Single-Point SCF Calculations

Computes the electronic ground state wave function of molecular systems using the Self-Consistent Field (SCF) method. SCF iteratively solves the electronic Schrödinger equation within an approximate mean-field framework, where the motion of each electron is averaged over the positions of all others. This process refines the electronic density matrix until self-consistency is achieved.

See SEQM Parameters for details on specifying the options for an SCF calculation.

The Electronic_Structure driver performs single-point ground-state SCF (and, with the right flags, excited-state) calculations on one or more molecules in a batch. Internally, it inherits from torch.nn.Module, so you can treat it like any other PyTorch model (e.g. move it to GPU, etc.).

Basic Usage

import torch
from seqm.seqm_functions.constants import Constants
from seqm.Molecule import Molecule
from seqm.ElectronicStructure import Electronic_Structure

torch.set_default_dtype(torch.float64)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

species = torch.as_tensor([[8,6,1,1],
                        [8,6,1,1],
                        [8,8,6,0]],
                        dtype=torch.int64, device=device)

coordinates = torch.tensor([
                            [
                            [0.00,    0.00,    0.00],
                            [1.22,    0.00,    0.00],
                            [1.82,    0.94,    0.00],
                            [1.82,   -0.94,    0.00]
                            ],
                            [
                            [0.00,    0.00,    0.00],
                            [1.22,    0.00,    0.00],
                            [1.82,    0.94,    0.00],
                            [1.82,   -0.94,    0.00]
                            ],
                            [
                            [0.00,    0.00,    0.00],
                            [1.23,    0.00,    0.00],
                            [1.82,    0.94,    0.00],
                            [0.0,     0.0,     0.0]
                            ]
                        ], device=device)


const = Constants().to(device)

seqm_parameters = {
'method': 'AM1',
'scf_eps': 1.0e-6,
'scf_converger': [2, 0.0],
}

molecule = Molecule(const, seqm_parameters, coordinates, species).to(device)

esdriver = Electronic_Structure(seqm_parameters).to(device)
esdriver(molecule)

print("Total Energy (eV):", molecule.Etot)
print("Forces (eV/Å):", molecule.force)

Accessing Results

After esdriver(molecule) has run, inspect the molecule attributes:

• molecule.Hf Heat of formation (eV)

• molecule.Etot Total Energy (electronic + nuclear energies), \(E_{\mathrm{tot}}\) (eV)

• molecule.Eelec Electronic Energy (eV)

• molecule.Enuc Nuclear-nuclear repulsion energy (eV)

• molecule.force Forces on atoms (eV/Å), equal to \(-\nabla E_{\mathrm{tot}}\)

• molecule.e_mo Molecular orbital eigenvalues (eV)

• molecule.e_gap HOMO–LUMO Gap (eV)

Batch Processing

SCF runs on all molecules in your batch simultaneously, so you’ll get all result attributes as batched tensors. For example, if you have 5 molecules in your input, then molecule.Etot is a length-5 tensor.