Edge-Sensitive Topological Invariant

• Edge-sensitive topological invariants are defined to capture boundary phenomena, such as gapless edge modes and defect states, through explicit edge mode counting and spectral flow analysis.
• They are derived using various methodologies, including gluing functions, scattering matrices, and many-body entanglement, demonstrating a versatile approach to topology.
• These invariants reveal both the precise counting of edge states and situations where nontrivial bulk topology does not guarantee gapless boundaries, emphasizing their contextual nature.

Searching arXiv for recent and foundational papers on edge-sensitive topological invariants. An edge-sensitive topological invariant is a bulk- or boundary-derived quantity whose nontrivial value is tied, directly or indirectly, to boundary phenomena such as gapless edge modes, edge spectral flow, boundary-localized zero modes, or boundary geometric response. In the literature, the phrase does not refer to a single universal object. Instead, it denotes a family of constructions in which topology is resolved relative to a boundary orientation, a surface momentum, a boundary condition, a scattering setup, or a boundary defect. A central theme is that some invariants are strongly edge-sensitive in the sense that they count edge modes or boundary crossings exactly, whereas others are only edge-sensitive through bulk-edge correspondence, and still others furnish counterexamples by remaining physically consequential even when edges are gapped (Fidkowski et al., 2011, Sun et al., 2019, Shulman et al., 2010).

1. Edge sensitivity as a topological criterion

In one common usage, an edge-sensitive invariant is a topological quantity whose value is determined by boundary spectral data. This viewpoint is explicit in the reformulation of the Kane–Mele $\mathbb Z_2$ invariant as the parity of the spectral flow of a family of $1+1$-dimensional Dirac operators equipped with a Kramers-induced Atiyah–Patodi–Singer global boundary condition. There the bulk invariant $\nu$ is identified with the mod-$2$ spectral flow,

$\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$

and, because the bulk remains gapped, only edge states contribute to the crossings; the parity of the number of Kramers pairs of gapless edge states is exactly the parity of the spectral flow (Yu et al., 2016).

A second strong notion of edge sensitivity appears in continuum topological elasticity. There the invariant is defined not from the dynamical matrix alone but from a continuum compatibility operator $\mathcal C$, and for a strip geometry with fixed surface momentum $\bar q_x$ one defines

$$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$

This winding number counts zeros of $\det \mathcal C(\bar q_x,q_y)$ in the relevant half-plane and therefore directly counts top-edge or bottom-edge zero modes at that momentum. The invariant depends explicitly on boundary orientation through the decomposition into tangential and transverse momenta and is therefore edge-sensitive in the strongest sense (Sun et al., 2019).

A broader but still precise formulation treats boundary sensitivity through a bulk object whose spectrum is continuously deformable to the edge spectrum. For translation-invariant band insulators, the relevant object is the gluing function

$$U_g(\vec{k}_\parallel)=\mathcal{P}\exp\!\left(i\int_{0}^{2\pi}A_\perp(\vec{k}_\parallel,k_\perp)\,dk_\perp\right),$$

whose eigenphases are hybrid Wannier centers. The topological content lies in the spectral flow of $1+1$0 as a function of $1+1$1, and this flow is continuously deformable to the physical edge-mode spectrum for the chosen cut (Fidkowski et al., 2011).

These constructions suggest a useful distinction. Some invariants are edge-defined or edge-counting; some are bulk-defined but edge-resolved; and some are bulk invariants whose physical significance is revealed only after a boundary is introduced.

2. Boundary-resolved formulations and exact counting statements

Several works formulate edge-sensitive topology through exact counting statements. In continuum Maxwell media, zero modes are fields satisfying $1+1$2, and the generalized Maxwell–Calladine theorem reads

$1+1$3

At the Maxwell point, $1+1$4 is square, and the edge-resolved winding $1+1$5 counts the number of long-wavelength edge-localized zero modes at fixed surface momentum. Because the contour excludes nonphysical gauge poles, this count is gauge-invariant and insensitive to microscopic choices such as unit cell convention or lattice connectivity (Sun et al., 2019).

In the scattering formulation of topological pumping, the edge invariant is the winding of scattering-matrix eigenvalues. For a measured $1+1$6 edge scattering matrix $1+1$7 with eigenvalues $1+1$8, the invariant is the integer winding of $1+1$9 as the boundary twist angle $\nu$0 is varied through a $\nu$1 cycle. In the topologically nontrivial regime the windings are $\nu$2, while in the trivial regime they vanish. The physical content is explicitly boundary-based: nonzero winding of the edge scattering eigenvalues implies topological edge states in the probed bulk gap (Hu et al., 2014).

For two-dimensional topological insulators in the single-particle setting, the edge invariant can be written as

$\nu$3

where $\nu$4 is the number of edge-state crossings of the Fermi level for $\nu$5, with endpoint crossings counted half. This edge parity equals a bulk $\nu$6 invariant defined from a bundle of decaying bulk solutions over $\nu$7, even when no full two-dimensional Brillouin zone is assumed (Graf et al., 2012).

In the mathematical coarse-geometric formulation of bulk-edge correspondence, the edge-sensitive object is a boundary-localized coarse index obstruction

$\nu$8

obtained by applying the exponential boundary map of a Roe-algebra exact sequence to a bulk spectral projection $\nu$9. Because this class lives in the $2$0-group of an algebra supported near $2$1, it is intrinsically boundary-localized. Its nontriviality forces a spectral gap-filling phenomenon after compressing the Hamiltonian to a domain with boundary, and its Fredholm pairing computes quantized boundary current channels that follow the boundary (Ludewig et al., 2020).

Across these examples, the common structure is exactness: the invariant is not merely correlated with edge physics but determines an edge count, an edge crossing parity, or a boundary current index.

3. Bulk invariants with edge-sensitive manifestation

Other constructions remain bulk-defined but acquire operational meaning through the boundary. In a time-reversal-invariant two-dimensional multiband $2$2-wave superconductor, the full Chern number vanishes by time-reversal symmetry, but one can define a partial Chern-number sum $2$3 from one representative block and its parity

$2$4

This parity diagnoses whether the system supports an odd number of helical Majorana edge-mode pairs. Thus the invariant is computed in the bulk but is edge-sensitive because its robust content is precisely the parity of helical Majorana Kramers pairs at the boundary (Deng et al., 2011).

A related phenomenon appears in two-dimensional topological metals with no global clean band gap. There the usual clean-band $2$5 invariant is not well defined, yet disorder localizes bulk states and produces a mobility-gap regime in which the spin Chern number

$2$6

becomes quantized, with $2$7. The invariant is computed from the disordered bulk with periodic boundary conditions, but its physical signature is edge-only transport: with open boundaries the conductance shows a plateau at $2$8, whereas with periodic boundaries it goes to zero (Chen et al., 2019).

For open subsystems coupled to an environment, the Green’s-function Chern invariant

$2$9

is edge-sensitive in a generalized sense because it counts the difference between gapless edge bands arising from poles and gapless edge blind bands arising from zeros of the subsystem Green’s function. This extends bulk-edge correspondence to open systems, where edge content is no longer exhausted by ordinary spectral branches (Zheng et al., 2017).

A further example is the $\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$0 spin-skyrmion invariant

$\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$1

built from sector-resolved pseudospin textures rather than occupied-state projectors. In the models studied, odd $\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$2 predicts robust helical edge states under open boundary conditions even when the conventional projector-based $\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$3 invariant is trivial (Flores-Calderon et al., 2023).

These cases suggest that edge sensitivity need not mean that the invariant is computed from edge data. It may instead mean that the invariant’s experimentally or spectrally decisive manifestation occurs only when edges are present.

4. Counterexamples and limits of edge sensitivity

A major line of work shows that nontrivial topology need not enforce gapless edge spectra. In a spin-$\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$4 Kane–Mele-type model, the clean topological phase has spin-Chern number

$\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$5

and nevertheless exhibits gapped edge states. Under spin-independent disorder, the model still supports protected extended bulk states, with level-statistics evidence of levitation and pair annihilation. By varying the Fermi level across the mobility edges, the authors show that the invariant transferred by these extended states is precisely the spin-Chern number. In this example the topology is physically consequential but not edge-gaplessness-sensitive in the narrow sense (Shulman et al., 2010).

A different limitation arises in rotationally invariant Abelian topological phases. There, coupling boundary modes to the extrinsic geometry of the boundary allows one to construct rotationally invariant, charge-neutral mass terms

$\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$6

or, for half-integer spin-vector entries,

$\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$7

The resulting gapped boundary theory carries a local induced action, the Gromov-Jensen-Abanov action,

$\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$8

which depends on the extrinsic geometry. The key conclusion is that rotational invariance does not protect gapless edge modes in the class studied; the shift $\mathrm{SF}\{D_t\}=N_{-\to \ge 0}-N_{\ge 0\to -},$9 survives as a boundary geometric response coefficient, not as a gap-protecting invariant (Sinha et al., 18 Mar 2025).

The possibility of defect-sensitive rather than edge-gapless topology further complicates the picture. In a modified Dirac model for time-reversal-invariant topological insulators, vacancies or holes act as internal boundaries and bind $\mathcal C$0-protected in-gap bound states. These states are descendants of ordinary edge or surface modes under shrinkage of an internal boundary and show that boundary sensitivity can extend to defects, not only to outer sample edges (Shan et al., 2010).

Together, these works establish that “edge-sensitive” should not be conflated with “forces gapless edge dispersion.” A nontrivial invariant may count edge states, control defect states, govern a boundary response, or instead obstruct complete localization in the bulk while leaving the edge gapped.

5. Non-Hermitian and symmetry-unprotected generalizations

Non-Hermitian systems broaden the notion of edge sensitivity further. In a class of non-Hermitian topological insulators obtained by similarity transformation of Hermitian parents, the bulk spectrum remains real and there is no non-Hermitian skin effect, yet the boundary sector can become anomalous. The paper introduces an inversion-protected total $\mathcal C$1 invariant. In one dimension,

$\mathcal C$2

and in the deformed two-dimensional Qi–Wu–Zhang model with $\mathcal C$3,

$\mathcal C$4

A nontrivial $\mathcal C$5 diagnoses the topological phase, but the edge manifestation is anomalous: under suitable chiral or generalized chiral symmetry, the boundary modes can become defective rather than ordinary independent edge states (Wang et al., 2020).

An even broader non-symmetry-protected construction exists for one-dimensional nearest-neighbor two-band models with arbitrary complex couplings. There the relevant invariant is defined on the analytically continued complex band structure and the associated eigenvector Riemann surface. In the general non-Hermitian case,

$\mathcal C$6

This invariant predicts the number of open-boundary edge states, which may be $\mathcal C$7, $\mathcal C$8, or $\mathcal C$9, even though the system has no symmetry protection in the usual sense and the transition does not coincide with the ordinary gap-closing condition (Zhong et al., 2024).

Non-Hermitian edge sensitivity can also appear as a response property rather than an invariant. In a finite non-Hermitian dimer chain in the topological regime $\bar q_x$0, edge-edge hybridization splits the topological edge states, and tuning gain and loss drives the boundary doublet to an exceptional point. Near that point, a perturbation applied specifically at the chain end produces enhanced square-root response,

$\bar q_x$1

whereas before the exceptional point the response is linear and beyond it the edge states are barely affected. This is an edge-sensitive topological sensing mechanism built on topological edge modes, not a new invariant (Guo et al., 2020).

These examples show that once Hermiticity or symmetry protection is relaxed, edge sensitivity may concern counting, defectiveness, or susceptibility of the boundary sector, and the invariant structure can migrate from ordinary Brillouin-zone topology to generalized Bloch, scattering, or Riemann-surface formulations.

6. Many-body, crystalline, and higher-order boundary sensitivity

In interacting settings, edge-sensitive invariants may be many-body quantities whose interpretation is most transparent from edge conformal field theory or boundary defects. For $\bar q_x$2D invertible fermionic phases with $\bar q_x$3 symmetry, a many-body invariant is defined from a tripartite entanglement construction,

$\bar q_x$4

Its quantized phase detects both the chiral central charge and an $\bar q_x$5-torsion index, and the edge-CFT interpretation reduces it to an overlap of a twisted symmetry action with an Ishibashi state. For symmetry-preserving boundaries, nontrivial phases are distinguished by zero modes at the intersection of the reflection axis and the boundary, so the invariant is sensitive not merely to a boundary line but to a special boundary point (Kobayashi et al., 2024).

The coarse-geometric bulk-edge framework likewise extends edge sensitivity beyond crystalline edges. In that setting, the numerical pairing

$\bar q_x$6

is invariant under admissible boundary deformations and coarse cobordisms. This suggests that edge-sensitive topology can be formulated without reference to a straight edge, a Brillouin zone, or even a periodic lattice, provided locality and coarse geometry are retained (Ludewig et al., 2020, Kubota, 2015).

A mathematically related but distinct framework for controlled topological phases defines bulk and edge indices as classes in twisted equivariant $\bar q_x$7-groups of Roe algebras. The edge index is defined after cutting space into a half-space and passing to a boundary/relative Roe algebra, so it is genuinely boundary-localized. Bulk-edge correspondence is then the statement

$\bar q_x$8

proved through the coarse Mayer–Vietoris exact sequence (Kubota, 2015).

These many-body and coarse-geometric approaches indicate that edge sensitivity is not restricted to single-particle band theory. It can also appear as a quantized many-body phase, a boundary defect signature, or a $\bar q_x$9-theoretic boundary class.

7. Conceptual synthesis

The literature supports at least four distinct meanings of an edge-sensitive topological invariant.

First, there are edge-counting invariants. These include the continuum compatibility-operator winding $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$0, the parity of edge spectral flow $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$1, and the scattering-matrix winding of $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$2. Their nontrivial values directly count edge modes, crossings, or edge-state spectral flow (Sun et al., 2019, Graf et al., 2012, Hu et al., 2014).

Second, there are bulk invariants with exact edge manifestation. The gluing-function spectrum, the partial Chern-number parity $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$3, the emergent $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$4 invariant in disordered topological metals, and the open-system Green’s-function Chern invariant fall in this category. They are not edge-defined, but their operational content is boundary-localized or boundary-selective (Fidkowski et al., 2011, Deng et al., 2011, Chen et al., 2019, Zheng et al., 2017).

Third, there are counterexamples to naive edge sensitivity. The spin-Chern invariant in systems with gapped edges and the geometric response coefficients $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$5 and $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$6 in rotationally invariant Abelian phases show that a nontrivial invariant may fail to protect gapless boundary modes while remaining physically meaningful (Shulman et al., 2010, Sinha et al., 18 Mar 2025).

Fourth, there are generalized boundary-sensitive structures in non-Hermitian, symmetry-unprotected, many-body, or coarse-geometric settings. These include inversion-based $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$7 indicators whose topological phases host defective edge states, Riemann-surface winding numbers that count open-boundary states without symmetry protection, tripartite many-body $$N_{T/B}(\bar q_x)=\frac{1}{2\pi i}\oint_{T/B}dq_y\; \mathrm{tr}\!\left(\mathcal C^{-1}\partial_{q_y}\mathcal C\right).$$8 phases that diagnose boundary-intersection zero modes, and Roe-algebra boundary classes that force edge-following current channels (Wang et al., 2020, Zhong et al., 2024, Kobayashi et al., 2024, Ludewig et al., 2020).

A plausible implication is that “edge-sensitive topological invariant” is best treated as a contextual term rather than a fixed technical label. In some settings it means an invariant that exactly counts edge modes; in others it means a bulk quantity whose nontriviality is revealed only by boundary response; and in still others the most important lesson is negative, namely that topology need not be edge-gaplessness-sensitive. The modern literature therefore replaces a single bulk-boundary paradigm with a more differentiated picture: edge spectra, mobility edges, defect modes, scattering windings, entanglement phases, and boundary geometric actions are all legitimate carriers of topological information, but they are not interchangeable (Shulman et al., 2010, Sun et al., 2019, Yu et al., 2016).