NOTE: My actual question is in the last paragraph. Feel free to skip down there first to see what exactly I'm asking.

Here's the question I've written a proof for:

Suppose $V$ is finite-dimensional and $\varphi_1, \dots, \varphi_n$ is a basis of $V'$. Show that there exists a basis of $V$ whose dual basis is $\varphi_1, \dots, \varphi_n$.

I have written a proof that I think is correct. Here it is:

By $\widehat{\operatorname{null}\varphi_j}$ denote $\operatorname{null}\varphi_1 \cap \cdots \cap \operatorname{null}\varphi_{j-1}\cap\operatorname{null}\varphi_{j+1}\cap \cdots \cap \operatorname{null}\varphi_n$. From [the previous exercise], we see that $\dim\widehat{\operatorname{null}\varphi_j} = 1$ for all $j=1,\dots, n$. I.e. for any $j=1,\dots, n$, there exists a nonzero $v_j\in\widehat{\operatorname{null}\varphi_j}$. As a further consequence of [the previous exercise], we see that $v_j\not\in \operatorname{null}\varphi_j$ because $\dim(\operatorname{null}\varphi_1\cap \cdots \cap \operatorname{null}\varphi_n) = \dim V - \dim V' = 0$ (i.e. $\operatorname{null}\varphi_1\cap \cdots \cap \operatorname{null}\varphi_n = \{0\}$). It follows that $\left\{\frac{v_1}{\varphi_1(v_1)}, \dots, \frac{v_n}{\varphi_n(v_n)}\right\}$ has the property $\varphi_j\left(\frac{v_i}{\varphi_i(v_i)}\right) = \delta_{ij}$ as desired.

To show that this is a basis it suffices to show that it is linearly independent. Let $a_1, \dots, a_n\in \Bbb F$ such that $$a_1\frac{v_1}{\varphi_1(v_1)} + \cdots + a_n\frac{v_n}{\varphi_n(v_n)} = 0.$$ Applying $\varphi_j$ to both sides yields $a_j=0$ for any $j=1,\dots, n$, completing the proof.$\square$

Now here's my question. This exercise is clearly stated as an existence proof. With my particular proof, have I also shown uniqueness? Afterall, while the choice of $v_j$ was arbitrary, it was in a $1$-dimensional space $\widehat{\operatorname{null}\varphi_j}$, which is just the multiples of this vector. Then I (kinda) normalized it by dividing by $\varphi_j(v_j)$. Meaning if instead I'd chosen $kv_j$, I'd just have $\frac{kv_j}{\varphi_j(kv_j)} = \frac{kv_j}{k\varphi_j(v_j)} = \frac{v_j}{\varphi_j(v_j)}$ in my basis. So doesn't that mean that each $\frac{v_j}{\varphi_j(v_j)}$ is unique?


Yes, the basis $\{w_1,\dots,w_n\}$ of which $\{\varphi_1,\dots,\varphi_n\}$ is the dual basis is uniquely determined. Indeed, if $v$ is a vector such that $\varphi_i(v)=1$ and $\varphi_j(v)=0$ for $j\ne i$, then $$ \varphi_j(v-w_i)=0 $$ for every $j=1,2,\dots,n$. In particular, every linear form vanishes over $v-w_i$, so $v-w_i=0$. A property of the dual space is that, for every nonzero vector, there is a linear form that doesn't vanish on it.

There is a possibly better proof. Consider the dual basis in $V''$ and pull it back to $V$ via the canonical isomorphism $V\to V''$.

I leave you to complete the details.

| cite | improve this answer | |
  • $\begingroup$ That's pretty clever (the top part, I don't get to the canonical isomorphism between $V$ and $V''$ for two more exercises). Accepted. $\endgroup$ – Bobbie D Mar 4 '17 at 21:35
  • $\begingroup$ @BobbieD OK! But, really, doing it with the double dual is much easier! Note that the above uniqueness proof doesn't need existence. $\endgroup$ – egreg Mar 4 '17 at 21:39

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.