The Regular Heptagon by Angle Trisection

and Other Constructions

Paul Yiu
Department of Mathematics
Florida Atlantic University

April 24, 2007

6
0 1 2 c−1 c
Contents
1 Which regular n-gons are constructible? . . . . . . . . . . . . . . . . . 3
2 Gauss’ Construction of the regular 17-gon . . . . . . . . . . . . . . . . 10
2.1 An explicit construction of a regular 17-gon . . . . . . . . . . . . . . . . 18
3 Angle trisection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1 A false trisection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Constructions with a marked ruler . . . . . . . . . . . . . . . . . . . . . 22
3.3 The quadratrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 The trisectrix (MacLaurin) . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 A recent trisection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.6 Angle trisection by paper folding . . . . . . . . . . . . . . . . . . . . . . 35
4 Archimedes’ construction of the regular heptagon (Arabic tradition) . . . 37
4.1 Archimedes’ construction . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 The heptagonal triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Division of segment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5 Construction of regular 7-gon by angle trisection . . . . . . . . . . . . . 44
6 Euler’s problem of construction of ABC from OHI . . . . . . . . . . . 47
6.1 Euler’s fundamental results . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Euler’s examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.3 Conic solution of Euler’s problem . . . . . . . . . . . . . . . . . . . . . 53

2
1 Which regular n-gons are constructible?
ζ2
ζ

ζ3

1
0

ζ4

ζ6
ζ5

ζ k , k = 1, 2, . . . , 6, are the roots of

x6 + x5 + x4 + x3 + x2 + x + 1 = 0.
These are constructible with ruler and compass if and
only if y = x + x1 = 2 cos k · 2π
7
are.

y 3 + y 2 − 2y − 1 = 0.

3
Theorem 1. Let F be a field.
The roots of a cubic polynomial p(x) ∈ F [x] are con-
structible by ruler and compass from F if and only if
p(x) has a root in F .
ζ2
ζ

ζ3

0
y3 y2 y1
1
2 2 2

ζ4

ζ6
ζ5

Corollary 2. The regular 7-gon is not constructible by

ruler and compass.
Proof. The cubic polynomial y 3 + y 2 − 2y − 1 has no
rational root, (the only possible candidates being ±1).

4
More examples
√
1. 3 2 cannot be constructed by ruler and compass.

Proof. x3 − 2 does not have a rational root.

2. The 60◦ angle cannot be trisected by ruler and com-

pass.

Proof. Since cos 60◦ = 12 , the trisection of 60◦ is equiv-

alent to the solution of cos 3θ = 12 . With x = cos θ,
this is
1
4x3 − 3x = ,
2
or 8x3 − 6x − 1 = 0. It is clear that this does not
have rational roots. This shows that 60 ◦ cannot be
trisected with ruler and compass.

5
Theorem 3. A regular n-gon is constructible if and only


if n = 2a j pj where
(i) a ≥ 0,
k
(ii) if j ≥ 1, each pj is a Fermat prime of the form 2 2 +1,
(iii) if j ≥ 2, the Fermat primes p j are all distinct.
Proof. (1) If a regular n-gon is constructible, then so is a
regular 2n-gon.
(2) If a regular n-gon is constructible, and m|n, then a
regular m-gon is constructible.
(3) If p is an odd prime and a regular p-gon is con-
structible, then p − 1 is a power of 2.
(4) If p is an odd prime of the form 2 k + 1, then a
regular p-gon is constructible (Gauss).

6
 (5) If gcd(m, n) = 1 and regular m- and regular n-
gons are constructible, then so is a regular mn-gon.
Find integers a and b such that am + bn = 1, and note that
2π 2π 2π
=b· +a· .
mn m n

It remains to show why the Fermat primes may not

repeat.
(6) Let p be an odd prime divisor of n. If a regular
n-gon is constructible, then p 2 cannot divide n.

7
 A regular n-gon is constructible if and only if the num-
bers tan k·2π
n , k = 1, 2, . . . , n−1, are constructible. These
are the roots of the polynomial
  
n
Pn(x) = (−1)j−1 x2j .
2j≤n−1
2j + 1
Proof.
 j n
  2j+1
2j+1≤n (−1) 2j+1 tan θ
tan nθ =   n  2j
j
2j≤n (−1) 2j tan θ
  n  2j
tan θ 2j+1≤n(−1)j 2j+1 tan θ
=    .
(−1) j n tan2j θ
2j≤n 2j

If n is odd, we may take P n to be a monic polynomial

(with leading coefficient 1).

8
 (6) Let p be an odd prime divisor of n. If a regular
n-gon is constructible, then p 2 cannot divide n.
Proof. Pp2 (x) = Pp(x)Q(x) for a polynomial Q(x).
(i) Q(x) ∈ Z, and is monic.
(ii) Each coefficient of Q(x), apart from the leading
one, is divisible by p.
(iii) The constant term is not divisible by p 2.
Therefore, Q(x) is irreducible by Eisenstein’s crite-
rion.
Example: For p = 3, P3 (x) = x2 − 3,

P9 (x) = x8 − 36x6 + 126x4 − 84x2 + 9,

P9 (x)
Q(x) = P3 (x) = x6 − 33x4 + 27x − 3.

Since deg Q = p2 −p = p(p−1) is not a power of 2, the

roots of Q(x) are not ruler-and-compass constructible.

9
2 Gauss’ Construction of the regular 17-gon
ζ5 ζ4
ζ3
ζ6
ζ2
ζ7
ζ
ζ8
1
0
ζ9
ζ 16
ζ 10
ζ 15
ζ 11
ζ 12 ζ 14
ζ 13

First entry of Gauss’ diary:

[1] The principles upon which the division of the

circle depend, and geometrical divisibility of the
same into seventeen parts, etc.
[1796] March 30 Brunswick

The details appeared in Section VII of his

Disquisitiones Arithmeticae (1801).

10
 ζ5 ζ4
ζ3
ζ6
ζ2

ζ7

ζ8

1
0
ζ9

ζ 16

ζ 10

ζ 15
ζ 11
ζ 14
ζ 12 ζ 13

ζ, ζ 2, . . . , ζ 15, ζ 16 are the roots of the equation

x17 − 1
= x16 + x15 + · · · + x + 1 = 0.
x−1

11
 Gauss’ construction depends on two key ideas.
(1) Solution of quadratic equations by ruler and com-
pass:
B

A M
b b

a a

K b A

(a) x2 = ax + b2 : (b) x2 + b2 = ax :
x1 = KA, x2 = −KB x1 = AM, x2 = BM

12
 (2) Gauss observed that, modulo 17, all the residues
are powers of 3 (which is a primitive root for the prime
17):

k 0 2 4 6 8 10 12 14
3k 1 9 13 15 16 8 4 2

k 1 3 5 7 9 11 13 15
3k 3 10 5 11 14 7 12 6
If we take
y1 = ζ + ζ 9 + ζ 13 + ζ 15 + ζ 16 + ζ 8 + ζ 4 + ζ 2,
y2 = ζ 3 + ζ 10 + ζ 5 + ζ 11 + ζ 14 + ζ 7 + ζ 12 + ζ 6,
then y1 + y2 = −1. More importantly,

13
the product y1 y2 does not depend on the choice of ζ.
y1 = ζ + ζ 9 + ζ 13 + ζ 15 + ζ 16 + ζ 8 + ζ 4 + ζ 2 ,
y2 = ζ 3 + ζ 10 + ζ 5 + ζ 11 + ζ 14 + ζ 7 + ζ 12 + ζ 6 .

This table gives ζ h · ζ k = ζ m, m ≡ h + k (mod 17).

1 9 13 15 16 8 4 2
3 4 12 16 1 2 11 7 5
10 11 2 6 8 9 1 14 12
5 6 14 1 3 4 13 9 7
11 12 3 7 9 10 2 15 13
14 15 6 10 12 13 5 1 16
7 8 16 3 5 6 15 11 9
12 13 4 8 10 11 3 16 14
6 7 15 2 4 5 14 10 8
Each of 1, 2, . . . , 16 appears exactly 4 times.

14
 y1 + y2 = ζ + ζ 2 + · · · + ζ 16 = −1,
y1y2 = 4(ζ + ζ 2 + · · · + ζ 16) = −4.
It follows that y1 and y2 are the roots of
y 2 + y − 4 = 0,
and are constructible. We may take
√ √
−1 + 17 −1 − 17
y1 = , y2 = .
2 2

15
Recall
y1 = ζ + ζ 9 + ζ 13 + ζ 15 + ζ 16 + ζ 8 + ζ 4 + ζ 2,
y2 = ζ 3 + ζ 10 + ζ 5 + ζ 11 + ζ 14 + ζ 7 + ζ 12 + ζ 6.
Now separate each of y 1 , y2 into two “groups” of four:
z1 = ζ + ζ 13 + ζ 16 + ζ 4,
z2 = ζ 9 + ζ 15 + ζ 8 + ζ 2;
z3 = ζ 3 + ζ 5 + ζ 14 + ζ 12,
z4 = ζ 10 + ζ 11 + ζ 7 + ζ 6 .
Then,
z1z2 = (ζ + ζ 13 + ζ 16 + ζ 4)(ζ 9 + ζ 15 + ζ 8 + ζ 2 )
= ζ + ζ 2 + · · · + ζ 16
= −1;
z3z4 = −1.

16
 Numbers quadratic equation
z1 , z2 z 2 − y1 z − 1 = 0
z3 , z4 z 2 − y2 z − 1 = 0
Finally, further separating z 1 = ζ + ζ 13 + ζ 16 + ζ 4 into
t1 = ζ + ζ 16, t2 = ζ 13 + ζ 4,
we obtain
t1 + t2 = z1,
t1t2 = (ζ + ζ 16)(ζ 13 + ζ 4 )
= ζ 14 + ζ 5 + ζ 12 + ζ 3
= z3.
Therefore, t1 = ζ + ζ 16 and t2 = ζ 13 + ζ 4 are the roots of
t2 − z1t + z3 = 0,
and are constructible.
17
2.1 An explicit construction of a regular 17-gon
  
2π 1 16
 1 √ √
cos = ζ +ζ = 17 − 1 + 34 − 2 17
17 2 16
 
1 √ √ √
+ 17 + 3 17 − 34 − 2 17 − 2 34 + 2 17.
8

Circle (O) with perpendicular diameters P Q and RS.

(1) A on OR with OA = 14 OP .

(2) Construct the bisectors OB and OC of angle OAP .

(3) D = OP ∩ C(A) and E = OP ∩ B(A).

Q P
C O D B E

18
(4) M = midpoint M of QD.

(5) F = OS ∩ M (Q).

(6) G on semicircle on OE, with OG = OF .

(7) H = OP ∩ E(G).

(8) P1 H ⊥ OP .
R

P1
A

Q P
M C O D B E H

F
G

19
3 Angle trisection

3.1 A false trisection

O D B

20
 A

x
2
E

θ y x
2θ
O 1 D B

Equal areas ⇒y sin 2θ = 2y sin θ

⇒ cos θ = 1 !!!
In this example, cos AOB = 25 , ∠AOB ≈ 66.422◦,
1
2θ − ∠AOB < 0.142◦.
3

21
3.2 Constructions with a marked ruler

(1) Pass a line through O such that the intercept between

the parallel and the perpendicular at A to the line OB is
2 · OA. Then this line is a trisector of angle AOB.

A E

2 · OA

O B

22
Proof.

A E

O B

23
(2) Archimedes [Book of Lemmas, Proposition 8]

Construct a line through A such that the intercept be-

tween the circle and the line BO has the same length as
the radius of the circle.
Then ∠AOC = 13 ∠AOB.

A A
C

24
3. The use of conics

B A
M P

If ∠AOB = 3∠AOC, then

AB : AC = BK : CK = BM : P M.
Since AB = 2 · BM , we have AC = 2 · P M .
Therefore, C lies on the hyperbola with focus A, di-
rectrix OM , and eccentricity 2.
25
 AC = 2 · P M ⇒ C lies on the hyperbola with
focus A, directrix OM , and eccentricity 2.

B A
M P

26
3.3 The quadratrix

A horizontal line HK (with initial position AB) falls vertically,

and a radius OP (with initial position OA) rotates about O,

both uniformly and arrive at OC at the same time.

The locus of the intersection Q = HK ∩ OP is the quadratrix.

A B

Q
H K

O C

27
 B

K
Q

A
O

To trisect angle AOB, let OB intersect the quadratrix

at P . Trisect the segment OP at K. Construct the paral-
lel through K to OA to intersect the quadratrix at Q.
Then OQ is a trisector of angle AOB.

28
3.4 The trisectrix (MacLaurin)

A B
O

T = AP ∩ perpendicular bisector of OA.

AT = QP .
The locus of Q is the trisectrix.

29
 P

A B
O

OP is a trisector of angle QOB.

30
 P

A B
O

31
3.5 A recent trisection

D. A. Brooks, A new method of trisection, College Math. Journal, 38 (2007) 78–81.

Given an acute angle AOB, let M be the midpoint of OA. Construct (i) the circle, center

M , passing through A, (ii) the perpendicular M Q from M to OB. If P is a point on M Q

such that the tangent from P to the circle makes a right angle with OP , then OP is a

trisector of angle OAB.

Q
P

O A
M

32
 Consider a triangle ABC with points X, Y , Z on the
side lines BC, CA, AB respectively.
Theorem 4 (Menelaus). The points X, Y , Z are collinear
if and only if
BX CY AZ
· · = −1.
XC Y A ZB
A

X B C

33
 B

Q
P

2θ M
O A
K 2θ

Proof. Let ∠P OA = ∠OAT = 2θ. We show that ∠HP T = θ. It will follow that
∠BOP = θ, and OP is a trisector of angle AOB.
Assume unit radius for the circle. Applying Menelaus’ theorem to triangle M HA
HT AK M P
with transversal T KP , we have · · = −1. Therefore,
T A KM P H

HT P H KM OA AK − AM sec 2θ − 1
=− · =− · = 2· = 2(1−cos 2θ) = 4 sin2 θ.
TA M P AK MO AK sec 2θ
Since T A = 1, we have T H = 4 sin2 θ. Now, P T = 2 sin 2θ = 4 sin θ cos θ. This
T = tan θ.
gives tan HP T = TPH

34
3.6 Angle trisection by paper folding

B. Casselman, If Euclid had been Japanese, Notices of AMS, 54

(2007) May issue, 626–628.

A M B A M B

P

P Q P

R

D

R S R

D C D C

35
(2) Cubic root of 2 by paper folding

A paper square ABCD is divided into three strips of

equal area by the parallel lines P Q and RS. The square
is then folded so that C falls on AD and S √ falls on P Q

(as C  in the second diagram). Then CACD = 2.
3

A B A B

P Q P
S

C

R S R

D C D C

36
4 Archimedes’ construction of the regular heptagon
(Arabic tradition)

4.1 Archimedes’ construction

Construct a segment AB with division points H and K

such that

b2 = a(c + a), c2 = (a + b)b.

K H
B A
b a c

37
 K H
B A
b a c

B A
K H

38
4.2 The heptagonal triangle

A : B : C = 1 : 2 : 4.
C

a
b

B A
c

39
 C

H
B A
b K a c

b2 = a(c + a), c2 = (a + b)b.

Proof. Applications of Ptolemy’s theorem.

40
 a

a c

b2 = a(c + a).

41
 b

a b

c2 = (a + b)b.

42
4.3 Division of segment
Let BHP Q be a square, with one side BH sufficiently extended.

Draw the diagonal BP . Place a ruler through Q, intersecting the diagonal BP at T ,

and the side HP at E, and the line BH at A such that the triangles AHE and T P Q have

equal areas. Then,

b2 = a(c + a), c2 = (a + b)b.

B K H A

b a c

Q L P

GSP

43
5 Construction of regular 7-gon by angle trisection

y 3 + py + q = 0 with discriminant

p3 q 2
D := + .
27 4
The cubic equation y 3 + py + q = 0 has
(i) three distinct real roots if D < 0,
(ii) three real roots with one of multiplicity ≥ 2 if D = 0,
(iii) one real root and two imaginary roots if D > 0.
The geometric solution of a nonsingular cubic equa-
tion reduces to one of the two problems:
(1) extraction of a real cube root if D > 0;
(2) angle trisection: if D < 0, then the equation
y 3 + py + q = 0 can be converted into
cos 3θ = f (p, q).

44
Construction of regular 7-gon by angle trisection

x6 + x5 + x4 + x3 + x2 + x + 1 = 0.

transformation equation
y = x + x1 y 3 + y 2 − 2y − 1 = 0

t−1
y= 3
t3 − 21t − 7 = 0

t = r cos θ r3 cos3 θ − 21r cos θ = 7

√ 1
t = 2 7 cos θ cos 3θ = √
2 7

45
 √
2π 2 7 cos θ − 1 1
cos = , cos 3θ = √ .
7 6 2 7

θ
6
0 1 2 c−1 c

46
6 Euler’s problem of construction of ABC from OHI

6.1 Euler’s fundamental results

I
O

B C

47
Euler line

O, H, and G are collinear, and HG : GO = 2 : 1.

A

O
G

B C

48
Euler’s formula

OI 2 = R(R − 2r).

O
d
I

r R

B C

49
Condition for existence

There is a triangle ABC with given O, H, I if and only if

I lies in the interior of the orthocentroidal disk (except
the midpoint of OH).
A

I O
G

B C

50
6.2 Euler’s examples

Given O, H, and I, Euler found a cubic equation whose

roots are the lengths of the sides of triangle ABC.
(1) HI 2 = 3, IO2 = 2, HO2 = 9.
The sides of the triangle are the roots of
3
√ 2 √
z − 71z + 22z − 2 71 = 0.
Euler noted that the roots of this equation is equivalent
to the trisection of the angle

71
α = arccos ≈ 41◦530 .
125
The side lengths are
√ √ √ √
71 2 5 1 71 2 5 1
− · cos α and + · cos(60◦ ± α).
3 3 3 3 3 3

51
 (2) Euler also showed that if OI = IH, the cubic poly-
nomial factors. This means that the triangle can be con-
structed by ruler and compass.

52
6.3 Conic solution of Euler’s problem

Given O, H, I, the circumcircle and the incircle of the

required triangle can be constructed, making use of
(1) Euler’s formula OI 2 = R(R − 2r),
(2) Feuerbach theorem: The nine-point circle (through
the midpoints of the sides and with center the midpoint
of OH) is tangent internally to the incircle.
A

Fe

I
O
NG
H

B C

53
Circumradius

1
OI 2 = R(R − 2r) and IN = (R − 2r)
2
2
⇒OI = 2R · IN.

2R

O N H

54
Circumcircle and incircle

2R

O N H

55
 Thus, given O, H, I, we can have constructed the cir-
cumcircle and the incircle.

I
O
N H

In fact, starting with an arbitrary point A on (O), by

drawing tangents, we can complete a triangle ABC with
incircle (I) and circumcircle (O).

56
 A

I
O H
N
H

The orthocenter, however, is not always correct!

GSP

57
 Note that ∠HAI = ∠OAI. If we put
O = (0, 0), H = (k, 0), I = (p, q),
the locus of point P for which ∠HP I = ∠OP I is the
curve

F :=2qx3 − (2p − k)x2y + 2qxy 2 − (2p − k)y 3

− 2(p + k)qx2 + 2(p2 − q 2 )xy + 2(p − k)qy 2
+ 2kpqx − k(p2 − q 2)y
=0.
Note that F (k, 0) = 0, i.e., F contains the point H.
The circumcircle has equation

2 2 (p2 + q 2 )2
G := x + y − .
(2p − k)2 + 4q 2

58
 We find a linear function L such that F − L · G has no
third degree terms. With L = 2qx − (2p − k)y − 2qk, we
have
F − L · G = E,
where
E := − 2pqx2 + 2(p2 − q 2 )xy + 2pqy 2
k2 (k − 4p)(p2 − q 2 ) + k(3p2 − 5q 2 )(p2 + q 2 ) + 2p(p2 + q 2 )2
− ·x
(2p − k)2 + 4q 2
2q(kp((2p − k)2 + 4q 2 ) + (p2 + q 2 )2 )
+ ·y
(2p − k)2 + 4q 2
2k(p2 + q 2 )2 q
− .
(2p − k)2 + 4q 2

The quadratic part factors as −2(px + qy)(qx − py).

This means that E represents a rectangular hyperbola
with asympotes with slopes pq and − pq .
Note that L represents a line through H = (k, 0).
Therefore, the conic E = 0 also contains H.

59
 The center of the rectangular hyperbola is a point which
can be described geometrically as follows.
T
I M
O H

T := intersection of tangents of circumcircle of OHI at O and H,

T I intersects the circle again at Q,

M is a trisection point of IH,

P divides M Q in the ratio 1 : 3.

60
 T
I
O H

61
The rectangular hyperbola through H

T
I
O H

62
The circumcircle and incircle

T
I
O
H
P

63
 A

T
I
O
H

64
 A

T
I
O
H

65