Sie sind auf Seite 1von 40

Probability and Stochastic processes

The Liouville equation


Brownian motion
The Master equation

Overview of the course


Complex systems: many particle systems with complex
patterns of interactions
Objective: physical theory that gives quantitative predictions
to be confronted with observations and experiments
Need mathematical model that indicates how some variables
evolve in time and their connection to measureable quantities
Randomness can appear in several ways
Finite precision on initial conditions (important with sensitive
dependence on initial conditions) e.g. coin tossing
Lack of information about all relevant variables or inability to
process them e.g. Brownian motion
Stochastic character of evolution laws e.g. animal behaviour
(arguably depending on physical and chemical processes that
constitute its brain and body, but not directly derivable from
them)
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Assume that individual degrees of freedom behave randomly


according to certain probabilistic rules
Consider many identical copies of the same system with
different realizations of the randomness: ensemble
Expect averages over ensembles exist and can be calculated
Get statistical properties of the motion, that can be
experimentally investigated by repeating experiments many
times (or making observation time very long)
Stochastic models fully described by probability distribution to
find system at time t in a certain configuration s
In thermal equilibrium probability distribution is given by the
Gibbs-Boltzmann p(s) eH(s)/T : equilibrium models are
defined by an energy function s H(s) (no notion of time)
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Non-equilibrium models are defined by a set of transition rates


and the probability distribution is obtained by solving the
Master equation
Analytical solutions of ME hardly available, two strategies:
numerical integration or Perturbative theory (e.g. Van
Kampen system size expansion, Kramers-Moyal expansion etc)
to cast ME into Fokker-Planck eqn
For many-particle systems often possible to deduce equations
for a small set of (macroscopic) variables that follow
approximately a deterministic law
eliminated variables are felt as a superimposed effective noise,
often referred as fluctuations (basis of Langevin approach)
Stochastic approach needed to study fluctuations (important
on nanoscales) and to determine range of validity of
macroscopic laws
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Stochastic dynamics: Objectives


At the end of this section youll be able to:
1

3
4
5

Derive the Liouville equation for systems evolving


deterministically
Write the Chapman-Kolmogorov equation for Markov
processes
Derive the Master equation for Markov processes
Use ME to derive equations for the average and fluctuations
Use detailed balance to prove convergence to equilibrium of
ergodic systems
Find the solution of a master equation via spectral
decomposition for systems which satisfy detailed balance
understand the difference between equilibrium and steady
states
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Outline
1

Probability and Stochastic processes

The Liouville equation

Brownian motion
Dimansional analysis and Scaling

The Master equation


Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Probability density
Stochastic variable X = variable whose value is unknown
Stochastic process X(t) = time evolution of stochastic var
Consider system which can be described in terms of X
P1 (x, t) = prob. density that X has value x at time t
P2 (x1 , t1 ; x2 , t2 ) = prob. density that X has value x1 at time
t1 and x2 at time t2
Pn (x1 , t1 ; . . . ; xn , tn ) = prob. density that X has value x1 at
time t1 , . . . , and xn at time tn
Pn 0 n (Non-negative)
R
dxn Pn (x1 , t1 ; ...; xn1 , tn1 ; xn , tn ) =
Pn1 (x1 , t1 ; ...; xn1 , tn1 ) (Marginal)
R
dx1 P1 (x1 , t1 ) = 1 (Normalization)
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Moments
Time-dependent moments
Z
hx(t1 )x(t2 )...x(tn )i = dx1 dx2 ...dxn Pn (xn , tn ; ...; x2 , t2 ; x1 , t1 )x1 x2 ...xn
Stationary processes:
Pn (x1 , t1 ; x2 , t2 ; ...; xn , tn ) = Pn (x1 , t1 +T ; x2 , t2 +T ; ...; xn , tn +T ) n, T

P1 (x1 , t1 ) = P1 (x1 )

(hx1 (t1 )i = M )

hx1 (t1 )x2 (t2 )i = C(|t1 t2 |)

Equilibrium stationarity
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Connected correlator
If value of x2 at t2 independent of x1 at t1
P2 (x1 , t1 ; x2 , t2 ) = P1 (x1 , t1 )P1 (x2 , t2 )
Z
hx1 (t1 )x2 (t2 )i = dx1 dx2 P2 (x2 , t2 ; x1 , t1 )x1 x2
Z
=

dx1 dx2 P1 (x2 , t2 )P1 (x1 , t1 )x1 x2 = hx1 (t1 )ihx2 (t2 )i

Connected correlator
hx1 (t1 )x2 (t2 )i hx1 (t1 )ihx2 (t2 )i
measures degree of correlation between two measures taken at
different times
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Conditional probability
P1|1 (x2 , t2 |x1 , t1 ) = conditional prob. dens. for X to have
value x2 at t2 given it had value x1 at t1
defined by Bayes:
P2 (x2 , t2 ; x1 , t1 ) = P1|1 (x2 , t2 |x1 , t1 )P1 (x1 , t1 )
properties:
R

dx1 P1|1 (x2 , t2 |x1 , t1 )P1 (x1 , t1 ) = P1 (x2 , t2 )

dx2 P1|1 (x2 , t2 |x1 , t1 ) = 1

Joint conditional prob. density


Pk|` (x`+1 , t`+1 ; . . . ; x`+k , t`+k |x1 , t1 ; ...; x` , t` )
=

Pk+` (x1 , t1 ; ...; x` , t` ; x`+1 , t`+1 ; . . . ; x`+k , t`+k )


P` (x1 , t1 ; . . . ; x` , t` )
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Markov processes
Markov property:

transition probability

z
}|
{
P1|n1 (xn , tn |xn1 , tn1 ; ...; x1 , t1 ) = P1|1 (xn , tn |xn1 , tn1 ),

t1 < ... < tn

Markov process fully determined by P1 and P1|1


P3 (x1 , t1 ; x2 , t2 ; x3 , t3 ) = P2 (x1 , t1 ; x2 , t2 )P1|2 (x3 , t3 |x1 , t1 ; x2 , t2 )
= P1 (x1 , t1 )P1|1 (x2 , t2 |x1 , t1 )P1|1 (x3 , t3 |x2 , t2 )

Int. over x2 , divide by P1 (x1 , t1 ): Chapman-Kolmogorov eqn


Z
P1|1 (x3 , t3 |x1 , t1 ) =

dx2 P1|1 (x3 , t3 |x2 , t2 )P1|1 (x2 , t2 |x1 , t1 )


Z

For t1 = t2 = t :

P1|1 (x3 , t3 |x1 , t) =


satisfied by

dx2 P1|1 (x3 , t3 |x2 , t)P1|1 (x2 , t|x1 , t)

P1|1 (x2 , t|x1 , t) = (x2 x1 )


A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Deterministic evolution
Assume system can be described by set of variables X
evolving according to autonomous ODE (Newtons,
Hamiltons, Schrodingers etc)
d
X(t) = f (X(t))
dt

e.g. classical system, N particles, 3-dim; q = (q1 , . . . , q3N ),


p = (p1 , . . . , p3N ), briefly X = (q, p)
Set of possible values of X determines phase space
Each possible state of the system determines a point in phase
space
Deterministic evolution: state X at time t univocally assigned
from initial state X 0 at time 0 as X(t, X 0 ), solution of ODE
with initial condition X(0) = X 0
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Randomness in deterministic evolution


1-dim for simplicity
d
dt x(t)

= f (x(t))
x(0) = x0

x(t, x0 )

Deterministic evolution so
P (x, t|x0 , 0) = (x x(t, x0 ))

However, x0 determined through measurements, subjects to


errors and finite precision
initial conditions should not be given as a point in phase
space, but as distribution P (x0 , 0)
P (x, t) =

(x x(t, x0 ))P (x0 , 0) = h(x x(t, x0 ))i

x0

A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

The Liouville equation


P (x, t)
t

dx(t, x0 )

i
h(x x(t, x0 ))i = h(x x(t, x0 ))
x
dt
t

= h(x x(t, x0 ))f (x(t, x0 ))i = [h(x x(t, x0 ))if (x)]


x
x
=

Distribution in phase space evolves according continuity


equation

P (x, t)
=
[P (x, t)f (x)]
t
x

In some situations if initial condition is sharply peaked around


x0 , P (x, t) remains peaked around x(t, x0 )
But this is not true for sensitive dependence on initial
condition and stochastic dynamics
probabilistic approach required when P (x, t) spreads over time
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Dimansional analysis and Scaling

The Brownian motion


historically first phenomenological theory of how fluctuating
phenomena arise
observation by Brown in 1827: a pollen grain suspended in
water is found in very animated and irregular motion
Spectacular evidence on macroscopic scale for discrete or
atomic nature of matter on the micro-scale
Paradigm theory for many-body systems in classical statistical
mechanics (noise, thermal bath, separation of time scale
between degrees of freedom, fluctuation-dissipation etc.)
first explanations:
Einstein (1905), Smoluchowski (1906): neglect inertia
Langevin (1908): account for inertia

1950s: clear that can apply theory of Brownian motion to any


observable in a macroscopic system generalized BM
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Dimansional analysis and Scaling

Einsteins explanation
motion caused by frequent impacts on the pollen grains by
molecules of the liquid
too complicated statistical descritpion
Assumptions:
motion of each particle independent of others
motion of the same particle in successive time intervals are
independent (timpact   tobs )

Simple: isotropy; can in fact look at one dimension


x(t + ) = x(t) + (t)

random, distributed according to () = ()


Z

Z
d () = 1,

Z
d () = 0,

induces distribution of x, P (x, t), with


A Annibale

d 2 () = a2

dx P (x, t) = 1

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Dimansional analysis and Scaling

Diffusion equation
Markov assumption
Z
P (x, t + ) =

dx0 P (x0 , t)(x x0 ) =

Z
d P (x , t)()

decays very rapidly, P broad


Z
P (x, t + ) =

d ()[P (x, t)

1
2

P (x, t) + 2 2 P (x, t) + . . .]
x
2
x

1
2
= P (x, t) + a2 2 P (x, t)
2 x

small P (x, t + ) = P (x, t) + P


t
P (x, t)
2 P (x, t)
=D
,
t
x2
A Annibale

1
0 2

D = lim

d ()2

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Dimansional analysis and Scaling

Solution by Fourier Transform


Solve in Fourier space for P (x, 0) = (x)
Z

dx P (x, t)eiqx ,

G(q, t) =

G(q, 0) = 1

2
2
G(q, t)
= Dq 2 G(q, t) G(q, t) = eDq t G(q, 0) = eDq t
t
Z
2
1
1
P (x, t) =
dq G(q, t)eiqx =
ex /4Dt
2
4Dt

Moments
hxi = 0,

hx2 i = 2Dt

one of central results in statistical physics: x(t)


diffusion
A Annibale

t for

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Dimansional analysis and Scaling

Contains many major concepts central in dynamical analysis of


stochastic processes
independence of the pushes on the previous history:
Markovian property
Kramers-Moyal expansion: approximation which effectively
replaces a process whose sample path need not be continuous
with one whose paths are continuous.
Fokker-Planck equation: equation for the probability
distribution to find the system in a certain state
Alternative: regard x(t) as a continuous function of time, but
a random function: write a stochastic differential equation for
the path (initiated by Langevin)
Tools for non-equilibrium analysis, such as Dimensional
analysis and Scaling
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Dimansional analysis and Scaling

Dimensional analysis
Get moments from dimensional analysis without solving eqn
2 P (x, t)
P (x, t)
=D
t
x2

Clearly, hxi = 0, as there is no bias


hx2 i non trivial, should depend on D and t
Let L=unit of length, T =unit of time
[x] = L,

[t] = T,

[hx2 i] = L2

[D] = L2 /T

hx2 i = C Dt

C e.g. from equation for moments


d 2
hx i = 2D
dt

Dimensional analysis works for much more complex problems


Very simple: should be used as first resort
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Dimansional analysis and Scaling

Scaling
R
[P ]
= L1 e.g. from dx P (x, t) = 1
so, Dt P (x, t) dimensionless

using x, t, D, can form dimensionless quantity = x/ Dt

Dt P (x, t) = ()

Scaling ansatz
1
P (x, t) = ()
Dt

Plug into eqn, get an ODE


200 + 0 + = 0
P (x, t) =
A Annibale

() = (4)1/2 e

/4

2
1
ex /4Dt
4Dt

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

The Master equation


Define transition probability per unit time
Wt (x0 |x) =

P1|1 (x0 , t + |x, t)


| =0

Assume can Taylor expand P1|1 (x0 , t + |x, t) for small


P1|1 (x0 , t + |x, t) = (x0 x) + Wt (x0 |x) + O( 2 )

Need correction to preserve normalization


Z

P1|1 (x0 , t + |x, t) = c(x0 x) + Wt (x0 |x) + O( 2 )


Z
dx0 P1|1 (x0 , t + |x, t) = 1 c = 1 dx0 Wt (x0 |x)

Define escape rate a(0) (x, t) =

dx0 Wt (x0 |x)

so

h
i
P1|1 (x0 , t + |x, t) = 1 a(0) (x, t) (x0 x) + Wt (x0 |x) + O( 2 )
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

The Master equation - 2


Recall

P2 (x,t+ ;x0 ,t0 )

Z
P1 (x, t + ) =

z
}|
{
dx0 P1 (x0 , t)P1|1 (x, t + |x0 , t)

For small
P1 (x, t + ) = P1 (x, t)[1 a(0) (x, t)] +

dx0 P1 (x0 , t)W (x|x0 )

For 0 get continuous time Master equation

P1 (x, t) =
t

dx0 [W (x|x0 )P1 (x0 , t) W (x0 |x)P1 (x, t)]

For discrete set of states, let pn (t) = P1 (n, t)


dpn (t) X
=
[Wn0 n pn0 (t) Wnn0 pn (t)]
dt
0
n

A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Comments
Master Equation is a gain-loss equation
Not invariant for t t irreversible dynamics towards
steady state where transitions cannot cause further changes to
probability distribution (P/t = 0)
Broad applicability (Markov process), only needed transition
probability over short time
ME also applies to all transition probabilities P1|1 (x, t|x0 , t0 )
Given X(t), defined by P1 (x1 , t1 ) and P1|1 (x2 , t2 |x1 , t1 ), can
always extract sub-ensemble X ? (t) with
P1? (x1 , t1 ) = P (x1 , t1 |x0 , t0 ) and
? (x , t |x , t ) = P (x , t |x , t ),
P1|1
2 2 1 1
1|1 2 2 1 1
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Equation for the average


for arbitrary function f (x):
Z

Z
P1 (x, t)
dx = f (x)[W (x|x0 )P1 (x0 , t)
t
Z
W (x0 |x)P1 (x, t)]dxdx0 = [f (x0 ) f (x)]W (x0 |x)P1 (x, t)dxdx0

d
hf (x)i =
dt

f (x)

f (x) = x, def a() (x, t) =


dhxi
=
dt

dx0 (x0 x) Wt (x0 |x) = 0, 1...

dx a(1) (x)P1 (x, t) = ha(1) (x, t)i

if a(1) is linear, ha(1) (x, t)i = a(1) (hxi, t), eqn closes
t hxi = a(1) (hxi, t) deterministic, ODE

if a(1) not linear, expand


aboutDynamical
hxi Analysis of Complex Systems CCMCS04
A Annibale

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Equation for the variance


If a(1) not linear need higher order moments. For f (X) = X 2
d
hX 2 i =
dt
Z
=

dx dx0 (x02 x2 )W (x0 |x)P1 (x, t)

dx dx0 [(x0 x)2 + 2x(x0 x)]W (x0 |x)P1 (x, t)


= ha(2) (X)i + 2hXa(1) (X)i

but generally depends on higher order moments..


often approximation required to close eqns e.g. neglect
fluctuations hX 2 i ' hXi2
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Bra and Ket


Reminder: for vector A in 3-dim Euclidean space have

A1
A = A1 e1 + A2 e2 + A3 e3 = A2
A3

em en = mn

Gen. to N -dim vector space over complex numbers, A C N

ket :

|Ai = A1 |1i + A2 |2i + . . . + AN |N i =

Hermitian conjugate |Ai+ = hA|

A1
A2
..
.

AN

hA| = A?1 h1|+A?2 h2|+. . .+A?N hN | = (A?1 , A?2 , . . . , A?N )


P
Inner (gen. scalar) product: hA|Bi hA|Ai = i |Ai |2
P
P
Identity operator 1: |Ai = i hi|Ai|ii
i |iihi| = 1

bra :

A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Master equation in Dirac notation


pn (t) =

[Wmn pm (t) Wnm pn (t)]

Define Lnm = Wmn mn

pn (t) =

n0

Wnn0
Lnm pm (t)

Associate with
P each possible configuration n = 1 . . . M a basis
vector |ni:
hm|ni = mn
n |nihn| = 1;
regard pn (t) as n-th component of a state vector |P (t)i

|P (t)i =

p1 (t)
p2 (t)
..
.

pN (t)

so, |P (t)i =

n pn (t)|ni,
A Annibale

i.e.

pn (t) = hn|P (t)i

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

L matrices
Recall
pn (t) =

Lnm pm (t)

Rates collected in matrix


L=

Lnm |nihm|,

hn|L|mi = Lnm

mn

t |P (t)i

= L|P (t)i

P
projection state vector hI| = (1, . . . , 1) = n hn|
P
1 = n pn (t) = hI|P (t)i t hI|P (t)i = 0 hI|L = 0
= 0 is always an eigenvalue with left eigenstate hI| and
right eigenstate |Peq i, as t |Peq i = 0
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

How do we choose the rates?


if dynamics converges to equilibrium steady state
equilibrium probabilities pn pn () generally known, e.g.
canonicalPensemble
|Peq i = n pn |ni indep on time, so L|Peq i = 0
also, equilibrium dynamics is reversible, i.e. a trajectory over
time t = mt, n0 . . . nm is as likely as nm . . . n0
this requires Detailed Balance (DB), i.e. no net probability
current:
Wmn pm = Wnm pn
n, m
DB guarantes that pn is a steady state:
X
X
Wnm pn =
Wmn pm

pn = 0

but is a stronger condition


A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Metropolis and Glauber rates


Focus on canonical ensemble, pn = Z 1 eH(n)
DB: exp[H(n)] Wnm = exp[H(m)] Wmn
choice not unique
Metropolis rates

Wnm =

1
e[H(m)H(n)]

Glauber
Wnm =

A Annibale

H(m) H(n)
H(m) > H(n)

1
1+

e[H(m)H(n)]

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Convergence to equilibrium
If rates are in detailed balance with pn can show for ergodic
systems that pn (t) pn from any pn (0)
Focus on canonical ensemble,
pn () =

1 H(n)
e
Z

Kullback-Leibler distance D(p||q) =


F (t) =

pn (t) ln

pn
n pn ln qn

X
pn (t)
=
pn (t)[ln pn (t) + H(n) + ln Z]
pn ()
n

F (t) is a Liapunov function


F (t) 0 (= 0 iff pn (t) = pn ())
F 0 (t) 0 (= 0 iff pn (t) = pn ())
dF
dt

n [ln pn (t)

n
+ H(n) + 1] dp
dt
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics
0

Use Master equation and DB: Wn0 n eH(n ) = Wnn0 eH(n)


dF
1X
=
Wnn0 eH(n) [(ln pn (t)+H(n))(ln pn0 (t)+H(n0 ))]
dt
2 0
nn

[eH(n )+ln pn0 (t) eH(n)+ln pn (t) ] 0

Used identity: (ex ey )(x y) 0 (x, y), equality iff x = y


System must reach a point where
0

eH(n) pn = eH(n ) pn0


pn = (n)e

H(n)

or Wnn0 = 0

, with (n) = (n0 ) n, n0 : Wnn0 6= 0

Sn set of states dynamically accessible from n:


(n0 ) = Zn1 n0 Sn
ergodic: (n) = Z 1 n Boltzmann unique equilibrium
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Steady state
Solution of ME hardly available
Formal solution to ME: |P (t)i = eLt |P (0)i
Need to diagonalize L: usually non-trivial, L non-hermitian
Easy when DB holds
stationary distribution of ME can be obtained by iterating
p2 =

W21
p1 ;
W12
with

p3 =
X

W32 W21
p1 ;
W23 W12

...

pn = 1

non-stationary distribution pn (t) can be otained from spectral


decomposition of ME
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Simmetrizing L
d|P (t)i
= L|P (t)i
dt
r
r
pn
pm
= pm Wmn
Wnm =
Wmn
pm
pn

ME :
DB :

pn Wnm

always possible to rewrite ME in terms of symmetric matrix U


r
Umn =

pn
Lmn =
pm

X
pn
Wnm mn
Wnn0 = Unm
pm
0
n

for transformed distributions


pn (t)
pn (t) = ;
pn

pn (t) = hn|P (t)i,

d|P (t)i
= U |P (t)i
dt
A Annibale

Unm = hn|U |mi

|P (t)i = eU t |P (0)i

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Spectral decomposition of ME
U symmetric complete orthonormal set of eigenvectors
hi |U = i hi |,

U |i i = i |i i,

|P (t)i = eU t |P (0)i =

i = 0, . . . , M 1

ei t |i ihi |P (0)i

XX
=
ei t |i ihi |mihm|P (0)i
i

pn (t) =

XX
i

pn (t) =

ei t hn|i ihi |mi


pm (0)

M
1 X
M
X
i=0 m=1

pn
pm (0)hi |miei t hn|i i,
pm

A Annibale

i 0 i

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Solution to the ME
Stationary solution
let (0) = 0 :

M r
X
pn
pm (0)h0 |mihn|0 i
t
p
m
m=1
X

pm (t) = 1 t hm|0 i = h0 |mi = pm

pn = lim pn (t) =

Time-dependent solution
pn (t) = pn +

M
1 X
M
X

i=1 m=1

pn
pm (0)hi |miei t hn|i i
pm

Equilibrium given by eigenvector with zero eigenvalue


Relaxation time given by second smallest eigenvalue (in
absolute value) = 1/mini>0 |(i) |
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Non-equilibrium dynamics
Some systems reach steady state (SS), but this does NOT
satisfy DB with L (e.g. biased random walker on cycle)
X
m

Wmn pm =

Wnm pn

but Wmn pm 6= Wnm pn

These are non-equilibrium systems: for these, SS can NOT in


general be described by Gibbs distribution
P eq(C) = Z 1 eE(C)/T ,

Z=

eE(C)/T

non-eq. systems do not normally have an energy function


E(C), but are described by their transition rates
lack of energy function common in irreversible processes,
where basic dynamical quantities, e.g. number of particles are
not conserved (aggregation, fragmentation, adsorption)
however possible sometimes to construct energy function
which describes SS (conservative process, e.g. biased RW)
A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Choice of rates makes part of the modelling


If DB does not hold, time-dependent solution available only in
special cases
when not, a numerical integration might be possible
or Perturbation theory: Kramers-Moyal (small jump)
expansion, Van Kampens (large size) expansion etc

A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Textbooks

N.G. Van Kampen (2007)


Stochastic Processes in Physics and Chemistry, Elsevier, 3rd
Edition
Linda E Reichl
A Modern Course in Statistical Physics, Wiley-VCH 2009

A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Probability and Stochastic processes


The Liouville equation
Brownian motion
The Master equation

Derivation
Equations for the moments
Dirac notation
Equilibrium dynamics
Spectral decomposition
Non-equilibrium dynamics

Recent research papers

P Moretti, A Baronchelli, A Barrat, R Pastor-Satorras (2011)


Complex Networks and Glassy Dynamics: walks in the energy
landscape J. Stat. Mech. P03032
B Waclaw, RJ Allen, M Evans (2010)
A dynamical phase transition in a model for evolution with
migration, Phys. Rev. Lett. 105, 268101
J Currie, M Castro, G Lythe, E Palmer, C Molina-Paris (2012)
A stochastic T cell response criterion JR Soc Interface
9:2856-2870

A Annibale

Dynamical Analysis of Complex Systems CCMCS04

Das könnte Ihnen auch gefallen