Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Taylor & Francis, Ltd. and Mathematical Association of America are collaborating with JSTOR to
digitize, preserve and extend access to The American Mathematical Monthly
Abstract. In plane geometry, Holditch’s theorem states that if a chord of fixed length is
allowed to rotate inside a convex closed curve, then the locus of a point on the chord a distance
p from one end and a distance q from the other is a closed curve whose area is less than that
of the original curve by π pq. In this article we obtain, first, sufficient conditions to ensure
the existence of the Holditch curve and, second, a version of Holditch’s theorem for convex
polygons where the ellipse involved is explicitly shown.
q
C p
C1
πpq
Our interest in this theorem started from the inclusion of this result as one of Clif-
ford Pickover’s 250 milestones in the history of mathematics [5]. In the two pages
of the book devoted to the theorem, the author echoes the question raised by Mark
Cooker in [2]. Namely, the statement refers to an ellipse, but. . . where is it? The proof
of Holditch’s theorem, after adding some necessary hypotheses as can be found in [1],
uses usual techniques from calculus, but no reference appears to any ellipse other than
the fact that the area is π pq.
In this article we will try to address this missing ellipse. The first steps were already
presented in the cited paper [2]: If C1 is a rectangle and the stick is shorter than the
shortest side, then the Holditch curve consists of four similar quarter-ellipses. More-
over, in the same paper it is also said that if the angle between two consecutive sides
is not a right angle, then a part of an ellipse still appears although the semiaxes are no
longer p and q. Our investigation starts by noting that in this case (not a right angle)
http://dx.doi.org/10.4169/amer.math.monthly.124.5.403
MSC: Primary 52A10
404
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
Holditch curves may have retrograde motion. This happens when one of the end-
points of the chord passes through the same point at least twice, as happens in an
equilateral triangle. If both maps g and h are homeomorphisms of the circle, then
there is no retrograde motion. When g is a homeomorphism, Holditch curves can be
reduced to
p α(s) + ( − p) α( f (s))
Hα (s) = , s ∈ S1, (2)
where f = h ◦ g −1 .
First, we give the conditions that ensure the existence of Holditch curves.
Proof. We will prove the existence of the Holditch curve as an application of the
implicit function theorem. Let us define F : S 1 × S 1 → R as
Condition i) ensures that there are s0 and t0 such that F(s0 , t0 ) = 2 . More-
over, condition ii) is equivalent to saying that if s, t ∈ S 1 are such that (α(s) −
α(t)) ⊥ α (t), then α(s) − α(t) = . Since α(s0 ) − α(t0 ) = , we have that
α(s0 ) − α(t0 )), α (t0 )
= 0. (Figure 3 illustrates the geometric interpretation of the
second condition.)
Therefore, ∂∂tF (s0 , t0 ) = 0 and we can apply the implicit function theorem. There
exist Us0 , a neighborhood of s0 , Vt0 , a neighborhood of t0 , and a C 1 -function f 0 :
Us0 → Vt0 such that f 0 (s0 ) = t0 and, for any s ∈ Us0 ,
F s, f 0 (s) = 2 .
This means that for any s ∈ Us0 , α(s) − α( f 0 (s)) = . We have shown that if there
is one possible position of a chord of length with endpoints on the trace of α, specif-
ically α(s0 ) and α(t0 ), then a piece of any -Holditch curve can be built in neighbor-
hoods of s0 and t0 .
The extension of f 0 , and therefore of the Holditch curve, to the whole S 1 follows
from a typical argument using connectedness properties of S 1 . Let A ⊂ S 1 be the
subset where f 0 can be extended; that is,
F(s, f 1 (s)) = 2 .
α(s) − α( f (s)) = ,
for all s ∈ S 1 , which is the map that defines the -Holditch curves in α, according to
expression (2). This means that g = Id (which is obviously a homeomorphism) and
h = f is well-defined if we are working under the hypothesis of Proposition 1. We
shall restrict ourselves to the case without retrograde motion. In such a case, the map
f : S 1 → S 1 can be defined as follows: if the chord has an endpoint at α(s), then the
other endpoint is α( f (s)) according to the orientation of the curve (see Figure 2).
In this case we will say that α is -Holditch admissible (or, simply, admissible)
and we will denote by C−ad 1
(resp., P−ad
1
) the space of simple closed C 1 (resp.,
piecewise C ) -Holditch admissible curves.
1
α ( f (s))
α (s)
406
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
r(s) α (f (s))
α (s)
Example 1. Computation
of the Holditch radius for an ellipse. Let α(s) =
a cos(s), b sin(s) be the usual parameterization of an ellipse with semiaxes a and b.
The computation of its Holditch radius is a typical minimization problem with restric-
tions. Let
We have to compute the minimum value of F(s, t) under the restriction α(s) −
α(t), α (t)
= 0. Since
= 2a 2 cos(s) − cos(t) sin(t) + 2b2 cos(t) − sin(s) + sin(t) ,
27a 4 b4
4a 2 , 4b2 , 0, 3 .
a 2 + b2
27a 4 b4
Excluding the 0, the minimum is the last value (a 2 +b2 )3
. For example, taking a = 2 and
b = 1, the Holditch radius is
27a 4 b4 432
3 = = 3.456.
a 2 + b2 125
1.0
0.5
−
−2
2 −1 1 2
− 0.5
−1.0
1.0
The next result shows a sufficient condition to have an admissible curve (i.e., no
retrograde motion).
Theorem 1. Let α be a convex simple closed curve, let R H > 0 be its Holditch radius,
and > 0. If < R H , then α is -Holditch admissible.
408
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
Now we know that F̄ is constant equal to 2 by definition of f . Thus F̄ (s) = 0 for all
s ∈ S1.
Since < R H , we have ∂∂tF (s, f (s)) = 0 for all s ∈ S 1 . If we have ∂∂sF (s, f (s)) = 0,
then f (s) = 0 for all s ∈ S 1 , which would imply the injectivity of f . Let us show that
∂F
∂s
(s, f (s)) = 0. We will prove that ∂∂sF (s, f (s)) = 0 implies ≥ R H . Let s ∈ S 1 . The
fact that
means that α(s) − α( f (s)) is one of the distances considered in the minimum
r ( f (s)), f (s) ∈ S 1 . By definition of R H , we have
α(s) − α( f (s)) ≥ R H ,
Remark 2. In Theorem 1 we have seen how to obtain Holditch curves with the map
f : S 1 → S 1 injective depending on the length of the Holditch chord. We will now
show the geometric interpretation that is behind this result. For any point α(s) on a
closed curve α, let us consider the intersection between the curve and circle centered
at α(s) with radius . If, for some point α(s0 ) the intersection is empty, then α is not
-Holditch admissible. If the intersection is reduced to just one point α(t0 ), then the
circle is tangent to α at α(t0 ), and thus α(s0 ) − α(t0 ), α (t0 )
= 0. If the intersection
has 3 or more points, then the function F(s0 , t) = α(s0 ) − α(t)2 will have some
local minimum at some t0 with F(s0 , t0 ) ≤ . Therefore, the case we were looking for
is when, for any point α(s), the intersection is made up of only two different points
(see Figure 5), and this happens only when we have no retrograde motion. According
to the orientation of the curve, one of the intersection points, let us say α(t1 ), will be
the one next to α(s0 ). Hence, the map f : S 1 → S 1 is well-defined as f (s0 ) = t1 and
the curve is -Holditch admissible.
Example 2. The Holditch radius can be zero but a Holditch curve still exists, as hap-
pens in a square or in an equilateral triangle. In this latter case, an equilateral triangle,
retrograde motion appears as the segment traces the Holditch curve.
Now let us study the continuity of the map that sends any admissible curve (thus,
without retrograde motion) to its Holditch curve. First we will need a preliminary
result.
α(f(s ))
2
α(s ) 2
α(f(s ))1
α(f(s ))
1
α(s )
1
α(s )
2
α(s )
1
Figure 5. If is less than the Holditch radius, the intersection between the curve and circle centered at α(s)
and with radius is reduced to only two points (left). If at some point that intersection has three or more points,
then we will have retrograde motion (right).
Lemma 1. Let α and β be two simple closed and admissible curves. Let us denote by
f (resp., g) : S 1 → S 1 the map of the circle defining the Holditch curve of α (resp.,
β); that is,
p α(s) + ( − p) α( f (s)) p β(s) + ( − p) β(g(s))
Hα (s) = , Hβ (s) = .
Given η > 0, there is δ > 0 such that, if α(s) − β(s) < δ for all s ∈ S 1 , then
f (s) − g(s) < η for all s ∈ S 1 .
Since α is -Holditch admissible, then for all s0 ∈ S 1 there is t0 = f (s0 ) such that
α(s0 ) − α(t0 ) = . Equivalently, F(s0 , 0, t0 , 0) = (2 , 0). Now, it is easy to check
that the 2 × 2 matrix given by the partial derivatives
∂F
∂t (s0 ,0,t0 ,0)
= −2 α(s0 ) − α(t0 ), α (t0 )
, 0 ,
∂F
∂v (s ,0,t ,0)
= (∗, −1),
0 0
has a maximal rank. We can then apply the implicit function theorem. There are neigh-
borhoods Us0 of (s0 , 0) and Vt0 of (t0 , 0) and a continuous map f s0 : Us0 → Vt0 such
that f s0 (s0 , 0) = (t0 , 0) and, for all (s, u) ∈ Us0 ,
F s, u, f s0 (s, u) = (2 , 0). (4)
s s
If we write f s0 (s, u) = f 1 0 (s, u), f 2 0 (s, u) , then (4) is equivalent to
s
f 2 0 (s, u) = u,
s s
α(s) + u h(s) − α( f 1 0 (s, u)) + u h( f 1 0 (s, u)) = . (5)
410
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
β(s) − α(s)
= α(s) + α(s) − β(s)
α(s) − β(s)
≤ α( f (s)) − β( f (s)) + β( f (s)) − β(g(s))
ε ε ε
≤δ+ < + = ε.
2 2 2
1
Hα (s) − Hβ (s) ≤ p α(s) − β(s) + ( − p) α( f (s)) − β(g(s))
1
≤ ( p ε + ( − p) ε) = ε.
Remark 3. Notice that, once we have proven the continuity of the Holditch map, the
existence result of Holditch curves can be extended to the piecewise case. Indeed, any
piecewise C 1 closed curve is the limit of a sequence of C 1 closed curves.
If, in addition, a horizontal translation of vector x0 is applied, the resulting linear map
will be denoted by Sc,x0 : R2 → R2 and its expression is
There are two main properties of shear mappings that we are going to need. First,
since their Jacobians are equal to 1, shear transformations preserve areas. Second,
shears transform orthogonal ellipses into oblique ellipses.
The following result indicates what kind of curve the Holditch curve is in the case
we are focusing on and shows a way to find out where the area of the Holditch ellipse
comes from. Let us recall, as stated in [6], page 65, that the Holditch curve associated
to an angle is a particular case of the ellipse construction invented by Leonardo da
Vinci.
412
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
Figure 6. Holditch’s curve is an arc of an oblique ellipse which in turn is the image by a shear of an orthogonal
ellipse with semiaxes p and − p. Its area is θ2 p ( − p), where θ is the external angle.
Proof. Let us denote by θ the external angle formed by the two semistraight lines and
let t be the angle between the Holditch chord and one of the lines. We will use the
quantity t to parameterize the Holditch curve. The situation is represented in Figure 7.
B
−p
C
sin t sin t
p sin θ
A t θ
cos t − sin t cot θ O sin t cot θ V
We set the origin of the coordinates O = (0, 0) at the intersection of our two semi-
straight lines. Let A = A(t) and B = B(t) be the points defined by the ends of the
chord in the semistraight lines. Then it is easy to notice that the only thing we need
to be able to parameterize the Holditch curve is to know the coordinates of A and B,
since such parameterization would be defined as
1
γ (t) = ( − p) A + p B , t ∈ [0, θ]. (6)
If we focus on the triangle OVB and call its hypotenuse v, then we have sin(t) =
v sin(θ), from which it follows that v = sin(θ)
sin(t)
. Thus, the side O V of the triangle has
length v cos(θ) = sin(t) cot(θ). From this we get
B = sin(t) cot(θ), sin(t) .
Now we focus on the triangle AVB. If the length of the side AO is called x, then
cos(t) = x + sin(t) cot(θ). From that we obtain x = cos(t) − sin(t) cot(θ).
γθ, p, (t) = −( − p) cos(t) + sin(t) cot(θ), p sin(t) , t ∈ [0, θ] (7)
Now that we know the explicit parameterization of the Holditch curve given in (7),
we can see what it is like in some easy examples. For instance, Figure 8 shows the inner
curve when the initial curve is a square as well as when it is an equilateral triangle.
In these examples, we can understand the usefulness of the shear map defined in the
previous proof at each vertex of the polygonal curve. In the square we have four equal
parts of an ellipse with semiaxes lengths 1 and 1 − p and in the triangle, three pieces
of an oblique ellipse (obtained by a shear transformation of an orthogonal one). From
the fact that shear maps preserve areas, we deduce that the area in both examples is the
same as the area of the Holditch ellipse, as stated in Proposition 2.
Figure 8. Holditch curve in a square and in an equilateral triangle with different choices of p.
We will use the previous basic result about Holditch curves in the following two
sections, in which we conduct a separate study depending on the length of the chord.
414
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
Proposition 3. The area between a convex closed polygonal curve with sides of length
larger than the length of the moving chord and its Holditch curve is equal to the area
of an ellipse with both semiaxes lengths being the same as those into which the point
divides the chord.
Proof. Suppose that the initial curve is an n-sided convex closed polygon with external
angles θi , i = 1, . . . , n, defined at each vertex. By Proposition 2, the area at each
vertex between the polygon and the Holditch curve is the area of a (0, θi )-sector of an
orthogonal ellipse with semiaxes lengths p and − p. That area is well known and
equal to
θi
p ( − p).
2
Therefore, if we have n vertices, the total Holditch area would be
n
θi 1 n
p ( − p) = p ( − p) θi .
i=1
2 2 i=1
n
Now i=1 θi = 2 π since it is the sum of all the external angles in a simple closed
polygon. Thus we conclude that the total Holditch area is π p( − p), the area of the
ellipse with semiaxes lengths p and − p that we were looking for.
5. CUTTING OFF A CORNER. In the previous section we have seen how to deal
with the case of a polygonal curve with all sides longer than the length of the moving
chord. Since our interest is to build a sequence of convex polygonal curves approxi-
mating a curve, the next step is to deal with a polygonal curve, some of the sides of
which are shorter than the chord.
The situation now is that the endpoints of the moving chord could not be on con-
secutive sides of the polygon. Then the Holditch curve will be piecewise defined, as
we show in Figure 9.
θ2
θ1 θ1+θ2
C1 C2
A2 A1 O2 O1
Lemma 2. Let C1 be the oblique ellipse defined as the image of an orthogonal ellipse
by the horizontal shear transformation Sc1 (x, y) = (x + c1 y, y), and let C2 be the
oblique ellipse defined as the image of the same orthogonal ellipse by a horizontal
shear transformation plus a horizontal translation Sc2 ,x0 (x, y) = (x + x0 + c2 y, y)
with x0 < 0 (see Figure 10). Suppose that c1 < c2 . Let Ai be the leftmost intersection
between the line joining the centers O1 and O2 of both ellipses and the ellipse Ci ,
i = 1, 2. Among the four possible intersection points between the two ellipses, let P
be the one located above that line and to the left. The area defined by the three points
A1 , A2 , and P is equal to the area of a triangle with a base of length −x0 and height
x0
c1 −c2
.
Proof. If we apply Sc1 −c2 to the second ellipse, then what we get is the same first ellipse
C1 but translated according to the vector (x0 , 0). The region (P, A2 , C) (see Figure 11)
is transformed into the region defined by points (Q, A2 , C). The area we are looking
for is the area of (P, A2 , C) minus the area of (P, A1 , C). Since shear transformations
preserve area, the area of (P, A2 , C) is the same as the area of (Q, A2 , C). Moreover,
the area of (P, A1 , C) is the same as the area of (Q, A2 , D). Therefore, the area of
(P, A2 , A1 ) is the same as the area of the triangle (Q, D, C). The length of its base is
the norm of the vector (x0 , 0) and its height is the height of the intersection point P. A
straightforward computation shows that the second coordinate of P is c1x−c 0
2
.
416
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
A2 D A1 C
The following idea illustrates the purpose of the previous result. The case we are
focusing on now, as we said before, is the one shown in Figure 9, which is none other
than the basic case of two semistraight lines that form an external angle θ1 + θ2 , but
cutting off its corner with another straight line that forms angles θ1 and θ2 with each
of the first two lines. Now, in this case we want to study the new area between the
Holditch curve, which is piecewise defined, and the sides of the polygon. The idea is
to prove that the area of the triangle removed by the new line is equal to the area of
both the new bumps of area that appear on each side. This is what the following results
deal with. See Figure 12.
Figure 12. The area of the removed triangle is equal to the sum of the areas defined by the two new pieces of
the Holditch curve.
In what follows in this section, for the sake of simplicity we will assume that = 1.
Clearly this can be achieved with a scale change.
Lemma 3. The area of the left bump is equal to p times the area of the removed
triangle.
Proof. We will apply Lemma 2 to the situation shown in Figure 9. In this case, the
two shears are defined by the parameters (according to the notation in the statement of
Lemma 2):
where c1 < c2 and x0 < 0. Therefore, the area of the left bump is equal to
On the other hand, let us compute the area of the removed triangle. We can get its
height from the system of equations
h
−x0 +a
= tan(θ1 ),
h
a
= tan(θ1 + θ2 ),
Proposition 4. The area of the removed triangle is equal to the sum of the areas of the
two new bumps of the Holditch curve.
1-p
p 1-p
Proof. Let AT be the area of the removed triangle. As we have seen, the area of the
left bump is equal to p AT . Thanks to a flip with respect to the bisecting line of the
inner angle at the corner, we can interchange both bumps (see Figure 13). Notice that
the flip also interchanges the lengths p and q = 1 − p in the chord. Hence, the area of
the second bump is equal to (1 − p) AT . Thus, the sum of the areas of both bumps is
equal to AT .
418
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
A removed
A small
Figure 14. Case when the straight line that cuts off the corner also cuts the Holditch curve.
We remark now that in the previous procedure we have supposed something implic-
itly. The fact is that, depending on the choice of p, there is another possibility that we
have not talked about. This appears when the straight line that cuts off the corner also
cuts the Holditch curve we had. In this case we do not really remove the full area
inside the triangle, only Aremoved (see Figure 14). It is easy to check that the sense of
the previous statement is still true in this case, i.e., the sum of the areas of both bumps
is equal to the removed area. Indeed, let us suppose, without loss of generality, that the
left bump is the one with problems. We have proved before that
Then, with the notation of Figure 14, since AT = Aremoved + Asmall and Aleft bump =
Anew bump + Asmall , we conclude that
as we wanted to see.
If the previous problem appears in both bumps, we can argue similarly with each of
them to reach the same conclusion, i.e., the sum of the areas of both bumps is equal to
the removed area.
Figure 16. The three first terms in the sequence toward the Holditch curve of a circle (above) and of a convex
planar curve (below).
Since the sequence of polygons {Pn }∞n=2 converges to the curve α and the map send-
ing any admissible curve to its Holditch curve is continuous (Theorem 2), then the
sequence of Holditch curves {Hn }∞n=2 converges to the Holditch curve of α. Intuitively,
this shows that the Holditch area is the result of an infinite number of “cutting off a
corner” steps which distribute the area of the Holditch ellipse by pasting small pieces
of it.
ACKNOWLEDGMENT. The authors wish to thank Mark Cooker for his interest and careful reading of the
preprint we sent him. We have taken note of his comments to improve some details in the writing. The first
author is partially supported by DGICYT grant MTM2012-33073.
REFERENCES
420
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124
JUAN MONTERDE received his Ph.D. from the University of Valencia in 1988. His interests range from
classical differential geometry to computer-aided geometric design.
Dept. of Mathematics, University of Valencia, Avd. Vicent Andrés Estellés, 1, E-46100-Burjassot (València),
Spain
monterde@uv.es
DAVID ROCHERA earned his master’s degree from the University of Valencia in 2015. He is especially
interested in the areas of classical differential geometry and applied mathematics.
Dept. of Mathematics, University of Valencia, Avd. Vicent Andrés Estellés, 1, E-46100-Burjassot (València),
Spain
David.Rochera@uv.es