Generalised Spin Structures in General Relativity

Generalised spin structures describe spinor fields that are coupled to both general relativity and gauge theory. We classify those generalised spin structures for which the corresponding fields admit an infinitesimal action of the space–time diffeomorphism group. This can be seen as a refinement of the classification of generalised spin structures by Avis and Isham (Commun Math Phys 72:103–118, 1980).


Introduction
In this paper, we study the space-time transformation behaviour of spinors that are coupled to general relativity (GR) as well as gauge theory.
In the absence of gauge fields, space-time transformations of spinors coupled to GR can be understood by considering pairs (g, ψ) of a metric g, together with a compatible spinor field ψ. The transformation behaviour is then governed not by a spin structure Q → M , but rather by the principal GL + (n, R)bundle Q → M associated to Q along the inclusion of the spin group in GL + (n, R). Since spinor fields acquire a minus sign upon a full rotation, the action of the space-time diffeomorphism group Diff(M ) does not lift from M to Q. It does, however, lift at the infinitesimal level, i.e. at the level of the Lie algebra Vec(M ) of vector fields. This implies that in the absence of gauge fields, the spinor fields carry an action of the universal cover of the connected component of unity of the space-time diffeomorphism group [8].
In the presence of gauge fields, the consistent description of spinors requires a so-called generalised spin structure or Spin G -structure [4,16]. This is a natural generalisation of a Spin c -structure and reduces to this in the case G = U(1) of electrodynamics. Generalised spin structures were classified in [2].
The aim of the present paper is to study the transformation behaviour of spinors in the presence of both gauge theory and GR. Just like in the case M is a spin manifold, the space-time diffeomorphism group does not admit a natural action on the spinor fields of M = CP 2 , not even at the infinitesimal level.

Spinors Coupled to GR and Gauge Fields
In view of the central role of this notion in the present paper, we give a more detailed description of infinitesimally natural bundles in Sect. 2.1. In Sects. 2.2-2.4, we then formulate the kinematics of spinors coupled to GR in terms of fibre bundles over the space-time manifold M , the main point being that the relevant bundles are infinitesimally natural. In Sect. 2.5, we describe spinors coupled to both GR and gauge theory. In this setting, the relevant bundles are associated to Spin G -structures rather than spin structures. In Sect. 2.6, we focus on the space-time transformation behaviour of these generalised spin structures and show that they are not necessarily infinitesimally natural.

Natural and Infinitesimally Natural Bundles
In a geometric setting, classical fields are sections of a fibre bundle π : Y → M over the space-time manifold M . Such a bundle is called natural if (locally defined) diffeomorphisms α of M lift to (locally defined) automorphisms Σ(α) of Y → M , in such a way that composition and inversion are preserved. More precisely, one requires that Σ(α) −1 = Σ(α −1 ) and Σ(α • β) = Σ(α) • Σ(β) for all composable local diffeomorphisms α and β on M . The space-time diffeomorphism group then acts naturally on the space of fields: a diffeomorphism α maps a field φ : M → Y to the field Σ(α) • φ • α −1 .
Natural bundles are perfectly suited for GR, providing a geometric framework not only for (mixed) tensor fields, but also for the more complicated transformation behaviour of (Levi-Civita) connections. They first appeared under the name 'geometric objects' [28,36,41], although the modern definition is due to Nijenhuis [29,30]. Natural bundles were fully classified by Palais and Terng [32], building on work of Salvioli [35] and Epstein and Thurston [12].
Unfortunately, the framework of natural bundles is quite unsuitable for field theories involving spinors. The reason is that a (local) full rotation of the space-time manifold M acts trivially on the (local) fields. Therefore, the minus sign associated to spinor rotation cannot be reproduced within the setting of natural bundles.
One way to deal with this is to give up on diffeomorphism invariance, and instead ask for invariance under the automorphism group of an underlying principal fibre bundle. This leads to the theory of gauge natural bundles [9,21]. Because the distinction between space-time symmetries and gauge symmetries is lost, it is rather hard to recover the distinction between the SEM-tensor and the gauge currents in this formalism [25,33].
In this paper, we propose a different solution. Rather than abandoning diffeomorphism invariance altogether, we require diffeomorphism invariance only at the infinitesimal level. A fibre bundle π : Y → M is called infinitesimally natural if it comes with a lift of infinitesimal diffeomorphisms [19].
More precisely, an infinitesimally natural bundle is a smooth fibre bundle π : Y → M , together with a Lie algebra homomorphism σ : Vec(M ) → aut(Y ). We require that the lift σ(v) of any vector field v ∈ Vec(M ) projects down to v again, π * • σ(v) = v. Remark 1. If π : Y → M is just a smooth fibre bundle, then aut(Y ) is the Lie algebra of projectable vector fields on Y , and aut V (Y ) denotes the Lie algebra of vertical vector fields. However, if Y has additional structure, then we will take aut(Y ) to be the corresponding subalgebra of infinitesimal automorphisms. For example, if Y is a principal H-bundle or a bundle of homogeneous spaces, then aut(Y ) = Vec(Y ) H is the Lie algebra of equivariant vector fields, and aut V (Y ) is the gauge algebra of vertical, equivariant vector fields.
Rephrasing the above definitions, one can say that a natural bundle has a (local) splitting of the sequence of groups, whereas an infinitesimally natural bundle has a splitting of the corresponding exact sequence Every natural bundle is of course infinitesimally natural, but the converse is not true. It turns out that the extra leeway provided by infinitesimally natural bundles is just enough to describe spin structures and certain types of generalised spin structures, while at the same time providing the extra structure needed to globally define a canonical SEM-tensor, cf. [13, p. 333], [14].
Throughout the paper, we assume that M is a smooth, connected, orientable manifold, and we fix a nondegenerate, bilinear form η : R n × R n → R. Unless stated otherwise, M will be of dimension n ≥ 3. The adaptations needed for the case n = 2 will be briefly discussed in Remark 6. We denote the group of orientation preserving linear transformations of R n by GL + (n, R), and we denote by SO(η) ⊆ GL + (n, R) the subgroup of transformations that preserve η. The principal GL + (n, R)-bundle of oriented frames is denoted by F + → M , or by F + (M ) → M if we need to emphasise the manifold. If g is a pseudo-Riemannian metric on M of signature η, then the principal SO(η)-bundle of oriented, g-orthonormal frames is denoted OF + g → M . We assume that the gauge group G is a compact Lie group and denote its Lie algebra by g.

General Relativity
The fundamental degrees of freedom in general relativity are a pseudo-Riemannian metric g of signature η on space-time M , and a connection ∇ on T M. It will be convenient to describe both g and ∇ as sections of a bundle of homogeneous spaces. We identify the metric g with a section of the bundle in the usual way, namely by associating to g x : We view the connection ∇ on T M as an equivariant connection on F + → M . This in turn can be identified with a section of These two, the metric g and the connection ∇, are conveniently combined into a single section Φ g,∇ of the fibre bundle By concatenating the section Φ g,∇ : M → J 1 (F + )/SO(η) with the projections J 1 (F + )/SO(η) → F + /SO(η) and J 1 (F + )/SO(η) → J 1 (F + )/GL + (n, R), one recovers the metric g and the connection ∇ from the section Φ g,∇ .
The fields Φ g,∇ transform in a natural fashion under the group Diff + (M ) of orientation preserving diffeomorphisms. Indeed, any α ∈ Diff + (M ) gives rise to an automorphism Σ(α) of the bundle J 1 (F + )/SO(η) → M of homogeneous spaces, defined by The diffeomorphism α then maps the field Φ g,∇ to Σ(α) • Φ g,∇ • α −1 , which is again a section of the bundle J 1 (F + )/SO(η) → M . Note that the group homomorphism splits the exact sequence of groups The derived Lie algebra homomorphism therefore splits the corresponding exact sequence of Lie algebras The bundle J 1 (F + )/SO(η) → M , whose sections Φ g,∇ describe a metric g together with a connection ∇, is therefore an infinitesimally natural bundle in the sense of [19]. Needless to say, the lift Σ : Diff + (M ) → Aut(J 1 (F + )/SO(η)) is of central importance in GR, since diffeomorphism invariance is one of the guiding principles for finding the Einstein-Hilbert action.

Spin Structures
The description of spinors coupled to general relativity (GR) involves a twofold cover of SO(η). In order to handle manifolds M that are oriented, but not necessarily time-oriented, we define SO(η) to be the twofold cover κ −1 (SO(η)) of SO(η) that arises as the restriction of the universal covering map κ : GL + (n, R) → GL + (n, R).
Let M be an orientable manifold of dimension n ≥ 3 with a pseudo-Riemannian metric g of signature η. Then a spin structure is by definition an SO(η)-bundle Q over M , equipped with a twofold cover u : Q → OF + g of the oriented, orthogonal frame bundle, such that the following diagram commutes: Recall that the twofold cover κ : SO(η) → SO(η) is the restriction of the universal covering map of GL + (n, R). A manifold is called spin if it admits a spin structure. We define the principal GL + (n, R)-bundle and denote the induced map Q → F + by u as well. As any cover of F + by a GL + (n, R)-bundle can be obtained in this way, there is a 1:1 correspondence between spin covers of OF + g (M ) and F + . In particular, whether or not M is spin depends neither on the metric nor on the signature.
For n ≥ 3, we identify the fundamental group π 1 (GL + (n, R)) of GL + (n, R) with Z 2 . The orbit map can be seen as fibre inclusion, so the Serre homotopy exact sequence gives rise to the exact sequence of groups Proposition 1. A spin structure exists if and only if ι * : Z 2 → π 1 (F + ) is injective and (4) splits as a sequence of groups. If spin structures exist, then equivalence classes of spin covers correspond to splittings of (4).
Proof. See e.g. [26]. Alternatively, this criterion for M to be spin is equivalent to the vanishing of the second Stiefel-Whitney class [23]. In the same vein, we have the following criterion for the universal cover of M to be spin.

Proposition 2. The universal cover of M is spin if and only if the map
Proof. The universal covering map p : M → M gives rise to the pushforward map Dp : F + ( M ) → F + (M ) of oriented frame bundles. If we denote by ι M and ι M the fibre inclusions for the oriented frame bundle of M and M , respectively, we find the following commutative diagram: On the level of homotopy groups, this yields the commutative diagram is a covering map, it induces an injective homomorphism Dp * : π 1 (F + ( M )) → π 1 (F + (M )) of homotopy groups. From the above diagram, one then infers that ι M * is injective if and only if ι M * is injective. By Proposition 1, this is the case if and only if M is spin.

Spinors Coupled to GR
Spinor fields are usually described as sections of a spinor bundle S g → M , associated to a spin structure Q → M along a unitary spinor representation 2 V of SO(η). This description is somewhat inconvenient to describe spinors coupled to GR, because variations in the metric g would change the very bundle S g → M of which the spinors are sections.
Although it is possible to deal with this problem, we prefer to sidestep it by using the composite bundle associated to the twofold cover u : Q → F + of Sect. 2.3 along the spinor representation V (cf. e.g. [18, p. 177] and [27]). From a section τ : M → Q × SO(η) V , one recovers both the metric g and the spinor field ψ. Indeed, Using this metric g, one then defines the spin structure Q g = u −1 (OF + g ) inside Q. From this, one constructs the spinor bundle In the same vein, we will describe physical fields by sections of the fibre bundle J 1 ( Q) × SO(η) V . This is equivalent to providing three sections: one of F + /SO(η), one of J 1 (F + )/GL + (n, R), and one of S g = u −1 (OF + g ) × SO(η) V . These correspond to the metric g μν , the (Levi-Civita) connection Γ α μβ , and the spinor field ψ a , respectively.
We investigate the transformation behaviour of this bundle. Note that it is not a natural bundle in the sense of [30] or [21]. As a spinor changes sign under a 2π-rotation, there is no hope of finding an interesting group homomorphism Diff + (M ) → Aut(J 1 ( Q) × SO(η) V ). There is, however, a canonical homomorphism at the level of Lie algebras, making it an infinitesimally natural bundle in the sense of [19].
Because the twofold cover u : Q → F + has discrete fibres, it has a unique flat, equivariant connection, yielding a Lie algebra homomorphism This can be combined with the canonical Lie algebra homomorphism for the natural bundle F + → M . At the point f ∈ F + , it is defined by the first-order derivative of the pushforward map, where t → exp(tv) is the flow on M generated by the vector field v. The composition σ := ∇ can • D is a Lie algebra homomorphism σ : Vec(M ) → aut( Q) that splits the exact sequence of Lie algebras This induces a splitting for J 1 ( Q) by prolongation (see e.g. [13]), and consequently also one for

Remark 4.
We would like to emphasise that even if a splitting at the level of groups does exist, it will not be physically relevant, since it cannot reproduce the minus sign under a full rotation that one expects in spinors. Take for example the spin structure Q = R n × SO(η) over R n , and lift , we see that sections of the spinor bundle S = Q × SO(η) V then transform under the trivial representation of the Lorentz group, producing Lorentz scalars rather than spin-1/2 particles. In general, using a different splitting results in an incorrect energy-momentum tensor [14].
The above remark shows that it is not only the bundle Q and the covering map u : Q → OF g that are relevant, but also the splitting It must satisfy u * •σ = D in order for the metric g ∈ Γ(F + /SO(η)) to transform properly.
Although a canonical splitting σ is naturally associated to any ordinary spin structure, this is no longer the case for the Spin G -structures used to describe spinors coupled to gauge fields.

Generalised Spin Structures
In the presence of gauge fields, the topological conditions on M in order to support a spin structure are more relaxed. Roughly speaking, this is because the gauge group G can absorb some of the indeterminacy that stems from the 2:1 cover of the Lorentz group. This is made more rigorous by the notion of a generalised spin structure or Spin G -structure [2,4,16]. For n ≥ 3, we identify the centre of SO(η) with Z 2 π 1 (GL(n, R)). If G is a Lie group with a central subgroup Z 2 ⊆ G isomorphic to Z 2 , then we define 3 We denote the map (x, g) → κ(x) by κ : Spin G → SO(η).

Definition 1.
A Spin G -structure is a Spin G -bundle Q over M , together with a map u : Q → OF + g that makes the following diagram commute: The Spin G -structure Q gives rise to the principal GL(n, R) × Z2 G-bundle Let V be a representation of Spin G . The bundle of which the physical fields are sections is then the fibre bundle A single section of J 1 ( Q) × Spin G V represents a metric g μν , a Levi-Civita connection Γ α μβ , a gauge field A μ , and a spinor field ψ a . The metric is the induced section of the bundle and the Levi-Civita connection that of J 1 (F + )/GL + (n, R) → M . One constructs the principal G/Z 2 -bundle and the gauge field is the induced equivariant connection on P , a section of the bundle

Infinitesimally Natural Generalised Spin Structures
We now focus on the generalised spin structures that have an appropriate transformation law under infinitesimal space-time diffeomorphisms. We will call a Spin G -structure Q → M infinitesimally natural if the associated bundle Q → M is infinitesimally natural in the sense of Sect. 2.1.

Definition 2.
An infinitesimally natural Spin G -structure is a Spin G -structure u : Q → OF + g , for which there exists a Lie algebra homomorphism that splits the exact sequence where aut( Q) is the Lie algebra of Spin G -equivariant vector fields on Q. Moreover, we require that the composition u * • σ of σ with the pushforward u * is equal to the canonical splitting D of Eq. (7).
The splitting of (15) comes from the physical requirement that fields should have a well-defined transformation behaviour under infinitesimal coordinate transformations. The requirement u * • σ = D corresponds to the fact that we need to interpret a section of Q/ GL(n, R) × Z2 G F + /SO(η) as a metric, and we know that its transformation behaviour is governed by D. The Spin G -structures thus appear as the underlying principal fibre bundles in classical field theories combining gravity, spinors and gauge fields. If they are infinitesimally natural, then these fields have a well-defined transformation behaviour under infinitesimal space-time transformations. In particular, a stress-energy-momentum tensor corresponding to space-time transformations is then well defined by [13,14].

Classification
This raises the question which of the Spin G -structures are infinitesimally natural, and which ones are not. This is answered by Theorem 3 in Sect. 3.1. The proof proceeds by adapting the classification theorem for infinitesimally natural principal bundles (Theorem 4.4 in [19]) to the specific case of Spin Gstructures. We review the necessary material in Sect. 3.2 and proceed with the proof of Theorem 3 in Sect. 3.3.

The Classification Theorem
Let G be a Lie group with a central subgroup Z 2 ⊆ G isomorphic to Z 2 .
If the Lie algebra g of G does not contain any subalgebra isomorphic to sl(n, R)-a requirement that is automatically fulfilled if G is compact-then we shall prove the following classification theorem for infinitesimally natural Spin G -structures.

Theorem 3. (Classification theorem) An oriented manifold M of dimension n ≥ 3 admits infinitesimally natural Spin G -structures if and only if its universal cover is spin. For every infinitesimally natural Spin
In other words, every infinitesimally natural Spin G -structure is associated to the universal cover OF g + of the oriented, orthogonal frame bundle, along a homomorphism τ : Remark 6. For Riemannian manifolds of dimension n = 2, the classification theorem 3 continues to hold if one makes the necessary adaptations to account for the fact that π 1 (SO(2)) = Z. In this context, a Spin G -structure can be defined as in Sect. 2.5 with Spin G := (R × G)/Z, where the action of Z on R = SO(2) is by translation, and the action on G comes from the unique nontrivial The classification theorem rather simplifies the data needed to construct the bundle of fields (12) in the infinitesimally natural case. Indeed, it suffices to have: -An orientable manifold M whose universal cover is spin. The homomorphism ι * : Z 2 → π 1 (F ) is then injective, and its image ι * (Z 2 ) ⊆ π 1 (F ) is a central subgroup. -A representation (ρ, V ) of SO(η) × Z2 π 1 (F ), which is unitary when restricted to π 1 (F ), and faithful on Z 2 . -A subgroup G ⊆ U (V ) that commutes with the image of SO(η) under ρ and contains the image of π 1 (F (M )).
One can then construct the Spin G -structure Q = F + × π1(F ) G, from which one recovers the bundle of fields As discussed in Sect. 2.5, a single section of this bundle provides the metric, Levi-Civita connection, gauge fields and spinors.
In particular, the bundle (5) describing spinors and metric is simply where R is the group R := SO(η) × Z2 π 1 (F ). The principal G/Z 2 -bundle (14) describing the gauge fields is necessarily the trivial bundle P = M × G/Z 2 .
According to Theorem 3, the above setting exhausts the possibilities in the infinitesimally natural case-at least under the natural assumption that V is a faithful, unitary representation for the group G, which is then automatically compact.

Infinitesimally Natural Principal Fibre Bundles
The proof of Theorem 3 relies on the result [19,22] that every infinitesimally natural principal fibre bundle is associated to the universal cover of a kth order frame bundle. The essential new ingredient in the proof of Theorem 3 is that in the particular case of Spin G -structures, the order k must be equal to one. Before proceeding with this proof, we therefore briefly recall some results on infinitesimally natural principal bundles.

The kth Order Frame Bundle and its Universal Cover.
A kth order frame f k x at a point x ∈ M is by definition the k-jet f k x = j k 0 φ at zero of an orientation preserving local diffeomorphism φ : R n → M with φ(0) = x. The oriented kth order frame bundle π : F k+ → M is defined by with projection π : F k+ → M given by π(j k 0 φ) = φ(0). It is a principal bundle with structure group In the trivial case k = 0, we have F 0+ = M , and G(0, n) = {1}. The first interesting example is k = 1, in which case the principal fibre bundle F 1+ → M is the oriented frame bundle F + , with structure group G(1, n) = GL + (n, R). For k ≥ 2, the natural projections F k+ → F + and G(k, n) → GL + (n, R) have contractible fibres, so that π 1 (F k+ ) π 1 (F + ), and π 1 (G(k, n)) π 1 (GL + (n, R)) Z 2 .
The universal cover F k+ of the oriented k-frame bundle is, therefore, essentially determined by the universal cover F + of the ordinary frame bundle. To determine the structure group of the principal fibre bundle F k+ → M , note that the orbit map ι : G(k, n) → F k+ , defined by ι(g) = f k x g, gives rise to a group homomorphism The structure group of the principal fibre bundle F k+ → M is, therefore, where Z 2 is identified with the central subgroup {(z, ι * (z −1 )) ; z ∈ Z 2 }.

Classification Results for Principal Bundles.
Every principal fibre bundle P → M gives rise to an exact sequence of Lie algebras where aut(P ) is the Lie algebra of equivariant vector fields on P , and aut V (P ) is the Lie algebra of vertical equivariant vector fields. The latter is isomorphic to Γ(Ad(P )), the Lie algebra of infinitesimal gauge transformations. A principal fibre bundle P → M is called infinitesimally natural if it comes with a Lie algebra homomorphism σ : Vec(M ) → aut(P ) that splits the exact sequence (19).
The kth order frame bundle F k+ is an infinitesimally natural principal fibre bundle, with section D k : Vec(M ) → aut(F k+ ) defined by where t → exp(tv) is the flow on M generated by the vector field v.

Remark 7. A Spin
infinitesimally natural as a principal fibre bundle, and if u * • σ = D 1 . This additional compatibility condition expresses that the covering map u : Q → F + is a morphism of infinitesimally natural principal fibre bundles.
Since F k+ → F k+ is a discrete cover, it has a canonical flat equivariant connection ∇ can : Vec(F k+ ) G(k,n) → Vec( F k+ ) G(k,M ) . It follows that also F k+ → M is an infinitesimally natural bundle, with splitting D k = ∇ can • D k . Proof. This is Theorem 4.4 in [19]. A version of this result was proven earlier by Lecomte in [22].

Proof of the Classification Theorem
By Theorem 4, we may assume that an infinitesimally natural Spin G -structure takes the form where the group H is defined as Proof. Define the covering map u 0 : Since both u 0 and u satisfy the equivariance equation u(qh) = u(q)κ(h) for h ∈ H, the two maps differ by a gauge transformation of F + . We have u(q) = u 0 (q)g(q) for a smooth map g : If v ranges over Vec(M ) andf k over F k+ , then D k (v)|f k ranges over the full tangent bundle T F k+ . It follows that g is (ỹ). From this, we deduce that κρ(ỹ) = c −1 yc.
Using this, we can bring the infinitesimally natural Spin G -structures in the following standard form.

Standard Form of the Homomorphism.
From Lemma 6, it follows that not only the principal bundle Q → M , but also the covering map u : Q → F + is entirely determined by the homomorphism ρ : G(k, M ) → H. We proceed by deriving a standard form for ρ.
Recall from (18) that G(k, M ) G(k, n) × Z2 π 1 (F + ). Further, we have G(k, n) GL + (n, R) GL >1 , where GL >1 denotes the subgroup of k-jets that are the identity to first order. Decomposing GL(n, R) SL(n, R) × R + , we may thus consider ρ as a map If the infinitesimal Spin G -structure is in the standard form of Lemma 6, then ρ takes the following form.
But on the other hand, we have [R, gl >1 ] = gl >1 , since R represents the multiples of the Euler vector field. This yieldṡ As the intersection is zero, we haveρ(gl >1 ) = {0}.

Conclusion of the Proof of Theorem 3.
The proof of the classification theorem is completed by combining the standard form of Q (Lemma 6) with that of the homomorphism ρ (Lemma 7).

Proof of Theorem 3.
Since ρ is trivial on the group GL >1 of k-frames that agree with the identity to order 1, one deduces from Lemma 6 that By Lemma 7 and Remark 8, the homomorphism ρ : G(1, M) → H depends only on the homomorphism τ : π 1 (F + ) → G, the homomorphism γ : π 1 (M ) → R + , and the element Λ of g. Recall that Q ⊆ Q is the preimage under u : Q → F + of the bundle OF + g of oriented, orthonormal frames for the metric g. Since u ([f, (x, y)]) = fx, one can fix the representativesf ∈ F + ,x ∈ GL + (n, R), and y ∈ G of the class [f, (x, y)] ∈ u −1 (OF + g ) so thatx = 1 and f ∈ OF + g . We thus find Q = {[f, (1, y)] ;f ∈ OF + g , y ∈ G}, and hence Q OF + g × τ G.

Classification of the Splittings.
The classification theorem ensures that every infinitesimally natural Spin G -structure is isomorphic to a Spin G -structure of the form Q τ := OF + g × τ G. Note that Q τ is itself an infinitesimally natural Spin G -structure. Indeed, the principal H-bundle Q τ = F + × τ G comes with a canonical splitting σ : Vec(M ) → aut( Q τ ), induced by the splitting D for F + .
This splitting, however, is not necessarily identical to the one induced by Q. To obtain a model for Q that yields the correct natural splitting σ as well as the correct covering map u, one proceeds as follows. The metric g on M gives rise to a volume form λ. Denote by F λ ⊆ F + the principal SL(n, R)-bundle of frames with volume 1. Identifying F λ with the quotient of F + by R + , we obtain a principal R + × π 1 (F )-bundle F + → F λ .
The desired bundle Q ρ is then obtained by associating R + × G to F + → F λ along the homomorphism ρ, that is, From the proof of Theorem 3, we then obtain the following corollary. In short, an infinitesimally natural Spin G -structure (Q, u) is determined by a homomorphism τ : π 1 (F ) → G that that identifies Z 2 ⊆ π 1 (F ) with Z 2 ⊆ G. For a given Spin G -structure (Q, u), the splittings are determined by an element Λ ∈ g Im(τ ) and a class log(γ) ∈ H 1 (M, R).

Applications
It was already recognised by Hawking and Pope [16] that the existence of generalised spin structures may place restrictions on the space-time manifold M . When generalised spin structures were classified by Avis and Isham [2], it was found that if the Lie group G contains SU(2), then 'universal spin structures' in the sense of [4] exist, irrespective of the topology of M . In particular, there are no topological obstructions to the existence of a Spin G -structure as soon as SU (2) This is no longer the case for infinitesimally natural generalised spin structures. In this setting, universal spin structures exist only for certain noncompact groups. For compact G, the requirement that there exist a homomorphism π 1 (F + ) → G that maps Z 2 ⊆ π 1 (F + ) onto Z 2 ⊆ G provides an obstruction on the space-time manifold M in terms of the group G of internal symmetries.
In this section, we work out these obstructions for a number of specific gauge theories. For concreteness, we assume that M is an oriented, timeoriented, Lorentzian manifold of dimension 4. The time-orientability allows us to replace SO(η) by SL(2, C).

Weyl and Dirac Spinors
Consider a single, massless, charged Weyl spinor coupled to a U (1) gauge field. In this setting, the gauge group G is U (1), and V = C 2 ⊗ C q is the two-dimensional defining representation of SL(2, C) tensored with the onedimensional defining representation of U (1). This representation descends to Spin c = SL(2, C) × Z2 U (1). Given a Spin c -structure Q, the configuration space consists of sections of the bundle If Q is infinitesimally natural, then Theorem 3 yields a homomorphism τ : π 1 (F + ) → U (1) that sends the image of π 1 (GL + (n, R)) in π 1 (F + ) to {± 1}. If π 1 (M ) is finitely generated, then Im(τ ) ⊆ U (1) is a finitely generated subgroup containing {± 1}, hence Im(τ ) Z n × (Z/2mZ) for certain n, m ∈ N. In particular, there exists a homomorphism π 1 (F + ) → Z/2mZ ⊆ U (1) that maps the image of π 1 (GL + (n, R)) in π 1 (F + ) to {± 1}. Since every such homomorphism yields an infinitesimally natural Spin c -structure by the procedure outlined in Sect. 3.1, we arrive at the following conclusion. The topological requirements on M for admitting infinitesimally natural Spin c -structures are more restrictive than those for admitting ordinary Spin cstructures. However, they are less restrictive than those for admitting spin structures. Indeed, if m is odd, then the sequence of groups is split. Every homomorphism π 1 (F + ) → Z/2mZ then induced a homomorphism π 1 (F + ) → Z/2Z, and hence a spin structure on M . If m is even, then this sequence does not split. In that case, M may admit infinitesimally natural Spin c -structures without admitting ordinary spin structures.

Dirac Spinors.
For Dirac spinors, V is the 4-dimensional representation C 4 ⊗ C q , where the Clifford representation C 4 splits into two identical irreducible representations C 2 ⊕ C 2 under SL(2, C), the left-handed and righthanded spinors.
Note that the unitary commutant of SL(2, C) in V is U (2) rather than U (1). For a discrete subgroup H ⊆ U (2), we can, therefore, form the group U (1) H generated by H and the gauge group U (1) and consider Spin G -structures Q with structure group G = U (1) H . The generic fibre of the bundle J 1 ( Q)× Spin G V is the same for G = U (1) as it is for G = U (1) H , so adding H will not change the space of local sections.
If H is a discrete group of global symmetries of the Lagrangian, then the action is well defined for sections of this bundle. Indeed, the action is invariant under constant H-valued transformations because H is a global symmetry group, and the part of the transition functions involving H will be constant since H is discrete.
For a massive Dirac spinor, where the Lagrangian contains a term of the form mψψ, the subgroup of U (2) which preserves the Lagrangian is precisely the diagonal U (1). This means that the infinitesimally natural Spin Gstructures are precisely the infinitesimally natural Spin c -structures classified above, and there is no possibility to add a discrete subgroup H.
For massless Dirac spinors, where the term mψψ is absent, the left and right Weyl spinors decouple, so that the relevant symmetry group is U L (1) × U R (1). Although the requirement on a manifold to carry a Spin G -structure does not change, this does give us more Spin G -structures for the same manifold.
More generally, we may enlarge the gauge group G by any group H of discrete symmetries of the Lagrangian in order to obtain infinitesimally natural Spin G -structures.
The fermion representation V for a single generation can be conveniently described (see e.g. [5]) by C 2 ⊗ ∧ • C 5 , the tensor product of the defining representation of SL(2, C) and the exterior algebra of the defining representation of SU (5). Under SL(2, C) × S(U (3) × U (2)), this decomposes into 12 irreps corresponding to left-and right-handed electrons, neutrinos, up and down quarks and their antiparticles.
As the gauge group alone is of no use when trying to find a Spin Gstructure, one has to involve the group of global symmetries of the standard model Lagrangian. It contains the gauge group G, but also (at least on the classical level) the global U (1) B × U (1) L -symmetries that rotate quarks and leptons independently (these are connected to baryon and lepton number).
We conclude that the only infinitesimally natural Spin G -structures relevant to the standard model are the ones associated to homomorphisms that preserve Z 2 , the subgroup of U (1) B × U (1) L generated by (− 1, − 1). In this expression,Ĝ is the group of global symmetries of the standard model Lagrangian, which at least contains We construct an example. Consider de Sitter space H = { x ∈ R 5 | − x 2 0 + x 2 1 + x 2 2 + x 2 3 + x 2 4 = 1}, which has a pseudo-Riemannian metric g with constant curvature induced by the Minkowski metric in the ambient R 5 . Its group of orientation preserving isometries is SO (1,4), and H SO(1, 4)/SO (1,3). Denote by OF +↑ g (H) the bundle of orthogonal frames with positive orientation and time orientation. By viewing OF +↑ g (H) as a submanifold of R 5 × SO(1, 4) 0 , one can see that SO(1, 4) 0 acts freely and transitively by x : f → x * f . Therefore, OF +↑ g (H) is diffeomorphic to SO(1, 4) 0 . Now let Γ ⊆ SO(4) be a discrete group which acts freely, isometrically and properly discontinuously on S 3 . Manifolds of the type Γ\S 3 are called spherical space forms (see [40] for a complete classification). As Γ includes into SO(1, 4) 0 , it acts on H, making M = Γ\H into a pseudo-Riemannian manifold with constant curvature.
As H is simply connected, we see that π 1 (M ) = Γ. We calculate the homotopy group of the frame bundle. Because OF +↑ g (M ) is just Γ\OF +↑ g (H), it is isomorphic to Γ\SO(1, 4) 0 . Going to the universal cover, we see that OF + g (M ) = Γ\ SO(1, 4) 0 . As Γ ⊆ SO(4), we may consider Γ to be the preimage of Γ in Spin (4). As the universal cover is simply connected, it is now clear that π 1 (OF +↑ g (M )) = Γ. We get for free a homomorphismΓ → Spin(4) SU(2) l ×SU(2) r , which maps the noncontractible loop in the fibre to (− 1, −1).
Triggered by the WMAP-data on cosmic background radiation, it has been proposed that space may carry the topology of I * \S 3 , where Γ = I * is the binary icosahedral group [24,34]. Although these views are far from universally accepted [20], it is nonetheless interesting in this connection to note that M = I * \H, which has spacelike hypersurfaces I * \S 3 , allows for infinitesimally natural Spin G -structures which do not stem from Spin c -structures.
Under the identification Spin(4) SU(2) l ×SU(2) r , we see that Γ = I * ×1 lives only in SU(2) l , so thatΓ is the direct product of I * × 1 and the Z 2 generated by (− 1, −1). One can, therefore, define a homomorphism (23) by identifying SU(2) l with SU(2) L ⊆ G, and mapping (− 1, −1) to (− 1, −1) ∈ U (1) B × U (1) L . This yields an infinitesimally natural Spin G -structure which uses the noncommutativity of the gauge group in an essential fashion. This means that M = I * \H carries more infinitesimally natural Spin G -structures than just the 'ordinary' Spin c -structures.

Extensions of the Standard Model
The fact that S(U (3) × U (2)) does not contribute to the obstruction of finding infinitesimally natural Spin G -structures on M is due to the fact that it never acts by −1 on V . This is not true for some GUT-type extensions of the standard model, such as the Pati-Salam SU(2) L × SU(2) R × SU(4) model and anything which extends it, for example Spin (10).
If N is the group of order 2 generated by (− 1, − 1, − 1), then infinitesimally natural Spin G -structures in the Pati-Salam model correspond, neglecting global symmetries, to homomorphisms π 1 (F + ) → SU(2) L × SU(2) R × SU(4)/N which take Z 2 to (− 1, − 1, 1) . It is therefore possible that a space-time manifold M admits infinitesimally natural Spin G -structures for the Pati-Salam model, but not for the standard model. Indeed, M has this property if the smallest quotient of π 1 (F (M )) containing ι * (Z 2 ) is a nonabelian subgroup of SU(2) L × SU(2) R × SU(4) containing (− 1, − 1, 1). Ann. Henri Poincaré Open Access. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/ by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.