"use_exp_form = true" flag often causes simulations to fail

[smpark7]

The use_exp_form = true flag allows users to track neutron flux values as exponent values instead of raw values (e.g. e^10 -> 10). Refer to moltres/src/base/ScalarTransportBase.C for the source code.

However, this flag requires careful use because it easily triggers the DIVERGED_FNORM_NAN error on nonlinear solve iterations. The error message does not indicate that use_exp_form is responsible for the failed solve.

I suspect that it triggers the error because Moltres tries to evaluate exp(x) and the value can easily exceed the maximum representable float/double.

This issue is more of a PSA for future users who encounter the DIVERGED_FNORM_NAN with this flag, which is true by default.

[gridley]

I think deprecating use_exp_form would be worth considering. It adds a pretty large expense to have to compute logarithms/exponentials at each quadrature point. In fact, I believe it would speed moltres up fairly to simply eliminate the computeQpConcentration. I can't think of any problem where this approach is actually necessary. Maybe there is some problem where this is required in plasma problems, but there are no neutronics problems documenting this in the literature which require this approach. Its accuracy is also debatable considering that discretization error is much larger than roundoff error in any case Moltres runs.

[smpark7]

Hi @gridley . Yea I also can't think of cases when computeQpConcentration would be useful. MOOSE provides better alternatives for scaling. I'll probably make a PR for this when I have the time.