Given any triangle $\triangle ABC$, we can build the parabola with directrix passing through the side $AB$ and focus in $C$. This curve intersects the other two sides in the points $D$ and $E$.

enter image description here

Similarly, we can build other two parabolas, one with directrix passing through $AC$ and focus in $B$ (red), and one with directrix passing through $BC$ and focus in $A$ (green), obtaining other $2$ couples of points $F,G$ and $H,I$.

enter image description here

My conjecture is that

The $6$ points $D,E,F,G,H,I$ always determine an ellipse.

enter image description here

How can I show this (likely obvious) result with a simple and compact proof?

Thanks for your help, and sorry for the trivial question!

This problem is related to this one.

  • 1
    $\begingroup$ $\overleftrightarrow{AB}$ meets the $A$-focused parabola at two points: $H$ and some $H^\prime$. Likewise, for each other side-line and the appropriate parabolas. The collection of six "other" points also appear to lie on a common conic, although that conic isn't always an ellipse. (It's possible that your ellipse isn't always an ellipse, either. I haven't checked.) $\endgroup$ – Blue Oct 17 '18 at 13:20
  • $\begingroup$ @Blue Thanks for your comment. I may have not fully understood your observation. The parabola with focus in $A$ has the directrix passing through $BC$ (the green one). How can it intersect the side $AB$ in two points? $\endgroup$ – user559615 Oct 17 '18 at 13:28
  • $\begingroup$ Extend side $\overline{AB}$ to a line. Since that line crosses the $A$-focused once, it must cross it again (ignoring the case where the line is parallel to the parabola's axis). $\endgroup$ – Blue Oct 17 '18 at 13:37
  • $\begingroup$ @Blue I think I understood what you mean. But the conjecture is restricted to its intersection with the side $\overline{AB}$. However, it looks a very interesting observation! $\endgroup$ – user559615 Oct 17 '18 at 13:39
  • 1
    $\begingroup$ That's okay. We'll say that my conjecture involves the other points. ;) The techniques used to solve one case will apply almost-immediately to the other, since constructions and algebra often ignore arbitrary restrictions. (For instance, a coordinate argument to find point $H$ involves solving a system involving a parabola equation and a line equation (not a segment equation). Such a system naturally admits two solutions —one corresponding to "your" point, one to "mine"— for which the algebra has no intrinsic preference.) So, we get two results for the price of one proof. $\endgroup$ – Blue Oct 17 '18 at 13:48

We'll take the following Ceva-like result as a given:

For $\triangle ABC$ with $D_B$ and $D_C$ on $\overleftrightarrow{BC}$, $E_C$ and $E_A$ on $\overleftrightarrow{CA}$, and $F_A$ and $F_B$ on $\overleftrightarrow{AB}$, those points lie on a common conic if and only if $$ \frac{BD_B}{D_BC}\cdot\frac{CE_C}{E_CA}\cdot\frac{AF_A}{F_AB} = \frac{CD_C}{D_CB}\cdot\frac{AE_A}{E_AC}\cdot\frac{BF_B}{F_BA} \tag{$\star$}$$

As is typical of Ceva-like results, the individual ratios in $(\star)$ are signed: a ratio is positive if the component directed segments point in the same direction; negative if they point in opposite directions.

Now, to the problem at hand ...

enter image description here

I've renamed points to match the statement above, where a subscript indicates the focus of the parabola through the point. Of course, if a side-line crosses a parabola, then it typically does so again (unless the line is parallel to the axis of the parabola). So, a line-parabola intersection typically consists of two points. For instance, there are two candidate positions where side-line $\overleftrightarrow{CA}$ meets the $C$-focused parabola; I've marked these $E_C^{+}$ and $E_C^{-}$, as the superscripted sign indicates whether the point is on the same side of $C$ as point $A$.

Now, from $E_C^{\pm}$, drop a perpendicular to $C^{\pm}$ on $\overleftrightarrow{AB}$, the directrix of the $C$-focused parabola. By definition of a parabola, $$C^{\pm} E_C^{\pm} = CE_C^\pm \tag{1}$$ Noting that, as an unsigned ratio, $$\frac{C^\pm E_C^\pm}{E_C^\pm A} = \sin A \tag{2}$$ we have the signed ratio $$\frac{CE_C^\pm}{E_C^\pm A} = \pm\sin A \tag{3}$$ Likewise, $$ \frac{AF_A^\pm}{F_A^\pm B}=\pm\sin B \qquad \frac{BD_B^\pm}{D_B^\pm C}=\pm\sin C \tag{4}$$ $$\frac{CD_C^\pm}{D_C^\pm B}=\pm \sin B \qquad \frac{AE_A^\pm}{E_A^\pm C}=\pm\sin C \qquad \frac{BF_B^\pm}{F_B^\pm A}=\pm\sin A$$

Clearly, these give that the ratios on the left- and right-hand sides of $(\star)$ match in absolute value (namely, $\sin A\sin B\sin C$); we make them match completely by choosing appropriate signs. There are $26$ ways to do this, hence $26$ common conics.

In particular, there is a common conic through $D_B^{+}$, $E_C^{+}$, $F_A^{+}$, $D_C^{+}$, $E_A^{+}$, $F_B^{+}$, as per OP's conjecture, and there is a common conic through $D_B^{-}$, $E_C^{-}$, $F_A^{-}$, $D_C^{-}$, $E_A^{-}$, $F_B^{-}$, as per the conjecture I suggested in a comment. $\square$

Note. This analysis doesn't show specifically that OP's conjectured conic is always specifically an ellipse. ("My" conjectured conic varies in nature. I haven't checked the other $24$.) I'll have to come back to that.

  • $\begingroup$ Very nice. Concerning "my" conjecture, I still think it is always an ellipse! I thought there could be a more elementary proof. Great, in any case! And thanks!!! $\endgroup$ – user559615 Oct 17 '18 at 19:22
  • $\begingroup$ What do you think about this math.stackexchange.com/q/2958984/559615 ? $\endgroup$ – user559615 Oct 17 '18 at 19:24

Here is a proof using barycentric coordinates. For non-degenerate triangle $ABC$, we are going to show the six points $D, E, F, G, H, I$ lies on an ellipse.


  • $a = |BC|, b = |CA|, c = |AB|$. WOLOG, we will assume $a \ge b, c$.
  • $R$ and $\Delta$ be the circumradius and area of $\triangle ABC$.
  • $\alpha = \frac{\sin A}{1 + \sin A} = \frac{a}{2R+a}$, $\beta = \frac{\sin B}{1 + \sin B} = \frac{b}{2R+b}$ and $\gamma = \frac{\sin C}{1 + \sin C} = \frac{c}{2R+c}$

For any point $P$ and line $\ell$, let $(u_P, v_P, w_P)$ be its barycentric coordinates with respect to $\triangle ABC$ and $d_\ell(P)$ be the distance between $P$ and line $\ell$.

Consider point $D$. Since it lies on $AC$, $(u_D, v_D, w_D)$ has the form $(u, 0, 1-u)$. Since $D$ lies on the parabola having $C$ as focus and $AB$ as directrix, we have

$$bu = |CD| = d_{AB}(D) = d_{AB}(C)(1-u) = \frac{2\Delta}{c}(1-u) = b\sin A(1-u)$$

This implies $u = \frac{\sin A}{1 + \sin A}$ and hence $D$ lies on the line $u = \alpha$ in barycentric coordinates.

By a similar argument, we have $G$ lies on line $u = \alpha$, $E$, $H$ lie on line $v = \beta$ and $F$, $I$ lie on line $w = \gamma$. As a result, the six points $D, E, F, G, H, I$ lies on the cubic

$$(u - \alpha)(v - \beta)(w - \gamma) = 0$$ in barycentric coordiantes. These six points belong to another cubic $uvw = 0$ which corresponds to the three lines supporting the triangle. So they belong to the "difference"

$$\mathcal{Q}(u,v,w) \stackrel{def}{=} uvw - (u-\alpha)(v-\beta)(w-\gamma) = 0$$ When we expand it, the cubic term cancel out and we are left with the equation of a conic $$(\alpha vw + \beta uw + \gamma uv) - (u\beta\gamma + v\alpha\gamma + w\alpha\beta) + \alpha\beta\gamma = 0\tag{*1}$$ To see what sort of conic is this, substitute $w = 1 - u - v$ into LHS and expand, we get

$${\rm LHS} = -\alpha v^2 + (\gamma - \alpha - \beta ) uv - \beta u^2 + ( \text{ linear/constant terms in } u, v )$$

Notice under affine transform, ellipse get mapped into ellipse. The conic above is an ellipse in Cartesian coordinates if and only if it is an ellipse in barycentric coordinates. In terms of coefficients in equation $(*1)$, we need $$\begin{align} & 4\alpha\beta - (\gamma - \alpha - \beta)^2 \stackrel{?}{>} 0\\ \iff & 2(\alpha\beta + \beta\gamma + \gamma\alpha) - \alpha^2 - \beta^2 - \gamma^2 \stackrel{?}{>} 0\\ \iff & \alpha(\beta + \gamma - \alpha) + \beta(\alpha+\gamma-\beta) + \gamma(\alpha+\beta-\gamma) \stackrel{?}{>} 0 \end{align} $$ Since $a \ge b, c$, we have $\alpha \ge \beta, \gamma$. The $2^{nd}$ and $3^{th}$ terms are already non-negative. Let's look at the $1^{st}$ term, we find

$$\begin{align}\beta + \gamma - \alpha &= \frac{b}{2R+b} + \frac{c}{2R+c} - \frac{a}{2R+a}\\ &= \left(\frac{b}{2R+b} - \frac{b}{2R+a}\right) + \left(\frac{c}{2R+c} - \frac{c}{2R+a}\right) + \frac{b+c-a}{2R+a}\\ &= \underbrace{\frac{b(a-b)}{(2R+a)(2R+b)}}_{\ge 0} + \underbrace{\frac{c(a-c)}{(2R+a)(2R+c)}}_{\ge 0} + \underbrace{\frac{b+c-a}{2R+a}}_{> 0, \text{ by triangle inequality}}\\ &> 0 \end{align} $$ Combine all these, we can deduce $4\alpha\beta - (\gamma - \alpha - \beta)^2 > 0$. As a result, the conic in $(*1)$ is an ellipse in barycentric coordinates and hence one in Cartesian coordinates.

A Side Note

Let $J$ be the center of above ellipse. I was originally drawn to this question by another question (currently deleted and OP decided to leave math.SE). Aside from labeling of the points, it is a conjecture about the areas of three triangles:

$$\verb/Area/(IHJ) = \verb/Area/(GFJ) = \verb/Area/(EDJ)\tag{*2}$$

In barycentric coordinates, we have

$$D = ( \alpha, 0, 1-\alpha), E = (0,\beta,1-\beta), F = (0,1-\gamma,\gamma),\\ G = ( \alpha,1-\alpha,0), H = (1-\beta,\beta,0), I = (1-\gamma,0,\gamma)$$ For any $P = (u,v,w)$, it is not hard to show

$$\begin{align} \verb/Area/(IHP) &= \Delta (v w - (v - \beta)(w - \gamma)) = \Delta \frac{\partial}{\partial u}\mathcal{Q}(u,v,w)\\ \verb/Area/(GFP) &= \Delta( u w - (u - \alpha)(w - \gamma)) = \Delta \frac{\partial}{\partial v}\mathcal{Q}(u,v,w)\\ \verb/Area/(EDP) &= \Delta( u v - (u - \alpha)(v - \beta)) = \Delta \frac{\partial}{\partial w}\mathcal{Q}(u,v,w) \end{align} $$ The conjecture about areas is equivalent to $J = (u_J,v_J,w_J)$ is a solution to the set of equations: $$ \frac{\partial}{\partial u}\mathcal{Q}(u,v,w) = \frac{\partial}{\partial v}\mathcal{Q}(u,v,w) = \frac{\partial}{\partial w}\mathcal{Q}(u,v,w)\tag{*3}$$

In Cartesian coordinates $(x,y)$, any conic can be viewed as the zero-set of a quadratic polynomial $\mathcal{P}(x,y)$. For circle/ellipse/hyperbola, the symmetric center is the unique point which satisfies

$$\frac{\partial}{\partial x}\mathcal{P}(x,y) = \frac{\partial}{\partial y}\mathcal{P}(x,y) = 0$$

Translate this to barycentric coordinates, the symmetric center of conic $\mathcal{Q}(u,v,w) = 0$ is the unique point which satisfies $(*3)$. This means $J$ is a solution of $(*3)$ and hence conjecture $(*2)$ is true.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy