The families analytic index for 1|1-dimensional Euclidean field theories

A trace class representation of {\sf sPath}_0(M) of degree n is the data of a degree n representation E of {\sf sPath}_0(M) and a section (Z(E),Z(E)_\ell)\in \Gamma(\mathcal{L}^{1|1}_0(M);{\sf{Pf}}^{\otimes -n}) satisfying the property

where {\sf Tr}_{{\mathbb{C}}{\rm l}_n} denotes the Clifford super trace.

For a trace class degree n representation, Z(E)\in \Omega_{\rm cl}^\bullet(M;C^\infty({\mathbb{R}}_{>0})) is a closed differential form in degrees equal to the parity of n (i.e., either even or odd degree). When E is a real field theory, Z(E) is a real differential form. When E is real and reflection positive, Z(E) is real and concentrated in degrees equal to n mod 4.

For the first, note that the element \Gamma_n\in {\rm Cl}_n\subset {\mathbb{C}}{\rm l}_n has parity equal to the parity of n, and so the endomorphism \Gamma_n\circ \rho_+ also has parity n mod 2. Hence, it follows for any degree n representation for which the trace (?) is defined, the resulting function is even or odd according to this parity. The real structure preserves \Gamma_n\in {\mathbb{C}}{\rm l}_{n}, and the reality condition (?) implies that Z(E) be equal to its conjugate. Since \Gamma_n\in {\rm Cl}_n determines a real trivialization (see (?)), this amounts to demanding that Z(E)\in C^\infty({\mathbb{R}}_{>0}\times {\sf Map}({\mathbb{R}}^{0|1},M))\simeq \Omega^\bullet(M;C^\infty({\mathbb{R}}_{>0})) is real as a differential form, for the standard conjugation action on {\mathbb{C}}-valued differential forms. When E is also reflection positive, we observe that \Gamma_n\mapsto i^n\Gamma_n under the *-superalgebra structure (6.29). This imposes an additional invariance property on Z(E), implying that it takes values in 4-periodic differential forms.

Assuming Hypothesis 5.9, a choice of orientation on {\mathbb{R}}^n determines an isomorphism {\sf{Pf}}^{\otimes n} \simeq \mathscr{T}^{\otimes n}(\mathcal{L}^{1|1}(M)/\!\!/ ({\mathbb{R}}^{1|1}\rtimes {\mathbb{Z}}/2)) between the Pfaffian line bundle and the valued of the degree n twist on \mathcal{L}^{1|1}(M)/\!\!/ ({\mathbb{R}}^{1|1}\rtimes {\mathbb{Z}}/2). In particular, a degree n field theory determines a section of the dual of {\sf{Pf}}^{\otimes n}.

It suffices to take M={\rm pt}. Then the value of the twist on \mathcal{L}^{1|1}({\rm pt}) is the line bundle with fiber

A choice of orientation \Gamma_n\in \Lambda^{\rm top}({\mathbb{C}}^n)\simeq {\mathbb{C}}{\rm l}_n/[{\mathbb{C}}{\rm l}_n,{\mathbb{C}}{\rm l}_n] trivializes this line, and so after this choice it remains to compute the equivariance structure. Under Hypothesis 5.9, the {\mathbb{R}}^{1|1}-action is trivial, and the {\mathbb{Z}}/2-action is by parity. This completes the proof.

Assuming Hypothesis 5.9, the restriction of a degree n field theory along (?) determines a trace class degree n representation of {\sf sPath}_0(M).

The argument is similar to [ST11, Lemma 3.20]. In the terminology of [STTraces], the morphisms in the image of the composition

are thick: this follows from Proposition 4.15. The categorical trace of a thick morphism is an endomorphism of the monoidal unit. Proposition 4.18 shows that this categorical trace is the image of the composition

Applying a degree n field theory to these families, Corollary 5.11 shows that the categorical trace of the family (?) is sent to the composition

The element \Gamma_n provides an identification {\mathbb{C}}{\rm l}_n\otimes_{{\mathbb{C}}{\rm l}_n\otimes {\mathbb{C}}{\rm l}_n} {\mathbb{C}}{\rm l}_n\simeq \underline{{\mathbb{C}}} with the trivial line bundle, and the above composition becomes the Clifford super trace. Finally, using that nuclear Fréchet spaces have the approximation property [Schaefer, page 109], we find that the categorical trace above is independent of the choice of thickener [STTraces, Theorem 1.7], completing the proof.

Let E denote a degree n representation of {\sf sPath}_0(M), and let {\mathbb{A}}_\pm denote the superconnections afforded by Lemma 4.24. Then E is infinitesimally generated if we have the equality of differential forms

compare (?).

An infinitesimally generated representation admits smooth energy cutoffs if furthermore the superconnections {\mathbb{A}}_\pm admits smooth index bundles in the sense of Definition ?.

We comment on the conditions imposed in Definition ?. Although one expects a field theory to be determined by infinitesimal data (implying the condition (?)), this is not automatic in the formalism [ST11] due to the failure of existence and uniqueness of solutions to differential equations in topological vector spaces; see Remark 1.5.

The existence of smooth index bundles is more subtle. For a family of Dirac operators, the existence of an index bundle follows from ellipticity, using that eigensections of the Dirac Laplacian are smooth, e.g., see [BGV, Proposition 9.10] or [LM, III, Theorem 5.8]. As such, the existence of smooth cutoffs can be thought of as an ellipticity property for field theories. In higher dimensions, one can formulate similar smooth cutoff conditions for subspaces \mathcal{H}^{\vec\lambda}\subset \mathcal{H} indexed by both energies and momenta, with the number of cutoff indices equal to the dimension of the field theory.

There is an equivalence of categories

from the category of real, {\rm Cl}_n-linear self-adjoint superconnections admitting smooth index bundles to the category 1|1\hbox{-{\sf eft}}^n(M).

For a {\rm Cl}_n-linear superconnection {\mathbb{A}} on a real super vector bundle V\to M, consider the representation of {\sf sPath}_0(M) defined on the vector bundles with sections

and super semigroup actions given by

A pairing for this representation is inherited from the hermitian pairing \overline{V}_{\mathbb{C}}\otimes V_{\mathbb{C}}\to \underline{{\mathbb{C}}} built in the usual way from the underlying metric on V. The real and reflection structures are evident from the construction, as is the reflection positivity condition. The trace structure uses that \exp(-t{\mathbb{A}}^2)\in \Omega^\bullet(M;{\rm End}(V_{\mathbb{C}})) is trace class (from the existence of smooth index bundles), and Z_\ell is determined by the Chern–Simons form for the 1-parameter family of superconnections gotten from Getzler rescaling of {\mathbb{A}}. Existence of cutoffs follows by assumption. This constructs the functor {\rm sPar} in (1.4) on objects. The construction on morphisms is straightforward: using ?, isomorphisms between superconnections are sent to isomorphisms between representations of {\sf sPath}_0(M). The functor is full and faithful since isomorphisms of representations of {\sf sPath}_0(M) are determined by isomorphisms of vector bundles. Finally, Proposition 1.10 verifies that {\rm sPar} is essentially surjective, and hence an equivalence.

The groupoid of self-adjoint {\rm Cl}_n-linear superconnections on finite-rank real vector bundles admits a functor {\rm sPar}

determined by super parallel transport along nearly constant superpaths.

This follows immediately from Proposition 1.4 by the existence of smooth index bundles in finite dimensions, e.g., see [BGV, §9.1].

Given E\in 1|1\hbox{-{\sf eft}}^n(M) and \lambda\in {\mathbb{R}}_{>0}, the \lambda-cutoff theory E_{<\lambda}\in 1|1\hbox{-{\sf eft}}^n(U_\lambda) is defined as the functor (?) applied to the super vector bundle \mathcal{H}^{<\lambda}\to U_\lambda with Clifford linear self-adjoint connection \nabla^{\mathcal{H}^{<\lambda}}, i.e.,

in the notation of Corollary ?.

The partition function of E and the partition function of the \lambda-cutoff theory are related by

This follows from the construction of the \eta-forms (?).

The existence of cutoffs is preserved under the renormalization group flow: for E\in 1|1\hbox{-{\sf eft}}^n(M), the \lambda-cutoff of {\sf RG}^\mu(E) is canonically isomorphic to the \mu\lambda-cutoff of E

Using Proposition ?, the description of the {\sf RG}-action in terms of Getzler rescaling allows us to analyze cutoffs. Unraveling Definition 1.11, the \mu\lambda-cutoff for E is the same data as the \lambda-cutoff for {\sf RG}^\mu(E): the cutoff bundles are the same subspace of \mathcal{H}\to U_\lambda with the same connection (note {\mathbb{A}}_1 is invariant under the {\sf RG}-flow). Hence, there is indeed a canonical isomorphism (?), and furthermore {\sf RG}^\mu(E) satisfies properties (1) and (2) of Definition 1.11. Property (3) follows from the Getzler rescaling description of the {\sf RG}-action, where the partition function of {\sf RG}^\mu(E) is computed by the Getzler-rescaled superconnection. Finally, by a change of variables one identifies the integral (?) for E and \lambda\mu with the integral for {\sf RG}^\mu (E) and \lambda. This completes the proof.

An object E\in 1|1\hbox{-{\sf eft}}^n(M) flows to zero under the renormalization group flow if for all \lambda\in {\mathbb{R}}_{>0} and compact submanifolds K\subset M, there exists \mu\in {\mathbb{R}}_{>0} such that {\sf RG}^\mu(E)|_K has the zero theory as its \lambda-cutoff, i.e., K\subset U_\lambda and \mathcal{H}^{<\lambda}|_K=\{0\}.

An object E\in 1|1\hbox{-{\sf eft}}^n(M) flows to zero under the renormalization group flow if and only if the degree zero component of its associated superconnection {\mathbb{A}}_{0} is invertible.

It suffices to prove the lemma when M is compact. Let \nu\in {\mathbb{R}}_{>0} be the minimum eigenvalue of {\mathbb{A}}_{0}^2. If \nu>0 (i.e., {\mathbb{A}}_{0} is invertible), for any \lambda we may choose some \mu\gg 0 such that \mu\nu>\lambda. By Lemma ? the \lambda-cutoff theory of {\sf RG}^\mu(E) is the zero theory. Conversely, if there exists some \mu such that {\sf RG}^\mu(E) has the zero bundle as its cutoff bundle, then \mu\nu must be larger than \lambda, and hence \nu\ne 0 and {\mathbb{A}}_0 is invertible.

For E\in 1|1\hbox{-{\sf eft}}^n(M), choose a discrete subset \Lambda\subset {\mathbb{R}}_{>0} such that \{U_\lambda\}_{\lambda\in \Lambda} is a locally finite cover. Then define the differential cocycle

in the description of differential {\rm KO}-theory from [DBEBis, §7].

The index map (?) applied to the self-adjoint superconnection underlying E\in 1|1\hbox{-{\sf eft}}^n(M) gives a map of the desired type, but a priori depends on a choice of subset \Lambda\subset {\mathbb{R}}_{>0} determining a choice of locally finite subcover \{U_\lambda\}_{\lambda\in \Lambda}. To show that the differential KO-class is independent of this choice, one applies the same argument that proves the differential families index is well-defined [DBEBis, §8.3]. To review this argument, for a different choice \Lambda'\subset {\mathbb{R}}_{>0}, we can compare the resulting {\rm Cl}_{-n}-bundles on the mutual refinement \{U_\lambda\}_{\lambda\in (\Lambda\bigcup \Lambda')}. For \lambda<\lambda', we obtain inclusions (?) of {\rm Cl}_{-n}-bundles with commuting {\rm Cl}_{-1}-actions on the orthogonal complement to the inclusion. This gives the data of a stable equivalence of {\rm Cl}_{-n}-bundles. By [DBEBis, Lemma 5.9], the {\rm Cl}_{-n}-bundles are therefore concordant and the differential KO-classes agree. This shows that (?) is independent of the choice of \Lambda, and the index map is well-defined.

By Lemma ?, objects in 1|1\hbox{-{\sf eft}}^n(M) that are sent to zero under the RG flow are also sent to the zero class under the composition

since the degree zero part of their associated superconnection is invertible, and hence the underlying KO-class is trivial. This completes the proof.

The composition of the first two arrows

sends a Clifford linear superconnection to the expected differential {\rm KO}-class, and hence the above composition sends a {\rm Cl}_n-linear superconnection {\mathbb{A}} to the differential form {\sf sTr}_{{\rm Cl}_n}(e^{- {\mathbb{A}}^2}) representing its Chern character.

Without loss of generality, we may assume M is compact. Choosing a cutoff valued \lambda larger than the maximum eigenvalue of {\mathbb{A}}_0^2, the index construction sends a vector bundle with Clifford linear superconnection to the associated differential class in \widehat{\rm KO}{}^n(M), and the curvature of this class is as stated above. This completes the proof.

Using Proposition 1.4, for a family \pi\colon X\to M of Riemannian spin manifolds and E\to X a real vector bundle the {\rm Cl}_{-n}-linear superconnection (?) acting on the Fréchet bundle V=\pi_*({\mathbb{S}}\times E)\otimes {\mathbb{C}} determines an object of 1|1\hbox{-{\sf eft}}^{-n}(B) via the representation of positively oriented superpaths,

The pairing and reflection positivity data are inherited from the hermitian pairing on \pi_*({\mathbb{S}}\times E)\otimes {\mathbb{C}}. The real structure comes from complex conjugation on \pi_*({\mathbb{S}}\times E)\otimes {\mathbb{C}}. The standard properties of the Bismut superconnection reviewed in §? imply the existence of a differential index bundle, so we do indeed obtain an object in 1|1\hbox{-{\sf eft}}^{-n}(M). Finally, applying the index map (?) to this object of 1|1\hbox{-{\sf eft}}^{-n}(B) is precisely the (real) family Clifford index [DBEBis, Theorem 1.3]. This completes the proof.

A real structure on a super vector space V is an isomorphism r\colon V\stackrel{\sim}{\to} \overline{V} of super vector spaces such that \overline{r}\circ r={{\rm id}}_V. The real subspace associated with a real structure is the fixed point set V_{\mathbb{R}}\subset V for r.

A hermitian form on a super vector space V is the data of a map \langle-,-\rangle\colon \overline{V}\otimes V\to {\mathbb{C}} making the diagram commute,

or equivalently, satisfying the formula \langle x,y\rangle =(-1)^{|x||y|}\overline{\langle y,x\rangle} for homogeneous elements. A hermitian form is positive if

A real hermitian form on a super vector space with real structure is a hermitian form that when restricted to (V_{\mathbb{R}})^{\rm ev}\subset V takes values in {\mathbb{R}}\subset {\mathbb{C}} and when restricted to (V_{\mathbb{R}})^{\rm odd}\subset V takes values in i^{-1}{\mathbb{R}}\subset {\mathbb{C}}.

Let V be a super vector space equipped with a positive hermitian form \langle-,-\rangle. The super adjoint of a linear map T\colon V\to V is characterized by

Let V be a super vector space with hermitian inner product. Then the super algebra {\rm End}(W) has the structure of a *-superalgebra determined by the super adjoint (?). A real structure on W determines a real structure on {\rm End}(W) for which this *-structure is real.

The Clifford algebras {\mathbb{C}}{\rm l}_n defined in (?) have the structure of *-superalgebras using (6.29). Complex conjugation gives {\mathbb{C}}{\rm l}_n the structure of a real super algebra for which this *-structure is real.

For a super vector space V with positive hermitian form \langle-,-\rangle, the super algebra {\rm End}(V) has the structure of a *-superalgebra via the super adjoint. It has the structure of a {\mathbb{Z}}/2-graded *-algebra via the adjoint with respect to the (ungraded) inner product determined by (?). These two notion of *-structure are related by (?).

In the case of the Clifford algebras {\mathbb{C}}{\rm l}_n, we obtain a {\mathbb{Z}}/2-graded *-algebra via the {\mathbb{C}}-antilinear extension of the map on generators,

This differs from the *-superalgebra structure (6.29) by a factor of i.

A Clifford module is a graded module over a Clifford algebra {\mathbb{C}}{\rm l}_n. The canonical isomorphism {\mathbb{C}}{\rm l}_n^{\sf{op}}\simeq {\mathbb{C}}{\rm l}_{-n} leads to equivalences of categories

which will allow us to translate any Clifford (bi)module into a left Clifford module.

A module V over a *-superalgebra A is self-adjoint if the structure map A\to {\rm End}(V) is a *-homomorphism for a specified hermitian form on V.

When A={\mathbb{C}}{\rm l}_n with the *-superalgebra structure (6.29), the above definition is equivalent to [BGV, Definition 3.3] via the translation to {\mathbb{Z}}/2-graded *-algebras in Examples ? and ?.

Define a left {\mathbb{C}}{\rm l}_{1}\otimes {\mathbb{C}}{\rm l}_{-1}-module whose underlying super vector space is {\mathbb{C}}^{1|1} with action determined by

The assignment (?) induces an isomorphism of super algebras, {\mathbb{C}}{\rm l}_{1}\otimes {\mathbb{C}}{\rm l}_{-1}\simeq {\rm End}({\mathbb{C}}^{1|1}). In other words, {\mathbb{C}}{\rm l}_{1}\otimes {\mathbb{C}}{\rm l}_{-1} is Morita equivalent to {\mathbb{C}}. This {\mathbb{C}}{\rm l}_{1}\otimes {\mathbb{C}}{\rm l}_{-1}-action is self-adjoint for the graded hermitian pairing on {\mathbb{C}}^{1|1} coming from the standard hermitian pairing on {\mathbb{C}}^2 under (?).

Let {\mathbb{R}}^{1|1} denote the supermanifold whose underlying topological space is {\mathbb{R}}, and structure sheaf is the sheaf of sections of \Lambda^\bullet \underline{{\mathbb{C}}}^\vee, where \underline{{\mathbb{C}}} is the trivial complex line bundle on {\mathbb{R}}. Then define {\mathbb{R}}^{1|1}_{\ge 0} and {\mathbb{R}}^{1|1}_{\le 0} by restricting the structure sheaf of {\mathbb{R}}^{1|1} to {\mathbb{R}}_{\ge 0}\subset {\mathbb{R}} and {\mathbb{R}}_{\le 0}\subset{\mathbb{R}}.

For a complex vector bundle E\to N over an ordinary manifold N, let \Pi E denote the supermanifold whose structure sheaf of {\mathbb{Z}}/2-graded algebras is given by

Every supermanifold S is isomorphic to \Pi E for some vector bundle E\to N=S_{\rm red} [Batchelor].

The supermanifold {\mathbb{R}}^{n|m} is the locally ringed space with underlying topological space ({\mathbb{R}}^{n|m})_{\rm red}={\mathbb{R}}^n and structure sheaf the sheaf of sections of \Lambda^\bullet (\underline{{\mathbb{C}}}^m)^\vee, the trivial rank m complex vector bundle on {\mathbb{R}}^n. The S-points of {\mathbb{R}}^{n|m} are

where the reality condition on the f_i is imposed on the reduced manifold of S.

Define {\sf Map}({\mathbb{R}}^{0|1},M) as the presheaf on supermanifolds whose value on S is the set of maps S\times {\mathbb{R}}^{0|1}\to M; this supermanifold is representable {\sf Map}({\mathbb{R}}^{0|1},M)\simeq \Pi TM_{\mathbb{C}}, e.g., see [HKST, Proposition 3.1]. Consider the precomposition action of {\mathbb{R}}^{0|1} on itself by translation,

Differentiation at 0\in {\mathbb{R}}^{0|1} gives an odd vector field on {\sf Map}({\mathbb{R}}^{0|1},M). Under the isomorphism C^\infty({\sf Map}({\mathbb{R}}^{0|1},M))\simeq C^\infty(\Pi TM_{\mathbb{C}})\simeq \Omega^\bullet(M), this vector field is the de Rham differential, acting on forms as an odd derivation.

Given supermanifolds \Sigma and X, define the mapping object

as a generalized supermanifold with the indicated S-points.

Given a {\mathbb{Z}}/2-graded topological vector space {\mathcal{V}}, define a generalized supermanifold by the S-points

where the tensor product is the {\mathbb{Z}}/2-graded projective tensor product.

There is an involution on the category of supermanifolds,

sending a supermanifold to the same locally ringed space but whose sheaf has complex numbers acting through complex conjugation. A real structure on a supermanifold S is an isomorphism r_S\colon S\stackrel{\sim}{\to} \bar S such that \overline{r}\circ r={{\rm id}}_S. A map f\colon S'\to S between supermanifolds with real structure is real if r_S\circ f=f\circ r_{S'}.

For a complex vector bundle E\to N, a real structure on E (as a vector bundle) determines a real structure on \Pi E (as a supermanifold)

Indeed, complex conjugation determines a {\mathbb{C}}-antilinear map C^\infty(\Pi E)\to C^\infty(\Pi E) which provides an isomorphism of supermanifolds \Pi E\stackrel{\sim}{\to} \overline{\Pi E}.

For any vector bundle {\mathcal{V}}\to {\sf Map}({\mathbb{R}}^{0|1},M), there is an isomorphism {\mathcal{V}} \simeq p^*V where V\to M is a vector bundle on M.

Let i\colon M\hookrightarrow {\sf Map}({\mathbb{R}}^{0|1},M) denote the inclusion of the reduced manifold. Given a vector bundle {\mathcal{V}}\to {\sf Map}({\mathbb{R}}^{0|1},M), our candidate for V\to M in the statement is V:=i^*{\mathcal{V}}. Indeed, the composition i\circ p is homotopic to the identity on {\sf Map}({\mathbb{R}}^{0|1},M). For example, consider the homotopy {\mathbb{R}}\times \Pi TM_{\mathbb{C}}\to \Pi TM_{\mathbb{C}} that rescales the odd fibers of \Pi TM_{\mathbb{C}}\to M by a scalar t\in {\mathbb{R}}; in the functor of points this map is (t,x,\psi)\mapsto (x,t\psi) for t\in {\mathbb{R}}(S) and (x,\psi)\in \Pi TM_{\mathbb{C}}(S)\simeq {\sf Map}({\mathbb{R}}^{0|1},M)(S). Now we use that pullbacks along homotopic maps yield isomorphic vector bundles,

concluding that p^*V=p^*i^*{\mathcal{V}}\simeq (i\circ p)^*{\mathcal{V}} is isomorphic to {\mathcal{V}}.

The d=2 case of conjecture (?) requires a generalization of the definitions below to fully-extended 2|1-dimensional Euclidean field theories of degree n. In particular, this requires that one enhance the 2|1-Euclidean bordism category to (some flavor of) bicategory that encodes cutting and gluing of 2|1-Euclidean manifolds with corners. The framework for such a definition is a subject of current investigation, and likely to depart in various ways from prior approaches. Indeed, extending the definitions in [ST11] to bicategories internal to symmetric monoidal stacks seems rather unwieldy. At present, smooth bordism categories inspired by iterated Segal spaces seem more tractable, e.g., see [ClaudiaTheo, GradyPavlov, LudewigStoffel, BEPTFT].

A family of (G,{\mathbb{M}})-supermanifolds is data

1. a smooth submersion p\colon Y\to S;

2. a maximal atlas \{U_i\} of Y equipped with open embeddings \varphi_i \colon U_i \hookrightarrow S \times {\mathbb{M}} over S;

3. transition data g_{ij}\colon p(U_i\bigcap U_j)\to G.

The embeddings \varphi_i are required to be compatible with transition data g_{ij} (using the G-action on {\mathbb{M}}) and the transition data must satisfy a further cocycle condition.

An isometry between (G,{\mathbb{M}})-families is a pullback f\colon Y\to Y' over a base change S\to S' with the property that the restriction of f to the atlases determining the (G,{\mathbb{M}})-structures is given by the action of G on open subspaces of {\mathbb{M}}. The collection of (G,{\mathbb{M}})-supermanifolds and isometries defines a stack (G,{\mathbb{M}})-{\sf SMfld} on the site of supermanifolds; see [ST11, §2.8] and [Powerops, Remark A.23].

A family of (G,{\mathbb{M}})-pairs is a (G,{\mathbb{M}})-family Y\to S with a codimension 1 submanifold Y^c\subset Y over S with the property that in a chart for the (G,{\mathbb{M}})-structure, the image of U_i\bigcap Y^c\subset U_i under \varphi_i lands in the subspace S\times {\mathbb{M}}^c\subset S\times {\mathbb{M}}. Isometries between (G,{\mathbb{M}})-pairs are isometries of (G,{\mathbb{M}})-supermanifolds Y\to Y' such that Y^c \subset Y is sent to Y'^c\subset Y'. This determines a stack on the site of supermanifolds whose objects are (G,{\mathbb{M}})-pairs.

Let {\mathbb{A}}^d be affine d-space, regarded as a Riemannian spin manifold. Its group of spin structure preserving isometries is {\mathbb{R}}^d\rtimes {\rm Spin}(d), where {\rm Spin}(d) acts on {\mathbb{R}}^d through the double cover {\rm Spin}(d)\to {\rm SO}(d). The d-dimensional Euclidean geometry is defined by the triple (G,{\mathbb{M}},{\mathbb{M}}^c)=({\mathbb{R}}^d\rtimes {\rm Spin}(d),{\mathbb{A}}^d,{\mathbb{A}}^{d-1}), where {\mathbb{A}}^{d-1}\subset {\mathbb{A}}^d comes from the standard inclusion of a totally geodesic hyperplane; after a choice of basepoint this is the standard inclusion {\mathbb{R}}^{n-1}\subset {\mathbb{R}}^n. The stack ({\mathbb{R}}^d\rtimes {\rm Spin}(d),{\mathbb{A}}^d)-{\sf SMfld} of d-dimensional Euclidean manifolds has as objects fiber bundles Y\to S of d-manifolds endowed with flat Riemannian metrics and spin structures. The stack of d-dimensional Euclidean pairs has as objects fiber bundles Y\to S of d-manifolds endowed with flat Riemannian metrics and spin structures together with a family of totally geodesic codimension 1 submanifolds Y^c\subset Y over S.

Fix a rigid geometry (G,{\mathbb{M}}). Given a homomorphism g\colon G\to G and a G-equivariant map f\colon {\mathbb{M}}\to {\mathbb{M}} relative to g that preserves the image of the inclusion {\mathbb{M}}^c\subset {\mathbb{M}}, there is a uniquely determined map of stacks

The proof is the same argument as [HST, Lemma 6.19]. Suppose we are given an S-family of (G,{\mathbb{M}})-manifolds Y\to S. We obtain a new (G,{\mathbb{M}})-family f(Y\to S) given by the same fibration Y\to S, but where local charts U_i\subset Y are modified by postcomposition with f,

Because f is G-equivariant relative to g, the gluing data g^{-1}g_{ij}g determines a (G,{\mathbb{M}})-structure on Y. Similarly, because f preserves the image of {\mathbb{M}}^c\subset {\mathbb{M}}, it determines an endofunctor on the category of (G,{\mathbb{M}})-pairs. Since the underlying map on the family Y\to S is the identity map, these endofunctors of (G,{\mathbb{M}})-manifolds determine maps of stacks.

A real structure on a rigid geometry (G,{\mathbb{M}}) is a real structure on {\mathbb{M}}^c, {\mathbb{M}} and G such that all the maps defining the rigid geometry are real in the sense of Definition ?.

A category {\sf C} internal to symmetric monoidal stacks is the data of symmetric monoidal stacks {\rm Ob}({\sf C}) and {\rm Mor}({\sf C}) of objects and morphisms with source, target, unit, and composition functors

along with additional coherence data (e.g., associator and unitor) satisfying compatibility properties.

For {\sf C}, {\sf D} categories internal to symmetric monoidal stacks, an internal functor F=(F_0,F_1,\mu,\varepsilon)\colon {\sf C}\to {\sf D} is defined as morphisms of stacks F_0\colon {\rm Ob}({\sf C})\to {\rm Ob}({\sf D}), F_1\colon {\rm Mor}({\sf C})\to {\rm Mor}({\sf D}), and 2-commuting data,

These data are required to satisfy further properties involving the associators and unitors.

An internal natural transformation (\eta,\rho)\colon F\Rightarrow G between internal functors F,G\colon {\sf C}\to {\sf D} is a morphism of stacks \eta\colon {\rm Ob}({\sf C})\to {\rm Mor}({\sf D}) and 2-commuting data \rho

These data are required to satisfy additional coherence properties. Define an isomorphism between internal natural transformations (\eta,\rho)\Rightarrow (\eta',\rho') as the the data of 2-morphism \eta\Rightarrow \eta' between maps of stacks with respect to which \rho and \rho' are compatible.

A super Lie category is a category internal to supermanifolds whose source and target maps are submersions. For a super Lie category {\sf C}, let {\rm Ob}({\sf C}) and {\rm Mor}({\sf C}) denote the supermanifolds of objects and morphisms, and let

denote the unit, source, target, and composition maps, respectively. A smooth functor F\colon {\sf C}\to {\sf D} is a pair of maps of supermanifolds F_0\colon {\rm Ob}({\sf C})\to {\rm Ob}({\sf D}), F_1\colon {\rm Mor}({\sf C})\to {\rm Mor}({\sf D}) satisfying the axioms for a functor between categories. A smooth natural transformation \eta\colon F\Rightarrow G is a map of supermanifolds \eta\colon {\rm Ob}({\sf C})\to {\rm Mor}({\sf D}) satisfying the axioms for a natural transformation between functors. Super Lie categories, smooth functors, and smooth natural transformations form a strict 2-category.

Any supermanifold determines a discrete super Lie category, i.e., a super Lie category with only identity morphisms.

Categories internal to stacks, internal functors, and internal natural transformations are not as well-behaved as one might hope. For example, an internal functor {\sf C}\to {\sf D} that is (appropriately) essentially surjective and fully faithful need not have an inverse functor. For example, let {\sf C} be the Čech groupoid of an open cover of X and {\sf D} be the (discrete) internal category determined by X; the evident internal functor {\sf C}\to {\sf D} has an inverse internal functor if and only if the cover admits a section. The “correct" framework for categories internal to stacks therefore requires a localization at an appropriate class of weak equivalences, analogous to the localization relating the 2-category of Lie groupoids with geometric stacks [Pronk]. For the manipulations in this paper, it suffices to work with categories internal to stacks before such a localization.

Fix an involution {\sf SMfld}\to {\sf SMfld} on the category of supermanifolds. For a category {\sf C} internal to (symmetric monoidal) stacks on the site of supermanifolds, an internal involution is the data of an internal functor \tau\colon {\sf C} \to {\sf C} covering the fixed involution of supermanifolds, an internal natural isomorphism \eta \colon {{\rm id}}_{\sf C}\Rightarrow \tau\circ\tau, and an isomorphism \delta between internal natural isomorphisms \tau\Rightarrow \tau in the diagram

The isomorphism \delta between internal natural transformations satisfies a compatibility condition with \eta for the quadruple composition \tau\circ\tau\circ\tau\circ\tau.

Let ({\sf C},\tau_{\sf C},\eta_{\sf C},\delta_{\sf C}) and ({\sf D},\tau_{\sf D},\eta_{\sf D},\delta_{\sf D}) be internal categories with internal involution, and let F\colon {\sf C}\to {\sf D} be a functor. Equivariance data for F is an internal natural transformation \phi\colon F\circ \tau_{\sf C}\Rightarrow \tau_{\sf D}\circ F, i.e., 2-commuting data

and an isomorphism \beta\colon \phi^2\circ \eta_{\sf C}\rightarrow \eta_{\sf D} between internal natural transformations in the diagram

where \beta is compatible with \delta_{\sf C} and \delta_{\sf D}.

Let ({\sf C},\tau_{\sf C},\eta_{\sf C},\delta_{\sf C}) and ({\sf D},\tau_{\sf D},\eta_{\sf D},\delta_{\sf D}) be internal categories with internal involution, and let F,G\colon {\sf C}\to {\sf D} be internal functors with equivariance data (\phi_F,\beta_F) and (\phi_G,\beta_G). Equivariance data for an internal natural transformation E\colon F\Rightarrow G is an isomorphism \sigma\colon \phi_G\circ E\simeq E\circ \phi_F of natural transformations compatible with \beta_F and \beta_G.

Define a category (G,{\mathbb{M}})\hbox{-{\sf Bord}}(M) internal to symmetric monoidal stacks as follows.

An object of the stack {\rm Ob}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)) is an equivalence class of the data:

1. a (G,{\mathbb{M}})-pair Y^c\subset Y\to S where Y^c\to S is proper;

2. a partition into two components Y^+\coprod Y^-\simeq Y\setminus Y^c\subset Y; and

3. a map of supermanifolds \phi\colon Y\to M.

The supermanifold Y^c is called the core. The supermanifold Y\supset Y^c in which the core sits is the collar. The collar is partitioned into an outgoing component Y^+ and an incoming component Y^-.

We apply the following equivalence relation to these data:

where W\subset Y is an open submanifold containing Y^c\subset Y, and W^+=W\bigcap Y^+. In short, we only remember the data of the map \phi on a half-open germ of a neighborhood of Y^c\subset Y of the outgoing collar. We often abbreviate the data of an object of {\rm Ob}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)) to \phi\colon Y\to M.

A morphism in {\rm Ob}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)), denoted (Y^c_0\subset Y_0,Y^\pm_0,\phi_0)\to (Y^c_1\subset Y_1,Y^\pm_1,\phi_1) is an equivalence class of open submanifolds V_0\subset Y_0 and V_1\subset Y_1 containing the cores and isometries f\colon V_0\to V_1 in the commutative diagram

These data are subject to the equivalence relation that f\colon V_0\to V_1 and f'\colon V_0'\to V_1' represent the same isomorphism if f|_{V_0\bigcap V_0'}=f'|_{V_0\bigcap V_0'}.

Disjoint union endows {\rm Ob}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)) with a symmetric monoidal structure.

An object of the stack {\rm Mor}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)) is given by the data:

1. a (G,{\mathbb{M}})-family \Sigma\to S,

2. a map of supermanifolds \Phi\colon \Sigma\to M,

3. a pair of objects of {\rm Ob}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)), denoted \phi_{\rm in}\colon Y_{\rm in}\to M, \phi_{\rm out}\colon Y_{\rm out}\to M,

4. spans of smooth maps Y_{{\rm out}} \hookleftarrow W_{{\rm out}} \stackrel{i_{{\rm out}}}{\to} \Sigma\stackrel{i_{{\rm in}}}{\leftarrow} W_{{\rm in}} \hookrightarrow Y_{{\rm in}}, where W_{{\rm out}/{\rm in}}\hookrightarrow Y_{{\rm out}/{\rm in}} are open embeddings.

Define the notation

Then \Sigma^c:=\Sigma\setminus (i_{\rm in}(W_{\rm in}^-)\cup i_{\rm out}(W_{\rm out}^+)) is the core. The above data are require to satisfy

1. the image W_{\rm in/out}\hookrightarrow Y_{\rm in/out} contains a neighborhood of the core Y^c_{\rm in/out};

2. \Sigma^c\to S is proper;

3. i^-_{\rm in} and i^-_{\rm out} are isometries onto their image in \Sigma\setminus i_{\rm out}(W_{\rm out}^+\cup Y_{\rm out}^c); and

4. the maps \phi_{\rm in/out}\colon Y_{\rm in/out} \to M are compatible with \Phi\colon \Sigma\to M.

Similar to (?), we impose an equivalence relation on objects

so again the equivalence class of \Phi is only the data of a half-open neighborhood of the core \Sigma^c\to M inside the outgoing collar.

A morphism in {\rm Mor}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)) is an equivalence classes of open sub supermanifolds \Omega_i\subset \Sigma_i, V_{i,{\rm in}}\subset Y_{i,{\rm in}}, V_{i,{\rm out}}\subset Y_{i,{\rm out}} satisfying

and isometries f_{\rm in} \colon V_{0,{\rm in}}\to V_{1,{\rm in}}, f_{\rm out} \colon V_{0,{\rm out}}\to V_{1,{\rm out}} and F\colon \Omega_0\to \Omega_1 in the commutative diagram of supermanifolds over M

where we also require F preserves the core of the bordism, F|_{\Sigma^c_0}\colon \Sigma^c_0\to \Sigma_1^c\subset \Omega_1. These data are equivalent (f,g,F)\sim (f',g',F') when

Disjoint union endows {\rm Mor}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)) with a symmetric monoidal structure. We often abbreviate the data of an object of {\rm Mor}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)) to Y_{\rm in}\coprod Y_{\rm out} \to \Sigma\xrightarrow{\Phi} M, or even more succinctly to \Phi\colon \Sigma\to M.

Unit, source, and target functors are given by the assignments on objects

with similar assignments on morphisms. For any S, let \emptyset\in {\rm Ob}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M))(S) and \emptyset\in {\rm Mor}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M))(S) denote the empty object over S, i.e., the symmetric monoidal unit over S for these symmetric monoidal stacks.

Composition is given by gluing bordisms along collars; see [ST11, page 20] and [HST, Definition 6.5].

The source and target functors

are fibrations of stacks.

We recall that a map of stacks is a fibration if it induces a fibration of groupoids over each supermanifold S, e.g., see [Hollander]. We prove the statement of the lemma for the functor {\sf s}; the case for {\sf t} is similar. It suffices to show that for any isomorphism f\colon {\sf s}(\Sigma_0,\Phi_0)\to (Y_1,\phi_1) in {\rm Ob}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M'))(S) where

there exists an isomorphism (f,g,F)\colon (\Sigma_0,\Phi_0)\to (\Sigma_1,\Phi_1) in {\rm Mor}((G,{\mathbb{M}})\hbox{-{\sf Bord}}(M'))(S) with {\sf s}(f,g,F)=f. We recall that the notation \Sigma_0 for a cobordism is an abbreviation for the data

Similarly, f\colon {\sf s}(\Sigma_0,\Phi_0)\to (Y_1,\phi_1) is an abbreviation for the data of open sub supermanifolds V_0\subset Y_0, V_1\subset Y_1 containing the core and an isometry f\colon V_0\to V_1 preserving the cores.

Using the notation above, we define (\Sigma_1,\Phi_1) by the data

where the relevant properties are inherited from (?). Next define an isomorphism (\Sigma_0,\Phi_0)\to (\Sigma_1,\Phi_1) by taking V_{0,{\rm out}}=V_{1,{\rm out}}=W_{{\rm out}}, \Omega_0=\Omega_1=\Sigma_0, V_{0,{\rm in}}=W_{0,{\rm in}}\bigcap V_0, V_{1,{\rm in}}=f(W_{0,{\rm in}}\bigcap V_0), f_{\rm out}={{\rm id}}_{W_{{\rm out}}}, F={{\rm id}}_{\Sigma_0} and f_{\rm in}=f|_{W_{0,{\rm in}}\bigcap V_0}. By construction, the diagram (?) commutes with this data and the requisite properties are satisfied to define an isomorphism in {\rm Mor}(1|1\hbox{-}{\sf EBord}(M)). The source functor applied to this isomorphism agrees with (f,V_0,V_1) on restriction to V_0\bigcap W_{0,{\rm in}} and f(V_0\bigcap W_{0,{\rm in}}), so this morphism is equal to f because morphisms are defined as equivalence classes. This completes the proof.

A map f\colon M\to M' of smooth manifolds induces an internal functor

and these internal functors are compatible with composition of maps between manifolds.

Define the category {\sf {Vect}} internal to symmetric monoidal stacks as follows. The stack {\rm Ob}({\sf {Vect}}) classifies bundles of {\mathbb{Z}}/2-graded, nuclear Fréchet spaces over {\mathbb{C}}, i.e., locally free sheaves of nuclear Fréchet super vector spaces. The stack {\rm Mor}({\sf {Vect}}) has objects maps between vector bundles, and as morphisms commuting squares

whose horizontal arrows are maps of vector bundles and whose vertical arrows are fiberwise isomorphisms of vector bundles, i.e., pullbacks along base changes. There are functors {\sf s,t}\colon {\rm Mor}({\sf {Vect}})\to {\rm Ob}({\sf {Vect}}) from restricting to the left and right halves of the square (?). Finally, endow {\rm Ob}({\sf {Vect}}) and {\rm Mor}({\sf {Vect}}) with symmetric monoidal structures from the projective tensor product of topological vector bundles.

The above deviates from the definition in [ST11] in two ways. First, we only work with vector bundles rather than general sheaves of topological vector spaces (i.e., we impose a locally free condition). Second, we specialize from general locally convex topological vector spaces to nuclear Fréchet spaces. We make this choice in order to guarantee that certain formal categorical structures in the functorial definition of quantum field theory agree with standard structures in physics: Fréchet spaces have the approximation property [Schaefer, page 109], leading to a well-behaved theory for traces of nuclear operators, see [STTraces, §4.5]. It also happens that all the desired constructions in index theory result in vector bundles of nuclear Fréchet spaces, so this restriction does not any exclude any of the desired examples. We refer to [costbook, Appendix 2] for a succinct overview of nuclear Fréchet spaces. Stolz and Teichner’s motivation to work with more general sheaves of topological vector spaces is explained in Remark 1.16.

Define the internal Morita category {\sf {Alg}} as a category internal to symmetric monoidal stacks as follows. The stack {\rm Ob}({\sf {Alg}}) classifies locally trivial bundles of nuclear Fréchet super algebras and isomorphisms of algebra bundles. The stack {\rm Mor}({\sf {Alg}}) classifies locally trivial bundles of nuclear Fréchet bimodules over topological algebras. Composition comes from the projective tensor product of bimodule bundles,

Viewing an algebra bundle as a bimodule over itself constructs the unit map {\sf u}\colon {\rm Ob}({\sf {Alg}})\to {\rm Mor}({\sf {Alg}}). Endow {\rm Ob}({\sf {Alg}}) and {\rm Mor}({\sf {Alg}}) with monoidal structures from the projective tensor product of algebra and bimodule bundles over the structure sheaf of the base.

Given any category {\sf C} internal to symmetric monoidal stacks, there is an internal functor

defined as follows. For Y\in {\rm Ob}({\sf C})(S), assign the trivial algebra bundle on S, i.e., the sheaf of algebras whose global sections are C^\infty(S). For \Sigma\in {\rm Mor}({\sf C})(S), assign the trivial bimodule bundle over S, i.e., the sheaf of bimodules whose global sections are C^\infty(S) as a bimodule over C^\infty(S). The 2-commuting data \mu is determined by the canonical isomorphism C^\infty(S)\otimes_{C^\infty(S)} C^\infty(S)\simeq C^\infty(S), while the 2-commuting data \epsilon comes from the canonical identification of the algebra C^\infty(S) with the identity bimodule over itself.

A geometric field theory over M is an internal functor

A twisted geometric field theory over M is an internal natural transformation E

where \mathscr{T} is a twist functor and \mathbbold{1} is defined in Example ?. Twisted field theories are the objects of a groupoid whose morphisms are isomorphisms between internal natural transformations.

An (untwisted) geometric field theory in the sense of Definition ? determines a twisted field theory in the sense of Definition ? for the trivial twist functor \mathscr{T}=\mathbbold{1} [ST11, Lemma 5.7].

Fix a rigid geometry (G,{\mathbb{M}}). Given a homomorphism g\colon G\to G and a G-equivariant map f\colon {\mathbb{M}}\to {\mathbb{M}} relative to g that preserves the image of the inclusion {\mathbb{M}}^c\subset {\mathbb{M}}, there is a uniquely determined internal functor

The endofunctors of (G,{\mathbb{M}})-manifolds from Lemma ? determine maps of stacks,

by taking the same collar data on bordisms, but different (G,{\mathbb{M}})-structures. By construction, these maps of stacks are (strictly) compatible with source, target, unit, and composition functors defining (G,{\mathbb{M}})\hbox{-{\sf Bord}}(M).

In the context of Lemma ?, suppose that {\sf fl} \colon {\mathbb{M}}\to {\mathbb{M}} is an involution commuting with the action of G and preserving {\mathbb{M}}^c\subset {\mathbb{M}}. This endows every family of (G,{\mathbb{M}})-manifolds with an involution. In practice, an involution {\sf fl}\colon {\mathbb{M}}\to {\mathbb{M}} comes from a spin flip automorphism, e.g., see [HST, §6.7, Lemma 6.39], [ST11, Example 2.39 and §4.1] and (3.8). Following [ST11, §2.6], such an involution {\sf fl} allows one to regard the bordism category as a category internal to symmetric monoidal stacks with involution.

Fix a rigid geometry (G,{\mathbb{M}}) with an involution {\sf fl}\colon {\mathbb{M}}\to {\mathbb{M}} as in Example ?. A flip-preserving geometric field theory is a functor (?) internal to the category of symmetric monoidal stacks with involution. Similarly, one defines flip-preserving twists and flip-preserving twisted geometric field theories as functors and natural transformations internal to symmetric monoidal stacks with involution.

Fix an orientation on the manifold {\mathbb{M}}_{\rm red}, and suppose that {\sf or} \colon {\mathbb{M}}\to {\mathbb{M}} is a G-equivariant map with the property that {\sf or}_{\rm red}\colon {\mathbb{M}}_{\rm red}\to {\mathbb{M}}_{\rm red} is an orientation reversing map and {\sf or}^2={\sf fl} is the spin flip. Then Lemma ? determines an orientation reversal functor

with a natural isomorphism {\sf Or}\circ {\sf Or}\Rightarrow {{\rm id}} (via the spin flip). Indeed, {\sf Or} determines an internal involution in the sense of Definition ?.

Given an S-family of (G,{\mathbb{M}})-supermanifolds (Y\to S), a real structure on the rigid geometry (G,{\mathbb{M}}) gives a unique (G,{\mathbb{M}})-stucture for the \overline{S}-family (\overline{Y}\to \overline{S}). This extends to an internal functor

The identification \overline{M}\simeq M uses that M is an ordinary manifold, and hence has a canonical real structure when regarded as a supermanifold. Via the forgetful functor (G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)\to {\sf SMfld} that sends an S-family of bordisms to S, the internal functor (?) covers the conjugation functor on supermanifolds. We observe that {\sf R} determines an internal involution in the sense of Definition ?.

Let (G,{\mathbb{M}}) be a rigid geometry with a chosen real structure and orientation reversing automorphism. A reflection structure for a twist \mathscr{T} is equivariance data for the involutions {\sf R}\circ {\sf Or} and (\overline{\phantom{-}}).

A reflection structure for a twisted field theory E\colon \mathbbold{1}\to \mathscr{T} is equivariance data for the internal natural transformation E.

Roughly, a reflection structure stipulates that orientation reversal of bordisms correspond to complex conjugation of vector spaces (which is data). In tandem with this structure is the reflection positivity property, which demands that a certain hermitian pairing is positive. Suggestively we write this pairing as

where the isomorphism h comes from the reflection structure, and \Sigma is a bordism with boundary \overline{Y}\coprod Y where Y is the orientation reversal of Y. Then the positivity property is the positivity of the pairing (?).

We provide the precise property for 1|1-Euclidean field theories in Definition 5.16.

Let (G,{\mathbb{M}}) be a rigid geometry with a chosen real structure, and \mathscr{T}\colon (G,{\mathbb{M}})\hbox{-{\sf Bord}}(M)\to {\sf {Alg}}. A real structure for \mathscr{T} is equivariance data for the involutions {\sf R} and (\overline{\phantom{-}}). Given a twisted field theory E\colon \mathbbold{1}\to \mathscr{T}, a real structure on E is equivariance data for the internal natural transformation.

In many cases, there is a nontrivial {\mathbb{R}}_{>0}-action on {\mathbb{M}} that commutes with the G-action. For example, the Euclidean rigid geometries from Example ? admit such a structure from the standard dilation action on {\mathbb{R}}^d. Applying Lemma ?, this leads to the action of the renormalization group,

which then acts on the category of (twisted) field theories by precomposition with the internal functor (?).

Let V\to M be a complex vector bundle. A superconnection {\mathbb{A}} on V is an odd, {\mathbb{C}}-linear map of Fréchet spaces satisfying a Leibniz rule

where v\in \Omega^\bullet(M;V) and \alpha\in \Omega^\bullet(M).

Suppose that \langle-,-\rangle\colon\overline{V}\otimes V\to \underline{{\mathbb{C}}} is a (super) hermitian pairing, where \otimes is the projective tensor product. Then a superconnection {\mathbb{A}} on V is self-adjoint if

where {\mathbb{A}}_k^* is the super adjoint of {\mathbb{A}}_k.

Translating from super adjoints to ordinary adjoints as in (?), the above is equivalent to the standard definition of self-adjoint superconnection [BGV, page 117].

The Chern form of a {\rm Cl}_n-linear superconnnection is the Clifford super trace (which need not exist for infinite rank bundles)

When it exists, the Chern form of a {\rm Cl}_n-linear superconnection {\mathbb{A}} is closed and has total degree n.

A topological category is a category in which the objects and morphisms are topological spaces for which the structure maps (source, target, unit, composition) are continuous. For a topological category \mathcal{C}, let \mathcal{C}_0 denote the space of objects and \mathcal{C}_1 the space of morphisms.

Let \mathcal{M}_n be the category whose objects are finite-dimensional {\rm Cl}_{n}-submodules W\subset \mathcal{H}_n and whose morphisms are

where W_0^\perp\subseteq W_1 denotes the orthogonal complement of the subspace W_0\subseteq W_1. Composition for a nested inclusion W_0\subseteq W_1\subseteq W_2 uses the direct sum of Clifford modules to obtain a {\rm Cl}_{n}\otimes {\rm Cl}_{-1}-module structure on the orthogonal complement of W_0\subseteq W_2. Endow the objects and morphisms of \mathcal{M}_n with the subspace topology for the embeddings

where the first inclusion identifies a submodule W_0\subset \mathcal{H}_n with the (finite-rank) projection \mathcal{H}_n\to W_0. For the second inclusion, a triple (W_0,W_1,e) determines a pair of finite rank projection operators and a bounded operator e.

The nerve of a topological category \mathcal{C} is the simplicial space whose k-simplices are N_k\mathcal{C}=\mathcal{C}_k for k=0,1, and for k>1 are length k chains of composable morphisms

topologized as the fibered product.

The face maps in N_\bullet \mathcal{C} are determined by composing morphisms, and degeneracies are determined by the unit map in \mathcal{C}.

The classifying space {\rm B} \mathcal{M}_n represents the functor {\rm KO}^{n}(-).

For Z a topological space and \{U_i\}_{i\in I} an open cover, the Čech category, denoted \check{{\rm C}}(U_i), is the topological category whose objects are \check{{\rm C}}(U_i)_0=\coprod_{\sigma} U_\sigma and whose morphisms are the space \check{{\rm C}}(U_i)_1=\coprod_{\sigma\subseteq \tau} U_\tau. The source and target maps are determined by the projection and inclusion U_\tau\subseteq U_\sigma, the unit is induced by the identity map U_\sigma \to U_\sigma, and composition comes from nested inclusions.

For Z a topological space and \{U_i\}_{i\in I} an open cover, the canonical map

is a homotopy equivalence.

A {\rm Cl}_n-bundle on a manifold M is data

1. an open cover \{U_i\}_{i\in I} of M,

2. for each finite subset \sigma \subset I, a metrized super vector bundle W_\sigma\to U_\sigma with {\rm Cl}_n-action such that the fibers of W_\sigma are self-adjoint {\rm Cl}_n-modules;

3. for each \sigma\subset \tau, a {\rm Cl}_n-equivariant, injective map of super vector bundles g_{\sigma\tau}\colon W_\tau\hookrightarrow W_\sigma, together with a {\rm Cl}_{n}\otimes {\rm Cl}_{-1}-action on W_\tau^\perp\subset W_\sigma on the orthogonal complement of the image extending the {\rm Cl}_n-action.

For \sigma\subset \tau\subset \rho, we require an equality of maps g_{\sigma\tau}\circ g_{\tau\rho}=g_{\sigma\rho}\colon W_\rho\hookrightarrow W_\sigma, and that the {\rm Cl}_{-1}-action on W_\rho^\perp\subset W_\tau \hookrightarrow W_\sigma restricts from the {\rm Cl}_{-1}-action on W_\tau^\perp\subset W_\sigma.

Given a {\rm Cl}_n-bundle on M, there exists a continuous functor

that is unique up to a contractible space of choices. After realization, this determines a map M\to B\mathcal{M}_n\simeq {\rm KO}^n into the nth space of {\rm KO} spectrum.

The functor (?) comes from choosing a {\rm Cl}_n-equivariant embedding W_\sigma\subset \mathcal{H}_n\times U_\sigma over U_\sigma for each \sigma\subset I. By Kuiper’s theorem [Kuiper], the space of such choices is contractible. Then the corollary follows from Proposition ? and Lemma ?.

Given a {\rm Cl}_n bundle on M, choices of self-adjoint {\rm Cl}_n-linear superconnections {\mathbb{A}}_\sigma on each W_\sigma uniquely determine a Čech–de Rham representative of the Pontryagin character {\rm Ch}(W)\in {\rm H}^\bullet(M;{\mathbb{R}}[u,u^{-1}]) of [W]=[\{U_\sigma,W_\sigma,g_{\sigma\tau},e_{\sigma\tau}\}]\in {\rm KO}^n(M).

Over each U_i, a choice of superconnection on W_i gives a differential form {\rm Ch}_n({\mathbb{A}}_i)\in \Omega^\bullet(U_i;{\mathbb{R}}[u,u^{-1}]) of total degree n. The data g_{\sigma\tau},e_{\sigma\tau} allow one to extend this to a Čech–de Rham cocycle.

A differential refinement of a {\rm Cl}_n-bundle V=\{V_\sigma,g_{\sigma\tau},e_{\sigma\tau}\} defined relative to an open cover \{U_i\}_{i\in I} is data

1. i. for each \sigma\subset I, a self-adjoint {\rm Cl}_n-linear superconnection {\mathbb{A}}_\sigma on V_\sigma\to U_\sigma, and

2. ii. a closed global form {\rm Ch}(V,{\mathbb{A}},\phi)\in \Omega^\bullet(M;{\mathbb{R}}[\alpha^{\pm 1}]) of total degree n;

3. iii. a coboundary \{\phi_\sigma\in \Omega^\bullet(U_\sigma;{\mathbb{R}}[\alpha^{\pm 1}])\} between {\rm Ch}(V,{\mathbb{A}},\phi) and the Čech–de Rham cocycle {\rm Ch}_n(\{U_i\},{\mathbb{A}}_i)\in \Omega^\bullet(U_*;{\mathbb{R}}[u^{\pm 1}]) in Lemma ?.

A differential refinement of a {\rm Cl}_n-bundle is a differential cocycle denoted by (V,{\mathbb{A}},\phi)=\{(V_\sigma,{\mathbb{A}}_\sigma,\phi_\sigma),(g_{\sigma\tau},e_{\sigma\tau})\}.

A self-adjoint Clifford linear superconnection {\mathbb{A}} admits smooth index bundles if

1. the collection of subsets \{U_\lambda\}_{\lambda\in {\mathbb{R}}_{>0}} defined in (?) form a smooth open cover;

2. the subsets \mathcal{H}^{<\lambda}\subset \mathcal{H}|_{U_\lambda}\to U_\lambda are smooth vector bundles;

3. the composition defines a smooth connection:

   \nabla^{<\lambda}:=p^\lambda\circ {\mathbb{A}}^{[1]}|_{U_\lambda}\circ p^\lambda\colon \Gamma(\mathcal{H}^{<\lambda})\to \Omega^1(U_\lambda;\mathcal{H}^{<\lambda}),

4. the limit of Chern–Simons forms

   \begin{aligned} \eta_{\lambda}(t):=\lim_{r\to \infty} \int_1^r {\sf sTr}_{{\rm Cl}_n}(\frac{d{\mathbb{A}}_{\lambda}(s)}{ds}\exp({\mathbb{A}}_{\lambda}(s)))ds \in \Omega^\bullet(U_\lambda ;C^\infty({\mathbb{R}}_{>0})[u^{\pm 1}]) \end{aligned}

   exists and satisfies

   \begin{aligned} d\eta_\lambda(t)={\rm Ch}({\mathbb{A}}(t))|_{U_\lambda}-{\rm Ch}(\nabla^{<\lambda}) \end{aligned}

   where in (?), {\mathbb{A}}_{\lambda}(s) is the 1-parameter family of superconnnections on V|_{U_\lambda} \oplus \Pi \mathcal{H}^{<\lambda},

   \begin{aligned} &&{\mathbb{A}}_{\lambda}(s)={\mathbb{A}}_{\lambda}(s)^{[0]}+{\mathbb{A}}^{[1]} \oplus \nabla^{<\lambda} +\sum_{i>1} s^{-i} {\mathbb{A}}^{[i]} \oplus 0_{\Pi \mathcal{H}^{<\lambda}},\quad {\mathbb{A}}_{\lambda}(s)^{[0]}=\left[\begin{array}{cc} {\mathbb{A}}^{[0]} & s\cdot i_\lambda \\ s \cdot p^\lambda & 0 \end{array}\right] \end{aligned}

   where i_\lambda and p^\lambda are the maps (?).

Given a {\rm Cl}_n-linear superconnection admitting cutoffs, a choice of locally finite open cover \{U_\lambda\}_{\lambda \in \Lambda} associated with a discrete subset \Lambda\subset {\mathbb{R}}_{>0} determines a map

to the nth space in the {\rm KO}-spectrum that is unique up to contractible choice. Given choices \Lambda\subset \Lambda' \subset {\mathbb{R}}_{>0} related by a refinement of open covers, there is a unique homotopy between the associated maps (?).

The {\rm Cl}_n-modules \mathcal{H}^{<\lambda}\to U_\lambda determine class [\mathcal{H}^{<\lambda}]\in {\rm KO}^n(U_\lambda) for each \lambda\in {\mathbb{R}}_{>0}. On overlaps U_\lambda\bigcap U_{\mu} for \lambda<\mu there are canonical inclusions

and the orthogonal complement to the image of g_{\lambda\mu} has a {\rm Cl}_{n}\otimes {\rm Cl}_{-1}-action extending the {\rm Cl}_n-action, determined by the invertible odd endomorphisms

that define the action of the generator e\in {\rm Cl}_{-1}. The inclusions g_{\lambda\mu} and odd endomorphisms e^{\lambda\mu} satisfy a compatibility property on triple intersections. By Corollary ?, we obtain a map to the nth space in the {\rm KO}-spectrum (?). A refinement of open covers associated with \Lambda\subset \Lambda'\subset {\mathbb{R}}_{>0}, determines a continuous functor between the associated Čech categories that after realization is a homotopy equivalence. This leads to a pair of functors (?) and restriction of the operators (?) furnish a natural isomorphism between these functors. By [Segalclassifying, Proposition 2.1], the resulting maps {\rm Ind}_\Lambda({\mathbb{A}}),{\rm Ind}_{\Lambda'}({\mathbb{A}})\colon M\to {\rm KO}^{-n} are homotopic.

Let {\mathbb{A}} be a self-adjoint Clifford linear superconnection that admits smooth index bundles. The differential forms (?) witness a coboundary in the Čech–de Rham complex between (?) and the (globally defined) cocycle {\rm Ch}({\mathbb{A}}). Hence,

exacts a differential cocycle from the superconnection {\mathbb{A}}.

The Čech–de Rham coboundary follows from (?). The differential cocycle (?) uses the description of differential {\rm KO}-theory from [DBEBis, §7].

A spin structure for a rank n real, oriented, metrized vector bundle V\to M is a {\rm Spin}(n)-principal bundle P_V\to M equipped with a double cover P_V\to {\rm Fr}(V) of the oriented frame bundle of V that is equivariant relative to {\rm Spin}(n)\to {\rm SO}(n).

For \pi\colon X\to M a bundle of Riemannian manifolds for which T(X/M) has a spin structure, the spinor bundle is

The sections of {\mathbb{S}}_{X/M} support an M-family of Dirac operators \slashed{D} that graded commute with the left {\rm Cl}_{-n}={\rm Cl}_n^{\sf{op}} action.

Any finite-rank self-adjoint Clifford linear superconnnection admits cutoffs. Self-adjoint Clifford linear superconnections adapted to a family of Clifford linear Dirac operators admit cutoffs.

The finite rank case is [BGV, Theorem 9.7]. For superconnections adapted to Dirac operators (without Clifford actions), this follows from [BGV, Proposition 9.10 and Theorem 9.26], see also [FreedLott, §7.2]. The straightforward extension of these results to Clifford linear Dirac operators is [DBEBis, §8.2-8.3].