2
$\begingroup$

Let $\mathcal{C}$ be an abelian category and $$ C_\bullet:C_n \rightarrow C_{n-1}\rightarrow \ldots \rightarrow C_1\rightarrow C_0$$ a complex in $\mathcal{C}$. Suppose we have for each $i$ a subobject $Y_i \leq H_i(C_\bullet)$. One of the properties is that $Y_i=0$ if and only if $H_i(C_\bullet)=0$.

If possible what could be methods to construct a complex $C^{'}_\bullet$ with homology $Y_i$?

$\endgroup$
2
  • 3
    $\begingroup$ Do you want $C_\bullet '$ to be a subcomplex of $C_\bullet$, or at least have a map of complexes (inducing your embedding on homology)? Otherwise you can just take direct sum of $Y_i[i]$ with zero differentials. $\endgroup$
    – user42024
    Commented Mar 18, 2021 at 13:13
  • $\begingroup$ You are right, I should be more precise. I hope for a subcomplex of $C_\bullet$ but at least a map of complexes as you mention seems to be necessary. $\endgroup$
    – KKD
    Commented Mar 18, 2021 at 15:00

1 Answer 1

2
$\begingroup$

$\newcommand{\dd}{\mathrm{d}}$Let $\mathcal C$ be the category of modules over the polynomial ring $R = k[x,y]$, where $k$ has characteristic zero. Let $D = k[x,y]\langle 1,\xi,\eta,\xi\eta\rangle$ be the Koszul resolution of the $R$-module $k$, i.e. $\xi,\eta$ live in degree one with $\dd\xi = X,\dd\eta=y,\dd(\xi\eta) = x\eta - y\xi$. Then there is a $R$-linear chain map $f:D\to D[2]$ which in degree $2$ sends $\xi\eta\mapsto 1$. Its mapping cone defines a complex $C$ (with $n=4$ in your notation) whose cohomology is $k$ in degrees $0$ and $1$ (and zero elsewhere). I claim that there is no complex $C'$ whose cohomology is $k$ concentrated in degree $0$ such that there is a chain map $C'\to C$ inducing the identity on zeroth cohomology. Indeed, we may assume that $C'$ is the projective resolution $D$, and by design there is an exact triangle $D[1]\to C\to D\to D[2]$ such that the identity of $D$ gets sent to the map $f$ which represents a nonzero element of $\operatorname{Ext}_R(k,k)$, so that the identity does not lift to $C$.

This example does not satisfy your additional condition since the submodule of the homology is zero in degree one. However, we can just take the sum with another chain complex with nontrivial first cohomology to essentially get rid of this condition.

In general, given two complexes $C',C$ there is a spectral sequence with $E_2^{p,q} = \prod_{k}\operatorname{Ext}^q(H_{q+k}(C'),H_{k}(C))$ converging to $\mathcal D(C'[p+q],C)$ (the signs are probably wrong, I'll try to fix them when I have more time). You are asking whether the inclusion of the submodule of homology, which defines an element in bidegree $(0,0)$, survives to the $E_\infty$-page. The example above is essentially obtained by forcing it to be the source of a nonzero $\dd_2$. However, since you haven't actually fixed the complex $C'$, there is considerable freedom in modifying the higher differentials in general.

$\endgroup$
2
  • $\begingroup$ Thanks for your detailed answer. At least interesting arguments! What is $\mathcal{D}(C'[p+q],C)$? $\endgroup$
    – KKD
    Commented Mar 19, 2021 at 14:39
  • $\begingroup$ Maps in the derived category, i.e. replace $C'$ and $C$ by chain complexes of projective/injectives, shift the source by $p+q$ and take chain homotopy classes of chain maps. $\endgroup$ Commented Mar 19, 2021 at 17:21

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .