6
$\begingroup$

I apologize for the vague question title, but I can't think of a better subject line.

Let $f : X\rightarrow S$ be a morphism of connected schemes. Let $x,y\in X$, $s\in S$ be geometric points with $f(x) = f(y) = s$.

Let $FEt_S$ be the category of finite etale schemes over $S$, and similarly with $X$. We have fiber functors $F_s : FEt_S\rightarrow Sets$, and $F_x,F_y : FEt_X\rightarrow Sets$, and $\pi_1(X,x),\pi_1(X,y),\pi_1(S,s)$ are just the automorphism groups of the associated fiber functors.

There are canonical isomorphisms $$\beta_x : F_x\circ f^*\stackrel{\sim}{\longrightarrow}F_s\qquad\text{and}\qquad \beta_y : F_y\circ f^* \stackrel{\sim}{\longrightarrow}F_s$$

Any isomorphism $\alpha : F_x\stackrel{\sim}{\longrightarrow} F_y$ induces an automorphism $$\overline{\alpha} := \beta_y\circ\alpha\circ\beta_x^{-1}\in Aut(F_s) = \pi_1(S,s)$$

Suppose the induced map $f_* : \pi_1(X,x)\rightarrow \pi_1(S,s)$ is trivial (sends everything to the identity), and similarly $f_* : \pi_1(X,y)\rightarrow\pi_1(S,s)$, then I believe I can prove that $\overline{\alpha}$ does not depend the choice of $\alpha$. However, I'm having a lot of trouble arguing that $\overline{\alpha}$ should always be $id\in Aut(F_s)$.

Is this even true? (I mean, surely it is right. "What else could it be?") How would one prove something like this?

EDIT 1: The specific case I'm interested in is where $X\rightarrow S$ actually factors through the geometric point $s$.

EDIT 2: Okay it seems that this additional assumption makes the statement pretty easy to prove, though I would be interested in a counterexample to the original question.

$\endgroup$
6
$\begingroup$

I don't think that $\bar\alpha$ is always the identity. You should think of $\alpha$ as a ``path'' from $x$ to $y$, so in principle it is possible that the loop obtained as the image of $\alpha$ is not nullhomotopic, even if the image of every loop is. In fact, if the map $f$ is the universal covering space you can get every possible element in $\pi_1(S,s)$ by choosing appropriately $x$ and $y$. To support this let me try to get you an example where $\bar \alpha$ is not in the image of $f_*$ (although $f_*$ is not trivial).

Fix the ground field $\mathbb{C}$. Take $f:\mathbb{A}^1\smallsetminus \{0\}\to \mathbb{A}^1\smallsetminus \{0\}$ be the square map, and take $x=1$ and $y=-1$. Let the isomorphism $\alpha: F_x\xrightarrow{\sim} F_y$ to be obtained by "sliding" the points of the fiber along the half unit circle in the positive semiplane. Then $\bar\alpha$ should be the generator of $\pi_1(\mathbb{A}^1\smallsetminus\{0\})=\hat{\mathbb{Z}}$, while the image of $f_*$ is the index 2 subgroup.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ So the case that I'm specifically interested in is where the map $X\rightarrow S$ factors through a geometric point, in which case topologically speaking any path in $X$ maps to the constant loop. I'll edit the question to reflect this. $\endgroup$ – Will Chen Feb 28 '16 at 4:31
  • $\begingroup$ @oxeimon Well, in that case $\bar\alpha$ is in the image of $\pi_1(\{s\},s)=*$, so there's not much to prove :). My problem with writing a counterexample is that I don't know any maps such that $f_*$ is trivial, I believe that almost any of them is a counterexample. $\endgroup$ – Denis Nardin Feb 28 '16 at 12:53

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.