2
$\begingroup$

Let $V$ be a finite dimensional vector space over the field $K$, and let $g$ be a bilinear form on $V$, written $\langle,\rangle$. In the previous part: for each $w\in V$, the map $v\to\langle v,w\rangle$ is a functional $L_w$ on $V$, and the map $w\to L_w$ is a linear map of $V$ into the dual space $V^{*}$.(This was already proven in a previous exercise). Show that the following conditions are equivalent:

(i) The kernel of the map $L$ above is {0}

(ii)The map $L$ is an isomorphism between $V$ and $V^{*}$

(iii) If $C$ is the matrix representing the bilinear form with respect to a basis of $V$, then $\det(C)\neq 0$.

A bilinear form satisfying the preceding three conditions is said to be non-degenerate.

I started answering using (ii)

$\dim V=\dim V^{*}$, since w\in V by assumption.

$\ker(L_w)=0$ iff $w=0$

Therefore there exists an isomorphism that was already proven in the book.

(ii) implies (i), since $w\in V$, the $Ker{L}=0$ iff $w=0$ or $v=0$, in which in this case $v=w$. Therefore the $ker{L}=0$ for arbitrary $w$.

I make use of the following theorem for the relation (ii) implies (iii).

Theorem: Let $F:V\to W$ be a linear map, and assume that F is injective and surjective. Then F is invertible.

Proof:Each $w\in W$ matches a v once F is surjective and and the v is uniquely determined for each w once F is injective. If we consider $G:W\to V$, then $G=F^{-1}$, we need to prove G is linear.

If $w_1,w_2\in W$ then $F(v_1)=w_1$ and $F(v_2)=w_2$ and since F is linear $F(v_1+V_2)=F(v_1)+F(v_2)$ Therefore $G(w_1+w_2)=v_1+v_2=G(v_1)+G(v_2)$.

If $c\in K$ then $F(cv_1)=cF(v_1)=cw_1$. Therefore $G(cw_1)=cv_1=cG(w_1)$ $\blacksquare$

Since the map $w\to L_w$ is an isomorphism by (ii) implies the map is invertible. For an arbitrary bilinear matrix C to be invertible it requires $\det(C)\neq 0$, implying (iii).

Questions:

1) Is my usage of the mapping $w\to L_w$ correct? Does it solve the question?

2) Is my poof right?

3) I made used of non-degeneracy in the proof. What is supposed to mean "A bilinear form satisfying the preceding three conditions is said to be non-degenerate."

Thanks in advance!

$\endgroup$
3
  • $\begingroup$ @DietrichBurde It means exercise. $\endgroup$ Jul 24, 2017 at 18:32
  • 1
    $\begingroup$ And what means "the map $L$ above"? I can't find a definition of $L$. $\endgroup$ Jul 24, 2017 at 18:37
  • $\begingroup$ @DietrichBurde I appreciate your attention. My sincere apologies. I forgot to write vital information for the exercise propose. Please check my new edit. $\endgroup$ Jul 24, 2017 at 20:04

1 Answer 1

2
$\begingroup$

I'm afraid you did not prove the statement.

Let's go in the usual way.

(i)$\implies$(ii) Since $\dim V=\dim V^*$ and $L\colon V\to V^*$ is injective (having zero kernel), it is also surjective, hence an isomorphism.

(ii)$\implies$(iii) Let $\{v_1,\dots,v_n\}$ be a basis of $V$; the matrix $C=[c_{ij}]$ is defined by $c_{ij}=\langle v_i,v_j\rangle$. Let's compute the matrix of $L$ with respect to $\{v_1,\dots,v_n\}$ and $\{v_1^*,\dots,v_n^*\}$ (the dual basis). By definition, we need to express $L_{v_i}$ as a linear combination of the vectors in the dual basis: $$ L_{v_i}=\alpha_{1}v_1^*+\dots+\alpha_{n}v_n^* $$ Now, for $j=1,2,\dots,n$, $$ L_{v_i}(v_j)=\alpha_1v_1^*(v_j)+\dots+\alpha_nv_n^*(v_j)=\alpha_j $$ so $\alpha_j=\langle v_i,v_j\rangle$.

We conclude the requested matrix is exactly $C$ (or $C^T$ depending on conventions); the matrix associated to an isomorphism is invertible.

(iii)$\implies$(i) Take $v\in\ker L$ and…

$\endgroup$
14
  • $\begingroup$ How would you go from (iii) to (i)? $\endgroup$ Jul 25, 2017 at 11:55
  • $\begingroup$ After reading more carefully your proof.$ L_{v_i}(v_j)=\alpha_1v_1^*(v_j)+\dots+\alpha_nv_n^*(v_j)=\alpha_j$ only if the basis of $V$ is an orthogonal basis, which is not assumed. $\endgroup$ Jul 25, 2017 at 12:13
  • $\begingroup$ @PedroGomes By definition, $v_i^*(v_j)=\delta_{ij}$. No orthogonality, which makes no sense here: $\{v_1^*,\dots,v_n^*\}$ is, by assumption, the dual basis of $\{v_1,\dots,v_n\}$. $\endgroup$
    – egreg
    Jul 25, 2017 at 12:21
  • $\begingroup$ It was just because $L_{v_i}(v_j)=\alpha_1v_1^*(v_j)+\dots+\alpha_nv_n^*(v_j)=\alpha_j$ in which $\alpha_j$ is a constant so that $a_j\langle v_j,v^{*}_j\rangle=a_j$ and if $j\neq i$ then $a_j\langle v_i,v^{*}_j\rangle=0$. That is the reason why I thought you assumed ortoghonality or another functional. Why write $a_j$ and now write $v_i^*(v_j)=\delta_{ij}$? $\endgroup$ Jul 25, 2017 at 12:46
  • $\begingroup$ @PedroGomes $\delta_{ij}$ is customary notation for $\delta_{ii}=1$ and $\delta_{ij}=0$ for $i\ne j$. You should know about the dual basis in $V^*$ corresponding to a basis of $V$. $\endgroup$
    – egreg
    Jul 25, 2017 at 12:59

You must log in to answer this question.

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