FieldMatrix::solve method should be used. The code assumed that
all diagonal elements were nonzero before truncation. In this case
any zero diagonal element was due to truncation and hence setting
it to one was alright.
Now instead, we do not assume nonzero diagonal elements. We check
for truncation of diagonal elements instead (and assert that they
were set to zero before in the truncation step). Then setting these
entries to one (but not the ones which were zero a-priori) is alright
and we do not solve a modified system.