Non-Hermitian Multi-Band Twister Models

• Non-Hermitian multi-band twister models are lattice Hamiltonians whose complex spectra form twisted, graph-like structures under open boundary conditions, unifying non-Bloch momentum and exceptional point phenomena.
• They utilize methods such as generalized Brillouin zones, Wiener–Hopf factorization, and permutation-group monodromy to classify spectral loops, branch counts, and braid structures.
• Experimental realizations in photonic, acoustic, and superconducting systems validate these models, offering practical insights into non-Hermitian band topology and edge mode localization.

Non-Hermitian multi-band twister models, an Editor’s term, can be understood as multi-band, multi-component non-Hermitian lattice Hamiltonians whose complex spectra exhibit twisted connectivity: planar spectral graphs under open boundary conditions (OBC), composite loops that switch band identity at exceptional points (EPs), or braid-like band trajectories whose closures realize knots and links. Across recent work, this class is described through non-Bloch momentum, generalized Brillouin zones (GBZs), spectral graphs, permutation-group monodromy, Wiener–Hopf factorization (WHF), and real-space twisting constructions, with representative realizations ranging from multi-band SSH-type chains to programmable acoustic, photonic, and superconducting-quantum platforms (Tai et al., 2022, Nehra et al., 2022, Ng et al., 29 Apr 2026).

1. Spectral-graph and non-Bloch foundations

A central formulation starts from a 1D non-Hermitian lattice under OBC, where momentum becomes complex, $k\to k+i\kappa$, with $\kappa=\operatorname{Im}k$ the inverse skin depth, and the bulk dispersion is encoded by the bivariate characteristic polynomial

$P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$

For each root $z_i(E)$, one defines $\kappa_i(E)=-\log|z_i(E)|$ as a surface over the complex-energy plane. An OBC eigenenergy $\bar E$ occurs where two or more roots satisfy the equal-decay condition

$$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$

Pairwise intersections of the $\kappa$-surfaces generate edges in the complex $E$-plane, and higher-order intersections generate vertices; the union is the OBC spectral graph. In this sense, the OBC spectrum is a planar graph embedded in $\mathbb C_E$, with line, loop, star, flower, and insect-like connectivities (Tai et al., 2022).

This spectral-graph viewpoint is distinct from conventional non-Hermitian band topology. Point-gap topology classifies the winding of $\kappa=\operatorname{Im}k$0 around a base energy $\kappa=\operatorname{Im}k$1, line-gap topology classifies spectra avoiding a line in the complex plane, and EP topology classifies eigenvalue or eigenvector winding around branch points. Spectral-graph topology instead classifies the connectivity of the entire OBC spectrum as a planar graph: the number of branches, loops, branching points, and adjacency relations. Two spectra may therefore share the same point-gap winding while possessing radically different graph topology (Tai et al., 2022).

The non-Bloch formulation supplies the corresponding momentum-space description. For a finite-range 1D model with $\kappa=\operatorname{Im}k$2 internal degrees of freedom and hopping range $\kappa=\operatorname{Im}k$3, one defines

$\kappa=\operatorname{Im}k$4

and solves

$\kappa=\operatorname{Im}k$5

If the $\kappa=\operatorname{Im}k$6 roots are ordered by magnitude,

$\kappa=\operatorname{Im}k$7

then the continuum OBC bands are characterized by

$\kappa=\operatorname{Im}k$8

The GBZ is the locus of the roots satisfying this equal-modulus condition, and the non-Bloch bulk bands are the eigenvalues of $\kappa=\operatorname{Im}k$9 evaluated on that contour (Yokomizo et al., 2020).

A further structural simplification comes from conformal equivalence in the energy plane. If $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$0 is holomorphic and invertible on the relevant domain, then $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$1 and the OBC spectrum transforms as $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$2. Spectra related by such conformal maps are graph-topologically equivalent, even when the microscopic Hamiltonians differ substantially. This equivalence organizes families of twisted spectra into canonical classes and explains emergent rotational or reflection symmetries of spectral graphs that need not be present in the underlying Hamiltonian (Tai et al., 2022).

2. Algebraic construction of multiband twister Hamiltonians

A large class of multiband twister models is organized by the characteristic polynomial

$P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$3

with $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$4, $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$5 polynomials in $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$6, $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$7, $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$8 polynomials in $P(E,z)=\det[H(z)-E\mathbb I]=0,\qquad z=e^{ik}.$9, and $z_i(E)$0 a tunable parameter. In the canonical mixed form

$z_i(E)$1

the degree $z_i(E)$2 equals the band number, while $z_i(E)$3 control how many branches emanate from the origin and how they reconnect at large $z_i(E)$4 (Tai et al., 2022).

The algebraic branch counts are explicit. Near $z_i(E)$5, the number of star branches is

$z_i(E)$6

At large $z_i(E)$7, the branch number is

$z_i(E)$8

If $z_i(E)$9, some branches must fuse into closed loops; if $\kappa_i(E)=-\log|z_i(E)|$0, additional outer branches emerge; and if $\kappa_i(E)=-\log|z_i(E)|$1, looping may still occur through relative rotational shifts between inner and outer structures. The loop number is determined by Euler’s formula,

$\kappa_i(E)=-\log|z_i(E)|$2

with $\kappa_i(E)=-\log|z_i(E)|$3, $\kappa_i(E)=-\log|z_i(E)|$4, and $\kappa_i(E)=-\log|z_i(E)|$5 the numbers of connected components, edges, and vertices of the spectral graph (Tai et al., 2022).

Representative canonical examples are summarized below.

Model class	Characteristic polynomial	Reported spectral graph
2-band triangular star	$\kappa_i(E)=-\log|z_i(E)|$6	triangle with three radial branches
4-band flower	$\kappa_i(E)=-\log|z_i(E)|$7	6-petal flower-like spectrum
4-band higher-order mixed model	$\kappa_i(E)=-\log|z_i(E)|$8	9 inner, 5 outer branches, 3 loops

The simplest separable subclass,

$\kappa_i(E)=-\log|z_i(E)|$9

already yields multiband star spectra. When $\bar E$0 is a monomial of degree $\bar E$1, the OBC spectrum forms a star with

$\bar E$2

equally spaced prongs. The two-band model

$\bar E$3

has $\bar E$4, is conformally equivalent to $\bar E$5, and produces a 6-prong star (Tai et al., 2022).

A separate constructive route starts from a Hermitian chiral-symmetric parent Hamiltonian

$\bar E$6

and adds a staggered imaginary potential $\bar E$7. The resulting non-Hermitian Bloch Hamiltonian

$\bar E$8

has band energies

$\bar E$9

This produces tunable Dirac-cone and exceptional-point reconnections in a multi-band setting and provides an analytically controlled way to deform a chiral parent spectrum into a non-Hermitian twisted spectrum (Lin et al., 2016).

3. Permutation topology, pseudo-Hermitian lines, and braided bands

A complementary description treats twisting as a problem of band permutation. In a 2D parameter space, multibands are called non-separable when there exists at least a one-dimensional periodic loop on which the initial state exchanges into a different state after a period. The induced permutation $$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$0 is classified by its conjugacy class $$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$1, so $$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$2 denotes one separable band and one exchanged pair, जबकि $$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$3 denotes a three-cycle. This identifies band twisting with monodromy on a multi-sheeted Riemann surface (Ryu et al., 2024).

Pseudo-Hermitian lines (PHLs) provide a mechanism for such twisting without requiring EPs on the loop itself. For a two-band Hamiltonian $$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$4, pseudo-Hermiticity is equivalent to

$$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$5

Real PHLs satisfy $$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$6, imaginary PHLs satisfy $$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$7, and their intersections are EPs. In a toroidal Brillouin zone, a non-contractible loop can intersect a non-contractible PHL an odd number of times, producing nontrivial band permutation even when there are no EPs in the 2D region. In the three-band model

$$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$8

with

$$P(\bar E,z_i)=P(\bar E,z_j)=0,\qquad |z_i(\bar E)|=|z_j(\bar E)|.$$9

the loop $\kappa$0 at fixed $\kappa$1 realizes the class $\kappa$2, and combining non-contractible $\kappa$3- and $\kappa$4-loops yields the class $\kappa$5 (Ryu et al., 2024).

Quadripartite non-Hermitian SSH chains realize the same idea in explicitly EP-centered form. Their complex-energy bands can form separate loops, two-band composite loops around finite-energy or zero-energy EPs, or a four-band composite loop encircling three EPs. The resulting topologies were compared with Möbius strips and Penrose triangles: two-band composite loops carry an effective $\kappa$6 periodicity, while the four-band composite loop carries an effective $\kappa$7 periodicity. Because adiabatic single-band evolution fails at the EPs, the topology is quantified by a nonadiabatic cyclic geometric phase built from the participating bands only (Nehra et al., 2022).

An explicit family of $\kappa$8-band twister Hamiltonians makes the braid structure fully constructive: $\kappa$9 where

$E$0

and

$E$1

For the pure twister $E$2, the eigenvalues are

$E$3

so the $E$4 strands cycle around the unit circle and the embedded trajectories form a torus link of type $E$5. Concrete 2-band and 4-band realizations support the Hopf link, unknot, unlink, Hopf chain, and Solomon’s knot, and their braid words can be reconstructed from pairwise winding data; the resulting braid closures were characterized by the Alexander and Jones polynomials (Ng et al., 29 Apr 2026).

4. Boundary localization, WHF/Amoeba theory, and real-space twisting

Because twisted spectra are inseparable from OBC physics, a major line of work studies the OBC spectral potential rather than the spectrum directly. For a finite-range 1D Bloch Hamiltonian

$E$6

define

$E$7

The Ronkin function is

$E$8

In single-band class A, the Amoeba theorem gives $E$9, but in multiband systems this simple optimization can fail because different channels can carry nonzero WHF partial indices even when the total winding vanishes (Kaneshiro et al., 14 Nov 2025).

WHF resolves the obstruction. For a matrix Laurent polynomial $\mathbb C_E$0,

$\mathbb C_E$1

the integers $\mathbb C_E$2 are the partial indices. In multiband non-Hermitian systems, they encode skin or edge modes of the Hermitian-doubled problem. The central applicability criterion is sharp: $\mathbb C_E$3 where $\mathbb C_E$4 are the residual partial indices at the optimizing $\mathbb C_E$5. When residual partial indices are nonzero, the OBC potential differs from the naïve minimum by root-exchange corrections determined by $\mathbb C_E$6 (Kaneshiro et al., 14 Nov 2025).

In class AII$\mathbb C_E$7, Kramers pairing makes the total Ronkin function even in $\mathbb C_E$8, so the naïve Amoeba minimization always returns $\mathbb C_E$9. The remedy is a symmetry-resolved construction. For

$\kappa=\operatorname{Im}k$00

the symplectic WHF produces band-resolved Ronkin functions

$\kappa=\operatorname{Im}k$01

with

$\kappa=\operatorname{Im}k$02

The generalized Szegő relation then becomes

$\kappa=\operatorname{Im}k$03

which rigorously recovers the symmetry-decomposed Amoeba formula and the localization lengths of the Kramers-paired channels (Kaneshiro et al., 25 Feb 2025).

A different real-space route replaces global non-Bloch momentum by local twisting. For a general 1D nonreciprocal chain

$\kappa=\operatorname{Im}k$04

a local scaling transformation defines site-dependent factors $\kappa=\operatorname{Im}k$05 and local twisting variables

$\kappa=\operatorname{Im}k$06

The cumulative twist

$\kappa=\operatorname{Im}k$07

controls the localization envelope, and the set

$\kappa=\operatorname{Im}k$08

defines the Zahlen-Brillouin Zone (ZBZ), which reduces to the GBZ in periodic systems but remains well-defined in nonperiodic and disordered lattices. This framework also introduces the generalized multiple-channel skin effect (MCSE) and the global skin index

$\kappa=\operatorname{Im}k$09

together with the extended skin index

$\kappa=\operatorname{Im}k$10

for distinguishing unidirectional and competing twisting regimes (Zhao et al., 25 May 2026).

5. OBC computation, decimation, and flat-band reductions

Direct OBC computation in multiband non-Hermitian systems is often recast as a recurrence problem for the characteristic polynomial

$\kappa=\operatorname{Im}k$11

For finite-range hopping, the determinants satisfy a linear recurrence in the system length,

$\kappa=\operatorname{Im}k$12

and for generic nearest-neighbor multiband chains this simplifies to a second-order recurrence,

$\kappa=\operatorname{Im}k$13

whose order is independent of the band number. In the thermodynamic limit, the bulk OBC spectrum is characterized by equality of the maximal-modulus recurrence roots, while the edge spectrum is characterized by vanishing of the coefficient of a uniquely dominant root. For periodic nearest-neighbor $\kappa=\operatorname{Im}k$14-band chains, the bulk OBC bands are obtained from

$\kappa=\operatorname{Im}k$15

and the edge spectrum follows from a low-degree polynomial constraint. This gives an algebraic alternative to GBZ methods, with particular efficiency for multiband chains and quasi-1D Hofstadter problems (Chen et al., 2024).

Exact real-space decimation provides a complementary reduction of multiband models to lower-dimensional effective Hamiltonians while preserving the full characteristic polynomial. In a 4-band generalized non-Hermitian SSH chain, two middle sites per supercell can be eliminated to obtain an effective two-band Hamiltonian

$\kappa=\operatorname{Im}k$16

with energy-dependent renormalized parameters. The reduced model reproduces the full 4-band spectrum exactly and is then used to derive GBZ equations, transfer matrices, non-Bloch van Hove singularities, and open-boundary spectra (Banerjee et al., 2023).

The same decimation framework generates tunable non-Hermitian flat bands. In the quasi-1D non-Hermitian Lieb chain, successive elimination of triply coordinated sites yields a reduced ladder Hamiltonian whose two dispersions satisfy

$\kappa=\operatorname{Im}k$17

For the symmetric gain/loss choice used in the paper, the coefficient of $\kappa=\operatorname{Im}k$18 in one branch vanishes, producing the exact flat band

$\kappa=\operatorname{Im}k$19

The corresponding compact localized states satisfy

$\kappa=\operatorname{Im}k$20

This motivates the paper’s hypothesis that quasi-1D bipartite non-Hermitian systems with flat bands decouple into SSH chains and compact localized states across various models (Banerjee et al., 2023).

6. Experimental realizations and present scope

The phenomenology of multiband twisting is no longer purely theoretical. The spectral-graph program explicitly points to photonic, mechanical, electrical, and cold-atom platforms as settings where Hatano–Nelson and non-Hermitian SSH spectra have already been realized, and where multi-component star-, flower-, and insect-like spectra become accessible by enlarging the unit cell, introducing longer-range asymmetric couplings, and engineering effective energy-dependent couplings (Tai et al., 2022).

A photonic-crystal realization of multiband permutation topology was constructed from triangular cavities with two sub-cavities per unit cell and refractive index $\kappa=\operatorname{Im}k$21. In that system, real and imaginary PHLs connect several pairs of EPs between bands 3 and 4. Along a loop parallel to $\kappa=\operatorname{Im}k$22 at fixed $\kappa=\operatorname{Im}k$23, the third and fourth bands exchange states after one adiabatic period, directly realizing non-separable band topology in a realistic lossy photonic medium (Ryu et al., 2024).

Experimental access to full non-Bloch band geometry has advanced through the non-Bloch supercell framework. There, real Bloch phase is imposed by twisted boundary conditions, while the imaginary part of momentum is controlled independently by an exponent-flattening protocol that redistributes $\kappa=\operatorname{Im}k$24 across bulk hoppings. Implemented in programmable one- and two-dimensional acoustic crystals, this framework reconstructs momentum-resolved complex energy surfaces $\kappa=\operatorname{Im}k$25 and biorthogonal eigenmodes from Green’s-function measurements, and the resulting data accurately predict open-boundary spectra and eigenstates verified in separate open-geometry experiments (Zhong et al., 23 Oct 2025).

The braid and knot sector has now been brought onto quantum hardware. The $\kappa=\operatorname{Im}k$26-band twister family was digitally simulated on a programmable superconducting quantum processor using a non-variational protocol based on dilated nonunitary evolution, Hamiltonian rotation for eigenstate selection, and measurement of a small set of local observables sufficient to reconstruct braid words. This yielded experimental reconstructions of the Hopf link, Hopf chain, Solomon’s knot, and related braid topologies, together with Alexander and Jones polynomials obtained without full spectral tomography (Ng et al., 29 Apr 2026).

Taken together, these developments position non-Hermitian multi-band twister models at the intersection of non-Bloch band theory, planar spectral-graph topology, permutation-group monodromy, and braid-based spectral topology. The unifying theme is not a single invariant but a hierarchy of structures: branch counts and loops in spectral graphs, permutation conjugacy classes on compact parameter spaces, WHF partial indices and Ronkin minima under OBC, and braid words or knot polynomials when the spectrum itself becomes a multi-strand object. This suggests that “twister” is best regarded not as a single model family but as a common geometric regime of multiband non-Hermitian band theory (Tai et al., 2022, Kaneshiro et al., 14 Nov 2025, Ng et al., 29 Apr 2026).