[TSI] Fix solver increment handling in partitioned solver

[rjoussen]

Description and Context

The increment vectors need to store the current states before an outer iteration. This is done in every outer_iteration_loop except for IterStagg with temperature as the coupling variable.

Consequently, the corresponding tests tsi_lincompression_iterstaggtemp.4C.yaml-p1 so far never detected that the outer iteration loop is actually converged, as can be seen by its output:

**************************************************************
      OUTER ITERATION LOOP 
      Time Step   9/2000 
**************************************************************
|    95/100    |   1.000E-06[L_2 ]      |  1.641E+02         |  1.704E+00         |
|    96/100    |   1.000E-06[L_2 ]      |  1.039E+03         |  5.085E-03         |
|    97/100    |   1.000E-06[L_2 ]      |  1.641E+02         |  1.704E+00         |
|    98/100    |   1.000E-06[L_2 ]      |  1.039E+03         |  5.085E-03         |
|    99/100    |   1.000E-06[L_2 ]      |  1.641E+02         |  1.704E+00         |
|   100/100    |   1.000E-06[L_2 ]      |  1.039E+03         |  5.085E-03         |

With this change, we get:

**************************************************************
      OUTER ITERATION LOOP 
      Time Step   9/2000 
**************************************************************
|     1/100    |   1.000E-06[L_2 ]      |  1.751E+01         |  1.704E-01         |
|     2/100    |   1.000E-06[L_2 ]      |  5.660E-02         |  5.426E-04         |
|     3/100    |   1.000E-06[L_2 ]      |  1.829E-04         |  1.754E-06         |
|     4/100    |   1.000E-06[L_2 ]      |  5.911E-07         |  5.667E-09         |

Related Issues and Pull Requests

[Copilot]

Pull request overview

Fixes convergence behavior in the TSI partitioned (iterative staggered) solver when temperature is the coupling variable by ensuring the “previous outer-iterate” state is captured before the sub-solves overwrite it. This enables convergence_check() to compute correct increments and stop the outer loop when converged (as intended by the existing tests).

Changes:

• Store the current temperature and displacement outer-iterates (tempnp, dispnp) at the start of each outer iteration for IterStagg + temperature coupling.
• Align increment-handling behavior with the other outer_iteration_loop branches that already snapshot these states prior to sub-solves.

[maxfirmbach]

Makes sense. @rjoussen FYI: Partitioned TSI is in generally in a really bad shape ...

[maxfirmbach]

@rjoussen Can you check if tsi_lincompression_iterstaggtemp.4C.yaml-p1 also runs on two processors and gives the same result?

[rjoussen]

Makes sense. @rjoussen FYI: Partitioned TSI is in generally in a really bad shape ...

Yeah that's my impression too, but it's at least conceptually doing the right things o believe. So when I want to make a material tsi-ready it's still worth to check. But yeah it needs an overhaul, just the algorithm class itself has a lot of duplicate code. Same for the thermo element's tsi capabilities