31

It is well-known that there is only one "kind" of line, and that there are three "kinds" of quadratic curves (the nature of which depends on the sign of a so-called "discriminant").

It is noteworthy that many of the named cubic curves look rather similar: the folium of Descartes, the trisectrix of Maclaurin, the (right) strophoid, and the Tschirnhausen cubic look very similar in form; the semicubical parabola and the cissoid of Diocles resemble each other as well.

I have deliberately placed the word "kind" in quotes since there does not seem to me an intuitive way of defining the term, so an answer to my question might have to define "kind" rigorously in the context of cubic curves. (An algebraic invariant, for instance... it is a pity that there does not seem to be an analogue of "eccentricity" for cubics!)

In here, it is noted that Newton classified cubics into 72 "kinds", and Plücker after him described 219 "kinds".

So, how does one algebraically distinguish one cubic curve from another, and with a rigorous definition of "kind", how many cubics are there?

J. M. ain't a mathematician
  • 76,840

2 Answers2

23

The most standard notion of "kind" is that of isomorphism in algebraic geometry, that is, there are polynomial maps from one curve to the other which, when composed, are the identity on the curves (they don't have to be the identity in other places).

For real, affine plane curves, this notion reproduces the fact that there is a unique line and exactly three smooth conics (over the complex numbers, ellipses and hyperbolas turn out to be the same, and in the projective plane, all are the same).

However, regardless of ground field (well, regardless of working over the real or complex numbers), there are infinitely many different cubics up to isomorphism, determined by the j-invariant, one for every element of the field.

A conclusion you can draw from this is that there really isn't a good algebraic way to say that there are only finitely many types of plane cubic, unless you want to lump all smooth cubics together (which you clearly don't, as you consider parabolas and hyperbolas and ellipses to be distinct). I don't know what method Newton and Plücker used to classify them, but isomorphism is the standard method in modern algebraic geometry.

Charles Siegel
  • 2,197
  • 1
    As you can probably tell, algebraic geometry isn't my expertise... :( so you're effectively saying there's no "nice" algebraic invariant with a geometric interpretation (analogous to eccentricity for conics) that can be used to distinguish one cubic from another? – J. M. ain't a mathematician Dec 20 '10 at 03:53
  • 2
    I'm saying that I'm not aware of one that behaves like eccentricity...and that eccentricity is, in some sense, not an algebraic invariant, in that algebraically we can't really distinguish the various conics from each other in any decent way. There may be some other geometric invariant that will behave that way, but it won't, in some sense, be algebraic. – Charles Siegel Dec 20 '10 at 04:00
16

In John Stillwell's Mathematics and its History it is observed that Euler criticized Newton's classification for lacking a general principle, but that a closer examination of Newton/s work reveals one. In fact his work gives a general classification into 5 types, depicted on page 112 of that book.

I believe the difference in the real and complex cases, is the fact that the "discriminant" locus of singular cubics has real codimension one in the real case, hence separates the space of smooth cubics into distinct connected components, and this may be the types sought for. By Ehresmann's theorem, a connected family of compact manifolds have the same topological type. Thus curves on different connected components of the complement of the discriminant locus can have different homeomorphism type.

At least one (two?) of Newton's types is also singular and could represent the general (and special) point of the discriminant locus. I am not an expert.

roy smith
  • 1,572
  • Thank you very much. As can be noticed, I had accepted an answer in the interim, but I am still interested in hearing from people with both algebraic geometry and mathematical history expertise. – J. M. ain't a mathematician Dec 30 '10 at 02:17
  • I came here based on questions after reading that book -- I think of the five "types" that Newton mentioned, two are singular (one has a cusp, the other has a double point). The book makes it sound like a projective classification of the cubics over the reals, but as the accepted answer to this question and a related answer to another question https://math.stackexchange.com/a/2814084/327486 make clear, projective classification alone would lead to uncountably infinitely many types. However more carefully reading the claim from Stillwell's book, it seems to be that in modern terms what – Chill2Macht Nov 26 '25 at 03:28
  • Newton showed (well whether it was "shown" depends on one's standards of rigor) is that any plane cubic curve over the reals can be projectively transformed (over the reals) to one of the form $y^2 = ax^3 + bx^2 + cx + d$, which isn't necessarily inconsistent with the above since it corresponds to four real parameters. So I think your answer explains in modern terms (I didn't know about Ehresmann's theorem, this helped me) what Newton essentially "did next" to group those infinitely many projective equivalence classes, each with representatives of the form $y^2 = ax^3 + bx^2 + cx + d$, into – Chill2Macht Nov 26 '25 at 03:32
  • finitely many classes (i.e. based on the discriminant). One can notice that the form $y^2 = ax^3 + bx^2 + cx + d$ is basically the same as the "modern standard" "Weierstrass normal form", $Y^2 = AX^3 + BX + C$, using the "standard" trick to kill the quadratic term of a cubic (keyword: "depressed cubic", requires characteristic not equal to 3, which we have over the reals). If all this is true, of course it would be nice to show that the discriminant locus does divide the parameter space of $y^2 = ax^3 + bx^2 + cx + d$ cubics into three components over the reals (corresponding to Newton's three – Chill2Macht Nov 26 '25 at 03:40
  • nonsingular types of real cubic), and one connected component over the complex numbers (corresponding to cubic curves of genus one, i.e. nonsingular cubic curves). Admittedly I'm not sure whether the two singular real types of the form $y^2 = ax^3 + bx^2 + cx+d$ over the reals (i.e. the self-intersecting / double point one and the cusp one) are also "supposed" to collapse into a single case or not over the complex numbers. I found this question because I'm still confused about these issues and how to explain why the definition of elliptic curve is the correct definition of "interesting cubic". – Chill2Macht Nov 26 '25 at 03:43