Georgia Institute of Technology
Parameterized Computational Framework for Density Evolution Equations with Applications in Wasserstein Gradient Flows, Games, and Kushner-Stratonovich Equation
Abstract
dc:description.abstractThis thesis presents a novel computational framework for numerical solution of density evolution equations, with application in computing Wasserstein gradient flow (WGF), game dynamics and Kushner-Stratonovich equation. The equations are stemmed from multiple areas including optimal transport (OT) theory, economics and nonlinear filtering. The main part of the thesis contains three sections. 1. In the first section, we first review a framework for computing Wasserstein gradient flows with parameterized equations - Parameterized Wasserstein gradient flow (PWGF). We demonstrate the effectiveness of the framework by solving the porous medium equation. Our method is supported by numerical experiments and error estimates in 2-Wasserstein metric. 2. In the second section, we extend the computational framework to non-gradient system on Wasserstein manifold. Unlike porous medium equation, most evolutional PDEs on Wasserstein manifold don't preserve a gradient flow structure and cannot be naturally fitted into PWGF framework. We derive a novel computational framework which fits for general evolutional PDEs of probability densities. Numerical examples for game dynamics are provided to illustrate the performance of the algorithms. Error estimate in Wasserstein-2 sense is provided. 3. In the third section, we consider numerical computation of Kushner-Stratonovich equation with our proposed framework. Kushner-Stratonovich equation is a stochastic partial differential equations describing the evolution of conditional probability densities, which therefore, provides the solution to nonlinear filtering problem. We derive a new McKean-Vlasov type stochastic differential equation for Kushner-Stratonovich equation in particle level and derive the parameter equation by finding the least square estimation of the state. Numerical examples with scalar and high-dimensional linear models are presented to illustrate the performance of the method and error estimate in Wasserstein-2 distance is discussed to facilitate the framework. \end{itemize}
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Mathematics
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jin, Yijie
- Advisor dc:contributor.advisor
-
- Zhou, Haomin
- Committee members dc:contributor.committeemember
-
- Kang, Sung Ha
- Liao, Wenjing
- Tao, Molei
- Ye, Xiaojing
Subjects
dc:subject × 1Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81998