It seems that the symmetrization of the force-constant matrix is performed with a single precision accuracy.
In particular, after symmetrization dyn mat at the edge of the brilluin zone (that should be real as q = -q + G) have a small nonvanishing imaginary part of the order of 10^-7, which is single precision accuracy.
Probably this arises in the subroutine subroutine symdynph_gq_new of same named file, in the symmetrization part.
This error is particularly cumbersome when ibrav=0 occurs (probably due to numerical precision in reading the coordinates), but even after manual symmetrization a small 10^-7 residual error persists.
This needs to be inverstigated further.
It seems that the symmetrization of the force-constant matrix is performed with a single precision accuracy.
In particular, after symmetrization dyn mat at the edge of the brilluin zone (that should be real as q = -q + G) have a small nonvanishing imaginary part of the order of 10^-7, which is single precision accuracy.
Probably this arises in the subroutine
subroutine symdynph_gq_newof same named file, in the symmetrization part.This error is particularly cumbersome when ibrav=0 occurs (probably due to numerical precision in reading the coordinates), but even after manual symmetrization a small 10^-7 residual error persists.
This needs to be inverstigated further.