Let $X$ be a projective normal variety over $\mathbb C$, I have several questions about semi-stable sheaves:

Question 1. Suppose that $E$ is a pure sheaf such that $HN_*(E)$ is the Harder-Narasimhan filtration of $E$. Let $H$ be an ample divisor and $D \in |aH|$ be a general element for $a\gg 1$. Then is $HN_{*}(E)|_D$ the HN-filtration for $E|_D$?

(I guess the statement is false because we even cannot guarantee that $HN_{*}(E)|_D$ is a filtration (i.e. the restriction of inclusion may not be inclusion). But the statement might be true if $D$ is replaced by a general complete intersection curve.)

Question 2. If $E, F$ are (slope) stable sheaves, then is $E \otimes F$ still stable?

Question 3. Suppose that $0=E^0 \subset E^1\subset \cdots \subset E^k=E$ is a filtration of $E$ such that $G^i=E^i/E^{i-1}$ is semi-stable with slopes $\mu(G^i)$ strictly decreasing. Then is above filtration the NH-filtration?

(The answer may be no. But I was wondering if there is a non-constructive way to formulate NH-filtration? If there is a such why, please indicate how it goes.)

Thank you very much!

  • $\begingroup$ Please confer Mehta-Ramanathan. $\endgroup$ Commented Oct 14, 2022 at 18:01

1 Answer 1


Concerning your Questions 1 & 2, since you adress to slope semistable sheaves and Metha-Ramanathan type results, I think every sheaf you mean is in fact torsion free (i.e. pure of dimension $\dim X$), right? Under this assumption, Question 1 is true if $D$ is replaced by a general complete intersection curve and Question 2 is also true.

Your Question 3 is just the definition of the Harder-Narasimhan filtration (thus is true), c.f. [HL10, Definition 1.3.2, p.16].

Let me explain how to prove your Question 1 & 2:

  • For Question 1 (the version for general complete intersection curve), the point is that a general complete intersection curve $C$ is disjoint from the non-locally free loci (which is of codimension $\geqslant 2$ by torsion-freeness) of the factor sheaves of the HN-filtration, then the inculsions and factor sheaves are preserved under the restriction to $C$. Indeed, let $$E_0\subset E_1\subset\cdots \subset E_r=E$$ be the HN-filtration of $E$. Then for each $i$, since $E_i/E_{i-1}$ is locally free near $C$, the restriction to $C$ of the exact sequence $$0\to E_{i-1}\to E_i\to (E_i/E_{i-1})\to 0,$$ remains exact. The remaining thing is to use Metha-Ramanthan (c.f. [HL10, Theorem 7.2.1, p.197]) to show the slope semistability of each $(E_i/E_{i-1})|_C$. This result should also hold over algebraically closed fields of positive characteristic, since Metha-Ramanthan holds for arbitrary characteristic.
  • As for Question 2, again let $C$ be a general complete intersection curve, then $E|_C$ and $F|_C$ are both locally free and slope stable (by Metha-Ramanthan, c.f. [HL10, Theorem 7.2.8, p.202]). And so is $(E\otimes F)|_C\simeq E_C\otimes F|_C$, as a consequence of Narasimhan-Seshadri theorem (Ulenbeck-Yau in the curve case); there is also a purely algebraic proof for this fact.


[HL10] Daniel Huybrechts & Manfred Lehn: The Geometry of Moduli Spaces of Sheaves (2nd ed.), Cambridge: CUP, 2010,

  • $\begingroup$ Thank you Juanyong! $\endgroup$
    – Li Yutong
    Commented Apr 17, 2023 at 2:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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