Date post: | 18-Dec-2015 |
Category: |
Documents |
Upload: | anthony-walters |
View: | 215 times |
Download: | 2 times |
2011 DSMC Workshop
Improvements to the Discrete Velocity Method for the Boltzmann Equation
Peter Clarke
D. Hegermiller, A.B. Morris , P.T. Bauman, P. L. Varghese, D. B. Goldstein
University of Texas at AustinDepartment of Aerospace Engineering
DSMC Workshop September 2011
Funding: Some of this material is based upon work supported by the DOE [National Nuclear Security Administration] under Award Number [DE-FC52-08NA28615] and NASA’s NSTRF Fellowship program
2011 DSMC Workshop Outline
• Motivation
• The Discrete Velocity Method
• Previous work
• Variance reduction using an interpolation scheme
• Non-uniform grids in velocity space
• Application of VHS and VSS collision models
• Future Work• Inclusion of internal energy in DVM
2
2011 DSMC Workshop Motivation
• Discrete velocity methods are comparable to DSMC
• But discrete velocity methods have several traditional problems
• High Mach number flows and other flows that require large velocity bounds.
• The inclusion of physics in the model such as varying types of molecular potentials, multi-species flow, internal
energy, and chemical reactions are often neglected in preliminary
DVM investigations.
• We wish to solve the first problem with the eventual application of
adaptive velocity grids.
• The first step towards this goal is the implementation of non uniform grids in velocity space.
3
2011 DSMC Workshop DVM Formulation
ˆˆˆ])ˆ(ˆ)ˆ(ˆ)'ˆ(ˆ)'ˆ(ˆ[
1ˆ
ˆˆ
ˆˆ
dVgKnxt T
depletingngreplenishi
depl Trepl T dVgdVgxt
ˆˆ ˆˆ)]ˆ(ˆ)ˆ(ˆ[ˆˆ)]'ˆ(ˆ)'ˆ(ˆ[ˆ
ˆˆ
ˆˆ
The collision integral is split into replenishing and depleting parts:
We begin with the scaled Boltzmann equation:
To solve the Boltzmann equation using DVM, we must discretize the integro-differential equation.
L
xx ˆ
r ˆ
m
Tk rbr
22
3ˆ
r
rn
Ltt r
ˆr
ˆ
r
gg
ˆ
Scaling Factors:
1, pseudo-maxwell 1, pseudo-maxwell
4
2011 DSMC Workshop DVM Formulation
collisionsconvectiont
ˆˆ'ˆ
ijk ijk
tt 33 )ˆ(ˆ)ˆ(ˆˆ)'ˆ(ˆ)'ˆ(ˆˆˆ'ˆ
We separate the convection and collision parts of the equation
4th order convection: 22111122' nnnonnn
We then approximate the collision integral with finite summations:
β
5Bobylev, A.V., 1976, Soviet Phys. Dokl., 20, 822-824.Krook, M., and Wu, T.T., 1977, Phys. Fluid, 20, 1589-1595
2011 DSMC Workshop DVM Formulation
DSMC – “Fixed mass, variable velocity particles.”
ηi
ηj
ϕ
ηi
ηj
DVM – “Fixed velocity, variable mass quasi-particles.”
ϕ
6
2011 DSMC Workshop Variance Reduction
ˆˆˆ EQ
Decompose ϕ into an equilibrium part and a deviation from equilibrium part
As has been previously presented by A. Morris we use a stochastic discrete velocity model:
Baker, L.L. and N.G. Hadjiconstantinou, "Variance Reduction for Monte Carlo Solutions of the Boltzmann Equation," Physics of Fluids, 17, 2005Morris, A.B., “Variance Reduction for a Discrete Velocity Gas” 2011
7
δ
2011 DSMC Workshop
ˆ
ˆ
ˆ
ˆˆ)]ˆ(ˆ)ˆ(ˆ)'ˆ(ˆ)'ˆ(ˆ[
ˆˆ)]ˆ(ˆ)ˆ(ˆ)'ˆ(ˆ)'ˆ(ˆ[2
ˆˆ)]ˆ(ˆ)ˆ(ˆ)'ˆ(ˆ)'ˆ(ˆ[
dVdg
dVdg
dVdgI
EQEQ
EQEQEQEQ
Variance Reduction
As has been previously presented by A. Morris we use a stochastic discrete velocity model:
Baker, L.L. and N.G. Hadjiconstantinou, "Variance Reduction for Monte Carlo Solutions of the Boltzmann Equation," Physics of Fluids, 17, 2005Morris, A.B., “Variance Reduction for a Discrete Velocity Gas” 2011
0
8
δ
2011 DSMC Workshop Variance Reduction
Calculate depletion mass:
Select random collision partners, either two from the deviation distribution or one from the deviation and one from the equilibrium distribution.
1; 0sgn( )
1; 0
xx
x
31,
ˆˆ)ˆsgn(ˆˆ
c
EQ
N
nnt
32,2
ˆˆ)ˆsgn()ˆsgn(ˆˆ
cN
nnt
or
β
9
spacingmesh
densitynumbern
collisionsofnumberNc
ˆ
2011 DSMC Workshop Interpolation
06/)(
)(
)(
)(
01
222
222
cbacbaf
fcf
fbf
faf
cbaf
e
extiz
extiy
extix
o
Begin with conservation equations:
MassMomentumEnergy
The system of equations is solved:
iz
ez
ex
ey
iy
xyz
ix
o
ab
c
Δϕ
Varghese, P.L., “Arbitrary Post-Collision Velocities in a Discrete Velocity Scheme for the Boltzmann Equation.” 2007 10
2011 DSMC Workshop Code Development
DVM has been implemented using modern software engineering principles that enhance maintainability and ease of testing to allow more thorough verification of the software and, thus, increases confidence that the implementation is correct. These practices include:
• Object-oriented code style to enhance encapsulation and minimize code duplication.
• Source code revision control using svn.
• Build system (Autotools) for portability between computing systems (code currently tested on Linux and Mac OS X environments)
• Build system also enables easy addition of unit and regresssion tests. Current suite is at 42 tests.
• Full documentation of code and algorithms
11
2011 DSMC Workshop Variance Reduction with Interpolation
We combine the variance reduction technique with the interpolation scheme that has been developed:
4th, 6th, and 8th moments of the relaxation of the BKW distribution
- Interpolation
- analytic
12
- No interpolation
No Interpolation
Interpolation
2011 DSMC Workshop Variance Reduction with Interpolation
- No interpolation
- Interpolation
Mach 2 Shock density profile:
13
2011 DSMC Workshop Non-uniform grids in velocity space
Additions to the Discrete Velocity Method:
Due to the interpolation scheme we can relax the requirement that β be a constant number
ijk ijk
kjikji tt )ˆ(ˆ)ˆ(ˆ)'ˆ(ˆ)'ˆ(ˆ'
βi
βj
βk
14
2011 DSMC Workshop Non-uniform grids in velocity space
3D homogeneous relaxation with variable grid:
The optimal configuration for the velocity grid is an area of active research15
2011 DSMC Workshop VHS and VSS
/1
1cos2d
b
deplrepldVgdVg
xt
ˆ)1(
ˆ)1( ˆ)]ˆ(ˆ)ˆ(ˆ[ˆ)]'ˆ(ˆ)'ˆ(ˆ[
ˆ
ˆˆ
ˆˆ
VSS is similar to VHS except the scattering is no longer isotropic .
When picking post-collision velocities, sample from the scattering distribution.
VHS VSS
refrefT g
g
Collision probability depends on relative speed.
The amount depleted during a collision is now proportional to collision probability.
16
2011 DSMC Workshop VHS Example
DVM
DSMC
A Mach 2 shock with VHS:
17
2011 DSMC Workshop
ηk
ηj
ηi
Erot Evib
Future Work
A major addition to the Discrete Velocity Method that allows for more accurate physics is the inclusion of internal energy
We assign a single internal energy to each location in velocity space.
Future work will allow a distribution of energies at every velocity location
Internal Energy:
18
2011 DSMC Workshop Future Work
An exchange in energy between translation and internal energy is calculated using a Landau-Teller-like equation.
The exchange changes the magnitude of the post-collision relative velocity vector as well as adding or subtracting from the internal energy distributions.
Interpolation allows for any post collision relative velocity vector length.
19
2011 DSMC Workshop Summary
We showed: • Comparison between DSMC and DVM
• Combination of Variance Reduction with Interpolation
• Non-uniform velocity grids
• Application of VHS and VSS collision models
Future Work:• Full implementation of internal energy including distributions of energy at every point in velocity space.
• Adaptable velocity grids.
20