Let $V$ be a vector space over a field $k$. Let me describe several different canonical maps which we shall compose in the question.

  • There is a canonical bilinear map $V \times V^* \to \text{End}(V)$ sending $v , \varphi \mapsto [w \mapsto \varphi(w) v]$, so the universal property of tensor product gives a linear map $\Phi: V \otimes V^* \to \text{End}(V)$. If $V$ is finite-dimensional (f.d.), this is an isomorphism. Its dual map $\Phi^* : \text{End}(V)^* \to (V \otimes V^*)^*$ is then also an isomorphism.
  • If $W$ is another $k$-vector space and there is a canonical bilinear map $V^* \times W^* \to (V \otimes W)^*$ sending $\varphi , \psi \mapsto [v \otimes w \mapsto \varphi(v)\psi(w)]$. Again if $V$ and $W$ are f.d., the induced map is also an isomorphism. In the special case when $W = V^*$ ($V$ f.d.), let's name this isomorphism $\Psi: V^* \otimes V^{**} \to (V \otimes V^*)^*$.
  • There is a canonical map $V \to V^{**}$ sending $v \mapsto \text{eval}_v$. Again when $V$ is f.d. this map is an isomorphism, hence we obtain an isomorphism $\Theta: V^* \otimes V \to V^* \otimes V^{**}$.
  • Finally, to be completely pedantic, there is a canonical isomorphism $\Gamma: V \otimes V^* \to V^* \otimes V$ given by swapping the order of the simple tensors.
  • Composing maps (f.d. case), we have a canonical isomorphism $F : \text{End}(V) \to \text{End}(V)^*$:

$$ \text{End}(V) \overset{\Phi^{-1}}{\longrightarrow} V \otimes V^* \overset{\Gamma} {\longrightarrow} V^* \otimes V \overset{\Theta}{\longrightarrow} V^* \otimes V^{**} \overset{\Psi}{\longrightarrow} (V \otimes V^*)^* \overset{(\Phi^*)^{-1}}{\longrightarrow} \text{End}(V)^*$$

  • In the f.d. case, there is a special element of $\text{End}(V)^*$, namely the trace. As an element of $(V \otimes V^*)^*$ it is given by tensor contraction: $\Phi^*(\text{tr})(v \otimes \varphi) = \varphi(v)$.

Actual Question:

This seems like it should be totally obvious, but I'm kinda stumped! What the heck element of $\text{End}(V)$ does the trace correspond to under the isomorphism $F$? i.e. what is $F^{-1}(\text{tr})$? And actually, while we're at it (or perhaps along the way), what is $\Psi^{-1}(\Phi^*(\text{tr}))$? It feels strange to have a distinguished element of $V^* \otimes V^{**}$. Well I suppose the image of $1_V \in \text{End}(V)$ is also distinguished... Hm.

  • 1
    $\begingroup$ In fact, I believe that you should find that $F:\operatorname{End}(V) \to \operatorname{End}(V)^*$ is the map $$ F(\alpha)(\beta) = \operatorname{tr}(\alpha \beta) $$ $\endgroup$ Aug 19 '20 at 8:28

As you say, the only distinguished element of $\text{End}(V)$ is $\text{id}_V$ and that's what you end up getting. I haven't checked but you should be able to verify this by writing everything out in terms of a basis $e_i$ of $V$ and the corresponding dual basis $e_i^{\ast}$ of $V^{\ast}$. You get

$$\text{id}_V = \sum_{i=1}^n e_i \otimes e_i^{\ast}.$$

  • $\begingroup$ Cool, thanks so much!! I guess it had to be that, by Occam's razor or something :). So is there no "basis free" way of writing $1_V \in V \otimes V^*$? Or equivalently $\text{tr} \in V^* \otimes V^{**}$? $\endgroup$ Aug 18 '20 at 23:54
  • 1
    $\begingroup$ You can almost formalize the intuition that it "has to be" the identity: whatever you get it has to be invariant under conjugation by automorphisms of $V$ (because every map you've written down is functorial wrt automorphisms) and the only elements of $\text{End}(V)$ with that property are the scalar multiples of the identity. $\endgroup$ Aug 19 '20 at 0:14
  • 1
    $\begingroup$ As for writing $\text{id}_V \in V \otimes V^{\ast}$ in a basis-free way you can transport $\text{id}_V \in \text{End}(V)$ along the natural automorphism. There's a whole larger story to tell here about how duals work: ncatlab.org/nlab/show/dualizable+object $\endgroup$ Aug 19 '20 at 0:15

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.