Complete hierarchical structure of the spectral bands in the Kohmoto model

Abstract: We study the Kohmoto model, a family of discrete Schrödinger operators with Sturmian potentials depending on a frequency and a coupling constant. We prove that, for all non-vanishing coupling constants, all spectral bands admit a hierarchical structure. This structure offers a variety of applications, including a detailed description of the Kohmoto butterfly and a central step towards the resolution of the dry ten Martini problem for Sturmian Hamiltonians, which we carry out in a subsequent work.

• The paper establishes that every spectral band of every rational periodic approximant has a definitive type A or B for all nonzero coupling strengths, extending Raymond’s result beyond V>4.
• The authors combine continued-fraction induction, Floquet–Bloch matrix interlacing, Chebyshev trace calculations, and Lipschitz spectral continuity to control band nesting even when bands overlap.
• Each approximant contains exactly q_k−q_{k−1} type-A bands and q_{k−1} type-B bands, with unique forward nesting that clarifies the self-similar structure of the Kohmoto butterfly.

The Kohmoto model is the family of discrete Schrödinger operators $H_{\alpha,V}$ on $\ell^2(\mathbb{Z})$ with Sturmian potentials $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$, indexed by a frequency $\alpha\in[0,1]$ and coupling constant $V$. For rational $\alpha=p/q$ (coprime), the operator is periodic with spectrum consisting of exactly $q$ closed intervals — the spectral bands — and plotting these spectra over $\alpha\in[0,1]\cap\mathbb{Q}$ produces the self-similar fractal known as the Kohmoto butterfly. The paper under review establishes that, for every non-zero coupling constant, all spectral bands of all periodic approximants admit a complete hierarchical classification into two types, $A$ and $B$, extending a classical result of Raymond from the large-coupling regime $\ell^2(\mathbb{Z})$0 to all $\ell^2(\mathbb{Z})$1 (2607.06361).

The main theorem

The central result states that for all $\ell^2(\mathbb{Z})$2 and every finite continued fraction expansion $\ell^2(\mathbb{Z})$3 with evaluation $\ell^2(\mathbb{Z})$4, every spectral band in $\ell^2(\mathbb{Z})$5 is either of type $\ell^2(\mathbb{Z})$6 or of type $\ell^2(\mathbb{Z})$7, and its type is independent of the value of $\ell^2(\mathbb{Z})$8 (respectively $\ell^2(\mathbb{Z})$9). This is a strong dichotomy: previously it was only available for $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$0 via Raymond's work, where bands are well separated and trace-map methods suffice. Extending to $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$1 is delicate because bands begin to overlap, so the relative position of bands can no longer be controlled by traces alone.

The authors introduce an "augmented" space $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$2 of finite continued fraction expansions, including non-standard expansions ending in $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$3 or $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$4. Crucially, the type of a band depends on the expansion $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$5 itself rather than on its rational value $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$6; since each rational in $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$7 has exactly two reduced expansions, a duality proposition shows that the two representations swap types $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$8. This is precisely why working in the full space $$\omega_\alpha(n)=\chi_{[1-\alpha,1)}(n\alpha \bmod 1)$$9, rather than with rationals, is essential.

Backward and forward types

Two notions combine to define the types. A band $\alpha\in[0,1]$0 of $\alpha\in[0,1]$1 is of backward type $\alpha\in[0,1]$2 if it is strictly contained in some band of $\alpha\in[0,1]$3, and of backward type $\alpha\in[0,1]$4 if strictly contained in a band of $\alpha\in[0,1]$5 (with weak variants allowing mere containment). The forward type encodes how a band $\alpha\in[0,1]$6 of $\alpha\in[0,1]$7 relates to bands of the next approximants $\alpha\in[0,1]$8 and $\alpha\in[0,1]$9: for type $V$0 there exist $V$1 bands of $V$2 strictly contained in $V$3 and not of weak backward type $V$4; for type $V$5 there exists, for each $V$6, a nested chain ("tower property") of $V$7 bands of backward type $V$8 inside $V$9; additionally an interlacing ordering property holds among these bands.

A band is of $\alpha=p/q$0-type $\alpha=p/q$1 (resp. $\alpha=p/q$2) if it satisfies both the corresponding backward and forward conditions, and of type $\alpha=p/q$3 (resp. $\alpha=p/q$4) if this holds for all $\alpha=p/q$5. The paper proves an equivalent characterization: type $\alpha=p/q$6 is equivalent to strict containment in a band of the previous convergent's spectrum, while type $\alpha=p/q$7 is equivalent to not being contained in that spectrum together with containment in the spectrum of the second-to-last convergent.

Proof architecture

The proof proceeds by induction over the space $\alpha=p/q$8, split into horizontal steps (increasing the number of digits) and vertical steps (varying the last digit). Two ingredients drive it:

• Backward implies forward: if each band of $\alpha=p/q$9 has a fixed backward type for all $q$0, then it satisfies the full $q$1-property for all $q$2, i.e., the associated critical thresholds vanish.
• Induction base: explicit analysis of the bands $q$3 (type $q$4) and $q$5 (type $q$6), using dilated Chebyshev polynomials $q$7 to compute transfer-matrix traces exactly.

A key technical device is a uniform Lipschitz bound $q$8, which allows properties established at large $q$9 to be propagated downward continuously, combined with trace estimates showing that candidate collision energies cannot actually be band edges.

Spectral tools

The spectral analysis rests on two complementary descriptions. Floquet–Bloch theory reduces spectra to unions of eigenvalues of finite Hermitian matrices $\alpha\in[0,1]\cap\mathbb{Q}$0, $\alpha\in[0,1]\cap\mathbb{Q}$1, with band edges given by eigenvalues at $\alpha\in[0,1]\cap\mathbb{Q}$2. An antisymmetry under a diagonal sign-flip unitary yields $\alpha\in[0,1]\cap\mathbb{Q}$3, reducing everything to $\alpha\in[0,1]\cap\mathbb{Q}$4.

The paper's main new analytic tool is an interlacing theorem for Floquet–Bloch matrices: since the diagonal of $\alpha\in[0,1]\cap\mathbb{Q}$5 concatenates those of $\alpha\in[0,1]\cap\mathbb{Q}$6 and $\alpha\in[0,1]\cap\mathbb{Q}$7, the matrix $\alpha\in[0,1]\cap\mathbb{Q}$8 differs from the direct sum $\alpha\in[0,1]\cap\mathbb{Q}$9 by a symmetric rank-two perturbation with zero trace — provided the triple $A$0 contains an even number of $A$1's, a condition termed admissibility. Weyl-type inequalities then give $A$2, with strict inequalities when $A$3 is simple. Admissibility admits a purely combinatorial characterization via band indices and left/right endpoint data, and also a trace-theoretic one: admissibility holds iff the product of the three relevant trace values is positive for all $A$4 — a criterion the authors flag as central to their companion resolution of the dry ten Martini problem.

Counting functions relate band indices to eigenvalue counts, and detailed index identities (e.g., $A$5) feed the interlacing arguments that establish the forward-type properties.

Consequences

Two corollaries quantify the structure. First, writing $A$6 for the denominators of the continued-fraction convergents of $A$7, the spectrum $A$8 contains exactly $A$9 bands of type $B$0 and $B$1 bands of type $B$2, for all $B$3. Second, the bands appearing in the forward-type nesting are unique for all $B$4 — strengthening Raymond's uniqueness statement, which held only for $B$5. These counting results mean the hierarchical decomposition is exhaustive and rigid across the entire coupling range, which should enable fractal-dimension estimates of the Kohmoto butterfly at arbitrary coupling, analogous to what the $B$6 structure afforded in earlier work.

Limitations and open questions

The theorem covers periodic approximants with $B$7; the case $B$8 is trivially excluded since all spectra collapse to $B$9. The irrational-frequency Sturmian operators themselves are treated only indirectly through rational approximation, with the application to the dry ten Martini problem deferred to subsequent work. The authors note that whether a band of type $\ell^2(\mathbb{Z})$00 may also satisfy the containment characterizing type $\ell^2(\mathbb{Z})$01 at particular values of $\ell^2(\mathbb{Z})$02 (it can, e.g., in $\ell^2(\mathbb{Z})$03) means the classification genuinely requires the augmented expansion space rather than the rational value alone. Open questions include quantitative fractal-dimension consequences of the hierarchy at small coupling and the precise number-theoretic description of the butterfly's self-similarity via continued fractions.

Conclusion

This work completes the $\ell^2(\mathbb{Z})$04 classification of spectral bands in the Kohmoto model for all non-vanishing couplings, removing the $\ell^2(\mathbb{Z})$05 restriction that had stood for three decades. The combination of an interlacing theorem for Floquet–Bloch matrices under an admissibility condition, uniform Lipschitz control of band edges, and Chebyshev-based trace computations provides a robust framework whose immediate payoff is a detailed structural description of the Kohmoto butterfly and a central step toward resolving the dry ten Martini problem for Sturmian Hamiltonians.