On transient hybrid Lattice Boltzmann–Navier-Stokes flow simulations
Introduction
Both Lattice Boltzmann (LB) and incompressible Navier-Stokes (NS) solvers are well established methods to compute and, thus, predict fluid flow behavior. Both approaches come with advantages and disadvantages. On the one hand, LB methods were shown to provide accurate prediction of fluid flow over a wide range of temporal and spatial scales, including fluctuating hydrodynamics for nanoflow systems [1] and rarefied gas flows in the slip [2] and transition regime [3]. From the algorithmic perspective, they are rather simple and very local in their computations. This allows for highly efficient, massively parallel simulations as well as block-adaptive GPU variants, see amongst others [4], [5]. Incompressible NS solvers, on the other hand, typically come along with lower memory consumption, and often allow the use of coarser grids and rigorous higher-order discretizations such as discontinuous Galerkin approaches [6], [7] which are rather uncommon in the LB framework [8], [9]. For more detailed comparisons of LB and NS approaches, see amongst others [10], [11]. Hence, one or the other solver, or even a combination of both to exploit their features to highest extent may be favorable for a particular flow problem.
Various works on coupling LB solvers with respective finite difference or finite volume discretizations of the underlying, asymptotically approximated PDEs have been carried out. In the Navier-Stokes context, Latt et al. [12] have provided a coupling method for steady state channel flow scenarios. Yeshala [13] partly employs this approach to address flow around cylinders and other scenarios. Our group has recently developed and evaluated different Schwarz-like domain decomposition techniques for spatial steady-state LB–NS simulations [14].
In this contribution, we extend our considerations to transient coupling of LB and NS solvers for laminar flows. One- and two-way coupled systems are investigated in Couette, Taylor–Green and Karman vortex flows, the latter based on the 2D2 benchmark [15]. We show that one-way coupling, i.e. using a NS solver to impose arbitrary boundary/flow field conditions to LB, yields accurate results in all of these scenarios. This result has several, but most importantly two, implications. First, this allows to generate boundary conditions from a computationally cheap, coarse-grained flow model and plug them into the LB method to construct accurate predictions on finer scales at higher computational cost. Second, incompressible Navier-Stokes solvers may—based on these results—be used to represent an accurate initial flow field predictor. Another potential application may be found in parallel-in-time CFD simulations where LB is employed for the fine-scale simulation, cf. amongst others [16], and NS is used as coarse-scale predictor. This would further improve the (strong) scaling limits of this method.
Moreover, we consider two-way coupling where the domain is decomposed into separate LB and NS overlapping domains. We show that the weak compressibility can have a severe impact on the overall solution quality of the coupling, and briefly point out the importance of a valid choice of LB and NS time steps at the example of the Taylor–Green scenario.
We revise the theory for LB, NS, and their implementations in Section 2. The coupling methods—one- and two-way coupling—are discussed in Section 3. The studies of Couette, Taylor–Green and Karman vortex scenarios are presented in Section 4. We close with a short summary and outlook to future work in Section 5.
Section snippets
Lattice Boltzmann
We consider the LB method with a standard D2Q9 discretization [17] together with the BGK collision operator [18]to compute the probability density function fi related to lattice velocity ci and its relaxation towards equilibrium stateat rate . The lattice weights and velocities are given byandDensity ρLB, pressure p
Coupling
We use time-explicit coupling algorithms to combine LB and NS solver. Both solvers use constant time steps dtNS, dtLB with dtNS = N · dtLB and integer N > 0. The LB solver operates on a finer mesh with mesh size dxLB = dxNS/M with integer M > 1. The LB algorithm is carried out on the nodes of the grid, i.e. the vertices of the fine Cartesian grid represent the LB degrees of freedom. This yields a closed layer of LB nodes on the “NS → LB” interface, cf. thick dashed line in Fig. 1.
Each simulation starts
Results
We investigated the coupled scheme in the following transient flow scenarios: Couette Flow, Oscillating Couette Flow, Taylor–Green, and Karman Vortex 2D2. Table 1 summarizes the parameters that are used to describe these scenarios.
Summary
We analyzed one- and two-way coupled LB–NS simulations in transient flow scenarios. The one-way coupling “NS → LB” exhibits excellent agreement between NS and LB solutions. NS solvers can thus be used as (potentially more efficient) preconditioners for LB simulations, or to impose time-dependent boundary conditions. Even the time-dependent behavior of two-way coupled Couette flow agrees very well with the analytic solution, without any particular measures to remove or reduce compressibility
Acknowledgements
This work is partially supported by the Award No. UK-C0020 made by King Abdullah University of Science and Technology (KAUST). P. Neumann particularly thanks Ravikishore Kommajosyula for preliminary studies on two-way LB–NS coupling for oscillatory channel flows.
Philipp Neumann studied Technomathematik at the FAU Erlangen-Nürnberg and graduated in 2008. He obtained his Ph.D. (Dr. rer. nat.) from the TU München in 2013 for his thesis on “Hybrid Multiscale Simulation Approaches for Micro- and Nanoflows”. He is currently employed as postdoctoral researcher at TU München.
References (23)
- et al.
Lattice Boltzmann modeling of microchannel flow in the slip flow regime
J. Comput. Phys.
(2009) - et al.
Multi-thread implementations of the lattice Boltzmann method on non-uniform grids for CPUs and GPUs
Comput. Math. Appl.
(2011) - et al.
A high order discontinuous Galerkin finite element solver for the incompressible Navier-Stokes equations
Comput. Fluids
(2011) - et al.
An implicit high-order discontinuous Galerkin method for steady and unsteady incompressible flows
Comput. Fluids
(2007) Fast discontinuous Galerkin lattice-Boltzmann simulations on GPUs via maximal kernel fusion
Comput. Phys. Commun.
(2013)- et al.
A spectral-element discontinuous Galerkin lattice Boltzmann method for nearly incompressible flows
J. Comput. Phys.
(2011) - et al.
Benchmark computations based on lattice-Boltzmann, finite element and finite volume methods for laminar flows
Proceedings of the First International Conference for Mesoscopic Methods in Engineering and Science
Comput. Fluids
(2006) - et al.
Comparison between lattice Boltzmann method and NavierStokes high order schemes for computational aeroacoustics
J. Comput. Phys.
(2009) - et al.
Parallel in time approximation of the lattice Boltzmann method for laminar flows
J. Comput. Phys.
(2014) - et al.
Lattice Boltzmann method with regularized pre-collision distribution functions
Math. Comput. Simul.
(2006)
Lattice Boltzmann simulations of soft matter systems
Adv. Polym. Sci.
Cited by (4)
A hybrid lattice Boltzmann - Navier-Stokes method for unsteady aerodynamic and aeroacoustic computations
2023, Journal of Computational PhysicsThe 24th International Conference on Discrete Simulation of Fluid Dynamics in Edinburgh, Scotland
2016, Journal of Computational ScienceGenerating Coupling Interfaces for Multiphysics Simulations with ExaStencils and waLBerla
2023, 2023 IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2023
Philipp Neumann studied Technomathematik at the FAU Erlangen-Nürnberg and graduated in 2008. He obtained his Ph.D. (Dr. rer. nat.) from the TU München in 2013 for his thesis on “Hybrid Multiscale Simulation Approaches for Micro- and Nanoflows”. He is currently employed as postdoctoral researcher at TU München.