# Criteria for a map of schemes to be an isomorphism

Let $X, Y$ be separated finite type schemes over an algebraically closed field $k$. Assume that $Y$ is reduced. Let $\phi : X \rightarrow Y$ be a morphism of schemes. Suppose that $\phi$ gives a bijection on $k$-points and an injection on $S$-points for all $k$-schemes $S$. Prove or disprove that $\phi$ is an isomorphism (or add some extra hypotheses to ensure that $\phi$ is an isomorphism).

When $k = \mathbb{C}$, $X$ is normal, an $Y$ is normal and irreducible, then I have a proof which uses the following crazy fact: If $X$ and $Y$ are irreducible varieties over $\mathbb{C}$ and $Y$ is normal, then a morphism $\phi$ inducing a bijection on $\mathbb{C}$-points is an isomorphism. My original question follows from this fact via a small tweaking of the usual Yoneda argument.

If anyone can give me a proof or reference for this last fact, I would be grateful too. I read it in the appendix of Kumar's book on Kac-Moody groups.

Edit: In light of some counterexamples, let me assume that X is irreducible and Y is normal and irreducible.

-
Why can't $X$ be the disjoint union of the affine line and a point, and $Y$ be the projective line (and the map be the obvious thing) for a counterexample? – Kevin Buzzard Sep 6 '10 at 14:14
<TeX-pedant>It is usually much better to write `\$X\$, \$Y\$' than '\$X, Y\$', because of the resulting spacing and, in some cases, the fact that the math comma may be different from the text comma.</TeX-pedant> – Mariano Suárez-Alvarez Sep 6 '10 at 14:24
See mathoverflow.net/questions/12767 for a related question. – VA. Sep 6 '10 at 14:37
And also mathoverflow.net/questions/16786 – Frank Sep 6 '10 at 14:40

Trivial counterexample when $X$ is not connected: let $F$ be closed an nonempty in $Y$, $U:=Y\setminus F$ (assumed nonempty), and $X$ the disjoint sum of $U$ and $F$.

A bit less trivial with $X$ and $Y$ irreducible: $Y$= an irreducible curve with a node $y$, $X'$:=its normalization, $X$= $X'$ with one of the two points over $y$ removed.

The property holds, indeed, if $X$ is irreducible and reduced and $Y$ normal, assuming only that $X(\mathbb{C})\to Y(\mathbb{C})$ is bijective. In fact, in this case $f$ must be quasifinite and birational, hence an isomorphism if $Y$ is normal.

-

[This is a minor comment on Laurent Moret-Bailly's answer; I'm too new to leave comments]

That the map $f$ in Laurent Moret-Bailly's answer is birational follows from Proposition 7.16 in [J. Harris, Algebraic Geometry, A First Course, GTM 133, 1992]; that it is an isomorphism from Zariski's Main Theorem.

-