Issue solving with more than one time independent variable

[pbrewick]

I have been using the new corrosion app. I tried to introduce displacement as a time independent variable, but I consistently get a strange error once the solve begins, shown here:

`solving...

wall time: 3.171e-06s
field 'mu' [auxiliary solve]: current solution: 5.319559e-01, current residual:1.918105e-14
field 'mupsi' [auxiliary solve]: current solution: 5.319559e-01, current residual:1.918105e-14
field 'Phi' [nonlinear solve]: current solution: 0.000000e+00, current residual:0.000000e+00
field 'Phi' [linear solve]: initial residual:0.000000e+00, current residual:0.000000e+00, nsteps:0, tolerance criterion:1.000000e-06, solution: 0.000000e+00, dU: 0.000000e+00
Relative difference between nonlinear iterations: 0 0 0
field 'irxn' [auxiliary solve]: current solution: 1.523358e+06, current residual:5.800746e-07
Relative difference between nonlinear iterations: 1.52336e+06 0 0
field ' u' [linear solve]: initial residual:1.156921e-10, current residual:1.156921e-10, nsteps:0, tolerance criterion:5.000000e-03, solution: 4.242641e-06, dU: 0.000000e+00
field 'Phi' [nonlinear solve]: current solution: 0.000000e+00, current residual:2.547818e-02

---

An error occurred in line <4322> of file </p/home/pbrewick/project/deal.ii-9.0.1/include/deal.II/matrix_free/fe_evaluation.h> in function
dealii::Tensor<1, n_components_, dealii::Tensor<1, dim, dealii::VectorizedArray > > dealii::FEEvaluationBase<dim, n_components_, Number, is_face>::get_gradient(unsigned int) const [with int dim = 2; int n_components_ = 2; Number = double; bool is_face = false]
The violated condition was:
this->gradients_quad_initialized==true
Additional information:
(none)`

You can see that the solve has already begun and that, at least initially, the code is able to solve for “u.” I spent some time searching through the PRISMS-PF users forum, but could not find any instances of the same error; and a further on-line search did not turn up any potential remedies, either.

Interestingly, if I remove the potential (Phi) as a field variable and only include displacement (u), the app works fine; the converse is also obviously true -- one can run the corrosion app as-is with just Phi. Thus, it appears as though the issue is related to having two (2) time independent variables. As the code this pertains to is related to not-yet-published work, I cannot include my files, but I believe that this issue would be fairly straightforward to re-create.

Thanks,

Patrick

[landinjm]

This may have to do with the fact that the change terms for the Newton's method are hard-coded to 0. This should produce the result above since we haven't dealt with any other linear/nonlinear fields.

phaseField/src/variableContainer/variableContainer.cc

Lines 138 to 149 in 5f4829a

	void variableContainer<dim,degree,T>::reinit_and_eval_change_in_solution(const vectorType &src, unsigned int cell, unsigned int var_being_solved){
	if (varChangeInfoList[var_being_solved].is_scalar) {
	scalar_change_in_vars[0].reinit(cell);
	scalar_change_in_vars[0].read_dof_values(src);
	scalar_change_in_vars[0].evaluate(varChangeInfoList[var_being_solved].need_value, varChangeInfoList[var_being_solved].need_gradient, varChangeInfoList[var_being_solved].need_hessian);
	}
	else {
	vector_change_in_vars[0].reinit(cell);
	vector_change_in_vars[0].read_dof_values(src);
	vector_change_in_vars[0].evaluate(varChangeInfoList[var_being_solved].need_value, varChangeInfoList[var_being_solved].need_gradient, varChangeInfoList[var_being_solved].need_hessian);
	}