Optimized Algorithms for Solving Structural Inverse Gravimetry and Magnetometry Problems on GPUs / Akimova Elena N.,Misilov Vladimir E.,Tretyakov Andrey I. // . - 2017. - V. 753, l. . - P. 144-155.

ISSN/EISSN:

1865-0929 / 1865-0937

Type:

Proceedings Paper

Abstract:

In this article, we construct new variants of iteratively regularized linearized gradient-type methods for solving structural inverse gravimetry and magnetometry problems, namely the regularized conjugate gradient method, the modified regularized conjugate gradient method, and the hybrid regularized conjugate gradient method. The main idea of the modification is to calculate the Jacobian matrix of the integral operator at a fixed point, without updating it during the entire iteration process. We also developed memory-optimized and time-efficient parallel algorithms and programs on the basis of the constructed modified methods. The memory optimization uses the block-Toeplitz structure of the Jacobian matrix. The algorithms were implemented on GPUs using the NVIDIA CUDA technology. We performed an efficiency and speedup analysis, and solved a model problem with synthetic disturbed data.

Author keywords:

Nonlinear gradient-type methods; Parallel algorithms; Gravimetry and magnetometry problems; Toeplitz matrix; GPU GRAVITY PROBLEM

DOI:

10.1007/978-3-319-67035-5\_11

Web of Science ID:

ISI:000450925000011

Соавторы в МНС:

—

Другие поля

Поле	Значение
Editor	Sokolinsky, L and Zymbler, M
Booktitle	PARALLEL COMPUTATIONAL TECHNOLOGIES, PCT 2017
Series	Communications in Computer and Information Science
Note	11th International Scientific Conference on Parallel Computational Technologies (PCT), Kazan, RUSSIA, APR 03-07, 2017
Organization	Russian Univ, Supercomputing Consortium; Fed Agcy Sci Org; Russian Fdn Bas Res; Intel; RSC Grp; NVIDIA; Hewlett Packard Enterprise; AMD
Publisher	SPRINGER-VERLAG BERLIN
Address	HEIDELBERGER PLATZ 3, D-14197 BERLIN, GERMANY
Language	English
EISSN	1865-0937
ISBN	978-3-319-67035-5; 978-3-319-67034-8
Keywords-Plus	GRAVITY PROBLEM
Research-Areas	Computer Science; Telecommunications
Web-of-Science-Categories	Computer Science, Hardware \& Architecture; Computer Science, Theory \& Methods; Telecommunications
Author-Email	aen15@yandex.ru out.mrscreg@gmail.com fr1z2rt@gmail.com
ORCID-Numbers	Akimova, Elena/0000-0002-4462-5817
Funding-Acknowledgement	Russian Foundation for Basic Research {[}15-01-00629a]; Ural Branch of the Russian Academy of Sciences {[}15-7-1-3]
Funding-Text	This work was partly supported by the Russian Foundation for Basic Research (project no. 15-01-00629a), and the Ural Branch of the Russian Academy of Sciences (project no. 15-7-1-3).
Number-of-Cited-References	12
Usage-Count-Last-180-days	1
Usage-Count-Since-2013	1
Doc-Delivery-Number	BL4WR