Sunteți pe pagina 1din 4

Questions for the exam:

Nonequilibrium statistical mechanics


Christian Maes
Instituut voor Theoretische Fysica, KU Leuven, Belgium
The exam is oral and open book. You can bring your notes and answers. Here
are the questions. During the exam further background and general ideas will be
discussed.
1. We want to discuss the Clausius heat theorem. What is the thermodynamic formu-
lation? How can you give a simple and yet precise version using Markov processes?
Show the calculation for a specic system of your choice, e.g. a two-state Markov
process. Check also in general and explicitly that

dE(, .)
d
d
E(, )

= d [E + log Z
,
]
where averages are taken with respect to
eq(,)
(x) = e
E(x,)
/Z
,
. Convince yourself
that this equality conrms the Clausius heat theorem for quasistatic processes.
2. Present evidence or proof that the Kac ring model relaxes to equilibrium. State
carefully what you mean by that, and discuss the solutions of the Zermelo and of
the Loschmidt paradoxes.
3. Discuss an H-theorem for the linear diusion equation
m
t
(x)
t
= Dm

t
(x)
with diusion constant D > 0, x [0, 1], m
t
(0) = a, m
t
(1) = b and m
t
(x) 0. We
need to understand both the cases a = b and a = b. Is there a relation between the
Lyapunov functional and the Boltzmann entropy?
4. Explain why for the Boltzmann equation, the H-functional
H(t)

R
3
d
3
v f(v, t) log f(v, t), t 0
2
is monotone in time (a Lyapunov funnction). We take zero external force F = 0 for a
homogeneous system with f(r, v, t) = f(v, t) in the Boltzmann equation:
d
dt
H(t) 0
The result follows from a specic calculation, that you should give, but also from a
more general argument.
5. Explain how the random walk converges to the linear diusion equation in a con-
tinuum limit.
6. Using the linear Langevin dynamics with standard white noise (t),
m
d
dt
v(t) = m v(t) +

2D
v
(t)
compute its exact solution that gives for example expressions for the covariance
v(t
1
)v(t
2
) =
D
v
m
2

e
|t
1
t
2
|
e
(t
1
+t
2
)

+ v
2
0
e
(t
1
+t
2
)
or the position variance
x
2
(t) =
2k
B
T
m
t +
k
B
T
m
2

4e
t
e
2t
3

+ v
2
0

(e
t
1)

2
say for x
0
= 0.
7. Using a Dyson series one can obtain the Agarwal-Kubo formula for linear response.
Show how you get from that AgarwalKubo formula to the standard Kubo formula of
linear response around equilibrium, using a Markov diusion dynamics. What do we
mean by the uctuationdissipation theorem?
8. Show the validity of the Einstein relation between mobility and diusion constant using
the Langevin dynamics. Can you also derive that relation in the frequency domain?
Start from
d
dt
v
t
= v
t
+
1
m
F(t) +

2
m
(t), t > 0
with (t) standard white noise and F(t) some time-dependent external force. Show
that
R() =
1
m
1
i
3
with imaginary part
R() =

2
G()
entering
v(t) =

dsR(t s) F(s)
and where G is the Fourier transform of the velocity autocorrelation function
G() =

dte
it
v(t) v(0)
eq
9. Use the path-space formalism for jump processes to discuss the linear response for
3state Markov processes. Take for example a dynamics on {a, b, c} with
k(a, b) = 1, k(b, a) = , k(a, c) = Ee
E
, k(c, a) = Ee
E
, k(b, c) = 1, k(c, b) =
with > 0, E 0. Start at time zero in its unique stationary distribution. Give an
expression (general formula parameterized by and E) for the probability at time
t > 0 that we are in state a to rst order in h > 0 if we perturb the above dynamics
by changing E E + h for all positive times. Show that we get back to equilibrium
response for E = 0.
10. Let X be a random variable taking values in [[1, K]], with probabilities p
k
, and take
N independent copies of X. Compute the large deviation function and the generating
function of the cumulants for the empirical vector
{V
k
=
1
N
N

i=1

X
i
,k
}
and check that they are Legendre transforms of one another. What should we do
to them if we want the large deviation function and the generating function of the
cumulants of the average of a quantity f instead, dened by
a =
K

k=1
f(k)V
k
?
Can you derive the Legendre transform property between those two from the previous
case ? As an application, consider the binomial distribution with parameter x
p
k
=
K!
k!(K k)!
x
k
(1 x)
1k
and compute the large deviation function of the average of f(k) = k in two dierent
ways (with and without going through the generating function of its cumulants).
4
11. Consider a two-state model 0 1 with transition rates (from 0 to 1) and (from
1 to 0). Write the generating function of the cumulants of:
the empirical vector at a given time averaged over N independent copies of the
system;
the empirical vector averaged over a long time in one realisation of the process.
How do these quantities relate to each-other ?
12. Write the deformed Markov matrix that measures the particle current in the periodic
ASEP of size L. How does it relate to that for the entropy production ? What
property of the generating function of the cumulants of the current does it imply ?
What happens to that property in the limit of q 0 ? And q 1 ? And L ?

S-ar putea să vă placă și