

July 15, 2024 at 5:18 pmSimonOlivierSubscriber
Hi,
I would like your help since I am a stuck with this problem.
Context:
I solve a transient and nonlinear TE simulation with standard NewtonRaphson.
 With 2023 R1 version, the model is very stable and converges very well (symmetric global matrix)
 With 2023 R2 version, the model no longer converges (non symmetric global matrix)
I noticed that:
 With 2023 R1 version , the model generates a symmetric global matrix
 With 2023 R2 version, the model generates a nonsymmetrical global matrix, possibly brought about by the addition of new terms in elementary calculation
My question:
 What caused this nonsymmetry in 2023 R2 version?
Is there an option somewhere that eliminates the consideration of this nonsymmetry (to generate a symmetric global matrix)?
Thank you very much for your help. 
July 16, 2024 at 2:12 pmAshish KhemkaForum Moderator
Hi Simon,
You can compare the input files  ds.dat for both the analysis. Right click on solution information and you will be able to see a file named ds.dat which is the input file to the solver. Comparing them will give you an idea on what is different in the setup for both the analysis.
Regards,
Ashish Khemka

July 22, 2024 at 7:21 pmSimonOlivierSubscriber
Â
Hi,Â
Thank you for the reply.
I compare the ds file for a very simple model and the only change is my node numerotation and I donâ€™t know why (same number of node/elements/quality statisticsâ€¦).
Also, the change in the global matrix formulation stays the same:
Â
 With 2023 R1 version, tsymmetric global matrix
 With 2023 R2 version,non symmetric global matrix
But this time, since the model is very simple, I can get a solution in the 2 cases.
Is there a reason for matrix symetric/nonsimmetric?
Thank you for your help
Â


July 24, 2024 at 7:57 amAshish KhemkaForum Moderator
Hi Simon,
Thanks for sharing this. Let me check and I will let you know.
Regards,
Ashish Khemka

July 24, 2024 at 1:38 pmdloomanAnsys Employee
Around that time we made the Seebeck effect strongly coupled.Â This is probably the issue.Â Set KEYOPT(2)=1 to go back to load vector coupling.Â LNSRCH,ON is frequently helpful in these analyses.

 You must be logged in to reply to this topic.

451

199

194

166

162
Â© 2024 Copyright ANSYS, Inc. All rights reserved.