This thesis analyzes and compares some of the most widely used numerical methods for solving large sparse linear systems, which frequently arise from the discretization of engineering problems governed by partial differential equations. In particular, LU factorization, the Jacobi and Gauss-Seidel iterative methods, and the Conjugate Gradient method are examined, with the latter representing the main focus of this work. The results show that, although LU factorization provides a direct solution, it is limited by its high computational cost and the *fill-in* phenomenon when applied to large-scale systems. The Jacobi and Gauss-Seidel methods prove to be poorly scalable due to both the large number of iterations required and their overall computational cost. In contrast, the Conjugate Gradient method is shown to be the most efficient approach for sparse, symmetric, positive-definite matrices, combining a low computational cost per iteration with fast convergence. These characteristics make it particularly well suited for large-scale engineering applications.
L'elaborato analizza e confronta alcuni tra i principali metodi numerici per la risoluzione di sistemi lineari sparsi di grandi dimensioni, che derivano frequentemente dalla discretizzazione di problemi ingegneristici descritti da equazioni alle derivate parziali. In particolare, vengono esaminati la fattorizzazione LU, i metodi iterativi di Jacobi e Gauss-Seidel e il metodo del Gradiente Coniugato, approfondito come principale oggetto di studio. I risultati evidenziano come la fattorizzazione LU, pur fornendo una soluzione diretta, sia penalizzata dall'elevato costo computazionale e dal fenomeno del fill-in per sistemi di grandi dimensioni. I metodi di Jacobi e Gauss-Seidel risultano poco scalabili, sia per il numero di iterazioni richieste sia per il costo computazionale. Il metodo del Gradiente Coniugato si dimostra invece la soluzione più efficiente per matrici sparse, simmetriche e definite positive, grazie al ridotto costo per iterazione e alla rapidità di convergenza, confermandosi particolarmente adatto ad applicazioni ingegneristiche di larga scala.
Implementazione, valutazione e approfondimento di metodi numerici per la risoluzione di sistemi lineari
SARTOR, NICOLETTA
2025/2026
Abstract
This thesis analyzes and compares some of the most widely used numerical methods for solving large sparse linear systems, which frequently arise from the discretization of engineering problems governed by partial differential equations. In particular, LU factorization, the Jacobi and Gauss-Seidel iterative methods, and the Conjugate Gradient method are examined, with the latter representing the main focus of this work. The results show that, although LU factorization provides a direct solution, it is limited by its high computational cost and the *fill-in* phenomenon when applied to large-scale systems. The Jacobi and Gauss-Seidel methods prove to be poorly scalable due to both the large number of iterations required and their overall computational cost. In contrast, the Conjugate Gradient method is shown to be the most efficient approach for sparse, symmetric, positive-definite matrices, combining a low computational cost per iteration with fast convergence. These characteristics make it particularly well suited for large-scale engineering applications.| File | Dimensione | Formato | |
|---|---|---|---|
|
Sartor_Nicoletta.pdf
Accesso riservato
Dimensione
1.11 MB
Formato
Adobe PDF
|
1.11 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/111730