2

Consider a cubic equation $ x^3 + 3a x^2 + 3 b x + c=0$ with distinct roots. Show that any two roots $x$, $y$ are connected by a homographic relation

$$(a^2-b) x y + \frac{1}{2}\ (\ (a b-c+\delta) x + ( a b -c-\delta) y\ ) + (b^2 - a c)=0 \ \ \ \ \ \ (*)$$

where $\delta = \pm\frac{\sqrt{\Delta}}{9}$ and $\Delta$ is the discriminant of the cubic.

Notes:

  1. This appears in Burnside and Panton -- The Theory of Equations, with the same title ( the notations changed a bit)

  2. Example: for the equation $x^3 - 4 x + x + 1=0$ with discriminant $13^2$ we get $ y= \frac{1}{1-x}$ or $y = \frac{x-1}{x}$

  3. For the case of a cubic equation

$$(x+w)^3 - \frac{3(u^3+v^3)}{u+v} (x+w) + (u^3+v^3)$$

with discriminant $\Delta = 9^2 (u-v)^2(u^2-u v + v^2)^2$ we get the transformations of the Galois group $\phi_{0,1,2}(x)$, with $\phi_0(x) = x$, \begin{eqnarray} \phi_{1}(x) = (u-w) + \frac{((v-w) - (u-w))^2}{v-w-x}\\ \phi_{2}(x) = (v-w) + \frac{((u-w) - (v-w))^2}{u-w-x} \end{eqnarray}

Note that $\phi_i \circ \phi_j = \phi_{i+j}$, where $i,j \in \mathbb{Z}/3$.

The case $w=0$ is even nicer.

  1. Consider a matrix $M_{p,q}= \left(\begin{matrix} -\frac{1}{2}-p & \frac{3/4 + p^2}{q}\\ q & -\frac{1}{2}+p\end{matrix} \right)$ with characteristic polynomial $t^2 + t + 1$. We have $M_{-p,-q} =M_{p,q}^2 = M_{p,q}^{-1}$. If we specialize $p = \frac{a b - c}{2 \delta}$, $q=\frac{a^2-b}{\delta}$ we get the transformations of the Galois group of the cubic $x^3 + 3 a x^2 + 3 b x + c$. These transformations form a cyclic group of order $3$ no matter what $\delta$ is. However, for the mappings obtained from the homographic relation $(*)$ we need the specific condition on $\delta$ to get order $3$.

  2. There are several question on this site (here and here) about finding the roots of a cubic with square discriminant in terms of one root. The expressions in polynomial form seems more involved. Moreover, every element of a cubic extension $K \subset K(x)$ is a fraction $\frac{\alpha x + \beta}{\gamma x + \delta}$, hence the problem finding such expressions for the other roots.

  3. This seems to be on the beaten path, so is posted as reference.

Any feedback would be appreciated!

$\bf{Added:}$ Let us note that for the equation

$(x+w)^3 - \frac{3(u^3 + v^3)}{u+v}(x+w) + (u^3 + v^3)$ the homographic transforms are given by the matrix

\begin{eqnarray} \left (\begin{matrix} \frac{w-u}{u-v} & \frac{ (u-v)^2+(v-w)^2+(w-u)^2}{2(u-v)}\\ -\frac{1}{u-v} & \frac{v-w}{u-v} \end{matrix} \right) \end{eqnarray}

Let us write $\alpha = u-v$, $\beta= v-w$, $\gamma = w-u$, with $\alpha + \beta + \gamma = 0$. So we have a matrix

\begin{eqnarray} \left (\begin{matrix} \frac{\gamma}{\alpha} & \frac{\alpha^2 + \beta^2 + \gamma^2}{2 \alpha} \\ -\frac{1}{\alpha} & \frac{\beta}{\alpha} \end{matrix} \right ) \end{eqnarray} with characteristic polynomial $t^2 + t+1$. With $\alpha$, $\beta$, $\gamma$ of sum $0$ we can parametrize all such matrices uniquely. So now we see a way to get all the cubics with Galois group containing a specific homographic transform of order $3$ ( maybe some problems here with lifting to SL2 here, I'll think about it later).

orangeskid
  • 56,630

1 Answers1

2

Multiplying by $18$ and rearranging, the equality to be proved is:

$$ 18(a^2-b) x y + 9(a b-c)(x+y) \pm 9\delta(x - y) + 18(b^2 - a c)=0 \tag{1} $$

Let the third root be $z$, then the elementary symmetric polynomials in $x,y$ can be written in terms of $z$ as:

$$ \begin{align} x+y = -3a-z \tag{2} \\ xy = -\frac{c}{z} \tag{3} \end{align} $$

Choosing the sign for $\pm \sqrt{\Delta}$ such that $\delta = \frac{1}{9}(x-y)(y-z)(z-x)\,$:

$$ \begin{align} -9\delta(x-y) &= (x-y)^2(z-x)(z-y) \\ &= \big((x+y)^2 - 4xy\big)(z^2 - (x+y)z + xy) \\ &= \big(z^2 + 6a z + 9a^2 + \frac{4c}{z}\big)\big(2z^2 + 3 a z - \frac{c}{z}\big) \\ &= \frac{1}{z^2}(z^3+6az^2+9a^2z+4c)(2z^3 + 3 a z^2 - c) \tag{4} \end{align} $$

$\big($Another way to read $(4)$ is as an expression for $\,|x-y|=\dfrac{\pm\,9\delta z}{2z^3 + 3 a z^2 - c}\,$ in terms of $\,z, \delta\,$.$\big)$

Substituting $(2),(3),(4)$ in $(1)\,$:

$$ - 18(a^2-b) \frac{c}{z} - 9(a b-c)(3a + z) - \frac{1}{z^2}(z^3+6az^2+9a^2z+4c)(2z^3+3az^2-c) + 18(b^2 - a c) = 0 $$

Clearing the denominators gives a sextic in $z$, which has the original cubic as a factor (verified by WA):

$$ \begin{align} &- 18(a^2-b) c z - 9(a b-c)(3a + z)z^2 - (z^3+6az^2+9a^2z+4c)(2z^3+3az^2-c) + 18(b^2 - a c)z^2 \\ &\;\;=\;\; -(z^3 + 3 a z^2 + 3 b z + c) \left(2 z^3 + 9 a z^2 + 3(3 a^2 - 2 b) z - 4 c\right) \\ &\;\;=\;\; 0 \end{align} $$

dxiv
  • 77,867
  • 1
    Great! So just a calculation using the values of coefficients in terms of roots $x$, $y$, $z$ does it, right? I am still thinking about the problem. For every cubic form in two variables there is a transformation of order $3$ that invariates the form. This since the form splits in a product of linear forms and the transform permutes cyclically the linear forms. Perhaps I should post this. All is old stuff, people like Cayley were quite familiar with it. – orangeskid Apr 24 '23 at 22:27
  • @orangeskid Thanks. That works for proving it, though not so well for finding it. It's fairly straightforward to get different relations between the sum, product and difference of two roots in terms of the coefficients and $\sqrt\Delta$. What's far less obvious, however, is how to get your nice and simple relation, at least not without some lucky guesses. So, yes, different approaches could provide more insight. – dxiv Apr 24 '23 at 22:41