5

This is actually exercise 12.2.H of Vakil's notes. In the notes, a k-scheme is defined to be k-smooth of dimension d if there exists a affine open cover(every is of form $A=k[x_1,...,x_n]/(f_1,...,f_r)$) where the Jacobian matrix has corank d at all points. Then 12.2.H says it suffices to check this at all closed points.

The hint says the points satisfying the condition can be described as locus where the Jacobian matrix has corank d can be described in terms of vanishing and nonvanishing of determinants of certain explicit matrices. I guess here he means the minors. I also know if some property is open(if a point x has property P, then there exists an open neiborhood U s.t. every y in U has property P), then it suffices to check it at closed points. Then I am stucked, could some one help me? Thanks!

2 Answers2

5

As you correctly noted, it is about the vanishing and non-vanishing of minors. To be precise, let $A=k[X_1,\ldots,X_n]/(f_1,\ldots, f_r)$ be a finitely generated $k$-algebra and let $x \in \operatorname{Spec}k[X_1,\ldots, X_n]$ containing $f_1,\ldots,f_r$, and let $i \geq 0$ be a fixed integer. Then the Jacobian matrix at $x$ has rank $\leq i$ if and only if all $l \times l$ - minors vanish for all $l > i$, i.e. if and only if $x \in \operatorname{V}(I_i,f_1,\ldots f_r)$, where $I_i$ denotes the ideal generated by all $l \times l$ - minors with $l>i$.

It follows that the points $\operatorname{V}(I_d, f_1,\ldots, f_r) \cap \operatorname{Spec}k[X_1,\ldots,X_n]\setminus \operatorname{V}(I_{d-1},f_1,\ldots,f_r)$ are precisely the points at which the Jacobian matrix has rank $d$.

Now, since $A$ is a finitely generated $k$-algebra, a consequence of Hilbert's Nullstellensatz tells us that the nilradical is equal to the Jacobson radical. But, if the rank at each closed point is equal to $d$, then $I_d$ is contained in the intersection of all maximal ideals containing $(f_1,\ldots, f_r)$, whence $I_d$ is contained in every prime containing $(f_1,\ldots, f_r)$. This proves that the rank at each point has to be less than or equal to $d$.
For the converse inequality, just note that each prime is contained in some maximal ideal.

Plankton
  • 1,098
3

Let me give a proof more in the spirit of the OP's approach and KReiser's hint.

Suppose that the Jacobian $J$ of $X = \operatorname{Spec}{ k[x_1, \dots, x_n]/(f_1, \dots, f_r)}$ has corank $d$ at all closed points. Thus, $$ V( (n-d)\text{-minors of }J) = \emptyset. $$ This shows that the Jacobian of $X$ has corank $\leq d$ at all points but we still want to show $\geq d$.

In particular, we now know that the Jacobian has rank $\geq n-d$ for all points in $X$. Therefore, now the points with exactly corank $d$ are given by $D((n-d+1)\text{-minors of } J)$ as those indicate that some $n-d+1$ minor does not vanish. But a polynomial not vanishing is an open condition. Since $X$ is quasicompact, this can now be checked on closed points.

Qi Zhu
  • 7,999
  • I think you probably need something slightly stronger than $X$ quasi-compact in the last line - you really want to make this about the density of closed points, which isn't implied by quasi-compactness (think about the spectrum of a DVR, for instance). – KReiser May 04 '21 at 19:20
  • @KReiser Hm, I don't get it, can you elaborate? I use that the closure of every point contains a closed point and this uses compactness. This is basically also written on p. 154 of Vakil's book. – Qi Zhu May 04 '21 at 19:28
  • Ah, I was mistaken - you're correct. The perils of commenting on an empty stomach! – KReiser May 04 '21 at 19:32
  • @KReiser Ok, thanks for the comments. Enjoy your lunch/dinner! – Qi Zhu May 04 '21 at 19:34
  • Surely the points with rank $\leq n-d$ lie in $V((n-d+1)-\text{minors of} , J)$, rather than $D((n-d+1)-\text{minors of}, J)$? So you really need to use the above answer's reasoning for this last part. – Sunny Sood Aug 16 '23 at 10:41