Reading through the proof of following theorem in 'Noncommutative Algebra by Farb and Dennis' :

If $R$ is a semisimple ring, then every simple $R$-module is isomorphic to a simple constituent of R.

The proof goes like this:

Suppose $R \cong \bigoplus_{i \in I} M_{i} $, and let $M$ be a simple R-module. Then, we have the maps :

$ \bigoplus_{i \in I} M_{i} \rightarrow R \rightarrow M $, where the second map is onto. Since, $M$ is simple, only one of the maps $ M_{i} \rightarrow M$ is non-zero, ........

I have problem with the bold part in the proof. The way I justified it was like this: Suppose $ i \neq j$, then $ M_{i} \ncong M_{j}$. So, if $ M_{i}\rightarrow M$ and $M_{j}\rightarrow M$ are non-zero, then $M_{i}, M_{j}$ and $M$ being simple, $M_{i} \cong M, M_{j} \cong M \implies M_{i} \cong M_{j}$. This is a contradiction.

Now, I am not quite satisfied with this argument, because 'if $ i \neq j$, then $ M_{i} \ncong M_{j}$' is true only if we combine together the simple summands of M, but then we will get $R \cong \bigoplus_{i \in I} M_{i}^{n_{i}} $, and $M_{i}^{n_{i}} $ are no longer simple. Please help me get this clear.

  • $\begingroup$ I think it is better to say only one map upto isomorphism of R-modules is non-zero. I get that in any case, the argument after this statement remains same, and the proposition is proved, but the statement in bold seems false to me. $\endgroup$ – P-addict Sep 10 '19 at 6:56

It seems already very clear to argue from first principles like this:

  1. There exists a homomorphism of right $R$ modules $\phi:R\to M$ that is onto.

  2. By the first isomorphism theorem, $R/\ker\phi\cong M$.

  3. Since $R$ is semisimple, $R=\ker\phi\oplus N$ for some submodule $N$ of $R$.

  4. By the second isomorphism theorem, $R/\ker\phi=(\ker\phi+N)/\ker\phi\cong N$.

Therefore $M\cong N$ where $N$ is a minimal right ideal of $R$.

If the argument given in the text is that among the injections of each $M_i$ into $R$ composed with the map $R\to M$, only one can be nonzero, then this is false.

For example, let $\phi: M_2(\mathbb R)\to \mathbb R^2$ simply extract the top row of the matrix, and express $M_2(\mathbb R)=\left\{\begin{bmatrix}a&b\\a&b\end{bmatrix}\right\}\oplus \left\{\begin{bmatrix}a&b\\0&0\end{bmatrix}\right\}$, the projection of both pieces is nonzero.

But if it is arguing that at least one is nonzero then this is fine. In your projection, you probably chose $1\mapsto m$ for some nonzero $m\in M$. Of course, $1=\sum m_i$ with each $m_i\in M_i$. Then $0\neq m=\sum\phi(m_i)$, and it cannot be that all the $\phi(m_i)$ are zero. Pick one that isn't zero, and you've proven that, for that particular $i$, the composed map from $M_i\to M$ is nonzero.

| cite | improve this answer | |
  • 1
    $\begingroup$ Thanks for your perfect answer! In the text, the proof goes like I have given, and yes, it says that only one of such maps can be nonzero. $\endgroup$ – P-addict Sep 10 '19 at 15:18
  • $\begingroup$ @P-addict No problem. Thanks for adding a quality post to the site! If only all users < 1000 rep took the time to ask as carefully as you did. $\endgroup$ – rschwieb Sep 10 '19 at 15:35

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.