Let $V$ be a vector space over a field $\Bbbk$. Let $V^*=Hom_\Bbbk(V,\Bbbk)$ be the vector space of linear functionals on $V$. Consider the map $V^*\times V\to End_\Bbbk(V)$ given by $(f,v_0)\to\phi$ where $\phi(v)=f(v)\cdot v_0$. This map is obviously bilinear and thus induces a linear map $V^*\otimes_\Bbbk V\to End_\Bbbk V$.

If $V$ is finite-dimensional, then a choice of basis $\{e_1,\dots,e_n\}$ and a dual basis $\{e_1^*,\dots,e_n^*\}$ establishes that the map is injective since evaluations at $e_i$ for the image of each basis element $e_i^*\otimes e_j$ of $V^*\otimes V$ shows that the corresponding endomorphisms are linearly independent. Identifying $End_\Bbbk(V)$ with the $n^2$-dimensional space of $n\times n$ matrices and the fact that $\dim V^*\otimes V$ is $n^2$ shows that the map is in fact an isomorphism.

Now onto my question. If $V$ is an infinite-dimensional vector space over $\Bbbk$, then:

  1. is the above (obvious?) map $V^*\otimes_\Bbbk V\to End_\Bbbk(V)$ still injective? If not, then under what conditions is it?
  2. given a $\phi\in End(V)$, is there a criterion to determine whether $\phi$ comes from an element of $V^*\otimes V$?
  • $\begingroup$ When you say the space of linear functionals, do you mean continuous linear functionals? There is a distinction in the infinite case. $\endgroup$ – Asaf Karagila Feb 27 '11 at 7:51
  • $\begingroup$ @Asaf: I assume he means just what is asked; there is no mention of topology or continuity. (But if there were, then my answer to question 2. would have to be modified to show that each $c_j$ is continuous.) $\endgroup$ – Jonas Meyer Feb 27 '11 at 8:07
  • $\begingroup$ Jonas: of course. I'm just being petty. :-) $\endgroup$ – Asaf Karagila Feb 27 '11 at 10:07
  1. Yes. If $\sum_{j=1}^n f_j\otimes e_j$ is an element of $V^*\otimes_k V$, you can assume WLOG that $\{e_1,\ldots,e_n\}$ is linearly independent. If $\sum_{j=1}^n f_j(v)e_j=0$ for all $v\in V$, then for each $j$, $f_j(v)=0$ for all $v\in V$, so that $f_1=f_2=\cdots=f_n=0$.

  2. They are the linear transformations with finite dimensional range. If the range of $\phi\in \mathrm{End}(V)$ has basis $e_1,\ldots,e_n$, then you can write $\phi(v)=\sum_{j=1}^n c_j(v)e_j$, and each coefficient map $c_j:V\to k$ is linear. Thus $\phi$ comes from $\sum_{j=1}^n c_j\otimes e_j$.

  • $\begingroup$ Thanks: this shows that the explicit finite-dimensionality (i.e. choosing basis for $V$) in my proof was really a red herring. $\endgroup$ – Vladimir Sotirov Feb 27 '11 at 15:47

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.