nLab transgression




Transgression is a tool in algebraic topology to transfer cohomology classes from one space to another without needing a morphism between them. Rather, one has a third space with morphisms to each of the two spaces. Of course, not just any third space will do.

The topological set up is as follows. We have two topological spaces, XX and YY, and a generalised cohomology theory E *()E^*(-). We want to be able to transfer ( transgress ) E *E^*-classes from XX to YY. We find a third space, ZZ, which has morphisms to both XX and YY. We usually write this in the following way:

Z X Y \begin{matrix} Z & \to& X \\ \downarrow \\ Y \end{matrix}

(Note to self: replace with SVG, write f:ZXf \colon Z \to X and g:ZYg \colon Z \to Y.)

An important example is that where we have some parameter space Σ\Sigma (for instance the circle Σ=S 1\Sigma = S^1) and set Z=Σ×[Σ,X]Z = \Sigma \times [\Sigma,X], the product of Σ\Sigma with the mapping space [Σ,X][\Sigma,X] of maps from Σ\Sigma to XX (for instance the loop space of XX). Then the morphism on the right is taken to be evaluation and the morphism on the left is projection onto the mapping space [Σ,X][\Sigma,X]:

Σ×[Σ,X] ev X p 2 [Σ,X]. \array{ \Sigma \times [\Sigma, X] &\stackrel{ev}{\to}& X \\ \downarrow^{\mathrlap{p_2}} \\ [\Sigma,X] } \,.

This setup induces transgression to mapping spaces in the following.

Now we apply the generalised cohomology theory to this diagram. As it is a cohomology theory, the arrows reverse. We have part of our route from E *(X)E^*(X) to E *(Y)E^*(Y), namely from E *(X)E^*(X) to E *(Z)E^*(Z). However, the way from E *(Y)E^*(Y) to E *(Z)E^*(Z) is blocked: it is a one-way street and we want to go the wrong way.

However, under certain special circumstances, cohomology theories admit push-forward maps. That is, if g:ZYg \colon Z \to Y is particularly nice, there is a map g !:E *(Z)E *(Y)g_! \colon E^*(Z) \to E^*(Y). The notation g !g_! (pronounced “g shriek”) has the explanation that it is a surprise that there is a map in this direction1. Note that the push-forward map, g !:E *(Z)E *(Y)g_! \colon E^*(Z) \to E^*(Y), often results in a change of degree. This is described at fiber integration and Pontrjagin-Thom collapse map.

Thus the key to transgression is to understand the conditions where the map g:ZYg \colon Z \to Y admits a push-forward map. The simplest case is where ZZ is a product space, Y×WY \times W, and WW is orientable for the generalised cohomology theory E *()E^*(-). This means that WW has a fundamental class in E *(W)E^*(W) and evaluation on this class gives a morphism E *(Y×W)E *(Y)E^*(Y \times W) \to E^*(Y).


Transgression of differential forms

See at transgression of differential forms,

By pullback and fiber integration


Transgression of forms through fiber bundles

A related construction called transgression is the following

For PXP \to X a fiber bundle with typical fiber i:FPi : F \hookrightarrow P and [ω]H dR n(F)[\omega] \in H_{dR}^n(F) a class in de Rham cohomology of the fiber, we say this is transgressive if

  • there exists a form csΩ n(P)cs \in \Omega^{n}(P) on the total space of the bundle;

  • such that restricted along FPF \hookrightarrow P to the fiber it is ω\sim \omega;

  • and such that dcsd cs is the pull-back of a form κΩ n+1(X)\kappa \in \Omega^{n+1}(X) on the base along the bundle projection .

See for instance section 9 of (Borel).

This construction really exhibits transgression as a special case of the connecting homomorphism in cohomology

H n(F)H n+1(X) H^n(F) \to H^{n+1}(X)

that is induced from the short exact sequence of cocycles

ker(i *)Ω (P)i *Ω (F) ker(i^*) \hookrightarrow \Omega^\bullet(P) \stackrel{i^*}{\to} \Omega^\bullet(F)

by restricting to those elements for which it factors through

Ω (X)ker(i *). \Omega^\bullet(X) \hookrightarrow ker(i^*) \,.

For more on this see Chern-Simons element.


Transgression in de Rham Cohomology

With a cohomology theory that has a good geometric model, such as de Rham cohomology, there is often a similar geometric model for the push-forward map. In the case of de Rham cohomology, it goes by the name of integration along the fibres.

Using the notation above, with the simple case of Z=Y×WZ = Y \times W, we have the formula

αH dR k(X) Wf *αH dR kdimW(Y) \alpha \in H^k_{dR}(X) \mapsto \int_W f^* \alpha \in H^{k - \dim W}_{dR}(Y)

Transgression of de Rham cohomology to loop spaces

A particular application of this is in respect to the cohomology of loop spaces, and in particular to iterated integrals. The aim here is to transfer information from a space XX to its free loop space, LXL X. The intermediate space in this situation is S 1×LXS^1 \times L X with evaluation as the map to XX and projection to LXL X. Note that although LXL X is rather large, the fibre is S 1S^1 and it is that which needs to be controlled.

If we worked in the realm of pure algebraic topology (i.e. didn’t care about geometry) this transfer would be almost trivial. Indeed, if we worked with based loops, it would be a tautology since then the push-forward H *(S 1ΩX)H *1(ΩX)H^*(S^1 \wedge \Omega X) \to H^{*-1}(\Omega X) is simply the suspension isomorphism. (We use the smash product as we are working with based spaces here.)

However, we wish to have a geometric interpretation of this, and so work in the realm of differential topology: i.e. with smooth manifolds (albeit possibly infinite dimensional). We therefore have the diagram:

S 1×LM M LM \array{ S^1 \times L M & \to & M \\ \downarrow \\ L M }

The general lore of transgression says that this induces a map in cohomology H dR k(M)H dR k1(LM)H^k_{dR}(M) \to H^{k-1}_{dR}(L M) via

α S 1ev *α. \alpha \mapsto \int_{S^1} ev^*\alpha.

As with any cohomological construction, there is always the question as to whether or not it is purely cohomological or whether or not it can be lifted to whatever-it-was that defined the cohomology theory. In this case, that is differential forms. In this case we are in the happy circumstance that the lift exists. Expressed purely in terms of differential forms, the formula doesn’t change. However, we can think of differential forms as “that which acts on tangent vectors” and ask for a formulation that involves tangent vectors. This can make the integration step clearer.

To reformulate transgression thus we need to understand tangent vectors on LML M. The simplest way to view a tangent vector on LML M is to identify TLMT L M with LTML T M. That is, we view a tangent vector xT γLMx \in T_\gamma L M as a loop in TMT M with the property that x(t)T γ(t)Mx(t) \in T_{\gamma(t)} M. In one view (TLMT L M), xx tells γ\gamma which direction to move in; in the other view (LTML T M), each x(t)x(t) tells each γ(t)\gamma(t) which direction to move in.

That done, we have an obvious evaluation map S 1×TLMTMS^1 \times T L M \to T M, xx(t)x \mapsto x(t) and so, given a differential form on MM, we can evaluate it on tangent vectors coming from LML M.

However, that’s not all we need. There’s a very special vector field on LML M which we want to throw into the mix. This is the vector field which assigns to γLM\gamma \in L M the tangent vector γLTM\gamma' \in L T M. This is the rotation vector field: if we rotate the circle on S 1×LMS^1 \times L M then to keep the evaluation map consistent, we have to rotate the loops in LML M as well.

So starting with a kk-form, say α\alpha, on MM, we fix a loop γLM\gamma \in L M, tS 1t \in S^1, and take k1k-1 tangent vectors at γ\gamma, say x 1,,x k1x_1, \dots, x_{k-1}. We evaluate these at tt to get x 1(t),,x k1(t)T γ(t)Mx_1(t), \dots, x_{k-1}(t) \in T_{\gamma(t)} M and also throw in γ(t)\gamma'(t). That gives kk vectors at T γ(t)MT_{\gamma(t)} M on which we can evaluate α\alpha:

α(x 1(t),x 2(t),,x k1(t),γ(t)) \alpha(x_1(t),x_2(t),\dots,x_{k-1}(t),\gamma'(t))

This is a number. Hence, by varying tS 1t \in S^1, it defines a function S 1S^1 \to \mathbb{R}. Integrating this function around S 1S^1 yields a number:

S 1α(x 1(t),x 2(t),,x k1(t),γ(t))dt \int_{S^1} \alpha(x_1(t),x_2(t),\dots,x_{k-1}(t),\gamma'(t)) dt

Thus we have a map k1T(LM)\bigotimes^{k-1} T(L M) \to \mathbb{R}. It can be shown that it is alternating and (k1)(k-1)-linear, and that it varies smoothly over LML M, hence defines an element of Ω k1(LM)\Omega^{k-1}(L M).

The reason why S 1ev *α\int_{S^1} ev^*\alpha takes the form of the above formula is the following:

ev *:Ω k(M)Ω k(S 1×LM)= i+j=kΩ i(S 1)Ω j(LM)=Ω 0(S 1)Ω k(LM)Ω 1(S 1)Ω k1(LM).ev^*: \Omega^k(M) \to \Omega^k(S^1\times LM)= \oplus_{i+j=k} \Omega^i(S^1) \otimes\Omega^j(LM) = \Omega^0(S^1) \otimes \Omega^k(LM) \oplus \Omega^1(S^1)\otimes \Omega^{k-1}(LM).

Thus we may write

ev *α=α 0α k+α 1α k1,ev^*\alpha= \alpha_0\otimes \alpha_k + \alpha_{1}\otimes \alpha_{k-1},

where α kΩ k(LM)\alpha_k\in \Omega^k(LM), α k1Ω k1(LM)\alpha^{k-1}\in \Omega^{k-1}(LM), and α 1Ω 1(S 1)\alpha_1\in \Omega^1(S^1), α 0Ω 0(S 1)\alpha^0\in \Omega^0(S^1). Then contracting with S 1S^1 via S 1\int_{S^1}, only the second term survives. Thus

(1)( S 1ev *α) γ(x 1,...,x k1)= S 1α 1α k1(x 1,...,x k1). (\int_{S^1} ev^*\alpha)_{\gamma} (x_1, ..., x_{k-1}) =\int_{S^1} \alpha_{1} \otimes \alpha_{k-1} (x_1, ..., x_{k-1} ) .

On the other hand, let (v i,x i)T tS 1×T γLM(v_i, x_i) \in T_t S^1 \times T_{\gamma}LM be tangent vectors of T(S 1×LM)T(S^1\times LM) for i=1,,ki=1, \dots, k. Then

ev *α (t,γ)((v 1,x 1),,(v k,x k))=α k1| γα 1| t((v 1,x 1),,(v k,x k))+α k| γα 0| t((v 1,x 1),,(v k,x k)).ev^*\alpha_{(t, \gamma)}((v_1, x_1), \dots, (v_k, x_k)) = \alpha_{k-1}|_{\gamma} \otimes \alpha_1 |_t((v_1, x_1), \dots, (v_k, x_k)) + \alpha_k|_{\gamma}\otimes \alpha_0|_t((v_1, x_1), \dots, (v_k, x_k)).

At the same time, T γev(v,x)=vγ+xT_{\gamma} ev( v, x) = v \gamma' + x. This can be seen by taking a variation (a small path) (t+vϵ,γ ϵ)(t+v\epsilon, \gamma^\epsilon) representing (v,x)(v, x) (thus γ 0=γ\gamma^0=\gamma). Then

T γev(v,x)=ddϵ| ϵ=0(γ ϵ(t+vϵ))=ddϵ| ϵ=0γ 0(t+vϵ)+ddϵ| ϵ=0γ ϵ(t)=vγ(t)+x(t),T_{\gamma} ev( v, x)= \frac{d}{d\epsilon}|_{\epsilon=0}(\gamma^\epsilon(t+v\epsilon))=\frac{d}{d\epsilon}|_{\epsilon=0}\gamma^0(t+v\epsilon)+ \frac{d}{d\epsilon}|_{\epsilon=0}\gamma^\epsilon(t) = v\gamma'(t)+x(t),

which is a tangent vector at γ(t)\gamma(t). Thus

ev *α| (t,γ)((v 1,x 1),,(v k,x k))=α γ(t)(T γev(v 1,x 1),,T γev(v k,x k))ev^*\alpha|_{(t, \gamma)}((v_1, x_1), \dots, (v_k, x_k))= \alpha_{\gamma(t)}(T_\gamma ev (v_1, x_1), \dots, T_\gamma ev (v_k, x_k))
=α γ(t)(x 1(t),,x k(t))+α γ(t)(v 1γ(t),x 2(t),,x k(t))+c.p. = \alpha_{\gamma(t)}(x_1(t), \dots, x_k(t)) + \alpha_{\gamma(t)}(v_1 \gamma'(t), x_2(t), \dots, x_{k}(t)) + c.p.

Thus α k=α\alpha_k=\alpha, α 0=1\alpha_0=1, α k1=ι(γ)α\alpha_{k-1}=\iota(\gamma')\alpha, and α 1=dt\alpha_1=dt. Combining with equation (1), we obtain the desired formula (might be up to a sign).

There is much more to tell in this particular story. It is possible to replacing S 1S^1 by a simplex (but keeping LML M the same) and so build up more complicated objects in Ω *(LM)\Omega^*(L M). This is the starting point of Chen’s theory of iterated integrals and further developments of the theory can be found in the work of Jones, Getzler, and Petrack.

Transgression in group cohomology

See at transgression in group cohomology.

Transgression to mapping spaces in terms of internal homs

just a rough remark, for the moment

At least for some cases of transgression to mapping spaces, the concept has an nPOV description in terms of internal homs.

This makes crucial use of the nPOV notion of cohomology, as described there.

Let H=\mathbf{H} = Top \simeq ∞Grpd. For XHX \in \mathbf{H} and for Σ\Sigma a kk-dimensional closed oriented manifold, and for KK an abelian group that is injective as a \mathbb{Z}-module, we may identify transgression to the mapping space [Σ,X][\Sigma,X] as the composite

trans Σ:H(X,B nK)[Σ,]H([Σ,X],[Σ,B nK])(τ nk) *H([Σ,X],τ nk[Σ,B nK])H([Σ,X],B nkK). trans_\Sigma : \mathbf{H}(X,\mathbf{B}^n K) \stackrel{[\Sigma,-]}{\to} \mathbf{H}([\Sigma,X], [\Sigma, \mathbf{B}^n K]) \stackrel{(\tau_{n-k})_*}{\to} \mathbf{H}([\Sigma,X], \tau_{n-k} [\Sigma, \mathbf{B}^n K]) \simeq \mathbf{H}([\Sigma,X], \mathbf{B}^{n-k} K) \,.

Here the first step is application of the internal hom, then the second is postcomposition with truncation, and then the last step uses the universal coefficient theorem. Note that the transgression map trans Σtrans_\Sigma depends on the choice of an equivalence τ nk[Σ,B nK]B nkK\tau_{n-k} [\Sigma, \mathbf{B}^n K]\simeq \mathbf{B}^{n-k} K. Up to homotopy, this choice is precisely the datum of the orientation on Σ\Sigma.

more details go here….

This plays a role in the quantization process that yields FQFTs. For an application see Dijkgraaf-Witten theory.

more details, and more polishing later…


The classical notion of transgression of forms through fiber bundles is described in section 9 of

  • Armand Borel, Topology of Lie groups and characteristic classes Bull. Amer. Math. Soc. Volume 61, Number 5 (1955), 397-432. (EUCLID)

Transgression as pullback in cohomology along the evaluation map followed by fiber integration over the base space is considered in:

for differential forms:

for Deligne cohomology (and with an eye towards defining the higher holonomy of bundle gerbes with connection):

Transgression of group cohomology for discrete groups to cohomology of inertia groupoids (see at transgression in group cohomology):

The image of the transgression operation from bundle gerbes (with connection) to complex line bundles (with connection) on the free loop space of their base space is characterized (as consisting of the “fusion bundles”) in:

  1. To the best of my knowledge, this remark is attributable to Ralph Cohen.

Last revised on December 11, 2022 at 17:03:52. See the history of this page for a list of all contributions to it.