Bethe Eigenlines in Integrable Systems

• Bethe eigenlines are one-dimensional spectral objects uniquely defined by Bethe Ansatz data, representing invariant lines spanned by on-shell Bethe vectors.
• They emerge in various integrable models—including spin chains, Gaudin models, and Bethe–Salpeter frameworks—facilitating analysis of eigenvalue trajectories and regularization methods.
• Studying Bethe eigenlines clarifies physical realizability, operadic coverings, and numerical schemes, thereby deepening our understanding of spectral theory in integrable systems.

Searching arXiv for recent and directly relevant papers on Bethe eigenlines and closely related usages. Bethe eigenlines are one-dimensional common eigenspaces arising in Bethe-ansatz-integrable spectral problems. In the strictest sense represented in the Gaudin-model literature, they are the joint eigenspaces of commutative Gaudin algebras acting on tensor products or multiplicity spaces, and they vary in families over real moduli spaces of marked curves (Kamnitzer et al., 17 Jul 2025). In the algebraic Bethe ansatz for spin chains, the same term is naturally attached to the line spanned by an on-shell Bethe vector, since the vector is defined only up to normalization and depends on Bethe-root data rather than on a canonical scale (Nepomechie et al., 2013). In numerical and functional treatments of Bethe–Salpeter equations, the closest analogue is a structured eigenline or paired eigendirection of a parameter-dependent matrix problem, often tracked through eigenvalue curves or preserved by structure-aware eigensolvers (Blank et al., 2010, Shao et al., 2015). Across these settings, the common theme is that Bethe data do not merely produce eigenvalues: they determine distinguished one-dimensional spectral objects whose geometry, regularization, monodromy, and numerical computation are central to the theory.

1. Bethe eigenlines in algebraic Bethe ansatz for spin chains

In the isotropic periodic spin-$\frac12$ Heisenberg chain, the standard ABA begins from the Hamiltonian

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$

with monodromy matrix

$$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$

transfer matrix

$t(\lambda)=A(\lambda)+D(\lambda),$

and pseudovacuum

$|0\rangle=\binom10^{\otimes N}.$

For distinct Bethe roots $\lambda_1,\dots,\lambda_M$, the standard Bethe vector is

$$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$

with roots constrained by the Bethe equations

$\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$

The transfer-matrix action has the standard wanted term plus unwanted terms, and the Bethe equations are exactly the conditions that remove all unwanted contributions (Nepomechie et al., 2013).

For generic nonsingular roots, this gives the usual correspondence: a solution of the Bethe equations determines a Bethe vector and therefore a common eigenline of the commuting transfer matrices and of the Hamiltonian (Nepomechie et al., 2013). The paper on singular solutions makes explicit that the eigenline, not the unnormalized vector, is the invariant object. This becomes decisive when the vector formula degenerates.

A closely related situation appears for the open XXX chain with general boundaries. There the transfer matrix contains both $\mathscr B(u)$ and $\mathscr C(u)$, so $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$0 symmetry is broken and the standard fixed-magnetization ABA no longer closes. The generalized construction replaces the usual Bethe vector by

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$1

where the modified creation operator is

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$2

with

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$3

When the corresponding Bethe equations hold, the off-shell transfer-matrix action reduces to a wanted term, so the vector becomes an eigenvector. Since it is meaningful only up to an overall factor, the associated object is again an eigenline (Belliard et al., 2013).

This suggests a broad ABA principle: Bethe equations parametrize eigenlines, while explicit Bethe vectors are representatives of those lines whose normalization is secondary. A plausible implication is that “Bethe eigenline” is often the conceptually correct formulation whenever Bethe vectors are only canonically defined projectively.

2. Singular solutions and the failure of naive root-to-eigenline correspondence

The clearest breakdown of the naive Bethe-solution/eigenvector dictionary occurs for singular solutions of the XXX chain containing

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$4

For $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$5, this pair is an exact formal solution of the Bethe equations, but it is singular in two ways. First, the energy formula

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$6

diverges because $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$7 at $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$8. Second, the Bethe vector becomes ill-defined: in the conventions of the paper, $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$9 is finite while $$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$0 is singular, producing components of indeterminate $$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$1 type (Nepomechie et al., 2013).

The central conceptual point is that a formal solution of the Bethe equations containing $$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$2 does not automatically define a genuine transfer-matrix eigenvector, or equivalently a genuine Bethe eigenline (Nepomechie et al., 2013). Root data alone are insufficient at the singular locus.

A naive regularization,

$$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$3

recovers the correct energy limit $$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$4 for the two-root state, but not the correct eigenvector. The reason is that the unwanted coefficients in the off-shell action have the wrong asymptotics: $$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$5 whereas one needs

$$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$6

The cure is the higher-order regularization

$$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$7

with $$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$8 fixed by generalized consistency conditions (Nepomechie et al., 2013).

For even $$T_a(\lambda)=L_{Na}(\lambda)\cdots L_{1a}(\lambda) = \begin{pmatrix} A(\lambda) & B(\lambda)\ C(\lambda) & D(\lambda) \end{pmatrix},$$9, both asymptotic conditions are satisfied by

$t(\lambda)=A(\lambda)+D(\lambda),$0

For odd $t(\lambda)=A(\lambda)+D(\lambda),$1, they cannot be satisfied simultaneously, so the singular two-string is unphysical (Nepomechie et al., 2013). In the general singular configuration

$t(\lambda)=A(\lambda)+D(\lambda),$2

compatibility reduces to the practical criterion

$t(\lambda)=A(\lambda)+D(\lambda),$3

A singular solution is physical iff this condition holds, equivalently iff one can choose $t(\lambda)=A(\lambda)+D(\lambda),$4 so that the regularized Bethe vector becomes a genuine eigenvector (Nepomechie et al., 2013).

This criterion sharply refines completeness questions. For $t(\lambda)=A(\lambda)+D(\lambda),$5, $t(\lambda)=A(\lambda)+D(\lambda),$6, there are $t(\lambda)=A(\lambda)+D(\lambda),$7 singular solutions of the formal Bethe equations, but only $t(\lambda)=A(\lambda)+D(\lambda),$8 is physical. For $t(\lambda)=A(\lambda)+D(\lambda),$9, $|0\rangle=\binom10^{\otimes N}.$0, there are $|0\rangle=\binom10^{\otimes N}.$1 singular solutions, but only $|0\rangle=\binom10^{\otimes N}.$2 are physical (Nepomechie et al., 2013). The practical conclusion is that the space of admissible Bethe eigenlines is strictly smaller than the space of formal Bethe-root solutions once singular roots are allowed.

3. Geometric and categorical Bethe eigenlines in Gaudin models

In the Gaudin-model setting, Bethe eigenlines are treated directly as geometric objects. For dominant weights $|0\rangle=\binom10^{\otimes N}.$3, write

$|0\rangle=\binom10^{\otimes N}.$4

For regular $|0\rangle=\binom10^{\otimes N}.$5, the inhomogeneous Gaudin algebra

$|0\rangle=\binom10^{\otimes N}.$6

acts on $|0\rangle=\binom10^{\otimes N}.$7, preserving weight spaces, and the paper denotes by

$|0\rangle=\binom10^{\otimes N}.$8

the set of eigenlines of $|0\rangle=\binom10^{\otimes N}.$9 on $\lambda_1,\dots,\lambda_M$0 (Kamnitzer et al., 17 Jul 2025). In the homogeneous case over $\lambda_1,\dots,\lambda_M$1, one similarly considers eigenlines of $\lambda_1,\dots,\lambda_M$2 on multiplicity spaces

$\lambda_1,\dots,\lambda_M$3

These eigenlines vary over real cactus flower moduli spaces. The families

$\lambda_1,\dots,\lambda_M$4

form a $\lambda_1,\dots,\lambda_M$5-coloured operadic covering of $\lambda_1,\dots,\lambda_M$6 (Kamnitzer et al., 17 Jul 2025). The key structural reason is factorization of Gaudin algebras on boundary strata, for example

$\lambda_1,\dots,\lambda_M$7

together with analogous formulas for mixed and homogeneous boundary gluings (Kamnitzer et al., 17 Jul 2025).

The resulting conceptual statement is stronger than a mere parametrization of eigenvectors. Isomorphism classes of operadic coverings of $\lambda_1,\dots,\lambda_M$8 are naturally one-to-one with equivalence classes of concrete coboundary monoidal categories satisfying certain semisimplicity and finiteness conditions, and the covering defined by Bethe eigenlines reconstructs the concrete coboundary category of Kashiwara $\lambda_1,\dots,\lambda_M$9-crystals (Kamnitzer et al., 17 Jul 2025). The paper proves an equivalence

$$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$0

sending simple objects $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$1 to crystals $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$2 (Kamnitzer et al., 17 Jul 2025).

This yields a monodromy interpretation. Over $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$3, Bethe eigenlines carry an action of the virtual cactus group $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$4, and there are bijections

$$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$5

compatible with the $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$6-actions (Kamnitzer et al., 17 Jul 2025). Over split and compact real loci of the trigonometric deformation, the monodromy is described by the mirabolic cactus group $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$7 and the extended affine cactus group $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$8, acting respectively by elementary transpositions or cyclic rotations together with the usual cactus-group action on tensor-product crystals (Kamnitzer et al., 17 Jul 2025).

In this geometric regime, Bethe eigenlines are literally points in topological covering spaces. Their importance lies not only in encoding spectra of Gaudin Hamiltonians, but in carrying the operadic and monodromic data from which crystal tensor combinatorics can be recovered. This suggests that “Bethe eigenline” is, in this context, a genuinely geometric notion rather than a shorthand for an eigenspace.

4. Bethe eigenlines, Jordan degeneration, and generalized eigenvectors

A different use of the eigenline perspective appears in non-diagonalizable limits of integrable spin chains. In the eclectic spin chain, obtained as a strong-twist limit of a finitely twisted $$|\lambda_1,\dots,\lambda_M\rangle = B(\lambda_1)\cdots B(\lambda_M)|0\rangle,$$9 XXX chain,

$\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$0

the limiting Hamiltonian becomes defective and develops Jordan blocks (García et al., 2021).

For finite twist, one has ordinary Bethe eigenvectors from nested coordinate Bethe ansatz. In the strong-twist limit, several of these eigenvectors collapse onto the same limiting “locked state”

$\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$1

so distinct finite-$\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$2 Bethe eigenlines coalesce (García et al., 2021). The missing spectral information is not lost; it is encoded in suitable singular linear combinations of the colliding Bethe vectors, which produce generalized eigenvectors and full Jordan chains.

The general mechanism is formulated for a family $\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$3 of diagonalizable matrices approaching a defective limit $\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$4. If eigenvectors $\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$5 coalesce, then limits of singular combinations such as

$\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$6

recover rank-2 generalized eigenvectors, and higher-rank constructions follow recursively (García et al., 2021). The paper proves that, when the limiting Jordan blocks are distinguishable, these limits reconstruct the full generalized eigenspace.

This is relevant to Bethe eigenlines because it shows that a family of ordinary Bethe eigenlines can degenerate into non-semisimple spectral data. The limit is not simply a set of surviving eigenlines; it carries extra extension data detectable only through the singular asymptotics of the Bethe vectors. A plausible implication is that the projective viewpoint is especially natural near exceptional points: the collision of eigenlines is the primary event, and Jordan structure emerges from how these lines approach one another.

5. Bethe–Salpeter eigenlines as structured eigendirections

In Bethe–Salpeter theory, the phrase “Bethe eigenline” is not standard, but several papers treat closely related objects: one-dimensional eigendirections, eigenvalue branches, and paired invariant subspaces of the Bethe–Salpeter Hamiltonian.

After discretization, the homogeneous Bethe–Salpeter equation takes the matrix form

$\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$7

and physical masses satisfy

$\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$8

Each $\left(\frac{\lambda_k+\frac{i}{2}}{\lambda_k-\frac{i}{2}}\right)^N = \prod_{\substack{j=1\j\neq k}^{M} \frac{\lambda_k-\lambda_j+i}{\lambda_k-\lambda_j-i}, \qquad k=1,\dots,M.$9 is an eigenvalue trajectory, and its crossing with $\mathscr B(u)$0 marks a bound state (Blank et al., 2010). In this sense, a “Bethe eigenline” is the $\mathscr B(u)$1-dependent eigendirection $\mathscr B(u)$2 associated with an eigenvalue branch.

The inhomogeneous equation,

$\mathscr B(u)$3

detects the same spectrum through poles of the resolvent (Blank et al., 2010). Poles appear precisely when some eigenline satisfies $\mathscr B(u)$4. Thus eigenline crossings and poles are two views of the same spectral event (Blank et al., 2010).

For the finite-dimensional Bethe–Salpeter Hamiltonian

$\mathscr B(u)$5

the spectrum has the symmetry

$\mathscr B(u)$6

and under the definiteness condition

$\mathscr B(u)$7

all eigenvalues are real and occur in $\mathscr B(u)$8 pairs (Shao et al., 2015). The corresponding right and left eigenvector matrices can be written

$\mathscr B(u)$9

with

$\mathscr C(u)$0

Thus the positive eigendirections are not isolated objects; each determines companion negative and dual directions through the BSE symmetry (Shao et al., 2015).

Later work recasts this geometry in explicit indefinite-inner-product language. For a definite Bethe–Salpeter Hamiltonian

$\mathscr C(u)$1

one has a structured spectral decomposition

$\mathscr C(u)$2

with

$\mathscr C(u)$3

If

$\mathscr C(u)$4

is an eigenvector for $\mathscr C(u)$5, then

$\mathscr C(u)$6

is the corresponding eigenvector for $\mathscr C(u)$7 (Shan et al., 9 Mar 2026). The relevant eigenspaces lie in

$\mathscr C(u)$8

meaning they admit bases that are simultaneously of Bethe–Salpeter block form and $\mathscr C(u)$9-orthonormal (Shan et al., 9 Mar 2026).

The algorithmic literature is built around preserving this structure. Structure-preserving dense and parallel algorithms reduce the BSE to a real Hamiltonian problem and then to a skew-symmetric eigenproblem (Shao et al., 2015, Penke et al., 2019). Structure-preserving $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$00QR and $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$01-Lanczos methods keep the $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$02-Hermitian form intact, so computed eigenvalues and eigenvectors retain the symmetry pairing of the original BSE matrix (Guo et al., 2018). A recent structure-preserving LOBPCG solver computes a few smallest positive eigenpairs while maintaining the $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$03-orthogonal paired geometry (Shan et al., 9 Mar 2026).

The precise terminology in these papers is “eigenvectors,” “paired eigendirections,” “invariant subspaces,” and “eigenvalue trajectories,” not “Bethe eigenlines.” Still, the notion is effectively present: physically relevant spectral data are organized into one-dimensional eigendirections or their paired counterparts, and preserving that organization is the central numerical concern.

6. Reconstruction, orthogonality, and computational access to Bethe eigenlines

Several papers address how Bethe-type spectral lines are reconstructed when standard ABA machinery is unavailable or when explicit orthogonality/completeness must be established by other means.

For integrable models solved by off-diagonal Bethe Ansatz, the inhomogeneous $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$04-$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$05 relation gives eigenvalues but not automatically eigenvectors. The retrieval method constructs an orthogonal basis of the Hilbert space from commuting monodromy entries and determines an eigenstate by its overlaps with this basis. In the XXZ spin torus model, the candidate Bethe state is

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$06

with a reconstructed reference state

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$07

For the open XXX chain, the reconstructed Bethe states are

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$08

In both cases the state is determined uniquely up to an overall scalar factor by its overlaps with a complete orthogonal basis, so what is reconstructed is naturally an eigenline (Zhang et al., 2014).

Orthogonality of Bethe eigenfunctions is the main result in a different continuum setting: the Laplacian on the hyperoctahedral Weyl alcove with Robin boundary conditions. The Bethe Ansatz eigenfunctions

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$09

form an orthogonal basis of $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$10 for $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$11 (Diejen et al., 2016). These functions arise as the continuum limit of an orthogonal basis of algebraic Bethe Ansatz eigenfunctions for a finite $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$12-boson system. In the finite model, the $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$13-particle Bethe states

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$14

diagonalize the commuting transfer operator and Hamiltonian, and distinct spectral points give orthogonal states (Diejen et al., 2016). This is precisely the setting in which each admissible Bethe solution determines a distinguished spectral line.

Recent computational work accesses Bethe-root-parameterized eigenstates directly. In the XXZ chain, Bethe states are used as the variational ansatz in VQE, with Bethe roots treated as variational parameters. The coordinate Bethe state

$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$15

is optimized by minimizing either energy or variance, thereby estimating Bethe roots corresponding to actual eigenstates (Raveh et al., 2024). The paper explicitly notes that this approach is not limited to real Bethe roots and that, when optimization converges to an eigenstate, the resulting roots are automatically physical (Raveh et al., 2024). It does not formulate a geometric eigenline theory, but it gives a practical parameter-space route to projective Bethe states.

A different computational direction appears in the 2026 study of three periodic spin-$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$16 chains solved by coordinate and nested Bethe Ansatz. For the scalar model $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$17, the eigenstate is parametrized by momenta $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$18; for the nested model $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$19, by $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$20 and auxiliary roots $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$21; for the model $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$22, by first-level roots $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$23 together with an eigenline of a non-$$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$24-symmetric free-fermionic auxiliary transfer matrix (Pozsgay et al., 31 Mar 2026). The last case is especially notable because the auxiliary problem is free-fermionic and lacks $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$25-invariance, so the spectral line at the nesting level is not encoded by a conventional auxiliary-magnon number sector (Pozsgay et al., 31 Mar 2026).

These examples show that the construction of Bethe eigenlines is not confined to standard ABA. They can be reconstructed from functional relations, proved orthogonal through finite-volume limits, or searched variationally through root-parameterized trial states.

7. Conceptual synthesis and common misconceptions

A recurring misconception is that a solution of Bethe equations automatically determines a physical eigenstate. The singular-root analysis of the XXX chain is the clearest counterexample: formal Bethe solutions may fail to define finite, nonzero transfer-matrix eigenvectors, and additional generalized consistency conditions are required (Nepomechie et al., 2013). Likewise, in non-diagonalizable limits, ordinary Bethe eigenvectors may collide, and their limits alone do not exhaust the spectral data; generalized eigenvectors must be extracted from singular combinations (García et al., 2021).

Another misconception is that Bethe eigenlines are always isolated one-dimensional objects with no extra structure. In Gaudin models they form coverings with operadic gluing and nontrivial monodromy, recovering coboundary tensor-category data (Kamnitzer et al., 17 Jul 2025). In Bethe–Salpeter problems they typically appear in paired $$H=\frac14 \sum_{n=1}^{N}\left(\vec{\sigma}_n\cdot \vec{\sigma}_{n+1}-1\right), \qquad \vec{\sigma}_{N+1}=\vec{\sigma}_1,$$26-symmetric families and are naturally embedded in structured invariant subspaces rather than treated as independent lines (Shao et al., 2015, Shan et al., 9 Mar 2026, Guo et al., 2018).

A third misconception is that the term applies uniformly across all “Bethe” contexts. The data suggest a more careful picture. In the strict Gaudin and ABA literature, “Bethe eigenline” refers quite directly to the line spanned by an on-shell Bethe vector or to the common eigenline of a commutative Bethe algebra (Nepomechie et al., 2013, Kamnitzer et al., 17 Jul 2025). In Bethe–Salpeter literature, the same wording is best understood as an interpretive bridge rather than established terminology: the actual objects are structured eigendirections, eigenvalue branches, and paired invariant subspaces (Blank et al., 2010, Shao et al., 2015).

Taken together, the literature supports a robust definition. A Bethe eigenline is the one-dimensional spectral object singled out by Bethe data—roots, auxiliary roots, or more general commuting-algebra eigenconditions—once normalization is quotiented out. Depending on context, that line may require regularization, may sit inside a monodromic family over moduli space, may degenerate into Jordan structure, or may come paired with a symmetry-related companion. The unifying point is that Bethe methods do not merely compute eigenvalues: they organize spectral theory around these line-level objects.