`sp`¶

pdb2reaction sp evaluates the MLIP energy and atomic forces (optionally the full Hessian) at a single geometry. Use it for a quick energy / forces / Hessian sanity check on a structure before running an optimization, for comparing backends head-to-head, or for generating reference values and Hessians outside the optimizer loop.

Examples¶

Command form:

pdb2reaction sp -i FILE -q INT -m INT [-b uma|orb|mace|aimnet2] [--hess] [options]

Energy + forces (UMA backend, neutral closed-shell):

# energy + forces (UMA backend, neutral closed-shell)
pdb2reaction sp -i structure.pdb -q 0 -m 1

Also compute the full Hessian (UMA → analytical; other backends → finite-difference):

# also compute the full Hessian (UMA → analytical; other backends → finite-difference)
pdb2reaction sp -i structure.pdb -q 0 -m 1 --hess

Outputs¶

sp writes its outputs under result_sp/ by default. The scalar energy and |force|_max are printed to stdout; forces.npy (and hessian.npy with --hess) are always written there.

file	contents	written
stdout	scalar energy (a.u.) and `	force
`forces.npy`	`(N, 3)` array of forces in atomic units (Hartree / Bohr)	always
`hessian.npy`	`(3N, 3N)` mass-unweighted Hessian (Hartree / Bohr²)	only with `--hess`
`result.json` / `summary.json`	machine-readable energy (a.u.), backend, charge/spin, paths to npy outputs, elapsed time	only with `--out-json`

sp writes no human-readable summary.log.

Hessian backend¶

When --hess is set, the backend choice picks the Hessian computation strategy:

--backend uma (default) → Analytical Hessian via the UMA torch autograd path
--backend orb / mace / aimnet2 → falls back to FiniteDifference (no analytical Hessian implementation upstream)

--hessian-calc-mode lets you force FiniteDifference even on UMA if you want a sanity check against the analytical implementation. Forcing Analytical on a non-UMA backend has no effect — those backends always fall back to FiniteDifference.

CLI options¶

The full flag list is in the generated command reference; the table below covers the options that need explanation.

flag	default	meaning
`-i, --input FILE`	—	PDB / XYZ / GJF structure file (required)
`-q, --charge INT`	—	total charge of the system (required for non-GJF; GJF inherits the template)
`-l, --ligand-charge TEXT`	—	per-residue charge mapping (e.g. `SAM:1,GPP:-3`), used to derive `-q` automatically
`-m, --multiplicity INT`	`1`	spin multiplicity, 2S+1 (optional; defaults to 1. GJF inherits the template)
`-b, --backend [uma\|orb\|mace\|aimnet2]`	`uma`	MLIP backend selection
`--hess / --no-hess`	`--no-hess`	also compute and write `hessian.npy`
`--hessian-calc-mode [Analytical\|FiniteDifference]`	auto	force a specific Hessian mode (only applies with `--hess`)
`-o, --out-dir PATH`	`./result_sp/`	output directory
`--precision [fp32\|fp64]`	`fp32`	numeric precision passed to the backend
`--config PATH`	—	YAML config providing `calc.`, `geom.` defaults
`--out-json / --no-out-json`	`--no-out-json`	also write a machine-readable `result.json` (mirrored to `summary.json`) into the output directory
`--show-config / --dry-run`	off	print effective merged config / validate without running

Run pdb2reaction sp --help-advanced for the full list (workers, solvent corrections, etc.).

Notes¶

For single-point DFT (gpu4pyscf / PySCF) benchmarking use dft instead.

sp¶