In this question a variety means any subset of complex projective space $\mathbb CP^n$ that is the set of common zeroes of a set of homogeneous polynomials. Thus if $A,B\subset{\mathbb C}P^n$ are varieties so is $A\cup B$.

Suppose $V_k$ is a sequence of varieties that converge to a subset $V\subset{\mathbb C}P^n$ in the Hausdorff topology on closed subsets of ${\mathbb C}P^n$. (This means after choosing a Riemannian metric on ${\mathbb C}P^n$ that given $\epsilon>0$ there is $K$ so that if $k>K$ then $V\subset N_{\epsilon}(V_k)$ and $V_k\subset N_{\epsilon}(V)$ where $N_{\epsilon}(X)$ is an $\epsilon$-neighborhood of $X$).

Also suppose that there is $d$ such that each $V_k$ is the set of common zeroes of some polynomials of degree at most $d$.

Is it true that $V$ is always a variety ?

This is a theorem of Ed Bishop if the $V_k$ are pure-dimensional.


For each subspace $M$ of degree $d$ homogeneous polynomials let $V(M)$ be the common vanishing locus. Because Grassmannians are compact you can assume your $V_k= V(M_k)$s where the $M_k$s converge to a limit $M_{\infty}$. I claim that $V(M_k)$ converges in Hausdorff topology to $V(M_{\infty})$.

It is enough to see that, if $M, M'$ are close in the natural topology on subspaces, then $V(M)$ and $V(M')$ are close in the Hausdorff topology.

This amounts to the following assertion: if a basis of $M$ takes small values on $x \in \mathbb{C}P^n$, then $x$ must actually be very close $V(M)$. That's a standard thing - Lojasiewicz inequality - it's much easier in the algebraic context.

| cite | improve this answer | |
  • 1
    $\begingroup$ But homogeneous polynomials do not "take values" on points of projective space. $\endgroup$ – Mariano Suárez-Álvarez Jul 16 '15 at 2:45
  • $\begingroup$ No, they don't. I was too lazy to normalize things, but I am sure you can manage it in your head. $\endgroup$ – category_student Jul 16 '15 at 3:08
  • 5
    $\begingroup$ Ah. I am sorry. I thought you might be trying to be helpful. My mistake. $\endgroup$ – Mariano Suárez-Álvarez Jul 16 '15 at 3:11
  • 4
    $\begingroup$ category_student: Thanks for telling me about Lojasiewicz inequality. I am a topologist and do not know any algebraic geometry. However it is not true that M and M' close implies V(M) is close to V(M') e.g. $M=\langle x^2y,xy^2\rangle$ and $M_k=\langle x^2y,xy^2+ (x+y)^3k^{-1} \rangle$. Then $M_k\to M$ as $k\to\infty$ but $V(M)$ and $V(M_k)$ have different dimensions. The problem seems to be there is a constant in the Lojasiewicz inequality that does not vary continuously with the data. $\endgroup$ – Daryl Cooper Jul 16 '15 at 14:44

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.