University of British Columbia
Differential Equation-Constrained Local Polynomial Regression Models
Abstract
dc:descriptionThis thesis develops a unified framework of differential equation-constrained local polynomial regression (DE-constrained LPR) for nonparametric estimation of a smooth mean function $g(x)$ whose derivative structure is governed by an ordinary differential equation (ODE). The motivation is the prevalence of sparse, irregularly spaced data in the life and environmental sciences, where knowledge encoded in an ODE can compensate for a scarcity of observations that would otherwise render conventional smoothers unreliable. The framework is developed progressively in two directions. The DE1-$k$ estimator is introduced first for the exponential growth model $g'(x) = \lambda g(x)$, and the constraint is then relaxed in stages: to the nonlinear quasi-exponential model g'(x) = \lambda gθ(x), to the general first-order linear ODE $g'(x) = a(x)g(x) + b(x)$ with arbitrary coefficient functions, and finally to the fully general form $g'(x) = F(x, g(x))$, which subsumes the preceding cases. The first-order restriction is then removed altogether, yielding DE2-$k$ and DE3-$k$ estimators for second- and third-order constraints, which require two and three local parameters respectively. For each model class, conditional asymptotic bias and variance expressions are derived, and data-driven bandwidth selection procedures are established. A key finding, consistent across all model classes, is that ODE constraints reduce bias without inflating variance relative to unconstrained local polynomial regression, with the largest gains in sparse-design settings where structure is most needed. Bias theorems for the DE2-$k$ and DE3-$k$ estimators that account for the Jacobian structure of the higher-dimensional nonlinear least squares problem are among the principal theoretical contributions. Three real-data applications demonstrate the breadth of the framework: mouse tumour growth (exponential and quasi-exponential), firebrand burning rates with as few as three observations per group (general linear), and Korean influenza-like illness surveillance analysed via the SIR model. In each case, DE-constrained LPR recovers smooth, interpretable curves that unconstrained methods cannot produce at these sample sizes.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy - PhD
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Mathematics
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ge, Chunlei
- Contributors dc:contributor
-
- Braun, W. John
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/95175