Monotonic Convergence Rebound

Torque_Converter · October 30, 2012, 13:01

I have noticed this twice in different simulations. I get beautiful convergence and imbalance progress; so much so I can go 1000 times the starting time scale. As soon as it gets to a certain point (in this case RMS residuals < 5e-4 and imbalances <0.5%) I get a massive rebound growth on both sides that is equally smooth and monotonic. An immediate attempt to limit these issues by lowering the time scale rarely resolves this and hours of work is lost. I haven't noticed a pattern as to why, nor any of the traditional reasons for poor convergence (poor mesh stats, ill defined or poor boundary conditions, trying to simulate extremely advanced physics).

Is this generally indicative of a specific mistake?

cdegroot · October 30, 2012, 16:02

I'm not sure about the reason for this, but to avoid "hours of work" being lost you should probably start saving backups. At least that way you don't have to start from scratch when you restart the simulation.

singer1812 · October 30, 2012, 17:24

are you running a simulation that might have a feature (high pressure wave, shock, or something like that) that is traveling in your domain in a relative "free" area then hits something (boundary of some sort)?

Even steady state solutions can have this phenomena. You might have to "travel" past this to get the correct solution field, or resolve this feature, and proceed to a converge solution.

ghorrocks · October 30, 2012, 17:49

Following on from Edmund's comment - Do not regard the convergence rebound as "lost work". Generally this means something has happened in the flow which has caused convergence to be harder. This could be an initial condition finally reaching the exit, or conditions being reached where a transient flow can stop up (eg vortex shedding).

The important thing to note is that these effects are real and the apparent good convegernce before-hand was just because the solver did not know that the simulation was harder than it thought. So previously you had an incorrect and optimistic idea of how well it had converged, and you are getting a more realistic picture now.

October 30, 2012, 13:01	Monotonic Convergence Rebound	#1
Torque_Converter Member Join Date: Jan 2012 Location: Edmonton, CA Posts: 87 Rep Power: 14	I have noticed this twice in different simulations. I get beautiful convergence and imbalance progress; so much so I can go 1000 times the starting time scale. As soon as it gets to a certain point (in this case RMS residuals < 5e-4 and imbalances <0.5%) I get a massive rebound growth on both sides that is equally smooth and monotonic. An immediate attempt to limit these issues by lowering the time scale rarely resolves this and hours of work is lost. I haven't noticed a pattern as to why, nor any of the traditional reasons for poor convergence (poor mesh stats, ill defined or poor boundary conditions, trying to simulate extremely advanced physics). Is this generally indicative of a specific mistake?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Convergence	Centurion2011	FLUENT	48	June 15, 2022 00:29
Force can not converge	colopolo	CFX	13	October 4, 2011 23:03
Convergence of CFX field in FSI analysis	nasdak	CFX	2	June 29, 2009 02:17
Defect correction and convergence	ganesh	Main CFD Forum	4	June 30, 2006 15:20
Convergence problems	Chetan	FLUENT	3	April 15, 2004 20:13

October 30, 2012, 16:02		#2
cdegroot Senior Member Chris DeGroot Join Date: Nov 2011 Location: Canada Posts: 414 Rep Power: 18	I'm not sure about the reason for this, but to avoid "hours of work" being lost you should probably start saving backups. At least that way you don't have to start from scratch when you restart the simulation.

October 30, 2012, 17:24		#3
singer1812 Senior Member Edmund Singer P.E. Join Date: Aug 2010 Location: Minneapolis, MN Posts: 511 Rep Power: 21	are you running a simulation that might have a feature (high pressure wave, shock, or something like that) that is traveling in your domain in a relative "free" area then hits something (boundary of some sort)? Even steady state solutions can have this phenomena. You might have to "travel" past this to get the correct solution field, or resolve this feature, and proceed to a converge solution.

October 30, 2012, 17:49		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,850 Rep Power: 144	Following on from Edmund's comment - Do not regard the convergence rebound as "lost work". Generally this means something has happened in the flow which has caused convergence to be harder. This could be an initial condition finally reaching the exit, or conditions being reached where a transient flow can stop up (eg vortex shedding). The important thing to note is that these effects are real and the apparent good convegernce before-hand was just because the solver did not know that the simulation was harder than it thought. So previously you had an incorrect and optimistic idea of how well it had converged, and you are getting a more realistic picture now.