C06 — C/T Skeleton Active-Space Pre-screening

Field Value
Domain Chemistry
System Any molecule with MP2 natural orbitals
Group U(n) × U(n) (orbital rotation group)
H^k tier H⁰ / H¹ / H²
ISA Forge (β ≈ β*)
Status Validated
Opcodes ORBIT · TWIST · BIND · LABEL
Papers doi:10.5281/zenodo.21300668 (Paper 588), Paper 596

Physical system

Given a molecule, run a cheap MP2 calculation to obtain natural orbital occupation numbers (NOONs). Each bonding/antibonding pair (nᵢ, 2−nᵢ) is classified into one of three strata by two ISA descriptors:

  • θ_G = 2·arccos(√(n/2)) — Grassmannian angle; measures departure from a frozen (ionic/covalent) pair
  • c₂ = min(n, 2−n, ñ, 2−ñ) — Weyl chamber coordinate; measures entanglement between bonding and antibonding partners

This produces a C/T skeleton (sCeleTon): a bipartite graph over GF(2) where C-boxes (frozen pairs, θ_G < δ) are labelled 0 and T-arrows (active pairs, θ_G ≥ δ) are labelled 1. The T-arrow set is the recommended CASSCF active space.


Target category

Vect(ℝ) with graded H^k structure — the category of real vector spaces stratified by cohomological degree:

  • H⁰: C-box pairs (NOON near 0 or 2) — tropical fixed points, handled by DFT
  • : T-arrow pairs with c₂ < δ* — Clifford/CCSD regime
  • : T-arrow pairs with c₂ ≥ δ* — magic/CASSCF regime

Interpretation functor

F: C → Vect(ℝ) defined by:

Opcode F(opcode)
ORBIT Eigenvalue of occupation number operator; output = NOON nᵢ ∈ [0,2]
TWIST Berry phase between bonding/antibonding pair; captured by θ_G
BIND Non-abelian holonomy of multi-pair entanglement; triggers CASSCF
LABEL Stratum assignment: H⁰ (C-box) / H¹ / H² (T-arrow)

ISA programme

NOON:    ORBIT[nᵢ for all pairs]        -- MP2 natural orbital occupations
ANGLE:   LABEL[θ_G = 2·arccos(√(n/2))] -- Grassmannian angle per pair
WEYL:    LABEL[c₂ = min(n,2-n,ñ,2-ñ)]  -- Weyl coordinate per pair
ROUTE:   LABEL[stratum | θ_G, c₂, δ*]  -- classify C-box vs T-arrow
BETA:    LABEL[β_eff = Δ_HL/(σ·(2S+1))]-- MGE inverse temperature
ACTIVE:  ORBIT[T-arrow indices]         -- recommended active space
METHOD:  TWIST or BIND                  -- route to CCSD (H¹) or CASSCF (H²)

Routing threshold: δ* = (π/4)√α where α = β_eff · ε (correlation energy per pair). This is derived from the MGE cost-accuracy tradeoff, not fitted to data.

Computable output

  • Active space (n_elec, n_orb): list of T-arrow orbital indices and electron count for CASSCF input. Reduces to 7–36× smaller active space than full valence (Paper 588, x588g: r = −0.979 correlation with CASSCF NOONs).
  • Weyl c₂: DFT failure predictor with r = 0.990 correlation to multi-reference correlation energy (Paper 596, x596a).
  • Speedup: 7–36× wall-time reduction over manual active-space selection, with no loss in accuracy for the validated benchmark set.

Validation

  • x588g (SHA 45dbc10): Weak Lifting Theorem confirmed numerically — 10/10 molecules, r = −0.979 between C/T stratum and CASSCF NOON deviation.
  • x596a–e (Papers 596): Weyl c₂ predicts DFT error r = 0.990 across N₂, Fe²⁺, and 6-molecule benchmark; MGE soft router p continuous 0.024→0.911 along N₂ dissociation curve.
  • alchemi (PyPI): reference implementation at github.com/roguetrainer/alchemi.

Part of the ISA Zoo. Categorical foundations: Paper 591.