A Multigroup M1 Model for Radiation Hydrodynamics and Applications R. Turpault CEA-CESTA, Bp2 33114 Le Barp, France and MAB, UMR CNRS 5466, LRC-CEA M03, Université Bordeaux I, 33400 Talence, France Abstract. Several applications of radiative transfer need to take into account quantities which are frequency dependant. However, when coupling radiation with other physical phenomena, it is crucial to use radiation models which are less expensive than kinetic methods. We introduce a moments model which aims at extending the possibilities of the M1 model proposed by B.Dubroca and J.L.Feugeas [3] in order to solve such applications and keeps the main mathematical and physical properties. INTRODUCTION OF THE PHYSICAL PROBLEM A lot of applications can be found involving several different domains of physics that introduce a form of coupling between hydrodynamics and radiative transfer. These applications include combustion, hypersonic processes -for example probe atmospheric (re)entry-, Inertial Confinement Fusion, hot plasmas and high temperature industrial processings such as material pyrolysis or glass manufacture. In order to get relevant results, it is very important to use convenient radiation models when dealing with this kind of problems. These models have to be both accurate enough to give a good approximation of the physical phenomena and cheap enough to avoid unreasonable calculation times, espacially for multi-dimensionnal coupling. Among the admissible models, the M1-model (for the grey formulation see[3]) is a good compromise. Very cheap, as a moments model, it keeps a lot of good physical properties such as the possibility to take into account strong directionnal non-equilibria (high anisotropy factors). However, as a grey model, it is still approximative when considering problems where the quantities are frequency dependant. Our goal is to extend it into a multigroup model in order to get the rid of this deficiency. This multigroup model is to be implemented inside a fully coupled hydro-rad code. MULTIGROUP M1 MODEL WITH ENTROPIC CLOSURE The radiative transfer equation describes the evolution of the radiative intensity I, which is linked the the photons’ distribution function. Since it does not conserves the energy, we have to couple it with an equation that describes the evolution of the material temperature. For the purpose of this article, we used a very simple one. More general energy equations, such as hydrodynamics equations for example, can be treated the same way. The system is then given by: 1 ∂ I Ω Ω ∇x Iν Ω σ a Bν T Iν Ω c t ν σd σ Iν Ω 4π d ρ Cv ∂t T 4π pν Ω Ω Iν Ω dΩ 0 σ a ν Bν T Iν d ν d µ CP663, Rarefied Gas Dynamics: 23rd International Symposium, edited by A. D. Ketsdever and E. P. Muntz © 2003 American Institute of Physics 0-7354-0124-1/03/$20.00 T is the material temperature, σ x t ν the opacities (σ a is the absorption opacity and σ d the scattering opacity), Ω the direction of propagation of the photons, ν the frequency, p ν Ω Ω the scattering angular redistribution function and Bν T Planck’s function which realizes the minimum of the radiative entropy. This kinetic system can be solved thanks to Monte-Carlo or deterministic (see [9]) methods. However, these methods are very expensive, especially for multidimensionnal problems and since our goal is to couple the radiation with other physics, we will look for cheaper macroscopic models. Hence we have to integrate our equations. The first step is to integrate the radiative transfer equation over all the directions of propagation to get the intermediate system: ∂t E ν ∇x F ν cσ a ν 4π B T ν E ν (1) ∂t F ν c2 ∇x P ν c σ a ν σ d 1 ν F ν (2) This intermediate system has now to be integrated over several groups of frequency, thus we consider ν 1 m 1 M 1 such as ν 1 0 and ν 1 ∞ and we integrate the previous system from ν 1 to ν 1 to m 2 get 1 the system: M 2 2 ∂t m m 2 ∇x m m ρ Cv ∂t T c ∑ m 2 (3) m m aθm4 T Em (4) m cσma F 0 Em cσma aθm4 T , m 2 m , m and m . Fm c Pm 0 c σma 1 m σmd 0 the mth group. Em , Fm and Pm are the radiative energy, the radiative flux vector and the radiative pressure tensor inside where: m θm has the dimension of a temperature and is defined by caθm4 S2 ν m ν m is given by: 1 2 1 2 Bν T d ν dΩ. The radiative entropy density 2kν 2 n lnnI nI 1 ln nI 1 c3 I c2 where nI is the occupation number defined by: nI I 2hν 3 ν The system 3 is not closed yet. We choose to close it by using the minimum entropy principle. Iν (5) 2hν 3 hν 1 exp m α m 1! ; ν 2 1 c k m m 1 m 2 2 realizes the minimum of the radiative entropy under the condition to adequately reconstruct the moments of the considered model. Furthermore, if Em Fm are physically relevant, then α m exist and are unique and independent from one another. Theorem 1. If we note α the lagrange multiplier, the distribution ν Ω ∑ 1 ν With this closure, we are able to know Pm as a function of Em and Fm so that 3-4 is a closed nonlinear system. Moreover, it is possible to write it in an Eddington form Pm ∑m Dm Em where Dm is the Eddington tensor defined by: Dm 1 χm 3 χm 1 Id m# " 2 2 " m (6) Fm $ and χm is the eigenvalue of Dm associated with m . " " Fm Unfortunatly, as opposed to the grey case, it cannot be explicited. We also have: where m $ Theorem 2. The multigroup M1 model has the following properties: 1. The system is hyperbolic. 1 We chose here to consider the opacities σ to be constants σ inside each group. The way to choose a correct mean is most of the times very m difficult and is not developped in this article. 2. The eigenvalues of the system are smaller than c in module. 3. The system has a total entropy which is locally dissipated. 4. The total energy of the system (radiative+matter) is conserved. Fm 5. The model has a natural limitation of the flux ie 1. cEm 6. The closure distribution allows to consider any physically relevant anisotropy. Furthermore, all the anisotropy lies in the direction colinear to the radiative flux. These properties are very useful to implement the model. The first one allows us to work with classical schemes used to solve hyperbolic problems, even at the boundaries. The third and fourth show that the system has a realistic physical behaviour. The fifth point is very important since it tells us that the model is relevant even far away from the equilibrium. As opposed to some other models, the M1-model still has a realistic behaviour even for nearly unidirectionnal flows. In particular, the photons will never travel faster than c, which can be greatly violated by P1 and other models. The last point, linked with (6), shows that even for multidimensionnal cases, only a scalar (χ m ) is needed to determine the radiative pressure. This means that the cost of 2D and 3D computations is nothing more than the ’usual’ cost. Another very important point is that this model is relevant for free-streaming as well as very diffusive regimes (see [1]) and in-between. Since these regimes can be found very close from one another in practice (due to a huge spatial variation of the opacities at the transition from a very opaque to a transparent zone such as smoke/air for example), all the calculation can be made without shifting models. NUMERICAL APPROXIMATION For numerical approximation, we use an explicit finite volumes scheme of the following form: n 1 i m n i m ∆t Ci ∑ n i j m j ∆t n i m n i m n i m (7) 1 n x dx where C are all the cells of the mesh. n i j m are the numerical fluxes normal to the i Ci Ci m interfaces. We used two different numerical fluxes: one associated with the HLLE approximate Riemann solver and a kinetic one based on a kinetic interpretation of 3 . with: n i m The numerical HLLE flux has the following form: n ij b b n i b b n j b b b b U jn Uin (8) The coefficients b and b are written to include the system’s eigenvalues. We can hence choose b b c, however, this choice gives some numerical diffusion. To reduce this numerical diffusion, it is possible to choose values which are closer to the system’s eigenvalues. The numerical scheme we get is positive, decreases the entropy and natu∆t rally limits the flux ( Fm ni c Em ni m i n) under a very restrictive CFL condition: c c∆t σma 1 m σmd 1. ∆x This condition is most of the times too restrictive when coupling the radiative transfer with other phenomena, such as hydrodynamics, which means that in these cases, implicit schemes are to be used. The kinetic numerical flux is computed as the sum of two different half-fluxes: Finj ν Fi j m m 1 2 S2 νm 1 2 ν Ω i d ν dΩ and Fi j m m 1 2 S2 νm 1 2 Ω j d ν dΩ, where S2 Ω S2 Ω n 0 et S2 Fi j n Fi j n with: Ω S2 Ω n 0 , n is the incoming normal vector. This method is slighly more expensive than an HLLE-type flux, but allows to use classic upwind schemes and does not require to evaluate the eigenvalues of the system. The main difficulty introduced by using a multigroup method instead of a grey one is that the value of the radiative pressure Pm is no longer an explicit function of the radiative energy and the radiative flux. In fact, it depends on ν η3 d η , which is only analytical if ν 0 or ν ∞. Since we obviously have to compute the function Ξ ν η 0 e 1 this function a lot of times during a time step, a lot of CPU time is wasted in the process. A way to overcome this drawback consists in tabulating the radiative pressure (and the half-fluxes in the cas of our kinetic half-fluxes scheme). It is nearly the same way that the pressure is tabulated for the resolution of the real-gas Euler equations. Since the variation of the Eddington factor χ (which is the only scalar we need to know in order to compute the pressure in our model) is smooth, the tabulation is very efficient: we do not need a huge amount of points and a linear interpolation is good enough to have a very accurate approximation. A boundary condition frequently used in radiative transfer consists in imposing the incoming radiative flux. For instance, in one dimension, the left boundary condition is done by giving Fi j m . Such a condition is easily implementable with the kinetic scheme since it is the natural condition one may expect for it. However, the HLLE scheme does not take half fluxes into account, hence this condition is not so obvious to implement for it. In fact, the whole flux at the boundary for HLLE is replaced by the kinetic one: the total flux at the boundary is given by the sum of the incoming half flux, which is given, and the outgoing half flux, which is computed. NUMERICAL RESULTS Comparison between the M1 model and flux-limited diffusion Here we are going to compare the results given by the M1 model with results given by flux-limited diffusion models. We consider a simple 1D application: radiation comes from the left boundary inside an initially cold medium. We use the following set: σ a 100, σ d 0, ρ Cv 10 3, Tm0 300K et Tmg 1000K. We look at the evolution of the M1 model, a flux-limited diffusion model using Kershaw limiter [5] and a kinetic model which solves the full radiative transfer equation [9] for 3 different times: t0 1 33ns t1 10t0 and t2 100t0 . 1000 M1 flux−limited diffusion/Kershaw kinetic model 900 800 700 600 500 400 300 0 0.02 0.04 0.06 Figure 2 (a)-Radiative temperatures 0.08 0.1 1000 900 800 700 600 500 400 300 0 0.02 0.04 0.06 0.08 0.1 Figure 2 (b)-Material temperatures 1000 900 P1 M1 kinetec RTE flux limited diffusion/Kershaw flux limited diffusion/Levermore 800 700 600 500 400 300 0 0.01 0.02 0.03 0.04 Figure 2 (c)-Material temperatures at t1 As expected, for t t2 the three models give similar results since we are in a diffusion regime. However for smaller times, the flux-limited diffusion overestimates the temperature and the error can be as great as 100 %. On the other hand, the results of the M1 model are still satisfying. Figure 2 c also gives the results for the P1 model and a flux-limited diffusion using Levermore’s limiter [5]. The P1 model is another moment model which is good near equilibrium. Since there are high anisotropy factors in the precursor, this model yield some errors in this region. The other flux-limited diffusion method is a little bit better than the previous one, but the difference is still important compared to the solution. Multigroup results The case which result is given in Figure 1 shows why it can be crucial to consider some frequency non-equilibrium to solve several applications: a hot wave penetrates inside an initially cold medium. Opacities are quite important inside IR (σ a 4) and UV(σ a 3 2) regions but very small inside visible region(σ a 10 3). In this 1D case, the scattering is neglected. The results showed here are given by 3 models: grey-M1 (with a mean opacity), multigroup-M1 and a kinetic model which solves the full radiative transfer equation. All the results are second-order and given for a time of approximately 1 05ns. Trad/grey−M1 Tmat/grey M1 Trad/kinetic Tmat/kinetic Trad/mg−M1 Tmat/m1−M1 5000 4000 3000 2000 1000 0 0.1 0.2 0.3 0.4 0.5 Figure 1-Material and radiative temperature of the 3 models. The difference between the 2 M1 models is obvious: in the grey model the thermal wave is too fast since it overestimates the effects of the visible regions and underestimates the effects of the IR and UV regions. The M1 model still conserves the energy, the result is that, as the radiation is too fast, the material heating is too slow. Hence it is not very relevant for unstationnary results. On the other hand, the multigroup model gives a good approximation compared to the kinetic model which is more expensive. REFERENCES 1. E. AUDIT, P. C HARRIER , J.-P. C HIÈZE AND B. D UBROCA, A radiation-hydrodynamics scheme valid from the transport to the diffusion limit, http://arxiv.org/list/astro-ph/0206281, submitted to J. Comput. Phys. 2. P. C HARRIER , B. D UBROCA , L. M IEUSSENS AND R. T URPAULT, Discrete-velocity models for numerical simulations in transitional regime for rarefied flows and radiative transfer, Accepted for publication in IMA Volumes in Mathematics and its Applications, 2002. 3. B. D UBROCA AND JL. F EUGEAS, Hiérarchie de Modèles aux Moments pour le Transfert Radiatif, 1998. 4. H. G RAD, On Kinetic Theory of the Rarefied Gases, Comm. Pure and Appl. Math., Vol II, 331-407, 1949. 5. D. L EVERMORE, Relating Eddington Factors to Flux Limiters, J. Quant. Spectrosc. Radiat. Transfer, vol 31, 1984. 6. D. L EVERMORE, Moment Closure Hierarchies for Kinetic Theories, J. Stat. Phys., vol. 83, 1996. 7. D. AND B. W. M IHALAS, Foundations of Radiation Hydrodynamics, Oxford University Press, 1984. 8. G. C. P OMRANING, The Equations of Radiation Hydrodynamics, Science Applications, 1973. 9. R. T URPAULT, Modélisation et approximation numérique des équations du transfert radiatif à l’aide d’une méthode volumes finis implicite discrète en direction et en fréquence, Mémoire de DEA, université de Bordeaux I, 2000. 10. R. T URPAULT, Construction d’un modèle M1-multigroupe pour les équations du transfert radiatif , C. R. Acad. Sci. Paris, Ser. I, 334 (2002) 1-6 ©2002 Académie des sciences. 11. R. T URPAULT, Numerical Solution of Radiative Transfer Problems with Finite Volumes, proceeding of Finite Volumes for Complex Applications III, 695-702.
© Copyright 2025 Paperzz