2010.10.05.

Different behavior
Physical processes lead to a temporal equilibrium in many cases.

analitical solutions Explicit Euler method:

Numerical approximations of derivatives and integralls

Gergely Kristf 27-th September 2010
x u

Implicit Euler method:

Euler method
uj j x From the Taylor polynomial we can express a differencing scheme of first order accuracy: uj+1 j+1 x Taylor polynomial of the solution from point j to point j+1: uj-1 j-1

CDS
uj j uj+1 j+1 x

u j +1 = u j + u' j x + o( x )
This is an integration scheme of first order accuracy. When the differencial equation is given in the explicit form:

u j +1 = u j + u' j x + u" j

x 2
2

+ o( x 2 )

u' j = f u j , x j

) ( )

u j 1 = u j + u' j ( x ) + u" j
u j +1 u j 1 2 x

x 2
2

+ o( x 2 )

u' j =

u j +1 u j

+ o( 1 )

we can integral step by step, by assuming:

Note that, the error term is one degree of magnitude higher.

u j +1 u j + f u j , x j x

u' j =

+ o( x )

Bacward Euler method

Another first order scheme: uj j x uj+1 j+1 x

An implicit differencing scheme with second order accuracy

uj-1 j-1 uj j uj+1 j+1 x

u j = u j +1 + u' j +1 ( x ) + o( x )
from the backward Euler scheme we get:

u' j +1 =

u j +1 u j

+ o( 1 )

u j = u j +1 + u' j +1 ( x ) + u" j +1

x 2
2

+ o( x 2 )

Now, we assume the differential equation in the form:

u j 1 = u j +1 + u' j +1 ( 2x ) + u" j +1 2x 2 + o( x 2 )

When F is evaluated in j+1, we may end up with a more complicated expression for uj+1. This kind of discretization is called implicit.

F u' j +1 ,u j +1 , x j +1 = 0

uj

u j 1 4

u j +1 u j F x ,u j +1 , x j +1 0

3 x u j +1 + u' j +1 + o( x 2 ) 4 2

3 1 u j +1 2u j + u j 1 2 u' j +1 = 2 + o( x ) x

2010.10.05.

Adams-Basforth scheme
uj-1 j-1 j uj+1 j+1 x

Approximation of the divergence operator

From the volume integral of the divergence operator we can obtain the cell average of the divergence term. The Gauss-Ostrogradskij theorem for a vector quantity u:

u j +1 = u j + u' j x + u" j

x 2
2

+ o( x 2 )
+ ...

u dV = u d A
x
2
V A

u' j 1 = u' j +u" j ( x ) + o( x )

For simplicity, we denote components of u vector by ui. The cell-average of the divergence operator is now:

3 1 u j +1 = u j + u' j x u' j 1 x + o( x 2 ) 2 2
An explicit integrating scheme with second order accuracy It is often used for integrating the Navier-Stoket equations.

~ ui =

u
k Ak

dA

VP

in which Ak are the faces of the cell. The surface integral for each face is a scalar product: 3 u dA = ui dAi in which ui is one component of u interpolated to the cell surface. = i 1 Ak

A 2 step 2nd order explicit Runge-Kutta type scheme

uj-1 j-1 1st step: uj j uj+1 j+1 x

Gradient
A direct consequence of the Gauss-Ostrogradskij theorem:

dV = d A
V A

Use the Euler method for geting into point j:

u j = u j 1 + u' j 1 x + o( x )
Evaluate the derivative in point j:

The i-th component of the approximate gradient can be evaluated according to the following expression:

d = f (u j + o( x ), x j ) = f (u j , x j ) + o( x ) = u' j +o( x )
2nd step: Use CDS scheme around point j:

~ =
i

dA
k Ak

VP

Ai is the i-th component of the surface vector in Descartes system.

u j +1 = u j 1 + d 2 x + o( x 2 ) = u j 1 + u' j 2 x + o( x 2 )

Spatial derivatives in finite volume methods

The generic transport equation in integral form:

The approximate Laplacian

=
When calculating the discrete approximation of the operator the gradient must be interpolated onto the face centroids. This is denoted by < > in the following formula:

r r + ( v ) = S A + ( ) + S v t
In which is the mass concentration of a conserved quantity (eg. in kg/kg). Spatial derivatives are always in div(...), grad(...) or div(grad(...)) forms. We only need to look for the discrete approximations of these operators, which is done - in the case of finite volume method - on the basis of surface and volume integrals along with some spatial interpolations. The numerical mesh around the cell having its center in point P: Cell centroid. Here we store P. Anything can be interpolated from cells to surfaces...

=
i

For most field variables - excepting for the pressure field the face normal component of the gradient vector can be calculated on a more simple way: from values stored in the centers of the adjacent cells. In this case the discrete form of the Laplacian operator can be calculated as a linear combination of P and the neighboring values:

P Face centroids. Defined by surface vectors.

= AP P + Al l
In which AP are constant values, depending only on the mesh parameters.