Unclear construction in Hitchins paper on Integrable Systems

107 Views Asked by At

In the paper Stable Bundles and Integrable Systems (1987), Hitchin shows that the cotangent bundles of moduli spaces of stable bundles can be viewed as integrable systems.

In section 4 of the paper, he constructs a mapping $$p:H^0(M;\ \text{ad}P\otimes K)\to H^0(M;\ K^d)$$ based on an invariant homogeneous polynomial $p$ of degree $d$ on a Lie algebra $\mathfrak{g}$ (which he later uses to construct hamiltonians). He justifies this construction by stating that the vector bundle $\text{ad}P$ is a bundle of Lie algebras each isomorphic to $\mathfrak{g}$.

Here, $P$ is a stable holomorphic principal $G$-bundle over a compact Riemann surface $M$ of genus $g>1$, with $G$ a semisimple complex Lie group, and $K$ is the canonical line bundle on $M$.

I do not understand how exactly we construct a mapping $H^0(M;\ \text{ad}P\otimes K)\to H^0(M;\ K^d)$, given the polynomial $p$?

All other discussions of the Hitchin system that I have seen (that came out some years after Hitchins original paper, for example here and here) employ a notion of Higgs bundles in the construction of the Hitchin system and do not directly mention the above mapping. I would therefore be very grateful for help.

Edit: Previously, I also included the following question (resolved due to comments below): I understand that each fibre of the associated bundle $\text{ad }P$ is a Lie algebra; however, I am not sure how to show that they are mutually isomorphic, and hence all isomorphic to a Lie algebra $\mathfrak{g}$. Furthermore, is $\mathfrak{g}$ in this case the Lie algebra associated to $G$ (I don't believe that it is for principle bundles and adjoint representations in general)?

1

There are 1 best solutions below

6
On

It is probably helpful to consider the case of vector bundles first, where we can replace $\text{Ad}(P)$ by $\text{End}(E)$ for a given vector bundle $E\to\Sigma$. In this case, a section of $\text{End}(E)$ can be viewed as a family of matrices which is parameterised by the base $\Sigma$. Say $\varphi\in H^{1,0}(\Sigma,\text{End}(E))\cong H^0(\Sigma,\text{End}(E)\otimes K)$ (Serre duality) and represent it by a $1$-form also denoted $\varphi$. Then we can define $\varphi\mapsto\text{tr}(\wedge^d\varphi)$ where $\wedge^d\varphi:\wedge^d E\to\wedge^d(E\otimes K)$. But now, we can use the fact that $\wedge^d(E\otimes K)=\wedge^dE\otimes K^{\otimes d}$, which is true because $K$ is a line bundle (see here for instance). So locally, we have that $\wedge^d\varphi$ can be written as $\sum A_i\otimes\eta_i$ with $A_i$ an endomorphism of $\wedge^dE$, and $\eta_i$ a section of $K^{\otimes d}$. The trace map is obtained by just taking the trace of each $A_i$. So we are left with $\sum \text{tr}(A_i)\eta_i\in H^0(U,K^{\otimes d})$, and likewise globally $\text{tr}(\wedge^d\varphi)\in H^0(\Sigma,K^{\otimes d})$.

So when $\mathfrak{g}=\text{End}(V)$ for a vector space $V$, then we have $\det(\lambda-\varphi)=\lambda^r+\sum_{k=1}^r\text{tr}(\wedge^k\varphi)\lambda^{r-k}$ and so these coefficients (which are known to be $\text{Ad}$-invariant homogeneous polynomials) are a natural choice. When we look at more general principal bundles, we have to fix some $\text{Ad}$-invariant homogeneous polynomials on the Lie algebra $\mathfrak{g}$. But besides that, the idea is exactly the same and you should be able to copy the construction I outlined above in the case of principal bundles. I'll leave that for you to write down as it's probably a useful exercise to make sure you understood the concept.