4

I am reading the fifth chapter of "An introduction to extremal Kaehler metrics" by Gabor Szekelyhidi. At the very beginning of that chapter, the author describes moment maps and Hamiltonian action. Here is a short description: Suppose that a connected Lie group $G$ acts on a Kaehler manifold $M$ preserving the Kaehler form $\omega.$ The derivative of the action gives rise to a Lie algebra map

$$ \rho: \mathfrak{g} \rightarrow \mathfrak{X}(M)$$

Where $\mathfrak{g}$ is the Lie algebra of $G$ and $\mathfrak{X}$ is the algebra of vector fields on $M.$ Then the action of $G$ on $M$ is said to be Hamiltonian if there exists a $G-$equivariant map

$$ \mu: M \rightarrow \mathfrak{g}^{*} $$

such that for any $\xi \in \mathfrak{g}$ the function $\langle\mu, \xi\rangle$ is a hamiltonian function for the vector field $\rho(\xi):$

$$ d\langle\mu, \xi\rangle = -\iota_{\rho(\xi)} \omega. $$

I got stuck with the following exercise: Let $M_n$ be the set of $n \times n$ complex matrices equipped with the Euclidean metric under the identification $M_n=\mathbb{C}^{n^2}.$ The unitary matrices $U(n)$ act on $M_n$ by conjugation, preserving this metric. I.e. $A \in U(n)$ acts by $M \mapsto A^{-1}MA.$ Find a moment map $\mu: M_n \rightarrow \mathfrak{u}(n)^{*}.$ So far, I've just noticed that if $n=1$ then the action of $U(1)$ by conjugation on complex number is the trivial action, what will be then the moment map? I am trying to see what happens also in dimension 2. I know that $\mathfrak{u}(2)$ is generated by skew-Hermitian matrices, so every skew-hermitian matrix can be diagonalized and the eigenvalues must be pure imaginary. How to find a moment map in this case?

davyjones
  • 659
Salvatore
  • 1,302
  • 1
    A definition of the moment map would be useful for those not enough fluent with the subject. – lisyarus Feb 07 '18 at 13:19
  • Your moment map for the 1-dimensional case does not work. In fact, it does not even make any sense - do you know what the Lie algebra looks like? – Amitai Yuval Feb 07 '18 at 16:12
  • the Lie algebra of U(1) is just the real line or the real line multiplied by the imaginary unit. Why it does not make sense? – Salvatore Feb 07 '18 at 16:14
  • Yes, the Lie algebra is a 1-dimensional real vector space, and so is its dual. How can $\mu(z)=z$ make any sense? – Amitai Yuval Feb 07 '18 at 16:53
  • You are totally right, I will edit my post right now – Salvatore Feb 07 '18 at 16:55
  • So, regarding the 1-dimensional case: the action is trivial, and this means that the vector field corresponding to any element in the Lie algebra is the zero vector field. Hence, the zero map seems like a promising candidate. – Amitai Yuval Feb 07 '18 at 19:20

3 Answers3

4

Ok, so first, you should know what is the Kähler metric that you are defining on the space $M_n$. Since it is not specified, I will assume it is the induced by the standard hermitian product $$h(A,B)=Tr(AB^\dagger)$$, where $B^\dagger$ denotes the adjoint (that is the transpose conjugate) of $B$. This of course has associated a scalar product $$ g(A,B)=Re(Tr(AB^\dagger))$$ and a symplectic form $$ \omega(A,B)=-Im(Tr(AB^\dagger))$$ (the minus sign here is just a convention).

As you noted, there $U(n)$ acts symplectically on $M_n$ via $A\mapsto U^\dagger A U$. Now, to see how the infinitesimal associated action $\mathfrak{u}(n)\rightarrow M_n$ is, just compute $$ \rho_X(A)=\left.\frac{d}{dt}\right|_{t=0}(e^{-Xt}Ae^{Xt})=-XA+AX=[A,X]. $$

Now we want to find a moment map $\mu:M_n \rightarrow \mathfrak{u}(n)^{\ast}$ for this action. To do this we will identify $\mathfrak{u}(n)$ with its dual using the pairing given by the hermitian metric, that is $$ \langle X,Y \rangle = Tr(XY^\dagger). $$ Therefore, I claim that the momentum map is given by the mapping $\mu:M_n \rightarrow \mathfrak{u}$ such that $$\mu(A)=\tfrac{i}{2}[A,A^\dagger].$$

Now note that $$ f_X(A)=\langle \mu(A),X \rangle = \tfrac{i}{2} Tr([A,A^\dagger]X), $$ so (check this out) $$ df_{X,A}(Y)=\left.\frac{d}{dt}\right|_{t=0}f_X(A+tY)= \tfrac{i}{2}Tr([Y,A^\dagger]X+[A,Y^\dagger]X). $$ Now, using the cyclic property of the trace ($Tr(ABC)=Tr(BCA)=Tr(CAB)$), is easy to check that this is indeed $$ df_{X,A}(Y)=-Im(Tr([X,A]Y^\dagger))=\omega([X,A],Y)=-\omega(\rho_X(A),Y). $$ So $\mu$ is indeed a moment map for this action.

This example is indeed very useful, for example, to find the moment map that gives Hitchin equations for Higgs Bundles, check out Section 4 on Hitchin's paper "The self-duality equations on a Riemann surface" if you want to read more about this.

  • Thanks for your very interesting answer. I noticed that during your carrying out the exercise, you claimed that the moment map has to have such a form (you write down the equation) and finally you checked that it satisfies the required condition. I was wondering if there is a way to understand the form of the moment map by using its definition property. It seems to me that in general it's not so obvious to give the exact form of the moment map and then check that it satisfies the properties. I hope that you get what I mean. – John117 Nov 28 '20 at 11:31
  • @John117 I understand that what you are asking is why one would know beforehand what the moment map should be. In this particular case I guess one can just deduce it easily as in the cited Hitchin’s paper (page 78). I really do not think I can give a satisfying answer for the general case... However, at least in examples coming from classical mechanics, moment maps are obtained as the conserved quantities yielded by Noether’s theorem. – G. Gallego Nov 29 '20 at 18:49
1

For the Hermitian case, there is a more direct formula which can avoid using the dual $\mathfrak{g}*$, and the tangent map.

Let's assume $(V,h)$ is an Hermitian vector space (hence $\mathrm{Im} h$ is a symplectic form) which endows with an Hermitian $G$-action (hence also Hamiltonian), pick $\langle\cdot,\cdot\rangle$ be an Ad-invariant Euclidean metric on $\mathfrak{g}$, then the moment map computes as $$\langle\mu(x),\xi\rangle=\frac{i}{2}h\left(\underline{\xi}(x),x\right)$$ where $\underline{\xi}(x)$ means the fundamental vector field associated to the $G$-action.

Notice that this formula makes sense, as the RHS is indeed taking values in $\Bbb{R}$, because the action is $G\longrightarrow U(V)$, thus the infinitesimal action is $\mathfrak{u}(V)$, which makes $\underline{\xi}(\cdot)$ a skew-Hermitian operator.

The proof of above formula is a straightforward computation.

Now back to our case, the $V$ is $\text{End}(\Bbb{C}^n)$, the Hermitian product is given by $$h(X,Y)=\text{Tr}(XY^*)$$ the Lie group $G$ is $U(n)$, the Ad-invariant metric on its Lie algebra is $$\langle\xi,\zeta\rangle=\text{Tr}(\xi\zeta^*)$$ the fundamental vector field is $$\underline{\xi}(X)=[\xi,X]$$ therefore, the moment should satisfy $$\begin{aligned}\text{Tr}(\mu(X)\xi^*)&=\frac{i}{2}\mathrm{Tr}\left([\xi,X]X^*\right)\\ &=-\frac{i}{2}\text{Tr}([X,X^*]\xi^*)\end{aligned}$$ for all $\xi$. So $$\mu(X)=-\frac{i}{2}[X,X^*]$$

Dialektik
  • 160
0

It is easier to compute the moment map of the left and right action $(A,M)\to M.A$ and $(A,M)\to A^{-1}M$, and one find respectively ${i\over 2} M M^*, {-i\over 2}M^*M)$, so that the moment map is ${i\over 2}(MM^*-M^*M)$.

Thomas
  • 8,588