Projection schemes for stochastic partial differential equations
The focus of the present work is to develop stochastic reduced basis methods (SRBMs) for solving partial differential equations (PDEs) defined on random domains and nonlinear stochastic PDEs (SPDEs). SRBMs have been extended in the following directions: Firstly, an h-refinement strategy referred to...
Main Author: | |
---|---|
Other Authors: | |
Published: |
University of Southampton
2009
|
Subjects: | |
Online Access: | https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.500829 |
id |
ndltd-bl.uk-oai-ethos.bl.uk-500829 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-bl.uk-oai-ethos.bl.uk-5008292018-09-05T03:28:02ZProjection schemes for stochastic partial differential equationsPrerapa, Surya MohanNair, Prasanth2009The focus of the present work is to develop stochastic reduced basis methods (SRBMs) for solving partial differential equations (PDEs) defined on random domains and nonlinear stochastic PDEs (SPDEs). SRBMs have been extended in the following directions: Firstly, an h-refinement strategy referred to as Multi-Element-SRBMs (ME-SRBMs) is developed for local refinement of the solution process. The random space is decomposed into subdomains where SRBMs are employed in each subdomain resulting in local response statistics. These local statistics are subsequently assimilated to compute the global statistics. Two types of preconditioning strategies namely global and local preconditioning strategies are discussed due to their merits such as degree of parallelizability and better convergence trends. The improved accuracy and convergence trends of ME-SRBMs are demonstrated by numerical investigation of stochastic steady state elasticity and stochastic heat transfer applications. The second extension involves the development of a computational approach employing SRBMs for solving linear elliptic PDEs defined on random domains. The key idea is to carry out spatial discretization of the governing equations using finite element (FE) methods and mesh deformation strategies. This results in a linear random algebraic system of equations whose coefficients of expansion can be computed nonintrusively either at the element or the global level. SRBMs are subsequently applied to the linear random algebraic system of equations to obtain the response statistics. We establish conditions that the input uncertainty model must satisfy to ensure the well-posedness of the problem. The proposed formulation is demonstrated on two and three dimensional model problems with uncertain boundaries undergoing steady state heat transfer. A large scale study involving a three-dimensional gas turbine model with uncertain boundary, has been presented in this context. Finally, a numerical scheme that combines SRBMs with the Picard iteration scheme is proposed for solving nonlinear SPDEs. The governing equations are linearized using the response process from the previous iteration and spatially discretized. The resulting linear random algebraic system of equations are solved to obtain the new response process which acts as a guess for the next iteration. These steps of linearization, spatial discretization, solving the system of equations and updating the current guess are repeated until the desired accuracy is achieved. The effectiveness and the limitations of the formulation are demonstrated employing numerical studies in nonlinear heat transfer and the one-dimensional Burger’s equation.510TJ Mechanical engineering and machineryUniversity of Southamptonhttps://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.500829https://eprints.soton.ac.uk/342800/Electronic Thesis or Dissertation |
collection |
NDLTD |
sources |
NDLTD |
topic |
510 TJ Mechanical engineering and machinery |
spellingShingle |
510 TJ Mechanical engineering and machinery Prerapa, Surya Mohan Projection schemes for stochastic partial differential equations |
description |
The focus of the present work is to develop stochastic reduced basis methods (SRBMs) for solving partial differential equations (PDEs) defined on random domains and nonlinear stochastic PDEs (SPDEs). SRBMs have been extended in the following directions: Firstly, an h-refinement strategy referred to as Multi-Element-SRBMs (ME-SRBMs) is developed for local refinement of the solution process. The random space is decomposed into subdomains where SRBMs are employed in each subdomain resulting in local response statistics. These local statistics are subsequently assimilated to compute the global statistics. Two types of preconditioning strategies namely global and local preconditioning strategies are discussed due to their merits such as degree of parallelizability and better convergence trends. The improved accuracy and convergence trends of ME-SRBMs are demonstrated by numerical investigation of stochastic steady state elasticity and stochastic heat transfer applications. The second extension involves the development of a computational approach employing SRBMs for solving linear elliptic PDEs defined on random domains. The key idea is to carry out spatial discretization of the governing equations using finite element (FE) methods and mesh deformation strategies. This results in a linear random algebraic system of equations whose coefficients of expansion can be computed nonintrusively either at the element or the global level. SRBMs are subsequently applied to the linear random algebraic system of equations to obtain the response statistics. We establish conditions that the input uncertainty model must satisfy to ensure the well-posedness of the problem. The proposed formulation is demonstrated on two and three dimensional model problems with uncertain boundaries undergoing steady state heat transfer. A large scale study involving a three-dimensional gas turbine model with uncertain boundary, has been presented in this context. Finally, a numerical scheme that combines SRBMs with the Picard iteration scheme is proposed for solving nonlinear SPDEs. The governing equations are linearized using the response process from the previous iteration and spatially discretized. The resulting linear random algebraic system of equations are solved to obtain the new response process which acts as a guess for the next iteration. These steps of linearization, spatial discretization, solving the system of equations and updating the current guess are repeated until the desired accuracy is achieved. The effectiveness and the limitations of the formulation are demonstrated employing numerical studies in nonlinear heat transfer and the one-dimensional Burger’s equation. |
author2 |
Nair, Prasanth |
author_facet |
Nair, Prasanth Prerapa, Surya Mohan |
author |
Prerapa, Surya Mohan |
author_sort |
Prerapa, Surya Mohan |
title |
Projection schemes for stochastic partial differential equations |
title_short |
Projection schemes for stochastic partial differential equations |
title_full |
Projection schemes for stochastic partial differential equations |
title_fullStr |
Projection schemes for stochastic partial differential equations |
title_full_unstemmed |
Projection schemes for stochastic partial differential equations |
title_sort |
projection schemes for stochastic partial differential equations |
publisher |
University of Southampton |
publishDate |
2009 |
url |
https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.500829 |
work_keys_str_mv |
AT prerapasuryamohan projectionschemesforstochasticpartialdifferentialequations |
_version_ |
1718729976174346240 |