13

I am looking for a proof of the easiest affine version of Chevalley's Theorem on constructible sets :

Theorem (Chevalley). The image of a constructible subset of $\mathbf C^n$ by a polynomial map $P:\mathbf C^n\rightarrow \mathbf C^m$ is constructible.

I am surprised that the model-theoretic version, prooving quantifier elimination of the first-order theory ${\rm Th}(\mathbf C)$ of $\mathbf C$ in the pure language of fields seems to involve almost no technicalities : a back-and-forth argument between to saturated models of ${\rm Th}(\mathbf C)$. Knowing what a minimal polynomial is, and maybe some elementary arithmetic on $\mathbf C[X]$ mostly suffices to understand the proof.

Question. Is there any elementary (non model-theoretic, not using schemes) proof of the above version of Chevalley's Theorem ?

Related MO questions:

Drike
  • 1,555
  • 3
    The proof in Harris' Algebraic Geometry, a first course, p. 40, is quite elementary. – abx Feb 25 '16 at 05:50
  • Thanks, I had a look, but I am afraid it is the projective version. – Drike Feb 25 '16 at 16:19
  • It implies the case you consider -- if a subset of $\Bbb{C}^m$ is constructible in $\Bbb{P}^m$, it is constructible. – abx Feb 25 '16 at 16:29
  • 2
    @Drike Can you define what you consider an "elementary" proof? All the algebro-geometric proofs I know use at least the concept of affine algebraic variety (by proving the more general statement that the image of any regular map between affine algebraic varieties is constructible). Are you trying to do without this? – Denis Nardin Feb 25 '16 at 16:31
  • @DenisNardin By elementary, I mean as elementary as possible, extracting precisely what about $\mathbf C$ makes the proof valid. For me an elementary proof is one every step of which can be explained ideally to a random old lady in the street, well if some time is allowed ! No I am not trying to do without the concept of affine algebraic variety. – Drike Feb 25 '16 at 17:05
  • 1
    Ok, so what abot the proof in the Stacks project? http://stacks.math.columbia.edu/tag/00FE. It will work exactly in the same way if $R,S$ are finite type algebras over $\mathbb{C}$ instad of general rings – Denis Nardin Feb 25 '16 at 17:15
  • 1
    I would say that the basic input in (one of the) proofs is Noether's normalisation lemma. I don't think that you can make it more elementary than that. – Damian Rössler Feb 26 '16 at 13:08

1 Answers1

2

$\DeclareMathOperator{\pp}{\mathbb{P}} \DeclareMathOperator{\kk}{\mathbb{K}} \DeclareMathOperator{\cc}{\mathbb{C}} \DeclareMathOperator{\ppp}{\mathfrak{p}}$ Having noticed this question 7 years after it was asked, my answer is that the proof from Mumford's "Algebraic Geometry I" is pretty elementary. The point is that

  • Chevalley's theorem (for an arbitrary algebraically closed field $\kk$ in place of $\cc$) is an "elementary" consequence of the following two facts:
    1. The one dimensional projective space $\pp^1$ is proper. In fact one needs the following fact (which is equivalent to properness): the projection map $\kk^m \times \pp^1 \to \kk^m$ is closed (in Zariski topology) for each $m$.
    2. The "basic" fact that the dimension of a proper (closed) subvariety of an irreducible variety $X$ is smaller than the dimension of $X$. (In "properness" I meant the property that
  • The properness of $\pp^n$ (for arbitrary $n \geq 1$) is an "elementary" consequence of the observation that a linear map $\phi:A \to B$ between vector spaces is surjective if and only if there is an $N \times N$ minor, where $N := \dim(B)$, of the matrix of $\phi$ with nonzero determinant.
  • The "basic fact" about dimension quoted above also has an "elementary" proof (if dimension of an irreducible variety is defined as the transcendence degree of its field of rational functions) - see e.g. the proof of Proposition 1.14 in "Algebraic Geometry I".

Below is a sketch of the first two points.

Proof of Chevalley's Theorem (assuming properness of $\pp^1$ plus the "basic fact" about dimension):

Step 1. Reduce the general statement of Chevalley's theorem to the following statement: let $\pi:\kk^{m+1} \to \kk^m$, $m \geq 0$, be the projection onto the first $m$-coordinates.

If $V$ is an irreducible subvariety of $\kk^{m+1}$ and $U$ is a nonempty open subset of $V$, then $\pi(U)$ contains a nonempty open subset of the closure of $\pi(V)$.

Step 2. Prove the above statement as follows: Let $W$ be the closure of $\pi(V)$ in $\kk^m$. Since $V$ is irreducible, $W$ is also irreducible. Moreover, $\kk[V] \cong \kk[W][x_{m+1}]/\ppp$ for some prime ideal $\ppp$ of $\kk[W][x_{m+1}]$. If $\ppp = 0$, then $V \cong W \times \kk$ and $\pi$ is the natural projection, and it is easy to see that the above statement is true. So assume $\ppp \neq 0$. Then since $\pi^*: \kk[W] \to \kk[V]$ is injective, it follows that $\dim(V) = \dim(W)$. Now consider the closure $\bar V$ of $V$ in $\kk^m \times \pp^1$. Then $\bar V$ is irreducible (since $V$ is), and $\pi(\bar V)$ is a closed (since $\pp^1$ is proper) irreducible subset of $\kk^m$ containing $W$. Since $\dim(W) = \dim(V) = \dim(\bar V) \geq \dim(\pi(\bar V))$, it follows that $$W = \pi(\bar V)$$ Let $V' := \bar V \setminus U$. Then $V'$ is a proper closed subset of $\bar V$, so that $\dim(V') < \dim(\bar V) = \dim(W)$. The properness of $\pp^1$ then implies that $\pi(V')$ is a proper closed subset of $W$. Since $\pi(U) \supset \pi(\bar V) \setminus \pi(V') \supset W \setminus \pi(V')$, the proof is complete.

Proof of properness of $\pp^n$: We prove the following statement (which is equivalent to properness of $\pp^n$):

The projection map $\kk^m \times \pp^n \to \kk^m$ is closed (in Zariski topology) for each $m,n$.

Let $Z$ be a closed subset of $\kk^m \times \pp^n$. Choose affine coordinates $(y_1, \ldots, y_m)$ on $\kk^m$ and homogeneous coordinates $[x_0: \cdots : x_n]$. Then $Z$ is the set of zeroes of polynomials $f_1, \ldots, f_k$ in $(x,y)$ which are homogeneous in $(x_0, \ldots, x_n)$. Given $b = (b_1, \ldots, b_m) \in \kk^m$, $b \not \in \pi(Z)$ if and only if there is $d \geq 1$ such that the vector space $P_d$ of homogeneous polynomials of degree $d$ is contained in the ideal of $\kk[x_0, \ldots, x_n]$ generated by $f_1(x,b), \ldots, f_k(x,b)$. In other words, $\kk^m \setminus \pi(Z) = \bigcup_{d \geq 1} U_d$, where $$U_d := \{b \in \kk^m: P_d \subseteq \langle f_1(x,b), \ldots, f_k(x,b) \rangle\}$$ (where $\langle \cdot, \cdots, \cdot \rangle$ denotes the ideal generated by the enclosed polynomials). Consequently, it suffices to show that $U_d$ is open for each $d$. Now let $d_j := \deg(f_j)$, $j = 1, \ldots, k$, and consider the map $$\phi_b:P_{d-d_1} \oplus \cdots \oplus P_{d-d_k} \to P_d$$ given by $$(g_1, \ldots, g_k) \mapsto \sum_j f_j(x,b)g_j$$ Then $b \in U_d$ if and only if $\phi_b$ is surjective if and only if there is an $m_d \times m_d$-minor of $\phi_b$ with nonzero determinant. Since each of these minors is a polynomial in $b$ (once you choose a fixed basis for each $P_e$), it follows that $U_d$ is Zariski open, as required.

pinaki
  • 5,064