21
$\begingroup$

One motivation for studying infinity categories is that the derived category does not satisfy Zariski descent, although the infinity categorical version does.

I would like to see an example of Zariski descent failing for the (one categorical) derived category of coherent sheaves on a scheme.

That is I would like an example of an (preferably affine) algebraic variety $X$, an open cover $\{U_i \hookrightarrow X\}$ and a descent datum in the derived category which does not glue uniquely to an element of the derived category of $X$.

$\endgroup$
3
  • 3
    $\begingroup$ My hunch is that you will have to settle for an example of non-uniqueness: two different objects on $X$ such that their descent data in the derived category is isomorphic. For etale descent you could have non-existence. $\endgroup$ Commented Mar 6, 2023 at 12:51
  • $\begingroup$ Thanks, I edited the question to reflect the fact I would be very happy with an example of non--uniqueness of gluing $\endgroup$
    – Mathmop
    Commented Mar 6, 2023 at 13:18
  • 2
    $\begingroup$ I guess you want something along the lines of page 8 example of the notes perso.math.univ-toulouse.fr/btoen/files/2012/04/swisk.pdf $\endgroup$
    – Edouard
    Commented Mar 7, 2023 at 19:30

1 Answer 1

12
$\begingroup$

Consider the canonical functor $H$ from the homotopy category of homotopy coherent descent data in the ∞-category of coherent sheaves to the category of descent data in the derived category of coherent sheaves.

Nonuniqueness of gluing amounts to $H$ not being a full functor and is observed for covers of cardinality 3 and higher.

Nonexistence of gluing amount to $H$ not being essentially surjective and is observed for covers of cardinality 4 and higher.

For covers of cardinality 0, 1, and 2, the functor $H$ is an equivalence.

Let's analyze first the case of covers of cardinality 3. Given three opens $U_1$, $U_2$, $U_3$, an object in the domain of $H$ will have chain complexes of coherent sheaves $F_i$ over $U_i$, isomorphisms $t_{i,j}:F_i→F_j$ of their restrictions to $U_i∩U_j$ ($1≤i<j≤3$, setting $t_{j,i}=t_{i,j}^{-1}$), and a choice of a homotopy $h$ (or rather its homotopy class) from the composition $t_{3,1}t_{2,3}t_{1,2}$ to the identity map on the restriction of $F_1$ to $U_1∩U_2∩U_3$. In particular, objects with the same data of $F_i$ and $t_{i,j}$, but nonhomotopic $h$ are nonisomorphic.

An object in the codomain of $H$ has a similar description, except that $h$ is merely required to exist, but its data is not included.

It is now clear how to show that $H$ is not full: pick two objects in the domain of $H$ with the same data of $F_i$ and $t_{i,j}$, but different (nonhomotopic) $h$. Their images under $H$ will see the data of $h$ discarded, so the two objects become isomorphic in the codomain of $H$, despite not being isomorphic in the domain of $H$.

Therefore, to construct a concrete example, one needs to specify $U_i$, $F_i$, and $t_{i,j}$ such that the identity map $p$ on $F_1$ restricted to $U_1∩U_2∩U_3$ admits a homotopy $h$ from $p$ to itself that is not homotopic to the identity homotopy from $p$ to itself. That is to say, we want $\def\Aut{\mathop{\sf Aut}\nolimits}\def\End{\mathop{\sf End}\nolimits} π_1(\Aut(F_1))$ to be nontrivial. For example, we can take $\def\cO{{\cal O}} F_1=\cO[0]⊕\cO[1]$, then $$\End(F_1)=\cO[-1]⊕\cO[0]⊕\cO[0]⊕\cO[1],$$ and a nontrivial 1-cocycle in $\Aut(F_1)$ yields such a homotopy $h$.

The case of covers of cardinality 4 is similar. As before, for an object in the domain of $H$ we have $U_i$ and $F_i$ ($1≤i≤4$), $t_{i,j}$ ($1≤i<j≤4$), $h_{i,j,k}$ ($1≤i<j<k≤4$), and now on $⋂_i U_i$ there is a homotopy $q$ from a certain composition of homotopies $h$ to the identity homotopy on the identity map on the restriction of $F_1$ to $⋂_i U_i$.

The functor $H$ discards the data of $h$ and $q$. In the codomain of $H$, some maps $h$ must exist (but their choice is not given to us), but $q$ need not exist at all. Indeed, by modifying the above example appropriately, we can construct descent data for the derived category in which the composition of homotopies $h$ is not homotopic to the identity homotopy, which means that the existence of gluing is violated for the derived category.

$\endgroup$
4
  • 5
    $\begingroup$ I can't really follow your examples. What is the variety, and what are open sets $U_i$? $\endgroup$ Commented Mar 6, 2023 at 21:34
  • 3
    $\begingroup$ @R.vanDobbendeBruyn: For example, take the projective space with coordinates z_i and U_i being given by the condition z_i≠0. $\endgroup$ Commented Mar 6, 2023 at 22:09
  • 2
    $\begingroup$ I don't know think the example works out as claimed (as far as I understood it), if by $k[2]$ you mean $\mathcal{O}[2]$: locally, $\mathcal{O}[2]$ has $\pi_2$, but it's space of endomorphisms is discrete. $\endgroup$ Commented Mar 8, 2023 at 20:36
  • $\begingroup$ @AchimKrause: Thanks, I rewrote the example and cleaned up the notation. $\endgroup$ Commented Mar 8, 2023 at 21:09

You must log in to answer this question.

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