Spin Foam and Loop approaches to Quantum Gravity reformulate Einstein’s theory of relativity in terms of connection variables. The metric properties are encoded in face bivectors/conjugate fluxes that are required to satisfy certain conditions, in order to allow for their geometric interpretation. We show that the (sub-)set of the so-called ‘volume simplicity constraints’ is not implemented properly in the current EPRL-FK spinfoam vertex amplitude, if extended to arbitrary polyhedra.

We then propose that a certain knot-invariant of the bivector geometry, induced on the boundary graph, encodes the missing conditions, allowing for reconstruction of a polytope from its two-dimensional faces. Implemented in the quantum amplitude, this leads to corrected semi-classical asymptotics for a hypercuboid, and is conjectured to be non-trivial in more general situations.

The analysis of linear version of ‘volume simplicity’ suggests to switch from hypersurface normals to edge lengths, that is from 3-forms directly to tetrads – in the extended configuration space of the Plebanski constrained formulation. We then give the corresponding dual version of linear simplicity constraints, which prescribe 3d volume for the polyhedral faces in the boundary of a 4d polytope.

We also analyse the status of metric/vielbein degrees of freedom and the role of local translations in the classical Einstein-Cartan gravity, viewed as a Poincare gauge theory. The relation with the diffeomorphism symmetry is established through the key concept of development, which generalizes parallel transport of vectors in the geometric theory of Cartan connections. We advocate the latter to be the natural gauge-theoretic framework for the theory of relativity.


Spinschaum-Modelle und Schleifenquantengravitation reformulieren Einsteins Allgemeine Relativitätstheorie mit Hilfe von Zusammenhangsvariablen. Die metrischen Eigenschaften sind in den konjugierten Flüssen kodiert. Diese müssen gewisse Bedingungen erfüllen, um eine geometrische Interpretation der Variablen zuzulassen. Wir zeigen, dass eine dieser Bedingungen, die ‘‘Volumenzwangsbedingung’’, nicht richtig im EPRL-FK Modell implementiert ist, sobald sie auf generelle Polyeder verallgemeinert wird.

Wir schlagen dann eine gewisse Knoten-Invariante der Bivektorgeometrie im Randgraphen vor, die die fehlende Bedingung enthatlen könnte. Diese erlaubt eine Rekonstruktion des Polyeders aus dessen zweidimensionalen Flächen. Implementiert in der Quantenamplitude, hat diese dann die richtige semiklassische Asymptotik für den Hyperquader. Es wird vermutet, dass sie für allgemeinere Polyeder ebenfalls anwendbar ist.

Die Analyse der linearen Volumenzwangsbedingung legt nahe, von den Hyperflächenormalen zu den Kantenlängen überzugehen, also von 3-Formen direkt zu den Dreibeinen im erweiterten Konfigurationsraum des Plebanskiformalismus. Wir entwickeln die duale Version der linearen Volumenzwangsbedingung, welche 3d Volumina für die polyhedralen Flächen im Rand der 4d Polytope vorgibt.

Ebenfalls analysieren wir den Status der metrischen/Vielbein-Freiheitsgrade, sowie die Rolle der lokalen Translationen in klassischer Einstein-Cartan-Gravitation, formuliert als Poincare-Eichtheorie. Wir stellen diese als den natürlichen eichtheoretischen Rahmen für Relativitätstheorie dar.


standard \pdfstringdefDisableCommands\pdfstringdefDisableCommands

On Geometry and Symmetries in Classical and Quantum Theories of Gauge Gravity


zur Erlangung des Doktorgrades

an der Fakultät für Mathematik,

Informatik und Naturwissenschaften

Fachbereich Physik

der Universität Hamburg

vorgelegt von

Vadim Belov

aus Sankt Petersburg



Gutachter der Dissertation: Dr. Benjamin Bahr
Prof. Dr. Gleb Arutyunov
Zusammensetzung der Prüfungskommission: Dr. Benjamin Bahr
Prof. Dr. Gleb Arutyunov
Prof. Dr. Sven-Olaf Moch
Prof. Dr. Jörg Teschner
Prof. Dr. Dieter Horns
Vorsitzender der Prüfungskommission: Prof. Dr. Dieter Horns
Datum der Disputation: 10.04.2019
Vorsitzender Fach-Promotionsausschusses PHYSIK: Prof. Dr. Michael Potthoff
Leiter des Fachbereichs PHYSIK: Prof. Dr. Wolfgang Hansen
Dekan der Fakultät MIN: Prof. Dr. Heinrich Graener

The Book of Nature is written in the language of mathematics, and the symbols are triangles, circles and other geometrical figures, without whose help it is impossible to comprehend a single word of it; without which one wanders in vain through a dark labyrinth. [G. Galilei]

I attach special importance to the view of geometry which I have just set forth, because without it I should have been unable to formulate the theory of relativity. [A. Einstein]

I think I can safely say that nobody understands quantum mechanics. [R. Feynman]


Chapter I Introduction

The two cornerstones of fundamental physics are the Quantum theory and General Relativity, they provide some of the deepest insights into the nature of observable universe. The first one is primarily concerned with the elementary constituents of matter and their behaviour at the smallest scales, resulting in the Standard Model of particle physics and the description of three fundamental interactions (electroweak and strong nuclear forces) in terms of Quantum Field Theory (QFT). Whereas the second one presents the classical theory of gravitation, prescribing the dynamics of the large collection of masses and extending its effects across the whole universe, studied in cosmology. It endows the space-time itself with dynamical properties, usually formulated in the language of (pseudo-)Riemannian geometry.

The two disciplines thus rarely overlap in their usual domains of applicability, where the effects of one theory are negligibly small as compared to the other. It is only in the exotic physical conditions of high energy densities of matter concentrated in the small region – such as in the center of black holes, or at the big bang inception of expanding universe – where both theories should work together, describing the quantum properties of a strongly curved space-time geometry. Up until now, consistent theory of Quantum Gravity – that would combine the essential features of both types of description of dynamics – has not been fully developed. A number of promising candidates have been put forward, however, each emphasizing various aspects of the problem 111Such as string theory, loop and foam approaches, dynamical triangulations and quantum Regge calculus, causal sets, asymptotic safety hypothesis, enthropic and modified gravity – just to name a few.. It seems clear that the usual methods of the perturbative QFT do not work properly, leading to intractable singularities, and one must explore other directions.

In some of the modern non-perturbative approaches to the quantization of gravity, the role of Riemannian metrics is very much reduced, in favour of mathematically more tractable connection variables. For instance, in the canonical Loop Quantum Gravity (LQG) [1, 2, 3], the densitiezed triad of the hypersurface plays the part of momenta variables, conjugate to the configuration of the certain Ashtekar-Barbero connection. The emerging picture of ‘quantum space’ may be read in terms of a certain ‘twisted geometries’ [4, 5, 6, 7], consisting of the discrete polyhedra of ‘fuzzy’ shapes, glued non-trivially alond their two-dimensional faces. In the closely related Spin Foam (SF) state-sum/path-integral quantization [8, 9, 10], the analogous general bivectors of the 2-dimensional surfaces are first integrated out completely, in order to obtain the vacuum state of the so-called BF Topological Quantum Field Theory (TQFT), which can be rigorously quantized. (Part of) the metric information is then restored via restriction of summation in the partition function to the states that could attain the meaning of some ‘discrete metric geometry’. Thereby one also says that the respective quantum topological BF model is being ‘reduced’ to that of General Relativity (GR).

Both these strategies then struggle to recover some more familiar spacetime geometry from their formalisms, for instance, within some sort of semi-classical approximation (‘infra-red’ limit). Their reliance on the discrete structures for the regularization also leads to the continuum limit problems (‘ultra-violet’ completion), that is consistency of the results under arbitrary refinements. The inability to identify what constitutes a ‘geometry’ in these models impedes the sensible interpretation of their outcomes. In result, one cannot achieve an unequivocal conclusion, whether the gravitational field has been satisfactorily quantized or not. In this thesis we address the nature of the degrees of freedom that are usually associated with the (pseudo-)Riemannian metric field g, both from the discrete viewpoint of the above Spin Foam quantum amplitudes, as well as in the classical continuum field-theoretic framework.

Our first contribution consists in realization that one of the most reliable and studied SF proposals so far – the Engle-Pereira-Rovelli-Livine-Freidel-Krasnov (EPRL/FK) amplitude [11, 12, 13, 14, 15] – originally constructed and working well-enough for simplicial discretizations, does not extend trivially to the arbitrary polyhedral cells. Such an extension is required to match the general-valence graph kinematical states of the boundary configurations (as LQG suggests). This was indeed put forward by Kamiński-Kisielowski-Lewandowski (KKL) [16, 17, 18], following the original EPRL-recipe. By scrutinizing the instructive case of the hypercuboid in Ch. V, we explicitly demonstrate that the part of the so-called ‘simplicity constraints’, reducing BF to GR, is not implemented properly in the model [19]. In result, the vertex amplitude cannot be assigned the meaning of describing contribution from some (unique) flat polyhedron, like it was for the rigid 4-simplex. This is evidenced by the appearance in the large- asymptotic limit of the same type face-mismatch (and torsion) as in canonical LQG. This constitutes a major problem for interpretation of such states in terms of some (semi-)classical 4d geometry 222We are basing our analysis on the numerical studies first performed in [20] and subsequently generalized in [21]. This part of our work provides a possible interpretation for their results..

Our second contribution – after identification of the problem – was to propose some tentative solution. This is done in Ch. VI, more or less straightforwardly. The purely mathematical offspring of the present analysis was the formula, expressing the volume of 4-dim polyhedron in terms of bivectors of its 2-dim faces, proved in [22, 23]. Based on these developments, we link the missing set of constraints with certain type of knot invariants, related to embedding of the boundary graph in 4d [24]. We show how the condition of independence on the chosen particular knot, imposed on the amplitude (weakly), is capable of resolving the above-indicated issue, in the case of hypercuboid.

As has been noted, Spin Foams usually avoid operating full-fledged metric/vielbein variables, preferring instead to constrain the bivectors accordingly. The follow up of our analysis in Ch. V was another another realization that the vielbein/co-frame field is inherently present in the formalism: its degrees of freedom are encoded in the hypersurface normals (relatives of lapse-shift variables of canonical theory) that are exploited for constraints imposition. Arguably, they are as independent as the connection variables (cf. [25]). This led us to suggest as the starting point not the Lorentz-BF, but the Poincaré-BF [19], whose Klein’s homogeneous space suits the idea of the flat Minkowski vacuum more naturally. The corresponding classical theory is developed in Ch. VII, as well as the novel version of simplicity constraints is proposed (dual to the currently used). This is our third contribution to the analysis of quantum SF models.

The above findings can be concisely put in the form of a question: ‘‘What are the independent degrees of freedom of gravitational field that are actually being quantized?’’ This naturally led us to address the status of the metric/vielbein most directly in the classical theory of Einstein-Cartan gravity that underlies the quantum framework. In particular: since the quantization of connections is fairly well understood, it is all the more natural to phrase GR in these terms. Indeed, this was the driving force of Loop Quantum Gravity since Ashtekar uncovered his variables – in the canonical setting – related to local Lorentz symmetry (at the point, or ‘internal’). It is desirable thus to relate metric/vielbein variables as well as the associated symmetry w.r.t. diffeomorphism transformations (‘external’) – which are distinguishing features of gravity – to the local translations of the Poincaré gauge theory 333In particular, given their controversial status in the discrete framework, cf. [26, 27].. This constitutes the first half of the thesis, where the comprehensive study of classical Cartan connections is performed. (Also, in order to set up the stage for the rest of the work.)

The idea is certainly not new and (re-)appeared in many forms [28, 29, 30, 31, 32, 33]. The most noticeable proponents of Cartan connections are R. W. Sharpe [34] in mathematics of differential geometry, and D. K. Wise [35, 33] in the physics of gravity, respectively. In particular, regarding the gauge theoretic description of gravity and its symmetries, those are frequently viewed as arising from the sort of ‘symmetry breaking’ of some related topological theory (conceptually similar to Spin Foams).

We choose more conservative standpoint, actually dating back to É. Cartan himself. The mathematical theory of connections is laid out in Ch. II in accordance with his geometric intuition, following very closely to the presentation in [34] (to which we refer for the most of proofs). Nevertheless, we make more extensive use of the theory of geometric -structures, adapted to the bundle framework of modern gauge theories [36], in what we called a ‘(locally) Klein bundle’. It enjoys the full action of the principal Poincaré group of geometry, while the role of a base manifold is diminished. The notion of Cartan connection then embodies the geometric picture of ‘‘rolling’’ the affine space on the ‘‘lumpy’’ surface of a curved manifold.

The main outcome of our analysis is the implementation of diffeomorphism transformations as the gauge group of translations, which is ‘not broken’ but an exact local symmetry. This requires a refined notion of tensors in affine space, which can be either ‘free’ or ‘bound to a point’. One then shows how the covariant derivative could be seen as a Lie dragging that acts both on vectors and points. In Ch. III, the physical theory of relativity is given in terms of Cartan geometry, following [37]. In particular, the geometric discretization of forms by means of integration is naturally seen in terms of vector summation. The series of results is obtained on torsion in relation to (non-) closure of surfaces with ‘defects’. We then discuss tangentially the possible implications for quantization, when the approaches of LQG and Spin Foams are briefly reviewed in Ch. IV, bridging the gap between ‘classical’ and ‘quantum’ parts of the thesis.

Chapter II Background of ‘background-independence’

The ultimate goal of the search for the theory of Quantum Gravity (QG), broadly defined, is to develop a conceptual and mathematical scheme which combines the relativistic field-theoretical description of gravitation with the main principles of quantum physics within a coherent framework. The standard methods of the perturbative QFT lead to intractable divergences when applied to the gravitational field and space-time itself, prompting to revisit the role of background structure(s). Let us retrace consecutively the basic assumptions and fundamental principles to be thoroughly implemented in the modern non-perturbative approaches to the quantization of gravity. The selection of material is quite subjective and may differ somewhat from the traditional presentation. Whereas the language may appear less formal, for the sake of accessibility, we supplied all the references to the rigorous proofs and statements wherever possible.

There are two pieces of the puzzle – 1) the gravity and 2) the quantum, respectively, that one has to specify first in order to make the whole subject tangible. The best description of the gravitational phenomena up to date is provided by the Einstein’s classical theory of General Relativity (GR), so that its basic assumptions are to be taken at face value and (at least some of them) realized quantum mechanically. This emphasis on GR postulates is characteristic to such conservative 111’Conservative’ in the sense of not altering the theory’s foundations without compelling evidences. This is what distinguishes it from other approaches where general relativistic description is viewed as effective (or emergent), such as string theory. approaches as Loop Quantum Gravity (LQG), Spin Foams (SF) and related programs. The specialty of gravity is that its field, in comparison with other known interactions of the Standard Model, directly relates to the mutual displacement of objects, prescribing the change of their relative situation in the manifold of conceivable ‘points-events’. According to Einstein, this type of gravitational pull is realized through the properties of the spacetime itself, that is endowing the abstractly defined (topological) with geometry.

The notion of the manifold is the standard one and intuitively clear, though we remind it promptly. Roughly speaking, it serves the purpose of introducing the way to label points and differentiate functions in a consistent manner. 222According to [34, p.2]: In the hierarchy of geometry (whose ‘‘spine’’ rises from homotopy theory through cell complexes, through topological and smooth manifolds to analytic varieties), the category of smooth manifolds and maps lies ‘‘halfway’’ between the global rigidity of the analytic category and the almost total flabbiness of the topological category. We might say that a smooth manifold possesses full infinitesimal rigidity governed by Taylor’s theorem while at the same time having absolutely no rigidity relating points that are not ‘‘infinitesimally near’’ each other, as is seen by the existence of partitions of unity . . . Smooth manifolds are sufficiently rigid to act as a support for the structures of differential geometry while at the same time being sufficiently flexible to act as a model for many physical and mathematical circumstances that allow independent local perturbations. The latter is precisely our main focus in the context of modern QG approaches, where this relation between global-and-local, discrete-and-continuous is often debated. [Since we are interested in the local field excitations, for the mathematical details on the notions from topology, concerning global aspects, we refer to the textbooks, e.g. [38, 39, 40]] The crucial part is that its points – and hence the coordinate labels that can be chosen arbitrarily – have no internal significance per se. They attain characterization only through the dynamical elements of the theory themselves, such as particle collisions (an example of Einstein’s ‘spacetime coincidences’, cf. [2, sec.2.2.5]) or certain field value measurements. In other words, whereas the fields of the Standard Model are localized w.r.t. inert/non-dynamical Minkowski (or de Sitter) space, in GR they are localized w.r.t. one another, what embodies the relational view of the world. The contiguity relation becomes dynamical entity in its own right that we know under the name of gravitational field. Quoting from [2, p.10]:

This absence of the familiar spacetime ‘‘stage’’ is called the background independence of the classical theory. Technically, it is realized by the gauge invariance of the action under (active) diffeomorphisms. A diffeomorphism is a transformation that smoothly drags all dynamical fields and particles from one region of the four-dimensional manifold to another. In turn, gauge invariance under diffeomorphism (or diffeomorphism invariance) is the consequence of the combination of two properties of the action: its invariance under arbitrary changes of coordinates and the fact that there is no nondynamical ‘‘background’’ field.

In respect, all the (non-gravitational) fields inhabiting the arena react backwards on the geometry, universally coupling it to energy-momentum – the quantity of motion contained in the ‘medium’. All this makes Einstein’s relativity stand out as the general theory of motion, with the ‘shape’ of spacetime actively responding to the configuration of the matter. The challenge of Quantum Gravity is precisely to fully incorporate this drastic novelty and to understand what is a general-relativistic, or background-independent, Quantum Field Theory (QFT).

Note that we kept discussion as general as possible so far and did not specify yet what we understand by ‘geometry’ and how exactly it relates to the displacement. This is to stress the universal significance of the above fundamental concepts, having roots in the very nature of how one experiences reality and describes observations. In fact, the physical content of GR can be expressed in multiple possible ways 333This is not to mention almost the infinitude of conceivable corrections and modifications of the Einstein’s GR, compatible with the empirical data., putting various its aspects in the forefront, depending on what is considered essential [e.g., metric or connection?]. It is already at this early stage, the preference given to one formalism over the other may play a crucial role, and the choice of the appropriate mathematical language is the first step in the quantization process.

When a long computation gives a short answer, then one looks for a better method,

– these words preceded in [41, p.344] the introduction of an effective computational device that, however, demands a heavier investment in the calculus of differential forms

than anyone would normally find needful for any introductory survey of relativity.

In contrast, the present chapter provides an account of the mathematical theory of connections from a geometric standpoint – avoiding as much as possible what Cartan 444According to [42, p.261] a powerful mathematician who possessed a remarkable geometrical intuition which enabled him to see the geometrical content of very complicated calculations. In fact Cartan often used geometrical arguments to replace some of the calculations, and a reader who does not possess this remarkable gift is often baffled by his arguments. It is hoped that as a result of reading this brief introduction the reader will be encouraged to make a serious study of Cartan’s book  [43]. We wholeheartedly subscribe under the above words, and may only recommend adding [44, 37, 34] to the list. called ‘‘les débauches d’indices’’ – in an effort to stress the naturality of this framework.

ii.1 (Principal) fiber bundles and the general notion of gauge

The rationale behind GR was the realization that the inertial motion has relative status, whereas the physical effects of force depend on the chosen observer. However, the laws of physics must be expressed in the form of mathematical equations, valid for any admissible observer’s frame of reference. Known as the (generalized) Galilei/Einstein’s principle of relativity, it is at the core of both GR and the gauge theories of particle interactions. Once the key notions of a frame and the transformations between them are specified, this leads to the geometrization of all the known forces of nature.

In a different vein, since all measurements are made relative to a choice of frame (broadly defined), and the measurement process can never be completely divorced from the aspect of the universe being measured, we are led to the concept of the bundle of reference frames, corresponding to the smooth concatenation of all admissible observers, each with its own measuring apparatus. The geometry (broadly defined) is all about the relations, left invariant under admissible frame transformations – in accordance with Klein’s Erlangen program [34, 45].

Figure II.1: Schematic picture of the manifold and consistent coordinate charts (from [36]).

The typical example of a reference frame (and the one which underlies the formulation of GR) is provided by the smooth structure on itself, making it into a differentiable manifold. Concretely, let be the atlas, consisting of coordinate charts , covering some (paracompact, Hausdorff) by open sets of the Euclidean space 555It is crucial that the affine structure of allows to canonically identify all the tangent spaces by the parallel transport of vectors to the common point , so that the notion of differentiation makes sense., such that the transition functions between overlapping charts are all smooth (see Fig. II.1). Then the local parametrization of via some backwards mapping extends to the derivative map , which establishes the linear isomorphism between the corresponding tangent spaces at the point and that of its image – this being referred to as a frame (at the point):


For a time being, let us use interchangeably for the point and its parametrization. Given any standard basis in at , such a map determines a basis in . Let be the set of all (linear) frames at , then put


to be the linear frame bundle. It comes equipped with a natural right action , given by for each , which may be regarded as the change of standard basis, used by observer to describe its local neighbourhood. This is an example of a more general construction of the principal fiber bundle (PFB), recall it to be precise.

Definition II.1.

A (locally trivial) fiber bundle with an abstract fiber consists of a quadruple of (smooth) manifolds (called the total space), (called the base), and the (standard) fiber , together with a submersion map (called projection, sometimes), such that is diffeomorphic by the local trivialization map to for an open set , containing , and the following diagram commutes

The pair is called a chart (or local bundle coordinate system, subordinate to the chosen coordinates on the base manifold ).

Definition II.2.

Suppose that a Lie group acts smoothly on as a group of diffeomorphisms. A  atlas for bundle is a collection of charts , covering , such that for each pair of charts in the map


called a coordinate change, has the form , where the smooth maps are called transition functions and satisfy:

  1. for all ;

  2. for all ;

  3. for all .

The two atlases are equivalent if their union is also a atlas (from which a unique maximal one can always thus be formed). A bundle is a , on which a structure is specified, that is an equivalence class of atlases.

The transition functions describe how the direct products glue together to form the total space. Indeed, can be considered as factorspace, obtained from the disjoint union via the equivalence relation, identifying the points and . Given the covering of by and the transition functions, satisfying 1-3, the bundle can be reconstruted with virtually any -module space as a typical fiber.

Definition II.3 (Pfb).

For a special type of principal bundle, its structure group is diffeomorphic to its standard fiber, on which acts by left translations. It can always be associated to by considering the fiber to consist of the (generalized) -frames at the point , and the diffeomorphism from the standard fiber acting as a change of basis (cf. [46, Prop.1.5.4]). Consequently, the total space is then endowed with a smooth right action (commuting with the left coordinate changes) that is fiber preserving and acts simply transitively on each fiber (which is just the orbit of through above .

In fact, the action alone is enough to fully characterize the PFB, as the following implies.

Theorem II.1 ([34, App. E]).

Let be a smooth manifold, a Lie group, and the (smooth) right action be free (i.e. if for some ) and proper (i.e. if and compact is compact). Then

  1. with a quotient topology is a topological manifold ;

  2. has a unique smooth structure for which the canonical projection is a submersion;

  3. The tuple is a smooth principal right G bundle.

The construction in (II.2) is a PFB with a structure group and an obvious projection . Other PFBs with different groups, such as unitary , are prolific in the standard model of particle physics, where the notion of a ‘frame’ could mean the choice of zero-phase angle or direction of axes in the (so-called ‘internal’) isospin space [47, 29, 36]. Since the ‘glueing’ of fibers is solely determined by transition functions, the bundle over with typical fiber can always be associated to by a smooth effective (left) action via representation . Denoted , this associated fiber bundle consists of equivalence classes, identifying the points , and can be obtained from as the orbit space by factoring through the group action


and the standard projection .

Definition II.4.

A local section of a bundle is the smooth map from the open subset such that . The space of (local) sections of over we will denote , or simply . When can be extended to the whole of , one speaks of global sections.

The configurations of a (matter) field are described by the elements in a coordinate-independent (geometric) manner. In [particle] physics, one usually works with the local (-valued) functions in a region , through the use of bundle charts . Implicit in this determination is the choice of some basis in , corresponding to certain states of the system. This fixed, but arbitrary choice of (moving) reference frame – known as gauge – is the necessary key ingredient for describing motions in a background independent fashion, referring only to immanent (and dynamical) elements of the system.

Definition II.5.

The gauge is the (local) section of a PFB with the group . Since has the local trivializations of the form , where satisfy the right equivariance , there is a natural 1-to-1 correspondence between trivializations and gauges: .

Presumably, for the physically meaningful quantities the choice of gauge in their description should be immaterial, in accord with the relativity postulate. Hence, some additional care should be taken in order to ensure that the expressions in various patches glue coherently into well-defined objects that do not actually depend on the local trivialization. The power of PFB manifests in that it allows for a neat balance between the two perspectives above, encoding the ‘geometrically’-given sections in all possible frames, so to say. If acts on as , one adopts the following

Definition II.6.

The space of (-valued) -equivariant maps


There is a natural bijective correspondence:


By the use of (II.6), the field should be regarded as a function on : if is the value of relative to , then is its value w.r.t. transformed frame . The good transformation properties of w.r.t. frame change are generally referred to as covariance, which will be the convention that we stick to. Given the gauge , we can pull-back down to function on . If is another gauge, then , showing how the locally-defined objects relate under the change of gauge. Expressing the transition functions as , the property 3 of Def. II.2 can be readily shown.

ii.1.1 Differential forms and operations with them

In general, for integration and differentiation purposes, the main objects of interest will also include the following

Definition II.7 (Forms).

An exterior vector-valued differential -form on is a smooth map (from Whitney sum) , whose restriction to any fiber is -times multilinear and totally skew-symmetric. If acts linearly via representation , one denotes the space of such forms, satisfying equivariance relation , i.e. for . If , so that , one speaks of just -forms (or invariant, scalar forms). Let denote the space of horizontal forms, which are vanishing on vertical vectors: if any . There is a natural 1-to-1 correspondence:


(That is, invariantly-defined skew-symmetric maps , akin to (II.6). Note: unlike horizontal vectors, the horizontal forms are naturally given, since the vertical vectors are canonically generated by the group action tangent to the fiber, cf. further discussion in Sec. II.3.2.)

The new forms are constructed in three natural ways:

  1. The exterior multiplication (or wedge-product) of and is obtained via anti-symmetrization in the arguments of the combined map between product spaces, namely

  2. The interior product with the given vector field is simply an evaluation , and for 0-forms. It satisfies the derivation property ;

  3. The exterior derivative is given by expression


    such that it reduces to on functions, and can be uniquely characterized by its properties (axiomatic definition, cf. [34, Lemma 1.5.13][48, §4.3]):

    1. (linearity),

    2. (derivation),

    3. (co-boundary).

    The latter property 33 – figuring in the Poincaré’s lemma – generalizes the local integrability condition on 1-form to be a total differential (potentiality of the gradient, leading to the path-independence of the integral , cf. footnote 20).

In the case the space of values is endowed with some multiplication , the sum of (infinite dimensional) vector spaces is made into graded differential algebra with combined multiplication


satisfying .

Definition II.8.

Let be the Lie algebra homomorphism, induced by the group action


Analogously to (II.10), for and , we can then define


which is just the wedge (II.8), followed by the action in the coefficient space. In particular, if in (II.10) is induced by the adjoint representation sending , that is sending , then we write: for (II.12). The graded Jacobi identity: is satisfied.

Classical interpretation.

The scalar forms constitute the natural arguments of multidimensional integration (cf. [48, Ch.4]), when pulled-back down to (via gauge choice). Through their definition and properties, the following familiar operations in the Euclidean space are concisely captured and abstracted: 1) the construction via of the volume-determinants built on (unspecified, or generic) argument vectors; 2) the value of the latter on the concrete infinitesimal elements of the -dim surface is obtained via -substitution; 3) the exterior derivative generalizes the ordinary , and of -vectors.

Moreover, being formulated in the explicitly coordinate-free language, they are naturally applicable to non-Euclidean manifolds, which by definition represent the smooth patchwork glued from the local regions of . Provided some local chart and the ‘natural’ basis of vectors and dual forms are given, the usual coordinate expressions:

are an easy consequence of the intrinsic definitions.

The line element of , however, is only applicable point-wise – by considering each tangent space as Euclidean. It is not until thus-induced scalar product/metric is specified smoothly at each point (a section of the tensor bundle over ), that the interior multiplication could be regarded as the inner product contraction – using the (non-canonical) identification via . We do not hasten to provide such an identification right away, though. The reason is that we wish to stick to the framework where the ‘‘amorphous’’ parameter space (which we tend to identify with domain ) is clearly delineated from the value space of ‘‘objects’’ (with canonically given structure), attached to it 666Accordingly, our notation marks with ordinary typeface, while is bold.. Arguably, this provides the cleanest strategy to isolate physical/geometrical content from possible artefacts due to coordinate choices (‘‘the former precedes the latter’’, so to say), and to ensure the re-parametrization invariance of the results. In fact, the whole method of forms and moving frames was originally developed by Cartan to tackle these subtle issues, in particular.

In the familiar cases of fields on the fixed Galilei/or Minkowski background, there is no such puzzle, for there is a global Cartesian coordinate system valid over the entire and canonically given inner product on forms. This can be used to construct the integrands from the physical quantities of interest (such as magnetic fluxes, or circulation of electric field), roughly speaking, by weighting the projections in the certain directions with the respective -volumes. (For instance, is the typical example, where are components of the electromagnetic stress-energy tensor.) The integration over -surface is then seen as simply ‘summation’ of the well-defined objects (scalars) in the limit of smaller and smaller divisions, connecting the ‘finer’ description to the ‘coarse-grained’ observables which are actually being measured.

The physical laws (e.g., that of Faraday-Maxwell) establish the relations between different such quantities in an invariant manner, accumulating the large number of experimental facts. The bridge to the formulation in partial (1st order) differential equations, relating the numerical values of physical quantities at two infinitesimally close points, is provided by the famous Stokes formula:


Incidentally, that also encapsulates how the new invariants, associated with the higher dimensional elements, may be constructed from the quantities at the boundary (consider the values of integrands in the decrementally small region, so that rather field or surface variations could be disregarded).

ii.1.2 Motivations from QG: discretization, continuum limit

Regularization/cut-off in general.

Arguably, the exact field values could never be measured pointwise (this is not even mentioning the relational character of observations), only to a sufficiently fine precision, like described above. For the practical purposes, one commonly adheres to regularization by introducing some sort of cut-off on physically interesting degrees of freedom. Broadly speaking, this allows to organize computations effectively, reducing the infinitude of field’s characteristics down to the manageable amount (e.g., corresponding to a finite series of measurements). Examples of this include the typical energy of collisions in particle physics, the lattice spacing in the Monte-Carlo simulations, as well as the Regge discretization [49] in gravity. The latter, akin to lattice gauge theory, replaces the smooth spacetime with simplicial complexes, conveying the gravitational d.o.f. (distributional) to the way how the simplexes glue together in a non-trivial manner.

Continuum limit troubles.

The characteristic of the ‘good’ theory requires consistency of the description as one removes the cut-off, by gradually taking into account more and more degrees of freedom. This is referred to as the continuum limit, corresponding to the infinitely fine (in principle) description of the system. The regularization should capture well the essential properties of the system, in particular, its behaviour w.r.t. symmetry transformations. This is currently not the case in the modern attempts to quantization of gravity based on the use of discrete structures for regularization, descendants of the Regge’s triangulations 777LQG is somewhat exceptional, in this regard, whose (kinematical) Hilbert space allows for the continuum limit, although well-behaved only w.r.t. the symmetries of 3d hypersuface. The dynamics is viewed in terms of hypersurface deformations and problematic.. The reason appears to be non-sufficient understanding of the diffeomorphism symmetry and its implementation, resulting in that one of the GR’s main tenets is lost under discretization [26, 27]. What is worse, it is not entirely clear how this should be realized in the first place.

Inspiration via analogies.

It is not hard to notice the similarity of the above narrative to that of surrounding the whole integration process, briefly sketched around (II.13). Indeed, the increments in the principal parts of the integrals utilize some linear approximations, reminiscent of the individual simplicial cells, whose geometry is known and can be used for contraction. The refinement of the subdivision adds more contributions to the sum and the better approximation, ideally. In fact, these analogies inspired (various) geometric discretization techniques, that were exploited, e.g., for putting the Maxwell’s equations on a grid (finite element method, cf. [50]), and in some TQFTs as well [51]. They were also incorporated into several discrete models of QG, mainly based on the topological nature of the latter in 3d.

The primary idea consists of discretizing the domain, s.t. it is represented via the collection of basic building blocks – called ‘cells’ – which topologically are closed -balls, ‘glued together’ into a cell-complex. The details may vary, and depending on the required properties of maps, one distinguishes purely topological CW complexes, piecewise-linear, combinatorial, and simplicial ones. In the piecewise-linear context [8], one usually talks about point as 0-cell, a 1-cell is a closed interval, a 2-cell is a polygon, and so on. Any lower dimensional face , contained in (or equal) is called its ‘face’ and denoted . A piecewise-linear cell-complex is a collection of cells in some such that: 1) if then ; 2) If then . One writes for the (set theoretic) union of cells contained in  888One notices that if the topological domain supports some non-trivial geometric structure, one should be careful that it does not run into conflict with the extra assumption on linearity of -cells, which is basically the content of our following remarks..

The algebraico-topological considerations then usually enter the discretization of objects on the domain as follows. Given the simplicial cell-complex , the singular homology is constructed by considering maps of the standard -simplex to a topological space, and composing them into formal sums. Define the -chains as maps from (combinatorial) -skeleton to the (abstract) abelian group, such that the image for each cell with reversed orientation: . Taking to be the space of all real-valued -chains, due to natural multiplicative structure (that gives the notion of ‘scale’), each element can be seen as the linear combination of unit chains , and – as a linear space over , spanned by the standard basis of elementary cells. Suppose that (the homeomorphic image of) somehow approximates the smooth manifold . Then the -forms are naturally discretized as co-chains via integration over simplices:


The isomorphism of the corresponding cohomologies is induced by this map, according to de-Rham’s theorem. By the Stokes formula (II.13), is the co-boundary operator dual to . In accord with our understanding of integration, the co-chains represent densities w.r.t. measures , imparted to the discretized support by -valued chains.

Problem posing.

As such, the purely combinatorial construction of a cell complex is well-suited to study the topological properties of , respectively, but what about geometry? Part of the geometrical information stripped away is restored by attaching signed -measures (lengths, areas, volumes, etc.) to the cells of , which may be suitable for theories with background metric space. Notwithstanding, we see couple of weak points, hindering the genuine extension of this scheme to gravity.

  1. In the background independent GR, the metric itself is dynamical and subject to variation, hence it should rather enter the ‘density’ part.

  2. The formal summation in makes little-to-no ‘geometric’ sense (or at least obscure for us) prior to integration, since it is only by the field the real values are associated to the (elementary) cells that play the role of ‘labels’. On the contrary, the integration itself is naturally viewed as summation.

  3. At last, the scalar forms are insufficient, since the geometric information contains not only -volumes of (elementary) surfaces, but also their directions (mutual 999To provide some more context: In the gauge invariant languages, using lengths/areas/volumes as discrete (metric) variables, the insufficiency of a subset of them to characterize the configuration of polyhedral cells locally [52], led to the introduction of numerous angle-variables [53, 54, 55]. The latter cannot be arbitrarily assigned but have to satisfy some non-trivial constraints. (The rigid simplex is exceptional, since its shape is uniquely determined by the edge-lengths in every dimension.) In contrast, we are talking about gauge covariant description, local to a region.) in spacetime.

With all the above said, we formulate the task of integration of vector-valued forms, attaching to the domain some sensible characteristic of geometry. One should be careful though, since the result may depend on the choice of frame. All this prompts to take a closer look on variables of the gravitational field in the background-independent GR, to analyse what constitutes its fundamental degrees of freedom in the continuum, keeping eye also on what happens in discrete QG proposals.

Let us first remind in Sec II.2 the very basic concepts from the ordinary (flat) Euclidean/Minkowskian spaces, permitting us to measure lengths and angles by means of a scalar product between two vectors (and, more generally, tensors). Then in Sec. II.3, this view will be ‘localized’ using the framework of Cartan affine connections. This strictly corresponds to the conventional way of thinking about manifolds in terms of plane tangent spaces, that are attached – or ‘soldered’ – to at the point. We return to the task in Sec. III.3, where it will be addressed in the geometric context of gravity theory.

ii.2 Elementary geometry

In this section we recount the basic notions from the geometry of the familiar Euclidean space , with the straightforward generalization to Minkowski spacetime , whose affine structure is carefully highlighted. The focus is on the multilinear algebra, describing plane elements of various dimensionality, where we mostly stick to [43, Ch.1]. The description exploits the existence of globally defined reference frames – including the choice of origin – having in mind their subsequent local and approximate character, upon ‘gauging’ and inclusion of (gravitational) interactions.

ii.2.1 Affine spaces and bases

Let be a point set and an -dimensional (real) vector space regarded as abelian group. Consider the action which is free (i.e. if for some ) and transitive (i.e. for any there is such that ). The triple is then called an affine space, whereas – the principal homogeneous space for the group of translational motions , the stabilizer (sub-)group of every point being trivial . The dimension of an affine space is defined as the dimension of the vector space of its translations.

Translations play a twofold role for . First, in an obvious manner any element defines the rigid motion of space as a whole via point set correspondence (globally defined diffeomorphism)


On the other side, for any fixed and an arbitrarily chosen linear basis in , the bijective correspondence is established via the group action between the points of -space and their parametrization via -elements:


The pair forms a global affine frame in the sense that any point can be written as (II.16) (w.r.t. some arbitrary and , of course), so that the chart provides the globally-defined coordinate system.

In the affine spaces, the ‘‘addition of points’’ is not allowed. One can only pass to new points by adding elements of , or meaningfully take the difference of two points to obtain the arrow-vector that they bound. In this regard, one draws the following distinction between two types of vectors. If the origin is fixed, the vector is called bound, or sliding, e.g. the ‘‘radius-vector’’ of some other point’s position as in (II.16). When only the magnitude and direction of the arrow matter, while the particular initial point is of no importance, then one speaks about free vector. They are normally obtained as the difference of two bound vectors. The typical example of the first is the force, being dependent on the point of application, while the total angular momentum is a free vector – indeed: due to in Newtonian mechanics 101010Usually, in the Euclidean space equipped with a choice of origin, this separation is non-relevant, since a free vector is equivalent to the bound one of the same magnitude and direction whose initial point is the origin..

ii.2.2 Group theoretic properties

On the manifold of all (global) affine frames of the natural right action of the affine group is defined (‘passive view’):


with the multiplication law of the semi-direct product . Here the components of describe the new frame’s position w.r.t. the original system. We also have used the following shorthand notation:


defining the respective projections on the original axes of the new frame’s unit vectors.

The following short sequence of group homomorphisms


is exact in the sense that the kernel of each homomorphism is equal to the image of the preceding one. The map , sending , provides the so-called split of (II.19) in the sense that there is a homomorphism , defined as , projecting to the identity: . One says that the semi-direct product is (non-trivial) extension of by .

In a different vein (‘active view’): whereas the (sub-)group of general linear transformations preserves the linear structure in the sense that , the affine group respects the affine structure of translations in the sense that . The kernel of is the (sub-)group of translations . Using this, can be identified with the stabilizer of the point, and thus the kernel of the canonical projection in the dual sequence:


(Exact, but not homomorphisms of groups, since is the homogeneous space of cosets.) The relation between two perspectives may be seen as follows.

The set is the principal homogeneous space for , on which the latter acts freely and transitively. Hence, for an arbitrarily fixed , there is a bijective correspondence between and . The group – called the principal group of a geometry (on 111111The name of the PFB has its origins in this translation of the German ‘‘hauptgruppe’’ (cf. [34, p.138]). – then acts on itself from the left. Decomposing , the factor-space of (left) cosets , w.r.t. sub-group of homogeneous transformations of the linear part of the frame, may be identified with the point set itself – in accord with Klein’s Erlangen program.

The projection map is a PFB, canonically associated with the arbitrary homogeneous Klein geometry, understood in the sense of a pair of (principal) group and its (closed) sub-group . Had we chosen the different origin, the respective stability sub-groups are related by conjugation , corresponding to different embeddings of in . [This base-point disappears in the infinitesimal description, thus making the choice of origin the same type of ‘‘integration constant’’ as .] The splitting of the sequence (II.20) (i.e. , s.t. ) defines the connection in the direct sense of choosing the ‘‘horizontal’’ lift, typical to the groupoid approach to this concept [56]. Both ‘active’, and ‘passive’ viewpoints on the transformation group are completely equivalent and allow generalization to arbitrary geometries (e.g., non-spacetime reductive, s.a. conformal or projective) [45, 57]; though, we have the tendency to use the language of frames as being more ‘‘visual’’.

It is convenient to represent – and its tangent (free) vectors – via an embedded -hyperplane in , where -subgroup of acts linearly as follows:


Due to parallelizability of , the frames with the -origin fixed may be used, s.t. the points are represented via bound vectors , having components , while the free vectors lie in the hyperplane altogether and transform linearly under -subgroup, correspondingly. Let us now concentrate on the basic notions of metric geometry of (free) vectors and their -dim generalizations.

ii.2.3 Metric properties

The affine geometry studies the properties of figures that remain unaffected under [via (II.21) the latter may be viewed as a subset of projective transformations/homographies leaving the hyperplane at infinity intact]. The notions of addition of vectors and their equality may thus be established, as well as the equivalence between sliding vectors (parallelism). However, the (ratio of) lengths of two vectors is available only for those which are parallel. Indeed, the arbitrary choice of non-collinear axes [e.g. pointing to some distant stars] sets up the most general system of Cartesian coordinates: the projections of the vector on each of the axes are made by the hyperplane parallel to the linear span of , the magnitudes being measured in terms of the individual units (rods and clocks) on each of the basis vectors of the oblique frame.

The particular unit of length, considered as absolute, may be set up through the following

Definition II.9.

The metric on is the symmetric non-degenerate -valued bilinear form


Introducing the coefficients w.r.t. arbitrary basis , it may be written as . This is used to compute the squared magnitude of the vector’s length and the scalar product:


the latter may be obtained as the -prefactor in the expansion of . The (planar) angle between two vectors is then deduced to be


The passage from the oblique axes to the orthonormal basis can always be performed, in which the g-matrix is diagonalized and have the entries


The number of positive and negative eigenvalues is preserved (Sylvester’s law) and called the signature. We are mainly interested in the positive-definite metric of the Euclidean space , and that of the Minkowski spacetime with -signature (in 121212An intriguing and phenomenologically viable option could be the principal group of the de Sitter space , with ‘translations’ replaced by non-commutative -module [akin to boosts ]. This allows to accomodate for a positive cosmological constant/fundamental length scale within the modified kinematics of special relativity [58, 59, 60].. Denoting , in general, the requirement of the preservation of the scalar product then restricts the sub-group of homogeneous transformations to the respective orthogonal group (or special-orthogonal/proper-orthochronous, depending on allowed discrete symmetries). In the latter case, the angle in (II.24) should be replaced by hyperbolic for the timelike vectors, i.e. the measure of boost in terms of additive rapidity.

The usual convention introduces the covariant components of the vector , s.t. the scalar product may be concisely written as , and . They are inversely related to their contravariant counterparts by means of , constructed from the (transpose) matrix of cofactors in the determinant expansion along any row or column . Although two types of components coincide (up to sign) in the orthonormal basis and often used interchangeably, they encode the different information, and the distinction is drawn in the general case: the first give the usual projections on axes , whereas the other projections are made orthogonally using the scalar product -hyperplane.

ii.2.4 (Simple) bivectors and multivectors

The (free) vector is seen as the directed line segment of certain length, i.e. it is characterized 1) by its measure, 2) direction, and 3) orientation in the sense of a certain order of its endpoints (their precise position is non-important for free vectors). This may be generalized to a plane elements of higher dimensionality, leading to the construction of a ‘geometric algebra’. Let us examine it in some greater detail using the instructive case of a 2-plane, following [43, Ch.1]. Being the first non-trivial dimension where the curvature reveals itself, it is representative of a more general situation. (Also, the bivectors constitute one of the basic constructive elements in Spin Foams and LQG theories.)

Definition II.10.

A (simple) bivector is defined to be the configuration formed by two vectors and , arranged in a certain order. To clarify its geometric meaning one formulates the conditions when two bivectors are considered to be equivalent, or equal. Namely, the two parallelograms built on the pairs of vectors as their edges should

  1. lie in the same (or parallel) plane,

  2. possess the same area, and

  3. the same orientation (circulation in the boundary).

How these conditions, defining the bivector, determine its coordinate representation? Consider the operation of taking the wedge product, or exterior multiplication of vectors 131313The definition of (II.8) represents the dual operation on linear functionals, applied pointwise to tangent vector spaces of an arbitrary manifold. One is straightforwardly related to the other through the concept of soldering., that is

  1. linear ;

  2. associative ;

  3. alternating