A new proof of Morleys theorem
par Alain Connes
It is now 22 years since IHES oered me its hos- understood the truth of the statement of J. Hadamard
pitality. I have learnt there most of the mathematics on the depth of the mathematical concepts coming
I know, mostly thanks to impromptu lunch conversa- from physics:
tions with visitors or permanent members. Not this short lived novelty which can too often
When I arrived, I was engrossed in my own work only influence the mathematician left to his own de-
and had the humbling experience of finding out how vices, but this infinitely fecund novelty which springs
little I understood of what was currently discussed. from the nature of things.
Dennis Sullivan took care of me, and gave me a In order to give some flavor of the atmosphere of
crash course in geometry which influenced the way friendly competition caracteristic of IHES, I have cho-
I thought for the rest of my life. sen a specific example of a lunch conversation of this
It is also in Bures, thanks to the physicists that I last spring which led me to an amusing new result.
Figure 1.
Around 1899, F. Morley proved a remarkable the- It was the first time I heard about Morleys result
orem on the elementary geometry of Euclidean trian- and when I came back home, following one of the ad-
gles: vices of Littlewood, I began to look for a proof, not
in books but in my head. My only motivation besides
Given a triangle A, B, C the pairwise intersec-
curiosity was the obvious challenge This is one of
tions , , of the trisectors form the vertices of an
the rare achievements of Bonaparte I should be able
equilateral triangle (cf. Fig. 1).
to compete with. After a few unsuccessful attempts
One of us mentioned this result at lunch, and I quickly realized that the intersections of consecutive
(wrongly) attributed it to Napoleon. Bonaparte had trisectors are the fixed points of pairwise products of
indeed studied mathematics at an early age, and, be- rotations gi around the vertices of the triangle (with
sides learning English, was teaching mathematics to angles two thirds of the corresponding angles of the
the son of Las Cases during his St Helens exile in triangle). It was thus natural to look for the three fold
Longwood. symmetry g of the equilateral triangle as an element of
*
Ce texte provient de louvrage fetant les 40 ans de lIHES intitule Les relations entre les mathematiques et la physique theorique,
IHES, 1998, pp. 43-46
**
College de France, Paris, et IHES, 91440 Bures-sur-Yvette, France.
2 Quadrature n 47
the group generated by the three rotations gi . Now, the translational part of g31 g32 g33 . The first condition is
it is easy to construct an example (in spherical geom- exactly j3 = 1. Note that j 1 by hypothesis. Next
etry) showing that Morleys theorem does not hold one has
in NonEuclidean geometry, so that the proof should
make use of special Euclidean properties of the group b = a21 + a1 + 1 b1 + a31 a22 + a2 + 1 b2
of isometries. + (a1 a2 )3 a23 + a3 + 1 b3 . (4)
I thus spent some time trying to find a formula for
g in terms of gi , using the easy construction (any isom- A straightforward computation, using a1 a2 a3 = j
etry with angle 2/n, n 2 is automatically of or- gives,
der n), of plenty of elements of order 3 in the group , b = ja21 a2 (a1 j)(a2 j)(a3 j) + j + j2 ,
such as g1 g2 g3 . After much eort I realized that this
(5)
was in vain (cf. Rem. 2 below) and that the relevant
group is the ane group of the line, instead of the where, , , are the fixed points
isometry group of the plane. a1 b2 + b1 a2 b3 + b2 a3 b1 + b3
The purpose of this short note is to give a concep- = , = , =
1 a1 a2 1 a2 a3 1 a3 a1
tual proof of Morleys theorem as a group theoretic (6)
property of the action of the ane group on the line. It
will be valid for any (commutative) field k (with arbi- Now, ak j 0 since by hypothesis the pairwise prod-
trary characteristic, though in characteristic 3 the hy- ucts of g j s are not translations. Thus, and whatever
pothesis of the theorem cannot be fulfilled). Thus we the characteristic of k is, we get that a) b).
let k be such a field and G be the ane group over k,
a b Corollary. Morleys theorem.
in other words the group of 2 2 matrices g =
0 1
where a k, a 0, b k. For g G we let, Proof. Take k = C and let g1 be the rotation with cen-
ter A and angle 2a, where 3a is the angle BAC and
(g) = a k . (1) similarly for g2 , g3 . One has g31 g32 g33 = 1 since each
g3i can be expressed as the product of the symmetries
By construction is a morphism from G to the multi- along the consecutive sides. Moreover for a similar
plicative group k of non zero elements of k, and the reason = fix(g1 g2 ), = fix(g2 g3 ), y = fix(g3 g1 ) are
subgroup T = Ker is the group of translations, i.e. the intersections of trisectors. Thus from a) b) one
the additive group of k. Each g G defines a transfor- gets + j + j2 = 0 which is a classical characteri-
mation, zation of equilateral triangles.
g(x) = ax + b x k, (2) Remark 1. Without altering the cubes g31 , g32 , g33 one
can multiply each gi by a cubic root of 1, one obtains
and if a 1, it admits one and only one fixed point, in this way the 18 nondegenerate equilateral triangles
of variants of Morleys theorem.
b
fix (g) = (3)
1a Remark 2. We shall now show that in general the ro-
Let us prove the following simple fact: tation g which permutes cyclically the points , ,
does not belong to the subgroup of G generated by
Theorem. Let g1 , g2 , g3 G be such that g1 g2 , g1 , g2 , g3 . Under the hypothesis of the theorem, we
g2 g3 , g3 g1 and g1 g2 g3 are not translations and let can assume that the field k contains a non trivial cubic
j = (g1 g2 g3 ). The following two conditions are root of unity, j 1, and hence that its characteristic is
equivalent, not equal to 3. The rotation which permutes cyclically
the points , , is thus the element of G given by,
a) g31 g32 g33 = 1.
j b
g= , 3b = (1 j) ( + + ). (7)
0 1
b) j3 = 1 and + j+ j2 = 0 where = fix(g1 g2 ),
a b
= fix(g2 g3 ), = fix(g3 g1 ). Now for any element g = , of the group gen-
0 1
erated by g1 , g2 , g3 , one has Laurent polynomials Pi ,
ai bi in the variables a j such that,
Proof. Let gi = . The equality g31 g32 g33 = 1 is
0 1
equivalent to (g31 g32 g33 ) = 1, and b = 0, where b is b = b1 P1 + b2 P2 + b3 P3 . (8)
Hiver 2002-2003 3
Thus, expressing, with the above notations, bi in terms We can thus assume that we have found Laurent poly-
of , , , nomials Qi such that for any a1 , a2 , a3 k , with
a1 a2 a3 = j, and any , , k with + j + j2 = 0,
b1 = (1 + j)1 a1
3 (a3 j) (9)
the following identity holds,
(a1 j) + a1 (a2 j))
b2 = (1 + j)1 (a2 (a3 j)
(1 j)( + + ) = 3 ((a3 j)Q1
+a1
1 (a1 j) (a2 j) +(a1 j)Q2 + (a2 j)Q3 ) . (11)
b3 = (1 + j)1 ((a3 j)
+a3 (a1 j) + a1
2 (a2 j) , We then choose a1 = j, a2 = j, a3 = j2 , = 0,
= j, = 1, and get a contradiction. Passing to a
we get Laurent polynomials Qi such that,
function field over k, this is enough to show that, in
b = (a3 j) Q1 + (a1 j)Q2 + (a2 j)Q3 . (10) general, g .
**********************
