7
$\begingroup$

Let $A$ be an associative algebra over $\mathbb{C}$ with irreducible finite-dimensional representations on $V$ and $W$. Then is the tensor product of representations on $V \otimes W$ semi-simple?

The below question is concerning representations of a group and in this case the answer is positive due to a result of Chevalley. I tried looking in the literature but I could not find any equivalent results for representations of an algebra.

Tensor product of simple representations

$\endgroup$
8
  • 10
    $\begingroup$ What is the $A$-module strucutre on $V \otimes W$? If it is not a Hopf algebra this might not be so canonical. Or do you view $V \otimes W$ as a module over the enveloping algebra $A \otimes A^{op}$ of $A$? $\endgroup$
    – Mare
    Commented Jun 17, 2023 at 7:01
  • 2
    $\begingroup$ Indeed in characteristic 0, $V\otimes W$ is semi-simple as a $A\otimes A^{op}$-module. $\endgroup$
    – abx
    Commented Jun 17, 2023 at 7:44
  • 1
    $\begingroup$ @Mare Sorry, I have realised that the tensor product is not in general defined for algebras. So to be more specific, the case I am interested in is when $A$ is the free algebra generated by $n$ elements (even $n = 2$ case is fine). Though, perhaps its too much to ask for such a specific case to have been treated in the literature. $\endgroup$
    – Nanoputian
    Commented Jun 17, 2023 at 8:02
  • 2
    $\begingroup$ What is your coproduct on the free algebra? $\endgroup$ Commented Jun 17, 2023 at 9:30
  • 3
    $\begingroup$ About Dave Benson's question, presumably you mean that the coproduct of a generator $x$ is $x\otimes x$, in other words a generator operates on $V\otimes W$ by $x(v\otimes w)=xv\otimes xw$. Assuming this, there is a counterexample in which $A$ is free on two generators and $V=W$ is $2$-dimensional. $\endgroup$ Commented Jun 17, 2023 at 11:17

2 Answers 2

10
$\begingroup$

By a famous theorem of R. Steinberg, STEINBERG, R. Complete sets of representations of algebras. PROC. Am. Math. Soc. 13 (1962), 746-747, if $V$ is a faithful representation for a monoid $M$, then $T(V)=\bigoplus_{n\geq 0}V^{\otimes n}$ is a faithful representation for the monoid algebra $KM$ for any field $K$. If $M$ is finite, then $KM$ is finite dimensional and so it follows easily that $\bigoplus_{n=0}^rV^{\otimes n}$ is a faithful $KM$-module for some $r$.

So you can get a counterexample by taking a finite monoid $M$ with a faithful complex semisimple representation which does not have a semisimple complex algebra. There are many such examples. Here is an easy one.

Consider $M$ the monoid of $3\times 3$ complex matrices consisting of zero, the identity, all matrix units $E_{ij}$ and the nilpotent matrix $\begin{bmatrix}0 & 1 &0\\ 0 &0 &1 \\ 0 & 0 &0 \end{bmatrix}$. Then $\mathbb CM$ has a faithful irreducible representation $V$, namely the one I used to define it. Irreducibility follows since all the matrix units are in the image of the representation, and so the map $\mathbb CM\to M_3(\mathbb C)$ is sujective.

But $\mathbb CM$ does not have a semisimple algebra. If you factor out the ideal of $\mathbb CM$ spanned by the matrices of rank at most 1, you get an algebra with a nilpotent ideal (spanned by the coset of the nilpotent matrix). So $\mathbb CM$ has a nonsemisimple quotient and hence is not semisimple. It follows that not all tensor powers of $V$ are semisimple by R. Steinberg's theorem.

Added. I think already $V\otimes V$ is not semisimple. It contains as a submodule the exterior power $\Lambda^2(V)$. This is annihilated by all the rank $\leq 1$ matrices but not by the nilpotent matrix. So the image of $\mathbb CM$ under this exterior power representation is not semisimple (it has a nilpotent ideal spanned by the image of the nilpotent matrix) and so this subrepresentation is not semisimple and hence $V\otimes V$ is not semisimple.

Added. Here is a two-dmensional example. Take the monoid consting of the 2x2 identity matrix and the matrices $\pm \begin{bmatrix} 1 & 1\cr 0 & 0\end{bmatrix}, \pm \begin{bmatrix} 1 & -1\cr 0 & 0\end{bmatrix}, \pm \begin{bmatrix} 0 & 0\cr 1 & 1\end{bmatrix}, \pm \begin{bmatrix} 0 & 0\cr 1 & -1\end{bmatrix}$. It generated as a monoid by $\begin{bmatrix} -1 & -1\cr 0 & 0\end{bmatrix}, \begin{bmatrix} 0 & 0\cr 1 & -1\end{bmatrix}$. It is easy too see there is no invariant subspace since these rank $1$ matrices span all $2\times 2$ matrices. So this is an irreducible represention $V$.

But $V\otimes V$ has a nonsemisimple submodule. The vectors $e_1\otimes e_1$ and $e_2\otimes e_2$ span an invariant subspace with $e_1\otimes e_1-e_2\otimes e_2$ spanning an invariant subspace $W$ with no complement. A complement would have to be fixed by all of the matrices (since $W/\mathbb C(e_1\otimes e_1-e_2\otimes e_2)$ is the trivial module) but no such vector exists.

$\endgroup$
1
  • 2
    $\begingroup$ There is a smaller example of a 2-generated 9 element monoid with an irreducible 2-dim rep whose tensor square is not semisimple if anybody wants to see it $\endgroup$ Commented Jun 17, 2023 at 17:06
7
$\begingroup$

Here is an example in which $A$ is freely generated by two elements $x,y$. Let $W$ be $\mathbb C^2$ with representation given by $$ x(a_1,a_2)=(0,a_1) $$$$ y(a_1,a_2)=(a_2,0). $$ This is irreducible. Let $V$ be again two-dimensional, with $x$ and $y$ acting by invertible maps $S:V\to V$, $T:V\to V$ respectively. So $V\otimes W$ can be described as $V\times V$ with $$ x(v_1,v_2)=(0,Sv_1) $$$$ y(v_1,v_2)=(Tv_2,0). $$ A proper nontrivial subrepresentation $U\subset V\otimes W$ corresponds precisely to a one-dimensional subspace $L\subset V$ such that $S(L)=T(L)$; take $U$ to consist of pairs $(v_1,v_2)$ such that $v_1\in L$ and $v_2\in S(L)$.

If $S$ and $T$ are chosen in such a way that $T^{-1}S$ has exactly one $1$-dimensional invariant subspace $L$, then $V\otimes W$ has exactly one proper nontrivial subrepresentation, so it is not semisimple. On the other hand, if we also choose $S$ in such a way that $L$ is not invariant under $S$, then $V$ will be irreducible.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .