Sign up ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

A common way to define a group is as the group of structure-preserving transformations on some structured set. For example, the symmetric group on a set $X$ preserves no structure: or, in other words, it preserves only the structure of being a set. When $X$ is finite, what structure can the alternating group be said to preserve?

As a way of making the question precise, is there a natural definition of a category $C$ equipped with a faithful functor to $\text{FinSet}$ such that the skeleton of the underlying groupoid of $C$ is the groupoid with objects $X_n$ such that $\text{Aut}(X_n) \simeq A_n$?

Edit: I've been looking for a purely combinatorial answer, but upon reflection a geometric answer might be more appropriate. If someone can provide a convincing argument why a geometric answer is more natural than a combinatorial answer I will be happy to accept that answer (or Omar's answer).

share|cite|improve this question
I have a copout answer involving a total order on X but I would really not like to introduce a total order to solve this problem. Somehow I feel that the essence of the structure necessary is less than that. – Qiaochu Yuan Sep 22 '10 at 4:02
You can just take X_n to be a Cayley graph of A_n with some natural generating set. You probably don't count this as natural enough though. – Alon Amit Sep 22 '10 at 4:08
What is the precise meaning of "natural definition"? – Hans Stricker May 27 '11 at 15:19
@Hans: in this context, there really isn't one. – Qiaochu Yuan May 27 '11 at 15:33

5 Answers 5

up vote 10 down vote accepted

The alternating group preserves orientation, more or less by definition. I guess you can take $C$ to be the category of simplices together with an orientation. I.e., the objects of $C$ are affinely independent sets of points in some $\mathbb R^n$ together with an orientation and the morphisms are affine transformations taking the vertices of one simplex to the vertices of another. Of course this is cheating since if you actually try to define orientation you'll probably wind up with something like "coset of the alternating group" as the definition. On the other hand, some people find orientations of simplices to be a geometric concept, so this might conceivably be reasonable to you.

share|cite|improve this answer
I considered this, but I wonder if the definition can be made as set-theoretic as possible, with as little geometry as possible. – Qiaochu Yuan Sep 22 '10 at 4:25
@Qiaochu, since orientation has to do with determinants, maybe you could whittle the above down to a discrete interpretation of determinants -- the determinants of permutation matrices, perhaps. Or would that be circular? The determinant does have an independent definition unrelated to the signature of a permutation. – Rahul Sep 22 '10 at 4:42
So the question is: what is an oriented set? Answer: It's a set with an orientation. So what is an orientation on a set? ${}\qquad{}$ – Michael Hardy May 15 at 21:51

$A_n$ is the symmetry group of the chamber of the Tits building of $\mathbb{P}GL_n$. The shape of this chamber is independent of what coefficients you insert into the group scheme $\mathbb{P}GL_n$, just the number and configuration of chambers changes. If you insert the finite fields $\mathbb{F}_p$ then you get finite simplicial complexes as buildings, and the smaller $p$ gets, the fewer chambers you have. You can analyse and even reconstruct the group in terms of its action on this building. The natural limit case would be just having one chamber and the symmetry group of this chamber - the Weyl group - is $A_n$. This is how Tits first thought that there is a limit case to the sequence of finite fields - which he called the field with one element.

Maybe somewhat more algebraically you can think in terms of Lie algebras - as I said the shape of the chamber does not change with different coefficients. The reason is that it is determined just by the Lie algebra of the group and thus describable by a Dynkin diagram or by a root system (ok, geometry creeps in again). The Wikipedia page about Weyl groups tells you that the Weyl group of the Lie algebra $sl_n$ is $S_n$. If have no experience with Lie algebras, but maybe you can get $A_n$ the same way.

If you can get hold of it, you can read Tits' original account, it's nice to read (but geometric) see the reference on this Wikipedia page.

Edit: Aha, I found a link now: Lieven Le Bruyn's F_un page is back online. You can look there under "papers" and find Tits' article. And, since you are picking up the determinant ideas, you should definitely take a look at Kapranov/Smirnov!

share|cite|improve this answer
Doesn't GL(n) involve ordering an n-element set (of basis vectors)? The question Qiaochu is raising is something like, can we construct A(X), with X a set or a set with "Alt-structure" weaker than an ordering. The A_n and S_n and GL_n with their standard embeddings into each other assume the standard ordering. One can discuss GL(V) for V a vector space, but constructing it as an algebraic group uses an order, at least in the usual presentation of the theory. – T.. Sep 29 '10 at 17:12
Does it? Aren't the group operations still there when you forget about the coordinates, which you used to define it via matrix multiplication? The construction of the building then uses just coordinate-free notions like "Borel subgroups", as far as I remember... – Who Sep 29 '10 at 17:25
I am also starting to find that strange now: Does an embedding of S_n into GL_n give me a choice of basis? Maybe via picking eigenvectors? If not one could define A_n as the intersection of S_n with O_n (which is definable coordinate-free) under any embedding - and then use coordinates to show independence of the embedding. – Who Sep 29 '10 at 17:32
I should have added that in addition to defining A(X) (which requires no structure except the finite set X), the problem here is to define even bijections from X to Y. This cannot be done without extra data, but does not require the full strength of an ordering. Functorial construction of Tits building from X and Y would, for any bijection, give a map from the X-building to the Y-building and asking whether it is an isomorphism of whatever structure in the building defines A(X) would define even bijections. So I suspect some sort of extra data is hidden in the construction. – T.. Sep 30 '10 at 7:30
I don't know whether the construction of the Tits building is functorial. I didn't mean to imply this with anything I wrote. It would be nice of course, then an embedding of S_n might correspond to the inclusion of a particular chamber. – Who Sep 30 '10 at 15:33

Here is one idea, although I do not find it very satisfying. An object of $C$ is a finite set $X$ equipped with $\frac{|X|!}{2}$ (or $1$ if $|X| = 1$) total orders, all of which are even with respect to each other (in other words, basically a coset of $A_n$ in $S_n$). A morphism between two objects in $X$ is a map of sets preserving these orders (in other words, take one of the orderings on $X$ and apply a function $f : X \to Y$ to its elements. The result, after throwing out repeats, must be compatible with an ordering on $Y$.)

This is more or less a discretization of Omar's answer. Again, I would like to do better than this, or at least see the data described above packaged in a more satisfying way.

share|cite|improve this answer

You can do with something very slightly weaker than an ordering: an identification of each $n$-element set with a single "universal" (unordered) $n$-set. This data canonically identifies any two $n$-element sets and thus associates a permutation to any bijection of such sets. Even bijections can then be defined in terms of the cycle structure of the permutations.

This is not much of an improvement, but you are in effect asking for a lifting of the alternating group to a groupoid of maps between finite sets. It is hard to see how to determine whether a bijection of finite sets is even (reducing to the usual notion when the sets are the same, and also "transporting structure" along the whole category) without having a coordinatization of the sets.

share|cite|improve this answer

(This is, essentially, just a «repackaging» of your answer. Still, I find this version somewhat more satisfying — at least, it avoids even mentioning total orders.)

For a finite set $X$ consider projection $\pi\colon X^2\to S^2 X$ (where $S^2 X=X^2/S_2$ is the symmetric square). To a section $s$ of the projection one can associate a polynomial $\prod\limits_{i\neq j,\,(i,j)\in\operatorname{Im}s}(x_i-x_j)$ — and since any two such products coincide up to a sign, this gives a partition of the set $\operatorname{Sec}(\pi)$ into two parts. Now, $A_X$ is the subgroup of $S_X$ preserving both elements of this partition. (I.e. the structure is choice of one of two elements of the described partition of $\operatorname{Sec}(\pi)$.)

share|cite|improve this answer

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.