Application of linear multifrequency-grey acceleration to preconditioned Krylov iterations for thermal radiation transport
Introduction
At high temperatures, thermal radiation (photons) born from blackbody emission dominates the internal energy in a system [1]. Even at more moderate temperatures, this radiation can be an important, nonlocal source of energy deposition. High-fidelity modeling uses a kinetic description of the radiation with a source term from the material, which is coupled to the radiation through photon absorption, emission, and scattering processes. At high material temperatures, absorption-reemission timescales can be orders of magnitude smaller than the desired timestep sizes — which resolve heating timescales — requiring the use of implicit time discretizations. The implicit material emission rate is nonlinear in the end-of-timestep material temperature. Linearization of emission about the material temperature is necessary for a linear method to have stability. It introduces within-timestep absorption-reemission term that depend implicitly on the end-of-timestep radiation distribution. The computation of that often requires an iterative solve of a stiff linear system. In this work, we explore a preconditioning technique, called linear multifrequency-grey (LMFG) acceleration, to improve convergence of the within-timestep iterative solution for the radiation distribution.
Formally, the radiation distribution is modeled with the time-, frequency-, angularly-, and spatially-dependent Boltzmann radiation transport equation. There are two common treatments for the dependence of the radiation distribution on photon frequency. The first treatment, referred to as the grey approximation, is to integrate the radiation equation over all frequencies and use approximate, frequency-averaged material interaction probabilities. The second treatment, referred to as the multigroup (MG) approximation, is to integrate the radiation equation over a collection of frequency groups and use material interaction probabilities averaged over each group, replacing the continuous-frequency radiation equation with a set of coupled, frequency-integrated equations called the MG radiation equations. A common simplifying assumption for the photon angular dependence, accurate for thick systems with large amounts of absorption-reemission, leads to a time-dependent radiation diffusion equation.
The grey or MG radiation diffusion equation can itself be used to calculate the radiation distribution, or it can be used to precondition the radiation transport. The grey radiation diffusion equation can also be used to precondition the MG radiation diffusion equation. LMFG is a linear preconditioner that accelerates the convergence of the MG radiation equations by solving a grey radiation diffusion equation. The LMFG acceleration technique has been proven to be effective for improving the iterative convergence of MG radiation diffusion calculations in high energy density simulations [2]. In [3], a grey radiation transport equation was used to accelerate the MG radiation transport equation. This method, called GTA, shifts the workload from MG radiation transport to grey radiation transport, resulting in improved computational efficiency.
The purpose of this paper is to investigate LMFG for MG thermal radiation transport calculations. Prior work has successfully demonstrated the effectiveness of LMFG to precondition the iterative solution of the MG radiation diffusion equation, especially when solved with Krylov method such as GMRES [4]. The transport setting is novel. In our MG transport setting, two distinct iterative formulations appear, one dealing with scalar fluxes and another dealing with absorption rate densities. We denote these methods as the scalar flux formulation (SFF) and the absorption rate formulation (ARF). In prior radiation diffusion work, only the ARF method was investigated. We investigate the iterative convergence properties of SFF and ARF in the absence and presence of physical scattering. As in [4], we formulate the methods as preconditioned linear systems that can be solved with Krylov iterative methods.
The remainder of this paper is organized as follows. In Section 2, we derive the discretized TRT equations, introduce the source-iteration scheme for iteratively solving these equations, and demonstrate the need for preconditioning. In Section 3, we introduce LMFG acceleration, SFF, and ARF for the transport equation, allowing for monoenergetic scattering. In Section 4, we investigate algorithms for dealing with the complications of this scattering. In Section 5, we present computational results comparing the performance of LMFG with Krylov methods to LMFG with source iteration; the performance of ARF to SFF; and the performance of various techniques for dealing with scattering. In Section 6, we give conclusions and recommendations for future work.
Section snippets
Deriving the discretized TRT equations
The TRT equations are statements of energy conservation for radiation and material energy densities. The Boltzmann radiation transport equation is a conservation statement of photon energy. We make the assumption that the electron collisional timescales are much shorter than the timescales of interest, allowing us to describe the electron distribution as a Maxwellian at the material temperature, which leads to material internal energy being represented as a density in physical space only. The
Derivation of the LMFG acceleration scheme
In this section, we introduce notation related to the discretized radiation equation and then introduce LMFG. We begin with the time-discretized, linearized transport equation for radiation transport with coherent, isotropic scattering. We introduce an iteration scheme to solve this linear problem.
Beginning with Eq. (4a), (4b), (4c), (4d), dropping the timestep index n and the explicit dependence on position — for the remainder of the paper, because the results and presentation are independent
Applying the inner iterations
Schemes one and three require inner iterations when scattering is extant. That is, they require iteratively solving for , given a source . Each of these inner iterations requires the application of , which is a discrete-ordinates sweep for every group.
The inner iterations can be solved in two ways, monolithically or sequentially. If monolithically, is applied to all groups at once, a Jacobi iteration in energy. The inner iteration terminates when the norm across
Summary of algorithms and test problems
We delineate two cases. The first comprises problems without true scattering. In this case, schemes two and three are equivalent and we compare schemes one and two. We refer to scheme two as the Scalar Flux Formulation (SFF) because it deals with scalar fluxes (group-dependent, angle-integrated intensities); we refer to scheme one as the Absorption Rate Formulation (ARF) because it deals with absorption rates. In truth, scheme one deals with reemission rates, although the idea is the same: this
Conclusions
Our results indicate that LMFG is an effective preconditioner, whether applied to SI or GMRES, in the absence or presence of scattering, with ARF and SFF, for a variety of problems and timesteps, for thermal radiation transport. LMFG is seen to reduce the required work per time step by reducing the number of required outer iterations and hence transport sweeps. We found sweeps to be a valid measure of workload, with the majority of the computational time spent therein. This is true at least
Acknowledgements
Los Alamos Report LA-UR-17-28830. The research described here was supported in part by the US Department of Energy Computational Science Graduate Fellowship, grant number DE-FG02-97ER25308, and by the Department of Energy at Los Alamos National Laboratory under contract DE-AC52-06NA25396.
References (18)
- et al.
A synthetic acceleration scheme for radiative diffusion calculations
J. Quant. Spectrosc. Radiat. Transf.
(1985) A grey transport acceleration method far time-dependent radiative transfer problems
J. Comput. Phys.
(1988)- et al.
Linear multifrequency-grey acceleration recast for preconditioned Krylov iterations
J. Comput. Phys.
(2007) - et al.
Nonlinear Krylov acceleration applied to a discrete ordinates formulation of the k-eigenvalue problem
J. Comput. Phys.
(2013) The Equations of Radiation Hydrodynamics
(2005)Radiative Transfer
(1960)- et al.
Krylov iterative methods and the degraded effectiveness of diffusion synthetic acceleration for multidimensional calculations in problems with material discontinuities
Nucl. Sci. Eng.
(2004) Coupled energy group iteration scheme for deterministic transport problems
Synthetic method solution of the transport equation
Nucl. Sci. Eng.
(1963)
Cited by (10)
Improved treatment of multi-material cells in thermal radiation transport codes
2023, Journal of Computational PhysicsDeterministic radiative transfer equation solver on unstructured tetrahedral meshes: Efficient assembly and preconditioning
2021, Journal of Computational PhysicsCitation Excerpt :The main challenge, for methods presented in this article, is the presence of in-scattering from the surface integral in eq. (1), which leads to coupled systems that are difficult to solve due to memory and convergence issues. Associated with the multigroup approximation [11,12] or spectral models [13,14], monochromatic RTE solvers are a stepping stone for designing multi-frequency RTE solvers in different scientific communities, e.g., combustion [15,16]. Now that the RTE has been discretized both in space and angles, it is possible to highlight the major contributions of this paper.
A Monolithic Preconditioned Iterative Solver and Parallel Computing for Three-dimensional Thermal Radiation Transport Equation
2021, Jisuan Wuli/Chinese Journal of Computational PhysicsNonlinear Elimination Applied to Radiation Diffusion
2020, Nuclear Science and Engineering