0

This is a second follow-up to this question. For the ellipse centered at $(h,k)$ defined implicitly as

$$ \frac{(x-h)^2}{a^2}+\frac{(y-k)^2}{b^2}-1=0 \tag{1} $$

Please show the steps for calculating its dual curve and provide the equation for the dual curve. Do not solve by redefining the function parametrically. I know how to solve for parametric functions, but need to see the general algorithm for implicit functions.

Wikipedia provides a trivial example that does not help me.

My Attempt

At point $(p,q)$,

$$ X = \lambda \frac{2(p-h)}{a^2} \implies p = h + \left( \frac{a^2}{2\lambda} \right) X $$

$$ Y = \lambda \frac{2(q-k)}{b^2} \implies q = k + \left( \frac{b^2}{2\lambda} \right) Y $$

For incidence to occur, the inner product of the line $[X,Y]$ and point $(p,q)$ must be zero:

$$ \frac{a^2}{2\lambda}X^2 + hX + \frac{b^2}{2\lambda}Y^2 + kY = 0 $$

$$ a^2X^2 + b^2Y^2 + 2\lambda (hX + kY) = 0 \tag{2} $$

How should $\lambda$ be eliminated at this point?

Thought 1:

Can we homogenize the linear terms by $\frac{\left( hX + kY \right)}{-2\lambda}$? If so, we get

$$ a^2X^2 + b^2Y^2 - (hX+kY)^2 = 0 $$

$$ a^2X^2 + b^2Y^2 - (h^2X^2 + 2hkXY + k^2Y^2) = 0 $$

$$ (a^2-h^2)X^2 + (b^2-k^2)Y^2 - 2hkXY = 0 $$

but I'm not sure if this is the correct equation for the dual. Is there a way to check?

Thought 2:

Alternatively, can I just say that $\lambda = 1$? In this case, I get:

$$ a^2X^2+b^2Y^2+2(hX+kY)=0 $$

However, now my answer in Thought 1 is different from Thought 2.

  • First multiply through with $a^2b^2$ expand and homogenize the equation:

    $$(a^2k^2+b^2h^2-a^2b^2)z^2-2a^2kyz-2b^2hxz+a^2y^2+b^2x^2=0$$

    Then the system is

    $$\begin{align}X-\lambda (2b^2p-2b^2hr)&=0\ Y-\lambda (2a^2q-2a^2kr)&=0\ Z-\lambda (2(a^2k^2+b^2h^2-a^2b^2)r-2a^2kq-2b^2hp)&=0\ Xp+Yq+Zr&=0\end{align}$$

    – Jan-Magnus Økland Mar 15 '21 at 18:37
  • So I homogenized the wrong equation in Thought 1? I should have homogenized the original conic (1) rather than the dual form (2)? – adam.hendry Mar 15 '21 at 18:45
  • @Jan-MagnusØkland Thank you so much for everything! And I will definitely look into Macaulay2 and Grobner bases! – adam.hendry Mar 15 '21 at 18:49
  • 1
    @Jan-MagnusØkland So, just to be clear, the moral of the story is that you must homogenize the original equation before pursuing the lambda method, not the dual equation? – adam.hendry Mar 15 '21 at 19:42

3 Answers3

1

The dual conics are for $ax^2+bxy+cy^2+dx+ey+f=0$ given by $$ (-\frac14e^2+cf)x^2+(\frac12de-bf)xy+(-\frac14d^2+af)y^2+(-cd+\frac12be)x+(\frac12bd-ae)y-\frac14b^2+ac=0$$ or $\begin{pmatrix}x&y&1\end{pmatrix}\begin{pmatrix}a&\frac{b}{2}&\frac{d}{2}\\\frac{b}{2}&c&\frac{e}{2}\\\frac{d}{2}&\frac{e}{2}&f\end{pmatrix}\begin{pmatrix}x\\y\\1\end{pmatrix}$ to $$\begin{pmatrix}x&y&1\end{pmatrix}\begin{pmatrix}-\frac14e^2+cf&\frac{\frac12de-bf}{2}&\frac{-cd+\frac12be}{2}\\\frac{\frac12de-bf}{2}&-\frac14d^2+af&\frac{\frac12bd-ae}{2}\\\frac{-cd+\frac12be}{2}&\frac{\frac12bd-ae}{2}&-\frac14b^2+ac\end{pmatrix}\begin{pmatrix}x\\y\\1\end{pmatrix}$$

from which you can see it's to do with the matrix of a conic and it's adjoint matrix which is the inverse when divided by the determinant.

For an explicit derivation see Perspectives on Projective Geometry by Jürgen Richter-Gebert:

A conic consists of all points p that satisfy an equation $p^TAp = 0.$ The set of all tangents to this conic can be described as $\{Ap | p^TAp = 0\}.$ We can describe this set of homogeneous coordinates for lines directly as a quadratic form by the following observation:

$p^TAp = p^TAA^{−1}Ap = p^TA^TA^{−1}Ap = (Ap)^TA^{−1}(Ap).$

The right side of the equation explains how the set of tangent lines of a conic may be directly interpreted as the zero set of a quadratic form with matrix $A^{−1}$.

Edit:

The $\lambda$-method works in Macaulay2:

R=QQ[a,b,c,d,e,f]
S=R[l]
T=S[p,q,r,X,Y,Z,MonomialOrder=>Lex]
I=ideal(X-l*(2*a*p+b*q+d*r),Y-l*(b*p+2*q*c+e*r),Z-l*(d*p+e*q+2*f*r),X*p+Y*q+Z*r)
gens gb I -- matrix {{(e^2-4*c*f)*X^2+(-2*d*e+4*b*f)*X*Y+(4*c*d-2*b*e)*X*Z+(d^2-4*a*f)*Y^2+(-2*b*d+4*a*e)*Y*Z+(b^2-4*a*c)*Z^2, (2*c*d^2-2*b*d*e+2*a*e^2+2*b^2*f-8*a*c*f)*l*r+(-2*c*d+b*e)*X+(b*d-2*a*e)*Y+(-b^2+4*a*c)*Z,(e^2-4*c*f)*l*r*X+(-d*e+2*b*f)*l*r*Y+(2*c*d-b*e)*l*r*Z-e*X*Y+2*c*X*Z+d*Y^2-b*Y*Z, (d*e-2*b*f)*l*r*X+(-d^2+4*a*f)*l*r*Y+(b*d-2*a*e)*l*r*Z-e*X^2+d*X*Y+b*X*Z-2*a*Y*Z,(2*c*d-b*e)*l*r*X+(-b*d+2*a*e)*l*r*Y+(b^2-4*a*c)*l*r*Z-2*c*X^2+2*b*X*Y-2*a*Y^2, (2*c*d-b*e)*l*q+(d*e-2*b*f)*l*r-d*Y+b*Z, (b*d-2*a*e)*l*q+(d^2-4*a*f)*l*r-d*X+2*a*Z, (b^2-4*a*c)*l*q+(b*d-2*a*e)*l*r-b*X+2*a*Y,(2*c*d^2-2*b*d*e+2*a*e^2)*q*Y^2+(2*b^2*e-8*a*c*e)*q*Y*Z+(-2*b^2*c+8*a*c^2)*q*Z^2+(d*e^2-4*c*d*f)*r*X*Y+(-b*e^2+4*b*c*f)*r*X*Z+(-d^2*e+4*a*e*f)*r*Y^2+(4*c*d^2-2*a*e^2-8*a*c*f)*r*Y*Z+(-4*b*c*d+b^2*e+4*a*c*e)*r*Z^2,(2*c*d-b*e)*q*X+(-b*d+2*a*e)*q*Y+(b^2-4*a*c)*q*Z+(d*e-2*b*f)*r*X+(-d^2+4*a*f)*r*Y+(b*d-2*a*e)*r*Z, e*l*q*X-d*l*q*Y+2*f*l*r*X-d*l*r*Z-X*Z, 2*c*l*q*X-b*l*q*Y+e*l*r*X-b*l*r*Z-X*Y, b*l*q*X-2*a*l*q*Y+d*l*r*X-2*a*l*r*Z-X^2,e*q*X*Y-2*c*q*X*Z-d*q*Y^2+b*q*Y*Z+2*f*r*X*Y-e*r*X*Z-d*r*Y*Z+b*r*Z^2, e*q*X^2-d*q*X*Y-b*q*X*Z+2*a*q*Y*Z+2*f*r*X^2-2*d*r*X*Z+2*a*r*Z^2, 2*c*q*X^2-2*b*q*X*Y+2*a*q*Y^2+e*r*X^2-d*r*X*Y-b*r*X*Z+2*a*r*Y*Z, d*l*p+e*l*q+2*f*l*r-Z,b*l*p+2*c*l*q+e*l*r-Y, 2*a*l*p+b*l*q+d*l*r-X, 2*a*p*Z-e*q*X+d*q*Y+b*q*Z-2*f*r*X+2*d*r*Z, d*p*Y-b*p*Z+e*q*Y-2*c*q*Z+2*f*r*Y-e*r*Z, 2*a*p*Y-2*c*q*X+2*b*q*Y-e*r*X+d*r*Y+b*r*Z, p*X+q*Y+r*Z}}

$$(e^2-4cf)X^2+(-2de+4bf)XY+(4cd-2be)XZ+(d^2-4af)Y^2+(-2bd+4ae)YZ+(b^2-4ac)Z^2=0$$

It even gives you via

oo // (gens I)

the combination of the generators that give this relation:

$$(X-\lambda (2ap+bq+dr))((e^2-4cf)X+(-de+2bf)Y+(2cd-be)Z)\\+(Y-\lambda (bp+2qc+er))((-de+2bf)X+(d^2-4af)Y+(-bd+2ae)Z)\\+(Z-\lambda (dp+eq+2fr))((2cd-be)X+(-bd+2ae)Y+(b^2-4ac)Z)\\ + (pX + qY + rZ)((2cd^2-2bde+2ae^2+2b^2f-8acf)\lambda)$$

0

I see now. So the general quadratic form in the projective plane is

$$ \begin{pmatrix} x & y & z \end{pmatrix} \begin{pmatrix} a & b & d \\ b & c & e \\ d & e & f \end{pmatrix} \begin{pmatrix} x \\ y \\ z\end{pmatrix} = 0 $$

which yields

$$ ax^2+cy^2+fz^2+2bxy+2dxz+2eyz=0, $$

but when we restrict ourselves to the affine plane,

$$ \begin{pmatrix} x & y & 1 \end{pmatrix} \begin{pmatrix} a & b & d \\ b & c & e \\ d & e & f \end{pmatrix} \begin{pmatrix} x \\ y \\ 1\end{pmatrix} = 0 $$

becomes

$$ ax^2+cy^2+f+2bxy+2dx+2ey=0. $$

The adjoint (signed cofactor matrix) is denoted $A^\Delta$:

$$ \begin{pmatrix} a & b & d \\ b & c & e \\ d & e & f \end{pmatrix} ^ \Delta = \begin{pmatrix} + \begin{vmatrix} c & e \\ e & f \end{vmatrix} & - \begin{vmatrix} b & e \\ d & f \end{vmatrix} & + \begin{vmatrix} b & c \\ d & e \end{vmatrix} \\ - \begin{vmatrix} b & d \\ e & f \end{vmatrix} & + \begin{vmatrix} a & d \\ d & f \end{vmatrix} & - \begin{vmatrix} a & b \\ d & e \end{vmatrix} \\ + \begin{vmatrix} b & d \\ c & e \end{vmatrix}& - \begin{vmatrix} a & d \\ b & e \end{vmatrix} & + \begin{vmatrix} a & b \\ b & c \end{vmatrix} \end{pmatrix} $$

0

First multiply through with $a^2b^2$ expand and homogenize the equation:

$$(a^2k^2+b^2h^2-a^2b^2)z^2-2a^2kyz-2b^2hxz+a^2y^2+b^2x^2=0$$

Then the system is

$$\begin{align}X-\lambda (2b^2p-2b^2hr)&=0\\ Y-\lambda (2a^2q-2a^2kr)&=0\\ Z-\lambda (2(a^2k^2+b^2h^2-a^2b^2)r-2a^2kq-2b^2hp)&=0\\ Xp+Yq+Zr&=0\end{align}$$

Solving explicitly is feasible by hand (showing the Sylvester matrix of which the resultant is the determinant):

resultant(X-l*(2*b^2*p-2*b^2*h*r),Y-l*(2*a^2*q-2*a^2*k*r),p);

$$\begin{pmatrix} 2\,a^2\,k \,l\,r-2\,a^2\,l\,q+Y\\\end{pmatrix}$$

resultant(X-l*(2*b^2*p-2*b^2*h*r),Z-l*(2*(a^2*k^2+b^2*h^2-a^2*b^2)*r-2*a^2*k*q-2*b^2*h*p),p);

$$\begin{pmatrix} -2\,b^2\, l&2\,b^2\,h\,l\,r+X\\ 2\,b^2\,h\,l&\left( -2\,a^2\,k^2-2\,b^2\,h^2+2\,a^2\,b^2\right)\,l\,r+2\,a^2\,k\,l\,q+Z \\ \end{pmatrix}$$

resultant(Y-l*(2*a^2*q-2*a^2*k*r),Z-l*(2*(a^2*k^2+b^2*h^2-a^2*b^2)*r-2*a^2*k*q-2*b^2*h*p),p);

$$\begin{pmatrix} 2\,a^2\,k \,l\,r-2\,a^2\,l\,q+Y\\\end{pmatrix}$$

resultant(X-l*(2*b^2*p-2*b^2*h*r),X*p+Y*q+Z*r,p);

$$\begin{pmatrix}-2\,b^2\, l&2\,b^2\,h\,l\,r+X\\ X&Z\,r+Y\,q\\ \end{pmatrix}$$

resultant(Y-l*(2*a^2*q-2*a^2*k*r),X*p+Y*q+Z*r,p);

$$\begin{pmatrix} 2\,a^2\,k \,l\,r-2\,a^2\,l\,q+Y\\ \end{pmatrix}$$

resultant(Z-l*(2*(a^2*k^2+b^2*h^2-a^2*b^2)*r-2*a^2*k*q-2*b^2*h*p),X*p+Y*q+Z*r,p);

$$\begin{pmatrix} 2\,b^2\,h \,l&\left(-2\,a^2\,k^2-2\,b^2\,h^2+2\,a^2\,b^2\right)\, l\,r+2\,a^2\,k\,l\,q+Z\\ X&Z\,r+Y\,q\\ \end{pmatrix}$$

Among these choose two and take resultant w.r.t $q$

resultant((2*X*a^2*k^2+2*X*b^2*h^2+2*Z*b^2*h-2*X*a^2*b^2)*l*r+(2*Y*b^2*h-2*X*a^2*k)*l*q-X*Z,2*a^2*k*l*r-2*a^2*l*q+Y,q);

$$\begin{pmatrix}\left(2\, Y\,b^2\,h-2\,X\,a^2\,k\right)\,l&\left(2\,X\,a^2\,k^2+2 \,X\,b^2\,h^2+2\,Z\,b^2\,h-2\,X\,a^2\,b^2\right)\,l\,r-X\,Z\\ -2\,a ^2\,l&2\,a^2\,k\,l\,r+Y\\ \end{pmatrix}$$

resultant((-(2*X*b^2*h+2*Z*b^2)*l*r)-2*Y*b^2*l*q-X^2,2*a^2*k*l*r-2*a^2*l*q+Y,q);

$$\begin{pmatrix} -2\,Y\,b^ 2\,l&\left(-2\,X\,b^2\,h-2\,Z\,b^2\right)\,l\,r-X^2\cr -2\,a^2\,l&2\,a^2\,k\,l\,r+Y\\ \end{pmatrix}$$

Finally the resultant w.r.t $r$

resultant(2*l*((2*Y*a^2*b^2*h*k+2*X*a^2*b^2*h^2+2*Z*a^2*b^2*h-2*X*a^4*b^2)*l*r-X*Y*a^2*k+Y^2*b^2*h-X*Z*a^2),-2*l*((b^2*(2*Y*a^2*k+2*Z*a^2)+2*X*a^2*b^2*h)*l*r+Y^2*b^2+X^2*a^2),r);

$$\begin{pmatrix} \left(4\, Y\,a^2\,b^2\,h\,k+4\,X\,a^2\,b^2\,h^2+4\,Z\,a^2\,b^2\,h-4\,X\,a^4\,b ^2\right)\,l^2&\left(-2\,X\,Y\,a^2\,k+2\,Y^2\,b^2\,h-2 \,X\,Z\,a^2\right)\,l\cr \left(-4\,Y\,a^2\,b^2\,k-4\,X\,a^2\,b^2\,h- 4\,Z\,a^2\,b^2\right)\,l^2&\left(-2\,Y^2\,b^2-2\,X^2\,a ^2\right)\,l\\ \end{pmatrix}$$

factor(%);
-8*X*a^4*b^2*(Y^2*k^2+2*X*Y*h*k+2*Y*Z*k+X^2*h^2+2*X*Z*h-Y^2*b^2-X^2*a^2+Z^2)*l^3

i.e. $Y^2k^2+2XYhk+2YZk+X^2h^2+2XZh-Y^2b^2-X^2a^2+Z^2=0.$

or in one fell swoop using the maxima CAS function eliminate

eliminate([X-l*(2*b^2*p-2*b^2*h*r),Y-l*(2*a^2*q-2*a^2*k*r),Z-l*(2*(a^2*k^2+b^2*h^2-a^2*b^2)*r-2*a^2*k*q-2*b^2*h*p),X*p+Y*q+Z*r],[p,q,r]);
[-16*Y*a^2*b^6*(Y^2*k^2+2*X*Y*h*k+2*Y*Z*k+X^2*h^2+2*X*Z*h-Y^2*b^2-X^2*a^2+Z^2)*l^4]

Setting $Z=1$ we have the dual by the equation in the standard affine in the dual projective plane

$$(h^2-a^2)X^2+2hkXY+(k^2-b^2)Y^2+2hX+2kY+1=0.$$

  • Thank you! Why does the conic need to be homogenized for the lambda method to work, but not need to be homogenized for the matrix method to work? – adam.hendry Mar 17 '21 at 20:27
  • @A.Hendry: The coefficient matrix is the same whether you homogenize or not. Projective duality as its name suggests works on the homogeneous coordinate level. The point line correspondence of classical projective geometry extends to curves (and even varieties) with the dual curve parametrizing the tangents. The duality is global, but works in the open sets $z=1$ in the projective plane and corresponding open $Z=1$ in the projective dual plane because $xX+yY+zZ=0$ specializes to $xX+yY+1=0.$ – Jan-Magnus Økland Mar 17 '21 at 20:53
  • Ah yes, you are correct. And I had forgotten we set $z=1$ in the conic equation in the accepted answer, so it was in fact already homogeneous, just "disguised". – adam.hendry Mar 17 '21 at 20:57
  • I have a follow-up question. I'll create a new question for this. – adam.hendry Mar 18 '21 at 21:41
  • My follow-up question is here – adam.hendry Mar 18 '21 at 21:57