Hi,
I'm investigating the behavior of the NPREC parameter in the MUMPS solver when running thermal non-linear analyses and I want to understand why singularity detection behaves differently between single-step and transient thermal problems.
Here is my case:
I have two nearly identical thermal models using THER_NON_LINE, please check the attached files.
If I use TIMELIST = DEFI_LIST_REEL(VALE=(0.0)) Simulation succeeds with NPREC=-1
if I use TIMELIST = DEFI_LIST_REEL(DEBUT=0.0, INTERVALLE=_F(JUSQU_A=0.0001, PAS=0.0001)) Simulation fails with NPREC=-1 (ALGELINE5_74)
║ <EXCEPTION> <ALGELINE5_74> ║
║ ║
║ You use a functionality which requires to know the degree of singularity of matrixes ║
║ associated with ║
║ linear systems. However, you decontaminated the detection of singularity with key word NPREC. ║
║ ║
║ Advices: ║
║ - Start again computation with NPREC > 0 (for example 8) under the key word factor SOLVER. ║
║ - If it is essential for you to decontaminate the detection of singularity, try to use ║
║ another linear solver, like MULT_FRONT for example. ║
Is this expected behavior? Should NPREC=-1 never be use in transient analyses? Or is this a bug I'm attaching both .comm and export file for reference (it is not possible to upload the med mesh!)
From my search this should be related to the commit 32952 in ca 16.3.25
Thanks!
Meysam
