QEncode uses fixed suite definitions so every run is reproducible and directly comparable across teams. All v4 molecules use the cc-pVDZ basis with chemistry-driven active spaces.
Pipeline: PySCF HF → [CASSCF] → CASCI reference · PennyLane Hamiltonian · Z2 tapering · VQE (COBYLA / L-BFGS-B / Adam)
16 certified molecules. All use cc-pVDZ basis.
| Molecule | Active space | JW qubits | Tapered | Encodings | Flags | Status |
|---|---|---|---|---|---|---|
| H₂ Hydrogen | [2e, 2o] | 4 | 1 | JW PAR BK | 6 entries | |
| HF Hydrogen Fluoride | [2e, 2o] | 4 | 1 | JW PAR BK | 6 entries | |
| LiH Lithium Hydride | [4e, 4o] | 8 | 5 | JW PAR | 3 entries | |
| BeH₂ Beryllium Hydride | [4e, 4o] | 8 | 3 | JW PAR | 4 entries | |
| H₂O Water | [4e, 4o] | 8 | 4 | JW PAR | 3 entries | |
| NH₃ Ammonia | [4e, 4o] | 8 | 5 | JW PAR | 3 entries | |
| H₄ Hydrogen Chain (H₄) | [4e, 4o] | 8 | 5 | JW PAR | 4 entries | |
| N₂ Nitrogen | [6e, 6o] | 12 | 8 | JW PAR | CASSCF | 3 entries |
| H₆ Hydrogen Chain (H₆) | [6e, 6o] | 12 | 9 | JW | CASSCF | 1 entries |
| H₂CO Formaldehyde | [4e, 4o] | 8 | 4 | JW PAR | 1 entries | |
| C₄H₆ 1,3-Butadiene | [4e, 4o] | 8 | 4 | JW PAR | 1 entries | |
| (H₂O)₂ Water dimer | [4e, 4o] | 8 | 5 | JW PAR | 4 entries | |
| C₄H₄ Cyclobutadiene | [4e, 4o] | 8 | 6 | JW PAR | CASSCF | 4 entries |
| Benzene Benzene (C₆H₆) | [6e, 6o] | 12 | 9 | JW PAR | CASSCF | 2 entries |
| H₈ Hydrogen Chain (H₈) | [8e, 8o] | 16 | 13 | JW | CASSCF | 1 entries |
| H₁₀ Hydrogen Chain (H₁₀) | [10e, 10o] | 20 | 18 | JW | CASSCF | 1 entries |
Encoding exclusion notes
BK excluded (LiH, BeH₂, H₂O, NH₃, H₄, N₂, H₆, H₂CO, C₄H₆, (H₂O)₂, C₄H₄, benzene, H₈, H₁₀) — PennyLane 0.45 introduces imaginary artefacts (>7 mHa) in BK tapering for active spaces larger than [2,2]. Only H₂ and HF pass the imaginary-strip check.
PAR/UCCSD excluded (LiH, H₂O, NH₃, N₂, benzene) — UCCSD excitation operators generated in the JW basis are not correctly adapted for Parity tapering in these active spaces. BeH₂ is the exception: D∞h linear symmetry keeps the operator space well-conditioned.
CASSCF required (C₄H₄, N₂, H₆, benzene, H₈, H₁₀) — HF orbitals do not cleanly partition the active space for strongly correlated systems. CASSCF pre-optimises orbitals before the VQE circuit is built.
Maps each spin-orbital to one qubit. Locality is preserved along the qubit chain. Supported for all Suite v4 molecules.
Supported: All molecules
Encodes parity information, enabling 2-qubit reduction. Implemented via OpenFermion bridge. UCCSD operators require care — excluded for LiH, H₂O, NH₃, N₂, benzene.
Supported: All molecules (HEA); select molecules (UCCSD)
Balances locality and non-locality. Tapering verified clean only for H₂ and HF in PennyLane 0.45. Excluded for all larger molecules due to imaginary artefacts.
Supported: H₂ and HF only
Unitary Coupled Cluster Singles and Doubles
Chemically-motivated ansatz. Excitation operators are generated from the active space. High accuracy but deep circuits — parameter count scales with active space size. N₂ UCCSD has 404 parameters.
When to use: Chemically preferred. Required for certified leaderboard entries on strongly correlated molecules.
Hardware-Efficient Ansatz
Generic parameterised circuit with alternating rotation and entanglement layers. Shallow, hardware-friendly, fast to run. Layer count (reps) is configurable. Sufficient for simple molecules, insufficient for strong multireference systems like N₂.
When to use: Preferred for near-term hardware experiments. Not always sufficient for certification.
Adaptive Derivative-Assembled Problem-Tailored VQE
Builds the circuit operator by operator, selecting from the UCCSD excitation pool by parameter-shift gradient magnitude at each step. Reaches UCCSD-class accuracy with a small fraction of the parameters, keeping the optimisation tractable where full UCCSD is not.
When to use: Certifies the medium and large molecules — H₂CO, C₄H₆, C₄H₄, H₆, benzene, H₈, H₁₀ — where UCCSD with COBYLA is infeasible.
Every Suite v4 entry is generated by a single script — scripts/generate_entry_v4.py — with a fully pinned environment (requirements-v4.txt). The molecule geometry, basis set, active space, encoding, and ansatz are all predetermined by suite rules.
Each entry is identified by a compound ID such as N2_ccpvdz_JW_UCCSD_v4_casscf_tapered__sha256_82e00cea5a20cd83 and includes a SHA-256 provenance hash and Ed25519 signature for tamper detection. All entries are stored in the public GitHub repository under releases/v4/db/.
Reference energies (HF, MP2, CCSD, CCSD(T), CASCI) are computed by PySCF at the same geometry and basis. VQE gaps are always measured against E_CASCI — never against full-system FCI or a classical approximation.
