Papers
Topics
Authors
Recent
Search
2000 character limit reached

Unitary Yang--Baxter Operators: Towards a Classification

Published 17 Aug 2026 in math.QA | (2608.16865v1)

Abstract: There is a well-known circle of conjectures relating unitary solutions of the Yang--Baxter equation, unitary braided fusion categories, topological quantum computation, and link invariants. Progress is limited by the lack of a classification of unitary Yang--Baxter operators. We propose a conjectural classification with three generating sources: monomial solutions, group-type solutions arising from Yetter--Drinfeld modules, and solutions from twisted group-algebra towers. We conjecture that, up to unit scalars and local unitary basis changes, all solutions are generated by these sources. We provide evidence for our classification conjecture by means of computational searches for unitary Yang--Baxter operators among Clifford groups and a polynomial-in-Pauli ansatz. This search yields only a few solutions that do not appear to be equivalent to monomial or group type solutions. Direct calculations identify their link invariants with specializations of the HOMFLYPT and BMW/Kauffman invariants; the latter is a squared Jones specialization at a primitive sixth root, and the corresponding qutrit operator strictly localizes the semisimple quotient of the BMW algebra specializations $\mathcal{C}_n(q<sup>3,e<sup>{π\ii/6})$ associated with the braided fusion category SO(4)2SO(4)_2. The quaternionic Family III specialization realizes a two-eigenvalue class and gives tensor-power realization of the semisimple Hecke algebra tower associated with (SU(3)_3). Significantly, this answers a recent existence question of Lechner, and verifies an open case of the Rowell-Wang localization conjecture from 2010.

Authors (2)

Summary

  • The paper proposes a classification framework in which unitary Yang–Baxter operators arise from monomial, group-type, or twisted group-algebra constructions combined through products, sums, and orthogonal-sum operations.
  • The metric-group families produce finite braid images and link invariants computable exactly in deterministic polynomial time, while generic cyclic dimension-8 families have infinite but virtually abelian braid images.
  • The paper combines exact low-dimensional searches with structural results, identifying new qutrit and ququart realizations, verifying cases of the Rowell–Wang conjecture, and clarifying that a complete classification remains open.

Overview and motivation

The paper develops a structural approach to the classification of unitary Yang–Baxter operators (YBOs), motivated by a circle of conjectures connecting unitary solutions of the Yang–Baxter equation, unitary braided fusion categories, topological quantum computation, and link invariants (2608.16865). A general classification of constant YBE solutions is known only in dimension 2 (Hietarinta), so the authors propose a conjectural classification: up to unit scalars and local unitary basis changes, every unitary YBO is generated by three sources — monomial solutions (set-theoretic solutions plus multiplicative cocycle data), group-type solutions from Yetter–Drinfeld modules, and solutions from twisted group-algebra towers — closed under external products, direct sums, and admissible internal orthogonal sums.

The main technical contribution is the abelian specialization of the twisted tensor-product construction of Gustafson–Kimball–Rowell–Zhang to twisted group-algebra towers Tn(G;ν,τ)T_n(G;\nu,\tau) built from a finite abelian group GG, a unitary 2-cocycle ν\nu, and a bicharacter τ\tau. The braid relation reduces to a finite coefficient equation for a function f ⁣:GCf\colon G\to\mathbb C, and unitarity to a finite coefficient condition Cfν(t)=δt,0C_f^\nu(t)=\delta_{t,0}; this converts the analytic matrix problem into finite algebraic data.

Three families from metric groups

For a finite metric abelian group (A,q)(A,q), the paper constructs three uniform families of unitary YBOs:

  • Family I is the Gaussian construction RQI=A1/2aQ(a)XaZaR^{\mathrm I}_Q=|A|^{-1/2}\sum_a Q(a)\,X_a\otimes Z_a, with two polarization signs ϵ=±1\epsilon=\pm1.
  • Family II uses two coupled quadratic phases Q,PQ,P on GG0, preserving the total label; a self-adjoint automorphism GG1 yields the reindexed Family GG2.
  • Family III has constant coefficients but nontrivially twisted local multiplication (GG3 not cohomologous to GG4); its local algebra is noncommutative.

All three admit faithful compatible GG5-representations on tensor powers via generalized Pauli operators, yielding unitary YBOs. Orthogonal decompositions of GG6 induce canonical factorizations of towers, spectra, and invariants, with primary decomposition automatic and block decompositions requiring GG7-stability choices — a stated dependence on auxiliary data rather than a canonical invariant.

Two structural consequences follow. First, each local operator normalizes a generalized Pauli group whose projective label group is finite, so by the Clifford exact sequence all ordinary braid images of the metric families are finite. Second, the identity-coefficient functional gives scalar Turaev enhancements, and the trace of any braid word reduces to a finite quadratic Gauss sum over the kernel of an explicitly computable homomorphism; using the algorithm of Cai–Chen–Lipton–Lu, the associated link invariants are evaluated exactly in deterministic polynomial time in strand number and word length for fixed metric data. This is a strong complexity result: it places these Gaussian-type invariants in P, contrasting with the generally hard evaluation of Reshetikhin–Turaev-type invariants.

Generically non-Gaussian cyclic families

Applying the admissible internal orthogonal-sum operation GG8 to two-torsion sectors of the dimension-8 cyclic tower produces eight one-parameter families GG9. For generic ν\nu0 these are non-Gaussian in the fixed Pauli realization, yet each contains signed-Gaussian specializations (e.g., ν\nu1 with ν\nu2). When ν\nu3 is not a root of unity, the local generator has infinite projective order and the braid image is infinite. Nevertheless, the paper proves these images are virtually abelian: comparing each parameter value with a signed-Gaussian base point places generator ratios in a common commutative two-torsion algebra, giving an abelian normal subgroup of index bounded by the finite base-point image. This separates finiteness from virtual abelianness and confirms Conjecture 1.3 (virtual abelianness of unitary YBO braid images) within this family — notably without contradicting the Rowell–Wang localization conjecture.

Low-dimensional computational evidence

Exact searches support the classification conjecture within declared search spaces. For the full-support cyclic Pauli-polynomial ansatz in dimensions ν\nu4, only Gaussian points survive unitarity; in ν\nu5 the positive-dimensional unitary components are exactly the eight families above. In square dimensions (4, 9), Clifford representations are monomial after a Zak basis change, so no new classes arise there — a limitation the authors state plainly.

The two-qutrit Clifford enumeration (over a projective group of order ν\nu6) yields 1750 strict YBOs reducing to two representatives not identified as monomial or group type:

Source Local dim Spectrum Quotient / invariant
Family ν\nu7, ν\nu8 3 ν\nu9 Temperley–Lieb; HOMFLYPT at τ\tau0
Family II, τ\tau1 3 τ\tau2 rank-one BMW, τ\tau3; Kauffman invariant = squared Jones at τ\tau4
Family III, τ\tau5 4 τ\tau6 Hecke, non-Temperley–Lieb

The Family II operator strictly localizes the semisimple BMW trace quotient τ\tau7 at τ\tau8, verifying an open case of the Rowell–Wang localization conjecture from 2010. The quaternionic Family III specialization realizes Lechner's exceptional two-eigenvalue class τ\tau9: it gives a faithful tensor-power realization of the semisimple Hecke tower f ⁣:GCf\colon G\to\mathbb C0 associated with f ⁣:GCf\colon G\to\mathbb C1, answering a recent existence question of Lechner and showing that 4 is the smallest possible local dimension for that class.

In the two-ququint case, exact characteristic-zero enumeration over f ⁣:GCf\colon G\to\mathbb C2 followed by finite-field (f ⁣:GCf\colon G\to\mathbb C3) gauge tests matches all remaining non-monomial orbits with signed Family I or Family f ⁣:GCf\colon G\to\mathbb C4 representatives. The authors emphasize this is a finite-field observation, not a characteristic-zero classification.

Limitations and open questions

The paper is explicit about scope. The classification conjecture remains open; the low-dimensional computations cover only specified Pauli directions and Clifford ansätze, exclude sparse solutions, and rely on equivalences broader than local unitary equivalence (Galois conjugation, conditional Doikou–Smoktunowicz f ⁣:GCf\colon G\to\mathbb C5-moves, finite-field gauge tests). The ququint identifications hold only over f ⁣:GCf\colon G\to\mathbb C6. The internal orthogonal-sum construction is analyzed one Pauli realization at a time rather than intrinsically, and the rigidity conjecture for fixed Pauli directions is stated only for f ⁣:GCf\colon G\to\mathbb C7. Whether the three generating sources suffice outside these search spaces, and whether characteristic-zero versions of the finite-field comparisons hold, remain unresolved.

Conclusion

The paper supplies a coherent algebraic framework — twisted group-algebra towers plus three combination operations — under which diverse known and new unitary YBOs become instances of a single coefficient calculus. Its concrete achievements include three uniform metric families with finite braid images and polynomial-time computable link invariants, generically non-Gaussian dimension-8 families with virtually abelian (possibly infinite) braid images, strict localizations resolving cases of the Rowell–Wang conjecture and Lechner's existence question, and substantial exact computational evidence for the proposed trichotomy. The residual gap between computational evidence and a genuine classification defines the remaining program.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.