Realizability-preserving monolithic convex limiting in continuous Galerkin discretizations of the M 1 M_{1} model of radiative transfer
Paul Moujaes, Dmitri Kuzmin, Christian Bäumer · 2025
Read Realizability-preserving monolithic convex limiting in continuous Galerkin discretizations of the M 1 M_{1} model of radiative transfer by Paul Moujaes, Dmitri Kuzmin, Christian Bäumer (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07689.txt.
Introduction
Paul Moujaes and 2 co-authors design a numerical scheme in 2025 for the M1 model of radiative transfer. The scheme never lets the computed states leave the physically allowed set. Radiative transfer tracks how radiation flows through matter. The M1 model follows its two lowest moments, summaries of the radiation field. An entropy-based rule closes the system. If computed moments step outside the realizable set, the model stops making physical sense.
The authors first build a low-order method that provably preserves the invariant domain. It uses intermediate bar states drawn from averaged exact solutions of Riemann problems. They prove these states are realizable in any number of space dimensions. High resolution comes from nonlinear correction terms. A two-step monolithic convex limiting algorithm constrains those terms. Reactive terms get a diagonally implicit treatment. The result is a scheme that stays physical by construction.
Catalog record
| CREATOR | Paul Moujaes, Dmitri Kuzmin, Christian Bäumer |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-07689
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Realizability-preserving monolithic convex limiting in continuous Galerkin discretizations of the M 1 M_{1} model of radiative transfer Paul Moujaes, Dmitri Kuzmin, Christian Bäumer Abstract We discretize the M 1 M_{1} model of radiative transfer using continuous finite elements and propose a tailor-made monolithic convex limiting (MCL) procedure for enforcing physical realizability. The M 1 M_{1} system of nonlinear balance laws for the zeroth and first moments of a probability distribution function is derived from the linear Boltzmann equation and equipped with an entropy-based closure for the second moment. To ensure hyperbolicity and physical admissibility, evolving moments must stay in an invariant domain representing a convex set of realizable states. We first construct a low-order method that is provably invariant domain preserving (IDP). Introducing intermediate states that represent spatially averaged exact solutions of homogeneous Riemann problems, we prove that these so-called bar states are realizable in any number of space dimensions. This key auxiliary result enables us to show the IDP property of a fully discrete scheme with a diagonally implicit treatment of reactive terms. To achieve high resolution, we add nonlinear correction terms that are constrained using a two-step MCL algorithm. In the first limiting step, local bounds are imposed on each conserved variable to avoid spurious oscillations and maintain positivity of the scalar-valued zeroth moment (particle density). The second limiting step constrains the magnitude of the vector-valued first moment to be realizable. The flux-corrected finite element scheme is provably IDP. Its ability to prevent nonphysical behavior while attaining high-order accuracy in smooth regions is verified in a series of numerical tests. The developed methodology provides a robust simulation tool for dose calculation in radiotherapy. 1 Introduction Radiative transfer models based on the linear Boltzmann equation (LBE) are widely used in computational radiotherapy [ 8 , 27 , 58 ] and other fields of medical physics. The transported variable of the LBE is a fluence that depends on space, time, energy, and direction of travel. Mathematically, LBE has the structure of a Fokker-Planck equation for a nonnegative probability density function. Practical use of deterministic LBE models as a healthcare simulation tool is currently restricted by exorbitant computational cost. An efficient alternative is provided by moment models , in which dependence on the angular variable is eliminated and the dimensionality of the problem is reduced [ 54 , 55 ] . Since such model reduction leads to a system with more unknowns than equations, a closure relation is required to express higher-order moments in terms of the retained ones. In entropy-based closures, the highest moment is modeled by a solution of an entropy optimization problem [ 2 , 13 , 14 , 18 , 25 , 35 , 42 ] or an approximation thereof [ 16 , 41 , 47 , 48 , 51 ] . The reduced models are strongly nonlinear and physically meaningful only if reconstructed moments correspond to a nonnegative particle distribution. The inexpensive M_{1} model [ 17 , 48 ] has already proven its worth in the context of dose calculations for radiotherapy [ 7 , 10 , 23 , 26 , 52 ] . For the underlying closure to be meaningful, the zeroth moment must remain positive, while the magnitude of the first moment must be bounded above by the zeroth moment [ 9 , 17 , 38 , 50 ] . These constraints define the set of admissible states, which forms a convex cone and is referred to as realizable set . To maintain physical consistency, numerical approximations must remain within this set. In the general context of hyperbolic problems, discretizations that guarantee this property are referred to as invariant domain preserving (IDP) [ 30 , 40 ] . While recent years witnessed significant advances in the development of IDP methods for hyperbolic flow models, the application of these techniques to the M_{1} model requires careful extensions and further analysis. Adapting property-preserving methods to the M_{1} system poses additional challenges due to the forcing terms resulting from particle sources as well as scattering and absorption processes. In the context of discontinuous Galerkin methods, flux and/or slope limiters can be applied to the discretized M_{1} system [ 3 , 17 , 50 ] , but yield unsatisfactory results in some cases [ 16 ] . In this work, we extend the monolithic convex limiting (MCL) framework introduced in [ 39 ] to a continuous finite element discretization of the inhomogeneous M_{1} model. The underlying low-order scheme preserves invariant domains by design. Key to its derivation are the so-called bar states , which represent spatial averages of exact solutions of the homogeneous Riemann problem [ 30 ] . Since realizability of the exact Riemann solutions is only proved in one space dimension [ 18 ] , we provide an alternative proof to ensure admissibility of the bar states also in the multidimensional case. We treat the reactive term that results from absorption and scattering implicitly, while employing explicit strong stability preserving Runge–Kutta (SSP-RK) methods [ 29 , 56 ] . By lumping the discrete reaction operator, we avoid solving a linear system in each forward Euler stage. The low-order method serves as the foundation for constructing high-resolution IDP schemes for the M_{1} system. To ensure numerical stability in the vicinity of shocks and steep fronts, we limit the antidiffusive fluxes that recover the high-order target scheme. The proposed MCL strategy constrains each component of a flux-corrected bar state individually before performing a synchronized IDP fix. The involved limiting steps are similar to those of sequential MCL algorithms for the compressible Euler equations [ 39 , 40 ] . We begin in Section 2 by presenting the M_{1} model and reviewing some physical properties. In Section 3 , we design a low-order discretization that is IDP for all physically admissible particle sources. Scattering and absorption terms are taken into account in a manner consistent with the requirement of realizability. In Section 4 , we introduce our customized MCL scheme for the M_{1} model. Finally, we present the results of our numerical experiments in Section 5 and draw conclusions in Section 7 . 2 M_{1} moment model Let \psi=\psi(\mathbf{x},t,\bm{\Omega}) denote a probability density (fluence) that depends on space location \mathbf{x}\in\mathcal{D}\subset\mathbb{R}^{d} , d\in\{1,2,3\} , time instant t\geq 0 , and orientation \bm{\Omega}\in\mathbb{S}_{d-1} , where \mathbb{S}_{d-1}=\{\bm{\Omega}\in\mathbb{R}^{d}:|\bm{\Omega}|=1\} is the unit sphere. In the context of radiation transport modeling, \psi(\mathbf{x},t,\cdot):\mathbb{S}_{d-1}\to\mathbb{R}_{+} represents the angular distribution of particles at a fixed space-time location (\mathbf{x},t) . In what follows, we write “ \gneq ” if we assume that \psi(\mathbf{x},t,\cdot)\in L^{2}(\mathbb{S}_{d-1}) is nonnegative with \|\psi(\mathbf{x},t,\cdot)\|_{L^{1}(\mathbb{S}_{d-1})}>0 . A detailed description of radiative transfer is provided by LBE models of the form [ 2 , 48 ] \frac{\partial\psi}{\partial t}+\bm{\Omega}\cdot\nabla\psi=-(\sigma_{s}+\sigma_{a})\psi+\frac{\sigma_{s}}{4\pi}\int_{\mathbb{S}_{d-1}}\psi(\bm{\Omega}^{\prime})\,\mathrm{d}\bm{\Omega}^{\prime}+Q, (1) where Q=Q(\mathbf{x},t,\bm{\Omega}) is a nonnegative source of particles. The absorption and scattering properties of the background material are characterized by \sigma_{a}\geq 0 and \sigma_{s}\geq 0 , respectively. In principle, approximate solutions to ( 1 ) can be obtained using numerical methods for transport-reaction equations (see, e.g., [ 28 , 34 , 60 ] ). However, the cost of evolving \psi(\mathbf{x},t,\bm{\Omega}) is very high considering that the domain \mathcal{D}\times\mathbb{R}_{+}\times\mathbb{S}_{d-1} is six-dimensional for d=3 . Therefore, it is common practice to approximate ( 1 ) by nonlinear evolution equations for N+1