Let $X$ be a quasiprojective variety, $Y$ a projective variety, and $f:X \rightarrow Y$ an open immersion. If $\mathcal{F}$ is a locally free coherent sheaf, what can be said about $f_\ast \mathcal{F}$? Is it coherent? Is it torsion free? Is it reflexive?

$\begingroup$ $F=\mathcal{O}_X$ is a counterexample for the first and the third question. $\endgroup$ – Martin Brandenburg Sep 14 '10 at 11:42

$\begingroup$ You also might find Section 1, of ``Generalized Divisors on Gorenstein Schemes'' a useful read. In particular Prop 1.11 and Thm 1.12. $\endgroup$ – Karl Schwede Sep 14 '10 at 15:34
About your new question:
Let $Y$ be a projective variety and let $X\subset Y$ be an open subset with complement the closed subset $S:=Y\setminus X$. Call $f:X\hookrightarrow Y$ the inclusion.
Let $\mathcal F$ be an algebraic coherent sheaf without torsion on $X$.
Theorem (SerreGrothendieck) Suppose that $Y$ is normal and that $S$ has codimension $\geq 2$. Then the sheaf $f_\ast \mathcal F$ is coherent.
Serre, Prolongement de faisceaux analytiques cohérents, Ann.Inst.Fourier 16 (1966), 363374

$\begingroup$ This is close to Torsten's answer, but Serre supposes that the sheaf is without torsion, rather than reflexive.The article also considers the analytic case. $\endgroup$ – Georges Elencwajg Sep 14 '10 at 14:04

$\begingroup$ In the context of locally free F this statement (replacing projective variety with integral scheme) follows immediately from algebraic Hartogs' lemma $\endgroup$ – Tomo Apr 28 '18 at 3:51

1
Dear Yemon,
a)The sheaf $f_\ast \mathcal{F}$ is not coherent in general since its stalk will not be finitely generated over the local ring of a point of $Y\setminus X$. For example take $P$ a point of $\mathbb P^1=Y$ and $X= \mathbb P^1 \setminus P=\mathbb A^1$. Then for $\mathcal F =\mathcal O_X$, you get $(f_\ast \mathcal{F})_P= Rat(Y)$
b) The direct image $f_\ast \mathcal{F}$ will be torsion free because an inductive limit of torsion free modules over a domain is torsion free ( I assume that variety means in particular integral scheme.)
c) I'm not sure reflexive is a reasonable concept for a noncoherent sheaf.

$\begingroup$ Thank you for your answer. My question was motivated by the fact that I would like to construct a reflexive coherent sheave $\mathcal{G}$ on $Y$ such that $\mathcal{G}_X = \mathcal{F}$. Is it possible? What if I suppose that $Y$ is normal and codim($Y\setminus X) \geq 2$? $\endgroup$ – Yemon Dai Sep 14 '10 at 12:50

$\begingroup$ This is different, you may always extend any coherent sheaf on $X$ to some coherent sheaf on $Y$ and then take its double dual. Under your supplementary conditions such an extension is equal to the direct image (which in particular is coherent and reflexive). $\endgroup$ – Torsten Ekedahl Sep 14 '10 at 12:56

$\begingroup$ Indeed, for the part on finding a coherent sheaf on $Y$ that restricts to $F$ on $X$, take a look at Hartshorne, chapter II, exercise 5.15 where this construction is done step by step. $\endgroup$ – Karl Schwede Sep 14 '10 at 15:16
By the way, assuming by varieties, you mean irreducible varieties, then for the second question, the answer is yes.
For the torsion freeness, suppose that $r \in H^0(U, O_X)$ kills some nonzero element $z \in H^0(U, f_* \mathcal{F}) = H^0(U \cap X, \mathcal{F})$. By restriction, $r$ is a nonzero element of $H^0(X \cap U, \mathcal{O}_Z)$. We still have $rz = 0$ even in this setting, and so by restricting to an affine cover of $X$, it still happens. This will contradict the torsionfreeness (and thus in particular the locallyfreeness) of $\mathcal{F}$.