I am trying to understand more about the Bidualspace (or double dual space). The whole idea is that $V$ and $V^{**}$ are canonically isomorphic to one another, which means that they are isomorphic without the choice of a basis, which means there exists an isomorphism between them which does not depend on choosing some basis (on either of the spaces). (suggestion by @DonAntonio)

Let $V$ be a finite dimensional Vectorspace with Basis $\beta = \lbrace v_1, \dots , v_n \rbrace $ (the infinite dimensional case is discussed at Canonical Isomorphism Between $\mathbf{V}$ and $(\mathbf{V}^*)^*$). I do understand that although we make a choice of Basis here, it will later on somehow be obsolete (which I don't see why)

Define: \begin{align} \begin{matrix} V & \overset{\Phi}{\longrightarrow}& V^* & \overset{\Phi^*}{\longrightarrow}&V^{**} \\ \sum_{i=1}^n \lambda_i v_i & \longmapsto & \sum_{i=1}^n \lambda_i v_i^* & \longmapsto & \sum_{i=1}^n \lambda_i v_i^{**} \end{matrix} \end{align} Discussion: I know and have already shown that $V$ and $V^*$ are isomorphic (not canonically isomorphic!) to one another, meaning that $\lbrace v_1^*, \dots , v_n^* \rbrace$ defines a Basis for $V^*$. Although I did not show it I am at peace with the statement that the mapping $\Phi^*$ introduces another isomorphism between $V^*$ and $V^{**}$. It is the canonical isomorphism between $V$ and $V^{**}$ that bothers me.

My tutors reasoning: In the following I will use $\checkmark$ to highlight whether or not I understand something.

  • $v_i^{**} \in V^{**}=\hom(V^{*},k)  \checkmark$, sure nothing to add here
  • $v_i^{**} (\sum_{j=1}^n \lambda_j v_j^{*})=\lambda_i \checkmark$, should be completely analogous to $v_i^*(v_j)=\lambda_i$
  • $\Phi^* \circ \Phi (v) =: \iota_v$ $$ \iota: \begin{cases} V & \longrightarrow V^{**} \\ v & \longmapsto \iota_v \end{cases}$$ where $\iota_v( \varphi)= \varphi(v)$ and $\varphi \in V^*$ is a linear functional. My tutor said that $\iota$ is 'suddenly' a canonical isomorphism, independent of the choice of the Basis $\beta$ because $\Phi$ and $\Phi^*$ are isomorphisms. This is the step where I understand nothing about.

What happened next (two more equations): I told my tutor that I don't see why $\iota$ is independent of a basis choice, because at $\Phi^* \circ \Phi= \iota$ we still make a choice for a basis at $\Phi$ namely $\beta$. My tutor said that this seems to be the case at first sight and made two more calculations which I in fact "understand": $$ \Phi^* \circ \Phi \left( \sum_{i=1}^n \lambda_i v_i \right) ( \varphi) = \Phi^* \left( \sum_{i=1}^n \lambda_i v_i^* \right) ( \varphi) = \left( \sum_{i=1}^n \lambda_i v_i^{**} \right)(\varphi) \\ = \sum_{i=1}^n \sum_{j=1}^n \lambda_i \mu_i \underbrace{v_i^{**}(v_j^*)}_{= \delta_{ij}} = \sum_{i=1}^n \lambda_i \mu_i $$ I understand this calculation, it's mainly an application of the above introduced definitions and the Kronecker-Delta. Apparently I am supposed to have an "aha" moment here, which unfortunately didn't occur yet. The next calculation he did was $$ \varphi (v) = \left(\sum_{j=1}^n \mu_j v_j^{*} \right) \left( \sum_{i=1}^n \lambda_i v_i \right)= \sum_{j=1}^n \sum_{i=1}^n \mu_j \lambda_i \underbrace{v_j^*( v_i)}_{\delta_{ij}} = \sum_{i=1}^n \lambda_i \mu_i $$ which evidently gives the same result as above. I would appreciate some wording on how those two equations (which calculations I understand) help me to see why there appears to be a canonical isomorphism between $V$ and $V^{**}$

  • 1
    $\begingroup$ "Isomorphic without the choice of a basis" could probably be a little inaccurate. I'd rather go with "there exists an isomorphism between them which doesn't not depend on choosing some basis (on either of the spaces)." $\endgroup$ – DonAntonio Apr 21 '14 at 16:29
  • $\begingroup$ Thank you @DonAntonio, I did add your suggestion $\endgroup$ – Spaced Apr 21 '14 at 16:47

The first thing you mention is that, if you fix a basis for $V$, you get a non-canonical isomorphism $V\cong V^*$. Similarly, fixing a basis for $V^*$, you get a non-canonical isomorphism $V^*\cong (V^*)^*=V^{**}$. The "magic" is that, when you compose these two isomorphisms, you get an isomorphism $V\cong V^{**}$ which of course depends on your choice of bases, but which is exactly equal to the canonical isomorphism $$\iota: V\to V^{**};\quad \iota(v)=\iota_v,$$ regardless of what choices you made to define $\Phi$ and $\Phi^*$. The two extra equations you give show that $$\Phi^*\circ\Phi = \iota.$$ What may confuse you is that, in order to show these two maps are equal, we choose bases. But this is normal: we need these bases to define $\Phi$ and $\Phi^*$ to start with. But note that the two extra equations hold regardless of what bases were chosen.

  • $\begingroup$ Thanks for your answer @MTurgeon, I did ponder about it for a while now and feel already better with it. Do the two discussed equations also highlight that they hold independent of the choice of the bases because the $v_i$ "vanish" during the calculations? Also is the resulting scalar in $V^{**}$? $\endgroup$ – Spaced Apr 22 '14 at 17:21

The canonical map $\theta$ from a vector space to its double-dual has a particularly clean pointwise forumula:

$$\theta(v)(\omega) = \omega(v)$$

where $v$ is a vector and $\omega$ is a covector (an element of $V^*$).

Since $\omega$ ranges over all covectors, this gives a pointwise definition of $\theta(v)$. And since $v$ ranges over all vectors, this in turn gives a pointwise definition of $\theta$.

This notation is a little strange at first, but it's quite useful once you're used to it. $\theta(v)$ is an element of $V^{**}$, and so it makes sense to evaluate $\theta(v)$ at $\omega$ to get a scalar.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.