PaperlayerView exact revision on arXiv

Source-linked AI summary

Spectra of Non-Self-Adjoint Almost Mathieu Matrices and the Scottish Flag Operator

Simon Becker, Izak Oltman

arXiv:2608.30191v1math.SPmath.NA

TL;DR

The paper studies when spectra of non-self-adjoint almost Mathieu matrices exhibit the Scottish flag’s two-line geometry. It derives a Chambers formula and analyzes the phase-independent polynomial, proving universal two-line containment for its zeros, aligned even-periodic confinement, and explicit limiting measures, while leaving irrational-frequency operators outside its scope.

  • Problem

    The Scottish flag matrix had a two-line spectrum observed numerically but not rigorously established, motivating an exact finite-dimensional analysis of this non-self-adjoint setting.

  • Method

    The paper derives a generalized Chambers formula, isolates Q_N,φ independent of potential phase and boundary twist, and uses zero-diagonal tridiagonal structure to analyze spectral geometry.

  • Results

    The zeros of Q_N,φ lie on two perpendicular lines for every N; for even N with ϑ ∈ 2πZ/N, the matrices have the same two-line confinement, and the Scottish flag matrix satisfies spec(B_N) ⊂ X_{π/4}.

  • Takeaways & Limitations

    The Scottish flag spectrum is rigorously explained by the paper’s two-line spectral structure, with explicit limiting eigenvalue measures for even periodic matrices.

  • Takeaways & Limitations

    The paper does not pursue irrational frequencies or the associated operator on ℓ²(Z), focusing on exact finite-dimensional identities and explicit even-periodic limits.

Abstract

from arXiv · show

For $N\geq 3$ and a potential phase $\vartheta\in\mathbb{R}$, we study the non-self-adjoint almost Mathieu matrix obtained by multiplying the discrete Laplacian by a complex phase with angle $\varphi\in\mathbb{R}$, $A_N(\varphi,\vartheta)=e^{i\varphi}(S+S^{-1})/2+\operatorname{diag}(\cos(2πj/N+\vartheta))_{j\in\mathbb{Z}/N\mathbb{Z}}$, where $S e_j=e_{j+1}$ is the periodic shift on $\mathbb{C}^N$. We derive a Chambers formula and isolate the part $Q_{N,\varphi}$ of the characteristic polynomial that depends only on $N$ and $\varphi$, but not on $\vartheta$ or on a change of boundary conditions for the shift operator. We then show, for every $N$, that the zeros of $Q_{N,\varphi}$ lie on the two perpendicular lines $e^{i\varphi/2}\mathbb{R}\cup e^{i(\varphi/2+π/2)}\mathbb{R}$. For even $N$, the same property holds for the matrices $A_N(\varphi,\vartheta)$ with $\vartheta\in 2π\mathbb{Z}/N$, and we compute their limiting eigenvalue measure explicitly. For $\varphi\in[-π,π]$, the eigenvalue distribution approximates elliptic-integral densities with masses $1-|\varphi|/π$ and $|\varphi|/π$, and maximal radii $2|\cos(\varphi/2)|$ and $2|\sin(\varphi/2)|$, respectively. At $\varphi=π/2$, the central polynomial $Q_{N,\varphi}$ factors into positive quartic factors. This proves that the Scottish flag matrix, after Trefethen and Chapman, has its spectrum on the two diagonal lines of the saltire.

1. Introduction and main results

The paper rigorously establishes the Scottish flag spectrum and develops exact finite-dimensional identities for non-self-adjoint almost Mathieu matrices. It proves two-line spectral confinement under aligned even-periodic conditions, isolates a phase-independent central polynomial, and derives explicit limiting eigenvalue distributions.

  • Motivation: Trefethen and Chapman’s Scottish flag spectrum was previously observed numerically on two orthogonal lines, and this paper establishes that structure rigorously.The Scottish flag matrix is identified with the family at coupling phase φ = π/2.
  • Scope: The analysis does not pursue irrational frequencies or the corresponding infinite-dimensional discrete operator, focusing instead on exact finite-dimensional identities and even periodic matrices.General rational frequencies and higher-dimensional extensions are left as future directions.
  • Two-line confinement: For even N and ϑ ∈ 2πZ/N, every eigenvalue lies on e^{iφ/2}R or e^{i(φ/2+π/2)}R, with respective radius bounds 2|cos(φ/2)| and 2|sin(φ/2)|.The restriction on ϑ is necessary; for N = 5 and (φ,ϑ) = (π/2,0), a root lies off the saltire.
  • Chambers’ formula: A Chambers formula separates the characteristic polynomial into a (τ, ϑ)-dependent constant term and a monic polynomial Q_N,φ independent of boundary twist and potential phase.This isolates the central polynomial governing the phase-independent part of the spectral problem.
  • Two-line confinement: For every N, a suitable phase and boundary twist produce a fibre whose spectrum lies on e^{iφ/2}R, while every zero of Q_N,φ lies on the two perpendicular lines.The Scottish flag consequence is spec(B_N) ⊂ X_{π/4} for N ≥ 3.
  • Limiting distribution: For even N, the eigenvalue measures converge to explicit densities on the two perpendicular lines, with supports determined by the corresponding radii.The limiting measures are identified through weak convergence and elliptic-integral density formulas.

2. Two-line spectra

A basis change puts A_N(φ, ϑ) into a cyclic zero-diagonal tridiagonal form whose edge products control its spectrum. These products share a phase, while their real factors determine whether one or two perpendicular spectral lines occur.

  • Structural reduction: A_N(φ, ϑ) is generally non-self-adjoint and non-normal, but a suitable orthonormal basis makes it cyclic tridiagonal with zero diagonal.Only nearest-neighbour entries remain in this representation.
  • Change of basis: For even N, the basis is constructed with a parity-dependent periodic exponential, with the even case using exp(πij²/N)ω^{jk}.The construction is well defined because the relevant exponent factor is N-periodic.
  • Change of basis: Multiplication by the shifted cosine and shifting the exponential produce the two nearest-neighbour coefficients αk,N and βk,N.Together they yield the zero-diagonal recurrence representation of A_N(φ, ϑ).
  • Edge products: The edge product ρj is the product of opposite off-diagonal entries across each edge, and all ρj lie on e^{iφ}R.The common phase alone does not force the spectrum onto two lines; the ordering and reflection reduction also matter.

2.2. A tridiagonal matrix calculation.

The tridiagonal calculation reduces characteristic polynomials to edge products and then splits a zero-diagonal matrix into odd and even sectors. A single sign change in the real edge-product factors is precisely what permits the spectrum to occupy two perpendicular lines.

  • Continuant recurrence: The continuant recurrence pk+1 = zpk − ρkpk−1 shows that the characteristic polynomial depends on each edge only through ρk = ukℓk.This makes opposite-entry products the relevant tridiagonal data rather than the individual entries.
  • Constant-sign products: If ρj = e^{2iθ}aj with aj > 0, diagonal similarity makes the matrix e^{iθ} times a real symmetric tridiagonal matrix, so its eigenvalues lie on e^{iθ}R and are simple.The real symmetric matrix has nonzero off-diagonal entries, which gives simplicity.
  • Odd–even decomposition: Odd–even permutation produces a block form with bidiagonal B, yielding det(zI − T) = z^{n−2s}det(z²I_s − B^T B).The companion identity for BB^T shows that the nonzero spectra of B^T B and BB^T coincide.
  • Odd–even decomposition: Every consecutive product wjwj+1 occurs in exactly one of B^T B and BB^T, enabling the exceptional sign-change product to be isolated in one Gram matrix.The other Gram matrix can then be rescaled to a real symmetric matrix.
  • Two-line conclusion: If the real factors cj change sign at most once on each nonzero block, then spec(T) lies in e^{iα}R ∪ e^{i(α+π/2)}R.Constant sign gives one line; one sign change contributes the perpendicular line through the odd–even reduction.

2.3. Reflection symmetry when ϑ ∈2πZ/N.

When the potential phase is an allowed lattice phase, reflection symmetry folds the cyclic matrix into two invariant tridiagonal blocks. Their edge products inherit the required single-sign-change structure, yielding the two-line spectral inclusion and unitary equivalence across allowed phases.

  • Folded edge products: Folding preserves interior edge products, while fixed endpoints acquire factors of 2 in the symmetric block; these positive factors do not change product signs.The antisymmetric block loses the endpoint edges because it contains neither fixed vertex.
  • Sign pattern: For ϑ = 0, the relevant real edge-product factors decrease strictly and therefore change sign at most once.This verifies the hypothesis needed for the two-line tridiagonal theorem in both reflection blocks.
  • Allowed phases: All matrices with ϑ ∈ 2πZ/N are unitarily equivalent, so the two-line spectral inclusion extends from ϑ = 0 to every allowed potential phase.The equivalence follows from cyclic translation of the diagonal potential while the hopping matrix commutes with the translation.
  • Line bounds: The eigenvalue equation separates the two line directions through Hermitian quadratic forms a and b, with real and imaginary parts weighted by cos(φ/2) and sin(φ/2).This supplies the corresponding bounds along the two spectral lines.

2.4. Arbitrary potential phase and boundary twist.

The determinant expansion isolates all dependence on the potential phase and boundary twist into explicit constant-term corrections. The remaining monic polynomial Q_N,φ is independent of both, while periodic boundary conditions remove the twist correction.

  • Phase dependence: Cyclic translation restricts the Laurent polynomial’s possible powers to ξ^{-N}, 1, and ξ^N.Only powers divisible by N survive the translation invariance, and the Laurent range is bounded by ±N.
  • Phase dependence: The ξ^{±N} determinant terms contribute −2^{1−N}cos(Nϑ), capturing the entire potential-phase dependence.The coefficients arise by selecting the corresponding hopping term from every diagonal factor in the determinant expansion.
  • Boundary twist: The boundary twist contributes −2^{1−N}e^{iNφ}cos κ through the two oriented cycle terms, while terms using both endpoint entries are twist-independent.Here τ = e^{iκ}, and the endpoint-product contribution ττ^{-1} cancels.
  • Chambers formula: Collecting all remaining terms defines a monic polynomial Q_N,φ(z) independent of both ϑ and the boundary twist τ.This decomposition is the generalized Chambers formula used throughout the paper.
  • Special fibres: For periodic boundary conditions τ = 1, the twist correction vanishes, whereas arbitrary potential phases can change the constant term and move eigenvalues off the distinguished lines.When ϑ lies in 2πZ/N, cos(Nϑ) = 1 and the periodic fibres are isospectral; for general ϑ this need not hold.

2.5. Phase opening of the central polynomial.

The phase-independent central polynomial can be realized as the characteristic polynomial of a zero-diagonal path after opening one vanishing edge. A sign-change argument then confines its zeros to two perpendicular lines, with corresponding real-root structure after a quadratic change of variables.

  • Path realization: The central polynomial is the characteristic polynomial of a zero-diagonal tridiagonal path with specified consecutive edge products.A phase choice makes one edge product vanish, opening the cyclic matrix into a path.
  • Path realization: The opened path determinant matches the central polynomial because the only omitted cyclic terms are the two oriented full-cycle permutations.The remaining fixed-point and transposition terms are exactly the path determinant expansion.
  • Zero localization: The edge-product signs change at most once, allowing the path sign-change theorem to be applied with α = φ/2.The sampled sine arguments occupy an interval shorter than π, so the relevant sine changes sign at most once after zero terms are removed.
  • Zero localization: Every zero of Q_N,φ lies on e^{iφ/2}R ∪ e^{i(φ/2+π/2)}R.Equivalently, t = e^{-iφ}z^2 is real for every zero.
  • Polynomial structure: The zeros of the transformed monic polynomial R_N,φ are real, establishing the stated real-coefficient representation.The quadratic transformation sends the two-line containment to the real axis.

3.1. Reversal identity for products ρj.

A reversal symmetry organizes the edge products of the relevant tridiagonal matrices into sign-controlled halves. This permits reflection decomposition and yields simple, symmetric spectra on the two-line set, with explicitly described zero Jordan blocks.

  • Spectral structure: A real symmetric tridiagonal block with nonzero off-diagonal entries has simple spectrum of the form {0, ±ν_1, …, ±ν_m}.The resulting nonzero eigenvalues are paired symmetrically around zero.
  • Reversal symmetry: The reversal identity makes the edge-product sequence positive on its first half and negative on its second, with at most one sign change after removing a possible zero.For odd M, the middle term is forced to vanish.
  • Reflection decomposition: Reflection symmetry decomposes the cyclic matrix into two tridiagonal path blocks whose edge-product lists inherit the same one-sign-change property.Passing to sublists and multiplying endpoint factors by 2 does not create additional sign changes.
  • Spectral localization: The sign-change theorem applies separately to both blocks, placing their spectra on the corresponding line set X_θ.The construction reduces the cyclic problem to path matrices with controlled edge-product signs.
  • Spectral structure: At φ = π/2, the factorization has distinct positive parameters, and z^4 = −η_j,N places the nonzero roots on the cross.The zero Jordan structure is also identified in the two parity cases.

3.2. The periodic φ = π/2 matrix.

For the periodic φ = π/2 family, squaring separates even and odd index subspaces into balanced cyclic tridiagonal blocks. Their edge products have exactly one negative member each, enabling strip bounds and exact exclusion from the diagonal lines away from aligned phases.

  • Block reduction: Squaring the φ = π/2 matrix preserves parity and produces two M × M cyclic tridiagonal blocks with real diagonal.The original matrix exchanges even and odd index subspaces, while its square preserves them.
  • Block reduction: Both squared blocks satisfy the balancing condition because their forward and backward cycle products contain all α_k,N and β_k,N factors once.This permits the almost-Hermitian localization lemma to be applied to each block.
  • Edge signs: Each block has exactly one negative edge product, since the cosine sequence changes sign twice and the two changes fall into opposite parity classes.All remaining edge products are positive.
  • Spectral exclusion: For nonaligned phases, spec A_N(π/2,ϑ) is disjoint from X_π/4.The proof combines the central-polynomial factorization with a strictly positive real term that cannot cancel a purely imaginary term.
  • Aligned phases: At aligned phases, the periodic matrix has spectrum on the cross, with every nonzero eigenvalue simple and quarter-turn symmetry.The central polynomial factorization also expresses each nonzero quartet through a factor z^4 + τ with τ > 0.

3.3. The original Scottish flag matrix.

The original Scottish flag matrix inherits the two-line spectral geometry and explicit quartic factorizations from the periodic φ = π/2 analysis. Across all dimensions, its nonzero eigenvalues are simple, while even dimensions require a distinction in the zero Jordan structure.

  • Dimension-dependent factorization: For N = 4m, 4m + 1, 4m + 2, and 4m + 3, the characteristic polynomial has the corresponding positive-parameter factorizations.The odd-dimensional cases use pairwise distinct positive parameters.
  • Spectral conclusion: The Scottish flag spectrum lies on the two diagonal lines, and every nonzero eigenvalue is simple for every N ≥ 3.This establishes the Scottish flag phenomenon in all dimensions.
  • Transfer from periodic case: For even N, conjugation and spectral scaling transfer the periodic φ = π/2 result to the original Scottish flag matrix.The two-line containment and factorizations follow from the periodic theorem.
  • Zero Jordan structure: For N ≡ 2 (mod 4), the zero eigenvalue has two size-one Jordan blocks, unlike the single larger zero block in the periodic comparison.The characteristic and geometric multiplicities are both two in this case.

4. Limiting eigenvalue distribution

The section proves limiting eigenvalue distributions by approximating slowly varying tridiagonal matrices with local coefficient models and controlling boundary and finite-rank effects. Applied to the even-dimensional almost Mathieu matrices, this yields two explicit limiting densities with stated masses and supports.

  • Trace asymptotics: Local walk expansions show that normalized traces of powers converge to integrals determined by the slowly varying coefficient profiles.Interior walks can be replaced by local coefficients, while boundary contributions are negligible in the normalized trace.
  • Trace asymptotics: O(n^-1) coefficient approximation and O(n^-1) boundary contributions give quantitative polynomial-moment estimates.The estimate assumes C1 coefficient profiles and O(n^-1) uniform coefficient errors.
  • Stability: Uniformly bounded rank perturbations change normalized traces by O(n^-1), so bounded exceptional rows and zero modes do not alter the limiting measure.The argument requires the spectra to remain in a common bounded interval.
  • Almost Mathieu application: For even N, reflection and odd–even decompositions reduce the spectrum, apart from O(1) zero eigenvalues, to positive semidefinite Jacobi matrices with limiting coefficient profiles.The resulting sectors contribute the two integrals that identify the limiting densities.
  • Almost Mathieu application: The limiting measures have densities supported on [−2aφ,+, 2aφ,+] and [−2aφ,−, 2aφ,−], with masses specified by the two sectors.The densities vanish outside their corresponding intervals, and the proof identifies the measures with the formulas in the main theorem.
  • Almost Mathieu application: At φ = π/2, the same two-integral formula holds along the odd subsequence, completing the limiting-distribution proof for all N.Combining odd and even subsequences establishes the stated polynomial-moment estimate.

Appendix A. Proof of the determinant formula under the reversal identity

The appendix uses a reversal symmetry to reduce the determinant problem to real symmetric tridiagonal matrices with nonzero off-diagonal entries. Their spectra are real, simple, and symmetric about zero, producing the required determinant structure.

  • Two-step reduction: The reversal identity selects an odd-dimensional two-step matrix whose entries share a common phase times real coefficients.The parity of the dimension determines whether BT B or BBT avoids the unique sign change.
  • Two-step reduction: After removing the common phase, the resulting matrix is real symmetric tridiagonal with every off-diagonal entry nonzero.This structure follows from products of consecutive nonzero weights.
  • Spectral symmetry: Its spectrum is real and simple, and reversal pairs each eigenvalue with its negative.Because the dimension is odd, zero is also an eigenvalue.
  • Zero structure: The determinant identities extend to z = 0 by polynomial continuation, with zero algebraic multiplicities determined by the tridiagonal recurrence.The zero Jordan blocks in the two determinant cases are J1(0) and J3(0).

Appendix B. Jordan blocks at zero when φ = π/2

For φ = π/2, the appendix determines the Jordan structure at zero by analyzing reflection sectors and their zero-product edges. Odd-dimensional symmetric tridiagonal blocks supply the zero eigenvectors, while triangular coupling can produce a J2(0) block.

  • N = 4m: For N = 4m, the two reflection sectors have odd dimensions and yield Jordan blocks J3(0) and J1(0) at zero.Each sector has a one-dimensional kernel, and the determinant factors distinguish the cubic and linear zero factors.
  • Jordan coupling: The endpoint coordinates of the block zero eigenvectors are nonzero, ensuring that the coupling controls the geometric multiplicity.Block triangularity gives algebraic multiplicity two, while the coupling equation determines the nullspace dimension.
  • N = 4m + 2: For N = 4m + 2, a zero-product edge makes one reflection sector block triangular rather than a direct sum.The singular sector has two odd diagonal path blocks, while the other sector has even diagonal blocks and is invertible at zero.
  • Jordan coupling: Coupling two odd-dimensional zero-diagonal symmetric tridiagonal blocks through a nonzero rank-one edge gives zero algebraic multiplicity two and geometric multiplicity one.The resulting Jordan form is J2(0).

Appendix C. The zero eigenspace when τ = −1

The appendix analyzes the zero eigenspace when τ = −1 by transforming the matrix into two coupled odd-dimensional tridiagonal blocks. Each block has a simple zero eigenvalue, and the coupling determines whether zero is semisimple.

  • Zero eigenspace: The zero eigenspace has dimension two, so the two zero eigenvalues produce two one-dimensional Jordan blocks.The algebraic multiplicity is also two.
  • Block decomposition: A diagonal phase transformation and reordered basis produce a block triangular matrix with two odd-dimensional zero-diagonal tridiagonal blocks.The off-diagonal products in the two blocks have opposite imaginary signs.
  • Block decomposition: Each diagonal block is diagonally similar to a phase multiple of a real symmetric tridiagonal matrix with nonzero off-diagonal entries.Consequently, each block has a simple zero eigenvalue.
  • Zero eigenspace: The coupling solvability condition creates a second independent zero eigenvector rather than a generalized eigenvector.The resulting zero Jordan form is J1(0) ⊕ J1(0).
Loading 2608.30191v1…