[Paper Review] Dynamical low-rank integrator for the linear Boltzmann equation: error analysis in the diffusion limit

Zhiyan Ding, Lukas Einkemmer+1 more|arXiv (Cornell University)|Jul 9, 2019

Model Reduction and Neural Networks44 references4 citations

TL;DR

This paper presents the first mathematical error analysis of a dynamical low-rank (DLR) integrator for the linear Boltzmann equation in the diffusion limit. It demonstrates that the DLR method, using a projector-splitting time integrator, can dynamically and accurately capture the intrinsic rank-one structure of the solution, with convergence proven under mild time step conditions when using the Crank-Nicolson-like implicit Euler (CNIE) scheme.

ABSTRACT

Dynamical low-rank algorithms are a class of numerical methods that compute low-rank approximations of dynamical systems. This is accomplished by projecting the dynamics onto a low-dimensional manifold and writing the solution directly in terms of the low-rank factors. The approach has been successfully applied to many types of differential equations. Recently, efficient dynamical low-rank algorithms have been applied to treat kinetic equations, including the Vlasov--Poisson and the Boltzmann equation, where it was demonstrated that the methods are able to capture the low-rank structure of the solution and significantly reduce numerical cost, while often maintaining high accuracy. However, no numerical analysis is currently available. In this paper, we investigate the error analysis for a dynamical low-rank algorithm applied to the multi-scale linear Boltzmann equation (a classical model in kinetic theory) to showcase the validity of the application of dynamical low-rank algorithms to kinetic theory. The equation, in its parabolic regime, is known to be rank one theoretically, and we will prove that the scheme can dynamically and automatically capture this low-rank structure. This work thus serves as the first mathematical error analysis for a dynamical low-rank approximation applied to a kinetic problem.

Motivation & Objective

• To establish the first rigorous mathematical error analysis for a dynamical low-rank approximation applied to a kinetic equation.
• To investigate whether the DLR method can dynamically capture the theoretically predicted low-rank structure of the linear Boltzmann equation in the diffusion limit.
• To analyze the role of the time integrator in preserving the low-rank structure, particularly comparing implicit Euler and CNIE schemes.
• To validate numerically that the DLR method maintains high accuracy and reduces computational cost while preserving the rank-one structure in multi-scale regimes.
• To provide theoretical justification for the use of DLR methods in kinetic theory, addressing a gap in analytical support despite strong numerical performance.

Proposed method

• The method employs a projector-splitting integrator to evolve the low-rank factors of the solution matrix directly, avoiding full-dimensional representation.
• The time integration uses a Crank-Nicolson-like implicit Euler (CNIE) scheme, which is shown to preserve the low-rank structure due to its inherent symmetry.
• The analysis is conducted in the diffusive scaling of the linear Boltzmann equation, where the solution is theoretically rank one.
• Error bounds are derived by comparing the DLR solution to the full-rank reference solution, with convergence established under mild time step restrictions.
• The method is applied to the linear Boltzmann equation with both constant and high-contrast cross sections, testing both kinetic and diffusive regimes.
• Numerical validation uses implicit Euler/upwind solvers and diffusion limit solutions as reference, with error measured in the Frobenius norm and rank-approximation error.

Experimental results

Research questions

• RQ1 Can a dynamical low-rank integrator accurately capture the intrinsic rank-one structure of the linear Boltzmann equation in the diffusion limit?
• RQ2 How does the choice of time integrator (implicit Euler vs. CNIE) affect the preservation of the low-rank structure in the DLR scheme?
• RQ3 Under what conditions does the DLR method maintain convergence and stability in the multi-scale regime?
• RQ4 Does the DLR method preserve the low-rank structure without requiring explicit enforcement, especially when the initial data is well-prepared?
• RQ5 What is the quantitative error behavior of the DLR method in comparison to full-rank solvers and the diffusion limit?

Key findings

• The DLR method successfully captures the theoretical rank-one structure of the linear Boltzmann equation in the diffusion limit, both numerically and analytically.
• The CNIE time integrator preserves the low-rank structure automatically for mild time step sizes when the initial data is well-prepared, unlike the implicit Euler method which requires very small steps.
• Numerical experiments show exponential decay of the rank-approximation error in both spatial and velocity directions, confirming the method's robustness.
• For ε = 10⁻³, the DLR solution closely matches the diffusion limit, validating its accuracy in the asymptotic regime.
• The method achieves high accuracy with significantly reduced degrees of freedom, demonstrating computational efficiency without sacrificing solution fidelity.
• The Frobenius norm of the error in the low-rank approximation decays rapidly with increasing rank r, confirming the method's convergence as r increases.