1
$\begingroup$

Is there a non trivial sequence $(T_{n})$ of linear operators $T_{n}$ on $M_{n}(\mathbb{C})$ such that $$T_{nm}(X\otimes Y)=T_{n}(X) \otimes T_{m}(Y) $$ where $X$ and $Y$ are in $M_{n}(\mathbb{C})$ and $M_{m}(\mathbb{C})$, respectively?

By trivial sequence we mean $T_{n}=Id$ for all $n$.

$\endgroup$

3 Answers 3

4
$\begingroup$

$T_n(X) = n \times n$ matrix with all entries $trace(X)$.
This doesn't satisfy $T_n(X_1 X_2) = T_n(X_1) T_n(X_2)$.

Edit: More examples:

  1. $T_n(X)$ has $trace(X)$ in upper left corner, 0 elsewhere.

  2. $T_n(X)$ has $trace(X)$ in lower right corner, 0 elsewhere.

  3. $T_n(X)$ has sum of anti-diagonal $(\Sigma x_{i,n-i})$ in upper right corner, 0 elsewhere.

  4. $T_n(X)$ has sum of anti-diagonal in lower left corner, 0 elsewhere.
  5. $T_n(X)$ has all entries = sum of all entries of $X$.

Edit: Another type:

  1. $T_n(X)=0$ ($n$ even); $T_n(X)$ has sum of middle row in every middle row entry, 0 elsewhere ($n$ odd).

  2. Same sum as in (6.) except $T_n(X)=0$ has only one non-zero entry at left end resp. middle resp. right end of middle row.

$\endgroup$
2
  • $\begingroup$ [This doesn't satisfy $T_n(X_1 X_2) = T_n(X_1) T_n(X_2)$.]---> Yes but this is not required and your example fits perfectly the requirements of the MO (+1). We can even mix your $T_n$ with thee ``logarithmic trick'' of Simon. $\endgroup$ Apr 26, 2016 at 11:46
  • $\begingroup$ [We can even mix...] ---> Yes: if $(S_n), (T_n)$ work then so does $(S_{n}T_{n})$. This example and the "logarithmic trick" examples all commute. $\endgroup$ Apr 26, 2016 at 13:04
2
$\begingroup$

What is a linear operator---do you require $T_n(X_1 X_2) = T_n(X_1) T_n(X_2)$? If not, $T_n(X) = c^n X$ for a fixed scalar $c$ works. Even if yes, I think that $T_n(X) = X$ if $n$ odd, $0$ if $n$ even should work ($T_n = Id$ for odd $n$, but not all $n$).

$\endgroup$
2
  • $\begingroup$ Thanks for your answer. Your last example works. but the first one does not work, for n=1 unless c=1. $\endgroup$ Apr 26, 2016 at 8:24
  • $\begingroup$ A revised version : is every sequence necessarily a sequence of 0 ,Id? $\endgroup$ Apr 26, 2016 at 8:31
2
$\begingroup$

Pick for each prime number $p$ a complex number $\alpha_p$.

For $n$ an integer, define $\alpha_n = \prod \alpha_p ^{\nu_p(n)}$ where $\nu_p(n)$ is the order of $p$ in the prime decomposition of $n$.

One has $\alpha_{nm} = \alpha_n \alpha_m$

Then you can take $T_n(M) = \alpha_n M$

Zach Teitler's example correspond to $\alpha_2=0$ and $\alpha_p =1$ for all odd prime $p$.

$\endgroup$
2
  • $\begingroup$ Thanks for your very interesting answer. what about if a trivial sequence would be meant as "a sequence of scalar operator"? is there a non trivial example? $\endgroup$ Apr 26, 2016 at 9:02
  • $\begingroup$ Maybe you can do the same construction with $\alpha_p$ an endomorphism of $M_p(\mathbb{C})$ for each prime $p$ ? But you need to be more carefull on how you identify $M_n \otimes M_m$ with $M_{nm}$ to see if this works... I'm not convinced. $\endgroup$ Apr 26, 2016 at 10:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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