I cannot manage to prove that a free monoid with operation concatenation, and with at least two generators is residually finite. If there is just one generator, the free monoid $\{a\}^*$ is isomorphic to $\langle N, +\rangle$ which is residually finite. What happens if there are two or more generators? I found some results about monoids with relations on the generators, but I don't see how to lift the result to the free monoid. I will be grateful for any help or reference.
-
1What definition of "residually finite" are you using? – Jul 20 '23 at 10:48
-
3One way to prove this would be to embed your free monoid into the free group over the same alphabet, and then apply the residual finiteness of the free group. Proofs of this result can be found in many books, or on MathOverflow. – ADL Jul 20 '23 at 10:57
-
2@ADL, the proof for monoids is much easier than for groups. – Benjamin Steinberg Jul 20 '23 at 13:34
-
4@BenjaminSteinberg Indeed it is much easier; my answer could even be compressed into a single line without any real loss of understanding ("the quotient of the free monoid by the ideal of all elements longer than $u$ and $v$ will be a finite monoid separating $u$ and $v$"). – Carl-Fredrik Nyberg Brodda Jul 20 '23 at 13:54
-
3@Carl-FredrikNybergBrodda that is the shortest argument, but one can get a much smaller monoid by assuming wlog that u is not a factor of v and taking A/AuA* which sends u to 0 but not v. This semigroup has size the number of factors of u+1 which is quardratic in the length of the larger guy at worse. – Benjamin Steinberg Jul 20 '23 at 14:01
-
1@BenjaminSteinberg the immediate argument I saw was to squeeze to 0 all words of length $>n$. This remains injective on words of length $\le n$. It seems even simpler, yet it doesn't yield a quadratic bound. – YCor Jul 20 '23 at 22:36
-
2@YCor, your argument is the same as Carl-FredrikNybergBrodda's answer. That's what Rees quotient means. The simple argument is to write down a determinstic partial automaton accepting the shorter word (so a straightline automaton reading that word) and then the other word cannot be read on this automaton. The transition monoid is basically what I wrote above. – Benjamin Steinberg Jul 20 '23 at 22:48
-
1I miswrote my first comment. I should have taken as my ideal all words which are not a factor of v. I accidentally wrote $A^uA^$ instead. Then modding out by this correct ideal consists of only the factors of v and 0. In particular u is sent to 0 and v is sent to itself. This semigroup is quadratic in length of v. – Benjamin Steinberg Jul 21 '23 at 00:33
-
1Thanks so much for your several proofs ! – iguessarian Jul 21 '23 at 18:16
3 Answers
The standard meaning of residually finite here is that for every pair of elements $u, v$ in the free monoid $A^\ast$, if $u \neq v$ then there is a homomorphism $\phi \colon A^\ast \to F$ with $F$ a finite monoid and $\phi(u) \neq \phi(v)$. Proving that free monoids are residually finite is conceptually quite simple; no fancy tools are needed.
So, given distinct $u, v \in A^\ast$, consider the ideal $I$ of $A^\ast$ which consists of all words of length $>|u|+|v|$ (here $|u|$ denotes the length of the word $u$). Obviously, this is a (two-sided) ideal; if I multiply any word in $A^\ast$ by an element from $I$, then I will remain in $I$, as the length of products in $A^\ast$ can only increase. We can then consider the Rees quotient semigroup $A^\ast/I$ of $A^\ast$ by $I$, which is just a fancy way of saying "collapse all elements of $I$ into a single element $0$, and define multiplication in the obvious way". Then $A^\ast / I$ is a finite monoid - indeed, its elements are all the words of length $\leq |u|+|v|$, and a new element called $0$, with multiplication as before (and, if the product is $>|u|+|v|$, then we set the product to be $0$). The obvious homomorphism from $A^\ast$ to $A^\ast / I$ will map $u$ and $v$ to distinct elements, as neither $u$ nor $v$ are long enough words to be killed by the quotient. Thus free monoids are residually finite.
(This proof does not carry over to free groups - after all, the set of words of length $\geq n$ in a free group does not form an ideal!)
-
1Doesn't this only work with finitely generated free monoids? If the alphabet is not finite, then the Rees quotient you construct is infinite. Or am I missing something? – Salvo Tringali Jul 20 '23 at 23:38
-
5@SalvoTringali, If you have two words in an infinitely generated free monoid then you can send all letters not in the words to 1 to reduce to the finitely generated case. Or you can add to your ideal all words containing a letter not appearing in any of the two words. – Benjamin Steinberg Jul 21 '23 at 00:28
-
1@BenjaminSteinberg I agree. It's just that this would be better mentioned in Carl-Fredrik Nyberg Brodda's answer. – Salvo Tringali Jul 21 '23 at 00:40
-
1@SalvoTringali To modify it for infinitely generated free monoids is just as BenjaminSteinberg puts it. Most answers (and the question there) on the free group question assume finite generation, so I did too. – Carl-Fredrik Nyberg Brodda Jul 21 '23 at 11:40
-
1@Carl-FredrikNybergBrodda I don't see any reference to finite alphabets in the OP (whence my comment). With that said, I love your proof. – Salvo Tringali Jul 21 '23 at 11:43
-
1@SalvoTringali Sure, in this question there is no reference to finite alphabets, so you're quite right to point it out! – Carl-Fredrik Nyberg Brodda Jul 21 '23 at 11:45
-
1Is the first sentence of your second paragraph supposed to be "consider the ideal $I$ of $A^*$"? – cody Jul 21 '23 at 15:54
-
1@cody Yep — thank you! ($F$ for finite, not free :-). – Carl-Fredrik Nyberg Brodda Jul 21 '23 at 17:43
-
1
-
1@einsupportsModeratorStrike I don't need it -- it's sufficient! Of course $\operatorname{max}(|u|, |v|)$ works too. – Carl-Fredrik Nyberg Brodda Jul 22 '23 at 21:21
Here is another proof that I like, although it is not as simple as modding out by a cofinite ideal as per @Carl-FredrikNybergBrodda's answer. This approach I learned from Stalling's old geometric group theory lecture notes. Hopefully I am correctly recalling it. Since all free monoids on a countable generating set embed in the free monoid on 2 generators $\mathbf 0$, $\mathbf 1$ it is enough to do that case.
I'll give a representation of the free monoid on $2$ generated by $2\times 2$ nonnegative matrices (invertible over $\mathbb Z[1/2]$) that is particularly easy to verify is faithful (but this representation does not extend faithfully to the free group). Since the monoid of $2\times 2$ integer matrices is clearly residually finite by taking the entries modulo large primes, we are done.
Map $\mathbf 0$ to $\begin{bmatrix}1 & 0 \\ 0 & 2\end{bmatrix}$ and map $\mathbf 1$ to $\begin{bmatrix} 1 & 1\\ 0 & 2\end{bmatrix}$. Then a simple induction shows a word $w$ maps to $\begin{bmatrix} 1 & d(w)\\ 0 & 2^{|w|}\end{bmatrix}$ where $|w|$ is the length of $w$ and $d(w)$ is the natural number whose base $2$ expansion is $w$ (with perhaps some leading zeroes). Knowing the length of $w$ and the integer $d(w)$ lets you recover $w$ since from the length you know how many leading zeroes to add.

- 37,275
-
1
-
1This one proves that the free monoid is residually a finite group (and not just residually finite). – YCor Jul 22 '23 at 12:55
-
1Yes it gives residually a finite $p$-group for p not equal 2 – Benjamin Steinberg Jul 22 '23 at 14:59
One can generalize the notion of residual finiteness to universal algebra. An algebraic structure $A$ is said to be residually finite if $A$ is isomorphic to a subdirect product of finite algebraic structures. Equivalently, $A$ is residually finite if for each $a,b\in A$ with $a\neq b$, there is a homomorphism $\phi:A\rightarrow B$ where $B$ is a finite algebraic structure and $\phi(a)\neq\phi(b)$.
As it was observed by ADL in the comments, one can use the fact that free groups are residually finite to conclude that free monoids are also residually finite as a corollary (on this site, here is a collection of proofs that free groups are residually finite).
We may also establish residual finiteness of $A^*$ using monoid homomorphisms $\phi:A^*\rightarrow M$ both when $M$ is always a group and when $M$ is far away from being a group.
If $R,S$ are sets, then $R^S$ denotes the set of all functions from $S$ to $R$.
Let $A,B$ be finite sets. Suppose $*:A\times B\rightarrow B$ is a binary operation $*$. For each $a\in A$, let $L_{a,*,n}:B^n\rightarrow B^n.$ be the mapping defined by letting $L_{a,*,n}(b_1,\dots,b_n)=(a*b_n,b_1,\dots,b_{n-1})$. We can define a monoid homomorphism $\phi_{*,n}:A^*\rightarrow (B^n)^{B^n}$ by letting $\phi_{*,n}(a_1\dots a_r)=\phi(a_1)\dots\phi(a_r)=L_{a_1,*,n}\dots L_{a_r,*,,n}$. Then $$\phi_{*,n}(a_1\dots a_n)(b_1,\dots,b_n)=(a_1*b_1,\dots,a_n*b_n).$$
Observation 1: If the operation $*$ is left cancellative, then the mapping $\phi_{*,n}(a_1\dots a_n)$ is injective, so $\phi_{*,n}[A^*]$ is a subgroup of $S(A^n)$. This shows that the free monoid embeds into a product of symmetric groups.
Observation 2: If $A=B$ and the operation $*$ satisfies $x*y=x$, then the semigroup $\phi_{*,n}[A^*]\setminus\{e\}$ satisfies the identity $x_1\dots x_ny_1\dots y_n=x_1\dots x_n$. This means that $A^*$ can be embedded into a product of monoids $M$ where $M\setminus\{e\}$ is a subsemigroup that satisfies the identity $x_1\dots x_ny_1\dots y_n=x_1\dots x_n$.
There are other constructions for embedding $A^*$ into a product of non-group-like finite semigroups.

- 27,791
-
1
-
1What does the notation $(A^n)^{A^n}$ mean? And what definition of residually finite are you using? – Carl-Fredrik Nyberg Brodda Jul 20 '23 at 12:59
-
1@Carl-FredrikNybergBrodda I am not aware of a definition of residual finiteness that is not trivially equivalent to a usual one. – Joseph Van Name Jul 20 '23 at 13:14