Fix to IBS force.cpp

[mrm24]

For LAPW+lo IBS is giving NaN in the case there in some grid configurations (if more than one process is assigned to the k-point group). See the following output of exciting with SIRIUS:

This patch solves it by initializing h1 and o1 (the derivatives) to 0 inside the Cartesian axis loop. The result looks like:

[toxa81]

Dear @mrm24
thank you for submitting the fix! If possible, can you please share your input file and MPI grid layout (you can send it to me to antonk@cscs.ch). When time allows, I need to have a look at them. Why I'm concerned: from the code zeroing of h1 and o1 is not required. Each element of those auxiliary matrices will be set explicitly. So while your fix is working, something is still not right

[toxa81]

cscs-ci run default

[toxa81]

I found the issue. It's the local orbital - local orbital part of the matrix has to be zeroed, because it doesn't contribute. Correct code is:

            /* zero lo-lo block */
            for (int icol = 0; icol < kp__->num_lo_col(); icol++) {
                for (int irow = 0; irow < kp__->num_lo_row(); irow++) {
                    h1(irow + kp__->num_gkvec_row(), icol + kp__->num_gkvec_col()) = la::constant<std::complex<double>>::zero();
                    o1(irow + kp__->num_gkvec_row(), icol + kp__->num_gkvec_col()) = la::constant<std::complex<double>>::zero();
                }
            }

[toxa81]

fixed in a different MR