Let $S$ be a semigroup such that for all $a,b \in S$ $$ aS \cup \{a\} = bS \cup \{b\} \quad \text{and} \quad Sa \cup \{a\} = Sb \cup \{b\}. $$ where $aS = \{as : s \in S \}$ and similarly $Sa$. I want to show that $S$ is a group.

The above conditions imply $$ a \in bS $$ for $a \ne b$, hence $S \setminus\{b\} \subseteq bS$ for all $b \in S$. If $S$ contains an idempotent $e$, then as $e \in eS$ we have $eS = S = Se$ and $e$ is the identity. With an identity element it is easy to see that $aS = S$ and $S = Sa$ for every $a \in S$, hence that $S$ is a group.

So, if I could show that $S$ contains an idempotent then it follows that $S$ is a group. But here I am stuck, so could anyone provide a hint how to proceed?

By the way, this is an exercise from J. Howie Fundamentals of Semigroup Theory (page 61).

  • $\begingroup$ Why is this downvoted? $\endgroup$ – Maria Mazur Jul 31 '18 at 12:51
  • $\begingroup$ Before downvoting please leave a comment what I could improve (or add a comment to your downvote at least)... I dont know either why this was downvoted! $\endgroup$ – StefanH Jul 31 '18 at 12:55
  • 1
    $\begingroup$ @DerekHolt That being said, if the purpose of actively downvoting is, as far as I can tell, to make this site a better place, then leaving a comment is the only way to teach the poster what they can do better. So downvoting without a comment serves little purpose except possibly scare new users away from the site (which, if I'm allowed to be cynical, might be a goal to some). $\endgroup$ – Arthur Jul 31 '18 at 13:33
  • 1
    $\begingroup$ I don't know much about semigroups, but i had a quick look at the book. Lemma 2.2.1 seems to say that multiplying by $s$ on the right is a bijection of $S$, and similarly on the left. So, for $a \in S$, there exists $e$ with $ae=a$, and then $ae^2=ae$, so $e^2=e$, and $e$ is an idempotent. $\endgroup$ – Derek Holt Jul 31 '18 at 14:01
  • 1
    $\begingroup$ @DerekHolt Okay, yes by my assummption there is just one $\mathcal L$-class! So this is fine, and also by this cancellability follows. $\endgroup$ – StefanH Jul 31 '18 at 14:06

As pointed out in the comments by DerekHolt, the key is Green's Lemma (as cited in the mentioned book on page 49). To formulate it, we start with an arbitrary semigroup $S$, and we define equivalence relations $\mathcal R$ and $\mathcal L$ on $S$ by setting \begin{align*} a \mathcal R b & :\Leftrightarrow \{a\} \cup aS = \{ b \} \cup bS, \\ a \mathcal L b & :\Leftrightarrow \{a\} \cup Sa = \{ b \} \cup Sb \end{align*} for all $a, b \in S$. These are part of what is known as Green's relations. Then we have:

(Green's Lemma) If $a \ne b$ and $a \mathcal R b$ with $as = b, a = bt$, then the maps $$ x \mapsto xs \quad \mbox{and} \quad x \mapsto xt $$ are mutually inverse bijections between the $\mathcal L$-classes of $a$ and $b$, which preserve $\mathcal R$-classes. A similar (dual) relation holds if $a\mathcal L b$.

Proof: Its easy to see that $\mathcal L$ is a left congruence and $\mathcal R$ is a right congruence. Hence if $x\mathcal L a$ then $xs \mathcal L as$, so $xs \mathcal L b$ because $b = as$. Similarly, right multiplication by $t$ maps the $\mathcal L$-class of $b$ into the one of $a$. So the two maps between the $\mathcal L$-classes are well-defined. Also, if $x = ua$ then $$ xst = uast = ubt = ua = x $$ and similarly $yts = y$ for all $y \mathcal L b$; hence they are mutually inverse on the corresponding $\mathcal L$-classes. Also, as $\mathcal R$ is a right congruence, the images of $\mathcal R$-related elements stay $\mathcal R$-related, but further as $x = (xs)t$ for $x \mathcal L a$ we have $x \mathcal R xs$ and so the mapping stays in the same $\mathcal R$-class for all elements $\mathcal L$-equivalent to $a$. $\square$

The assumption of the question gives that we just have a single $\mathcal R$-class and a single $\mathcal L$-class. So pick arbitrary $a \ne b$ and $as = b$ and $a = bt$, then for every $x \in S$ by the above we have $$ xst = x $$ giving that $e := st$ is idempotent (set $x := e$ above) and a right-identity. Further as the equations $ux = e$ and $xu = e$ are solvable for each $u \in S$ it must be a group.

Remark 1: With further theory from the book it is shown that every non-empty intersection of an $\mathcal L$-class and an $\mathcal R$-class is a group.

Remark 2: In the paper Intersections of maximal ideals in semigroups by P. Grillet I found

An ideal $I$ is maximal iff $S - I$ is a $\mathcal J$-class.

And the same proof works in case of right-ideals and $\mathcal R$-classes and left-ideals and $\mathcal L$-classes. So if $aS = S \setminus \{a\}$, then as $aS$ is a right ideal it is maximal and $\{a\}$ would be a single $\mathcal R$-class, which is excluded. Hence $aS = S$ for all $a \in S$. Similar $Sa = S$ for all $a \in S$ so that $S$ is a group. And the result follows without Green's lemma.


Your question is indeed related to Green's relations. The hypothesis tells you that all the elements of your semigroup are $\mathcal{R}$-equivalent and $\mathcal{L}$-equivalent. Therefore, there are all $\mathcal{H}$-equivalent. It now suffices to apply the following standard result (see for instance [1, Proposition 1.4]).

Let $H$ be an $\mathcal{H}$-class of a semigroup. The following conditions are equivalent:

  1. $H$ contains an idempotent,
  2. there exist $s$, $t \in H$ such that $st \in H$,
  3. $H$ is a group.

[1] P.A. Grillet, Semigroups, an introduction to the structure theory, Marcel Dekker, Inc., New York, 1995


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.