12
$\begingroup$

It is well known that any smooth curve $C\in |\mathcal{O}_{\mathbf{P}^1\times\mathbf{P}^1}(2,2)| $ has geometric genus equal to 1, so its isomorphism class is determined by its $j$-invariant. Nevertheless we have that $\dim(\mathbf P(H^0(\mathcal{O}_{\mathbf{P}^1\times\mathbf{P}^1}(2,2))))=8$ and $\dim(\operatorname{Aut}(\mathbf P^1\times\mathbf P^1))=6$. So the dimension of the GIT quotient $$ \mathbf P(H^0(\mathcal{O}_{\mathbf{P}^1\times\mathbf{P}^1}(2,2)))//\operatorname{Aut}(\mathbf P^1\times\mathbf P^1) $$ is 2. If we assume that the base field has characteristic 0, then by Castelnuovo theorem the quotient above is a rational surface. So, at least for a dense subset, there should be two numbers $I_1,I_2$ parametrizing the orbits of the action of $\operatorname{Aut}(\mathbf P^1\times\mathbf P^1)$.

My questions are:

1) Are these numbers known (in terms of the coefficients of the defining polynomial of $C$)?

2) How are they related to the $j$-invariant?

Thanks in advance.

$\endgroup$
6
  • 7
    $\begingroup$ Geometrically we can see not just a $j$-invariant but also a divisor class on Jac$(C)$, coming from the difference of the degree-$2$ classes ${\mathcal O}(1,0)$ and ${\mathcal O}(0,1)$. Conversely, given a genus-$1$ curve $C$ and two degree-$2$ divisors $D_1,D_2$ you can recover the embedding $\iota: C \to {\bf P}^1 \times {\bf P}^1$ from the two degree-$2$ maps $f_i : C \to {\bf P}^1$ obtained from sections of $D_i$. $$ $$ Somebody else will probably post a description or reference to the invariant theory of $(2,2)$ forms before I have the chance to work it out or look it up. $\endgroup$ Commented May 7, 2020 at 16:03
  • 3
    $\begingroup$ I added the tag "classical invariant theory" which is the one most relevant to the question, I think. That's because what is asked is to figure out some explicit generators for the ring of invariants of bihomogeneous forms $F(x,y)$ of bidegree $(2,2)$ in $x=(x_1,x_2)$ and $y=(y_1,y_2)$ under the group $SL_2\times SL_2$. I suspect one should be able to construct three explicit invariants $J_1,J_2,J_3$ and take for $I_1$ and $I_2$ suitable ratios of powers of these invariants to make them absolute. $\endgroup$ Commented May 7, 2020 at 21:02
  • $\begingroup$ There is a single (up to scale) invariant $J_1$ in degree two, and $J_2$ in degree three. I still have to think about $J_3$. $\endgroup$ Commented May 7, 2020 at 21:11
  • 1
    $\begingroup$ The ring of invariants is ${\bf C}[J_2,J_3,J_4]$ where $J_k$ (each $k=2,3,4$) is a polynomial invariant of degree $k$. The Jacobian of $C$ is $y^2 = x (x-J_2)^2 - 4 J_4 x + J_3^2$, with a visible rational point $(x,y) = (0,J_3)$. This must be classical but it's more fun to work it out than to try to hunt it down in the literature. I'll post one approach soon. $\endgroup$ Commented May 7, 2020 at 22:21
  • 2
    $\begingroup$ @NoamD.Elkies: I guess it makes more sense to index the $J$'s by their degree...:) $\endgroup$ Commented May 7, 2020 at 22:40

1 Answer 1

11
$\begingroup$

[EDITED to exhibit $j$ as a rational function of $J_2,J_3,J_4$, and to fix various local errors etc.]

The action of ${\rm SL_2} \times {\rm SL_2}$ on the $9$-dimensional space of $(2,2)$ forms has a polynomial ring of invariants, with generators in degrees $2,3,4$. If we write a general $(2,2)$ form $P(x_1,x_2;y_1,y_2)$ as $(x_1^2, x_1 x_2, x_2^2) M_3 (y_1^2, y_1 y_2, y_2^2)^{\sf T}$ where $M_3$ is the $3 \times 3$ matrix $$ M_3 = \left( \begin{array}{ccc} a_{00} & a_{01} & a_{02} \cr a_{10} & a_{11} & a_{12} \cr a_{20} & a_{21} & a_{22} \end{array} \right) \; , $$ then $J_k$ ($k=2,3,4$) can be taken to be the $x^{4-k}$ coefficient of the characteristic polynomial $\chi^{\phantom.}_{M_4}$ of the $4 \times 4$ matrix $$ M_4 = \left( \begin{array}{cccc} \frac12 a_{11} & -a_{10} & -a_{01} & 2 a_{00} \cr a_{12} & -\frac12 a_{11} & -2 a_{02} & a_{01} \cr a_{21} & -2 a_{20} & -\frac12 a_{11} & a_{10} \cr 2 a_{22} & -a_{21} & -a_{12} & \frac12 a_{11} \end{array} \right) . $$ This matrix is characterized by the identity $$ P(x_1,x_2;y_1,y_2) = (z_{11},z_{12},z_{21},z_{22}) M_4 (z_{22},-z_{21},-z_{12},z_{11})^{\sf T} $$ where each $z_{ij} = x_i y_j$, together with the requirement that $M_4$ has trace zero and becomes symmetric when its columns are listed in reverse order and columns $2,3$ are multiplied by $-1$. The invariants of degree $2$ and $3$ can also be written as $$ J_2 = -\frac12 a_{11}^2 + 2(a_{01} a_{21} + a_{10} a_{12}) - 4 (a_{00} a_{22} + a_{20} a_{02}), \quad J_3 = -4 \det M_3; $$ of course $J_4 = \det M_4$. The (Jacobian of the) genus-$1$ curve $P=0$ is isomorphic with the elliptic curve $$ y^2 = x (x-J_2)^2 - 4 J_4 x + J_3^2. $$ In particular this lets us compute the $j$-invariant of this curve as a rational function of $J_2,J_3,J_4$: $$ j = \frac{256 (J_2^2 + 12 J_4)^3}{16 J_2^4 J_4 - 4 J_2^3 J_3^2 - 128 J_2^2 J_4^2 + 144 J_2 J_3^2 J_4 + 256 J_4^3 - 27 J_3^4} \, . $$

One way to obtain these results is as follows. First compute the Hilbert series of the invariant ring. We find that it is $1 / \bigl( (1-t^2) (1-t^3) (1-t^4) \bigr)$; this suggests a polynomial ring of invariants with generators of degrees $2,3,4$, and shows that if we find independent invariants $J_2,J_3,J_4$ of those degrees then ${\bf C}[J_2,J_3,J_4]$ is the full invariant ring.

Now use the basis $\{z_{ij}\}$ of the four-dimensional space, call it $Z$, of sections of ${\cal O}(1,1)$; it is well-known that $\{z_{ij}\}$ embeds ${\bf P}^1 \times {\bf P}^1$ into ${\bf P}^3$ as the quadric $z_{11} z_{22} = z_{12} z_{21}$, identifying ${\rm SL_2} \times {\rm SL_2}$ with the special orthogonal group ${\rm SO}(Q)$ where $Q$ is the quadratic form $z_{11} z_{22} - z_{12} z_{21}$. This identifies $P$ with some other quadratic form $\tilde P$ in the $z_{ij}$, determined uniquely modulo ${\bf C} Q$.

Now $Q$ is nondegenerate, so it identifies $Z$ with its dual $Z^*$, and thus identifies quadratic forms on $Z$ with self-adjoint maps $T: Z \to Z$, with $Q$ itself mapping to the identity map. It is known that generically ${\rm SO}(q)$ orbits of such $T$ are determined by their spectrum, and thus by the characteristic polynomial $\chi^{\phantom.}_T$. There is a unique translate $\tilde P + cQ$ of trace zero, represented by the above matrix $M_4$. Hence the coefficients $J_2,J_3,J_4$ of $\chi^{\phantom.}_{M_4}$ are invariant and independent, as claimed.

To identify the elliptic curve, write $C$ as a double cover of one of the ${\bf P}^1$'s by taking the discriminant of $P$ with respect to the other ${\bf P}^1$, and then use classical formulas for the Jacobian of a genus-$1$ curve $y^2 = {\rm quartic}$. The formulas, though not pretty to look at, are short enough to let us identify the coefficients with polynomials in $J_2,J_3,J_4$. The resulting curve has a rational point whose $x$-coordinate is a multiple of $J_2$; translating $x$ to put this point at $x=0$ yields the model $y^2 = x (x-J_2)^2 - 4 J_4 x + J_3^2$ exhibited above. The visible rational point $(x,y) = (0,J_3)$ ought to correspond to the difference between the divisors ${\mathcal O(0,1)}_C$ and ${\mathcal O(1,0)}_C$, but I haven't checked this.

$\endgroup$
6
  • $\begingroup$ Nice answer. Does my polarization argument in my comments above help at all with the $j$ invariant question? It seems to provide an embedding of the moduli space of elliptic curves inside the OP's moduli space, but I don't know if the two ways computing $j$ coincide on this locus. $\endgroup$ Commented May 8, 2020 at 13:48
  • $\begingroup$ I'm not sure. (In any case the natural map goes the other way, from the moduli of $(2,2)$ curves to moduli of elliptic curves.) I did use the invariant theory of binary quartics to write the elliptic curve in terms of $J_2,J_3,J_4$. The $j$-invariant can be computed from my answer -- e.g. ellinit([0,-2*J2,0,J2^2-4*J4,J3^2]).j in gp -- though inevitably it's not very pretty. $\endgroup$ Commented May 8, 2020 at 15:29
  • $\begingroup$ @AbdelmalekAbdesselam The composition of your map with Noam's is quadratic - it sends a binary quartic $f$ to $ (d^2f /dx_1^1) (d^2 f/dx_2^2) - (d^2 f/dx_1 dx_2)^2$, I think - so it shouldn't preserve the $j$ invariant. $\endgroup$
    – Will Sawin
    Commented May 9, 2020 at 19:07
  • $\begingroup$ @WillSawin: my map is linear it sends the binary quartic $f(x_1,x_2)$ to $D^2 f$ where $D$ is the differential operator $y_1\frac{\partial}{\partial x_1}+y_2\frac{\partial}{\partial x_2}$. The result is a biform in $x$ and $y$ of bidegree $(2,2)$. What you wrote sounds more like the Hessian of $f$ which is another binary quartic. Sorry if my notations were confusing. $\endgroup$ Commented May 9, 2020 at 19:12
  • $\begingroup$ @AbdelmalekAbdesselam Sorry, I should have been more clear. There are actually two maps from biforms of bidegree $(2,2)$ to binary quartics which preserve the $j$ invariant. These are given by taking the discriminant in one of the two variables. They were mentioned as the "natural map" by Noam. To see what your map does to the $j$ invariant, we can compose it with either of them. Doing this gives the Hessian. Thus your map preserves $j$ if and only if the Hessian does. (But it could still help somehow even if it does not preserve $j$.) $\endgroup$
    – Will Sawin
    Commented May 9, 2020 at 19:22

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .