Bug in m_gsphere.F90 affecting gwcomp behavior

GW, Bethe-Salpeter …

Moderators: maryam.azizi, bruneval

Posts: 3
Joined: Wed Mar 20, 2019 4:42 pm

Bug in m_gsphere.F90 affecting gwcomp behavior

Post by yow5110 » Wed Mar 20, 2019 4:55 pm

I was performing convergence studies for 2D systems using the extrapolar method (gwcomp) and found that:

playing with gwencomp at the Sigma level does increase convergence speed as promised but playing with gwencomp at the Epsilon level gave me results that neither converges nor makes physical sense..

I have traced the error back to the gsph_in_fftbox() subroutine on Line 659 of m_gsphere.F90, which is called when gwcomp is active. Looks to me that this line should be
vec(3) = n3_max(dir3)
instead of
vec(3) = n1_max(dir3)
since what this section does is finding the G-sphere that's internally tangent to the G-box (same dimensions as the FFT box).

After I fixed this, all the self energies now make sense. Previously it squashed the g-sphere along z so that the summation over GG' (in the "delta part" of the extrapolar method) is very far from converged.

I assume this was not found earlier because you would need a geometry with a large aspect ratio to see this error manifest. In my case c/a=8.

Yuanxi Wang