In this recent question (which now has an answer), Richard Thomas asked whether any projective $k$-scheme $X$ of (local) embedding dimension $d(X)$ can be embedded in a smooth $k$-scheme of dimension $d(X)$. If $i \colon X \hookrightarrow Y$ is such an embedding, then in particular we get a surjection $i^*\Omega_Y \twoheadrightarrow \Omega_X$. My (so far unsuccessful) strategy was to obstruct such a surjection from existing.

For a coherent sheaf $\mathscr F$, write $d_x(\mathscr F) = \dim_{\kappa(x)} \mathscr F_x \otimes_{\mathcal O_{X,x}} \kappa(x)$ and $$d(\mathscr F) = \max \left\{d_x(\mathscr F)\ |\ x \in X\right\}.$$

Question. If $X$ is a quasi-projective $k$-scheme, and $\mathscr F$ a coherent sheaf, does there exist a surjection $\mathscr E \twoheadrightarrow \mathscr F$ from a locally free sheaf of rank $d(\mathscr F)$?

Already if $X = \mathbf A^n$ this seems false to me; for example there should exist finite modules $M$ with $d(M) = 2$ that cannot be generated by $2$ elements (here I am using the Quillen–Suslin theorem that a finite projective module on $\mathbf A^n$ is free). But I don't know so many ways to prove that something is not generated by $2$ elements, except for a local obstruction $d_x(\mathscr F) > 2$.

I think it should be possible to give a negative answer to Thomas's question along these lines, by exhibiting a finite flat cover $\pi \colon X \to \mathbf A^n$ such that $\pi_*\Omega_X$ does not admit a surjection from a vector bundle of rank $\deg(\pi) \cdot d(\Omega_X)$. A great answer would incorporate something like this, but I would already be very happy with some global obstruction to surjecting from a vector bundle of a given rank.

  • 4
    $\begingroup$ You may want to look at the literature around Foster-Swan theorem. $\endgroup$ – Hailong Dao Jan 28 at 20:23

Let me explain a simple example.

Let $C \subset \mathbb{P}^3$ be a twisted cubic curve. It is a locally complete intersection of codimension 2, hence its ideal $I_C$ is locally generated by two sections. Let me show that there are no surjections $E \twoheadrightarrow I_C$ from a locally free sheaf $E$ of rank 2.

Indeed, assume such a surjection exists. Its kernel is a reflexive sheaf of rank 1, hence is a line bundle, so we have an exact sequence $$ 0 \to L \to E \to I_C \to 0. $$ Restricting to $C$ we obtain an exact sequence $$ 0 \to \det N^* \to L\vert_C \to E\vert_C \to N^* \to 0, $$ where $N^*$ is the conormal bundle. But $N^*$ is locally free of rank 2, hence the surjection $E\vert_C \to N^*$ is an isomorphism, hence the middle arrow is zero, hence $$ \det N^* \cong L\vert_C. $$ But the adjunction fromula shows that $\det N^* \cong \mathcal{O}_C(-10)$, and this line bundle does not restrict from $\mathbb{P}^3$ (because 10 is not divisible by 3). This contradiction proves that no surjection from $E$ as above exists.

Of course, the same argument works for many other lci of codimension 2.

  • $\begingroup$ Could you say a word about where $\det N^*$ comes from? (I can do a $\mathscr Tor$ computation locally, but the Koszul resolution doesn't globalise by the very statement you're trying to prove.) $\endgroup$ – R. van Dobben de Bruyn Jan 29 at 15:12
  • 1
    $\begingroup$ In fact, it does (sometimes this is called the fundamental local isomorphism): $Tor_i(\mathcal{O}_Z,\mathcal{O}_Z) \cong \wedge^iN^*$ for any lci scheme $Z$. $\endgroup$ – Sasha Jan 29 at 15:53
  • $\begingroup$ I don't mean that the resolution generalizes, but its consequence (the isomorphism for $Tor_i$) does. $\endgroup$ – Sasha Jan 29 at 16:10
  • 1
    $\begingroup$ See "Les K-groupes d'un schéma éclaté..." by Thomason (Inventiones 1993), Lemme 3.2 (the indexing is correct). $\endgroup$ – Riza Hawkeye Jan 29 at 21:03
  • 1
    $\begingroup$ Maybe it's worth mentioning that in many "practical" situations, by the Serre construction, the existence of an isomorphism $\text{det} N^*\cong L|_C$ is exactly what is needed to give a global resolution of $I_C$. $\endgroup$ – Yosemite Stan Jan 31 at 6:58

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.