I think there must be a standard answer to this, for people in the know.

Let $X\subseteq\mathbb{A}^{n}$ be an affine (closed) variety of dimension $\geq d$, and fix some set $\{f_{i}\}_{i}$ of polynomials defining $X$. Let $\mathcal{G}=\mathrm{Gr}(n-d,\mathbb{A}^{n})$ be the Grassmannian of $(n-d)$-dimensional linear subspaces in $\mathbb{A}^{n}$. Since a generic $L\in\mathcal{G}$ is supposed to have non-empty intersection with $X$, my question is: is there a way to describe the closure of \begin{equation*} W=\{L\in\mathcal{G}|L\cap X=\emptyset\} \end{equation*} as a subvariety of $\mathcal{G}$ solely in terms of the $f_{i}$?

I would even settle for a description of any (closed) subvariety $W'\supseteq W$ inside $\mathcal{G}$, rather than $\overline{W}$ itself, as long as $\dim(W')<\dim(\mathcal{G})$.

  • 2
    $\begingroup$ Let $\overline{X}$ be the closure of $X$ in $\mathbb{P}^n$ and $Y$ be the intersection of $\overline{X}$ with the hyperplane $\mathbb{P}^{n-1}\cong\mathbb{P}^n-\mathbb{A}^n$. Further, let $G$ be the Grassmannian of $\mathbb{P}^{n-d}$'s in $\mathbb{P}^n$; $\mathcal{G}$ is an open subset of $G$. We have a closed subvariety $\tilde{W}$ of $G$ consisting of $M$ such that $M\cap\overline{X}\subset M\cap Y$. The variety $W$ should be its intersection with $\mathcal{G}$. $\endgroup$ – Kapil Apr 30 at 17:37
  • $\begingroup$ Thanks, I was also suggested in real life to think projectively: here we are saying that $M$ would intersect $X$ only at infinity, if I understand your idea correctly. But then I have the same standard question: why/how is $\tilde{W}$ itself a closed subvariety describable in terms of the $f_{i}$? $\endgroup$ – D. Dona May 2 at 12:47

$L \cap X = \emptyset$ is the same as saying that $\{f_i\} \cup \{\ell_j\}$ has no solution, where the $\ell_j$ are the linear functions defining $L$. By the Nullstellensatz, that's the same as saying there exist polynomials $h_i,g_j$ such that $\sum_i h_i f_i + \sum_j g_j \ell_j = 1$. By an Effective Nullstellensatz there is a bound (doubly-exponential, but still finite) on the degrees of the $h_i,g_j$ that can possibly be needed here. So you can treat the coefficients of the $h_i,g_j$ as variables, and then the preceding equation is a (really big) linear equation $Ax = b$, where the $x$ vector consists of the coefficients of $h_i,g_j$, there is one equation for each monomial up to the relevant degree bound, and the $b$ vector is $e_1$ (because of the 1 on the RHS of the above equation). The values appearing in $A$ are the coefficients of the $f_i,\ell_j$.

So $L \cap X = \emptyset$ precisely when this big linear equation has solutions, which is to say when $e_1$ is in the column span of $A$. The latter can be written in terms of various minors vanishing, where those minors depend only on the coefficients of the $f_i, \ell_j$. (1)

Now, if you treat the $\ell_j$ themselves as variables, each such minor gives you an equation in the $\ell_j$.

Next, one has to convert back from the description of $L$ via its equations $\ell_j$ to a point on the Grassmannian. Local coordinates on the Grassmannian $G(k,n)$ are essentially given by the entries of a $k \times n$ matrix (with each chart determined by requiring that a fixed $k \times k$ submatrix is full rank), where the corresponding subspace is the rowspan of this matrix. Let $\mathcal{L}$ be the matrix with columns being the $\ell_j$; then the matrix $M$ corresponding to the space $L$ is just the left kernel of $\mathcal{L}$. So we get the equations $M \mathcal{L} = 0$. (2)

Finally, since we want the $\ell_j$ themselves to be variables, we have the equations (1) and (2) with the $\ell_j$ as variable, and then project onto the coordinates of $M$, and this should be the variety you want. Note that the only part here that requires anything like Gröbner bases is this projection step - the rest is pretty straightforward from the coefficients of the $f_i$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.