Contemporary Mathematics

First integrals and gradient flow for a generalized

Darboux-Halphen system

S. Chakravarty and R.G. Halburd

Abstract. First integrals are explicitly constructed for a third-order system

of ODEs that arises as a reduction of the self-dual Yang-Mills equations and in
the theory of hypercomplex manifolds. These first integrals are branched func-
tions of the phase space variables, even in cases for which the general solution
is single-valued. This branching is characterized in terms of the monodromy of
the hypergeometric equations. The first integrals are then used to formulate
a Nambu-Poisson structure of the system. A representation of the generalized
Darboux-Halphen system as a gradient flow is also given.

1. Introduction
The system
ω̇1 = ω2 ω3 − ω1 (ω2 + ω3 ) + τ 2 ,
(1.1) ω̇2 = ω3 ω1 − ω2 (ω3 + ω1 ) + τ 2 ,
ω̇3 = ω1 ω2 − ω3 (ω1 + ω2 ) + τ 2 ,
where
τ 2 = α2 (ω1 − ω2 )(ω3 − ω1 ) + β 2 (ω2 − ω3 )(ω1 − ω2 ) + γ 2 (ω3 − ω1 )(ω2 − ω3 ),
was first studied by Halphen [14] as a natural generalization of the classical Darboux-
Halphen (DH) system, which corresponds to setting τ = 0 in equation (1.1). The
classical Darboux-Halphen system first arose in Darboux’s study of triply orthog-
onal surfaces [11] and was later solved by Halphen [15]. The classical DH system
has also appeared in studies of self-dual Bianchi-IX metrics with Euclidean signa-
ture [4, 13] and in reductions of the associativity equations on a three-dimensional
Frobenius manifold [12]. Furthermore, if (ω1 , ω2 , ω3 ) is a solution of the classical
Darboux-Halphen system, then
(1.2) y := −2(ω1 + ω2 + ω3 )
1991 Mathematics Subject Classification. Primary 34A34, 34A05; Secondary 37K10.
Key words and phrases. Darboux-Halphen, Schwarzian.
The first author was partially supported by a Grigsby-Sensenbaugh Foundation grant from
the University of Colorado.
The second author was partially supported by grant number NAL/00344/G from the Nuffield
Foundation.

2002
c American Mathematical Society

1
2 S. CHAKRAVARTY AND R.G. HALBURD

satisfies the Chazy equation,

 2
d3 y d2 y dy
(1.3) = 2y − .
dt3 dt2 dt
In [3] it was shown that y defined by equation (1.2) solves the generalized Chazy
equation,
 2  2
d3 y d2 y dy 4 dy
(1.4) − 2y 2 + = 6 −y 2
,
dt3 dt dt 36 − n2 dt
where the ωi ’s solve the generalized Darboux-Halphen system for the special choices
of parameters (α, β, γ) given by (2/n, 2/n, 2/n) and (1/3, 1/3, 2/n), et cyc. Note
that equation (1.3) corresponds to the limit n → ∞ in equation (1.4). Equations
(1.3) and (1.4) were first studied by Chazy in [8, 9, 10].
The system (1.1) arises in the study of the equation
T
(1.5) Ṁ = (adj M ) + M T M − (Tr M )M,
for a 3 × 3 matrix valued function M (t) where adj M is the adjoint of M satisfying
(adj M ) M = (det M )I, M T is the transpose of M and the dot denotes differenti-
ation with respect to t. Equation (1.5) was obtained in [6] as a reduction of the
self-dual Yang-Mills equations with an infinite dimensional gauge group of diffeo-
morphisms Diff(S 3 ) of the three-sphere. Equation (1.5) also describes a class of
self-dual Weyl Bianchi IX space-times with Euclidean signature [5]. More recently,
equation (1.5) was used to describe SU(2) invariant hypercomplex 4-manifolds [16].
In section 2 we will review the reduction of equation (1.5) to the generalized
DH system (1.1). The general solution will be constructed and a special case will
be studied. In section 3, first integrals and “action-angle” variables are given for
equation (1.1). The first integrals involve hypergeometric functions and are non-
meromorphic, even in cases where the general solution is single-valued.

2. The solution of the generalized Darboux-Halphen system

In this section the solution of equation (1.5) is given by a factorization method
which first appeared in [1]. The solution can also be obtained via associated linear
problems. In [2], the solution was obtained via an evolving monodromy problem
that arises as a reduction of the isospectral problem for the self-dual Yang-Mills
equations. In [16], the solution was obtained via an isomonodromy problem which
describes the Riccati solutions of the sixth Painlevé equation. Degenerate cases
were discussed in [3].
We begin by decomposing the matrix M into its symmetric (Ms ) and antisym-
metric (Ma ) parts. Furthermore, the eigenvalues of Ms are assumed to be distinct,
so that it can be diagonalized using a complex orthogonal matrix. Thus we have
(2.1) M = Ms + Ma = P (d + a)P −1 ,
where P ∈ SO(3, C), d = diag(ω1 , ω2 , ω3 ) with ωi = ωj , i = j, and the elements
3
of the skew-symmetric matrix a are given by aij = εijk τk , where εijk is totally
k=1
skew-symmetric in its indices and ε123 = 1. Using the transformation (2.1), the
 A GENERALIZED DARBOUX-HALPHEN SYSTEM 3

diagonal part of equation (1.5) yields the system (1.1), where τ 2 = τ12 + τ22 + τ32 .
The skew-symmetric part gives
(2.2) τ̇1 = −τ1 (ω2 + ω3 ), τ̇2 = −τ2 (ω3 + ω1 ), τ̇3 = −τ3 (ω1 + ω2 ) ,
and the off-diagonal symmetric part gives
(2.3) Ṗ = −P a,
which is a linear equation for P .
Taking differences of equations in system (1.1) gives
d
(2.4) (ω1 − ω2 ) = −2ω3 (ω1 − ω2 ), et cyc.
dt
Using equations (2.2) and (2.4), we can solve the τi ’s in terms of the ωi ’s as
(2.5) τ12 = α2 (ω1 − ω2 )(ω3 − ω1 ),
τ22 = β 2 (ω2 − ω3 )(ω1 − ω2 ), τ32 = γ 2 (ω3 − ω1 )(ω2 − ω3 ) ,
where α, β, and γ are integration constants.
In terms of the cross ratio
ω1 − ω3
(2.6) s := ,
ω2 − ω3
it follows from equation (1.1) that the ωi ’s can be parameterized as
1d ṡ 1d ṡ 1d ṡ
(2.7) ω1 = − ln , ω2 = − ln , ω3 = − ln .
2 dt s(s − 1) 2 dt s − 1 2 dt s
where s satisfies
   2
d s̈ 1 s̈ ṡ2
(2.8) − + V (s) = 0 ,
dt ṡ 2 ṡ 2
with    2
d s̈ 1 s̈
{s, t} := −
dt ṡ 2 ṡ
being the Schwarzian derivative and V is given by
1 − β2 1 − γ2 β 2 + γ 2 − α2 − 1
(2.9) V (s) = + + .
s2 (s − 1)2 s(s − 1)
Equation (2.8) is the Schwarzian equation, which describes the conformal map-
pings of the upper-half s-plane to the interior of a region of the complex sphere
bounded by three regular circular arcs. If α, β, and γ are non-negative real num-
bers such that α + β + γ < 1, then the angles subtended at the vertices s = 0, s = 1,
and s = ∞ of this triangle are απ, βπ, and γπ. Furthermore, if α, β, and γ are
chosen to be either reciprocals of integers or zero, then s is analytic on the interior
of a circle on the complex sphere and cannot be analytically extended across this
circle, which is a natural barrier [19].
The general solution of equation (2.8) is given implicitly by
u1 (s)
(2.10) t(s) = ,
u2 (s)
where u1 (s) and u2 (s) are independent solutions to the Fuchsian equation
d2 u 1
(2.11) + V (s)u = 0.
ds2 4
4 S. CHAKRAVARTY AND R.G. HALBURD

Equation (2.11) is equivalent to the hypergeometric equation,

d2 χ dχ
(2.12) s(1 − s) + [c − (a + b + 1)s] − abχ = 0,
ds2 ds
where a = (1 + α − β − γ)/2, b = (1 − α − β − γ)/2, c = 1 − β, and
(2.13) u(s) = sc/2 (1 − s)(a+b−c+1)/2 χ(s).

A fifth order reduction

We will now consider the case in which M has the special form
 
M11 M12 0
(2.14) M = M21 M22 0 .
0 0 M33
This special form of M was considered in [5, 2], where equation (1.5) was analyzed
using an associated evolving monodromy problem. Here we will show that quantities
that arise naturally from this monodromy analysis can be obtained in a straight-
forward manner from the factorization method described above.
Consider the factorization of M given by equation (2.1) where M is given by
equation (2.14). Due to special block structure of M, its symmetric part can be
diagonalized by an orthogonal matrix of the form
 
cos ψ sin ψ 0
(2.15) P = − sin ψ cos ψ 0 ,
0 0 1
where ψ is a (generally complex) function of t to be determined. That is, Ms =
P dP −1 , where d = diag(ω1 , ω2 , ω3 ). Furthermore, the skew-symmetric part of M
is unchanged by the adjoint action of P. So
 
0 τ3 (t) 0
1
(2.16) a = P −1 Ma P = Ma = (M − M T ) = −τ3 (t) 0 0 ,
2
0 0 0
where τ3 (t) = 12 (M12 − M21 ). Since τ1 = τ2 = 0 in this case, the ωi ’s are given by
equation (2.7) where s solves equation (2.8) with α = β = 0. From equation (2.5)
and equation (2.7), we have
iγ ṡ
τ3 (t) = √ ,
2 s(s − 1)
where γ is a constant. With the τ3 (t) given above, equation (2.3) can be readily
integrated to give √ 
iγ s−1
ψ= log √ + ψ0 ,
2 s+1
where ψ0 is a constant. Finally, the matrix M in (2.14) is reconstructed from the
various components P , d and a according to equation (2.1). Note that in order to
obtain any solution of equation (1.5) where M is given by (2.14) we must fix the
two constants γ and ψ0 and choose a solution to equation (2.8) with α = β = 0
and the fixed value of γ.
In [5] the general solution of equation (1.5) with M of the form (2.14) was found
via a different method which involved the analysis of a certain evolving monodromy
problem. This led to the following combination of the matrix elements of M
α± = (M11 − M22 ) ∓ i(M12 + M21 ), R2 = α+ α− ,
 A GENERALIZED DARBOUX-HALPHEN SYSTEM 5

β± = ω ± i(M21 − M12 ), ω = M11 + M22 − 2M33 ,

together with the conserved quantity
(M12 − M21 )2
a2 = .
R2 − ω 2
In the factorization method outlined above, these variables arise naturally from the
component matrices P , d and a as follows
R = ω1 − ω 2 , α± = Re∓2iψ , ω = (ω1 − ω3 ) + (ω2 − ω3 ), M12 − M21 = 2τ3 ,
and a = γ.

3. First Integrals
In this section, following [7], we will use the method of solution given in section
2 to construct first integrals for equation (1.1). We begin by constructing explicit
first integrals for the Schwarzian equation (2.8), as the formulas are much simpler
in this case. Let u1 and u2 be two linearly independent solutions of equation (2.11)
satisfying the Wronskian condition W (u1 , u2 ) = u1 u2 −u2 u1 = 1. Then any solution
to equation (2.8) is given implicitly by
J2 u1 (s) − J1 u2 (s)
(3.1) t(s) = ,
I2 u1 (s) − I1 u2 (s)
where Ik and Jk , k = 1, 2, are constants satisfying
(3.2) I1 J2 − I2 J1 = 1.
Hence any three of the constants I1 , I2 , J1 , J2 can be taken as independent first
integrals.
Differentiating equation (3.1) with respect to t gives
(3.3) I2 u1 − I1 u2 = ṡ1/2 .
Differentiation of equation (3.3) gives
1 −3/2
(3.4) I2 u2 − I1 u2 = ṡ s̈.
2
Solving the system (3.3–3.4) for I1 and I2 gives
dφk
(3.5) Ik = , φk = ṡ−1/2 uk (s), k = 1, 2.
dt
The constants J1 and J2 are given by the solution of equations (3.1), (3.5) and the
normalization condition (3.2). This yields Jk = tIk − φk , k = 1, 2.
In terms of the gDH variables, we have
 1/2  
√ (ω2 − ω3 ) ω1 − ω3
φk = 2 uk ,
(ω1 − ω2 )(ω1 − ω3 ) ω2 − ω3
 1/2  
√ (ω1 − ω3 )(ω1 − ω2 )  ω1 − ω3
(3.6) Ik = 2 uk
(ω2 − ω3 ) ω2 − ω3
 1/2  
(ω2 − ω3 ) ω1 − ω3
−(ω1 − ω2 − ω3 ) uk .
2(ω1 − ω3 )(ω1 − ω2 ) ω2 − ω3
6 S. CHAKRAVARTY AND R.G. HALBURD

Using the new variables φk and Ik instead of the gDH variables ωi ’s, equation (1.1)
can be formulated as a Hamiltonian system
∂H ∂H I12 + I22
(3.7) φ̇k = = Ik , I˙k = − = 0, H= , k = 1, 2,
∂Ik ∂φk 2
together with the constraint
(3.8) φ1 I2 − φ2 I1 = W (u1 , u2 ) = 1.
Although I1 and I2 are constant functions of t, they are multivalued functions
of {ω1 , ω2 , ω3 } and of the Schwarzian variables {s, ṡ, s̈}. In terms of the solutions
χ1 , χ2 of the hypergeometric equation (2.12), the first integrals are given by
χ1 (s) χ2 (s)
(3.9) I1 I2 = σ [λ 1] ,
χ1 (s) χ2 (s)
where
a + b + 1 − cs s̈
σ (s, ṡ) = sc/2 (1 − s)(a+b−c+1)/2 ṡ1/2 , and λ(s, ṡ, s̈) = − 2.
2s(1 − s) 2ṡ
Next we will discuss the dependence of I1 and I2 on s, ṡ, and s̈. Clearly Ik , k = 1, 2,
is single-valued as a function of s̈ and has square-root branch points as a function
of ṡ about ṡ = 0 and ṡ = ∞. In fact, the conserved quantities I12 and I22 are single-
valued as functions of ṡ. Holding ṡ and s̈ fixed, Iµ can only admit branch points at
s = 0, s = 1, and s = ∞. Let γ0 and γ1 be two closed curves with a common base
point in the finite complex s-plane enclosing the points s = 0 and s = 1 respectively,
and traversed once in the positive direction. Analytic continuation of σ along γ0
and γ1 gives
γ0 : σ → eiπc σ, γ1 : σ → eiπ(a+b−c) σ.
Analytic continuation along γ0 and γ1 transforms the fundamental matrix of solu-
tions of equation (2.12) according to
   
χ1 (s) χ2 (s) χ1 (s) χ2 (s)
γµ : → Mµ , µ = 0, 1.
χ1 (s) χ2 (s) χ1 (s) χ2 (s)
For generic values of a, b, c, and for the choice of basis solutions: χ1 = F (a, b, c; s),
χ2 = F (a, b, a + b − c + 1; 1 − s) of the hypergeometric equation, the monodromy
matrices Mµ are given by [20]
   −2πi(a+b−c) 
1 e−2πib − e−2πic e 0
M0 = and M = .
0 e−2πic 1
1 − e−2πi(a−c) 1
So under analytic continuation, the first integrals I1 , I2 transform as
γ0 : I1 I2 → I1 I2 M0 eiπc , γ1 : I1 I2 → I1 I2 M1 eiπ(a+b−c) .
Analyic continuation around s = ∞ is equivalent to a loop around s = 0 and s = 1.
Hence the branching of the first integrals I1 and I2 is characterized in terms of the
monodromy group for the hypergeometric equation.
The first integrals in equation (3.9) for the classical DH system (α = β =
γ = 0) are expressed in terms of the special hypergeometric equation (2.12) with
a = b = 1/2, c = 1. In this case, the monodromy matrices are given by
   
1 −2 1 0
M0 = and M1 =
0 1 2 1
 A GENERALIZED DARBOUX-HALPHEN SYSTEM 7

relative to the choice of basis χ1 = F (1/2, 1/2, 1; s) and χ2 = iF (1/2, 1/2, 1; 1 − s).
Note that in this case, I1 and I2 are still branched, even though the solution itself
is single-valued. The non-existence of meromorphic first integrals for the classical
Darboux-Halphen system was proved in [17]. We show by explicit construction
that the first integrals do indeed exist but they are non-algebraic and multi-valued.

4. Nambu-Poisson Structure and Gradient flow

The gDH system (1.1) can also be viewed as a complex dynamical system given
by (ω̇1 , ω̇2 , ω̇3 ) = X where the vector field

X = ω 2 ω 3 − ω 1 ω 2 − ω 3 ω 1 + τ 2 , ω3 ω 1 − ω 2 ω 3 − ω 1 ω 2 + τ 2 , ω 1 ω 2 − ω 3 ω 1 − ω 2 ω 3 + τ 2 .
The generalized DH flow given by the integral curves of X lie on the intersection
of the level sets of the first integrals (I1 = constant and I2 = constant) in a three-
dimensional complex manifold M 3 := C \ {ωi = ωj , i = j}. Since the Ik ’s are
conserved under the gDH flow,
dIk
= X · ∇Ik = 0, k = 1, 2,
dt
it follows that the vector field X is proportional to ∇I1 × ∇I2 . Explicit calculation
shows that,
1
X= ∇I1 × ∇I2
4∆
where
∆ = (ω2 − ω3 )(ω3 − ω1 )(ω1 − ω2 ).
The gDH equations can be expressed as

(4∆)−1 ∇I2 · (∇ωj × ∇I1 ) =: {ωj , I1 }1 ,
ω̇j = X · ∇ωj =
−(4∆)−1 ∇I1 · (∇ωj × ∇I2 ) =: {ωj , I2 }2 .
It can be verified that the brackets
B1 (g, h) = {g, h}1 = (4∆)−1 ∇I2 · (∇g × ∇h),
B2 (g, h) = {g, h}2 = −(4∆)−1 ∇I1 · (∇g × ∇h),
are Poisson (i.e., they are bi-linear, anti-symmetric and satisfy the Jacobi identity).
So X is a Hamiltonian vector field with respect to the two Poisson strucures B1
and B2 with Hamiltonians I1 and I2 respectively.
The Poisson structures B1 and B2 are degenerate (rank 2) and admit Casimir
functions I2 and I1 respectively. This is easily verified by using the vector triple
product identity
{I2 , g}1 = (4∆)−1 ∇I2 · (∇I2 × ∇g) = (4∆)−1 ∇g · (∇I2 × ∇I2 ) = 0,
{I1 , g}2 = (4∆)−1 ∇I2 · (∇I1 × ∇g) = (4∆)−1 ∇g · (∇I1 × ∇I1 ) = 0,
for any smooth function g on M . Furthermore, since {I1 , I2 }1 = {I1 , I2 }2 = 0, the
integrals of the gDH system are in involution. The Poisson structures B1 and B2
are compatible in the sense that
B := λ1 B1 + λ2 B2 , λi = λi (I1 , I2 )
is also a Poisson structure for gDH with a Hamiltonian H(I1 , I2 ) satisfying
∂H ∂H
λ1 + λ2 = 0.
∂I1 ∂I2
8 S. CHAKRAVARTY AND R.G. HALBURD

Thus the gDH system is bi-Hamiltonian.

The symmetric representation of the gDH system using both Hamiltonians I1 ,
I2 is
∂(ωi , I1 , I2 )
ω̇j = (4∆)−1 ∇ωj · (∇I1 × ∇I2 ) = (4∆)−1 =: {ωj , I1 , I2 }.
∂(ω1 , ω2 , ω3 )
This is an example of a Nambu-Poisson bracket similar to rigid body dynamics in
three dimensions [18, 21].
The Darboux-Halphen system (1.1) is also a gradient flow. In terms of local
coordinates ωi , it can be written as

3
∂Φ
ω̇i = g ij ,
j=1
∂ωj
ij
where g is a constant contravariant metric and Φ is a potential function. The
metric is given by
 
m(α, β, γ) κ + 4γ 2 κ + 4β 2
(g ij ) =  κ + 4γ 2 m(β, γ, α) κ + 4α2  ,
2
κ + 4β κ + 4α2 m(γ, α, β)
where
κ = (α + β + γ)(α + β − γ)(α − β + γ)(α − β − γ) − 1,
and
m(α, β, γ) = (1 − α2 + (β + γ)2 )(1 − α2 + (β − γ)2 ).
The potential Φ is a homogeneous polynomial of degree 3 in the ωi ’s and is invariant
under the simultaneous cyclic permutation of {ω1 , ω2 , ω3 } and {α, β, γ}. In terms
of the function
F (ω1 , ω2 , ω3 ; α, β, γ) = (1 − α2 )(3α2 − 2β 2 − 2γ 2 + 1) + (β 2 − γ 2 )2 ×
ω1 α2 ω12 + 3β 2 ω22 + 3γ 2 ω32 + (1 − α2 − 3β 2 − 3γ 2 )ω2 ω3 ,
the potential is expressed as
4
Φ=− [F (ω1 , ω2 , ω3 ; α, β, γ) + F (ω2 , ω3 , ω1 ; β, γ, α) + F (ω3 , ω1 , ω2 ; γ, α, β)] .
3 det(g ij )
With respect to the metric g ij , the constant Φ surfaces are orthogonal to the curves
obtained by the intersection of the constant I1 and I2 surfaces.
In the classical Darboux-Halphen case (α = β = γ = 0), we have
 
1 −1 −1
g ij = −1 1 −1 and Φ = ω1 ω2 ω3 .
−1 −1 1
For the fifth order reduction (α = β = 0) discussed in section 2,
 
(1 + γ 2 )2 γ 4 + 4γ 2 − 1 γ4 − 1
g = γ + 4γ − 1
ij 4 2
(1 + γ 2 )2 γ4 − 1 
γ −1
4
γ4 − 1 (1 − γ 2 )2
and the corresponding potential function is given by
γ 2 ω32 [3(1 − γ 2 )(ω1 + ω2 ) + (1 + 3γ 2 )ω3 ] + 3(1 − γ 2 )2 ω1 ω2 ω3
Φ= .
3(1 − γ 2 )3
 A GENERALIZED DARBOUX-HALPHEN SYSTEM 9

References
1. M.J. Ablowitz, S. Chakravarty, and R. Halburd, Darboux-Halphen systems and the singularity
structure of its solutions, Proc. Fourth Int. Conf. on Mathematical and Numerical Aspects of
Wave Propagation (Philadelphia) (J.A. DeSanto, ed.), SIAM, 1998, pp. 408–412.
2. , On Painlevé and Darboux-Halphen type equations, The Painlevé Property, One Cen-
tury Later (Berlin) (R. Conte, ed.), CRM Series in Mathematical Physics, Springer, 1998.
3. , The generalized Chazy equation from the self-duality equations, Stud. Appl. Math.
103 (1999), 75–88.
4. M.F. Atiyah and N.J. Hitchin, The geometry and dynamics of magnetic monopoles, Princeton
University Press, Princeton, New Jersey, 1988.
5. S. Chakravarty and M.J. Ablowitz, Integrability, monodromy, evolving deformations, and
self-dual Bianchi IX systems, Phys. Rev. Lett. 76 (1996), 857–860.
6. S. Chakravarty, M.J. Ablowitz, and L.A. Takhtajan, Self-dual Yang-Mills equation and new
special functions in integrable systems, Nonlinear Evolution Equations and Dynamical Sys-
tems (Singapore) (M. Boiti, L. Martina, and F. Pempinelli, eds.), World Scientific, 1992.
7. S. Chakravarty and R. Halburd, First integrals of a generalized Darboux-Halphen system,
Loughborough University Preprint (2002).
8. J. Chazy, Sur les équations différentielles dont l’intégrale générale est uniforme et admet des
singularities essentielles mobiles, C.R. Acad. Sc. Paris 149 (1909), 563–565.
9. , Sur les équations différentielles dont l’intégrale générale possède une coupure essen-
tielle mobile, C.R. Acad. Sc. Paris 150 (1910), 456–458.
10. , Sur les équations différentielles du troisième et d’ordre supérieur dont l’intégrale
générale à ses points critiques fixés, Acta Math. 34 (1911), 317–385.
11. G. Darboux, Sur la théorie des coordonnées curvilignes et les systémes orthogonaux, Ann.
Ec. Normale Supér. 7 (1878), 101–150.
12. B. Dubrovin, Geometry of 2D topological field theories, Integrable systems and quantum
groups (Montecatini Terme, 1993), Springer, Berlin, 1996, pp. 120–348.
13. G.W. Gibbons and C.N. Pope, The positive action conjecture and asymptotically Euclidean
metrics in quantum gravity, Commun. Math. Phys. 66 (1979), 267–290.
14. G. Halphen, Sur certains systèmes d’équations différentielles, C. R. Acad. Sci. Paris 92 (1881),
1404–1406.
15. , Sur un système d’équations différentielles, C. R. Acad. Sci. Paris 92 (1881), 1101–
1103.
16. N. Hitchin, Hypercomplex manifolds and the space of framings, The geometric universe (Ox-
ford, 1996), Oxford Univ. Press, Oxford, 1998, pp. 9–30.
17. A.J. Maciejewski and J. Strelcyn, On the algebraic non-integrability of the Halphen system,
Phys. Lett. A 201 (1995), 161–166.
18. Y. Nambu, Generalized Hamiltonian dynamics, Phys. Rev. D (3) 7 (1973), 2405–2412.
19. Z. Nehari, Conformal mapping, McGraw-Hill, New York (1952).
20. J. Plemelj, Problems in the sense of Riemann and Klein, Interscience Publishers John Wiley
& Sons Inc. New York-London-Sydney, 1964.
21. L. Takhtajan, On foundation of the generalized Nambu mechanics, Comm. Math. Phys. 160
(1994), 295–315.

Department of Mathematics, University of Colorado, Colorado Springs, CO 80933-

7150, USA
E-mail address: chuck@math.uccs.edu

Department of Mathematical Sciences, Loughborough University, Loughborough,

Leicestershire, LE11 3TU, UK
E-mail address: R.G.Halburd@lboro.ac.uk