3
$\begingroup$

Let $V$ and $W$ be two vector spaces over field $\mathbb{k}$. Suppose $\beta : V \otimes W \rightarrow \mathbb{k}$ is a non-degenerate bilinear pairing in the variable $V$, then it means that for $0 \ne v \in V$, there exists an element $w \in W$ such that $v \otimes w \mapsto k$ where $k \ne 0$. This is the definition I remember (I hope I got it correct. Please correct me if I need to refine it).

However, recently I saw that there is another definition on non-degenerate bilinear pairing. It goes like this:

Let $V$ and $W$ be two vector spaces over field $\mathbb{k}$. A pairing $\beta : V \otimes W \rightarrow \mathbb{k}$ is called non-degenerate in the variable $V$ if there exists a linear map $\gamma : \mathbb{k} \rightarrow W \otimes V$, called co-pairing, such that the following composite is equal to the identity map of $V$:

\begin{equation*} V = V\otimes \mathbb{k} \xrightarrow[]{id_V \otimes \gamma} V \otimes (W \otimes V) = (V \otimes W) \otimes V \xrightarrow[]{\beta \otimes id_V} \mathbb{k} \otimes V = V \end{equation*}

Is this definition talking about the same non-degenerate bilinear pairing as the one I have in mind? If it is, then how are they equivalent? (They felt equivalent to me, but I don't see how)

$\endgroup$
0

1 Answer 1

4
$\begingroup$

If $\gamma \colon \mathbb{k} \to W \otimes V$ is a linear map then we dentote the composition $$ V = V \otimes \mathbb{k} \xrightarrow{\operatorname{id}_V \otimes \gamma} V \otimes W \otimes V \xrightarrow{\beta \otimes \operatorname{id}_V} \mathbb{k} \otimes V = V $$ by $f_\gamma$. Notice that if $\gamma(1) = \sum_{i=1}^n w_i \otimes v_i$ then $$ f_\gamma(v) = \sum_{i=1}^n \beta(v,w_i) v_i. $$


Suppose that such a a copairing exists, and let $v \in V$ with $v \neq 0$. If $\gamma(1) = \sum_{i=1}^n w_i \otimes v_i$ then $$ 0 \neq v = f_\gamma(v) = \sum_{i=1}^n \beta(v,w_i) v_i, $$ so at least one of the coeffients $\beta(v,w_i)$ must be nonzero. Hence $\beta$ is non-degenerate in $V$.


For the existence we need $V$ to be finite-dimensional:

Notice that if $\gamma \colon \mathbb{k} \to W \otimes V$ is any linear map, then for $\gamma(1) = \sum_{i=1}^n w_i \otimes v_i$ we have by the above formula that $$ f_\gamma(v) = \sum_{i=1}^n \beta(v,w_i) v_i \in \langle v_1, \dotsc, v_n \rangle_{\mathbb{k}} \quad \text{for all $v \in V$}. $$

Suppose that $\gamma$ is a copairing. Then $f_\gamma(v) = v$ for all $v \in V$, and we get that $v \in \langle v_1, \dotsc, v_n \rangle_\mathbb{k}$ for all $v \in V$. Thus $\langle v_1, \dotsc, v_n \rangle_\mathbb{k} = V$, i.e. $V$ is spanned by the finitely many vectors $v_1, \dotsc, v_n$, and is therefore finite-dimensional.

That shows that the existence of a copairing implies that $V$ is finite-dimensional.


So suppose that $V$ is finite-dimensional, say $\dim V =: n$, and that $\beta$ is non-degenerate in $V$.

Claim. The map $B \colon W \to V^*$, $w \mapsto \beta(-,w)$ is surjective.

Proof. Suppose $B$ is not surjective. Then $\operatorname{im} B$ is a proper subspace of $V^*$, say $\dim \operatorname{im} B =: r < n$. We take a basis $(\phi_1, \dotsc, \phi_r)$ of $\operatorname{im} B$ and extend this to a basis $(\phi_1, \dotsc, \phi_n)$ of $V^*$.
Let $(v_1, \dotsc, v_n)$ be the corresponding dual basis of $V$, i.e. for all $i,j = 1, \dotsc, n$ we have $\phi_i(v_j) = \delta_{ij}$. Then $\phi_i(v_n) = 0$ for all $i = 1, \dotsc, r$ and thus $\phi(v_n) = 0$ for all $\phi \in \operatorname{im} B$. With this we have $v_n \neq 0$ but $$ \beta(v,w) = B(w)(v) = 0 \quad \text{for all $w \in W$}, $$ contradicting the non-degeneracy of $\beta$ in $V$.

Let $(v_1, \dotsc, v_n)$ be an ordered basis of $V$ and denote by $(v_1^*, \dotsc, v_n^*)$ the corresponding dual basis of $V^*$, i.e. $v_i^*(v_j) = \delta_{ij}$ for all $i,j = 1, \dotsc, n$. Because $B$ is surjective there exists for every $i = 1, \dotsc, n$ some $w_i \in W$ with $v_i^* = B(w_i) = \beta(-,w_i)$.

Let $\gamma \colon \mathbb{k} \to W \otimes V$ be the unique linear map with $\gamma(1) = \sum_{i=1}^n w_i \otimes v_i$. Then for all $j = 1, \dotsc, n$ we have $$ f_\gamma(v_j) = \sum_{i=1}^n \beta(v_j, w_i) v_i = \sum_{i=1}^n B(w_i)(v_j) v_i = \sum_{i=1}^n v_i^*(v_j) v_i = \sum_{i=1}^n \delta_{ij} v_i = v_j, $$ and thus $f(v) = v$ for all $v \in V$ by the linearity of $f$.


So the existence of a copairing is equivalent to $V$ being finite-dimensional and $\beta$ being non-degenerate in $V$.

$\endgroup$
4
  • 1
    $\begingroup$ I see, thanks for the explanation. But there are a few parts I need clarifications. Firstly, "So for $\gamma$ to be a copairing we need $V$ to be finite-dimensional", why is it so? Is it because $n$ is finite? Why can't $n$ be infinite? Lastly, there you wrote $\beta(−,w_i)$. What does the dash represents? Is it a place holder for something not important? $\endgroup$
    – Jerry
    Aug 14, 2016 at 1:47
  • 1
    $\begingroup$ (1) I added a short diagram about $V$ being finite-dimensional. (2) Every element of the tensor product $W \otimes V$ is a finite sum simple tensors (i.e. elements of the form $w \otimes v$ with $w \in W$ and $v \in V$), hence the element $\gamma(1)$ is a finite sum $\gamma(1) = \sum_{i=1}^n w_i \otimes v_i$ with $w_1, \dotsc, w_n \in W$ and $v_1, \dotsc, v_n \in V$. It does not make sense for $n$ to be infinite, because infinite sums make no sense. $\endgroup$ Aug 14, 2016 at 4:21
  • 1
    $\begingroup$ (3) If $w \in W$ then I denote by $\beta(-, w)$ the linear map $V \to \mathbb{k}$ which is given by $v \mapsto \beta(v,w)$. So the dash is a place holder for the elements of $V$. $\endgroup$ Aug 14, 2016 at 4:22
  • 2
    $\begingroup$ Ah yes, because by definition an element of a vector space is a linear combination of finite subset of a basis. And now I see how the notation $\beta(-,w)$ works. Thanks! $\endgroup$
    – Jerry
    Aug 14, 2016 at 4:41

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .