Many engineering applications can be modeled by partial differential equations (PDEs) on complicated domains. However, when classical Finite Element methods (FEMs) are employed, the computational cost may become unaffordable due to the need for generating fine computational meshes composed of standard element shapes such as tetrahedra or hexahedra. To overcome this issue, various solutions have been proposed, including approaches that involve solving PDE problems on grids composed of polytopic elements. In this context, variants of Discontinuous Galerkin (DG) methods, such as discontinuous Galerkin methods on polytopic grids (PolyDG), have been introduced in the literature, as they are naturally suited to support polytopic meshes. In real-life applications, where the management of hundreds of millions of unknowns is common, scalability becomes a necessity. Scalability is essential not only for handling numerous degrees of freedom but also for reducing simulation walltimes. This work focuses on enhancing the LYMPH3D library, implemented in Fortran, to solve a vector-value problem like the linear elasticity in a scalable manner using PolyDG methods. Gauss-Legendre quadrature formulas are employed for the computation of integrals, and their extension to standard element shapes is considered. We revisit theoretical aspects of PolyDG, Gauss-Legendre quadrature formulas over the triangle and tetrahedron, the algebraic formulation, and the structure of the library with additional modifications. Finally, three numerical tests are conducted to validate the proposed enhancements. The first two tests involve a convergence analysis of the method on a unit cube, assessing its optimality. Additionally, the performance of scalability is evaluated to showcase its effectiveness in managing extensive computational loads. The last test demonstrates a practical applicability, wherein a cantilever beam under self-weight is studied.
Molte applicazioni ingegneristiche possono essere modellate mediante equazioni differenziali alle derivate parziali (PDE) su domini complicati. Tuttavia, quando vengono utilizzati i classici metodi degli elementi finiti (FEM), il costo computazionale può diventare proibitivo a causa della necessità di generare mesh computazionali fini composte da elementi in forma standard come tetraedri o esaedri. Per superare questo inconveniente, sono state proposte varie soluzioni, comprese quelle che coinvolgono la risoluzione di problemi di PDE su griglie composte da elementi poligonali. In questo contesto, sono stati introdotti nella letteratura varianti dei metodi di Galerkin discontinuo (DG), come i metodi di Galerkin discontinuo su griglie poligonali (PolyDG), poiché sono naturalmente adatti a supportare mesh poligonali. Nelle applicazioni reali, dove la gestione di centinaia di milioni di incognite è comune, la scalabilità diventa una necessità. La scalabilità è essenziale non solo per gestire numerosi gradi di libertà, ma anche per ridurre i tempi di simulazione. Questo lavoro si concentra sullo sviluppo della libreria LYMPH3D, implementata in Fortran, per risolvere in modo scalabile un problema a valori vettoriali come l’elasticità lineare tramite metodi PolyDG. Per il calcolo degli integrali vengono utilizzate le formule di quadratura di Gauss-Legendre, e vengono considerate la loro estensione agli elementi in forma standard. Rivisitiamo gli aspetti teorici di PolyDG, le formule di quadratura Gauss-Legendre sul triangolo e sul tetraedro, la formulazione algebrica e la struttura della libreria con modifiche aggiuntive. Infine, vengono condotti tre test numerici per convalidare le modifiche proposte. I primi due test coinvolgono un’analisi di convergenza del metodo su un cubo unitario, valutando la sua ottimalità. Inoltre, viene valutata la scalabilità per mostrare la sua efficacia nella gestione di carichi computazionali estesi. L'ultimo test dimostra un’applicazione pratica, studiando una trave in sbalzo soggetta al proprio peso.
Enhancing LYMPH3D library for solving the linear elasticity problem
Gorlezza, Lorenzo
2023/2024
Abstract
Many engineering applications can be modeled by partial differential equations (PDEs) on complicated domains. However, when classical Finite Element methods (FEMs) are employed, the computational cost may become unaffordable due to the need for generating fine computational meshes composed of standard element shapes such as tetrahedra or hexahedra. To overcome this issue, various solutions have been proposed, including approaches that involve solving PDE problems on grids composed of polytopic elements. In this context, variants of Discontinuous Galerkin (DG) methods, such as discontinuous Galerkin methods on polytopic grids (PolyDG), have been introduced in the literature, as they are naturally suited to support polytopic meshes. In real-life applications, where the management of hundreds of millions of unknowns is common, scalability becomes a necessity. Scalability is essential not only for handling numerous degrees of freedom but also for reducing simulation walltimes. This work focuses on enhancing the LYMPH3D library, implemented in Fortran, to solve a vector-value problem like the linear elasticity in a scalable manner using PolyDG methods. Gauss-Legendre quadrature formulas are employed for the computation of integrals, and their extension to standard element shapes is considered. We revisit theoretical aspects of PolyDG, Gauss-Legendre quadrature formulas over the triangle and tetrahedron, the algebraic formulation, and the structure of the library with additional modifications. Finally, three numerical tests are conducted to validate the proposed enhancements. The first two tests involve a convergence analysis of the method on a unit cube, assessing its optimality. Additionally, the performance of scalability is evaluated to showcase its effectiveness in managing extensive computational loads. The last test demonstrates a practical applicability, wherein a cantilever beam under self-weight is studied.File | Dimensione | Formato | |
---|---|---|---|
Executive_Summary_Lorenzo_Gorlezza.pdf
Open Access dal 11/03/2025
Descrizione: Executive Summary
Dimensione
984.48 kB
Formato
Adobe PDF
|
984.48 kB | Adobe PDF | Visualizza/Apri |
Master_Thesis_Lorenzo_Gorlezza.pdf
Open Access dal 11/03/2025
Descrizione: Thesis
Dimensione
1.42 MB
Formato
Adobe PDF
|
1.42 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/217827