Hamiltonian Formalism

f
Legendre transformations u
x
f f
f ( x, y ) df  dx  dy  udx  vdy f
x y v
y
g  f  ux dg  df  udx  xdu  udx  vdy  udx  xdu
g
v
g g y
dg  vdy  xdu  dy  du g ( y, u )
y u g
x
• Legendre transformation: u

g ( y, u )  f ( x, y )  ux
g g
v ;x  
y u Adrien-Marie
Legendre
(1752 –1833)
 g ( y, u )  f ( x, y )  ux
What is H?
g g
M
L v ;x  
H  qm  L y u
m 1 qm
L
• Conjugate momentum  pm
q m
• Then
M
 H  L(q1 ,..., qM , q1 ,..., q M , t )   pm q m
m 1
• So
H (q1 ,..., qM , p1 ,..., pM , t )
M
  L(q1 ,..., qM , q1 ,..., q M , t )   pm q m
m 1
 L
What is H?  pm
M q m
H   L   pm q m
m 1
M   L
M
 
dH  dL  d   pm q m     
L
 dqm  dqm 
 m1  m1  qm qm 
L M
 dt    pm dq m  q m dpm 
t m 1

M
L
   p m dqm  q m dpm   dt
L d  L  t
   m 1

qm dt  qm 
  pm   p m
d
dt
 What is H?
M
L
dH    p m dqm  q m dpm   dt
m 1 t

H H H L dH
  p m  q m  
qm pm t t dt

dH  H
M
H  H
   qm  p m  
dt m1  qm pm  t
M
H dH H
   p m q m  q m p m   
m 1 t dt t
 What is H?
• If L  L0 (q1 , q2 ,..., qM , t )   l1i (q1 , q2 ,..., qM , t )qi
i

  l2ij (q1 , q2 ,..., qM , t )qi q j  L0  L1  L2

i, j

• Then H  L2  L0
1 
T   mi ri  2
• Kinetic energy
i 2  
• In generalized coordinates ri  ri (q1 , q2 ,..., qM , t )
  2
1  ri ri 
T   mi  q j 
i 2
 t q 
 j j 
 What is H?
   
2
 2
1  ri ri 
T   mi  q j   mi  i 
 
1 r
 t  
i 2  j q j  i 2  t 
       
 r r 1  r  r
    mi i
 q j     mi
i i
 i q j q k

j  i q  t  2  q  q 
j  j ,k  i j k 
• For scleronomous generalized coordinates
 
ri  ri (q1 , q2 ,..., qM )
   
1  ri ri 
T    mi  q j q k  L2
2 j ,k  i q j qk 
• Then H  L2  L0  T  L0
• If V   L0 H  T  V  Emec
 What is H?
L  L0  L1  L2
• For scleronomous generalized coordinates, H is a
total mechanical energy of the system (even if H
depends explicitly on time)

• If H does not depend explicitly on time, it is a

constant of motion (even if is not a total mechanical
energy)

• In all other cases, H is neither a total mechanical

energy, nor a constant of motion
 Hamilton’s equations
M
• Hamiltonian: H (q1 ,..., qM , p1 ,..., pM , t )   pm q m
m 1

 L(q1 ,..., qM , q1 ,..., q M , t )

• Hamilton’s equations of motion:

H H H L
q m  p m   
pm qm t t

Sir William Rowan

Hamilton
(1805 – 1865)
 Hamiltonian formalism
• For a system with M degrees of freedom, we have
2M independent variables q and p: 2M-dimensional
phase space (vs. configuration space in Lagrangian
formalism)

• Instead of M second-order differential equations in

the Lagrangian formalism we work with 2M first-order
differential equations in the Hamiltonian formalism

• Hamiltonian approach works best for closed

holonomic systems

• Hamiltonian approach is particularly useful in

quantum mechanics, statistical physics, nonlinear
physics, perturbation theory
 Hamiltonian formalism for open
systems
d  L  L
  Qj
dt  q j  q
 j

dp j L
  Qj
dt q j

H H
q m  p m    Qm
pm qm
 Hamilton’s equations in symplectic
notation
• Construct a column matrix (vector) with 2M
elements
 j  q j ; M  j  p j
• Then
 H  H  H  H
   ;   
 η  j q j  η  M  j p j
• Construct a 2Mx2M square matrix as follows:
1 0 ... 0 0 0 ... 0
 0 1 0 1 ... 0  0 0 ... 0 
J  1  0
  1 0  ... ... ... ... ... ... ... ...
   
0 0 ... 1 0 0 ... 0
 Hamilton’s equations in symplectic
notation
• Then the equations of motion will look compact in
the symplectic (matrix) notation:

H
η  J
η
• Example (M = 2):
 q1   0 0 1 0  H / q1 
 q   0 0  
0 1  H / q 
 2  2

 p1   1 0 0 0  H / p1 
    
 p 2   0  1 0 0 H / p2 
 Lagrangian to Hamiltonian
• Obtain conjugate momenta from a Lagrangian
L
pm 
q m
• Write a Hamiltonian
M
H   L   pm q m
m 1

• Obtain qm  qm (q1 ,..., qM , p1 ,..., pM , t ) from

• Plug qm  qm (q1 ,..., qM , p1 ,..., pM , t ) into the

Hamiltonian to make it a function of coordinates,
momenta, and time
 Lagrangian to Hamiltonian bmn  bnm
• For a Lagrangian quadratic in generalized velocities
L(q1 ,..., qM , q1 ,..., q M , t )  L0 (q1 ,..., qM , t )
  am (q1 ,..., qM , t )q m   bmn (q1 ,..., qM , t )q m q n
m m,n
~ 1~
• Write a symplectic notation: L  L0  q  a  q bq
~ 2
• Then a Hamiltonian H  q p  L
~ ~ 1~ ~ 1~
 q p  L0  q a  q bq  q (p  a)  L0  q bq
2 2
• Conjugate momenta
L   ~ 1~ 
p  ~  ~  L0  q a  q bq   a  bq
q q  2 
 Lagrangian to Hamiltonian

p  a  bq
• Inverting this equation
~ ~ 1 ~
p  a  bq 1
b (p  a)  q (p  a )b  q
~ 1 ~
• Then a Hamiltonian H  q (p  a)  L0  q bq
2
~ ~ 1 1 ~ ~ 1 1
 (p  a )b (p  a)  L0  (p  a )b bb (p  a)
2
1 ~ ~ 1
H  (p  a )b (p  a)  L0
2
 Example: electromagnetism
 
m(r  r )  
L  q  q(r  A)
2
 
L p j  qAj  p  qA
pj   mrj  qAj rj  r
rj m m
 
H  pr  L
    2  
 p  qA m  p  qA   p  qA  
H  p     q  q  A 
m 2  m   m



 Example: electromagnetism
    2  
 p  qA m  p  qA   p  qA  
H  p     q  q  A 
m 2  m   m



         
p  p  qp  A p  p  2qp  A  q 2 A  A
   q
m 2m
   
p  A  qA  A
q
  m       
2 p  p p  p  q2 A  A 2qp  A  2q 2 A  A
   q 
2m  2m  2m
  
p  p q A A 2qp  A
 2

2

   q  p  qA
2m 2m 2m H  q
2m
 Hamilton’s equations from the
variational principle
• Action functional :
t2

I   L(q1 ,..., qM , q1 ,..., q M , t )dt

t1

M 
t2

    pm q m  H (q1 ,..., qM , p1 ,..., pM , t ) dt

t1  m 1 
• Variations in the phase space :

qm (t ,  )  qm 0 (t )  m (t )  m (t1 )   m (t2 )  0

pm (t ,  )  pm 0 (t )   m (t )  m (t1 )   m (t2 )  0
 Hamilton’s equations from the
variational principle
 dpm dH 
t2
dI M
dq m
  q m  pm  dt
d t1  m1 d d d 
 H H 
t2 M

    m q m   m pm  m   m dt
t1 m 1 
qm pm 
• Integrating by parts
 H H 
t2 M M
    m q m   m p m  m   m dt    m pm t
t2

t1 m 1  q m  p m  m 1
1

t2 M
  H    H  
     q m   m   p m    m dt

t1 m 1  
p m    q m  
 Hamilton’s equations from the
variational principle

dI
t2 M
 H    H  
     q m   m   p m    m dt  0

d t1 m 1   pm   q m  

• For arbitrary independent variations

H H
q m  ; p m  
pm qm
 Conservation laws

H H dH H
p m   ; q m  ; 
qm pm dt t

• If a Hamiltonian does not depend on a certain

coordinate explicitly (cyclic), the corresponding
conjugate momentum is a constant of motion

• If a Hamiltonian does not depend on a certain

conjugate momentum explicitly (cyclic), the
corresponding coordinate is a constant of motion

• If a Hamiltonian does not depend on time explicitly,

this Hamiltonian is a constant of motion
 Higher-derivative Lagrangians
  i ηm ( xn )  i  0,1,2,3... ?
• Let us recall: L  , x 
 x i n  x  x, y, z , t ,... ?
 n  n

 L dxdydz  L d rm (ti
) / dt , t 
i rm  rx , ry , rz

• Lagrangians with i > 1 occur in many systems and

theories:
1. Non-relativistic classical radiating charged particle
(see Jackson)
2. Dirac’s relativistic generalization of that
3. Nonlinear dynamics
4. Cosmology
5. String theory
6. Etc.
 Higher-derivative Lagrangians
• For simplicity, consider a 1D case:
L  L( x, x, x, x,...)
• Variation x(t ,  )  x0 (t )  (t )
 L L L L
t2 t2
dI dL 
 dt            ...dt
  
d t1 d t1 
x x x x 

L d  L  L
t2 t2 t2

t x dt  t  dt  x dt  x  t

1 1 1

Mikhail Vasilievich
Ostrogradsky
(1801 - 1862)
 Higher-derivative Lagrangians
 L L L L
t2 t2
dI dL 
 dt             ...dt
d t1 d t1 
x x x x 

L d  L  L
t2 t2 t2

t x dt  t  dt  x dt  x  t

1 1 1

t2
d  L  L d  L 
t2 2 t2

  2  dt     
t1
dt  x  x t1 dt  x  t1
 Higher-derivative Lagrangians
 L L L L
t2 t2
dI dL 
 dt             ...dt
d t1 d t1 
x x x x 
L d  L  L
t2 t2 t2

t xdt  t  dt  x dt  x t

1 1 1

t2
d  L  L d  L 
t2 2 t2

  2  dt     

t1
dt  x  x t1 dt  x  t1
t2 t2
d  L  L d  L  d  L 
t2 3 t2 2
   3  dt       2  
t1
dt  x  x t1 dt  x  t1 dt  x  t
1
 Higher-derivative Lagrangians
 L d  L  d  L  d  L  
t2 2 3
dI
       2    3    ... dt
d t1  x dt  x  dt  x  dt  x  
 L d  L  d 2  L   t2 p1
      2    ... t
 x dt  
x  dt  
x   1
p2
 L d  L   t2 p3
      ... t
  x dt   
x   1 ...
k i
 L   t2  d
...
L
   ...  t  ...  0 pi     
 x  1 k i  dt  k
d
  x
• Generalized coordinates/momenta:  dt 
q1  x; q2  x; q3  x; q4  x; ...
 Higher-derivative Lagrangians
dI
t2
  d 
k 1
L   d 
k 1

    
 dt   pk     t  0
t2

d t1  k 1  dt  qk  k 1  dt  1

• Euler-Lagrange equations:
k 1
 d L
  
k 1  dt  qk
0

• We have formulated a ‘higher-order’ Lagrangian

formalism

• What kind of behavior does it produce?

 Example
L  L( x, x , x)
L d L L
q1  x p1   p1   p 2
x dt x x
q2  x L  L
p2  p1  p 2 
x x

L d L  d  L
2
L
   0  p 1
x dt x  dt  x x

H  p1q1  p2q2  L( x, x, x)

 L
Example  p 1
x
H  p1q1  p2q2  L( x, x, x)
L L
dH  p1dq1  q1dp1  p2 dq 2  q 2 dp2  dx  dx
x x
L
 dx  p1dq2  q1dp1  p2 dq 2  q 2 dp2  p1dq1
x
  p1  p 2 dq2  p2 dq 2
 q1dp1  q2 dp2  p1dq1  p 2dq2

H H dH
q m  ; p m   ; 0
pm qm dt
L L
 p2  p1  p 2
x x
 q1  x  q2
Example
H  p1q1  p2q2  L
• H is conserved and it generates evolution – it is a
Hamiltonian!

H  p1q1  p2q2  L  p1q2  p2q2  L

• Hamiltonian linear in momentum?!?!?!

• No low boundary on the total energy – lack of

ground state!!!

• Produces ‘runaway’ solutions: the system becomes

highly unstable - collapse and explosion at the same
time
 ‘Runaway’ solutions
• Unrestricted low boundary of the total energy
produces instabilities

• Additionally, we generate new degrees of freedom,

which require introduction of additional (originally
unknown) initial conditions for them

• These problems are solved by means of

introduction of constraints

• Constraints restrict unstable behavior and eliminate

unnecessary new degrees of freedom
 9.1
Canonical transformations
• Recall gauge invariance (leaves the evolution of the
system unchanged):
dF
L'  L  L  {qm }; L'  {Qm }
dt
• Let’s combine gauge invariance with Legendre
transformation:
M M
dF

m 1
pm q m  H   PmQm  K 
m 1

dt
 K  K dK K
Qm  ; Pm   ; 
Pm Qm dt t
• K – is the new Hamiltonian (‘Kamiltonian’ )

• K may be functionally different from H

 9.1
Canonical transformations
M M
dF

m 1
pm q m  H   PmQm  K 
m 1

dt
• Multiplying by the time differential:
M M

p
m 1
m dqm  Hdt   Pm dQm  Kdt  dF
m 1
M
dF    pm dqm  Pm dQm   K  H dt
m 1
• So
F F F
pm  ; Pm   ;K  H 
qm Qm t
F  F (q1 , q2 ,..., qM , Q1 , Q2 ,..., QM , t )
 9.1
Generating functions
F  F1 (q1 , q2 ,..., qM , Q1 , Q2 ,..., QM , t )
• Such functions are called generating functions of
canonical transformations

• They are functions of both the old and the new

canonical variables, so establish a link between the
two sets

• Legendre transformations may yield a variety of

other generating functions
 9.1
Generating functions
F  F1 (q1 , q2 ,..., qM , Q1 , Q2 ,..., QM , t )

• We have three additional choices:

M
F  F2 (q1 , q2 ,..., qM , P1 , P2 ,..., PM , t )   Qm Pm
m 1
M
F  F3 ( p1 , p2 ,..., pM , Q1 , Q2 ,..., QM , t )   qm pm
m 1
M
F  F4 ( p1 , p2 ,..., pM , P1 , P2 ,..., PM , t )   qm pm  Qm Pm 
m 1
• Canonical transformations may also be produced by
a mixture of the four generating functions
 9.2
An example of a canonical
transformation
M
F F F
F1   qmQm pm  ; Pm   ;K  H 
m 1 qm Qm t

Qm  pm ; Pm  qm ; K  H

• Generalized coordinates are indistinguishable from

their conjugate momenta, and the nomenclature for
them is arbitrary

• Bottom-line: generalized coordinates and their

conjugate momenta should be treated equally in the
phase space
 9.4
Criterion for canonical transformations
Q  Q(q, p); P  P(q, p) q  q(Q, P); p  p(Q, P)
• How to make sure this transformation is canonical?
 Q Q H Q H Q
Q  q  p  
q p p q q p
• On the other hand
H H p H q
 
P p P q P
• If
Q p Q q
 ; 
q P p P
• Then
 H
Q
P
 9.4
Criterion for canonical transformations
• Similarly,  P P H P H P
P q  p  
q p p q q p

H H p H q
 
Q p Q q Q
• If P p P q
 ; 
q Q p Q

• Then  H
P
Q
 9.4
Criterion for canonical transformations
• So,

H H  H  H
q  ; p   Q  ;P   ;
p q P Q

• If

P q Q p P p Q q
 ;  ;  ; 
p Q q P q Q p P
 9.4
Canonical transformations in a
symplectic form
H  j  q j ; M  j  p j
η  J
η
• After transformation ζ  ζ (η)  j  Q j ;  M  j  Pj
 i  i
ζ  Aη  AJ H
2M
i  
  j Aij 
j 1  j  j η
• On the other hand

H 2 M H  j H ~ H ~ H
 A ζ  AJA
i j 1  j i η ζ ζ
F
 0 K  H
t
 9.4
Canonical transformations in a
symplectic form
~ H
ζ  AJA
ζ H

• For the transformations to be canonical: ζ  J
ζ
• Hence, the canonicity criterion is:
~
AJA  J
• For the case M = 1, it is reduced to (check yourself)

P q Q p P p Q q
 ;  ;  ; 
p Q q P q Q p P
 9.3
1D harmonic oscillator
p 2 kq2
H  E
2m 2
• Let us find a conserved canonical momentum
H
P  const P0 
 H  H (P )
Q
• Generating function F  F (q, Q)
1

F F
P p
Q q
 F  F / q  kq
2 2 2 2
p kq
H   H     
2m 2  Q  2m 2
 9.3
1D harmonic oscillator
H  F / Q  
F / q  kq2
2

2m 2
• Nonlinear partial differential equation for F 

• Let’s try to separate variables

H  A' b  
 Ab'

2
kq 2
F  A(Q)b(q)
2m 2
• Let’s try b '  q H ( P)  P
 A' q   A
2 2
k 2 A 2
A 2
 km
H       q  A'  k 
 2   2m 2  m m
dA A  km
2
 dA m 1  A 
 Q  m 2  cot  
dQ m A  km k  km 
 9.3
1D harmonic oscillator

Q  m / k cot 1 A / km  
A  km cot Q k / m 
 
2
q
F  A(Q)b(q)  km cot Q k / m
2
• We found a generating function!
H
H ( P)  P  E 
Q 1 Q  t  t0
P
F 2
q k
P 
Q 2 sin 2 (Q k / m )

2E
q sin(( t  t0 ) k / m )
k
 9.3
1D harmonic oscillator
 
2
q 2E
F  km cot Q k / m q  sin(( t  t ) k / m )
2 k
0

p
F
q
 q km cot Q k / m  
Q  t  t0
  2Em cos((t  t0 ) k / m )
2E
q sin(( t  t0 ) k / m )
k
p   2Em cos((t  t0 ) k / m )

p 2 / 2m  kq2 / 2  E
 9.3
1D harmonic oscillator

q, p Q, P
 9.5
Canonical invariants
• What remains invariant after a canonical ζ  ζ (η)
transformation? 2M
 i
d i   Aij d j Aij 
j 1  j
• Matrix A is a Jacobian of a space transformation

• From calculus, for elementary volumes: ~

2M 2M 2M AA
 d
i 1
i det( Aij ) di  A  di
i 1 i 1 ~
• Transformation is canonical if AJA  J
~
AJ A  J  0 1 ~
A A 1 A 1
2

~ J    1 0 J  1
AJAJ   A  1
 9.5
Canonical invariants
2M 2M 2M

 d
i 1
i  A  di   d i
i 1 i 1

• For a volume in the phase space

2M
V    d i    di  V
2M

i 1
 i 1

• Magnitude of volume in the phase space is invariant

with respect to canonical transformations:

V  V
 9.5
Canonical invariants
• What else remains invariant after canonical ζ  ζ (η)
transformations?
v ~ v  i ~
v  v ( η, t ) A Aij  AJA  J
η ζ  j
~ ~ ~
u ~ u  u   ~ u   u 
u  u ( η, t ) A     A     A
η ζ  η   ζ   ζ 
~ ~ ~
 u  v  u  ~ v  u  v
  J    A J A    J
  η  η  ζ  ζ  ζ  ζ
 9.5
Canonical invariants
~ ~
 u  v  u  v
  J    J
 η  η  ζ  ζ
• For M = 1
 v   v 
   
 u u  0 1  q   u u  p  u v
      
 q p   1 0  v   q p   v  q p
 p   q 
    v u
• For many variables 
~ q p
 u  v  u v v u 
  J     
 η  η i  qi pi qi pi 
 9.5
Poisson brackets
• Poisson brackets:
~
 u  v  u v v u 
  J       [u, v]
 η  η i  qi pi qi pi 
• Poisson brackets are invariant with respect to any
canonical transformation
 u v v u 
i  q p  q p 
 i i i i 

 u v v u 
    
i  Qi Pi Qi Pi  Siméon Denis Poisson
(1781 – 1840)
 9.5
Poisson brackets
• Properties of Poisson brackets :

[F , F ]  0
[ F , G ]  [G, F ]
[ F , G  X ]  [ F , G]  [ F , X ]
[ F , GX ]  [ F , G ] X  G[ F , X ]

[aF  bG, X ]  a[ F , X ]  b[G, X ]

[ F , [G, X ]]  [G, [ X , F ]]  [ X , [ F , G ]]  0
 9.5
Poisson brackets
~
 η  η
[ η, η]    J 1J1 J
 η  η
• In matrix element notation: [ η, η]lm  [ηl ,ηm ]  J lm
[qi ,q j ]  0 [ qi ,p j ]   ij [ pi ,p j ]  0 [ pi ,q j ]   ij
• In quantum mechanics, for the commutators of
coordinate and momentum operators:

[qˆi ,qˆ j ]  [ pˆ i ,pˆ j ]  0

[qˆi ,pˆ j ]  [ pˆ i ,qˆ j ]  i ij
 9.6
Poisson brackets and equations of
u  u ( η, t ) motion
~ ~
du  u  u  u  H u u
   η     J   [u , H ] 
dt  η  t  η  η t t
H du u
η  J   [u , H ]
η dt t
dH H dH H
  [H , H ] 
dt t dt t
dη η η  [ η, H ]
  [ η, H ]
dt t
 9.6
Poisson brackets and conservation
laws
du u
  [u , H ]
dt t
• If u is a constant of motion

u
du / dt  0  [H , u]
t
• If u has no explicit time dependence

[H , u]  0
• In quantum mechanics, conserved quantities
commute with the Hamiltonian
 9.6
Poisson brackets and conservation
laws
• If u and v are constants of motion with no explicit
time dependence
[ H , u ]  0; [ H , v]  0
• For Poisson brackets:
[ F , [G, X ]]  [G, [ X , F ]]  [ X , [ F , G ]]  0
[u, [v, H ]]  [v, [ H , u ]]  [ H , [u, v]]  0
d [u , v]
0
dt
• If we know at least two constants of motion, we can
obtain further constants of motion
 9.4
Infinitesimal canonical transformations
• Let us consider a canonical transformation with the
following generating function (ε – small parameter):
M
F   G (q1 , q2 ,..., qM , P1 , P2 ,..., PM )   (qm  Qm ) Pm
m 1
• Then M


M M
dF  P Q  K

m 1
pm q m  H   PmQm  K 
m 1

dt m 1
m m

 G
M
G   M
    q m  Pm    (q m  Q m ) Pm
m 1  qm Pm  m 1
M
  (qm  Qm ) Pm
m 1
 9.4
Infinitesimal canonical transformations
• Multiplying by dt
M M

p
m 1
m dqm  Hdt   Pm dQm  K dt
m 1

 G
M
G  M
    dqm  dPm    ( Pm dqm  Pm dQm )
m 1  qm Pm  m 1
M
  (qm dPm  Qm dPm )
m 1

• Then
G G
Pm  pm   Qm  qm   KH
qm Pm
 9.4
Infinitesimal canonical transformations
• Infinitesimal canonical transformations:
G G G
Pm  pm   ; Qm  qm    qm     ...
2

qm Pm pm

• In symplectic notation:
G ζ   G 
ζ  η J A  1  J 
η η η  η 
~
 u  v G
[u , v]    J ζη J
 η  η η
ζ  η   [ η, G ]
v G
[ η, v]  1 J [ η, G]  J
η η
 9.6
Evolution generation
ζ  η   [ η, G ] ζ  η  dt [ η, H ]  dt η  dη
  dt G  H [ η, H ]  η ζ  η  dη

• Motion of the system in time interval dt can be

described as an infinitesimal transformation
generated by the Hamiltonian

• The system motion in a finite time interval is a

succession of infinitesimal transformations,
equivalent to a single finite canonical transformation

• Evolution of the system is a canonical

transformation!!!
 9.9
Application to statistical mechanics
• In statistical mechanics we deal with huge numbers
of particles

• Instead of describing each particle separately, we

describe a given state of the system

• Each state of the system represents a point in the

phase space

• We cannot determine the initial conditions exactly

• Instead, we study a certain phase volume –

ensemble – as it evolves in time
 9.9
Application to statistical mechanics
• Ensemble can be described by its density – a
number of representative points in a given phase
volume N
D
V
• The number of representative points does not
change
N  const
• Ensemble evolution can be thought as a canonical
transformation generated by the Hamiltonian

• Volume of a phase space is a constant for a

canonical transformation
V  const
 9.9
Application to statistical mechanics
• Ensemble is evolving so its density is evolving too
dD D
  [ D, H ]
dt t N const
• On the other hand D   const
V const
D
 [ H , D]
t
• Liouville’s theorem

• In statistical equilibrium
D
 0 [ H , D ]  0
t Joseph Liouville
(1809 -1882)
 10.1
Hamilton–Jacobi theory
• We can look for the following canonical
transformation, relating the constant (e.g. initial)
values of the variables with the current ones:

qi 0  qi 0 (q1 ,..., qM , p1 ,..., pM , t )

pi 0  pi 0 (q1 ,..., qM , p1 ,..., pM , t )

• The reverse transformations will give us a complete

solution

qi  qi (q10 ,..., qM 0 , p10 ,..., pM 0 , t )

pi  pi (q10 ,..., qM 0 , p10 ,..., pM 0 , t )
 10.1
Hamilton–Jacobi theory
• Let us assume that the Kamiltonian is identically
zero K  0
Q i  K / Pi  0; Pi  K / Qi  0;
• Then
Qi  const ; Pi  const ;
• Choosing the following generating function
F  F2 (q1 ,..., qM , P1 ,..., PM , t )   Qi Pi
i
• Then, for such canonical transformation:
pi  F2 / qi Qi  F2 / Pi K  H  F2 / t
H (q1 ,..., qM , p1 ,..., pM , t )  F2 / t  0
F2 F2 F2
H (q1 ,..., qM , ,..., , t)  0
q1 qM t
 10.1
Hamilton–Jacobi theory
F2 F2 F2
H (q1 ,..., qM , ,..., , t)  0
q1 qM t
Sir William Rowan
• Hamilton–Jacobi equation
Hamilton
(1805 – 1865)
• Conventionally: S  F2 Hamilton’s principal function
S S S
H (q1 ,..., qM , ,..., , t)  0
q1 qM t
• Partial differential equation

• First order differential equation

Karl Gustav Jacob
• Number of variables: M + 1 Jacobi
(1804 – 1851)
 10.1
Hamilton–Jacobi theory
H (q1 ,..., qM , S / q1 ,..., S / qM , t )  S / t  0
• Suppose the solution exists, so it will produce M + 1
constants of integration:
S  S (q1 ,..., qM ,1 ,..., M 1 , t )
• One constant is evident:
S (q1 ,..., qM , 1 ,...,  M 1 , t )
H  S / t  0  S (q1 ,..., qM , 1 ,...,  M , t )   M 1
• We chose those M constants to be the new
momenta P 
i i

• While the old momenta

S (q1 ,..., qM , 1 ,...,  M , t )
pi 
qi
 10.1
Hamilton–Jacobi theory
• We relate the constants with the initial values of our
old variables:
S (q1 ,..., qM , 1 ,...,  M , t )
pi 0 
qi q q
i i 0 ;t  t 0
• The new coordinates are defined as:

S (q1 ,..., qM , 1 ,...,  M , t )

Qi   i 
 i q q
i i 0 ;t  t 0

• Inverting those formulas we solve our problem

qi  qi (1 ,...,  M , 1 ,...,  M , t )

pi  pi (1 ,...,  M , 1 ,...,  M , t )
 10.1
Have we met before?
dS S S
 qi 
dt i qi t
S
pi  H  S / t  0
q i

dS
  pi qi  H  L S   Ldt  const
dt i

• Remember action?

t2
I   Ldt  S (t2 )  S (t1 )
t1
 10.1
Hamilton’s characteristic function
• When the Hamiltonian does not depend on time
explicitly
dH / dt  H / t  0
• Generating function (Hamilton’s characteristic
function)
F2  W (q1 ,..., qM ,1 ,..., M )
W W
Qi    W W
Pi  i H  H (q1 ,..., qM , ,..., )
W q1 qM
pi 
qi
W W    pi dqi  const
qi   pi qi
dW

dt i qi i i
 10.3
Hamilton’s characteristic function
W W
• Now we require: H (q1 ,..., qM , ,..., )  1
q1 qM
• So:
W K  K
KH K  1 
Pi   ; Qi 
t Qi Pi
Pi  0 Pi   i ;

Q i   i1 ; Q1  t  1  W / 1;
Qi  i  W /  i ; i  1
• Detailed comparison of Hamilton’s characteristic vs.
Hamilton’s principal is given in a textbook (10.3)
 10.3
Hamilton’s characteristic function
• What is the relationship between S and W ?

H  S / t  0  K S H  1  KW

• One of possible relationships (the most

conventional):

S (q1,..., qM ,1 ,..., M , t )  W (q1 ,..., qM ,1 ,..., M )  1t

H  1  0
 10.6
Periodic motion
• For energies small enough we have periodic
oscillations (librations) – green curves

• For energies great enough we msy have periodic

rotations – red curves

• Blue curve – separatrix trajectory – bifurcation

transition from librations to rotations
 10.6
Action-angle variables
• For either type of periodic motion let us introduce a
new variable – action variable (don’t confuse with
J   pdq
action!):

H ( q, p )   p  p ( q,  ) J  J ( ) W  W ( q, J )
• A generalized coordinate conjugate to action
variable is the angle variable: W
w
J
• The equation of motion for the angle variable:

H ( J ) S   Ldt  const
w   v( J )  const
J W    pi dqi  const
i
 10.6
Action-angle variables
w  v w  vt  
• In a compete cycle
w W 2
 W d
w   dq   dq   dq   pdq
q qJ J q dJ
d
 J  1  (v(t   )   )  (vt   ) 1  v
dJ
v  1/ 
• This is a frequency of the periodic motion

W W
w p J   pdq
J q
 10.2
Example: 1D Harmonic oscillator
H
1
2m
 p  m  q   E;
2 2 2 2
  k/m
2

1   W  
2
W 2 2 2
p H    m  q  
q 2m   q  
 
m q2 2
W  2m  1 dq
2
m 2 q 2
S  2m  1  dq  t
2
S m dq 1 m2



2  m 2 q 2
 t  arcsin q
 2
t
1
2
 10.2
Example: 1D Harmonic oscillator
1 m 2
2
  arcsin q t q sin(   t )
 2 m 2

W 2 2 2  2m cos(  t )
p  2m  m  q
q
( p 2  m2 2 q 2 ) / 2m  E    E

2E
q sin(   t ) p  2mE cos(  t )
m 2

W  2m  1
m 2 q 2
dq
mq0 / p0  tan(   t0 )
2
 10.6
Action-angle variables for 1D harmonic
oscillator 2
q
W m
sin z
2

p  2m  m  q
2 2 2

q
2
2 2
J   pdq   2m  m  q dq   cos zdz 
2 2 2 2

 0 
J
H  
2
• Therefore, for the frequency:

H  k /m
v  
J 2 2
 10.4
Separation of variables in the Hamilton-
Jacobi equation
• Sometimes, the principal function can be
successfully separated in the following way:
S   Si (qi , 1 ,...,  M , t )
i
Si Si
H i (qi , , 1 ,...,  M , t )  0
qi t
• For the Hamiltonian without an explicit time
dependence:
Wi
Si  Wi   i t H i (qi , , 1 ,...,  M )   i
qi
• Functions Hi may or may not be Hamiltonians