Maximum overlap when partitioning increases from steady-state to transient

lowdo · March 25, 2021, 07:38

Hi all,

I'm running some cases in CFX which involve a steady state run which is used to initialise a transient run. I've noticed that for the transient phase, the maximum overlap between partitions increases significantly compared to the steady state phase (it doubles, as a minimum), despite using the same number of partitions. I've tested this on a couple of different hardware configurations and it seems to be consistent. Why is this? And does it matter? I'm trying to take account of the guidance in the CFX manual that the maximum overlap should not exceed 20%, but this would mean using fewer partitions for the transient run, although I've seen a suggestion on this forum that that figure may be pessimistic.

Thanks,

Chris

Opaque · March 25, 2021, 08:04

You have not given enough information about your setup.

For example, for a single domain model, there is no way the spatial partitioning be that much different between steady-state and transient. Therefore, I can only assume you are running a multi-domain model.

On the multi-domain model, you must be doing something specific at the domain interfaces that could change the behavior of the partitioning algorithm.

A few examples:

1 - if you are using a general connection interface, the partitions should be identical or very close between steady and transient models

2 - if you are using a frame change model, say mixing-plane or frozen rotor, the partitions, as far as I understand them, should be nearly the same again

3 - Now, if you are using a sliding mesh (transient rotor-stator) model, then spatial partitioning at the domain interface must account for the possible configuration the interface could have during the simulation, correct? In that case, a partition that spans across the domain interface may need full information from connecting domain all the way around the interface, i.e. a greater overlap.

Which example does your model fall into?

lowdo · March 25, 2021, 08:07

Thanks, I think that probably explains the difference then and was something I had wondered about. The transient part of the run is a transient rotor-stator simulation.

March 25, 2021, 07:38	Maximum overlap when partitioning increases from steady-state to transient	#1
lowdo New Member Chris Join Date: Feb 2013 Posts: 17 Rep Power: 13	Hi all, I'm running some cases in CFX which involve a steady state run which is used to initialise a transient run. I've noticed that for the transient phase, the maximum overlap between partitions increases significantly compared to the steady state phase (it doubles, as a minimum), despite using the same number of partitions. I've tested this on a couple of different hardware configurations and it seems to be consistent. Why is this? And does it matter? I'm trying to take account of the guidance in the CFX manual that the maximum overlap should not exceed 20%, but this would mean using fewer partitions for the transient run, although I've seen a suggestion on this forum that that figure may be pessimistic. Thanks, Chris

March 25, 2021, 08:04		#2
Opaque Senior Member Join Date: Jun 2009 Posts: 1,880 Rep Power: 33	You have not given enough information about your setup. For example, for a single domain model, there is no way the spatial partitioning be that much different between steady-state and transient. Therefore, I can only assume you are running a multi-domain model. On the multi-domain model, you must be doing something specific at the domain interfaces that could change the behavior of the partitioning algorithm. A few examples: 1 - if you are using a general connection interface, the partitions should be identical or very close between steady and transient models 2 - if you are using a frame change model, say mixing-plane or frozen rotor, the partitions, as far as I understand them, should be nearly the same again 3 - Now, if you are using a sliding mesh (transient rotor-stator) model, then spatial partitioning at the domain interface must account for the possible configuration the interface could have during the simulation, correct? In that case, a partition that spans across the domain interface may need full information from connecting domain all the way around the interface, i.e. a greater overlap. Which example does your model fall into? __________________ 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
Transient & Steady state velocity profile	TheMightySpartan	FLUENT	2	September 24, 2018 21:38
Effect of initial condition for steady state vs Transient	prasa	ANSYS	0	August 22, 2018 05:45
Convergence in steady state simulations vs transient ones	cardioCFD	CFX	5	January 21, 2018 11:59
time step and iterations in steady state problem using transient solver	jing113cn	FLUENT	2	January 15, 2010 04:18
About the difference between steady and unsteady problems	Lisa	Main CFD Forum	11	July 5, 2000 15:37

March 25, 2021, 08:07		#3
lowdo New Member Chris Join Date: Feb 2013 Posts: 17 Rep Power: 13	Thanks, I think that probably explains the difference then and was something I had wondered about. The transient part of the run is a transient rotor-stator simulation.