Solving the Navier–Stokes problem using large-scale optimization techniques

This thesis explores the application of large-scale optimization techniques to solve the Navier-Stokes equations for incompressible fluids in the steady-state regime. Due to the nonlinear and coupled nature of these equations, traditional solvers often face convergence difficulties and suffer from l...

Descripción completa

Detalles Bibliográficos
Autor: Tarrés Jané, Pau
Tipo de recurso: tesis de maestría
Fecha de publicación:2025
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/447014
Acceso en línea:https://hdl.handle.net/2117/447014
Access Level:acceso abierto
Palabra clave:Navier-Stokes equations
Mathematical optimization
Computational fluid dynamics
Navier-Stokes
Stokes
Large-scale
Convex optimization
Proximal gradient
Dual ascent
Augmented Lagrangian method
Equacions de Navier-Stokes
Optimització matemàtica
Dinàmica de fluids computacional
Àrees temàtiques de la UPC::Matemàtiques i estadística::Anàlisi numèrica::Mètodes numèrics
Àrees temàtiques de la UPC::Enginyeria mecànica::Mecànica de fluids
Descripción
Sumario:This thesis explores the application of large-scale optimization techniques to solve the Navier-Stokes equations for incompressible fluids in the steady-state regime. Due to the nonlinear and coupled nature of these equations, traditional solvers often face convergence difficulties and suffer from low computational efficiency. As an alternative, this work investigates convex optimization techniques and algorithms, with a focus on the proximal gradient method, dual ascent, and the augmented Lagrangian method, with the aim of developing more scalable solvers. The finite element method was used to discretize the problem, and the implementations were made in MATLAB within the Swan framework. The methodology was first validated on simpler, structurally similar problems (linear elasticity and Stokes flow) before being applied to the Navier-Stokes equations. The developed methods were compared with monolithic formulations in terms of accuracy, computational cost, memory usage, and convergence behaviour. The results show that, although optimization-based solvers do not always outperform monolithic approaches in computational time, they require fewer memory resources, making them particularly well-suited for large-scale problems and a promising alternative for solving fluid dynamics systems.