# Conference Agenda

Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).

Please note that all times are shown in the time zone of the conference. The current conference time is: 18th Aug 2022, 09:44:58pm CEST

 Session Overview
 Date: Monday, 27/June/2022 9:30am - 10:45am SC1: Short Courses 1Location: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle University 9:30am - 10:30amVarious Collision Models for LBE Luo, Li-Shi CSRC, China, People's Republic of Various Collision Models for LBE 10:45am - 11:00am Coffee Break 11:00am - 12:15pm SC2: Short Courses 2Location: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle University 11:00am - 12:00pmTwo approximations of the Euler equations using a non conservative formulation Abgrall, Rémi University of Zurich, Switzerland Two approximations of the Euler equations using a non conservative formulation 12:15pm - 1:45pm LunchLocation: Room MSI 218, Maison des Sciences de l'Ingénieur (MSI), La Rochelle University 1:45pm - 3:00pm SC3: Short Courses 3Location: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle University 1:45pm - 2:45pmRecursive asymptotic analysis for the lattice Boltzmann method Geier, Martin TU Braunschweig, Germany The lattice Boltzmann equation was first proposed as a rule based physical model mimicking the behavior of undistinguishable particles in a gas, rather than a discretization of a partial differential equation. Different asymptotic techniques have since been proposed with the aim of linking the lattice Boltzmann equation to a particular differential equation, to analyze boundary conditions and to improve the accuracy of the method. These techniques include the Chapman Enskog method, the plain Taylor expansion method and the asymptotic expansion in a numerical smallness parameter. Expansions are often applied directly to the distribution functions and moments are obtained through a transformation matrix as known form Multiple Relaxation Time methods. In this lecture we will investigate a matrix free expansion in countable raw moments. Through recursion of the collision operators of different moments this expansion can be reduced to an expansion in the primitive variables and hence leads very naturally to the partial differential equation being solved. The purpose of this method is to make analyzing lattice Boltzmann methods as simple as possible, both in the manual and the computer algebra aided context. 3:00pm - 3:15pm Coffee Break 3:15pm - 4:30pm SC4: Short Courses 4Location: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle University 3:15pm - 4:15pmLBM as a general numerical method: fluid dynamics and beyond Sagaut, Pierre Aix-Marseille University, France LBM as a general numerical method: fluid dynamics and beyond 4:30pm - 5:45pm SC5: Short course 5Location: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle University Equivalent finite difference schemes for the magic TRT lattice Boltzmann method Dellar, Paul University of Oxford, United Kingdom Equivalent finite difference schemes for the magic TRT lattice Boltzmann method 5:45pm - 7:00pm Q&A for short coursesLocation: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle University
 Date: Thursday, 30/June/2022 9:30am - 10:30am LBM and Complex FlowsLocation: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle UniversitySession Chair: Michel Berthier, La Rochelle University 9:30am - 9:50amFlow of plant protein doughs in extruder dies van der Sman, Ruud Wageningen University & Research, Netherlands, The In this contribution we present a numerical model of the flow of plant protein dough through an extruder die, as occurs in the production of plant-based meat replacers. This model is to be coupled to another model describing the flow of the material in the extruder screw section. The combined model will be used for process design. The protein dough is assumed to behave as a Herschel-Bulkley fluid, having a yield stress, and exhibiting shear thinning. These parameters have been measured as function of moisture content and temperature. It follows that the yield stress depends on Tg/T, the ratio of the (moisture dependent) glass transition temperature and the actual temperature. The shear thinning exponent is independent of temperature or moisture content. The cooling die receives the protein dough at high temperatures in the range of 120-160$^oC$, and it will be cooled down to temperatures of 60-100$^oC$. The extruder will be operated at a constant throughput. The cooling of the dough in the die will involve a strong increase of the yield stress. Hence, in the middle of the flow channel an unyielded plug flow will develop. Shear rate gradients develop in a small boundary layer along the wall. Above all, we expect a lubrication layer will develop along the wall due to expulsion of moisture, leading to wall slip. Towards the end the plug flow will cover the entire cross section of the die. Hence, for proper process design the flow model needs to be coupled to an energy balance solving the temperature profile within the extruder die. There will be a two-way coupling between the flow model and the energy balance: the dough rheology is strongly temperature dependent, and the viscous dissipation can increase temperature. The flow model is implemented using Lattice Boltzmann model, which has the advantage that shear stresses are directly locally available, and it can later be extended to multiphase flow, as plant protein doughs consists often of two different biopolymers. The issue of the singularity in the effective viscosity in the Herschel-Bulkley model is resolved via a biviscosity regularization. Wall slip will be modelled using the general formalism by Kalyon. A scaling analysis of the governing equations show that a 2D model will suffice for process design purposes. The overall boundary condition is complex, which is governed by a constant throughput. The related flow profile is an unknown, and will develop over the length of the extruder die, due to the non-isothermal conditions. The required pressure gradient is the desired outcome of the simulation, and thus it can be prescribed directly. Hence, as a boundary condition we impose a pressure gradient, which will be varied such that the flow profiles in all cross sections matches that of the imposed throughput. 9:50am - 10:10amInvestigation of the Capture of Coarse Air-borne Particles at different Fibre Aspect Ratios using Lattice Boltzmann Method Bhardwaj, Utsav1,2; Abraham, Jijo Derick1,2; Ray, Bahni1; Das, Dipayan1; Das, Apurba1; Mitchell, Travis2; Leonardi, Christopher2 1Indian Institute of Technology Delhi; 2The University of Queensland Air pollution is one of the most burning and life-threatening global issues, for which the fibrous air filters have emerged extensively as a remedial solution. In the present study, the capture dynamics of coarse airborne particles and associated variation of filtration performance parameters (efficiency and pressure drop) with respect to the variations in aspect ratio of a single rectangular fibre have been investigated numerically in two dimensions. With the length scales corresponding to the fibre and particles being a few microns each, the mesoscopic method, viz. lattice Boltzmann method (LBM) has been used for simulations of airflow across the fibre. For simulations of the motion of particles and their capture by fibre under the influence of aerodynamic drag and gravity, the Lagrangian approach has been used. One-way coupling has been employed between the airflow and particles (where only airflow field determines the motion of particles), and the inter-particle interactions have not been considered as the concentration of particles is pretty small. In case of LBM, the D2Q9 lattice configuration has been used. Also, the single relaxation time model has been used to define the collision operator via BGK approximation. The fibre-air interface has been provided with no-slip boundary condition via bounce-back scheme whereas the upper and lower boundaries of the computational domain have been provided with either symmetry or free-slip boundary condition using specular reflection. The study presents the characteristic variations observed in the filtration performance parameters as the fibre aspect ratio is varied. 10:10am - 10:30amCompressible lattice Boltzmann methods based on numerical collision Thyagarajan, Karthik; Coreixas, Christophe; Latt, Jonas University of Geneva/ Battelle, Switzerland Over the past three decades, lattice Boltzmann methods (LBMs) have grown as serious alternatives to Navier-Stokes-Fourier (NSF) solvers for the simulation of isothermal and weakly compressible flows past realistic geometries. Nevertheless, most compressible LBMs available in the literature have difficulties to compete with NSF solvers due to an increased (1) size of the lattice required to get the correct macroscopic behavior, and/or (2) complexity of the numerical scheme to ensure stable simulations . Recently, a GPU-accelerated and purely LB approach based on numerical equilibria was proposed as an efficient and accurate alternative. In this approach, equilibria are computed through a root-finding algorithm that imposes an arbitrary number of equilibrium moments, which alleviates common accuracy and stability limits encountered with relatively small lattices, polynomial equilibria and BGK collision model. Yet, the BGK approximation also leads to a number of physical and numerical limitations, e.g., fixed Prandtl number and stability issues. In this talk, we propose to extend the numerical computation of equilibrium populations to their post-collision counterpart, hence, allowing for the direct control of diffusive fluxes through constraints on non-equilibrium moments. This idea was originally proposed in the context rarefied gas dynamics to impose the correct heat flux, and is further extended here to any non-equilibrium moment. As a proof of concept, the D2Q37 lattice is adopted to study the effects of the numerical collision on both stability and accuracy. Particular attention is paid to the computation of gradients, which are used as constraints for non-equilibrium moments, and that can either be done locally (LB scheme) or thanks to a finite-difference scheme. The proposed approach can seamlessly run on CPUs or GPUs thanks to the recent upgrade of the C++ STL library and NVIDIA compiler. The numerical collision is tested for several benchmark problems of increasing complexity, and corresponding results provide a positive feedback to proceed further. 10:30am - 10:50am Coffee Break 10:50am - 11:50am Method and Analysis IILocation: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle UniversitySession Chair: Li-Shi Luo, CSRC 10:50am - 11:10amA consistent discrete divergence-free condition in lattice Boltzmann magnetohydrodynamics: a data-driven approach Dellar, Paul University of Oxford, United Kingdom Magnetohydrodynamics combines the Navier-Stokes and Maxwell equations to describe the flow of electrically conducting fluids in magnetic fields. Maxwell’s equations require the magnetic field to evolve in a way that keeps the divergence of the magnetic field zero. This is analogous to the incompressibility condition for a fluid, except there is no analogue of a pressure in Maxwell’s equations. Lattice Boltzmann magnetohydrodynamics relies upon kinetic representations of both the fluid and the electromagnetic field. There is an extra kinetic degree of freedom that represents the divergence of the magnetic field for slowly varying solutions, but its relation to any discrete approximation to the divergence of the magnetic field on the lattice has previously been mysterious. We show empirically, using data from simulations, that there is an optimal finite difference stencil for approximating the discrete divergence of the magnetic field, with parameters that depend on the relaxation times for the different components of the kinetic representation of the electromagnetic field. We then show that the parameters for the optimal stencil can be derived analytically using operator algebra techniques, and that they resemble formulas for the optimal placement of no-slip boundaries between lattice points. Finally, we show that adjusting the relaxation time for the kinetic degree of freedom mentioned above implements an extended magnetohydrodynamics with an extra scalar field to maintain the divergence-free condition. The joint evolution of this kinetic degree of freedom and the optimal finite difference approximation for the discrete divergence of the magnetic field matches analytical solutions of the extended magnetohydrodynamics equations. 11:10am - 11:30amConservative models for the compressible hybrid lattice Boltzmann method Wissocq, Gauthier; Coratger, Thomas; Farag, Gabriel; Zhao, Song; Boivin, Pierre; Sagaut, Pierre Aix-Marseille Université, France A new methodology is introduced to build conservative numerical models for fluid simulations based on segregated schemes, where mass, momentum and energy equations are solved by different methods. It is here designed for developing new numerical discretizations of the total energy equation, adapted to a thermal coupling with the lattice Boltzmann method (LBM). The proposed methodology is based on a linear equivalence with standard discretizations of the entropy equation, which, as a characteristic variable of the Euler system, allows efficiently decoupling the energy equation with the LBM. To this extent, any LBM scheme is written under a finite-volume formulation involving fluxes, which are included in the total energy equation as numerical corrections. Three models are subsequently derived: a first-order upwind, a Lax-Wendroff and a MUSCL-Hancock schemes. They are assessed on standard academic test cases for compressible flows, with and without discontinuitities. Three key features are exhibited: 1) the models are conservative by construction, recovering correct jump relations across shock waves, 2) the stability and accuracy of entropy modes can be explicitly controlled, 3) the low dissipation of the LBM for isentropic phenomena is preserved. 11:30am - 11:50amLimit consistency of lattice Boltzmann equations Simonis, Stephan; Krause, Mathias J. Karlsruhe Institute of Technology (KIT), Germany We establish the notion of limit consistency as a novel technique to formally prove the consistency of lattice Boltzmann equations (LBE) to a given partial differential equation (PDE). For the purpose of illustration, the incompressible Navier–Stokes equations (NSE) are used as a paragon. Based upon the proven diffusion limit [L. Saint-Raymond (2003), doi: 10.1016/S0012-9593(03)00010-7] of the BGK Boltzmann equation (BGKBE) towards the NSE, we provide a successive discretization by nesting conventional Taylor expansions and finite differences. Tracking the discretization state of the domain for the particle distribution functions, we measure truncation errors at all levels within the derivation procedure. Via parametrizing equations and proving limit consistency of the resulting sequences, we retain the path towards the targeted PDE at each step of discretization, i.e. for the discrete velocity BGKBE (DVBGKBE) and the space-time discretized lattice BGKBE (LBGKBE). As a direct result, we unfold the discretization technique of lattice Boltzmann methods as chaining finite differences and provide a generic top-down derivation of the numerical scheme. 11:50am - 1:20pm LunchLocation: Room MSI 218, Maison des Sciences de l'Ingénieur (MSI), La Rochelle University 2:00pm - 10:00pm Social Event & Gala DinnerLocation: Boat trip + Museum (corderie royale) + Gala Dinner
 Date: Friday, 01/July/2022 8:30am - 9:30am Invited speakerLocation: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle UniversitySession Chair: Michel Berthier, La Rochelle University From Langevin dynamics to kinetic Monte Carlo: the quasi-stationary distribution approach Lelièvre, Tony Ecole des Ponts ParisTech, France We will present a mathematical framework to draw a rigorous connection between microscopic and mesoscopic models to describe the evolutions of materials at the atomistic scale: molecular dynamics, namely Langevin or overdamped Langevin dynamics, and kinetic Monte Carlo (a.k.a. Markov State Models), namely jump Markov processes with values in a discrete state space. This analysis is useful to analyze and justify numerical methods which use the jump Markov model underlying the molecular dynamics as a support to efficiently sample the state-to-state dynamics (accelerated dynamics techniques à la D. Perez and A.F. Voter). It also provides a mathematical framework to justify the use of transition state theory and the Eyring-Kramers formula to build kinetic Monte Carlo or Markov state models. References: - G. Di Gesù, T. Lelièvre, D. Le Peutrec and B. Nectoux, Sharp asymptotics of the first exit point density, Annals of PDE, 5(1), 2019. - T. Lelièvre, Mathematical foundations of Accelerated Molecular Dynamics methods, In: W. Andreoni and S. Yip (Eds), Handbook of Materials Modeling, Springer, 2018. - T. Lelièvre, M. Ramil and J. Reygner, Quasi-stationary distribution for the Langevin process in cylindrical domains, part I: existence, uniqueness and long-time convergence, Stochastic Processes and their Applications, 144, 176-201, (2022). 9:30am - 11:10am Phase-Field, Multiphase and Other Methods for Complex FluidsLocation: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle UniversitySession Chair: François Dubois, CNAM & Univ. Paris-Saclay 9:30am - 10:30amThree-phase contact implementations to progress towards rough-walled fractures Mitchell, Travis Ryan1; Sashko, Dmytro1; Laniewski-Wollk, Lukasz1,2; Leonardi, Christopher1 1School of Mechanical and Mining Engineering, The University of Queensland, Brisbane, QLD4072, Australia; 2Institute of Aeronautics and Applied Mechanics, Warsaw University of Technology, Warsaw00-665, Poland The propagation of two-phase flows through complex geometries are ubiquitous in nature and have numerous application in engineering fields. Of particular interest in this work is the risk of migration of liquid and gas products adjacent to defective oil & gas wells, potentially causing leakage pathways to aquifers and or the surface. In this setting (i.e. defective cement), the flow geometries have a high degree of randomness generally forming complex microannular voids and fractures either through the wellbore cement or between the cement-formation and or cement-pipe boundaries. This work looks to describe these defects as fractures in which the surfaces are correlated self-affine random fields, which allows the random nature of fractures and cracks to be quantified based on a handful of measurable parameters. The aim is to investigate the dependence of relative permeability on these parameters. In this work, the phase-field lattice Boltzmann model originally proposed by Fakhari et al. [1,2] is extended to account for three-phase contact angles using both the free-energy and geometric approaches for staircase and smoothly defined solid normals. These are often investigated in two-dimensional settings, with non-trivial extensions to complex, three-dimensional geometries. As such, a projection of the phase gradient on the fluid nodes to the plane tangent to the surface is proposed, simplifying the need to determine consistent normal vectors that correspond with the reference frame on every boundary node. The existing implementations outlined in the literature along with various discretisation methods for the gradients are investigated. This is done through examining the performance and accuracy obtained when studying the Washburn law and interaction of droplets on curved boundaries. Finally, the methods are applied to a stochastically generated, three-dimensional fracture to show the variations in fluid behaviour that can be observed based on the implementation of the three-phase contact. [1] Fakhari, A., Mitchell, T., Leonardi, C., Bolster, D., Improved locality of the phase-field lattice-Boltzmann model for immiscible fluids at high density ratios, Physical Review E 96, 053301, 2017. [2] Mitchell, T., Leonardi, C., Fakhari, A., Development of a three-dimensional phase-field lattice Boltzmann method for the study of immiscible fluids at high density ratios, International Journal of Multiphase Flow 107, 1-15, 2018. 10:30am - 10:50amEngulfment of a drop on solids coated by thin and thick fluid films Zhao, Chunheng The City College of New York, United States of America The short-time evolutions of water droplets propagating on both the thin and the thick oil films are explored by the conservative phase-field Lattice Boltzmann method (LBM). The simulation is conducted to investigate the effect of oil layer thickness on the spreading and the engulfing processes. The simulation and the mathematical derivation are used to demonstrate the scaling analysis of those processes. The simulation findings show that the spreading process on the thin film $H/R<<1$ follows the same scaling rule for both $Oh<<1$ and $Oh>>1$ at the short time: $r/R~ (T/t)^{0.5}$. Besides, 2D simulation produces a similar result comparing to 3D simulation for droplet spreading on thin film. When we increase the film thickness gradually, the effect of the film thickness on the spreading process disappears. This argument is also approved by the viscous dissipation from the flow field. Through the comparison between our simulation results and the experimental results, the short time spreading radius of $Oh<<1$ on a thick film follows the scaling rule: $r/R~(T/t_ \eta)^{0.6}$. 10:50am - 11:10amImposing Ratios of Outlet Flow Rates on Large Arterial Networks with Two-Element Windkessel Model: Parametric Analysis Lo, Sharp Chim Yui1; McCullough, Jon1; Coveney, Peter V.1,2 1University College London, United Kingdom; 2University of Amsterdam, Netherlands Substantial effort is being invested in the creation of a virtual human --- a model which will improve our understanding of human physiology and diseases and assist clinicians in the design of personalised medical treatments. A central challenge of achieving blood flow simulations at full-human scale is the development of an efficient and accurate approach to imposing boundary conditions on many outlets. A previous study proposed an efficient method for implementing the two-element Windkessel model to control the flow rate ratios at outlets. However, no study to date has examined the conditions for this approach to hold in complex geometries. Here we clarify the general role of the resistance and capacitance in this approach. We show that the error of the flow rate ratios decreases exponentially as the resistance increases. The errors fall below 4\% in a simple five-outlets model and 7\% in a human artery model comprising 10 outlets. Moreover, the flow rate ratios converge faster and suffer from weaker fluctuations as the capacitance decreases. Our findings also establish constraints on the parameters controlling the numerical stability of the simulations. The findings from this work are directly applicable to larger and more complex vascular domains encountered at full-human scale. 11:10am - 11:30am Coffee Break 11:30am - 12:50pm LBM for MHD and Other Complex FlowsLocation: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle UniversitySession Chair: Li-Shi Luo, CSRC 11:30am - 11:50amThe Role of Hydrodynamics in the Dynamic Response of Magnetic Particles in a Polymer Suspension Kreissl, Patrick; Holm, Christian; Weeber, Rudolf Institute for Computational Physics, University of Stuttgart, Germany Composites of magnetic nanoparticles and polymers are interesting materials, as they combine viscoelastic properties of the polymers with the possibility for external control via a magnetic field. Possible applications include micro-actuation and drug delivery. Many variants of such composite materials have been realized experimentally, spanning magnetic particles in polymer suspensions, hydrogels, and rubber-like materials. While those materials have been synthesized successfully, accessing microscopic detail experimentally is challenging. This is true in particular for the coupling between the magnetic nanoparticles and the polymers. For these questions, simulations are a very useful tool, as one can control the nanoparticle-polymer interaction and observe the corresponding dynamic response of the system. In this contribution, we will focus on a model system consisting of magnetically blocked nanoparticles in a polymer suspension. The magnetic moment being blocked means that a re-alignment of the magnetic moment implies a rotation of the particle, and vice versa. This is a suitable approximation for cobalt ferrite nanoparticles with a size of several nanometers, as they are commonly used in experiments. Magnetic AC susceptibility measurements, i.e., determining the magnetic response to a small applied AC magnetic field, are employed to probe the local environment of the magnetic particles[1]: assuming a Stokes-like friction of the particle, Germain-Dimarzio-Bishop theory can be used to calculate the local viscoelastic moduli as they are felt by the magnetic nanoparticles in their polymeric environment. In our presentation, we will report on simulations corresponding to those experiments[2]. We use a combination of coarse-grained particles treated via molecular dynamics simulations coupled to a thermalized lattice-Boltzmann solver. By controlling the interactions between the nanoparticles and the polymers, we can determine the individual contributions to the magnetic susceptibility response, and hence the degree of mobility the particles possess in their local environment. Our results demonstrate that hydrodynamic interactions alone can reproduce the trends observed in experiments. Hydrodynamics therefore plays a key role in understanding magnetic particle-polymer coupling in those systems. [1] E. Roeben, L. Roeder, S. Teusch, M. Effertz, U. Deiters, A. M. Schmidt. Colloid and Polymer Science, 292, 2014. [2] P. Kreissl, C. Holm, R. Weeber. Soft Matter, 17, 2021. 11:50am - 12:10pmDouble MRT-LBM Analysis of the Coupled Radiation-Convection in a Square Cavity Filled with a Newtonian and non-Newtonian Hybrid Nanofluids under the magnetic field effects chtaibi, khalid1,2; HASNAOUI, Mohammed1; Dahani, Youssef1; Ben Hamed, Haïkel2; Amahmid, Abdelkhalek1 1UCA, Faculty of Sciences Semlalia, Department of Physics, LMFE, B.P. 2390 Marrakesh, Morocco; 2UPJV, University Institute of Technology, LTI, Amiens, France In recent decades, the Lattice Boltzmann Method (LBM) has become one of the most widely used numerical tools by the scientific community for simulating the heat transfer generated by natural convection of Newtonian and non-Newtonian fluids. The attractiveness of this mesoscopic method (simplicity, robustness and ease adaptation to different geometries) has facilitated its breakthrough and allowed it to compete seriously with the classical methods used for simulations purposes. In the present study, Double-Multiple-Relaxation-Time LBM was used to study heat transfer by MHD natural convection in a square configuration filled with a non-Newtonian Fe3O4-MWCNTs/water hybrid nanofluid in the presence of thermal radiation. The numerical simulation results were obtained at given Prandtl (Pr = 6.2) and Rayleigh (Ra = 105) numbers, while the remaining governing parameters were varied in wide ranges that are (0 ≤ Ha ≤ 50) for the Hartmann number, (0 ≤ φ ≤4%) for the nanoparticles volume fraction, (0.6 ≤ n ≤1.4) for the power-law index, and (0 ≤ Rd ≤ 2) for the radiation parameter. The findings of the present study illustrate the combining effects of these parameters in terms of streamlines, isotherms, and mean Nusselt numbers. Attenuation effects, characterized by a decrease in the flow intensity and a degradation of heat transfer process, are observed by increasing either the Hartmann number or the power-law index. However, increasing the radiation parameter or the nanoparticle volume fraction has led to an opposite effect that promotes both the heat transfer and the flow intensity. 12:10pm - 12:30pmNon linear stability of Lattice Boltzmann scheme for under resolved simulation using global optimisation Dubois, François1,2; Saint-Jean, Christophe3; Tekitek, Mohamed Mahdi3 1Laboratoire de Mathématiques d'Orsay, bâtiment 307, F-91405 Orsay, France; 2Conservatoire National des Arts et Métiers, LMSSC laboratory, F-75003 Paris, France; 3MIA laboratory, La Rochelle University, 17000 La Rochelle, France Previous works [2,3] showed that D2Q9 BGK are unstable for test case given in [1] but D2Q9 MRT still stable for such nonlinear problem. In other hand, to investigate the stability of LB scheme, it is possible only numerically using von Neumann analysis [4], and only for linear case. In this work, the Minion et al. [1] test case stability is investigated for a fixed viscosity. Regarding relaxation parameters(free, no effect up to order 2), the stability zone is investigated and characterized using a decision tree, a machine learning technique focused on interpretability. In order to go further, a simple global optimization method (genetic algorithm) is used to yield a set of stable relaxation parameters for the Minion et al. [1] and Taylor-Green test cases. Finally, we show that this optimization method also leads to find a stable non-trivial (non-physical) LB parameter set for the non-linear case. [1] M.L. Minion, D.L. Brown, Performance of under-resolved two-dimensional incompressible flow simulations II, J. Comput. Phys., vol. 138, (1997). [2] P.J. Dellar, Bulk and shear viscosities in lattice Boltzmann equations, Phys. Rev. E, 64 (2001). [3] D. Ricot, Simon Marié, P. Sagaut, C. Bailly, Lattice Boltzmann method with selective viscosity filter, J. Comput. Phys., vol. 228, (2009). [4] P. Lallemand, L.-S. Luo, Theory of the lattice Boltzmann method: Dispersion, dissipation, isotropy, Galilean invariance, and stability, Phys. Rev. E, vol. 61, 2000 12:30pm - 12:50pmStructure-preserving machine learning moment closure models for the radiative transfer equation Huang, Juntao Michigan State University, United States of America In this talk, we present our work on structure-preserving machine learning (ML) moment closure models for the radiative transfer equation. Most of the existing ML closure models are not able to guarantee the stability, which directly causes blow up in the long-time simulations. In our work, with carefully designed neural network architectures, the ML closure model can guarantee the stability (or hyperbolicity). Moreover, other mathematical properties, such as physical characteristic speeds, are also discussed. Extensive benchmark tests show the good accuracy, long-time stability, and good generalizability of our ML closure model. 12:50pm - 1:20pm ICMMES Awards & ClosingLocation: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle UniversitySession Chair: Li-Shi Luo, CSRCSession Chair: Catherine Choquet, La Rochelle University 1:20pm - 2:50pm LunchLocation: Room MSI 218, Maison des Sciences de l'Ingénieur (MSI), La Rochelle University