Source-linked AI summary
A stable rank-adaptive step-and-truncate finite volume method for Vlasov transport on domains with piecewise linear boundaries
André Uschmajew, Andreas Zeiser
TL;DR
The paper addresses low-rank simulation of linear Vlasov transport on bounded domains with inflow boundaries and potentially unstructured meshes. It combines a tensorized finite volume discretization with a rank-adaptive step-and-truncate integrator, proving inherited stability and truncation-error bounds while demonstrating low-rank treatment of high-dimensional problems.
Problem
Linear Vlasov transport on bounded domains has high-dimensional phase space, while unstructured meshes and inflow boundaries complicate separated space–velocity representations.
Method
The method tensorizes a finite volume discretization with separated numerical fluxes and applies a rank-adaptive step-and-truncate integrator to the resulting matrix ODE.
Results
The scheme inherits the L2 stability and CFL restriction of the full forward Euler FV method for homogeneous inflow, with explicit truncation-error bounds and numerical confirmation in 1d1v and 2d2v.
Takeaways & Limitations
The experiments indicate that SAT retains underlying FV stability while controlling low-rank truncation error and reducing effective complexity for high-dimensional transport problems.
Takeaways & Limitations
The analysis does not address conservation or positivity preservation, and extending the approach to higher-order discretizations and nonlinear Vlasov–Poisson remains future work.
Abstract
from arXiv · showhide
We consider the numerical solution of the linear Vlasov transport equation on bounded spatial domains with inflow boundary conditions based on low-rank approximation. We combine a finite volume discretization with a rank-adaptive step-and-truncate scheme for the resulting matrix ODE. The spatial and velocity meshes may be unstructured, while suitable numerical fluxes retain a separated space-velocity representation. For homogeneous inflow, we show that the low-rank scheme inherits the $L_2$ stability and CFL restriction of the underlying full finite volume forward Euler method. In addition, the low-rank approximation error is bounded explicitly in terms of the truncation tolerances, avoiding the modeling error associated with tangent-space projections in dynamical low-rank approximation. Numerical experiments in 1d1v and 2d2v confirm the predicted error behavior. In 2d2v, the method handles an unstructured spatial mesh with nonzero inflow and a full tensor-product discretization of approximately $5.8\cdot10^{10}$ phase-space cells while the numerical rank is at most twelve.
1 Introduction
The paper develops a rank-adaptive finite volume approach for linear Vlasov transport on bounded piecewise linear domains with inflow boundaries and potentially unstructured meshes. It establishes stability and explicit low-rank error control while addressing the phase-space dimensionality challenge.
- Motivation: The linear Vlasov equation evolves particle density in a high-dimensional phase space, motivating low-rank approximations that exploit approximate space–velocity separability.In three spatial dimensions, the density depends on six phase-space variables in addition to time.
- Finite volume formulation: The method starts from a finite volume discretization because it enables transparent forward Euler stability and error analysis while retaining unstructured meshes and inflow boundaries.The FV setting is presented as a basis for future higher-order discontinuous Galerkin extensions.
- Separated flux representation: Piecewise linear boundaries permit boundary flux operators to retain separated space–velocity form through finite sums of tensor-product terms despite varying face normals.The construction uses upwind fluxes at boundary faces and can extend to other fluxes with sufficiently short separated expansions.
- Stability and error: The rank-adaptive step-and-truncate integrator inherits the L2 stability and CFL restriction of the full forward Euler finite volume method under homogeneous inflow.The intermediate projected step and final SVD truncation are both nonexpansive in the Frobenius norm.
- Stability and error: Low-rank errors are explicitly controlled by truncation tolerances relative to the full finite volume forward Euler solution, without assuming a priori low-rank approximability.With suitable initial approximation and tolerance choices, the additional error remains O(τ).
- Scope and contributions: The formulation supports tensorized finite volume discretizations with separated fluxes on bounded domains, including inflow boundaries and unstructured spatial and velocity meshes.The electric field is prescribed, while self-consistent Vlasov–Poisson models remain future work.
2 Finite volume semi-discretization
The paper discretizes the linear Vlasov equation with a tensor-product finite volume space on separately meshed spatial and velocity domains. Carefully selected fluxes and separated inflow data preserve a tensorized matrix representation, including for piecewise linear boundaries.
- Discretization domain: The unbounded velocity domain is truncated to a bounded polytope, and spatial and velocity domains are discretized separately with finite volume meshes.Both meshes may be unstructured, and their product defines the phase-space discretization.
- Finite volume spaces: L2-normalized cell indicators form orthonormal bases, so the phase-space finite volume function is represented by a coefficient matrix.The Frobenius norm of this matrix equals the L2 norm of the finite volume function.
- Cell balance: The conservative Vlasov equation is integrated over cells and normal fluxes are replaced by numerical fluxes on interior and boundary faces.The phase-space velocity is divergence-free, making conservative and nonconservative forms equivalent.
- Numerical fluxes: Global Lax–Friedrichs fluxes retain separated representations, whereas upwind fluxes require one tensor-product term per distinct face normal.Piecewise linear boundaries have finitely many boundary normal directions, enabling tensorized upwind treatment there; unstructured interior meshes may have too many directions.
- Tensorized formulation: The resulting semidiscrete equations are a sum of matrix products and low-rank boundary terms when the inflow data has a separated representation.An approximate separated inflow representation would introduce an additional approximation error not analyzed here.
3 Low-rank integrator
The proposed integrator applies a rank-adaptive step-and-truncate procedure to the finite volume matrix ODE. It uses factorized matrices, augmented bases, and truncated SVDs to approximate forward Euler updates while retaining low-rank computational costs.
- Matrix ODE: The finite volume right-hand side has an affine-linear matrix form with sparse spatial and velocity operators plus low-rank boundary terms.The semidiscrete equations fit this matrix-ODE structure directly.
- Low-rank representation: The method represents the coefficient matrix as U_n = X_nS_nV_n^T, separating spatial and velocity variables.The factors have dimensions determined by the spatial and velocity cell counts and the adaptive rank.
- Truncation: The truncated SVD is computed from a small intermediate matrix and preserves nonexpansiveness in the Frobenius norm.This property can also be supplied by other low-rank approximation methods with suitable error bounds.
- Rank-adaptive update: Each step first approximates the right-hand side by truncated SVD, augments the current bases, performs forward Euler in the enlarged tensor-product space, and truncates again.The first tolerance controls the update-direction approximation, while the second controls the rank of the updated solution.
- Computational cost: If nL, nB, and r are much smaller than both mesh dimensions, the low-rank integrator is significantly cheaper than full forward Euler.Sparse operators require linear work in the spatial or velocity dimension, while the augmented rank is bounded by r* = (nL + 1)r + nB.
4 Stability and conditional convergence
The analysis shows that the rank-adaptive SAT scheme inherits the full forward Euler method’s L2 stability under the same CFL restriction. It also bounds the difference from the full scheme through the prescribed truncation tolerances, including with common inhomogeneous inflow.
- Error control: The low-rank approximation introduces explicitly controlled errors rather than the modeling error associated with tangent-space dynamical low-rank approximation.Step-and-truncate evolves in the ambient discretized space before compression.
- Forward Euler stability: Under homogeneous inflow and τ ≤ τCFL, the full finite volume forward Euler scheme is L2 stable.The CFL threshold follows from the divergence-free phase-space velocity and the finite volume flux assumptions.
- Rank-adaptive stability: The rank-adaptive SAT integrator inherits the same L2 stability condition because its intermediate projection and final SVD truncation are nonexpansive.The proof treats the forward Euler update as projected into enriched tensor-product spaces before truncation.
- Conditional convergence: For nonhomogeneous inflow, the scheme’s difference from full forward Euler is bounded by the initial error and accumulated truncation tolerances.The common affine inflow terms cancel when comparing the two schemes, leaving a homogeneous-inflow stability estimate.
- Conditional convergence: If the initial approximation is exact or δ = O(τ), suitable tolerance scaling keeps the low-rank error relative to full forward Euler at O(τ).This scaling targets the first-order accuracy of the reference forward Euler method.
- Scope of the heuristic: The tolerance heuristic controls accuracy but does not guarantee that the computed solutions have low numerical rank.The experiments indicate that low-rank solutions may nevertheless occur in several cases.
5 Numerical experiments
The experiments test truncation effects, stability, convergence, and scalability of the rank-adaptive SAT integrator in 1d1v and 2d2v settings. Results remain consistent with the theoretical bounds while enabling simulations on unstructured meshes and very large phase-space discretizations.
- 5.1 Free transport on an interval (1d1v): The 1d1v experiment tests consistency without truncation, stability, tolerance dependence, and time-step scaling against full forward Euler.The setup uses free transport with homogeneous inflow, a smooth initial bump, uniform meshes, and τ = 5 · 10^-4.
- 5.1 Free transport on an interval (1d1v): 1.2 · 10^-13 maximum difference confirms agreement between untruncated SAT and full forward Euler up to accumulated floating-point errors.The comparison uses identical initial data, spatial discretization, and time step.
- 5.1 Free transport on an interval (1d1v): The measured SAT–forward Euler difference stays below the Theorem 4.3 bound across tolerance studies; ε2 shows clear first-order dependence, while ε1 eventually levels off near 10^-9.For coupled tolerances, the error is approximately first-order in C and remains below 2TCτ.
- 5.1 Free transport on an interval (1d1v): Both SAT and full forward Euler normalized discrete L2 norms are nonincreasing over the complete interval, confirming the predicted nonexpansive behavior.For T = 0.2 and C = 10, the observed errors decrease with time-step refinement and remain below the upper bound 4τ.
- 5.2 Transport on a triangular-shaped spatial domain (2d2v): Orthogonal projections and SVD truncations can violate entrywise nonnegativity and discrete conservation: the SAT minimum is umin = −6.99 · 10^-7 and its maximum relative mass deviation is 8.02 · 10^-5.The corresponding full forward Euler solution remains nonnegative.
- 5.2 Transport on a triangular-shaped spatial domain (2d2v): In 2d2v, the density enters through the left boundary, is deflected downward by the electric field, and exits through the lower boundary on an unstructured triangular mesh.The refinement study shows decreasing estimated L2 errors and only moderate numerical-rank growth.
6 Conclusion
The paper develops an adaptive low-rank SAT method for Vlasov finite volume discretizations, proving stability and truncation-error bounds while supporting challenging geometries and inflow conditions.
- The method combines finite volume discretization with adaptive step-and-truncate low-rank integration for the Vlasov equation.
- It accommodates bounded domains, inflow boundary conditions, and unstructured meshes while inheriting the underlying finite volume scheme’s L2 stability.
- The analysis derives error bounds in terms of truncation tolerances, and experiments in 1d1v and 2d2v confirm the theoretical predictions.
- The present analysis does not address conservation or positivity preservation, and higher-order discretizations and nonlinear Vlasov extensions remain open.
A Stability of the fully discrete forward Euler scheme
The fully discrete forward Euler finite volume scheme is analyzed for divergence-free transport on bounded polyhedral domains with homogeneous inflow, establishing stability under a CFL restriction.
- The stability result concerns divergence-free linear transport on bounded polyhedral domains with homogeneous inflow boundary conditions.
- A polyhedral mesh and finite volume space define the fully discrete forward Euler sequence using numerical fluxes on cell faces.
- The flux parameters are assumed uniformly bounded, with upwind fluxes used on boundary faces and Lax–Friedrichs choices distinguished by face-dependent parameters.
- For time steps satisfying 0 < τ ≤ τCFL, the forward Euler update is stable in the stated finite volume norm.
- The proof rewrites each update as a convex combination of current neighboring and ghost-cell values, then applies Jensen’s inequality.
- Nonnegative update coefficients follow from the CFL condition and flux properties, completing the stability proof.