Choosing a method
Start from what you want to compute. Each row names a method that ferric
implements for that task, where to run it (CLI method.kind or Python only),
how far it is validated, and a shipped example to copy. Grades, and the anchors behind them, are on
Capabilities and validation. Cost is given as formal scaling
with system size N, not as timings.
This page only recommends what the code and its tests support. Where ferric has no validated option for a task, the row says so.
By task
| Task | Recommended | Where | Grade | Scaling | Example |
|---|---|---|---|---|---|
| Geometry optimization | KS-DFT (e.g. PBE, B3LYP) or RHF, task = "optimize" | CLI, Python (run_optimize is RHF) | Proven | N⁴ | h2-lda-opt.toml, h2_opt.toml |
| Harmonic frequencies | KS-DFT or RHF/UHF/ROHF, task = "frequencies" (analytic Hessian for RHF/UHF/ROHF with exact J/K, and for RHF, UHF and ROHF with RI-J and/or RI-K; the analytic moving-grid Hessian for closed-shell KS LDA/GGA, pure or global hybrid; both under the default hessian = "auto"; otherwise finite differences of the analytic gradient, 6N gradients) | CLI, Python run_frequencies | energies Proven; check the printed Hessian asymmetry | 6N × N⁴ | water-frequencies.toml |
| Transition state | run_saddle (P-RFO), then run_irc to confirm which minima it connects | Python only, closed shell | see Capabilities and validation | ~2(6N+1) gradients + steps | — |
| Conformer or reaction energies, routine | KS-DFT + D3(BJ) | CLI ksdft + [dft] dispersion = "d3bj", Python run_dft(dispersion=...) | DFT Proven; D3(BJ) (with or without ATM) matches simple-dftd3 to <1e-12 Ha | N⁴ (DFT) | water-pbe-d3bj.toml |
| Correlated energies, small to medium | RI-MP2 | CLI rimp2, Python run_rimp2 | Proven | N⁵ | water-rimp2.toml |
| Correlated energies, benchmark quality, small | CCSD(T) (closed shell) | Python run_ccsd_t only; CCSD also CLI ccsd | CCSD Proven; (T) matches PySCF ~1e-6 Ha on H2O/cc-pVDZ | N⁶ / N⁷ | water-ccsd.toml (H2) |
| Non-covalent interaction energies | Attenuated MP2 in the basis its parameters were fitted for: SCS-MP2(2terfc) or MP2-V in aug-cc-pVTZ, erfc att-MP2 in aug-cc-pVDZ (the basis of the 2012 paper); or KS-DFT + D3(BJ) | CLI scs-mp2-2terfc, mp2-v, att-rimp2, ksdft | att-MP2, 2terfc and MP2-V Proven (narrow) as implementations; the MP2-V parameterization is unbenchmarked | N⁵ | water-scs-mp2-2terfc.toml, water-mp2v.toml, water-attmp2.toml |
| Long-range correlation from response | RS-MP2 + LR-RPA, or PDEP-RPA | CLI rs-mp2-rpa, pdep-rpa | RS-MP2+RPA Smoke; PDEP-RPA Proven | N⁵ (MP2 part); N⁴ (RI-RPA) | water-rs-mp2-rpa.toml, water-pdep-rpa.toml |
| Ionization potentials / electron affinities | G0W0@PBE (or @HF); U-GW for open shells | CLI gw, Python run_gw, run_u_gw | Smoke, about ±0.3 eV | — | water-g0w0-pbe.toml, oh-ugw.toml |
| Excitation energies | BSE-TDA on G0W0@HF; or TDA/TDDFT (CIS/TDHF on an HF reference) | CLI bse-tda, tda, tddft; Python run_bse_tda, run_tddft | BSE-TDA Smoke; TDA/TDDFT Proven (narrow, closed shell) | — | water-bse-tda.toml, water-tda.toml |
| Static polarizability | PDEP-RPA properties, or RPAx@KS (static α only) | CLI pdep-rpa with compute_polarizability, tdhf-static-polarizability | PDEP-RPA Proven (energy); RPAx Smoke | N⁴ (RI) | h2o-pdep-rpa-props.toml, water-tdhf-static-alpha.toml |
| \( C_6 \) dispersion coefficients | PDEP-RPA dynamic α (c6_source = "pdep") or TS/MBD, in an augmented basis. Not RPAx: its \( C_6 \) is ~63% low | CLI pdep-rpa + [rpa] compute_c6 | which source is better is not established | N⁴ (RI) | water-c6-pdep.toml, argon-c6-rpa-pbe.toml |
| Implicit solvation | IEF-PCM (CLI [pcm], Python solvent=) or COSMO (CLI [cosmo]) | see SCF and DFT | both cross-checked against PySCF on water | SCF cost | — |
| Embedding in a protein or solvent | QM/MM | Python run_qmmm; CLI [qmmm] (point charges) | see QM/MM | SCF cost | water-qmmm.toml |
| Electron-transfer coupling | constrained DFT + Wu–Van Voorhis \( H_{ab} \) | Python run_cdft, cdft_coupling (no CLI) | no external reference | SCF cost × outer λ loop | — |
| Atomic charges, ESP | Löwdin, Hirshfeld, CHELPG, RESP; ESP at nuclei or on the surface | Python | see Capabilities and validation | SCF cost | — |
Notes on the table:
- "Proven" for an attenuated MP2 method means the implementation reproduces
its reference, not that the method is accurate on your system. Its
parameters are fitted to interaction-energy benchmarks in one basis (the
aug-cc-pVTZ sets without counterpoise, with frozen core); in another basis,
with counterpoise correction, or with
frozen_core = 0(the default), you are extrapolating. See The MP2 family. - KS-DFT uses density fitting by default (
def2-universal-jkfit), and the fitting error grows with system size. Turn it off or match it before comparing with an exact-Coulomb code; see SCF and DFT. - CC results are sensitive to the aux basis: at CC accuracy the RI error depends on the auxiliary set. See Coupled cluster.
Pairing an orbital basis with its auxiliary basis
Every correlated method needs an RI auxiliary basis ([mp2] auxbasis or
[rpa] auxbasis; the auxbasis argument in Python). Use the set built for
your orbital basis. The bundled pairs are:
| Orbital basis | RI (correlation) aux | JK aux (for df_j_aux / df_k_aux) |
|---|---|---|
| cc-pVDZ / TZ / QZ | cc-pvdz-rifit (alias cc-pvdz-ri), cc-pvtz-rifit, cc-pvqz-rifit | def2-universal-jkfit; cc-pvtz-jkfit / cc-pvqz-jkfit (H, B-F, Al-Cl, Ga-Br only) |
| aug-cc-pVDZ / TZ / QZ | aug-cc-pvdz-rifit, aug-cc-pvtz-rifit, aug-cc-pvqz-rifit | aug-cc-pvdz-jkfit, aug-cc-pvtz-jkfit, aug-cc-pvqz-jkfit (H, B-Ne, Al-Ar, Ga-Kr; Ne/Ar/Kr extrapolated), else def2-universal-jkfit |
| cc-pVXZ-PP, aug-cc-pVXZ-PP | <orbital>-pp-rifit (H-Ar, without Li and Na in the aug sets; Sc-Ni at TZ and QZ only; Y-Pd; Hf-Pt. K-Ni at DZ, K and Ca at TZ/QZ, Cu-Sr, Ag-Lu and Au-Rn error) | def2-universal-jkfit |
| def2-SVP(D), TZVP(P)(D), QZVP(P)(D) | def2-<x>-rifit for each | def2-universal-jkfit |
Each RIFIT set in the table also loads under ORCA's <orbital>/c name (for
example def2-tzvp/c), each JKFIT set under <orbital>/jk (for example
aug-cc-pvtz/jk), and def2-universal-jkfit as def2/jk. STO-3G, the
Pople sets and the pcseg family have no matching RI set; pair them with a
def2 or cc RIFIT set of similar quality. Any other orbital set goes in
through [basis] path as a Gaussian-94 file (orbital basis only).
def2-universal-jkfit is the default JK set for KS-DFT.
When nothing here fits
- Analytic Hessians cover RHF, UHF and ROHF (exact J/K, or RI-J / RI-K with no external potential; an ECP is fine; basis up to f functions; RHF and UHF with exact J/K also allow an external potential, ROHF allows none on any route), and closed-shell KS LDA/GGA (pure or global hybrid, moving-grid Hessian). Other KS-DFT (UKS, ROKS, meta-GGA, range-separated hybrids) and correlated methods' frequencies are by finite differences of the analytic gradient.
- Open-shell coupled cluster and open-shell TS search are not available
(the one open-shell CC-type method, LinLCCD(hh), is library-only). Open-shell
KS-DFT is available from the CLI:
ksdftwithmultiplicity > 1runs UKS, anduhf/rohfwith[dft] functionalrun UKS/ROKS. See Capabilities and validation: open shells in the CLI. - TDDFT with the XC kernel exists only as a library-only spike; the CLI and Python TDDFT omit it. See RPA, GW and excited states.