2

Let $\pi:P\longrightarrow M$ be a principal $G-$bundle over a manifold $M$ (you can impose some extra conditions on $M$, such as compact, oriented, Riemannian, etc.), where $G$ is a complex Lie group. Let $P^r\subset P$ be a subbundle with the structure group $G^r$, where $G^r$ is a real form of $G$, i.e, the complexification of $G^r$ is $G$.

Then the affine space of $G-$connections on $P$, say $\mathcal{A}(P)$, can be viewed as a cotangent bundle $$\mathcal{A}(P)=\mathcal{A}(P^r)\times \Omega^1(M;\mathrm{ad}P^r)=T^*\mathcal{A}(P^r)$$ which is a natrural symplectic manifold with the standard symplectic structure $\omega$. Let $\mathcal{G}^r$ be the gauge group of the bundle $P^r$, it acts by gauge transformations on $T^*\mathcal{A}(P^r)$.

Is this action a Hamiltonian action? If so, how to compute the moment map $\mu:\mathcal{A}(P)\longrightarrow \mathrm{Lie}(\mathcal{G}^r)\cong\Gamma(\mathrm{ad}P^r)$?

Z. Liu
  • 843
  • When $M$ is a compact Kähler manifold, then yes. This is essentially the story of Higgs bundles over Kähler manifolds, and the Kähler form is used for defining the momentum map. So I don't think this would be true in much more generality, although the Kähler condition can probably be loosened somewhat. – Quaere Verum Jul 26 '24 at 16:21
  • @QuaereVerum I think the essence of the requirement of a Kähler structure is the metric induced by the Kähler form. I add an answer (where I proved it is true) under the assumption that $(M,g)$ is compact oriented Riemannian manifold. – Z. Liu Jul 27 '24 at 12:30
  • 1
    Your argument seems correct. In the argument that I am familiar with, the Lie algebra is identified with sections of the top exterior power of the adjoint bundle, and then the momentum map is $A\mapsto F_A\wedge\omega^{n-1}$ with $\omega$ being the Kahler form. But using the adjoint of the differential seems to work out just fine, so good on you, answering your own question. – Quaere Verum Jul 27 '24 at 13:54
  • @QuaereVerum Yes, the top exterior power of adjoint bundle can be viewed as the dual of the Lie algebra of the gauge group. In this setting, the paring of $\mathrm{Lie}(\mathcal{G}^r)$ and its dual is given by integration over $M$, and in the Kahler case, they can be identified by Hodge star operator. In Kahler case, to agree with the expression in my argument, we can take $d^_a\beta$ which is $d_a\beta\wedge\omega^{n-1}$, where $d_a\beta$ can be understood as the curvature of the "imaginary part" of the connection $A$. – Z. Liu Jul 27 '24 at 17:40
  • Yes that's right, so this more general argument is pretty neat, good job :) – Quaere Verum Jul 27 '24 at 18:30

1 Answers1

2

I think it is true when we assume $M$ is a compact oriented Riemannian manifold. Under this assumption, we can define an $L^2$-inner product on the space $\Omega^k(M;\mathrm{ad}P^r)$ by: $$\begin{aligned}(\xi_1,\xi_2)_{L^2}&:=\int_M\langle\xi_1,\xi_2\rangle_x\mathrm{vol_M}\\&=\int_M\xi_1\wedge *\xi_2\end{aligned}$$ where $\mathrm{vol}_M$ is Riemannian volume form on $M$, $*$ is the Hodge *-operator and $\langle\cdot,\cdot\rangle_x$ is the fiber-wise $\mathrm{Ad}G^r-$invariant inner product on $\mathrm{ad}P^r$.

In particular, we can identify the Lie algebra $\mathrm{Lie}(\mathcal{G}^r)\cong\Omega^0(\mathrm{ad}P^r)$ with its dual, it will be convenient for us write a moment map.

Now, to proof my claim, I need to establish some computaional results at first.

Recall that a connection in $\mathscr{A}(P)$ is a $\mathfrak{g}-$valued 1-form defined on the principal bundle $\pi:P\longrightarrow M$, by identifying $$\mathscr{A}(P)=T^*\mathscr{A}(P^r)=\mathscr{A}(P^r)\times \Omega^1(M;\mathrm{ad}P^r)$$ we can write the connection $A$ as $$A=a+(\pi^*\beta)i:=(a,\beta)$$ where $a$ is a connection on the subbundle $P^r$, $\beta\in\Omega^1(M;\mathrm{ad}P^r)$ is an $\mathrm{ad}P^r-$valued 1-form on $M$.

A tangent vector $\xi\in T_A\mathscr{A}\cong\Omega^1(M;\mathrm{ad}P)$ can be decoposed as $$\xi=(\xi_1,\xi_2)\in T_a\mathscr{A}(P^r)\times\Omega^1(M;\mathrm{ad}P^r)=\Omega^1(M;\mathrm{ad}P^r)\times \Omega^1(M;\mathrm{ad}P^r)$$ where $\xi_1$ is simply $(dp_1)_A\xi$, and $p_1:A=(a,\beta)\mapsto a$ is the projection onto the 1st component.

Lemma 1: For $X\in\mathrm{Lie}(\mathcal{G}^r)$, the fundamental vector field $\underline{X}$ at a connection $A=(a,\beta)$ is $$\underline{X}(A)=(d_aX,0)\in T_a\mathscr{A}(P^r)\times\Omega^1(M;\mathrm{ad}P^r)$$where $d_a$ is the induced de Rham differential $\Omega^k(M;\mathrm{ad}P^r)\longrightarrow\Omega^{k+1}(M;\mathrm{ad}P^r)$.

The proof of this lemma is a very common argument in gauge theory. We shall notice that $\underline{X}(A)$ has only the first component is becasue the gauge group $\mathcal{G}^r$ acts by restriction of the total gauge group $\mathcal{G}(P)$, hence it acts by gauge transformation on $a$ but trivial on $\beta$.

Lemma 2: The Liouville 1-form $\alpha$ on $\mathscr{A}(P)$ is $$\alpha_A(\xi)=(\beta,\xi_1)_{L^2}$$ where $A=(a,\beta)$, $\xi_1=(dp_1)_A\xi$ is the first component of the tangent vector $\xi$.

This is can be computed directly by the definition of Liouville 1-form on cotangent bundle.

Lemma 3: For two vector fields $\xi,\eta$ on $\mathscr{A}(P)$, the value of $\xi(\alpha(\eta))$ at the connection $A=(a,\beta)$ is $$\alpha_A([\xi,\eta])+(\eta_1,\xi_2)_{L^2}$$

Proof: Let $\gamma(t)=(\gamma_1(t),\gamma_2(t))$ be the curve on $\mathscr{A}(P)$ such that $$\xi\circ \gamma(t)=\gamma'(t),\quad \gamma(0)=A$$ then by definition and Leibniz law, we have $$\begin{aligned}\xi(\alpha(\eta))(A)&=\left.\frac{d}{dt}\right|_{t=0}\left((dp_1)_{\gamma(t)}\eta(\gamma(t)),\gamma_2(t)\right)_{L^2}\\&=\alpha_A([\xi,\eta])+(\eta_1,\xi_2)_{L^2}\end{aligned}$$ . $\clubsuit$

Hence we can write the symplectic form immediately:

Corollary 4: The standard symplectic form on $\mathscr{A}(P)$ is $$\begin{aligned}\omega_A(\xi,\eta)&=(d\alpha)_A(\xi,\eta)\\&=\alpha_A([\xi,\eta])+(\eta_1,\xi_2)_{L^2}-(\eta_2,\xi_1)_{L^2}\end{aligned}$$ where $\xi,\eta\in T_A\mathscr{A}(P)$

With these results in mind, we can compute what is $\iota_{\underline{X}}\omega$. For $X\in\mathrm{Lie}(\mathcal{G}^r)$, and for an arbitrary vector field $\eta$ on $\mathscr{A}(P)$, we have at $A=(a,\beta)$ that $$\begin{aligned}(\iota_{\underline{X}}\omega)_A(\eta)&=\omega_A((d_aX,0),\eta)\\&=\alpha_A([d_aX,\eta])-(\eta_2,d_aX)_{L^2}\end{aligned}$$

Finally, I claim that the map $$\begin{aligned}\mu:\mathscr{A}(P)&\longrightarrow\mathrm{Lie}(\mathcal{G}^r)\\A=(a,\beta)&\mapsto d_a^*\beta\end{aligned}$$ is the moment map associated to the gauge action of $\mathcal{G}^r$. Where $d_a^*=-*\circ d_a\circ *$ is the Hodge co-differential.

It suffices to compute $((d\mu)_A\eta,X)_{L^2}$. Indeed, a parametrized curve $\gamma(t)$ in $\mathscr{A}(P)$ passing through $A$ with $\gamma'(0)=\eta$ is $$\begin{aligned}A+t\eta&=a+(\pi^*\beta)i+\pi^*(\eta_1+i\eta_2)t\\&=a+t\pi^*\eta_1+i\pi^*(\beta+\eta_2t)\end{aligned}$$ hence by the formal adjointness of the Hodge co-differential and Leibniz law, we have $$\begin{aligned}((d\mu)_A\eta,X)_{L^2}&=\left(\left.\frac{d}{dt}\right|_{t=0}d^*_{a+t\pi^*\eta_1}(\beta+\eta_2t),X\right)_{L^2}\\&=\left.\frac{d}{dt}\right|_{t=0}\left(\beta+\eta_2t,d_{a+t\pi^*\eta_1}X\right)_{L^2}\\&=(\eta_2,d_aX)_{L^2}+\left(\beta,\left.\frac{d}{dt}\right|_{t=0}\underline{X}(\gamma(t))\right)_{L^2}\\&=(\eta_2,d_aX)_{L^2}-\alpha_A([d_aX,\eta])\\&=-(\iota_\underline{X}\omega)_A(\eta)\end{aligned}$$ It is indeed the moment map.

To sum up, we obtained the following result:

Theorem 5: For a complex Lie group $G$, and a principal $G-$bundle $P$ over a compact oriented Riemannian manifold $(M,g)$, the restriction action of a real gauge group $\mathcal{G}^r$ on the affine space of connections $\mathscr{A}(P)$ is Hamiltonian. The moment map is given by $$\mu(A)=\mu(a+(\pi^*\beta)i)=d_a^*\beta$$

Z. Liu
  • 843