I am studying Joe Harris' Algebraic Geometry: A First Course, the section on Moduli Spaces, pg 278. I am stuck in a subtle point. Harris gives on p 279 an argument why there is no coarse moduli space of plane cubics and I not understand this argument:

Example 21.12. Plane Cubics
The fundamental example of a moduli space is one we encountered before in Example 10.16, that of plane cubics. Based on our previous discussion, we see that even a coarse moduli space does not exist for plane cubics. This is due to the various inclusions among closures of orbits of the action of $PGL_3 \ K$ ($K$ field with $char(K) \neq 2,3$)n on the space $\mathbb{P}^9$ of plane cubics. For example, if $\mathcal{M}$ is the set of isomorphism classes of plane cubics, $\mathcal{M}$ will have one point $p$ corresponding to irreducible plane cubics with a node, and another point $q$ corresponding to cuspidal cubics. But by what we saw in Example 10.16, the point $q$ would have to lie in the closure of the point $p$!

Explantions on terminology & references:

  1. the coarse moduli space is informally introduced at pges 278/279 as: Let $\{X_{\alpha}\}$ a collection of certain varieties (e.g. like genus $g$ curves, etc.). Then a variety $\mathcal{M}$ is called a coarse moduli space with respect this collection $\{X_{\alpha}\}$ if
  • as underlying set $\mathcal{M}$ is bijective to the set of isomorphism classes of $X_{\alpha}$.

  • for any reduced family $\pi: \mathcal{V} \to B$ (ie a flat surjection such that the general fiber is reduced) such that every fiber $\pi^{-1}(b) = X_b $ is a member of the collection $\{X_{\alpha}\}$ the canonical set-theoretic map

$$ \phi_{\pi}: B \to \mathcal{M} $$

given by sending each point $b \in B$ to the point of $\mathcal{M}$ representing the isomorphism class $[X_b]$ of the fiber $X_b$ over $b$ is a regular map.

  1. Above is also refered to Example 10.16. ( How $PGL_3 \ K$ Acts on $\mathbb{P}^9$) (page 121). It states that for base field $K$ with $char(K) \neq 2,3$ there exist a natural action of $PGL_3 \ K$ on the space of cubic polynomials on $\mathbb{P}^2$.

The interesting result was that this action has some interesting closure relationships among diverse orbits under this action. For example the orbit consisting of smooth cubics with $j$-invariant $j$ contains in its closure the locus of cuspidal cubics (i.e., the orbit of cubics projectively equivalent to $Y^2Z - X^3$).

This also implies that as stated above that the closure of the orbit of point $p$ corresponding to irreducible plane cubics with a node contains point $q$ corresponding to cuspidal cubic.

Question: Why this observation heuristically indicates that if $q \in \overline{ \{p \} } \subset \mathcal{M}$ then $\mathcal{M}$ cannot be a coarse moduli space of plane cubics? (and therefore a coarse moduli space of plane cubics not exist)

  • $\begingroup$ why the downvote? $\endgroup$ May 24, 2021 at 22:05
  • $\begingroup$ I have upvoted this stupid downvote. It happens sometimes on this site, without any reason in your case. $\endgroup$
    – Jean Marie
    May 24, 2021 at 23:03

1 Answer 1


The idea is that the $PGL_3 (K)$ action preserves isomorphism classes, but since $PGL$ is non-proper, the orbit closures contain limiting cases that are not of the same isomorphism class as the general fiber. So e.g. $y^2 = x^3 - tx$ is a family of nodal cubics over $\mathbb A^1$ with central fiber a cusp. Since all nodal cubics are abstractly isomorphic to $\mathbb P^1$ with $0$ and $\infty$ identified, the image of $\mathbb A^1$ induced by this family in $\mathcal M$ is a pair of isolated points ($0\mapsto [cusp]$ and $x \mapsto [node]$ for $x \in \mathbb A^1 \setminus 0$), showing that the map is not even continuous.

  • $\begingroup$ I think there is a subtle point here which should be clarified. We consider the space $\mathcal{M}$ as a variety, therefore it is endowed with Zariski topology which is not Hausdorff. Therefore we cannot separate arbitrary two points therefore and I'm not sure why it is immediately that your map $f: \mathbb{A}^1 \to \mathcal{M}, 0 \mapsto [cusp], x \neq 0 \mapsto [node]$ can't be continuous with resp underlying topologies. Clearly in analytic topology where continuity can be defined by sequences as $f: X \to Y$ is continuous $\endgroup$ May 25, 2021 at 23:21
  • $\begingroup$ iff for every converging sequence $x_n \to x$ the image sequence converge as well $f(x_n) \to f(x)$ your map is discontinuous. $\endgroup$ May 25, 2021 at 23:21
  • $\begingroup$ But here our topologies are coarser and we cannot separate $p=[node]$ and $q=[cusp]$ since we observed above that if $\mathcal{M}$ would be coarse moduli space then $q \in \overline{ \{p\} }$. Question: what are open sets in the image $f(\mathbb{A}^1)= \{ [node], [cusp] \} \subset \mathcal{M}$? In order to show that $f$ is discontinuous we have to show that $[cusp]$ is open in $f(\mathbb{A}^1)$, that is that there exist open $U \subset \mathcal{M}$ which contains $[cusp]$ but not $[node]$, but that's not obvious to me. $\endgroup$ May 25, 2021 at 23:21
  • 1
    $\begingroup$ Suppose $f$ is continuous. We know from point-set topology that the continuous image of a connected set is connected. But since the nodal cubic and cuspidal cubic are each represented by closed points of $\mathcal M$, the set $\left\{[node],[cusp]\right\}$ is disconnected. Since $\mathbb A^1$ is connected, this is a contradiction. $\endgroup$ May 26, 2021 at 1:57
  • $\begingroup$ If $\mathcal{M}$ would be our moduli space why $[node], [cusp]$ should be closed points of $\mathcal{M}$? That's exactly main reason for my confusion: a variety in general can of course have points (=primes) which are not closed. Why in this situation $[node], [cusp]$ you assume to be closed? $\endgroup$ May 26, 2021 at 23:19

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.