On Kleins icosahedral solution of the quintic

Oliver Nash
February 5, 2012
Abstract
We present an exposition of the icosahedral solution of the quintic
equation rst described in the classic work [17]. Although we are
heavily inuenced by [17] we follow a slightly dierent approach which
enables us to arrive at the solution more directly.
1 Introduction
In 1858, Hermite published a solution of the quintic equation using modular
functions [12]. His work received considerable attention at the time and
shortly afterward Kronecker and Brioschi also published solutions, but it
was not till Kleins seminal work [17] in 1884 that a comprehensive study of
the ideas was provided.
Although there is no modern work covering all of the material in [17],
there are several noteworthy presentations of some of the main ideas. These
include an old classic of Dickson [4] as well Slodowys article [28] and the
helpful introduction he provides in the reprinted edition [18] of [17]. In
addition Kleins solution is discussed in both [31], [32] as well as [15]. Finally
the geometry is outlined briey in [22] and a very detailed study of a slightly
dierent approach is presented in the book [26].
Perhaps surprisingly, we believe there is room for a further exposition
of the quintics icosahedral solution. For one thing, all existing discussions
arrive at the solution of the quintic indirectly as a result of rst studying
quintic resolvents of the icosahedral eld extension. Even Klein admits that
he arrives at the solution somewhat incidentally
1
and each of the accounts
listed above, except [22] and [26], exactly follow in Kleins footsteps. In
addition, we believe the icosahedral solution deserves a short, self-contained
account.
1
His words in the original German are gewissermassen zufalligerweise.
1
We thus follow Klein closely but take a direct approach to the solution of
the quintic, bypassing the study of resolvents of the icosahedral eld exten-
sion. In fact our approach is closely related to Gordons work [9] and indeed
Klein discusses the connection but, having already achieved his goal by other
means, he contents himself with an outline.
The direct approach enables us to present the solution rather more con-
cisely than elsewhere and we hope this may render it more accessible; part
of our motivation for writing these notes was provided by [20]. In addition
our derivation of the icosahedral invariant of a quintic produces a dierent
expression than that which appears elsewhere and which is more useful for
certain purposes (for example our formula can be easily evaluated along the
Bring curve).
Finally it is worth highlighting the geometry that connects the quintic
and the icosahedron. Using a radical transformation, a quintic can always be
put in the form y
5
+5y
2
+5y + = 0. The vector of ordered roots of such
a quintic lies on the quadric surface

y
i
=

y
2
i
= 0 in P
4
and the reduced
Galois group A
5
acts on the two families of lines in this doubly-ruled surface
by permuting coordinates. The A
5
actions on these families, parameterized
by P
1
, are conjugate to the action of the group of rotations of an icosahedron
on its circumsphere and the quintic thus denes a point in the quotients
the icosahedral invariants of a quintic. We discuss this in detail below but
rst we collect those facts about the icosahedron that we will need.
2 The icosahedron
Given an icosahedron, we can identify its circumsphere S with the extended
complex plane, and so also with P
1
, using the usual stereographic projection:
(x, y, z)
x+iy
1z
. Orienting our icosahedron appropriately, the 12 vertices
have complex coordinates:
0,

( +
1
), ,

(
2
+
2
) = 0, 1, . . . , 4 (1)
where = e
2i/5
.
Projecting radially from the centre, we can regard the edges and faces of
the icosahedron as subsets of S. With the sole exception of gure 2 below,
we shall always regard the faces and edges as subsets of S C . The
picture of the icosahedron we should have in mind is thus:
2
Figure 1: The icosahedron, projected radially onto its
circumsphere.
We may inscribe a tetrahedron in an icosahedron by placing a tetrahedral
vertex at the centre of 4 of the 20 icosahedral faces as shown:
Figure 2: The icosahedron with inscribed tetrahedron.
Note that for each icosahedral vertex, exactly one of the 5 icosahedral faces
to which it belongs has a tetrahedral vertex at its centre. If we pick an axis
3
joining two antipodal icosahedral vertices, we can consider the 5 inscribed
tetrahedra obtained by rotating this conguration through 2/5 for =
0, 1, . . . , 4. None of these tetrahedra have any vertices in common and so each
of the 20 faces of the icosahedron are labeled by a number {0, 1, . . . , 4}.
Figure 3 below exhibits such a numbering after stereographic projection.
0
0
0
0
1
1
1
1
2
2
2
2
3
3
3
3
4
4
4
4
Figure 3: Tetrahedral face numbering of the icosahedron under
stereographic projection (the outer radial lines meet at the vertex
at innity).
The symmetry group of the icosahedron acts transitively on the set
of 20 faces with stabilizer of order 3 at each face and so has order 60. The
symmetry group also acts faithfully
2
on the set of 5 tetrahedra as constructed
above and so we obtain an embedding S
5
. Since A
5
is the only subgroup
of S
5
of order 60 we must thus have:
A
5
It will be useful later to have explicit generators for . Thus let S be
a rotation through 2/5 about the axis of symmetry joining the antipodal
2
Consider how many rotations simultaneously stabilize at least three of the tetrahedra
(either of gures 2 or 3 may help).
4
vertex pair 0, and let T be a rotation through about the axis of symmetry
joining the midpoints of the antipodal edge pair [0, +
1
], [,
2
+
2
]. Using
the face numbering in gure 3, S, T correspond to the permutations:
S = (01234)
T = (12)(34)
(2)
We note in passing that since these two permutations generate A
5
we can use
this to see that the action on tetrahedra is faithful. In any case we thus have
generators for .
3
In addition, under the embedding of symmetry groups:
PSL(2, C) associated to the identication of the circumsphere of the
icosahedron with P
1
we have:
S =
_

3
0
0
2
_
T =
1

5
_
(
4
)
2

3

4
_ (3)
Having pinned down the symmetry group and its generators, we turn our
attention to the branched covering:
P
1
P
1
/
We wish to construct an explicit isomorphism P
1
/ P
1
. We thus study the
invariant elements in the homogeneous coordinate ring: C[P
1
]

.
Consider rst the possible stabilizer subgroups for the action of on P
1
.
The action is free except on the three exceptional orbits that are the sets of
vertices, edge midpoints and face centres where it has stabilizer subgroups
of order
3
= 5,
1
= 2,
2
= 3 respectively (corresponding to the rotations
about axes of symmetry through antipodal pairs of these points). Each of
these exceptional orbits is the divisor of an invariant homogeneous polyno-
mial. Using (1) we can calculate the polynomial corresponding to the vertices
directly. We obtain:
f(z
1
, z
2
) = z
1
z
2
(z
10
1
+ 11z
5
1
z
5
2
z
10
2
) (4)
We could also calculate the polynomials corresponding to the edge midpoints
and face centres directly but this would be somewhat cumbersome. Instead
we calculate the Hessian and Jacobian covariants of f (see, e.g., [5], [26]) and
3
In fact ST is a rotation about a face centre and we obtain the presentation of as a
triangle group: < S, T | T
2
, (ST)
3
, S
5
> that Hamilton used for his Icosian Calculus.
5
obtain
4
:
H(z
1
, z
2
) = (z
20
1
+z
20
2
) + 228(z
15
1
z
5
2
z
5
1
z
15
2
) 494z
10
1
z
10
2
T(z
1
, z
2
) = (z
30
1
+z
30
2
) + 522(z
25
1
z
5
2
z
5
1
z
25
2
) 10005(z
20
1
z
10
2
+z
10
1
z
20
2
)
(5)
In view of their degrees and the fact that there are no other exceptional orbits,
the divisors of these invariant polynomials must be the the face centres and
edge midpoints respectively.
Next note that since the divisor of any -invariant polynomial p is a sum
of -orbits, repeated according to multiplicity, p must be of the form:
p = f
e
1
H
e
2
T
e
3

j
p
j
(6)
where 0 e
i
<
i
for i = 1, 2, 3 and deg(p
j
) = 60. In particular a degree-60
-invariant polynomial vanishes on a unique orbit.
Now the space of degree-60 -invariant polynomials on P
1
corresponds to
degree-1 polynomials on P
1
/ and so forms a 2-dimensional space. In fact
it is easy to see that any set of three elements must be linearly dependent
directly. Suppose that f
1
, f
2
, f
3
are degree-60 -invariant polynomials on P
1
.
Let [u
i
, v
i
] be a zero of f
i
for i = 1, 2. We suppose that f
1
, f
2
are both
non-zero, that neither is a multiple of the other so that the orbits containing
[u
1
, v
1
], [u
2
, v
2
] are distinct and consider the polynomial (dened up to scale):
g = f
2
(u
1
, v
1
)f
3
(u
2
, v
2
)f
1
+f
3
(u
1
, v
1
)f
1
(u
2
, v
2
)f
2
f
2
(u
1
, v
1
)f
1
(u
2
, v
2
)f
3
We must then have g = 0 for if not we would have an degree-60 -invariant
polynomial vanishing on both the orbit containing [u
1
, v
1
] as well as the orbit
containing [u
2
, v
2
]. This is the required linear relationship.
In particular we must have a linear relationship between f
5
, T
2
, H
3
. Using
the above construction we obtain cf
5
T
2
H
3
= 0 for some constant c.
Expanding and comparing coecients of z
60
1
we nd c = 1728 and thus obtain
the important syzygy:
H
3
+T
2
= 1728f
5
(7)
Moving on from polynomials, we consider -invariant rational functions.
Because has no non-trivial characters such a function r must be of the
form r = p/q where p, q are -invariant homogeneous polynomials of the
4
We think it worthwhile following the notation of [17] as closely as possible to aid the
reader who wishes to compare. It is unfortunate that we must thus use T to denote both
the rotation mentioned above as well as the invariant polynomial introduced here but we
trust that context will protect us from confusion.
6
same degree. Furthermore we can assume e
i
= 0 in (6) for both p, q since
the only solution of deg(f
e
1
H
e
2
T
e
3
) deg(f
e

1
H
e

2
T
e

3
) mod 60 is e
i
= e

i
. If
f
1
, f
2
are a basis for the degree-60 -invariant homogeneous polynomials we
can thus write:
r =

j
a
j
f
1
+b
j
f
2
c
j
f
1
+d
j
f
2
= R(f
1
/f
2
)
where R is the rational function z

j
(a
j
z + b
j
)/(c
j
z + d
j
). We see that,
as expected, the space of -invariant rational functions is simply the space
of rational functions in f
1
/f
2
for any basis f
1
, f
2
as above. Following Klein
we make the choice of basis such that the rational function f
1
/f
2
sends the
vertices, edge midpoints and face centres to , 1, 0, respectively, i.e., we use
the rational function:
I =
H
3
1728f
5
(8)
and we can summarise by saying that we have an isomorphism of function
elds
5
:
C(P
1
)

C(I)
and a corresponding equivalence:
I : P
1
/ P
1
Lastly we note a property of f, H, T, that we shall need later: they are
invariant under composition of antipodal map z 1/ z and reection z z
since the product of these two orientation-reversing icosahedral symmetries
is necessarily a rotation. In particular:
f(z
2
, z
1
) = f(z
1
, z
2
) (9)
and similarly for H, T.
If we were now to follow the usual approach to the icosahedral solution
of the quintic, we would next study the quintic resolvents, of the Galois
extension C(P
1
) C(P
1
)

. These are obtained by taking subgroups of index

5 in the Galois group A
5
corresponding to the tetrahedron. However, as
we have already mentioned, we follow a slightly dierent approach and so
immediately turn our attention to the solution of the quintic. We rst quickly
gather the few facts about Tschirnhaus transformations that we need.
5
In fact we can do better than birational equivalence; indeed we have an isomorphism
of graded algebras C[P
1
]

C[f, H, T]/(1728f
5
T
2
H
3
) where f, H, T are given weights
12, 30, 20 respectively. However we do not need this.
7
3 Tschirnhaus and the canonical equation
A common approach when solving the polynomial equation:
x
n
+a
1
x
n1
+ +a
n
= 0 (10)
is to begin by making the ane substitution y = x a
1
/n and so eliminate
the term of degree n 1. This substitution is a special case of the so-called
Tschirnhaus transformation [33] in which y is allowed to be a polynomial ex-
pression q in x. If
i
are the roots of (10), the coecients of the transformed
equation:

i
(y q(
i
)) = 0
are polynomials in the a
i
by S
n
-invariance (or Newtons identities).
Using a Tschirnhaus transformation we can simultaneously eliminate fur-
ther terms in the original polynomial. For example if n 3 and a
1
= 0, it is
easy to check that the substitution:
y = x
2
+b
1
x +b
2
simultaneously eliminates the terms of degree n 1 and n 2 provided the
coecients b
1
, b
2
satisfy the auxiliary polynomial conditions [26]:
b
2
p
2
/n = 0
p
2
b
2
1
+ 2p
3
b
1
+ (p
4
p
2
2
/n) = 0
where p
j
=

j
i
are the power sums of the roots.
Thus, provided we are willing to allow ourselves the auxiliary square root
necessary to solve the above quadratic for b
1
, we may take the general form
of the quintic to be:
y
5
+ 5y
2
+ 5y + = 0 (11)
(We include the factors of 5 for consistency with [17].)
In fact it is possible to simultaneously eliminate the terms of degrees n1,
n2 and n3 (where the coecients of the substitution are determined by
polynomials of degree strictly less than n). Thus, as rst shown by Bring [2]
and subsequently by Jerrard [16], the general quintic can be reduced to the
so-called Bring-Jerrard form:
y
5
+y + = 0
However it is not in this form that the quintic most easily reveals its icosa-
hedral connections and so, except for section 8.1 and appendix A, we shall
take the quintic in the form (11).
8
4 The space of roots and the icosahedral in-
variant
Given the quintic (11), if we exclude the trivial equation with = = = 0
we may regard the roots y
i
as homogeneous coordinates of a point in P
4
.
More precisely, since the roots are unordered we obtain an orbit of S
5
in P
4
,
where S
5
acts by permuting coordinates. Furthermore because the quintic
lacks terms of degree 4 and 3, this orbit lies in the non-singular S
5
-invariant
quadric surface:
Q =
_
[y
0
, . . . , y
4
] P
4
|

y
i
=

y
2
i
= 0
_
Now the quadric surface is ruled by two P
1
-families of lines, the (reduced)
Galois group acts on these and the quintic thus denes a point in the quotient
of each family. Each quotient is isomorphic to the map P
1
P
1
/ of section
2 and the points the quintic obtained are the icosahedral invariants of the
quintic. We shall describe this explicitly in section 5 but rst we discuss the
geometry slightly more abstractly.
Thus let V = {(y
0
, . . . , y
4
) C
5
|

y
i
= 0} and recall [34] that the
space of lines in P(V ) is naturally the Klein quadric:
K =
_
[] P(
2
V ) | = 0
_
The line determined by a point [] K is L
[]
= P({v V | v = 0}).
Now the quadratic form q =

i
y
2
i
restricted to V provides an isomorphism
V V

and so also an isomorphism:

q :
2
V (
2
V )

Combining this with the natural isomorphism provided by the wedge product:
:
2
V (
2
V )

4
V
we thus have an isomorphism:

2
V
2
V
4
V
Furthermore this isomorphism is equivariant with respect to the orthogo-
nal group O(V ) = O(V, q). If we now x an orientation on V we have an
SO(V )-equivariant automorphism of
2
V . This is of course simply the (com-
plex) Hodge -operator and in this setting is self-inverse. Thus q provides a
decomposition:

2
V =
+

9
of
2
V into the positive and negative eigenspaces of ; this is a decomposition
into irreducible SO(V ) spaces. K meets the two planes P(

) in two non-
singular conics C

and the natural map (intersection of lines) provides an

SO(V )-equivariant isomorphism:
C
+
C

Q (12)
This is the well known double ruling of the non-singular quadric surface.
Now the representation of S
5
on
4
V is the sign representation and so in
view of the above it is desirable to restrict our representation S
5
O(V ) to
A
5
SO(V ). To do this we must be able to associate an A
5
-orbit to our
quintic rather than a full S
5
orbit. We achieve this by supplying a square
root of the discriminant:

2
= 3125

i<j
(y
i
y
j
)
2
= 108
5
135
4

2
+ 90
2

2
320
3
+ 256
5
+
4
(13)
in addition to , , .
Thus, given a quintic, which for simplicity we assume has full Galois
group S
5
, together with we obtain an orbit of A
5
in each of the conics C

.
Since these actions are non-trivial and A
5
is simple, we must have icosahedral
actions. We thus obtain two points:
Z

/A
5
P
1
Note, for the sake of deniteness, that we are using the unique isomorphism
C

/A
5
P
1
sending the orbits with order 60/
i
to 1, 0, respectively for

i
= 2, 3, 5. We thus have a (possibly innite) numerical invariant Z

asso-
ciated to our quintic.
By implicitly employing the isomorphism C

P
1
above, we are using
the fact that a non-singular conic is rational. In fact these conics have natural
rational parameterizations if we supply V with just a little extra structure
(a complex spin structure). Specically we take two 2-dimensional complex
vectors spaces S

with structure group SL(2, C) and x an isomorphism:

S
+
S

V (14)
commuting with the SL(2, C) SL(2, C) and SO(4, C) actions. Using this
extra data we can represent everything in terms of S

. In particular

Sym
2
S

and the symmetric square of S

in Sym
2
S

is (the cone over) the

conic C

. We thus naturally have C

P(S

). Using this, the double ruling

(12) becomes the more familiar Segre embedding:
P(S
+
) P(S

) Q
10
The corresponding statement for homogeneous coordinate rings is that we
can express C[Q] as a Segre product:
C[Q]
n0
C[P(S
+
)]
n
C[P(S

)]
n
(15)
where C[P(S

)]
n
= H
0
(P(S

), O(n)) is the degree n summand of the graded

ring C[P(S

)].
In the following section, where we actually calculate Z

in terms of
, , , , we give a more concrete, explicit construction of Z

.
5 Calculating the invariant
We wish to calculate Z

in terms of , , , (which we now regard as

parameters). This is mostly just an exercise in classical invariant theory for
the groups S
5
and A
5
.
The rings of invariants of P(V ) are well known:
C[P(V )]
S
5
C[q, , , ]
C[P(V )]
A
5
C[q, , , , ]/(
2
D(q, , , ))
where D is the discriminant of the quintic y
5
+qy
3
+ 5y
2
+ 5y +.
The rings of invariants of Q for its embedding in P(V ) are:
C[Q]
S
5
C[, , ]
C[Q]
A
5
C[, , , ]/(
2
D(, , ))
(16)
To see that these hold, let G denote S
5
or A
5
and take G-invariants of the
exact sequence dening the homogeneous coordinate ring of Q. A priori we
get the half-exact sequence:
0 (q)
G
C[P(V )]
G
C[Q]
G
If we can show that the map to C[Q]
G
is in fact surjective then (16) follow
trivially. Demonstrating surjectivity amounts to showing that any function
d : G (q) such that d(gh) = d(g) + g d(h) for all g, h G is of the form
d(g) =

d g

d for some

d (q). This is easily seen to be the case by taking

d = 1/|G|

gG
d(g).
6
Note also that in view of the relation
2
= D(, , ), any A
5
-invariant
polynomial h on Q can be written uniquely in the form:
h = h
s
+h
a

6
Of course what were really showing is that the group cohomology H
1
(G, (q)) = 0.
11
for unique polynomials h
s
, h
a
in , , determined by:
h
s
= (h +h

)/2
h
a
= (h h

)/2
(17)
where h

is the polynomial obtained by transforming h under any odd per-

mutation.
In view of (15) and (16), any S
5
-invariant element of the ring C[P(S
+
)]
n

C[P(S

)]
n
can be written uniquely as a polynomial in , , . We shall make
repeated use of this essential fact below. We now introduce coordinates to
make all this explicit.
We rst change basis from y
0
, . . . , y
4
to the (Fourier-dual) basis in which
S is diagonal:
p
k
=

kj
y
j
(18)
where as before = e
2i/5
. Our subspace V now has equation p
0
= 0 and
our quadratic form q, restricted to V , is dened in terms of the p
k
by 25q =
10(p
1
p
4
+ p
2
p
3
). We let S
+
have coordinates
1
,
2
, S

have coordinates

1
,
2
and dene the map (14) via the bilinear map:
p
1
= 5
1

1
p
2
= 5
2

1
p
3
= 5
1

2
p
4
= 5
2

2
(19)
Thus [
1
,
2
] are homogeneous coordinates on C
+
and [
1
,
2
] homogeneous
coordinates on C

. The A
5
-action on the
i
is given by the formulae (2), (3)
and the dual action on the
i
is obtained by replacing with
2
in the same
formulae. The S
5
-action of course interchanges the
i
and
i
, indeed the odd
permutation R = (1243) acts as ([
1
,
2
], [
1
,
2
]) ([
2
,
1
], [
1
,
2
]).
Recalling our formulae for f, H, T from section 2 we dene:
f
1
= f(
1
,
2
) f
2
= f(
1
,
2
)
and similarly we dene H
1
, H
2
and T
1
, T
2
. Note that by (9), R interchanges
f
1
, f
2
and similarly for the H
i
, T
i
. From now on we also denote Z

by Z
1
, Z
2
.
In this notation we now have:
Z
1
=
H
3
1
1728f
5
1
We wish to express this in terms of , , , . We thus write:
Z
1
=
H
3
1
f
5
2
1728(f
1
f
2
)
5
(20)
12
Now f
1
f
2
xed by R and so is S
5
-invariant. It thus lies in C[, , ]. Since
it is of degree 12, it must be a linear combination of
4
,
3
, . To x
the coecients we compare leading coecients as polynomials in
i
,
i
. It is
straightforward to verify that:
=
3
1

2
1

2
1

3
2

2
2

3
1
+
3
2

2
2
=
4
1

3
2
+
3
1

4
1
+ 3
2
1

2
2

2
1

2
2

3
2

4
2
+
4
2

3
1

2
=
5
1
(
5
1
+
5
2
) + 10
4
1

3
1

2
2
10
3
1

2
2

4
2

10
2
1

3
2

4
1

2
10
1

4
2

2
1

3
2
+
5
2
(
5
1

5
2
)
The coecient of
12
1
in f
1
f
2
is 0 whereas the same coecients in
4
,
3
,
are
8
1

4
2
,
3
1

9
2
,
8
1

4
2

3
1

9
2
respectively. From this we see that we must
have f
1
f
2
= A(
4

3
+ ) for some constant A. Furthermore, upon
noting that the coecient of
11
1

2

11
1

2
in f
1
f
2
is 1 whereas it is 0 in
4
,
3
and 1 in we learn that A = 1. In other words we obtain:
f
1
f
2
=
4

3
+
This deals with the denominator in (20); we turn our attention to the nu-
merator.
Decomposing the numerator of (20) using (17) and recalling that our odd
permutation R interchanges the f
1
, f
2
as well as H
1
, H
2
, we get:
H
3
1
f
5
2
=
H
3
1
f
5
2
+H
3
2
f
5
1
2
+
H
3
1
f
5
2
H
3
2
f
5
1
2
= p +q (21)
where p, q are polynomials in , , .
We could now attempt to calculate p, q by the same means that we cal-
culated f
1
f
2
above but this would be an enormous amount of work since p, q
have degrees 60, 50 respectively. Instead, recall that we have the syzygies:
T
2
i
= 12
3
f
5
i
H
3
i
Multiplying these together and rearranging we obtain:
2p = H
3
1
f
5
2
+H
3
2
f
5
1
= 12
3
(f
1
f
2
)
5
+ 12
3
(H
1
H
2
)
3
12
3
(T
1
T
2
)
2
We will thus have the required expression for p in terms of , , as soon as
we express H
1
H
2
and T
1
T
2
in these terms. To do this we do follow the same
procedure as that used to nd f
1
f
2
above and (admittedly with somewhat
more eort) we obtain:
H
1
H
2
=
4
+ 40
2

2
192
5
120
3
+ 640
4

2
144
5
T
1
T
2
=
6
+ 60
2

4
+ 576
5

3
180
3

3
+ 648
5

2
2760
4

2
+
7200
7
1728
10
+ 9360
3

4
2080
6

3
16200
2

6
13
It remains only to calculate q. This time the trick we use is to note that
as well as (21) above, we have H
3
2
f
5
1
= p q and so:
(H
1
H
2
)
3
(f
1
f
2
)
5
= p
2

2
q
2
It follows that taking our above polynomial expressions for H
1
H
2
, f
1
f
2
,
2
=
D(, , ), p we must nd a factorization of ((H
1
H
2
)
3
(f
1
f
2
)
5
p
2
)/D(, , ).
From this we determine
7
:
2q = (8
5
40
4

2
+ 10
2

2
+ 45
3
81
5

4
)
(64
10
+ 40
7
160
6

3
+
5

5
4

2
+ 5
3

4
25
2

2
)
The two signs corresponding to the two invariants: Z
1
, Z
2
. With this formula
in hand we have achieved our goal of expressing Z
i
in terms of , , , .
Our derivation and formulae for Z
i
in terms of , , , dier from
Kleins ([17], pg. 213) and indeed from any other accounts we have come
across which all reproduce Kleins formulae exactly. In fact our approach is
more closely connected to Gordons work [9]. Klein does outline the connec-
tion but he leaves it unnished as he has already obtained his formulae for
Z
i
by other means. Although the formulae for Z
i
in [17] have the advantage
of being more concise than ours, they have the disadvantage that Z
i
is rep-
resented as a rather complicated rational expression (rather than in terms
polynomials) and that Z
i
cannot be evaluated along the Bring curve = 0
easily.
6 Obtaining the roots
The invariants Z
i
dened above are obtained by applying the icosahedral
function I to the coordinates
i
,
i
. These coordinates determine a point in
the projective space of vectors of roots P(V ), and so if we could invert the
icosahedral function I and express
i
,
i
in terms of the invariants Z
i
, we
would be able to solve the quintic. We will show how to invert I in section 7
below but rst we ll in the details of how an inverse for I yields a solution
of the quintic.
We begin by using (18), (19) to express roots in terms of
i
,
i
:
y

=
4

1
+
2

2
+

2
(22)
7
That q itself turns out to factorize like this is unexpected, at least to the author.
14
Now suppose [
1
,
2
] is a solution of the icosahedral equation:
I(
1
,
2
) = Z
1
(23)
The rst thing to notice is that we do not also need to nd an inverse for Z
2
.
Concretely, the idea is that if we can nd A
5
-invariant forms that are linear
in
i
then we can use these to eliminate the
i
in (22) and so express y

in
terms of just , , and our solution [
1
,
2
] of (23). We thus enlarge the
ring of invariant polynomials we are studying from the Segre product (15)
to the full tensor product C[P(S
+
)] C[P(S

)]. The two A

5
-invariant forms
linear in
i
of lowest degree in
i
are:
N
1
= (7
5
1

2
2
+
7
2
)
1
+ (
7
1
+ 7
2
1

5
2
)
2
M
1
= (
13
1
39
8
1

5
2
26
3
1

10
2
)
1
+ (26
10
1

3
2
+ 39
5
1

8
2
+
13
2
)
2
(24)
There are a number of ways to derive these expressions. We follow Gordon
[9] and use transvectants. We thus recall (see for example [3] or [5]) that if
f, g are two homogeneous polynomials in
1
,
2
then the r
th
transvectant of
f, g is given by:
(f, g)
r
=
r

i=0
(1)
i
i!(r i)!

r
f

ri
1

i
2

r
g

i
1

ri
2
We extend this to homogeneous polynomials in both
i
,
i
using bilinearity,
i.e., if:
f =

i,j
f
ij

i
1

ai
2

j
1

bj
2
g =

k,l
g
kl

k
1

ck
2

l
1

dl
2
then we dene the (r, s)-transvectant:
(f, g)
r,s
=

i,j,k,l
f
ij
g
kl
(
i
1

ai
2
,
k
1

ck
2
)
r
(
j
1

bj
2
,
l
1

dl
2
)
s
Using these transvectants, we can construct the invariants we need. Indeed
it is straightforward to verify that:
(, )
0,3
= 6N
1
((, )
0,2
, N
1
)
0,1
= 8M
1
as required.
15
As a brief aside we comment on the geometry of N
1
, M
1
. Since they are
linear in the
i
they correspond to A
5
-equivariant branched covers P(S
+
)
P(S

) of degrees 7, 13 respectively. From the Riemann-Hurwitz formula,

the ramication index of these maps is 12, 24 respectively and so the branch
locus in P(S

) must be the set of vertices of an icosahedron in P(S

) in each
case (since they are stable under the A
5
action). For N
1
, there are 12 doubly-
ramied points in P(S
+
) which must also be the vertices of an icosahedron.
There are 5 remaining unramied points in the bre over each of the 12
points in the branch locus. These 60 points are the A
5
-orbit containing
the point
1
/
2
= 7
1/5
, i.e., the orbit with icosahedral invariant 64/189.
It is interesting to wonder if it is possible to characterise the conguration
of these points geometrically. The set of vertices of an Archimedean solid
with icosahedral symmetry might seem like a plausible candidate but in fact
this does not work as can be seen easily since the point 7
1/5
lies outside
the eld of coecients of these solids (though the points are remarkably
close to the vertices of the truncated dodecahedron). Similarly there are two
distinguished orbits for M
1
. If it is possible to come up with a geometric
characterisation of these various orbits then it should also apply to the other
two A
5
-equivariant branched covers (also identied by Gordon in [9]) which
have degrees 17 and 23.
Returning to the task at hand we solve the 2 2 system (24) and express

i
in terms of M
1
, N
1
and using (22) obtain:
y

= H
1
1
b

M
1
+H
1
1
c

N
1
(25)
Here H
1
appears as it is the determinant of the matrix which we invert and
the coecients b

, c

are given by:

_
b

=
_

4

2

2

1
+

_

7
1
+ 7
2
1

5
2
26
10
1

3
2
39
5
1

8
2

13
2
7
5
1

2
2

7
2

13
1
39
8
1

5
2
26
3
1

10
2
_
Now in (25) we can calculate H
1
, b

, c

using our solution of (23). We

deal with M
1
, N
1
turning them into forms of the same degree in
i
and
i
which we can thus express in terms of , , , . We thus rewrite (25) as:
y

=
b

f
1
H
1

M
1
f
2
f
1
f
2
+
c

T
1
H
1
f
2
1

N
1
f
2
1
T
2
T
1
T
2
The methods described in section 5 then allow us to calculate:
M
1
f
2
= (11
3
+ 2
2

2
)/2 /2
N
1
f
2
1
T
2
= r +s
16
where:
2r =
2

4
+ 53
4

3
+ 64
7

2
7
4

3
225
3

12
6

2
+ 216
9
+ 717
2

5
464
5

4
720
7
2s =
2

3
+ 3
2

2
9
4
4
4
8
7
80
3

3
and since we already have formulae for f
1
f
2
and T
1
T
2
we have the required
expression for y

in terms of , , , ,
1
,
2
.
It thus remains only to show how to solve (23) to nd
1
,
2
.
7 Solving the icosahedral equation
We wish to invert the equation:
I(z) = Z (26)
(In this section we regard I as a function of the single variable z = z
1
/z
2
.)
This problem was essentially solved by Schwarz in 1873 when he determined
the list parameters for which the hypergeometric dierential equation has
nite monodromy. We thus need only cast our problem in the right form and
appeal to general theory. The solution can be written in terms of Schwarz
s-functions (as discussed in [23] for example).
We must identify domains of injectivity for I (equivalently, fundamental
domains for the action of on P
1
). We thus note that if r is any reection
about a plane of symmetry of the icosahedron then:
I r =

I
for the reasons explained when discussing (9). Since there is a plane of
symmetry through any edge of the icosahedron as well as a plane of symmetry
through each of the altitudes of any face of the icosahedron, it follows that the
edges and altitudes of the faces of the icosahedron constitute the preimage of
RP
1
= R under I. The altitudes divide each face into six triangles with
angles /2, /3, /5 = /
i
. I sends the vertices of each triangle to 0, 1,
and is injective on the interior. It maps three of them biholomorphically to
upper half space H
+
and three of them biholomorphically to lower half space
H

, according to whether their vertices are sent to 0, 1, in anti-clockwise

or clockwise order respectively. Subdividing faces like this, gure 1 becomes:
17
Figure 4: Icosahedral tiling of sphere. I maps the interior of each
light and dark triangle biholomorphically onto the upper and
lower half-planes respectively.
Under stereographic projection, the subdivision of the face with vertices
0, +
1
,
2
+ 1 is:
0 t
h
+
-1

2
+1
Figure 5: Domains of injectivity for I under stereographic
projection. The points h, t are the images of the face centre and
edge midpoint respectively.
We shall construct an inverse for the restriction of I to the interior of the
triangle with vertices 0, t, h. I maps the interior biholomorphically to the
18
lower half-plane H

and the boundary homeomorphically to RP

1
, sending
the points 0, t, h to , 1, 0 respectively.
Now the equation (26) we wish to solve is a degree 60 polynomial but it
has the special property that its monodromy lies in PSL(2, C), i.e., the 60
branches of a local inverse are related by a Mobius transformation. Since
the Schwarzian derivative is invariant under Mobius transformations, the
Schwarzian derivative of the inverse is independent of the branch; moreover
it is easily calculated. This yields a dierential equation for the inverse which,
using standard methods, may be transformed to Gausss hypergeometric dif-
ferential equation. It then follows that an inverse may be expressed as a
ratio of linearly independent solutions to this equation, i.e., as a Schwarz
s-function.
Indeed, as shown in [23], the inverse we seek is determined up the scale
by the angles in the triangle, /
i
, and is given by:
s : H

C
Z C
Z
c1
2
F
1
(a

, b

; c

; Z
1
)
2
F
1
(a, b; c; Z
1
)
Here
2
F
1
is the analytic continuation of Gausss hypergeometric series to
C[1, ), Z
c1
is dened using the principle branch of log on C(, 0].
The values of the constants are:
a =
1
2
_
1
1

3
+
1

1
_
b =
1
2
_
1
1

1
_
c = 1
1

3
a

= a c + 1 b

= b c + 1 c

= 2 c
To determine C we compare the coecients of Z on both sides of the
power series identity:
Z H
3
(z
1
, z
2
) = f
5
(z
1
, z
2
)
with z
1
= CZ
1c
2
F
1
(a

, b

; c

; Z) and z
2
=
2
F
1
(a, b; c; Z). This yields C =
1728
1/5
and so s is completely determined.
8
In summary then, for imZ < 0, our inverse is:
s(Z) =
2
F
1
(
31
60
,
11
60
;
6
5
; Z
1
)
(1728Z)
1/5
2
F
1
(
19
60
,
1
60
;
4
5
; Z
1
)
8
An alternate means of determining C would be to impose the condition s(1) = t
and to use Gausss expression for
2
F
1
(a, b, c, 1) in terms of -functions. This yields C =
t
(
4
5
)(
41
60
)(
61
60
)
(
6
5
)(
29
60
)(
49
60
)
. Since t = 2 (cos(/10) cos(/5)) =
_
5+

5
2

1+

5
2
and we already
know C = 1728
1/5
we obtain a -function identity.
19
Using a similar expression for im(Z) > 0, we could extend this function to the
open set H
+
H

(0, 1) so that we would have an inverse for the restriction

of I to the interior of the triangle with vertices 0, +
1
, h. (Indeed using
Schwarz reection, we can extend s across any of the edges of the triangle
and of course this process generates the monodromy A
5
.)
8 Further properties and parting words
Our focus in these notes has been to present the icosahedral solution of the
quintic as concisely as possible, subject to the conditions of remaining as
explicit as [17] and as self-contained as possible. As a result we have been
forced us to omit discussion of many related matters. We comment briey
on some of these here.
8.1 Brings curve and Keplers great dodecahedron
We mentioned in section 3 that it is possible to reduce the general quintic to
the so-called Bring-Jerrard form:
y
5
+y + = 0
but that we would work with the quintic in the form (11). We did this
because we were following [17], because (11) is more general and because it
is easy to bring out the icosahedral connection using the A
5
actions on the
lines in the doubly-ruled quadric surface. However there is an appealing way
to connect the icosahedron with the quintic in Bring-Jerrard form which is
worth mentioning. The construction below is described in [11].
Firstly note that the family of quintics in Bring-Jerrard form is the smooth
genus 4 curve B cut out of P
4
by the equations

y
i
=

y
2
i
=

y
3
i
= 0.
This is known as the Bring curve and has automorphism group S
5
correspond-
ing to the general Galois group. The branched covering B B/A
5
P
1
allows us to dene an invariant as before and connects with our work above.
Secondly, starting with an icosahedron in R
3
we form Keplers great do-
decahedron G
D
. This regular solid, which self-intersects in R
3
, has one face
for each vertex of the icosahedron. It is formed by spanning the ve neigh-
bouring vertices of each vertex of the icosahedron with a regular pentagon
and then dismissing the original icosahedron. G
D
thus has the same 12 ver-
tices and 30 edges as the icosahedron but only 12 faces. Projection onto
the common circumsphere S yields a triple covering G
D
S with a double
branching at the 12 vertices and after identifying S with P
1
provides G
D
with
20
a complex structure. Evidently G
D
has Euler characteristic 6 and so genus
4. In fact, as explained in [11], G
D
is isomorphic to the Bring curve.
The isomorphism G
D
B can be used to bring out the relationship
between the quintic and the icosahedron.
8.2 Modular curves, Ramanujans continued fraction
and Shiodas modular surface
From one point of view, the exceptional geometry of the quintic is a result
of the exceptional isomorphism:
A
5
PSL(2, 5)
Corresponding to the exact sequence dening the level 5 principal con-
gruence subgroup of the modular group:
0 (5) PSL(2, Z) PSL(2, 5) 0
there is a factorization of the modular quotient:
j : H

j
5
X(5)

I
X(1)
where H

= H
+
QP
1
is the upper half-plane together with the PSL(2, Z)-
orbit of and X(N) is the compactied modular curve of level N. The
curves X(5), X(1) are rational and the map

I : X(5) X(1) is a quotient
by PSL(2, 5) and is thus an icosahedral quotient. We can use this to nd an
inverse for the icosahedral function I in terms of Jacobi -functions (provided
we are willing to invert Kleins j-invariant). Indeed the map j
5
may be
expressed as:
j
5
() = q
2/5

Z
q
5n
2
+3n

Z
q
5n
2
+n
= q
3/5

1
(; q
5
)

1
(2; q
5
)
(27)
where q = e
i
and we are using the -function notational conventions of
[35]. Thus given Z as in section 7, if satises j() = 1728Z then z = j
5
()
is a solution to I(z) = Z.
In fact there is another expression for j
5
, it is none other than Ramanu-
jans continued fraction:
j
5
() =
q
1/5
1 +
q
1 +
q
2
1 +
q
3
1 +
21
Furthermore because we know that the icosahedral vertices, edge midpoints
and face centres in X(5) lie above the points , 1, 0 in X(1), we can calculate
the values of this continued fraction at those orbits in H

which
1
1728
j maps
to , 1, 0. For example j(i) = 1728 and the corresponding edge midpoint
equality:
j
5
(i) = t =

5 +

5
2

1 +

5
2
is one of the identities that famously caught Hardys eye when Ramanujan
rst wrote to him. An excellent account of these results together with a proof
of (27) can be found in [6].
This method of inverting I treats the icosahedral invariant Z of a quintic
as a point in the moduli space of elliptic curves X(1). It is natural to wonder
if the quintic in fact provides a geometric realization of the elliptic curve with
j-invariant Z. Since X(5) is the moduli space of elliptic curves together with
level 5 structure, inverting I would then correspond to nding a basis for the
5-torsion of such a curve in which the Weil pairing is in standard form. If
this were the case we would expect a link with the elliptic curves embedded
in P
4
dened in [14] as an intersection of quadrics. The union of all of these
curves is an irreducible surface (with singularities) whose normalization is
Shiodas modular surface (a compactication of the universal elliptic curve
with level 5 structure).
8.3 Parting words
There is of course much more to say beyond even those remarks in sections
8.1 and 8.2 above. Other matters which we shall not discuss at all but which
seem worth mentioning include the following:
The group PSL(2, 5) is one of a trinityof groups, the other two being
PSL(2, 7) and PSL(2, 11). Higher degree equations can have these
latter two groups as their Galois group and for these there exists a
higher genus version of the icosahedral solution of the quintic.
The rational parameterization of the singularity T
2
+ H
3
= 1728f
5
we have described can be used to nd solutions of the Diophantine
equation a
2
+b
3
+c
5
= 0. See Beukers [1] for details.
We have worked exclusively over C. Needless to say, this is not neces-
sary, the eld Q(e
i/5
) suces for most of the above. Serre makes some
helpful comments in this direction in his letter [27].
22
The icosahedral solution of the quintic is not usually the most ecient
technique for nding the roots. More practical formulae appear in [30]
for example.
A An earlier solution
In addition to the techniques described above, there is another approach to
the solution of the quintic discovered by Lambert
9
[19] in 1758 and again by
Eisenstein [7] in 1844.
Consider the quintic in Bring-Jerrard form (up to a sign):
y
5
y + = 0 (28)
Viewing y as a function of , we claim that the branch of y such that y(0) = 0
has power series:
y() =

k0
_
5k
k
_

4k+1
4k + 1
(29)
(This can also be expressed in terms of the generalized hypergeometric func-
tion as: y() =
5
F
4
_
1,
4
5
,
3
5
,
2
5
,
1
5
;
5
4
, 1,
3
4
,
1
2
; 5(
5
4
)
4
_
.)
This appealing result is established using analytic methods (Lagrange in-
version) in [24] and [29] as well as [21]. However since this statement is really
an identity of binomial coecients it is desirable to have a combinatorial
proof for the identity (28) satised by the generating function (29).
Now the coecients in (29) are a special case of the sequence:
p
d
k
=
1
(p 1)k + 1
_
pk
k
_
which specializes to the Catalan numbers for p = 2. This sequence, con-
sidered long ago by Fuss [8], was studied in some detail in [13]. Just as
various identities for the Catalan numbers can be established by observing
that the they count (amongst many other things) certain lattice paths, so
too can those identities for
p
d
k
which we seek for p = 5 be established by
demonstrating that these coecients count certain paths introduced in [13].
Although the results in [13] thus provide a combinatorial proof of the
generating function identity (28), there is a more direct combinatorial proof
9
It should be pointed out that Lambert would not have been aware that his method
provided a solution of the general quintic since the reduction the Bring-Jerrard form was
not known in his time.
23
presented in [10] based on an observation of Raney [25]. He noticed that if
a
1
, . . . , a
m
is any sequence of integers that sum to 1, then exactly one of the
m cyclic permutations of this sequence has all of its partial sums positive.
With this in mind we consider the problem of counting sequences a
0
, . . . , a
kp
such that:
a
0
+ +a
kp
= 1
All partial sums are positive
Each a
i
is either 1 or 1 p
Using Raneys observation it is clear that the number of such sequences is
p
d
k
.
The natural recursive structure of such sequences provided by concatenation
of p such sequences, followed by a terminating value of 1p then corresponds
to the identity we seek. The interested reader will nd details in [10].
References
[1] F. Beukers. The diophantine equation ax
p
+ by
q
= cz
r
. Duke Mathe-
matical Journal, 91(1), 1998.
[2] E.S. Bring. Meletemata quaedam mathematematica circa transforma-
tionem aequationum algebraicarum. Lund University, Promotionschrift,
1786.
[3] W. Crawley-Boevey. Lectures on representation theory and invariant
theory. Erganzungsreihe Sonderforschungsbereich 343 Diskrete Struk-
turen in der Mathematik, 90-004, Bielefeld University, 1990.
[4] L.E. Dickson. Modern Algebraic Theories. Ben J.H Sanborn & Co.,
1926.
[5] I.V. Dolgachev. Lectures on Invariant Theory. Number 296 in London
Mathematical Society Lecture Note Series. Cambridge University Press,
2003.
[6] W. Duke. Continued fractions and modular functions. Bulletin of the
American Mathematical Society, 42(2):137162, 2005.
[7] F.G.M. Eisenstein. Allgemeine auosung der gleichungen von den ersten
vier graden. J. Reine Angew. Math., 27:8183, 1884.
24
[8] N. Fuss. Solutio quaestionis, quot modis polygonum n laterum in polyg-
ona m laterum, per diagonales resolvi quaeat. Nova acta academiae
scientiarum imperialis Petropolitanae, 9:243251, 1791.
[9] P. Gordon. Ueber die auosung der gleichungen vom f unften grade.
Mathematische Annalen, 13:375404, 1869.
[10] R.L. Graham, D.E. Knuth, and O. Patashnik. Concrete mathematics.
Addison Wesley, 1994.
[11] M.L. Green. On the analytic solution of the equation of the fth degree.
Compositio Mathematica, 37:233241, 1978.
[12] C. Hermite. Sur la resolution de lequation du cinqu`eme degre (1858).
Oevres de Charles Hermite, 2:521, 2009.
[13] P. Hilton and J. Pedersen. Catalan numbers, their generalization, and
their uses. Mathematical Intelligencer, 13(2):6475, 1991.
[14] K. Hulek. Geometry of the horrocks-mumford bundle. In Proceedings
of Symposium in Pure Math, volume 46, pages 6985. American Math-
ematical Society, 1987.
[15] B. Hunt. The geometry of some special arithmetic quotients. Springer,
1996.
[16] G.B. Jerrard. Mathematical researches. William Strong, Bristol, 2, 1834.
[17] F. Klein. Lectures on the icosahedron and the equation of the fth degree
(trans.). Cosimo Classics, 1884.
[18] F. Klein and P. Slodowy. Vorlesungen uber das Ikosaeder und die
Auosung der Gleichungen vom f unften Grade. Birkhauser, 1993.
[19] J.H. Lambert. Observationes variae in mathesin puram. Acta Helvetica,
3(1):128168, 1758.
[20] Mathoverow. Do there exist modern expositions of kleins icosahedron?
http://mathoverflow.net/questions/9474, 2009.
[21] Mathoverow. What is lagrange inversion good for? http://
mathoverflow.net/questions/32099/#32261, 2010.
[22] H. McKean and V. Moll. Elliptic Curves. Cambridge University Press,
1997.
25
[23] Z. Nehari. Conformal mapping. McGraw-Hill, 1952.
[24] S.J. Patterson. Eisenstein and the quintic equation. Historia mathemat-
ica, 17:132140, 1990.
[25] G.N. Raney. Functional composition patterns and power series reversion.
Transactions of the American Mathematical Society, 94:441451, 1960.
[26] M.J. Schurman. Geometry of the quintic. Wiley Interscience, 1997.
[27] J.P. Serre. Extensions icosaedriques. Seminaire de Theorie des Nombres
de Bordeaux, 19(123):550552, 1979.
[28] P. Slodowy. Das ikosaeder und die gleichungen f unften grades. Arith-
metik und Geometrie. Vier Vorlesungen (Mathematischen Miniat uren,
Band 3), 1986.
[29] J. Stillwell. Eisensteins footnote. Mathematical Intelligencer, 17(2):58
62, 1995.
[30] B. Sturmfels. Solving algebraic equations in terms of a-hypergeometric
series. In Department Of Mathematics, Texas A&M University, College
Station, TX 77843. (EM), pages 171181, 1996.
[31] G. Toth. Finite Mobius groups, minimal immersions of spheres, and
moduli. Springer, 2002.
[32] G. Toth. Glimpses of Algebra and Geometry. Springer, second edition,
2002.
[33] E. Tschirnhaus. Methodus auferendi omnes terminos intermedios ex
data equatione. Acta Eruditorum, II, 1683.
[34] R. S. Ward and R. O. Wells, Jr. Twistor geometry and eld theory.
Cambridge Monographs on Mathematical Physics. Cambridge Univer-
sity Press, Cambridge, 1990.
[35] E.T. Whittaker and G.N. Watson. A Course of Modern Analysis. Cam-
bridge University Press, 1927.
26