Open cortner opened 6 years ago
If we just integrate over the BZ, then the expression we get is something like (if I'm not mistaken)
G(x) - G(0) = \int_BZ g(k) sin^2(k . x / 2). dk
This is analytic away from k = 0. so how about splitting it into two integrands
\int_BZ f_1(x; k) + f_2(x; k)
where f_1 is supported in a ball B_r(0). and f_2(x; k) is C^inf in BZ. Then we can integrate f_2 using a standard uniform grid (MP) an we can integrate f_1 by transforming it to spherical coordinates -> that should completely remove the singularity?
@jjbraun no rush at all, but if you manage to check whether your surface integral formulation could be made to work, can you put a brief summary here?
A quick remark:
how about splitting it into two integrands
is basically Dallas' approach to computing the LGF.
Collecting different options how to implement LGFs.