Solving fully coupled non-linear hygro-thermo-mechanical problems relative to the behaviour of concrete at high temperatures using monolithic models is nowadays a very interesting and challenging computational problem. These models require an extensive use of computational resources, such as main memory and computational time, due to the great number of variables and the numerical characteristics of the coefficients of the linear systems involved. In this paper, a number of different variants of a frontal solver used within HITECOSP, an application developed within the BRITE Euram III ‘HITECO’ EU project, to solve multiphase porous media problems, are presented and evaluated with respect to their numerical accuracy and performance. When developing the variants, several optimization techniques have been adopted, such as data structure, cache and branches optimizations. Specifically, numerical accuracy has been evaluated using a modified componentwise backward error analysis. The main result of this work is a new solver which is both much faster and more accurate than the original one. Specifically, the code runs over five times faster and numerical errors are reduced by up to three orders of magnitude
A frontal solver tuned for fully-coupled non-linear hygro-thermo-mechanical problems
BILARDI, GIANFRANCO;PESAVENTO, FRANCESCO;PUCCI, GEPPINO;
2003
Abstract
Solving fully coupled non-linear hygro-thermo-mechanical problems relative to the behaviour of concrete at high temperatures using monolithic models is nowadays a very interesting and challenging computational problem. These models require an extensive use of computational resources, such as main memory and computational time, due to the great number of variables and the numerical characteristics of the coefficients of the linear systems involved. In this paper, a number of different variants of a frontal solver used within HITECOSP, an application developed within the BRITE Euram III ‘HITECO’ EU project, to solve multiphase porous media problems, are presented and evaluated with respect to their numerical accuracy and performance. When developing the variants, several optimization techniques have been adopted, such as data structure, cache and branches optimizations. Specifically, numerical accuracy has been evaluated using a modified componentwise backward error analysis. The main result of this work is a new solver which is both much faster and more accurate than the original one. Specifically, the code runs over five times faster and numerical errors are reduced by up to three orders of magnitudePubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.