9512.net
甜梦文库
当前位置:首页 >> >>

Quantum general relativity and the classification of smooth manifolds

Quantum general relativity and the classi?cation of smooth manifolds

arXiv:gr-qc/0404088v2 17 May 2004

Hendryk Pfei?er? DAMTP, Wilberforce Road, Cambridge CB3 0WA, United Kingdom Emmanuel College, St Andrew’s Street, Cambridge CB2 3AP, United Kingdom May 17, 2004

Abstract The gauge symmetry of classical general relativity under space-time di?eomorphisms implies that any path integral quantization which can be interpreted as a sum over spacetime geometries, gives rise to a formal invariant of smooth manifolds. This is an opportunity to review results on the classi?cation of smooth, piecewise-linear and topological manifolds. It turns out that di?erential topology distinguishes the space-time dimension d = 3 + 1 from any other lower or higher dimension and relates the sought-after path integral quantization of general relativity in d = 3 + 1 with an open problem in topology, namely to construct non-trivial invariants of smooth manifolds using their piecewise-linear structure. In any dimension d ≤ 5 + 1, the classi?cation results provide us with triangulations of space-time which are not merely approximations nor introduce any physical cut-o?, but which rather capture the full information about smooth manifolds up to di?eomorphism. Conditions on re?nements of these triangulations reveal what replaces block-spin renormalization group transformations in theories with dynamical geometry. The classi?cation results ?nally suggest that it is space-time dimension rather than absence of gravitons that renders pure gravity in d = 2 + 1 a ‘topological’ theory. PACS: 04.60.-m, 04.60.Pp, 11.30.-j keywords: General covariance, di?eomorphism, quantum gravity, spin foam model

1

Introduction

Space-time di?eomorphisms form a gauge symmetry of classical general relativity. This is an immediate consequence of the fact that the theory can be formulated in a coordinate-free fashion and that the classical observables are able to probe only the coordinate-independent aspects of space-time physics, i.e. they probe the ‘geometry’ of space-time which always means ‘geometry up to di?eomorphism’. The classical histories of the gravitational ?eld in the covariant Lagrangian language are therefore the equivalence classes of space-time geometries modulo di?eomorphism. In the present article, we consider path integral quantizations of classical general relativity in d space-time dimensions in which the path integral is a sum over the histories of the gravitational ?eld. For any given smooth d-dimensional space-time manifold M with boundary
?

E-mail: H.Pfeiffer@damtp.cam.ac.uk

1

2

Quantum general relativity and the classification of smooth manifolds

?M , the path integral [1] with suitable boundary conditions is supposed to yield transition amplitudes between quantum states (equation (3.2) below). The states correspond to wave functionals on a suitable set of con?gurations that represent 3-geometries on the boundary ?M . The only background structure involved is the di?erentiable structure of M . It is known that path integrals of this type are closely related to Topological Quantum Field Theories (TQFTs) [2,3]. In order to avoid possible misunderstandings of this connection right in the beginning, we stress that in the case of general relativity in d = 3 + 1, there is no reason to require that the vector spaces [or modules] in the axioms of [2] be ?nite-dimensional [?nitely generated]. In fact, it is believed that TQFTs with ?nite-dimensional vector spaces would be insu?cient and unable to capture the propagating modes of general relativity in d = 3 + 1. A similar argument is thought to apply to general relativity in d = 2 + 1 if coupled to certain matter, for example, to a scalar ?eld. We also stress that in the literature, the letter ‘T’ in TQFT does not necessarily refer to topological manifolds. In fact, the entire formalism of TQFTs is usually set up in the framework of smooth manifolds [2], and unless d ≤ 2 + 1, the structure of smooth manifolds is in general rather di?erent from that of topological manifolds. One should therefore distinguish smooth from topological manifolds and relate the path integral of general relativity to a TQFT that uses smooth manifolds. We call such a theory a C ∞ -QFT. In contrast to the smooth case, we will subsequently use the term C 0 -QFT for a TQFT that refers to topological manifolds. One application of the connection of general relativity with C ∞ -QFTs is that the partition function which can be computed from the path integral, forms (at least formally) an invariant of smooth manifolds [2, 3]. Results on the classi?cation of topological, piecewise-linear and smooth manifolds which we review in this article, can then be used in order to narrow down some properties of the path integral. In dimension d ≤ 2 + 1, smooth manifolds up to di?eomorphism are already characterized by their underlying topological manifolds up to homeomorphism. This means that in these dimensions, the C ∞ -QFT of general relativity is in fact a C 0 -QFT which makes the colloquial assertion precise that pure general relativity in d = 2+1 is a ‘topological’ theory. The same is no longer true in d = 3 + 1. A given topological 4-manifold can rather admit many inequivalent di?erentiable structures so that in d = 3 + 1, there is a highly non-trivial di?erence between C ∞ -QFTs and C 0 -QFTs. The invariants of Donaldson [4] or Seiberg– Witten [5] can indeed be understood as partition functions of (generalized) C ∞ -QFTs [2] which are sharp enough to detect the inequivalence of di?erentiable structures on the same underlying topological manifold. The partition function of quantum general relativity in d = 3 + 1, if it can indeed be constructed, will o?er the same potential. This relationship is the main theme of the present article. The special role of space-time dimension d = 3 + 1 in di?erential topology is summarized by the following result. (without boundary if Theorem 1.1. Let M be a compact topological d-manifold, d ∈ d = 5). If M admits an in?nite number of pairwise inequivalent di?erentiable structures, then d = 4. This is a corollary of several theorems by various authors. We explain in this article why this result is related to the search for a quantum theory of general relativity. Since in d ≥ 3 + 1, smooth manifolds up to di?eomorphism are in general no longer characterized by their underlying topological manifolds up to homeomorphism, general relativity is no longer related to a C 0 -QFT. There is, however, another classi?cation result that is still

?

Quantum general relativity and the classification of smooth manifolds

3

applicable in any d ≤ 5+1: smooth manifolds up to di?eomorphism are characterized by their Whitehead triangulations up to equivalence (PL-isomorphism). We explain these concepts in greater detail below. They provide us with a very special type of triangulations that can be used in order to discretize smooth1 manifolds in a way which is not merely an approximation nor introduces a physical cut-o?, but which rather captures the full information about the equivalence class of di?erentiable structures. This suggests that the path integral of general relativity in d ≤ 5 + 1 admits a discrete formulation on such triangulations. General relativity in d ≤ 5 + 1 is therefore related to what we call a PL-QFT, i.e. a TQFT based of piecewise-linear manifolds. In particular, a path integral quantization of general relativity in d ≤ 5 + 1 is related to the construction of invariants of piecewise-linear manifolds. From the classi?cation results, we will see that this is most interesting and in fact an unsolved problem in topology, precisely if d = 3 + 1. It is the decision to take the di?eomorphism gauge symmetry seriously which singles out d = 3 + 1 this way. The di?eomorphism invariance of the classical observables then implies in the language of the triangulations that all physical quantities computed from the path integral, are independent of which triangulation is chosen. The discrete formulation on some particular triangulation therefore amounts to a complete ?xing of the gauge freedom under space-time di?eomorphisms. The relevant triangulations can furthermore be characterized by abstract combinatorial data, and the condition of equivalence of triangulations can be stated as a local criterion, in terms of so-called Pachner moves. ‘Local’ here means that only a few neighbouring simplices of the triangulation are involved in each step. A comparison of Pachner moves with the block-spin or coarse graining renormalization group transformations in Wilson’s language reveals what renormalization means for theories with dynamical geometry for which there exists no a priori background geometry with which we could compare the dynamical scale of the theory. We will ?nally see that the absence of propagating solutions to the classical ?eld equations, for example in pure general relativity in d = 2 + 1, is related 1. neither to the question of whether the path integral corresponds to a C 0 -QFT (as opposed to a C ∞ -QFT), 2. nor to the question of whether the vector spaces of this C 0 -QFT or C ∞ -QFT are ?nitedimensional, 3. nor to the question of whether the theory admits a triangulation independent discretization. The present article is structured as follows. In Section 2, we review the classical theory and its gauge symmetry. The formal properties of path integrals and their connection with C ∞ -QFTs and manifold invariants are summarized in Section 3. In Section 4, we then compile the relevant results on the classi?cation of the various types of manifolds and discuss their physical signi?cance. In Section 5, we sketch the special case of d = 2 + 1. Section 6 ?nally contains speculations on the coincidence of open problems in physics and mathematics and on how to narrow down the path integral in d = 3 + 1. We try to make this article selfcontained by including a rather extensive Appendix which contains all relevant de?nitions from topology.
1 This might be unexpected at ?rst sight, but in generic dimension d ≤ 5 + 1, it is smooth rather than topological manifolds that correspond to triangulations in this way.

4

Quantum general relativity and the classification of smooth manifolds
?2 ? ??1 1

?d
?1

?d
?2

U12 U1 U2 M

Figure 1: A manifold M with two coordinate systems (U1 , ?1 ) and (U2 , ?2 ) whose patches have some overlap U12 = U1 ∩ U2 , and the corresponding transition function ?2 ? ??1 . 1

2

The classical theory

This section contains merely review material: Subsection 2.1 on smooth manifolds, Subsection 2.2 on the ?rst order formulation of general relativity, Subsection 2.3 on its gauge symmetries, Subsection 2.4 on the role of di?erential topology in the study of general relativity, and Subsection 2.5 on space-time di?eomorphisms. Each of these subsections can be safely skipped. We nevertheless include the material here in order to ?x the terminology, in particular in order to resolve the various misunderstandings that can arise in the discussion of di?eomorphisms in general relativity, just because there seems to exist no standardized terminology in the literature.

2.1

Smooth manifolds

We ?rst review some basic facts about smooth manifolds. Detailed de?nitions can be found in the Appendix. A d-dimensional manifold M is a suitable topological space which is covered by coordinate systems (Figure 1). A coordinate system (Ui , ?i ) is a patch Ui ? M together with a one-toone map ?i : Ui → ?i (Ui ) ? d onto some subset of the standard space d . We can use the coordinate maps ?i in order to assign d real coordinates ?? (p), ? = 0, . . . , d ? 1, with each i point p ∈ Ui . The coordinates take values in the subset ?i (Ui ) ? d . As soon as two coordinate systems (Ui , ?i ), (Uj , ?j ) have a non-empty overlap Uij = Ui ∩ Uj = ?, we can change the coordinates by means of the transition function ?ji := ?j ? ??1 : ?i (Uij ) → ?j (Uij ) which is a map from a subset of d to d . i A scalar function α : M → can be described in any of the coordinate systems in terms of the functions α ? ??1 : ?i (Ui ) → from some subset of d to . i

?

?

?

?

?

?

?

?

?

Quantum general relativity and the classification of smooth manifolds

5

If we require all coordinate maps ?i to be homeomorphisms, i.e. continuous with continuous inverse, we obtain the de?nition of a topological manifold. In this case, the transition functions are homeomorphisms, too, and it makes sense to study continuous functions α : M → and homeomorphisms f : M → N between manifolds. If we want to talk about di?erential equations on a space-time manifold, we have to know in addition how to di?erentiate functions and therefore have to impose additional structure. This can be accomplished by restricting the transition functions from homeomorphisms to some special subclass of functions. A smooth manifold, for example, is a topological manifold for which all transition functions ?ji and their inverses are C ∞ , i.e. all partial derivatives of all orders exist and are continuous. A covering of M with such coordinate systems is known as a di?erentiable structure. A continuous function α : M → is called smooth if all its coordinate representations α ? ??1 : ?i (Ui ) → are C ∞ -functions. A homeomorphism f : M → N i between smooth manifolds is called a di?eomorphism if all coordinate representations ψj ? f ? ??1 : ?i (Ui ) → ψj (Vj ) and their inverses are C ∞ . Here (Vj , ψj ) denotes the coordinate i systems of N . A substantial part of the present article is concerned with the various structures one can impose on topological manifolds by restricting the transition functions, and with the question of how to compare these structures.

?

?

?

2.2

Two pictures for general relativity

For background on classical general relativity, we refer to the textbooks, for example [6, 7]. We are interested in general relativity in d-dimensional space-time and, for simplicity, often restrict ourselves to pure gravity without matter. Space-time is given by a smooth oriented d-manifold M . Second order metric picture. We denote the (smooth) metric tensor by g?ν from which one can calculate the unique metric compatible and torsion-free connection ?? , its Riemann curvature tensor R? νρσ , the Ricci tensor R?ν = Rρ ?ρν and the scalar curvature R = g?ν R?ν . The Einstein–Hilbert action (without matter) reads, S[g] = 1 16π G
M

(R ? 2Λ)

| det g| dx0 ∧ · · · ∧ dxn?1 ,

(2.1)

where G is the gravitational and Λ the cosmological constant. Variation with respect to the metric yields the Einstein equations (without matter), 1 R?ν ? (R ? 2Λ)g?ν = 0, 2 which are second order di?erential equations for the metric tensor g?ν . First order Hilbert–Palatini picture. We sometimes contrast this second order metric picture with the ?rst order Hilbert–Palatini formulation. For a recent review, see, for example [8]. We therefore choose local orthonormal basis vectors eI ∈ Tp M , I = 0, . . . , d ? 1, of the tangent spaces at each point p ∈ M , i.e. e? eν g?ν = ηIJ where η = diag(?1, 1, . . . , 1) denotes I J ? the standard Lorentzian bilinear form. The dual basis (eI ) of 1-forms eI = eI dx? ∈ Tp M , ? I (e ) = δ I , is called the coframe ?eld. Denote by A an SO(1, d ? 1)-connection whose i.e. e J J (2.2)

6

Quantum general relativity and the classification of smooth manifolds

curvature 2-form we write as an so(1, d ? 1)-valued 2-form F I J = dAI J + AI K ∧ AK J on M . Classical general relativity can be formulated as a ?rst order theory with the Hilbert–Palatini action, S[A, e] = 1 16 (d ? 2)! π G
M

εIJK1 ···Kd?2 eK1 ∧ · · · ∧ eKd?2 ∧ F IJ ?

2Λ eI ∧ eJ , (2.3) d(d ? 1)

where F IJ = F I K η KJ . Variation with respect to the connection A and the cotetrad e yields the ?eld equations, 0 = εIJP K1 ···Kd?3 eK1 ∧ · · · ∧ eKd?3 ∧ F IJ ? 0 = 2Λ eI ∧ eJ , (d ? 1)(d ? 2) (2.4a) (2.4b)

I δL d + 2 AI L ∧ εLJ K1 ···Kd?2 eK1 ∧ · · · ∧ eKd?2 ,

for all I, J, P . These are coupled ?rst order di?erential equations for A and e. Whenever the cotetrad is non-degenerate, the ?rst of these equations is equivalent to (2.2) for the metric tensor g?ν = eI eJ ηIJ , (2.5) ? ν while the second equation states that the connection is torsion-free. Note that any SO(1, d ? 1)-connection is always metric compatible with respect to (2.5). One reason for introducing the Hilbert–Palatini formulation is to demonstrate that we can easily adopt a point of view in which the metric tensor is not a fundamental ?eld of the classical theory. This illustrates once more that a generic smooth manifold M is the only input of the theory. The ?elds A and e and, as a consequence, the metric g are determined by the dynamics of the theory.

2.3

Gauge symmetries

Let us review the gauge symmetries of the classical theory in the ?rst order picture. Local Lorentz symmetry. The choice of local orthonormal bases (eI ) is unique only up to a local Lorentz transformation. Such transformations can be expressed in a coordinate system U ? M , ? : U → d by some smooth Lorentz-group valued function Λ : U → SO(1, d ? 1). We write x = ?(p), p ∈ U , and denote the old variables by (A, e) and the transformed ones by (A, e),

?

eI (x) = ΛI J (x)eJ (x), ? ? A? I J (x) = ΛI K (x)A? K L (x)ΛJ L (x) + ΛI K (x) ? ΛJ K (x), ?x?

(2.6a) (2.6b)

where we write ΛJ K = ηJI η KL ΛI L . The metric (2.5) is obviously invariant under (2.6a). Space-time di?eomorphisms. Any two geometries (M, g) and (M, g′ ) of M are physically identical as soon as they are related by a space-time di?eomorphism f : M → M , i.e. g′ = f ? g. Let us describe the action of some di?eomorphism f : M → M on the various ?elds in detail. Consider a smooth function α, a smooth vector ?eld X = X ? ?? , a smooth 1-form

Quantum general relativity and the classification of smooth manifolds

7

ω = ω? dx? , and the coframe ?eld eI dx? . For p ∈ M , choose a coordinate system (U, ?) such ? that p ∈ U and write x = ?(p). The di?eomorphism acts in coordinates as follows, (f ? α)(x) = α(f (x)), (f X) (x) = (Df
? ? ? ?1

(2.7a)
? ν

(f (x)))

X (f (x)),
ν

ν

(2.7b) (2.7c) (2.7d) (2.7e)

(f ω)? (x) = ων (f (x)) (Df (x)) ? , (f ? e)I (x) = eI (f (x)) (Df (x))ν ? , ν ? (f ? A)? I J (x) = Aν I J (f (x)) (Df (x))ν ? ,

where (Df (x))? ν := ?f ? (x)/?xν denotes the Jacobi matrix of f , written in coordinates, too. These rules also determine the action of the di?eomorphism on higher rank tensors2 .

2.4

Di?erential topology versus di?erential geometry

The problem of classical general relativity can be summarized as follows. Second order picture. Given a smooth oriented d-manifold M with boundary ?M , ?nd a smooth metric tensor g?ν that satis?es the Einstein equations (2.2) in the interior of M and suitable boundary conditions on ?M . Study existence and uniqueness of the solutions. ′ Any two solutions g?ν , g?ν that are related by a space-time di?eomorphism f : M → M , i.e. g′ = f ? g, are physically identical. First order picture. Given a smooth oriented d-manifold M with boundary ?M , ?nd a smooth SO(1, d ? 1)-connection A and a smooth non-degenerate coframe ?eld e that satisfy the ?rst order ?eld equations (2.4) in the interior of M and suitable boundary conditions on ?M . Study existence and uniqueness of the solutions. Any two solutions (A, e), (A′ , e′ ) that are related by a local Lorentz transformation (2.6) or by a space-time di?eomorphism (2.7), are physically identical. Interpretation. We stress that the input for classical general relativity is a smooth manifold, i.e. a topological manifold with a di?erentiable structure. We are therefore in the realm of di?erential topology. This has to be contrasted, for example, with the non-generally relativistic treatment of Yang–Mills theory in an a priori ?xed space-time geometry. Such a background geometry is described by a Riemannian manifold (M, g), i.e. by a smooth manifold M with a metric tensor g?ν , which renders such a theory a problem of di?erential geometry rather than di?erential topology. Boundary conditions. We do not comment on how the boundary and/or initial conditions a?ect the existence and uniqueness of the classical solutions. For simplicity, we use the following speci?cation of boundary data in the ?rst order formulation. In the variational principle, we ?x the connection A|?M at the boundary ?M and assume that the variations δA and δe are supported only in the interior of M . The ?eld equations (2.4) in the interior of M can then be derived without additional boundary terms for the action.
2 We have de?ned the di?eomorphism action on eI so that it acts only on the cotangent index ?, but not ? on the internal index I.

8

Quantum general relativity and the classification of smooth manifolds

This choice is made entirely for convenience. It is known to be satisfactory in the toy model of pure gravity in d = 2 + 1, but might have to be revised in order to ?nd the correct path integral in d = 3 + 1.

2.5

Gauge symmetries and quantization

Assuming that there exists a quantization of general relativity, will the quantum theory respect the classical gauge symmetries, or not? In order to get some more insight into this question, let us recall why general relativity possesses such symmetries in the ?rst place. The answer is quite standard [7], but nevertheless worth spelling out in detail, in particular because this symmetry plays an important role in the remainder of the present article. The entire framework of smooth manifolds has been set up in order to describe physical quantities by mathematical constructions that have a meaning independently of the coordinate systems which we choose in order to represent them. Consider, for example, some physical ?eld which is given by a smooth scalar function α : M → . There are various possible coordinate systems (Ui , ?i ) in order to represent this scalar ?eld. The reader may think of ?at space-time with Cartesian or polar coordinates which give rise to coordinate representations of the ?eld, α ? ??1 : ?(Uj ) → , and whose j transition functions ?ji = ?j ? ??1 : ?i (Ui ∩ Uj ) ? d → ?j (Ui ∩ Uj ) ? d prescribe how to i change the coordinates. One might now be tempted to think that the physical reality is described literally by the set M with its points p, q ∈ M which would symbolize space-time events. Starting from this ‘reality’, one would then construct various coordinate systems ?i : Ui ? M → d in order to represent this ‘reality’ in terms of coordinates. The transition functions guarantee that iterated coordinate changes always give consistent results. Certainly, the collection of all the ranges ?j (Uj ) ? d of the coordinate systems together with the transition functions ?j ? ??1 and the coordinate representations α ? ??1 of the i j scalar function constitute the maximum information about our ‘reality’ that can be written down by the experimenters (who always use coordinate systems). This raises the question of whether, conversely, we can reconstruct M together with its points p, q ∈ M and the function α: M → from such a collection of coordinate ranges, transition functions and coordinate representations. The well-known answer to this question is ‘no’ [7]. One can reconstruct3 M and α only up to di?eomorphism. The actual reality is therefore not given literally by the set M and the function α : M → , but rather by equivalence classes [(M, α)] modulo di?eomorphism. Here (M, α) and (M ′ , α′ ) are considered equivalent if and only if there exists a di?eomorphism f : M → M ′ such that α′ = f ? α. We have called this a gauge symmetry simply because there are several di?erent mathematical ways (M, α), (M ′ , α′ ), . . . of specifying a single physical history. If we now attempt to quantize such a theory, we have to remember that the classical con?gurations (histories) are the equivalence classes [(M, α)] rather than the particular representatives (M, α). The above argument indicates that the quantum theory should not violate this type of gauge symmetry because otherwise the outcome of quantum experiments would depend, very roughly speaking, on whether the classical observer who performs the measurement, uses Cartesian or polar coordinates in order to write down the observations.

?

?

?

?

?

?

?

?

A closely related result in mathematics is the fact that ?bre-bundles can be reconstructed from their transition functions, but only up to bundle automorphisms, i.e. compositions of local frame transformations and di?eomorphisms.

3

Quantum general relativity and the classification of smooth manifolds

9

As classical general relativity is formulated in the language of smooth manifolds, it is automatically compatible with the action of space-time di?eomorphisms4 f : M → M and therefore well-de?ned as a theory on the equivalence classes of geometries. It is thus safe to choose a particular representative of the geometry whenever convenient and to perform the relevant calculations for the representative. When one tries to apply a quantization procedure to the theory, the same is no longer obvious. For many quantization schemes, one is forced to choose representatives and to write down all the physical ?elds as particular functions α : M → , i.e. as ?elds on space-time. The study of the equivalence classes and the central question of whether the quantization scheme was indeed compatible with space-time di?eomorphisms, are often postponed. In the subsequent sections, we insist on working with the equivalence classes. This leads us in particular to the question of when two given smooth manifolds are di?eomorphic. What is the di?erence between general relativity and other theories such as the nongenerally relativistic treatment of Yang–Mills theory? That theory, too, can be formulated in a coordinate-free fashion and therefore shares the same type of gauge symmetry, i.e. any two representations (M, g, A, ψ) and (M ′ , g′ , A′ , ψ ′ ) are physically equivalent if and only if they are related by a di?eomorphism f : M → M ′ , i.e. g′ = f ? g, A′ = f ? A, and ψ ′ = f ? ψ. Here (M, g) is the underlying Riemannian manifold and A and ψ denote the additional ?elds, in the Yang–Mills case a connection and some charged fermion ?elds. Up to this point, there is no di?erence compared with general relativity at all. Even non-relativistic Newtonian mechanics can be written down in a coordinate-free fashion and shares all the properties mentioned here, see, for example p. 300 of [6]. In the non-generally relativistic treatment of ?eld theories, for example of Yang–Mills theory, the space-time metric g is, however, non-dynamical. This allows us to ?x a special pair (M, g) forever and to study the isometries of the Riemannian manifold (M, g), i.e. those di?eomorphisms f : M → M for which f ? g = g. In the common non-generally relativistic terminology, these isometries are often called active transformations because they ‘actively’ move the ?elds A and ψ relative to the ?xed background (M, g). These active transformations do not form a gauge symmetry because one can actually measure whether some object in the laboratory has been translated or not. Of course, the laboratory is here viewed as ?xed with respect to the background (M, g). If (M, g) does not have enough isometries, one may resort to the study of (local) Killing vector ?elds, but the interpretation would not change. The notion of active transformation requires an a priori decomposition of the physical ?elds into those that form the background, usually the space-time metric, which is non-dynamical, plus other ?elds that live on this background and which are treated as dynamical. The term passive transformation usually refers to the coordinate changes via transition functions, for example to the transition from Cartesian to polar coordinates in some ?at geometry. The space-time di?eomorphisms f : M → M ′ that relate two representatives (M, g, A, ψ) and (M ′ , g′ , A′ , ψ ′ ) of the same physical con?guration and which always form a gauge symmetry of the theory, do not have any special name in the jargon of non-generally relativistic physics. The standard treatment of non-generally relativistic ?eld theories completely focuses on the active transformations and does not discuss the generic di?eomorphism gauge symmetry which is nevertheless present. In full general relativity, however, the metric is considered

?

Strictly speaking, the generic transformations are di?eomorphisms f : M → M ′ , but there are various di?eomorphisms f1 , f2 , . . . : M → M ′ between the same pair of manifolds, and we can use the ?rst, f1 , in order to identify M ′ ≡ M and then obtain a di?eomorphism M → M from the second, etc.. One can therefore say that the symmetry is given by space-time di?eomorphisms M → M .

4

10

Quantum general relativity and the classification of smooth manifolds
Σ2 Σ1 M

˙ Figure 2: A d-manifold M with boundary ?M = Σ1 ∪Σ2 .

dynamical so that the separation into a Riemannian manifold plus ?elds that live is this geometry, is no longer available. The active transformations of non-generally relativistic physics therefore have no correspondence in the full theory of general relativity.

3

Path integrals

In this section, we review the formal properties of path integrals for general relativity (Subsection 3.1) and exhibit their close relationship with the axioms of TQFT, more precisely C ∞ -QFT [2] (Subsection 3.2). The connection of C ∞ -QFT with theories that have di?eomorphisms as a gauge symmetry, is already familiar from Witten’s work on Chern–Simons theory and knot invariants [3] although in three dimensions, the equivalence classes of smooth manifolds up to di?eomorphism are in one-to-one correspondence with those of topological manifolds up to homeomorphism. Barrett, Crane and Baez–Dolan [9–11] have explicitly proposed C ∞ -QFT as a framework for the quantization of general relativity in d = 3+1. We here recall the central ideas of this connection and outline its relationship with the classi?cation of manifolds (Subsection 3.3). In the subsequent sections, we explain why this framework is highly dimension-dependent and why the di?eomorphism gauge symmetry alone is already su?cient to single out d = 3 + 1. We ?nally speculate about an extension of the framework of C ∞ -QFT in order to better deal with the notion of time in general relativity (Subsection 3.4).

3.1

Formal properties

Path integral quantizations of general relativity, see, for example [1], are supposed to share the following formal properties. The subsequent discussion is purely heuristic. Hilbert spaces. We associate Hilbert spaces H(Σ) with closed (d ? 1)-manifolds Σ. Such a (d ? 1)-manifold Σ represents, for example, a space-like hyper-surface on which the canonical variables of the theory are de?ned. Choosing a polarization, we have a position representation H = L2 (A) with suitable wave functionals on some set A of canonical coordinates. For de?niteness, let us imagine that we choose a connection representation so that A denotes the set of all connections A|Σ . Transition amplitudes. The path integral is supposed to describe transition amplitudes from H(Σ1 ) to H(Σ2 ) for hyper-surfaces Σ1 and Σ2 , by summing over all histories of the gravitational ?eld, i.e. over all space-time geometries, that interpolate between the ‘initial’

Quantum general relativity and the classification of smooth manifolds

11

Σ1

M1

Σ2 M2 M

Σ3

˙ Figure 3: Two d-manifolds M1 , M2 with boundaries ?M1 = Σ? ∪Σ2 and ?M2 = 1 ˙ Σ? ∪Σ3 glued together along their common boundary component Σ2 . The result 2 is M = M1 ∪Σ2 M2 .

Σ1 and the ‘?nal’ hyper-surface Σ2 . We therefore choose a d-manifold M whose boundary ˙ ?M = Σ1 ∪Σ2 is the disjoint union of Σ1 and Σ2 in order to support these histories (Figure 2). For connection eigenstates A|Σj ∈ H(Σj ), j = 1, 2, the transition amplitudes are given by the (suitably normalized) path integral, A|Σ2 | T (M ) |A|Σ1 =
A|Σ1 ,A|Σ2

DA De exp

i

S[A, e] .

(3.1)

These are the matrix elements of the linear transition map T (M ) : H(Σ1 ) → H(Σ2 ). The path integration DA De comprises the integration over all connections A compatible with the boundary conditions, A|Σ1 and A|Σ2 , and over all coframe ?elds e. Notice that the connection representation for the Hilbert spaces H(Σ) is compatible with our choice of boundary conditions made above, namely to ?x the connection A|?M on the boundary. We work with ‘real’ time so that there is an i in the exponent. For generic states ψ(A|Σj ) ∈ H(Σj ), j = 1, 2, the path integral reads, ψ(A|Σ2 )| T (M ) |ψ(A|Σ1 ) = DA De ψ(A|Σ2 )ψ(A|Σ1 ) exp i S[A, e] , (3.2)

where the integration over connections A is now unrestricted. Gauge symmetry. All manifolds and functions are assumed to be smooth, and everything should be speci?ed only ‘up to local Lorentz transformations’ and ‘up to di?eomorphisms’ in a suitable way in order to implement the gauge symmetry of general relativity. The Hilbert spaces therefore represent states of spatial (d ? 1)-geometries up to di?eomorphism, etc..

3.2

Axioms of C ∞ -QFT

The framework sketched above is remarkably similar to the axiomatic de?nition of C ∞ QFT [2]. The main advantage of axiomatic C ∞ -QFT over Subsection 3.1, and almost the only di?erence, is that the di?eomorphism gauge symmetry is carefully taken into account. We give here a slightly simpli?ed account of the original axioms. Note that the condition of

12

Quantum general relativity and the classification of smooth manifolds

?nite-dimensionality guarantees that everything is well de?ned, but may ?nally have to be relaxed for general relativity. In the following, all manifolds are smooth and oriented, and all di?eomorphisms are orientation preserving. A d-dimensional C ∞ -QFT, (S1) assigns to each closed (d ? 1)-manifold Σ a [?nite-dimensional] complex vector space H(Σ), and (S2) assigns to each compact d-manifold M with boundary ?M a vector T (M ) ∈ H(?M ), such that the following axioms (A1)–(A4) hold. (A1) The vector space of some opposite oriented manifold Σ? is the dual vector space, H(Σ? ) = H(Σ)? (Bra-vectors live in the space dual to that of ket-vectors). (A2) The vector space associated with a disjoint union of (d ? 1)-manifolds Σ1 , Σ2 is the ˙ tensor product, H(Σ1 ∪Σ2 ) ? H(Σ1 ) ? H(Σ2 ) (The Hilbert space of a composite system = is the tensor product of the spaces of the constituents). (A3) For each di?eomorphism f : Σ1 → Σ2 between closed (d?1)-manifolds Σj , there exists a linear isomorphism f ′ : H(Σ1 ) → H(Σ2 ). If f and g : Σ2 → Σ3 are di?eomorphisms, then (g ? f )′ = g′ ? f ′ . Furthermore, for each di?eomorphism f : M1 → M2 between compact d-manifolds Mj with boundary ?Mj = Σj , it is required that T (M2 ) = f |′ 1 (T (M1 )), Σ where f |Σ1 : Σ1 → Σ2 denotes the restriction to the boundary (This is the implementation of the di?eomorphism gauge symmetry). The axioms (A1) and (A2) imply that any d-manifold M whose boundary is a disjoint union of ˙ the form ?M = Σ? ∪Σ2 , is assigned a vector T (M ) ∈ H(Σ1 )? ?H(Σ2 ) ? Hom (H(Σ1 ), H(Σ2 )), = 1 i.e. a linear map T (M ) : H(Σ1 ) → H(Σ2 ) (These are the desired transition maps (3.2)). ˙ ˙ (A4) If compact d-manifolds M1 , M2 with boundaries ?M1 = Σ? ∪Σ2 and ?M2 = Σ? ∪Σ3 are 1 2 glued together along their common boundary component Σ2 (Figure 3), the resulting manifold M = M2 ∪Σ2 M1 yields the composition of the transition maps, T (M2 ∪Σ2 M1 ) = T (M2 ) ? T (M1 ) which is a linear map from H(Σ1 ) to H(Σ3 ). Notice that the local Lorentz symmetry was not explicitly mentioned. It is usually taken into account by an appropriate choice of vector spaces H(Σ). The change of spatial topology (more precisely of the spatial smooth manifold) is possible in this framework if Σ1 and Σ2 in Figure 2 are not di?eomorphic. The precise form of this topology change is encoded in the choice of M . The d-manifold M itself, however, is always ?xed. Generalizations that include the ‘superposition’ of di?erent d-manifolds M are not covered by the correspondence principle and would therefore require new physical assumptions.

3.3

Invariants of smooth manifolds

The axioms have some interesting consequences. 1. Axiom (A4) applied to the case of disjoint manifolds M1 and M2 , i.e. ?M1 = Σ1 , ?M2 = Σ2 , Σ1 ∩ Σ2 = ?, implies that the associated vector is just the tensor product ˙ of vectors, T (M1 ∪M2 ) = T (M1 ) ? T (M2 ) ∈ H(Σ1 ) ? H(Σ2 ) (State of a system that is composed from independent constituents). 2. Axiom (A2) applied to Σ2 = ? shows that H(?) = or otherwise all H(Σ) are null. 3. Axiom (A4) applied to M1 = M2 = ? implies that T (?) ∈ H(?) = satis?es T (?) = 1 or otherwise all T (M ) = 0.

Quantum general relativity and the classification of smooth manifolds

13

Σ

? =

Σ

Figure 4: Gluing together two cylinders Σ × I along one of their boundaries Σ gives another cylinder (up to di?eomorphism).

M Σ1 Σ2 Mf

f

˙ Figure 5: If a compact d-manifold M has got some boundary ?M = Σ? ∪Σ2 whose 1 components are di?eomorphic, one can use the di?eomorphism f : Σ2 → Σ1 in order to glue them together and form the closed manifold Mf .

4. Axiom (A4) applied to a cylinder M = Σ×I where Σ is a closed (d?1)-manifold and I = [0, 1] the unit interval (Figure 4), shows that the transition map T (M ) : H(Σ) → H(Σ) is a projection operator, T (M )2 = T (M ). For any d-manifold that can be written as a cylinder glued to something else, this projector is e?ective so that what matters is only its image. Therefore, one often imposes the additional condition that T (Σ×I) = idH(Σ) . It is well-known [2] that the appearance of these projection operators is closely related to the vanishing of the Hamiltonian, H = 0, in the corresponding canonical formulation of the theory. This is in fact a prominent feature of general relativity [1] as soon as the 3+1 splitting for the Hamiltonian formulation is done with respect to coordinate time. In this sense, both axiomatic C ∞ -QFT and canonical general relativity are ‘non-dynamical’ theories. This is in both cases a direct consequence of the di?eomorphism gauge symmetry. 5. For closed d-manifolds, ?M = ?, so that T (M ) ∈ H(?) = . Therefore, T (M ) is just a number that depends on the di?eomorphism class of the smooth manifold M . It is an invariant of smooth manifolds! In analogy with statistical mechanics, this number is called the partition function of M and denoted by Z(M ) := T (M ). 6. More generally, for a compact d-manifold M whose boundary is of the form ?M = ˙ Σ? ∪Σ2 where Σ1 and Σ2 are di?eomorphic, one can use such a di?eomorphism f : Σ2 → 1 Σ1 in order to glue M to itself at its boundary so that one obtains a closed manifold Mf (Figure 5). In this case, by axiom (A3), there is an isomorphism of vector spaces,

14

Quantum general relativity and the classification of smooth manifolds f ′ : H(Σ2 ) → H(Σ1 ) so that the partition function of Mf can be written as the trace, Z(Mf ) = trH(Σ1 ) (f ′ ? T (M )). (3.3)

In the language of the path integral (3.1), this calculation of the trace corresponds to integrating over all possible boundary conditions on the identi?ed Σ1 ≡f Σ2 . If actually Σ1 = Σ2 = Σ and f = idH(Σ) , Z(MidΣ ) = trH(Σ) T (M ) = DA De exp i S[A, e] . (3.4)

This formula for Z(MidΣ ) in terms of the unrestricted path integral (over the manifold MidΣ ) motivates the term partition function. The most important implications are the items (5.) and (6.). It is essentially a consequence of the di?eomorphism gauge symmetry that the partition function Z(M ) which can be computed from the path integral, is an invariant of smooth manifolds. Although the partition function itself does not have any physical meaning, it is therefore mathematically very valuable. The physically relevant objects are the matrix elements (3.2). They are more general than just the partition function which is their trace, but for the beginning, let us focus on the partition function. In the subsequent sections, we ask in which space-time dimensions we can expect interesting theories, for example, ? Are smooth manifolds up to di?eomorphism already characterized by simpler structures, for example, by their underlying topological manifolds up to homeomorphism or even up to homotopy equivalence? ? In which space-time dimensions is it possible to compute the partition function more e?ciently in a purely combinatorial context using suitable triangulations of the spacetime manifold? Which role do these triangulations play in general relativity?

3.4

Extensions of the framework of C ∞ -QFT

Dimensionality. As mentioned before, for general relativity it may be necessary to drop the words ‘?nite-dimensional’ from (S1) above. In this case, one has to be more careful with the notion of dual vector space and with the construction of the traces which may become in?nite. We illustrate this below in Section 5 for the example of quantum gravity in d = 2 + 1. Hermitean structures. A Hermitean scalar product on the vector spaces H(Σ) gives rise to an isomorphism H(Σ? ) = H(Σ)? ? H(Σ) where H(Σ) denotes the vector space H(Σ) with = the complex conjugate action of the scalars. In this case, it is possible to add the axiom, (A5) For each d-manifold M , the vector T (M ) ∈ H(?M ) satis?es T (M ? ) = T (M ). ˙ If ?M = Σ? ∪Σ2 and T (M ) : H(Σ1 ) → H(Σ2 ) is viewed as a linear map, this axiom relates 1 orientation reversal of space-time with the hermitean adjoint of the transition map, T (M ? ) = T (M )? .

Quantum general relativity and the classification of smooth manifolds

15

Unitarity. Assume that the axiom (A5) is satis?ed. If the transition map T (M ) : H(Σ1 ) → ˙ H(Σ2 ) for ?M = Σ? ∪Σ2 was unitary, i.e. T (M )? = T (M )?1 , orientation reversal of M would 1 just invert the transition map, T (M ? ) = T (M )?1 . It is tempting to think that orientation reversal was in this way related with time reversal. Unitarity in quantum theory, however, expresses the conservation of probability with respect to a global time parameter and therefore cannot be expected to hold in general relativity without additional assumptions. Consider, for example, the path integral (3.1) for the d-manifold M of Figure 2 where the boundary data A|Σ1 and A|Σ2 impose space-like geometries on Σ1 and Σ2 , respectively. The path integral will generically contain histories in which Σ2 is in the causal future of Σ1 and also some in which Σ2 is in the past of Σ1 , unless we impose additional conditions. Following the ideas of Oeckl [12], one can consider more general manifolds of the form M = S × I where S is a compact (d ? 1)-manifold with boundary ?S = ?. The boundary ?M then consists of ?M = (S × {0}) ∪ (S × {1}) ∪ (?S × I). The idea is now to impose space-like geometries on S × {0} (initial preparation of the experiment) and S × {1} (?nal measurement of the outcome) and in addition a time-like geometry on ?S × I which represents the clock in the classical laboratory that surrounds the quantum experiment and which ensures that S × {1} is in the future of S × {0}. Notice that this is the situation in which textbook quantum theory makes sense without conceptual extensions and in which it has been con?rmed experimentally: the quantum experiment is limited both in size and duration, and all measurements are performed by classical observers who use classical clocks.

Corners and higher level. If we just plugged this choice of M = S × I into the axioms of Section 3.2, we would get a single Hilbert space associated with the boundary ?M . This is not quite what we want. We would rather prefer to obtain Hilbert spaces for the boundary components S × {0} and S × {1} between which the transition map acts, but not for ?S × I. The framework of higher level TQFT or TQFT with corners, see for example [11], might be appropriate to treat this situation. We do not go into details here, but rather focus on the structure of manifolds in the subsequent sections.

4

Classi?cation of manifolds

Given some path integral (3.1) of general relativity in d-dimensional space-time, let us consider the partition function Z(M ) of (3.4), assuming for a moment that it is well de?ned and can be computed for some class of space-time manifolds. Whenever Z(M ) = Z(M ′ ), then M and M ′ are not di?eomorphic. The partition function therefore forms a tool for the classi?cation of smooth manifolds up to di?eomorphism. Although the framework of general relativity is by de?nition that of smooth manifolds and smooth maps (the Einstein equations are di?erential equations after all), it is instructive to contrast it with other types of manifolds. Barrett [9] has already remarked that the space-time dimension d = 3 + 1 plays a special role. Let us explain in more detail why. The presentation in this section is a rather informal overview. The detailed de?nitions and theorems to which we refer, can be found in the Appendix.

16

Quantum general relativity and the classification of smooth manifolds

Topological manifolds up to homeomorphism. Each smooth manifold has got an underlying topological manifold (Section 2.1), and any two di?eomorphic smooth manifolds have homeomorphic underlying topological manifolds. Can we use the information about topological manifolds up to homeomorphism in order to classify smooth manifolds up to di?eomorphism? The answer is ‘yes’ in space-time dimension d ≤ 2 + 1, but ‘no’ in general. Indeed, in d ≤ 2 + 1, each topological manifold admits a di?erentiable structure, and any two homeomorphic topological manifolds have di?erentiable structures so that the resulting smooth manifolds are di?eomorphic. This result is one of the explanations of why quantum general relativity in d = 2 + 1 is particularly simple. In fact, its path integral quantization is closely related to a C 0 -QFT, using topological manifolds up to homeomorphism, and exploiting the fact that in d = 2 + 1, there is no di?erence between C 0 - and C ∞ -QFTs. Examples of such theories are given in Section 5 below, but before we can state them, we need some more theoretical background. In d ≥ 3 + 1, no analogous result is available. There exist countably in?nite families of (compact) smooth 4-manifolds [13, 14] which are pairwise non-di?eomorphic, but which have homeomorphic underlying topological manifolds. There is therefore a considerable discrepancy between C ∞ - and C 0 -QFTs in d = 3 + 1 space-time dimensions. The most striking result even concerns the standard space 4 [15, 16].

?

Theorem 4.1. Consider the topological manifold

?d, d ∈ ?. ?

? If d = 4, then there exists a di?erentiable structure for d which is unique up to di?eomorphism. ? If d = 4, then there exists an uncountable family of pairwise non-di?eomorphic di?erentiable structure for d .

?

The di?erentiable structure of 4 induced from × × × is called standard and the others exotic. Non-uniqueness of di?erentiable structures persists in higher dimensions, for example, there are 28 inequivalent di?erentiable structures on the topological sphere S 7 , or 992 inequivalent di?erentiable structures on S 11 [17], but in dimension d ≥ 4 + 1 (d ≥ 5 + 1 if the manifold has a non-empty boundary), there never exists more than a ?nite number of non-di?eomorphic di?erentiable structures on the same underlying topological manifold. The space-time dimension d = 3 + 1 is distinguished by the feature that there can exist an in?nite number of homeomorphic, but pairwise non-di?eomorphic compact smooth manifolds. Topological manifolds up to homotopy equivalence. There is another way of classifying topological manifolds. This relation is known as homotopy equivalence. Two topological manifolds M , N are called homotopy equivalent if there exists a pair of continuous maps f : M → N and g : N → M such that f ? g is homotopic to the identity map idM of M , i.e. it ‘can be continuously deformed’ to idM , and g ? f is homotopic to idN (see Appendix A.7 for details). The concept of homotopy equivalence of topological manifolds is weaker than that of homeomorphism: any two homeomorphic topological manifolds are also homotopy equivalent. The converse implication is true, for example, in d ≤ 1+1, i.e. any two compact topological manifolds that are homotopy equivalent, are also homeomorphic. But it does not hold in

?

? ? ? ?

Quantum general relativity and the classification of smooth manifolds

17

k=0

k=1

k=2

k=3

Figure 6: k-simplices in

?3.

d = 2 + 1: there exist compact topological 3-manifolds which are homotopy equivalent, but not homeomorphic. In d ≤ 1 + 1, quantum general relativity is therefore even simpler than in d = 2 + 1. It is not only given by a C 0 -QFT (as opposed to a generic C ∞ -QFT), but even by what one could call an hQFT (TQFT up to homotopy equivalence5 ). Quantum general relativity in d = 2 + 1, in contrast, is not an hQFT. A topological invariant closely related to its partition function has been con?rmed to distinguish homotopy equivalent topological 3-manifolds that are not homeomorphic. Quantum general relativity in d = 2 + 1 is thus as generic as topology allows. Piecewise-linear manifolds up to PL-isomorphism. Before we can cite the next relevant classi?cation result, we have to introduce yet another type of manifolds: piecewise-linear (PL-) manifolds (see Appendix A.3 for details). A k-simplex in standard space d is the smallest convex set that contains k + 1 points that span a k-dimensional hyperplane (Figure 6). A polyhedron is a locally ?nite union of simplices. d itself, for example, is a polyhedron. A piecewise-linear map is a map f : P → Q between polyhedra which maps simplices onto simplices. We can now obtain another type of manifold by restricting the transition functions of a topological manifold to piecewise-linear functions. A piecewise-linear (PL-) manifold is a topological manifold such that all transition functions are piecewise-linear. This is illustrated in Figure 7. Notice that not M itself is triangulated, but rather the coordinate systems are. PL-manifolds are classi?ed up to PLisomorphism (Appendix A.3), i.e. up to homeomorphisms that are piecewise-linear in coordinates.

?

?

Triangulations of smooth manifolds. If some topological manifold admits both a piecewise-linear and a smooth structure, satisfying a compatibility condition (see Appendix A.4 for details), we say that the di?erentiable structure is a smoothing of the piecewise-linear structure. There is a close relationship between smooth and piecewise-linear manifolds given by Whitehead’s theorem: For each smooth manifold M , there exists a PL-manifold MPL , called its Whitehead triangulation, so that M is di?eomorphic to a smoothing of MPL . MPL is unique up to PL-isomorphism.
5 The name homotopy quantum ?eld theory and the abbreviation HQFT are already gone for a di?erent concept.

18

Quantum general relativity and the classification of smooth manifolds
?2 ? ??1 1

?d
?1

?d
?2

U12 U1 U2 M

Figure 7: A piecewise-linear (PL-) manifold.

Whitehead’s theorem is therefore a license to triangulate space-time. But does the Whitehead triangulation capture all features of the given smooth manifold? The answer is ‘yes’, at least in space-time dimension d ≤ 5 + 1: each PL-manifold admits a smoothing, and the resulting smooth manifold is unique up to di?eomorphism. The equivalence classes of smooth manifolds up to di?eomorphism are therefore in one-to-one correspondence with those of PL-manifolds up to PL-isomorphism. We are free to choose either framework at any time. We therefore know that the path integral of general relativity in d ≤ 5 + 1 is closely related to a TQFT that is de?ned for PL-manifolds up to PL-isomorphism, i.e. to a PLQFT. Whitehead triangulations provide us with a way of ‘discretizing’ space-time which is not merely some approximation nor introduces a physical cut-o?, but which is rather exact up to di?eomorphism. Combinatorial manifolds Whitehead triangulations are widely used in topology because they facilitate e?cient computations which are most conveniently performed in a purely combinatorial language. It is known that each d-dimensional PL-manifold M is PL-isomorphic to a single polyhedron P in n for some n (Figure 8). Such a polyhedron which itself forms a PL-manifold, is called a combinatorial manifold (see Appendix A.5 for details) or a global triangulation of M . If M is compact, P can be described in terms of a ?nite number of simplices. In order to characterize M up to PL-isomorphism, it su?ces to characterize the PL-manifold P up to PL-isomorphism. A very convenient way of stating the condition of PL-isomorphism, and for our purposes the best intuition, is provided by Pachner’s theorem. We give here the version for closed manifolds: any two closed combinatorial manifolds are PL-isomorphic if and only if they are related by a ?nite sequence of Pachner moves. Pachner moves are local modi?cations of the triangulation by joining simplices or by

?

Quantum general relativity and the classification of smooth manifolds

19

?d

?d

?d

?d

?n

P M

Figure 8: Each piecewise-linear (PL-) manifold M is PL-isomorphic to some combinatorial manifold P .

splitting some polyhedra up into smaller pieces. Figure 9 shows the Pachner moves for closed manifolds in d = 1 + 1 and d = 2 + 1. A systematic way of listing the moves in any dimension is explained in Appendix A.6. Pictures for d = 3 + 1 can be found in [18, 19], and the moves for manifolds with boundary in [20]. The following procedure is now available in order to decide whether two given smooth d-manifolds, d ≤ 5 + 1, are di?eomorphic. Start with two closed smooth manifolds M (1) (2) (1) and M (2) . Construct their Whitehead triangulations MPL and MPL . Find combinatorial (1) (2) manifolds P (1) and P (2) which are PL-isomorphic to MPL and MPL , respectively. M (1) and (2) are di?eomorphic if and only if P (1) and P (2) are related by a ?nite sequence of Pachner M moves. Scenario for quantum gravity. We have reached a ?rst goal: the di?eomorphism gauge symmetry of general relativity on a closed space-time manifold has been translated into a purely combinatorial problem involving triangulations that consist of only a ?nite number of simplices, and their manipulation by ?nite sequences of Pachner moves. If not only the partition function, but also the full path integral of general relativity in d ≤ 5 + 1 is given by a PL-QFT, we know that all observables are invariant under Pachner moves. The partition function of quantum general relativity is an invariant of PL-manifolds, too, and can be computed by purely combinatorial methods for any given combinatorial manifold. A generic expression of such a partition function is the state sum, Z= (amplitudes),
{ colourings } { simplices }

(4.1)

where the sum is over all labellings of the simplices with elements of some set of colours, and the integrand is a number that can be computed for each such labelling. In Section 5 below,

20

Quantum general relativity and the classification of smooth manifolds

(a)

(b)

(c)

(d)

d=2

d=3

Figure 9: Pachner moves in d = 2: (a) The 1 ? 3 move subdivides a triangle into three. (b) the 2 ? 2 move joins two triangles and then splits the result in a di?erent way. Pachner moves in d = 3: (c) the 1 ? 4 move subdivides one tetrahedron into four. (d) the 2 ? 3 move changes the subdivision of a diamond from two tetrahedra to three ones (glued along the dotted line in the bottom picture).

we give examples and illustrate that the partition function of quantum general relativity in d = 2 + 1 is precisely of this form. If quantum general relativity in d = 3+1 is indeed a PL-QFT, the following two statements which sound philosophically completely contrary, ? Nature is fundamentally smooth. ? Nature is fundamentally discrete. are just two di?erent points of view on the same underlying mathematical structure: equivalence classes of smooth manifolds up to di?eomorphism. Further classi?cation results which are compiled in the Appendix, indicate that the partition function in d ≥ 4 + 1 would be much less interesting than in d = 3 + 1 because smooth manifolds up to di?eomorphism are already essentially classi?ed by their underlying topological manifold up to homeomorphism, subject to only ?nite ambiguities in choosing a di?erentiable structure. In fact, one could even use the homotopy type of space-time plus some additional information about the structure of the tangent bundle. Quantum general relativity in d ≥ 4+1 would therefore be closely related to an hQFT supplemented by additional data in order to specify the tangent bundle and to resolve the ambiguities. On the mathematical side, the path integral quantization of general relativity is closely related to the problem of classifying smooth manifolds up to di?eomorphism by classifying their Whitehead triangulations up to PL-isomorphism. Precisely in d = 3 + 1, topology is rich enough to (potentially) provide in?nitely many non-trivial partition functions that are able to distinguish non-PL-isomorphic PL-structures on the same underlying topological manifold. It is an open problem in topology to construct these invariants. Unless d = 2+ 1 or d = 3+ 1, the problem of constructing interesting partition functions is

Quantum general relativity and the classification of smooth manifolds d≤2 C∞ PL Top htpy d=3 C∞ PL Top htpy d=4 C∞ PL Top htpy d≥5 C∞ ..... PL ..... Top ..... htpy

21

Table 1: The relationship between the classi?cations of various types of manifolds of dimension d: smooth manifolds up to di?eomorphism (C ∞ ), piecewise-linear manifolds up to PL-isomorphism (PL), topological manifolds up to homeomorphism (Top), and topological manifolds up to homotopy equivalence (htpy). Equivalence of manifolds of the type shown in one row implies equivalence of manifolds of the type shown in the rows below. If the rows are not separated by any line, the converse implication holds as well. If the rows are separated by a dotted line, the converse implication holds up to some obstruction and ambiguity. A solid line indicates that the converse implication is seriously violated. For details, we refer to the Appendix, in particular to Appendix A.8. This table is rather sketchy, and a number of subtleties have been suppressed, so we ask the reader not to consider this table as a theorem without actually having read the small-print in the Appendix. ?nally dominated by the study of topological manifolds up homotopy equivalence which would render general relativity ‘suspiciously simple’. All these considerations apply to the partition function (3.4), but not necessarily to the matrix elements (3.2), i.e. before tracing out the boundary conditions. In the next section, we illustrate how the examples in d = 2 + 1 show that the expression for the state sum (4.1) already suggest the appropriate Hilbert spaces and boundary conditions. Finding the partition function is therefore a key step. Notice that we are not claiming that the universe is a smooth 4-manifold M that has an ‘exotic’, i.e. non-standard, di?erentiable structure. This may or may not be true, and the answer to this question is independent of the considerations presented so far. Some consequences of exotic di?erentiable structures on space-time have been explored in [21, 22]. The crucial observation is rather that, just because the Einstein equations are di?erential equations, the path integral of general relativity ought to be sensitive to the di?erentiable structure of M , even if it is merely the standard di?erentiable structure. It may eventually turn out that in many cases the smooth structure on the boundary ?M already determines the smooth structure of the entire M , i.e. that the smooth background can be viewed as part of the classical boundary data. Table 1 ?nally summarizes the classi?cation of smooth, PL- and topological manifolds in the various dimensions. Refer to the Appendix for details. Renormalization group transformations. The 1 ? d + 1 Pachner move (Figure 9) always subdivides one d-simplex into d + 1 d-simplices. It obviously resembles the block spin transformations familiar from the Statistical Mechanics treatment of renormalization, but is here applied to the Whitehead triangulations of smooth manifolds as opposed to discretizations of Riemannian manifolds. In our case, the simplices do not have any metric size as there is no background geometry associated with the smooth space-time manifold M .

22

Quantum general relativity and the classification of smooth manifolds

In non-generally relativistic ?eld theories on some given Riemannian manifold (M, g), renormalization is the comparison of the dynamical scale of the theory, i.e. the relevant correlation lengths, for any given cut-o? and bare parameters, with the scale of the background metric g, in order to determine the relation between cut-o? and bare parameters for which the physical predictions are constant. In general relativity, there is no background metric and therefore no way of (and no need to) introduce a cut-o?. The di?eomorphism gauge symmetry implies the invariance of all observables under Pachner moves so that one can say that the theory is readily renormalized or, depending on the personal taste, that there is no need to renormalize theories in which the geometry is dynamical. We stress that Whitehead triangulations neither introduce a cut-o? nor break any of the symmetries. Refer to [23] for the implications on the notion of locality, on the appearance of the Planck scale and on the compatibility with ideas in the context of the holographic principle.

5

Examples

This section is a brief overview over some results on quantum general relativity in d = 2 + 1 space-time dimensions. For more details, see the review articles [24–26]. Turaev–Viro invariant. We have stressed above the importance of the partition function and that it forms an invariant of smooth manifolds up to di?eomorphism. In d = 2 + 1, we know that we can equivalently study an invariant of PL-manifolds up to PL-isomorphism which is given by a state sum (4.1) for a closed combinatorial manifold. The Turaev–Viro invariant [27] is such a state sum, Z=
1 j : ?1 →{0, 2 ,1,..., k?2 } σ1 ∈?1 2

dimq j(σ1 )
σ3 ∈?3

{6j}q (σ3 ) .

(5.1)

Here k = 1, 3 , 2, . . . is a ?xed half-integer, ?1 and ?3 denote the sets of 1-simplices and 2 3-simplices, respectively. The sum is over all ways of colouring the 1-simplices σ1 ∈ ?1 with 1 half-integers j(σ1 ) ∈ {0, 2 , 1, . . . , k?2 }. The quantum dimension dimq j(σ1 ) can be computed 2 for each j(σ1 ), and the quantum-6j-symbol {6j}q (σ3 ) depends on the labels j(σ1 ) associated with all 1-simplices σ1 in the boundary of each 3-simplex σ3 , see [27] for details. The halfintegers j(σ1 ) in fact characterize the ?nite-dimensional irreducible representations of the quantum group Uq (sl2 ) for the root of unity q = eiπ/k . Ponzano–Regge model. The ?rst connection with quantum gravity in d = 2 + 1 can be seen in the limit k → ∞ in which q → 1 and the quantum group Uq (sl2 ) is replaced by the envelope U (sl2 ) which is dual to the algebra of functions on the local Lorentz group SU (2) = Spin(3) (up to complexi?cation). The j(σ1 ) then characterize the ?nite-dimensional irreducible representations of SU (2). In this limit, Z agrees with the partition function of the Ponzano–Regge model [28], a non-perturbative quantization of the toy model of general relativity in d = 2 + 1 with Riemannian signature η = diag(1, 1, 1). In the limit, the partition function diverges and is no longer a mathematically well-de?ned invariant, but the Pachner move invariance of (5.1) persists formally if one accepts to divide out in?nite factors. For the model with the realistic Lorentzian signature η = (?1, 1, 1), Spin(3) is replaced by Spin(1, 2) so that the representation labels become continuous [29, 30].

Quantum general relativity and the classification of smooth manifolds

23

Cosmological constant. In the Ponzano–Regge model, the large spin limit of the 6jsymbols yields the connection with the classical action of general relativity [28] and shows 1 that the labels j(σ1 ) represent dynamically assigned lengths6 G(j(σ1 )+ 2 ) for the 1-simplices σ1 . The analogous argument for the Turaev–Viro invariant indicates [31,32] that the Turaev– Viro invariant is the partition function of general relativity with Riemannian signature and quantized positive cosmological constant Λ = 4π 2 /( Gk)2 . The limit k → ∞ then sends Λ → 0 as expected. TQFT. Although we have so far concentrated on the partition function, one can easily read o? from the state sum (5.1) a consistent choice of boundary ?elds and Hilbert spaces [27]: ?x the j(σ1 ) for all 1-simplices in the boundary Σ = ?M in order to characterize a state. The Hilbert space H(Σ) then has a basis whose vectors are labelled by all these j(σ1 ). In the limit k → ∞, the Hilbert spaces become in?nite-dimensional so that the rule (S1) in Section 3.2 has to be relaxed for the Ponzano–Regge model. The Hilbert spaces admit a precise interpretation as spaces of states of 2-geometries. Although our focus on the partition function instead of the full path integral with boundary conditions (3.1), seemed to be somewhat narrow at ?rst sight, the examples show that state sums such as the Turaev–Viro invariant (5.1) already carry the information about boundary ?elds and Hilbert spaces. In fact, once the set of colourings of the state sum has been speci?ed, it automatically determines the vector spaces and boundary ?elds. Of course, there can be several di?erent formulae of state sums (4.1) with di?erent sets of colours which yield the same invariant. This corresponds to di?erent path integrals (3.1) with di?erent Hilbert spaces that have the same trace (3.4). What we are looking for in the case of general relativity, is the full TQFT and not just the partition function. Since all the classi?cation results are available for manifolds with boundary, the key step is to ?nd the physical interpretation for the boundary ?elds, expressed on triangulations of the boundary, for the set of colours of the state sum (4.1). Towards 3+1. We can now come back to the claim of the introduction that the absence of gravitons in d = 2 + 1 is related 1. neither to the question of whether the path integral corresponds to a C 0 -QFT (as opposed to a C ∞ -QFT), 2. nor to the question of whether the vector spaces of this C 0 -QFT or C ∞ -QFT are ?nitedimensional, 3. nor to the question of whether the theory admits a triangulation independent discretization. Whereas the question of C 0 -QFT versus C ∞ -QFT is related to the classi?cation of topological versus smooth manifolds and therefore depends on the space-time dimension d, the question of ?nite-dimensionality is presently understood only in d = 2 + 1. We have seen examples for both alternatives, ?nite-dimensional (Turaev–Viro model) and in?nite-dimensional (Ponzano– Regge model). Both are constructed from classical theories that have only constant curvature geometries as their solutions and therefore no propagating modes. The answer to question (3.) above is ?nally ‘yes’ in any d ≤ 5 + 1 from the classi?cation results. Concrete examples
6

In d = 2 + 1, G is the Planck length.

24

Quantum general relativity and the classification of smooth manifolds

have so far been constructed in d = 2 + 1 for both pure general relativity (Turaev–Viro model and Ponzano–Regge model) and for a special case of general relativity with fermions [33], and in d = 3 + 1 only for BF-theory without [34] or with [35, 36] cosmological constant. For general relativity in d = 3 + 1, the results of Section 4 still suggest that we should expect a Pachner move invariant state sum although the ?nite-dimensionality of the Hilbert spaces might be lost in a more drastic fashion than in d = 2 + 1, and there are the remarks of Section 3.4 on the role of time. A possible approach to d = 3 + 1 in order to narrow down the path integral is suggested by the special properties of smooth 4-manifolds as we sketch in the ?nal section.

6

Physics meets Mathematics

Connecting the gauge symmetry of classical general relativity with results on the classi?cation of smooth manifolds up to di?eomorphism, we have revealed the coincidence of an open problem in topology, namely to construct a non-trivial invariant of piecewise-linear 4-manifolds, with an open problem in theoretical physics, namely to ?nd a path integral quantization of general relativity in d = 3 + 1 space-time dimensions.

6.1

Mathematical aspects

The mathematical question is whether one can construct invariants of piecewise-linear 4manifolds that are non-trivial in the sense that they can distinguish inequivalent di?erentiable structures on the same underlying topological manifold. From the classi?cation results, it is known that the Whitehead triangulation of any given smooth manifold captures the full information about its di?erentiable structure up to diffeomorphism (Appendix A.4). This suggests that a state sum which probes the abstract combinatorial information contained in a triangulation, will be su?cient (Appendix A.6). On the other hand, there exist already way too many topological manifolds in order to extract the complete classi?cation information by any conceivable algorithm (Appendix A.8.1). The interesting question is therefore whether one can ?nd a suitably restricted class of piecewiselinear 4-manifolds, for example closed connected and simply connected manifolds, for which a non-trivial invariant can be constructed. The existence of the Donaldson [4] and Seiberg– Witten [5] invariants in the smooth framework is encouraging because it demonstrates that some non-trivial information can indeed be extracted. Question 6.1. Do there exist state sum invariants of piecewise-linear 4-manifolds which are able to distinguish inequivalent PL-structures on the same underlying topological manifold? Di?erential topologists have been considering this question for a long time, see, for example the introduction of [37]. Some state sum invariants of piecewise-linear 4-manifolds have already been constructed, for example the Crane–Yetter invariant [36] and Mackaay’s state sum [19]. So far, it has not been con?rmed that any of these constructions is indeed non-trivial in the above sense. The Crane–Yetter invariant is known to depend only on the homotopy type of the underlying topological manifold [38] although it o?ers a novel combinatorial method for computing the signature (of the intersection form) of M . Mackaay’s state sum has so far been explored only for very special cases in which it, too, depends only on the homotopy type [39]. Nevertheless, in order to appreciate this state sum, a more detailed comparison with the 3-dimensional case is very instructive.

Quantum general relativity and the classification of smooth manifolds

25

Just as the Turaev–Viro invariant [27] of 3-manifolds can be constructed for a certain class of spherical categories [40], Mackaay’s state sum [19] is de?ned for a class of spherical 2-categories. The situations for which Mackaay’s state sum has been carefully studied [39] involve rather special spherical 2-categories for which this state sum resembles a 4-dimensional generalization of the Dijkgraaf–Witten model [41]. It is, however, known that the Dijkgraaf– Witten invariant agrees for lens spaces as soon as they have the same homotopy type [42]. Lens spaces are the standard examples [43] of topological 3-manifolds in order to show that some invariant can distinguish manifolds of the same homotopy type that are not homeomorphic. In order to render the Turaev–Viro invariant non-trivial, the construction of the (modular categories of representations of the) quantum groups Uq (sl2 ), q = e2πi/? , ? ∈ , seems to be essential. In this case, the Turaev–Viro invariant indeed distinguishes non-homeomorphic lens spaces of the same homotopy type7 . The same cannot be accomplished with the category of representations of an ordinary group. This comparison with the 3-dimensional case therefore suggests that Mackaay’s state sum should be studied for su?ciently sophisticated spherical 2-categories which are as ‘generic’ in the context of 2-categories as are the modular categories of representations of Uq (sl2 ) in the context of 1-categories.

?

6.2

Physical aspects

Questions about observables of general relativity are questions about smooth manifolds and smooth functions. The physical answers are speci?ed only up to space-time di?eomorphism. Whenever we have made use of classi?cation results on smooth, piecewise-linear or topological manifolds, we have exploited this gauge freedom in an essential way. The classi?cation results show that unless d = 3 + 1, at least the partition function Z(M ) (3.4) of the path integral can be computed without explicitly referring to the di?erentiable structure of the space-time manifold M . In d = 2 + 1, Z(M ) depends only on the underlying topological manifold M up to homeomorphism (Appendix A.8.3), and this is the mechanism that renders quantum general relativity in d = 2 + 1 space-time dimensions particularly simple. This indeed justi?es the jargon ‘topological’. In d ≤ 1 + 1 or d ≥ 4 + 1, the information required in order to determine Z(M ) is essentially the homotopy type of M (assuming that M admits a di?erentiable structure) together with information on the tangent bundle and one ?nite number which resolves the ambiguities in constructing ?rst a PL-structure for the topological manifold M (Appendix A.8.3) and then a smoothing of this PL-structure (Appendix A.8.2). Although we have de?ned the partition function Z(M ) in (3.4) using the path integral of general relativity in a way that seems to employ the di?erentiable structure of space-time, the above results suggest that unless d = 3 + 1, there exists an alternative way of computing Z(M ), either from the underlying topological manifold up to homeomorphism (d = 2 + 1), or even essentially from its homotopy type together with additional choices (d ≤ 1 + 1 or d ≥ 4 + 1). It happens only in d = 3 + 1, that there can be more than ?nitely many pairwise inequivalent di?erentiable structures for the topological manifold underlying M . This means that there is generically no short cut available in order to compute Z(M ) from the underlying
By the theorem of Turaev and Walker [38], the Turaev–Viro invariant is the squared modulus of the Reshetikhin–Turaev invariant [44] which is known [45] to distinguish, for example, the lens spaces L(7, 1) and L(7, 2).
7

26

Quantum general relativity and the classification of smooth manifolds

topological manifold alone. Only in d = 3 + 1, di?erential topology is rich enough in order to provide us with a large number of non-trivial Z(M ) in the context of smooth manifolds. This is very reassuring because in d = 3 + 1, one might expect quite a number of quantum ?eld theories with a gauge symmetry under space-time di?eomorphisms even though the theories other than general relativity are usually treated in a di?erent framework. Besides the partition function, will the entire C ∞ -QFTs in space-time dimensions other than d = 3 + 1 be necessarily less interesting? This question cannot be ultimately answered yet, but one can expect that the dominance of homotopy type in this problem in d ≤ 1 + 1 or d ≥ 4 + 1 (d ≥ 5 + 1 if there is a non-empty boundary) would provide strong constraints. Concerning d = 3 + 1, we arrive at, Question 6.2. Does there exist a particular state sum invariant of piecewise-linear 4-manifolds, a degenerate limit of which yields the path integral quantization of pure general relativity in d = 3 + 1 space-time dimensions (in the special case in which no boundary conditions are imposed)? Let us again compare the situation with the toy model of general relativity in d = 2 + 1. The state sum that yields the actual topological invariant, is the Turaev–Viro invariant [27]. 3 This state sum for Uq (sl2 ), q = eiπ/k , k = 1, 2 , 2, . . ., corresponds to the partition function of quantum gravity with Riemannian signature η = diag(1, 1, 1) and quantized positive cosmological constant Λ = 4π 2 /( Gk)2 [31, 32]. This model de?nes [27] a proper TQFT in the strict sense [2], based on topological manifolds up to homeomorphism and involving ?nitely generated modules. Quantum gravity with Riemannian signature, but Λ = 0, can then be understood as a limit of this invariant for k → ∞, Λ → 0, q → 1, in which the Hilbert spaces become in?nite-dimensional and the partition function diverges. The partition function is therefore no longer a well-de?ned invariant of topological manifolds. Nevertheless, this degenerate limit precisely agrees with the original model of Ponzano–Regge [28] which had been invented as a non-perturbative quantization of general relativity in d = 2+1 with Riemannian signature, well before it was realized that this framework is closely related to invariants of topological manifolds. Quantum gravity in d = 2 + 1 with the realistic Lorentzian signature η = diag(?1, 1, 1) is ?nally even more complicated, replacing Spin(3) by Spin(2, 1) and the discrete representation labels of the Ponzano–Regge model with continuous ones [29, 30]. If we try to extrapolate this experience from d = 2 + 1 to d = 3 + 1, on the mathematical side we may well expect that the proper invariant of piecewise-linear 4-manifolds involves some highly non-trivial 2-category which may be very di?cult to guess. Its physical counterpart, the sought-after path integral of general relativity, however, needs to be ‘just’ a degenerate limit of the proper invariant. Given the experience from d = 2 + 1 and noting that the Ponzano–Regge model (although divergent and therefore not well de?ned in the mathematical sense) is so much easier than the Turaev–Viro invariant, it seems to be a reasonable strategy to approach the state sum invariant of piecewise-linear 4-manifolds via theoretical physics. This means to proceed in two steps: ?rst to construct a physically motivated, but degenerate limit of the invariant which corresponds to general relativity; second to study the deformation theory of the relevant categories and to aim for the actual invariant. The interesting perspective is here the combination of techniques developed in mathematics on how to lift combinatorial and algebraic structures from three to four dimensions, see, for example the introductory sections of [19, 39] and also [11, 18] for references, with methods

Quantum general relativity and the classification of smooth manifolds

27

from theoretical physics on how to construct discrete physical models related to the path integral of general relativity, see, for example, the review articles [24–26]. The framework outlined here, ? Does not involve any new physical assumptions. It just combines quantum theory (represented by the axioms (A1), (A2), (A4) of axiomatic X-QFT for X = C 0 , C ∞ , PL, h) with the properties of generally relativistic theories (represented by axiom (A3) and the choice X = C ∞ ). ? Singles out the space-time dimension d = 3 + 1. ? Explains that, if mathematicians solve Question 6.1, this will provide physicists with (a family of) rigorously de?ned path integrals which have the same symmetries as general relativity. ? Explains why the di?eomorphism gauge symmetry takes care of renormalization. ? Contains the spin foam models of (2 + 1)-dimensional quantum gravity as well-studied examples. ? Applies to other coordinate-free formulated ?eld theories as well (c.f. Section 2.5), not necessarily generally relativistic. In the search for a path integral quantization of general relativity in d = 3 + 1, we are facing a coincidence of open questions in mathematics with open questions in theoretical physics. In exploring these connections, we have just scratched the surface.

Acknowledgments
The author is indebted to Marco Mackaay for explaining various results on the topology of four-manifolds. I would like to thank Louis Crane, Gary Gibbons, Robert Helling and Daniele Oriti for discussions.

A

Topological, piecewise-linear and smooth manifolds

For mathematical background on topological, smooth and piecewise-linear manifolds, we refer to various textbooks [6, 7, 46–48] as well as to the introductory sections of [16, 37]. We review here only those facts that are relevant to the classi?cation of the various types of manifolds, hoping that this compilation of de?nitions and results will make the literature more accessible.

A.1

Topological manifolds

We ?rst recall the de?nitions of topological and smooth manifolds. De?nition A.1. Let M be a topological space, and ?x some d ∈

?, called the dimension. ?

1. A chart or coordinate system for M is a pair (U, ?) of an open set U ? M and a homeomorphism ? : U → ?(U ) onto an open subset ?(U ) ? d of the half-space + d := {x ∈ d : x ≥ 0}. 0 +

?

?

2. A TOPd -atlas A for M is a family A = {(Ui , ?i ) : i ∈ I} of coordinate systems for M such that the Ui form an open cover of M , Ui = M.
i∈I

(A.1)

28

Quantum general relativity and the classification of smooth manifolds Here I denotes some index set. 3. On the non-empty overlaps Uij := Ui ∩ Uj = ?, i, j ∈ I, there are homeomorphisms, ?ji := ?j ? ??1 |?i (Uij ) : ?i (Uij ) → ?j (Uij ), i which are called transition functions. 4. The boundary ?M of M is the set of all p ∈ M for which ?i (p) ∈ d : x = 0}. 0 (A.2)

?

?d , ?d := {x ∈ 0 0

5. Two atlases are called equivalent if their union is an atlas. These are the most general atlases we are interested in. They give rise to topological manifolds. De?nition A.2. Fix some dimension d ∈

?.

1. Let M be a topological space. A TOPd -structure [A] for M is an equivalence class of TOPd -atlases. 2. A topological d-manifold (M, [A]) is a paracompact Hausdor? space M equipped with a TOPd -structure [A]. 3. Two topological manifolds (M, [A]), (N, [B]) are called equivalent if M and N are homeomorphic. 4. A topological manifold (M, [A]) is called compact if the underlying topological space M is compact. It is called closed if it is compact and ?M = ?. We often write just M rather than (M, [A]). The boundary ?M of any topological dmanifold M forms a topological (d ? 1)-manifold with empty boundary. A 0-manifold is just a set of points with the discrete topology. Due to paracompactness, each atlas of any topological manifold admits a locally ?nite re?nement for which the ?i (Ui ) are contained in compact subsets of d . + The remainder of this section is a technical detail which is necessary in order to combine results from various di?erent sources in the literature. The conditions that the underlying topological space in De?nition A.2(2.) be paracompact and Hausdor?, are the same as those used in [7] which are those that correspond to the physically relevant space-times [49]. The following alternative choices are common in the literature,

?

1. metrizable [50, 51], 2. second countable and Hausdor? [46], 3. separable and Hausdor? [16]. Concerning (1.), by a theorem of Stone [52], each metrizable topological space is paracompact and Hausdor?. Conversely, any paracompact Hausdor? space with a TOPd -atlas inherits a metric from d on each chart and, by employing a continuous partition of unity, can be shown to be metrizable. Concerning (2.), each topological space with a TOPd -atlas is locally compact, and any locally compact and second countable Hausdor? space is metrizable [52]. Conversely, by a theorem of Alexandro? [52], each locally compact metrizable space admits a countable basis of

?

Quantum general relativity and the classification of smooth manifolds

29

open sets for each connection component. As soon as the discussion is restricted to topological spaces with a countable number of connection components, (2.) is therefore equivalent to (1.). Concerning (3.), each second countable topological space is separable, but (3.) is in general weaker than (2.). We note, however, that a given topological space is paracompact and locally compact if and only if it is a free union of spaces that are σ-compact (unions of countably many compact sets) [53]. The relevant examples of [16] are all of this type.

A.2

Smooth manifolds

We can impose additional structure on manifolds by restricting the transition functions to appropriate sub-families of homeomorphisms. Recall that a map f : U → n , for some open set U ? m , is called C k if all k-th partial derivatives exist and are continuous on U . A C k -di?eomorphism is an invertible C k -map with a C k -inverse.

?

?

De?nition A.3. Let M be a topological d-manifold, d ∈

?, and k ∈ ?0 ∪ {∞}.

1. A C k -atlas for M is a TOPd -atlas A = {(Ui , ?i ) : i ∈ I} such that for all i, j ∈ I with Ui ∩ Uj = ?, the transition functions ?ji are C k -di?eomorphisms. 2. A C k -structure [A] for M is an equivalence class of C k -atlases. This de?nition includes the topological case (De?nition A.1) for k = 0. De?nition A.4. Let k ∈

?0 ∪ {∞}. ?, is a topological d-manifold M equipped

1. A d-dimensional C k -manifold (M, [A]), d ∈ with a C k -structure [A].

2. Let (M, [A]), (N, [B]) be C k -manifolds with atlases A = {(Ui , ?i ) : i ∈ I} and B = {(Vj , ψj ) : j ∈ J}, not necessarily of the same dimension. A map f : M → N is called C k if f is continuous and if for all p ∈ M such that p ∈ Ui , f (p) ∈ Vj , for some i ∈ I, j ∈ J, the map, ψj ? f ? ??1 : ?i (Ui ) → ψj (Vj ) (A.3) i is a C k -map. 3. A C k -di?eomorphism f : M → N is an invertible C k -map whose inverse is C k . 4. Two C k -manifolds are called equivalent if they are C k -di?eomorphic. The boundary ?M of any d-dimensional C k -manifold is a (d? 1)-dimensional C k -manifold without boundary. De?nition A.5. 1. A C k -atlas or C k -structure, k ≥ 1, is called oriented if all transition functions ?ji are orientation preserving, i.e. if they have positive Jacobi determinants. 2. An oriented d-dimensional C k -manifold is a topological d-manifold with an oriented C k -structure. 3. A C k -di?eomorphism is called orientation preserving if all its coordinate representations (A.3) are orientation preserving.

30

Quantum general relativity and the classification of smooth manifolds

The boundary ?M of any oriented C k -manifold is oriented. We have de?ned oriented manifolds only for the cases C k , k ≥ 1 (although this can also be done for topological manifolds). If we nevertheless mention oriented topological manifolds in the following, we do this only if the topological manifold admits some oriented C k -structure that is unique up to orientation preserving C k -di?eomorphism. Obviously, each C k -structure is also a C ? -structure for any 0 ≤ ? < k, including the topological case ? = 0. As long as we are only interested in C r -manifolds up to equivalence, this time excluding the topological case, i.e. 1 ≤ r ≤ ∞, we can restrict ourselves to C ∞ manifolds as the following theorem of Whitney shows, see, for example [48]. Theorem A.6. Let M be a C r -manifold, 1 ≤ r ≤ ∞, with some C r -structure [A(r) ] and some k with r ≤ k ≤ ∞. 1. There exists a C k -structure [A(k) ] for M . 2. The C r -manifolds (M, [A(r) ]) and (M, [A(k) ]) are C r -di?eomorphic. 3. Let [B (k) ] denote any other C k -structure for M . Then (M, [A(k) ]) and (M, [B (k) ]) are C r -di?eomorphic. Of all the C ? -manifolds, 0 ≤ ? ≤ ∞, the relevant types for the purpose of classi?cation are therefore the topological (or C 0 -) and the C ∞ -manifolds. A C ∞ -structure is usually called a di?erentiable structure. C ∞ -maps and d-dimensional C ∞ -manifolds are known as smooth maps and smooth d-manifolds, and C ∞ -di?eomorphisms are often called just di?eomorphisms.

A.3

Simplices and piecewise-linear manifolds

In this section, we de?ne another type of manifold by restricting the transition functions of topological manifolds to piecewise-linear (PL) maps. These are the maps that are compatible with triangulations of the sets ?i (Ui ) ? d for the coordinate systems (Ui , ?i ). Let us start with the notion of a simplex.

?

De?nition A.7. Fix some dimension d ∈

?. ?d , k ∈ ?0, is the set ?d .
(A.4)
k

1. The convex hull of some ?nite set of points A = {p0 , . . . , pk } ?
k

[A] = [p0 , . . . , pk ] := {
j=0

p j λj :

λj ≥ 0,
j=0

λj = 1 } ?

We also de?ne [?] := ?.

2. A ?nite set of points A = {p0 , . . . , pk } ? d , k ∈ , is called a?ne independent if the set of vectors {p1 ? p0 , . . . , pk ? pk?1 } is linearly independent. Sets containing only one point are by de?nition a?ne independent. 3. A k-simplex in d , k ∈ 0 , is a set σ ? d of the form σ = [p0 , . . . , pk ] for some a?ne independent set {p0 , . . . , pk } ? d . A (?1)-simplex is by de?nition the empty set.

?

?

?

?

?

?

4. Let σ = [A] and τ = [B] be simplices in d so that A and B are a?ne independent sets. The simplex σ is called a face of τ if A ? B. In this case we write σ τ .

?

Quantum general relativity and the classification of smooth manifolds

31

Note that the de?nition of a?ne independence does not depend on the numbering of the points. For each simplex σ ? d , there is a unique ?nite set A ? d such that A is a?ne independent and σ = [A]. In this case, we call Vert σ := A the set of vertices of σ. Each face of a given k-simplex, k ∈ 0 , is an ?-simplex for some ? ≤ k. Each 0-simplex has got precisely one face, the empty set, and ? σ for all simplices σ. The relation ‘ ’ is a partial order on the set of all simplices in d .

?

?

?

?

De?nition A.8. 1. A set S of simplices in d is called locally ?nite if each p ∈ d has got a neighbourhood U such that U ∩ σ = ? only for a ?nite number of simplices σ ∈ S. 2. A polyhedron P ?

?

?

?d is a set of the form

|S| :=
σ∈S

σ,

(A.5)

for some locally ?nite set S of simplices in 3. A simplicial complex K in (a) (b) if τ, σ ∈ K then τ ∩ σ

?d is a locally ?nite set of simplices in ?d such that, whenever σ ∈ K and τ σ for any simplex τ ? ?d , then also τ ∈ K, and
τ and τ ∩ σ σ. {p} ∈ K } the set

?d .

4. A simplicial complex K is called ?nite if K is a ?nite set. 5. If K is a simplicial complex in of its vertices.

?d, we denote by Vert K := { p ∈ ?d :

6. If K is a simplicial complex, the set |K| is called its underlying polyhedron. 7. If a polyhedron P is of the form P = |K| for some simplicial complex K, then K is called a triangulation of P . Each polyhedron P ? d has got a triangulation. If P is compact, then there exists a triangulation of P which is a ?nite simplicial complex. We can now de?ne piecewise-linear (PL) maps to be the maps compatible with the triangulations of polyhedra. De?nition A.9. Let P ? K.

?

?m, Q ? ?n be polyhedra and P

= |K| for some triangulation

1. A map f : P → Q is called piecewise-linear (PL) if it is continuous and if the restrictions f |σ are a?ne maps for each simplex σ ∈ K. 2. The map f is called piecewise di?erentiable (PD) if it is continuous and if the f |σ are C ∞ -maps of maximum rank for each σ ∈ K. The inverse of any piecewise-linear homeomorphism is also piecewise-linear. The notion of piecewise di?erentiability plays a role when we discuss the compatibility of piecewise-linear and smooth structures in Appendix A.4 below. De?nition A.10. Let K and L be simplicial complexes. A map f : |K| → |L| is called simplicial is it is continuous, if f maps vertices to vertices, i.e. f (p) ∈ Vert L for all p ∈ Vert K, and if for each k-simplex [p0 , . . . , pk ] ∈ K, the simplex generated by the images [f (p0 ), . . . , f (pk )] is contained in L. A simplicial isomorphism is a simplicial map that is a homeomorphism.

32

Quantum general relativity and the classification of smooth manifolds

Let P and Q be polyhedra. A map f : P → Q is PL if and only if f is a simplicial map for some triangulations K and L such that P = |K| and Q = |L|. De?nition A.11. ? An oriented k-simplex (σ, ≤) is a k-simplex σ together with a linear order ‘≤’ on the set of its vertices Vert σ. In the bracket notation [q0 , . . . , qk ] for ksimplices, we can write for the oriented simplex, ε · [pτ (0) , . . . , pτ (k) ], (A.6)

where τ ∈ Sk+1 is a permutation such that p0 ≤ · · · ≤ pk and ε := sgn τ ∈ {?1, +1} is the sign of the permutation. For example, for 2-simplices, [p1 , p0 , p2 ] = ?[p0 , p1 , p2 ]. (A.7)

? The faces of an oriented k-simplex σ = [p0 , . . . , pk ] ? d are the oriented simplices with the induced orientation, (?1)j · [p0 , . . . , pj , . . . , pk ], (A.8) where the hat denotes the omission of a vertex. ? An oriented simplicial complex (K, ≤) is a simplicial complex K together with a partial order ‘≤’ on the set of vertices Vert K that restricts to a linear order on Vert σ for each σ ∈ K. ? A simplicial isomorphism f : |K| → |L| between oriented simplicial complexes (K, ≤) and (L, ≤) is called orientation preserving if it is compatible with the partial order, i.e. if p ≤ q implies f (p) ≤ f (q) for all p, q ∈ Vert K. Each d-simplex in d , σ = [p0 , . . . , pd ], inherits an orientation from d such that the linear order of its vertices is p0 ≤ · · · ≤ pk up to an even permutation precisely when det(p0 ? p1 , . . . , pd?1 ? pd ) > 0. Each subset U ? d that is open in d , is contained in some + + polyhedron which has a triangulation in terms of an oriented simplicial complex, compatible with the orientation inherited from d . We are now ready to restrict the transition functions of manifolds to piecewise-linear maps.

?

?

?

?

?

?

De?nition A.12. Let M be a topological d-manifold, d ∈

?.

1. A PLd -atlas for M is a TOPd -atlas A = {(Ui , ?i ) : i ∈ I} such that for all i, j ∈ I with Ui ∩ Uj = ?, the transition functions ?ji are piecewise-linear homeomorphisms. 2. An oriented PLd -atlas for M is a PLd -atlas such that the transition functions ?ji are orientation preserving simplicial maps with respect to the orientation inherited from d.

?

3. An [oriented] PLd -structure [A] for M is an equivalence class of [oriented] PLd -atlases. The following is as usual. De?nition A.13. 1. A d-dimensional [oriented] PL-manifold (M, [A]), d ∈ logical d-manifold M equipped with an [oriented] PL-structure [A].

?, is a topo-

2. Let (M, [A]), (N, [B]) be PL-manifolds with atlases A = {(Ui , ?i ) : i ∈ I} and B = {(Vj , ψj ) : j ∈ J}. A map f : M → N is called PL if f is continuous and if for any p ∈ M such that p ∈ Ui , f (p) ∈ Vj for some i ∈ I, j ∈ J, the map, ψj ? f ? ??1 : ?i (Ui ) → ψj (Vj ) i is piecewise-linear. (A.9)

Quantum general relativity and the classification of smooth manifolds 3. A PL-isomorphism f : M → N is an invertible PL-map whose inverse is PL.

33

4. A PL-isomorphism is called orientation preserving if all its coordinate representations (A.9) are orientation preserving simplicial maps. 5. Two PL-manifolds are called equivalent if they are PL-isomorphic. The boundary ?M of each d-dimensional [oriented] PL-manifold is a (d ? 1)-dimensional [oriented] PL-manifold without boundary.

A.4

Triangulations of smooth manifolds

A smoothing of some PL-manifold is a di?erentiable structure that is compatible with the PL-structure in the following way. De?nition A.14. Let M be a PL-manifold with the PL-atlas A = {(Vi , ψi ) : i ∈ I}. A di?erentiable structure represented by some C ∞ -atlas {(Uj , ?j ) : j ∈ J} on M is called a smoothing of the PL-structure [A] if on each non-empty overlap Uj ∩ Vi = ?, i ∈ I, j ∈ J, the homeomorphism
?1 ?j ? ψi |ψi (Uj ∩Vi ) : ψi (Uj ∩ Vi ) → ?j (Uj ∩ Vi )

(A.10)

is piecewise-di?erentiable. Whitehead’s theorem [54] guarantees that any smooth manifold of arbitrary dimension can be triangulated in this way. Theorem A.15 (Whitehead). For each smooth manifold M , there exists a PL-manifold MPL , called its Whitehead triangulation, such that M is di?eomorphic to a smoothing of MPL . Any two di?eomorphic smooth manifolds have PL-isomorphic Whitehead triangulations.

A.5

Combinatorial manifolds

We have de?ned PL-manifolds above as topological manifolds subject to the additional condition that their transition functions are PL. In this section, we review the concept of combinatorial manifolds which reduces the study of an entire PL-manifold to the study of a single simplicial complex and its combinatorics. Notice that the underlying polyhedron |K| ? n of any simplicial complex K in n , forms a paracompact Hausdor? space with the relative topology induced from n .

?

?

?

De?nition A.16. The underlying polyhedron |K| of a simplicial complex K in n for some n ∈ , is called a combinatorial d-manifold if it forms a d-dimensional PL-manifold, d ∈ .

?

?

?

In order to describe the relationship between PL-manifolds and combinatorial manifolds, we need the notion of the join and the link of simplices. De?nition A.17. Let σ, τ ? A, B ? d .

?

?d be simplices, σ = [A], τ = [B] with a?ne independent sets

1. σ and τ are called joinable if the set A ∪ B is a?ne independent. 2. If σ, τ are joinable, their join is de?ned to be the set σ · τ := [A ∪ B].

34

Quantum general relativity and the classification of smooth manifolds

De?nition A.18. Let K be a simplicial complex and σ ∈ K. The link of σ in K is the following set of simplices, lkK (σ) := {τ ∈ K : σ and τ are joinable and σ · τ ∈ K }. (A.11)

Notice that lkK (σ) is itself a simplicial complex. We also need the de?nition of PL-balls and PL-spheres. De?nition A.19. Let k ∈

? and {p0 , . . . , pk } ? ?d be a?ne independent.
B k := [p0 , . . . , pk ].

1. A piecewise-linear k-ball is a polyhedron which is PL-isomorphic to the k-simplex, (A.12) σ} which contains σ

2. For each simplex σ, we de?ne the simplicial complex σ := {τ : τ together with all its faces.

3. A piecewise-linear (k ? 1)-sphere is a polyhedron which is PL-isomorphic to the polyhedron, (A.13) S k?1 = ?B k := |{σ ∈ B k : σ = B k }|. A 0-ball is a set containing one point, and a 0-sphere a set with precisely two points. ?-balls and ?-spheres for ? < 0 are by de?nition the empty set. PL-manifolds can be characterized by combinatorial manifolds as follows. Theorem A.20. Each PL-manifold is PL-isomorphic to some combinatorial manifold. Conversely, the underlying polyhedron |K| ? n of some simplicial complex K admits a PLd structure, n, d ∈ , if and only if for each k-simplex σ ∈ K, the polyhedron | lkK (σ)| is a PL (d ? k ? 1)-ball or a PL (d ? k ? 1)-sphere.

?

?

This theorem implies that each PL-manifold M can be described by a single simplicial complex K with certain properties. If M is compact, then K can be chosen to be ?nite. Otherwise, due to paracompactness, K can always be built up from only a countable number of simplices. Furthermore, σ ∈ K is contained in the boundary ?M if and only if | lkK (σ)| is a PL (d ? k ? 1)-ball (as opposed to a sphere). Each (d ? 1)-simplex in K is the face of at most two d-simplices in which it occurs with opposite induced orientations.

A.6

Abstract triangulations

It is ?nally even possible to forget the surrounding standard space are contained, and to concentrate only on their combinatorics.

?n in which the simplices

De?nition A.21. An abstract simplicial complex (K 0 , K) is a set K 0 , called the set of abstract vertices, together with a family K ? P(K 0 ) of ?nite subsets of K 0 , called the set of abstract simplices, such that 1. {p} ∈ K for all p ∈ K 0 , and 2. If σ ∈ K and τ ? σ, then also τ ∈ K. For some subset A ? K of abstract simplices, we denote by (A , A) the smallest abstract simplicial complex that contains all simplices of A, i.e. A = { σ ∈ K : σ ? τ, τ ∈ A } and 0 A = { p ∈ K 0 : {p} ∈ A }. We often write just K rather than (K 0 , K). An oriented abstract simplicial complex is an abstract simplicial complex with a partial order on the set K 0 that restricts to a linear order on each σ ∈ K.
0

Quantum general relativity and the classification of smooth manifolds

35

Given some simplicial complex L in n , its abstraction is the abstract simplicial complex (K 0 , K) for which K 0 = Vert L and {p0 , . . . , pk } ∈ K if and only if [p0 , . . . , pk ] ∈ L. Conversely, given some ?nite abstract simplicial complex (K 0 , K), there exists a simplicial 0 complex L in |K |?1 whose abstraction is (K 0 , K). L is unique up to simplicial isomorphism. If (K 0 , K) is a ?nite abstract simplicial complex, we de?ne its underlying polyhedron as |K| := |L| which is well de?ned up to simplicial isomorphism. Recall the additional conditions (Theorem A.20) if this is supposed to yield a PL-manifold. We can now study the question of whether any two given PL-manifolds are PL-isomorphic in the context of the abstractions of their global triangulations. This is accomplished in a systematic way for any dimension by Pachner’s theorem [20]. We mention here the version for closed manifolds.

?

?

Theorem A.22 (Pachner). Let K and L be the ?nite abstract simplicial complexes associated with closed oriented combinatorial d-manifolds |K| and |L|, d ∈ . The combinatorial manifolds |K| and |L| are PL-isomorphic if and only if the abstract simplicial complexes K and L are related by a ?nite sequence of bistellar moves, the so-called Pachner moves.

?

The Pachner moves in dimension d ∈ can be described as follows. Consider an oriented (d + 1)-simplex σ = [0, 1, . . . , d + 1] where we write integer numbers for the linearly ordered vertices. Write down the set of faces of σ. These are d + 2 oriented d-simplices, { +[1, 2, . . . , d + 1], ?[0, 2, 3, . . . , d + 1], +[0, 1, 3, 4, . . . , d + 1], . . . , (?1)d [0, 1, . . . , d] }. (A.14) For each ? ∈

?

?, 1 ≤ ? ≤ d/2 + 1, partition this set into the following subsets,
A? = { +[1, 2, . . . , d + 1], . . . , (?1)??1 [0, . . . , ? ? 1, . . . , d + 1] }, B? = { ?(?1) [0, . . . , ?, . . . , d + 1], . . . , ?(?1) [0, 1, . . . , d] },
? d

(A.15a) (A.15b)

where the hat denotes omission of a vertex and where we have reversed the orientation of all simplices in B? . Both A? and B? generate oriented abstract simplicial complexes A? and B? . Observe that both polyhedra |A? | and |B? | have PL-isomorphic boundaries. The bistellar ?-move consists of cutting out the simplices of A? from the given abstract simplicial complex and gluing in those of B? . The Pachner moves in dimension d are the bistellar ?-moves, 1 ≤ ? ≤ d/2+1, and their inverses. For manifolds with boundary, the Pachner moves are the so-called elementary shellings [20].

A.7

Homotopy type

Topological manifolds can not only be compared by studying whether they are homeomorphic or not, but there is also the weaker notion of homotopy type. De?nition A.23. Let M , N be topological spaces. 1. Continuous maps f, g : M → N are called homotopic (f ? g) if there exists a continuous map F : [0, 1] × M → N such that F (0, p) = f (p) and F (1, p) = g(p) for all p ∈ M . The map F is called a homotopy. 2. M and N are called homotopy equivalent or of the same homotopy type (M ? N ) if there exist continuous maps f : M → N and h : N → M such that f ? h ? idN and h ? f ? idM .

36

Quantum general relativity and the classification of smooth manifolds

A.8

Comparison of structures

By Theorem A.15, any two di?eomorphic smooth manifolds have got PL-isomorphic PLmanifolds as their Whitehead triangulations. Each PL-manifold has got some underlying topological manifold, and PL-isomorphic PL-manifolds obviously have homeomorphic underlying topological manifolds. Finally, homeomorphic topological manifolds are always of the same homotopy type. We therefore have the following hierarchy, di?eomorphic =? PL-isomorphic =? homeomorphic =? homotopy equivalent (A.16)

Which of these arrows can be reversed? It turns out that the answer to this question strongly depends on the dimension. Before we comment of the individual arrows in greater detail, we mention the following algebraic limitation to the classi?cation of manifolds. A.8.1 Impossibility of complete classi?cation

A complete classi?cation of manifolds of dimension d ≥ 4 is not possible, even if the question is restricted to the classi?cation of topological manifolds up to homotopy equivalence. Manifolds M ? N of the same homotopy type have isomorphic fundamental groups π1 (M ) ? π1 (N ). = The problem is that for any d ≥ 4 and for any group G given by a ?nite presentation in terms of generators and relations, there exists a closed topological d-manifold M with π1 (M ) = G, see, for example [16]. But there exists no algorithm that decides for any two given ?nite group presentations whether they describe isomorphic groups or not [55]. This algebraic di?culty is the reason why some results on d-manifolds, d ≥ 4, are usually stated only for closed, connected and simply connected manifolds. In the case of some generic closed and connected manifold M , one would therefore study its universal covering M ?rst, but then the covering map itself whose ?bres are just π1 (M ), can be extremely complicated. A.8.2 PL-isomorphism versus di?eomorphism

In dimensions d ≤ 6, the equivalence classes of smooth manifolds up to di?eomorphism are in one-to-one correspondence with those of PL-manifolds up to PL-isomorphism as the following theorem shows. Theorem A.24. Let d ≤ 6. Each d-dimensional PL-manifold admits a smoothing, and any two PL-isomorphic PL-manifolds of dimension d give rise to di?eomorphic smoothings. For d ≤ 3, this theorem is a classical result which already holds for the underlying topological manifolds and the unique smooth structure they admit [56]. For d ∈ {5, 6}, it follows from a theorem of Kirby–Siebenmann [50] whereas in d = 4, it is a consequence of a result by Cerf [57] in the context of the general smoothing theory, see, for example [51, 58]. For a given d-dimensional PL-manifold, d ≥ 7, there can be obstructions to the existence of smoothings as well as a ?nite ambiguity, i.e. there can exist a ?nite number of pairwise non-di?eomorphic smoothings, see, for example [50]. An example for such an ambiguity is provided by the exotic smooth structures of S 7 [17], and there exists an 8-dimensional PLmanifold that does not admit any di?erentiable structure at all [50]. Usually, the smoothings of some given PL-manifold (and also the PL-structures of some given topological manifold) are classi?ed not up to di?eomorphism, but up to the stronger relation of isotopy [50,51]. We have formulated the results above (Theorem A.15 and A.24) in such a way that this subtlety is

Quantum general relativity and the classification of smooth manifolds

37

hidden: If two d-dimensional PL-manifolds, d ≤ 6, are PL-isomorphic, then their smoothings are di?eomorphic. The corresponding di?eomorphism, however, is in general not the same map as the PL-isomorphism with which one has started, but only isotopic to it. A.8.3 Homeomorphism versus PL-isomorphism

In dimension d ≤ 3, the equivalence classes of PL-manifolds up to PL-isomorphism are in oneto-one correspondence with those of topological manifolds up to homeomorphism because of the following theorem [56]. Theorem A.25 (Bing, Moise). Let d ≤ 3. Each topological d-manifold admits a PLstructure (proof of the triangulation conjecture). Homeomorphic topological d-manifolds give rise to PL-isomorphic PL-structures (proof of the Hauptvermutung of Steinitz). This is no longer true if d = 4. The work of Freedman [59] and Donaldson [4] has lead to the construction of families of smooth manifolds that are homeomorphic, but pairwise non-di?eomorphic. There exists, for example, an uncountable family of pairwise nondi?eomorphic di?erentiable structures on the topological manifold 4 [15, 16], and there are countably in?nite families of homeomorphic, but pairwise non-di?eomorphic compact smooth manifolds [13, 14]. Furthermore, there exist topological 4-manifolds that do not admit any di?erentiable structure at all [59]. Together with Theorem A.15 and Theorem A.24, this implies that in d = 4, there is a substantial discrepancy between the classi?cation of topological manifolds up to homeomorphism and that of PL-manifolds up to PL-isomorphism. In dimension d ≥ 5, there can also be both obstructions to the existence of a PL-structure on some given topological manifold as well as an ambiguity, but the ambiguity leaves only a ?nite number of choices, see, for example [50]. Special attention has to be paid to 5-manifolds whose boundary is non-empty because of the special features in d = 4.

?

A.8.4

Homotopy equivalence versus homeomorphism

It is a classical result that any two closed oriented topological 2-manifolds that are homotopy equivalent, are in fact homeomorphic. The only invariant required in order to determine the homotopy type of a closed oriented topological 2-manifold, is the genus. This correspondence disappears in dimension d = 3 in view of the lens spaces L(p, q). There exist closed oriented topological 3-manifolds that are homotopy equivalent, but not homeomorphic, for example L(7, 1) and L(7, 2) [43]. In dimension d ≥ 4, however, homotopy type is again very important in the classi?cation of topological manifolds. Very strong results are provided by the h-cobordism theorems of Smale [60] (d ≥ 5, even in the smooth case) and of Freedman [59] (d = 4, restricted to the topological case). In particular, there is Freedman’s theorem [59]. Theorem A.26 (Freedman). Let M , N be connected and simply connected closed topological 4-manifolds. If M and N are homotopy equivalent, then they are homeomorphic. Furthermore, the generalized Poincar? conjecture is true in d = 4 [59] as well as in d ≥ e 5 [60, 61]. Theorem A.27 (generalized Poincar? conjecture). Let M be a topological d-manifold, e d ≥ 4. If M is homotopy equivalent to S d , then M is homeomorphic to S d .

38

Quantum general relativity and the classification of smooth manifolds

If one wants to study smooth manifolds and start from the underlying manifold up to homotopy type, one has to supply additional data on the structure of the tangent bundle and then resolve a ?nite ambiguity, see, for example [62]. A.8.5 Summary (Table 1 of Section 4)

In dimension d = 2, all four arrows of (A.16) can be reversed, at least for closed oriented manifolds. In dimension d = 3, each topological manifold still admits a di?erentiable structure that is unique up to di?eomorphism, but homotopy equivalence no longer implies homeomorphism. In dimension d = 4, each PL-manifold still admits a smoothing that is unique up to di?eomorphism. Topological manifolds of the same homotopy type are also homeomorphic, at least for closed, connected and simply connected manifolds, but homeomorphism no longer implies di?eomorphism. There can rather exist in?nitely many pairwise inequivalent di?erentiable structures for the same underlying topological manifold. In dimension d ≥ 5, there can ?nally be obstructions and ambiguities both for having PL-structures on a given topological manifold and for smoothing a given PL-manifold. All the ambiguities in d ≥ 5, however, are ?nite [50]. What we have summarized so far, ?nally implies Theorem 1.1 of the Introduction. There are therefore two very striking ‘gaps’ in which the material we have reviewed so far, does not su?ce in order to reverse the arrows of (A.16), ? in d = 3 to study topological manifolds that are homotopy equivalent but not homeomorphic, ? in d = 4 to study smooth manifolds that are homeomorphic, but not di?eomorphic. The Turaev–Viro invariant [27], closely related to the partition function Z(M ) of quantum gravity in d = 2+1, provides non-trivial topological information on the ?rst of these questions. We speculate that quantum gravity in d = 3 + 1 will provide new insight into the second question.

References
[1] C. W. Misner: Feynman quantization of general relativity. Rev. Mod. Phys. 29, No. 3 (1957) 497–509. [2] M. F. Atiyah: Topological quantum ?eld theories. IHES Publ. Math. 68 (1988) 175– 186. [3] E. Witten: Quantum ?eld theory and the Jones polynomial. Comm. Math. Phys. 121 (1989) 351–399. [4] S. K. Donaldson: Polynomial invariants for smooth four-manifolds. Topology 29, No. 3 (1990) 257–315. [5] E. Witten: Monopoles and four-manifolds. hep-th/9411102. Math. Res. Lett. 1 (1994) 769–796,

[6] C. W. Misner, K. S. Thorne and J. A. Wheeler: Gravitataion. Freeman & Co., New York, 1973.

Quantum general relativity and the classification of smooth manifolds [7] R. M. Wald: General relativity. The University of Chicago Press, Chicago, 1984.

39

? [8] P. Peldan: Actions for gravity with generalizations: a review. Class. Quant. Grav. 11 (1994) 1087, gr-qc/9502010. [9] J. W. Barrett: Quantum gravity as topological quantum ?eld theory. J. Math. Phys. 36, No. 11 (1995) 6161–6179, gr-qc/9506070. [10] L. Crane: Clock and category: Is quantum gravity algebraic? J. Math. Phys. 36, No. 11 (1995) 6180–6193, gr-qc/9504038. [11] J. C. Baez and J. Dolan: Higher dimensional algebra and topological quantum ?eld theory. J. Math. Phys. 36, No. 11 (1995) 6073–6105, q-alg/9503002. [12] R. Oeckl: Schr¨dinger’s cat and the clock: Lessons for quantum gravity. Class. Quant. o Grav. 20 (2003) 5371–5380, gr-qc/0306007. [13] R. Friedman and J. W. Morgan: On the di?eomorphism types of certain algebraic surfaces. I. J. Di?. Geom. 27 (1988) 297–369. [14] C. Okonek and A. Van der Ven: Stable bundles and di?erentiable structures on certain elliptic surfaces. Invent. Math. 86 (1986) 357–370. [15] C. H. Taubes: Gauge theory on asymptotically periodic 4-manifolds. J. Di?. Geom. 25 (1987) 363–430. [16] R. E. Gompf and A. I. Stipsicz: 4-Manifolds and Kirby calculus. AMS Graduate Studies in Mathematics 20. American Mathematical Society, Providence, 1999. [17] M. A. Kervaire and J. W. Milnor: Groups of homotopy spheres. I. Ann. Math. 77, No. 3 (1963) 504–537. [18] J. S. Carter, L. H. Kauffman and M. Saito: Structures and Diagrammatics of four dimensional topological lattice ?eld theories. Adv. Math. 146 (1999) 39–100, math.GT/9806023. [19] M. Mackaay: Spherical 2-categories and 4-manifold invariants. Adv. Math. 143, No. 2 (1999) 288–348, math.QA/9805030. [20] U. Pachner: PL homeomorphic manifolds are equivalent by elementary shellings. Eur. J. Comb. 12 (1991) 129–145. [21] C. H. Brans: Localized exotic smoothness. Class. Quant. Grav. 11, No. 7 (1994) 1785–1792, gr-qc/9404003. [22] T. A?elmeyer-Maluga and C. H. Brans: Cosmological anomalies and exotic smoothness structures. Gen. Rel. Grav. 34 (2002) 1767–1771, gr-qc/0110043. [23] H. Pfeiffer: Di?eomorphisms from ?nite triangulations and the absence of ‘local’ degrees of freedom (2003). Preprint hep-th/0312060. [24] J. C. Baez: An introduction to spin foam models of BF theory and quantum gravity. In Geometry and Quantum Physics, Lecture Notes in Physics 543. Springer, Berlin, 2000, pp. 25–93, gr-qc/9905087.

40

Quantum general relativity and the classification of smooth manifolds

[25] D. Oriti: Space-time geometry from algebra: Spin Foam Models for nonperturbative quantum gravity. Rep. Prog. Phys. 64, No. 12 (2001) 1703–1756, gr-qc/0106091. [26] A. Perez: Spin foam models for quantum gravity. Class. Quant. Grav. 20, No. 6 (2003) R43–R104, gr-qc/0301113. [27] V. G. Turaev and O. Y. Viro: State sum invariants of 3-manifolds and quantum 6j-symbols. Topology 31, No. 4 (1992) 865–902. [28] G. Ponzano and T. Regge: Semiclassical limit of Racah coe?cients. In Spectroscopic and group theoretical methods in physics, ed. F. Bloch. North Holland Publications, Amsterdam, 1968, pp. 1–58. [29] S. Davids: A state sum model for (2 + 1) Lorentzian quantum gravity. Phd thesis, University of Nottingham (2001). Preprint gr-qc/0110114. [30] L. Freidel: A Ponzano–Regge model of Lorentzian 3-dimensional gravity. Nucl. Phys. Proc. Suppl. B 88 (2000) 237–240, gr-qc/0102098. [31] S. Mizoguchi and T. Tada: Three-dimensional gravity from the Turaev–Viro invariant. Phys. Rev. Lett. 68, No. 12 (1992) 1795–1798, hep-th/9110057. [32] J. W. Barrett: Geometrical measurements in three-dimensional quantum gravity. J. Phys. A: Math. Gen. 18 Suppl. (2003) 97–113, gr-qc/0203018. [33] E. R. Livine and R. Oeckl: Three-dimensional quantum supergravity and supersymmetric spin foam models (2003). Preprint hep-th/0307251. [34] H. Ooguri: Topological lattice models in four dimensions. Mod. Phys. Lett. A 7, No. 30 (1992) 2799–2810, hep-th/9205090. [35] J. C. Baez: Four-dimensional BF theory as a topological quantum ?eld theory. Lett. Math. Phys. 38 (1996) 129–143, q-alg/9507006. [36] L. Crane, L. H. Kauffman and D. N. Yetter: State-sum invariants of 4-manifolds. J. Knot Th. Ramif. 6, No. 2 (1997) 177–234, hep-th/9409167. [37] D. S. Freed and K. K. Uhlenbeck: Instantons and four-manifolds. Springer, New York, 1984. [38] J. E. Roberts: Skein theory and Turaev–Viro invariants. Topology 34 (1995) 771–787. [39] M. Mackaay: Finite groups, spherical 2-categories, and 4-manifold invariants. Adv. Math. 153, No. 2 (2000) 353–390, math.QA/9903003. [40] J. W. Barrett and B. W. Westbury: Invariants of piecewise-linear 3-manifolds. Trans. AMS 348, No. 10 (1996) 3997–4022, hep-th/9311155. [41] R. Dijkgraaf and E. Witten: Topological gauge theories and group cohomology. Comm. Math. Phys. 129 (1990) 393–429. [42] D. Altshuler and A. Coste: Invariants of three-manifolds from ?nite group cohomology. J. Geom. Phys. 11 (1993) 191–203.

Quantum general relativity and the classification of smooth manifolds [43] D. Rolfsen: Knots and Links. Publish or Perish, Wilmington, 1976.

41

[44] N. Reshetikhin and V. G. Turaev: Invariants of 3-manifolds via link polynomials of quantum groups. Invent. Math. 103 (1991) 547–597. [45] D. S. Freed and R. E. Gompf: Computer calculation of Witten’s 3-manifold invariant. Comm. Math. Phys. 141 (1991) 79–117. [46] J. F. P. Hudson: Piecewise linear topology. W. A. Benjamin, New York, 1969. [47] C. P. Rourke and B. J. Sanderson: Introduction to piecewise-linear topology. Ergebnisse der Mathematik und ihrer Grenzgebiete 69. Springer, Berlin, 1972. [48] J. R. Munkres: Elementary di?erential topology. Annals of Mathematics Studies 54. Princeton University Press, Princeton, 1963. [49] R. P. Geroch: Spinor structure of space-times in general relativity. I. J. Math. Phys. 9, No. 11 (1968) 1739–1744. [50] R. Kirby and L. C. Siebenmann: On the triangulation of manifolds and the Hauptvermutung. Bull. AMS 75 (1969) 742–749. [51] M. W. Hirsch and B. Mazur: Smoothings of piecewise linear manofolds. Annals of Mathematics Studies 80. Princeton University Press, Princeton, 1974. [52] J. G. Hocking and G. S. Young: Topology. sachusetts, 1961. Addison Wesley, Reading, Mas-

[53] J. Dugundji: Topology. Allyn and Bacon, Boston, 1975. [54] J. H. C. Whitehead: On C 1 -complexes. Ann. Math. 41, No. 4 (1940) 809–824. [55] M. O. Rabin: Recursive unsolvability of group theoretic problems. Ann. Math. 67, No. 1 (1958) 172–194. [56] E. E. Moise: A?ne structures in 3-manifolds V. The triangulation theorem and Hauptvermutung. Ann. Math. 56, No. 1 (1952) 96–114. [57] J. Cerf: Sur les di??omorphismes de la sph`re de dimension trois (Γ4 = 0). Lecture e e Notes in Mathematics 53. Springer, Berlin, 1968. [58] N. H. Kuiper: On ‘La nullit? de π0 (Di?S 3 )’ by J. Cerf. Mathematical Reviews e No. 6641 a–d Vol. 33 (1963) 1126. [59] M. H. Freedman: The topology of four-dimensional manifolds. J. Di?. Geom. 17 (1982) 357–453. [60] S. Smale: On the structure of manifolds. Amer. J. Math. 84 (1962) 387–399. [61] M. H. A. Newman: The engul?ng theorem for topological manifolds. Ann. Math. 84 (1966) 555–571. [62] D. Sullivan: In?nitesimal computations in topology. IHES Publ. Math. 47 (1977) 269–332.



学霸百科 | 新词新语

All rights reserved Powered by 甜梦文库 9512.net

copyright ©right 2010-2021。
甜梦文库内容来自网络,如有侵犯请联系客服。zhit325@126.com|网站地图