5

I need to find the rational parametrization of the complex curve with affine equation $$y^2-6x^2y-3x^4+4x^3y+4x^3=0.$$

I did find a solution, I don't know if there is a more elegant solution, but this is what I did nonetheless.

I took the lines passing through the origin $y=kx$ and intersecting these I find $$x(k)=\frac{(4-6k)+(4i(k-1))^2\sqrt {k-1}}{2k^2}.$$

Now this is obviously not a rational function, but composing the function $k(t)=t^2+1$ rationalizes the square root and it should give a solution to the problem.Is there a better way to solve this?

Wastaken
  • 187
  • 2
  • 12

3 Answers3

4

Here's a different method using the anticanonical map. The method is very general, but the downside is that some of the intermediate computations are not easily done by hand. See by Rational Parametrizations of Algebraic Curves using a Canonical Divisor by van Hoeij for full algorithmic details.

The basic idea is that we find a birational map $\alpha: C \to Q$ where $Q$ is a plane conic. We are then reduced to finding a parametrization for a conic, which can be done by considering a pencil of lines through a given point. Once we have a parametrization $\varphi: \mathbb{P}^1 \to Q$, we can simply compose with $\alpha^{-1}$ to obtain a parametrization $\newcommand{\P}{\mathbb{P}} \alpha^{-1} \circ \varphi: \P^1 \to C$.

Let $C$ be a curve. In order for there to be rational parametrization for $C$ in the first place, then $C$ must have (geometric) genus $g=0$. In this case, a canonical divisor $K$ of $C$ has degree $$ \deg(K) = 2g - 2 = -2 \, , $$ so its negative, the anticanonical divisor $-K$, has degree $2$. Since $K < 0$, then $$ \require{cancel} \ell(-K) - \cancelto{0}{\ell(2K)} = 1 - g + \deg(-K) = 1 - 0 + 2 = 3 $$ by Riemann-Roch. Let $f_1, f_2, f_3$ be a basis for $L(-K)$. (For how to compute such a basis, see the paper cited above, or Computing Riemann-Roch spaces in algebraic function fields and related topics by Florian Hess.) Since $-K$ is very ample, then the induced map \begin{align*} \alpha: C &\hookrightarrow \P^2\\ P &\mapsto [f_1(P) : f_2(P) : f_3(P)] \end{align*} is a closed embedding. Moreover, since $\deg(-K) = 2$, then its image is a conic $Q$. If $Q$ has a rational point, we can then parametrize it as mentioned above; otherwise we may have to pass to a quadratic extension of the ground field in order to obtain a point.

I used the following Magma code to compute the canonical map for your curve. (You can try it yourself using the online Magma calculator.)

QQ := Rationals();
R<x,y> := PolynomialRing(QQ,2);
f := y^2 - 6*x^2*y - 3*x^4 + 4*x^3*y + 4*x^3;
C0 := Curve(AffineSpace(R),f);
C := ProjectiveClosure(C0);
P<X0,X1,X2> := Ambient(C);
K := CanonicalDivisor(C);
Support(K);
mp := DivisorMap(-K);
P<Y0,Y1,Y2> := Codomain(mp);
mp;

which returned

Mapping from: CrvPln: C to Prj: P
with equations :
X0^3*X2 - 2*X0^2*X2^2 + X0*X1*X2^2
2*X0^3*X2 - 3*X0^2*X2^2 + X1*X2^3
X0^4 - 2*X0^3*X2 + X0^2*X2^2

Running Q := Image(mp); calculates the conic $Q$, which in this case is $$ Q: Y_0^2 - Y_0 Y_1 + 4 Y_0 Y_2 - 2 Y_1 Y_2 = 0 \, . $$

Thus we have reduced the problem to parametrizing the conic $Q$. Consider the affine patch where $Y_2 \neq 0$ and let $u = Y_0/Y_2$ and $v = Y_1/Y_2$ on this patch. Then $Q$ is given by $$ u^2 - uv + 4u - 2v = 0 $$ on this patch. We see that the point $(0,0)$ lies on $Q$, so we consider the pencil of lines $v = tu$ through this point. Substituting this into the equation above yields \begin{align*} 0 = u^2 - t u^2 + 4u - 2tu = (1-t) u^2 + (4 - 2t) u = u ((1-t) u + 4 - 2t) \, . \end{align*} The root $u = 0$ corresponds to the origin, so we set the other factor equal to $0$ and solve for $u$ in terms of $t$: $$ 0 = (1-t) u + 4 - 2t \implies u = \frac{2t - 4}{1-t} \, . $$ This allows us to solve for $v$ in terms of $t$ as well: $$ v = t u = t \frac{2t - 4}{1-t} \, , $$ which yields a rational parametrization of the conic $Q$: \begin{align*} \psi: \mathbb{A}^1 &\to Q\\ t &\mapsto (u,v) = \left(\frac{2t - 4}{1-t}, t \frac{2t - 4}{1-t}\right) \, . \end{align*}

With some wrestling with Magma, we can find the inverse of $\alpha$:

KC<xx,yy> := FunctionField(C);
KQ<u,v> := FunctionField(Q);
mpQC := map< Q -> C | [Pushforward(mpCQ, xx), Pushforward(mpCQ, yy), 1]>;
mpQC;

Mapping from: CrvPln: Q to CrvPln: C with equations : -1/4Y0^2Y2^2 + 1/2Y0Y2^3 - 1/4Y2^4 1/16Y0^4 - 3/8Y0^2Y2^2 + 1/2Y0Y2^3 - 3/16Y2^4 Y0Y2^3

Composing these two maps yields the parametrization \begin{align*} \varphi: \mathbb{A}^1 &\to C\\ t &\mapsto \left( -\frac{1}{t(t-2)}, \frac{-2t + 1}{t^3 (t-2)} \right) \, . \end{align*}

You can see from the plots here that this parametrization indeed traces out the curve defined by your implicit equation.

Viktor Vaughn
  • 19,278
  • 1
    Nice. R=QQ[s,t,x,y,z,MonomialOrder=>Eliminate 2] I=ideal(x+t^2*s^2,y-(s-2*t)*s^3,z-t^3*(t-2*s)) gens gb I -- 3*x^4-4*x^3*y-4*x^3*z+6*x^2*y*z-y^2*z^2 – Jan-Magnus Økland Jun 08 '21 at 07:22
  • 1
    This is actually the way I ended up doing it before looking at your solution and I was about to share. Some of the theoretical results you mention are slightly out of hand for me, but what I did was basically transform the singular points to (1,0,0),(0,1,0),(0,0,1), and this was easy to compute, so then the α is just X=1/x,Y=1/y,Z=1/z and it becomes a conic, then I parametrize the conic and go back and undo all the maps. Our professor likes to make us dabble with these things but it's just a second year course so he introduces things in small bits, but I think the idea is the same as yours. – Wastaken Jun 08 '21 at 18:28
  • @SamueleMonitto That's one of the drawbacks of using a computer. The general algorithms will work, but they often don't give the nicest or smallest answer. Interesting, so there's some relationship with the Cremona involution. Please do post your answer (at least how you get the conic); I'd be interested to read it. – Viktor Vaughn Jun 08 '21 at 22:02
3

Using the above mentioned approach of adjoint curves in the pdf by V. Pillwein with the cusps $(0:1:0),(0:0:1),(1:1:1)$ and the simple point $(-4:32:5)$, found using the M2 package RationalPoints2

rationalPoints(ideal(y^2-6*x^2*y-3*x^4+4*x^3*y+4*x^3), Bound=>100) -- {{0, 0}, {4/3, 0}, {1, 1}, {-1/3, 1}, {9/5, -27/5}, {-4/5, 32/5},  {-1/3, -5/27}, {4/3, 32/27}, {-1/8, 5/32}, {9/8, 27/32}}

I get in maxima CAS

solve([a3,a5,a0+a1+a2+a3+a4+a5,(-4)*(-4)*a0+(-4)*32*a1+(-4)*5*a2+32*32*a3+32*5*a4+5*5*a5],[a0,a1,a2,a3,a4,a5]); # [[a0 = 3*%r1-5*%r2,a1 = %r1,a2 = 4*%r2-4*%r1,a3 = 0,a4 = %r2,a5 = 0]]

setting %r1 to $1$ and %r2 to $t$ and then

factor(resultant(y^2-6*x^2*y-3*x^4+4*x^3*y+4*x^3,(3-5*t)*x^2+x*y+4*(t-1)*x+t*y,y)); # (x-1)^2*x^2*(5*x+4)*(4*t*x-3*x+4*t^2-8*t+4)
solve(4*t*x-3*x+4*t^2-8*t+4,x); # [x = -(4*t^2-8*t+4)/(4*t-3)]
factor(resultant(y^2-6*x^2*y-3*x^4+4*x^3*y+4*x^3,(3-5*t)*x^2+x*y+4*(t-1)*x+t*y,x)); # (y-1)^2*y^2*(5*y-32)*(4*t*y-3*y-16*t^4+48*t^3-48*t^2+16*t)
solve(4*t*y-3*y-16*t^4+48*t^3-48*t^2+16*t,y); # [y = (16*t^4-48*t^3+48*t^2-16*t)/(4*t-3)]

$(x,y) = (-\frac{4(t-1)^2}{4t-3},\frac{16(t-1)^3t}{4t-3})$

Lastly, I check in M2 that the implicitization is the same again:

R=QQ[s,t,x,y,z,MonomialOrder=>Eliminate 2]
I=ideal(x+4*(t-s)^2*s^2,y-16*(t-s)^3*t,z-(4*t-3*s)*s^3)
gens gb I -- 3*x^4-4*x^3*y-4*x^3*z+6*x^2*y*z-y^2*z^2

and it is, when $z=1,$ the negative of your form.

  • Nice answer. +1 But I think your last point should be $(9: \color{red}{-}27: 5)$. Does using adjoint curves always yield a parametrization? I was under the impression that it only worked for curves with ordinary singularities, but it seems to work just fine on this example, which has cusps. – Viktor Vaughn Jun 08 '21 at 07:06
  • @ViktorVaughn: I actually used $(-4:32:5).$ – Jan-Magnus Økland Jun 08 '21 at 08:55
1

Let's work with the closure of the curve and with some variables swapped for practical purposes, the equation is $V(F): xz^2-6xy^2z-3y^4+4y^3z+4xy^3=0$ . The singularities are $(1:0:0),(0:0:1),(1:1:1)$. It can be proved that any quartic with 3 singularities in the points $(1:0:0),(0:1:0),(0:0:1)$ is of the form $g(xy,xz,yz)=0$ where g is the equation of a conic. First of all, we need a projectivity $ω$ to transform our singular points into the three points $(1:0:0),(0:1:0),(0:0:1)$ . One of the projectivities that does the job is the one associated to the matrix $\begin{pmatrix} 1 & -1 & 0 \\ 0 &-1 & 0 \\ 0 &-1 &1 \end{pmatrix}$ . To compute the image of the curve, we substitute $(x:y:z)=(x'-y':-y':z'-y')$ and eliminate the apexes, which gives us the equation $(xy)^2+(xz)^2+(yz)^2-2(xy)(xz)-2(xy)(yz)-2(xz)(yz)=0$ . This substitution is just the polynomial $F∘ω^{−1}$ , the idea is that, if $G=F∘ω^{-1}$ since $F(ω^{−1}(ω(P)))=F(P)$ then $G(ω(P))=0⟺F(P)=0$ , so $V(G)= ω(V(F))$ . Now we use the birational mapping $$φ:\mathbb{P}^{2} \to \mathbb{P}^{2}\\ (x:y:z) \mapsto (xy:xz:yz)=(\frac{1}{z}:\frac{1}{y}:\frac{1}{x})=(X:Y:Z)$$

All that is left to do is, parametrize the conic $X^2+Y^2+Z^2-2XY-2XZ-2YZ=0$ , let's call the rational parametrization $f'$, then $ω^{-1}\circ φ^{-1} \circ f'=f$ is the parametrization for the original curve.

Wastaken
  • 187
  • 2
  • 12
  • Very nice. Your method of finding a map to a conic reminds me of this post. There the problem is to show that the Cremona involution takes a family of conics to a family of lines. Your approach is like a higher degree version of that, taking a degree $4$ curve to a conic. – Viktor Vaughn Jun 10 '21 at 17:29