Basis-set library roadmap

Reconciled against current development: 2026-10-19. The original 2026-05-08 review is historical. The pre-split basissetdev branch is not a pending integration dependency.

This roadmap tracks library coverage, not the separately versioned optimizer in vibe-basis. A bundled numerical record, an implemented Hamiltonian and scientific acceptance are three different facts.

GitLab milestone and weekly reconciliation

The development project’s Milestones page tracks this work as QC basis-library authoritative store (project milestone ID 8). The existing umbrella #295 and its focused store/provenance/cache/import issues remain the implementation and acceptance records. Missing library families are separate additions, not an unlimited release-completion requirement.

Review library coverage and tracker acceptance weekly against actual QVF records, generated renderings, resolver/import source and accepted checks. Use the main roadmap’s weekly review policy: implementation, qualification, release readiness and tagged delivery are different states. Advance QC’s feature target when the selected criteria pass; the library has no invented independent version line. The optimizer’s separate release milestones stay in vibe-basis.

Current implementation and priorities

• QVF JSON records in python/vibeqc/basis_library/qvf/ are authoritative; .g94 and .ecp files are generated renderings. The library guide documents conversion, provenance, citation routing and rendering checks.

• Molecular on-demand BSE lookup ships through the optional [bse] extra and run_job(fetch_from_bse=True). It reads the installed BSE catalogue; it is not a missing HTTP downloader. Periodic run_periodic_job does not expose that keyword.

• ECP integration, the pc/pcseg families, diffuse Pople variants, modified 3c bases, vDZP, LANL and the listed PP orbital families exist. They are not waiting on libecpint or on a new branch merge.

• The named user basis store, provenance records and resolver are implemented. Remaining tracker acceptance must be distinguished from missing source. Do not edit rendered files as the primary data or add a second store.

• Remaining work is targeted missing coverage, per-element provenance/parity, relativistic Hamiltonians for the all-electron relativistic families, and auxiliary-basis research. Automatic Stoychev-style AutoAux remains a proposal; a citation entry or a registry-selected fitting basis is not its implementation.

Cross-cutting prerequisites

Prerequisite	Current state	Remaining limitation
ECP integrals	Implemented through libecpint; basis/ECP records are loaded together	Per-element accuracy and route coverage, not initial integration
Scalar-relativistic X2C / DKH2 / ZORA	Future method work	Meaningful use of the corresponding all-electron recontracted sets
GPW/GAPW	Gaussian plus auxiliary plane-wave routes exist; GAPW remains experimental	MOLOPT/GTH pairing and route qualification are separate from merely adding a basis file
Pure-PW / PAW / ONCV / NAO	Not the current Gaussian library architecture	No implied commitment to import incompatible library families
NMR shielding / J coupling	Future property work	A bundled pcS/pcJ basis does not implement the observable
AutoAux and bespoke pob JK fits	Research / missing generation workflow	Published fit data, generation algorithm and numerical qualification

The table below preserves the review’s priority recommendations. ✅ means a matching bundled orbital record is present, not full element coverage or a validated method; ❌ means no matching bundled record was found; 🚫 identifies a missing method or research prerequisite. Optional BSE catalogue availability is separate from bundled coverage. Historical counts of “45 missing” are removed because aliases and later additions made them unreliable.

The big table

Everything the review recommends, organised by family. “BSE name” is the catalogue identifier, not necessarily the bundled file stem. Use the current BSE loader or basis conversion tools rather than the old proposed fetch script.

Pople family

Name	BSE id	Status	Priority	Notes
STO-3G	STO-3G	✅	low	Pedagogy / SCF guess. UI label: “for guess only”.
STO-6G	STO-6G	✅	low	Same role as STO-3G; slightly better atomic energies.
MINI	MINI (Huzinaga)	✅	low	Niche.
3-21G	3-21G	✅	low	UI label: “unpolarised, avoid for production”.
6-31G	6-31G	✅	low	Same UI label as 3-21G.
6-31G(d)	6-31G*	✅	medium	Legacy default; keep.
6-31G(d,p)	6-31G**	✅	medium	Same.
6-31+G(d,p)	6-31+G**	✅	high	Diffuse-aug; 6-31+g**.g94 shipped.
6-31++G(d,p)	6-31++G**	✅	high	Best DZ basis per Pitman 2024; 6-31++g**.g94 shipped.
6-311G(d,p)	6-311G**	✅	medium	UI label: “polarised 6-311G is poorly parameterised, prefer pcseg-2 / def2-TZVP” (Pitman 2024).
6-311+G(d,p)	6-311+G**	✅	medium	Diffuse variant; 6-311+g**.g94 shipped.
6-311+G(2d,p)	6-311+G(2d,p)	✅	low	6-311+g(2d,p).g94 shipped.
6-311+G(2df,p)	6-311+G(2df,p)	❌	low	Same, still missing.

Karlsruhe def2 family

Name	BSE id	Status	Priority	Notes
def2-SV	def2-SV	✅	medium
def2-SV(P)	def2-SV(P)	✅	medium
def2-SVP	def2-SVP	✅	high	Default DZ.
def2-SVPD	def2-SVPD	✅	medium
def2-TZVP	def2-TZVP	✅	highest	Production default.
def2-TZVPD	def2-TZVPD	✅	high
def2-TZVPP	def2-TZVPP	✅	high
def2-TZVPPD	def2-TZVPPD	✅	high	Inorganic main-group default.
def2-QZVP	def2-QZVP	✅	high	Benchmark DFT.
def2-QZVPP	def2-QZVPP	✅	high
def2-QZVPPD	def2-QZVPPD	✅	medium
ma-def2-TZVP	ma-def2-TZVP	❌	high	Truhlar minimally-augmented.
ma-def2-TZVPP	ma-def2-TZVPP	❌	medium
ma-def2-SVP	ma-def2-SVP	❌	medium
def2-universal-J	(built into def2-J)	✅	high	aux-fit, Coulomb
def2-universal-JK	(built into def2-JK)	✅	high	aux-fit, Coulomb+exchange

Karlsruhe def2, relativistic (blocked on rel. Hamiltonian)

Name	BSE id	Status	Priority	Notes
x2c-SVPall	x2c-SVPall	🚫	high	Pollak-Weigend 2017. Blocked on X2C / DKH.
x2c-TZVPall	x2c-TZVPall	✅	high	Same.
x2c-QZVPall	x2c-QZVPall	🚫	high	Franzke 2020.
x2c-TZVPall-s	x2c-TZVPall-s	✅	medium	NMR-tuned.
dhf-SVP	dhf-SVP	✅	medium	Dirac-Hartree-Fock-derived def2-equivalent; dhf-svp.g94/.ecp shipped.
dhf-TZVP	dhf-TZVP	✅	medium	dhf-tzvp.g94/.ecp shipped.
dhf-QZVP	dhf-QZVP	✅	low	dhf-qzvp.g94/.ecp shipped.
def2-TZVP-DKH	def2-TZVP-DKH	🚫	medium	DKH-recontracted.

Karlsruhe def2, modified for 3c (bundled)

Name	BSE id	Status	Priority	Notes
def2-mSVP	def2-mSVP	✅	high	PBEh-3c, B3LYP-3c carrier.
def2-mTZVP	def2-mTZVP	✅	high	B97-3c carrier.
def2-mTZVPP	def2-mTZVPP	✅	high	r²SCAN-3c carrier.

Dunning correlation-consistent

Name	BSE id	Status	Priority	Notes
cc-pVDZ	cc-pVDZ	✅	high
cc-pVTZ	cc-pVTZ	✅	high	Post-HF default.
cc-pVQZ	cc-pVQZ	✅	high	CBS extrapolation.
cc-pV5Z	cc-pV5Z	✅	medium
cc-pV6Z	cc-pV6Z	✅	low
cc-pV7Z	cc-pV7Z	✅	low
aug-cc-pVDZ	aug-cc-pVDZ	✅	high
aug-cc-pVTZ	aug-cc-pVTZ	✅	high	Anions, response.
aug-cc-pVQZ	aug-cc-pVQZ	✅	medium
aug-cc-pV5Z	aug-cc-pV5Z	✅	medium
cc-pV(D+d)Z	cc-pV(D+d)Z	✅	medium	“tight d” for hypervalent S/P.
cc-pV(T+d)Z	cc-pV(T+d)Z	✅	medium
cc-pV(Q+d)Z	cc-pV(Q+d)Z	✅	medium
aug-cc-pV(T+d)Z	aug-cc-pV(T+d)Z	✅	medium
cc-pCVDZ	cc-pCVDZ	✅	medium	Core-valence.
cc-pCVTZ	cc-pCVTZ	✅	medium	Core-valence.
cc-pCVQZ	cc-pCVQZ	✅	medium	Core-valence.
aug-cc-pCVTZ	aug-cc-pCVTZ	✅	medium	Core-valence.
cc-pwCVTZ	cc-pwCVTZ	✅	low	Weighted core-valence.
cc-pVTZ-F12	cc-pVTZ-F12	✅	medium	Explicit-correlation.
cc-pVQZ-F12	cc-pVQZ-F12	✅	medium

Dunning pseudopotential-adapted (Peterson/Figgen small-core PPs)

Shipped 2026-09-08. Before that the library carried sixteen -PP-RIFIT fitting sets and zero PP orbital sets: auxiliary bases for orbital bases the registry would refuse to load. Each orbital set covers 39 elements (Cu-Kr, Y-Xe, Hf-Rn) and carries its ECP blocks in the same BSE file; scripts/setup_basis_library.sh splits them into a .g94 + .ecp pair.

Note the coverage asymmetry that remains, by design: the RI fits are narrower than the orbital sets (15 elements for cc-pV*Z-PP-RIFIT, 21 for cc-pwCV*Z-PP-RIFIT, 8 for aug-cc-pwCV*Z-PP-RIFIT). An RI job on an element the fit omits is refused by the aux coverage guard naming the element (#480). No JK fit is published for the family at all.

Name	BSE id	Status	Priority	Notes
cc-pVDZ-PP	cc-pVDZ-PP	✅	high	Small-core PP; heavy-element orbital set.
cc-pVTZ-PP	cc-pVTZ-PP	✅	high
cc-pVQZ-PP	cc-pVQZ-PP	✅	medium
cc-pV5Z-PP	cc-pV5Z-PP	✅	low
cc-pwCVDZ-PP	cc-pwCVDZ-PP	✅	medium	PP core-valence.
cc-pwCVTZ-PP	cc-pwCVTZ-PP	✅	medium
cc-pwCVQZ-PP	cc-pwCVQZ-PP	✅	low
cc-pwCV5Z-PP	cc-pwCV5Z-PP	✅	low
aug-cc-pVDZ-PP	aug-cc-pVDZ-PP	✅	high	PP + diffuse.
aug-cc-pVTZ-PP	aug-cc-pVTZ-PP	✅	high
aug-cc-pVQZ-PP	aug-cc-pVQZ-PP	✅	medium
aug-cc-pV5Z-PP	aug-cc-pV5Z-PP	✅	low
aug-cc-pwCVDZ-PP	aug-cc-pwCVDZ-PP	✅	medium
aug-cc-pwCVTZ-PP	aug-cc-pwCVTZ-PP	✅	medium
aug-cc-pwCVQZ-PP	aug-cc-pwCVQZ-PP	✅	low
aug-cc-pwCV5Z-PP	aug-cc-pwCV5Z-PP	✅	low

Jensen polarization-consistent (segmented + general)

Name	BSE id	Status	Priority	Notes
pc-0	pc-0	✅	medium	Unpol DZ.
pc-1	pc-1	✅	medium	DZ-pol.
pc-2	pc-2	✅	medium	TZ-pol.
pc-3	pc-3	✅	low	QZ-pol.
pc-4	pc-4	✅	low	5Z-pol.
pcseg-0	pcseg-0	✅	high	Segmented contraction.
pcseg-1	pcseg-1	✅	highest	DFT-optimal DZ.
pcseg-2	pcseg-2	✅	highest	DFT-optimal TZ, best on diet-GMTKN55 (Pitman 2024).
pcseg-3	pcseg-3	✅	medium
pcseg-4	pcseg-4	✅	low
aug-pcseg-1	aug-pcseg-1	✅	high
aug-pcseg-2	aug-pcseg-2	✅	high	TDDFT, anions.
pcS-0…3	pcS-{0,1,2,3}	✅	medium	pcs-0 through pcs-3 data bundled; NMR shielding kernel remains open.
pcJ-0…3	pcJ-{0,1,2,3}	🚫	medium	NMR J-coupling.

vDZP (Grimme 2023, bundled)

Name	BSE id	Status	Priority	Notes
vDZP	vDZP	✅	high	ωB97X-3c carrier; uses ECPs; vdzp.g94/.ecp shipped.

ANO families

Name	BSE id	Status	Priority	Notes
ANO-RCC (full)	ANO-RCC	✅	high	Generally contracted.
ANO-RCC-MB	ANO-RCC-MB	✅	low	Minimal.
ANO-RCC-VDZP	ANO-RCC-VDZP	✅	medium	Specific contraction.
ANO-RCC-VTZP	ANO-RCC-VTZP	✅	high	Multireference TM standard.
ANO-RCC-VQZP	ANO-RCC-VQZP	✅	medium
ANO-R0	ANO-R0	✅	medium	Zobel-Widmark-Veryazov 2020.
ANO-R1	ANO-R1	✅	medium
ANO-R2	ANO-R2	✅	medium
ANO-R3	ANO-R3	✅	low

LANL ECP (deprecated by review, ship for compatibility)

Name	BSE id	Status	Priority	Notes
LANL2DZ	LANL2DZ	✅	low	UI label: “deprecated, use def2-TZVP + def2-ECP”. ECP integration is available.

Periodic Gaussian

Name	Source	Status	Priority	Notes
pob-TZVP	Bredow tarball	✅	highest	Verified byte-clean f059b37.
pob-DZVP-rev2	Bredow scrape	✅	high	Verified.
pob-TZVP-rev2	Bredow tarball	✅	highest	Verified. S d-shell bug fixed.
pob-DZVP (original)	Bredow archive	❌	medium	Predecessor of rev2; for reproducibility.
pob-DZVPP	Bredow archive	❌	medium	Polar augmentation.
pob-TZVPP	Bredow archive	❌	medium	Polar augmentation.
pob-TZVP-rev2 (Rb-I, fifth period)	Bredow tarball	✅	high	ECP-backed fifth-period records are present; precision/parity remains per element.
pob-TZVP-jk	(to design)	🚫	high	Goal 5 of basissetdev.
MOLOPT-SZV	CP2K source	🚫	medium	Pairing and route qualification remain; GPW infrastructure exists.
MOLOPT-DZVP	CP2K source	🚫	medium
MOLOPT-TZVP	CP2K source	🚫	medium
MOLOPT-TZV2P	CP2K source	🚫	medium
dcm-TZVP	Daga/Civalleri/Maschio 2020	❌	low	System-specific (diamond, graphene, carbyne).

Property-specific (data and property implementation are separate)

Name	BSE id	Status	Priority	Blocks
EPR-II	EPR-II	🚫	medium	EPR g-tensor / hyperfine kernel
EPR-III	EPR-III	🚫	medium	Same
Sadlej-pVTZ	Sadlej pVTZ	✅	medium	Polarizability, kernel exists
LPolX (Maroulis)	(BSE varied)	❌	low	Same
ccX-DK	ccX-DK	❌	low	Core-level X-ray

Plane-wave / NAO / LAPW (out of scope today)

The review recommends VASP-PAW, PseudoDojo (ONCV), SSSP, GBRV, SG15, JTH PAW, FHI-aims tier 1-4, SIESTA NAO, and (L)APW+lo. vibe-qc is a Gaussian / atom-centred-orbital code today; none of these libraries fit the current architecture. The roadmap captures them under “deferred to a future periodic-PW chat”:

• PseudoDojo (ONCV) is the recommended starting point if / when vibe-qc grows a plane-wave path. Open-source, well-curated, has scalar and fully-relativistic variants.

• SSSP precision is the natural complement for high-throughput materials work.

• PAW (VASP / GPAW / ABINIT JTH) is a substantial implementation lift via the projector formalism; defer until ONCV is stable.

• FHI-aims tier-1/2 NAO would require a numerical-orbital integral path entirely separate from libint.

Inventory and maintenance

The library README, registry and QVF records define the live inventory. Check both element coverage and ECP pairing before using a family. Adding a record does not close a numerical acceptance issue, and a pending acceptance issue does not make existing data absent.

1. Select a genuinely missing record or coverage gap, checking the bundle and optional BSE catalogue first.

2. Convert/import into the existing QVF record system; preserve upstream identity, numerical content hash and per-element citations.

3. Render through scripts/render_basis_library.py or the supported setup script and check record/rendering parity. Keep ECP blocks with the correct orbital family and refuse uncovered elements.

4. Validate the affected numerical use before making new accuracy claims.

5. Keep auxiliary-basis design and production optimized-basis campaigns in their own milestones; they do not hold unrelated library additions.

Cross-references

• PLAN.md, top-level basissetdev plan; this roadmap is rolled in as a new long-term goal slot.

• REQUIREMENTS-PERIODIC.md, historical design record to the periodic chat; the periodic-Gaussian items above partly depend on R1-R5 there.

• GOAL4_DESIGN.md, basis-optimisation engine design; note: the optimiser will eventually generate basis sets that themselves go into this library (Goal 5: bespoke pob-*-jk; Goal 6: periodic metals basis).