You are on page 1of 53

SLE

The Stochastic Liouville Equation


How to Win Friends and Influence Functionals

Gerard McCaul
Kings College London
gerard.mccaul@kcl.ac.uk

July 17, 2017

Gerard McCaul | Kings College London | July 17, 2017 1/1


SLE

Introductions

lev2.jpg chris.jpg
Lev Kantorovich
Chris Lorenz
CANESlogo.png

Gerard McCaul | Kings College London | July 17, 2017 2/1


SLE

This talk

This talk is designed to sketch the derivation and implementation of


the Stochastic Liouville Equation (SLE)
The SLE is an exact description for the dynamics of arbitrary open
quantum system interacting with a harmonic environment
This is a conceptually simple, numerically implementable result
It is also possible to generalise this result to arbitrary initial
conditions.

Gerard McCaul | Kings College London | July 17, 2017 3/1


SLE

Gerard McCaul | Kings College London | July 17, 2017 4/1


SLE | Open Quantum Systems

Gerard McCaul | Kings College London | July 17, 2017 5/1


SLE | Open Quantum Systems

A little motivation

Quantum theory is very good a describing the behaviour of closed


systems.
Very few systems are truly closed.
What can we do if we want to describe open, dissipative systems?
Is it possible to describe the effect of the environment on an arbitrary
quantum system in simple terms?

Gerard McCaul | Kings College London | July 17, 2017 6/1


SLE | Open Quantum Systems

Serious attempts to incorporate dissipation into quantum theory begin


with the work of Kanai (1948), Stevens (1958) and Kerner(1958).
The models proposed to describe these dissipative cavities were based
on the classical damped harmonic oscillator.
As expected, this model once quantised gives observable expectation
values that decay.
Unfortunately, later work by Senitzky (1960) showed that any
phenomenological model with damping leads to a decaying
fundamental commutation relation.

Gerard McCaul | Kings College London | July 17, 2017 7/1


SLE | Open Quantum Systems

Serious attempts to incorporate dissipation into quantum theory begin


with the work of Kanai (1948), Stevens (1958) and Kerner(1958).
The models proposed to describe these dissipative cavities were based
on the classical damped harmonic oscillator.
As expected,
Damped HarmonicthisOscillator
model once quantised gives observable expectation
values that decay.
mx + x = m 2 x
Unfortunately, later work by Senitzky (1960) showed that any
phenomenological model with damping leads to a decaying
fundamental commutation relation.

Gerard McCaul | Kings College London | July 17, 2017 7/1


SLE | Open Quantum Systems

Serious attempts to incorporate dissipation into quantum theory begin


with the work of Kanai (1948), Stevens (1958) and Kerner(1958).
The models proposed to describe these dissipative cavities were based
on the classical damped harmonic oscillator.
As expected, this model once quantised gives observable expectation
values that decay.
Unfortunately, later work by Senitzky (1960) showed that any
phenomenological model with damping leads to a decaying
fundamental commutation relation.

Gerard McCaul | Kings College London | July 17, 2017 8/1


SLE | Open Quantum Systems

Serious attempts to incorporate dissipation into quantum theory begin


with the work of Kanai (1948), Stevens (1958) and Kerner(1958).
The models proposed to describe these dissipative cavities were based
on the classical damped harmonic oscillator.
As expected,
Decaying this model once quantised gives observable expectation
Commutation
values that decay.
[x, p] = i~ exp (Rt)
Unfortunately, later work by Senitzky (1960) showed that any
phenomenological model with damping leads to a decaying
fundamental commutation relation.

Gerard McCaul | Kings College London | July 17, 2017 8/1


SLE | Open Quantum Systems

Explicit environments

This is a problem. Is there a model where dissipation (and


fluctuations) can be included, while still preserving
quantum-mechanical rules?
Consider a quantum system where the environment is modelled
explicitly:

Htot (Q, X) = HQ (Q) + HX (X) + HI (Q, X) (1)

Gerard McCaul | Kings College London | July 17, 2017 9/1


SLE | Open Quantum Systems

This mixed system of interest is described by a density matrix tot .


We can trace over the environmental coordinates to obtain the
reduced density matrix:
This provides an effective description of the open system in terms of a
modified closed system:

Reduced Density Matrix


Z
0
dx tot q, x; q 0 , x

(q, q ) =

To summarise:
Find !

Gerard McCaul | Kings College London | July 17, 2017 10 / 1


SLE | The Model

Gerard McCaul | Kings College London | July 17, 2017 11 / 1


SLE | The Model

A Model for the environment

Take a simple model Hamiltonian


Explicitly, we model the environment as one of M harmonic
oscillators with an interaction linear in the oscillator coordinates.
We assume the system is initially partitioned.
To find (t), we evolve tot
0 before tracing over the environment.

Gerard McCaul | Kings College London | July 17, 2017 12 / 1


SLE | The Model

A Model for the environment

Take a simple model Hamiltonian


Model Hamiltonian
Explicitly, we model the environment as one of M harmonic
oscillators with an interaction
M linear in the M oscillator coordinates.
M
1X 2 1X 2 2 X
H
We tot (q, ) = H (q)
assume the system
q + m
is initially +
i ipartitioned.i i f (q) ci i
2 2
i=1 i i
To find (t), we evolve tot
0 before tracing over the environment.

Gerard McCaul | Kings College London | July 17, 2017 12 / 1


SLE | The Model

A Model for the environment

Take a simple model Hamiltonian


Explicitly, we model the environment as one of M harmonic
oscillators with an interaction linear in the oscillator coordinates.
We assume the system is initially partitioned.
To find (t), we evolve tot
0 before tracing over the environment.

Gerard McCaul | Kings College London | July 17, 2017 13 / 1


SLE | The Model

A Model for the environment

Take a simple model Hamiltonian


Explicitly, we model the environment as one of M harmonic
Partitioning
oscillators with an interaction linear in the oscillator coordinates.
tot
We assume the system is initially X
0 = 0partitioned.
0
To find (t), we evolve tot
0 before tracing over the environment.

Gerard McCaul | Kings College London | July 17, 2017 13 / 1


SLE | The Model

A Model for the environment

Take a simple model Hamiltonian


Explicitly, we model the environment as one of M harmonic
oscillators with an interaction linear in the oscillator coordinates.
We assume the system is initially partitioned.
To find (t), we evolve tot
0 before tracing over the environment.

Gerard McCaul | Kings College London | July 17, 2017 14 / 1


SLE | The Model

A Model for the environment

Take a simple model Hamiltonian


Reduced Density matrix
Explicitly, we model the Z environment as one of M harmonic
0
oscillatorswith
t (q, qan dxdx0 dx
) =interaction linear in0 U
dqdq the(q,oscillator
x, t; q, x,coordinates.
0)
We assume the system is initially partitioned.
tot
tot0 0
q 0 , x0 , 0;over
q 0 , x,the
 
To find (t), weevolve
0 0q ,before
q,x; x U tracing t environment.

Gerard McCaul | Kings College London | July 17, 2017 14 / 1


SLE | The Stochastic Liouville Equation

Gerard McCaul | Kings College London | July 17, 2017 15 / 1


SLE | The Stochastic Liouville Equation

We have an expression for the reduced density matrix, but can we


evaluate it?
Feynman & Vernon (1963) developed a powerful formalism for
dealing with systems of this type, using path integrals to describe the
system of interest without reference to the environment.
Z 
0 1
dqdq Dq (t) Dq (t) F q (t) , q 0 (t)
0 0
  
tf q; q =
Z
 
0
 i i  0 
0 q; q exp Sq [q (t)] Sq q (t)
~ ~

Gerard McCaul | Kings College London | July 17, 2017 16 / 1


SLE | The Stochastic Liouville Equation

We have an expression for the reduced density matrix, but can we


evaluate it?
Feynman & Vernon (1963) developed a powerful formalism for
dealing with systems of this type, using path integrals to describe the
systemFunctional
Influence of interest without reference to the environment.
F [q (t) , q 0 (t)] is the  Influence
Z Functional. It characterises the 

0 1
dqdq Dq (t) Dq (t) F q (t) , q 0 (t)
0 0

environments tf effect
q; q on = the system.
Z
 
0
 i i  0 
0 q; q exp Sq [q (t)] Sq q (t)
~ ~

Gerard McCaul | Kings College London | July 17, 2017 16 / 1


SLE | The Stochastic Liouville Equation

We have an expression for the reduced density matrix, but can we


evaluate it?
Feynman & Vernon (1963) developed a powerful formalism for
dealing with systems of this type, using path integrals to describe the
system of interest without reference to the environment.
Z 
0 1
dqdq Dq (t) Dq (t) F q (t) , q 0 (t)
0 0
  
tf q; q =
Z
 
0
 i i  0 
0 q; q exp Sq [q (t)] Sq q (t)
~ ~

Gerard McCaul | Kings College London | July 17, 2017 17 / 1


SLE | The Stochastic Liouville Equation

We have an expression for the reduced density matrix, but can we


evaluate it?
Feynman & Vernon (1963) developed a powerful formalism for
dealing with systems of this type, using path integrals to describe the
system
System of interest without reference to the environment.
Action
Sq [q (t)] is the action  of 1the isolated
Z system evolving
 under the
0
dqdq Dq (t) Dq (t) F q (t) , q 0 (t)
0 0
 
HamiltoniantH q; q =
f tot = Hq
Z
 
0
 i i  0 
0 q; q exp Sq [q (t)] Sq q (t)
~ ~

Gerard McCaul | Kings College London | July 17, 2017 17 / 1


SLE | The Stochastic Liouville Equation

The Influence functional is a path integral which can be evaluated


analytically.
Unfortunately the reduced density matrix is now also expressed as a
path integral, which in general cannot be solved analytically.
Is it possible to use this to find an effective Hamiltonian to describe
the reduced density matrix?
h i
i~t (t) = Hb eff (t) , (t)
?

Gerard McCaul | Kings College London | July 17, 2017 18 / 1


SLE | The Stochastic Liouville Equation

The Influence functional is a path integral which can be evaluated


analytically.
Unfortunately the reduced density matrix is now also expressed as a
path integral, which in general cannot be solved analytically.
TheIsInfluence
it possibleFunctional
to use this to find an effective Hamiltonian to describe
theisreduced
...but horrible.density matrix?
h i
i~t (t) = Hb eff (t) , (t)
?

Gerard McCaul | Kings College London | July 17, 2017 18 / 1


SLE | The Stochastic Liouville Equation

The Influence functional is a path integral which can be evaluated


analytically.
TheUnfortunately the reduced density matrix is now also expressed as a
Influence Functional
path integral, which
 in0 general cannot
 be solved  analytically.
 1  0 
Is it possible toFuseq,this expan
q , qto=find q, qHamiltonian
effective
~
, q to describe
the reduced density matrix?  X
q, q 0 , q = h ij q, q 0 , q i
  
b eff (t) , (t)
i~t (t) = ij H
?

Gerard McCaul | Kings College London | July 17, 2017 18 / 1


SLE | The Stochastic Liouville Equation

More Influence Functional


 0 
The Influence functional isij a path
q, q ,integral
q = which can be evaluated
analytically.
Z ~ Z ~
0 1
 e 0 0
o also fj 0 as a
  
Unfortunately
d the
dreduced
fi ( ) density
Lij matrix
isLnow
ij expressed
path0integral,0 which 2in general cannot be solved analytically.
Is it possible to use this totf find an effective Hamiltonian to describe
Z ~ Z
the reduced i d
density matrix? dt vi (t) Lij (t i ) fj ( )
0 0
h i
Z tf Z tf b eff (t)
i~0 t (t) =
 H 0 ,I (t)
dt vi (t) ? t t Lij t t0 rj t0
  
+2i dt
0 0
Z tf Z tf
1
dt0 vi (t) LR 0 0
 
+ dt ij t t vj t
2 0 0

Gerard McCaul | Kings College London | July 17, 2017 18 / 1


SLE | The Stochastic Liouville Equation

Yet More Influence Functional


The Influence functional is a path integral which can be evaluated
1
vi (t) = fi (t) fi0 (t) and ri (t) =
analytically. fi (t) + f 0i (t)

2
Unfortunately the reduced density matrix is now also expressed as a
path integral, which1 in X general
ei ecannot be1solved analytically.
 
e j
Lij ( ) = coth ~ cosh ( )
Is it possible to use
mi mthis
j to find 2an effective 2 Hamiltonian to describe

the reduced density matrix?
1 X ei ej
Loij ( ) = h sinh (
i )
i~m m 2 , (t)
b eff
t i(t)j = H (t)
?
X ei ej  
R 1 1
Lij (t) = coth ~ cos ( t)
mi mj 2 2

Gerard McCaul | Kings College London | July 17, 2017 18 / 1


SLE | The Stochastic Liouville Equation

The Influence functional is a path integral which can be evaluated


analytically.
Revenge of The Influence Functional
Unfortunately the reduced density matrixis now also
 expressed as a
path integral, which1in general
X eicannot
ej 1
be solved analytically.
LIij (t) = coth ~ sin ( t)
Is it possible to usemthis
i mj to find2

effective2Hamiltonian to describe
an
the reduced density matrix?   
~
1 X eihej cosh i2 + it
Lij (t i ) = i~ (t) = H b eff (t) , (t) 1

mi mjt 2 2 sinh 2 ~
?

Gerard McCaul | Kings College London | July 17, 2017 18 / 1


SLE | The Stochastic Liouville Equation

The Influence functional is a path integral which can be evaluated


analytically.
Unfortunately the reduced density matrix is now also expressed as a
path integral, which in general cannot be solved analytically.
Is it possible to use this to find an effective Hamiltonian to describe
the reduced density matrix?
h i
i~t (t) = Hb eff (t) , (t)
?

Gerard McCaul | Kings College London | July 17, 2017 19 / 1


SLE | The Stochastic Liouville Equation

Yes!

Gerard McCaul | Kings College London | July 17, 2017 20 / 1


SLE | The Stochastic Liouville Equation

Use the Hubbard-Stratonovich Transformation!

The HS transformation can be considered as converting a system of


two body potentials into a set of independent particles in a
fluctuating field.
This removes the temporally non-local terms in the influence
functional, and allows for the identification of effective forwards and
backwards propagators for the system.

Gerard McCaul | Kings College London | July 17, 2017 21 / 1


SLE | The Stochastic Liouville Equation

Explicitly Use the Hubbard-Stratonovich Transformation!


* " #+
X Z tf
The HS transformation
exp i can bedtconsidered
zi (t) ki (t)as converting a system of
two body potentials into 0
i a set of independent particles in a
{z(t)}
fluctuating
field.
1 X Z tf Z tf 
This
= removes
exp the temporally dt non-local terms
dt0 ki (t) Ain the0 influence
 0 
ij t, t kj t

2 allows 0for the 0identification
functional, and of effective forwards and
ij
backwards propagators for  theD system. E
A
ij t, t0
= z
i (t) z 0
j t
{z(t)}

Gerard McCaul | Kings College London | July 17, 2017 21 / 1


SLE | The Stochastic Liouville Equation

Going Forwards by Going Backwards

After the HS transformation, we obtain the Stochastic Liouville Equation


(SLE): h i
i~t e (t) = H b q (t) , (t)

 h i h i 
~
(t) f (q, t) , (t) + (t) f (q, t) , (t)
2 +

Gerard McCaul | Kings College London | July 17, 2017 22 / 1


SLE | The Stochastic Liouville Equation

Going Forwards by Going Backwards

After the HS transformation, we obtain the Stochastic Liouville Equation


(SLE):
Noises h i
i~t e (t) = H b q (t) , (t)
Here and are complex Gaussian coloured noises, with correlation
 i 
functions determined by the influence ~functional.
h i h

(t) f (q, t) , (t) + (t) f (q, t) , (t)
2 +

Gerard McCaul | Kings College London | July 17, 2017 22 / 1


SLE | The Stochastic Liouville Equation

Going Forwards by Going Backwards

Single-trajectory
After density matrix
the HS transformation, we obtain the Stochastic Liouville Equation
(SLE):
e (t) is the single-trajectory reduced h density matrix.i The physical density
i~t e (t)
matrix is found by averaging this=object.
H
b q (t) , (t)

 h i h i 
he
(t)i ~= (t)
(t) f (q, t) , (t) h,i
+ (t) f (q, t) , (t)
2 +
Importantly, any single realisation of e (t) is meaningless

Gerard McCaul | Kings College London | July 17, 2017 22 / 1


SLE | Generating Noise

Gerard McCaul | Kings College London | July 17, 2017 23 / 1


SLE | Generating Noise

The SLE is an exact equation, but this does not mean a priori that it
is possible to generate the required noises numerically.
To proceed well use a linear filtering ansatz, relating the correlation
functions to a linear filtering kernel, e.g:
Z
dt0 G t t0 x t0
 
( ) =
0

x (t) x t0 = t t0

 
Z
0
dt00 G t t00 G t0 t00

  
(t) t =
0

Gerard McCaul | Kings College London | July 17, 2017 24 / 1


SLE | Generating Noise

etaeta.png

Gerard McCaul | Kings College London | July 17, 2017 25 / 1


SLE | Generating Noise

etanu.png

Gerard McCaul | Kings College London | July 17, 2017 26 / 1


SLE | Generalisations of the SLE

Gerard McCaul | Kings College London | July 17, 2017 27 / 1


SLE | Generalisations of the SLE

Lets return to the assumption that the system is initially partitioned. It is


possible to generalise this statement:
H tot
tot
0 =e 6= 0 X
0

Gerard McCaul | Kings College London | July 17, 2017 28 / 1


SLE | Generalisations of the SLE

britain.png

Gerard McCaul | Kings College London | July 17, 2017 29 / 1


SLE | Generalisations of the SLE

A New Initial Condition

Using a path integral expression for the initial density matrix, we find that
the system can be described using the same Liouville equation, but with a
new, more general initial condition that we find via the following
differential equation:

( ) 1h i
= Hq + f(q, )( ) ( ) (2)
~

h(~)i,, = (0) (3)

Gerard McCaul | Kings College London | July 17, 2017 30 / 1


SLE | Generalisations of the SLE

A New Initial Condition

Using a path integral expression for the initial density matrix, we find that
the system can be described using the same Liouville equation, but with a
new, more general initial condition that we find via the following
differential
A note onequation:

In this more general case,
( ) the1 h is not independent i of the noises in the
real-time evolution, = H
and has non-zeroq + f (q, )( )
correlations with( ) (2)
~

h(~)i,, = (0) (3)

Gerard McCaul | Kings College London | July 17, 2017 30 / 1


SLE | Generalisations of the SLE

matrixevolutionwitharrows.jpeg

Gerard McCaul | Kings College London | July 17, 2017 31 / 1


SLE | Generalisations of the SLE

mumu.png

Gerard McCaul | Kings College London | July 17, 2017 32 / 1


SLE | Generalisations of the SLE

cross-corrfirst.png

Gerard McCaul | Kings College London | July 17, 2017 33 / 1


SLE | Generalisations of the SLE

crosscorrtauslice50kruns.png

Gerard McCaul | Kings College London | July 17, 2017 34 / 1


SLE | Generalisations of the SLE

References

R.P. Feynman (1948)


Rev. Mod. Phys. 20, 367

R.P. Feynman, F.L. Vernon Jr. (1963)


Ann. Phys. 24, 118-173

H. Grabert, P. Schramm. G. Ingold (1988)


Physics Reports 3, 115-207

H. Kleinert, S.V. Shabanov (1995)


Phys. Lett. A. 200, 224-232

J. Stockburger (2004)
J. Chem. Phys. 296, 159-169

G.M.G McCaul, C.D. Lorenz, L. Kantorovitch (2017)


Phys. Rev. B 95, 125124

Gerard McCaul | Kings College London | July 17, 2017 35 / 1


SLE | Generalisations of the SLE

Thanks for listening!

Gerard McCaul | Kings College London | July 17, 2017 36 / 1


SLE | Generalisations of the SLE

noiseeffect.png

Gerard McCaul | Kings College London | July 17, 2017 37 / 1

You might also like