't Hooft anomalies of discrete gauge theories and non-abelian group cohomology (Q2181949)

This review has three parts labelled I, II, III. \bigskip I. Introduction. This article deals with \textit{anomalies} possessed by a certain class of quantum field theories with a gauge group: namely those for which the gauge group is a finite group. These latter -- usually called Dijkgraaf-Witten theories -- are \textit{topological quantum field theories}. The main result can be summarised briefly as follows: Suppose that \(M\) is an \(n\)-dimensional manifold, on which there is a gauge theory with finite group \(D\), and that this gauge theory also has a \textit{global symmetry group} \(G\), also finite. It transpires that there are examples where the global symmetry \(G\) acting on the fields is realised projectively in \(D\) -- i.e. we have a a representation of \(G\) modulo elements of \(D\) -- and so one has an extension \(\widehat G\) of \(G\) by \(D\). Given these circumstances it turns out that one is naturally led to interesting topological conditions for the existence of an anomaly to gauging the global symmetry \(G\). These topological conditions are the subject of this paper; also described in detail are mechanisms for anomaly cancellation. For the remainder of this review we must now elaborate on what we have said above. This article leans very heavily towards the mathematical side of its subject rather than the physical: it makes extensive use of a functorial approach to quantum field theory, as well as employing, algebraic topology, spectral sequences and group cohomology. As a consequence it is a challenging article to read and to review. Hence, in meeting this challenge, it is helpful to first provide some background on both anomalies and topological quantum field theories. We shall begin with anomalies and then move on to topological quantum field theories. \bigskip II. Background. The story of anomalies begins with \textit{S. Adler} [``Axial-vector vertex in spinor electrodynamics'', Phys. Rev. 177, 2426--2438 (1969; \url{doi:10.1103/PhysRev.177.2426}); Lectures on elementary particle physics and quantum field theory. Cambridge, MA - London: The MIT Press (1970)] and \textit{J. S. Bell} and \textit{R. Jackiw} [``A PCAC puzzle: $\pi^0\to\gamma\gamma$ in the $\sigma$-model'', Nuovo Cimento A 60, 47--61 (1969; \url{doi:10.1007/BF02823296})] and has to do with strange or even pathological behaviour of \textit{chiral Fermions} when coupled to Abelian gauge fields. Subsequently many other kinds of anomalies have been found to occur in a variety of theories: these theories include Yang-Mills theories, theories of gravity and string theory. In this wider context it is more appropriate to consider an anomaly as some obstruction to the defining of a quantum theory with the same invariance group as its underlying classical theory. A common feature of many of these anomalies is that this obstruction has a topological origin: the topology involved belonging to the cohomology of the invariance groups of the theory. The invariance groups that are encountered can be the group of gauge transformations \(\mathcal{G}\) of an Abelian or non-Abelian theory gauge theory with gauge group \(G\), the group of orientation preserving general coordinate transformations \(\mathrm{Diff}^+(M)\) of a manifold \(M\), or a mixture of the two, and, in string theory, we find manifestations of anomalies with all these varieties of gauge group. The string theory examples include the celebrated conformal anomaly of the Bosonic string where the \(\bar\partial\) operator -- which is the Dirac operator in two dimensions -- plays the leading rôle; this anomaly vanishes when spacetime has the critical dimension \(26\). To be found amongst these examples, also, is the seminal anomaly cancellation success of \(N=1\) supersymmetric string theory in critical dimension \(10\), this latter singling out the two \(496\)-dimensional gauge groups \(E_8\times E_8\) and \(O(32)\). If, as an example, we consider the Yang-Mills case, then one has to consider certain elements of \(H^*(\mathcal{G})\). In constructing these elements it is frequently useful to consider families of Dirac operators \({\rlap/\partial}_A\) parametrised by a connection \(A\in\mathcal{A}\) in the space of all connections, and look for an obstruction to the existence of a gauge invariant determinant for \({\rlap/\partial}_A\) (or more precisely \(\sqrt{\det({\rlap/\partial}_A^*{\rlap/\partial}_A)}\)): the point is that under a gauge transformation \(g\in \mathcal{G}\) where \(A\longmapsto A_g\) one has \[ {\rlap/\partial}_{A_g}=g^{-1}{\rlap/\partial}_A g \] The determinant of an operator \(O\) vanishes if \(\ker O\) is non-empty and so the index of \({\rlap/\partial}_A\) enters; further, as \(A\in\mathcal{A}\) varies we generate a family of such objects, so the index theorem for the family \(\operatorname{Index}{\rlap/\partial}\) given by \[ \operatorname{Index} {\rlap/\partial}=\{\ker {\rlap/\partial}_A\mid A\in\mathcal{A} \}-\{\ker {\rlap/\partial}^*_A\mid A\in\mathcal{A} \} \] is needed; we see that \(\operatorname{Index} {\rlap/\partial}\in K(\mathcal{A/G})\), where \(K(\mathcal{A/G})\) denotes the K-theory of \(\mathcal{A/G}\). The upshot of all this is that the absence of a gauge invariant Dirac determinant means that \(\sqrt{\det({\rlap/\partial}_A^*{\rlap/\partial}_A)}\) does not descend from the space of connections \(\mathcal{A}\) to a \textit{function} on the orbit space \(\mathcal{A/G}\): instead the determinant is a \textit{section} of the non-trivial determinant line bundle \[ \det\operatorname{Index}{\rlap/\partial} \] which is characterised by having non-zero Chern class \(c_1(\det\operatorname{Index}{\rlap/\partial})\in H^2(\mathcal{A/G})\). This in turn means that one does not obtain a gauge invariant Jacobian \(\sqrt{\det({\rlap/\partial}_A^*{\rlap/\partial}_A)}\) by first integrating out the Fermions, and so one cannot then integrate over the orbit space \(\mathcal{A/G}\) to construct the quantum theory: in blunt terms the quantum theory does not exist. Finally we say that a theory in question has an anomaly when one or more of these topological obstructions is non-zero. The anomalies mentioned thus far can, in the main, be calculated using de Rham cohomology which is constructed using local quantities and so can be thought of as \textit{local anomalies}. However there are also topological obstructions which are \textit{not local} and must vanish for a theory to be anomaly free. Global anomalies enter when \(\mathcal{G}\) or \(\mathrm{Diff}^+(M)\) have more than one connected component: i.e. when \(\pi_0(\mathcal{G})\) and \(\pi_0(\mathrm{Diff}^+(M))\) are non-trivial. This means, for example, that they contain elements not continuously connected to the identity. Such discreteness is not detectable using methods of curvature and de Rham cohomology -- these methods are only sensitive to objects in the tangent space. The situation is closely analogous to the calculation of torsion in homology and cohomology. Unfortunately, torsion calculations are typically more difficult than free cohomology calculations for which the de Rham method may well apply. If \(\dim M=d\), then thus far, we needed \(d\) to be \textit{even} in order for chiral Fermions to exist; \textit{odd dimensions} will now occur in the following way: the checking process for global anomalies can be done by working with a one parameter family of Dirac operators: this will be a family of connections \(A_t\), \(t\in S^1\), in the Yang-Mills case, or a family \(\rho_t\), \(t\in S^1\), of metrics in the gravitational case, where \(\rho\) denotes a metric on spacetime \(M\). In each case these \(S^1\) families of Dirac operators can be promoted to be a \textit{single} Dirac operator \({\rlap/\hskip-0.2em D}\) on certain manifold \(N\) dimension \(d+1\): thus \(\dim N\) is odd and the Dirac operator \({\rlap/\hskip-0.2em D}\) is necessarily \textit{self-adjoint} and so that its index is zero. However, in odd dimensions, a \textit{self-adjoint} operator such as \({\rlap/\hskip-0.2em D}\), with a spectrum \(\{\lambda_n\}\) unbounded from above and below, possesses an important spectral invariant \(\eta_{\rlap/\hskip-0.2em D}(s)\) defined by \[ \eta_{\rlap/\hskip-0.2em D}(s)=\sum_{\lambda_n\not=0}\frac{sign\,(\lambda_n)}{\vert\lambda_n\vert^s} \] for which the number \(\eta_{\rlap/\hskip-0.2em D}(0)\) is always finite and is distinguished by the fact that \(\eta_{\rlap/\hskip-0.2em D}(0)\) appears as part of the boundary contribution to the index of an elliptic operator on a manifold \textit{with boundary}. All these matters are relevant but suffice it to say, here, that the operator \({\rlap/\hskip-0.2em D}\) is, in general, \textit{not periodic} in \(t\), and we find that, on going once round \(S^1\), \[ \det{\rlap/\hskip-0.2em D}\longmapsto \exp\left[\frac{\pi i}{2} \eta_{\rlap/\hskip-0.2em D}(0)\right]\det{\rlap/\hskip-0.2em D} \] We are dealing with a determinant line bundle \(\det {\rlap/\hskip-0.2em D}_{S^1}\) over \(S^1\) which, for \(\eta_{\rlap/\hskip-0.2em D}(0)\not=0\bmod 4\), is non-trivial -- the theory then possessing a global anomaly. It is vital to check for the presence or absence of global anomalies in these theories. Two useful papers which illustrate this are Witten [\textit{E. Witten}, ``An SU (2) anomaly'', Phys. Lett. B 117, No. 5, 324--328 (1982; \url{doi:10.1016/0370-2693(82)90728-6}); \textit{E. Witten}, Commun. Math. Phys. 100, 197--229 (1985; Zbl 0581.58038)]. In [Commun. Math. Phys. 100, 197--229 (1985; Zbl 0581.58038)] \textit{E. Witten} analyses global gauge and gravitational anomalies in \(\mathcal{G}\) and \(\mathrm{Diff}^+(M)\) for supersymmetric string theory where the gauge group \(G\) is \(E_8\times E_8\) or \(O(32)\), and the spacetime \(M=S^{10}\): there are various operators contributing a non-zero term of the type \(\eta_{\rlap/\hskip-0.2em D}(0)\) -- one for each Fermion type in the supersymmetry algebra -- but they contribute additively to the calculation, and their sum is zero mod 4 which means that the global anomaly is absent. In [\textit{E. Witten}, ``An SU (2) anomaly'', Phys. Lett. B 117, No. 5, 324--328 (1982; \url{doi:10.1016/0370-2693(82)90728-6})] a simple, but important, global anomaly is identified and calculated for \(SU(2)\) Yang-Mills theory. There is another well known anomaly which is expressed in terms of an \(\eta\)-invariant: this is the \textit{parity anomaly} for Fermions in \textit{odd dimensions} [\textit{A. J. Niemi} and \textit{G. W. Semenoff}, ``Axial-anomaly-induced fermion fractionization and effective gauge-theory actions in odd-dimensional space-times'', Phys. Rev. Lett. 51, 2077--2080 (1983; \url{doi:10.1103/PhysRevLett.51.2077}); {A. N. Redlich}, ``Gauge noninvariance and parity nonconservation of three-dimensional fermions'', Phys. Rev. Lett. 52, 18--21 (1984; \url{doi:10.1103/PhysRevLett.52.18}); \textit{L. Alvarez-Gaumé} et al., Ann. Phys. 163, 288--317 (1985; Zbl 0584.58049)]. In brief the situation is as follows. Let \({\rlap/\hskip-0.2em D}_A={\rlap/\partial}+{\rlap/\hskip-0.2em A} \) be the self adjoint Dirac operator in an odd dimension coupled to a gauge field \(A\). Let the spacetime coordinates be \((x_1,\ldots,x_{n-1},t)\) with \(n\) odd, then under the `parity' transformation \(t\mapsto -t\) the eigenvalues \(\lambda_i\) of \({\rlap/\hskip-0.2em D}_A\) change sign. The standard Fermion effective action \[ e^{-\Gamma(A)}=\int \mathcal{D}\psi \mathcal{D}\bar \psi exp\left[-\int_M \bar\psi\,{\rlap/\hskip-0.2em D}_A\psi \right] \] changes under the parity transformation and one easily checks that the change in \(\Gamma(A)\) is pure imaginary and given by \[ \frac{\pi i}{2} \eta_{\,{\rlap/\hskip-0.2em D}_A}(0) \] and this is the parity anomaly: a non-perturbative result. An important property of these results about anomalies is that they are topological and so are deformation invariants, and are typically discrete: for example one has \(\operatorname{index}{\rlap/\partial}_A\in\mathbb{Z}\) and \(\pi_0(\mathrm{Diff}^+(S^{10}))=\mathbb{Z}_{992}\); this means that they do not receive corrections from perturbation theory and so are exact; despite this, some of them can be detected in perturbation theory, e.g. at one loop in the shape of the famous triangle diagram anomaly. For many more details on the material alluded to so far cf. Nash [\textit{C. Nash}, Differential topology and quantum field theory. New York: Academic Press (1991; Zbl 0752.57001)] and the papers in [\textit{M. F. Atiyah} and \textit{I. M. Singer}, Proc. Natl. Acad. Sci. USA 81, 2597--2600 (1984; Zbl 0547.58033); \textit{K. Fujikawa}, ``Path integral measure for gauge invariant Fermion theories'', Phys. Rev. Lett. 42, 1195--1198 (1979; \url{doi:10.1103/PhysRevLett.42.1195}); \textit{L. Alvarez-Gaumé} and \textit{P. Ginsparg}, Ann. Phys. 161, 423--490 (1985; Zbl 0579.58038); \textit{O. Alvarez}, ``Conformal anomalies and the index theorem'', Nucl. Phys. B 286, 175--188 (1987; \url{doi:10.1016/0550-3213(87)90436-6}); \textit{M. B. Green} and \textit{J. H. Schwarz}, ``Anomaly cancellations in supersymmetric D = 10 gauge theory and superstring theory'', Phys. Lett. B 149, 117--122 (1984; \url{doi:10.1016/0370-2693(84)91565-X}); \textit{O. Alvarez} et al., Commun. Math. Phys. 96, 409--417 (1984; Zbl 0587.58042); \textit{D. S. Freed}, Commun. Math. Phys. 107, 483--513 (1986; Zbl 0606.58013)]. Now we leave the topological formalism to introduce an important, influential, result [\textit{G. 't Hooft}, ``Naturalness, chiral symmetry, and spontaneous chiral symmmetry breaking'', NATO Adv. Study Inst. Ser. B Phys. 59, 135--157 (1980; \url{doi:10.1007/978-1-4684-7571-5_9}] due to 't Hooft which used the exactness of gauge theory anomalies to predict the existence of extra massless Fermions; to understand the result, one needs to digest the following four points: \begin{itemize} \item[(i)] One takes a quantum field theory with whose Fermions \(\psi\) a possess a \textit{global} symmetry group \(G\). If, just for simplicity, we take an Abelian example, \(G=U(1)\), and \(a\) is a \textit{a constant}, then under the map \(\psi\longmapsto \exp[i a\gamma_5]\psi\) the Fermion kinetic term in the action is invariant; hence, in the absence of any other Fermion coupling, one has a global symmetry. Noether's theorem then tells us that the axial current \[ j^\mu_5 =\bar\psi\gamma^\mu \gamma_5\psi \] is conserved. \item[(ii)] If one takes the further step of coupling the Fermions to a \textit{fixed gauge field} \(A\), then \(j^\mu_5 \) is no longer conserved -- instead, if \(F_{\mu\nu}\) is the field strength of \(A\) and \(g\) is its Fermionic coupling constant, one finds the famous result that \[ \partial_\mu j^\mu_5=\frac{g^2}{16 \pi^2} \epsilon_{\mu\nu\rho\sigma}F^{\mu \nu}F^{\rho\sigma} \] and we have an anomaly. \item[(iii)] If, next, we wish promote the gauge field \(A\), from being fixed, to being \textit{dynamical}-something that would mean that we would integrate over \(A\) in the functional integral -- then we are stuck because such a procedure cannot make gauge invariant sense as we pointed out already: the desired theory with a \textit{local gauge group} \(G\) simply does not exist because of the anomaly. This phenomenon is called a \textit{'t Hooft anomaly}. Summarising, a 't Hooft anomaly is as an obstacle to the gauging of a global symmetry. It is in this context that 't Hooft anomalies appear in the paper under review. \item[(iv)] 't Hooft also gave an argument about the existence of bound states: since this aspect of a 't Hooft anomaly is not directly relevant to this paper we only give a sketch of the logic here. His reasoning was as follows: one begins with a theory of chiral Fermions with a global symmetry \(G\) and a non-zero anomaly constant \(A_G\), say, one then introduces what are called \textit{spectator Fermions}, these are massless Fermions which also have a non-zero anomaly constant \(A_{spectator}\), say. Now we choose the second anomaly to cancel the first -- anomalies are additive between Fermions so one can do this -- that is one sets \[ A_{spectator}=-A_G. \] This choice is known as \textit{'t Hooft matching}. The theory has become anomaly free and we can now gauge the global symmetry \(G\). This, we do, but choose the gauge coupling \(g\) of the spectator Fermions to be very small, here \textit{small}, means small enough so that the spectator Fermions do not change the massless particle spectrum of the original theory -- note that however small \(g\) is, the anomaly constant \(A_{spectator}\) stays the same. Still more, though the theory may be described by different degrees of freedom at differing energy scales, the theory remains anomaly free whatever the energy scale. \end{itemize} Hence, in the case where the \textit{original Fermions are confined}, these original Fermions are degrees of freedom that are not seen at low energies; but the spectator Fermions will now create an anomaly unless there are some \textit{new massless Fermions} (presumed to be bound states of the original Fermions) which contribute an anomaly constant \(A_{new}\) such that \[ A_{new}=A_G \] so that the total anomaly constant \(A_{new}+A_{spectator}\) remains zero. Thus we have made a fundamental deduction about the massless Fermion spectrum. Finally we point out that the global symmetry \(G\) must remain exact and, in particular must not be broken spontaneously since then there would be new massless particles: Goldstone Bosons. This last point can even be exploited in reverse to deduce that some symmetry \(G\) \textit{is spontaneously broken}: it suffices to consider the case \(G=U(1)_V\times SU(N)_L\times SU(N)_R\) (cf. [\textit{D. Tong}, ``Gauge theory'', \url{http://www.damtp.cam.ac.uk/user/tong/gaugetheory.html}; \textit{S. Weinberg}, The quantum theory of fields. Cambridge University Press. Volume I (1996; Zbl 0959.81002); Volume II (1996; Zbl 0885.00020)]) where one can deduce that there is no possible set of bound states which can give the desired value of \(A_{new}\), the only alternative is that \(G\) is spontaneously broken and there are massless Goldstone Bosons. Condensed matter physics now makes extensive use of topological methods and has entered the arena of anomalies. These developments -- cf. [\textit{X. Chen, Z.-C. Gu, Z.-X. Liu} and \textit{X.-G. Wen}, ``Symmetry protected topological orders and the group cohomology of their symmetry group'', Phys. Rev. B 87, 155114 (2013; \url{doi:10.1103/PhysRevB.87.155114}); \textit{A. Kapustin} and \textit{R. Thorngren}, ``Anomalies of discrete symmetries in various dimensions and group cohomology'', Preprint, \url{arXiv:1404.3230}; \textit{E. Witten}, ``Fermion path integrals and topological phases'', Rev. Mod. Phys. 88, 35001 (2016; \url{doi:10.1103/RevModPhys.88.035001});\textit{E. Witten} and \textit{K. Yonekura}, ``Anomaly inflow and the $\eta$-invariant'', Preprint, \url{arXiv:1909.08775}; \textit{J. Wang, X.-G. Wen} and \textit{E. Witten}, ``A new SU (2) anomaly'', J. Math. Phys. 60, 052301 (2019; \url{doi:10.1063/1.5082852}); \textit{E. Witten}, ``The `parity' anomaly on an unorientable manifold'', Phys. Rev. B 94, 195150 (2016; \url{doi:10.1103/PhysRevB.94.195150})] -- are proving very illuminating and constitute an important bridge connecting condensed matter physics and quantum field theory. The condensed matter fields that deserve mention here include, the integer quantum Hall effect, topological insulators, topological superconductors, and these constitute examples of \textit{symmetry protected topological phases} or SPT phases. An SPT phase is a gapped phase that admits a description in terms of 't Hooft anomalies: in this description an SPT phase on a manifold \(N\) of dimension \(d\) has a global symmetry \(G\) with a 't Hooft anomaly, but \(N\) is required to be the boundary of a manifold \(M\) called the bulk, and on \(M\) there is a certain topological field theory structured so that the combined theory on boundary and bulk is \textit{anomaly free}: one pictures this anomaly cancellation as a \textit{flow} from the bulk to the boundary. This interplay of the anomaly between bulk and boundary is called \textit{anomaly inflow} -- a tying together of bulk and boundary is also a key content of the \textit{holographic principle} of 't Hooft and Susskind [\textit{G. 't Hooft}, ``Dimensional reduction in quantum gravity'', Preprint, \url{arXiv:gr-qc/9310026}; \textit{L. Susskind}, ``The world as a hologram'', J. Math. Phys. 36, 6377--6396 (1995; \url{doi:10.1063/1.531249})]. The calculational details for Fermion anomalies in dimension \(n\) connects them to Chern-Simons theories in dimension \(n+1\). This is what is being manifest in the anomaly inflow mechanism with Fermions -- e.g. in the integer quantum Hall effect. However the theories studied in this paper are Dijkgraaf-Witten theories [\textit{R. Dijkgraaf} and \textit{E. Witten}, Commun. Math. Phys. 129, No. 2, 393--429 (1990; Zbl 0703.58011)], and such theories have two special characteristics: they have a finite gauge group and are \textit{topological quantum field theories}, about which latter we now give a short outline. Topological quantum field theories grew out of work on two dimensional conformal field theories. The only requisite data for conformal field theories is a Riemann surface \(\Sigma\) and its complex structure: with these the Hilbert space of the particular theory can be constructed. Topological quantum field theories require even less data: just a manifold, not even a metric. Without a metric there are no distance measurements or forces and so no conventional dynamics. The Hamiltonian \(\mathcal{H}\) of the theory has only zero eigenstates and the Hilbert space of the theory, unlike the conformal case, is usually finite dimensional. The non-triviality of the theory is reflected in the existence of tunnelling between vacua. An axiomatic approach to conformal field theories by Segal [\textit{G. Segal}, ``Two dimensional conformal field theories and modular functors'', I. A. M. P. Congress, Swansea, 1988. Institute of Physics (1989)] has been very influential in that subject, and elsewhere: in particular it gave rise to an axiomatic definition by Atiyah [\textit{M. Atiyah}, Publ. Math., Inst. Hautes Étud. Sci. 68, 175--186 (1988; Zbl 0692.53053)] for a \textit{topological quantum field theory or TQFT} and this, in its turn, has stimulated the creation of various axiomatic variants. Topological quantum field theories seek to connect the possible spacetime manifolds of a quantum field theory, with their possible Hilbert spaces, in a flexible topological and relativistic like manner. Inspired by the string theory picture where a closed string sweeps out a Riemann surface, an \(n\)-dimensional TQFT takes a closed spatial manifold \(\Sigma_1\) (\(\dim \Sigma_1=n-1\)) at time \(t_1\) and views its evolution to another manifold \(\Sigma_2\) at a later time \(t_2\) as a \textit{cobordism} \(M\), say, from \(\Sigma_1\) to \(\Sigma_2\). The spatial manifolds \(\Sigma\) and their cobordisms are assembled into a cobordism category, which we denote by \(\operatorname{Cob}_n\), and the associated Hilbert spaces form the category \(\operatorname{Vect}_F\) of vector spaces over some field \(F\) (very often \(F=\mathbb{C}\)). The morphisms of the two categories \(\operatorname{Cob}_n\) and \(\operatorname{Vect}_F\) being cobordism and linear maps respectively; both categories possess symmetric products: disjoint union \(\sqcup\), for \(\operatorname{Cob}_n\), and standard tensor product \(\otimes\), for \(\operatorname{Vect}_F\). A category with such a product is called \textit{monoidal}. Finally the Hilbert spaces are assigned functorially by giving a `partition function' \textit{functor} \(Z\) which acts on \(\operatorname{Cob}_n\) and respects the two products: \[Z:(\operatorname{Cob}_n,\sqcup)\longrightarrow (\operatorname{Vect}_F,\otimes)\] \(Z\) can be used to compute partition functions, quantum states or vectors, including vacuum states, and other significant quantities. The entire machinery is assembled together in order to declare that an \(n\)-dimensional topological quantum field theory is simply an instance of the functor \(Z\) just described -- cf. [\textit{C. Nash}, Differential topology and quantum field theory. New York: Academic Press (1991; Zbl 0752.57001)][\textit{G. Segal}, ``Two dimensional conformal field theories and modular functors'', I. A. M. P. Congress, Swansea, 1988. Institute of Physics (1989); \textit{M. Atiyah}, Publ. Math., Inst. Hautes Étud. Sci. 68, 175--186 (1988; Zbl 0692.53053); \textit{J. Kock}, Frobenius algebras and 2D topological quantum field theories. Cambridge: Cambridge University Press (2004; Zbl 1046.57001); \textit{D. S. Freed}, Bull. Am. Math. Soc., New Ser. 46, No. 2, 221--254 (2009; Zbl 1166.81037); \textit{J. Lurie}, in: Current developments in mathematics, 2008. Somerville, MA: International Press. 129--280 (2009; Zbl 1180.81122)] and references therein for important successes, examples, and fuller information. In the article under review the discussion cannot proceed without adding in extra fields -- for example one wishes to have gauge fields. These fields do not fluctuate, and so are non-dynamical background fields: we denote the set of these background fields by \(\mathcal{F}\). Later, if one is successful in ridding the theory of anomalies, one may allow the elements of \(\mathcal{F}\) to be dynamical and one will integrate over them in the basic functional integral, or, in the case of Dijkgraaf-Witten theories, where \(\mathcal{F}\) is a finite set, one can actually sum over \(\mathcal{F}\). The category \(\operatorname{Cob}_n\) with background fields included is written as \(\operatorname{Cob}_n^{\mathcal{F}}\) and, setting \(F=\mathbb{C}\), the TQFT functor \(Z\) now looks like \[ Z:\operatorname{Cob}_n^{\mathcal{F}}\longrightarrow \operatorname{Vect}_{\mathbf C} \] The incorporation of anomalies into the functorial picture requires us to focus attention topological quantum field theories known as \textit{invertible theories}: an invertible QFT is one for which the functor \(Z\) has the property that, \textit{for all} \(M\) and \(\Sigma\) as above, one has \[ Z(M)\not=0\; \text{ and } \dim Z(\Sigma)=1 \] The term invertibility stems from the fact that, if \( V=Z(\Sigma)\) is one dimensional, there exists another vector space \(V^\prime\) for which \(V\otimes V^\prime\simeq \mathbb{C}\) -- something which requires \(\dim V=1\). The \textit{trivial} invertible theory, whose functor we denote by \(\mathbf 1\), assigns constant values to its basic constituents: in this case that means that \[ {\mathbf 1}(M)=1\text{ and }\; {\mathbf 1}(\Sigma)=\mathbb{C} \] The connection of anomalies to invertibility is because invertibility is the functorial manifestation of the determinant line bundle characterisation of anomalies found in the geometric approach described above: the evaluation of the quantity \(Z(\Sigma) \) can be regarded as assigning a complex line to the partition function of the theory on \(\Sigma\), thus we make contact with the determinant line approach. A final pair of additions to the functorial machine is needed to accommodate anomalies and their inflow: these additions are the notions of \textit{extended} and \textit{relative} field theories. We turn first to extended theories. This first addition is that of including manifolds whose dimensions are successively lower and lower than \(n=\dim M\). Thus far we have had manifolds \(M\) and \(\Sigma\) of dimensions \(n\) and \(n-1\) respectively; one can also include manifolds \(N\), say, of dimension \(n-2\), \(n-3\) etc. Such a quantum field theory is known as an \textit{extended quantum field theory} -- cf. [\textit{D. S. Freed}, Bull. Am. Math. Soc., New Ser. 50, No. 1, 57--92 (2013; Zbl 1280.57025)]. In the paper under review one only goes down as far as \(n-2\). One now allows \(n-1\) dimensional manifolds \(\Sigma\) to have boundaries, and since these boundaries are \(n-2\) dimensional and closed, these alone require us to include closed \(n-2\)-dimensional manifolds. However this is enough to promote the category \(\operatorname{Cob}_n^\mathcal{F}\) to be a \(2\)-category, which, in brief, means that as well as the objects and morphisms of \(\operatorname{Cob}_n^\mathcal{F}\) one has morphisms between the original morphisms; still more the category \(\operatorname{Vect}_{\mathbf C}\) is replaced by a \(2\)-category of \(2\)-vector spaces which is denoted by \(2\operatorname{Vect}_{\mathbf C}\), for more details cf. [\textit{J. Lurie}, in: Current developments in mathematics, 2008. Somerville, MA: International Press. 129--280 (2009; Zbl 1180.81122); \textit{D. S. Freed}, Bull. Am. Math. Soc., New Ser. 50, No. 1, 57--92 (2013; Zbl 1280.57025)]. We shall denote the functor for an extended theory by \(E\), rather than \(Z\), and \(E\) evaluates on the various manifolds as follows: for the case where \(M,\Sigma\) and \(N\) are all closed, one has \begin{align*} &\dim M=n && E(M)\in \mathbb{C}\\ &\dim \Sigma=n-1 && E(\Sigma)\in \operatorname{Vect}_{\mathbf C}\\ &\dim N=n-2 && E(N)\text{ a linear category} \end{align*} In order to display all the dimensions involved, \(\operatorname{Cob}_n^\mathcal{F}\) is now denoted by \(\operatorname{Cob}_{n,n-1,n-2}^\mathcal{F}\) so that, for the extended theory, one now writes \[E:\operatorname{Cob}_{n,n-1,n-2}^\mathcal{F}\longrightarrow 2\operatorname{Vect}_{\mathbf C}\] Now we turn to the second addition referred to above that of a relative theory. For this we need two field theories \(Z\) and \(E\), say, with \(E\) in one dimension higher than \(Z\). Let \(Z\) be a quantum field theory in dimension \(n-1\), \(E\) be an \textit{extended quantum field theory} in dimension \(n\), and \({\mathbf 1}\) be the trivial theory in dimension \(n-1\). Then \(Z\) is said to be a \textit{theory relative to \(E\)} if, when we \textit{truncate the top dimensional data in \(E\)} by including only manifolds up to dimension \(n-1\), there is a homorphism \[Z:{\mathbf 1}\longrightarrow \operatorname{Trunc}_{n-1} (E)\] where \(\operatorname{Trunc}_{n-1} (E)\) denotes the truncation of \(E\). It may help to know that a quantum field theory \(Z\) relative to the trivial theory \(\mathbf 1\) is simply a standard \(n-1\) dimensional quantum field theory -- also called an \textit{absolute} theory. With this revamped functorial machine in place an anomaly is simply defined by adding invertibility the theory \(E\) above. In other words a theory \(Z\) that is relative to an \(n\) dimensional \textit{invertible extended theory} \(E\) is called anomalous -- some authors then call the anomaly the theory \(E\), cf. [\textit{D. S. Freed}, Proc. Symp. Pure Math. 88, 25--45 (2014; Zbl 1321.81058); \textit{D. S. Freed} and \textit{C. Teleman}, Commun. Math. Phys. 326, No. 2, 459--476 (2014; Zbl 1285.81057)]. The anomaly inflow picture can now be seen, thinly veiled by this functorial definition: On the \(n\)-dimensional bulk manifold \(M\) we have an extended invertible theory \[E:\operatorname{Cob}_{n,n-1,n-2}^\mathcal{F}\longrightarrow 2\operatorname{Vect}_{\mathbf C}\] while on the boundary we have a quantum field theory \[Z:\operatorname{Cob}_{n-1}^\mathcal{F}\longrightarrow \operatorname{Vect}_{\mathbf C}\] It is necessary to incorporate a group \(G\) into these functors so as to discuss global symmetries and gauge theories. This is done using \(G\)-equivariant versions of \(\operatorname{Cob}_{n-1}^\mathcal{F}\) and \(\operatorname{Cob}_{n,n-1,n-2}^\mathcal{F}\); these are denoted by \(G\)-\(\operatorname{Cob}_{n-1}^\mathcal{F}\) and \(G\)-\(\operatorname{Cob}_{n,n-1,n-2}^\mathcal{F}\), their constructions are found in sections \(2\) and \(3\). One should now be able to verify that when anomaly inflow occurs the anomaly is absent from the combined \((E,Z)\) theory. This verification is carried out successfully in Section 4. With this TQFT outline complete we must describe the salient features of Dijkgraaf-Witten theories since they are the gauge theories studied in the article under review. A Dijkgraaf-Witten theory -- cf. [\textit{R. Dijkgraaf} and \textit{E. Witten}, Commun. Math. Phys. 129, No. 2, 393--429 (1990; Zbl 0703.58011)] Section 6 -- is a gauge theory with a finite gauge group; many topological features are now absent but there is still plenty of interesting content to be understood. For a spacetime \(M\) of dimension \(n\) with a \(G\) connection \(A\) the action \(S(A)\) of the theory is the integral of a suitably normalised Chern-Simons or secondary characteristic class -- e.g. for \(n=3\) one has the familiar formula \[ S(A)=\frac{k}{8\pi^2}\int_M tr\left(A\wedge d A +\frac{2}{3}A\wedge A\wedge A\right),\;k\in\mathbb{Z} \] and partition function \[ Z(M)=\int_\mathcal{A} \mathcal{DA}\, e^{2\pi i S(A)} \] Under a gauge transformation \(g\) where \(A\mapsto A_g\), the action \(S(A)\) is not invariant: instead \(S(A_g)=S(A)+m\), \(m\) an integer. Hence \(e^{2\pi i S(A)}\) is still single valued, however we should take \(S(A)\) itself to be \(\mathbb{R}/\mathbb{Z}\) valued. \(G\)-bundles over \(M\) are specified by choosing an element \(f\) of \([M,BG]\) where \(BG\) is the classifying space of \(G\) and, using \(f\), one finds that \(S(A)\) determines an element of \(H^n(BG;\mathbb{R}/\mathbb{Z})\). Moreover for finite groups one has the isomorphism \[ H^n(BG;\mathbb{R}/\mathbb{Z})\simeq H^{n+1}(BG;\mathbb{Z}) \] so that \(H^{n+1}(BG;\mathbb{Z})\) classifies the Chern-Simons actions for \(n\)-dimensional spacetimes \(M\) -- also it is this, cohomology \(H^*(BG;\mathbb{Z})\) of \(BG\), that is the \textit{group cohomology} of \(G\). \bigskip III. The paper. Now, turning to the paper under review, we find that the problem addressed is that of identifying 't Hooft anomalies in Dijkgraaf-Witten theories and making precise any obstruction to their removal. First we point out that, thus far, apart from the parity anomaly, we have concentrated on anomalies concerning \textit{chiral Fermions} on an \(n\)-dimensional spacetime and this meant that \(n\) was required to be even. However purely \textit{Bosonic theories} can also have anomalies, and such anomalies can occur for both odd and even dimensional spacetimes \(M\). This paper deals with anomalies of Bosonic gauge theories with no restriction on the parity of \(n=\dim M\). It is solely anomalies of the 't Hooft kind that are investigated, and the various groups involved are all finite -- however they need not always be Abelian. In any case we must specify a global symmetry group which is coupled to a connection \(A\) and look for an obstruction to gauging this group. The details are somewhat involved and present a delicate interplay involving anomaly inflow into the closed spacetime \(M\), from its interior \(N\) (thus \(M=\partial N\)); this interplay can result in cancellation of the anomaly (as is typical in Fermionic examples) but, in these Bosonic systems, examples are given where anomaly inflow cannot cancel the anomaly. We begin with a spacetime \(M\) (\(\dim M=n\)) on which one has a Dijkgraaf-Witten theory whose finite gauge group is \(D\). Let this theory have a global symmetry group \(G\) which is finite and which acts as automorphisms on the principal \(D\)-bundle -- i.e. on \([M,BD]\). When \(G\) acts on \(BD\), and thus on \(D\), because of the projectivity of the action (cf.[\textit{A. Vishwanath} and \textit{T. Senthil}, ``Physics of three dimensional Bosonic topological insulators: surface deconfined criticality and quantized magnetoelectric effect'', Phys. Rev. X 3, 011016 (2013; \url{doi:10.1103/PhysRevX.3.011016})]), there arises a cohomology class \[ c\in H^2(BG;D) \] or equivalently, since \(G\) is a finite group, a class in \(H^2(G;D)\); one knows that such classes classify extensions \(\widehat G\) of \(G\) by \(D\). The exact sequence for the extension is \[ 1\longrightarrow D\longrightarrow \widehat G \longrightarrow G \longrightarrow 1 \] and this, one might hope, should elevate the theory with group \(D\), to one with group \(\widehat G\). We note in passing that, even when \(D\) and \(G\) are Abelian, \(\widehat G\) can be non-Abelian. Hence we now have Dijkgraaf-Witten theories with groups \(D\) and \(\widehat G\), and these theories are classified by \(\omega\) and \(\widehat \omega\) respectively where \begin{align*} \omega &\in H^n(BD;\mathbb{R}/\mathbb{Z})\\ \widehat\omega&\in H^n(B\widehat G;\mathbb{R}/\mathbb{Z}) \end{align*} Still more, the inclusion \(D\longrightarrow \widehat G\) induces an embedding \[ e:BD\longrightarrow B\widehat G \] which pulls back \(\widehat\omega\) to a form \[ e^*\widehat\omega=\omega \in H^n(B\widehat G ;\mathbb{R}/\mathbb{Z}) \] But if the map \(e^*\) is not surjective then, for some \(\omega\), an element \(\widehat\omega\) does not exist. In that case one has failed to gauge the global symmetry \(\widehat G\): i.e. there is a 't Hooft anomaly. The authors point out that out that there are topological obstructions to the existence of \(\widehat \omega\) which come from the spectral sequence of the group extension. Recall that a spectral sequence, in the usual notation, is a sequence of differential groups \(\{E_r, d_r\}\) with the hierarchical property that each \(E_r\) is the (co)homology of its predecessor. If the sequence becomes stationary, then the stationary \(E_r\) is denoted by \(E_\infty\). One can regard the terms of the sequence as successive approximations to some (co)homological quantity \(H\): if \(E_\infty\) actually equals \(H\), then the spectral sequence is said to converge to \(H\). There is a particular spectral sequence developed to deal with group extensions known as the Lyndon-Hochschild-Serre sequence, or LHS sequence [\textit{L. Evens}, The cohomology of groups. Oxford etc.: Clarendon Press (1991; Zbl 0742.20050); \textit{L. Evens}, Trans. Am. Math. Soc. 212, 269--277 (1975; Zbl 0331.18022)]. Hence this is the spectral sequence used in this article; also relevant here is that the LHS sequence of an extension of \textit{finite groups} always stops after a finite number of steps. In any case, returning to topological obstructions to the existence of \(\widehat \omega\), then, using the LHS spectral sequence it is found that there are \(n\) possible topological obstructions: these are cohomology classes belonging to the list \begin{align*} &H^2(BG; H^{n-1}(BD;\mathbb{R}/\mathbb{Z}))\\ &H^3(BG; H^{n-2}(BD;\mathbb{R}/\mathbb{Z}))\\ &\qquad\qquad\vdots \\ &H^n(BG; H^1(BD;\mathbb{R}/\mathbb{Z}))\\ &H^{n+1}(BG;\mathbb{R}/\mathbb{Z}) \end{align*} It is important to realise that the hierarchical property of the spectral sequence, when examined more closely, means that the existence (or definition) of a given class in the list requires the vanishing of its predecessor. Various obstructions may vanish, but, most important here are the following two special cases. \begin{itemize} \item[(i)] All \(n\) obstructions vanish. \item[(ii)] The first \(n-1\) obstructions vanish, but not the last (this is the more interesting case). \end{itemize} If case (i) holds, then \(e^*\) is surjective and all is well: the symmetry group \(G\) can be gauged; further \(D\) and \(G\) can be combined into the extended gauge group \(\widehat G\) with a single \(\widehat G\)-connection \(A\). If case (ii) holds then there is a 't Hooft anomaly to gauging \(G\). However in case (ii), there is another fact highly relevant for condensed matter theory namely: the anomaly on the closed manifold \(M\) is that of an \textit{SPT phase} so that there is \textit{anomaly inflow} from the \(n+1\) dimensional bulk \(N\) cancelling the anomaly on its boundary \(M\). The composite theory of bulk and boundary is then anomaly free, and \(G\) can be gauged. The \(n+1\) dimensional bulk theory can be constructed and is another Dijkgraaf-Witten theory with group \(G\) so that it is specified by an element of \[ H^{n+1}(BG;\mathbb{R}/\mathbb{Z}) \] If neither (i) nor (ii) hold, then one has a 't Hooft anomaly to gauging \(G\) which cannot be cancelled by anomaly inflow and one has to examine things case by case. However there are examples where there is a non-vanishing 't Hooft anomaly which cannot be cancelled by anomaly inflow -- cf. [\textit{A. Kapustin} and \textit{R. Thorngren}, ``Anomalies of discrete symmetries in various dimensions and group cohomology'', Preprint, \url{arXiv:1404.3230}] section 4.2 where such an example is given; its content is as follows: one has \(n=3\), and only the first obstruction vanishes so that the leading obstruction occurs in second last place and thus belongs to \(H^3(BG,H^1(BD;\mathbb{R}/\mathbb{Z}))\); also , for \(p\) an odd prime, one has \(D=\mathbb{Z}_p\) and \(G=\mathbb{Z}_p\times \mathbb{Z}_p\). A key fact here, relevant to the failure of anomaly inflow, is that the gauge variation of the \(n=3\) action depends on the \(n=3\) gauge field \(A\). The authors point out that their work can be considered as a functorial formulation of the main results of [\textit{A. Kapustin} and \textit{R. Thorngren}, ``Anomalies of discrete symmetries in various dimensions and group cohomology'', Preprint, \url{arXiv:1404.3230}]; they go further however and deal with the case where the group \(D\) is non-Abelian -- for example the case \(D=D_8\) where \(D_8\) denotes the dihedral group of order \(8\); [\textit{A. Kapustin} and \textit{R. Thorngren}, ``Anomalies of discrete symmetries in various dimensions and group cohomology'', Preprint, \url{arXiv:1404.3230}] only deals with the non-Abelian cases where the extended group is given by \(\widehat G=D_8\) or \(\widehat G=Q_8\) where \(Q_8=\{\mp 1,{\mathbf \mp i, \mp j, \mp k}\}\) is the group of quaternionic units. In any case, in the present article the authors encounter non-Abelian group cocycles and have to deploy non-Abelian group cohomology; in addition they provide a detailed functorial formulation of the bulk boundary correspondence. The paper is a kind of functorial tour de force which brings with it several advantages, on the other hand performing detailed calculations requires some extra work as one has to the unpack the various functors involved.