Currently:
At each integration grid point, rho = sum_ab{c_a*D_ab*c_b) with a and b running
over all basis functions. This scales badly (~N_gridpoints x N_basis^3).
To be done:
Can be made more efficient by exploiting the fact that basis functions are
localized at atoms and have only a finite extent. Shells (a) outside a certain
cutoff of the grid point [fix version], or (b) whose functions' values at the
grid point fall below a threshold can be removed from the summation. Then
N_basis^eff << N_basis per grid point. (plus extra tricks)
Original issue reported on code.google.com by bjoern.b...@gmail.com on 26 Oct 2014 at 6:45
Original issue reported on code.google.com by
bjoern.b...@gmail.comon 26 Oct 2014 at 6:45