FluxNumerical Methods
FluxNumerical Methods
FluxNumerical Methods
Flux Basics
The finite-volume formulation of the conservation equations resulted in the equation
dQ = PF dt
where F was the flux of the flow across the control surface resulting from the approximation of the surface integral. For a finite-volume cell, the flux was expressed as
F = Ff
Where
f =1
nf
= [ ( v g ) Q D] ( n dS ) Ff f f
It was assumed that the flux was uniform over the cell face.
F=
where again,
nf = 1 f =1
( n dS ) f
= [ ( v g ) Q D] ( n dS ) Ff f f
Ff
Face f ( f is an index for the face )
= [ ( v g ) Q D] ( n dS ) Ff f f
The normal area vector ( n dS ) f is usually easily defined for a quadrilateral or triangular cell face. The focus of the rest of this discussion is on numerical methods for computing
[ ( v g ) Q D] f
at a cell face.
g is a known velocity for the cell face. We first will assume that
F f = F f ( QL , QR )
One can define
FL = F f ( QL )
FR = F f ( QR )
QL
( n dS ) f
Ff
QR
FL = FR
= 1F +F Ff L R 1
The central-difference method works okay for elliptic components of the flux because there is no preferred direction for the propagation of information. A simple central difference is often unstable, especially in the presence of strong gradients. One solution is to add some second-order and fourth-order dissipation (artificial viscosity) to the flux.
) ) = 1F + F + D ( 1 + D ( 1 Ff L R 1
Methods for computing D(2) and D(4) vary, but generally use second and fourth-order differences with switches to handle variations in Q.
Upwind Methods
We expressed the non-convective portion of the flux of the Navier-Stokes equation as
D NS = D INS + DV NS
F f = ( v g ) QNS D INS DV NS
or
]
I
F f = F fI + F fV
We will now focus on computing the inviscid flux F f using upwind methods. The focus will be on the use of Roes Upwind Flux-Difference Splitting Method.
I = 1F I + F I 1 F Ff f L R 1 1
F = F + + F = ( + ) r w F m m m
+ 1 m =1 1
F = (m ) rm wm
m =1
The wm are the Riemann invariants and represent the strength of the wave, p w1= 1 c w1 = n1w n1u
w1= n1u n1v p w1 = + n1u + n1v + n1w c p w1= n1u + n1v + n1w c
10
= R L p = p R p L u = u R u L v = v R v L w = wR wL
Flow properties at the face are computed using Roe-averaging
1= R L
/ / 11 L + 11 R u L R u u= / / 11+ 11 L R
11
Higher-Order Projection
The choice of values of QL and QR have several option: 1) Use the values of the finite-volume cells to the left and right of the face. This is a zero-order evaluation and will result in a spatially first-order flux. 2) Use an extrapolation of neighboring finite-volume cells to form a firstorder evaluation of Q at the face. This will result in a spatially secondorder flux.
1 QL = Qi + ( Qi Qi 1 ) 1
1 QR = Qi +1 ( Qi + 1 Qi +1 ) 1
QL QR Qi-1 Qi
Ff Q i+1
Qi+2
12
Variation Limiting
The simple extrapolation formulas assume a smooth variation of Q; however, discontinuities in Q are possible (i.e. shocks). Need some mechanism to sense such discontinuities and limit the variation of Q in these extrapolation formulas. Modify the extrapolations by introducing a limiter ,
1 QL = Qi + ( Qi Qi 1 ) 1 1 QR = Qi +1 ( Qi + 1 Qi +1 ) 1
This gets into the topic of TVD (Total Variational Diminishing) flux limiting methods, which we will not get into here. The essential role of the limiter is to make 0 in the presence of large variations, which make the flux spatially first-order. 13
Examples of Limiters
The possible functions (and theory) for limiters is varied. A couple examples include: Superbee: Chakravarthy:
Where r is some ratio of the flow properties and indicates the amount of variation in the solution. An example is
Qi +1 r= Qi
The is a compression parameter 1 2, where a value toward 1 makes the limiter more dissipative.
14
F fI = F + + F
van Leers flux-vector splitting has the general form of
mass F = f 1 q 1 vn
)v c 1 [( 1 n 1 ] +
1
[(
1 1 1
)]
mass
) c ( M n 1 = 1
vn = v n
vn Mn = c 15
16
Next