Spectral Topological Invariants

• Spectral topological invariants are defined as robust, quantized measures extracted from operator spectra, including mod-2 Fredholm indices, winding numbers, and localizer signatures.
• They leverage methodologies like spectral flow, asymptotic phase analysis, and determinant mapping to ensure stability under perturbations and characterize bulk–edge correspondence.
• These invariants enable classification in diverse settings—from topological insulators to non-Hermitian systems and Floer theory—providing practical insights for experimental validation.

A spectral topological invariant is a topological quantity extracted from spectral data of an operator, an operator family, or a spectrally defined filtration. In the cited literature, the term encompasses mod-$2$ Fredholm indices of gapped time-reversal-invariant Hamiltonians, winding numbers determined by asymptotic symbols, essential spectra stable under compact perturbations, signatures and determinant signs of spectral localizers, vorticities of complex eigenenergies at exceptional points, conductivity sum rules governed by band topology and quantum geometry, intersection-theoretic classes attached to spectral curves, and Floer-theoretic spectral numbers. The common structure is that topological information is encoded by spectral behavior that is stable under an appropriate notion of deformation, such as norm-continuous homotopy, compact perturbation, locality-preserving truncation, or filtered continuation (Fonseca et al., 2019, Tanaka, 2020, Su et al., 2020, Qi et al., 2024, Eynard, 2011, Kawamoto et al., 2023).

1. Core meaning and representative forms

Across the operator-theoretic literature, spectral topological invariants are typically defined from one of four spectral mechanisms: a Fredholm defect, a winding of a determinant or eigenvalue phase, a localizer signature or determinant sign, or an integrated spectral response. Each mechanism produces a quantity that survives perturbations preserving the relevant gap, locality, symmetry, or asymptotic phase. Representative formulas used in the cited works include the mod-$2$ Fredholm index $\operatorname{ind}_2 A=\dim\ker A \pmod 2$, the winding-number formula for one-dimensional strictly local operators, the exceptional-point vorticity $w$, the real-space localizer sign $C_{x,\bar\epsilon}$, the spectral-localizer signature $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$, and the flat-band conductivity sum rule $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$ (Fonseca et al., 2019, Tanaka, 2020, Su et al., 2020, Qi et al., 2024, Jezequel et al., 31 Jul 2025, Kruchkov et al., 2023).

Formulation	Representative expression	Typical setting
Fredholm parity	$\operatorname{ind}_2 A=\dim\ker A \pmod 2$	2D TRI insulators
Asymptotic-symbol index	$$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$	1D two-phase strictly local operators
Spectral winding	$$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$	Non-Hermitian exceptional points
Localizer sign / signature	$2$0, $2$1	Dense-spectrum or disordered systems
Optical sum-rule invariant	$2$2	Flat Chern bands

This diversity is substantive rather than merely terminological. In some works the invariant is a parity-valued obstruction to trivialization; in others it is a geometric winding, a local sign change, or an integrated spectral weight. A plausible implication is that “spectral topological invariant” functions as a family of constructions rather than as a single canonical definition.

2. Fredholm defects, spectral flow, and bulk–edge correspondence

A central operator-theoretic realization appears for spinful, non-interacting electrons on a two-dimensional lattice with a spectral gap at the Fermi energy and fermionic time-reversal symmetry $2$3. On the bulk Hilbert space $2$4, with Fermi projection $2$5 and flux-insertion unitary $2$6, the compressed operator

$2$7

satisfies the $2$8-odd relation $2$9. Its ordinary Fredholm index vanishes, but the mod-$\operatorname{ind}_2 A=\dim\ker A \pmod 2$0 index survives, and the bulk $\operatorname{ind}_2 A=\dim\ker A \pmod 2$1 invariant is

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$2

For a compatible half-space edge Hamiltonian $\operatorname{ind}_2 A=\dim\ker A \pmod 2$3 on $\operatorname{ind}_2 A=\dim\ker A \pmod 2$4, the paper defines an edge Fredholm operator $\operatorname{ind}_2 A=\dim\ker A \pmod 2$5, with edge invariant $\operatorname{ind}_2 A=\dim\ker A \pmod 2$6, and proves the bulk–edge correspondence

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$7

The proof is homotopy-theoretic and relies on locality near the boundary, together with a Fredholm criterion for the family $\operatorname{ind}_2 A=\dim\ker A \pmod 2$8 when $\operatorname{ind}_2 A=\dim\ker A \pmod 2$9 is confined in the second direction. The same work also gives a local trace formula

$w$0

showing that the invariant is encoded in the spectral defect from unitarity (Fonseca et al., 2019).

A complementary analytic description uses spectral flow. For the Fu–Kane spin pump, the spectral flow of a one-parameter family $w$1 of $w$2-dimensional Dirac operators with APS boundary condition is the integer counting net crossings through zero. The paper identifies the Kane–Mele invariant with this spectral flow modulo $w$3,

$w$4

and states that the parity of the number of Kramers pairs of gapless edge states equals the parity of the spectral flow. The reduced eta-invariant supplies the APS index-theoretic bridge between momentum-space $w$5 data and real-space edge crossings (Yu et al., 2016).

The same logic extends beyond topological-insulator Dirac families. For self-adjoint Morse–Sturm systems, a determinant map $w$6 on a complexified parameter rectangle $w$7 defines a Brouwer-degree invariant

$w$8

and the main spectral-flow formula is

$w$9

This converts a spectral crossing count into a topological degree of a determinant map and relates it to Hill-type determinant formulas (Portaluri et al., 2020).

These formulations show that the invariant need not be an observable such as conductance. In the TRI case, the ordinary index vanishes identically, yet a parity-valued spectral obstruction remains.

3. Asymptotic phases, essential spectrum, and one-dimensional classification

For one-dimensional strictly local operators on $C_{x,\bar\epsilon}$0, the cited work isolates two spectral/topological invariants that are stable under compact perturbations: the Fredholm index and the essential spectrum. A strictly local matrix entry has the form

$C_{x,\bar\epsilon}$1

and in the two-phase case the coefficients admit limits at $C_{x,\bar\epsilon}$2 and $C_{x,\bar\epsilon}$3. The resulting operator is a compact perturbation of the direct sum of its asymptotic bulks,

$C_{x,\bar\epsilon}$4

The paper proves that $C_{x,\bar\epsilon}$5 is Fredholm exactly when

$C_{x,\bar\epsilon}$6

and then

$C_{x,\bar\epsilon}$7

The essential spectrum is likewise asymptotic: $C_{x,\bar\epsilon}$8 The topological content is therefore determined entirely by the mismatch of asymptotic phases, even when the operator varies arbitrarily in the interior (Tanaka, 2020).

The split-step quantum walk provides a concrete realization. For the chiral pair $C_{x,\bar\epsilon}$9 with evolution $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$0, the Fredholm condition is

$I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$1

and the Witten index can take values in $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$2. The same asymptotic inequality also controls the spectral gap at $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$3: $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$4 In this setting, the essential spectrum is not merely background spectral data; it is part of the topological classification.

This asymptotic framework clarifies a recurrent principle: spectral topology can be encoded at infinity. Bulk inhomogeneity does not destroy the invariant so long as the asymptotic phases are well defined and the relevant compact-perturbation class is preserved.

4. Spectral localizers, dense spectra, and mobility gaps

When band-gap methods fail, the spectral localizer furnishes a real-space invariant built from position and Hamiltonian data. For a one-dimensional time-quasiperiodically driven class-D superconductor, the frequency-domain Hamiltonian $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$5 has an effectively dense quasienergy spectrum, so conventional gap-based bulk invariants are inadequate. The localizer is

$I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$6

and at particle-hole-symmetric quasienergies $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$7 the invariant is the sign of the determinant of the real matrix

$I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$8

A jump in $I_{\mathrm{SL}}=\frac12\,\operatorname{Sig}(L)$9 as $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$0 is swept from the boundary into the bulk signals a protected Majorana mode at quasienergy $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$1. The paper interprets $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$2 as a PT-symmetric non-Hermitian matrix and requires $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$3, where $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$4 is the exceptional-point threshold beyond which topological equivalence fails (Qi et al., 2024).

A subsequent work shows that the spectral localizer index is explicitly equal to standard local real-space markers in the small-$$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$5 regime. For even $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$6,

$$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$7

while for odd $$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$8,

$$\int d\omega\,\mathrm{Re}(\sigma_{xx}+\sigma_{yy})=C\Delta e^2$$9

The invariant

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$0

is proved to satisfy

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$1

The derivation is a perturbative expansion in $\operatorname{ind}_2 A=\dim\ker A \pmod 2$2, and the Clifford algebra forces the local Chern or winding marker to appear at leading nontrivial order (Jezequel et al., 31 Jul 2025).

The same method extends to strong disorder and mobility gaps. For random Altland–Zirnbauer Hamiltonians with a fractional moments bound, the finite-volume localizer

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$3

provides a computable representative of the strong topological invariant. The paper proves continuity of the probability distribution of the invariant along averaged-continuous homotopies preserving a mobility gap; for ergodic families, the almost sure invariant is therefore constant on such paths. It also proves that if two mobility-gapped bulks have different strong invariants, then an interface between them cannot remain mobility-gapped and must support delocalized interface states (Stoiber, 2024).

Taken together, these works move the subject from momentum-space band topology toward finite-volume, local, and disorder-compatible formulations.

5. Non-Hermitian spectral winding and direct experimental access

In non-Hermitian systems, a spectral topological invariant can be carried by the complex eigenvalues themselves rather than by eigenstates. For a two-level non-Hermitian system with complex eigenenergies $\operatorname{ind}_2 A=\dim\ker A \pmod 2$4, the invariant studied in exciton polaritons is

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$5

It is nonzero only when the loop encloses an exceptional point, and in the cited experiment one finds the half-integer values

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$6

The half-integer charge reflects the square-root branch structure of $\operatorname{ind}_2 A=\dim\ker A \pmod 2$7. The same work emphasizes that eigenstate winding and spectral winding are not equivalent: eigenstate topology can change under weak chiral perturbations, while the spectral winding around exceptional points remains half-integer and stable until the exceptional points annihilate (Su et al., 2020).

The direct measurability of spectral invariants is illustrated in photonics. In a 12-core topological photonic crystal fiber realizing an SSH chain, the winding number $\operatorname{ind}_2 A=\dim\ker A \pmod 2$8 is inferred from the wavelength-resolved weighted intensity difference

$\operatorname{ind}_2 A=\dim\ker A \pmod 2$9

with

$$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$0

when spectral averaging suppresses the oscillatory correction. The method requires a sufficiently broad input spectrum; the relevant criterion is $$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$1, with $$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$2 about $$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$3 in the experiment. The paper reports that the extraction remains reliable for broad spectra and becomes unreliable below about $$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$4 RMS width in the experiment, while also remaining robust to asymmetric spectral distributions up to a point (Roberts et al., 2023).

A distinct but related route uses integrated optical response. For dispersionless flat Chern bands, the cited sum rule is

$$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$5

For dispersive Chern bands, the rule is governed by the quantum metric and the Luttinger invariant rather than by a pure Chern-number quantization. The paper is explicit that this is not a DC Hall statement: the relevant object is the frequency integral of the longitudinal conductivity, and the invariant arises from spectral sum rules rather than from zero-frequency transport (Kruchkov et al., 2023).

A common misconception is that topological invariants must be read from eigenstate textures or Hall plateaus. These examples show that the topology may instead reside in complex-energy phase winding, broadband output spectra, or integrated optical spectral weight.

6. Geometric, contact, and Floer-theoretic extensions

Outside band theory, spectral topological invariants appear in moduli-space geometry. For any regular spectral curve

$$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$6

the cited work constructs a cohomology class $$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$7 on the moduli space $$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$8 of $$\operatorname{ind}(A)=\operatorname{wn}(\det\hat A(\cdot,+\infty))-\operatorname{wn}(\det\hat A(\cdot,-\infty))$$9-colored stable curves such that the Eynard–Orantin recursion correlators are intersection numbers: $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$0 The construction uses local branchpoint data and Laplace transforms of $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$1 and $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$2, and the paper identifies the Laplace transform as the mirror map between the B-model spectral curve and the A-model intersection theory (Eynard, 2011).

In stable homotopy theory, the $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$3-invariant is a tertiary spectral invariant. The analytic side is a tertiary index built from elliptic genus Dirac operators and eta-type boundary corrections,

$$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$4

while the homotopy-theoretic side is Laures’ $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$5-invariant on $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$6. The cited tertiary index theorem states that the analytic and topological constructions agree, thereby relating spectral asymmetry of elliptic operators to second-stage Adams–Novikov data (Bunke et al., 2008).

On CR Seifert manifolds with transverse circle action, contact analytic torsion and eta-like invariants admit simultaneous spectral, topological, and dynamical interpretations. The torsion heat trace

$$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$7

can be rewritten both as an index series over Fourier modes of the Reeb flow and as a Selberg-type sum over closed Reeb orbits weighted by holonomy. The contact analytic torsion $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$8 and the contact eta invariant are thus determined not only by operator spectra but also by orbifold characteristic classes and periodic-orbit data (Rumin, 2024).

Floer theory supplies another established usage. In Hamiltonian Floer homology, the spectral invariant of a class $$w=-\frac{1}{2\pi}\oint_C \nabla_{\mathbf k}\arg[E_+(\mathbf k)-E_-(\mathbf k)]\cdot d\mathbf k$$9 is the min–max action level

$2$00

and the spectral norm is $2$01. The coefficient ring matters sharply: for $2$02 with $2$03, the cited work proves

$2$04

whereas over every field $2$05,

$2$06

In gauged Floer theory, vortex spectral invariants

$2$07

satisfy spectrality, Lipschitz continuity, monotonicity, a triangle inequality, and Hamiltonian invariance, and they are used to define partial quasi-morphisms and quasi-states on the symplectic quotient (Kawamoto et al., 2023, Wu et al., 2015).

These extensions show that the adjective “spectral” does not force a single analytic template. The invariant may be encoded in operator spectra, eta corrections, recursion kernels, or filtered action values.

7. Terminological breadth and major caveats

The term also appears in settings where “spectral” refers to a spectral sequence rather than to the spectrum of an operator. In crystalline band theory, the Atiyah–Hirzebruch spectral sequence organizes irreducible representations at high-symmetry cells, compatibility relations, topological gapless states, and singular points through the pages $2$08. In that framework, $2$09 is the compatibility relation, higher differentials encode higher-dimensional obstructions, and $2$10 yields graded pieces of the full equivariant $2$11-theory classification. Here the topological invariants are spectral-sequence outputs rather than spectral defects of a Hamiltonian spectrum (Shiozaki et al., 2018).

A further caveat concerns highly specialized frameworks that present a “spectral topological invariant” as a complete classifier of geometry. One cited paper claims that the deformation spectrum of a custom self-adjoint operator $2$12, together with a flow $2$13, is a complete invariant of smooth geometry for simply-connected closed manifolds, with the limiting condition $2$14 forcing $2$15. The same source explicitly notes that this is “not a conventional theorem of spectral geometry as usually understood,” that it depends on a paper-specific operator and flow, and that its claims are restricted to that internal framework and its rigidity assumptions (Alexa, 4 Aug 2025).

This breadth has two consequences. First, the phrase “spectral topological invariant” is best read contextually: in one paper it may mean a mod-$2$16 Fredholm kernel parity, in another a half-integer exceptional-point vorticity, in another a spectral-sequence obstruction, and in another a tertiary index class. Second, robustness is always conditional. The invariant is only as stable as the preserved structure—spectral gap, mobility gap, asymptotic phase, symmetry, localizer gap, branchpoint regularity, or filtered continuation—on which its definition is built.