An algorithm for the computation of the solution of the Fermi-Pasta-Ulam problem (a one-dimensional string of particles coupled with non-linear forces) on vector computers is presented. The algorithm has been tested on a FPS-100 attached Array Processor using the CPFORTRAN language with calls to the vector routines. Detailed analysis of the algorithm shows that when the order of the system of ordinary differential equations increases, the efficiency becomes unitary. It was found that the half performance length is n_1/2 = 120 and that 90% of the maximum performance of the machine (8 Mega FLOPS) is exceeded if the order of the system is greater than 1800. The algorithm can be easily moved to other vector computers and the method can be used to solve similar problems.
E' presentato un algoritmo per it calcolo della soluzione del problema Fermi-Pasta-Ulam (una stringa monodimensionale di particelle accoppiate con forze non lineari) su elaboratori vettoriali. L'algoritmo è stato provato su un Array Processor FPS-100 utilizzando il linguaggio CPFORTRAN con chiamate alle routine vettoriali. Un'analisi dettagliata dell'algoritmo mostra che quando l'ordine del sistema di equazioni differenziali ordinarie aumenta l'efficienza diventa unitaria. Risulta che la dimensione per la quale si ottiene la metà della velocità massima é n_1/2= 120 e che si supera il 90% della velocità massima del processore (8 Mega FLOPS) se l'ordine del sistema è maggiore di 1800. L'algoritmo può essere facilmente portato su altri elaboratori vettoriali ed il metodo può essere usato per risolvere problemi simili.
An algorithm for the solution of the Fermi-Pasta-Ulam problem on a vector processor
A Corana;M Muselli;
1988
Abstract
An algorithm for the computation of the solution of the Fermi-Pasta-Ulam problem (a one-dimensional string of particles coupled with non-linear forces) on vector computers is presented. The algorithm has been tested on a FPS-100 attached Array Processor using the CPFORTRAN language with calls to the vector routines. Detailed analysis of the algorithm shows that when the order of the system of ordinary differential equations increases, the efficiency becomes unitary. It was found that the half performance length is n_1/2 = 120 and that 90% of the maximum performance of the machine (8 Mega FLOPS) is exceeded if the order of the system is greater than 1800. The algorithm can be easily moved to other vector computers and the method can be used to solve similar problems.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.