Freeze the nuclei before solving for electrons
Put the two hydrogen nuclei on the horizontal axis at and . The positive separation is in bohr. The Born–Oppenheimer approximation treats this nuclear geometry as fixed while the electronic state is solved. Repeating the calculation at many separations produces an electronic potential energy curve for nuclear motion.
The electronic Hamiltonian contains kinetic energy, attraction to both nuclei and repulsion between the two electrons. Nuclear repulsion is a separate constant at each geometry. It must be added to every electronic method before comparing total curves or differentiating them.
Reading: A. E. DePrince III · Psi4 manual
Three Gaussians make one atomic basis function
The hydrogen STO-3G basis represents each atomic function by a fixed sum of three Gaussian primitives. Their exponents and contraction coefficients are part of the basis definition. They are not extra orbital coefficients for the SCF calculation to optimize.
There is one contracted spatial function on each hydrogen, so the molecule has two spatial basis functions. An orbital is a linear combination . Electron spin is an intrinsic angular momentum; along a chosen axis we use two spin labels. Each spatial orbital has two corresponding spin functions. Thus there are four spin orbitals, and placing two electrons in them gives determinants, antisymmetric occupation states whose sign reverses when two electron coordinates are exchanged. Selecting a spin sector keeps states with specified spin properties.
Adding a larger basis changes the available spatial shapes. Mixing more determinants at a fixed basis changes how the electrons share those shapes. These are different improvements, and their errors should be inspected separately.
Reading: Basis Set Exchange · Psi4 manual
The same integrals feed every electronic method
An overlap matrix records nonorthogonality. A one-electron matrix records kinetic energy and nuclear attraction. Four-index integrals record Coulomb repulsion between pairs of basis-function products. Here the indices label atomic functions, and the vertical bar separates the two electron coordinates.
Once these arrays and the nuclear repulsion are fixed, RHF, UHF and FCI solve different trial-state problems for the same Hamiltonian. Their energy differences can then be attributed to the method. Independently changing an integral or shifting one curve would destroy that comparison.
Reading: A. E. DePrince III
Count electrons with the overlap included
For real orbitals, let , where is the spatial-orbital occupation. Expanding the density in atomic functions and integrating gives . The trace adds the diagonal entries of the matrix product. It is not generally the trace of the density matrix alone.
As a worked check, take one doubly occupied symmetric orbital with atomic overlap . Its coefficients are both , so every density-matrix entry is . The diagonal contributions sum to , and the two overlap contributions add . The electron count is two.
This numerical diagnostic tests a physical normalization in addition to checking iterative convergence. A density picture should be derived from this same matrix and these same basis functions.
Reading: A. E. DePrince III