Change of connection formula for Chern Weil characteristic forms

Exact formula relating characteristic forms computed from two different principal connections
Change of connection formula for Chern Weil characteristic forms

Let π:PM\pi:P\to M be a , and let ω0,ω1\omega_0,\omega_1 be two on PP with connection 1-forms (also denoted) ω0,ω1Ω1(P;g)\omega_0,\omega_1\in \Omega^1(P;\mathfrak g) as in . Write their curvature 2-forms as Ω0,Ω1Ω2(P;g)\Omega_0,\Omega_1\in \Omega^2(P;\mathfrak g) (see ).

Let pSymk(g)Gp\in \operatorname{Sym}^k(\mathfrak g^*)^G be an Ad\operatorname{Ad}-invariant homogeneous polynomial of degree kk, so that the associated p(Ω)p(\Omega) is a closed 2k2k-form on MM (via basicness on PP), as in .

Define the difference

η:=ω1ω0Ω1(P;g). \eta := \omega_1-\omega_0 \in \Omega^1(P;\mathfrak g).

By , η\eta is horizontal and equivariant (hence “tensorial”), so it corresponds to an ad(P)\operatorname{ad}(P)-valued 1-form on MM.

Consider the straight-line path of connections

ωt:=ω0+tη,t[0,1], \omega_t := \omega_0 + t\,\eta,\qquad t\in[0,1],

with curvature

Ωt:=dωt+12[ωtωt] \Omega_t := d\omega_t + \tfrac12[\omega_t\wedge \omega_t]

(compare ).

Change-of-connection (transgression) formula

Define the transgression form

Tp(ω0,ω1):=k01p(ηΩtk1)dt, T_p(\omega_0,\omega_1) := k \int_0^1 p\bigl(\eta \wedge \Omega_t^{k-1}\bigr)\,dt,

a (2k1)(2k-1)-form on PP which is basic and therefore descends to a (2k1)(2k-1)-form on MM (often also denoted Tp(ω0,ω1)T_p(\omega_0,\omega_1)).

Then the characteristic forms satisfy

p(Ω1)p(Ω0)  =  dTp(ω0,ω1), p(\Omega_1) - p(\Omega_0) \;=\; d\,T_p(\omega_0,\omega_1),

so the two forms differ by an exact form on MM. In particular, the de Rham class of p(Ω)p(\Omega) does not depend on the chosen connection, which is the mechanism behind .

This construction is the standard transgression mechanism (see and ), and in low degrees it produces the usual .

Examples

  1. Degree 1 (abelian-style) case.
    If k=1k=1 and p:gRp:\mathfrak g\to\mathbb R is an Ad\operatorname{Ad}-invariant linear functional, then

    Tp(ω0,ω1)=p(η),p(Ω1)p(Ω0)=dp(η). T_p(\omega_0,\omega_1)=p(\eta),\qquad p(\Omega_1)-p(\Omega_0)=d\,p(\eta).

    For G=U(1)G=U(1) this recovers the familiar fact that changing a connection 1-form changes the curvature 2-form by an exact 2-form (compare the local picture in ).

  2. Degree 2 and the Chern Simons 3-form.
    For k=2k=2 and an invariant quadratic polynomial pp (for instance a suitably normalized trace form on a matrix Lie algebra), the transgression is a 3-form

    Tp(ω0,ω1)=201p(ηΩt)dt, T_p(\omega_0,\omega_1) = 2\int_0^1 p(\eta\wedge \Omega_t)\,dt,

    and the formula says p(Ω1)p(Ω0)p(\Omega_1)-p(\Omega_0) is exact with this primitive. On a trivial bundle, taking ω0=0\omega_0=0 and ω1=A\omega_1=A yields the usual Chern Simons expression in terms of AA and dAdA, matching the standard on a chart.

  3. Pontryagin forms from different connections.
    Let EME\to M be a real vector bundle with two 0,1\nabla^0,\nabla^1. Using the induced connections on the (via ), the change-of-connection formula implies that the associated Pontryagin forms differ by an exact form. Hence the resulting are independent of the chosen connection.