# The restriction of a global section which is not a zero divisor is still an non-zero divisor?

Let X be a scheme. U is an open subscheme of X. Assume f is a global section on X which is not a zero divisor, then the restriction of f to U is still an non-zero divisor?

If X is affine, the answer is obvious true. I don't know the answer for a general scheme.

This is a question raised in the definition of sheaf of total fraction rings. Some author claim U|-> total fraction ring of sections over U is a presheaf, but I can't see the reason.

-
The authors who claim that are wrong, even if Grothendieck is among them. See mathoverflow.net/questions/28553/… – Georges Elencwajg Aug 29 '11 at 8:01

Here's a counterexample.

Let $P=\mathbb{P}^1$, $X=\mathbb{A}^1$, and attach $X$ to $P$ along a single point $\{x\}$. Then there is a global section $f$ which is nonzero on $X$ except at $x$, and is identically zero on $P$. Moreover, $f$ restricts to a zero divisor on the open subvariety $X\cup P/\lbrace y\rbrace$, where $y$ is any point of $P$ other than $x$.

Now suppose $fg=0$. Then $g\equiv 0$ on $X/\lbrace x\rbrace$ implies $g(x)=0$. Since $g\mid_P$ vanishes at one point, $g\mid_P\equiv 0$, so $g= 0$. $\Box$

The correct definition of this presheaf is given in Hartshorne II.6. Basically, you only consider global sections which are not zero divisors in each local ring.

-
Very nice, Kevin! – Georges Elencwajg Aug 29 '11 at 8:11
Thanks! But excuse me， what is attach along a closed subscheme? – MZWang Aug 29 '11 at 12:34
@unknown I do not know. But the attaching along the point $x$ in the example is clear: Locally around the point $x$ it looks like the union of two axes in the affine plane. And away from the point it just looks like it looked before the attaching. If you wish you can look at the result as a push out, say as the coequalizer of a map $x\to \mathbb P^1$ and a map $x\to \mathbb A^1$. At the ring level this is locally given as a fibered product. – Wilberd van der Kallen Aug 29 '11 at 13:37
@unknown: What Wilberd said is correct. The subvariety $X\cup P/\lbrace y\rbrace$ will be affine with global sections isomorphic to $k[S,T]/\langle ST\rangle$. In general, it is possible to make sense of gluing along closed subschemes. See this question: mathoverflow.net/questions/64294/gluing-along-closed-subschemes. – Kevin Ventullo Aug 29 '11 at 18:09
No gluing is necessary. Consider $\mathbb P^2_k$ (with coordinates (u:v:w)). Kevin's scheme is the locally closed subscheme obtained by first taking the cross $uv=0$ (a closed subscheme of $\mathbb P^2_k$) and then deleting the closed point $(0:1:0)$ (say) from that cross. – Georges Elencwajg Aug 29 '11 at 19:40