EPJ manuscript No.

(will be inserted by the editor)

Real Hamiltonian forms of Hamiltonian systems

V. S. Gerdjikov1,3 , A. Kyuldjiev1 , G. Marmo2 , and G. Vilasi3
arXiv:nlin/0310005v1 [nlin.SI] 8 Oct 2003

1
Institute for Nuclear Research and Nuclear Energy, 72 Tzarigradsko chaussée, 1784 Sofia, Bulgaria,
e-mail: gerjikov@inrne.bas.bg, kyuljiev@inrne.bas.bg
2
Dipartimento di Scienze Fisiche, Università Federico II di Napoli and Istituto Nazionale di fisica Nucleare, Sezione di Napoli,
Complesso Universitario di Monte Sant’Angelo, Via Cintia, 80126 Napoli, Italy, e-mail: marmo@na.infn.it
3
Dipartimento di Fisica ”E.R.Caianiello”, Universita di Salerno Istituto Nazionale di fisica Nucleare, Gruppo Collegato di
Salerno, Salerno, Italy, e-mail: vilasi@sa.infn.it

Received: date / Revised version: date

Abstract. We introduce the notion of a real form of a Hamiltonian dynamical system in analogy with the
notion of real forms for simple Lie algebras. This is done by restricting the complexified initial dynamical
system to the fixed point set of a given involution. The resulting subspace is isomorphic (but not sym-
plectomorphic) to the initial phase space. Thus to each real Hamiltonian system we are able to associate
another nonequivalent (real) ones. A crucial role in this construction is played by the assumed analyticity
and the invariance of the Hamiltonian under the involution. We show that if the initial system is Liouville
integrable, then its complexification and its real forms will be integrable again and this provides a method
of finding new integrable systems starting from known ones. We demonstrate our construction by finding
real forms of dynamics for the Toda chain and a family of Calogero–Moser models. For these models we
also show that the involution of the complexified phase space induces a Cartan-like involution of their Lax
representations.

PACS. 02.30.Ik Integrable systems – 45.20.Jj Lagrangian and Hamiltonian mechanics

1 Introduction
Recently the so-called complex Toda chain (CTC) was
shown to describe N -soliton interactions in the adiabatic
approximation [1,2,3]. The complete integrability of the
CTC is a direct consequence of the integrability of the real
(standard) Toda chain (TC); it was also shown that CTC
allows several dynamical regimes that are qualitatively dif-
ferent from the one of real Toda chain [2]. These results
as well as the hope to understand the algebraic structures
lying behind the integrability of CTC (such as, e.g. Lax
representation) were the stimulation for the present work.
We start from a standard (real) Hamiltonian system
H ≡ {ω, H, M} with n degrees of freedom and Hamil-
tonian H depending analytically on the dynamical vari-
ables. It is known that such systems can be complexified
and then written as a Hamiltonian system with 2n (real)
degrees of freedom. Our main aim is to show that to each
compatible involutive automorphism Ce of the complexified
phase space we can relate a real Hamiltonian form of the
initial system with n degrees of freedom. Just like to each
complex Lie algebra one associates several inequivalent
real forms, so to each H we associate several inequivalent
real forms HR ≡ {ωR , HR , MR }. Like the initial system
H, the real form is defined on a manifold MR with n real
e
degrees of freedom. Provided C(H) = H the dynamics on
the real form will be well defined and will coincide with
the dynamics on MC restricted to MR . We show that if
the initial system H is integrable then its real Hamiltonian
forms will also be integrable. We pay special attention to
the connection with integrable systems and the possibility
they offer to define a class of new integrable systems start-
ing from an initial one. Recently a procedure to obtain
new integrable systems by composing known integrable
ones has been elaborated in the framework of coproducts
[4]. Here we are not concerned with this approach.
Examples of indefinite-metric Toda chain (IMTC) have
already been studied by Kodama and Ye [5]. In particu-
lar they note that while the solutions of the TC model
are regular for all t, the solutions of the IMTC model
develop singularities for finite values of t. Particular ex-
amples of non-standard (or “twisted”) real forms of 1 + 1-
dimensional Toda field theories have already been studied
by Evans and Madsen [6] in connection with the problem
of positive kinetic energy terms in the Lagrangian descrip-
tion and with emphasis on conformal WZNW models.
The approach we follow here is different and more gen-
eral than the ones in [5,6]. Its main ideas were reported in
[7]; here we elaborate the proofs the details and provide
new classes of examples.
2 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems

2 Complexified Hamiltonian Dynamics Obviously, dim MC = 4n and therefore HC may be

considered as a real dynamical system with 2n degrees of
We start with a real Hamiltonian system with n degrees freedom. To elaborate on this, we start from the complex-
of freedom H ≡ M(n) , H, ω where M(n) is a 2n dimen- ified equations of motion:
sional vector space and
dpC
k ∂H C
=− C, (6)
n
X dt ∂qk
ω= dpk ∧ dqk (1)
dqkC ∂H C
k=1 =
dt ∂pC
k
Let’s consider its complexification:
The right hand side of (6) contain the partial derivatives
 of both H0 and H1 . Since we assumed analiticity, H0 and
H C ≡ MC , H C , ω C (2)
H1 will satisfy the Cauchy-Riemann equations:
(2n) ∂H0 ∂H1 ∂H0 ∂H1
where MC can be viewed as a linear space M(n) over = , =− (7)
the field of complex numbers: ∂qk,0 ∂qk,1 ∂qk,1 ∂qk,0
(2n) which means that the derivatives in the right hand sides
MC = M(n) ⊕ iM(n) . of (6) are equal to:

In other words the dynamical variables pk , qk in MC now

(n) ∂H C ∂H0 ∂H0 ∂H1 ∂H1
= −i = +i . (8)
may take complex values. We assume that observables F , ∂qkC ∂qk,0 ∂qk,1 ∂qk,1 ∂qk,0
G and the Hamiltonian H are real analytic functions on
(2n) Analogous formulae hold for the derivatives with respect
M and can naturally be extended to MC . to pk,0 and pk,1 . Thus all terms in the right hand sides of
The complexification of the dynamical variables F , G (6) can be expressed through the partial derivatives of H0
and H means that they become analytic functions of the only:
complex arguments:
dpk,0 ∂H0 dqk,0 ∂H0
=− , = ,
pk C = pk,0 + ipk,1 , qk C = qk,0 + iqk,1 , k = 1, . . . , n± (3) dt ∂qk,0 dt ∂pk,0
dpk,1 ∂H0 dqk,1 ∂H0
and we can write: = , =− (9)
dt ∂qk,1 dt ∂pk,1
H C = H(pC C
k , qk ) = H0 + iH1 . (4) Obviously (9) are standard Hamiltonian equations of mo-
tion for a dynamical system with 2n degrees of freedom
The same goes true also for the complexified 2–form: corresponding to:
n
X
C
H0 = Re H C (pC C
k , qk ), (10)
ω = dpC C
k ∧ dqk = ω0 + iω1 (5) n
X
k=1 ω0 = Re ω C = (dpk,0 ∧ dqk,0 − dpk,1 ∧ dqk,1 )
k=1
Note that each of the symplectic forms ω0 and ω1
are non-degenerate. However the linear combination ω C = We denote the related real dynamical vector field by Γ0 .
ω0 + iω1 can be written down in the form: The system (6) allows a second Hamiltonian formula-
  tion with:
dpk,0
Xn
 dp  H1 = Im H C (pC C
k , qk ), (11)
ω= (dpk,0 , dpk,1 , dqk,0 , dqk,1 ) B0  k,1  X n
dqk,0
k=1
dqk,1 ω1 = Im ω C = (dpk,0 ∧ dqk,1 + dpk,1 ∧ dqk,0 )
k=1
where the matrix B0 and also real dynamical vector field Γ1 . Due to the ana-
  lyticity of H C Cauchy-Riemann equations yield that these
0 0 1 i two vector fields actually coincide:
1 0 0 i −1 
B0 = 
2 −1 −i 0 0  Γ0 = Γ1 .
−i 1 0 0
So Γ0 is a bi-Hamiltonian vector field and the correspond-
ing recursion operator is:
obviously has the property B02 = 0.
T = ω0−1 ◦ ω1 = (12)
Remark 1 The kernel of ω C is spanned by the antiholo- ∂ ∂ ∂ ∂
morphic vector fields. We could also choose the antiholo- − ⊗ dp0 + ⊗ dp1 − ⊗ dq0 + ⊗ dq1
morphic (anti-analytic) functions in the complexification ∂p1 ∂p0 ∂q1 ∂q0
procedure. This would lead to equivalent results. T 2 = −1.
 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems 3

3 Complexification and Liouville integrability In the complexified case both the integrals Ik ≡ IkC
and the dynamical variables zj become complex-valued.
Let us now assume that our initial system is Liouville Their independence can be formulated by:
integrable and analyze what consequences will have this
on the complexified system. dI1 ∧ . . . ∧ dIn ∧ dI1∗ ∧ . . . ∧ dIn∗ = (18)
2n
X X
Proposition 1 Let the initial system H have n function- = Wi1 ,...,ik ,jk+1 ,...,j2n−k dzi1 ∧ . . . ∧ dzik ∧
ally independent integrals Ik , k = 1, . . . , n which are real k=1 i1 <...<ik
analytic (meromorphic) functions of qk and pk . Let also jk+1 <...<j2n−k

Ik be in involution: ∧dzj∗k+1 ∧ . . . ∧ dzj∗2n−k 6= 0,

Xn  
∂Ik ∂Ij ∂Ik ∂Ij where
{Ik , Ij } ≡ − = 0, (13)  
s=1
∂ps ∂qs ∂qs ∂ps W 11 W 12
W = , (19)
 W 21 W 22
C
Then the complexified system M , H0 , ω0 is also Li-
∂Ij ∂Ij∗
ouville integrable, i.e. it has 2n functionally independent 11
Wjk = , 22
Wjk = 11 ∗
= (Wjk ) ,
integrals Ik,0 , Ik,1 , k = 1, . . . , n in involution. ∂zk ∂zk∗
12 ∂Ij 21
∂Ij∗ 12 ∗
Proof Obviously after the complexification each of the in- Wjk = ∗, Wjk = = (Wjk ) .
tegrals Ik becomes complex-valued IkC = Ik,0 + iIk,1 . Since ∂zk ∂zk
Ik is a real analytic function then IkC satisfies Cauchy- Therefore we have to prove now that rank W = 2n. But
Riemann equations with respect to each of the complexi- due to the analyticity of all Ik the matrix elements of W 12
fied variables. Keeping this in mind let us complexify the and W 21 vanish and W becomes a block-diagonal matrix
dynamical variables in (13). Separating the real and the each block being of rank n. The proposition is proved.
imaginary parts and making use of Cauchy-Riemann equa-
tions we get by direct calculation that: Remark 2 If the transition to action-angle variables is not
analytic then Proposition 1 is not directly applicable.
{Ik,0 , Ij,0 }0 = {Ik,0 , Ij,1 }0 = {Ik,1 , Ij,1 }0 = 0, (14)

where 4 Hamiltonian Reductions and Real

Xn 
∂F ∂G ∂F ∂G Hamiltonian forms
{F, G}0 ≡ − − (15)
∂p s,0 ∂qs,0 ∂q s,0 ∂ps,0
s=1
 In this Section we will show how we associate with each
∂F ∂G ∂F ∂G Hamiltonian system H a family of RHF. We will do this
− + = 0, by using a special type of reductions.
∂ps,1 ∂qs,1 ∂qs,1 ∂ps,1
There are several methods to reduce Hamiltonian sys-
are the Poisson brackets related to the symplectic form ω0 tems to systems with lower number of degrees of freedom.
(10). One of the best known is the reduction by using inte-
Quite analogously we can prove that this set of inte- grals of motion [8]. Indeed, if we constrain our dynami-
grals are in involution also with respect to the Poisson cal system by fixing the values of p integrals of motion
brackets {·, ·}1 related
 to the symplectic form ω1 . Thus Ij (qk , pk ) = const , j = 1, . . . , p, then such constraint is
we have proved that MC , H0 , ω0 has 2n integrals in in- compatible with the dynamics and leads to reduced sys-
volution. tem with n − p degrees of freedom. Here, we will follow
The next step is to prove that these 2n integrals are somewhat different route by restricting the dynamics to
functionally independent provided the initial ones Ik are. MR which is the fixed point set of a Cartan-like involu-
The independence of Ik can be expressed by tive automorphism C, e defined below.
X The approach we will follow in this paragraph is in-
dI1 ∧. . .∧dIn ≡ Wi1 ,...,in dzi1 ∧. . .∧dzin 6= 0, (16) spired by the basic idea of construction of real forms for
i1 <...<in simple Lie algebras [9]. A basic tool in our construction is
the automorphism Ce , which plays the rôle of a “complex
where Wi1 ,...,in is the minor of order n of the n×2n matrix conjugation operator”.
W with matrix elements Let us introduce involutive canonical automorphism C
∂Ij (i.e. C 2 = 1 , C ∗ ω = ω) on the phase space M(n) and on
Wjk = , j = 1, . . . , n, k = 1, . . . , 2n. (17) its dual1 by:
∂zk
determined by the columns 1 ≤ ii < . . . < in ≤ 2n. Here C ({F, G}) = {C(F ), C(G)} , C 2 = 11, (20)
and below we will denote by z the 2n-component vector 1
In general we should have different notations of C in these
with components zi = qi and zi+n = pi for i = 1, . . . , n. In spaces [8]. However, since our phase space is a vector space
other words the independence of Ik means that rank W = with some abuse of notations we will use the same letter for
n. both realizations.
4 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems

(n)
where F , G ∈ F (M(n) ) are real analytic functions on The real form MR of the phase space is the symplec-
M(n) . The involution acts on them by: (2n)
tic (with respect to ω0 ) subspace of MC invariant with
e
respect to C:
C(F (p1 , . . . , pn , q1 , . . . , qn ))
= F ((C(p1 ), . . . , C(pn ), C(q1 ), . . . , C(qn ))). (21) (n) (n ) (n )
MR = M+ + ⊕ iM− − (27)
In terms of vector fields X, Y ∈ T M and the lifted
(n)
involution T C : T M → T M we have: Indeed any element of MR can be represented as:
2
ω(T C(X), T C(Y )) = C(ω(X, Y )), (T C) = 11 (22) (n)
Z = X + iY ∈ MR . (28)
where ω is the symplectic form corresponding to the Pois-
(n )
son brackets in (20). where X and Y are real-valued elements of M+ + and
Since C has eigenvalues 1 and −1, it naturally splits (n )
M− − respectively. The reality condition means that
(n ) (n )
M(n) into two subspaces M(n) = M+ + ⊕ M− − such
that ˜
C(Z) ≡ C(Z∗ ) = C(X − iY) = X + iY = Z (29)
(n )
CX = X for X∈ M+ + ,
where we have made use of Eq. (23).
(n )
CY = −Y for Y∈ M− − (23) (n)
Here and below by FR and GR ∈ F (MR ) we denote
(n )
the restriction of the observables F, G by restricting their
where n± = dim M± ± , n = n+ + n− . We also assume arguments to MR . Then FR and GR satisfy the analog
that the starting Hamiltonian H is invariant with respect of Eq. (21) with C replaced by C. e Due to Eq. (20) their
to C: (n) (n)
Poisson bracket {FR , GR } ∈ F (MR ) too, i.e. F (MR )
C(H) = H. (24)
becomes a Poisson subalgebra. If we choose the Hamilto-
Equation (20) guarantees that each of the subspaces (n)
nian HR ∈ F (MR ) then
(n ) (n ) (2n)
M+ + , M− − is a symplectic subspace of MC . On
(n ) (n )
M+ + and M− − we have e R ) = HR .
C(H (30)
n+ n−
X X The evolution associated with HR of the dynamical vari-
ω+ = dp+ +
k ∧ dqk . ω− = dp− −
k ∧ dqk able FR :
k=1 k=1
dFR
where p+ + − − + (n ) = {HR , FR }, (31)
k , qk (resp. pk , qk ) are basis in M+ (resp. dt
(n− )
M− ). With respect to the automorphism C they satisfy: (n)
defines a dynamics on MR . Rewriting (31) into its equiv-
C(p± ±
k ) = ±pk , C(qk± ) = ±qk± (25) alent form:
ω(XHR , ·) = dHR · (32)
for all k = 1, . . . , n± .
(n) and making use of (22) we see that the vector field XHR
In MC along with C we can introduce also the com-
plex conjugation ∗. In this construction obviously C com- must also satisfy
mutes with ∗ and their composition Ce = C ◦ ∗ = ∗ ◦ C is
e H ) = Xe
C(X
again an involutive automorphism on MC .
(2n) R C(HR ) = XHR (33)

Corollary 1 Let Ij , j = 1, . . . , n be n integrals of motion (n)

on MR . The symplectic form restricted on MR
(n)
also
in involution of our initial system H, depending analyti- becomes real and equals
cally on qk and pk . Let us denote their extensions to MC
by Ij = Ij,0 + iIj,1 . Then both Ij and n+
X n−
X
ωR = dp+
k,0 ∧ +
dqk,0 − dp− −
k,1 ∧ dqk,1 (34)
e j) = I∗
C(I (26) k=1 k=1
j

are also integrals of motion in involution for the complex- where p+ + − −

k,0 , qk,0 , k = 1, . . . , n+ and pk,1 , qk,1 , k = 1, . . . , n−
ified system HC . (n)
are the basic elements in MR .
Using eq. (20) it is easy to check that the dynamics (2n)
If MC is endowed with Hamiltonian which is “real”
generated by a Hamiltonian which is invariant with re-
spect to C (24) will have the subspaces F+ and F− of C- with respect to Ce and whose vector field XH satisfies
(n)
invariant and C-antiinvariant functions as invariant sub- (33) then the restriction of the dynamics on {MR , ωR ,
spaces. Moreover F+ turns out to be also an invariant HR } is well defined and coincides with the dynamics on
(2n) (n)
subalgebra. {MC , ω0 , H0 } restricted to MR .
 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems 5

The complexified equations of motion take the form: MR = M+ ⊕ iM− , HR (p, q) = H0 (pk , qk )|MR ,
n+ n−
dp+ dp+ X X
k,0 ∂H0 k,1 ∂H0 ω R = ω 0 |M R = dp+ ∧ +
dqk,0 − dp̃− −
=− + , = + , k,0 k,1 ∧ dq̃k,1
dt ∂qk,0 dt ∂qk,1 k=1 k=1

dp−
k,1 ∂H0 dp−
k,0 ∂H0 where dp̃−
k,1 = −idp−
k,1 ,
−
dq̃k,1 = −
−idqk,1 . The proposition
= − , =− − ,
dt ∂qk,1 dt ∂qk,0 is proved.
+ +
dqk,0 ∂H0 dqk,1 ∂H0 Thus we have proved that the equations (35) can be
= , =− , (35) consistently restricted to MR and give rise to a well de-
dt ∂p+
k,0
dt ∂p+
k,1
fined dynamical system with n degrees of freedom HR ≡
− −
dqk,1 ∂H0 dqk,0 ∂H0 {MR , ωR , HR } which we call a real Hamiltonian form (RHF)
=− , = of the initial Hamiltonian system H ≡ {M, ω, H}.
dt ∂p−
k,1
dt ∂p−
k,0
Let us now consider the set of observables FR , GR re-
(2n) lated to our RHF HR . They can be obtained from the
We shall show that the dynamics on {MC , ω0 , H0 }
observables of the complexified system HC by restricting
e
reduces naturally to MR which is the fixed point set of C. their variables to MR . We remind that we are selecting
the class of observables which depends analytically on the
Proposition 2 Let the Hamiltonian system HC and the
dynamical variables. This means, that after restricting on
involution C˜ be as above. Then the second class Dirac con-
MR they satisfy
straints

p+ +
p− − ∂F0 ∂F0 ∂F0 ∂F0
k,1 = 0, qk,1 = 0, k,0 = 0, qk,0 =0 (36) = = = =0
+ −
∂qk,1 ∂qk,0 ∂p+
k,1 ∂p−
k,0
MR MR MR MR
on the subspace MR invariant with respect to C˜ are com- (39)
patible with the dynamics of the complexified Hamiltonian and analogous relations for the partial derivatives of G. If
system HC . The reduced equations of motion take the form: we now calculate the Poisson brackets (15) between two
observables FR and GR we easily find that due to eqs. (39)
dp+
k,0 ∂HR they will simplify to
=− + ,
dt ∂qk,0 !
n+
X ∂FR ∂GR ∂FR ∂GR
dp−
k,1 ∂HR {FR , GR }0 ≡ + − (40)
= − , (37) ∂p+
s,0 ∂qs,0
+
∂qs,0 ∂p+s,0
dt ∂qk,1 s=1
!
X n
+
dqk,0 ∂FR ∂GR ∂FR ∂GR
∂HR − − −
= , ∂p− −
∂p−
dt ∂p+ s=n +1 + s,1 ∂qs,1 ∂qs,1 s,1
k,0
−
dqk,1 ∂HR Obviously the Poisson brackets (40) correspond to the
=−
dt ∂p−
k,1
symplectic form ωR .

where HR = H0 |MR . Proposition 3 The RHF HR corresponding to a Liouville

integrable Hamiltonian system H is Liouville integrable.
Proof Since H is real analytic function, then
Proof We start with a H which has n integrals in involu-
H0 (. . . , p+ + + + tion Ik depending analytically on the dynamical variables.
k,1 , qk,1 , . . .) = H0 (. . . , −pk,1 , −qk,1 , . . .)
The complexification provides us with 2n integrals of mo-
is an even function of p+ + tion Ik and Ik∗ which are also in involution. Let us now
k,1 and qk,1 . Besides from the con-
dition C(H) = H it follows, that restrict ourselves to MR . To this end we use the involution
e It is easy to check that n of the integrals are preserved:
C.
H0 (. . . , p− − − −
k,0 , qk,0 , . . .) = H0 (. . . , −pk,0 , −qk,0 , . . .)
+ 1  e k)

Ik,R = Ik + C(I (41)
and H0 is an even function also of p− − 2 MR
k,0 and qk,0 . Obviously
the first derivatives of H0 with respect to these variables
and become invariant with respect to Ce while the other n
are odd functions and therefore:
integrals vanish:
∂H0
=
∂H0
=
∂H0
=
∂H0
= 0. − 1  e k)

+
∂qk,1 −
∂qk,0 ∂p+ ∂p− Ik,R = Ik − C(I = 0, (42)
MR MR k,1 MR k,0 MR 2 MR
(38) +
The real Hamiltonian form is determined by: The fact that the integrals Ik,R are in involution with
respect to the Poisson brackets {·, ·}R follows from (40).
HR ≡ {MR , HR (p, q), ωR } , The proposition is proved.
6 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems

Those Ij which have definite C-parity will produce real and we have to satisfy:
first integrals Ij,R for the real form dynamics and those
with minus C-parity will produce purely imaginary inte- ∂H ∂HR ∂H ∂HR
+ = + =
grals. ∂qk ∂qk+ ∂pk ∂p+
k
Remark 3 We proved that after restricting HC to MR the ∂H ∂HR ∂H ∂HR
− = − − =− . (44)
integrals of motion satisfy (42). One may ask whether the ∂qj ∂qj− ∂pj ∂p−
j
inverse statement also holds true. Namely, assume that we
restrict HC by using simultaneously all n constraints in Comparing the equations from the first and second
(42) together with their symplectic conjugate ones. Since rows of (44) we obtain that all mixed derivatives of H
−
Ik,R are all independent on MC then such a procedure should vanish:
would lead to a dynamical system with n degrees of free-
dom. Will this new system coincide with HR ? We believe ∂2H ∂2H
= 0, − = 0,
that the answer to this question is positive, though we do ∂qk+ ∂qj− ∂p+
k ∂pj
not yet have a rigorous proof of this fact.
∂2H ∂2H
If we restrict our dynamical variables on MR then only = 0, − =0 (45)
∂qk+ ∂p−
j ∂p+
k ∂qj
the integrals invariant with respect to Ce survive.
Let us assume now that we have the complexified sys- which is exactly the separability condition i) of the Hamil-
tem HC restricted by (42). Let us now pick a point m ∈ tonian. Now ii) part of the Proposition follows trivially
MR which will correspond to the initial condition of our from the way we construct HR .
dynamical equations. Obviously the evolution can not take Due to the closedness of ω (and ωR ) we have:
this point out of MR . Indeed, if we assume the opposite
we easily find that the corresponding Hamiltonian vector LΓ ωR = d(Γ ωR ) (46)
e This means that the
field is not invariant with respect to C. !
X 2
∂ H 2
∂ H
Hamiltonian H is not invariant with respect to C which is =2 + − − +
dqk ∧ dpj + − + dqj ∧ dpk
a contradiction. j,k
∂qk+ ∂p−
j ∂qj ∂pk
Following the same line we can prove that if the initial
point belongs to M\MR then the evolution can not take and vanishing of all mixed derivatives in eq.(45) is the con-
it out of M\MR . dition for the required compatibility and/or local Hamil-
As a result the set of constraints (42) split MC into tonianity. Since the separability of H is equivalent to the
two orthogonal invariant subspaces. separability of HR we have similarly compatibility of ΓR
For some special choices of H it may happen that the with ω and local Hamiltonianity of ΓR with respect to ω.
dynamics of H and HR coincide.
Despite the apparent simplicity of the transition to the When the conditions of the Proposition 4 are fulfilled
real form dynamics it is essentially different from the ini- our procedure will give us a bi-Hamiltonian description of
tial one—the real form dynamical vector field satisfying the initial dynamics with a recursion operator
ΓR ωR = −dHR is generally not even a (locally) Hamil-
tonian vector field for the initial ω. As a result we may T = ω −1 ◦ ωR = 11+ − 11−
expect that the new and the initial dynamical vector fields
will coincide only for very few cases. Conditions for this (n )
where 11± are the identity tensors on M± ± . So, whatever
are settled in the following the outcome we gain either new dynamics or a bi-Hamil-
Proposition 4 ΓR = Γ iff tonianity of the old one.
i) the Hamiltonian is separable in a sum of two parts For instance, if we haveP a collection of uncoupled har-
depending on as follows: H = H+ (p+ + − − monic oscillators H = i (p2i + qi2 ) and a compatible in-
k , qk ) + H− (pj , qj )
volution: C(pi ) = Pǫi pi , C(qi ) = ǫi qi with ǫi = ±1 then we
where qk+ , p+ − −
k (resp. qj , pj ) are elements of M+ (resp. will have HR = i ǫi (p2i + qi2 ) and ΓR = Γ i.e. coinciding
M− ), and initial and real form dynamics. The same will be also true
ii) H− (ip− − − −
j , iqj ) = −H− (pj , qj ). for U (ip− − − −
j , iqj ) = −U (pj , qj ).
Also, condition i) is equivalent both to the local Hamil- At present, we do not have a receipt how to construct
tonianity of Γ with respect to ωR and to the compatibility and classify all involutions of M consistent with a given
of Γ with ωR i.e. LΓ ωR = 0 . Hamiltonian. However in many important cases it is pos-
Proof The dynamical vector field satisfying ΓR ωR = sible to provide a number of such non-trivial involutions.
−dHR will be
n+  
X ∂HR ∂ ∂HR ∂
ΓR = + −
5 RHF of Completely Integrable systems
k=1
∂p+
k ∂qk ∂qk+ ∂p+k
n−
! Here we understand the notion of completely integrable
X ∂HR ∂ ∂HR ∂
+ − (43) systems in a broader sense [10]. We will say that a dy-
j=1
∂qj− ∂p−
j ∂p−j ∂qj
−
namical system with n degrees of freedom is completely
 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems 7

integrable if we find 2n functions {Ik , φk }, k = 1, . . . , n In terms of Ik± , φ±

k the tensor field T and ω1 can be ex-
such that Ik are integrals of motion, i.e.: pressed by:
n+ n− n+ n−
LΓ Ik = 0, k = 1, . . . , n, (47) X X X X
T = Tk+ + Tk− , ω1 = +
ω1,k + −
ω1,k ,
and φk are nilpotent with respect to the vector field LΓ k=1 k=1

k=1

k=1
of order 2: ∂ ∂
Tk± = ζk± (Ik± ) dIk± ⊗ ± + dφ±
k ⊗ ,
∂Ik ∂φ±k
LΓ LΓ φk = 0, k = 1, . . . , n. (48) ±
ω1,k = ζk± (Ik± )dIk± ∧ dφ±
k (53)
If there is one or more functions of the φ’s which are
constants of the motion, the system is said to be superin- where ζk± (Ik± ) are some functions of Ik± ; we assume that
tegrable [11]. We also assume that {Ik , φk } introduce local they are real analytic functions of their variables.
coordinates on our phase space. In order that ω1 also satisfy eq. (22) it is enough that
Such set of functions {Ik , φk } generalize the standard C(T ) = T . In particular this means that ζk− must be even
notion of action-angle variables (AAV) [10]. While the functions of Ik− . If this is so then we can repeat our con-
standard AAV can be introduced only for compact mo- struction also for ω1 ; i.e., we can complexify it and then
tion on a torus, our variables {Ik , φk } can be derived also restrict it onto MR with the result:
for non-compact dynamics. This is important to our pur- n+
X n−
X
poses, especially when we analyze how the transition from TR = +
TR,k + −
TR,k , (54)
one RHF to another changes the character of the motion k=1 k=1
from compact to a non-compact one and vice versa, see  
+ ∂ ∂
Section 8 below. Therefore from now on, with some abuse TR,k = ζk+ (Ik+ ) +
dIk ⊗ + + dφk ⊗ +
,
of language we will say that {Ik , φk } are our AAV. ∂Ik ∂φ+ k
 
The complexification renders all Ik± and φ± k complex: − ∂ ∂
C,± ± ± C,± ± ± TR,k = ζk− (iIk− ) dIk− ⊗ − + dφ− k ⊗ ,
Ik = Ik,0 + iIk,1 and φk = φk,0 + iφk,1 and the auto- ∂Ik ∂φ− k
morphism Ce has the form: n+
X n−
X
ω1,R = ζk+ (Ik+ )dIk+ ∧ dφ+k − ζk− (iIk− )dIk− ∧ dφ−
k.
e C,± ) = ±(dI C,± )∗ ,
C(dI e C,± ) = ±(dφC,± )∗ (49)
C(dφ k=1 k=1
k k k k
Thus by construction, the restriction to other real forms
The automorphism Ce obviously satisfies the condition (33). preserves the Nijenhuis property of T and the condition
In order to satisfy also (24) we need to assume that H is for double degenerate and nowhere constant eigenvalues.
an even function of all Ik− ∈ M− . Then the restriction Also it is easy to check that it is preserved by the dy-
on MR according to (28) and (49) means restricting all namical flow: LΓR TR = 0. Due to the fact that existence
Ik+ , φ+ − −
k ∈ M+ to be real, while all Ik , φk ∈ M− become of recursion operators is equivalent [14] to integrability at
purely imaginary. Then: least in the non-resonant case, this line of argumentation
+ + − −
HR = H(I1,0 , . . . , In+ ,0 , iI1,1 , . . . , iIn− ,1 ), gives us another instrument to treat the integrability of
real forms.
n+ n−
X X The separability of H (51) looks rather restrictive con-
+ + − −
ωR = dIk,0 ∧ dφk,0 − dIk,1 ∧ dφk,1 (50) dition. However, all integrable systems obtained by reduc-
k=1 k=1 ing a soliton equation (like, e.g. the nonlinear Schrödinger
equation) on its N -soliton sector are separable. Finite-
and obviously we have again a completely integrable Hamil-
dimensional systems allowing Lax pairs like Toda chains,
tonian system.
Calogero-Moser systems etc. also possess this property, see
Till the rest of this section we also assume that the
eq. (70) below.
Hamiltonian is separable, i.e.
n+ n−
X X
H= h+ +
k (Ik ) + h− −
k (Ik ). (51) 6 Examples
k=1 k=1
We illustrate our construction by several paradigmatic ex-
Obviously the condition (24) requires that h− − amples which include several types of Toda chain models
k (Ik ) must
−
be even functions of Ik . and Calogero-Moser models.
Another important approach to completely integrable Example 1 Toda chain related to the sl(n, C) algebra. We
systems [12] is based on the notion of the recursion opera- consider simultaneously the conformal and the affine case
tor – a (1,1) tensor field with vanishing Nijenhuis torsion by introducing the parameter c0 which takes two values:
[13]. This tensor field reflects the possibility to introduce c0 = 0 (conformal case) and c0 = 1 (affine case).
a second symplectic structure ω1 on M through:
n
X n−1
X
p2
1 HTC = k
+ eqk+1 −qk + c0 eq1 −qn ,
ω1 (X, Y ) = (ω(T X, Y ) + ω(X, T Y )) . (52) 2
2 k=1 k=1
8 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems

n
X with:
ω= dpk ∧ dqk . (55) 
g 2 V−n , for An−1
k=1 U=
g 2 (V−n + V+n ) + g12 V1n + g22 V2n , for BCn .
We choose the involution as:
Here
X X
C(pk ) = −pk̄ , C(qk ) = −qk̄ . (56) V±n = v(qj ± qk ), V1n = v(qj ),
j<k≤n j≤n
where k̄ = n + 1 − k. We also introduce new symplectic X
coordinates adapted to the action of C: V2n = v(2qj ), (62)
j≤n
pk − pk̄ qk − qk̄
p+
k = √ , qk+ = √ , (57) and the function v(x) is one of the following:
2 2
pk + pk̄ qk + q 1 1 1 1
pe−
k = √ , qek− = √ k̄ , k = 1, . . . , r = [n/2] v(q) = , , , + ωq 2 , ℘(q) (63)
2 2 q2 sin2 q sinh2 q q2
where n = 2r + r0 with r0 = 0 or 1; for r0 = 1 we have to where ℘(q) is the Weierstrass function.
add also: pe = pr+1 , qe = qr+1 . These coordinates are such The BCn case contains in itself the CMS for the alge-
that bras Bn , Cn and Dn ; they are obtained by putting g2 = 0,
g1 = 0 and g1 = g2 = 0 respectively.
C(p± ±
k ) = ±pk , C(eqk± ) = ±e
qk± , (58) All these models are invariant under the involution
C(e
p) = −ep, C(e
q) = −e q (56). For convenience we again use the coordinates (58)
together with the following complex coordinates
As a result we obtain the following real forms of the TC 1
1

model: i) for n = 2r + 1: zj = √ qk+ + iqk− , zj∗ = √ qk+ − iqk− ,
2 2
r
1X + 2 1 − 2 k = 1, . . . , r.
HTC1 = ((pk ) − (p− 2
k ) ) − (pr+1 ) + In order to obtain the Hamiltonian of the RHF of the
2 2
k=1 CMS we have to express HCMS (61) in the new coordinates
r−1
X +
−
qk+1
+ − q−
√ and assume that pe− k,qek− , pe and qe are purely imaginary and
+2 e(qk+1 −qk )/√ k 2
cos then calculate its real part. We shall denote the imaginary
k=1
2 parts of these variables by the same letter but without
 
+
√
− q − tilde and so all tilde-less variables will be real. As a result:
+ 2e−qr / 2 cos qr+1 − √r , (59)
2 n r r
r r+1 X p2j 1X + 2 1X − 2 1
X X Re = (pk ) − (pk ) − r0 p2 ,
ωR = dp+ +
k ∧ dqk − dp− −
k ∧ dqk . 2 2 2 2
j=1 k=1 k=1
k=1 k=1
r
X
ii) for n = 2r: Re V1n = 2Re v(zj ) − r0 v(q),
k=1
r
1X + 2 √
− 2qr+ Xr
HTC2 = ((pk ) − (p− 2
k) )+e + (60) Re V2n = 2Re v(2zj ) − r0 v(2q),
2
k=1 k=1
r−1 X
X + +
√ q− − q− Re V−n = 2Re [v(zj − zk ) + v(zj + zk∗ )] +
+2 e (qk+1 −qk )/ 2
cos k+1√ k ,
k=1
2 j<k≤r
r
X X
r
X r
X √
ωR = dp+ +
dp− − + v( 2qk+ ) + 2r0 Re v(zj − iq),
k ∧ dqk − k ∧ dqk .
k=1 j≤r
k=1 k=1 X
These models are generalizations of the well known Re V+n = 2Re [v(zj + zk ) + v(zj − zk∗ )] +
j<k≤r
Toda chain models associated to the classical Lie algebras
r
[15]; indeed if we put qk− ≡ 0 and p−
k ≡ 0 we find that (59)
X √ X
goes into the Br TC while (60) provides the Cr TC. − v( 2qk− ) + 2r0 Re v(zj + iq).
k=1 j≤r
Example 2 The real Hamiltonian forms for the Calogero– To illustrate the form of the new Calogero-Moser poten-
Moser systems (CMS) [16]. The CMS corresponding to the tials we note that:
root systems of A, B, C, D and BC-series [17] are defined
by Hamiltonians of the type: 1 (sin x cosh y)2 − (cos x sinh y)2
Re 2 = ,
sin (x + iy) ((sin x)2 + (sinh y)2 )2
n
X
HCMS = 1/2 p2j + U (q) (61) 1 (sinh x cos y)2 − (cosh x sin y)2
Re 2 = .
j=1 sinh (x + iy) ((sinh x)2 + (sin y)2 )2
 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems 9


Together with + v(zi + zj∗ ) + v(zi + zj ) + v(zi − zj∗ ) +
r
X X r
r
X
+ 2gS2 Re v(zj ) + 2gL 2
Re v(2zj ).
ωR = dp+ + − −
k ∧ dqk − dpk ∧ dqk − r0 dp ∧ dq (64)
j=1 j=1
k=1

they define the real form dynamics which is obviously not Note that: v(iy) = −v(y).
confined to the standard Calogero-Moser models. Example 3 In the case of involution C2 = Se1 −e2 :
More explicitly: C2 (q1 ) = q2 , C2 (q2 ) = q1 , C2 (p1 ) = p2 , C2 (p2 ) = p1 ,
for n = 2r + 1 and g ≃ An−1 :
we obtain for g ≃ BCn :
r
1 X + 2  1 − 2 " #
HCM,R = (pk ) − (p−
k)
2
− (pr+1 ) + 1 Xn
2
k=1
2 HCM,R = (p+ 2 − 2
1 ) − (p1 ) + (p+
k)
2
+
2
Xr k=3
2
   
+ 2g Re v(zi − zj ) + v(zi + zj∗ ) + Xn
i<j + gS2 2Re v(z1 ) + Re v(qj ) +
r
X r
X
√ j=3
+ g2 v( 2qj+ ) + 2g 2 −
Re v(zi − iqr+1 ). 
 X n
 
j=1 j=1
+ g2 2 Re v(z1 − qj+ ) + v(z1 + qj+ ) −

for n = 2r and g ≃ An−1 : j=3
√ − √
r r −v( 2q1 ) + v( 2q1+ ) +
1 X + 2 − 2
 X √ 
HCM,R = (pk ) − (pk ) + g 2
v( 2qj+ ) + Xn 
2
k=1 j=1 + [v(qk − qj ) + v(qk + qj )] +
r 
X   3≤k<j
2
+ 2g Re v(zi − zj ) + v(zi + zj∗ ) .  
X n
i<j 2 
+ gL 2Re v(2z1 ) + Re v(2qj )
for n = 2r + 1 and g ≃ BCn : j=3

r Example 4 In the case of involution C3 = −11:

1 X + 2  1 − 2
HCM,R = (pk ) − (p−
k)
2
− (pr+1 ) +
2 2 C2 (qk ) = −qk , C2 (pk ) = −pk ,
k=1
r h
X i
√ √ qk− = qk , p−
+ g2 v( 2qj+ ) − v( 2qj− ) + k = pk , dim M− = 2n.

j=1 We obtain for Toda chains with the algebra An−1 ≃

r
X sl(n):
2
+ 2g Re [v(zi − zj ) + n n−1
i<j 1 X 2 m2 X β(αk ,q)
 HTC = pk + 2 e + c0 eβ(α0 ,q) ,
+ v(zi + zj∗ ) + v(zi + zj ) + v(zi − zj∗ ) + 2 β
k=1 k=1
r
X  − −
 where c0 = 0 stands for conformal TC and c0 = 1 for
+ 2g 2 Re v(zi − iqr+1 ) + v(zi + iqr+1 ) −
affine TC. Now all pk and qk become purely imaginary.
j=1
We choose also β = iβ0 . As a result we obtain Toda chain
− −
− gS2 v(qr+1 2
) − gL v(2qr+1 )+ with purely imaginary interaction constant.
Xr Xr Similarly, for CMS we obtain:
+ 2gS 2
Re v(zj ) + 2gL2
Re v(2zj ). A) If v(q) = 1/q 2 or v(q) = 1/q 2 +ω 2 q 2 then we obtain:
j=1 j=1
HCM,R = −H, which is equivalent to t ↔ −t.
for n = 2r and g ≃ BCn :
B) If v2 (q) = a2 / sin2 (aq) and v3 (q) = a2 / sinh2 (aq)
r
X then the real form dynamics is obtained by the exchange:
1  
HCM,R = (p+ 2 − 2
k ) − (pk ) +
2 v2 (q) ↔ −v3 (q), t ↔ −t.
k=1
r h
X i
√ √ C) If v4 (q) = a2 ℘(aq|ω1 , ω2 ) then the real form dy-
+ g2 v( 2qj+ ) − v( 2qj− ) +
namics is obtained by the exchange:
j=1
r
X v4 (q) ↔ −v4′ (q), t ↔ −t.
2
+ 2g Re [v(zi − zj ) +
i<j where v4′ (q) = −a2 ℘(aq|iω1 , iω2 ).
10 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems

Remark 4 Several of the RHF of CMS have been obtained y22 + y32 . These coordinates are connected to the pseudo-
earlier by Calogero [18] using the so-called ‘duplication spherical coordinates (r, ϑ, ϕ) by:
procedure’. Our results above show that each ‘duplication’
is related to a Cartan-like automorphism of the relevant y1 = r cosh ϑ, y2 = r sinh ϑ cos ϕ, y3 = r sinh ϑ sin ϕ.
Lie algebra. This fact can be used to classify all inequiva-
lent RHF of CMS. The equation y 2 = −r2 , i.e. the locus of points equidistant
from the origin, specifies a hyperboloid of two sheets inter-
Example 5 An example from general relativity: transition secting the y1 axis at the points ±r called poles in analogy
from SO(3) to SO(2, 1) symmetry. with the sphere. Either sheet (say the upper sheet) mod-
Recall that the geodesic Schwarzschild flow, correspond- els an infinite spacelike surface without a boundary; hence,
ing to a metric g = gij dxi dxj on M, can be viewed as the Minkowski metric becomes positive definite (Rieman-
the flow, generated by the Hamiltonian field XH with nian) upon it. This surface has constant Gaussian curva-
H = 41 g ij pi pj (pi ’s are conjugated to xi ’s coordinates on ture (R = −1/r2 ), and it is the only simply connected sur-
T ∗ M). face with this property. Other embeddings of the pseudo-
For example, the geodesic flow for the Schwarzschild sphere in the 3-dimensional Euclidean space are also avail-
metric able, for example it can be regarded as the 2 -dimensional
 2  surface generated by the tractrix, but they are not global.
rc − 2M G 2 rc2 2 It is worth noting that metrics g1 and g2 are so (3)-
g1 = ǫ1 dt − 2 dr +
rc2 rc − 2M G invariant and so (2, 1)-invariant2, respectively, and that


+ ǫ2 r2 dϑ2 + sin2 ϑ dϕ2 , (65) geodesic flows, corresponding to the model Ricci-flat met-
rics, associated with 3-dimensional Killing algebras, are
where ǫ1 = ±1, ǫ2 = ±1, is described by the (completely integrable. To see that one may observe that, for instance,
integrable) Hamiltonian Hamiltonian H2 possesses five independent first integrals
  2  1
1 rc − 2M G 2 rc2 − 2M G 2 H2 , pτ , p2ϑ + p2ϕ , I1 , I2
H1 = ǫ1 p − p +
4 rc 2 t
rc 2 r sinh2 ϑ
 
ǫ2 1
+ 2 p2ϑ + p 2
ϕ . (66) where I1 and I2 are generators of a noncommutative two-
r sin2 ϑ dimensional subalgebra of so (2, 1), for example
h  i
The complexification and real projection procedure leads I = 1 + √2 cos ϕ + sin ϕ p +
1 ϑ
to the following Hamiltonian h  
√ √  i
  2 2
 + 1 + 2 + coth ϑ cos ϕ − 1 + 2 sin ϕ pϕ
1 rc − 2M G 2 rc − 2M G 2
H2 = ǫ1 pt − pr − √
4 rc2 rc2 I2 = 2 [cos ϕ + sin ϕ] pϑ +
  h √ i
ǫ2 2 1 2 + 2 + 2 coth ϑ (cos ϕ − sin ϕ) pϕ .
− 2 pϑ + pϕ (67)
r sinh2 ϑ
The 5 first integrals span a rank 3 Lie algebra A, and since
which (is still completely integrable and) is associated with
the metric
rank A + dim A = dim T ∗ M and rank A < dim A
 2 2

rc − 2M G 2 rc
g2 = ǫ1 2
dt − 2 dr2 − the system is noncommutatively integrable in the sense
rc rc − 2M G

[22,23].
−ǫ2 r2 dϑ2 + sinh2 ϑ dϕ2 , (68) Geodesic flows corresponding to the metrics invariant
for a 3-dimensional Lie algebra are discussed in [20]. Note
solution of vacuum Einstein field equations [19,20]. also that the above proposition is no more valid for the
The above notation for coordinates might be mislead- geodesic flow of Ricci-flat metrics with nonextendable two-
ing because, in g1 and H1 , r ∈ ]0, ∞[ , ϑ ∈ ]−π, π[ , ϕ ∈ dimensional Killing algebras.
[0, 2π[ denote spherical coordinates, while, in g2 and H2 ,
r ∈ ]0, ∞[ , ϑ ∈ R, ϕ ∈ [0, 2π[ denote pseudo-spherical co-
ordinates.
The 2-dimensional surfaces endowed with metric g2
7 Lax operators
restricted by (r, t = const ) may be identified with one of
the sheets of the two-sheeted space-like hyperboloid. They It is well known that a number of completely integrable
are also known as pseudo-spheres. systems admit Lax representation. By it we mean the ex-
The pseudo-sphere is a surface with constant negative istence of two matrices L(p, q) and M (p, q) depending
Gaussian curvature R = −1/r2 . It can be globally embed- 2
In the pseudo-spherical coordinates, the so(2, 1) Lie alge-
ded in a 3-dimensional Minkowskian space. Let y1 , y2 , y3 bra [X1 , X2 ] = X3 , [X2 , X3 ] = −X1 , [X3 , X1 ] = −X2 ,, is
denote the coordinates in the Minkowskian space, where spanned by X1 = sin ϕ∂ϑ + cos ϕ coth ϑ∂ϕ , X2 = − cos ϕ∂ϑ +
the separation from the origin is given by y 2 = −y12 + sin ϕ coth ϑ∂ϕ , X3 = ∂ϕ [21].
 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems 11

explicitly on the dynamical variables and such that the where αk is the set of simple roots of g, α0 is the minimal
Lax equation root of g,
dL 1
= [L, M ] (69) ak = exp((q, αk )), (74)
dt 2
is equivalent to the corresponding equations of motion of and c0 was introduced in (55) above. By Eα and Hk above
H. Let us also remind that from (69) there follows imme- we denote the Cartan-Weyl generators of g; they satisfy
diately that the eigenvalues ζs of L are integrals of the the commutation relations:
motion in involution. One can use also as such
[Hk , Eα ] = (α, ek )Eα ,
n
X [Eα , E−α ] = Hα , (75)
Ik = tr Lk (p, q) = ζsk , k = 1, . . . , n, (70)
[Eα , Eβ ] = Nα,β Eα+β ,
s=1
where α and β ∈ ∆ are any two roots of g and Nα,β = 0
and as a rule the Hamiltonian H is a linear combination if α + β 6∈ ∆.
of Ik ’s; for example, for the Toda chain and for the CMS The involution C induces an involutive automorphism
H = 2 I2 . C in the algebra g. Indeed, both p and q can be viewed
If the matrices L and M are analytic (meromorphic) as vectors in the root space En ≃ h∗ which is dual to the
functions of pk and qk then obviously these properties will Cartan subalgebra h ⊂ g. Besides, C must preserve the
hold true also for the integrals Ik . The analytic properties Toda chain Hamiltonian:
of ζs as functions of pk and qk require additional consid-
n n
erations. Indeed, ζs are roots of the corresponding charac- 1 1X 2 X 2
teristic equation which is of order n so the mapping from HTC = tr L2TC = pk + ak + c20 a20 . (76)
2 2
pk and qk to ζs may have essential singularities. k=1 k=1

The Lax representation ensures also the separability of This means that C in the conformal (resp. affine) case must
all Ik (including H) in terms of ζs . For some of the best preserve the system of simple (resp. admissible) roots of g.
known cases like for the Toda chain, the matrices L and This allows one to associate with the involutive automor-
M take values in the normal real form of some simple Lie phism C an involutive automorphism C # of the algebra g.
algebra g. This means that L is a linear combination of Obviously, if C is defined on M by
the Cartan-Weyl generators with real valued coefficients.
n
X n
X
The complexification procedure renders pk and qk com-
plex and as a result both L and M become complex- C(pk ) = cks ps , C(qk ) = cks qs , (77)
valued: s=1 s=1
Pn
where cks are such that s=1 cks csm = δkm then the ac-
L → LC = L0 +iL1 , M → M C = M0 +iM1 , L, M ∈ gC .
tion of C # on h can be determined by duality as:
Note however, that the complexified Lax equation n
X
C # (Hk ) ≡ HC(ek ) = cks Hs . (78)
dLC s=1
= [LC , M C ], (71)
dt
On the root vectors Eα the automorphism C # acts as fol-
can again be written down in terms of purely real Lax lows:
matrices with doubled dimension: C # (Eα ) = nα EC # (Eα ) , nα = ±1. (79)
   
dL L0 L1 M0 M1 An well known fact [9] is that the subgroup of the au-
= [L, M], L = , M= . tomorphism group preserving h is determined by Ad h ⊗
dt −L1 L0 −M1 M0
(72) Wg ⊗ Vg , where Ad h is the subgroup of inner automor-
This fact we will use below when we discuss general- phisms by elements of the Cartan subgroup, Wg is the
izations of our approach. Weyl group and Vg is the group of outer automorphisms.
Vg is isomorphic to the symmetry group of the Dynkin
Let us now consider the Lax representation for the
diagram of g.
Toda chain related to the simple Lie algebra g with rank
Since the involution C for the conformal Toda chain
n:
must preserve the set of simple roots of g it must be related
n
X to a symmetry of the Dynkin diagram; therefore C # must
LTC = p k Hk + (73) correspond to an outer automorphism of g. For the affine
k=1 Toda chains C # must be related to a symmetry of the
Xn extended Dynkin diagram which allows for a larger set of
+ ak (Eαk + E−αk ) + c0 a0 (Eα0 + E−α0 ), choices for C # . These symmetries have been classified in
k=1 [24].
n
X
MTC = ak (Eαk − E−αk ) + c0 a0 (Eα0 − E−α0 ), Remark 5 The involution C # dual to C in eq. (56) is the
k=1
outer automorphism of sl(n).
12 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems

Analogously we consider the Lax representation (69) This means that the number of inequivalent choices for
of the Calogero-Moser models is of the form [17]: the involution C and therefore, the number of inequivalent
RHF of the Calogero-Moser systems is much bigger. To
n
X X classify all of them one has to consider the equivalence
L(p, q) = p j Hj + i gα x((q, α))Eα , (80) classes of W (g) and pick up just one element from each
j=1 α∈∆ class of second order elements. It would be also natural to
Xn X relate to each Satake diagram of g, or a subalgebra of g a
M (p, q) = z j Hj + gα y((q, α))Eα , (81) RHF of CMS. A detailed study will be reported elsewhere.
j=1 α∈∆ The Cartan subalgebra h and the algebra g, just like
M, can be splitted into direct sums:
The constant gα depends only on the length of the root
α. The three functions x(q), y(q) and z(q) satisfy a set of h = h+ ⊕ h− , g = g+ ⊕ g− , (86)
functional equations #
compatible with the involution C ; i.e.:
′′
x (ξ) C # (X) = X, for any X ∈ h+ , resp., for any X ∈ g+ ,
y(q) = −x′ (q), z(ξ) = , (82)
2x(ξ) #
C (Y ) = −Y, for any Y ∈ g− , resp., for any X ∈ h+ ,
x(ξ)x′ (η) − x(ηx′ (ξ) = x(ξ + η)[z(ξ) − z(η)], (83)
If we compose C with −† we obtain the Cartan involution
#

The Hamiltonian H = 1
2 tr L
2
and the function v(q) are Cf# which selects a real form of the algebra g .
R
related to them by:

v(q) = −x(q)x(−q). 8 Dynamics of the Toda chain and its RHF

The solutions to these equations are given by [16,17,18]: In this Section on the example of the Toda chains, we
1 briefly discuss how the two basic steps in constructing the
 , I RHF’s may affect the dynamical regimes of the Hamilto-



 ξ nian systems. These two steps are:
 −1
a) complexification and,
a coth(aξ), a sinh (aξ), II
x(ξ) = (84)

 a cotan (aξ), a sin−1 (aξ), III ˜
b) imposing the involution C.

 a cn (aξ) a dn (a ξ) a The dynamical regimes of the CTC were studied re-

 , , IV
sn (aξ) sn (aξ) sn (aξ) cently from the point of view of their applications to the
adiabatic N -soliton interactions [1]. It was already known
then the corresponding v(ξ) provide the choices in (63). that CTC has a much richer class of dynamical regimes
The involution C # induces a Cartan involution C on than the real TC. This shows that step a) (the complexifi-
∼ cation) changes drastically the character of the dynamics
gC by C (Z) = −C(Z † ) has all the properties of Cartan in- of out initial Hamiltonian system.
∼
volution. As a result the invariance condition with respect Indeed, it is well known [26,27] that the solutions of
to C restricts to a real form of g. The invariance condition the conformal Toda chains are parametrized by the eigen-
∼
values ζj and the first components rj of the normalized
for L has the form:
eigenvectors of the Lax matrix:
# #


C L(C (p), C (q), g = L(p, q, −g). (85) Lwj = ζj wj , rj = wj ,
(1)
(87)
∼
n
X
For the case of the example above with g ≃ sl(n) and (s) (s)
(wj , wk ) ≡ wj wk = δjk . (88)
x(q) = −x(−q) this can be easily shown by realizing that s=1
C(X)P can be written in the form C(X) = T XT −1 where
T = k=1 Ek,k̄ . One can check that (85) and an analo- For the Toda chain related to the algebra sl(n) these so-
n

gous relation on M (p, q, g) leave the Lax representation lutions take the form:
invariant. Compare this with Mikhailov reduction group X n

[25]. In this example we miss the spectral parameter, but q1 (t) = ln A 1 (t), A 1 (t) = rj2 e−2ζj t ,
we also have the interaction constant g and (85) involves j=1
non-trivial action also on g. Ak+1
The substantial difference as compare to the Toda chain qk (t) = ln , k = 2, 3, . . . , n, (89)
Ak
case is that the Calogero-Moser Hamiltonian is invariant X
with respect to the Weyl group of g as well as with respect Ak (t) = W 2 (j1 , . . . , jk )rj21 . . . rj2k × (90)
to the group of outer automorphisms of g. Therefore any j1 <...<jk
involutive element of Wg ⊗ Vg can be used to construct × exp(−2(ζj1 + . . . + ζjk )t)
a RHF of the CMS. Obviously two such RHF’s will be Yn
equivalent if the corresponding automorphisms C1 and C2 An = rj2 W 2 (1, 2, . . . , n), An+1 = 1,
belong to the same conjugacy class of Wg . k=1
 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems 13

where Wj1 ,...,jk is the Wandermonde determinant: For CTC one can derive the analogs of the relations
  (93):
1 1 ... 1
 ζj1 ζj2 . . . ζjk  lim (qk (t) − vk± t) = βk± , (98)
  t→±∞
Wj1 ,...,jk = det  .. .. . . . . (91)
 . . . ..  vk− = −2κk , vk+ = −2κn+1−k ; (99)
ζjk−1
1
ζjk−1
2
. . . ζjk−1
k
which shows that now it is the set of real parts {κj } of the
It is well known that for the real Toda chain: eigenvalues ζj which determine the asymptotic velocities.
1) all ζj are real and pair-wise different; Even if we assume3 that ζj 6= ζk for j 6= k then we still may
2) all rj are also real [28]. have κj = κk ; note that equal asymptotic velocities allow
This ensures that Ak (t) are positive for all t and there- for the possibility of bound states. Skipping the details
fore the solutions qk (t) are regular for all t. In addition one (see [1,2]) we just list the different types of asymptotic
can evaluate the asymptotic behavior of qk (t) for large |t|. regimes for the CTC which are determined by the so-called
If we assume that the eigenvalues ζk satisfy the sorting sorting condition:
condition:
ζ1 > ζ2 > . . . > ζn , (92) κ1 ≥ κ2 ≥ . . . ≥ κn . (100)

then we get the result: This sorting condition differs from (92) for the real TC in
that it allows equalities. Therefore the CTC allows several
lim (qk (t) − vk± t) = βk± , (93) non-degenerate regimes:
t→±∞
A) Asymptotically free regime: κj 6= κk for j 6= k;
vk− = −2ζk , vk+ = −2ζn+1−k ; (94) B) n-particle bound state for κ1 = κ2 = . . . = κn ;
C) A number of mixed regimes when we have two or
for the constants βk± one can derive explicit expressions in more groups of equal κj ’s;
terms of ζj . Another difference with respect to the real TC case is
If we interpret qk (t) as the trajectory of the k-th par- that Ak (t) are no more positive definite. For some choices
ticle then 2ζj characterize their asymptotic velocity. The of the ζj and rj they may vanish even for finite values of
property 1) above means that the (real) Toda chain allows t = t0 which means that qk (t) may develop singularity for
only for (non-compact) asymptotically free motion of the t → t0 .
particles. Let us now analyze the effect of the involutions C and
The set of coefficients {ζj , ln rj } are the action-angle C˜ ≡ C ◦ ∗ on the spectral data {ζk , ln rk } of L. It will allow
variables in the generalized sense, introduced in the be- us to describe the dynamical regimes of the RHF of CTC.
ginning of Section 5. Of course they are convenient for Eq. (56) leads to:
solving the TC model because: i) they satisfy canonical
Poisson brackets and ii) the TC Hamiltonian takes the ˜ k ) = −qn+1−k
C(q ∗
, ˜ k ) = −p∗n+1−k ,
C(p (101)
simple form:
Xn
and consequently:
HTC = 2ζj2 . (95)
j=1 ˜ k ) = −b∗n+1−k ,
C(b ˜ k ) = a∗n−k .
C(a (102)
Therefore the equations of motion for {ζj , ln rj2 } take the
These constraints mean that the Lax matrix L satisfies:
form:
˜ k ), C(q
S0 L(C(p ˜ k ))S −1 = −L(pk , qk )† , (103)
dζj d ln rj2 
0

= 0, = −2ζj , j = 1, . . . , n. (96) 0 0 ... 0 1
dt dt
 0 0 . . . −1 0 
 . .. . . .. .. 
The situation changes substantially after the complex- S0 =  . . . .
ification, see [2]. In fact the same formulae (89), (90) pro-  . . . (104)
 0 (−1)n−1 . . . 0 0 
vide the solution also for the complex case. But now we
(−1)n 0 ... 0 0
have neither of the properties 1) or 2). The eigenvalues ζj ,
as well as rj become complex:
This means that the eigenvalues and the matrix of eigen-
vectors of L satisfy:
ζj = κj + iηj , ρj + iφj = ln rj eq1 (0) , j = 1, . . . , n,
(97) S0 ZS0−1 = −Z ∗ , Z = diag (ζ1 , . . . , ζn ) or
where
n
Y ˜ ∗
C(ζk ) = −ζn+1−k . (105)
e−nq1 (0) = An = rj2 W 2 (1, . . . , n). 3
j=1 In principle we may have ζj = ζk for k 6= j which leads
to degenerate solutions; we will not consider such cases below,
This substantially modifies the properties of the solutions. see e.g. [2].
14 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems

Technically it is a bit more difficult to derive the action The energy of these RHF of sl(4)-CTC is given by:
˜ k ). Indeed, this must ensure that the functions Ak
of C(r
satisfy Hsl(4) = 8(κ21 − η12 ). (114)
˜ k ) = A∗ ,
C(A (106)
n−k
Due to the symmetry the ‘particles’ q3 (t) and q4 (t)
for all values of t. After some calculations we get that the
also form a bound state; the corresponding formulae eas-
relations
ily follow from (112) and (113) and q4 (t) = −q1∗ (t) and
ρn+1−k + iφn+1−k = −ρk + iφk + ln wk , (107) q3 (t) = −q2∗ (t).
W (1, . . . , k − 1, k + 1, . . . , n) This example demonstrates several peculiarities of the
wk = RHF dynamics. The first one is that configurations in
W (1, 2, . . . , n)
which all n particles form a bound state is impossible.
∗
ensure that Bn−k = Bk where Indeed, such bound state may take place if κj = 0 for all
j = 1, . . . , n. But it is easy to check that the sorting con-
Bk (t) = Ak (t)e−kq1 (0) . (108) dition (100) and the symmetry of the eigenvalues (109)
Let us illustrate this by first putting n = 4 and choos- lead to degeneracy of the spectrum of L for κj = 0.
ing The RHF dynamics allows either asymptotically free
regimes or mixed regimes with special symmetry, namely
ζ1 = −ζ4∗ = κ1 + iη1 , ζ2 = −ζ3∗ = κ1 − iη1 , we can have only even number of bound states which move
π
ρ1 = ρ0 + iφ0 , ρ3 = −ρ0 − i(φ0 + α + ), (109) with opposite asymptotic velocities.
2 The above considerations on the properties of the dy-
π namical regimes are not specific for the Toda chain. Sim-
ρ2 = ρ0 − iφ0 , ρ4 = −ρ0 + i(φ0 + α + ).
2 ilar conclusions can be derived also for the RHF of the
With these notations we can write down the solution Calogero-Moser systems. Here we have richer variety of
for the sl(4)-CTC in the form: RHF’s due to the fact that the CMS Hamiltonian is in-
variant with respect to the whole Weyl group of g. More
B2 (t)
q1 (t) = ln B1 (t), q2 (t) = ln , (110) detailed investigation of the CMS dynamical regimes will
B1 (t) be done elsewhere.
B3 (t)
q3 (t) = ln = −q2∗ (t), q4 (t) = − ln B3 (t) = −q1∗ (t),
B2 (t)
where
9 Discussion
1  −2κ1 t+2ρ0
B1 (t) = p e cos(2η1 t − 2φ0 ) 1. The method goes also for non-integrable models, though
25 κ1 η1 κ1 + η12 2 we treated mainly integrable ones.
 Other related aspects of our method concern more
− e2κ1 t−2ρ0 cos(2η1 t − 2φ0 − 2α) (111)
complicated objects such as dynamical classical r-matrices
−1 
B2 (t) = η 2 cosh(4η1 t − 4φ0 ) or quantum R-matrices due to their algebraic struc-
27 κ21 η12 (κ21 + η12 ) 1 tures will also satisfy relations of the form (85).

+ κ21 cos(4η1 t − 4φ0 − 2α) + κ21 + η12 , 2. Our list of examples can be substantially extended
η1 with Ruijsenaars-Sneider models and their generaliza-
B3 (t) = B1∗ , α1 = arg ζ1 = arctan . tions being among the obvious candidates.
κ1
3. The present approach can naturally be generalized also
The choice of ζ1 = ζ2∗ in (109) combined with the invo- for infinite-dimensional integrable models such as the
lution ζ1 = −ζ4∗ , ζ2 = −ζ3∗ ensures that q1 (t) and q2 (t) will 1 + 1-dimensional Toda field theories. This will allow
have equal asymptotic velocities and will form a bound one to construct new classes of real Hamiltonian forms
state. Indeed, from (110) and (111) we get: of Toda field theories extending the results of [6].
  4. The method can be used repeatedly: Since the com-
1 iπ
lim (q1 (t) + q2 (t)) ∓ 2κ1 t = ∓ 2ρ0 plexified model can be viewed as real Hamiltonian sys-
t→±∞ 2 2
  tem with 2n degrees of freedom, then we can start and
1 complexify it getting a Hamiltonian system with 4n de-
− ln 28 κ21 (κ21 + η12 ) + O e−4κ1 |t| , (112)
2 grees of freedom producing a real form dynamics with
which means that the center of mass of the particles q1 (t) 2n degrees of freedom.
and q2 (t) for t → ±∞ undergoes non-compact asymptoti- If the initial system allows Lax representation that de-
cally free motion with asymptotic velocities ±2κ1 . At the pends analytically on the dynamical parameters:
same time the relative asymptotic motion is compact:
dL0
= [L0 , M0 ]. (115)
q1 (t) − q2 (t) −→ iπ − ln 4η12 + ln cos2 (Φ± (t)) (113) dt
t→±∞
 
+O e−2κ1 |t| , After the complexification we get complex-valued L =
L0 + iL1 and M = M0 + iM1 which provide the Lax
Φ+ (t) = 2η1 t − 2φ0 − 2α1 , Φ− (t) = 2η1 t − 2φ0 . representation for the complexified system. But the
 V. S. Gerdjikov et al.: Real Hamiltonian forms of Hamiltonian systems
complexified system can be viewed as real Hamilto-
nian system with 2n degrees of freedom. It allows Lax
representation with real-valued L and M of the form
(72).
5. Solutions of the initial system that depend analytically
on the initial parameters ‘survive’ the complexification
procedure. As we see from the CTC case discussed
above, solutions that have been regular for all time af-
ter such procedure may develop singularities for finite
t. Another substantial difference is in the asymptotic
dynamical regimes.
Projecting onto the RHF some of these singularities
remain. Their asymptotic dynamical regimes need ad-
ditional studies.
10 Acknowledgements
One of us (VSG) acknowledges financial support in part
from Gruppo collegato di INFN at Salerno, Italy and proj-
ect PRIN 2000 (contract 323/2002). The work of GM and
GV was partially supported by PRIN SINTESI. One of
us (VSG) thanks Professor F. Calogero for the hospitality
at Rome University and the nice discussions concerning
this paper. One of us (AK) is grateful to the Department
of Physics and to for the hospitality of the University of
Salerno and to INFN group at Salerno for their warm
hospitality.
References
1. V.S. Gerdjikov, I.M. Uzunov, E.G. Evstatiev, G.L. Di-
ankov, Phys. Rev. E 55, 6039 (1997);
V.S. Gerdjikov, D.J. Kaup, I.M. Uzunov, E.G. Evstatiev.
Phys. Rev. Lett. 77, 3943 (1996)
2. V.S. Gerdjikov, E.G. Evstatiev, R.I. Ivanov, J. Phys. A:
Math & Gen. 31, 8221 (1998); J. Phys. A: Math & Gen.
33, 975 (2000)
3. J.M. Arnold, Phys. Rev. E 60, 979 (1999)
4. J. Grabowski, G. Marmo, P. Michor, Modern Physics Let-
ters A 14, 2109 (1999);
A. Ballesteros, O. Ragnisco, J. Phys. A 31, 3791 (1998)
5. Y. Kodama, J. Ye, Physica D 91, 321 (1996); Physica D
121, 89 (1998)
6. J. Evans, Nucl. Phys. B 390, 225 (1993);
J. Evans, J. Madsen, Phys. Lett. B 384, 131 (1996);
J. Evans, J. Madsen, Nucl. Phys. B 536, 657 (1998);
Erratum-ibid. 547, 665 (1999)
7. V.S. Gerdjikov, A. Kyuldjiev, G. Marmo, G. Vilasi, Eur.
Phys. J. B 29, 177 (2002)
8. G. Marmo, E.G. Saletan, A. Simoni, B. Vitale, Dynami-
cal systems: a differential geometric approach to symmetry
and reduction, (John Wiley & Sons, N.Y., 1985);
J. Grabowski, G. Landi, G. Marmo, G. Vilasi, Fortschr.
Phys. 42, 393 (1994)
9. J.F. Cornwell, Group Theory in Physics, Vol. 3. (Tech-
niques of Physics, 7), (Academic Press, London 1984);
S. Helgason, Differential geometry, Lie groups and sym-
metric spaces, (Academic Press, 1978)
15
10. G. Landi, G. Marmo, G. Vilasi, J. Group Theory in Phys.
2, 1 (1994)
11. M. Giordano, G. Marmo, A. Simoni, F. Ventriglia, in Non-
linear Physics: Theory and Experiment. II edited by M.
Ablowitz, M. Boiti, F. Pempinelli, B. Prinari (World Sci-
entific, 2003), p. 179.
12. G. Vilasi, Hamiltonian Dynamics, (World Scientific, Sin-
gapore, 2001)
13. G. Landi, G. Marmo, G. Vilasi, J. Math. Phys. 35, 808
(1994)
14. S. De Filippo, G. Marmo, M. Salerno, G. Vilasi, Nuovo
Cim. B83, 97 (1984)
15. A.V. Mikhailov, M.A. Olshanetzky, A.M. Perelomov,
Comm. Math. Phys. 79, 473 (1981)
16. J. Moser, in Progr. Math. 8 (Birkhäuser, Boston, 1980);
F. Calogero, Physica D 18, 280 (1986)
17. A. Perelomov, Integrable Systems of Classical Mechanics
and Lie Algebras, (Birkhaüser, Basel, 1990)
18. F. Calogero, Classical Many-Body Problems Amenable to
Exact Treatments, Lecture Notes in Physics. Monographs
66, (Springer, 2001); J. Math. Phys. 39, 5268 (1998)
19. M.O. Katanaev, T. Klosch, W. Kummer, Ann. Phys. 276,
191 (1999)
20. G. Sparano, G. Vilasi, A.M. Vinogradov, Phys. Lett. B
513, 142 (2001); Diff. Geom. Appl. 16, 95 (2002); Diff.
Geom. Appl. 17, 15 (2002)
21. G. Vilasi, P. Vitale, Classical and Quantum Gravity 19,
3333 (2002)
22. A.S. Mishenko, A.T. Fomenko, Funct. Anal. Appl. 12, n.2,
113 (1978)
23. G. Sparano, G. Vilasi, J. Geom. Phys. 36, 270 (2000)
24. S.P. Khastgir, R. Sasaki, Prog. Theor. Phys. 95 485 (1996);
Prog. Theor. Phys. 95 503 (1996)
25. A.V. Mikhailov, Physica D 3, 73 (1981)
26. M.A. Olshanetzky, A.M. Perelomov, Phys. Reports 71, 313
(1983)
27. M.A. Olshanetsky, A.M. Perelomov, A.G. Reyman, M.A.
Semenov-Tian-Shansky, Itogi Nauki i Techniki, VINITI
AN SSSR, Contemporary problems in mathematics 16, 86
(1987)
28. J. Moser, in Dynamical Systems, Theory and Applications.
Lecture Notes in Physics, 38, (Springer, 1975); Adv. Math.
16, 197 (1975)