Consider the figure of an ellipse below: enter image description here

where the coordinate pair $ (x_e , y_e) $ denote a point on the positive half of the ellipse, the red stars denote the foci of the ellipse (at $x = \pm c$) , and $d$ denotes a distance centered on the foci as shown.

If we draw two lines which connect the points $(-c - \frac{d}{2}, 0)$ and $(-c+\frac{d}{2},0)$ to the point $(x_e,y_e) $ these lines will form an angle $\theta$ between them.

The question is:

For a given d, for what pair $(x_e, y_e)$ does the angle $\theta$ maximize?

My work so far

I use the law of cosines to find $\theta$ :

$$ d^2 = s_1^2 + s_2^2 - 2 \cdot s_1 \cdot s_2 \cdot cos(\theta)$$


$$ s_1 = \sqrt{(x_e - (-c + \frac{d}{2}))^2 + y_e^2} $$ and
$$s_2 = \sqrt{(x_e - (-c - \frac{d}{2}))^2 + y_e^2}$$

so $$cos(\theta) = \frac{s_1^2 + s_2^2 - d^2}{2 \cdot s_1 \cdot s_2}$$

I get that $s_1^2 + s_2^2 -d^2$ becomes:

$$2\cdot \left((x_e + c)^2 + y_e^2 - (\frac{d}{2})^2 \right)$$

and the full expression becomes:

$$cos(\theta) = \frac{(x_e + c)^2 + y_e^2 - (\frac{d}{2})^2}{\sqrt{\left((x_e + c - \frac{d}{2})^2 + y_e^2\right)\left((x_e + c + \frac{d}{2})^2 + y_e^2\right)}}$$

because $y_e = b \cdot \sqrt{1 - (\frac{x_e}{a})^2} $ it means that (for a constant $d$) $cos(\theta)$ is a function of only $x_e$

so finally I got:

$$\theta(x_e) = \cos^{-1}\left(\frac{(x_e + c)^2 + (b^2 \cdot (1 - (\frac{x_e}{a})^2) - (\frac{d}{2})^2}{\sqrt{\left((x_e + c - \frac{d}{2})^2 + (b^2 \cdot (1 - (\frac{x_e}{a})^2)\right)\left((x_e + c + \frac{d}{2})^2 + (b^2 \cdot (1 - (\frac{x_e}{a})^2)\right)}}\right)$$

$x_e$ goes from $[-a,a]$ and I think that $d \le 2\cdot (a - c)$ so that $-c - \frac{d}{2}$ lies within [-a,a]

Going back to the question posed, I have two sub-questions:

1) Is this the correct equation for $\theta (x_e)$ ?

2) If this equation is correct, would I just have to set the derivative equal to $0$ to find the critical points (i.e. $x_e$ which gives the maximum $\theta$) ?

Extra plot:

I plotted the above function $\theta(x)$ (with the given constants for a, b, c, and d and $\theta(x)$ is given in degrees) in python, where the red vertical lines show the x values of the foci ($x = \pm c$), and I got this:

enter image description here

what prompted me to ask this question is that it seems the maximum inscribed angle occurs at an x position which is not the focus $ x = - c$, even though intuitively, I would have thought it would have been at that position. Did I just mess up in the derivation of $\theta(x)$ ?


No you did not mess up - although I have not checked your calculation.

The maximum angle $\theta$ results from two possibly competing factors :

  1. For a fixed direction of the point $E(x_e, y_e)$ from the focus $C$, angle $\theta$ is largest when the distance $EC$ is smallest.
  2. For a fixed distance $EC$, angle $\theta$ is largest when the direction of $E$ from $C$ is perpendicular to the axis on which $d$ lies.

The 2nd factor favours the point which you expected. However points to the left of that are closer so they are favoured by the 1st factor. A balance between these two factors determines where $\theta$ is maximum.

| cite | improve this answer | |

This problem has a simple geometric interpretation. The locus of points within the upper half-plane at which the segment of width $2d$ is subtended by an angle $\theta$, is a circular arc having the same endpoints as that segment, thus with its center on the perpendicular axis of the segment. The larger is the radius, the smaller is $\theta$. Hence the maximum value of $\theta$ corresponds to the smallest such circle intersecting the ellipse, with its center on the perpendicular axis of the segment (and in the upper half plane) and internally touching the ellipse.

A quick sketch made with GeoGebra, with $a=1$, $b=1/2$, $d=(2-\sqrt3)/4$, gives a maximum angle of about $38°$, corresponding to $x_e\approx-0.96$, which is in good agreement with your graph.

enter image description here


The above geometric setting can also lead to a quantitative result. If we parameterise point $E=(a\cos t,b\sin t)$ then a normal vector to the ellipse at $E$ is ${\mathbf n}=(b\cos t,a\sin t)$, and we can write: $C=E+s{\mathbf n}$.

Parameter $s$ can then be found from $x_C=-c$ and to find $t$ we can write the equation $CE^2=CD^2$, where $D=(-c-d,0)$ is an endpoint of the given segment. The resulting equation for $x=\cos t=x_E/a$ is quite simple: $$ e^2x^3-(1+2e^2-d^2/a^2)x-2e=0, $$ where $e=c/a$ is the ellipse eccentricity. Solving this with the given values ($a=1$, $d=(2-\sqrt3)/4$, $e=\sqrt3/2$) gives $x\approx -0.959788$ and $\theta=\arctan(d/y_C)\approx 38.0735$.


I realised only after writing my answer that I gave the given segment a width of $2d$, whereas at the beginning of the question its width is named $d$. And in computing numerical results I used a value for $d$ which is double of that given in the question.

| cite | improve this answer | |
  • $\begingroup$ @OscarLanzi I don't understand your remark above: as far as I know, $\theta=\arctan d/\sqrt{4r^2-d^2}$, which is monotonically decreasing with increasing values of $r$. $\endgroup$ – Intelligenti pauca May 1 at 12:50
  • $\begingroup$ That's true if $\theta$ is acute. But $\theta$ could be obtuse and then it's $\pi-\arctan(d/\sqrt{4r^2-d^2})$ instead. Both functions are monotonic for their respective domains, but in opposite directions. The same values of $d$ and $r$ will give both an acute angle by your formula and a supplementary obtuse angle by mine. $\endgroup$ – Oscar Lanzi May 1 at 13:01
  • 1
    $\begingroup$ I took for granted that the center of the circle lies in the upper half plane. I'll state that explicitly. $\endgroup$ – Intelligenti pauca May 1 at 13:06
  • $\begingroup$ So, why must the center be in the upper half-plane? You should explain that the point lies outside circle centered directly on the major axis, where $\theta=\pi/2$. That's what really makes $\theta$ acute in order to reach the ellipse. $\endgroup$ – Oscar Lanzi May 1 at 13:33
  • $\begingroup$ @OscarLanzi You are right, but that should be self-evident. Intelligenti pauca. $\endgroup$ – Intelligenti pauca May 1 at 13:35

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.