Ring Closure as Holonomy

Cyclic peptide backbones modelled as cellular sheaves over a cycle graph. The dihedral mismatch accumulated on ring closure enters the sheaf Laplacian exactly where a magnetic flux enters a tight-binding ring, which puts the frustration spectrum in closed form.

lambda_k^{+/-} = 2 - 2 cos( (2 pi k +/- theta) / N )

Verified against numerical diagonalisation to 4e-15.

Status

The mathematics is verified and reproducible. The chemistry interpretation is partially supported and still open. Read FINDINGS.md before citing anything here.

Claim Status
Closed-form spectrum Verified, 4e-15
Flux / Aharonov-Bohm correspondence Verified, unitary equivalence
Strain laws (spectral gap and total) Verified, exact for all N
Betti invariance under stiffness Verified
theta extractable from coordinates Machinery validated; identification with the model's theta unresolved
Dilution with ring size Direction confirmed (p = 9e-38); exponent undetermined

The two strain laws

Two normalisations answer two different questions, and conflating them is the most likely source of a wrong exponent:

quantity normalisation closed form limit
spectral gap unit-norm section 2 - 2cos(theta/N) theta^2 / N^2
total strain unit per-residue amplitude 2N(1 - cos(theta/N)) theta^2 / N

E_tot = N * lambda_min exactly. The second is the Kirchhoff elastic-rod law, recovered without assuming elasticity anywhere.

Quick start

pip install -r requirements.txt
python verify.py     # theory checks, ~10 s, no network
python scaling.py    # both strain laws, N = 4..256
python run_p1.py     # downloads 8 CCDC structures, measures theta

holonomy_extract.py is the piece most likely to be useful standalone: correct NeRF backbone construction, CIF and PDB readers, Bishop-frame holonomy, a Gauss-Bonnet cross-check, and the loop-closure Jacobian.

What this is not

Not a structure predictor and not a competitor to conformational sampling. It answers a prior question: given a backbone's frame mismatch, can the ring close without strain, and if not how much is irreducible. Applies to macrocycles only, since an open backbone is a tree and carries no holonomy.

Caveats before reuse

  • Author line and affiliation in the manuscript are placeholders.
  • Two 2026 arXiv references in the related-work section came from a literature pass, not from direct reading. Verify before citing.
  • The mathematics stands on its own; the biological identification does not yet.
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support