XFlow 2022

THEORY GUIDE

©2021 Dassault Systèmes. All rights reserved. 3DEXPERIENCE®, the Compass icon, the 3DS logo, CATIA, BIOVIA, GEOVIA, SOLIDWORKS, 3DVIA, ENOVIA, EXALEAD, NETVIBES, MEDIDATA, CENTRIC PLM, 3DEXCITE, SIMULIA, DELMIA, and IFWE are commercial
trademarks or registered trademarks of Dassault Systèmes, a French “société européenne” (Versailles Commercial Register # B 322 306 440), or its subsidiaries in the United States and/or other countries. All other trademarks are owned by their respective owners. Use
of any Dassault Systèmes or its subsidiaries trademarks is subject to their express written approval.
SIMULIA XFlow is © 2011 - 2021 Dassault Systèmes España, SLU .

Trademarks
XFlow, 3DEXPERIENCE, the Compass logo and the 3DS logo, CATIA,
SOLIDWORKS, ENOVIA, DELMIA, SIMULIA, GEOVIA, EXALEAD, 3D VIA,
BIOVIA, NETVIBES, and 3DEXCITE are commercial trademarks or registered
trademarks of Dassault Systèmes, a French ”société européenne” (Versailles
Commercial Register # B 322 306 440), or its subsidiaries in the U. S. and/or
other countries. All other trademarks are owned by their respective owners. Use
of any Dassault Systèmes or its subsidiaries trademarks is subject to their express
written approval.
DS Offerings and services names may be trademarks or service marks of Dassault
Systèmes or its subsidiaries.
Legal Notices
XFlow and this documentation may be used or reproduced only in accordance
with the terms of the software license agreement signed by the customer, or,
absent such an agreement, the then current software license agreement to which
the documentation relates.
This documentation and the software described in this documentation are subject
to change without prior notice.
Dassault Systèmes and its subsidiaries shall not be responsible for the consequences
of any errors or omissions that may appear in this documentation.
SIMULIA XFlow is © 2011 - 2021 Dassault Systèmes España, SLU .
For additional information concerning trademarks, copyrights, and licenses, see the
Legal Notices in the XFlow Installation Guide.
XFlow 2022 Theory Guide
 Contents

Contents i

1 Introduction 1

2 The challenge of the traditional CFD 3

2.1 Discretization scheme . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Turbulence modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Numerical scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Particle-Based Methods 7
3.1 Lattice schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Lattice Gas Automata . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Lattice-Boltzmann method . . . . . . . . . . . . . . . . . . . . . . . 9
3.4 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4 XFlow: an Innovative LBM Approach 17

4.1 Octree lattice structure . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 Central-moment LBM . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3 Wall-adapting local eddy viscosity . . . . . . . . . . . . . . . . . . . 19
4.4 Advanced wall treatment . . . . . . . . . . . . . . . . . . . . . . . . 20
4.5 Advanced dynamic geometries . . . . . . . . . . . . . . . . . . . . . . 21
4.6 Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

References 22
ii
 1
Introduction

The Computational Fluid Dynamics (CFD) allows the analysis of flow behavior in
virtual environments, lowering costs and times, and being a real alternative to the
analysis in experimental facilities. Furthermore, CFD analysis provides valuable
information for the understanding of flow behavior, allowing the analysis and
measurement of fluid magnitudes through the whole domain without interference.
Benefits from CFD analysis are nowadays clearly stated and computational
simulation is providing new insights at all stages of product life-cycle management
and specifically at design stage. These are not just as a replacement to the
measurement of loads at experimental facilities but also in order to better understand
why the performance of some models may dramatically change under slight
modifications.
The need for robust and reliable CFD solutions is therefore increasing strongly, as
well as the complexity of the simulations. To provide quick and accurate results to
complex and realistic engineering problem is consequently essential for companies
to be competitive.
The main numerical methodologies investigated so far were the classical finite
volume and finite elements methods, based on the Navier-Stokes equation. However,
even though those methods have been widely investigated they still face important
drawbacks limiting their application to real industrial problems. Alternative
particle-based methods are now emerging and consist in a real alternative to
overcome these drawbacks. Among them, the lattice-Boltzmann method (LBM)
is becoming one of the most interesting alternative, and the development of new
commercial LBM codes such as XFlow provide new alternatives for companies to
solve complex engineering problems.
2 Introduction
 2
The challenge of the traditional CFD

2.1 Discretization scheme

The traditional CFD solvers solving the Navier-Stokes equations are generally mesh-
based solutions, which therefore implies the need to go through a time-consuming
meshing process. For every CFD study, a large amount of time will dedicated to the
meshing and therefore represents an important cost for companies.
Many studies are constrained to use simplified geometries in order to make the
meshing process possible or to avoid convergence issues due to bad quality mesh
(proportioned aspect ratio, high orthogonality, low skewness, smooth expansion
transition, etc.). Small gaps may also give troubles in the mesh generation, as
for example the gaps between flap, slat and wing of an aircraft in the high-lift
configuration as illustrated on Figure 2.1.

Figure 2.1: Example of mesh of an aircraft wing section in high-lift configuration.

The mesh quality is not only constraining for the numerical stability, it also strongly
influences the quality of the solution. Several meshes of equivalent size could provide
different solutions, which introduces uncertainties on the solution. Engineers must
carefully chose the type of mesh elements adapted to a given problem (structured,
unstructured, hybrid, etc.) and also correctly anticipate the areas of refinement
(wake, interface, etc.).
4 The challenge of the traditional CFD

Finally, the moving geometries cannot be handled in a straight-forward manner as

they imply changes of the fluid domain topology. Despite methods such as sliding
mesh exist to model moving geometries, they requires a very time-consuming and
unstable re-meshing process.

2.2 Turbulence modeling

The turbulence is a complex mechanism of the fluid dynamics and its modeling has
always been a real challenge in CFD. In practice, almost all engineering applications
are turbulent nowadays and it is therefore a key issue to accurately model the
turbulence.
The Navier-Stokes equations have no analytic solution unless you are considering
strong simplifications and assumptions, and their direct resolution (Direct Numerical
Simulation, known as DNS) is therefore almost impossible on real engineering cases
due to computational limitations and is limited to academic applications only.
Models are thus required in order to take into account the turbulence effects on
the solution in turnaround times that can satisfy the industry.
The turbulence modeling first started with Joseph Boussinesq who proposed to
introduce the concept of turbulent viscosity in order to relate turbulence stresses
to mean flow. This turbulent viscosity models the transfer of momentum caused
by turbulent eddies, and has been then complemented by Prandtl’s mixing length
concept in order to take into account the wall effects more effectively.
Many turbulence models have been developed based on these hypothesis to solve
the Reynolds-Averaged Navier-Stokes equations and are known as RANS turbulence
models. The most common ones are the Spalart-Allmaras, k − , k − omega, and
the SST. The RANS models are nowadays widely applied in the industry mostly
because of their low computational cost and their reasonable accuracy.
However, they are empirical models and thus face the drawback of relying a
large number of constants that one must carefully calibrate in order to model the
turbulence correctly. Every of those model may include different sets of constants
and may predict a considerably different solution as separation prediction will
strongly depend on the turbulence. The choice of the correct model therefore highly
depends on the considered application, and introduces major uncertainties when
the product design goes through important changes. Also, the RANS turbulence
models are modeling all eddy scales without distinction and filter all instantaneous
variations induced by the turbulence since based on time-averaged equations. They
are mostly limited to steady-state analysis and the actual turbulent energy spectrum
described by Kolmogorov cannot be entirely and accurately captured. This is an
important drawback especially for aeroacoustics applications.
2.3 Numerical scheme 5

An intermediate class of turbulence models was first proposed by Joseph

Smagorinsky in 1964 to generalize the use of the turbulent viscosity applied on sub-
grid scales that are unresolved. This turbulence modeling approach is named the
Large Eddy Simulation (LES) and proposes a good intermediate solution between
the RANS and DNS, since most of the turbulence is directly resolved but smaller
scales are modeled to thus reach accessible computation times as illustrated on
Figure 2.2.

Figure 2.2: Turbulence modeling approaches comparison.

The LES turbulence models are not widely employed however, since the computation
cost is relatively high compared to RANS models with the Navier-Stokes solvers.
They are used for high-end applications or in industries that require transient
simulations with highly turbulent and separated flows, where the RANS models
tend to have poor predictions.

2.3 Numerical scheme

The mathematical formulation of the Navier-Stokes equations is non-trivial to

resolve numerically. The numerical analysis of these equations is a wide research
topic and plenty of schemes have been developed, each of them with their
advantages and drawbacks depending on the assumptions considered (compressible
or incompressible, steady or unsteady, etc.).
Due to the nonlinear character of the behavior of fluids, the solution of the
Navier-Stokes system of equations requires a suitable linearization of the discrete
formulation of the original system. An Eulerian mathematical formulation is the
natural approach for finite elements and finite volumes, but this introduces many
difficulties in the non-linear advection term of the momentum equations.
6 The challenge of the traditional CFD

Most of traditional codes artificially stabilize the solution by adding artificial

numerical viscosity that strongly changes and dampens the physics, specifically
for the higher frequencies, and with a dramatic negative effect in the accuracy for
unsteady flow and acoustics.
 3
Particle-Based Methods

Particle-based methods have been developed for several decades now and start to
be emerging recently. Among them, the Lattice-Boltzmann Method (LBM) which
tackles many of the drawbacks presented by the traditional CFD methods: the
meshing process is removed as the simulation relies on an automatically generated
lattice which is organized in an Octree structure, the LBM scheme allows the
use of Large Eddy Simulation (LES) turbulence model at low computational cost,
and finally the mathematical formulation of the Boltzmann transport equation is
naturally straightforward to resolve numerically.
Over the last few years, schemes based on minimal kinetic models for the Boltzmann
equation are becoming increasingly popular as a reliable alternative to conventional
CFD approaches.
The Lattice Boltzmann method (LBM) was originally developed as an improved
modification of the Lattice Gas Automata (LGA) to remove statistical noise and
achieve better Galilean invariance ([1, 2]). Due to the flexibility afforded by its close
connection to kinetic theory, the LBM can be adapted to model several physical
phenomena. Recent research has led to major improvements, including physically
consistent models for multiphase and multi-component flow and fully compressible
flow ([3, 4]).

3.1 Lattice schemes

Some of the particle-based methods such as LGA and LBM work on a spatial
discretization named lattice, consisting of a Cartesian distribution of discrete points
with a discrete set of velocity directions.
The discrete set of velocities has a common terminology refereed to the dimension
of the problem and the number of velocity vectors DnQm, where n represents the
dimension of a problem and m refers to the number of velocities direction.
8 Particle-Based Methods

For a two-dimensional model, the most common scheme is the D2Q9 model
illustrated on Figure 3.1 and involves 9 velocity vectors defined by:


 (0, 0)
 i=0
~ei = (1, 0), (0, 1), (−1, 0), (0, −1) i = 1, 2, 3, 4 (3.1)

(1, 1), (−1, 1), (−1, −1), (1, −1) i = 5, 6, 7, 8


Figure 3.1: D2Q9 Lattice model of velocities discretization

For a three dimensional model, the most scheme is the D3Q19 model illustrated on
Figure 3.2 and involves 19 velocity vectors defined by:


 (0, 0, 0)
 i=0
~ei = (±1, 0, 0), (0, ±1, 0), (0, 0, ±1) i = 1, 2, ..., 5, 6 (3.2)

(±1, ±1, 0), (±1, 0, ±1), (0, ±1, ±1) i = 7, 8, ..., 17, 18


Figure 3.2: D3Q19 Lattice model of velocities discretization

Other LBM implementations use the scheme D3Q27 which provides a higher number
of velocity directions, and thus provide a higher order scheme despite of a slightly
higher computational cost. Indeed, the more velocity directions and the more
equations are to solve.
3.2 Lattice Gas Automata 9

3.2 Lattice Gas Automata

The Lattice Gas Automata (LGA) is a simple scheme for modeling the behavior
of gases. The basic idea behind the LGA is that particles with specific velocities
(ei , i = 1, ..., b) propagate through a d-dimensional lattice, at discrete times t =
0, 1, 2, ... and collide according to specific rules designed to preserve the mass and
the linear momentum when different particles reach the same lattice position.
The simplest LGA model is the HPP approach, introduced by Hardy, Pomeau and
de Pazzis, in which particles move in a two-dimensional square lattice and in four
directions (d = 2, b = 4). The state of an element of the lattice at instant t is given
by the occupation number ni (r, t), with ni = 1 being presence and ni = 0 absence
of particles with velocity ei .
The stream-and-collide equation that governs the evolution of the system is

ni (r + ei , t + dt) = ni (r, t) + Ωi (n1 , ..., nb ), i = 1, ..., b, (3.3)

where Ωi is the collision operator that computes a post-collision state conserving

mass and linear momentum. If one were to assume Ωi = 0, only an streaming
operation would be performed.
From a statistical point of view, the system is made up of a large number of elements
which are macroscopically equivalent to the problem investigated. The macroscopic
density and linear momentum can be computed as:
b
1X
ρ = ni (3.4)
b
i=1
b
1 X
ρv = ni ei (3.5)
b
i=1

3.3 Lattice-Boltzmann method

The Lattice Boltzmann Method (LBM) is steadily becoming a relevant method for
solving the fluid flow equations. Its direct connection to the kinetic theory of gases
allows for a deeper understanding and more detailed modeling of physical phenomena
[3, 4]. This is leading to its adoption by many groups as the de-facto tool for studying
multiphase/multicomponent flow ([5]). New developments ([6]) are also enabling its
application in the study of compressible flow.
There are many approaches to the LBM, and one can indeed think of it as group of
methods aimed at solving systems of equations for nonlinear hyperbolic conservation
laws. The common characteristic to all the LBM models (and Lattice Gas Automata
([1, 2])) is their time-stepping model, based on a propagate-collide scheme, on top
10 Particle-Based Methods

of a lattice discretization. The propagation step performed on a lattice enforces a

constant timestep dt and a discrete set of velocities (ei , i = 1, ..., b) that ensure that
the positions of the simulated particles motion is restricted to lattice sites. The
set of velocities thus generates the lattice and for each lattice site, b probability
distribution functions (PDFs) fi (r, t) are stored.
In the continuum space with discrete velocities, the Boltzmann transport equation
can be written as follows:

∂fi
+ ei · ∇fi = Ωi , i = 1, ..., b, (3.6)
∂t
where Ωi is the collision operator that computes a post-collision state conserving
mass and linear momentum. If one were to assume Ωi = 0, only a streaming
operation would be performed. This equation is discretized on the lattice as:

fi (r + ei , t + dt) = fi (r, t) + Ωi (f1 , ..., fb ), i = 1, ..., b. (3.7)

The stream-and-collide scheme of the LBM can be interpreted as a discrete

approximation of the continuous Boltzmann equation. The streaming or propagation
step models the advection of the particle distribution functions along discrete
directions, while most of the physical phenomena are modeled by the collision
operator which also has a strong impact on the numerical stability of the scheme.
As in the continuum Boltzmann equation, the macroscopic variables can be derived
from the statistical moment of the PDFs:
b
X
ρ = fi (3.8)
i=1
b
X
ρv = fi ei (3.9)
i=1

Moment of zero-order corresponds to the macroscopic density, and the moments

of first-order provide the momentum in the three directions. Further moments of
higher orders can be computed to obtain higher order quantities.

3.3.1 Single relaxation time

The collision operator is generally modelled as a relaxation of the PDFs towards an

equilibrium state. A single-relaxation time (SRT) based on the Bhatnagar-Gross-
Krook (BGK) approximation is used in the most common approach:

1 eq
ΩBGK
i = (f − fi ), (3.10)
τ i
3.3 Lattice-Boltzmann method 11

where τ is the relaxation time, and fieq is the local equilibrium function towards
which the state is relaxed. The Chapman-Enskog expansion applied to the LBM
equation allows to identify the kinematic viscosity ([7]) with the relaxation time,
and provides the following relation
1
ν = c2s (τ − ). (3.11)
2
where cs is the speed of sound. The relaxation of the collision step controls the
response to equilibrium, thus modeling fluid kinematic viscosity.
The local equilibrium function fieq is generally derived from a Maxwell-Boltzmann
distribution with the same macroscopic variables of the pre-collision state, ensuring
mass and momentum conservation. It is a function only of the macroscopic
properties of the flow, and usually defined as
  
eiα uα uα uβ eiα eiβ
fieq = ρwi 1 + + − δ αβ . (3.12)
c2s 2c2s c2s
where u is the macroscopic velocity, δ the Kronecker delta and the wi are weighting
constants built to preserve the isotropy. The α and β subindexes denote the
different spatial components of the vectors appearing in the equation and Einsteins
summation convention over repeated indices has been used.
By means of the Chapman-Enskog expansion the resulting scheme can be shown
to reproduce the hydrodynamic regime for low Mach numbers ([8, 9]). Recently it
has been shown that there are similarities between the LBM method and the family
of artificial compressibility methods ([10]), and an artificial compressibility method
can be constructed which is very similar to LBM at the algorithmic level.
The SRT collision operator, although still commonly used for many applications,
has several shortcomings that preclude its use. The Mach number is limited to low
values, and the viscosities employed are conditioned by the τ value which becomes
unstable near values of 0.5 [11]. Several alternative methods have been devised to
overcome these problems. For instance, an alternative method is the entropic lattice
Boltzmann (ELBM) scheme, which may rely on a single-relaxation time where the
attractors of the particle distribution functions are not simple polynomial expansions
of the Maxwell-Boltzmann distribution, but instead based on the minimization of a
Lyapunov-type functional enforcing the H-theorem locally in the collision step. This
ensures the stability of the method, however, it is expensive from the computational
point of view ([12]) and thus not used on practical engineering applications.

3.3.2 Multiple relaxation time

Due to the single-relaxation time collision operator limitations, the multiple-

relaxation time (MRT) collision operator was introduced by [13] where the collision
12 Particle-Based Methods

pror le cess is carried out in momentum space instead of the usual velocity space.
The development was continued by [14] for the D2Q9 scheme and by [15] for the
D3Q19 scheme. It has been demonstrated that the MRT models are more stable
than their BGK counterparts ([14]), because the different relaxation times can be
individually adjusted to achieve a good stability.
Analogically to Equations (3.8) and (3.9), the raw moment µ of the probability
distribution function f can be expressed in a more general form by the relation

b
X
µ xk y l z m = fi ekix eliy em
iz (3.13)
i=1
where k, l, and m are respectively the orders of moments taken in x, y, and z
directions respectively. The raw moment order is therefore k + l + m. Denoting µi
as a raw moment µxk yl z m of a given combination of (k, l, m), the relation between
the PDFs and the raw moments can be expressed in the following matrix form

µi = Mij fj (3.14)

where Mij is the transformation matrix. The number of raw moments µi is therefore
limited to the number of discrete velocities.
Similarly to Equation (3.10), the multiple-relaxation time collision operator ΩMRT
can be calculated as a relaxation in momentum space followed by the inverse
transformation to density function space

ΩMRT
i = −Mij−1 Ŝij (µeq
i − µi ) (3.15)
where Ŝij is the diagonal relaxation matrix, and µeq
i is the raw moment at
equilibrium.
Once the post-collision µi is obtained, the post-collision PDFs can be recovered from
inverse transformation matrix
fj = Mij−1 µi (3.16)

As an example for the D2Q9 lattice scheme [16], the following moment vector can
be defined
µ = (µ0 , µ1 , ..., µ8 )T = (ρ, e, , jx , qx , jy , qy , pxx , pyy )T (3.17)
where ρ is the density, e is the energy kinetic,  is related to kinetic energy square,
jx = ρux and jy = ρuy correspond to components of momentum, qx and qy
correspond to the energy components and pxx and pyy correspond to the symmetric
traceless viscous stress tensors. These moments are ordered in function of the tensor
order first, and then by moments order. For example, the first three components
correspond to the zeroth-order tensors (ρ, e, and ), and are respectively zeroth,
second, and fourth order moments of fi (µx0 y0 , µx1 y1 , and µx2 y2 ).
3.3 Lattice-Boltzmann method 13

Using the Equation 3.13 and applying the Gram-Schmidt orthogonalization

procedure ([15]), the raw moments contribution for each fi can be written as
ρ |i = | e i |0 = 1
e |i = −4 | ei |0 +3(e2i,x + e2i,y )
 |i = 4 | ei |0 −21/2(e2i,x + e2i,y ) + 9/2(e2i,x + e2i,y )2
jx |i = ei,x
0
qx | i = ei,x [−5 | ei | +3(e2i,x + e2i,y )] (3.18)
jy |i = ei,y
qy | i = ei,y [−5 | ei |0 +3(e2i,x + e2i,y )
pxx |i = (e2i,x − e2i,y )ρ
pxy |i = ei,x ei,y /ρ

Hence the transformation matrix M is

   
ρ |i 1 1 1 1 1 1 1 1 1
 e |i  −4 −1 −1 −1 −1 2 2 2 2
   
   
  |i   4 −2 −2 −2 −2 1 1 1 1
   
 jx |i   0 1 0 −1 0 1 −1 −1 1 
   
   
M =  qx |i  =  0 −2
   0 2 0 1 −1 −1 1 
 (3.19)
 jy |i   0 0 1 0 −1 1 1 −1 −1
   
   
 qy | i   0 0 −2 0 2 1 1 −1 −1
   
pxx |i   0 1 −1 1 −1 0 0 0 0
   

pxy |i 0 0 0 0 0 1 −1 1 −1

Ŝij = diag(s0 , s1 , s2 , s3 , s4 , s5 , s6 , s7 , s8 ) is the diagonal relaxation matrix. s0 , s3

and s5 are related with density and momentum which have to be conserved and
are therefore set to zero. In this approach the kinematic viscosity ν and the bulk
viscosity µ are related with the following relaxations parameters

1 1 1 1 5 − 9c2s 1 1
ν = c2s (− ) = c2s ( − ), µ = ( − ). (3.20)
s7 2 s8 2 9 s1 2
For stability reason each si value must lie between 0 and 2.
Finally, applying the transformation matrix to the equilibrium PDF of Equation
(3.12), the raw moments at the equilibrium are
eeq = −2ρ + 3(jx2 + jy2 )
eq = ρ − 3(jx2 + jy2 )
qxeq = −jx
(3.21)
qyeq = −jy
peq
xx = jx2 − jy2
peq
xy = jx jy
14 Particle-Based Methods

The moments ρeq , jxeq , and jyeq are not required as they will be multiplied by s0 , s3 ,
and s5 which are zero.

3.4 Boundary conditions

The boundary conditions (BCs) are an important aspect of the numerical

simulations, and their implementations varies depending on the method used. They
are critical for the stability and the accuracy of the simulation.
In the lattice-Boltzmann method, the BCs provide the value for the unknown
PDFs created after the streaming step. The Figure 3.3 shows as an example a
wall boundary condition on a D2Q9 lattice scheme present from the left-side. On
this illustration lattice points are shown on the left of the wall which are used to
implement the wall boundary condition. The unknown PDFs are f1 , f5 and f8 , for
which no incoming PDFs are available from the left-side since no lattice nodes are
present. Note that depending on the wall locations and direction, the unknown PDF
will change. In this example, the boundary condition will therefore impose those
three distribution functions.

Figure 3.3: Unknown PDFs on left wall after the streaming step.

Throughout the years, many boundary condition implementation methods have been
developed. The basic and straight-forward implementations are presented below.

3.4.1 Periodicity

This is the simplest boundary condition implementation. The periodicity boundary

condition takes the unknown PDFs values from the periodic boundary pair.
3.4 Boundary conditions 15

3.4.2 Bounce-back

The bounce-back boundary condition is used to implement the no-slip condition on

a geometry wall. As its name suggests, the incoming PDFs at the wall node are
reflected back to the initial fluid node giving value to the unknown PDFs ([4]). On
the illustration Figure 3.4, at a time t the distribution function f7 is reflected back
to f5 , f3 to f1 , and f6 provides f8 value.

Figure 3.4: Bounce-back boundary condition.

It is quite intuitive and is particularly simple to implement. This approach supposes

that the geometry walls are just in the middle between the lattices nodes, however
this normally does not occur. To address this limitation, advanced implementations
([17, 18]) take into account the distance between the fluid node and the wall node
to get a better accuracy.

3.4.3 Velocity and pressure boundaries

The main inlet and outlet boundary conditions used in CFD are velocity and
pressure, since this is a data that is usually known or can be easily measured. [19]
developed originally the velocity and pressure boundary conditions.
The approach consists in the formulation of a linear system with the first two
moments conservation (mass and momentum). This linear system will provide the
value of the unknown PDFs created after the streaming step as well as the value
of ρ if a velocity boundary condition is imposed, or u a pressure condition at the
boundary. According to the Figure 3.3 example, after rearranging the moments
equations (3.4) and (3.5) (density, x-momentum, y-momentum) the linear system
can be written as
16 Particle-Based Methods

f1 + f5 + f8 = ρ − (f0 + f2 + f4 + f3 + f6 + f7 )
f1 + f5 + f8 = ρu + f3 + f6 + f7 (3.22)
f5 − f8 = ρv + f2 + f4 − f6 + f7

A fourth equation is required to close the system and compute the value of velocity
or pressure. A valid assumption is to apply the bounce-back rule for the non-
equilibrium part of the particle distribution normal to the boundary. Therefore,
considering an inlet boundary is imposed on the left nodes of Figure 3.3 the fourth
equation is

f1 − f1eq = f3 − f3eq (3.23)

Solving the linear system, the unknown PDFs are

2
f1 = ρu
3
1 1 1
f5 = f7 − (f2 − f4 ) + ρu + ρv (3.24)
2 6 2
1 1 1
f8 = f6 − (f2 − f4 ) + ρu − ρv
2 6 2

Hence for velocity boundary condition the linear system gives:

1
ρ= [f0 + f2 + f4 + 2(f3 + f6 + f7 )] (3.25)
1−v

For a pressure boundary condition, taking v = 0, the u is defined by:

[f0 + f2 + f4 + 2(f3 + f6 + f7 )]
u=1− (3.26)
ρ

This approach is quite stable and straightforward, however due to the unsteady and
compressible character of LBM, reflections may occur at outlet and inlets. There
are some boundary conditions developments techniques to address this issue, known
as non-reflecting boundary conditions ([20]).
 4
XFlow: an Innovative LBM Approach

The CFD software XFlow developed by Next Limit Technologies provides a

proprietary fully Lagrangian particle-based kinetic approach, based on the lattice-
Boltzmann method. However, unlike existing LBM codes, XFlow employs an
innovative approach to LBM which provides it unique features.
Among the key features, XFlow provides a unique collision operator scheme and
implementation offering higher order and stability. The octree lattice structure
handles multi-resolution and adaptive refinement, and dynamic geometries can
be handled in a straightforward manner. The turbulence modeling and near-
wall treatment is based on a state-of-the-art Wall-Modeled Large Eddy Simulation
(WMLES) that offers advanced turbulence prediction.
XFlow is able to solve a wide range of problems involving internal and external
flows and it is particularly useful for the computation of transient aerodynamics,
free surface and multiphase flows, and problems involving complex geometries.

4.1 Octree lattice structure

The lattice structure featured by XFlow is the D3Q27 organized as an octree

structure, which is the maximum number of discrete velocities for a 3D scheme.
The octree structure offers the possibility to use a non-uniform lattice structure,
and therefore have different spatial scales at different locations of the fluid domain.
The pre-processor generates the initial octree lattice structure based on the input
geometries, the user-specified lattice resolution for each geometry, as well as the far-
field resolution. User-defined regions (sphere, box, cylinder, etc.) can also be created
to refine arbitrary regions at the specified lattice resolution. The different spatial
scales employed are hierarchically arranged. Each level solves spatial and temporal
scales twice smaller than the previous level, thus forming the aforementioned octree
structure (Figure 4.1). This is particularly efficient as the ratio dx/dt is maintained
18 XFlow: an Innovative LBM Approach

in the entire fluid domain, thus ensuring a constant CFL condition and speed of
sound at any location of the fluid domain. Finite volume and elements normally use
a global time step which is inefficient for the coarser mesh cells, whereas the local
time step approach in XFlow allows to have always an adapted time step for every
lattice size of your fluid domain.

Figure 4.1: Octree lattice structure with different lattice resolution.

This initial lattice structure can be modified during the simulations by the XFlow
solver based on several criteria. First, if the computational domain changes due to
presence of moving geometries, the lattice can dynamically be refined to follow the
new position of the geometry every time step. Other adaptive refinement criteria
to adapt the flow physics are also available. A refinement algorithm based on the
level of vorticity is effective to dynamically refine the wake region (Figure 4.2),
characterized by high vorticity. Also, the free surface and multiphase flows can be
refined dynamically at the free surface or interface, and therefore have an efficient
tracking of the interface including droplets and splashing.

Figure 4.2: Example of lattice structure with adaptive wake refinement.

4.2 Central-moment LBM 19

4.2 Central-moment LBM

The collision operator in XFlow is based on a multiple relaxation time scheme.

However, as opposed to standard MRT employed by most LBM implementations,
the scattering operator is implemented in central momentum space. The central
moments are defined as
N
X
µ̃xk yl z m = fi (eix − ux )k (eiy − uy )l (eiz − uz )m . (4.1)
i

where ux , uy , and uz are the components of the macroscopic velocity.

Similarly to the MRT scheme, the central moments are computed and relaxed
independently and the inverse transformation is then performed to recover the
probability distribution functions. By shifting the discrete particle velocities with
the local macroscopic velocity, the Galilean invariance and the numerical stability
for a given velocity set are greatly improved ([21, 22]). The Mach number limitation
is also enhanced, and global Mach number up to 0.6 can be handled while the local
Mach number is limited to 1.0. This collision operator provides a low numerical
diffusion even for very low viscosity.

4.3 Wall-adapting local eddy viscosity

The approach used for turbulence modeling is the Large Eddy Simulation (LES).
This scheme introduces an additional viscosity, called turbulent eddy viscosity νt , in
order to model the subgrid turbulence. The LES model used in XFlow is the Wall-
Adapting Local Eddy (WALE) viscosity model, that provides a consistent local
eddy-viscosity and near wall behavior ([23]).
The actual implementation is formulated as follows

(Gdαβ Gdαβ )3/2

νt = ∆2f (4.2)
(Sαβ Sαβ )5/2 + (Gdαβ Gdαβ )5/4
gαβ + gβα
Sαβ = (4.3)
2
1 2 1
Gdαβ = 2
(gαβ + gβα 2
) − δαβ gγγ (4.4)
2 3
∂uα
gαβ = (4.5)
∂xβ

where ∆f = Cw ∆x is the filter scale, S is the strain rate tensor of the resolved scales
and the constant Cw is typically 0.325.
The strain rate tensor (gαβ ) is locally available with the lattice-Boltzmann method
as the second-order moment, which makes extremely efficient the implementation
20 XFlow: an Innovative LBM Approach

of state-of-the-art LES models. For finite volume and elements methods, the strain
rate tensor requires information from the neighbor nodes to be computed, since
being a spatial derivative. Besides the additional operations it requires, the access
in memory to get the neighbor values makes the process extremely inefficient in
comparison to LBM.
Furthermore, the Cartesian lattice structure is well suited for the LES turbulence
model. Because of isotropic character of the turbulence out of the boundary layer,
the LES turbulence model require cells with proportioned aspect ratio. The lattice
completely fulfills this requirement.

4.4 Advanced wall treatment

In addition to the LES turbulence modeling, XFlow uses wall function in order to
model the near-wall region and boundary layer, and therefore employs the so-called
Wall-Modeled LES approach (WMLES). The isotropy of the lattice structure would
imply an unreasonably high number of elements to resolve the boundary layer. This
issue is addressed by the use of a generalized law of the wall.
The boudary layer is modeled by a generalized law of wall given by Shih et al. [24]
based on a previous work of [25]. The approach takes into account the effect of
adverse and favorable pressure gradients
U U1 + U2 uτ U1 up U2
= = + (4.6)
uc uc uc uτ uc up
   
τw uτ + uτ dpw /dx up + up
= f1 y + f2 y (4.7)
ρu2τ uc uc |dpw /dx| uc uc
uc y
y+ = (4.8)
ν
uc = uτ + up (4.9)
p
uτ = |τw | /ρ (4.10)
ν dpw 1/3
 
up = . (4.11)
ρ dx

y is the normal distance from the wall, x is the local flow direction tangentially
to the wall, uτ is the skin friction velocity, τw is the turbulent wall shear stress,
dpw /dx is the wall pressure gradient, up is a characteristic velocity of the adverse
wall pressure gradient and U is the mean velocity at a given distance from the wall.
The interpolating functions f1 and f2 are depicted in Figure 4.3.
The velocity field of the boundary layer is obtained through the y + which depends
on the distance between the first lattice from the geometry wall y and the velocity
of this first lattice uc .
XFlow projects the set of discrete velocities on the geometry tessellation to obtain
4.5 Advanced dynamic geometries 21

25 45
f1 (y + uτ/uc ) f2 (y + up /uc )
40
20 35
30
15
25
f1

f2
20
10
15

5 10
5
0 1 0
10 0
10 10 2
100 101 102
y + uτ/uc y + up /uc

Figure 4.3: Unified laws of the wall

the wall distance and thus discretize the geometry as depicted Figure 4.4. This
implies a high level of details for the geometry discretization as one lattice node can
detect up to 27 geometry projections. These projected velocities are also used to
calculate the curvature of the surface that is taken into account for the wall function.

Figure 4.4: Wall distance and curvature calculation.

4.5 Advanced dynamic geometries

The flexibility of the octree lattice structure and the advanced near-wall treatment
proposed by XFlow allows to address one of the most important difficulties faced by
the traditional CFD: the fluid-structure interaction. XFlow proposes two different
options to handle dynamic geometries: the enforced behavior, and the rigid body
dynamics behavior. The enforced behavior moves a geometry based on an input
position and orientation law, enforcing the motion of the object. The rigid body
dynamics behavior couples the fluid equations resolved by the XFlow LBM solver to
a Next Limit Technologies proprietary multi-body solver, that handles rigid bodies
allowing up to 6 degrees of freedom.
For both dynamic behaviors, the lattice structure is updated every time step to
mark the lattice nodes that belong to the fluid region, and those that belong to solid
region as depicted Figure 4.5. The discrete velocities are also projected every time
22 XFlow: an Innovative LBM Approach

step in order to compute the new distance to the wall required for the wall function.

Figure 4.5: Lattice nodes identification in moving geometry process.

4.6 Scalability

The lattice-Boltzmann method has the advantaged to have an easy implementation,

since the main algorithm function consists of a loop to calculate the streaming,
the collision, the macroscopic variables, and the boundary conditions which is an
intrinsic parallel structure allowing high computing performance.
XFlow is fully parallelized with near-linear scalability for both multi-core processors
on shared memory (SMP) and distributed memory (DMP) environments. It
therefore integrates into workstations as well as HPC environments. The scalability
curves are shown

Figure 4.6: Serail and distributed XFlow performance.

 References

[1] U. Frisch, B. Hasslacher, and Y. Pomeau. Lattice-gas automata for the navier-
stokes equation. Physical review letters, 56(14):1505–1508, 1986.

[2] G. R. McNamara and G. Zanetti. Use of the Boltzmann equation to simulate

lattice-gas automata. Physical Review Letters, 61:2332–2335, November 1988.

[3] S. Chen and G.D. Doolen. Lattice boltzmann method for fluid flows. Annual
review of fluid mechanics, 30(1):329–364, 1998.

[4] Sauro Succi. The Lattice-Boltzmann Equation. Oxford university press, Oxford,
2001.

[5] M Sbragaglia, Roberto Benzi, Luca Biferale, Sauro Succi, K Sugiyama, and
Federico Toschi. Generalized lattice boltzmann method with multirange
pseudopotential. Physical Review E, 75(2):026702, 2007.

[6] Ai-Guo Xu, Guang-Cai Zhang, Yan-Biao Gan, Feng Chen, and Xi-Jun Yu.
Lattice boltzmann modeling and simulation of compressible flows. Frontiers of
Physics, 7(5):582–600, 2012.

[7] Zhaoli Guo and Chang Shu. Lattice Boltzmann method and its applications in
engineering. World Scientific, 2013.

[8] Y. H. Qian, D. D’Humières, and P. Lallemand. Lattice BGK models for Navier-
Stokes equation. EPL (Europhysics Letters), 17:479, February 1992.

[9] F. J. Higuera and J. Jiménez. Boltzmann approach to lattice gas simulations.

EPL (Europhysics Letters), 9:663, August 1989.

[10] Pietro Asinari, Taku Ohwada, Eliodoro Chiavazzo, and Antonio F Di Rienzo.
Link-wise artificial compressibility method. Journal of Computational Physics,
231(15):5109–5143, 2012.
24 XFlow: an Innovative LBM Approach

[11] Zhaoli Guo, Chuguang Zheng, and TS Zhao. A lattice bgk scheme with general
propagation. Journal of scientific computing, 16(4):569–585, 2001.

[12] Santosh Ansumali and Iliya V Karlin. Entropy function approach to the lattice
boltzmann method. Journal of statistical physics, 107(1-2):291–308, 2002.

[13] Dominique d’Humieres. Generalized lattice-boltzmann equations. Rarefied gas

dynamics- Theory and simulations, pages 450–458, 1994.

[14] Pierre Lallemand and Li-Shi Luo. Theory of the lattice boltzmann method:
Dispersion, dissipation, isotropy, galilean invariance, and stability. Physical
Review E, 61(6):6546, 2000.

[15] Dominique d’Humières. Multiple–relaxation–time lattice boltzmann models in

three dimensions. Philosophical Transactions of the Royal Society of London.
Series A: Mathematical, Physical and Engineering Sciences, 360(1792):437–451,
2002.

[16] X. Shan and H. Chen. A general multiple-relaxation-time boltzmann collision

model. International Journal of Modern Physics C, 18(4):635–643, 2007.

[17] Li Chen, Yang Yu, and Guoxiang Hou. Sharp-interface immersed boundary
lattice boltzmann method with reduced spurious-pressure oscillations for
moving boundaries. Physical Review E, 87(5):053306, 2013.

[18] XD Niu, C Shu, YT Chew, and Y Peng. A momentum exchange-based

immersed boundary-lattice boltzmann method for simulating incompressible
viscous flows. Physics Letters A, 354(3):173–182, 2006.

[19] Qisu Zou and Xiaoyi He. On pressure and velocity boundary conditions for the
lattice boltzmann bgk model. Physics of Fluids (1994-present), 9(6):1591–1598,
1997.

[20] Martin B Schlaffer et al. Non-reflecting boundary conditions for the lattice
Boltzmann method. PhD thesis, Universitätsbibliothek der TU München, 2013.

[21] M Geier, A Greiner, and JG Korvink. A factorized central moment lattice

boltzmann method. The European Physical Journal Special Topics, 171(1):55–
61, 2009.

[22] K.N. Premnath and S. Banerjee. On the three-dimensional central moment

lattice boltzmann method. Journal of Statistical Physics, pages 1–48, 2011.

[23] F. Ducros, F. Nicoud, and T. Poinsot. Wall-adapting local eddy-viscosity

models for simulations in complex geometries. In Proceedings of 6th ICFD
Conference on Numerical Methods for Fluid Dynamics, pages 293–299, 1998.
4.6 Scalability 25

[24] T.H. Shih, L.A. Povinelli, N.S. Liu, M.G. Potapczuk, and JL Lumley. A
generalized wall function. TM 113112 NASA Technical Report, July 1999.

[25] Hendrik Tennekes and John Leask Lumley. A first course in turbulence. MIT
press, 1972.
 or registered trademarks of Dassault Systèmes, a French “société européenne” (Versailles Commercial Register # B 322 306 440), or its subsidiaries in the United States and/or other countries. All other trademarks are owned by their respective owners. Use of any Dassault
©2020 Dassault Systèmes. All rights reserved. 3DEXPERIENCE, the Compass icon, the 3DS logo, CATIA, BIOVIA, GEOVIA, SOLIDWORKS, 3DVIA, ENOVIA, EXALEAD, NETVIBES, MEDIDATA, CENTRIC PLM, 3DEXCITE, SIMULIA, DELMIA, and IFWE are commercial trademarks

Systèmes or its subsidiaries trademarks is subject to their express written approval.

Our 3DEXPERIENCE® platform powers our brand applications, serving 11 industries, and provides
a rich portfolio of industry solution experiences.
Dassault Systèmes, the 3DEXPERIENCE Company, is a catalyst for human progress. We provide business and people with collaborative virtual environments
to imagine sustainable innovations. By creating ‘virtual experience twins’ of the real world with our 3DEXPERIENCE platform and applications, our
customers push the boundaries of innovation, learning and production.

Dassault Systèmes’ 20,000 employees are bringing value to more than 270,000 customers of all sizes, in all industries, in more than 140 countries. For
more information, visit www.3ds.com.

Americas Europe/Middle East/Africa Asia-Pacific

Dassault Systèmes Dassault Systèmes Dassault Systèmes K.K.
175 Wyman Street 10, rue Marcel Dassault ThinkPark Tower
Waltham, Massachusetts CS 40501 2-1-1 Osaki, Shinagawa-ku,
02451-1223 78946 Vélizy-Villacoublay Cedex Tokyo 141-6020
USA France Japan