Vector-Valued Topological Invariants

• Vector-valued topological invariants are defined as assignments of structured outputs—such as cohomology classes, diagram algebras, or vector spaces—that capture intricate geometric and homotopical data.
• They unify various approaches including equivariant characteristic classes, obstruction theories, and finite-type invariants, offering precise classification tools in low-dimensional and quantum settings.
• Applications span classifying ‘Quaternionic’ vector bundles in topological insulators, detecting obstructions in continuous fields of Cuntz algebras, and constructing diagram invariants for 3-manifolds.

A vector-valued topological invariant is a topological invariant whose values lie not in a single scalar set such as $\mathbb Z$ or $\mathbb Z_2$, but in a structured target such as a cohomology group, a generalized cohomology set, or a finite-dimensional vector space. In the contemporary literature, this terminology encompasses several distinct but related patterns: characteristic classes of equivariant vector bundles valued in relative equivariant cohomology, obstruction classes for continuous fields valued in generalized cohomology, and manifold invariants valued in diagram algebras. A recurrent theme is that the richer codomain records geometric or homotopical data that cannot be compressed to a single number without loss of structure (Nittis et al., 2022, Sogabe, 2020, Shimizu, 2013).

1. Conceptual form and codomains

A topological invariant is vector-valued in the strict sense when its output is an element of a vector space or of a structured algebraic object that behaves like a “multi-component” receptacle for topology. One explicit instance is the invariant of rational homology $3$-spheres

$\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$

where $\mathcal A_n(\emptyset)$ is the real vector space of degree-$n$ Jacobi diagrams modulo the AS and IHX relations (Shimizu, 2013). Another is the obstruction for continuous fields of the Cuntz algebra, which takes values in Dadarlat–Pennig’s generalized cohomology set

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$

rather than in ordinary cohomology (Sogabe, 2020).

A broader usage occurs when the fundamental geometric datum is itself a vector field. In the velocity-field approach to quantum systems, the primary object is the Bloch-electron velocity field

$$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$

or, after setting $\hbar=1$,

$$\mathbf{v}_n(\mathbf{k},\lambda)=\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$

and the associated topological information is extracted from its zero modes and their indices (Fan et al., 2024, Fan et al., 2024). This is not vector-valued in the same codomain sense as diagram-valued or cohomology-valued invariants, but it is a vector-field-based topological invariant in the sense adopted in that literature.

The codomain is often decisive for interpretation. Relative equivariant cohomology groups support pullback, exact sequences, and characteristic-class formalisms; generalized cohomology carries obstruction-theoretic meaning; diagram vector spaces encode finite-type information. This suggests that “vector-valued” should be understood as a family of rigorously different constructions unified by non-scalar output and by the preservation of topological information under the relevant equivalence relation.

2. Equivariant characteristic classes for “Quaternionic” vector bundles

A central example is the FKMM invariant for “Quaternionic” vector bundles over involutive spaces. If $\mathbb Z_2$0 carries an involution and one considers odd time-reversal symmetry with $\mathbb Z_2$1, then the associated spectral bundles are “Quaternionic” vector bundles, the bundle-theoretic model for class AII systems (Nittis et al., 2022). In the low-dimensional range, the invariant is a cohomological map

$\mathbb Z_2$2

with target the second Borel equivariant cohomology of the pair $\mathbb Z_2$3 with local coefficients $\mathbb Z_2$4 (Nittis et al., 2022).

The conceptual advance is that this invariant is the fundamental characteristic class for classifying “Quaternionic” vector bundles in dimension less than, or equal to three. The paper establishes that

$\mathbb Z_2$5

is a bijection for any $\mathbb Z_2$6-CW complex $\mathbb Z_2$7 with $\mathbb Z_2$8 (Nittis et al., 2022). In that range, the FKMM invariant is simultaneously an obstruction class, a characteristic class, and a complete classification label.

The cohomological interpretation had already been developed in a more general form for involutive spaces with free involution or with a non-finite fixed-point set. For a “Quaternionic” vector bundle $\mathbb Z_2$9 over $3$0, the determinant line bundle carries induced equivariant structure, and the generalized FKMM invariant is defined as

$3$1

through the determinant construction and the canonical Real trivialization over the fixed-point set (Nittis et al., 2016). In this formulation, the invariant is invariant under isomorphism, natural under pullback, vanishes on $3$2-trivial bundles, and is additive under Whitney sum: $3$3

A further structural feature is universality. The tautological $3$4-bundle over the quaternionic Grassmannian defines a universal class

$3$5

and for any classified bundle one has pullback along the classifying map (Nittis et al., 2016). In the later low-dimensional classification theorem, the universal FKMM class also refines ordinary Chern-theoretic data through the relation

$3$6

which identifies the invariant as a genuine refinement of the first Chern class in the presence of involution (Nittis et al., 2022).

The decisive methodological point is the reinterpretation through Bredon equivariant cohomology. The bridge theorem

$3$7

makes the FKMM invariant accessible to equivariant homotopy and to the comparison of the classifying space of “Quaternionic” bundles with a $3$8-equivariant Eilenberg–Mac Lane space (Nittis et al., 2022). This gives the invariant the status of a natural equivariant characteristic class rather than an ad hoc $3$9-quantity from topological-insulator physics.

3. Generalized cohomology obstructions for continuous fields of Cuntz algebras

A second major class of vector-valued invariants appears in operator algebras. For a continuous field of the Cuntz algebra $\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$0 over a finite CW complex $\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$1, a topological invariant is defined as

$\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$2

with target the Dadarlat–Pennig generalized cohomology set

$\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$3

In the case relevant to the paper, $\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$4 and the target is

$\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$5

The invariant is therefore not an ordinary cohomology class but a class in a generalized cohomology theory built from automorphism groups of stabilized strongly self-absorbing $\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$6-algebras (Sogabe, 2020).

Its construction uses the Cuntz–Toeplitz extension

$\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$7

and, after tensoring with $\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$8,

$\widetilde z_n(Y)\in \mathcal A_n(\emptyset),$9

A key technical input is that the quotient map

$\mathcal A_n(\emptyset)$0

is a weak homotopy equivalence, allowing homotopical information to be transferred to the stabilized ideal, where Dadarlat–Pennig theory applies (Sogabe, 2020).

The main theorem identifies the invariant as a precise obstruction to classical origin. For $\mathcal A_n(\emptyset)$1, the class vanishes exactly when the continuous field comes from a vector bundle via Pimsner’s construction: $\mathcal A_n(\emptyset)$2 Thus the invariant is a cohomological obstruction to being of vector-bundle origin (Sogabe, 2020). In that sense it is analogous to a Dixmier–Douady-type class, but its natural codomain is generalized cohomology rather than $\mathcal A_n(\emptyset)$3.

In low-dimensional situations the invariant reduces to classical obstruction theory. If $\mathcal A_n(\emptyset)$4, then

$\mathcal A_n(\emptyset)$5

and the invariant becomes the Bockstein map

$\mathcal A_n(\emptyset)$6

This low-dimensional reduction shows how a generalized, non-scalar invariant can recover more familiar cohomological obstructions in a specific range (Sogabe, 2020).

4. Diagram-valued invariants of $\mathcal A_n(\emptyset)$7-manifolds via vector fields

A particularly literal form of vector-valued topological invariant occurs in the theory of rational homology $\mathcal A_n(\emptyset)$8-spheres. For such a manifold $\mathcal A_n(\emptyset)$9, the invariant

$n$0

takes values in the real vector space of degree-$n$1 Jacobi diagrams modulo the AS and IHX relations (Shimizu, 2013). Here the target is a finite-dimensional diagram algebra, so “vector-valued” is exact.

The construction starts from a family of admissible vector fields

$n$2

on $n$3 and defines

$n$4

where the principal term is given by configuration-space integrals and the anomaly term corrects the dependence on auxiliary choices (Shimizu, 2013). The principal term has the form

$n$5

The geometric mechanism is mediated by admissible vector fields, their normalized graphs in the unit sphere bundle, and propagators on the two-point compactified configuration space. The role of the correction term is to ensure that the resulting class is independent of all such auxiliary data. The main theorem is precisely that the difference

$n$6

does not depend on the admissible family $n$7, and hence yields a topological invariant of $n$8 (Shimizu, 2013).

This construction also clarifies how vector-valued invariants can unify previously distinct formalisms. The invariant generalizes both the Kontsevich–Kuperberg–Thurston invariant and Watanabe’s Morse homotopy invariant, proving the equivalence of these two invariants in the rational homology $n$9-sphere setting (Shimizu, 2013). A plausible implication is that vector-valued targets such as diagram spaces are particularly effective when the topology being recorded is finer than a single numerical degree but still amenable to finite-type organization.

5. Velocity-field topological invariants in quantum systems

In quantum two-band models, a different usage centers on the velocity field as the geometric carrier of topological information. For a Bloch Hamiltonian

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$0

the Bloch-electron velocity is defined by

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$1

and, for a two-level Hamiltonian

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$2

it is written locally on the manifold as

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$3

The zero modes

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$4

are treated as effective topological charges or defects, and their local indices are determined from the sign of the Jacobian determinant (Fan et al., 2024, Fan et al., 2024).

The global invariant is the sum of these local indices: $$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$5 in accordance with the Poincaré–Hopf theorem (Fan et al., 2024). In this framework, the Euler characteristic of the associated compact orientable manifold is obtained from the zero-mode structure of the velocity field, provided the zero modes are isolated and the basis correspondence between the Brillouin zone and the manifold is compatible (Fan et al., 2024).

The quantum sphere and quantum torus models supply explicit demonstrations. For the sphere model, the total index is

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$6

matching $$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$7 (Fan et al., 2024). For the torus model, after accounting for boundary and corner multiplicities, the total index is

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$8

or equivalently

$$E_D^1(X)=[X,B\!\operatorname{Aut}(\mathbb K\otimes D)],$$9

matching $$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$0 (Fan et al., 2024, Fan et al., 2024).

These papers emphasize that the velocity-field invariant is different from the Chern number. The former is built from zeros of $$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$1, captures the global index of the flow, and equals the Euler characteristic of the submanifold; the latter is a homotopy invariant associated with the map $$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$2 and with Berry curvature, and changes when the energy gap closes (Fan et al., 2024). Thus the velocity-field construction is topological, but its output is ultimately a scalar Euler characteristic or winding number rather than an element of a cohomology group or vector space. This distinction is essential to avoid conflating vector-field-based invariants with vector-valued codomain invariants.

6. Boundaries, misconceptions, and adjacent notions of invariance

The phrase “vector-valued topological invariant” can be misread in at least two ways. First, not every invariant attached to vector-valued or operator-valued objects is topological. In the study of Brownian shifts on vector-valued Hardy spaces, the main classification uses the Type I / Type II dichotomy, the inner function $$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$3, the canonical decomposition

$$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$4

the multiplicity spaces, and the angle $$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$5, but the paper explicitly does not define a topological invariant like an index, winding number, or vector bundle invariant (Das et al., 29 Jul 2025). The invariants there are unitary equivalence invariants, not topological ones.

Second, some papers use the word “invariant” for rigidity properties of embeddings rather than for a classifying label of spaces or bundles. A topological group has invariant linear span if all linear spans of the group under arbitrary embeddings into topological vector spaces are isomorphic as topological vector spaces (Pernecká et al., 2020). This is an invariant property of a group, but not a vector-valued topological invariant in the characteristic-class or obstruction-theoretic sense.

A related adjacent case arises in the theory of perturbed Toeplitz operators in vector-valued Hardy spaces. There the decisive structure is nearly $$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$6-invariance with finite defect, and kernels are represented through defect spaces and backward-shift-invariant parameter spaces (Chattopadhyay et al., 2020). This is invariant-subspace theory rather than topology in the sense of cohomological or manifold invariants.

Taken together, these contrasts delimit the term. In the strongest sense, a vector-valued topological invariant assigns to a topological object an element of a structured codomain that is functorial or obstruction-theoretic and remains unchanged under the relevant equivalence. The FKMM class in relative equivariant cohomology, the Dadarlat–Pennig obstruction class for Cuntz fields, and the Jacobi-diagram invariant of rational homology $$\mathbf{v}_n(\mathbf{k},\lambda)=\frac{1}{\hbar}\nabla_{\mathbf{k}}E_n(\mathbf{k},\lambda),$$7-spheres exemplify this strict form (Nittis et al., 2022, Sogabe, 2020, Shimizu, 2013). The velocity-field literature represents a looser but still mathematically substantive usage, where the invariant is extracted from a vector field and its zero-mode topology rather than taking values in a vector space (Fan et al., 2024, Fan et al., 2024).