CamCASP/Programming/8

CamCASP => Programming => Density Fitting

We need to change our strategy. Numerical evidence suggests that the auxiliary basis should always be of the DC or DC+ types. But the routines we use to evaluate the S and T matrices needed for DF were optimized for the MC type of basis. This could means we waste resources and needs to be looked into.

T-matrix Dimer

This object is of the form: <math>

 T_{ab,k} = < \phi_a \phi_b | \chi_k >

</math> where <math>\phi_a</math> is a MO of molecule A, <math>\phi_b</math> of B and <math>\chi_k</math> is an auxiliary basis function of the dimer. At present we construct the auxiliary basis of the dimer my merging the two bases using subroutine merge_bases from module basis_operations. This seems to remove duplicate functions. So the auxiliary basis of the dimer does not contain duplicate functions.

But the main basis sets are merged without any attempt to see if basis functions are the same. So there is quite a lot of duplication with the MC+ basis type and a doubling of the basis with the DC or DC+ types. This is awful given that our integral modules are already slow.