Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

What is the right definition of an exact sequence of monoid homomorphisms?

I can't seem to find a consistent in my searches; indeed Balmer (Remark 2.6, http://www.math.ucla.edu/~balmer/research/Pubfile/Prod.pdf) calls it a "slippery notion".

Among other things, I'll know that it's a good definition if I can take an exact sequence of groups and forget inverses and obtain an exact sequence of monoids.

One possible answer is just to take the definition for groups: a sequence $$A \xrightarrow{f} B \xrightarrow{g} C $$ of monoid homomorphisms is exact if $\mathrm{ker}\ g = \mathrm{img}\ f$, i.e., $$ \{ y \in B : g(y) = 1 \} = \{ f(x) : x \in A \}. $$ But, according to Bergman (3.10.6--7, http://math.berkeley.edu/~gbergman/245/Ch.3.pdf), this definition of kernel does not determine the image. Instead, we look at the \emph{kernel congruence} $$ K_g = \{(y,z) \in B \times B : g(y)=g(z)\}. $$ The set $K_f$ defines a \emph{congruence} on $A$, an equivalence relation compatible with the operation on $A$, i.e., if $(y,z),(y',z') \in K_g$ then $(yy',zz') \in K_g$.

So then how should we define the image $I_f$ so that our sequence is exact if and only if $I_f=K_g$. One proposal is given by Arturo Magidin (http://math.stackexchange.com/questions/18387/is-there-an-analogue-of-short-exact-sequences-for-semigroups): he defines $$ I_f = (f(A) \times f(A)) \cup \Delta(B) $$ where $\Delta(B)=\{(b,b) : b \in B\}$.

This notion, though probably useful in some ways, has a grave defect. Let $$ A \xrightarrow{f} B \xrightarrow{g} C, $$ be an exact sequence of groups, so $$ \mathrm{ker} g = \{y \in B : g(y) = 1\} = \mathrm{img} f = \{f(x) : x \in A\}. $$ I would like to verify that the sequence is exact as a sequence of monoids, so $I_f=K_g$. We have $(y,z) \in K_g$ if and only if $g(yz^{-1}) = 1$ since g is a group homomorphism, if and only if $yz^{-1} \in K = \mathrm{ker} g$ if and only if $y = kz$ for some $k \in K$. Thus $$ K_g = \{(kz,z) \in B \times B : k \in K, z \in B \}. $$

On the other hand, suppose g is not injective and that f is not surjective (not an atypical situation). Then there exists $z \in B$ such that $z \not\in f(A)$ and there exists $k \in K$ such that $k \neq 1$. Then $(kz,z) \in K_g$ by definition, but $(kz,z) \not\in I_f$: it does not belong to $f(A) \times f(A)$ since $z \not \in f(A)$ and does not belong to $\Delta(B)$ since $kz \neq z$. So $K_g \not\subseteq I_f$.

I can see why one must add the diagonal to $f(A) \times f(A)$, since one wants a reflexive relation on B. It follows that $I_f$ is an equivalence relation. I would think that one should go farther and take the image to be the congruence closure of $I_f$, i.e., the submonoid $\overline{I}_f$ of $B \times B$ generated by $f(A) \times f(A)$ and $\Delta(B)$. Then it is clear that $(kz,z) = (k,1)(z,z)$ in $\overline{I}_f$, and so $K_g \subseteq \overline{I}_f$. Conversely, it is clear that $\overline{I}_f$ is a subset of $K_g$, so then I'm happy again.

But this definition has a different problem: an exact sequence $$ A \xrightarrow{f} B \to 1$$ of monoids is exact if $f$ is surjective but not conversely: the inclusion map $$ \mathbb{N} \xrightarrow{f} \mathbb{Z} $$ of the natural numbers in the integers has $\overline{I}_f = \mathbb{Z} \times \mathbb{Z}$ but is not surjective. Arturo Magidin pointed out to me in an e-mail that this map $f$ is an epimorphism in the category of monoids; maybe I'll have to accept that.

Can anyone shed some light on this matter? I'd already be pretty happy with a reference.

share|improve this question
The category of monoids is an exact category in the sense of Barr, see ncatlab.org/nlab/show/exact+category. So it has a natural notion of exact sequence. –  Benjamin Steinberg Dec 9 '11 at 23:02
To add to Benjamin's and James's responses, any category that is algebraic (or monadic) over $Set$ is Barr-exact. –  Todd Trimble Dec 9 '11 at 23:50
I am always wary of the word 'right' in the first line of the question. "Horses for courses", but given that there was a very nice preprint by Charlie Wells: cwru.edu/artsci/math/wells/pub/pdf/catext.pdf which might be of use. The original dated to 1980. –  Tim Porter Dec 10 '11 at 7:00
add comment

2 Answers

up vote 13 down vote accepted

Hi John. I'd say there is no generalization of short exact sequence to the category of monoids, although I suppose it really depends on what you want to do with it. What you probably want is an internal equivalence relation. So you could say a diagram $A\rightrightarrows B \to C$ of monoid maps (where the two compositions $A\rightrightarrows C$ agree) is short exact if the map $B\to C$ is surjective and the induced map $A\to B\times_C B$ is an isomorphism. This is equivalent to requiring the induced map $A\to B\times B$ to be injective, its image to be an equivalence relation, and the induced map $B/A\to C$ is to be an isomorphism.

My general feeling is that this is the right concept in most categories of sets with algebraic structure (e.g. the category of sets itself, semi-rings). It's only in categories where the objects have some group structure that you can re-express it using kernels.

share|improve this answer
I see Ben said the same thing in a comment to the original post. –  JBorger Dec 9 '11 at 23:17
This doesn't exactly answer my question, since it only gives me short exact sequences, but I'll take it. Thanks James (and Ben)! –  John Voight Dec 10 '11 at 17:17
John, the monoid analogue of a long exact sequence is probably a simplicial monoid that is contractible (as a simplicial set), or something like that. –  JBorger Dec 11 '11 at 5:55
add comment

For the special case of commutative monoids, or more generally semimodules over a semiring, my related preprints on arXiv might be helpful (see below). For arbitrary monoids, the general definition is similar, however it is difficult to apply since the notion of the cokernel of a morphism of monoids is really "slippery":



share|improve this answer
add comment

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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