2
$\begingroup$

Let $R$ be a ring with unity and $M$ any right $R$-module. A submodule $X$ of $M$ is called square-root in $M$ if $X \oplus X$ embeds in $M$ (i.e., there exists a monomorphism $X \oplus X \to M$). Call $M$ square-full if every non-zero submodule of $M$ contains a non-zero square-root in $M$. We say that the ring $R$ is left (resp., right) square-full if the module ${}_RR$ (resp., $R_R$) is square-full.

Now, let $R$ be a semisimple ring. Is it true that $R$ is left square-full if and only if $R$ is right square-full?. Unfortunately, I don't have any contribution to this question. So, I appreciate any help.

Thanks in advance.

$\endgroup$
2
  • $\begingroup$ @rschwieb I had thought so as well, but what is written is that every non-zero submodule $N$ of $M$ contains a (some) non-zero square-root of $M$, say $L\subseteq N$. Hence $L\oplus L$ embeds into $M$ (and not necessarily $N\oplus N$). Thus if you take $M=E\oplus E$, with $E$ simple, then $M$ is square-full, as any non-zero submodule $N$ is either $M$ or simple. But if it is simple, then $N\simeq E$ and $E\oplus E$ embeds into $M$. $\endgroup$ Commented Oct 2 at 16:29
  • $\begingroup$ @ChristianLomp I see. So the doubled copy of the submodule is not necessarily nested with the submodules... not quite what I was thinking when I first read it! I understood your version, of course, but thought something stronger could be said. $\endgroup$
    – rschwieb
    Commented Oct 2 at 16:49

2 Answers 2

6
$\begingroup$

I am rewriting my answer since I haven't understood correctly your question. If I understand you correctly, then a submodule $X$ of $M$ is square-free, if $M$ contains a submodule that is a direct sum of two isomorphic copies of $X$. While you call $M$ square-full, if any non-zero submodule $Y$ of $M$ contains some non-zero submodule $X\subseteq Y$ such that $M$ contains a submodule that is a direct sum of two isomorphic copies of $X$.

If I understand your notions correctly, then no simple module $M$ can be square-full, as $M$ is the only non-zero submodule and $M\oplus M$ does not embed into $M$.

If $E$ is a simple module and $M\simeq E^{(\Lambda)}$ is a direct sum of $|\Lambda|$-copies of $E$, where $\Lambda$ is some set, possibly infinite, then $M$ is square-full if and only if $|\Lambda|>1$, because any non-zero submodule $Y$ of $M$ will contain a simple submodule $X$ isomorphic to $E$. If $|\Lambda|>1$, then $X\oplus X \simeq E\oplus E$ embeds into $M$. If $|\Lambda|=1$, then $M$ is simple and cannot be square-full.

Thus if $R$ is semisimple Artinian, then $R \simeq M_{n_1}(D_1) \times \cdots \times M_{n_k}(D_k)$, for division rings $D_i$ and numbers $n_i\geq 1$. For each block $i$, there exists a unique simple left $R$-module $E_i$ (eg. the first column of $M_{n_i}(D_i)$). Moreover, $E_i\not\simeq E_j$, for all $i\neq j$. Any left $R$-module $M$ is a finite direct sum of submodules $M_i \simeq E_i^{(\Lambda_i)}$, for some set $\Lambda_i$. Also here, $\mathrm{Hom}(M_i,M_j)=0$, for all $i\neq j$. In particular, $R = E_1^{n_1} \oplus \cdots \oplus E_k^{n_k}$. If $n_i\geq 2$, for all $i$, then each $E_i^{n_i}$ is square-full and hence $R$ is a square-full left $R$-module, while if there exists one $n_i=1$, then $R$ contains the simple submodule $E_i$ and $E_i\oplus E_i$ does not embed into $R$. Hence $R$ is not a square-full left $R$-module.

The same argument goes through when dealing with right $R$-modules. Hence a semisimple Artinian ring $R$ is square-full as a left $R$-module if and only if all the matrix sizes $n_i$ are greater than $1$ if and only if $R$ is square-full as a right $R$-module.

$\endgroup$
3
  • 1
    $\begingroup$ Doesn't the definition in the question only require that every nonzero submodule contains a square-root, rather than is a square-root? $\endgroup$ Commented Oct 2 at 7:32
  • $\begingroup$ Oh, you are right and I misread the question. I'll fix my answer. $\endgroup$ Commented Oct 2 at 9:17
  • $\begingroup$ Dear prof. @ChristianLomp I really really appreciate your effort. Your thoughts and arguments are wonderful. I could write my own proof based on your argument and I'll post it as an answer below. It'll be nice to take a look at my argument. $\endgroup$ Commented Oct 3 at 15:12
0
$\begingroup$

Assume that $R$ is a semisimple left square-full ring. Since ${}_RR$ is a semisimple module, ${}_RR= S_1^{(n_1)} \oplus \ldots \oplus S_k^{(n_k)}$ where $k\in \mathbb{Z}^+$, $n_1,\ldots,n_k$ are positive integers, and each $S_\ell$ is a simple left $R$-module with $S_i \not\cong S_j$ for $i\neq j$. We show, under the hypothesis that ${}_RR$ is square-full, that $n_\ell \geq 2$ for each $\ell=1,\ldots,k$. Assume on the contrary that $n_t=1$ for some $t=1,\ldots,k$. If there exists an embedding $\gamma: S_t \oplus S_t \to {}_RR$, then $S_t \oplus S_t \cong \gamma(S_t \oplus S_t) \cong \bigoplus_{\lambda\in \Lambda} S_\lambda^{(a_\lambda)}$ where $\Lambda \subset \lbrace 1,\ldots, k \rbrace$ and $a_\lambda$ are positive integers with $a_\lambda \leq n_\lambda$ for each $\lambda\in \Lambda$. Observe that $\bigoplus_{\lambda\in \Lambda} S_\lambda^{(a_\lambda)}$ has at most one copy of $S_t$ whereas $S_t \oplus S_t$ is a direct sum of two copies of $S_t$, a contradiction. Consequently, $S_t \oplus S_t$ can not embedded in ${}_RR$ which implies that $S_t$ does not contain a nonzero square-root of ${}_RR$, contradicting the hypothesis that ${}_RR$ is square-full. Thus, $n_\ell \geq 2$ for each $\ell=1,\ldots,k$. Since $R_R$ is semisimple, then $R_R = U_1^{(b_1)} \oplus \ldots \oplus U_s^{(b_s)}$ where $s\in \mathbb{Z}^+$, $b_1,\ldots,b_s$ are positive integers, and each $U_m$ is a simple right $R$-module with $U_i \not\cong U_j$ for $i \neq j$. Put $\mathfrak{A}_\ell=\text{End}(S_\ell)$ and $\mathfrak{B}_m = \text{End}(U_m)$ for each $\ell=1,\ldots,k$ and $m=1,\ldots,s$. Since $\lbrace S_1,\ldots,S_k \rbrace$ and $\lbrace U_1,\ldots,U_s \rbrace$ are two collections of pairwise non-isomorphic simple modules, then $\text{Hom}\bigl(S_i^{(n_i)}, S_j^{(n_j)} \bigr)=0=\text{Hom}\bigl(U_\alpha^{(b_\alpha)}, U_\beta^{(b_\beta)} \bigr)$ for $i \neq j$ and $\alpha \neq \beta$. We now have the ring isomorphisms \begin{align*} R \cong \text{End}({}_RR) \cong \text{End}\bigl( S_1^{(n_1)} \oplus \ldots \oplus S_k^{(n_k)} \bigr) \cong \prod_{\ell=1}^k \text{End}(S^{(n_\ell)}_\ell) \cong \prod_{\ell=1}^k \mathbb{M}_{n_\ell}(\mathfrak{A}_\ell), \\ R \cong \text{End}(R_R) \cong \text{End}\bigl(U_1^{(b_1)} \oplus \ldots \oplus U_s^{(b_s)} \bigr) \cong \prod_{m=1}^s \text{End}\bigl(U^{(b_m)}_m \bigr) \cong \prod_{m=1}^s \mathbb{M}_{b_m}(\mathfrak{B}_m). \end{align*} Thus, $\prod_{\ell=1}^k \mathbb{M}_{n_\ell}(\mathfrak{A}_\ell) \cong \prod_{m=1}^s \mathbb{M}_{b_m}(\mathfrak{B}_m)$. By the uniqueness part of Wedderburn-Artin theorem, $k=s$ and $\mathfrak{B}_1,\ldots,\mathfrak{B}_s$ can be reindexed so that $b_t=n_t \geq 2$ for each $t=1,\ldots,k$. \ Now, let $X$ be a nonzero submodule of $R_R=U_1^{(b_1)} \oplus \ldots \oplus U_k^{(b_k)}$. Since $X$ is a summand of $R_R$, there exist $\Omega \subset \lbrace 1,\ldots,k \rbrace$ and positive integers $d_1,\ldots,d_k$ with $d_\ell \leq b_\ell$, for each $\ell=1,\ldots,k$, such that $X \cong \bigoplus_{\omega \in \Omega} U_\omega^{(d_\omega)}$. Let $Y$ be a simple submodule of $X$. Thus, $Y \cong U_\omega$ for some $\omega\in \Omega$ and hence $Y \oplus Y \cong U_\omega \oplus U_\omega \hookrightarrow R_R$ (as $b_\omega \geq 2$). That is to say $Y$ is a square-root in $R_R$ that is contained in $X$. Hence, as $X_R \subseteq R_R$ is arbitrary, $R_R$ is square-full. i.e. $R$ is right square-full. Analogously, $R$ is left square-full if $R$ is right square-full.

$\endgroup$

You must log in to answer this question.

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