4

Let $R,T$ be two rings and a homomorphism $f:M_n(R)\longrightarrow T$. Show that there exists a ring $S$ and a ring isomorphism $g:T\longrightarrow M_n(S)$.

This is an exercise from a ring theory course. I have no idea how to start this and find such a ring $S$. If I could find such a ring, I think I could construct such an isomorphism. Any hints on how to begin this?

  • Is there any other context? Any requirement about a relationship between $S$ and $R$ or $T$? Otherwise this seems too open-ended, as the construction of $g$ has nothing to do with $f$. One possibility: just let $S = T$ and embed $T$ on the diagonal: $g(t) = tI_n$. But there are other maps too. – Sammy Black Nov 10 '23 at 01:13
  • @SammyBlack No, the exercise was given as I have written it. Maybe that's why I could not find something, as I was focused on $f$. Thanks! – TOP STIN ELLADA Nov 10 '23 at 01:22
  • @SammyBlack Also, I don't think that $g(t)=tI_n$ would be an isomorphism as it cannot be surjective. If we take a matrix $A$ with a non-zero non-diagonal element, then there is no $t\in T$ such that $g(t)=A$. Is this wrong? Anyhow, I appreciate the idea and I am gonna look further upon it. – TOP STIN ELLADA Nov 10 '23 at 01:29
  • Ohh. I misread, yeah, the second map is meant to be an isomorphism, essentially saying that any map out of a matrix ring lands in another matrix ring. – Sammy Black Nov 10 '23 at 01:47
  • 1
    The kernel of $f$ is an ideal of $R$, i.e. $M_n(I)$ for some ideal $I$ in $R$. Then $S = R/I$. – Amateur_Algebraist Nov 10 '23 at 11:32
  • 2
    @Amateur_Algebraist This just reduces to the case $\ker(f)=0$, which is not clear either. – Martin Brandenburg Nov 10 '23 at 16:42
  • @TOPSTINELLADA What is your convention for rings? Commutative and unital? Are your homomorphisms unital? Otherwise we can take $R=0$ and would get that every ring is isomorphic to a matrix ring (while I don't believe to be true, but I'd have no idea how to show that). – Severin Schraven Nov 10 '23 at 17:24
  • Rings are unital (and hence also homomorphisms), but not commutative. https://math.mit.edu/~poonen/papers/ring.pdf – Martin Brandenburg Nov 10 '23 at 17:25
  • @SeverinSchraven The rings are not necessarily commutative but they are unital. As for the homomorphisms they are unital. – TOP STIN ELLADA Nov 10 '23 at 17:26
  • 1
    I have solved the problem if $f$ is surjective, using the Morita Equivalence idea from above. I can't seem to find a way, though, if $f$ is not necessarily surjective. Any idea for this? – TOP STIN ELLADA Nov 10 '23 at 17:27
  • 2
    The case of surjective $f$ is covered by https://math.stackexchange.com/questions/22629/why-is-the-ring-of-matrices-over-a-field-simple/45061#45061 already. We may in fact assume that $M_n(R) \subseteq T$ is a subring. The problem is really weird ... It feels it must be wrong, but everything I have a "counterexample" it is actually no counterexample. – Martin Brandenburg Nov 10 '23 at 17:31
  • 2
    In the general case where $f$ is not necessarily surjective, maybe set $S := { x \in T \mid f(E_{11}) x = x = x f(E_{11}) }$, try to show $S$ is a subring, and then define a function $g : T \to M_n(S)$ by $x \mapsto (f(E_{1i}) x f(E_{j1}))$ and try to show that's an isomorphism of rings? – Daniel Schepler Nov 10 '23 at 17:55
  • @DanielSchepler This does not work since for example $f(E_{11}) \in S$. – Martin Brandenburg Nov 10 '23 at 17:56
  • 1
    OK, I guess $S$ is not a sub-unital-ring, but maybe it's close enough using $f(E_{11})$ as an idempotent to become the unit of $S$. – Daniel Schepler Nov 10 '23 at 17:58
  • No, I mean try your attempt with $T = M_n(R)$. It does not reconstruct $R$. – Martin Brandenburg Nov 10 '23 at 17:58
  • 1
    In the case $T = M_n(R)$ with $f$ the identity, I would think my construction gives $S$ as the set of matrices with all zero entries except for the $(1,1)$ entry, which is isomorphic to $R$. – Daniel Schepler Nov 10 '23 at 18:00

3 Answers3

8

The claim is that if a ring admits a homomorphism from an $n \times n$-matrix ring, then it is an $n \times n$-matrix ring as well. This follows from the main result in the paper

  • G. Agnarsson, S. A. Amitsur and J. C. Robson, Recognition of matrix rings II, link / pdf

It proves that a ring is an $n \times n$-matrix ring if and only if it contains elements $f,a,b$ such that $$f^n = 0 \quad \text{and} \quad a f^{n-1} + f b = 1.$$ This property is clearly preserved by homomorphisms.

I am not aware of a more direct proof of the claim.

Other characterization theorems have appeared in the literature as well:

  • J. C. Robson, Recognition of matrix rings, link
  • P. R. Fuchs, A characterization result for matrix rings, link
  • T. Y. Lam, A. Leroy, Recognition and Computations of Matrix Rings, link
3

The proof that Daniel Schepler sketched in comments works, but here's a less fiddly proof that uses a little more technology.

First, $f$ allows us to view $T$ as a left $M_n(R)$-module.

Let $X$ be the right $M_n(R)$-module of row vectors. Then as a right $M_n(R)$-module, $M_n(R)\cong X^n$, the direct sum of $n$ copies of $X$.

So as a right $T$-module $$T\cong M_n(R)\otimes_{M_n(R)}T\cong(X\otimes_{M_n(R)}T)^n.$$

So as a ring $$T\cong\operatorname{End}_T(T)\cong\operatorname{End}_T\left((X\otimes_{M_n(R)}T)^n\right)\cong M_n(\operatorname{End}_T\left(X\otimes_{M_n(R)}T)\right).$$

  • Very nice proof, and much more direct than mine. – Martin Brandenburg Nov 11 '23 at 11:26
  • I assume that we can generalize this argument with any ring that is Morita equivalent to R? – Martin Brandenburg Nov 11 '23 at 11:31
  • @MartinBrandenburg I’m not sure what conclusion you want if we generalize to a ring Morita equivalent to $R$? – Jeremy Rickard Nov 11 '23 at 14:52
  • I'm not sure either - this was part of my question. I just noticed that your proof is essentially making use of the equivalence $\mathbf{Mod}R \cong \mathbf{Mod}{M_n(R)}$ (and its explicit form). But maybe I am wrong, I didn't think through this yet. – Martin Brandenburg Nov 11 '23 at 15:16
  • Ah this seems really straightforward but I can't understand it yet with my current knowledge. Thanks for the answer though. – TOP STIN ELLADA Nov 11 '23 at 20:42
  • @TOPSTINELLADA May I ask which parts of the answer are not clear / go beyond the current knowledge you have? Maybe it's not as hard as you think. :) I can also post a more elaborate form of that proof if you wish, but for that it is useful to know your prerequisites (for example, if you have seen tensor products before). – Martin Brandenburg Nov 12 '23 at 23:28
  • 1
    @MartinBrandenburg Hello, and thanks very much for your time and effort!! I am currently studying an introduction to Ring Theory and haven't yet reached the modules (or the tensor products). As of now, I have seen homomorphisms of rings, algebras and an elementary proof of the equivalence of $M_n(R)$ and $R$. The goal of the course is to show an elementary proof of the Wedderburn-Artin theorem. – TOP STIN ELLADA Nov 13 '23 at 00:19
3

This problem is essentially Corollary 3.4 in P. M. Cohn's Algebra, Volume 2, second edition (Section 4.3, pages 136-137). The idea is that the homomorphism from $M_n(R)$ to $T$ gives us a complete set of matrix units in $T$, and any ring with a complete set of matrix units is a full matrix ring.

Let $E_{ij}$ denote the standard matrix unit in $M_n(R)$, with a $1$ in the $i,j$ position and zeroes elsewhere. Set $e_{ij}=f(E_{ij})$. Then the $e_{ij}$, $1\le i,j\le n$, satisfy the rules $e_{ij}e_{kl}=\delta_{jk}e_{il}$ and $\sum_i e_{ii}=1$. Such a collection of elements is called a complete set of matrix units. Proposition 3.3 from Cohn's text says that if $T$ has a complete set of matrix units, then $T$ is isomorphic to $M_n(S)$, where $S$ is the centralizer of the matrix units. That is, $S=\{\,t\in T\mid e_{ij}t=te_{ij}\text{ for all }i,j\,\}$.

Here is an outline of the proof. Convention: $A,B$ will always be $n\times n$ matrices with $i,j$ entries $a_{ij}$, $b_{ij}$ respectively. We define $h:M_n(S)\to T$ by $h(A)=\sum_{i,j} a_{ij}e_{ij}$ and $g:T\to M_n(S)$ by $g(t)=A$, where $a_{ij}=\sum_k e_{ki}te_{jk}$. We should immediately verify that $a_{ij}\in S$. First note $e_{pq}a_{ij}=\sum_k e_{pq}e_{ki}te_{jk}=e_{pi}te_{jq}$, since the only nonzero summand occurs when $k=q$. A similar calculation shows $a_{ij}e_{pq}=e_{pi}te_{jq}$, so each $a_{ij}\in S$. Since the elements of $S$ commute with all the $e_{ij}$, the map $h$ is clearly a ring homomorphism. All that remains is to show $g$ and $h$ are inverses.

We show directly that $g$ and $h$ undo each other. First suppose $t\in T$, so $h(g(t))=\sum_{i,j,k} e_{ki}te_{jk}e_{ij}=\sum_{j,k} e_{kk}te_{jj}=(\sum_k e_{kk})t(\sum_j e_{jj})=1t1=t$. Next suppose $A\in M_n(S)$ and write $h(A)=\sum_{p,q}a_{pq}e_{pq}$, where the $a_{pq}$ commute with all of the matrix units. Then $g(h(A))=B$, where $b_{ij}=\sum_{k,p,q} e_{ki}a_{pq}e_{pq}e_{jk}=\sum_k a_{ij}e_{ki}e_{ij}e_{jk}$, since the summands are $0$ unless $p=i$ and $q=j$. This last sum equals $a_{ij}\sum_k e_{kk}=a_{ij}$. Thus $g(h(A))=A$.

Allen Bell
  • 1,242