You are on page 1of 7

266 THE MATHEMATICAL GAZETTE

Conics and convexity


K. ROBIN McLEAN

1. Introduction
In [1], W. D. Munn proved the following result.

Theorem 1: Infmitely many ellipses pass through the four vertices of a


given convex quadrilateral.

Much of the geometry that I studied as an undergraduate in the 1950s


concerned complex projective space, in which convexity plays no part. So I
found Theorem 1 especially piquant and sought to understand it better. This
article is the result. After examining the convexity of quadrilaterals in
general, especially those inscribed in conics, I consider the following
problem. Let P be a variable point in the plane, distinct from the vertices of
a given convex quadrilateral ABCD. It is well known that there is a unique
conic, S (P), through the five points A, B, C, D and P. How does the nature of
this conic depend on the position of P? As a spin-off, we get a very short
proof of Theorem 1. Finally I look at what happens when the quadrilateral
ABCD is not convex. In this case, S (P) is always a hyperbola, but the
distribution of A, B, C and D on its branches is still of interest.

2. When is a quadrilateral convex?


There are many equivalent ways to define convexity for a quadrilateral,
and readers may like to check the equivalence of our definitions. We say that
a set of points, X, in a plane is convex if, for every A, B E X, the whole of the
line segment AB lies in X. A simple polygon, one with no crossings, is convex
if its interior is a convex set, or equivalently if none of its interior angles is
reflex. Finally, a simple quadrilateral ABCD is convex if it is convex as a
polygon, as above, or equivalently if its diagonal line-segments AC and BD
have a point in common. (See Figure 1.) A simple quadrilateral (we shall
consider no others) that is not convex is said to be concave or re-entrant.

o"
C

./
B\PA
D (a) A D (b)

FIGURE 1: (a) A simple convex qudrilateral ABCD. Its diagonal line segments AC
and BD meet at the point O. (b) A simple concave or re-entrant quadrilateral ABCD.
The line segments AC and BD do not meet.

3. Convexity properties of quadrilaterals inscribed in conics


Since opposite angles of a cyclic quadrilateral ABeD add up to 1800, no

https://doi.org/10.1017/S0025557200001303 Published online by Cambridge University Press


CONICS AND CONVEXITY 267

interior angle can be reflex so, by a criterion of the previous section, ABCD
is convex. This is the first of four well-known and easily visualised results:
(l) Any quadrilateral inscribed in a circle is convex.
(2) Any quadrilateral inscribed in an ellipse is convex. (This follows from
(1) because, in a suitable coordinate system, we can map the ellipse
into a circle and back by transformations of the form (x, y) ~ (x, Ay).
These do not affect convexity.)
(3) Any quadrilateral inscribed in a parabola is convex.
(4) Any quadrilateral inscribed in a single branch of a hyperbola is
convex.
Less well known is the following more general result, of which (4) is a
special case:
(5) A quadrilateral inscribed in a hyperbola is convex if, and only if, an
even number of its vertices lie on each branch of the hyperbola.

Examples of each of the three different types of quadrilateral inscribed


in a hyperbola are shown on Figure 2. When all four vertices lie on a single
branch, or when two vertices lie on each branch, the quadrilateral is convex,
as in (a) or (b). When three vertices lie on one branch and one vertex lies on
the other branch, as in (c), the quadrilateral is concave.

) ~~ It (a) (b)
HaVRE 2: Qudrilaterals inscribed in a hyperbola.
(c)

All these results can, if desired, be proved algebraically. For example, to


prove (5), take the asymptotes as oblique axes (see, for example, [2, p.108]).
Then, with suitable scaling, we may let the hyperbola have parametric
equations x = t, Y = 1 I t with A, B, C, D given by t = a, b, e, d
respectively. The diagonal AC has equation x + aey = a + e. Let the point
of intersection, 0, of AC and BD divide BD in the ratio /1 : A and AC in the
ratio /1' : A'. Then 0 is
Ab + Ild, A(llb) + /1(1ld)).
(
A+/1 A+/1
It is easy to verify that substituting this into the equation of AC gives
A b (d - a)(d - e)
- = (1)
/1 deb - a)(b - e)
Similarly, swapping a with b and e with d,
A' a(e - b)(e - d)
(2)
/1' e(a - b)(a - d)

https://doi.org/10.1017/S0025557200001303 Published online by Cambridge University Press


268 THE MATHEMATICAL GAZETTE

From (1) and (2),


A A' ab(d - d (3)
Ii Ii' cd (b - af·
If ABCD is convex, then 0 lies on both line segments AC and BD, so that
M' 1!JIl' > O. By (3), this means abl cd > 0, so that an even number of
vertices of the quadrilateral lie on each branch of the hyperbola. To reverse
the argument, note that we assume ABCD to be simple so, if M' 1!JIl' > 0,
then Alii > 0 and A'Iii' > O.
4. How does the conic S (P) depend on P?
Turning to the problem posed earlier, let the diagonals AC and BD of the
~ ~
given convex quadrilateral ABCD meet at O. Choosing OA, OB as the
positive directions along oblique axes Ox, Oy, we let A be (a,O) and B be
(0, f3), where a, f3 > O. Since ABCD is convex, C and D have the forms
(-y,O) and (0, -l5) respectively for some y, l5 > O. The conic
S == (Jl5(x - a)(x + y) + ay(y - (J)(y + l5) + a(Jyl5 = 0 (4)
clearly goes through A, B, C, D as does the line pair xy = 0 and any conic
of the form S + 2hxy = O. Now S + 2hxy = 0 is an ellipse if, and only if,
a(Jyl5 > h2• So if flo = .ja(Jyl5, then for each of the infinitely many values
of h in the range -ho < h < ho we get an ellipse through A, B, C, D, giving
a proof of Theorem 1.
Next, write S, == S + 2hoxy = 0 and S2 == S - 2hoxy = 0,
each of which is a parabola or a parallel line pair. In the regions where
xy > 0, we have
Sl > S + 2hxy > S2,
so that if (x, y) is on the ellipse S + 2hxy = 0, then at (x, y), Sl > 0 > S2,
which says (x, y) is outside SI and inside S2. For xy < 0 the inequalities,
and the conclusions, are reversed.
Figure 3 shows the general situation when ABCD is not a trapezium. In
this case Sl and S2 are parabolas, indeed the only parabolas of the pencil of
conics through A, B, C and D. The three line-pairs and two parabolas of this
pencil separate the real projective plane (see [3]) into 20 regions, numbered
in Figure 3, each having precisely two of the vertices A, B, C, D on its
boundary. Each of the other conics of the pencil passes through exactly four
of these 20 regions in a predetermined cyclic order. So our 20 regions can be
partitioned into five 4-cycles as follows:
A B 0 C
~ 2 ~ 3 ~ 4 ~
A B C 0
5 ~ 6 ~ 7 ~ 8 ~ 5
A C B 0
9 ~ 10 ~ 11 ~ 12 ~ 9
A 0 C B
13 ~ 14 -7 15 ~ 16 ~ 13
A B C 0
17 ~ 18 ~ 19 -7 20 ~ 17

https://doi.org/10.1017/S0025557200001303 Published online by Cambridge University Press


CONICS AND CONVEXITY 269

FIGURE 3: The three line-pairs and two parabolas through the vertices of a convex
quadrilateral ABeD divide the real projective plane into 20 regions. Each of the
regions 14, 16, 18 and 20 is bounded by an arc of a parabola and a straight line
segment and is too thin to be easily visible in this figure.

The first cycle indicates that any conic of the pencil that passes into
region 1 must go through A into region 2, then through B into 3, via D into 4
and via C back into region 1. Regions 1 and 3 cross the line at infinity.
Moreover, in going from 1 to 3 the conic passes through A and B, whilst in
returning from 3 to 1 it passes through D and C. So if P lies in any of the
four regions of this first cycle, then the conic S (P) is a hyperbola with A and
B on one branch and C and D on the other branch. Readers can use similar
arguments to check that when P lies in a region of the fourth or the fifth
cycle, S (P) is a hyperbola with A, B, C and D on the same branch and that
when P lies in a region of the third cycle, S (P) is a hyperbola with A and D
on one branch and Band C on the other. No region in the second cycle
crosses the line at infinity, although regions 5 and 6 extend to infinity and
each has one infinite point on its boundary. So when P lies in any second
cycle region, S (P) is an ellipse.
In the special case when ABCD is a trapezium, but not a parallelogram,
we may suppose that AB is parallel to CD. One of the two parabolas of the
pencil now degenerates to this parallel line-pair and we lose a complete
cycle of regions (the fifth) from Figure 3.
When ABCD is a parallelogram, there are no parabolas in the pencil and
a further cycle of regions (the fourth in our list) disappears, leaving the 12
regions shown in Figure 4 that constitute the first three of our original 4-
cycles.

https://doi.org/10.1017/S0025557200001303 Published online by Cambridge University Press


270 THE MATHEMATICAL GAZETTE

AGURE4: The three line-pairs through the vertices of a parallelogram divide the real
projective plane into 12 regions.

FIGURE5: The three line-pairs through the vertices of a concave quadrilateral


divide the real projective plane into 12 regions.

5. Concave quadrangles
When the quadrilateral ABCD is concave, it can have no circumscribing
ellipses or parabolas, so every non-degenerate conic through its vertices
must be a hyperbola. Moreover three vertices must lie on one branch whilst
one vertex, which I shall call isolated, lies on the other branch. Since ABCD

https://doi.org/10.1017/S0025557200001303 Published online by Cambridge University Press


CONICS AND CONVEXITY 271

is concave, one of its vertices lies inside the triangle formed by the other
three vertices. This interior vertex (denote it by A) cannot be isolated for, by
continuity, the branch through A must cut at least one of the sides (say BC)
of the surrounding triangle BCD, and it must do so at either B or C, for
otherwise BC would meet the hyperbola in three distinct points. The six
lines AB, AC, AD, BC, BD and CD divide the real projective plane into 12
regions. (See Figure 5.)
As in the convex case, these 12 regions fall into 4-cycles:
A C B D
~ 2 ~ 3 ~ 4 ~
A DeB
5~6~7~8~5
A B D C
9 ~ 10 ~ 11 ~ 12 ~ 9
In the first cycle, regions 3 and 4 cross the line at infinity. The point B is
the common boundary of these regions so, if P is a point in any region of the
first cycle, then the conic S (P) is a hyperbola with B isolated on one branch
whilst A, C and D lie on the other branch. Similarly when P lies in a region
of the second cycle, C is isolated and, when P lies in a third cycle region, D
is the isolated vertex.
Figure 6 shows some of the conics of the pencil through A, B, C and D.
All are hyperbolas or line-pairs, as predicted.

FIGURE 6: Three of the hyperbolas that pass through the vertices of a concave
quadrilateral. Each corresponds to one of the the three 4-cycles identified in the text.

Acknowledgements
I am grateful to Professor Peter Giblin for interesting discussions and
for showing me how Cinderella software (see [4]) can be used in visual
exploration of what happens to the unique conic through five points, A, B, C,

https://doi.org/10.1017/S0025557200001303 Published online by Cambridge University Press


272 THE MATHEMATICAL GAZEITE

D and P when P is dragged around the screen whilst the other four points
remain fixed.
The referee deserves special mention for commenting on earlier
versions of this article so fully and helpfully. As a result, several arguments
have been improved and new ones suggested. In particular, the observation
in §§4 and 5 that regions fall into 4-cycles is due to the referee, who also
unearthed an 1847 article ([5]) containing essentially the same argument as I
used in the first part of §4.
I would also like to thank the Editor for his patience in allowing me so
much time for revising the article.

Dedication
I never knew Douglas Munn but, from what I have read ([6, 7, 8]), he
was a fine teacher. Many readers of the Gazette have been fortunate to have
such teachers. So I would like to dedicate this article to the memory of those
who taught me mathematics at Bradford Grammar School and whose
example led me to become a teacher myself: Margaret Baker, from whom I
first learned about congruent triangles; Walter Thornton, from whom I
learned about pencils of conics; and Keith Turner, who once showed us
seven different solutions to a Cambridge Scholarship geometry problem and
who inspired so many of his pupils.

References
1. W. D. Munn, Ellipses circumscribing convex quadrilaterals, Math. Gaz.
92 (November 2008) pp. 566-568.
2. H. S. M. Coxeter, Introduction to geometry (2nd edn.), John Wiley &
Sons (1969).
3. H. S. M. Coxeter, The real projective plane (2nd edn.), Cambridge
University Press (1955).
4. P. J. Giblin, Review of Cinderella (Version 1.2) interactive geometry
software, Math. Gaz. 85 (July 2001) pp. 364-366.
5. S. Fenwick, On the plane quadrilateral, The Mathematician, Vol.lI,
No.6 (July 1847) pp. 285-295, also available at
http://books.google.co.uklbooks?id=59YLAAAAYAAJ&pg=PA285.
6. Scotsman Newspapers Limited, Douglas Munn (12 November 2008),
available at http://thescotsman.scotsman.com!obituaries!Professor-
Douglas- Munn.468381 O.jp
7. J. M. Howie, Tribute to Douglas Munn, Semigroup Forum 59 (1999)
pp.I-7.
8. N. R. Reilly, Walter Douglas Munn 1929-2008, Semigroup Forum 78
(2009) pp. 1-6.
K. ROBIN McLEAN
Department of Mathematical Sciences, University of Liverpool, Liverpool
L697ZL

https://doi.org/10.1017/S0025557200001303 Published online by Cambridge University Press

You might also like