STAR-CCM+ Stopping simulations after residuals convergence

skhan547 · September 21, 2018, 18:46

Hi all,

I've been wondering how to stop a simulation based on residuals convergence.

When I've got MATLAB models, I usually exit my numerical analysis when error hits a predetermined lower limits, such as 1e-4. The only thing I've found built into STAR-CCM+ is the 'Stopping Criteria', which seems to be based off number of iterations. I want to optimise here, and not run more iterations than needed.

I'm proficient at writing macros, so if there's anything in the API you'd like to point me to, I'd appreciate the reference.

Thanks!

me3840 · September 21, 2018, 23:41

There are stopping criterion you can add from the residuals. Right click the stopping criterion folder in the tree. Then add one based on a monitor. A monitor exists for all residuals.

Though I will add deciding convergence on residuals alone is not a good idea. You should be monitoring engineering quantities and ensure they converge as well.

skhan547 · September 22, 2018, 00:20

Thanks for the help, that worked great for what I wanted!

However, I thought that convergence of engineering quantities was proportionally reflected in the convergence of error (ie, residuals). I may have a flawed understanding of residuals then; will try and do more reading.

me3840 · September 22, 2018, 00:51

The residuals are just one measure of convergence. The value you get is just a volume weighted average of sorts of all residuals in the domain. I'm sure you can imagine several ways such a value can fool you. In academic problems it can be easier to actually get them to machine zero, but for engineering problems this is often impossible. They may just flatline or oscillate about a low value as the linear solver struggles to reduce error further. Many times this would make evaluation of convergence ambiguous without another quantity to measure.

It may or may not be true for your problem, but I would encourage you to monitor both and find out. Ideally you should have a mix of important localized values (probes in a shear layer or expected area of instability, for example) and integrated quantities (drag, nusselt number, mass flow rate in a passage, etc). Monitor a few and see what happens on different cases, you may find yourself surprised.

September 21, 2018, 18:46	STAR-CCM+ Stopping simulations after residuals convergence	#1
skhan547 New Member Sameh Khan Join Date: Sep 2018 Posts: 2 Rep Power: 0	Hi all, I've been wondering how to stop a simulation based on residuals convergence. When I've got MATLAB models, I usually exit my numerical analysis when error hits a predetermined lower limits, such as 1e-4. The only thing I've found built into STAR-CCM+ is the 'Stopping Criteria', which seems to be based off number of iterations. I want to optimise here, and not run more iterations than needed. I'm proficient at writing macros, so if there's anything in the API you'd like to point me to, I'd appreciate the reference. Thanks!

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
convergence and residuals	ebm	CFX	2	February 2, 2009 17:55
error in star ccm	maurizio	Siemens	3	October 16, 2007 05:17
convergence of DPM simulations	fluid	FLUENT	0	October 12, 2007 09:11
Convergence - scaled vs unscaled residuals	HS	FLUENT	1	November 7, 2005 05:45
[Commercial meshers] Trimmed cell and embedded refinement mesh conversion issues	michele	OpenFOAM Meshing & Mesh Conversion	2	July 15, 2005 04:15

September 21, 2018, 23:41		#2
me3840 Senior Member Join Date: Nov 2010 Location: USA Posts: 1,232 Rep Power: 24	There are stopping criterion you can add from the residuals. Right click the stopping criterion folder in the tree. Then add one based on a monitor. A monitor exists for all residuals. Though I will add deciding convergence on residuals alone is not a good idea. You should be monitoring engineering quantities and ensure they converge as well.

September 22, 2018, 00:20		#3
skhan547 New Member Sameh Khan Join Date: Sep 2018 Posts: 2 Rep Power: 0	Thanks for the help, that worked great for what I wanted! However, I thought that convergence of engineering quantities was proportionally reflected in the convergence of error (ie, residuals). I may have a flawed understanding of residuals then; will try and do more reading.

September 22, 2018, 00:51		#4
me3840 Senior Member Join Date: Nov 2010 Location: USA Posts: 1,232 Rep Power: 24	The residuals are just one measure of convergence. The value you get is just a volume weighted average of sorts of all residuals in the domain. I'm sure you can imagine several ways such a value can fool you. In academic problems it can be easier to actually get them to machine zero, but for engineering problems this is often impossible. They may just flatline or oscillate about a low value as the linear solver struggles to reduce error further. Many times this would make evaluation of convergence ambiguous without another quantity to measure. It may or may not be true for your problem, but I would encourage you to monitor both and find out. Ideally you should have a mix of important localized values (probes in a shear layer or expected area of instability, for example) and integrated quantities (drag, nusselt number, mass flow rate in a passage, etc). Monitor a few and see what happens on different cases, you may find yourself surprised.