1

I am working in the following problem from my algebraic geometry course:

Let $X$ be a projective set of $\mathbb{P}^n$. Prove that there exists a finite regular morphism $F:X\to\mathbb{P}^k$, where $k=\dim X$.

(In case of doubt finite means that $k[X]$ is integral over $F^*(k[\mathbb{P}^k])$ and $F(X)$ is dense in $\mathbb{P}^n$, which is equivalent with $F^*$ injective)

Any help or hint will be appreciated, thanks :)

Edit: Here is my attempt:

If $X=\mathbb{P}^n$ we are done, since we can take $F$ to be the identity. If $X\neq\mathbb{P}^n$ there exist $x\in\mathbb{P}^n\setminus X$, so we can consider the map $f_1$ as the maps that projects $X$ from $x$. If $f_1(X)=\mathbb{P}^{n-1}$ we can take $F=f_1$. If $f_1(X)\neq\mathbb{P}^{n-1}$ there exist $y\in\mathbb{P}^{n-1}\setminus f_1(X)$, so we can consider the map $f_2$ as the maps that projects $f_1(X)$ from $y$. If $f_2(f_1(X))=\mathbb{P}^{n-2}$ we can take $F=f_2\circ f_1$,....

Since this process will end when we arrive $\mathbb{P}^k$ we can say that $F=f_{n-k}\circ\dots\circ f_{1}$. Of course this $F$ is well defined and it must be the solution to the problem, but I don't know how to prove that it is finite. If I can show that $f_i$ (a projection from a point) is a finite regular morphism we will be done, but I don't have ideas for this.

Edit: I know @KReiser said it could be a duplicate, but there the answer just work for characteristic $0$ and it gives us a 'geometric solution' that i don't like so much. I would like to have a pure algebraic solution and, if possible, an explicit formula for $F$ or a constructive method.

Marcos
  • 1,861
  • 1
    The map you describe is not defined at any point of $X$, all the coordinate functions vanish. You want to use Noether Normalization for this problem. – Smn Oct 31 '21 at 16:02
  • @Mentos Thanks, I added what I came up with, but I dont know how to finish from this. – Marcos Oct 31 '21 at 21:08
  • Potential duplicate of https://math.stackexchange.com/questions/679768/projective-noether-normalization – KReiser Oct 31 '21 at 21:15
  • @KReiser yes and no, it is the same result but I would like a more algebraic solution, not just a simply geometric interpretation, which is more intuitive, but not what I am looking for. – Marcos Oct 31 '21 at 23:21

1 Answers1

3

Here is the more algebraic solution you're asking for. First, we should require that our base field is infinite - we could bypass this requirement if we were working in the affine/non-graded case (see for instance a very nice proof by Mel Hochster here which works over any field and in fact generalizes to arbitrary domains), but the structure of the argument we're using here does require this.

Let $k$ be an infinite field and let $R$ be a graded $k$-algebra which is finitely generated by $x_0,\cdots,x_n$ in degree one. Then there are $y_0,\cdots,y_m\in R$ so that each $y_i$ is a $k$-linear combination of the $x_j$, the $y_i$ are independent, and $R$ is finite over $k[y_0,\cdots,y_m]$. We prove this by induction on $n$: the base case where $R$ is generated by $x_0$ is clear, as either $R$ is a polynomial ring in $x_0$ or $R$ is finite over $k$.

For the inductive step, let $\varphi:k[x_0,\cdots,x_n]\to R$ be the obvious graded surjection. If $\ker\varphi=0$, we're done. Else, let $P\in \ker\varphi$ be a nonzero homogeneous element of degree $d$. I claim that up to a change of variables of the form $x_i=x_i+\lambda_ix_n$ and $x_n=\lambda_nx_n$, we may assume $P=x_n^d+\cdots$: writing $P$ as a polynomial in $x_0$ and expanding out after this change of variables, we have $$P=P(\lambda_0,\cdots,\lambda_n)x_n^d+\cdots$$ where the $\cdots$ represents terms of degree less than $d$ in $x_n$. As any nonzero polynomial over an infinite field has a nonzero value, we can find $\lambda_i$ so that $P(\lambda_0,\cdots,\lambda_n)\neq 0$ and therefore $k[x_0,\cdots,x_n]/(P)$ is finite over $k[x_0,\cdots,x_{n-1}]$ and surjects on to $R$. Now consider the image of $\varphi(k[x_0,\cdots,x_{n-1}])$ in $R$: by the previous sentence, $R$ is finite over this, and by induction, we can find an appropriate selection of $y_i$ so that $\varphi(k[x_0,\cdots,x_{n-1}])$ is module-finite over $k[y_0,\cdots,y_m]$. As the composition of finite extensions is finite, we're done.

KReiser
  • 65,137
  • Maybe I am stupid or something, but I understand what you did but I dont get why this implies the result... – Marcos Nov 03 '21 at 23:45
  • 1
    This is the map on coordinate algebras corresponding to your projections. How familiar are you with the interplay between coordinate algebras and varieties? – KReiser Nov 03 '21 at 23:55
  • I know the relation, but what I dont see is why this map correspond to a projection. – Marcos Nov 04 '21 at 07:29
  • 2
    Let's look at $k[x_0,\cdots,x_{n-1}]\to k[x_0,\cdots,x_n]$ first. The preimage of the homogeneous prime ideal $(a_ix_j-a_jx_i){0\leq i\leq j\leq n}$ corresponding to the point $[a_0:\cdots:a_n]$ is $(a_ix_j-a_jx_i){0\leq i\leq j <n}$, which corresponds to the point $[a_0:\cdots:a_{n-1}]$. That's projection. The inclusion map $X\to\Bbb P^n$ is given on the level of rings by the surjection $k[x_0,\cdots,x_n]\to R$, and so the composite map is the projection of $X$ in to the $\Bbb P^{n-1}$ with $x_n=0$. – KReiser Nov 04 '21 at 07:35
  • Right, now I get It, thank you for your nice answer. – Marcos Nov 04 '21 at 07:47