Mathematical Proceedings of the Cambridge

Philosophical Society
http://journals.cambridge.org/PSP

Additional services for Mathematical

Proceedings of the
Cambridge Philosophical Society:

Email alerts: Click here

Subscriptions: Click here
Commercial reprints: Click here
Terms of use : Click here

On the construction of the regular hendecagon by marked

ruler and compass

ELLIOT BENJAMIN and C. SNYDER

Mathematical Proceedings of the Cambridge Philosophical Society / Volume 156 / Issue 03 / May 2014, pp 409
- 424
DOI: 10.1017/S0305004113000753, Published online: 24 January 2014

Link to this article: http://journals.cambridge.org/abstract_S0305004113000753

How to cite this article:

ELLIOT BENJAMIN and C. SNYDER (2014). On the construction of the regular hendecagon by
marked ruler and compass . Mathematical Proceedings of the Cambridge Philosophical Society,
156, pp 409-424 doi:10.1017/S0305004113000753

Request Permissions : Click here

Downloaded from http://journals.cambridge.org/PSP, IP address: 131.220.245.205 on 11 Mar 2015

Math. Proc. Camb. Phil. Soc. (2014), 156, 409–424 
c Cambridge Philosophical Society 2014 409
doi:10.1017/S0305004113000753
First published online 24 January 2014

On the construction of the regular hendecagon by marked ruler and

compass

B Y ELLIOT BENJAMIN
Mathematics Department, CALCampus, N.H. 03461, U.S.A.
e-mail: [email protected]

AND C. SNYDER
Department of Mathematics and Statistics, UMaine, Orono, ME 04469, U.S.A.
e-mail: [email protected]

(Received 13 September 2013)

In Memory of Ali Erhan Özlük, 1952–2012

Abstract
We prove that the regular hendecagon (11-gon) is constructible by marked ruler and
compass.

1. Introduction
We develop a theory that characterizes certain types of points and numbers which can be
constructed by a marked ruler with two marks one unit apart and a (Euclidean) compass. As
an application of the theory we prove that the regular hendecagon can be constructed with
these tools. This solves a long-standing open problem in the area of geometric constructions.
This is also a first result in a program to try to characterize all points constructible by
marked ruler and compass, a problem whose solution seems to be a long way off.

2. Preliminaries
In this section we examine various types of constructions involving different sets of tools.
All these constructions are applied in a Cartesian plane, R2 , where R is the field of real
numbers. As our first example we recall the classical straightedge and compass construc-
tions. (Cf. [3, chapter 2], for example.)
Straightedge and Compass Constructions
A point is constructible by straightedge and compass if it is the last point in a finite
sequence of points such that either the point is in the “starter set” {(0, 0), (1, 0)} or is a point
of intersection of:
(i) two distinct lines each passing through a pair of points earlier in the sequence;
(ii) two distinct circles each passing through points earlier in the sequence and with
earlier points as their centers;
(iii) a line and a circle as determined in (i) and (ii), respectively.
410 B ENJAMIN AND S NYDER
A real number x is said to be constructible by straightedge and compass if it is the x-
coordinate of a point on the x-axis that is constructible by straightedge and compass. (Actu-
ally, this is equivalent to declaring that x is the x-coordinate of a constructible point, since
projection onto the x-axis through a constructible point is a constructible line. A similar
consideration shows that the y-coordinate of a constructible point is constructible, too.) As
is well known, a number is constructible by straightedge and compass if and only if it lies
in a real 2-tower of Q, the field of rational numbers; i.e., a real number x is constructible by
straightedge and compass if and only if there exists a tower of fields K 0 , . . . , K n such that
Q = K 0 ⊆ K 1 ⊆ · · · ⊆ K n ⊆ R,
with [K j+1 : K j ] = 1 or 2, for j = 0, . . . , n − 1, and with x ∈ K n .
In the first part of the nineteenth century, Gauss and Wantzel characterized those regular
n-gons that can be constructed by straightedge and compass, namely those n > 2 of the form
n = 2a p1 · · · pk (for a, k nonnegative integers) where the pi are distinct Fermat primes, i.e.,
primes of the form 2m + 1, cf. [3, chapter 2].
Marked ruler constructions
We now consider a marked rule with two marks one unit apart on our straightedge and
allow for what is called verging, cf. [3, pp. 124–125]. By verging through a point V between
two geometric objects (for us lines and circles) we mean determining two points Q 1 and Q 2
obtained by drawing a line through the point V that intersects the two geometric objects at
points Q 1 on one object and Q 2 on the other such that Q 1 and Q 2 are one unit apart. Notice
that this can be done by a marked ruler once we are given a verging point and a pair of
geometric objects. The points Q i are thus said to be constructible by verging. With respect
to geometric constructions, a marked ruler can be used in two ways: for drawing a line
through two distinct points like a classical unmarked straightedge or as a tool for verging.
A point is said to be constructible by a marked ruler if it is the terminal point in a finite
sequence of points in R2 such that the point is either in the “starter set” {(0, 0), (1, 0), (0, 1)}
or is obtained in one of the following two ways:
(i) as the point of intersection of two distinct lines each passing through a pair of earlier
points in the sequence;
(ii) as either of two points obtained by verging through an earlier point between two
distinct lines as given in (i).
A real number x is constructible by marked ruler, or a marked ruler number, for short, if
it is the x-coordinate of a point on the x-axis constructible by a marked ruler (or equivalently
is the x- or y-coordinate of a point constructible by marked ruler).
Even though we have not included the compass as a tool, it turns out that any point con-
structible by straightedge and compass can be constructed by a marked ruler only, cf. [3].
Also there is a characterization of marked ruler numbers: A real number is constructible
by a marked ruler if and only if it lies in a real 2-3 tower of Q. Here, the tower is defined
analogously to a 2 tower except that we allow the degrees [K j+1 : K j ] = 1, 2, or 3, as the
terminology suggests.
Pierpont (late nineteenth century) characterized those regular n-gons that can be construc-
ted with marked ruler (more precisely, by a method equivalent to the marked ruler), namely
those n > 2 of the form n = 2a 3b p1 · · · pk (for a, b, k nonnegative integers) where the pi
are distinct Pierpont primes, i.e., primes of the form 2 3m + 1, cf. [3, chapter 9].
 Constructing the regular hendecagon 411
Notice that the regular hendecagon is the regular polygon with minimal number of sides
which cannot be constructed with only a marked ruler.

Restricted marked ruler and compass constructions

We now add a compass to our marked ruler constructions. This allows us to construct
circles and hence to add verging between a line and a circle or a pair of circles. Now, in the
classical case of straightedge and compass, we allow intersections of two lines, a line and
a circle, and two circles. However, recall that nothing is lost by restricting to the first two
cases. In a similar manner, in order to simplify some of the algebra, we restrict to verging
between pairs of lines and between a line and a circle, but not between two circles — hence
we will use the term restricted marked ruler and compass constructions. We do not know if
this restriction necessarily yields fewer points than no restriction, but it is sufficient to prove
the main result of the paper. Here is a formal definition.
A point is Restricted Marked ruler and Compass constructible or an RMC point, for short,
if it is the terminal point in a finite sequence of points in R2 such that the point is in the
“starter set” {(0, 0), (1, 0)} or is obtained in one of the following ways:
(i) as the intersection of two distinct lines each passing through a pair of points earlier
in the sequence;
(ii) as the intersection of two distinct circles each passing through points earlier in the
sequence and with earlier points as their centers;
(iii) as the intersection of a line and a circle as determined in (i) and (ii), respectively;
(iv) as one of two points obtained by verging through an earlier point between two distinct
lines as given in (i);
(v) as one of two points obtained by verging through an earlier point between a line and a
circle as given in (i) and (ii), respectively. We call the point on the circle the primary
point and the point on the line the secondary point.
For convenience we call a sequence of points as given in the definition an RMC sequence.
A restricted marked ruler and compass number or RMC number, again for brevity, is
defined analogously to the previous cases.
For the record, marked ruler and compass points and numbers may be defined similarly
by adding verging between two circles (with a minor restriction).
Now let
C = {x ∈ R : x is an RMC number}.
Notice that C is a subfield of R containing all real 2-3 towers of Q. In [1] Baragar showed
that any marked ruler and compass number (and thus also any RMC number) lies in a real
tower {K j } of Q with [K j+1 : K j ]  6. From this he was able to prove that any regular
n-gon for which n is divisible by a prime, p, such that there is a prime > 5 dividing p − 1,
cannot be constructed with marked ruler and compass (for example, n = 29). However,
Baragar was not able to determine what happens when all the primes p have p − 1 divisible
by only the primes 2, 3, 5 and 5 occurring at least once. In particular, n = 11 is the smallest
such number.
In the following sections we will characterize RMC numbers and then use this character-
ization to show that 2 cos(2π/11) is an RMC number, thus implying that the regular 11-gon
is constructible by marked ruler and compass.
412 B ENJAMIN AND S NYDER
3. The setup
It is known, cf. [1], that determining one point of the pair of points obtained by verging
between a line and some other geometric object can be given as a point of intersection of a
conchoid associated with the line (and the verging point) and the other geometric object. Let
V be a (verging) point and L a line not passing through V . For any point P on L consider
the line through V and P. Let Q and Q  be the two points on this line that are of distance
one to P. The conchoid Con L ,V associated with the verging point V and axis L is the set of
all points Q, Q  as given above as P varies over all points of L.
Now we determine an equation of the conchoid. To simplify the equation we let V =
(0, 0) be the verging point and L be a vertical line with equation x = a where a > 0. Hence
in polar coordinates L has an equation r cos θ = a or equivalently
L : r = a sec θ.
Thus the conchoid has equation
Con x=a,(0,0) : r = a sec θ ± 1.
Converting to rectangular coordinates yields
Con : (x − a)2 (x 2 + y 2 ) = x 2 ,
cf. [1]. (Notice that in rectangular coordinates, the origin is always on the graph.)
In the procedures for constructing restricted marked ruler and compass numbers, notice
that steps (i)–(iii) produce numbers that are in fields of degree at most degree 2 over the
previously constructed fields, whereas (iv) produces an extension of degree at most 4, cf. [1]
or [3]. Hence to be able to get an extension of degree 5 or 6 in one step we must apply step
(v). Thus by the previous consideration, we analyze the points of intersection of a conchoid
and a circle. This will be done in the following section.

4. Characterization of RMC numbers

Following Baragar, [1], we now consider the construction of the coordinates of the points
of intersection (if any) of the conchoid
Con : (x − a)2 (x 2 + y 2 ) = x 2 ,
associated with the origin and the vertical line x = a, and the circle C = C(b,c),s
C : (x − b)2 + (y − c)2 = s 2 ,
with center (b, c) and radius s. For the moment we require only that a, b, c, s are real num-
bers and without loss of generality that a > 0, c  0, s > 0. Rewriting the equation of the
circle as
C : x 2 + y 2 = s 2 − b2 − c2 + 2bx + 2cy
allows us to represent the coordinates of a point of intersection of the conchoid and circle as
Con  C : (x − a)2 (s 2 − b2 − c2 + 2bx + 2cy) = x 2 ,
or equivalently
A(x)y = B(x),
 Constructing the regular hendecagon 413
where
A(x) = 2c(x − a)2
and
B(x) = x 2 − (x − a)2 (s 2 − b2 − c2 + 2bx),
for short.
Notice that when c = 0, i.e., the circle is centered on the x-axis, then x satisfies the cubic
equation
B(x) = x 2 − (x − a)2 (s 2 − b2 + 2bx) = 0.
Hence, we need not consider this case for constructing quintic or sextic irrationalities in one
step. (See the last paragraph of the previous section.)
From now on we assume c > 0.
From above, we have
A(x)(y − c) = B(x) − c A(x)
so that
A2 (x)(y − c)2 = (B(x) − c A(x))2
and whence
A2 (x)(s 2 − (x − b)2 ) = (B(x) − c A(x))2 ,
which gives an equation for the x-coordinate of a point of intersection of the conchoid and
circle.
It turns out that it will be more convenient to work with
x
z=
x −a
rather than x. Notice that if we “know” z, then we know x, since
az
x= ,
z−1
and conversely. Also notice that if the point (x, y) is on the conchoid, then from the defining
equation of the conchoid, z satisfies the relation
z2 = x 2 + y2,
and thus z is a “signed distance” of the point (x, y) to the origin (0, 0). (The sign is easily
determined. Namely, z > 0 for x > a or x < 0, and z < 0 for 0 < x < a. Also, clearly
z = 0 if and only if x = 0.) One advantage of working with the signed distance is that we
may assume, without loss of generality, that our verging point is the origin and that the line
associated with the conchoid is vertical; this simplifies a lot of our computations.
We now construct a sixth-degree polynomial having z as a root. To this end, notice first
that
x az a
z= , x= , x −a = ,
x −a z−1 z−1
as is easily verified. In terms of z,
2ca 2
A(x) = 2c(x − a)2 =
(z − 1)2
414 B ENJAMIN AND S NYDER
and
 
a2 z2 a2 2abz
B(x) = − s −b −c +
2 2 2
(z − 1)2 (z − 1)2 z−1
2
a
= ((z 2 (z − 1) + (b2 + c2 − s 2 )(z − 1) − 2abz)
(z − 1)3
a2
= (z 3 − z 2 + (b2 + c2 − 2ab − s 2 )z + s 2 − b2 − c2 ).
(z − 1)3
For later use, notice that we can obtain y in terms of z (along with x):
az
x= , (4·1a)
z−1
B(x) z 3 − z 2 − (s 2 − b2 − c2 + 2ab)z + s 2 − b2 − c2
y= = . (4·1b)
A(x) 2c(z − 1)
Also, we have
 az 2
s 2 − (x − b)2 = s 2 − −b
z−1
1
= (s 2 (z − 1)2 − (az − bz + b)2 )
(z − 1)2
1  2 
= (s − (a − b)2 )z 2 − 2(s 2 + b(a − b))z + s 2 − b2 .
(z − 1) 2

Thus
(z − 1)6
A2 (x)(s 2 − (x − b)2 ) = 4c2 ((s 2 − (a − b)2 )z 2 − 2(s 2 + b(a − b))z + s 2 − b2 ).
a4
On the other hand,
a2 2c2 a 2
B(x) − c A(x) = (z 3
− z 2
+ (b 2
+ c 2
− 2ab − s 2
)z + s 2
− b 2
− c 2
) −
(z − 1)3 (z − 1)2
2
a
= (z 3 − z 2 + (b2 − c2 − 2ab − s 2 )z + s 2 − b2 + c2 ).
(z − 1)3
But then
(z − 1)6
(B(x) − c A(x))2
a4
= (z − z − (s − b2 + c2 + 2ab)z + s 2 − b2 + c2 )2
3 2 2

= z 6 − 2z 5 + (1 − 2(s 2 − b2 + c2 + 2ab))z 4 + 4(s 2 − b2 + c2 + ab)z 3

+ ((s 2 − b2 + c2 + 2ab)2 − 2(s 2 − b2 + c2 ))z 2
− 2(s 2 − b2 + c2 )(s 2 − b2 + c2 + 2ab)z + (s 2 − b2 + c2 )2 .
Thus z is a root of the polynomial
(X − 1)6 2 (X − 1)
6
f (X ) = (B(x) − c A(x))2 − A 2
(x)(s 2
− (x − b) ) ,
a4 a4
i.e.,
f (X ) = X 6 + a1 X 5 + a2 X 4 + a3 X 3 + a4 X 2 + a5 X + a6 ,
 Constructing the regular hendecagon 415
where

a1 = −2,
a2 = 1 − 2(s 2 − b2 + c2 + 2ab),
a3 = 4(s 2 − b2 + c2 + ab),
a4 = (s 2 − b2 + c2 + 2ab)2 − 2(s 2 − b2 + c2 ) − 4c2 (s 2 − (a − b)2 ),
a5 = 2 (ab)2 − 2(s 2 − b2 − c2 + ab)2 ,
a6 = (s 2 − b2 − c2 )2 .

Summarizing, we see that if z is the signed distance associated with a point of intersection
of the conchoid and circle given above, then z is a root of the polynomial f (X ) whose coef-
ficients a j satisfy the above equations. Hence f (X ) ∈ K [X ], where K = Q(a 2 , ab, c2 , s 2 ).
Notice, too, that by reversing the steps in the argument above we see that any real root of
f (X ) is a signed distance.
The polynomial f (X ) constructed above is defined as a verging polynomial; the field K
is the verging field; and the numbers a, b, c, and s are the verging parameters.
Now we want to find necessary conditions for when a monic sixth-degree polynomial is
a verging polynomial and when it is, how to determine the verging parameters. To this end,
suppose f (X ) = X 6 + a1 X 5 + a2 X 4 + a3 X 3 + a4 X 2 + a5 X + a6 ∈ R[X ] is a verging
polynomial with verging parameters a, b, c, s. Then clearly we must have a1 = −2. Since
the coefficients a j must have the form in the equations above, we see in particular that
1 a2
− ,
s 2 − b2 + c2 + 2ab =
2 2
a3
s − b + c + ab = .
2 2 2
4
Subtracting these two equations thus yields
m
ab = , where m = 2 − 2a2 − a3 .
4
Hence we can get b if we know a. From this expression and the formula for a3 above, we
see that
a3 −1 + a2 + a3
s 2 − b2 + c2 = − ab = .
4 2
Now, we consider

(2a6 + a5 )2 = (2(s 2 − b2 − c2 )2 + 2(ab)2 − 2(s 2 − b2 − c2 + ab)2 )2

= 4((ab)2 + (2(s 2 − b2 − c2 ) + ab)(−ab))2
= 16(s 2 − b2 − c2 )2 (ab)2 = a6 m 2

and thus
(2a6 + a5 )2 = a6 m 2 .

Hence, if we know a2 , a3 , a5 , then we know a6 at least up to two values:


m 2 − 4a5 ± m m 2 − 8a5
a6 = .
8
416 B ENJAMIN AND S NYDER
Also notice that if a6 satisfies the above equation, then clearly a6  0. Moreover, we claim
m − 8a5  0. This is clear if a6 = 0, for then a5 = 0 in which case m 2 − 8a5 = m 2  0. If
2

a6  0, then
(2a6 + a5 )2 (2a6 − a5 )2
m 2 − 8a5 = − 8a5 = .
a6 a6
Now consider a5 . We have
a5 = 2(ab)2 − 2(2c2 − (s 2 − b2 + c2 + ab))2
m2 1
= − (8c2 − a3 )2
8 8
whence
(8c2 − a3 )2 = m 2 − 8a5
and so

8c2 = a3 + ε m 2 − 8a5 ,
for ε = ±1. Thus, we must have

a3  −ε m 2 − 8a5 ,
in which case

1  
c = cε = a3 + ε m 2 − 8a5 .
8
Next, notice that
 
1 a2 2
a4 = − + 1 − a2 − a3 − 4c2 (s 2 − b2 + c2 + 2ab − a 2 − c2 )
2 2
 
1 a2 1 
= 4c4 − 4c2 − − a 2 + 1 − a2 − a3 + 1 − 2a2 + a22
2 2 4
2
5 3 a
= 4c4 + (4a 2 + 2a2 − 2)c2 + − a2 + 2 − a3 .
4 2 4
Also from above, we have
1 a 2 a3 a2 a3
a5 = −8c4 + 2a3 c2 + − a2 + 2 − + .
2 2 2 2
Thus
5 a2 a3
2a4 + a5 = (8a 2 + 4a2 + 2a3 − 4)c2 + 3 − 4a2 + a22 − a3 +
2 2
and therefore
(8a 2 − 2m)c2 = B,
where
5 a2 a3
B = 2a4 + a5 − 3 + 4a2 − a22 + a3 − .
2 2
We now assume, as before, that c > 0. From all this we see we must have
B
m>− ,
2cε2
 Constructing the regular hendecagon 417
in which case

m B
a = aε = + 2.
4 8cε
Thus
m
b = bε = ,
4aε
from above.
Finally,
a3
s 2 = −c2 + b2 − ab +
4
m2 1
= −c2 + 2
− (1 − a2 − a3 ).
16a 2
Thus we must have
m2
> 16cε2 + 8(1 − a2 − a3 ),
aε2
in which case

m2 1
s = sε = − (1 − a2 − a3 ) − cε2 .
16aε2 2
We are now in a position of stating our first main result on verging between a line and a
circle. To economize on the length of the theorem, we first isolate some properties:
Let
f (X ) = X 6 + a1 X 5 + a2 X 4 + a3 X 3 + a4 X 2 + a5 X + a6 ∈ R[X ].
We say f (x) satisfies (**), if its coefficients satisfy the following conditions:
a1 = −2;
a2 = 1 − 2(s 2 − b2 + c2 + 2ab);
a3 = 4(s 2 − b2 + c2 + ab);
a4 = (s 2 − b2 + c2 + 2ab)2 − 2(s 2 − b2 + c2 ) − 4c2 (s 2 − (a − b)2 );
a5 = 2 (ab)2 − 2(s 2 − b2 − c2 + ab)2 ;
a6 = (s 2 − b2 − c2 )2 .
T HEOREM 1 (Verging Theorem).
(i) Let L be the vertical line in R2 with equation x = a, a > 0, and C the circle centered
at (b, c), c > 0, with radius s. If P = (x, y) is a point of intersection of the conchoid
Con L ,(0,0) and the circle C, then the signed distance
x
z=
x −a
is a root of the polynomial
f (X ) = X 6 + a1 X 5 + a2 X 4 + a3 X 3 + a4 X 2 + a5 X + a6
satisfying the conditions (**) above. Hence f (X ) is a verging polynomial with ver-
ging field K = Q(a 2 , ab, c2 , s 2 ).
418 B ENJAMIN AND S NYDER
(ii) Now suppose
f (X ) = X 6 + a1 X 5 + a2 X 4 + a3 X 3 + a4 X 2 + a5 X + a6 ∈ R[X ]
is a polynomial and let
5 a2 a3
m = 2 − 2a2 − a3 , B = 2a4 + a5 − 3 + 4a2 − a22 + a3 − .
2 2
Moreover, suppose that the following conditions are satisfied:
a1 = −2 (4·2)

(2a6 + a5 )2 = a6 m 2 (4·3)

a3 > −ε m 2 − 8a5 (ε = 1, or − 1) (4·4)
in which case

1  
c = cε = a3 + ε m 2 − 8a5 ;
8
B
m>− (4·5)
2cε2
in which case

m B
a = aε = + 2
4 8cε
and whence
m
b = bε = ;
4aε
m2
> 16cε2 + 8(1 − a2 − a3 ) (4·6)
aε2
and then

m2 1
s = sε = 2
− (1 − a2 − a3 ) − cε2 .
16aε 2
Finally,
 2 2
sε − bε2 − cε2 = a6 . (4·7)
Then f (x) is a verging polynomial with verging parameters aε , bε , cε , and sε given
above.
Proof. Most of the proof has been carried out above. But one still needs to check that if
conditions (4·2)–(4·7) are satisfied, then from the parameters aε , bε , cε , sε given above we
retrieve by the conditions (**) our original coefficients a1 , . . . , a6 . As the reader may verify,
these parameters uniquely determine a2 , a3 , a4 , a5 , but (sε2 − bε2 − cε2 )2 is only determined up
to one of the solutions of condition (4·3) in the theorem. More precisely, one can show that

 2 
2 2 m 2 − 4a5 + ε m m 2 − 8a5
sε − bε − cε =
2
.
8
Thus we must check to see if this value is actually equal to a6 — hence the need for condition
(4·7) above.
 Constructing the regular hendecagon 419
We are now in a position to characterize the numbers in C, but we first define an RMC
tower, for convenience.
A real RMC tower of Q is a tower of fields {K j }nj=0 for some nonnegative integer n such
that
Q = K 0 ⊆ K 1 ⊆ · · · ⊆ K n ⊆ R,
where the degree [K j+1 : K j ]  6, and such that if [K j+1 : K j ] = 5 or 6, then K j+1 =
K j (z j+1 ) for some root z j+1 of a verging polynomial f (X ) ∈ K j [X ].

T HEOREM 2 (Characterization Theorem). A real number α is an RMC number, i.e., α ∈

C, if and only if there is a real RMC tower {K j }nj=0 of Q such that α ∈ K n .

Proof. Let P j = ( p j , q j ), for j = 1, . . . , m, be an RMC sequence of points. We claim

that there is a real RMC tower {K j }nj=0 of Q such that p j , q j ∈ K n for all j = 1, . . . , m. We
use induction on m. For m = 1, P1 = ( p1 , q1 ) must be in the starter set {(0, 0), (1, 0)} and
so p1 , q1 ∈ Q = K 0 , an RMC tower.
Now suppose m > 1 and that the statement above holds for any RMC sequence of less
than m points. Let P j = ( p j , q j ), for j = 1, . . . , m, be an RMC sequence of m points.
Hence by the inductive hypothesis, there is a real RMC tower K 0 , . . . , K n of Q such that
p j , q j ∈ K n for all j < m. We need to show that the p j , q j lie in some RMC tower for all
j  m. To this end, notice that Pm can be obtained in at least one of the five ways given in
Section 2.
(i)-(iii) In these cases, arguing as in the case of the classical straightedge and com-
pass shows that Pm has coordinates in an extension K n+1 of degree 1 or 2 over K n . Thus
K 0 , . . . , K n+1 is an RMC tower with p j , q j ∈ K n+1 , for all j  m.
(iv) This case is essentially the situation found in [3, chapter 9], also see [1], and involves
a tower of extensions of degree  4 of K n containing the coordinates of point Pm . This
extended tower is thus an RMC tower of Q.
(v) Finally, suppose Pm is obtained by verging through an earlier point V = P j , j < m,
between a line L, not containing V , passing through two earlier points and a circle C with
center (b, c) = ( pk , qk ) = Pk , k < m and passing through an earlier point P ,  < m. (If V
lies on L, then we are reduced to cases (ii)-(iii).) Notice that the square of the radius s is
s 2 = ( p − b)2 + (q − c)2 ∈ K n .
First, assume that Pm is a primary point. As we know from Section 3, Pm is a point of
intersection of the conchoid, Con L ,V , and the circle C.
Now let L ⊥ be the line through V perpendicular to L and let P = (u, v) be their point of
intersection. Since L and L ⊥ have equationswith coefficients in K n , we see that u, v ∈ K n .
Let a be the distance from V to P, i.e., a = (u − p j )2 + (v − q j )2 , and thus a 2 ∈ K n . We
now apply a rigid motion of the plane to move V to the origin and move L to the vertical
line x = a. Namely, let
(X  )t = a −1 A(X t − V t ),
where the t means transpose, X = (x, y), X  = (x  , y  ), and
 
α β
A= ,
−β α
420 B ENJAMIN AND S NYDER
with
α = u − pj, β = v − qj.
Notice that α, β ∈ K n .
Also notice that
X t = A−1 (a(X  )t ) + V t = a −1 At (X  )t + V t ,
since A−1 = (1/a 2 )At .
If O is any subset in the plane, then denote its image under the rigid motion by O . Thus
we have V  = (0, 0), L  : x = a, and the circle C  with radius s and center (b , c ) which
satisfies
(b , c )t = a −1 A((b, c)t − V t ),
and thus notice that ab and ac are in K n . Let Pm = ( pm , qm ), where
(Pm )t = a −1 A(Pmt − V t ).
We consider three cases according as c > 0, c = 0, c < 0.
Case 1. Suppose c > 0. Define
pm
zm = .
pm − a
Hence z m is the signed distance between V  = (0, 0) and Pm and thus by the Verging The-
orem, z m is a root of the polynomial f (X ) = X 6 + a1 X 5 + a2 X 4 + a3 X 3 + a4 X 2 + a5 X + a6
with the ai determined by the parameters a, b , c , s as in Equations (**) of the theorem. But
since s 2 , b2 , c2 , a 2 , ab ∈ K n , the ai are therefore in K n .
By our equations (4·1a ,b) earlier in this section, we have
az m z m3 − z m2 − (s 2 − b2 − c2 + 2ab )z m + s 2 − b2 − c2
pm = , qm = .
zm − 1 2c (z m − 1)
This implies that a −1 pm , a −1 qm ∈ K n (z m ).
But now notice that
Pmt = At (a −1 Pm )t + V t ,
which shows that pm , qm ∈ K n (z m ). Therefore, K 0 , . . . , K n+1 = K n (z m ) is a real RMC
tower of Q containing the coordinates of all m points P1 , . . . , Pm .
Case 2. Suppose c = 0. Let
K n+1 = K n (a), K n+2 = K n+1 ( pm ), K n+3 = K n+2 (qm ).
Then we have [K n+1 : K n ]  2, [K n+2 : K n+1 ]  3, as we saw earlier in this section, and
[K n+3 : K n+2 ]  2, since (qm ) ∈ K n+2 . But notice that
Pmt = At (a −1 Pm )t + V t
implies pm , qm ∈ K n+3 . Thus P1 , . . . , Pm are contained in the RMC tower K 0 , . . . , K n+3 .
Case 3. Suppose c < 0. Then reflect the plane through the x-axis and so any point (x  , y  )
is mapped to (x  , −y  ). We are thus reduced to Case 1.
Now assume Pm is the secondary point. If the primary point is not V , then the coordinates
of the secondary point lie in the same RMC tower as those of the primary point, since Pm
 Constructing the regular hendecagon 421
is the intersection of the line L with the line passing through V and the primary point. If,
on the other hand, the primary point is V , then the secondary point may be found as the
intersection of L and the circle of radius 1 centered at V . Hence the coordinates lie in a
quadratic extension of K n .
Now, we consider the converse. Suppose K 0 , . . . , K n is a real RMC tower of Q. We need
to show K n ⊆ C and to do so use induction on n  0. For n = 0, K n = Q ⊆ C.
Now suppose n > 0 and that the result holds for any RMC tower of length at most n − 1
and let K 0 , . . . , K n be any real RMC tower. Hence K n−1 ⊆ C. We consider the degree
[K n : K n−1 ].
If [K n : K n−1 ] = 1, then, of course, K n = K n−1 ⊆ C.
If [K n : K n−1 ] = 2, 3, or 4, then the elements in K n can be constructed (over K n−1 ) by a
marked ruler only; cf. [3, chapter 9].
Finally, if [K n : K n−1 ] = 5 or 6, then K n ⊆ C by the definitions of an RMC number and
tower.

5. A proof that 2 cos(2π/11) is an RMC number

Notice that to prove that a regular hendecagon is constructible by marked ruler and com-
pass, it suffices to prove that 2 cos(2π/11) is an RMC number. This we now do.
Let ζ = ζ11 = e2πi/11 , and define
α j = ζ j + ζ − j = 2 cos(2 jπ/11),
for any integer j. Let α = α1 . As is well known, Q(α)/Q is a Galois extension with cyclic
Galois group of order 5. (For properties of Galois extensions and related concepts, refer, for
example, to [2].) Moreover, the minimal polynomial of α over Q is given by

5
p(X ) = X 5 + X 4 − 4X 3 − 3X 2 + 3X + 1 = (X − α j ).
j=1

P ROPOSITION 1. If α (given above) is an RMC number, then there exists an RMC tower
K 0 , K 1 , . . . , K n such that K n = K n−1 (α) and [K n : K n−1 ] = 5.
Proof. Assuming α ∈ C, Theorem 2 implies that there is an RMC tower K 0 , . . . , K n , with
n minimal, say, such that α ∈ K n − K n−1 . Since Q(α)/Q is a Galois extension, by shifting
the base field to K n−1 , we see that K n−1 (α)/K n−1 is a Galois extension with Galois group
isomorphic to a subgroup of Gal(Q(α)/Q). But then we have
1 < [K n−1 (α) : K n−1 ] | [Q(α) : Q] = 5
and thus [K n−1 (α) : K n−1 ] = 5. Hence since
K n−1  K n−1 (α) ⊆ K n ,
we have 5 | [K n : K n−1 ]. But by the definition of an RMC tower, [K n : K n−1 ]  6. This
forces [K n : K n−1 ] = 5, as desired.
Here is how we will use this proposition.
C OROLLARY 1. The number α is an RMC number if and only if there exists a subfield K
of C with α ^ K , numbers a, b, c, s ∈ K with a, c, s > 0, and a point
(x, y) = P ∈ Con x=a,(0,0)  C(b,c),s ,
422 B ENJAMIN AND S NYDER
such that the signed distance z = x/(x − a) from the origin to P satisfies
K (α) = K (z).
(Notice that when this happens
z = u 1 α1 + u 2 α2 + u 3 α3 + u 4 α4 + u 5 α5
for uniquely determined u j ∈ K .)

Proof. The proof is an immediate consequence of the definition of RMC numbers, The-
orem 2, and Proposition 1 in which we take K = K n−1 . (The parenthetical remark follows
from the observation that α1 , . . . , α5 form a basis of Q(α)/Q.)

Now we are in a position of showing that α is an RMC number. Set z = 5j=1 u j α j and
try to exhibit u 1 , . . . , u 5 so that z is the root of a verging polynomial and that the u j lie in a
subfield of C not containing α. This in general can be a daunting task. Hence we try to pick
the simplest case in which all but one of the u j vanish. If that does not work, then we could
assume that all but two of the u j vanish. Of course, the computations get more complicated
as fewer u j vanish.
With this in mind we assume that
u1 = u and u 2 = u 3 = u 4 = u 5 = 0,
and set z = uα. Also, let z j = uα j . Let q(X ) be the minimal polynomial of z over K =
Q(u). Hence

5
q(X ) = (X − z j ) = X 5 + b1 X 4 + · · · + b5 ,
j=1

where
b1 = u, b2 = −4u 2 , b3 = −3u 3 , b4 = 3u 4 , b5 = u 5 .

By the Verging Theorem, if z = uα is a signed distance, then there exists a verging

polynomial
f (X ) = X 6 + a1 X 5 + · · · + a6 , a j ∈ K , a1 = −2,
with z as a root and with K = Q(u) the verging field. Hence q(X ) | f (X ) in K [X ] and
consequently there is an element η ∈ K such that
f (X ) = (X − η)q(X ).
Multiplying out we have
−2 = a1 = b1 − η, a2 = b2 − ηb1 , a3 = b3 − ηb2 ,
a4 = b4 − ηb3 , a5 = b5 − ηb4 , a6 = −ηb5 .
But then
η = 2 + u, a2 = −u(5u + 2), a3 = u 2 (u + 8),
a4 = 6u 3 (u + 1), a5 = −2u 4 (u + 3), a6 = −u 5 (u + 2).
 Constructing the regular hendecagon 423
We now need to find a value of u so that f (X ) is a verging polynomial; this means that
f (X ) must satisfy conditions (4·2)–(4·7) of the Verging Theorem. For use below, notice that
m and B defined in the Verging Theorem can be given in terms of u:
1 5 5
m = −u 3 + 2u 2 + 4u + 2, B= u + 2u 4 + u 3 − 4u 2 − 8u − 3.
2 2
Since we are already assuming a1 = −2, condition (4·2) is satisfied.
Writing condition (4·3) as (2a6 + a5 )2 − m 2 a6 = 0, we have
g(u) = 5u 12 + 22u 11 + 48u 10 + 76u 9 + 84u 8 + 64u 7 + 36u 6 + 8u 5 = 0.
Now a “miracle” occurs:
g(u) = u 5 (5u + 2)(u 3 + 2u 2 + 2u + 2)2 ,
where the cubic factor is irreducible over Q (2-Eisenstein) with one real root. The real roots
of g(u) are
 
2 1 3 √ 3 √
u = 0, u = − , u = −2 + −17 + 297 + −17 − 297 ≈ −1.544.
5 3
Clearly u = 0 is of no interest.
For u = −2/5, it turns out that condition (4·5) is not satisfied.
Now let u be the third root; call it u 0 for brevity. In this case, we have the following
approximate values (exact values in terms of square and cube roots can also be given):
a2 ≈ −8.828, a3 ≈ 15.385, a4 = 12, a5 ≈ −16.540, a6 = 4,

m ≈ 4.270, B ≈ −2.405, and m 2 − 8a5 ≈ 12.270.
Conditions (4·3)–(4·6) turn out to be satisfied for both ε = 1 and ε = −1. By some
calculations we get the following approximations of the verging parameters:
From condition (4·4), we have
c+ ≈ 1.860, c− ≈ 0.624;
condition (4·5) yields
a+ ≈ 0.990, b+ ≈ 1.078, a− ≈ 0.544, b− ≈ 1.963;
and condition (4·6) gives
s+ ≈ 0.696, s− ≈ 2.499.
Now, with respect to condition (4·7), we have
(s+2 − b+
2
− c+ ) ≈ 17.097  a6 ,
2 2
whereas (s−2 − b−
2
− c− ) ≈ 4 = a6 .
2 2

Since one of these two cases gives the value of a6 , it must be the second and so ε = −1.
Thus z = u 0 α is an RMC number with verging parameters a− , b− , c− , s− , and since u 0 is a
marked ruler number, hence also RMC, we conclude that α itself is an RMC number.
We have thus proved

T HEOREM 3 (Main Theorem). A regular hendecagon can be constructed by marked

ruler and compass.
424 B ENJAMIN AND S NYDER
Acknowledgment. We would like to take this opportunity to thank Dr. Oscar Lanzi
for his interest in some of our work and especially for prompting us to look at the regular
hendecagon.

REFERENCES
[1] A. BARAGAR . Constructions Using a Compass and Twice-Notched Straightedge. MAA Monthly 109
(2002), 151–164.
[2] D. C OX . Galois Theory (Wiley, 2003).
[3] G. E. M ARTIN . Geometric Constructions (Springer, 1998).