electroket 0.1.0

A NetKet extension for continuous space problems, with a focus on electronic structure

electroket

electroket is an experimental extension of [NetKet](https://www.netket.org/) that adds utilities for solving continuous space problems with a focus on electronic structure calculations. It reexports NetKet so it can be used as a drop‑in replacement while providing additional modules tailored for continuous degrees of freedom.

Installation

pip install electroket

Usage

import electroket as ek

print(ek.show_version())  # prints electroket's version

Library contents

electroket exposes several modules aimed at working with particles in continuous space:

• hilbert – Hilbert spaces for particles and spinful particles in boxes or free space.
• geometry – simulation cells supporting periodic and open boundary conditions.
• molecule – helpers to define molecules and atomic properties including a small periodic table.
• operator – kinetic, potential and interaction operators for continuous systems.
• models – simple variational ansätze like Gaussian or Slater wave functions.
• sampler – Metropolis samplers with rules suitable for continuous variables.
• random – routines to generate random particle configurations.
• pyscf_utils – utilities to interface with PySCF and obtain SCF orbitals.

These modules are imported in electroket.__init__ so they can be accessed directly from the top-level package.

Examples

Below is a minimal SR-VMC calculation for the H$_2$ molecule extracted from the Examples directory.

import netket as nk
import electroket as ek
from electroket.models import MolecularSlater

h2 = ek.Molecule(
    atoms=[
        ("H", [0.0, 0.0, -0.35]),
        ("H", [0.0, 0.0, 0.35]),
    ],
    units="angstrom",
)

ham = ek.operator.KineticEnergy(h2) + ek.operator.CoulombInteraction(h2)
model = MolecularSlater(h2)
sampler = ek.sampler.MetropolisSampler(
    h2,
    ek.sampler.GaussianRule(sigma=0.1),
)

vstate = nk.vqs.MCState(sampler, model, n_samples=500)

sr = nk.optimizer.SR(diag_shift=0.05)
opt = nk.optimizer.Sgd(learning_rate=0.01)

driver = nk.driver.VMC(
    hamiltonian=ham,
    optimizer=opt,
    variational_state=vstate,
    preconditioner=sr,
)

driver.run(n_iter=1000)