# Diffusion term

(Difference between revisions)
 Revision as of 01:28, 15 September 2005 (view source)Zxaar (Talk | contribs)← Older edit Latest revision as of 07:32, 22 January 2010 (view source)Dani1978 (Talk | contribs) (difference vector incorrectly defined (compare with picture!)) (19 intermediate revisions not shown) Line 1: Line 1: - ==Discretisation of Diffusive Term == + ==Discretisation of the diffusion term == === Description=== === Description===

- Note: The approaches those are discussed here are applicable to non-orthoganal meshes as well as orthogonal meshes. + For a general control volume (orthogonal, non-orthogonal), the discretization of the diffusion term can be written in the following form
-
+ - A control volume in mesh is made up of set of faces enclosing it. The figure 1.1 shows a typical situation. Where '''A''' represent the magnitude of area of the face. And '''n''' represents the normal unit vector of the face under consideration. + :$\int_{S}\Gamma\nabla\phi\cdot{\rm{d\vec S}} = \sum_{faces}\Gamma _f \nabla \phi _f \cdot{\rm{\vec S_f}}$
-
+ where - [[Image:Nm_descretisation_diffusionterms_01.jpg]]
+ *S denotes the surface area of the control volume - '''Figure 1.1'''
+ *$S_f$ denotes the area of a face for the control volume - :
+ As usual, the subscript f refers to a given face. The figure below describes the terminology used in the framework of a general '''non-orthogonal''' control volume
- $\vec r_{0}$ and $\vec r_{1}$ are position vector of centroids of cells cell 0 and cell 1 respectively.
+ [[Image:non_orthogonal_CV_terminology.jpg]]
- ${\rm{d\vec s}} = \vec r_{1} - \vec r_{0}$ + '''A general non-orthogonal control volume'''
+ + + If $\vec r_{P}$ and $\vec r_{N}$ are position vector of centroids of cells P and N respectively. Then, we define
+ :$\overrightarrow{d_{PN}}= \vec r_{N} - \vec r_{P}$ +

- We wish to approaximate $D_f = \Gamma _f \nabla \phi _f \bullet {\rm{\vec A}}$ at the face. + We wish to approaximate the diffusive flux $D_f = \Gamma _f \nabla \phi _f \cdot{\rm{\vec S_f}}$ at the face. === Approach 1 === === Approach 1 === - Another approach is to use a simple expression for estimating the gradient of scalar normal to the face.
+ A first approach is to use a simple expression for estimating the gradient of a scalar normal to the face.
:$:[itex] - D_f = \Gamma _f \nabla \phi _f \bullet \vec A = \Gamma _f \left[ {\left( {\phi _1 - \phi _0 } \right)\left| {{{\vec A} \over {d\vec s}}} \right|} \right] + D_f = \Gamma _f \nabla \phi _f \cdot \vec S_f = \Gamma _f \left[ {\left( {\phi _N - \phi _P } \right)\left| {{{\vec S_f} \over {\overrightarrow{d_{PN}}}}} \right|} \right]$
[/itex]
- where $\Gamma _f$ is suitable face averages.
+ where $\Gamma _f$ is a suitable face average.
- This approach is not very good when the non-orthogonality of the faces increases. Instead for the fairly non-orthogonal meshes, it is advisable to use the following approaches.
+ This approach is not very good when the non-orthogonality of the faces increases. If this is the case, it is advisable to use one of the following approaches.
=== Approach 2 === === Approach 2 === - We define vector + We define the vector - $+ :[itex] - \vec \alpha {\rm{ = }}\frac{{{\rm{\vec A}}}}{{{\rm{\vec A}} \bullet {\rm{d\vec s}}}} + \vec \alpha {\rm{ = }}\frac{{{\rm{\vec {S_f}}}}}{{{\rm{\vec S_f}} \cdot {\overrightarrow{d_{PN}}}}}$ [/itex] giving us the expression:
giving us the expression:
:$:[itex] - D_f = \Gamma _f \nabla \phi _f \bullet {\rm{\vec A = }}\Gamma _{\rm{f}} \left[ {\left( {\phi _1 - \phi _0 } \right)\vec \alpha \bullet {\rm{\vec A + }}\bar \nabla \phi \bullet {\rm{\vec A - }}\left( {\bar \nabla \phi \bullet {\rm{d\vec s}}} \right)\vec \alpha \bullet {\rm{\vec A}}} \right] + D_f = \Gamma _f \nabla \phi _f \cdot{\rm{\vec S_f = }}\Gamma _{\rm{f}} \left[ {\left( {\phi _N - \phi _P } \right)\vec \alpha \cdot {\rm{\vec S_f + }}\bar \nabla \phi_f \cdot {\rm{\vec S_f - }}\left( {\bar \nabla \phi_f \cdot {\overrightarrow{d_{PN}}}} \right)\vec \alpha \cdot {\rm{\vec S_f}}} \right]$
[/itex]
where $\bar \nabla \phi _f$ and $\Gamma _f$ are suitable face averages.
where $\bar \nabla \phi _f$ and $\Gamma _f$ are suitable face averages.
+ === Orthogonal correction approaches === + In non-orthogonal grids, the gradient direction that will yield an expression involving the values at the neighboring control volumes will have to be along the line joining the centroids of the two control volumes. If this direction has a unit vector denoted by $\vec e$ then, by definition
+ :$+ \vec e {\rm{ = }} \frac{{{\rm{\overrightarrow{d_{PN}}}}}} {\left| {\overrightarrow{d_{PN}}} \right|} +$
+ then the gradient in the direction of $\vec e$ can be written as
+ :$\nabla \phi _f \cdot \vec e = \frac {\partial \phi_f} {\partial e} = \frac { \phi_N - \phi_P} {\left| {\overrightarrow{d_{PN}}} \right|}$
+ + If the surface vector $\vec {S_f}$ is written as the summation of two vectors $\vec {E}$ and $\vec {T}$
+ :$\vec {S_f} = \vec {E} + \vec {T}$
+ where $\vec {E}$ is in the direction joining the centroids of the two control volumes, we will then be able to express the diffusive flux in terms of the neighboring control volumes plus an additional correction. This is done as follows
+ + :$\nabla \phi_f \cdot \vec {S_f} = \nabla \phi_f \cdot \vec {E} + \nabla \phi_f \cdot \vec {T}$
+ + :$\nabla \phi_f \cdot \vec {S_f} = E \nabla \phi_f \cdot \vec {e} + \nabla \phi_f \cdot \vec {T}$ .... (where E is the magnitude of $\vec E$
+ At the outset, one obtains
+ :$\nabla \phi_f \cdot \vec {S_f} = E \frac { \phi_N - \phi_P} {\left| {\overrightarrow{d_{PN}}} \right|} + \nabla \phi_f \cdot \vec {T}$
+
+ The first term in the above equation can be thought of as the orthogonal contribution to the diffusive flux, while the second term represents the non-orthogonal effects. At this point, the vector $\vec {T}$ has not been defined yet. There are three main methods to define this vector. + + ==== Minimum correction ==== + In the minimum correction approach, the vectors are defined as
+ :$\vec E = (\vec e \cdot \vec S_f)\cdot \vec e = S_f \cos\theta \vec e$
+ :$\vec T = \vec S - \vec E= S_f (\vec n - \cos\theta \vec e)$
+ [[image:Non_orthogonal_CV_minimum_correction.jpg]]
+ '''Minimum Correction Approach''' + + ==== Orthogonal correction ==== + In the orthogonal correction approach, the vectors are defined as
+ :$\vec E = S_f \vec e$
+ :$\vec T = \vec S - \vec E = S_f (\vec n - \vec e)$
+ [[image:Non_orthogonal_CV_orthogonal_correction.jpg]]
+ '''Orthogonal Correction Approach''' + + ==== Over relaxed correction ==== + Finally, in the over relaxed approach, we define
+ :$\vec E = \frac {\vec S_f \cdot \vec S_f}{\vec S_f \cdot \vec e} \vec e = \frac {S_f}{\cos \theta} \vec e$
+ :$\vec T = \vec S - \vec E = S_f (\vec n - \frac{1}{\cos \theta} \vec e)$
+ [[image:Non_orthogonal_CV_Over_relaxed_correction.jpg]]
+ '''Over Relaxed Correction Approach''' + == References == - == Reference == + #{{reference-book|author=Ferziger, J.H. and Peric, M.|year=2001|title=Computational Methods for Fluid Dynamics|rest=ISBN 3540420746, 3rd Rev. Ed., Springer-Verlag, Berlin.}} + #{{reference-paper|author=[http://www.h.jasak.dial.pipex.com/ Hrvoje, Jasak]|year=1996|title=Error Analysis and Estimation for the Finite Volume Method with Applications to Fluid Flows|rest=PhD Thesis, Imperial College, University of London ([http://www.h.jasak.dsl.pipex.com/HrvojeJasakPhD.pdf download])}} + #{{reference-paper|author=[http://webfea-lb.fea.aub.edu.lb/fea/me/CFD/ Darwish, Marwan]|year=2003|title=CFD Course Notes|rest=Notes, American University of Beirut}} + #{{reference-paper|author=[http://jedi.knows.it/ Saad, Tony]|year=2005|title=Implementation of a Finite Volume Unstructured CFD Solver Using Cluster Based Parallel Computing|rest=Thesis, American University of Beirut}} - #'''Ferziger, J.H. and Peric, M. 2002'''. "Computational Methods for Fluid Dynamics", 3rd rev. ed., Springer-Verlag, Berlin. + ---- - # '''Jasak Hrvoje''', ''PhD. Thesis'', "Error Analysis and Estimation for the Finite Volume Method with Applications to Fluid Flows " + Return to [[Numerical methods | Numerical Methods]]

## Discretisation of the diffusion term

### Description

For a general control volume (orthogonal, non-orthogonal), the discretization of the diffusion term can be written in the following form

$\int_{S}\Gamma\nabla\phi\cdot{\rm{d\vec S}} = \sum_{faces}\Gamma _f \nabla \phi _f \cdot{\rm{\vec S_f}}$

where

• S denotes the surface area of the control volume
• $S_f$ denotes the area of a face for the control volume

As usual, the subscript f refers to a given face. The figure below describes the terminology used in the framework of a general non-orthogonal control volume

A general non-orthogonal control volume

If $\vec r_{P}$ and $\vec r_{N}$ are position vector of centroids of cells P and N respectively. Then, we define

$\overrightarrow{d_{PN}}= \vec r_{N} - \vec r_{P}$

We wish to approaximate the diffusive flux $D_f = \Gamma _f \nabla \phi _f \cdot{\rm{\vec S_f}}$ at the face.

### Approach 1

A first approach is to use a simple expression for estimating the gradient of a scalar normal to the face.

$D_f = \Gamma _f \nabla \phi _f \cdot \vec S_f = \Gamma _f \left[ {\left( {\phi _N - \phi _P } \right)\left| {{{\vec S_f} \over {\overrightarrow{d_{PN}}}}} \right|} \right]$

where $\Gamma _f$ is a suitable face average.

This approach is not very good when the non-orthogonality of the faces increases. If this is the case, it is advisable to use one of the following approaches.

### Approach 2

We define the vector

$\vec \alpha {\rm{ = }}\frac{{{\rm{\vec {S_f}}}}}{{{\rm{\vec S_f}} \cdot {\overrightarrow{d_{PN}}}}}$

giving us the expression:

$D_f = \Gamma _f \nabla \phi _f \cdot{\rm{\vec S_f = }}\Gamma _{\rm{f}} \left[ {\left( {\phi _N - \phi _P } \right)\vec \alpha \cdot {\rm{\vec S_f + }}\bar \nabla \phi_f \cdot {\rm{\vec S_f - }}\left( {\bar \nabla \phi_f \cdot {\overrightarrow{d_{PN}}}} \right)\vec \alpha \cdot {\rm{\vec S_f}}} \right]$

where $\bar \nabla \phi _f$ and $\Gamma _f$ are suitable face averages.

### Orthogonal correction approaches

In non-orthogonal grids, the gradient direction that will yield an expression involving the values at the neighboring control volumes will have to be along the line joining the centroids of the two control volumes. If this direction has a unit vector denoted by $\vec e$ then, by definition

$\vec e {\rm{ = }} \frac{{{\rm{\overrightarrow{d_{PN}}}}}} {\left| {\overrightarrow{d_{PN}}} \right|}$

then the gradient in the direction of $\vec e$ can be written as

$\nabla \phi _f \cdot \vec e = \frac {\partial \phi_f} {\partial e} = \frac { \phi_N - \phi_P} {\left| {\overrightarrow{d_{PN}}} \right|}$

If the surface vector $\vec {S_f}$ is written as the summation of two vectors $\vec {E}$ and $\vec {T}$

$\vec {S_f} = \vec {E} + \vec {T}$

where $\vec {E}$ is in the direction joining the centroids of the two control volumes, we will then be able to express the diffusive flux in terms of the neighboring control volumes plus an additional correction. This is done as follows

$\nabla \phi_f \cdot \vec {S_f} = \nabla \phi_f \cdot \vec {E} + \nabla \phi_f \cdot \vec {T}$
$\nabla \phi_f \cdot \vec {S_f} = E \nabla \phi_f \cdot \vec {e} + \nabla \phi_f \cdot \vec {T}$ .... (where E is the magnitude of $\vec E$

At the outset, one obtains

$\nabla \phi_f \cdot \vec {S_f} = E \frac { \phi_N - \phi_P} {\left| {\overrightarrow{d_{PN}}} \right|} + \nabla \phi_f \cdot \vec {T}$

The first term in the above equation can be thought of as the orthogonal contribution to the diffusive flux, while the second term represents the non-orthogonal effects. At this point, the vector $\vec {T}$ has not been defined yet. There are three main methods to define this vector.

#### Minimum correction

In the minimum correction approach, the vectors are defined as

$\vec E = (\vec e \cdot \vec S_f)\cdot \vec e = S_f \cos\theta \vec e$
$\vec T = \vec S - \vec E= S_f (\vec n - \cos\theta \vec e)$

Minimum Correction Approach

#### Orthogonal correction

In the orthogonal correction approach, the vectors are defined as

$\vec E = S_f \vec e$
$\vec T = \vec S - \vec E = S_f (\vec n - \vec e)$

Orthogonal Correction Approach

#### Over relaxed correction

Finally, in the over relaxed approach, we define

$\vec E = \frac {\vec S_f \cdot \vec S_f}{\vec S_f \cdot \vec e} \vec e = \frac {S_f}{\cos \theta} \vec e$
$\vec T = \vec S - \vec E = S_f (\vec n - \frac{1}{\cos \theta} \vec e)$

Over Relaxed Correction Approach

## References

1. Ferziger, J.H. and Peric, M. (2001), Computational Methods for Fluid Dynamics, ISBN 3540420746, 3rd Rev. Ed., Springer-Verlag, Berlin..
2. Hrvoje, Jasak (1996), "Error Analysis and Estimation for the Finite Volume Method with Applications to Fluid Flows", PhD Thesis, Imperial College, University of London (download).
3. Darwish, Marwan (2003), "CFD Course Notes", Notes, American University of Beirut.
4. Saad, Tony (2005), "Implementation of a Finite Volume Unstructured CFD Solver Using Cluster Based Parallel Computing", Thesis, American University of Beirut.