betaFeedback

WorksPhysics and Astronomy

Analytical treatment of reentrant localization via Lyapunov branch structure

Abstract

In reentrant localization, states become localized under a growing quasiperiodic potential partly become extended again before localizing once more. So far this has been demonstrated only numerically. Here we prove it in a chain of weakly coupled dimers in which we vary only the overall strength of a fixed onsite potential. As the strength increases, the lower orbital of each dimer shifts its weight between the two sites. At a particular strength, the quasiperiodic potential it feels cancels, while the hopping between dimers stays finite. When the energy scales are well separated, we prove that, over a finite range of potential strengths, all states in a central part of the lower band, one third of its states, have a vanishing Lyapunov exponent that is stable against small changes of phase, the certification of extended states. On both sides of this range, all states are localized. Numerical calculations confirm that the recovery covers the whole lower band, while the upper band stays localized.

Introduction

Anderson localization shows how in a disordered medium, interference can confine waves and suppress diffusion [1]. In one dimension, uncorrelated onsite disorder produces localization even at weak disorder [2]. Quasiperiodic potentials provide a different setting: their deterministic correlations allow localization transitions at finite modulation strength. The Aubry-André (AA) model [3, 4, 5] is the standard example, with a self-dual transition criteria, the potential amplitude equals twice the hopping [6]. Localization in such models is most often characterized through the Lyapunov exponent, or equivalently the localization length, and through participation ratios and the multifractal structure of eigenstates [7, 8]. Beyond these probes, localization in AA-type models has also been studied through bulk polarization and geometric Binder cumulants [9, 10].

Extensions of the AA model reveal a richer spectral structure. Longer-range hopping and changes in the onsite potential can produce mobility edges, separating localized and extended states at different energies [11, 12, 13, 14, 15, 16, 17, 18, 19]. Non-Hermitian models extend these questions to asymmetric hopping and complex onsite potentials, where localization is connected to changes in the complex spectrum and, in some models, symmetry breaking [20, 21, 22, 23, 24, 25, 26, 27, 28, 29]. The AA model and its generalizations have also been studied on ladders [30, 31], two-dimensional lattices [32, 33], diamond chains [34], and helical structures [35, 36]. These examples show that localization depends on both the strength and the structure of the modulation.

Quasiperiodic localization has been observed with ultracold atoms in optical lattices [37] and light in photonic waveguide arrays [38]. In atomic systems, interactions can be tuned while keeping the spatial potential fixed. This connects quasiperiodic localization to many-body localization, where an isolated interacting system can retain local memory and fail to thermalize [39, 40]. Experiments with interacting fermions in a quasiperiodic optical lattice have observed a density imbalance that persists over the accessible times [41]. Related studies of interacting localization have used trapped-ion simulators [42] and superconducting circuits [43].

At the single-particle level, the standard AA model gives a monotonic response to increasing modulation: its states become localized beyond a single threshold and remain localized as the potential grows. Reentrant localization changes this sequence. A further increase in an onsite control can restore spatial extent in part of the spectrum, followed by localization again at larger coupling. Such behavior has been reported in dimerized quasiperiodic chains [44, 45, 46, 47, 48], models with an additional staggered potential [49, 50, 51], lattices with longer-range hopping [52, 53, 54], helical systems [55, 56], and non-Hermitian quasiperiodic lattices [57, 58, 59, 60].

Here we study this question in a nearest-neighbor chain of weakly coupled dimers. All hoppings and the spatial onsite profile remain fixed while one overall potential amplitude is varied. This amplitude changes both the local detuning and the quasiperiodic modulation. The resulting redistribution of dimer weights cancels the first-order potential in the lower band while leaving its hopping finite. The upper band has no such cancellation, so the recovery is band selective. Wang et al. [61] identified orbital-weight cancellation numerically in a trimer ladder; here we establish its consequences for the full dimer chain, beyond the projection. We place the weak hopping between the residual modulation scale near cancellation and the larger modulation away from it.

Near cancellation, the exact transfer matrix remains uniformly close to a constant rotation throughout a fixed complex strip of phases. This bound limits the complexified Lyapunov exponent on the strip. Since its slopes are quantized by Avila’s global theory [62], the exponent must be flat near the real axis, and a flat positive exponent is excluded on the spectrum. For sufficiently small modulation parameter η\eta, every spectral energy in a central lower-band window is therefore subcritical, with a finite zero-exponent plateau, throughout a recovery interval of couplings of width of order α/η\alpha/\eta. Continuity of the rotation number shows that this window holds one third of the lower-band states. The exponents are uniformly positive at comparison couplings before and after recovery.

The branches of the complexified Lyapunov exponent also provide a diagnostic at finite parameters. A signed branch intercept summarizes the loss and recovery of finite zero-exponent plateaus across the spectrum and remains informative when the physical exponent vanishes. Spectrum-wide participation maps and plateau scans show that the recovery covers the entire lower band and is absent in the upper band. The numerically determined window edges coincide with the crossings predicted by the first-order projection, and the window width follows the predicted a/ηa/\eta scaling as η\eta decreases.

Section II introduces the model and its dimer projection. Section III develops the Lyapunov branch diagnostic, and Sec. IV establishes the controlled reentrant regime. Section V presents the numerical results, and Sec. VI discusses their implications and scope. Technical derivations are collected in the appendices.

Model and dimer mechanism

We consider an infinite chain with Hamiltonian

H=∑m[bcA,m†cB,m+acB,m†cA,m+1+H.c.]+∑m[εA,m(λ)nA,m+εB,m(λ)nB,m].(1)H=\sum_m\left[b c^\dagger_{A,m}c_{B,m}+a c^\dagger_{B,m}c_{A,m+1}+\mathrm{H.c.}\right]+\sum_m\left[\varepsilon_{A,m}(\lambda)n_{A,m}+\varepsilon_{B,m}(\lambda)n_{B,m}\right]. \tag*{(1)}

with a,b>0a,b>0 and a≪ba\ll b. The strong bond joins AmA_m and BmB_m while the weak bond joins neighboring cells. The onsite terms are

εA,m(λ)=λ[1+3ηcos⁡θm],εB,m(λ)=−λ[1+ηcos⁡θm],θm=2πβm+ϕ.(2)\varepsilon_{A,m}(\lambda)=\lambda[1+3\eta\cos\theta_m], \qquad \varepsilon_{B,m}(\lambda)=-\lambda[1+\eta\cos\theta_m], \qquad \theta_m=2\pi\beta m+\phi. \tag*{(2)}

The frequency β\beta is irrational to ensure quasiperiodicity. The parameters aa, bb, η>0\eta>0, and β\beta remain fixed while λ≥0\lambda\ge0 is varied. Thus H=Hhop+λFH=H_{\mathrm{hop}}+\lambda F with one fixed onsite profile FF.

The quasiperiodic terms have opposite signs, unequal magnitudes, and a common phase within each cell. This choice is central to solve analytically. It differs from a cosine sampled at every original site, where adjacent sites have a phase difference.

Projected potential and hopping

We first solve the strong dimers exactly, keeping the detuning λ\lambda, and then treat the quasiperiodic modulation and weak bonds as perturbations. For a,b>0a,b>0, the isolated dimer has

h0(λ)=(λbb−λ),E±(λ)=±b2+λ2.(3)h_0(\lambda)= \begin{pmatrix} \lambda& b\\ b & -\lambda \end{pmatrix}, \qquad E_\pm(\lambda)=\pm\sqrt{b^2+\lambda^2}. \tag*{(3)}

Its level separation is at least 2b2b. The lower state has site probabilities

wA=12(1−λb2+λ2),wB=12(1+λb2+λ2).(4)w_A=\frac{1}{2}\left(1-\frac{\lambda}{\sqrt{b^2+\lambda^2}}\right), \qquad w_B=\frac{1}{2}\left(1+\frac{\lambda}{\sqrt{b^2+\lambda^2}}\right). \tag*{(4)}

and can be written as

∣m,−⟩=wA ∣A,m⟩−wB ∣B,m⟩.(5)|m,-\rangle=\sqrt{w_A}\,|A,m\rangle-\sqrt{w_B}\,|B,m\rangle. \tag*{(5)}

As λ\lambda grows, its weight shifts toward the lower-energy site BB.

All lower dimer states have the same unperturbed energy. We therefore allow the perturbations to mix these states and calculate their diagonal and nearest-neighbor matrix elements. Coupling through the upper dimer states contributes at higher order.

The quasiperiodic onsite perturbation is

δhm=ηλcos⁡θm(300−1).(6)\delta h_m=\eta\lambda\cos\theta_m \begin{pmatrix} 3 & 0\\ 0 & -1 \end{pmatrix}. \tag*{(6)}

Its first-order contribution to the local energy is

V−,m=⟨m,−∣δhm∣m,−⟩=ηλ(3wA−wB)cos⁡θm=ηλ(1−2λb2+λ2)cos⁡θm.(7)\begin{aligned} V_{-,m}&=\langle m,-|\delta h_m|m,-\rangle\\ &=\eta\lambda(3w_A-w_B)\cos\theta_m\\ &=\eta\lambda\left(1-\frac{2\lambda}{\sqrt{b^2+\lambda^2}}\right)\cos\theta_m. \tag*{(7)} \end{aligned}

Writing V−,m=v−(λ)cos⁡θmV_{-,m}=v_-(\lambda)\cos\theta_m, the modulation amplitude is thus the probability-weighted sum of the two opposite-sign onsite potentials. The contributions cancel at

λ∗=b3,wA(λ∗)=14,wB(λ∗)=34.(8)\lambda_*=\frac{b}{\sqrt{3}}, \qquad w_A(\lambda_*)=\frac{1}{4}, \qquad w_B(\lambda_*)=\frac{3}{4}. \tag*{(8)}

The smaller probability on AA exactly compensates its larger modulation coefficient.

The weak bond between adjacent dimers,

Ha,m=a∣B,m⟩⟨A,m+1∣+H.c.,(9)H_{a,m}=a|B,m\rangle\langle A,m+1|+\mathrm{H.c.}, \tag*{(9)}

gives the first-order hopping matrix element

teff(λ)=⟨m,−∣Ha,m∣m+1,−⟩=a(−wB)wA=−awAwB.(10)t_{\mathrm{eff}}(\lambda)=\langle m,-|H_{a,m}|m+1,-\rangle =a(-\sqrt{w_B})\sqrt{w_A} =-a\sqrt{w_Aw_B}. \tag*{(10)}

Since

wAwB=14(1−λ2b2+λ2)=b24(b2+λ2),(11)w_Aw_B=\frac{1}{4}\left(1-\frac{\lambda^2}{b^2+\lambda^2}\right) =\frac{b^2}{4(b^2+\lambda^2)}, \tag*{(11)}

its magnitude is

∣teff(λ)∣=ab2b2+λ2,and∣teff(λ∗)∣=34a.(12)|t_{\mathrm{eff}}(\lambda)|=\frac{ab}{2\sqrt{b^2+\lambda^2}}, \qquad \text{and}\qquad |t_{\mathrm{eff}}(\lambda_*)|=\frac{\sqrt{3}}{4}a. \tag*{(12)}

The effective hopping therefore remains finite when the first-order modulation cancels.

Writing the state in this description as ∣Ψ⟩=∑mxm∣m,−⟩|\Psi\rangle= \sum_m x_m|m,-\rangle, first-order perturbation gives

(E+b2+λ2)xm=−∣teff∣(xm+1+xm−1)+v−(λ)cos⁡θm xm.(13)\left(E+\sqrt{b^2+\lambda^2}\right)x_m=-|t_{\mathrm{eff}}|(x_{m+1}+x_{m-1})+v_{-}(\lambda)\cos\theta_m\,x_m. \tag*{(13)}

The change of amplitudes xm=(−1)mumx_m=(-1)^m u_m gives xm±1=−(−1)mum±1x_{m\pm1}=-(-1)^m u_{m\pm1} and reverses the hopping sign. The effective equation becomes

(E+b2+λ2)um=∣teff∣(um+1+um−1)+v−(λ)cos⁡θm um.(14)\left(E+\sqrt{b^2+\lambda^2}\right)u_m=|t_{\mathrm{eff}}|(u_{m+1}+u_{m-1})+v_{-}(\lambda)\cos\theta_m\,u_m. \tag*{(14)}

To this order, the lower band is therefore described by an AA chain with modulation-hopping ratio

ρ(λ)=∣v−(λ)∣2∣teff(λ)∣=ηλ∣b2+λ2−2λ∣ab.(15)\rho(\lambda)=\frac{|v_{-}(\lambda)|}{2|t_{\mathrm{eff}}(\lambda)|}=\frac{\eta\lambda\left|\sqrt{b^2+\lambda^2}-2\lambda\right|}{ab}. \tag*{(15)}

On its spectrum, this effective AA model has zero Lyapunov exponent for ρ≤1\rho\le1 and a positive exponent for ρ>1\rho>1 [3]. For a≪ηba\ll\eta b, the ratio can exceed one before λ∗\lambda_\ast, vanishes at λ∗\lambda_\ast, and grows again (Figure 1).

Figure 1. Projected modulation-to-hopping ratio for a/b=10−3a/b=10^{-3} and η=10−2\eta=10^{-2}. Shading marks ρ<1\rho<1; the arrow indicates first-order cancellation.

For the upper dimer state, the site probabilities are interchanged, giving the first-order modulation amplitude

v+(λ)=ηλ(1+2λb2+λ2).(16)v_{+}(\lambda)=\eta\lambda\left(1+\frac{2\lambda}{\sqrt{b^2+\lambda^2}}\right). \tag*{(16)}

It has no cancellation at positive λ\lambda, so the first-order recovery is band selective.

The projection predicts the physical route to recovery, but its AA criterion is exact only for Eq. (14). The original chain retains interband processes and higher-order modulation. Their control is developed in Sec. IV.

Lyapunov branch structure

Transfer matrices and imaginary phase

Write a stationary wavefunction as (xm,ym)(x_m,y_m) on (Am,Bm)(A_m,B_m). The site equations are

(E−εA,m)xm=bym+aym−1,(E−εB,m)ym=bxm+axm+1.(17)\begin{aligned} (E-\varepsilon_{A,m})x_m&=by_m+ay_{m-1},\\ (E-\varepsilon_{B,m})y_m&=bx_m+ax_{m+1}. \tag*{(17)} \end{aligned}

With P=E−εA(θ)P=E-\varepsilon_A(\theta) and Q=E−εB(θ)Q=E-\varepsilon_B(\theta), the exact cell transfer matrix is

\begin{pmatrix}x_{m+1}\\y_m\end{pmatrix} =M_E(\theta_m)\begin{pmatrix}x_m\\y_{m-1}\end{pmatrix}, \qquad M_E(\theta)=\frac{1}{ab}\begin{pmatrix}PQ-b^2&-aQ\\aP&-a^2\end{pmatrix}. \tag*{(18)}\end{pmatrix}

Its determinant is one. The chain samples this one matrix-valued function along θ↦θ+2πβ\theta\mapsto\theta+2\pi\beta; this transfer evolution is the cocycle. Complexify the phase as ϕ→ϕ+ih\phi\to\phi+ih. Let

MN(E,ϕ+ih)=ME(ϕ+(N−1)2πβ+ih)⋯ME(ϕ+ih)(19)\mathcal{M}_N(E,\phi+ih)=M_E(\phi+(N-1)2\pi\beta+ih)\cdots M_E(\phi+ih) \tag*{(19)}

and define

L(E,h)=lim⁡N→∞1N12π∫02πln⁡∥MN(E,ϕ+ih)∥ dϕ.(20)L(E,h)=\lim_{N\to\infty}\frac{1}{N}\frac{1}{2\pi}\int_0^{2\pi}\ln\left\|\mathcal{M}_N(E,\phi+ih)\right\|\,d\phi. \tag*{(20)}

This exponent is per cell. The physical exponent per original site reads as

γ(E,λ)=12L(E,0).(21)\gamma(E,\lambda)=\frac{1}{2}L(E,0). \tag*{(21)}

The imaginary phase is an analytical probe of the real chain. The identity cos⁡(θ+ih)=(e−h+iθ+eh−iθ)/2\cos(\theta+ih)=(e^{-h+i\theta}+e^{h-i\theta})/2 shows how it amplifies one component. L(E,h)L(E,h) is even and convex, with integer slopes for this analytic determinant-one cocycle [62].

For our model and h≥0h\ge0, the complexified Lyapunov exponent is piecewise linear with allowed slopes zero, one, and two. The upper bound follows from the Fourier degree of the cell transfer matrix. We characterize the slope-one and slope-two branches by their intercepts,

c1(E,λ)=min⁡h≥0[L(E,h)−h],(22)c_1(E,\lambda)=\min_{h\ge0}[L(E,h)-h], \tag*{(22)}
c2(λ)=ln⁡3η2λ24ab=2ln⁡λλH,(23)c_2(\lambda)=\ln\frac{3\eta^2\lambda^2}{4ab}=2\ln\frac{\lambda}{\lambda_H}, \tag*{(23)}

where

λH=2ab3η.(24)\lambda_H=\frac{2\sqrt{ab}}{\sqrt{3}\eta}. \tag*{(24)}

For λ>λH\lambda>\lambda_H, the inequality L(E,0)≥c2(λ)L(E,0)\ge c_2(\lambda) forces every spectral exponent to be positive. Thus λH\lambda_H is an upper bound on the survival of zero exponents, rather than an exact transition point. The slope restriction, c1c_1 construction, the intercept formulas, and the resulting bound are detailed in Appendix A.

Diagnostic across the transition

For an energy in a band, a flat segment can occur only at height zero. A positive flat segment would imply uniform hyperbolicity and place the energy outside the spectrum.

We have the complexified Lyapunov exponent as

L(E,h)=max⁡{0,c1(E,λ)+h,c2(λ)+2h}(25)L(E,h)=\max\{0,c_1(E,\lambda)+h,c_2(\lambda)+2h\} \tag*{(25)}

For a zero-exponent energy, let hch_c be the largest h≥0h\geq0 for which L(E,h)=0L(E,h)=0; set hc=0h_c=0 when the physical exponent is positive. Eq. (25) gives

c1(E,λ)={2γ(E,λ),γ(E,λ)>0,−hc(E,λ),γ(E,λ)=0.(26)c_1(E,\lambda)= \begin{cases} 2\gamma(E,\lambda), & \gamma(E,\lambda)>0,\\ -h_c(E,\lambda), & \gamma(E,\lambda)=0. \end{cases} \tag*{(26)}

Thus c1>0c_1>0 identifies a positive exponent, c1<0c_1<0 a subcritical energy with a finite zero plateau, and c1=0c_1=0 a critical energy. The plateau width measures stability under imaginary phase. To follow the chain as a whole, we define

C1(λ)=min⁡E∈σ(Hλ)c1(E,λ),D(λ)=min⁡E∈σ(Hλ)γ(E,λ),C_1(\lambda)=\min_{E\in\sigma(H_\lambda)}c_1(E,\lambda),\qquad D(\lambda)=\min_{E\in\sigma(H_\lambda)}\gamma(E,\lambda),
W(λ)=max⁡E∈σ(Hλ)hc(E,λ).(27)W(\lambda)=\max_{E\in\sigma(H_\lambda)}h_c(E,\lambda). \tag*{(27)}

where σ(Hλ)\sigma(H_\lambda) denotes the spectrum. They obey

2D=max⁡{0,C1},W=max⁡{0,−C1}.(28)2D=\max\{0,C_1\},\qquad W=\max\{0,-C_1\}. \tag*{(28)}

A positive C1C_1 means that every spectral energy has a positive exponent. A negative value measures the widest subcritical plateau. At C1=0C_1=0, critical energies can remain even though W=0W=0. Recovery of finite plateaus between positive-exponent regimes is therefore the sequence C1>0C_1>0, C1<0C_1<0, C1>0C_1>0.

Below λH\lambda_H, the slope-two line limits every zero plateau to

hc(E,λ)≤hmax⁡(λ)=−c2(λ)2=ln⁡λHλ.(29)h_c(E,\lambda)\leq h_{\max}(\lambda)=-\frac{c_2(\lambda)}{2}=\ln\frac{\lambda_H}{\lambda}. \tag*{(29)}

For E∈σ(Hλ)E\in\sigma(H_\lambda) and 0<λ<λH0<\lambda<\lambda_H, Eq. (25) then gives a useful identity,

c1(E,λ)=L(E,hmax⁡)−hmax⁡.(30)c_1(E,\lambda)=L(E,h_{\max})-h_{\max}. \tag*{(30)}

reentrant regime

Controlled subcritical recovery

The first-order perturbation predicts cancellation of the lower-cell modulation at λ∗=b/3\lambda_* = b/\sqrt{3}. We now estimate the modulation that remains beyond first order and compare it with the hopping between cells.

The isolated cell can be diagonalized exactly, giving the lower energy

e−(θ,λ)=ηλcos⁡θ−b2+λ2(1+2ηcos⁡θ)2.(31)e_-(\theta,\lambda)=\eta\lambda\cos\theta-\sqrt{b^2+\lambda^2(1+2\eta\cos\theta)^2}. \tag*{(31)}

At λ∗\lambda_*, the unperturbed lower-cell energy used in the projection is

e−(0)(λ∗)=−b2+λ∗2=−2b3.e_-^{(0)}(\lambda_*)=-\sqrt{b^2+\lambda_*^2}=-\frac{2b}{\sqrt{3}}.

Expanding the exact energy around λ=λ∗\lambda=\lambda_*

e−(θ,λ∗)=e−(0)(λ∗)−34η2bcos⁡2θ+O(η3b).(32)e_-(\theta,\lambda_*)=e_-^{(0)}(\lambda_*)-\frac{\sqrt{3}}{4}\eta^2b\cos^2\theta+O(\eta^3b). \tag*{(32)}

The first-order correction in η\eta vanishes, leaving a correction of order η2b\eta^2b. Using cos⁡2θ=[1+cos⁡(2θ)]/2\cos^2\theta=[1+\cos(2\theta)]/2, we separate this correction into a uniform energy shift and a second-harmonic modulation. The uniform shift changes every cell energy equally; the second harmonic produces the remaining energy differences between cells.

The exact bound

−23η2b≤e−(θ,λ∗)−e−(0)(λ∗)≤0(33)-\frac{2}{3}\eta^2b\leq e_-(\theta,\lambda_*)-e_-^{(0)}(\lambda_*)\leq0 \tag*{(33)}

holds for every θ\theta, including all higher-order corrections. Thus the lower-cell energies lie within an interval of width at most 2η2b/32\eta^2b/3.

Equation (33) limits the spread of the lower-cell energies to order η2b\eta^2b. The hopping between these states remains of order aa, with leading value teff=3a/4t_{\mathrm{eff}}=\sqrt{3}a/4 at λ∗\lambda_*. If η2b≪a\eta^2b\ll a, the energy differences between cells are small compared with their coupling, allowing the lower-cell states to hybridize strongly.

Away from cancellation, at couplings of order bb, the first-order modulation is instead of order ηb\eta b. Choosing a≪ηba\ll\eta b makes this modulation strong compared with the hopping. We therefore place the hopping between the two modulation scales

η2b≪a≪ηb\eta^2b\ll a\ll\eta b

This hierarchy provides the energy-scale separation needed for recovery near λ∗\lambda_* between regimes of stronger modulation.

We enforce both inequalities by choosing

a=bηp,1<p<2(34)a=b\eta^p,\qquad1<p<2 \tag*{(34)}

with fixed b>0b>0, pp, and β\beta, and taking η\eta sufficiently small. Indeed, η2b/a=η2−p→0\eta^2b/a=\eta^{2-p}\to0 and a/(ηb)=ηp−1→0a/(\eta b)=\eta^{p-1}\to0.

To control propagation in the original chain, we introduce r=a/br=a/b and write

E=−2b3+ax.E=-\frac{2b}{\sqrt{3}}+ax.

Thus xx measures energy relative to the unperturbed lower-cell level in units of the weaker hopping. At λ∗\lambda_*, as η→0\eta\to0 with r=ηpr=\eta^p and fixed xx, the exact transfer matrix converges to a phase independent matrix. We denote this limiting matrix by

M0(x)=(−4x/31/3−30).(35)M_0(x)= \begin{pmatrix} -4x/\sqrt{3} & 1/\sqrt{3}\\ -\sqrt{3} & 0 \end{pmatrix}. \tag*{(35)}

We restrict the scaled energy with ∣x∣≤3/4|x| \le\sqrt{3}/4. Throughout this window,

∣tr⁡M0(x)∣=4∣x∣3≤1,|\operatorname{tr} M_0(x)| = \frac{4|x|}{\sqrt{3}} \le1,

so the limiting matrices remain uniformly inside the elliptic regime ∣tr⁡M0∣<2|\operatorname{tr} M_0| < 2. Each can therefore be transformed into a rotation by a change of basis.

Near cancellation, the exact transfer matrix is a small quasiperiodic perturbation of this elliptic matrix. The uniform matrix estimate also constrains the Lyapunov branches. After a bounded constant change of basis, the exact transfer matrix is close to a rotation throughout a fixed complex strip. Its Lyapunov exponent is therefore small throughout that strip. Since every nonzero slope is at least one, a rising branch cannot begin close to h=0h=0 without exceeding this bound. For sufficiently small η\eta, the exponent must have a plateau around the real axis.

At a spectral energy, this plateau must have zero height. A positive exponent that is locally independent of hh would imply uniform hyperbolicity, which is excluded on the spectrum [63]. Thus every spectral energy E=−2b/3+axE=-2b/\sqrt{3}+ax with ∣x∣≤3/4|x| \le\sqrt{3}/4 satisfies

L(E,h)=0,∣h∣≤h0,(36)L(E,h)=0,\qquad|h|\le h_0, \tag*{(36)}

where h0>0h_0>0 can be chosen uniformly over this energy window. All spectral energies in the window are therefore subcritical.

The window is also occupied. Near cancellation, the rotation number of the exact transfer matrix stays close to that of M0(x)M_0(x), which sweeps an interval of length 1/61/6 across the window. Through the integrated density of states, the window therefore holds a fraction 1/3+O(δη)1/3+O(\delta_\eta) (with δη\delta_\eta defined in Eq. (58)) of the lower-band states, all of them subcritical (Appendix B § 0.9.2).

Recovery persists over a finite range of λ\lambda. Moving away from λ∗\lambda_* restores modulation of order η∣λ−λ∗∣\eta|\lambda-\lambda_*|, while the hopping remains of order aa. Their competition sets the natural recovery scale a/ηa/\eta.

Appendix B § 0.9.3 shows that there is a constant c∗>0c_* > 0 such that, for sufficiently small η\eta, at every coupling satisfying

∣λ−λ∗∣<c∗aη.(37)|\lambda-\lambda_*|<c_*\frac{a}{\eta}. \tag*{(37)}

the controlled window holds a fraction close to one third of the lower-band states, and every spectral energy in it is subcritical. This bound guarantees recovery within the stated interval without locating the transition points. The recovered energies follow the lower dimer level −b2+λ2-\sqrt{b^2+\lambda^2} as λ\lambda varies.

Within the first-order AA description, the condition ρ<1\rho<1 gives the leading width estimate

∣λ−λ∗∣≲2a3η.(38)|\lambda-\lambda_*|\lesssim\frac{2a}{\sqrt{3}\eta}. \tag*{(38)}

Its derivation is also given in Appendix B § 0.9.3.

Positive exponents before and after recovery

At the earlier comparison point λe=b/4\lambda_e=b/4, the original transfer matrix gives a uniform bound over the spectrum,

γ(E,λe)≥12[ln⁡ηba−C0]>0,(39)\gamma(E,\lambda_e)\ge\frac{1}{2}\left[\ln\frac{\eta b}{a}-C_0\right]>0, \tag*{(39)}

for sufficiently small η\eta under Eq. (34). The constant C0C_0 is independent of η\eta. Appendix B § 0.9.4 derives this estimate from transfer growth at a fixed imaginary phase and uses the maximum slope two to obtain the real-axis bound.

For the later comparison point, choose λ1=2λH=4ab/(3η)\lambda_1=2\lambda_H=4\sqrt{ab}/(\sqrt{3}\eta). The bound in Eq. (56) gives

γ(E,λ1)≥ln⁡λ1λH=ln⁡2(40)\gamma(E,\lambda_1)\ge\ln\frac{\lambda_1}{\lambda_H}=\ln2 \tag*{(40)}

at every spectral energy. Moreover,

λ1λ∗=4ηp/2−1⟶∞.(41)\frac{\lambda_1}{\lambda_*}=4\eta^{p/2-1}\longrightarrow\infty. \tag*{(41)}

The guaranteed recovery width, measured in units of bb, is 2c∗aηb=2c∗ηp−1→02c_*\frac{a}{\eta b}=2c_*\eta^{p-1}\to0. Together with λe<λ∗\lambda_e<\lambda_* and Eq. (41), this places the comparison points on opposite sides of the recovered interval for sufficiently small η\eta.

The minimum physical exponent therefore satisfies

D(λe)>0,D(λ1)≥ln⁡2,D(\lambda_e)>0,\qquad D(\lambda_1)\ge\ln2,
D(λ)=0for∣λ−λ∗∣<c∗a/η.(42)D(\lambda)=0\quad\text{for}\quad|\lambda-\lambda_*|<c_*a/\eta. \tag*{(42)}

The recovered energies have finite zero-exponent plateaus, giving the branch signs

C1(λe)>0,C1(λ∗)<0,C1(λ1)>0.(43)C_1(\lambda_e)>0,\qquad C_1(\lambda_*)<0,\qquad C_1(\lambda_1)>0. \tag*{(43)}

At λ=0\lambda=0, the chain is periodic and its spectral bands have zero exponent. The fixed-profile sweep therefore exhibits an initial closing, a recovery, and a later closing in the exponent-based sense.

The analytical bounds establish reentrance for sufficiently small η\eta under Eq. (34). The numerical parameters used below follow the same scaling with p=3/2p=3/2, but their finite value of η\eta is not certified by our conservative estimates.

Numerical results

Unless stated otherwise, we use a/b=10−3a/b=10^{-3}, η=10−2\eta=10^{-2}, and the irrational cell frequency β=5−2\beta=\sqrt{5}-2. Eigenstates are computed with open boundary conditions for Nc=213,…,217N_c=2^{13},\ldots,2^{17} two-site cells. Lyapunov estimates use transfer products of up to 2×1062\times10^6 cells.

Recovery and its spectral extent

Figure 2(a) shows the sampled global intercept C1(λ)C_1(\lambda). Its sign follows the sequence negative, positive, negative, and positive as λ\lambda increases. The positive intervals have positive exponents at all sampled energies, while the recovered negative interval contains the cancellation point λ∗\lambda_*.

Figure 2. (a) Signed intercept C1(λ)C_1(\lambda) and (b) cell IPR at the marked couplings for a/b=10−3a/b = 10^{-3} and η=10−2\eta= 10^{-2}. The IPR uses lower-band label q=0.125q = 0.125. The vertical dashed line marks λ∗\lambda_*. (c) Scaled recovery half-width along a/b=η3/2a/b = \eta^{3/2}. Error bars show its variation across 24 lower-band quantiles. The dashed line marks the first-order prediction 2/32/\sqrt{3}.

To follow spectral populations, we use the quantile q≃j/(2Nc)q \simeq j/(2N_c) of the ordered finite-chain eigenvalues, re-computing its energy at each coupling and size. For a normalized eigenstate, the cell probability and cell inverse participation ratio are

pm=∣xm∣2+∣ym∣2,I=∑mpm2.(44)p_m = |x_m|^2 + |y_m|^2,\qquad\mathcal{I} = \sum_m p_m^2. \tag*{(44)}

Summing within each cell removes the effect of shifting weight between its two sites.

Figure 2(b) shows the IPR at the lower-band quantile q=0.125q = 0.125, geometrically averaged over four spectral phases. At λ∗\lambda_*, I\mathcal{I} decreases approximately as Nc−1N_c^{-1} across the five sizes, consistent with an extended state. At λ/b=0.275\lambda/b = 0.275 and 0.9000.900, it remains nearly independent of size, consistent with localization. The same spectral population therefore recovers spatial extent between two localized regimes.

The participation map in Figure 3(a) shows that this recovery extends across the sampled lower band. The upper band remains strongly concentrated over the same coupling range. This band selectivity agrees with the cancellation of the lower-state modulation and the absence of a corresponding cancellation in the upper state. The numerical boundaries coincide with the first-order ρ=1\rho= 1 crossings; their median over quantiles differs from

Figure 3. Recovery in the lower band for a/b=10−3a/b = 10^{-3} and η=10−2\eta= 10^{-2}. (a) Cell IPR, shown as log⁡10I\log_{10} \mathcal{I}, versus coupling and spectral quantile qq. White dashed lines mark the first-order ρ=1\rho= 1 crossings; the red line marks λ∗\lambda_*. (b) Normalized zero-exponent plateau width hc/hmax⁡h_c/h_{\max} at λ∗\lambda_*, showing subcritical recovery in the lower band. The horizontal line at q=1/2q = 1/2 separates the two bands.

them by less than 4×10−5b4 \times10^{-5}b.

Figure 4. Complexified Lyapunov exponent L(E,h)L(E,h) per cell at q=0.125q = 0.125 for the indicated couplings, with a/b=10−3a/b = 10^{-3} and η=10−2\eta= 10^{-2}. The inset enlarges the recovery corner. Dashed lines show branch extrapolations, and numbers label their slopes.

Lyapunov plateaus and recovery width

Figure 4 compares the complexified Lyapunov exponent at q=0.125q = 0.125 for the same three couplings. The spectral energies are selected at ϕ=0\phi= 0, and the profiles are averaged over three transfer phases. Before and after recovery, L(E,0)>0L(E,0) > 0. At λ∗\lambda_*, the exponent is consistent with zero throughout a finite interval of imaginary phase. The inset resolves a short slope-one segment between the zero plateau and the slope-two branch. This finite plateau identifies the recovered state as subcritical. Figure 3(b) extends this test across the spectrum at λ∗\lambda_*. The plateau width hc/hmax⁡h_c/h_{\max} is close to its allowed maximum throughout the sampled lower band, while the upper band has no resolved zero plateau. The recovered region therefore combines spatial spreading with finite subcritical plateaus.

Finally, Figure 2(c) tests the predicted recovery scale by varying η\eta along a/b=η3/2a/b=\eta^{3/2}. For each η\eta, we use Nc=217N_c=2^{17}, 2×1062\times10^6 transfer cells, and three transfer phases. Recovery edges are located separately at 24 lower-band quantiles q∈[0.02,0.48]q\in[0.02,0.48], using sign changes of c1c_1 from Eq. (30). For each quantile, we locate the two couplings at which recovery begins and ends, and take ww as half the distance between them. The bars show how ww varies across the sampled quantiles. As η\eta decreases, w/(a/η)w/(a/\eta) approaches 2/32/\sqrt{3} and its variation across the band narrows. Extrapolating the three smallest values of η\eta to η→0\eta\to0 gives 1.154, in agreement with 2/3≃1.1552/\sqrt{3}\simeq1.155. This supports the predicted a/ηa/\eta recovery scale and its first-order prefactor.

For sufficiently small nonzero η\eta under Eq. (34), the analytical result establishes a finite recovery interval in which every spectral energy of a central window, holding one third of the lower-band states, is subcritical. This interval is bracketed by uniform positive-exponent regimes. The interval is sufficient and does not identify exact critical couplings. Positivity of the exponent alone is also not a universal point localization theorem. The result provides a controlled example of reentrance driven by changes in the internal dimer weights during a fixed-profile potential sweep.

Discussion

Increasing λ\lambda strengthens the modulation and shifts the weight between the two sites of each dimer. This redistribution can reduce the potential seen by the lower-band states even as the bare onsite amplitudes grow. At first-order cancellation, the weak hopping remains finite and the residual modulation is of order η2b\eta^2 b. The hierarchy η2b≪a≪ηb\eta^2 b\ll a\ll\eta b places the hopping above this residual but below the modulation away from cancellation, providing the scale separation for recovery.

The projected cancellation also persists under a change of modulation ratio. Replacing the coefficients 3 and 1 in Eq. (2) by fixed sA>sB>0s_A>s_B>0 gives

v−(s)(λ)=ηλ2[(sA−sB)−(sA+sB)λb2+λ2],(45)v_-^{(s)}(\lambda)=\frac{\eta\lambda}{2}\left[(s_A-s_B)-(s_A+s_B)\frac{\lambda}{\sqrt{b^2+\lambda^2}}\right], \tag*{(45)}

with projected cancellation point

λ∗(s)=b(sA−sB)2sAsB.(46)\lambda_*^{(s)}=\frac{b(s_A-s_B)}{2\sqrt{s_As_B}}. \tag*{(46)}

This predicts how the cancellation shifts with the modulation ratio.

Cancellation through redistribution of orbital weights also appears in related molecular-band models [61]. Here, control of the residual terms and the exact transfer matrix connects the projected mechanism to the full dimer chain. At the finite parameters studied, the reentrant sign sequence of C1C_1, the spectrum-wide participation map, the maximal zero-exponent plateaus across the lower band, and the a/ηa/\eta scaling of the recovery window support the same picture. For C1<0C_1<0, the quantity −C1-C_1 gives the widest plateau width, retaining information when the minimum physical exponent vanishes.

Acknowledgments

This work was supported by the Scientific and Technological Council of Türkiye (TÜBİTAK) under Grant No. 125F435 and the Turkish Academy of Sciences (TÜBA) under Grant No. AD-2026. The numerical calculations reported in this paper were performed at TÜBİTAK ULAKBİM, High Performance and Grid Computing Center (TRUBA resources).

Slope bound and large-phase asymptote

We take a,b,η>0a,b,\eta>0 and λ>0\lambda>0. In Eq. (18), the onsite factors at complex phase z=θ+ihz=\theta+ih are

P=E−λ−3ηλcos⁡z,P=E-\lambda-3\eta\lambda\cos z,
Q=E+λ+ηλcos⁡z.(47)Q=E+\lambda+\eta\lambda\cos z. \tag*{(47)}

Their product is

PQ=E2−λ2−2ηλ(E+2λ)cos⁡z−3η2λ2cos⁡2z.(48)PQ=E^2-\lambda^2-2\eta\lambda(E+2\lambda)\cos z -3\eta^2\lambda^2\cos^2z. \tag*{(48)}

Thus every entry of MEM_E has Fourier degree at most two. Since cos⁡(θ+ih)=O(eh)\cos(\theta+ih)=O(e^h) uniformly in θ\theta,

L(E,h)≤ln⁡sup⁡θ∥ME(θ+ih)∥≤2h+C,(49)L(E,h)\le\ln\sup_\theta\lVert M_E(\theta+ih)\rVert\le2h+C, \tag*{(49)}

where CC is independent of hh. Convexity makes the slopes of LL nondecreasing. A slope greater than two would therefore contradict this bound. Evenness gives nonnegative slopes for h≥0h\ge0, while acceleration quantization restricts them to integers. Thus the allowed slopes are zero, one, and two.

To determine the slope-two intercept, we use

cos⁡(θ+ih)=12eh−iθ[1+O(e−2h)].(50)\cos(\theta+ih)=\frac{1}{2}e^{h-i\theta}\left[1+O(e^{-2h})\right]. \tag*{(50)}

The quadratic term in PQPQ then gives

ME(θ+ih)=−3η2λ24abe2h−2iθ(1000)+O(eh),(51)M_E(\theta+ih)=-\frac{3\eta^2\lambda^2}{4ab}e^{2h-2i\theta} \begin{pmatrix}1&0\\0&0\end{pmatrix}+O(e^h), \tag*{(51)}

uniformly in θ\theta. After removing the scalar prefactor, the matrices approach diag⁡(1,0)\operatorname{diag}(1,0) with error O(e−h)O(e^{-h}). For sufficiently large hh, the remaining Lyapunov exponent is therefore O(e−h)O(e^{-h}), giving

L(E,h)=2h+ln⁡3η2λ24ab+O(e−h).(52)L(E,h)=2h+\ln\frac{3\eta^2\lambda^2}{4ab}+O(e^{-h}). \tag*{(52)}

This establishes Eq. (23). The intercept depends on neither EE nor the irrational frequency β\beta.

The slope-one intercept is defined by requiring the line c+hc+h to lie below L(E,h)L(E,h) for every h≥0h\ge0. This requires c≤L(E,h)−hc\le L(E,h)-h at every height, so the largest admissible intercept is

c1(E,λ)=min⁡h≥0[L(E,h)−h].(53)c_1(E,\lambda)=\min_{h\ge0}[L(E,h)-h]. \tag*{(53)}

The large-hh asymptote gives L(E,h)−h=h+c2(λ)+o(1)→∞L(E,h)-h=h+c_2(\lambda)+o(1)\to\infty, so the minimum is attained at a finite height. Consequently,

L(E,h)≥c1(E,λ)+h,(54)L(E,h)\ge c_1(E,\lambda)+h, \tag*{(54)}

with equality at a minimizing height. If a slope-one segment is present, c1c_1 is its intercept. The definition remains valid when the supporting line touches only a corner or the endpoint h=0h=0, without an exposed slope-one segment. The slope bound makes L(E,h)−2hL(E,h)-2h nonincreasing, with limit c2(λ)c_2(\lambda), so

L(E,h)≥c2(λ)+2h.(55)L(E,h)\ge c_2(\lambda)+2h. \tag*{(55)}

At h=0h=0, using L(E,0)≥0L(E,0)\ge0 and the per-site normalization, we obtain

γ(E,λ)=12L(E,0)≥max⁡{0,12c2(λ)}=max⁡{0,ln⁡λλH},λH=2ab3η.(56)\begin{aligned} \gamma(E,\lambda)&=\frac{1}{2}L(E,0)\\ &\ge\max\left\{0,\frac{1}{2}c_2(\lambda)\right\}\\ &=\max\left\{0,\ln\frac{\lambda}{\lambda_H}\right\},\qquad \lambda_H=\frac{2\sqrt{ab}}{\sqrt{3}\eta}. \tag*{(56)} \end{aligned}

Analytical bounds for the original chain

Recovery at the cancellation point

We first show that the cancellation suppresses the modulation in the exact transfer matrix. Set r=a/br=a/b, E=−2b/3+axE=-2b/\sqrt{3}+ax, and c=cos⁡θc=\cos\theta. At λ=λ∗=b/3\lambda=\lambda_\ast=b/\sqrt{3}, Eq. (18) gives

M11=−4x3+rx2−2ηx3c−η2rc2,M12=13−rx−η3c,M21=−3+rx−3ηc,M22=−r.(57)\begin{aligned} M_{11}&=-\frac{4x}{\sqrt{3}}+rx^2-\frac{2\eta x}{\sqrt{3}}c-\frac{\eta^2}{r}c^2,\\ M_{12}&=\frac{1}{\sqrt{3}}-rx-\frac{\eta}{\sqrt{3}}c,\\ M_{21}&=-\sqrt{3}+rx-\sqrt{3}\eta c,\qquad M_{22}=-r. \tag*{(57)} \end{aligned}

We restrict the scaled energy to I=[−3/4,3/4]I=[-\sqrt{3}/4,\sqrt{3}/4].

To control the analytic dependence on phase, we write θ=ϑ+ih\theta=\vartheta+ih, with real ϑ\vartheta and hh, and consider the fixed strip ∣h∣<1|h|<1. Throughout this domain,

∣x∣≤12,∣cos⁡θ∣≤cosh⁡1<2.|x|\le\frac{1}{2},\qquad|\cos\theta|\le\cosh1<2.

Let Mη,xM_{\eta,x} denote the exact matrix in Eq. (57) and M0(x)M_0(x) its limit in Eq. (35). Applying these bounds to the entries of Mη,x−M0(x)M_{\eta,x}-M_0(x), and bounding the operator norm by the sum of their absolute values, gives

∥Mη,x−M0(x)∥strip≤6δη,δη=r+η+η2r.(58)\left\|M_{\eta,x}-M_0(x)\right\|_{\mathrm{strip}}\le6\delta_\eta,\qquad \delta_\eta=r+\eta+\frac{\eta^2}{r}. \tag*{(58)}

The strip norm denotes the supremum of the operator norm over all real ϑ\vartheta and ∣h∣<1|h|<1. The estimate holds uniformly for every x∈Ix\in I.

The relevant smallness is set by δη\delta_\eta: choosing r=ηpr=\eta^p with 1<p<21<p<2 gives

δη=ηp+η+η2−p⟶0as η→0.\delta_\eta=\eta^p+\eta+\eta^{2-p}\longrightarrow0\qquad\text{as }\eta\to0.

The exact transfer matrix therefore approaches the limiting matrix throughout the chosen energy window.

The limiting propagation is oscillatory throughout II: ∣tr⁡M0∣≤1|\operatorname{tr}M_0|\le1, is inside the elliptic regime ∣tr⁡M0∣<2|\operatorname{tr}M_0|<2. Each limiting matrix can be transformed into a rotation by a change of coordinates. Because the whole interval stays away from ∣tr⁡M0∣=2|\operatorname{tr}M_0|=2, these transformations remain uniformly bounded. In the same coordinates, the exact matrix is a perturbation of a constant rotation of order δη\delta_\eta.

Subcriticality in the controlled window

The uniform matrix estimate constrains the entire spectral window. Let TxT_x be the change of basis that transforms M0(x)M_0(x) into a rotation RxR_x. The transformed exact matrix satisfies

M~η,x=Tx−1Mη,xTx,∥M~η,x−Rx∥strip≤Krotδη.(59)\widetilde{M}_{\eta,x}=T_x^{-1}M_{\eta,x}T_x,\qquad \left\|\widetilde{M}_{\eta,x}-R_x\right\|_{\mathrm{strip}}\le K_{\mathrm{rot}}\delta_\eta. \tag*{(59)}

where KrotK_{\mathrm{rot}} is independent of x∈Ix\in I. Since a rotation has operator norm one, each transformed matrix has norm at most 1+Krotδη1+K_{\mathrm{rot}}\delta_\eta. Multiplying these bounds and using invariance of the Lyapunov exponent under the change of basis gives

0≤L(E,h)≤ln⁡(1+Krotδη),∣h∣<1.(60)0\le L(E,h)\le\ln(1+K_{\mathrm{rot}}\delta_\eta),\qquad|h|<1. \tag*{(60)}

The lower bound follows from the unit determinant.

Fix a height 0<H<10<H<1. Since δη→0\delta_\eta\to0, one can take η\eta sufficiently small that the upper bound in Eq. (60) is less than H/2H/2. In our phase normalization, every positive slope is at least one. If such a slope appeared at t≤H/2t\le H/2, convexity would give

L(E,H)≥L(E,t)+(H−t)≥H/2.(61)L(E,H)\ge L(E,t)+(H-t)\ge H/2. \tag*{(61)}

contradicting Eq. (60). The exponent must therefore be constant for 0≤h≤H/20 \le h \le H/2. Its evenness extends this plateau to ∣h∣≤H/2|h| \le H/2.

We now restrict to energies in the original spectrum. If the plateau had positive height, the Lyapunov exponent would be positive and would not change with hh near the real axis; such a cocycle is called regular. Avila’s theorem would then imply uniform hyperbolicity. For nonzero hoppings aa and bb, uniform hyperbolicity of the cell transfer matrix is excluded at spectral energies. Consequently,

L(E,h)=0,∣h∣≤H/2,E∈σ(Hλ∗),L(E,h)=0,\qquad|h|\le H/2,\qquad E\in\sigma(H_{\lambda_*}),
E=−2b3+ax,x∈I.(62)E=-\frac{2b}{\sqrt{3}}+ax,\qquad x\in I. \tag*{(62)}

Every spectral energy in the controlled window is therefore subcritical. We now show that the window carries a finite fraction of the spectrum.

In the coordinates of (B3), each transfer step differs from the rotation RxR_x by at most KrotδηK_{\mathrm{rot}}\delta_\eta. The angular advance per cell therefore differs from that of RxR_x by O(δη)O(\delta_\eta), uniformly in x∈Ix\in I, and the rotation number satisfies

∣ϱ(x)−ϱ0(x)∣≤Bδη,(63)|\varrho(x)-\varrho_0(x)|\le B\delta_\eta, \tag*{(63)}

where ϱ0(x)\varrho_0(x) is the rotation number of M0(x)M_0(x). Since M0(x)M_0(x) has unit determinant and ∣tr⁡M0∣<2|\operatorname{tr}M_0|<2, it is conjugate to a rotation by the angle 2πϱ02\pi\varrho_0, with

2cos⁡(2πϱ0(x))=tr⁡M0(x)=−4x3.(64)2\cos(2\pi\varrho_0(x))=\operatorname{tr}M_0(x)=-\frac{4x}{\sqrt{3}}. \tag*{(64)}

At the window edges x=∓3/4x=\mp\sqrt{3}/4 the trace equals ∓1\mp1, giving 2π∣ϱ0∣=π/32\pi|\varrho_0|=\pi/3 and 2π/32\pi/3. The negative lower-left entry of M0M_0 fixes the direction, so ϱ0\varrho_0 covers [−1/3,−1/6][-1/3,-1/6] across II.

The cell matrix MEM_E is the product of the two one-site transfer matrices, (xm,ym−1)→(ym,xm)→(xm+1,ym)(x_m,y_{m-1})\to(y_m,x_m)\to(x_{m+1},y_m). As in oscillation theory, each state below EE adds a half-turn to the solution at energy EE, so the rotation number determines the integrated density of states [64, 65]. Across the full lower band, ∣x∣≤3/2|x|\le\sqrt{3}/2, the trace runs from 22 to −2-2 and ∣ϱ0∣|\varrho_0| covers [0,1/2][0,1/2]. The window covers an interval of length 1/61/6 within it, and therefore contains a fraction 1/3+O(δη)1/3+O(\delta_\eta) of the lower-band states. Since the integrated density of states increases only on the spectrum, these states are in the spectrum, and by Eq. (62) all of them are subcritical.

Finite recovery interval

We now show that recovery survives a small change in λ\lambda, rather than occurring only at the cancellation point. Write Δλ=λ−λ∗\Delta\lambda=\lambda-\lambda_* and follow the lower-cell level by setting

E=−b2+λ2+ax,x∈I.E=-\sqrt{b^2+\lambda^2}+ax,\qquad x\in I.

The exact transfer entries are

M11=−2b2+λ2bx+rx2−2ηλab[2λ−b2+λ2+ax]cos⁡θ−3η2λ2abcos⁡2θ,M12=b2+λ2−λ−ax−ηλcos⁡θb,M21=−b2+λ2−λ+ax−3ηλcos⁡θb,M22=−r.(65)\begin{aligned} M_{11}&=-\frac{2\sqrt{b^2+\lambda^2}}{b}x+rx^2\\ &\quad-\frac{2\eta\lambda}{ab}\left[2\lambda-\sqrt{b^2+\lambda^2}+ax\right]\cos\theta\\ &\quad-\frac{3\eta^2\lambda^2}{ab}\cos^2\theta,\\ M_{12}&=\frac{\sqrt{b^2+\lambda^2}-\lambda-ax-\eta\lambda\cos\theta}{b},\\ M_{21}&=\frac{-\sqrt{b^2+\lambda^2}-\lambda+ax-3\eta\lambda\cos\theta}{b},\\ M_{22}&=-r. \tag*{(65)} \end{aligned}

The cancellation is controlled by 2λ−b2+λ22\lambda-\sqrt{b^2+\lambda^2}. This expression vanishes at λ∗=b/3\lambda_* = b/\sqrt{3}, and its derivative there is 3/23/2. For ∣Δλ∣≪b|\Delta\lambda|\ll b,

2λ−b2+λ2=32Δλ+O(Δλ2b).(66)2\lambda-\sqrt{b^2+\lambda^2}=\frac{3}{2}\Delta\lambda+O\left(\frac{\Delta\lambda^2}{b}\right). \tag*{(66)}

Its contribution to M11M_{11} is therefore of order η∣Δλ∣/a\eta|\Delta\lambda|/a. This is the restored first-order modulation measured relative to the hopping.

The phase-independent terms change by order ∣Δλ∣/b|\Delta\lambda|/b. The terms proportional to axcos⁡θax\cos\theta and cos⁡2θ\cos^2\theta remain of order η\eta and η2/r\eta^2/r, respectively, while the weak-hopping corrections remain of order rr. Using the same bounds on xx and cos⁡θ\cos\theta as at cancellation gives

∥ME−M0(x)∥strip≤C[δη+∣Δλ∣b+η∣Δλ∣a],(67)\left\|M_E-M_0(x)\right\|_{\mathrm{strip}}\le C\left[\delta_\eta+\frac{|\Delta\lambda|}{b}+\frac{\eta|\Delta\lambda|}{a}\right], \tag*{(67)}

for x∈Ix\in I and λ\lambda in a fixed neighbourhood of λ∗\lambda_*. The constant CC is independent of η\eta and Δλ\Delta\lambda in this neighbourhood.

To keep this error small, choose ∣Δλ∣<c∗ ⁣a/η|\Delta\lambda|<c_*\!a/\eta, with fixed c∗>0c_*>0. Since a/b=ηpa/b=\eta^p, Eq. (67) becomes

∥ME−M0(x)∥strip≤C[δη+c∗ηp−1+c∗].(68)\left\|M_E-M_0(x)\right\|_{\mathrm{strip}}\le C\left[\delta_\eta+c_*\eta^{p-1}+c_*\right]. \tag*{(68)}

First choose c∗c_* so that Cc∗Cc_* is below the smallness threshold required by the recovery argument. Then take η\eta sufficiently small: both δη\delta_\eta and ηp−1\eta^{p-1} vanish because 1<p<21<p<2. The same branch argument applies throughout this interval. Equation (B11) keeps the transformed matrix uniformly close to a rotation when c∗c_* is chosen sufficiently small and then η\eta is reduced. Hence every spectral energy in the window E=−b2+λ2+axE=-\sqrt{b^2+\lambda^2}+ax, x∈Ix\in I, is subcritical. The rotation-number estimate of Appendix B applies with the same uniformity, so at every coupling in this interval the window holds a fraction 1/3+O(δη)1/3+O(\delta_\eta) of the lower-band states, all subcritical. This establishes Eq. (37).

The first-order AA description gives a separate leading estimate of the interval width. Its projected modulation is

v−(λ)=ηλ(1−2λb2+λ2).v_-(\lambda)=\eta\lambda\left(1-\frac{2\lambda}{\sqrt{b^2+\lambda^2}}\right).

Using Eq. (66),

v−(λ)=−3η4Δλ+O(ηΔλ2b).(69)v_{-}(\lambda)=-\frac{3\eta}{4}\Delta\lambda+O\left(\frac{\eta\Delta\lambda^{2}}{b}\right). \tag*{(69)}

The leading hopping at cancellation is ∣teff(λ∗)∣=3a/4|t_{\mathrm{eff}}(\lambda_{\ast})|=\sqrt{3}a/4. Linearizing the modulation and evaluating the hopping at this point gives

ρ(λ)=∣v−(λ)∣2∣teff(λ)∣≃3η2a∣Δλ∣.\rho(\lambda)=\frac{|v_{-}(\lambda)|}{2|t_{\mathrm{eff}}(\lambda)|}\simeq\frac{\sqrt{3}\eta}{2a}|\Delta\lambda|.

On the scale ∣Δλ∣∼a/η|\Delta\lambda|\sim a/\eta, ∣Δλ∣/b∼ηp−1→0|\Delta\lambda|/b\sim\eta^{p-1}\to0. The linearization and the use of the hopping at λ∗\lambda_{\ast} are therefore consistent to leading order. The AA condition ρ<1\rho<1 gives the estimated width

∣λ−λ∗∣≲2a3η,|\lambda-\lambda_{\ast}|\lesssim\frac{2a}{\sqrt{3}\eta},

as quoted in Eq. (38).

Positive exponent before recovery

We show that every spectral energy has a positive physical exponent at the earlier coupling λe=b/4\lambda_{\mathrm{e}}=b/4. Assume η≤10−4\eta\leq10^{-4} and r=a/b≤η/4r=a/b\leq\eta/4. The weak hopping and modulation shift the spectrum by at most ηb\eta b from the isolated-cell energies ±17b/4\pm\sqrt{17}b/4. Consequently,

∣E2−b2−λe2∣≤94ηb2,∣E+λe∣≥3b4,∣E+2λe∣≥b2.(70)|E^{2}-b^{2}-\lambda_{\mathrm{e}}^{2}|\leq\frac{9}{4}\eta b^{2},\qquad |E+\lambda_{\mathrm{e}}|\geq\frac{3b}{4},\qquad |E+2\lambda_{\mathrm{e}}|\geq\frac{b}{2}. \tag*{(70)}

These bounds hold throughout both spectral bands.

Eliminating the BB amplitudes gives an exact equation for the AA amplitudes xmx_m:

xm+1=QE(θm)xm−RE(θm)xm−1.x_{m+1}=Q_E(\theta_m)x_m-R_E(\theta_m)x_{m-1}.

Define

dE(θ)=E+λe+ηλecos⁡θ.d_E(\theta)=E+\lambda_{\mathrm{e}}+\eta\lambda_{\mathrm{e}}\cos\theta.

The coefficients are

RE(θ)=dE(θ)dE(θ−2πβ),R_E(\theta)=\frac{d_E(\theta)}{d_E(\theta-2\pi\beta)},
QE(θ)=1ab[E2−b2−λe2−2ηλe(E+2λe)cos⁡θ−3η2λe2cos⁡2θ]−rRE(θ).(71)Q_E(\theta)=\frac{1}{ab}\left[E^{2}-b^{2}-\lambda_{\mathrm{e}}^{2}-2\eta\lambda_{\mathrm{e}}(E+2\lambda_{\mathrm{e}})\cos\theta-3\eta^{2}\lambda_{\mathrm{e}}^{2}\cos^{2}\theta\right]-rR_E(\theta). \tag*{(71)}

We first establish growth at the imaginary phase h=ln⁡100h=\ln100. This shift enhances the cosine: ∣cos⁡(θ+ih)∣≥sinh⁡h>49|\cos(\theta+ih)|\geq\sinh h>49. The term linear in cos⁡θ\cos\theta therefore dominates the phase-independent term, while the extra factor of η\eta keeps the quadratic correction small. The denominator also stays away from zero: b/2≤∣dE(θ+ih)∣≤3b/2b/2\leq|d_E(\theta+ih)|\leq3b/2. Using Eq. (70), we obtain

∣RE(θ+ih)∣≤3,∣QE(θ+ih)∣≥8ηr.|R_E(\theta+ih)|\leq3,\qquad|Q_E(\theta+ih)|\geq\frac{8\eta}{r}.

Thus the coefficient multiplying the current amplitude is much larger than the coefficient multiplying the previous one.

To see how this produces sustained growth, start with a transfer vector satisfying ∣v2∣≤∣v1∣|v_2|\leq|v_1|. The next vector is (v1′,v2′)=(QEv1−REv2,v1)(v'_1,v'_2)=(Q_Ev_1-R_Ev_2,v_1), so

∣v1′∣≥(∣QE∣−∣RE∣)∣v1∣≥7ηr∣v1∣,∣v2′∣≤∣v1′∣.|v'_1|\geq\left(|Q_E|-|R_E|\right)|v_1|\geq\frac{7\eta}{r}|v_1|,\qquad|v'_2|\leq|v'_1|.

The same condition therefore holds after every step, and the amplitude grows by at least 7η/r7\eta/r per cell. Eliminating the BB amplitudes is a bounded change of transfer coordinates and does not change the Lyapunov exponent. Hence

L(E,h)≥ln⁡7ηr.L(E,h)\geq\ln\frac{7\eta}{r}.

The maximum slope of L(E,h)L(E,h) is two. Returning from h=ln⁡100h=\ln100 to the real axis can therefore reduce the exponent by at most 2ln⁡1002\ln100. With the per-site normalization γ=L/2\gamma=L/2, this gives

γ(E,λe)≥12ln⁡7η104r.(72)\gamma(E,\lambda_{\mathrm{e}})\geq\frac{1}{2}\ln\frac{7\eta}{10^{4}r}. \tag*{(72)}

For r=ηpr=\eta^{p} with p>1p>1, the ratio η/r\eta/r diverges as η→0\eta\to0. The bound is therefore positive at every spectral energy for sufficiently small η\eta.

Together with the recovered subcritical spectral set and the later positive-exponent bound in Eq. (56), this establishes recovery between two regimes with uniformly positive exponents.

References

References

  1. [1]P. W. Anderson, Absence of Diffusion in Certain Random Lattices, Phys. Rev. 109, 1492 (1958).
  2. [2]E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V. Ramakrishnan, Scaling theory of localization: Ab- sence of quantum diffusion in two dimensions, Phys. Rev. Lett. 42, 673 (1979).
  3. [3]S. Aubry and G. André, Analyticity breaking and Anderson localization in incommensurate lattices, Ann. Israel Phys. Soc 3, 18 (1980).
  4. [4]P. G. Harper, Single band motion of conduction electrons in a uniform magnetic field, Proc. Phys. Soc. Sect. A 68, 874 (1955).DOI
  5. [5]D. R. Hofstadter, Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields, Phys. Rev. B 14, 2239 (1976).
  6. [6]S. Y. Jitomirskaya, Metal-insulator transition for the almost Mathieu operator, Ann. Math. 150, 1159 (1999).arxiv.org/abs/math/9911265
  7. [7]B. Kramer and A. MacKinnon, Localization: theory and experiment, Rep. Prog. Phys. 56, 1469 (1993).DOI
  8. [8]F. Evers and A. D. Mirlin, Anderson transitions, Rev. Mod. Phys. 80, 1355 (2008).
  9. [9]B. Hetényi, Scaling of the bulk polarization in extended and localized phases of a quasiperiodic model, Phys. Rev. B 110, 125124 (2024).DOI
  10. [10]B. Hetényi and I. Balogh, Numerical study of the localization transition of aubry-andré type models, Phys. Rev. B 112, 144203 (2025).DOI
  11. [11]J. Biddle, B. Wang, D. Priour Jr, and S. Das Sarma, Localization in one-dimensional incommensurate lattices beyond the Aubry-André model, Phys. Rev. A 80, 021603 (2009).DOI
  12. [12]X. Li and S. Das Sarma, Mobility edge and intermediate phase in one-dimensional incommensurate lattice potentials, Phys. Rev. B 101, 064203 (2020).DOI
  13. [13]S. Ganeshan, J. Pixley, and S. Das Sarma, Nearest neighbor tight binding models with an exact mobility edge in one dimension, Phys. Rev. Lett. 114, 146601 (2015).pubmed.ncbi.nlm.nih.gov/25910146
  14. [14]T. Liu, X. Xia, S. Longhi, and L. Sanchez-Palencia, Anomalous mobility edges in one-dimensional quasiperiodic models, SciPost Phys. 12, 027 (2022).DOI
  15. [15]Z. Wang, Y. Zhang, L. Wang, and S. Chen, Engineering mobility edges in quasiperiodic lattices with exact mobility edges, Phys. Rev. B 108, 174202 (2023).
  16. [16]Y.-C. Zhang and Y.-Y. Zhang, Lyapunov exponent, mobility edges, and critical region in the generalized aubry-andré model with an unbounded quasiperiodic potential, Phys. Rev. B 105, 174206 (2022).DOI
  17. [17]A. Goswami, Exact mobility edges in a slowly varying quasiperiodic ladder model, Physical Review B 114, 094205 (2026).DOI
  18. [18]Y. Wang, X. Xia, L. Zhang, H. Yao, S. Chen, J. You, Q. Zhou, and X.-J. Liu, One-dimensional quasiperiodic mosaic lattice with exact mobility edges, Phys. Rev. Lett. 125, 196604 (2020).DOI
  19. [19]Y. Hu, Y. Wang, and Y. Wang, Exact mobility edges and coexisting extended, critical, and localized states in an extended quasiperiodic mosaic model, Phys. Rev. B 114, 134206 (2026).
  20. [20]N. Hatano and D. R. Nelson, Localization transitions in Non-Hermitian quantum mechanics, Phys. Rev. Lett. 77, 570 (1996).DOI
  21. [21]S. Longhi, Metal-insulator phase transition in a non-Hermitian Aubry–André–Harper model, Phys. Rev. B 100, 125157 (2019).DOI
  22. [22]Y.-P. Wang, C.-K. Chang, R. Okugawa, and C.-H. Hsu, Quasiperiodicity-induced bulk localization with self-similarity in non-hermitian systems, Phys. Rev. Res. 7, 043353 (2025).DOI
  23. [23]Y.-Q. Zheng, S.-Z. Li, and Z. Li, Emergent multiloop nested point gap in a non-hermitian quasiperiodic lattice, Phys. Rev. B 111, 104204 (2025).DOI
  24. [24]X.-P. Jiang, Z. Liu, Y. Hu, H. Hou, and L. Pan, Localization and mobility edges in non-hermitian continuous quasiperiodic systems, New J. Phys. 27, 083201 (2025).DOI
  25. [25]G.-Q. Li, Z.-Y. Lin, Y.-J. Dong, Y.-F. Xue, C.-Y. Ren, and P. Peng, Multiple mobility rings in non-hermitian Su-Schrieffer-Heeger chain with quasiperiodic potentials, Phys. Lett. A, 131426 (2026).
  26. [26]S.-R. He, J. F. Fan, L. Li, R. Wang, D.-H. Xu, and Z.-H. Wang, Exact mobility edges and hidden self-duality in non-Hermitian quasiperiodic network models, Phys. Rev. B 113, 224205 (2026).DOI
  27. [27]W. Wang, T. Li, and W. Yi, Anomalous localization and duality in non-hermitian quasiperiodic models, Phys. Rev. B 114, 144205 (2026).DOI
  28. [28]S. Longhi, Topological phase transition in non-Hermitian quasicrystals, Phys. Rev. Lett. 122, 237601 (2019).DOI
  29. [29]Y. Liu, X.-P. Jiang, J. Cao, and S. Chen, Non-Hermitian mobility edges in one-dimensional quasiperiodic systems with parity-time symmetry, Phys. Rev. B 101, 174205 (2020).DOI
  30. [30]M. Rossignolo and L. Dell’Anna, Localization transitions and mobility edges in coupled Aubry-André chains, Phys. Rev. B 99, 054211 (2019).DOI
  31. [31]R. Wang, X. M. Yang, and Z. Song, Localization transitions and mobility edges in quasiperiodic ladder, J. Phys.: Condens. Matter 33, 365403 (2021).DOI
  32. [32]A. Szabó and U. Schneider, Mixed spectra and partially extended states in a two-dimensional quasiperiodic model, Phys. Rev. B 101, 014205 (2020).DOI
  33. [33]C. W. Duncan, Critical states and anomalous mobility edges in two-dimensional square quasiperiodic systems, Phys. Rev. B 109, 014210 (2024).DOI
  34. [34]A. Ahmed, A. Ramachandran, I. M. Khaymovich, and A. Sharma, Flat band based multifractality in the all-band-flat diamond chain, Phys. Rev. B 106, 205119 (2022).DOI
  35. [35]T. Yildiz, B. Tanatar, and B. Hetényi, Localization transitions in a half-filled helical Aubry-André model, Phys. Rev. B 114, 094206 (2026).DOI
  36. [36]T. Yildiz and B. Tanatar, Localization and persistent currents in a quasiperiodic disordered helical lattice, Scientific Reports 15, 37307 (2025).DOI
  37. [37]G. Roati, C. D’Errico, L. Fallani, M. Fattori, C. Fort, M. Zaccanti, G. Modugno, M. Modugno, and M. Inguscio, Anderson localization of a non-interacting Bose–Einstein condensate, Nature 453, 895 (2008).
  38. [38]Y. Lahini, R. Pugatch, F. Pozzi, M. Sorel, R. Morandotti, N. Davidson, and Y. Silberberg, Observation of a localization transition in quasiperiodic photonic lattices, Phys. Rev. Lett. 103, 013901 (2009).pubmed.ncbi.nlm.nih.gov/19659147
  39. [39]D. M. Basko, I. L. Aleiner, and B. L. Altshuler, Metal-insulator transition in a weakly interacting many-electron system with localized single-particle states, Ann. Phys. 321, 1126 (2006).
  40. [40]D. A. Abanin, E. Altman, I. Bloch, and M. Serbyn, Colloquium: Many-body localization, thermalization, and entanglement, Rev. Mod. Phys. 91, 021001 (2019).DOI
  41. [41]M. Schreiber, S. S. Hodgman, P. Bordia, H. P. Lüschen, M. H. Fischer, R. Vosk, E. Altman, U. Schneider, and I. Bloch, Observation of many-body localization of interacting fermions in a quasirandom optical lattice, Science 349, 842 (2015).DOI
  42. [42]J. Smith, A. Lee, P. Richerme, B. Nevenhuis, P. W. Hess, P. Hauke, M. Heyl, D. A. Huse, and C. Monroe, Many-body localization in a quantum simulator with programmable random disorder, Nat. Phys. 12, 907 (2016).DOI
  43. [43]P. Roushan, C. Neill, J. Tangpanitanon, V. M. Bastidas, A. Megrant, R. Barends, Y. Chen, Z. Chen, B. Chiaro, A. Dunsworth, A. Fowler, B. Foxen, M. Giustina, E. Jeffrey, J. Kelly, E. Lucero, J. Mutus, M. Neeley, C. Quintana, D. Sank, A. Vainsencher, J. Wenner, T. White, H. Neven, D. G. Angelakis, and J. Martinis, Spectroscopic signatures of localization with interacting photons in superconducting qubits, Science 358, 1175 (2017).DOI
  44. [44]S. Roy, T. Mishra, B. Tanatar, and S. Basu, Reentrant Localization Transition in a Quasiperiodic Chain, Phys. Rev. Lett. 126, 106803 (2021).DOI
  45. [45]S. Roy, S. Chattopadhyay, T. Mishra, and S. Basu, Critical analysis of the reentrant localization transition in a one-dimensional dimerized quasiperiodic lattice, Phys. Rev. B 105, 214203 (2022).DOI
  46. [46]T. F. Allard and G. Weick, Reentrant localization transition in a dimerized quasiperiodic dipolar chain, J. Phys. Condens. Matter 38, 225401 (2026).DOI
  47. [47]Y. Yang, X. Zhang, and X. Li, Multiple topological and localization transitions in Su–Schrieffer–Heeger model with off-diagonal quasiperiodic modulation, Commun. Theor. Phys. 78, 095702 (2026).DOI
  48. [48]T. Yildiz and B. Tanatar, Reentrant localization length in a quasiperiodic thue-morse chain, Scientific Reports 10.1038/s41598-026-71207-9 (2026).DOI
  49. [49]A. Padhan, M. K. Giri, S. Mondal, and T. Mishra, Emergence of multiple localization transitions in a one-dimensional quasiperiodic lattice, Phys. Rev. B 105, L220201 (2022).DOI
  50. [50]X. Wei, Reentrant localization transition induced by a composite potential, Phys. Rev. B 111, 134204 (2025).DOI
  51. [51]P.-J. Chang, M.-X. Zheng, Q.-F. Lu, Z.-Y. Ma, D. Ruan, and G.-L. Long, Dimer-driven multiple reentrant localization with composite potential, Phys. Rev. A 113, 012206 (2026).arxiv.org/abs/2509.21338
  52. [52]L. Gong, Y. Feng, and Y. Ding, Anderson localization in one-dimensional quasiperiodic lattice models with nearest-and next-nearest-neighbor hopping, Phys. Lett. A 381, 588 (2017).DOI
  53. [53]P.-J. Chang, Q.-B. Zeng, J. Pi, D. Ruan, and G.-L. Long, Investigation of reentrant localization transition in one-dimensional quasi-periodic lattice with long-range hopping, New J. Phys. 27, 053501 (2025).arxiv.org/abs/2412.13518
  54. [54]H. Wang, X. Zheng, J. Chen, L. Xiao, S. Jia, and L. Zhang, Fate of the reentrant localization phenomenon in the one-dimensional dimerized quasiperiodic chain with long-range hopping, Phys. Rev. B 107, 075128 (2023).DOI
  55. [55]S. Ganguly, S. Sarkar, K. Mondal, and S. K. Maiti, Phenomenon of multiple reentrant localization in a double-stranded helix with transverse electric field, Sci. Reports 14, 3059 (2024).DOI
  56. [56]S. Ganguly, S. Chattopadhyay, K. Mondal, and S. K. Maiti, Critical analysis of multiple reentrant localization in an antiferromagnetic helix with transverse electric field: Hopping dimerization-free scenario, SciPost Phys. 8, 012 (2025).DOI
  57. [57]A. Padhan, S. R. Padhi, and T. Mishra, Complete delocalization and reentrant topological transition in a non-Hermitian quasiperiodic lattice, Phys. Rev. B 109, L020203 (2024).DOI
  58. [58]C. Wu, J. Fan, G. Chen, and S. Jia, Non-Hermiticity-induced reentrant localization in a quasiperiodic lattice, New J. Phys. 23, 123048 (2021).DOI
  59. [59]X.-P. Jiang, Y. Qiao, and J.-P. Cao, Mobility edges and reentrant localization in one-dimensional dimerized non-Hermitian quasiperiodic lattice, Chin. Phys. B 30, 097202 (2021).DOI
  60. [60]H. Wang, X. Zheng, L. Xiao, S. Jia, J. Chen, and L. Zhang, Coexistence of reentrant localization and dynamical delocalization in a one-dimensional non-Hermitian quasiperiodic lattice, Phys. Rev. B 112, 054202 (2025).
  61. [61]X. Wang, P.-J. Chang, Q. Lu, M. Zheng, D. Ruan, and G.-L. Long, Orbital cancellation and reentrant localization in a quasiperiodic trimer ladder (2026), arXiv:2610.16197 [cond-mat.dis-nn].arxiv.org/abs/2610.16197
  62. [62]A. Avila, Global theory of one-frequency Schrödinger operators, Acta Math. 215, 1 (2015).DOI
  63. [63]C. A. Marx and S. Jitomirskaya, Dynamics and spectral theory of quasi-periodic Schrödinger-type operators, Ergodic Theory Dynam. Systems 37, 2353 (2017), arXiv:1503.05740.DOI
  64. [64]F. Delyon and B. Souillard, The rotation number for finite difference operators and its properties, Commun. Math. Phys. 89, 415 (1983).DOI
  65. [65]G. Teschl, Jacobi operators and completely integrable nonlinear lattices, Vol. 72 (American Mathematical Society, Providence, RI, 2000).DOI

Paper details

vlnc:W_t8z56hqg
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