Given a group $G$, find subsets $A,B$ such that $G=A\sqcup B$ and $A$ and $B$ are closed under multiplication: $x,y\in A$ (corr. $x,y\in B$) implies $xy\in A$ (corr. $xy\in B$).

For example, if $G$ is finite then all splitting are trivial: if $1\in A$ then $A=G$ and $B=\emptyset$.

If $G=\mathbb{Z}^d$ then any splitting is determined by a halfspace $\Pi^+$ in $\mathbb R^d\supset\mathbb Z^d$: $A=\Pi^+\cap\mathbb Z^d$. (More precisely, by a flag of halfspaces $\Pi^+\supset \Pi_1^+\supset\dots$, because we can distribute the points of the hyperplane $\partial\Pi^+$ between $A$ and $B$.)

What about the free group $G=F_d$? Are there splitting which are not induced by the projection $F_d\to\mathbb Z^d$?

  • 5
    $\begingroup$ You describe your motivation in a comment below, but, for future reference, such motivation should ideally go in the question itself. As is, the imperative ("find subsets …") and lack of context make this appear like a homework problem, and it could have been closed. $\endgroup$
    – LSpice
    Jul 24, 2021 at 20:28

1 Answer 1


This is a way to rediscover a quite well-studied class of groups:

Proposition Let $G$ be a group. Equivalent statements: (a) $G$ admits a partition $G=A\sqcup B$ with $1\in A$, $B$ nonempty, and both $A,B$ subsemigroups. (b) $G$ admits a nontrivial order-preserving action on some totally ordered set [which can be chosen to be the real line, or $\mathbf{Q}$, if $G$ is countable] (c) $G$ has a nontrivial left-orderable quotient.

The equivalence between (b) and (c) is classical. If $G$ has a non-trivial orderable quotient $p:G\to Q$, there is (by definition) a submonoid $K$ of $Q$ such that $K\cap K^{-1}=\{1_Q\}$ and $K\cup K^{-1}=Q$; pulling back to $G$ yields the desired decomposition.

Conversely, suppose that $G=A\cup B$ as above (with $1\in A$). Define $H=A\cap A^{-1}$: this is a subgroup (call it "core" of the decomposition).

Observe that $HB=BH=B$: indeed if $h\in H$ and $b\in B$, suppose by contradiction $a:=hb\in A$: so $b=ah^{-1}\in AA\subset A$, contradiction, same contradiction if $bh\in A$.

On $G/H$ define a total order by $gH\le g'H$ if $g^{-1}g'\in A$. Note that this doesn't depend on the choices of $g,g'$ modulo $H$ on the right. This relation is $G$-invariant, and is easily seen to be a total order. It is not reduced to a singleton (since $H=A$ would force $B$ empty, since the complement of a subgroup can't be closed under multiplication, unless empty).

Edit: this argument shows something more precise, not only describing the class of groups, but describing these decompositions:

Proposition 2: Let $G$ be a group. (1) If $G$ acts on a totally ordered set $X$ in a order-preserving and $x\in X$, define $A=\{g\in G:gx\le x$ and $B=\{g\in G:gx>x\}$. Then $G=A\sqcup B$ is a decomposition with the given axioms (and $1\in A$). (2) Conversely, any such decomposition of $G$ occurs in this way (for some such action and $x$, where we can assume in addition the action to be transitive).

Note that $B$ is empty iff $G$ fixes $x$.

Note that this shows that if $G$ has such a decomposition, it has one for which the core is a normal subgroup (while in general the core need not be normal).

Also, since non-abelian free groups are themselves left-orderable, they admit such decomposition that are not of the prescribed form (namely have such a decomposition with trivial core).

  • 1
    $\begingroup$ So, the problem belongs to the filed of ordering of groups. My goal is to describe such splittings for the fundamental groups of surfaces including the free groups. Now I see I should look for actions of those groups on totally ordered sets. Thank you very much! $\endgroup$
    – nim
    Jul 24, 2021 at 18:50
  • 1
    $\begingroup$ Very nice answer, but I’m confused by your last observation, “…non-abelian free groups […] admit such decomposition that are not of the prescribed form”, which seems to contradict Prop. 2(2). Presumably I’m misunderstanding what you mean by “of the prescribed form” here; could you clarify that? $\endgroup$ Jul 25, 2021 at 7:50
  • 1
    $\begingroup$ @PeterLeFanuLumsdaine "of the prescribed form" refers to OP's question, namely preimages of (some) orderings of the abelianization. $\endgroup$
    – YCor
    Jul 25, 2021 at 9:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.