5
$\begingroup$

Let $k$ be a field and $\text{Mat}_n(k)$ be $n \times n$ matrices over $k$. Let's consider $\text{Mat}_n(k)$ as an associative algebra and denote $gl_n(k)$ be the same $k$-linear space as $\text{Mat}_n(k)$ but considered as a Lie algebra. We can form the universal enveloping algebra $U(gl_n(k))$ of $gl_n(k)$, which satisfies the universal property, $\forall$ unital associative $k$-algebra $A$, we have $$ Hom_{k-alg}(U(gl_n(k)),A)\cong Hom_{Lie}(gl_n(k),F(A)), $$ where $F$ is the forgetful functor from $k$-alg to Lie.

Now we know that the associative algebra $\text{Mat}_n(k)$ is not the universal enveloping algebra $U(gl_n(k))$. However, by the universal property, we have $$ Hom_{k-alg}(U(gl_n(k)),\text{Mat}_n(k))\cong Hom_{Lie}(gl_n(k),gl_n(k)), $$ which makes $\text{Mat}_n(k)$ "special" in some sense.

$\textbf{My question}$ is: Could we make the above observation more precise? Does $\text{Mat}_n(k)$ have some universal properties similar to $U(gl_n(k))$?

$\endgroup$
  • 1
    $\begingroup$ It seems odd to think about matrices as an object having a universal property (in the same way that it seems odd to think about $S_n$ as an object having a universal property). I would rather say that it's the endomorphism monoid of an object in a category I care about (namely vector spaces). $\endgroup$ – Qiaochu Yuan Sep 17 '13 at 21:26
  • $\begingroup$ @QiaochuYuan, yes, you are right, but/and this could be revised-away from the question, just as the coordinates could be removed from $GL_n(\mathbb R)$ and such... $\endgroup$ – paul garrett Sep 17 '13 at 22:31
8
$\begingroup$

Of course, what's a universal property is open to interpretation, but $\mathrm{Mat}_n(k)$ corepresents something very nice: $\mathrm{Hom}_{k-alg}(A,\mathrm{Mat}_n(k))$ is the space of representations of $A$ on $k^n$ (not up to any kind of equivalence, of course). This can also be interpreted as the $k$ points of an affine algebraic $k$-variety $\mathrm{Rep}_n(A)$, as long as $A$ is finitely generated: if $a_1,\dots, a_m$ are generators of $A$, and $r_1(\mathbf{a}), \dots, r_\ell(\mathbf{a})$ are the non-commutative polynomials giving relations, then $\mathrm{Rep}_n(A)$ is defined by taking $n\times n$ matrices $M_1,\dots, M_m$ with each matrix coefficient a variable, and letting the relations be $r_k(\mathbf{M})=0$.

In fancier language, we can define its points over any $k$-algebra $R$ by $$\mathrm{Rep}_n(A,R)=\mathrm{Hom}_{R-alg}(A\otimes_k R,\mathrm{Mat}_n(R)).$$

You can interpret this as saying that the functor from commutative to non-commutive $k$-algebras given by $\mathrm{Mat}_n$ is the right adjoint to the functor sending $A$ to $k[\mathrm{Rep}_n(A)]$, much like the universal enveloping algebra is left adjoint to the forgetful functor from algebras to Lie algebras that just remembers the bracket.

$\endgroup$
  • 2
    $\begingroup$ I think this is "the right answer". :) $\endgroup$ – paul garrett Sep 17 '13 at 22:29
  • $\begingroup$ Based on the OP's earlier interest in maps out of the universal enveloping algebra I would have guessed that the OP was interested in maps out of the matrix algebra, not maps in. $\endgroup$ – Qiaochu Yuan Sep 18 '13 at 1:19
  • 1
    $\begingroup$ In the words of Mick Jagger: you can't always get what you want. $\endgroup$ – Ben Webster Sep 18 '13 at 13:13

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.