Julia code to compare SCF and direct minimization algorithms for the high-spin ROHF model. This code uses the PySCF python library to handle the generation of the atomic basis set and the computation of electronic integrals.

Requirements:

Julia 1.8 and above (tested up until Julia 1.10.0-rc2).

Installing all dependancies

Open a Julia shell with julia --project in your local copy of this repository and call

using Pkg; Pkg.instantiate(".")

to install all the needed dependencies. When everything is installed, you should be able to import the ROHFToolkit package with

using ROHFToolkit

Interface with PySCF

In order to guarantee that the code runs at optimal speed, it is better to use the python environment created by Conda.jl. In a fresh julia shell, install the Conda.jl with

using Pkg; Pkg.add("Conda")

Then install the pyscf library in Conda environment with

using Conda; Conda.add("pyscf")

Now open a shell in the ROHFToolkit project as in the previous section. Set up the PyCall python environment with

ENV["PYTHON"] = "/home/username/.julia/conda/3/bin/python_with_pyscf"

and call

using Pkg; Pkg.build("PyCall")

The code should be running fine, and fast.

Launch a ROHF computation

Let us compute the ground state of oxygen in a 6-31G basis, from the core Hamiltonian guess, using a preconditioned Riemannian conjugate gradient algorithm. The oxygen molecule as a PySCF object has been defined in the test_cases directory:

using ROHFToolkit
using LineSearches

include("test_cases/oxygen.jl")
x_init = ROHFState(oxygen("6-31G"), guess=:hcore)

# Launch minimization. (Note that these kwargs are the default one)
output = compute_ground_state(x_init; solver=ConjugateGradient, preconditioned=true, linesearch=HagerZhang());

# Extract final state and energy
x_min = output.final_state
e_min = output.energy

We can then generate a molden file for visualization with

generate_molden(x_min, "oxygen_min_orbitals.mol")

Contact

This is research code, not user-friendly, actively maintained, extremely robust or optimized. If you have questions contact me at: Laurent(dot)vidal(at)enpc(dot)fr