh. Ch.1, 2 [Difference schemes on adaptive meshes. V. 1, 2]. Novosibirsk: NGU; 2009: (In Russ.)
[2] Borovkov, A. A. Teoriya veroyatnostey [Probability theory]. Moscow: Nauka; 1986: 432. (In Russ.)
[3] Mikhaylov, G. A., Voytishek, A. V. Chislennoe statisticheskoe modelirovanie. Metody Monte-Karlo [Numerical statistical modelling. Monte Carlo
Methods]. Moscow: Izd. Tsentr «Akademiya»; 2006: 368. (In Russ.)
[4] Berezin, I. S., Zhidkov, N. P. Metody vychisleniy. T. 1 [Numerical methods. V. 1]. Moscow: Gosudarstvennoye izdatelstvo fiziko-matematicheskoy literatury; 1959: 464. (In Russ.)
[5] Kohonen, T. Self-Organizing Maps. Springer-Verlag; 2001: 502.
[6] Nechaeva, O. I. Neyrosetevye modeli, algoritmy i kompleks programm dlya po-stroeniya adaptivnykh setok [Neural network models, algorithms and program complex for constructing adaptive meshes] ... Novosibirsk: NGU; 2007: 120. (In Russ.)
[7] Marchuk, G. I., Agoshkov, V. I. Vvedenie v proektsionno-setochnye metody [Introduction in projective-mesh methods]. Moscow: Nauka; 1981: 416. (In Russ.)
[8] Voytishek, A. V., Kablukova, E. G. Using the approximation functional bases in Monte-Carlo methods. Russian Journal of Numerical Analysis and Mathematical Modelling. 2003; 18(6):521–542.
[9] Bessmeltsev, M. V., Voytishek, A. V. Modification of geometrical program modules, tied with construction of modelled probabilistic densities. Computational Technologies. 2011; 16(4):19–36. (In Russ.)
[10] Miloserdov, V. V. Diskretno-stokhasticheskie chislennye algoritmy so splayn-vospolneniyami [Discrete-stochastic numerical algorithms with spline approximations]. Novosibirsk: NGU; 2006: 83. (In Russ.)
[11] Voytishek, A. V., Khmel, D. S. Analytical description for application of the 1D Kohonen scheme for constructing adaptive grids. Numerical Analysis and Applications. 2011; 4(2):105–113.
[12] Voytishek, A. V., Khmel, D. S. Analytical approach to study the boundary effect in the 1D randomized numerical algorithm for constructing adaptive grids. Zhurnal Vychislitel'noi Matematiki i Matematicheskoi Fiziki. 2013; 53(2):195–208. (In Russ.)
Bibliography link:
Voytishek A.V., Prasol D.A. On the choice of distribution densities in adaptive mesh nodes for stochastic algorithms of numerical integration and function approximation // Computational technologies. 2017. V. 22. ¹ 1. P. 3-16
|
