Análisis empírico de algoritmos de inversión de matrices, aplicados al cálculo de propiedades moleculares.
; Aucar, Gustavo Adolfo 
Otros responsables
Fecha
12/2000
Editorial
Universidad Nacional del Nordeste. Facultad de Ciencias Exactas y Naturales y Agrimensura
Revista
Facena
Título del libro
Nombre del evento
Fecha del evento
Institución organizadora
ISSN
0325-4216
eISSN
1851-507X
ISBN
Idioma
Español
Tipo del evento
Clasificación temática
Palabras clave
Resumen
En este articulo se evahia la performance en cuanto a tiempos de ejecuci6n de dos algoritmos de inversi6n de matrices: uno tradicionalmente utilizado en calculos de propiedades moleculares y uno nuevo implementado en nuestro grupo de investigaci6n y que utiliza un desarrollo en sene de potencias de los elementos matriciales. Se probaron distintas plataformas, sistemas operativos, compiladores y subrutinas de algebra lineal (BLAS). Se realizaron calculos para compuestos modelo que requieren el tratamiento de matrices conteniendo entre 80.000 y 4.000.000 de elementos. Se obtuvieron los tiempos de ejecuci6n de dichos algoritmos a traves de la evaluaci6n con el metodo de propagadores de polarizaci6n de propiedades moleculares tipo triplete y singulete. Se observ6 que el rendimiento del algoritmo de la serie aumenta al crecer la dimensi6n de la matriz a invertir. Para todas las plataformas utilizadas se obtuvo una performance ganancioso de la serie con respecto al algoritmo tradicional.
In this article, the performance of two algorithms for matrix inversion are evaluated considering as a parameter the times of execution: one traditionally used in the calculations of molecular properties and a new one implemented in our research group, that makes use of a series development of matrix elements. Different platforms, operating sistems, compilers and subroutines of lineal algebra (BLAS) were investigated. Calculations for model compounds that require the treatment of matrix containing a number of elements between 80.000 and 4.000.000 was carried out. The calculation of molecular properties of both, singlet and triplet type was used to evaluate the execution time of these algorithms. It was observed that the performance of the series algorithm increases when the dimension of the given matrix grows. For all plataforms, a gainful performance of the series was obtained, compared to the tradicional algorithm.
In this article, the performance of two algorithms for matrix inversion are evaluated considering as a parameter the times of execution: one traditionally used in the calculations of molecular properties and a new one implemented in our research group, that makes use of a series development of matrix elements. Different platforms, operating sistems, compilers and subroutines of lineal algebra (BLAS) were investigated. Calculations for model compounds that require the treatment of matrix containing a number of elements between 80.000 and 4.000.000 was carried out. The calculation of molecular properties of both, singlet and triplet type was used to evaluate the execution time of these algorithms. It was observed that the performance of the series algorithm increases when the dimension of the given matrix grows. For all plataforms, a gainful performance of the series was obtained, compared to the tradicional algorithm.
Colecciones
Aprobación
Revisión
Complementado por
Referenciado por
Licencia Creative Commons
Excepto donde se indique lo contrario, la licencia de este ítem se describe como by-nc-sa
