does only write grids, without any data.
-Look at this code from error_estimator.cc:
- vector<vector<Tensor<1,dim> > > tmp (n_components,
- neighbor_psi);
- fe_face_values_neighbor.get_function_grads (solution, tmp);
- neighbor_psi.swap (tmp[selected_component]);
- I believe it is wrong in that it assumes that the outermost index
- in tmp runs over the components, but get_function_grads returns it
- the other way round, so it should only work if
- n_components==n_q_points. there are several such places, fix them.
+In 'KellyErrorEstimator' add new argument 'const vector<double> weight_
+components' for using the error indicators of all components of a
+system in a linear combination. This vector must have the size of the
+number of components and includes the weights. This could replace the
+old argument 'const unsigned int selected_component'.