betaFeedback

WorksPhysics and Astronomy

Thouless's bandwidth conjecture: a proof by elliptic periods

Abstract

We prove Thouless's bandwidth conjecture for the critical rational Harper operator: the denominator times the total bandwidth tends to 32G/π32G/π, uniformly in the reduced numerator, where GG is Catalan's constant. We identify a universal limit of the normalized even coarea density. Positive real part and evenness give an integrable endpoint bound, so the local law determines the bandwidth. An exact finite-matrix formula reduces the local law to one normalized period for each fixed odd numerator. Explicit boundary polynomials, point blowups, and intersections determine a compact genus-one family. Nodal-annulus calculations fix its period equation and hence the universal density. An exact rational Fourier identity and uniform closed-path bounds extend the result to all reduced numerators and both denominator parities. The singular matrix limit, including its subtraction and normalization, is proved separately in detail.

Introduction

The total bandwidth

For a rational parameter α=p/q\alpha=p/q, the spectrum of the critical Harper operator consists of finitely many bands. Their number increases with the denominator qq, and their total length tends to zero. Thouless’s conjecture predicts the precise rate: the denominator times the total length tends to a universal constant involving Catalan’s constant. We prove this prediction uniformly in the numerator.

We use the normalization

(Hα,ϑu)n=un+1+un−1+2cos⁡(2παn+ϑ)un,u∈ℓ2(Z).(1)(H_{\alpha,\vartheta}u)_n=u_{n+1}+u_{n-1}+2\cos(2\pi\alpha n+\vartheta)u_n,\qquad u\in\ell^2(\mathbb Z). \tag*{(1)}

The coefficients of un−1u_{n-1} and un+1u_{n+1} are one, and the coefficient of the cosine is two. The real parameters are α\alpha and ϑ\vartheta. For relatively prime integers 1≤p<q1\le p<q, define

Sp,q=⋃ϑ∈R/2πZσ(Hp/q,ϑ).(2)S_{p,q}=\bigcup_{\vartheta\in\mathbb R/2\pi\mathbb Z}\sigma(H_{p/q,\vartheta}). \tag*{(2)}

Here σ(H)\sigma(H) denotes the spectrum of HH. The union over ϑ\vartheta is part of the definition. Its Lebesgue measure ∣Sp,q∣|S_{p,q}| is the total bandwidth considered here.

Theorem 1.1 (Thouless’s bandwidth asymptotic). For the normalization (1),

lim⁡q→∞sup⁡1≤p<q(p,q)=1∣q∣Sp,q∣−32Gπ∣=0,G=∑j=0∞(−1)j(2j+1)2.(3)\lim_{q\to\infty}\sup_{\substack{1\le p<q\\(p,q)=1}}\left|q|S_{p,q}|-\frac{32G}{\pi}\right|=0,\qquad G=\sum_{j=0}^{\infty}\frac{(-1)^j}{(2j+1)^2}. \tag*{(3)}

Here GG is Catalan’s constant.

The constant is 32G/π=9.329948928…32G/\pi=9.329948928\ldots. In particular, the result holds along every sequence of reduced rational parameters approaching an irrational parameter, without any assumption on its continued fraction. The uniform statement (3) is stronger than this usual formulation.

Historical background

The bandwidth question grew out of the spectral theory of the Harper equation. Harper introduced the difference equation in his study of electrons on a lattice in a uniform magnetic field [8]. For a rational parameter, Chambers separated the dependence of the finite determinant on the two phase variables [5]. This relation reduces the union of the spectra over the phases to the inverse image of an interval under a single polynomial. Hofstadter’s calculations subsequently displayed the hierarchy of bands and gaps as the rational parameter varies [10]. The total bandwidth extracts one numerical quantity from this intricate set: the sum of the lengths of all its bands.

Thouless’s 1983 paper [23] studied this quantity both analytically and numerically. At the critical coupling, his computations suggested decay proportional to the inverse denominator, with a leading constant close to 9.33 over a broad range of numerators. His later analysis of the numerator-one regime identified the candidate 32G/π32G/\pi [24]. Here the rational parameter tends to zero as 1/q1/q, so the argument concerns a particular family of long-period operators. Thouless and Tan investigated corrections to this scaling [25]; the slow corrections in the numerator-one case help explain the difficulty of extrapolating the constant numerically. Their comparison with numerator two also emphasized that the approach to the limit can depend strongly on arithmetic features of the rational parameter.

Avron, van Mouche and Simon established important exact identities and estimates for spectral measures and rational approximation [3, 4]. Their analysis distinguishes the union of the spectra over the phase from their intersection and gives, in particular, precise results away from the critical coupling. This distinction matters for the bandwidth problem, whose rational spectrum is the union. Building on their intersection-spectrum identity, Last and Wilkinson proved the exact sum rule [18]

∑Pp,q(E)=01∣Pp,q′(E)∣=1q.\sum_{P_{p,q}(E)=0} \frac{1}{\lvert P'_{p,q}(E)\rvert} = \frac{1}{q}.

Here Pp,qP_{p,q} denotes the Chambers polynomial, defined in (4) below. It fixes the central value exactly and yielded the lower bound ∣Sp,q∣≥4/q\lvert S_{p,q}\rvert\ge4/q. The higher Taylor coefficients of the coarea density, which are needed to recover the leading constant from this local information, are additional quantities.

Last then proved the general bounds [16]

2(5+1)q<∣Sp,q∣<8eq.\frac{2(\sqrt{5}+1)}{q} < \lvert S_{p,q}\rvert< \frac{8e}{q}.

These bounds established the correct order of magnitude uniformly in the reduced numerator. The upper bound was also central to his proof of zero spectral measure at the critical coupling for a full-measure class of irrational parameters. The determination of the leading constant required finer information. Helffer and Kerdelhue supplied rigorous semiclassical proofs of the Thouless asymptotic for 1/q1/q and for 2/q2/q with qq odd [9]. Their proof analyzes the spectrum near the saddle energy of the symbol cos⁡x+cos⁡ξ\cos x+\cos\xi, where the principal contribution to the bandwidth is concentrated. Their operator has half the energy scale used here, so their constant 16G/π16G/\pi is the same constant after normalization. In both theorems the numerator is fixed while the denominator tends to infinity.

Krasovsky obtained exact formulas and asymptotic information for the derivative of the discriminant at the central energy [14]. For arbitrary fixed numerator and increasing odd denominator, his expansion gives detailed information about the central band and the clustering of nearby bands. It concerns the discriminant at one energy, whereas the bandwidth involves all inverse branches over the full interval [−4,4][-4,4]. Ouvry and Wu later proposed extensions of the Thouless formula to higher moments for numerator one [19]. Their moments are alternating sums of powers of band-edge energies, and their asymptotic analysis has numerical support; the paper explicitly leaves a firmer proof of the higher-moment extension open. These moments differ from the derivatives of the coarea density considered here.

A substantial improvement of the general upper bound was obtained by Jitomirskaya, Konstantinov and Krasovsky [12]:

∣Sp,q∣≤4πq.|S_{p,q}| \le\frac{4\pi}{q}.

Their proof uses a different gauge, a corresponding Chambers-type formula and matrix eigenvalue inequalities. Figueras and Puig’s recent computer-assisted work on open gaps also examines bandwidth scaling and stronger finite-denominator inequalities [7]. Its validated finite computations give useful evidence while leaving the asymptotic law as a conjecture. An asymptotic statement by itself does not establish those stronger finite-denominator inequalities.

The rational bandwidth problem is related to, but more quantitative than, the vanishing measure of the irrational critical spectrum. Avila and Krikorian established the measure formula for every irrational parameter [2]; Avila and Jitomirskaya proved the Cantor-spectrum assertion of the Ten Martini Problem [1]. Neither assertion specifies the leading rate of decay of rational bandwidths. The bandwidth conjecture is often formulated along rational sequences approaching a fixed irrational parameter. The formulation in Theorem 1.1 is uniform over the reduced numerators. It therefore also encompasses the fixed-numerator regimes described above. The contact theorem supplies the local information needed for this uniform conclusion, and the integral estimate in Section 2 transfers that information to the total bandwidth.

A local question behind the total bandwidth

Chambers’ formula gives a monic real polynomial Pp,qP_{p,q} of degree qq such that

Sp,q=Pp,q−1([−4,4]).(4)S_{p,q}=P_{p,q}^{-1}([-4,4]). \tag*{(4)}

We construct this polynomial explicitly in Section 3. Its zeros are simple and real. A critical value means a number Pp,q(Ec)P_{p,q}(E_c) with Pp,q′(Ec)=0P'_{p,q}(E_c)=0. All such values lie outside (−4,4)(-4,4):

Pp,q′(Ec)=0⟹∣Pp,q(Ec)∣≥4.P'_{p,q}(E_c)=0 \quad\Longrightarrow\quad|P_{p,q}(E_c)|\ge4.

A proof of this basic, well-known fact is given in Subsection 3.3. For ∣t∣<4|t|<4, set

ρp,q(t)=∑E∈RPp,q(E)=t1∣Pp,q′(E)∣.(5)\rho_{p,q}(t)=\sum_{\substack{E\in\mathbb{R}\\P_{p,q}(E)=t}}\frac{1}{|P'_{p,q}(E)|}. \tag*{(5)}

The variable EE is the original spectral variable; t=Pp,q(E)t=P_{p,q}(E) is a polynomial level. Thus ρp,q(t) dt\rho_{p,q}(t)\,dt is the total length on the EE-axis lying over a small interval of levels of length dtdt. The change-of-variables formula gives

∣Sp,q∣=∫−44ρp,q(t) dt.(6)|S_{p,q}|=\int_{-4}^{4}\rho_{p,q}(t)\,dt. \tag*{(6)}

The Last–Wilkinson identity [18] is

qρp,q(0)=q∑Pp,q(E)=01∣Pp,q′(E)∣=1.(7)q\rho_{p,q}(0)=q\sum_{\substack{P_{p,q}(E)=0}}\frac{1}{|P'_{p,q}(E)|}=1. \tag*{(7)}

It is natural to seek a local coarea law before taking the integral in (6): does the normalized density approach one function, independently of the rational parameter? For the bandwidth it is enough to answer the symmetric version of this question. Put

Rp,q(t)=q2(ρp,q(t)+ρp,q(−t)).(8)R_{p,q}(t)=\frac{q}{2}\left(\rho_{p,q}(t)+\rho_{p,q}(-t)\right). \tag*{(8)}

Then Rp,qR_{p,q} is even, Rp,q(0)=1R_{p,q}(0)=1, and q∣Sp,q∣=∫−44Rp,q(t) dtq|S_{p,q}|=\int_{-4}^{4}R_{p,q}(t)\,dt. Symmetrization avoids imposing an unnecessary parity condition on the original density.

A useful guide is the case of a fixed numerator equal to one. Its limiting matrix calculation, carried out below, reduces to the angular average of (1−t2cos⁡2ψ/16)−1/2(1-t^2\cos^2\psi/16)^{-1/2} over −π/2≤ψ≤π/2-\pi/2\le\psi\le\pi/2.

This suggests looking for the same function at every rational parameter. The Last–Wilkinson normalization (7) already fixes its value at zero. The candidate limiting profile is

G(t)=∑m=0∞(2mm)2256mt2m=2F1(12,12;1;t216),∣t∣<4.(9)\mathcal{G}(t)=\sum_{m=0}^{\infty}\frac{\binom{2m}{m}^{2}}{256^{m}}t^{2m}={}_2F_1\left(\frac{1}{2},\frac{1}{2};1;\frac{t^2}{16}\right),\qquad|t|<4. \tag*{(9)}

The symbol 2F1{}_2F_1 denotes the Gauss hypergeometric function, defined for ∣z∣<1|z|<1 by

2F1(a,b;c;z)=∑m=0∞(a)m(b)m(c)mm!zm,c∉{0,−1,−2,…}.{}_2F_1(a,b;c;z)=\sum_{m=0}^{\infty}\frac{(a)_m(b)_m}{(c)_m m!}z^m,\qquad c\notin\{0,-1,-2,\ldots\}.

Here (a)0=1(a)_0=1 and (a)m=a(a+1)⋯(a+m−1)(a)_m=a(a+1)\cdots(a+m-1) for $m\geq 1; this is the rising factorial. The subscripts indicate two upper parameters a,ba,b and one lower parameter cc. The square-lattice representation of G\mathcal{G} makes the proposed universality plausible:

G(t)=1(2π)2∫02π∫02πdx dy1−t16(2cos⁡x+2cos⁡y).(10)\mathcal{G}(t)=\frac{1}{(2\pi)^2}\int_0^{2\pi}\int_0^{2\pi}\frac{dx\,dy}{1-\frac{t}{16}(2\cos x+2\cos y)}. \tag*{(10)}

Indeed, expansion of this integral counts closed square-lattice walks of even length; their number at length 2m2m is (2mm)2\binom{2m}{m}^{2}. This explains the proposed coefficients, but does not prove that the Harper inverse-slope density has them. Establishing that identification is the main task of the paper.

Rather than attempt a direct limit of all inverse branches on the whole interval, we first examine derivatives at the central level t=0t=0. Each derivative can be computed by finite matrix algebra; the positive-real estimate below will recover the local function and justify integration up to the endpoints. We call ρp,q(2m)(0)\rho_{p,q}^{(2m)}(0) the even central contacts; these are simply the even Taylor derivatives of the coarea density.

References

  1. [1]A. Avila and S. Jitomirskaya. The ten Martini problem. Annals of Mathematics, 170(1):303–342, 2009. DOI: 10.4007/annals.2009.170.303.DOI
  2. [2]A. Avila and R. Krikorian. Reducibility or nonuniform hyperbolicity for quasiperiodic Schrödinger cocycles. Annals of Mathematics, 164(3):911–940, 2006.DOI
  3. [3]J. E. Avron, P. H. M. van Mouche, and B. Simon. On the measure of the spectrum for the almost Mathieu operator. Communications in Mathematical Physics, 132(1):103–118, 1990.
  4. [4]J. E. Avron, P. van Mouche, and B. Simon. Erratum: On the measure of the spectrum for the almost Mathieu operator. Communications in Mathematical Physics, 139:215, 1991. DOI: 10.1007/BF02102736.
  5. [5]W. G. Chambers. Linear-network model for magnetic breakdown in two dimensions. Physical Review, 140(1A):A135–A143, 1965. DOI: 10.1103/PhysRev.140.A135.DOI
  6. [6]G. Chinta, J. Jorgenson, and A. Karlsson. Zeta functions, heat kernels, and spectral asymptotics on degenerating families of discrete tori. Nagoya Mathematical Journal, 198:121–172, 2010. DOI: 10.1215/0027 7630-2009-009.arxiv.org/abs/0806.2014
  7. [7]J.-L. Figueras and J. Puig. Computer validation of open gaps for the almost Mathieu operator with critical coupling. Communications in Nonlinear Science and Numerical Simulation, 152:109407, 2026. Available online 10 October 2025. DOI: 10.1016/j.cnsns.2025.109407.DOI
  8. [8]P. G. Harper. Single band motion of conduction electrons in a uniform magnetic field. Proceedings of the Physical Society. Section A, 68(10):874–878, 1955. DOI: 10.1088/0370-1298/68/10/304.DOI
  9. [9]B. Helffer and P. Kerdelhue. On the total bandwidth for the rational Harper’s equation. Communications in Mathematical Physics, 173(2):335–356, 1995. DOI: 10.1007/BF02101237.DOI
  10. [10]D. R. Hofstadter. Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields. Physical Review B, 14(6):2239–2249, 1976. DOI: 10.1103/PhysRevB.14.2239.DOI
  11. [11]D. Husemöller. Fibre bundles, second edition, volume 111 of Graduate Texts in Mathematics. Springer, New York, 2004. DOI: 10.1007/b97292.
  12. [12]S. Jitomirskaya, L. Konstantinov, and I. Krasovsky. On the spectrum of critical almost Mathieu operators in the rational case. Journal of Spectral Theory, 12(1):11–21, 2022. DOI: 10.4171/JST/392.DOI
  13. [13]J. Jorgenson, A. Karlsson, and L. Smajlovic. The resolvent kernel on the discrete circle and twisted cosecant sums. Journal of Mathematical Analysis and Applications, 538(2):128454, 2024. DOI: 10.1016/j.jmaa.2024.128454.DOI
  14. [14]I. V. Krasovsky. On the discriminant of Harper’s equation. Letters in Mathematical Physics, 52(2):155–163, 2000.DOI
  15. [15]I. Krasovsky. Central spectral gaps of the almost Mathieu operator. Communications in Mathematical Physics, 351(1):419–439, 2017. DOI: 10.1007/s00220-016-2774-9.DOI
  16. [16]Y. Last. Zero measure spectrum for the almost Mathieu operator. Communications in Mathematical Physics, 164(2):421–432, 1994. DOI: 10.1007/BF02101708.DOI
  17. [17]Y. Last. Spectral theory of Sturm–Liouville operators on infinite intervals: a review of recent developments. In W. O. Amrein, A. M. Hinz, and D. Pearson, editors, Sturm–Liouville Theory: Past and Present, pp. 99–120. Birkhäuser, Basel, 2005. DOI: 10.1007/3-7643-7359-8_5.DOI
  18. [18]Y. Last and M. Wilkinson. A sum rule for the dispersion relations of the rational Harper equation. Journal of Physics A: Mathematical and General, 25(22):6123–6133, 1992. DOI: 10.1088/0305-4470/25/22/033.DOI
  19. [19]S. Ouvry and S. Wu, Thouless bandwidth formula in the Hofstadter model, J. Phys. A: Math. Theor. 50 (2017), no. 49, 495204. doi:10.1088/1751-8121/aa9042. arXiv:1703.09634.DOI
  20. [20]M. Reid. Chapters on algebraic surfaces. In Complex Algebraic Geometry (Park City, UT, 1993), IAS/Park City Mathematics Series, vol. 3, pp. 3–159. American Mathematical Society, Providence, RI, 1997. arXiv:alg-geom/9602006; doi:10.1090/pcms/003/02.DOI
  21. [21]B. Simon. Orthogonal Polynomials on the Unit Circle. Part 1: Classical Theory, volume 54, Part 1 of American Mathematical Society Colloquium Publications. American Mathematical Society, Providence, RI, 2005. DOI: 10.1090/coll/054.1.
  22. [22]G. Teschl. Jacobi Operators and Completely Integrable Nonlinear Lattices, volume 72 of Mathematical Surveys and Monographs. American Mathematical Society, Providence, RI, 2000. DOI: 10.1090/surv/072.DOI
  23. [23]D. J. Thouless. Bandwidths for a quasiperiodic tight-binding model. Physical Review B, 28(8):4272–4276, 1983. DOI: 10.1103/PhysRevB.28.4272.DOI
  24. [24]D. J. Thouless. Scaling for the discrete Mathieu equation. Communications in Mathematical Physics, 127(1):187–193, 1990. DOI: 10.1007/BF02096501.DOI
  25. [25]D. J. Thouless and Y. Tan. Total bandwidth for the Harper equation. III. Corrections to scaling. Journal of Physics A: Mathematical and General, 24(17):4055–4066, 1991. DOI: 10.1088/0305-4470/24/17/022.DOI
  26. [26]B. Totaro. Euler and algebraic geometry. Bulletin of the American Mathematical Society, 44(4):541–559, 2007. DOI: 10.1090/S0273-0979-07-01178-0.DOI
  27. [27]F. Y. Wu. Number of spanning trees on a lattice. Journal of Physics A: Mathematical and General, 10(6):L113–L115, 1977. DOI: 10.1088/0305-4470/10/6/004.DOI
  28. [28]B. Xie, Y. Zhao, and Y. Zhao. Special values of spectral zeta functions of graphs and Dirichlet L-functions. Journal of Number Theory, 256:136–159, 2024. DOI: 10.1016/j.jnt.2023.09.012.DOI

Paper details

vlnc:W_dzs6hx98
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 1

downloads · 0

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

  • Physics and Astronomy

OpenAI’s ChatGPT, Codex

View statement
Looking for related papers…
Contents