Solving the quintic by iteration in three dimensions

SOLVING THE QUINTIC
BY ITERATION IN THREE DIMENSIONS

SCOTT CRASS

Abstract. The requirement for solving a polynomial is a means of breaking
its symmetry, which in the case of the quintic, is that of the symmetric group
S5. Induced by its five-dimensional linear permutation representation is a
three-dimensional projective action. A mapping of complex projective 3-space
vfith this 55 symmetry can provide the requisite symmetry-breaking tool.

The article describes some of the 1S5 geometry in CP'^ as well as several
maps with particularly elegant geometric and dynamical properties. Using a
rational map in degree six, it culminates with an explicit algorithm for solv-
ing a general quintic. In contrast to the Doyle-McMuUen procedure — three
1-dimensional iterations, the present solution employs one 3-dimensional iter-
ation.

1. Overview

In I Doyle and McMuUen 1989 1, a solution to the quintic takes place in three iter-
ative steps — a tower of algorithms each of which involves iteration in one complex
dimension. Given almost any quintic p and almost any initial point in C , the series
of algorithms produces a root of p. Their method is geometrically distinguished in
that the tower has the S5 symmetry of the general quintic. Its central feature is a
map on the Riemann sphere with icosahedral (^^5) symmetry.

The present paper describes a solution to a full measure's worth of quintics that
runs as a single iteration in three dimensions. That the procedure produces a root
for almost any initial point in complex projective 3-space (CP'^) is conjectural at the
moment. At its core is a map on CP'^ with ^5 symmetry. Motivating this general
project is a desire to develop solutions to equations that utilize geometrically elegant
dynamical systems.

The work unfolds in three stages: 1) some background geometry, 2) special maps
with symmetry, and 3) a solution to the quintic based on the preceding stages.

Section ^ ^5 geometry. The setting here is CP^ upon which the symmetric
group S5 acts. Finding a map with special S5 geometry requires some familiarity
with this action. We will consider some features associated with the maps that
emerge in the second stage. Indeed, the discovery of these maps derives from an
awareness of the geometric landscape:

• coordinate systems

• the structure of an iSs-invariant quadric surface

• the structure of certain special orbits of points, lines, planes, and conies.

In addition, the system of 55-invariant polynomials plays a fundamental role in the
search for maps.

1

2

SCOTT CRASS

Section |^: Maps with 55 symmetry. At this stage, we exploit our geometric
understanding to discover empirically several maps with special qualities. Appear-
ing here are families of maps associated with the icosahedron, the dodecahedron,
and the complete graph on five vertices. The known features of their geometric
and dynamical behavior come under discussion. However, they are not known to
possess several desired properties. In light of significant experimental evidence, I
leave claims concerning these properties as conjectures.

Section ^: Dynamical solution to the quintic. Following the Doyle-
McMuUen framework, a special family of quintics corresponds to a rigid family
£ of iSs maps on CP'^. 'Rigidity' means that each member of £ is conjugate to a
single reference map / with elegant geometry and dynamics. The solution is general
since almost any quintic p transforms into the special family. Thus, associated with
p is a map g that we iterate. Using tools, its output — conjecturally, a single ^5
orbit — provides for an approximate solution to {p = 0}.

A subsequent paper extends the m ethod to the octic in a way that might gener-

alize to higher degree. [jCrass 1999c

2. S5 ACTS ON CP^
The permutation action of the on preserves the hyperplane

and, thereby, restricts to a faithful four-dimensional irreducible representation.
(Since there will be two variables that describe the hyperplane, the subscript x
appears here.) This induces an 55 action on CP'^. Let Q120 denote the correspond-
ing subgroup of PGL4C.

2.1. Coordinates. For many purposes, the most perspicuous geometric descrip-
tion of G120 employs five coordinates that sum to zero. One advantage is the simple
expression of the 5i2o-duality between points and planes. In general, for a finite ac-
tion Q whose matrix representatives are unitary, a point a is Q-dual to a hyperplane
C if

C^{a-x = 0}.

Consequently, a and C have the same stabilizer in Q. By the orthogonal action of
on C*, a point

a = [ai, 02, as, 04, 05]^ ^^=0 G CP^
corresponds to the plane

{a • X 0} < > flfe Xfc =

(Square brackets indicate a point in projective space.)

A system of four it-coordinates also describes the hyperplane 7i„. These hyper-
plane coordinates arise from the "hermitian" change of variable

/ 1 W5 ujI w| \

u^Hx x^H^u H=^ % ^ 4 %

V5 1 W5 W5

V 1 W| UjI UjI UJ5 J

SOLVING THE QUINTIC BY ITERATION

3

where lo^ = '/^ and the choice of scalar factor gives

(1)

(I

V

/-4

1

1

1

1\

1

-4

1

1

1

1

1

-4

1

1

1

1

1

-4

1

1

1

1

-V

2.2. Invariant polynomials. The fundamental result on symmetric fimctions
states that the n elementary symmetric functions of degrees one through n generate
the ring of <S„-invariant polynomials. Since the action on CP'' occurs where the
degree-1 symmetric polynomial vanishes, there are four generating 5i2o-invariants.
By Newton's identities, the power sums

5

A; = 2,..., 5

Fk{x) = Va;^

also generate the ^120 invariants. In hyperplane coordinates, these are

$2(w) = F2{H'^U) = 2 (mi W4 + W2 Ms)
$3(m) = -7= (mi U% + M? M3 + U% U4 + U2 u\)

Vo

2/3 "2 2 33 22 3\

$4(w) = - (2Wi W2 + 3M2 M3 + 2 Wl M3 + 2 U2 U4 + 12 Ul U2 Us M4 + 3Mi M4 + 2 W3 W4)

5

$6 (w) = — («! + W2 + 20 wi U2 U3 + 30 Ul U2 W3 + m| + 30 U2 U4
5 V 5

+ 20 Wi Us ua + 20 U2 W3 ua + 30 M2 us u\ + 30 Wi ui, u\ + 20 ui U2 ua + "4) •

In classical invariant theory, relative invariants result from taking the deter-
minant of 1) the hessian H{F) of an invariant F and 2) the "bordered hessian"
B{F, G) of two invariants F and G

B{F,G)

(

\ 9G

H{F)

dG

dXn

dG \

do

/

A polynomial F is relatively invariant if, for all T e ^,

FoT = aTF

where a is a character on Q.

Proposition 1. Given T e GL„(C) and invariants F,G,

\H{F{Tx))\=\T\-'\H{Fix))\
\B{F{Tx),G{Tx))\ = \T\-^\B{F{x\G{x))\.
Here, \ ■ \ indicates the determinant.

For the permutation action of iSs, these give absolute invariants — the character is
trivial. Thus, each is expressible in terms of the generators $fc. The following result

will serve a subsequent computational purpose. (Note: Many of this work's results
derive from calculation. For this purpose, I used Mathematica. I will refer to them
as "Facts." )

4

SCOTT CRASS

Fact 1. With G4 = \H{^3)\ and G5 = \B{^3, $2)!, the "power-sum" invariants of
degrees four and five are given by

$5 = -i-(720$2$3 + G5).

864 ^ '

2.3. Quadric surface. The degree-2 invariant defines an 55-invariant surface in
CP^

Q = {$2 = 0}.
The quadratic form associated with Q is given by

$2 = 2 det U with U

Ui -U2
U3 Ui

Accordingly, Q is ruled by two families of lines

Ui — U2\ fn n\ TTU ( ^'^2\ f bl\ fO

U3 Ui I 1 62 / Vo

Alternatively, the "a-ruling" is defined by

C/^a = 0.

Each ruling forms a projective line CP;^, CP^ respectively.

Given a point u = 1*2, M3, U4] on Q, the matrices U and C/'^ each have rank
one. Thus, distinct lines in CP;'^ (or CP^) are skew while exactly one a-line and
one 5-line intersect at u. This gives the quadric a CPj x CP^ structure. (See
|Hodge and Pedoe 196^ , Ch. XIII: Quadrics.)

Furthermore, as a set, each ruling has an A5 stabilizer Qqq and, hence, CP^
and CP^ have icosahedral geometry. The "odd" elements Q120 — Geo exchange the
a-ruling with the 6-ruling.

2.4. Special orbits. The 3-dimensional ^5 action comes in both real and complex
versions. This means that, in the standard x coordinates, G120 acts on TZ — the
RP'^ of points with real components. Table ^ in Appendix ^ enumerates some
special orbits contained in TZ while Table || describes elements of Q that are fixed
by members of Gi2o- For ease of expression, I will refer to special points (or lines,
planes, etc.) in terms of the orbit size: "20-points" (10-lines, 5-planes). Also,
these points get a symbolic description in reference to orbit size (superscript) and
coordinate expression (subscript).

Corresponding to each special point a is the plane {a - x — 0}. In the case of the
10-points

[1,-1,0,0,0], ... ,[0,0,0,1,-1],

there are 10-planes

{xi = X2}, ... ,{X4 = X5}

that are pointwise fixed by the involutions

Xi ^ X2, ... ,X4 ^ X5.

These ten transpositions gener ate ^120 making it the pro jective image of a real or

complex reflection group. (See [ Shcphard and Todd 1954 1.)

Another noteworthy orbit is that of the five 54-stable coordinate planes
£l^{x,^0} z = l,...,5

SOLVING THE QUINTIC BY ITERATION

5

as well as the five octahedral conies

Ql = Qncl.

Some data for special two-dimensional orbits appear in Table ^ I describe these
sets in terms of dimension (superscript), orbit-size (subscript), and coordinate ex-
pression (sub-subscript).

Finally, a number of special lines appear as intersections of the 5-planes and
10-planes. Table ^summarizes the situation.

2.5. Configurations. Some of the geometry that will have dynamical significance
shows up in various collections of lines. First, the 10-lines

form a complete graph on the 5-points. Figure |l] illustrates this in two ways. The
pentagon-pentagram figure displays a 5-fold symmetry while the double pyramid
exhibits the structure of a single 10-line. (The illustration suppresses the '10'
subscript.)

The intersections of "complementary" pairs of 10-planes yield an orbit of 15-lines

This forms a graph on 15 vertices — the 5-points and 10-points pj^^.

• At a 5-point , there are three 15-lines

• On a 15-line , there is one 5-point where i ^ j, k, £, m.

• At a 10-point pjj^, there are three 15-lines

• On a 15-hne ^l^.. there are two 10-points p\^^, Pm^-

Within each of the icosahedral rulings on Q there are three special line-orbits.
These correspond to the 12 vertices, 20 face-centers, and 30 edge-midpoints of the
icosahedron. Intersections of lines between rulings give special point structures.

• Two 20-line C/go-orbits form ten "quadrilaterals" at two pairs of 20-points.
(See Figure |.)

• Two 12-line C/go-orbits form six quadrilaterals at 24-points.

• Two 30-line t/go-orbits orbits form 15 quadrilaterals at two pairs of 30-points.

Since Q120 — ^eo, exchanges the orbits in CP^ with those in CP^, these give overall
line-orbits of sizes 40, 24, and 60.

3. Equivariant maps
The primary tool to be used in solving the general quintic is a rational map

/ : CP''' — > CP3
with 55 symmetry. In algebraic terms, this means that

foT = Tof for all T e ^120.

Furthermore, such an equivariant map (or simply equivariant) should have reliable
dynamics: its attractor
1) is a single G120 orbit

6

SCOTT CRASS

2) has a corresponding basin with fuh measure in CP'^
2') alternatively, has a corresponding basin that is dense in CP^.
Recall that a point a in a space X is attracting when, for all x in some neigh-
borhood of a,

A point s is superattracting in a direction L if the derivative /'(s) has a zero
eigenvalue in the L direction. The basin of attraction of a is the set of all points
attracted to a;

Ba = {x(^X: fix) a}.
Also, the attractor of / is the set of all attracting points.

3.1. Basic maps. A finite group action Q on C" induces an action on the as-
sociated exterior a lgebra. Moreover, ^-invariant [n — l)-forms correspond to Q-
equivariant maps. [ Crass forthcoming Briefly, let

dZ^ = {-Ifdz,, A • • • A dz,^_,

where / is the ordered set

{ii,

I is the single index in {1, .
If

■ ,in-l} il < ■ ■■ < in-1,

n} — /, and ct/ is the sign of the permutation
1 2 ... n

I il ... in-l.

1=1

is a ^-invariant (n — l)-form, then the map

/(2)-(/l(z),...,/„(z))

Figure 1. Configuration of 10- lines and 5-points

SOLVING THE QUINTIC BY ITERATION

7

is relatively tj-equivariant (a multiplicative character appears under the action of
g on /).

For a reflection group, the number of generating 0- forms (i.e., polynomials) is
the dimension of the action. jShephard and Todd 1954, p. 282] From a result in
complex reflection groups, this is also the number of generating 1-forms and (n— 1)-
forms. [Orlik and Terao 1992, p. 232] Indeed, the 1-forms are exterior derivatives
of the 0-forms while the (n — l)-forms are wedge products of 1-forms.

Proposition 2. With = —4:X^ + ' /om?~ maps

h{x) = [Xl Xlxl XlX^] = 1, . . . , 4

generate the module of Qi2q equivariants over the ring of Qi2o-invariants.

These maps are projections onto the hyperplane Ti^ along [1,1,1,1,1] of the
power maps

r k k k k kl
[Xi , X2, X^, X^, Xg^ .

2 n T,^,,^Q

Q n z

Q n T„2oQ

2 n T,.o Q

Figure 2. Configuration of 40-hnes and 20-points on Q. At a
20-point qfj_ or qfji._ there are two 40-lines — one in each ruling
on the quadric. This pair of lines is the intersection of Q with the
tangent plane to Q at the respective 20-point. Also indicated are
the 10-lines determined by a pair of antipodal 20-points.

8

SCOTT CRASS

Proposition 3. Under an orthogonal action an invariant F{x) gives rise to an
equivariant f{x) by means of a formal gradient

fix) = V,F(x)

dF OF .

-— a; , . . . , - — [x)

OXi OXn

Proof. For a homogeneous polynomial F{x) of degree m, the Euler identity gives

mF{x) = VxF{x)'^x = VxF{x) ■ x = f{x) ■ x.
Invariance of F yields

mF{x) = mF{Ax) = y^^FiAxfAx.
Using an auxiliary variable y,

V^F{Ax) = A^ VyF{y)\y^A. = A^ f{y) = A^ f{Ax).
By orthogonality of A,

mF{x)=A-'^f{Ax)-x.
Equating expressions for m F{x) reveals equivariance:

A-^f{Ax) = fix).

□

Note that the <S5 -equivariant fkix) is not equal to VxFk+iix), but is a multiple of

WFk+iix)^^x^.

While this may be a source of confusion, it does not cause problems, since we are
working on the hyperplane Hx. When using hyperplane coordinates on Hu, the
discrepancy disappears.
A map on Hx produces

= HfiWu)
on Hu. Expressing the generating M-equivariants

M^) = HfkiH^u)
in terms of the basic M-invariants $fe('u) will be useful.
Definition 1. Let

/O 1\

10
10

\i oy

represent the reversed identity and reversed gradient.

and V;f(m) =i?V„F(u)

Proposition 4. In Hu coordinates, the map <piu) = H fiH'^u) is given by

<Piu) = Vl^iu)

where $(u) = FiWu) = Fix).

SOLVING THE QUINTIC BY ITERATION

Proof. For the change of variable u = H x and x = H'^ u, the chain rule yields

fix) = V^F{x) = V^$(u) = H^VuHu).

Since HH'^ = R,

Hf(x) = HH^V,Mu)
HfiWu) = RVu^u)

Thus, the basic maps in u are

□

5

Explicitly,

0l(w) = 2 [ui,U2,U3,U4]

2

02 (w) = — p [U3 + 2U2U4,Ui + 2U3U4,2uiU2 + U4,U2 +2ui Us]

v5

03 (w) = -r [lt2 + 6mi W2 W3 + 3 Ml M4 + 3 M3 W4, 3U2 W3 + 3 Wl M3 + 6mi M2 W4 + W4,

o

3 2 2 2 3 2l
Wl + 3 W2 M3 + 3 M2 W4 + 6 Ml W3 M4 , 3 Wi W2 + W3 + 6 M2 W3 M4 + 3 Wl W4J

04 (m) = -7= [4 Ml U2 + 4Ui U3 + 4m2 M3 + 12lt2 M3 M4 + 12 Ml M3 M4 + 12 Ml M2 M4 + M4,

V5

32 43 2 22 2

4 Ml M2 + 12 Ml M2 M3 + M3 + 4Mi M4 + 12 M2 M3 M4 + 6 M2 M4 + 12 Ml M3 M4,

4 2222 3 23
M2 + 12 Ml M2 M3 + 6 Ml M3 + 12 Ml M2 M4 + 4 U3 U4 + 12 M2 M3 M4 + 4 Ml M4,

4 3 2 2 2 2 2 3"!

Ml + 4 M2 M3 + 12 Ml M2 M3 + 12 Ml M2 M4 + 12 Mi M3 M4 + 6 M3 M4 + 4 M2 M4J .

3.2. A fixed point property. For a 5i2o-equivariant / and a point a that an
element T € ^120 fixes,

Tf{a)=f{Ta)=f{a).

Hence, equivariants preserve fixed points of a group element.
Being pointwise fixed by the involution

Xi ^ ^ Xj ,

a 10-plane

either maps to itself or collapses to its companion 10-point

=[...0..., 1 ,...0..., -l,...0...]^£fo„-

In the former generic case, the map preserves the 10-line and 15-line orbits MIq.
and £}r , , that are intersections of 10-planes.

10

SCOTT CRASS

3.3. Families of equivariants. The G120 equivariants form a module over the Q12Q
invariants for which degree provides a grading. This means that for an invariant
Fi and equivariant of degrees £ and m, the product

Pe ■ g-m

is an equivariant of degree £ + m. When looking for a map in a certain degree
k with special geometric or dynamical properties, my approach is to express the
entire family of "fc-maps" and by manipulation of parameters, locate a subfamily
with the desired behavior.

3.4. Quadric-preserving maps. The rich geometry of the quadric Q provides
an intriguing setting for dynamical exploration. Are there 55-symmetric maps that
send Q to itself? If so, how do they behave on and off Q? I will describe discoveries
of two species of such maps: one associated with the icosahedron and the other with
the octahedron.

Maps that preserve icosahedral rulings. Were a t/i2o-equivariant to preserve the
^5 rulings on Q, its restriction to either ruling CP^ or CP J would express itself
in terms of the basic equivariants under the one-dimensional icosahedra l action.

Such maps occur in degrees 11, 19, and 29. [ Doyle and McMuUen 1989 , p. 166]

Consequently, the 20-parameter family of 11-maps comes under scrutiny:

/ii = (ai F| + a2Fi Fi + Q3 ^2 F4 + 04 F| F4 + 05 F2 F| + Qe F2 F3 F5 + 07 Fi) /i
+ (as F| F,i + Q9 Fi + Qio F2 Fz F4 + an F| F5 + «i2 F4 Fs)f2
+ (ai3 Fi + Qi4 F2 Fi + ai5 F| F4 + «i6 f! + F3 F5) /g

+ (dig Fi F3 + aig F3 F4 + 020 F2 F5) fi.

From the geometric description of the icosahedral 11-map on CP^ or CP^
( |Doyle and McMuUen 1989| , p. 163]), a rulin g- preserving 11-niap would exchange

antipodal pairs of 20-hnes {C-20ii ^202} or {^20^, C^q.^} and 30-lines while fixing 12-
lines. (Re call that these are Qqq orbits.) Imposed on the configurations described

in Section 'Lb, these conditions require analogous behavior at the associated points:

„20 20 20 20

lijke * lijkl

„30 30 30 30

The specified action occurs automatically for qfj_, qfjke^ ^^'^ iTjk^- After solving
two linear equations associated with the remaining two conditions

f (n^O \ _ 20 f / 30 ^ _ „30

as well as four linear equations

/ll('C20i) = -^202

SOLVING THE QUINTIC BY ITERATION

11

that arrange for the exchange of an antipodal pair of 20-Unes in either ruhng we
obtain a 13-parameter family of ruling-preserving maps

gii = 4 (16 ai F| + 16 02 F2 + 16 Q3 F| F4 + 67 F| F4

+ 16 Q5 F2 Fi + 16 aa F2 F3 F5 + 45 F^) /i

+ 4 (16 as Fi F3 + 16 F| + 16 aio F2 F3 F4 + 16 an F| F5 - 135 F4 F5) /a
+ (64 ai3 + 64 Qi4 F2 Fi + 64 ais F| F4 + 405 F| - 720 F3 F5) /s
+ 4 (16 ai8 F| F3 - 225 F3 F4 + 16 Q20 Fa F5) /4.

When restricted to the ruling CP;^ and expressed in the homogeneous ruling coor-
dinates [ai,a2], the map has the elegant appearance

5iilcpi ■ [01,02] — > [fli {- al" + 66 al al + 11 al"), a2{nal°- 66 alal-al")].

Of course, the same form appears for the 6-ruling.

Restricted to a ruling, the dynamics of each gu is completely understood. The
20-lines are period-2 and the only elements of the critical set. (Recall that 20-lines
in Q are dodecahedral vertices in CP;'^ or CP^.) This implies that almost every line
in the ruling belongs to the basin of one of the ten pairs of the superattracting set.
(Sec [ [Doyle and McMuUcn 1989| , p. pp. 166-167] and Figure | in Appendix

Thus, for almost every point qa on Q, there is an "antipodal" pair of intersections
between 20-lines in each ruling toward which gu attracts the trajectory:

5W(go) ^ {^0, n /:k,^2o. n4o,} ihj} = {1,2}.

As a result, the global behavior of each gu depends on its dynamics off Q. Should
the quadric attract or repel? If Q were attracting, then the 400 intersections of
20-lines would attract in all directions. One way to arrange for this is to force these
points to be critical in the off-quadric direction. However, this situation does not
conform to the model of reliable dynamics. The attractor would not be a single
Gi2o orbit of points, though it might be the set of intersections of a single Hne-orbit.
I have not explored the case of a repelling quadric.

Interestingly, the quadric resists criticality. Computation reveals that no member
of is critical on all of Q. Is there a geometric reason for this? The next example
reveals that this is not a universal trait of quadric-preserving maps.

An octahedral map. Since the orbit of five planes >C|^ has fundamental geometric
significance, a map that preserves these sets might exhibit interesting dynamics.
Arranging for this spends four of the twenty parameters of the family /n.

The intersection of a 5-plane £|^and Q is a conic Q^, with 54 symmetry and,
thereby, octahedral structure. One of the special equivariants for the octahedral ac-
tion on CP""^ is a 5-map that attracts almost every point to the eight face-centers —
vertices of the dual cube. Geometrically, the map stretches each face F of the
cube symmetrically over the five faces in the complement of the face antipodal
to F. As a face stretches, it makes a half-turn so that the vertices land on their
antipodes. This makes each vertex critical and period-2; locally, the map is squar-
ing. Since these are the only critical points, their basins have full measure. (See
Doyle and McMuUen 1989| , p. 156] and Figure | in Appendix |c|.) Under C/120,

antipodal pairs of octahedral face-centers are the 20-points qfj^ , qf^^ .

The idea is to look for a reliable map with the 20-points as its only attractor.
In degree five there are too few parameters for the purpose. However, the 11-maps

12

SCOTT CRASS

provide enough freedom to arrange for elegant geometry. The goal demands that
the desired map hn preserve the 5-conics and then decay to the octahedral
5- map there. One way to realize this is to self-map the quadric Q. This takes six
of the remaining 16 parameters the expenditure of one of which assures that the
20-points do not blow up.

Intriguingly, when any member gn of the resulting 10-parameter family restricts
to Q, it decays into a 5-map

5ii|q = -^J^3 (2i^3/2-^4/l)|Q.

This decadence occurs unexpectedly, since most octahedral 1 1-maps exchange pairs
of face-centers and are non-degenerate. When restricted to an "affine" part of the
quadric

Qn{m ^0},

the maps have the simple form

+ 3y-2xy^ 3x'^ + 2y + x^y^

5ii|cn{«i#o} : ix,y)

2x + 3x^y'^ — 2/3 ' l + 2x^y — 3xy^

Is there a geometric description of the restricted map?

Every member of the gn family preserves the iSa-symmetric conic

each of which contains a pair of 20-points g?? , qfj^ . In coordinates where these
points are and oo,

Of course, the period-2 points and oo are critical. By experiment, the remaining
six critical points belong to their supcrattracting basin. Such circumstances force
almost every point on a conic to belong to the basin.

Octahedral 11-maps generically exchange antipodal pairs of vertices. Such a pair
corresponds to the 30-points Qi'jj^^ , Qi'jk2 ■ ^ degenerate member of the family,
the 5-map fixes these points. These conditions require each gn to blow up the
30-points. Also blowing up are the 24-points.

Now the issue is behavior off Q. Since the desired attractor lies on Q and
the dynamics there appears to be reliable, a map for which the quadric is itself
attracting comes to mind. Because octahedral face-centers are superattracting on
the respective conies, each gn is critical at the associated 20-points. The maps
are also critical at the blown-up points. Arranging for critical behavior at the
three quadric orbits consisting of the non-octahedral 20-points, 30-points and the
octahedral 60-points costs three parameters. The result is a seven parameter family
of 11-maps for which the entire quadric is critical and each octahedral 20-point is
superattracting in three directions.

Each of the 10-lines jO-Iq.. contains a pair of antipodal 20-points. A map that

1) preserves these lines,

2) attracts almost every point on the line to the 20-points, and

3) is critical in the directions off the line

SOLVING THE QUINTIC BY ITERATION

13

would act as a "superattracting pipe" to the quadric. Expenditure of four of the
remaining seven parameters purchases a map with these properties. Indeed, when
restricted to each CIq.^ , the map is

1

with the pair of 20-points at and oo.

The final three parameters allow for a map hn with a non-attracting pipe to Q
at the 10-lines MIq.^^:

/ill = (-21 F| + 56 f| + 66 Fi F4 + 48 F4 - 48 F2 f| - 96 F2 F3 F5) fi

- 24(4F|-9F2F3F4 + 3F|F5)/2+ 12 (5 Fj' + 8 Fa F| - 10 F| F4) /g

- 96FIF3/4.

Such a line contains the pairs of 20-points qfji.-^ , q^jk^ ■ coordinates where these
points are and cxd, the restriction of hn to J^w..,, is

1

On Q these 20-points are repelhng. Indeed, they belong to the conies

Qc^cl^^^, Qn£?o,., Q^t^w,,

on which the basins of the pair of 20-points ql^_ , , have full measure.

Experiment reveals that nearby points belong to the basins of the other 20-point
orbit.

Due to its geometry, hn preserves the various CP^ intersections of 5-planes and
10-planes. The two such lines not yet considered are the 15-lines ^{5^. and the
30-lines £30. . In "symmetrical" coordinates where the intersections with Q are
at and 00, the restricted maps are:

19 z2

z2(9z2 - 19)

llz2 + 9
"z2(-9^2 iiy

In the former case, the map has attracting fixed points at the pair of 10-points
Pi% ' Pkt2 ^'^'^ ^ period-2 superattractor at qf^f^i^ , qf^ki^, ■ Overall, these are saddle
points where the map repels off the line. A similar state of affairs occurs on the
30-lines. Here, the pair of attracting fixed points is pf^jf^g, PTjkm ^^'^ period-2
superattractor is at (zfj^^, Qi^jk2- Oiice again, at these points hu is repelhng off
the line. Dynamical experiments on the respective lines show that these points
attract all six critical points. Thereby, the associated basins have full measure on
the lines. Basin portraits for these restrictions appear in Figures ^ and ^ Since
almost every point on these these lines is in the basin of an overall saddle point,
the lines themselves behave as saddles and, thereby, are measure-zero pieces of the
Julia set J^^^ .

Since the pair of 15-lines and A^is.^ are pointwise fixed by the involu-

tion

a t/120-equivariant that does not smash down CIq._ permutes these lines as sets.

Xk < — > xe,
•2

-lOi

14

SCOTT CRASS

Fact 2. Under hu, M.\^.- ^f. maps to itself. With the pair of30-points q^ij^^; 1mij2
at and oo,

I z{z^ + 6)

^^ij, ki D + 1

This map has non-critical, attracting fixed points at p^j*^ , p\^_^ . Since the four critical
points belong to the associated basins, the dynamics on the line is reliable. Also
passing through the attracting 10-point pjj^ are three 10-lines ^lo^^ (fc,^ ^ i,j) so
that, at this point, hu repels away from the line. Hence, this line also lies in the
Julia set.

The special geometry of hu forces a number of points to blow up:

„5 10 10 15 20 30 30 24

-Pi 1 yiji^ "ii2 ' yiJk'J yijki^ t^ij,kt^ ^i,jk-J ^ijki'

Experimental evidence suggests that neighborhoods of these blown-up points are
filled by basins of the octahedral 20-points. Indeed, the CP^ of directions through
a 10-point plj_^ maps to the point itself. Lying at the intersection of three 10-lines
^lOki (^'^ 7^ such a location might be called super-repelling. In contrast, the
directions through a 30-point p^ji^g blow up onto the superattracting 10-line CIq.^
whose "basin" is that of the 20-points qfj_ .

Since the coefficients of hu are real, the map also preserves TZ — the iSs-symmetric
RP"^ — as well as the RP^ intersections of TZ with £5 . and C^q.^ . In the former case
there are four RP^ intersections of the RP^ with the 10-lines jCIq.^ while in the
latter there is a single such intersection. The stabilizer of the respective 5-plane or
10-plane fixes its resident RP^s. Thus, each such RP^ is an "equatorial slice" of
the associated CP"'^. Being equivalent to the map

1

z^

on the unit circle {\z\ = 1}, hu acts chaotically when restricted to such a slice.
Hence, each RP^ is a chaotic attractor on the respective RP^. A basin portrait for
the 5-plane reveals no basins other than those of the four 10-lines. (See Figure 0.)
The RP^ dynamics on the 10-plane shows, in addition to the chaotic line-attractor,
three additional basins at the 30-points pfjjfe^- (See Figure ^.) A 30-point belongs
to the 10-line 'CJq^^, which intersects the 10-plane >Cfg. . transversely. Thus, in a
neighborhood of the 30-point, but off the 10-plane, there is only the "pipe-basin"
of the 20-points qfj_ . Hence, the basins on the 10-plane are 2-dimcnsional.

Conjecture 1. The 20-point orbit is the attractor for hu and the corresponding
basins have full measure in CP"^.

Iteration experiments on TZ reveal attraction only to the ten chaotically attract-
ing RP^ intersections TZ n £10^^ •

Conjecture 2. The S^-invariant RP^ is non- attracting (repelling?) and so be-
longs to hu 's Julia set.

3.5. What to look for in an attractor. A pair of 20-points g^j'^, qfj^ associates
canonically with an orbit of ten lines. However, there is no such correspondence
between a pair of 20-points and an orbit of size five; the 20-points do not decompose
into five sets of four ^4 orbits. An association of this kind makes for a natural
solution to the quintic. What could serve the purpose better than a map whose
attractor is the 5-point orbit?

SOLVING THE QUINTIC BY ITERATION

15

3.6. A special map in degree six. In the configuration of 10- lines -MIq..^ each.
5-point hes at the intersection of four hues. (See Section |2.4 ) Moreover, these
are the only intersections of 10-lines. To take advantage of this structure, a map
could have superattracting pipes along the 10-lines and basins of attraction at the
5-points.

The family of 6-maps has (homogeneous) dimension six. Obtaining maps for
which the 10-lines arc critical in the "off-line" directions uses four parameters. For
the remaining two, we get a map /g whose restriction to a 10-line is

in coordinates where the 5-points on are and oo. In hyperplane

coordinates,

06 = 2 (9 $2 $3 - 10 $5) 01 - 2 ($i - 5 $4) (f>2 + 20 -1'3 03 -f 15 $2 04

= [2 Ml — 4 Ml M2 — 74 Ml it2 M3 — 46 Ml U2 M3 — 14 M2 M3 — 2 Ml M3 — 38 Mi M2 M4

4 32 3 42 22

— 44 Ml M3 M4 — 50 M2 M3 114 — 122 Ml M2 M3 M4 — 14 M2 M4 — 152 mi M2 M3 M4

— 68 Ml M3 u\ — 72 u\ U2 m| — 22 U3 u\ — 29 M2 u\ — u\u%,

— 2 Ml M2 -|- 2 U2 — 44 Ml U2 ^3 ~ 68 Ml M2 M3 — 22 u'l M3 — U2 M3

23 3 23 4 42

— 46 Ml M2 M4 — 122 Ml M2 U3 U4 — 72 M2 II3 M4 — 29 Ml M3 M4 — 14 Ml M4

32 22 23 23 24 5

— 38 M2 M3 M4 — 152 Ml M2 U3 U4 — 74 Itl M2 M4 — 50 Ml M3 M4 — 14 M3 M4 — 4 U2 U4 ,

— 14 Ml M2 — 4 Ml M3 — M2 M3 — 72 Ml M2 M3 — 38 Ml U2 M3 -f 2 M3 — 29 Ml M2 M4

22 32 4 32 222

— 152 Ml M2 M3 M4 — 74 Ml M3 M4 — 44 M2 M3 M4 — 50 Ml M2 M4 — 68 M2 M3 M4

— 46 Ml M3 M4 — 22 M2 M4 — 122 Ml M2 M3 M4 — 14 Ml M4 — 2 M3 M4 ,

— 22 Ml M2 — 29 Ml M2 M3 — 14 M2 u\ — 50 mi M2 M3 — 14 Mi M3 — mJ" M4 — 2 M2 M4

— 122 Ml M2 M3 M4 — 152 Ml M2 M3 M4 — 4 M3 M4 — 68 Ml M2 M4 — 72 Ml M3 M4

— 74 M2 M3 M4 — 46 M2 M3 M4 — 38 Ml M3 M4 — 44 Ml M2 M4 + 2 M4] .

By construction, /g self-maps each 53-symmetric 10-plane . The 10-point
pI^^ and 5-points p| (fc ^ form ^3 orbits on C\q.. of sizes one and three.

Furthermore, /e preserves TZ — the 55-symmetric RP'^. We can get a picture of the
map's restricted dynamics by plotting basins of attraction on the RP^ intersection

(See Figures ^, through |l2| in Appendix ^) The plot shows attraction to the 5-
points and the 10-point. However, the 10-point lies on the "equator" of an A^io^^^
{k,i,m 7^ i,j) where /g repels in the off-plane direction. Thus, the 2-dimensional
basin of a 10-point is a measure-zero part of J/g . No other attracting sets appear.
Moreover, regions of positive measure that do not belong to one of these four
"restricted basins" are not evident. The plot is consistent with the claim that the
only fully 3-dimensional basins are those of the 5-points.

A 15-linc contains one 5-point p^„, one 15-point p^jj (m ^ k,£), and

two 10-points Pij^, Pm^- In coordinates where the 5-point is 0, the 15-point is 00,
and the 10-points are ±1 the map restricts to

^ ' -3-z2 + 35z4_^ 17^6 ■

16

SCOTT CRASS

The critical points of the restricted map are

0, ±1, ±1

'9±4\/2T
17

with 0, ±1 fixed. Experiment reveals that the four non-fixed critical points belong
to the basins of the three superattracting points. Hence, these basins have full
measure on the 15-line. (Figures |l^ and ^ display portraits.) As a member of
three 15-lines ^l^.^ a 10-point pY^^ superattracts in these directions. However,
these three lines lie in the 10-plane C\q.. so that, as seen above, Jq is completely
superattracting in the plane at p\^^ .

Another distinction for /g is its action on a 15-line which, by equivari-

ance, must map either to itself or .

Fact 3. Under /g, Ml^-- maps to C\^.^ . Effectively, this creates a second orbit
of superattracting pipes to the 5-points.

This is what led me to 6-maps, each of which send the 10-point pjj_^ to the 10-point

Finally, noting that (/ig has real coefficients, it must preserve the RP"^ whose
points have real u coordinates. This is not the iSs-symmetric TZ. Rather it seems
to be associated with the 54 stabilizer of pf which is [1, 1, 1, 1] in the u space. This
RP'^ intersects the 10-planes C\q^^ and in an RP^ with Z2 x Z2 symmetry.

In addition to p\ this RP'^ contains the 10-points P25i ; -P252 ' P342 ^ ^'^^^
RP^ through p}^_^ and P25, • Since this line is an equatorial slice through ^310125 '
attracts chaotically along the line. (See Figure |l^ for a basin portrait.)

Graphical and experimental evidence supports the claim of reliability for /g.

Conjecture 3. The attractor for /g is the 5-point orbit the basins of which fill up
CP^ in measure.

4. Solving the quintic

To compute a root of a polynomial, one must overcome the symmetry present.
For a general equation of degree n the obstacle is 5„. Klein described a means
to this end: given values for an "independent" set of 5„-invariant homogeneous
polynomials

ai = Gi{x), . . . , a„ = Grnix),

find the 5„ orbits of solutions x to these equations. [ Klein 1956 , pp. 69ff] This task
of inverting the Gk is the form problem on 5„. It also has a rational manifestation:
for m — 1 given values, invert m — 1 invariant rational functions of degree zero.

An Sn equivariant with reliable dynamics breaks the obstructing symmetry. In
effect, this provides a mechanism for solving the form problem and, hence, the nth
degree equation. What follows is one way to use C/120-symmetry in multiple settings
to assemble a procedure that solves almost any quintic.

4.1. Parameters. The G120 rational form problem is to solve

<i>2(u)2 $2(^)3 $2 (u) $3 (w)

SOLVING THE QUINTIC BY ITERATION

17

As functions, the Ki define the ^120 quotient map

[K,, K2, K^, 1] = [$2 $3 $4, $1, $2 ^5, $2 ^3]

on CP^ \ {$2 = $3 = 0}. The generic fiber over points in CP^ is a ^120 orbit given

by

{$4 = i^i $2} n = i^2 n {$5 = ii^s $2 $3}.

Exceptional locations are [0, 1, 0, 0] and [0, 0, 1, 0] where the respective fibers include
the quadric and cubic surfaces {$2 = 0} and {$3 = 0}.

Between quintic equations and ^120 actions the parameters Ki forge a link. The
connection consists in if-paramctrizations of each regime. From a parametrized
family of ^120 actions, we can extract parametrized families of S5 invariants and
equivariant 6-maps. In this way, a choice of parameter K produces a quintic Rk
as well as a system of invariants $2jc (w), . . . , ^5j^{w), and a 6-map (j)K(w) — a con-
jugate of 4>q{u) — on a parametrized w-space.

4.2. A family of 1S5 quintics. Let Qy be a version of that acts on a v-
coordinatized CP^. This will be a parameter space — the coordinate v merely stands
in for u. The linear polynomials

Xk{x) = -^Xk + ^Xi
form an orbit of size five. In hyperplane coordinates, these are

LM =

-VS (Wl + 1^2 + W3 + ^^4)

U{v) =

— -/S W5 (uj\ V\ + U2 +

U{v) =

— -v/S W5 til + W2 +

W3 +

CJs ?74 )

U{v) =

— VS 0^5 (ws Vx + wf U2 +

W3 +

^5^4)

U{v) =

-a/5 W5 {v\ + W5 i;2 +

wf ^74)

The rational functions

<E>2(^')ivfc(f)

also give a 5-orbit. Taking the as roots of a polynomial

fe=l fe=0

yields a family of quintics whose members generically have symmetry. Since
permutes the Sk{v), each coefficient Ck{v) is CJ„-invariant and hence, expressible
in terms of the basic forms ^k{v) and, ultimately, in terms of the Ki. Of course,
Co{v) = 1. Since there is no degree-1 ^120 invariant, Ci{v) = 0. Direct calculation

18

SCOTT CRASS

determines the remaining coefficients:
125$2(^J)^ _ 125

C2{v)
C4{v)

2$3(«)2 2^-2(1;)
625V5$2{vf _ 625 V5

15625 {^2{vf - 2$2{v)'^^4{v)) _ 15625 (1 - 2 7^1(1;))

8$3(«)4 ^ 8K2{vY

-15625 (5V5$2(«)^<&3(«)-6^/5$2(^;)''$5(^^)) _ -15625 ^ (6 7^3 (w) - 5)

6<E'3(w)5 6A'2(w)2
Members of the 3-parameter family of quintic G120 resolvents

^ , 125 . 625V5 , 15625 -1 + 2 if 1 15625^5 -5 + 6^3)
Rji (s) = H — ^ ^ — s H — ^-^ —

are particularly well-suited for an iterative solution that employs 06- For selected
values of the Ki^ a solution to the resulting form problem yields a root of Rk- Use
of symmetry will provide a means of finding such a solution without explicitly
inverting the Ki equations (|^).

4.3. Reduction of the general quintic to a Q120 resolvent. By means of a well-
known linear Tschirnhaus transformation the general quintic becomes the standard
4-parameter resolvent

q{y) = + &2 + ^3 + 64 y + 65-
Application of another linear Tschirnhaus transformation

y

converts the 3-parameter family Rk{s) into a Gi2a resolvent

^kAv) = A' Rk (I) = + C2 y' + C3 y^ + C4 y + A^ C5

in the four parameters Ki, K2, K3, and the auxiliary A.
The functions

bk = A'= Ck

relate the coefficients of q and E/f,A- These invert to

6^-264 -952 5(6263-65) , -363
^ J^2 — 5~ = A = ^

2bl 8&3 66263 IOV562

Thus, almost any quintic descends to a member of Rk- The reduction fails when

-2ai + 5a2 = 562 = or 4 - 15 ai 02 + 25 03 = 25 63 = 0.

A solution to the special resolvent Rk then ascends to a solution to the general
quintic.

SOLVING THE QUINTIC BY ITERATION

19

4.4. A family of ^5 actions. With the basic t/^-maps, construct the parametrized
change of coordinates

\U4/

u = TyW = 2_^{^e-i{'") (t)i{v))wi.
1=1

A matrix form results from taking the (pkiv) as column vectors:

^f,{v) <l)i{v) : ^i{v)<l>2{v) : ^■i{v)<l>3{v) : ^2(v) <l>4,{v)
For a choice of parameter w,

is linear in w and gives rise to a parametrized family of Q120 groups

The setup here is as follows.

• ^/ii is a version of Q120 that acts on a reference space CP^.

• C/t, is a version of Q12Q that acts on a parameter space CP^.

• Qu and Qy have identical expressions in their respective coordinates.

• 5^ is a version of ^120 that acts on a parametrized space CP^,.

• The iteration that solves quintics in Rk will take place in CPf„.

Each Q"^ has its system of invariants and equivariants. From this point of view,
we can see, in the resolvents and equivariants, a connection between quin-
tics and dynamical systems. Furthermore, each invariant and equivariant is
expressible in the Ki.

The first thing to notice is that, by construction, t^w possesses an equivariance
property:

TAvW ^ At^w for Ae Gv,Gu-

The determinant of will enter into upcoming calculations and so, demands some
attention. Since

(3)

\TAv\

\A\\

\ttj \ is invariant under the subgroup Qqo oiQv but only relatively invariant under
the full 55 group Gi2o- The even transformations have determinant 1 while the odd
elements have determinant —1. Furthermore,

|-r„| = ^2{v)^3{v)^4{v)^5{v) \Mv) I Mv) I Mv) I Mv)\

= $2(f) $3(w) ^4{v) $5(f) *lo(f)

where 5* 10 is a scalar multiple of the product of the ten linear forms associated with
the ten planes of reflection that generate Gv Reflection group theory tells us that
this is the only form in degree ten that is invariant under Geo but not Gi20- From
(||), the degree-48 square of |r„| is ^120-invariant. Let

\T,\' = ^l\v)tK

determine its /^-expression. The explicit form of tx appears in Appendix

20

SCOTT CRASS

4.5. A family of ^5 invariants. The equivariance in v of TyW implies that ^2(tvw)
is tJij-invariant. Thus, each w coefficient of ^2{Tyw) inherits the same invariance.
Since

deg„ ^2{tvw) = deg„ $2(") • deg^ TyW = 2 • 6 = 12,
the rational function

is degree zero in v and thereby, expressible in the Ki. Let

(4) $2W$2,,H-$2(w)

define the basic degree-2 Q"!^ invariant $2a- i"*^)- Solving a system of linear equations
whose dimension is that of the dcgrce-12 invariants yields an explicit expression
in the Ki for each w-coefficient of ^2iTyw). Similar considerations apply in degree
three where

(5)

The results appear in Appendix |b|.

By Fact |l|, the degree-4 and degree-5 invariants derive from those in degrees two
and three. First of all, the chain rule determines transformation formulas for the
hessian and bordered hessian.

Proposition 5. For y — Ax,

H^{F{y)) = A^Hy{F{y))A

B.AF{y)My))= ^^By{F{y)My)){^Q 5)

where the subscript indicates the variable of differentiation. Thus,

\H,iFiy))\ = \A\^\HyiFiy))\
\BAF{y),G{y))\ = \A\^\By{F{y),G{y))\.

Applied to the parametrized change of variable w — t^^u,

Gi{u) = |i7„($3(w))|

ff„($2(«)'$3AH)

{^2{vf

^2iv)'
tK

^2{vY^G^Aw

SOLVING THE QUINTIC BY ITERATION

21

and

G5{u) = |B„($3(w.),4>2(w))|

= S4$2(«)'$3kH,*2(^^)'$2^H)

B^(*3kH,$2kH)

$2(v)

15

tK

Employed here are the obvious definitions

tK

tK

With natural definitions for ^4^{w) and <I>5^(w),

*4(n) = ^(f2W'-5G4(u))
1

324 (Mv)'^ $2. («^)' - 5 ^2{v)'' G,, (w))

and

= ^(720$2(U)$3(«)+G5(W))
oD4

= ^ (720 $2(1^)'' $2. {w) $3k H + $2(t;)'' Gs, («;))

= $2(^;)'' ^5^H-

4.6. A family of ^5 equivariant 6-maps. Emerging from each action is a ver-
sion (Pq{tvw) of (j)&{u). Being ^^-invariant, these maps also admit parametriza-
tion by K. Thereby, each quintic Rk enters into association with a dynamical
system on CP^.

The reversed identity R and gradient V = iiV appeared in the context of a
change from five x coordinates to four u coordinates. In the present setting, a
reversed transpose arises.

Definition 2. The repose of an n x n matrix A is its reflection through the
reversed diagonal — the entries whose subscripts sum to n + 1. Alternatively,

A'' ^RA^ R.

Proposition 6. For a change of coordinates u = Aw and a polynomial $(u)
the reversed gradient map transforms by

Vmu)=A^Vmw).

22

SCOTT CRASS

Proof. Noting that ^ I,

= RA^ RRV^^{w)

□

For the generating Qi2a maps,
Thus,

Using the description on the left-hand side, a straightforward calculation reveals
this map to be invariant in v so that the matrix Ty has entries that are degree-12
Qy invariants. Hence, the matrix product has a X-expression:

r-ry = <i>2{vfTK or {t^ Ty)-' -

(See Appendix |b| for the explicit form.) Using this to express the transformation
of basic equivariants yields

= ci,2(z;)3(^-l)r„</>,,H

where

Finally, we can identify a iC-parametrized 6- map 4>k{w) that is conjugate to
06 (w). The map's expression in basic terms appears after substitution into the
formula found in Section 3.6. (See Appendix |b|.)

4.7. Root selection. Being conjugate to (t>(,{u) each shares the former's

conjectured reliable dynamics. Accordingly, the attractor for each choice of Ki is
the 5-point orbit in the corresponding CP^ so that for almost every wq G CP^,

</'k(w'o) — > T~''Pi for some 5-point e CP^.

To solve the resolvent Rk, the output of the iteration must link with the roots of
Rk- With this, we see that solving Rk amounts to inverting Ty — the form problem
in yet another guise. With the assistance of a 5i2o tool, this is effectively what the
dynamics of (j)K accomplishes. (This clever device is due to McMullen.)

To manufacture the root-selecting tool, we begin with an orbit of quadratic 1S4-
invariants

SOLVING THE QUINTIC BY ITERATION

23

These form a ^120 orbit of size five. Their hyperplane expressions are
Qi{u) = 2 (2 + 4 wi W2 + 2 M2 + 4 wi W3 + 3 M2 W3 + 2 M3 + 3 Ml W4
+ 4 U2 W4 + 4 wa W4 + 2 W4)

( 3 2 2 2 2 2

2 Ml + 4 LUr, ui M2 + 2 U2 + 4 wi wa + 3 U2 wa — 2 W3 — 2 0)5 wa

— 2 0)5 Ma — 2 Olf M3 + 3 Ml W4 — 4 M2 M4 — 4 M2 M4 — 4 ^5 M2 M4

— 4 M2 M4 + 4 1^)5 M3 M4 + 2 M4)

Qa (m) = — 2 (—2 0)5 Ml + 4 Ml M2 + 4 tj.s mi M2 + 4 a;| mi M2 + 4 a;^ mi M2 — 2 m^

— 4 0)5 Ml Ma — 3 M2 Ma — 2 M3 — 3 mi M4 — 4 W5 M2 M4 — 4 Ma M4
+ 2 M4 + 2 U4 + 2 U4 + 2 U4)

( 2 22232 32

2 Ml + 2 Ml + 2 0)5 Ml + 2 0^5 Ml — 4 Ml M2 — 2 0^5 U2

— 4 Ml Ma — 3 M2 Ma — 2 M3 — 3 mi M4 — 4 015 M2 M4 + 4 Ma M4

+ 4 aj5 Ma M4 + 4 oj'^ Ma M4 + 4 a)| Ma M4 — 2 0^5 M4)

Q5 (m) = 2 (2 Ml + 4 0)5 Ml M2 — 2 M2 — 2 [jJ5 M2 — 2 tJs M2 — 2 CiJs M2 — 4 Ml Ma

— 4 oJs Ml Ma — 4 Ml Ma — 4 Ml Ma + 3 M2 Ma + 2 Ma + 3 Ml M4
+ 4tJ5 M2 M4 + 4a;5 Ma M4 + 2tj| M4).

Furthermore, each of the five forms

3

Gkiu) = -—Lk{uf + Qkiu) k = l,...,5

vanish at the 5-points with £ k but not at

Now, to draw the roots of the quintics Rk{s) into the game, consider the rational
function

where a is a constant to be determined. Since the u-dcgrcc of the numerator and
denominator is 15 = 2 • 6 + 3 while the w-degree is 2, the function is rationally
degree zero in both variables. At a 5-point T~^p| in CP^ four of the five terms in
Jy vanish; this leaves

Setting

^ <i>2(pf) _ _ $2(pi) _ 1

" GM) "' Gabi) 15
"selects" the root Se{v) of Rk{s). Since the iterative "output" of (j)K{w) is a single
5-point in CP^, the dynamics produces one root.

The root-selector Jy (w) has invariance properties that allow it to exhibit a useful
form. Let

5

rv{w) = ^Gfc(T„w) Lk{v).

fe=i

Since Qy permutes its terms, Fy is invariant under the action and hence, expressible
in K:

24

SCOTT CRASS

(The explicit form of Tk appears in Appendix |^.) Finally, application of (^) yields

Jv{w)

Jk{w)

15$3(w) <^2{TvW)

Tk{w)

4.8. The procedure summarized.

1. Select a general 5-parameter quintic p{x).

2. Tschirnhaus transform p{x) into a member Rk{s) of the 3-parameter family
of Qi2o quintics — this determines values for Ki, K2, K3 as well as the auxiliary
parameter A.

3. For the selected K values compute the invariants {i = 2,3,4,5),
the 6-map the form Tj^iw), and the root-selector Jk{w). (In fact,
a rather leng thy once-and-for-all expression for (ppciw) is easy to compute.
[ Crass 1999a Such a formula renders calculations of ^2k, ^3k^
superfluous .)

4. From an arbitrary initial point wo iterate 0_ff until convergence:

(I^Kiwa) > Woo-

Conjecturally, the output Woo is a 5-point in CP^.

5. Compute a root S = J k{w 00) of Rk-

6. Transform S into a root oip{x)-

(At ICrass 1999a , there are Mathematica data files and a notebook that implement
the iterative solution to the quintic.)

SOLVING THE QUINTIC BY ITERATION

25

Appendix A. Special orbit data

For ease of reference, the following tables provide descriptions of the special ^120
orbits that bear upon the quintic-solving algorithm.

Table 1. Special points on {[a;i,a;2,a;3,a;4,a;5] | Xk € R} — RP^

Size

Representative

Descriptor

Stabilizer

5

[-4,1,1,1,1]

Pl

10

[0,0,0,1,-1]

S3 X Z2

10

[2,2,2, -3, -3]

P452

.Sa X Z2

15

[0,1,1,-1,-1]

Pl%3 = P\%5

20

[0,-3,1,1,1] [-3,0,1,1,1]

20 20
Pl,345 P2,345

S3

30

[0,0,1,1,-2]

pIU

Z2 X Z2

Table 2. Special points on Q = {J^l^i xl = 0}

Size

Representative

Descriptor

Stabilizer

Remarks

20

[0,0,l,u;3,c^|]
[0,0,l,a;|,W3]

9l2i
9l22

Ze

antipodal pair of
eight octahedral
face-centers on

Ql,Qh

Q\ = ^i, n Q

20

[1,1, 1, a, a] [1,1,1, a, a]

?123i 91232

.Sa

a = -3+/^'

24

[l,a;i,a;i,wf,w|]

Hijkt

Z5

30

[0,l,i,-l,-i]
[0,l,-i,-l,i]

30 _ 30

91.241 — 91,352
„30 _ „30

91.242 — 9l,35i

Z4

antipodal pair of
six octahedral
vertices on Qi

30

[1,1, A A -2(1 + /?)]
[1,1,^,^,-2(1 + ^)]

9l2,34i
912,342

Z2 X Z2

/3 =

60

[0,1,1,7,7]
[0,1,1,7,7]

9l,12i
9l,122

Z2

antipodal pair
of 12 octahedral
edge- midpoints
on Ql;

7 = -l + V2i

26

SCOTT CRASS

Table 3. Some fundamental CP^ orbits

Size

Algebraic
definition

Corresponding

point

Descriptor

Set-wise

stabilizer

Point- wise

stabilizer

Restricted

action

5

{xi = 0}

54

Zi

54

10

{Xi =Xj}

Fiji

53 X Za

Za

53

10

{Xi = —Xj}

53 X Za

Zi

53 X Za

Table 4. Special CP^ orbits

Size

Algebraic definition

Descriptor

Set- wise

stabilizer

Point- wise

stabilizer

Restricted

action

10

53 X Za

Za

53

10

^10,, n £i()^.^ n £i().^

53 X Za

53

Za

15

Vi

Za X Za

Za

15

Vi

Za

Za X Za

30

Za X Za

Za

Za

Appendix B. Parametrized forms

Each case below requires Qy invariants to be expressed in terms of the basic
invariants <^i{v). This amoimts to solving a system of linear equations whose di-
mension is that of the respective space of invariants. Direct substitution into the
basic-invariant expressions then leads to the descriptions in K.

Basic invariants. Each «;-coefficient of $£(t„w) is a degree-6^ invariant in v. In
terms of K, the forms in degrees two and three are:

$a(r„w)

48

(240 K2 kI wl + 480 X 1 K2 K3 wi W2 - 48 wi -|- 240 Kf wl

-I- 480 Kx K2 K3 wi W3 - 96 Ki K2 wa wz + 480 Ki K2 K3 wa wa - 30 K2 wf
+ 180 Ki K2 wl -h 32 kI wl + 480 K2 kI wi wa - 60 Ki W2 wa
-\- 264 /sTi W2 Wi + 160 Ki K2 wa W4 — 140 K2 wz wa -\- 184 K\ K2 ws Wi
+ 336 K2 JTa wa W4 - 15 u;4 -I- 60 Ki wl + 12 Kf wl + 128 K2 K3 wl)

l9

-|- (-43200 Ki Ki Wi + 25920 Ki K2 Ki wl W2 - 129600 Ki K2 Ki W2

1728

+ 51840 Kf K2 Kz Wl wl - 129600 kI K2 Ki wi wl + 1944 kI wl

- 6480 Kt wl - 14400 kI K2 wl + 25920 kI Ki wl wz - 129600 kI Ki wl wz

SOLVING THE QUINTIC BY ITERATION

27

+ 32400 Ki K2 K3 wi W2 W3 - 142560 K2 K3 wi W2 W3

- 34560 Ki K2 K3 wi W2 W3 + 27432 kI K2 wl W3 - 49680 kI K2 wl W3

- 38880 kI K2 Ka wl W3 + 37800 K3 wi W3 - 23760 Ki kI K3 wi wl

- 90720 kI K'i Wl wi + 4860 Ki K2 W2 wl - 12960 kI K2 W2 wl

- 32400 K2 W2 wl ~ 1728 Ki kI W2 wl - 17280 Ki kI Ka W2 wl
+ 4860 kI -afi + 3240 Ki kI wl + 384 kI wl - 9720 kI K3 wt

- 19440 Ki K'2 K3 wi + 16200 K2 K'I wf W4 - 71280 Ki K2 K'I wf W4

- 43200 K2 Ki wl W4 + 75600 Ki K2 K3 wi W2 W4 - 99360 kI K2 K3 wi W2 W4

- 129600 Ki K2 kI Wl W2 W4 + 1620 K^ wl Wi - 3888 kI wl Wi

- 6480 Kt wl Wi + 17280 kI K2 wl Wi - 69120 Kf K2 K3 wl W4
+ 16200 Ki K3 Wl W3 W4 - 64800 Ki K2 K3 wi W3 W4

- 12960 K'i K2 Ki Wl W3 W4 - 86400 kI kI wi W3 W4 + 27000 Ki K2 W2 wg W4

- 48816 Kl^ K2 W2 W3 W4 - 11520 Ki kI W2 ws W4 - 22032 Ki K2 K3 W2 W3 W4

- 64800 Kl K2 K3 W2 W3 W4 + 2025 K2 wl W4 - 3240 Ki K2 wl W4

- 21060 Kl K2 wl W4 + 2880 Ki w'l W4 - 7488 Ki K'i wl W4

- 6912 Kl K3 wl W4 - 25920 kI kI wl W4 + 24300 K2 K3 wi wl

- 48600 Kl K2 K3 Wl wl - 14400 Kl K3 wi wl - 29160 K2 Kl wi wl

- 6480 Kl K2 Kl Wl wl + 405 Ki W2 wl - 5508 Ki W2 wl

+ 18000 Kl K2 W2 wl - 18720 kI K2 W2 wl - 29376 Ki K2 K3 W2 wl

- 25920 Kl K2 Kl W2 wl + 5805 K2 W3 wl - 8640 Ki K2 ws wl

- 3348 K'} Kz W3 wl - 34992 Ki K2 K3 W3 wl - 17856 Ki K3 W3 wl

+ 405 Kl w'i - 1620 Kf wl + 324 Kf wl + 3600 K2 wl - 7200 Ki K2 w'l

- 1600 Kl wl - 3456 Ki K2 K3 wl - 10368 K2 kI w|)

Change of coordinates. Computing the square of the determinant |r^| amounts
to expressing the degree-20 invariant ^'io(w)^ in terms of the basic forms:

Tk =- ~

$3(t;)^ ^ijvf $5(«)' 'j^ioQ;)'

" $2(V)6 $2(W)4 $2(l')2 4>3(w)2 $2(«)"'
Q1 oc: 7v^2

OL^D i\.l I\2 ^3 ( a^r I n/icnr^ Kionnr^^ , lornnnT^^^ i too nn

13824

-675 + 9450 Kl - 51300 K{ + 135000 Kl - 172800 Kl

+ 86400 K\ + 23700 Ki - 147600 Ki Kz + 111600 Ki Kz + 436800 Ki K2

- 271800 Ki + 424800 Ki Ki + 7200 kI kI + 25600 Ki - 79200 K2 K3

+ 535680 Kl K2 K3 - 777600 Kl K2 K3 - 576000 Ki if 2 K3 + 1552320 A'| K3

- 1238400 JsTI K3 - 30720 JsTI ^3 + 68256 K2 kI - 475200 Ki K2 kI
+ 864000 Kl K2 Kl - 3628800 kI kI + 864000 Ki kI kI

+ 4032000 kI kI - 1728000 kI kI) .

28

SCOTT CRASS

Each entry of r„ is a degree-12 invariant in v. The matrix product's expression
in K is

Tk =

2 Ki (-15 + 66K1+ 40 K2)
48K1K2 {-I + 5K3)

240 Ki K2 K3

200 Ki K2 K3
2AOK1K2K3
\ 2AOK2KI

2 K2 (-35 + 84 Kz) 5 (-15 + 60 Ki + 12 Kf + 128 K2 K3) \

2 K2 (-15 + 90 iCi + 16 2 K2 (-35 + 46 iCi + 84 ifs)

240 Jfi ATa (- 1 + 5 iTg) 2 i^i (- 15 + 66 T^i + 40 K2)

24OK1K2K3 24OK2KI J

The inverse of Tk results from application of Cramer's rule:

T^Cof

^ " \Tk\

where T'^^ is the matrix of cofactors.
Note that tK = \TK\.

Root-selector. The w-coefficients of r„ (w) are degree-13 t;-invariants. Expressed
in

Tk{w)

r„(w)

^2{vY ^iiv)

-125^

36

720 K2 kI wl - 288 Ki K3 wi 102 + 1440 K3 wi W2

- 288 Ki wl + 720 Kf K3 wl - 288 K2 K3 wi W3 + 1440 K2 kI wi W3

- 180 Ki W2 W3 + 792 Ki W2 W3 + 192 Ki K2 W2 W3 - 210 K2 wl

+ 132 Ki K2 wl + 504 K2 K3 wl - 180 K3 wi W4, + 792 Ki K3 wi W4
+ 480 K2 K3 Wl wa - 420 W2 wa + 552 kI W2 wa + 720 Ki K3 W2 wa

- 90 W3 Wa + 360 Ki W3 wa + 72 W3 wa + 480 K2 K3 W3 wa

- 135w4 +270 7^1 W4 +80 /s:2W4 + 162 ^3 W4 + 36 /fi /fs W4).

The 6-maps. Prom the expression for 4>q{u) in basic invariants and equivariants,
a -fT-parametrized 6-map (f)K{w) emerges:

<P6 (w) = (v) r„ (2 (9 $2^ (w) $3^ (w) - 10 {w))<Pi^{w)
-2(*LH-5*4^H) <P2^iw)

+ 20 *3^ (W) <p3K (W) + 15 ^2k (w) (t>AK (W))

= ^Wv) r„ (2 (9 3-2^ (w) *3;. (w) - 10 (w)) V;$2^ {w)
- 2 ($i^ (W) - 5 $4k (w)) V;3'3;. (w)
+ 20 *3,^ (w) VI^^ak (w) + 15 $2k (w) V;;,*5,f (w))

= $2^(v)r„<?!.KH.

SOLVING THE QUINTIC BY ITERATION

29

Appendix C. Basin portraits

The plots that follow are productions of the program Dynamics 2 running on
a Dell Dimension XPS with a Pentium II processor. Its BA process produced
Figure p| while the BA S routine generated the remaining plots. (See the manual
[Nusse and Yorke 1998 |.) Each procedure divides the screen into a grid of cells and

then colors each cell according to which attracting point its trajectory approaches.
If it finds no such attractor after 60 iterates, the cell is black. The BA algorithm
finds the attractor whereas BAS requires the user to specify a candidate attracting
set of points. Each portrait exhibits the highest resolution available — a 720 x 720

grid. Color versions of the images appear at |Crass 1999b|.

In Figure ^, we have the dodecahedral 11-map. Each of the ten pairs of antipodal
dodecahedral vertices — black dots — is a period-2 superattractor. Their basins fill
up CP^ in measure. (Bear in mind that points in the space of this plot correspond
to hues on the quadric surface Q.)

Figure || indicates the behavior of hn restricted to an 54-symmetric conic Qj.
The four pairs of antipodal vertices of the cube are period-2 superattracting 20-
points whose basins have full measure on the conic.

Figures |^ and ^ show the behavior of the octahedral map hn on a 15-line and
a 30-line respectively. In the former case, the critical points at and 00 are a
pair of 30-points on Q that hu exchanges. A pair of fixed 10-points accounts for
the remaining two basins. At each of these attracting points, the map repels in at
least one direction away from the line. Although the line has Z2 symmetry under
G120, the plot displays that of Z2 x Z2. This is a manifestation of an additional
antiholomorphic symmetry

that extends Gi2n by degree two.

On the 30-line, the critical points at and 00 are a pair of octahedral 60-points
on Q that hn exchanges. The remaining two basins belong to a pair of 20-points
on TZ. At each of these attracting points, the map repels in at least one direction
away from the line. Again, Z2 x Z2 symmetry appears.

In Figures |^ and || we see the restriction of hn to an RP^ with ^4 symmetry and
an RP^ with S3 symmetry. Each case involves a chaotic attractor. In the former,
the attractor consists of the four RP^ intersections of TZ, C\., and the 10-lines
£}q.^. . The six intersections occur at 10-points p^^^ {k,t 7^ i). (In the picture, two
of these intersections occur on the line at infinity.) The pictured "lines" are the
images of small circles centered along the edges of the inner square. This graphical
technique specifically relies on the chaotic behavior of hn along each RP^.

In the 1S3 plane, the attracting line is the RP"'^ intersection of TZ, C\q^. and the 10-
line C\q^_ at infinity — the light gray basin. The three "attracting" 30-points — they
are blowing up — are the vertices

of an equilateral triangle about (0, 0).

The remaining images illustrate the dynamics of the quintic-solving 6-map /g.
In Figures |^ through |l^, we see the restriction to the RP^ determined by £f g . . n TZ.
Since this plane is iSa -symmetric, the affine coordinates here are chosen with the

30

SCOTT CRASS

three 5-points at

Three of the superattracting pipes form a triangle on these points. Indeed, the
image in Figure |9| of the circle

1'

2 I 2

is nearly this triangle. The attractor at (0, 0) is the 1-point orbit in the 10-plane —
overall, the 10-point p\'j^ . In the direction away from the plane, /g repels at this site
along the superattracting pipe A^io^f^ {k,£,m ^ The three "spokes" at basin
boundaries are pieces of 15-lines each of which passes through a secondary

basin that contains a preimage of the central 10-point. The boxed region is the

approximate content of Figure 11

Figure 10 show /in's critical set {minus the three "doubly-critical" 10-lines)
superimposed on the blurry basin portrait. The critical contour is a Mathematica
plot. Of course, the higher order intersections occur at the 5-points. All but six
critical points appear to belong to the basin of either a 5-point or the central 10-
point Pij^- The six exceptions lie on the 15-lines at basin boundaries. If this is
so, then there is no other attracting site — provided that a basin contains critical
points.

In Figures 13 and |lj we see the map restricted to a 15-line. The coordinates of

this image place the single 5-point at and the two fixed superattracting 10-points
at ±1. At the latter points, the map repels in all directions off the line. Figure [l^
approximately shows the boxed region.

In Figure the space is the RP^ intersection of an 54-invariant RP'^ and a
10-plane ^Wij- The RP^ intersection of the RP^ and the 10-line C\q.^ is {x = 0}.
By plotting the trajectory of one of its generic points, this line reveals itself as
a chaotic attractor; the plot shows roughly 20, 000 iterates. The map attracts at
(1,0), (—1,0) — the 5-point p| (k ^ i,j) and 10-point pj? respectively.

SOLVING THE QUINTIC BY ITERATION

31

Figure 4. Four basins of attraction for the octahedral 5-map

Figure 8. Chaotic attractor for hn on an RP^ with S3 symmetry

Figure 10. Critical set of fe restricted to an RP^

Figure 12. Detail of the left cusp in Figure |T|

Figure 14. Magnified view of the boxed region in Figure

13

38

SCOTT CRASS

References

[Crass forthcoming] S. Crass. Solving the sextic by iteration: A study in complex geometry and

dynamics. To appear in Experiment. Math.
[Crass 1999a] S. Crass, 1999. Mathematica notebook and data files that implement the quintic-

solving algorithm based on the dynamics of fa. See www.bufTalostate.edu/ "crasssw.
[Crass 1999b] S. Crass, 1999. "Solving the quintic by iteration in three dimensions." Preprint at

xxx.lanl.gov/abs/math.DS/9903054.
[Crass 1999c] S. Crass, 1999. "Solving the octic by iteration in six dimensions." Preprint at

XXX. lanl . gov /abs/math.DS/.
[Doyle and McMuUcn 1989] P. Doyle and C. McMuUen. Solving the quintic by iteration. Acta

Math. 163 (1989), 151-80.
[Hodge and Pedoe 1968] W. Hodge and D. Pedoe. Methods of Algebraic Geometry. Cambridge

University Press, 1968.

[Klein 1956] F. Klein. Lectures on the Icosahedron and the Solutions of Equations of the Fifth
Degree. Translated by G. Morrice. Dover, 1956.

[Nussc and Yorke 1998] H. Nusse and J. Yorkc. Dynamics: Numerical Explorations, 2e Springer-
Verlag, 1998. Computer program Dynamics 2 by B. Hunt and E. Kostelich.

[Orlik and Terao 1992] P. Orlik and H. Terao. Arrangements of Hyperplanes. Springer- Verlag,
1992.

[Shcphard and Todd 1954] G.C. Shcphard and T.A. Todd. Finite unitary reflection groups.
Canad. J. Math. 6 (1954), 274-304.

Mathematics Institute, University of Warwick, Coventry CV4 7AL UK (1999-2000)
E-mail address: scrasswSmaths. Warwick. ac.uk

Department of Mathematics, SUNY College at Buffalo, Buffalo, NY 14222
E-mail address: crassswSbuffalostate.edu