Favre-averaged stress failure on forced reset at restart

[spdomin]

Favre stress on forced reset at a restart is polluted with nans due to a divide by zero:

const double rhoAOld = (currentTimeFilter_*rhoAk - rhok*dt)/oldTimeFilter;
double uiAOld = (currentTimeFilter_*rhoAk*uiA - rhok*ui*dt)/oldTimeFilter/rhoAOld;

Above, rhoAOld is zero due to the fact that the current filter and time step are the same. Moreover, a forced reset would note that rhoAk and rhok are the same value:

const double rhok = rho[k];
const double rhoAk = rhoA[k];

with

const std::string densityAName = "density_ra_" + averageBlockName;
stk::mesh::FieldBase *density = metaData.get_field(stk::topology::NODE_RANK, "density");
stk::mesh::FieldBase *densityA = metaData.get_field(stk::topology::NODE_RANK, densityAName);

const double *rho = (double*)stk::mesh::field_data(*density, b);
const double *rhoA = (double*)stk::mesh::field_data(*densityA, b);

It is not clear to me why an initial run is sane...

[spdomin]

A clue:

 // avoid computing stresses when when oldTimeFilter is not zero
 // this will occur only on a first time step of a new simulation
 if (oldTimeFilter > 0.0 ) {
    if ( avInfo->computeFavreStress_ ) {
        compute_favre_stress(avInfo->name_, oldTimeFilter, zeroCurrent, dt, s_all_nodes);
    }
  }