Title Information
Title
Time-dependent Karhunen-Loeve type decomposition methods for SPDEs
Name: Personal
Name Part
Choi, Minseok
Role
Role Term: Text
creator
Origin Information
Copyright Date
2014
Physical Description
Extent
xiii, 126 p.
digitalOrigin
born digital
Note
Thesis (Ph.D. -- Brown University (2014)
Name: Personal
Name Part
Karniadakis, George
Role
Role Term: Text
Director
Name: Personal
Name Part
Rozovsky, Boris
Role
Role Term: Text
Reader
Name: Personal
Name Part
Themistoklis, Sapsis
Role
Role Term: Text
Reader
Name: Corporate
Name Part
Brown University. Applied Mathematics
Role
Role Term: Text
sponsor
Genre (aat)
theses
Abstract
A new hybrid methodology for the stochastic partial differential equations (SPDEs) is developed based on the dynamically-orthogonal (DO) and bi-orthogonal (BO) expansions; both approaches are an extension of the Karhunen-Loeve (KL) expansion. The original KL expansion provides a low-dimensional representation for square integrable random processes since it is optimal in the mean square sense. The solution to SPDEs is represented in a way that it follows the characteristics of KL expansion on-the-fly at any given time. To this end, both the spatial and stochastic basis in the representation are time-dependent unlike the traditional methods such as polynomial chaos (PC), where only one of them is time-dependent. In order to overcome the redundancy the DO imposes the dynamical constraints on the spatial basis while the BO imposes the static constraints on the spatial and stochastic basis. We examine the relation of the BO and DO and prove theoretically and illustrate numerically their equivalence, in the sense that one method is an exact reformulation of the other by deriving an invertible and linear transformation matrix governed by a matrix differential equation that connects the BO and the DO. We also examine the pathology of the BO that occurs when there is an eigenvalue crossing leading to the numerical instability. On the other hand we observe that the DO suffers numerically when there is a high condition number of the covariance matrix for the stochastic basis. To this end, we propose a unified hybrid framework of the two methods by utilizing an invertible and linear transformation between them. We also present an adaptive algorithm to add or remove modes to better capture the transient behavior. Several numerical examples, linear and nonlinear, are presented to illustrate the DO and BO methods, their equivalence, and adaptive strategies. It is also shown numerically that two methods converge exponentially fast with respect to the number of modes giving the same levels of accuracy, which is comparable with the PC method but with substantially smaller computational cost compared to stochastic collocation, especially when the involved parametric space is high-dimensional.
Subject
Topic
uncertainty quantification
Subject
Topic
reduced-order models,high dimensions
Subject
Topic
stochastic collocation
Subject
Topic
dynamical orthogonality
Subject
Topic
bi-orthogonality
Subject (FAST) (authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/1133516")
Topic
Stochastic partial differential equations
Record Information
Record Content Source (marcorg)
RPB
Record Creation Date (encoding="iso8601")
20141006
Language
Language Term: Code (ISO639-2B)
eng
Language Term: Text
English
Identifier: DOI
10.7301/Z0VT1QF0
Access Condition: rights statement (href="http://rightsstatements.org/vocab/InC/1.0/")
In Copyright
Access Condition: restriction on access
Collection is open for research.
Type of Resource (primo)
dissertations