Carleton University
Scalable Domain Decomposition Methods for Nonlinear and Time-Dependent Stochastic Systems
Abstract
dc:description.abstractComputational modelling is one of the most important tools to understand and predict real-life physical processes. However, the accuracy of their predictions becomes questionable when the uncertainties associated with model parameters, assumptions to the mathematical models and the noise in experimental data are not properly accounted for. Sampling-based approaches to handle these uncertainties become overwhelmingly expensive for large-scale models with high resolution discretizations in space/time. This thesis proposes a sampling-free intrusive stochastic Galerkin-based approach to handle the uncertainties associated with model parameters for time-dependent and nonlinear problems. The increased cost of solving high resolution models using this sampling-free approach is handled using domain decomposition (DD)-based solvers by efficiently distributing the workload to many processes. Developing parallel scalable iterative solvers for uncertainty quantification of these high-resolution models in high performance computing (HPC) environments is the main objective of this thesis. An acoustic wave propagation model with a random field representation of wave speed is handled using a non-overlapping DD method. The symmetric and positive-definite coefficient matrix of the system can be solved using a conjugate-gradient iterative method and associated Neumann-Neumann vertex-based preconditioner in two dimensions. However, the complex spatial coupling and the coupling among the stochastic expansion coefficients can affect the scalabilities of the solver in three dimensions. Hence, a wirebasket-based preconditioner is utilized to enrich the coarse grid allowing better global error propagation and improved scalability for the elastic wave propagation model. For nonlinear stochastic partial differential equations (PDEs), the coefficient matrix of the associated linearized algebraic system is non-symmetric which requires the use of generalized minimum residual (GMRES) method-based iterative solvers. A multilevel Schwarz preconditioner combining DD and algebraic multigrid method is proposed for efficient error reduction for large-scale models. The scalabilities of the proposed solver are demonstrated for the PDE-based compartmental model of the geospatial spread of COVID-19 considering uncertain population movement in a large geographical domain of Southern Ontario.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (Ph.D.)
- Level thesis:degree_level
- Doctoral
- Discipline thesis:degree_discipline
- Engineering, Civil
- Grantor dc:publisher
- Carleton University
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Padillath Vasudevan, Sudhi Sharma
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Copyright © 2023 the author(s). Theses may be used for non-commercial research, educational, or related academic purposes only. Such uses include personal study, distribution to students, research and scholarship. Theses may only be shared by linking to the Carleton University Institutional Repository and no part may be copied without proper attribution to the author; no part may be used for commercial purposes directly or indirectly via a for-profit platform; no adaptation or derivative works are permitted without consent from the copyright owner.
- Language dc:language.iso
- en