Polynomial integrability of the Temperley-Lieb loop model: non-degenerate principal-Hamiltonian spectra and the transfer-operator-to-Hamiltonian decomposition for all system sizes

The Temperley-Lieb (TL) loop model on a strip of width $n$ with loop fugacity $\beta$ has a row-transfer operator $T_n(u,\beta)$ that is an element of the algebra $\mathrm{TL}_n(\beta)$ and commutes with itself at all values of the spectral parameter $u$ . Its Hamiltonian limits at the identity points $u_*=0$ and $u_*=-2/\beta$ yield two principal Hamiltonians: the familiar $h_0=-(e_1+\cdots+e_{n-1})$ and a second element $h_{-2/\beta}$ that had not been discussed before the source work. Poncini and Rasmussen (2023) proved that the model is polynomially integrable for $\beta>2\cos(\pi/n)$ , in the sense that the transfer operator is a polynomial in some spectral-parameter-independent element, and verified computationally that $T_n(u,\beta)$ is a polynomial in $h_0$ for all but finitely many $\beta$ for $n\le17$ , and in $h_{-2/\beta}$ for $n\le6$ ; the corresponding statements for general $n$ are left as conjectures. The open problem is to settle these for all system sizes: prove or disprove the generic non-degeneracy of the spectrum of $\rho_n(h_{u_*})$ (Conjecture 4.15), which would imply that $h_{u_*}$ generates its own centraliser and hence that $T_n(u,\beta)$ is polynomial in $h_{u_*}$ ; establish the unique decomposition of $T_n(u,\beta)$ in powers of $h_{u_*}$ with monic denominator $f_{n,u_*}(\beta)$ and its conjectured factorization (Conjectures 4.19-4.20, with the refined decompositions of Conjectures B.3-B.4); and classify the exceptional $\beta$ -values for which polynomiality fails, including their relation to the roots of $f_{n,u_*}$ .

Background

A central tool in the exact solution of two-dimensional lattice models is the transfer matrix: the partition function on an $L\times n$ lattice is the trace of the $L$ -th power of a row-transfer operator. In the quantum inverse scattering method, integrability rests on an $R$ -matrix acting on a tensor product of local vector spaces and satisfying the Yang-Baxter equation, from which one obtains a commuting family of transfer matrices. Loop models based on the Temperley-Lieb algebra do not fit this tensorial mould, and the source paper develops an alternative, planar-algebraic framework for them.

The Temperley-Lieb algebra $\mathrm{TL}_n(\beta)$ is the associative algebra generated by $e_1,\dots,e_{n-1}$ subject to $e_i^2=\beta e_i$ , $e_ie_{i\pm1}e_i=e_i$ , and $e_ie_j=e_je_i$ for $|i-j|>1$ . Diagrammatically, its elements are linear combinations of non-crossing matchings of $2n$ boundary points drawn in a rectangle, and concatenation of diagrams erases each closed loop at the cost of a multiplicative factor $\beta$ , the loop fugacity. A planar algebra, in the sense of the formalism recalled in the source paper, organises such diagram spaces across all sizes and allows compositions more general than mere concatenation; the Temperley-Lieb loop model is built in the Temperley-Lieb planar algebra. Its transfer operator $T_n(u,\beta)\in\mathrm{TL}_n(\beta)$ is constructed by Baxterisation, that is, from an $R$ -operator $R(u)=\mathbb{1}_2+ue$ acting non-tensorially as a planar tangle; a local generalised Yang-Baxter equation then implies $[T_n(u,\beta),T_n(v,\beta)]=0$ for all $u,v,\beta$ , so the model is Yang-Baxter integrable.

The source paper re-examines integrals of motion algebraically: a $u$ -independent element of the algebra commuting with the transfer operator family. A model is called polynomially integrable if the transfer operator is a polynomial in a single spectral-parameter-independent element $b$ , in which case the entire commuting family collapses to univariate polynomial algebra. Identity points are values $u_*$ at which $T_n(u_*,\beta)$ is proportional to the identity; for $\beta\notin\{-2,0,2\}$ they are $\{0,-\beta/2,-2/\beta\}$ . Expanding $T_n$ about an identity point defines a principal Hamiltonian. At $u_*=0$ this is the usual $h_0=-\sum_{i=1}^{n-1}e_i$ ; at $u_*=-2/\beta$ (and equivalently $u_*=-\beta/2$ up to scale) one obtains a second Hamiltonian $h_{-2/\beta}$ , expressed through certain spider-like connectivity diagrams $S^{(n)}_{j,k}(u)$ that also enter the exact decomposition $T_n(u,\beta)=(\beta[n+1]_{u^2}+2u[n]_{u^2})\mathbb{1}_n+u(\beta+2u)(2+\beta u)\sum_{j,k}u^{2k}S^{(n)}_{j,k}(u)$ , where $[m]_x=1+x+\cdots+x^{m-1}$ .

The spectral analysis uses the cellularity of $\mathrm{TL}_n(\beta)$ : its standard modules $V_{n,d}$ , labelled by the number $d\in\{n,n-2,n-4,\dots\}$ of through-lines, are spanned by link states (half-diagrams), with $\dim V_{n,d}=\binom{n}{(n-d)/2}-\binom{n}{(n-d)/2-1}$ . The direct sum $\rho_n=\bigoplus_d\rho_{n,d}$ has dimension $c_n=\binom{n}{\lfloor n/2\rfloor}$ and is a faithful representation of $\mathrm{TL}_n(\beta)$ whenever the standard-module Gram determinants do not vanish, in particular when $\beta$ is an indeterminate. A matrix is called non-derogatory when its minimal polynomial equals its characteristic polynomial, which for $\rho_n(h_{u_*})$ is equivalent to the spectrum being non-degenerate (all eigenvalues distinct) and to the minimal polynomial of $h_{u_*}$ having degree $c_n$ . Non-degeneracy would mean that $h_{u_*}$ generates its own centraliser, so that every element commuting with it, including $T_n(u,\beta)$ , is a polynomial in $h_{u_*}$ . The paper proves polynomial integrability (with some generator) for $\beta>2\cos(\pi/n)$ using self-adjointness and semisimplicity, verifies non-degeneracy of $\rho_n(h_0)$ for $n=2,\dots,17$ and of $\rho_n(h_{-2/\beta})$ for $n=2,\dots,6$ at generic sample $\beta$ -values, and computes the minimal polynomials $m^{(n)}_0$ explicitly for $n\le7$ , of degrees $2,3,6,10,20,35=c_n$ . It further establishes, from polynomiality of the matrix entries in $\beta$ , that spurious spectral degeneracies occur for at most finitely many $\beta$ , so $T_n(u,\beta)$ is polynomial in $h_{u_*}$ for all but finitely many $\beta$ at least for the verified ranges of $n$ .

For the small values of $n$ where the expansions were computed, $T_n(u,\beta)$ admits a unique decomposition in powers of $h_{u_*}$ whose only $\beta$ -dependence, apart from polynomial numerators, is a single monic denominator polynomial $f_{n,u_*}(\beta)$ ; for example $f_{3,0}=f_{4,0}=1$ , $f_{5,0}=(\beta^2+4)(\beta^2-\tfrac12)(\beta^4-27\beta^2+121)$ , and $f_{5,-2/\beta}=(\beta^{10}-1024)^3p_5(\beta)$ with $p_5$ an even polynomial of degree 198. Values of $\beta$ for which $T_n(u,\beta)$ fails to be polynomial in $h_{u_*}$ are called $h_{u_*}$ -exceptional; they form a finite set $E_{n,u_*}$ contained in the root set $Z_{n,u_*}$ of $f_{n,u_*}$ . Known computations give $Z_{5,0}=E_{5,0}$ , $Z_{6,0}\setminus E_{6,0}=\{0\}$ , $Z_{n,-2/\beta}=E_{n,-2/\beta}$ for $n=3,4,5$ , and $E_{n,0}\cap E_{n,-2/\beta}=\emptyset$ for $n=3,4,5$ , so that for these $n$ the transfer operator is polynomial in at least one of the two Hamiltonians for every $\beta\in\mathbb{C}$ .

Problem Statement

For the Temperley-Lieb loop model on the strip with indeterminate loop fugacity $\beta$ , settle whether the transfer operator is polynomial in its principal Hamiltonians at every system size $n$ , and characterize that polynomiality exactly. The single objective comprises three linked deliverables, the first being the engine for the other two.

(1) Spectral non-degeneracy (Conjecture 4.15 of the source). Prove or disprove: for every $n\ge2$ and each $u_*\in\{0,-2/\beta\}$ , the spectrum of $\rho_n(h_{u_*})$ is non-degenerate when $\beta$ is an indeterminate. Equivalent formulations to be settled: the minimal polynomial of $h_{u_*}$ , whose coefficients are polynomials in the indeterminate $\beta$ , has degree $c_n=\binom{n}{\lfloor n/2\rfloor}$ ; $h_{u_*}$ is non-derogatory; $h_{u_*}$ generates its own centraliser in $\mathrm{TL}_n(\beta)$ . An affirmative answer implies, via the paper's Proposition 4.18 together with Corollary 4.17, that $T_n(u,\beta)$ is polynomial in $h_{u_*}$ for all but finitely many $\beta$ .

(2) Transfer-operator-to-Hamiltonian decomposition (Conjectures 4.19-4.20, refined in Appendix Conjectures B.3-B.4). Prove or disprove: for every $n\ge3$ and each $u_*\in\{0,-2/\beta\}$ , $T_n(u,\beta)$ admits a unique decomposition

T_n(u,\beta)=\big(\beta[n+1]_{u^2}+2u[n]_{u^2}\big)\mathbb{1}_n +\frac{u(\beta+2u)(2+\beta u)}{f_{n,u_*}(\beta)}\sum_{i=1}^{c_n-1}a^{n,u_*}_i(u,\beta)\,h_{u_*}^{i},

where $[m]_x=1+x+\cdots+x^{m-1}$ , $f_{n,u_*}(\beta)$ is monic, and no root of $f_{n,u_*}$ is a root of any $a^{n,u_*}_i(u,\beta)$ ; and that the denominator factorizes as $f_{n,-2/\beta}(\beta)=(\beta^{2n}-4^n)^{n-2}p_n(\beta)$ with $p_n(\beta)$ a non-degenerate (no repeated roots) even polynomial. In the source, the sum runs over $i=1,\dots,l^{(n)}_{u_*}-1$ , where $l^{(n)}_{u_*}=c_n$ is the minimal-polynomial degree asserted in deliverable (1). The refined versions assert the corresponding unique decompositions of the rescaled operator $\tilde T_n(x)$ , with $T_n(u)=u^n\tilde T_n(u+1/u)$ , into Chebyshev polynomials of the second kind $U_k$ in $x$ (with $U_k(\cos\theta)=\sin((k+1)\theta)/\sin\theta$ , entering through the prefactor $\beta U_n(x/2)+2U_{n-1}(x/2)$ ) and powers of $h_0$ (respectively $h_{-2/\beta}$ ), together with the degree and parity structure observed for the coefficient polynomials $\tilde a^{n,u_*}_{i,k}(\beta)$ (their degrees follow a fixed formula in $i$ and $k$ in each branch, affine in $i$ and $k$ for $u_*=-2/\beta$ and carrying an additional parity-dependent term in $k$ for $u_*=0$ , and their parity matches that of $n+i+k$ up to a fixed offset). Deliverable (2) bundles the following independently refutable sub-claims, each a mandatory target for every $n\ge3$ and both $u_*$ : (a) the decomposition itself (existence and uniqueness of the expansion in powers of $h_{u_*}$ with the stated prefactor and monic denominator $f_{n,u_*}$ ); (b) coprimality of $f_{n,u_*}$ with every numerator $a^{n,u_*}_i$ ; (c) the denominator factorization $f_{n,-2/\beta}(\beta)=(\beta^{2n}-4^n)^{n-2}p_n(\beta)$ with $p_n$ a non-degenerate even polynomial; (d) the refined decompositions of $\tilde T_n(x)$ into Chebyshev polynomials $U_k$ and powers of $h_0$ (respectively $h_{-2/\beta}$ ); and (e) the stated degree and parity structure of the coefficient polynomials $\tilde a^{n,u_*}_{i,k}$ . A counterexample to any one sub-claim refutes that sub-claim and deliverable (2) for the corresponding $n$ and $u_*$ , leaving the others open; only proofs of all sub-claims for every $n\ge3$ and both $u_*$ resolve deliverable (2) affirmatively.

(3) Exceptional-specialisation classification. For general $n$ , determine the finite sets $E_{n,u_*}$ of $h_{u_*}$ -exceptional $\beta$ -values, for which $T_n(u,\beta)$ fails to be polynomial in $h_{u_*}$ , and settle their relation to the root sets $Z_{n,u_*}$ of $f_{n,u_*}(\beta)$ : currently only the inclusion $E_{n,u_*}\subseteq Z_{n,u_*}$ , itself conditional on the decomposition conjecture (2) that defines $f_{n,u_*}$ , is available in general, together with the small- $n$ identifications listed in the Background. Deliverable (3) is the conjunction of two mandatory demands, for general $n$ and each $u_*\in\{0,-2/\beta\}$ : an explicit determination of the exceptional sets $E_{n,u_*}$ , and their relation to the root sets $Z_{n,u_*}$ . It is resolved by settling both: either by establishing $E_{n,u_*}=Z_{n,u_*}$ for all $n$ and both $u_*$ , or, wherever the two sets differ, by determining which elements of $Z_{n,u_*}$ are not exceptional. A result that proves only the relation, or that determines $E_{n,u_*}$ for isolated values of $n$ , resolves only that part or that case, not deliverable (3) in general.

Deliverable (1) implies polynomiality for all but finitely many $\beta$ ; deliverable (2) gives the explicit structure of that polynomiality; deliverable (3) delimits exactly where it holds. Together they settle whether the Temperley-Lieb loop model is polynomially integrable with respect to its principal Hamiltonians at all system sizes. Broader directions flagged by the authors but not part of this problem's completion boundary include how finite-size polynomial integrability manifests in the continuum scaling limit, and extensions of the analysis to affine/periodic tangle settings and to other planar-algebraic models.

References

Xavier Poncini and Jorgen Rasmussen, 'Integrability of planar-algebraic models', J. Stat. Mech. 2023 (2023) 073101. DOI: 10.1088/1742-5468/acdce7; arXiv:2206.14462; https://arxiv.org/abs/2206.14462
Xavier Poncini and Jorgen Rasmussen, 'Integrable models from singly generated planar algebras', Nucl. Phys. B 994 (2023) 116308. DOI: 10.1016/j.nuclphysb.2023.116308; arXiv:2302.11712; https://arxiv.org/abs/2302.11712
Yacine Ikhlef and Alexi Morin-Duchesne, 'Fusion in the periodic Temperley-Lieb algebra: general definition of a bifunctor', SciPost Phys. Core 9 (2026) 053. DOI: 10.21468/SciPostPhysCore.9.3.053; arXiv:2509.11756; https://arxiv.org/abs/2509.11756
Yacine Ikhlef and Alexi Morin-Duchesne, 'Fusion of irreducible modules in the periodic Temperley-Lieb algebra', SciPost Phys. 17 (5) (2024) 132. DOI: 10.21468/SciPostPhys.17.5.132; arXiv:2312.14837; https://arxiv.org/abs/2312.14837
Remy Adderton and Murray T. Batchelor, 'The planar parafermion algebra: the Z_N clock model and the coupled Temperley-Lieb algebra', J. Phys. A: Math. Theor. 59 (2026) 045201. DOI: 10.1088/1751-8121/ae36d5; arXiv:2508.01672; https://arxiv.org/abs/2508.01672
Paul A. Pearce and Jorgen Rasmussen, 'Ocneanu algebra of seams: Critical unitary E6 RSOS lattice model', in '2024 MATRIX Annals, Part II', MATRIX Book Series (Springer, 2026), pp. 113-135. DOI: 10.1007/978-3-032-16206-9_8; arXiv:2409.06236; https://arxiv.org/abs/2409.06236