Deligne cohomology


Differential cohomology

(,1)(\infty,1)-Topos Theory

(∞,1)-topos theory





Extra stuff, structure and property



structures in a cohesive (∞,1)-topos



Deligne cohomology – or Deligne-Beilinson cohomology – is an abelian sheaf cohomology that models ordinary differential cohomology.

The Deligne complex is like a truncated de Rham complex but, crucially, with the sheaf of 0-forms – the structure sheaf 𝒪\mathcal{O} - replaced by the multiplicative group 𝒪 ×\mathcal{O}^\times under the exponential map

[𝒪 ×dlogΩ 1dΩ 2ddΩ n]. \left[ \mathcal{O}^\times \stackrel{d log}{\longrightarrow} \Omega^1 \stackrel{d}{\longrightarrow} \Omega^2 \stackrel{d}{\longrightarrow} \cdots \stackrel{d}{\longrightarrow} \Omega^n \right] \,.

Deligne cohomology H conn n+1(X,)H^{n+1}_{conn}(X, \mathbb{Z}) in degree (n+1)(n+1) is the abelian sheaf cohomology with coefficients in this chain complex of sheaves of abelian groups (“hypercohomology”).

This was introduced in (Deligne 71) in the context of analytic geometry (hence using holomorphic differential forms) as a Hodge-filtered version of singular cohomology, designed to be a target for the Beilinson regulator from motivic cohomology. But the form of the definition applies more generally, in particular also in smooth differential geometry, a fact amplified and popularized in (Brylinski 93).

In smooth differential geometry the typical minor variant has the sheaf U̲(1)=C (,U(1))\underline{U}(1) = C^\infty(-,U(1)) of circle group-valued smooth functions in degree nn:

[C (,U(1))dlogΩ 1dΩ 2ddΩ n]. \left[ C^\infty(-,U(1)) \stackrel{d log}{\longrightarrow} \Omega^1 \stackrel{d}{\longrightarrow} \Omega^2 \stackrel{d}{\longrightarrow} \cdots \stackrel{d}{\longrightarrow} \Omega^n \right] \,.

Given any manifold XX, then the resulting complex of abelian groups is, under the Dold-Kan correspondence, the n-groupoid of circle n-bundles with connection whose underlying circle (n-1)-group-principal infinity-bundle is trivialized. Passing to the abelian sheaf cohomology implicitly corresponds to considering the infinity-stackification of this nn-groupoid valued presheaf, and in this way Deligne cohomology computes equivalence classes of circle n-bundles with connection. Another way to say this is that under the Dold-Kan correspondence and infinity-stackification, the above Deligne complex defines a smooth infinity-stack B nU(1) conn\mathbf{B}^n U(1)_{conn} which is the moduli infinity-stack for circle n-bundles with connection, and Deligne cohomology computes the homotopy classes of maps (of infinity-stacks) into this (FSS 10)

H conn n+1(X,)π 0(XB nU(1) conn). H^{n+1}_{conn}(X,\mathbb{Z}) \simeq \pi_0(X \to \mathbf{B}^n U(1)_{conn}) \,.

In this way Deligne cohomology, or rather the collection of Deligne cocycles with coefficients in the Deligne complex that defines it, is considerably richer than other models for ordinary differential cohomology such as Cheeger-Simons differential characters, which see only the cohomology group, but not the full moduli n-stack.

Explicitly, computing the abelian sheaf cohomology with coefficients in the Deligne complex via Cech cohomology gives that a cocycle A¯\overline{A} on some space XX is represented with respect to a suitable covering {U iX}\{U_i \to X\} by a collection of differential forms and functions

A¯={A i 0,,i kΩ nk(U i 0,i k)} k=0 n{g i 0,,i n𝒪 ×(U i 0,,i n+1)} \overline{A} = \left\{ A_{i_0, \cdots, i_k} \in \Omega^{n-k}(U_{i_0, \cdots i_k}) \right\}_{k = 0}^{n} \cup \{ g_{i_0, \cdots, i_n} \in \mathcal{O}^\times(U_{i_0, \cdots, i_{n+1}}) \}

such that the failure of the (nk+1)(n-k+1)-forms to glue on (k+1)(k+1)-fold intersections of charts is given by the de Rham differential of the (nk)(n-k)-forms

j=0 k(1) jA i 0,,i j1,i j+1,,i k+1=d dRA i 0,,i k+1. \sum_{j = 0}^k (-1)^j A_{i_0, \cdots, i_{j-1}, i_{j+1}, \cdots, i_{k+1}} = d_{dR} A_{i_0, \cdots, i_{k+1}} \,.

This evidently generalizes the familiar Cech cocycle data for traditional line bundles with connection.

As the notation indicates, Deligne cohomology is a differential cohomology refinement of ordinary cohomology with integer coefficients, exhibited by a canonical forgetful map

H conn n+1(X,) H n+1(X,) \array{ H^{n+1}_{conn}(X,\mathbb{Z}) \\ & \searrow \\ && H^{n+1}(X,\mathbb{Z}) }

which is induced by the evident morphism of chain complexes. This is one map in an exact differential hexagon which exhibits Deligne cohomology as the differential refinement of ordinary integral cohomology by closed curvature differential form data.

0 0 Ω n(X)/Ω int n(X) d Ω cl n+1(X) a F () H n(X,) H conn n+1(X,) H n+1(X,) DD H n(X,U(1)) β H n+1(X,) 0 0 connectionformsontrivialbundles deRhamdifferential curvatureforms curvature deRhamtheorem flatdifferentialforms geometricbundleswithconnection rationalizedbundles topol.class Cherncharacter flatconnections Bocksteinhomomorphism shapeofbundle \array{ 0 & && && && & 0 \\ & \searrow && && && \nearrow \\ && \Omega^{n}(X)/\Omega^n_{int}(X) && \stackrel{\mathbf{d}}{\longrightarrow} && \Omega^{n+1}_{cl}(X) \\ & \nearrow && \searrow^{\mathrlap{a}} && {}^{\mathllap{F_{(-)}}}\nearrow && \searrow \\ H^{n}(X, \mathbb{R}) && && H^{n+1}_{conn}(X,\mathbb{Z}) && && H^{n+1}(X,\mathbb{R}) \\ & \searrow && \nearrow && \searrow^{\mathrlap{DD}} && \nearrow \\ && H^{n}(X,U(1)) && \underset{\beta}{\longrightarrow} && H^{n+1}(X,\mathbb{Z}) \\ & \nearrow && && && \searrow \\ 0 & && && && & 0 \\ \\ && {{connection\;forms} \atop {on\;trivial\;bundles}} && \stackrel{de\;Rham\;differential}{\longrightarrow} && {{curvature} \atop {forms}} \\ & \nearrow & & \searrow & & \nearrow_{\mathrlap{curvature}} && \searrow^{\mathrlap{de\;Rham\;theorem}} \\ {{flat} \atop {differential\;forms}} && && {{geometric\;bundles} \atop {with \;connection}} && && {{rationalized} \atop {bundles}} \\ & \searrow & & \nearrow & & \searrow^{\mathrlap{topol.\;class}} && \nearrow_{\mathrlap{Chern\;character}} \\ && {{flat} \atop {connections}} && \underset{Bockstein\,homomorphism}{\longrightarrow} && {{shape} \atop {of\;bundle}} }


In any context where these symbols make the evident sense, the Deligne complex of degree (n+1)(n+1) is the chain complex 𝒪 ×dlogΩ 1dΩ 2Ω n\mathcal{O}^\times \stackrel{d log}{\to}\Omega^1 \stackrel{d}{\to} \Omega^2 \to \cdots \to \Omega^n, and Deligne cohomology in degree (n+1)(n+1) is the abelian sheaf cohomology with coefficients in this complex.

More generally one considers any discrete group AA and inclusion A𝒪A \hookrightarrow \mathcal{O} into the structure sheaf, then the corresponding Deligne complex is A𝒪dΩ 1Ω nA \hookrightarrow \mathcal{O} \stackrel{d}{\to} \Omega^1 \to \cdots \to \Omega^n.

For definiteness we consider here in detail the Deligne complex in the context of differential geometry modeled on smooth manifolds. All variants work essentially directly analogously, but it may be useful to have a specific case in hand. This discussion overlaps with and is put into a broader context at geometry of physics -- principal connections.

Preliminaries on sheaf cohomology

In order to be somewhat self-contained, this section reviews some elements of abelian sheaf cohomology specified to the context that we need. It also sets up some notation. The definition of the Deligne complex itself is below in def. 6.


Write CartSp for the site whose

Write PSh(CartSp)=Func(CartSp op,Set)PSh(CartSp) = Func(CartSp^{op},Set) for the category of presheaves over this site. Write

Smooth0TypeSh(CartSp) Smooth0Type \coloneqq Sh(CartSp)

for its category of sheaves, also called the cohesive topos of smooth spaces.


Instead of the site CartSp of def. 1 one could use the site SmoothMfd of all smooth manifolds. All of the statements and constructions in the following go through in that case just as well. In fact CartSp is a dense subsite of SmoothMfd. On the one hand this implies that the abelian sheaf cohomology is the same for both sites, but on the other hand means that it is convenient to restrict to the much “smaller” site of Cartesian spaces. In fact since the stalks of sheaves over smooth manifolds are evaluations on small open balls and since every open ball is diffeomorphic to a Cartesian space, many statements that are true (only) stalkwise over SmoothMfd are actually true globally over CartSpCartSp. It is the “descent” or “infinity-stackification” which is implicit in abelian sheaf cohomology that takes care of these global statements over CartSp to translate into the same local statements as one gets over SmoothMfd.


The assignment C (,): nC ( n,)C^\infty(-,\mathbb{R}) \colon \mathbb{R}^n \mapsto C^\infty(\mathbb{R}^n,\mathbb{R}) of smooth functions with values in the real numbers is a sheaf. Since this is representable we are entitled to identify this with the smooth manifold \mathbb{R} (the real line) itself, and just write Sooth0Type\mathbb{R} \in Sooth0Type.

Similarly for XX any other smooth manifold, it represents a sheaf on CartSp and we just write XSmooth0TypeX \in Smooth0Type for this.

Of particular interest below is the case where X=S 1=/=U(1)X = S^1 = \mathbb{R}/\mathbb{Z} = U(1) is the circle, to be regarded as the circle group.

Notice that traditionally the sheaf represented by \mathbb{R} or U(1)U(1) is indicated by an underline as in ̲\underline{\mathbb{R}} and U̲(1)\underline{U}(1), but we do not follow this tradition here.

Instead, if we consider the other sheaf that might deserve to be denoted by \mathbb{R}, namely the constant sheaf on \mathbb{R}, which sends each UCartSpU \in CartSp to the set underlying \mathbb{R}, then we write \flat \mathbb{R} for that. Similarly

U(1)Smooth0Type \flat U(1) \in Smooth0Type

is the sheaf sending each test manifold to the set of points in the circle, and each smooth function between Cartesian spaces to the identity function on that set.


For kk \in \mathbb{N} write

Ω kSh(CartSp) \mathbf{\Omega}^k \in Sh(CartSp)

for the sheaf Ω k:UΩ k(U)\mathbf{\Omega}^k \colon U \mapsto \Omega^k(U) of smooth differential k-forms on XX. The de Rham differential extends to a morphism of sheaves

d:Ω kΩ k+1. \mathbf{d} \colon \mathbf{\Omega}^k \to \mathbf{\Omega}^{k+1} \,.

For positive kk its kernel is the sub-sheaf

0Ω cl kΩ kdΩ k+1 0 \to \mathbf{\Omega}^k_{cl} \hookrightarrow \mathbf{\Omega}^k \stackrel{\mathbf{d}}{\longrightarrow} \mathbf{\Omega}^{k+1}

of closed differential forms; and for k=0k = 0 its kernel is the sub-sheaf of constant functions

0dΩ 1. 0 \to \flat \mathbb{R} \hookrightarrow \mathbb{R} \stackrel{\mathbf{d}}{\longrightarrow} \mathbf{\Omega}^1 \,.

In the background, what plays a role for the following is the full cohesive homotopy theory of smooth ∞-groupoids. This receives a map from the following coarse homotopy theory of chain complexes of abelian sheaves, which is all that is necessary for the present purpose.



Ch +(Smooth0Type)=Ch +(Sh(CartSp))=Sh(CartSp,Ch +) Ch_+(Smooth0Type) = Ch_+(Sh(CartSp)) = Sh(CartSp,Ch_+)

for the category of chain complexes in the smooth sheaves of def. 1, hence for the 1-category whose objects are chain complexes of abelian sheaves on CartSpCartSp.

Regard this as equipped with the structure of a category of fibrant objects induced by the projective model structure on chain complexes, hence with classes of morphisms labeled as follows: a chain map f :A B f_\bullet \colon A_\bullet \to B_\bullet is called


For our purpose the main use of this structure is to compute homotopy fibers via the factorization lemma. Namely

  1. every chain map may be replaced, up to weak equivalence of its domain, by a fibration;

  2. the homotopy fiber of a chain map is the ordinary fiber of any of its fibration replacements.


That the properties in def. 2 are interpreted in sheaves simply means that they apply stalk-wise. For instance a morphism of chain complexes of presheaves f :A B f_\bullet \colon A_\bullet \to B_\bullet is a weak equivalence precisely if the underlying presheaf of chain complexes becomes a quasi-isomorphism for each point xx in each Cartesian space n\mathbb{R}^n after restricting (via the presheaf structure maps) to a small enough open neighbourhood of that point. Similarly for epimorphisms.


There is a canonical map of homotopy theories from Ch +(Smooth0Type)Ch_+(Smooth0Type) to the full (∞,1)-topos Smooth∞Grpd which is given by applying the Dold-Kan correspondence followed by ∞-stackification. The key point is that this map preserves homotopy fiber products, which is the universal construction that already captures most of the relevant properties of the Deligne complex. In this way it is sufficient to concentrate on Ch +(Smooth0Type)Ch_+(Smooth0Type) for much of the theory.

When writing out the components of chain complexes we will use square brackets always denote the group in degree-0 to the far right, and the group in degree kk being kk steps to the left from that.


For AAb(Smooth0Type)=Sh(CartSp,Ab)A \in Ab(Smooth0Type) = Sh(CartSp,Ab) any abelian sheaf and for nn \in \mathbb{N} we write

(B nA) A[n]=[A00] (\mathbf{B}^n A)_{\bullet} \coloneqq A[-n] = \left[ A \to 0 \to \cdots \to 0 \right]

for the chain complex of sheaves concentrated on AA in degree nn.


There is a weak equivalence, def. 2,

[]U(1) \left[\mathbb{Z}\to \mathbb{R} \right] \stackrel{\simeq}{\longrightarrow} U(1)

given by the chain map

mod 0 U(1) \array{ \mathbb{Z} &\hookrightarrow& \mathbb{R} \\ \downarrow && \downarrow^{\mathrlap{mod\,\mathbb{Z}}} \\ 0 &\to& U(1) }

(which is just the exponential sequence regarded as a chain map). That this is a weak equivalence is the statement that every smooth U(1)U(1)-valued function is locally the quotient of a smooth \mathbb{R}-valued function by a \mathbb{Z}-valued function. In fact on Cartesian spaces this is of course true even globally.

The de Rham differential extends through this equivalence to produce a morphism denoted dlog\mathbf{d} log:

() d Ω 1 dlog U(1). \array{ (\mathbb{Z} \to \mathbb{R}) &\stackrel{\mathbf{d}}{\longrightarrow}& \mathbf{\Omega}^1 \\ \downarrow^{\mathrlap{\simeq}} & \nearrow_{\mathrlap{\mathbf{d} log}} \\ U(1) \,. }

On a given U(1)U(1)-valued function this is given by representing the function by a smooth \mathbb{R}-valued function under mod-\mathbb{Z}-reduction (which is always possible over a Cartesian space) and applying the de Rham differential to that.

The kernel of that is the constant sheaf U(1)\flat U(1) of example 1

0U(1)U(1)dlogΩ 1. 0 \to \flat U(1) \hookrightarrow U(1) \stackrel{\mathbf{d} log}{\longrightarrow} \mathbf{\Omega}^1 \,.

Under addition of differential forms, the sheaves Ω k\mathbf{\Omega}^k of example 2 becomes abelian sheaves, and we will implicitly understand them this way now.


Write (B n+1)^ Ch(Smooth0Type)\widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet \in Ch(Smooth0Type) for the complex of sheaves given by the truncated de Rham complex:

(B n+1)^ [dΩ 1dΩ 2ddΩ cl n+1]. \widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet \coloneqq \left[ \mathbb{R} \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^1 \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^2 \stackrel{\mathbf{d}}{\to} \cdots \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^{n+1}_{cl} \right] \,.

The morphism

(B n+1) (B n+1)^ (\flat \mathbf{B}^{n+1}\mathbb{R})_\bullet \longrightarrow \widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet

given by the canonical chain map

0 0 0 d Ω 1 d Ω 2 d d Ω cl n+1 \array{ \flat \mathbb{R} &\stackrel{}{\to}& 0 &\stackrel{}{\to}& 0 &\stackrel{}{\to}& \cdots &\stackrel{}{\to}& 0 \\ \downarrow && \downarrow && \downarrow && \cdots && \downarrow \\ \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^2 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n+1}_{cl} }

is a weak equivalence in the sense of def. 2.


By the Poincaré lemma. This is the Poincaré Lemma.

Every A Ch +(Smooth0Type)A_\bullet \in Ch_+(Smooth0Type) serves as the coefficients for an abelian sheaf cohomology theory on smooth manifolds. Abelian sheaf cohomology has a general abstract characterization (see at cohomology) in terms of derived hom-spaces. For definiteness, we recall the model for this construction given by Cech cohomology .


(Čech complex)

Let XX be a smooth manifold and let A Ch +(Smooth0Type)A_\bullet \in Ch_+(Smooth0Type) be a sheaf of chain complexes. Let {U iX}\{U_i \to X\} be a good open cover of XX, i.e. an open cover such that each finite non-empty intersection U i 0,,i kU_{i_0, \cdots, i_k} is diffeomorphic to an open ball/Cartesian space.

The Čech cochain complex C ((X,{U i}),A )C^\bullet((X,\{U_i\}),A_\bullet) of XX with respect to the cover {U iX}\{U_i \to X\} and with coefficients in A A_\bullet is in degree kk \in \mathbb{N} given by the abelian group

C k((X,{U i}),A ) l,nk=nl i 0,i 1,,i nA l(U i 0,,i n) C^k((X,\{U_i\}),A_\bullet) \coloneqq \oplus_{{l,n} \atop {k = n-l}} \oplus_{i_0, i_1, \cdots, i_n} A_l(U_{i_0, \cdots, i_n})

which is the direct sum of the values of A A_\bullet on the given intersections as indicated; and whose differential

d:C k((X,{U i}),A )C k+1((X,{U i}),A ) d \colon C^{k}((X,\{U_i\}),A_\bullet) \longrightarrow C^{k+1}((X,\{U_i\}),A_\bullet)

is defined componentwise (see at matrix calculus for conventions on maps between direct sums) by

(da) i 0,,i k+1 ( Aa+(1) kδa) i 0,,i k+1 Aa i 0,,i k+1+(1) k 0jk+1(1) ja i 0,,i j1,i j+1,,i k+1| U i 0,,i k+1 \begin{aligned} (d a)_{i_0, \cdots, i_{k+1}} & \coloneqq (\partial_A a + (-1)^k \delta a)_{i_0, \cdots, i_{k+1}} \\ & \coloneqq \partial_A a_{i_0, \cdots, i_{k+1}} + (-1)^k \sum_{0 \leq j \leq k+1} (-1)^{j} a_{i_0, \cdots, i_{j-1}, i_{j+1}, \cdots, i_{k+1}} |_{U_{i_0, \cdots, i_{k+1}}} \end{aligned}

where on the right the sum is over all components of aa obtained via the canonical restrictions obtained by discarding one of the original (k+1)(k+1) subscripts.

The Cech cohomology groups of XX with coefficients in A A_\bullet relative to the given cover are the chain homology groups of the Cech complex

H Cech k((X,{U i}),A )H k(C ((X,{U i}),A )). H_{Cech}^k((X,\{U_i\}), A_\bullet) \coloneqq H^k(C^\bullet((X,\{U_i\}),A_\bullet)) \,.

The Cech cohomology groups as such are the colimit (“direct limit”) of these groups over refinements of covers

H Cech k(X,A )lim {U iX}H Cech k((X,{U i}),A ). H^k_{Cech}(X, A_\bullet) \coloneqq \underset{\longrightarrow}{\lim}_{\{U_i \to X\}} H_{Cech}^k((X,\{U_i\}), A_\bullet) \,.

Often Cech cohomology is considered for the case that A A_\bullet is concentrated in a single degree, in which case the first term in the sum defining the differential in def. 4 disappears. When A A_\bullet is not concentrated in a single degree, then for emphasis one sometimes speaks of hypercohomology. This is the case of relevance for Deligne cohomology.


The Cech chain complex in def. 4 is the total complex of the double complex whose vertical differential is that of A A_\bullet and whose horizontal differential is the Cech differential δ\delta given by alternating sums over restrictions along patch inclusions

A A iA 1(U i 0) δ i 0,i 1A 1(U i 0,i 1) δ A A iA 0(U i) δ i 0,i 1A 0(U i 0,i 1) δ \array{ \vdots && \vdots \\ \downarrow^{\mathrlap{\partial_A}} && \downarrow^{\mathrlap{\partial_A}} \\ \oplus_i A_1(U_{i_0}) &\stackrel{\delta}{\longrightarrow}& \oplus_{i_0, i_1} A_1(U_{i_0, i_1}) &\stackrel{\delta}{\longrightarrow}& \cdots \\ \downarrow^{\mathrlap{\partial_A}} && \downarrow^{\mathrlap{\partial_A}} \\ \oplus_i A_0(U_i) &\stackrel{\delta}{\longrightarrow}& \oplus_{i_0, i_1} A_0(U_{i_0, i_1}) &\stackrel{\delta}{\longrightarrow}& \cdots }

For analyzing the properties of Deligne cohomology below, all one needs is the following fact about Cech cohomology, which is discussed for instance at infinity-cohesive site:

For XX a smooth manifold (in particular paracompact),

  1. XX admits a good open cover {U iX}\{U_i \to X\} (by charts U iU_i all whose finite non-empty intersections are diffeomorphic to an open ball/Cartesian space n\mathbb{R}^n);

  2. for any such good open cover the Cech complex C ((X,{U i}),A )C^\bullet((X,\{U_i\}),A_\bullet) already computes Cech cohomology (i.e. there is no further need to form the colimit of Cech complexes over refinements of covers);

  3. the functor C ((X,{U i}),):Ch +(Smooth0Type)Ch +C^\bullet((X,\{U_i\}),-) \colon Ch_+(Smooth0Type) \to Ch_+ preserves weak equivalences and fibrations.

This means in particular that if X Y Z X_\bullet \to Y_\bullet \to Z_\bullet is a homotopy fiber sequence in Ch +(Smooth0Type)Ch_+(Smooth0Type), then also

C ((X,{U i}),X )C ((X,{U i}),Y )C ((X,{U i}),Z ) C^\bullet((X,\{U_i\}), X_\bullet) \to C^\bullet((X,\{U_i\}), Y_\bullet) \to C^\bullet((X,\{U_i\}), Z_\bullet)

is a homotopy fiber sequence of chain complexes, and therefore the cohomology groups sit in the long exact sequence in homology of this sequence of chain complexes.


For A =(B n+1) A_\bullet = (\mathbf{B}^{n+1}\mathbb{Z})_\bullet as in example 3, then for XX a smooth manifold

H Cech 0(X,(B n+1) )H n+1(X,) H^0_{Cech}(X, (\mathbf{B}^{n+1}\mathbb{Z})_\bullet) \simeq H^{n+1}(X,\mathbb{Z})

is the ordinary cohomology of XX with integer coefficients, the cohomology which is also computed as the singular cohomology of the underlying topological space of XX.

Similarly for A =(B nU(1)) A_\bullet = (\flat \mathbf{B}^n U(1))_{\bullet} then

H Cech 0(X,(B nU(1)) )H n(X,U(1)) H^0_{Cech}(X, (\flat\mathbf{B}^{n}U(1))_\bullet) \simeq H^{n}(X,U(1))

is the ordinary cohomology of XX with circle group coefficients, the cohomology which is also computed as the singular cohomology of the underlying topological space of XX with U(1)U(1)-coefficients.


Passing to abelian sheaf cohomology (e.g. via def. 4), then prop. 1 is the de Rham theorem.

We will have need to give names to truncations of the de Rham complex. One is this:


For nn \in \mathbb{N} write

Ω nCh +(Smooth0Type) \mathbf{\Omega}^{\bullet \leq n} \in Ch_+(Smooth0Type)

for the chain complex of the form

Ω n[dΩ 1ddΩ n1dΩ n] \mathbf{\Omega}^{\bullet \leq n} \coloneqq \left[ \mathbb{R} \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^1 \stackrel{\mathbf{d}}{\to} \cdots \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^{n-1} \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^{n} \right]

with all nn-forms, not just the closed ones, in degree 0.


The abelian sheaf cohomology of the truncated de Rham complex in def. 5 is Ω n(X)/im(d)\Omega^n(X)/im(\mathbf{d}).

The Deligne complex


For nn \in \mathbb{N} the smooth Deligne complex of degree nn

(B nU(1) conn) Ch +(Smooth0Type) (\mathbf{B}^n U(1)_{conn})_\bullet \in Ch_+(Smooth0Type)

is the chain complex of abelian sheaves given by

(B nU(1) conn) [U(1)dlogΩ 1ddΩ n] (\mathbf{B}^n U(1)_{conn})_\bullet \; \coloneqq \; \left[ U(1) \stackrel{\mathbf{d} log}{\to} \mathbf{\Omega}^1 \stackrel{\mathbf{d}}{\to} \cdots \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^n \right]

with U(1)U(1) in degree nn and with the differentials as in def. 2 and example 4.

We write

H conn n+1(X,)H Cech 0(X,(B nU(1) conn) ) H^{n+1}_{conn}(X,\mathbb{Z}) \coloneqq H^0_{Cech}(X,(\mathbf{B}^n U(1)_{conn})_\bullet)

for its abelian sheaf cohomology.


By example 4 the obvious chain map

d Ω 1 d d Ω n ()/ id id 0 U(1) dlog Ω 1 d d Ω n \array{ \mathbb{Z} &\hookrightarrow& \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n \\ \downarrow && \downarrow^{\mathrlap{(-)/\mathbb{Z}}} && \downarrow^{id} && && \downarrow^{id} \\ 0 &\to& U(1) &\stackrel{\mathbf{d} log}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n }

is a weak equivalence, def. 2, and one could define the top chain complex here as “the” Deligne complex, just as well. In the context of homotopy theory/homological algebra, all that matters is the complex up to zig-zags of weak equivalences.


In def. 6 the de Rham complex is truncated to the right by discarding what would be the next differentials, without passing to their kernel, i.e. in degree 0 the Deligne complex has all differential nn-forms, not just the closed nn-fomrs. This simple point is the key aspect of the Deligne complex. If one instead truncates while preserving the chain homology in the lowest degree, then one obtains the following complex with the sheaf Ω cl n\mathbf{\Omega}^{n}_{cl} of closed forms in lowest degree, which gives ordinary cohomology.


For nn \in \mathbb{N} the flat smooth Deligne complex of degree nn

(B nU(1) flat) Ch +(Smooth0Type) (\mathbf{B}^n U(1)_{flat})_\bullet \in Ch_+(Smooth0Type)

is the chain complex of abelian sheaves given by

(B nU(1) flat) [U(1)dlogΩ 1ddΩ cl n] (\mathbf{B}^n U(1)_{flat})_\bullet \; \coloneqq \; \left[ U(1) \stackrel{\mathbf{d} log}{\to} \mathbf{\Omega}^1 \stackrel{\mathbf{d}}{\to} \cdots \stackrel{\mathbf{d}}{\to} \mathbf{\Omega}^n_{cl} \right]

with U(1)U(1) in degree nn and with the differentials as in def. 2 and example 4, and with the closed nn-forms on the right.



(B nU(1)) =(B nU(1)) Ch +(Smooth0Type) (\flat\mathbf{B}^n U(1))_\bullet = (\mathbf{B}^n \flat U(1))_\bullet \in Ch_+(Smooth0Type)

for the chain complex of abelian sheaves given by

(B nU(1)) [U(1)00] (\flat \mathbf{B}^n U(1))_\bullet \; \coloneqq \; \left[ \flat U(1) \stackrel{}{\to} 0 \stackrel{}{\to} \cdots \stackrel{}{\to} 0 \right]

with the constant sheaf U(1)\flat U(1) of example 1 in degree nn.


For (B nU(1)) (\flat \mathbf{B}^n U(1))_\bullet as in def. 8, then the morphism

(B nU(1)) (B nU(1) flat) (\flat \mathbf{B}^n U(1))_\bullet \stackrel{\simeq}{\longrightarrow} (\mathbf{B}^n U(1)_{flat})_\bullet

given by the chain map

U(1) 0 0 U(1) dlog Ω 1 d d Ω cl n \array{ \flat U(1) &\to& 0 &\to& \cdots &\to& 0 \\ \downarrow && \downarrow && \cdots && \downarrow \\ U(1) &\stackrel{\mathbf{d}log}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n_{cl} }

(with the vertical morphism on the left being the inclusion of example 4) is a weak equivalence, def. 2.


By the Poincaré lemma, this is just an immediate variant of prop. 1.

Cup product in Deligne cohomology

The cup product on ordinary cohomology refines to Deligne cohomology.

For more on this see at Beilinson-Deligne cup-product.


Curvature and characteristic classes

We discuss the construction of two canonical morphisms out of Deligne cohomology, and two canonical morphisms into it. Below these are shown to form two interlocking exact sequences and in fact an exact differential cohomology hexagon which accurately characterizes Deligne cohomology as the differential cohomology extension of integral ordinary cohomology by differential forms.

Throughout, for ease of notation, we assume nn \in \mathbb{N} to be positive,

n1. n \geq 1 \,.

The remaining case n=0n = 0 describes “circle 0-bundles with connection”, which are just U(1)U(1)-valued functions, and is hence essentially trivial in itself.

In the following XX is any smooth manifold.


Let (B n+1) Ch +(Smooth0Type)(\mathbf{B}^{n+1}\mathbb{Z})_\bullet \in Ch_+(Smooth0Type) be as in example 3. Write

(B nU(1) conn) DD (B n+1) (\mathbf{B}^n U(1)_{conn})_{\bullet} \stackrel{\simeq}{\longleftarrow} \stackrel{DD_\bullet}{\longrightarrow} (\mathbf{B}^{n+1} \mathbb{Z})_{\bullet}

for the zig-zag of chain complexes where the left weak equivalence is that of remark 8, i.e. for the chain maps given by

0 0 0 id d Ω 1 d d Ω n 0 mod id id 0 U(1) dlog Ω 1 d d Ω n \array{ \mathbb{Z} &\to& 0 &\to& 0 &\to& \cdots &\to& 0 \\ \uparrow^{\mathrlap{id}} && \uparrow^{\mathrlap{}} && \uparrow^{\mathrlap{}} && \cdots && \uparrow^{\mathrlap{}} \\ \mathbb{Z} &\hookrightarrow& \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n \\ \downarrow^{\mathrlap{0}} && \downarrow^{\mathrlap{mod\,\mathbb{Z}}} && \downarrow^{\mathrlap{id}} && \cdots && \downarrow^{\mathrlap{id}} \\ 0 &\to& U(1) &\stackrel{\mathbf{d} log}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n }

Passing to abelian sheaf cohomology this gives, by def. 6 and example 5, a morphism

H conn n+1(X,) DD H n+1(X,) \array{ H^{n+1}_{conn}(X,\mathbb{Z}) \\ & \searrow^{\mathrlap{DD}} \\ && H^{n+1}(X,\mathbb{Z}) }

from Deligne cohomology to ordinary cohomology with integer coefficients in degree n+1n+1.

For []H conn n+1(X,)[\nabla] \in H^{n+1}_{conn}(X,\mathbb{Z}) we call DD()H n+1(X,)DD(\nabla) \in H^{n+1}(X,\mathbb{Z})
the Dixmier-Douady class of the underlying circle n-bundle.



(F ()) :(B nU(1) conn) Ω cl n+1 (F_{(-)})_\bullet \colon (\mathbf{B}^n U(1)_{conn})_\bullet \stackrel{}{\longrightarrow} \mathbf{\Omega}^{n+1}_{cl}

for the morphism given by the chain map which is just the de Rham differential in degree 0

0 U(1) dlog Ω 1 d d Ω n1 d Ω n d 0 0 0 0 Ω cl n+1 \array{ 0 &\to& U(1) &\stackrel{\mathbf{d} log}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n-1} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n \\ \downarrow && \downarrow && \downarrow && \cdots && \downarrow^{\mathrlap{}} && \downarrow^{\mathrlap{\mathbf{d}}} \\ 0 &\to& 0 &\to& 0 &\to& \cdots &\stackrel{}{\to}& 0 &\to& \mathbf{\Omega}^{n+1}_{cl} }

Passing to abelian sheaf cohomology this gives a morphism of the form

Ω cl n+1(X) F () H conn n+1(X,) \array{ && \Omega^{n+1}_{cl}(X) \\ & {}^{\mathllap{F_{(-)}}}\nearrow \\ H^{n+1}_{conn}(X,\mathbb{Z}) }

We call this the curvature map, i.e. for []H conn n+1(X,)[\nabla] \in H^{n+1}_{conn}(X,\mathbb{Z}) the class of a Deligne cocycle, we call

F Ω cl n+1(X) F_\nabla \in \Omega^{n+1}_{cl}(X)

its curvature form.


Consider the zig-zag

Ω n(B nU(1) conn) \mathbf{\Omega}^{\bullet\leq n} \stackrel{}{\longrightarrow} (\mathbf{B}^n U(1)_{conn})_\bullet

out of the complex of def. 5, given by the chain maps

0 d Ω 1 d d Ω n1 d Ω n id id id d Ω 1 d d Ω n1 d Ω n id id id 0 U(1) dlog Ω 1 d d Ω n1 d Ω n \array{ 0 &\to& \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n-1} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n \\ \downarrow && \downarrow && \downarrow^{\mathrm{id}} && \cdots && \downarrow^{\mathrm{id}} && \downarrow^{\mathrm{id}} \\ \mathbb{Z} &\hookrightarrow& \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n-1} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n \\ \downarrow && \downarrow && \downarrow^{\mathrm{id}} && \cdots && \downarrow^{\mathrm{id}} && \downarrow^{\mathrm{id}} \\ 0 &\to& U(1) &\stackrel{\mathbf{d}log}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n-1} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n }

where the bottom quasi-isomorphism is from remark 8.

On passing to abelian sheaf cohomology this gives, by example 7, a morphism

Ω n(X)/im(d) H conn n+1(X,) \array{ \Omega^n(X)/im(\mathbf{d}) \\ & \searrow \\ && H^{n+1}_{conn}(X,\mathbb{Z}) }

Consider the canonical morphism

(B nU(1)) (B nU(1) flat) (B nU(1) conn) (\flat \mathbf{B}^n U(1))_\bullet \stackrel{\simeq}{\longrightarrow} (\mathbf{B}^n U(1)_{flat})_\bullet \longrightarrow (\mathbf{B}^n U(1)_{conn})_\bullet

via def. 7, prop. 2.

Passing to abelian sheaf cohomology this induces a morphism

H conn n+1(X,) H n(X,U(1)) \array{ && H^{n+1}_{conn}(X,\mathbb{Z}) \\ & \nearrow \\ H^n(X,U(1)) }

We call this map the inclusion of the flat infinity-connections into all circle n-connections.

Combining what we have so far:


The composite of the morphisms of def. 11 and of the curvature morphism of def. 10

Ω n(X)/im(d) d Ω cl n+1(X) F () H conn n+1(X,) \array{ \Omega^n(X)/im(\mathbf{d}) && \stackrel{\mathbf{d}}{\longrightarrow} && \Omega^{n+1}_{cl}(X) \\ & \searrow && \nearrow_{\mathrlap{F_{(-)}}} \\ && H^{n+1}_{conn}(X,\mathbb{Z}) }

is given by the de Rham differential d\mathbf{d} on differential forms.

The composite of the morphisms of def. 12 and def. 9 is the Bockstein homomorphism:

H conn n+1(X,) DD H n(X,U(1)) β H n+1(X,) \array{ && H^{n+1}_{conn}(X,\mathbb{Z}) \\ & \nearrow && \searrow^{\mathrlap{DD}} \\ H^n(X,U(1)) && \underset{\beta}{\longrightarrow} && H^{n+1}(X,\mathbb{Z}) }

By composing the defining zig-zags of chain maps the statement is immediate.

The Chern character

While the explicit definition of the Deligne complex in def. 6 is easy enough, all its good abstract properties are best understood by realizing that it is the homotopy fiber product of a kind of higher abelian Chern character map with the closed differential forms Ω cl n+1\mathbf{\Omega}^{n+1}_{cl}. This is the content of prop. 4 below.



ch :(B n+1) (B n+1) ch_\bullet \colon (\mathbf{B}^{n+1}\mathbb{Z})_\bullet \longrightarrow (\flat \mathbf{B}^{n+1}\mathbb{R})_\bullet

for the morphism given as the composite

(B n+1) =(B n+1) (B n+1) (\mathbf{B}^{n+1}\mathbb{Z})_\bullet = (\flat \mathbf{B}^{n+1}\mathbb{Z})_\bullet \longrightarrow (\flat \mathbf{B}^{n+1}\mathbb{R})_\bullet

where the second morphism is induced by the canonical inclusion \mathbb{Z} \hookrightarrow \mathbb{R}.

Passing to abelian sheaf cohomology this induces a morphism

H n+1(X,) ch H n+1(X,) \array{ && H^{n+1}(X,\mathbb{R}) \\ & \nearrow_{\mathrlap{ch}} \\ H^{n+1}(X,\mathbb{Z}) }

The morphism in def. 13 is just the traditional map from ordinary cohomology with integer coefficients to that with real numbers coefficients given for instance via singular cohomology simply by forming the tensor product of abelian groups with \mathbb{R}. In the broader context of differential cohomology however it is useful to think of this map as the Chern character, whence the notation.

While this is easy enough to construct in itself, the underlying chain map here is not a fibration in the sense of def. 2 (having as non-trivial component the inclusion \mathbb{Z} \hookrightarrow \mathbb{R}, which is evidently not an epimorphism). But the homotopy fiber of this map plays a crucial role in the theory, and so in view of remark 2 we consider now a fibration resolution of this map


Consider the chain complex

(B n+1)^ [ d Ω 1 d d Ω n1 d Ω n id +id id ±id id d Ω 1 d Ω 2 d d Ω n ] \widehat {(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet \coloneqq \left[ \array{ \mathbb{Z} &\stackrel{}{\hookrightarrow}& \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n-1} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n \\ \oplus &\nearrow_{-\mathrlap{id}}& \oplus &\nearrow_{+\mathrlap{id}}& \oplus &\nearrow_{-\mathrlap{id}}& \cdots &\nearrow_{\pm\mathrlap{id}}& \oplus &\nearrow_{\mp\mathrlap{id}}& \\ \mathbb{R} &\underset{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\underset{\mathbf{d}}{\to}& \mathbf{\Omega}^2 &\underset{\mathbf{d}}{\to}& \cdots &\underset{\mathbf{d}}{\to}& \mathbf{\Omega}^{n} && } \right]

where we use matrix calculus-notation as for mapping cones (see at mapping cone – Examples – In chain complexes).


(B n+1)^ ch^ (B n+1)^ \widehat {(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet \stackrel{\widehat{ch}_\bullet}{\longrightarrow} \widehat {(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet

for the morphism to the chain complex of def. 3 which is given by the chain map that in positive degree projects onto the lower row in the above direct sum expression and in degree 0 is given by the de Rham differential:

d Ω 1 d d Ω n1 d Ω n id +id id ±id id d Ω 1 d Ω 2 d d Ω n d d Ω 1 d Ω 2 d d Ω n d Ω n+1 \array{ \mathbb{Z} &\stackrel{}{\hookrightarrow}& \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n-1} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^n \\ \oplus &\nearrow_{-\mathrlap{id}}& \oplus &\nearrow_{+\mathrlap{id}}& \oplus &\nearrow_{-\mathrlap{id}}& \cdots &\nearrow_{\pm\mathrlap{id}}& \oplus &\nearrow_{\mp\mathrlap{id}}& \\ \mathbb{R} &\underset{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\underset{\mathbf{d}}{\to}& \mathbf{\Omega}^2 &\underset{\mathbf{d}}{\to}& \cdots &\underset{\mathbf{d}}{\to}& \mathbf{\Omega}^{n} && \\ \downarrow && \downarrow && \downarrow && \cdots && \downarrow && \downarrow^{\mathrlap{\mathbf{d}}} \\ \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^2 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n+1} }

Finally write

(B n+1) (B n+1)^ (\mathbf{B}^{n+1}\mathbb{Z})_\bullet \longrightarrow \widehat {(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet

for the morphism given by the chain map which in degree n+1n+1 is given by

\mathbb{Z} \to \mathbb{Z} \oplus \mathbb{R}
n(n,n). n \mapsto (n,n) \,.

The construction in def. 14 gives a fibration resolution of the Chern character morphism of def. 13 in that it gives a commuting diagram of chain maps

(B n+1) ch (B n+1) (B n+1)^ ch^ (B n+1)^ \array{ (\mathbf{B}^{n+1}\mathbb{Z})_\bullet &\stackrel{ch_\bullet}{\longrightarrow}& (\flat\mathbf{B}^{n+1} \mathbb{R})_\bullet \\ \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{\simeq}} \\ \widehat {(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet &\stackrel{\widehat {ch}_\bullet}{\longrightarrow}& \widehat {(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet }

with, on the right, the weak equivalence of prop. 1, where

  1. the left vertical morphism is a weak equivalence;

  2. the bottom horizontal morphism ch^ \widehat{ch}_\bullet is a fibration.


That the diagram commutes is a straightforward inspection, unwinding the definitions. That ch^ \widehat{ch}_\bullet is a fibration according to def. 2 is by its very construction, being a projection in positive degree. That the left morphism is a weak equivalence comes down to the Poincaré lemma, in a slight variant of the simple argument that proves prop. 1.



Ω cl n+1ι B n+1^ (B n+1) \mathbf{\Omega}^{n+1}_{cl} \stackrel{\iota_\bullet}{\longrightarrow} \widehat{\flat \mathbf{B}^{n+1}\mathbb{R}}_\bullet \stackrel{\simeq}{\longleftarrow} (\flat \mathbf{B}^{n+1}\mathbb{R})_\bullet

for the zig-zag whose right morphism is the weak equivalence of prop. 1 and whose left morphism is given by the chain map

0 0 0 Ω cl n+1 id d Ω 1 d d Ω n d Ω cl n+1. \array{ 0 &\to& 0 &\to& \cdots &\to& 0 &\to& \mathbf{\Omega}^{n+1}_{cl} \\ \downarrow && \downarrow && \vdots && \downarrow && \downarrow^{\mathrlap{id}} \\ \mathbb{R} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^1 &\stackrel{\mathbf{d}}{\to}& \cdots &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n} &\stackrel{\mathbf{d}}{\to}& \mathbf{\Omega}^{n+1}_{cl} } \,.

The chain maps

  • (F ()) (F_{(-)})_\bullet, def. 10;

  • DD DD_\bullet, def. 9;

  • ι \iota_\bullet, def. 15

  • ch^ \widehat{ch}_\bullet, def. 14

fit into a commuting diagram

Ω cl n+1 (F ()) ι (B nU(1) conn) (B n+1)^ DD ch^ (B n+1)^ \array{ && \mathbf{\Omega}^{n+1}_{cl} \\ & {}^{\mathllap{(F_{(-)})_\bullet }}\nearrow && \searrow^{\mathrlap{\iota_\bullet}} \\ (\mathbf{B}^n U(1)_{conn})_\bullet && && \widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet \\ & {}_{\mathllap{DD_\bullet}}\searrow && \nearrow_{\mathrlap{\widehat{ch}_\bullet}} \\ && \widehat{(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet }

which is a pullback diagram in Ch +(Smooth0Type)Ch_+(Smooth0Type). This exhibits the Deligne complex (B nU(1) conn) (\mathbf{B}^n U(1)_{conn})_\bullet as the homotopy pullback of the inclusion of Ω cl n+1\mathbf{\Omega}^{n+1}_{cl} along the Chern character map ch ch_\bullet.


The first statement follows straightforwardly by inspection, using that pullbacks of chain complexes are computed componentwise. From this the second statement follows then since by 1 ch^ \widehat{ch}_\bullet is a fibration resolution of ch ch_\bullet.

The exact sequences for curvature and characteristic classes



Ω int n+1(X)Ω cl n+1(X) \Omega^{n+1}_{int}(X) \hookrightarrow \Omega^{n+1}_{cl}(X)

for the inclusion of those closed differential forms whose periods (integration over (n+1)(n+1)-cycles) takes values in the integers.


The image of the curvature map F ():H conn n+1(X,)Ω cl n+1(X)F_{(-)} \colon H^{n+1}_{conn}(X,\mathbb{Z}) \longrightarrow \Omega^{n+1}_{cl}(X) of def. 10 are the integral forms of def. 16.


The homotopy pullback characterization of of prop. 4 implies that the image consists of precisely those closed differential forms which under the de Rham theorem, remark 6, represent real cohomology classes that are in the image of integral cohomology classes. These are the differential forms with integral periods.


(curvature exact sequence)

The Deligne cohomology group fits into a short exact sequence (of abelian groups) of the form

0H n(X,U(1))H conn n+1(X,)F ()Ω int n+1(X)0 0 \to H^n(X,U(1)) \longrightarrow H^{n+1}_{conn}(X,\mathbb{Z}) \stackrel{F_{(-)}}{\longrightarrow} \Omega^{n+1}_{int}(X) \to 0

where F ()F_{(-)} is the curvature map of def. 10.


By prop. 5 the morphism on the right is indeed an epimorphism. It remains to determine its kernel.

To that end, consider the pasting diagram of homotopy pullbacks obtained form the homotopy pullback in prop. 4. Using the pasting law and the fact that the loop space object of a 0-truncated object such as Ω cl n+1\mathbf{\Omega}^{n+1}_{cl} is trivial, this is of the form

0 (B n(/))^ 0 0 (B nU(1) conn) (F ()) Ω cl n+1 DD ι (B n+1)^ ch^ (B n+1)^ . \array{ 0 &\longrightarrow& \widehat{(\flat \mathbf{B}^{n}(\mathbb{R}/\mathbb{Z}))}_\bullet &\longrightarrow& 0 \\ \downarrow && \downarrow && \downarrow^{} \\ 0 &\longrightarrow& (\mathbf{B}^n U(1)_{conn})_\bullet &\stackrel{(F_{(-)}_\bullet)}{\longrightarrow}& \mathbf{\Omega}^{n+1}_{cl} \\ && \downarrow^{\mathrlap{DD_\bullet}} && \downarrow^{\mathrlap{\iota_\bullet}} \\ && \widehat{(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet &\stackrel{\widehat{ch}_\bullet}{\longrightarrow}& \widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet } \,.

Passing to abelian sheaf cohomology and applying the induced long exact sequence in homology, in view of remark 7, implies the claim.


(characteristic class exact sequence)

The Deligne cohomology group fits into a short exact sequence (of abelian groups) of the form

0Ω n(X)/Ω int n(X)H conn n+1(X,)DDH n+1(X,)0 0 \to \Omega^{n}(X)/\Omega^n_{int}(X) \stackrel{}{\longrightarrow} H^{n+1}_{conn}(X,\mathbb{Z}) \stackrel{DD}{\longrightarrow} H^{n+1}(X,\mathbb{Z}) \to 0

where DDDD is the characteristic class map of def. 9.


The chain map that represents the Dixmier-Douady class by def. 9 is manifestly a fibration in the sense of def. 2. Therefore its ordinary fiber is already its homotopy fiber. That ordinary fiber is evidently the domain of the morphism constructed in def. 11, in its second weakly equivalent incarnation as displayed there.

Therefore the long exact sequence in homology, induced by the chain map DD DD_\bullet under passage to abelian sheaf cohomology (in view of remark 7) goes as

H n(X,)Ω n(X)/im(d)H conn n+1(X,)DDH n+1(X,) \cdots \longrightarrow H^n(X,\mathbb{Z}) \longrightarrow \Omega^n(X)/im(\mathbf{d}) \stackrel{}{\longrightarrow} H^{n+1}_{conn}(X,\mathbb{Z}) \stackrel{DD}{\longrightarrow} H^{n+1}(X,\mathbb{Z})

As in the proof of prop. 5 it follows that the rightmost morphism is an epimorphism. Hence we get a short exact sequence by dividing out the image of H n(X,)H^n(X,\mathbb{Z}) in Ω n/im(d)\Omega^n/im(\mathbf{d}). That image is Ω int n(X)\Omega^n_{int}(X). Since this image contains im(d)im(\mathbf{d}) (as the closed differential forms all whose periods are 00 \in \mathbb{N} ) the resulting quotient is Ω n(X)/Omega int n(X)\Omega^n(X)/Omega^n_{int}(X) and the claim follows.


In words the statement of prop. 6 and prop. 7 is that Deligne cohomology groups H conn n+1(X,)H^{n+1}_{conn}(X,\mathbb{Z}) constitute a group extension

  1. of integral ordinary cohomology by differential nn-forms modulo integral forms;

  2. of closed (n+1)(n+1)-forms by U(1)U(1)-valued ordinary cohomology.

The first statement is what gives the name to “differential cohomology” as it makes precise how H conn n+1(X)H^{n+1}_{conn}(X) is a combination of ordinary integral cohomology with differential form-data.

The second statement is secretly of the same flavor, if maybe not as manifestly so: the U(1)U(1)-valued ordinary cohomology is really what classifies flat circle n-connections on circle n-group principal infinity-bundles (either, depending on perspective, by def. 7 or else, more intrinsically, by the very statement of prop. 6) and hence again describes a combination of underlying bundles with differential form data.

The exact differential cohomology hexagon

Summing up, the homotopy pullback square of prop. 4 together with the maps of prop. 3 form a commuting diagram in Ch +(Smooth0Type)Ch_+(Smooth0Type) of the form.

Ω n d Ω cl n+1 (F ()) ι (B nU(1) conn) (B n+1)^ DD ch^ (B nU(1))^ β (B n+1)^ \array{ \mathbf{\Omega}^{\bullet \leq n} && \stackrel{\mathbf{d}}{\longrightarrow} && \mathbf{\Omega}^{n+1}_{cl} \\ &\searrow& & {}^{\mathllap{(F_{(-)})_\bullet }}\nearrow && \searrow^{\mathrlap{\iota_\bullet}} \\ && (\mathbf{B}^n U(1)_{conn})_\bullet && && \widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet \\ &\nearrow& & {}_{\mathllap{DD_\bullet}}\searrow && \nearrow_{\mathrlap{\widehat{ch}_\bullet}} \\ \widehat{(\flat \mathbf{B}^n U(1))}_\bullet && \underset{\beta}{\longrightarrow} && \widehat{(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet }

This extends to a diagram in Ch +(Smooth0Type)Ch_+(Smooth0Type) of the form

Ω n d Ω cl n+1 (F ()) ι (B n)^ (B nU(1) conn) (B n+1)^ DD ch^ (B nU(1))^ β (B n+1)^ \array{ && \mathbf{\Omega}^{\bullet \leq n} && \stackrel{\mathbf{d}}{\longrightarrow} && \mathbf{\Omega}^{n+1}_{cl} \\ & \nearrow& &\searrow& & {}^{\mathllap{(F_{(-)})_\bullet }}\nearrow && \searrow^{\mathrlap{\iota_\bullet}} \\ \widehat{(\flat \mathbf{B}^n \mathbb{R})}_{\bullet} && && (\mathbf{B}^n U(1)_{conn})_\bullet && && \widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet \\ &\searrow & &\nearrow& & {}_{\mathllap{DD_\bullet}}\searrow && \nearrow_{\mathrlap{\widehat{ch}_\bullet}} \\ && \widehat{(\flat \mathbf{B}^n U(1))}_\bullet && \underset{\beta}{\longrightarrow} && \widehat{(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet }

such that

  1. both square are homotopy pullback squares;

  2. both diagonals are homotopy fiber sequences;

  3. the two outer sequences are long homotopy fiber sequences.


For the first statement consider the pasting of homotopy pullback diagrams as in the proof of prop. 6, now extended to the left, via the pasting law, as

(B n)^ (B n(/))^ 0 Ω n (B nU(1) conn) (F ()) Ω cl n+1 DD ι (B n+1)^ ch^ (B n+1)^ . \array{ \widehat{(\flat \mathbf{B}^n \mathbb{R})}_\bullet &\longrightarrow& \widehat{(\flat \mathbf{B}^{n}(\mathbb{R}/\mathbb{Z}))}_\bullet &\longrightarrow& 0 \\ \downarrow && \downarrow && \downarrow^{} \\ \mathbf{\Omega}^{\bullet \leq n} &\longrightarrow& (\mathbf{B}^n U(1)_{conn})_\bullet &\stackrel{(F_{(-)}_\bullet)}{\longrightarrow}& \mathbf{\Omega}^{n+1}_{cl} \\ && \downarrow^{\mathrlap{DD_\bullet}} && \downarrow^{\mathrlap{\iota_\bullet}} \\ && \widehat{(\mathbf{B}^{n+1}\mathbb{Z})}_\bullet &\stackrel{\widehat{ch}_\bullet}{\longrightarrow}& \widehat{(\flat \mathbf{B}^{n+1}\mathbb{R})}_\bullet } \,.

That the NE-diagonal is a homotopy fiber sequence is the statement in the proof of prop. 6. That the SE-diagonal is a homotopy fiber sequence follows by inspection as remarked in the proof of prop. 7.

From this the last statement now is implied by using the pasting law yet once more, as show in the proof here.


The form of the exact hexagon characterizing the Deligne complex via prop. 8 is in fact a general abstract consequence of the fact that all universal constructions in Ch +(Smooth0Type)Ch_+(Smooth0Type) considered here indeed may be understood as taking place in the cohesive homotopy theory of smooth ∞-groupoids, via remark 4. This is discussed in detail at differential cohomology hexagon.


The Deligne complex is naturally defined in smooth differential geometry as well as in complex analytic geometry as well as in algebraic geometry over the complex numbers. In the spirit of GAGA it is of interest to know how Deligne cohomology in these different settings relates.

One useful statement is: given an smooth algebraic variety over the complex numbers, then a sufficient condition for a complex-analytic Deligne cocycle over its analytification to lift to an algebraic Deligne cocycle is that its curvature form is an algebraic form (Esnault 89, corollary 1.3).

Moduli and deformation theory

The moduli spaces of holomorphic Deligne cohomology groups are closely related to intermediate Jacobians, see there fore more.

The deformation theory of Deligne cohomology groups is given by Artin-Mazur formal group, see there for more

moduli spaces of line n-bundles with connection on nn-dimensional XX

nnCalabi-Cau n-foldline n-bundlemoduli of line n-bundlesmoduli of flat/degree-0 n-bundlesArtin-Mazur formal group of deformation moduli of line n-bundlescomplex oriented cohomology theorymodular functor/self-dual higher gauge theory of higher dimensional Chern-Simons theory
n=0n = 0unit in structure sheafmultiplicative group/group of unitsformal multiplicative groupcomplex K-theory
n=1n = 1elliptic curveline bundlePicard group/Picard schemeJacobianformal Picard groupelliptic cohomology3d Chern-Simons theory/WZW model
n=2n = 2K3 surfaceline 2-bundleBrauer groupintermediate Jacobianformal Brauer groupK3 cohomology
n=3n = 3Calabi-Yau 3-foldline 3-bundleintermediate JacobianCY3 cohomology7d Chern-Simons theory/M5-brane
nnintermediate Jacobian

Interpretation in terms of higher parallel transport

There is a natural way to understand the Deligne complex of sheaves as a sheaf which assigns to each patch the Lie nn-groupoid of smooth higher parallel transport n-functors.

We start by discussing this in low degree.

There is path groupoid P 1(X)P_1(X) whose smooth space of objects is XX and whose smooth space of morphisms is a space of classes of smooth paths in XX. Every smooth 1-form AΩ 1(X)A \in \Omega^1(X) induces a smooth functor tra A:P 1(X)BU(1)tra_A : P_1(X) \to \mathbf{B}U(1) from P 1(X)P_1(X) to to the smooth groupoid BU(1)\mathbf{B} U(1) with one object and U(1)U(1) as its smooth space of morphisms by sending each path γ:[0,1]X\gamma : [0,1] \to X to exp(2πi 0 1γ *A)\exp (2 \pi i\int_0^1 \gamma^* A). This map from 1-forms to smooth functors turns out to be bijective: every smooth functor of this form uniquely arises this way. Similarly, one finds that smooth natural transformation η f:tra Atra A\eta_f : tra_A \to tra_{A'} between two such functors is in components precisely a smooth function f:XU(1)f : X \to U(1) such that A=A+dlogfA' = A + d log f.

Since the analogous statements are true for every open subset UXU \subset X this defines a sheaf of Lie groupoids

Funct (P 1(),BU(1)):Op(X) opLieGrpd. Funct^\infty(P_1(-), \mathbf{B}U(1)) : Op(X)^{op} \to LieGrpd \,.

By the Dold-Kan correspondence this sheaf of groupoids corresponds to a sheaf of complexes of groups. This complex of sheaves is nothing but the degree 2 Deligne complex

Funct (Π 1(),BU(1))(2) D . Funct^\infty(\Pi_1(-), \mathbf{B}U(1)) \simeq \mathbb{Z}(2)^\infty_D \,.

This way Deligne cohomology is realized as computing the stackification of the pre-stack Funct (P 1(),B(1))Funct^\infty(P_1(-), \mathbf{B}(1)) of smooth U(1)U(1)-valued parallel transport functors.

The identification generalizes: for all nn there is a path n-groupoid P n(X)P_n(X) whose kk-morphisms are kk-dimensional smooth paths in XX. Smooth nn-functors tra C: n(X)B nU(1)tra_C : _n(X) \to \mathbf{B}^n U(1) are canonically identified with smooth nn-forms CΩ n(X)C \in \Omega^n(X) and under the Dold-Kan correspondence the Deligne-complex in degree n+1n+1 is identified with the sheaf of nn-groupoids of such smooth nn-functors

nFunct (P n(),B n)(n+1) D . n Funct^\infty(P_n(-), \mathbf{B}^n) \simeq \mathbb{Z}(n+1)^\infty_D \,.


  • John Baez, Urs Schreiber, Higher Gauge Theory (arXiv)

The full proof for n=1n=1 this is in

  • Urs Schreiber, Konrad Waldorf, Parallel transport and functors (arXiv);

for n=2n=2 in

  • Urs Schreiber, Konrad Waldorf, Smooth functors versus differential forms (arXiv)

For more on this see infinity-Chern-Weil theory introduction.

For higher nn there is as yet no detailed proof in the literature, but the low dimensional proofs have obvious generalizations.


As described in some detail at electromagnetic field in abelian higher gauge theories the background field naturally arises as a Čech–Deligne cocycle, i.e. a Čech cocycle representative with values in the Deligne complex.


Deligne cohomology was introduced in complex analytic geometry (by a chain complex of holomorphic differential forms) in

  • Pierre Deligne, Théorie de Hodge II , IHES Pub. Math. (1971), no. 40, 5–57 (pdf)

with applications to Hodge theory and intermediate Jacobians. The same definition appears in

  • Barry Mazur, William Messing, Universal extensions and one-dimensional crystalline cohomology, Springer lecture notes 370, 1974

  • Michael Artin, Barry Mazur, section III.1 of Formal Groups Arising from Algebraic Varieties, Annales scientifiques de l’École Normale Supérieure, Sér. 4, 10 no. 1 (1977), p. 87-131 numdam, MR56:15663

under the name “multiplicative de Rham complex” (and in the context of studying its deformation theory by Artin-Mazur formal groups). The theory was further developed in

with the application to Beilinson regulators. Later the evident version of the Deligne complex in differential geometry over smooth manifolds gained more attention and is still referred to as “Deligne cohomology”.

Surveys and introductions in the context of differential geometry include

Review with more emphasis on complex analytic geometry and the theory of (Beilinson 85) with more details spelled out is in

  • Hélène Esnault, Eckart Viehweg, Deligne-Beilinson cohomology in Rapoport, Schappacher, Schneider (eds.) Beilinson’s Conjectures on Special Values of L-Functions . Perspectives in Math. 4, Academic Press (1988) 43 - 91 (pdf)

  • Hélène Esnault, On the Loday-symbol in the Deligne-Beilinson cohomology, K-theory 3, 1-28, 1989 (pdf)

See also

Discussion of Deligne cohomology in terms of simplicial presheaves and higher stacks includes

See also the references given at differential cohomology hexagon – Deligne coefficients.

Revised on March 6, 2015 09:15:15 by Urs Schreiber (