convergence rate is very small

Sidharthkp · February 18, 2020, 07:21

In a transient simulation of mixed convection(rotation of outer cylinder and buoyancy) in an annular gap with a time step size of 0.0001s, i am getting convergence rate value of 1 in first iteration and near zero in second iteration for any given timestep.Attachment 74917

I dont know much about transcient simulation. am i doing it right by this result...
....please see the attachment....

Sidharthkp · February 18, 2020, 07:53

Also, I initialized the solver using the steady state result of the case i ran using the same model (omega reynolds stress) until a satisfying result is obtained.....

Also have another doubt, that how will i determine the total time that I have to set if I initialize a transient solution with results from a steady state run.

ghorrocks · February 18, 2020, 16:37

This FAQ was written for steady state simulation, but many of the comments are relevant to transient simulations as well: https://www.cfd-online.com/Wiki/Ansy...gence_criteria

My comments on your specific case are:
* You are running Reynolds Stress turbulence. These models are always much harder to converge. Only use these models if you really need them as they slow things down a lot.
* Re Stress models are highly sensitive to mesh quality. Make sure your mesh quality is as good as you can get it.
* Use smaller time steps and double precision numerics.
* You should determine your time step size by a sensitivity study. Don't guess, especially for Reynolds Stress models.

Sidharthkp · February 19, 2020, 00:53

Thank you for the replay,
I need to use RSM for this problem. I am using structured quad mesh made using Hypermesh , exported to CFX.(attached picture)

I have some more doubts:

1. I am really confused with understanding convergence in a transient problem because each timestep is converged to the limit and runs for the Total time specified is reached. So I guess the final(total) time we specify is important (But how?.)

2. In the previous attachment, the run which gave extremely opposite convergence rate during inner loop iteration (near 1 in 1st and near 0 in 2nd) failed giving insensible results(inthe transient run, not for steady laminar and RANS cases), But same case with a finer mesh gave acceptable output. So I guess as you said, Unsteady RANS is highly sensible to mesh refinement also.(Or I mght have made some mistake).

Please give your comments...

ghorrocks · February 19, 2020, 17:49

Your mesh is obviously high quality - but watch out for the transitions in mesh size and the distorted elements in the 4 corners you have defined in the circle section. These can cause problems.

I do not understand your first question. Note that the transient terms are discretised like all the other terms, and just because it converges does not mean it is accurate. That is why you need to do a sensitivity study on the time step size to check it is accurate.

Antanas · March 3, 2020, 01:36

Convergence rate in the output tables is the ratio between the normalized residual at current iteration and the normalized residual at previous iteration. So the lower convergence rate the better. If you'll RTFM, you'll know that "A residual reduction rate of 0.95 or smaller is considered typical for most situations, while a rate of 0.85 or smaller is considered to be very good" and that imbalances < 0.01 (i.e. <1%) are also good.

February 18, 2020, 07:53	initialisation	#2
Sidharthkp New Member Sidharth K PIllai Join Date: Aug 2019 Location: INDIA Posts: 12 Rep Power: 6	Also, I initialized the solver using the steady state result of the case i ran using the same model (omega reynolds stress) until a satisfying result is obtained..... Also have another doubt, that how will i determine the total time that I have to set if I initialize a transient solution with results from a steady state run.

February 18, 2020, 16:37		#3
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,705 Rep Power: 143	This FAQ was written for steady state simulation, but many of the comments are relevant to transient simulations as well: https://www.cfd-online.com/Wiki/Ansy...gence_criteria My comments on your specific case are: * You are running Reynolds Stress turbulence. These models are always much harder to converge. Only use these models if you really need them as they slow things down a lot. * Re Stress models are highly sensitive to mesh quality. Make sure your mesh quality is as good as you can get it. * Use smaller time steps and double precision numerics. * You should determine your time step size by a sensitivity study. Don't guess, especially for Reynolds Stress models. __________________ Note: I do not answer CFD questions by PM. CFD questions should be posted on the forum.

February 19, 2020, 17:49		#5
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,705 Rep Power: 143	Your mesh is obviously high quality - but watch out for the transitions in mesh size and the distorted elements in the 4 corners you have defined in the circle section. These can cause problems. I do not understand your first question. Note that the transient terms are discretised like all the other terms, and just because it converges does not mean it is accurate. That is why you need to do a sensitivity study on the time step size to check it is accurate. __________________ Note: I do not answer CFD questions by PM. CFD questions should be posted on the forum.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
shear rate and strain rate	Lilly	FLUENT	1	August 29, 2014 00:46
error and convergence rate	HaKu	Main CFD Forum	0	April 28, 2011 17:39
Convergence rate of SimpleFoam	boddyouareboy	OpenFOAM	0	January 25, 2011 03:49
Convergence in Fluent at small (order 1) Re	Sergei Chernyshenko	FLUENT	0	January 10, 2008 05:39
Convergence Rate Study	Khurram	Main CFD Forum	1	May 15, 2002 05:05

March 3, 2020, 01:36		#6
Antanas Senior Member Join Date: Feb 2011 Posts: 496 Rep Power: 18	Convergence rate in the output tables is the ratio between the normalized residual at current iteration and the normalized residual at previous iteration. So the lower convergence rate the better. If you'll RTFM, you'll know that "A residual reduction rate of 0.95 or smaller is considered typical for most situations, while a rate of 0.85 or smaller is considered to be very good" and that imbalances < 0.01 (i.e. <1%) are also good.