Intrinsic non-Hermitian topological phases

Ken Shiozaki · 2025

Read Intrinsic non-Hermitian topological phases by Ken Shiozaki (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-06879.txt.

Introduction

Phases that exist only because a system is non-Hermitian are separated from impostors by Ken Shiozaki in a solo 2025 condensed matter paper. Non-Hermitian topology comes in point-gap and line-gap flavors. Some phases reduce to plain Hermitian or anti-Hermitian ones; others are truly new, with no Hermitian counterpart.

Natural maps from line-gap phases to point-gap phases let Shiozaki draw the line precisely: extrinsic phases are those reducible in that way, intrinsic phases are the rest. Classification tables for all symmetry classes existed in earlier work, but this paper develops a single shared treatment and works through every internal symmetry class. One shared treatment matters because researchers comparing results across symmetry classes need one consistent language for what counts as an intrinsically non-Hermitian phase.

Catalog record

CREATORKen Shiozaki
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE123 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-06879

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Intrinsic non-Hermitian topological phases Ken Shiozaki Abstract We study the interplay of non-Hermitian topological phases under point- and line-gap conditions. Using natural homomorphisms from line-gap to point-gap phases, we distinguish extrinsic phases, reducible to Hermitian or anti-Hermitian line-gapped phases, from intrinsic phases, which are genuinely non-Hermitian without Hermitian counterparts. Although classification tables for all symmetry classes were already presented in earlier work, the present paper develops a unified formulation and provides explicit computations for all internal symmetries. 1 Introduction Non-Hermitian quantum systems have recently attracted broad interest across condensed matter physics, open quantum systems, photonic and acoustic platforms, and mechanical metamaterials, providing a setting for novel phenomena absent in Hermitian systems (see reviews [ 1 , 2 , 3 , 4 , 5 ] ). Compared with Hermitian systems, non-Hermitian systems exhibit several distinctive features: eigenvalues are generally complex, right and left eigenvectors differ, and Hamiltonians may fail to be diagonalizable, giving rise to exceptional points. Moreover, spectral properties often show strong boundary dependence, which is called the non-Hermitian skin effect [ 6 ] , where bulk spectra under periodic and open boundary conditions differ drastically and an extensive number of eigenstates accumulate at the boundary. In such cases, the bulk-boundary correspondence must be reformulated in terms of the generalized Brillouin zone [ 6 , 7 ] . These intrinsic properties of non-Hermitian systems require new topological principles, motivating systematic studies of gap conditions and classification schemes for non-Hermitian Hamiltonians. One approach to non-Hermitian topology is based on the point-gap condition, where the spectrum avoids a reference point in the complex plane [ 8 ] . The point gap gives rise to non-Hermitian-specific topological invariants, such as spectral winding numbers. Alternatively, line gaps with respect to the real or imaginary axis [ 9 ] relate non-Hermitian phases to Hermitian topological insulators and superconductors, thereby connecting non-Hermitian classifications to those for Altland–Zirnbauer (AZ) classes [ 10 , 11 , 12 , 13 ] . This naturally raises the question: what phenomena are genuinely intrinsic to non-Hermitian systems, i.e., not realizable within Hermitian settings? For instance, gapless edge states of a Chern insulator remain stable under non-Hermitian perturbations, indicating their essentially Hermitian origin. By contrast, the non-Hermitian skin effect emerges from the spectral winding number that vanishes in Hermitian systems and therefore represents a genuine non-Hermitian topological phenomenon [ 14 , 15 ] . In this work, we propose a systematic framework to classify such intrinsic non-Hermitian topological phases by subtracting line-gap phases from point-gap phases. Since every line-gapped Hamiltonian is also point-gapped, this defines natural homomorphisms from line-gap to point-gap classifications. We identify the image of these maps with extrinsic phases, while point-gap phases outside the image are defined as intrinsic non-Hermitian topological phases. Explicit examples of intrinsic non-Hermitian phases have been reported in various dimensions and symmetry classes: one- and two-dimensional \mathbb{Z}_{2} skin effects with a reciprocal symmetry [ 15 ] , an isolated exceptional point appearing on the surface [ 16 ] and magnetic field induced non-Hermitian skin effect in three dimensions [ 17 , 18 ] , boundary states associated with PT symmetry breaking trajectories [ 19 , 20 ] , and momentum-space defect structures [ 21 ] . Systematic analyses of intrinsic non-Hermitian phases are also discussed in [ 20 , 19 ] . As a related development, the recently established detachability of gapless boundary states in Hermitian systems [ 22 ] has been interpreted as an instance of imaginary-line-gap extrinsic phases [ 23 , 24 ] . In this paper, we present a unified formalism that incorporates crystalline symmetries and provides explicit computations for all internal symmetry classes. In particular, we obtain classification results for all 38 independent non-Hermitian symmetry classes, distinguishing extrinsic from intrinsic phases (see Tables 5 – 11 ). While the classification results already appeared in Supplemental Material of Ref. [ 15 ] , the present work develops a comprehensive formulation together with detailed computations for internal symmetries. The remainder of this paper is organized as follows. In Sec. 2 we review basic ingredients of non-Hermitian free-fermion systems, including the definition of symmetry, the Hermitization procedure, and gap conditions. In Sec. 3 we introduce point-gap and line-gap topological phases and formulate the natural homomorphisms from line-gap to point-gap classifications. Sec. 4 presents explicit computations of these homomorphisms for all internal symmetry classes and provides the resulting classification tables. We summarize our findings and discuss future directions in Sec. 5. 2 Some generalities In this section, we review several basic ingredients of non-Hermitian free-fermion systems: the definition of symmetry, the Hermitization procedure, and gap conditions. 2.1 Symmetry We begin with symmetries of non-Hermitian matrices H_{\bm{k}} defined over a general parameter space X , with {\bm{k}}\in X . Let G be a finite group, and denote its left action on X by G\times X\ni(g,{\bm{k}})\mapsto g{\bm{k}}\in X . In a non-Hermitian setting, the Hamiltonian is not necessarily Hermitian, and symmetry transformations may involve complex conjugation, transpose, or their combination, the Hermitian conjugate. To distinguish these possibilities, we introduce two homomorphisms \displaystyle\phi,\kappa:G\to\mathbb{Z}/2=\{\pm 1\}, (1) and adopt the following notation for a complex number or a matrix M : \displaystyle M^{\phi_{g}}=\begin{cases}M&(\phi_{g}=1),\\ M^{*}&(\phi_{g}=-1),\end{cases} (2) \displaystyle M^{\phi_{g},\kappa_{g}}=\begin{cases}M&(\phi_{g}=1,\kappa_{g}=1),\\ M^{*}&(\phi_{g}=-1,\kappa_{g}=1),\\ M^{\top}&(\phi_{g}=-1,\kappa_{g}=-1),\\ M^{\dagger}&(\phi_{g}=1,\kappa_{g}=-1).\end{cases} (3) For a family of non-Hermitian matrices on X , we consider unitary matrices u_{\bm{k}}(g) and v_{\bm{k}}(g) , defined continuously on X , that satisfy the symmetry constraint \displaystyle u_{\bm{k}}(g)\,H_{\bm{k}}^{\phi_{g},\kappa_{g}}\,v_{\bm{k}}(g)^{\dagger}=H_{g{\bm{k}}},\quad g\in G. (4) Note that, unlike in the Hermitian case, it is not required that u_{\bm{k}}(g)=v_{\bm{k}}(g) . For g,h\in G , comparing two decompositions, \displaystyle H_{gh{\bm{k}}} \displaystyle=u_{\bm{k}}(gh)H_{\bm{k}}^{\phi_{gh},\kappa_{gh}}v_{\bm{k}}(gh)^{\dagger} \displaystyle=\begin{cases}u_{h{\bm{k}}}(g)\,u_{\bm{k}}(h)^{\phi_{g}}\,H_{\bm{k}}^{\phi_{gh},\kappa_{gh}}\left(v_{h{\bm{k}}}(g)\,v_{\bm{k}}(h)^{\phi_{g}}\right)^{\dagger}&(\kappa_{g}=1),\\ u_{h{\bm{k}}}(g)\,v_{\bm{k}}(h)^{\phi_{g}}\,H_{\bm{k}}^{\phi_{gh},\kappa_{gh}}\left(v_{h{\bm{k}}}(g)\,u_{\bm{k}}(h)^{\phi_{g}}\right)^{\dagger}&(\kappa_{g}=-1),\end{cases} (5) we obtain \displaystyle\begin{cases}u_{\bm{k}}(gh)^{\dagger}\,u_{h{\bm{k}}}(g)\,u_{\bm{k}}(h)^{\phi_{g}}\,H_{\bm{k}}^{\phi_{gh},\kappa_{gh}}=H_{\bm{k}}^{\phi_{gh},\kappa_{gh}}\,v_{\bm{k}}(gh)^{\dagger}\,v_{h{\bm{k}}}(g)\,v_{\bm{k}}(h)^{\phi_{g}}&(\kappa_{g}=1),\\ u_{\bm{k}}(gh)^{\dagger}\,u_{h{\bm{k}}}(g)\,v_{\bm{k}}(h)^{\phi_{g}}\,H_{\bm{k}}^{\phi_{gh},\kappa_{gh}}=H_{\bm{k}}^{\phi_{gh},\kappa_{gh}}\,v_{\bm{k}}(gh)^{\dagger}\,v_{h{\bm{k}}}(g)\,u_{\bm{k}}(h)^{\phi_{g}}&(\kappa_{g}=-1).\end{cases} (6) For simplicity, we assume that ( 6 ) imposes only trivial constraints. In other words, we suppose there exists a U(1) phase z_{\bm{k}}(g,h) such that \displaystyle\begin{cases}u_{\bm{k}}(gh)^{\dagger}\,u_{h{\bm{k}}}(g)\,u_{\bm{k}}(h)^{\phi_{g}}=\left(v_{\bm{k}}(gh)^{\dagger}\,v_{h{\bm{k}}}(g)\,v_{\bm{k}}(h)^{\phi_{g}}\right)^{\dagger}=z_{\bm{k}}(g,h)&(\kappa_{g}=1),\\ u_{\bm{k}}(gh)^{\dagger}\,u_{h{\bm{k}}}(g)\,v_{\bm{k}}(h)^{\phi_{g}}=\left(v_{\bm{k}}(gh)^{\dagger}\,v_{h{\bm{k}}}(g)\,u_{\bm{k}}(h)^{\phi_{g}}\right)^{\dagger}=z_{\bm{k}}(g,h)&(\kappa_{g}=-1),\end{cases} (7) holds. The U(1) phase z_{\bm{k}}(g,h) must then satisfy the following cocycle condition depending on {\bm{k}}\in X : \displaystyle z_{h,l}({\bm{k}})^{\phi_{g}}\,z_{gh,l}({\bm{k}})^{-1}\,z_{g,hl}({\bm{k}})\,z_{g,h}(l{\bm{k}})^{-1}=1,\quad g,h,l\in G,\ {\bm{k}}\in X. (8) 2.2 Hermitization For a non-Hermitian Hamiltonian H_{\bm{k}} , we double the matrix size and define the Hermitian Hamiltonian \displaystyle\tilde{H}_{\bm{k}}:=\begin{pmatrix}&H_{\bm{k}}\\ H_{\bm{k}}^{\dagger}&\end{pmatrix}_{\sigma}, (9) where the subscript \sigma indicates the block structure associated with the doubled internal degrees of freedom. We denote by {\sigma}_{\mu} (\mu=0,x,y,z) the Pauli matrices acting on this doubled space. The corresponding doubled symmetry operators are defined by \displaystyle\tilde{u}_{\bm{k}}(g):=\begin{cases}\begin{pmatrix}u_{\bm{k}}(g)&\\ &v_{\bm{k}}(g)\\ \end{pmatrix}&(\kappa_{g}=1),\\[5.16663pt] \begin{pmatrix}&u_{\bm{k}}(g)\\ v_{\bm{k}}(g)&\\ \end{pmatrix}&(\kappa_{g}=-1),\end{cases}\quad g\in G, (10) so that the symmetry relations ( 4 ) and ( 7 ) can be written more compactly in the Hermitianized form: \displaystyle\tilde{u}_{\bm{k}}(g)\,\tilde{H}_{\bm{k}}^{\phi_{g}}\,\tilde{u}_{\bm{k}}(g)^{\dagger}=\tilde{H}_{g{\bm{k}}},\quad g\in G, \displaystyle\tilde{u}_{h{\bm{k}}}(g)\,\tilde{u}_{\bm{k}}(h)^{\phi_{g}}=z_{\bm{k}}(g,h)\,\tilde{u}_{\bm{k}}(gh),\quad g,h\in G. (11) The doubling in ( 9 ) and ( 10 ) also implies the chiral symmetry \displaystyle{\sigma}_{z}\tilde{H}_{\bm{k}}{\sigma}_{z}=-\tilde{H}_{\bm{k}}, (12) together with the algebraic relation between \tilde{u}_{\bm{k}}(g) and the chiral operator, \displaystyle{\sigma}_{z}\tilde{u}_{\bm{k}}(g)=\kappa_{g}\,\tilde{u}_{\bm{k}}(g){\sigma}_{z},\quad g\in G. (13) Conversely, the conditions ( 11 ), ( 12 ), and ( 13 ) imply the original non-Hermitian symmetry relation ( 4 ). Thus, the topological properties of a non-Hermitian Hamiltonian H_{\bm{k}} can be studied equivalently through its Hermitianized counterpart \tilde{H}_{\bm{k}} . 2.3 Point-gap condition For a family of non-Hermitian Hamiltonians H_{\bm{k}} , we would like to define a notion of topologically stable classification. Since any two Hamiltonians H_{\bm{k}},H_{\bm{k}}^{\prime} can be linearly interpolated while preserving the symmetry ( 4 ) as \displaystyle(1-t)H_{\bm{k}}+tH_{\bm{k}}^{\prime},\quad t\in[0,1], (14) no nontrivial classification can be obtained without an additional constraint. To obtain a

Read the complete text