asdeasoft.net

Thanks!

STKO Team wrote: ↑

Wed Aug 28, 2024 8:17 am

1) You are using a convergence tolerance that is too high, so that also wrong solutions will be accepted
2) You are using some sort of Explicit algorithm (for example IMPLEX in contact), that require small time steps

For UmfPack and Newton:
If I change the test tolerance from 0.01 to 0.001 and keep the time steps at 100, it can't converge.
If I keep the test tolerance at 0.01 and change the time steps to 1000, it converges and the result doesn't change.
If I change the test tolerance to 0.001 and change the time steps to 1000, it also can't converge.

For SparseSYM and Krylov-Newton:
If I change the test tolerance from 0.01 to 0.001 and keep the time steps at 100， it converges and the result doesn't change.
If I keep the test tolerance at 0.01 and change the time steps to 1000, it converges and the result doesn't change.
If I change the test tolerance to 0.001 and change the time steps to 1000, it has a complex root.

I think the tolerance and time steps are enough. But the difference still has.
Is there any other way to make the two results more consistent？

Hi, actually 1e-6 can be set as a strict lower limit for the tolerance if the unit system is in N-mm. Either 0.01 or 0.001 are too high! You can check it. For the explicit algorithm, you can try using a time step based on the shear wave velocity information of the model's material.

I debugged your file and it seems that the issue is due to the MinMax material.
I have to check if it's due to nature of the sudden fracture simulated by the MinMax, or a Bug in the source code.
Please, verify this on your side (simply switch off all -min -max strain limits in your MinMax wrappers), and let's see.
If that is the issue I will try to modify the OpenSees source code for that material (It is not our development so it may take a while)

Note:
Please, double-check your model. All those min-max material are really confusing. I think there is something wrong with your material settings.

STKO Team wrote: ↑

Thu Aug 29, 2024 11:09 am

Please, verify this on your side (simply switch off all -min -max strain limits in your MinMax wrappers), and let's see.

Yes, when I remove all the MinMax wrappers, the results of different solvers are the same as those obtained with UmfPack and the Newton solver before removing MinMax.

As for the results from SparseSYM and Krylov-Newton without removing MinMax, the larger computed values are likely due to the accumulation of numerical errors, which caused the material strain to exceed the MinMax limits.

Ok, for the moment you can use the hysteretic material for steel if you want to include a simple softening for fracture. We will have a look at the MinMax material as soon as possible to see if it's a bug or a pathological behavior of a sudden failure