Lectures on Integrable Models in Field/String Theory

Lecture 1
Introduction
Hamilton’s principle. Noether’s theorem. Gauge symmetry. Noether’s second theorem. Hamil-
tonian description. Faddeev-Jackiw formalism.

Lecture 2
Symplectic geometry
The Hamiltonian dynamics. Definitions and notations. Useful formulas and identities. Hamil-
tonian vector fields. Darboux theorem. Symplectic structure on T Q, T ∗ Q and on the space of
solutions. Moment map. Co-cycles of Lie algebras and central extensions.

Lecture 3
The SL(2, R) group
The sl(2, R) algebra. The Killing form. The exponential map: sl(2, R) → SL(2, R). The adjoint
representation. Coordinates on SL(2, R). Functions, vector fields, 1-forms and the metric on the
SL(2, R) group manifold.

Lecture 4
Particle dynamics on symmetric spaces
The Liouville model. The dynamics of a particle in SU (2) (classical and quantum theories).
Dynamics of a relativistic particle in SL(2, R). Particle dynamics in AdS space. The dynamics
of a massless particle.

Lecture 5
Gauging and Hamiltonian reduction
Gauging of Noether symmetries. Singular Lagrangian. First order formalism. Reductions of
differential forms. Examples: Mechanical model of QED. Gauging of the particle dynamics on
SU (2) and SL(2, R) group manifolds.

Lecture 6
The method of co-adjoint orbits
Co-adjoint representation of Lie groups. Co-adjoint orbits. Symplectic forms and Hamiltonian
vector fields on co-adjoint orbits. Geometric quantization. Choice of polarization. Irreducible
representations.

Lecture 7
Geometric quantization and coherent states
Symmetries and coherent states. Examples: Weyl group, SL(2, R) and SU (2) coherent states.
Symbol calculus. Moyal quantization. Coherent state formalism and geometric quantization.
Lecture 8
The Lagrangian formulation of SL(2, R) WZW theory
σ-models in 2-dimensions. The SL(2, R) target space. 2-forms on SL(2, R) group manifold and
the SL(2, R) WZW Lagrangian. The general solution and global symmetries. The SU (2) WZW
Lagrangian and the WZ term. Symmetries and integration of dynamical equations.

1
Lecture 9
The Symplectic structure of 2d free-field theory
Free field theory on a cylinder and a strip. Canonical form. Chiral fields and the chiral symplectic
form. The Poisson brackets algebra of chiral fields. ‘Vertex functions’ and their algebra. The
energy momentum tensor and the conformal symmetry.

Lecture 10
The Hamiltonian formulation of WZW theory
Canonical structure of WZW theory. The chiral symplectic form. The Poisson brackets algebra
of chiral WZ fields. Kac-Moody algebra. The Sugawara energy momentum tensor. SU (2) and
SL(2, R) WZW models.

Lecture 11
Gauging of WZW theory
Vector and axial gauging of SL(2, R) WZW theory. U (1) gauging and SL(2, R)/U (1) black hole
model. R1 gauging. Nilpotent gauging and Liouville theory. Hamiltonian reduction and free-field
parametrization.

Lecture 12
Canonical quantization of 2d CFT
Canonical quantization of free-field theory. 2d conformal symmetry and Virasoro algebra. Vertex
operators and their algebra. Canonical map to Liouville theory. Construction of Liouville vertex
operators and calculation of the reflection amplitude.

Lecture 13
Geometric quantization of infinite dimensional symmetries
The co-adjoint orbits of Virasoro group. Symplectic structure and Poisson brackets. Transfor-
mation to free-field variables. Coherent states of infinite dimensional translation group and 2d
conformal group. Transition amplitudes between the coherent states. Kac-Moody group.

Lecture 14
String dynamics in Minkowsky space
Dynamical equations and gauge fixing. Integration of equations of motion. Light-cone gauge
quantization. Covariant quantization and critical dimension. Polyakov method and non-critical
strings. Static gauge quantization.

Lecture 15
The AdS/CF T correspondence
String dynamics in AdS and AdS ×S spaces. Lax pair representation of the dynamical equations.
Integration by the Pohlmeyer method. Light-cone quantization of AdS5 × S 5 string dynamics.
Static gauge quantization of AdS strings. N = 4 supersymmetric Yang-Mills theory. Main ideas
of the correspondence. Integrable structures of the dual theories.

2
References
[1] K. Pohlmeyer, “Integrable Hamiltonian Systems And Interactions Through Quadratic Con-
straints,” Commun. Math. Phys. 46 (1976) 207.

[2] K. Sundermeyer, Lect. Notes Phys. 169 (1982) 1.

[3] E. Witten, “Nonabelian bosonization in two dimensions,” Commun. Math. Phys. 92, 455
(1984).

[4] A. M. Perelomov, “Generalized coherent states and their applications,” Berlin, Germany:
Springer (1986) 320 p

[5] M. B. Green, J. H. Schwarz and E. Witten, “Superstring Theory. Vol. 1: Introduction,”

Cambridge, Uk: Univ. Pr. ( 1987) 469 P. (Cambridge Monographs On Mathematical Physics)

[6] L. D. Faddeev and R. Jackiw, “Hamiltonian Reduction of Unconstrained and Constrained

Systems,” Phys. Rev. Lett. 60 (1988) 1692.

[7] N.M.J. Woodhouse, Geometric Quantisation, Clarendon, Oxford, 1992.

[8] P. Di Francesco, P. Mathieu and D. Snchal, “Conformal Field Theory, Springer-Verlag, ”

New York, 1997

[9] J. M. Maldacena, “The large N limit of superconformal field theories and supergravity,”
Adv. Theor. Math. Phys. 2 (1998) 231 [Int. J. Theor. Phys. 38 (1999) 1113] [arXiv:hep-
th/9711200].

[10] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, “Large N field theories,
string theory and gravity,” Phys. Rept. 323 (2000) 183

[11] J. Teschner, “Liouville theory revisited,” Class. Quant. Grav. 18 (2001) R153 [arXiv:hep-
th/0104158].

[12] A. Rogers, “Supermanifolds: Theory and applications,” Hackensack, USA: World Scientific
(2007) 251 p

[13] M. Grigoriev and A. A. Tseytlin, “On reduced models for superstrings on AdSn × S n ,” Int.
J. Mod. Phys. A 23 (2008) 2107

[14] N. Beisert, C. Ahn, L. F. Alday, Z. Bajnok, J. M. Drummond, L. Freyhult, N. Gromov and

R. A. Janik et al., “Review of AdS/CFT Integrability: An Overview,” Lett. Math. Phys.
99, 3 (2012)

[15] L. V. Prokhorov and S. V. Shabanov, “Hamiltonian mechanics of gauge systems,” Cambridge,

UK: Univ. Pr. (2011) 466 p

By George Jorjadze
Free University of Tbilisi and
Razmadze Mathematical Institute of Javakhishvili State University

3
Lecture 1
Introduction

Hamilton’s principle
Let us consider a D-dimensional spacetime X with coordinates xµ (µ = 0, 1, 2, ..., D − 1)
and a set of fields Φk (x) (κ = 1, 2, ..., N ) on X, i.e. x = (x0 , x1 , . . . , xD−1 ). Similarly,
Φ = (Φ1 , Φ2 , . . . , ΦN ) will denote the set of fields Φk (x) and ∂Φ the set of their first order
derivatives ∂µ Φk (x). The action of the system is then introduced by
∫
S[Φ, Ω] = dD x L(Φ(x), ∂Φ(x), x) , (1.1)
Ω

where Ω is a domain in X and L is a Lagrangian. The equations of motion

( )
∂L ∂L
− ∂µ =0 (1.2)
∂Φk ∂(∂µ Φk )

are obtained by the Hamilton’s principle, which states that the action functional (1.1) is
stationary on the solutions of the equation of motion with fixed values of the fields Φk (x)
at the boundary of Ω.
Note that point mechanics corresponds to D = 1, x0 = t and Φk (x) = q k (t). In this
case, the Euler-Lagrange equations (1.2) are equivalent to

Wkl (q, q̇, t) q̈ l = Vk (q, q̇, t) , (1.3)

with
∂ 2L ∂L ∂ 2L l
Wkl = , Vk = − q̇ . (1.4)
∂ q̇ k ∂ q̇ l ∂q k ∂ q̇ k ∂q l
The matrix Wkl is called Hessian. Equation (1.3) takes a Newtonian form

q̈ k = F k (q, q̇, t) (1.5)

for a non-degenerated Hessian and the corresponding Lagrangian is called regular.

For a regular Lagrangian the map from the velocities q̇ k to the canonical momenta
∂L
pk = (1.6)
∂ q̇ k

is invertible, q̇ k = v k (p, q, t), and the system of second order equations (1.3) is equivalent
to
∂H ∂H
q̇ k = , ṗk = − k , (1.7)
∂pk ∂q
where H is the canonical Hamiltonian
( )
∂L k
H= q̇ − L |q̇k =vk (p,q,t) . (1.8)
∂ q̇ k

1
 The solutions of the Hamilton equations (1.7) are stationary points of the action
∫ tf
S= dt[pk q̇ k − H(p, q, t)] . (1.9)
ti

The coordinates and momenta (q k , pk ) are canonical variables on the phase space and
an observable A is a function of the canonical variables and time A = A(p, q, t).
The Poisson brackets
∂A ∂B ∂A ∂B
{A, B} = k
− k (1.10)
∂pk ∂q ∂q ∂pk
define the Lie algebra of observables and allow to write the Hamilton equations (1.7) in a
symmetric form
q̇ k = {H, q k } , ṗk = {H, pk } . (1.11)
The dynamical equation of an observable A(p, q, t) then becomes

Ȧ = ∂t A + {H, A} . (1.12)

The canonical momenta in field theory are introduced similarly to (1.6)

∂L
Πk (t, ⃗x) = , (1.13)
∂ Φ̇k
where the following notations are used t = x0 , ⃗x = (x1 , . . . , xD−1 ) and Φ̇k = ∂0 Φk . The
Euler-lagrange equations (1.2) are usually equivalent to the Hamilton equations
δH δH
Φ̇k (⃗x) = , Π̇k (⃗x) = − , (1.14)
δΠk (⃗x) δΦk (⃗x)
with ∫ ( )
H= d⃗x Πk (⃗x)Φ̇k (⃗x) − L(⃗x) . (1.15)

Here, the integration is performed over (D − 1) spatial coordinates, at fixed time t, and
certain boundary conditions for the functions Πk (⃗x) and Φk (⃗x) are assumed. Boundary
conditions should provide consistency of the Hamilton equations (1.14) and their equiv-
alence to the Euler-Lagrange equations (1.2). Some standard boundary conditions are
discussed below for a 2-dimensional filed theory. ( )
Observables are functionals of the canonical variables Πk (⃗x), Φk (⃗x) and the Poisson
brackets are defined by
∫ ( )
δA δB δA δB
{A, B} = d⃗x − . (1.16)
δΠk (⃗x) δΦk (⃗x) δΦk (⃗x) δΠk (⃗x)

This gives to the Hamilton equations the same form as in (1.11)

Φ̇k (⃗x) = {H, Φk (⃗x)} , Π̇k = {H, Πk (⃗x)} , (1.17)

only now the canonical variables are labeled by the index k and the spatial coordinate ⃗x.

2
 As an example we consider 2-dimensional (2d) field theory with the Lagrangian
1
L = − ∂µ Φ(x)∂ µ Φ(x) , (1.18)
2
where ∂ µ = η µν ∂ν and η µν = diag(−1, 1) is the metric tensor of 2d Minkowski space.
From (1.2) we obtain the wave equation (∂02 − ∂12 )Φ = 0, which we write in the form

Φ̈(τ, σ) − Φ′′ (τ, σ) = 0 , (1.19)

where dot and prime denote time (τ = x0 ) and space (σ = x1 ) derivatives, respectively.
By (1.13) and (1.18), we get Π(σ) = Φ̇(σ) and the Hamiltonian (1.15) becomes
∫
1 [ ]
H= dσ Π2 (σ) + Φ′ 2 (σ) , (1.20)
2
with some integration domain in σ. The case −∞ < σ < ∞ corresponds to 2d Minkowski
space. Two other standard cases are the field theory on the cylinder 0 ≤ σ ≤ 2π and on
the strip 0 ≤ σ ≤ π. For these cases we consider the following boundary conditions:

1. − ∞ < σ < ∞ , Φ′ (σ) → 0 , Π(σ) → 0 , s → ±∞ ; (1.21)

2. 0 ≤ σ ≤ 2π , Φ(0) = Φ(2π), Π(0) = Π(2π) ; (1.22)
3. 0≤σ≤π , Φ (0) = Φ (π) = 0 , Π′ (0) = Π′ (π) = 0 .
′ ′
(1.23)

The case (1.21) is called the open boundary conditions and it usually assumes vanishing
of fields and their derivatives at the spatial infinity. The second case (1.22) corresponds to
the periodic boundary conditions and the third case to the Neuman boundary conditions.
These tree cases lead to the Hamilton equations

Φ̇(σ) = Π(σ) , Π̇(σ) = Φ′′ (σ) , (1.24)

which are equivalent to (1.19). As it was mentioned above, boundary conditions in field
theory have to provide consistency of the Hamilton equations.
Noether’s theorems
Noether’s theorem relates symmetry transformations to conservation laws.
We consider so called point transformations of coordinates and fields

xµ 7→ x̃µ = y µ (x) , Φk (x) 7→ Φ̃k (x̃) = Ψk (Φ, x) . (1.25)

Here, each y µ is a function of D coordinates xν , each Ψk is a function of N fields Φl and

D coordinates xν . The map (1.25) is assumed invertible.
The Lagrangian for the fields Φ̃k (x̃) is obtained by the condition S̃[Φ̃, Ω̃] = S[Φ, Ω],
which provides
∂x
L̃(Φ̃(x̃), ∂˜Φ̃(x̃), x̃) = L(Φ(x), ∂Φ(x), x) , (1.26)
∂ x̃
( ∂xµ )
where ∂x
∂ x̃
= det ∂ x̃ν
is the Jacobian of the coordinate transformation.

3
 Introducing the Euler derivative
( )
∂L ∂L
Ek = − ∂µ , (1.27)
∂Φk ∂(∂µ Φk )

from (1.26) be obtain

∂x ∂Φl
Ẽk (x̃) = El (x) . (1.28)
∂ x̃ ∂ Φ̃k
Thus, the fields Φ̃(x̃) also satisfy the Euler-Lagrange equations with the Lagrangian (1.26).
However, it does not mean that the Euler-Lagrange equations are form-invariant.
The transformation (1.25) is called a symmetry transformation if the new Lagrangian
L̃ has the same functional form as the initial one, i.e.
∂x
L(Φ̃(x̃), ∂˜Φ̃(x̃), x̃) = L(Φ(x), ∂Φ(x), x) . (1.29)
∂ x̃
In this case the form of the equations of motion for the fields Φ̃k (x̃) and Φk (x) is the same.
Now we consider a special type of symmetry transformations related to a Lie group G.
Let G be a group with M real parameters ϵα (α = 1, 2, . . . , M ) which acts both on the
spacetime X and on the space of fields Φ. Assuming that ϵ = 0 corresponds to the unit
element of the group, we can write the infinitesimal action of the group in the form

xµ 7→ x̃µ = xµ + ϵα vαµ (x) + O(ϵ2 ) , Φk 7→ Φ̃k = Φk + ϵα Vαk (Φ) + O(ϵ2 ) . (1.30)

Note that the operators

v̂a = vαµ ∂µ , V̂α = Vαk ∂k , (1.31)
satisfy the commutation relations

[v̂α , v̂β ] = Cαβ γ v̂γ , [V̂α , V̂β ] = Cαβ γ V̂γ , (1.32)

where Cαβ γ are the structure constants of G.

If (1.30) is a symmetry transformation then the first Noether’s theorem states that

∂µ Jαµ + Eκ Vαk (Φ) = 0 , (1.33)

where Eκ are Euler derivatives (1.27) and

∂L ( k )
Jαµ = vaµ L + Vα − v ν
α ∂ ν Φk
. (1.34)
∂(∂µ Φk )

Hence,
∂µ Jαµ = 0 , (1.35)
due to the Euler-Lagrange equations, and one finds the conserved observables
∫
Qα = d⃗x Jα0 (t, ⃗x) . (1.36)

4
These Qα are called Noether charges and Jαµ (x) conserved currents.
A symmetric action, that leads to conservation laws, can be constructed by some in-
variants. However, such constructions usually involve non physical degrees of freedom and
create gauge symmetries. Note that in gauge transformations group parameters are arbi-
trary functions of spacetime coordinates. For illustration we consider two simple examples.
1. A relativistic particle in Minkowski space is described by the action
∫ √
S = −m dτ −q̇ 2 (1.37)

where q̇ 2 = ηµν q̇ µ q̇ n and ηµν = diag(−1, 1, 1, 1) is the metric tensor of Minkowski space.
This action is proportional to the length of the particle worldline and therefor it is invariant
under the Poincare transformations

q µ 7→ q̃ µ = Λµ ν q ν + aµ , (1.38)

which define the isometry group of the Minkowski space. However, at the same time, the
length is invariant under the reparametrizations t 7→ f (t), with arbitrary monotonic f (t).
These are the gauge transformations mentioned above.
The Noether charges for the symmetry transformations (1.38) read
mq̇ µ
Pµ = √ , Mµν = Pµ qν − Pν qµ , (1.39)
−q̇ 2
and their conservation is provided by the equations of motion obtain from (1.37)
[ ]
m q̇µ (q̇ q̈)
√ q̈µ − =0. (1.40)
−q̇ 1/2 q̇ 2

2. The action for the electro-magnetic field is given by

∫
1
S=− d4 xFµν F µν (1.41)
4
where Fµν = ∂µ Aν − ∂ν Aµ . This action leads to the Poincare covariant field equations

∂µ F µν = 0 , (1.42)

and at the same time it is invariant under the gauge transformations

Aµ (x) 7→ Aµ (x) + ∂µ λ(x) . (1.43)

A similar situation is in Yang-Mills theory, in the standard model and in string theory.
Note that general relativity is also a gauge invariant theory.
An important structure of gauge theories is provided by the second Noether’s theorem.
It states that the Euler derivatives of a gauge theory are linearly related to each other.
One usually considers two different type of gauge transformations. The first corresponds
to spacetime diffeomorphisms and the second to gauge transformations only of fields, as it
was illustrated in the above mentioned two examples. We present these cases separately.

5
 First we consider gauge transformations of coordinates and the corresponding transfor-
mations of fields written in the following infinitesimal form

x̃µ = xµ + ϵµ (x) + O(ϵ2 ) , Φ̃k (x̃) = Φk (x) + ∂ν ϵµ (x) Vµk, ν (Φ) + O(ϵ2 ) . (1.44)

If the action of the system is invariant under these gauge transformations, then the linear
relations between the Euler derivatives take the form
( )
∂µ Φk Ek − ∂ν Vµk,ν Ek = 0 . (1.45)

In the second type gauge symmetries one transforms only fields

Φk (x) 7→ Φ̃k (x) = Φk (x) + ϵα (x) Vακ (Φ) + ∂ν ϵα (x) Vαk, ν (Φ) + O(ϵ2 ) . (1.46)

and the linear relations between the Euler derivatives become

( )
Vαk Ek − ∂ν Vαk,ν Ek = 0 . (1.47)

Note that the number of linear relations in both cases coincide with the number of
gauge group parameters and in the first case it is given by the spacetime dimension.
Linear relations between the Euler derivatives implies a degeneracy of the Hessian.
The Lagrangian of a gauge theory, therefore, is singular and the standard Hamiltonian
description fails.
Hamiltonian treatment of singular systems was introduced by Dirac and later on there
were many attempts to develop the method due to its importance for the fundamental
theories. Here we present the Faddeev-Jackiw formalism which appeared quite effective for
the description of gauge theories and other constraint systems as well.
Faddeev-Jackiw formalism
Let us consider a mechanical system with coordinates q = (q 1 , q 2 , . . . , q N ) and the action1
∫
S[q] = dt L(q, q̇) . (1.48)

One can introduce 2N additional coordinates p = (p1 , p2 , . . . , pn ), v = (v 1 , v 2 , . . . v n ) and

the new action ∫
[ ]
S̃[q, p, v] = dt pk (q̇ k − v k ) + L(q, v) , (1.49)

where L(q, v) is the same function as in (1.48). The variation of (1.49) with respect to pk
provides q̇ k = v k and their insertion in (1.49) leads to the initial action (1.48).
Thus, these two actions provide the same dynamics for the coordinates q k .
One can also consider the variation of (1.49) with respect to v k , which yields
∂L(q, v)
pk = . (1.50)
∂v k
1
Since the equations of motion and other local properties of a system are independent on boundary
conditions, the integration domain sometimes is not important and it is not indicated in the action.

6
For a regular Lagrangian these equations allow to express the variables v k in terms of (p, q)
and one ends up with the Hamiltonian system (1.9).
However, if the Lagrangian is singular, one can solve only a part of the variables v k .
Suppose, these are (v 1 , v 2 , . . . v N −M ). Without loss of generality one can assume that after
solving the first N − M variables (v 1 , v 2 , . . . v N −M ) the dependance on the rest variables
v a (a = N − M + 1, . . . , N ) is linear and the action (1.49) takes the following form
∫
[ ]
S̃ = dt pk q̇ k − H(p, q) − v a ϕa (p, q) . (1.51)

Then varying v a lead to the constraints

ϕa (p, q) = 0 . (1.52)
These constraints define (2N − M )-dimensional surface in (p, q) space. The point on the
surface can be parameterized by 2N − M coordinates ξ α (α = 1, 2, . . . , 2N − M )
q k = q k (ξ) , pk = pk (ξ) , (1.53)
and the action (1.51) in the new coordinates becomes
∫ [ ]
˙
S̃[ξ] = dt θα (ξ)ξ − H(ξ) ,
α
(1.54)

where θa (ξ) = pk (ξ)∂a q k (x). For further analysis one has to calculate the rank r of the
antisymmetric matrix
ωαβ (ξ) = ∂α θβ (ξ) − ∂β θα (ξ) , (1.55)
which is even r = 2n. According to the Darboux’s theorem there exists a transformation
to new variables (Pi , Qi , η γ ), where i = 1, 2, . . . , n and γ = 1, 2, . . . , 2N − M − 2n, such
that
θα dξ α = Pi dQi + dF (P, Q, η) . (1.56)
Neglecting the derivative term dF (P, Q, η), one gets from (1.54)
∫ [ ]
S̃ = dt Pi Q̇ − H(P, Q, η) .
i
(1.57)

Varying now η γ , one finds algebraic equations for η’s

∂γ H(P, Q, η) = 0 . (1.58)
Either all η’s can be excluded from these equations or one gets new constraints on (P, Q)
and the reduction procedure has to be repeated again. In this way one finally obtains an
ordinary Hamiltonian system (1.9).
This reduction procedure for a gauge invariant theory usually stops at the level (1.57)
and at the level (1.51) the functions ϕa (p, q) and H provide the Poisson brackets algebra
{ϕa , ϕb } = Cab c ϕc , {H, ϕa } = 0 , (1.59)
where Cab c are the structure constants of the gauge group.

7
Exercises
1. Derive the Euler-Lagrange equations (1.2) from the Hamilton’s principle.
2. Check the equivalence of the equations of motion (1.3) and (1.7).
3. Derive the Hamilton’s equation (1.7) by the variation of (1.9) and specify the boundary
conditions.
4. Check the canonical Poisson brackets

{Πk (⃗x), Πl (⃗y )} = 0 = {Φk (⃗x), Φl (⃗y )} , {Πk (⃗x), Φl (⃗y )} = δkl δ(⃗x − ⃗y ) . (1.60)

5. Derive the Hamilton equations (1.24), using the Hamiltonian (1.20) and the boundary
conditions (1.21)-(1.23).
6. Check the relation (1.28) for the Euler derivatives.
7. Derive the commutation relations (1.28).
8. Let gϵ be a one parameter group which acts on the time coordinate t by the rule

gϵ2 (gϵ1 (t)) = gϵ1 +ϵ2 (t) , (1.61)

and its infinitesimal form is given by

gϵ (t) = t + ϵ v(t) + O(ϵ2 ) . (1.62)

Find the global form of gϵ (t) in terms of the function v(t).

9. Let us consider a set of transformations (1.62) labeled by integer n

gϵn (t) = t + ϵn vn (t) + O(ϵ2 ) , (1.63)

with vn (t) = tn+1 . Show that the vector fields v̂n form a Lie algebra and calculate its struc-
ture constants. Show that v−1 , v0 and v1 form a subalgebra and calculate the corresponding
global transformations of t.
10. Let us consider a mechanical system with Noether charges
∂L ( k )
Qα = vα (t)L(q, q̇, t) + k
Vα (q) − vα q̇ k . (1.64)
∂ q̇
Check whether the Poisson brackets of Qα form the algebra (1.32). Investigate the same
problem in field theory.
11. Derive the equations of motion (1.37) and check the linear relation (1.45).
12. Derive the equations of motion (1.41) and check the linear relation (1.47).
13. Check the equations (1.45) and (1.47).
14. Apply the Faddeev-Jackiw reduction to the relativistic particle action (1.37).
15. Apply the Faddeev-Jackiw reduction to the action of electro-magnetic field (1.41).

8
16. Let us consider a Lagrangian with second order derivatives
∫
S[Φ] = dT L(Φk , ∂µ Φk , ∂µν
2
Φk ) . (1.65)
Ω

Formulate the stationary action principle and derive the equations of motion
( ) ( )
∂L ∂L ∂L
− ∂µ 2
+ ∂µν =0. (1.66)
∂Φk ∂(∂µ Φk ) 2 Φk )
∂(∂µν

17. Consider a symmetry transformation for the action with second order derivatives and
construct the corresponding conserved current Jαµ .
18. Describe symmetries of the model (1.18) and find the corresponding conserved currents.

References
[1] K. Sundermeyer, “Constrained Dynamics With Applications To Yang-mills Theory,
General Relativity, Classical Spin, Dual String Model,” Lect. Notes Phys. 169 (1982).
(Chapters 1.1-1.3).

[2] L. D. Faddeev and R. Jackiw, “Hamiltonian Reduction of Unconstrained and Con-

strained Systems,” Phys. Rev. Lett. 60 (1988) 1692.

[3] R. Jackiw, “(Constrained) quantization without tears,” In *Jackiw, R.: Diverse topics
in theoretical and mathematical physics* 367-381. [hep-th/9306075].

[4] N.M.J. Woodhouse, “Geometric Quantisation,” Clarendon, Oxford, (1992).

(Chapter 1.4).

[5] L. V. Prokhorov and S. V. Shabanov, “Hamiltonian mechanics of gauge systems,”

Cambridge, UK: Univ. Pr. (2011).
(Chapters 1.1-1.6, 3.1-3.2).

9
Lecture 2
Symplectic Geometry
1. The Hamiltonian dynamics
The Hamilton equations
∂H ∂H
q̇ a = , ṗa = − (a = 1, ..., N ) (2.1)
∂pa ∂q a
can be written in the form
1 mn ∂H
η̇ m =
ω , (2.2)
2 0 ∂η n
where η n (n = 1, ..., 2N ) combines the phase-space coordinates
η 1 = p1 , ... , η N = pN , η N +1 = q 1 , ... , η 2N = q N (2.3)
and ω0mn is the 2N × 2N antisymmetric matrix
( )
mn 0 −I
ω0 = 2 . (2.4)
I 0
The coefficient 1/2 in (2.2) and the corresponding normalization of the matrix ω0 is chosen just
for further convenience. Eq. (2.2) in arbitrary coordinates η 7→ ξ becomes
1
ξ˙m = ω mn ∂n H , (2.5)
2
with
∂ξ m kl ∂ξ n ∂
ω mn = k
ω0 l
and ∂n ≡ n . (2.6)
∂η ∂η ∂ξ
Since the Jacobian of the transformation η 7→ ξ is non-zero, the matrix ω mn is non-degenerated
and its inverse ωmn provides ω ml ωln = δ m n . The matrixes ωmn and ω mn transform as covariant
and contravariant 2-tensors, respectively. They remain antisymmetric, but, in general, they are
coordinate dependent. If ω mn = ω0mn the transformation η 7→ ξ is canonical. In this case the
initial canonical form of Hamilton equations (2.1) is preserved.

Calculating the canonical Poisson bracket

∂f ∂g ∂f ∂g
{f, g} = − (a = 1, ..., N ) (2.7)
∂pa ∂q a ∂q a ∂pa
in the coordinates ξ n , one finds
1 mn
{f, g} = ω ∂n f ∂m g . (2.8)
2
Then, the Jacobi identity
{{f, g}, h} + {{g, h}, f } + {{h, f }, g} = 0 , (2.9)
implies the condition
ω kl ∂k ω mn + ω km ∂k ω nl + ω kn ∂k ω lm = 0 , (2.10)
which for the inverse matrix ωmn leads to
∂l ωmn + ∂m ωnl + ∂n ωlm = 0 . (2.11)
Exercise 1-5.

1
2. Definitions and notations from differential geometry

Before introducing the notions of symplectic geometry we recall basic definitions, notations and
some useful formulas form differential geometry (see also the appendix of the lecture 3).

• M -dimensional manifold M locally looks like a domain of RM .

• ξ n (n = 1, ..., M ) are local coordinates in a domain of M.

• Smooth functions C ∞ (M); f ∈ C ∞ (M), f : M 7→ R.

• Vector fields V (M); V ∈ V (M), V = V n (ξ) ∂n .

• 1-forms Ω1 (M); Θ ∈ Ω1 (M), Θ = Θn (ξ) dξ n .

• The action of 1-forms on vector fields Θ(V ) = Θn V n is equivalent to the rule

dxm (∂n ) = δ m n . (2.12)

• p-forms Ωp (M); Λ ∈ Ωp (M), Λ = Λkl...n dξ k ∧ dξ l ∧ · · · ∧ dξ n .

• The action of Λ on p vector fields Λ(V1 , ..., Vp ) = Λkl...n V1k V2l ...Vpn is provided by

Λ(∂k , ∂l , ..., ∂n ) = Λkl...n . (2.13)

• The anti-symmetrization of tensors is given by

1 ∑
T[kl...n] = sing(σ) Tσ(k)σ(l)...σ(n) , (2.14)
p! σ

where the summation is over the permutations σ of p-indices k, l, ..., n.

• The exterior product of Λ ∈ Ωp (M) and Θ ∈ Ωq (M); Λ ∧ Θ ∈ Ωp+q (M),

(Λ ∧ Θ)kl...n = Λ[kl... Θm...n] . (2.15)

• The differential of a p-form; dΛ ∈ Ωp+1 (M),

(dΛ)kl...n = ∂[k Λl...n] . (2.16)

• The Lie derivative of a p-form; LV (Λ) ∈ Ωp (M),

LV (Λ)kl...n = V m ∂m Λkl...n + pΛm[l...n ∂k] V m . (2.17)

• The contraction of V with Λ; V ⌋Λ ∈ Ωp−1 (M),

(V ⌋Λ)l...n = p V m Λml...n . (2.18)

• The commutator (the Lie bracket) of two vector fields V and W ; [V, W ] ∈ V (M),

[V, W ]n = V m ∂m W n − W m ∂m V n . (2.19)

2
 Useful formulas for Λ ∈ Ωp (M) , Θ ∈ Ωq (M) :

d(dΛ) = 0 , (2.20)

Λ ∧ Θ = (−1)pq Θ ∧ Λ , (2.21)

d(Λ ∧ Θ) = dΛ ∧ Θ + (−1)p Λ ∧ dΘ , (2.22)

V ⌋(Λ ∧ Θ) = (V ⌋Λ) ∧ Θ + (−1)p Λ ∧ (V ⌋Θ) , (2.23)

LV Λ = V ⌋dΛ + d (V ⌋Λ) , (2.24)

LV (dΛ) = d (LV Λ) , (2.25)

LV (W ⌋Λ) = [V, W ]⌋Λ + W ⌋LV Λ , (2.26)

(LV LW − LW LV ) Λ = L[V,W ] Λ . (2.27)

• A form Λ is called closed if dΛ = 0, and it is called exact if Λ = dΘ.

• Due to (2.20), an exact form is closed, but a closed form is not always exact.

Examples with 1- and 2-forms (λ = λn dξ n , θ = θn dξ n , ω = ωmn dξ m ∧ dξ n ) :

1
(λ ∧ θ)mn = (λm θn − λn θm ) , (2.28)
2
1
(dθ)mn = (∂m θn − ∂n θm ) , (2.29)
2

(LV θ)n = V m ∂m θn + θm ∂n V m , (2.30)

V ⌋θ = V n θn , (2.31)

1
(ω ∧ θ)mnl = (ωmn θl + ωnl θm + ωlm θn ) , (2.32)
3
1
(dω)lmn = (∂l ωmn + ∂m ωnl + ∂n ωlm ) , (2.33)
3

(LV ω)mn = V l ∂l ωmn + ωln ∂m V l − ωlm ∂n V l , (2.34)

(V ⌋ω)n = 2V m ωmn . (2.35)

Exercise 6-8.

3
3. Symplectic manifold, Hamiltonian vector fields and Darboux theorem

A 2-form ω is called non-degenerated if the equation for a vector field V

V ⌋ω = θ (2.36)

has an unique solution for any 1-form θ. Writing ω in local coordinates

ω = ωmn dξ m ∧ dξ n , (2.37)

by (2.35) one finds the equation

2 V m ωmn = θn , (2.38)
and the non-degeneracy condition is equivalent to det ωmn ̸= 0. Thus, for a non-degenerated
2-form there exists the inverse matrix ω mn with

ω ml ωln = δ m n . (2.39)

Since ωmn is anti-symmetric, it can be non-degenerated only if the dimension of M is even.

A 2-form ω is called symplectic if it is non-degenerated and closed ( dω = 0 ).
A manifold M equipped with a symplectic form is a symplectic manifold.
By (2.33), the condition dω = 0 is just eq. (2.11). To discuss other similarities with the
phase space structure, let us consider eq. (2.36) with θ = −df , where f is a function on M

df + Vf ⌋ω = 0 . (2.40)

The field Vf is called the Hamiltonian vector field. It is associated with a function f and has the
components
1
Vfn = ω nm ∂m f . (2.41)
2
Due to (2.24) and (2.40) the Hamiltonian vector fields preserve the symplectic form

LVf ω = 0 . (2.42)

Poisson bracket of two functions f and g is defined by

{ f, g } = 2ω(Vf , Vg ) . (2.43)

Using (2.41), it can be written in local coordinates as Eq. (2.8)

1 mn
{ f, g } = 2ωmn Vfm Vgn = ω ∂n f ∂m g . (2.44)
2
From (2.44) follow other, equivalent to (2.43), expressions

{ f, g } = Vf (g) = −Vg (f ) . (2.45)

The Poisson bracket (2.43) is obviously anti-symmetric. As it was mentioned in Section 1, the
Jacobi identity (2.9) is equivalent to (2.10). On the other hand, for a non-degenerated ω, equation
(2.10) follows from (2.11) (see the exercise 4). It means, that the Jacobi identity is the consequence
of the non-degeneracy and the closer of ω.

4
 Finally note that the Hamiltonian vector fields satisfy the commutation relation
[ Vf , Vg ] = V{f,g} . (2.46)
To check this, one can act by the left and right hand sides of (2.46) on a function h. With the
help of (2.45) one can verify that the obtained relation is just the Jacobi identity.
Thus, the set of smooth functions on a symplectic manifold form a Lie algebra with respect
to the Poisson brackets and the map of this algebra to the Hamiltonian vector fields is a repre-
sentation of this Lie algebra.
An important example of a symplectic manifold is a standard phase space with canonical
coordinates pa , q a (a = 1, ..., N ) and the canonical symplectic form
ω = dpa ∧ dq a . (2.47)
In this case the Hamiltonian vector fields are given by
∂H ∂ ∂H ∂
VH = a
− a , (2.48)
∂pa ∂q ∂q ∂pa
and the Poisson bracket (2.43) becomes (2.7). As we have seen in Section 1, transformations to
arbitrary phase-space coordinates reproduce the formalism of symplectic geometry. It is natural
to ask the question: whether a symplectic manifold has the canonical coordinates. The answer
is given by Darboux’s theorem:
Let (M, ω) be a 2N -dimensional symplectic manifold and let m ∈ M. Then there is a neighbor-
hood U of m and a coordinate system (pa , q a ), (a = 1, ..., N ) on U such that ω = dpa ∧ dq a .
In general, canonical (Darboux) coordinates exist only locally. It should be mentioned that
their explicit construction sometimes is not easy, even if the canonical structure is global.
In physical applications a symplectic manifold (M, ω) usually arises in a Hamiltonian reduc-
tion of a gauge theory to physical (gauge invariant) variables. The obtained symplectic manifold
is called the physical phase space.
As we have seen a function on M plays two roles. It is an observable and at the same time
it is a generator of a one parameter group of transformations. A set of functions could generate
a Lie group transformations, if their Poisson brackets form a Lie algebra.
Exercise 9-12

4. Symplectic structure on T ∗ Q, T Q and on the space of solutions

The Lagrangian L(q, v) is a function on T Q, where T Q denotes the tangent space to the configu-
ration space Q. The dynamical trajectories are solutions of the variational equation δS = 0, with
the action ∫ t1
S= dt L(q, q̇). (2.49)
t0
The corresponding Euler-Lagrange equations
( )
d ∂L ∂L
a
− a =0 (2.50)
dt ∂ q̇ ∂q
can be written in the first order form for the 2N variables (q a , v a )
( )
d ∂L ∂L
− a =0, and q̇ a = v a . (2.51)
dt ∂v a ∂q

5
The Lagrangian is called regular if
( )
∂2L
det ̸= 0 . (2.52)
∂v a ∂v b
In this case the equations
∂L
pa = (2.53)
∂v a
define velocities va as functions of the coordinates q a and the momenta pa . Eqs. (2.51), then
become equivalent to the Hamilton equations (2.1) with the Hamilton function
∂L a
H= v −L . (2.54)
∂v a
The Hamilton equations have the form
a
q̇ a = VHq , ṗa = VHpa , (2.55)

where
a ∂H ∂H
VHq = , VHpa = − (2.56)
∂pa ∂q a
are the components of the Hamiltonian vector field (2.48) for the canonical symplectic form (2.47).
Note that the Hamiltonian (2.54), the canonical symplectic form (2.47) and the canonical
1-form θ = pa dq a are invariant under the coordinate transformations on Q.
The transformations from the canonical coordinates (q a , pa ) to (q a , v a ) can be considered as
a change of coordinates on the symplectic manifold. Due to (2.53), the symplectic form in the
new coordinates becomes
∂2L ∂2L
ωL = dq a
∧ dq b
+ dv a ∧ dq b . (2.57)
∂q a ∂v b ∂v a ∂v b
Introducing the corresponding Hamiltonian vector field VH related to the Hamilton function
(2.54)
VH ⌋ωL + dH = 0 , (2.58)
and writing it as
∂ ∂
VH = q̇ a a
+ v̇ a a , (2.59)
∂q ∂v
from (2.57) one finds
∂2L a b ∂2L a b ∂2L
VH ⌋ωL = q̇ dq − q̇ dq + (v̇ a dq b − q̇ a dv b ) . (2.60)
∂q a ∂v b ∂q b ∂v a ∂v a ∂v b
Now differentiating the Hamiltonian (2.54)
∂2L a b ∂2L a b ∂L
dH = a b
v dv + a b
v dq − a dq a (2.61)
∂v ∂v ∂v ∂q ∂q
and inserting (2.60)-(2.61) in eq. (2.58), one indeed obtains the dynamical equations (2.51)
[ 2 ]
∂2L ∂ L b ∂2L ∂L
VH ⌋ωL + dH = a b (v − q̇ ) dv +
a a b
q̇ + a b v̇ − a dq a
b
(2.62)
∂v ∂v ∂v a ∂q b ∂v ∂v ∂q
∂2L
+ a b (v a − q̇ a ) dq b = 0 .
∂v ∂q

6
 The space of solutions (motions) M is the set of functions q a = q a (t), which satisfy the
dynamical equations (2.50). Note that the trajectories q a (t) and q a (t + s), in general, are distinct
points of M , even though the two trajectories occupy the same points in Q.
Symmetry properties of integrable systems is usually easier to formulate for M . This space
becomes a manifold if we parameterize the solutions q a = q a (t) by the initial data (q a (t0 ), q̇ a (t0 )).
A Hamiltonian field VH is called complete if the solutions with all admissible initial data can be
continued for arbitrary t. When VH is complete and L is regular, the map

ϕ(t0 ) : M 7→ T Q : q a (t) 7→ (q a , v a ) = (q a (t0 ), q̇ a (t0 )) (2.63)

defines a diffeomorphism and the symplectic form on T Q (given by ωL ) induces the symplectic
form on M . There is another, but equivalent, way to define the symplectic structure of M .
Let us consider the following function on M
∫ t1
S(t0 , t1 ) = dt L(q, q̇) , (2.64)
t0

where the action is calculated on the solutions q = q(t). A tangent vector U to M at a solution
q = q(t) is a function u = u(t), which satisfies the linearized equation of motion
( 2 )
d ∂ L ∂2L ∂2L ∂2L b
u̇ b
+ ub
− u̇b
− u =0. (2.65)
dt ∂v a ∂v b ∂v a ∂q b ∂q a ∂v b ∂q a ∂q b

The derivative of the function (2.64) along U is given by

∫ t1 ( )
∂L a ∂L a
U ⌋dS(t0 , t1 ) = dt u + a u̇ (2.66)
t0 ∂q a ∂v
[ ]t1 ∫ t1 [ ( )]
∂L a ∂L d ∂L
u + dt − ua .
∂v a t0 t0 ∂q a dt ∂v a

The last integral vanishes because q(t) is a solution of the equation of motion. Introducing, for
each t, a 1-form θt on M by
∂L
U ⌋θt = ua (t) a , (2.67)
∂v
we find
dS(t0 , t1 ) = θt1 − θt0 . (2.68)
Here the right-hand side of (2.67) is calculated at (q, v) = (q(t), q̇(t)).
The closed 2-form
ω = dθt (2.69)
does not depend on t and it coincides with the induced form from T Q.
Comparing the dynamical pictures on M and T Q, one finds a similarity with the quantum
case. The dynamics on M is the analog of the Heisenberg picture, while the Schrodinger picture
corresponds to the dynamics on T Q.
Exercise 13-14.

7
Appendix

Moment map

Let (M, ω) be a symplectic manifold and let G be a Lie algebra. The action of G on M is a linear
map from A 7→ VA
V[A,B] = [VA , VB ] , (2.70)

Hamiltonian action
h[A,B] = {hA , hB } . (2.71)

Moment
µ : m 7→ fm , fm (A) = hA (m) (2.72)
‘Moment’ because such a map generalizes the momentum and angular momentum associated with
translations and rotations respectively.

Co-cycles of Lie algebras

Let G be a Lie algebra G and G ∗ its dual. A cocycle on G is an anti-symmetric bilinear form
α ∈ G ∗ ∧ G ∗ , such that

α([A, B], C) + α([B, C], A) + α([C, A], B) = 0 (2.73)

for every A, B, C. Any f ∈ G ∗ defines an element δf of G ∗ ∧ G ∗

1
δf (A, B) = f ([A, B]) . (2.74)
2
Two cocycles are said to be equivalent if they differ by δf . The set of equivalence classes forms
a group under addition. It is called the second cohomology group of G and is denoted by H 2 G.

A canonical action A 7→ VA of a Lie algebra G on a symplectic manifold (M, ω) determines an

element Ω of H 2 G. If there is a moment, then Ω = 0.

If Ω = 0 and if each of the vector fields VA is Hamiltonian, then the action of G on M is

Hamiltonian.

The usefulness of this statement is that H 2 G = 0.

[G, G] = G and H 2 G = 0 for any semi-simple Lie algebra.

The Abelian Lie algebras: [A, B] = 0 for any A, B ∈ G. [G, G] = 0 and H 2 G = G ∗ ∧ G ∗ .

8
Exercises
1. Prove that if ω mn is given by (2.6), then its inverse is

∂η k 0 ∂η l
ωmn = ω , (E.1)
∂ξ m kl ∂ξ n
where ( )
0 1 0 I
ωkl = (E.2)
2 −I 0
inverts (2.4).

2. Derive eq. (2.10) from eq. (2.11).

3. Verify that for N = 1, eq. (2.11) is satisfied by any antisymmetric ω.

4. Check that ωmn given by eq. (E.1) satisfies eq. (2.11).

5. Let us consider the transformation (p, q) 7→ (ξ 1 , ξ 2 ) from the canonical coordinates (p, q) to
√ √
H H
ξ1 = p α + , ξ2 = q α + . (E.3)
2 2
Here α is a non-negative parameter and H is the harmonic oscillator hamiltonian

p2 + q 2
H= . (E.4)
2
Check that √
{ξ 1 , ξ 2 } = H + α = (ξ 1 )2 + (ξ 2 )2 + α2 , (E.5)
and that the Poisson brackets of the functions ξ 1 , ξ 2 and ξ 0 = H + α form the sl(2, R) algebra.

6. Let us consider on R2 \ (0, 0) the 1-form

x y
λ= dy − 2 dx . (E.6)
x2 +y 2 x + y2
Check that λ is closed, but not exact.

7. Derive (2.34) from (2.17).

8. Similarly to the exercise 7, derive the formula for the Lie derivative of a metric tensor.

9. A field V on a symplectic manifold (M, ω) is called locally Hamiltonian if it preserves ω

LV ω = 0 . (E.7)

Let us consider on R2 \ (0, 0) the canonical form ω = dp ∧ dq and the vector field
p q
V = ∂p + 2 ∂q . (E.8)
p2 + q 2 p + q2

Check that the field (E.8) is locally Hamiltonian, but not Hamiltonian.

9
10. Check that the commutator of two locally Hamiltonian vector fields V and W is Hamiltonian

[V, W ] = Vf , (E.9)

with
f = 2ω(V, W ) . (E.10)

11. Let us consider the 2-form

1
ω=√ dx ∧ dy (E.11)
α 2 + x2 + y 2

on R2 . Here α is a positive parameter. Find a 1-form θ, such that dθ = ω.

12. Let us consider the 2-form

ω = z dx ∧ dy + x dy ∧ dz + y dz ∧ dx (E.12)

on R3 and its reduction on the unit sphere x2 + y 2 + z 2 = 1. Calculate the reduced 2-form in
the spherical coordinates. Check that the reduced form is closed, but not exact. Find the local
canonical coordinates for the reduced 2-form.

13. The space of solutions for a free particle is

q(t) = x + vt , (E.13)

where x and v are the coordinate and the velocity at t = 0.

Check that this space is invariant under:
• Translations q(t) 7→ q(t) + a + bt.
• The conformal transformations q(t) 7→ q(ϕ(t)) ϕ̇(t)−1/2 , with

αt + β
ϕ(t) = , αδ − βγ = 1 . (E.14)
γt + δ
Find the corresponding transformations of the parameters x, v.

14. The space of solutions of the harmonic oscillator is given by

q(t) = x cos t + v sin t , (E.15)

where x and v are the coordinate and the velocity at t = 0.

Check that this space is invariant under the translations q(t) 7→ q(t) + a cos t + b sin t.

Consider the following transformation of the initial data

( ) ( )( )
x α β x
7→ , with αδ − βγ = 1 . (E.16)
v γ δ v

Check that (E.16) is a canonical transformation and find the corresponding transformation of the
space of solutions (E.15).

10
References
[1] N.M.J. Woodhouse, “Geometric Quantisation,” Clarendon, Oxford, (1992).
(Chapters 1.1-1.7, 2.1-2.4, 3.1-3.3).

[2] V.I. Arnold, “Mathematical methods of classical mechanics,” Springer, New York (1978).
(Chapters 7.32-7.34, 8.37-8.39).

[3] K. Sundermeyer, “Constrained Dynamics With Applications To Yang-mills Theory, General

Relativity, Classical Spin, Dual String Model,” Lect. Notes Phys. 169 (1982).
(Chapter 1.4).

[4] L. V. Prokhorov and S. V. Shabanov, “Hamiltonian mechanics of gauge systems,” Cambridge,

UK: Univ. Pr. (2011).
(Chapters 1.1-1.6).

11
Lecture 3

The SL(2, R) group

1. The sl(2, R) algebra

A Lie algebra G is a linear space with a multiplication rule, which is bilinear, antisymmetric and
satisfies the Jacobi identity. The product of two vectors A ∈ G and B ∈ G is called the Lie bracket
and it is denoted by [ A , B ]. Thus, [ A , B ] ∈ G and it satisfies the conditions

[λ A + B , C ] = λ [ A , C ] + [ B , C ] , (3.1)

[A, B ] = −[B , A] , (3.2)

[[A, B ], C ] + [[B , C ], A] + [[C , A], B ] = 0 . (3.3)
The numbers λ are real for real algebras and to distinguish the real and complex algebras one
uses the letters R and C, respectively.
A representation of G is a linear map of G to a space of linear operators A 7→ ÔA , such that

[ A , B ] 7→ ÔA ÔB − ÔB ÔA . (3.4)

Let us consider the operator adA acting on G by

adA (B) = [ A , B ] . (3.5)

The Jacobi identity (3.3) provides that

ad[ A , B ] = adA adB − adB adA . (3.6)

Thus, the map A 7→ adA defines a representation of G. It is called the adjoint representation.
The sl(2, R) algebra is a remarkable example, which arises in many mathematical and physical
constructions. Its elements are 2×2 real traceless matrices and the Lie bracket is the commutator
[ A , B ] = AB − BA. Since the commutator of two matrixes is traceless, [ A , B ] ∈ sl(2, R). The
conditions (3.1)-(3.3) are obviously fulfilled.
One can use the following basis in sl(2, R)
( ) ( ) ( )
0 −1 0 1 1 0
T0 = , T1 = , T2 = . (3.7)
1 0 1 0 0 −1

Any A ∈ sl(2, R) can be written as A = An Tn , with real numbers An . The basis elements Tn
(n = 0, 1, 2) satisfy the relations

Tm Tn = −ηmn I + ϵl mn Tl , (3.8)

where I is the unit matrix, ηmn = diag(+, −, −) form the metric tensor of 3d Minkowski space
and ϵmnl is antisymmetric, with ϵ012 = 1. Lover and upper indices are provided by the metric
tensor ηmn and its inverse η mn (η ml ηln = δnm ). Due to (3.8), the commutators of Tn are given by

[ Tm , Tn ] = 2ϵl mn Tl . (3.9)

The numbers 2ϵl mn are called the structure constants of the sl(2, R) algebra (see the exercise 1).
Exercises 1-6.

1
2. The Killing form
Let us introduce a bilinear and symmetric form B(A, B) on sl(2, R)

B(A, B) = ⟨ A B ⟩ , (3.10)

where the brackets ⟨ · ⟩ denote the following normalized trace of 2 × 2 matrices

1
⟨M ⟩ = − Tr (M ) . (3.11)
2
Deu to (3.8), the normalization (3.11) yields

⟨ Tm Tn ⟩ = ηmn . (3.12)

Expanding then A and B in the basis (3.7): A = Am Tm , B = B n Tn , we obtain

⟨ A B ⟩ = ηmn Am B n . (3.13)

Note that the maps An 7→ A = An Tn and A 7→ An = ⟨ A T n ⟩ are inverse to each other.

Thus, the scalar product (3.10) makes the sl(2, R) algebra isometric to 3d Minkowski space.
Similarly to the Minkowski space, a non-zero element A is called time-like if ⟨ A A ⟩ > 0,
space-like if ⟨ A A ⟩ < 0 and light-like if ⟨ A A ⟩ = 0. For example, T0 is time-like, whereas T1 and
T2 are space like. An example of a light-like element is
( )
1 0 0
T+ = (T0 + T1 ) = , (3.14)
2 1 0

which is a nilpotent matrix: T+2 = 0.

In general, a bilinear symmetric form on a Lie algebra G is introduced by the Killing form

K(A, B) = Tr(adA adB ) , (3.15)

where adA is the operator for the adjoint representation (3.5). To calculate the Killing form, one
can choose a basis in G and associate to adA a matrix (adA )m n in a standard way (see (E.4)).
Then one finds
K(A, B) = (adA )m n (adB )n m . (3.16)
Note that this calculation does not depend on the choice of a basis en .
For A = Am em and B = B n en , the Killing form can be written as

K(A, B) = Kmn Am B n , (3.17)

with
Kmn = ck ml cl nk , (3.18)
where ck ml are the structure constants of G (see the exercises 1 and 2).
This calculation for the sl(2, R) algebra gives

K(A, B) = −8⟨ A B ⟩ . (3.19)

Exercises 7-9.

2
3. The exponential map
Due to (3.8), the square of any A ∈ sl(2, R) is proportional to the unit matrix

A · A = −⟨ A A ⟩ I . (3.20)

This formula helps to find a compact form of eA

√ A
eA = cos θ I + sin θ Â , with θ = ⟨AA⟩ , Â = , if ⟨ A A ⟩ > 0 ; (3.21)
θ
√ A
eA = cosh λ I + sinh λ Â , with λ = −⟨ A A ⟩ , Â = , if ⟨ A A ⟩ < 0 ; (3.22)
λ
eA = I + A , if ⟨ A A ⟩ = 0 . (3.23)

In particular,
( ) ( ) ( )
θ T0 cos θ − sin θ λ T2 eλ 0 ρ T+ 1 0
e = , e = , e = . (3.24)
sin θ cos θ 0 e−λ ρ 1

Exercise 10.

4. The adjoint representation of SL(2, R)

The SL(2, R) group is the set of 2 × 2 real matrixes with unit determinant and the standard
multiplication rule of matrices. An element g ∈ SL(2, R) and its inverse g −1 can be written as
( ) ( )
a b −1 d −b
g= , g = , with ad − bc = 1 . (3.25)
c d −c a

Since det eA = eTr A , the exponentials (3.21)-(3.23) have the unit determinant, but note that this
map covers only a part of SL(2, R). In particular, one can not get the elements with Tr g < −2.
Let us introduce the operators Adg , which act on sl(2, R) by

Adg (A) = g A g −1 . (3.26)

It is a representation of SL(2, R) (Adg Adg′ = Adg g′ ), which is called the adjoint representation.
The transformations (3.26) obviously leave the bilinear form (3.10) invariant

⟨ Adg (A) Adg (B)⟩ = ⟨ A B⟩ . (3.27)

The corresponding transformations of the coordinates An = ⟨ T n A ⟩ are

An 7→ Λn m Am , (3.28)

with
Λn m = ⟨ T n g Tm g −1 ⟩ . (3.29)
Since the sl(2, R) algebra is isometric to 3d Minkowski space, we find that (3.28) is a 3d Lorentz
transformation. Thus, eq. (3.29) provides a map from the SL(2, R) group to the Lorentz group.
As far as the SL(2, R) group manifold is connected (see the next section), the matrices Λn m
belong to the SO↑ (1, 2) subgroup. Note that (3.29) maps g and −g to the same Lorentz matrix.
Exercises 11-12.

3
5. Coordinates on SL(2, R)
SL(2, R) is a three dimensional manifold. Its parameterization can be obtained as a combination
of one parameter subgroups of the type (3.24). One of these combinations (see the exercise 10)
can be written as
( )
cosh λ cos α + sinh λ cos β − cosh λ sin α + sinh λ sin β
g(λ, α, ρ) = . (3.30)
cosh λ sin α + sinh λ sin β cosh λ cos α − sinh λ cos β

These parameters become ‘global coordinates’ on SL(2, R) for

λ ≥ 0, 0 ≤ α < 2π , 0 ≤ β < 2π . (3.31)

To prove this statement, we consider another useful parameterization.

Adding the unit matrix to the basis (3.7), one gets a basis in the space of 2 × 2 matrices.
Therefore, any g ∈ SL(2, R) can be written as

g = c I + un Tn , (3.32)

with some real numbers c and un . The matrix form of (3.32) is

( )
c − u2 −u0 − u1
g= , (3.33)
u0 − u1 c + u2

where the parameters (un , c) satisfy the condition

c2 + (u0 )2 − (u1 )2 − (u2 )2 = 1 . (3.34)

To parameterize g in terms of independent variables it is convenient to consider the planes (u1 , u2 )

and (u0 , c). A point from the plane (u1 , u2 ) fixes the radial distance on the plane (u0 , c) by
c2 + (u0 )2 = 1 + (u1 )2 + (u2 )2 and the freedom remains in the polar angle only. Thus, the
SL(2, R) group manifold can be treated as R2 × S 1 , where R2 stands for the plane (u1 , u2 ) and
S 1 for the polar angle on (u0 , c).
u2 c
c2 + (u0 )2 ≥ 1
'$

u1 u0
&%

Comparing (3.33) and (3.30) we find

u1 = − sinh λ sin β , u0 = cosh λ sin α , (3.35)

u2 = − sinh λ cos β , c = cosh λ cos α . (3.36)

Hence, sinh λ and β play the role of polar coordinates on the plane (u1 , u2 ) and α is the polar
angle on (u0 , c).
Exercises 13-14.

4
6. Functions, vector fields, 1-forms and the metric on SL(2, R)
The matrix elements gαβ are functions on the SL(2, R) group manifold, which are related by
g11 g22 − g12 g21 = 1.
Vector fields V̂ are given as linear operators acting on functions (see the appendix)

∂f (x)
V̂ [f ] = V µ (x) . (3.37)
∂xµ

Here xµ are coordinates and V µ (x) denote the components of V̂ in these coordinates. The
solutions of the equation ẋµ = V µ (x) define a flow on the manifold.
Let us introduce two vector fields L̂A and R̂A , labeled by A ∈ sl(2, R) and defined as

L̂A (gαβ ) = (A g)αβ , R̂A (gαβ ) = (g A)αβ . (3.38)

Writing this equations without matrix indices

L̂A (g) = A g , R̂A (g) = g A , (3.39)

one immediately finds the corresponding flows (see the appendix for a definition)

g 7→ g(t) = etA g , g 7→ g(t) = g etA , (3.40)

as left and right multiplications for L̂A and R̂A , respectively.

As it follows from (3.39)

[L̂A , L̂B ] = −L̂[A,B] , [R̂A , R̂B ] = R̂[A,B] , [L̂A , R̂B ] = 0 . (3.41)

If A and B are the basis vectors (3.7), these commutators become

[L̂m , L̂n ] = −2ϵl mn L̂l , [R̂m , R̂n ] = 2ϵl mn R̂l , [L̂m , R̂n ] = 0 , (3.42)

where we have used the notations L̂n = L̂Tn , R̂n = R̂Tn .

1-forms θ = θµ (x) dxµ are characterized by co-vector fields θµ (x). They act on vector fields
and give functions
θ [ V̂ ] = θµ (x) V µ (x) . (3.43)
Particular examples of 1-forms are differentials of functions df and the rule (3.43) provides

df [V̂ ] = V̂ [f ] . (3.44)

Similarly to the vector fields, let us introduce the left and the right 1-forms parameterized by the
elements of sl(2, R)
LA = ⟨A dg g −1 ⟩ , RA = ⟨A g −1 dg⟩ . (3.45)
The action of these 1-forms on the vector fields (3.39) are given by

LA [L̂B ] = ⟨A B⟩ , RA [R̂B ] = ⟨A B⟩ , (3.46)

and
LA [R̂B ] = ⟨A g B g −1 ⟩ , RA [L̂B ] = ⟨A g −1 B g ⟩ . (3.47)

5
Taking A and B as the basis vectors (3.7) and using the notations Lm = LTm , Rm = RTm we get

Lm [L̂n ] = ηmn , Rm [R̂n ] = ηmn . (3.48)

Thus, the left and right 1-forms are dual to the corresponding vector fields. One also has

Lm [R̂n ] = ⟨Tm g Tn g −1 ⟩ , Rm [L̂B ] = ⟨Tm g −1 Tn g ⟩ . (3.49)

The last two functions are just the matrix components of Lorentz transformations (3.29).
One can also consider the left and the right 1-forms with values in the sl(2, R) algebra (Maurer-
Cartan form)

Ln T n = dg g −1 , Rn T n = g −1 dg . (3.50)

Here we have used the summation property (E.12) for the matrices Tn .
The metric tensor on SL(2.R) is defined by

g = ⟨ (g −1 dg) ⊗ (g −1 dg)⟩ . (3.51)

It is a symmetric 2-form with the components

gµν = ⟨ g −1 ∂µ g g −1 ∂ν g⟩ . (3.52)

In terms of the left or the right 1-forms the metric (3.51) reads

g = Ln ⊗ Ln = Rn ⊗ Rn , (3.53)

and the components are

gµν = η mn Lm , µ Ln , ν = η mn Rm , µ Rn , ν, (3.54)

with
Lm , µ = ⟨Tm ∂µ g g −1 ⟩ , Rm , µ = ⟨Tm g −1 ∂µ g⟩ . (3.55)
The metric tensor (3.52) is invariant under the left (g 7→ hg) and the right (g 7→ gh) multi-
plications of g by the group elements h ∈ SL(2, R).
Exercises 15.

Remarks

A Lie algebra G is called semi-simple if its Killing form is non-degenerated.

A Lie algebra G is called Abelian, if [A, B] = 0 for any A, B ∈ G.
Examples of semi-simple Lie algebras are o(n), for n > 2, su(N ), sp(n), o(p, q), su(p, q).
Exercise 16.

6
Appendix
Covariant tensors T (x) have down components Tmn... (x). They act on vector fields
T (v, w, ...) = Tmn... v m wn ... . (A.1)
Contravariant tensors T (x) have upper components T mn... (x). They act on functions
T (f, g, ...) = T mn... ∂m f ∂n g... . (A.2)
Functions, forms and metric are covariant tensors, while vector fields are contravariant ones.
A map ϕ : M 7→ M̃ from a manifold M to a manifold M̃ creates:
• ϕ∗ - pull-back map of covariant tensors.
• ϕ∗ - push-forward map of contravariant tensors.

One has
ϕ(x) = x̃ , ϕ∗ f˜ = f˜ ◦ ϕ , (A.3)
ϕ∗ v (f˜) = v(ϕ∗ f˜) , ϕ∗ θ̃(v) = θ̃(ϕ∗ v) . (A.4)
Here x̃ is a point in M̃, f˜ is a function on M̃, f˜ ◦ ϕ denotes the composition of f˜ and ϕ; v is
a vector field on M and θ̃ is a co-vector field on M̃. The generalization to higher order tensors
is straightforward. Then, one gets in components
∂ x̃m̃ ∂ x̃ñ
(ϕ∗ T̃ )mn = T̃m̃ñ , (A.5)
∂xm ∂xn
∂ x̃m̃ ∂ x̃ñ mn
(ϕ∗ T )m̃ñ = T . (A.6)
∂xm ∂xn

ϕ∗ : v 7→ ṽ
v ṽ
·
x → ·
ϕ : x 7→ x̃ x̃
θ θ̃

M θ ← θ̃ : ϕ∗ M̃

The pull-back map of the metric tensor is called the induced metric.
A flow ϕt on M is a one parameter additive family of transformations of M
ϕt : M 7→ M , with ϕt ◦ ϕt′ = ϕt+t′ . (A.7)
A vector field v(x) provides the dynamical equation ẋ = v(x). The solutions of this equation
x(t) with all possible initial data x(t)|t=0 = x define a flow ϕvt (x) = x(t). This flow acts on
covariant and contravariant tensors according to (A.5) and (A.6), respectively. The Lie derivative
Lv of a tensor filed T is defined by
d
Lv T = (ϕv ∗ T ) |t=0 . (A.8)
dt t
The components of Lv T (see the next lecture) can be obtained from (A.5), replacing ϕ∗ by
ϕvt ∗ , and using that
ϕvt (x) = x + tv(x) + O(t2 ),
for small t.

7
Exercises
1. Let en be a basis of a Lie algebra G. Since [ em , en ] ∈ G, one has

[ em , en ] = el cl mn , (E.1)

with some constants cl mn , which are called the structure constants of G.

Check that the structure constants satisfy the relations

cl mn = −cl nm , (E.2)
k j k j k j
c lm c kn +c mn c kl +c nl c km =0. (E.3)

2. Let (adA )m n be the matrix associate with adA in a basis en

adA (en ) = em (adA )m n . (E.4)

Check that
(adel )m n = cm ln , (E.5)
and relate the identity (E.3) to the commutator of the basis vectors in the adjoint representation.
3. Check that the commutators of the basis vectors (3.7) are

[ T1 , T2 ] = 2 T0 , [ T1 , T0 ] = 2 T2 , [ T0 , T2 ] = 2 T1 . (E.6)

4. Verify the following relation

(Tn )αβ (T n )α′ β ′ = δαβ δα′ β ′ − 2δαβ ′ δβα′ . (E.7)

5. Prove that  
ηil ηim ηin
ϵijk ϵlmn = det  ηjl ηjm ηjn  . (E.8)
ηkl ηkm ηkn
6. Using (E.8) derive the summation rules

ϵij k ϵkmn = ηim ηjn − ηin ηjm , ϵi jk ϵjkn = 2ηin , ϵijk ϵijk = 6 . (E.9)

7. By (3.8) check that

⟨ Tl Tm Tn ⟩ = ϵlmn . (E.10)
8. Using (E.9) derive that

⟨ Tk Tl Tm Tn ⟩ = ηkm ηln − ηkn ηlm − ηkl ηmn . (E.11)

9. Let A ∈ sl(2, R) and B ∈ sl(2, R). Using (E.7) check that

⟨ Tn A ⟩ ⟨ T n B ⟩ = ⟨ A B ⟩ . (E.12)

10. Calculate the product of three exponents given below and show that
( )
θ T0 λ T2 γ T0 cosh λ cos α + sinh λ cos β − cosh λ sin α + sinh λ sin β
e e e = , (E.13)
cosh λ sin α + sinh λ sin β cosh λ cos α − sinh λ cos β

with α = θ + γ and β = θ − γ.

8
11. Prove that:
a) any g ∈ SL(2, R) with −2 < Tr g < 2 is given by (3.21).
b) any g ∈ SL(2, R) with Tr g > 2 is given by (3.22).
c) any g ∈ SL(2, R) with Tr g < −2 is given by g = −eA with a space-like A.
d) any g ∈ SL(2, R) with Tr g = 2 is given by (3.23).
12. Using (E.12), prove that the matrixes Λn m defined by (3.29) satisfy the conditions

Λn m Λn l = ηml . (E.14)

13. Prove that the inverse to (3.32) is

g −1 = c I − un Tn . (E.15)

14. The SU (1, 1) group is defined as the set of 2 × 2 complex matrixes with unit determinant
which preserve the following scalar product

( ψ|χ ) = ψ1∗ χ1 − ψ2∗ χ2 . (E.16)

Show that:
a) a group element g̃ ∈ SU (1, 1) is given by
( )
z u
g̃ = , (E.17)
u∗ z ∗

where the z and u are complex numbers, with |z|2 − |u|2 = 1.

b) the map ( )
−1 1 1 i
g̃ 7→ g = U g̃ U with U=√ , (E.18)
2 i 1
transforms the SU (1, 1) matrixes to the SL(2, R) ones. Thus, these two groups are equivalent.
15. Let xn be the normal coordinates on SL(2, R) defined by
n
g = ex Tn
. (E.19)

Calculate in theses coordinates :

a) The components of the vector fields L̂n and R̂n .
b) The components of the co-vector fields for the 1-forms Ln and Rn .
c) The components of the metric tensor g.

16. Check that the Killing form is non-degenerated for the su(2) algebra, but it is degenerated
for u(2).

17. Let us consider two maps from the sl(2, R) algebra to the SL(2, R) group

A 7→ g = eA and B 7→ g = −eB . (E.20)

Check that they together cover SL(2, R). Describe the domain on SL(2R), where these maps
intersect and find there the relation between A and B.

18. Check that the Lie derivative of the metric tensor on the SL(2, R) group manifold vanishes
for the left and right vector fields.

9
19. Let us consider the 4-dimensional space R2,2 with coordinates (u1 , u2 , u0 , c) and the metric

ds2 = dc2 + du20 − du21 − du22 . (E.21)

Its isometry group is O(2, 2). The hyperboloid

c2 + u20 − u21 − u22 = 1 (E.22)

embedded in R2,2 is invariant under the O(2, 2) transformations. Verify that the induced metric
on the hyperboloid (E.22) coincides with the metric on SL(2, R).

References
[1] B.A. Dubrovin, A.T. Fomenko, S.P. Novikov. Modern Geometry-Methods and Applications
Vol. I. Springer-Verlag New York. 1984
(Chapter 3, PP 24, Pages 212-224)

10
Lecture 4

Particle dynamics on manifolds

In this lecture we consider simple models of particle dynamics and demonstrate how the mathe-
matical tools introduced in the previous lectures work for them.

1. The Liouville model

The first example is the dynamics of a particle in the exponential potential. Here we obtain the
space of motions M , calculate the symplectic form on M and find its tangent vectors.
The Liouville field theory is described by the dynamical equation
( 2 )
∂τ − ∂σ2 φ(τ, σ) + 4m2 e2φ(τ,σ) = 0 , (4.1)

where m > 0 is a coupling constant and (σ, τ ) are space-time coordinates. This model of expo-
nentially self-interacting field theory is integrable. Its general solution in terms of two arbitrary
functions A and B can be written in the form
1 A′ (τ + σ) B ′ (τ − σ)
φ(τ, σ) = log . (4.2)
2 [1 + m2 A(τ + σ) B(τ − σ)]2
Let us consider the homogeneous field configurations ∂σ φ = 0. These fields are time dependent
φ(τ, σ) ≡ x(τ ) and they describe the dynamics of a particle in the exponential potential

ẍ(τ ) + 4m2 e2x(τ ) = 0 . (4.3)

The Lagrangian of this model is

1 2
L= ẋ − 2m2 e2x , (4.4)
2
and it provides the standard symplectic form on T Q (see (2.57))

ωL = dv ∧ dx . (4.5)

Using the conservation of energy

1 2
E= ẋ + 2m2 e2x , (4.6)
2
the equation of motion (4.3) can be integrated in the form
p
x(t) = log , (4.7)
2m cosh(q + pτ )
√
where p = 2E and q is an integration constant. The variables (p, q) parameterize the space of
motions and since the energy is positive, these variables are given on the half-plane p > 0. Note
that the solution (4.7) is obtained from (4.2) for m A(x) = m B(x) = e(q+px) .
The symplectic form on the space of motions (4.7) is induced by (4.5) as the pull-back map
(see the appendix of Lecture 3 and (2.63)) for

ϕ(t0 ) : x(t) 7→ (x, v) = (x(t0 ), ẋ(t0 )) . (4.8)

Calculating first the velocity from (4.7)

ẋ(t) = −p tanh(q + pτ ) , (4.9)

1
and then taking the differentials of (4.7) and (4.9) with respect to p and q

dp
dx(t) = − tanh(q + pτ ) (dq + τ dp) , (4.10)
p
p
dẋ(t) = − tanh(q + pτ ) dp − 2 (dq + τ dp) , (4.11)
cosh (q + pτ )

by direct calculation we obtain

dẋ(t0 ) ∧ dx(t0 ) = dp ∧ dq . (4.12)

As it was expected, this symplectic form does not depend on t0 .

The equation for the tangent vectors (2.65) in this case reads

2p2
ü(τ ) + 2 u(τ ) = 0 . (4.13)
cosh (q + pτ )

This second order linear equation has two linear independent solutions with the unit Wronskian

u1 (τ ) = tanh(q + pτ ) and u2 (τ ) = τ tanh(q + pτ ) − 1/p . (4.14)

This tangent vectors are also obtained from (4.7), differentiating it with respect to q and p.
Exercise 1-2.

2. A free particle in a curved space

In this section we consider the dynamics of a free particle in a curved space and show how the
Hamilton equations on T Q reproduce the equations for geodesics.
Let us consider N -dimensional configuration space Q with coordinates q µ (µ = 1, ..., N ) and the
matric tensor gµν (q). A free particle dynamics in this space is described by the Lagrangian

1
L= g (q) q̇ µ q̇ ν . (4.15)
2 µν
The metric tensor is assumed positively defined to realize the minimal action principle. Multi-
plying the Euler-Lagrange equations obtained from (4.15) by gµν (q), which is the inverse to the
′
metric tensor (gµµ gµ′ ν = δ µ ν ), one finds the equations for geodesics

q̈ µ + Γµνσ (q) q̇ ν q̇ σ = 0 . (4.16)

Here Γµνσ (q) are the Christoffel symbols

1 µµ′ ( )
Γµνσ (q) = g ∂ν gµ′ σ + ∂σ gµ′ ν − ∂µ′ gνσ . (4.17)
2
The Lagrangian (4.15) provides on T Q the following symplectic form (see (2.57))

ωL = gµν dv ν ∧ dq µ + ∂σ gµν v ν dq σ ∧ dq µ , (4.18)

and the Hamiltonian (see (2.54))

1
H= g (q) v µ v ν . (4.19)
2 µν

2
Differentiating this Hamiltonian, one can solve the equation dH +VH ⌋ωL = 0 for the Hamiltonian
vector field VH and obtain
∂ 1 ′ ( ) ∂
VH = v µ µ
+ gµµ ∂µ′ gνσ − 2∂σ gµ′ ν v ν v σ µ . (4.20)
∂q 2 ∂v
The corresponding Hamilton equations

q̇ a = v a , v̇ a = −Γabc (q) v b v c , (4.21)

are indeed equivalent to the equations of geodesics (4.16), written in the first order form.

Exercise 3-5.

3. Particle dynamics in SU (2)

Now we consider the dynamics of a particle on the SU (2) group manifold. We integrate the
equations of motion and describe the symplectic and Poisson bracket structures of the system.
Some details on the SU (2) group manifold are given in the Appendix (see also the exercise 6).
The SU (2) group manifold geometrically is realized as the 3-dimensional sphere S 3 . The metric
tensor on SU (2) is defined similarly to the SL(2, R) case by (A.10). Then, the free particle
Lagrangian (4.15) on SU (2) takes the form
1 −1 −1
L= ⟨g ġg ġ⟩ . (4.22)
2
The Euler-Lagrange equations ( ) ( )
d ∂L ∂L
= , (4.23)
dt ∂ ġαβ ∂gαβ
can be reduced to the matrix equation
d ( −1 )
g ġ = 0 . (4.24)
dt
Here we have used that
∂gα−1
′β′
= −gα−1 −1
′ α gββ ′ , (4.25)
∂gαβ
( )
which is a consequence of g −1 g αβ = δαβ , and also

∂L 1 ( −1 −1 ) ∂L 1 ( −1 −1 −1 )
=− g ġg βα
, = g ġg ġg βα
. (4.26)
∂ ġαβ 2 ∂gαβ 2

Integrating (4.24), first one finds

g −1 ġ = r , (4.27)
with a time independent su(2) matrix r. Further integration of ġ = g r leads to

g(t) = g0 ert , (4.28)

where g0 is the initial value for g(t): g0 = g(0).

Eq. (4.28) defines the space of motions M and it is parameterized by the pair (g0 , r).

3
 To find the symplectic form of the system one can introduce an equivalent to (4.22) Lagrangian
written in the first order form
1
L̃ = ⟨rg −1 ġ⟩ − ⟨r r⟩ . (4.29)
2
The variation of (4.29) with respect to r gives r = g −1 ġ and plug in it back into (4.29), one indeed
gets (4.22). The first order formalism leads to the Hamiltonian formulation. By (4.29) one can
introduce the 1-form
θ = ⟨rg −1 dg⟩ , (4.30)
called the symplectic potential. Its differential is the symplectic form

ω = ⟨dr ∧ g −1 dg⟩ − ⟨r g −1 dg ∧ g −1 dg⟩ , (4.31)

which can be rewritten as

ω = drm ∧ Rm − ϵklm rk Rl ∧ Rm . (4.32)
Here
rm = ⟨em r⟩ , Rm = ⟨em g −1 dg⟩ , (4.33)
em (m = 1, 2, 3) are the basis vectors of the su(2) algebra and, in addition, we have used eq.
(A.7) for the normalized traces.
The Hamiltonian vector field Vrn , associated with rn , is obtained from the equation

drn + Vrn ⌋ω = 0 , (4.34)

which by (4.32) leads to

drn + Vrn (rm ) Rm − Rm (Vrn ) drm − ϵklm rk Rl (Vrn ) Rm + ϵklm rk Rm (Vrn ) Rl = 0 . (4.35)

Since drm and Rm are six linearly independent 1-forms, we obtain

Rm (Vrn ) = δmn , Vrn (rm ) = 2ϵknm rk . (4.36)

The second equation here provides the Poisson brackets (see (2.45))

{rn , rm } = 2ϵknm rk , (4.37)

and the first one leads to (see (4.33)) Vrn (gαβ ) = (g en )αβ , or equivalently

{rn , g} = g en . (4.38)

The equation for the Hamiltonian vector field Vgαβ , given by dgαβ +Vαβ ⌋ω = 0, yields the relations
Rn (Vgαβ ) = 0, which defines the Poisson brackets

{gαβ , gα′ β ′ } = 0 . (4.39)

The generalization of this scheme to any semi simple Lie group is straightforward. Then
one concludes that the dynamics of a free particle on a semi-simple Lie group G is described
by the phase space G × G, with the pre-symplectic form (4.30). The Poisson brackets of the
corresponding variables (g, r) are given by (4.39), (4.38) and (4.37), replacing there 2ϵknm by the
structure constants of G.

Exercises 6-10.

4
4. A relativistic particle in AdS spaces
A manifold M with a Lorentzian metric can be interpreted as a curved space-time. The dynamics
of a relativistic particle on M is described by the action proportional to the length of a timelike
trajectory. Typical examples of such manifolds are de Sitter (dS) and anti de Sitter (AdS) spaces,
which have a constant curvature. This constant is positive for the dS spaces and negative for AdS.
The corresponding metric tensor satisfies the Einstein equation with a cosmological constant.
The (N + 1)-dimensional de Sitter space is defined as the following hyperboloid of a radius R

∑
N +1
Y02 − Yn2 + R2 = 0 (4.40)
n=1

embedded in the (N + 2)-dimensional Minkowski space R1, N +1 . One can check that the induced
metric tensor on the hyperboloid (4.40) has a Lorentzian signature (see the exercise 11), where
the space part is given by the N -dimensional sphere and the time coordinate is unbounded.
In this section we concentrate on the AdS spaces. We describe their symmetries and study
the dynamics of a relativistic particle.
The (N + 1)-dimensional AdS space is represented as the hyperboloid

∑
N
X02′ + X02 − Xn2 = R2 (4.41)
n=1

embedded in the (N + 2)-dimensional flat space R2, N with coordinates X A , A = (0′ , 0, 1, ..., N )
and the metric tensor GAB = diag(+, +, −, ..., −). Like in (4.40), the parameter R is called
the radius of the hyperboloid. To find the structure of the induced metric, we parameterize the
hyperboloid (4.41) by N + 1 coordinates xµ (µ = 0, 1, ..., N )
′
X 0 = r sin θ , X 0 = r cos θ , X n = xn (n = 1, ..., N ) ,
√
where θ = x0 and r = R2 + xn xn . (4.42)

The induced metric tensor gµν = GAB ∂µ X A ∂ν X B has a Lorentzian signature, since
xm xn
g00 = r2 , g0n = gn0 = 0 , gmn = −δmn + (4.43)
r2
(see also the exercise 12). Due to (4.42)-(4.43), the cyclic coordinate θ ∈ S 1 is identified with
time. In this way the hyperboloid (4.41) becomes a space-time manifold with a compact time
coordinate. Its isometry group is O(2, N ) and an analog of the proper Lorentz transformations is
the SO↑ (2, N ) subgroup, which is represented as a composition of the SO(2) × SO(N ) rotations
and the boosts in the planes (X0 , Xn ) and (X0′ , Xn ).
Unwrapping the time coordinate by θ 7→ t ∈ R1 , one gets the space, which is an universal
covering of the hyperboloid (4.41). The AdSN +1 space is usually associated with this covering
space. Since the hyperboloid (4.41) and AdSN +1 are locally isometric, particle dynamics on these
spaces is similar. The difference is that the closed timelike curves on the hyperboloid (4.41) are
not closed in AdSN +1 .
Let us consider the 2-dimensional case as an illustrative example. The corresponding hyper-
boloid X02′ + X02 − X12 = R2 , embedded in R2, 1 , is visualized on Fig. 1.

5
 X0

X1

X0 ′
Fig. 1

This hyperboloid can be mapped onto the cylinder (θ, σ), where θ ∈ S 1 is the time coordinate
(4.42) and σ ∈ (0, π) parameterizes X1 by X1 = R cot σ. Note that the induced metric tensor in
the coordinates (θ, σ) becomes conformally flat
( )
R2 1 0
gµν = . (4.44)
sin2 σ 0 −1

Unwrapping the cylinder (θ, σ) one gets AdS2 as a strip (t, σ), where t ∈ R1 .
In 3-dimensions the equation for the hyperboloid (4.41)

X02′ + X02 − X12 − X22 = R2 (4.45)

is equivalent to the condition ad − bc = 1 with

X0′ + X2 X1 + X0 X1 − X0 X0′ − X2
a= , b= , c= , d= . (4.46)
R R R R
Therefore, this space is described by the SL(2, R) group manifold. Moreover, the induced metric
on the hyperboloid dX A dXA coincides with the standard left-right invariant metric on the group
manifold ⟨ g −1 dg g −1 dg⟩ (see the exercise 19 of the lecture 3). Note that the left-right symmetry
of the metric on SL(2, R) corresponds to the decomposition SO↑ (2, 2) = SO↑ (2, 1) × SO↑ (2, 1).

The dynamics of a relativistic particle of a mass m moving on the hyperboloid (4.41) can be
described by the action
∫ [ ]
Ẋ A ẊA em2 µ A
S = − dτ + + (X XA − R2 ) . (4.47)
2e 2 2

Here e and µ are Lagrange multipliers and τ is an evolution parameter. To have the kinetic term
of the space coordinates Ẋn Ẋn with a positive coefficient, one assumes e > 0. Note that for e > 0

6
and m > 0, after elimination of the Lagrange multipliers e and µ, the action (4.47) reduces to
the standard form S = −ml, where l is the length of a world-line.
To fix the time direction, we also assume θ̇ > 0, which by (4.42) is equivalent to

X0 Ẋ0′ − X0′ Ẋ0 > 0 . (4.48)

The SO↑ (2, N ) symmetry of (4.47) provides the Noether’s conserved quantities

JAB = PA XB − PB XA , (4.49)

where PA are the canonical momenta PA = (∂L)/(∂ Ẋ A ) = −ẊA /e. The dynamical integrals J0n
and J0′ n are related to the above mentioned boosts, while J00′ and Jmn to the SO(2) and SO(N )
rotations, respectively. We use the notations J0n = Kn , J0′ n = Ln and J00′ = E. Since θ is the
time coordinate, E is associated with the particle energy, and due to (4.48) it is positive

E = P0 X0′ − P0′ X0 = e−1 (X0 Ẋ0′ − X0′ Ẋ0 ) > 0 . (4.50)

The dynamical integrals (4.49) allow to represent the set of all trajectories geometrically
without solving the dynamical equations. From (4.49) we find N equations as identities in the
(P, X)-variables
E Xn = Kn X0′ − Ln X0 , (n = 1, ..., N ) . (4.51)
Since E, Kn , Ln are constants, eq. (4.51) defines a 2-dimensional plane, which goes through
the origin of the embedding space R2, N . The intersection of this plane with the hyperboloid
(4.41) gives a trajectory, which is a timelike geodesic. This line can be parameterized by the time
coordinate θ (see the exercise 13).
The action (4.47) is invariant under reparametrizations τ → f (τ ), with the corresponding
transformations of the Lagrange multipliers (µ → µ /f ′ , e → e/f ′ ). The gauge symmetry leads
to dynamical constraints. Applying the Dirac’s procedure, we find three constraints

X A XA − R 2 = 0 , PA P A − m 2 = 0 , PA X A = 0 . (4.52)

They fix the quadratic Casimir number of the symmetry group by

1
JAB J AB = m2 R2 . (4.53)
2
This equation can be written as

E 2 + J 2 = K 2 + L2 + α 2 , (4.54)

where
1
J2 = Jmn Jmn , K 2 = Kn Kn , L2 = Ln Ln and α = mR . (4.55)
2
A set of other quadratic relations follows from (4.49) as the identities in the (P, X)-variables

JAB JA′ B ′ = JAA′ JBB ′ − JAB ′ JBA′ . (4.56)

These equations are nontrivial in terms of the dynamical integrals, if all indexes A, B, A′ , B ′ are
different. Taking A = 0, B = 0′ , A′ = m and B ′ = n (m ̸= n) we obtain

E Jmn = Km Ln − Kn Lm , (4.57)

7
and it provides
E 2 J 2 = K 2 L2 − (K · L)2 . (4.58)
From eq. (4.54) and (4.58) follows a quadratic equation for E 2 and one finds
1 ( 2 √ )
E2 = K + L2 + α2 + α4 + 2α2 (K 2 + L2 ) + (K 2 − L2 )2 + 4(KL)2 . (4.59)
2
We neglect the small root of the quadratic equation, since it does not describe a real trajectory.
Indeed, using the result of the exercise 13, the small root gives an imaginary r(θ) in eq. (E.11).
According to (4.59) α is the lowest value of energy. Two other inequalities
E 2 ≥ Kn Kn , E 2 ≥ Ln Ln , (4.60)
also follow from (4.59). They are similar to the relation between the energy and momentum in
the Minkowski space. On the basis of these inequalities one can show that the choice of the time
direction by (4.48) is invariant under the SO↑ (2, N ) transformations (see the exercise 14).
Eqs. (4.59) and (4.57) define E and Jmn as functions of (Kn , Ln ) and, therefore, (Kn , Ln )
are global coordinates on the space of dynamical integrals JAB . The physical phase space can be
identified with the space of dynamical integrals, since they form a complete set of gauge invariant
variables. Thus, the 2N variables (Kn , Ln ) form global coordinates on the physical phase space.
The canonical brackets {PA , X B } = δA B provide the o(2, N ) algebra of the generators (4.49)
{JAB , JA′ B ′ } = GAA′ JBB ′ + GBB ′ JAA′ − GAB ′ JBA′ − GBA′ JAB ′ . (4.61)
Due to the gauge invariance of JAB , the Poisson bracket relations (4.61) remain the same af-
ter reduction to the physical phase space, where the generators E and Jmn become non-linear
functions of the independent variables (Kn , Ln ).
The Poisson brackets on the physical phase space can be obtained by inversion of the sym-
plectic form, which corresponds to the reduced canonical form dPA ∧ dX A . To calculate the
reduction of this canonical form, it is useful to note that on the constraint surface (4.52) the
following relation holds
1
dPA ∧ dX A = JAB dJ AC ∧ dJ B C . (4.62)
2α2
The righthand side of this equation can be easily recalculated in terms of the coordinates (Kn , Ln ).
We demonstrate this procedure in 2-dimensions.
In this case there are only three dynamical integrals E, K = K1 and L = L1 . They are related
by the Casimir condition (4.54)
E 2 = K 2 + L2 + α 2 . (4.63)
Since E > 0, this equation defines the upper hyperbola in the space of dynamical integrals. The
calculation of the symplectic form (4.62) then yields
dL ∧ dK
ω= , (4.64)
E
√
with E = K 2 + L2 + α2 . The inversion of this symplectic form provides the Poisson bracket
relations of the o(2, 1) algebra
{L, K} = E , {E, K} = L , {E, L} = −K . (4.65)
Note the generators E, K, L have an oscillator representation with E = H + α, where H is the
harmonic oscillator hamiltonian (see the exercise 5 of the lecture 2).
The 3-dimensional case with the hyperboloid (4.45) is discussed in the exercise 15.

8
Appendix
The SU (2) group is given as the set of 2×2 unitary matrices with the unit determinant. Similarly
to SL(2, R) (see Lecture 3), an element g ∈ SU (2) and its inverse g −1 can be written as
( ) ( )
a b −1 d −b
g= , g = , with ad − bc = 1 . (A.1)
c d −c a
But now the numbers a, b, c, d are complex. The unitarity condition g + g = I is equivalent to
g + = g −1 and from (A.1) one obtains d = a∗ , c = −b∗ . As a result, the SU (2) group elements
are parameterized by
( )
a b
g= , with |a|2 + |b|2 = 1 . (A.2)
−b∗ a∗
Introducing the real parameters a = u1 + iu2 , b = u3 + iu4 , one gets the equation for S 3
u21 + u22 + u23 + u24 = 1 . (A.3)
Hence, the SU (2) group is a real 3-dimensional manifold, which is identified with S 3 .
The corresponding Lie algebra su(2) is formed by the anti-hermitian 2 × 2 traceless matrices.
Choosing the basis en = −iσn , n = (1, 2, 3), where σn are the Pauli matrixes
( ) ( ) ( )
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = , (A.4)
1 0 i 0 0 −1
one obtains
em en = −δmn I + ϵlmn el , (A.5)
and the commutators
[ em , en ] = 2ϵlmn el . (A.6)
Thus, the su(2) algebra is an Euclidean version of sl(2, R) and the definitions and constructions
discussed in Lecture 3 have a natural generalization. Below we give the corresponding list of
formulas, which are similar to the SL(2, R) case.
• The normalized trace: ⟨ M ⟩ = −1/2 Tr(M ).
• The normalized traces of products of the basis vectors:
⟨ em en ⟩ = δmn , ⟨ el em en ⟩ = ϵlmn . (A.7)
• The maps between su(2) and R3 : A = An en ∈ su(2), A 7→ An , An = ⟨ en A⟩.
• The square of an element A ∈ su(2): A2 = −⟨ A A⟩ I.
• The exponential map:
√ A
eA = cos θ I + sin θ Â , with θ = ⟨ A A ⟩ , Â = . (A.8)
θ
• The adjoint representation: Adg (A) 7→ g Ag −1 and the map from SU (2) to SO(3):
g 7→ Omn = ⟨ em g en g −1 ⟩ . (A.9)
• The left-right vector fields on SU (2): L̂n (g) = en g, R̂n (g) = g en .
• The left-right 1-forms on SU (2) : Ln = ⟨en dg g −1 ⟩, Rn = ⟨en g −1 dg⟩.
• The metric tensor on SU (2):
gµν = ⟨ g −1 ∂µ g g −1 ∂ν g⟩ . (A.10)

9
Exercises
1. The space of motions M for the oscillator problem ẍ + x = 0, can be written as

x(t) = A cos(t + α) . (E.1)

Calculate the symplectic form on M in terms of the parameters (A, α).

2. Show that the parameter p in (4.7) corresponds to the asymptotic in-momentum of the
Liouville particle and −p is the out-momentum, respectively.
Find the in and out coordinates in terms of p and q.
Constuct the canonical transformation from in to out variables, show that it has the form

(pin , qin ) 7→ (pout , qout ) = (−pin , −qin + β ′ (pin )) , (E.2)

and find the function β(p) (up to a constant).

3. The Lie derivative of a metric tensor gµν in components is given by (see the exercise 2.9)

(LV g)µν = V σ ∂σ gµν + gµσ ∂ν V σ + gσν ∂µ V σ . (E.3)

Check that if (LV g)µν = 0, then the free particle Lagrangian (4.15) is invariant under the in-
finitesimal transformations q µ 7→ q µ + ϵ V µ (q).
4. Check that if (LV g)µν = 0 (see the previous exercise), then C = gµν (q) V µ (q)q̇ ν is a dynamical
integral Ċ = 0.
5. Describe the geodesics on the sphere x2 + y 2 + z 2 = R2 .
6. Check that the metric tensor on the SU (2) group manifold (A.10) coincides with the standard
metric on S 3 induced from R4

ds2 = du21 + du22 + du23 + du24 . (E.4)

7. From (4.27) verify that the su(2) valued matrix l = ġg −1 , is also a dynamical integral (l˙ = 0)
and it is related to r by l = grg −1 .
8. Check that l and r (see the previous exercise) have the same norm ⟨l2 ⟩ = ⟨r2 ⟩.
9. Check that the variable ln = ⟨l en ⟩ (see the exercise 7) is given by ln = Onm rm , with
Omn = ⟨en g em g −1 ⟩ ∈ SO(3).
10. Using the Poisson brackets (4.37), (4.38) and (4.39) check that ln satisfies the following
Poisson bracket relations

{ln , lm } = −2ϵknm lk , {ln , g} = en g , {lm , rn } = 0 . (E.5)

11. Let us parameterize the (N + 1)-dimensional hyperboloid (4.40) by

Y0 = R sinh θ , Yn = R cosh θ un , (E.6)

where un is an unit vector (un un = 1) in RN +1 , which defines the N -dimensional sphere. Check
that the induced metric is given by

ds2 = R2 (dθ)2 − R2 cosh2 θ (dun dun ) . (E.7)

10
12. In the AdS case, the hyperboloid (4.41) can also be parameterized by

X0′ = R cosh ρ sin θ , X0 = R cosh ρ cos θ , Xn = R sinh ρ un , (E.8)

where now un is an unit vector in RN . Check that the induced metric in these coordinates is
[ ]
ds2 = R2 cosh2 ρ (dθ)2 − R2 (dρ)2 + sinh2 ρ (dun dun ) . (E.9)

13. Using the parametrization (4.42) of the hyperboloid (4.41), form (4.51) one finds the following
form of the trajectories
r(θ)
X0 = r(θ) cos θ , X0′ = r(θ) sin θ , Xn = (Kn sin θ − Ln cos θ) . (E.10)
E
Check that the function r(θ) can be written as
√
2 RE
r(θ) = √ , (E.11)
2E − [K + L − (K − L2 ) cos 2θ − K · L sin 2θ]
2 2 2 2

where K 2 and L2 are given by (4.55) and K · L = Kn Ln .

Note that eqs. (E.10)-(E.11) provide a parametrization of the trajectories through the dy-
namical integrals E, Kn , Ln and the evolution coordinate θ.
14. Check that the condition for the choice of time direction (4.48) is invariant under a boost
in the (X0 , Xn ) plane. One can use the form of the trajectories (4.51) and the inequality (4.60).
The invariance of (4.48) with respect to the rotations SO(2) × SO(N ) is obvious.
15. Here we consider the dynamics on the hyperboloid (4.45). In this case there is only one
quadratic relation (4.57), and together with the Casimir condition (4.54) one gets two equations

E 2 + J12
2
= K12 + K22 + L21 + L22 + α2 , E J12 = K1 L2 − K2 L1 . (E.12)

Let us introduce ‘left’ and ‘right’ variables (El , Kl , Ll ) and (Er , Kr , Lr ) defined by

2El = E + J12 , 2Kl = K1 + L2 , 2Ll = L1 − K2 ; (E.13)

2Er = E − J12 , 2Kr = K1 − L2 , 2Lr = L1 + K2 .

Using the algebra (4.61), check that the left variables have zero Poisson brackets with the right
variables and they both satisfy the Poisson bracket relations of the o(2, 1) algebra (4.65).
Check also that the two conditions of (E.12) are equivalent to

El2 = Kl2 + L2l + (α/2)2 , Er2 = Kr2 + L2r + (α/2)2 . (E.14)

These equations define two hyperbolas (4.63) with the lowest values of El and Er equal to α/2.
Find the oscillator representation of the symmetry generators (E.13), using the above men-
tioned splitting o(2, 2) = o(2, 1) ⊕ o(2, 1) and the result of the exercise 5 of the lecture 2.
Express the o(2, 2) generators E, J12 , Kn , Ln (n = 1, 2) in terms of the ‘creation-annihilation’
variables defined by
al + ar al − ar
a1 = √ , a2 = i √ , (E.15)
2 2
where al and ar are the ‘annihilation’ variables for the left (El , Kl , Ll ) and the right (Er , Kr , Lr )
generators, respectively.

11
References
[1] A.M. perelomov, “Integrable Systems of Classical Mechanics and Lie Algebras,” Birkhiiuser
Verlag Basel (1990).
(Chapters 1.1-1.8, 1.11-1.12).

[2] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, “Large N field theories,
string theory and gravity,” Phys. Rept. 323 (2000).
(Chapters 2.2.1-2.2.2).

[3] H. Dorn and G. Jorjadze, “Oscillator quantization of the massive scalar particle dynamics
on ads spacetime,” Phys. Lett. B 625 (2005).

12