Inconsistency in Beta Value in runScripts_v2.py for FieldInversion Task #595
-
Dear Professor He, I am facing an uncertain issue in parallel computing. I used runScripts_v2.py for the FieldInversion task. To my surprise, while running the runScripts_v2 version, I noticed that the beta_fieldInversion term in primalSolver is not zero, which seems illogical. On further investigation, I identified the problem here. In this code section, I found that the values of dvDict and DV_listGlobal are different, leading to a beta value, which should be 1, being set to 0 at every alternate position in the list. This error only occurs under the "--task opt" setting. I didn't encounter this issue while running runPrimal and runAdjoint. I have attempted to debug this issue but am struggling to determine where dvDict is initialized and setDesignVars is called in the "opt" mode. |
Beta Was this translation helpful? Give feedback.
Replies: 4 comments 4 replies
-
In my tests, runScript.py does not exhibit this issue, but it has an additional bug that hinders parallel processing. While runScript_v2.py is free from the parallel bug, the beta's dvDict seems to have some problems. |
Beta Was this translation helpful? Give feedback.
-
I have successfully identified where the problem occurs.But further debugging is somewhat beyond my capabilities. |
Beta Was this translation helpful? Give feedback.
-
This is strange. Can you share your case (please minimize its size) so that we can reproduce the issue? |
Beta Was this translation helpful? Give feedback.
-
We reproduced the issue you had. It is likely to be caused by the mix of real and complex numbers in DVGeo. Instead of digging into it, I suggest you try our new API for setting non-geometry design variables (there is no need to set DVGeo, USMesh, DVCon, etc). See attached for a runScript template. It should fix the issue you have. |
Beta Was this translation helpful? Give feedback.
We reproduced the issue you had. It is likely to be caused by the mix of real and complex numbers in DVGeo. Instead of digging into it, I suggest you try our new API for setting non-geometry design variables (there is no need to set DVGeo, USMesh, DVCon, etc). See attached for a runScript template. It should fix the issue you have.
runScript_v2.py.zip