QUANTUM GRAVITY AND THE

FUNCTIONALRE NORMALIZATION GROUP

The Road towards Asymptotic Safety

During the past two decades the gravitational Asymptotic Safety scenario has
undergone a major transition from an exotic possibility to a serious contender
as a realistic theory of quantum gravity. It aims at a mathematically consis-
tent quantum description of the gravitational interaction and the geometry
of spacetime within the realm of quantum field theory, keeping its predictive
power at the highest energies. This volume provides a self-contained pedagogical
introduction to Asymptotic Safety and introduces the functional renormalization
group techniques used in its investigation, along with the requisite computational
techniques. The foundational chapters are followed by an accessible summary of
the results obtained thus far. It is the first detailed exposition of Asymptotic
Safety, providing a unique introduction to quantum gravity. The text assumes
no previous familiarity with the renormalization group and thus serves as an
important resource for both practicing researchers and graduate students enter-
ing this maturing field.

M a r t i n R e u t e r is a professor of Theoretical Physics at the Johannes

Gutenberg University Mainz. He previously worked at the European Labora-
tory for Particle Physics, CERN, in Geneva, the DESY laboratory in Hamburg,
and the Leibniz University of Hanover. In the 1990s, he initiated the exploration
of quantum gravity and the Asymptotic Safety scenario using the functional
renormalization group and subsequently played a key role in developing the
program into its present form.

F r a n k S a u e r e s s i g is an assistant professor at the Radboud University

Nijmegen. Before his appointment, he held research positions at Utrecht Univer-
sity, the Institut de Physique Théorique at CEA/Saclay, and an Emmy Noether
fellowship at the Johannes Gutenberg University Mainz. With more than 70
scientific publications in the field of quantum gravity he is among the leading
young talents in the field.
 CAMBRIDGE MONOGRAPHS ON MATHEMATICAL PHYSICS

S. J. Aarseth Gravitational N-Body Simulations: Tools and Algorithms †

J. Ambjørn, B. Durhuus and T. Jonsson Quantum Geometry: A Statistical Field Theory Approach †
A. M. Anile Relativistic Fluids and Magneto-fluids: With Applications in Astrophysics and
Plasma Physics
J. A. de Azcárraga and J. M. Izquierdo Lie Groups, Lie Algebras, Cohomology and Some
Applications in Physics †
O. Babelon, D. Bernard and M. Talon Introduction to Classical Integrable Systems †
F. Bastianelli and P. van Nieuwenhuizen Path Integrals and Anomalies in Curved Space †
D. Baumann and L. McAllister Inflation and String Theory
V. Belinski and M. Henneaux The Cosmological Singularity †
V. Belinski and E. Verdaguer Gravitational Solitons †
J. Bernstein Kinetic Theory in the Expanding Universe †
G. F. Bertsch and R. A. Broglia Oscillations in Finite Quantum Systems †
N. D. Birrell and P. C. W. Davies Quantum Fields in Curved Space †
K. Bolejko, A. Krasiński, C. Hellaby and M-N. Célérier Structures in the Universe by Exact
Methods: Formation, Evolution, Interactions
D. M. Brink Semi-Classical Methods for Nucleus-Nucleus Scattering †
M. Burgess Classical Covariant Fields †
E. A. Calzetta and B.-L. B. Hu Nonequilibrium Quantum Field Theory
S. Carlip Quantum Gravity in 2+1 Dimensions †
P. Cartier and C. DeWitt-Morette Functional Integration: Action and Symmetries †
J. C. Collins Renormalization: An Introduction to Renormalization, the Renormalization Group
and the Operator-Product Expansion †
P. D. B. Collins An Introduction to Regge Theory and High Energy Physics †
M. Creutz Quarks, Gluons and Lattices †
P. D. D’Eath Supersymmetric Quantum Cosmology †
J. Dereziński and C. Gérard Mathematics of Quantization and Quantum Fields
F. de Felice and D. Bini Classical Measurements in Curved Space-Times
F. de Felice and C. J. S Clarke Relativity on Curved Manifolds †
B. DeWitt Supermanifolds, 2nd edition †
P. G. O. Freund Introduction to Supersymmetry †
F. G. Friedlander The Wave Equation on a Curved Space-Time †
J. L. Friedman and N. Stergioulas Rotating Relativistic Stars
Y. Frishman and J. Sonnenschein Non-Perturbative Field Theory: From Two Dimensional
Conformal Field Theory to QCD in Four Dimensions
J. A. Fuchs Affine Lie Algebras and Quantum Groups: An Introduction, with Applications in
Conformal Field Theory †
J. Fuchs and C. Schweigert Symmetries, Lie Algebras and Representations: A Graduate Course
for Physicists †
Y. Fujii and K. Maeda The Scalar-Tensor Theory of Gravitation †
J. A. H. Futterman, F. A. Handler, R. A. Matzner Scattering from Black Holes †
A. S. Galperin, E. A. Ivanov, V. I. Ogievetsky and E. S. Sokatchev Harmonic Superspace †
R. Gambini and J. Pullin Loops, Knots, Gauge Theories and Quantum Gravity †
T. Gannon Moonshine beyond the Monster: The Bridge Connecting Algebra, Modular Forms and
Physics †
A. Garcı́a-Dı́az Exact Solutions in Three-Dimensional Gravity
M. Göckeler and T. Schücker Differential Geometry, Gauge Theories and Gravity †
C. Gómez, M. Ruiz-Altaba and G. Sierra Quantum Groups in Two-Dimensional Physics †
M. B. Green, J. H. Schwarz and E. Witten Superstring Theory Volume 1: Introduction
M. B. Green, J. H. Schwarz and E. Witten Superstring Theory Volume 2: Loop Amplitudes,
Anomalies and Phenomenology
V. N. Gribov The Theory of Complex Angular Momenta: Gribov Lectures on Theoretical Physics †
J. B. Griffiths and J. Podolský Exact Space-Times in Einstein’s General Relativity †
T. Harko and F. Lobo Extensions of f(R) Gravity: Curvature-Matter Couplings and Hybrid
Metric-Palatini Gravity
S. W. Hawking and G. F. R. Ellis The Large Scale Structure of Space-Time †
F. Iachello and A. Arima The Interacting Boson Model †
F. Iachello and P. van Isacker The Interacting Boson-Fermion Model †
C. Itzykson and J. M. Drouffe Statistical Field Theory Volume 1: From Brownian Motion to
Renormalization and Lattice Gauge Theory †
C. Itzykson and J. M. Drouffe Statistical Field Theory Volume 2: Strong Coupling, Monte Carlo
Methods, Conformal Field Theory and Random Systems †
G. Jaroszkiewicz Principles of Discrete Time Mechanics
G. Jaroszkiewicz Quantized Detector Networks
C. V. Johnson D-Branes †
P. S. Joshi Gravitational Collapse and Spacetime Singularities †
J. I. Kapusta and C. Gale Finite-Temperature Field Theory: Principles and Applications, 2nd
edition †
V. E. Korepin, N. M. Bogoliubov and A. G. Izergin Quantum Inverse Scattering Method and
Correlation Functions †
J. Kroon Conformal Methods in General Relativity
M. Le Bellac Thermal Field Theory †
Y. Makeenko Methods of Contemporary Gauge Theory †
S. Mallik and S. Sarkar Hadrons at Finite Temperature
A. Malyarenko and M. Ostoja-Starzewski Tensor-Valued Random Fields for Continuum Physics
N. Manton and P. Sutcliffe Topological Solitons †
N. H. March Liquid Metals: Concepts and Theory †
I. Montvay and G. Münster Quantum Fields on a Lattice †
P. Nath Supersymmetry, Supergravity, and Unification
L. O’Raifeartaigh Group Structure of Gauge Theories †
T. Ortı́n Gravity and Strings, 2nd edition
A. M. Ozorio de Almeida Hamiltonian Systems: Chaos and Quantization †
M. Paranjape The Theory and Applications of Instanton Calculations
L. Parker and D. Toms Quantum Field Theory in Curved Spacetime: Quantized Fields and Gravity
R. Penrose and W. Rindler Spinors and Space-Time Volume 1: Two-Spinor Calculus and
Relativistic Fields †
R. Penrose and W. Rindler Spinors and Space-Time Volume 2: Spinor and Twistor Methods in
Space-Time Geometry †
S. Pokorski Gauge Field Theories, 2nd edition †
J. Polchinski String Theory Volume 1: An Introduction to the Bosonic String †
J. Polchinski String Theory Volume 2: Superstring Theory and Beyond †
J. C. Polkinghorne Models of High Energy Processes †
V. N. Popov Functional Integrals and Collective Excitations †
L. V. Prokhorov and S. V. Shabanov Hamiltonian Mechanics of Gauge Systems
S. Raychaudhuri and K. Sridhar Particle Physics of Brane Worlds and Extra Dimensions
A. Recknagel and V. Schiomerus Boundary Conformal Field Theory and the Worldsheet Approach
to D-Branes
M. Reuter and F. Saueressig Quantum Gravity and the Functional Renormalization Group
R. J. Rivers Path Integral Methods in Quantum Field Theory †
R. G. Roberts The Structure of the Proton: Deep Inelastic Scattering †
C. Rovelli Quantum Gravity †
W. C. Saslaw Gravitational Physics of Stellar and Galactic Systems †
R. N. Sen Causality, Measurement Theory and the Differentiable Structure of Space-Time
M. Shifman and A. Yung Supersymmetric Solitons
Y. M. Shnir Topological and Non-Topological Solitons in Scalar Field Theories
H. Stephani, D. Kramer, M. MacCallum, C. Hoenselaers and E. Herlt Exact Solutions of Einstein’s
Field Equations, 2nd edition †
J. Stewart Advanced General Relativity †
J. C. Taylor Gauge Theories of Weak Interactions †
T. Thiemann Modern Canonical Quantum General Relativity †
D. J. Toms The Schwinger Action Principle and Effective Action †
A. Vilenkin and E. P. S. Shellard Cosmic Strings and Other Topological Defects †
R. S. Ward and R. O. Wells, Jr Twistor Geometry and Field Theory †
E. J. Weinberg Classical Solutions in Quantum Field Theory: Solitons and Instantons in High
Energy Physics
J. R. Wilson and G. J. Mathews Relativistic Numerical Hydrodynamics †

†
Available in paperback
 Quantum Gravity and the
Functional Renormalization Group
The Road towards Asymptotic Safety

MARTIN REUTER
Johannes Gutenberg Universität Mainz

FRANK SAUERESSIG
Radboud Universiteit Nijmegen
 University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India
79 Anson Road, #06-04/06, Singapore 079906

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107107328
DOI: 10.1017/9781316227596
⃝
c Martin Reuter and Frank Saueressig 2019
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2019
Printed and bound in Great Britain by Clays Ltd, Elcograf S.p.A.
A catalogue record for this publication is available from the British Library.
ISBN 978-1-107-10732-8 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
 Contents

Preface page xi
List of Abbreviations xiii

1 A Quantum Field Theory of Gravity 1

1.1 Renormalizing the Unrenormalizable 1
1.2 Background Independence 3
1.3 All Backgrounds Is No Background 6
1.4 Asymptotic Safety in a Nutshell 8
1.5 Continuum or “Atoms of Spacetime”? 17

2 The Functional Renormalization Group 19

2.1 The Concept of the Effective Average Action 19
2.2 Theory Space and Its Truncation 28
2.2.1 The Space of Actions 28
2.2.2 The FRGE in Component Form 29
2.2.3 Truncations of Theory Space 31
2.2.4 The Local Potential Approximation 35
2.2.5 Adjusted Cutoffs 45
2.3 Decoupling 45
2.4 Background Fields 48

3 The Asymptotic Safety Mechanism 54

3.1 Geometry and RG Dynamics on Theory Space 54
3.1.1 The Beta Functional 55
3.1.2 From the Beta Functional to the Beta Functions 56
3.1.3 The Cutoff Scale as a Unit 56
3.1.4 The Geometric Interpretation 61
3.2 Fixed Points 66
3.2.1 The Zeros of β 66
3.2.2 The Linearized Flow 67
3.2.3 Universality of the Critical Exponents 69
3.2.4 The Ultraviolet Critical Hypersurface 70
3.2.5 Gaussian vs. non-Gaussian Fixed Points 71
3.2.6 Linearization about the Scalar GFP 71
viii Contents

3.3 Asymptotic Safety: The Basic Idea 74

3.3.1 The Input: T and β 75
3.3.2 The Quest for Complete Trajectories 76
3.3.3 The “Asymptotic Safety Scenario” of the UV Limit 76
3.3.4 Predictive Power 78
3.3.5 Generalizations and Refinements 81
3.4 A First Example: Gravity in 2 + ε Dimensions 82

4 A Functional Renormalization Group for Gravity 85

4.1 What is “Metric” Quantum Gravity? 86
4.2 What is “Coarse Graining” in Gravity? 89
4.2.1 Background Independence 90
4.2.2 Gauge Invariance 91
4.2.3 Toward a Notion of Coarse Graining 94
4.2.4 The Background Field Method 103
4.3 Introducing an EAA for Gravity 110
4.3.1 The Gauge-Fixed Functional Integral 110
4.3.2 Classical BRST Invariance 112
4.3.3 Background-Type Gauge-Fixing Conditions 112
4.3.4 Generalized Harmonic Gauges 113
4.3.5 The Mode Suppression Term 114
4.3.6 Sources and Expectation Values 115
4.3.7 Formal Definition of the Gravitational EAA 117
4.4 Properties of the Gravitational EAA 118

5 Truncations of Single-Metric Type 127

5.1 General Classes of Truncations 128
5.1.1 Freezing the Ghost Sector 128
5.1.2 A Class of Bi-Metric Truncations 129
5.1.3 The Class of Single-Metric Truncations 130
5.2 The Einstein–Hilbert Truncation 132
5.2.1 Derivation of the β-Functions for g and λ 133
5.2.2 Structure of the β-Functions for g and λ 139
5.3 Properties of the Einstein–Hilbert Flow 143
5.3.1 The Subtruncation of Vanishing λ 144
5.3.2 The Subtruncation of Constant G 148
5.3.3 Gaussian and non-Gaussian Fixed Points 150
5.3.4 The Phase Portrait 155
5.3.5 Close Relatives of the EH Truncation 158
5.3.6 Testing the Reliability of Truncations 161
5.3.7 Evidence for Asymptotic Safety 164
 Contents ix

6 Bi-Metric Truncations 175

6.1 Level Expansion of Γgrav
k 175
6.2 Level One and the Tadpole Equation 176
6.3 A Bi-Metric Einstein–Hilbert Ansatz 178
6.4 The Plethora of Couplings 179
6.5 Results from the Bi-Metric EH Truncation 181
6.5.1 The 2D Flow on the “Dyn” Subspace 182
6.5.2 Screening vs. Antiscreening 183
6.5.3 Flow on the 4D Parameter Space 184
6.5.4 Imposing Split Symmetry 184
6.5.5 Discussion 185

7 Conformally Reduced Gravity 187

7.1 From Gravity to ϕ4 Theory 188
7.2 The Untruncated Conformally Reduced Theory 190
7.3 Coarse Graining Operators: Gravity vs. Matter 192
7.4 The Reduced Einstein–Hilbert Truncation 194
7.5 An Infinite-Dimensional Truncation 198
7.6 Bi-Metric Truncations and the Split Ward Identity 204

8 The Reconstruction Problem 207

8.1 In Search of a Bare Theory 207
8.2 The EAA in the Presence of a UV Cutoff 210
8.3 A One-Loop Reconstruction Formula 212
8.4 The Twofold Einstein–Hilbert Truncation 215
8.5 Further Remarks 218

9 Alternative Field Variables 220

9.1 Quantum Einstein–Cartan Gravity 220
9.2 Implementations of Gauge Invariance 223
9.3 Truncations of Hilbert–Palatini Type 230
9.4 Quantum Arnowitt–Deser–Misner Gravity 234
9.5 Quantum ADM Gravity in the Einstein–Hilbert Truncation 238
9.6 Observables in Quantum ADM Gravity 242

10 Matter Coupled to Quantum Gravity 245

10.1 Free Matter Fields in the Einstein–Hilbert Truncation 245
10.2 Interacting Gravity-Matter Fixed Points 251
10.2.1 QED on Minkowski Space 251
10.2.2 QED Coupled to QEG 252
10.3 Two non-trivial Fixed Points 254
10.4 Approximate Analytical Solution 255
x Contents

10.5 Asymptotic Safety Construction at NGFP2 258

10.6 Numerical Solution 259
10.7 Can We Compute the Fine-Structure Constant? 260

11 Towards Phenomenology 263

11.1 Scale-Dependent Riemannian Structure 263
11.2 Cutoff Identifications 264
11.3 Structure of “Quantum Spacetime” 266
11.4 Black Holes 271
11.5 Cosmology 276

12 Miscellanea 281

Appendix A: Notation and Conventions 293

Appendix B: Organizing the Derivative Expansion 295

B.1 An Explicit Basis 295
B.2 Prototypical Background Spaces 296

Appendix C: Metric Variations 300

Appendix D: Heat Kernel Techniques 301

D.1 Early Time Expansion 302
D.2 The Off-Diagonal Heat Kernel 304
D.3 Integral Transforms 307

Appendix E: Cutoff- and Threshold Functions 311

E.1 Exponentially Decaying Cutoffs 312
E.2 The Sharp Cutoff 312
E.3 The Optimized Cutoff 315
E.4 Regulators for Complex Frequencies 316
E.5 Mass-Type Cutoff 316

Appendix F: Field Decompositions 317

References 319
Index 339
 Preface

This book grew out of a series of lectures the authors have given at various
universities and summer schools. Its intention is to provide an easily accessible,
pedagogical account of the basic conceptual ideas and methods underlying the
asymptotic safety approach to quantum gravity. Knowledge of General Relativ-
ity and quantum field theory at a master-course level should suffice to follow
the exposition. The necessary technical background is developed from the begin-
ning and no previous familiarity with functional renormalization group methods
is assumed. As much as possible we try to supplement formal derivations by
intuitive arguments. Our hope is that the book provides a valuable resource
for graduate students and researchers, enabling them to follow the cutting-edge
research in this field and placing it into the broad context.
It is impossible to thank all people here who directly or indirectly contributed
to our work on quantum gravity and to this book ultimately. However, M. R.
is particularly grateful to Christof Wetterich for the crucial and highly inspiring
collaborations during the early days of the functional renormalization group and
more than three decades of creative exchange and to Roberto Percacci for sharing
the dream about asymptotic safety right from the start and for his decisive work
toward making it a reality. It is also a pleasure to thank Alfio Bonanno for
an enjoyable collaboration from very early on in an effort to understand the
phenomenological consequences of asymptotic safety. Special heartfelt thanks go
to Ennio Gozzi and Walter Dittrich for their support and guidance across all of
physics, and well beyond.
F. S. thanks Jan Ambjørn, Renate Loll, and the quantum gravity group at
Nijmegen for the continuous exchange of ideas and countless discussions, which
have significantly influenced his view on quantum gravity in the last few years.
A loving thank you goes to his wife Tetyana and daughter Sophia: without their
continuous support and patience this exposition would not have been the same.
Much of the work reported in this book would not have seen the light of the
day without the enthusiasm and ingenuity of the students and young researchers
who collaborated with us on the asymptotic safety program: N. Alkofer, D.
Becker, D. Benedetti, J. Biemans, A. Codello, A. Contillo, J.-E. Daum, M. Dem-
mel, G. D’Odorico, J.-W. Goossens, K. Groh, U. Harst, W. B. Houthoff, A.
Kurov, O. Lauscher, P. Machado, E. Manrique, A. Nink, C. Pagani, A. Platania,
xii Preface

S. Rechenberger, G. M. Schollmeyer, M. Schutten, J.-M. Schwindt, E. Tuiran, F.

Versteegen, H. Weyer, and O. Zanusso.
We are particularly indebted to Gregor M. Schollmeyer for his invaluable input
and untiring care in typesetting, shaping, and improving this monograph. We are
also grateful to Peter van Dongen for suggesting numerous improvements, and
Robin Ooijer and Arthur Vereijken for their help in proofreading the (almost)
final manuscript.
 List of Abbreviations

ADM Arnowitt–Deser–Misner
BRST Becchi–Rouet–Stora–Tyutin
CDT Causal Dynamical Triangulation
CREH conformally reduced Einstein–Hilbert
EAA Effective Average Action
EH Einstein-Hilbert
FRG Functional Renormalization Group
FRGE Functional Renormalization Group Equation
GFP Gaussian fixed point
GR General Relativity
IR infrared
LHS left-hand side
LPA local potential approximation
LQG Loop Quantum Gravity
LSZ Lehmann–Symanzik–Zimmermann
NGFP non-Gaussian fixed point
ODE ordinary differential equation
QCD Quantum Chromodynamics
QED Quantum Electrodynamics
QEG Quantum Einstein Gravity
RG renormalization group
RHS right-hand side
UV ultraviolet
 1
A Quantum Field Theory of Gravity

Today we know of four kinds of fundamental interactions which seem to underlie

all elementary processes observed in nature. Three of them, the electromagnetic,
the weak, and the strong interactions, are combined in the standard model of
elementary particle physics, which has received striking experimental confirma-
tion during the past decades. Regarded as a classical field theory, the model
employs geometrically natural and mathematically well-understood structures
such as connections of the Yang–Mills type, for example. Furthermore, being
renormalizable in perturbation theory, we also know how the model can be ele-
vated to the level of a perturbatively defined quantum field theory. Beyond this
stage there are ongoing efforts directed toward a non-perturbative definition and
evaluation of at least certain sectors of the theory. Here, modern concepts of sta-
tistical field theory have proven invaluable. They explain, for instance, how the
renormalization properties of the original continuum theory are related to the
behavior of appropriate statistical mechanics models on spacetime lattices when
they approach a second-order phase transition. These insights opened the door
for employing Monte-Carlo simulations as a non-perturbative tool in quantum
field theory, and in particular, as a device to test for the “existence” of a theory
beyond the realm of perturbation theory.

1.1 Renormalizing the Unrenormalizable

As for our theoretical understanding of the fourth of the fundamental interac-
tions, gravity, the situation is markedly different from the other three forces of
nature. With Einstein’s general theory of relativity we have a classical field the-
ory at our disposal which is spectacularly successful in explaining gravitational
phenomena on scales that span many orders of magnitude. However, when we
try to quantize General Relativity (GR) along the same lines as the standard
model, we find that this road is blocked since the theory is non-renormalizable
within perturbation theory [1, 2]. At higher orders of the loop expansion the
2 A Quantum Field Theory of Gravity

calculations must cope with an increasing number of new types of divergences

and they all must be absorbed by suitable counter terms. The finite parts of
their coefficients are left undetermined by the theory itself and so they must be
taken from experiment. While this does not exclude the possibility of comput-
ing quantum corrections to predictions of the classical theory, it implies that
those corrections involve an increasing number of undetermined parameters as
the perturbation order is increased.
As long as one restricts attention to a regime where only a few such new cou-
pling constants play a role, quantized General Relativity (GR) has the status of
an effective quantum field theory [3–5]. Similar to chiral perturbation theory [6]
it makes unambiguous predictions for certain leading quantum corrections. With
increasing energy increasingly high loop orders must be included and hence the
predictions unavoidably involve a growing number of undetermined parameters.
At this stage the theory gradually loses its predictive power, and ultimately it
may break down completely. Increasing the order of perturbation theory beyond
this point, would then have the paradoxical consequence of diminishing the the-
ory’s net predictive power as the hoped-for better precision is more than offset
by the new undetermined parameters it introduces.
This loss of predictivity at high-energy or short-distance scales is a strong
motivation to search for a fundamental quantum theory of gravity, i.e., a theory
that is predictive on all scales and that admits potentially large quantum effects.
Ideally this hypothetical theory would contain only a few free parameters whose
values are not fixed by the theory itself. Similar to the familiar situation in
perturbatively renormalizable models it would express all predictions as well-
defined, computable functions of those few measured parameters.
Given that GR is not renormalizable in standard perturbation theory, it has
commonly been argued that a satisfactory microscopic theory of the gravita-
tional interaction cannot be set up within the realm of quantum field theory, at
least not without adding further symmetries, extra dimensions, or new principles
such as holography, for instance. By contrast, the Asymptotic Safety program
retains quantum field theory without such additions to the theoretical arena.
Instead, it abandons the traditional techniques of perturbation theory, the con-
cepts of perturbative renormalization, and of perturbative renormalizability in
particular.
The Asymptotic Safety approach is based on the generalized notion of renor-
malization shaped by Kadanoff and Wilson [7–9] and the use of a “functional”
or “exact” renormalization group (RG) equation. Hence, concepts from modern
statistical field theory play a pivotal role. They provide a unified framework for
approaching the problems with both continuum and discrete methods.
In the new setting one can conceive of non-perturbatively renormalizable
quantum field theories, i.e., models free from physically harmful divergences
that remain predictive up to the highest energies even though they are
non-renormalizable in the perturbative sense.
 1.2 Background Independence 3

The idea that there might exist a non-perturbatively renormalizable, or as he

called it, “asymptotically safe” quantum theory of gravity was first proposed
by S. Weinberg in the late 1970s [10]. At that time no efficient tools to test this
scenario were available. However, an epsilon-expansion valid near two dimensions
indicated that the new path could indeed be viable, at least for an unphysical
dimensionality of spacetime. Further encouragement came from certain matter
field theories that likewise were not renormalizable within perturbation theory
but could be shown to be non-perturbatively renormalizable. In a paper entitled
“Renormalizing the Non-renormalizable,” Gawedzki and Kupiainen [11] used a
1/N -expansion to prove the non-perturbative renormalizability of the Gross–
Neveu model in three dimensions [12, 13].
The systematic exploration of the Asymptotic Safety scenario in four dimen-
sions began only in the 1990s when powerful functional renormalization group
methods became available for the gravitational field [14]. The exposition of
these non-perturbative methods and their use in scrutinizing the viability of
the Asymptotic Safety route to a fundamental quantum field theory of gravity
is the main topic of this book.

1.2 Background Independence

While the various approaches trying to unify the principles of quantum mechan-
ics and General Relativity are based upon rather different physical ideas and are
formulated in correspondingly different mathematical frameworks,1 they all must
cope with the problem of Background Independence in one way or another. What-
ever the ultimate theory of quantum gravity, a central requirement we impose on
it is that it should be Background Independent in the same sense as GR. Loosely
speaking, this means the spacetime structure that is actually realized in nature
should not be part of the theory’s definition but rather arise as a solution to
certain dynamical equations [18–20].
In classical General Relativity the spacetime structure is encoded in a
Lorentzian metric on a smooth manifold, and this metric, via Einstein’s equation,
is a dynamical consequence of the matter present in the universe. In quantum
gravity we would like to retain the fundamental idea that “matter tells space how
to curve, and space tells matter how to move,” but describe both “matter” and
“space” quantum mechanically. While, today, it is fairly well understood how to
set up quantum field theories of matter systems, the open key problem is the
quantum mechanical description of “space.”
In this book we will mostly explore the possibility of constructing a quantum
field theory of gravity in which the spacetime metric carries the dynamical degrees
of freedom which we associate to “space.” Even though this property is taken

1 For reviews of the attempts to obtain a quantum theory of gravitation see, for example,
[15–17].
4 A Quantum Field Theory of Gravity

from GR, the fundamental dynamics of those metric degrees of freedom is allowed
to be different from that in the classical theory.
The quantum theory of gravity we are searching for will be required to respect
the following principle of Background Independence: In the formulation of the
theory no special metric should play any distinguished role. The actual metric of
spacetime should arise as the expectation value of the quantum field (operator)
gbµν with respect to some state: gµν = ⟨b gµν ⟩.
This requirement is in sharp contradistinction to the traditional setting of
quantum field theory of matter systems on Minkowski space whose conceptual
foundations heavily rely on the availability of a non-dynamical (rigid) Minkowski
spacetime as a background structure.
The principle of Background Independence2 can be rephrased more precisely
as follows. We require that none of the theory’s basic rules and assumptions,
and none of its predictions, therefore, may depend on any special metric that has
been fixed a priori. All metrics of physical relevance must result from the intrinsic
quantum gravitational dynamics.
A possible objection against this working definition [21] could be as follows: A
theory can be made “Background Independent” in the above sense, but neverthe-
less has a distinguished rigid background if the latter arises as the unique solution
to some field equation which is made part of the “basic rules.” For instance, rather
than introducing a Minkowski background directly one instead imposes the field
equation Rµνρσ = 0. However, this objection can apply only in a setting where
the dynamics, the field equations, can be chosen freely. In asymptotically safe
gravity this is impossible since, as we shall see, the dynamical laws are dictated
by the fixed-point action. They are thus a prediction rather than an input.
If we try to set up a continuum quantum field theory for the metric itself,
even assuming we are given some plausible candidate for a microscopic dynam-
ics, described by, say, a diffeomorphism invariant bare action functional S, then
already well before one encounters the notorious problems related to the UV
divergences, profound conceptual problems arise. Just to name one, in absence
of a rigid background when the metric is dynamical, there is no preferred time
direction, for instance, hence no notion of equal time commutators, and clearly
the usual rules of quantization cannot be applied straightforwardly.
Many more problems arise when one tries to apply the familiar concepts and
calculational methods of quantum field theory to the metric itself without intro-
ducing a rigid background structure. Some of them are conceptually deep while
others are of a more technical nature.
The problems are particularly severe if one demands that the sought-for
theory can also describe potential phases of gravity in which ⟨b gµν ⟩ is degenerate

2 Here, and in the following, we write “Background Independence” with capital letters when
we refer to this principle rather than simply to the independence of some quantity with
respect to the background field.
 1.2 Background Independence 5

(non-invertible) or completely vanishing in the most extreme case. Interpreting

⟨b
gµν ⟩ as an order parameter analogous to the magnetization in a magnetic system,
a non-degenerate classical metric ⟨b gµν ⟩ would signal a spontaneous breaking of
diffeomorphism invariance that leaves only the stability group of ⟨b gµν ⟩ unbroken,
i.e., the Poincaré group for example, when ⟨b gµν ⟩ is given by the Minkowski met-
ric. Conversely, ⟨b gµν ⟩ ≡ 0 would then be the hallmark of a phase with completely
unbroken diffeomorphism invariance.
The analogy to magnetic systems suggests that this “unbroken phase” is much
easier to deal with than those with ⟨b gµν ⟩ ̸= 0. However, in practice this is not the
case, and again the reason is that the traditional toolbox of quantum field theory
as shaped by the requirements of particle or condensed matter physics has very
little to ∫offer as soon as gµν vanishes.
∫ √ µν The familiar actions for matter fields such
√ µν
as, say, gg Dµ ϕDν ϕ or gg g αβ Fµα Fνβ , can no longer be written down
since they require an invertible gµν , and problems of this kind are clearly only
the tip of the iceberg.
A similar difficulty shows up when one tries to conceive an appropriate notion
of a “functional renormalization group” in the realm of quantum gravity. In stan-
dard field theory on a rigid background spacetime typical regularization schemes
(by higher derivative regulators, for example) which are used to make the cal-
culations well defined both in the infrared (IR) and the ultraviolet (UV) make
essential use of the metric provided by this background spacetime. As a result, it
is not obvious whether and how such schemes can carry over to quantum gravity.
This problem is particularly acute for non-perturbative approaches employ-
ing any kind of functional renormalization group equation (FRGE) that would
implement a Wilson-like or “exact” renormalization group flow by a repeated
coarse graining [22–35].
In conventional Euclidean field theory as it is employed in statistical mechanics,
for instance, every such coarse-graining step comes equipped with an associated
length scale. In the case of, say, block-spin transformations it measures the size
of the spacetime blocks within which the microscopic degrees of freedom were
averaged. But when the metric is dynamical and no rigid background is available,
this concept becomes highly problematic since it is not clear in terms of which
metric one should measure the physical, i.e. proper extension of a given spacetime
block.
From a continuum viewpoint, in one way or another all techniques of functional
renormalization involve a mode decomposition of the (field) configurations that
are summed or integrated over in the partition function or functional integral.
In the standard case the modes are often taken to be plane waves, characterized
by a momentum vector pµ . They should be thought of as the eigenfunctions of
the Laplacian δ µν ∂µ ∂ν . These modes are grouped into two classes then, namely
long wavelength (or IR) modes, and short-wavelength (or UV) modes, respec-
tively, depending on whether the Euclidean magnitude of their momentum,
(δ µν pµ pν )1/2 , is smaller or bigger than a certain value. The short-wavelength
6 A Quantum Field Theory of Gravity

modes are then “integrated out” and the resulting effective dynamics of the
long-wavelength modes is deduced.
In the absence of an intrinsically given metric comparable to δ µν this proce-
dure fails for (at least) the following obvious reasons: There is neither a natural,
physically motivated way of choosing the basis of field modes, nor is it clear
how to discriminate between IR and UV modes and to fix the order in which
the individual modes belonging to some (ad hoc) basis of field modes should be
integrated out.
Thus, the potential danger one faces in applying the ideas of coarse grain-
ing and RG flows to a continuum formulation of gravity is that in the absence
of a naturally provided metric there is a considerable degree of arbitrariness
in the flow that might ruin the power the exact RG has otherwise. After all,
most of its celebrated successes on both the foundational or conceptual side
(understanding the nature of continuum limits, etc.) and the practical side (com-
puting useful effective descriptions of a given fundamental theory) heavily rely
on the rule “short wavelengths first, long wavelengths second” when it comes to
integrating out degrees of freedom. Trivial as it sounds, nothing the like of it
is available in a manifestly Background Independent continuum formulation of
gravity.

1.3 All Backgrounds Is No Background

There are two quite different strategies for complying with the requirement of
Background Independence:

(i) One can try to define the theory and work out its implications without
ever employing a background metric or a similar non-dynamical structure.
This is the path taken in Loop Quantum Gravity [36–39] and the discrete
approaches to quantum gravity [40–48], for instance, where manifest Back-
ground Independence has dramatic consequences for the structure of the
theory [49]. As we saw, it seems very hard, if not impossible, to realize it in
a continuum field theory, however.3
(ii) One takes advantage of an arbitrary classical background metric ḡµν at the
intermediate steps of the quantization, but verifies at the end that no phys-
ical prediction depends on which metric was chosen. This background field
method is at the heart of the continuum-based gravitational average action
approach [14], which we shall employ in our investigation of asymptotic
safety.

3 Some of the difficulties are reminiscent of those encountered in the quantization of topological
Yang–Mills theories. Even when the classical action can be written down without the need
of a metric, the gauge fixing and quantization of the theory usually requires one. Hence, the
only way of proving the topological character of some result is to show its independence of
the metric chosen.
 1.3 All Backgrounds Is No Background 7

The two strategies have complementary advantages and disadvantages. Follow-

ing the path (i), Background Independence is implemented strictly. Hence, it is
manifest at all intermediate steps of the constructions and calculations, but one
must then cope with the above profound difficulties. Taking the path (ii) instead,
Background Independence is not manifest during the intermediate steps and
requires an effort to reestablish it at the end. This strategy has the invaluable
benefit that basically the entire arsenal of general concepts and technical tools
of conventional background-dependent quantum field theory is applicable.
In the simplest variant of the background field method one parameterizes the
quantum metric as gbµν = ḡµν + b hµν or a non-linear generalization thereof, and
then quantizes the fluctuation b hµν in essentially the same way one would quantize
a matter field in a classical spacetime with metric ḡµν . In this way, all of the
conceptual problems alluded to above, in particular the difficulties related to the
construction of regulators, disappear.
Technically the quantization of gravity proceeds then almost as in standard
field theory on a rigid classical spacetime, with one essential difference, though:
In the latter, one concretely fixes the background ḡµν typically as ḡµν = ηµν or as
ḡµν = δµν in the Euclidean case. In “background independent” quantum gravity
instead, the metric ḡµν is never specified concretely. All objects that one has
to compute in this setting, generating functionals, say, are functionals of the
variable ḡµν .
An example is the effective action Γ[hµν ; ḡµν ] that generates the dynamical
equations for the expectation value hµν ≡ ⟨b hµν ⟩ of bhµν and its higher n-point
functions. It depends on both the background metric and the fluctuation expecta-
tion value. Similarly all n-point correlation functions of b hµν which one computes
from it have a parametric dependence on ḡµν . To stress this fact we sometimes
write hµν [ḡ] ≡ ⟨b
hµν ⟩ḡ for the 1-point function, for example.
Thus, in a sense, the Background Independent quantization of gravity amounts
to its quantization on all possible backgrounds simultaneously.
There are now two metrics in the game which are equally important: the
background ḡµν and the expectation value of the full metric,

gµν ≡ ⟨b
gµν ⟩ = ḡµν + hµν , hµν ≡ ⟨b
hµν ⟩. (1.1)

Alternatively we may regard the effective action as a functional of the two metrics
rather than hµν and ḡµν . We define

Γ[gµν , ḡµν ] ≡ Γ[hµν ; ḡµν ] . (1.2)

hµν =gµν −ḡµν

Because of the almost symmetric status enjoyed by the two metrics we refer to
this setting as the “bi-metric” approach to the Background Independence problem.
As for the notion of an “exact renormalization group” in quantum gravity, we
will introduce the Effective Average Action (EAA) as a scale-dependent version
of the ordinary effective action with a built-in IR cutoff at a variable mass scale
8 A Quantum Field Theory of Gravity

k and derive in particular a functional RG equation for it. As we will explain in

more detail below, the construction of the EAA and its RG equation are only
possible due to the presence of the classical background spacetime.
Despite the unavoidable bi-metric appearance of the background field method,
the expectation value of the microscopic metric, gµν , and the variable back-
ground metric ḡµν , enter physical quantities (observables) not independently
but instead are constrained by a symmetry requirement. Obviously the full
metric gbµν ≡ ḡµν + bhµν is invariant under the split symmetry transformation
δbhµν = εµν , δḡµν = −εµν with an arbitrary symmetric tensor field εµν . At the
quantum level, this transformation implies Ward identities for the n-point func-
tions and the effective (average) action similar to those implied by gauge or
Becchi–Rouet–Stora–Tyutin (BRST) invariance. In either case one must make
sure that in the end the quantum theory constructed actually satisfies these Ward
identities.
In a way, this is the point where one is paying the price for the many advan-
tages the background field technique brings about. However, it will become clear
that while the extra work necessary to implement split symmetry, and thus Back-
ground Independence at the quantum level, is a hard technical challenge, it does
not involve insoluble problems of principle.

1.4 Asymptotic Safety in a Nutshell

In this section we give a concise preview of what Asymptotic Safety is about.
Technical details and refinements are omitted as much as possible. They will be
delivered later on in this book.

(1) The problem. The ultimate goal of the Asymptotic Safety program consists
in giving a mathematically precise meaning to, and∫ actually compute,
( )functional
integrals over “all” spacetime metrics of the form Db gµν exp iS[b
gµν ] , or
∫
Z = Db gµν e−S[bgµν ] , (1.3)

from which all quantities of physical interest can be deduced then. Here S[b gµν ]
denotes the classical or, more correctly, the bare action. It is required to be dif-
feomorphism invariant, but is kept completely arbitrary otherwise. In general it
differs from the usual Einstein–Hilbert action. This generality is essential in the
Asymptotic Safety program: the viewpoint is that the functional integral would
exist only for a certain class of actions S and the task is to identify this class.

(2) The problem, reformulated. Following the approach proposed in [14] one
attacks this problem in an indirect way: rather than dealing with the integral per
se, one interprets it as the solution of a certain differential equation, a functional
renormalization group equation, or “FRGE”. The advantage is that, contrary to
the functional integral, the FRGE is manifestly well defined. It can be seen as an
 1.4 Asymptotic Safety in a Nutshell 9

“evolution equation” in a mathematical sense, defining an infinite dimensional

dynamical system in which the RG scale plays the role of time.
Loosely speaking, this reformulation replaces the problem of defining functional
integrals by the task of finding evolution histories of the dynamical system that
extend to infinitely late times. According to the Asymptotic Safety conjecture
the dynamical system possesses a fixed point which is approached at late times,
yielding well-defined, fully extended evolutions, which in turn tell us how to con-
struct (or “renormalize”) the functional integral.

(3) From the functional integral to the FRGE. Let us now be slightly more
explicit about the passage from the functional integral to the FRGE.

(3a) Formal character of the integral. Recall that in trying to put the purely
formal functional integrals on a solid basis one is confronted with a number of
obstacles:

(i) As in every field theory, difficulties arise since one tries to quantize infinitely
many degrees of freedom. Therefore, at the intermediate steps of the con-
struction one keeps only finitely many of them by introducing cutoffs at very
small and very large distances, Λ−1 and k −1 , respectively. We shall specify
their concrete implementation in a moment. The ultraviolet and infrared
cutoff scales Λ and k, respectively, have the dimension of a mass, and the
original system is recovered for Λ → ∞, k → 0.
(ii) We mentioned already that the most severe problem one encounters when
trying to quantize gravity is the requirement of Background Independence.
In the approach to Asymptotic Safety along the lines of [14] we follow
the spirit of DeWitt’s background field method [50, 51] and introduce a
(classical, non-dynamical) background metric ḡµν , which is kept arbitrary.
We then decompose the integration variable as gbµν ≡ ḡµν + b hµν , or a non-
linear generalization thereof, and replace Db gµν with an integration over
the fluctuation, Db hµν . In this way one arrives at a conceptually easier task,
namely the quantization of the matter-like field b hµν in a generic, but classical
background ḡµν .
The availability of the background metric is crucial at various stages of
the construction of an FRGE. However, the final physical results do not
depend on the choice of a specific background.
(iii) As in every gauge theory, the redundancy of gauge-equivalent field config-
urations (diffeomorphic metrics) has to be carefully accounted for. Here
we
∫ √employ the Faddeev–Popov method and add a gauge-fixing term Sgf ∝
ḡḡ µν Fµ Fν to S where Fµ ≡ Fµ (b h; ḡ) is chosen such that the condition
Fµ = 0 picks a single representative from each gauge orbit. The resulting vol-
ume element on orbit space, the Faddeev–Popov determinant, is expressed
as a functional integral over Grassmannian ghost fields C µ and C̄µ , governed
by an action Sgh .
10 A Quantum Field Theory of Gravity
∫ ( )
In this way (1.3) gets replaced by Z[ e Φ̄] = DΦ b exp −Stot [Φ,b Φ̄] . Here the
total bare action Stot ≡ S + Sgf + Sgh depends on the dynamical fields Φ b≡
b
(hµν , C , C̄µ ), the background fields Φ̄ ≡ (ḡµν ), and possibly also on (both dynam-
µ

ical and background) matter fields, which for simplicity are not included here.

(3b) Standard effective action. Using the gauge fixed and regularized inte-
gral ∫we can compute arbitrary (Φ̄-dependent!) expectation values ⟨O(Φ)⟩ b ≡
e −1 b b b Φ̄]
−Stot [Φ,
Z DΦ O(Φ) e ; for instance, n-point functions where O consists
of strings Φ(x b 1 )Φ(xb 2 ) . . . Φ(x
b n ). For n = 1 we use the notation Φ ≡ ⟨Φ⟩ b ≡
(hµν , ξ , ξµ ), i.e., the elementary field expectation values are hµν ≡ ⟨b
µ ¯
hµν ⟩, ξ µ ≡
⟨C µ ⟩ and ξ¯µ ≡ ⟨C̄µ ⟩. Thus, the full dynamical metric has the expectation value
gµν ≡ ⟨b gµν ⟩ = ḡµν + hµν .
The dynamical laws which govern the expectation value Φ(x) have an elegant
description in terms of the effective action Γ. It is a functional depending on Φ
similar to the classical S[Φ] to which it reduces in the classical limit. Requiring
stationarity, S yields the classical field equation (δS/δΦ)[Φclass ] = 0, while Γ gives
rise to a quantum mechanical analog satisfied by the expectation values, the
effective field equation (δΓ/δΦ)[⟨Φ⟩] b = 0.
If, as in the case at hand, Γ ≡ Γ[Φ, Φ̄] ≡ Γ[hµν , ξ µ , ξ¯µ ; ḡµν ] also depends on back-
ground fields, the solutions of this equation inherit this dependence and thus
hµν ≡ ⟨b hµν ⟩ functionally depends on ḡµν .
Technically,∫ Γ is obtained from a functional integral with Stot replaced by
J
Stot ≡ Stot − dx J(x)Φ(x). b The new term couples the dynamical fields to exter-
nal, classical sources J(x) and repeated functional differentiation (δ/δJ)n of
e Φ̄] yields the n-point functions. In particular, Φ = δ ln Z/δJ.
ln Z[J, e It is a stan-
dard result that Γ[Φ, Φ̄] equals exactly the Legendre transform of ln Z[J, e Φ̄], at
fixed background fields Φ̄.
The importance of Γ also resides in the fact that it is the generating functional
of special n-point functions from which all others can be easily reconstructed.
Therefore, finding Γ in a given quantum field theory is often considered equiva-
lent to completely “solving” the theory.

(3c) Notions of gauge invariance. In practical applications of Γ[Φ, Φ̄] it is

advantageous to employ a gauge-breaking condition Fµ that fixes a gauge belong-
ing to the distinguished class of the so-called background gauges. To see the
benefit, recall that the original gauge transformations read δb gµν = Lv gbµν where
Lv , denotes the Lie derivative with regards to the vector field v.
When we decompose gbµν = ḡµν + b hµν we can distribute the gauge variation of
gbµν in different ways over ḡµν and bhµν . In particular, this gives rise to what is
known as quantum gauge transformations,

δQb
hµν = Lv (ḡµν + b
hµν ) , δ Q ḡµν = 0 (1.4)
 1.4 Asymptotic Safety in a Nutshell 11

and background gauge transformations,

δBb
hµν = Lv b
hµν , δ B ḡµν = Lv ḡµν . (1.5)

Note that the “ordinary” or “true” gauge invariance the Faddeev–Popov

method has to take care of is the δ Q -invariance. It must be gauge-fixed by the
condition Fµ (bh; ḡ) = 0.
Interestingly enough, there exist Fµ ’s, a variant of the harmonic coordinate
condition, for example, which indeed fix the δ Q -transformations, but at the
same time transform covariantly under δ B -transformations: δ B Fµ = Lv Fµ . They
implement the background gauges, and from now on we assume that one of those
is employed.
Then, as a consequence, the effective action Γ[Φ, Φ̄] is invariant under
background-gauge transformations which include the ghosts: δ B Γ[Φ, Φ̄] = 0 where
all fields transform as δ B Φ = Lv Φ, δ B Φ̄ = Lv Φ̄.
We emphasize that this property should not be confused with another notion of
“gauge independence,” which the above Γ[Φ, Φ̄] actually does not have: It is not
independent of which particular Fµ is picked from the class with δ B Fµ = Lv Fµ .
This Fµ -dependence will disappear only at the level of observables.

(3d) Effective Average Action. Heading now toward the concept of a func-
tional renormalization group for gravity we recall that the above definition of
Γ is based on the functional integral regularized in the IR and UV and hence
depends on the corresponding cutoff scales: Γ ≡ Γk,Λ [Φ, Φ̄]. It is this object for
which we derive a FRGE, more precisely a closed evolution equation governing
its dependence on the IR cutoff scale k. This is possible only if the IR regulariza-
tion is implemented appropriately, as in the so-called Effective Average Action
(EAA)[22, 23].
The EAA is related to the modified integral,
∫
b e−Stot b
e−∆Sk [Φ,Φ̄] ≡ Zk,Λ [J, Φ̄]
J
DΦ (1.6)

whose second exponential factor in the integrand, containing the cutoff action
∆Sk , is designed to achieve the IR regularization. To see how this works, assume
b = (b
the integration variable Φ h, C, C̄) is expanded in terms of eigenfunctions φp of
the covariant tensor Laplacian related to the background metric, D̄2 ≡ ḡ µν D̄µ D̄ν .
b ∑
Writing −D̄2 φp = p2 φp we have, symbolically, Φ(x) = p αp φp (x). The αp′ s are
generalized Fourier coefficients, and so the functional integration over Φ b amounts
to integrating over all αp :
∏ ∫ ∞ ( J ′ )
Zk,Λ [J, Φ̄] = dαp exp −Stot [{α}, Φ̄] . (1.7)
p2 ∈[0,Λ2 ] −∞
12 A Quantum Field Theory of Gravity

J ′ J b b Φ̄] with the expansion for Φb inserted.

Here Stot equals Stot [Φ, Φ̄] + ∆Sk [Φ,
In (1.7) we implemented the UV regularization by retaining only eigenfunctions
(or “modes”) corresponding to −D̄2 -eigenvalues (or squared “momenta”) smaller
than Λ2 . The IR contributions, i.e. those corresponding to eigenvalues between
p2 = 0 and about p2 = k 2 are cut off smoothly instead, namely by a p2 -dependent
suppression factor arising from ∆Sk .
To obtain a structurally simple FRGE, ∆S∫k should be chosen quadratic in the
dynamical fields. Usually one sets ∆Sk = 12 dx ΦR b kΦb with an operator Rk ∝
k R (−D̄ /k ) containing a dimensionless function R(0) . In the −D̄2 -basis we
2 (0) 2 2
∑
have then ∆Sk ∝ k 2 p R(0) (p2 /k 2 )αp2 , which shows that ∆Sk represents a kind
of p2 -dependent mass term: A mode with eigenvalue p2 acquires a (mass)2 of the
order k 2 R(0) (p2 /k 2 ).
We require R(0) (p2 /k 2 ) to have the qualitative properties of a smeared step
function, which, around p2 /k 2 ≈ 1, drops smoothly from R(0) = 1 for p2 /k 2 . 1
to R(0) = 0 for p2 /k 2 & 1. This achieves precisely ∫ the desired IR regularization:
In the product over p2 in (1.7), ∆Sk equips all dαp -integrals pertaining to the
low momentum modes, i.e., those with p2 ∈ [0, k 2 ], with a Gaussian suppression
factor e−k αp since for such eigenvalues R(0) (p2 /k 2 ) ≈ 1. The high momentum
2 2

modes, having p2 ∈ [k 2 , Λ2 ], yield R(0) (p2 /k 2 ) ≈ 0 and thus remain unaffected

by ∆Sk .
At least on a flat background, low (high) momentum modes φp (x) have long
(short) wavelengths. Therefore, when one lowers k from k = Λ down to k = 0 one
“unsuppresses” modes of increasingly long wavelengths, thus proceeding from
the UV to the IR.4
Note that at this point the availability of the background metric is crucial
for the whole construction: Via the associated covariant Laplacian D̄2 it defines
which modes are high or low momentum. Hence, tuning k from high to low scales,
the background metric also decides in which order the various field modes get
integrated out.
∫ The cutoff action for the gravitational field itself has the structure ∆Sk ∝
√
dx ḡ bhµν R(0) (−D̄2 /k 2 )b
hµν and hence is invariant under general coordinate
transformations if both b hµν and ḡµν transform as tensors. This highly desirable
property could not have been realized without having the background metric in
the game.
This process of encoding the contributions from an increasing number of modes
in a scale-dependent or “running” action functional is precisely a renormalization
in the modern sense due to Wilson [7].
The EAA, Γk,Λ [Φ, Φ̄], is defined to be the Legendre transform of ln Zk,Λ [J, Φ̄]
given by (1.7), with respect to J, for k, Λ, and Φ̄ fixed (and with ∆Sk [Φ, Φ̄]
subtracted from the result of the transformation, which is not essential here).

4 This is called the “integrating out” of the high momentum modes since in older approaches
the low momentum modes were completely discarded, rather than just suppressed.
 1.4 Asymptotic Safety in a Nutshell 13

The EAA has a number of important features not realized in other functional
RG approaches:

(i) Since no fluctuation modes are taken into account in the k = Λ → ∞ limit,
the EAA approaches (essentially) the bare (i.e., unrenormalized) action,
ΓΛ,Λ ∼ Stot . In the limit k → 0, it yields the standard effective action (with
an UV cutoff).
(ii) The EAA satisfies an exact FRGE independent of S and can be computed
by integrating this FRGE toward low k, with the initial condition ΓΛ,Λ =
Stot + . . . at k = Λ. (The dots stand for a computable correction term which
is unimportant here.)
(iii) The functional Γk,Λ [Φ, Φ̄] is invariant under background gauge transforma-
tions δ B for all values of the cutoffs. This property is preserved by the
FRGE: the RG evolution does not generate δ B -noninvariant terms.
(iv) The FRGE continues to be well behaved when the UV regularization is
removed (Λ → ∞). Denoting solutions to the UV cutoff-free FRGE by
Γk [Φ, Φ̄], it reads:
[( )−1 ]
1 (2)
k∂k Γk [Φ, Φ̄] = STr Γk [Φ, Φ̄] + Rk [Φ̄] k∂k Rk [Φ̄] . (1.8)
2
(2)
Here STr denotes the functional supertrace, and Γk stands for the matrix
of second functional derivatives of Γk with respect to Φ at fixed Φ̄. The
interplay of the regulators Rk in the numerator and denominator ensures
that the argument of the supertrace is strongly peaked around p2 = k 2 and
vanishes for p2 ≫ k 2 . As a consequence, the trace is perfectly finite both in
the IR and the UV, and this is why sending Λ → ∞ was unproblematic.
(v) At least in non-gauge theories, Γk is closely related to a generating func-
tional for field averages over finite domains of size k −1 ; hence the name EAA
[22, 23]. Thanks to this property, when treated as a classical action, Γk can
provide an effective field theory description of quantum physics involving
typical momenta near k. This property has been exploited in numerous
applications of the EAA to particle and condensed matter physics, but it
plays no role in the present context. Rather, it is its interpolating prop-
erty between S and Γk=0 which is instrumental in the Asymptotic Safety
program.

(3e) Theory space. The arena in which the RG dynamics takes place is the
infinite dimensional theory space, T . It consists of all well-behaved action func-
tionals (Φ, Φ̄) 7→ A[Φ, Φ̄], which depend on a given set of fields and respect certain
subsidiary conditions and symmetry constraints. In metric gravity the “points”
of T are δ B -invariant functionals A[gµν , ḡµν , ξ µ , ξ¯µ ].
The RHS of the FRGE (1.8) defines a vector field β on T . Its natural orien-
tation is such that β points from higher to lower momentum scales k, from the
14 A Quantum Field Theory of Gravity

UV to the IR. (This is the direction of increasing “coarse-graining” in which the

microscopic dynamics is “averaged” over increasingly large spacetime volumes.)
The integral curves of this vector field, k 7→ Γk , are the RG trajectories, and the
pair (T , β) is called the RG flow . It constitutes the dynamical system alluded to
earlier.

(3f ) Coordinates, a.k.a. coupling constants. One usually assumes that

∑∞
every A ∈ T can be expanded as A[Φ, Φ̄] = α=1 ūα Iα [Φ, Φ̄] where the set {Iα }
forms a basis of invariant functionals. Writing the RG trajectory correspond-
∑∞
ingly, Γk [Φ, Φ̄] = α=1 ūα (k)Iα [Φ, Φ̄], one encounters infinitely many running
coupling constants, ūα (k), whose k-dependence is governed by an infinite coupled
system of differential equations: k∂k ūα (k) = β̄ α (ū1 , ū2 , · · · ; k). The dimension-
ful beta functions β̄ α arise by expanding the RHS of the FRGE: 21 STr[· · · ] =
∑∞ α α
α=1 β̄ Iα [Φ, Φ̄]. The coefficients β̄ are similar to the familiar beta functions
of perturbative quantum field theory (where, however, only the finitely many
beta functions of the relevant couplings are considered).
Reexpressing the RG equations in terms of dimensionless couplings uα ≡
k −dα ūα with dα the canonical mass dimension of ūα , the resulting FRGE
in component form is autonomous, i.e,. its β-functions have no explicit k-
dependence: k∂k uα (k) = β α (u1 (k), u2 (k), · · · ). The coupling constants (uα ) ≡ u
serve as local coordinates on T , and the β α s are the components of the vector field
β ≡ (β α (u)).

(4) Asymptotic Safety construction of the UV limit. The construction

of a quantum field theory involves finding an RG trajectory that is infinitely
extended in the sense that it is a curve, entirely within theory space, with well-
defined limits k → 0 and k → ∞, respectively. Asymptotic Safety is a proposal for
ensuring the existence of the second limit. Its crucial prerequisite is a non-trivial
RG fixed point on T .

(4a) Fixed points. By definition, at a fixed point of the RG flow the vector
field β vanishes, β = 0, so its coordinates (uα ∗ ) = u∗ satisfy the infinitely many
conditions β α (u∗ ) = 0. The fixed points UV critical hypersurface, SUV , or syn-
onymously its unstable manifold, is defined to consist of all points in T , which
are pulled into the fixed point under the inverse RG flow, i.e., for increasing
scale k. ( )
∑
Linearizing the flow about u∗ one has k∂k uα (k) = γ B α γ uγ (k) − uγ∗ with
the stability matrix B = (B α γ ), B α γ ≡ ∂γ β α (u∗ ). If the eigenvectors of B form a
∑ ( )
α k0 θI
basis, its solution reads uα (k) = uα ∗ + I CI VI k . Here the CI ’s are con-
stants of integration and the VI ’s denote the right-eigenvectors of B with
∑
eigenvalues −θI , i.e., γ B α γ VIγ = −θI VIα . In general, B is not symmetric and
the critical exponents θI are complex.
 1.4 Asymptotic Safety in a Nutshell 15

Along eigendirections with Re θI > 0 (Re θI < 0) deviations from uα ∗ grow

(shrink) when k is lowered from the UV toward the IR; they are termed relevant
(irrelevant).
A trajectory uα (k) within SUV , by definition, approaches uα (k → ∞) = uα ∗ in
the UV. For the constants CI in its linearization this implies that CI = 0 for all
I with Re θI < 0. Hence, the trajectories in SUV are labeled by the remaining
CI′ s related to the critical exponents with Re θI > 0. (For simplicity, we assume
all Re θI non-zero.)
As a consequence, the dimensionality of the critical hypersurface SUV equals
the number of critical exponents with Re θI > 0, i.e., the number of relevant
directions.
A fixed point is called Gaussian if it corresponds to a free field theory. Its
critical exponents agree with the canonical mass dimension of the corresponding
operators. A fixed point whose critical exponents differ from the canonical ones
is referred to as non-trivial or as a non-Gaussian fixed point (NGFP).

(4b) Asymptotically safe RG trajectories. Now we return to the key prob-

lem of ensuring the existence of the k → ∞ limit. So let us assume that there is
indeed a fixed point of the RG flow on theory space. Then it is sufficient to pick
any of the trajectories within its hypersurface SUV to be sure that the trajectory
has a singularity-free UV behavior. In this case it is certain that the trajectory
will always hit the fixed point for k → ∞, thus excluding the possibility that it
ultimately leaves the space of acceptable actions, T .
There exists a dim(SUV )-parameter family of such “asymptotically safe” tra-
jectories. Assuming the existence of a suitable
{ IR-limit, any
} of them constitutes
a complete family of action functionals, Γk , k ∈ (0, ∞) , which describes the
integrating out of all modes of the field, and hence may be seen as defining a
quantum field theory.
Most probably a UV fixed point is not only sufficient but also necessary for an
acceptable theory without divergences. Therefore, in the simplest case when there
exists only one, the physically inequivalent asymptotically safe quantum theo-
ries one can construct are labeled by the dim(SUV ) parameters characterizing
trajectories inside SUV .
Thus, the degree of predictivity of asymptotically safe theories is essentially
determined by the number of relevant eigendirections. If dim(SUV ) is a finite
number, it is sufficient to measure only dim(SUV ) of the couplings {uα (k)}
characterizing Γk in order to predict the infinitely many others. In particular, at
k = 0 the standard effective action Γ ≡ Γ0 is obtained which “knows” all possible
predictions.

(4c) Background Independence of the second kind. The only input

required in order to start the search for asymptotically safe quantum field theories
16 A Quantum Field Theory of Gravity

is the theory space T , i.e., the desired field content and the symmetries. It fully
determines the structure of the FRGE and, as a consequence, the RG flow and
in particular its fixed point properties.
In the EAA approach, the functional flow equation with the UV regulator
removed is completely independent of the bare action S in the original, merely
formal functional integral from which we started. Hence, the entire RG flow
and in particular (if our search is successful) its fixed points enjoy an existence
independent of any preconceived bare action, and so the same is true for the
asymptotically safe trajectories related to those fixed points.
We refer to this property as Background Independence of the second kind. This
term expresses the independence of our approach from a preferred action, in the
same spirit as the familiar Background Independence (of the first kind) refers to
the absence of a distinguished metric.

(5) Construction of a bare action. By the general properties of the EAA

in the presence of an explicit UV regulator, the functional ΓΛ,Λ ≡ Γk=Λ,Λ for
Λ → ∞ is closely related to the bare action S ≡ SΛ in the UV regularized func-
tional integral, ΓΛ,Λ = SΛ + · · · This asymptotic action should not be confused
with the limit Γk=Λ→∞ of a solution Γk to the UV regulator-free flow equation
(1.8). Under certain (technical) conditions the two can be related, provided one
more piece of input is supplied, namely a UV-regularized measure (and of course
a UV-regularization scheme, in the first place). The information contained in
an asymptotically safe trajectory Γk near the fixed point should then suffice
to determine how the Λ-dependent couplings in the bare action SΛ→∞ must
be tuned so that, together with the particular measure selected, a mathemati-
cally well-behaved functional integral arises in the limit Λ → ∞. This last step
in the Asymptotic Safety program is sometimes referred to as the reconstruction
problem, i.e., the task of a posteriori finding a bare, or “classical” theory that
reproduces a given effective one.
Thus, contrary to the usual situation in quantum mechanics or quantum field
theory the implicitly underlying classical theory is actually computed here. It is
an output rather than an input, giving special predictive power to the approach.
Based on the EAA, the Asymptotic Safety program may be thought of as a sys-
tematic research process among quantum theories, rather than the quantization
of a classical system known beforehand.
It has become customary to call Quantum Einstein Gravity, or QEG, any quan-
tum field theory of metric-based gravity, regardless of its bare action, which is
defined by a complete trajectory on the theory space T ≡ TQEG . So conceptually
QEG has a priori no reason to be related to the familiar Einstein–Hilbert action,
or any other simple diffeomorphism invariant functional of the metric one might
guess naively.
 1.5 Continuum or “Atoms of Spacetime”? 17

1.5 Continuum or “Atoms of Spacetime”?

We saw that the EAA-based search for a fundamental quantum theory of gravity
and the geometry of spacetime makes use of classical or at least classical-looking
fields in order to describe expectation values like ⟨b gµν (x)⟩ ≡ gµν (x) and that
the entire physical contents of such a theory would be encoded in an effective
action functional defined over such fields. This might cause concerns that the
whole program is perhaps doomed to failure right from the start because its
mathematical setting is still “too classical” and therefore incapable of successfully
coping with the presumably very “exotic” physics of strongly non-perturbative
gravity at microscopic scales.
Such concerns were nurtured, for instance, by the discovery of quantized area
and volume elements in the framework of Loop Quantum Gravity (LQG), which
led to the intuitive picture of spacetime being built up from tiny discrete building
blocks, the “atoms of spacetime.” It is not yet clear whether or not this kind
of fundamental discreteness (at the kinematical level) can have any imprint on
observable physics. But for the sake of the argument, let us assume it has, and let
us explain why this is nevertheless not an argument against the continuum-based
Asymptotic Safety program.
It is indeed possible to employ the language of continuum functions on classical
manifolds in order to describe “quantum spacetimes,” which are fundamen-
tally different from those in General Relativity, differentiable manifolds furnished
with classical metrics. This probably includes even highly non-classical quantum
spacetimes with discrete or lattice-like features.
A proof of principle can already be found in elementary quantum mechan-
ics, namely in the Wigner–Weyl–Moyal formalism, which represents all states
(density matrices) and operators by ordinary commuting functions on the sys-
tem’s classical phase space manifold. The canonical commutation relations are
nevertheless fully accounted for since the operator product gets replaced by a
non-commutative star product for such functions.
This is perhaps the simplest instance of a much more general methodology
also used in modern non-commutative geometry [52, 53], namely defining and
characterizing “non-commutative manifolds” in terms of an algebra of functions
defined over them. In the quantum mechanics example this algebra consists of the
phase space functions equipped with the star product. It encodes, for example,
the well-known partitioning of (2-dimensional) phase space in elementary blocks
which realize the fundamental unit for the symplectic volume set by Planck’s
constant. These are the Planck cells which underlie the familiar counting rule
“one state per Planck cell.”
It is instructive to see how the continuum language of the classically looking
phase space functions manages to express this lattice-like discreteness. In this
18 A Quantum Field Theory of Gravity

language, coherent states |p, q⟩ are represented by Gaussian functions5 on phase

space that are peaked at a given point (p, q) ∈ R × R. Because of the one-to-one
correspondence between such points and Gaussians we may think of phase space
as the set of all Gaussians (with a given width fixed by Planck’s constant) which
in turn are in one-to-one correspondence
{ with} the coherent states |p, q⟩.
Now, note that the set |p, q⟩ , p ∈ R, q ∈ R seen as a Hilbert space basis is
overcomplete. In order to reduce it to a subset which is complete but not over-
complete, one can restrict the labels p, q of the states |p, q⟩ to the discrete points of
the von Neumann lattice. It is a regular lattice in the p-q plane whose elementary
cell has a volume of precisely one Planck unit [56].
These remarks are certainly not to say that it is really a von Neumann lattice
that is at the heart of spacetime on microscopic scales.6 They make it plausible
though that the framework we are going to advocate in this book, namely a
continuum-based background field approach, is in principle capable to also cope
with potentially quite formidable “deformations” of classical geometry such as
the replacement of a continuum by a discrete point set.
In summary, classical functions over a continuum should be seen merely as a
language to formulate theories, and as such it is largely unrelated to the physical
content such theories may have.
The discrete and the continuum languages have complementary advantages
and disadvantages. The former, in its incarnations of Loop Quantum Gravity or
Causal Dynamical Triangulations, for example, is applied most easily in the deep
UV, i.e., for a description of the building blocks or “atoms” of spacetime, which
may or may not manifest themselves as a physical, observable discreteness in the
end. Moving toward the IR in these approaches is comparatively difficult though
since assembling many such building blocks amounts to solving an increasingly
complex dynamical problem. As a result, proving the emergence of a classical
spacetime is one of the hardest problems in these approaches.
Adopting the continuum language the situation is exactly the opposite. While
the desired classical regime of the theory is also not for free in this setting, it
is much more easily explored by the background field method and a continuum
description because its mathematical language is already similar to that of classi-
cal General Relativity. In the extreme UV, on the other hand, a continuum-based
theory must “work very hard,” that is it must be able to correctly handle the
large quantum effects which become unavoidable should the microstructure of
spacetime be very non-classical.

5 Like for all state vectors they are the “Weyl symbol” of the associated pure state density
matrix, a.k.a. the Wigner function of the pertinent wave function [54, 55].
6 It is amusing though that the microscopic building blocks were “Planck cells” in a double
sense of the word then. For concrete realizations of this idea see also [57, 58].
 2
The Functional Renormalization Group

In this chapter we introduce the general idea of the Effective Average Action
and we describe in particular the Functional Renormalization Group Equation
it gives rise to [22–28, 31–35].1 To avoid inessential technical complications we
choose the simplest setting possible, namely a scalar matter field theory on a
non-dynamical, flat Euclidean space. We will mostly focus on those aspects that
later on will be important in quantum gravity, referring to the literature for a
more exhaustive survey.

2.1 The Concept of the Effective Average Action

We start by considering a single-component real scalar field ϕ, which is defined
on the d-dimensional Euclidean spacetime (Rd , δµν ). Its dynamics is governed
by the{ bare action S[ϕ]. Typically the} functional S has the structure S[ϕ] =
∫ d
d x 12 (∂µ ϕ)2 + 12 m2 ϕ2 + interactions , but we will not assume any specific
form of S in the following.
In all applications of quantum or statistical field theory the ultimate goal
consists in computing expectation values ⟨O⟩ of observables O, given for instance
by the (a priori formal) functional integral2
∫
{ }
⟨O⟩ = Dϕb O(ϕ) b exp −S[ϕ]
b . (2.1)

The most important example of such expectation values are Green functions,
or
⟨ n-point functions, ⟩ where O consists of a string of elementary fields,
b 1 )ϕ(x
ϕ(x b 2 ) · · · ϕ(x
b n ) . Generating functionals are a technically convenient way

1 General references on functional integration include [59–62].

2 b serve
Here and in the following our convention is to use a caret to denote fields which, like ϕ,
as integration variables. The same symbol without a caret, like ϕ, will always denote the
corresponding expectation value or “classical fields”.
20 The Functional Renormalization Group

of treating all n-point functions in one stroke by representing them as a multiple

functional derivative with respect to a “source” function J:
⟨ ⟩ δ n Z[J]
b 1 )ϕ(x
ϕ(x b 2 ) · · · ϕ(x
b n) = . (2.2)
δJ(x1 ) · · · δJ(xn ) J=0

After coupling ϕ(x) to a source J(x) we can write down a path integral repre-
sentation for it, and for W [J], the generating functional of all connected Green
functions:
∫ { ∫ }
Z[J] ≡ eW [J] = Dϕb exp −S[ϕ] b + dd x ϕ(x)J(x)
b . (2.3)

A third generating functional is the well-known standard effective action, Γ[ϕ],

which is given by the Legendre transform of W [J]. It depends on the ⟨ vari-
⟩
able “dual” to the source J, namely the field expectation value ϕ(x) ≡ ϕ(x) b =
δW [J]/δJ(x).
Here and in the following the notation ⟨ · · · ⟩ is understood to denote
connected n-point functions in the presence of a ⟨source; hence J(x)⟩ is
not put to zero after the functional differentiation: ϕ(x b 1 )ϕ(x
b 2 ) · · · ϕ(x
b n) =
δ W [J]/δJ(x1 ) · · · δJ(xn ). The functional Γ in turn can be shown to gener-
n

ate all 1-particle irreducible Green functions by a similar multiple functional

differentiation with respect to ϕ(x).
In order to make the functional integral well-defined a UV cutoff is needed.
One could replace Rd by a d-dimensional lattice Zd , for example. The formal
∏
measure Dϕb would then turn into the regularized product x∈Zd dϕ(x). b In the
following we implicitly assume the presence of some sort of UV regularization,
but we leave its detailed form unspecified and use continuum notation for the
fields and their Fourier transforms.
The construction of the Effective Average Action [22] starts out from Wk [J], a
modified form of the functional W [J] which has a built-in IR cutoff at a variable
mass scale, k. This scale is used to classify the Fourier modes of ϕb as “short wave-
length” or “long wavelength,” depending on whether their momentum square
p2 ≡ pµ pµ is, respectively, larger or smaller than k 2 . We would like the modes
with p2 > k 2 to contribute without any suppression to the functional integral
defining Wk [J], while those with p2 < k 2 should contribute only with a reduced
weight or even be suppressed altogether, depending on which variant of the for-
malism is used. The new functional Wk [J] can be thought of as the conventional
generating functional belonging to a modified theory whose bare action S + ∆Sk
includes an additional cutoff action ∆Sk [ϕ]: b
∫ { ∫ }
{ }
Zk [J] = exp Wk [J] = Dϕb exp − S[ϕ] b − ∆Sk [ϕ]
b + dd x ϕ(x)J(x)
b . (2.4)

b serves the purpose of suppressing the “IR modes”

The factor exp{−∆Sk [ϕ]}
b
contained in the field ϕ(x), i.e., those of its Fourier components which have
 2.1 The Concept of the Effective Average Action 21

p2 < k 2 . In momentum space, the new term in the action is taken to be of the
form ∫
b 1 dd p
∆Sk [ϕ] ≡ Rk (p2 ) |ϕbp |2 , (2.5)
2 (2π)d
with Rk (p2 ) a certain cutoff function, and ϕbp the Fourier transform of ϕ(x) b
defined in Appendix A. The precise details of the function p 7→ Rk (p ) are arbi-
2 2

trary to a large extent; what matters is only its limiting behavior for p2 ≫ k 2 and
p2 ≪ k 2 , respectively. In the simplest case3 we require that
{ 2
k for p2 ≪ k 2 ,
Rk (p2 ) ≈ (2.6)
0 for p2 ≫ k 2 .

The first condition leads to a suppression of the small momentum modes by a soft
mass-like IR cutoff, the second guarantees that the large momentum modes are
integrated out in the usual way. Adding ∆Sk to the above bare action leads to
∫
( )
b =1 dd p [ 2 ]
S + ∆Sk [ϕ] p + m 2
+ Rk (p2
) |ϕbp |2 + interactions. (2.7)
2 (2π)d

Obviously the cutoff function Rk (p2 ) has the interpretation of a momentum

dependent mass square which vanishes for p2 ≫ k 2 , while it assumes the con-
stant value k 2 for p2 ≪ k 2 . How precisely Rk (p2 ) interpolates between these two
regimes is a matter of calculational convenience. In explicit calculations one often
uses the exponential cutoff
[ ]−1
Rk (p2 ) = p2 exp(p2 /k 2 ) − 1 , (2.8)

but many other choices are possible as well [31, 32, 63–65].
One could also suppress the p2 ≪ k 2 modes completely. This is achieved by
allowing Rk (p2 ) to diverge for p2 ≪ k 2 so that exp{−∆k S[ϕ]}b → 0 for modes with
p ≪ k . While this behavior of Rk (p ) seems most natural from the viewpoint of
2 2 2

a Kadanoff–Wilson type coarse graining, its singular behavior makes the resulting
generating functional problematic to deal with technically. For this reason, and
since it will still allow us to derive an exact RG equation later on, we prefer to
work with a smooth cutoff satisfying (2.6).
At the non-perturbative path integral level, the∫ presence of ∆Sk suppresses the
long-wavelength modes by a factor exp{− 12 k 2 |ϕ| b 2 }. In perturbation theory,
instead, according to (2.7), the ∆Sk term leads to the modified propagator
[p2 + m2 + Rk (p2 )]−1 , which equals [p2 + m2 + k 2 ]−1 for p2 ≪ k 2 . Thus, when
computing loops with this propagator, k 2 acts indeed as a conventional IR cutoff
if m2 ≪ k 2 . (It plays no role in the opposite limit m2 ≫ k 2 in which the physical
particle mass acts like an IR cutoff.)

3 We will discuss a slight generalization of these conditions at the end of this section.
22 The Functional Renormalization Group

We also note that by replacing p2 with −∂ 2 in the argument of Rk (p2 ) the

cutoff action can be written in a way which no longer makes any explicit reference
b
to the Fourier decomposition of ϕ:
∫
b 1 b 2 b
∆Sk [ϕ] = dd x ϕ(x)R k (−∂ )ϕ(x). (2.9)
2

The next steps toward the definition of the Effective Average Action are similar
to the standard
⟨ procedure.
⟩ One defines the (now k-dependent) field expectation
b
value ϕ(x) ≡ ϕ(x) = δWk [J]/δJ(x) and computes the Legendre transform of
Wk [J] with respect to J(x), at fixed k. If the functional relationship ϕ = ϕ[J] can
be solved for the source to yield J = Jk [ϕ], it assumes the form
∫
[ ]
e
Γk [ϕ] = dd x Jk [ϕ](x)ϕ(x) − Wk Jk [ϕ] , (2.10)

where the notation Jk [ϕ](x) indicates that Jk is both a functional of the function
e k is given by the Legendre–
ϕ and a function of the point x. In the general case, Γ
Fenchel transform [66]:
{∫ }
e k [ϕ] = sup
Γ dd x J(x)ϕ(x) − Wk [J] . (2.11)
J(x)

ek :
The actual EAA, denoted by Γk [ϕ], is obtained by subtracting ∆Sk [ϕ] from Γ
∫
e k [ϕ] − 1
Γk [ϕ] ≡ Γ dd x ϕ(x)Rk (−∂ 2 )ϕ(x). (2.12)
2

The rationale for this specific definition becomes clear when we look at the list
of properties enjoyed by the Effective Average Action (EAA):

(1) The EAA gives rise to the effective field equation

δΓk [ϕ]
+ Rk (−∂ 2 )ϕ(x) = J(x). (2.13)
δϕ(x)

This scale-dependent source–field relationship follows from the definition (2.12)

e k [ϕ]/δϕ(x) = J(x). The latter equation is obtained by differenti-
together with δ Γ
ating (2.10) and noting that the terms that are due to the ϕ-dependence of Jk [ϕ]
mutually cancel by virtue of ϕ(y) = δWk [J]/δJ(y):

ek ∫ ∫
δΓ δJk δWk δJk
= Jk [ϕ](x) + dd y ϕ(y) − dd y
δϕ(x) δϕ(y) δJ(y) J=Jk δϕ(y) (2.14)
= Jk [ϕ](x).
 2.1 The Concept of the Effective Average Action 23

(2) The EAA satisfies a functional renormalization group equation, namely the
Wetterich equation:
[( )−1 ∂ ]
∂ 1 (2)
k Γk [ϕ] = Tr Γk [ϕ] + Rk k Rk . (2.15)
∂k 2 ∂k

Here the RHS of (2.15) uses a compact infinite dimensional matrix notation.
(2)
In the position space representation, the operator Γk has the matrix elements
(2)
Γk (x, y) ≡ δ 2 Γk /δϕ(x)δϕ(y), i.e., it is the Hessian of the EAA. Likewise, the cut-
off operator has matrix elements Rk (x, ∫ dy) ≡ Rk (−∂x )δ(x − y), and the functional
2

trace Tr corresponds to an integral d x.

At the level of the functional RG equation (2.15) the implicit UV cutoff can be
removed trivially now. This( (2) is most)−1
easily seen in the momentum (representation.
There the product of Γk + Rk , which equals essentially p2 + Rk (p2 ) +
)−1
··· ∂
, and k ∂k Rk (p2 ), considered a function of p2 , is significantly different
from zero only in a narrow interval centered around p2 = k 2 . Hence, the trace
receives contributions from a thin shell of momenta p2 ≈ k 2 only and is therefore
well convergent both in the UV and IR.
The derivation of the FRGE (2.15) proceeds as follows [22]. Taking the
k-derivative of (2.11) with (2.4) and (2.9) inserted one finds
∫ ∫
∂ e 1 ⟨ ⟩ ∂
k Γ k [ϕ] = dd
x b ϕ(y)
dd y ϕ(x) b k Rk (x, y) . (2.16)
∂k 2 ∂k
The angular brackets denote a normalized expectation; for a general observable
O it is given by
∫ { ∫ }
⟨O⟩ ≡ e−Wk [J] Dϕb O(ϕ)
b exp −S − ∆Sk + J ϕb . (2.17)

Note that ⟨O⟩ depends on both J and k, but we do not indicate this notationally.
Next, it is convenient to introduce the (likewise J and k dependent) connected
two-point function Gxy ≡ G(x, y) ≡ δ 2 Wk [J]/δJ(x)δJ(y) and the ϕ and k depen-
dent Hessian of Γe k : (Γe (2) )xy ≡ δ 2 Γ
e k [ϕ]/δϕ(x)δϕ(y). Since Wk and Γ
e k are related
k
by a Legendre transformation one shows in the usual way that
∫ e k [ϕ]
δ 2 Wk [J] δ2 Γ
dd z = δ(x − y). (2.18)
δJ(x)δJ(z) δϕ(z)δϕ(y)
Hence G and Γ e (2) are mutually inverse matrices: GΓ
e (2) = 1. Furthermore, taking
two J-derivatives of (2.4) one obtains the following decomposition of the two-
point function in a connected plus a disconnected part:
⟨ ⟩
b ϕ(y)
ϕ(x) b = G(x, y) + ϕ(x)ϕ(y). (2.19)
Substituting this representation for the two-point function into (2.16) we arrive at
∫
[ ]
e k [ϕ] = 1 Tr ∂k Rk G + 1 dd x ϕ(x) ∂k Rk (−∂ 2 ) ϕ(x).
∂k Γ (2.20)
2 2
24 The Functional Renormalization Group
[ ]
In terms of the EAA (2.12),∫ this equation boils down to ∂ k Γk [ϕ] = 1
2 Tr ∂ k Rk G .
The cancellation of the 12 ϕ ∂k Rk ϕ term is a first motivation for the definition
(2.12) where this term is subtracted from the Legendre transform Γ e k . The deriva-
( (2) )−1
e (2) −1
tion of the FRGE is completed by noting that G = [Γ ] = Γk + Rk ,
(2) e (2)
where the second equality follows by differentiating (2.12): Γk = Γk − Rk .

(3) The EAA satisfies the following integro-differential equation:

∫ { ∫ }
{ }
exp −Γk [ϕ] = Dϕb exp −S[ϕ] b + dd x (ϕb − ϕ) δΓk [ϕ]
δϕ(x)
{ 1∫ } (2.21)
× exp − dd x (ϕb − ϕ)Rk (−∂ 2 )(ϕb − ϕ) .
2

This equation is easily derived by combining (2.4), (2.10), and (2.12) and by
e k /δϕ = J.
using the effective field equation δ Γ

(4) For k → 0 the EAA approaches the ordinary effective action, limk→0 Γk = Γ,
and for k → ∞ the bare action, Γk→∞ = S:
k→0 k→∞
Γ ←−−−−− Γk −−−−−→ S. (2.22)

The k → 0 limit is a direct consequence of (2.6); Rk (p2 ) vanishes for all p2 > 0
when k → 0. The derivation of the k → ∞ limit makes use of the integro-
differential equation (2.21). A formal version of the argument is as follows. Since
Rk (p2 ) approaches∫ k 2 for k → ∞, the second exponential on the RHS of (2.21)
becomes exp{−k 2 dd x(ϕb − ϕ)2 } in this limit, which, up to a normalization fac-
tor, approaches a delta-functional δ[ϕb − ϕ]. The ϕb integration can be performed
trivially then and one ends up with limk→∞ Γk [ϕ] = S[ϕ]. In a more careful
treatment [22] one shows that the saddle-point approximation of the functional
integral in (2.21) about the point ϕb = ϕ becomes exact in the limit k → ∞. As a
result, limk→∞ Γk and S differ at most by the infinite mass limit of a one-loop
determinant, which we omit here since it plays no role in typical applications
(see [30] for a more detailed discussion).

(5) There exists a second, equivalent form of the FRGE :

[ )]
∂ 1 D ( (2)
k Γk [ϕ] = Tr k ln Γk [ϕ] + Rk . (2.23)
∂k 2 Dk
Here D/Dk is a special scale derivative which by definition acts only on the
(2)
k-dependence of Rk , but not of Γk .
While (2.23) is exact, it is instructive to see how it relates to the one-loop
approximation of the EAA. The latter is given by the formal expression
1 ( )
Γk [ϕ] = S[ϕ] + Tr ln S (2) [ϕ] + Rk + O(2 loops), (2.24)
2
 2.1 The Concept of the Effective Average Action 25

which one derives most easily from the functional integral (2.4) by going through
exactly the same steps as in the standard case [59]. The only difference to be taken
into account is the regulator term ∆Sk in the bare action which contributes the
( )(2)
term ∆Sk = Rk to its Hessian. In this way, the familiar one-loop expression
ln det(S ) = Tr ln(S (2) ) immediately leads to (2.24).
(2)

Note that (2.24) makes sense only in the presence of a UV cutoff since the
functional trace in (2.24) would not exist otherwise.
Taking a k-derivative on both sides of (2.24) leads to
[ )]
∂ 1 D ( (2)
k Γk [ϕ] = Tr k ln S [ϕ] + Rk + O(2 loops). (2.25)
∂k 2 Dk

Obviously (2.25) differs only in one respect from the exact equation (2.23): the
argument of the logarithm contains S (2) rather than the Hessian of the quantum-
(2)
corrected action, Γk . In writing down (2.25) we tacitly pulled the k-derivative
under the trace. This amounts to a UV regularization since upon taking the
scale derivative of the logarithm we get the same suppression factor ∂Rk /∂k as
in (2.15).
So the conclusion is that the FRGE has almost the structure of a one-loop
result, but with two crucial modifications, which render it an exact relation:
First, a mode-suppression term ∆Sk is added which serves as a variable IR cut-
off, and second, in the one-loop determinant the bare action is replaced with
(2)
the EAA. The replacement S (2) → Γk is an example of what is called an RG
improvement, a topic to which we will come back to later on.

(6) The FRGE by itself is completely independent of the bare action S. The
latter comes into play only via its initial condition, ΓΛ = S. In fact, in the EAA
approach, we trade the problem of performing the functional integral for the task
of solving the RG equation all the way down from some very high UV scale k = Λ,
Λ → ∞, where the initial condition ΓΛ = S is imposed, down to k = 0, where the
EAA equals the ordinary effective action Γ, the object which we actually would
like to know; see Figure 2.1 for an illustration.

0 k k k

p
0
k k k S

Figure 2.1. The figure shows the |p|-axis corresponding to the momenta of the
field modes ∝ exp(ipµ xµ ). Solving the FRGE for Γk subject to initial conditions
imposed at k = Λ amounts to “integrating out” the momentum eigenmodes
with |p| ∈ [k, Λ]. Changing the IR cutoff infinitesimally from k to k + δk, the
[( (2) )−1 ]
EAA changes by the amount δΓk = 21 Tr Γk + Rk ∂k Rk δk, according to
the FRGE.
26 The Functional Renormalization Group

(7) Since the FRGE depends on the cutoff operator Rk chosen, so do its solutions
in general. The change of Γk [ϕ] caused by a small “deformation” Rk → Rk + δRk
is governed by the exact functional equation
[( )−1 ]
1 (2)
δΓk [ϕ] = Tr Γk + Rk δRk . (2.26)
2

The derivation of this relation parallels the derivation of the FRGE.

(8) The EAA is closely related to a generating functional for correlators of field
averages, hence its name. To see this, consider an arbitrary field configuration ϕb
and associate the following new field to it:
∫
ϕ(x) = b
dd y Fk (x − y) ϕ(y). (2.27)

Here Fk denotes
∫ a spherically symmetric and normalized weight function, i.e.,
Fk ≥ 0 and dd xFk (x) = 1, which is nonzero and approximately constant at small
distances |x − y| ≪ k −1 , and vanishes for |x − y| ≫ k −1 . Near |x − y| ≈ k −1 it
interpolates between the two regimes smoothly. As a result, the value of ϕ at the
point x is a (weighted) average of the original field ϕb over a ball in the Euclidean
spacetime, which is centered at x and has a radius of order k −1 . The average
value field ϕ is said to be obtained from the original one, ϕ, b by a process of
coarse graining. While ϕ and ϕb show the same features at large distance scales,
ϕ is lacking the finer short-distance details of ϕb that got “averaged away”.
If we express ϕ(x)b and ϕ(x) in terms of their Fourier transforms ϕbp and
ϕp , respectively, the convolution in (2.27) turns into a simple multiplication in
momentum space:
ϕp = fk (p2 ) ϕbp . (2.28)

Back in position space it amounts to applying the operator fk (−∂ 2 ):

b
ϕ(x) = fk (−∂ 2 ) ϕ(x). (2.29)

Here fk is related to the kernel Fk (x − y) by

∫
dd p
Fk (x − y) = fk (−∂ ) δ(x − y) =
2
fk (p2 ) eip(x−y) . (2.30)
(2π)d

For example, if we choose

( 1 a2 Gaussian ) weight function in position space,
−d
Fk (x − y) = (2π) k exp − 2 k (x − y) , its counterpart in momentum space
2 d 2

is Gaussian as well: ( p2 )
fk (p2 ) = exp − 2k 2 . (2.31)
Using (2.31) in (2.28) makes it obvious that the low-momentum Fourier com-
ponents of ϕ and ϕb coincide essentially, while the average field has vanishing
 2.1 The Concept of the Effective Average Action 27

high-momentum components: ϕp = ϕbp if p2 ≪ k 2 , and ϕp = 0 if p2 ≫ k 2 , as it

should be.
Next, we turn to ensembles of field configurations and define a partition
function which is restricted to fields giving rise to a prescribed average field:
∫
b
e−Γk [ϕ] ≡ Dϕb δ[ϕ − fk (−∂ 2 )ϕ] b e−S[ϕ]
AA
. (2.32)

Clearly, the (proper) average action ΓAA k is the continuum analog of a Kadanoff–
Wilson block spin action, with ϕb and ϕ playing the role of spin configurations on
the fine and blocked lattice, respectively [7, 9]. Note that the original partition
∫
function Z = Dϕb e−S[ϕ] can be recovered from any of the functionals ΓAA
b
k ,
∫ −Γk
AA[ϕ]
k ∈ [0, ∞), by integrating over the average field, Z = Dϕ e .
In the continuum it is highly non-trivial to make sense of the delta-functional
in (2.32), which, formally, consists of an infinite product of Dirac delta-functions,
one for each point of spacetime. In [67, 68] the authors proposed approxi-
mating those functions by a Gaussian of finite width, whereby the width is
allowed to depend∫ on the momentum carried by the fields. As a result, (2.32)
becomes e−Γk = Dϕb e−S−Sconstraint , with an additional contribution to S, the
AA

“constraint action” Sconstraint . It has the general structure

∫ ( ) ( )
1
Sconstraint = dd x ϕ − fk (−∂ 2 )ϕb K ϕ − fk (−∂ 2 )ϕb . (2.33)
2
Herein K is an operator which controls how precisely the width of the
approximating Gaussian depends on the momentum.
We interpret the formal equation (2.32) as follows now:
∫ d
∫ ∫ d ∫ d
b 1 b b b
e−Γk [ϕ] = e− 2 d x ϕKϕ Dϕb e−S[ϕ]− 2 d x ϕRk ϕ+ d x J ϕ .
AA 1
(2.34)

Here we denoted J ≡ Kfk (−∂ 2 )ϕ, and tentatively set Rk ≡ fk (−∂ 2 )2 K. In fact,
if we assume that K is such that the Rk defined in this way satisfies the require-
ments for a cutoff operator, the functional integral in (2.34) coincides ∫ d exactly
with the one representing Wk in (2.4) above, whence Γk [ϕ] = 2 d x ϕKϕ −
AA 1

Wk [J] J=Kf ϕ . Thus, we observe that (up to an explicitly known term ∝ ϕKϕ)
k
the action ΓAA k [ϕ] is given by the generating functional related to the EAA,
Wk [J], evaluated at a ϕ-dependent source. So in essence, ΓAA k and Γk are related
by a Legendre transformation. [ ]−1
In [68] the choice K = (−∂ 2 ) 1 − fk (−∂ 2 )2 has been advocated. With the
[ ]−1
weight function (2.31) it entails Rk = (−∂ ) exp(−∂ 2 /k 2 ) − 1 , which equals
2

precisely the cutoff operator (2.8) written in position space.

(9) The continuum functional ΓAA k is known as the (proper) Average Action
[67, 68].4 It has been generalized to gauge fields [25, 26] and was employed in a
number of practical calculations [70–75].

4 Its non-derivative part at k = 0 coincides with the so-called constraint effective potential [69].
28 The Functional Renormalization Group

Historically, in the search for a non-perturbative renormalization group in

the continuum, the Average Action preceded the EAA, Γk , reviewed above and
introduced in [22].
In this book we focus almost entirely on the EAA, which is more suitable for
our purposes. Nevertheless, keeping in mind its relation to the “old” average
action helps understanding in what sense precisely Γk is the result of averaging
over increasingly larger volumes in spacetime when k is lowered.

2.2 Theory Space and Its Truncation

The first step toward making the initial value problem of the FRGE precise
consists in fixing the space of functionals over which the equation is supposed
to be defined, and from which eligible initial points ΓΛ are drawn, for example.
It has become customary to refer to this space comprising, in a sense, all action
functionals that are possible in principle, as theory space T . This space contains
bare actions, the standard effective actions, and Effective Average Actions on a
completely equal footing.

2.2.1 The Space of Actions

To illustrate the concept of a theory space, let us return to the above scalar
example and define T by the following requirements:

(i) All functionals A[ · ] ∈ T are Z2 -symmetric, i.e., A[ϕ] = A[−ϕ] for any ϕ, and
invariant under O(d) transformations x′µ = Λµ ν xν , the Euclidean analog of
Lorentz transformations, i.e., A[ϕ] = A[ϕ′ ] for all fields ϕ and ϕ′ related by
the scalar transformation law ϕ′ (x) = ϕ(Λ−1 x).
(ii) The space T contains all functionals that can occur in a derivative expan-
sion, i.e., arbitrary field monomials, integrated over spacetime, consisting of
any number of fields and derivatives of any order which act on them in all
ways possible.

Thus, schematically, elements of theory space would look as follows:

∫ {
A[ϕ] = dd x ū0,0 + ū0,2 ϕ2 (x) + ū0,4 ϕ4 (x) + ū0,6 ϕ6 (x) + · · ·
( )2 ( )2 ( )2
+ ū2,0 ∂µ ϕ + ū2,2 ϕ2 ∂µ ϕ + ū2,4 ϕ4 ∂µ ϕ + · · ·
+ ū4,0 ∂µ ϕ ∂ µ ϕ ∂ν ϕ ∂ ν ϕ + · · ·
..
.
( )m ( )p
+ · · · + ūm,n,p ∂µ ϕ ∂ µ ϕ ϕn ∂ν ∂ ν ϕ + · · ·
.. }
. ... . (2.35)
 2.2 Theory Space and Its Truncation 29

We do not attempt a systematic enumeration of all possible terms here. However,

from (2.35) it should be clear that the generic A ∈ T has an expansion of the form
∑
A[ϕ] = ūα Iα [ϕ]. (2.36)
α

The infinitely many “basis functionals” Iα [ · ] are monomials built from powers of
the field and its derivatives, all evaluated at the same point x, and integrated over
spacetime. The coefficients ūα should be thought of as the “components” of the
functional A[ · ] with respect to the basis {Iα }. The set of coefficients {ūα }, often
referred to as generalized coupling constants can also be seen as local coordinates5
on the “manifold” T . Indeed, there is an obvious one-to-one correspondence
between functionals, regarded usually as maps ϕ 7→ A[ϕ] and sets {ūα }:

T ∋ A ←→ {ūα }. (2.37)

2.2.2 The FRGE in Component Form

Let us return to the FRGE, (2.15). Its solutions, called the renormalization group
trajectories, are one-parameter families of actions Γk of the type (2.35). For every
fixed value of k they can be expanded as in (2.36):
∑
Γk [ϕ] = ūα (k) Iα [ϕ]. (2.38)
α

The scale dependence of Γk is now carried by the running coupling constants,

ūα (k). Note that the basis {Iα } is independent of k. Geometrically speaking, the
RG trajectories are parametrized curves on T .

(1) By virtue of the correspondence (2.37) the abstract form of the FRGE in
(2.15) is equivalent to a coupled system of differential equations for the functions
k 7→ ūα (k). To derive it, we insert the expansion (2.38) into (2.15) and obtain6
the following equation, valid at any field argument ϕ:
[ ]
∑ 1 (∑ )−1
α
k∂k ū (k) Iα [ϕ] = Tr ū (k) Iα [ϕ] + Rk
α (2)
k∂k Rk . (2.39)
α
2 α

While the RHS of this equation is an extremely complicated functional of ϕ, at

least in principle we can apply the standard algorithms for derivative expansions

5 The notation ūα with an overbar is chosen for later convenience. It indicates that the generic
ūα has a non-zero canonical dimension. Later on we will slightly modify the definition of T
and employ a dimensionless variant of the couplings as coordinates. For the time being this
difference is inessential.
6 Here and in the following the superscript (2) denotes the Hessian of the corresponding
functional. We also abbreviate ∂k ≡ ∂/∂k.
30 The Functional Renormalization Group

to it. Since, by its very definition, T contains all functionals that could possibly
[ ]
be produced by those algorithms, it follows that the ϕ-dependence of Tr · · ·
can be expanded in the basis {Iα }. Denoting the corresponding coefficients b̄α ,
(2.39) then reads:
∑ ∑ ( )
k∂k ūα (k) Iα [ϕ] = b̄α ū(k); k Iα [ϕ]. (2.40)
α α

The functions b̄α have both an implicit and explicit k-dependence. The former is
due to their dependence on the set of all running couplings, ū(k) ≡ {ūα (k)}, and
the latter stems from the operator Rk under the trace. Since the Iα ’s are linearly
independent we may equate their coefficients on both sides of (2.40), yielding
( )
k∂k ūα (k) = b̄α ū(k); k ∀α. (2.41)

This is the coupled system of infinitely many ordinary differential equations which
determines the RG trajectories. We will refer to it as the FRGE in component
form and to b̄α as the beta function of the running coupling constant ūα (k).7

(2) The reader might wonder whether the enormous complexity of the system
(2.41), consisting of infinitely many differential equations, is really unavoidable,
and if so, what do we get in return if we really try to come to terms with the
system? After all, in perturbative field theory RG equations are used, too, but
they are way simpler and in particular there are just a few of them [76].
The answer to the first question is indeed affirmative: We really do need the
infinite set of coupled equations in general; with any attempt of reducing it
to some subset we lose information. Intuitively it is quite plausible that the
equations for {ūα (k)} cannot be anything simple since, within the framework of
the EAA, the running couplings parameterize an entire functional, namely Γk .
Now, even if we impose initial conditions at k = Λ and choose ΓΛ as simple as,
say, ∫ {1 1 λΛ 4 }
ΓΛ [ϕ] = d4 x ∂µ ϕ ∂ µ ϕ + m2Λ ϕ2 + ϕ , (2.42)
2 2 4!
the FRGE tells us that, at any lower scale k < Λ, the EAA will look as complicated
as the functional in (2.35). Integrating out the field modes with |p| ∈ [k, Λ] gener-
ates a plethora of complicated terms (with higher derivatives and higher powers
of ϕ) that were not present initially. This is actually as it must be since, by con-
struction, the running functional Γk approaches the ordinary effective action Γ in
the limit k → 0, and from standard quantum field theory we know that a typical
effective action is anything but simple. Usually Γ will contain all sorts of “exotic”
terms which never appear in the bare action of a perturbatively renormalizable

7 We reserve the standard notation β α for the beta function of the dimensionless coupling to
be introduced later.
 2.2 Theory Space and Its Truncation 31

theory; a typical example in the scalar model at hand is the Coleman–Weinberg

potential, ϕ4 ln(ϕ) [77].
When applied to one of the familiar, perturbatively renormalizable theories,
the EAA trajectory k 7→ Γk may be regarded as a smooth interpolation between
the bare action at k = Λ → ∞, and the effective one at k = 0. Hence, the RG evo-
lution of the EAA describes how the complexity of Γ emerges out of the simplicity
of S.

(3) The effective action Γ is a concise way of summarizing all predictions of a

field theory model. Thus, loosely speaking, computing Γ amounts to completely
“solving” this model. This answers the second question above: If we manage to
find the solution to the FRGE subject to the initial condition ΓΛ = S we have
derived all possible predictions of the model, i.e., we have “solved” this theory.
It is clear therefore that the FRGE should have a level of complexity comparable
to that of the functional integral we started from.
This property of the FRGE, providing a complete “solution of the theory”,
is in sharp contradistinction to the various types of RG equations used in per-
turbation theory [78]. They are finite in number and control only the coupling
constants related to “renormalizable interactions.” Therefore, the amount of
information encapsulated in the solutions to their RG equations is insufficient
to fully determine Γ.

2.2.3 Truncations of Theory Space

Up to this point, our construction did not involve any approximation. However,
when it comes to actually solving the FRGE or the system (2.41) we clearly
must resort to some sort of approximation. An obvious possibility is perturba-
tion theory in one or several couplings which are assumed small; in this way
one could recover the RG equations of perturbative renormalization theory, for
instance [79].
Here we will adopt a different strategy, the method of truncated theory spaces,
which has the essential advantage that it can yield non-perturbative approximate
solutions. They can go beyond the realm of perturbation theory by summing
up contributions from all orders of the small coupling expansion or the loop
expansion, for instance. Such solutions might even depend on the small param-
eters in which perturbation theory is trying to expand in a non-analytical way
so that at any finite order perturbation theory is “blind” to the corresponding
contributions.
The basic idea is to project the exact RG evolution on T as described by (2.41)
onto a simpler, often finite dimensional subspace Ttrunc ⊂ T . Ideally, the truncated
theory space, Ttrunc , should be chosen such that, on the one side, it is general
enough for the projected evolution to retain the essential physical features of its
32 The Functional Renormalization Group

exact precursor, and on the other side, is simple enough to make the necessary
calculations technically feasible.
In the simplest case, the projection from T onto a, say, N -dimensional subspace
Ttrunc proceeds as follows. To (partly) specify a truncation, we make an ansatz
for the functionals in the subspace:

∑
N
Γk [ϕ] = ūi (k) Ii [ϕ]. (2.43)
i=1

Here {Ii [ · ], i = 1, · · · , N } is a certain subset of the full basis {Iα [ · ]}, which is
selected such that it spans the subspace chosen. Inserting (2.43) into the exact
FRGE we are confronted with the following relation:
 
∑N ( ∑N ) −1
? 1
k∂k ūj (k) Ij [ϕ] = Tr  k∂k Rk  .
(2)
ūj (k) Ij [ϕ] + Rk (2.44)
j=1
2 j=1

As it stands, this equation will have no solution for the N functions ūj (k) in
general. The reason is that even though only[ the] basis elements Ij spanning
Ttrunc appear under the functional trace Tr · · · , its derivative expansion will
generate terms outside Ttrunc as well:

1 [ ] ∑N ({ } )
Tr · · · = b̄i ūj j =1 , ··· , N ; k Ii [ϕ]
2 i=1
∑ ({ } )
+ b̄α ūj j =1 , ··· , N ; k Iα [ϕ]. (2.45)
α, Iα ∈T
/ trunc

Typically such terms are generated by expanding the propagator in terms of

background quantities. Upon inserting (2.45) into (2.44), those terms on the
RHS of the resulting equation which involve basis elements Iα ∈ / Ttrunc have no
counterparts on its LHS.
At this point an additional assumption is needed in order to render the system
of equations consistent. One assumes that it is legitimate to replace the prefactors
of all invariants outside Ttrunc by zero:

b̄α 7→ 0 for all α such that Iα ∈

/ Ttrunc . (2.46)

With (2.43) and (2.46) the FRGE boils down to a system of N coupled
equations:
( )
k∂k ūj (k) = b̄j ū1 (k), · · · , ūN (k); k ∀j = 1, · · · , N. (2.47)

At least when N is a (small) finite number one can investigate this system with
standard methods and try to justify the assumption underlying the truncation.
 2.2 Theory Space and Its Truncation 33

Several comments are in order here.

(i) The projection. The rule (2.46) describes what we referred to as the
“projection” of the RG dynamics from T onto Ttrunc ; together with the
specification of Ttrunc in (2.43) it defines a certain truncation of theory space.
As for specifying a truncation uniquely there is a subtlety which is
overlooked sometimes and at first sight might seem surprising perhaps.
Namely, the beta functions for the couplings retained, ūj , depend not only
on the choice of a basis on Ttrunc , but also on how the other basis elements
Iα ∈/ Ttrunc are chosen even though their coupling constants are set to zero
by the truncation ansatz.
To see the reason note that theory space in general does not carry a natural
inner product that would define, in particular, a notion of orthogonality
and orthogonal projections. While the restricted setting of the derivative
expansion adopted here does equip T with the structure of a vector space,
it still does not supply a natural inner product.
As a consequence, given an arbitrary point P ∈ T , which does not happen
to lie on Ttrunc , it is the matter of an explicit definition to specify its image
under the projection, Pproj ∈ Ttrunc . The choice of a projection map has the
same conceptual status as the choice of a subspace, Ttrunc , but is logically
independent from it.
A projection is conveniently specified in terms of an adapted basis on T :
a subset of its basis vectors spans Ttrunc , and we define the projection to
be parallel to all those basis vectors that are not tangent to Ttrunc . With
other words, the coordinates of Pproj ∈ Ttrunc are obtained by setting to zero
all coordinates of its preimage P ∈ T except those related to the subbasis
spanning Ttrunc . This is exactly what is done in (2.46).
In Figure 2.2 this is illustrated for the simple example of the one-dimen-
sional truncation of a 2D theory space. In Figure 2.2a, the subspace Ttrunc
is fixed by picking a single basis vector on the plane; it defines the “u1 -
direction” and identifies Ttrunc with the “u1 -axis”. No notion of a projection
is available yet. In Figure 2.2b a second basis vector is selected which gives
rise to the “u2 -axis”. The projection T → Ttrunc , P 7→ Pproj , is defined to be

u2
P P
? ?
?
trunc trunc
u1 u1
Pproj
(a) (b)

Figure 2.2. A 1-dimensional truncation of the 2-dimensional plane. (a) The

definition of a first basis vector specifies the truncation subspace. (b) The
definition of a second basis vector defines a projection onto this subspace.
34 The Functional Renormalization Group

parallel to this axis. It is clear then that the location of Pproj on Ttrunc , i.e.,
its u1 -coordinate, depends on our choice of the u2 -axis.
(ii) Projected solution vs. solution of the projected equation. Let k 7→
Γk be an exact solution of the FRGE on the untruncated theory space T .
Furthermore, let γ be the set of all points visited by this trajectory. We can
visualize it as an oriented curve on T , as shown in Figure 2.3.
Now we choose a truncation by specifying a subspace Ttrunc plus a pro-
jection. By applying the latter to the exact RG trajectory γ we obtain its
projection in Ttrunc , which we denote by γproj .
The truncation chosen gives rise to an approximate RG equation of the
type (2.47). Let us solve this approximate equation subject to the condition
that it passes through a point P0 ∈ Ttrunc , which is also visited by the exact
γ. We denote this resulting (exact!) solution to the approximate RG equation
by γtrunc .
We emphasize that, typically, γ, γproj , and γtrunc , are three distinct curves.
While γ can explore the full theory space in principle, both γproj and γtrunc
“live” on Ttrunc by definition. In general, γproj and γtrunc are different from
one another, and this is of course what makes the truncation, at best, an
approximation.
The trajectories γ and γproj are both exact, the only difference being
that γproj represents only partial information about the running coupling
constants, namely only about those which coordinatize Ttrunc . The third

u2

u3

P0
gproj gtrunc

u1

Figure 2.3. A 2-dimensional truncation (u1 –u2 -plane) of a 3-dimensional

space. The exact trajectory γ, its projection γproj , and γtrunc pass through
a common point P0 on Ttrunc . While γ gradually departs from Ttrunc , γproj and
γtrunc stay close to each other if the truncation is reliable.
 2.2 Theory Space and Its Truncation 35

curve, γtrunc , is an approximation (to γproj ) since it is the solution of an

incomplete system of differential equations which is ignorant of the exact
form of the RG dynamics; see Figure 2.3 for an illustration.
Thus, a reliable truncation is characterized by an approximate equality
of γproj and γtrunc , at least within a certain interval of scales k. It is in
general a difficult but clearly important task to judge the reliability of a
given truncation. We will address this issue repeatedly later on.

2.2.4 The Local Potential Approximation

The local potential approximation (LPA) is a prototypical example of a trunca-
tion of the scalar theory space discussed above. The “points” A[ · ] ∈ T have the
general structure indicated in (2.35). The LPA deals with an infinite-dimensional
subspace Ttrunc . It retains from (2.35) all the non-derivative terms of the first line,
( )2
and the first term in the second, i.e., ū2,0 ∂µ ϕ . Thus, the generic A[ · ] ∈ Ttrunc
has the form
∫ { }
A[ϕ] = dd x ū2,0 ∂µ ϕ(x) ∂ µ ϕ(x) + U (ϕ(x)) . (2.48)

Here all possible zero-derivative terms are collected in the potential function
U ( · ).
The projection onto Ttrunc is defined in the spirit of the derivative expansion:
The projected functional Aproj ∈ Ttrunc related to a general A ∈ T is obtained
by inserting for ϕ a plane wave, expanding A[ϕ] in powers of its amplitude and
momentum, and setting to zero all terms except those of zeroth order in the
momentum (and any order in the amplitude), and those of second order in both
the momentum and the amplitude.
Thus, the general form of the EAA on the subspace reads
∫ {1 }
Γk [ϕ] = dd x Zk ∂µ ϕ(x) ∂ µ ϕ(x) + Uk (ϕ(x)) . (2.49)
2
Here Uk denotes the scale-dependent effective average potential, and we also
introduced the customary notation Zk ≡ 2ū2,0 (k). Actually the ansatz (2.49) with
a running prefactor of the kinetic term, Zk , amounts already to an extended
version of the LPA. The LPA proper, on which we will focus in the following,
makes the additional assumption that the k-dependence of Zk is actually weak,
and it sets Zk ≡ 1 in the ansatz (2.49). As a result, the only k-dependence that has
to be tracked in the (projected) FRGE is that of the running potential function
Uk ( · ).
To derive a differential equation for Uk , we insert (2.49) with Zk ≡ 1 into the
exact form of the FRGE, (2.15), and obtain in a first step:
∫ [( )−1 ]
1 (2)
d x k∂k Uk (ϕ(x)) = Tr Γk [ϕ] + Rk
d
k∂k Rk . (2.50)
2 proj
36 The Functional Renormalization Group

Here we indicated symbolically that the RHS of (2.50), regarded as a functional

of ϕ, must be projected onto Ttrunc ultimately. The Hessian of the truncation
ansatz, to be inserted on the RHS of (2.50), is given by the second functional
derivatives of (2.49):
δ 2 Γk [ϕ] [ ]
= −∂ 2 + Uk′′ (ϕ(x)) δ(x − y). (2.51)
δϕ(x)δϕ(y)
A prime will always denote the derivative with respect to the argument, so here
Uk′′ ≡ ∂ 2 Uk /∂ϕ2 . In the following we interpret Γk as the ϕ-dependent differential
(2)

operator acting on the δ-distribution in (2.51):

Γk [ϕ] = −∂ 2 + Uk′′ (ϕ(x)).
(2)
(2.52)
To set up the RG equation for Uk ( · ), we first introduce a cutoff function
according to the prescription (2.7). Subsequently we perform the projection T →
Ttrunc , A 7→ Aproj of the special functional that is defined by the RHS of (2.50)
with (2.52):
[( )−1 ]
1 ′′
A[ϕ] ≡ Tr −∂ + Rk (−∂ ) + Uk (ϕ(x))
2 2
k∂k Rk (−∂ ) .
2
(2.53)
2
This calculation proceeds as follows. ( )2
The functionals in Ttrunc are space integrals over Zk ∂µ ϕ plus all possible
non-derivative terms built from ϕ(x). Since the standard LPA makes the addi-
tional approximation Zk ≡ 1 we actually
( )2 do not need to know the component of
the projection Aproj [ · ] along the ∂µ ϕ -direction of Ttrunc , so it suffices to find
the non-derivative terms of Aproj [ · ]. They are precisely those which, upon insert-
ing a plane wave into A[ · ], are of zeroth order in the plane wave’s momentum.
Hence, it is convenient to insert a constant field configuration ϕ(x) = const ≡ C
since it sets to zero the uninteresting terms automatically. As a consequence,
the operator under the trace (2.53) loses its explicit x-dependence and is easily
diagonalized in the momentum eigen-basis of the Laplacian on Rd . Thus, for any
real value C of the constant field:
∫ ∫
1 d dd p k∂k Rk (p2 )
A[C] = d x . (2.54)
2 (2π)d p2 + Rk (p2 ) + Uk′′ (C)
Furthermore, it follows that Aproj [C] = A[C], since the original functional A ∈ T
and its projection Aproj ∈ Ttrunc agree on constant functions.
Let us emphasize at this point that usually Aproj and A are assumed to have
the same domain, i.e., Aproj [ϕ] and A[ϕ] are defined over the same space of
ϕ’s. So setting ϕ = const for projection purposes is not meant to imply that
Aproj is defined for constant fields only. Rather, the correct interpretation is that
Aproj [ϕ( · )] for any, in general non-constant ϕ ≡ ϕ(x), has the same structure
as Aproj [C], but with C replaced by ϕ(x):
∫
( )2
Aproj [ϕ( · )] = A[C] + ak dd x ∂µ ϕ . (2.55)
C→ϕ(x)
 2.2 Theory Space and Its Truncation 37

( )2
For completeness we also added the ∂µ ϕ -component of the projection, so that
(2.55) covers all terms contained in the ansatz (2.49). Since the scale dependence
of Zk will not be considered in the sequel, we will not compute the coefficient ak
here and refer to Appendix A of [80] for the details of its derivation.
In fact, the assumption Zk ≡ 1, implying ∂k Zk = 0, amounts to assuming that
ak is small and we may replace ak → 0. This becomes obvious when we insert
(2.55) into the FRGE (2.50): Comparing the terms under the space integral,
( )2
we see that the term ak ∂µ ϕ on the RHS has no counterpart on the LHS to
match with since ∂k Zk = 0 was used there; thus consistency is achieved if ak is
negligibly small.8
Furthermore, equating the non-derivative terms under the x-integral yields the
desired relation for the potential:
∫
1 dd p k∂k Rk (p2 )
k∂k Uk (ϕ) = . (2.56)
2 (2π) p + Rk (p2 ) + Uk′′ (ϕ)
d 2

At this point it has become immaterial whether ϕ is a constant number or a

function of x: (2.56) is a partial differential equation involving a first-order scale
and a second-order ϕ-derivative, which makes no reference to spacetime points.
Still with a general cutoff function Rk , we can simplify (2.56) by performing
the angular part of the momentum integration. The relevant identity, valid for
any function F with suitable falloff behavior, reads
∫ ∫ ∞
1 dd p
dw w 2 −1 F (w)
d
µ
d
F (pµ p ) = vd (2.57)
2 (2π) 0

with the dimension-dependent constant

[ d ]−1
vd ≡ 2 (4π) 2 Γ( d2 ) . (2.58)
1
Special values of vd include v2 = 8π , v3 = 8π1 2 , v4 = 32π
1
2.

This yields the following projected flow equation:

∫ ∞
k∂k Rk (w)
dw w 2 −1
d
k∂k Uk (ϕ) = vd . (2.59)
0 w + Rk (w) + ∂ϕ2 Uk (ϕ)

The partial differential equation (2.59) is the main result of the LPA. It has a
number of rather typical properties which we shall re-encounter frequently later
on in quantum gravity, albeit in a far more complicated setting. Next, we discuss
some of them.

(1) Rk dependence. The RG equation (2.59) depends manifestly on the cutoff

function chosen, and the same is true for most properties of its solutions. Proper-
ties of, or quantities derived from, a solution which are equal for all functions Rk

8 For a detailed discussion of the circumstances under which this is the case we must refer to
the literature [31].
38 The Functional Renormalization Group

with the asymptotic behavior (2.6) are called universal. For instance, observables
must be universal quantities in this sense.

(2) Explicit form. The w-integration in (2.59) can be performed analytically

for a number of special choices of Rk , including, for example, the “optimized”
cutoff function [63],
Rk (p2 ) = (k 2 − p2 ) Θ(k 2 − p2 ), (2.60)
which gives rise to the partial differential equation

4vd k d+2
k∂k Uk (ϕ) = . (2.61)
d k + ∂ϕ2 Uk (ϕ)
2

While this equation is of a non-standard form, a number of techniques have been

developed to analyze and solve it. This includes, in particular, numerical [31] or
pseudospectral methods [81, 82].

(3) Approaching convexity. Since Γk equals the usual effective action Γ in the
limit k → 0, the non-derivative part of the former, Uk (ϕ), should approach the
non-derivative part of the latter, the familiar effective potential [59]: Veff (ϕ) =
limk→0 Uk (ϕ).
Furthermore, by its very definition Γ is given by a Legendre transform, and
the same is true for Veff , by restriction to constant fields. Standard properties of
the Legendre transformation [66] thus imply that Γ is a convex functional, and
′′
Veff a convex function, Veff (ϕ) ≥ 0.
This should be contrasted with Γk and Uk at non-zero scales k > 0 where the
∆Sk -term in (2.12) prevents them from being pure Legendre transforms. As a
consequence, they have no reason to be convex as long as k > 0, but they must
become so in the limit k → 0. Indeed, in this limit ∆Sk [ϕ] vanishes.
While these properties apply to the exact EAA and its likewise exact non-
derivative part, they can easily get ruined by inadequate approximations. Con-
vexity poses a notorious problem in perturbation theory (loop expansion); see, for
instance, [59]. Therefore, it is even more remarkable that the LPA flow equation
(2.61) is powerful enough to correctly describe the approach to convexity.
This is demonstrated in Figure 2.4, which shows the effective average potential
Uk (ϕ) at various scales. It is obtained by numerically integrating (2.61) “down-
ward”, i.e., toward decreasing values of k. The initial potential imposed at k = Λ
was chosen to be “W-shaped”: UΛ (ϕ) = −c1 ϕ2 + c2 ϕ4 with c1,2 > 0. When k is
lowered at first the two minima of the “W” move inward, then come to a halt at
a certain value ±ϕmin , and the non-convex part of the curve flattens more and
more, until it finally approaches a function U0 (ϕ) which is indeed convex, and
constant for ϕ between −ϕmin and +ϕmin .
This is, in fact, precisely what one expects for the exact effective potential in
the phase where the Z2 symmetry under ϕ → −ϕ is spontaneously broken; see
[69] and [31] for a detailed discussion.
 2.2 Theory Space and Its Truncation 39

0.15

0.10

0.05

U¢ (r)
0.00

–0.05
–0.10
0.0 0.1 0.2 0.3 0.4 0.5
r

Figure 2.4. Plot of the first derivative effective average potential, U ′ (ρ) as a
function of ρ = 12 ϕ2 in d = 3. Moving from the bottom to the top, the curves are
obtained for decreasing RG times ln(k/Λ) = 0, −0.5, −1, −1.5, −1.7, −2, −2.1.
Starting from UΛ = −0.05ϕ2 + 0.0625ϕ4 , the potential turns convex for
decreasing RG scale k. (Obtained by solving the LPA equation via pseudospec-
tral methods [82].)

(4) Dimensionless variables. The example of (2.61) illustrates that the dif-
ferential equation for Uk (ϕ) has an explicit dependence on the scale, k. It can be
removed though, by re-expressing it in terms of a dimensionless potential and a
dimensionless field variable, which are obtained from the corresponding dimen-
sionful quantities by multiplying them with appropriate powers of the cutoff scale.
Assigning the canonical mass dimensions, in d spacetime dimensions,

[xµ ] = −1, [∂µ ] = +1, [k] = +1,

[S] = [Γ] = [Γk ] = 0 =⇒ [Uk ] = d, [ϕ] = d−2
2 , (2.62)

the dimensionless field and its potential are given by, respectively,

φ ≡ k−
d−2
2 ϕ
−d
( d−2 )
uk (φ) ≡ k Uk k 2 φ ≡ k −d Uk (ϕ). (2.63)

Furthermore, we introduce a dimensionless version of Rk , namely the shape

function R(0) :
2
Rk (p2 ) = k 2 R(0) ( kp2 ). (2.64)

By virtue of (2.6), R(0) is an essentially arbitrary dimensionless function, depend-

ing on a dimensionless argument z ≡ p2 /k 2 , and required to comply with the
asymptotics:

R(0) (z) −→ 0 for z → ∞, and R(0) (z) −→ 1 for z → 0, (2.65)

whereby the transition between the two regimes occurs near z ≈ 1.

We also employ the dimensionless renormalization group “time”:

t ≡ ln(k/k0 ), (2.66)
40 The Functional Renormalization Group

which takes over the role of k as the evolution parameter. (The fixed mass
parameter k0 just defines the units in which k is measured.)
In terms of the dimensionless variables, the general LPA equation is
1
∂t uk (φ) + duk (φ) − (d − 2) φ u′k (φ)
2
∫ ∞ (2.67)
R(0) (z) − zR(0)′ (z)
dz z 2 −1
d
= 2vd .
0 z + R(0) (z) + u′′k (φ)

The example (2.61) which employs the cutoff (2.60), or equivalently the shape
function
R(0) (z) = (1 − z) Θ(1 − z), (2.68)
assumes the following form now:

1 4vd 1
∂t uk (φ) + duk (φ) − (d − 2) φ u′k (φ) = . (2.69)
2 d 1 + u′′k (φ)

We see that (2.67) and (2.69) are indeed autonomous equations, i.e., they have
no explicit dependence on k, or t, any longer. (While the running potential is
actually a function of t now, it is customary to keep using the notation uk .)

(5) Fixed points. As for solutions uk (φ) to the dimensionless LPA equation,
a particularly interesting class are the fixed point solutions which satisfy (2.67)
with ∂t uk (φ) = 0. These special potentials will be denoted u∗ (φ). Picking the
shape function (2.68) they satisfy the following ordinary differential equation, if
they exist:
1 4vd 1
du∗ (φ) − (d − 2) φ u′∗ (φ) = . (2.70)
2 d 1 + u′′∗ (φ)

There is one obvious solution to this equation which exists in any number of
spacetime dimensions, namely the trivial, or Gaussian fixed point (GFP) with the
φ-independent potential u∗ (φ) = 4v d2 = const. This fixed point is termed “Gaus-
d

sian” since the only important

∫( field
)2 monomial in the corresponding action is the
1
quadratic kinetic term 2 ∂µ ϕ , and this makes the functional integral over ϕ
a Gaussian one.
Concerning non-trivial, or “non-Gaussian,” fixed points with potentials that
actually depend on φ the situation strongly depends on the dimensionality of
spacetime: There are infinitely many such fixed points in two dimensions, there
is one (the Wilson–Fisher fixed point) in three dimensions, but none is known to
exist in four dimensions. ( d−2 )
Note that the dimensionful potential Uk (ϕ) ≡ k d uk k − 2 ϕ related to a fixed
point u∗ is by no means scale independent. It “runs” in a particularly simple way:
( d−2 )
Uk (ϕ) = k d u∗ k − 2 ϕ . (2.71)
 2.2 Theory Space and Its Truncation 41

We will re-encounter this kind of behavior in the case of quantum gravity

later on.
Fixed points in scalar systems have been investigated by many authors, using
a variety of methods. This is an important topic in its own right and which is
beyond the scope of this book though. The interested reader is referred to the
pedagogical introduction in [66] and the comprehensive list of references therein.

(6) Power series expansion. In (2.59) or (2.61) the LPA equation for Uk (ϕ)
has the appearance of a partial differential equation in two variables, k and ϕ.
We can obtain an alternative representation in “component form” by expanding
the potential in a power series:
∞
∑
Uk (ϕ) = λ̄2n (k) ϕ2n . (2.72)
n=0

In this manner the partial differential equation becomes a system of ordinary

ones for the expansion coefficients λ̄2n ≡ ū0,2n :
( )
k∂k λ̄2n (k) = b̄2n λ̄0 (k), λ̄2 (k), · · · ; k . (2.73)

Since
[λ̄2n ] = [b̄2n ] = d − (d − 2)n, (2.74)

the corresponding dimensionless coupling constants are

λ2n (k) = k (d−2)n−d λ̄2n (k). (2.75)

Their RG equations read

( )
∂t λ2n (k) = β2n λ0 (k), λ2 (k), λ4 (k), · · · (2.76)

with the dimensionless beta functions

( ) [ ]
β2n λ0 , λ2 , · · · ≡ (d − 2)n − d λ2n + k (d−2)n−d b̄2n , (2.77)

where the λ̄′ s in the argument of b̄2n are now to be expressed in terms of their
dimensionless pendants.
The functions (2.77) have a structure which is typical of all β-functions
we will encounter in this book. They consist of two parts: First, a classical,
or canonical term linear in the respective coupling which owes its existence
to using the variable scale k as the unit in which all dimensionful quantities
are expressed, and second, a dedimensionalized b̄-function; the latter describes
the true renormalization effects that are caused by the statistical or quantum
fluctuations.
42 The Functional Renormalization Group

The explicit form of the b̄′2n s is found by inserting (2.72) into (2.59), expanding
in ϕ2 , and comparing equal powers of ϕ2 . The structure of the first few examples
looks as follows:

b̄0 = J1 , (2.78)
b̄2 = −12 λ̄4 J2 , (2.79)
b̄4 = −30 λ̄6 J2 + 144 (λ̄4 ) J3 ,
2
(2.80)
b̄6 = −56 λ̄8 J2 + 720 λ̄4 λ̄6 J3 − 1728 (λ̄4 ) J4 , 3
(2.81)
b̄8 = −90 λ̄10 J2 + 1344 λ̄4 λ̄8 J3 + 900 (λ̄6 ) J3 2

−12960 λ̄6 (λ̄4 )2 J4 + 20736 (λ̄4 )4 J5 , (2.82)

b̄10 = · · · . (2.83)

Here the J ′ s are a family of convergent momentum integrals. They depend on

λ̄2 = k 2 λ2 , but on no further coupling constants:
∫ ∞
k∂k Rk (w)
dw w 2 −1 [
d
Jn ≡ vd ]n
0 w + Rk (w) + 2λ̄2
∫ ∞
R(0) (z) − zR(0)′ (z)
dz z 2 −1 [
d
= 2 vd k d+2−2n
]n . (2.84)
0 z + R(0) (z) + 2λ2

Generically these integrals depend on the cutoff function in a non-trivial way.

Exceptions occur, however, at λ2 = 0 if n = (d + 2)/2. The integrand turns out
to be a total derivative then, and it is sufficient to exploit that R(0) (0) = 1 and
R(0) (∞) = 0 in order to evaluate the integral:
∫ ∞
R(0) (z) − zR(0)′ (z)
dz z 2 −1 [
d
J d+2 = 2 vd ] d+2
2 λ2 =0 0 z + R(0) (z) 2
 
∫ ∞ d
d 2 z2 
= 2 vd dz
0 dz d [z + R(0) (z)] d2
[( ) d2 ]∞
4vd z
=
d z + R(0) (z)
0
4vd
= . (2.85)
d
The coefficients J d+2 at λ2 = 0 are examples of universal quantities which are
2
insensitive to the details of the cutoff function. They are dimensionless, pure
numbers.
It should be stressed, however, that universality is rather the exception than
the rule; typical coefficients in b̄-, or β-functions, are indeed sensitive to the
full profile of the function R(0) (z). This dependence on unphysical features of
the regulator makes it quite clear that, in the EAA approach, the majority of
 2.2 Theory Space and Its Truncation 43

the beta functions and running couplings cannot have an immediate physical
meaning per se. In general, the cutoff scheme dependence will drop out only at
a late stage in the calculation of observables. Up to this point a concrete shape
function must be employed, but any such function with the correct asymptotics
will do the job.
With (2.68), for example, the above coefficients Jn for arbitrary n and d, and
any value of λ2 , are explicitly given by
( 1 )n
Jn = 4 vd d−1 k d+2−2n . (2.86)
1 + 2λ2

Because of the general structure of the b̄-functions in (2.78)–(2.83), we observe

that the hierarchy of equations k∂k λ̄2n = b̄2n does not close at any finite order.
The scale derivative of a given coupling constant, apart from its own value,
depends on couplings which are both lower and higher( up in the) hierarchy. In
(2.82), for instance, it is seen that k∂k λ̄8 depends on λ̄2 , λ̄4 , λ̄6 , on λ̄8 itself,
but also on λ̄10 . Thus, we cannot hope to easily find an exact solution to these
equations.

(7) Polynomial truncations. Under certain conditions it can be legitimate to

use a truncation which is simpler than the full-fledged LPA and to replace the
∑N
power series (2.72) by a polynomial, Uk (ϕ) = n=0 λ̄2n (k) ϕ2n , N < ∞.
Note that trying to set λ̄2n (k) = 0 for n > N at all scales assumes not only that
the couplings thus discarded are negligibly small; it also imposes constraints
on the couplings retained, namely that b̄2n ≈ 0 for n > N , and this also con-
strains λ̄′2n s at n ≤ N . This latter requirement must be met for consistency since
otherwise a coupling that is small at one scale could grow large at another.
If λ̄2n and b̄2n can be replaced by zero for n > N , the hierarchy indeed collapses
to its first N + 1 equations.
As an example, let us consider the case N = 2, i.e., a quadratic polynomial:

Uk (ϕ) = λ̄0 (k) + λ̄2 (k)ϕ2 + λ̄4 (k)ϕ4 . (2.87)

Now, according to (2.81), there are three terms contributing to b̄6 . The first two
of them, containing λ̄8 and λ̄6 , do indeed vanish upon setting the higher couplings
to zero, but not so the third term ∝ (λ̄4 )3 J4 . Demanding that it can be neglected
obviously implies conditions on λ̄4 and λ̄2 . Once a solution to the RG equations
of the polynomial truncation has been found one must check a posteriori whether
or not this solution does indeed meet these conditions.
With the ansatz (2.87), what remains are the three equations:

k∂k λ̄0 = J1 , (2.88)

k∂k λ̄2 = −12 λ̄4 J2 , (2.89)
k∂k λ̄4 = 144 (λ̄4 ) J3 .
2
(2.90)
44 The Functional Renormalization Group

Noting that the J ′ s depend on λ̄2 we see that the equations for λ̄2 and λ̄4 are
indeed coupled, while the equation for λ̄0 stands alone and can be integrated
straightforwardly once λ̄2 (k) is known.
Let us stick to d = 4 now; λ̄0 and λ̄2 have canonical mass dimension 4 and 2
then, and λ̄4 ≡ λ4 is dimensionless, hence β4 ≡ b̄4 . Let us furthermore specialize
1
for the regime where the running mass, i.e., (2λ̄2 ) 2 , is much smaller than the
cutoff:
λ̄2 ≪ k 2 ⇐⇒ λ2 ≪ 1. (2.91)
Then letting λ̄2 → 0 in J3 , the equation for the quartic coupling λ4 gets separated
from that of the mass, and the coefficient in its β-function happens to be one of
the universal quantities (2.85), namely J3 (0) = v4 = 32π 1
2 . As a consequence, the

resulting RG equation,
18 2
k∂k λ4 = β4 (λ4 ) = λ , (2.92)
(2π)2 4
is seen to have the same form and to contain the same numerical constant as the
familiar result for the RG equation in one-loop perturbation theory. (Usually the
latter is derived using dimensional regularization (minimal subtraction) and k is
to be identified with the mass scale µ, see [76].)
Again, the coincidence of some b̄-function descending from the FRGE with a
perturbative β-function is a very rare exception. In general, this happens only
in the case of universal coefficients. In fact, it takes as little as going beyond
the regime of small masses, allowing for arbitrary λ2 , to obtain a R(0) -dependent
answer which has no direct analog in perturbation theory. And, what is even more
important, the superficial similarity of the three differential equations (2.88)–
(2.90) with the perturbative RG equation of ϕ4 -theory arises only because we
employed a rather severe approximation of the full FRGE, namely the LPA and
on top of it the restriction to a quartic potential.
Adopting the special shape function (2.68), equation (2.86) provides us with
the mass dependence of β4 :
18 1
β4 (λ2 , λ4 ) = λ2 . (2.93)
(2π) (1 + 2λ2 )3 4
2

In comparison with the formula for λ2 = 0 the new feature of this β-function is
the factor { 1
1 k2 1 if k ≫ |2λ̄2 | 2
= = 1 (2.94)
1 + 2λ2 k 2 + 2λ̄2 0 if k ≪ |2λ̄2 | 2 .
This factor makes our result (2.93) powerful enough to quantitatively describe
the phenomenon known as the decoupling of heavy modes.9
Equation (2.93) displays a threshold at the (mass)2 -scale set by 2λ̄2 (k). For k
values well above the threshold, β4 is the same function as for a massless field, but

9 See [83] for its description within perturbation theory and [84] for the example of QED.
 2.3 Decoupling 45

it approaches zero when the cutoff scale falls below the threshold. As a result,
when we integrate the RG equations with (2.93) for decreasing k, the quartic
coupling stops running; it “decouples” when k approaches the threshold from
above: ∂k λ4 (k) → 0. In the following we will see how we can take advantage of
this decoupling phenomenon as a powerful computational tool.
A more detailed discussion of the solutions to the scalar RG equations would
lead us too far afield, and thus we refer to the literature here [31].

2.2.5 Adjusted Cutoffs

The first step in going beyond the local potential approximation consists in giving
up the restriction Zk ≡ 1, thus allowing for a running wave function normalization
constant Zk in (2.49). Using truncations of this type one should employ a slightly
different normalization of the cutoff function, such that Rk (p2 ) = Zk k 2 if p2 ≪ k 2 .
(2)
As a result, Rk combines in the FRGE with Γk to the inverse propagator
(2)
Γk + Rk = Zk (p2 + k 2 ) + · · · , which is valid for low momentum modes. This is
precisely as it should be if we want the IR cutoff to give a squared mass of the
size k 2 , rather than k 2 /Zk , to the low momentum modes.
This adjustment is particularly important in more complicated systems with
more than one field. Typically their kinetic terms will contain different Zk -factors,
and so we must make sure that all fields are cut off at precisely the same scale,
namely k 2 . This is achieved by a cutoff of the type
2
Rk (p2 ) = Zk k 2 R(0) ( kp2 ). (2.95)

As before, R(0) is a shape function satisfying the normalized boundary conditions

R(0) (0) = 1 and R(0) (∞) = 0. Furthermore, Zk is a matrix in field space which is
adjusted to the normalization of the various fields in such a way that they are
all cut off at p2 = k 2 . If the field modes of type “i” have an inverse propagator
(i) (i)
Zk p2 + · · · and the modes do not mix, Zk is a diagonal matrix with Zk = Zk
in the sector of the “i” modes.

2.3 Decoupling
In this section we return to the decoupling mechanism which we encountered
earlier already and demonstrate how it can be exploited in order to deduce the
presence of certain terms in the EAA without actually going through the explicit
EAA-based calculation [70, 85].
For the sake of simplicity we employ again the example of a single real scalar
field on a flat 4-dimensional space whose Effective Average Action is a solution
to the flow equation
[( )−1 ]
1 (2)
k∂k Γk [ϕ] = Tr Γk [ϕ] + Rk k∂k Rk . (2.96)
2
46 The Functional Renormalization Group

For the purposes of the present discussion it is sufficient to work at the level of
the LPA ansatz for Γk . After a slight simplification of the notation it reads:
∫ {1 }
1 1
Γk [ϕ] = d4 x Z(k)∂µ ϕ ∂ µ ϕ + m̄2 (k)ϕ2 + λ(k)ϕ4 + · · · . (2.97)
2 2 12

To start with, we neglect the running of the kinetic term and approximate Z(k) ≡
1. For functionals of this type, and in a momentum basis where −∂ 2 ≡ p2 , the
denominator appearing under the trace of (2.96) reads

(2)
Γk + Rk = p2 + m̄2 (k) + k 2 + λ(k)ϕ2 + · · · . (2.98)

To make the discussion as transparent as possible we inserted a mass-type cut-

off Rk = k 2 here even though the corresponding profile function is not strictly
admissible in the sense of the requirements (2.65).10
In a diagrammatic loop calculation of Γk the inverse of (2.98) appears as the
effective propagator in all loops. It contains the IR cutoff at the scale k. For
the specific choice (2.98) this adds the k 2 -term to the running mass parameter
m̄2 (k). Therefore, the pµ -modes (plane waves) are integrated out fully only in
the domain p2 & m̄2 + k 2 + λϕ2 + · · · . In the opposite case, all loop contributions
are suppressed because of the non-zero effective mass square given by m̄2 + k 2 +
λϕ2 + · · · . It receives contributions from both the “artificial” cutoff k 2 introduced
by hand and the “physical” cutoff terms m̄2 (k) + λ(k)ϕ2 + · · · . As a consequence,
Γk can display a significant dependence on k in the regime k 2 & m̄2 (k) + λ(k)ϕ2 +
· · · only. Otherwise, k 2 is negligible relative to m̄2 + λϕ2 + · · · in all propagators;
it is then the physical cutoff scale m̄2 + λϕ2 + · · · rather than k 2 which delimits
the range of p2 -modes, which are integrated out.
Typically, for k very large, k 2 is larger than the physical cutoffs so that Γk
“runs” very fast. Lowering k it might happen that the running of m̄(k) comes
to a halt at some k = kdec . For k < kdec the “artificial” cutoff k becomes smaller
than m̄(k) = const. At this point the physical mass starts playing the role of the
actual cutoff; its effect overrides that of k so that Γk becomes approximately
independent of k for k < kdec . As a result, Γk ≈ Γkdec for all k below the threshold
scale given by kdec . In particular, the ordinary effective action Γ = Γ0 does not
differ from Γkdec significantly. This is the prototype of a decoupling or threshold
phenomenon.
The situation is even more interesting when m̄2 is negligible and k 2 com-
petes with λϕ2 for the role of the actual cutoff. (Here we assume that ϕ is
x-independent.) The running of Γk , evaluated at a fixed ϕ, stops once k . kdec (ϕ).
The field-dependent decoupling scale is determined by the implicit equation

10 It is not difficult to convince oneself that a general cutoff function Rk (p2 ) would lead to
identical conclusions and that they are properties of the exact RG flow valid beyond the
LPA truncation.
 2.3 Decoupling 47

2
( )
kdec = λ kdec ϕ2 and decoupling occurs for sufficiently large values of ϕ. The
RG evolution below kdec is negligible then and

Γ[ϕ] = Γk [ϕ] . (2.99)

k=kdec (ϕ)

This relationship represents a valuable practical tool. Under favorable circum-

stances it allows us to go beyond the truncation (2.97) without having to derive
and solve a more complicated flow equation. Indeed, because of the additional
ϕ-dependence which comes into play via kdec (ϕ), (2.99) can predict certain terms
that must be present in Γ, but were not included into the truncation ansatz.
A simple example illustrates this point. For k large, the quartic truncation for
the potential gives rise to a logarithmic running of the ϕ4 -coupling:
( ) ∝ ln(k).
λ(k)
As a result, (2.99) suggests that Γ should contain a term ∝ ln kdec (ϕ) ϕ4 . Since,
in leading order of the implicit equation for the decoupling scale, kdec ∝ ϕ, this
leads us to the prediction of a ϕ4 ln(ϕ)-term in the conventional effective action.
This prediction, including the prefactor of the term, is known to be correct actu-
ally: the Coleman–Weinberg potential of massless ϕ4 -theory does indeed contain
a ϕ4 ln(ϕ)-term [76, 77]. Note that this term admits no power series expansion
in ϕ, so it lies outside the space of functionals spanned by the original ansatz
(2.97).
This example nicely illustrates the power of the decoupling arguments. They
can be applied even when ϕ is taken x-dependent as it is necessary for computing
n-point functions by differentiating Γk [ϕ]. The running inverse propagator, say,
(2)
is given by Γk (x − y) = δ 2 Γk /δϕ(x)δϕ(y). Here a new competitor for the role of
the actual cutoff scale enters the game: the momentum q conjugate to the dis-
tance x − y. When q serves as the acting IR cutoff, the running of Γ e (2) (q), the
k
(2)
Fourier transform of Γk (x − y), stops once k 2 is smaller than kdec2
= q 2 . Hence
e (2) (q) ≈ Γ
Γ e (2) (q) √ for k 2 . q 2 , provided no other physical scales intervene.
k k
k= q2
As a result, if one has computed the running wave function renormalization Zk
in(the
√ truncation
) (2.97) one is able to predict an inverse propagator
(√ )of the type
Z q 2 q 2 in the standard effective action. Note that ϕ Z −∂ 2 ∂ 2 ϕ corre-
sponds to a non-local term in Γ, even though the truncation ansatz was perfectly
local. (See [70] for an example.)
The methodology of expressing k in terms of dynamical quantities, such as
physical masses, field values, momenta, etc., is referred to as a renormalization
group improvement. Based upon the above ideas on decoupling, it can be a pow-
erful tool, a kind of “shortcut” from the UV to the IR, provided one is able to
identify the actual physical cutoff mechanism without laboriously solving for the
full RG flow. Typically this requires a detailed case-by-case study of the various
physical scales displayed by the system at hand. The identification of the physi-
cal cutoff is greatly facilitated by symmetry arguments and, in particular, near a
scale-invariant fixed point, by dimensional analysis. But even in the generic case
48 The Functional Renormalization Group

it is sometimes possible to get a handle on the essential physics by a careful anal-

(2)
ysis of the eligible cutoff candidates in Γk . We will come back to this issue when
we try to find the quantum gravity corrections to black holes and cosmological
spacetimes.

2.4 Background Fields

Later on when we address the difficulties specific to quantum gravity and not
shared by matter field theories on a classical spacetime, the concept of back-
ground fields [51] will play a pivotal role. While the background technique can
be applied to such matter theories as well, in typical calculations of particle or
condensed matter physics this is avoided usually. Occasionally it has been used in
Yang–Mills theory where, still not compulsory, it offers the convenient option of
a gauge-invariant generating functional [86, 87]. In this section we illustrate the
approach by means of a scalar “toy system” where it appears to be a triviality
almost. Nevertheless, it is helpful to see the background method at work in a
setting which is much simpler than quantum gravity [59, 88, 89].

(1) The EAA in the background formalism, ΓB k , is based on a variant of the gen-
erating functional in (2.4), designated as WkB [J; ϕ̄]. The essential modification
consists in writing the variable of integration, ϕ,b as the sum of a prescribed,
non-dynamical or “background” field ϕ̄, and a dynamical or “fluctuations”
b
field, φ:
b = ϕ̄(x) + φ(x).
ϕ(x) b (2.100)
We emphasize that φ b is by no means assumed to be numerically small relative to
ϕ̄, and that no expansion in φ b is implied by (2.100). Indeed, in the background
approach φ b serves as the new integration variable. By virtue of the translation
invariance of the measure Dϕb we have Dϕb = Dφ, b and so we define the new, now
ϕ̄( · )-dependent, generating functional by
∫ {
B
ZkB [J; ϕ̄] ≡ eWk [J;ϕ̄] = Dφ b exp −S[ϕ̄ + φ]−∆S
b B
b ϕ̄]
k [φ;
∫ } (2.101)
b
+ dd x φ(x)J(x) .

The new cutoff action possesses the structure

∫
1
B
b ϕ̄] =
∆Sk [φ; b RB
dd x φ(x) b
k [ϕ̄] φ(x), (2.102)
2
whereby RB b which has the same general properties
k [ϕ̄] is an operator acting on φ,
as the ones we employed earlier. The only novelty is that it may be chosen
ϕ̄-dependent now.
 2.4 Background Fields 49

As the notation indicates, the generating functionals in the background setting,

ZkB [J; ϕ̄]
and WkB [J; ϕ̄], respectively, will in general depend on our choice for the
background field ϕ̄(x). In fact, (2.101) differs from its “non-background” analog
in two ways that go beyond simply shifting a dummy variable:

b only, i.e., not to the full field ϕ̄ + φ.

(i) The source J couples to the fluctuation φ b
As a consequence, multiple differentiation of WkB with respect to J gives rise
to connected correlation functions ⟨φ(x
b 1 )φ(x
b 2 ) · · · φ(x
b n )⟩ that apply in the
presence of an external field, namely ϕ̄. The one-point function ⟨φ(x)⟩,b for
example, reads in full detail

δWkB
[J; ϕ̄] = φk [J; ϕ̄](x). (2.103)
δJ(x)

Often we omit the plethora of arguments and write ⟨φ(x)⟩ b = φ(x)⟨simply.

⟩
Furthermore, ϕ̄ is a classical, i.e., non-fluctuating field, implying ϕ̄ = ϕ̄,
and thus the expectation value of the complete field is recovered as
⟨ ⟩
b
ϕ(x) ≡ ϕ(x) = ϕ̄(x) + ⟨φ(x)⟩
b ≡ ϕ̄(x) + φ(x). (2.104)

The actual motivation for the seemingly ad hoc φ b J-coupling in the defini-
tion (2.101) would become clear only in more complicated theories containing
Yang–Mills fields, for instance. For a gauge field, the decomposition analo-
gous to (2.100) reads Abµ (x) = Āµ (x) + b
aµ (x) whereby both the full field, A bµ ,
and the background field, Āµ , gauge-transform as a connection, i.e., inhomo-
geneously. It then follows that b aµ must transform homogeneously, i.e., in a
tensorial way. Coupling b aµ rather than A bµ to the source is therefore a first
crucial step in ultimately arriving at a gauge-covariant formalism.
(ii) The new cutoff action is designed to produce a mass term for the low momen-
tum modes of φ b In (2.102), ∆S B is bilinear in the fluctuation
b rather than ϕ. k
field and depends on ϕ̄ at most via RB k [ϕ̄]. In fact, in scalar field theory
on a classical spacetime there is no cogent reason that would enforce a ϕ̄-
dependent cutoff operator; a choice of the form RB 2 (0)
k =k R (−∂ 2 /k 2 ) is also
the most natural option here. However, admitting also ϕ̄-dependent oper-
ators brings our scalar play ground closer to quantum gravity where deep
general principles will make it impossible to construct cutoff actions that
would not depend on the background field.

The EAA of the background approach, ΓB k [φ; ϕ̄], is defined in analogy with
Γk [ϕ] as the Legendre transform of WkB [J; ϕ̄] with respect to the source with the
cutoff action subtracted. In this way we replace J by the expectation value of the
fluctuation, φ = ⟨φ⟩,
b as the independent variable. In the non-singular case when
50 The Functional Renormalization Group

the field-source relationship (2.103), i.e., φk [J; ϕ̄](x) = φ(x) with φ(x) given, can
be solved for the source as J(x) = JkB [φ; ϕ̄](x), we have explicitly
∫
k [φ; ϕ̄] ≡
ΓB dd x φ(x) JkB [φ; ϕ̄](x) − WkB [JkB [φ; ϕ̄]; ϕ̄]
∫ (2.105)
1
− dd x φ(x) RB k [ϕ̄] φ(x).
2

Obviously the EAA has inherited a parametric ϕ̄-dependence from WkB . (Note
that the background field is just a spectator in the Legendre transformation.)

(2) The properties of ΓBk are easily established following the steps outlined in
Section 2.1. For example, by differentiating (2.105) with respect to φ we obtain
the corresponding effective field equation:

δΓB
k
[φ; ϕ̄] + RB
k [ϕ̄]φ(x) = J(x). (2.106)
δφ(x)
This equation expresses the relationship between sources and fields in a form
which is conjugate to (2.103). Furthermore, the EAA in the background setting
is easily seen to satisfy an FRGE of the form
[( )−1 ]
1 B(2)
∂k Γ B
k [φ; ϕ̄] = Tr Γ k [φ; ϕ̄] + R B
k [ϕ̄] ∂ R
k k
B
[ ϕ̄] . (2.107)
2
B(2)
Here Γk stands for the Hessian matrix that comprises all second derivatives
with respect to φ at fixed ϕ̄:

B(2) δ2
Γk [φ; ϕ̄](x, y) ≡ ΓB [φ; ϕ̄]. (2.108)
δφ(x)δφ(y) k

(3) Sometimes it is helpful to think of the EAA as depending on the two inde-
pendent fields ϕ ≡ ϕ̄ + φ and ϕ̄, rather than φ together with ϕ̄, and to employ
the redefined action functional11

k [ϕ, ϕ̄] ≡ Γk [φ; ϕ̄]

ΓB B
. (2.109)
φ=ϕ−ϕ̄

Consider, for instance, the special case when ΓB k [φ; ϕ̄] happens to
⟨ depend
⟩ on the
fields φ and ϕ̄ only via their sum, the full, undecomposed field ϕ ≡ ϕb ≡ ϕ̄ + ⟨φ⟩b ≡
B
ϕ̄ + φ. The new functional Γk [ϕ, ϕ̄] is independent of its second argument then:

k [φ; ϕ̄] = F [ϕ̄ + φ] ⇐⇒ Γk [ϕ, ϕ̄] = F [ϕ].

ΓB B
(2.110)

11 The two functionals in (2.109) are distinguished by the “comma vs. semicolon” notation
that has been customary for the two variants of the background EAA since very early on
[23, 24].
 2.4 Background Fields 51

Therefore, a non-trivial dependence of ΓBk [ϕ, ϕ̄] on ϕ̄, at fixed ϕ indicates an extra
background field dependence over and above the one that matches the dependence
on φ so that it can be joined with it, resulting in a complete field ϕ̄ + φ ≡ ϕ.
A given field ϕ can be decomposed into background and fluctuation fields
in infinitely many different ways. Different decompositions are related by the
background-quantum field split symmetry transformations, involving an arbitrary
function α:
δα φ(x) = α(x) , δα ϕ̄(x) = −α(x). (2.111)
Obviously the complete field ϕ is invariant under such split transformations,
δα ϕ = δα φ + δα ϕ̄ = 0, and the same holds true for any functional of ϕ alone. In
fact, the response of the EAA to a split symmetry transformations, δα ΓB k [ϕ, ϕ̄] ≡
Γk [ϕ, ϕ̄ − α] − Γk [ϕ, ϕ̄], is a measure for its extra background-field dependence.
B B

If we require for instance that δα ΓB k = 0 ∀ α, it follows that the EAA has no extra
ϕ̄ dependence.
More generally, one can derive a Ward identity for the split symmetry by
applying (2.111) under the functional integral and working out the consequences
for the EAA. In the scalar toy model the result has a similar structure as the
flow equation [90]:
[( )−1 δRB [ϕ̄] ]
δ 1 B(2)
Γ [ϕ, ϕ̄] = Tr Γk [ϕ, ϕ̄] + Rk [ϕ̄]
B B k
. (2.112)
δ ϕ̄(y) k 2 δ ϕ̄(y)

This kind of split symmetry Ward identity was first derived in [29] for Yang–
Mills theory. Its applications include testing the reliability of truncations [90]. As
they are derived from the same functional integral, consistency between Ward
identity and FRGE is guaranteed for exact RG trajectories, but within trunca-
tions it may be violated to some extent.

(4) Despite the differences noted in (i), (ii), the background approach and
the one without background fields are strictly equivalent in the scalar case.
This equivalence is made manifest by the following relationship between their
respective generating functionals:
∫ ( )
1 [ ]
WkB [J; ϕ̄] = − dd x J ϕ̄ + ϕ̄ RB
k [ ϕ̄] ϕ̄ + Wk′ J + RB
k [ϕ̄] ϕ̄ . (2.113)
2
Herein, Wk′ [J], which appears with a shifted argument in (2.113), represents the
usual generating functional of the non-background formalism, albeit with the
possibly ϕ̄-dependent cutoff operator RB k built into it:
∫ ( ∫
eWk [J] ≡ Dϕb exp −S[ϕ]
′
b − 1 dd x ϕb RB [ϕ̄] ϕb
k
2
∫ ) (2.114)
b
+ dd x J(x)ϕ(x) .
52 The Functional Renormalization Group

Exactly as outlined in Section 2.1, we associate an EAA, in this context denoted

Γ′k [ϕ], to the above Wk′ [J].
The primes at Wk′ , Γ′k , and at the quantities derived from them are meant to
remind us of the “weak” ϕ̄-dependence those quantities will acquire if we decide
to employ a cutoff RB k , which depends on the background field. If instead we
select a ϕ̄-independent one (which is an option in the scalar toy model but not
in quantum gravity) all primes may be omitted and the corresponding quantities
will be exactly those of the non-background approach.
Differentiating both sides of (2.113) with respect to J(x) we are led to
δWkB δWk′ [ ]
[J; ϕ̄] + ϕ̄(x) = J + RB
k [ϕ̄] ϕ̄ . (2.115)
δJ(x) δJ(x)
The interpretation of this identity becomes clear when we recall (2.103) and the
analogous equation for Wk′ ,
δWk′
[J] = ϕk [J](x). (2.116)
δJ(x)
b
b and ϕ,
Namely, for J and ϕ̄ fixed, (2.115) says that the expectation values of φ
respectively, are related by
[ ]
φk [J; ϕ̄](x) + ϕ̄(x) = ϕ′k J + RB
k [ϕ̄] ϕ̄ (x). (2.117)

Obviously this formula is the “effective” analog of the decomposition ϕb = ϕ̄ + φ.

b
It applies at the level of expectation values, however, rather than to the bare
quantities.
We can obtain a conjugate form of the result (2.117) if we fix the expectation
values of the fields and ask which source produces them:
′
JkB [φ; ϕ̄](x) + RB
k [ϕ̄] ϕ̄(x) = Jk [ϕ̄ + φ](x). (2.118)

Here JkB [ · ]( · ) is the same functional as above, namely the solution to (2.103),
but we read (2.118) as an equation for J in which φk [J; ϕ̄] ≡ φ is fixed. Anal-
ogously the solution to (2.116), with ϕk [J] ≡ ϕ fixed, is denoted Jk′ [ϕ]. On the
RHS of (2.118) the latter functional is evaluated at ϕ̄ + φ = ϕ.
Equipped with the tool of (2.118) it is easy to express the EAA of the back-
′
ground field approach, ΓB k , in terms of Γk , the ordinary EAA with a ϕ̄-dependent
cutoff possibly. After the dust has settled we find
′
ΓB
k [φ; ϕ̄] = Γk [φ + ϕ̄]. (2.119)

This result looks even simpler if we rewrite it in terms of the alternate version
of the EAA, ΓB k [ϕ, ϕ̄] ≡ Γk [ϕ − ϕ̄; ϕ̄], which regards ϕ and ϕ̄ as the independent
B

arguments:
′
ΓB
k [ϕ, ϕ̄] = Γk [ϕ]. (2.120)
 2.4 Background Fields 53

(5) The interpretation of (2.119) is particularly clear in the limit k → 0 with

′
RBk → 0. Then Γ0 = Γ0 = Γ coincides with the standard effective action of the
non-background formalism. According to (2.119) it may be written as Γ[φ + ϕ̄] =
0 [φ; ϕ̄], or after setting φ ≡ 0:
ΓB

Γ[ϕ̄] = ΓB
0 [0; ϕ̄]. (2.121)

This simple statement shows that the standard one particle irreducible Green
functions given by multiple derivatives of Γ can also be obtained by differentiating
ΓB0 with respect to the background field, at φ = 0:

δ n Γ[ϕ] δn
= ΓB [φ; ϕ̄] . (2.122)
δϕ(x1 ) · · · δϕ(xn ) δ ϕ̄(x1 ) · · · δ ϕ̄(xn ) 0 ϕ̄=ϕ,φ=0

As a consequence, the entire physical contents of the theory is encapsulated in

the background EAA for a vanishing expectation value of the dynamical field,
i.e., in the restricted map ϕ̄ 7→ ΓB
0 [0; ϕ̄].
This, then, is one of the main advantages of the background field method: it
permits a diagrammatic computation of the relevant effective action by summing
vacuum graphs only, i.e., graphs without external lines of the quantum fields.

(6) Returning to (2.119) in the functional RG context we see that at scales k > 0
there is a minor difference between ΓB k [0; ϕ̄] and the non-background EAA, the
reason being that Γ′k employs the possibly ϕ̄-dependent cutoff operator RB k [ϕ̄].
This is its only ϕ̄-dependence though, and since this ϕ̄-dependence merges into the
intrinsic arbitrariness of the cutoff it follows that Γk and Γ′k , and hence ΓB
k φ=0 ,
can differ at most in unphysical features which will cancel out somewhere on the
way to observable quantities.
As an aside we mention that in gauge theories there exists another source
of such unphysical differences between the analogs of Γ′k and Γk if ΓB k employs
a background field-dependent gauge-fixing condition. As Γ′k inherits this gauge
fixing, it acquires a corresponding dependence on the background gauge field
which, in this case, merges into the general arbitrariness of the gauge-fixing
condition.
As a consequence, the difference cannot have any influence on quantities that
are insensitive to the gauge-fixing condition, observables in particular.
A generic n-point function, on the other hand, is gauge-fixing dependent and
in this way acquires in particular a certain dependence on the background gauge
field. But, for example, in constructing scattering matrix elements from it via
the LSZ method, which in particular includes going “on shell”, this dependence
disappears so that Γ, Γ′ , and ΓB0 become indeed fully equivalent at the observable
level [87, 91].
 3
The Asymptotic Safety Mechanism

In this chapter we give an elementary introduction to the general ideas Asymp-

totic Safety rests upon without embarking on the particular intricacies of gravity.
Our presentation emphasizes the geometry of theory space and its natural struc-
tures while the discussion of the primarily technical aspects of quantum field
theory is kept to a bare minimum.

3.1 Geometry and RG Dynamics on Theory Space

{ 1 2 }
We consider a general set of fields Φ b= Φ b ,Φb , · · · with arbitrary spin
and
{ 1 Grassmann } parity, along with the corresponding background fields Φ̄ =
Φ̄ , Φ̄ , · · · and fluctuations φ
2
b≡Φ b − Φ̄. Their expectation values are collec-
⟨ ⟩
tively denoted Φ = Φ b and φ = ⟨φ⟩b = Φ − Φ̄, respectively. We assume that the
effective dynamics of this system of fields is governed by an Effective Average
Action (EAA) of the form Γk [φ; Φ̄] ≡ Γk [Φ, Φ̄].
In the previous chapter we described in detail how to construct the EAA in
the special case of scalar fields. Since the discussion in the present chapter will
mostly deal with the “kinematic” aspects of the RG evolution on theory space,
it is at this point sufficient to know that we can indeed generalize the EAA to
arbitrary systems of (matter, gauge, gravitational, etc.) fields while retaining
most of the general properties relevant to the scalar case. The specific difficulties
posed by gauge and gravitational fields will not yet play an important role, and
we postpone their discussion to the next chapter. Suffice it to say that even for
general field systems, which include quantized gravity, an appropriately defined
action functional Γk will ultimately turn out to satisfy a closed functional RG
equation again, and the structural appearance of this FRGE is analogous to that
of the scalar flow equation from the previous chapter:
[( )−1 ]
1 (2)
k∂k Γk [Φ, Φ̄] = STr Γk [Φ, Φ̄] + Rk [Φ̄] k∂k Rk [Φ̄] . (3.1)
2
 3.1 Geometry and RG Dynamics on Theory Space 55

(2)
Here Γk denotes the Hessian matrix with respect to all dynamical fields at fixed
background fields, δ 2 Γk /δΦi (x)δΦj (y), and the supertrace STr includes a sum-
mation over the field components and takes care of extra minus signs occurring
for Grassmann fields.
We assume that, besides the nature of the field multiplets, all additional
requirements to be satisfied by the EAA (symmetries, etc.) have been speci-
fied in some way. This concerns, in particular, the function spaces from which Φ
and Φ̄ are drawn, henceforth denoted F and F̄, respectively. These data define
the (dimensionful) theory space T . Its “points” A ∈ T are action functionals
A : F × F̄ → R, (Φ, Φ̄) 7→ A[Φ, Φ̄] with certain prescribed (symmetry, regularity,
etc.) properties.

3.1.1 The Beta Functional

When written in the pointwise style of (3.1) the FRGE expresses an equality
between two real numbers which must hold true for all pairs of fields (Φ, Φ̄) ∈
F × F̄. We may equivalently regard it as an equality of functionals and write it
compactly as [( ]
1 )−1
(2)
k∂k Γk = STr Γk + Rk k∂k Rk . (3.2)
2
As we shall see, the most geometric way to think of the FRGE is as follows.
For every fixed value of k we consider the map

Bk : T × F × F̄ −→ R ,
(A, Φ, Φ̄) 7−→ B k {A}[Φ, Φ̄], (3.3)

which returns a real number for every given action and every pair of field
arguments, and is defined explicitly by
[( )−1 ]
1
Bk {A}[Φ, Φ̄] ≡ STr A [Φ, Φ̄] + Rk [Φ̄]
(2)
k∂k Rk [Φ̄] . (3.4)
2

The number B k {A}[Φ, Φ̄] should be seen as a certain action functional, denoted
Bk {A}, which is being evaluated at the point (Φ, Φ̄).
Thus B k : A 7→ Bk {A} constitutes a map of actions onto actions. For lack of a
better word we refer to B k as the beta functional.
The usefulness of the definitions (3.3) and (3.4) is obvious, of course. It casts
the FRGE into the form

k∂k Γk [Φ, Φ̄] = B k {A}[Φ, Φ̄] ∀ (Φ, Φ̄), (3.5)

which is equivalent to the functional relationship

k∂k Γk = Bk {Γk } . (3.6)

56 The Asymptotic Safety Mechanism

In particular, for structural considerations the equation (3.6), and the inter-
pretation that comes with it, is the most convenient way to express the
FRGE.

3.1.2 From the Beta Functional to the Beta Functions

Let us assume that T has the structure of a vector space, and let {Iα } denote
a complete set of (k-independent!) basis functionals Iα : (Φ, Φ̄) 7→ Iα [Φ, Φ̄]. Every
∑
A ∈ T has an expansion A = α ūα Iα with real components ūα then.
In the case of the EAA, the components, or generalized coupling constants, as
they are called here, carry the entire scale dependence:
∑
Γk = ūα (k) Iα . (3.7)
α

To derive RG equations for the k-dependence of the running coupling constants

we insert (3.7) into the functional equation (3.6):
∑ {∑ }
k∂k ūα (k) Iα = Bk ūα (k) Iα . (3.8)
α α

As the map Bk { · } returns an action functional, and since the Iα s form a basis
of theory space, we know that the RHS of (3.8) possesses a unique expansion in
this basis: {∑ } ∑ ( )
Bk ūα (k) Iα = b̄α ū(k); k Iα . (3.9)
α α

Here ū(k) ≡ {ū (k)} denotes the list of all running coupling constants. Upon
α

using (3.9) in (3.8) and equating coefficients of equal basis elements we are led
to the exact RG equation in component form:
( )
k∂k ūα (k) = b̄α ū(k); k ∀ α. (3.10)

Note that besides an implicit k-dependence via ū(k), the functions b̄α explicitly
depend on k since the map Bk does.
We note that in the present context one might regard Γk [ · ] and k∂k Γk [ · ]
simply as generating functionals for two infinite families of functions, namely
their components {ūα } and {b̄α }, respectively. This viewpoint deemphasizes, or
at least shifts, the role of the fields; they are needed merely in order to describe
the mathematical nature of the basis elements Iα and the space they span.

3.1.3 The Cutoff Scale as a Unit

A generic running coupling constant ūα will have a nonzero canonical mass
dimension [ūα ] ≡ dα . Since Γk [Φ, Φ̄] is dimensionless so are all terms ūα Iα [Φ, Φ̄]
 3.1 Geometry and RG Dynamics on Theory Space 57
[ ]
in its component representation and thus Iα [Φ, Φ̄] = −[ūα ] = −dα . Furthermore,
(3.10) shows that b̄α must have the same dimension as the coupling it belongs
to: [b̄α ] = dα .
Let us now switch to a formalism in which all dimensionful quantities, cou-
plings, fields, beta functions, etc., are expressed in units of the cutoff scale, k, or
appropriate powers thereof.

(1) First, we introduce dimensionless coupling constants:

uα ≡ k −dα ūα . (3.11)

Usually uα is considered a function of the RG time t ≡ ln( kk0 ), which is also

dimensionless; it involves an arbitrary, fixed reference scale k0 . Thus, inserting
ūα (k) = k dα uα (t) into (3.10), we arrive at the exact RG equation in dimensionless
component form:
( )
∂t uα (t) = β α u(t) . (3.12)

It involves the dimensionless β-functions given by

β α (u) = −dα uα + bα (u). (3.13)

The bα s without the overbar are defined as

( )
bα (u) = k −dα b̄α {k dα uα }; k . (3.14)

Here all explicit factors of k on the RHS must cancel, the reason being that this
term is dimensionless, k is the only dimensionful quantity available and so the
term is of zeroth order in k.
The new β-functions depend only on u ≡ {uα } and when u ≡ u(t) is inserted
on the RHS of (3.12) they of course acquire an implicit scale dependence. This
shows that, in contrast to the original dimensionful couplings, the dimensionless
ones satisfy a system of autonomous differential equations.

(2) In general, the various fields in the sets Φ = {Φi }, φ = {φi }, and Φ̄ =
{Φ̄i } come with different canonical dimensions δi ≡ [Φi ] = [φi ] = [Φ̄i ]. We define
dimensionless fields in the obvious way:

e i ≡ k −δi Φi ,
Φ ei ≡ k −δi φi ,
φ ē i ≡ k −δi Φ̄i .
Φ (3.15)

For the sets Φ e ≡ {Φ e i }, etc., we use the convenient shorthand notations Φ e=

k −[Φ] Φ, φ
e = k −[φ] φ, and Φ ē = k −[Φ̄] Φ̄, where [Φ] = {[Φi ]} and, by definition of the
linear split, [Φ] = [φ] = [Φ̄].
We already know that the basis functionals Iα [Φ, Φ̄] consist of monomials built
from the dimensionful fields in such a way that their dimensions sum up to
58 The Asymptotic Safety Mechanism
[ ]
the overall value Iα [Φ, Φ̄] = −dα . Since the fields are the only1 dimensionful
quantities available out of which the Iα s can be constructed, dimensional analysis
implies that
[ ]
Iα c[Φ] Φ, c[Φ̄] Φ̄ = c−dα Iα [Φ, Φ̄] (3.16)
for any constant c > 0. Evaluating this relation at c = k −1 we find

Iα [Φ, ē ≡ I [k −[Φ] Φ, k −[Φ̄] Φ̄] = k dα I [Φ, Φ̄].

e Φ] (3.17)
α α

The quantity (3.17), in any of the three versions, is dimensionless. Thus we learn
that when we insert dimensionless field arguments into the basis functionals
in place of the original Φ and Φ̄, the Iα s lose their nonzero mass dimension:
[ ē ] = 0.
e Φ]
Iα [Φ,
For later use we also remark that by acting with c ∂c on (3.16) and setting
c = 1 thereafter we obtain the useful identity

XIα [Φ, ē = −d I [Φ,

e Φ] ē
e Φ] (3.18)
α α

with the operator

∫ { }
∑ δ δ
X= δi X i , Xi ≡ d
d x ei
Φ (x) ē i (x)
+Φ . (3.19)
e i (x)
δΦ ē i (x)
δΦ
i

When applied to a given field monomial the operator Xi counts the number of
“i”-type fields contained in it. Note that while equation (3.18) is written entirely
in terms of dimensionless quantities, the eigenvalue (−dα ) on its RHS equals
exactly the would-be canonical dimension of Iα , i.e., the one it assumes when
dimensionful fields are plugged into it.

(3) If parametrized by the dimensionless couplings, the EAA reads

∑
Γk [Φ, Φ̄] = uα (k) k dα Iα [Φ, Φ̄] (3.20)
α

or, after making use of (3.17),

∑ [ ]
Γk [Φ, Φ̄] = uα (k) Iα k −[Φ] Φ, k −[Φ̄] Φ̄
α
∑ (3.21)
= uα (k) Iα [Φ, ē
e Φ].
α

1 Note that, by definition, Iα does not contain any coupling constants. Furthermore, according
to the assignment of dimensions we use in this section all coordinates xµ are dimensionless.
This comes at the price of considering the metric, even when non-dynamical, as a field. Car-
rying a nonzero dimension, the transition to dimensionless fields, (3.15), entails a nontrivial
rescaling, ē
g µν = k2 ḡµν .
 3.1 Geometry and RG Dynamics on Theory Space 59

Let us employ the set of dimensionless running couplings {uα (k)} appear-
ē 7→ A [Φ,
e Φ)
ing in (3.20) in order to define a new action functional Ak : (Φ, ē
e Φ]
k
according to
∑
A [Φ, ē ≡
e Φ] ē
e Φ].
uα (k) I [Φ, (3.22)
k α
α

Here Φ ē should be thought of as dimensionless fields that can be chosen

e and Φ
freely. They are the independent arguments of Ak and thus play the same role
for Ak as the dimensionful ones for Γk .
If some pair of dimensionless arguments (Φ, e Φ)ē is fixed first, and then a
dimensionful pair (Φ, Φ̄) is associated to it by virtue of (3.15), equation (3.21)
amounts to
A [Φ, ē = Γ [Φ, Φ̄].
e Φ] (3.23)
k k

Despite the numerical equality expressed by this relation, Ak and Γk are dif-
ferent functionals. The independent arguments of the former (latter) are the
dimensionless (dimensionful) fields.
If corresponding sets of dimensionless and dimensionful fields are related by
the transformation (3.15), then Ak attributes the k-dependence of this transfor-
mation to (Φ, Φ̄), while from the perspective of Γk it is the dimensionless pair
e Φ)
(Φ, ē that carries the scale dependence. As a consequence, when we take the
k-derivatives of Ak and Γk , keeping their respective field arguments fixed, we
obtain different results:
∑
k∂k Ak [Φ, ē =
e Φ] ē
e Φ],
k∂k uα (k) Iα [Φ, (3.24)
α
∑{ }
k∂k Γk [Φ, Φ̄] = k∂k uα (k) + dα uα (k) k dα Iα [Φ, Φ̄]. (3.25)
α

The extra term proportional to dα uα (k) in (3.25) stems from the derivative of
the factor k dα in (3.20).

(4) If we insert the explicit β-functions (3.13) and (3.14) into (3.25) we are led
back to ∑ ( )
k∂k Γk [Φ, Φ̄] = b̄α ū(k); k Iα [Φ, Φ̄]. (3.26)
α

This equation appeared above where it defined the expansion coefficients b̄α . The
analogous equation at the dimensionless level follows from (3.24) by inserting
(3.12):
∑ ( )
k∂k Ak [Φ, ē =
e Φ] ē
e Φ].
β α u(k) Iα [Φ, (3.27)
α

Obviously the scale-derivatives of Γk and Ak may be interpreted as generating

functionals for the functions {b̄α } and {β α }, respectively.
60 The Asymptotic Safety Mechanism

Finally we set up a truly functional RG equation also in the dimensionless

setting; it evolves actions per se rather than their basis-dependent components.
It is defined on the theory space proper, namely a space of functionals

A : F × F̄ −→ R , (Φ, ē 7−→ A[Φ,

e Φ) ē
e Φ], (3.28)

which, abstractly, is not different from the “dimensionful” theory space consid-
ered earlier, just the assignment of dimensions has changed.
Henceforth we refer to this reinterpreted space of actions depending on dimen-
sionless fields as “the” theory space. Keeping in mind that the dimensionless
theory space is isomorphic to the dimensionful one introduced at the beginning
of this section, we denote it by T as well.
Inserting (3.13) with (3.14) into (3.27) we obtain
∑ ∑ ( )
k∂k Ak = − uα (k) dα Iα + b̄α ū(k); k k −dα Iα . (3.29)
α α

Taking advantage of the identity (3.18) the first term on the RHS of this equation
∑
is seen to be equal to X α uα (k)Iα = XAk , while the second term equals Bk up
to various rescalings. The latter are written most conveniently with the help of
the simple map
( )
σk : (Φ, Φ̄) 7→ k −[Φ] Φ, k −[Φ̄] Φ̄ , (3.30)
which turns dimensionful into dimensionless fields.
In this manner we obtain the final result for the functional RG equation
governing the EAA of the dimensionless fields, namely

∂t Ak ≡ β{Ak } (3.31)

with the beta functional of the dimensionless approach:

β{A} = XA + B{A}. (3.32)

Here, X is the counting operator (3.19) which depends on the field dimensions
δi , and B denotes the dimensionless counterpart of B k ,

B{A}[Φ, ē ≡ B {A ◦ σ }[σ −1 (Φ,

e Φ] ē ].
e Φ) (3.33)
k k k

The new map B has no explicit scale dependence.2 As a result, the beta func-
tional (3.32) is also k-independent, so that (3.31) is indeed an autonomous,
basis-independent evolution equation. After fixing a basis it would lead back
to the component equations (3.12), (3.13), of course.

2 Cf. the reasoning in (1) above. Note also that we returned to the RG time as a manifestly
dimensionless evolution parameter (t = ln k + const, k∂k = ∂t ). For simplicity we keep using
the notation Ak though.
 3.1 Geometry and RG Dynamics on Theory Space 61

3.1.4 The Geometric Interpretation

The evolution of the Effective Average Action Ak resulting from a change of the
RG scale has an appealing geometric interpretation as a dynamical system on an
infinite dimensional space.

(A) Finite dimensional flows. To make this interpretation explicit, let us first
consider an arbitrary smooth, n-dimensional
( ) manifold, Mn . We furnish it with
a system of local coordinates, ξ ≡ ξ i , i = 1, · · · , n, giving rise to a correspond-
ing basis ∂i ≡ ∂ξ∂ i in the tangent spaces Tξ Mn , and a dual basis dξ i in Tξ∗ Mn .
Furthermore, we assume that in addition to the manifold structure we are given
a vector field V on Mn , locally represented as V (ξ) = V i (ξ)∂i .
The pair (Mn , V ), which consists of a manifold and a vector field, is referred
to as a flow. This setup plays a central role in the general theory of dynamical
systems, e.g., [92–96].
A well-known special case is Hamilton’s version of classical mechanics, where
the (now even dimensional) manifold Mn plays the role of the system’s phase
space. Hamiltonian dynamical systems are special in the sense that the admissible
vector fields V are “symplectic gradients” V i = J ij ∂j H. Here the Hamilton func-
tion H is an arbitrary scalar on Mn and the Poisson tensor J ij is an independent
ingredient that defines the mechanical system along with H and the manifold
structure. In local Darboux coordinates ξ i = (q, p) and for n = 2 its components
are J 12 = −J 21 = 1, J 11 = J 22 = 0. The vector field V ≡ VH associated with H
then reads
∂H ∂ ∂H ∂
VH = − (3.34)
∂p ∂q ∂q ∂p
and similarly for more than one degree of freedom.
The name “flow” for the pair (Mn , V ) has then been coined since those data
are precisely what is needed in order to set up the “flow equation”

˙
ξ(λ) ≡ ∂λ ξ(λ) = V (ξ(λ)). (3.35)

This equation determines the integral curves λ 7→ ξ(λ) the vector field V gives rise
˙
to. By definition, their “velocity” ξ(λ) ∈ Tξ(λ) Mn is tangent to V at all points
visited by the curve.
Denoting the solution to (3.35) subject to the initial condition ξ(0) = ξ0 by
λ 7→ ξ(λ; ξ0 ), we may think of ξ(λ; ξ0 ) as the location of a fictitious particle at
”time” λ moving on Mn and released from ξ0 at the time λ = 0. Conversely we
can fix the final time, setting λ = 1, say, and ask how the position of the parti-
cle at this time depends on its initial position at λ = 0. Allowing ξ0 to be any
point of the manifold, answering this question amounts to constructing the map
Mn → Mn , ξ0 7→ ξ(1; ξ0 ) by solving the differential equation subject to all possi-
ble initial conditions. Under appropriate conditions this map is a diffeomorphism.
It may be interpreted (actively) as describing the motion of a swarm comprising
62 The Asymptotic Safety Mechanism

many particles released from different initial points, or (passively) as a general

coordinate transformation on Mn .
In Hamiltonian mechanics the flow equation (3.35) refers to the physical time
λ ≡ t and uses a symplectic vector field of the form (3.34). At the level of its
components it is nothing else than Hamilton’s equations of motion, q̇ = ∂H/∂p,
ṗ = −∂H/∂q. In this particular case the above diffeomorphism is volume preserv-
ing (by Liouville’s theorem). A generic flow (Mn , V ) will not have this property.

(B) The RG flow. Returning now to the Effective Average Action of the
dimensionless setting, Ak ≡ A(t), we see that the RG equation which it satisfies,

∂t A(t) = β{A(t)} , (3.36)

has the same structure as the flow equation (3.35), at least formally. The role of
the evolution parameter λ is played by the RG time, the “manifold” on which
the dynamics take place is the theory space T of “all” actions A : F × F̄ → R,
and β{A} ≡ XA + B{A} with B defined in (3.33) takes the place of the vector
field V .
Invoking this formal correspondence we would interpret the map β when eval-
uated at some point A ∈ T as a vector in the “tangent space to T at A,” and
instead of V (ξ) ∈ Tξ Mn we would write β{A} ∈ TA T . At the level of this corre-
spondence the customary term renormalization group flow then indeed refers to
a flow in the above sense:

RG flow of the EAA : (T , β). (3.37)

It should be clear that the “RG flow of the EAA” goes significantly beyond
the mathematically well studied finite dimensional case. Theory space is infinite
dimensional and it is notoriously difficult to make the notion of a “space of all
actions” mathematically precise. Therefore, we cannot expect that the results
established for finite-dimensional systems (Mn , V ) carry over straightforwardly
to the RG case, at least not completely.
The drastic nature of the step from (Mn , V ) to (T , β) is perhaps appreciated
best by recalling that ∂t A(t) = β{A(t)} expresses the equality of two functionals
ē = β{A(t)}[Φ,
e Φ]
and is hence equivalent to ∂t A(t)[Φ, ē for all field configurations
e Φ]
ē ∈ F × F̄. Introducing the “index notation” ∂ A(Φ,
e Φ)
(Φ,
ē
e Φ) ē and
e Φ],
(t) ≡ ∂t A(t)[Φ,
t
similarly for β, the equation assumes the somewhat arcane-looking form
e ē e ē ē ∈ F × F̄.
e Φ)
∂t A(Φ,Φ) (t) = β (Φ,Φ) {A(t)} ∀ (Φ, (3.38)

We would like to interpret this equation as the component form of (3.36). The
analogy to the finite-dimensional equation

∂λ ξ i (λ) = V i (ξ(λ)) ∀ i ∈ {1, 2, · · · , n}, (3.39)

 3.1 Geometry and RG Dynamics on Theory Space 63

then suggests that there is a correspondence between the vector index i on

the one side, assuming finitely many discrete values only, and full-fledged field
configurations on the other side:

i ←→ (Φ, ē
e Φ). (3.40)

Thus, loosely speaking, the tangent space TA T would have as many dimensions
as there are fields in F × F̄.
It is certainly not easy to give a mathematically precise meaning to such
statements and to find an appropriate infinite-dimensional generalization of the
classical geometrical concepts in full generality. It also seems clear that this will
not be possible without imposing additional technical conditions on the various
ingredients of the setup. Those conditions should be strong enough to allow for
proving general mathematical theorems, but at the same time not so strong as
to rule out some or, possibly, all applications of physical interest.
Here and in the following we make first of all the technical assumption that
T has at least the structure of a vector space. This setting will be sufficient for
all our general discussions and explicit examples. We invoked this requirement
already earlier on when we assumed that every A ∈ T can be expanded in a set
∑
of basis functionals, A = α uα Iα . Moreover, we already discussed a nontrivial
example where we saw that this is indeed possible, namely the framework of the
derivative expansion.
Under this assumption we can identify all tangent spaces TA T with T itself
∑
and reinterpret any point A = α uα Iα ∈ T as a vector, which we view as the
“position vector” of the point A, having components u ≡ (uα ), with respect to
the basis chosen (“I-basis”).
Let us recast the expansion A[Φ, e Φ] ē = ∑ uα I [Φ, e Φ]ē in the “vector index”
α α
ē
e Φ)
language used in (3.38). Setting Iα
(Φ,
≡ Iα [Φ, ē this leads us to interpret
e Φ]

e ē
∑ e ē
A(Φ,Φ) = uα Iα(Φ,Φ) (3.41)
α

as the components of the position vector in the “field basis.” Seen from a formal
( (Φ, ē )
e Φ)
perspective, Iα is an infinite-dimensional matrix whose rows and columns
are labeled by “indices” (Φ, ē and α, respectively, and which describes a basis
e Φ)
change from the I-basis to the field basis. The completeness of the functionals
Iα assumed, this matrix is invertible and so we can freely switch back and forth
between the two bases, trading the (Φ, ē
e Φ)-index for α and vice versa.
e ē
In particular, we can now regard i ↔ (Φ, Φ) as a correspondence between the
vector indices i and α, and we may apply the basis change (3.41) to both sides
of the “arcane” equation (3.38). In this way we end up with the expected result,

∂t uα (t) = β α (u(t)) ∀ α, (3.42)

64 The Asymptotic Safety Mechanism

where ∂t uα and β α (u(t)) are the components in the I-basis of ∂t A and β{A(t)},
respectively, and uα (t) are the coordinates of the point A(t), or the components
of its position vector.
The components of β in (3.38) and (3.42), respectively, are related by a matrix
( (Φ, ē )
e Φ)
multiplication with the transformation matrix Iα :

e ē
∑ ∑ e ē
β (Φ,Φ) {A} ≡ β{A}[Φ, ē =
e Φ] β α (u)Iα [Φ, ē ≡
e Φ] β α Iα(Φ,Φ) . (3.43)
α α

The first and the last equalities in (3.43) merely amount to a trivial change
of notation, while the second one represents the expansion of a certain func-
∑
tional in a basis, namely β{A}[ · ], yielding α β α (u)Iα [ · ]. The point A, having
coordinates u, is held fixed thereby.
Of course (3.42) is not new, but precisely the FRGE in component form (3.12)
from which we started. However, we went through this rederivation since it is
often quite illuminating to think of the flow equation in terms of these differential
geometric concepts and the formal calculus they entail.
For example, at first sight the set of all beta functions {u 7→ β α (u)} appears to
be a mathematical object of a quite different nature than the supertrace func-
tional STr[· · · ] on the RHS of the FRGE. However, in the geometric interpretation
e ē
the latter is essentially the same as β (Φ,Φ) {Ak }, which, as (3.43) shows, is related
to β α (u(k)) simply by a change of basis, though admittedly a colossal one.

(C) General coordinate transformations on T . The functional RG equation

∂t A(t) = β{A(t)} enjoys a manifestly coordinate independent status. While its
component form ∂t uα (t) = β α (u(t)) depends of course on the system of coor-
dinates chosen, it is form invariant under t-independent general coordinate
′
( ′α ) on theory space, u 7→ u (u). Since under this transformation,
transformations
′α ∂u γ
∂t u = ∂uγ ∂t u , the transformed RG equation is seen to be of the same
form as the old one, ∂t u′α (t) = β ′α (u′ (t)), provided the beta functions transform
according to
∂u′α γ
β ′α (u′ ) = β (u), (3.44)
∂uγ

which is exactly the transformation law satisfied by the components of a vector

field.
Note that theory space carries no natural metric or connection, and therefore
there is no intrinsically given covariant derivative with which we could act on β.
The partial derivatives of the components β α ,

∂ α
B α γ (u) ≡ β (u) , (3.45)
∂uγ
 3.1 Geometry and RG Dynamics on Theory Space 65

transform at a generic point u of T like

∂u′α κ ∂uδ ∂uδ ∂ 2 u′α

B ′α γ (u′ ) = κ
B δ (u) ′γ + β κ (u) ′γ δ κ . (3.46)
∂u ∂u ∂u ∂u ∂u

This transformation law is non-tensorial unless the second term on the RHS of
(3.46) vanishes for some reason. There is an important situation where this is
indeed the case, namely when β happens to vanish at the particular point con-
sidered, β α (u) = 0. Then the matrix B = (B α γ ) evaluated at this special point
does indeed transform as a (1, 1)-tensor under a change of coordinates on T . This
simple fact will become relevant when we discuss the critical exponents of fixed
points in the next section.

(D) RG-time independence of β. Throughout our interpretation of the func-

tional RG equation as an infinite-dimensional extension of ∂λ ξ i (λ) = V i (ξ(λ)) we
tacitly exploited already that the RHS of this equation has only an implicit
dependence on λ, namely via ξ(λ). This property does not come for free. The
original FRGE in (3.1), i.e., the one written in terms of dimensionful quantities
does not have this property. It is rather of the type ∂λ ξ i (λ) = V i (ξ(λ); λ) where
the vector field V ( · ; λ) has an explicit λ-dependence as well. In the FRGE it is
due to the explicitly k-dependent B k of (3.4) with the components b̄α ( · ; k); see
(3.9). But following the steps leading from the dimensionful differential equations
(3.10) to the dimensionless ones in (3.12) and (3.13), it becomes clear that the
explicit scale dependence disappears, and β becomes RG time independent, if we
employ k as the universal unit of all dimensionful quantities. This is the reason
why we usually prefer the Ak -variant of the EAA over Γk .

(E) Background independence of the first and second kind. Looking at

the explicit definition of β given in (3.32) with (3.4) and (3.19), it is obvious that
on a given theory space (i.e., for a fixed set of fields, symmetries, etc.) it is only
Rk that remains to be specified before we can evaluate β explicitly. The cutoff
operator is indeed arbitrary to a large extent. While physical observables must
be independent of it, the RG flow, i.e., β and the functions β α (u1 , u2 , · · · ), will
usually depend on it.
Having decided for some Rk we can “turn the crank” and compute the func-
tional traces on the RHS of the FRGE. The vector field β is found then by
evaluating the components β{A}[Φ, ē of (3.32) for all points A of the theory
e Φ]
e Φ)
space and all points (Φ, ē of field space. This shows that nothing more than the
arguments plugged into β{ · }[ · ] appear in the trace computations: the functional
A from the first “slot” and the fields (Φ, ē from the second.
e Φ)
This simple observation has an important consequence: Neither does β depend
on any special field configuration, i.e., a distinguished point of field space, nor
66 The Asymptotic Safety Mechanism

does it depend on any special action functional, i.e., a distinguished point in

theory space.
We refer to these properties of β as Background Independence of the first and
the second kind, respectively.
These names are motivated by the traditional use of the term “Background
Independence” in quantum gravity where it expresses that no special spacetime
and no metric are singled out. This is an example of “Background Independence
of the first kind.” In fact, when it comes to applying the EAA formalism to
ē will include both a dynamical and a background metric.
e Φ)
gravity, the fields (Φ,

3.2 Fixed Points

Fixed points are particularly important and distinguished “inhabitants” of theory
space. They play a prominent role in virtually any RG application, and quantum
gravity is no exception in this respect. In this section we shall discuss some of
their general properties as a preparation for our exposition of Asymptotic Safety
later on.

3.2.1 The Zeros of β

By definition, a fixed point in the dimensionless theory space, A∗ ∈ T , is a point
at which the vector field β has a zero: β{A∗ } = 0. At a fixed point the “velocity”
of any RG trajectory will vanish: ∂t A(t) = β{A∗ } = 0.
Depending on whether one prefers the field or component language the
condition for a fixed point reads, more explicitly, either

β{A∗ }[Φ, ē = 0 ∀ (Φ,

e Φ] ē
e Φ) (3.47)

or
β α (u∗ ) = 0 ∀ α. (3.48)

Within the LPA approximation we saw already a concrete example of (3.47),

namely the differential equation (2.70) for the potential u∗ (φ). Instead, in (3.48),
∗ } denotes the components of A∗ , i.e.,
where u∗ ≡ {uα
∑
ē =
e Φ]
A∗ [Φ, uα e ē
∗ Iα [Φ, Φ] , (3.49)
α

the fixed-point condition has the appearance of infinitely many coupled equa-
tions for infinitely many unknowns: β α (u1∗ , u2∗ , u3∗ , · · · ) = 0, α = 1, 2, 3, · · · . It is a
highly nontrivial question whether this system of equations does actually admit
solutions, and if so, what their characteristic properties are.
Notice that it is the dimensionless couplings that are required to become scale
independent: uα (k) ≡ k −dα ūα (k) = const ≡ uα ∗ . The dimensionful ones continue
 3.2 Fixed Points 67

to be k-dependent even at a fixed point, but their RG evolution is very simple

there. It amounts to a set of power laws

ūα (k) = uα dα
∗ k , (3.50)

in which all couplings scale according to their canonical dimensions dα .

Every fixed point (FP) gives rise to a special RG trajectory, the
fixed point trajectory. At the dimensionful level it is described by the running
action ∑
k 7→ ΓFP
k [Φ, Φ̄] = uα
∗ k
dα
Iα [Φ, Φ̄] . (3.51)
α

This is indeed a nontrivial curve on the (dimensionful) theory space. In the

dimensionless setting, the analogous action reads
∑
k 7→ AFP e ē e ē e ē
k [Φ, Φ] = ∗ Iα [Φ, Φ] = A∗ [Φ, Φ] .
uα (3.52)
α

Logically, AFP
k is also defined for all k ∈ [0, ∞), but it happens to be time inde-
pendent and so the curve degenerates to a point on the dimensionless theory
space, T . In the dynamical system interpretation AFP k corresponds to a particle
that sits at one and the same point at all times. This is possible since at the
point A∗ the “force” β acting on the particle vanishes.
With the fixed-point trajectories ΓFPk and Ak we have found an example of
FP

an exact solution to the dimensionful and dimensionless forms of the FRGE,

respectively.

3.2.2 The Linearized Flow

Next, we ask about the behavior of RG trajectories that stay close to the fixed
point for a while without necessarily sitting on top of it at all times. To answer
the question we Taylor expand the vector field β to first order in deviations from
the fixed point u∗ . The exact FRGE k∂k uα = β α (u) then leads to the following
linear differential equation:
∑ ( )
k∂k uα (k) = B α γ uγ (k) − uγ∗ . (3.53)
γ

Here we encounter the stability matrix B ≡ (B α γ ) of the fixed point under

consideration:
∂ α
Bαγ = β . (3.54)
∂uγ u=u∗

Generically, this matrix has no reason to be symmetric. As a consequence, we

may not expect that it possesses a complete system of eigenvectors in general,
and even if it does, the left and right eigenvectors may be different. Furthermore,
its eigenvalues are not necessarily real [96].
68 The Asymptotic Safety Mechanism

Rather than discussing all cases that could occur we assume here and in the
following that B does indeed admit a complete system of right eigenvectors VJ ≡
(VJα ), J = 1, 2, 3, · · · :
∑
B α γ VJγ = −θJ VJα , ∀ J. (3.55)
γ

In all RG investigations of quantum gravity, up to now, this assumption indeed

turned out to be satisfied. Nevertheless, the stability matrices that occur are
typically not symmetric, and as a consequence, the eigenvalues (−θJ ) often have
a nonzero imaginary part, and the eigenvectors are not orthogonal. The complex
numbers θJ are referred to as stability exponents, critical exponents, or scaling
exponents.
Equipped with a complete system {VJ }, it is easy to write down the general
solution to the linearized RG equation:

∑ ( )θJ
α k0
u (k) = uα
∗ + CJ VJα . (3.56)
k
J

Here, the CJ s denote constants of integration which can be expressed in terms

of the initial conditions,
( ) and k0 is a fixed scale which we identify with the one
present in t ≡ ln kk0 . Note that if some critical exponent θJ is not real, the
solution (3.56) contains an oscillatory component:
( )θJ [ ] ( k )Re θJ
k0 −θJ t 0
=e = cos(Im θJ t) − i sin(Im θJ t) . (3.57)
k k
In a physically relevant solution the CJ s are such that all imaginary parts cancel
in the sum (3.56).
When we move forward along the trajectory (3.56), i.e., when we decrease k,
equation (3.57) implies that the component of u(k) − u∗ in the direction of a
given vector VJ will increase if Re(θJ ) > 0, decrease if Re(θJ ) < 0, and is purely
oscillatory if Re(θJ ) = 0. The basis vector (or “scaling operator,” as it is some-
times called in this context) is then termed relevant, irrelevant, or marginal,
respectively.
So “relevant” is synonymous to “growing under downward evolution” from the
UV to the IR. Conversely, under upward evolution (increasing k) all contributions
to the sum in (3.56) belonging to relevant directions get damped, while those
along irrelevant directions blow up if the pertinent coefficients CJ are nonzero.
Since, by assumption, the vectors {VJ } form a complete system, they provide
us with a (not necessarily orthogonal) basis of the tangent space to T at the fixed
point, TA∗ T . The stability matrix supplies a natural decomposition of TA∗ T in
a relevant, irrelevant, and marginal subspace, which are spanned by the VJ s with
Re(θJ ) > 0, Re(θJ ) < 0, and Re(θJ ) = 0, respectively.
 3.2 Fixed Points 69

3.2.3 Universality of the Critical Exponents

An important property of the critical exponents θJ is that they depend neither
on the system of coordinates on T , {uα }, nor on the cutoff scheme, i.e., on the
operator Rk : They are universal quantities [97]. As an immediate consequence,
when we decompose the tangent space TA∗ T in relevant, marginal, and irrel-
evant subspaces the respective dimensionalities of those subspaces are likewise
universal.

(1) Coordinate independence. The coordinate independence of the θJ s is easy

to see. In (3.46) we wrote down how the partial derivatives B α γ (u) ≡ ∂u∂ γ β α (u)
respond to a general coordinate transformation on theory space, u 7→ u′ (u). There
we mentioned that their transformation law becomes tensorial at points where β
vanishes, i.e., at fixed points. At a fixed point, B α γ (u∗ ) ≡ B α γ in (3.45) coincides
exactly with the above stability matrix. From (3.46) we infer that in going from
the old coordinates uα to the new ones, u′α , the corresponding stability matrices
B and B ′ are related according to
∂u′α κ ∂uδ
B ′α γ (u′∗ ) = B δ (u∗ ) . (3.58)
∂uκ ∂u′γ
Here u∗ and u′∗ are the old and new fixed-point coordinates, respectively,
β ′α (u′∗ ) = 0 = β α (u∗ ). Thus
( B and B
)
′
are connected by a similarity transforma-
tion, B ′ = SBS −1 , S ≡ ∂uγ u′α (u∗ ) , which implies that their spectra coincide.
∂

Therefore, we can be sure to find the same critical exponents, no matter which
coordinate system we use.

(2) Scheme independence. The scheme independence of the critical expo-

nents, i.e., their independence of the cutoff operator Rk , is a consequence of
their coordinate independence. The reason is that a deformation Rk → Rk + δRk
modifies the linear flow at u∗ by not more than a coordinate transformation.
This can be seen most directly from the functional relation (2.26) whose general
structure carries over from scalars to general field systems.
Proceeding in complete analogy with the derivation of the FRGE in component
∑
form, we may expand Γk = α ūα (k) Iα and rewrite (2.26) in terms of the dimen-
sionless cousins of the ūα s. The resulting equations tell us how the individual
couplings respond to the change Rk → Rk + δRk . They assume the form
( )
δuα (k) = f α u(k) , (3.59)
where the f α s depend linearly on δRk . These equations do not couple dif-
ferent scales and describe an “instantaneous” change of the RG trajectories,
uα (k) → uα (k) + δuα (k). Hence, for k → ∞ and u(k) → u∗ in the case of interest,
we see that modifying the cutoff can be compensated by a (local) coordinate
transformation uα → uα + f α , which is known to leave the spectrum of the
stability matrix unaltered.
70 The Asymptotic Safety Mechanism

3.2.4 The Ultraviolet Critical Hypersurface

By definition, the unstable manifold, or synonymously the UV critical hypersur-
face SUV of a given fixed point consists of all points A ∈ T that are “pulled” into
the fixed point under the inverse flow, i.e., when k is increased toward infinity.
Suitable regularity conditions assumed, this set of points indeed constitutes a
submanifold of theory space. It should be visualized as a curved surface embedded
in T , which, while tangent to the unstable, or relevant subspace of TA∗ T precisely
at the fixed point A∗ , describes the true behavior of the flow lines even far away
from the fixed point where the linearized differential equation loses its validity.
Clearly, if a trajectory k 7→ u(k) constituting a solution to the exact RG equa-
tion is such that u(k1 ) ∈ SUV at one scale k1 , then u(k) ∈ SUV for all other scales
as well.
Let us determine the dimensionality of SUV . Close to the fixed point, the gen-
eral solution to the linear RG equation given in (3.56) provides a valid description
of all trajectories, whether they run inside SUV or not. Now, letting k → ∞ in
(3.56), those terms in the sum that correspond to the “relevant” eigenvectors
with Re(θJ ) > 0 vanish in this limit independently of the values of the associated
coefficients CJ . The terms associated with the “irrelevant” directions, having
Re(θJ ) < 0, will diverge, however, unless their coefficients CJ are zero. Finally,
in marginal directions linearized trajectories deviate in the limit k → ∞ from u∗
by a finite, but nonzero amount should their CJ s be nonzero.
Hence, it follows that trajectories running inside SUV , close to the fixed point,
are approximated by (3.56) with CJ = 0 for all J which refer to an irrelevant or
marginal eigenvector VJ :

∑ ( )θJ
k0
uα (k) = uα
∗ + CJ VJα on SUV . (3.60)
k
Re(θJ )>0

Here the sum is restricted to include the “relevant” directions only. All trajec-
tories (3.60) indeed have the property that limk→∞ uα (k) = uα ∗ for any choice of
the remaining, in general, nonzero coefficients CJ .
Conversely, near A∗ , every trajectory inside SUV has the form (3.60) for some
choice of these coefficients. As a consequence, dim(SUV ) equals the number of
independent CJ s that can still be chosen freely, or, stated differently, the dimen-
sionality of the “relevant” subspace of TA∗ T . Therefore, we obtain the important
result that the dimension of SUV equals the number of eigenvalues of B which
possess a positive real part:
{ }
∆UV ≡ dim(SUV ) = # θJ | Re(θJ ) > 0 . (3.61)

Henceforth we denote the number of relevant directions by ∆UV .

At this point it is not obvious that the notion of a “number of relevant eigen-
vectors” makes any sense, given that B is an infinite-dimensional matrix. Later
 3.2 Fixed Points 71

on we shall see however that for a typical fixed point in the physical systems we
are familiar with, ∆UV is indeed a finite, and usually small, number.

3.2.5 Gaussian vs. non-Gaussian Fixed Points

To begin with a simple example leading to a trivial, or Gaussian fixed point
(GFP), consider a set of β-functions which behaves like

β α (u) = −dα uα + O(u2 ) (3.62)

for small values of the couplings. According to (3.13) this amounts to bα (u) =
O(u2 ) and dα ≡ [ūα ]. In this case, uα∗ = 0 for all α with dα ̸= 0 is an obvious
α
solution to β (u∗ ) = 0. The exact stability matrix at this fixed point is very
simple, namely B α γ = −dα δγα . It happens to be diagonal and its negative eigen-
values, the critical exponents, equal exactly the canonical mass dimensions of the
couplings ūα :
θα = dα at a GFP. (3.63)

This equality is the motivation for defining the θs with the explicit minus sign
in (3.55).
In this book we adopt the definition proposed in [98] and refer to any fixed
point whose critical exponents are all equal to the canonical mass dimensions as
a Gaussian fixed point. Conversely, at a non-Gaussian fixed point (NGFP), by
definition, some or all critical exponents differ from their canonical value.
While the defining property (3.63) is a coordinate-independent statement, the
beta functions near a GFP are usually more complicated than (3.62) when
a generic system of coordinates {uα } is used. A linear change of variables
(diagonalization of B α γ ) is then necessary to find the critical exponents.

3.2.6 Linearization about the Scalar GFP

In our investigation of scalar field theory within the local potential approxima-
tion we encountered an example of a Gaussian fixed point. In the LPA, the
fixed-point condition for the dimensionless potential u∗ (φ) is given by the (ordi-
nary) differential equation (2.70), and we remarked already that it admits the
constant solution u∗ (φ) = 4vd /d2 = const.

(1) Let us compute the critical exponents and the scaling operators for this fixed
point. The calculation can be done either in an algebraic way by expanding the
potential as in (2.72) and solving for the coefficients λ2n individually, or “in one
stroke” by using the (partial) differential equation for the complete potential
function.
72 The Asymptotic Safety Mechanism

Here we follow the second strategy and make the following linearization ansatz
for the running dimensionless potential:
( )θ
k0
uk (φ) = u∗ + Y (φ) ≡ u∗ + Y (φ)e−θt . (3.64)
k

We insert (3.64) into the flow equation (2.67) and expand to linear order in Y ,
obtaining
[ ( ) ]
d2 d−2 d ( )
−σd 2 + φ − d Y (φ) = −θ Y (φ) (3.65)
dφ 2 dφ

with the dimension-dependent positive constant

∫ ∞
R(0) (z) − zR(0)′ (z)
dz z 2 −1
d
σd ≡ 2vd [ ]2 . (3.66)
0 z + R(0) (z)

The differential equation (3.65) represents an infinite-dimensional eigenvalue

equation BY = (−θ)Y for the eigenvectors Y and eigenvalues (−θ). The role
of the stability matrix is played by the second-order differential operator on the
LHS of (3.65).
Assuming d > 2 from now on the analysis of (3.65) simplifies by redefining the
function Y in the following way:
x2
√
Y (φ) ≡ e 4 ψ(x) with x ≡ d−2 2σd φ. (3.67)

In terms of ψ, the above eigenvalue condition boils down to an equation that

looks like the time-independent Schrödinger equation of a harmonic oscillator:

1
−ψ ′′ (x) + x2 ψ(x) = Eψ(x) (3.68)
2

with the identification

1 d−θ
E≡ +2 . (3.69)
2 d−2
(2) At this point a very general question arises which we shall re-encounter
in gravity later on. Namely, what precisely is the space of functions the devia-
tions from the fixed point are supposed to live in? Or concretely, what are the
subsidiary or boundary conditions to be imposed on eligible solutions to (3.68)?
If we interpret the eigenvalue problem (3.68) as a quantum mechanical
Schrödinger equation, we know that ψ must be square-integrable. However, in
the present case ψ is essentially a potential, and there is no obvious physical
necessity for this potential to be square-integrable.

(3) Nevertheless,
∫∞ let us begin our discussion by finding all solutions to (3.68)
with −∞ dx|ψ|2 < ∞. From quantum mechanics we know that such solutions
 3.2 Fixed Points 73

exist only for a discrete set of energy eigenvalues, E = En ≡ n + 12 , n = 0, 1, 2, · · · ,

and that the corresponding eigenfunctions are Hermite polynomials times an
x2
exponential, ψn (x) ∝ Hn ( √x2 )e− 4 . This implies the following solutions to the
original equation (3.65):
( √ )
Yn (φ) ∝ Hn 1
2
d−2
σd φ ,
(d ) (3.70)
θn = d − 2 −1 n

We observe that the scaling operators are given by polynomials of degree n. They
are even (odd) functions of φ if n is even (odd). If we impose symmetry under
φ → −φ, only the even eigenfunctions Y2n (φ) are permitted, and so the gen-
eral solution to the linearized flow equation is the following linear superposition
analogous to (3.56):
∞
∑ ( )θ2n
k0
uk (φ) = u∗ + C2n Y2n (φ) . (3.71)
n=0
k

The pertinent scaling exponents are given by θ2n = d − (d − 2)n. A comparison

with (2.74) reveals that the θ2n s are precisely the canonical mass dimen-
∫ d of2nthe coupling constants λ̄2n multiplying the basis functionals I2n [ϕ] ≡
sions
d xϕ (x). Thus, having shown that all critical exponents equal the classical
dimensions, θ2n = [λ̄2n ], we have established that the fixed point u∗ (φ) = const is
indeed a Gaussian one.

(4) We also observe that the vast majority of eigenvectors are associated with
irrelevant directions since θ2n = d − (d − 2)n is negative for most values of n.
Relevant directions (θ2n > 0) can occur only as long as n is smaller than d/(d − 2).
In d = 4 dimensions, e.g., there exist two relevant, one marginal, and infinitely
many irrelevant directions. This is summarized in Table 3.1.
In the third column of the table we show the general structure of the
eigenfunctions, suppressing numerical coefficients.

Table 3.1 Relevant, marginal, and irrelevant scaling

fields (eigen-directions) at the scalar GFP in d = 4
dimensions according to the LPA

n θ2n = 4 − 2n Y2n (φ) Type

0 4 1 Relevant
1 2 φ2 + 1 Relevant
2 0 φ4 + φ2 + 1 Marginal
3 –2 φ6 + φ4 + φ2 + 1 Irrelevant
. . .
.. .. .. ”
74 The Asymptotic Safety Mechanism

Thus, for the special example of the GFP, and within the local potential
approximation, we see explicitly that “relevant” eigendirections are indeed an
exception, while irrelevant ones are the rule.

(5) Actually it is easy to generalize this result beyond the LPA. If we invoke a
double expansion of Γk in both the number
∫ of fields and the number of derivatives,
additionally including all terms ūm d4 x ϕ(∂µ ∂ µ )m ϕ, for example, into the action
(m = 2, 3, 4, · · · ), then the new higher derivative terms are all irrelevant. In this
example the running couplings have the dimension [ūm ] = 2 − 2m, which can only
become smaller by increasing m.
This is also the general situation: Given a monomial containing n powers of
ϕ and 2m derivatives acting in an arbitrary way on the fields, it is sufficient
to know that [∂µ ] = +1 in order to conclude that adding derivatives unavoidably
decreases the dimension of the coupling and, as a consequence, the corresponding
eigenvalue of the stability matrix. Hence the “most relevant” monomials of this
kind are those with only two derivatives.

(6) So far we have focused on square integrable solutions to the Schrödinger-like

eigenvalue equation (3.68). For completeness we mention that if we relax this
condition a continuous infinity of further solutions will arise. For every θ ∈ R
there exists an eigenfunction
x2
( )
ψθ (x) = e− 4 1 F1 − d−θ , 1 x2
d−2 2 2, , (3.72)

where 1 F1 denotes the confluent hypergeometric (or Kummer’s) function. While

the solution (3.72) is even under x → −x, there also exists a similar odd one. For
all θ ̸= θn the functions ψθ (x) and the corresponding Yθ (φ) grow very rapidly for
x → ±∞ or φ → ±∞:
x2
( ( ) )
ψθ (x) ∼ e 2 , Yθ (φ) ∼ exp + d−2
4σd φ
2
. (3.73)

This asymptotic behavior makes it obvious that, for θ ̸= θn , all Yθ s are non-
polynomial functions of the field. As a result, perturbation theory is unable to
detect these so called “Halpern-Huang directions.” Hence their discovery, by non-
perturbative methods, came as quite a surprise [99]. Their status and physical
role are not yet fully understood [100, 101].

3.3 Asymptotic Safety: The Basic Idea

After the detailed preparations in the previous sections we are now in the position
to outline the key idea of Asymptotic Safety in a concise way. The goal of this
section is to describe the basic picture in a language as geometric as possible.
The discussion of the technical issues related to the investigation of this scenario
is relegated to subsequent chapters of this book.
 3.3 Asymptotic Safety: The Basic Idea 75

At first let us recall the main questions which the Asymptotic Safety program
as a whole wants to address:
∫
(i) Given the fact that functional integrals such as DΦ b e−S[Φ̂,Φ̄] are mathemat-
ically ill defined and make sense only in the presence of a UV cutoff, we would
like to find out for which measure and bare actions S the limit of an infinite
UV cutoff can be taken consistently. This limit must give rise to a quantum
field theory which respects all basic physical principles which we consider
indispensable (Hilbert space positivity, Background Independence, etc.).
Perturbation theory can give only a partial answer here; it identifies the
perturbatively renormalizable actions. However, many potentially interest-
ing cases (including quantum gravity) lie beyond the scope of perturbation
theory. Hence an appropriate notion of non-perturbative renormalizability is
called for.
(ii) For the class of bare actions that can have an acceptable limit we would like
to know whether it is unique, and if not, to what extent there are ambigu-
ities, and most importantly how the bare parameters in S must depend on
the cutoff when the UV limit is taken.
(iii) Once it is known how to take the limit(s), we are confronted with a well-
defined, but presumably very complicated, functional integral which may
contain non-perturbative information. Therefore, we need a corresponding
non-perturbative calculational method in order to evaluate it and derive
predictions for observable quantities.

As we shall see, the framework of the Effective Average Action is well suited
for exploring all three of the above problems in a self-contained manner. In the
EAA framework, they amount to the following three steps in the analysis of the
RG flow, which we assume as given for the time being:

(i) Determine all fixed points of the RG flow.

(ii) Compute the linearized flow at a given fixed point and determine its
subspace of “relevant” directions.
(iii) Compute the nonlinear extension of the latter, i.e., the fixed point’s entire
UV critical hypersurface, SUV .

Let us now describe the various parts of this program in more detail.

3.3.1 The Input: T and β

Our search for non-perturbatively renormalizable theories begins by specifying
a theory space of well-behaved action functionals, T . In this step we fix the
nature of the fields on which the actions are supposed to depend, the symmetry,
covariance, and regularity properties of the actions and their field arguments, as
well as any additional subsidiary conditions we might want the functionals to
satisfy. As a further, purely technical input we also pick a cutoff operator Rk .
76 The Asymptotic Safety Mechanism

Given these “input data” we have all that is needed in order to write down
the associated FRGE and the associated vector field β which determines the RG
flow on T .

3.3.2 The Quest for Complete Trajectories

As discussed, the FRGE is not plagued by any UV divergences. In the previ-
ous chapter we derived the flow equation from a functional integral which was
equipped with a UV cutoff at some scale Λ. While it is unclear whether or not
the integral per se admits a meaningful limit for Λ → ∞, we saw that this limit
can be taken straightforwardly at the level of the flow equation.
So, if the FRGE is perfectly finite, where have all the notorious questions
related to the existence of a UV limit gone? The answer is simple: Now they
manifest themselves as questions about the nature of the solutions to the FRGE,
i.e., the RG trajectories.
The construction of a quantum field theory with the UV cutoff removed
amounts to integrating out all field modes. At the level of Γk this means lowering
the IR cutoff from “k = ∞” to k = 0. In fact, we saw already that Γk→∞ is closely
related to the bare action S, while, at the endpoint of the RG trajectory, Γk=0
equals the ordinary effective action.
Hence, if we require that there is a fundamental quantum field where all field
modes can be integrated out, it is necessary that there exists a well defined action
functional Γk for each value of the scale parameter k ∈ [0, ∞).
Geometrically speaking, and using the dimensionless language now, these
actions constitute a complete RG trajectory, i.e., a trajectory that never leaves
the space T of “acceptable” actions, neither at finite scales, nor in the limits
k → ∞ or k → 0, respectively.
Thus, the EAA-based search for the (non-)perturbatively renormalizable
theories, i.e., those admitting a meaningful limit k → ∞, essentially consists in
finding all complete RG trajectories which exist on the theory space chosen:
A• : [0, ∞) → T , k 7→ Ak .
This reformulation turns the original renormalization problem into a question
about the long-time behavior of trajectories in the infinite-dimensional dynamical
system related to (T , β).
As such, it is on a par with many other important, but notoriously difficult,
problems in physics and mathematics. They range from establishing ergodicity or
the various forms of chaoticity to problems in biology, medicine, and economics,
to mention just a few of the fields [94].

3.3.3 The “Asymptotic Safety Scenario” of the UV Limit

When trying to construct complete RG trajectories the most severe difficulty one
must overcome is to make sure that the trajectories do not leave theory space in
the limit k → ∞, i.e., develop an unacceptable behavior at high momentum scales.
 3.3 Asymptotic Safety: The Basic Idea 77

This dangerous possibility is the new incarnation of the old UV renormalization

problem.
Given the previous discussion, it suggests itself to use a fixed point of the RG
flow in order to control the k → ∞ limit of the complete RG trajectories. In fact,
Asymptotic Freedom (well-known from QCD, for example) is a realization of
this idea restricted to the case where the underlying fixed point is the Gaussian
one. For the Einstein–Hilbert action, Asymptotic Freedom fails, however, due to
the negative mass dimension of Newton’s coupling. Phrased differently, General
Relativity is not contained in the unstable manifold of the corresponding GFP.
In his seminal paper [10], Steven Weinberg formulated a conjecture about the
existence of the infinite cutoff limit in quantum gravity, which became known as
the Asymptotic Safety Scenario.
This conjecture has been the main inspiration for the whole Asymptotic Safety
program in its modern EAA-based form, and also gave the name to this entire
direction of research.
Weinberg’s conjecture consists of two parts. In a somewhat simplified form
(refinements will be discussed later), and in the same language as used above,
they read as follows:

(W1) The theory space contains a non-trivial fixed point, and this fixed point
has a low-dimensional UV critical hypersurface.

(W2) Every RG trajectory that does not hit this fixed point in the UV limit
ultimately leaves theory space, developing unacceptable singular behavior at high
scales.

If both (W1) and (W2) are true, a given RG trajectory has an acceptable UV
limit, and hence signals the existence of a “non-perturbatively renormalizable”
field theory, if and only if its endpoint in the ultraviolet is given by the non-trivial
fixed point of the RG flow. The theory is then safe from the threat of divergences
in physical quantities at asymptotically high momentum scales since we know for
sure that when k → ∞ the trajectory always stays (increasingly) close to an inner
point of theory space, namely the fixed point. This behavior is exactly what gave
rise to the name “Asymptotic Safety.”
Let the functional A∗ ∈ T , having components uα ∗ , be a nontrivial fixed-point
action, i.e., β (u∗ ) = 0 ∀α, and furthermore let the trajectory k 7→ uα (k) be an
α

“asymptotically safe” one, i.e., limk→∞ uα (k) = uα ∗ . It is clear then that this tra-
jectory has a very small ”velocity” ∂t uα = β α when it approaches the fixed point
since, directly at A∗ , the β-functions vanish completely. As a result, the trajectory
“uses up” an infinite amount of RG time while staying within an infinitesimally
small neighborhood of A∗ .
During this eternal approach of the fixed point’s exact location, the dimension-
less Effective Average Action Ak almost does not move, while at the dimensionful
78 The Asymptotic Safety Mechanism

level the components of Γk evolve according to the power laws ūα (k) = uα dα
∗ k ,
as we saw in Section 3.2.1.

3.3.4 Predictive Power

It is essential to understand in what sense and to which degree theories based on
asymptotically safe RG trajectories can have predictive power.

(1) Such theories are intended to constitute fundamental quantum field theories
which remain consistent and predictive even at the highest momentum scales.
This distinguishes them from effective field theories, which require an increasing
number of coupling constants at high momenta that are in no way constrained
theoretically and therefore must be taken from experiments.
As we explain next, Asymptotic Safety can “tame” this plethora of infinitely
many undetermined constants and reduce them to just a few parameters that
must be extracted from experiments. The situation is then comparable to an ordi-
nary perturbatively renormalizable theory, a prime example being QED, which
has only two theoretically undetermined parameters, namely the electron’s mass
and charge.

(2) According to (W1), theory space contains a fixed point, A∗ . As described

in Section 3.2.2, we explore the RG flow in its vicinity by linearizing it about
A∗ . This leads to a decomposition of the tangent space TA∗ T into the subspaces
spanned by the relevant, marginal, and irrelevant eigendirections of the stabil-
ity matrix. These subspaces correspond to critical exponents θJ with positive,
vanishing, and negative real parts, respectively.
Furthermore, (W1) and (W2) imply that every complete trajectory runs into
the fixed point when k is increased toward infinity, or in other words, when we
proceed along the trajectory in the direction of the inverse RG flow, i.e., in the
direction from the IR to the UV.
Recalling the discussion in Section 3.2.4 we see that this is tantamount to
saying that a trajectory is complete if, and only if, it lies in the UV critical
hypersurface of an RG fixed point.
Indeed, in Section 3.2.4 we defined the set of all points A ∈ T that are dragged
into a fixed point A∗ under the inverse flow to constitute the UV critical surface
SUV ≡ SUV (A∗ ).

(3) If uα (k1 ) is an arbitrary point on a complete trajectory k 7→ uα (k), this point

is situated within the UV critical surface when k = k1 . By the very definition of
SUV this also applies to all other points visited by uα (k) at higher and lower
scales k.
 3.3 Asymptotic Safety: The Basic Idea 79

Theory space

UV critical
surface

Figure 3.1. A two-dimensional UV critical hypersurface embedded in a three-

dimensional theory space

In the limit k → ∞, when the linearization about A∗ is a good approximation,

the trajectory uα (k) is therefore given by the sum (3.60). It involves only terms
belonging to the relevant subspace.
The situation is illustrated in Figure 3.1, which shows examples of trajectories
that hit A∗ for k → ∞, as well as others that run outside SUV . The latter fail
to reach the fixed point and get driven away from it under the inverse flow.3
According to (W2), these trajectories will leave theory space ultimately when
k → ∞ and thus they do not correspond to quantum field theories with a contin-
uum limit.

(4) Let us switch for a moment from the inverse flow (increasing k) to the
forward flow (decreasing k). So now we follow the trajectories of Figure 3.1
in the direction of the arrows. From this perspective, trajectories outside SUV
are attracted by A∗ toward SUV along the irrelevant directions. Conversely,
the trajectories inside SUV , those corresponding to field theories admitting a
continuum limit, are pushed away from the fixed point along one of its relevant
directions.
From this geometric picture and the assumed validity of (W1) and (W2) we can
derive an estimate of how many complete trajectories and therefore inequivalent
fundamental quantum field theories can exist.
According to the discussion leading to (3.61) the dimensionality of the UV-
critical hypersurface, dim(SUV ) ≡ ∆UV , is given by the number of relevant
directions, i.e., the number of θJ s with Re(θJ ) > 0, which, according to (W1),
is a small finite number.

3 Recall again that the (forward) flow amounts to decreasing the mass scale k, while the inverse
flow means increasing it. Note also that in this book (and most parts of the literature) the
arrows in all figures point in the forward direction, i.e., toward smaller values of k.
80 The Asymptotic Safety Mechanism

As a consequence, there exists a (∆UV − 1)-parameter family of RG trajecto-

ries approaching A∗ in the UV, since we must specify (∆UV − 1) constants of
integration in order to pick a specific trajectory on SUV .
In the linear approximation, these constants of integration are the CJ ’s in
(3.56), up to an irrelevant common factor that can be absorbed into the refer-
ence scale k0 . This leaves us with the (∆UV − 1) free parameters which label
asymptotically safe trajectories and their inequivalent quantum theories.

(5) In practice one could find those trajectories as follows. Within SUV , we draw
a small ∆UV -dimensional ball of radius ε around A∗ ; its surface is a (∆UV − 1)-
sphere. Then we pick a point on this sphere, use it as an initial condition for
the RG equation, and solve the flow equation in the forward direction toward
smaller values of k, proceeding from the UV to the IR. Finally taking the limit
ϵ → 0 we obtain a trajectory which is indeed complete, the existence of the
limit being guaranteed by the (conjectured) validity of (W1) and (W2). We can
repeat the procedure with all other points of the (∆UV − 1)-sphere, which shows
that its points are in a one-to-one correspondence with the asymptotically safe
trajectories the fixed point gives rise to.
So, again, the conclusion is that the family of UV complete, or “fundamental”
quantum theories that can be obtained by the Asymptotic Safety construction
is labeled by (∆UV − 1) parameters. It is therefore necessary to perform a cor-
responding number of measurements in order to find out which one of those
theories is actually realized in nature.

(6) In this scenario there is a simple way of visualizing the totality of all predic-
tions that are implied by a given fixed point: they are encoded in the way SUV is
imbedded into theory space. Near the fixed point, SUV is linearly approximated
by the “relevant” subspace of TA∗ T . But away from A∗ when the nonlinearities
in the beta functions become significant, this linear subspace of the “tangent
plane” gets deformed and becomes a complicated curved surface in general.
Locally the embedding of SUV into T is parametrized by functions of the form

( ) ( )
w1 , w2 , · · · , w∆UV →7 uα w1 , w2 , · · · , w∆UV . (3.74)

( )
We may use the parameters w = wi , i = 1, . . . , ∆UV , as local coordinates on
SUV , and then describe the complete trajectories on SUV concisely as k 7→ wi (k).
Thus, since on all scales the physical system under consideration must be rep-
resented by a point on the, by assumption, low-dimensional hypersurface SUV ,
knowledge of the few parameters w1 , · · · , w∆UV is sufficient to determine all the
infinitely many couplings uα by using the embedding map (3.74). This makes it
clear that, geometrically speaking, the shape of SUV inside T encapsulates all
possible predictions due to A∗ .
 3.3 Asymptotic Safety: The Basic Idea 81

Note that the predictions that can be made on the basis of a single given
asymptotically safe theory, based on a certain trajectory k 7→ w(k), is only a sub-
set of all possible predictions compatible with Asymptotic Safety with respect
to A∗ since this specific theory evaluates the embedding functions only along a
line, k 7→ uα (w(k)). On the other hand, the shape of SUV is really a geometrical
way of summarizing the predictions of all possible asymptotically safe theories
that are constructable at A∗ .

(7) Of course, there is also the possibility that nature does not adhere to the
principle of Asymptotic Safety, which could actually be tested by setting up
experiments searching for violations of the embedding maps (3.74). Thus the
Asymptotic Safety hypothesis is falsifiable, at least in principle.

The wealth of information contained in the embedding maps (3.74) is indeed

the main difference between an effective field theory and an asymptotically safe
fundamental one [102, 103]. In an effective field theory the RG trajectories explore
increasingly higher dimensional subspaces of theory space at higher scales, while
in Asymptotic Safety the trajectories will always stay on a low-dimensional
submanifold. This is precisely what “tames” the flood of phenomenological
parameters that tend to render effective theories inefficient at high scales.

3.3.5 Generalizations and Refinements

Above we have drawn a slightly simplified picture of the Asymptotic Safety
approach, trying to be as geometrical and intuitive as possible. In concrete
realizations of this framework a number of generalizations, refinements, and addi-
tional technical assumptions might be required. In this subsection we discuss
some of them.

(1) Essential vs. inessential couplings. It might happen that some of the
couplings uα do not affect the explicit expressions for observables. Therefore,
the observables may have a well-defined UV-limit even if these special couplings
diverge for k → ∞ rather than assume fixed-point values. Following Weinberg [10]
we refer to such couplings that can be tolerated to diverge in the UV limit without
ruining the theory’s predictivity and mathematical consistency as inessential
couplings. In standard field theory a well-known example of inessential couplings
are the wave function renormalizations associated with different fields. They do
not enter observables like reaction rates or cross sections and can be changed by
field redefinitions. Conversely, all other couplings, which are required to approach
finite fixed-point values, are termed essential couplings.
As a consequence, the coupling constants appearing in the general picture of
Asymptotic Safety developed above should include the essential ones only.
82 The Asymptotic Safety Mechanism

(2) Several fixed points. In general there can be more than one fixed point
on a given theory space of (essential) couplings which is suitable for constructing
an asymptotically safe UV limit. In this case the theory space supports multiple
“universality classes” of qualitatively different quantum field theories based on
different UV fixed points.

(3) Infinite dimensional SUV . In concrete examples one also might encounter
fixed points with an infinite-dimensional UV critical hypersurface. Even in this
case Asymptotic Safety may result in predictive theories if the number of pre-
dictions grows faster than linearly with ∆UV when ∆UV → ∞. (We refer to [98]
for a detailed discussion and an example of this possibility.)
Furthermore, there may be additional physics principles beyond renormal-
izability (stability criteria, unitarity requirements, etc.) which put additional
restrictions on the admissible RG trajectories within SUV and allow to fix a
priori free parameters on theoretical grounds.

3.4 A First Example: Gravity in 2 + ε Dimensions

Returning now to quantum field theories of gravity it is certainly far from obvious
that one can define a nonperturbative infinite cutoff limit à la Asymptotic Safety.
In particular there is no general reason why there should exist an RG fixed point
with the right properties.
Nonetheless, already at the time of Weinberg’s seminal paper [10], a simple
toy model was available which displayed such a fixed point, namely gravity in
d = 2 + ε dimensions, with a small ε > 0.
In exactly two dimensions, Newton’s ∫ dconstant is dimensionless and thus a
√
theory based on the bare action ∝ G 1
d x gR(g) with d = 2 is formally power
counting renormalizable in perturbation theory. However, in d = 2, this action
happens to be a topological invariant, essentially the Euler number, and so in
order to maintain a nontrivial kinetic term for the metric fluctuations the best
we can do is to compute the beta functions in d ̸= 2 dimensions and study their
behavior near d = 2 thereafter.
In the “pre-FRGE” era, dimensional regularization had been used for this
purpose, which gives rise to two types of poles in 1/ε. Besides the usual ones
due to UV divergences there are also “kinematical poles” which originate from
the vanishing of the kinetic term for ε → 0. Absorbing both types of divergences
in renormalized quantities, the bare Newton constant is turned into a floating
coupling constant G(µ), where µ is the familiar normalization mass scale of the
dimensional regularization scheme. Since µ is not an observable, only the floating
of Newton’s constant relative to another observable has intrinsic meaning; in
practice, the cosmological constant or certain matter couplings have been used
for this purpose.
 3.4 A First Example: Gravity in 2 + ε Dimensions 83

A number of perturbative calculations had been performed in 2 + ε dimen-

sions. They differed in their precise computational setup, but they all led to the
following Callan–Symanzik equation at one-loop order of perturbation theory:
d ( )
µ g(µ) = ε − bg g ≡ β pert (g). (3.75)
dµ
Here g(µ) ≡ µε G(µ) is the dimensionless Newton constant, and the coefficient b
in the perturbative beta function is dimensionless as well.
While all calculations agree that b is positive for pure gravity, one can distin-
guish three different physical settings based on the value of b they give rise to.
Within each of these settings b is universal though.
For quantum gravity coupled to ns real, minimally coupled scalar matter fields
the results are
b1 = 23 (1 − ns ) [10, 104, 105]
b2 = 23 (19 − ns ) [10, 106–109] (3.76)
b3 = 23 (25 − ns ) [110–116]
Obviously the metric fluctuations and the scalar fields always drive g(µ) in oppo-
site directions, but the graviton contribution is different in the three cases: it
can offset 1, 19, or 25 scalars, respectively. (Later on the very same numbers
reappeared also in FRGE-based calculations.)
The explanation for the differences among the groups are two critical choices
which underlie any of the calculations:
First, for spacetimes with a boundary, a scale-dependent G(µ) ∫ can be deduced
2+ε √
from both the coefficient of the “bulk” gravitational action d x gR and
the corresponding Gibbons-Hawking surface term. In general, the resulting beta
functions are found to disagree. In fact, b1 above was obtained from the surface
term, while b2 and b3 are obtained from the bulk action instead.
Second, the calculations may employ different field parametrizations for
the gravitational fluctuations: the derivation of b1 and b2 employs the linear
background split gµν = ḡµν + hµν , while b3 is obtained with the exponential
parametrization gµν = ḡµρ (eh )ρν . In either case, ḡµν is a classical background
metric and h = (hµν ) is the elementary quantum field.
The deeper meaning of these classes of beta functions is still not fully
understood; see [117] for a discussion.
The perturbative RG equation (3.75) displays two fixed points, a trivial “Gaus-
sian” one at g∗ = 0 and, remarkably enough, also a non-trivial one of the kind we
are looking for in relation to Asymptotic Safety. It is located at
g∗ = b−1 ε (3.77)
and thus merges with the trivial one in the limit ε → 0. The non-vanishing
fixed-point value of Newton’s constant is indicative of an interacting, or
84 The Asymptotic Safety Mechanism

“non-Gaussian” theory at small but nonzero values of ε. Since g∗ can be made

arbitrary small, one may hope that perturbation theory is actually reliable in this
case so that the perturbative fixed point (3.77) “lifts” to a full-fledged NGFP on
the infinite-dimensional theory space.
On the basis of this result (3.77), Weinberg argued that gravity in 2 + ε
dimensions could indeed be asymptotically safe [10].
Since at the time he proposed Asymptotic Safety the modern functional RG
methods were not available yet, extensions to the non-perturbative level and/or
to four spacetime dimensions were not possible. Thus the perturbatively accessi-
ble fixed point in d = 2 + ε dimensions remained for a long time the only evidence
in support of the conjecture that pure quantum gravity is non-perturbatively
renormalizable. This only changed with the advent of the Effective Average
Action for gravity, which is introduced in the next chapter.
 4
A Functional Renormalization Group
for Gravity

The purpose of this chapter is to motivate and explain a particular proposal

for a generally covariant and Background Independent notion of “coarse grain-
ing” or “averaging” that is applicable in a gravitational context, to implement
it concretely in a suitably constructed scale dependent action functional, and
finally to establish an exact renormalization group equation for the functional
thus defined [14].
This setting, employing the gravitational Effective Average Action, has been
used in almost all modern investigations of Asymptotic Safety.1 Nevertheless,
one should emphasize that there exists no such thing as the functional renor-
malization group for gravity. In principle, other approaches and formulations are
conceivable as well.
Even if one focuses, as we do here, on continuum-based approaches and tries
to follow the example of the scalar EAA from Chapter 2 as closely as possible
it is not immediately obvious how to proceed. When one leaves the safe terrain
of matter field theories on classical spacetimes, a considerable number of general
concepts and methods become inapplicable. Typically there exists no “canonical”
solution to these problems that would indicate the way to go. Thus the gravi-
tational EAA as we introduce it here arose in a long process of trial and error
as a compromise between practical applicability and mathematical elegance or
“naturalness.”
It should also be stressed that finding a generally covariant and Background
Independent generalization of the exact renormalization group is of great impor-
tance in its own right and should not only be seen in connection to Asymptotic
Safety. Its potential applications go well beyond the search for asymptotically
safe theories. For instance, the entire framework of the gravitational EAA also
offers useful tools for the effective field theory approach to gravity. The EAA

1 Notable exceptions are the non-standard perturbative approach in [118–120] and an exact
treatment of the restricted functional integral over all metrics with two Killing vector fields
[121–123]. (See also [98, 124].)
86 A Functional Renormalization Group for Gravity

defines an effective field theory at a certain scale, and the corresponding flow
equation can be used to “evolve” it to another scale. In fact, the exploration of
effective infrared theories of gravity provided a second strong motivation for the
original work [14] independently of Asymptotic Safety.
As we will see explicitly later in this chapter, one of the advantages of the
gravitational EAA framework is that it completely decouples the notorious UV
problems of quantum gravity from the “foundational” questions related to the
search for a suitable notion of an exact renormalization group in gravity, within a
fully Background Independent and generally covariant setting. Intuitively speak-
ing the reason for this decoupling is the difference between the functional integral
and the FRGE-based quantization methods which we observed already: Work-
ing with the functional integral directly, the UV divergences are “defects” of the
basic mathematical object itself. The indirect way via the FRGE bypasses these
difficulties: The FRGE is perfectly finite, and the UV divergences only resurface
at the level of its solutions.
For this reason, the problems related to UV divergences will not play an impor-
tant role in the present chapter. Instead, the focus is on the conceptual differences
between simple non-gauge matter theories on rigid spacetimes, such as the scalar
theory from Chapter 2, and gravity. These differences are mostly due to the
requirements of Background Independence and general covariance.

4.1 What is “Metric” Quantum Gravity?

In this chapter we are going to explain all relevant aspects by means of metric
quantum gravity as the so far best studied example. Thus the first question to
be clarified is what exactly should be “metric” in the quantum theory we are
aiming at.
A successful quantum theory of gravity should be able to reproduce classi-
cal General Relativity in a certain limit. We know that classical gravity can
be described in a large variety of physically equivalent but mathematically dif-
ferent ways, employing different field variables for instances (metric, tetrads,
spin-connection, etc.). Hence, at the quantum level, there is potentially a sim-
ilar multitude of different theories, based on different variables. These theories
may or may not be equivalent beyond the classical regime. In this sense, the
weakest requirement on a theory that deserves being called “metric” would be
roughly as follows: It is a quantum theory containing a certain operator gbµν (x),
symmetric in µ and ν, which in some states |Ψ⟩ possesses an expectation value
⟨Ψ|bg µν (x)|Ψ⟩ ≡ gµν (x) with the properties of a classical Riemannian metric, i.e.,
it is a symmetric (0, 2) tensor field defining a non-degenerate positive-definite
bilinear form.2

2 Unless stated otherwise we assume an Euclidean signature, but the same remarks apply to
the Lorentzian case mutatis mutandis.
 4.1 What is “Metric” Quantum Gravity? 87

Here it is left open whether gbµν is an elementary field variable, or a composite

operator built of elementary fields. The most familiar example of the latter case
is the representation gbµν = ebaµ ebbν ηab with the tetrad field ebaµ now considered
the elementary one.3 At this stage we make our definition of “metric quantum
gravity” more precise and require that gbµν is not a composite operator.
At the level of Lagrangian functional integrals, we would then integrate over
commuting functions gbµν which take the place of the metric operator gbµν .4 As
a consequence we would now say that “metric quantum ∫ gravity,” by defini-
tion, is a theory defined via an integral of the form Db gµν exp(−S) in which
the commuting analog of the metric operator, gbµν , serves as the variable of
integration.
Henceforth we shall use the term metric quantum gravity, or synonymously
Quantum Einstein Gravity, in this general and still somewhat vague sense. It is
vague not only because the ultimate “theory” will depend on how the UV limit of
the functional integral (if any) is taken, but also for the following more mundane
reason.
Recall that above we insisted that there exist some states |Ψ⟩ in which gµν (x) ≡
⟨Ψ|bg µν (x)|Ψ⟩ has the properties of a non-degenerate metric. In the functional
integral language this expectation value reads
∫
gµν (x) = Db gαβ gbµν (x) e−S[bg] (4.1)

with suitable boundary conditions imposed and sources included in S. In general,

one would expect that besides the “classically appearing” states |Ψ⟩ there exist
e in which the expectation value of gbµν has
also other physical states, say |Ψ⟩,
e g µν |Ψ⟩
no interpretation as a classical metric. For example, gµν = ⟨Ψ|b e may not be
smooth, or the matrix gµν may degenerate, det(gµν ) = 0, so that neither a volume
√
element g nor an inverse metric g µν exists. One could even imagine a state |Ψ0 ⟩
in which the expectation value of the metric operator vanishes identically all over
spacetime:
⟨Ψ0 |b
g µν (x)|Ψ0 ⟩ = 0. (4.2)
We would then expect the gravitational system to be in a state of unbroken
diffeomorphism invariance.
These terms allude to an analogy with spontaneous symmetry breaking in
magnetic systems. The correspondence is between the magnetization density M c
and the metric gbµν , respectively. If they develop an expectation value this can
lead to a breakdown of the original vacuum symmetry, namely invariance under
spatial SO(3) rotations in the first, and general coordinate transformations in
the second case.

3 We are cavalier here concerning the definition of operator products at the same point.
4 According to the conventions used throughout this book both integration variables and
operators are indicated by a hat. The difference will be clear from the context and no
confusion should arise.
88 A Functional Renormalization Group for Gravity

From this perspective, gravity with ⟨Ψ|b g µν |Ψ⟩ ≡ ηµν being equal to the
Minkowski metric amounts to a broken phase of the underlying theory. The
“vacuum condensate” ηµν plays a role analogous to the spontaneous magneti-
zation of a ferromagnet. It leaves a small subset of coordinate transformations
unbroken, namely those that form the Poincaré group.
The possibility of an unbroken phase of gravity might appear somewhat
“exotic” at first sight, but there are also good reasons to believe that it is actu-
ally not, and that the unbroken phase even admits to the simpler description
[125]. After all, in typical examples of spontaneous symmetry breaking in parti-
cle and condensed matter physics the much harder problem is usually to analyze
the broken phase. In the case of gravity we are perhaps biased as we have no
experimental or observational access to the unbroken phase.
The reason for mentioning the possibility of an unbroken phase at this point of
our discussion is because in the above we agreed that it is plausible to assume that
the integral (4.1) for the expectation value gµν (x) is a bona fide classical metric
in some states (i.e., for some boundary conditions, sources in S, etc.), but not
in others. Therefore, when we leave the state unspecified, the expectation value
gµν (x) is at best a symmetric tensor field, but not a metric, since it does not
necessarily satisfy the additional requirements of constituting a non-degenerate
positive-definite quadratic form.
Thus accepting that the left-hand side of (4.1) is “just” a tensor field without
additional properties, it then is natural to ask the question: What precisely are
the properties of the functions gbµν on the right-hand side of (4.1) over which we
perform the functional integral? Are they metrics, and therefore non-degenerate
by definition, or generic tensor fields that can degenerate in all possible ways,
or perhaps something in between with only some “mild” form of degeneracy
allowed?
Technically speaking it would seem easiest to integrate over generic tensors
since they form a vector space. If instead we restrict the integral to true metrics
then the domain of integration is a complicated curved submanifold thereof.
In the first case “Db gµν ” could be a simple translational invariant measure, but
in the second case we would have to cover the manifold of all metrics with a
nontrivial atlas of coordinate charts and express the functional integral as the sum
of their respective contributions [126]. On each chart there are local coordinates
pb1 ( · ), pb2 ( · ), · · · , pbd(d+1)/2 ( · ) then, themselves functions over spacetime, which
parametrize the genuine metrics in a suitable way: gbµν ≡ gbµν (b p1 , pb2 , · · · ). In this
∏
case “Db gµν ” would boil down to i Db pi times the Jacobian factor implied by
the parametrization.
At this point there is no way of answering the question about the “correct” or
“best” way of choosing the integration variables. For the time being we therefore
subsume all of the above options under the heading of metric quantum grav-
ity. When compared to the direct construction of the functional integral, the
flow-equation based approach has the advantage that the choice of integration
 4.2 What is “Coarse Graining” in Gravity? 89

variables can be deferred until a later stage of development. At this point compu-
tational results are already available which may guide us toward the physically
interesting choices.
When we try to develop a renormalization group for gravity along the lines of
the scalar EAA in Chapter 2, the main role of the functional integral consists
of determining the theory space to be considered, i.e., the space T of all “well-
behaved” actions A[ · ] on which the RG flow is to take place. In particular, the
type of field arguments on which the actions depend and the symmetries get
fixed in this manner.
In the case at hand, the natural first guess about the theory space is that
its points A ≡ A[gµν ] should be functionals of the expectation value gµν ≡ ⟨b gµν ⟩
that are invariant under general coordinate transformations if gµν is transformed
tensorially under such transformations.5 However, this guess turns out to be too
naive ultimately, for reasons which we shall discuss in detail in the next section.
To summarize, let us be explicit about the desired properties of the yet-to-be
found theory of metric quantum gravity, or Quantum Einstein Gravity (QEG).
These properties will be the input for our RG-based search strategy:

(i) The gravitational degrees of freedom are carried by a symmetric tensor field
gbµν which acts as an elementary field variable.
(ii) The theory is a gauge theory, covariant under gauge transformations implied
by the spacetime diffeomorphisms (general coordinate transformations).
(iii) The theory is Background Independent.

These three properties are inspired by the formal structure of classical General
Relativity. In fact, they are all that we take over from GR; in particular, the
microscopic dynamics of gbµν is still completely open. The name Quantum Ein-
stein Gravity refers only to this formal structure but not, for instance, to the
use of the Einstein–Hilbert action. It plays no conceptually distinguished role in
the Asymptotic Safety program. This is in marked contrast to other quantum
gravity programs such as canonical quantum gravity [39].

4.2 What is “Coarse Graining” in Gravity?

In the previous chapter we emphasized that for non-gravitational theories, given
a theory space, the form of the FRGE and, as a result, the vector field β,
are basically completely fixed, so that the RG flow and in particular its fixed
point structure can be studied without any further input. In the case of gravity,
however, it is much harder to make this program work in a concrete way.
In this section we are going to discuss the new problems which arise specifically
in quantum gravity when one tries to develop a renormalization group framework

5 This tentative definition is insensitive to the allowed degeneration properties of gµν , which
could be imposed in addition.
90 A Functional Renormalization Group for Gravity

by following the route sketched in Chapter 2 that led to the scalar Effective
Average Action.

4.2.1 Background Independence

The theory of quantum gravity we are aiming at should be formulated in a
Background Independent way. It should explain rather than presuppose the exis-
tence and the properties of the spacetime we live in. Hence, no special spacetime
manifold, in particular no special causal, let alone Riemannian structure, should
play a distinguished role at the fundamental level of the theory.
Moreover, we might require the theory to also cover phases in which the metric
can degenerate or even have a vanishing expectation value.
Since almost our entire repertoire of quantum field theory methods relies on the
availability of an externally given spacetime manifold, usually Minkowski space,
this is a severe conceptual problem. In fact, it is the central challenge for basically
all approaches to quantum gravity, in one guise or another [16, 36–38, 127].
To give an extreme example of the dramatic consequences it may entail in
connection with an EAA, let us consider again the unbroken phase of a (hypo-
thetical) metric theory based on a theory space consisting of action functionals
A[gµν ] whose domain includes fields gµν which vanish identically: gµν = 0. When
we try to write down examples of such actions we soon realize how extremely
√
difficult this is as we have neither a volume element g nor an inverse matrix
g µν at our disposal.
To get an impression of the problem it suffices to leave the gravitational dynam-
ics aside for a moment and simply try to construct matter Lagrangians, which
remain meaningful when gµν = 0. This requirement rules out basically all actions
we are
∫ familiar with from theories like the Standard Model: Scalar
∫ potential terms
√ √
like dd x g V (ϕ) vanish identically, the scalar kinetic term dd x g g µν ∂µ ϕ∂ν ϕ
is undefined when gµν → ∫ 0dsince g µν does not exist then, and for the same reason
√ µν αβ a a
the Yang–Mills action
∫ d d x g g g Fµα Fνβ no longer makes sense. Likewise,
the Dirac action d x (det e) ψ̄γ a eµa Dµ ψ is ruled out as the inverse tetrad eµa , the
“square root” of g µν , is undefined now.
Even if we (tentatively) insist that gµν must be non-degenerate everywhere,
other problems still remain, most importantly those related to the “averaging”
of field configurations, an issue we discuss in detail below.
At this point a remark is in order concerning the mathematical structures that
constitute a “background”, and to what extent the approach to quantum gravity
we are going to develop will eventually be independent of those structures.
At the most basic level, “spacetime” for us is a set of points, called the “events,”
with no further structure. Then, at the next higher, slightly more refined level of
description we assume that this set is a topological space, and that it carries a
differentiable structure. These assumptions will not be proven or justified; their
motivation resides solely in their importance for classical physics.
 4.2 What is “Coarse Graining” in Gravity? 91

The theory we try to construct will only be able to deal with, and make predic-
tions for, those levels in the description of spacetime that go beyond its (assumed)
property of being a differentiable manifold. In particular, it can “decide” about
the Riemannian structure this manifold also may or may not carry in addition.
The theory renders the Riemannian structure dynamical. In contrast, the topo-
logical and the differentiable structures are given a priori and thus cannot be
“explained.”
Therefore, strictly speaking a “background”, or more precisely a background
spacetime, consists of a topological, differentiable and a Riemannian structure
(metric). The form of “Background Independence” we are actually able to achieve
refers to independence with respect to a (preferred) metric only. All of the fol-
lowing discussions of asymptotically safe gravity are based on an externally given
manifold with unchangeable topological and differential structure.

4.2.2 Gauge Invariance

The theory we are searching for is supposed to share a number of properties with
GR: The elementary field carrying the gravitational degrees of freedom is a ten-
sor field gµν , it is a gauge theory covariant under spacetime diffeomorphisms, and
it is Background Independent. In this section we discuss specifically the implica-
tions of gauge invariance.

(1) General coordinate transformations. In Section 4.1 we already raised

the question about how the gauge theory aspects affect the correct choice of
the theory space of QEG, T ≡ TQEG . The gauge transformations in question are
(infinitesimal) coordinate transformations generated by arbitrary vector fields v µ
which act on gµν by the Lie derivative Lv :

δgµν = Lv gµν ≡ v ρ ∂ρ gµν + ∂µ v ρ gρν + ∂ν v ρ gρµ . (4.3)

Again, referring back to General Relativity, we observe that the underlying

Einstein–Hilbert action SEH [gµν ] is gauge invariant in the sense that δSEH = 0
under gµν → gµν + δgµν , and we are thus motivated to define TQEG as the theory
space of all invariant functionals:
{ }
A[gµν ]|δA = 0 under δgµν = Lv gµν ≡ Tinv . (4.4)

Natural and simple as it seems, this choice would cause considerable problems
once we try to set up a gravitational RG framework along the lines of the scalar
EAA.

(2) Zero modes and FRGE. To see the origin of the difficulties, recall that
in the scalar theory the k-dependence of the EAA is governed by the Wetterich
92 A Functional Renormalization Group for Gravity
[( )−1 ]
(2)
equation ∂t Γk = 12 Tr Γk + Rk ∂t Rk . If we try to apply an analogous equa-
tion to a running action Γk ∈ Tinv on our tentative theory space (4.4) we find
immediately that there is a clash between the required diffeomorphism invari-
ance of A[gµν ] and the general structure of this FRGE. This clash occurs even if
we restrict gµν to be non-degenerate.
The reason is as follows. Naively replacing Γk [ϕ] by some Γk [gµν ], the Hessian
(2)
operator Γk under the trace in the FRGE is concretely given by the second
δ2
derivatives of a diffeomorphism invariant functional, δgµν (x)δg αβ (y)
Γk [g]. This
(2)
implies that Γk will possess zero modes if it is evaluated “on shell.” There exists
(2)
an infinite-dimensional subspace of the space traced over on which [ Γ−1
k is zero
]
effectively
[ and
] thus contributes an unacceptable divergent piece Tr R k ∂t Rk =
Tr ∂t ln Rk to the right-hand side of the flow equation.
Those zero modes are a direct consequence of diffeomorphism invariance: Con-
sider an arbitrary invariant functional A[g]. It satisfies δA ≡ A[g + δg] − A[g] = 0,
which, after Taylor expanding to first order in v µ , implies
∫
δA[g]
dd x Lv gµν (x) = 0. (4.5)
δgµν (x)

Taking another functional derivative of (4.5) yields

∫ ∫
δ 2 A[g] δA[g] δ
d
d x Lv gµν (x) = − dd x Lv gµν (x). (4.6)
δgαβ (y)δgµν (x) δgµν (x) δgαβ (y)

This equation holds true for all tensor fields gµν . Let us now specialize them
for the stationary points of A, i.e., those satisfying the “equation of motion”
δA[g]
δgµν = 0. For them,
∫
δ 2 A[g]
dd x Lv gµν (x) = 0. (4.7)
δgαβ (y)δgµν (x)

This equation expresses the fact that “on shell,” i.e., when the corresponding
equations of motion are fulfilled, the Hessian of an invariant functional A has
eigenvectors with eigenvalue zero, namely the “gauge modes” Lv gµν stemming
from arbitrary vector fields v µ . This is what we wanted to show.

(3) Gauge fixing. The above reasoning is the gravitational analog of a well-
known argument in quantum electrodynamics and Yang–Mills theory. It suggests
that some sort of gauge-fixed flow equation is called for that does not suffer from
the problems due to gauge zero modes.
We take this as motivation to reconsider the derivation of a flow equation, but
this time starting out from a gauge fixed functional integral.
In principle, it is straightforward to use the Faddeev–Popov trick in order to
“divide out” the redundant integration over gauge equivalent fields. The method
 4.2 What is “Coarse Graining” in Gravity? 93

parallels that in quantum electrodynamics or Yang–Mills theory, and like there,

the extended bare action that emerges in this way is invariant under appropriate
BRST transformations, which in turn give rise to certain Ward identities.
As we will see, the BRST Ward identities constitute the symmetry constraint
that codetermines the elements of the actual theory space TQEG .

(4) Manifest invariance of Γ. Opting for a gauge-fixed functional integral as

the starting point for the derivation of a flow equation, while solving one problem,
leads to a new one, namely the issue of manifest gauge invariance of the actions
A ∈ TQEG .
Considering generic functionals A[T1 , T2 , T3 , · · · ], which may depend on any
set of tensor fields T1 , T2 , · · · we say that A is manifestly gauge invariant if
it does not change under the transformations where all the tensor arguments
simultaneously are changed by an amount δTi = Lv Ti , i = 1, 2, · · · .6
For example, if A ≡ A[gµν ] as before, then the functionals in Tinv of (4.4) are
manifestly gauge invariant in this sense.
Let us leave the functional RG aside for a moment, and let us also sidestep the
difficulties related to Background Independence by postulating some ad hoc back-
ground metric (for flat Euclidean space, say) relative to which we quantize the
fluctuations of gbµν . The problem that∫ arises then is that if we straightforwardly
−S
apply the Faddeev–Popov trick to Db gµν e and follow the lines of typical per-
turbative calculations in particle physics [128], we end up with an 1PI-generating
functional, an effective action Γ[gµν ], which is not manifestly gauge invariant.
The reason is exactly the same as in the standard and perhaps better-
known functional integral quantization of Yang–Mills theories. If one gauge fixes
the functional integral with an ordinary (covariant) gauge-fixing condition like
∂ µ Aaµ = 0,7 couples the non-abelian gauge field Aaµ to a source, and constructs
the ordinary effective action, the resulting functional Γ[Aaµ ] is not invariant
under the gauge transformations of Aaµ , Aaµ 7→ Aaµ + Dµab (A) ω b . The straightfor-
wardly defined Γ[Aaµ ] communicates the fixed gauge from the bare to the effective
level [128].
Only at the level of the observable quantities deduced from Γ[Aaµ ] (S-matrix
elements, for instance) gauge invariance is recovered. So from the physical point of
view there is nothing wrong about the non-invariance of Γ[Aaµ ]. All gauge-variant
features of Γ[Aaµ ] are forgotten along the way to the observables. In perturbative
particle physics calculations one is therefore, more often than not, willing to live
with this non-invariance as only a minor deficiency of Γ[Aaµ ].

6 The explicit form of the Lie derivative Lv depends on the type of the tensor, of course. Fields
Ti with odd Grassmann parity are included here as well.
7 In this section indices a, b, · · · refer to the adjoint representation of the Yang–Mills gauge
group. Gauge group generators in some given representation are denoted T a . Our conven-
tions follow [88]. In particular the generators T a , satisfying [T a , T b ] = if abc T c , are taken
hermitian.
94 A Functional Renormalization Group for Gravity

In a non-perturbative functional RG setting the requirements are somewhat

different though. Assume we manage in some way to introduce a scale-dependent
version of this effective action, Γk [Aaµ ], then it will also not display manifest
gauge invariance either. When it comes to computing RG flows this is a severe
disadvantage, however, since one can build many more gauge non-invariant action
monomials from Aaµ than invariant ones. Therefore, the structure of the RG flow
is much simpler in an invariant than in a non-invariant framework. In the former
case, truncations that achieve a given level of accuracy will require much fewer
terms than in the latter.
These observations apply to gravity and Yang–Mills theory alike. The BRST
quantization of gbµν relative to a fixed background metric leads to a Γ[gµν ], which
is not manifestly gauge invariant. Introducing an ad hoc IR cutoff in this situation
can make things even worse.
In the later chapters of this book we will see repeatedly that the limiting factor
of practical RG calculations in quantum gravity is always the theory’s enormous
algebraic complexity. For this reason one should take advantage of all available
devices that can simplify the structure of the RG flow and its truncations.
Motivated by these remarks, we decide to “repair” the non-invariance of the
naive Γ[gµν ]. In the following we describe a scale-dependent action Γk which
enjoys manifest gauge invariance.
We emphasize already at this point that the manifest gauge invariance is not a
substitute for the BRST Ward identities mentioned in (3). The actions belonging
to TQEG will be constrained by manifest gauge invariance and the Ward identities
independently.

4.2.3 Toward a Notion of Coarse Graining

A particularly hard problem in gravity consists in establishing a suitable general-
ized concept of “coarse graining” which respects Background Independence and
diffeomorphism invariance. Deeply rooted conceptually, the problem pervades
both the continuum and discrete approaches.8
A functional RG equation which implements the, yet to be found, new general-
ized concept of coarse graining should, on the one hand, be as simple as possible,
but on the other hand should also involve a type of scale-dependent action which
is as close to observable physics as possible; “simple,” i.e., computationally feasi-
ble approximations, should suffice in order to extract the physical essence of the
RG flow.
Before embarking on gravity we consider Yang–Mills theory on a flat Euclidean
spacetime in order to disentangle the problems arising from gauge (diffeomor-
phism) invariance, from those due to Background Independence.

8 See [129–131] for a corresponding discussion within Loop Quantum Gravity.

 4.2 What is “Coarse Graining” in Gravity? 95

In Yang–Mills theory a “coarse graining” based on a naive Fourier decomposi-

tion of Aaµ (x) is not gauge covariant and hence cannot be physically meaningful.
In fact, if one were to gauge transform a slowly varying Aaµ (x) using a parameter
function ω a (x) with a fast x-variation, a gauge field with a fast x-variation would
arise which, however, still describes the same physics.
In a non-gauge theory coarse graining is usually performed by expanding the
field in terms of eigenfunctions of the (positive) operator −∂ 2 and declaring its
eigenmodes to be of “long” or “short” wavelength depending on whether the
corresponding eigenvalue p2 is smaller or larger than a given k 2 .
In a Yang–Mills theory there is an obvious clash between this way of distin-
guishing “coarse” and “fine” and the basic requirement of gauge covariance, since
the ordinary Laplacian ∂ 2 is not covariant under gauge transformations.
It seems therefore natural that in a gauge theory ∂ 2 should be replaced by the
covariant Laplacian D2 = δ µν Dµ Dν . It involves the covariant derivative Dµ ≡
∂µ − iAµ with the Lie algebra-valued field Aµ ≡ Aaµ T a in a, for the time being,
arbitrary representation of the gauge group generators T a .
While from a differential geometric point of view replacing ∂ 2 by D2 is only
too natural, it ruins to some extent the possibility of intuitively grasping the
difference between a “fine” and a “coarse” field configuration. Our intuition for
such matters stems from the classical example of Fourier analysis and synthesis
that is naturally connected to the operator ∂ 2 with its plane-wave eigenfunctions.
But this “free field”-like situation is atypical as the relevant operator, ∂ 2 , happens
to be field independent. Its non-abelian pendant D2 ≡ (∂µ − iAµ )2 has an explicit
dependence on the dynamical field Aµ instead.
Moreover, when we go on along this route and consider the eigenvalue problem
of the field-dependent operator,
−D2 (A) ψn = En ψn (4.8)
we are confronted with eigenfunctions ψn ≡ ψn (A) and eigenvalues En ≡ En (A)
which have inherited the parametric dependence on the gauge field. This field
dependence causes a number of severe problems, as we discuss next.

(1) Mode functions. Gauge covariance comes at the price of eigenfunctions that
are no longer simple plane waves, but rather reflect the details of the underlying
gauge-field configuration, including its gauge dependence. Typically the ψn s will
not possess any (translation, rotation, etc.) symmetries, unlike the plane waves
that are momentum eigenstates. Moreover, the relation between the eigenvalue
En and the position space dependence displayed by the associated ψn is usually
far from obvious. In the case of −∂ 2 the eigenvalue E ≡ p2 of a given eigenfunction
can easily be read off “with the bare eye” from the single length scale √ featured
by the corresponding ψ(x) ∝ exp(ip · x), namely its wavelength 2π/ p2 . For a
generic ψn (A) this is usually impossible.
96 A Functional Renormalization Group for Gravity

(2) Ordered sets of modes. Let us assume now that a meaningful notion of
nonabelian coarse graining can be based on the identification

“fine vs. coarse” ←→ “large vs. small eigenvalues of −D2 (A)”. (4.9)

For Aµ fixed, we order the (real, positive) eigenvalues9 of −D2 (A) in an ascending
sequence and determine the concomitant ordering of the eigenfunctions:

0 ≤ E1 (A) ≤ E2 (A) ≤ · · · ≤ En (A) ≤ · · · ⇒ ψ1 (A), ψ2 (A), · · · , ψn (A), · · · . (4.10)

As n increases, the associated sequence of eigenfunctions proceeds from ψs

considered “coarse” to increasingly “fine” ones.
This sequence of eigenfunctions can be thought of as an ordered set10 of
“templates,” which we can match against the field configuration to be coarse
grained.
This concept of coarse vs. fine is a gauge-invariant notion. In fact, if ψn is
an eigenfunction with eigenvalue En for the gauge field Aµ , then ψn′ = U ψn
is an eigenfunction with the same eigenvalue for the gauge-transformed field
A′µ = U Aµ U −1 − i(∂µ U )U −1 with U ≡ exp(−iω a (x)T a ). As a consequence, the
order of the ψs in (4.10) depends only on the gauge equivalence class Aµ
belongs to.

(3) Field-dependent bases. Under certain technical conditions the eigenfunc-

tions obtained with a given gauge field, {ψn (A)}, form a complete set. Hence,
each Aµ gives rise to its own basis.
We may use any of the bases {ψn (A)} in order to expand arbitrary matter
fields Ψ(x) that are coupled to Aµ , provided they transform as Ψ 7→ Ψ′ = U Ψ,
i.e., homogeneously and in the T a -representation. Then,
∑
Ψ(x) = cn (A) ψn (x; A). (4.11)
n

By assumption, Ψ is independent of Aµ . Therefore, its expansion coeffi-

cients cn (A) must depend on the gauge field in order to compensate for the
Aµ -dependence of the basis functions.
Given the expansion (4.11), a natural notion of coarse-graining matter fields
emerges: It simply consists in truncating the summation in (4.11) at some large
n = nmax , thus omitting extremely “fine” basis functions from the expansion.
Conversely, we can introduce the IR cutoff necessary for the formulation of
an FRGE by omitting from the sum (4.11) the “coarse” terms below a certain
n = nmin .

9 For the sake of this argument we assume a discrete spectrum.

10 In the case of degenerate eigenvalues it is only partially ordered. In general, degeneracies are
unlikely to occur however since a generic field configuration Aaµ (x) will break all symmetries.
 4.2 What is “Coarse Graining” in Gravity? 97

Note that we cannot apply this strategy to the Yang–Mills field itself. Being a
connection, it transforms inhomogeneously under gauge transformations, and so
unlike the matter fields Ψ, the gauge field admits no covariant expansion of the
form (4.11).
The actually difficult problem in Yang–Mills theory is finding a suitable notion
of coarse graining or averaging for the connection. Before returning to it let us
take a closer look at the matter fields.

(4) Invariant cutoff for matter fields. Based on this definition of “fine vs.
course” it is now straightforward to introduce an EAA for quantum matter fields
Ψ interacting with an external, i.e., a classical gauge field that is not quantized.
The relevant partition function
∫
b
b e−S M [Ψ,A]
Z M [A] = DΨ (4.12)

is invariant under gauge transformations of Aµ if S M [Ψ,b A] is invariant under

b
combined Ψ- and A-transformations (in absence of anomalies).
Following Chapter 2, it is not difficult to equip the integral over Ψ b with
a regulator term which suppresses the IR modes, while preserving its gauge
invariance: ∫
b
b e−S M [Ψ,A] M b
Zk [A] = DΨ
M
e−∆Sk [Ψ,A] . (4.13)

Every cutoff action of the type

∫ ( ( ) )
b A] = 1 k 2 b R(0) − D (A) b ,
2
∆SkM [Ψ, dd x Ψ, k 2 Ψ (4.14)
2

where the covariant derivatives are in the representation of Ψ,b and ( · , · ) denotes
an invariant inner product, will retain gauge invariance. Furthermore, R(0) ( · ) is a
shape function obeying (2.65). As a result, the factor e−∆Sk suppresses precisely
b
the Ψ-modes with eigenvalues En . k 2 .
So we see that for the matter fields it is indeed possible to implement a covari-
ant notion of coarse graining at the (formal) functional integral level. Moreover,
it is straightforward to start out from ZkM , define the corresponding EAA, and
derive its FRGE. The steps parallel those for the scalar theory, and lead to a
(parametrically A-dependent) flow equation with the same structure as in the
scalar case.
For later comparison we emphasize that the RG flow obtained by varying the
parameter k in (4.13), or the equivalent EAA, admits a clearcut interpretation:

(i) The absolute scale of the cutoff parameter k enjoys an invariant meaning:
it is a particular point in the gauge-invariant spectrum of a gauge-covariant
operator.
98 A Functional Renormalization Group for Gravity

(ii) Concerning the significance of the relative order of any two scales k1 and
k2 < k1 , we can be sure that when we lower the cutoff and proceed from the
UV to the IR the mode with eigenvalue E = k12 is integrated out first, the
other with k22 second. Decreasing the parameter k from infinity down to zero,
the individual modes of Ψ are integrated out11 exactly according to the fine
vs. coarse classification in (4.10), from right to left.
While at first sight they seem relatively trivial and self-evident, these two prop-
erties are quite “precious” actually. They do not readily generalize to Yang–Mills
and gravitational fields, as we shall see in a moment.
Again for later comparison, let us also note that the second property above
is seen most explicitly by rewriting the functional integral over the functions Ψ
as an integration over all their expansion coefficients {cn } appearing in (4.11).
Formally this amounts to a unitary transformation, with a unit Jacobian, for any
basis {ψn (A)}, i.e., any gauge field Aµ . Hence, somewhat symbolically,
∫ ∏ ∑ (0) En
dcn e−S [{cn },A] e− 2 k n R ( k2 )cn .
M 1 2 2
ZkM [A] = (4.15)
n

The second exponential in this equation makes it manifest that the mode ψn gets
unsuppressed exactly when k 2 passes the eigenvalue En .

(5) The core of the problem. The proposed notion of coarse graining applies
only to fields transforming homogeneously under gauge transformations, Ψ′ =
U Ψ, thus excluding the most interesting case of the gauge field itself.
A first idea that comes to mind is that, perhaps, one should coarse grain a
given Aµ -configuration indirectly via the field strength Fµν ≡ Fµν a
T a . This has
the advantage that Fµν = ∂µ Aν − ∂ν Aµ + f Aµ Aν transforms homogeneously
a a a abc b c

with U in the adjoint representation of the gauge group. Being given some Aµ
field configuration, we would then compute its field strength Fµν ≡ Fµν (A), coarse
coarse
grain it in some way to obtain a certain Fµν , and then search for a new
coarse coarse
connection Aµ which reproduces it, Fµν = Fµν (Acoarse
µ ).
Unfortunately, at least for individual Aµ configurations, this simple idea is not
viable. At this point we mention only one of several reasons: In non-abelian gauge
theories, contrary to abelian ones, not all gauge potentials Aµ that produce a
given Fµν are gauge equivalent, i.e., can be related by a gauge transformation.
coarse
Because of this so-called Wu–Yang ambiguity [132] the computed Fµν fails to
coarse
determine Aµ uniquely modulo gauge transformations.
A second approach one might try is to give up the ambition of coarse graining
individual configurations as we did for Ψ, b but rather jump ahead and modify
ba
the functional integral over Aµ right away by some sort of nonstandard mode
suppression term ∆S YM [A]. b The hope would be that a clever choice of ∆S YM [A] b
k k

11 Recall that getting “integrated out” has the concrete meaning of “becoming unsuppressed”
when R(0) crosses over from R(0) ≈ 1 to R(0) ≈ 0.
 4.2 What is “Coarse Graining” in Gravity? 99

might give rise to an interesting and useful RG flow, even though the separation
of modes (suppressed vs. unsuppressed) might not have much to do with their
“fineness” but perhaps some other feature.
Since the cutoff action should be gauge invariant, the main building blocks
are the field-strength tensor and covariant derivatives acting upon it. A simple
example is
∫
( 2 b )ab
b = 1 k 2 dd x Fµν
∆SkYM [A] a b b
(A)R(0) − D k(2A) F bµν (A) (4.16)
4
with R(0) a shape function with the standard characteristics. The cutoff action
(4.16) is perfectly gauge invariant. Adding it to the bare action S[A] we are led
to the functional integral (before gauge fixing)
∫
b −∆S YM [A]
b e−S[A] b
ZkYM = DA e k . (4.17)

At first sight this scale-dependent integral seems to be similar to (4.13) for matter
b in place of Ψ.
fields with Fµν (A) b However, as soon as we try to base a full-fledged
functional renormalization on it, severe difficulties emerge.
The first problem concerns the interpretation of (4.17). What is it really that
gets suppressed by ∆SkYM , and what is the significance of some value of k being
smaller than another one?
For a fixed configuration A bµ contributing to the integral we might solve the
tensorial eigenvalue equation for −D2 (A) b in the adjoint representation of the
gauge group, the one appropriate for Fµν , and then imitate (4.11) and expand
it; symbolically, ∑
a b b ψn (A)
b aµν .
Fµν (A) = cn (A) (4.18)
n

This time the same field A bµ enters both the operator D2 (A)b whose eigenbasis
b
{ψn (A)} is used for the expansion and the function to be expanded. This com-
pletely obscures the interpretation of such an eigenfunction expansion, if any,
bµ via its Fµν (A)
and suggests again that the indirect coarse graining of an A b is a
dead end.
More importantly, when we try to functionally integrate over the gauge field,
we clearly cannot trade the integration variable Abµ (x) for the collection of expan-
sion coefficients {cn }, as we did for Ψ b in the derivation of (4.15). However, as
we discussed in the previous section, only after bringing the partition function
to the form (4.15) can we get a clear understanding of the order in which field
configurations are integrated out along RG trajectories, cf. property (ii) in (4)
above. Concerning ZkYM it is meaningless to say that “at scale k 2 the eigenmodes
with eigenvalue E = k 2 are integrated out”, since at each point of the domain of
integration the pertinent operator D2 (A)b is a different one.
Even if we decide to ignore these interpretational issues and go ahead, trying to
derive an FRGE on the basis of ZkYM , we soon realize that the present approach
leads to a disaster from the practical point of view.
100 A Functional Renormalization Group for Gravity

The reason is that ∆SkYM [A], b contrary to its scalar precursor, is no longer
bilinear in the integration variable A bµ . Higher order terms in A bµ arise both
b
because of the A-dependence (0)
in the argument of R and the ∂ A b+A b2 structure
of the nonabelian field strength. As a consequence ∂t ∆Sk [A], b too fails to be
bilinear, and so it is impossible to express its expectation value ⟨∂t ∆Sk [A]⟩,b
the analog of (2.16), solely in terms of the two-point function and its inverse,
e (2) ≡ Γ(2) + Rk . Recall that this possibility was instrumental in the derivation
Γ k k
of the scalar FRGE, cf. the steps leading from (2.16) to (2.20). In fact, the
bilinearity of ∆Sk was the deeper reason why the FRGE turned out to contain
only second-order functional derivatives and no higher ones.
Employing instead our candidate (4.16) or any similar gauge invariant cutoff
action, we can reexpress ⟨∂t ∆Sk [A]⟩b in terms of Γk itself at best at the price of
(2) (n)
utilizing not only Γk but also the plethora of all higher n-point functions Γk .
This would amount to an extremely complicated FRGE in which arbitrarily high
δ n Γk
derivatives δA···δA appear on the right-hand side.
Obviously there is a rather generic clash between a simple structure of the
FRGE, building on the propagators derived from Γk only, and gauge invariance,
which enforces the presence of the quantum field in places where it does not
appear in non-gauge theories.
Since the prospects for making sense of an infinite-order functional differential
equation seem to be rather dim, progress will depend on overcoming this “infinite-
order problem” in one way or another.

(6) Gravity: problems and options. So far we have considered separately

the typical problems that arise from the requirements of Background Indepen-
dence and gauge invariance. If we now return to our main goal, finding a gen-
eralization of the Wilsonian renormalization group that continues to make sense
in the realm of quantum gravity, we must deal with both types of difficulties in
combination.
The overall problem presents itself somewhat differently in typical discrete and
continuum approaches.

(6a) Discrete approaches to quantum gravity based on any sort of “atoms of

spacetime” exhibit a certain similarity to the many body problems of condensed
matter physics. Concepts of the exact RG naturally lend themselves to a descrip-
tion of the (real or purely mathematical) process in which large clusters of atoms
build up iteratively by the aggregation of further such atoms. There is a clearcut
notion of an RG step and hence of the direction of the RG flow in this case.
Determining the degree of “infraredness” of a given cluster, i.e., its position on
the RG trajectory, is thus a simple matter of counting.
The same is true if one considers an infinite “solid” or “crystal” made of such
atoms and follows the familiar idea of block spin transformations. One would
combine a few of the atoms into blocks, determine the properties of the entire
 4.2 What is “Coarse Graining” in Gravity? 101

blocks as implied by the given properties of the atoms, and then declare those
blocks with their “renormalized” properties to be the new “atoms” for the next
RG step. Again, if the rules for the “blocking” step are given, a clear notion
of an exact RG flow emerges. The direction of the RG trajectories defines a
relative notion of coarse vs. fine blocks, and also here, finding their location on
the trajectory only requires elementary counting, of the number of blocking steps
in this case.
As these two examples illustrate, within appropriate discrete formulations of
quantum gravity it may be rather obvious, at least in principle, how to formulate
a natural notion of a coarse-graining step and the resulting renormalization group
without invoking ad hoc background structures.
It goes without saying that it is by no means straightforward to implement this
kind of RG transformations at a technically explicit level in spin-foam models or
similar theories, but first results have been obtained [133, 134].

(6b) Let us now return to the continuum-based brand of the exact renormal-
ization group, the functional RG, ∫ and more concretely, let us try to modify
formal functional integrals like Db gµν e−S[bg] by some, yet-to-be found Back-
ground Independent and generally covariant “mode” suppression factor with
similar properties as e−∆Sk in the scalar case.
Unquestionably, it seems natural to replace the standard Laplacian ∂ 2 ≡
δ ∂µ ∂ν with a covariant one, something like, say, D2 ≡ gµν Dµ Dν . This imme-
µν

diately raises the questions: what is the tensor gµν and how should one define
the connection entering the covariant derivative Dµ ? Taking Background Inde-
pendence literally, the only building block available for this construction is the
tensor gbµν itself.
The simplest option is to identify gµν with the inverse of the matrix gbµν and
Dµ ≡ Dµ (b g ) with the associated Levi–Civita connection. This would make D2
the Laplace–Beltrami operator related to the dynamical field, the integration
variable gbµν :
g ) ≡ gbµν Dµ (b
D2 (b g )Dν (b
g ). (4.19)

This is not a viable road though if our ambition is to functionally integrate over
all symmetric tensor fields gbµν , and not only metrics that are non-degenerate
by definition. Above we argued that it is desirable to be open-minded about
including arbitarily degenerate, i.e., non-invertiable gbµν s into the integration,
but for those (4.19) is undefined.
Thus, let us be modest at this point and assume that it is actually meaningful
to integrate over genuine metrics only, so that (4.19) does make sense. Then we
can proceed as in Yang–Mills theory, and consider the eigenvalue problem of
D2 (bg ), acting on tensor fields of any type:

−D2 (b
g ) ψn = En ψn . (4.20)
102 A Functional Renormalization Group for Gravity

Leaving technical issues aside, the eigenfunctions {ψn (bg )} form a metric-
dependent basis, and we may expand arbitrary matter fields as
∑
Ψ(x) = g ) ψn (x; gb).
cn (b (4.21)
n

With this expansion we have all tools that are needed to introduce an EAA for
matter fields interacting with an external gravitational field, via
∫
b g]
b e−S M [Ψ,b
g ] = DΨ
Z M [b . (4.22)

The steps are analogous to those in Yang–Mills theory leading from (4.12) to the
representation (4.15) of the modified partition function. The latter representa-
tion was instrumental in establishing the absolute and relative meaning of the
cutoff scale k, and the discussion is perfectly analogous for quantum matter fields
interacting with external gravitational rather than Yang–Mills fields.
While this leads to the desired diffeomorphism-invariant notion of coarse
graining matter fields, the other key problem, Background Independence, is not
addressed yet since from the point of view of Ψ, the field gbµν is a given, fixed
classical metric.

(6c) Thus we are back to the core of the problem, the coarse graining of gbµν itself.
At first sight the situation seems more favorable than in Yang–Mills theory, since
in metric gravity the elementary field variable transforms under the pertinent
gauge transformations as a tensor rather than a connection. Hence, as far as gauge
invariance is concerned, nothing would forbid us to write down straightforward
mode-suppression terms like
∫ √ 2
∆Sk ∝ k 2 dd x gb gbµν R(0) (− Dk(b
2
g)
) gbµν (4.23)

since D2 can act on gbµν directly. This is an illusion, though, since the covariant
derivative employed is metric-compatible with gbµν , and thus R(0) gets evaluated
2
at zero argument for any metric. Using (2.65), R(0) (− Dk(b 2
g)
) gbµν = gbµν . Hence,
(4.23) is actually nothing more than a contribution to the cosmological constant.

(6d) Another option is to suppress gbµν -configurations in the functional integral

according to their curvature; for instance, using terms involving the Riemann
tensor, like
∫ √ 2
∆Sk ∝ k dd x gb R(b
2
g )µνρσ R(0) (− Dk(b
2
g)
) R(b
g )µνρσ . (4.24)

Indeed, it would be quite interesting, and could teach us a lot about quantum
gravity
∫ to investigate the k-dependence of the modified partition function Zk =
Dbgµν exp (−S[bg ] − ∆Sk [b
g ]) with cutoff actions like (4.24).
 4.2 What is “Coarse Graining” in Gravity? 103

However, in order to explore this sort of RG flow the typical strategies of the
scalar EAA approach are of little help: ∆Sk depends on the dynamical fields, the
components of gbµν , in a complicated non-polynomial manner, and so an FRGE
derived from Zk that is modeled along the scalar example would not obey a
sufficiently simple, practically useful FRGE with only finitely many functional
derivatives. It would suffer from the same infinite-order problem we encountered
already in Yang–Mills theory.

4.2.4 The Background Field Method

In the preceding sections we outlined the difficulties and obstacles specific to
quantum gravity. They arise from the basic requirements of Background Inde-
pendence and general covariance, and also because no distinguished, natural
notion of coarse graining suggests itself in this setting.
We saw that there is no unique ideal way to define the renormalization group
for quantum gravity that would apply to all, or at least several of the different
approaches one would like to compare on the basis of their RG properties.
At a concrete technical level the options for defining what “coarse graining”
is to be are quite different in discrete- and continuum-based approaches. But
even when we focus only on continuum approaches and search for a physically
meaningful scale-dependent functional “Γk ” roughly similar to the scalar EAA,
there is still a variety of routes one may follow.
Experience has shown that the most useful definition of “Γk ” hardly can be
found by abstract reasoning but rather as a subtle compromise between desirable
formal properties of the functional on one side and its practical computability on
the other. As such, it can only emerge in a process of trial and error. In the rest
of this chapter we describe the result of this process, the gravitational Effective
Average Action proposed in [14].

(1) Three problems, one solution. The construction of the gravitational

EAA begins with a critical decision with far-reaching consequences, namely the
decision to approach the Background Independence problem using the background
field method.
Surprisingly enough, what at first sounds like a contradiction in terms, namely
to use background fields in order to implement Background Independence, does
not only solve the first of the three main problems discussed above, but the
other two, coarse graining and gauge invariance, as well: The background field
method supplies an almost canonical, natural notion of coarse graining, and,
by a standard technique (background-type gauge fixing), it allows us to define
the associated scale-dependent action as a manifestly diffeomorphism invariant
functional.
In the simpler case of Yang–Mills theory when Background Independence is
not an issue the background field method likewise leads to both a gauge-covariant
104 A Functional Renormalization Group for Gravity

notion of coarse graining, and to an EAA with manifest gauge invariance. We

will not go into the details of the Yang–Mills case here but refer the reader to
the original articles on the subject: The EAA for abelian and non-abelian gauge
fields (coupled to matter) was introduced in a series of papers [23–26]. In the
first two of them, [25, 26], the proper “average action” for gauge theories was
constructed, while the related “Effective Average Action” for Yang–Mills theory
was introduced in [23, 24]. Early applications include [27–29], and for reviews see
[33–35].12

(2) All backgrounds = no background! Pioneered by Schwinger and later

on in particular by DeWitt [50, 51], the basic idea behind the background field
approach is to decompose, in one way or another, the dynamical field gbµν (x) as
the “sum” of a given classical background metric ḡµν (x) “plus” a dynamical quan-
tum field bhµν (x) that is then promoted to an operator, in canonical quantization,
or is integrated over in the corresponding functional integral. This amounts to
quantizing the fluctuation degrees of freedom, b hµν , similar to a matter field, on
a given classical spacetime manifold which carries a bona fide metric ḡµν .
As a result, we free ourselves of all the conceptual difficulties specifically
related to the Background Independent quantization of gravity and may resort
to well-developed methods and calculational tools of standard, i.e., background-
dependent, quantum field theory.
The problems that originate in the quantization of spacetime itself have not
disappeared, however; they resurface in the (logically) second step of the proce-
dure: To achieve Background Independence we must now repeat the quantization
of b
hµν for all background metrics ḡµν .13 Only the totality of all quantum theories
together, describing the dynamics of b hµν on arbitrary, but classical background
spacetimes, contains sufficient information to implicitly account for a single
“quantum spacetime.”
So, loosely speaking, the background-field method allows us to reinterpret the
quantum theory of spacetime itself as a set of infinitely many “ordinary” quantum
field theories of a matter-like field on given classical spacetimes. These matter-like
theories are distinguished by a label which identifies the background spacetime
manifold in question. Besides its topological and differentiable structure, which
we consider fixed here, the label specifies the metric it is equipped with, ḡµν .
At least in principle, we can now use the established methods of background-
dependent quantum field theory to “solve” the theory of the b hµν field on one

12 Note that the (older, proper) “average action” [67] and the “Effective Average Action”
[22] are two differently defined scale-dependent functionals. While the latter is the type
of running action considered in this book exclusively, the former is closer in spirit to an
iterated Kadanoff–Wilson block spin transformation; in the case of scalar fields it may be
thought of as a scale-dependent generalization of the “constraint effective potential” [69].
13 A mathematically more precise meaning of “all background metrics” will emerge later in
the discussion.
 4.2 What is “Coarse Graining” in Gravity? 105

fixed background, and then repeat this conceptually clear step for all possible
backgrounds.
The outcome of the corresponding calculations are ḡ-dependent expecta-
tion values ⟨O(b h)⟩ḡ , for example, and in particular n-point functions like
⟨b
h(x1 ) · · · b
h(xn )⟩ḡ . The latter can be encoded in a generating functional Γ[h; ḡ],
which, besides h ≡ ⟨b h⟩, depends on a second argument, ḡ. As for the domain of
this background-effective action, we tentatively assume that the space of back-
grounds for which it can be defined, F̄ ≡ {ḡµν }, is roughly as large as the space
from which the dynamical metric is drawn.
Because of the comparable importance enjoyed by ḡµν and ⟨b gµν ⟩ ≡ gµν this
effective action is sometimes said to be of the bi-metric type.

(3) The natural notion of coarse graining. Having interpreted the quantum
theory of spacetime itself as the totality of all b
hµν -theories on a fixed background,
there is an obvious natural notion of coarse graining now. Within the theory with
fixed “label” ḡµν , it is based, in the sense explained above, on the ordered set of
eigenmodes of D2 (ḡ) ≡ D̄2 ≡ ḡ µν D̄µ D̄ν where D̄µ involves the Christoffel symbols
of ḡµν :
−D2 (ḡ) h(n) (n)
µν = En hµν . (4.25)

Now hµν appears as a matter field similar to Ψ above, which at this point is
completely unrelated to ḡµν . Hence, we are no longer in the confusing and almost
circular situation described above where the very same field that should be coarse
grained occurs also in the covariant Laplacian defining the modes.
In each of the fixed-ḡ theories the background metric implies a clearcut concept
of “long-wavelength” and “short-wavelength” modes of b hµν , or corresponding
low-momentum and high-momentum modes.
Furthermore, as we will see, the discrimination between the two types of modes
under the functional integral can be achieved by means of a cutoff action ∆Sk
which is bilinear in b hµν , and thus also solves the “infinite order problem” we
were facing before.

(4) Parametrizing gbµν . The background field approach expresses the func-
tional integral over gbµν as an integral over a new field, b hµν , which is then
considered the elementary quantum mechanical variable. Technically, this step
depends on whether or not the field configurations gbµν contributing to the orig-
inal integral are allowed to degenerate.
∫
(4a) If the integral Db gµν · · · extends over all symmetric tensor fields, including
arbitrarily degenerate matrices and even gbµν = 0, the domain of the gb-integration
is easily parametrized in terms of fluctuations b hµν using a linear split:

gbµν = ḡµν + b
hµν . (4.26)
106 A Functional Renormalization Group for Gravity
∫ ∫
The functional integral Db gµν · · · ≡ Db hµν · · · involves an unconstrained inte-
gration over all symmetric tensors b hµν then, with a translation-invariant measure
like in standard matter field theories.
∫
(4b) If Db gµν · · · extends only over the subset of all metrics with a given sig-
nature, the requirement of non-degeneracy implies certain nonlinear constraints
among the components of gbµν , in particular det(b gµν ) ̸= 0. In this case, the domain
of integration, F ≡ {nondeg., sym. tensor fields gbµν }, is no longer a linear space
but a curved infinite-dimensional “manifold”. The task then consists in finding
a suitable parametrization gb = gb(b h; ḡ) of this manifold in terms of unconstrained
fields b hµν . A typical example for such a parametrization is provided by the
exponential split (4.33) constructed below.
Regarding the differential geometry of the infinite dimensional manifold F, the
functions gb are merely coordinates on some local patch of F, i.e., no tensorial
quantities. Instead, we would prefer b h to behave as a vector, since only then
a functional integral over b h will be of the familiar type we know from matter
fields [135].
At least in the finite-dimensional case, there is a well-known general construc-
tion, the exponential map, which supplies precisely the sought-for parametriza-
tion [136]. It interprets b h as a tangent vector to the field space F at some base
b
point ḡ, that is, h ∈ Tḡ F, and then parametrizes arbitrary points gb ∈ F in a neigh-
borhood of ḡ by setting up a one-to-one correspondence between those points on
F, and vectors b h in a fixed tangent space, namely Tḡ F. Of course, the base point
ḡ is identified with the background metric in our case.
To write down this correspondence concretely let us switch to a con-
densed index notation and identify ϕi ≡ gbµν (x), ϕ̄i ≡ ḡµν (x), hi ≡ b hµν (x). Indices
i, j, k, · · · will combine the spacetime indices µ, ν, · · · with the continuous x
coordinates, and the “summation” over repeated Latin indices is understood
to include an integration over x.
Let us assume now that the manifold in question,14 in our case F, is equipped
with an arbitrary linear connection, Γkij , so that we may talk about geodesics on
F, i.e., parametrized curves τ 7→ ϕ(τ ) satisfying

ϕ̈i (τ ) + Γijk (ϕ(τ )) ϕ̇j (τ )ϕ̇k (τ ) = 0. (4.27)

Here dots denote derivatives with respect to the parameter τ .

After having fixed a base point ϕ̄ ∈ F, we now associate arbitrary points ϕ ∈ F
in a neighborhood of ϕ̄ to vectors h ∈ Tϕ̄ F in a neighborhood of the origin 0 ∈ Tϕ̄ F

14 In GR the exponential map makes its appearance in setting up systems of Riemann normal
coordinates. In this case the manifold is spacetime itself instead of F, and Γkij is the Levi–
Civita connection of its metric [137]. For a similar application on symplectic manifolds
(phase spaces) see [138].
 4.2 What is “Coarse Graining” in Gravity? 107

by demanding that ϕ and ϕ̄ are connected by the segment of a geodesic, {ϕ(τ ),

τ ∈ [0, 1]}, i.e.,
ϕi (τ = 0) = ϕ̄i , ϕi (τ = 1) = ϕi (4.28)
with the “initial velocity” given by the vector
ϕ̇i (τ = 0) = hi . (4.29)
The relationship h ↔ ϕ provided by the exponential map is indeed a diffeomor-
phism of some neighborhood of the origin in Tϕ̄ F with a neighborhood of ϕ̄ on
F [136]. It can be worked out explicitly by Taylor expanding the geodesic at the
initial point,
1 ( n i )
∑∞
ϕi (τ ) = ∂τ ϕ (τ ) τn (4.30)
n=0
n! τ =0

and recursively eliminating all higher derivatives by means of the geodesic equa-
tion. This expresses the right-hand side of (4.30) in terms of ϕ(0) and h = ϕ̇(0)
only. In this manner we obtain at τ = 1 the desired formula for the endpoint ϕ,
namely
ϕi = ϕ̄i + H i (h; ϕ̄) (4.31)
with the series representation
1 i j k
H i (h; ϕ̄) = hi − Γ̄ h h
2! jk
1( i ) (4.32)
− lk Γ̄mj − Γ̄lj Γ̄km h h h + O(h ).
Γ̄jk,l − Γ̄m i m i j k l 4
3!
Here the overbar indicates that the connection coefficients and their derivatives
are evaluated at ϕ̄.

(4c) The next step consists of using the nonlinear background quantum field
b
split given by (4.31) in order to replace the ϕ-integral with an integration over
b
h, taking the corresponding Jacobian factor properly into account.
Note that while the connection Γijk has not been specified yet, there are various
natural options.
In background-dependent field theory, there exist well-developed techniques for
the quantization of nonlinear σ-models, pioneered by Honerkamp [139], Friedan
[140], and others [141, 142], based on the normal coordinate expansion (4.32).
In this case the situation simplifies since the ϕi s are coordinates on the model’s
(finite-dimensional) target space only, rather than the huge space of all fields, F.
Nevertheless, many of the methods developed for the perturbative quantization
of nonlinear σ-models carry over to the infinite-dimensional context of the space
of all metrics, F.15

15 The relation to the σ-models becomes particularly clear if one imposes Killing vector con-
straints on the metrics in F; see [121–123] for an in-depth analysis of the σ-model describing
the quantum theory of metrics with two Killing symmetries.
108 A Functional Renormalization Group for Gravity

In σ-models with a Riemannian target space the natural choice of Γkij is clearly
the Levi–Civita connection of its target space metric.
On the field space underlying General Relativity, F, a number of possibilities
are conceivable, for instance the Levi–Civita connection pertaining to the family
of ultralocal DeWitt metrics on F [126, 143], or the Vilkovisky–DeWitt connec-
tion [144], which is adapted to the action of the gauge group on the points of F.
Furthermore, there even exists a choice of the connection for which the general
exponential map boils down to, literally, the exponential of the matrix hµν =
ḡ µα hαν [145–147]: ( )ρ
−1
g = ḡ eḡ h ⇐⇒ gµν = ḡµρ eh ν . (4.33)
This exponential parametrization establishes a one-to-one correspondence
between symmetric tensors hµν and metrics gµν for Euclidean signature, but
it fails to do so in the Lorentzian case [146].

(5) The split symmetry. The base point ϕ̄ is chosen freely on the manifold,
and so it is natural to ask how the vector h representing a given ϕ ∈ F responds
to a change of ϕ̄. From (4.31) it follows that the pairs (ϕ̄, h) and (ϕ̄ + δ ϕ̄, h + δh),
where h ∈ Tϕ̄ F and (h + δh) ∈ Tϕ̄+δϕ̄ F, respectively, describe the same point ϕ
on the manifold, if
δ ϕ̄i = −εi ,
(4.34)
δhi = −Fki (h; ϕ̄) εk ,

where εi is an infinitesimal vector and Fki is determined by the requirement

δϕi = 0,16 i.e.,
∂H i l ∂H
i
δki + + Fk = 0. (4.35)
∂ ϕ̄k ∂hl
The equations (4.34) constitute the general, i.e., nonlinear form of what is known
as a background quantum field split symmetry transformation, or simply “split
transformation” for short [140].
Given the series representation of H i in (4.32) we can work out the solution
to (4.35) as a similar Taylor series in h:

Fki (h; ϕ̄) = −δki − Γikl (ϕ̄)hl + O(h2 ). (4.36)

Furthermore, it is not difficult to infer from (4.35) that the functions Fki satisfy
the relation
∂Fki ∂Fji ∂Fki l ∂Fji l
− + F − F = 0, (4.37)
∂ ϕ̄j ∂ ϕ̄k ∂hl j ∂hl k
which can be thought of as a kind of zero-curvature condition. In fact, the
relation (4.37) implies that the split symmetry transformations (4.34) form an

16 We use the notation of finite-dimensional manifolds here. Otherwise, all derivatives should
be replaced by functional derivatives.
 4.2 What is “Coarse Graining” in Gravity? 109

abelian group. Thus, their order does not matter when performing consecutive
transformations.
Friedan [140] showed that, in more geometrical terms, Fki (h; ϕ̄) has the interpre-
tation of a nonlinear flat connection on the tangent bundle. It defines a parallel
transport of vectors h from Tϕ̄ F to another tangent space, at, say, ϕ̄′ such that
the original h together with ϕ̄, and the transported vector together with ϕ̄′ ,
describe one and the same point ϕ on the manifold.
Assume we are given a set of functions Ti1 ···in (h; ϕ̄) with an arbitrary depen-
dence on h and ϕ̄, and we ask whether they are the components of a unique,
globally well-defined tensor field on F, expressed via the exponential map based
at ϕ̄. The condition for this to be the case is that those functions must be
covariantly constant with respect to the nonlinear connection Fki (h; ϕ̄). Explicitly
[140, 142, 141]:

∂
n ∑
l ∂ ∂
k
Ti 1 ···i n + F k l
T i1 ···i n + Ti1 ···ir−1 jir+1 ···in ir Fkj = 0. (4.38)
∂ ϕ̄ ∂h r=1
∂h

Analogous globality conditions can be set up for more general tensors and
spinors.17 If T carries no indices, (4.38) simply states that T (h; ϕ̄) = S(ϕ̄ +
H(h; ϕ̄)) for some globally defined scalar S.
It is clear that if the manifold F in question happens to support a trivial
connection with Γijk ≡ 0, the expansions (4.32) and (4.36) lead us back to the
linear split, ϕi = ϕ̄i + H i , where H i = hi , with its obvious split transformation
δ ϕ̄i = −εi , δhi = εi . We mentioned already that this split is appropriate to coor-
dinatize the unconstrained space of all tensor fields, but not the manifold of
metrics:
gbµν = ḡµν + b
hµν ,
(4.39)
δḡµν = −εµν , δb hµν = εµν .

In the sequel we will mostly discuss this case where unconstrained symmetric
tensors rather than genuine metrics are integrated over. The motivation is first
of all simplicity and the fact that it has been employed in the vast majority of
Asymptotic Safety studies.
It is clear that scrutinizing more advanced theory spaces by taking into account
higher orders of the normal coordinate expansion will be a central future research
direction. First results along these lines, employing the exponential parametriza-
tion, were reported in [145]. For a systematic examination of different ways to
parametrize the graviton field and the related functional measures we refer to
[148, 149].
Nevertheless, we emphasize that the linear split is important not only as a
first approximation to a nonlinear expansion but also in its own right. It is well

17 For the corresponding discussion of metaplectic spinor fields see [138].

110 A Functional Renormalization Group for Gravity

conceivable that the integral over unconstrained bhµν s possesses an Asymptoti-

cally Safe continuum limit different from the one over metrics, i.e., that it probes
another “universality class” of fundamental theories. (See [117] for a discussion
of this point.)

4.3 Introducing an EAA for Gravity

In this section we describe in detail the Effective Average Action for gravity,
focusing mostly on those aspects that are different from the scalar case outlined
in Chapter 2. The presentation follows [14] to which we refer for further details.

4.3.1 The Gauge-Fixed Functional Integral

We start out from a functional integral over tensors gbµν of the form
∫
Db
gµν exp (−S[b gµν ] + source terms). (4.40)

The fields gbµν live on a given spacetime manifold with a fixed topological class and
smooth differential structure. The bare action S[b gµν ] is assumed to be invariant
under general coordinate transformations which act on the metric according to

gµν = Lv gbµν ≡ v ρ ∂ρ gbµν + ∂µ v ρ gbρν + ∂ν v ρ gbρµ .

δb (4.41)

Here Lv denotes the Lie derivative with respect to the generating vector field v µ .
Furthermore, to render the ensuing mathematical manipulations well defined
we assume that the functional integral has been UV-regularized in some way. We
shall not need to specify how this is done explicitly at this point. It is sufficient
to keep the corresponding cutoff implicit.
Invoking the linear background split for explicitness, we decompose the
variable of integration according to gbµν = ḡµν + b hµν , where ḡµν is a fixed back-
ground
∫ metric. We interpret the measure Dbgµν as Db hµν , which leaves us with
Dbhµν exp(−S[ḡ + bh] + · · · ) then. 18

The next step consists in “cutting out” gauge-equivalent field configurations

from this functional integral. The gauge transformations which we have to gauge-
fix under the integral over b hµν are the so-called “true,” or “quantum” gauge
transformations δ Q . They follow from (4.41) under the assumption that ḡµν
does not change:
( )
δQb
hµν = Lv gbµν = Lv ḡµν + b
hµν , δ Q ḡµν = 0. (4.42)

18 bµν
The construction outlined in this section is easily adapted to other decompositions of g
into background and fluctuation fields. In particular, there are no additional conceptual
difficulties when constructing the corresponding EAAs.
 4.3 Introducing an EAA for Gravity 111

Picking any gauge-fixing function Fµ (b h; ḡ), the familiar Faddeev–Popov trick

can be applied straightforwardly to this type of gauge invariance [91, 143, 150–
152]. It leads us to the modified integral

∫ ( )
Db
hµν DC µ DC̄µ exp −S[ḡ + b
h] − Sgf [b
h; ḡ] − Sgh [b
h, C, C̄; ḡ] . (4.43)

Let us explain its various ingredients.

g ] ≡ S[ḡ + b
First of all, the classical action S[b h] was augmented by a gauge-fixing
term, Sgf , which has the general structure
∫
1 √
Sgf [b
h; ḡ] = dd x ḡ ḡ µν Fµ (b
h; ḡ)Fν (b
h; ḡ). (4.44)
2α

The gauge-fixing function Fµ (b

h; ḡ) is still quite arbitrary here. In particular, we
allow it to depend on the background metric ḡ.
Furthermore, Sgh is the action for the associated Faddeev–Popov ghost and
antighost fields, which we denote C µ and C̄µ , respectively:
∫
√ ∂Fν ( )
Sgh [b
h, C, C̄; ḡ] = −κ−1 dd x ḡ C̄µ ḡ µν LC ḡαβ + b
hαβ . (4.45)
∂b
hαβ
∫
This action Sgh ∝ dd x C̄MC displays the ghost kinetic operator M and is
obtained along the same lines as in the familiar case of Yang–Mills theory: one
first applies a gauge transformation δ Q to Fµ , then contracts the result with C̄µ ,
and finally replaces the parameters v µ by the anticommuting field C µ [153].
Functional integration over the ghost C µ and the likewise anticommuting
antighost C̄µ exponentiates the Faddeev–Popov determinant, then
[ ] ∫
[ ] δFµ ∫
det M = det ν
= DC µ DC̄µ e− C̄MC , (4.46)
δv

whereby the integration rules for Grassmann variables are applied. The determi-
nant of the infinite-dimensional “matrix” M may be seen as the volume element
on the “gauge slice” of fields b
h satisfying the gauge condition Fµ (b
h; ḡ) = 0.
For the time being, the gauge-fixing parameter α and the ghost normalization
κ are arbitrary constants. Since κ has the dimension of a mass we will often set
in the following

κ ≡ (32π Ḡ)− 2 ,
1
(4.47)

where Ḡ is a constant reference value of Newton’s constant.

112 A Functional Renormalization Group for Gravity

4.3.2 Classical BRST Invariance

The gauge-fixed action under the integral (4.43) is invariant under the following
BRST transformation:
( )
δ BRST b
hµν = εκ−2 LC gbµν = εκ−2 LC ḡµν + b
hµν ,
ε
δ BRST
ε ḡµν = 0,
(4.48)
δ BRST
ε C µ = εκ−2 C ν ∂ν C µ ,
δ BRST
ε C̄µ = ε(ακ)−1 Fµ .
Here ε is an anticommuting, x-independent parameter. It is not difficult to check
that the total action is indeed invariant:
( )
δ BRST
ε S[ḡ + b
h] + Sgf [b
h; ḡ] + Sgh [b
h, C, C̄; ḡ] = 0. (4.49)

Furthermore, one can verify that δ BRST is nilpotent, even “off-shell,” when acting
on gbµν , ḡµν , b
2
hµν , and C µ , while (δ BRST ) C̄µ vanishes only when the classical
equation of motion for the antighost field is used.

4.3.3 Background-Type Gauge-Fixing Conditions

As we will see more explicitly in a moment, using a generic gauge-fixing condition
Fµ (b
h; ḡ) = 0, the (ordinary) effective action resulting from the above gauge-
fixed functional integral fails to be a diffeomorphism-invariant functional of its
arguments [87, 91, 143, 150–152].
But the situation is different for the special class of the so-called gauge-
fixing conditions of the background type. While – as any gauge-fixing condition
does – they break the invariance under the true gauge transformations (4.42),
their defining property is to be invariant under the so-called background gauge
transformations defined by

δBb
hµν = Lv b
hµν , δ B ḡµν = Lv ḡµν . (4.50)

The key idea is that the full metric gbµν = ḡµν + bhµν transforms in the required
fashion (4.41) both under the background gauge transformations (4.50) and
the true, quantum gauge transformations (4.42): δ B gbµν = δ Q gbµν = Lv gbµν . The
transformations δ Q and δ B distribute the given variation δb gµν = Lv gbµν of gbµν =
ḡµν + b
hµν differently over ḡµν and b hµν . The quantum gauge transformations δ Q
keep ḡµν unchanged and ascribe the entire variation of gbµν to b hµν , while the
background transformations δ B of (4.50) change both ḡµν and b hµν , but only by
their own Lie derivative.
These rules imply δ B Fµ = Lv Fµ , and as a consequence, both the classical action
and the gauge-fixing action are invariant under δ B :

δ B S[ḡ + b
h] = 0, δ B Sgf [b
h; ḡ] = 0. (4.51)
 4.3 Introducing an EAA for Gravity 113

Of course, one of the consequences entailed by the differences between δ Q and δ B

is that Sgf , despite its δ B -invariance, is not invariant under δ Q -transformations:
after all, its very raison d’être is to break the quantum gauge invariance!
It can be verified that for every background-type choice of Fµ (b h; ḡ), the ghost
b
action Sgh [h, C, C̄; ḡ] is also invariant under the extended background gauge
transformations given by (4.50) together with the rules

δ B C µ = Lv C µ , δ B C̄µ = Lv C̄µ , (4.52)

for the ghost and antighost fields, respectively. So, taking everything together we
may conclude that the total action including the gauge-fixing and ghost terms is
invariant under background gauge transformations:
( )
δ B S[ḡ + b
h] + Sgf [b
h; ḡ] + Sgh [b
h, C, C̄; ḡ] = 0. (4.53)

This crucial property, among others, will allow us to set up a fully δ B -covariant
Effective Average Action.

4.3.4 Generalized Harmonic Gauges

Clearly there exist many possible gauge-fixing actions Sgf [b h; ḡ] of the form (4.44)
which break (4.42) and at the same time are invariant under (4.50). It is con-
venient to choose gauge functions of the form Fµ ∝ Fµ αβ [ḡ] b hαβ . They are linear
in the metric fluctuation, and employ some δ -covariant operator Fµ αβ acting
B

on it.
An example of this sort which has often been employed in practical calculations
is the one-parameter family of generalized harmonic gauge conditions:
√
Fµ (b
h; ḡ) ≡ 2 κ Fµ αβ [ḡ] b
hαβ with Fµ αβ = δµβ ḡ αγ D̄γ − ϖḡ αβ D̄µ . (4.54)

Here ϖ is a free parameter, and Fµ αβ is taken to be a first-order differential

operator built from the covariant derivative D̄µ , which involves the Christoffel
symbols Γ̄ρµν of the background metric.
For the value ϖ = 21 , (4.54) indeed reduces to the background version of the
harmonic coordinate condition [91, 143, 150–152]: on a flat background with
ḡµν = δµν , for example, the condition Fµ = 0 assumes the familiar form ∂ µ hµν =
1 µ
2 ∂ν hµ .
The ghost action for the gauge condition (4.54) reads
√ ∫ d √
b
Sgh [h, C, C̄; ḡ] = − 2 d x ḡ C̄µ M[b
g , ḡ]µ ν C ν . (4.55)

The Faddeev–Popov operator involves the metric-compatible covariant deriva-

tives constructed from both gbµν and ḡµν :
( )
M[bg , ḡ]µ ν = ḡ µρ ḡ σλ D̄λ gbρν Dσ + gbσν Dρ − 2ϖ ḡ ρσ ḡ µλ D̄λ gbσν Dρ . (4.56)
114 A Functional Renormalization Group for Gravity

As always, the covariant derivatives related to gbµν and ḡµν are denoted Dµ and
D̄µ , respectively.

4.3.5 The Mode Suppression Term

Next, we equip the integral (4.43) with an infrared regulator, concretely with
a higher-derivative mode suppression that damps the contribution of the low-
momentum modes to the functional integral, while leaving the high-momentum
modes untouched. Thanks to the presence of the background metric, a natural
notion of “low-momentum” vs. “high-momentum” is available now.
In analogy with the flat-space and non-gauge case, where the momentum
square equals the eigenvalue of the negative Laplacian −∂ 2 , we now define the
“momentum square” to be the negative eigenvalue of the covariant tensor Lapla-
cian D̄2 ≡ ḡ µν D̄µ D̄ν built from the background metric. Low- and high-momentum
modes, by definition, are those eigenfunctions of −D̄2 whose eigenvalues are
smaller or larger, respectively, than a given scale k 2 .
To implement the IR regulator we multiply the integrand of (4.43) by the extra
factor
b
e−∆Sk [h,C,C̄;ḡ] . (4.57)
The cutoff action ∆Sk suppresses the low-momentum modes of the metric
fluctuations b hµν and the Faddeev–Popov ghosts.
To avoid the fatal “infinite-order problem” we take another critical decision at
this point, namely we insist that ∆Sk must be bilinear in the dynamical fields,
i.e., in the integration variables b
h, C, C̄:
2 ∫ √ ∫ d √
κ d √ b grav µνρσ b
∆Sk = d x ḡ hµν Rk [ḡ] hρσ + 2 d x ḡ C̄µ Rgh µ
k [ḡ] C . (4.58)
2
As we saw already, this decision sacrifices geometric naturalness to some extent
in favor of practical computability.
The ḡµν -dependent cutoff operators Rgravk and Rgh
k must be designed in such
a way that they achieve the desired discrimination between high-momentum
and low-momentum modes: Eigenmodes of −D̄2 with eigenvalues p2 ≫ k 2 are
left untouched, whereas modes with small eigenvalues p2 ≪ k 2 get suppressed.
Both operators, Rgrav k and Rghk , have the same general structure Rk [ḡ] =
Zk k R (−D̄ /k ), where the dimensionless shape function R(0) interpolates
2 (0) 2 2

smoothly between R(0) (0) = 1 and R(0) (∞) = 0; see (2.65).

The factors Zk are different for the graviton and the ghost cutoff. They are
determined by the condition that all fluctuation spectra must be cut off at pre-
(2)
cisely the same value k 2 . This is the case, for example, if Rk combines with Γk
(2)
to an inverse propagator like Γk + Rk = Zk (p2 + k 2 R(0) (p2 /k 2 )) + · · · . For low-
momentum modes it gives rise to a (mass)2 of order k 2 , rather than the wrong
value (Zk )−1 k 2 . Following our earlier discussion of adjusted cutoffs in Section
2.5 this is precisely as it should be.
 4.3 Introducing an EAA for Gravity 115

Applying this construction principle in the ghost sector, Zk ≡ Zkgh is a pure

number, whereas for the metric fluctuation Zk ≡ Zkgrav is a certain tensor con-
structed from the background metric. We will be more explicit about the
adjustment of the Zk s when we discuss concrete examples later on.
A first general property of ∆Sk which is of importance for the practical appli-
cability of the resulting RG equations is that the modes of b hµν and the ghosts are
organized according to their eigenvalues with respect to the background Laplace
operator D̄2 = ḡ µν D̄µ D̄ν rather than D2 = g µν Dµ Dν , which pertains to the full
quantum metric. Using D̄2 , the functional ∆Sk is indeed bilinear in the quan-
tum fields and thus will give rise to a flow equation, which contains only second
functional derivatives of Γk but no higher ones.
A second general property of ∆Sk which significantly simplifies the resulting
EAA formalism is its invariance under the background gauge transformations
(4.50) with (4.58):
δ B ∆Sk [b
h, C, C̄; ḡ] = 0. (4.59)

This property further motivates the particular form (4.58) we have given to ∆Sk .
On the other hand, for scales k ̸= 0 the cutoff action is not BRST invariant:

δ BRST
ε ∆Sk [b
h, C, C̄; ḡ] ̸= 0. (4.60)

BRST invariance can only be recovered in the limit k → 0.

4.3.6 Sources and Expectation Values

Now we promote the integral (4.43) with the extra factor (4.57) included to a
generating functional and define:
( )
exp Wk [tµν , σ µ , σ̄µ ; β µν , τµ ; ḡµν ]
∫ (
= Db hµν DC µ DC̄µ exp −S[ḡ + b h] − Sgf [bh; ḡ] (4.61)
)
− Sgh [b
h, C, C̄; ḡ] − ∆Sk [b h, C, C̄; ḡ] − Ssource .

Here the last part of the action,

∫
√ {
Ssource = − dd x ḡ tµν b hµν + σ̄µ C µ + σ µ C̄µ
( ) } (4.62)
+ β µν LC ḡµν + b
hµν + τµ C ν ∂ν C µ

contains sources for the three dynamical fields b hµν , C µ , and C̄µ , which are
µν µ
denoted by t , σ̄µ , and σ , respectively. The sources for the ghosts are anti-
commuting objects themselves. The two additional source functions β µν and τµ
couple to the BRST variations of b
hµν and C µ , respectively. We do not need them
116 A Functional Renormalization Group for Gravity

at this point, but later on they will play a role in the formulation of the Ward
identities that represent the BRST invariance at the quantum level.
This leads us to a somewhat lengthy list of arguments for the generating
functional:
Wk [tµν , σ µ , σ̄µ ; β µν , τµ ; ḡµν ] ≡ Wk [J; J BRST ; ḡµν ]. (4.63)

Functional derivatives of Wk with respect to J ≡ (tµν , σ µ , σ̄µ ) and J BRST ≡

(β µν , τµ ) generate the expectation values of the dynamical fields and BRST vari-
ations, respectively. Besides on these sources, Wk also functionally depends on
the background metric.
The expectation values of φ b ≡ (b
hµν , C µ , C̄µ ) are denoted by φ ≡ (hµν , ξ µ , ξ¯µ )
and we have explicitly

⟨ ⟩ 1 δWk µ 1 δWk ¯ ⟨ ⟩ 1 δWk

hµν = b
hµν = √ µν
, ξ = ⟨C µ ⟩ = √ , ξµ = C̄µ = √ , (4.64)
ḡ δt ḡ δσ̄µ ḡ δσ µ

or, in a more compact notation,

⟨ i ⟩ 1 δ
b (x) = √
φi (x) = φ Wk [J; J BRST ; ḡ]. (4.65)
ḡ(x) i (x)
δJ
√
The explicit factors of ḡ in (4.64) and (4.65) guarantee that both the sources and
the fields transform under general coordinate transformations as proper tensors
rather than tensor densities.
Let us now switch from the sources to the field expectation values as the
independent variables. Correspondingly we introduce the Legendre transform Γ ek
of Wk with respect to the dynamical sources J. All other variables Wk depends
on, J BRST , ḡµν , and k, are just spectators in this transformation and are kept
fixed.
In the general case Γ e k is given by the Legendre–Fenchel supremum formula
(2.11). In the regular case when the relations given by (4.65), φ ≡ φ[J; J BRST ; ḡ],
can be solved for J, it assumes the form
∫
e k [h, ξ, ξ;
¯ β, τ ; ḡ] = √ { µν }
Γ dd x ḡ t hµν + σ̄µ ξ µ + σ µ ξ¯µ − Wk [t, σ, σ̄; β, τ ; ḡ]
(4.66)
with the inverted φ-J-relationship J ≡ (t, σ, σ̄) = Jk [φ; J ; ḡ] substituted on the
BRST

right-hand side.
As usual, the Legendre transform gives rise to source–field relations,

ek
δΓ √ ek
δΓ √ ek
δΓ √
= ḡ tµν , = − ḡ σ µ , = − ḡ σ̄µ , (4.67)
δhµν δ ξ¯µ δξ µ

which are ”dual” to the original ones in (4.64).

 4.3 Introducing an EAA for Gravity 117

4.3.7 Formal Definition of the Gravitational EAA

Having explained all the ingredients which go into the functional integral (4.61)
for the generating functional Wk we finally introduce the actual Effective Average
Action for metric gravity. It is obtained from the Legendre transform Γ e k of Wk
by subtracting the cutoff action ∆Sk with the classical fields inserted:

Γk [h, ξ, ξ; e k [h, ξ, ξ;
¯ β, τ ; ḡ] = Γ ¯ β, τ ; ḡ] − ∆Sk [h, ξ, ξ;
¯ ḡ]. (4.68)

The relations (4.61), (4.66), and (4.68) constitute what one might call the
formal definition of the EAA since it is based on a functional integral which only
makes mathematical sense in the presence of a UV regulator, and this regulator
is implicit in all of the above equations.
At this point it is an open question, whether, or how, the UV regulator can be
removed consistently. Before we can address this issue we first derive a number
of general properties of the gravitational EAA, or more precisely, of the EAA in
presence of the UV regulator, unless otherwise stated.
A remark concerning the notations we use for the EAA might be in place here.
It is natural to introduce the expectation value of the full metric,

gµν (x) = ⟨b
gµν (x)⟩ ≡ ḡµν (x) + hµν (x), (4.69)

as the classical analogue of the quantum metric gbµν ≡ ḡµν + b

hµν , and to consider
Γk as a functional of gµν rather than hµν :

¯ β, τ ] ≡ Γk [h, ξ, ξ;
Γk [g, ḡ, ξ, ξ; ¯ β, τ ; ḡ] . (4.70)
h=g−ḡ

Apart from the ghosts and BRST sources, the action Γk [h, ξ, ξ; ¯ β, τ ; ḡ], by defi-
nition, depends on the fluctuations’ expectation value hµν and the background
metric as the independent field variables. This should be contrasted with the
second variant of the EAA, Γk [g, ḡ, ξ, ξ; ¯ β, τ ], which, by definition, depends on
two fully fledged metrics, namely the expectation value of the quantum metric,
⟨b
gµν (x)⟩ ≡ gµν (x) and the background metric ḡµν (x).
The “semicolon variant” of the EAA, Γk [h, ξ, ξ; ¯ β, τ ; ḡ], emphasizes the point
of view that the fluctuations of the metric, hµν , may be regarded as matter-like
excitations on a classical spacetime with metric ḡµν .
Instead, the “comma variant” Γk [g, ḡ, ξ, ξ; ¯ β, τ ] ≡ Γk [ḡ + h, ḡ, ξ, ξ;
¯ β, τ ], makes
it explicit that the EAA suffers from an extra ḡ-dependence over and above the
one that combines with h to build up a full metric g = ḡ + h. In fact, it is precisely
¯ β, τ ] that amounts to the extra ḡ-dependence.
the second argument of Γk [g, ḡ, ξ, ξ;
This seemingly trivial difference of the viewpoint will become important when
we discuss the restoration of background independence later on. It is no exag-
geration to say that, in the EAA framework, the most profound and deeply
118 A Functional Renormalization Group for Gravity

rooted difference between the standard matter quantum field theory on a clas-
sical spacetime and quantum gravity is the inevitability of an extra Background
Dependence.

4.4 Properties of the Gravitational EAA

Let us now collect the main features of the EAA introduced in the previous sec-
tion. We begin by streamlining our notation a bit. Note that ḡµν is the only
background field in the case at hand. In principle, one could also introduce
background fields for the ghosts, but this is unnecessary here.
For brevity we shall often use the following collective notation for the various
fields, their expectation values and sources:
( )
φbi ≡ bhµν , C µ , C̄µ integration variable,
( )
Φ̄ ≡ ḡµν , 0, 0
i
background field,
( µ ¯
)
φ ≡ hµν , ξ , ξµ
i
expectation value (fluctuation),
( µ ¯
) (4.71)
Φ ≡ gµν , ξ , ξµ
i
expectation value (full field),
( )
Ji ≡ tµν , σ̄µ , σ µ b
sources for φ,
( µν )
J BRST
≡ β , τµ sources for (δ BRST hµν , δ BRST C µ ) .
[ ] [
Thus the EAA reads Γk φ; J BRST ; Φ̄ or Γk Φ, Φ̄; J BRST ], respectively.

(1) Background gauge invariance. The functionals Γ e k and Γk are invariant

under infinitesimal general coordinate transformations where all its arguments
transform as tensors of the corresponding rank:
[ ] [ ]
Γk φ + Lv φ; J BRST + Lv J BRST ; Φ̄ + Lv Φ̄ = Γk φ; J BRST ; Φ̄ . (4.72)

Note that (4.72) expresses the invariance of Γk under background gauge trans-
formations δ B lifted to the level of the expectation values. Hence, contrary to the
“quantum gauge transformation” (4.42), also the background metric transforms
nontrivially here: δḡµν = Lv ḡµν .
Equation (4.72) is a consequence of the invariance enjoyed by the generating
functional Wk [J; J BRST ; Φ̄]:

Wk [J + Lv J; J BRST + Lv J BRST ; Φ̄ + Lv Φ̄] = Wk [J; J BRST ; Φ̄]. (4.73)

This property follows from the functional integral (4.61) if one performs a com-
pensating transformation (4.50), (4.52) on the integration variables b hµν , C µ , C̄µ ,
and uses the δ -invariance of S[ḡ + b
B
h], Sgf , Sgh , and ∆Sk , respectively. At this
point we also assume that the functional measure in (4.61) is diffeomorphism
invariant.
The background gauge invariance of Γk , expressed in (4.72), is highly welcome
from the practical point of view. It implies that if the initial functional contains
 4.4 Properties of the Gravitational EAA 119

only δ B -invariant terms, its RG evolution will not generate non-invariant terms.
Very often this reduces the number of terms to be retained in a reliable trunca-
tion ansatz quite considerably.

(2) The infrared limit. Since the Rk s vanish for k = 0, the limit k → 0 of
Γk [gµν , ḡµν , ξ µ , ξ¯µ ] leads us to the standard effective action. It still depends on two
metrics, though. The “ordinary” effective action Γ[gµν ], having only one metric
argument is obtained from this functional by setting ḡµν = gµν , or equivalently
hµν = 0 [87, 91, 143, 150–152]:

¯ β, τ ]
Γ[gµν ] ≡ lim Γk [g, ḡ, ξ, ξ;
k→0 g=ḡ, ξ=0=ξ̄, β=0=τ
(4.74)
¯ β, τ ; ḡ]
≡ lim Γk [h, ξ, ξ; .
k→0 h=0, ξ=0=ξ̄, β=0=τ

This action brings about the “magic property” of the background field formal-
ism: a priori the 1PI n-point functions of the metric are obtained by an n-fold
functional differentiation of Γ0 [h, 0, 0; 0, 0; ḡ] with respect to hµν . Hereby ḡµν is
kept fixed; the background metric acts simply as an externally prescribed func-
tion which specifies the form of the gauge-fixing condition. Hence, the functional
Γ0 and the resulting off-shell Green functions do depend on ḡµν , while the on-
shell Green functions, related to observable scattering amplitudes, do not depend
on ḡµν . In this respect, ḡµν plays a role that is conceptually similar to the gauge
parameter α in the standard approach.
Remarkably enough, the very same on-shell Green functions can be obtained by
differentiating the functional Γ[gµν ] of (4.74) with respect to gµν , or equivalently
Γ0 [h = 0, 0, 0; 0, 0; ḡ = g], with respect to its ḡ argument. (In this context, “on-
shell” means that the metric satisfies the effective field equation δΓ0 [g]/δgµν = 0.)

(3) The functional RG equation. The derivation of an exact evolution equa-

tion for Γk proceeds as follows. Taking a derivative of the functional integral
(4.61) with respect to the renormalization group time t ≡ ln k one obtains, in
matrix notation:
[ ) ] [ ) ]
1 ⟨ ⟩( ⟨ ⟩(
− ∂t W k = Tr b h⊗b bk
h ∂t R bk
− Tr C̄ ⊗ C ∂t R . (4.75)
2 hh ξ̄ξ

b k is a matrix in field space whose non-zero entries are taken to be:

Here R

( )µνρσ
bk
R = κ2 (Rgrav [ḡ])
µνρσ
,
k
( )hh
√ (4.76)
Rbk = 2 Rgh [ḡ].
k
ξ̄ξ
120 A Functional Renormalization Group for Gravity

The right-hand side of (4.75) can be re-expressed directly in terms of Γk by

noting that the connected two-point function
⟨ i ⟩
Gij (x, y) ≡ φ
b (x)φbj (y) − φi (x)φj (y)
1 δ 2 Wk (4.77)
=√
ḡ(x)ḡ(y) δJi (x)δJj (y)

ek
and the Hessian of the Legendre transform Γ

( (2) ) 1 ek
δ2 Γ
e
Γ (x, y) ≡ √ (4.78)
k ij
ḡ(x)ḡ(y) δφ (x)δφj (y)
i

are inverse matrices in the sense that

∫ √ ( (2) ) δ(x − z)
e
dd y ḡ(y) Gij (x, y) Γ (y, z) = δli √ . (4.79)
k jl
ḡ(z)

Thus we arrive at the following exact functional RG equation for the gravitational
Effective Average Action:

¯ β, τ ; ḡ]
∂t Γk [h, ξ, ξ;
[( )−1 ( ) ]
1 (2) bk bk
= Tr Γk + R ∂t R
2 hh hh (4.80)
[{( ) ( )−1 } ( ) ]
1 −1
− Tr
(2)
Γk + R bk (2)
− Γk + R bk bk
∂t R .
2 ξ̄ξ ξ ξ̄ ξ̄ξ

If one wants to express the right-hand side of this equation in terms of position-
space matrix elements∫ then
√ the integration implied by the functional traces has
to be interpreted as dd x ḡ(x), and correspondingly the matrix elements of the
(2)
Hessian operator Γk are given by

( (2) ) ⟨ (2) ⟩ 1 δ 2 Γk
Γk (x, y) ≡ x, i Γk y, j = √ . (4.81)
ij
ḡ(x)ḡ(y) δφ (x)δφj (y)
i

The analogous matrix elements in the ghost sector are defined in terms of left
derivatives; for example,
(( )
(2) )
ν 1 δ 1 δ
Γk ξ̄ξ (x, y) = √ √ Γk . (4.82)
µ ḡ(x) δξ µ (x) ḡ(y) δ ξ¯ν (y)

For any cutoff operator R b k with the correct qualitative properties the traces
on the right-hand side of (4.80) are well convergent, both in the IR and the
b k , the dominant contributions to the traces, now
UV. By virtue of the factor ∂t R
interpreted as a sum over eigenvalues, come from a narrow band of generalized
momenta centered around k. Large momenta are exponentially suppressed.
 4.4 Properties of the Gravitational EAA 121

Therefore, at this point, it is easily possible to remove the (implicit) UV reg-

ulator that we introduced above in order to render the derivation of the FRGE
from the functional integral meaningful.
Henceforth we regard (4.80) as a UV cutoff-free flow equation which has lost
all memory about the cutoff. The equation describes the change of Γk when its
scale is lowered from k to k − dk and it is, in this sense, a “local” statement in
eigenvalue space.

(4) The exact integro-differential equation. From the definition of the EAA
in terms of the formal functional integral one easily derives the following, likewise
formal, exact integro-differential equation satisfied by the gravitational average
action:
( )
¯ β, τ ; ḡ]
exp −Γk [h, ξ, ξ;
∫ [
b
= DhDCDC̄ exp −S[ ebh, C, C̄; β, τ ; ḡ]
∫ { }] (4.83)
( ) δΓk ( µ )
µ δΓk
( ) δΓk
d b
+ d x hµν − hµν + C −ξ ¯
+ C̄µ − ξµ ¯
δhµν δξ µ δ ξµ
( )
× exp −∆Sk [b h − h, C − ξ, C̄ − ξ; ¯ ḡ] .

The action Se which appears under the above integral is given by:
∫
√ { ( ) }
S ≡ S + Sgf + Sgh − dd x ḡ β µν LC ḡµν + b
e hµν + τµ C ν ∂ν C µ . (4.84)

Contrary to the FRGE, this functional integro-differential equation for Γk

requires the presence of a UV regularization in order to be well defined.

(5) The modified BRST Ward identities. We mentioned already that the
classical action plus the gauge-fixing and ghost terms are invariant under the
BRST transformations (4.48). Therefore, the BRST variation of the total bare
action Stot ≡ S + Sgf + Sgh + ∆Sk + Ssources receives contributions only from the
cutoff and the source terms.
If we apply a BRST transformation to the functional integral defining Wk , and
assume that its measure is BRST invariant, we obtain

⟨δ BRST
ε Ssources + δ BRST
ε ∆Sk ⟩ = 0, (4.85)

where the expectation value is to be interpreted as

∫
⟨O⟩ ≡ e−Wk Db hDCDC̄ O e−Stot . (4.86)

Our goal is to convert (4.85) to a statement about the average action Γk .

122 A Functional Renormalization Group for Gravity

Because the BRST transformation (4.48) is off-shell nilpotent when acting on

b
hµν and on C µ (but not on C̄µ ) one has:
∫ {
d √
( )
δ BRST
ε S sources = −εκ −2
d x ḡ tµν LC ḡµν + b hµν
} (4.87)
− σ̄µ C ν ∂ν C µ − κα−1 σ µ Fµ (ḡ, b
h) .

If we take the expectation value of (4.87) and then express Wk in terms of Γk we

are led to
∫ { ′ }
ε 1 δΓk δΓ′k δΓ′k δΓ′k ε
⟨δ BRST
ε S sources ⟩ = dd
x √ + + 2 Yek , (4.88)
κ2 ḡ(x) δhµν δβ µν δξ µ δτµ κ

with
∫ { ( ) }
1 δ∆Sk δΓ′k δ∆Sk δΓ′k √ κ√
Yek ≡ dd x √ µν
+ − 2 ḡFµ (ḡ, h)Rgh
k ξ µ
.
ḡ δhµν δβ δξ µ δτµ α
(4.89)
Here we introduced the convenient abbreviation:

Γ′k ≡ Γk − Sgf [h; ḡ]. (4.90)

⟨ ⟩
We also exploited the antighost’s quantum equation of motion: δStot /δ C̄µ = 0,
which can be cast in the form
[ ]
1 δ √ µν ρσ 1 δ
√ − 2 ḡ Fν √ ¯ β, τ ; ḡ] = 0. (4.91)
Γk [h, ξ, ξ;
ḡ(x) δ ξ¯µ (x) ḡ(x) δβ ρσ (x)

Furthermore, the BRST variation of the cutoff action gives rise to:
ε ( )
⟨δ BRST
ε ∆Sk ⟩ = − 2 Yk + Yek , (4.92)
κ
with the following traces on its right-hand side:
[ ( )−1 ]
µνρσ (2) bk δ 2 Γk
Yk ≡ κ2 Tr (Rgrav
k ) Γ k + R √ √
hρσ φ ḡδφ ḡδβ µν
[ ( )−1 ]
√ δ 2 Γk
− 2 Tr Rgh Γ
(2)
+ Rbk √ √ (4.93)
k k
ξ̄µ φ ḡδφ ḡδτµ
[ ( )−1 ]
− 2α−1 κ2 Tr Rgh F µ
ρσ
Γ
(2)
+ Rb k .
k k µ hρσ ξ
{ }
Herein the field components φ ≡ h, ξ, ξ¯ are summed over.
From (4.88) and (4.92) we obtain the modified BRST Ward identities in their
final form:
∫ { ′ }
1 δΓk δΓ′k δΓ′k δΓ′k
d x√
d
+ µ = Yk . (4.94)
ḡ(x) δhµν δβ µν δξ δτµ
 4.4 Properties of the Gravitational EAA 123

Equation (4.94) has to be compared to the ordinary gravitational Ward

identities [153] which are similar to (4.94) but have a vanishing right-hand side.
The contribution Yk is due to the cutoff and therefore it vanishes for k → 0 since
Rk ∝ k 2 → 0 in this limit. Hence the standard effective action Γ = Γ0 indeed obeys
the familiar BRST identities,
∫ { }
1 δΓ′ δΓ′ δΓ′ δΓ′
dd x √ + = 0, (4.95)
ḡ(x) δhµν δβ µν δξ µ δτµ

where Γ′ ≡ Γ − Sgf . Thus standard BRST invariance is restored for k → 0.

(6) The Ward identity for split symmetry. The amount of “extra” ḡ-
dependence the EAA is suffering from is governed by the following exact
functional equation:

δ ¯ β, τ ]
Γk [g, ḡ, ξ, ξ;
δḡµν (x)
[ ] (4.96)
1 ( )−1 δ Se(2) [g, ḡ, · · · ]
(2)
= STr Γk [g, ḡ, · · · ] + Rk [ḡ] .
2 δḡµν (x)

It should be noted that this equation is written in terms of the functional

Γk [g, ḡ, · · · ] for which gµν and ḡµν are the independent variables. Therefore, the
derivative (4.96) really measures only that dependence on ḡµν , which does not
combine with hµν and thus is “extra.”
The relation (4.96) can be seen as the Ward identity associated with a linear
split symmetry transformation.19 The ḡ-derivative given by (4.96) would vanish
identically if there was no breaking of split symmetry. This requires the Hes-
sian Se(2) [g, ḡ, · · · ] to be independent of ḡ, which is clearly impossible to achieve
because of the unavoidable split symmetry breaking by the coarse-graining oper-
(2)
ator ∆Sk = Rk [ḡ] and the gauge-fixing sector.

(7) Ultraviolet behavior. In Section 4.1 we saw already in the scalar case
that the exact integral equation satisfied by Γk is a convenient starting point for
deducing its behavior at large values of k.
For a heuristic derivation one observes that when k → ∞ the last exponential
in (4.83) becomes proportional to a δ-functional, which equates the quantum
fields (b
h, C, C̄) to their classical counterparts:

e−∆Sk ∼ δ[b ¯
h − h] δ[C − ξ] δ[C̄ − ξ]. (4.97)
k→∞

19 This type of Ward identity was first introduced in the context of Yang–Mills theory [29]. See
Appendix A of [29] for a detailed derivation and discussion. It was first applied to gravity
in [90].
124 A Functional Renormalization Group for Gravity

As a consequence, the Effective Average Action for k → ∞ reads:

¯ β, τ ; ḡ] = S[ḡ + h] + Sgf [h; ḡ] + Sgh [h, ξ, ξ;

Γk [h, ξ, ξ; ¯ ḡ]
∫
√ { ( ) }
− dd x ḡ β µν Lξ ḡµν + hµν + τµ ξ ν ∂ν ξ µ + · · · .

(4.98)

At the level of the restricted functional,

¯ β, τ ]
Γ̄k [gµν ] ≡ Γk [g, ḡ, ξ, ξ; (4.99)
g=ḡ, ξ=0=ξ̄, β=0=τ

the UV behavior (4.98) boils down to:

Γ̄k [gµν ] = S[gµν ] + · · · . (4.100)

A more precise treatment employs a saddle-point approximation of the functional

integral appearing in (4.83) and also determines the subleading terms which are
indicated by the dots in the above equations. We will come back to this point in
Chapter 8.

(8) Extended theory space and bi-metric character of the EAA.

With the restricted effective action (4.99) we succeeded in constructing a
diffeomorphism-invariant generating functional for the n-point functions of the
metric which depends only on a single metric. Thanks to (4.72), the actions
Γ[gµν ] and Γ̄k [gµν ] are indeed invariant under general coordinate transformations,
δgµν = Lv gµν .
However, the simplified functional Γ̄k [gµν ] is afflicted with the drawback that
it does not satisfy any closed functional RG equation we could use to compute
it without obtaining the full-fledged Γk first.
As a consequence, the actual RG evolution must be performed at the level of
the much more complicated functional Γk [g, ḡ, ξ, ξ; ¯ β, τ ]. Only after having solved
the FRGE we may equate the two metrics, i.e., let ḡ = g in the list of arguments
of Γk , along with setting the ghosts and BRST sources to zero.
As a result, the true theory space of metric gravity,
{ }
¯ β, τ ] ,
T ≡ A[g, ḡ, ξ, ξ; (4.101)

consists of functionals which depend not only on one, but rather two metrics, and
in addition on Faddeev–Popov ghosts and BRST sources. Thus, unavoidably, the
running action functionals are of the bi-metric type. Furthermore, the actions A ∈
T are required to comply with a set of constraints, including the invariance under
background gauge transformations, (4.72), the BRST Ward identities (4.94), and
the Ward identities for the split symmetry (4.96).
These constraints are “instantaneous” conditions on the form of the EAA; all
quantities are evaluated at the same RG time t = ln(k/k0 ). Importantly enough,
 4.4 Properties of the Gravitational EAA 125

they are consistent with the RG time evolution given by the flow equation since
they are derived from the same underlying functional integral.
Coming back to the flow equation it is suggestive to abbreviate the FRGE
(4.80) as
[( )−1 ]
1 (2) bk bk ,
∂t Γk = STr Γk + R ∂t R (4.102)
2
where the supertrace “STr” takes care of the minus sign in (4.80) due to the odd
Grassmann
{ parity
} of the ghosts, and also implies a summation over the various
¯
fields, h, ξ, ξ .
This makes it obvious that we did reach one of our goals, namely to preserve
the simple formal structure displayed by the scalar flow equation even in the
realm of Background Independent quantum gravity.
The structural simplicity of the flow equation comes at a considerable price,
however, namely a highly complicated space of functionals on which it must be
defined, and solved ultimately.

(9) Running field equations and self-consistent backgrounds. The EAA

gives rise to a scale-dependent source–field relationship which includes an explicit
cutoff term linear in the fluctuation field. From (4.67) with (4.68) we obtain, in
compact notation,20

1 δΓk [φ; Φ̄]

√ + Rk [Φ̄]i j φj (x) = J i (x). (4.103)
ḡ(x) δφi (x)

These coupled field equations generalize the classical Euler–Lagrange equations

and determine the expectation values of the metric fluctuations and the ghosts
in dependence on ⟨the external
⟩ sources, the background, and the RG scale. The
solutions φi (x) ≡ φ
bi (x) = φik [J; Φ̄](x) describe how the quantum fluctuations
respond to changes of the environment they are placed in. The latter is modeled
by the background field Φ̄ and the sources J.
The magnitude of the quantum fluctuations decides to what extent the dynam-
ically determined, actual expectation value ⟨Φ⟩ b deviates from the background
which is prescribed externally:
⟨ i⟩
Φb = Φ̄i + φik [J; Φ̄]. (4.104)

We are particularly interested in the distinguished backgrounds ⟨ ⟩ in which

the quantum fluctuations, at J = 0, are “too weak” to move Φ b away from
Φ̄. By definition, such self-consistent⟨ ⟩backgrounds Φ̄ ≡ Φ̄ have the property
sc

b = Φ̄sc . In particular, in a self-consistent

that φk [J = 0; Φ̄ = Φ̄sc ] = 0, whence Φ
⟨ ⟩ ⟨ ⟩
background, b hµν = 0, and so gbµν = ḡµν sc
[154].

20 For simplicity we suppress the BRST sources here.

126 A Functional Renormalization Group for Gravity

A background is self-consistent if, with this background inserted, the field

equation (4.103) for J ≡ 0 admits the solution φ(x) ≡ 0. It satisfies the tadpole
equation, expressing a vanishing one-point function:

δ
Γk [φ; Φ̄] = 0. (4.105)
δφi (x) φ=0;Φ̄=Φ̄sc
k

It is clear that in general solutions to (4.105) will have an explicit scale

dependence, hence the notation Φ̄sc k .
Let us be slightly more specific and focus on solutions involving vanishing
ghost fields, ξ µ = 0 = ξ¯µ .21 Then self-consistent background metrics gksc satisfy

δ
Γk [h, 0, 0; ḡ] = 0. (4.106)
δhµν (x) sc
h=0;ḡ=ḡk

Note again that this is not a differential equation for hµν , but a constraint on
ḡµν . The tadpole equation (4.106) often serves as a powerful tool for extracting
physical information from the EAA.
This completes our summary of the main properties enjoyed by the gravita-
tional average action. In later chapters we will come back to them repeatedly.

21 Note that ξ µ = 0 = ξ̄µ is always a possible solution of (4.103) since the ghost-charge
neutrality of Γk forces ghost and antighost fields to occur in pairs.
 5
Truncations of Single-Metric Type

In quantum gravity, as in basically all applications of the functional renormal-

ization group, finding non-perturbative solutions to the pertinent flow equations
is an extremely difficult task. It may come as a surprise that, solving the RG
equations in gravity is usually the easier part of the problem: The true challenge
consists in deriving them in an explicit form in the first place.
In fact, both for deep conceptual reasons related to the implementation of
general covariance and Background Independence, for instance, and substantial
technical difficulties, it is particularly hard to actually compute explicit RG flows
from the rather abstract, albeit exact, functional differential equation for the
gravitational Effective Average Action. The tensor character of the fields renders
the analytical evaluation of the functional traces in the FRGE significantly more
demanding than in the scalar theory encountered in Chapter 2 or similar matter
field theories in flat space.
In practice, both on a rigid spacetime and in quantum gravity, the most impor-
tant strategy for finding approximate, while still non-perturbative, solutions to
the flow equation is the truncation of theory space. Hereby one projects the
exact flow on the infinite-dimensional theory space onto a “smaller,” technically
tractable subspace, often parametrized by just a few generalized couplings, and
then, possibly after some iterations of “trial and error,” tries to prove that the
truncated flow thus obtained approximates the exact one sufficiently well.
As a starting point, a still rather general class of truncations for the gravita-
tional EAA has been proposed in [14], and for more than a decade all examples
worked out explicitly were based on truncations belonging to this class. They are
referred to as single-metric truncations, for a reason that will become clear in a
moment.
In this chapter we first explain the general structure of the single-metric
truncations, in Section 5.1. The subsequent sections then discuss a number of rep-
resentative examples and the results achieved with them. The discussion focuses
on the question of whether and to what extent those truncated RG flows support
128 Truncations of Single-Metric Type

the Asymptotic Safety hypothesis in four-dimensional (and more generally, d

dimensional) quantum gravity.

5.1 General Classes of Truncations

5.1.1 Freezing the Ghost Sector
As a first step toward simplifying the field dependence of the gravitational average
action one can try to neglect the evolution of the ghost action. This amounts to
making an ansatz of the following form:

¯ β, τ ] = Γgrav [g, ḡ] + Sgf [g − ḡ; ḡ] + Sgh [g − ḡ, ξ, ξ;

Γk [g, ḡ, ξ, ξ; ¯ ḡ]
k
∫
√ { } (5.1)
− dd x ḡ β µν Lξ gµν + τµ ξ ν ∂ν ξ µ .

In (5.1) we pulled out the classical gauge-fixing and ghost actions Sgf and Sgh
from Γk , and also assumed that the coupling to the BRST variations, for all
k, has the same form as in the bare action. The remaining functional Γgravk is
allowed to depend on both gµν and ḡµν .
Without loss of generality it can be decomposed as

Γgrav b k [g, ḡ],

[g, ḡ] ≡ Γ̄k [g] + Γ (5.2)
k

where Γ̄k [g] ≡ Γgrav [g, g] is the restriction of Γgrav bk ≡

to the diagonal, and Γ
k k
grav
Γk − Γ̄k denotes the resulting remainder. Note that Γ̄k is the same functional
we encountered in (4.99),

Γ̄k [g] = Γk [g, g, 0, 0; 0, 0], (5.3)

bk
and it approaches the one in (4.74) for k → 0. Furthermore, by definition, Γ
vanishes whenever its two arguments are equal:

b k [g, g] = 0.
Γ (5.4)

b k contains in particular quantum corrections to the gauge-fixing

Note also that Γ
term, which also vanishes for ḡ = g.
At this point Γgrav
k [g, ḡ] is still a rather general “bi-metric” functional, i.e., it
depends on two independent metrics, gµν and ḡµν . However, it is constrained
by the requirement that it must be invariant under the background gauge
transformations (4.50), i.e.,

δ B Γgrav
k [g, ḡ] = 0. (5.5)

b k [g, ḡ] are also δ B -invariant.

As a consequence, Γ̄k [g] and Γ
 5.1 General Classes of Truncations 129

5.1.2 A Class of Bi-Metric Truncations

Let us now list a number of further properties of the special class of bi-metric
truncations based on (5.1) satisfying (5.5):

(i) All actions of the form (5.1) are invariant under the background gauge
transformations (4.50) with (4.52):

¯ β, τ ] = 0.
δ B Γk [g, ḡ, ξ, ξ; (5.6)

(ii) The ansatz (5.1) complies with the large-k behavior (4.98) if, to lowest loop
order,
b k [g, ḡ] = 0 + · · · for k → ∞.
Γ̄k [g] = S[g] + · · · , Γ (5.7)

(iii) The ansatz (5.1) satisfies the antighost quantum equation of motion, (4.91),
exactly.

(iv) Substituting the ansatz (5.1) into (4.94) one finds that the BRST Ward
b k -part of Γgrav :
identity implies a non-trivial constraint on the Γ k

∫ b k [g, ḡ]
δΓ
dd x Lξ gµν = −Yk . (5.8)
δgµν (x)

Obviously the functional differential operator applied to Γ b k [g, ḡ] on the left-
hand side of (5.8) causes a gauge transformation on the first argument of
b k [g, ḡ] only. Despite the δ B -invariance of Γ
Γ b k , the result can be non-zero
b
because of the “extra ḡ-dependence” Γk [g, ḡ] may possess.
We observe that Γ b k = 0 is a good approximation provided we may neglect
Yk on the right-hand side of (5.8). The traces which define Yk amount to
loop integrals, and if we think in terms of a loop expansion, Yk amounts to
a higher loop effect and may be neglected in a first approximation.
At the non-perturbative level one can still try to set Γ b k = 0 and inves-
tigate the consequences in concrete examples. In Yang–Mills theory the
analogous truncation has been tested under more controlled conditions,
and it led to rather encouraging results [22, 23]. In the following sections
we perform various explicit calculations of the gravitational EAA in this
approximation.
(v) The Ward identity for the split symmetry, (4.96), when projected down onto
Γgrav , yields likewise a rather complicated condition on the Γ b k -part; it has
k
δ b
the general structure ( δḡµν )Γk [g, ḡ] = 2 STr[· · · ]. Like (5.8), it is satisfied by
1

b k = 0 if the corrections from the supertrace can be neglected.

Γ
(vi) Inserting the truncation ansatz (5.1) into the exact FRGE (4.80) and dis-
carding all types of terms absent from the ansatz we obtain the following
130 Truncations of Single-Metric Type

reduced functional RG equation:

[( )−1 ]
1 −2 (2) grav grav
∂t Γk [g, ḡ] = Tr κ Γk [g, ḡ] + Rk [ḡ] ∂t Rk [ḡ]
2
[( )−1 ] (5.9)
gh gh
− Tr −M[g, ḡ] + Rk [ḡ] ∂t Rk [ḡ] .

The first trace on the right-hand side of this equation is due to the metric
fluctuations, hµν , and the second one stems from the Faddeev–Popov ghosts.
The equation is written in terms of the bi-metric action

Γk [g, ḡ] ≡ Γk [g, ḡ, 0, 0; 0, 0]

= Γgrav
k [g, ḡ] + Sgf [g − ḡ; ḡ] (5.10)
b k [g, ḡ].
= Γ̄k [g] + Sgf [g − ḡ; ḡ] + Γ

(2)
Furthermore, Γk denotes the Hessian of Γk [g, ḡ] with respect to gµν at
(2) ( δ )2
fixed ḡµν . Symbolically, Γk [g, ḡ] ∝ δg Γk [g, ḡ]. Choosing the generalized
harmonic coordinate condition, for example, the classical kinetic operator
of the ghosts, M, is given by (4.56).

(vii) In Chapter 4 we discussed self-consistent backgrounds and saw in particular

that those with vanishing ghosts are determined by the simplified tadpole
equation (4.106). For the family of actions (5.1), (5.2) it assumes the form
[154]
δ { }
b k [g, ḡ]
Γ̄k [g] + Γ = 0. (5.11)
δgµν (x) sc
g=ḡ;ḡ=ḡk

∫
Note that the gauge-fixing term Sgf ∝ (Fh)2 drops out from this equation
since, being bilinear in h ≡ g − ḡ, its derivative vanishes at h = 0.

5.1.3 The Class of Single-Metric Truncations

Solving the reduced functional RG equation (5.9) is still not an easy task. There-
fore, it seems advisable to restrict the truncation ansatz even further for a first
analysis of this equation. A well-motivated subclass of the functionals (5.1) are
the so-called single-metric truncations in which one either sets Γ b k [g, ḡ] to zero
exactly, or at most allows it to be of the same form as Sgf , though with a different
prefactor:
b k [g, ḡ] ≡ 0
Γ or b k ∝ Sgf .
Γ (5.12)
 5.1 General Classes of Truncations 131

Adopting (5.12), the only scale dependence in the EAA enters via Γ̄k [g],
and thus the truncated theory space boils down to diffeomorphism-invariant
functionals of a single metric: Ttrunc = {Ā[g]|δ B Ā[g] = 0}.
The definition of the truncation is completed by specifying how a generic func-
tional A[g, ḡ] is projected onto the subspace Ttrunc . We define this projection to
act on bi-metric functionals by equating the two metrics,

A[g, ḡ] 7→ A[g, ḡ] ≡ Ā[g] (5.13)

ḡ=g

or equivalently in the “semicolon notation,” by setting the fluctuation field to

zero:
A[h; ḡ] 7→ A[h; ḡ] ≡ Ā[g]. (5.14)
h=0, ḡ=g

Note that since Ā ∈ Ttrunc depends on one field only, it does not matter whether
we denote this argument by gµν or ḡµν . Therefore, replacing ḡ by g in (5.14) is
only a trivial renaming of the independent variable.
When we apply this projection to both sides of the reduced FRGE (5.9) we
obtain a flow equation for the single-metric functional. The only place in this
equation where the intrinsic bi-metric nature of the gravitational EAA still plays
a role is the gauge-fixing term. Under the first trace in (5.9) we need the projected
(2) (2) (2)
Hessian of Γk [g, ḡ] = Γ̄k + Sgf , i.e., Γk [g, ḡ] = Γ̄k [g] + Sgf [g − ḡ; ḡ], evaluated
(2)
at h ≡ g − ḡ = 0. Projecting the gauge-fixing term we encounter at first Sgf [0; ḡ]
( δ )2
which must be read as δh Sgf [h; ḡ] h=0 : the differentiation with respect to hµν
is performed prior to setting hµν = 0, of course. As a second, trivial step one may
(2) ( δ )2
rename g → ḡ, so that then, symbolically, Sgf [0; g] ∝ δh Sgf [h; ḡ] h=0, ḡ=g .
In this manner (5.9) projects onto the following functional RG equation for
the single-metric EAA:

[( )−1 ]
1
Tr κ−2 Γ̄k [g] + κ−2 Sgf [0; g] + Rgrav
(2) (2) grav
∂t Γ̄k [g] = k [g] ∂ R
t k [g]
2
[( )−1 ] (5.15)
− Tr −M[g, g] + Rgh k [g] ∂t Rgh
k [g] .

This equation inherits the background gauge invariance from its exact pre-
cursor. As a consequence, if a solution is invariant under general coordinate
transformations at one scale, δ B Γ̄k [g] = 0, it is so at any other scale as well.
As it stands, (5.15) applies to the case when Γ b k is put to zero exactly. If instead
b (2)
Γk is allowed to be proportional to Sgf , the prefactor of Sgf in (5.15) changes
correspondingly.
The ultimate justification of the truncation leading to (5.15) will consist in
systematically enlarging the subspace Ttrunc and demonstrating that the results
132 Truncations of Single-Metric Type

stabilize at a certain point so that a not too large subspace Ttrunc is already
sufficient to achieve the desired level of accuracy.
There is a heuristic argument which indicates that the single-metric trunca-
tions are a sensible first guess to study: According to (ii) above, Γb k does indeed
vanish for k → ∞ and so it is plausible to try setting Γb k = 0 also at lower scales.
We know that this is in conflict with the BRST Ward identity (5.8), which tells
us that Γ b k = 0 is only a good approximation provided we may neglect Yk . The
functional traces which define Yk amount to loop integrals, so if we think in
terms of a systematic loop expansion Yk is indeed a higher-order effect and may
be neglected in a first approximation.
At the non-perturbative level one can still employ the ansatz Γ b k = 0, try to
justify it a posteriori by non-perturbative means, and investigate its consequences
in concrete examples. In the next section we perform an explicit calculation in
this approximation.

5.2 The Einstein–Hilbert Truncation

In this section we illustrate the use of the reduced FRGE (5.9) using the perhaps
most natural example of a single-metric truncation. It is inspired by the classical
Einstein–Hilbert action, in d dimensions:
∫
1 √ { }
S= dd x g −R(g) + 2Λ . (5.16)
16πG

We are going to use a truncation that replaces in (5.16) the classical Newton
constant G and cosmological constant Λ by k-dependent functions Gk and λ̄k ,
respectively. We then determine the scale dependence of these running coupling
constants by solving the flow equation (5.9) with an ansatz of the single-metric
type:
b k + Sgf ,
Γk [g, ḡ] ≡ Γ̄k [g] + Γ (5.17)
b k ∝ Sgf . Explicitly, the Einstein–Hilbert truncation amounts to the choice
with Γ
∫
1 √ { }
Γ̄k [g] = dd x g −R(g) + 2λ̄k ,
16πGk
( ) (5.18)
b k [g, ḡ] = α Ḡ − 1 Sgf [g − ḡ; ḡ].
Γ Gk α

The first equation of (5.18) is the expected k-dependent generalization of the

classical Einstein–Hilbert action, while the second one exploits the possibility to
change the gauge sector by a non-zero Γ b k proportional to the classical Sgf . This
b
choice for Γk is motivated by the algebraic simplifications it will entail later on.
In (5.18), α is the gauge-fixing
∫ d √parameter that occurs in the prefactor of the
gauge-fixing term, Sgf ≡ 2α
1
d x ḡ ḡ µν Fµ Fν , and Ḡ denotes the k-independent
 5.2 The Einstein–Hilbert Truncation 133

reference value of the (dimensionful) Newton constant which we had introduced

into the gauge-fixing functions of (4.54) via the constant mass parameter

κ ≡ (32π Ḡ)− 2 .
1
(5.19)

With the above ansatz the calculation of the RG flow of Gk and λ̄k
will be simplest
√ if one adopts a special member of the family of gauges
Fµ [h; ḡ] = 2κ Fµ αβ [ḡ]hαβ with Fµ αβ [ḡ]hαβ ≡ D̄ν hµν − ϖ D̄µ ḡ αβ hαβ . Letting
ϖ = 21 , which amounts to the standard harmonic gauge, and choosing in addition
the gauge-fixing parameter α = 1, simplifies the algebra considerably.
It remains to clarify why we are allowed to freely add any convenient Γ b k ∝ Sgf
b
to the truncation ansatz in the first place. After all, the chosen Γk replaces the
constant gauge-fixing parameter α in the prefactor of the original Sgf by the
scale-dependent ratio Gk /Ḡ. However, Sgf [h; ḡ] is bilinear in the fluctuation field
h ≡ g − ḡ, hence “bi-metric,” and therefore, by the single-metric hypothesis, the
scale dependence of its coefficient is assumed to have negligible, or acceptably
small, impact on Γ̄k , i.e., on Gk and λ̄k . Hence, even if α is now k-dependent with
a k-dependence inherited from Gk , which is probably not the correct one, it is
consistent within the single-metric setting to neglect this effect. (Only in a much
more ambitious
∫ bi-metric calculation would one include a running gauge-fixing
√
action 2α1 k dd x ḡ ḡ µν Fµ Fν in the ansatz and then compute the actual running
of αk using the FRGE; see [155].)
It is convenient to refer the running Newton constant to its fixed reference
value Ḡ by means of the dimensionless function ZNk ≡ Ḡ/Gk :

Ḡ
Gk ≡ . (5.20)
ZNk

The complete truncation ansatz then assumes the following form:

∫
√ { }
Γk [g, ḡ] = 2κ2 ZNk dd x g −R(g) + 2λ̄k
∫ (5.21)
√ ( )( )
2
+ κ ZNk dd x ḡ ḡ µν Fµ αβ gαβ Fν ρσ gρσ .

In writing the gauge-fixing term we exploited that Fµ αβ gαβ = Fµ αβ hαβ since

D̄µ ḡαβ = 0.

5.2.1 Derivation of the β-Functions for g and λ

In order to determine the functions ZNk and λ̄k ∫we must project
∫√ the evolution
√
equation on the space spanned by the operators g and gR. After having
inserted the ansatz into the reduced evolution equation (5.9) and performed the
134 Truncations of Single-Metric Type

gµν -derivatives we may set ḡµν = gµν so that the gauge-fixing term in (5.21)
vanishes. As a result, the left-hand side of the evolution equation reads:
∫
√ [ ( )]
∂t Γk [g, g] = 2κ2
dd x g −R(g)∂t ZNk + 2∂t ZNk λ̄k . (5.22)

On the right-hand side of (5.9) we must perform a derivative

∫√ ∫expansion and
√
retain only the terms proportional to the two invariants g and gR, respec-
tively. Equating the result to (5.22) we can read off the sought-for system of
ordinary differential equations for ZNk and λ̄k .

(1) Under the trace appearing in the reduced FRGE we need the second
b k + Sgf at fixed ḡµν . We expand
functional derivative of Γk [g, ḡ] ≡ Γ̄k + Γ

Γk [ḡ + h, ḡ] = Γk [ḡ, ḡ] + O(h) + Γquad

k [h; ḡ] + O(h3 ), (5.23)
∫
√ [ ]
Γquad
k [h; ḡ] = ZNk κ 2
dd x ḡ hµν −K µν ρσ D̄2 + U µν ρσ hρσ . (5.24)

Here and in the following, indices are always raised and lowered with ḡµν .
Furthermore, the tensors K and U are given by
1[ µ ν ]
K µν ρσ = δρ δσ + δσµ δρν − ḡ µν ḡρσ (5.25)
4
and
1[ µ ν ]( ) 1 [ µν ]
U µν ρσ = δ δ + δσµ δρν − ḡ µν ḡρσ R̄ − 2λ̄k + ḡ R̄ρσ + ḡρσ R̄µν
4 ρ σ 2 (5.26)
1[ ] 1[ ν µ ]
− δρµ R̄νσ + δσµ R̄νρ + δρν R̄µσ + δσν R̄µρ − R̄ ρ σ + R̄νσµρ .
4 2
In (5.26) all geometrical quantities with an overbar are constructed from the
background metric.1

(2) In order to partially diagonalize the quadratic form (5.24) we decompose hµν
as the sum of a traceless tensor h̊µν and a trace part involving ϕ ≡ ḡ µν hµν :

hµν = h̊µν + d−1 ḡµν ϕ, ḡ µν h̊µν = 0. (5.27)

As a consequence, (5.24) becomes

∫ { [ ]
d √ 1
Γquad
k [h; ḡ] =ZNk κ 2
d x ḡ h̊µν − D̄ 2
− 2λ̄k + R̄ h̊µν
2
( ) [ ]
d−2 d−4
− ϕ −D̄2 − 2λ̄k + R̄ ϕ (5.28)
4d d
}
d−4
− R̄µν h̊νρ h̊µρ + R̄αβνµ h̊βν h̊αµ + ϕR̄µν h̊µν .
d

1 Our conventions for curvature tensors are summarized in Appendix A.

 5.2 The Einstein–Hilbert Truncation 135

(3) The differential

∫ √ equations
∫ √ for ZNk and λ̄k are obtained by comparing the
coefficients of g and gR on both sides of the evolution equation at ḡµν =
gµν . For this purpose we may insert an arbitrary ∫√ family
∫ √ of metrics ḡµν which
is general enough to identify the terms ḡ and ḡ R̄ and to distinguish
∫√ 2
them
∫ √ µν from higher-order terms in the derivative expansion, such as ḡ R̄ or
ḡ R̄ D̄µ D̄ν R̄, for instance.
We exploit this freedom by assuming that ḡµν corresponds to a maximally
symmetric space so that its Riemann and Ricci tensor, respectively, have the
structure [156]
1
R̄µνρσ = [ḡµρ ḡνσ − ḡµσ ḡνρ ] R̄,
d(d − 1)
(5.29)
1
R̄µν = ḡµν R̄.
d
From now on the curvature scalar R̄ parametrizes the family of metrics inserted,
and it should be regarded as an externally prescribed number rather than a
functional of the metric.
For a maximally symmetric background the quadratic action reads
∫ {
quad 1 [ ]
Γk [h; ḡ] = ZNk κ 2
d x h̊µν −D̄2 − 2λ̄k + CT R̄ h̊µν
d
2
( ) } (5.30)
d−2 [ ]
− ϕ −D̄ − 2λ̄k + CS R̄ ϕ ,
2
2d
with the dimension-dependent constants
d(d − 3) + 4 d−4
CT ≡ , CS ≡ . (5.31)
d(d − 1) d
(4) Before continuing we need to specify the precise form of the cutoff operators
Rgrav
k and Rghk which describe the transition from the high-momentum to the
low-momentum regime. Either of them has the structure
2
Rk [ḡ] = Zk k 2 R(0) (− D̄
k2 ) (5.32)

whereby the matrix Zk should be adjusted in such a way that for every low-
momentum mode the cutoff combines with the kinetic term of this mode to
(−D̄2 + k 2 ) times a constant.
Looking at (5.30) we see that the respective kinetic terms for h̊µν and ϕ differ
by a factor of −(d − 2)/2d. This suggests the following choice:
[ ]
grav µνρσ µνρσ d − 2 µνρσ
(Zk ) = (1 − Pϕ ) − Pϕ ZNk . (5.33)
2
( )
Here (1)µν ρσ = 21 δµρ δνσ + δνρ δµσ is the unit on the space of symmetric 2-tensors
and
ḡµν ḡ ρσ
(Pϕ )µν ρσ = (5.34)
d
136 Truncations of Single-Metric Type

is the projector on the trace part of the metric. For the traceless tensor, (5.33)
yields Zkgrav = ZNk 1, while for ϕ the different normalization is taken into account.
With this cutoff we obtain the following operators in the h̊ and the ϕ-sector,
respectively:
( ) [ ( ) ]
κ−2 Γk [g, g] + Rgrav
(2) 2
k = ZNk −D2 + k 2 R(0) − Dk2 − 2λ̄k + CT R
h̊h̊
( ) d−2
κ−2 Γk [g, g] + Rk
(2) grav
=− (5.35)
ϕϕ 2d
[ ( 2
) ]
× ZNk −D2 + k 2 R(0) − Dk2 − 2λ̄k + CS R .

At this point we may set ḡ = g, thus omitting the bars from the metric and the
curvature in the following.

(5) The last ingredient that is required before we can assemble the entire RG
equation is the Faddeev–Popov operator. From (4.56) one obtains, at ḡ = g,
[ ]
M[g, g]µν = δνµ D2 + Rµν = −δνµ −D2 + CV R , (5.36)

with the constant

1
CV ≡ − . (5.37)
d
The second equality in (5.36) follows upon using (5.29) for a maximally symmet-
ric background. Since we neglect all renormalization effects in the ghost action
we set Zkgh ≡ 1 in Rgh
k and obtain
( )
D2
− M + Rgh
k = −D 2
+ k 2 (0)
R − k 2 + CV R. (5.38)

(6) Equipped with the cutoff and Hessian operator (5.36) and (5.38) we can now
return to the FRGE.
Let us write Sk (R) for the right-hand side of the reduced RG equation (5.9)
evaluated at ḡ = g. Inserting (5.36) and (5.38) there we arrive at
[ ] [ ]
Sk (R) = TrT N (A + CT R)−1 + TrS N (A + CS R)−1
[ ] (5.39)
− 2 TrV N0 (A0 + CV R)−1 ,

with the two operators

( 2
)
A ≡ −D2 + k 2 R(0) − D
k 2 − 2λ̄k ,
[ ( )]
−1 2
N ≡ (2ZNk ) ∂t ZNk k 2 R(0) − D k2 (5.40)
[ ] ( ) ( )
1 2
D2
= 1 − ηN (k) k 2 R(0) − D k 2 + D 2 (0)′
R − k 2 .
2
 5.2 The Einstein–Hilbert Truncation 137

Here a prime denotes the derivative with respect to the argument, and

ηN ≡ −∂t ln ZNk (5.41)

is the anomalous dimension related to the running Newton constant. Further-

more, the operators N0 and A0 are defined similarly to (5.40) but with λ̄k = 0
and ZNk = 1, i.e., ηN (k) = 0.
The right-hand side of the flow equation, (5.39), involves traces of functions
of the covariant Laplacian D2 ≡ g µν Dµ Dν acting on traceless symmetric tensors
(“T ”), scalars (“S”), and vectors (“V ”). Because we need only the zeroth and
the first order in the curvature scalar we can expand in R:
[ ] [ ] [ ]
Sk (R) = TrT N A−1 + TrS N A−1 − 2 TrV N0 A−1 0
( [ ] [ ] [ ])
− R CT TrT N A−2 + CS TrS N A−2 − 2CV TrV N0 A−2 0 (5.42)
+ O(R2 ).

(7) We evaluate the traces in (5.42) by taking advantage of the familiar heat
kernel expansion2
[ ] ( i )d/2 ∫
√
{
1
}
Tr e−isD =
2
tr(I) dd x g 1 − isR + O(R2 ). (5.43)
4πs 6
Here I denotes the unit matrix of the space of fields on which D2 acts. Hence,
tr(I) is the number of independent field components. In the case at hand we
encounter in particular
trS (I) = 1,
trV (I) = d, (5.44)
1
trT (I) = (d − 1)(d + 2).
2
As for generic functions of the covariant Laplacian, let
∫ ∞us considerisz
an arbitrary
f f
function W with Fourier transform W , i.e., W (z) = −∞ dsW (s)e . Then the
expansion of the trace
∫ ∞ [ ]
Tr[W (−D2 )] = f (s) Tr e−isD2
ds W (5.45)
−∞

is given by [14]:
{ ∫
2 −d/2 √
Tr[W (−D )] = (4π) tr(I) Qd/2 [W ] dd x g
∫ } (5.46)
1 d √
+ Qd/2−1 [W ] d x gR + O(R2 ).
6

2 See the Appendix for a brief review of the heat kernel and Laplace transform techniques
needed here.
138 Truncations of Single-Metric Type

Here we introduced the “Q-functionals”

∫ ∞
Qn [W ] ≡ f (s).
ds (−is)−n W (5.47)
−∞

It is not difficult to reexpress (5.47) in terms of the original function W (z) itself.
We obtain [14]:

Q0 [W ] = W (0),
∫ ∞
1
Qn [W ] = dz z n−1 W (z) for n > 0, (5.48)
Γ(n) 0
( d )|n|
Qn [W ] = − W (z) for n < 0.
dz z=0

(8) The next step toward the beta functions consists of using (5.46) in order to
evaluate the traces in (5.42) and to combine the resulting expression Sk (R) with
the left-hand side of the evolution
∫√equation, (5.22).
Comparing the coefficients of g we read off the following equation:
( ) {
∂t ZNk λ̄k = (4κ2 )−1 (4π)−d/2 trT (I)Qd/2 [N /A]
} (5.49)
+ trS (I)Qd/2 [N /A] − 2 trV (I)Qd/2 [N0 /A0 ] .
∫√
Likewise, equating the prefactors of gR gives rise to:

∂t ZNk = − (12κ2 )−1 (4π)−d/2

[ { }
× trT (I) Qd/2−1 [N /A] − 6CT Qd/2 [N /A2 ]
{ } (5.50)
+ trS (I) Qd/2−1 [N /A] − 6CS Qd/2 [N /A2 ]
{ }]
− 2 trV (I) Qd/2−1 [N0 /A0 ] − 6CS Qd/2 [N0 /A20 ] .

In (5.49) and (5.50), N and A are considered functions of the real variable z now
which replaces −D2 in (5.40).
To proceed it is convenient to introduce the following normalized threshold
functions [14]:
∫ ∞
1 R(0) (z) − zR(0)′ (z)
Φpn (w) ≡ dz z n−1 [ ]p ,
Γ(n) 0 z + R(0) (z) + w
∫ ∞
(5.51)
e pn (w) ≡ 1 R(0) (z)
Φ dz z n−1
[ ]p .
Γ(n) 0 z + R(0) (z) + w

Here p = 1, 2, 3, · · · , and n ≥ 0. A careful evaluation of the limit n ↘ 0 yields:

e p (w) = (1 + w)−p .
Φp0 (w) = Φ (5.52)
0
 5.2 The Einstein–Hilbert Truncation 139

After having decided about a concrete cutoff shape function R(0) (z) it is straight-
forward to perform the z-integrations and thus obtain the threshold functions
related to this shape function; see Appendix E for examples.
When expressed in terms of these threshold functions, (5.49) assumes the form
( ) [ ( )
∂t ZNk λ̄k = (16κ2 )−1 (4π)−d/2 k d 2d(d + 1)Φ1d/2 − 2kλ̄2k
( )] (5.53)
e1
− 8dΦ1d/2 (0) − d(d + 1)ηN Φd/2 − k2
2λ̄k
.

Similarly the second equation (5.50) becomes

∂t ZNk = −(24κ2 )−1 (4π)−d/2 k d−2

[ { ( ) 1 ( )}
× d(d + 1) Φ1d/2−1 − 2kλ̄2k − ηN Φ e1
d/2−1 − k2
2λ̄k
2
{ ( ) 1 ( )} (5.54)
e2
− 6d(d − 1) Φ2d/2 − 2kλ̄2k − ηN Φ − 2λ̄k
d/2 k 2
]2
− 4dΦ1d/2−1 (0) − 24Φ2d/2 (0) .

At this point we could integrate the two equations (5.53), (5.54) and obtain the
running dimensionful Newton and cosmological constant, respectively.

5.2.2 Structure of the β-Functions for g and λ

Let us switch to dimensionless coupling constants now. We introduce the
dimensionless running Newton constant
−1
gk ≡ k d−2 Gk ≡ k d−2 ZNk Ḡ (5.55)

and the dimensionless cosmological constant:

λk ≡ k −2 λ̄k . (5.56)
−1
Recall that Gk ≡ ZNk Ḡ is the dimensionful running Newton constant at scale k,
and that [Gk ] = 2 − d and [λ̄k ] = 2 in d spacetime dimensions.
Applying ∂t to (5.55) we are led to the following RG equation for the
dimensionless Newton constant:
[ ]
∂t gk = d − 2 + ηN (gk , λk ) gk . (5.57)

From (5.54) we obtain an implicit equation which allows us to determine the

anomalous dimension ηN as a function of gk and λk :

ηN (k) = gk B1 (λk ) + ηN (k) gk B2 (λk ). (5.58)

140 Truncations of Single-Metric Type

The coefficient functions B1 and B2 read explicitly:

1
B1 (λk ) ≡ (4π)1−d/2 ×
3 [
× d(d + 1)Φ1d/2−1 (−2λk ) − 6d(d − 1)Φ2d/2 (−2λk )
]
− 4dΦ1d/2−1 (0) − 24Φ2d/2 (0) , (5.59)
1
B2 (λk ) ≡ − (4π)1−d/2 ×
6[ ]
e1
× d(d + 1)Φ (−2λk ) − 6d(d − 1)Φe 2 (−2λk ) .
d/2−1 d/2

We can easily solve the relation (5.58) for ηN in terms of gk and λk :

gk B1 (λk )
ηN (gk , λk ) = . (5.60)
1 − gk B2 (λk )

Note that the anomalous dimension has no explicit k-dependence. It depends on

the scale only via gk and λk .
As for the cosmological constant, the scale derivative of λk is related to (5.53)
according to
∂t λk = −(2 − ηN )λk + 32πgk κ2 k −d ∂t (ZNk λ̄k ) (5.61)
so that we obtain the following RG equation for the cosmological constant:

1
∂t λk = βλ (gk , λk ) = −(2 − ηN )λk + gk (4π)1−d/2
[ 2
× 2d(d + 1)Φd/2 (−2λk ) − 8dΦ1d/2 (0)
1 (5.62)
]
− d(d + 1)ηN Φ e 1 (−2λk ) .
d/2

The pair of equations (5.57) and (5.62) constitutes a system of coupled

differential equations for the dimensionless Newton and cosmological constant.
This system of RG equations has the general structure
[ ]
∂t gk = βg (gk , λk ) ≡ d − 2 + ηN (gk , λk ) gk ,
(5.63)
∂t λk = βλ (gk , λk ).

On its right-hand side it displays the two-component vector field β = (βλ , βg )

comprised of the dimensionless β functions βg and βλ , respectively. As expected
on general grounds, the system has indeed become autonomous after switching
to dimensionless couplings.
The gk and λk dependence of the anomalous dimension ηN (gk , λk ) has the
structure of (5.60) whereby the two functions B1 and B2 are explicitly given by
(5.59). Furthermore, βλ is defined by the right-hand side of (5.62).
 5.2 The Einstein–Hilbert Truncation 141

As we can see, the non-trivial content of this system of RG equations is

encoded in the three functions B1 (λ), B2 (λ), and βλ (g, λ), respectively, which we
expressed in terms of the standardized threshold functions Φpn and Φ e pn defined in
(5.51).

(1) Threshold functions and scheme dependence. The functions Φpn and
e p are manifestly cutoff-scheme dependent via the shape function R(0) . The
Φ n
latter is required to interpolate smoothly between R(0) (0) = 1 and R(0) (∞) = 0
but is quite arbitrary otherwise. In addition, typical regulators R(0) do not vanish
too quickly for small values of its argument, i.e., z + R(0) (z) ≥ 1, ∀z ≥ 0.
The definition of the threshold functions (5.51) immediately implies that
Φpn (w), Φ e pn (w) → 0 for w → ∞.3 Furthermore, one notes by examining the deno-
minator of the integrands that the functions Φpn (w) and Φ e pn (w) are well defined
only for arguments w > −1. For w < −1 the denominator vanishes for some z in
the region of integration, yielding a non-integrable singularity. Therefore, Φpn (w)
and Φ e pn (w) are finite and well defined for w ∈ (−1, ∞) only.
In βg and βλ the threshold functions are being evaluated at w = −2λ. As a
consequence, the beta functions are well defined for λ < 12 only.

(2) Sharp cutoff. A particularly convenient choice of R(0) (z) for which the z-
integration can be done analytically is the sharp cutoff (sc) defined in (E.4). It
leads to a perfectly explicit final form of the RG equations for spacetimes of any
dimensionality d.
Upon substituting the threshold functions (E.5), the corresponding functions
B1 , B2 , and βλ read as follows:

{
1 d(d + 1)
sc
B1 (λ) = (4π) 1−d/2
− ln(1 − 2λ) + d(d − 3)φd/2−1
3 Γ(d/2 − 1)
}
6d(d − 1) 1 24
− − ,
Γ(d/2) 1 − 2λ Γ(d/2)
1 d(d + 1)
B2 (λ)sc = − (4π)1−d/2 , (5.64)
6 Γ(d/2)
[
1 2d(d + 1)
βλ (λ, g) = −(2 − ηN )λ + (4π)
sc sc 1−d/2
g − ln(1 − 2λ)
2 Γ(d/2)
]
d(d + 1) sc
+ 2d(d − 3)φd/2 − η .
Γ(d/2 + 1) N

3 If z + R(0) (z) ≥ ε we may use 1

≤ 1
[ε+w]p
under the z-integrals of (5.51). Since
[z+R(0) (z)+w]p
the right-hand side of this estimate is independent of z, we may pull it out of the integral,
which still converges then thanks to the fall-off properties of R(0) . As a consequence, this
prefactor causes the threshold functions (with p > 0) to vanish in the limit w → ∞.
142 Truncations of Single-Metric Type

As always, ηN (g, λ) is of the form (5.60), where the functions B1,2 ≡ B1,2 sc

are now given by the above expressions. Furthermore, the φn s are numerical
constants whose values are given in (E.12).
Note that the sharp cutoff is special in the sense that B2 (λ)sc turns out to be
a constant, i.e., independent of λ.
In the following sections we analyze the RG flow implied by the Einstein–
Hilbert truncation in detail. We close this section with a number of further
comments.

(3) Resummed perturbative contributions. The above RG equations while

approximate are nevertheless of a manifestly non-perturbative nature. Taylor
expanding for example (5.60) with respect to the dimensionless Newton constant,
we obtain
∞
∑ ( )n
ηN (g, λ) = B1 (λ) g n+1 B2 (λ) . (5.65)
n=0

Obviously the Einstein–Hilbert truncation sums up certain contributions to ηN

and βg ≡ [d − 2 + ηN ]g, which are of arbitrarily high order in the coupling con-
stant g.

(4) The “Z = ζ”-rule. Let us return to the idea of an adjusted cutoff, which
was briefly mentioned in Section 5.2.5.
There we described a rule for fixing the matrix Zk that appears in the ansatz
for the cutoff operator, Rk = Zk k 2 R(0) (−D2 /k 2 ). We applied this rule in the
above calculation where it led to the form (5.33) of the submatrix Zkgrav for
the graviton sector. However, there is a rather subtle issue related to the choice
(5.33), as we discuss next.
Let us try to apply the rule formulated in Section 5.2.5 to the cutoff for hµν . We
leave the truncation arbitrary for a moment and concentrate on the contribution
(2)
of a fixed mode χ contained in hµν . We assume that χ is an eigenfunction of Γk
2 2
with eigenvalue ζk p , where p is a positive eigenvalue of some covariant kinetic
(2)
operator, typically of the form −D̄2 + R̄-terms. For theories with Γk > 0, the
wave-function normalization ζk is positive. In this case our general rule is to set
Zk = ζk because this guarantees that for the low-momentum modes the effective
(2)
inverse propagators Γk + Rk become ζk (p2 + k 2 ), as it should be.
(2)
In the Einstein–Hilbert truncation, the condition Γk > 0 is met for the
traceless part of the metric fluctuation, h̊µν . From (5.30) we can read off the
corresponding ζk , and it is found to be positive.
The trace part ϕ, on the other hand, is seen to possess a negative normalization
constant ζk in all dimensions d > 2. So the important question is how do we choose
Zk if ζk is negative?
If we continue to use Zk = ζk , the FRGE is perfectly well defined because the
inverse propagator −|ζk |(p2 + k 2 ) never vanishes, and the functional traces on
 5.3 Properties of the Einstein–Hilbert Flow 143

its right-hand side do not suffer from any IR problems. For instance, if we write
down their perturbative expansion we see that all propagators are correctly cut
off in the IR, and that loop momenta smaller than k are properly suppressed.
Moreover, due to the projective structure of the trace arguments, the overall
minus sign drops out so that the conformal mode contributes like a scalar field
with a standard kinetic term.
On the other hand, if we set Zk = −ζk , we are led to −|ζk |(p2 − k 2 ), which
introduces a spurios singularity at p2 = k 2 , and so the cutoff operator fails to
make the theory IR finite in this case.
At first sight the choice Zk = |ζk | might have appeared ( ∫ the )more natural
one because only if Zk > 0 the factor exp(−∆Sk ) ∼ exp − Rk ϕ2 is a damped
exponential which suppresses the low momentum modes in the usual way.
Nevertheless, there are good reasons to adopt the rule of setting Zk = ζk for
either sign of ζk .
As for the conformal factor of the metric, our above choice (5.33) complies with
this rule. At least in the Einstein–Hilbert truncation the evolution equations are
well defined when Zk = ζk (but not with Zk = |ζk |) even though it appears even
more difficult than usual to give a meaning
( ∫ to the ) functional integral itself then.
In the case Zk = ζk < 0 the factor exp + |Rk |ϕ is a growing exponential and it
2

might seem that this enhances rather than suppresses the low momentum modes.
However, as suggested by the perturbative argument above, this conclusion is too
naive probably. (If one invokes, for example, Hawking’s prescription of rotating
the contour of integration over ϕ so that it is parallel to the imaginary axis, both
the kinetic term and the cutoff lead to damped exponentials.)
The “Z = ζ”-rule may find further justification when one deals with quantum
mechanics-type functional integrals over oscillating exponentials eiS rather than
e−S . For the FRGE this change is by far less dramatic than for the integral. Apart
from the obvious substitutions Γk → −iΓk , Rk → −iRk , the evolution equation
remains structurally unaltered. For Zk = ζk it has all desired features, and ζk < 0
does not seem to pose a special problem.

5.3 Properties of the Einstein–Hilbert Flow

In order to gain a first understanding of the RG trajectories from the Einstein–
Hilbert truncation, arising as the solutions of the system (5.63), we begin by
studying the two possible “subtruncations” which it admits.
At first, in Section 5.3.1 we will omit the cosmological constant term in the
ansatz for Γk , and solve the remaining RG equation for Newton’s constant alone.
∫ √ in Section 5.3.2, we freeze the running of Gk , setting Gk ≡ const
Conversely,
in the gR-term, and study the resulting simplified scale dependence of the
cosmological constant.
In Section 5.3.3 and 5.3.4 we then explore the fixed points of the full coupled
system for gk and λk , and analyze its entire phase portrait.
144 Truncations of Single-Metric Type

Finally, after also introducing some close “relatives” of the Einstein–Hilbert

truncation, and discussing the available methods to judge the reliability of trun-
cations, we summarize the status of the Asymptotic Safety conjecture in light of
those findings in Section 5.3.7.
From now on we focus on four spacetime dimensions (d = 4). The generalization
to d ̸= 4 would be straightforward; we will comment on the corresponding results
later on.

5.3.1 The Subtruncation of Vanishing λ

Let us consider the one-dimensional truncation∫ with g(k) as the only running
√
coupling constant. Omitting the term λ̄(k) dd x g from the ansatz for Γk gives
rise to a single differential equation for Newton’s constant then. It is obtained
from (5.63) by letting λ ≡ 0 in the first equation, and discarding the second one.4
Thus we are left with
[ ]
k∂k g(k) = 2 + ηN (g(k)) g(k) ≡ βg (g(k)), (5.66)

with ηN and βg given by

B1 g 1 − ω′ g
ηN (g) = ⇐⇒ βg (g) = 2g . (5.67)
1 − B2 g 1 − B2 g

At this level, B1 and B2 denote simple, albeit cutoff-dependent constants:

1
B1 ≡ B1 (0) = − [24Φ22 (0) − Φ11 (0)],
3π (5.68)
1 e 22 (0) − 5Φ
e 11 (0)].
B2 ≡ B2 (0) = [18Φ
6π
For convenience we also abbreviate
1
ω ≡ − B1 , ω ′ ≡ ω + B2 . (5.69)
2
Picking, for example, the exponential cutoff (E.1) we have explicitly

π2
Φ11 (0)exp = , Φ22 (0)exp = 1,
6 (5.70)
e 11 (0)exp = 1,
Φ e 22 (0)exp = 1 ,
Φ
2
which entails the following two positive constants:
4( π2 ) 2
ω= 1− , B2 = . (5.71)
π 144 3π
( )
4 We shall often write g(k) ≡ gk , λ(k) ≡ λk , etc., for RG trajectories, k 7→ g(k), λ(k) , and
(g, λ) for the coordinates on the dimensionless theory space.
 5.3 Properties of the Einstein–Hilbert Flow 145

While the precise value of ω is cutoff dependent, it turns out that

1
ω ≡ − B1 (0) > 0 (5.72)
2

for any admissible choice of the shape function R(0) . The relevance of a positive
value of ω will become clear in a moment.

(1) We observe that βg (g) has two zeros, namely at g = 0 and g = ω1′ , respectively.
As a result, the evolution equation (5.66) displays two fixed points g∗ , β(g∗ ) = 0,
namely an infrared-attractive Gaussian fixed point at g∗GFP = 0 and a ultraviolet-
attractive non-Gaussian fixed point at

1
g∗NGFP = . (5.73)
ω′
The non-Gaussian fixed point (NGFP) separates a weak coupling regime (g <
g∗NGFP ) from a strong coupling regime where g > g∗NGFP . Since the β-function
is positive for g ∈ [0, g∗NGFP ] and negative otherwise the solutions g(k), the
“renormalization group trajectories,” fall into the following three classes:

(i) Trajectories with g(k) < 0 for all k. They are attracted toward g∗GFP for
k → 0.
(ii) Trajectories with g(k) > g∗NGFP for all k. They are attracted toward g∗NGFP
for k → ∞.
(iii) Trajectories with g(k) ∈ [0, g∗NGFP ] for all k. They are attracted toward
g∗GFP = 0 for k → 0 and towards g∗NGFP for k → ∞.

Only the trajectories in (iii) are relevant to us. We will not allow for a negative
Newton constant, and we also discard the solutions of (ii) as they are entirely
within the strong coupling region and do not connect to a classical large distance
regime.

(2) The differential equation (5.66) with (5.67) is easily integrated analytically,
yielding
( )2
g g(k0 ) k
ω = [ ] ω′ . (5.74)
′
(1 − ω g) ′ k
ω ′
1 − ω g(k0 ) ω 0

This expression cannot be solved for g = g(k) in closed form. However, it is obvi-
ous that this solution indeed interpolates between the IR behavior g(k) ∝ k 2 for
k 2 → 0 and g(k) → ω1′ for k → ∞.
In order to obtain an approximate analytic expression for the running Newton
constant we observe that the ratio ω ′ /ω is actually very close to unity. With the
exponential shape function one has ω ≈ 1.2, B2 ≈ 0.21, ω ′ ≈ 1.4, g∗NGFP ≈ 0.71 so
that ω ′ /ω ≈ 1.18 is indeed not far from unity.
146 Truncations of Single-Metric Type

Replacing ω ′ /ω → 1 in (5.74) yields a rather accurate approximation with the

same general features as the exact solution. With this approximation we can
easily solve (5.74):
g(k0 )k 2
g(k) = . (5.75)
ωg(k0 )k + [1 − ωg(k0 )]k02
2

This function is actually an exact solution to the renormalization group

equation containing the approximate anomalous dimension

ηN (g) = −2ωg + O(g 2 ), (5.76)

which is nothing but the first term in the perturbation series of (5.65):
[ ∞
]
∑
ηN (g) = −2ωg 1 + n
(B2 g) . (5.77)
n=1

It is reassuring that for the trajectory (5.75) the product B2 g(k) remains neg-
ligibly small for all values of k. It assumes its largest value at the UV fixed point
where B2 g∗NGFP ≈ 0.15. Thus the linear approximation ηN ∝ g and the resulting
equation (5.75) provide us with a reliable approximation to the exact solution of
(5.66) and (5.67).
Within this approximation, the fixed point coordinate (5.73) gets replaced with
g∗NGFP = ω1 , and (5.75) indeed approaches this value asymptotically:

lim g(k) = g∗NGFP . (5.78)

k→∞

(3) We mentioned already that the constant ω is always positive. This implies
a negative anomalous dimension for the physically relevant case of a positive
Newton constant:

ηN < 0 if g>0 (antiscreening). (5.79)

Recalling the definition of the anomalous dimension,

k∂k Gk
ηN ≡ −k∂k ln ZNk ≡ −k∂k ln GḠk = , (5.80)
Gk
we therefore conclude that the dimensionful Newton constant is a decreasing
function of the scale k.
This behavior predicted by the Einstein–Hilbert truncation, first obtained in
[14], is known as gravitational antiscreening: Newton’s constant Gk decreases as
k increases, it is small in the UV, and grows larger as we evolve it toward the
infrared.
The sign of this effect is the same as for the non-abelian gauge coupling in
Yang-Mills theory and it is opposite to the one in QED.
Interpreting high momentum scales k as short distances, we see that gravity is
antiscreening in the sense that at large distances Newton’s constant is larger than
 5.3 Properties of the Einstein–Hilbert Flow 147

at small distances. This fits well with the intuitive picture that the gravitational
charge, the mass, is not screened by quantum fluctuations but rather receives an
additional positive contribution from the virtual particles and graviton excita-
tions surrounding it. This behavior is opposite to the one found in QED where
virtual charges screen the electromagnetic field.

(4) In terms of the dimensionful Newton constant G(k) ≡ g(k)/k 2 the solution
(5.75) reads
G(k0 )
G(k) = . (5.81)
1 + ωG(k0 )[k 2 − k02 ]
Henceforth, we set k0 = 0 for the reference scale and identify G0 ≡ G(0) with the
observed value of Newton’s constant in the extreme IR. This leads to the fol-
lowing very simple, but instructive formula for the running dimensionful Newton
constant [157]:
G0
G(k) = . (5.82)
1 + ω G0 k 2
This equation describes a smooth, monotonic interpolation between a classi-
cal regime in the infrared where, approximately, G(k) = const = G0 , and a fixed
point regime in the ultraviolet where instead g(k) = const = g∗NGFP . The crossover
between those regimes takes place when k ≈ mPl , i.e., near the Planck scale
−1/2
defined by mPl ≡ G0 .
For k ≪ mPl we may expand (5.82),

G(k) = G0 − ωG20 k 2 + O(G30 k 4 )

[ ( k )2 ( k 4 )] (5.83)
≡ G0 1 − ω +O .
mPl m4Pl

Obviously the leading quantum corrections at low scales are suppressed by a

( )2
factor of mkPl .
In the other extreme, for k ≫ mPl , the fixed point behavior prevails, i.e., G(k)
“forgets” its IR value G0 and evolves according to the power law:

1 1 g∗NGFP
G(k) = = . (5.84)
ω k2 k2

As it should be, the corresponding dimensionless Newton constant g(k) ≡ k 2 G(k)

approaches the constant value g∗NGFP there.

(5) Even at the exact, i.e., untruncated, level and in all truncations more general
than the present one, the existence of a NGFP, for purely dimensional rea-
sons, always leads to the power law G(k) = g∗NGFP k 2−d near the fixed point,
i.e., G(k) ∝ k −2 in d = 4. According to the definition (5.80), the k −2 -behavior of
148 Truncations of Single-Metric Type

the dimensionful Newton constant is tantamount to the fixed point value of the
anomalous dimension
NGFP
ηN = −2. (5.85)

Therefore, by continuity, we always have ηN < 0, hence antiscreening, on trajec-

tories near to a NGFP. However, there is no general reason that would guarantee
ηN to be negative for all scales along a given trajectory. This happened to be the
case for the trajectories we computed above, but as we will see later on, more
general (bi-metric) truncations can change the picture for k → 0.

5.3.2 The Subtruncation of Constant G

Turning to the second specialization of the Einstein–Hilbert system of RG equa-
tions, we now consider a segment of an RG trajectory, which is near the classical
regime so that we may ignore the k-dependence ∫ √ of the dimensionful Newton con-
stant on all scales of interest. We retain the gR-term in the truncation ansatz,
however, furnishing it with a constant prefactor in which G(k) ≡ const ≡ G0 . At
the dimensionless level this translates into g(k) = G0 k 2 . This simple function is
then inserted into the RG equation (5.62) for the cosmological constant.
Exploiting that ηN = 0 when G is constant we obtain the following differential
equation for the dimensionful cosmological constant, λ̄k :

1 4 [ ( ) ]
k∂k λ̄(k) = k G0 10Φ12 − 2λ̄(k)
k 2 − 8Φ 1
2 (0) . (5.86)
2π

We observe that the value of λ̄(k) backreacts in a potentially complicated way on

its own RG running via the threshold function Φ12 . This possibility is typical of
the FRGE-based approach; it is not found in a standard perturbative treatment.
Equation (5.86) simplifies further when λ̄ ≪ k 2 so that it is legitimate to replace
λ̄(k)/k 2 → 0 in the argument of the threshold function. The equation is easy to
integrate then. Imposing initial conditions at an arbitrary scale k1 , the solution
reads
1 1 [ ]
λ̄(k) = Φ (0)G0 k 4 − k14 + λ̄(k1 ). (5.87)
4π 2

This equation describes how the cosmological constant changes upon integrating
out the metric fluctuations, or “virtual gravitons,” which have momenta between
k and k1 .

(1) With (5.87) we have recovered the well-known, quartically cutoff-dependent

cosmological constant λ̄(k) ∝ k 4 which one encounters in basically any quantum
field theory model when performing perturbative loop calculations. Nevertheless,
more often than not it is simply ignored as one is interested in physics on a non-
dynamical flat spacetime only, or one compensates it by a finetuned counter term
 5.3 Properties of the Einstein–Hilbert Flow 149

(which amounts to the same), or one tries to find regularization schemes that
put it to zero formally (as in a version of dimensional regularization).
The perhaps best known calculation leading to the quartic cutoff dependence
is based on the following semiclassical argument, presumably due to Pauli [158]:
A non-zero cosmological constant λ̄ ≡ (8πG0 )ρvac represents a certain vacuum
energy density ρvac . The vacuum energy in turn receives contributions from all
(matter, gauge, graviton, etc.) field modes, each of them supplying its ground
state energy “ 12 ~ω(p).” The total vacuum energy is then identified with a sum
(integral) over all modes, i.e., all 3-momenta p. For a massless free field with
ω(p) = |p|, for example, we obtain
∫
~ d3 p
ρvac = |p| + const. (5.88)
2 (2π)3

This integral is highly UV divergent and requires regularization. Restricting |p|

to a finite interval one finds a quartic dependence of ρvac on both the UV and
the IR cutoff. This is indeed what we also found in (5.87).
Even if we restrict the integration to field modes between |p| = 0 and, say, the
scale of atomic physics only, the contribution to λ̄ which one obtains is already
larger than its measured value by many orders of magnitude. As a result, the
corresponding counter term, the constant in (5.88), must be finetuned at an
enormous level of precision, which is usually considered “unnatural.”
With this finetuning we encounter one face of the notorious cosmological
constant problem [159, 158].
We will return to this issue repeatedly. Even if Asymptotic Safety per se does
not solve the problem fully, it puts it into a new perspective that will become
clear eventually.
At this point we continue with a number of remarks related to (5.86) and (5.87).

(2) Taken at face value, (5.87) entails the same finetuning issue as the mode-
summation argument: Leaving the possibility of an asymptotically safe UV limit
aside for a moment, we adjust the constant of integration in (5.87) at the Planck
−1/2
scale, k1 = mPl ≡ G0 and assume that at this scale the cosmological constant
assumes a “natural” value of the order of the Planck scale, i.e., λ̄(mPl ) = Cm2Pl
with a dimensionless constant C = O(1). From (5.87) we then obtain the following
value at k = 0: ( )
1 1
λ̄(0) = C − Φ (0) m2Pl . (5.89)
4π 2

For the sake of argument we further assume that λ̄(0) and G0 ≡ 1/m2Pl can be
identified with the cosmological constant and Newton constant measured in real
nature. Then λ̄(0) is known to be about 120 orders of magnitude
( smaller than
)
m2Pl . Hence the constant C must be fixed in such a way that C − 4π 1
Φ12 (0) ≈
10−120 , whereby both C and (1/4π)Φ12 (0) are of order unity. This implies that
150 Truncations of Single-Metric Type

the initial value λ̄(k1 ) must be finetuned with a precision of about 120 digits,
and this, again, amounts to the “cosmological constant problem.”

(3) However, there are various reasons to believe that within the FRGE and
Asymptotic Safety context the above reasoning is way too simplistic and per-
haps even wrong. First of all, (5.87) is the result of drastic truncations and
approximations. It might be that this formula for λ̄(k) is incorrect even at the
qualitative level. In fact, later on we shall see that the Einstein–Hilbert truncation
is likely to become questionable at too low momentum scales, i.e., in the infrared.

(4) Furthermore, even if (5.87) was qualitatively correct, λ̄(k) may not be
straightforwardly inserted into the standard Einstein field equation in order to
relate it to, say, the observed expansion rate of the universe.
It is obvious that λ̄(k) cannot have the status of an observable, neither in the
approximation (5.87), nor at the exact level, since the “RG trajectory” λ̄(k) is
manifestly cutoff-scheme dependent.
In (5.87), the prefactor Φ12 (0) depends on the shape function R(0) , and the
exact RG equations that determine λ̄(k) together with all other couplings contain
further, likewise non-universal threshold functions. It is therefore structurally
impossible to relate λ̄(k) to an observable such as the Hubble rate by means of
the standard Einstein equation since, by itself, it “knows” nothing about our
choice of R(0) , and hence cannot compensate for the unphysical R(0) -dependence
of λ̄(k). This indicates that a qualitatively reliable and correct way to proceed
from running couplings (here λ̄ only) to observables (the Hubble rate) is actually
more involved and requires further running couplings.

(5) This example illustrates the general situation: Generically, the individual
running coupling constants have no direct physical relevance, they are no observ-
ables, in particular they are cutoff scheme and gauge fixing dependent. Quantities
with the status of an observable depend on very special combinations of several
such couplings in which the scheme and gauge dependencies mutually cancel (at
least in an exact treatment).
A typical example of such “universal” quantities are the critical exponents of
fixed points, the topic to which we turn next.

5.3.3 Gaussian and non-Gaussian Fixed Points

Now we return to the full system of RG equations (5.63) that determines both
g(k) and λ(k). We begin by searching for fixed points on the two-dimensional
g-λ-theory space. Their coordinates (g∗ , λ∗ ) would be common zeros of the two
beta functions: [ ]
βg (g∗ , λ∗ ) ≡ d − 2 + ηN (g∗ , λ∗ ) g∗ = 0,
(5.90)
βλ (g∗ , λ∗ ) = 0.
 5.3 Properties of the Einstein–Hilbert Flow 151
[
The first ] one of these conditions is satisfied if either g∗ = 0 or d − 2 +
ηN (g∗ , λ∗ ) = 0. Both possibilities are indeed realized in the Einstein–Hilbert
truncation. They lead to a Gaussian fixed point (GFP) and a non-Gaussian
fixed point (NGFP), which are, respectively, located at,
( GFP GFP )
g∗ , λ∗ = 0,
( NGFP NGFP ) (5.91)
g∗ , λ∗ ̸= 0.

Concerning the GFP, it is obvious from the beta function on the right-hand side
of (5.62) that g∗ = 0 together with λ∗ = 0 is indeed also a zero of βλ .

(1) Coordinates of the NGFP. When searching for non-trivial fixed points it
is convenient to start from the condition
( )
ηN g∗NGFP , λNGFP
∗ = 2 − d. (5.92)

Using (5.60) this relation can be solved for g∗ as a function of λ∗ :

( ) d−2
g∗NGFP λNGFP
∗ = . (5.93)
(d − 2)B2 (λNGFP
∗ ) − B1 (λNGFP
∗ )
We exploit (5.93) in order to eliminate the g-dependence of βλ at the non-trivial
fixed point. So it remains to solve
( ( ) NGFP )
βλ g∗NGFP λNGFP
∗ , λ∗ =0 (5.94)

for λNGFP
∗ . Due to its complicated structure this equation can only be investigated
numerically.
In d = 4, and with the sharp cutoff, the numerical analysis establishes the
existence of precisely one non-Gaussian fixed point. It is located at

λNGFP
∗ = 0.330, g∗NGFP = 0.403 (sc). (5.95)

An interesting characteristic property of fixed points is the value which the

anomalous dimension assumes there. In the case at hand, the GFP has ηN = 0,
while the NGFP displays a non-zero, in fact large,( integer anomalous
) dimension,
ηN = 2 − d . In particular, in four dimensions: ηN g∗NGFP , λNGFP
∗ = −2.

(2) Linearization. In order to determine the stability properties of the fixed

points and to compute their critical exponents we linearize the flow of the
couplings ui ≡ (u1 , u2 ) ≡ (λ, g), i = 1, 2, about ui∗ = (λ∗ , g∗ ):
[ ∂β ∂β ]
∑ ∂λ
λ
∂g
λ
i
∂t u = B j (u − u∗ ), B ≡ [B j ] = ∂βg ∂βg
i j j i
(5.96)
j=1,2 ∂λ ∂g .

All derivatives defining the matrix elements B ij are taken at u = u∗ .

152 Truncations of Single-Metric Type

The partial derivatives of the β functions are easily found by making use of
the following recursion formulas satisfied by the threshold functions:

d p d ep e p+1
Φ (w) = −pΦp+1
n (w), Φ (w) = −pΦ n (w). (5.97)
dw n dw n

These relations are easily proven by interchanging the derivative with respect to
w with the z-integration in the integral representations (5.51).
For any shape function R(0) , and at a generic point (λ, g) we thus obtain:
∂βλ ( g )
= − (2 − ηN ) + λ − (4π)1−d/2 d(d + 1)Φ e 1 (−2λ) ∂ηN
d/2
∂λ ( 2 ∂λ )
g
+ (4π) 1−d/2
4d(d + 1)Φd/2 (−2λ) − 2d(d + 1)ηN Φ̃d/2 (−2λ) ,
2 2
2
∂βλ ( g )
e 1 (−2λ) ∂ηN + 1 (4π)1−d/2 ×
= λ − (4π)1−d/2 d(d + 1)Φ d/2
∂g 2 ∂g 2
( ) (5.98)
× 2d(d + 1)Φd/2 (−2λ) − 8dΦd/2 (0) − d(d + 1)ηN Φ
1 1 e 1 (−2λ) ,
d/2

∂βg g2 ( )
= B1′ (λ) + ηN B2′ (λ) ,
∂λ 1 − gB2 (λ)
( )
∂βg gB2 (λ)
= d−2+ 2+ ηN .
∂g 1 − gB2 (λ)
Furthermore, the partial derivatives of ηN are given by
( )
∂ηN 1 B2 (λ)
= + ηN ,
∂g g 1 − gB2 (λ)
∂ηN g ( ) (5.99)
= B1′ (λ) + ηN B2′ (λ) .
∂λ 1 − gB2 (λ)
′
Here B1,2 (λ) denotes the derivatives of B1,2 (λ) with respect to λ:
1 ( )
B1′ (λ) = (4π)1−d/2 2d(d + 1)Φ2d/2−1 (−2λ) − 24d(d − 1)Φ3d/2 (−2λ) ,
3
1 ( )
′
B2 (λ) = − (4π)1−d/2 2d(d + 1)Φ e2 e3
d/2−1 (−2λ) − 24d(d − 1)Φd/2 (−2λ) .
6
(5.100)

Since these equations make no explicit use of the fixed point values λ∗ and g∗
they can be used to investigate the trivial and the non-trivial fixed point alike.
We will discuss them in turn.
As we mentioned in Chapter 3 the eigenvalues and right eigenvectors of B,
evaluated at the corresponding fixed point, determine its critical exponents and
scaling fields, respectively. Since, generically, B is not symmetric, its eigenvalues
are not real, and the eigenvectors are not orthogonal in general.
We define the stability coefficients θI , I = 1, 2, as the negative eigenvalues of B
satisfying the equation BV I = −θI V I , where V I are the right eigenvectors of B.
 5.3 Properties of the Einstein–Hilbert Flow 153

(3) Critical exponents (GFP). Substituting λ∗ = 0, g∗ = 0 into (5.96) the

stability matrix simplifies to
[ ]
−2 (4π)1−d/2 d(d − 3)Φ1d/2 (0)
BGFP = . (5.101)
0 d−2

Diagonalizing (5.101) leads to the following two real stability coefficients with
their corresponding right eigenvectors:

θ1 = 2 with V 1 = (1, 0)T ,

( )T (5.102)
θ2 = 2 − d with V 2 = (4π)1−d/2 (d − 3)Φ1d/2 (0), 1 .

Here, finally, we see that as we have anticipated the fixed point at g∗ = λ∗ = 0 is

indeed a “Gaussian” one according to our general definition in Chapter 3: The
two critical exponents θ1 = 2 and θ2 = 2 − d equal precisely the canonical ones,
i.e., the mass dimensions of the running couplings in question, λ̄(k) and G(k),
respectively.
Note also that while V 1 points exactly in the λ-direction, the second “scaling
field” V 2 is slightly rotated relative to the g-axis. Interestingly enough, this
rotation is absent in precisely three spacetime dimensions.
In d = 4, the positive (negative) sign of θ1 = +2 (θ2 = −2) indicates that the
GFP is repulsive (attractive)5 in the direction of V 1 (V 2 ).
This behavior is seen explicitly if we use the eigenvalues and -vectors in order
to write down the linearized solution:
M2 k d−2
λ(k) ≈ α1 + α2 (4π) 1−d/2
(d − 3)Φ1
d/2 (0) ,
k2 M d−2 (5.103)
k d−2
g(k) ≈ α2 d−2 .
M
Here α1 and α2 are constants of integration allowing one to adjust the solution
to given initial conditions. They are made dimensionless by separating off an
appropriate power of the mass scale M , which is thus not independent of α1
and α2 .
Switching to dimensionful coupling constants equations (5.103) read:

G(k) ≈ G0 ,
(5.104)
λ̄(k) ≈ λ̄0 + (4π)1−d/2 (d − 3)Φ1d/2 (0)G0 k d .

Here we chose M = mPl by setting α2 = 1, and we identified α1 = λ̄0 /m2Pl . We see

that both G(k) and λ̄(k) run toward constant and non-zero values G0 and λ̄0 in

5 Recall that in the nomenclature used here the terms repulsive and attractive are syn-
onymous to “IR repulsive” and “IR attractive,” respectively, indicating whether a certain
eigen-perturbation increases or decreases when k is lowered.
154 Truncations of Single-Metric Type

the limit k → 0, unless we set α1 = 0, which means λ̄0 = 0. There is no special rea-
son for α1 to be zero, however. So we see that λ̄0 depends on the free parameter
α1 and therefore on the trajectory considered. Hence, the Gaussian fixed point
does not determine the value of the cosmological constant in the infrared, λ̄0 .

(4) Critical exponents (NGFP). The computation of the critical exponents

for the NGFP is slightly more tedious. By using equations (5.64) and (5.60) for
the sharp cutoff we can compute the entries of the stability matrix analytically.
But since the fixed-point coordinates are the result of a numerical calculation,
the diagonalization of B must also be performed numerically.
For the NGFP one finds a complex conjugate pair of critical exponents

θ1 ≡ θ′ + iθ′′ and θ2 ≡ θ′ − iθ′′ , (5.105)

where, with the sharp cutoff [160],

θ′ = 1.941, θ′′ = 3.147 (sc). (5.106)

As the real part of θ1 and θ2 is positive, the NGFP is seen to have two relevant
eigendirections, meaning that it is repulsive (or equivalently UV attractive) in
both directions. As a result, the two-parameter Einstein–Hilbert truncation leads
to a two-dimensional UV critical hypersurface: ∆UV = dim SUV = 2.
The general solution to the linearized flow equations reads, with the RG time,
t ≡ ln(k/k0 )

( )T {
λ(k), g(k) = (λ∗ , g∗ )T + 2 [Re C cos(θ′′ t) + Im C sin(θ′′ t)] Re V
} ′ (5.107)
+ [Re C sin(θ′′ t) − Im C cos(θ′′ t)] Im V e−θ t .

Here C ≡ C1 = (C2 )∗ is an arbitrary complex number labeling the trajectory

under consideration, and V ≡ V 1 = (V 2 )∗ denotes the right-eigenvector of B with
eigenvalue −θ1 = −θ2∗ .
Due to the positivity of θ′ , all trajectories are seen to hit the fixed point as
t is sent to infinity. The non-vanishing imaginary part θ′′ has no impact on the
stability. However, it influences the shape of the trajectories: they spiral around,
and ultimately into, the fixed point for k → ∞.
Solving the full, non-linear flow equations [160] shows that the asymptotic
scaling region, where the linearization (5.107) is valid, extends from k = ∞ down
−1/2
to about k ≈ mPl with the Planck mass defined as mPl ≡ G0 . Here the Planck
scale mPl seems to play a role similar to the dynamically generated mass scale
ΛQCD of QCD for example: it marks the lower boundary of the asymptotic scaling
region.
 5.3 Properties of the Einstein–Hilbert Flow 155

5.3.4 The Phase Portrait

The information supplied by the linearizations about the two fixed points is
almost sufficient in order to draw the qualitative structure of the vector field
β = (βλ , βg ) and its integral curves, the RG trajectories.
Alternatively one can solve the coupled differential equations numerically,
which leads to the phase portrait shown in Figure 5.1. It was first obtained in
[160] using the sharp cutoff, which has the advantage that all threshold functions
can be evaluated analytically.
While in [160] the system of RG equations has been investigated in detail for
any dimensionality of spacetime, d, here we focus on d = 4 dimensions first.

(1) According to (5.64) and (5.60) the sharp cutoff (with s = 1, see Appendix E)
leads to the following explicit and comparatively simple-looking RG equations:
[ ]
gk 5
∂t λk = −(2 − ηN )λk − 5 ln(1 − 2λk ) − 2ζ(3) + ηN ,
π 2
∂t gk = (2 + ηN )gk where (5.108)
[ ]
2gk 18
ηN = − + 5 ln(1 − 2λk ) − ζ(2) + 6 .
6π + 5gk 1 − 2λk

Note that the beta functions in (5.108) have poles of first order as well as log-
arithmic singularities at λ = 21 . These divergences are the concrete incarnation,

0.75

0.5
Type IIa
0.25
Type Ia Type IIIa
λ
−0.2 −0.1 0.1 0.2 0.3 0.4 0.5

−0.25
Type IIIb
−0.5
Type Ib
−0.75

Figure 5.1. The phase portrait of the RG flow in the g-λ-plane. The arrows
point in the direction of decreasing k. (Taken from [160].)
156 Truncations of Single-Metric Type

for the sharp cutoff, of the divergences at w = −1 that plague Φpn (w) and Φe pn (w)
(0) 1
for any R . As a result, trajectories hitting the λ = 2 -line terminate there at a
non-zero value of k.

(2) Figure 5.1 displays a number of typical RG trajectories obtained by numer-

ically integrating (5.108). The RG flow is indeed seen to be dominated by two
fixed points: the GFP at g∗ = λ∗ = 0, and the NGFP with g∗ > 0 and λ∗ > 0.
There exist three types of trajectories emanating from the NGFP : trajectories
of Type Ia and Type IIIa which run toward negative and positive cosmological
constants, respectively, and a single trajectory of Type IIa (also known as the
separatrix ), which hits the GFP for k → 0.
The separatrix describes a “crossover” from the NGFP to the GFP, and
it separates the qualitatively different trajectories of Type Ia and Type IIIa,
respectively.6
The UV behavior of any of these trajectories is governed by the NGFP: for
k → ∞, they approach and asymptotically hit this point.

(3) Since it is the dimensionless coupling constants g(k) and λ(k) that approach
constant, non-zero values at the NGFP the corresponding dimensionful quantities
keep running for k → ∞ according to the power laws

G(k) = g∗NGFP k −2 , λ̄(k) = λNGFP

∗ k2 . (5.109)

Hence, for k → ∞, the dimensionful Newton constant vanishes, while the cosmo-
logical constant diverges. This confirms the behavior of Newton’s constant we
already saw in the first subtruncation.

(4) The onset of the asymptotic scaling law (5.109) is shown explicitly in Fig-
ure 5.2, which displays the k-dependence of both the dimensionless and the
dimensionful Newton and cosmological constant along the separatrix. The dimen-
sionful quantities, including k itself, are expressed in Planckian units, the Planck
mass being defined in terms of the value of G at k = 0, that is, mPl = G(k = 0)−1/2 .
In Figure 5.2 we can see that it is indeed near k ≈ mPl where the trajectory
“crosses over” from the vicinity of the GFP, where G, λ̄ are approximately inde-
pendent of k, to the asymptotic scaling regime governed by the NGFP where
instead g, λ = const.

(5) There is one feature of the RG flow that cannot be seen in Figures 5.1 and
5.2, namely that, according to the Einstein–Hilbert truncation, the trajectories
of Type IIIa terminate at a non-zero value of k at the moment they reach the
λ = 21 -line. The beta functions develop singularities there so that the differential

6 For a complete classification of all trajectories, including also “exotic” ones disconnected
from the classical regime, see [160].
 5.3 Properties of the Einstein–Hilbert Flow 157

Ğ (k)
1 ˘ (k)
λ
Type IIa 0.25
0.8
Type IIa
0.2
0.6
0.15
0.4 0.1

0.2 0.05
k/mP1
k/mP1 0.2 0.4 0.6 0.8 1
0.5 1 1.5 2 2.5 3 3.5 4 –0.05

log10 (Ğ (k))

˘ (k))
log10 (λ
1
0 5
k0 2.5
–1
0 k2
–2
–2.5
–3 k–2 –5
–4 –7.5
–5 –10 k4
Type IIa Type IIa
k/mP1 –12.5
k/mP1
0.01 0.1 1 10 100 0.01 0.1 1 10 100

log10 (g(k)) log10 ((λ(k))

1 1
0 0
−1 k0 −1 k0
−2 −2
−3 −3
−4 k2 −4 k2
−5 −5
−6 Type IIa −6 Type IIa
k/mPl k/mPl
0.01 0.1 1 10 100 0.01 0.1 1 10 100

Figure 5.2. Scaling laws for the crossover trajectory IIa, the separatrix. The
diagrams show the k-dependence of both the dimensionless couplings g and λ
and the dimensionful ones expressed in Planck units, Ğ = G m2Pl , λ̆ = λ̄/m2Pl ,
−1/2
mPl ≡ G0 . For k/mPl ≪ 0.1 and k/mPl ≫ 10 the scale dependence is seen to
be governed by the scaling laws of the GFP and the NGFP, respectively. The
transition between these power laws occurs at k ≈ mPl . (Taken from [160].)

equations cannot be integrated beyond this point. Therefore, the λ = 21 -line can
be seen as a natural boundary of the g-λ theory space.

(6) These general properties of the flow (are confirmed

) by the diagrams display-
ing the anomalous dimension ηN (k) ≡ ηN λ(k), g(k) evaluated along the sample
trajectories of Type Ia, Type IIa, and Type IIIa shown in Figure 5.3.
In the region governed by the non-trivial fixed point it is seen that ηN ≈ −2,
∗ ∗
while ηN ≈ 0 in the infrared region. Recalling that ηN = −2 and ηN = 0 at the
NGFP and the GFP in d = 4, respectively, this is exactly what we expect.
158 Truncations of Single-Metric Type

g
0.5 ηN (k)
0
0.4
−1
0.3
−2
0.2
−3
0.1 Type IIIa −4
Type Ia Type Ia
Type IIa k/mPl
λ
−0.4 −0.2 0.2 0.4 0.01 0.1 1 10 100

ηN (k) ηN(k)
0 0

−1 −1

−2 −2

−3 −3

−4 −4
Type IIa Type IIIa
k/mPl k/M
0.01 0.1 1 10 100 0.01 0.1 1 10 100

Figure 5.3. Anomalous dimension ηN evaluated along the Type Ia, IIa, and
IIIa trajectories shown in the first diagram. (Taken from [160].)

The termination of the Type IIIa trajectory is accompanied by a steep decrease

of ηN , caused by the divergence of B1 (λ)sc at constant B2 (λ)sc for λ approaching
the boundary line λ = 21 . (This can easily be checked from (5.64).)
The “IR divergence” of ηN for the trajectories of Type IIIa suggests that the
Einstein–Hilbert truncation may not be sufficient to describe the RG flow close
to the boundary line λ = 12 .

5.3.5 Close Relatives of the EH Truncation

Before turning to a detailed assessment of the reliability of the Einstein–Hilbert
(EH) truncation and the implications for Asymptotic Safety let us first introduce
a number of immediate generalizations it gave rise to. To put the EH trunca-
tion into the broader context, we will analyze it further together with those
generalizations.
The natural next step beyond the Einstein–Hilbert truncation consists in gen-
eralizing the functional Γ̄k [g], while keeping the gauge-fixing and ghost sector
classical, as in (5.1). During the RG evolution the flow generates all possible
diffeomorphism-invariant terms in Γ̄k [g] which one can construct from gµν . Typ-
ical invariants contain strings of curvature tensors, and covariant derivatives
acting upon them, with any number of tensors and derivatives; see Appendix
B for examples and [161] for a classification of such terms.
 5.3 Properties of the Einstein–Hilbert Flow 159

Let us now list a number of ansätze for Γ̄k [g] that have been investigated in
the literature, always accompanied by single-metric type gauge-fixing and ghost
terms.
Considering those ansätze the reader should always keep in mind that the
canonical mass dimension of a field monomial is not a reliable measure for its
importance, as this would be the case in perturbation theory.
We are aiming at describing a non-perturbative fixed point with potentially
large anomalous dimensions. Thus every new truncation ansatz usually amounts
to a “step in the dark” on the largely uncharted theory space.

(1) The R2 -truncation. The first truncation of this class that was worked out
b k as
completely [162, 163] is the R2 truncation defined by (5.10) with the same Γ
2
before, and the (curvature) action:

∫ { }
√ 1 [ ]
Γ̄k [g] = d
d x g −R + 2λ̄k + β̄k R .
2
(5.110)
16πGk

In this case the truncated theory space is three-dimensional. Its natural dimen-
sionless coordinates are (g, λ, β), where βk ≡ k 4−d β̄k , while g and λ are the usual
dimensionless Newton and cosmological constant. Even though (5.110) contains
only one additional invariant, the derivation of the corresponding RG equations
is far more complicated than in the Einstein–Hilbert case. As the beta functions
for the three couplings would fill many pages we cannot reproduce them here but
summarize the results obtained with them below.

(2) The R2 + C 2 -truncation. A natural extension of the R2 truncation

consists of retaining all gravitational four-derivative terms in the truncation
subspace:
∫ {
√ 1 [ ] ωk 2
Γ̄k [g] = d4 x g −R + 2λ̄k − R
16πGk 3σk
} (5.111)
1 θk
+ Cµνρσ C µνρσ + E .
2σk σk

Here Cµνρσ C µνρσ denotes the square of the (conformal) Weyl tensor, and
E = Cµνρσ C µνρσ − 2Rµν Rµν + 23 R2 is the integrand of the (topological) Gauss–
Bonnet term in four dimensions. Using this ansatz and the FRGE, the perturba-
tive one-loop beta functions for higher-derivative gravity have been reanalyzed
in [119, 164–167], while the first non-perturbative results have been obtained in
[168–170].
The key ingredient in the projection of the flow onto the space of actions
(5.111) was the generalization of the background metric ḡ from the maximally
symmetric spheres S d , that had been employed in the R2 -truncation, to a generic
160 Truncations of Single-Metric Type

Einstein space. Inserting an S d metric, the RG equations collapse into a single

flow equation for βk = − 3σ
ωk
k
θk
+ 6σ k
. While working with a generic Einstein metric
allows one to find the beta functions for two independent combinations of the
four-derivative couplings:

ωk θk 1 θk
βk = − + , γk = + . (5.112)
3σk 6σk 2σk σk

Ideally, the new class of backgrounds should be general enough to disentangle

the coefficients multiplying R2 , C 2 , and E and, at the same time, simple enough
to avoid the appearance of non-minimal higher-derivative differential operators
inside the functional trace, which would be difficult to deal with. While the S d
backgrounds are insufficient in the former respect, a generic compact Einstein
background (without Killing or conformal Killing vectors and without bound-
ary for simplicity), satisfying R̄µν = R̄4 ḡµν , meets both criteria and allows us to
determine the non-perturbative beta functions of the linear combinations (5.112).
And surprisingly enough, projecting the flow equation resulting from the ansatz
(5.111) with a generic Einstein background, the differential operators appear-
ing on its right-hand side are found to organize themselves into second order
differential operators of the Lichnerowicz form,

∆2L ϕµν ≡ −D̄2 ϕµν − 2R̄µανβ ϕαβ ,

∆1L ϕµ ≡ −D̄2 ϕµ − R̄µν ϕν , (5.113)
∆0L ϕ ≡ −D̄ ϕ, 2

which, moreover, commute with all the other curvature terms inside the trace.
This feature makes the traces amenable to standard heat kernel techniques for
minimal second-order differential operators.

(3) Polynomial f (R) truncations. A different direction to go from the EH

ansatz is to allow for arbitrary powers of the curvature scalar, but instead discard
all other types of invariants in order to keep the calculational problem manage-
able. The truncation ansatz for Γ̄k [g] contains an arbitrary function of R then:

∫
√
Γ̄k [g] = dd x g fk (R). (5.114)

Later on we will allow fk ( · ) to be a running function of a rather general type.

Here we specialize to polynomials, which may have any degree, N , though:

∑
N ( )n
R
fk (R) = un (k) k d . (5.115)
n=0
k2
 5.3 Properties of the Einstein–Hilbert Flow 161

In this case the truncated theory space is (N + 1)-dimensional and can be coor-
dinatized by the already dimensionless coefficients (u0 , u1 , · · · , uN ). Clearly, if
N = 1 or N = 2 we are back to the EH and R2 -truncations, respectively.

(4) General gauge-fixing parameter α. There are also results concerning

the gauge-fixing term. Even if one uses the ansatz (5.10) for Γk [g, ḡ] in which
the gauge-fixing term has the classical (or more appropriately, bare) structure
one should in principle still treat its prefactor as a running coupling constant:
α ≡ αk .
While the beta function of α is not easily computed from the FRGE, there is
a simple argument which allows us to bypass this calculation.
In non-perturbative Yang-Mills theory, and in perturbative quantum gravity,
α = αk = 0 is known to be a fixed point of the α evolution. The following reason-
ing suggests that the same is true within the non-perturbative FRGE approach to
gravity. In the formal functional integral the limit α → 0 corresponds to a sharp
implementation of the gauge-fixing condition,
∫ i.e., exp(−Sgf ) becomes propor-
tional to δ[Fµ ]. The domain of the Db hµν integration consists of those bhµν s
which satisfy the gauge-fixing condition exactly, Fµ = 0. Adding the IR cutoff at
k amounts to suppressing some of the b hµν modes while retaining the others. But
since all of them satisfy Fµ = 0, a variation of k cannot change the domain of the
b
hµν integration. The delta functional δ[Fµ ] continues to be present for any value
of k if it was there originally. As a consequence, α vanishes for all k, i.e., α = 0
is a fixed point of the α evolution [171]. For gravity, this fixed point has been
confirmed in [155].
As a consequence, it is possible to mimic the dynamical treatment of a running
α by setting the gauge-fixing parameter to the constant value α = 0.
The calculation for α = 0 is more involved than at α = 1, but for the Einstein–
Hilbert truncation the α-dependence of βg and βλ , for arbitrary constant α, has
been found in [172, 173]. The (curvature)2 -truncations could be analyzed only
in the simpler α = 1 gauge, but the results from the Einstein–Hilbert truncation
suggest that the fixed-point quantities of interest do not change much between
α = 0 and α = 1 [163, 173].

(5) Including scale-dependent couplings in the ghost sector. General-

izations of the classical ghost action including a scale-dependent wave-function
renormalization or a scale-dependent ghost-curvature coupling have been con-
sidered in [174, 175] and [176], respectively. Both extensions confirm the phase
diagram obtained from the single-metric Einstein–Hilbert truncation, justifying
the approximation of a classical ghost sector a posteriori.

5.3.6 Testing the Reliability of Truncations

Since the method of the truncated theory spaces does not come with a natu-
ral small expansion parameter, and since in general the relative importance of
162 Truncations of Single-Metric Type

the various terms in the ansatz for Γk is not simply related to their canoni-
cal dimension, a crucial part of the method consists of testing the qualitative
reliability, and the quantitative precision that can be achieved with a given
truncation.
Generally speaking we can distinguish two types of tests, namely those that
apply to individual truncations, and those that address several truncations
jointly.

(1) The first type of tests examines the reliability or “robustness” of the predic-
tions obtained within a given truncation under changes of the various unphysical
elements in the computational setup, the cutoff operator being the prime example.
The idea behind this method is quite simple.
Let us assume we know from general principles that a certain quantity Q
that can be computed from the RG flow in some way is universal, i.e., cutoff
scheme independent. Hence, if we change the cutoff operator from Rk to Rk +
δRk , the theory’s exact answer for this quantity will not change. Now, typically,
approximations and truncations destroy this property of scheme independence,
i.e., approximate computations of Q are Rk -dependent, and in fact, they are
even more so the poorer the quality of the underlying approximation.
This observation provides us with a necessary condition for the reliability of
a truncation (or any sort of approximation), which can be checked straightfor-
wardly: If a quantity that is known to be universal at the exact level changes
under a deformation of Rk by a certain amount, then typical errors of the
pertinent calculation are at least of the order of this amount, or larger even.
An example of quantities known to be cutoff scheme-independent when deter-
mined exactly are the critical exponents of a fixed point. In Section 3.2 we
demonstrated already that the θα ’s are invariant under general coordinate trans-
formations on theory space, and as a consequence, invariant under changes
of Rk .
At a less general level, for Einstein–Hilbert-like truncations, the dimensionless
(d/2−1)
combination g∗ λ∗ is also approximately scheme independent and often has
been studied in its dependence on Rk .

(2) In practice one can modify Rk both by changing its matrix structure in field
space and the shape function R(0) it involves. As for the latter, typical choices
of R(0) that have been used in quantum gravity include:
(i) The exponential cutoff [14]:
z
R(0) (z)exp = ≡ R(0) (z; 1)exp . (5.116)
ez −1
(ii) The one-parameter family of generalized exponential cutoffs [177, 14]:
sz
R(0) (z; s)exp = (s > 0). (5.117)
esz − 1
 5.3 Properties of the Einstein–Hilbert Flow 163

(iii) The one-parameter family of cutoffs with compact support [163, 173], with
shape parameter b ∈ [0, 32 ):


 if z ≤ b,
1 [ ( ) [( )−1 ]]
3 −1
R (z; b) = exp z − 2
(0) cp
exp b − z if b < z < 32 , (5.118)


0 if z ≥ 2 .
3

(iv) The sharp cutoff [160, 178, 179]:

b Θ(1 − z),
R(0) (z)sc = R b→∞
R (5.119)

and its one-parameter family of generalizations R(0) (z; s)sc , described in

Appendix E.
(v) The optimized cutoff [180]:

R(0) (z)opt = (1 − z)Θ(1 − z). (5.120)

For checking the degree of scheme independence exhibited by a certain quantity

one may perform both “small” changes of R(0) by varying the free parameters
in one of the families, as well as “large” changes by switching from one type to
another. Among the families of cutoffs the generalized sharp cutoff R(0) (z; s)sc
is particularly convenient since it has both a tunable free parameter and allows
for an analytic evaluation of the threshold functions Φ and Φe [160].

(3) Another test that can be applied to truncations individually suggests itself
when the truncated Γk contains a certain coupling constant, say u(k), which is
scale dependent at the exact level, but whose scale dependence is neglected as
part of the approximations that define the truncation. If we set u(k) = const ≡
u(0) with an arbitrary constant u(0) , the test consists in checking how the quanti-
ties of interest vary with the value of u(0) chosen in the calculation. Then, again,
the “error bar” to be attached to those quantities should be at least of the order
of their variation with u(0) , unless further information is available.
The prototype of a coupling of this kind is the running gauge-fixing constant
α(k). Treating it as k-independent amounts to an approximation on top of a
truncation. Letting α(k) = const ≡ α it is instructive to see how the answers pro-
vided by the truncation depend on the constant α. In this way one gets a first
impression of the precision that can be expected.

(4) The tests discussed up to now refer to individual truncations. While they
are comparatively easy to apply, it is clear that further justification of a given
truncation, or a hierarchy of nested truncations, must come from a stepwise
extension of the truncated theory space. Including further invariants into Γk one
should observe a certain degree of stabilization or “convergence” of the physical
predictions.
164 Truncations of Single-Metric Type

Calculations of this sort, while straightforward conceptually, are extremely

hard in practice. Usually every enlargement of the truncated gravitational theory
space entails new computational challenges which are often quite formidable and
require developing completely new strategies and calculational techniques. The
problems are mostly due to the functional traces in the FRGE which must be
projected analytically on the chosen Ttrunc and whose evaluation is notoriously
difficult.
In the next subsection we apply this type of reliability test to the hierarchy of
nested polynomial f (R)-truncations (5.115) whose members are labeled by the
degree of the polynomial, N .

5.3.7 Evidence for Asymptotic Safety

Next, we list and compare the results obtained with the Einstein–Hilbert trun-
cation and its various relatives, detailing in particular how they score in the
reliability tests, and collecting the pieces of evidence which they contribute in
support of the Asymptotic Safety hypothesis.

(1) Evidence from the EH truncation. The Einstein–Hilbert truncation

predicts the existence of a NGFP with exactly the properties needed for the
Asymptotic Safety construction. Of course, the immediate question is whether
this NGFP is the projection of a true fixed point in the exact theory or whether
it is merely the artifact of an insufficient approximation.
We start by summarizing the properties of the NGFP in the Einstein–Hilbert
truncation itself. Except for point (v) below, the results refer to d = 4.
All findings listed are independent pieces of evidence supporting the conjecture
that QEG is indeed asymptotically safe in four dimensions.

(i) Universal existence (EH): The non-Gaussian fixed point exists for all
cutoff schemes and shape functions implemented to date. It seems impossible
to find an admissible cutoff which destroys the fixed point in d = 4. This
result is highly non-trivial since in higher dimensions (d & 5) the existence
of the NGFP depends on the cutoff chosen [160].
(ii) Positive Newton constant (EH): While the position of the fixed point is
scheme dependent, all cutoffs yield positive values of g∗ and λ∗ . A negative g∗
might have been problematic for stability reasons, but there is no mechanism
in the flow equation which would exclude it on general grounds.
(iii) Stability (EH): For any cutoff employed, the NGFP is found to be UV
attractive in both directions of the λ-g-plane. Linearizing the flow equation
we always obtain a pair of complex conjugate critical exponents, θ1 = θ2∗ ,
and they have a positive real part θ′ > 0 for all cutoffs.
(iv) Scheme and gauge dependence (EH): Analyzing the cutoff-scheme
dependence of θ′ , θ′′ , and g∗ λ∗ as a measure for the reliability of the
 5.3 Properties of the Einstein–Hilbert Flow 165

truncation, the critical exponents were found to be reasonably constant

within about a factor of 2 within the one-parameter family of exponential
cutoffs (5.117). For α = 1 and α = 0, for instance, they assume values in
the ranges 1.4 . θ′ . 1.8, 2.3 . θ′′ . 4 and 1.7 . θ′ . 2.1, 2.5 . θ′′ . 5, respec-
tively. The universality properties of the product g∗ λ∗ are even more
impressive. Despite the rather strong scheme dependence of g∗ and λ∗ sep-
arately, their product has almost no visible s-dependence for not too small
values of s. Its value is
{
0.12 for α = 1,
g∗ λ∗ ≈ (5.121)
0.14 for α = 0.
The difference between the “correct” (i.e., fixed point) value of the gauge
parameter, α = 0, and the technically more convenient, α = 1, are at the level
of about 10 to 20 percent. Modifying the coarse-graining operator appearing
in the argument of Rk gives rise to variations of a similar magnitude [181].
(v) Lower and higher dimensions (EH): The beta functions implied by the
FRGE are continuous functions of the spacetime dimensionality and it is
instructive to analyze them for d ̸= 4.
In [14] it has been shown that for d = 2 + ε, |ε| ≪ 1, the FRGE reproduces
3
Weinberg’s [10] fixed point for Newton’s constant, g∗ = 38 ε, and in addition
supplies a corresponding fixed-point value for the cosmological constant,
λ∗ = − 383 1
Φ1 (0)ε.
The g∗ value from the FRGE, g∗ = b−1 ε with b = 23 · 19, agrees with the
perturbative result for pure gravity, b2 in (3.76).
The latter had been found using the same linear background split which
also underlies the FRGE described here. Furthermore, it turned out that the
analogous FRGE with an exponential background split also reproduces the
corresponding perturbative coefficient, b3 in (3.76), namely b = 23 · 25 [145].
After the FRGE-based beta function for the Gibbons-Hawking surface
term had been derived [182] it also turned out [183, 117] that the non-
perturbative flow equation even reproduces the perturbative answer, b1 in
(3.76), that pertains to the Newton constant in the surface term (and a
linear background split).
In each of the three calculations the coefficient b is perfectly independent
of R(0) , i.e., a universal quantity, in this sense.
In the sequel we return to the linear split and the “bulk” Newton constant.
For arbitrary d, and a generic cutoff, the RG flow is found to be quanti-
tatively similar to the four-dimensional one for all d smaller than a certain
critical dimension dcrit , above which the existence or non-existence of the
NGFP becomes cutoff dependent. The critical dimension is scheme depen-
dent, but for any admissible cutoff it lies well above d = 4. As d approaches
dcrit from below, the scheme dependence of the universal quantities increases
drastically, indicating that the EH-truncation becomes insufficient near dcrit .
166 Truncations of Single-Metric Type

l* q¢
0.5 5
sc s = 1
0.4 4 4 sc s = 30
0.3 3.5 Exp s = 1
3.5 4 3
4
0.2
3.5 4
3 3.5 sc s = 1 2 22.5
0.1 sc s = 30 2.5 3 3.5 3.5
2.5 3 Exp s = 1 4
2 1 4
g* 3 3.5
2.5 0.1 0.2 0.3 0.4 0.5
–0.1 q¢¢
1 2 3 4 5

q¢
17.5
6
15
12.5
10
5.5 6
7.5
5 5
6 5.5
4.5
2.5 2 4 5
4.5
3 4 4 q¢¢
2.5 5 7.5 10 12.5 15 17.5

Figure 5.4. Comparison of λ∗ , g∗ , θ′ , and θ′′ for different cutoff functions in

dependence on the dimension d. Two versions of the sharp cutoff (sc) and the
exponential cutoff with s = 1 (Exp) have been employed. The upper diagrams
show that for 2 + ε ≤ d ≤ 4 the cutoff-scheme dependence of the results is rather
small. The lower diagram shows that increasing d beyond about 5 leads to a
significant difference in the results for θ′ , θ′′ obtained with different cutoff
functions. (Taken from [160].)

In Figure 5.4 we show the d-dependence of g∗ , λ∗ , θ′ , and θ′′ for two

versions of the sharp cutoff (with s = 1 and s = 30, respectively) and for the
exponential cutoff with s = 1. For 2 + ε ≤ d ≤ 4 the scheme dependence of
the critical exponents is seen to be rather weak; it becomes appreciable only
near d ≈ 6 [160].
Figure 5.4 suggests that the Einstein–Hilbert truncation in d = 4 performs
almost as well as near d = 2. (Its validity can be slightly extended toward
larger dimensionalities by optimizing the shape function [180, 184].)

(2) Evidence from the R2 truncation. The ultimate justification of a given

truncation consists in proving that if one adds further terms to the ansatz, its
predictions do not change much. The first step toward testing the robustness of
the Einstein–Hilbert truncation against the inclusion of other invariants has been
taken in [162, 163] where the beta functions for the three generalized couplings
g, λ and β entering into the R2 -truncation of (5.110) have been derived and
analyzed. Concerning Asymptotic Safety, the main results are as follows.

(i) Position of the NGFP (R2 ): Also on the basis of the generalized trunca-
tion (5.110) the NGFP is found to exist for all admissible cutoffs. Figure 5.5
 5.3 Properties of the Einstein–Hilbert Flow 167

0.8
g*
0.6

0.4

λ∗
0.2
g*λ∗
s
1.5 2 2.5 3 3.5 4 4.5 5
(a)

β*
0.01

0.008

0.006

0.004

0.002

s
(b) 5 10 15 20 25 30

Figure 5.5. (a) g∗ , λ∗ , and g∗ λ∗ as functions of s for 1 ≤ s ≤ 5, and (b) β∗ as

a function of s for 1 ≤ s ≤ 30, using the family of exponential shape functions.
(Taken from [162].)

shows its coordinates (λ∗ , g∗ , β∗ ) for the family of exponential shape func-
tions (5.117). For every shape parameter s, the values of λ∗ and g∗ are
almost the same as in the Einstein–Hilbert truncation. In particular, the
product g∗ λ∗ is constant with a very high accuracy. For s = 1, for instance,
one obtains (λ∗ , g∗ ) = (0.359, 0.272) from the Einstein–Hilbert truncation
and (λ∗ , g∗ , β∗ ) = (0.330, 0.292, 0.005) from the generalized truncation.
It is remarkable that β∗ is always significantly smaller than λ∗ and g∗ ,
thus “disturbing” the earlier analysis on the basis of the EH-truncation only
very little. Within the limited precision of our calculation this means that in
the three-dimensional parameter space the fixed point practically lies on the
λ-g-plane with β = 0, i.e., on the parameter space of the Einstein–Hilbert
subtruncation.
(ii) Eigenvalues and vectors (R2 ): The NGFP of the R2 -truncation proves
to be UV attractive in any of the three directions of the (λ, g, β)-space for
all cutoffs used. The linearized flow in its vicinity is always governed by
168 Truncations of Single-Metric Type

a pair of complex conjugate critical exponents θ1 = θ′ + iθ′′ = θ2∗ with θ′ > 0

and a single real, positive critical exponent θ3 > 0. For the exponential shape
function with s = 1, for instance, we find θ′ = 2.15, θ′′ = 3.79, θ3 = 28.8. The
first two eigenperturbations again yield spiral type trajectories, while the
third one amounts to a straight line.7

For any cutoff, the numerical results display a number of robust features that
are quite remarkable. They all indicate that, close to the NGFP, the RG flow is
rather well approximated by the pure Einstein–Hilbert subtruncation:

(a) The eigenvectors associated with the spiraling directions span a plane which
virtually coincides with the g-λ-subspace at β = 0, i.e., with the parameter space
of the Einstein–Hilbert truncation. As a consequence, the corresponding normal
modes are essentially the same trajectories as the “old” normal modes already
found without the R2 -term. The corresponding θ′ - and θ′′ values also coincide
within the scheme dependence.

(b) The additional eigenvalue θ3 introduced by the R2 term is significantly larger

than θ′ . When a trajectory approaches the fixed point from below (t → ∞), the
“old” normal modes are proportional to exp(−θ′ t), but the new one is propor-
tional to exp(−θ3 t), so that it decays much quicker. For every trajectory running
into the fixed point we therefore find that once t is sufficiently large the tra-
jectory practically lies entirely in the β = 0-plane. Due to the large value of θ3 ,
the new scaling field is highly “relevant.” However, when we start at the fixed
point (t = ∞) and lower t it is only at the low energy scale k ≈ mPl (t ≈ 0) that
exp(−θ3 t) reaches unity, and only then, i.e., far away from the fixed point, the
new scaling field starts growing rapidly.
Thus it seems that close to the fixed point the RG flow is essentially two-
dimensional, and that the two-dimensional flow is well approximated by the RG
equations of the EH truncation.
In Figure 5.6 we show a typical trajectory which has all three normal modes
excited with equal strength. All the way down from k = ∞ to about k = mPl it
is indeed confined to a very thin box surrounding the β = 0-plane.

(iii) Scheme dependence (R2 ): The scheme dependence of the critical expo-
nents and of the product g∗ λ∗ turns out to be of the same order of magnitude
as in the case of the Einstein–Hilbert truncation. Figure 5.7 shows the cut-
off dependence of the critical exponents, using the family of shape functions
(5.117). For the cutoffs employed, θ′ and θ′′ assume values in the ranges
2.1 . θ′ . 3.4 and 3.1 . θ′′ . 4.3, respectively. While the scheme dependence
of θ′′ is weaker than in the case of the Einstein–Hilbert truncation one

7 The value obtained for θ3 obtained within the R2 -truncation is rather large. It stabilizes
quickly to θ3 ≈ 1.5 once higher powers of R are included; see Table 5.3.
 5.3 Properties of the Einstein–Hilbert Flow 169

0.1 g
0.15 g
t=1 0.2
0.25
0.3
Im V

0.35 –V3

t=∞ λ
0.3 Re V λ

0.02
0 β β
–0.02
(a) (b)

Figure 5.6. Trajectory of the linearized flow equation obtained from the R2 -
truncation for 1 ≤ t = ln( kk0 ) < ∞. In (b) we depict the eigendirections and the
“box” to which the trajectory is confined. (Taken from [162].)

5
θ″
4
θ′
3

1
s
(a) 5 10 15 20 25 30

θ3
30

25

20

15

10

s
(b) 5 10 15 20 25 30

Figure 5.7. (a) θ′ = Re θ1 and θ′′ = Im θ1 , and (b) θ3 as functions of s, using

the family of exponential shape functions. (Taken from [162].)
170 Truncations of Single-Metric Type

finds that it is slightly larger for θ′ . The exponent θ3 suffers from relatively
strong variations as the cutoff is changed, 8.4 . θ3 . 28.8, but it is always
significantly larger than θ′ .
The product g∗ λ∗ again exhibits an extremely weak scheme dependence.
Figure 5.5(a) displays g∗ λ∗ as a function of s. It is impressive to see how the
cutoff dependences of g∗ and λ∗ cancel almost perfectly. Figure 5.5(a) sug-
gests the universal value g∗ λ∗ ≈ 0.14. Comparing this value to those obtained
from the Einstein–Hilbert truncation we find that it differs slightly from the
one based on the same gauge α = 1. The deviation is of the same size as the
difference between the α = 0- and the α = 1-results of the Einstein–Hilbert
truncation.

(iv) Dimensionality of SUV (R2 ): Since the real parts of θ1 , θ2 , and θ3 are
positive throughout, all three scaling fields are relevant, which suggests that
∆UV ≡ dim SUV is at least equal to 3.
According to the canonical dimensional analysis, the (curvature)n invari-
ants in four spacetime dimensions are classically marginal for n = 2 and
irrelevant for n > 2. So, if the classical scaling dimensions were a reliable
guideline near the NGFP even, we would expect no further relevant direc-
tions to arise when we generalize the R2 -truncation by adding terms of
canonical dimension 6 or higher to it. However, there could be large non-
classical contributions to the scaling dimensions so that there might be
relevant operators perhaps even beyond n = 2. Within the R2 -truncation
it is therefore not yet possible to determine their number, ∆UV , at least not
in d = 4.
But nevertheless, as Weinberg pointed out already [10], it is hardly con-
ceivable that the quantum effects are powerful enough to change the signs
of infinitely many, arbitrarily large (negative) classical scaling dimensions.
Therefore, we expect ∆UV to be a finite number at least.
If correct, this expectation will manifest itself as follows. Let us consider
a sequence of increasingly refined truncations with a growing dimension-
ality of the truncated theory space, dtrunc ≡ dim Ttrunc . Then, as long as
the truncation is still comparatively simple, we expect to find only rele-
vant directions, hence ∆UV = dtrunc , but from a certain point onward ∆UV
should stabilize and no longer grow with increasing dtrunc . At this point,
once ∆UV < dtrunc , the truncated flow is precise enough to describe a “low”
dimensional hypersurface embedded in a “high” dimensional theory space.
This allows us to make dtrunc − ∆UV > 0 predictions after having measured
only ∆UV couplings.
The first confirmation of this picture, the demonstration that enlarging
the truncation ultimately adds only irrelevant scaling fields, came from the
R2 -calculation in d = 2 + ε dimensions [163]. There the dimensional count is
shifted downward by two units. In this case we indeed find that already the
 5.3 Properties of the Einstein–Hilbert Flow 171

Table 5.1 Fixed point coordinates and critical exponents of the R2 truncation
in 2 + ε dimensions. Only the respective leading order in ε of these quantities is
shown. The negative value θ3 < 0 implies that SUV is an (only!) two-dimensional
surface in the three-dimensional theory space. (From [163].)

s λ∗ g∗ β∗ θ1 θ2 θ3

1 −0.131ε 0.087ε −0.083 2 0.963ε −1.968

5 −0.055ε 0.092ε −0.312 2 0.955ε −1.955
10 −0.035ε 0.095ε −0.592 2 0.955ε −1.956

third scaling field is irrelevant, for any cutoff employed: θ3 < 0. In Table 5.1
we present the corresponding numerical results for selected values of the
shape parameter s and in the leading order of ε.
For all cutoffs used we obtain three real critical exponents; the first two
are positive and the third is negative. This suggests that in d = 2 + ε the
dimensionality of SUV is likely to be as small as ∆UV = 2. Hence, measuring
only two free parameters, the renormalized Newton constant G0 and the
renormalized cosmological constant λ̄0 , for instance, we are able to to predict
the third coupling.

(3) Evidence from polynomial f (R) truncations. Next, we summarize the

main results obtained with the (N + 1)-parameter truncations of (5.114) with
(5.115).

(i) Position of the NGFP (f (R), d = 4): The generalized RG equations,

too, give rise to a NGFP with g∗ > 0, λ∗ > 0, whose N -dependent position
is shown in Table 5.2. The product g∗ λ∗ = u0 /(32π(u1 )2 ) displayed in the
last column is remarkably constant. It is in excellent agreement with the
Einstein–Hilbert truncation (5.121), and the R2 -truncation, Figure 5.5(a).
Only the value obtained in the case N = 2 shows a mild deviation from the
R2 -computations in [163], which can, most probably, be attributed to the
use of a different gauge-fixing procedure, cutoff shape function, and ansatz
bk .
for Γ
(ii) Dimensionality of SUV (f (R), d = 4): The critical exponents resulting
from the stability analysis of the NGFP in the polynomial f (R)-truncation
are summarized in Table 5.3. We see that only three of the eigendirections
associated to the NGFP are relevant, i.e., UV attractive. Including higher-
derivative terms Rn , n ≥ 3 in the truncation creates irrelevant directions
only. Thus, consistent with the R2 -truncation, we find ∆UV = 3, i.e., the
UV-critical surface associated with the fixed point is a three-dimensional
submanifold in the (N + 1)-dimensional theory space. Its dimensionality
∆UV is stable with respect to increasing N [185, 186].
172 Truncations of Single-Metric Type

Table 5.2 Location of the NGFP in the polynomial f (R)-truncation (5.115) of

degree N . The table illustrates the stabilization of the fixed point coordinates as
N increases. (From [185].)

N u∗0 u∗1 u∗2 u∗3 u∗4 u∗5 u∗6 g ∗ λ∗

1 0.00523 −0.0202 0.127

2 0.00333 −0.0125 0.00149 0.211
3 0.00518 −0.0196 0.00070 −0.0104 0.134
4 0.00505 −0.0206 0.00026 −0.0120 −0.0101 0.118
5 0.00506 −0.0206 0.00023 −0.0105 −0.0096 −0.00455 0.119
6 0.00504 −0.0208 0.00012 −0.0110 −0.0109 −0.00473 0.00238 0.116

Table 5.3 Stabilization of the critical exponents of the NGFP for increasing
dimension N + 1 of the truncation subspace. The first two critical exponents
form a complex pair: θ′ ± iθ′′ . (From [185].)

N θ′ θ′′ θ2 θ3 θ4 θ5 θ6

1 2.38 −2.17
2 1.26 −2.44 27.0
3 2.67 −2.26 2.07 −4.42
4 2.83 −2.42 1.54 −4.28 −5.09
5 2.57 −2.67 1.73 −4.40 −3.97 + 4.57i −3.97 − 4.57i
6 2.39 −2.38 1.51 −4.16 −4.67 + 6.08i −4.67 − 6.08i −8.67

A specific RG trajectory tracing out this surface is determined by fixing the

three relevant couplings. These describe in which direction tangent to SUV
the trajectory moves away from the fixed point. All remaining couplings, the
irrelevant ones, are predictions from Asymptotic Safety.
Subsequently, these results have been extended to N = 8 [187] and more
recently to N = 35 [188], providing even stronger evidence for the robustness
of the RG flow under the inclusion of further invariants constructed from the
Ricci scalar.

(4) Evidence from the R2 + C 2 -truncation in d = 4. The results obtained

on the basis of the ansatz (5.111) can be summarized as follows.

(i) Existence of the NGFP (R2 + C 2 ): The R2 + C 2 -truncation predicts a

NGFP with positive Newton and cosmological constant at

g∗ = 1.960, λ∗ = 0.218, β∗ = 0.008, γ∗ = −0.005. (5.122)

The finite values for β∗ and γ∗ also imply a non-zero value of σ∗ , via (5.112).
 5.3 Properties of the Einstein–Hilbert Flow 173

This should be contrasted to the one-loop result σ∗ = 0 obtained from

the perturbative quantization of fourth-order gravity. Obviously the non-
perturbative corrections captured by the FRGE modify the fixed point prop-
erties in such a way that Asymptotic Freedom is replaced with Asymptotic
Safety.
(ii) Stability properties (R2 + C 2 ): Linearizing the RG flow at the fixed
point (5.122) along the lines of Section 5.2.2 the critical exponents are
found as

θ0 = 2.51, θ1 = 1.69, θ2 = 8.40, θ3 = −2.11. (5.123)

We observe that the inclusion of the C 2 -term leads to entirely real stability
coefficients. This is in marked contrast to the complex stability coefficients
and the corresponding spiraling approach to the NGFP characteristic for
f (R)-type truncations.

Moreover, (5.123) shows that the “new” direction added by extending the R2 -
truncation is UV-repulsive, i.e., irrelevant. Thus, in a four-dimensional truncation
space, SUV remains three-dimensional: ∆UV = 3, dtrunc = 4. The condition for a
trajectory being inside the UV-critical surface then imposes one constraint among
the coupling constants. It can be used to express, say, γ in terms of the other
coupling constants contained in the ansatz. In the linear regime we can do this
even in closed form:

γ = γ(g, λ, β) = −0.116 + 0.030 λ g −1 + 0.049 g −1 + 11.06 β. (5.124)

This relationship parametrizes the hyperplane tangent to SUV at the NGFP.

The non-linear structure of SUV away from the fixed point can be determined
numerically.
Comparing the results (5.122) and (5.123) to their counterparts in the
Einstein–Hilbert and R2 -truncation, the C 2 -term is seen to lead to a moderate
shift of the NGFP properties. In particular, the product g∗ λ∗ turns out enhanced
by about a factor of three, g∗ λ∗ = 0.427, and all stability coefficients of the fixed
point are rendered real. Thus, the C 2 -term affects the fixed point structure more
drastically than the inclusion of the R2 -term, or working with different cutoff
schemes or gauge-fixing functions within the Einstein–Hilbert truncation.

Summary: The above results strongly suggest that the non-Gaussian fixed point
occurring in the Einstein–Hilbert truncation is not a truncation artifact but
rather the projection of a genuine fixed point of the full RG flow on the exact
theory space.
The fixed point and all its qualitative properties are stable against variations
of the cutoff and the inclusion of further invariants in the truncation. It is par-
ticularly remarkable that within the scheme dependence the additional R2 -term
174 Truncations of Single-Metric Type

has essentially no impact on the fixed point. Moreover, truncations involving

higher-order polynomials in R or the tensor structure C 2 fully confirm the quali-
tative picture suggested by the simpler Einstein–Hilbert truncation and provide a
strong indication that the corresponding quantum field theories are characterized
only by a finite number of free parameters.
We interpret the above results and their mutual consistency as quite non-trivial
indications supporting the conjecture that four-dimensional QEG possesses a
fixed point with precisely the properties needed for Asymptotic Safety.
 6
Bi-Metric Truncations

Obviously, there are many directions in which the truncations described in the
previous chapter can be extended. We may include more complicated invariants
into Γ̄k [g], for instance, or generalize the ghost and gauge-fixing terms, or allow
for a nontrivial extra background field dependence of the EAA, making it a
genuine bi-metric action. It is the latter kind of generalization we are going to
discuss in this chapter.
In Sections 5.1.1 and 5.1.3 we already introduced a broad class of bi-metric
truncations. The pertinent ansatz for the EAA (5.1) consisted of a δ B -invariant
functional Γgravk [g, ḡ], plus the classical gauge-fixing and ghost terms with, by
assumption, scale-independent prefactors. So far we have analyzed only a special
subsector of those truncations, namely the so-called single-metric truncations
described in Section 5.1.3. While they are less general and therefore probably
less reliable than a generic EAA of the form (5.1), they have the great advantage
of a considerably simplified RG equation. In the single-metric sector, the FRGE
boils down to the functional differential equation (5.15) for a running action with
only one metric argument, Γ̄k [g].
In this section we go beyond the single-metric approximation and consider
generic truncation ansätze of the type (5.1). We are thus forced to cope with
the considerably more complicated FRGE in (5.9) which RG evolves a generic
functional Γgrav
k [g, ḡ] of two independent metrics.

6.1 Level Expansion of Γgrav

k

Often it is natural to regard Γgrav

k as depending on the pair h, ḡ rather than g, ḡ
as the set of independent field variables:
Γgrav
k [h; ḡ] ≡ Γgrav
k [g, ḡ] . (6.1)
g=ḡ+h

Furthermore, if Γgrav
k [h; ḡ] admits a functional Taylor series expansion in hµν
about hµν = 0, we can classify the terms contributing to Γgrav
k [h; ḡ] according to
176 Bi-Metric Truncations

their level, i.e., their degree of homogeneity in the fluctuation field h. The EAA
then admits a level expansion of the form
∞
∑
Γgrav
k [h; ḡ] = Γ̌pk [h; ḡ], (6.2)
p=0

where the level-p contribution, Γ̌pk , by definition, is homogeneous with respect

to h, i.e., Γ̌pk [ch; ḡ] = cp Γ̌pk [h; ḡ] for any constant c > 0. In a symbolic notion,
Γ̌pk [h; ḡ] ∼ γ̌ pk [ḡ] · (hµν )p where the dot represents a summation and integration
over the discrete and continuous indices carried by the p factors of h, and γ̌ pk [ḡ]
is a tensor-valued functional which depends only on the background metric. At
the level p = 1, for example, we have explicitly:
∫
√
Γ̌1k [h; ḡ] = dd x ḡ hµν (x) γ̌ 1k [ḡ]µν (x). (6.3)

By virtue of the level expansion, we may consider the bi-metric action

Γgrav
k [g, ḡ] equivalent to an infinite family of single-metric functionals,
{γ̌ pk [ḡ]}p=0,1,2,··· , carrying a non-trivial tensor structure.
When we insert the level expansion into the FRGE and project on a fixed
level p, we see that ∂t Γ̌pk gets equated to an expression exclusively involv-
ing {Γ̌qk | q = 2, · · · , p + 2}. Hence, the noncanonical parts of all level-p beta
functions depend on the level-q couplings, with q = 2, · · · , p + 2, only. Gener-
ically the FRGE amounts to an infinite hierarchy of equations ∂t Γ̌pk = · · · for
p = 0, 1, 2, 3, · · · which does not terminate at any finite level. Only if it was possi-
ble to realize split-symmetry exactly this tower of equations collapses to a single
equation that governs all levels.

6.2 Level One and the Tadpole Equation

In the expansion (6.2) the level p = 1 is of special interest since it supplies the
tadpole equation with the help of which self-consistent backgrounds can be com-
puted. According to (5.11) with Γgrav b k it reads for the present class of
≡ Γ̄k + Γ
k
actions, and in the sector with vanishing ghosts:
δ
Γgrav [h; ḡ] = 0. (6.4)
δhµν (x) k sc
h=0;ḡ=ḡk

Inserting the expansion (6.2) into (6.4) we see that the tadpole equation is
tantamount to

γ̌ 1k [ḡ]µν sc
= 0. (6.5)
ḡ=ḡk

This condition can be regarded as a generalization of Einstein’s equation in

classical relativity. Indeed, if we identify Γgrav
k [h; ḡ] with the Einstein–Hilbert
 6.2 Level One and the Tadpole Equation 177
∫ √
term dd x gR(g)|g=ḡ+h , the left-hand side of (6.5) is nothing but the standard
Einstein tensor Ḡµν .
Nevertheless, a word of caution might be appropriate here. It is well known
that the usual Gµν has vanishing covariant divergence, Dµ Gµν = 0, and that
the same is true also for all generalizations of Gµν that can be obtained as the
functional derivative of an action S[g], which, first, depends only on one metric
and, second, is invariant under diffeomorphisms:
[ ]
1 δS[g]
Dµ √ = 0. (6.6)
g δgµν (x)

The standard argument leading to this conclusion1 fails however, if we replace

S[g] by a generic Γgrav b k [g, ḡ]. Returning to the “comma nota-
[g, ḡ] ≡ Γ̄k [g] + Γ
k
tion” where g and ḡ are the independent variables, (6.2) and (6.3) tell us
that
1 δ
γ̌ 1k [g]µν (x) ∝ √ Γgrav
k [g, ḡ] . (6.7)
δg
g(x) µν (x) ḡ=g

The generalized “Einstein tensor” γ̌ 1k is indeed a functional derivative but of an

action which depends on an additional field. Stated differently, it suffers from
an extra background field dependence, and this extra dependence entails that in
general
Dµ γ̌ 1k [g]µν ̸= 0. (6.8)

Only the more restrictive single-metric class of truncations which has Γ b k ∝ Sgf
grav
or, more generally, Γk [g, ḡ] = Γ̄k [g] + O((g − ḡ) ) yields a conserved tensor,
2

Dµ γ̌ 1k [g]µν = 0.
The non-conservation (6.8) is easy to understand: it expresses that while the
total 4-momentum is conserved, the dynamical metric can exchange momentum
with all other fields present, in particular with matter fields, if included, and
with the “extra” component of ḡµν .
When matter is coupled to gravity its energy-momentum tensor T µν replaces
the zero on the right-hand side of (6.5). The condition for the consistency of the
classical field equation, Dµ (g)T µν = 0, is then replaced by a more complicated
condition involving ḡµν . (A detailed discussion of this point can be found in
[189].)
At the observable level the “extra” part of the background metric will carry no
physical energy and momentum, however. At k = 0, in the gauge-invariant sector,
the restored split symmetry must erase all traces of the background metric we
introduced as an intermediate technical device.

1 Under an infinitesimal transformation δgµν = Lv gµν or, equivalently, δgµν = Dµ vν + Dν vµ

∫ ∫ √
the functional S[g] changes by an amount δS = dd x δgµν δgδS = 2 dd x g (Dµ vν ) √1g δgδS =
∫ d √ ( ) µν µν

−2 d x g vν Dµ √1g δgδS . Since vν is arbitrary, invariance (δS = 0) implies (6.6).

µν
178 Bi-Metric Truncations

6.3 A Bi-Metric Einstein–Hilbert Ansatz

The analysis of truncations with a non-trivial extra ḡ-dependence only started rel-
atively recently, the reason being that the derivation of projected flow equations
in the presence of two independent metrics g, ḡ tends to be a rather formidable
task. In order to project and concretely evaluate the functional traces in the
FRGE one has to carefully disentangle g and ḡ-dependencies, but no ready-made
efficient techniques are available for such calculations.2
The bi-metric truncations were introduced and tested within conformally
reduced gravity in [90]. Subsequently they were applied to matter-induced gravity
in [189], and finally, in [190, 191], to fully dynamical Quantum Einstein Gravity.
Special features of the RG flow in full gravity were analyzed further in [154, 192],
namely the (dis-)appearance of antiscreening in QEG and the existence of a cer-
tain c-function like quantity, respectively. Building on the level expansion (or
synonymously vertex expansion), flows of the Effective Average Action on a flat,
Euclidean background have been studied in [193–197] while the generalization to
curved backgrounds has been discussed in [155].
In the sequel we mostly focus on the investigations in [190, 191] which are based
on the following special truncation ansatz,3 referred to as a bi-metric Einstein–
Hilbert truncation:
∫ ( )
grav 1 d √ Dyn
Γk [g, ḡ] =− d x g R(g) − 2Λ k
16πGDyn
k
∫ ( ) (6.9)
1 d √
− d x ḡ R(ḡ) − 2Λk .
B
16πGB k

The action (6.9) consists of two structurally identical parts of the Einstein–
Hilbert type which depend on different metrics gµν and ḡµν . The ansatz involves
a total of 4 running coupling constants: the pair GDyn k , ΛDyn
k in the “dynamical,”
B B
i.e., gµν -dependent sector and Gk , Λk in the pure background part.
∫√ While the two ∫ √ invariants of the single-metric Einstein–Hilbert truncation,
gR(g) and g, are the only terms with at most two derivatives of the metric,
one can easily write down many more δ B -invariant zero- and second-derivative
terms
∫ √ depending ∫ √ on g∫ and ḡ than those appearing in (6.9). Examples include
√ µν
g R(ḡ), ḡ R(g), ḡ ḡ Rµν (g), etc., or more complicated mixtures of the
two metrics. In fact, ∫ √ there exists already an infinite variety of non-derivative
terms of the form g Y (gµν , ḡµν ), where Y is an arbitrary function.4

2 An exception are bi-metric computations on a flat background. In this case one may resort
to momentum space techniques developed within the framework of standard quantum field
theory.
3 In
∫ daddition
√ (√ to√ the)n ansatz (6.9), [190] also considered the mixed bi-metric invariant
d x ḡ ḡ/ g , n ∈ N, which will be omitted here.
4 In the case of matter-induced gravity, the infinite-dimensional RG flow of the full function
Y ≡ Yk has been analyzed in [189]. The generalization to the gravitational setting has been
carried out in [198].
 6.4 The Plethora of Couplings 179

Like in the single-metric case, we expect large anomalous dimensions to occur.

Hence, the importance of a given term in the ansatz cannot be estimated before-
hand by canonical power counting. As before, an important part of the analysis
will consist in the validation of the truncation a posteriori.

6.4 The Plethora of Couplings

In order to generate a list of all invariants that can occur in a bi-metric EAA,
it is advantageous to employ the h, ḡ-language and the level-expanded form of
Γgrav
k [h; ḡ], represented by the coefficients {γ̌ pk [ḡ]}p=0,1,2,··· .
As they depend on one metric only we can then derivative-expand the γ̌s
with respect to ḡ in the usual way.5 This provides us with a double expansion
of Γgrav
k [h; ḡ] in h and ḡ from which, at least in principle, a complete set of bi-
metric functionals can be read off. More precisely, this set may be considered
complete, i.e., a basis of theory space, provided the functionals to be expanded,
say the traces in the FRGE, are evaluated in conformity with this algorithm
(level expansion with subsequent derivative expansion in ḡ).

(1) In the “semicolon language,” we can set up truncations for Γgravk [h; ḡ] by
retaining some of the invariants in the list generated, while discarding others.
As an, admittedly not quite typical, example let us recast the double Einstein–
Hilbert action (6.9) in a form that makes its level structure explicit. Inserting
g = ḡ + h into (6.9) and expanding in h the first few terms read
∫
1 √ ( (0)
)
Γgrav
k [h; ḡ] = − (0)
dd x ḡ R(ḡ) − 2Λk
16πGk
∫
1 √ [ (1)
]
+ (1)
dd x ḡ Ḡµν + Λk ḡ µν hµν (6.10)
16πGk
∫ [ ]
1 √ ( (2) )
+ (2)
dd x ḡ hµν −K µνρσ D̄2 + 2Λk + U µνρσ hρσ
16πGk
3
+O(h ).

Here Ḡµν is the Einstein tensor for the background metric and K µνρσ and
µνρσ
U are given in (5.25) and (5.26) except that λ̄k is set to zero in U µνρσ .
In writing down (6.10) we designated the couplings appearing in a term of level
(p)
p by the superscript (p) . Since (6.10) descends from (6.9), the coefficients Gk
(p)
and Λk are, of course, not all independent. They can be expressed in terms of

5 For a partial classification of such monomials see [199].

180 Bi-Metric Truncations

the four couplings in (6.9):

(0) GDyn GB (1) (2)

Gk = k k
, Gk = Gk = · · · = GDyn
k ,
GDyn
k + GB
k
(6.11)
(0) ΛDyn GB B Dyn
k + Λk Gk (1) (2)
Λk = k
, Λk = Λk = · · · = ΛDyn
k .
GDyn
k + GB
k

(2) If we wanted to go beyond the double EH truncation we could relax these rela-
(1) (2)
tions and consider, say, Gk and Gk , as two independent functions of k. Note,
(p) (p)
however, that even if we treat all Gk , Λk , p = 0, 1, 2, · · · , ∞, as independent
we are still far from exhausting all possible invariants.
For example, there exists a further level-one term that also involves two deriva-
tives of the background metric. Including it into (6.10) the level-one part of the
EAA would generalize to
∫ [ ]
1 d √ (1) µν
(1)
d x ḡ Ḡµν
− 1
E
2 k ḡ µν
R̄ + Λ k ḡ hµν , (6.12)
16πGk

with an independent coupling constant Ek . In this case the tensor γ̌ 1k [ḡ]µν is,
(1)
up to a constant, given by Ḡµν − (1/2)Ek ḡ µν R̄ + Λk ḡ µν . While it has indeed
vanishing covariant divergence D̄µ γ̌ 1k [ḡ]µν = 0 if Ek = 0, the new term is seen to
spoil this property, D̄µ γ̌ 1k [ḡ]µν ∝ Ek ∂ν R̄.
When we invoke the single-metric approximation,

Γgrav
k [g, ḡ] = Γ̄k [g] (single-metric approx.), (6.13)

the Ek -term does not occur. And consistent with that, the δ B -invariance of Γ̄k [g]
implies by the usual argument that γ̌ 1k [ḡ]µν ∝ δgδµν Γ̄k [g]|g=ḡ is covariantly con-
served then: D̄µ γ̌ 1k [ḡ]µν = 0.

(3) Specializing (6.10) for the single-metric case of (6.13), the Newton and
cosmological constants of all levels, including p = 0, become equal:
(0) (1) (2) (0) (1) (2)
Gk = Gk = Gk = · · · and Λk = Λk = Λk = · · · . (6.14)

In the language of equation (6.9) this special case amounts to

1 ΛB
k
=0 and = 0. (6.15)
GBk GB
k

(p) (p)
Then GDyn
k = Gk , ΛDyn
k = Λk for all p = 0, 1, 2, · · · , ∞.

(4) The defining properties of theory space include a number of constraints, in

particular the Ward identity for split symmetry given by (4.96). In the truncation
 6.5 Results from the Bi-Metric EH Truncation 181

at hand it reduces to an equation of the form δḡµνδ (x) Γgrav

k [g, ḡ] = (1/2) STr[ · · · ]
with a complicated right-hand side that involves Γgrav
k itself. Projecting the Ward
identity on the space of action functionals (6.10) results in an infinite set of
(p) (p)
relations among all Gk and Λk , respectively.
At the lowest order, i.e., when we neglect loop contributions, STr[ · · · ] → 0,
those relations are nothing but (6.14), or equivalently (6.15). They characterize
the limiting case of vanishing extra ḡ-dependence.

6.5 Results from the Bi-Metric EH Truncation

In the sequel our presentation will mostly follow [191]. We consider the four-
parameter ansatz (6.9), or equivalently its level-expanded version, subject to the
relations (6.11). This amounts to setting Ek = 0,6 in particular.
After multiplying the Gk s and Λk s by the usual factors of k d−2 and k −2 , respec-
tively, the RG trajectories are parametrized by the four dimensionless couplings
(gkDyn , λDyn
k , gkB , λB
k ). Their RG equations read:
[ ]
∂t gkDyn = βgDyn (gkDyn , λDyn
k ) ≡ d − 2 + η Dyn (gkDyn , λDyn
k ) gkDyn ,
∂t λDyn
k = βλDyn (gkDyn , λDyn
k ),
[ ] (6.16)
∂t gkB = βgB (gkDyn , λDyn
k , gkB ) ≡ d − 2 + η B (gkDyn , λDyn
k , gkB ) gkB ,
B Dyn
∂t λB
k = βλ (gk , λDyn
k , gkB , λB
k ).

Generalizing (5.80), the anomalous dimensions η Dyn and η B are given by the loga-
rithmic scale derivatives of GDyn
k and GB k , respectively. Notably the two equations
for the “Dyn” sector close among themselves and do not contain the “B” cou-
plings, while conversely the “Dyn” couplings do appear in the beta functions
of the “B” sector. This triangular structure can be understood by noticing that
terms containing the background couplings are, by definition, independent of the
fluctuation fields so that they can not appear in the right-hand side of the flow
equation.
The equivalent system of equations in level language can be written in terms
(0) (0) (1) (1)
of the four independent couplings (gk , λk , gk , λk ) since, within the present
(p) (1)
truncation, all couplings at higher levels are equal to those at level one: gk = gk
(p) (1)
and λk = λk for all p ≥ 1.
The β-functions for the bi-metric Einstein–Hilbert ansatz have been derived in
[190, 191] for two different but, as it turned out, essentially equivalent projections.
They differ with respect to the gauge-fixing conditions and parameters, as well
as the field parametrization in use. In [190] the operator Rk was adapted to a
transverse-traceless (TT) decomposed field basis for hµν (cf. Appendix F), while
in [191] no TT decomposition was necessary.

6 An analysis of the RG behavior of Ek within matter-induced gravity can be found in [189].

182 Bi-Metric Truncations

The explicit formulae for the β-functions are too lengthy to be reproduced here
so that we must limit ourselves to a qualitative discussion of the RG flow they
give rise to. Again, we specialize to d = 4 dimensions.

6.5.1 The 2D Flow on the “Dyn” Subspace

{ }
On the four-dimensional parameter space T ≡ (g Dyn , λDyn , g B , λB ) the RG
flow decomposes hierarchically according to (g Dyn , λDyn ) → g B → λB . The “Dyn”
equations closing among themselves
{ allows
} us to first compute the flow on the
dynamical subspace TDyn ≡ (g Dyn , λDyn ) without reference to the background
couplings yet. Hence, the solution to the first two equations in (6.16) yields the
exact projection of the four-dimensional flow.
The two-dimensional projection onto the g Dyn -λDyn -plane is shown in
Figure 6.1. It is strikingly similar to the phase portrait of the correspond-
ing single-metric truncation in Figure 5.1: The g Dyn -λDyn -plane displays both
a Gaussian fixed point and a non-Gaussian one, denoted by GDyn -FP and
NGDyn -FP, respectively. Moreover, we can identify the same classes of trajec-
tories as in the single-metric discussion, namely those of Type Ia, IIa, and IIIa,
depending on whether the cosmological constant is heading for −∞, 0, or +∞
in the IR. Typical trajectories display a generalized crossover transition which
connects a fixed point in the UV to a classical regime in the IR. The latter is
located on the trajectory’s lower, almost horizontal, branch where g ≪ 1.

1.5

1.0
gDyn

0.5

0.0

1.0 0.8 0.6 0.4 0.2 0.0 0.2 0.4

lDyn

Figure 6.1. Phase portrait on the g Dyn -λDyn -plane obtained by projecting the
four-dimensional bi-metric flow. This projection is qualitatively identical with
the corresponding single-metric Einstein–Hilbert flow, displaying in particular
the projection of a 4-dimensional non-Gaussian fixed point. (Adapted from
[192].)
 6.5 Results from the Bi-Metric EH Truncation 183

The accessible part of theory space has a boundary, a line on which the
beta functions diverge. In Figure 6.1, and similar diagrams that will follow, this
boundary is indicated by a dashed line.
The properties of NGDyn -FP are qualitatively identical and numerically sim-
ilar to those of the NGFP found earlier in the single-metric truncations. The
critical exponents of NGDyn -FP form a complex conjugate pair with a non-
zero imaginary part leading to spiral shaped trajectories. The fixed point is UV
attractive in both directions. The precise numerical values of the critical expo-
nents still differ appreciably for the two different projections used in conjunction
with the ansatz (6.9). According to [190] and [191] they are 4.47 ± 4.23i and
3.6 ± 4.3i, respectively.

6.5.2 Screening vs. Antiscreening

There is an important difference between the single metric and the g Dyn -λDyn
flow which cannot be seen in the phase portraits: While according to the single-
metric EH truncation the anomalous dimension ηN (g, λ) is always negative for
g > 0, this is not the case for its bi-metric counterpart.
The anomalous dimension belonging to the dynamical Newton constant
η Dyn Dyn
(g , λDyn ) ≡ ∂t ln GDyn
k is found to change its sign at a certain critical
value of the dimensionless cosmological constant, λDyn crit . On the half-space with
g Dyn > 0, the anomalous dimension is positive in the domain where λDyn < λDyn crit ,
and negative when λDyn > λDyn crit ; see Figure 6.2.

1.5 0.6

0.5

1.0 0.4 –10

gDyn

0.3 –2

0.2 –0.5
0.5
–0.1
0.1

0.0

–1.0 –0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4

lDyn

Figure 6.2. The contour plot shows the lines of constant anomalous dimen-
sion η Dyn implied by the bi-metric calculation [191, 192]. The shaded regions
correspond to η Dyn > 0. The function η Dyn changes its sign on a straight line
λDyn = λDyn Dyn
crit with λcrit > 0. Near the Gaussian Fixed Point at the origin, η
Dyn

is seen to be positive. (Adapted from [192].)

184 Bi-Metric Truncations

Since λDyn
crit > 0, the former domain contains a vicinity of the Gaussian Fixed
Point, whereas the NGFP is located within the latter. After all, for purely dimen-
sional reasons, η Dyn must assume the value −2 directly at the NGFP; see (6.16)
with d = 4.
The dynamical Newton constant, GDyn k , increases (decreases) with decreasing
RG scale k, hence indicating a regime of gravitational antiscreening (screening),
provided η Dyn < 0 (η Dyn > 0).
In contrast to all earlier single-metric based calculations which unanimously
find a negative anomalous dimension everywhere, the double Einstein–Hilbert
truncation predicts that the gravitational antiscreening is lost at sufficiently
small values of the dimensionless cosmological constant. In the semiclassical
regime near the Gaussian Fixed Point g∗Dyn = 0 = λDyn ∗ , for example, GDyn
k shows
a screening behavior: When we start out at k = 0 and increase k, we first pass a
regime (extending to about the Planck scale) in which GDyn k grows slightly and
then reaches a maximum before it ultimately approaches zero according to the
fixed point scaling behavior GDynk = g∗Dyn k −2 , k → ∞.
On the other hand, according to the single-metric Einstein–Hilbert truncation,
Gk is still almost constant at the scales where GDyn k grows.

6.5.3 Flow on the 4D Parameter Space

Let us now return to the system of all four differential equations. Every
given two-dimensional trajectory k 7→ (gkDyn , λDyn k ), together with initial con-
ditions for the background couplings g B and λB , fixes a unique trajectory
on the (g Dyn , λDyn , g B , λB )-parameter space. Analyzing the corresponding four-
dimensional flow it turns out that the fixed point NGDyn -FP discussed
above “lifts” to two fixed points in four dimensions, GB ⊕ NGDyn -FP and
NGB ⊕ NGDyn -FP. As the notation indicates, they are, respectively, Gaussian
and non-Gaussian with respect to the background couplings.
The doubly non-Gaussian Fixed Point NGB ⊕ NGDyn -FP can be seen as
the bi-metric refinement of the familiar single-metric fixed point in the Einstein–
Hilbert truncation. It is the natural candidate for the construction of an
asymptotically safe UV limit. Within the present bi-metric truncation, it has four
relevant (i.e., UV-attractive) directions on the (g Dyn , λDyn , g B , λB )-parameter
space.

6.5.4 Imposing Split Symmetry

Every point of the (g Dyn , λDyn , g B , λB )-parameter space represents a specific
action of the form (5.1). It complies with some, but not yet all, defining prop-
erties the functionals forming the true theory space must possess. While it is
δ B -invariant and satisfies the BRST Ward identities to the lowest order, it does
not satisfy the Ward identity for split symmetry in general; see Section 5.1.3.
 6.5 Results from the Bi-Metric EH Truncation 185

Both types of Ward identities are known to be consistent with the RG flow in
an exact calculation, i.e., if they are fulfilled at one point of an RG trajectory,
they hold at any other point of the trajectory as well. On the other hand, in
a truncated calculation, consistency is expected to be attainable at best in an
approximate sense.
Investigations of this issue are particularly demanding technically since the
functional traces that occur in the Ward identities are even more complicated
than those of the FRGE. Therefore, the first investigation of the role played
by the split symmetry Ward identities [191] was based upon a leading-order
approximation analogous to the one invoked for the BRST identities; it consists
of neglecting the respective STr[ · · · ] terms which are due to higher loop contri-
butions.7 At the end of Section 6.4 we mentioned already that the split symmetry
Ward identities assume the simple form (6.14) or, equivalently, (6.15) then.
Those relations were imposed on the initial conditions of the RG trajectories
at some low infrared scale kIR and it was then checked numerically how well they
are satisfied at other scales.
Concerning the existence of asymptotically safe RG trajectories the main result
is the following:
Since the NGB ⊕ NGDyn -FP is UV-attractive in all four directions, its UV-
critical hypersurface SUV is 4-dimensional as long as one ignores the constraints
coming from the Ward identities. Hence, within the truncation considered, there
exists a four-parameter family of trajectories which emanate from the fixed
point in the UV and stay on SUV while heading for the IR. Generically, they
will not satisfy the constraints there. However, a detailed analysis has shown
[191] that there exists a two-parameter subfamily of trajectories which are both
asymptotically safe and satisfy the constraints due to the Ward identities.
Thus, effectively, the dimensionality of the true theory space and of SUV gets
lowered from 4 to 2 by imposing split symmetry or, more generally speaking,
Background Independence.
The significance of these results resides in the fact that they demonstrate, at
least within a truncation, that the two key requirements of Asymptotic Safety and
Background Independence can be attained simultaneously; they are not mutually
exclusive.

6.5.5 Discussion
A detailed comparison of the single- and the bi-metric Einstein–Hilbert trunca-
tion reveals that, quite unexpectedly, the former is a rather precise approximation
to the latter in the vicinity of the non-Gaussian fixed point.

7 For first steps toward solving the FRGE and modified split symmetry Ward-identities
simultaneously see [200, 201].
186 Bi-Metric Truncations

In the far IR the split-symmetry restoring trajectories, by construction, give

rise to another regime in which the two truncations agree well. However, in
between there are even qualitative differences, for instance with respect to the
sign of the dynamical anomalous dimension η Dyn .
Furthermore, substantial quantitative differences between the critical expo-
nents in both settings are found, despite the “miraculous” precision of the
single-metric truncation near the NGFP. They clearly show the limitations of
the single-metric approximation.
The general lesson that emerges from a comprehensive reliability analysis [154,
191, 192] is that from a certain degree of precision onward it no longer makes sense
to keep including further invariants in the truncation ansatz that are built from
the dynamical metric alone. Rather, to the same extent we allow the Effective
Average Action to depend on gµν in a more complicated way, also its dependence
on the background metric ḡµν must be generalized.
While the general picture conveyed by the single-metric truncations seems cor-
rect concerning the crucial question of an asymptotically safe UV limit, they are
insufficient for quantitatively precise calculations, the determination of critical
exponents being a prominent example.
Finally we refer the reader to [202] for the analysis of a different bi-metric
truncation that takes the influence of the anomalous dimensions of the hµν
and ghost-fields into account. Within this truncation, a NGFP suitable for
Asymptotic Safety is also found.
 7
Conformally Reduced Gravity

Conformal reduction in the sense we use the term here describes an approxima-
tion scheme in which only the conformal modes of the metric are quantized, i.e.,
integrated over in the functional integral; all other types of metric fluctuations
are neglected, leading to considerable technical simplifications. This approxi-
mation may be applied over and above any given truncation of theory space,
the prime example being the conformally reduced Einstein–Hilbert truncation
(CREH).
Reduced gravity theories are an instructive theoretical laboratory, which,
thanks to its mathematical simplicity, has led to a number of valuable general
insights that carry over to the full-fledged theory where all fluctuations are quan-
tized. The CREH model illuminates in particular, in a setting as transparent as
possible, how the special status of the gravitational field leads to RG flows that
are substantially different from those of typical matter systems.
The conformally reduced FRGE framework was introduced in [80] and
[203, 204], respectively, where the CREH truncation and an LPA generalization
thereof were analyzed. The reduced setting has often served as a testbed for new
developments. Bi-metric truncations [90], for example, or the “reconstruction” of
a regularized functional integral from the NGFP [205, 206], were first explored in
this setting. The investigation of Asymptotic Safety on infinite-dimensional the-
ory spaces [203] also started with conformally reduced gravity. Many of the new
techniques required by the f (R) truncations, for example, were developed in this
context.
Going beyond the CREH model one might also explore the effective dynam-
ics of the conformal factor within full-fledged QEG, i.e., with all fluctuation
modes quantized. At this level, the implications of Asymptotic Safety have been
analyzed in [207, 208].
188 Conformally Reduced Gravity

7.1 From Gravity to ϕ4 Theory1

To prepare the stage for the CREH truncation we start out from the Einstein–
Hilbert action
∫
1 √ ( )
SEH [gµν ] = − dd x g R(g) − 2Λ (7.1)
16πG
and assume that its argument, gµν , equals a certain conformal factor times a
fixed, non-dynamical reference metric g̊µν . We would like to parameterize this
conformal factor in terms of a scalar function ϕ(x) in such a way that the kinetic
term for ϕ has the standard form ∝ (∂µ ϕ)2 . This is indeed possible, for any
dimensionality of spacetime, provided one expresses the conformal factor appro-
priately in terms of the independent field variable, ϕ. In fact, introducing ϕ
according to [209],
gµν = ϕ2ν(d) g̊µν , (7.2)
with the dimension-dependent exponent
2
ν(d) = , (7.3)
d−2
yields the following result for SEH evaluated on metrics of the form (7.2):
∫ √ (
1
SEH [ϕ] = − dd x g̊ 12 g̊ µν ∂µ ϕ ∂ν ϕ + 21 ξ(d) R̊ ϕ2
8π ξ(d) G
) (7.4)
− ξ(d) Λ ϕ2d/(d−2) .

Here R̊ denotes the curvature scalar of the reference metric g̊µν , and

d−2
ξ(d) ≡ . (7.5)
4(d − 1)

Employing an exponent ν different from (7.3) the kinetic term in SEH [ϕ] would
fail to be bilinear in the field ϕ.
In f our dimensions, for example, we have ν(4) = 1 and ξ(4) = 61 so that

gµν = ϕ2 g̊µν (7.6)

converts the Einstein–Hilbert action to a kind of “ϕ4 -theory”:

∫ √ ( µν )
3
SEH [ϕ] = − dd x g̊ 12 g̊ ∂µ ϕ ∂ν ϕ + 1
12 R̊ ϕ
2
− 16 Λ ϕ4 . (7.7)
4πG

1 Our presentation in Sections 7.1–7.4 mostly follows [80] to which the reader is referred for
further details.
 7.1 From Gravity to ϕ4 Theory 189

We refer to the action (7.4) and its special case (7.7) as the conformally reduced
Einstein–Hilbert or “CREH” action.
For d > 2, the case we will always assume in the following, the kinetic term in
(7.4) is negative definite due to the “wrong sign” of its prefactor. As a result, the
action is unbounded below: with a ϕ(x) which varies sufficiently rapidly, SEH [ϕ]
becomes arbitrarily negative. This is the notorious conformal factor instability.
Quantizing the conformal factor in the CREH approximation2 with the bare
action SEH [ϕ] thus seems similar to quantizing a scalar theory with an action of
the type ∫ { }
2
S[ϕ] = c d4 x − 12 (∂ϕ) + U (ϕ) , (7.8)

where c is a positive constant. For the sake of the argument let us assume that
g̊µν = δµν is the flat metric on R4 . Then SEH of (7.7) is indeed of the form (7.8)
with the potential U (ϕ) = 16 Λ ϕ4 and c = 4πG
3
> 0. If the cosmological constant is
positive, as in the case which will be relevant later on, the potential term in (7.8)
is positive definite.
Now we would like to understand the quantum theory based upon the
functional integral ∫
eb
I ≡ Dϕb eiS[ϕ] , (7.9)

where Se is the Wick rotated version of S, with the Lorentzian (∂ ϕ) b 2≡

η µν ∂µ ϕb ∂ν ϕ.
b One would expect that in this theory the wrong sign of the kinetic
term drives the condensation of spatially inhomogeneous (x-dependent) modes,
i.e., the formation of a “kinetic condensate” [222]. The amplitude of the inhomo-
geneous modes cannot grow unboundedly though since this would cost potential
energy.
Next, consider the closely related theory with the “inverted” action Sinv [ϕ] ≡
−S[ϕ]. Thus, ∫ { }
2
Sinv [ϕ] = c d4 x + 12 (∂ϕ) + V (ϕ) (7.10)

involves the negative potential V (ϕ) ≡ −U (ϕ) ≤ 0. In pulling out a global minus
sign from S the instability has been shifted from the kinetic to the potential
term. According to the picture conveyed by Sinv , the kinetic energy now assumes
its minimum for x-independent field configurations ϕ = const, while the new
potential V (ϕ) = − 16 Λ ϕ4 is unbounded below when Λ > 0 so that it causes an
instability.
Even though S and Sinv appear to be plagued by instabilities of a rather dif-
ferent nature, they should describe the same physics (possibly up to a time

2 For the quantization of a more general bare action S[ϕ] including the non-local terms induced
by the conformal anomaly of matter fields [210] we refer to the work of Antoniadis, Mazur,
and Mottola [211–217]. Implications for the cosmological constant problem are discussed in
[218–221].
190 Conformally Reduced Gravity

reflection). Indeed, the path integrals involving S and Sinv , respectively, are
related by complex conjugation:
∫
eb
Iinv ≡ Dϕb e−iS[ϕ] = I ∗ . (7.11)

One refers to the formulations in terms of S and Sinv as the original picture and
the inverted picture, respectively. Due to the projective nature of the Wetterich
equation (2.15) one expects that the overall sign drops out so that both pictures
lead to identical RG flows.
So we see that for pure gravity in the CREH approximation the “wrong” sign
of the kinetic term can be traded for an upside down potential. The FRGE for-
malism we are going to set up will effectively correspond to the inverted picture,
employing an action with a positive kinetic, but negative potential term when
Λ > 0.
The scalar ϕ4 -theory with a negative coupling constant was already discussed
by Symanzik [223] in the early 1970s. He showed that the coupling strength
decreases at short distances, thus providing the first example of an asymptotically
free quantum field theory [224].

7.2 The Untruncated Conformally Reduced Theory

Let us now turn to a general reduced quantum theory of the conformal factor
and the associated FRGE, which is still exact in the sense that no truncation
beyond the reduction is performed. ∫
we are led to investigate the “toy integral” Dϕb e−S[ϕ] instead of the
b
Thus,
∫ −S[b
full Db gµν e gµν ]
, where the integration is only over metrics conformal to the
3
reference metric :
gbµν = ϕb2ν(d) g̊µν . (7.12)
Here the microscopic, or quantum field, i.e., the field that would be promoted to
an operator in canonical quantization, is denoted by ϕ. b It differs from the actual
conformal factor, which is a certain power of it, ϕb2ν(d) . As a result, the func-
tional measure for ϕb as it descends from the full integral differs from the usual
translation-invariant measure Dϕb by a Jacobian that is neglected in the simplest
form of the CREH model.4 Yet, the quantum conformal factor is a composite
field now,5 and so its expectation value is not directly related to that of the
elementary field [229].

3 To make contact with the notation used in the literature [80, 203] one should make the
following replacements in the above and following equations: g̊µν → g b → χ, φ
bµν , ϕ b → f , φ → f¯,
ϕ̄ → χB , a → φ.
4 For a discussion of the measure see in particular [135, 225, 226] and, in relation to CDT,
[47, 227, 228].
5 If d = 4 and g̊µν = δµν we may interpret the composite operator representing the metric as the
product of two conformally reduced tetrad fields eba b a bµν = ϕb2 δµν =
µ = ϕ δµ , since in this case g
eba b
e b δ .
µ ν ab
 7.2 The Untruncated Conformally Reduced Theory 191

To make the analogy to full gravity as explicit as possible we compute the ϕ-b
integral in the indirect way by means of using a functional RG equation employing
the background field technique. More precisely, the field variable integrated over
is split linearly,
ϕb = ϕ̄ + φ,
b (7.13)
b
with a classical background ϕ̄ and a dynamical fluctuation field φ.
The starting point for the derivation of the FRGE is the following generating
functional with an infrared cutoff:
∫ (
( )
exp Wk [J; ϕ̄] = Dφ b exp −S[ϕ̄ + φ]
b − ∆Sk [φ;b ϕ̄]
∫ √ ) (7.14)
d
+ d x g̊ J(x)φ(x)b .

To mimic the situation in full gravity, the mode suppression term ∆Sk is chosen
bilinear in the fluctuation rather than in the full field, and the cutoff operator
Rk ≡ Rk [ϕ̄] is allowed to depend on the background field:
∫ √
1
b ϕ̄] =
∆Sk [φ; b Rk [ϕ̄] φ(x).
dd x g̊ φ(x) b (7.15)
2
By the same series of steps as in Chapter 2, the generating functional (7.14)
leads to the corresponding Effective Average Action Γk [φ; ϕ̄] and its FRGE. It
is similar to that of a matter field theory on a non-dynamical spacetime, albeit
a possibly curved one with arbitrary metric g̊µν :
[( )−1 ]
1 (2)
k∂k Γk [φ; ϕ̄] = Tr Γk [φ; ϕ̄] + Rk [ϕ̄] k∂k Rk [ϕ̄] (7.16)
2
∫ √
Here, in bra-ket notation, Tr( · · · ) ≡ dd x g̊⟨x|( · · · )|x⟩, and the Hessian oper-
δ 2 Γk [φ;ϕ̄]
ator is given by ⟨x|Γk [φ; ϕ̄]|y⟩ = √ 1√
(2)
g̊(x) δφ(x)δφ(y) .
g̊(y)
For the field expectation values with respect to the above functional integral
we use the notation φ ≡ ⟨φ⟩b as well as ϕ ≡ ⟨ϕ⟩ b = ϕ̄ + ⟨φ⟩
b ≡ ϕ̄ + φ. Compared to
the full gravity theory the analogous field variables are gµν ↔ ϕ, ḡµν ↔ ϕ̄, hµν ↔
φ. However, there is no counterpart of g̊µν in the full-fledged quantum gravity
theory.
Also note that the linear split of ϕb amounts to a non-linear parametrization
of the full dynamical metric in terms of φ b if we write the conformal factor as in
(7.12) with (7.13).
Besides the reference metric g̊µν , we must distinguish the following conformally
reduced metrics now: the background metric,

ḡµν = ϕ̄2ν(d) g̊µν , (7.17)

b
the quantum metric, a non-linear function of ϕ,

gbµν = ϕb2ν(d) g̊µν = (ϕ̄ + φ)

b 2ν(d) g̊µν , (7.18)
192 Conformally Reduced Gravity

the expectation value of the quantum metric,

gµν ⟩ = ⟨ϕb2ν(d) ⟩g̊µν = ⟨(ϕ̄ + φ)
gµν ≡ ⟨b b 2ν(d) ⟩g̊µν , (7.19)
and a novel metric with conformal factor ϕ2ν(d) :
ğµν ≡ ϕ2ν(d) g̊µν ≡ (ϕ̄ + φ)2ν(d) g̊µν . (7.20)
In general, gµν and ğµν differ, but they are approximately equal if the quantum
fluctuations are negligible.
In d = 4 where ν = 1, for instance, we have
[ ]
gµν = ḡµν + 2 ϕ̄ ⟨φ⟩
b + ⟨φb2 ⟩ g̊µν ,
[ ] (7.21)
ğµν = ḡµν + 2 ϕ̄ ⟨φ⟩
b + ⟨φ⟩
b 2 g̊µν .
[ 2 ]
Hence, gµν − ğµν = ⟨φ b ⟩ − ⟨φ⟩
b 2 g̊µν is proportional to the variance of φ so that
gµν and ğµν are not too different if the fluctuations of φ are small. In order to
make this statement precise one would have to give a mathematical meaning to
the expectation value of the operator product φ b2 with both operators at the same
point, something we will not attempt here; see [229]. But notice that ğµν reduces
to ḡµν if φ = 0. For gµν this is not necessarily the case since gµν = ḡµν + ⟨φ
b2 ⟩g̊µν .

7.3 Coarse Graining Operators: Gravity vs. Matter

The FRGE (7.16) is almost the same as the one governing the RG flow of a scalar
matter field on a classical spacetime with metric δµν , (2.107). The only minor
difference is the replacement of δµν by a possibly more general but likewise non-
dynamical metric g̊µν . So, what can this simple system teach us about quantum
gravity?
This brings us to an important issue at the heart of the gravitational Effective
Average Action and its interpretation: the interplay of Background Independence
and the deeper reason for the existence of a non-Gaussian fixed point.

(1) Recall that the mass scale k parameterizing Γk [g, ḡ, · · · ] equals a special
eigenvalue of a certain ḡµν -dependent cutoff operator, the covariant Laplacian
typically. Namely, k 2 is the eigenvalue of the “last” eigenmode that is integrated
out without suppression.
We insist that the same property holds true in the conformally reduced setting
as well. This is a non-trivial decision since, besides ḡµν ≡ ϕ̄2 g̊µν , we could also
use g̊µν itself in order to construct the operator which discriminates the modes.
But this option would not have an interpretation at the exact level though.
To find the correct Rk in (7.15), we think of the integration variable φ b in (7.14)
as expanded in terms of the eigenfunctions of the Laplace–Beltrami operator
pertaining to ḡµν :

¯ ≡ √1 ∂µ √ḡ ḡ µν ∂ν .
 (7.22)
ḡ
 7.3 Coarse Graining Operators: Gravity vs. Matter 193

As always, Rk must suppress modes with eigenvalues smaller than k 2 by giving

them a mass of the order k, while those with larger eigenvalues are unsuppressed.
b
In the simplest case when the φ-modes have a kinetic operator proportional to 
¯
(2)
itself, the rule is that the correct Rk , when added to Γk , causes the replacement
( ¯)
(−) ¯ + k 2 R(0) − 
¯ −→ (−) (gravity), (7.23)
k2

with R(0) (z) an arbitrary shape function satisfying (2.65). Hence, the effective
inverse propagators of the long- and short-wavelength modes are − ¯ + k 2 and
−, respectively, and the long/short-transition is at the −-eigenvalue k 2 , as it
¯ ¯
should be.
In this manner the mass scale k acquires a “quasiphysical” status. More pre-
cisely, its inverse k −1 has the interpretation of a proper length with respect to
the metric ḡµν . At least for simple geometries, this length is directly related to
the “diameter,” ℓ, of the spacetime volumes averaged over by the coarse-graining.

(2) The coarse-graining scale ℓ = ℓ(k) corresponding to a given value of k can be

estimated by investigating the properties of the specific −-eigenfunction
¯ that
2
possesses the eigenvalue k , the so-called cutoff mode [230, 231]: one determines
its typical scale of variation with respect to x (a period, say) and converts this
coordinate length to a physical, i.e., proper, length by using ḡµν . The result,
ℓ(k), is an approximate measure for the extension of the spacetime volumes up
to which the dynamics has been coarse-grained. If ḡµν is close to a flat metric,
ℓ(k) equals approximately π/k. (See [230, 231] for a detailed discussion.)
In this sense the background metric ḡµν , or rather its conformal factor ϕ̄,
determines the physical, i.e., proper scale of k. Furthermore, for self-consistent
backgrounds we have ⟨b gµν ⟩ = ḡµν . So we can even go one step further and say
that k is proper, at the level of expectation values, with respect to the dynamical
quantum metric.

(3) While the above choice of Rk , based on the rule (7.23), is the correct one in
quantum gravity, the standard quantization that treats ϕ as an ordinary scalar
matter field uses a different cutoff, namely one based on the Laplace–Beltrami
˚ ≡g̊ −1/2 ∂µ g̊ 1/2 g̊ µν ∂ν . For a
operator pertaining to the reference metric g̊µν , 
matter field, Rk is usually designed to implement the replacement:
( ˚)
(−) ˚ + k 2 R(0) − 
˚ −→ (−)
k2 . (matter) (7.24)

As a consequence the proper scale of k is now determined by the metric g̊µν , i.e.,
a metric that has no physical meaning within Background Independent quantum
gravity. It “knows” nothing about the true “on-shell” metrics of spacetime, like
the self-consistent background metrics, for example, which are determined by the
b + ⟨φ
dynamics such that, in d = 4, 2ϕ̄⟨φ⟩ b2 ⟩ = 0.
194 Conformally Reduced Gravity

The scheme (7.24) is the correct choice if one interprets ϕ as a standard matter
field on a non-dynamical spacetime with metric g̊µν , on flat space (g̊µν = δµν ),
for instance. The average action formalism based on (7.24) then reproduces in
particular all the familiar results of perturbation theory, the ln(k)-running of the
quartic coupling in ϕ4 -theory, for example.

(4) Contrary to ḡµν , the metric g̊µν is an absolutely rigid object. The reference
metric never gets changed and thus constitutes an “absolute element” in the
RG formalism which destroys its Background Independence. This is the reason
it should not be used for gravity. Only the cutoff that employs ḡµν respects
Background Independence.
As we will see, the RG flow based on the -scheme
¯ (7.23) is extremely different
from the one for standard scalars. The reason is that, via the ϕ̄ dependence of
,
¯ the gravitational field itself sets the scale of k. The difference between (7.23)
and (7.24) becomes manifest when we recall that the Laplacians of g̊µν and
ḡµν = ϕ̄2ν(d) g̊µν are related by

¯ = ϕ̄−2ν(d) 
 ˚ + O(∂ ϕ̄). (7.25)

The factor ϕ̄−2ν(d) leads to considerable modifications of the RG flow within the
conformally reduced Einstein–Hilbert truncation, the topic we discuss next.

7.4 The Reduced Einstein–Hilbert Truncation

So far we have applied the conformal reduction to the exact, untruncated theory.
In order to be able to perform explicit calculations, we now combine it with a
truncation of theory space. As a first example of the single-metric type, we turn
to the conformally reduced Einstein–Hilbert truncation (CREH). For simplicity
we specialize for d = 4 in the sequel.

(1) As an ansatz for Γk [φ; ϕ̄] we choose SEH [ϕ̄ + φ] from (7.7) with a k-dependent
Newton constant Gk and cosmological constant Λk :
∫ √ {
3 ( ) ( )
Γk [φ; ϕ̄] = − d4 x g̊ − 12 ϕ̄ + φ ˚ ϕ̄ + φ
4πGk
)4 }
(7.26)
( )2 (
1
+ 12 R̊ ϕ̄ + φ − 61 Λk ϕ̄ + φ .

From (7.26) we obtain the Hessian operator

(2) 3 { ˚ ( )2 }
Γk [φ; ϕ̄] = − −x + 16 R̊(x) − 2Λk ϕ̄(x) + φ(x) , (7.27)
4πGk

to which we come back in a moment.

 7.4 The Reduced Einstein–Hilbert Truncation 195

(2) To find the beta functions of Gk and Λk we insert (7.26) into (7.16) and
perform a derivative and field expansion of the functional trace, retaining only
terms proportional to the monomials ϕϕ, ˚ R̊ϕ2 , and ϕ4 , respectively, where
ϕ ≡ ϕ̄ + φ. It is sufficient to perform the projection at a constant background
field ϕ̄(x) ≡ ϕ̄.
Since Gk appears both in the kinetic and the R̊ϕ2 -term of the potential in the
ansatz (7.26), it is possible to compute the anomalous dimensions ηN ≡ k∂k ln Gk
in two different ways, namely either from the kinetic or the potential term. The
kin pot
respective anomalous dimensions, ηN and ηN , have no reason to be exactly
equal given the approximations we made, but they should at least turn out com-
parable if the truncation is reliable.

(3) In quantum gravity the cutoff must be imposed on the spectrum of , ¯ not
that of .
˚ Since ϕ̄ = const in the case at hand, the two operators are related by


˚ = ϕ̄2 
¯ (7.28)
(2)
so that we may re-express Γk as

(2) 3 { 2¯ 1 }
Γk [φ; ϕ̄] = − −ϕ̄  + 6 R̊ − 2Λk (ϕ̄ + φ)2 . (7.29)
4πGk
Now we fix Rk in a way that effects the replacement (7.23):
(2)
Γk [φ; ϕ̄] + Rk [ϕ̄]
3 { 2[ ¯ ( ¯ )] } (7.30)
=− ϕ̄ − + k 2 R(0) − k2 + 16 R̊ − 2Λk (ϕ̄ + φ)2 .
4πGk
As a consequence, the cutoff operator acquires an explicit dependence on the
background field :

3 ( ¯) 3 ( ˚ )
Rk [ϕ̄] = − ϕ̄2 k 2 R(0) − k2 = − ϕ̄2 k 2 R(0) − ϕ̄2k2 . (7.31)
4πGk 4πGk

By contrast, installing the cutoff in the spectrum of  ˚ instead and applying

the rule (7.24), we find the cutoff operator for standard matter fields,

3 ( ˚)
Rstandard
k =− k 2 R(0) − k2 . (7.32)
4πGk

This operator is field independent.

Even though the field present in the operator (7.31) is not the dynamical one,
ϕ, but the background, ϕ̄, the two RG equations based on Rk [ϕ̄] and Rstandard
k ,
are quite different. In fact, in projecting out the monomials ϕϕ,˚ R̊ϕ2 , and ϕ4
from the FRGE we set ϕ = ϕ̄ after computing the Hessian so that ϕ and ϕ̄ are
indistinguishable from that point on.
196 Conformally Reduced Gravity

Hence, in the quantum gravity setting, there are additional ϕ-dependent terms
which affect the beta functions. They originate from the cutoff operator Rk and
are absent in the standard matter field theory.

(2)
(4) Note also the overall minus sign displayed by Γk in (7.27) and, as a con-
sequence, Rk in (7.31), which is a reflection of the conformal factor instability.
However,
( (2)upon inserting
)−1 them into the flow equation, the minus signs cancels
from Γk + Rk ∂k R k .
This is exactly the point where in the FRG setting the transition from the
“original” to the “inverted” picture takes place: the negative kinetic term is
traded for an upside-down potential.

(5) The detailed derivation of the beta functions for gk ≡ k d−2 Gk and λk ≡
k −2 Λk , with an arbitrary shape function and for any spacetime dimensional-
ity, can be found in [80]. Here we only quote the results for the optimized shape
function (5.120) and d = 4.
The coupled RG equations [have the same] structure as in the full Einstein–
Hilbert truncation: k∂k gk = 2 + ηN (gk , λk ) gk , k∂k λk = βλ (gk , λk ). For the
anomalous dimension coming from the kinetic term we obtain

(kin) 2 gλ2
ηN (g, λ) = − . (7.33)
3π (1 − 2λ)4

The one derived from the potential has the familiar structure

(pot) g B1 (λ)
ηN (g, λ) = , (7.34)
1 − g B2 (λ)

with the following B-functions:

()
1 1 1
B1 (λ) = −λ 2,
3π 4 (1 − 2λ)
( ) (7.35)
1 1 1
B2 (λ) = − −λ 2.
12π 3 (1 − 2λ)

For βλ one finds

( )
g 1 1
βλ (g, λ) = − (2 − ηN ) λ + 1 − ηN , (7.36)
4π 6 1 − 2λ
(kin) (pot)
where either ηN or ηN is to be inserted for ηN .

(6) The solutions to the above RG equations deviate considerably from those in
a standard scalar matter theory, which employ the operator Rstandard
k . Here is a
particularly striking example.
 7.4 The Reduced Einstein–Hilbert Truncation 197

Let us consider an RG trajectory in a regime where the anomalous dimen-

sion is small so that we may approximate ηN = 0. Then Gk = const ≡ G0 , and the
equation for λk involves the correspondingly simplified β-function (7.36) with
gk ≡ G0 k 2 . In terms of the dimensionful cosmological constant Λk ≡ k 2 λk this
RG equation reads:
G0 k6
k∂k Λk = . (7.37)
4π k 2 − 2Λk
In particular, when Λk ≪ k 2 it simplifies to
1
k∂k Λk = G0 k 4 . (7.38)
4π
Obviously the RG equations of the CREH truncation imply a quartic running of
the cosmological constant as long as Λ is small and G is approximately constant.
This behavior is exactly the same as the one found on the basis of the full-
fledged Einstein–Hilbert truncation, (5.87).
From the quantum gravity perspective it is unsurprising, being the well known
result of summing up the zero-point energies of all field modes.
On the other hand, from the matter field perspective, the quartic running is
a surprise. In the CREH ansatz for Γk the cosmological constant Λk plays the
role of a ϕ4 -coupling constant which, quite unusually, behaves as Λk ∝ k 4 here.
This comparatively strong scale dependence has to be contrasted with the much
weaker, merely logarithmic k-dependence one finds with Rstandard
k in an ordinary
scalar matter theory on a 4-dimensional flat spacetime.
The origin of this difference is clear: The conformal factor determines the
proper scale of the cutoff, while a scalar matter field does not.
In this manner the distinguished status of the conformal factor comes into play.
We emphasized already that if the coarse-graining scale is to be given a proper
as well as Background Independent meaning, k must be a cutoff in the spectrum
of the background field dependent operator .¯ This enforces the substitution rule
(7.23) which differs from the one for standard matter. The rule (7.23) in turn
is crucial for the occurrence of a NGFP. This establishes the deep connection
between Background Independence and Asymptotic Safety we alluded to above.

(7) Analyzing the RG flow numerically, the main results are as follows.
On the g-λ-plane, the RG equations of the CREH truncation give rise to a
phase portrait which, at the qualitative level, is identical to the one in Figure 5.1
that was obtained in the full Einstein–Hilbert truncation. This striking similarity
between the reduced and the full flow occurs with both variants of the anomalous
(kin) (pot)
dimension, ηN and ηN , respectively.
It is particularly astonishing that the CREH flow, coming from a simple scalar-
looking theory, displays a non-Gaussian fixed point. The product of its coordinate
values g∗ and λ∗ in the “kin” and “pot” schemes are (g∗ λ∗ )(kin) ≈ 1.296, and
(g∗ λ∗ )(pot) ≈ 1.106, respectively.
198 Conformally Reduced Gravity

The unexpectedly small discrepancy between them is of the same order of

magnitude as the uncertainty in the full EH truncation. This suggests that the
CREH truncation could have a certain degree of internal consistency. On the
other hand these values differ substantially from the result in the full EH trunca-
tion, which is (g∗ λ∗ )(full EH) ≈ 0.137 for the same cutoff. But this is no reason for
concern, since we are comparing theories with a different number of fields here.
The critical exponents of the NGFP show a stronger variation. They form com-
plex conjugate pairs, θ(kin) = 4.0 ± 6.2 i and θ(pot) = 1.47 ± 9.30 i, which should
be compared to θ(full EH) = 1.48 ± 3.04 i. Both CREH calculations agree on a pos-
itive real part. Hence, in conformity with the full EH truncation, the NGFP is
UV attractive in both directions.
These findings are consistent with the general expectation that the CREH
model could represent a theory sharing certain properties with full quantum
gravity. But it should not be regarded as an approximation thereof, though.
For further details on the CREH truncation the reader is referred to the
literature [80]. Within the finite-dimensional (and single-metric) framework,
conformally reduced gravity has been analyzed further in [232] using a more
advanced truncation which involves four-derivative terms.

7.5 An Infinite-Dimensional Truncation

Among other developments, conformally reduced gravity also allowed for the first
investigation of Asymptotic Safety on a theory space, which, while truncated, is
still infinite dimensional [203].
The corresponding truncation is similar to the local potential approximation
(LPA) which we encountered in Section 2.2.4 for standard scalar matter fields.
The ansatz reads:
∫ √ (
3 ) ( )
Γk [φ; ϕ̄] = + d4 x g̊ ϕ̄ + φ 
˚ ϕ̄ + φ
8πGk
∫ √ (7.39)
( )
+ d4 x g̊ Uk ϕ̄ + φ .

Here Uk ( · ) is an arbitrary, k-dependent potential function ( 2 for ϕ̄ + )φ ≡ ϕ,

generalizing the simple polynomial UkCREH (ϕ) = − 48πG 3
k
R̊ϕ − 2Λ k ϕ4
used
above.
With the ansatz (7.39) the projected FRGE consists of a partial differential
equation coupled to an ordinary one. Together they determine the scale depen-
dence of Uk ( · ) and Gk , respectively. So the corresponding truncated
( theory
) space
is indeed infinite dimensional. We coordinatize it by pairs g, Y ( · ) where g
denotes the familiar dimensionless Newton constant and Y ( · ) is a dimensionless
potential depending on a likewise dimensionless field, kϕ ≡ a:
3
Uk (ϕ) ≡ − Yk (a), a ≡ kϕ. (7.40)
4πgk
 7.5 An Infinite-Dimensional Truncation 199

Using the same operator Rk as for the CREH truncation, the flow is governed
by the partial differential equation

k∂k Yk (a) = (2 + ηN ) Yk (a) − a Yk′ (a)

( )
gk 1 − 2N a2 ρ(a) + 12 ηN ρe(a)
η (7.41)
− .
2π a2 + Yk′′ (a)
[ ( )]
It must be solved together with k∂k gk = 2 + ηN gk , [Yk ] gk .
The Y ( · )-dependent anomalous dimension, read off from the kinetic term, is
explicitly given by
[ ]2
( ) gk a31 Yk′′′ (a1 )
ηN gk , [Yk ] = − [ ]4 . (7.42)
24π 2
a1 + Yk′′ (a1 )

Here a1 is a normalization point that ultimately will be sent to infinity, a1 → ∞.

The partial differential equation (7.41) contains the two functions
[ ] [ ]
ρ(a) ≡ Tr Θ(a2 + )
˚ , ρe(a) ≡ Tr (−)Θ(a
˚ 2
+ )
˚ , (7.43)

which are sensitive to the spectrum of , ˚ the Laplace–Beltrami operator per-

taining to the reference metric g̊µν . This brings us to the first general lesson we
can learn from this truncation.

(1) Topology dependence. The truncation at hand is the first one we

encounter that allows us to see a phenomenon which we expect on general
grounds, namely that the RG flow depends on the topology of the spacetime
manifold.
Considering manifolds of topology S 4 and R4 as examples, the spectral func-
tions ρ, ρe have been calculated [203] with g̊µν , given by the metric of a round
unit sphere6 in the first, and by δµν in the second case.

(i) The spectrum of the Laplacian on the unit sphere S 4 is well known. The
eigenvalues En of (−)
˚ and their degeneracies Dn , n = 0, 1, 2, · · · , for (−)
˚
acting on scalars are given by [233–235]

En = n(n + 3), Dn = 61 (n + 1)(n + 2)(2n + 3). (7.44)

To evaluate ρ and ρe we note that

∞
∑ ∞
∑
ρ(a) = Dn Θ(a2 − En ), ρe(a) = En Dn Θ(a2 − En ) (7.45)
n=0 n=0

6 The radius of this sphere is inessential. A change could be absorbed by a redefinition of the
field variable.
200 Conformally Reduced Gravity

are functions of the form

( ) ( )
ρ(a) = J4 N (a) , ρe(a) = Je4 N (a) , (7.46)

where N (a) is the largest positive integer n such that En = n(n + 3) < a2 ,
and
∑
N ∑
N
J4 (N ) ≡ Dn , Je4 (N ) ≡ En Dn . (7.47)
n=0 n=0

The finite sums can be worked out explicitly:

1
J4 (N ) = 12 N4 + 2
3 N3 + 23 2 7
12 N + 3 N + 1,
(7.48)
Je4 (N ) = 1
18
6
N + 2
3 N 5 + 18
55
N 4 + 20
3 N
3
+ 62
9 N2 + 8
3 N.

Since the largest contributing eigenvalue has the quantum number

{
a for a ≫ 1,
N (a) ≈ (7.49)
0 for a ≪ 1,

we observe that for very large and small dimensionless fields, respectively,
{ {
1 4
a for a ≫ 1 a for a ≫ 1
1 6
ρ(a) ≈ 12 , ρe(a) ≈ 18 . (7.50)
1 for a ≪ 1 0 for a ≪ 1

On S 4 the spectrum of  ˚ is completely discrete. As a consequence, the

exact formulas for ρ(a) and ρe(a) display discontinuities at the a-values at
which N (a) jumps. To analyze the RG equation smooth functions ρ(a) and
ρe(a) clearly would be easier to deal with. For this reason it was proposed to
adopt a certain smoothing procedure that replaces the original functions ρ
and ρe by smooth interpolating functions [203].
The way of performing this interpolation, or smoothing, is in no way
unique. Choosing a specific smoothing procedure has the same conceptual
status as choosing a particular cutoff function R(0) : It amounts to specifying
how precisely the transition from the high- to the low-momentum regime
takes place, i.e., how the modes are suppressed when their eigenvalue is below
the threshold value given by k. Observable quantities derived from the RG
flow must be independent of both R(0) and the smoothing procedure.
Denoting the smoothened functions by ρ and ρe again, the numerical anal-
ysis has shown that for most purposes the following approximation suggested
by (7.50) is perfectly sufficient:

ρ(a) = 1 + 1
12 a4 , ρe(a) = 18
1 6
a . (7.51)

(ii) Turning to the case of R4 , with g̊µν = δµν , the spectrum of  ˚ is continuous,
and the functions ρ and ρe are easily evaluated in the plane wave eigenbasis.
One finds that, for all values of a, ρ(a) ∝ a4 and ρe(a) ∝ a6 . Here we omitted
 7.5 An Infinite-Dimensional Truncation 201

a constant factor proportional to the infinite volume of R4 . Choosing a nor-

malization convention consistent with the one for S 4 , the partial differential
equation on R4 becomes

k∂k Yk (a) = (2 + ηN ) Yk (a) − a Yk′ (a)

gk ( ) a6 (7.52)
− 1 − 16 ηN 2 .
24π a + Yk′′ (a)

This equation coincides with the corresponding S 4 result if in the latter

the a ≫ 1-approximations from (7.50) are used for all values of a. Indeed,
for a ≫ 1 the spectral sums of S 4 are dominated by many densely spaced
eigenvalues forming a quasicontinuum.
The difference between R4 and S 4 is most pronounced for small values of a.
There the finite volume of S 4 and the resulting discreteness of the spectrum
strongly modify the spectral density ρ. On flat space it is proportional to a4 ,
while on the sphere it approaches ρ(0) = 1 for a ≪ 1.
For a ≪ 1 the functional trace in the FRGE of S 4 is dominated by a single
eigenvalue, namely the zero mode of . ˚ For the R4 topology, on the other
hand, the ρ ∼ a -behavior extends down to a = 0.
4

(2) Fixed functions. The RG flow on the infinite-dimensional( theory

) space
possesses both a Gaussian and a non-Gaussian fixed point g∗ , Y∗ ( · ) . The “fixed
function” Y∗ ( · ) is determined by the ordinary differential equation one obtains
by setting ∂k Yk ≡ 0.
The GFP is characterized by g∗ = 0 together with the fixed point potential
Y∗ (a) = c a2 with c = 0 for R4 and c = 1 for S 4 .
For the R4 topology, the result for the NGFP turns out to be exactly the
same as we found earlier in the CREH approximation, namely g∗ = g∗CREH ≈
4.650 and Y∗ (a) = y∗ − 61 λ∗ a4 , where λ∗ = λCREH
∗ ≈ 0.279. The constant y∗ is
not determined by the flow equation.
Except for this constant, the infinite-dimensional RG equations have not taken
advantage of the possibility to generalize the fixed point potential beyond the
functional form of the CREH approximation. This robustness under extensions
of the truncation is quite remarkable.
For the S 4 topology the situation is different, and the properties of the NGFP
show significant deviations from the earlier CREH analysis. In this case the fixed
point potential can only be determined numerically. In Figure 7.1 we display the
result obtained with the smooth spectral functions (7.51).

(3) Infinite-dimensional stability analysis. In order to determine the fixed

point’s critical exponents and scaling fields, we insert

gk = g∗ + ε yg e−θt ,
(7.53)
Yk (a) = Y∗ (a) + ε Υ(a) e−θt
202 Conformally Reduced Gravity

–Y*
–Y*
30
105
20
8 ·104
10
6 ·104
a
0 1 2 3 4 5
4 ·104
–10
2 ·104
–20
a
0 10 20 30
–30
–2 ·104

Figure 7.1. The negative fixed point potential −Y∗ (a) for S 4 . The dashed line
is the CREH potential with the same value for λ∗ . The figure on the left-hand
side magnifies the small-a region of the one on the right-hand side; the latter
shows that Y∗ coincides with the CREH potential asymptotically. (Taken from
[203].)

into the RG equations and expand them to(first order ) in ε. The resulting condition
on the allowed values of θ and the pairs yg , Υ( · ) , which generalize the eigen-
vectors of the stability matrix, involves a linear ordinary differential equation.
(See Section 2.2.4 for a similar discussion.)
Focusing on the NGFP and the R4 topology from now on, the main results are
as follows. ( )
There exist only two scaling fields, yg ̸= 0, Υ( · ) , with a non-vanishing g com-
ponent. Their only other non-vanishing component points in the a4 -direction of
the function space {Y ( · )}, which we can identify with the λ-direction of the
CREH truncation. Those two scaling fields are relevant, have complex critical
exponents with θ′ = 4, θ′′ ≈ 6.18, and coincide exactly with those of the CREH
approximation. ( )
All other scaling fields, 0, Υ( · ) , have vanishing g-component and correspond
to perturbations of the potential alone. There are both relevant and irrelevant
fields of this type. Generically they correspond to non-polynomial oscillatory
functions like
′ ′
Yk (a) − Y∗ (a) ∝ an e−θ t cos(n′′ ln a − θ′′ t + δ) , (7.54)

which arise as the real and imaginary parts of complex powers, Υ(a) ∝ an , where
n ≡ n′ + in′′ and, as usual, θ ≡ θ′ + iθ′′ . The eigenvalue condition provides the
relationship θ ≡ θ(n), and moreover it restricts the exponent to be in a certain
domain of the complex plane, n ∈ D ⊂ C.7

7 A graphical representation of D can be found in [203].

 7.5 An Infinite-Dimensional Truncation 203

It is unclear a priori whether one should admit all n ∈ D, or only the sub-
set of, say, all real, or all integer exponents. Depending on which scaling fields
(eigenfunctions) are considered admissible, the number of relevant and irrelevant
directions changes. This affects in particular the dimensionality of the UV-critical
manifold, SUV .
If we were to admit all n ∈ D, we would find infinitely many relevant scaling
fields, i.e., SUV is infinite dimensional. On the other hand, dim(SUV ) is finite if
n is restricted to the integers in D.

(4) Precise specification of theory space. These findings bring us to another

general lesson which is nicely illustrated by the model system at hand.
Namely, at a certain point of the Asymptotic Safety program we must decide
about the mathematically precise specification of theory space, T . Only then can
we distinguish “acceptable” and “unacceptable” action functionals, depending
on whether or not they are elements of T .
Finding the optimum theory space is an involved problem: T should be small
enough to eliminate all actions with physically harmful singular behavior, but at
the same time large enough to contain as many asymptotically safe and complete
RG trajectories as possible. Finding the optimum space will require a certain
understanding of the observables in the theory and their dependence on the
running couplings (which we are lacking for the time being).
At this point the mathematical setting of the linear differential equation for
the scaling fields will also become fully concrete so that one can meaningfully
talk about the number of relevant eigendirections. Loosely speaking, the scaling
fields live in the tangent space to T , a notion that must be made precise in this
step.
Let us consider the following general situation exemplified by the present trun-
cation. Assume that we start out from some preliminary “trial” theory space T0
and, in the ideal case, are able to compute all RG trajectories on it. We then
discover that there exists a subspace T ′ ⊂ T0 which is left invariant by the flow,
i.e., if a trajectory begins on T ′ it will stay on it for ever. If T ′ furthermore
contains the complete and asymptotically safe trajectories we are after, it is then
possible to declare T ′ the actual theory space.
Narrowing down the set of eligible actions from T0 to T ′ can, but not necessarily
must, lead to enhanced predictive power of the theory constructed. The decisive
question is whether there is a conclusive physical reason for excluding the actions
in T0 \ T ′ .
In our example, the space of all potentials, {Y ( · )}, plays the role of T0 . It
contains functions that are indeed never generated by the flow if they are absent
initially. They include potentials that grow faster than Y (a) ∝ a4 for a → ∞, or
examples like Y (a) ∝ a−p , p = 1, 2, 3, · · · , which are singular at a = 0.
In [203] various possible restrictions of the function spaces {Y ( · )} and {Υ( · )}
at the linearized level were considered as examples of T ′ . They lead to the
204 Conformally Reduced Gravity

differing results for dim(SUV ) which we mentioned in (3) above. We will not
go into the details here. Suffice it to say that no compelling physical arguments
preferring one definition of theory space over the other are known for the time
being.

(5) Phases of unbroken diffeomorphism invariance. If one numerically

solves the non-linear RG equations one can distinguish two essentially different
classes of potentials Uk (ϕ) along the RG trajectories, namely potentials having
their minimum at ϕ ̸= 0 or ϕ = 0, respectively.
Recalling that gµν ≡ ⟨bgµν ⟩ = ϕ2 δµν , it is tempting to interpret ϕ as an “order
parameter” whose vanishing (non-vanishing) expectation value indicates a phase
of gravity with unbroken (broken) diffeomorphism invariance. If one lowers k,
“phase transitions” from one regime to the other are observed. They can be of
both first- and second-order type [203].

7.6 Bi-Metric Truncations and the Split Ward Identity

The truncated actions for conformally reduced gravity which we considered in
the previous sections all depend on φ and ϕ̄ via the combination φ + ϕ̄ ≡ ϕ only.
In the (g, ḡ)-, or here, (ϕ, ϕ̄)-language, the functional Γk [ϕ, ϕ̄] ≡ Γk [φ; ϕ̄] φ=ϕ−ϕ̄ is
completely independent of ϕ̄. It thus amounts to the conformally reduced variant
of a single-metric truncation.
As a first orientation beyond the single-metric level the following “bi-field”
ansatz was considered in [90]:8
∫ { } ∫
3 √ 1 1 √
Γk [ϕ, ϕ̄] = + 4
d x g̊ Dyn
ϕ  ϕ + B ϕ̄  ϕ̄ + d4 x g̊ Uk (ϕ, ϕ̄).
˚ ˚
8π Gk Gk
(7.55)

This action leaves room for an “extra” background field dependence by an inde-
pendent kinetic term for ϕ̄ and its appearance in the potential. It covers (the
conformally reduced version of) the bi-metric Einstein–Hilbert truncation (6.9)
in which case
ΛDyn ΛB
Uk (ϕ, ϕ̄) = k
Dyn
ϕ4 + k
ϕ̄4 . (7.56)
8πGk 8πGBk

(We specialize for the R4 topology in this section.)

The FRGE projected on the ansatz (7.55) yields two independent anomalous
Dyn
dimensions ηN B
, ηN related to gkDyn ≡ k 2 GDyn
k and gkB ≡ k 2 GBk , respectively, as
well as a partial differential equation for the potential Yk (a, b) ≡ − 4π3 gk Uk (ϕ, ϕ̄).

8 In [75, 236, 237] an earlier, similar calculation with two independent conformal factors in a
perturbatively renormalizable gravity model can be found.
 7.6 Bi-Metric Truncations and the Split Ward Identity 205

Here a ≡ kϕ and b ≡ k ϕ̄ denote the dimensionless field variables. The partial

differential equation reads explicitly:

( Dyn ) g Dyn ( η Dyn ) b6

k∂k + a∂a + b∂b − ηN − 2 Yk (a, b) = − k 1− N 2 2
.
24π 6 b + ∂a Yk (a, b)
(7.57)

The solutions of the coupled equations have been studied both within
a( finite-dimensional ) “subtruncation” of Uk and on the infinite-dimensional
g , g , Y ( · , · ) -theory space. Thereby the essential qualitative results of the
Dyn B

earlier analysis could be confirmed. A detailed discussion can be found in [90].

More interestingly, within the same setting also a first analysis of the Ward
identity for split symmetry (4.96), has been possible [90]. In conformally reduced
gravity it assumes the form:
[( )−1 δ ]
δ 1 (2)
Γk [ϕ, ϕ̄] = Tr Γk [ϕ, ϕ̄] + Rk [ϕ̄] Rk [ϕ̄] . (7.58)
δ ϕ̄(x) 2 δ ϕ̄(x)

For the truncation (7.55), at the dimensionless level, this Ward identity boils
down to:
∂ g Dyn b6
b Yk (a, b) = − k 2 2
. (7.59)
∂b 24π b + ∂a Yk (a, b)
While its structure is similar to the RG equation (7.57), the Ward identity is
seen to be an “instantaneous” relation, i.e., it involves no scale derivative.
If it was not for the truncation, the Ward identity would hold true at any point
of an RG trajectory if it holds at one of its points. Hence, the mutual consistency
of the truncated Ward identity and the flow equation, both considered at the
same level of accuracy, can shed light on the reliability of the truncation.
As the RG trajectories are approximate common solutions of both the flow
equation and the Ward identity, the same should be true for any combination
thereof. It is therefore tempting to combine (7.57) and (7.59) in such a way that
the complicated term involving ∂a2 Yk (a, b) in the denominator disappears [90]:
[ ( ) ]
Dyn 1 Dyn
k∂k − 2 + ηN + a∂a + ηN b∂b Yk (a, b) = 0. (7.60)
6

It is particularly instructive to consider (7.60) at the fixed points, ∂k Yk = 0. At

the GFP we have g∗Dyn = 0 and ηN Dyn
= 0. Hence, (7.59) tells us that b∂b Y∗ (a, b) = 0
and (7.60) yields [a∂a − 2] Y∗ (a, b) = 0. Taken together these relations imply that

Y∗GFP (a, b) = c a2 , (7.61)

where c is a constant. Note that this fixed point potential depends only on the
dynamical field a, not on the background.
206 Conformally Reduced Gravity

Dyn
At a NGFP, instead, one has ηN = −2, and (7.60) boils down to
[ ]
1
a∂a − b∂b Y∗NGFP (a, b) = 0. (7.62)
3

This equation when taken at face value would imply that Y∗NGFP is a function of
a single variable, b3 a. However, there exists no exact fixed point solution to the
flow equation (7.57) of the form Y∗NGFP (a, b) = f (b3 a).
These examples illustrate a general difficulty: As the flow equation and the
Ward identity are the result of approximations, we should be prepared to find
spurious contradictions or inconsistencies if we commit the mistake and consider
them true exactly. Clearly it is somewhat subtle to meaningfully talk about
“approximate consistency,” and so one must resort to a detailed case-by-case
analysis.
The qualitative conclusion we can draw from (7.62) is that, at a NGFP, the
fixed point potential Y∗NGFP is likely to involve both the dynamical field a and the
background b, with a clear trend for the background field to be predominant. In
fact, if we change the normalization of Rk and replace 1/GDynk in its prefactor by
Dyn
a constant, the ηN /6 term disappears from the flow equation and from (7.60).
As a consequence, (7.62) gets replaced by a∂a Y∗ (a, b) = 0, implying that Y∗NGFP
is a function of b alone!
As we emphasized above, we are dealing with approximate solutions to approx-
imate equations so that this result cannot be taken at face value. Yet, it highlights
the importance of the background field for the structure of the fixed point action.9
Further quantitative tests of the compatibility between the two equations com-
pared the values of the fixed point coordinates as predicted by (7.57) and (7.59),
respectively, within a finite-dimensional subtruncation. In this context, indica-
tions of a non-trivial “conspiracy” between the two equations have indeed been
observed.
For further details concerning the CREH model we must refer to the literature
[90]. Related work with a somewhat different scope includes a comprehensive con-
ceptual discussion and explicit analysis of the RG flow in Weyl-invariant theories
[239–241] and an investigation of the background-scale Ward identity [201].

9 Further investigations along these lines can be found in [238].

 8
The Reconstruction Problem

This chapter is devoted to the so-called reconstruction problem, i.e., the problem
of finding a UV regularized functional integral, which, first, has a mathematically
well-defined continuum limit and, second, reproduces a given RG trajectory, a
complete solution to the UV-cutoff-free FRGE.

8.1 In Search of a Bare Theory

In order to put the reconstruction problem into a proper perspective, let us
recall that we started out from an ill-defined functional integral that is plagued
by short-distance singularities. We equipped this integral with a UV regulator,
then associated a FRGE to the regularized integral, and finally observed that
in the flow equation the regulator is actually unnecessary and can be removed
almost trivially.

(1) After this last step, our efforts toward a quantum theory of gravity com-
pletely decoupled from the initial, and rather formal, functional integral. While
we did employ the integral as inspiration for a promising theory space and for
the structure of the FRGE, our actual calculations were all based on the UV-
cutoff-free flow equation. We accepted this equation as the starting point of all
mathematically rigorous developments, because it replaces the problem of giving
a meaning to the functional integral by a new one which, we believe, is easier to
handle, namely finding fully extended RG trajectories {Γk }k∈[0,∞) .
The consequence is that, even if we have a complete RG trajectory at our
disposal, we are unable to specify a regularized functional integral without
additional input. The information that is missing consists of a prescription for
regularizing the measure along with the bare action, Db g → DΛ gb, S → SΛ , and a
rule for sending Λ → ∞ in ∫
DΛ gb e−SΛ [bg] , (8.1)
208 The Reconstruction Problem

which guarantees a meaningful limit. The integral (8.1) depends on a number of

Λ-dependent “bare” parameters, and the rule would include information about
how one must finetune the Λ-dependence in order to achieve this limit.

(2) At this point the decision to base our search for consistent theories within
the Asymptotic Safety program on a variant of the effective action, namely the
EAA, rather than a running bare action becomes significant.
The latter option would seem natural in the Wilson–Kadanoff renormalization
group describing a discrete sequence of successive coarse-graining steps consisting
of a block-spin transformation, for instance [7–9]. For continuum-based gravity
and gauge theories we rejected this route as it entails profound difficulties related
to gauge invariance.
Note that a true Wilsonian action, SΛW , has the status of a bare action, i.e.,
it is meant to be used under a regularized path integral. It depends on the UV
cutoff, and the dependence on Λ is governed by an RG equation which differs
from the one for the EAA both conceptually and structurally.
There is indeed a considerable difference between the renormalization groups
behind the EAA and the Wilsonian approach:
In a sense, SΛW for different values of Λ is a set of actions for the same system:
the Green functions have to be computed from SΛW by a further functional inte-
gration over the low momentum modes, and this integration renders them inde-
pendent of Λ. By contrast, the EAA can be thought of as the standard effective
action for a family of different systems, namely those with the bare actions SΛ +
∆Sk labeled by k ∈ [0, Λ). The corresponding n-point functions, computed simply
as functional derivatives of Γk without any further integration, are scale depen-
dent and constitute an effective field theory description valid near the scale k [31].
Also these general differences between the EAA and a genuine Wilson action
make it quite clear that our approach cannot by itself imply a regularized path
integral.

(3) At first sight one might think that the functional integral is dispensable and
that a complete RG trajectory Γk suffices to extract all properties of interest
from the corresponding quantum field theory; after all, its n-point functions are
all provided by the derivatives of Γk at k = 0.
Nevertheless, there are a number of reasons why the presentation of the asymp-
totically safe theory as a bona fide functional integral is desirable.

(a) Working with the EAA alone we have no information about the micro-
scopic (or “classical”) system whose standard quantization would give rise to this
particular effective action. The functional integral representation of an asymp-
totically safe theory will allow the reconstruction of the microscopic degrees of
freedom that we implicitly integrated out in solving the FRGE as well as their
fundamental interactions.
 8.1 In Search of a Bare Theory 209

Based on the action resulting from the path integral we can reconstruct the
Hamiltonian description by a kind of generalized Legendre transformation. From
this phase space formulation we can in turn read off a classical dynamical system
whose quantization (also by other methods, e.g., canonically [242]) leads to the
given effective action. We expect that this classical system is rather complicated
so that it may not be guessed easily.

(b) Many physically interesting observables involve composite operators which

do not appear in Γk , either because they are not even elements of the exact theory
space, or they are absent as the consequence of a truncation. Typical examples
include geometrically motivated length or volume operators [229]. Their expec-
tation values and correlation functions cannot be inferred from the usual EAA.
However, a functional integral representation would allow us to extend the scope
of the EAA and its renormalization group beyond the correlators of the elemen-
tary quantum fields. (See [229, 243] for a detailed discussion.)

(c) Ultimately we would like to understand the relation of QEG with other
approaches to quantum gravity, such as canonical quantization [242], Loop Quan-
tum Gravity [36–39], or Monte Carlo simulations of statistical systems [244], for
example. In these approaches the bare action plays a central role.
In the Monte Carlo simulations of the Regge or dynamical triangulations for-
mulation, for instance, the starting point is a regularized path integral involving
some discretized version of SΛ . In order to take the continuum limit one must
finetune the bare parameters in SΛ in a suitable way. If one is interested in the
asymptotic scaling, for instance, and wants to compare the analytic QEG pre-
dictions to the way the continuum is approached in the simulations, one must
convert the Γk -trajectory to a SΛ -trajectory first.
The map from Γk to SΛ depends explicitly on how the path integral is dis-
cretized; so each alternative formulation of QEG associates its own regularized
measure and bare action SΛ to one and the same Γk -trajectory.

(4) It should be clear from these remarks that the reconstruction step of the
Asymptotic Safety program must begin with a decision about, first, the nature
of the microscopic variables integrated or summed over, and second, a UV
regularization of the corresponding functional integral or partition function.
The sought-for map relating effective and bare couplings will strongly depend
on these two choices.
In the sequel of this chapter we illustrate these ideas using the specific example
in which the microscopic variables are chosen to be continuum fields gbµν (x) and
the UV regularization is realized by a sharp mode cutoff.
We show how to find a corresponding regularized functional integral which
reproduces a given asymptotically safe RG trajectory {Γk , 0 ≤ k < ∞}, a solu-
tion of the FRGE without UV regulator. In particular, we determine how the
210 The Reconstruction Problem

bare couplings parameterizing SΛ must behave for Λ → ∞ in order to yield a

continuum limit in accordance with the fixed point of the EAA.
Our presentation follows the work in [205, 206].

8.2 The EAA in the Presence of a UV Cutoff

As a preparation we briefly revisit the derivation of the EAA and its flow equation
from the functional integral, paying special attention to the regularization in the
UV now. In order to make the integral mathematically well defined we install a
UV regulator at a cutoff scale Λ and then define an Effective Average Action on
the basis of the regularized integral. To indicate the presence of the UV cutoff
we denote it Γk,Λ .
Many regularization schemes are conceivable here. For concreteness we use a
“finite-mode regularization,” which will also be ideally suited for implementing
Background Independence in QEG.
To avoid inessential complications we first consider a simple scalar field on flat
space as we did in Chapter 2.

b
(1) So, let ϕ(x) be a real scalar matter field on a flat d-dimensional Euclidean
spacetime. In order to discretize momentum space we compactify it to a d-
torus. The eigenfunctions of the Laplacian p̂2 ≡ −δ µν ∂µ ∂ν are plane waves
up (x) ∝ exp (ip · x) with discrete momenta pµ and eigenvalues p2 . For√a given
cutoff scale Λ there are only finitely many eigenfunctions with |p| ≡ p2 ≤ Λ.
We regularize the path integral in the UV by restricting the integration to those
modes. Therefore, the field ϕb and its source J have expansions:
∑ ∑
b =
ϕ(x) αp up (x) and J(x) = Jp up (x). (8.2)
|p|∈[0,Λ] |p|∈[0,Λ]

We define a UV-regularized generating functional:

∫ ( ∫ )
( )
b b b d b
exp Wk,Λ [J] ≡ DΛ ϕ exp −SΛ [ϕ] − ∆Sk [ϕ] + d x J(x)ϕ(x) . (8.3)

The notation in (8.3) is symbolic. Its right-hand side involves only finitely many
integrations and thus is not a genuine functional integral. The measure DΛ ϕb
stands for an integration over the Fourier coefficients αp with p2 below Λ2 :
∫ ∏ ∫ ∞
b
DΛ ϕ = dαp M −[αp ] . (8.4)
|p|∈[0,Λ] −∞

The arbitrary mass parameter M was introduced in order to give the canonical
dimension zero to (8.4). Furthermore, as always in the EAA construction, the IR
modes with∫|p| < k are suppressed in Wk,Λ by means of using the cutoff action
∆Sk [ϕ] b
b = 1 dd x ϕ(x)R 2 b
k (p̂ )ϕ(x).
2
 8.2 The EAA in the Presence of a UV Cutoff 211

In (8.3) the bare action SΛ is allowed to depend on the UV cutoff. Ultimately

we would like to fix this Λ-dependence in such a way that, for every finite k and
J, the path integral has a well-defined limit for Λ → ∞.
We introduce the UV regular EAA as
∫
e k,Λ [ϕ] − 1 dd x ϕ(x)Rk (p̂2 )ϕ(x),
Γk,Λ [ϕ] ≡ Γ (8.5)
2

e k,Λ [ϕ] is the Legendre transform of Wk,Λ [J] with respect to the finitely
where Γ
b obtained
many variables of J and ϕ = {ϕp }|p|∈[0,Λ] is the expectation value ϕ ≡ ⟨ϕ⟩
by differentiating Wk,Λ [J] with respect to the source J, cf. Section 2.1.

(2) Repeating the steps described in Section 2.1 one finds that Γk,Λ satisfies the
following exact FRGE:
[( )−1 ]
1 (2)
k∂k Γk,Λ [ϕ] = TrΛ Γk,Λ [ϕ] + Rk k∂k Rk . (8.6)
2

Here, TrΛ denotes the trace restricted to the subspace spanned by the eigenfunc-
tions of p̂2 with eigenvalues smaller than Λ2 :
[ ]
TrΛ [ · · · ] = Tr Θ(Λ2 − p̂2 )[ · · · ] . (8.7)

The UV-regularized EAA satisfies the integro-differential equation

∫ ( ∫
( ) δΓk,Λ [ϕ]
exp −Γk,Λ [ϕ] = DΛ f exp − SΛ [ϕ + f ] + dd x f (x)
δϕ(x)
∫ ) (8.8)
1
− dd x f (x)Rk (p̂2 )f (x) ,
2

b
with the measure (8.4) and the fluctuation field f (x) ≡ ϕ(x) − ϕ(x). Equation
(8.8) will be instrumental in the next section.

(3) The question arises about whether the UV cutoff can be removed from the
FRGE. Indeed, for this to be possible, it is sufficient to assume that the IR-cutoff
operator is chosen such that k∂k Rk (p2 ) decreases rapidly for large p2 so that the
trace in the flow equation (8.6) does not receive contributions from eigenvalues
p2 ≫ k 2 and hence continues to exist in the limit Λ → ∞. As a result, the “Λ-free”
FRGE without UV cutoff, valid for all k ≥ 0, takes the familiar form:
[( )−1 ]
1 (2)
k∂k Γk [ϕ] = Tr Γk [ϕ] + Rk (p̂ ) 2
k∂k Rk (p̂ ) .
2
(8.9)
2

We denote the solutions of (8.9) as {Γk , 0 ≤ k < ∞} and those of the FRGE (8.6)
with UV cutoff as {Γk,Λ , 0 ≤ k < Λ}.
212 The Reconstruction Problem

k=∞

ΓΛ = ΓΛ, Λ
k=Λ

Incomplete trajectory Complete trajectory

{Γk, Λ, 0 ≤ k < Λ} {Γk, 0 ≤ k < ∞}

k = k1

k=0

Figure 8.1. With the optimized cutoff, every complete solution to the Λ-free
FRGE gives rise to a solution of the FRGE with UV cutoff, valid up to any
value of Λ.

(4) Remarkably, the flow equations for Γk and Γk,Λ are essentially identical
as long as k ≪ Λ [205]. Generically, when k approaches Λ from below, there
exist some small corrections due to the UV cutoff which affect Γk,Λ and cause
it to differ from Γk . However, it is always possible to choose a special IR cutoff
Rk (p2 ) such that those corrections vanish, an example being the optimized cutoff
Rk (p2 ) = (k 2 − p2 )Θ(k 2 − p2 ).
As a result, the functional Γk,Λ satisfies the same FRGE as Γk but is defined in
the interval k ≤ Λ only. For identical initial conditions, the relationship between
the respective solutions of the two flow equations is thus quite simple:

Γk,Λ = Γk when 0 ≤ k ≤ Λ. (8.10)

For a fixed, but arbitrary, finite scale Λ, {Γk,Λ , 0 ≤ k < Λ} is the restriction of the
complete solution {Γk , 0 ≤ k < ∞} to the interval k < Λ. Thus, sending Λ → ∞
in (8.10) is a trivial step. The situation is illustrated in Figure 8.1.

8.3 A One-Loop Reconstruction Formula

The problem we want to address now is how we can determine the Λ-dependence
of the bare action SΛ , given some complete solution of the Λ-free flow equation,
{Γk , k ∈ [0, ∞)}. According to (8.10), this complete solution implies a solution
 8.3 A One-Loop Reconstruction Formula 213

with a UV cutoff: {Γk,Λ , k ∈ [0, Λ]}. Setting k = Λ we have in particular ΓΛ,Λ = ΓΛ

or, more explicitly,
Γk=Λ,Λ = Γk=Λ . (8.11)
Thus, knowing Γk for all k means that we know ΓΛ,Λ for all Λ.
Next, we explain how the bare action SΛ can be found once Γk is given. Setting
k = Λ we know ΓΛ,Λ = ΓΛ and since Γk is a solution for all values of k, we may
assume that ΓΛ,Λ is known for all values of Λ.

(1) Using (8.8) we can obtain the desired relation between Γk and SΛ . For this
purpose we evaluate the functional integral on the right-hand side of (8.8) by a
saddle point expansion. Let the fluctuation field be f (x) ≡ f0 (x) + h(x) where f0
is the stationary point of the total action:
∫ ∫
δΓk,Λ [ϕ] 1
Stot [f ; ϕ] ≡ SΛ [ϕ + f ] − d xf (x)
d
+ dd xf (x)Rk (p̂2 )f (x). (8.12)
δϕ(x) 2

Now, expanding Stot to second order in h and performing the Gaussian integral
over h we obtain the following relationship between the bare and the Effective
Average Action:
∫ ∫
δΓk,Λ [ϕ] 1
Γk,Λ [ϕ] = SΛ [ϕ + f0 ] − dd xf0 + dd xf0 Rk f0
δϕ 2
[( 2 ) ] (8.13)
1 δ SΛ [ϕ + f0 ] −2
+ TrΛ ln + Rk M + · · · .
2 δϕ2

Noting that also the stationary point f0 has an expansion in powers of the loop-
counting parameter ~, (8.13) yields, in a slightly symbolic notation:

Γk,Λ [ϕ] − SΛ [ϕ]

∫ ∫
δ ( ) 1 (
(2)
)
= − f0 Γk,Λ − SΛ [ϕ] + f0 SΛ [ϕ] + Rk f0
δϕ 2
+ O(f03 ) (8.14)
~ {[ ] }
+ TrΛ ln SΛ [ϕ] + SΛ [ϕ]f0 + SΛ [ϕ]f0 f0 + · · · + Rk M −2
(2) (3) (4)
2
+ O(~2 ) .

Together with the expansion of the stationarity condition (δStot /δf )[f0 ] = 0 the
above equation is solved self-consistently if f0 = 0 + O(~) and Γk,Λ [ϕ] − SΛ [ϕ] =
O(~). This leads to the following one-loop formula for the difference between the
average and the bare action:

~ {[ ] }
TrΛ ln SΛ [ϕ] + Rk M −2 .
(2)
Γk,Λ [ϕ] − SΛ [ϕ] = (8.15)
2
214 The Reconstruction Problem

Setting k = Λ we arrive at the final result:

~ {[ ] }
TrΛ ln SΛ [ϕ] + RΛ M −2 .
(2)
ΓΛ,Λ [ϕ] − SΛ [ϕ] = (8.16)
2

This is a differential equation which determines SΛ . It tells us how the bare action
SΛ must depend on Λ in order to give rise to the prescribed ΓΛ,Λ . It will be our
main tool for finding the path integral that belongs to a known solution of the
FRGE.

(2) A comment concerning the mass parameter M might be in order here. Even
though M was introduced in (8.4) only in order to make the measure dimension-
less, it has a non-trivial impact on the solution of (8.16) for the bare action SΛ .
Indeed, different choices of M can lead to quite different actions, but all of them
are physically equivalent. In a sense, changing M amounts to shifting contribu-
tions from the measure into ∫ the bare action and vice versa. This illustrates the
general fact that neither DΛ ϕb nor e−SΛ have a physical meaning separately,
only their combination has. (See [205] for a detailed discussion.)

(3) In QEG, the analogous construction begins with the UV-regularized func-
tional integral:
∫ ( )
DΛ b
h DΛ C DΛ C̄ exp −SeΛ [b
h, C, C̄; ḡ] − ∆Sk [b
h, C, C̄; ḡ] . (8.17)

Here the total bare action SeΛ ≡ SΛ + Sgf,Λ + Sgh,Λ depends on Λ and includes
the gauge-fixing term Sgf,Λ and the ghost action Sgh,Λ . The UV cutoff is imple-
mented by restricting the expansion of (b
h, C, C̄) in terms of eigenfunctions of the
background-covariant Laplacian −D̄ ≡ −ḡ µν D̄µ D̄ν to those with eigenvalues κ
2

smaller than a given Λ2 . Hence, the measure reads in analogy with (8.4),
∫ ∏ ∏∫ ∞
b
DΛ h = dhκm M −[hκm ] , (8.18)
κ∈[0,Λ2 ] m −∞

and likewise for the ghosts. Here, m is a degeneracy index, and hκm are the
expansion coefficients of b hµν (x).
Note that at this point the background metric ḡµν (x) is instrumental not only
for the IR cutoff and the gauge fixing but also for implementing the UV cutoff
in a δ B -covariant fashion.
The remaining steps are analogous to the scalar case. One considers the UV-
regular EAA Γk,Λ [h, ξ, ξ;¯ ḡ] pertaining to the integral (8.17) and shows that it
satisfies a FRGE of the form (8.6), albeit with a modified supertrace that carries
an explicit ḡ-dependence:
[ ]
STrΛ [ · · · ] ≡ STr Θ(Λ2 + D̄2 )[ · · · ] . (8.19)
 8.4 The Twofold Einstein–Hilbert Truncation 215

¯ ḡ], we establish
Denoting, as usual, solutions to the Λ-free FRGE as Γk [h, ξ, ξ;
that at k = Λ → ∞, and for an appropriate Rk ,
¯ ḡ] = ΓΛ [h, ξ, ξ;
ΓΛ,Λ [h, ξ, ξ; ¯ ḡ], (8.20)
and that the latter functional is related to the total bare action by
[( ) ]
ΓΛ,Λ [h, ξ, ξ; ¯ ḡ] + 1 STrΛ ln Se(2) + RΛ [h, ξ, ξ;
¯ ḡ] = SeΛ [h, ξ, ξ; ¯ ḡ] N −1 .
Λ
2
(8.21)

Here N is a block diagonal normalization matrix, equal to M d and M 2 in the

graviton and the ghost sector, respectively.

8.4 The Twofold Einstein–Hilbert Truncation

Solving the functional differential equation (8.21) for SeΛ [h, ξ, ξ;
¯ ḡ] is clearly diffi-
cult. In practice one has to restrict the spaces of actions from which Γk and SeΛ
are drawn by truncating them to a tractable number of invariants. The simplest
possibility, which we discuss now, is to employ the Einstein–Hilbert truncation
for both the effective and the bare action.
As in Section 5.2 we make the ansatz
∫
¯ =− 1 √ ( )
¯ ḡ]
Γk [g, ḡ, ξ, ξ] dd x g R(g) − 2λ̄k + Sgh [g − ḡ, ξ, ξ;
16πGk
∫ (8.22)
1 √ ( )( )
+ dd x ḡ ḡ µν Fµαβ gαβ Fνρσ gρσ .
32πGk
The last term of (8.22) is the gauge-fixing action, with α = 1, for the har-
monic coordinate condition, involving Fµαβ ≡ δµβ ḡ αγ D̄γ − 12 ḡ αβ D̄µ . Furthermore,
we make an analogous ansatz for the bare action:
∫
1 √ ( ˇ ) + S [g − ḡ, ξ, ξ;
SeΛ [g, ḡ, ξ, ξ]
¯ =− dd x g R(g) − 2λ̄ Λ gh
¯ ḡ]
16π ǦΛ
∫ (8.23)
1 √ ( )( )
+ dd x ḡ ḡ µν Fµαβ gαβ Fνρσ gρσ .
32π ǦΛ
The EAA (8.22) contains the familiar running parameters Gk and λ̄k . The cor-
responding bare Newton and cosmological constant are denoted by ǦΛ and λ̄ˇ ,
Λ
respectively.

(1) Computing the Hessians and setting ξ = ξ¯ = 0 and ḡ = g afterwards, the

supertrace in (8.21) has a derivative expansion of the form
1 [( ) ]
STrΛ ln SeΛ + RΛ [0, 0, 0; ḡ] N −1
(2)
2 ∫ ∫ (8.24)
√ √
= B0 Λd dd x g + B1 Λd−2 dd x gR(g) + · · · .
216 The Reconstruction Problem

The dimensionless coefficients B0 and B1 can be evaluated with the same heat
kernel methods used in Chapter 5. In d = 4 one finds [205]:

1 [ ]
B0 = 5 ln(1 − 2λ̌Λ ) − 5 ln(ǧΛ ) + Q Λ ,
32π 2
1
B1 = B0 + ∆B1 ,
3
1 2 − λ̌Λ
∆B1 ≡ , (8.25)
16π 2 1 − 2λ̌Λ
(Λ)
QΛ ≡ 12 ln + b0 ,
M
b0 ≡ −5 ln(32π) − ln 2 .

Using (8.24) in (8.21) and equating the coefficients of the independent

invariants we obtain two equations relating the effective to the bare parameters:

1 1 λ̄Λ ˇ
λ̄Λ
− = −16πB1 Λd−2 , − = 8πB0 Λd . (8.26)
GΛ ǦΛ GΛ ǦΛ

In terms of the dimensionless quantities defined by gΛ ≡ Λd−2 GΛ , ǧΛ ≡ Λd−2 ǦΛ ,

and analogously for the two cosmological constants, we get

1 1 λΛ λ̌
− = −16πB1 , − = 8πB0 . (8.27)
gΛ ǧΛ gΛ ǧΛ

This algebraic system of equations contains the information we are interested in.
It allows us to determine ǧΛ and λ̌Λ for given gΛ and λΛ .

(2) While it is impossible to solve the system (8.27) analytically, the numerical
analysis performed in [205] established a well-defined map (g, λ) 7→ (ǧ, λ̌) for all
g > 0 and λ < 12 . The analysis covered a wide range of values for the constant
Q = 12 ln c + b0 which labels different choices for the measure.1

(3) By this map, the fixed point behavior limk→∞ (gk , λk ) = (g∗ , λ∗ ) on the effec-
tive side is mapped onto an analogous fixed point behavior at the bare level. The
image of the GFP is always at ǧ∗ = λ̌∗ = 0, while the coordinates of the “bare”
NGFP, ǧ∗ and λ̌∗ , depend on the value of Q.

1 Initially the map (g, λ) 7→ (ǧ, λ̌) is explicitly Λ-dependent because of the parameter QΛ ≡
Λ
12 ln( M ) + b0 . This Λ-dependence can be removed by including appropriate factors of the
UV cutoff into the measure. In the present discussion we set M = Λ c
with an arbitrary c > 0,
hence the quantity Q ≡ QcM = 12 ln c + b0 becomes a Λ-independent constant. As a result,
the map (g, λ) 7→ (ǧ, λ̌) has no explicit dependence on any (UV or IR) cutoff.
 8.4 The Twofold Einstein–Hilbert Truncation 217

g ǧ
1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

l lˇ
–0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4 –0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4
(a) (b)
ǧ ǧ
1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

lˇ lˇ
–0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4 –0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4
(c) (d)

Figure 8.2. The diagram (a) shows the phase portrait of the effective RG flow
on the (g, λ)-plane. The other diagrams are its image on the (ǧ, λ̌)-plane of
bare parameters for three different values of the parameter Q characterizing the
measure, namely (b) Q = +1, (c) Q = −0.1167 where λ̌∗ = 0, and (d) Q = −1,
respectively. (Adapted from [205].)

These features are illustrated in Figure 8.2, where we apply the map (g, λ) 7→
(ǧ, λ̌) for different values of Q to a set of representative effective RG trajectories
on the half-plane g > 0.
We emphasize again that all choices of Q are physically equivalent. Varying Q
simply amounts to shifting contributions back and forth between the action and
the measure.

(4) It is instructive to determine the linearized flow near the two “bare” fixed
points and to determine the corresponding critical exponents, if such exponents
can be defined.
Both the “effective” and the “bare” NGFP are inner points of the correspond-
ing coupling constant space. The flow in the vicinity of one is the diffeomorphic
image of the flow near the other. The RG running of the respective scaling fields
is ∝ k −θ and ∝ Λ−θ , respectively, with the same critical exponents θ.
The “effective” GFP shows the familiar pure power-law scaling (5.102). But
the “bare” GFP is located on the line ǧ = 0, i.e., on the boundary of the domain
on which the map from the effective to the bare couplings is defined. In its
218 The Reconstruction Problem

vicinity (on the half-plane with ǧ > 0) the “bare” running is characterized by
logarithmically corrected power laws. This is seen by expanding the relations
(8.27) near the GFP:

ǧ = g + O(g 2 , λ2 ),
g ( ) gλ [ ] (8.28)
λ̌ = λ − Q − 5 ln g + 3 − Q + 5 ln g + O(g 2 , λ2 ).
4π 6π
These are the first few terms of a power-log series. Thus, contrary to the param-
eters in the EAA, the running of the bare parameters follows logarithmically
corrected power laws when Λ → ∞.

8.5 Further Remarks

We have only discussed some first steps toward solving the reconstruction prob-
lem for asymptotically safe theories. Yet it has become clear that, after specifying
a UV-regularization scheme and a measure, every solution of the flow equation
for the Effective Average Action without a UV cutoff gives rise to a regular-
ized path integral with a well-defined limit Λ → ∞. We close with a number of
remarks.

(1) The basic idea behind the reconstruction step is to complete the Asymptotic
Safety program by identifying a classical system, which, in a hitherto hidden way,
got quantized by picking a certain RG trajectory of the EAA which extends from
the NGFP in the UV to the IR.
The advantage of the strategy we adopted, selecting a theory on the basis of
its effective rather than bare action, is that in this manner it is easier to see that
it has indeed an “asymptotically safe” short-distance behavior since the effective
action is already closer to the observables. The consequence is that extra work
and further structural input is needed to obtain the eligible microscopic systems.

(2) Once we have a (regularized) functional with a well-defined continuum limit

in our hands, involving a particular bare action, we can attempt a kind of “Leg-
endre transformation” to find appropriate phase space variables, a microscopic
Hamiltonian, and thus a canonical description of the bare theory. Only at this
level we can identify the degrees of freedom that got quantized as well as their
fundamental interactions.
Since the Hamiltonian is unlikely to turn out quadratic in the momenta, the
“Legendre transformation” involved is to be understood as a generalized, i.e.,
quantum mechanical one. In the simplest ∫ case (it consists
) in reformulating a
given configuration space path integral DΦ b exp iS[Φ]
b as a phase space inte-
∫ ∫ ( ∫ )
gral DΦ b DΠ b exp i ΠbΦḃ − H[Π,
b Φ]
b . In a way, we must undo the integration
over the momenta.
 8.5 Further Remarks 219

However, given the complexity of the fixed point action which most probably
contains higher derivative terms, a generalized Ostrogradsky-type phase space
formalism [245, 246] will presumably emerge. In the same spirit, non-localities
also suggest introducing additional auxiliary fields that render the action local,
thereby further enlarging the phase space.

(3) Being interested in a canonical description of the bare fixed point action, one
might wonder if there exists an alternative to the EAA approach that would deal
directly with the RG flow of Hamiltonians rather than Lagrangians. Here the
problem is that, if we apply a coarse-graining step to an action which contains
only, say, first derivatives of the field, the result will also contain higher deriva-
tives in general. This poses no special problem in a Lagrangian setting, but for the
Hamiltonian formalism it implies that new momentum variables must be intro-
duced. As a result, the coarse-grained Hamiltonian “lives” on a different phase
space (in the sense of Ostrogradsky’s method) than the original one. Therefore,
there exists no obvious Hamiltonian analog of the flow on the space of actions.
A way to bypass this obstacle could be to abandon the spacetime averaging
in favor of a purely spatial averaging. In this way Hamiltonian RG flows that do
not generate higher-order time derivatives are conceivable.
 9
Alternative Field Variables

In the previous chapters the field carrying the gravitational degrees of freedom
was chosen to be a symmetric tensor gµν . However, it is well known that General
Relativity admits many reformulations which employ different field variables,
such as tetrads and/or spin connections, for example. Every such reformulation
is characterized by a specific set of fields and a certain covariance group act-
ing on them. This suggests considering a theory space comprised of functionals
which depend on this new set of field variables and are invariant under the cor-
responding group of transformations. Since classical equivalence of theories does
not imply their quantum equivalence, the RG flow on the new space may or may
not be physically equivalent to that of metric gravity. Searching for asymptoti-
cally safe theories of gravity it is therefore important to explore such alternative
settings as well. This is the topic of the present chapter.
Furthermore, elementary field variables different from those considered so far
can even occur within the metric approach when particular parametrizations
of gµν are invoked. The example we are going to discuss here is the ADM
decomposition of the metric, which suggests considering invariant functionals
of a three-metric together with the lapse and shift functions.

9.1 Quantum Einstein–Cartan Gravity

In General Relativity there exists a rich variety of variational principles which
give rise to Einstein’s equation, or equations equivalent to it but expressed
in terms of different field variables [247–249]. The best-known examples employ
the Einstein–Hilbert action expressed in terms of the metric, SEH [gµν ], or the
tetrad, respectively, SEH [eaµ ]. The latter action functional is obtained by insert-
ing the representation of the metric in terms of vielbeins, gµν = ηab eaµ ebν , into
the former.1

1 In this chapter coordinate (frame) indices are denoted by Greek (Latin) letters and ηab is
the frame metric. In the Euclidean setting, ηab = δab , the four-dimensional Lorentz group is
replaced with O(4).
 9.1 Quantum Einstein–Cartan Gravity 221

Another formulation which is classically equivalent, at least in the absence

of spinning matter, is provided by the first-order Hilbert–Palatini action
SHP [eaµ , ω abµ ], which, besides the tetrad, depends on the spin connection ω abµ
assuming values in the Lie algebra of O(1, 3) or O(4), respectively [247, 248, 250].
The variation of SHP with respect to ω abµ leads, in vacuo, to an equation of
motion stating that this connection has vanishing torsion. It can be solved
algebraically as ω = ω(e), which, when inserted into SHP , brings us back to
SEH [e] ≡ SHP [e, ω(e)].
Still another equivalent formulation is based on the self-dual Hilbert–Palatini
sd (+) ab
action SHP [eaµ , ω µ ], which only depends on the self-dual projection of the
(+) ab
spin connection, ω µ [16, 36–39]. This action in turn is closely related to
the Plebański action [251], containing additional two-form fields, and to the
Capovilla–Dell–Jacobson action [252, 253] which involves essentially only a
self-dual connection. Similarly, Krasnov’s diffeomorphism-invariant Yang–Mills
theories [254, 255] allow for a “pure connection” reformulation of General
Relativity.
The above variational principles are Lagrangian in nature; the fields employed
provide a parametrization of configuration space. The corresponding Hamilto-
nian description in which the “carrier fields” of the gravitational interaction
parametrize a phase space is obtained by applying a Legendre transform. In this
way the ADM-Hamiltonian [256] and, most importantly, Ashtekar’s Hamiltonian
[257], make their appearance.
Regarding our search for an asymptotically safe quantum theory of gravity
this multitude of classical formalisms offers many equally plausible possibilities
to explore. Each of the classical systems mentioned above “inspires” a particular
space of functionals and group of transformations under which they are invariant.
In principle we can calculate the RG flow on these theory spaces without any
further input data, in particular without selecting a bare action.
We are going to discuss the functional RG approach to “quantum Einstein–
Cartan gravity” as an example here, following the original works [258–261].
Throughout the discussion we set d = 4.
In Einstein–Cartan gravity the field variables are the tetrad, or vielbein eaµ
and a spin connection ω abµ , the latter assuming values in the Lie algebra of the
(Euclidean) “Lorentz” group, here taken to be O(4). This entails an enlargement
of the group of gauge transformations from the diffeomorphisms, Diff(M), to
Diff(M) n O(4)loc , so that gauge-invariant functionals A[e, ω] constitute a new
“universality class” replacing that of metric gravity with its diffeomorphism-
invariant functionals A[g]. By augmenting the number of field components from
10 in the case of gµν , to 40 for a pair (eaµ , ω abµ ), Einstein–Cartan gravity, in
fact, generalizes metric gravity: In particular, ω abµ allows for the possibility of
spacetime torsion [247, 248, 262].
Classically, the dynamics of pure Einstein–Cartan gravity are determined by
the Hilbert–Palatini action SHP [e, ω]. Since the resulting equations of motion give
222 Alternative Field Variables

rise to vanishing torsion, this functional can be regarded as the counterpart of

the Einstein–Hilbert action of metric gravity. However, quantum mechanically,
generic configurations (e, ω) contributing to the functional integral will have non-
zero torsion even if torsion should happen to vanish at the classical level: Hence,
classical equivalence does by no means imply quantum equivalence. Therefore,
the additional fields of Einstein–Cartan gravity can be expected to crucially affect
the renormalization.
The Hilbert–Palatini action is often generalized to the so-called Holst action
SHo [e, ω] which contains an additional term that only exists in four dimensions
[263, 264]; its prefactor is the dimensionless Immirzi parameter, γ [265, 266].
Since this monomial vanishes for vanishing torsion, it is absent in the metric
approach. While the classical field equations are independent of γ, the corre-
sponding quantum theory is expected to depend on it. In this respect, γ can
be compared to the θ-parameter of QCD; even though the latter multiplies a
topological term and therefore does not affect the classical equations of motion,
quantum observables may depend on it.
The Holst action underlies several approaches to the quantization of grav-
ity, including canonical quantum gravity with Ashtekar’s variables [36–38], Loop
Quantum Gravity (LQG) [39], spin-foam models [267], and group field theory
[268, 269]. Within LQG, for instance, γ enters the spectrum of area and volume
operators as well as the entropy formula for black holes [38], which exemplifies the
quantum significance of γ stated above. Furthermore, when coupled to fermions
in a non-minimal way, the presence of the Immirzi term induces a CP violating
four-fermion interaction that might be interesting for phenomenological reasons,
e.g., in the early universe [270].
While in LQG, for example, γ is a constant parameter that labels distinct
quantum theories, a consistent application of the functional RG in the context
of Einstein–Cartan gravity should treat γ as a running coupling, associated with
a certain monomial in the action functional, that is subject to renormalization.
From the perspective of the non-perturbative renormalization group the crucial
question is whether the RG flow on the theory space T made up by all function-
als of eaµ and ω abµ (and the necessary ghosts) that respect a background-type
realization of the new gauge invariance possesses zero, one, or even several fixed
points suitable for the Asymptotic Safety construction.
As we will see, the first non-perturbative studies indeed suggest the existence
of two NGFPs, which both seem suitable for defining a quantum field theory of
Einstein–Cartan gravity.
Since the underlying “universality class” of flows is different from that of metric
gravity, there is no general principle implying that the corresponding quantum
field theories are equivalent to their metric counterpart QEG.
In particular, even if we analyze the fixed points of Einstein–Cartan and metric
gravity, respectively, using truncations that happen to be equivalent classically
there are no conceptual reasons or mathematical mechanisms that would enforce
 9.2 Implementations of Gauge Invariance 223

equivalent results. In fact, the mere existence of fixed points in the (e, ω) uni-
versality class already constitutes a novel result, completely independent of the
analogous findings obtained in metric gravity.
Logically, Asymptotic Safety of metric gravity is neither necessary nor suffi-
cient for Asymptotic Safety on the Einstein–Cartan theory space, which may well
probe different universality classes of microscopic theories.

9.2 Implementations of Gauge Invariance

When setting up an Effective Average Action for the Einstein–Cartan theory
space, many of the relevant considerations parallel those in metric gravity. How-
ever, additional difficulties arise due to the more complicated structure of the
gauge algebra. In the following we will therefore mainly
∫ focus on those
{ aspects.
}
The formal starting point is the integral Z = Dê µ Dω̂ µ exp −S[ê, ω̂] ,
a ab

where the quantum fields êaµ and ω̂ abµ are defined on a fixed differentiable
manifold without boundary, M, and the bare action S is invariant both under dif-
feomorphisms, Diff(M), and local Lorentz rotations. We consider the Euclidean
version of the theory, so that the relevant group of gauge transformations is the
semidirect product G = Diff(M) n O(4)loc .
( )
(1) The pairs êaµ , ω̂ abµ determine an O(4)-covariant derivative

ˆ µ ≡ ∂µ + 1 ω̂ abµ Mab ,
∇ (9.1)
2
where Mab are the generators in the corresponding representation. The associated
curvature and torsion tensors read

F̂ abµν ≡ ∂µ ω̂ abν + ω̂ acµ ω̂ cbν − (µ ↔ ν),

(9.2)
T̂ aµν ≡ ∂µ êaν + ω̂ acµ êcν − (µ ↔ ν).

(2) Under the O(4)loc transformations δ L with parameters λab we have

δ L (λ)êaµ = λab êbµ ,

(9.3)
δ L (λ)ω̂ abµ = −∂µ λab + λac ω̂ cbµ + λbc ω̂ ac µ ≡ −∇
ˆ µ λab .

The spacetime diffeomorphisms δ D act as

δ D (w)êaµ = Lw êaµ , δ D (w)ω̂ abµ = Lw ω̂ ab µ , (9.4)

where Lw denotes the Lie derivative along the generating vector field w.
Upon gauge fixing we also need to consider diffeomorphism ghosts and
antighosts, C µ and C¯µ , respectively, and likewise Faddeev–Popov ghosts Σab and
Σ̄ab for the local O(4) transformations. We require all ghost and antighost fields
to transform under Diff(M) and O(4)loc as tensors of the corresponding type.
224 Alternative Field Variables

It then follows that the gauge transformations satisfy the Lie algebra relations

[δ D (w1 ), δ D (w2 )] Φ = δ D ([w1 , w2 ]) Φ

[δ L (λ1 ), δ L (λ2 )] Φ = δ L ([λ1 , λ2 ]) Φ (9.5)
[δ D (w), δ L (λ)] Φ = δ L (Lw λ) Φ

where Φ is any field from the set {êaµ , ω̂ abµ , C µ , C¯µ , Σab , Σab }. Furthermore
[w1 , w2 ] denotes the Lie bracket of the vector fields w1 and w2 , and [λ1 , λ2 ] is
the commutator of two matrices. The algebra constituted by (9.5) is a semidirect
product, Diff(M) n O(4)loc , with the local Lorentz transformations playing the
role of the invariant subalgebra.
In order to implement the gauge transformations on the theory space compris-
ing functionals A[êaµ , ω̂ abµ , C µ , C¯µ , Σab , Σ̄ab ] it is convenient to introduce Ward
operators WD and WL such that δ D,L A = −WD,L A to linear order in the
transformation parameters. Explicitly,
∫ (
δ δ
WD (w) = − d4 x δ D (w)êaµ (x) + δ D (w)ω̂ abµ (x) ab
δêaµ (x) δ ω̂ µ (x)
δ δ
+ δ D (w)C µ (x) µ + δ D (w)C¯µ (x) ¯ (9.6)
δC (x) δ Cµ (x)
)
δ δ
+ δ D (w)Σab (x) ab + δ D (w)Σ̄ab (x)
δΣ (x) δ Σ̄ab (x)

and analogously for WL . The Ward operators satisfy

[WD (w1 ), WD (w2 )] = WD ([w1 , w2 ]),

[WL (λ1 ), WL (λ2 )] = WL ([λ1 , λ2 ]), (9.7)
[WD (w), WL (λ)] = WL (Lw λ).

Gauge-invariant functionals A[ê, ω̂, C, C,

¯ Σ, Σ̄] are characterized by the conditions
WD (w) A = 0 = WL (λ) A for all w and λ.

(3) In order to construct a flow equation with the desired invariance properties
it is important to notice that the (ordinary) diffeomorphisms δ D (w) are not
covariant under O(4)loc . This is obvious from the fact that the Lie derivative
involves partial rather than O(4)-covariant derivatives.
As a remedy we covariantize the diffeomorphisms by combining them with an
appropriate O(4) transformation. Namely, introducing

δf
D (w) ≡ δ D (w) + δ L (w · ω̂), (9.8)
 9.2 Implementations of Gauge Invariance 225

where (w · ω̂)ab ≡ wµ ω̂ abµ , it can be shown that these modified diffeomorphisms

δf
D act on all fields by covariant derivatives ∇µ only:
ˆ

δf
D (w)ê µ = w ∇ρ ê µ + (∇µ w )ê ρ ,
a ρˆ a ˆ ρ a

δf
D (w)ω̂ µ = −F̂ µρ w ,
ab ab ρ

δf
D (w)C = w ∇ρ C − (∇ρ w )C = w ∂ρ C − (∂ρ w )C ,
µ ρˆ µ ˆ µ ρ ρ µ µ ρ
(9.9)
δf
D (w)Cµ = w ∇ρ Cµ + (∇µ w )Cρ ,
¯ ρˆ ¯ ˆ ρ ¯

δf
D (w)Σ = w ∇ρ Σ ,
ab ρˆ ab

δf
D (w)Σ̄ab = w ∇ρ Σ̄ab .
ρˆ

Associating as above Ward operators W g D (w) to the modified diffeomorphisms

leads to the following covariantized form of the gauge algebra:

[ ]
Wg g g
D (w1 ), WD (w2 ) = WD ([w1 , w2 ]) − WL (w1 w2 · F̂ ),
[ ]
WL (λ1 ), WL (λ2 ) = WL ([λ1 , λ2 ]), (9.10)
[ ]
Wg D (w), WL (λ) = 0.

Here (w1 w2 · F̂ )ab ≡ w1µ w2ν F̂ abµν . Note that while the modified diffeomorphisms
commute with local Lorentz transformations, they no longer close among them-
selves; their commutator contains an O(4)loc transformation whose parameter
involves F̂ , the curvature of ω̂.
Note that gauge-invariant functionals A can now also be characterized by the
conditions W g D (w) A = 0 = WL (λ) A for all w and λ.

(4) As in metric gravity, we employ the background field technique to cope

with the requirement of Background Independence. We introduce arbitrary back-
ground fields ēaµ and ω̄ abµ , whereby the background vielbein ēaµ is assumed to be
non-degenerate. As a result, it gives rise to a well-defined inverse (ēaµ ) ≡ (ēaµ )−1 ,
to a non-degenerate background metric ḡµν ≡ ēaµ ēbν δab , and to the completely
covariant derivative D̄ ≡ ∂ + ω̄ + Γ̄ ≡ ∇ ¯ + Γ̄ where Γ̄ ≡ Γ̄(ē, ω̄) is fixed by the
a
requirement D̄µ ē ν = 0 [248].
We decompose the variables of integration as êaµ ≡ ēaµ + ε̂aµ , ω̂ abµ ≡ ω̄ abµ +
τ̂ abµ , and then apply the Faddeev–Popov method in order to gauge fix the
functional integral:
∫
{ }
Z= Dε̂aµ Dτ̂ abµ exp −S[ē + ε̂, ω̄ + τ̂ ] − Sgf [ε̂, τ̂ ; ē, ω̄]
∫ (9.11)
{ }
× DC µ DC¯µ DΣab DΣ̄ab exp −Sgh .
226 Alternative Field Variables

Here Sgf and Sgh denote the gauge-fixing and ghost actions, respectively, C µ and
C¯µ are the diffeomorphism ghosts, and Σab and Σ̄ab those related to the local
O(4). The gauge fixing is of the form
∫ ∫
1 1
Sgf = d4 x ē ḡ µν Fµ Fν + d4 x ē G ab Gab (9.12)
2αD · 16πG 2αL
where Fµ and G ab break the Diff(M) and O(4)loc gauge invariance, respectively.
In order to render Γk manifestly Diff(M) n O(4)loc invariant we employ gauge
conditions of the “background type” for which Sgf [ε̂, τ̂ ; ē, ω̄] is invariant under
the combined background gauge transformations δ B D,L acting on both (ε̂, τ̂ ) and
(ē, ω̄) while, of course, it is not invariant under the “true” (or “quantum”) gauge
transformations, denoted by δ Q Q
D and δ L , respectively.

(5) The true diffeomorphisms act on the decomposed fields according to

δQ a
D (w)ē µ = 0,

δQ
D (w)ε̂ µ = Lw (ē µ + ε̂ µ ),
a a a
(9.13)
δQ ab
D (w)ω̄ µ = 0,

δQ
D (w)τ̂ µ = Lw (ω̄ µ + τ̂ µ )
ab ab ab

and the true O(4) gauge transformations are given by

δQ a
L (λ)ē µ = 0,

δQ a a b b
L (λ)ε̂ µ = λ b (ē µ + ε̂ µ ),
(9.14)
δQ ab
L (λ)ω̄ µ = 0,

δQ
L (λ)τ̂ µ = −∂µ λ
ab ab
+ λac (ω̄ cbµ + τ̂ cbµ ) + λbc (ω̄ acµ + τ̂ acµ ).

On the other hand, the background diffeomorphisms act as

D (w)ē µ = Lw ē µ ,
δB a a

D (w)ε̂ µ = Lw ε̂ µ ,
δB a a
(9.15)
D (w)ω̄ µ = Lw ω̄ µ ,
δB ab ab

D (w)τ̂ µ = Lw τ̂ µ
δB ab ab

and the background O(4) gauge transformations are

δB a a b
L (λ)ē µ = λ b ē µ ,

δB a a b
L (λ)ε̂ µ = λ b ε̂ µ ,
(9.16)
L (λ)ω̄ µ = −∂µ λ + λac ω̄ cbµ + λbc ω̄ acµ ≡ −∇
δB ab ab ¯ µ λab ,
δB ab a cb b ac
L (λ)τ̂ µ = λ c τ̂ µ + λ c τ̂ µ ,

where ∇
¯ denotes the O(4) covariant derivative (9.1) constructed from ω̄ abµ .
 9.2 Implementations of Gauge Invariance 227

Since no background split is introduced for the ghost fields, their true and
background gauge transformations coincide. We require that under δ Q D,L and
δBD,L they transform tensorially according to their respective index structure.
Introducing Ward operators W B D , W L for the background gauge transforma-
B
Q Q
tions, and W D , W L for the “quantum” or “true” ones, we can verify that the
former satisfy the algebra
[ B ]
W D (w1 ), W B
D (w2 ) = W D ([w1 , w2 ]),
B
[ B ]
W L (λ1 ), W B
L (λ2 ) = W L ([λ1 , λ2 ]),
B
(9.17)
[ B ]
W D (w), W B L (λ) = W L (Lw λ),
B

while the latter obey the relations

[ Q ]
W D (w1 ), W Q Q
D (w2 ) = W D ([w1 , w2 ]),
[ Q ]
W L (λ1 ), W Q Q
L (λ2 ) = W L ([λ1 , λ2 ]), (9.18)
[ Q ]
W D (w), W Q Q
L (λ) = W L (Lw λ).

Like their precursors before the background split, these commutation relations
are not O(4)loc covariant. This time within the background field setting, we
therefore define modified diffeomorphisms:

f
δf
D (w) ≡ δ D (w) + δ L (w · ω̄),
B B B
(9.19)
f
f
δQ Q Q
D (w) ≡ δ D (w) + δ L (w · ω̄).

In terms of their Ward operators, the modified background diffeomorphisms are

then seen to satisfy the commutation relations

[g
g g
g ] gg
W B (w ), W
D 1
B (w ) = W
D 2
B ([w , w ]) − W B (w w · F̄ ),
D 1 2 L 1 2
[ B ]
W L (λ1 ), W B (9.20)
L (λ2 ) = W L ([λ1 , λ2 ]),
B

[g
g ]
W B (w), W B (λ) = 0.
D L

Their “quantum” counterparts have the following Lie algebra relations:

[gg g
g ] gg
WQ Q Q Q
D (w1 ), W D (w2 ) = W D ([w1 , w2 ]) + W L (w1 w2 · F̄ ),
[ Q ]
W L (λ1 ), W Q Q (9.21)
L (λ2 ) = W L ([λ1 , λ2 ]),
[g
g ]
WQ Q Q
D (w), W L (λ) = W L (w · ∇λ).
¯
228 Alternative Field Variables

Both algebras, (9.20) and (9.21), are of immediate relevance to the functional
RG equation: The “background” transformations constrain the theory space on
which the flow takes place, while the algebra of the “quantum” transformations
determines the ghost action [271].

(6) Let us choose the following convenient gauge conditions which are linear in
ε̂aµ and independent of τ̂ abµ [272]:
[ ]
Fµ = ēaν D̄ν ε̂aµ + βD D̄µ ε̂aν ,
1 [ ] (9.22)
G ab = ḡ µν ε̂aµ ēbν − ε̂bµ ēaν ≡ ε̂[ab] .
2
With the new parameter βD there is a total of three gauge-fixing parameters
now: αD , αL , and βD . Using (9.22) in (9.12) we can verify that Sgf [ε̂, τ̂ ; ē, ω̄] is
indeed background gauge invariant:

g
g
WB
L Sgf = 0 = W D Sgf ⇔ W L Sgf = 0 = W D Sgf .
B B B
(9.23)

However, the ghost sector requires some care. We would like the ghost
action Sgh [ε̂, τ̂ , C, C,
¯ Σ, Σ̄; ē, ω̄] to be background gauge invariant, too. How-
ever, straightforwardly applying the Faddeev–Popov procedure to the original
transformations  
Q
δ (w)
δQ =  D  (9.24)
δQL (λ)

we obtain, in the Σ̄ − C-sector, the ghost action2

∫ ( )
∂G ab Q
Sgf [C, Σ̄; ē, ω̄] = − d x ē Σ̄ab c δ D (C)ε̂ ν
Σ̄−C 4 c
, (9.25)
∂ ε̂ ν ε̂=0

which, with (9.22), evaluates to

∫
Σ̄−C
Sgf [C, Σ̄; ē, ω̄] = − d4 x ē Σ̄ab ēbµ LC ēaµ . (9.26)

While this latter functional is invariant under background diffeomorphisms, as

it should be, it fails to be invariant under the O(4)loc transformations δ B L (λ).
The reason is that the Lie derivative of an O(4) tensor does not define an O(4)
tensor. Rather, we have LC (λab ēbµ ) ̸= λab LC ēbµ , since λab (x) is a spacetime scalar
that transforms non-trivially under diffeomorphisms. Stated differently, O(4)loc
transformations and (ordinary) diffeomorphisms do not commute. This is exactly
what the above algebra relations express.

2 Here we already specialized to the case ε̂aµ = 0, which will be sufficient for all truncations of
“single-field” type [258].
 9.2 Implementations of Gauge Invariance 229

The way out of this difficulty consists in applying the Faddeev–Popov proce-
dure to the O(4)loc -covariantized (true) gauge transformations rather than the
original ones:  
f
f
f Q
δf
δ (w) 
Q
= D . (9.27)
δQ
L (λ)
They are gauge fixed by the 10 gauge conditions
 
F ( )
 µ  ≡ QI (9.28)
G ab

( ) ( )
for which
( I ) we( use
) a uniform notation here, with QI ≡ Fµ for I = 1, · · · , 4
and Q ≡ G ab for I = 5, · · · , 10. Denoting,
( I ) in( the same
) fashion, the ten
µ a
parameters of the gauge transformations as Λ = w , λ b , the Faddeev–Popov
determinant reads ( I )
δQ (x)
det . (9.29)
δΛJ (y) Λ=0
Exponentiating it we obtain a ghost action which has the structure
 T   
∫
C¯µ Ωµν Ωµcd Cν
− d4 x ē     . (9.30)
Σ̄ab Ωabν Ωabcd Σcd

The Faddeev–Popov operator Ω appearing in (9.30) is rather complicated; we

must refer to [258] for its explicit form. However, it is important to note that one
can now check that the ghost action (9.30) is indeed invariant under background
gauge transformations:
g
g
L Sgh = 0 = W D Sgh
WB B ⇔ L Sgh = 0 = W D Sgh .
WB B
(9.31)

This property is a central prerequisite for ultimately arriving at a background

gauge-invariant Effective Average Action.

(7) The functional integral (9.11) gives rise to the associated Effective Average
Action [14] in the by∫now familiar way: one adds a δ B -invariant mode cutoff to the
bare action, ∆Sk ∝ d4 x ē (ε̂, τ̂ ) Rk (ε̂, τ̂ )T , couples ε̂ and τ̂ to sources, Legendre
transforms the resulting generating functional ln Zk , and finally subtracts ∆Sk
for the expectation value fields in order to arrive at the running action:
¯ Υ, Ῡ; ē, ω̄] ≡ Γk [e, ω, ē, ω̄, ξ, ξ,
Γk [ε, τ, ξ, ξ, ¯ Υ, Ῡ]. (9.32)

In the list of arguments the expectation value fields are ε ≡ ⟨ε̂⟩, τ ≡ ⟨τ̂ ⟩, and for
the ghosts
ξ µ ≡ ⟨C µ ⟩, ξ¯µ ≡ ⟨C¯µ ⟩, Υab ≡ ⟨Σab ⟩, Ῡab ≡ ⟨Σ̄ab ⟩. (9.33)
230 Alternative Field Variables

In addition,

eaµ ≡ ⟨êaµ ⟩ = ēaµ + εaµ , ω abµ ≡ ⟨ω̂ abµ ⟩ = ω̄ abµ + τ abµ . (9.34)

The Effective Average Action Γk may be considered a functional of either the

fluctuations εaµ and τ abµ or the complete classical fields eaµ and ω abµ .
Obviously the action Γk is defined on a rather complicated theory space,
T . It consists of functionals depending on two independent vielbein variables
(e, ē), two spin connections (ω, ω̄), as well as on the diffeomorphism and O(4)
ghosts and antighosts, respectively. The functionals A[·] in T are constrained
by the requirement of background gauge invariance: W B D (w) A = 0 ∧ W L (λ) A =
B

0, ∀w , λ b . Additionally they must satisfy Ward identities for BRST and split-
µ a

symmetry transformations, in analogy with the metric case.

(8) Given this theory space and the above definition of Γk , it is now straightfor-
ward to derive its FRGE. Again, it has the same structure as the previous flow
equation:
1 [ ]
k∂k Γk = STr (Γk + Rk )−1 k∂k Rk .
(2)
(9.35)
2
With Rk [ē, ω̄] specified appropriately, the equation indeed defines a flow on T ,
i.e., it does not generate background gauge invariance violating terms.

9.3 Truncations of Hilbert–Palatini Type

As a first concrete example [260, 261], the flow equation for Γk [e, ω, · · · ] has been
solved on a three-dimensional truncated theory space spanned by actions of the
generalized Hilbert–Palatini, i.e., Holst type:
∫ [ ( )
1 1
Γk [e, ω, · · · ] = − d4 x e eaµ ebν F abµν − ⋆ F abµν
16πGk γk
] (9.36)
− 2Λk + Sgf + Sgh .

This ansatz consists of the Hilbert–Palatini action known from classical Einstein–
Cartan gravity, plus the Immirzi term which only exists in four dimensions; in
fact, ⋆F abµν ≡ 21 εabcd F cdµν is the dual of the curvature of ω, F ≡ F (ω), with respect
to the frame indices.
Besides Gk , (9.36) contains two more running parameters: the cosmological
constant Λk and the dimensionless Immirzi parameter γk . The gauge fixing
and ghost terms are assumed to retain their classical form for all k, and the
parameters αD , αL and βD are treated as constants. Thus locally the truncated
theory space T can be coordinatized by triples (g, λ, γ) where gk ≡ k 2 Gk and
λk ≡ k −2 Λk .
 9.3 Truncations of Hilbert–Palatini Type 231

(1) The non-perturbative beta functions related to the ansatz (9.36) have been
obtained for two different projection schemes in [260, 261]. The two computations
differed also with respect to certain additional approximations. They have been
applied on top of the truncation in order to cope with the formidable algebra
behind the functional traces that must be evaluated.3
Focusing on the first scheme from now on, we can discuss its results only
briefly here; for a detailed discussion we refer to [260], and to [258, 259] for
shorter accounts.
The projected RG equations are of the form ∂t gk = βg ≡ (2 + ηN )gk , ∂t λk = βλ ,
∂t γk = βγ with the anomalous dimension of Newton’s constant and the other two
beta functions given by:

ηN (g, λ, γ) = 16π g f+ (λ, γ),

[ ]
βγ (g, λ, γ) = 16π g γ γ f− (λ, γ) − f+ (λ, γ) , (9.37)
[ ]
βλ (g, λ, γ) = −2 λ + 8π g 2 λ f+ (λ, γ) + f3 (λ, γ) .

The functions f± and f3 are extremely complicated and we must refer to [260]
for their explicit form. Besides λ and γ, they depend parametrically also on
the three gauge-fixing constants and an additional mass parameter µ̄ which is
needed to give a uniform dimension to ε̄aµ and τ̄ abµ . Indeed, only after rescaling
ε̄aµ → µ̄ 2 ε̄aµ , τ̄ abµ → µ̄− 2 τ̄ abµ the inverse propagator Γk constitutes an operator
1 1 (2)

with well-defined spectrum and trace.4 The resulting RG flow is reflection sym-
metric under γ → −γ.

(2) Being particularly interested in the regions near γ = 0 and γ = ±∞, we are
led to coordinatize T by an atlas consisting of two charts.
In order to cover the neighborhood of the submanifold γ = ±∞ in T , we intro-
duce a new coordinate γ̂. In the overlap |γ| ∈ ]0, +∞[ of the (g, λ, γ)- and the
(g, λ, γ̂)-chart, the coordinates γ and γ̂ are related by the transition function
γ̂(γ) = γ −1 . With βγ̂ (g, λ, γ̂) = −γ̂ 2 βγ (g, λ, γ̂ −1 ), the flow equation in the γ̂-chart
has the structure

ηN (g, λ, γ) = 16π g f+ (λ, γ̂ −1 ),

[ ]
βγ̂ (g, λ, γ̂) = 16π g γ̂ f+ (λ, γ̂ −1 ) − γ̂ −1 f− (λ, γ̂ −1 ) , (9.38)
[ ]
βλ (g, λ, γ̂) = −2 λ + 8π g 2 λ f+ (λ, γ̂ −1 ) + f3 (λ, γ̂ −1 ) .

3 In [260] the proper time approximation of the FRGE has been used (as in [273] for metric
gravity), while [261] employed a novel “Wegner-Houghton-like” functional RG equation which
follows from (9.35) under certain assumptions.
4 At a more abstract level, µ̄ characterizes a Riemannian metric in the space of fields which
is necessary to set up the FRGE [274].
232 Alternative Field Variables

(3) The beta functions βg , βγ , βγ̂ , and βλ have simple poles at γ = γ̂ = ±1,
presumably artifacts of the approximations employed. In fact, the analysis shows
that for γ not too close to ±1 the functions f± and f3 are actually independent
of γ. For such values of γ it is a rather precise approximation to replace them by
functions f˜± and f˜3 that only depend on λ:
[ ]
∂t gk = 2 + 16π gk f˜+ (λk ) gk ,
[ ]
∂t γk = 16π gk γk γk f˜− (λk ) − f˜+ (λk ) , (9.39)
[ ]
∂t λk = −2 λk + 8π gk 2 λk f˜+ (λk ) + f˜3 (λk ) .

and likewise for the γ̂-chart. While the equations (9.39) are manifestly equiv-
alent to (9.37) when |γ| ̸≈ 1, a detailed analysis [260] indicates that, somewhat
surprisingly, for |γ| → 1, too, the regular beta functions (9.39) rather than those
of (9.37) are likely to apply.

(4) The singularities at γ = ±1 are a consequence of the fact that for these val-
ues of the Immirzi parameter the Holst action involves the projection operator
on (anti-)selfdual field strength tensors, P ± = 12 (1 ± ⋆). It can be shown that the
(anti-)selfdual projection of the field-strength tensor of a generic spin connection
equals the field strength tensor of the (anti-)selfdual part of this spin connec-
tion: (P ± F )ab (ω) = F ab (P ± ω). Hence, if γk = ±1, the (anti-)selfdual part of ω
completely drops out from the action (9.36). However, the underlying functional
integral still includes the integration over the decoupled projection. It is this
divergence which is mirrored by the flow equation.
If one is interested in “chiral gravity” mediated by an (anti-)selfdual spin con-
nection ω ± an independent calculation should be set up, beginning with a theory
space of functionals A[e, ω ± ] which depend on either a selfdual or an anti-selfdual
connection. There is no functional integration over the discarded set of fields then,
and the resulting RG equations and beta functions are well behaved [275].

(5) The system (9.39) and its analogue in the γ̂-chart imply βγ = 0 and βγ̂ = 0
for γ ∗ = 0 and γ̂ ∗ = 0, respectively. For each of the two sets of equations we find a
fixed point NGFP0 ≡ (g0∗ , λ∗0 , γ ∗ ) and NGFP∞ ≡ (g∞ ∗
, λ∗∞ , γ̂ ∗ ) of (9.37), (9.38)
∗
with g0,∞ > 0, but λ∗0,∞ < 0 and g0∗ ̸= g∞ ∗
, λ∗0 ̸= λ∗∞ . Both of them seem eligible
for the Asymptotic Safety construction.
At either fixed point, the g and λ directions are to a very good approxima-
tion eigendirections of the linearized flow on T . This is even exactly true for
the γ- and γ̂-directions, respectively. At NGFP0 and NGFP∞ , both the g
and λ directions are relevant scaling fields, i.e., their associated critical expo-
nents θ1 and θ2 are real and positive. In contrast, at NGFP0 the Immirzi
parameter γ is irrelevant (θγ < 0), whereas at NGFP∞ its inverse γ̂ is relevant
(θγ̂ > 0).
 9.3 Truncations of Hilbert–Palatini Type 233

(6) The existence of the two non-Gaussian fixed points in the new “universality
class” based on the elementary fields e and ω is a significant result clearly, a
first hint at the viability of the Asymptotic Safety program in Einstein–Cartan
gravity.
As we pointed out already, their existence is logically and computationally
independent of, and not ∫ implied by, any known properties of the metric theory.
Defining I ≡ 16πG
1
d4
x e ε µνρσ a
T µν T bρσ δab , the Immirzi term appears in the
k { 1 }
functional integral as exp − γ · I + surface term . Therefore, for γ → 0+ , con-
figurations with I > 0 get strongly suppressed whereas those with I < 0 are
enhanced. For γ → 0− , the situation is just reversed. These two cases are related
by parity, and neither of them leads to a complete suppression of torsion. Hence
there is no general reason for metric gravity to be recovered at any value of γ.

(7) By letting λ = 0 we obtain the (g, γ)- and (g, γ̂)-subtruncation, respectively,
with βg , βγ , and βγ̂ given by (9.37) and (9.38), but with the functions f±
) the results are compatible with βγ = 0 = βγ̂ ⇔
evaluated at λ = 0.( In this case
f+ (0, γ)|γ=0, ±∞ = γ f− (0, γ) |γ=0, ±∞ . As a consequence, the renormalization
of the Immirzi term is only due to the running of the overall prefactor 1/Gk .
This result needs to be corroborated by a more precise treatment. If correct, the
Immirzi parameter owes its RG running to a non-zero cosmological constant in
an essential way.

(8) Within the λ = 0-subtruncations there exists an intriguing duality relating

small and large values of the Immirzi parameter [260]. The RG flow is then
invariant under the mapping γ 7→ 1/γ. A non-zero value of the cosmological con-
stant breaks this invariance, however.

(9) Summarizing the results obtained with both the first [260] and the second
projection scheme [261] it can be said that there is indeed non-trivial evidence
for the Asymptotic Safety of pure gravity in the Einstein–Cartan setting. There
seem to exist at least two NGFPs, located at γ = 0 and γ = ±∞, respectively,
which are suitable for taking the continuum limit there. Interestingly, no reliable
fixed point with a finite, non-zero value of γ is found.
The stability and robustness properties of the Einstein–Cartan results turned
out somewhat less pronounced than those of comparable truncations in met-
ric gravity. The reason is probably the enlargement of the gauge sector by the
local Lorentz transformations. In fact, similar observations were made in “tetrad-
only gravity” whose theory space consists of functionals A[eaµ ], which are also
Diff(M) n O(4)loc invariant [276].5

5 See also [277, 278] for a perturbative investigation of Einstein–Cartan gravity in the special
case λ = 0.
234 Alternative Field Variables

(10) Besides the space of Einstein–Cartan-type actions A[e, ω] a number of

related theory spaces, based on other field variables, have been scrutinized for
the possibility of asymptotically safe theories.
This includes in particular (anti-)selfdual gravity. It employs actions A[e, ω (±) ]
depending on the vierbein field along with an O(4)-valued spin connection, which
is required to be self-dual (or alternatively, anti-selfdual) with respect to its frame
indices [275].
This theory space is closely related to the Euclidean analog of Ashtekar’s
variables [36, 37, 257, 279].6 In fact, Ashtekar showed that if one starts from
the classical Einstein–Cartan theory and performs a suitably chosen canonical
transformation to new variables, the pertinent spin connection can be cho-
sen as self-dual (or anti-selfdual). For the Lorentzian signature and structure
group O(1, 3) self-dual connections are necessarily complex, which complicates
their quantization. In the Euclidean case they are real, however, and the condi-
tion of self-duality precisely halves the number of the connection’s independent
components.
Within the Euclidean setting, a variant of the calculations described above
led to the conclusion that self-dual gravity is also likely to be asymptotically safe
[275].
Similar results were also obtained in studies where the metric and the torsion
tensor T λ µν were considered the independent field variables of the “Holst trunca-
tion” [274], and in a more general truncation, comprising all invariants quadratic
in the torsion and non-metricity tensor [281].

9.4 Quantum Arnowitt–Deser–Misner Gravity

An alternative set of fields carrying the gravitational degrees of freedom arises
from the Arnowitt–Deser–Misner (ADM) or (3 + 1)-formulation of gravity [256,
282], see [283] for a pedagogical introduction.
This construction equips spacetime with a foliation structure by welding
together spatial slices on which the time coordinate τ is constant. This introduces
a preferred direction which may be interpreted as an (Euclidean) time direction
and provides a causal structure. As a byproduct it gives access to interesting
observables comprising, e.g., the expectation value for spatial volumes V3 (τ ) as
a function of time. This facilitates the direct comparison with results obtained
from the Causal Dynamical Triangulations program [244] which evaluates the
gravitational partition sum by Monte Carlo methods.
In this section we review the construction of the corresponding FRGE and high-
light the connections to the covariant formulation of Quantum Einstein Gravity.

6 See also [280] for the relation between self-duality and helicity.
 9.4 Quantum Arnowitt–Deser–Misner Gravity 235

(1) At the classical level, the ADM formulation of gravity starts from a d-
dimensional (Euclidean) spacetime M with metric gµν carrying coordinates xµ .
Subsequently, the construction introduces a time function τ (x) which assigns a
specific time to each spacetime point. The points with the same value of τ (x)
form D-dimensional spatial slices Στ ≡ {x : τ (x) = τ }. Here D ≡ d − 1.
The gradient of the time function, nµ ≡ N ∂µ τ (x), defines a vector nµ normal
to the spatial slices. The lapse function N (x) ensures that gµν nµ nν = 1. One
then introduces a new coordinate system xµ 7→ (τ, y i ), i = 1, · · · , D where the y i
provide coordinates on Στ .7 Defining the vector field tµ through the relation
tµ ∂µ τ = 1, the coordinate systems on neighboring spatial slices are related by
requiring that the coordinates y i are constant along the integral curves of tµ .
The tangent space at a point in M can then be decomposed into a subspace
spanned by vectors tangent to Στ and its complement. The corresponding basis
vectors are constructed from the Jacobians
∂xµ ∂xµ
tµ = , ei µ = . (9.40)
∂τ yi ∂y i τ

The normal vector thus satisfies gµν nµ ei ν = 0, and the metric induced on the
spatial slices is given by
σij (τ, y) ≡ ei µ ej ν gµν . (9.41)
In general tµ is neither tangent nor normal to the spatial slices. The Jacobians
(9.40) imply that its decomposition into components normal and tangent to Σ is
given by
t µ = N n µ + N i ei µ , (9.42)
where N i (τ, y) is called the shift vector. Furthermore, (9.40) implies that the
coordinate one-forms in the two coordinate systems are related by

dxµ = tµ dτ + ei µ dy i = N nµ dτ + ei µ (dy i + N i dτ ). (9.43)

Combining the relations (9.42) and (9.43) with the normal property of nµ and
the definition of σij , the line element ds2 = gµν dxµ dxν can be recast in terms of
the ADM fields {N, Ni , σij }:

ds2 = N 2 dτ 2 + σij (dy i + N i dτ )(dy j + N j dτ ). (9.44)

For the components of the metric tensor the decomposition (9.44) implies the
relations
( 2 ) ( )
Nj
N + Ni N i Nj µν N
1
2 −N 2
gµν = , g = Ni i j , (9.45)
Ni σij −N 2 σ ij + NNN 2

7 In this section Greek letters are used for spacetime indices while tangent indices with respect
to the spatial slices are denoted by Latin letters i, j, · · · .
236 Alternative Field Variables

where spatial indices i, j are raised and lowered with σij . This entails that the
relation between gµν and the ADM fields is actually non-linear.

(2) An infinitesimal coordinate transformation v µ (τ, y) acting on the metric

can be
( expressed in) terms of the Lie derivative, δgµν = Lv gµν . Decomposing
v ≡ f (τ, y), ζ (τ, y) into its temporal and spatial parts, this transformation
µ i

determines the transformation properties of the component fields under Diff(M):

δN = ∂τ (f N ) + ζ k ∂k N − N N i ∂i f,
δNi = ∂τ (Ni f ) + ζ k ∂k Ni + Nk ∂i ζ k + σki ∂τ ζ k + Nk N k ∂i f + N 2 ∂i f, (9.46)
k k k
δσij = f ∂τ σij + ζ ∂k σij + σjk ∂i ζ + σik ∂j ζ + Nj ∂i f + Ni ∂j f.

Thus, at the level of the component fields the action of Diff(M) is non-linear.
Notably, Diff(M) contains the subgroup of foliation preserving diffeomorphisms,
Diff(M, Σ), where f = f (τ ) is restricted to be independent of the spatial
coordinates. Inspecting (9.46), one observes that the restriction f (τ, y) → f (τ )
eliminates all non-linear terms from the transformation laws so that the compo-
nent fields transform linearly with respect to this subgroup.

(3) The construction of the FRGE tailored to the ADM formalism again builds on
the background field formalism. In this case the gravitational degrees of freedom
are encoded in the lapse function N (τ, y), the shift vector Ni (τ, y), and the metric
on the spatial slices σij (τ, y).
Following the metric construction, these fields are decomposed into fixed but
arbitrary backgrounds (marked with bars) and fluctuations (marked with hats).
The linear ADM split uses
b,
N = N̄ + N bi ,
Ni = N̄i + N bij .
σij = σ̄ij + σ (9.47)

For convenience, we denote the collection of physical fields, background fields,

b, respectively, and set σ
and the fluctuations by χ, χ̄, and χ b ≡ σ̄ ij σ
bij . The linear
split of σij can be replaced by
[ ]k
σij = σ̄ik eσb j , (9.48)

which defines the exponential ADM split.

The properties of these choices are conveniently discussed by writing the
determinant of the spacetime metric in terms of the ADM fields:

det g = N 2 det σ. (9.49)

Combining this relation with the decomposition (9.47) one concludes that the
fluctuations can not change the signature in the (Euclidean) time direction. The
exponential ADM split furthermore fixes the signature of the metric on the spatial
slices.
 9.4 Quantum Arnowitt–Deser–Misner Gravity 237

(4) At this stage it is useful to note that there exists a local map relating the
fluctuation fields of the covariant and the ADM formulation. This map allows
us to understand the ADM construction as a particular background-dependent
redefinition of the fluctuation fields.
The explicit relation between the two settings is found by starting from the
linear split in the metric setting, gµν = ḡµν + hµν , and carrying out an ADM
decomposition of both gµν and ḡµν according to (9.45). The result allows us to
express the components of hµν in terms of N b, N
bi , and σ
bij and their background
values:

b +N
h00 = 2N̄ N b 2 + σ ij (N̄i + N
bi )(N̄j + N
bj ) − σ̄ ij N̄i N̄j ,
bi ,
h0i = N (9.50)
bij .
hij = σ

Note that the resulting map is again non-linear. Due to the presence of σ ij in
bij to arbitrary high orders. At the linear
h00 it involves the spatial fluctuations σ
level the map (9.50) boils down to

b −σ
h00 ≈ 2N̄ N bi N̄j ,
bij N̄i N̄j + 2 σ̄ ij N bi ,
h0i = N bij .
hij = σ (9.51)

(5) The construction of the FRGE for the Effective Average Action Γk then
proceeds along the lines of Chapter 2 and gives [284]
[( )−1 ]
1 (2)
χ; χ̄] = STr Γk + Rk
∂t Γk [b ∂t Rk . (9.52)
2

(2)
The Hessian Γk now comprises the second functional derivative of Γk with
respect to all fluctuation fields appearing in the ADM formulation.
A further difference is that the background gauge transformations preserved
by the flow equation (9.52) are actually not the full background diffeomorphism
group but its foliation preserving subgroup only.
This feature results from the requirement that the k-dependent mass term ∆Sk
related to the regulator Rk must be quadratic in the fluctuation fields in order
to obtain a FRGE of the standard form. Thus, the regulator term is invariant
only under linear transformations acting on the fluctuation fields. At the same
time (9.46) indicates that the ADM fields transform non-linearly under diffeo-
morphisms. Therefore, (9.52) is invariant under foliation preserving background
transformations only.8

8 In principle the map (9.50) may be used to construct a regulator ∆Sk preserving full
background diffeomorphism invariance. At the level of the ADM fields this leads to the
infinite-order problem discussed in Section 4.2.
238 Alternative Field Variables

9.5 Quantum ADM Gravity in the Einstein–Hilbert Truncation

We illustrate the working of the FRGE by approximating the gravitational part
of Γk by the (Euclidean) Einstein–Hilbert action. In terms of the ADM fields
this implies
∫
grav 1 √ [ ]
Γk = dτ dD y N σ Kij K ij − K 2 − (D) R + 2Λk , (9.53)
16πGk
where the extrinsic curvature is defined as
1 ( )
Kij ≡ ∂τ σij − Di Nj − Dj Ni , K ≡ σ ij Kij , (9.54)
2N
and (D) R and Di are, respectively, the Ricci scalar and covariant derivative con-
structed from σij . In order to lighten the notation we omit the prescript from
the intrinsic curvature from now on, denoting R ≡ (D) R.
The ansatz for Γgrav
k comprises two scale-dependent couplings, the cosmolog-
ical constant Λk and Newton’s constant Gk . The beta functions encoding the
scale dependence of these couplings may be obtained by substituting the ansatz
into the FRGE and extracting suitable geometric terms on both sides. The form
of (9.53) suggests evaluating the flow equation at the background level, i.e., set-
ting χb = 0 after taking the necessary functional derivatives. In this case, it then
suffices to keep track of terms that are at most quadratic in the fluctuation fields
(2)
as these are necessary to construct the Hessian Γk .

(1) Similarly to the metric computation in Chapter 5, the evaluation of the

flow equation can be simplified by choosing a suitable background. In order to
facilitate the comparison with the metric setting, it is useful to choose a class of
backgrounds with topology S 1 × S D , with
D
S
N̄ (τ, y) = 1, N̄i (τ, y) = 0, σ̄ij (τ, y) = σ̄ij (y). (9.55)
D
S
Here σ̄ij (y) denotes the metric on the D-sphere with radius r, which is taken
to be independent of τ . At the geometric level this entails:

K̄ij = 0, R̄
R̄ijkl = D(D−1) (σ̄ik σ̄jl − σ̄il σ̄jk ) , 1
R̄ij = D σ̄ij R̄. (9.56)

The scale dependence of the cosmological constant and Newton’s constant can
be read off from the terms proportional to the background volume and the inte-
grated intrinsic background curvature, respectively.

(2) A key obstacle in the actual construction of flows in the ADM formalism
results from the lapse N and shift Ni which enter (9.53) as Lagrange multipliers.
At the level of the FRGE, which is based on an off-shell formalism, this leads to
(2)
a Hessian Γk which is degenerate. Imposing the proper-time gauge that would
fix N and Ni to their background values does not resolve this problem.
 9.5 Quantum ADM Gravity in the Einstein–Hilbert Truncation 239

An interesting route for addressing this obstacle efficiently is to start from

the background gauge-fixing procedure that we had used in the covariant
formulation:
∫
1 √
Γgf
k = dd x ḡ ḡ µν Fµ Fν . (9.57)
32πGk α

In the generalized harmonic gauge (4.54), Fµ is linear in the fluctuation field,

Fµ = D̄ν hµν − ϖ D̄µ h. (9.58)

Here h ≡ ḡ µν hµν , and α, ϖ are free gauge parameters. The harmonic gauge is
obtained by setting α = 1, ϖ = 21 while Landau-type gauges take the limit α →
0. By construction it is clear that all gauge-fixing terms in this class preserve
background diffeomorphism invariance.
The analogous gauge conditions in the ADM formalism are obtained by sub-
stituting the map (9.50) into (9.58). At the linear level, and for the specific
background (9.55), this results in

( )
F = 2∂τ Nb + ∂i Nb i − ϖ ∂τ 2Nb +σb ,
( ) (9.59)
bi + D̄j σ
Fi = ∂τ N bij − ϖ D̄i 2Nb +σb ,

where Fµ = (F, Fi ) has been decomposed in a spatial and time component. The
ghost action for this class of gauge fixings is again given by (4.55).

(3) At this stage it is instructive to combine the gravitational and gauge-fixing

terms and write down the part of the action quadratic in the fluctuation fields
on flat space, R̄ = 0. Abbreviating D̄2 ≡ σ̄ ij D̄i D̄j and setting α = 1 this yields
( )
(32πGk ) 12 δ 2 Γgrav
k + Γgf
k
∫
√ { [ ] ij [ ]
= dτ dd y σ̄ 12 σ bij −∂τ2 − D̄2 − 2Λk σ b +N b i −∂τ2 − D̄2 Nbi
[ ]
b (1 − 2ϖ2 )(−∂τ2 − D̄2 ) − Λk σ
− 21 σ b (9.60)
[ ]
+ 4N b (1 − ϖ)2 (−∂τ2 ) − ϖ2 D̄2 N b
[ ]
− 2N b 2ϖ(1 − ϖ)(−∂τ2 ) − (1 − 2ϖ2 )D̄2 − Λk σ b
( )}
− (1 − 2ϖ)(2N b +σ b) D̄i D̄j σ
bij + 2∂τ D̄i Nbi .

This shows that there is a unique gauge choice, corresponding to the harmonic
gauge ϖ = 21 , where all derivatives combine into the d-dimensional Laplacian on
flat space. We will adhere to this choice in the sequel.
240 Alternative Field Variables

(4) The beta functions governing the scale dependence of Gk and Λk are obtained
by restoring the intrinsic curvature terms in (9.60) and evaluating the resulting
operator traces via standard heat kernel techniques. The result has been obtained
in [285] and is conveniently expressed in terms of the dimensionless couplings λk ≡
Λk k −2 and gk = Gk k d−2 . Restricting to d = 3 + 1 dimensions and the optimized
regulator (5.120) it reads
∂t λk = βλ (gk , λk ), ∂t gk = βg (gk , λk ), (9.61)
where the beta functions are given by
βg (g, λ) = (2 + ηN ) g,
( )
g 36 − 6ηN 12 − 2ηN (9.62)
βλ (g, λ) = (ηN − 2) λ − 30 + 3ηN − − ,
24π 1 − 2λ 2 − 3λ
and the anomalous dimension of Newton’s coupling is
gB1 (λ)
ηN = . (9.63)
1 − gB2 (λ)
The functions B1 and B2 depend only on λ and read
( )
1 8 2 2 9
B1 = −9 − + − − ,
6π 1 − 2λ 2 − 3λ (1 − 2λ)2 (2 − 3λ)2
( ) (9.64)
1 24 6 4 18
B2 = − 11 − + − − .
72π 1 − 2λ 2 − 3λ (1 − 2λ)2 (2 − 3λ)2
(5) Clearly, the most important feature of the beta functions are their fixed points
where βλ (g∗ , λ∗ ) = 0, βg (g∗ , λ∗ ) = 0. The GFP is located at the origin, (λ∗ , g∗ ) =
(0, 0), and its stability coefficients are given by the classical mass dimension of
the couplings. In addition, there exists a unique NGFP,
NGFPADM : g∗ = 0.901, λ∗ = 0.222, g∗ λ∗ = 0.200. (9.65)
which comes with a complex pair of critical exponents,
NGFPADM : θ1,2 = 1.432 ± 2.586i, (9.66)
indicating that it acts as a spiraling UV attractor for the RG trajectories
in its vicinity. (This is the same characteristic behavior of the NGFP found
when evaluating the RG flow on foliated spacetimes using the Matsubara for-
malism [284, 286] and on Friedmann–Lemaı̂tre–Robertson–Walker backgrounds
[287, 288].)
It is interesting to compare the properties of this NGFP to the one seen in
QEG. Evaluating the fixed point equations based on (5.63) using the same gauge
fixing and the same cutoff shape function gives
NGFPQEG : g∗ = 0.707, λ∗ = 0.193, g∗ λ∗ = 0.137, (9.67)
 9.5 Quantum ADM Gravity in the Einstein–Hilbert Truncation 241

g g
1.4 1.4
1.2 1.2
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
l l
–1.0 –0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4 –1.0 –0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4

Figure 9.1. Phase portraits obtained by integrating the RG equations

from Quantum ADM Gravity (left) and QEG (right) in the Einstein–Hilbert
truncation using the optimized regulator and harmonic gauge.

and the critical exponents

NGFPQEG : θ1,2 = 1.475 ± 3.043i. (9.68)

Thus, the fixed point properties in the two cases are strikingly similar. The differ-
ence arises from the non-linearity of the map (9.50), which gives rise to additional
(2)
terms in Γk .

(6) The full phase diagram is obtained by integrating the system (9.61) numeri-
cally. It is displayed in the left diagram of Figure 9.1. The phase diagram resulting
from the QEG beta functions using the same gauge fixing and regulator profile
is shown in the right diagram for comparison.
In both cases the structure of the flow in the physically interesting region is
determined by three features: A singular line associated with a divergence of the
anomalous dimension ηN (thick dashed line) bounds the region to the right. The
flow to the left of this line is governed by the interplay of the NGFP and the
GFP. Flows in the vicinity of the GFP are characterized by gk ≪ 1 and exhibit
classical properties in the sense that the dimensionful couplings Gk and Λk are
(approximately) k-independent in this regime.
Depending on whether the trajectories flow to the left (right) of the separation
line (bold) the value of Λk in this classical regime is negative (positive). Following
the classification introduced in Section 5.3.4, these classes of solutions have been
termed Type Ia and Type IIIa, respectively. The single trajectory ending at the
GFP gives rise to a vanishing IR value of the cosmological constant and is referred
to as Type IIa.
The diagram also illustrates that the NGFP of Quantum ADM Gravity is
appropriate for providing the high-energy completion of the RG trajectories
leaving the classical regime for increasing k.
242 Alternative Field Variables

(7) Two of the most important results provided by the FRGE tailored to the
ADM formalism may be summarized as follows:

(i) Initial works on functional renormalization group flows in the ADM formal-
ism [284, 286] used a background S 1 × S d and proper time gauge, fixing the
lapse function and shift vector to their background values. The radius of
S d is taken to be time independent so that the (Euclidean) time direction
associated with the S 1 constitutes a global Killing vector field. This feature
allows us to analytically continue the flow equation to Lorentzian signature.
The fluctuations along the time direction can then be taken into account
through Matsubara sums, which, in contrast to the heat kernel computa-
tions, can be carried out in both Euclidean and Lorentzian signature. This
shows that the NGFP known from Euclidean signature computations persists
for Lorentzian signature [286].
(ii) The analysis of ADM-based RG flows on flat Friedmann–Lemaı̂tre–
Robertson–Walker backgrounds [287, 288] supports the existence of a NGFP
suitable for Asymptotic Safety. Supplementing the gravitational sector by
minimally coupled (non-selfinteracting) matter fields [287] showed that
many gravity-matter systems, including the matter content of the standard
model of particle physics, actually possess NGFPs as well. Depending on
the details of the matter sector the spiraling UV attractor seen in the pure-
gravity case is turned into a UV attractor with real critical exponents. The
overall picture obtained from the ADM framework thereby resembles the one
seen within the covariant approach [289–291] if the same approximations and
coarse-graining schemes are employed.

9.6 Observables in Quantum ADM Gravity

A primary motivation for considering the FRGE in the presence of a foliation
structure is its close relation to the Causal Dynamical Triangulation (CDT)
program [244].
The latter regularizes the gravitational partition function by introducing piece-
wise linear building blocks and evaluates the resulting sum using Monte Carlo
methods. The presence of a causal structure is thereby essential for obtaining
structures resembling a macroscopic spacetime from these elementary building
blocks [44, 46]. Since the simulations are capable of tracking a finite number
of building blocks only, CDT geometries have periodic boundary conditions in
the time direction and are necessarily compact. Their topology is fixed to either
S 1 × S 3 or S 1 × T 3 , where T 3 denotes the flat three-torus.

(1) A quantity that lends itself to a direct comparison of the FRG and CDT
approach is the profile of the physical volumes, V3 (τ ), as a function of Euclidean
time τ . From the FRGE perspective, this information can be obtained by solving
 9.6 Observables in Quantum ADM Gravity 243

the equations of motion arising from the Effective Average Action Γk . Utilizing
that the ansatz (9.53) is nothing but the Einstein–Hilbert action, the equa-
tions determining the self-consistent backgrounds are Einstein’s equations in the
presence of a k-dependent cosmological constant:

δΓk [χ̂; χ̄]

= 0 ⇐⇒ Rµν − 12 gµν R + gµν Λk = 0. (9.69)
δ χ̂ χ̂=0

(2) We first focus on the solutions of this equation for RG trajectories of Type
IIIa where Λk > 0 throughout. In this case, the solutions are given by four-
spheres:
( )
ds2 = r2 dτ 2 + r2 sin2 (τ ) dψ 2 + sin2 ψ dθ2 + sin2 ψ sin2 θ dϕ2 , (9.70)

with radius r2 = Λ3k . The time variable τ associated with the foliation takes values
τ ∈ [0, π), mimicking the periodic boundary conditions encountered in the CDT
framework.
The profile for the spatial volumes V3 (τ ) is obtained by integrating (9.70) over
the compact slices where τ is constant,

( )3/2
V3 (τ ) = 2π 2 3
Λk sin3 (τ ). (9.71)

Thus V3 (τ ) possesses the sin3 (τ )-profile characteristic for Euclidean de Sitter

space. In the classical regime, where Λk ≃ Λ0 is k-independent, V3 (τ ) inherits
this property. The volume is then set by Λ0 and can be used to identify the
underlying RG trajectory.
Typical volume profiles obtained as a self-consistent solution of the Effective
Average Action in the classical regime and from the CDT program are shown
in Figure 9.2. In both cases the τ -dependence of the volume distribution follows
the sin3 (τ )-profile.

(3) The RG trajectory of Type IIa, connecting the NGFP to the GFP, plays
a special role in the phase diagram shown in Figure 9.1. In this case the value
of the cosmological constant vanishes in the infrared, Λ0 = 0. The self-consistent
(compact) solutions of (9.69) then have the topology S 1 × T 3 and the spatial
slices are given by three-tori with arbitrary, time-independent volume V3 (τ ) =
V3 = const.
This solution together with the volume profiles measured on CDT configura-
tions with topology S 1 × T 3 are also shown in Figure 9.2. While in CDT the
total volume of the configuration is fixed by the number of simplices, the value
V3 determined from the Effective Average Action is a free parameter which can
be adjusted to match the CDT result.
244 Alternative Field Variables

V3(t) 〈nt〉
20 8000

15 6000

10 4000

5 2000

t t
0.0 0.5 1.0 1.5 2.0 2.5 3.0 20 40 60 80

Figure 9.2. Volume profiles obtained from solving the self-consistency con-
ditions of the Effective Average Action (left) and from the CDT simulations
[292–294] (right). The CDT results are obtained from simulations using
topologies S 1 × T 3 (horizontal line) and S 1 × S 3 (peaked line) and show the
expectation value ⟨nt ⟩ of the number of simplices found in a spatial slice. Their
FRGE counterparts result from selecting trajectories of Type IIa (horizontal
line) and Type IIIa (peaked line), respectively.

Following up on the comparison of spectral dimensions [295], the matching of

the long-distance properties arising from the microscopic quantum descriptions
of spacetime constitutes a prototypical example for connecting the FRG and
CDT frameworks in a systematic way.
 10
Matter Coupled to Quantum Gravity

The previous chapters focused on Asymptotic Safety in the context of pure grav-
ity. But a realistic description of our world should also include matter degrees
of freedom. At the level of the FRGE these can easily be incorporated by sup-
plementing the theory space of pure gravity by the fields carrying the matter
degrees of freedom. The UV completion based on a non-Gaussian fixed point
is not necessarily limited to gravity, and it is conceivable that the Asymptotic
Safety mechanism is also realized in suitable gravity-matter theories. Initiated in
[289, 290, 296] the systematic study of asymptotically safe gravity-matter sys-
tems is still in its infancy and we limit the discussion to the basic ideas and refer
to [297] for a recent overview.

10.1 Free Matter Fields in the Einstein–Hilbert Truncation

A natural starting point studies the effect of adding free matter fields on the flow
of Newton’s constant and the cosmological constant. In this case the gravita-
tional sector of the Effective Average Action is provided by the Einstein–Hilbert
truncation (5.18). The truncation ansatz is supplemented by NS scalar fields
ϕi , NV abelian vector fields Aiµ , and ND Dirac spinors ψ i , all of them non-
interacting among themselves but with a minimal coupling to gravity via the
scale-independent action

S matter = S scalar + S vector + S fermion . (10.1)

Here the scalar and fermionic contributions are

∫
1∑
N S
√
S scalar = dd x g ϕi (−D2 ) ϕi ,
2 i=1
(10.2)
ND ∫
∑ √
S fermion
=i d x g ψ̄ i ∇
d / ψi .
i=1
246 Matter Coupled to Quantum Gravity

The covariant derivative in the Dirac action is given by the O(d) covariant deriva-
tive (9.1) specialized to the case of Dirac fermions: ∇µ = ∂µ + 81 [γa , γb ] ω abµ . Here
the spin connection ω abµ is determined in terms of the vielbein ea µ in the usual
way,
ω abµ = eν a Γν σµ eσb + eν a ∂µ eνb , (10.3)
with Γα µν being the (torsionless) Christoffel symbol.1
The action for the vector fields comprises NV copies of the gauge-invariant
Maxwell action and the corresponding gauge-fixing and ghost contributions:
∫
1∑
N
V
√
S vector = dd x g g µν g αβ Fµα
i i
Fνβ
4 i=1
[ ] (10.4)
NV ∫
∑ √ 1 ( µν )
i 2
+ dd x ḡ i 2 i
ḡ D̄µ Aν + C̄ (−D̄ ) C .
i=1
2ξ

The field-strength tensor for this abelian U(1)NV gauge theory reads Fµν i
≡
∂µ Aν − ∂ν Aµ , and the kinetic term for the vector fields is supplemented by a
i i

gauge-fixing action and scalar ghost fields C̄, C exponentiating the (ḡ-dependent!)
Faddeev–Popov determinant. We adopt the Feynman gauge, setting ξ = 1 in the
sequel.
The beta functions of this truncation can be obtained following the com-
putation outlined in Section 5.2. The coupled RG equations controlling the
scale dependence of the dimensionless Newton constant gk and the cosmological
constant λk have the structure (5.63),
[ ]
∂t gk = βg (gk , λk ) ≡ d − 2 + ηN (gk , λk ) gk
(10.5)
∂t λk = βλ (gk , λk ).

The modified functions βλ (g, λ) and ηN (g, λ) depend parametrically on the

number of matter fields NS , NV , and ND now. Explicitly,
1 1−d/2
βλ = − (2 − ηN )λk + gk (4π)
[ 2
e 1 (−2λk )
× 2d(d + 1)Φ1d/2 (−2λk ) − d(d + 1)ηN Φ (10.6)
d/2
( ) ]
− 8dΦ1d/2 (0) + 4 NS + (d − 2)NV − 2d/2 ND Φ1d/2 (0) ,

while the anomalous dimension of Newton’s constant is given by

gk B1 (λk )
ηN (gk , λk ) = (10.7)
1 − gk B2 (λk )

1 For a more detailed discussion of structural aspects related to the inclusion of fermions in
the gravitational Effective Average Action we refer to [298, 299].
 10.1 Free Matter Fields in the Einstein–Hilbert Truncation 247

with the following functions of the cosmological constant:

1
B1 (λk ) ≡ (4π)1−d/2
3[
× d(d + 1)Φ1d/2−1 (−2λk ) − 6d(d − 1)Φ2d/2 (−2λk )
− 4dΦ1d/2−1 (0) − 24Φ2d/2 (0)
( )
+ 2 NS + (d − 2)NV − 2d/2 ND Φ1d/2−1 (0) (10.8)
( ) ]
− 3 4NV − 2d/2 ND Φ2d/2 (0) ,
1
B2 (λk ) ≡ − (4π)1−d/2
6[ ]
e1
× d(d + 1)Φ (−2λ k ) − 6d(d − 1) e 2 (−2λk ) .
Φ
d/2−1 d/2

Since the contribution of minimally coupled matter fields does not give rise to
terms proportional to the anomalous dimension of Newton’s constant, B2 (λk ) is
identical to the one in the Einstein–Hilbert truncation of pure gravity, (5.59).
Thus, including free matter fields leads to a two-parameter “deformation” of
the beta functions governing the scale dependence of Newton’s constant and
the cosmological constant in the pure gravity case. Introducing the ratio of the
threshold functions,
Φ2d/2 (0)
ΦM ≡ , (10.9)
Φ1d/2−1 (0)
the corresponding contributions to βλ and βg , i.e., to B1 , actually can be encoded
in two “deformation parameters,” namely2

dλ = NS + (d − 2)NV − 2d/2 ND ,
( ) (10.10)
dg = NS + (d − 2)NV − 2d/2 ND − 3 2 NV − 2d/2−1 ND ΦM .

Specifying to d = 4 and the optimized cutoff shape function (E.14) these relations
reduce to
dλ = NS + 2NV − 4ND ,
(10.11)
dg = NS − NV − ND .

For any choice of the shape function the map (10.10) assigns a particular set of
“coordinates” (dg , dλ ) to a given matter system. For example, the matter content

2 The relation between the number of matter fields and the parameters dλ and dg actually
depends on the choice of the coarse-graining operator. Following the terminology introduced
in [187], one may also resort to coarse-graining operators of Type II, which, besides the
Laplacian, also include specific endomorphism terms proportional to the Ricci scalar. In this
case one obtains dg = NS + (d − 8) NV + 2d/2−1 ND while dλ is unaffected. More details on
the fixed point structure of gravity-matter systems implementing the Type II regularization
can be found in [187, 291].
248 Matter Coupled to Quantum Gravity

of the standard model of particle physics contains 12 vector fields (eight gluons,
three gauge bosons for the weak interactions, and the photon), 45/2 fermionic
degrees of freedom (three generations of fermionic matter fields with the lightest
containing one Dirac field for electron, 1/2 Dirac field for left-handed neutrino,
three Dirac fields for the up quark (three colors), and three Dirac fields for the
down quarks), and four scalar fields. Evaluating the relations (10.11) thus assigns
the coordinates (dg , dλ ) = (−61/2, −62) to the standard model field content.
At this point it is instructive to study the fixed point structure of the beta
functions (10.5) as a function of the deformation parameters (dg , dλ ).
By solving the equation ηN (g∗NGFP , λNGFP
∗ ) = −2 for g∗NGFP and substituting the
NGFP NGFP
resulting expression into βλ (g∗ , λ∗ ) = 0 it follows that the beta functions
admit at most three NGFPs. Moreover, the resulting equation for λNGFP ∗ (dg , dλ )
possesses a singular locus at dg = 14 so that there are no fixed points for this
particular value.
The number of physically interesting NGFPs, i.e., fixed points situated at
g∗NGFP > 0 and λNGFP
∗ < 1/2, is shown in Figure 10.1. We see that for a generic
gravity-matter system the occurrence of zero or one NGFP is rather common.
Matter configurations supporting more than one NGFP appear in a compact
region close to the origin of the dg -dλ plane only.
Testing whether a NGFP may provide a suitable flow of Newton’s constant
and the cosmological constant includes determining its stability properties. A
NGFP will be denoted as UV-FP (IR-FP) if it attracts the (projected) RG flow
in its vicinity as k → ∞ (k → 0). In addition the NGFP can be a saddle point
coming with one UV-attractive and one UV-repulsive eigendirection.

100

50
dl

–50

–100
–100 –50 0 50 100
dg

Figure 10.1. Number of NGFP solutions of the system (10.5) as a function

of the deformation parameters dg and dλ . The system supports regions with
zero (white area), one (gray area), two (dark gray area), and three (light gray
area) NGFPs. There are no fixed point solutions on the line dg = 14.
 10.1 Free Matter Fields in the Einstein–Hilbert Truncation 249

The stability properties complementing the classification of the fixed points are
shown in Figure 10.2. The stability properties of the zero and one NGFP solutions
are classified in the left diagram. The dark-gray area extending from the upper-
left quadrant indicates a UV-FP with complex critical exponents, exhibiting
the spiraling attractor behavior typical for gravity without additional matter
fields. The two disjoint light-gray regions in the lower-left and upper-right corner
support UV-FPs with two real critical exponents. Saddle points or IR-FPs occur
within the gray wedge paralleling the dg -axis only.
The special values of dg , dλ giving rise to more than one NGFP are detailed
in the right diagram of Figure 10.2.
Notably, the classification provided in Figure 10.2 only relies on the numerical
value of the deformation parameters dg , dλ . As a consequence it is independent of
the cutoff shape functions used in the matter sector since changes in these quanti-
ties can be absorbed in a redefinition of dg and dλ . The location of a given matter
model within the dg -dλ -plane does depend on the specific regularization scheme.
For illustrative purposes, we use the map (10.11) to study the fixed point
structure arising for the matter field content of the standard model together
with its most common extensions. The results are summarized in Table 10.1.
We see that all matter systems considered are located in regions of the dg -dλ -
plane which support a single UV-attractive NGFP. In particular, the matter con-
tent of the standard model (and its minimal extensions) is located in the lower-left
quadrant. As a consequence the stability coefficients of the single UV-FP are real.
Note that the position and stability coefficients of the NGFPs appearing in this
sector are remarkably insensitive to the precise matter field content of the model.

100
5
50
0
dl

dl

0
–5
–50
–10
–100
–100 –50 0 50 100 0 10 20 30 40
(a) dg (b) dg

Figure 10.2. Stability properties of the NGFPs appearing in Figure 10.1. The
left diagram covers the cases of zero (white area) and one NGFP (gray shaded
regions). The shading indicates the occurrence of a UV-FP with complex
critical exponents (dark-gray area), a UV-FP with real critical exponents
(light gray area), or the existence of a single saddle point or IR-FP (gray
area). The right diagram magnifies the black region supporting two (open
symbols) or three NGFPs (filled symbols). The stability properties of the fixed
points associated with the filled squares are, respectively, UV/UV/saddle,
while the filled circles represent points with characteristics UV/saddle/IR.
The stability properties of the areas supporting two fixed points are UV/IR
(open diamond), UV/saddle (open triangle), UV/UV (open square), and
saddle/IR (open circles).
Table 10.1 Non-Gaussian fixed points of selected gravity-matter models as obtained with the Type I map. All models, apart from
the minimally supersymmetric standard model (MSSM) and the grand unified theories (GUT), sit in the lower-left quadrant of
Figure 10.2. All matter configurations, apart from the SO(10) GUT, possess a single UV-attractive NGFP with real critical
exponents. Except for the MSSM, the Type II map gives rise to the same qualitative picture.

Matter sector NS ND NV dg dλ g∗NGFP λNGFP

∗ θ1 θ2

Pure gravity 0 0 0 0 0 0.707 0.193 1.475 ± 3.043i

Quantum electrodynamics 0 1 1 −2 −2 0.836 0.157 1.226 ± 2.531i
45
Standard model (SM) 4 2
12 − 61
2
−62 0.894 −1.174 3.831 1.962
45
SM + dark matter scalar (DM) 5 2
12 − 59
2
−61 0.916 −1.185 3.829 1.963
SM + 3 right-handed neutrinos 4 24 12 −32 −68 0.863 −1.239 3.848 1.969
SM + 3 neutrinos + DM + axion 6 24 12 −30 −66 0.905 −1.264 3.845 1.970
61 13
Minimal supersymmetric standard model 49 2
12 2
−49 5.763 −6.424 3.933 2.135
SU(5) grand unified theory 124 24 24 76 76 0.156 0.371 12.51 6.972
SO(10) grand unified theory 97 24 45 28 91 0.139 0.368 8.131 ± 1.210i
 10.2 Interacting Gravity-Matter Fixed Points 251

At this stage the following remark is in order. At the time of writing the
investigation of Asymptotic Safety in gravity-matter systems is still in the early
stages. Nevertheless, the rather generic occurrence of UV-FPs is a strong indi-
cation that the Asymptotic Safety mechanism is not limited to the gravitational
interactions and may also be realized within gravity-matter systems. This result
is also supported by complementary and significantly more complex projections
[291, 300].
As we saw, the inclusion of minimally coupled matter fields leads to new
NGFPs with different stability characteristics. Conversely, it is expected that
the gravitational interactions at the NGFP turn on interactions in the matter
sector [301]. On this basis the fixed point solutions discussed in this section
should be seen as the projection of NGFPs on higher dimensional theory spaces
including such interactions onto the subspace where the matter-interactions are
switched off.

10.2 Interacting Gravity-Matter Fixed Points

A rather remarkable feature of interacting gravity-matter fixed points is that
their UV-critical surface may have a lower dimensionality than the one of the
corresponding Gaussian matter fixed point in the absence of gravity, thus giving
rise to an enhanced degree of predictivity.
In this section we illustrate this mechanism by considering standard Quantum
Electrodynamics (QED) coupled to asymptotically safe gravity, QEG. Follow-
ing the analyses in [271, 302, 303] we present evidence for the existence of an
asymptotically safe combined theory in which the fine-structure constant α is
computable from first principles.

10.2.1 QED on Minkowski Space

With the advent of perturbative renormalization theory in the late 1940s of
the last century QED matured to a physical theory of unprecedented predictive
power. With only two input parameters, the electron’s mass and charge, it can
explain a wealth of experimental data, often with spectacular precision. While
QED is considered the prototype of a perturbatively renormalizable theory, it
is natural to ask about its status beyond perturbation theory. Does it exist as
a fundamental theory? And more concretely, is there a non-Gaussian RG fixed
point with respect to which it is asymptotically safe?
Already in the early years of QED a non-trivial UV fixed point was speculated
about, albeit for a somewhat different reason [304]. The well-known one loop
formula for the renormalized charge,
( )
1 1 1 Λ
− 2 = 2 ln , (10.12)
e2ren eΛ 6π mren
252 Matter Coupled to Quantum Gravity

suggests that it might be difficult to obtain an interacting theory in the limit

when the UV cutoff Λ is removed: Keeping in (10.12) the value eΛ of the bare
charge fixed and sending Λ to infinity one finds that the renormalized charge eren
vanishes, so one is left with a trivial theory. Conversely, keeping eren fixed, it is
impossible to let Λ → ∞ since eΛ diverges at a finite value of Λ, the Landau pole.
Clearly, if the exact version of (10.12) displays a UV fixed point such that
eΛ → e∗ for Λ → ∞ the prospects for an interacting, cutoff-free theory are much
better. However, presumably no such fixed point exists. Comprehensive lattice
simulations [305–307] and studies using non-perturbative FRGE methods [308]
suggest that QED is very likely to be a trivial theory in four dimensions.
As we shall see, the situation changes when QED gets coupled to quantum
gravity: The combined theory QED + QEG appears to have significantly better
chances to exist non-perturbatively than QED on its own.

10.2.2 QED Coupled to QEG

Following [302] we describe the non-perturbative RG behavior of QED coupled
to QEG in terms of a three-dimensional truncation of theory space, including the
charge e(k), or equivalently the fine-structure constant α(k) ≡ e(k)2 /(4π), along
with Newton’s constant and the cosmological constant as running quantities. The
truncated EAA reads in the simplest case,
∫ ∫
1 d √ 1 √
Γk = d x g F µν F µν
+ dd x g (−R + 2λ̄k ) + · · · , (10.13)
4e(k)2 16πGk

where Fµν ≡ ∂µ Aν − ∂ν Aµ denotes the field-strength tensor of the abelian gauge

field Aµ , the photon. The dots in (10.13) represent the gauge-fixing and ghost
terms for both general coordinate and U(1) invariance, as well as the Dirac action
for the electron which, by assumption, contains no running parameters.
According to [302] the ensuing dimensionless RG equations can be approxi-
mated by the following simplified system:
[ ]
∂t g = βg ≡ 2 + ηN (g, λ) g,
∂t λ = βλ (g, λ), (10.14)
[ 6 ]
∂t α = βα ≡ A h2 (α) − Φ11 (0) g α,
π
with the coefficient
2
A≡ nF . (10.15)
3π
We consider a variant of quantum electrodynamics with nF “flavors” of electrons
here, and for simplicity, we stick to d = 4 dimensions. Several remarks are in order
at this point.
 10.2 Interacting Gravity-Matter Fixed Points 253

(1) The first two equations in (10.14) are taken to be those of pure gravity in the
Einstein–Hilbert truncation [14] which we discussed in Section 5.2.2. Neglecting
the backreaction of the matter fields on the renormalization in the gravity sec-
tor is at least partially justified by the above investigations of free matter fields
coupled to gravity. They show that a Maxwell field and one or a few Dirac fields
do not qualitatively alter the RG flow of g and λ.

(2) Following (5.65), the anomalous dimension ηN can be expanded for small g
( )
ηN ≡ B1 (λ) h1 (λ, g) = B1 (λ) g + B2 (λ)g 2 + · · · , (10.16)

with the coefficient functions B1 and B2 given in (5.59). From the analysis there
we know that B1 (λ) < 0 for all λ and that g(k) does not change very much if one
approximates λ(k) ≈ 0, B1 (λ) ≈ B1 (0), B2 (λ) ≈ 0, whence

ηN (g, λ) ≈ B1 (0)g < 0. (10.17)

(3) The beta function for α in the third equation of (10.14) involves a pure
matter part, written as A h2 (α) α, and as the most important new ingredient
the quantum gravity contribution ∝ g obtained in [271]. The former part is well
known from perturbation theory, its first two terms being (for nF = 1)
[ ( ) ]
2 α 1 ( α )2
βα (α)|g=0 ≡ A h2 (α) α = α + + O(α3 ) . (10.18)
3 π 2 π
For a qualitative understanding it will be sufficient to employ the one-loop
approximation, i.e.,
h2 (α) = α. (10.19)
Indeed, the lattice and flow equation studies mentioned above indicate that there
exists no non-trivial continuum limit for QED (without gravity), and this means
that βα (α)|g=0 has no zero at any α > 0. Therefore, h2 (α) = βα /(Aα) starts out
as h2 (α) = α in the perturbative regime α . 1, and for larger α it is still known
to be an increasing function: h′2 (α) > 0. To be able to solve the RG equations
analytically we set h2 (α) = α for all values of α. This is a qualitatively reliable
approximation since, as we will see, at most a zero of h2 (α) could change the
general picture.

(4) As it stands, βα applies only above the threshold due to the mass of the
electron at k = me . At low scales k . me the fermion loops no longer renormalize
α. In the full-fledged EAA this decoupling is described by a certain threshold
function. Here a simplified description will be sufficient where we set A = 0 if
k < me .

(5) The most interesting term in the RG equations is the contribution to βα

proportional to the product g α which describes the renormalization of the gauge
254 Matter Coupled to Quantum Gravity

coupling due to graviton loops. Remarkably enough, its sign is such that it coun-
teracts the fermionic contribution, i.e., it opposes the electromagnetic screening
of charges. It drives α(k) towards smaller values for increasing k, in particular
when k → ∞.
In the EAA framework the gravity contribution to βα was first derived in
[271, 303] where both abelian and non-abelian gauge fields were considered.
In either case it was found that the pure gauge theory (no fermions) becomes
asymptotically free when coupled to QEG, i.e., α(k) → 0 for k → ∞.3

10.3 Two non-trivial Fixed Points

Let us start the analysis of the RG equation (10.14) by finding its fixed points
(g∗ , λ∗ , α∗ ). First of all there exists an obvious Gaussian fixed point, GFP,
situated at g∗ = λ∗ = α∗ = 0.
Furthermore we know that the subsystem of flow equations for pure gravity
in the Einstein–Hilbert truncation, (5.63), admits for a NGFP at (g0∗ , λ∗0 ) ̸= 0.
This fixed point lifts to a NGFP of the full system located at (g0∗ , λ∗0 , α∗ = 0).
We denote it by NGFP1 . This fixed point is trivial from the QED perspective;
the electromagnetic interaction is “switched off” there, while the gravitational
selfinteraction is the same as in pure gravity.
There exists a second NGFP though, non-trivial also in the QED sense,
provided the equation
6 g∗ 1
h2 (α) = Φ (0) (10.20)
πA 1
admits a solution for some α = α∗ ̸= 0. We demonstrate below that this is indeed
the case and that α∗ > 0. This fixed point will be called NGFP2 .

(1) We are particularly interested in an Asymptotic Safety scenario at the second

NGFP. In its vicinity the linearized flow is governed by
∑ ( )
∂t ui (k) = Bij uj (k) − u∗j , (10.21)
j

where u ≡ (g, λ, α), and the stability matrix is of the form

 
∂g βg ∂λ βg 0
 
B = (Bij ) = ∂g βλ ∂λ βλ 0  . (10.22)
∂g βα ∂λ βα ∂α βα u=u∗

3 The same behavior occurs also in an earlier perturbative calculation [309]; see also [310–318]
for a controversial discussion of different perturbative schemes.
 10.4 Approximate Analytical Solution 255

Two of its eigenvalues coincide with those of pure gravity in the Einstein–Hilbert
truncation, giving rise to the familiar two UV-attractive directions [160, 173]. The
third eigenvalue is given by
( )
6
∂α βα (u∗ ) = Ah2 (α∗ ) − Φ11 (0)g∗ + Ah′2 (α∗ )α∗ = Ah′2 (α∗ )α∗ . (10.23)
π
As α∗ > 0, the sign of ∂α βα (u∗ ) agrees with the sign of h′2 (α∗ ).
At this point we exploit the information from lattice and flow equation studies
which tried to find a non-trivial continuum limit of QED without gravity. We
saw that their negative results suggest that h′2 (α∗ ) > 0 holds true even beyond
perturbation theory.
As a consequence, the third eigenvalue ∂α βα (u∗ ), corresponding to the α
direction in g-λ-α-theory space, is UV repulsive. With two UV-attractive
and one repulsive direction the UV-critical hypersurface SUV pertaining
to NGFP2 is a two-dimensional surface in a three-dimensional space, i.e.,
dim SUV (NGFP2 ) = 2.

(2) Let us also determine the critical exponents of the other fixed point NGFP1 .
As βg and βλ do not depend on α in our approximation the first two of them
remain the same as for NGFP2 . However, for the third eigenvalue we now obtain
( )
6 1 6
∂α βα (u∗ ) = Ah2 (α∗ ) − Φ1 (0)g∗ + Ah′2 (α∗ )α∗ ∗= − Φ11 (0)g∗ < 0.
α =0
π π
(10.24)
Hence, the third direction is UV attractive as well, and NGFP1 has a three-
dimensional UV-critical hypersurface, dim SUV (NGFP1 ) = 3.
We conclude that dim SUV (NGFP2 ) < dim SUV (NGFP1 ). This reflects an
enhanced predictivity of the Asymptotic Safety construction with respect to
NGFP2 as compared to NGFP1 .

10.4 Approximate Analytical Solution

Let us now analyze the flow in a simple analytically tractable approximation. For
that we expand the functions h1 and h2 to first order in g and α, respectively:

h1 (g) = g + O(g 2 ) and h2 (α) = α + O(α2 ). (10.25)

Furthermore, we neglect the running of the cosmological constant and fix λ = λ0

to a constant value. The remaining system of flow equations reads
[ ]
∂t g = 2 + B1 (λ0 ) g g,
( )
6 (10.26)
∂t α = Aα − Φ11 (0) g α.
π
256 Matter Coupled to Quantum Gravity

In this approximation the existence of the NGFP2 is obvious and we can read
off its coordinates:
2 6 g∗ 1
g∗ = − >0 and α∗ = Φ (0) > 0. (10.27)
B1 (λ0 ) πA 1

Note that as we already anticipated α∗ indeed turns out positive.

In the sequel we eliminate the constant B1 (λ0 ) in favor of g∗ , setting B1 (λ0 ) =
−2/g∗ everywhere.
The solution to the decoupled first equation in (10.26) then reads:

G0 k 2
g(k) = 2 . (10.28)
1 + Gg0∗k

The constant of integration G0 ≡ limk→0 g(k)/k 2 can be interpreted as the IR

value of the running Newton constant. The solution (10.28) for g shares a crucial
feature with any asymptotically safe trajectory of the exact system for pure
gravity, namely that it connects the classical regime g(k) ≈ G0 k 2 for k ≪ mPl ≡
−1/2
G0 and the fixed point regime g(k) ≈ g∗ for k ≫ mPl .
The simplified second RG equation (10.26) for α is a differential equation of the
Riccati type which can be solved in closed form without specifying the function
g(k). Its general solution reads, with Φ11 ≡ Φ11 (0),
( ∫ ) ∫ k ( ∫ )
1 1 6 1 k g(k ′ ) ′ 6 1 k g(k ′′ ) ′′ dk ′
= exp Φ dk − A exp Φ dk ,
α(k) α0 π 1 k0 k ′ k0 π 1 k′ k ′′ k′
(10.29)
where α0 = α(k0 ) is the value of the fine-structure constant at a fixed reference
scale k0 . If we insert the function g(k) of (10.28) we can perform the integrations
in (10.29) and find

1 ( 2
) π3 Φ1 g∗
1

= gg∗∗ +G0k
2
+G0 k0
α(k)
[ ( ) ( )]
× α10 − α1∗ 1 + Gg0∗k2 2 F1 1, 1, 1 + π3 Φ11 g∗ ; − Gg0∗k2 (10.30)
( ) (
0
) 0

+ α∗ 1 + G0 k2 2 F1 1, 1, 1 + π Φ1 g∗ ; − G0 k2 ,
1 g∗ 3 1 g∗

where 2 F1 (a, b, c; z) denotes the (ordinary) hypergeometric function.

From (10.30) we infer that there exist three kinds of possible UV behaviors for
α(k). The (k-independent) value of the terms inside the square brackets [ · · · ] on
the right-hand side of (10.30) determines which one of them is realized.

(1) For a strictly positive value [ · · · ] > 0 we see that limk→∞ α(k) = 0 as the pref-
6 1
actor of [ · · · ] diverges proportional to k π Φ1 g∗ when k → ∞. This corresponds to
 10.4 Approximate Analytical Solution 257

an asymptotically free fine-structure constant. The corresponding RG trajecto-

ries of the full system are asymptotically safe with respect to NGFP1 .

(2) For a negative value [ · · · ] < 0 there exists a scale kL at which the two terms
on the right-hand side of (10.30) cancel, so that α diverges at the finite scale
k = kL . This amounts to a Landau type singularity.

(3) A third type of limiting behavior is obtained for the case that the bracket van-
ishes exactly: [ · · · ] = 0. We are then left with the second term on the right-hand
side of (10.30), and since 2 F1 (a, b, c; 0) = 1, we find limk→∞ α(k) = α∗ , corre-
sponding to an asymptotically safe trajectory with a non-zero gauge coupling at
NGFP2 . This is precisely the behavior to be expected due to the UV-repulsive
direction of the fixed point. Since for k → ∞ the trajectory will only flow into
NGFP2 for one specific value of α0 , this value of α0 , and hence the whole tra-
jectory α(k), can be predicted under the assumption of Asymptotic Safety with
respect to NGFP2 .
The situation is illustrated by the g-α-phase portrait in Figure 10.3. Bearing
in mind that the arrows always point toward the IR, we see that NGFP1 is
IR repulsive in both directions shown, while NGFP2 is IR attractive in one
direction. This is consistent with our earlier discussion which showed that in the
three-dimensional g-λ-α-space NGFP1 has three and NGFP2 has only two IR
repulsive (or equivalently, UV attractive) eigendirections.
In Figure 10.3, the trajectories inside the GFP-NGFP1 -NGFP2 triangle are
those corresponding to the case [ · · · ] > 0 above; they are asymptotically safe with
respect to NGFP1 . The NGFP2 → GFP boundary of this triangle is the unique

2.5

2.0

1.5 NGFP1 NGFP2

g

1.0

0.5
GFP
0.0
0 5 10 15 20 25
a

Figure 10.3. The RG flow on the g-α-plane implied by the simplified equa-
tions (10.26). It is dominated by two non-Gaussian fixed points. (Adapted
from [302].)
258 Matter Coupled to Quantum Gravity

trajectory (heading toward smaller g and α values), which is asymptotically safe

with respect to the second non-trivial fixed point, NGFP2 .
The diagram in Figure 10.3 corresponds to a massless electron for which A
keeps its non-zero value at arbitrarily small scales. In reality the α-running due
to the fermions stops near me , of course.

10.5 Asymptotic Safety Construction at NGFP2

Let us investigate the unique asymptotically safe trajectory emanating from
NGFP2 in more detail. We note that the condition of a vanishing bracket [ · · · ]
in (10.30) is self-consistent in the sense that the resulting function α0 (k0 ) is of
identical form as the remaining function α(k):
( ) ( )
1 1 g∗ 3 1 g∗
= 1+ 2 F1 1, 1, 1 + Φ g∗ ; − . (10.31)
α(k) α∗ G0 k 2 π 1 G0 k 2

(1) Let us approximate this function for scales k ≪ mPl much below the Planck
−1/2
scale mPl ≡ G0 . Later on we will need it at k = me , for instance, where
( )2
me is the mass of the electron. Then the argument G0gm ∗
2 = g∗
mPl
me ≈ 1044
e
is extremely large and this will be an excellent approximation. Hence, we may
safely truncate the general series expansion of the hypergeometric function,
∑∞
Γ(c) −a (a)n (1 − c + a)n −n
2 F1 (a, a, c; z) = (−z) z
Γ(a)Γ(c − a) n=0
(n!)2
( )
× ln(−z) + 2ψ(n + 1) − ψ(a + n) − ψ(c − a − n) ,

after its first term, and approximate the resulting factor 1 + G0 k 2 /g∗ ≈ 1, such
that our final result for scales k ≪ mPl reads:
[ ( ) ( )]
1 g∗ 3 1 g∗ 3 1
= · Φ1 · ln − γ − ψ Φ g∗ . (10.32)
α(k) α∗ π G0 k 2 π 1
Here ψ denotes the Digamma function and γ is Euler’s constant. Using (10.27)
in order to re-express g∗ /α∗ in terms of A ≡ ( 3π
2
)nF we can also write (10.32) in
the following form:
[ ( ) ( )]
1 A g∗ 3 1
= ln −γ−ψ Φ g∗ . (10.33)
α(k) 2 G0 k 2 π 1

For k ≪ mPl , we recover the logarithmic running α(k)−1 = −A ln k + const famil-

iar from pure QED.

(2) For k comparable to, or even larger than, the Planck mass the gravity cor-
rections set in, stop the logarithmic running of α(k) and cause the coupling to
 10.6 Numerical Solution 259

freeze at a finite value α(k → ∞) = α∗ . Obviously, along this RG trajectory no

Landau pole singularity is encountered!

(3) Equation (10.33) shows that α(k) ∝ 1/nF at all scales. As a consequence, if
we consider a toy model with a large number of electron flavors, all α-values that
appear along the RG trajectory can be made arbitrarily small, which renders
perturbation theory in α increasingly precise. At the fixed point we have, for
instance,
g∗
α∗ = 9 Φ11 (0) . (10.34)
nF
(4) In an Asymptotic Safety scenario based on the fixed point NGFP2 , the
infrared value of the fine-structure constant αIR ≡ limk→0 α(k) is a computable
number.
Using (10.33) to calculate αIR we must remember however that as it stands it
holds true only for k & me . When k drops below the electron mass the standard
QED contribution to the running of α(k) goes to zero, and the gravity corrections
are negligible there anyway. Hence, approximately, ∂t α(k) = 0 for 0 ≤ k . me .
Thus (10.33) leads to the following prediction for αIR ≈ α(me ):
[ ( ) ( )]
1 A mPl 3 1
= 2 ln + ln(g∗ ) − γ − ψ Φ g∗ . (10.35)
αIR 2 me π 1

As the fixed point coordinates are an output of the RG equations, the only
input parameter needed to predict αIR in this approximation is the electron
mass expressed in Planck units, me /mPl .

(5) On the other hand, if QED + QEG is not asymptotically safe with respect
to NGFP2 then its RG trajectory is one of those inside the GFP-NGFP1 -
NGFP2 triangle in Figure 10.3. In this case it is asymptotically safe with respect
to the other non-trivial fixed point, NGFP1 . The resulting quantum field theory
is also free from divergences and valid at all energies. In this case the U(1)
coupling is not a prediction but an experimental input, though.

10.6 Numerical Solution

The analytically accessible RG equations solved above follow from the original
systems of equations (10.14) by a number of additional approximations. Staying
within the one-loop approximation of βα , it has been checked with numerical
methods that these additional approximations cause only minor changes of the
flow. The analytic solution correctly describes all qualitative properties of the
RG flow, and is precise even at a “semiquantitative” level [302].
The numerical solution confirms the existence of the two non-Gaussian fixed
points NGFP1 and NGFP2 as well as their stability properties. In accord
260 Matter Coupled to Quantum Gravity

l a
–0.5 0 0.5 10 5 0

2.0 2.0

g g
1.0 1.0

10 0.5
5a l 0
0 –0.5 0.0
0.0

2.0

g 1.0

0.0
10 0.5
5 0
α l
0 –0.5

Figure 10.4. The diagrams show three projections of the same family of
trajectories running inside the UV-critical surface SUV of NGFP2 in g-λ-α-
space. They are seen to form a two-dimensional surface in a three-dimensional
theory space. (Taken from [302].)

with the discussion above, NGFP2 is seen to have two UV-attractive and one
repulsive direction. All RG trajectories approaching NGFP2 for k → ∞ lie in its
two-dimensional UV-critical surface SUV . This surface is visualized in Figure 10.4
by a family of trajectories starting on SUV close to the FP, which were traced
down to lower scales k.

10.7 Can We Compute the Fine-Structure Constant?

(1) To summarize: We coupled Quantum Electrodynamics to quantized gravity
and explored the possibility of an asymptotically safe UV limit of the combined
system. Using a three-parameter truncation we found evidence indicating that
this is indeed possible. There exist two non-trivial fixed points which lend them-
selves for an Asymptotic Safety construction. Using the first one, NGFP1 , the
ultraviolet value of the fine-structure constant is zero. Its infrared value αIR is a
free parameter which is not fixed by the theory itself but has to be taken from
 10.7 Can We Compute the Fine-Structure Constant? 261

experiment. Basing the theory on the second non-Gaussian fixed point NGFP2
instead, the fixed point value α∗ is non-zero, and the (“renormalized”) low-energy
value of the fine-structure constant αIR can be predicted in terms of the electron
mass in Planck units.

(2) In either case the coupled theory QED + QEG is well behaved in the ultra-
violet. In particular, there is no Landau pole singularity, and no triviality problem,
i.e., the continuum limit is an interacting theory.
We see that its coupling to quantum gravity cures the notorious UV problems
standard QED is suffering from despite its perturbative renormalizability.

(3) On theory spaces with more than one eligible fixed point a comparison of pre-
dictions to experimental data can reveal in principle which of them is the actual
UV fixed point. It is therefore tempting to insert numbers into our prediction
(10.35) for the low-energy value of α.
With me = 5.11 × 10−4 GeV and mPl = 1.22 × 1019 GeV one finds me /mPl =
4.19 × 10−23 , and the cutoff (E.13), for example, yields Φ11 = 1. The coordinate
g∗ = −2/B1 (λ0 ) depends on the value chosen for λ0 . For λ0 = 0 or the fixed point
value of λ in the Einstein–Hilbert truncation, λ0 = λ∗ ≈ 0.193, we get g∗ ≈ 1.71
or g∗ ≈ 0.83, respectively. From that we obtain the result

1 λ0 =0 1 λ0 =λ∗
≈ 10.91 nF or ≈ 10.96 nF . (10.36)
αIR αIR

It is remarkably insensitive to the value of g∗ and/or Φ11 .

For nF = 1, this estimate differs from the fine-structure constant measured in
real nature, α ≈ 1/137, by a factor of roughly 13.
It is clear though that even within the limits of our crude approximation
(10.25) a serious comparison with experiment must include the renormalization
effects due to the other particles besides the electron, all those of the standard
model, and possibly beyond. Within the “nF flavor QED” considered here we
could mimic their effect by appropriately choosing nF . We would then conclude
that the observed αIR is consistent with Asymptotic Safety at NGFP2 if nF = 13.

(4) Let us turn to the full standard model finally. Applying the above discussion
to the weak hypercharge rather than the electromagnetic U(1) one again has a
one-loop equation of the type ∂t α1 = Aα12 , but this time with A = 41/(20π) and
α1 ≡ 5α/(3 cos2 θW ), where θW is the Weinberg angle [319].
It is convenient to compare the prediction of Asymptotic Safety to the exper-
imental value for α at the Z mass. From (10.35) with the new value of A, and
me replaced by MZ , we obtain (with λ0 = 0)

α1SM (MZ ) ≈ 1/25.7. (10.37)

262 Matter Coupled to Quantum Gravity

This prediction differs from the experimental value α1exp (MZ ) ≈ 1/59.5 by only
about a factor of two.
The similarity of the two numbers is certainly quite impressive if one recalls
that the rather simple RG equations we employed here must bridge more than
22 orders of magnitude in scale.
Nevertheless, we are not yet in the position to say whether the gravity-matter
system observed in nature is actually asymptotically safe with respect to a gener-
alization of NGFP1 or NGFP2 . It goes without saying that in order to address
this question much more accurate RG calculations for more general field systems
are needed.

(5) The above discussion of the coupled theory QED + QEG also sheds light on
another puzzle related to the four fundamental interactions, namely the hierarchy
problem, i.e., the question of why the typical energy scales of particle physics to
which the standard model applies are so much smaller than the Planck scale,
mPl = 1.22 × 1019 GeV [320]. Within the present setting, for example, it is natural
to ask why in real nature there is the huge hierarchy between the electron mass
me and the Planck scale.
If we adopt the point of view that the low-energy quantities αIR ≈ 1/137 and
me are known from the experiment, then (10.33) or (10.35) can be used to pre-
dict the value of mPl /me . Using only non-gravitational input data, αIR and me ,
this leads without further ado to a ratio mPl /me which is naturally large, as a
consequence of the logarithm in (10.35). The number we find is not too different
from the experimental value mPl /me ≈ 2.39 × 1022 , and comes even closer when
the argument is applied to the hypercharge.

(6) In conclusion it can be said that the present example provides a proof of prin-
ciple which indicates that, by coupling the standard model to asymptotically safe
gravity, it might well be possible to compute some of its otherwise undetermined
free parameters from first principles and to understand the scales it displays.
Moreover, according to the analysis in [321] also its Higgs sector acquires a
higher degree of predictivity and this allowed for an Asymptotic Safety-based
prediction of the Higgs mass.
 11
Towards Phenomenology

In this chapter we review the first steps undertaken in order to extract physical
information directly from the RG flow. Focusing on how the quantum gravity
effects might modify the classical picture of spacetime, the investigations are
restricted to the level of expectation values. They involve further approximations
on top of the truncations of theory space and therefore come with a higher
degree of uncertainty presumably. For the time being much less is known about
the observable predictions of QEG than about its RG flows. Thus, the results
presented here should still be considered preliminary.

11.1 Scale-Dependent Riemannian Structure

Let us fix an RG trajectory for the EAA as well as a manifold M with a given
topological and smooth structure to model spacetime. Then the self-consistent
metrics that can persist on this manifold are determined by the system of tadpole
equations (4.105). If we focus on the sector with vanishing ghosts only the simpler
equation (4.106) needs to be solved. Let us assume there is a solution ḡksc which
depends smoothly on k, either for a certain range of scales only, or even for all k ∈
[0, ∞). This solution promotes M to an entire family of Riemannian manifolds
whose members are labeled by the scale: (M, ḡksc ). If it is based on a complete
asymptotically safe RG trajectory defined for all k ∈ [0, ∞), this family, as a
whole, can be regarded as the metric structure of a “quantum spacetime” at the
mean field level.
Heuristically, one would try to interpret the family (M, ḡksc ) in the following
way. Assuming that Γk defines an effective field theory valid near typical scales
of order k, the infinitely many metrics ḡksc all refer to the same physical system,
the “quantum spacetime,” but describe its effective metric structure on different
scales. An observer using a “microscope” with a resolving power ℓ ≈ k −1 will
perceive the universe to be a Riemannian manifold with a particular metric
expectation value, namely gµν = ḡµν ≡ ḡkscµν . Even though at every fixed k, ḡksc
264 Towards Phenomenology

is a smooth classical metric the quantum spacetime as a whole can acquire very
non-classical and in particular fractal-like features due to the scale dependence
of the metric.
The family of metrics ḡkscµν can be determined from the tadpole equation result-
ing from the EAA. For bi-metric truncations this equation boils down to (5.11),
and for the special case of single-metric truncations it reads simply

δ
Γ̄k [g] = 0. (11.1)
δgµν (x) sc
g=ḡk

Becoming even more specific, for the single-metric and bi-metric Einstein–
Hilbert truncations, defined by (5.21) and (6.10), respectively, the corresponding
tadpole equations have the form of Einstein’s equation in classical General
Relativity, with a k-dependent Newton’s coupling and cosmological constant
though:
1
Rµν (ḡksc ) − R(ḡksc ) ḡkscµν = −Λk ḡkscµν + 8πGk ⟨Tµν ⟩k . (11.2)
2

(1)
Here Λk stands for λ̄k and Λk in the single- and the bi-metric Einstein–
Hilbert truncation, respectively. Moreover, ⟨Tµν ⟩k is the energy momentum
tensor describing the matter system, if any. Supplementing (11.2) by the equa-
tion of motion for the matter fields, also implied by the EAA, leads to a closed
system of equations which allows to determine {ḡkscµν }.

11.2 Cutoff Identifications

Having some family of metrics {ḡkscµν } at our disposal there are various ways to
proceed. The first one tries to extract information directly from the full stack of
metrics, without ascribing any specific meaning to the parameter which labels
them, k.
A second possibility is the approach of the cutoff identifications which tries to
identify k with a certain geometrical or dynamical scale. The idea is the same as
in Section 2.3 where we saw how sometimes decoupling arguments can be used
to “guess” difficult terms in Γk=0 from a simpler truncation provided we know
something about the relevant threshold, if any. Analogously it may be possible
to find the metric expectation value proper, ⟨b g (x)⟩ ≡ ḡk=0
sc
(x), by inserting an
appropriate function k = k(x) into the family: ḡk=k(x) (x) = ⟨b
sc
g (x)⟩.
The “resolving power” ℓ of the microscope is in general a complicated function
of k, and the challenge is to express ℓ and the cutoff by the “correct” function
on spacetime:
xµ 7→ k(xµ ). (11.3)
 11.2 Cutoff Identifications 265

On general grounds this function should behave as a scalar under a change

of coordinates. The details of this map strongly depend on the particular
physical situation under consideration. Explicit maps may be obtained from
symmetry considerations, spectral arguments [322], or imposing conditions
like the conservation of the stress-energy tensor [323, 324]. Typical choices
employed in the literature relate the scale k to the proper distance between
two points d(x, x′ ), computed from a classical reference metric, physical quanti-
ties like the Hubble scale H(t), or a scalar quantity constructed from curvature
invariants:

Type I: k 2 = ξ 2 d(x, x′ )−2 , (11.4)

2 2 2
Type II: k = ξ H(t) , (11.5)
√
Type III: k 2 = ξ 2 Rµνρσ Rµνρσ . (11.6)

Here ξ is a free constant parameter of order one. Substituting the identifica-

tion k ≡ k(x) into the family ḡkscµν results in a single effective metric for the
spacetime geometry. At scales where Λk and Gk are (approximately) scale-
independent this effective metric reduces to the one of General Relativity. The
running of the couplings induces distinct modifications controlled by the beta
functions and the choice of k(x). Figure 5.2 indicates that these corrections will
set in when k 2 & m2Pl generically, which is a natural scale for quantum gravity
effects.
An alternative to the RG improvement of the equations of motion described
above can be borrowed from QED and QCD [325–327]. It consists in carrying
out the cutoff identification k → k(x) at the level of the running action and
deriving the equations of motion thereafter [328–330]. Instead of calculating
complicated terms in the effective action directly, a shortcut is sometimes avail-
able. Symmetry and decoupling arguments as in Section 2.3 may be used then
to obtain those terms by inserting a certain field dependent cutoff identifica-
tion k(x) into a comparatively simple k-dependent action, typically comprising
only the perturbatively relevant terms with the corresponding scale depen-
dence. The leading-log approximation of QED and QCD in strong slowly varying
fields, for example, can be obtained in this manner, letting k 4 → Fµν (x)F µν (x)
[326, 327].
Applying the same approach to gravity allows to construct improved actions
by promoting k to an external field on spacetime directly, k = k(xµ ), or by iden-
tifying it with a function of the field strength. The latter corresponds to a cutoff
identification of Type III then. This procedure leads to additional terms in the
equations of motion. A more detailed discussion of the approach can be found in
[328–333].
In view of the somewhat heuristic character of these strategies it needs to
be emphasized that in principle cutoff identifications can be avoided completely,
albeit at the expense of by far harder explicit calculations.
266 Towards Phenomenology

11.3 Structure of “Quantum Spacetime”

We start by discussing the implications of (11.2) for the case of an empty space-
time without matter. For concreteness we base the discussion in this chapter on
the single-metric Einstein–Hilbert truncation. Occasionally the resolving power
of the microscope is approximated by ℓ ≈ 1/k, implementing (11.4) on not too
strongly curved spacetimes.

(1) Example: RG trajectories of Type IIIa. Considering solutions of (11.2)

that exist on all scales from the UV down to k = 0, the simplest situation arises
when all metrics ḡksc , k ∈ [0, ∞), can be put on the same manifold M. Then the
“running spacetime” can have the same topological structure on all scales and the
delicate issue of a topology change does not arise. This situation is realized along
the RG trajectories of the Type IIIa. Along them, the cosmological constant is
always positive, Λk > 0, k ∈ [0, ∞); see Section 5.3.4.
Under those circumstances any fixed metric g̊µν on M that solves Ein-
stein’s equation with cosmological constant Λ̊, i.e., Gµν (g̊) = −Λ̊g̊µν , immediately
implies a family of self-consistent backgrounds [322]:

Λ̊
ḡkscµν = g̊µν . (11.7)
Λk

Taking advantage of the identity Rµν (cg) = ( 1c )Rµν (g), valid for any constant
c > 0, we see that the rescaled metrics (11.7) indeed solve the tadpole equation
(11.2).
The Type IIIa trajectories feature a classical regime at low scales where the
dimensionful couplings do not run. Henceforth we identify Λ̊ with the value of
the cosmological constant in this regime, and interpret g̊µν as the metric of the
classical macroscopic world. When we increase k from the IR towards higher
scales, Λk increases monotonically. As a consequence, the running self-consistent
metric “shrinks” relative to the classical one. While Λk ∝ k 4 in the semiclassical
regime, near k ≈ mPl the trajectory enters the fixed point regime where

1 Λ̊
Λk = k 2 λ∗ , and ḡkscµν = g̊µν . (11.8)
k 2 λ∗

The behavior (11.8) applies to all asymptotically safe RG trajectories near the
NGFP.

(2) Self-similarity in the fixed point regime [163]. Let us consider a refer-
ence spacetime (M,g̊) which displays a certain feature (radius of curvature, etc.)
with a characteristic length L0 . We consider L0 a proper distance with respect
to g̊, and assume that it is determined by the only dimensionful quantity
√ in Ein-
stein’s equation, Λ̊. Hence, up to factors of order unity, L0 ≈ 1/ Λ̊. Then, by
 11.3 Structure of “Quantum Spacetime” 267

(11.7) the very same feature, but this time measured with the metric ḡksc has the
proper size:
√ √
Λ̊ 1
L(k) = L0 ≈ . (11.9)
Λk Λk

If we explore spacetime at a certain length scale, ℓ, characteristic of the experi-

ment or observation performed, the action Γk at, roughly, k ≈ 1/ℓ, should provide
the corresponding effective field theory description.
So, the characteristic length of the spacetime observed, measured with
the optimally suited metric ḡksc k≈ℓ−1 , is perceived to be L(ℓ) ≡ L(k = ℓ−1 ) ≈
−1/2
(Λk=ℓ−1 ) .
In the special case of scales belonging to the fixed point regime where Λk ∝
k 2 ∝ 1/ℓ2 we thus have

L(ℓ) ∝ ℓ. (11.10)

This is a remarkable result: when we observe a given structure in spacetime with

a “microscope” of variable resolution ℓ ≪ ℓPl , this structure seems to possess
a characteristic length which is always of the same order as the experiment’s
resolution, no matter to which level of precision we tune the microscope.
If we try to see finer details in the picture, we decrease ℓ by increasing k.
This increases Λk , and the larger cosmological constant in turn leads to a shorter
√
proper length L(k) ∝ 1/ Λk . In the fixed point regime, the two opposing trends
“increasing resolving power for higher k” and “shrinking detail of a shrinking
universe due to larger Λk ∝ k 2 ” cancel precisely, and so the picture we see in the
microscope does not become more detailed when we turn its adjusting screw.
The scaling relation (11.10) suggests that the QEG spacetime when observed
below the Planck length displays a self-similar structure and more generally may
share certain properties with a fractal.
On sub-Planckian length scales the QEG spacetime has no intrinsic scale
because in the fractal regime, i.e., when the RG trajectory is still close to
the NGFP, the parameters which usually set the scales of the gravitational
interaction, Gk and λ̄k , are not yet “frozen out.” This happens only later on,
somewhere half way between the non-Gaussian and the Gaussian fixed point, at
a characteristic scale of the order of mPl .1
Below the Planck scale, Gk and Λk stop running and, as a result, the best-
suited metric ḡksc becomes independent of k so that L(ℓ) = const for ℓ ≫ ℓPl . In
this regime spacetime is well described by a single classical Riemannian manifold
and we are in the familiar realm of General Relativity.

1 Recall that the Planck mass and length are defined in terms of Newton’s constant in the
−1/2
classical regime, Gclass , i.e., mPl ≡ ℓ−1
Pl ≡ Gclass .
268 Towards Phenomenology

(3) Fundamental limitation on the distinguishability of spacetime

points [230]. It has been argued (in the example M = S 4 , but the conclusions
probably hold much more generally) that there exists a fundamental obstruction
which makes it impossible to physically distinguish spacetime points that are too
close [230].
The underlying mechanism is the same which led to the self-similarity above:
In the Planck regime, the (familiar) effect that a more energetic probe yields
a better resolution is precisely compensated for by the backreaction of the
spacetime.
Roughly speaking, the more energetic the probe is, the larger the relevant
value of the cosmological constant, and the smaller the overall scale of the uni-
verse. This renders the S 4 “fuzzy” in the sense that there is a minimal angular
separation below which two points cannot be resolved by any experiment.
But the interpretation of this effect is subtle, though [230, 334]. In particular,
care is required in associating a smallest possible proper length to the indis-
tinguishably close spacetime points. As it turns out, there does indeed exist a
non-zero lower bound on the proper distances computed with the fixed metric
of the classical regime; however, there is no such bound for proper distances
referring to the family of scale-dependent metrics.
Possible implications of this effect for black holes and cosmological spacetimes
have been investigated in [231]. See also [335] for a related discussion.

(4) Dimensional reduction and graviton propagator [173]. A further argu-

ment indicating that in QEG the spacetime may have properties reminiscent of a
fractal, namely a non-integer effective dimensionality is based on the anomalous
dimension ηN ≡ ∂t ln Gk .
The RG trajectories of the Einstein–Hilbert truncation (within its domain of
validity) have ηN ≈ 0 for k → 0 and ηN ≈ −2 for k → ∞, the smooth change by
two units occurring near k ≈ mPl . As a consequence, the effective dimensionality
defined as deff ≡ d + ηN equals 4 for ℓ ≫ ℓPl and 2 for ℓ ≪ ℓPl .
We know that the UV fixed point has ηN (g∗ , λ∗ ) = −2 and can use this
information in order to determine the momentum dependence of the dressed
graviton propagator for momenta p2 ≫ m2pl . Expanding Γk about flat space and
omitting the standard tensor structures [173] we find the inverse running propa-
gator Gek (p)−1 ∝ G−1 e
k p . The conventional dressed propagator G(p) contained in
2

e
Γ ≡ Γk=0 is obtained from Gk in the limit k → 0.
For p2 > k 2 ≫ m2Pl the relevant cutoff scale is the physical√
momentum p2 itself.
Hence, the k-evolution of Gek (p) stops at the threshold k = p2 . Therefore,

e −1 ∝ p2 G−1
G(p) √
η
∝ (p2 )1− 2 (11.11)
k
k= p2

because G−1
k ∝k
−η
when η ≡ ηN (g, λ) is approximately constant.
 11.3 Structure of “Quantum Spacetime” 269

In flat spacetimes of d dimensions, and for η ̸= 2 − d, the Fourier transform of

e
G(p) ∝ 1/(p2 )1−η/2 yields the following propagator in position space:

1
G(x; y) ∝ deff −2
with deff ≡ d + η. (11.12)
|x − y|

This form of the two-point function is well known from the theory of critical
phenomena, for instance. (In the latter case it applies to large distances.)
Equation (11.12) is not valid directly at the NGFP. For, say, d = 4 with η = −2
e = 1/p4 , and it has the following representation in
the dressed propagator is G(p)
position space:
1
G(x; y) = − 2 ln(µ|x − y|) . (11.13)
8π
Here µ is an arbitrary constant with the dimension of a mass. Obviously (11.13)
has the same structure as a 1/p2 -propagator in two dimensions.
Slightly away from the NGFP, before other physical scales intervene, the prop-
agator is of the form (11.12). This demonstrates that the quantity ηN has indeed
the interpretation of an anomalous contribution to the spacetime dimension:
fluctuation effects modify the decay properties of G so as to correspond to a
spacetime of effective dimensionality deff = 4 + ηN .
Thus the general properties of the RG trajectories imply a remarkable
dimensional reduction: Spacetime, probed by a “graviton” with p2 ≪ m2Pl is four-
dimensional, but it appears to be two-dimensional for a graviton with p2 ≫ m2Pl
[173].
More generally, when the classical spacetime is d-dimensional, the anomalous
∗
dimension at the fixed point is ηN = 2 − d. Therefore, for any d, the effec-
tive dimensionality as implied by the RG running of Newton’s constant is
∗
deff = d + ηN = 2 [163, 173].

(5) Spectral dimension [295, 322]. Standard fractals admit various, in gen-
eral inequivalent, notions which generalize the classical concept of a dimension
[336, 337]. One of them, the spectral dimension, is particularly well suited
for an application to quantum spacetimes. It has been evaluated both within
the EAA-based continuum approach to Asymptotic Safety [322] and by Monte
Carlo simulations of Causal Dynamical Triangulations [46, 47], and many other
approaches to quantum gravity [338].
To define the spectral dimension we consider first the diffusion process where
a spinless test particle performs a Brownian random walk on an ordinary Rie-
mannian manifold with a fixed classical metric gµν (x). This process is described
by the heat kernel Kg (x, x′ ; T ) discussed in Appendix D. The heat kernel gives
the probability density for a transition of the particle from x to x′ during the
fictitious time T . It satisfies the heat equation

∂T Kg (x, x′ ; T ) = −∆g Kg (x, x′ ; T ), (11.14)

270 Towards Phenomenology

where ∆g ≡ −D2 . In flat space, this equation is solved by

∫
dd p ip·(x−x′ ) −p2 T
Kg (x, x′ ; T ) = e e . (11.15)
(2π)d
In general, the heat kernel is a matrix element of the operator exp(−T ∆g ). In
the random walk picture its trace per unit volume,
∫ √
−1
Pg (T ) = V dd x g(x) Kg (x, x; T ) ≡ V −1 Tr exp(−T ∆g ), (11.16)
∫ √
with V ≡ dd x g(x), has the interpretation of an average return probability.
We know that Pg possesses an asymptotic early time expansion (for T → 0) of
∑∞
the form Pg (T ) = (4πT )−d/2 n=0 An T n , with An denoting the Seeley–DeWitt
coefficients. This expansion motivates the definition of a scale-dependent spectral
dimension Ds (T ) as the logarithmic derivative
d ln Pg (T )
Ds (T ) ≡ −2 . (11.17)
d ln T
The standard definition of the spectral dimension is obtained from this expression
in the limit ds ≡ limT →0 Ds (T ). On classical manifolds, where the early time
expansion of Pg (T ) is valid, ds agrees with the topological dimension d of the
manifold.
Coming back to quantum gravity where we integrate over all metrics on a given
manifold, the idea is to define a spectral dimension for quantum spacetimes by
replacing the return probability Pg (T ) in (11.17) with its expectation value:
∫
( )
P (T ) = ⟨Pgb(T )⟩ ≡ Db g DCDC̄ Pgb(T ) exp −Sbare [b
g , C, C̄] . (11.18)

Here gbµν denotes the microscopic metric, and Sbare is the bare action related to
the UV fixed point with the gauge-fixing and ghost terms included.
In [322] the expectation value (11.18) has been evaluated by mean field tech-
niques thereby making essential use of the family of metrics {ḡksc }. The result
was that along the RG trajectory the spectral dimension interpolates smoothly
between its macroscopic value Ds (T ) = d in the classical regime, and
1
Ds (T ) = d (NGFP regime) (11.19)
2
near the non-Gaussian fixed point. For the derivation of this result we refer to
[322].
It is remarkable that the spectral dimension shows the same dimensional reduc-
tion from four macroscopic to two microscopic dimensions which was found on
the basis of the logically independent effective dimension deff ≡ d + ηN . It is per-
haps even more remarkable that deff and ds agree in the NGFP regime if, and
only if, d = 4.
 11.4 Black Holes 271

A more detailed discussion of the fractal-like properties of the QEG spacetime

can be found in [295, 322]. See also [191] and [339] for analyses using higher
derivative and bi-metric truncations, respectively.
Furthermore, it is intriguing that the numerical simulations in the CDT
approach also find evidence for a similar dimensional reduction from four to
about two dimensions [46, 47]. See [295, 340] for a detailed comparison.

(6) The NGFP: entrance gate to a new phase of gravity? In the fixed
point regime, the relation (11.8) implies that proper lengths computed with
respect to ḡkscµν and g̊µν , respectively, differ by a constant factor. It is proportional
to 1/k and vanishes in the limit k → ∞ therefore: ḡkscµν ∝g̊µν /k 2 → 0.
With all due caution, one may speculate that the vanishing of all running
lengths measured with ḡkscµν relative to the corresponding classical lengths given
by g̊µν reflects that, at asymptotically high scales, a new phase of gravity is
approached in which the metric has a vanishing expectation value. This hints
at the possibility of a considerable symmetry enhancement at the fixed point as
presumably diffeomorphism invariance is unbroken in the new phase.

11.4 Black Holes

In this section we specialize to a particularly interesting class of effective space-
times, namely black holes. We follow the work in [157] and consider the case of
static, spherically symmetric spacetimes with a point mass M situated at the ori-
gin. The self-consistent metrics obtained in this case are given by a one-parameter
family of Schwarzschild metrics:
( )
ds2 = −f (r) dt2 + f (r)−1 dr2 + r2 dθ2 + sin2 θ dϕ2 , (11.20)

where the lapse function contains the scale-dependent Newton’s constant:

2 Gk M
f (r) = 1 − . (11.21)
r
The flow of Gk depicted in Figure 5.2 then implies that the one-parameter family
of self-consistent metrics interpolates between the Schwarzschild solution in the
classical regime and flat Minkowski space as k → ∞. This feature has drastic
consequences for the structure of the quantum spacetime close to the central
singularity at r = 0.
In order to understand these consequences, it is plausible to apply the cutoff
identification (11.4). Furthermore, following [157] we choose d(x, x′ ) as the proper
distance between the origin and the point x, given by the classical line element,
along a purely radial curve Cr . The resulting function d(r) is given by
√ √
r
d(r)|r<2G0 M = 2G0 M arctan − r(2G0 M − r) (11.22)
2G0 M − r
272 Towards Phenomenology
(√ √ )
r r
d(r)|r>2G0 M = πG0 M + 2G0 M ln + −1
2G0 M 2G0 M (11.23)
√
+ r(r − 2G0 M )
for points inside and outside the classical horizon, respectively. The two
expressions match at r = 2G0 M . For r ≪ 2G0 M , (11.22) possesses the expansion
2 1
d(r)|r≪2G0 M ≈ √ r3/2 + O(r5/2 ), (11.24)
3 2G0 M
while the asymptotic behavior for large r resulting from (11.23) is

d(r)|r≫2G0 M ≈ r + O(r0 ). (11.25)

For analytic calculations it is convenient to approximate the distance function by

√
2r3
d(r) = . (11.26)
2r + 9G0 M
This simple formula has the correct qualitative properties, and in particular,
interpolates smoothly between the two scaling regimes obtained as r → 0 and
r → ∞, respectively.
Substituting the interpolation function (11.26) into the cutoff identification
k(r) = ξ/d(r) gives √
2r + 9G0 M
k(r) = ξ . (11.27)
2r3
We now insert this cutoff identification into the solution (11.21). It would be
straightforward to carry out this improvement on the basis of the scale depen-
dence encoded in the beta functions (5.108) using numerical methods. In order
to get analytic access to the system, one can approximate the scale dependence
of Newton’s constant by the one-loop result (5.82). Combining it with (11.27)
then yields the improved radial function:

4G0 M r2
f (r) = 1 − . (11.28)
2r3 + ω̃G0 (2r + 9G0 M )

Here we abbreviated ω̃ ≡ ωξ 2 . This lapse function determines all properties of the

quantum-improved black hole such as its horizon structure, asymptotic behavior,
and thermodynamics.
Figure 11.1 shows the improved radial function for various values of M . The
analysis leads to the following main conclusions:

(1) Asymptotic behavior. For large r the improved lapse function approaches
the classical one,
( ) ( )
2G0 M G0 1
f (r) = 1 − 1 − ω̃ 2 + O . (11.29)
r r r4
 11.4 Black Holes 273

f(r)
1.0

0.8

0.6

0.4

0.2

0.0 r
5 10 15 20 25
0.2

0.4

118
Figure 11.1. The radial function (11.29) for G0 = 1, ω̃ = 15π , and differ-
ent values of the mass, M = 2.5, 3.5027, 5 (top, middle, bottom). The classical
behavior represented by the dashed lines is shown for comparison.

This can also be seen in Figure 11.1. Furthermore, (11.29) allows us to read off
the leading correction to Newton’s potential. For ω̃ = 118/15π this correction is
in agreement with the perturbative result obtained in [3]. This agreement can be
used to fix the hitherto undetermined parameter ξ.

(2) Regularity. The classical Schwarzschild solution possesses a curvature

singularity at r = 0 where the square of the Riemann tensor (Kretschmann
scalar) diverges. For the improved spacetime the fate of this singularity can be
investigated by expanding (11.28) for small r:

4r2
f (r) ≈ 1 − + O(r3 ). (11.30)
9ω̃G0
We see that the improved radial function is actually well-behaved at r = 0 and
has the form of the de Sitter metric with an effective cosmological constant
Λeff = 4/(3ω̃G0 ) > 0. Assuming that the RG improvement is reliable even for
r → 0, this feature would amount to a de Sitter core of the black hole [157]. The
regularity of the improved spacetime is confirmed by calculating the square of
−1/2
its Riemann tensor. With mPl ≡ G0 ,
3
Rµναβ Rµναβ = 2 (4/3) ω̃ 2 m4Pl + O(r). (11.31)

In contrast to the classical Schwarzschild black hole, Rµναβ Rµναβ is finite, point-
ing at the resolution of the classical black hole singularity.

(3) Horizon structure. The horizon structure encoded in the zeros of the
improved radial function (11.28) turns out to be richer than in the classical case.
As it can be seen from Figure 11.1 the number of horizons, NH , depends on
274 Towards Phenomenology

whether M is larger or smaller than a certain critical mass Mcr :


 0 for M < Mcr
NH = 1(double) for M = Mcr . (11.32)

2 for M > Mcr

For M > Mcr there exists both an outer and an inner horizon, situated at r+ and
r− , respectively. If the mass of the black hole equals the critical mass, the two
horizons coincide, while for M < Mcr there is no horizon at all. For the interpo-
118
lation (11.26) with ω̃ = 15π the critical mass is found to be Mcr ≈ 3.5 mPl .

(4) Causal structure. Apart from its regularity at r = 0, the causal structure
of the spacetime with the improved radial function is very similar to the one of a
classical Reissner–Nordström black hole [341]. For M > Mcr , one can distinguish
five main regions:

I and V : r+ < r < ∞, (11.33)

II and IV : r− < r < r+ ,
III and III′ : 0 < r < r− .

It is straightforward to calculate the geodesics of this spacetime. A typical

geodesic is depicted in the Penrose diagram of Figure 11.2. Starting in region
I (r < ∞) at rest, the test mass will fall into the black hole, pass the horizon r+ ,
and transit the region II with inverted time direction. When passing region III,
the test mass is repelled outwards, back to regions with larger r, namely IV and
V. This is in full analogy to the Reissner–Nordström case. However, one difference
is that the improved lapse function allows a particle with sufficient energy (posi-
tive kinetic energy at infinity) to actually reach the now non-singular point r = 0.

(5) Thermodynamics. Following the standard discussion of black hole thermo-

dynamics in Euclidean time and imposing regularity at the cone of the resulting
R2 × S 2 topology, the temperature of a horizon in the spacetime with line element
(11.20) is given by [342]
1
TBH = ∂r f (r) (11.34)
4π r=rH

For the improved lapse function (11.28) this relation yields the following
temperature of the outer black hole horizon at r+ ≡ r+ (M ):
( ( ))
1
2
r+ − ω̃ G0 1 + 9 ω̃ Gr0+M
TBH = M G0 ( )2 . (11.35)
2π 2 + ω̃ G (1 + 9G0 M )
r+ 0 2 r+

Since also the radius r+ is a function of M , we plot the resulting temperature

TBH (M ) in Figure 11.3. One observes that for masses M > Mcr the temperature
 11.4 Black Holes 275

J+ J+

J– J–
C
IV

r=0 III III’ r=0

b)
a) II H
J+ J+
E
I

J– J–

Figure 11.2. Penrose diagram for the improved Schwarzschild spacetime

[157]. Depicted are two geodesics of radially infalling particles, one of a particle
with positive kinetic energy at infinity, and another one of a particle starting
at rest at finite r. One observes that the latter does not reach the origin at
r = 0.

TBH (M)

0.015

0.010

0.005

0.000 M
2 4 6 8 10 12 14

Figure 11.3. Mass dependence of the Hawking temperature for the improved
118
Schwarzschild black hole in Planck units (solid line) for ω̃ = 15π . For compar-
ison the standard semiclassical temperature is also shown (dashed line).
276 Towards Phenomenology

follows the same 1/M dependence as in the semiclassical case [342]. However,
when M approaches Mcr , the temperature of the improved black hole drops to
zero. This might indicate the formation of a stable black hole remnant as the
endpoint of the evaporation process.
The original analysis of RG improved black holes [157] which we outlined here
has been extended in various ways; see [343] for a detailed review.
In particular, the dynamics of black hole evaporation has been studied on the
basis of the Vaidya metric in [344], while the RG improved Kerr metric has been
constructed and analyzed in [345, 346]. The structure of the Cauchy horizon
singularity of a black hole formed in a generic collapse was studied in [347]. In
[348] the results of [157] have been reanalyzed from a thermodynamics perspec-
tive, and the role of higher derivative terms in the Effective Average Action has
been explored in [349]. Furthermore, RG improvements including scale-dependent
surface terms have been carried out in [182] and an equation highlighting the
state-counting properties of Γk has been proposed. Moreover, the effect of a
scale-dependent cosmological constant has been studied in [350, 351]. Further
aspects of RG improved black hole geometries have been explored in [352–357].
Quite surprisingly, it has been found [358] that the improved Schwarzschild space-
time shares many properties of the “Planck stars” found in the context of Loop
Quantum Gravity.
In general, these works showed that quantum effects originating in the vicinity
of the nontrivial UV fixed point lead to drastic modifications of the classical
picture at short distances or high momenta, while the long-distance properties
of the improved black hole spacetime essentially agree with classical General
Relativity.

11.5 Cosmology
As a last example, let us discuss the possible implications of QEG in the
Friedmann–Lemaı̂tre cosmological models [323, 359, 360].
A natural starting point for investigating the imprints of Asymptotic Safety
in cosmology studies the self-consistent solutions of (11.2) for the homogeneous
and isotropic, spatially flat Robertson–Walker metrics:
[ ]
ds2 = −dt2 + a(t)2 dx2 + dy 2 + dz 2 . (11.36)

This ansatz is supplemented by a stress-energy tensor corresponding to a perfect

fluid,
Tµ ν = diag[−ρ, p, p, p], (11.37)
satisfying the equation of state p = wρ. The effect of scale-dependent couplings
is then captured by the dynamics of the scale factor a(t).
The complete cosmic histories resulting from this framework have been studied
in a series of works [323, 331, 359–361]. This section briefly summarizes the main
 11.5 Cosmology 277

results of these investigations. A more detailed review can be found in [362]; see
also [363, 364].
Substituting the ansatz for the metric, (11.36), and the stress-energy tensor
(11.37) into (11.2), results in the standard Friedmann equations with Newton’s
coupling and the cosmological constant depending on the scale k. Promoting
k to a function of the cosmic time, k → k(t), the dynamics is governed by the
improved Friedmann and continuity equations:

8π 1
H2 = G(t)ρ + Λ(t),
3 3
(11.38)
Λ̇ + 8πρ Ġ
ρ̇ + 3H(ρ + p) = − .
8πG(t)

The second equation arises by combining the Bianchi identity and Einstein’s
equation, Dµ [Λ(t) gµν − 8πG(t) Tµν ] = 0. The extra term on its right-hand side
has the interpretation of an energy transfer between the gravitational degrees
of freedom and matter. Introducing the critical density ρcrit ≡ 3H(t)2 /(8πG(t))
and defining the relative densities Ωmatter = ρ/ρcrit and ΩΛ = ρΛ /ρcrit the first
equation is equivalent to Ωmatter + ΩΛ = 1.

(1) Cosmology in the fixed point regime. We first focus on the very
early part of the cosmological evolution where the RG running of Gk and Λk
is controlled by the NGFP:

Gk = g∗ k −2 , Λk = λ∗ k 2 . (11.39)

Identifying the scale k with the Hubble parameter, as suggested by (11.5), the
system (11.38) has the analytic solution
α [ ]−1
H(t) = , a(t) = A tα , α = 32 (1 + w)(1 − Ω∗Λ ) , (11.40)
t
together with
ρ(t) = ρb t−4 , b t2 ,
G(t) = G b t−2 .
Λ(t) = Λ (11.41)
Here A is a positive parameter. The constants G,b Λ,b and ρb are fixed by the
coordinates of the NGFP and the free parameter ξ,
( )
3 ξ 2 α4 1 b = g∗ , Λ b = λ∗ ξ 2 α2 .
ρb = 1 − λ∗ ξ , G
2
(11.42)
8π g∗ 3 ξ 2 α2
The vacuum energy density in the fixed point regime,

Ω∗Λ = 13 λ∗ ξ 2 , (11.43)

takes values in the interval 0 < Ω∗Λ < 1. For radiation dominance, w = 1/3, and the
fixed point (5.95), one typically has 1/2 < α < 2, but α may be made arbitrarily
large by choosing the value of ξ such that Ω∗Λ ≃ 1.
278 Towards Phenomenology

The cosmological solutions (11.40) possess no particle horizon if α ≥ 1, while

for α < 1 there is a horizon of radius rH = t/(1 − α).
Moreover, the universe described by the fixed point solution is in a phase
of power law inflation if α > 1. Assuming radiation dominance, w = 1/3, this
requires Ω∗Λ > 1/2. For the NGFP coordinates (5.95) this corresponds to 2.13 <
ξ < 3.02.
Remarkably, the asymptotic behavior of the solution for t → 02 is actually
independent of the chosen improvement scheme: all choices entail k ∝ t−1 +
subleading, corroborating the robustness of the improvement procedure.

(2) Quantum gravity-driven inflation. Realizing an inflationary phase dur-

ing the fixed point regime by having Ω∗Λ ≥ 1/2 is a rather attractive scenario.
Being driven by quantum gravity effects, more precisely by the originally
large cosmological constant, inflation ends automatically at a certain tran-
sition time ttr when the RG flow enters into the classical regime and Λ(t)
is sufficiently small. For t > ttr the evolution is then given by a classical
Friedmann–Lemaı̂tre–Robertson–Walker universe.
Thus the NGFP-driven inflationary phase occurs in a completely natural way
and does not require any extra ingredients like an inflaton or even a specific
inflaton potential.
Quite remarkably, the NGFP-driven inflation may leave observable imprints
in the cosmic fluctuation spectrum.
The transition time ttr is determined by the scale k where the underlying
RG trajectory enters into the classical regime when k . mPl , H(ttr ) ≈ mPl . Since
ξ = O(1) the relation H(t) = α/t then leads to the estimate

ttr = α tPl . (11.44)

If Ω∗Λ is very close to one, i.e., if α ≫ 1, the cosmic time ttr when the Hubble
parameter is of order mPl can be much larger than the Planck time. The latter
is then located well within the NGFP regime.
We now consider the evolution of a fluctuation with comoving length ∆x. The
corresponding proper length is L(t) = a(t)∆x. During the NGFP regime, L(t) is
related to the proper length at the transition time ttr via L(t) = (t/ttr )α L(ttr ).
The ratio of L(t) and the Hubble radius ℓH (t) then evolves as
( )α−1
L(t) t L(ttr )
= . (11.45)
ℓH (t) ttr ℓH (ttr )

2 In [365, 366] an analogous fixed point cosmology due to a hypothetical IR fixed point of the
RG flow has been used to describe the acceleration of the universe at late times (t → ∞). It
turned out to be consistent with all observational data available [367].
See also [368] for a general analysis of cosmologies with a time-dependent cosmological
constant.
 11.5 Cosmology 279

For α > 1 the proper length of the object grows faster than the Hubble radius.
Fluctuations which are of sub-Hubble size at early times can therefore cross the
Hubble radius and become “super-Hubble”-size at later times.
For definiteness, let us consider a fluctuation which, at the transition time
ttr , is eN times larger than the Hubble radius. For this fluctuation, (11.45)
implies
( )α−1
L(t) N t
=e . (11.46)
ℓH (t) ttr
The time tN where this fluctuation crosses the Hubble radius, L(tN ) = ℓH (tN ),
is given by
( )
N
tN = ttr exp − . (11.47)
α−1
Thus, even for moderate values of α, NGFP-driven inflation easily magni-
fies fluctuations to a size many orders of magnitude larger than the Hubble
radius.
Interestingly, the cosmological structures we observe today may indeed have
crossed the Hubble radius during the NGFP regime. Considering the largest
structures visible today and using the classical evolution equations to backtrace
them in time to the point where H = mPl , we find that their size back then
was about e60 ℓPl . Setting N = 60 the time t60 when these structures crossed the
horizon can be estimated from (11.47). For α = 25, t60 = ttr /12.2 = 2.05tPl . Thus
t60 is one order of magnitude smaller than ttr . Hence, in this setting structures
observed today may have their origin in the quantum gravity regime controlled
by the NGFP.

(3) Scale-free fluctuation spectrum. The NGFP also offers a natural

mechanism for generating a nearly scale-free spectrum of primordial fluctua-
tions [359, 360]. This can be seen as follows: due to the anomalous dimension
at the NGFP, ηN = −2, the associated effective graviton propagator (at the
background level) has a characteristic 1/p4 dependence. This implies a loga-
rithmic dependence for the two-point graviton correlator in configuration space,
⟨hµν (x)hρσ (y)⟩ ∼ ln(x − y)2 . As a consequence, curvature fluctuations δR ∝ ∂ 2 h
(where R stands for any component of the Riemann tensor) must behave as,
symbolically, ⟨δR(x, t)δR(y, t)⟩ ∝ 1/|x − y|4 .
Now we make the natural assumption that the primordial matter density
fluctuations δρ originate from the fluctuations of the geometry itself.
The classical Einstein equations then provide the relation δρ ∝ δR. As a
consequence, the correlation function ξ(x) ≡ ⟨δ(x)δ(0)⟩ of the density contrast
δ(x) ≡ δρ(x)/ρ behaves as [359, 360]

1
ξ(x) ∝ , (11.48)
|x|4
280 Towards Phenomenology

provided the physical distance a(t)|x| is smaller than the Planck length. From
the three-dimensional Fourier transform of (11.48) we immediately get |δk |2 ∝
|k|, which amounts to a scale-invariant power spectrum, with spectral index
n = 1 [369].
Clearly, small deviations from n = 1 are to be expected as the prediction of an
exactly scale-free spectrum holds only directly at the NGFP. Since the NGFP
is supposed to govern the dynamics of the very early universe this entails, in
particular, that the power spectrum can acquire non-trivial corrections during
the later cosmological epochs.

(4) Including anisotropy. In [370] the framework of isotropic Friedmann–

Lemaı̂tre–Robertson–Walker cosmologies has been generalized to anisotropic
Bianchi IX spacetimes. It was found that the chaotic approach to the initial
singularity is removed by the scale-dependent couplings: the series of classical,
chaotic Kasner oscillations ceases, and the solutions uniformly approach a point
singularity as t → 0. The transition between the chaotic Kasner oscillations and
the non-chaotic approach to the singularity is triggered by quantum effects dom-
inating over the classical dynamics when the value of the running cosmological
constant becomes large.

(5) Toward realistic cosmologies. An interesting alternative to the analysis

of the self-consistent solutions arising from (11.2) carries out a cutoff identifica-
tion of Type III, (11.6), at the level of the action rather than a specific solution.
For k 2 (x) = ξ 2 R with R being the Ricci scalar, ∫this procedure results in quan-
√
tum gravity inspired actions of the form S[g] = d4 x −g f (R). In [333] it was
found that the resulting modified gravity actions give rise to fluctuation spectra
compatible with the data published by the Planck collaboration in 2015 [371].
 12
Miscellanea

In the previous chapters of this monograph we introduced the basic ideas and
methods underlying the “Asymptotic Safety program,” which aims at establish-
ing a concrete realization of Weinberg’s scenario on the basis of the functional
renormalization group.
Being a rapidly growing direction of research, it is clear that we could not
cover all developments here. This concerns in particular the large and steadily
increasing number of truncated RG flows that have been computed and analyzed
explicitly. However, regarding the viability of the Asymptotic Safety program,
the developments confirm the picture we have drawn here without any exception.
In this final chapter we list and briefly summarize a number of further, dis-
connected developments and investigations. A detailed, pedagogical account is
likewise well beyond the scope of this book, and the reader is invited to consult
the literature for further details.

(1) Infinite-dimensional truncations of f (R)-type. In Section 5.3.5 we

discussed finite-dimensional truncation subspaces where the gravitational part
of the Effective Average Action has the form
∫
√
Γ̄k [g] = dd x g fk (R), (12.1)

with fk (R) being a polynomial in the Ricci scalar. The results there rest on
the fact that one can derive a partial differential equation (PDE) that gov-
erns the scale dependence of the function fk (R) without making any assumption
on its functional form [185, 186]. The polynomial approximation of fk (R) can
be thought of as an expansion approximating the flow in a regime where the
dimensionless background curvature r ≡ R̄/k 2 is small. In order to establish the
consistency of the NGFP seen in these expansions, it is interesting to understand
under which conditions these polynomial expansions can be extended to a global
282 Miscellanea

solution of the PDE, which is well-defined for all values of r. Since the result-
ing functions fk (R) and in particular the fixed function f∗ (R) carry information
about an infinite number of coupling constants, these computations are said to
employ an infinite-dimensional truncation.
Starting from [372], the properties of this class of truncated flow equations
has been studied extensively in both d = 3 [373–375] and in d = 4 dimensions
[376–383].
In general, the PDE governing the flow of fk (R) is not unique. It depends para-
metrically on the choices for the gauge-fixing term and coarse-graining operator.
All constructions studied to date evaluate the flow on a maximally symmetric
background satisfying (5.29). Furthermore, they employ a transverse-traceless
decomposition [384, 385] of the fluctuation field:

2 1
µν + D̄µ ξν + D̄ν ξµ + 2D̄µ D̄ν σ − ḡµν D̄ σ +
hµν = hT 2
ḡµν ϕ. (12.2)
d d
The term transverse-traceless decomposition indicates that the component fields
are transverse with respect to the covariant derivative constructed from the
background metric and satisfy the differential constraints

D̄µ hT
µν = 0, ḡ µν hT
µν = 0, D̄µ ξ µ = 0, ḡ µν hµν = ϕ . (12.3)

In practice the transverse-traceless decomposition has the highly welcome con-

sequence that all terms containing a contraction of a covariant derivative with
a fluctuation field vanish. In combination with a maximally symmetric back-
ground this ensures that all differential operators combine into Laplacians
D̄2 ≡ ḡ µν D̄µ D̄ν . The operators under the traces on the right-hand side of the
FRGE are then functions of D̄2 and R̄. They can be evaluated using heat kernel
techniques or as spectral sums.
In order to describe the new ideas and structures appearing in the context
of the f (R)-type truncations, we focus here on the specific setting introduced
in [376, 377]. This construction uses the freedom in choosing a gauge-fixing and
coarse-graining scheme in such a way that the contributions of the gauge degrees
of freedom, the ghost sector, and the Jacobians coming from the transverse-
traceless decomposition precisely cancel. In other words, this setting is minimal in
the sense that ∂t Γk receives non-zero contributions from the transverse-traceless
modes hT µν and the trace modes ϕ only. The two remaining traces are evaluated
as a sum over eigenvalues weighted by the corresponding degeneracies. Due to the
regulator Rk they receive contributions from a finite number of terms only and
thus converge. Applying a smoothing procedure for the step functions results in a
two-parameter family of smooth partial differential equations. It is conveniently
expressed in terms of the dimensionless function φk (r), defined through
( )
fk (R̄) ≡ k 4 φk R̄k −2 , r ≡ R̄ k −2 . (12.4)
 Miscellanea 283

Explicitly, the PDE for φk (r) is given by [377]:

c̃1 φ′k + c̃2 (φ̇′k − 2rφ′′k )

φ̇k + 4φk − 2rφ′k =
3φk − ((3α2 + 1) r − 3)φ′k
c1 φ′k + c2 φ′′k + c3 φ̇′k + c4 (φ̇′′k − 2rφ′′′
k )
+ .
2φk + (3 − (3α0 + 2) r)φ′k + (3α0 r + r − 3)2 φ′′k
(12.5)

Here dots and primes denote derivatives with respect to the renormalization
group time t and the dimensionless curvature r, respectively. The coefficients ci
and c̃i are certain polynomials in r. They read

( )
((6α0 − 1) r − 6) α0 (6α0 − 1) r 2 + (10 − 48α0 )r + 42
c1 = − ,
2304π 2
( ( ) 3 ( ) )
((6α0 − 1) r − 6) α0 54α0 − 3α0 − 1 r + 270α0 + 42α0 − 35 r 2 − 39(18α0 + 1)r + 378
2 2
c2 = − ,
4608π 2
(α0 r − 1)((6α0 − 1)r − 6) 2
c3 = − ,
2304π 2
(α0 r − 1)((6α0 − 1)r − 6)2 ((9α0 + 5)r − 9)
c4 = ,
4608π 2
(12.6)
and
(( ) )
5(6α2 r + r − 6) 18α22 + 9α2 − 2 r2 − 18(8α2 + 1)r + 126
c̃1 = − ,
6912π 2 (12.7)
5(6α2 r + r − 6)((3α2 + 2)r − 3)((6α2 − 1)r − 6)
c̃2 = − .
6912π 2
Furthermore, the parameters α0 and α2 in (12.5) capture the freedom of including
an endomorphism term proportional to the Ricci scalar in the coarse-graining
operator.
The first and second term on the right-hand side of (12.5) are the contributions
of the transverse-traceless and the conformal fluctuations, respectively. Hence,
the order of the PDE is determined by the conformal sector. The conformally
reduced version of the full equation is obtained by dropping the first term. Thus
the conformally reduced and the full equation have the same structure, suggesting
that the conformally reduced setting provides a good approximation to the full
equation.
By definition, fixed functions φ∗ (r) are the k-stationary solutions of the evo-
lution equation (12.5). In general, a proper fixed function should satisfy the
following criteria:

1. φ∗ (r) should be globally well-defined: By the general discussion of Chapter

3, a fixed point corresponds to a special type of complete RG trajectory
where Γk = Γ∗ for all k ∈ [0, ∞). When working on a maximally symmetric
284 Miscellanea

background with fixed, positive scalar curvature, R̄ > 0, the transition to

the dimensionless curvature r = R̄k −2 implies that the fixed function φ∗ (r)
should be regular on the entire half-line r ∈ [0, ∞].
2. φ∗ (r) should be an isolated solution: A fixed point of the renormalization
group corresponds to a single point in theory space. At the level of functional
truncations, this feature is reflected by the condition that φ∗ (r) should not
be part of a multi-dimensional family of solutions.
3. φ∗ (r) should admit a discrete spectrum of globally well-defined deforma-
tions: Once a suitable fixed function satisfying the conditions 1. and 2. has
been constructed, one should verify that the solution gives rise to a dis-
crete set of globally well-defined deformations χn (r). This can be studied
by substituting the ansatz

φk (t, r) = φ∗ (r) + ϵ e−θn t χn (r) (12.8)

into the PDE (12.5) and looking for globally well-defined solutions χn (r)
of the linearized equation. The condition that χn (r) is globally well-defined
is expected to put conditions on the admissible values of θn such that the
stability coefficients are bounded from above and discrete. Since φ∗ (r) is
typically known only numerically, carrying out this calculation in practice is
computationally quite challenging. The best indications that the θn indeed
form a discrete spectrum is then provided by the polynomial expansion of
fk (R) as displayed in Table 5.3.
We now illustrate the implementation of these conditions for the PDE (12.5).
Imposing that φk (r) is independent of k leads to an ordinary differential equation
(ODE) for the fixed function φ∗ (r):

c̃1 φ′ − 2c̃2 rφ′′

4φ − 2rφ′ =
3φ − (3α2 r + r − 3)φ′
(12.9)
c1 φ′ + c2 φ′′ − 2c4 rφ′′′
+ .
(3α0 r + r − 3)2 φ′′ + (3 − (3α0 + 2)r)φ′ + 2φ

The coefficients ci and c̃i are again given by (12.7) and (12.6). Since there is no
risk of confusion we omit the asterisk and denote the fixed function by φ(r) from
now on.
Equation (12.9) is a non-linear third-order equation for φ(r). Thus, locally,
there is a three-parameter family of solutions to this equation. The condition
that a fixed function should be globally well-defined may reduce the number
of these free parameters. In particular, one may expect isolated fixed functions
if there is a balance between the order of the ODE and its singularity structure
[383]:
Degree of ODE − number of singularities = 0. (12.10)
Thus any singularity reduces the number of free parameters by one.
 Miscellanea 285

For fixed singularities this feature can be understood as follows: Let rising
denote the position of a fixed singularity of order one. Casting the ODE (12.9)
into normal form by solving for φ′′′ (r), the right-hand side of the resulting
equation admits a Laurent expansion,
ei (φ′′ (ri ), φ′ (ri ), φ(ri ), ri )
φ′′′ (r) = + regular terms, (12.11)
r − rising

where the explicit form of the ei can be derived from (12.9). Demanding that
φ′′′ (r) does not diverge at rising requires that the functions ei must vanish at r =
rising . This puts a non-trivial boundary condition on admissible solutions which
reduces the number of free parameters.
In principle the LHS of (12.10) also takes into account movable singularities,
which arise from a φ-dependent denominator developing a zero dynamically.
Their occurrence requires studying explicit solutions. However, so far, only fixed
singularities played a role in restricting the fixed function φ(r).
The set of fixed singularities for (12.9) is found by casting the equation into
normal form by solving for φ′′′ (r). The resulting expression indicates that, in
general, the third derivative will be infinite if the prefactor rc4 multiplying φ′′′ (r)
vanishes. From (12.6) one finds that this prefactor is a fifth order polynomial with
roots located at
9 1 6
r1sing = 0, r2sing = , r3sing = , sing
r4,5 = . (12.12)
5 + 9α0 α0 6α0 − 1

Obviously the position of the fixed singularities rising , i = 2, 3, 4, 5, depends on the

“endomorphism” parameter α0 which characterizes Rk . The fixed singularities
reveal that there are two distinguished values for α0 : r2sing is infinite if α0 = − 95 ,
sing
while the double root r4,5 is shifted to infinity for α0 = 16 . The latter choice
reduces the number of fixed singularities from five to three so that the balance
equation (12.10) is satisfied and one expects that the system admits isolated
global solutions.
The value of α2 is not fixed by the global structure of the ODE. A conceivable,
but far less cogent, selection principle for this value is provided by the condition
of equal lowest eigenvalues, i.e., one implements that the smallest eigenvalues
of the coarse-graining operators in the transverse-traceless and conformal sector
agree. These conditions fix the free parameters to

α0 = 16 , α2 = − 12 . (12.13)

The set of fixed singularities for this choice is then

r1sing = 0, r2sing = 18
13 , r3sing = 6 . (12.14)

At this stage the task of finding admissible fixed functions has been reduced to
searching the solution space of the ODE, verifying that there are indeed globally
286 Miscellanea

j(r)
5

j (r)
4 0.30
0.25
0.20
3 0.15
0.10
0.05
2 r
1 2 3 4
–0.05

r
2 4 6 8 10

–1

Figure 12.1. The isolated, globally well-defined fixed function φ∗ (r) con-
structed from imposing regularity at the three fixed singularities (12.14). The
fixed function is bounded from below, comes with a positive cosmological
constant λ∗ and Newton’s constant g∗ , and displays a global minimum at
rmin ≈ 2.5. Together with φ∗ (0) > 0 this minimum ensures that the solution
passes the redundancy test [382]. (From [377].)

well-defined solutions satisfying the non-local boundary conditions imposed by

the fixed singularities.
Practically, this is done by applying a shooting method. The initial conditions
for solving the ODE are set up at r = 0. The algorithm then uses analytic expan-
sions around rising to move away from, or extrapolate a solution through, a fixed
singularity. These expansions are connected by numerical solutions obtained on
the intervals [rising + ϵ, ri+1
sing
− ϵ], with ϵ = 10−3 , say, where the ODE is regular.
This strategy indeed identifies an isolated, globally well-defined fixed function
φ∗ (r). This solution is shown in Figure 12.1.
The expansion of the solution for r ≪ 1 allows us to read off the values of the
dimensionless cosmological constant and Newton’s constant associated with the
fixed function and the universal product τ∗ = g∗ λ∗ :
λ∗ = 0.411, g∗ = 0.547, τ∗ = 0.224. (12.15)
Thus, φ∗ (r) gives rise to a positive fixed point value for λ∗ and g∗ in agreement
with the results found in the Einstein–Hilbert truncation. The value of τ∗ is
in reasonable agreement with the results of the R2 -truncation (see Figure 5.5).
Moreover, a stability analysis of the ODE (12.9) using a polynomial approxima-
tion of the fixed function shows that the stability coefficients obtained in this
way are quite similar to the ones reported in Table 5.2. This provides a strong
 Miscellanea 287

indication that the NGFP seen in finite-dimensional truncations indeed possesses

an extension to a fixed function which includes an infinite number of coupling
constants.

(2) The perturbative two-loop counterterm is asymptotically safe

[386]. The perturbative quantization of the Einstein–Hilbert action gives rise
to a divergence at two-loop order [2, 387, 388],
∫
1 209 1 √
ΓGS = 2 2
d4 x g Cµν κλ Cκλ ρσ Cρσ µν , (12.16)
ϵ 2880 (16π )
with Cµνρσ being the Weyl-tensor. This divergence is not of the form of the bare
action and can thus not be absorbed into the bare coupling constants. It requires
adding a new counterterm with an undetermined coefficient. In combination with
power-counting arguments, this is taken as a signal that the perturbative quan-
tization of the Einstein–Hilbert action requires fixing an infinite number of free
parameters from experimental data, i.e., the Einstein–Hilbert action leads to a
perturbatively non-renormalizable quantum field theory.
The occurrence of the two-loop divergence (12.16) triggered the investigation
of a variety of alternative routes toward quantizing gravity, e.g., by modifying
the quantization scheme or introducing new degrees of freedom and symme-
tries. Eventually, any theory of quantum gravity has to explain the fate of this
divergence though.
For the Asymptotic Safety program, this question has been clarified in [386].
For this purpose the Einstein–Hilbert truncation, discussed in Section 5.2, is
supplemented by the interaction term
∫
√
Γk = σ̄k d4 x g Cµν κλ Cκλ ρσ Cρσ µν .
GS
(12.17)

The beta functions k∂k gi ≡ βgi (λ, g, σ) for the three dimensionless couplings
gi ≡ {λ, g, σ} can be constructed using the heat kernel techniques introduced in
Appendix D. In particular, the computation of βσ can be simplified by choosing
a Ricci-flat background, R̄µν = 0.
The resulting system of beta functions exhibits two rather remarkable prop-
erties. Firstly, the new coupling σk does not enter the beta functions of the
Einstein–Hilbert sector. The beta functions for g and λ are thus given by (5.63).
For the optimized shape function (E.13) they read

βg = (2 + ηN ) g,
( )
g 5 5 1 (12.18)
βλ = (ηN − 2) λ + − 4 − ηN .
2π 1 − 2λ 6 1 − 2λ
The anomalous dimension of Newton’s constant is of the usual form,
g B1
ηN = , (12.19)
1 − g B2
288 Miscellanea

with
[ ] [ ]
1 5 9 1 5 1 3
B1 = − − 5 , B 2 = − − .
3π 1 − 2λ (1 − 2λ)2 6π 2 (1 − 2λ) (1 − 2λ)2
(12.20)

Secondly, βσ is cubic in σ,

βσ = c0 + (2 + c1 ) σ + c2 σ 2 + c3 σ 3 , (12.21)

with the following coefficients [386]:

[ ]
1 2 − ηN 6 − ηN 5ηN
c0 = + − ,
64π 2 (1 − 2λ) 2(1 − 2λ) (1 − 2λ)3 378
[ ]
3g 23(8 − ηN ) 7(10 − ηN )
c1 = 5(6 − η N ) + − ,
16π(1 − 2λ)2 8(1 − 2λ) 10(1 − 2λ)2
[ ] (12.22)
g2 233(12 − ηN ) 9(14 − ηN )
c2 = − ,
2(1 − 2λ)3 10 7(1 − 2λ)
6πg 3 (18 − ηN )
c3 = .
(1 − 2λ)4
As the coefficient c3 (g, λ) is non-zero, the structure of (12.18) and (12.21)
then guarantees that there exists always at least one NGFP which generalizes
the NGFP of the Einstein–Hilbert truncation. For the optimized shape function,
this fixed point is located at

NGFPGS : λ∗ = 0.193, g∗ = 0.707, σ∗ = −0.305, (12.23)

and its stability coefficients are given by

θ1,2 = 1.475 ± 3.043 i, θ3 = −79.39. (12.24)

Thus, the inclusion of the two-loop counterterm in the truncation does not
vanquish the NGFP responsible for Asymptotic Safety. Moreover, the negative
stability coefficient indicates that there is no additional free parameter related to
the enlarged truncation space. This explicit computation strongly supports the
expectation that perturbative counterterms are actually harmless for the Asymp-
totic Safety mechanism.

(3) Paramagnetic dominance. One may ask if there is a simple physical

property of the quantum dynamically interacting non-linear gravitons which
is responsible for their inclination to form a NGFP. In [183] such a mecha-
nism, referred to as paramagnetic dominance, has been identified, and intuitive
analogies with better-understood physical phenomena were drawn (magnetism
in solids, asymptotic freedom of gluons).
Quite generally, Γk [h; g] gives rise to an inverse propagator for the hµν -
fluctuations which consists of a kinetic term involving covariant derivatives
 Miscellanea 289

as well as potential terms built from the background Riemann tensor R̄µναβ
and its contractions. For the Einstein–Hilbert truncation, say, we saw in (5.24)
that
∫ [ the relevant
] terms read, symbolically and with suppressed indices, Γquad
k ∼
h D̄ + R̄ h. By analogy with elementary magnetic systems, hD̄2 h and hR̄h,
2

respectively, describe dia- and paramagnetic-type interactions of the metric

fluctuations hµν with a given background field ḡµν .
It was shown that in d > 3 the gravitational antiscreening and the formation of
the NGFP in QEG is entirely due to a strong dominance of the ultralocal para-
magnetic interactions over the diamagnetic ones that favor screening. (In d < 3
both the dia- and paramagnetic effects support antiscreening.) In [183, 389] the
spacetimes of QEG are interpreted as a polarizable medium with a “paramag-
netic” response to external perturbations. In this respect they show similarities
with the vacuum state of Yang–Mills theory.1

(4) Composite operators. In applications of the EAA we may encounter

coevolving quantities, Qk , which satisfy their own flow equation, with a beta
functional depending on Γk though:

′
∂t Qk = Bk {Qk , Γk } and ∂t Γk = B k {Γk }. (12.25)

A typical example are composite operators that are not already included in Γk .
In [243] a general EAA-based method was introduced to define and renormalize
such operators.2 In [229] it has been employed to compute the anomalous scaling
dimensions of geometrically motivated observables (length and volume operators)
appearing in quantum gravity.
The basic idea of this method is to repeat the derivation of the FRGE with the
modified bare action S + ε · O which couples the composite operators of interest,
Oi (x), to sources εi (x).3 One may then take any number of functional derivatives
of the FRGE with respect to the sources before setting ε = 0 at the end. For a
single operator insertion, for example, the flow equation for [Ok ]i ≡ δΓk /δεi then
reads
[( )−1 ( )( )−1 ]
( ) 1 (2) (2)
∂t ε · [Ok ] = − STr Γk + Rk ε · [Ok ](2) Γk + Rk ∂t R k .
2
(12.26)
To solve (12.26) one expands [Ok ]i in a k-independent basis of operators,
∑
[Ok ]i = j Zij (k)Oj , so that (12.26) becomes a differential equation for the

1 Along a different line of research further similarities (vacuum condensates) were found in
[222, 390].
2 For earlier work in a non-gravitational context also see [34, 391].
3 The dot in expressions like ε · O symbolizes an integration over spacetime and a summation
over any type of indices carried by ε and O.
290 Miscellanea
( )
mixing matrix Zij (k). Letting γZ,ij ≡ Z −1 ∂t Z ij it reads

∑ [( )−1 ( )( )−1 ]
1 (2) (2) (2)
γZ,ij Oj (x) = − STr Γk + Rk Oi (x) Γk + Rk ∂t R k .
j
2
(12.27)
Projecting the right-hand side of (12.27) on the basis {Oj } results in explicit
expressions for γZ,ij in terms of the couplings which parametrize Γk .
It can be shown [243] that the quantum-corrected scaling dimensions of the
pertinent scaling fields are given by the eigenvalues of the matrix di δij + γZ,ij ,
with di denoting the classical mass dimension of Oi .
For a detailed discussion we refer to [243] where also the relation of those
eigenvalues to the critical exponents θi is explained, and to [229] for applications
in gravity.

(5) Gibbons–Hawking boundary term. On spacetime manifolds with a

non-empty boundary, the action functionals for gravity and matter typically
include certain boundary terms in order to admit a consistent classical varia-
tional principle [137]. The Einstein–Hilbert action, for example, is augmented by
the Gibbons–Hawking surface term [392] with a prefactor which is fixed by that
of the “bulk” term and therefore involves Newton’s constant.4 This surface term
plays a crucial role in the Euclidean approach to black hole thermodynamics, for
instance [342].
In [182], Quantum Einstein Gravity and its Asymptotic Safety were explored
on spacetimes with a non-trivial boundary using the Einstein–Hilbert ansatz
generalized by a Gibbons–Hawking term. The running prefactor of the latter
determines a kind of boundary Newton constant which is allowed to differ from
the familiar one in the bulk. Indeed, the two versions of Newton’s constant were
found to display opposite RG running. Invoking a scale-dependent variant of the
ADM mass, it was argued that this is precisely the signature of gravitational anti-
screening, and that the problem of the “bulk-boundary matching” [395] presents
itself differently in the background field setting. In an application to black holes,
indications were found that their specific heat capacity might turn positive at
Planckian scales [182, 396].

(6) Counting of modes and a C-function. The gravitational EAA has nat-
ural mode counting and monotonicity properties. In [154] they were exploited
to construct with the help of Γk a function C , which has properties similar
to Zamolodchikov’s C-function [397, 398] but without being restricted to two
dimensions though. This function C becomes stationary both at critical points
and in classical regimes, and it decreases monotonically along RG trajectories

4 For different boundary corrections see also [393, 394].

 Miscellanea 291

if split-symmetry is broken sufficiently weakly. The C -function gives a precise

meaning to the “number” of field modes, N , that are integrated out between two
scales k1 and k2 , respectively: Nk1 ,k2 ≡ Ck2 − Ck1 .
For the particularly interesting case of generalized crossover trajectories con-
necting the NGFP in the UV to a classical regime with positive cosmological
constant in the IR, the C -function depends on a gravitational instanton which
constitutes the background spacetime. In the case of four-dimensional de Sitter
dS
space, its Bekenstein–Hawking entropy SBH was found to play a role analogous
to the central charge in Zamolodchikov’s theorem. As a direct implication of
Asymptotic Safety the total number of modes turns out to be finite:

N = C UV − C IR ≈ 3π/Gcr Λcr = SBH

dS
. (12.28)

This motivates interpreting the entropy of de Sitter space as the total number
N of metric and ghost fluctuation modes integrated out along an asymptotically
safe RG trajectory ending in a classical regime with parameters Gcr and Λcr ,
respectively.
In [154] the above result has also been compared to the so-called N -bound
hypothesis 5 [399, 400] and a conjectured Λ-N -connection 6 [399–401] which had
emerged from entirely different considerations, but pointing in a similar direction
though.
For details the reader is referred to [154, 402].

(7) Finiteness of entanglement entropy. Asymptotic Safety can give finite

values to certain quantities which “count” the number of active degrees of free-
dom or field modes and would be badly divergent in standard field theory. One
example is N above, another are entanglement entropies of quantum fields on
the dynamical spacetimes of QEG [403].
For matter fields living on a classical spacetime the entanglement entropy
across a given surface of area A is typically of the form Sent ∝ A/ε2 . Here ε
is a short distance cutoff, and Sent is quadratically divergent [404, 405]. Loosely
speaking this divergence is caused by the “too many” degrees of freedom standard
field theories have in the ultraviolet. In QEG instead it can be shown [403] that
Asymptotic Safety renders the corresponding entropy finite.
This is yet another way of seeing that the spacetimes of asymptotically safe
gravity theories host far less degrees of freedom at short distances than the
familiar matter theories on Minkowski space.

5 In its stronger form [399], the claim is that in any universe with a positive cosmological
constant, containing arbitrary matter, the observable entropy Sobs is bounded by Sobs ≤
3π/GΛ ≡ N .
6 Arguments from string theory [399, 400] and Loop Quantum Gravity [401] motivated the
conjecture that all universes with a positive cosmological constant are described by a micro-
scopic quantum theory which has only a finite number of degrees of freedom, and that this
number is determined by Λ.
 Appendix A
Notation and Conventions

In this book, we follow the conventions used in [14] and the bulk of the Asymp-
totic Safety literature. The dimension of the spacetime manifold M is denoted
by d. Our spacetime metric gµν uses mostly plus conventions

Lorentzian metric : sign(gµν ) = (−, +, +, +, . . .) , (A.1)

Euclidean metric : sign(gµν ) = (+, +, +, +, . . .) . (A.2)

As usual, the inverse metric is indicated by upper indices g µα gαν = δνµ . Sym-
metrization and anti-symmetrization are indicated by round and square brackets,
respectively, and we normalize with unit strength, for example,

Hαβ = H(αβ) + H[αβ] . (A.3)

Covariant quantities constructed from gµν use their usual symbol while covariant
objects constructed from the background metric ḡµν are indicated with a bar.
The covariant derivative Dµ is defined by the Christoffel symbols, i.e.,

Dµ H α β = ∂µ H α β + Γµσ α H σ β − Γµβ σ H α σ (A.4)

which are given by

Γµν λ = 12 g λσ (∂µ gνσ + ∂ν gµσ − ∂σ gµν ) . (A.5)

The Riemann tensor Rµνρ λ , Ricci tensor Rµν , and the Ricci scalar R, respectively,
are defined as

Rµνρ λ ≡ ∂ν Γµρ λ − ∂µ Γνρ λ + Γµρ σ Γνσ λ − Γνρ σ Γµσ λ , (A.6)

Rµν ≡ Rµλν , λ
(A.7)
R ≡ g µν
Rµν . (A.8)

The curvature tensor satisfies the first and second Bianchi identity

Rµ[ναβ] = 0, Rµν[αβ;ρ] = 0. (A.9)

294 Notation and Conventions

The second Bianchi identity implies

Rµναβ ;β = Rµα;ν − Rνα;µ , Rµν ;ν = 12 R;µ . (A.10)

The commutator of two covariant derivatives satisfies

[Dµ , Dν ] Hλ = Rµνλ ρ Hρ ,
(A.11)
[Dµ , Dν ] Hλσ = Rµνλ ρ Hρσ + Rµνσ ρ Hλρ .

The Weyl tensor Cµνρσ is related to the Riemann tensor by

( )
Cµνρσ = Rµνρσ − d−2
2
gµ[ρ Rσ]ν − gν[ρ Rσ]µ + (d−1)(d−2)
2
R gµ[ρ gσ]ν . (A.12)

Occasionally we write
∆ ≡ −g µν Dµ Dν ≡ −D2 (A.13)
for the (negative) tensor Laplacian.
On a d-dimensional Euclidean spacetime (Rd , δµν ) it is often convenient to
expand fields in plane waves. Our convention regarding Fourier transforms is
∫
ϕp = dd xϕ(x)e−ipx (A.14)

implying the inverse transformation

∫
dd p
ϕ(x) = ϕp eipx . (A.15)
(2π)d

Throughout the work quantum fields like ϕb appearing as an integration variable

in a partition function will be denoted with a caret. The corresponding classical
fields given by their expectation values use the same symbol without the caret,
b Background fields like the background metric ḡµν are distinguished by
ϕ = ⟨ϕ⟩.
a bar and tensors carrying a bar like R̄µν are understood to be constructed from
the corresponding background quantities.
 Appendix B
Organizing the Derivative Expansion

A natural ordering principle for the interaction terms entering into the Effective
Average Action is provided by their number of derivatives. Due to the Bianchi
identities satisfied by the curvature tensors, constructing a basis of invariants at
each order in the derivative expansion is nontrivial. Such a basis, constructed
from the Riemann tensor, its contractions and covariant derivatives, may be
obtained based on the representation theory of the symmetric, general linear,
and orthogonal groups [161].

B.1 An Explicit Basis

On a manifold without boundary (or equivalently neglecting total derivatives)
this basis contains one element with, respectively, zero and two covariant deriva-
tives, three elements with four derivatives, and ten monomials built from six
derivatives.
For practical purposes it is convenient to trade the Riemann tensor for the
Weyl tensor via (A.12). An explicit basis containing all curvature monomials
with up to six derivatives is then given by the following set:

R0 = 1 ,
R1 = R ,
R21 = R2 , R22 = Cµναβ C µναβ
R23 = E
R31 = R∆R , R32 = Rµν ∆Rµν , (B.1)
R33 = R3 , R34 = RRµν Rµν ,
R35 = Rµ ν Rν α Rα µ , R36 = Rµν Rαβ C µανβ ,
R37 = RCαβµν C αβµν , R38 = Rµν C µαβγ C ν αβγ ,
R39 = Cµν ρσ
Cρσ αβ
Cαβ µν
, R310 = C α µ β ν C µ ρ ν σ C ρ α σ β .
296 Organizing the Derivative Expansion

In this basis the integrand of the Gauss-Bonnet term, constituting a topological

invariant if d = 4, is denoted by

E = R2 − 4Rµν Rµν + Rµναβ Rµναβ . (B.2)

In the context of the FRGE, other Riemann monomials involving four

or six derivatives such as Rµνρσ Rρνµσ , Rµν Dσ Dν Rµ σ , Rαρβσ Rαβµν Rρσ µν ,
Rαβ ρσ Rαµβν Rρµσν , and Rαρβσ Rαµβν Rρ ν σ µ can be encountered. These are not
part of the basis (B.1), but can be expressed in terms of the basis elements by
exploiting the following geometrical identities:

Rµνρσ Rρνµσ = 12 Rµνρσ Rµνρσ , (B.3)

Rµν Dσ Dν Rµ σ = 12 Rµν Dµ Dν R + Rµ ν Rν ρ Rρ µ − Rµν Rρσ Rµρνσ , (B.4)
1
Rαρβσ Rαβµν Rρσ µν = Rαβρσ Rαβµν Rρσ µν , (B.5)
2
1
Rαβ ρσ Rαµβν Rρµσν = Rαβρσ Rαβµν Rρσ µν , (B.6)
4
1
Rαρβσ Rαµβν Rρ ν σ µ = Rαρβσ Rα µ β ν Rρµσν − Rαβρσ Rαβµν Rρσ µν . (B.7)
4
For manifolds with sufficiently low dimensions some of the basis elements
become linearly dependent, or even vanish.
If d < 6 the contraction of three Riemann tensors, antisymmetrized over all
lower indices, satisfies
R[αβ αβ Rρσ ρσ Rµν] µν = 0 . (B.8)
This identity allows to express R310 in terms of the remaining basis elements.
For d = 4 the contraction of the Weyl tensor furthermore satisfies

Cµνρα C µνρβ = 14 Cµνρσ C µνρσ δαβ . (B.9)

Thus in four dimensions R38 can be expressed in terms of R37 . As a consequence,

the derivative expansion of the gravitational Effective Average Action in d = 4,
at sixth order, contains only eight of the ten basis elements listed in (B.1).
For d ≤ 3 the set of basis elements is reduced further due to the vanishing
of the Weyl tensor. Finally, two-dimensional manifolds are Einstein spaces such
that the Ricci tensor can be expressed in terms of the Ricci scalar and the metric.

B.2 Prototypical Background Spaces

Expanding an arbitrary functional of gµν in the basis (B.1) the corresponding
expansion coefficients can be found by evaluating this functional on a number of
inequivalent metrics. Those metrics will give certain known values to the basis ele-
ments and their linear combinations. Hence, if we know the value of the functional
 B.2 Prototypical Background Spaces 297

for a large enough set of such “backgrounds,” we are able to deduce the func-
tional’s expansion coefficients. Next, we list a number of typical spaces employed
for this purpose.

(1) The S 4 background. Most of the computations carried out within the
Asymptotic Safety program utilize the one-parameter family of four-spheres as a
background. Since these backgrounds are maximally symmetric, using, in d = 4,
[ ( ( ))]
ds2 = a2 dθ12 + sin2 θ1 dθ22 + sin2 θ2 dθ32 + sin2 θ3 dφ2 (B.10)

leads to significant simplifications. In particular, the Riemann tensor and Ricci

tensor satisfy
1 R
Rµν = gµν R , Rµνρσ = (gµρ gνσ − gµσ gνρ ) . (B.11)
4 12
Hence, R = 12/a2 , and the Riemann tensor is covariantly constant in this case,
Dα Rµνρσ = 0. Moreover, Cµνρσ = 0.

(2) The CP 2 background. In order to include the effects stemming from

the Weyl tensor, non-spherical backgrounds are needed. One class of such back-
grounds is the complex projective space CP 2 equipped with the one-parameter
family of (scaled) Fubini-Study metrics:

a2 [ ]
ds2 = i 2
(1 + zi z̄ i )dzj dz̄ j − z̄ j zi dzj dz̄ i , i, j = 1, 2 . (B.12)
(1 + zi z̄ )
This background is still Einstein but has a nontrivial, covariantly constant Weyl
tensor:
1
Rµν = gµν R , Cµνρσ ̸= 0 , Dα Cµνρσ = 0 . (B.13)
4
Thus, on CP 2 the Goroff-Sagnotti term [2, 387, 388] is given a non-vanishing
value.1
For completeness we evaluate the basis elements (B.1) for the particular back-
grounds (B.10) and (B.12). The result is summarized in Table B.1.

(3) Ricci-flat backgrounds. The computation [388] is carried out on a Ricci-

flat background, implying that the curvature tensors satisfy

R=0, Rµν = 0 , Cµνρσ ̸= 0 . (B.14)

An example for a compact space satisfying these properties is a K3 manifold.2

Unfortunately, it is very hard to write down the explicit form of the metric in

1 The RG flow of the Goroff-Sagnotti term was explored in the Asymptotic Safety context in
[386].
2 A non-compact example is a pp-wave background.
298 Organizing the Derivative Expansion

Table B.1 Evaluation of the basis elements on the four-sphere, the complex
projective space CP 2 , the K3 manifold, a generic Einstein manifold, and the
non-Einstein background S 2 × S 2 .

Einstein
S4 CP 2 K3 space S2 × S2

8π 2 a4 π 2 a4
Vol 3 2
Vol Vol 16π 2 a21 a22
( 2 )
R1 12
a2
24
a2
0 R1 a2
2
2 a1 + a2
2
1 a2
( 2 )
2 2
R21 144
a4
576
a4
0 R21 a4
4
4 a1 + a2
1 a2
( 2 )2
R22 0 96
a4
R22 R22 4
3a4 a4
a1 + a22
1 2
R23 24
a4
192
a4
R22 R22 + 16 R21 a2
8
2
1 a2
R31 0 0 0 0 0
R32 0 0 0 0 0
( 2 )3
R33 1728
a6
13824
a6
0 R33 a6
8
6 a1 + a22
1 a2
( 2 )( 4 )
R34 432
a6
3456
a6
0 R
1 3
4 3 a6
4
6 a1 + a2
2 a1 + a42
1 a2
( 6 )
R35 108
a6
864
a6
0 1
R3
16 3 a6
2
6 a1 + a2
6
1 a2
( 2 ) (
2 2 a2 + a2
)
R36 0 0 0 0 3a6
2
a 6 a1 − a2 1 2
1 2
( 2 )
2 3
R37 0 2304
a6
0 R37 3a6
8
6 a1 + a2
1 a2
( 2 )
2 3
R39 0 384
a6
R39 R39 9a6
4
6 a1 + a2
1 a2
( 2 )
2 3
R38 0 576
a6
0 R
1 3
4 7 3a6
2
6 a1 + a2
1 a2
( 2 )3
R310 0 192
a6
R
1 3
2 9
R310 2
9a6 a6
a1 + a22
1 2

this case, so one is forced to manipulate the curvature monomials in an abstract

way.
Important information is thereby contained in the identity

Rµνρσ D2 Rµνρσ = 4Rµνρσ Dµ Dρ Rνσ + 2Rµν Rµ ρσλ Rνρσλ

(B.15)
− Rµν ρσ Rρσ αβ Rαβ µν − 4Rα µ β ν Rµ ρ ν σ Rρ α σ β .

On a Ricci-flat manifold of dimension d < 6 this identity reduces to

Cµνρσ D2 C µνρσ = −3Cµν ρσ Cρσ αβ Cαβ µν . (B.16)

Thus, we conclude that a non-vanishing value for the invariant R39 requires that
Dα Cµνρσ ̸= 0. A computation on a Ricci-flat background therefore entails that
Cµνρσ does not commute with covariant derivatives and one has to keep track of
all terms containing two powers of the Weyl tensor and two covariant derivatives
acting on them.
 B.2 Prototypical Background Spaces 299

(4) Generic Einstein background. By definition, the Ricci tensor of an

Einstein space is proportional to the metric,
1
Rµν = R gµν . (B.17)
d
By virtue of the Bianchi identities (A.10) this entails that

Dµ R = 0 , Dσ Cµνρσ = 0 . (B.18)

Thus, evaluating the flow equation on a generic Einstein background allows us

to distinguish two of the three fourth-order invariants, as well as four out of the
ten sixth-order invariants.

(5) The background S 2 × S 2 . In order to extract complementary information

on the RG flow non-Einstein backgrounds are needed. The simplest class of such
backgrounds is given by the direct product of two two-spheres with different
radii: ( ) ( )
ds2 = a21 dθ12 + sin2 θ1 dθ22 + a22 dθ32 + sin2 θ3 dθ42 . (B.19)
For a1 ̸= a2 this metric does not satisfy the Einstein condition. But the resulting
Weyl tensor is covariantly constant though:

Dα Cµνρσ = 0 . (B.20)

This class of backgrounds allows us to distinguish two of the three fourth-order

invariants, as well as three out of the ten sixth-order invariants.
Notably, the information content of this projection is complementary to the
information obtained on a generic Einstein background. Carrying out analo-
gous computations using both the background classes S 4 and S 2 × S 2 allows
to determine the RG flow of all fourth-order couplings and three sixth-order
couplings.
 Appendix C
Metric Variations

The construction of the flow equation typically requires the functional differen-
tiation, or equivalently variation of covariant objects constructed from gµν and
its derivatives. The behavior under gµν → gµν + δgµν of various basic quantities
is summarized in Table C.1.
The variations of typical invariants building up Γk can be obtained from these
basic variations via the product rule. For example
(√ 2 ) √ √ √ 2 √
δ2 gR = R2 δ 2 g + 4R (δ g) (δR) + 2 g (δR) + 2 gRδ 2 R. (C.1)

Once the complexity of the monomials increases, it is economical to construct

the variations of the more complicated invariants using suitable computer algebra
packages. In this case the Table C.1 can serve as a benchmark which allows to
test the implementation.

Table C.1 Behavior of selected covariant objects under variation of the metric,
gµν → gµν + δgµν . Indices are raised with g µν .

δgµν ≡ hµν
δg µν = −hµν
( )
δΓµν λ = 1 λσ
2
g Dµ hνσ + Dν hµσ − Dσ hµν
√ 1 √
δ g = 2
g g µν hµν
( )
δRµνρ λ = 1
2
−Dµ Dρ hν λ + Dν Dρ hµ λ + Dµ Dλ hνρ − Dν Dλ hµρ − Rµνρ σ hσ λ + Rµνσ λ hρ σ
1
(
δRµν = 2 Rµσ hν + Rν hσµ + 2Rσµνλ h + Dν Dσ hσ µ
σ σ σλ
)
−Dµ Dν hα α − Dλ D λ hµν + Dµ Dσ hσν
( )
δR = −Rµν hµν + Dβ Dα hαβ − Dβ hα α
√ √ ( )
δ2 g = 12 g 12 g µν g ρσ hµν hρσ − hµν hµν
δ2 R = Rβµ hβγ hγ µ − Rαβγρ hβγ hαρ − 3hβγ Dγ Dα hα β + 2hβγ Dβ Dγ hα α + 2hβγ Dλ Dλ hβγ
− hβγ Dα Dβ hα γ − (Dα hβγ )(Dβ hα γ ) + 32 (Dλ hβγ )(Dλ hβγ ) − 2(Dγ hβγ )(Dα hα β )
+ 2(Dβ hβγ )(Dγ hα α ) − 12 (Dλ hγ γ )(Dλ hα α )
 Appendix D
Heat Kernel Techniques

Structurally the right-hand side of the functional renormalization group equation

is given by a trace over an operator-valued function. Typically the underlying
differential operators are quite involved. For specific choices of the background
geometry and gauge-fixing conditions they may simplify to differential operators
whose spectral properties have already been studied in the mathematical physics
literature.
A typical example are differential operators of second order,

∆ ≡ −g µν ∇µ ∇ν + E, (D.1)

where ∇µ ≡ Dµ + Aµ is a covariant derivative which may include a Yang–Mills

connection Aµ in addition to the Levi–Civita connection implicit in Dµ . The
endomorphism E is a linear map acting on spacetime and internal indices carried
by the field.
In such cases, a powerful tool for evaluating the flow equations are heat kernel
techniques [406–409]. The central idea is to introduce the heat kernel

K(s; x, y; ∆) ≡ ⟨x|e−s∆ |y⟩. (D.2)

The heat kernel satisfies the heat equation

(∂s + ∆x ) K(s; x, y; ∆) = 0, (D.3)

subject to the initial condition

lim K(s; x, y; ∆) = δ(x, y). (D.4)

s→0

For example, on a flat manifold Rd with Aµ = 0 and the endomorphism E = m2

given by the mass of a scalar field the heat kernel is known exactly:
( )
1 (x − y)2
K(s; x, y; ∆) = exp − − sm .
2
(D.5)
(4πs)d/2 4s
302 Heat Kernel Techniques

The trace of the operator e−s∆ can be expressed in terms of the heat kernel
evaluated at coincident points:
∫
[ ] √
Tr e−s∆ = dd x g K(s; x, x; ∆) . (D.6)

The right-hand side of this equation can then be studied by applying various
complementary approximation schemes to K, like the early or the late time
expansion.

D.1 Early Time Expansion

The early time expansion organizes the operator trace (D.6) in terms of an
asymptotic series in s:
∫ ∞
[ ] 1 √ ∑
Tr e−s∆ = d
d x g tr (an ) sn . (D.7)
(4πs)d/2 n=0

Here tr denotes a trace over the internal space on which Aµ and E act as matrices.
Comparing with (D.6) indicates that the early-time expansion of K(s; x, x; ∆) is
given by
∑∞
1
K(s; x, x; ∆) = tr (an ) sn . (D.8)
(4πs)d/2 n=0

Dimensional analysis shows that the mass dimension of the expansion coeffi-
cients an is given by [an ] = 2n. This makes the early time expansion particularly
suitable for evaluating the gravitational FRGE in a derivative expansion since
terms built from 2n derivatives appear in the coefficient an only.
The an can be expressed in terms of the curvature tensors constructed from
gµν , Aµ , and the endomorphism E. For n = 0, 1, 2 their explicit form can be found
in [408].1 For the operator (D.1) they are given by

a0 = 1, (D.9)
a1 = R 1 − E,
1
6 (D.10)
( µνρσ )
1
a2 = 180 R Rµνρσ − Rµν Rµν + 25 R2 + 6∇2 R 1
1
+ 12 Ωµν Ωµν − 16 R E + 21 E2 − 16 ∇2 E. (D.11)

1 Our conventions differ from the ones used in [408] by a relative minus sign in the endomor-
phism E and the Riemann tensor. The Ricci tensor and Ricci scalar agree. Results given in
[409] can be converted to ours by applying the replacement rules is 7→ s, Rµν 7→ Ωµν and a
rescaling ak 7→ k! ak . The conventions for the curvature tensors agree in this case.
 D.1 Early Time Expansion 303

Here Ωµν = [∇µ , ∇ν ] is the total curvature of the connection and 1 denotes the
unit on the internal space. For scalars (S), vectors (V ), and symmetric second-
rank tensors (ST ) one has, for instance,
( )
1S = 1, [1V ]µ ν = δµν , [1ST ]µν αβ = 21 δµα δνβ + δµβ δνα . (D.12)

The Ω-dependent terms capture the dependence of the heat kernel coefficients
on the matrix properties of the fields. For any fixed set of internal indices, they
give rise to specific matrix-valued contributions built from contractions of the cur-
vature tensors and their covariant derivatives. Their explicit expressions can be
obtained by evaluating the commutators on a “test” field carrying the appropriate
set of internal indices.
We illustrate the role of Ω by various examples. On flat space one has

Ωµν Ωµν = Fµν Fµν (D.13)

with Fµν the field strength associated with the Yang-Mills connection. In this
case the trace tr is over matrices constructed from the generators of the gauge
group.
In curved space and for Aµ = 0 we see that Ω vanishes when the heat kernel
is evaluated for scalar fields. For a vector field ξα one has instead

Ωµν Ωµν ξα = −Rµνρα Rµνρβ ξβ , (D.14)

where we used the relations (A.11) for converting the commutators to Riemann
curvature tensors. From (D.14) one then reads off the matrix

[Ωµν Ωµν ]α β = −Rµνρα Rµνρβ . (D.15)

The analogous computation for symmetric second-rank tensors yields

[Ωµν Ωµν ]αβ ρσ = −Rµνγα Rµνγρ δβσ − Rµνγβ Rµνγσ δαρ + 2Rµνα ρ Rµν β σ . (D.16)

The expansion (D.7) contains the trace of matrix valued quantities whereby
the outer indices are contracted. For the unit matrix, these traces count the
number of independent components:

trS 1 = 1, trV 1 = d, trST 1 = 12 d(d + 1). (D.17)

Tracing the outer indices in (D.15) and (D.16), taking into account the
appropriate symmetrizations, furthermore yields

trV [Ωµν Ωµν ] = −Rµνρσ Rµνρσ ,

(D.18)
trST [Ωµν Ωµν ] = −(d + 2) Rµνρσ Rµνρσ .
304 Heat Kernel Techniques

For completeness, we also display the coefficient a3 . Its general expression is

given in [408]. Upon dropping surface terms, integrating by parts, and using the
Bianchi identities, the result reads [187, 409]:
[( )3 ( ) ( )
a3 = 16 16 R 1 − E + 12 61 R 1 − E ∇2 16 R 1 − E
(1 ) ( µνρσ )
6R 1 − E Rµνρσ − Rµν Rµν + ∇2 R
1
+ 30 R
( )
+ 12 16 R 1 − E Ωµν Ωµν − 10 1
(∇µ Ωµα ) (∇ν Ωνα )
1
+ 30 (2Ωµ ν Ων ρ Ωρ µ − 2Rµ ν Ων ρ Ωρ µ + Rµνρσ Ωµν Ωρσ )
− 630
1
R ∇2 R 1 + 140
1
Rµν ∇2 Rµν 1
(
1
+ 7560 48Rµν Rρσ Rµρνσ − 64Rµ ν Rν ρ Rρ µ + 6Rµν Rµ αβγ Rναβγ
) ]
+ 17Rµν ρσ Rρσ αβ Rαβ µν − 28Rα µ β ν Rµ ρ ν σ Rρ α σ β 1 . (D.19)

D.2 The Off-Diagonal Heat Kernel

The early-time expansion of the heat kernel, discussed in the previous section,
provides a powerful tool for evaluating operator traces where the trace argument
is given by a minimal second-order differential operator. The off-diagonal heat
kernel techniques constitute an important generalization of these methods.
First introduced by DeWitt [50, 410] as a technique to compute the standard
heat kernel coefficients appearing in (D.8), the methods are also useful when
evaluating FRGEs [411]. Their key benefit is that they allow to construct the
early-time expansion of operator traces of the form
[ ] [ ]
Tr O e−s∆ = Tr Oµ1 ...µa Dµ1 · · · Dµa e−s∆ . (D.20)
Here O is an operator insertion which may be matrix-valued in field space. Typ-
ically the covariant derivatives will be with respect to the background metric
and capture derivatives that do not organize themselves into Laplacian type
operators. A frequent example is the insertion O = R̄µν D̄µ D̄ν .
Without loss of generality, one can assume that the tensors Oµ1 ...µa are totally
symmetric under all permutations of indices. Any antisymmetric contribution
in Oµ1 ...µa gives rise to commutators of covariant derivatives which can be re-
expressed in terms of curvature tensors via the standard commutator relations
(A.11). The resulting terms are again of the form (D.20) with the resulting
components Oµ1 ...µa having a lower rank.
Writing the RHS of (D.20) in terms of matrix elements,
∫ √ ∫ √
[ ]
Tr O e−s∆ = tr dd x g(x) dd y g(y)⟨ y | O | x ⟩⟨ x |e−s∆ | y ⟩, (D.21)
 D.2 The Off-Diagonal Heat Kernel 305

relates the traces with operator insertions O to the heat kernel K(s, x, y) at
non-coincident points. Precisely, (D.21) entails the master formula
∫
[ ] √
Tr O e−s∆ = tr dd x g Oµ1 ...µa Kµ1 ...µa , (D.22)

where tr denotes a trace over internal field indices.

In order to ease our notation, we indicate expressions evaluated in the coinci-
dence limit y → x by an overline. The tensors Kµ1 ...µa are covariant derivatives of
the off-diagonal heat kernel K(s; x, y) with respect to its argument x evaluated
at the coincidence point y = x:

Kµ1 ...µa ≡ D(µ1 . . . Dµa ) K(s; x, y). (D.23)

The definition (D.23) involves the heat kernel at non-coincidence points. There-
fore, the evaluation of operator traces via the master formula (D.22) is commonly
referred to as the off-diagonal heat kernel technique.
The tensors Kµ1 ...µa possess an early-time expansion which has the same struc-
ture as the one encountered for the diagonal heat kernel K(s; x, x) in (D.8).
The coefficients appearing in this expansion can be obtained by noting that the
off-diagonal heat kernel has the asymptotic expansion

∞
∑
−d/2 −
σ(x,y)
K(s; x, y) = (4πs) e 2s sn An (x, y). (D.24)
n=0

subject to the boundary condition

A0 (x, y) = 1. (D.25)

The world function σ(x, y) which appears in (D.24) is given by half the geodesic
distance between x and y and satisfies

1
2 σ;µ σ
;µ
= σ, σ(x, y) = σ(x, y);µ = 0, σ(x, y);µ(α1 ...αa ) = 0, a ≥ 2. (D.26)

A non-trivial contribution from σ(x, y) arises when exactly two covariant

derivatives act on it in a symmetric way:

σ(x, y);(µν) = gµν (x). (D.27)

The properties of the world function allow to express the tensors Kµ1 ...µa in
terms of the metric and symmetrized covariant derivatives of the off-diagonal
306 Heat Kernel Techniques

heat kernel coefficients An evaluated at the coincidence point. For the ten-
sors with rank up to four, the resulting combinatorics leads to the following
formulae:
1 ∑
K= sn An ,
(4πs)d/2 n≥0
1 ∑
Kµ = d/2
sn Dµ An ,
(4πs) n≥0
∑ ( )
1 1
K(µν) = sn−1
− gµν An + D(µ Dν) An−1 ,
(4πs)d/2 n≥0 2
∑ ( ) (D.28)
1 3
K(µνρ) = sn−1
− g D
(µν ρ) nA + D D D A
(µ ν ρ) n−1 ,
(4πs)d/2 n≥0 2
∑ (
1 3
K(µνρσ) = d/2
sn−2
g(µν gρσ) An − 3g(µν Dρ Dσ) An−1
(4πs) 4
n≥0
)
+ D(µ Dν Dρ Dσ) An−2 .

It is also useful to consider the tensors K(α1 ...α2a ) in the limit where all
curvatures and the endomorphism piece vanish. Then the only non-vanishing con-
tribution in the formulas (D.28) comes from the A0 -term without any covariant
derivatives. In this limit the combinatorics can be worked out in full generality,
giving the “flat space” result
( )a
1 1 [ ]
K(α1 ...α2a ) = d/2
− gα1 α2 · · · gα(2a−1) α2a + · · · . (D.29)
(4πs) 2s
( )
Here the dots indicate the remaining (2a)! a
2 a! − 1 index permutations of the
first term. The validity of this result can be verified by applying 2a partial
derivatives to the exact off-diagonal heat kernel on flat space (D.5) and setting
m = 0.
Evaluating the expressions (D.28) requires the computation of derivatives
of the off-diagonal heat kernel coefficients at coincident points. They can be
obtained systematically as follows.
Substituting the ansatz (D.24) into the heat equation (D.2) leads to a recursion
relation for the coefficients An and their covariant derivatives:
( )
n − d2 + 12 σ;µ µ An + σ ;µ An;µ − An−1;µ µ + E An−1 = 0, n ≥ 0. (D.30)

Equipping this relation with the boundary condition A−1 = 0, and taking suitably
symmetrized covariant derivatives of it allows to compute the coincidence limit
 D.3 Integral Transforms 307

of the coefficients An and their covariant derivatives. The results for the terms
including up to four covariant derivatives are thus found to be given by

A0 = 1,
Dµ A0 = 0,
D(ν Dµ) A0 = 16 Rµν , (D.31)
D(α Dν Dµ) A0 = 14 R(µν;α) ,
3 1 1 γ δ
D(β Dα Dν Dµ) A0 = 10 R(µν;αβ) + 12 R(αβ Rµν) + 15 Rγ(β|δ|α R µ ν) ,

together with

A1 = 16 R − E,
1
Dµ A1 = 12 R;µ − 16 Ωνµ ;ν − 12 E;µ ,
1
D(ν Dµ) A1 = 90 Rαβγ µ Rαβγν + 1
90 R
αβ
Rαµβν − 45
1
Rµα Rα ν + 36
1
RRµν (D.32)
− 1
60 ∆Rµν + 20 R;µν + 6 Ωα(µ Ω ν) − 6 Ωα(µ; ν)
1 1 α 1 α

− 1
6 Rµν E − 13 E;(µν) ,

and
( µνρσ )
A2 = 1
180 R Rµνρσ − Rµν Rµν + 52 R2 + 6∇2 R 1
1
+ 12 Ωµν Ωµν − 16 R E + 12 E2 − 16 ∇2 E. (D.33)

Notably, the coefficients A0 , A1 , and A2 coincide with the standard early-time

heat kernel coefficients (D.9), (D.10), and (D.11).
The tensors Kµ1 ...µa with a ≤ 6 as well as the coefficients Dµ1 . . . Dµa An with
up to six covariant derivatives can be found in [412, 413]. The complexity of the
coefficients increases rather quickly for increasing a and n [414]. Since they are
typically manipulated within a computer algebra program we refrain from giving
explicit expressions for them.

D.3 Integral Transforms

The FRGE for the Effective Average Action contains functional traces over
functions W (∆) of certain differential operators:

Tr [W (∆)] . (D.34)

Typically, the differential operator, ∆, is constructed from covariant derivatives

built from the background metric. If the theory contains additional gauge fields,
the connection may contain these fields as well. In many cases ∆ reduces to an
308 Heat Kernel Techniques

operator of the type considered above, i.e. a second order, Laplace-type opera-
tor, ∆ = −∇2 1 + E, where the endomorphism E is a linear map acting on the
spacetime and internal indices carried by the fields.

(1) The derivative expansion of the operator trace (D.34) can be obtained from
the early-time expansion of the heat kernel (D.7). Expressing W (z) by its Fourier
transform, as in Chapter 5, or alternatively through its inverse Laplace transform
f (s), according to,
W
∫ ∞
W (z) ≡ f (s) e−sz ,
ds W (D.35)
0

the trace (D.34) can be related to the traced heat kernel:

∫ ∞ [ ]
Tr [W (∆)] = f (s) Tr e−s∆ .
ds W (D.36)
0

Substituting the early-time expansion of the heat kernel (D.7) yields

∑∞ ∫
1 √
Tr [W (∆)] = Q d/2−k [W ] dd x g tr(ak ) (D.37)
(4π)d/2 k=0

with the “Q-functionals” defined analogous to (5.47):

∫ ∞
Qn [W ] ≡ f (s).
ds s−n W (D.38)
0

The factor on the right-hand side of (D.37) involving the spacetime integral
carries the information about the various field monomials, while the Q-functionals
encode the properties of the function W (z).
For positive n, the Q-functional (D.38) is related to W (z) through a Mellin-
transform:
∫ ∞
1
Qn [W ] = dz z n−1 W (z), n > 0. (D.39)
Γ(n) 0

The equivalence of (D.38) and (D.39) is proven by substituting (D.35) into

(D.39) and using the integral representation of the gamma function.
The index n can be continued to negative values. Denoting the k-th derivative
of W (z) by W (k) (z), (D.35) implies
∫
(k)
W (z) = (−1)
k f (s) e−sz .
ds sk W (D.40)

As a consequence,
[ ]
k
Qn [W ] = (−1) Qn+k W (k) . (D.41)
 D.3 Integral Transforms 309

For negative values n ≤ 0, one can then chose an integer k such that n + k > 0.
Following the proof for positive n leads to the general formula
∫ ∞
(−1)k
Qn [W ] = dz z n+k−1 W (k) (z) . (D.42)
Γ(n + k) 0

Given a specific function W (z), (D.39) and (D.42) determine the moments Qn [W ]
in the expansion (D.37) to any order.

(2) In many practical computations, the relevant values of n are either integers or
half-integers, depending on whether the dimension of spacetime is even or odd. In
these cases the general formula (D.42) simplifies considerably. Evaluating (D.35)
and its derivatives with respect to z at z = 0 establishes that for m = 0, 1, 2, · · · :
m
Q−m [W ] = (−1) W (m) (0). (D.43)

If n = −(2m + 1)/2 is a negative half-integer, it is convenient to choose k = m + 1,

so that
m+1 ∫ ∞
(−1)
Q−m [W ] = √ dz z −1/2 W (m+1) (z) , m ∈ N0 . (D.44)
π 0

(3) In typical computations the cutoff operator Rk is often proportional to the

identity on the internal space, and the endomorphism E is proportional to the
identity matrix in field space: E = q1. Then, with ∆ = (−∇2 + q)1, the argument
W of the Qn functionals reduces to a scalar function of −∇2 + q, or its eigenvalues
p2 + q actually.2
The most common and prototypical form of W (z) is therefore as follows:
2 −1 −p
W (z) ≡ W ( kp2 ) = (Zk ) (Pk + q) ∂t (Zk Rk ) . (D.45)

It contains a normalization constant Zk , a certain power of the regularized inverse

propagator Pk (p2 ) ≡ p2 + k 2 R(0) (p2 /k 2 ), and the scale derivative of the cutoff
operator ( 2)
p
Rk (p2 ) = k 2 R(0) . (D.46)
k2
Here R(0) (z) is a dimensionless shape function capturing the freedom of selecting
a largely arbitrary cutoff (see Appendix E).
It is convenient to employ the dimensionless variables
p2 q
z≡ , w≡ , (D.47)
k2 k2

2 Here and also often in the main text the suggestive notation p2 is used for the eigenvalues
of arbitrary Laplacians −∇2 . Note that no flat spacetime is implied.
310 Heat Kernel Techniques

and to express Qn [W ] in terms of an appropriate basis set of standard dimen-

sionless threshold functions. In the case at hand the following two families of
such functions are sufficient [14]:
∫ ∞
1 R(0) (z) − zR(0)′ (z)
Φpn (w) ≡ dz z n−1 [ ]p ,
Γ(n) 0 z + R(0) (z) + w
∫ ∞
(D.48)
e pn (w) ≡ 1 R(0) (z)
Φ dz z n−1
[ ]p .
Γ(n) 0 z + R(0) (z) + w

Substituting (D.45) into (D.39) and rewriting the Q-functional in terms of the
dimensionless variables (D.47) then leads to the useful relation:
[ ] ( )
∂t (Zk Rk ) e pn (w) .
Qn p =k
2n−2p+2
2Φpn (w) − η Φ (D.49)
Zk (Pk + q)

Here η is the “anomalous dimension” associated with the running normalization

constant Zk :
−1
η ≡ − (Zk ) ∂t Zk . (D.50)
(4) The threshold functions (D.48) encode the regulator-dependence of the beta
functions, i.e., they are manifestly R(0) -dependent in general. But certain combi-
nations of threshold functions assume universal values though, i.e., they do not
depend on the specific choice of shape function but only on the generic properties
of the regulator (2.65).
An example for such a universal behavior are threshold functions where the
integrand happens to be a total derivative:
∫ ∞ ( )
1 d zn
n+1
Φn (0) = dz ( )n . (D.51)
Γ(n + 1) 0 dz z + R(0) (z)

Using only that limz→∞ R(0) (z) is required to be finite, this implies that

1
Φn+1
n (0) = , n ≥ 0, (D.52)
Γ(n + 1)

independently of all local details of the function R(0) (z).

For n = 0 the result (D.52) is established by a direct evaluation of the Q-
functional via (D.43).
The identity (D.52) serves as a prototypical example showing how renormal-
ization scheme-independent quantities arise from the FRGE.
A second instructive example showing how universality emerges is the EAA-
based computation of the trace anomaly in two dimensions, which involves the
interplay of several threshold functions with complicated arguments; see [415] for
the details.
 Appendix E
Cutoff- and Threshold Functions

The explicit calculation of renormalization group trajectories makes it necessary

2
to concretely specify the cutoff, or regulator function Rk (p2 ) = Zk k 2 R(0) ( kp2 ).
One requires that it satisfies the following general properties:
(i ) Rk (p2 ) → 0 for k → 0. This property ensures that limk→0 Γk = Γ0 coincides
with the usual effective action.
(ii ) Rk (p2 ) approaches zero for p2 ≫ k 2 , entailing that the high-momentum
modes with p2 ≫ k 2 are not affected by the regulator. At the level of
the path integral the corresponding modes are integrated out without
suppression factor.
(iii ) Rk (p2 ) ∝ k 2 for p2 . k 2 , i.e., the low-momentum modes acquire an effective
mass of order k, suppressing their contribution in the flow equation. More-
over, this requirement ensures that the regulator diverges if k is sent to
infinity, completely suppressing the propagation of all modes in this limit.
These general properties can be met in a number of ways. Quite often their
precise implementation is tailored to the problem under investigation. The most
frequently used cutoff shape functions R(0) (z) are discussed below and listed in
Table E.1. For clarity the wave function factor Zk is set to one in the sequel.

Table E.1. Cutoff shape functions R(0) (z) and their respective virtues

Analytical
threshold Complex
Regulator R(0) (z) functions Smooth frequencies

z
Exponential (exp) ez −1
X
Sharp (sc) rb Θ(1 − z) X
Optimized (opt) (1 − z)Θ(1 − z) X
1
Analytic poly 1+c1 z+c2 z 2 +···
X
z
Analytic exp 2 X X
ez −1
Mass 1
312 Cutoff- and Threshold Functions

E.1 Exponentially Decaying Cutoffs

The characteristic feature of this class of cutoffs is that they vanish exponentially
for p2 ≫ k 2 . The prototype of a cutoff shape function R(0) belonging to this
class is
z
R(0) (z)exp = z , (E.1)
e −1
and its one-parameter generalization
sz
R(0) (z; s)exp = , s > 0. (E.2)
esz − 1

The key advantage of these regulators is that they are smooth, infinitely differen-
tiable functions. This makes them an attractive choice when computing derivative
expansions to high order.
As a drawback the integrals appearing in the threshold functions can, in gen-
eral, not be carried out analytically. The beta functions are then given by a set of
coupled integrodifferential equations, entailing that computing RG trajectories
may be numerically more expensive than for regulators with threshold functions
that can be evaluated analytically.
For vanishing arguments the threshold functions related to the generalized
exponential cutoff R(0) (z; s)exp in (E.2) can be obtained analytically. Using the
integral
( ) ∫representations for polylogarithms
( ) ∫ and ν−1
the Riemann ζ-function, Liν (s) =
1 ∞ sz ν−1 1 ∞ z
Γ(ν) 0
dz ez −s and ζ(ν) = Γ(ν) 0 dz ez −1 [416], respectively, one easily
verifies that, for example,
n{ }
Φ1n (0; s)exp = ζ(n + 1) − Lin+1 (1 − s) ,
sn
1 (E.3)
Φ2n (0; s)exp = n−2 Lin−1 (1 − s).
s (1 − s)

For non-vanishing arguments no analytic solution to the integrals is known.

E.2 The Sharp Cutoff

The sharp cutoff (sc) is best described using dimensionful variables. It reads [160]
( )
b q2
Rk (q ) = R Θ 1 − 2 .
2 sc
(E.4)
k

Here Θ denotes the Heaviside step function and the limit R b → ∞ is taken after
the q -integration or the integration over z ≡ q /k in the threshold functions.
2 2 2

This choice makes the coarse-graining procedure particularly transparent: the

cutoff vanishes for the UV modes while the IR modes acquire an infinite mass
term in the limit Rb → ∞. A highly welcome feature of the sharp cutoff is that
 E.2 The Sharp Cutoff 313

it permits an explicit analytical evaluation of the associated threshold functions

(D.48) [160]:
1
Φ1n (w)sc = − ln(1 + w) + φn ,
Γ(n)
1 1 1
Φpn (w)sc = , p > 1,
Γ(n) (p − 1) (1 + w)p−1 (E.5)
e 1n (w)sc = 1
Φ ,
Γ(n + 1)
e pn (w)sc = 0,
Φ p > 1.

Furthermore, an additional bonus of the sharp cutoff is that via the constants
φn ≡ φn (s), s > 0, which appear in (E.5), this cutoff actually amounts to a family
of shape functions labeled by a parameter s > 0 similar to the generalized expo-
nential cutoffs (E.2).

(1) It is instructive to understand why the s-independent function (E.4) actually

gives rise to a family of threshold functions. To see this, recall that the z-integrals
(5.51) for the threshold functions resulted from setting Rk (q 2 ) = k 2 R(0) (q 2 /k 2 ).
Since (E.4) is not of this form we first reintroduce Rk (q 2 ) and rewrite (5.51)
accordingly.1 The evaluation of the integrals defining the threshold functions
then proceeds as follows. With z ≡ q 2 /k 2 and w ≡ v/k 2 they read

(v) ∫ ∞
1 2 n−1 R (q /k 2 ) − (q 2 /k 2 )R(0)′ (q 2 /k 2 )
(0) 2
Φpn = (k 2 p−n
) dq 2
(q ) .
k2 Γ(n) 0 (q 2 + Rk (q 2 ) + v)p
(E.6)
Here the prime denotes the derivative of R(0) with respect to its argument. The
integrand in equation (E.6) can then be written as a derivative with respect to
k, for any p > 1:
(v) ∫ ∞
(k 2 )p−n−1 D (q 2 )n−1
Φpn = − dq 2
k
k2 2Γ(n)(p − 1) 0 Dk (q 2 + Rk (q 2 ) + v)p−1
2 p−n−1 ∫ ∞
(E.7)
(k ) ∂ (q 2 )n−1
=− dq 2 k .
2Γ(n)(p − 1) 0 ∂k (q 2 + Rk (q 2 ) + ṽ)p−1 ṽ=v

Here the derivative D/Dk acts by definition only on the k-dependence of Rk ,

but not on v. In order to rewrite D/Dk in terms of ∂/∂k we introduced the
constant ṽ, which is strictly independent of k. Only at the end of the calculation
it is identified with v ≡ wk 2 .

1 b 2 . How-
Formally we have R(0) (z)sc = rb Θ(1 − z) with the dimensionless parameter rb ≡ R/k
b → ∞ and rb → ∞ are equivalent only for finite and non-zero values
ever, note that the limits R
of k, which may require special care.
314 Cutoff- and Threshold Functions

For p = 1 the formula (E.7) breaks down since its right-hand side is no longer
well defined. Assuming p > 1, we interchange the q-integration and the derivative
with respect to k. This is allowed since the integral in (E.7) is absolutely con-
b Θ(1 − q 2 /k 2 ) one
vergent. If one now substitutes the sharp cutoff Rk (q 2 )sc ≡ R
finds:
( v )sc ∫ ∞
(k 2 )p−n−1 ∂ (q 2 )n−1
Φpn = − k dq 2
( ) .
k2 2Γ(n)(p − 1) ∂k 0 b Θ(1 − q22 ) + ṽ p−1
q2 + R k ṽ=v
(E.8)
b → ∞ restricts the momentum integration to q 2 > k 2 :
Taking the limit R
( v )sc ∫ ∞
(k 2 )p−n−1 ∂ (q 2 )n−1
Φpn = − k dq 2 2 . (E.9)
k 2 2Γ(n)(p − 1) ∂k k2 (q + ṽ)p−1 ṽ=v

The resulting integral is trivially evaluated by acting with the k-derivative on

the lower integration limit. This yields our final result for the threshold functions
with a sharp cutoff,

1 1 1
Φpn (w)sc = for p > 1, (E.10)
Γ(n) (p − 1) (1 + w)p−1

which is in accord with (E.5).

(2) One easily verifies that for p > 1 the threshold functions (E.10) satisfy the
recursion relation (5.97) even though the latter was derived for a smooth cutoff.
We now use this recursion relation in order to define what Φ1n (w)sc at p = 1
d
is supposed to be. We insist that it solves the differential equation dw Φ1n (w)sc =
− Γ(n) 1+w that arises from substituting the known Φn (w) into (5.97). Its
1 1 2 sc

solution contains a free constant of integration:

1
Φ1n (w)sc = − ln(1 + w) + Φ1n (0)sc . (E.11)
Γ(n)

The integration constants Φ1n (0)sc ≡ φn are arbitrary numbers a priori. They
parametrize a residual cutoff scheme dependence which is still present after
having opted for a sharp cutoff.
To parametrize this residual cutoff scheme dependence it is often convenient to
focus on a one-parameter family of numbers {φn }. In [160] it has been proposed to
relate them to the threshold functions obtained with the s-dependent exponential
cutoff functions (E.2). This is done by requiring that the threshold functions for
the sharp and exponential cutoffs agree at w = 0. For this value of the argument
we may use the analytic formula (E.3) then:

!
φn (s) ≡ Φ1n (0)sc(s) = Φ1n (0; s)exp . (E.12)
 E.3 The Optimized Cutoff 315

ϕ1 ϕ2
7
3.5
6
3
2.5 5
2 s=0.8 4
3 s=0.8
1.5 s=1
s=1
1 s=5 2
s=10
0.5 s=30 1 s=5 s=10 s=30
s s
5 10 15 20 25 30 5 10 15 20 25 30

Figure E.1. Dependence of the constants of integration φ1 and φ2 , (E.12),

on the shape parameter s.

In particular, one has φn (1) = nζ(n + 1) with ζ(n) being the Riemann zeta func-
tion. The resulting functions φ1 (s) and φ2 (s), which incidentally are the only ones
needed for the Einstein–Hilbert truncation in d = 4, are shown in Figure E.1.
e pn -functions the integral appearing in (D.48) can be evaluated unam-
(3) For all Φ
biguously after substituting the sharp cutoff. In a calculation similar to the one
above we obtain the results given in (E.5).
Thanks to its explicitly computable threshold functions and the simple
possibility of “deforming” it to a full family of cutoffs, the sharp cutoff rep-
resents a convenient tool for reliability tests of truncations using universal, i.e.,
scheme-independent quantities. (See [160] for further details.)

E.3 The Optimized Cutoff

Another convenient choice for the cutoff function is the so-called “optimized”
shape function [64], of the form

R(0) (z)opt = (1 − z) Θ(1 − z) (E.13)

with Θ again denoting the Heaviside step function. In this case the integral
defining the threshold functions (D.48) can be carried out analytically, yielding

1 1
Φpn (w)opt = ,
Γ(n + 1) (1 + w)p
(E.14)
e pn (w)opt = 1 1
Φ .
Γ(n + 2) (1 + w)p

In special cases it is possible to show that the regulator (E.13) is “optimal” in the
sense that it minimizes the systematic error when computing a physical quantity
within a given truncation; in this sense it “optimizes” the physics predictions
[64]. Moreover, the Q-functionals appearing in (D.49) vanish identically when
evaluated with R(0) (z)opt if n is sufficiently negative [185, 186].
316 Cutoff- and Threshold Functions

E.4 Regulators for Complex Frequencies

When computing real-time correlation functions the argument of the regulator
may actually become complex. In this case it becomes important that one can
control the residues induced by the regulator within the complex plane. In this
context one may resort to regulator functions featuring an algebraic decay [417],
[ ]−1
R(0) (z) = 1 + c1 z + c2 z 2 + · · · , (E.15)

with ci free constants, or an exponential decay [418]

[ 2 ]−1
R(0) (z) = z ez − 1 . (E.16)

Regulators of this type are relevant in setting up flow equations on spacetimes

with Lorentzian signature.

E.5 Mass-Type Cutoff

Strictly speaking, the mass-type cutoff defined by

Rk (p2 ) = k 2 ⇔ R(0) (z) = 1 (E.17)

is not a valid regulator function since it does not fall off for p2 ≫ k 2 and thus
violates the general requirement (ii ). Its main advantage is that it leads to very
simple momentum integrals and therefore is often used for a first orientation. Its
main disadvantage is that with (E.17) the factor ∂t Rk under the supertrace in the
FRGE is now p2 -independent and no longer renders the traces UV convergent.
Therefore, if one uses (E.17), an explicit UV regularization of some sort must be
introduced.
 Appendix F
Field Decompositions

A key step in the evaluation of the (truncated) functional renormalization group

(2)
equation (4.80) is the inversion of the Hessian (Γk + Rk ). This step can often
be facilitated by rewriting the fluctuation fields in terms of suitable component
fields. This appendix introduces the two most commonly used decompositions.
A typical example for a field decomposition of the metric fluctuations hµν is
the traceless decomposition encountered in (5.27). In this case hµν is expressed
in terms of a traceless symmetric tensor h̊µν and its trace ϕ according to
1
hµν = h̊µν + ḡµν ϕ. (F.1)
d
The component fields are then subject to the constraints

ḡ µν h̊µν = 0, ϕ = ḡ µν hµν . (F.2)

Note that the Jacobian related to the transverse decomposition of the field is
trivial.
The transverse-traceless (TT) decomposition [384, 385] provides a useful
refinement of the transverse decomposition, setting
2 1
µν + D̄µ ξν + D̄ν ξµ + 2D̄µ D̄ν σ − ḡµν D̄ σ +
hµν = hT 2
ḡµν ϕ. (F.3)
d d
In this case all component fields appearing on the right-hand side are transverse
with respect to the background covariant derivative and satisfy the differential
constraints

D̄µ hT
µν = 0, ḡ µν hT
µν = 0, D̄µ ξ µ = 0, ḡ µν hµν = ϕ . (F.4)

Typically the transverse-traceless (TT) decomposition of the metric fluctuations

is accompanied by a transverse decomposition of the vector fields according to

Cµ = CµT + D̄µ c, D̄µ CµT = 0. (F.5)

318 Field Decompositions

Using the component fields appearing in (F.3) and (F.5) has the advantage that
all terms containing contractions of fluctuation fields with a background covari-
ant derivative vanish. Typically this results in a drastic simplification of the
(2)
differential operators contained in (Γk + Rk ); see [411] for a detailed discussion.
Let us summarize the most important properties of the TT decomposition:

(1) Zero modes: From the structure of (F.3) and (F.5), it is clear that not all
modes of the component fields contribute to the metric fluctuations hµν or the
vector fields Cµ . In the metric decomposition, the constant mode of σ, vectors
Cµ ≡ D̄µ σ satisfying the conformal Killing equation D̄µ Cν + D̄ν Cµ − d2 ḡµν D̄α C α =
0, and transversal vectors solving the Killing equation D̄µ ξν + D̄ν ξµ = 0, do not
contribute to hµν . Analogously, the decomposition of a vector field is indepen-
dent of a constant mode c. These modes are unphysical and must be excluded
from the spectrum.

(2) Nontrivial Jacobian: The field decompositions (F.3) and (F.5) give rise
to nontrivial Jacobi determinants
( )1/2 ( )1/2
Jgrav = det′(1T,0) [M ] , Jvector = det′0 [−D̄2 ] . (F.6)

Here [ [ ] ]
(µ,ν) −2 ḡ µν D̄2 + R̄µν −2R̄µλ D̄λ
M = d−1 (F.7)
2D̄λ R̄λν 2 2 µν
d (D̄ ) + D̄µ R̄ D̄ν

is a (d + 1) × (d + 1)-dimensional matrix in field space whose first d components

act on transverse vectors and the last component on scalars, respectively. Fur-
thermore, the prime indicates that the zero modes have been excluded from the
determinants. The Jacobians may be taken into account by introducing suit-
able sets of auxiliary fields along the lines of the Faddeev–Popov trick. The TT
decomposition is( particularly
) powerful on backgrounds satisfying the Einstein
condition R̄µν = d1 ḡµν R̄ where the operator M is block diagonal.
 References

1. G. ’t Hooft and M. Veltman, One-loop divergencies in the theory of gravitation,

Annales de l’I.H.P. Physique théorique 20, 69 (1974).
2. M. H. Goroff and A. Sagnotti, Quantum gravity at two loops, Phys. Lett. B 160,
81 (1985).
3. J. F. Donoghue, Leading quantum correction to the Newtonian potential, Phys.
Rev. Lett. 72, 2996 (1994), [arXiv:gr-qc/9310024].
4. J. F. Donoghue, General relativity as an effective field theory: The leading
quantum corrections, Phys. Rev. D 50, 3874 (1994), [arXiv:gr-qc/9405057].
5. J. F. Donoghue, Introduction to the effective field theory description of gravity,
Lectures given at the Advanced School on Effective Theories, Almuñécar, Spain
(1995), [arXiv:gr-qc/9512024].
6. S. Scherer and M. R. Schindler, A Primer for Chiral Perturbation Theory,
Springer, Berlin (2012).
7. K. G. Wilson and J. B. Kogut, The renormalization group and the ϵ expansion,
Phys. Rept. 12, 75 (1974).
8. K. G. Wilson, The renormalization group: Critical phenomena and the Kondo
problem, Rev. Mod. Phys. 47, 773 (1975).
9. L. P. Kadanoff, Statistical Physics: Statics, Dynamics and Renormalization, World
Scientific, Singapore (2000).
10. S. Weinberg, Ultraviolet divergences in quantum theories of gravitation, in:
Hawking, S. W. and Israel, W. (Eds.), General Relativity: An Einstein Centenary
Survey, pp. 790–831, Cambridge University Press, Cambridge (1979).
11. K. Gawȩdzki and A. Kupiainen, Renormalizing the nonrenormalizable, Phys. Rev.
Lett. 55, 363 (1985).
12. K. Gawȩdzki and A. Kupiainen, Exact renormalization for the Gross–Neveu model
of quantum fields, Phys. Rev. Lett. 54, 2191 (1985).
13. K. Gawȩdzki and A. Kupiainen, Gross–Neveu model through convergent pertur-
bation expansions, Commun. Math. Phys. 102, 1 (1985).
14. M. Reuter, Nonperturbative evolution equation for quantum gravity, Phys. Rev.
D 57, 971 (1998), [arXiv:hep-th/9605030].
15. A. Ashtekar and R. P. Geroch, Quantum theory of gravitation, Rept. Prog. Phys.
37, 1211 (1974).
16. C. Kiefer, Quantum Gravity, Third Edition, Oxford University Press, Oxford
(2012).
17. R. Percacci, An Introduction to Covariant Quantum Gravity and Asymptotic
Safety, World Scientific, Singapore (2017).
18. C. Kiefer, Quantum gravity: General introduction and recent developments,
Annalen Phys. 15, 129 (2005), [arXiv:gr-qc/0508120].
19. D. Giulini and C. Kiefer, The canonical approach to quantum gravity: Gen-
eral ideas and geometrodynamics, in: Seiler, E. and Stamatescu, I.-O. (Eds.),
320 References

Approaches to Fundamental Physics: An Assessment of Current Theoretical Ideas,

pp. 131–150, Springer, Berlin (2007), [arXiv:gr-qc/0611141].
20. C. Kiefer, Conceptual problems in quantum gravity and quantum cosmology, ISRN
Math. Phys. 2013, 509316 (2013), [arXiv:1401.3578 [gr-qc]].
21. D. Giulini, Remarks on the notions of general covariance and background inde-
pendence, in: Seiler, E. and Stamatescu, I.-O. (Eds.), Approaches to Fundamental
Physics: An Assessment of Current Theoretical Ideas, pp. 105–120, Springer,
Berlin (2007), [arXiv:gr-qc/0603087].
22. C. Wetterich, Exact evolution equation for the effective potential, Phys. Lett. B
301, 90 (1993).
23. M. Reuter and C. Wetterich, Effective average action for gauge theories and exact
evolution equations, Nucl. Phys. B 417, 181 (1994).
24. M. Reuter and C. Wetterich, Exact evolution equation for scalar electrodynamics,
Nucl. Phys. B 427, 291 (1994).
25. M. Reuter and C. Wetterich, Average action for the Higgs model with abelian
gauge symmetry, Nucl. Phys. B 391, 147 (1993).
26. M. Reuter and C. Wetterich, Running gauge coupling in three dimensions and the
electroweak phase transition, Nucl. Phys. B 408, 91 (1993).
27. M. Reuter, Effective average action of Chern–Simons field theory, Phys. Rev. D
53, 4430 (1996), [arXiv:hep-th/9511128].
28. M. Reuter, Renormalization of the topological charge in Yang–Mills theory, Mod.
Phys. Lett. A 12, 2777 (1997), [arXiv:hep-th/9604124].
29. M. Reuter and C. Wetterich, Gluon condensation in nonperturbative flow
equations, Phys. Rev. D 56, 7893 (1997), [arXiv:hep-th/9708051].
30. M. Reuter and C. Wetterich, Quantum Liouville field theory as solution of a flow
equation, Nucl. Phys. B 506, 483 (1997), [arXiv:hep-th/9605039].
31. J. Berges, N. Tetradis, and C. Wetterich, Non-perturbative renormalization flow
in quantum field theory and statistical physics, Phys. Rept. 363, 223 (2002),
[arXiv:hep-ph/0005122].
32. C. Wetterich, Effective average action in statistical physics and quantum field
theory, Int. J. Mod. Phys. A 16, 1951 (2001), [arXiv:hep-ph/0101178].
33. M. Reuter, Effective average actions and nonperturbative evolution equations,
Lectures given at the 5th Hellenic School and Workshops on Elementary Particle
Physics, Corfu, Greece (1995), [arXiv:hep-th/9602012].
34. J. M. Pawlowski, Aspects of the functional renormalisation group, Annals Phys.
322, 2831 (2007), [arXiv:hep-th/0512261].
35. H. Gies, Introduction to the functional RG and applications to gauge theories, in:
Schwenk, A. and Polonyi, J. (Eds.), Renormalization Group and Effective Field
Theory Approaches to Many-Body Systems, pp. 287–348, Springer, Berlin (2012),
[arXiv:hep-ph/0611146].
36. A. Ashtekar, Lectures on Non-Perturbative Canonical Gravity, World Scientific,
Singapore (1991).
37. A. Ashtekar and J. Lewandowski, Background independent quantum gravity: A
status report, Class. Quant. Grav. 21, R53 (2004), [arXiv:gr-qc/0404018].
38. C. Rovelli, Quantum Gravity, Cambridge University Press, Cambridge (2004).
39. T. Thiemann, Modern Canonical Quantum General Relativity, Cambridge Uni-
versity Press, Cambridge (2007), [arXiv:gr-qc/0110034].
40. H. W. Hamber, Phases of four-dimensional simplicial quantum gravity, Phys. Rev.
D 45, 507 (1992).
 References 321

41. H. W. Hamber, Gravitational scaling dimensions, Phys. Rev. D 61, 124008 (2000),
[arXiv:hep-th/9912246].
42. H. W. Hamber, Quantum gravity on the lattice, Gen. Rel. Grav. 41, 817 (2009),
[arXiv:0901.0964 [gr-qc]].
43. T. Regge and R. M. Williams, Discrete structures in gravity, J. Math. Phys. 41,
3964 (2000), [arXiv:gr-qc/0012035].
44. J. Ambjørn, J. Jurkiewicz, and R. Loll, Emergence of a 4D world from causal
quantum gravity, Phys. Rev. Lett. 93, 131301 (2004), [arXiv:hep-th/0404156].
45. J. Ambjørn, J. Jurkiewicz, and R. Loll, Semiclassical universe from first principles,
Phys. Lett. B 607, 205 (2005), [arXiv:hep-th/0411152].
46. J. Ambjørn, J. Jurkiewicz, and R. Loll, The spectral dimension of the universe is
scale dependent, Phys. Rev. Lett. 95, 171301 (2005), [arXiv:hep-th/0505113].
47. J. Ambjørn, J. Jurkiewicz, and R. Loll, Reconstructing the universe, Phys. Rev.
D 72, 064014 (2005), [arXiv:hep-th/0505154].
48. J. Ambjørn, J. Jurkiewicz, and R. Loll, The universe from scratch, Contemp. Phys.
47, 103 (2006), [arXiv:hep-th/0509010].
49. A. Ashtekar, Some surprising implications of background independence in canon-
ical quantum gravity, Gen. Rel. Grav. 41, 1927 (2009), [arXiv:0904.0184 [gr-qc]].
50. B. S. DeWitt, Dynamical Theory of Groups and Fields, Gordon and Breach, New
York (1965).
51. B. S. DeWitt, The Global Approach to Quantum Field Theory. Vols. 1–2,
Clarendon Press, Oxford (2003).
52. A. Connes, Non-commutative differential geometry, Publications Mathématiques
de l’IHÉS 62, 41 (1985).
53. A. Connes, Noncommutative Geometry, Academic Press, San Diego, (1994).
54. R. G. Littlejohn, The semiclassical evolution of wave packets, Phys. Rept. 138,
193 (1986).
55. E. Gozzi and M. Reuter, Quantum deformed canonical transformations, W∞ -
algebras and unitary transformations, Int. J. Mod. Phys. A 9, 5801 (1994),
[arXiv:hep-th/0306221].
56. A. M. Perelomov, Generalized Coherent States and Their Applications, Springer,
Berlin (1986).
57. L. Freidel and E. R. Livine, 3D quantum gravity and effective noncommutative
quantum field theory, Phys. Rev. Lett. 96, 221301 (2006), [arXiv:hep-th/0512113].
58. S. Doplicher, K. Fredenhagen, and J. E. Roberts, The quantum structure of space-
time at the Planck scale and quantum fields, Commun. Math. Phys. 172, 187
(1995), [arXiv:hep-th/0303037].
59. R. J. Rivers, Path Integral Methods in Quantum Field Theory, Cambridge
University Press, Cambridge (1987).
60. J. Glimm and A. Jaffe, Quantum Physics: A Functional Integral Point of View,
Springer, New York (1987).
61. E. Gozzi, E. Cattaruzza, and C. Pagani, Path Integrals for Pedestrians, World
Scientific, Singapore (2016).
62. W. Dittrich and M. Reuter, Classical and Quantum Dynamics: From Classical
Paths to Path Integrals, Fifth Edition, Springer, Berlin (2017).
63. D. F. Litim, Optimisation of the exact renormalisation group, Phys. Lett. B 486,
92 (2000), [arXiv:hep-th/0005245].
64. D. F. Litim, Optimized renormalization group flows, Phys. Rev. D 64, 105007
(2001), [arXiv:hep-th/0103195].
322 References

65. D. F. Litim, Mind the gap, Int. J. Mod. Phys. A 16, 2081 (2001), [arXiv:hep-
th/0104221].
66. A. Wipf, Statistical Approach to Quantum Field Theory, Springer, Berlin (2013).
67. A. Ringwald and C. Wetterich, Average action for the N -component φ4 theory,
Nucl. Phys. B 334, 506 (1990).
68. C. Wetterich, Average action and the renormalization group equations, Nucl.
Phys. B 352, 529 (1991).
69. L. O’Raifeartaigh, A. Wipf, and H. Yoneyama, The constraint effective potential,
Nucl. Phys. B 271, 653 (1986).
70. C. Wetterich, The average action for scalar fields near phase transitions, Z. Phys.
C 57, 451 (1993).
71. N. Tetradis and C. Wetterich, The high temperature phase transition for ϕ4
theories, Nucl. Phys. B 398, 659 (1993).
72. M. Reuter, N. Tetradis, and C. Wetterich, The large-N limit and the high-
temperature phase transition for the ϕ4 theory, Nucl. Phys. B 401, 567 (1993),
[arXiv:hep-ph/9301271].
73. C. Wetterich, Quadratic renormalization of the average potential and the nat-
uralness of quadratic mass relations for the top quark, Z. Phys. C 48, 693
(1990).
74. A. Bonanno, Coarse graining and renormalization group in the Einstein universe,
Phys. Rev. D 52, 969 (1995), [arXiv:gr-qc/9505051].
75. R. Floreanini and R. Percacci, Average effective potential for the conformal factor,
Nucl. Phys. B 436, 141 (1995), [arXiv:hep-th/9305172].
76. P. Ramond, Field Theory: A Modern Primer, Second Edition, Westview Press,
Boulder (1997).
77. S. R. Coleman and E. J. Weinberg, Radiative corrections as the origin of
spontaneous symmetry breaking, Phys. Rev. D 7, 1888 (1973).
78. M. Le Bellac, Quantum and Statistical Field Theory, Clarendon Press, Oxford
(1992).
79. A. Codello, M. Demmel, and O. Zanusso, Scheme dependence and universal-
ity in the functional renormalization group, Phys. Rev. D 90, 027701 (2014),
[arXiv:1310.7625 [hep-th]].
80. M. Reuter and H. Weyer, Background independence and asymptotic safety in
conformally reduced gravity, Phys. Rev. D 79, 105005 (2009), [arXiv:0801.3287
[hep-th]].
81. J. Borchardt and B. Knorr, Global solutions of functional fixed point equations
via pseudospectral methods, Phys. Rev. D 91, 105011 (2015), [Erratum: Phys.
Rev. D 93, 089904 (2016)], [arXiv:1502.07511 [hep-th]].
82. J. Borchardt and B. Knorr, Solving functional flow equations with pseudo-spectral
methods, Phys. Rev. D 94, 025027 (2016), [arXiv:1603.06726 [hep-th]].
83. J. C. Collins, Renormalization: An Introduction to Renormalization, the Renor-
malization Group and the Operator–Product Expansion, Cambridge University
Press, Cambridge (1984).
84. R. Coquereaux, Renormalization schemes in QED, Annals Phys. 125, 401 (1980).
85. M. Reuter and H. Weyer, Renormalization group improved gravitational actions: A
Brans–Dicke approach, Phys. Rev. D 69, 104022 (2004), [arXiv:hep-th/0311196].
86. L. F. Abbott, Introduction to the background field method, Acta Phys. Polon. B
13, 33 (1982).
87. M. Böhm, A. Denner, and H. Joos, Gauge Theories of the Strong and Electroweak
Interaction, B. G. Teubner, Stuttgart (2001).
 References 323

88. W. Dittrich and M. Reuter, Selected Topics in Gauge Theories, Springer, Berlin
(1986).
89. G. ’t Hooft, The background field method in gauge field theories, in: Lopuszanski,
J. and Jancewicz, B. (Eds.), Functional and Probabilistic Methods in Quantum
Field Theory, Vol. 1, 12th Karpacz Winter School of Theoretical Physics, pp. 345–
369, Acta Universitatis Wratislaviensis No. 38, Wroclaw (1975).
90. E. Manrique and M. Reuter, Bimetric truncations for quantum Einstein gravity
and asymptotic safety, Annals Phys. 325, 785 (2010), [arXiv:0907.2617 [gr-qc]].
91. L. F. Abbott, The background field method beyond one loop, Nucl. Phys. B 185,
189 (1981).
92. D. V. Anosov, S. K. Aranson, V. I. Arnold, et al., Dynamical Systems I: Ordi-
nary Differential Equations and Smooth Dynamical Systems, Springer, Berlin
(1988).
93. V. I. Arnold and A. Avez, Ergodic Problems of Classical Mechanics, W. A.
Benjamin, New York (1968).
94. P. Castiglione, M. Falcioni, A. Lesne, and A. Vulpiani, Chaos and Coarse Graining
in Statistical Mechanics, Cambridge University Press, Cambridge (2008).
95. R. Abraham and J. E. Marsden, Foundations of Mechanics, Second Edition,
Addison-Wesley, Redwood City (1987).
96. L. Perko, Differential Equations and Dynamical Systems, Third Edition, Springer,
New York (2001).
97. F. J. Wegner, Some invariance properties of the renormalization group, J. Phys.
C: Solid State Phys. 7, 2098 (1974).
98. M. Niedermaier and M. Reuter, The asymptotic safety scenario in quantum
gravity, Living Rev. Rel. 9, 5 (2006).
99. K. Halpern and K. Huang, Nontrivial directions for scalar fields, Phys. Rev. D 53,
3252 (1996), [arXiv:hep-th/9510240].
100. A. Bonanno, Nonperturbative scaling in the scalar theory, Phys. Rev. D 62, 027701
(2000), [arXiv:hep-th/0001060].
101. H. Gies, Flow equation for Halpern–Huang directions of scalar O(N ) models, Phys.
Rev. D 63, 065011 (2001), [arXiv:hep-th/0009041].
102. J. F. Donoghue, The quantum theory of general relativity at low energies, Helv.
Phys. Acta 69, 269 (1996), [arXiv:gr-qc/9607039].
103. J. F. Donoghue and B. R. Holstein, Low energy theorems of quantum gravity
from effective field theory, J. Phys. G: Nucl. Part. Phys. 42, 103102 (2015),
[arXiv:1506.00946 [gr-qc]].
104. R. Gastmans, R. Kallosh, and C. Truffin, Quantum gravity near two dimensions,
Nucl. Phys. B 133, 417 (1978).
105. S. M. Christensen and M. J. Duff, Quantum gravity in 2 + ε dimensions, Phys.
Lett. B 79, 213 (1978).
106. H.-S. Tsao, Conformal anomalies in a general background metric, Phys. Lett. B
68, 79 (1977).
107. L. S. Brown, Stress-tensor trace anomaly in a gravitational metric: Scalar fields,
Phys. Rev. D 15, 1469 (1977).
108. H. Kawai and M. Ninomiya, Renormalization group and quantum gravity, Nucl.
Phys. B 336, 115 (1990).
109. I. Jack and D. R. T. Jones, The ϵ-expansion of two-dimensional quantum gravity,
Nucl. Phys. B 358, 695 (1991).
110. H. Kawai, Y. Kitazawa, and M. Ninomiya, Scaling exponents in quantum gravity
near two dimensions, Nucl. Phys. B 393, 280 (1993), [arxiv:hep-th/9206081].
324 References

111. H. Kawai, Y. Kitazawa, and M. Ninomiya, Ultraviolet stable fixed point and
scaling relations in (2 + ϵ)-dimensional quantum gravity, Nucl. Phys. B 404, 684
(1993), [arxiv:hep-th/9303123].
112. H. Kawai, Y. Kitazawa, and M. Ninomiya, Quantum gravity in 2 + ϵ dimensions,
Prog. Theor. Phys. Suppl. 114, 149 (1993).
113. T. Aida, Y. Kitazawa, H. Kawai, and M. Ninomiya, Conformal invariance and
renormalization group in quantum gravity near two dimensions, Nucl. Phys. B
427, 158 (1994), [arxiv:hep-th/9404171].
114. J. Nishimura, S. Tamura, and A. Tsuchiya, R2 gravity in (2 + ϵ)-dimensional
quantum gravity, Mod. Phys. Lett. A 9, 3565 (1994), [arxiv:hep-th/9405059].
115. H. Kawai, Y. Kitazawa, and M. Ninomiya, Renormalizability of quantum gravity
near two dimensions, Nucl. Phys. B 467, 313 (1996), [arxiv:hep-th/9511217].
116. T. Aida and Y. Kitazawa, Two-loop prediction for scaling exponents in (2 + ϵ)-
dimensional quantum gravity, Nucl. Phys. B 491, 427 (1997), [arxiv:hep-th/
9609077].
117. A. Nink and M. Reuter, The unitary conformal field theory behind 2D asymptotic
safety, J. High Energy Phys. 1602, 167 (2016), [arXiv:1512.06805 [hep-th]].
118. M. Niedermaier, Can a nontrivial gravitational fixed point be identified in
perturbation theory?, PoS CLAQG 08, 005 (2008).
119. M. R. Niedermaier, Gravitational fixed points from perturbation theory, Phys.
Rev. Lett. 103, 101303 (2009).
120. M. Niedermaier, Gravitational fixed points and asymptotic safety from perturba-
tion theory, Nucl. Phys. B 833, 226 (2010).
121. P. Forgács and M. Niedermaier, A fixed point for truncated quantum Einstein
gravity, [arXiv:hep-th/0207028].
122. M. Niedermaier, Renormalization and asymptotic safety in truncated quan-
tum Einstein gravity, J. High Energy Phys. 0212, 066 (2002), [arXiv:hep-th/
0207143].
123. M. Niedermaier, Dimensionally reduced gravity theories are asymptotically safe,
Nucl. Phys. B 673, 131 (2003), [arXiv:hep-th/0304117].
124. M. Niedermaier, The asymptotic safety scenario in quantum gravity: An intro-
duction, Class. Quant. Grav. 24, R171 (2007), [arXiv:gr-qc/0610018].
125. E. Witten, 2 + 1 dimensional gravity as an exactly soluble system, Nucl. Phys. B
311, 46 (1988).
126. B. S. DeWitt, Quantum theory of gravity. I. The canonical theory, Phys. Rev.
160, 1113 (1967).
127. H. W. Hamber, Quantum Gravitation: The Feynman Path Integral Approach,
Springer, Berlin (2009).
128. L. H. Ryder, Quantum Field Theory, Second Edition, Cambridge University Press,
Cambridge (1996).
129. B. Dittrich, The continuum limit of loop quantum gravity: A framework for solving
the theory, in: Ashtekar, A. and Pullin, J. (Eds.), Loop Quantum Gravity: The
First 30 Years, pp. 153–179, World Scientific, Singapore (2017), [arXiv:1409.1450
[gr-qc]].
130. B. Dittrich and S. Steinhaus, Time evolution as refining, coarse graining and
entangling, New J. Phys. 16, 123041 (2014), [arXiv:1311.7565 [gr-qc]].
131. B. Dittrich, From the discrete to the continuous: Towards a cylindrically consistent
dynamics, New J. Phys. 14, 123004 (2012), [arXiv:1205.6127 [gr-qc]].
132. T. T. Wu and C. N. Yang, Some remarks about unquantized non-abelian gauge
fields, Phys. Rev. D 12, 3843 (1975).
 References 325

133. B. Dittrich, S. Mizera, and S. Steinhaus, Decorated tensor network renormalization

for lattice gauge theories and spin foam models, New J. Phys. 18, 053009 (2016),
[arXiv:1409.2407 [gr-qc]].
134. B. Dittrich, M. Martin-Benito, and S. Steinhaus, Quantum group spin nets:
Refinement limit and relation to spin foams, Phys. Rev. D 90, 024058 (2014),
[arXiv:1312.0905 [gr-qc]].
135. E. Mottola, Functional integration over geometries, J. Math. Phys. 36, 2470
(1995), [arXiv:hep-th/9502109].
136. S. Kobayashi and K. Nomizu, Foundations of Differential Geometry. Vols. I–II,
John Wiley & Sons, New York (1996).
137. T. Padmanabhan, Gravitation: Foundations and Frontiers, Cambridge University
Press, Cambridge (2010).
138. M. Reuter, Quantum mechanics as a gauge theory of metaplectic spinor fields,
Int. J. Mod. Phys. A 13, 3835 (1998), [arXiv:hep-th/9804036].
139. J. Honerkamp, Chiral multi-loops, Nucl. Phys. B 36, 130 (1972).
140. D. H. Friedan, Nonlinear models in 2 + ε dimensions, Annals Phys. 163, 318
(1985).
141. P. S. Howe, G. Papadopoulos, and K. S. Stelle, The background field method and
the non-linear σ-model, Nucl. Phys. B 296, 26 (1988).
142. S. P. Sorella, Algebraic characterization of the topological σ-model, Phys. Lett. B
228, 69 (1989).
143. B. S. DeWitt, Quantum theory of gravity. II. The manifestly covariant theory,
Phys. Rev. 162, 1195 (1967).
144. B. S. DeWitt, The effective action, in: Batalin, I. A., Islam, C. J., and Vilkovisky,
G. A. (Eds.), Quantum Field Theory and Quantum Statistics, Essays in Honour
of the Sixtieth Birthday of E. S. Fradkin, Vol. 1, pp. 191–222, Adam Hilger, Bristol
(1987).
145. A. Nink, Field parametrization dependence in asymptotically safe quantum
gravity, Phys. Rev. D 91, 044030 (2015), [arXiv:1410.7816 [hep-th]].
146. M. Demmel and A. Nink, Connections and geodesics in the space of metrics, Phys.
Rev. D 92, 104013 (2015), [arXiv:1506.03809 [gr-qc]].
147. H. Gies, B. Knorr, and S. Lippoldt, Generalized parametrization dependence in
quantum gravity, Phys. Rev. D 92, 084020 (2015), [arXiv:1507.08859 [hep-th]].
148. N. Ohta, R. Percacci, and A. D. Pereira, Gauges and functional measures in
quantum gravity I: Einstein theory, JHEP 1606, 115 (2016), [arXiv:1605.00454
[hep-th]].
149. N. Ohta, R. Percacci, and A. D. Pereira, Gauges and functional measures in
quantum gravity II: Higher-derivative gravity, Eur. Phys. J. C 77, 611 (2017),
[arXiv:1610.07991 [hep-th]].
150. M. T. Grisaru, P. van Nieuwenhuizen, and C. C. Wu, Background-field method
versus normal field theory in explicit examples: One-loop divergences in S matrix
and Green’s functions for Yang–Mills and gravitational fields, Phys. Rev. D 12,
3203 (1975).
151. D. M. Capper, J. J. Dulwich, and M. Ramón Medrano, The background field
method for quantum gravity at two loops, Nucl. Phys. B 254, 737 (1985).
152. S. L. Adler, Einstein gravity as a symmetry-breaking effect in quantum field
theory, Rev. Mod. Phys. 54, 729 (1982), [Erratum: Rev. Mod. Phys. 55, 837
(1983)].
153. N. Nakanishi and I. Ojima, Covariant Operator Formalism of Gauge Theories and
Quantum Gravity, World Scientific, Singapore (1990).
326 References

154. D. Becker and M. Reuter, Towards a C-function in 4D quantum gravity, JHEP

1503, 065 (2015), [arXiv:1412.0468 [hep-th]].
155. B. Knorr and S. Lippoldt, Correlation functions on a curved background, Phys.
Rev. D 96, 065020 (2017), [arXiv:1707.01397 [hep-th]].
156. S. Weinberg, Gravitation and Cosmology: Principles and Applications of the
General Theory of Relativity, John Wiley & Sons, New York (1972).
157. A. Bonanno and M. Reuter, Renormalization group improved black hole space-
times, Phys. Rev. D 62, 043008 (2000), [arXiv:hep-th/0002196].
158. N. Straumann, The mystery of the cosmic vacuum energy density and the
accelerated expansion of the universe, Eur. J. Phys. 20, 419 (1999), [arXiv:astro-
ph/9908342].
159. S. Weinberg, The cosmological constant problem, Rev. Mod. Phys. 61, 1 (1989).
160. M. Reuter and F. Saueressig, Renormalization group flow of quantum gravity
in the Einstein–Hilbert truncation, Phys. Rev. D 65, 065016 (2002), [arXiv:hep-
th/0110054].
161. S. A. Fulling, R. C. King, B. G. Wybourne, and C. J. Cummins, Normal forms for
tensor polynomials: I. The Riemann tensor, Class. Quant. Grav. 9, 1151 (1992).
162. O. Lauscher and M. Reuter, Is quantum Einstein gravity nonperturbatively
renormalizable?, Class. Quant. Grav. 19, 483 (2002), [arXiv:hep-th/0110021].
163. O. Lauscher and M. Reuter, Flow equation of quantum Einstein gravity in
a higher-derivative truncation, Phys. Rev. D 66, 025026 (2002), [arXiv:hep-
th/0205062].
164. A. Codello and R. Percacci, Fixed points of higher-derivative gravity, Phys. Rev.
Lett. 97, 221301 (2006), [arXiv:hep-th/0607128].
165. K. Groh, S. Rechenberger, F. Saueressig, and O. Zanusso, Higher derivative grav-
ity from the universal renormalization group machine, PoS EPS-HEP 2011, 124
(2011), [arXiv:1111.1743 [hep-th]].
166. N. Ohta and R. Percacci, Higher derivative gravity and asymptotic safety in
diverse dimensions, Class. Quant. Grav. 31, 015024 (2014), [arXiv:1308.3398
[hep-th]].
167. N. Ohta and R. Percacci, Ultraviolet fixed points in conformal gravity and general
quadratic theories, Class. Quant. Grav. 33, 035001 (2016), [arXiv:1506.05526 [hep-
th]].
168. D. Benedetti, P. F. Machado, and F. Saueressig, Asymptotic safety in higher-
derivative gravity, Mod. Phys. Lett. A 24, 2233 (2009), [arXiv:0901.2984 [hep-th]].
169. D. Benedetti, P. F. Machado, and F. Saueressig, Taming perturbative divergences
in asymptotically safe gravity, Nucl. Phys. B 824, 168 (2010), [arXiv:0902.4630
[hep-th]].
170. D. Benedetti, P. F. Machado, and F. Saueressig, Four-derivative interactions in
asymptotically safe gravity, AIP Conf. Proc. 1196, 44 (2009), [arXiv:0909.3265
[hep-th]].
171. D. F. Litim and J. M. Pawlowski, Flow equations for Yang–Mills theories in general
axial gauges, Phys. Lett. B 435, 181 (1998), [arXiv:hep-th/9802064].
172. S. Falkenberg and S. D. Odintsov, Gauge dependence of the effective average action
in Einstein gravity, Int. J. Mod. Phys. A 13, 607 (1998), [arXiv:hep-th/9612019].
173. O. Lauscher and M. Reuter, Ultraviolet fixed point and generalized flow equation
of quantum gravity, Phys. Rev. D 65, 025013 (2001), [arXiv:hep-th/0108040].
174. K. Groh and F. Saueressig, Ghost wavefunction renormalization in asymp-
totically safe quantum gravity, J. Phys. A: Math. Theor. 43, 365403 (2010),
[arXiv:1001.5032 [hep-th]].
 References 327

175. A. Eichhorn and H. Gies, Ghost anomalous dimension in asymptotically safe

quantum gravity, Phys. Rev. D 81, 104010 (2010), [arXiv:1001.5033 [hep-th]].
176. A. Eichhorn, H. Gies, and M. M. Scherer, Asymptotically free scalar curvature–
ghost coupling in quantum Einstein gravity, Phys. Rev. D 80, 104003 (2009),
[arXiv:0907.1828 [hep-th]].
177. W. Souma, Non-trivial ultraviolet fixed point in quantum gravity, Prog. Theor.
Phys. 102, 181 (1999), [arXiv:hep-th/9907027].
178. M. Reuter and F. Saueressig, A class of nonlocal truncations in quantum Einstein
gravity and its renormalization group behavior, Phys. Rev. D 66, 125001 (2002),
[arXiv:hep-th/0206145].
179. M. Reuter and F. Saueressig, Nonlocal quantum gravity and the size of the
universe, Fortsch. Phys. 52, 650 (2004), [arXiv:hep-th/0311056].
180. D. F. Litim, Fixed points of quantum gravity, Phys. Rev. Lett. 92, 201301 (2004),
[arXiv:hep-th/0312114].
181. G. Narain and R. Percacci, On the scheme dependence of gravitational beta
functions, Acta Phys. Polon. B 40, 3439 (2009), [arXiv:0910.5390 [hep-th]].
182. D. Becker and M. Reuter, Running boundary actions, asymptotic safety, and black
hole thermodynamics, JHEP 1207, 172 (2012), [arXiv:1205.3583 [hep-th]].
183. A. Nink and M. Reuter, On the physical mechanism underlying asymptotic safety,
JHEP 1301, 062 (2013), [arXiv:1208.0031 [hep-th]].
184. P. Fischer and D. F. Litim, Fixed points of quantum gravity in extra dimensions,
Phys. Lett. B 638, 497 (2006), [arXiv:hep-th/0602203].
185. P. F. Machado and F. Saueressig, Renormalization group flow of f (R) gravity,
Phys. Rev. D 77, 124045 (2008), [arXiv:0712.0445 [hep-th]].
186. A. Codello, R. Percacci, and C. Rahmede, Ultraviolet properties of f (R)-gravity,
Int. J. Mod. Phys. A 23, 143 (2008), [arXiv:0705.1769 [hep-th]].
187. A. Codello, R. Percacci, and C. Rahmede, Investigating the ultraviolet properties
of gravity with a Wilsonian renormalization group equation, Annals Phys. 324,
414 (2009), [arXiv:0805.2909].
188. K. Falls, D. F. Litim, K. Nikolakopoulos, and C. Rahmede, Further evidence
for asymptotic safety of quantum gravity, Phys. Rev. D 93, 104022 (2016),
[arXiv:1410.4815 [hep-th]].
189. E. Manrique, M. Reuter, and F. Saueressig, Matter induced bimetric actions for
gravity, Annals Phys. 326, 440 (2011), [arXiv:1003.5129 [hep-th]].
190. E. Manrique, M. Reuter, and F. Saueressig, Bimetric renormalization group flows
in quantum Einstein gravity, Annals Phys. 326, 463 (2011), [arXiv:1006.0099
[hep-th]].
191. D. Becker and M. Reuter, En route to background independence: Broken split-
symmetry, and how to restore it with bi-metric average actions, Annals Phys. 350,
225 (2014), [arXiv:1404.4537 [hep-th]].
192. D. Becker and M. Reuter, Propagating gravitons vs. ‘dark matter’ in asymptoti-
cally safe quantum gravity, JHEP 1412, 025 (2014), [arXiv:1407.5848 [hep-th]].
193. N. Christiansen, D. F. Litim, J. M. Pawlowski, and A. Rodigast, Fixed points
and infrared completion of quantum gravity, Phys. Lett. B 728, 114 (2014),
[arXiv:1209.4038 [hep-th]].
194. N. Christiansen, B. Knorr, J. M. Pawlowski, and A. Rodigast, Global flows in
quantum gravity, Phys. Rev. D 93, 044036 (2016), [arXiv:1403.1232 [hep-th]].
195. N. Christiansen, B. Knorr, J. Meibohm, J. M. Pawlowski, and M. Reichert,
Local quantum gravity, Phys. Rev. D 92, 121501(R) (2015), [arXiv:1506.07016
[hep-th]].
328 References

196. N. Christiansen, Four-derivative quantum gravity beyond perturbation theory,

[arXiv:1612.06223 [hep-th]].
197. T. Denz, J. M. Pawlowski, and M. Reichert, Towards apparent convergence
in asymptotically safe quantum gravity, Eur. Phys. J. C 78, 336 (2018),
[arXiv:1612.07315 [hep-th]].
198. B. Knorr, Infinite order quantum-gravitational correlations, Class. Quant. Grav.
35, 115005 (2018), [arXiv:1710.07055 [hep-th]].
199. Y. Décanini and A. Folacci, FKWC-bases and geometrical identities for classical
and quantum field theories in curved spacetime, [arXiv:0805.1595 [gr-qc]].
200. P. Labus, T. R. Morris, and Z. H. Slade, Background independence in a back-
ground dependent renormalization group, Phys. Rev. D 94, 024007 (2016),
[arXiv:1603.04772 [hep-th]].
201. R. Percacci and G. P. Vacca, The background scale Ward identity in quantum
gravity, Eur. Phys. J. C 77, 52 (2017), [arXiv:1611.07005 [hep-th]].
202. A. Codello, G. D’Odorico, and C. Pagani, Consistent closure of renormalization
group flow equations in quantum gravity, Phys. Rev. D 89, 081701(R) (2014),
[arXiv:1304.4777 [gr-qc]].
203. M. Reuter and H. Weyer, Conformal sector of quantum Einstein gravity in the
local potential approximation: Non-Gaussian fixed point and a phase of unbro-
ken diffeomorphism invariance, Phys. Rev. D 80, 025001 (2009), [arXiv:0804.1475
[hep-th]].
204. M. Reuter and H. Weyer, The role of background independence for asymp-
totic safety in Quantum Einstein Gravity, Gen. Rel. Grav. 41, 983 (2009),
[arXiv:0903.2971 [hep-th]].
205. E. Manrique and M. Reuter, Bare action and regularized functional inte-
gral of asymptotically safe quantum gravity, Phys. Rev. D 79, 025008 (2009),
[arXiv:0811.3888 [hep-th]].
206. E. Manrique and M. Reuter, Bare vs. effective fixed point action in asymptotic
safety: The reconstruction problem, PoS CLAQG 08, 001 (2011), [arXiv:0905.4220
[hep-th]].
207. J.-E. Daum and M. Reuter, Effective potential of the conformal factor: Gravita-
tional average action and dynamical triangulations, Adv. Sci. Lett. 2, 255 (2009),
[arXiv:0806.3907 [hep-th]].
208. J.-E. Daum and M. Reuter, The effective potential of the conformal factor
in quantum Einstein gravity, PoS CLAQG 08, 013 (2011), [arXiv:0910.5401
[hep-th]].
209. R. Jackiw, C. Núñez, and S.-Y. Pi, Quantum relaxation of the cosmological
constant, Phys. Lett. A 347, 47 (2005), [arxiv:hep-th/0502215].
210. R. J. Riegert, A non-local action for the trace anomaly, Phys. Lett. B 134, 56
(1984).
211. I. Antoniadis and E. Mottola, Four-dimensional quantum gravity in the conformal
sector, Phys. Rev. D 45, 2013 (1992).
212. I. Antoniadis, P. O. Mazur, and E. Mottola, Conformal symmetry and central
charges in 4 dimensions, Nucl. Phys. B 388, 627 (1992), [arXiv:hep-th/9205015].
213. I. Antoniadis, P. O. Mazur, and E. Mottola, Scaling behavior of quantum
four-geometries, Phys. Lett. B 323, 284 (1994), [arXiv:hep-th/9301002].
214. I. Antoniadis, P. O. Mazur, and E. Mottola, Quantum diffeomorphisms and
conformal symmetry, Phys. Rev. D 55, 4756 (1997), [arXiv:hep-th/9509168].
215. I. Antoniadis, P. O. Mazur, and E. Mottola, Physical states of the quantum
conformal factor, Phys. Rev. D 55, 4770 (1997), [arXiv:hep-th/9509169].
 References 329

216. I. Antoniadis, P. O. Mazur, and E. Mottola, Criticality and scaling in 4D quantum

gravity, Phys. Lett. B 394, 49 (1997), [arXiv:hep-th/9611145].
217. P. O. Mazur and E. Mottola, Weyl cohomology and the effective action for
conformal anomalies, Phys. Rev. D 64, 104022 (2001) [arXiv:hep-th/0106151].
218. I. Antoniadis, P. O. Mazur, and E. Mottola, Conformal invariance and cosmic
background radiation, Phys. Rev. Lett. 79, 14 (1997), [arXiv:astro-ph/9611208].
219. I. Antoniadis, P. O. Mazur, and E. Mottola, Fractal geometry of quantum
spacetime at large scales, Phys. Lett. B 444, 284 (1998), [arXiv:hep-th/9808070].
220. I. Antoniadis, P. O. Mazur, and E. Mottola, Cosmological dark energy: Prospects
for a dynamical theory, New J. Phys. 9, 11 (2007), [arXiv:gr-qc/0612068].
221. E. Mottola and R. Vaulin, Macroscopic effects of the quantum trace anomaly,
Phys. Rev. D 74, 064004 (2006), [arXiv:gr-qc/0604051].
222. O. Lauscher, M. Reuter, and C. Wetterich, Rotation symmetry breaking conden-
sate in a scalar theory, Phys. Rev. D 62, 125021 (2000), [arXiv:hep-th/0006099].
223. K. Symanzik, A field theory with computable large-momenta behaviour, Lett.
Nuovo Cim. 6, 77 (1973).
224. G. ’t Hooft, The birth of asymptotic freedom, Nucl. Phys. B 254, 11 (1985).
225. P. O. Mazur and E. Mottola, The path integral measure, conformal factor
problem and stability of the ground state of quantum gravity, Nucl. Phys. B 341,
187 (1990).
226. A. Dasgupta and R. Loll, A proper-time cure for the conformal sickness in
quantum gravity, Nucl. Phys. B 606, 357 (2001), [arXiv:hep-th/0103186].
227. J. Ambjørn, A. Dasgupta, J. Jurkiewicz, and R. Loll, A Lorentzian cure
for Euclidean troubles, Nucl. Phys. B Proc. Suppl. 106, 977 (2002),
[arXiv:hep-th/0201104].
228. J. Ambjørn, A. Görlich, J. Jurkiewicz, and R. Loll, Nonperturbative quantum de
Sitter universe, Phys. Rev. D 78, 063544 (2008), [arXiv:0807.4481 [hep-th]].
229. C. Pagani and M. Reuter, Composite operators in asymptotic safety, Phys. Rev.
D 95, 066002 (2017), [arXiv:1611.06522 [gr-qc]].
230. M. Reuter and J.-M. Schwindt, A minimal length from the cutoff modes in asymp-
totically safe quantum gravity, JHEP 0601, 070 (2006), [arXiv:hep-th/0511021].
231. M. Reuter and J.-M. Schwindt, Scale-dependent metric and causal structures in
quantum Einstein gravity, JHEP 0701, 049 (2007), [arXiv:hep-th/0611294].
232. P. F. Machado and R. Percacci, Conformally reduced quantum gravity revisited,
Phys. Rev. D 80, 024020 (2009), [arXiv:0904.2510 [hep-th]].
233. M. A. Rubin and C. R. Ordóñez, Eigenvalues and degeneracies for n-dimensional
tensor spherical harmonics, J. Math. Phys. 25, 2888 (1984).
234. M. A. Rubin and C. R. Ordóñez, Symmetric-tensor eigenspectrum of the
Laplacian on n-spheres, J. Math. Phys. 26, 65 (1985).
235. K. Pilch and A. N. Schellekens, Formulas for the eigenvalues of the Laplacian on
tensor harmonics on symmetric coset spaces, J. Math. Phys. 25, 3455 (1984).
236. R. Floreanini and R. Percacci, Mean-field quantum gravity, Phys. Rev. D 46,
1566 (1992).
237. R. Floreanini and R. Percacci, Renormalization-group flow of the dilaton
potential, Phys. Rev. D 52, 896 (1995), [arXiv:hep-th/9412181].
238. J. A. Dietz and T. R. Morris, Background independent exact renormalization
group for conformally reduced gravity, JHEP 1504, 118 (2015), [arXiv:1502.07396
[hep-th]].
239. R. Percacci, Renormalization group flow of Weyl invariant dilaton gravity, New
J. Phys. 13, 125013 (2011), [arXiv:1110.6758 [hep-th]].
330 References

240. A. Codello, G. D’Odorico, C. Pagani, and R. Percacci, The renormalization group

and Weyl invariance, Class. Quant. Grav. 30, 115015 (2013), [arXiv:1210.3284
[hep-th]].
241. C. Pagani and R. Percacci, Quantization and fixed points of non-integrable Weyl
theory, Class. Quant. Grav. 31, 115005 (2014), [arXiv:1312.7767 [hep-th]].
242. C. Kiefer, Quantum geometrodynamics: Whence, whither?, Gen. Rel. Grav. 41,
877 (2009), [arXiv:0812.0295 [gr-qc]].
243. C. Pagani, Note on scaling arguments in the effective average action formalism,
Phys. Rev. D 94, 045001 (2016), [arXiv:1603.07250 [hep-th]].
244. J. Ambjørn, A. Görlich, J. Jurkiewicz, and R. Loll, Nonperturbative quantum
gravity, Phys. Rept. 519, 127 (2012), [arXiv:1203.3591 [hep-th]].
245. M. Ostrogradsky, Mémoires sur les équations différentielles, relatives au problème
des isopérimètres, Mem. Acad. St. Petersbourg VI (4), 385 (1850).
246. R. P. Woodard, Ostrogradsky’s theorem on Hamiltonian instability, Scholarpedia
10 (8), 32243 (2015), [arXiv:1506.02210 [hep-th]].
247. M. Blagojević and F. W. Hehl (Eds.), Gauge Theories of Gravitation: A Reader
with Commentaries, Imperial College Press, London (2013).
248. T. Ortı́n, Gravity and Strings, Cambridge University Press, Cambridge (2004).
249. P. Peldán, Actions for gravity, with generalizations: A review, Class. Quant.
Grav. 11, 1087 (1994), [arXiv:gr-qc/9305011].
250. F. W. Hehl, J. D. McCrea, E. W. Mielke, and Y. Ne’eman, Metric-affine gauge
theory of gravity: Field equations, Noether identities, world spinors, and breaking
of dilation invariance, Phys. Rept. 258, 1 (1995), [arXiv:gr-qc/9402012].
251. J. F. Plebański, On the separation of Einsteinian substructures, J. Math. Phys.
18, 2511 (1977).
252. R. Capovilla, T. Jacobson, and J. Dell, General relativity without the metric,
Phys. Rev. Lett. 63, 2325 (1989).
253. R. Capovilla, J. Dell, and T. Jacobson, A pure spin-connection formulation of
gravity, Class. Quant. Grav. 8, 59 (1991).
254. K. Krasnov, Gravity as a diffeomorphism-invariant gauge theory, Phys. Rev. D
84, 024034 (2011), [arXiv:1101.4788 [hep-th]].
255. K. Krasnov, Pure connection action principle for general relativity, Phys. Rev.
Lett. 106, 251103 (2011), [arXiv:1103.4498 [gr-qc]].
256. R. L. Arnowitt, S. Deser, and C. W. Misner, The dynamics of general relativity,
in: Witten, L. (Ed.), Gravitation: An Introduction to Current Research, pp.
227–264, John Wiley & Sons, New York (1962), [arXiv:gr-qc/0405109].
257. A. Ashtekar, New Hamiltonian formulation of general relativity, Phys. Rev. D
36, 1587 (1987).
258. J.-E. Daum and M. Reuter, Renormalization group flow of the Holst action,
Phys. Lett. B 710, 215 (2012), [arXiv:1012.4280 [hep-th]].
259. J.-E. Daum and M. Reuter, Running Immirzi parameter and asymptotic safety,
PoS CNCFG 2010, 003 (2010), [arXiv:1111.1000 [hep-th]].
260. J.-E. Daum and M. Reuter, Einstein–Cartan gravity, asymptotic safety, and the
running Immirzi parameter, Annals Phys. 334, 351 (2013), [arXiv:1301.5135
[hep-th]].
261. U. Harst and M. Reuter, A new functional flow equation for Einstein–Cartan
quantum gravity, Annals Phys. 354, 637 (2015), [arXiv:1410.7003 [hep-th]].
262. M. Blagojević, Gravitation and Gauge Symmetries, Institute of Physics
Publishing, Bristol (2002).
 References 331

263. S. Holst, Barbero’s Hamiltonian derived from a generalized Hilbert–Palatini

action, Phys. Rev. D 53, 5966 (1996), [arXiv:gr-qc/9511026].
264. P. Baekler and F. W. Hehl, Beyond Einstein–Cartan gravity: Quadratic torsion
and curvature invariants with even and odd parity including all boundary terms,
Class. Quant. Grav. 28, 215017 (2011), [arXiv:1105.3504 [gr-qc]].
265. G. Immirzi, Real and complex connections for canonical gravity, Class. Quant.
Grav. 14, L177 (1997), [arxiv:gr-qc/9612030].
266. J. F. Barbero G., Real Ashtekar variables for Lorentzian signature space-times,
Phys. Rev. D 51, 5507 (1995), [arXiv:gr-qc/9410014].
267. A. Perez, Spin foam models for quantum gravity, Class. Quant. Grav. 20, R43
(2003), [arXiv:gr-qc/0301113].
268. D. Oriti, The group field theory approach to quantum gravity, in: Oriti, D. (Ed.),
Approaches to Quantum Gravity: Toward a New Understanding of Space, Time
and Matter, pp. 310–331, Cambridge University Press, Cambridge (2009).
269. L. Freidel, Group field theory: An overview, Int. J. Theor. Phys. 44, 1769 (2005),
[arXiv:hep-th/0505016].
270. L. Freidel, D. Minic, and T. Takeuchi, Quantum gravity, torsion, parity violation,
and all that, Phys. Rev. D 72, 104002 (2005), [arXiv:hep-th/0507253].
271. J.-E. Daum, U. Harst, and M. Reuter, Running gauge coupling in asymptotically
safe quantum gravity, JHEP 1001, 084 (2010), [arXiv:0910.4938 [hep-th]].
272. R. Floreanini and R. Percacci, A multiplicative background field method, in:
Koloskov, A. (Ed.), Gravitation Theory and Geometric Methods in Field Theory,
volume in honour of D. Ivanenko’s 90th jubilee, Moscow (1994).
273. A. Bonanno and M. Reuter, Proper time flow equation for gravity, JHEP 0502,
035 (2005), [arXiv:hep-th/0410191].
274. M. Reuter and G. M. Schollmeyer, The metric on field space, functional renor-
malization, and metric-torsion quantum gravity, Annals Phys. 367, 125 (2016),
[arXiv:1509.05041 [hep-th]].
275. U. Harst and M. Reuter, On selfdual spin-connections and asymptotic safety,
Phys. Lett. B 753, 395 (2016), [arXiv:1509.09122 [hep-th]].
276. U. Harst and M. Reuter, The “tetrad only” theory space: Nonperturbative renor-
malization flow and asymptotic safety, JHEP 1205, 005 (2012), [arXiv:1203.2158
[hep-th]].
277. D. Benedetti and S. Speziale, Perturbative quantum gravity with the Immirzi
parameter, JHEP 1106, 107 (2011), [arXiv:1104.4028 [hep-th]].
278. D. Benedetti and S. Speziale, Perturbative running of the Immirzi parameter, J.
Phys. Conf. Ser. 360, 012011 (2012), [arXiv:1111.0884 [hep-th]].
279. A. Ashtekar, M. Reuter, and C. Rovelli, From general relativity to quantum
gravity, in: Ashtekar, A., Berger, B. K., Isenberg, J., and MacCallum, M. (Eds.),
General Relativity and Gravitation: A Centennial Perspective, pp. 553–611,
Cambridge University Press, Cambridge (2015), [arXiv:1408.4336 [gr-qc]].
280. A. Ashtekar, A note on helicity and self-duality, J. Math. Phys. 27, 824 (1986).
281. C. Pagani and R. Percacci, Quantum gravity with torsion and non-metricity,
Class. Quant. Grav. 32, 195019 (2015), [arXiv:1506.02882 [gr-qc]].
282. R. L. Arnowitt, S. Deser, and C. W. Misner, Dynamical structure and definition
of energy in general relativity, Phys. Rev. 116, 1322 (1959).
283. E. Gourgoulhon, 3+1 Formalism in General Relativity – Bases of Numerical
Relativity, Springer lecture notes in Physics Vol. 846, Springer, Berlin (2012),
[arXiv:gr-qc/0703035].
332 References

284. S. Rechenberger and F. Saueressig, A functional renormalization group equation

for foliated spacetimes, JHEP 1303, 010 (2013), [arXiv:1212.5114 [hep-th]].
285. W. B. Houthoff, A. Kurov, and F. Saueressig, Impact of topology in foliated
quantum Einstein gravity, Eur. Phys. J. C 77, 491 (2017), [arXiv:1705.01848
[hep-th]].
286. E. Manrique, S. Rechenberger, and F. Saueressig, Asymptotically safe Lorentzian
gravity, Phys. Rev. Lett. 106, 251302 (2011), [arXiv:1102.5012 [hep-th]].
287. J. Biemans, A. Platania, and F. Saueressig, Renormalization group fixed points
of foliated gravity-matter systems, JHEP 1705, 093 (2017), [arXiv:1702.06539
[hep-th]].
288. J. Biemans, A. Platania, and F. Saueressig, Quantum gravity on foliated
spacetimes: Asymptotically safe and sound, Phys. Rev. D 95, 086013 (2017),
[arXiv:1609.04813 [hep-th]].
289. R. Percacci and D. Perini, Constraints on matter from asymptotic safety, Phys.
Rev. D 67, 081503(R) (2003), [arXiv:hep-th/0207033].
290. R. Percacci and D. Perini, Asymptotic safety of gravity coupled to matter, Phys.
Rev. D 68, 044018 (2003), [arXiv:hep-th/0304222].
291. P. Donà, A. Eichhorn, and R. Percacci, Matter matters in asymptotically safe
quantum gravity, Phys. Rev. D 89, 084035 (2014), [arXiv:1311.2898 [hep-th]].
292. J. Ambjørn, A. Görlich, J. Jurkiewicz, et al., The semiclassical limit of causal
dynamical triangulations, Nucl. Phys. B 849, 144 (2011), [arXiv:1102.3929
[hep-th]].
293. J. Ambjørn, Z. Drogosz, J. Gizbert-Studnicki, et al., Impact of topology in causal
dynamical triangulations quantum gravity, Phys. Rev. D 94, 044010 (2016),
[arXiv:1604.08786 [hep-th]].
294. J. Ambjørn, J. Gizbert-Studnicki, A. Görlich, K. Grosvenor, and J. Jurkiewicz,
Four-dimensional CDT with toroidal topology, Nucl. Phys. B 922, 226 (2017),
[arXiv:1705.07653 [hep-th]].
295. M. Reuter and F. Saueressig, Fractal space-times under the microscope: A
renormalization group view on Monte Carlo data, JHEP 1112, 012 (2011),
[arXiv:1110.5224 [hep-th]].
296. D. Dou and R. Percacci, The running gravitational couplings, Class. Quant.
Grav. 15, 3449 (1998), [arXiv:hep-th/9707239].
297. A. Eichhorn, Status of the asymptotic safety paradigm for quantum gravity
and matter, Talk given at Black Holes, Gravitational Waves and Spacetime
Singularities, Rome, Italy (2017), [arXiv:1709.03696 [gr-qc]].
298. P. Donà and R. Percacci, Functional renormalization with fermions and tetrads,
Phys. Rev. D 87, 045002 (2013), [arXiv:1209.3649 [hep-th]].
299. H. Gies and S. Lippoldt, Fermions in gravity with local spin-base invariance,
Phys. Rev. D 89, 064040 (2014), [arXiv:1310.2509 [hep-th]].
300. J. Meibohm, J. M. Pawlowski, and M. Reichert, Asymptotic safety of gravity-
matter systems, Phys. Rev. D 93, 084035 (2016), [arXiv:1510.07018 [hep-th]].
301. A. Eichhorn, Quantum-gravity-induced matter self-interactions in the asymptotic-
safety scenario, Phys. Rev. D 86, 105021 (2012), [arXiv:1204.0965 [gr-qc]].
302. U. Harst and M. Reuter, QED coupled to QEG, JHEP 1105, 119 (2011),
[arXiv:1101.6007 [hep-th]].
303. J.-E. Daum, U. Harst, and M. Reuter, Non-perturbative QEG corrections to
the Yang–Mills beta function, Gen. Rel. Grav. 43, 2393 (2011), [arXiv:1005.1488
[hep-th]].
 References 333

304. M. Gell-Mann and F. E. Low, Quantum electrodynamics at small distances,

Phys. Rev. 95, 1300 (1954).
305. M. Göckeler, R. Horsley, V. Linke, et al., Is there a Landau pole problem in
QED?, Phys. Rev. Lett. 80, 4119 (1998) [arXiv:hep-th/9712244].
306. S. Kim, J. B. Kogut, and M.-P. Lombardo, On the triviality of textbook quantum
electrodynamics, Phys. Lett. B 502, 345 (2001), [arxiv:hep-lat/0009029].
307. S. Kim, J. B. Kogut, and M.-P. Lombardo, Gauged Nambu–Jona–Lasinio studies
of the triviality of quantum electrodynamics, Phys. Rev. D 65, 054015 (2002),
[arxiv:hep-lat/0112009].
308. H. Gies and J. Jaeckel, Renormalization flow of QED, Phys. Rev. Lett. 93, 110405
(2004), [arXiv:hep-ph/0405183].
309. S. P. Robinson and F. Wilczek, Gravitational correction to running of gauge
couplings, Phys. Rev. Lett. 96, 231601 (2006), [arXiv:hep-th/0509050].
310. A. R. Pietrykowski, Gauge dependence of gravitational correction to running of
gauge couplings, Phys. Rev. Lett. 98, 061801 (2007), [arXiv:hep-th/0606208].
311. D. J. Toms, Quantum gravity and charge renormalization, Phys. Rev. D 76,
045015 (2007), [arXiv:0708.2990 [hep-th]].
312. D. Ebert, J. Plefka, and A. Rodigast, Absence of gravitational contributions to
the running Yang–Mills coupling, Phys. Lett. B 660, 579 (2008), [arXiv:0710.1002
[hep-th]].
313. Y. Tang and Y.-L. Wu, Gravitational contributions to running of gauge couplings,
Commun. Theor. Phys. 54, 1040 (2010), [arXiv:0807.0331 [hep-ph]].
314. D. J. Toms, Cosmological constant and quantum gravitational corrections
to the running fine structure constant, Phys. Rev. Lett. 101, 131301 (2008),
[arXiv:0809.3897 [hep-th]].
315. D. J. Toms, Quantum gravity, gauge coupling constants, and the cosmological
constant, Phys. Rev. D 80, 064040 (2009), [arXiv:0908.3100 [hep-th]].
316. D. J. Toms, Quantum gravitational contributions to quantum electrodynamics,
Nature 468, 56 (2010), [arXiv:1010.0793 [hep-th]].
317. M. M. Anber, J. F. Donoghue, and M. El-Houssieny, Running couplings and
operator mixing in the gravitational corrections to coupling constants, Phys. Rev.
D 83, 124003 (2011), [arXiv:1011.3229 [hep-th]].
318. J. Ellis and N. E. Mavromatos, On the interpretation of gravitational corrections
to gauge couplings, Phys. Lett. B 711, 139 (2012), [arXiv:1012.4353 [hep-th]].
319. U. Amaldi, W. de Boer, P. H. Frampton, H. Fürstenau, and J. T. Liu, Consistency
checks of grand unified theories, Phys. Lett. B 281, 374 (1992).
320. R. Percacci, The renormalization group, systems of units and the hierarchy
problem, J. Phys. A: Math. Theor. 40, 4895 (2007), [arXiv:hep-th/0409199].
321. M. Shaposhnikov and C. Wetterich, Asymptotic safety of gravity and the Higgs
boson mass, Phys. Lett. B 683, 196 (2010), [arXiv:0912.0208 [hep-th]].
322. O. Lauscher and M. Reuter, Fractal spacetime structure in asymptotically safe
gravity, JHEP 0510, 050 (2005), [arXiv:hep-th/0508202].
323. M. Reuter and F. Saueressig, From big bang to asymptotic de Sitter: Com-
plete cosmologies in a quantum gravity framework, JCAP 0509, 012 (2005),
[arXiv:hep-th/0507167].
324. B. Koch, P. Rioseco, and C. Contreras, Scale setting for self-consistent
backgrounds, Phys. Rev. D 91, 025009 (2015), [arXiv:1409.4443 [hep-th]].
325. A. B. Migdal, Vacuum polarization in strong non-homogeneous fields, Nucl. Phys.
B 52, 483 (1973).
334 References

326. S. L. Adler, Dynamical applications of the gauge-invariant effective action formal-

ism, in: Christensen, S. M. (Ed.), Quantum Theory of Gravity, Essays in Honor of
the 60th Birthday of Bryce S. DeWitt, pp. 388–402, Adam Hilger, Bristol (1984).
327. W. Dittrich and M. Reuter, Effective Lagrangians in Quantum Electrodynamics,
Springer, Berlin (1985).
328. M. Reuter and H. Weyer, Running Newton constant, improved gravita-
tional actions, and galaxy rotation curves, Phys. Rev. D 70, 124028 (2004),
[arXiv:hep-th/0410117].
329. M. Reuter and H. Weyer, Quantum gravity at astrophysical distances?, JCAP
0412, 001 (2004), [arXiv:hep-th/0410119].
330. M. Reuter and H. Weyer, On the possibility of quantum gravity effects at
astrophysical scales, Int. J. Mod. Phys. D 15, 2011 (2006), [arXiv:hep-th/0702051].
331. B. Koch and I. Ramirez, Exact renormalization group with optimal scale and its
application to cosmology, Class. Quant. Grav. 28, 055008 (2011), [arXiv:1010.2799
[gr-qc]].
332. M. Hindmarsh and I. D. Saltas, f (R) gravity from the renormalization group,
Phys. Rev. D 86, 064029 (2012), [arXiv:1203.3957 [gr-qc]].
333. A. Bonanno and A. Platania, Asymptotically safe inflation from quadratic
gravity, Phys. Lett. B 750, 638 (2015), [arXiv:1507.03375 [gr-qc]].
334. M. Reuter and J.-M. Schwindt, Scale-dependent metric and minimal length in
QEG, J. Phys. A: Math. Theor. 40, 6595 (2007), [arXiv:hep-th/0610064].
335. R. Percacci and G. P. Vacca, Asymptotic safety, emergence and minimal length,
Class. Quant. Grav. 27, 245026 (2010), [arXiv:1008.3621 [hep-th]].
336. D. ben-Avraham and S. Havlin, Diffusion and Reactions in Fractals and
Disordered Systems, Cambridge University Press, Cambridge (2000).
337. B. B. Mandelbrot, The Fractal Geometry of Nature, W. H. Freeman and
Company, New York (1982).
338. S. Carlip, Dimension and dimensional reduction in quantum gravity, Class.
Quant. Grav. 34, 193001 (2017), [arXiv:1705.05417 [gr-qc]].
339. S. Rechenberger and F. Saueressig, R2 phase diagram of quantum Einstein gravity
and its spectral dimension, Phys. Rev. D 86, 024018 (2012), [arXiv:1206.0657
[hep-th]].
340. M. Reuter and F. Saueressig, Quantum Einstein gravity, New J. Phys. 14, 055022
(2012), [arXiv:1202.2274 [hep-th]].
341. S. W. Hawking and G. F. R. Ellis, The Large Scale Structure of Space-Time,
Cambridge University Press, Cambridge (1973).
342. G. W. Gibbons and S. W. Hawking (Eds.), Euclidean Quantum Gravity, World
Scientific, Singapore (1993).
343. B. Koch and F. Saueressig, Black holes within asymptotic safety, Int. J. Mod.
Phys. A 29, 1430011 (2014), [arXiv:1401.4452 [hep-th]].
344. A. Bonanno and M. Reuter, Spacetime structure of an evaporating black hole in
quantum gravity, Phys. Rev. D 73, 083005 (2006), [arXiv:hep-th/0602159].
345. M. Reuter and E. Tuiran, Quantum gravity effects in rotating black holes,
in: Kleinert, H., Jantzen, R. T., and Ruffini, R. (Eds.), The Eleventh Marcel
Grossmann Meeting, On Recent Developments in Theoretical and Experimental
General Relativity, Gravitation and Relativistic Field Theories, pp. 2608–2610,
World Scientific, Singapore (2008), [arXiv:hep-th/0612037].
346. M. Reuter and E. Tuiran, Quantum gravity effects in the Kerr spacetime, Phys.
Rev. D 83, 044041 (2011), [arXiv:1009.3528 [hep-th]].
 References 335

347. A. Bonanno and M. Reuter, Quantum gravity effects near the null black hole
singularity, Phys. Rev. D 60, 084011 (1999), [arXiv:gr-qc/9811026].
348. K. Falls and D. F. Litim, Black hole thermodynamics under the microscope,
Phys. Rev. D 89, 084002 (2014), [arXiv:1212.1821 [gr-qc]].
349. Y.-F. Cai and D. A. Easson, Black holes in an asymptotically safe gravity theory
with higher derivatives, JCAP 1009, 002 (2010), [arXiv:1007.1317 [hep-th]].
350. B. Koch and F. Saueressig, Structural aspects of asymptotically safe black holes,
Class. Quant. Grav. 31, 015006 (2014), [arXiv:1306.1546 [hep-th]].
351. R. Torres, Nonsingular black holes, the cosmological constant, and asymptotic
safety, Phys. Rev. D 95, 124004 (2017), [arXiv:1703.09997 [gr-qc]].
352. H. Emoto, Asymptotic safety of quantum gravity and improved spacetime of
black hole singularity by cutoff identification, [arXiv:hep-th/0511075].
353. B. F. L. Ward, Planck scale remnants in resummed quantum gravity, Acta Phys.
Polon. B 37, 1967 (2006), [arXiv:hep-ph/0605054].
354. K. Falls, D. F. Litim, and A. Raghuraman, Black holes and asymptotically safe
gravity, Int. J. Mod. Phys. A 27, 1250019 (2012), [arXiv:1002.0260 [hep-th]].
355. S. Basu and D. Mattingly, Asymptotic safety, asymptotic darkness, and the hoop
conjecture in the extreme UV, Phys. Rev. D 82, 124017 (2010), [arXiv:1006.0718
[hep-th]].
356. R. Casadio, S. D. H. Hsu, and B. Mirza, Asymptotic safety, singularities,
and gravitational collapse, Phys. Lett. B 695, 317 (2011), [arXiv:1008.2768
[gr-qc]].
357. A. Bonanno, B. Koch, and A. Platania, Cosmic censorship in quantum Einstein
gravity, Class. Quant. Grav. 34, 095012 (2017), [arXiv:1610.05299 [gr-qc]].
358. F. Saueressig, N. Alkofer, G. D’Odorico, and F. Vidotto, Black holes in
asymptotically safe gravity, PoS FFP 14, 174 (2016), [arXiv:1503.06472
[hep-th]].
359. A. Bonanno and M. Reuter, Cosmology of the Planck era from a renormalization
group for quantum gravity, Phys. Rev. D 65, 043508 (2002), [arXiv:hep-th/
0106133].
360. A. Bonanno and M. Reuter, Entropy signature of the running cosmological
constant, JCAP 0708, 024 (2007), [arXiv:0706.0174 [hep-th]].
361. A. Bonanno and S. Carloni, Dynamical system analysis of cosmologies with
running cosmological constant from quantum Einstein gravity, New J. Phys. 14,
025008 (2012), [arXiv:1112.4613 [gr-qc]].
362. A. Bonanno and F. Saueressig, Asymptotically safe cosmology – A status report,
Comptes Rendus Physique 18, 254 (2017), [arXiv:1702.04137 [hep-th]].
363. M. Reuter and F. Saueressig, Asymptotic safety, fractals, and cosmology, in:
Calcagni, G., Papantonopoulos, L., Siopsis, G., and Tsamis, N. (Eds.), Quan-
tum Gravity and Quantum Cosmology, pp. 185–226, Springer, Berlin (2013),
[arXiv:1205.5431 [hep-th]].
364. A. Bonanno and M. Reuter, Entropy production during asymptotically safe
inflation, Entropy 13, 274 (2011), [arXiv:1011.2794 [hep-th]].
365. A. Bonanno and M. Reuter, Cosmology with self-adjusting vacuum energy
density from a renormalization group fixed point, Phys. Lett. B 527, 9 (2002),
[arXiv:astro-ph/0106468].
366. A. Bonanno and M. Reuter, Cosmological perturbations in renormaliza-
tion group derived cosmologies, Int. J. Mod. Phys. D 13, 107 (2004),
[arXiv:astro-ph/0210472].
336 References

367. E. Bentivegna, A. Bonanno, and M. Reuter, Confronting the IR fixed

point cosmology with high-redshift observations, JCAP 0401, 001 (2004),
[arXiv:astro-ph/0303150].
368. M. Reuter and C. Wetterich, Time evolution of the cosmological “constant”,
Phys. Lett. B 188, 38 (1987).
369. T. Padmanabhan, Structure Formation in the Universe, Cambridge University
Press, Cambridge (1993).
370. G. D’Odorico and F. Saueressig, Quantum phase transitions in the Belinsky–
Khalatnikov–Lifshitz universe, Phys. Rev. D 92, 124068 (2015), [arXiv:1511.00247
[gr-qc]].
371. P. A. R. Ade et al. [Planck Collaboration], Planck 2015 results. XIII. Cos-
mological parameters, Astron. Astrophys. 594, A13 (2016), [arXiv:1502.01589
[astro-ph.CO]].
372. D. Benedetti and F. Caravelli, The local potential approximation in quan-
tum gravity, JHEP 1206, 017 (2012), [Erratum: JHEP 1210, 157 (2012)],
[arXiv:1204.3541 [hep-th]].
373. M. Demmel, F. Saueressig, and O. Zanusso, Fixed-functionals of three-dimensional
quantum Einstein gravity, JHEP 1211, 131 (2012), [arXiv:1208.2038 [hep-th]].
374. M. Demmel, F. Saueressig, and O. Zanusso, Fixed functionals in asymptotically
safe gravity, in: Jantzen, R., Rosquist, K., and Ruffini, R. (Eds.), The Thir-
teenth Marcel Grossmann Meeting, On Recent Developments in Theoretical and
Experimental General Relativity, Astrophysics and Relativistic Field Theories,
pp. 2227–2229, World Scientific, Singapore (2013), [arXiv:1302.1312 [hep-th]].
375. M. Demmel, F. Saueressig, and O. Zanusso, RG flows of quantum Einstein grav-
ity on maximally symmetric spaces, JHEP 1406, 026 (2014), [arXiv:1401.5495
[hep-th]].
376. M. Demmel, F. Saueressig, and O. Zanusso, RG flows of Quantum Einstein
gravity in the linear-geometric approximation, Annals Phys. 359, 141 (2015),
[arXiv:1412.7207 [hep-th]].
377. M. Demmel, F. Saueressig, and O. Zanusso, A proper fixed functional for four-
dimensional quantum Einstein gravity, JHEP 1508, 113 (2015), [arXiv:1504.07656
[hep-th]].
378. N. Ohta, R. Percacci, and G. P. Vacca, Flow equation for f (R) gravity and some
of its exact solutions, Phys. Rev. D 92, 061501(R) (2015), [arXiv:1507.00968
[hep-th]].
379. N. Ohta, R. Percacci, and G. P. Vacca, Renormalization group equation and
scaling solutions for f (R) gravity in exponential parametrization, Eur. Phys. J.
C 76, 46 (2016), [arXiv:1511.09393 [hep-th]].
380. S. Gonzalez-Martin, T. R. Morris, and Z. H. Slade, Asymptotic solutions in
asymptotic safety, Phys. Rev. D 95, 106010 (2017), [arXiv:1704.08873 [hep-th]].
381. I. H. Bridle, J. A. Dietz, and T. R. Morris, The local potential approximation in the
background field formalism, JHEP 1403, 093 (2014), [arXiv:1312.2846 [hep-th]].
382. J. A. Dietz and T. R. Morris, Redundant operators in the exact renormalisation
group and in the f (R) approximation to asymptotic safety, JHEP 1307, 064
(2013), [arXiv:1306.1223 [hep-th]].
383. J. A. Dietz and T. R. Morris, Asymptotic safety in the f (R) approximation,
JHEP 1301, 108 (2013), [arXiv:1211.0955 [hep-th]].
384. J. W. York, Jr., Conformally invariant orthogonal decomposition of symmetric
tensors on Riemannian manifolds and the initial-value problem of general
relativity, J. Math. Phys. 14, 456 (1973).
 References 337

385. J. W. York, Jr., Covariant decompositions of symmetric tensors in the theory of

gravitation, Annales de l’I.H.P. Physique théorique 21, 319 (1974).
386. H. Gies, B. Knorr, S. Lippoldt, and F. Saueressig, Gravitational two-loop
Counterterm is asymptotically safe, Phys. Rev. Lett. 116, 211302 (2016),
[arXiv:1601.01800 [hep-th]].
387. M. H. Goroff and A. Sagnotti, The ultraviolet behavior of Einstein gravity, Nucl.
Phys. B 266, 709 (1986).
388. A. E. M. van de Ven, Two-loop quantum gravity, Nucl. Phys. B 378, 309 (1992).
389. A. Nink and M. Reuter, On quantum gravity, asymptotic safety, and paramagnetic
dominance, Int. J. Mod. Phys. D 22, 1330008 (2013), [arXiv:1212.4325 [hep-th]].
390. A. Bonanno and M. Reuter, Modulated ground state of gravity theories with stabi-
lized conformal factor, Phys. Rev. D 87, 084019 (2013), [arXiv:1302.2928 [hep-th]].
391. Y. Igarashi, K. Itoh, and H. Sonoda, Realization of symmetry in the ERG
approach to quantum field theory, Prog. Theor. Phys. Suppl. 181, 1 (2009),
[arXiv:0909.0327 [hep-th]].
392. G. W. Gibbons and S. W. Hawking, Action integrals and partition functions in
quantum gravity, Phys. Rev. D 15, 2752 (1977).
393. A. Ashtekar, J. Engle, and D. Sloan, Asymptotics and Hamiltonians in a first-order
formalism, Class. Quant. Grav. 25, 095020 (2008), [arXiv:0802.2527 [gr-qc]].
394. R. B. Mann, D. Marolf, R. McNees, and A. Virmani, On the stress ten-
sor for asymptotically flat gravity, Class. Quant. Grav. 25, 225019 (2008),
[arXiv:0804.2079 [hep-th]].
395. T. Jacobson and A. Satz, On the renormalization of the Gibbons–Hawking
boundary term, Phys. Rev. D 89, 064034 (2014), [arXiv:1308.2746 [gr-qc]].
396. D. Becker and M. Reuter, Asymptotic safety and black hole thermodynamics,
in: Jantzen, R., Rosquist, K., and Ruffini, R. (Eds.), The Thirteenth Marcel
Grossmann Meeting, On Recent Developments in Theoretical and Experimental
General Relativity, Astrophysics and Relativistic Field Theories, pp. 2230–2232,
World Scientific, Singapore (2013), [arXiv:1212.4274 [hep-th]].
397. A. B. Zamolodchikov, “Irreversibility” of the flux of the renormalization group in
a 2D field theory, JETP Lett. 43, 730 (1986), [Russian: Pis’ma Zh. Eksp. Teor.
Fiz. 43, 565 (1986)].
398. A. B. Zamolodchikov, Renormalization group and perturbation theory about
fixed points in two-dimensional field theory, Sov. J. Nucl. Phys. 46, 1090 (1987),
[Russian: Yad. Fiz. 46, 1819 (1987)].
399. R. Bousso, Positive vacuum energy and the N -bound, JHEP 0011, 038 (2000),
[arXiv:hep-th/0010252].
400. T. Banks, Cosmological breaking of supersymmetry?, Int. J. Mod. Phys. A 16,
910 (2001), [arXiv:hep-th/0007146].
401. L. Smolin, Quantum gravity with a positive cosmological constant,
[arXiv:hep-th/0209079].
402. D. Becker and M. Reuter, Is there a C-function in 4D quantum Einstein gravity?,
in: Finster, F., Kleiner, J., Röken, C., and Tolksdorf, J. (Eds.), Quantum
Mathematical Physics: A Bridge between Mathematics and Physics, pp. 19–41,
Birkhäuser, Basel (2016), [arXiv:1502.03292 [hep-th]].
403. C. Pagani and M. Reuter, JHEP 1807, 039 (2018), [arXiv:1804.02162[gr-qc]].
404. S. N. Solodukhin, Entanglement entropy of black holes, Living Rev. Rel. 14, 8
(2011), [arXiv:1104.3712 [hep-th]].
405. T. Jacobson and A. Satz, Black hole entanglement entropy and the renormalization
group, Phys. Rev. D 87, 084047 (2013), [arXiv:1212.6824 [hep-th]].
338 References

406. A. O. Barvinsky and G. A. Vilkovisky, The generalized Schwinger–DeWitt

technique in gauge theories and quantum gravity, Phys. Rept. 119, 1 (1985).
407. P. B. Gilkey, Invariance Theory, the Heat Equation and the Atiyah–Singer Index
Theorem, Second Edition, CRC Press, Boca Raton (1994).
408. D. V. Vassilevich, Heat kernel expansion: User’s manual, Phys. Rept. 388, 279
(2003), [arXiv:hep-th/0306138].
409. I. G. Avramidi, Heat Kernel and Quantum Gravity, Springer, Berlin (2000).
410. B. S. DeWitt and R. W. Brehme, Radiation damping in a gravitational field,
Annals Phys. 9, 220 (1960).
411. D. Benedetti, K. Groh, P. F. Machado, and F. Saueressig, The universal RG
machine, JHEP 1106, 079 (2011), [arXiv:1012.3081 [hep-th]].
412. Y. Décanini and A. Folacci, Off-diagonal coefficients of the DeWitt–Schwinger
and Hadamard representations of the Feynman propagator, Phys. Rev. D 73,
044027 (2006), [arXiv:gr-qc/0511115].
413. K. Groh, F. Saueressig, and O. Zanusso, Off-diagonal heat-kernel expansion and
its application to fields with differential constraints, [arXiv:1112.4856 [math-ph]].
414. S. M. Christensen, Computational challenges in heat kernel calculations, in:
Fulling, S. A. (Ed.), Heat Kernel Techniques and Quantum Gravity, pp. 47–64,
Texas A&M University, College Station (1995).
415. A. Codello, Polyakov effective action from functional renormalization group
equation, Annals Phys. 325, 1727 (2010), [arXiv:1004.2171 [hep-th]].
416. L. Lewin, Polylogarithms and Associated Functions, Elsevier North Holland, New
York (1981).
417. S. Floerchinger, Analytic continuation of functional renormalization group
equations, JHEP 1205, 021 (2012), [arXiv:1112.4374 [hep-th]].
418. J. M. Pawlowski and N. Strodthoff, Real time correlation functions and the func-
tional renormalization group, Phys. Rev. D 92, 094009 (2015), [arXiv:1508.01160
[hep-ph]].
 Index

action entropy, 291

average, 27 remnant, 276
bare, 16, 208, 217 RG improvement, 272
effective, 20, 30, 208 singularity, 273
Effective Average, see EAA thermodynamics, 274
Holst, 230 BRST transformation, 112
ADM formalism
geometry, 235–236 C function, 291
anomalous dimension, 310 Capovilla–Dell–Jacobson action,
Newton constant, 137 221
Ashtekar variables, 234 Causal Dynamical Triangulation,
Asymptotic Safety, 2, 14 18, 242, 244
enhanced predictivity, 255 coarse-graining, 26
predictive power, 80 interpretation, 97
Asymptotic Safety scenario, 77 matter fields, 96
average action, 27 Coleman–Weinberg potential, 47
composite operator, 289
background field method, 8, 48 conformal factor instability, 189
background geometry convexity, scalar potential, 38
S 4 , 135, 297 correlation function, 19
CP 2 , 297 cosmological constant problem,
S 2 × S 2 , 299 149
Einstein, 298 coupling constant, 14
Ricci-flat, 297 essential, 81
Background Independence generalized, 29, 56
first kind, 3–4, 6–8, 65 inessential, 81
second kind, 15–16, 65 running, 29
beta function, 14, 30, 41 critical exponent, 68
ADM formalism, 240 Einstein–Hilbert truncation, 154
bimetric Einstein–Hilbert, 181 curvature monomial
dimensionless, 57 basis, 295
Einstein–Hilbert truncation, 140 identities, 296
Hilbert–Palatini formalism, 231 curvature tensor, 293
QED coupled to QEG, 252 cutoff
beta functional, 55 adjusted, 45, 142
conformal factor (LPA), 199 compactly supported, 163
dimenionless, 60 complex frequency, 316
scalar field theory, 37 exponential, 21, 162, 312
topology dependence, 201 generalized exponential, 162, 312
Bianchi identity, 293 mass-type, 316
bimetric approach, 7 optimized, 163, 315
black hole sharp, 163, 312–315
causal structure, 274 cutoff action, 20–22
340 Index

cutoff function, 21 gravity, 120

general properties, 311 Hilbert–Palatini formalism, 230
cutoff identification, 265 reduced, 130
cutoff mode, 193 scalar field, 23
single-metric truncation, 131
decoupling, heavy modes, 44 functional RG equation, see FRGE
diffeomorphism, 91
foliation preserving, 236 gauge fixing
modified, 225 background type, 112
dimensional reduction, 269 generalized harmonic gauge,
dimensionless coupling constants, 57 113, 239
Einstein–Hilbert truncation, 139 proper time, 238
gauge transformation
EAA, 7, 11 background, 11, 112
gravity, 117 quantum, 10, 110
integral form, 121 General Relativity, 1
Hilbert–Palatini formalism, 229 Gibbons–Hawking term, 290
level expansion, 176 gravitational antiscreening, 146
scalar fields, 22 gravity
Effective Average Action, broken phase, 88
see EAA unbroken phase, 88
effective field equation, 10, 22 gravity-matter fixed point, 249
effective field theory, 81 enhanced predictivity, 251
Einstein–Cartan gravity, 220
gauge fixing, 223
Halpern–Huang direction, 74
entanglement entropy, 291
harmonic gauge, see gauge fixing
exponential map, 106
heat kernel, 301
coefficient, 302–304
Faddeev–Popov determinant, 111
early time expansion, 302
Faddeev–Popov operator, 113
off-diagonal, 305
field
hierachy problem, 262
ADM formalism, 235
Hilbert–Palatini action, 221
averaged, 26
Holst action, 222
mode expansion, 96
field theory
effective, 2 Immirzi parameter, 222
fundamental, 2 renormalization, 232–233
fine-structure constant, 251 small-large duality, 233
prediction, 259 infinite-order problem
fixed functional gauge theory, 100
f (R)-type, 286 gravity, 102
properties, 283–284 inflation
scalar, 40 NGFP-driven, 278
stability analysis, 201 irrelevant operator, 68
fixed singularity, 285
flow Landau pole, 252
dynamical system, 61 Laplacian, 294
linearized, 67–68 background, 114
Fourier transform, 294 gauge theory, 95
fractal, 267 Lichnerowicz, 160
FRGE, 5, 9–13 spectrum S 4 , 199
component form, 14, 30, 56 lapse function, 235
conformally reduced gravity, 191 Lie derivative, 91
equivalent form, 24 Loop Quantum Gravity, 17
 Index 341

marginal operator, 68 Quantum Einstein Gravity,

metric see QEG
background, 7 quantum spacetime, 263
degenerate, 5 effective dimension, 268
exponential split, 108
fluctuation, 7 reconstruction formula, bare
linear split, 7, 109 action, 214
operator, 87 reconstruction problem, 16
reference, 266 relevant operator, 68
Robertson–Walker, 276 renormalizability
Schwarzschild, 271 nonperturbative, 2
spatial, 235 renormalization
metric fluctuation nonperturbative, 75
traceless decomposition, 134, 317 perturbative, 75
transverse-traceless decomposition, renormalization group
282, 317 flow, 5, 14, 62
metric quantum gravity, time, 39
see QEG trajectory, 14, 29
minimal length scale, 268 complete, 76
mode expansion Wilson–Kadanoff, 207
ordered set, 96 RG fixed point
moving singularity, 285 Gaussian, 15, 71
non-Gaussian, 15, 71
NGFP trajectory, 67
2 + ε-dimensional gravity, 83 RG improvement, 47, 265
R2 -truncation, 168 black hole, 272
R2 + C 2 -truncation, 172 cosmology, 277
ADM formalism, 240 Riemann tensor, 293
conformally reduced Einstein–Hilbert, 198
Einstein–Hilbert truncation, 151, 164–166 scalar field theory
gravity-matter, 249 beta function, 42
Hilbert–Palatini formalism, 232 beta functional, 37
polynomial f (R)-truncation, 171 EAA, 22
QED coupled to QEG, 254 FRGE, 23
two-loop counterterm, 288 LPA, 35
noncommutative geometry, 17 scaling operator, 68, 290
scalar field theory, 73
paramagnetic dominance, 288 self-consistent background, 125
phase diagram volume profile, 243
bimetric Einstein–Hilbert, 182 self-similarity, spacetime, 267
conformally reduced Einstein–Hilbert, 197 shift vector, 235
Einstein–Hilbert truncation, 156 shooting method, 286
QED coupled to QEG, 257 spacetime
phase space Bianchi IX, 280
Hamiltonian, 61 classical, 18
Planck cell, 17 QEG, 90
Planck scale, 147 quantum, 17, 267
Plebański action, 221 spacetime torsion, 221
primordial fluctuations, 279 spectral dimension, 269–270
spin connection, 246
Q-functional, 138, 308–309 split symmetry restoration, 184
QEG, 16, 87 split symmetry transformation, 8,
building blocks, 89 51, 108–109
space-time, 90 stability matrix, 67
342 Index

Standard Model, 1 single metric, 130

Asymptotic Safety, 248 two-loop counterterm, 287
stress-energy conservation, 177
supertrace, 55 universal quantity, 38, 42, 69
unstable manifold, 14, 70
tadpole equation, 126 dimension, 70, 174
tetrad, 220 parameterization, 173
Theory Space, 13, 28–29 UV critical hypersurface,
connection, 108 see unstable manifold
flat, 109
coordinate transformation, 64 vertex expansion,
geometrical picture, 55–60 see EAA, level expansion
metric gravity, 124
QEG, 94 Ward identity
reduced, 203 BRST, 123
threshold function, 138, 310 modified BRST, 122
torsion tensor, 223 modified split symmetry, 123, 185
truncation, 31–35 split symmetry, 205
R2 , 159 Ward operator, 224
R2 + C 2 , 159 weak hypercharge, 261
bimetric ansatz, 128 Weinberg, Steven, 77
bimetric EH, 178 Wetterich equation, 23
conformal factor (LPA), 198 Weyl tensor, 294
conformally reduced Einstein–Hilbert, 194 Wigner–Weyl–Moyal formalism, 17
Einstein–Hilbert, 132 Wilson–Fisher fixed point, 40
Holst type, 230 Wu-Yang ambiguity, 98
infinite-dimensional, 282
LPA, 35 Yang–Mills theory, 92
polynomial f (R), 160 Yang–Mills connection, 301
QED coupled to QEG, 252
reliability test, 162 zero mode, 92