Riemann surfaces, separation of variables and classical and quantum
integrability.
O. Babelon 1 and M. Talon1
September 2002 |
Abstract:
We show that Riemann surfaces, and separated variables immediately
provide classical Poisson commuting Hamiltonians.
We show that Baxter's equations for separated variables immediately
provide quantum commuting Hamiltonians.
The construction is simple, general, and does not rely on the
Yang–Baxter equation.
1 Introduction.
We know since Liouville that integrability means commuting Hamiltonians.
It is the primary role of Lax matrices and the Yang-Baxter equation to provide
non trivial such Hamiltonians.
In the classical theory, additional benefits are the spectral curve Γ
and the ability to separate variables by considering g = genus (Γ) points
on it [1].
In the quantum theory, the analog construction is Sklyanin's method of separation of variables
and Baxter's equations [2, 3]. Despite the beauty of this result,
the route from a Yang–Baxter defined quantum integrable model
to the separated variables is usually long and difficult, especially
in the non hyperelliptic case.
Here, we show that we can reverse the strategy. We start from separated variables and
consider Baxter's equations as equations for the Hamiltonians.
We then prove that these Hamiltonians commute under very general
hypothesis.
By its generality, its simplicity and its close analogy to the classical case,
this result could provide a good starting
point to build a theory of quantum integrable systems.
2 The main theorem.
Consider a curve in C2
Γ (λ,μ) ≡ R0(λ,μ) + |
|
Rj(λ,μ) Hj =0
(1) |
where the Hi are the only dynamical moduli, so that R0(λ,μ) and Ri(λ,μ) do not contain any dynamical variables.
If things are set up so that Γ is of genus g and there are exactly g Hamiltonian
Hj (see below for realizations of this setup), then the curve is completely determined
by requiring that it passes through g points (λi, μi), i=1,⋯, g. Indeed, the moduli Hj are determined by
solving the linear system
|
|
Rj(λi,μi) Hj + R0(λi,μi) = 0, i=1,⋯ ,g
(2) |
whose solution is
where
H = |
⎛
⎜
⎜
⎜
⎜
⎝ |
|
⎞
⎟
⎟
⎟
⎟
⎠ |
,
B = |
⎛
⎜
⎜
⎜
⎜
⎝ |
R1(λ1,μ1) |
⋯ |
Rg(λ1,μ1) |
⋮ |
|
⋮ |
R1(λi,μi) |
⋯ |
Rg(λi,μi) |
⋮ |
|
⋮ |
R1(λg,μg) |
⋯ |
Rg(λg,μg) |
|
⎞
⎟
⎟
⎟
⎟
⎠ |
,
V = |
⎛
⎜
⎜
⎜
⎜
⎝ |
R0(λ1,μ1) |
⋮ |
R0(λi,μi) |
⋮ |
R0(λg,μg) |
|
⎞
⎟
⎟
⎟
⎟
⎠ |
Here, of course, we assume that generically detB ≠ 0.
Theorem 1
Suppose that the variables (λi, μi) are separated i.e. they Poisson commute for
i ≠ j:
{ λi,λj} = 0, { μi,μj}=0,
{ λi,μj} = p(λi, μi) δij
(4)
Then the Hamiltonians Hi, i=1⋯ g, defined by eq.(3) Poisson commute
{ Hi , Hj } = 0
Proof. Let us compute
B1 B2 { (B−1V)1,(B−1V)2 } |
= |
{ B1, B2 } (B−1V)1 (B−1V)2 |
|
|
− { B1, V2 } (B−1V)1
− { V1, B2 } (B−1V)2
+ { V1, V2 } |
Taking the matrix element i,j of this expression, we get
( B1 B2 { (B−1V)1,(B−1V)2 })ij |
= |
δij |
|
{ Bik, Bil } (B−1V)k (B−1V)l |
|
|
|
− δij |
|
{ Bik, Vi } (B−1V)k
− δij |
|
{ Vi, Bil } (B−1V)l
+ δij{ Vi, Vi } =0 |
|
where δij occurs because the variables are separated.
It can hardly be simpler. The only thing we use is that the Poisson bracket
vanishes between different lines of the matrices, and then the antisymmetry.
We did not even need to specify the Poisson bracket between λi and
μi. The Hamiltonian are in involution whatever this Poisson bracket is.
This is the root of the multihamiltonian structure of integrable systems.
Can we make it quantum ? Let us consider a set of separated variables
[ λi,λj ] = 0, [ μi,μj] =0,
[ λi,μj] = p(λi, μi) δij
We want Baxter's equation, so we start
from the linear system
|
|
Rj(λi,μi) Hj + R0(λi,μi) = 0
(5) |
Here the Hj are on the right, and in Rj(λi,μi),
R0(λi,μi), we assume some order between λi,μi,
but the coefficients in these functions are non dynamical. Hence we start from the linear system
We notice that we can define unambiguously the left inverse of B.
First, the determinant D of B is well defined because it never
involves a product of elements on the same line. The same is true
for the cofactor Δij of the element Bij (we include the sign (−1)i+j in the definition of Δij).
Define
Bij−1 ≡ (B−1)ij = D−1 Δji
We have
But Δki does not contain any element Bkl, hence
the product Δki Bkj is commutative, and the usual
construction of the inverse of B is still valid. Since the left and right
inverse coincide in an associative algebra with unit, we have the identities
(BB−1)ij= |
|
Bik Bkj−1
= |
|
Bik D−1 Δjk = δij
(7) |
We write the solution of eq.(6) as
Theorem 2
The quantities Hi defined by eq.(8), which
solve Baxter's equations eqs.(5),
are all commuting
[ Hi , Hj ] = 0
Proof. Using that Vk and Vl commute, [ Vk,Vl ] = 0, we compute
|
[ Hi, Hj ] |
= |
|
(9) |
|
= |
|
[Bik−1, Bjl−1 ] Vk Vl
− Bik−1 [Bjl−1, Vk ] Vl
+ Bjl−1 [Bik−1, Vl ] Vk
|
|
|
|
Using
[A−1, B−1 ] =A−1 B−1[A,B]B−1A−1=
B−1 A−1[A,B]A−1B−1
so that
[ Bik−1, Bjl−1 ] |
= |
|
Bir−1Bjr'−1
[Brs, Br's'] Bs'l−1Bsk−1 |
|
|
= |
|
Bjr'−1 Bir−1
[Brs, Br's']Bsk−1 Bs'l−1 |
|
the first term can be written
|
= |
Σ |
|
Bir−1Bjr'−1 [Brs, Br's']
( Bs'l−1 Bsk−1 + Bs'k−1 Bsl−1 ) VkVl |
|
|
= |
Σ |
|
Bjr'−1 Bir−1 [Brs, Br's']
( Bsk−1Bs'l−1 + Bsl−1 Bs'k−1 ) VkVl |
|
Using that [Brs, Br's'] = δrr'[Brs, Brs'] and is therefore
antisymmetric in ss', and setting
Kss' = |
|
(
Bs'l−1 Bsk−1 + Bs'k−1 Bsl−1
−Bsl−1 Bs'k−1 − Bsk−1 Bs'l−1
) VkVl
|
we get
|
= |
|
|
|
Bir−1Bjr−1 [Brs, Brs'] Kss' |
|
|
= |
− |
|
|
|
Bjr−1 Bir−1 [Brs, Brs']Kss' |
|
|
= |
|
|
|
[Bir−1, Bjr−1] [Brs, Brs']Kss' |
|
The last two terms in eq.(9) are simpler, we get
|
Bjl−1 [Bik−1, Vl ] Vk
− Bik−1 [Bjl−1, Vk ] Vl =
|
|
[Bir−1, Bjr−1] [Brs, Vr] Bsk−1 Vk |
|
The quantities Hi will commute if
[
Bir−1,
Bjr−1] = 0, ∀
i,
j,
r
(10)
This is true as shown in the next Lemma.
The condition eq.(10) says that the elements on the same column of
B−1 commute among
themselves. In a sense this is a condition dual to the one on B. It is true
semiclassically because
{ Bir−1, Bjr−1 } = |
|
Bia−1 Bja'−1
{ Bab, Ba'b' } Bbr−1 Bb'r−1 =
|
|
Bia−1 Bja−1
{ Bab, Bab' } Bbr−1 Bb'r−1=0
|
where in the last step we use the antisymmetry of the Poisson bracket.
We show that it is also true quantum mechanically.
Lemma 1
Let B be a matrix whose elements commute if they do not belong
to the same line
[ Bik , Bjl ] = 0 if i ≠ j
Then the inverse B−1 of B is defined without ambiguity and
moreover elements on a same column of B−1 commute
[ Bir−1 , Bjr−1 ] = 0
Proof. We want to show that
ΔriBjr−1 = ΔrjBir−1
denote by βi(r) the vector with components Bki, k ≠ r. Then
we have (with j>i) (blue symbols are omitted):
ΔriBjr−1 |
= |
(−1)r+i
β1(r) ∧ β2(r) ∧ ⋯ βi(r)
∧ ⋯ βj(r) ∧ ⋯βg(r)
Bjr−1 |
|
= |
(−1)r+i+g−jβ1(r) ∧ β2(r) ∧ ⋯ βi(r)
∧ ⋯ βj(r) ∧ ⋯βg(r) ∧
βj(r) Bjr−1 |
|
= |
(−1)r+i+g−j+1β1(r) ∧ β2(r) ∧ ⋯ βi(r)
∧ ⋯ βj(r) |
∧ ⋯βg(r) ∧
|
|
βk(r) Bkr−1 |
|
|
= |
(−1)r+i+g−j+1β1(r) ∧ β2(r) ∧ ⋯ βi(r)
∧ ⋯ βj(r) ∧ ⋯βg(r) ∧
βi(r) Bir−1 |
|
= |
(−1)r+jβ1(r) ∧ β2(r) ∧ ⋯ βi(r)
∧ ⋯ βj(r) ∧ ⋯βg(r)
Bir−1 |
|
= |
ΔrjBir−1 |
In the above manipulations, we never have two operators Bij on the same line
so we can use the usual properties of the wedge product. Moreover it is important that
the line r is absent in the definition of β(r).
Remark that this equation can also be written
ΔriD−1 Δrj= ΔrjD−1 Δri which is a Yang–Baxter type equation.
With this Lemma, we have completed the proof of our theorem.
It is remarkable that, again, only the separated nature of the variables λi, μi
is used in this construction, but the precise commutation relations between
λi, μi does not even need to be specified. This is the origin of
the multi Hamiltonian structure of integrable systems, here extended to the quantum
domain.
3 Choosing the right number of dynamical moduli.
Let us explain how one can set up things in order that
the number of dynamical moduli is equal to the genus of the Riemann surface.
To understand the origin of the conditions we will write,
let us explain first what happens in the setting of general rational Lax matrices
described in [4, 9].
Quite generally, a Lax matrix L(λ) depending
rationally on a spectral parameter λ, with poles at points λ k
can be written as
where L0=Diag(a1,⋯,aN) is a constant diagonal matrix and
Lk(λ) is
the polar part of L(λ ) at λ k, ie. Lk(λ)=Σr=−nk−1
Lk,r (λ −λ k)r.
In order to have a good phase space to work with, we assume that Lk(λ)
lives in a coadjoint orbit of the group of N× N matrix regular in the vicinity of
λ =λ k, i.e.
Lk = (gk Ak gk−1)−
Here Ak(λ ) is a diagonal matrix with a pole of order
nk at λ =λ k, and gk has a regular expansion at λ =λ k. The notation ()− means
taking the singular part at λ =λ k. This singular part only
depends on the singular part (Ak)− and the first nk
coefficients of the expansion of gk in powers of (λ −λ k).
The matrix (Ak)− is an orbit invariant which specifies the coadjoint orbit, and is
not a dynamical variable. It is in the center of the Kirillov bracket which
as shown in [4] induces the Poisson bracket
eq.(4), with p(λi,μi)=1, on the separated variables.
The physical degrees of freedom are contained in the first nk
coefficients of gk(λ ). Note however that since Ak commutes
with diagonal
matrices one has to take the quotient by gk→ gk dk
where dk(λ ) is a regular diagonal matrix, in order to
correctly describe the dynamical variables on the orbit.
The dimension of the orbit of Lk is thus N(N−1)nk so that
L(λ ) depends on Σk N(N−1)nk degrees of freedom.
Finally, the form and analyticity properties of
L(λ ) are invariant under conjugation by constant matrices. To
preserve the normalization, L0, at ∞ these matrices have to be diagonal
(if all the ai's are different). Generically, these
transformations reduce the dimension of the
phase space by 2(N−1), yielding:
The spectral curve is
|
Γ : R(λ, μ ) ≡ det( L(λ )−μ 1 |
)
=(−μ)N + |
|
rq(λ) μq =0
|
|
|
|
(12) |
|
The coefficients rq(λ) are polynomials in the matrix elements
of L(λ ) and therefore have poles at λ k. The curve is naturally presented as a N–sheeted covering of the λ-plane. We call
μj(λ) the N branches over λ. Using the Riemann–Hurwitz
formula, we can compute the genus of Γ [4]:
It is important to observe that the
genus is half the dimension of phase space.
So the number of action variables occurring as
independent parameters in the eq.(12) should also
be equal to g. Let us verify it.
Since rj(λ) is the symmetrical function
σj(μ1,⋯,μN),
it is a rational function of λ .
It has a pole of order jnk at
λ =λ k.
Its value at λ = ∞ is known since
μj(λ)→ aj. Hence it can be expressed on jΣk nk
parameters
namely the coefficients of all these poles. Altogether we have
1/2 N(N+1) Σk nk parameters. They are not all independent
however. Above λ =λ k the various branches can be written:
where all the coefficients c1(j),⋯,cnk(j)
are fixed and non–dynamical because they are the matrix elements of
the diagonal matrices (Ak)−, while the regular part is
dynamical. This implies on rj(λ) that the coefficients of its nk highest
order pole terms are fixed. Summing over j, we get
Nnk constraints and we are left with 1/2 N(N−1) Σk nk
parameters, that is g+N−1 parameters.
It remains to take the quotient by the action of constant diagonal matrices.
The generators of this action are the Hamiltonians
Hn=(1/n) resλ =∞Tr
(Ln(λ )) dλ ,
i.e. the term in 1/λ in Tr (Ln(λ )).
Setting
around the point Qj = (∞,aj), we have Hn = Σj ajn−1 bj .
After Hamiltonian reduction these quantities are to be set to fixed
(non–dynamical) values. So, both ai (by definition) and bi are non dynamical.
On the functions rj(λ ) this implies that their expansion at infinity
starts as rj(λ ) = rj(0)+ rj(−1)/λ + ⋯,
with rj(0) and rj(−1) non dynamical.
Hence when the system is properly reduced we
are left with exactly g action variables.
The constraints eqs.(13, 14) can be summarized in a very elegant
way [5, 9].
Introduce the differential δ with respect to the dynamical moduli.
Then our constraints mean that the differential
δ μ dλ is regular everywhere on the spectral curve because the coefficients of the various poles being non dynamical, they are killed by δ:
Since the space of holomorphic differentials is of dimension g, the right hand side
of the above equation is spanned by g parameters which are the g independent
action variables we were looking for. Notice that these action variables are coefficients
in the pole expansions of the functions rj(λ), and thus appear linearly
in the equation of Γ. Hence eq.(12) can be written in the form
eq.(1). Clearly, these considerations can be adapted
by considering more general conditions such as
4 Examples.
Let us show how well known models fit into our scheme. For the
hyperelliptic ones, things are so simple that we can directly check
the commutation of the Hamiltonians.
4.1 Neumann model.
The spectral curve can be written in the form [9]:
Performing the birational transformation s= μ Q(λ), we get:
which is an hyperelliptic curve of genus g=N−1. The polynomial Q(λ)
is non dynamical. We have (N−1) independent
dynamical quantities, namely the (N−1) symmetrical functions of the bi, coefficients of
P(λ).
We have
δ μ dλ = |
|
dλ =
|
|
dλ = holomorphic |
|
Asking that a curve of the form eq.(15) passes through
the g points (λi,μi) determines the polynomial P(λ).
The solution of Baxter's equations
P(λi ) = Q(λi) μi2
simplifies in this case because the matrix B depends only on the λi.
It is equivalent to Lagrange interpolation formula:
P(λ) = P(0)(λ) + P(2)(λ)
with
P(0)(λ) = |
|
(λ − λi),
P(2)(λ) = |
|
Sj (λ) Q(λj)μj2,
Sj(λ) = |
|
Introducing the canonical commutation relations
[ μj , λk ] = i ℏ δjk
so that
[ μj, f(λj) ] = iℏ ∂λj f(λj),
[ μj2, f(λj) ] = 2 iℏ ∂λj f(λj) μj
+ (iℏ)2 ∂λj2 f(λj) |
We can check that [ P(λ), P(λ') ] =0 is a consequence of
∂λjP(0)(λ) = − Πk≠ j ( λj − λk) Sj(λ), and the identities
Sj(λ) ∂λjnSi(λ') − Sj(λ') ∂λjnSi(λ) =0, ∀ n >0
These identities follow from the remark that, if we define the translation operators
tj λi = λj + σ δij, then
Sj(λ) tj Si(λ') − Sj(λ') tj Si(λ) =
|
|
(λi − λj) (λ − λ')
(17) |
is independent of σ.
4.2 Toda Chain.
The spectral curve can be written in the form [8, 9]:
where 2P(λ)=λn+1 − Σi=1n+1 pi λn+⋯
is a polynomial of degree
(n+1). The spectral curve is hyperelliptic since it can
be written as
|
s2 = P2(λ) − 1, with s = μ −P(λ)
|
|
|
(19) |
|
The polynomial P2(λ)
is of degree 2(n+1) so the genus of the curve is g = n. The number
of dynamical moduli is g=n in the center of mass frame Σi=1n+1 pi =0.
We have
|
dλ = |
|
dλ =
|
|
dλ = holomorphic |
|
Asking that the curve eq.(18) passes through the n points (λi,μi),
we get Baxter's equations.
2 P(λi) = μi + μi−1
Their solution is again given by
Lagrange interpolation formula:
2P(λ) = P(0)(λ) + P(1)(λ)
where
P(0)(λ) = (λ + |
|
λi) |
|
(λ − λi),
P(1)(λ) = |
|
Si(λ) (μi + μi−1)
|
The polynomial P(0)(λ) is of degree n+1, vanishes for λ = λi and has no λn term.
Let the commutation relations of the separated variables be given by:
μj λj = q λj μj , μj λi = λi μj,
i ≠ j
Then again [ P(λ), P(λ') ] =0 as a result of eq.(17),
where tj is interpreted as tj λj = q λj, and the
facts that
Sj(λ) tj± 1 P(0)(λ')
− Sj(λ') tj± 1 P(0)(λ) |
= |
P(0)(λ') Sj(λ) − P(0)(λ) Sj(λ') |
|
|
|
4.3 A non–hyperelliptic model.
We consider the model studied in [2, 6, 7] .
The spectral curve can be written in the form:
R(λ,μ) ≡ μ
N +
t(1)(λ) μ
N−1 + ⋯
t(N)(λ) = 0
(20)
The polynomials t(k)(λ) are such that
degree t(k)(λ) ≤ kn −1 and
degree t(N)(λ) = Nn −1
for some integer n. The genus of this curve is
Assuming that there is no singular point at finite distance, the homomorphic differentials are
ωkl = |
|
dλ,
0 ≤ l < N−1, 0 ≤ k < (N−l−1)n −1
|
We have
This will be holomorphic if δ t(1)(λ) = 0 and
δ t(k)(λ) =δ H1(k) λ + ⋯ + δ H(k−1)n −1(k)λ(k−1)n −1
Baxter's equations and the commutation of the Hamiltonians
where proved in this case, starting from the definition of the quantum model
through it Lax matrix and the Yang–Baxter equation. Our approach gives a very simple
proof of this result.
5 Conclusion.
We have shown that starting from the separated
variables, one can give an easy definition of a quantum integrable system.
The next step is to reconstruct the Lax matrix and the original dynamical
variables of the model. While this is a well understood problem in the classical theory
[1, 9],
(it is the essence of the classical inverse scattering method), its quantum
counterpart will require a deeper understanding of the quantum affine Jacobian [10].
Acknowledgements. We thank D. Bernard and F. Smirnov for discussions.
References
- [1]
- B.A. Dubrovin, I.M. Krichever, S.P. Novikov,
Integrable Systems I. Encyclopedia of Mathematical Sciences,
Dynamical systems IV. Springer (1990) p.173–281.
- [2]
- E.K. Sklyanin, Separation of variables
in quantum integrable models related to the Yangian Y[sl(3)].
hep-th/9212076 (1992)
- [3]
- E.K. Sklyanin, Separation of variables.
Prog. Theor. Phys. (suppl), 185 (1995) p.35.
- [4]
- O.Babelon, M. Talon, The symplectic
structure of rational Lax pair systems.
solv-int/9812009 Physics Letters A 257 (1999) p.139-144.
- [5]
- I.M. Krichever, D.H. Phong, On the
integrable geometry of soliton equations and N=2 supersymmetric
gauge theories. J. Diff. Geom. 45 (1997) p.349–389.
- [6]
- F.A. Smirnov, Separation of variables
for quantum integrable models related to Uq(^slN).
math-ph/0109013
- [7]
- F.A. Smirnov, V. Zeitlin, Affine Jacobians of spectral Curves
and Integrable Models. math-ph/0203037
- [8]
- L. Faddeev, L. Takhtajan, Hamiltonian Methods in the Theory
of Solitons. Springer 1986.
- [9]
- O.Babelon, D. Bernard, M. Talon, Introduction to Classical Integrable Systems. Cambridge University Press (to appear).
- [10]
- F.A. Smirnov, Dual Baxter equations and quantization of Affine Jacobian.
math-ph/0001032
- 1
- L.P.T.H.E. Universités Paris VI–Paris VII (UMR 7589),
Boîte 126, Tour 16, 1er étage,
4 place Jussieu, F-75252 PARIS CEDEX 05
This document was translated from LATEX by
HEVEA.