betaFeedback

WorksPhysics and Astronomy

Semiclassical Asymptotics of Entanglement in Torus-Link States

Abstract

We investigate the entanglement entropy of topological quantum states associated with Td, dnT_{d,\,dn} torus links in (2+1)-dimensional SU(2) Chern–Simons theory at level kk. These links consist of dd components with pairwise linking number nn, and the corresponding link-complement states provide a natural setting to study multipartite entanglement in topological quantum field theories. We analyze the semiclassical limit k→∞k\to\infty of the entanglement entropy by deriving the asymptotic behavior of the moments of the reduced density matrix. To achieve this, we decompose the integrable-representation space into left-edge, right-edge, and bulk regions and employ the stationary-phase method to evaluate their respective contributions. We find that the leading asymptotic behavior is governed entirely by the two edge regions, while the bulk contribution is asymptotically suppressed. The resulting expression for the limiting entanglement entropy is given in terms of Hurwitz zeta functions and exhibits a dependence on the residue class of kk modulo nn. We further investigate the dependence of the limiting entanglement entropy on the winding parameter nn and the number of link components dd. In the large-dd limit, the entropy reduces to a binary entropy determined by two normalized weights associated with the left and right edges. Finally, we show that the two iterated limits, k→∞k\to\infty and d→∞d\to\infty, commute, reproducing the result obtained previously by taking the large-party limit first.

Introduction

Quantum entanglement is a fundamental feature of quantum systems and plays a central role in quantum information theory, quantum computation, and the characterization of strongly correlated quantum systems. Despite its importance, studying entanglement in generic quantum field theories is often technically challenging, owing to the infinitely many degrees of freedom and the nontrivial dependence on the geometry of the underlying spacetime. In this regard, topological quantum field theories (TQFTs) provide a particularly useful setting for studying quantum entanglement. Since physical observables and correlation functions in a TQFT are independent of the spacetime metric, such theories isolate the topological aspects of quantum correlations and provide an analytically tractable framework for investigating how entanglement encodes global and topological information.

Chern–Simons theory in (2+1)(2+1) dimensions [1] provides a prominent realization of a TQFT and establishes a remarkable connection between quantum field theory, knot and link invariants, and the topology of three-manifolds. In particular, the path integral of Chern–Simons theory on a three-manifold with multiple boundary components naturally prepares a multipartite quantum state in the tensor product of the Hilbert spaces associated with these boundaries. Knot and link complements therefore furnish a rich class of geometrically and topologically nontrivial quantum states whose entanglement properties can be studied systematically. Investigating quantities such as the Rényi and von Neumann entropies of these states can provide insight into how the topology of knots, links, and three-manifolds is encoded in quantum correlations, while also offering a controlled setting for exploring more general connections between topology and quantum entanglement. Over the past decade, several works have investigated different aspects of topological entanglement measures for such multi-boundary states. A non-exhaustive list of relevant studies includes [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16].

An important feature of quantum states associated with link complements is that their probability amplitudes are directly related to quantum invariants of the corresponding links [2]. More precisely, these amplitudes are given by expectation values of Wilson line operators supported on the links, which can be expressed in terms of the associated quantum link invariants. For instance, in SU(2)\mathrm{SU}(2) Chern–Simons theory, these amplitudes are related to colored Jones polynomials evaluated at a root of unity, q=exp⁡(2πik)q = \exp\left(\frac{2\pi i}{k}\right), where kk denotes the Chern–Simons level. Consequently, the entanglement measures of the resulting quantum states are determined by suitable combinations of these quantum invariants, thereby establishing a direct connection between quantum entanglement and the topological properties of links.

An important direction in this line of research is to investigate the asymptotic behavior of entanglement measures in the semiclassical regime, corresponding to the large-level limit k→∞k \to\infty. This limit is particularly interesting because it provides a bridge between quantum invariants and the underlying classical and geometric structures of the theory. Moreover, the asymptotic behavior of quantum invariants often reveals unexpected connections with geometry and number theory, involving quantities associated with moduli spaces, special values of zeta functions, and related arithmetic structures. Studying the large-kk behavior of entanglement measures therefore offers a natural way to explore whether such geometric and arithmetic information is also encoded in the entanglement structure of link-complement states.

Among the various classes of link complements, in this work we focus on torus links, and in particular on the family Td,dnT_{d,dn}. This family provides a useful setting for our analysis because the corresponding link-state coefficients can be expressed in terms of modular matrices, allowing the entanglement structure of the associated states to be investigated analytically. The primary goal of this work is to determine the semiclassical (large-kk) limit of the entanglement entropy for the Td,dnT_{d,dn} family in SU(2)\mathrm{SU}(2) Chern–Simons theory. Throughout this work, we refer to this large-kk limit as the ‘limiting entanglement entropy’ and denote it by EE∞\mathrm{EE}_{\infty}.

Earlier works investigating the entanglement structure of torus-link states include [9, 11, 12, 17]. In particular, Refs. [9, 17] have made progress towards analyzing the large-kk limits of entanglement measures for Td,dnT_{d,dn} link complements. For example, Ref. [9] obtained the large-kk limits of the entropies. for n=1,2,3,4,6n = 1,2,3,4,6. This analysis relied on the observation that the traces of the mmth powers (or moments) of the unnormalized reduced density matrix form sequences of positive integers as the level kk is varied. Exploiting this property, the moments were expressed as polynomial functions of kk for each Rényi index mm. By identifying the dependence of the leading-order polynomial coefficients on mm, the large-kk limit of the mmth Rényi entropy was obtained, and subsequently taking the limit m→1m \to1 yielded the limiting entanglement entropy. This approach, however, relies crucially on identifying the functional dependence of the polynomial coefficients on the Rényi index mm. Such a dependence could not be established for n=5n = 5, or more generally for larger values of nn, thereby limiting the applicability of this method.

In the present work, we employ Poisson summation and the method of stationary phase to derive the large-kk asymptotics of the eigenvalues of the unnormalized reduced density matrix for arbitrary nn. A key finding is that the leading asymptotic behavior of the density-matrix moments is governed by contributions from the two edges of the integrable-representation space. More specifically, only eigenvalues associated with the left edge R/k→0R/k \to0 and the right edge R/k→1R/k \to1 contribute at leading order, whereas the contribution from the bulk 0<R/k<10 < R/k < 1 is asymptotically suppressed. This edge-dominance mechanism enables us to derive the large-kk limit of the entanglement entropy for the entire Td,dnT_{d,dn} family, for arbitrary nn. Remarkably, the resulting expressions for the limiting entropies involve finite sums of Hurwitz zeta functions, with coefficients that retain information about the residue class of kk modulo nn.

The paper is organized as follows. In § 2, we introduce the theoretical framework of SU(2)k\mathrm{SU}(2)_k Chern–Simons theory, construct the quantum states associated with link complements, and specialize to the Td,dnT_{d,dn} torus-link family. In § 3, we derive the large-kk asymptotics of the density-matrix moments by analyzing the contributions from the left-edge, right-edge, and bulk regions of the integrable-representation space. In § 5, we use these asymptotic results to obtain the large-kk limits of the Rényi and the entanglement entropies. In § 6, we present numerical and analytical results illustrating the dependence of the limiting entanglement entropy on the winding parameter nn, the number of link components dd, and the residue class of kk modulo nn. We also examine the large-dd behavior and its relation to the semiclassical limit. Finally, we summarize our results and discuss possible directions for future work in § 7.

Set-up

Chern-Simons theory and link complement states

The Chern-Simons action for a gauge group GG at level kk on a three-manifold MM is given by

S(A)=k4π∫MTr⁡(A∧dA+23A∧A∧A),(1)S(A) = \frac{k}{4\pi}\int_M \operatorname{Tr}\left(A \wedge dA + \frac{2}{3}A \wedge A \wedge A\right), \tag*{(1)}

where A=AμdxμA = A_\mu dx^\mu is the gauge connection on a principal GG-bundle over MM. The fundamental gauge-invariant observables of Chern–Simons theory are Wilson loop operators. For an oriented knot K⊂M\mathcal{K} \subset M, carrying a representation RR of GG, the Wilson loop is defined by:

WR(K)=Tr⁡RPexp⁡(i∮KA).(2)W_R(\mathcal{K}) = \operatorname{Tr}_R \mathcal{P}\exp\left(i\oint_{\mathcal{K}} A\right). \tag*{(2)}

In the absence of Wilson loop insertions, the partition function of the manifold is

Z(M)=∫eiS(A) dA,(3)Z(M) = \int e^{iS(A)}\,dA, \tag*{(3)}

which is a topological invariant of MM. If MM contains a link L=K1∪K2∪⋯∪Kd\mathcal{L} = \mathcal{K}_1 \cup\mathcal{K}_2 \cup\cdots\cup\mathcal{K}_d, whose components are colored by representations R1,R2,…,RdR_1,R_2,\ldots,R_d, the corresponding path integral becomes

Z(M;L)=∫eiS(A)∏i=1dWRi(Ki) dA.(4)Z(M;\mathcal{L}) = \int e^{iS(A)}\prod_{i=1}^{d} W_{R_i}(\mathcal{K}_i)\,dA. \tag*{(4)}

For a closed three-manifold, this partition function computes the expectation value of the Wilson loops. By Witten’s seminal result [1], these expectation values coincide with the quantum link invariants. When the manifold has a boundary Σ\Sigma, the path integral is performed with the boundary condition A∣Σ=QA|_{\Sigma}=Q. The resulting partition function is interpreted as a quantum state,

∣Ψ⟩≡ZQ(M;L)=∫A∣Σ=QeiS(A)∏i=1dWRi(Ki) dA(5)|\Psi\rangle\equiv Z_Q(M;\mathcal{L})=\int_{A|_{\Sigma}=Q} e^{iS(A)}\prod_{i=1}^{d} W_{R_i}(\mathcal{K}_i)\,\mathrm{d}A \tag*{(5)}

which belongs to the Hilbert space HΣ\mathcal{H}_{\Sigma} associated with the boundary surface. Throughout this work, we take the boundary to be a torus, Σ=T2\Sigma=T^2. The Hilbert space HT2\mathcal{H}_{T^2} admits a natural basis obtained by performing the Chern-Simons path integral on a solid torus with a Wilson line in representation RR inserted along its non-contractible cycle. The allowed labels RR are the integrable representations of the affine Lie algebra associated with GG. For the SU(2) group, these are the representations whose Dynkin label aa satisfies the condition 0≤a≤k0\leq a\leq k. Consequently, HT2\mathcal{H}_{T^2} is finite-dimensional, with a basis

basis⁡(HT2)={∣e~a⟩:0≤a≤k}.(6)\operatorname{basis}(\mathcal{H}_{T^2})=\{|\tilde e_a\rangle:0\leq a\leq k\}. \tag*{(6)}

Each basis state ∣e~a⟩|\tilde e_a\rangle is obtained by the path integral on a solid torus carrying a Wilson line in representation aa along its non-contractible cycle.

A manifold MM with dd disjoint torus boundaries can be interpreted as a dd-party quantum state which belongs to the tensor product Hilbert space of dd copies of HT2\mathcal{H}_{T^2} and can be written as:

∣Ψ⟩=∑a1,…,ad=0kCa1,a2,…,ad∣e~a1,e~a2,…,e~ad⟩(7)|\Psi\rangle=\sum_{a_1,\ldots,a_d=0}^{k} C_{a_1,a_2,\ldots,a_d}|\tilde e_{a_1},\tilde e_{a_2},\ldots,\tilde e_{a_d}\rangle \tag*{(7)}

where ∣e~a1,e~a2,…,e~ad⟩|\tilde e_{a_1},\tilde e_{a_2},\ldots,\tilde e_{a_d}\rangle is shorthand for ∣e~a1⟩⊗∣e~a2⟩⊗⋯⊗∣e~ad⟩|\tilde e_{a_1}\rangle\otimes|\tilde e_{a_2}\rangle\otimes\cdots\otimes|\tilde e_{a_d}\rangle. The complex coefficients Ca1,a2,…,adC_{a_1,a_2,\ldots,a_d} encode the topological information of the three-manifold MM and determine the entanglement properties of the associated quantum state.

When M=S3\LM=S^3\backslash\mathcal{L} is a link complement, where L\mathcal{L} is a dd-component link, then the coefficients Ca1,a2,…,adC_{a_1,a_2,\ldots,a_d} are obtained using the gluing technique prescribed in [2] and are identified with the Chern-Simons partition function of S3S^3 in presence of the link L\mathcal{L}. We call such a state the link complement state and denote it as ∣L⟩|\mathcal{L}\rangle given by:

∣L⟩=∑a1,…,ad=0kZ(S3;L[a1,a2,…,ad])∣e~a1,e~a2,…,e~ad⟩(8)|\mathcal{L}\rangle=\sum_{a_1,\ldots,a_d=0}^{k} Z(S^3;\mathcal{L}[a_1,a_2,\ldots,a_d])|\tilde e_{a_1},\tilde e_{a_2},\ldots,\tilde e_{a_d}\rangle \tag*{(8)}

where Z(S3;L[a1,a2,…,ad])Z(S^3;\mathcal{L}[a_1,a_2,\ldots,a_d]) denotes the S3S^3 partition function in the presence of link L\mathcal{L} whose components are colored by a1,a2,…,ada_1,a_2,\ldots,a_d. For SU(2) Chern-Simons theory, these partition functions are proportional to the colored Jones polynomials of L\mathcal{L} evaluated at the root of unity q=exp⁡(2πik+2)q=\exp\left(\frac{2\pi i}{k+2}\right). Thus, for SU(2) Chern-Simons theory, the amplitudes of the link complement state are completely determined by the colored Jones polynomials of the corresponding link.

Entanglement structure of link complement states

To obtain the bipartite entanglement structure of the link complement state (8), we bipartition the total system as (A∣B)=(p∣d−p)(A|B)=(p|d-p), where Alice’s subsystem has pp parties, and Bob’s system has (d−p)(d-p) parties. The corresponding bipartitioning of the total Hilbert space will be:

H=HT2⊗HT2⊗⋯⊗HT2⏟p⊗HT2⊗HT2⊗⋯⊗HT2⏟d−p=HA⊗HB.(9)\mathcal{H}=\underbrace{\mathcal{H}_{T^2}\otimes\mathcal{H}_{T^2}\otimes\cdots\otimes\mathcal{H}_{T^2}}_{p}\otimes\underbrace{\mathcal{H}_{T^2}\otimes\mathcal{H}_{T^2}\otimes\cdots\otimes\mathcal{H}_{T^2}}_{d-p}=\mathcal{H}_A\otimes\mathcal{H}_B. \tag*{(9)}

The total density matrix for the link state will be:

ρtotal=∣L⟩⟨L∣⟨L∣L⟩.(10)\rho_{\mathrm{total}}=\frac{|\mathcal{L}\rangle\langle\mathcal{L}|}{\langle\mathcal{L}|\mathcal{L}\rangle}. \tag*{(10)}

The bipartite measures are obtained by taking the partial trace over any one of the subsystems. Taking the partial trace of ρtotal\rho_{\mathrm{total}} with respect to HB\mathcal{H}_B, we get a reduced density matrix:

ρ=Tr⁡B(∣L⟩⟨L∣)⟨L∣L⟩≡σTr⁡[σ],(11)\rho= \frac{\operatorname{Tr}_B\left(|\mathcal{L}\rangle\langle\mathcal{L}|\right)}{\langle\mathcal{L}|\mathcal{L}\rangle} \equiv\frac{\sigma}{\operatorname{Tr}[\sigma]}, \tag*{(11)}

where, for computational ease, we have defined an unnormalized matrix σ≡Tr⁡B(∣L⟩⟨L∣)\sigma\equiv\operatorname{Tr}_B\left(|\mathcal{L}\rangle\langle\mathcal{L}|\right). All the bipartite entanglement measures associated with the link state can be obtained using matrix ρ\rho or σ\sigma. For example, the Rényi entropy of index mm can be calculated as:

Rm=ln⁡[Tr⁡ρm]1−m=11−mln⁡[Tr⁡σm(Tr⁡σ)m].(12)R_m = \frac{\ln\left[\operatorname{Tr}\rho^m\right]}{1-m} = \frac{1}{1-m}\ln\left[\frac{\operatorname{Tr}\sigma^m}{(\operatorname{Tr}\sigma)^m}\right]. \tag*{(12)}

The entanglement entropy can be obtained by taking the m→1m \to1 limit of the Rényi entropy:

EE=lim⁡m→1Rm.(13)\mathrm{EE} = \lim_{m\to1} R_m. \tag*{(13)}

Td,dnT_{d,dn} torus links and associated quantum state

A Td,dnT_{d,dn} torus link consists of dd copies of T1,nT_{1,n} knots:

Td,dn=T1,n⊔T1,n⊔⋯⊔T1,n,⏟d copies(14)T_{d,dn} = T_{1,n} \sqcup T_{1,n} \sqcup\cdots\sqcup T_{1,n}, \qquad\underbrace{\hspace{3cm}}_{d\ \mathrm{copies}} \tag*{(14)}

where each component T1,nT_{1,n} is isotopic to the unknot and can be embedded on the surface of a torus so that it winds once around the meridian and nn times around the longitude. Moreover, every pair of distinct components has the same linking number lk=n\mathrm{lk}=n. Thus, the Td,dnT_{d,dn} family consists of dd unknotted components with a uniform pairwise linking number nn, making it a particularly simple and symmetric class of links.

For the Td,dnT_{d,dn} family, the Chern-Simons partition function on S3S^3 in the presence of a Td,dnT_{d,dn} link admits a remarkably simple closed-form expression in terms of the modular SS and TT matrices, which furnish a unitary representation of the modular group SL(2,Z)\mathrm{SL}(2,\mathbb{Z}). Following Refs. [18, 19], the partition function for the SU(2)\mathrm{SU}(2) group is:

Z(S3;Td,dn[a1,a2,…,ad])=∑b=0k(STnS)b0(S0b)d−1(Sa1bSa2b⋯Sadb),(15)Z\left(S^3;T_{d,dn}[a_1,a_2,\ldots,a_d]\right)=\sum_{b=0}^{k}\frac{(ST^nS)_{b0}}{(S_{0b})^{d-1}}\left(S_{a_1b}S_{a_2b}\cdots S_{a_db}\right), \tag*{(15)}

where the modular matrix elements for the SU(2)\mathrm{SU}(2) group are:

Sab=2k+2sin⁡[(a+1)(b+1)πk+2],Tab=exp⁡[iπa(a+2)2(k+2)−iπk4(k+2)]δab.(16)S_{ab}=\sqrt{\frac{2}{k+2}}\sin\left[\frac{(a+1)(b+1)\pi}{k+2}\right], \qquad T_{ab}=\exp\left[\frac{i\pi a(a+2)}{2(k+2)}-\frac{i\pi k}{4(k+2)}\right]\delta_{ab}. \tag*{(16)}

For convenience, we introduce K≡k+2K\equiv k+2. Since K/k→1K/k\to1 in the large-kk limit, we will use KK and kk interchangeably in algebraic prefactors and power-law asymptotic estimates whenever their difference is subleading. However, this replacement must be handled with care in the oscillatory phases arising from the modular TT matrix elements. In particular, the Rényi entropies and their large-level limits can depend on the level modulo nn, making the distinction between kk modulo nn and k+2k+2 modulo nn important. We therefore use KK as a bookkeeping device to preserve the exact phase information, keeping the distinction between kk and KK explicit in the phase factors whenever necessary, while suppressing it elsewhere when it does not affect the leading asymptotics.

Substituting the partition function (15) into (8), the corresponding link complement state becomes:

∣Td,dn⟩=∑a1,…,ad=0k∑b=0k(STnS)b0(S0b)d−1(Sa1bSa2b⋯Sadb)∣ea1,ea2,…,ead⟩.(17)|T_{d,dn}\rangle=\sum_{a_1,\ldots,a_d=0}^{k}\sum_{b=0}^{k}\frac{(ST^nS)_{b0}}{(S_{0b})^{d-1}}\left(S_{a_1b}S_{a_2b}\cdots S_{a_db}\right)|e_{a_1},e_{a_2},\ldots,e_{a_d}\rangle. \tag*{(17)}

A more convenient representation of the state is obtained by performing the same local unitary transformation on each copy of the Hilbert space. Since local unitary transformations preserve all entanglement measures, they do not alter the entanglement properties of the state. Using the unitarity of the modular SS matrix, we define a new basis:

∣eai⟩=∑ci=0k(Saici)∗∣fci⟩.(18)|e_{a_i}\rangle= \sum_{c_i=0}^{k} (S_{a_i c_i})^* |f_{c_i}\rangle. \tag*{(18)}

for the iith Hilbert space. Using the symmetric property of SS matrix and employing the unitarity relation

∑xSxySxz∗=δyz,(19)\sum_x S_{xy} S_{xz}^* = \delta_{yz} , \tag*{(19)}

the sum over the representation labels collapses immediately, and the state assumes the GHZ-like form:

∣Td,dn⟩=∑R=0k(STnS)R0(S0R)d−1∣fR,fR,…,fR⟩.(20)|T_{d,dn}\rangle= \sum_{R=0}^{k} \frac{(S T^n S)_{R0}}{(S_{0R})^{d-1}} |f_R,f_R,\ldots,f_R\rangle. \tag*{(20)}

The above GHZ-like structure of the state tells us that the total density matrix is a diagonal matrix of order (k+1)d(k+1)^d, which has only (k+1)(k+1) non-zero diagonal elements. This diagonal structure ensures that the reduced density matrix ρ\rho obtained by any bi-partitioning of the total Hilbert space remains diagonal. Since the diagonal entries are also the eigenvalues of ρ\rho, and the zero eigenvalues do not contribute to the entanglement measures, the entanglement structure is insensitive to how many parties are traced out. So without loss of generality, we can assume the bipartition (A∣B)=(1∣d−1)(A|B)=(1|d-1). Tracing out the Bob system will give the reduced density matrix ρ\rho of order (k+1)(k+1), where the eigenvalues are labeled by the integrable representations, given as follows:

λR=∣FR∣2∣F0∣2+∣F1∣2+…+∣Fk∣2 ;FR=(STnS)R0(S0R)d−1.(21)\lambda_R = \frac{|F_R|^2}{|F_0|^2+|F_1|^2+\ldots+|F_k|^2} \ ; \quad F_R = \frac{(S T^n S)_{R0}}{(S_{0R})^{d-1}} . \tag*{(21)}

For our calculation, we define the unnormalized matrix σ\sigma such that ρ=σ/Tr⁡[σ]\rho=\sigma/\operatorname{Tr}[\sigma]. The unnormalized eigenvalues ΛR\Lambda_R of matrix σ\sigma are given as:

ΛR=∣FR∣2.(22)\Lambda_R = |F_R|^2 . \tag*{(22)}

Large-k asymptotics of the density matrix moments

In this section, we investigate the large-kk asymptotics of the density-matrix moments for the Td,dnT_{d,dn} torus-link states introduced in § 2.3. Our analysis focuses on the SU⁡(2)k\operatorname{SU}(2)_k Chern–Simons theory and exploits the modular-matrix representation of the corresponding link-state coefficients. We organize the asymptotic analysis into contributions from the left edge, bulk, and right edge of the integrable-representation space, and identify the leading contribution to the moments in the large-kk limit.

As discussed earlier, the Rényi entropies can be expressed in terms of the moments of the matrix σ\sigma, which is the unnormalized version of the reduced density matrix. For the present class of states, these moments take the form

Tr⁡[σm]=∑R=0k(ΛR)m=∑R=0k∣FR∣2m.(23)\operatorname{Tr}[\sigma^m] = \sum_{R=0}^{k}(\Lambda_R)^m = \sum_{R=0}^{k}|F_R|^{2m} . \tag*{(23)}

To determine the large-kk behavior of the entanglement measures, we therefore need to understand the asymptotic behavior of these moments. To begin with, consider the function FRF_R defined in Eq. (21). Using the fact that the modular TT-matrix is diagonal, it can be written as

FR=1(S0R)d−1∑a=0kSRaTaanSa0.(24)F_R = \frac{1}{(S_{0R})^{d-1}} \sum_{a=0}^{k} S_{Ra} T_{aa}^{n} S_{a0} . \tag*{(24)}

There is an intermediate sum over the variable aa in the function FRF_R written in this form. Let us first analyze the large-kk or large-KK behavior of this sum. In this limit, the summand takes the leading-order form:

SRaTaanSa0⟶GK(a/K,R)=2sin⁡[π(R+1)aK]Keiπ(K2a2K2+aK−14)2sin⁡[πaK]K(25)S_{Ra}T^{n}_{aa}S_{a0}\longrightarrow G_K(a/K,R)=\frac{\sqrt{2}\sin\left[\pi(R+1)\frac{a}{K}\right]}{\sqrt{K}}e^{i\pi\left(\frac{K}{2}\frac{a^2}{K^2}+\frac{a}{K}-\frac{1}{4}\right)}\frac{\sqrt{2}\sin\left[\pi\frac{a}{K}\right]}{\sqrt{K}} \tag*{(25)}

where we have explicitly written the function GG in the variable a/Ka/K to enable us to apply the Poisson summation formula, which we review below.

Review: Poisson Summation Formula

For a sufficiently regular function ff and its Fourier transform f^\widehat{f},

∑m∈Zf(m)=∑s∈Zf^(s)=∑s∈Z∫−∞∞f(y)e−2πisy dy.\sum_{m\in\mathbb{Z}}f(m)=\sum_{s\in\mathbb{Z}}\widehat{f}(s)=\sum_{s\in\mathbb{Z}}\int_{-\infty}^{\infty}f(y)e^{-2\pi i sy}\,dy.

Applying the Poisson summation formula for the function GG, we will have:

∑a∈ZGK(a/K,R)=∑s∈Z∫−∞∞GK(y/K,R)e−2πisy dy.(26)\sum_{a\in\mathbb{Z}}G_K(a/K,R)=\sum_{s\in\mathbb{Z}}\int_{-\infty}^{\infty}G_K(y/K,R)e^{-2\pi i sy}\,dy. \tag*{(26)}

However, in (24), the sum over aa is restricted to integers from 0 to kk. To apply the Poisson summation formula to the finite sum, we extend GG from integer values of aa to a sufficiently smooth function of a continuous variable yy, and introduce a smooth cutoff function χ(y/K)\chi(y/K) that equals unity on the interval 0≤y/K≤10\le y/K\le1 and has compact support in a slightly larger interval. The finite sum can then be represented by the Poisson-resummed expression

∑a=0kGK(a/K,R)=∑s∈Z∫−∞∞GK(y/K,R)χ(y/K)e−2πisy dy.(27)\sum_{a=0}^{k}G_K(a/K,R)=\sum_{s\in\mathbb{Z}}\int_{-\infty}^{\infty}G_K(y/K,R)\chi(y/K)e^{-2\pi i sy}\,dy. \tag*{(27)}

where the cutoff ensures that only the desired range of the summation contributes. Next, we invoke the variable change x=y/Kx=y/K in the integral on the right-hand side. Now the cut-off function χ(x)\chi(x) equals unity on the interval 0≤x≤10\le x\le1 and smoothly decays to zero outside this interval. Since the cutoff is unity throughout the original integration domain and suppresses contributions away from it, we may effectively restrict the resulting integral to [0,1][0,1], giving the final resummed expression:

∑a=0kGK(a/K,R)=K∑s∈Z∫01GK(x,R)e−2πiKsx dx.(28)\sum_{a=0}^{k}G_K(a/K,R)=K\sum_{s\in\mathbb{Z}}\int_{0}^{1}G_K(x,R)e^{-2\pi i Ksx}\,dx. \tag*{(28)}

Thus, we arrive at the expression:

FR∼K(S0R)d−1∑s∈Z∫01GK(x,R)e−2πiKsx dx.(29)F_R\sim\frac{K}{(S_{0R})^{d-1}}\sum_{s\in\mathbb{Z}}\int_{0}^{1}G_K(x,R)e^{-2\pi i Ksx}\,dx. \tag*{(29)}

Since the moments Tr⁡[σm]\operatorname{Tr}[\sigma^m] require summing ∣FR∣2m|F_R|^{2m} over various values of RR, we need to understand the asymptotic behavior of FRF_R for various values of RR.

Asymptotic framework and decomposition of the representation space

A key feature of the Td,dnT_{d,dn} family is that the asymptotic contributions to Tr⁡[σm]\operatorname{Tr}[\sigma^m] can be organized into three distinct regions of the representation space. To characterize the different asymptotic regimes, we introduce the rescaled variable

r≡RK.r\equiv\frac{R}{K}.

with K=k+2K = k + 2. Since R=0,1,…,kR = 0,1,\ldots,k, the variable rr can be treated as a continuous variable in the large-kk limit, with r∈[0,1]r \in[0,1]. We distinguish three regimes according to how RR scales with kk. The ‘left-edge region’ corresponds to R=O(1)R = O(1), for which r→0r \to0 as k→∞k \to\infty. Similarly, the ‘right-edge region’ corresponds to k−R=O(1)k-R = O(1), for which r→1r \to1. The remaining representations, for which R=O(k)R = O(k) and rr remains finite in the open interval 0<r<10 < r < 1, constitute the ‘bulk region’.

A central observation of this work is that the eigenvalues associated with the two edge regions provide the leading contribution to Tr⁡[σm]\operatorname{Tr}[\sigma^m], whereas the contribution from the bulk region is asymptotically subleading. Accordingly, we decompose the moments as

Tr⁡[σm]=Tr⁡[σm]left+Tr⁡[σm]bulk+Tr⁡[σm]right,(30)\operatorname{Tr}[\sigma^m] = \operatorname{Tr}[\sigma^m]_{\mathrm{left}} + \operatorname{Tr}[\sigma^m]_{\mathrm{bulk}} + \operatorname{Tr}[\sigma^m]_{\mathrm{right}}, \tag*{(30)}

where the contribution Tr⁡[σm]bulk\operatorname{Tr}[\sigma^m]_{\mathrm{bulk}} is subleading compared to Tr⁡[σm]left\operatorname{Tr}[\sigma^m]_{\mathrm{left}} and Tr⁡[σm]right\operatorname{Tr}[\sigma^m]_{\mathrm{right}}. In the following, we use the method of stationary phase to evaluate the contributions from these three regions and thereby determine the large-kk asymptotics of Tr⁡[σm]\operatorname{Tr}[\sigma^m].

Left-edge contribution

The left-edge region corresponds to R=O(1)R = O(1) representations such that r=R/K→0r = R/K \to0. So we set R=αR = \alpha, where α\alpha is held fixed as k→∞k \to\infty. In the large-kk limit, up to leading order, we have:

S0R=S0α=2k+2sin⁡[(α+1)πk+2]∼2π(α+1)K3/2.(31)S_{0R} = S_{0\alpha} = \sqrt{\frac{2}{k+2}} \sin\left[\frac{(\alpha+1)\pi}{k+2}\right] \sim\frac{\sqrt{2}\pi(\alpha+1)}{K^{3/2}}. \tag*{(31)}

Furthermore, using (25), the leading form of GK(x,R)G_K(x,R) in the left-edge region is

GK(x,α)∼2Ksin⁡[πx]sin⁡[π(α+1)x]exp⁡[iK(nπx22+nπxK−nπ4K)].(32)G_K(x,\alpha) \sim\frac{2}{K}\sin[\pi x]\sin[\pi(\alpha+1)x]\exp\left[iK\left(\frac{n\pi x^2}{2}+\frac{n\pi x}{K}-\frac{n\pi}{4K}\right)\right]. \tag*{(32)}

Note that we have written the phase in the form eiK(⋯ )e^{iK(\cdots)} to make explicit the large parameter (KK) governing the oscillatory integral. This form is convenient for applying the method of the stationary phase in the subsequent analysis. Substituting the leading forms of S0RS_{0R} and GK(x,R)G_K(x,R) into (29), we get FRF_R in the left-edge region as:

Fα=K(3d−3)/22(d−3)/2πd−1(α+1)d−1∑s∈Z∫01sin⁡[πx]sin⁡[π(α+1)x]eiKΦs(x) dx.(33)F_\alpha= \frac{K^{(3d-3)/2}}{2^{(d-3)/2}\pi^{d-1}(\alpha+1)^{d-1}} \sum_{s\in\mathbb{Z}}\int_0^1 \sin[\pi x]\sin[\pi(\alpha+1)x]e^{iK\Phi_s(x)}\,dx. \tag*{(33)}

where the phase function is

Φs(x)=nπx22+nπxK−nπ4K−2πsx.(34)\Phi_s(x) = \frac{n\pi x^2}{2}+\frac{n\pi x}{K}-\frac{n\pi}{4K}-2\pi s x. \tag*{(34)}

For determining the leading-order stationary points in the large-kk limit, the terms suppressed by 1/K1/K in Φs(x)\Phi_s(x) can be neglected. The resulting leading-order phase is therefore

Φs(x)=nπx22−2πsx.(35)\Phi_s(x) = \frac{n\pi x^2}{2}-2\pi s x. \tag*{(35)}

Thus,

Fα=K(3d−3)/22(d−3)/2πd−1(α+1)d−1∑s∈ZIs,(36)F_\alpha= \frac{K^{(3d-3)/2}}{2^{(d-3)/2}\pi^{d-1}(\alpha+1)^{d-1}} \sum_{s\in\mathbb{Z}} I_s, \tag*{(36)}

where the integral is given as:

Is=∫01sin⁡[πx]sin⁡[π(α+1)x]eiKΦs(x) dx.(37)I_s = \int_0^1 \sin[\pi x]\sin[\pi(\alpha+1)x]e^{iK\Phi_s(x)}\,dx. \tag*{(37)}

The large-kk asymptotics of these integrals can then be obtained using the method of stationary phase, which we briefly review below.

Review: Stationary phase approximation

The stationary phase approximation [20] is used to perform asymptotic analysis of functions obtained by integrating against a rapidly varying complex exponential. Consider the integral:

I=∫abf(x)eiKϕ(x) dx(38)I = \int_a^b f(x)e^{iK\phi(x)}\,dx \tag*{(38)}

where KK is a large parameter. The functions f(x)f(x) and ϕ(x)\phi(x) are smooth and ϕ′(x)\phi'(x) has only a finite number of zeros in the interval [a,b][a,b]. In other words, the function ϕ(x)\phi(x) has finitely many stationary points in [a,b][a,b]. The method of the stationary phase tells us that the leading order contribution in the large-KK expansion of II will come from the immediate neighborhoods of stationary points in [a,b][a,b] of the phase function ϕ(x)\phi(x). Let x0x_0 be a stationary point in the interval [a,b][a,b]. Let us assume that ϕ′′(x0)>0\phi''(x_0)>0, which is applicable for our calculation. In the neighborhood of x0x_0, we have Taylor expansions:

f(x)=∑j=0∞f(j)(x0)j!(x−x0)j ;ϕ(x)=ϕ(x0)+ϕ′′(x0)2!(x−x0)2+⋯(39)f(x)=\sum_{j=0}^{\infty}\frac{f^{(j)}(x_0)}{j!}(x-x_0)^j \ ;\quad\phi(x)=\phi(x_0)+\frac{\phi''(x_0)}{2!}(x-x_0)^2+\cdots \tag*{(39)}

where we ignore the higher-order derivative terms in the expansion of ϕ(x)\phi(x). Thus, we can approximate the integral about a small neighborhood around x0x_0 as:

I∼∫nbh⁡(x0)∑j=0∞f(j)(x0)j!(x−x0)jeiKϕ(x0)eiKϕ′′(x0)2(x−x0)2 dx(40)I\sim\int_{\operatorname{nbh}(x_0)}\sum_{j=0}^{\infty}\frac{f^{(j)}(x_0)}{j!}(x-x_0)^j e^{iK\phi(x_0)}e^{iK\frac{\phi''(x_0)}{2}(x-x_0)^2}\,dx \tag*{(40)}

Here, the neighborhood nbh⁡(x0)\operatorname{nbh}(x_0) will depend on the location of the stationary point. If x0=ax_0=a, then nbh⁡(a)=[a,a+ε]\operatorname{nbh}(a)=[a,a+\varepsilon] for some small positive number ε\varepsilon. If x0=bx_0=b, then nbh⁡(b)=[b−ε,b]\operatorname{nbh}(b)=[b-\varepsilon,b]. For a<x0<ba<x_0<b, we have nbh⁡(x0)=[x0−ε,x0+ε]\operatorname{nbh}(x_0)=[x_0-\varepsilon,x_0+\varepsilon]. Next, introduce a change of variable u=(x−x0)Kϕ′′(x0)/2u=(x-x_0)\sqrt{K\phi''(x_0)/2}.

I∼2eiKϕ(x0)Kϕ′′(x0)∑j=0∞[A(j)2j/2f(j)(x0)j!(Kϕ′′(x0))j/2](41)I\sim\frac{\sqrt{2}e^{iK\phi(x_0)}}{\sqrt{K\phi''(x_0)}}\sum_{j=0}^{\infty}\left[A(j)\frac{2^{j/2}f^{(j)}(x_0)}{j!(K\phi''(x_0))^{j/2}}\right] \tag*{(41)}

where the function A(j)A(j) is generalization of the standard Fresnel integrals:

A(j)=∫range⁡(u)ujeiu2 du(42)A(j)=\int_{\operatorname{range}(u)}u^j e^{iu^2}\,du \tag*{(42)}

Here range⁡(u)\operatorname{range}(u) is the range of the integration variable uu and will again depend on the location of the stationary point. If x0=ax_0=a, then range⁡(u)=[0,∞]\operatorname{range}(u)=[0,\infty]. If x0=bx_0=b, then range⁡(u)=[−∞,0]\operatorname{range}(u)=[-\infty,0]. For a<x0<ba<x_0<b, we have range⁡(u)=[−∞,∞]\operatorname{range}(u)=[-\infty,\infty]. Since A(j)A(j) is independent of KK, the (41) provides the large-KK expansion of II to all orders in 1/K1/K in the neighborhood of the stationary point x0x_0. If ϕ(x)\phi(x) has multiple isolated stationary points in the interval [a,b][a,b], the contributions from all such stationary points must be summed to obtain the complete large-KK asymptotic expansion of II.

Let us now apply the stationary phase analysis to the integral IsI_s in (37). We identify

f(x)=sin⁡[πx]sin⁡[π(α+1)x] ;ϕ(x)=nπx22−2πsx(43)f(x)=\sin[\pi x]\sin[\pi(\alpha+1)x]\ ;\quad\phi(x)=\frac{n\pi x^2}{2}-2\pi sx \tag*{(43)}

The stationary points x0∈[0,1]x_0\in[0,1] are determined by ϕ′(x0)=0\phi'(x_0)=0 giving

x0=2sn.(44)x_0=\frac{2s}{n}. \tag*{(44)}

Since s∈Zs \in\mathbb{Z}, the condition 0≤x0≤10 \leq x_0 \leq1 restricts the the allowed values of ss to

s={0,1,2,…,n2,for n=even0,1,2,…,n−12,for n=odd(45)s = \begin{cases} 0,1,2,\ldots,\frac{n}{2}, & \text{for } n = \text{even} \\ 0,1,2,\ldots,\frac{n-1}{2}, & \text{for } n = \text{odd} \end{cases} \tag*{(45)}

The corresponding stationary points are

x0={0,2n,4n,…,n−2n,1,for n=even0,2n,4n,…,n−3n,n−1n,for n=odd(46)x_0 = \begin{cases} 0,\frac{2}{n},\frac{4}{n},\ldots,\frac{n-2}{n},1, & \text{for } n = \text{even} \\ 0,\frac{2}{n},\frac{4}{n},\ldots,\frac{n-3}{n},\frac{n-1}{n}, & \text{for } n = \text{odd} \end{cases} \tag*{(46)}

At the two endpoints x0=0x_0 = 0 and x0=1x_0 = 1, both f(x0)f(x_0) and f′(x0)f'(x_0) vanish. Consequently, the corresponding endpoint stationary-point contributions begin at order K−3/2K^{-3/2}, whereas the interior stationary points contribute at order K−1/2K^{-1/2}. The leading contribution, therefore, comes only from the stationary points satisfying 0<x0<10 < x_0 < 1. These stationary points are given by x0=2s/nx_0 = 2s/n where ss takes the following values:

s=1,2,3,…,⌊n−12⌋.(47)s = 1,2,3,\ldots,\left\lfloor\frac{n-1}{2}\right\rfloor. \tag*{(47)}

For these stationary points, we have:

f(x0)=sin⁡[2πsn]sin⁡[2π(α+1)sn] ;ϕ(x0)=−2πs2n ;ϕ′′(x0)=nπ.(48)f(x_0) = \sin\left[\frac{2\pi s}{n}\right]\sin\left[\frac{2\pi(\alpha+1)s}{n}\right] \ ;\quad\phi(x_0) = -\frac{2\pi s^2}{n} \ ;\quad\phi''(x_0) = n\pi. \tag*{(48)}

Applying the stationary phase formula (41) to the integral IsI_s in (37), we get, at leading order:

Is∼2e−2iπKs2nKnπsin⁡[2πsn]sin⁡[2π(α+1)sn]A(0) ,(49)I_s \sim\frac{\sqrt{2}e^{-\frac{2i\pi K s^2}{n}}}{\sqrt{K n \pi}}\sin\left[\frac{2\pi s}{n}\right]\sin\left[\frac{2\pi(\alpha+1)s}{n}\right] A(0) \ , \tag*{(49)}

where the integral A(0)A(0) is given as:

A(0)=∫−∞∞eiu2 du=eiπ4π .(50)A(0) = \int_{-\infty}^{\infty}e^{iu^2}\,du = e^{\frac{i\pi}{4}}\sqrt{\pi} \ . \tag*{(50)}

The large-kk asymptotic of FαF_\alpha in (36) will be:

Fα∼eiπ4(3d−4)/22d−42n1/2πd−1(α+1)d−1∑s=1⌊n−12⌋e−2iπKs2nsin⁡[2πsn]sin⁡[2π(α+1)sn] .(51)F_\alpha\sim\frac{e^{\frac{i\pi}{4} (3d-4)/2}}{2^{\frac{d-4}{2}} n^{1/2}\pi^{d-1}(\alpha+1)^{d-1}}\sum_{s=1}^{\left\lfloor\frac{n-1}{2}\right\rfloor}e^{-\frac{2i\pi K s^2}{n}}\sin\left[\frac{2\pi s}{n}\right]\sin\left[\frac{2\pi(\alpha+1)s}{n}\right] \ . \tag*{(51)}

Finally, we arrive at the leading order large-kk asymptotic of the corresponding eigenvalues ΛR=∣FR∣2\Lambda_R = |F_R|^2 in the left-edge region:

Λα∼K3d−42d−4nπ2d−2(α+1)2d−2∣∑s=1⌊n−12⌋e−2iπKs2nsin⁡[2πsn]sin⁡[2π(α+1)sn]∣2 .(52)\Lambda_\alpha\sim\frac{K^{3d-4}}{2^{d-4}n\pi^{2d-2}(\alpha+1)^{2d-2}}\left|\sum_{s=1}^{\left\lfloor\frac{n-1}{2}\right\rfloor}e^{-\frac{2i\pi K s^2}{n}}\sin\left[\frac{2\pi s}{n}\right]\sin\left[\frac{2\pi(\alpha+1)s}{n}\right]\right|^2 \ . \tag*{(52)}

We can now write the leading contribution from the left-edge region to Tr⁡[σm]\operatorname{Tr}[\sigma^m] as

Tr⁡[σm]left=∑α=0∞(Λα)m ,(53)\operatorname{Tr}[\sigma^m]_{\mathrm{left}} = \sum_{\alpha=0}^{\infty}(\Lambda_\alpha)^m \ , \tag*{(53)}

where the large-KK asymptotic form of Λα\Lambda_\alpha is given in (52). At finite KK, the sum over α\alpha terminates at k=K−2k = K - 2. In the left-edge scaling limit, α\alpha is held fixed as K→∞K \to\infty, and the upper limit consequently tends to infinity. Thus, provided the resulting large-α\alpha series is convergent, the limiting left-edge contribution is represented by the infinite sum above.

To carry out the sum over α\alpha, we first note from (52) that the α\alpha-dependent oscillatory factor contains

sin⁡[2π(α+1)sn].\sin\left[\frac{2\pi(\alpha+1)s}{n}\right].

Since ss is an integer, this factor is periodic in α\alpha with period nn. We can therefore decompose the non-negative integer α\alpha as

α=np+q,p∈Z≥0,q=0,1,…,n−1.\alpha=np+q,\qquad p\in\mathbb{Z}_{\geq0},\qquad q=0,1,\ldots,n-1.

This parametrization covers all non-negative integers α\alpha exactly once. Hence, we can rewrite the left-edge contribution as:

Tr⁡[σˉm]left=∑p=0∞∑q=0n−1(Λnp+q)m.(54)\operatorname{Tr}[\bar{\sigma}^{m}]_{\mathrm{left}}=\sum_{p=0}^{\infty}\sum_{q=0}^{n-1}(\Lambda_{np+q})^{m}. \tag*{(54)}

The advantage of this decomposition is that the oscillatory factor in (52) becomes independent of pp:

sin⁡[2π(α+1)sn]=sin⁡[2π(np+q+1)sn]=sin⁡[2π(q+1)sn].\sin\left[\frac{2\pi(\alpha+1)s}{n}\right]=\sin\left[\frac{2\pi(np+q+1)s}{n}\right]=\sin\left[\frac{2\pi(q+1)s}{n}\right].

The remaining pp-dependence can therefore be summed independently. In particular,

∑p=0∞1(np+q+1)2dm−2m=n2m−2dmζ(2dm−2m,q+1n),(55)\sum_{p=0}^{\infty}\frac{1}{(np+q+1)^{2dm-2m}}=n^{2m-2dm}\zeta\left(2dm-2m,\frac{q+1}{n}\right), \tag*{(55)}

where ζ(s,a)\zeta(s,a) denotes the Hurwitz zeta function. Since our set-up has d≥2d\geq2 and m≥1m\geq1, the series converges because 2dm−2m>12dm-2m>1. Performing the sum over pp, we obtain

Tr⁡[σˉm]left=K3dm−4m2dm−4mπ2dm−2mn2dm−m∑q=0n−1X(q)mζ(2dm−2m,q+1n),(56)\operatorname{Tr}[\bar{\sigma}^{m}]_{\mathrm{left}}=\frac{K^{3dm-4m}}{2^{dm-4m}\pi^{2dm-2m}n^{2dm-m}}\sum_{q=0}^{n-1}X(q)^{m}\zeta\left(2dm-2m,\frac{q+1}{n}\right), \tag*{(56)}

where we have defined:

X(q)=∣∑s=1⌊n−12⌋e−2iπKs2nsin⁡[2πsn]sin⁡[2π(q+1)sn]∣2.(57)X(q)=\left|\sum_{s=1}^{\left\lfloor\frac{n-1}{2}\right\rfloor}e^{-\frac{2i\pi Ks^{2}}{n}}\sin\left[\frac{2\pi s}{n}\right]\sin\left[\frac{2\pi(q+1)s}{n}\right]\right|^{2}. \tag*{(57)}

Thus, the left-edge contribution scales as

Tr⁡[σˉm]left∼k3dm−4m,(58)\operatorname{Tr}[\bar{\sigma}^{m}]_{\mathrm{left}}\sim k^{3dm-4m}, \tag*{(58)}

with the leading coefficient given by the expression in (56). We have thus obtained the leading contribution from the left-edge region, R=O(1)R=O(1). We next analyze the opposite boundary of the representation space, corresponding to k−R=O(1)k-R=O(1), and determine the associated right-edge contribution to Tr⁡[σˉm]\operatorname{Tr}[\bar{\sigma}^{m}].

Right-edge contribution

The right-edge region consists of representations RR such that k−R=O(1)k-R=O(1), corresponding to r=R/K→1r=R/K\to1. We therefore set R=k−βR=k-\beta, where β\beta is held fixed as k→∞k\to\infty. It should be noted that the modular SS matrix has symmetry that S0,k−β=S0βS_{0,k-\beta}=S_{0\beta}. Therefore, in the large-kk limit, up to leading order, we have:

S0R=S0,k−β=S0β∼2π(β+1)K3/2.(59)S_{0R}=S_{0,k-\beta}=S_{0\beta}\sim\frac{\sqrt{2}\pi(\beta+1)}{K^{3/2}}. \tag*{(59)}

Further, note the following important property in the right-edge region:

SRa=Sk−β,a=(−1)aSβa.(60)S_{Ra}=S_{k-\beta,a}=(-1)^aS_{\beta a}. \tag*{(60)}

Therefore, for the right-edge region

SRaTaanSa0=(−1)aSβaTaanSa0∼(−1)a2sin⁡[π(β+1)aK]Keinπ(K2a2K2+aK−14)2sin⁡[πaK]K.(61)S_{Ra}T_{aa}^{n}S_{a0}=(-1)^aS_{\beta a}T_{aa}^{n}S_{a0}\sim(-1)^a\frac{\sqrt{2}\sin\left[\pi(\beta+1)\frac{a}{K}\right]}{\sqrt{K}}e^{in\pi\left(\frac{K}{2}\frac{a^2}{K^2}+\frac{a}{K}-\frac{1}{4}\right)}\frac{\sqrt{2}\sin\left[\pi\frac{a}{K}\right]}{\sqrt{K}}. \tag*{(61)}

So, we extract the leading form of GK(x,R)G_K(x,R) in (25) with x=a/Kx=a/K, for the right-edge region as:

GK(x,k−β)∼2Ksin⁡[πx]sin⁡[π(β+1)x]exp⁡[iK(nπx22+nπxK−nπ4K+πx)].(62)G_K(x,k-\beta)\sim\frac{2}{K}\sin[\pi x]\sin[\pi(\beta+1)x]\exp\left[iK\left(\frac{n\pi x^2}{2}+\frac{n\pi x}{K}-\frac{n\pi}{4K}+\pi x\right)\right]. \tag*{(62)}

The factor (−1)a=eiπa=eiKπx(-1)^a=e^{i\pi a}=e^{iK\pi x} therefore introduces an additional phase term eiKπxe^{iK\pi x} in the above expression. Substituting these leading order forms into (29), we get FRF_R in the right-edge region as:

Fk−β=K(3d−3)/22(d−3)/2πd−1(β+1)d−1∑t∈Z∫01sin⁡[πx]sin⁡[π(β+1)x]eiKΦt(x) dx.(63)F_{k-\beta}=\frac{K^{(3d-3)/2}}{2^{(d-3)/2}\pi^{d-1}(\beta+1)^{d-1}}\sum_{t\in\mathbb{Z}}\int_0^1\sin[\pi x]\sin[\pi(\beta+1)x]e^{iK\Phi_t(x)}\,dx. \tag*{(63)}

where the phase function, after neglecting the terms suppressed by 1/K1/K, which do not contribute to determining the leading-order stationary points, is

Φt(x)=nπx22+πx−2πtx.(64)\Phi_t(x)=\frac{n\pi x^2}{2}+\pi x-2\pi tx. \tag*{(64)}

Thus,

Fk−β=K(3d−3)/22(d−3)/2πd−1(β+1)d−1∑t∈ZIt.(65)F_{k-\beta}=\frac{K^{(3d-3)/2}}{2^{(d-3)/2}\pi^{d-1}(\beta+1)^{d-1}}\sum_{t\in\mathbb{Z}}I_t. \tag*{(65)}

where the integral is given as:

It=∫01sin⁡[πx]sin⁡[π(β+1)x]eiKΦt(x) dx.(66)I_t=\int_0^1\sin[\pi x]\sin[\pi(\beta+1)x]e^{iK\Phi_t(x)}\,dx. \tag*{(66)}

We now apply the stationary phase analysis to ItI_t, with

f(x)=sin⁡[πx]sin⁡[π(β+1)x] ;ϕ(x)=nπx22+πx−2πtx.(67)f(x)=\sin[\pi x]\sin[\pi(\beta+1)x]\ ;\quad\phi(x)=\frac{n\pi x^2}{2}+\pi x-2\pi tx. \tag*{(67)}

The stationary points are determined by ϕ′(x0)=0\phi'(x_0)=0, giving

x0=2t−1n.(68)x_0=\frac{2t-1}{n}. \tag*{(68)}

As in the left-edge analysis, an interior stationary point gives a contribution of order K−1/2K^{-1/2}, whereas an endpoint stationary point for which both f(x0)f(x_0) and f′(x0)f'(x_0) vanish contributes only at order K−3/2K^{-3/2}. We therefore retain only the stationary points satisfying 0<x0<10<x_0<1. This gives

t=1,2,3,…,⌊n2⌋.(69)t=1,2,3,\ldots,\left\lfloor\frac{n}{2}\right\rfloor. \tag*{(69)}

The corresponding stationary points will be x0=(2t−1)/nx_0=(2t-1)/n, for which we will have:

f(x0)=sin⁡[π(2t−1)n]sin⁡[π(β+1)(2t−1)n] ;ϕ(x0)=−π(2t−1)22n ;ϕ′′(x0)=nπ.(70)f(x_0)=\sin\left[\frac{\pi(2t-1)}{n}\right]\sin\left[\frac{\pi(\beta+1)(2t-1)}{n}\right]\ ;\quad\phi(x_0)=-\frac{\pi(2t-1)^2}{2n}\ ;\quad\phi''(x_0)=n\pi. \tag*{(70)}

Therefore, using (41), the leading large-KK asymptotics of ItI_t is

It∼2e−iπK(2t−1)22nKnπeiπ4πsin⁡[π(2t−1)n]sin⁡[π(β+1)(2t−1)n].(71)I_t \sim\frac{\sqrt{2}e^{-\frac{i\pi K(2t-1)^2}{2n}}}{\sqrt{Kn\pi}}e^{\frac{i\pi}{4}}\sqrt{\pi}\sin\left[\frac{\pi(2t-1)}{n}\right]\sin\left[\frac{\pi(\beta+1)(2t-1)}{n}\right]. \tag*{(71)}

Finally, using (65), the leading large-kk asymptotic of the eigenvalues Λk−β=∣Fk−β∣2\Lambda_{k-\beta}=\lvert F_{k-\beta}\rvert^2 in the right-edge region is

Λk−β∼K3d−42d−4nπ2d−2(β+1)2d−2∣∑t=1⌊n2⌋e−iπK(2t−1)22nsin⁡[π(2t−1)n]sin⁡[π(β+1)(2t−1)n]∣2.(72)\Lambda_{k-\beta}\sim\frac{K^{3d-4}}{2^{d-4}n\pi^{2d-2}(\beta+1)^{2d-2}}\left\lvert\sum_{t=1}^{\left\lfloor\frac{n}{2}\right\rfloor}e^{-\frac{i\pi K(2t-1)^2}{2n}}\sin\left[\frac{\pi(2t-1)}{n}\right]\sin\left[\frac{\pi(\beta+1)(2t-1)}{n}\right]\right\rvert^2. \tag*{(72)}

The leading contribution from the right-edge region to Tr⁡[σm]\operatorname{Tr}[\sigma^m] can therefore be written as

Tr⁡[σm]right=∑β=0∞(Λk−β)m,(73)\operatorname{Tr}[\sigma^m]_{\mathrm{right}}=\sum_{\beta=0}^{\infty}(\Lambda_{k-\beta})^m, \tag*{(73)}

where, as in the left-edge analysis, the infinite sum represents the K→∞K\to\infty limit of the finite sum over 0≤β≤k0\leq\beta\leq k, with β\beta held fixed in the right-edge scaling regime. We now use the same decomposition as in the left-edge analysis,

β=np+q,p∈Z≥0,q=0,1,…,n−1.\beta=np+q,\qquad p\in\mathbb{Z}_{\geq0},\qquad q=0,1,\ldots,n-1.

This parametrization covers all non-negative integers β\beta exactly once. Moreover,

sin⁡[π(β+1)(2t−1)n]=sin⁡[π(np+q+1)(2t−1)n]=(−1)psin⁡[π(q+1)(2t−1)n].\sin\left[\frac{\pi(\beta+1)(2t-1)}{n}\right]=\sin\left[\frac{\pi(np+q+1)(2t-1)}{n}\right]=(-1)^p\sin\left[\frac{\pi(q+1)(2t-1)}{n}\right].

Since the factor (−1)p(-1)^p is independent of tt, it factors out of the sum and drops out upon taking the modulus squared in (72). The remaining pp-dependence is therefore entirely contained in the algebraic factor (np+q+1)−2d+2(np+q+1)^{-2d+2}, and its sum gives the same Hurwitz zeta function encountered in the left-edge analysis. We consequently obtain

Tr⁡[σm]right=K3dm−4m2dm−4mπ2dm−2mn2dm−m∑q=0n−1Y(q)mζ(2dm−2m,q+1n),(74)\operatorname{Tr}[\sigma^m]_{\mathrm{right}}=\frac{K^{3dm-4m}}{2^{dm-4m}\pi^{2dm-2m}n^{2dm-m}}\sum_{q=0}^{n-1}Y(q)^m\zeta\left(2dm-2m,\frac{q+1}{n}\right), \tag*{(74)}

where

Y(q)=∣∑t=1⌊n2⌋e−iπK(2t−1)22nsin⁡[π(2t−1)n]sin⁡[π(q+1)(2t−1)n]∣2.(75)Y(q)=\left\lvert\sum_{t=1}^{\left\lfloor\frac{n}{2}\right\rfloor}e^{-\frac{i\pi K(2t-1)^2}{2n}}\sin\left[\frac{\pi(2t-1)}{n}\right]\sin\left[\frac{\pi(q+1)(2t-1)}{n}\right]\right\rvert^2. \tag*{(75)}

Thus, the right-edge contribution scales as

Tr⁡[σm]right∼K3dm−4m,(76)\operatorname{Tr}[\sigma^m]_{\mathrm{right}}\sim K^{3dm-4m}, \tag*{(76)}

with the leading coefficient given by (74).

An interesting feature of the result is that, for odd nn, the left-edge and right-edge contributions to Tr⁡[σm]\operatorname{Tr}[\sigma^m] are identical. This follows from the symmetry ΛR=Λk−R\Lambda_R=\Lambda_{k-R} for odd nn, which exchanges the left and right edges of the representation space and explains the equality of their contributions for odd nn. This translates into an identical leading order asymptotic contribution in left-edge and right-edge contributions. We provide a proof for this below.

Proof that Tr⁡[σm]left=Tr⁡[σm]right\operatorname{Tr}[\sigma^m]_{\mathrm{left}} = \operatorname{Tr}[\sigma^m]_{\mathrm{right}} for odd values of nn

The values of Tr⁡[σm]left\operatorname{Tr}[\sigma^m]_{\mathrm{left}} and Tr⁡[σm]right\operatorname{Tr}[\sigma^m]_{\mathrm{right}} are given in (56) and (74) respectively. All terms are the same except the functions X(q)X(q) and Y(q)Y(q) in the two cases, respectively. These functions are given in (57) and (75) respectively, and we need to prove that X(q)=Y(q)X(q) = Y(q) for odd values of nn. Take n=2w+1n = 2w + 1 with w∈Zw \in\mathbb{Z}. So the upper limit of the sum in both functions becomes ww, and both summations contain the exact same number of terms. In the function YY, we change the summation variable t→st \to s, where s=w+1−ts = w + 1 - t. Since tt runs from 11 to ww, the new variable ss takes values from ww to 11. The two sine terms in the function YY will be modified in terms of the new summation variable as follows:

sin⁡[π(2t−1)n]→sin⁡[2πsn] ;sin⁡[π(q+1)(2t−1)n]→(−1)qsin⁡[2π(q+1)sn].(77)\sin\left[\frac{\pi(2t-1)}{n}\right] \to\sin\left[\frac{2\pi s}{n}\right] \ ;\quad\sin\left[\frac{\pi(q+1)(2t-1)}{n}\right] \to(-1)^q\sin\left[\frac{2\pi(q+1)s}{n}\right]. \tag*{(77)}

These are precisely the sine terms in the function X(q)X(q) in (57) with additional phase (−1)q(-1)^q. Since we need to take the modulus square after the summation, this overall phase will not matter, and we see that the trigonometric factors of Y(q)Y(q) map perfectly to those of X(q)X(q). Let us now check the phase factors. The phase factor in YY under the change of variable t→st \to s where s=w+1−ts = w + 1 - t will change as follows:

exp⁡[−iπk(2t−1)22n]→e−iπkn2e2iπksexp⁡[−2iπks2n]=e−iπkn2exp⁡[−2iπks2n].(78)\exp\left[-\frac{i\pi k(2t-1)^2}{2n}\right] \to e^{-\frac{i\pi kn}{2}}e^{2i\pi ks}\exp\left[-\frac{2i\pi ks^2}{n}\right] = e^{-\frac{i\pi kn}{2}}\exp\left[-\frac{2i\pi ks^2}{n}\right]. \tag*{(78)}

This is again the same phase factor in X(q)X(q) up to an overall phase of e−iπkn2e^{-\frac{i\pi kn}{2}}, which will not matter once we take the modulus square of the sum. Hence, the phase factors of Y(q)Y(q) also map perfectly to those of X(q)X(q). Thus, X(q)=Y(q)X(q) = Y(q) for odd values of nn. Hence the proof.

We have thus determined the leading contributions to Tr⁡[σm]\operatorname{Tr}[\sigma^m] arising from the two edges of the representation space, corresponding to R=O(1)R = O(1) and k−R=O(1)k - R = O(1), respectively. It remains to examine the contribution from the bulk region, where R=O(k)R = O(k) and r=R/Kr = R/K remains finite in the interval 0<r<10 < r < 1. In the following subsection, we analyze the large-KK behavior of the eigenvalues in this region and determine the corresponding contribution to Tr⁡[σm]\operatorname{Tr}[\sigma^m].

Bulk contribution and its asymptotic suppression

For the bulk region, R=O(k)R = O(k) such that r=R/Kr = R/K is finite in the interval 0<r<10 < r < 1. We fix r∈(0,1)r \in(0,1) and set

R=⌊Kr⌋=Kr−{Kr},R = \lfloor Kr \rfloor= Kr - \{Kr\},

where {Kr}\{Kr\} is the fractional part of KrKr, satisfying 0≤{Kr}<10 \le\{Kr\} < 1. In the large-kk limit with r=R/Kr = R/K and x=a/Kx = a/K, the leading forms of the modular matrix elements are:

S0R∼2sin⁡[πr]K,Sa0∼2sin⁡[πx]K,Taa∼eiK(πxK−π4K+πx22).(79)S_{0R} \sim\frac{\sqrt{2}\sin[\pi r]}{\sqrt{K}},\qquad S_{a0} \sim\frac{\sqrt{2}\sin[\pi x]}{\sqrt{K}},\qquad T_{aa} \sim e^{iK\left(\frac{\pi x}{K}-\frac{\pi}{4K}+\frac{\pi x^2}{2}\right)}. \tag*{(79)}

For the remaining modular matrix element, we have

SRa=2sin⁡[π(a+1)(R+1)K]K=2sin⁡[π(Kx+1)(Kr−{Kr}+1)K]K=2sin⁡[π(r−{Kr}K+1K)(Kx+1)]K.(80)S_{Ra} = \frac{\sqrt{2}\sin\left[\frac{\pi(a+1)(R+1)}{K}\right]}{\sqrt{K}} = \frac{\sqrt{2}\sin\left[\frac{\pi(Kx+1)(Kr-\{Kr\}+1)}{K}\right]}{\sqrt{K}} = \frac{\sqrt{2}\sin\left[\pi\left(r-\frac{\{Kr\}}{K}+\frac{1}{K}\right)(Kx+1)\right]}{\sqrt{K}}. \tag*{(80)}

Consequently, using the definition of GK(x,R)G_K(x,R) in (25), its leading form in the bulk region is

GK(x,r)=2Ksin⁡[π(r−{Kr}K+1K)(Kx+1)]sin⁡[πx]exp⁡[iK(nπxK−nπ4K+nπx22)].(81)G_K(x,r) = \frac{2}{K}\sin\left[\pi\left(r-\frac{\{Kr\}}{K}+\frac{1}{K}\right)(Kx+1)\right]\sin[\pi x]\exp\left[iK\left(\frac{n\pi x}{K}-\frac{n\pi}{4K}+\frac{n\pi x^2}{2}\right)\right]. \tag*{(81)}

Applying the Poisson summation formula, the function FRF_{\mathrm{R}} in (24), along with (28), will have the following asymptotics:

FR=Fbulk(r)∼(21−d2sin⁡1−d[πr]Kd−12)(K∑u∈Z∫01GK(x,r)e−2πiKux dx).(82)F_{\mathrm{R}}=F_{\mathrm{bulk}}(r)\sim\left(2^{\frac{1-d}{2}}\sin^{1-d}[\pi r]K^{\frac{d-1}{2}}\right)\left(K\sum_{u\in\mathbb{Z}}\int_{0}^{1}G_{K}(x,r)e^{-2\pi iKux}\,\mathrm{d}x\right). \tag*{(82)}

Now, the argument of the first sine term in the function GK(x,r)G_{K}(x,r) in (81) is at O(K)O(K) because

(r−{Kr}K+1K)(Kx+1)=Krx+r+x−x{Kr}−{Kr}K+1K.\left(r-\frac{\{Kr\}}{K}+\frac{1}{K}\right)(Kx+1)=Krx+r+x-x\{Kr\}-\frac{\{Kr\}}{K}+\frac{1}{K}.

This makes this sine term highly oscillatory in the large-KK limit, and it must be included in the rapidly varying complex exponential. For this, we write:

sin⁡[π(r−{Kr}K+1K)(Kx+1)]=eiπ(Krx−{Kr}K+1K+r−x{Kr}+x)−e−iπ(Krx−{Kr}K+1K+r−x{Kr}+x)2i.\sin\left[\pi\left(r-\frac{\{Kr\}}{K}+\frac{1}{K}\right)(Kx+1)\right]=\frac{e^{i\pi\left(Krx-\frac{\{Kr\}}{K}+\frac{1}{K}+r-x\{Kr\}+x\right)}-e^{-i\pi\left(Krx-\frac{\{Kr\}}{K}+\frac{1}{K}+r-x\{Kr\}+x\right)}}{2i}.

Substituting this decomposition into the Poisson-resummed expression gives two contributions:

Fbulk(r)∼sin⁡1−d[πr]Kd−122d−12i(∑u∈ZIu(1)−∑u∈ZIu(2)),(83)F_{\mathrm{bulk}}(r)\sim\frac{\sin^{1-d}[\pi r]K^{\frac{d-1}{2}}}{2^{\frac{d-1}{2}}i}\left(\sum_{u\in\mathbb{Z}}I_{u}^{(1)}-\sum_{u\in\mathbb{Z}}I_{u}^{(2)}\right), \tag*{(83)}

where the two integrals are given as:

Iu(1)=∫01sin⁡[πx]eiKΦ+(x) dx ;Iu(2)=∫01sin⁡[πx]eiKΦ−(x) dx.(84)I_{u}^{(1)}=\int_{0}^{1}\sin[\pi x]e^{iK\Phi_{+}(x)}\,\mathrm{d}x\ ;\qquad I_{u}^{(2)}=\int_{0}^{1}\sin[\pi x]e^{iK\Phi_{-}(x)}\,\mathrm{d}x. \tag*{(84)}

The terms that are suppressed by 1/K1/K in the phase do not affect the leading-order stationary points. Hence the two phase functions, after omitting such terms, will be given as:

Φ+(x)=nπx22+πrx−2πux ,Φ−(x)=nπx22−πrx−2πux.(85)\Phi_{+}(x)=\frac{n\pi x^{2}}{2}+\pi rx-2\pi ux\ ,\qquad\Phi_{-}(x)=\frac{n\pi x^{2}}{2}-\pi rx-2\pi ux. \tag*{(85)}

For a fixed r∈(0,1)r\in(0,1) and a given value of nn, the stationary points for the two phase functions, determined by Φ+′(x+)=0\Phi_{+}'(x_{+})=0 and Φ−′(x−)=0\Phi_{-}'(x_{-})=0, are

x+=2u−rn ,x−=2u+rn.x_{+}=\frac{2u-r}{n}\ ,\qquad x_{-}=\frac{2u+r}{n}.

Requiring these stationary points to lie in the interior of the integration interval 0<x±<10<x_{\pm}<1 restricts the allowed integer values of uu in the two sums in (83). Since both phase functions satisfy

Φ±′′(x)=nπ,\Phi_{\pm}''(x)=n\pi,

each interior stationary point gives a contribution of order 1/K1/21/K^{1/2}, which follows from the stationary-phase formula (41). In particular, the leading contributions are

Iu(1)∼2eiKΦ+(x+)KΦ+′′(x+)eiπ4πsin⁡[πx+]=2Kne−iπK(r−2u)22neiπ4sin⁡[π(2u−r)n]=C+(u,r,n)KI_{u}^{(1)}\sim\frac{\sqrt{2}e^{iK\Phi_{+}(x_{+})}}{\sqrt{K\Phi_{+}''(x_{+})}}e^{\frac{i\pi}{4}}\sqrt{\pi}\sin[\pi x_{+}]=\frac{\sqrt{2}}{\sqrt{Kn}}e^{-\frac{i\pi K(r-2u)^{2}}{2n}}e^{\frac{i\pi}{4}}\sin\left[\frac{\pi(2u-r)}{n}\right]=\frac{C_{+}(u,r,n)}{\sqrt{K}}
Iu(2)∼2eiKΦ−(x−)KΦ−′′(x−)eiπ4πsin⁡[πx−]=2Kne−iπK(r+2u)22neiπ4sin⁡[π(2u+r)n]=C−(u,r,n)K.(86)I_{u}^{(2)}\sim\frac{\sqrt{2}e^{iK\Phi_{-}(x_{-})}}{\sqrt{K\Phi_{-}''(x_{-})}}e^{\frac{i\pi}{4}}\sqrt{\pi}\sin[\pi x_{-}]=\frac{\sqrt{2}}{\sqrt{Kn}}e^{-\frac{i\pi K(r+2u)^{2}}{2n}}e^{\frac{i\pi}{4}}\sin\left[\frac{\pi(2u+r)}{n}\right]=\frac{C_{-}(u,r,n)}{\sqrt{K}}. \tag*{(86)}

Here C±C_{\pm} denote the corresponding stationary-phase coefficients. These coefficients depend only on KK modulo nn via the phase term, but their magnitudes are independent of KK. Hence, the two integrals asymptotically behave as 1/K1/21/K^{1/2}. Substituting these in (83), we get

Fbulk(r)∼sin⁡1−d[πr]Kd−222d−12i[∑uC+(u,r,n)−∑uC−(u,r,n)].(87)F_{\mathrm{bulk}}(r)\sim\frac{\sin^{1-d}[\pi r]K^{\frac{d-2}{2}}}{2^{\frac{d-1}{2}}i}\left[\sum_{u}C_{+}(u,r,n)-\sum_{u}C_{-}(u,r,n)\right]. \tag*{(87)}

where the two sums run over appropriate integers uu corresponding to their interior stationary points, as we have discussed above. Hence, for fixed r∈(0,1)r \in(0,1), the leading bulk asymptotics can be expressed as

Fbulk(r)∼C(r,n)Kd−22.(88)F_{\mathrm{bulk}}(r) \sim C(r,n)K^{\frac{d-2}{2}} . \tag*{(88)}

where

C(r,n)=sin⁡1−d[πr]2d−12i[∑uC+(u,r,n)−∑uC−(u,r,n)].(89)C(r,n)=\frac{\sin^{1-d}[\pi r]}{2^{\frac{d-1}{2}}i}\left[\sum_{u}C_{+}(u,r,n)-\sum_{u}C_{-}(u,r,n)\right] . \tag*{(89)}

Since ΛR=∣FR∣2\Lambda_R=|F_R|^2, the bulk eigenvalues at fixed rr therefore have the leading scaling

ΛR≡Λbulk(r)∼∣C(r,n)∣2Kd−2.(90)\Lambda_R\equiv\Lambda_{\mathrm{bulk}}(r)\sim|C(r,n)|^2K^{d-2} . \tag*{(90)}

By contrast, the left-edge and right-edge eigenvalues scale as

Λleft∼K3d−4,Λright∼K3d−4,\Lambda_{\mathrm{left}}\sim K^{3d-4},\qquad\Lambda_{\mathrm{right}}\sim K^{3d-4},

as obtained in (52) and (72), respectively. Since (3d−4)>(d−2)(3d-4)>(d-2) for all d≥2d\ge2, the bulk eigenvalues grow more slowly with KK than the edge eigenvalues. Since ΛR=O(Kd−2)\Lambda_R=O(K^{d-2}) for bulk representations RR, the number of such representations is at most O(K)O(K), the bulk contribution to the trace satisfies

Tr⁡[σm]bulk=∑R∈bulk(ΛR)m=O(K)O(Km(d−2))=O(Kdm−2m+1).(91)\operatorname{Tr}[\sigma^m]_{\mathrm{bulk}}=\sum_{R\in\mathrm{bulk}}(\Lambda_R)^m=O(K)O(K^{m(d-2)})=O(K^{dm-2m+1}) . \tag*{(91)}

On the other hand, the left- and right-edge contributions scale as

Tr⁡[σm]left∼K3dm−4m,Tr⁡[σm]right∼K3dm−4m.\operatorname{Tr}[\sigma^m]_{\mathrm{left}}\sim K^{3dm-4m},\qquad\operatorname{Tr}[\sigma^m]_{\mathrm{right}}\sim K^{3dm-4m}.

Since 3dm−4m>dm−2m+13dm-4m>dm-2m+1 for all d≥2d\ge2 and m≥1m\ge1, the bulk contribution is subleading compared with the edge contributions.

Leading asymptotics and residue-class dependence

Having determined the leading large-KK asymptotics of the Tr⁡[σm]\operatorname{Tr}[\sigma^m] in the left-edge, right-edge, and bulk regions, we can combine these contributions to obtain the leading asymptotic behavior of the moments. Since the bulk contribution is subleading, we have

Tr⁡[σm]∼Tr⁡[σm]left+Tr⁡[σm]right.(92)\operatorname{Tr}[\sigma^m]\sim\operatorname{Tr}[\sigma^m]_{\mathrm{left}}+\operatorname{Tr}[\sigma^m]_{\mathrm{right}} . \tag*{(92)}

Adding the two edge contributions obtained in (56) and (74), we find

Tr⁡[σm]∼Cm2dm−4mπ2dm−2mn2dm−mK3dm−4m,(93)\operatorname{Tr}[\sigma^m]\sim\frac{C_m}{2^{dm-4m}\pi^{2dm-2m}n^{2dm-m}}K^{3dm-4m} , \tag*{(93)}

where

Cm=∑q=0n−1[X(q)m+Y(q)m]ζ(2dm−2m,q+1n).(94)C_m=\sum_{q=0}^{n-1}[X(q)^m+Y(q)^m]\zeta\left(2dm-2m,\frac{q+1}{n}\right) . \tag*{(94)}

Here X(q)X(q) and Y(q)Y(q) are defined in (57) and (75), respectively. These functions contain KK-dependent oscillatory phase factors exp⁡[−2iπKs2/n]\exp[-2i\pi Ks^2/n] and exp⁡[−iπK(2t−1)2/n]\exp[-i\pi K(2t-1)^2/n] respectively. Since ss and tt are integers, their values depend on the congruence class of KK modulo nn. Since our primary interest is in the semiclassical limit of the entropies as the Chern–Simons level kk tends to infinity, we express this dependence in terms of the residue class of kk. To this end, we write

k=nk′+ℓ,ℓ=0,1,…,n−1,k=nk'+\ell,\qquad\ell=0,1,\ldots,n-1 ,

and take k′→∞k' \to\infty, with ℓ\ell fixed. Along each such subsequence, K=k+2K=k+2 has a fixed residue class modulo nn, determined by (ℓ+2)(\ell+2) mod nn. The coefficients CmC_m therefore depend on ℓ\ell, and we denote them by Cm(ℓ)C_m^{(\ell)}. Thus, although the leading power of kk is the same for all residue classes, its coefficient can depend on ℓ\ell. So we have

Tr⁡[σm]∼Cm(ℓ)2dm−4mπ2dm−2mn2dm−mk3dm−4m,(95)\operatorname{Tr}[\sigma^m]\sim\frac{C_m^{(\ell)}}{2^{dm-4m}\pi^{2dm-2m}n^{2dm-m}}k^{3dm-4m}, \tag*{(95)}

where

Cm(ℓ)=∑q=0n−1[Xℓ(q)m+Yℓ(q)m]ζ(2dm−2m,q+1n),(96)C_m^{(\ell)}=\sum_{q=0}^{n-1}\left[X_\ell(q)^m+Y_\ell(q)^m\right]\zeta\left(2dm-2m,\frac{q+1}{n}\right), \tag*{(96)}

with ℓ\ell dependent functions XX and YY given as

Xℓ(q)=∣∑s=1⌊n−12⌋e−2iπ(ℓ+2)s2nsin⁡[2πsn]sin⁡[2π(q+1)sn]∣2X_\ell(q)=\left|\sum_{s=1}^{\left\lfloor\frac{n-1}{2}\right\rfloor}e^{-\frac{2i\pi(\ell+2)s^2}{n}}\sin\left[\frac{2\pi s}{n}\right]\sin\left[\frac{2\pi(q+1)s}{n}\right]\right|^2
Yℓ(q)=∣∑t=1⌊n2⌋e−2iπ(ℓ+2)t(t−1)nsin⁡[π(2t−1)n]sin⁡[π(q+1)(2t−1)n]∣2.(97)Y_\ell(q)=\left|\sum_{t=1}^{\left\lfloor\frac{n}{2}\right\rfloor}e^{-\frac{2i\pi(\ell+2)t(t-1)}{n}}\sin\left[\frac{\pi(2t-1)}{n}\right]\sin\left[\frac{\pi(q+1)(2t-1)}{n}\right]\right|^2. \tag*{(97)}

Large-kk limit of the entanglement entropy for the Td,dnT_{d,dn} torus links

In this section, we use the asymptotic expressions for the density-matrix moments derived in the previous section to obtain the large-kk limits of the Rényi and entanglement entropies for the Td,dnT_{d,dn} torus-link states. Although the leading moments exhibit a common power-law dependence on kk, their coefficients retain a dependence on the residue class of kk modulo nn, which can lead to distinct limiting values of the entropies along different residue-class subsequences.

Using the leading large-kk asymptotics derived in (95), we see that

Tr⁡[σm]Tr⁡[σ]m∼Cm(ℓ)k3dm−4m+O(k3dm−4m−1)[C1(ℓ)k3d−4+O(k3d−5)]m→k→∞k≡ℓ (mod⁡n)Cm(ℓ)[C1(ℓ)]m.(98)\frac{\operatorname{Tr}[\sigma^m]}{\operatorname{Tr}[\sigma]^m}\sim\frac{C_m^{(\ell)}k^{3dm-4m}+O(k^{3dm-4m-1})}{\left[C_1^{(\ell)}k^{3d-4}+O(k^{3d-5})\right]^m}\xrightarrow[k\to\infty]{k\equiv\ell\ (\operatorname{mod} n)}\frac{C_m^{(\ell)}}{\left[C_1^{(\ell)}\right]^m}. \tag*{(98)}

The powers of kk cancel in this ratio. Consequently, the Rényi entropy has a finite semiclassical limit along each fixed residue-class subsequence:

Rm,∞(ℓ)≡lim⁡k→∞Rm(ℓ)=lim⁡k→∞11−mln⁡[Tr⁡[σm]Tr⁡[σ]m]=11−mln⁡[Cm(ℓ)/(C1(ℓ))m].(99)\mathcal{R}_{m,\infty}^{(\ell)}\equiv\lim_{k\to\infty}\mathcal{R}_m^{(\ell)}=\lim_{k\to\infty}\frac{1}{1-m}\ln\left[\frac{\operatorname{Tr}[\sigma^m]}{\operatorname{Tr}[\sigma]^m}\right]=\frac{1}{1-m}\ln\left[C_m^{(\ell)}/\left(C_1^{(\ell)}\right)^m\right]. \tag*{(99)}

The entanglement entropy is obtained from the Rényi entropy by taking the limit m→1m\to1, which gives

EE=−ddm[(1−m)Rm]∣m=1.(100)\mathrm{EE}=-\left.\frac{d}{dm}\left[(1-m)R_m\right]\right|_{m=1}. \tag*{(100)}

Consequently, the large-kk limit of the entanglement entropy for the Td,dnT_{d,dn} torus-link states is

EE∞(ℓ)=ln⁡C1(ℓ)−1C1(ℓ)dCm(ℓ)dm∣m=1,(101)\mathrm{EE}_\infty^{(\ell)}=\ln C_1^{(\ell)}-\left.\frac{1}{C_1^{(\ell)}}\frac{dC_m^{(\ell)}}{dm}\right|_{m=1}, \tag*{(101)}

where ℓ=0,1,2,…,n−1\ell=0,1,2,\ldots,n-1 labels the residue class k≡ℓ(modn)k\equiv\ell\pmod n. The notation EE∞(ℓ)\mathrm{EE}_\infty^{(\ell)} allows for the possibility that the limiting entanglement entropy depends on this residue class. Although Cm(ℓ)C_m^{(\ell)} was initially defined for integer Rényi index mm, its explicit expression involves powers Xℓ(q)mX_\ell(q)^m and Yℓ(q)mY_\ell(q)^m, together with a Hurwitz zeta function whose first argument depends continuously on mm. This expression therefore admits a natural continuation to real mm in a neighbourhood of m=1m = 1, allowing the derivative in (101) to be evaluated.

The expressions derived above determine the large-kk limits of the Rényi and entanglement entropies in terms of finite sums involving the coefficients Xℓ(q)X_{\ell}(q), Yℓ(q)Y_{\ell}(q), and the Hurwitz zeta function. In particular, they reveal how the limiting entropies depend on the link parameters and on the residue class of kk modulo nn. In the following section, we evaluate these expressions for representative values of nn and dd, and investigate the resulting patterns of residue-class degeneracies and the dependence of the limiting entanglement entropy on the number of link components.

Examples and numerical results

In this section, we examine the analytical results obtained earlier through explicit evaluations and numerical comparisons. We first study the dependence of the limiting entanglement entropy on the residue class of kk modulo nn, identifying cases in which different residue classes yield identical limiting values. We then investigate how these limiting values vary with the number of link components dd and the winding parameter nn. These examples illustrate the structure of the large-kk limits and provide a basis for examining the patterns that emerge as the link parameters are varied.

Residue-class dependence and degeneracies

In Table 1, we present the limiting entanglement entropies EE∞(ℓ)\mathrm{EE}^{(\ell)}_{\infty}, calculated using (101), for various values of nn and ℓ\ell, with d=2d = 2. For a given nn, different residue classes ℓ\ell can yield identical limiting entropy values. We group such residue classes together and list only the distinct limiting values. For example, for n=5n = 5, we obtain

nnDistinct limiting valuesResidue classes ℓ\ell
20.888{0,1}\{0,1\}
31.938{0,1,2}\{0,1,2\}
41.581, 1.524{0,2},{1,3}\{0,2\},\{1,3\}
51.828, 2.439, 1.341{0,1},{2,4},{3}\{0,1\},\{2,4\},\{3\}
61.388, 1.245, 1.929, 1.189{0,2},{1},{3,5},{4}\{0,2\},\{1\},\{3,5\},\{4\}
71.774, 2.104, 2.777, 1.083{0,3},{1,2},{4,6},{5}\{0,3\},\{1,2\},\{4,6\},\{5\}
81.581, 1.185, 0.888, 2.218, 1.007{0,4},{1,3},{2},{5,7},{6}\{0,4\},\{1,3\},\{2\},\{5,7\},\{6\}
91.741, 1.938, 2.342, 3.030, 0.952{0,5},{1,4},{2,3},{6,8},{7}\{0,5\},\{1,4\},\{2,3\},\{6,8\},\{7\}
101.466, 1.365, 1.231, 0.651, 2.443, 0.911{0,6},{1,5},{2,4},{3},{7,9},{8}\{0,6\},\{1,5\},\{2,4\},\{3\},\{7,9\},\{8\}

Table 1. Distinct limiting values of entanglement entropy for different values of nn, together with the corresponding degenerate residue classes. We have fixed d=2d = 2. The limiting values of EE are calculated using Eq. (101) and reported to three decimal places.

EE∞(ℓ)(n=5,d=2)≈{1.828, for ℓ=0,12.439, for ℓ=2,41.341, for ℓ=3(102)\mathrm{EE}^{(\ell)}_{\infty}(n = 5,d = 2) \approx \begin{cases} 1.828 & ,\ \text{for }\ell= 0,1 \\ 2.439 & ,\ \text{for }\ell= 2,4 \\ 1.341 & ,\ \text{for }\ell= 3 \end{cases} \tag*{(102)}

Figure 1 shows the variation of the entanglement entropy with the level kk for n=5n = 5 and d=2d = 2. The black curve connects the entropy values obtained at successive levels, while the coloured markers distinguish the five residue classes k=ℓ mod 5k = \ell\bmod5. As kk increases, the entropy exhibits a clear convergence pattern: the five residue-class subsequences approach three distinct limiting values, approximately 1.828, 2.439, and 1.341. Specifically, the classes ℓ=0,1\ell= 0,1 converge to 1.828, the classes ℓ=2,4\ell= 2,4 converge to 2.439, and the class ℓ=3\ell= 3 converges to 1.341, as indicated by the horizontal dashed lines. This behavior illustrates how the residue-class dependence persists in the large-kk limit, while distinct residue classes can yield identical limiting entanglement entropies.

Figure 1. Entanglement entropy as a function of the level kk for n=5n = 5 and d=2d = 2. The colored markers distinguish the residue classes corresponding to k=ℓ mod 5k = \ell\bmod5. The horizontal dashed lines indicate the three distinct limiting values, EE∞(ℓ)≈1.828\mathrm{EE}^{(\ell)}_{\infty} \approx1.828, 2.439, and 1.341, corresponding to the residue-class groups {0,1}\{0,1\}, {2,4}\{2,4\}, and {3}\{3\}, respectively.

Variation of limiting entanglement entropy with winding parameter

In this subsection, we investigate the dependence of the limiting entanglement entropy EE∞(ℓ)\mathrm{EE}^{(\ell)}_{\infty} on the winding parameter nn, keeping the number of link components fixed at d=2d = 2. We focus on the residue class ℓ=0\ell= 0, which corresponds to restricting the level kk to integer multiples of nn and then taking the limit k→∞k \to\infty. Using the analytical expression derived in (101), we evaluate EE∞(0)\mathrm{EE}^{(0)}_{\infty} for n=2,3,…,100n = 2,3,\ldots,100, and the values are presented in the Figure 2.

Figure 2. Limiting entanglement entropy EE∞(0)\mathrm{EE}_{\infty}^{(0)} as a function of the winding parameter nn for d=2d=2. The data are grouped according to the residue class nn mod 4. When nn is a multiple of 4, the limiting entropy is independent of nn and approximately equals 1.58099. The other three subsequences approach this value as nn increases, as suggested by the numerical results.

Figure 2 exhibits four distinct patterns corresponding to the subsequences classified by n mod 4n \bmod4. For n=4Zn = 4\mathbb{Z}, the limiting entanglement entropy is independent of nn and is given exactly by

EE∞(0)=−2γ−143ln⁡2+24ln⁡A,\mathrm{EE}^{(0)}_{\infty} = -2\gamma- \frac{14}{3}\ln2 + 24\ln A ,

where γ\gamma is the Euler–Mascheroni constant and AA is the Glaisher–Kinkelin constant. Numerically, this gives EE∞(0)≈1.58099\mathrm{EE}^{(0)}_{\infty} \approx1.58099. The subsequences with n=4Z+1n = 4\mathbb{Z} + 1 and n=4Z+3n = 4\mathbb{Z} + 3 decrease monotonically towards this value as nn increases, whereas the subsequence with n=4Z+2n = 4\mathbb{Z} + 2 increases from below and gradually approaches it. Although the four subsequences differ appreciably at small nn, their values become progressively closer as nn increases. The numerical results up to n=100n = 100 therefore suggest that all four subsequences converge to the same value. This behavior indicates that the dependence on the residue class of nn, which is prominent at small nn, becomes less pronounced in the large-nn regime.

Variation of limiting entanglement entropy with number of link components

In this subsection, we investigate the dependence of the limiting entanglement entropy EE∞(ℓ)\mathrm{EE}_{\infty}^{(\ell)} on the number of link components dd, for fixed winding parameter nn and residue class ℓ\ell of the level kk modulo nn. We consider the cases n=5n=5 and n=6n=6, for which we evaluate EE∞(ℓ)\mathrm{EE}_{\infty}^{(\ell)} for d=2,…,10d=2,\ldots,10 using the analytical expression derived in (101). The resulting behavior is shown in Figure 3. The numerical results reveal a clear dependence on the parity of nn and on the residue class ℓ\ell, and suggest a connection with the large-party limit studied in [16]. We will explore these next.

Figure 3. Limiting entanglement entropy EE∞(ℓ)\mathrm{EE}^{(\ell)}_{\infty} as a function of the number of link components dd for n=5n=5 (left panel) and n=6n=6 (right panel). Different curves correspond to the residue classes ℓ≡k(modn)\ell\equiv k \pmod{n}. The horizontal dashed line indicates the value ln⁡2\ln2.

For n=5n=5, the limiting entanglement entropy EE∞(ℓ)\mathrm{EE}_{\infty}^{(\ell)} for all residue classes ℓ=0,1,2,3,4\ell=0,1,2,3,4 approaches ln⁡2\ln2 as dd increases. In contrast, for n=6n=6, the subsequences corresponding to ℓ=1,3,5\ell=1,3,5 approach zero, while the remaining residue classes exhibit a different behaviour. These observations indicate that the large-dd limit depends on both the winding parameter and the residue class of the level. In particular, the numerical results suggest that the limiting entropy approaches ln⁡2\ln2 for odd nn, whereas for even nn it vanishes for odd residue classes ℓ\ell. To understand these patterns, we next derive the large-dd limit directly from the limiting density-matrix moments.

Analytical large-dd limit. To derive the large-dd behavior, we start from the leading order coefficient in the large-kk expansion of the moments Tr⁡[σm]\operatorname{Tr}[\sigma^m]. These coefficients Cm(ℓ)C_m^{(\ell)} are given in (96). The dd-dependence in these coefficients comes explicitly from the Hurwitz zeta function,

ζ(2m(d−1),q+1n).\zeta\left(2m(d-1),\frac{q+1}{n}\right).

Now, for ζ(s,a)\zeta(s,a) with s→∞s \to\infty and a fixed a>0a > 0, the leading order behavior is given as

ζ(s,a)=∑i=0∞1(a+i)s→s→∞1as.(103)\zeta(s,a)=\sum_{i=0}^{\infty}\frac{1}{(a+i)^s}\xrightarrow{s\to\infty}\frac{1}{a^s}. \tag*{(103)}

Since m≥1m \ge1, the first argument of the Hurwitz zeta function s=2m(d−1)→∞s=2m(d-1)\to\infty as d→∞d\to\infty. Thus, for a given value of qq and nn, the large-dd expansion can be

ζ(2m(d−1),q+1n)∼(nq+1)2m(d−1).(104)\zeta\left(2m(d-1),\frac{q+1}{n}\right)\sim\left(\frac{n}{q+1}\right)^{2m(d-1)}. \tag*{(104)}

Hence, from (96), the large-dd behavior of Cm(ℓ)C_m^{(\ell)} will be dominated by the q=0q=0 term, provided that Xℓ(0)+Yℓ(0)>0X_{\ell}(0)+Y_{\ell}(0)>0, which we have verified for various nn and ℓ\ell values. The subleading contributions from q≥1q\geq1 are exponentially suppressed relative to the q=0q=0 contribution as d→∞d\to\infty, and remain negligible upon differentiation with respect to mm in a neighborhood of m=1m=1. Thus, in the large-dd regime, we will have:

Cm(ℓ)∼n2m(d−1)(Xℓ(0)m+Yℓ(0)m),(105)C_m^{(\ell)}\sim n^{2m(d-1)}\left(X_{\ell}(0)^m+Y_{\ell}(0)^m\right), \tag*{(105)}

where,

Xℓ(0)=∣∑s=1⌊n−12⌋e−2iπ(ℓ+2)snsin⁡2[2πsn]∣2;Yℓ(0)=∣∑t=1⌊n2⌋e−2iπ(ℓ+2)t(t−1)nsin⁡2[π(2t−1)n]∣2.(106)X_{\ell}(0)=\left|\sum_{s=1}^{\left\lfloor\frac{n-1}{2}\right\rfloor}e^{-\frac{2i\pi(\ell+2)s}{n}}\sin^2\left[\frac{2\pi s}{n}\right]\right|^2;\qquad Y_{\ell}(0)=\left|\sum_{t=1}^{\left\lfloor\frac{n}{2}\right\rfloor}e^{-\frac{2i\pi(\ell+2)t(t-1)}{n}}\sin^2\left[\frac{\pi(2t-1)}{n}\right]\right|^2. \tag*{(106)}

Substituting this into the (101), the large-dd limit will be:

lim⁡d→∞EE∞(ℓ)=ln⁡[Xℓ(0)+Yℓ(0)]−Xℓ(0)ln⁡Xℓ(0)+Yℓ(0)ln⁡Yℓ(0)Xℓ(0)+Yℓ(0).(107)\lim_{d\to\infty}\mathrm{EE}^{(\ell)}_{\infty}=\ln\left[X_{\ell}(0)+Y_{\ell}(0)\right]-\frac{X_{\ell}(0)\ln X_{\ell}(0)+Y_{\ell}(0)\ln Y_{\ell}(0)}{X_{\ell}(0)+Y_{\ell}(0)}. \tag*{(107)}

Introducing the normalized weights,

pℓ=Xℓ(0)Xℓ(0)+Yℓ(0),qℓ=Yℓ(0)Xℓ(0)+Yℓ(0),(108)p_{\ell}=\frac{X_{\ell}(0)}{X_{\ell}(0)+Y_{\ell}(0)},\qquad q_{\ell}=\frac{Y_{\ell}(0)}{X_{\ell}(0)+Y_{\ell}(0)}, \tag*{(108)}

with pℓ+qℓ=1p_{\ell}+q_{\ell}=1, this limit can be written as:

lim⁡d→∞EE∞(ℓ)=−(pℓln⁡pℓ+qℓln⁡qℓ).(109)\lim_{d\to\infty}\mathrm{EE}^{(\ell)}_{\infty}=-(p_{\ell}\ln p_{\ell}+q_{\ell}\ln q_{\ell}). \tag*{(109)}

Thus, the large-dd limit of the limiting entanglement entropy is precisely the binary entropy associated with the normalized weights pℓp_\ell and qℓq_\ell. Consequently, it satisfies

0≤lim⁡d→∞EE∞(ℓ)≤ln⁡2.0 \le\lim_{d\to\infty} \mathrm{EE}^{(\ell)}_\infty\le\ln2 .

The upper bound is attained when the two weights are equal, while the entropy vanishes when either of the two weights is zero.

Dependence on the parity of nn. When nn is odd, we have Xℓ(0)=Yℓ(0)X_\ell(0)=Y_\ell(0), thus the two weights pℓp_\ell and qℓq_\ell are equal. Hence, the large-dd limit of the limiting entanglement entropy achieves its maximum value of ln⁡2\ln2. This is also evident from the n=5n=5 plot in Figure 3. To prove that Xℓ(0)X_\ell(0) and Yℓ(0)Y_\ell(0) are identical for all odd nn values, we write the two coefficients in (5.5) as:

Xℓodd(0)=∣∑s=1n−12e−2iπ(ℓ+2)snsin⁡2[2πsn]∣2 ;Yℓodd(0)=∣∑t=1n−12e−2iπ(ℓ+2)t(t−1)nsin⁡2[π(2t−1)n]∣2.(110)X_\ell^{\mathrm{odd}}(0)=\left|\sum_{s=1}^{\frac{n-1}{2}}e^{-\frac{2i\pi(\ell+2)s}{n}}\sin^2\left[\frac{2\pi s}{n}\right]\right|^2 \ ;\qquad Y_\ell^{\mathrm{odd}}(0)=\left|\sum_{t=1}^{\frac{n-1}{2}}e^{-\frac{2i\pi(\ell+2)t(t-1)}{n}}\sin^2\left[\frac{\pi(2t-1)}{n}\right]\right|^2 . \tag*{(110)}

In Yℓodd(0)Y_\ell^{\mathrm{odd}}(0), we make the change of summation variable s=n+12−ts=\frac{n+1}{2}-t. Under this transformation, 2t−1=n−2s2t-1=n-2s and t(t−1)=s2−ns+n2−14t(t-1)=s^2-ns+\frac{n^2-1}{4}. It follows that the sine-squared factor becomes identical to that in Xℓodd(0)X_\ell^{\mathrm{odd}}(0), while the exponential factor differs only by an overall phase independent of ss. Since this phase has unit modulus, we obtain Xℓodd(0)=Yℓodd(0)>0X_\ell^{\mathrm{odd}}(0)=Y_\ell^{\mathrm{odd}}(0)>0. As a result, pℓ=qℓ=1/2p_\ell=q_\ell=1/2 and we get

lim⁡d→∞EE∞(ℓ)=ln⁡2,(for odd n).\lim_{d\to\infty}\mathrm{EE}^{(\ell)}_\infty=\ln2,\qquad\text{(for odd $n$)} .

On the other hand, when nn is even and k≡ℓ=odd(modn)k\equiv\ell=\text{odd}\pmod n, one of the coefficients Xℓ(0)X_\ell(0) or Yℓ(0)Y_\ell(0) vanishes, and hence the corresponding large-dd limit of the limiting entanglement entropy vanishes. This is clear from the n=6n=6 plot in Figure 3. To see this behavior, consider n=2u=evenn=2u=\text{even} and write the two coefficients as Xℓeven(0)=∣SX∣2X_\ell^{\mathrm{even}}(0)=|S_X|^2 and Yℓeven(0)=∣SY∣2Y_\ell^{\mathrm{even}}(0)=|S_Y|^2, where:

SX=∑s=1u−1e−iπ(ℓ+2)s2usin⁡2[πsu] ;SY=∑t=1ue−iπ(ℓ+2)t(t−1)usin⁡2[π(2t−1)2u].(111)S_X=\sum_{s=1}^{u-1}e^{-\frac{i\pi(\ell+2)s^2}{u}}\sin^2\left[\frac{\pi s}{u}\right]\ ;\qquad S_Y=\sum_{t=1}^{u}e^{-\frac{i\pi(\ell+2)t(t-1)}{u}}\sin^2\left[\frac{\pi(2t-1)}{2u}\right] . \tag*{(111)}

In SXS_X, we make the change of summation variable s′=u−ss'=u-s. Using the invariance of the sine-squared factor under this transformation, we obtain SX=e−iπℓuSXS_X=e^{-i\pi\ell u}S_X. Similarly, in SYS_Y, we make the change of variable t′=u+1−tt'=u+1-t, which gives SY=e−iπℓ(u+1)SYS_Y=e^{-i\pi\ell(u+1)}S_Y. We see that for odd ℓ\ell, SXS_X vanishes for odd uu and SYS_Y vanishes for even uu. Consequently, for even nn and odd ℓ\ell, exactly one of Xℓ(0)X_\ell(0) and Yℓ(0)Y_\ell(0) vanishes, while the other is nonzero. We therefore obtain

lim⁡d→∞EE∞(ℓ)=0,(for n even and ℓ odd).(112)\lim_{d\to\infty}\mathrm{EE}^{(\ell)}_\infty=0,\qquad\text{(for $n$ even and $\ell$ odd)}. \tag*{(112)}

The double limit commutes. We now compare our results with those obtained in earlier work [16], where the large-party (large-dd) limit of the entanglement entropy for torus links was studied first, followed by its semiclassical (large-kk) limit for the class of Td,dnT_{d,dn} torus links. We find that our result in (5.8) exactly reproduces the result obtained in Ref. [16]. In particular, the two results take the form of a binary entropy involving two normalized weights, which precisely coincide with our weights pℓp_\ell and qℓq_\ell. This agreement establishes that the two iterated limits commute:

lim⁡d→∞(lim⁡k→∞EE)=lim⁡k→∞(lim⁡d→∞EE)=−(pℓln⁡pℓ+qℓln⁡qℓ).(113)\lim_{d\to\infty}\left(\lim_{k\to\infty}\mathrm{EE}\right)=\lim_{k\to\infty}\left(\lim_{d\to\infty}\mathrm{EE}\right)=-(p_\ell\ln p_\ell+q_\ell\ln q_\ell) . \tag*{(113)}

Conclusion and Discussion

In this work, we have investigated the semiclassical asymptotics of entanglement entropy for quantum states associated with Td,dnT_{d,dn} torus links in SU(2)k\mathrm{SU}(2)_k Chern–Simons theory. Our analysis focused on the large-level limit k→∞k \to\infty, with the number of link components dd and the winding parameter nn held fixed. By exploiting the modular-matrix representation of the link-state coefficients, we derived the asymptotic behavior of the moments of the reduced density matrix and, consequently, obtained an analytical expression for the limiting entanglement entropy.

To determine the large-kk behavior of the moments, we divide the representation space into three regions: the left-edge region, where the representation label RR remains finite as k→∞k \to\infty; the right-edge region, where k−Rk-R remains finite; and the bulk region, where RR scales with kk, such that R/kR/k approaches a finite value strictly between zero and one. A central result of our analysis is that the leading asymptotic behavior of the density-matrix moments is governed by contributions from the two edges of the integrable-representation space, whereas the bulk contribution is subleading. Using these asymptotic expressions for the moments, we obtain an analytical expression for the large-kk limit of the entanglement entropy for Td,dnT_{d,dn} torus links in terms of finite sums involving Hurwitz zeta functions. Although the entropy converges in the large-kk limit, its limiting value generally depends on the residue class ℓ=k mod n\ell= k \bmod n. This dependence reflects the sensitivity of the semiclassical limit to the arithmetic properties of the Chern–Simons level.

We have further explored the dependence of the limiting entanglement entropy on the winding parameter nn, keeping the number of link components dd fixed. Our numerical results reveal that different residue classes can lead to distinct limiting values, with certain classes exhibiting degeneracies. The degenerate classes, together with the corresponding numerical values of the limiting entanglement entropy, are listed in Table 1 for several small values of nn.

We have also investigated the behavior of the limiting entropy in the large-dd regime, with nn held fixed. By employing the large-dd asymptotic form of the Hurwitz zeta functions, we derive an analytical expression for this limit. As shown in Eq. (109), the resulting expression takes the form of a binary entropy involving two normalized weights associated with the left- and right-edge contributions. We find that these weights are equal for odd nn, yielding the maximum binary entropy ln⁡2\ln2. For even nn and even residue classes ℓ\ell, one of the weights vanishes, resulting in a zero large-dd limit of the entropy.

An interesting outcome is the agreement between the two possible orders of taking the large-kk and large-dd limits. We have explicitly verified that the expression obtained by first taking the semiclassical limit (k→∞k \to\infty) and subsequently the large-party limit (d→∞d \to\infty) coincides with the result obtained by reversing the order of these limits, as previously studied in [16]. Thus, for the class of Td,dnT_{d,dn} links considered here, the two iterated limits commute. This agreement provides a nontrivial consistency check of our asymptotic analysis and connects the finite-party semiclassical description with the previously obtained large-party result.

The present analysis also raises several interesting questions for future investigation. It would be worthwhile to extend the asymptotic framework to more general torus links Tp,qT_{p,q}, for which the structure of the link-state coefficients may lead to a richer pattern of contributions from different regions of the representation space. Another natural direction is to investigate whether the edge-dominance mechanism and the associated residue-class dependence persist for other compact gauge groups and more general classes of link states. Finally, a systematic analytical study of the large-nn behavior of the limiting entanglement entropy could provide a deeper understanding of the numerical trends observed in this work. These questions may help clarify the extent to which the asymptotic structure found here is a special feature of Td,dnT_{d,dn} links or reflects a more general property of topological entanglement in Chern–Simons theory.

Acknowledgment

This work is supported by the ‘ANRF (erstwhile SERB)’, Government of India, through the Start-up Research Grant (SRG) [Grant No. SRG/2023/001023]. VG acknowledges the ‘Council of Scientific and Industrial Research (CSIR)’ for its support through the research fellowship.

References

  1. [1]E. Witten, Quantum Field Theory and the Jones Polynomial, Commun. Math. Phys. 121 (1989) 351.
  2. [2]V. Balasubramanian, J. R. Fliss, R. G. Leigh and O. Parrikar, Multi-Boundary Entanglement in Chern-Simons Theory and Link Invariants, JHEP 04 (2017) 061 [1611.05460].DOI
  3. [3]S. Dwivedi, V. K. Singh, S. Dhara, P. Ramadevi, Y. Zhou and L. K. Joshi, Entanglement on linked boundaries in Chern-Simons theory with generic gauge groups, JHEP 02 (2018) 163 [1711.06474].DOI
  4. [4]V. Balasubramanian, M. DeCross, J. Fliss, A. Kar, R. G. Leigh and O. Parrikar, Entanglement Entropy and the Colored Jones Polynomial, JHEP 05 (2018) 038 [1801.01131].DOI
  5. [5]G. Salton, B. Swingle and M. Walter, Entanglement from Topology in Chern-Simons Theory, Phys. Rev. D95 (2017) 105007 [1611.01516].DOI
  6. [6]L.-Y. Hung, Y.-S. Wu and Y. Zhou, Linking Entanglement and Discrete Anomaly, JHEP 05 (2018) 008 [1801.04538].DOI
  7. [7]Y. Zhou, 3d One-form Mixed Anomaly and Entanglement Entropy, JHEP 07 (2019) 091 [1904.06924].DOI
  8. [8]S. Dwivedi, A. Addazi, Y. Zhou and P. Sharma, Multi-boundary entanglement in Chern-Simons theory with finite gauge groups, JHEP 04 (2020) 158 [2003.01404].DOI
  9. [9]S. Dwivedi, V. K. Singh and A. Roy, Semiclassical limit of topological Rényi entropy in 3d Chern-Simons theory, JHEP 12 (2020) 132 [2007.07033].DOI
  10. [10]A. Dwivedi, S. Dwivedi, B. P. Mandal, P. Ramadevi and V. K. Singh, Topological entanglement and hyperbolic volume, JHEP 10 (2021) 172 [2106.03396].DOI
  11. [11]S. Dwivedi, Topological entanglement and number theory, Phys. Rev. D 113 (2026) 086009 [2410.01492].DOI
  12. [12]A. Saini and S. Dwivedi, Analyzing reduced density matrices in su(2) chern-simons theory, in Proceedings of the International Conference on Knots, Quivers and Beyond (ICKQB 2025), pp. 65–77, Atlantis Press, 2025, DOI.arxiv.org/abs/2504.10098
  13. [13]C. Cummings, Entanglement in typical states of Chern-Simons theory, Phys. Rev. D 112 (2025) 025014 [2503.21894].DOI
  14. [14]M.-K. Yuan, M. Li and Y. Zhou, Multi-entropy from Linking in Chern-Simons Theory, 2510.18408.DOI
  15. [15]S. Akella, A. Gadde and J. Pandey, Multi-invariants in stabilizer states, 2601.16258.arxiv.org/abs/2601.16258
  16. [16]S. Sain and S. Dwivedi, Large-party limit of topological entanglement entropy in Chern-Simons theory, JHEP 05 (2026) 295 [2601.00406].DOI
  17. [17]P. Caputa, S. Purkayastha, A. Saha and P. Sulkowski, Musings on SVD and pseudo entanglement entropies, JHEP 11 (2024) 103 [2408.06791].DOI
  18. [18]S. Stevan, Chern-Simons Invariants of Torus Links, Annales Henri Poincare 11 (2010) 1201 [1003.2861].
  19. [19]A. Brini, B. Eynard and M. Marino, Torus knots and mirror symmetry, Annales Henri Poincare 13 (2012) 1873 [1105.2012].
  20. [20]E. T. Copson, The method of stationary phase, p. 27–35. Cambridge Tracts in Mathematics. Cambridge University Press, 1965.DOI

Paper details

vlnc:W_4r4bcfx4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents