Numerics of the lattice boltzmann method on nonuniform grids:
Standard LBM and finite-difference LBM
Abbas Fakhari, Taehun Lee
⇑
Department of Mechanical Engineering, City College of City University of New York, New York, NY 10031, USA
article info
Article history:
Received 24 June 2014
Received in revised form 3 November 2014
Accepted 8 November 2014
Available online 20 November 2014
Keywords:
Adaptive mesh refinement
Lattice Boltzmann method
Finite difference method
abstract
The present study is focused on the comparison between the standard ‘‘collide-and-stream’’ lattice Boltz-
mann method (LBM) and the Lax-Wendroff-based finite-difference LBM (FDLBM) on block-structured
nonuniform grids with an adaptive mesh refinement (AMR) strategy. While the standard LBM (SLBM)
is found to be slightly faster than the FDLBM, the latter is shown to be more stable at higher Reynolds
numbers. Although both approaches are as accurate in simulation of fluid flow problems, the SLBM
has a more complicated algorithm and its implementation is more involved; this is mainly because, in
applying SLBM, the AMR blocks at different refinement levels do not advance in time simultaneously.
On the other hand, the underlying differences between the cell-center and cell-vertex data structures
are explained and their advantages and disadvantages are highlighted. In general, the cell-center data
structure is favorable because it is more efficient in terms of computational time and memory. The effect
of the interpolation schemes on the order of accuracy of the LBM is also investigated. It is reestablished
that the popular linear interpolation degrades the order of accuracy of LBM to first order. A variety of
benchmark studies, including Taylor–Green decaying vortex, gravity-driven Poiseuille flow, thin shear
layer instability, and unsteady flow past a square cylinder, are carried out to assess SLMB and FDLBM with
a multiple-relaxation-time collision operator.
Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction
Recent evolutions in computer technology have made it easier
to perform numerical simulations using computational fluid
dynamics (CFD) techniques, especially on highly parallel comput-
ing machines. Along with the advancements in computer
technology, the numerical methods for the solution of fluid flow
have also evolved. The complicated adaptive mesh refinement
(AMR) strategies and the load balancing techniques associated
with them are now very common in daily simulations [1,2].
As a CFD tool to tackle fluid flow problems, the lattice Boltz-
mann method (LBM) has been in use for more than two decades
[3]. It has been harnessed and developed by many researchers in
different fields [4]. The implementation of the lattice Boltzmann
equation (LBE) on nonuniform grids has been facilitated by the
interpolation-supplemented LBM [5,6]. To improve the geometri-
cal flexibility of the LBM, several finite-difference [7–11],
finite-volume [12–15], and finite-element [16] models have been
proposed. The majority of the proposed LBE models employ a static
grid refinement scheme at the beginning of the simulation [17–19].
Only few works have been devoted to develop an LBM with
dynamic mesh refinement [20–23,10].
In the previous work [10], we proposed a new adaptive mesh
refinement technique that does not need to maintain or
manipulate a tree-type data structure. Therein, we highlighted
the differences between the standard LBM (SLBM) and the finite-
difference LBM (FDLBM) on nonuniform grids but we did not make
a comparison between them. This is the subject of the present
study to assess both models in terms of efficiency, stability, and
ease of implementation. We start by outlining the constituent
elements of the AMR and elaborating the differences between
the cell-vortex and cell-center data structures in Section 2.We
then describe the SLBM and the FDLBM in Section 3, followed by
numerical studies and detailed analysis of the two models in
Section 4. Finally, we summarize the paper in Section 5.
2. Adaptive mesh refinement strategy
In our previous work [10], we proposed an AMR with a cell-
vertex data structure (see Fig. 1(a)). In the current study, we
employ a cell-center data structure as shown in Fig. 1(b), and
describe the differences between the two. The algorithm of the
http://dx.doi.org/10.1016/j.compfluid.2014.11.013
0045-7930/Ó 2014 Elsevier Ltd. All rights reserved.
⇑
Corresponding author.
E-mail addresses: afakhar00@citymail.cuny.edu (A. Fakhari), thlee@ccny.cuny.
edu (T. Lee).
Computers & Fluids 107 (2015) 205–213
Contents lists available at ScienceDirect
Computers & Fluids
journal homepage: www.elsevier.com/locate/compfluid