Adjoint matrix with respect to the Hermitian product

566 Views Asked by At

Consider a Hermitian vector bundle $\pi: E \to X$ where $X$ is a complex manifold. Being Hermitian means that there exists a hermitian inner product for each fiber $\pi^{-1}(p)$. So if $X,Y \in \mathbb{C}^n$ for even $n$ are two sections on the fiber $E_p$ over $p \in X$ then the Hermitian inner product is defined as $$ (X,Y) = X^T H \bar{Y} $$ where the bar means complex conjugation. I am bit confused now on what this actually means. The product is defined w.r.t. a Hermitian matrix $H$. Does this $H$ vary as we move around the base? And, of course, what happens if the fibers are isomorphic to $\mathbb{C}^{r \times r}$? This is still a complex vector space. What is the definition of the inner product there?

The reason I am confused is because in the book Kobayashi-Hitchin correspondence by Lubke and Teleman, in page 20 they define a bilinear form $$ (f_1,f_2)= \int_X \text{tr} (f_1 \cdot f_2^{*}) \text{vol}_g $$ where $g$ is a Riemannian metric on $X$ and $f^{*}$ means adjoint of $f$ with respect to $g$. I do not understand this either :(

Any help would be appreciated.

1

There are 1 best solutions below

2
On BEST ANSWER

I'll break down the linear algebra first. Then the point will be that what I'm saying holds on each fiber, as long as we imagine the hermitian matrix $H$ as varying with the base (read: section of $\operatorname{End}(E)$).

Start with a vector space $V$ over $\mathbb{C}$. If $H$ is a hermitian matrix, we'll define a hermitian inner product $h(u,v) = v^T H \bar{u}$. Now, given any matrix $A \in \operatorname{End}(V)$, we can define its adjoint $A^*$ with respect to the metric $h$ by the formula $$ h(Au, v) = h(u, A^*v). $$ Then we can derive a formula for $A^*$: $$ \begin{align*} h(Au, v) &= v^T H \bar{A} \bar{u}\\ &= v^T (H\bar{A} H^{-1}) H \bar{u}\\ &= \bigg((H \bar{A} H^{-1})^T v \bigg)^T H \bar{u}\\ &= h(u, A^* v), \end{align*} $$ where now $A^* = (H \bar{A} H^{-1})^T = \bar{H} \bar{A}^T \bar{H}^{-1}$. This gives you a very concrete way of computing an adjoint of a matrix with respect to any hermitian inner product.

We can also define an induced hermitian inner product on the vector space $\operatorname{End}(V)$, which I will call $\tilde{h}$, and here's how I define it: if $A, B \in \operatorname{End}(V)$, then $$ \tilde{h}(A,B) = \operatorname{tr}(A^* B), $$ where $A^*$ is the same adjoint as above.

(By the way, a good exercise to do here is write down what $\operatorname{tr}(\bar{A}^T B)$ (so the simple case $H = \operatorname(id)$) is in terms of the entries of $A$ and $B$. You should find that it corresponds with the usual inner product of vectors in $\mathbb{C}^{r^2}$, so in some sense this is the most natural way we could have defined the inner product on $\operatorname{End}(V)$.)

So now think about a hermitian structure $h$ on a bundle $E\to X$. You can pretend that $h$ is defined by a matrix, which is absolutely true when you work in a local frame: if $e_1,\dots, e_r$ is a local frame for $E$, you get a matrix of smooth functions $H$, where the $i,j$ entry is the smooth function $H_{ij} = h(e_j,e_i)$.

Also, just like the linear algebra above, any endomorphism $f$ of $E$ has an adjoint $f^*$, defined again by $h(fs, t) = h(s,f^*t)$ for sections $s,t$ of $E$. In local coordinates we will again have the formula $$f^* = \bar{H} \bar{f}^T \bar{H}^{-1}.$$ Just beware: the transpose operation $(-)^T$ does not make sense globally! Only the adjoint $(-)^*$ makes sense globally.

Finally, this bilinear form that Teleman and Lübke define is just the globalized version of the induced metric $\tilde{h}$ I defined above: if $f_1,f_2$ are two endomorphisms, $\operatorname{tr}(f_1 f_2^*)$ defines a smooth function on $X$ (because fiberwise it gives a number), so integrating it over $X$ produces a number.