# Bi-embeddability vs. isomorphism

Can anybody give me an example of a "naturally-occurring" algebraic category $C$ in which:

1. $C$ has two non-isomorphic objects $A$ and $B$ which are bi-embeddable via monic maps; but

2. $C$ does NOT have any infinite collection $A_{1}$, $A_{2}$, ... of objects which are pairwise bi-embeddable (via monic maps) and pairwise nonisomorphic?

Alternatively: can anybody give a reason why, under some reasonable hypothesis about the category $C$, property 1 should imply that 2 fails ("there is an infinite collection of pairwise bi-embeddable, pairwise nonisomorphic objects")?

-

I don't know if this counts as a "naturally-occuring" algebraic category but as I scanned through my internal list of categories of modules over a ring, I spotted that the following has (1) and (2).

Take $G$ to be the "$p$-adic dihedral group" that is a semidirect product of the p-adic integers and a cyclic group of order $2$ where the latter acts by $-1$ on the former. Then form the completed group algebra (Iwasawa algebra) with coefficients in the field with $p$ elements, $R=F_{p}\left[\left[G\right]\right]$.

Now there are precisely two isomorphism classes of indecomposable projective modules [$P_{1}$] and [$P_{2}$] and $P_{1}$ and $P_{2}$ are bi-embeddable via monic maps (see section 9.6 of http://www.dpmms.cam.ac.uk/~sjw47/char.pdf for the proof of both these claims).

It follows that if we take the category of all finitely generated projective modules over $R$ then it consists of objects that are direct sums of m copies of $P_{1}$ and $n$ copies of $P_{2}$, say. Two such objects will be bi-embeddable if and only if they have the same value of $m+n$ thus whilst there are arbitrarily large finite collections of pairwise bi-embeddable but pairwise nonisomorphic objects in this category there are no infinite ones.

I am sure there will be many more examples along these lines, I just happen to spend a lot of time thinking about Iwasawa algebras.

-
Thanks! This is exactly the kind of example I was looking for. Abstracting a bit, it seems that if you ever have a "reasonable" category of algebraic objects with two "indecomposables" A and B which are bi-embeddable but nonisomorphic, then you ought to be able to take the subcategory of all objects "finitely generated by A and B" and get a similar thing satisfying both 1 and 2. I'm too lazy at the instant to think what's the optimal way to make this precise -- maybe it should be a statement about abelian categories? –  John Goodrick Oct 15 '09 at 22:54
If by finitely generated you mean under direct sum in an additive category then I think that will be true. –  Simon Wadsley Oct 16 '09 at 7:05

Natural example: the category of function fields of supersingular elliptic curves over a fixed algebraically closed field of characteristic $p$ (you pick $p$, say bigger than $13$ or so to ensure more than one isomorphism class). Note that this is -- up to translations -- the opposite category to the Brandt module category, in which the objects are the ss elliptic curves themselves and the morphisms are isogenies.

As $p$ varies, this gives a family of (essentially) finite categories such that any two objects are mutually embeddable and the number of isomorphism classes of objects tends to infinity with $p$.

If I may be so bold, I spent much of a paper talking about the relation between two function fields that each is embeddable in the other, which I called (borrowing from the theory of elliptic curves) "isogeny".

PLC, On elementary equivalence, isomorphism and isogeny.
J. Théor. Nombres Bordeaux 18 (2006), no. 1, 29--58.

http://math.uga.edu/~pete/logic.pdf

-