3
$\begingroup$

I have a basic question about how to calculate molecular-orbitals from atomic-orbitals in PySCF using the self-consistent field approach. My understanding mathematically is as follows: we start with a set of non-orthonormal atomic-orbitals $\{\phi_i\}$ with overlap matrix $S_{ij} = \langle \phi_i, \phi_j \rangle$. Application of Hartree-Fock produces a new set of orbitals $\{\psi_i\}$ via a unitary orbital transformation $U$. The relationship between the old and new orbitals is: $$ \psi_i = \sum_j U_{ij} \phi_j. $$

Since $U$ is unitary, the overlap matrix should be invariant under this transformation, implying that:

$$ \langle \psi_i, \psi_j \rangle = \sum_{kl} U_{ik} U_{jl} S_{kl} = S_{ij}. $$

However, when I check this, I get that the molecular-orbitals are actually orthonormal, as their overlap matrix is the identity (see below for the code I used), and moreover, the transformation $U$ I obtain is not unitary. My question is: since the self-consistent field method should produce a unitary transformation of the orbitals, what is the approach used to obtain a non-unitary transformation. A follow-up question is: is there a benefit to orthonormalizing the MOs, rather than orthnormalizing the AOs first before obtaining the MOs?


import numpy as np
from pyscf import gto, scf

mol = gto.M(
    atom='H 0 0 0 ; Li 0 0 1',
    basis='sto-3g'
)

hf = mol.RHF()
hf.scf()
S = hf.get_ovlp()
U = hf.mo_coeff.T # this defines the transformation from AOs to MOs

S_MOs = np.einsum('ik,jl,kl->ij', C, C, S)

# S_MOs is an identity matrix
```
$\endgroup$

1 Answer 1

4
$\begingroup$

Solving the Hartree-Fock equations is equivalent to unitary rotations of the orbitals, but you are missing the key piece: before Hartree-Fock, the basis set is orthonormalized to produce an orthonormal orbital basis.

In typical programs, this happens by the process of canonical orthonormalization due to Löwdin; see section 7 of the open access review Molecules 25, 1218 (2020). PySCF does not employ this approach by default; instead, it follows an alternative but equivalent approach of solving the generalized eigenvalue problem of the Roothaan equation ${\bf FC} = {\bf SCE}$ with the eigh function of SciPy.

In either case, you will find out that the produced orbital coefficients satisfy ${\bf C}^\text{T} {\bf SC} = {\bf 1}$, that is, the orbital coefficients are orthonormal in the overlap metric. Note that ${\bf C}^\text{T} {\bf C} \neq {\bf 1}$ except in the case of an orthonormal basis, $ {\bf S} = {\bf 1}$,

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .