[BUG] TSI `output()` writes output without calling `new_step()`

[rjoussen]

Bug Description: What's going wrong?

TSI::Algorithm::output() mainly writes output for the thermo field and the structure field:

thermo_field()->output(forced_writerestart);
structure_field()->output(forced_writerestart);

Besides that, TSI::Algorithm::output() also writes some cross-outputs, i.e., the displacement into the thermo field and the temperature into the structure field using the old output format, applying the mortar mapping if necessary. For that, it calls the disc_writers of the respective fields directly. The problem is that it does so regardless of whether or not the output methods of the individual fields actually wrote restart/old output and hence called new_step() already. If new_step() was not called before, this fails with a HDF5 error.

Expectations

It shouldn't fail.

Possible Solution

There are 3 possible solutions:

• TSI::Algorithm::output() needs to determine if the individual fields called new_step already/have datawritten. Since the output methods don't and shouldn't return anything, this would require extensive checks based on the input parameters.
• Since the faulty cross-outputs are not needed for restarts, we could move the output of temperature in the structure field and displacement in the thermo field to runtime vtk output. The problem with that is that the vtk writers of the individual fields are not available in this method and would need to be callable from the adapters, I assume
• Remove this cross-output altogether. This would make TSI_Algorithm::output() much slimmer (just calling output of the fields). Displaying the temperature on the deformed geometry for matching grids is trivial in Paraview; for non-matching grids, it can probably be interpolated somehow.

The last two approaches would also mean that we can get rid of the TSI specific old post_monitor writers, which would be helpful to afterwards remove the TermoWriter (since the TSI writer inherits from the thermo writer)

Steps to Reproduce

This behaviour can be reproduced for instance by modifying the volmortar3D_tsi_error.4C.yaml test:

IO:
  STRUCT_DISP: false # false doesnt work
  THERM_TEMPGRAD: no # no doesnt work
TSI DYNAMIC:
  RESTARTEVERY: 0 # this leads to no restart output being written, hence no new_step

Your Environment

• Relevant repo SHA1s:
• Relevant configure flags or configure script:
• Operating system and version:
• Compiler and TPL versions:

Related Issues and Pull Requests

• Blocks [TSI] Use WRITE_INITIAL_STATE #1678 (since no restart output is written at step=0 and volmortar3D_tsi_error.4C.yaml doesn't set IO/THERM_TEMPGRAD to not no.
• Follows
• Related to
• Part of
• Composed of

Interested Parties

fyi @c-p-schmidt

[maxfirmbach]

@rjoussen My question might be naive, but are there other fields that do something similar e.g. fsi? This is of course not volume coupled, but many of the "mechanisms" in 4C are very similar 😃. Maybe this can also help to find a proper solution to this issue.

[rjoussen]

@maxfirmbach Yes, I looked into that briefly, and this would really support removing the "cross-outputs" (temperature in the structure field and displacement in the thermo field) altogether. No other output method of other coupled problems does that currently.

Here is a quick overview of what I found:

• FSI

void FSI::Algorithm::output()
{
  // Note: The order is important here! In here control file entries are
  // written. And these entries define the order in which the filters handle
  // the Discretizations, which in turn defines the dof number ordering of the
  // Discretizations.
  structure_field()->output();
  mb_fluid_field()->output();
}

• FPSI

void FPSI::MonolithicBase::output()
{
  poro_field()->output();
  fluid_field()->output();
  ale_field()->output();
}

• SSI

void SSI::SsiMono::output()
{
  // output scalar transport field
  scatra_field()->check_and_write_output_and_restart();
  if (is_scatra_manifold())
  {
    // domain output
    scatra_manifold()->check_and_write_output_and_restart();
    // coupling output
    if (manifoldscatraflux_->do_output()) manifoldscatraflux_->output();
  }

  // output structure field
  structure_field()->output();

  if (print_matlab_) print_system_matrix_rhs_to_mat_lab_format();
}

void SSI::SSIPart2WC::update_and_output()
{
  constexpr bool force_prepare = false;
  structure_field()->prepare_output(force_prepare);

  if (ssi_interface_contact())
  {
    // re-evaluate the contact to re-obtain the displ state
    const auto& model_eval = structure_field()->model_evaluator(Inpar::Solid::model_structure);
    const auto& cparams = model_eval.eval_data().contact_ptr();
    nitsche_strategy_ssi()->integrate(*cparams);
  }

  structure_field()->update();
  scatra_field()->update();

  scatra_field()->evaluate_error_compared_to_analytical_sol();

  structure_field()->output();
  scatra_field()->check_and_write_output_and_restart();
}

• STI

void STI::Algorithm::output()
{
  scatra_->scatra_field()->check_and_write_output_and_restart();
  thermo_->scatra_field()->check_and_write_output_and_restart();
}

• SSTI

void SSTI::SSTIMono::output()
{
  scatra_field()->check_and_write_output_and_restart();
  thermo_field()->check_and_write_output_and_restart();
  structure_field()->output();
}

[maxfirmbach]

Alright, for consistency reasons I would also prefer to keep it in the style of the other coupled fields. I can't comment however on how important the respective output for tsi is.

[rjoussen]

I would assume that no one did a TSI simulation with a non-matching grid and postprocessed the output with the old postprocessor for a while. Honestly, I don't even know how that works. I will look into how it could be removed (together with the corresponding post-processing code).