Type of Document Dissertation Author Chen, Zheng Author's Email Address email@example.com URN etd-07092007-141850 Title Anova for Parameter Dependent Nonlinear PDES and Numerical Methods for the Stochastic Stokes Equations Degree Doctor of Philosophy Department Mathematics, Department of Advisory Committee
Advisor Name Title Max Gunzburger Committee Chair Fred Huffer Committee Member Janet Peterson Committee Member Xiaoqiang Wang Committee Member Keywords
- Stochastic Stokes Equations
- Finite Element Methods
- White Noise
- Colored Noise
Date of Defense 2007-07-02 Availability unrestricted AbstractThis dissertation includes the application of analysis-of-variance (ANOVA) expansions to analyze solutions of parameter dependent partial differential equations and the analysis and finite element approximations of the Stokes equations with stochastic forcing terms.
In the first part of the dissertation, the impact of parameter dependent boundary conditions on the solutions of a class of nonlinear PDEs is considered. Based on the ANOVA expansions of functionals of the solutions, the effects of different parameter sampling methods on the accuracy of surrogate optimization approaches to PDE constrained optimization is considered. The effects of the smoothness of the functional and the nonlinearity in the PDE on the decay of the higher-order ANOVA terms are studied. The concept of effective dimensions is used to determine the accuracy of the ANOVA expansions. Demonstrations are given to show that whenever truncated ANOVA expansions of functionals provide accurate
approximations, optimizers found through a simple surrogate optimization strategy are also
relatively accurate. The effects of several parameter sampling strategies on the accuracy
of the surrogate optimization method are also considered; it is found that for this sparse
sampling application, the Latin hypercube sampling method has advantages over other well-known sampling methods. Although most of the results are presented and discussed in the context of surrogate optimization problems, they also apply to other settings such as stochastic ensemble methods and reduced-order modeling for nonlinear PDEs.
In the second part of the dissertation, we study the numerical analysis of the Stokes
equations driven by a stochastic process. The random processes we use are white noise, colored noise and the homogeneous Gaussian process. When the process is white noise, we deal with the singularity of matrix Green's functions in the form of mild solutions with the aid of the theory of distributions. We develop finite element methods to solve the stochastic Stokes equations. In the 2D and 3D cases, we derive error estimates for the approximate solutions. The results of numerical experiments are provided in the 2D case that demonstrate the algorithm and convergence rates. On the other hand, the singularity of the matrix Green's functions necessitates the use of the homogeneous Gaussian process. In the framework of theory of abstract Wiener spaces, the stochastic integrals with respect to the homogeneous Gaussian process can be defined on a larger space than L2 . With some conditions on the
density function in the definition of the homogeneous Gaussian process, the matrix Green's
functions have well defined integrals. We have studied the probability properties of this kind
of integral and simulated discretized colored noise.
Filename Size Approximate Download Time (Hours:Minutes:Seconds)
28.8 Modem 56K Modem ISDN (64 Kb) ISDN (128 Kb) Higher-speed Access chen_PhD.pdf 963.65 Kb 00:04:27 00:02:17 00:02:00 00:01:00 00:00:05