Correlation as a distance on the Grassmannian
A single angle — the geodesic distance from the classical fixed point on the Grassmannian — tracks the onset of strong correlation. Computed in two domains, conjectured in others, and a reformulation of a standard criterion rather than a replacement for it.
The claim
Every system with $k$ correlated degrees of freedom embedded in an $n$-dimensional ambient space has a natural home: the Grassmannian $\mathrm{Gr}(k, n)$, the manifold of $k$-dimensional subspaces of $\mathbb{R}^n$ (or $\mathbb{C}^n$). The Grassmannian carries the Fubini-Study metric, which measures the angle between two subspaces.
Define
\[\theta_G = \arccos(\sigma_0)\]where $\sigma_0$ is the leading singular value of the system’s state matrix — the largest overlap between the correlated state and its closest single-reference (classical) approximation. Then:
- θ_G = 0°: the system is entirely classical — one configuration dominates
- θ_G = 45°: the system is a perfect biradical / maximally entangled / fully correlated in two directions
The claim is that a threshold near θ_G ≈ 20° separates the weakly correlated regime (classical simulation efficient) from the strongly correlated regime (classical simulation fails). It has been computed in two domains and is conjectured, not measured, in the other two — the table marks which:
| Domain | System | What θ_G measures | Crossing | Status |
|---|---|---|---|---|
| Chemistry | Molecular wavefunction | Distance from Hartree-Fock reference | Bond breaking, R ≈ 1.5 Å | computed (pyscf) |
| Condensed matter | Hubbard model | Deviation from mean-field ground state | Mott transition, U/t ≈ 1.8 | computed (pyscf) |
| Quantum error correction | Code subspace | Distance from nearest Pauli-error image | Fault-tolerance threshold p* | conjectured — never computed |
| Finance | Factor subspace | Distance from prior-period factor structure | Systemic crisis onset | conjectured — never computed |
Only the first two rows are measurements. θ_G has not been evaluated for a stabiliser code subspace or for an empirical factor structure; those rows state what the construction would mean, not what it returns. A four-domain universality claim rests on two domains.
The threshold is not fitted separately in each domain. It emerges from the same geometric condition: the leading Schmidt singular value dropping below $\sigma_0 \approx \cos(20°) \approx 0.94$, so that $\sigma_0^2 \approx 0.88$ — the point at which no single configuration accounts for more than 88% of the total weight.
Why it matters
Different fields measure “how correlated is this system?” differently: quantum chemists use diagnostics like the T1 amplitude norm and natural-orbital occupations, condensed matter physicists use U/t, QEC theorists use the fault-tolerance threshold p*, and risk managers use factor-model diagnostics. The Grassmannian picture proposes that these are projections of one geometric quantity — the distance from a classical fixed point to the true correlated state in the Fubini-Study metric on $\mathrm{Gr}(k,n)$.
An important qualification. In the chemistry case θ_G is a monotone function of the natural-orbital occupation numbers. It therefore orders systems exactly as occupation-number thresholding does, and flags the same molecules at the same geometries. It is a reformulation of a criterion the field already uses, in coordinates that happen to generalise, not a more sensitive diagnostic. Anyone adopting it should expect the same answers, more portably expressed — not better ones.
The value, if there is one, is portability: the same SVD applies to any state matrix. Whether that portability yields anything a domain specialist wants is the open question, and the two unmeasured rows above are where it would be settled.
The evidence
Chemistry: molecular bond breaking (Papers 563, 570)
SA-CASSCF calculations on H₂, H₂O, N₂ and benzene put the single-reference breakdown near θ_G ≈ 20° in each case. The crossing occurs at bond lengths where CCSD(T) degrades and multi-reference treatment becomes necessary.
Four closed-shell molecules is a small and homogeneous sample, and “universal across all molecules” is not what it supports. Open-shell systems, transition-metal complexes and excited states are untested.
The Hubbard model (1D, half-filling) crosses θ_G ≈ 20° at U/t ≈ 1.8, the Mott metal-insulator transition — the same threshold, on the condensed-matter side.
Quantum error correction
A [[n,k,d]] stabiliser code is a point $p_C \in \mathrm{Gr}(2^k, 2^n)$. The code distance d is the Fubini-Study distance from $p_C$ to the nearest Pauli-error image. The fault-tolerance threshold p* is the β* snap: the noise level at which this geodesic distance collapses to zero. The ISA chain complex (Paper 571) makes ∂² = 0 (stabilisers commute) tautological — it is the chain complex condition.
Nuclear physics
The deuteron’s S/D mixing angle θ_G ≈ 13° — determined by the tensor force from one-pion exchange — is the nuclear analogue of the bond-breaking angle in H₂. The alpha particle (4-nucleon FUSE orbit) sits at large θ_G, which is why alpha decay is the dominant heavy-nucleus instability. The Hoyle state of ¹²C is a two-level ISA system at the H²/H² interface.
Finance (Paper 580)
Systematic risk factor subspaces are points in $\mathrm{Gr}(k, n)$ (k factors, n assets). The Gaussian copula placed the 2008 CDO market at a point in $\mathrm{Gr}(1, n)$ while the true crisis structure required $\mathrm{Gr}(3, n)$ — a distance of 66°, well past any sensible β* threshold. The subspace velocity θ_G(t) = d_FS(p_t, p_{t-1}) began rising four quarters before Lehman.
The geometric picture
The same three-tier H^k structure governs all four domains:
| ISA tier | Opcode | Chemistry | QEC | Finance |
|---|---|---|---|---|
| H⁰ | RESOLVE | HF fixed point (σ₀ = 1) | Stabiliser code | Single-factor model |
| H¹ | TWIST | Mild correlation (θ_G < 20°) | Mild errors, correctable | Model risk accumulating |
| H² | FUSE | Strong correlation (θ_G > 20°) | Threshold failure | Systemic snap |
The β* snap is the geometric transition where the H⁰ fixed point loses stability. Below it, the leading singular value σ₀² > 0.88 and the system is well-approximated by a single dominant configuration. Above it, no single configuration dominates and the system enters the strongly correlated / H² regime.
The snap is sharp — not a smooth crossover — because the Grassmannian is a compact manifold and the distance function d_FS = arccos(σ₀) has zero derivative at σ₀ = 1 (near the classical fixed point) and large derivative near σ₀ = cos(20°) ≈ 0.94. The transition is a Morse-theoretic event on the Grassmannian, not a perturbative correction.
Connection to the Hilbert syzygy theorem
The three-tier structure terminates at H² — there is no H³ for generic quantum systems. Paper 578 proves this categorically: the relevant module category has global homological dimension ≤ 2 (Hilbert syzygy theorem, 1890). The Grassmannian encodes why: $\pi_2(\mathrm{Gr}(k,n)) = \mathbb{Z}$ for $k > 0$, which supports FUSE (H² holonomy) but $\pi_3 = 0$ generically, which is why there is no H³ opcode.
What would falsify it
The universality claim is falsified if:
- A domain is found where the snap threshold is substantially different from 20° for reasons not attributable to domain-specific renormalisation (e.g., different definition of the reference state). A threshold of 35° in chemistry and 5° in QEC, with no principled connection between them, would falsify the universality.
- The Hubbard model Mott transition is shown to occur at U/t substantially different from the value consistent with θ_G ≈ 20° when both are computed with the same definition of σ₀.
- The 2008 financial crisis is shown, on further analysis, to have been well-described by a rank-1 factor structure at the time — which would mean the 66° diagnosis is an artefact of hindsight data selection.
Open questions
- Paper 583 (planned): Is the CHSH Bell inequality violation threshold the same β* snap on $\mathrm{Gr}(2, 4)$? The Schmidt angle θ_G between the two halves of a bipartite quantum system is the natural entanglement measure. The CHSH violation requires |⟨CHSH⟩| > 2, and Tsirelson’s bound gives maximum $2\sqrt{2}$ at maximum entanglement. Does the β* snap at θ_G ≈ 20° coincide with a specific CHSH violation threshold?
- Paper 574 (amplituhedron): The positive Grassmannian of scattering amplitudes (Arkani-Hamed/Trnka) uses the same manifold. Is the β* snap related to the spurious-pole cancellation in BCFW recursion?
- Cross-domain calibration: The 20° threshold is calibrated separately in each domain. Is there a first-principles derivation from the geometry of $\mathrm{Gr}(k, n)$ that predicts this threshold without domain-specific fitting?
See also: Universal Chemical Bonding (#570) · QEC as Grassmannian parallel transport (#577) · The Grassmannian of Systematic Risk (#580) · Why exactly three tiers? (#578) ·