Faster Convergence

marco.bianchini · December 6, 2016, 04:37

Hi everyone,
it might sound basic for you, but I am new to CFD with star ccm+.
I am doing my first simulation and it is taking a lot of iteration to converge.
The question is, when I will make a new simulation with a finer mesh, how do I use the coarser one to reach convergence faster?
I am sorry if someone else already posted the same question, I could not find the answer.
Thank you.

Marco.

me3840 · December 6, 2016, 11:10

If you get your solution on the coarse mesh and then remesh to fine, the old solution is automatically interpolated onto the new mesh. There isn't anything you have to do then.

kirrer · December 6, 2016, 16:14

In addition, the Grid Sequencing option of the coupled solver allows for this to be done automatically. CCM+ will build progressively finer hexahedral meshes, approximating your geometry, solving the flow field inviscidly to initialize velocity; then it will apply the velocity field to your actual mesh as initial condition.

Also, my understanding of the algebraic multigrid approach is that it too creates progressively smaller virtual grids on which the solution is solved before solving the finer details on the full mesh, but my understanding stops there (and only because a support engineer explained it to me). I don't have all the details of that approach.

In essence, if you're using either of these options you may already have good convergence. The coupled solver also has the continuum convergence accelerator, but I've little experience with that.

If you look into these and still feel that you're converging slowly, your next task might be to quantify or describe what "taking a lot of iterations" means - 1000 iterations? 10,000? How big of a model or what type of physics are being solved?

me3840 · December 6, 2016, 20:05

Just to be precise, grid sequencing works on any mesh (not just hexahedral) and works for all primal flow equations (not just velocity).

AMG is a little different since the cells are agglomerated algebraically just based on the system matrix. Grid sequencing is technically GAMG, geometric algebraic multigrid, since it combines based on space.

If you are new to CFD I would not recommend turning on CCA. It takes some experience to use it to good effect.

December 6, 2016, 04:37	Faster Convergence	#1
marco.bianchini New Member Marco Bianchini Join Date: Dec 2016 Posts: 8 Rep Power: 9	Hi everyone, it might sound basic for you, but I am new to CFD with star ccm+. I am doing my first simulation and it is taking a lot of iteration to converge. The question is, when I will make a new simulation with a finer mesh, how do I use the coarser one to reach convergence faster? I am sorry if someone else already posted the same question, I could not find the answer. Thank you. Marco.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Convergence	Centurion2011	FLUENT	48	June 14, 2022 23:29
faster convergence of temperatures	keeper	CFX	3	March 30, 2016 10:43
Speed of Convergence faster with smaller time step	Torque_Converter	CFX	1	May 15, 2012 19:22
Force can not converge	colopolo	CFX	13	October 4, 2011 22:03
Convergence of CFX field in FSI analysis	nasdak	CFX	2	June 29, 2009 01:17

December 6, 2016, 11:10		#2
me3840 Senior Member Join Date: Nov 2010 Location: USA Posts: 1,232 Rep Power: 24	If you get your solution on the coarse mesh and then remesh to fine, the old solution is automatically interpolated onto the new mesh. There isn't anything you have to do then.

December 6, 2016, 16:14		#3
kirrer New Member Kevin Join Date: Oct 2012 Posts: 29 Rep Power: 13	In addition, the Grid Sequencing option of the coupled solver allows for this to be done automatically. CCM+ will build progressively finer hexahedral meshes, approximating your geometry, solving the flow field inviscidly to initialize velocity; then it will apply the velocity field to your actual mesh as initial condition. Also, my understanding of the algebraic multigrid approach is that it too creates progressively smaller virtual grids on which the solution is solved before solving the finer details on the full mesh, but my understanding stops there (and only because a support engineer explained it to me). I don't have all the details of that approach. In essence, if you're using either of these options you may already have good convergence. The coupled solver also has the continuum convergence accelerator, but I've little experience with that. If you look into these and still feel that you're converging slowly, your next task might be to quantify or describe what "taking a lot of iterations" means - 1000 iterations? 10,000? How big of a model or what type of physics are being solved?

December 6, 2016, 20:05		#4
me3840 Senior Member Join Date: Nov 2010 Location: USA Posts: 1,232 Rep Power: 24	Just to be precise, grid sequencing works on any mesh (not just hexahedral) and works for all primal flow equations (not just velocity). AMG is a little different since the cells are agglomerated algebraically just based on the system matrix. Grid sequencing is technically GAMG, geometric algebraic multigrid, since it combines based on space. If you are new to CFD I would not recommend turning on CCA. It takes some experience to use it to good effect.