successful time step for 2D problem failing for 3D case of same problem.

mandman · March 9, 2015, 09:02

Hello,
I am trying to simulate 3D convective diffusive unsteady problem using SIMPLER technique (Finite Volume Method). I got successful 2D case of the problem. For 3D I am keeping the size of space step same as in 2D, with one extra direction.
The problem is in choosing the time step. When I keep it same as in 2D the simulation diverges and in order to converge it I have to reduce the time step by order of 4.

Is this thing common or I am committing some other problem.

Unluckily if its common than it will take ages for this simulation.

BTW I am running this code with gfortran compilers.

FMDenaro · March 9, 2015, 12:17

Quote:

Originally Posted by mandman

Hello,
I am trying to simulate 3D convective diffusive unsteady problem using SIMPLER technique (Finite Volume Method). I got successful 2D case of the problem. For 3D I am keeping the size of space step same as in 2D, with one extra direction.
The problem is in choosing the time step. When I keep it same as in 2D the simulation diverges and in order to converge it I have to reduce the time step by order of 4.

Is this thing common or I am committing some other problem.

Unluckily if its common than it will take ages for this simulation.

BTW I am running this code with gfortran compilers.

yes, the numerical stability involves the relation of dimensionality therefore the constraint is more strict when going from 1D, 2D, 3D.

adrin · March 9, 2015, 22:32

Though it is true that the stability constraint for 3D is stricter than in 2D, it is difficult to believe that the time step size would have to drop by 4 _orders_. By 4 _orders_ I understand you're saying 10^4 (10,000 times). Otherwise, if the drop is _factor_ 4 (4 times) it is somewhat more reasonable, though it depends (very much) on the problem and what physics are introduced by the addition of the third dimension (and 4x increase doesn't really translate to "ages" of simulation time in added cost)

adrin

agd · March 10, 2015, 09:31

If you are saying that the time step has to be 10000 times smaller rather than 4 times smaller, I would look for an error in the coding. A factor of 10000 is not reasonable for a jump to 3D, in my experience.

mandman · March 10, 2015, 15:18

by O(4) I mean 10000.

adrin · March 10, 2015, 15:40

Then you most likely (definitely?) have a bug in the code! What are the 2- and 3-D problems you were solving and what are the corresponding CFL values?

adrin

mandman · March 10, 2015, 20:01

Ya!! you guys are right. I found THE "bug", it was resting in expression calculating the difference between pressure and pressure correction.
I was struggling with this problem for last 2 months, and I was so much confident about my code that I thought that its common for 3D problems. But you guys conviction made me doubtful.
I even learned OpenMP and parallelised this code, but there’s no way I could have compensated for O(4) time increase.
Thanks all

I am working on double diffusive convection problem, let me know if you want to know more. You can email me on manoj_d@students.iitmandi.ac.in

March 9, 2015, 09:02	successful time step for 2D problem failing for 3D case of same problem.	#1
mandman New Member Manoj Dhiman Join Date: Dec 2014 Posts: 4 Rep Power: 11	Hello, I am trying to simulate 3D convective diffusive unsteady problem using SIMPLER technique (Finite Volume Method). I got successful 2D case of the problem. For 3D I am keeping the size of space step same as in 2D, with one extra direction. The problem is in choosing the time step. When I keep it same as in 2D the simulation diverges and in order to converge it I have to reduce the time step by order of 4. Is this thing common or I am committing some other problem. Unluckily if its common than it will take ages for this simulation. BTW I am running this code with gfortran compilers.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
a problem with convergence in buoyantSimpleFoam	skuznet	OpenFOAM Running, Solving & CFD	6	November 15, 2017 13:12
High Courant Number @ icoFoam	Artex85	OpenFOAM Running, Solving & CFD	11	February 16, 2017 14:40
bubbleFoam tutorial case with turbulence	setzberg	OpenFOAM Running, Solving & CFD	6	October 15, 2015 14:29
Superlinear speedup in OpenFOAM 13	msrinath80	OpenFOAM Running, Solving & CFD	18	March 3, 2015 06:36
convergence problem when use pisoFoam, LES for wind tunnel case	Forrest_Lei	OpenFOAM	3	July 19, 2011 07:00

March 9, 2015, 22:32		#3
adrin Senior Member adrin Join Date: Mar 2009 Posts: 115 Rep Power: 17	Though it is true that the stability constraint for 3D is stricter than in 2D, it is difficult to believe that the time step size would have to drop by 4 _orders_. By 4 _orders_ I understand you're saying 10^4 (10,000 times). Otherwise, if the drop is _factor_ 4 (4 times) it is somewhat more reasonable, though it depends (very much) on the problem and what physics are introduced by the addition of the third dimension (and 4x increase doesn't really translate to "ages" of simulation time in added cost) adrin

March 10, 2015, 09:31		#4
agd Senior Member Join Date: Jul 2009 Posts: 358 Rep Power: 19	If you are saying that the time step has to be 10000 times smaller rather than 4 times smaller, I would look for an error in the coding. A factor of 10000 is not reasonable for a jump to 3D, in my experience.

March 10, 2015, 15:18		#5
mandman New Member Manoj Dhiman Join Date: Dec 2014 Posts: 4 Rep Power: 11	by O(4) I mean 10000.

March 10, 2015, 15:40		#6
adrin Senior Member adrin Join Date: Mar 2009 Posts: 115 Rep Power: 17	Then you most likely (definitely?) have a bug in the code! What are the 2- and 3-D problems you were solving and what are the corresponding CFL values? adrin

March 10, 2015, 20:01		#7
mandman New Member Manoj Dhiman Join Date: Dec 2014 Posts: 4 Rep Power: 11	Ya!! you guys are right. I found THE "bug", it was resting in expression calculating the difference between pressure and pressure correction. I was struggling with this problem for last 2 months, and I was so much confident about my code that I thought that its common for 3D problems. But you guys conviction made me doubtful. I even learned OpenMP and parallelised this code, but there’s no way I could have compensated for O(4) time increase. Thanks all I am working on double diffusive convection problem, let me know if you want to know more. You can email me on manoj_d@students.iitmandi.ac.in