0
$\begingroup$

I am trying to prove the following Bertini type theorem:

Given a non-constant morphism $f:X \rightarrow C$, where $X \subset \mathbb{P}^n$ is a smooth irreducible variety and $C$ is a smooth curve, then the set of hyperplanes $H \subset \mathbb{P}^n$ such that $f|_{X \cap H}$ is still non-constant, is zariski open in $(\mathbb{P}^n)^*=$ the set of hyperplanes in $\mathbb{P}^n$.

The statement seems natural and obvious but I'm stuck. Any suggestion?

$\endgroup$

2 Answers 2

1
$\begingroup$

First observe that we may assume that $X$ is projective: Let $\bar X$ be the closure of $X$ in $\mathbb P^n$. Since $f$ maps to a curve, it extends to $\bar X$, call that $\bar f$ (otherwise one could resolve the indeterminacies) and $X\cap H$ is dense in $\bar X\cap H$ for a general $H$, so if $\bar f$ is non-constant, so is $f$. Se we'll assume that $X$ is projective.

Next, notice that $f|_{X\cap H}$ is constant if and only if $X\cap H$ consists of only a fiber of $f$. However, $X\cap H$ is very ample. If $\dim X>1$, this means immediately that it has to intersect all fibers, so it cannot consist of a single fiber (in other words, in this case a fiber is not even ample). If $\dim X=1$, then the divisors corresponding to the fibers form a $1$-dimensional subset of the corresponding linear system, but unless $X=\mathbb P^1=\mathbb P^n$, that linear system has to be larger. Consequently a general $H$ will intersect $X$ in something different than a fiber, in other words $f|_{X\cap H}$ is not constant. (Of course, this also means that there is a trivial counter-example to your statement: If $X=\mathbb P^1=\mathbb P^n$, then $X\cap H$ is a single point for any $H$, so $f|_{X\cap H}$ is constant. However, the above shows that that's the only way it fails and obviously you would want to exclude that case anyway.)

$\endgroup$
10
  • $\begingroup$ And when $\dim X>1$, the argument shows that $f_{\vert X\cap H}$ is nonconstant for every $H$. $\endgroup$ Commented Feb 13, 2014 at 8:54
  • 1
    $\begingroup$ Yes, if $X$ is projective. Otherwise there may be a few hyperplanes that do not intersect $X$. I didn't want to bother with all the details... :) $\endgroup$ Commented Feb 13, 2014 at 17:59
  • $\begingroup$ Sándor: Pardon me for my ignorance. For $\dim X >1$, I don't understand why the fact that $X\cap H$ is very ample implies it has to intersect all fibers. My understanding of very ample divisors is that they induce a closed embedding to $\mathbb{P}^k$. I will appreciate if you can give me a reference in knowing more about very ample divisors. (I checked Hartshorne but didn't find it helpful on this.) Thank you. $\endgroup$
    – wongpin101
    Commented Feb 13, 2014 at 18:39
  • $\begingroup$ Actually for that you only need that $X\cap H$ is ample, but I wrote "very ample", because it is actually very ample. The reason is exactly what you are saying. If there were a fiber that $H$ did not intersect then the intersection cycle of $H$ and that fiber would be $0$. But then any other member of the linear system $|H|$ would have the same intersection cycle and hence all members would have to either contain or be disjoint from that fiber. If that happens, then the fiber would be mapped to a single point, so the morphism induced by $|H|$ would not be an embedding. $\endgroup$ Commented Feb 13, 2014 at 19:37
  • $\begingroup$ In other words the intersection of an ample divisor with any effective cycle is always positive. $\endgroup$ Commented Feb 13, 2014 at 19:39
2
$\begingroup$

$f |_{X \cap H}$ is nonconstant if and only if there are two points of $X \cap H$ which $f$ sends to different points of $C$. For a pair of points in $C$, being different is an open condition - it's just the complement of the diagonal in $C \times C$.

Consider the subvariety $(Y \in \mathbb P^n)^*$ which sends consists of the points in $\mathbb P^n$ lying in the hyperplane in $(\mathbb P^n)^*$. Let $X \cap Y$ be its intersection with $X$ in $\mathbb P^n$. Taking the fiber product of this with itself over $\mathbb P^n$, we obtain a family over $(\mathbb P^n)^*$ whose fiber over each point is the space of pairs of points in $X \cap H$. This maps to $C \times C$, and the pullback of the complement of the diagonal is an open subset which consisits of pairs of points which map to different kinds of $C$.

The desired locus is just the pushforward of this down to $(\mathbb P^n)^*$. We need to check that the pushforward of this open set is open, which means we need the morphism to be open. This is clear because it is flat of finite type. It's flat because it is the fiber product of two hypersurfaces in constant families.

However there is one caveat. To get $X \cap H$ flat, we need $H$ to never contain $X$ ( the zero-divisor condition.) If $H$ is contained in a hyperplane, you could have problems. For instance if $X = \mathbb P^1$, $C= \mathbb P^1$, $f$ the identity, $f|_{X\cap H}$ is constant unless $H$ contains $X$, which is a closed ondition.

$\endgroup$

You must log in to answer this question.

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