Track: DFT

Compute the SCF ground-state energy and NMR shieldings for a small molecule. The platform decides where each block of the graph runs; you decide the algorithm, the basis, and the molecule.

What ships

• starter.ipynb — H₂O at STO-3G, full preflight → submit → compare-to-PySCF workflow.
• baseline.py — PySCF RHF reference for accuracy comparison.

SDK surface you’ll use

from uniqx.domains.chemistry.basis import extract_basis
from uniqx.domains.chemistry.hartree_fock import rhf_module
from uniqx.domains.chemistry.nmr_full import nmr_full_module, scf_module

scf_module(geometry, basis_info, max_iter=N) returns a traced iterative SCF; rhf_module returns the closed-form analytic module. nmr_full_module returns SCF energy plus isotropic shieldings plus Fermi-contact J-couplings.

All four take the same six runtime inputs derived from basis_info:

runtime_inputs = [
    list(info.exps_flat),
    list(info.coeffs_flat),
    list(info.centers_flat),
    list(info.ang_flat),
    list(info.atom_coords_flat),
    list(info.charges_flat),
]

Where to push beyond the starter

Direction	What changes
Bigger basis (6-31G, cc-pVDZ)	Cost column climbs; accuracy improves; graph fans out
Bigger molecule (methane → methanol → alanine)	Gateway splits the graph into more execution blocks
Geometry optimization	Wraps SCF in an outer minimization loop
Tighten max_iter / convergence	Trades accuracy for runtime — read the max_error_rate column

Reference values

H₂O at STO-3G: RHF total energy ≈ −74.96 Ha (PySCF reference). The starter checks within 5 mHa.

H₂ at STO-3G: RHF total energy ≈ −1.117 Ha — used by the SDK’s own validation script.