SnGradSchemes

sega · December 15, 2008, 06:58

Hello!

I'm using a mesh which contains a small amount of tetrahedral cells.

Do I have to change the snGradScheme from corrected to uncorrected or limited in this case?

immortality · February 27, 2013, 14:14

is uncorrected for tetra grids?

immortality · April 2, 2013, 06:03

whats usage of corrected and uncorrected?

Tobias Adam · August 7, 2014, 08:16

I´d like to know this too :-)

I don´t know which setting is best for unstructured mesh and for structured mesh

tcarrigan · August 10, 2014, 01:12

The decision to use either corrected, uncorrected, or limited is a function of the orthogonality of the mesh, or more specifically, non-orthogonality. OpenFOAM defines non-orthogonality as the angle between the vector connecting two adjacent cell centers and the face normal vector as illustrated by theta in the attached image. The larger this value, the more non-orthogonal the grid becomes.

nonOrthogonality.PNG

In your case, the quad elements are highly orthogonal. However, the delaunay like triangulation in the corner has generated cells that contain slight non-orthogonality. Running checkMesh may only flag one or two faces as non-orthogonal, but overall the triangular mesh isn't bad at all. However, you may want to consider using a non-orthogonal correction.

When evaluating the surface normal gradient, the computation will be exact if and only if the face normal vector aligns with the vector connecting the two adjacent cell centers. However, for the triangular portion of your grid, and for more complex grids in general, this isn't necessarily going to be the case. Therefore, for a non-orthogonal mesh, OpenFOAM provides schemes for adding an explicit non-orthogonal correction to the surface normal gradient calculation. This correction is evaluated by interpolating the gradients at the adjacent cell centers through a differencing operation.

The corrected scheme is going to use the explicit non-orthogonal correction in its entirety. Uncorrected, as its name suggests, will not utilize the non-orthogonal correction. Whereas limited allows you to specify how much of a correction you desire depending on the severity of the non-orthogonality.

So, when do you use these schemes? After running checkMesh you should get a sense for your grid quality. Pay particular attention to the non-orthogonality. If you find your grid to be non-orthogonal, or if you have a few triangles in your mesh, you may want to use the limited scheme with a value of 0.333 to begin with and turn it up until right before the solver becomes unstable. This may take a few tries, but the goal is to select a value that provides a good balance between stability and solution accuracy.

Tobias Adam · October 13, 2014, 05:51

hmmm, didn´t you do a misstake with the limited scheme?

when you turn up the value of limited, it becomes more like corrected and therefore should be more stable.
limited 0 corresponds to uncorrected. Therefore it does not have orthogonal correction and is less stable, isn´t it?

Best regards Tobi

RodriguezFatz · October 13, 2014, 06:31

Tobias,
yes, limited "0" corresponds to uncorrected, but:
No, "corrected" is unstable, "uncorrected" is stable.
Try to find the thesis of Eugene de Villiers (pdf), he explains it on page 112:

While the diffusion term is bounded on orthogonal meshes, addition of the nonorthogonal correction can produce negative coefficients leading to unboundedness and instability, particularly if the mesh non-orthogonality is high. Under such conditions, the non-orthogonal contribution has to be limited, which reduces the formal accuracy of the diffusion term.

pici · December 12, 2014, 13:49

Hi to everybody.
I'm using a mesh created by Salomé, that is made by thetra and prims (generated by Viscous Layer creation).
Result of checkMesh utility is this:
-------------------------------------------------------------------
Checking geometry...
Overall domain bounding box (-0.005 -0.005 0) (0.008 0.005 0.02)
Mesh (non-empty, non-wedge) directions (1 1 1)
Mesh (non-empty) directions (1 1 1)
Boundary openness (-3.33448e-17 1.39317e-16 -1.78686e-17) OK.
Max cell openness = 1.04357e-15 OK.
Max aspect ratio = 24.5239 OK.
Minimum face area = 2.07388e-09. Maximum face area = 8.30801e-07. Face area magnitudes OK.
Min volume = 1.03026e-13. Max volume = 2.56118e-10. Total volume = 1.57646e-06. Cell volumes OK.
Mesh non-orthogonality Max: 66.7712 average: 15.0561
Non-orthogonality check OK.
Face pyramids OK.
Max skewness = 2.47683 OK.
Coupled point location match (average 0) OK.
----------------------------------------------------
My question is: how do I have to read 'red' infos? do I have to to use the limited or corrected scheme?
Last (but not list, for me) can you suggest the right dictionary where to add this option?
Many thanks
Piero

KateEisenhower · September 3, 2015, 07:58

Quote:

Originally Posted by pici

Hi to everybody.
I'm using a mesh created by Salomé, that is made by thetra and prims (generated by Viscous Layer creation).
Result of checkMesh utility is this:
-------------------------------------------------------------------
Checking geometry...
Overall domain bounding box (-0.005 -0.005 0) (0.008 0.005 0.02)
Mesh (non-empty, non-wedge) directions (1 1 1)
Mesh (non-empty) directions (1 1 1)
Boundary openness (-3.33448e-17 1.39317e-16 -1.78686e-17) OK.
Max cell openness = 1.04357e-15 OK.
Max aspect ratio = 24.5239 OK.
Minimum face area = 2.07388e-09. Maximum face area = 8.30801e-07. Face area magnitudes OK.
Min volume = 1.03026e-13. Max volume = 2.56118e-10. Total volume = 1.57646e-06. Cell volumes OK.
Mesh non-orthogonality Max: 66.7712 average: 15.0561
Non-orthogonality check OK.
Face pyramids OK.
Max skewness = 2.47683 OK.
Coupled point location match (average 0) OK.
----------------------------------------------------
My question is: how do I have to read 'red' infos? do I have to to use the limited or corrected scheme?
Last (but not list, for me) can you suggest the right dictionary where to add this option?
Many thanks
Piero

Hi Piero,

the limited scheme is the same as the corrected scheme if you set the coefficient to 1. You'll want to look for the snGradSchemes subdictionary located at /system/fvschemes directory.

Best regards,

Kate

squick · September 6, 2015, 12:24

The settings do not matter. OpenFoam does not converge on unstructured meshes.

See
http://www.cfd-online.com/Forums/ope...ed-meshes.html
http://www.paralgo.fr/publications/C...ce-OF-2015.pdf

The best alternative is to use AMR meshes.

inginheiro · December 1, 2016, 08:53

Please pay attention with those statements. Of corse OpenFoam can handle unstructured meshes. The solver settings just need to be adapted accordingly.

December 15, 2008, 06:58	Hello! I'm using a mesh whi	#1
sega Senior Member Sebastian Gatzka Join Date: Mar 2009 Location: Frankfurt, Germany Posts: 729 Rep Power: 20	Hello! I'm using a mesh which contains a small amount of tetrahedral cells. Do I have to change the snGradScheme from corrected to uncorrected or limited in this case? __________________ Schrödingers wife: "What did you do to the cat? It's half dead!"

August 10, 2014, 01:12		#5
tcarrigan Senior Member Travis Carrigan Join Date: Jul 2010 Location: Arlington, TX Posts: 161 Rep Power: 15	The decision to use either corrected, uncorrected, or limited is a function of the orthogonality of the mesh, or more specifically, non-orthogonality. OpenFOAM defines non-orthogonality as the angle between the vector connecting two adjacent cell centers and the face normal vector as illustrated by theta in the attached image. The larger this value, the more non-orthogonal the grid becomes. nonOrthogonality.PNG In your case, the quad elements are highly orthogonal. However, the delaunay like triangulation in the corner has generated cells that contain slight non-orthogonality. Running checkMesh may only flag one or two faces as non-orthogonal, but overall the triangular mesh isn't bad at all. However, you may want to consider using a non-orthogonal correction. When evaluating the surface normal gradient, the computation will be exact if and only if the face normal vector aligns with the vector connecting the two adjacent cell centers. However, for the triangular portion of your grid, and for more complex grids in general, this isn't necessarily going to be the case. Therefore, for a non-orthogonal mesh, OpenFOAM provides schemes for adding an explicit non-orthogonal correction to the surface normal gradient calculation. This correction is evaluated by interpolating the gradients at the adjacent cell centers through a differencing operation. The corrected scheme is going to use the explicit non-orthogonal correction in its entirety. Uncorrected, as its name suggests, will not utilize the non-orthogonal correction. Whereas limited allows you to specify how much of a correction you desire depending on the severity of the non-orthogonality. So, when do you use these schemes? After running checkMesh you should get a sense for your grid quality. Pay particular attention to the non-orthogonality. If you find your grid to be non-orthogonal, or if you have a few triangles in your mesh, you may want to use the limited scheme with a value of 0.333 to begin with and turn it up until right before the solver becomes unstable. This may take a few tries, but the goal is to select a value that provides a good balance between stability and solution accuracy. Hisham, sharonyue, Julien and 21 others like this. __________________ Travis Carrigan Manager, Business Development Pointwise, Inc.

October 13, 2014, 06:31		#7
RodriguezFatz Senior Member Philipp Join Date: Jun 2011 Location: Germany Posts: 1,297 Rep Power: 26	Tobias, yes, limited "0" corresponds to uncorrected, but: No, "corrected" is unstable, "uncorrected" is stable. Try to find the thesis of Eugene de Villiers (pdf), he explains it on page 112: While the diffusion term is bounded on orthogonal meshes, addition of the nonorthogonal correction can produce negative coefficients leading to unboundedness and instability, particularly if the mesh non-orthogonality is high. Under such conditions, the non-orthogonal contribution has to be limited, which reduces the formal accuracy of the diffusion term. Tobias Adam, bikooo3878, Mansur and 2 others like this. __________________ The skeleton ran out of shampoo in the shower.

December 12, 2014, 13:49	about non-orthogonality	#8
pici Member Pierluigi Cirrottola Join Date: Jun 2013 Posts: 52 Rep Power: 12	Hi to everybody. I'm using a mesh created by Salomé, that is made by thetra and prims (generated by Viscous Layer creation). Result of checkMesh utility is this: ------------------------------------------------------------------- Checking geometry... Overall domain bounding box (-0.005 -0.005 0) (0.008 0.005 0.02) Mesh (non-empty, non-wedge) directions (1 1 1) Mesh (non-empty) directions (1 1 1) Boundary openness (-3.33448e-17 1.39317e-16 -1.78686e-17) OK. Max cell openness = 1.04357e-15 OK. Max aspect ratio = 24.5239 OK. Minimum face area = 2.07388e-09. Maximum face area = 8.30801e-07. Face area magnitudes OK. Min volume = 1.03026e-13. Max volume = 2.56118e-10. Total volume = 1.57646e-06. Cell volumes OK. Mesh non-orthogonality Max: 66.7712 average: 15.0561 Non-orthogonality check OK. Face pyramids OK. Max skewness = 2.47683 OK. Coupled point location match (average 0) OK. ---------------------------------------------------- My question is: how do I have to read 'red' infos? do I have to to use the limited or corrected scheme? Last (but not list, for me) can you suggest the right dictionary where to add this option? Many thanks Piero

December 1, 2016, 08:53		#11
inginheiro New Member Join Date: Aug 2015 Posts: 8 Rep Power: 10	Please pay attention with those statements. Of corse OpenFoam can handle unstructured meshes. The solver settings just need to be adapted accordingly. jairoandres likes this.

February 27, 2013, 14:14		#2
immortality Senior Member Ehsan Join Date: Oct 2012 Location: Iran Posts: 2,208 Rep Power: 26	is uncorrected for tetra grids?

April 2, 2013, 06:03		#3
immortality Senior Member Ehsan Join Date: Oct 2012 Location: Iran Posts: 2,208 Rep Power: 26	whats usage of corrected and uncorrected?

August 7, 2014, 08:16		#4
Tobias Adam Member Tobias Adam Join Date: Oct 2013 Location: Siegen Posts: 55 Rep Power: 12	I´d like to know this too :-) I don´t know which setting is best for unstructured mesh and for structured mesh

October 13, 2014, 05:51		#6
Tobias Adam Member Tobias Adam Join Date: Oct 2013 Location: Siegen Posts: 55 Rep Power: 12	hmmm, didn´t you do a misstake with the limited scheme? when you turn up the value of limited, it becomes more like corrected and therefore should be more stable. limited 0 corresponds to uncorrected. Therefore it does not have orthogonal correction and is less stable, isn´t it? Best regards Tobi

September 6, 2015, 12:24		#10
squick New Member Gerard Join Date: Dec 2014 Posts: 7 Rep Power: 11	The settings do not matter. OpenFoam does not converge on unstructured meshes. See http://www.cfd-online.com/Forums/ope...ed-meshes.html http://www.paralgo.fr/publications/C...ce-OF-2015.pdf The best alternative is to use AMR meshes.