Manifolds and cobordisms

Quantum field theory



General idea

What is called a homotopy quantum field theory is a TQFT defined on cobordisms that are equipped with the extra structure of a continuous function into a fixed topological space BB.

Hence if Bord n(B)Bord_n(B) denotes a category of cobordisms suitably equipped with maps into BB, then an HQFT is a monoidal functor

Z:Bord n(B) Vect . Z \;\colon\; Bord_n(B)^{\coprod} \longrightarrow Vect^{\otimes} \,.


HQFTs were first defined (under a different name) by Graeme Segal as early as 1988 in (Segal 88).

Starting from 1991, several papers by Dan Freed, including FreedQuinn1991, further developed the notion of a functorial field theory, where bordisms in the domain category are equipped with a map to a target space (such as the Eilenberg–MacLane space K(G,1)K(G,1) for a finite group GG). See, in particular, Theorem 1.7 and the preceding discussion on page 6 of FreedQuinn1991.

From 1999, HQFTs were studied systematically by Vladimir Turaev (Turaev 99) for 2-dimensional manifolds/cobordisms and extended to 3-dimensional ones in (Turaev 00). Turaev also introduced the term “homotopy quantum field theory”.

At about the same time, (Brightwell-Turner 00) looked at what they called the homotopy surface category and its representations. There are two viewpoints which interact and complement each other. Turaev’s seems to be to see HQFTs as an extension of the tool kit for studying manifolds already given by TQFTs, whilst in Brightwell and Turner’s, it is the ‘background space’, which is probed by the surfaces in the sense of sigma-models.

In the proof of the cobordism hypothesis in (Lurie 09) the concept of HQFTs was refined to extended TQFT by considering an (∞,n)-category of cobordisms Bord n(X)Bord_n(X) with maps to a given homotopy type XX. For these the cobordism hypothesis essentially says (see at For framed cobordisms in a topological space) that Bord n(X) Bord_n(X)^\coprod is the free construction (∞,n)-category with duals on the fundamental ∞-groupoid Π(X)\Pi(X) of XX.


An HQFT is going to be defined as assigning data to BB-cobordisms. We first introduce these

and then define

themselves in terms of these.


Let BB be a pointed topological space.


A BB-manifold is a pair (X,g)(X, g), where XX is a closed oriented nn-manifold (with a choice of base point m im_i in each connected component X iX_i of XX), and gg is a continuous function g:XBg : X \to B, called the characteristic map, such that g(m i)=*g(m_i) = \ast for each base point m im_i.


A BB-isomorphism between BB-manifolds, ϕ:(X,g)(Y,h)\phi : ( X, g) \to ( Y, h) is an isomorphism ϕ:XY\phi : X \to Y of the manifolds, preserving the orientation, taking base points into base points and such that hϕ=gh\phi = g.


If as is often the case, the manifolds under consideration will be smooth manifolds and then ‘isomorphism’ is interpreted as ‘diffeomorphism’, but equally well we can position the theory in the category of PL-manifolds or triangulable topological manifolds with the obvious changes. In fact for some of the time it is convenient to develop constructions for simplicial complexes rather than manifolds, as it is triangulations that provide the basis for the combinatorial descriptions of the structures that we will be using.


Denote by Man(n,B)\mathbf{Man}(n,B) the category of nn-dimensional BB-manifolds and BB-isomorphisms. We define a ‘sum’ operation on this category using disjoint union. The disjoint union of BB-manifolds is defined by

(X,g)⨿(Y,h):=(X⨿Y,g⨿h),( X, g) \amalg ( Y, h) := ( X\amalg Y, g\amalg h),

with the obvious characteristic map, g⨿h:X⨿YBg\amalg h : X \amalg Y \to B. With this ‘sum’ operation, Man(n,B)\mathbf{Man}(n,B) becomes a symmetric monoidal category with the unit being given by the empty BB-manifold, \emptyset, with the empty characteristic map. Of course, this is an nn-manifold by default.

These BB-manifolds are the objects of interest, but they have to be related by the analogue of cobordisms for this setting.


A BB-cobordism, (W,F)(W,F), from (X 0,g)(X_0,g) to (X 1,h)(X_1,h) is a cobordism W:X 0X 1W : X_0 \to X_1 endowed with a homotopy class relative to the boundary of a continuous function F:WBF : W \to B such that

F| X 0=g,F| X 1=h F|_{X_0} = g \,, \;\;\;\; F|_{X_1} = h

Generally, unless necessary in this entry, we will not make a notational distinction between the homotopy class FF and any of its representatives.


A BB-isomorphism of BB-cobordisms, ψ:(W,F)(W ,F )\psi : (W,F) \to (W^\prime, F^\prime), is an isomorphism ψ:WW \psi : W \to W^\prime such that

ψ( +W)= +W ,ψ( W)= W ,\psi (\partial_+W) = \partial_+W^\prime, \quad \psi (\partial_-W) = \partial_-W^\prime,

and F ψ=FF^\prime \psi = F, in the obvious sense of homotopy classes relative to the boundary.

We can glue BB-cobordisms along their boundaries, or more generally, along a BB-isomorphism between their boundaries, in the usual way. This gives rise to a symmetric monoidal category HCobord(n,B)\mathbf{HCobord}(n,B) of BB-cobordisms

The detailed structure of BB-cobordisms and the resulting category HCobord(n,B)\mathbf{HCobord}(n,B) is given in (Rodrigues 03, appendix), at least in the important case of smooth BB-manifolds. This category is a monoidal category with strict dual objects.

Homotopy Quantum Field Theories

The general absract definition of an HQFT is now the following.

Fix an integer n0n \geq 0 and a field, KK. All vector spaces will be tacitly assumed to be finite dimensional. In general KK can be replaced by a commutative ring merely by replacing finite dimensional vector spaces by projective KK-modules of finite type, but we will not do this here.


A homotopy quantum field theory is a symmetric monoidal functor from HCobord(n,B)\mathbf{HCobord}(n,B) to the category, Vect, of finite dimensional vector spaces over the field KK.

This definiting unwinds to the following structure in components


A (n+1)(n + 1)-dimensional homotopy quantum field theory, τ\tau, with background BB assigns

  • to any nn-dimensional BB-manifold, (X,g)(X,g), a vector space, τ(X,g)\tau{(X,g)},

  • to any BB-isomorphism, ϕ:(X,g)(Y,h)\phi : (X, g) \to ( Y, h), of nn-dimensional BB-manifolds, a KK-linear isomorphism τ(ϕ):τ(X,g)τ(Y,h)\tau(\phi) : \tau{(X, g)} \to \tau{( Y, h)},


  • to any BB-cobordism, (W,F):(X 0,g 0)(X 1,g 1)(W,F) : (X_0,g_0) \to (X_1,g_1), a KK-linear transformation, τ(W):τ(X 0,g 0)τ(X 1,g 1)\tau(W) : \tau{(X_0,g_0)} \to \tau{(X_1,g_1)}.

These assignments are to satisfy the following axioms:

  1. τ\tau is functorial in Man(n,B)\mathbf{Man}(n,B), i.e., for two BB-isomorphisms, ψ:(X,g)(Y,h)\psi: (X, g) \to ( Y, h) and ϕ:(Y,h)(P,j)\phi : ( Y, h) \to (P,j), we have τ(ϕψ)=τ(ϕ)τ(ψ),\tau(\phi\psi) = \tau(\phi)\tau(\psi), and if 1 (X,g)1_{(X,g)} is the identity BB-isomorphism on (X,g)(X,g), then τ(1 (X,g))=1 τ(X,g)\tau(1_{(X,g)}) = 1_{\tau{(X,g)}}

  2. There are natural isomorphisms

    c (X,g),(Y,h):τ((X,g)⨿(Y,h))τ(X,g)τ(Y,h),c_{(X,g),(Y,h)} : \tau((X,g)\amalg (Y,h)) \cong \tau(X,g)\otimes \tau(Y,h),

    and an isomorphism, u:τ()Ku : \tau(\emptyset) \cong K, that satisfy the usual axioms for a symmetric monoidal functor.

  3. For BB-cobordisms, (W,F):(X,g)(Y,h)(W,F) : (X,g) \to (Y,h) and (V,G):(Y ,h )(P,j)(V,G): (Y^\prime, h^\prime) \to (P,j) glued along a BB-isomorphism ψ:(Y,h)(Y ,h )\psi :(Y,h) \to (Y^\prime,h^\prime), we have τ((W,F)⨿ ψ(V,G))=τ(V,G)τ(ψ)τ(W,F).\tau((W,F)\amalg_\psi (V,G))= \tau(V,G)\tau(\psi)\tau(W,F).

  4. For the identity BB-cobordism, 1 (X,g)=(I×X,1 g)1_{(X,g)} = (I\times X, 1_g), we have τ(1 (X,g))=1 τ(X,g).\tau( 1_{(X,g)}) = 1_{\tau(X,g)}.

  5. For BB-cobordisms (W,F):(X,g)(Y,h)(W,F) : (X,g) \to (Y,h) and (V,G):(X ,g )(Y ,h )(V,G) : (X^\prime,g^\prime) \to (Y^\prime,h^\prime) and (P,J):(P,J): \emptyset \to \emptyset, some fairly obvious diagrams are commutative.


These axioms are slightly different from those given in the original paper of Turaev in 1999. The really significant difference is in axiom 4, which is weaker than as originally formulated, where any BB-cobordism structure on I×XI \times X was considered as trivial. The effect of this change is important as it is now the case that the HQFT is determined by the (n+1)(n+1)-type of BB, cf. (Rodrigues 03).

With the revised version of the axioms, it becomes possible to attempt to classify HQFTs with a given nn and BB. Turaev did this in the original paper with n=2n = 2 and BB an Eilenberg-MacLane space, K(G,1)K(G,1). The results of Brightwell and Turner essentially gave the solution for BB a K(A,2)K(A,2).


1+1 dimensional HQFTs with background a K(G,1)K(G,1)

If we look at the case n=1n= 1 and with background an Eilenberg-Mac Lane space K(G,1)K(G,1) for a discrete group GG, then HQFTs correspond to crossed G-algebras, in much the same way that commutative Frobenius algebras correspond to 2d TQFTs. There the correspondence is given by a 2d TQFT, ZZ, corresponds to the Frobenius algebra, Z(S 1)Z(S^1). This is because the circle S 1S^1 is a Frobenius algebra, sometimes called a Frobenius object, in the category Bord 2Bord_2 of 2d-cobordisms between 1-manifolds.

In the case of HQFTs, the role of the circle is replaced by the family of circles with characteristic maps to BB. Each one gives, combinatorially, a circle together with a labelling of the boundary by an element of GG. (It does not seem to be known how to get a GG-graded version of an abstract Frobenius object that will correspond to this situation, although this is probably not too hard to do.)


In (Moore-Segal 06) are discussed GG-equivariant TFT?s and it is shown that they naturally correspond to a simple case of Turaev’s HQFTs. They relate (1+1) equivariant TFTs to Turaev’s crossed G-algebras (which they call Turaev algebras).


The original definition is due to Graeme Segal, who introduced them under the term elliptic objects?. Specifically, in Segal 88, we read in §6:

6. Speculation about the definition of elliptic cohomology

For any space XX let 𝒫 X\mathcal{P}_X be the category whose objects are the points of XX and whose morphisms from x 0x_0 to x 1x_1 are the paths in XX from x 0x_0 to x 1x_1, two such paths being identified if they differ only by reparametrization. A functor from 𝒫 X\mathcal{P}_X to finite dimensional vector spaces is essentially the same thing as a vector bundle on XX with a connection. (The functor must be continuous in a suitable sense.) It is well known how K-theory is constructed from such objects.

I have described elsewhere [23] a category 𝒞\mathcal{C} whose objects are all compact oriented one-dimensional manifolds, and whose morphisms from S 0S_0 to S 1S_1 are pairs (Σ,α)(\Sigma,\alpha), where Σ\Sigma is a Riemann surface with boundary Σ\partial\Sigma, and α\alpha is an isomorphism between X\partial X and S 1S 0S_1-S_0. Two pairs (Σ,α)(\Sigma,\alpha), (Σ,α)(\Sigma',\alpha') are identified if they are isomorphic. For any space XX one can now define a category 𝒞 X\mathcal{C}_X. Its objects are pairs (S,s)(S,s), where SS is an object of 𝒞\mathcal{C} and s:SXs \colon S \to X is a map. Its morphisms from (S 0,s 0)(S_0,s_0) to (S 1,s 1)(S_1,s_1) are triples (Σ,α,σ)(\Sigma,\alpha,\sigma), where (Σ,α):S 0S 1(\Sigma,\alpha)\colon S_0 \to S_1 is a morphism in 𝒞\mathcal{C}, and σ:ΣX\sigma\colon\Sigma\to X is a map compatible with (s O,s 1)(s_O,s_1). The category 𝒞 X\mathcal{C}_X is a natural analogue of the category 𝒫 X\mathcal{P}_X which gives rise to vector bundles.

It is appropriate to consider functors from 𝒞\mathcal{C} to the category 𝒱\mathcal{V} of topological vector spaces and trace-class maps. If such a functor EE is holomorphic in the natural sense then E(S 1)E(S^1) is a positive energy representation of Diff(S 1)Diff(S^1) of finite type. More precisely, as is familiar in the representation theory of Diff(S 1)Diff(S^1), one must consider projective representations of 𝒞\mathcal{C} of some definite positive integral level kk. Imposing a further condition — the contraction condition below — on the functor EE ensures that the character of E(S 1)E(S^1) is a modular form of weight kk.

Now let us define an elliptic object of level kk on XX as a projective functor E:𝒞 X𝒱E\colon\mathcal{C}_X\to \mathcal{V} of level kk which is holomorphic and satisfies the contraction condition. Such an object consists of an infinite dimensional vector bundle on the loop space X\mathcal{L}X, equivariant under Diff(S 1)Diff(S^1), together with some additional data amounting to a kind of connection. The primary example is the spin bundle of X\mathcal{L}X, which is defined when XX is a spin manifold with p 1=0p_1=0.

I have nothing precise to say about elliptic objects, but it seems to me quite likely that the objects of each level lead to an interesting cohomology theory, and that the theories for different levels are related by “Bott maps”. That would fit in well with Theorem (5.3), for just as elements of K(BG)K(B G) are elated to flat bundles, so elements of Ell *(BG)Ell^*(B G) seem to be related to flat elliptic objects, i.e. ones such that the operator associated to (Σ,α,σ)(\Sigma,\alpha,\sigma) depends on σ\sigma only up to homotopy, and is therefore a homomorphism π 1(Σ)G\pi_1(\Sigma)\to G.

I should mention that the category 𝒞\mathcal{C} can be modified by equipping the Riemann surfaces Σ\Sigma with chosen spin structures. That is certainly needed to obtain genuine elliptic cohomology.

Finally I return to the “contraction property”. This is motivated by the path-integral point of view. If a surface Σ\Sigma is a morphism from SS to itself then the trace of the operator E(Σ):E(S)E(S)E(\Sigma)\colon E(S)\to E(S) associated to Σ\Sigma must depend only on the closed surface Σˇ\check\Sigma obtained by attaching the two boundary pieces of Σ\Sigma to each other. Thus if Σ τ\Sigma_\tau is the annulus {zC|e iτ||z|1}\{z \in \mathbf{C} \mid |e^{i\tau}|\le |z|\le 1\} then Σˇ τΣˇ τ\check\Sigma_\tau\cong \check\Sigma_{\tau'}, when τ=1/τ\tau'=-1/\tau, and therefore the trace of E(Σ τ)E(\Sigma_\tau) is invariant under τ1/τ\tau\mapsto -1/\tau.

Brylinski [9] has proposed a similar approach to elliptic cohomology.

Postscript. After giving this talk, I learnt of the work [29] 1 which gives a good account of the Dirac operator on loop space from the path integral point of view.


The original definition is due to Graeme Segal (who introduced them under the name (flat) elliptic objects?), see §6 of

These ideas were further developed in

The theory of HQFTs was developed in

  • Vladimir Turaev, Homotopy field theory in dimension 2 and group-algebras (arXiv:math.QA/9910010)

  • Vladimir Turaev, Homotopy field theory in dimension 3 and crossed group-categories (arXiv:math.GT/0005291).

  • Vladimir Turaev, Homotopy Quantum Field Theory, Tracts in Mathematics 10, (with Appendices by Michael Muger and Alexis Vurelizier), European Mathematical Society, June 2010.

  • M. Brightwell and P. Turner, Representations of the homotopy surface category of a simply connected space, J. Knot Theory and its Ramifications, 9 (2000), 855–864.

  • G. Rodrigues, Homotopy Quantum Field Theories and the Homotopy Cobordism Category in Dimension 1 + 1, J. Knot Theory and its Ramifications, 12 (2003) 287–317 (arXiv:math.QA/0105018).

  • T. Porter and V. Turaev, Formal Homotopy Quantum Field Theories, I: Formal Maps and Crossed CC-algebras, Journal of Homotopy and Related Structures 3(1), 2008, 113–159. (arXiv:math.QA/0512032).

  • Tim Porter, Formal Homotopy Quantum Field Theories II: Simplicial Formal Maps, Cont. Math. 431, p. 375 - 404 (Streetfest volume: Categories in Algebra, Geometry and Mathematical Physics - edited by A. Davydov, M. Batanin, and M. Johnson, S. Lack, and A. Neeman) (arXiv:math.QA/0512034)

A treatment of HQFTs that includes some details of the links with TQFTs is given in HQFTs meet the Menagerie, which is a set of notes prepared by Tim Porter for a school and workshop in Lisbon, Feb. 2011.

Related ideas are discussed in

See also:

Understanding higher parallel transport of circle n-bundles with connection as an extended homotopy field theory:

Last revised on June 29, 2023 at 09:35:39. See the history of this page for a list of all contributions to it.