Let $R$ be a reasonable ring (maybe I mean a PID, or $\mathbb{Z}$, and when sufficiently desperate, a field). Now consider fixed sequences $C_n$ and $H_n$ of $R$-modules, which are tame in every possible sense -- finite rank, zero for large enough $n$, etc. Here's the problem:

What is the set (space? category?) of all $R$-linear maps $d_n:C_n \to C_{n-1}$ so that $(C_\bullet,d_\bullet)$ forms a chain complex with homology groups $H_n$?

I've tried to figure things out in baby cases without much luck in terms of seeing the bigger picture. You can easily create situations where no $d_\bullet$s will suffice, but most fascinating are the cases where several different sequences of boundary operators will work!

A very general thing to do is fix a basis of all chain groups in sight and examine conditions that entries of various $d_\bullet$-representing matrices have to satisfy. Just requiring the $d_\bullet$s to produce a chain complex yields an algebraic variety generated by a system of multivariate quadratic equations (which sounds nightmarish). So unless forcing the homology to be prescribed by $H_n$s greatly simplifies things, I should not expect a very explicit answer. However, I'd like to know what the obstructions to getting such an answer might be.

In the cases of interest, all my $d_\bullet$s can be assumed to have an upper-triangular form as $R$-matrices, although I'd also be interested in a solution that ignores such structural constraints.

Update: Thanks to David Speyer's answer, which provides a wealth of information on the problem when $R$ is a field. I'm still seeking answers for more general choices of $R$.

  • $\begingroup$ I speak from memory so I may not be that precise. You should look at the derived category of finite complexes of $R$-modules. If $R=\mathbb{Z}$, then any complex of Abelian groups is quasi-isomorphic to direct sums of trivial complexes $\cdots 0\to C_n\stackrel{0}{\to} C_{n+1}\to 0\cdots$, $n\in \mathbb{Z}$. $\endgroup$ – Liviu Nicolaescu Sep 16 '15 at 16:45
  • 1
    $\begingroup$ Am I misunderstanding you? If $H_n=C_n$ for all $n$, doesn't this force all $d_n$ to vanish? $\endgroup$ – Dan Petersen Sep 16 '15 at 16:52
  • $\begingroup$ @DanPetersen No, I just wrote something stupid. Fixing it now. $\endgroup$ – Vidit Nanda Sep 16 '15 at 17:03
  • 1
    $\begingroup$ The complex you wrote is quasi-isomorphic to a trivial complex. Two complexes are quasi-isomorphic if there exists a morphism of complexes between them that induces an isomorphism in homology. $\endgroup$ – Liviu Nicolaescu Sep 16 '15 at 17:14
  • 2
    $\begingroup$ "when sufficiently desperate, a field" I have often wondered about this question over $\mathbb{C}$, even just for exact sequences. For short exact sequences, we recover (both forms of the) Grassmannian, which is the primordial moduli space, and an endless supply of fascinating questions. I guess it would be fun to work over $\mathbb{Z}$ one day, but I don't think any desperation comes into it :) $\endgroup$ – John Wiltshire-Gordon Sep 16 '15 at 18:43

Over a field $k$, this is a special case of quiver loci. The general framework is that you are given a triangular array $(r_{ij})_{1 \leq i \leq j \leq n}$ of nonnegative integers. Consider a list of $n-1$ matrices $M_1$, $M_2$, …, $M_{n-1}$ of size $r_{ii} \times r_{(i+1)(i+1)}$. The quiver locus is the set where $\mathrm{rank}(M_i M_{i+1} \cdots M_{j-1}) \leq r_{ij}$. Your case is where $r_{ij}=0$ for $j-i \geq 2$.

See Knutson, Miller and Shimozono, and the many citations therein.

To be a bit more explicit, if you want $\dim V_i = d_i$ and $\dim H_i(V_{\bullet})=h_i$, then take $r_{ii} = d_i$, compute $r_{i(i+1)}$ by solving the linear equations $r_{(i-1)i} + r_{i(i+1)} + h_i= d_i$, and put all other $r_{ij}=0$.

  • $\begingroup$ This certainly gives a chain complex, but how does one go about constraining kernel mod image so that we get the desired homology? $\endgroup$ – Vidit Nanda Sep 17 '15 at 1:46
  • 2
    $\begingroup$ I've edited in an answer to that question. $\endgroup$ – David E Speyer Sep 17 '15 at 1:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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