Khimich O.M., Popov O.V., Chistyakov O.V., Kokhanovskyi V.O.
Adaptive algorithms for solving eigenvalue problems in a variable computer environment of supercomputers

The authors propose software for the analysis and solution to the algebraic eigenvalue problem using an MIMD computer with GPUs, which includes parallel algorithms and programs with the functions of automatic adaptive configuration of the variable computer environment (multilevel parallelism, variable topology of interprocessor communications, mixed word length, caching, etc.) on the mathematical properties of the problem identified in the computer and the architectural features to ensure the reliability of the solution results and the efficient use of computing resources. © 2023, Springer Science+Business Media, LLC, part of Springer Nature.

Keywords: adaptive algorithms, algebraic eigenvalue problems (AEVPs), changeable computer environment, gradient conjugation method, multi-core computers of MIMD-architecture with graphic processors, subspace iteration method, Eigenvalues and eigenfunctions, Iterative methods, Program processors, Reliability analysis, Supercomputers, Algebraic eigenvalue problem, Algebraic eigenvalue problems, Changeable computer environment, Computer environments, Gradient conjugation method, Graphic processors, MIMD architecture, Multi-core computer of MIMD-architecture with graphic processor, Multi-cores, Subspace iteration method, Adaptive algorithms

Cite:
Khimich O.M., Popov O.V., Chistyakov O.V., Kokhanovskyi V.O. (2023). Adaptive algorithms for solving eigenvalue problems in a variable computer environment of supercomputers. Cybernetics and Systems Analysis, 59 (3), 141–156. doi: https://doi.org/10.1007/s10559-023-00583-1 http://jnas.nbuv.gov.ua/article/UJRN-0001402517 [In Ukrainian].