Prof.

Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 1

Problem 1: Scale covariance of correlation functions

(a) The power-law correlation function,

C(r) = C0 r−2x , (1)
with r ≡ |r| and a constant C0 > 0, transforms covariantly under scale transformations,
r → b−1 r, C → b2x C. (2)
Show, by applying an infinitesimal scale transformation, that the transformation law (2) can
equivalently be expressed by the flow equation
 
d
r + 2x C(r) = 0. (3)
dr

(b) Now consider a correlation function of the form

 
−2x r r
C(r) = r Φ , , (4)
a ξ
which also depends on the molecular scale a and on the correlation length ξ. The function Φ
expressing these dependencies is called a scaling function. Generalise the scale transformation law
(2) to this case, by including the two additional length scales. Then apply again an infinitesimal
scale transformation to derive the generalized form of the flow equation (3).
(c) The spin correlation function of the Ising model is of the form (4) with a so-called canonical
scaling exponent,  
−2x0 r r d−2
C(r) = r Φ , with x0 = , (5)
a ξ 2
where d is the spatial dimension. This scaling exponent has been derived in the lecture from power
counting. The scaling function has the form
   
r r r r
Φ , =φ exp − . (6)
a ξ a ξ
Consider a measurement of the correlation function, C(r) ∼ r−2x , in the regime a  r  ξ.
Assuming a singular dependence on the molecular scale, φ ∼ a−η , compute the resulting “true”
decay exponent x. Note: in the system to be measured, a has a fixed value.
(d) Recall from elementary statistical mechanics that the spin correlation function is related to the
magnetic susceptibility, Z
χ = C(r) dd r. (7)

Show that the measurable scaling behaviour of the correlation function derived in (c) implies that
the susceptibility grows as a power of the correlation length,
χ ∼ ξζ , (8)
and compute the exponent ζ.

1
To be discussed on: Mon, October 28

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

2
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 2

Problem 2: Fully connected Ising model

In the lecture, we have discussed the mean field approximation for the general ferromagnetic Ising
model with couplings Jrr0 . In one important instance, the so-called infinite-range or fully connected
Ising model, every spin is coupled with all the others with the same strength. As we will show here,
the mean-field approximation becomes exact in this case.

(a) The Hamiltonian of N Ising spins si ∈ {±1} in the fully connected model has the form
N
X
H(s1 , . . . , sN ) = −J si sj . (1)
i,j=1

Since every spin is coupled with every other spin, we have dropped the positions r. Show that
1 PN
H(s) can be written as a simple function of the magnetization density m = N i=1 si (simply
called magnetization from now on). Do you notice any problem with the system-size dependence
of the Hamiltonian? Define a new rescaled Hamiltonian H0 in order to solve this problem.

(b) We now fix the magnetization of the system. Show that the number of states with magnetization
m is  
N
n(m) = (2)
N 1+m
2
Hint: express the magnetization as a function of the number of up spins N+ and down spins N− .
Now use Stirling approximation log(n!) ' n log(n) − n and show that the number of states with
magnetization m can be written as n(m) = eS(m) , with
    
1+m 1+m 1−m 1−m
S(m) = −N kb log + log . (3)
2 2 2 2

(c) UsingPthe results of (a) and (b), and taking the continuum limit, show that the partition function
0
Z = s1 =±1...sN =±1 e−βH (s) can be written as
Z 1
Z= dm e−βF (m) , (4)
−1

and express the free energy F(m) in terms of the energy and the entropy.

(d) To derive the saddle-point approximation for the integral (4), consider a function f : Rd → R that
has a maximum at the position x0 ∈ Rd . By expanding f (x) around x0 , show that
Z 
1 N f (x)
lim log dx e = f (x0 ). (5)
N →∞ N x∈Rd

Hint: try to obtain a Gaussian integral in Eq. (5). Inspired by this argument, discuss what
happens to the system in the thermodynamic limit in terms of the free energy density f = F/N .
What would happen if we kept the unrescaled Hamiltonian H in Eq. (1)?

1
(e) While the internal energy and the entropy of the system are always functions of the magnetization,
the fully connected model is the only case in which the mean field approximation is exact in the
thermodynamic limit. Point out which step of the calculation would be different in the general
case.

Note: The fully connected Ising model has an important application in statistical genetics. Here, m is
a quantitative genetic trait, i = 1, . . . , N are the genomic sites encoding the trait, s1 , . . . , sN are the
sequence letters (alleles) at these sites, and −H(m) is the fitness.

To be discussed on: Mon, November 4

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

2
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 3

Problem 3: An introduction to the 1d Ising Model.

The Ising model is, perhaps, the most studied model in statistical physics. As introduced in the lecture,
it consists of a discrete set of magnetic degrees of freedom (or spins) si . In this case, the system is
uniaxial anisotropic, which means that the spins align along an unique axis (for example the z-axis).
We say then that si = ±1. In the case of a 1d system (an spins chain), the system is completely
solvable (Ising 1926) and this is what we will explore in this exercise. Consider a one-dimensional
Ising model with nearest-neighbor interactions, in the absence of a magnetic field. The Hamiltonian
of this system is
X N
H = −J si si+1 (1)
i=1

with si = ±1 and J > 0. A given spin state consists of domains, i.e., contiguous stretches of up or
down spins. The domains are separated by boundaries called domain walls, which are marked by dots
in Fig. 1. The system has two ground states, in which all spins are aligned and there are no domain
walls.

Figure 1: A configuration of spins (arrows) and of domain walls (dots) in the one-dimensional Ising
model.

(a) Show that the energy of a spin state depends only on the number of domain walls, M , and compute
the excitation energy ∆EM = EM −E0 , where E0 is the ground state energy. Compute the number
of states with M domain walls (note that each configuration of domain walls corresponds to exactly
two spin states).
Hint: The computation is easier if one assumes free boundary conditions for the spins. Alterna-
tively one can assume periodic boundary conditions ( positions N + 1 and 1 are identical).This
forces an even number of domain walls. The leading behavior in the thermodynamic limit (N  1)
will not depend on the boundary conditions.

(b) Compute the partition function Z at inverse temperature β as a sum over domain wall configura-
tions.

(c) Compute the mean density of domain walls, ρ = hM/N i, the mean energy density u = U/N , and
the specific heat per unit length, c, at temperature β −1 . Evaluate the asymptotic behavior of
these quantities in the low-temperature limit.

1
(d) The spin-spin correlation function is defined by
N
X
C(r) ≡ N −1 hsi si+r i. (2)
i=1

Using the domain wall representation, show that this correlation depends exponentially on the
distance,
C(r) ∼ exp(−|r|/ξ), (3)
and compute the correlation length as a function of the domain wall density, ξ(ρ). Using the result
of (c), then compute ξ(β −1 ) and evaluate the asymptotic behavior in the low-temperature limit.
Hint: First convince yourself that the correlation function can be written as an inhomogeneous
magnetisation with a constraint for one of the spins, sk = 1:

C(r) = hsk+r is =1 ≡ m(r). (4)
k

Define the vector pr with two components p+ − +

r and pr = 1 − pr , which denote the probabilities of
the two spin states at position r. Write down a recursion relation pr = T pr−1 with a 2x2 matrix
T and express the coefficients of this matrix in terms of the domain wall density ρ. Simplify to a
recursion for the position-dependent magnetisation m(r) = p+ −
r − pr and solve this recursion.

(e) More generally, the recursive formalism covers the entire statistics of the system. Consider the
σ,σ 0
partition function ZN in a system of size N with constrained boundary spins, s1 = σ and
0
sN +1 = σ . Compute the 2x2 matrix
 ++
Z1+−

Z1
T= , (5)
Z1−+ Z1−−

which is called the transfer matrix of the system (each of these partition functions consists of a
single term). By complete induction, show that the partition functions in a system of size N with
constrained boundary spins are simply related to the transfer matrix,
 ++ +−

ZN ZN
−+ −− = TN (6)
ZN ZN

(it is useful to write out the matrix product TN explicitly for, say, N = 2, 3). In particular, the
partition function with periodic boundary conditions takes the form ZN = Tr TN . Compute ZN
using the asymptotic identity Tr AN ' λN 1 , where λ1 is the largest eigenvalue of A, and recover
the result of (b).

Problem 4: Mean Field Theory of the XY−model

The Ising model discussed above is a particular case of a broader type of systems with many magnetic
degrees of freedom, in which the spins tend to align along a particular axis, e.g the z−axis, such that
sz = ±1. Alternatively, unit-length spins may prefer to lay down on a plane, e.g. the xy−plane, from
which follows that (sx )2 + (sy )2 = 1. This is known as the planar or XY −model, for which we discuss
the mean field theory in this exercise.

(a) Starting with the Hamiltonian

X X
H=− J(~r − ~r0 )~s~r · ~s~r0 − ~h · ~s~r (7)
~ r0
r,~ ~
r

~ = (M x , M y ), expand the spin vector functions to first

and defining the magnetisation vector M
order around the magnetisation and rewrite the mean field hamiltonian HM F in terms of the norm

2
 Figure 2: At each site in position ~r there is a spin ~s~r that lies down on the xy−plane. In this figure,
we define the angle between ~s~r and the x−axis as φ~r . On the same plane lies the magnetisation vector
M~ . We define the angle between M ~ and the x−axis as θ. For convenience, the external magnetic field
~h lies along the x−axis.

of the magnetisation M and the corresponding

P angles0 θ and φ~r (See the figure above).
Hint: As has been done before, define J ≡ ~r J(~r −~r ). As shown in the picture, for convenience,
one can define the coordinates such that the external magnetic field lies down along the x−axis,
~h = hx̂.

(b) Write the mean field partition function ZM F . In particular, think about how to properly define
the sum over all the configurations in this case.

(c) Using the following integral

Z 2π p
ea cos(θ)+b sin(θ) dθ = 2πI0 ( a2 + b2 ) (8)
0

with I0 (x) the modified Bessel function of the first class, calculate the free energy of the system
FM F .

(d) Argue that the mean field values for the critical exponents of the XY −model are the same as
those for the Ising model that were calculated in the lecture.

To be discussed on: Mon, November 11

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

3
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 4

Problem 5: Block spin renormalization of the 1d Ising model.

One of the main ideas of the renormalization group is to express the parameters of a coarse-grained
Hamiltonian in terms of the parameters of the original Hamiltonian for the same system. This coarse-
graining process is performed such that the long-range physics is left unaltered. In this problem, we
coarse-grain the 1d Ising chain in the absence of an external magnetic field. We choose a specific
procedure with a coarse-graining factor b = 3, i.e., we coarse-grain groups of three neighboring spins,
{si−1 , si , si+1 }, to a single new spin s0j = si at the midpoint position (see figure).

(a) Write the partition function of the system as a product of Boltzmann factors eJsi si+1 , where J is
the reduced coupling constant (measured in units of kB T ). Note there are three factors, containing
the spin products s2 s3 , s3 s4 , and s4 s5 , between the two midpoint spins s01 ≡ s2 and s02 = s5 .
(b) Using the coarse-graining rule defined above, write this triple product in terms of the corresponding
new spins (s01 , s02 ) by performing the sum over the original spins in between, s3 , s4 = ±1. Hint:
Use eksi sj = cosh(k(1 + si sj tanh k)).
(c) Generalize the result of (b) to write the entire partition function of the coarse-grained system,
0 0
Z = {s0 } e−H ({s }) with a new Hamiltonian H 0 {s0 } depending only on the coarse-grained spins
P

{s0 }, X
H 0 (s0 ) = N g(J, J 0 ) − J 0 s0i s0i+1 . (1)
i
Compute the new coupling constant J 0 (J) and the function g(J, J 0 ). Give an interpretation of g.
(d) Defining x ≡ tanh(J) and x0 ≡ tanh(J 0 ), find the relation between x and x0 . This relation tells us
how the nearest-neighbour coupling constant of the Hamiltonian changes under the coarse-graining
procedure. What is the physical interpretation of the limits x → 0+ and x → 1− ?
(e) Suppose we iterate the procedure: x → x0 → x00 ..., etc. What are the two fixed points of this
dynamics? Are they stable or unstable? Give a physical interpretation.

To be discussed on: Mon, November 25th

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

1
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 5

Problem 6: Dilute Ising model, tricitical behavior, and φ6 field theory

In this problem, we study an Ising model with spin variables s(r) taking the values s = −1, 0, +1 at
each point of the lattice. The Hamiltonian takes the form
1X X X
H(s) = − J(r − r0 )s(r)s(r0 ) + ∆ s2 (r) − h s(r). (1)
2 0 r r
r,r

This model describes, for example, a ferromagnet with vacancies.

(a) Discuss the zero-temperature physics of this model, by evaluating

Pthe energy of the homogeneous
spin configurations s = −1, 0, 1 (use the coupling constant J ≡ q J(q) as in the standard Ising
model). Depending on the parameters J, ∆, and h, what are the possible degeneracies and first-
order transitions at low temperature? For what choice of ∆ does the system reduce to the standard
Ising model?

(b) Convince yourself that the lowest-order polynomial field theory consistent with symmetry and
ground states of this spin model takes the form
Z  
1 2 2 4 6
H(φ) = (∇φ) (r) + τ φ (r) + δφ (r) + λφ (r) − hφ(r) dr. (2)
2

Consider now the continuum model (2). Let us put ourselves close to the so-called tricritical point
τ = δ = h = 0, λ > 0. Play around with the perturbations of τ, δ and discuss the change in the
shape of the energy landscape and the number of minima.
Convince yourself that:

• Turning on τ generates a paramagnetic (as in the standard Ising model) (τ > 0) or an ordered
phase (τ < 0).
• Turning on δ > 0 brings us back to the usual φ4 -theory and decouples the spin state s = 0
but preserves the Ising critical point of the spin states s = ±1. What does this tell you about
the relation of δ and ∆?

(c) Review the scaling dimensions of the local fields φk and their conjugate couplings λk (k = 1, 2, . . . )
at the Gaussian fixed point as functions of the dimension d. Use scaling arguments (as in problem
1) to compute the singularities of the correlation length, the magnetic susceptibility, and the
specific heat at the tricritical point in d = 3,

ξ ∼ |λk |−νk , χ ∼ |λk |−γk , c ∼ |λk |−αk (k = 2, 4), (3)

where λ2 ≡ τ and λ4 ≡ δ.

(d) Show that the upper critical dimension corresponding to tricritical behavior is d = 3. Assuming
that the field φ6 has a term φ6 (r)φ6 (r0 ) ∼ C|r − r0 |−x6 φ6 (r+ ) in its operator product expansion,
compute the perturbative fixed point describing tricritical behavior to first order in (3 − d).

1
To be discussed on: Mon, December 2nd

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

2
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 6

Problem 6: A special kind of phase transition

For this problem, we return to the XY-model as introduced in problem set 3: unit-length spins lay
down on a 2-dimensional lattice, from which follows that (S x )2 + (S y )2 = 1.
The hamiltonian is given as:
K X
H=− Sr · Sr 0 , (1)
2 0
hrr i

where the sum is over the nearest-neighbours.

(a) Show that the Hamiltonian in the continuum limit is given as H = 21 K dr(∇φ(r))2 .
R

Hint: Assume that the spins are slowly varying differentiable functions.

Consider the following configurations of spins, in which there exists a vortex around which the spins
revolve in a system of size L and lattice constant a. The winding number of this vortex is n = −1.

Figure 1: A vortex (red point) in the XY-model

(b) Given that the change in φ along any curve encircling a vortex with winding number n has to be
2πn, what is the energy E of the vortex? Can a single vortex exist in the thermodynamic limit?

The situation is different when we consider a pair of vortices with opposite winding number. The two
vortices cancel each other out at distances r that are large compared to the separation R between the
two vortices.

1
 Figure 2: The red points are the two vortices at distance R from each other.

(c) What is the energy E for such a vortex pair?

(d) Consider the vortices as point particles in the XY -plane. What is the entropy of the vortex-
pair? Write down the free energy for a vortex-pair at distance R. Consider now an effective force
F = − ∂F
∂R . When is it attractive, and when is it repulsive?

(e) Analyse the free energy and consider the different phases that can exist. What is the critical
temperature? How is this phase transition different from the phase transition in the normal Ising
system?

To be discussed on: Mon, December 9th

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

2
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 7

Problem 7: Correlations in two dimensions

In this problem set we are going to compute the correlation function for the XY model in two dimen-
sions. This provides us a nice excuse to review some useful calculations for Gaussian distributions.

(a) We first warm up with a simple calculation about one-dimensional gaussian densities. Let Z(γ)
be defined as Z
x2
Z(γ) = e− 2σ2 +γx dx = Z(0)heγx iγ=0 . (1)

Compute Z(γ) and show that every moment of the gaussian distribution N (0, σ 2 ) can be computed
by the taking its derivatives. For this reason, heγx i is called the moment generating function.

(b) We now move to higher dimensions. Let ~x ∈ Rd , with a multivariate Gaussian distribution
T
defined by p(x) ∝ e−~x Π~x where Π is a positive definite symmetric matrix. We now want to
compute the moment generating function in d dimensions, Z(~γ )/Z(0), with ~γ ∈ Rd . By using the
transformation: ~x = ~y + ~v , with a suitable ~v , show that
1 T Σ~
Z(~γ ) = Z(0) e 2 ~γ γ
, (2)

where Σ = Π−1 . In the same way as in (a), show that every moment can be calculated by
computing the partial derivatives of Eq. (2). In particular, compute hxi xj iγ=0

(c) Let us move to even higher dimensions. We study the infinite-dimensional spaces of scalar functions
in R2 , where we extend the objects that we already know in finite dimensions.

T
R
~x ~y →R Rd ϕ(r)ψ(r)
 Scalar product
0 0
Π~x → Rd Π(r, r )ϕ(r )dr 0 Matrix-vector product . (3)

 T 0 0 0
R
~x Π~x → Rd ×Rd ϕ(r)Π(r, r )ϕ(r )dr dr Quadratic form

In addition,
R the role of the identity is played by the Dirac-delta function in virtue of its definition
ϕ(x) = Rd δ(x − y)ϕ(y). We need only a final step in order to compute our correlation functions.
by parts, we see that the Hamiltonian H = κ r (∇ϕ(r))2 can be written as
R
If we
R integrate
−κ r ϕ(r)∇2 ϕ(r), which is effectively a quadratic form defined by the linear operator ∇2 . We say
that the probability of the field ϕ(r) is given by
ϕ∇2 ϕ
R
P [ϕ(r)] ∝ e−κ . (4)

In order to repeat the calculation in (b) we need to find its inverse G(r, r0 ), defined by:

− κ∇2 G(r − r0 ) = δ(r − r0 ), (5)

where we have expressed G(r, r0 ) = G(r − r0 ) due to the translational invariance of the problem.
Use Gauss’ Theorem in a sphere of radius R centered in r0 to show that 0

G(r − r ) in d dimensions
decays as R −(d−2) 1 R
and that in the case of d = 2, G(R) = − 2πκ log a + const.

1
(d) All the ingredients necessary to compute the spin-spin correlation functions are now ready. Com-
pute the value of hei(ϕ(r1 )−ϕ(r2 ) i and from there deduce
R the spin-spin correlation hS(r1 ) · S(r2 )i.
Hint: You may want to use the relation ϕ(ri ) = r δ(r − ri )ϕ(r), considering this as a scalar
product.

To be discussed on: Mon, December 16

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

2
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 8

Problem 8: Diffusion in a quadratic potential

Here, we consider the relaxation of a diffusion process towards equilibrium:

∂ D ∂2 ∂
P (x, t) = P (x, t) − [v(x)P (x, t)] (1)
∂t 2 ∂x2 ∂x
d
with a deterministic velocity field v(x) = −Γ dx U (x) and potential
1 2
U (x) = x . (2)
2

(a) Show that the equilibrium distribution is a Gaussian distribution

 
1 1 2
Peq (x) = p exp − (x − µeq ) , (3)
2π∆eq 2∆eq

and calculate its mean µeq and variance ∆eq .

(b) Let the initial distribution at time t = 0 be Gaussian with mean µ0 and variance ∆0 . Show that
the time-dependent solution Pt (x) ≡ P (x, t) of the diffusion equation for t > 0 is also Gaussian
and calculate the time-dependent mean µ(t) and variance ∆(t). Sketch the results for the following
cases:

(i) µ0 = µeq , ∆0  ∆eq ,

(ii) µ0 = µeq , ∆0  ∆eq ,
(iii) µ0 > µeq , ∆0 = ∆eq .

Hint: Assume that P (x, t1 ) at a given time t1 is Gaussian distributed. Consider a small time-
step δt and show that at time t2 = t1 + δt the distribution P (x, t2 ) is in first order of δt of
the form A(t1 ) exp[B(x, t1 , t2 )], where B(x, t1 , t2 ) is quadratic in x. B can hence be written as
−(x − µ(t2 ))2 /2∆(t2 ) + C(t1 , t2 ). Express µ(t2 ) and ∆(t2 ) as functions of µ(t1 ) and ∆(t1 ). From
that follow differential equations for µ(t) and ∆(t), which you have to solve. Check, if P (x, t2 ) is
normalized properly.

(c) Calculate the Shannon entropy of time-dependent solution (minus its equilibrium value)

S(t) − Seq ≡ S(P (t)) − S(Peq ), (4)

the expectation value of the energy (minus its equilibrium value)

Z Z
U(t) − Ueq ≡ dx U (x)P (x, t) − dx U (x)Peq (x), (5)

and the resulting free entropy (minus its equilibrium value)

Φ(t) − Φeq ≡ (S(t) − Seq ) − β(U(t) − Ueq ) (6)

1
 with β = 2Γ/D. Discuss the behaviour for all three observables for the cases given in (b).
Remark: 1/β is the constant temperature of the reservoir, with which the system is in contact. In
this diffusive problem the motion is over-damped, i.e. the kinetic energy is in equilibrium during
the whole process and the dynamics only depend on the potential energy U .

To be discussed on: Mon, January 13th

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html

2
 Prof. Michael Lässig
Universität zu Köln Dr. Fernanda Pinheiro
Institut für Biologische Physik Denny Trimcev
Matthijs Meijers

Advanced Statistical Mechanics (WS 2019/20)

Problem Set 9

Problem 9: Detailed balance in a field theory

The dynamics of a field φ(r, t) is given by

∂ δH(φ)
φ(r, t) = −Γ + ζ(r, t), (1)
∂t δφ(r, t)

where φ : r → φ(r, t) denotes field configurations at a given time, H(φ) is the Hamiltonian, and the
noise ζ(r, t) is Gaussian with moments

hζ(r, t)i = 0, hζ(r, t)ζ(r0 , t0 )i = D δ(r − r0 ) δ(t − t0 ). (2)

This dynamics leads to an equilibrium of the form

1 2Γ
Peq (φ) = exp[−βH(φ)], β= . (3)
Z D

(a) For a field Hamiltonian of your choice, compute the propagator Gτ (φ0 |φ), where τ is a short time
interval and φ0 , φ are close field configurations (in a sense to be made more precise).

(b) Establish the detailed balance relation

Gτ (φ0 |φ) Peq (φ) = Gτ (φ|φ0 ) Peq (φ0 ). (4)

Problem 10: Dynamic scaling

Consider a dynamical field theory with variables φ(r, t) that is invariant under scale transformations

r → b−1 r, t → b−z t, φ → bx φ, (5)

where x is the scaling dimension of the field and z denotes the dynamical exponent.

(a) Establish the differential (Callan-Symanzik) form of the scale transformation for the two-point
correlation function
C(r1 − r2 , t1 − t2 ) ≡ hφ(r1 , t1 )φ(r2 , t2 )i, (6)
assuming translation invariance in space and time.

(b) Show that the solution of the Callan-Symanzik equation takes the form
 
−2x |t1 − t2 |
C(r1 − r2 , t1 − t2 ) = |r1 − r2 | G (7)
|r1 − r2 |z

with a scale-invariant, dimensonless function G (called a scaling function).

1
(c) From the solution (7), derive the scaling of temporal correlations,

C(0, t1 − t2 ) = |t1 − t2 |−2x/z . (8)

(d) Now identify the field variable with the local height of a surface, h(r, t), with a so-called roughness
exponent ζ ≡ −x > 0. The width of the surface, ∆(r, t), describes local height fluctuations,

∆2 (r, t) ≡ h(h(r, t) − hh(r, t)i)2 i. (9)

Assuming scale invariance of the form (5), compute the scaling of the width for the following two
processes:
(i) surface dynamics in an infinite system with a flat initial configuration h(r, t = 0) = 0, generating
a time-dependent width
∆(t) ∼ t? ; (10)
(ii) stationary surface dynamics in a finite system of size R, generating a size-dependent width

∆(R) ∼ R? . (11)

To be discussed on: Mon, January 20th

Course information: http://www.thp.uni-koeln.de/~lassig/teaching.html