MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Is the condition that a module is reflexive an open condition?

That is, if $X$ is a smooth projective complex variety, $T$ a quasi-projective variety, and $F$ a finitely presented module on $X \times T$ that is $T$-flat, then we can form the locus $T' \subset T$ of points $t$ such that the restriction of $F$ to $X \times t$ is reflexive.

Is $T' \subset T$ open?

If not, is it locally closed?

Recall that a coherent sheaf $F$ is said to be reflexive if the natural map $F \to (F^{\vee})^{\vee}$ to the double dual is an isomorphism.

share|cite|improve this question
up vote 9 down vote accepted

This locus is indeed open. I will explain why using Kollar's "Hulls and Husks" (arXiv:0805.0576). More generally, this article studies in great detail when taking the double dual commutes with base change.

First, we may restrict to the open locus of $T$ where $F_t$ is torsion-free (because reflexive sheaves are torsion-free). Then, we choose an ample line bundle $H$ on $X$, and we compute the Hilbert polynomials relatively to $H$. The Hilbert polynomial $P(F_t)$ of $F_t$ is constant by flatness. From the exact sequence $0\to F_t\to F^{\vee\vee}_t\to F^{\vee\vee}_t/F_t\to 0$, we see that $F_t$ is reflexive exactly when $P(F^{\vee\vee}_t)=P(F_t)$, i.e. exactly when $P(F^{\vee\vee}_t)$ takes its minimal value. But the polynomial $P(F^{\vee\vee}_t)$ is constructible and upper semicontinuous by Proposition 28 (3) of Hulls ans Husks. This proves that this locus is open.

share|cite|improve this answer
@Olivier Benoist: Thanks! – jlk Apr 2 '12 at 21:10
Olivier, how do you know that the polynomial $P(F_t^{\vee\vee})$ is constructible? The problem is that a priori $F_t^{\vee\vee}$ is just a collection of sheaves on the fibers, it is NOT given by restrictions of one coherent sheaf to the fibers... – Sasha Apr 3 '12 at 2:46
@Sasha : It is part of the statement of Kollar's article (arXiv:0805.0576 Proposition 28 (3)), and it is proven there. By noetherian induction, you only need to show that it is constant on an open subset of $T$, hence, you will only have to show that it is "given by restrictions of one coherent sheaf to the fibers" on an open subset of $T$. – Olivier Benoist Apr 3 '12 at 6:54

[EDIT: as Sasha points out, this does not answer the question. Please see it as a comment explaining why $X$ should be proper!]

The answer is no in general if $X$ is not proper: take $X=\mathrm{Spec}\,\mathbb{C}[x]$, $T=\mathrm{Spec}\,\mathbb{C}[t]$, and $F=$ the structure sheaf of $Z=\mathrm{Spec}\,(\mathbb{C}[t,x]/(1-tx))$. Then $T'$ is just the origin.

Variant: if instead you take $T=\mathrm{Spec}\,\mathbb{C}[t,u]$ and $Z=\mathrm{Spec}\,(\mathbb{C}[t,u,x]/(u,1-tx))$ (i.e. the same $Z$ as before, but embedded in 3-space), then $T'$ is the union of the origin and the complement of the $t$-axis, hence not locally closed (but still constructible).

Of course the point here is that $Z$ "goes to infinity" at the origin. I don't have a counterexample where $X$ is proper, but the main problem then is "taking the dual in the fibers", as in Sasha's comment above.

share|cite|improve this answer
In original question $X$ was proper (even projective). – Sasha Apr 2 '12 at 15:23
@Sasha: oops, sorry! I ha read "quasiprojective" but that was about $T$. – Laurent Moret-Bailly Apr 2 '12 at 17:02

I'll assume you meant to say that $F$ was locally finitely presented or coherent in your second sentence. The locus where $F$ is reflexive is the complement of the union of the supports of the kernel and cokernel of $F\to (F^\vee)^\vee$. This will be open.

Postscript As Sasha points out the argument is incomplete because it is not clear that duals commute with restriction to the fibres.

PPS Now Laurent has a counter example, so I guess that finishes it.

share|cite|improve this answer
Donu, the problem is that operation of taking the dual sheaf does not commute with restriction to the fiber over $T$. So, it is not clear why $((F^\vee)^\vee)_{|X\times\{t\}} \cong ((F_{|X\times\{t\}})^\vee)^\vee$. – Sasha Apr 2 '12 at 12:00
what happens if you take the derived dual? – Yosemite Sam Apr 2 '12 at 15:28
Well, in the bounded derived category of a smooth variety, everything is reflexive in the sense that $$F\cong RHom(RHom(F, O_X), O_X)$$ – Donu Arapura Apr 2 '12 at 16:08
ah, of course, silly me. it's an open condition then! :) – Yosemite Sam Apr 3 '12 at 23:54

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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