I have posted this question on math.stackexchange, without success. I'll make it brief:

Let $E\rightarrow M$ be an orientable vector bundle of rank n equipped with some Riemannian metric, $P:=F_{SO(n)}(E)$ its orthonormal frame bundle. Set $P^{c}:=F_{SO(n)}(E)\times_{SO(n)} SO(n,\mathbb{C})$.

I am trying to understand why the following statement is true:

If we have a principal connection $\omega^{c}$ on $P^{c}$, we can decompose it uniquely into $(\omega,\phi)$, with $\omega$ a principal connection on $P$ and $\phi\in \Omega^{1}_{M}(iad P)$.

What I understand is how to decompose the connection so that $\phi\in\Omega^1_M(iadP^c)$. It should be possible in general, as long as we have a setting where $W\subset Q$ is a reduction of principal bundles s.t. the quotient of their respective lie groups form a reductive space.

I do not understand how we can reduce $\phi$ even further. Is it possible that $w^c$ needs to be flat?

Thank you very much.

  • $\begingroup$ I don't understand your question. You want to know why you can decompose the connection on $P^c$. You know how to decompose it. Do you not know how to prove uniqueness of the decomposition? Or do you want to further decompose $\phi$ (which is not part of the statement you want to prove)? The adjoint representation of $SO(n)$ doesn't decompose at all, unless $n=4$. $\endgroup$
    – Ben McKay
    Apr 23, 2014 at 19:59
  • $\begingroup$ In my (false) previous understanding, $\phi$ took values in the wrong bundle. Andy Sanders answer cleared things up for me! $\endgroup$ Apr 24, 2014 at 10:01

1 Answer 1


The bundle $P$ is obtained by reducing the structure group of the $SO(n,\mathbb{C})$ bundle $P^c$ to the maximal compact subgroup $SO(n).$ Let's call the inclusion \begin{align} i: P\rightarrow P^c. \end{align} Pulling back the connection form gives $i^{*}(\omega^c)\in \Omega^1(P,\mathfrak{so}_n(\mathbb{C})).$ The Lie algebra of $SO(n,\mathbb{C})$ admits an invariant (with respect to the Adjoint action of $SO(n)$) splitting, \begin{align} \mathfrak{so}_n(\mathbb{C})=\mathfrak{so}(n)\oplus i\mathfrak{so}(n). \end{align} Decomposing the pull-back into these pieces gives, \begin{align} i^{*}\omega^c=\omega + \phi. \end{align} Then, $\omega$ is a connection on $P$ and $\phi$ descends to the base taking values in the adjoint bundle, \begin{align} P\times_{Ad(SO(n))} i\mathfrak{so}(n)=i\ \text{ad}(P). \end{align} That is to say, $\phi\in \Omega^1(M,i\ \text{ad}(P)).$

If the original connection is flat, then the pair $(\omega,\phi)$ satisfy some equations: \begin{align} F(\omega)+\frac{1}{2}[\phi,\phi]=0 \\ d^{\omega}\phi=0. \end{align} Here, $F(\omega)$ is the curvature of $\omega,$ the bracket operation combines the wedge product of forms and the Lie bracket, hence is symmetric so that the term $[\phi,\phi]$ does not (necessarily) vanish. The operator $d^{\omega}$ is the exterior covariant derivative operator associated to the connection $\omega.$ Notice that the first equation is of two forms with values in $ad(P)$ and the latter is of two forms with values in $i\ ad(P).$ The first equation is one of a pair of equations known as Hitchin's self-duality equations. You can read all about them (in the case of $SO(3,\mathbb{C})$) in the original beautiful paper of Hitchin: http://people.maths.ox.ac.uk/hitchin/hitchinlist/Hitchin%20THE%20SELF-DUALITY%20EQUATIONS%20ON%20A%20RIEMANN%20SURFACE%20(PLMS%201987).pdf

  • $\begingroup$ I understand it now, thank you! Sadly, I can't upvote you yet (too little rep). :( $\endgroup$ Apr 24, 2014 at 10:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.