A few eccentric

notes on conic
sections

Jon Cooper
c. 2021 cc0

Background image from Cyclopædia, or an Universal Dictionary of Arts and Sciences (1728)
Ed. Ephraim Chambers. Vol. 1, p. 304. From Wikipedia, the free encyclopedia.

0
Introduction

A picture (like the one below adapted from this free textbook [1]) is worth a thousand words.

My interest in this subject was triggered by this excellent article [2] which, although a bit too
long and complicated for me to follow, gave me the idea of trying to derive the equations of
the conic sections from just that – plane sections of a cone! The fairly simple approach here
[3] suggested it might work!

The proofs

Start with a right-circular double conic (i.e. the central axis is perpendicular to the base of
each cone). Assuming that a plane which crosses both halves of the conic, is a hyperbola,
x2 y2
proof of the equation − =1 is easiest if we make the intersecting plane parallel with the
a2 b 2
cone axis, as shown below.

1
By similar triangles we can see that the circle at the base of the cone has radius xb/a. From
the right-angled triangle drawn in this circle we get:

b 2
b2 + y 2=( x )
a
y x2
2
∴ 1+ 2 = 2
b a
x y2
2
− =1
a 2 b2

The angle between the asymptotes of the hyperbola is the angle at the tip of the cone, i.e. 2
arctan(b/a), and when this is 90°, i.e. a = b, the hyperbola is said to be rectangular, e.g. in the
Michaelis-Menten equation. The plane will also intersect the left hand cone and give a
symmetrical curve. Indeed, when the plane is not parallel with the cone axis, both halves of
the hyperbola are still symmetrical, as shown roughly below using CalcPlot3D.

Proof of the general form of the parabola. When the plane intersects the double conic parallel
with the surface it will only cross one of the cones and this defines a parabola. In the figure
below, note that the axes are swapped and oriented such that y and a are parallel with the
cone surface.

2
Considering the right-angled triangle on the base of the cone (redrawn on the right) gives:

b 2 2
b
x2 =( y ) −( y −2b)
a a
b 2 b 2 b
=( y ) −(( y ) −4 b( y )+ 4 b 2)
a a a
2
b
=4 y −4 b 2
a
a
∴ y =( 2 )x 2 +a
4b

and the form of a simple parabola y= Ax2 where A is constant, is proved. The general form
of a simple parabola is, of course y= Ax2 + Bx+C , but the effect of the extra terms (Bx and
C) is simply to shift the parabola to different locations, as shown in the fooplot.com graph
below.

Note that this derivation has been for a vertical parabola, whereas most sources e.g. this
excellent one [4] , derive the formula for a horizontal one i.e. y 2=4 ax .

When the plane crosses only one of the cones, forming a closed curve, this is an ellipse. In
the figure below (adapted from here [5]), line OG is the generator of a cone which is occupied
by two Dandelin spheres of different size, that are not touching each other. Each sphere
makes a circular line of contact with the inside of the cone and these two circles of contact are
perpendicular to the cone axis (not drawn) and are therefore parallel with each other.

3
The distance Q1Q2 is therefore a constant, regardless of where the generator (OG) lies on the
surface of the cone. The figure above shows a dark grey plane, passing through the cone,
which touches the spheres at points F 1 and F2 and the generator at P. Lines PF 1 and PQ1 are
therefore tangents to the sphere on the left and lines PF 2 and PQ2 are tangential to the sphere
on the right, meaning that PF 1 = PQ1 and PF2 = PQ2. Hence, PF1 + PF2 = PQ1 + PQ2. Since,
PQ1 + PQ2 is a constant, regardless of where P lies on the ellipse, PF 1 + PF2 must also be
constant. This is the classic definition of the ellipse, allowing it to be drawn by fixing the ends
of a piece of string to the page with pins and sliding the tip of a pencil inside the enclosed
loop. F1 and F2 are therefore the familiar foci of the ellipse, i.e. the points on the dark grey
plane where the drawing pins would have to be placed in order for a pencil at P to trace out
the ellipse. When the pencil is at either end of the long axis, its distance from the nearest pin
will be the same on each side of the ellipse. Hence, the foci must be equidistant from the
centre of the ellipse.

Deriving the equation of the ellipse from a cone (e.g. here [6]) is too hard so we’ll imagine that
the cone is infinitely long i.e. it degenerates to a cyclinder. A plane intersecting a cylinder at
right angles to its axis will give a circular cross-section. At any other angle, the intersection will
be elongated. We can imagine two Dandelin spheres inside the cylinder, although, of course,
they would have to be of equal size this time. We can have an intersecting plane which
touches both spheres, making the treatment essentially the same as above, but it is easier
just to do the trigonometry, as below, without Dandelin spheres.

Consider a circle (shown in the vertical direction above) which is formed by a plane crossing a
cylinder at right-angles to its long axis. If we tilt this plane through angle φ around the Y-axis
which, as shown in the figure, is also perpendicular to the cylinder axis, we get the grey
shaded ellipse and sinφ is referred to as its eccentricity (e). The semi-major and semi-minor
axes of the ellipse are a and b, respectively, and b is also radius of the original circle. Point P
has coordinates (x, y) with respect to the X- and Y-axes in the tilted grey plane. The X-axis of
the original circle is shown as X'. The projection of P onto the plane of the original circle has
coordinates (xcosφ, y) with respect to the X'- and Y-axes and since cosφ=b/a, the coordinates
b
become ( x , y ) and we can use Pythagoras as follows:
a

4
 b 2 2 2 x2 y2
( x ) + y =b giving + =1 as the equation for the ellipse.
a a2 b2

This is not so easy to do with a plane crossing a cone at right-angles because the centre of
the ellipse that is formed by tilting the plane moves away from the centre of the circle [6].
Right, so we have an equation for an ellipse, but is it consistent with what we derived from the
Dandelin spheres in a cone, i.e. that the sum of the distances of P from its two foci must be a
constant? Given that we could have two Dandelin spheres of equal size in a cylinder and
derive the same result from a plane touching both, its looking pretty promising.

But, to be sure lets go back to the cylinder. If we were to place the point P in the above figure
exactly on the positive side of the X-axis (i.e. at x = a and y = 0), if this really is an ellipse
then, since the foci must be equidistant from the centre, PF 1+PF2 must be 2a. If instead we
place P exactly on the positive Y-axis (as shown below) then x = 0 and y = b. Since PF1 = PF2
and PF1+PF2 = 2a then PF1 = PF2 = a. Hence, the triangle formed by P, F 2 and the origin is
identical with that formed by X, X' and the origin. Hence, the distance of each focus from the
origin is given by the following equivalent expressions: √ (a 2−b2 ) , asinφ or ae, and is
shown as c below. This distance, c, is called the linear eccentricity.

So, if we take the shape formed by intersecting a plane with a cylinder and consider the foci
to be at ±ae on the semi-major axis, then we have a shape that is identical in form to the
ellipse formed by the intersection of a plane with a cone, I thik.

In the figure above we placed P at the special position (0, b) and showed that the distance
from each focus is a. When P is at the general position (x, y), its distance from each of the
foci is a bit harder to work out, but not too bad. If we consider F 2 which is at position (c, 0), the
2 2
square of its distance from P is given by PF 22=(c−x )2+ y 2 where c=ae= √(a −b ) . Given
that the equation for the ellipse can be rearranged as y 2=b 2( 1−x 2 /a2) we get:

5
 x2 2 2 2
2 2 2 2 2 2 2 x 2 2 2 2 2 x 2 2 2 x
PF =(c−x ) + b (1− 2 ) =c −2cx + x +b −b 2 =a −b −2 cx+ x +b −b 2 =a −2 cx + x −b 2
2
a a a a
2 2 2
b a −b
=a2−2cx + x 2 (1− 2 ) =a 2−2 cx + x2 ( 2
) =a2−2 cx + x 2 e 2 =a2−2 aex + x 2 e 2 =(a−xe)2
a a

Hence, PF2 = a – xe and since PF 21=(c + x)2 + y 2 we can see that PF1 = a + xe, i.e. the sum
of the distances of P from the foci is, as expected, 2a. Hence, the formula for the ellipse
derived from the cylindrical section is consistent with the conic.

A summary of the ellipse formulae is given below.

Eccentricity is defined as the ratio of c/a i.e. the distance from centre to focus divided by the
distance from centre to vertex. For a circle the centre is the focus so e = 0. For the ellipse, we
have derived the formula of e=√( a2−b 2)/a and this will be in the range 0 < e < 1 since
when the eccentricity approaches 1, the ellipse becomes a parabola. One advantage of the
conic visualisation of the ellipse over the cylindrical one is that we can see that if the
intersecting plane becomes very tilted, i.e. φ approaches 90°, the ellipse becomes infinitely
long and open on one side but one vertex and focus will remain in the finite realm.

For the hyperbola, which is basically two open curves with their vertices on opposite sides of
the origin, the distance of each vertex from the centre is a, as in the earlier figure, but what is
the distance to the focus? Hmmm. We can use a Dandelin sphere in each of the cones of the
double conic which touches the plane of the hyperbola. Redrawing the hyperbola figure so
that we are looking exactly along the Y-axis with the X-axis horizontal gives the following view
of one half of the double-conic.

6
Line CG is the plane of the hyperbola which has a vertex, C, at a horizontal distance of a on
the X-axis, where the radius of the cone is b. A Dandelin sphere with centre at D and touching
the plane of the hyperbola at F, the focus, is shown. Since triangles ABC and AED are
congruent, the focal distance AD =AC =c= √ (a2 +b 2) . Hence for the hyperbola, the
eccentricity, e or c/a, is greater than 1. Note that AB = AE = a, i.e. the circle of contact
between the sphere and the cone is at distance a from the tip of the cone, as measured on
the surface of the cone. Note also that we are relying on the assumption that wherever the
Dandelin sphere touches the plane containing the curve is the focus of the curve. We need to
look more at what a focus actually is and will do this later on.

Now to the parabola and first we need to find where the focus is, which can be done using a
similar treatment to the above. In the figure below, line BE is a sideways-on view of the plane
containing the parabola which has its vertex at B and focus at F. The line through AB is the
generator of the cone and DB is its radius at the vertex of the parabola. Since triangles ABD,
ACD, BCD and BDF are all similar, we can see that BF is b2/a, i.e. the focus of the parabola is
at this distance from the vertex.

7
A problem arises with the parabola because it doesn’t have a centre from which to measure
distances to the vertex and focus. This is because the centre is at infinity i.e. it’s like the two
curves of a hyperbola or a closed ellipse with one of the two halves infinitely far away. So how
do we measure the eccentricity without a centre? Well, it seems you can use something
which I have been deliberately avoiding till now, namely the directrix. Most textbooks start with
the directrix, which is just a line, and derive everything about conic sections from it, but to me
it seemed better to start with an actual cone and relate everything to that, since these are
called conic sections after all! So what is this distractrix all about then? Apparently it is a line
from which you measure an orthogonal distance to point P on the curve. The distance of P
from the focus is then expressed as a fraction of the distance from the directrix and this is the
eccentricity. The textbook definition of a conic section is that this ratio must be constant for
any given curve. Fine, but confusing without a diagram.

Consider the first diagram of the parabola from which we derived y=(ax 2 /4 b 2)+a . If,
instead, we had placed the origin at the vertex of the parabola, we would have derived the
following simpler equation: y=(ax 2 /4 b 2) , and we’ve just shown that the focal distance of the
parabola (f) is b2/a, so we can now say: y=x 2 /( 4 f ) . This is drawn below and gives an
interesting result that when y = f, x = 2f, i.e. the full horizontal width of the parabola at the
focus (referred to as the latus rectum) is therefore 4f.

With the parabola, the latus rectum running horizontally through F (not shown) and the
directrix are parallel with the X-axis and are equidistant from it. The ratio PF/PD should
therefore be constant and give us the eccentricity. PD is given by:

x2
PD= y + f = +f
4f
x2 2
∴ PD 2=( +f )
4f
2 2
x x2
=( ) + 2( )f + f 2
4f 4f
2 2
x x2 2
=( ) + +f
4f 2

PF is determined by Pythagoras as:

Hence, PF = PD and the ratio PF/PD or eccentricity of the parabola is 1.

To prove that the focus really is a focus, consider a ray of light travelling in the direction of PD
which is being reflected by a parabolic mirror with angle of incidence, θ, equal to the angle of
reflection. Given that all of the triangles below are similar and PF = PD, FB = BE = f, etc., we
can see that FA = AD, i.e. line PA bisects the angle between PF and PD such that line AC
x2 x
behaves as a perfect mirror plane at P. The gradient of AC is PG/AG or ( ÷ ) or x/(2f)
4f 2
which is the same as the gradiant of the parabola that can be obtained by differentiation.
Hence, line AC must be a tangent to the parabola which reflects the ray towards the focus, F.

So, having introduced the directrix, where is it in relation to the Dandelin sphere? This
question is addressed in this online essay [7]. The figure on the left below shows that it lies on
the intersection of the plane of the circular contact between the cone and the Dandelin sphere
(line through DC) and the plane of the parabola (line through DE). Both planes are
perpendicular to the page, i.e. they appear as lines, and the line of intersection, or directrix, is
also perpendicular to the page and is shown as a black dot at D. The figure on the below right
shows the situation with the hyperbola where the lengths of DB and BF are not equal. We
showed earlier that the focus of the hyperbola is at a distance of (c-a) or a(e-1) from the
vertex (shown as horizontal line BF). Since CB = BF and DB = CB cosφ , we can see that the
distance of the directrix from the vertex, DB = a(e-1)a/c or (e-1)a/e.

9
A simplified Dandelin sphere treatment of the hyperbola is shown for the double conic below
which has generator G1G2. The plane of the hyperbola is parallel with the cone axis and
therefore both Dandelin spheres are the same size. Using the same considerations as for the
ellipse we see that PF1 = PQ1 and PF2 = PQ2. Point P traces out a hyperbola by swivelling the
generator G1G2 around the cone axis. The difference PF 1 - PF2, which equals PQ1 – PQ2, is
therefore constant and equal to the distance along the generator between the circles of
contact between each cone and its Dandelin sphere. We showed above that these circles are
at distance a from the tip of each cone, as measured on the surface of the cone, and we can
therefore say that PF1 – PF2 = 2a. This equation for the hyperbola is analogous to the one we
derived for the ellipse (PF 1 + PF2 = 2a) using Dandelin spheres. The figure below was
adapted from [8][9] and shows the plane of the hyperbola in the plane of the page with the
Dandelin spheres below it, for greater simplicity.

As we did with the ellipse, we can check that the equation PF1 – PF2 = 2a agrees with the
equation relating the Cartesian x, y coordinates. If we consider F2 which is at position (c, 0),
the square of its distance from P is given by PF22=(x−c )2+ y 2 where c=ae= √ (a2 +b 2) .
Given that the equation for the hyperbola can be rearranged as y 2=b 2( x2 /a 2−1) we get:

x2 2
2 x b2
PF 22=(x−c )2+ b2 ( 2 2 2 2
2
−1) =x −2 xc+ c + b 2
−b =x (1+ 2
)−2 xc +c 2−b2
a a a
2 2 2
a +b c
=x 2 ( 2 )−2 xc+ a2 =x 2 ( 2 )−2 xc +a 2 =x 2 e2−2 aex +a 2 =(xe−a)2
a a
10
Hence, PF2 = xe-a and similarly PF1 = xe+a, i.e. the difference between the distances of P
from the foci is, as expected, 2a.

The next thing to check is that the directrix whose position we determined on the right hand
side of the figure before the last one agrees with the rule that PF/PD should equal the
eccentricity. We can see in the figure below that PD = x-a/e or (xe-a)/e and we have just
shown that PF is (xe-a), i.e. PF/PD = e.

Right, since we’re dawdling around Dandelin spheres, where are the directrices of an ellipse?
The ellipse formed by two Dandelin spheres in a cone is shown on the left with the generator
OG in the fully upright position, i.e. P is collinear with the foci, F 1 and F2. From the ellipse
summary diagram we know that PF 1 and PQ1 are equal to a(1-e), but the rest of the maths for
a cone is too hard, so I swapped back to the cylinder representation with two Dandelin
spheres of equal size and P at a vertex of the ellipse, as on the right. This time, we are
viewing the plane of the ellipse exactly sideways on so the directrix D is orthogonal to the
page. The line DP shows the distance of the directrix from the vertex of the ellipse, which can
be determined from the triangle DPQ 1 since DP = PQ1 / sin φ or a(1-e)/e. The distance of D
from the centre O is OP + DP (or a+a(1-e)/e) which works out easily as a/e. Reassuringly,
these formulae are the same as those we derived for the hyperbola.

We now know where the directrices are for the ellipse and the hyperbola, but can we show
what features the foci have e.g. are they really foci as with a parabolic mirror?

11
The case of an ellipse is shown below and is adapted from a slightly scary one on Stack
Exchange [10], but I am not expert enough to fault it. The idea is to extend the line F 1P by the
length of F2P and there place the point F 2'. If we then place the point M so as to bisect F 2'F2
and draw a line through P and M, then with all the similar triangles we have something which
looks very much like a tangent, with angle of incidence (θ) equalling the angle of reflection,
etc., but is it really a tangent? The idea is that it only touches the ellipse at one point (P) so it
must be a tangent because any other line through P would have to cross the ellipse
somewhere else, too. But can we be sure that this line does only touch the ellipse once? The
answer to that is that if we take any other point on the line, represented by Q, then since it is
outside the grey shaded area of the ellipse, we can, almost Venn-diagram style, say that F 1Q
+ F2Q must always be bigger than 2a and therefore Q can never obey the sum of the
distances to the foci rule. Bit hand-wavy, but saves doing pages of algebra and it seems to
confirm that waves emitted at F 1 (shown as thick grey arrows) would all be reflected towards
F2 and vice versa, examples being elliptical rooms, allowing someone at one end to overhear
people whispering at the other.

On re-reading the proof, the clever bit is that F 2P = F2'P i.e. the length of the line from F 1 to F2'
is 2a. Considering triangle F1QF2', the sum of F1Q and F2'Q must be greater than 2a (or it
wouldn’t be a triangle, aka the triangle inequality) and since F2'Q = F2Q, it is easy to see that
F1Q + F2Q > 2a, i.e. Q must always be outside the ellipse, so the line PM must be a tangent.

Using the same notation, I adapted this proof for the hyperbola which confirms its focus
property. The main difference is that the line F 1P is extended backwards by the distance F 2P
and the line PM is drawn such that it bisects F 2F2'. Since F1P – F2P = 2a = F1P - PF2' we can
see that F1F2' must be 2a. Considering triangle F1F2'Q the triangle inequality dictates that F 1Q
< F1F2' + F2'Q or F1Q – F2'Q < F1F2'. Hence F1Q – F2Q < 2a and Q therefore cannot lie on the
hyperbola. Note also that the tangent can never cross the other half of the hyperbola since its
slope is always limited by the asymptote (grey dashed lines). Wikipedia has the same proof,
so it can’t be wildly wrong, and confirms that hyperbolic mirrors have a wider field of view than
parabolic ones, as the diagram below suggests.

12
Gradients of ellipse and hyperbola

The equation of the ellipse can be rearranged as follows and this allows implicit differentiation
(posh for chain rule) to be done (as per Stack Exchange [11]).

x2 y2
+ =1
a2 b 2
∴ b 2 x 2+ a2 y 2=a2 b2
dy
2b 2 x +2 a2 y =0
dx
2
dy −b x
∴ = 2
dx a y

The same treatment for the hyperbola gives the same equation, but of opposite sign.

x2 y 2
− =1
a2 b2
2
dy b x
∴ = 2
dx a y

Trying these formulae out for an ellipse and a hyperbola with a = 1, b = 0.5 gives promising
results for the tangent lines at the x and y values shown below, via fooplot.com:

13
Subjects still to be covered

The catenary, but that’ll do for now!

Bibliography
1: https://opentextbc.ca/algebratrigonometryopenstax/chapter/rotation-of-axes/
2: https://www.maa.org/press/periodicals/convergence/can-you-really-derive-conic-formulae-
from-a-cone-introduction
3: https://math.stackexchange.com/q/2246279
4: https://www.mathcentre.ac.uk/resources/uploaded/mc-ty-conics-2009-1.pdf
5: https://www.cut-the-knot.org/proofs/conics.shtml
6: https://math.stackexchange.com/q/4079720
7: https://www.yellowpigs.net/classes/ConicSections.pdf
8: http://nonagon.org/ExLibris/dandelin-spheres-conic-sections
9: http://www.nabla.hr/PC-ConicsProperties2.htm
10: https://math.stackexchange.com/q/1063977
11: https://math.stackexchange.com/q/1250919

14