DMET

A minimal DMET prototype based on PySCF, currently implementing:

• RHF reference state
• Lowdin-orthogonalized AO local basis
• Environment 1-RDM based bath/core orbital construction
• Fragment embedding Hamiltonian
• FCI as embedding solver
• One-shot DMET total energy estimation

This version focuses on "minimum runnable", and does not yet include DMET correlation potential self-consistency, chemical potential fitting, or more rigorous fragment energy partitioning.

Quick Start

from pyscf import gto

from dmet import RHFDMET, make_atom_fragments

mol = gto.M(
    atom="""
    H 0.0 0.0 0.0
    H 0.0 0.0 0.8
    H 0.0 0.0 1.6
    H 0.0 0.0 2.4
    """,
    basis="sto-3g",
    unit="Angstrom",
    verbose=0,
)

dmet = RHFDMET(mol)
result = dmet.kernel(make_atom_fragments(mol))

print("RHF energy:", result.mean_field_energy)
print("DMET energy estimate:", result.total_energy)
for fragment in result.fragments:
    print(
        fragment.fragment.name,
        "bath=",
        fragment.bath_dimension,
        "embed=",
        fragment.embedding_dimension,
        "E_emb=",
        fragment.embedding_energy,
    )

API

• Fragment(name, ao_indices): Manually define a fragment
• make_atom_fragments(mol): One fragment per atom
• RHFDMET(mol, mf=None): Build a one-shot DMET solver
• RHFDMET.kernel(fragments): Returns DMETResult

For minimal end-to-end usage, you can also check the test file test_dmet.py.

Development

just test

License

MIT