I found a theorem for cubic, the theorem is a generalization of the Carnot theorem for conic. I'm an electrical engineer, not a mathematician. I don't know how to prove this result.

Let ABC be a triangle, Let three points $A_1, A_2, A_3$ lie on $BC$, three points $B_1, B_2, B_3$ lie on $CA$, let three points $C_1, C_2, C_3$ lie on $AB$. There is a cubic passing through the nine points $A_1, A_2, A_3, B_1, B_2, B_3, C_1, C_2, C_3$ if only if:

$$\frac{\overline{A_1B}}{\overline{A_1C}}.\frac{\overline{A_2B}}{\overline{A_2C}}.\frac{\overline{A_3B}}{\overline{A_3C}}. \frac{\overline{B_1C}}{\overline{B_1A}}.\frac{\overline{B_2C}}{\overline{B_2A}}.\frac{\overline{B_3C}}{\overline{B_3A}}. \frac{\overline{C_1A}}{\overline{C_1B}}.\frac{\overline{C_2A}}{\overline{C_2B}}.\frac{\overline{C_3A}}{\overline{C_3B}}=1$$

enter image description here


2 Answers 2


Any nine points lie on a cubic; the nine points in the question already lie on a cubic: namely, the degenerate cubic consisting of the union of the triangle's edge-lines. What the question intends is that the nine points, under the stated condition, lie on a second cubic (and therefore, determine an pencil of cubics). That is, whereas we expect nine points in "general position" to determine a unique cubic, the condition indicates that these particular nine points are in "not-quite-general position".

This can be proven with the approach I used in this answer to prove Carnot's Theorem for Conics. Namely, we assign coordinates $$A := (x_A, y_A) \quad\quad B = (x_B, y_B) \quad\quad C := ( x_C, y_C )$$ and define $$ A^{(i)} = \frac{B + a_i C}{1 + a_i} \qquad B^{(i)} = \frac{C + b_i A}{1 + b_i} \qquad C^{(i)} = \frac{A + c_i B}{1 + c_i} \qquad $$ then, writing $P_x$ and $P_y$ for the coordinates of $P$, we use a determinant to express the equation of "the" cubic through our nine points:

$$\left|\begin{array}{ccccccccccc} x^3 & x^2 y & x y^2 & y^3 & x^2 & x y & y^2 & x & y & 1 \\ (A^{\prime}_x)^3 & (A^{\prime}_x)^2 A^{\prime}_y & A^{\prime}_x (A^{\prime}_y)^2 & (A^{\prime}_y)^3 & (A^{\prime}_x)^2 & A^{\prime}_x A^{\prime}_y & (A^{\prime}_y)^2 & A^{\prime}_x & A^{\prime}_y & 1 \\ (A^{\prime\prime}_x)^3 & (A^{\prime\prime}_x)^2 A^{\prime\prime}_y & A^{\prime\prime}_x (A^{\prime\prime}_y)^2 & (A^{\prime\prime}_y)^3 & (A^{\prime\prime}_x)^2 & A^{\prime\prime}_x A^{\prime\prime}_y & (A^{\prime\prime}_y)^2 & A^{\prime\prime}_x & A^{\prime\prime}_y & 1 \\ (A^{\prime\prime\prime}_x)^3 & (A^{\prime\prime\prime}_x)^2 A^{\prime\prime\prime}_y & A^{\prime\prime\prime}_x (A^{\prime\prime\prime}_y)^2 & (A^{\prime\prime\prime}_y)^3 & (A^{\prime\prime\prime}_x)^2 & A^{\prime\prime\prime}_x A^{\prime\prime\prime}_y & (A^{\prime\prime\prime}_y)^2 & A^{\prime\prime\prime}_x & A^{\prime\prime\prime}_y & 1 \\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots \\ (C^{\prime\prime\prime}_x)^3 & (C^{\prime\prime\prime}_x)^2 A^{\prime\prime\prime}_y & C^{\prime\prime\prime}_x (C^{\prime\prime\prime}_y)^2 & (C^{\prime\prime\prime}_y)^3 & (C^{\prime\prime\prime}_x)^2 & C^{\prime\prime\prime}_x C^{\prime\prime\prime}_y & (C^{\prime\prime\prime}_y)^2 & C^{\prime\prime\prime}_x & C^{\prime\prime\prime}_y & 1 \end{array}\right| = 0$$

A computer algebra system like Mathematica really helps to crunch the symbols, returning this:

$$\begin{align} 0 = &\phantom{\cdot}\;\left( x ( y_A - y_B ) - y ( x_A - x_B ) + x_A y_B - y_A x_B \right) \\ &\cdot\left( x ( y_B - y_C ) - y ( x_B - x_C ) + x_B y_C - y_B x_C \right) \\ &\cdot\left( x ( y_C - y_A ) - y ( x_C - x_A ) + x_C y_A - y_C x_A \right) \\ &\cdot|\triangle ABC|^7 \\ &\cdot(a_1 - a_2) (a_2 - a_3) (a_3 - a_1)\\ &\cdot(b_1 - b_2) (b_2 - b_3) (b_3 - b_1)\\ &\cdot(c_1 - c_2) (c_2 - c_3) (c_3 - c_1)\\ &\cdot(1 + a_1 a_2 a_3 b_1 b_2 b_3 c_1 c_2 c_3) \end{align}$$

(There must be a better way to derive this than by a brute-force calculation of a $10$-by-$10$ determinant. Of course, it helps to take convenient starting coordinates like $A=(0,0)$, $B = (x_B,0)$, $C = (x_C,y_C)$.)

The first three factors describe the triangle's edge-lines; their product represents the unique conic through our nine points, unless one of the other factors vanishes.

Of course, $|\triangle ABC| \neq 0$ for a non-degenerate triangle. Taking our points to be distinct guarantees that $a_i - a_j \neq 0$, and so forth. We may then draw the following conclusion:

Nine distinct points $A^{(i)}$, $B^{(i)}$, $C^{(i)}$ on the edge-lines of a non-degenerate triangle determine a pencil of cubics if and only if $$a_1 \; a_2 \; a_3 \; b_1 \; b_2 \; b_3 \; c_1 \; c_2 \; c_3 \;=\; -1 \tag{$\star$}$$ Equivalently, we can say that those nine points determine a unique cubic (the union of the triangle's edge-lines) if and only if $$a_1 \; a_2 \; a_3 \; b_1 \; b_2 \; b_3 \; c_1 \; c_2 \; c_3 \;\neq\; -1$$

To convert to the formula in the question, we note that the constants represent signed ratios of oriented segment lengths (in the grand Ceva tradition); a ratio is positive if the segments point in the same direction along a line, and negative if the segments point in opposite directions. Thus,

$$a_1 = \frac{|B A^{\prime}|}{|A^{\prime} C|} = - \frac{|A^{\prime} B|}{|A^{\prime}C|}$$

and so forth. An odd number of sign changes ensures that $(\star)$ matches the formula in the question. $\square$

  • $\begingroup$ Holy s********** ! And thanks to Mathematica for the brute force. $\endgroup$ Sep 19, 2015 at 2:08

Let $S$ be a hypersurface of degree $d$ of equation $\phi(x) = 0$ in some $k^n$, $a$ a point not on $S$. Consider a line $a + t v$ through $a$ that intersects $S$ in $n$ points given by the equation in $t$ $$\phi(a + t v ) = 0$$ Now expanding the LHS wr to the powers of $t$ we get $$\phi(a) + \cdots + \phi_d(v) t^d =0$$ (where $\phi_d$ is the $d$-homogenous part of $\phi$)). Therefore, the product of the roots of the above equation is ( by Viete) $$(-1)^d\frac{\phi(a)}{\phi_d(v)}$$

Therefore: if $a$, are $b$ points not on $S$ and the line $ab$ intersects $S$ in $d$ points $c_1$, $\ldots$, $c_d$ then $$\frac{ac_1 \cdot ac_2 \cdots ac_d}{b c_1\cdot bc_2 \cdots bc_d} = \frac{\phi(a)}{\phi(b)}$$

Consequence ( Carnot's theorem)

Let $a_1$, $\ldots$, $a_n$ points not on $S$. For every $i$ consider the intersection of the line $a_i$, $a_{i+1}$ with $S$ as $c_{i1}$, $\ldots$, $c_{id}$. Then we have

$$\prod_{i=1}^n \frac{a_i c_{i1} \cdot a_i c_{i2} \cdots \cdot a_i c_{id}}{a_{i+1} c_{i1}\cdot a_{i+1} c_{i2} \cdots \cdot a_{i+1} c_{id}}=1$$

Indeed, the product equals $$\prod_{i=1}^n \frac{\phi(a_i)}{\phi(a_{i+1})} = 1$$

Notes: the theorem about the product of chords determined by an algebraic curve is due to Newton ( it generalizes the power of a point)

The theorem of Carnot follows from the theorem of Newton, and generalizes results like Menelaus theorem.


You must log in to answer this question.

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