8.

334: Statistical Mechanics II Problem Set # 6

Fluctuations in low space dimension.
1. 1D Ising model, correlation length, susceptibility
P
a) Consider a pair spin correlation function in the 1D Ising model H = i Ksi si+1 , si = ±1. This
problem can be solved by the transfer matrix method, discussed in recitation, or by the method of
auxiliary link variables ξi = si si+1 , discussed in Lecture 6 [Eqs.(10,11,12)]. Show that in terms of
the link variables, the correlation function is
Y
hsi1 si2 i = h ξj i , i1 < i2 . (1)
i1 ≤j<i2

Using the hamiltonian for ξ’s (Eq.(11), Lec. 6), average the product (1). Comparing to the asymp-
totic formula C2 (x − x0 ) ∝ exp(−|x − x0 |/ξ), determine the correlation length ξ as a function of
temperature.
In terms of the link variables the Hamiltonian is
X X
H=K si si+1 = K ξi . (2)
i i

Hence !|i−i0 |
Y eK − e−K 0 0
hsi si0 i = h ξj i = = (tanh K)|i−i | = e−|i−i |/ξ (3)
i≤j<i0 eK + e−K
Therefore 
 e2J/T , T  J
ξ = −1/ ln tanh K = (4)
 1 ,T J
ln(T /J)

b) At distances much larger than ξ the system is disordered, while at distances less than ξ the spins are
strongly correlated. In RG language, for block sizes ≥ ξ the block spins are effectively decoupled from
each other. (Indeed, renormalized temperature becomes very large at the length scale ξ, indicating
decoupling.)
Consider spin susceptibility χ = dm/dh|h=0. Treating the problem as a system of independent block
spins of size ξ, each taking values ±ξ, show that χ ∼ ξ/T .
For one block of size ξ the spin susceptibility is χblock ∼ ξ 2 /T (Curie’s law for spin s = ±ξ. Multiplying
this by the block density ξ −1 , obtain χ ∝ ξ/T .
c) It is interesting to look at the susceptibility in the ξ variables representation. Start with the general
expression χ = d2 ln Z/dh2 |h=0 , and write the susceptibility in terms of ξi . Average over ξ’s using the
method of part a) and compare the result with the above estimate for independent block spins.

 
dm d2 ln Z 1X 1X |i−i0 | L 2 e2K
χ= = = hs s
i i0 i = (tanh K) = − 1 = L (5)
dh dh2 T i,i0 T ii0 T 1 − tanh K T
where L is the system size. This agrees with the result of part b).
2. Magnetization and susceptibility of a 2D Heisenberg ferromagnet
a) Use the thermodynamic potential of spin excitations of a 2D Heisenberg ferromagnet at low tem-
perature (Problem 2, PS#1), and find magnetization m = d ln Z/dh as a function of magnetic field
h. Show that the zero temperature value m(0) is reduced by thermal fluctuations so that the difference
δm = m(0) − m(T ) diverges logarithmically as h → 0.
The free energy is
Z   d2 k
F (h) = T ln 1 − e−βω(k) − |m0 |h (6)
(2π)2
where ω(k) = h + Jk 2 is the spin wave spectrum. The first term in the expression for the free energy
is the contribution of thermal fluctuation and the second term with |m0 | = 1 is the is the ground
state energy.
The average magnetization is
Z Z
dF βe−βω(k) dω(k) d2 k 1 d2 k
m=− =1−T =1− (7)
dh 1 − e−βω(k) dh (2π)2 eβω(k) − 1 (2π)2
Consider the contribution of the small k excitations, ω(k) ≤ T :
√
T Z ω(k)<T d2 k T Z T /J k dk T T
m=1− 2 =1− = 1− ln (8)
4π 0 ω(k) 2π 0 h + Jk 2 4πJ h
where we assumed that h is small, h  T . Therefore the difference δm = m(0) − m(T ) diverges
logarithmically, δm ∝ ln(T /h)
b) Relate the divergence found in part a) with the logarithmic divergence from the Mermin and Wagner
theorem. The latter is concerned with transverse magnetization fluctuations, so one has to connect
the depletion of spin polarization along the field with its transverse fluctuations.
Introduce transverse magnetization δm⊥ . Then
q 1
mz = 1 − δm2⊥ = 1 − δm2⊥ + O(δm4⊥ ) (9)
2
Using Mermin-Wagner theorem (see comment in Lecture 8) find
Z
2T d2 k T T
hδm2⊥ i = = ln (10)
(2π)2 Jk + h
2 2πJ h
Hence obtain for mz the same expression as in part a).
c) It is interesting to consider the zero field susceptibility of this system. From RG analysis of the
nonlinear sigma model, we know that the effective coupling becomes very small at some temperature-
dependent length scale ξ(T ). Use the arguments similar to those in part b) of Problem 1 to estimate
the susceptibility. Does your estimate agree with the above result for m(T ) extrapolated to h = 0?
For blocks of size ξ = a exp(2π/g), the effective spin is s ∼ ξ 2 . Thus the susceptibility is
1 2 2 −2 ξ 2 a2
χ≈ (ξ ) ξ = = e4π/g (11)
T T T
since a2 ≈ J/T , the susceptibility is
J 4πJ/T
χ≈ e . (12)
T2
Compare this to the results of parts a) and b)

T 1
χ(h) = (13)
4πJ h
This expression is valid at sufficiently large h, so that the logarithmic term ln(T /h) is small: T /h <
exp(4πJ/T ). Hence h > T exp(−4πJ/T ). Using this estimate in the expression for χ(h), obtain
1 4πJ/T
χ(0) ≈ e (14)
4πJ
The large exponential factor is reproduced correctly, but the prefactor is different by (T /J)2 .
3. The sine-Gordon model
Consider a 2D gaussian problem of a scalar field φ(x) perturbed by a small cos term:
ZZ  
1
H= (∂µ φ)2 + λ cos(βφ) d2 x (15)
2
Here λ is a coupling constant, and β is a dimensionless parameter.
Apply the field-theoretic RG transformation described in Lecture 8 (coarse grain + rescale + renor-
malize hamiltonian) to this problem, treating the cos term as a perturbation. On each RG step,
split the field φ into the fast and slow component, and average the Hamiltonian1 (15) over the fast
fluctuations, assuming them to be described by the quadratic part of (15).
a) Show that the hamiltonian form is preserved upon RG and derive the RG flow equation, first order
in a weak coupling constant:
dλ/dt = −f (β) λ , t ≡ ln(Λ/Λ0) (16)
Find the function f (β) and show that, depending on the value of β, the coupling λ either decreases, or
increases. For λ decreasing, the problem turns into a freely fluctuating gaussian field, corresponding to
a disordered state, whereas for λ increasing, the fluctuations freeze at a certain length scale (estimate
it), corresponding to the ordered state.
RG flow can be obtained by averaging λ cos βφ over fluctuations with Λ0 < q < Λ:
1 1  1 2 2

hλ cos βφi = λheiβφslow +iβδφ + c.c.i = λ eiβφslow e− 2 β hδφ i + c.c. (17)
2 2
where
X 1 1
hδφ2 i = = ln Λ/Λ0. (18)
q q2 2π
1
It would be more appropriate to average the exponent e−H , which is a slightly more difficult task. However, to
the first order in λ the result is the same, so averaging the Hamiltonian in this problem is sufficient. The reason for
this, as we discuss later, is that “the first loop RG approximation” is nonzero, and so the transition arises already in
the lowest order of perturbation theory. Wait till PS#7 to see an example of RG based on averaging e−H .
Hence obtain !β 2 /4π
Λ0
hλ cos βφi = λ cos βφslow . (19)
Λ
After rescaling x → Λ/Λ0 x, obtain
  β2

0 Λ 2− 4π
λ = λ. (20)
Λ0
Near fixed point β 2 = 8π, variation of λ is slow and one can cast the RG in a differential form:
!
dΛ β2
= 2− λ. (21)
dl 4π

Solution to this ODE reproduces the result (20).

b) Add a perturbation µ cos(β 0 φ) to the problem (15). Show that this perturbation is irrelevant near
the phase transition found in part a) when β 0 > β.
On the contrary, a perturbation with β 0 < β is relevant. How does it affect the phase transition?
The perturbation µ cos β 0φ is renormalized as
! 
0 2
2− (β4π)
dµ (β 0 )2 Λ
= 2− µ , or µ0 = µ. (22)
dl 4π Λ0

If β 0 > β, the perturbation decreases near 2 0

√ the critical point β 2 = 8π. Call it irrelevant. If β < β,
µ grows much faster than λ at β bellow 8π. Critical point β = 8π is destroyed and replaced by
another critical point (β 0 )2 = 8π, controlled by µ cos β 0φ term.
4. Magnetization in the XY problem.
Consider an XY model for planar ferromagnet in a weak in-plane magnetic field. The magnetization
is described by a 2-component unit vector n = cos θx̂ + sin θŷ.
a) Show that the hamiltonian, written in terms of the angular variable θ, has the form
ZZ  
1
H= J(∂µ θ)2 − h cos θ d2 x (23)
2
(here the field h is applied along the x axis). The hamiltonian is formally identical to the sine-
Gordon problem (15), except that in the XY case the variable θ lives on a circle, not on a line. For
the moment, however, let us ignore this difference.
Starting from Z  
1
H= J(∂µ n)2 − h · n d2 x (24)
2
and introducing phase θ: n1 = cos θ, n2 = sin θ, obtain
Z  
1
H= J(∂µ θ)2 − h cos θ d2 x (25)
2
where h k x.
b) The magnetization induced by the field h is given by m̄ = hcos θi. Consider the limit of a weak
field, when thermal fluctuations suppress magnetization. Assume that the temperature is below the
Kosterlitz-Thouless temperature, so that vortices can be excluded from the analysis.
To find the magnetization, go back to the RG analysis of Problem 3 a). At low temperature, when
f (β) < 0 and coupling grows under renormalization, determine the length scale Λ∗ at which the
second term of the hamiltonian (23) reaches the first term. Argue that at larger length scales the
fluctuations freeze out and the system becomes ordered. By averaging cos θ over the fluctuations with
wavelengths up to Λ∗ , find the magnetization as a function of magnetic field.
Using the results of Problem 3, find
!1/4πK
Λ0
mΛ0 = hcos θi = (26)
Λ

where K = J/T . Look for scale Λ0 at which h cos θ becomes of the order of the first term. At this
scale (see Problem 3)
 2− 4πK 1
Λ
hΛ0 = hΛ . (27)
Λ0
Therefore the condition on Λ is
  1
Λ0 2− 4πK
hΛ0 = h ≈ K. (28)
Λ0
Hence,
1
Λ∗ = Λ0 (h/K)α , where α = 1 . (29)
2 − 4πK
The fluctuations become massive at length scales bigger then (Λ0 )−1 and freeze out. Therefore the
magnetization can be estimated as
1
mΛ∗ ≈ (h/K)α/4πK = (h/K)γ , where γ = (30)
8πK − 1