Lattice Boltzmann Methods for Fluid Dynamics

Department of Mathematics Yale University

Steven Orszag

In collaboration with Hudong Chen, Isaac Goldhirsch, and Rick Shock

Transient flow around a car

LBGK vs. CONVENTIONAL CFD

Real Fluid
Free molecules in continous space

Continuum Kinetic Theory

Microscopic particles (Boltzmann Equation)

Conventional CFD Methods ___________________________ Construction of fluid equations

Navier-Stokes equations (PDE)

Lattice Based Method _________________________________ Discrete formulation of kinetic theory

Lattice Boltzmann equations

No further approximation
The equations are already in discrete form

Discrete approximation of PDE

Finite difference, finite element, etc

Numerical integration
Solve on lattices and apply kinetic based BC

Numerical integration
Solve the equations on a given mesh and apply PDE boundary conditions

Simple conversion to fluid variables

These are theoretically shown to obey the required fluid equations

Results
Fluid dynamic quantities at discrete points in space and time

Lattice Boltzmann (or BGK) Methods

Particles only have a finite number of discrete velocity values

! ! v ! {ci ; i = 0,1, " , b} " " " f ( x , v , t ) ! ni ( x , t ); i = 0,1,! , b ! Number density for particles with velocity c i
The choice is not arbitrary!
Satisfy foundational symmetry requirements (up to required orders) Avoid spurious invariants

4 0 5

b in 3D ~ 20 - 30

Lattice BGK method

LBGK:
! ! ! ! ni ( x + ci !t , t + !t ) = ni ( x , t ) + Ci ( x , t )
Coupled (via

Ci ) algebraic difference equations

D3Q19 lattice ! Fluid quantities obtained via averaging over c and space-time: i ! = " i ni
D2Q6 lattice

BGK form:

1 eq Ci = " (ni " ni ) !

! ! ! u = " i ci ni ! ! 2 1 ! T = D " i (ci # u ) ni

x
D2Q4

Remarks on LBGK
Lattice

BGK yields the Navier-Stokes equations Chapman-Enskog asymptotic expansion in powers of Knudsen number /L or /T << 1

Easy to compute time dependent flows Relaxation time defines viscosity No need to compute pressure explicitly Boundary conditions are fully realizable Stability is ensured Parallel performance with arbitrary geometry

Brief Comparison of LBGK and Conventional CFD

Conventional CFD
1. Nonlinear dynamic advection 2. Non-local limited parallel performance 3. Issues with boundary conditions (BC) 4. Geometry setup slow 5. 3D time dependent ows costly to simulate 6. Complex physics (like multi-phase ows) require complex physical models

LBGK
1. Linear advection 2. Local and fully parallel 3. BC are fully realizable for arbitrary geometry 4. Geometry setup fast 5. Time dependent ows straightforward especially important in 3D 6. Complex physics (like multiphase ows) involve simple physical models

2D Cylinder

2D Cylinder
Re = 100

2D Cylinder
Friction Drag

How do you derive N-S from LBGK?

Chapman-Enskog (moment expansion) procedure in powers of Knudsen number /L Navier-Stokes equations are independent of orientation of coordinate system BUT lattice BGK is highly anisotropic REMARKABLE FACT isotropy of velocity moments only up to a xed nite order are required

Isothermal Navier-Stokes equations at low Mach numbers

Density/momentum b
! =1

" ( x, t ) # $ f! ( x, t ), "u( x, t ) # $ c! f! ( x, t )
! =1

Momentum ux tensor
Pij ( x, t ) " # c! ,i c! , j f! ( x, t )
b

Energy ux tensor
Qijk ( x, t ) " # c! ,ic! , jc! ,k f ! ( x,t )
! =1
b

! =1

Navier-Stokes requires isotropy of velocity moments only up to 4th order

High-order models
Non-isothermal low Mach number NavierStokes equations requires velocity moment isotropy up to 6th order Other physically relevant models reuqire even higher-order velocity moment isotropy, further restricting the discrete velocity set used in lattice BGK For example, non-isothermal ow with Burnett corrections requires 8th order isotropy

Relation between rotational symmetry and order of moment isotropy in 2D

b velocities C = {c = (cos ( 2"! ), sin ( 2"! ); ! = 0,, b # 1} b b is invariant under rotations by multiples of 2/b Isotropy of the nth order basis moment tensor b
!

requires that # (c! " v) = A where A is a constant and ! v = (cos ! , sin ! ) is any unit vector b $1 ( n) n 2"! $ # ) be This requires that hb (# ) % & cos ( b b #1 ! =0 independent of , which holds if $ ei (2 j #n ) = 0 " =0 ie (2j-n)/b is not a nonzero integer for j=0,,n CONCLUSION: Isotropy for so n ! b " 2 hexagonal lattice gives 4th order isotropy, etc.
n
n
2 !" b

M ( n ) " # c! c! !c! ! !! ! ! b !

3D Moment Isotropy
nth order basis moment tensor
M
( n)

= # w! c! " c! " ! " c! !!!!!!! ! =1

n

Isotropy requires
( n ! 1)c 2 n !2) M i!i = M i(!i ! D + n ! 2 !!2 n n
( n)

( n ! 3) c 2 n !2) M i!i jj = M i(!i jj ! D + n ! 2 !!4 n!2 n

( n)

M i!i jjjj

( n)

and so on

n!4

( n ! 5) c 2 ( n !2) = M i!i jjjj D + n ! 2 !!6 n

Generation of Nth order isotropic lattices

Using these relations an Nth order isotropic lattice can be constructed by a union of (N-2)nd order isotropic lattices and its rotated realizations Example: 6th order set with 59 velocities 1 (0,0,0) 12 {(1, 1, 0), ( 1, 0, 1), (0, 1, 1)} 6 {(1, 0, 0), (0, 1, 0), (0, 0, 1)} 8 {(1, 1, 1)} 6 {(2, 0, 0)} 12 {(2, 2, 0), (2, 0, 2), (0, 2, 2)} 8 {(2, 2, 2)} 6 {(4, 0, 0), (0, 4, 0), (0, 0, 4)}

Boltzmann- Turbulence Modeling

Turbulence modeled via a modied relaxation time

"f 1 = ! ( f ! f eq ) "t #
1 1 1 1 1 = + + + + ... ! ! turb ! shear ! buoyancy ! swirl

Advantages of Boltzmann- Method - I

Realizability of the turbulence model Boltzmann- has guaranteed realizability
Requires only turb > 0 Stable numerical results Positive eddy viscosity

Navier-Stokes-based turbulence models can have signicant difculties with realizability

Divergent turbulence quantities Negative eddy viscosities,

Advantages of Boltzmann- Method - II

With the BGK model in terms of space, time, and velocity as independent coordinates, simple approximations (like models) may be extraordinarily complex in uid (velocity/pressure) variables Fluid velocity/pressure are projections of the BGK variables onto a lower-dimensional space In contrast to higher-order ChapmanEnskog projections, the BCs on BGK are well dened and easy to implement

An Opel in a Wind Tunnel

Centerline Velocity at 140 km/h

Dodge RAM Exterior/underhood/ under-carriage ow centerline velocity

Dodge RAM Acoustic Impact of Headlight/Hood Design on A-Pillar and Door Seals: ! = " 2 # S 2
2

3D Streamlines of Flow Past a Large Truck

Drag Development
Drag increase due to base of trailer Deflector Suction (-110 cnts) Secondary stagnation region (110 cnts) Steep drag increase in stagnation region (90 cnts)

Drag increase due to cab rear wheels (60 cnts)

Drag increase due to trailer friction (20 cnts)

Drag increase from cab-trailer gap (40 cnts)

Conclusions
Lattice BGK allows straightforward mix of complex uids, complex physics, and complex geometries Appropriate lattice structures can be derived to assure accurate and efcient ow computations, even with turbulence and other complex physics included