Back to search

Cornell University

Theory and Applications of Manifold Interpolation for Learning Empirical Green's Functions

Abstract

dc:description.abstract

This dissertation develops two complementary, mesh-independent frameworks for data-driven discovery and interpolation of empirical Green’s functions for one-dimensional parameterized linear operators. It also establishes theoretical groundwork for manifold interpolation in an infinite dimensional setting, and suggests some future research in that direction. Chapter 2 introduces $\bf{nsegf}$, a discrete approach requiring only input–output pairs \lbrace{fi,ui\rbrace}. We show that the empirical Green’s function integral operator matrix G\in\mathbb RN\times N can be recovered by a least-squares formula G = U F+W-1without adjoint solves, making it applicable to non-self-adjoint operators. This also avoids the need to train neural networks, as is done in our approach in Chapter 3. We then interpolate the low-rank SVD factors \((U,\Sigma,V)\) across parameters by lifting orthonormal bases onto the finite dimensional compact Stiefel manifold, SK(\mathbb{R}n), performing polynomial interpolation in the tangent space, and retracting via a \(QR\)-based map. Chapter 3 presents $\bf{chebgreen}$, which learns continuous Green’s functions via Rational Neural Networks and computes high-accuracy Singular Value Expansions using a Python Chebfun implementation. By storing the orthonormal bases from the SVE as “quasimatrices” in the Hilbert space H = (L2(\Omega))K, we generalize the interpolation to the infinite-dimensional Hilbert–Stiefel manifold, establish its injectivity radius, and derive analogous error bounds. Numerical experiments on Poisson, advection–diffusion, Airy, and fractional Laplacian problems achieve sub-percent accuracy with just 100 samples—even under 50 \% output noise. Both frameworks drastically reduce data requirements compared to prior neural-network-only methods. Furthermore, both frameworks allow for learning of Green's functions from a broader class of problems than in previous approaches as the assumption that the associated differential operator is self-adjoint is not required. Chapter 3 also includes a theoretical framework for interpolation using quasimatrices, and establishes rigorous error bounds. Chapter 4 suggests some future directions for Hilbert manifold interpolation theory and applications beyond the Green's function application. It suggests considering subspace interpolation problems in the Hilbert Grassmannian manifolds, and lays out a framework for functional principal component analysis using Hilbert Stiefel manifold interpolation.

Degree

thesis:*
Name thesis:degree_name
Ph. D., Applied Mathematics
Level thesis:degree_level
Doctor of Philosophy
Discipline thesis:degree_discipline
Applied Mathematics
Grantor
Cornell University
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Brown, Jacob
Chair dc:contributor.chair
  • Earls, Christopher
Committee members dc:contributor.committeemember
  • Townsend, Alex
  • Bindel, David

Subjects

dc:subject × 4

Rights

dc:rights
Statement dc:rights
  • Attribution-NoDerivatives 4.0 International
Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
17422649
Dc Identifier Other
ProQuest Submission ID: 15127
ProQuest Publication ID: 32166460

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Brown, Jacob. Theory and Applications of Manifold Interpolation for Learning Empirical Green's Functions. Doctor of Philosophy thesis, Cornell University, 2025. https://hdl.handle.net/1813/120950