ALESSIO MISCIOSCIA

S TAT I S T I C A L P H Y S I C S

CHECKED AND CORRECTED BY ANDREA SCANU

"Ludwig Boltzman, who spent much of his life studying statistical mechanics, died in 1906, by his own
hand. Paul Ehrenfest, carrying on the work, died similarly in 1933. Now it is our turn to study statistical
mechanics. Perhaps it will be wise to approach the subject cautiously."
D.L. Goodstein
Contents

Introduction and purposes 5

Recap of thermodynamics 7

Thermodynamics of phase transition 15

Recap of theory of ensamble 21

Statistical mechanics and phase transitions 25

Models 31

Dimensions, symmetries and range of interaction 47

Mean Field Theory 57

Landau Theory 73

Ginzburg-Landau theory 79

Scaling theory and renormalization group 89

4

Spontaneous symmetry breaking 97

Exercises 101

Bibliography 109

Index 111
Introduction and purposes

These notes are the transcription of Statistical Mechanics lecture’s

course held by Professor E. Orlandini. There are surely some mistake
and typos.
Let’s introduce the purposes of this notes, to go on it is necessary
to know the fundamentals of thermodynamics, statistical mechanics
and physics in general. These notes topics are systems composed
by a large number of particles N and we will study these systems in
their density of state Ω( E) which depends on the type of particles we
are considering. We distinguish three kind of particles:

fermions associate with Fermi-Dirac statistics;

bosons associate with Bose-Einstein statistics;

boltzmanions associate with Boltzman statistics, this are the classical

particles.

You can have a constant energy and we will call the ensamble mi-
crocanonical ensamble, you can have a constant temperature, for
example putting the system in contact with a reservoir, and you can
define the canonical ensamble or you can have a system in which
the chemical potential is constant and so you get the grancanonical
ensamble.
The role of statistical mechanics is to connect thermodynamics
with macroscopic phenomena.
In this notes we will study only equilibrium system, but let’s note
that we can have many kind of equilibrium. For instance you can find
the same material in solid phase, liquid phase or gas phase, it means
Figure 1: General phase diagram.
that at a particular temperature you will have a phase transition.
The system which is not at equilibrium
This is the prototype of this course study case and we are interesting is one of the main research activities in
in understand the physics of phase transitions. You will see that statistical mechanics.
phase transitions are associated with thermodynamical potentials
singularity.
Another fundamental case of study are the response functions
which are for instance the specific heat C, the compressibility KT and
the magnetic susceptibility χ M .

Figure 2: Possible behaviors of C near a

critical temperature.
6

You will see that this functions in presence of phase transition

present jump or divergence and in particular some of these functions,
near the critical point present a power law behavior, so that we can
write
T − Tc
 
C ' t−d t= d∈N
Tc
where Tc is the critic temperature. The critical exponent d do not
depend on microscopic detail of the system: this fact is a universality
principle. For instance the fluids have the same critical exponent
independently on the material, so that water and liquid helium will
have the same one (but not necessarily the same critical temperature).
On the other hands there are some parameter which is important:
the dimension, the range of interactions, the symmetry of the system
and others. Some of these symmetries are valid only in precise con- Spontaneous symmetry breaking is
dition so it is possible to observe a spontaneous symmetry breaking important also in particles physics.

phenomena when these condition are not present in the system.

How is it possible to define a critical points? The simplest way is
this: let ξ be a typical length of correlation between particles, Tc will
be the critical temperature if

ξ ' t−d

for some natural d. So ξ is divergent in the critical point.

This definition is scale invariant because the physics of critical
state does not change in a appreciate way if you observe the system
using a different observation scale. This is the deeper reason why the
critical exponent do not depends on the microscopic physics.
The next question is: is there some mathematical instruments
that help us to describe the phase transition phenomena? This is the
renormalization groups and it will be discussed in the last part of
this notes.
Let’s observe that the universality justify the use of a model, in fact
we can construct a model for a large set of phenomena. The approach
will be to simplify the model, but in those cases in which it will be
no possible there will be introduced approximations and so will be
provided methods to use approximations properly.
Recap of thermodynamics

In this chapter there will be a brief recap of fundamental of thermo-

dynamics which is the macroscopic theory of matter.
Let’s consider a system in equilibrium, for example in an external
bath, and with a large number of particles N. Quantitatively let’s
consider some mole of matter, we know that one mole corresponds
on
N = Avogadro number ' 6 1023

Two approaches are possible: the phenomenological one which con-

sist in start from experiment to construct theory and the axiomatic
approach which consist in construct theory from axioms and then
test the theory with experiment.
Since this is only a recap it will be enough to introduce the ax-
iomatic approach. For the ones who are interesting in
the phenomenological approach we
Let’s introduce extensive variable: variable, or observable, that
suggest
scale with the system. The main extensive variable are the volume E. Fermi. Termodinamica. November
V, the internal energy U, the magnetisation M ~ and the entropy 1977
S (which we will see better). It is necessary to introduce also the
thermodynamical constrains. A constrain could be a wall or a sort of
separation between two or more system, so that we distinguish

Adiabatic walls through which no heat flux can pass. If you remove
an adiabatic wall you are in presence of a diathermic wall through
which can pass some heat but the system keeps the temperature
fixed;

Rigid walls which stop every mechanical work; if you remove it you
have a flexible wall usually represented by a piston;

impermeable wall through which no flux of particles can pass, but if

you remove it you are in presence of a permeable wall.

Thermodynamic studies equilibrium states which are defined as

macroscopic states totally described by extensive parameters.
Thermodynamic processes are not simple in general, we can
idealize some of them as a quasi-static transformation: we think at
8

this set of processes as a sequence of equilibrium states. It should We can reproduce a quasi-static trans-
be obvious that an infinitesimal transformation is a quasi-static formation in a laboratory performing
an extremely slow transformation.
transformation. The first law of thermodynamics says us that for
infinitesimal transformation, if we define δW the work done by the
system and δQ the heat absorbed by the system We use d for exact form and δ for not
exact form.
dU = δQ − δW

where dU is a differential obtained from internal energy U.

You can describe the fundamental problem of thermodynamics
in this way: you start from an equilibrium state and then we remove
some constrains, this operation changes the walls properties, now
you want to find the new equilibrium state. How to solve this prob-
lem in general? You have to use a new function whose existence is an
axiom and it depends only on extensive variable U, Xi . This function
will be defined entropy

S = S(U, Xi )

The entropy’s fundamental properties are contained the following

axioms:

1. S is an additive function respect to subsystems. So if you have n

subsystem labeled by an index α you can write
n
S= ∑ S(α)
α =1

and is obvious that S is extensive.

2. S is differentiable and monotonically increasing with respect to U:

∂S
>0
∂U
in particular you get
∂S
6= 0
∂U
and so you can invert the entropy respect to U, so that you have

U = U (S, Xi )

3. For each subsystem (at equilibrium) you have

 
(α)
S ( α ) = S ( α ) U ( α ) , Xi

4. S is an homogeneous function of first order, in other words

S(λU, λXi ) = λS(U, Xi ) ∀λ

 9

By the entropy is possible to postulate that the new values taken

by extensive variable in the new equilibrium state are the ones that
maximize S. This axiom is called the maximum entropy postulate .
Mathematically you impose that the new extensive variables satisfy

dS = 0 d2 S < 0

Another important postulate is the third law of thermodynamics

which says that the entropy is zero for equilibrium state in which Usually this condition is written as
T = 0, but
∂U ∂U
=T
=0 ∂S
∂S so is only a different way to write the
Now you can differentiate the internal energy so that same thing.

∂U ∂U
dU = dS + ∑ dX j
∂S j
∂X j

The derivatives of U respect on the extensive variables are scale-

independent, in fact

∂(λU ) λ ∂U ∂U
U → λU S → λS ⇒ = =

∂(λS) λ ∂S
 ∂S

This derivatives will be called intensive variable. There are some of

them that you have to define
∂U
Temperature T = ;
∂S
∂U
Pressure P = − ;
∂V
∂U
electro-chemical potential µ = ;
∂N
It should be obvious that the intensive parameters are functions of
extensive variable. The equation Some important equations of states are
the equation of states for ideal gas
T = T (S, Xi ) or Pi = Pi (S, Xi ) PV = Nk b T
The Van der wall equation for non ideal
is called equation of states. Let’s remark that if we know all the gas
equations of states we know U up to a constant.
aN 2
 
As regards intensive variables definition, these variables are homo- P + 2 (V − Nb) = Nk b T
V
geneous function of zero order, so and the Curie law
0 CH
T (λS, λXi ) = λ T (S, Xi ) = T (S, Xi ) M=
T
where
In other words at equilibrium the temperature of the whole system is ∂U
H=
equal everywhere in the system and for every subsystem. And in an ∂M
is the magnetic field.
similar way
P(λS, λXi ) = P(S, Xi )
10

This formula is also called Pascal law.

The equilibrium condition is

dU = 0

and so you get a relation between intensive variable in differential

form
TdS + ∑ Pj dX j = 0
j

but you can also prove the Gibbs-Duhem relation We will not prove it but a proof are
provide in this book
SdT + ∑ X j dPj = 0 Callen and Herbert. Thermodynamics
j and an Introduction to Thermostatistics.
Graphics Press, LLC, second edition,
So in conclusion all the thermodynamical information is stored in a 1985
function, that could be S or U, but if you want to construct an exper- This formula for a fluid is quite famous
iment you will have practical difficulties in studying the entropy. So SdT − VdP + Ndµ = 0
you will have to construct other functions that, like S and U, which
give the thermodynamical information. This functions are called
thermodynamical potentials and you can get them performing a
Legendre transformation on S or U (or in general another thermo-
dynamical potential). In this way, we start from a thermodynamical Legendre transformation are taken
potential Y = Y ( X0 , X1 , · · · , Xn ) which is strictly convex respect to for granted: it should has known
from mathematical physics course.
the variable Xk , so that Substantially the geometrical idea is
∂2 Y that you want to use the derivative
>0 of the function as a variable of the
∂Xk2
function itself. A simple introduction
and we change the variables such that is provide by http://copernico.dm.
unipi.it/~milani/dinsis/node37.html
∂Y
Xk → Pk =
∂Xk
For instance, since is difficult to use the entropy in an experiment you
can Legendre transform the internal energy U, so that We have to recall that
∂U
∂U T=
A=U− S = U − TS ∂S
∂S
The potential A is called Helmholtz free energy. First of all you have
to understand from which variables A depends: Since
dU = TdS + ∑ Pj dX j
dA = dU − TdS − SdT = ∑ Pj dXj − SdT j
j

So A is a function of T and X j that is what you wanted: A do not

depend on the entropy.
Since A contain all the information of thermodynamic system, is it
possible to solve the fundamental problem?
To answer this question let’s consider a simple case: an isothermal
spontaneous transformation. We have

dA = dU − TdS − SdT = δQ − δW − TdS − SdT

 11

And so we find that

≤0
z }| {
δW = δQ − TdS − dA −   ≤ −dA
SdT

Where we use the second law of thermodynamics δQ ≤ TdS and the

fact that the path is isothermal (dT = 0).
Let’s consider a spontaneous transformation so that δW = 0 and
so dA ≤ 0. So spontaneously A goes to a minimum and so you can
substitute the maximal entropy rule with the minimum Helmholtz
free energy rule.
In the same way you could be interested in use the pressure in-
stead of the volume and so you have the Hentalpy function To show that Hentalpy do not depend
on the volume you consider a system in
∂U which dU = TdS − PdV and so
H =U− V = U + PV
∂V dH = dU + PdV + VdP =
We also could perform a double transformation. For example = TdS − +
PdV  + VdP =
PdV

if you want to use the temperature instead of the entropy and the = TdS + VdP

pressure instead of the volume (which is better for an experimental

physicist) you can decide to use the Gibbs Potential To show that Gibbs potential depends
on P and T you consider a system in
∂U ∂U which dU = TdS − PdV and so
G=U− S− V = U − TS + PV = A + PV
∂S ∂V  − Sdt − (
dG = 
dU (+(PdV
TdS
(
( + VdP =
This potential is also useful for the magnetization. In fact in an = VdP − SdT

experiment you are not able to change magnetization, but you can
easily change the magnetic field so you want to perform M ~ → H.~
You use the Gibbs G = G ( T, P, H ~ ) (sometimes in literature we
~
found G = G ( T, H ) for some system which have a fixed length as
crystal). The last potential is the Gran canonical potential which is
the internal energy in which N → µ and S → T so that It is possible to show that Ω depends
only on T, V ab µ: let’s consider dU =
Ω( T, V, µ) = U − TS − µN = A − µN TdS − PdV + µdN

dΩ = dU − dTS − TdS − µdN − dµN =

Sometimes we also find a variation = −SdT − PdV − Ndµ
G
Ω ( T, P, µ) = U − TS − µN + PV Similar steps can be taken for ΩG .

Since that the function U is a homogeneous function of degree 1. and

since that for this class of functions exist mathematical result known
as Euler equation you can conclude that

U = TS − PV + µN

This formula tells more than the differential one and gives the precise
form of the internal energy.
What about the second derivative of a thermodynamical potential?
You can use the Schwarz Theorem and so if you have a poten-
t ( t + 1)
tial which depends on t + 1 variables you find relations
2
12

between the variables known as Maxwell relations. For example for

U (S, V, N ) we have
∂T ∂2 U ∂P
= =−
∂V ∂V∂S ∂S
It is important to remark that physically a partial derivative means
that you have to change the function varying only the variable in
∂U
which you perform the derivative so that means that we change
∂V
U changing V but we have to keep S and N fixed. So you can write
in this way    
∂T ∂P
=−
∂V S,N ∂S V,N
where the subscripts remember which variables are kept fixed.
From a theoretical point of view the Maxwell relations could
seem obvious but from an experimental point of view change the
entropy and the pressure is more difficult that changing volume and
temperature.
You can get the Maxwell relations for every thermodynamic
potential, for example the Helmholtz free energy
   
∂P ∂S
− =−
∂T V,N ∂V T,N

Let’s define the response functions: roughly speaking this functions

tell how some observables change varying some variables. You have
already seen that the observables are substantially first derivative of
a thermodinamical potential, and so should be clear that a response
function is a second derivative of a thermodynamical potential.
Let’s consider some important examples. The thermal expansion
is defined  
1 ∂V
αP =
V ∂T P,N
You have to keep P and N fixed and use T as a variable, so the sim-
plest potential in this case is the Gibbs one. So that

1 ∂2 G
αP =
V ∂P∂T
You can use Gibbs potential also to define the molar heat capacity (at
constant pressure) In the step ! you have to use the re-
versibility of the path.
∂2 G
     
δQ ! ∂S
CP = =T = −T
∂T P,N ∂T P,N ∂T 2 P,N

And in the same way we can define the isothermal compressibility

∂2 G
   
1 ∂V 1
kT = − =−
V ∂P T,N V ∂P2 T,N
 13

Is not difficult to find some important relations we can show the first relation
∂P ∂P ∂V V ∂P ∂V α
= = = P
   
∂P α ∂S
= P = −Vα P ∂T ∂V ∂T V ∂V ∂T KT
∂T V,N KT ∂P T,N where we only use the properties of
the derivative and the definitions of the
This relations is important because is difficult to measure the member response functions.The second relation
in the left but is simple to measure the member in the right, so this 
∂V
  
∂S
−Vα P = − =
relations give a way to measure how the pressure change varying the ∂T P,N ∂P T,N
temperature or how the entropy change varying the pressure. where we use Maxwell relation
Another example is the (isothermal)magnetic susceptibility. We
   
∂V ∂S
− =
first give the definition for a uniaxial magnet, that is a system in ∂T P,N ∂P T,N

which M ~ = M~a where ~a is a versor.

   2 
∂M ∂ G
χT = =−
∂H T ∂H 2 T

~ (and so a generic
You can generalize this definition for a generic M
~
H) This is a tensor!!
!
∂Mα
χαβ =
∂Hβ
T
This functions is very useful, not only for experimental physicist in
order to measure something but allow us to define the stability of
some system. In fact you can define the thermal stability of a system

cV ≥ 0 cP ≥ 0

or in other words a system is in thermal stability if c P and cV are

non negative functions. In the same way you can define a system in
mechanical stability as a system in which k T ≥ 0. This definition is the deep reason for
Similar step could be done for the other response functions. the sign − in the definition of k T .

Following some useful formulas In order to show the first relation let’s
use some multi-variable differential
2 2 calculus rules to get the triple product
TVα2
 
T ∂V T ∂M
c P − cV = = c H − cK = rule
KT Vk T ∂T P χT ∂T H      
∂P ∂V ∂P
= −1
∂T V ∂P T ∂V P
and so we get
c P ≥ cV ≥ 0 cH ≥ cM ≥ 0 The differential form of the entropy
S( T, V ) is
   
∂S ∂S
dS = dT + dV
∂T V ∂V T
And so
       
∂S ∂S ∂S ∂V
− =
∂T P ∂T V ∂V T ∂T P
Using the definition of c P and cV and
the the triple product rule

∂V 2
   
∂P
c P − cV = − T
∂V T ∂T P
Let’s use the α P and the k T definitions
α2P
c P − cV = − TV
kT
Similar steps can be done for the second
relation.
Thermodynamics of phase transition

Equilibrium states are defined by a minimum or a maximum of a

thermodynamical potential. How can you visualize different phase?
The best way it is to associate for each phase a different potential, for
instance a molar Gibbs potential

gα = gα ( T, P)

where the α index represent the phase.

In order to find equilibrium states you have to minimize gα ∀α ,
and then choose the minimum between all of them to minimize it. A
pictorial (but not rigorous) description is given by figure 3. Figure 3: A pictorial representation of
Let’s consider only two phases: when gα < g β the equilibrium the behavior of g respect to P and T.
phase is the one corresponding to α, when gα > g β the equilibrium
phase is the one corresponding to β and when gα = g β the system is
in a coexistence of two phases (α and β).
For better understand let’s study the matter phase diagram. The
point ( Tt , Pt ) is called triple point and is the point at which coexist 3
phases (in this case the solid, liquid and gas phase). The critical point
( TC , PC ) is a point in which we can not distinguish a liquid phase
from a gas phase. Let’s fix P = P? > Pt to study the phase transitions.
There are two important points: a which is the point that intersects
the melting line and b the point that intersects the sublimation line as
in figure 4. In the same figure there is represented the behavior of gα
versus the temperature. In particular at the temperature



gα T = g β T

g is a continuous function, but if you try to draw the graph (s, T )

where s is the molar entropy (figure 4)
Figure 4: A phase diagram in the top
 
∂g picture. In the middle there is the
s=− behavior of g varying T and fixing
∂T P? P = P? . In the down it has been
it is possible to observe two jumps. One in correspondence of Ta and represented the molar entropy varying
T.
one in correspondence of Tb .
This example is coherent with the Ehrenfest definition of a phase
transition: A point is a phase transition of the order n if the nth
derivative display a jump. To be precise a modern word could be
"singularity", later there will be a better
discussion on this definition.
16

The graph of the molar entropy show that, since δQ = ∆sT, in

correspondence of a phase transition there is a laten heat, in our case

λ a = ∆s a Ta λb = ∆sb Tb

Let’s follow the same procedure fixing a temperature T = T ? .

Figure 5: A phase diagram in which we
Some graph are reported in figure 5. In this case have fixed T = T ? .
 2   
∂ G ∂V
= = −Vk T < 0 ⇒ V > 0
∂P2 T ? ∂P T ?

All this phase transitions is transitions of 1th order, but there are
also transitions of the 2th order, for instance the transition between a
conductor to a superconductor.
Let’s introduce a magnetic system. We want to study it in the
Gibbs picture so that P → H and V → M and G = G ( H, T ). The
phase diagram is plotted in 6. Figure 6: A phase diagram of a mag-
netic system. The red line is the coexis-
Let’s remark the meaning of the red line: From a statistical me- tense line.
chanics point of view it is the coexistense of M > 0 and M < 0, but
H = 0 and so there is not a field to study (experimentally).
It is possible to observe that there is a critical point TC . So there
are two phases of the system

ferromagnetic system for T < TC ;

paramagnetic system for T > TC ;

To see that there are critical points in correspondence of the red line
Figure 7: Some plot of M varying H at
we can plot the M varying H for a generic T < TC and for T = TC (7). T < T ? and T = T ? .
You can see that there is a jump for T < T ? but the gap goes to zero
when T goes to TC .
Let’s consider the coexistence curve of the phase transition be-
tween 1 (to which it corresponds g1 ( T, P)) and 2 (to which it corre-
sponds g2 ( T, P)) and assume that you know a point ( TK , PK ) in this
curve.
How to get other points on coexistence curve?
Surely Figure 8: Representation of the situa-
g1 ( TK , PK ) = g2 ( TK , PK ) tion.

Let’s consider two point a, b on the tangent space of the coexitance

line in the point ( TK , PK ). Assuming that these two points are close
enough ( TK , PK )
( a) ( a) (b) (b)
g1 = g2 g2 = g2
and so since
(b) ( a)
dgα = gα − gα
you get
dg1 = dg2
 17

From the Gibbs Duhem equation Where α = 1, 2

dgα = duα = −sα dT + vα dP

and so the derivative of pressure P varying the temperature T along

the coexistence curve is : Let’s propose a pair of simple exercises
of Clausius-Clapeyron equation.
s − s1 ∆s ∆S
 
∂P
= 2 = = Melting on Everest : A point in the
∂T coex. v2 − v1 ∆v ∆V melting line of the water is T =
273K and P = 1atm. At the Everest
This is the proof of the Clausius- Clapeyron equation P = 0.36atm and for the water it is
known that δQ = 6.01kJ/mol and
∆v = −1.75cm3 /mol. What is the
∆S
 
∂P melting temperature on the Everest?
=
∂T coex. ∆V From the Clasius-Clapeyron equa-
tion
Since the latent heat is defined by λ = ∆S T it’s possible to rewrite ∂P ∆s δQ J bar
= = ' −1.29 104 3 = −1.29 3
this equation ∂T δv T∆v m m
 
∂P λ And so
=
∂T coex. T∆V ∆P (1 − 0.36)atm
∆T = = = −0.5◦C
∂P
−1.29 bar
m3
Let’s show a pair of applications of this equation considering the ∂T

liquid gas transition. Since the coexistence line is an increasing And so the melting temperature on
Everest is Tm = 273.5◦C.
straight it will be true that
Boiling on Everest : The pressure at
the Everest is P = 0.36atm and
∆S
 
∂P ρ(100◦C) = 0.598kg/m3 . The latten
>0⇒ >0
∂T coex. ∆V heat is L = 2.257 103 J/m3 . The den-
sity of the vapour is approximately
But you know that the volume occupated by the gas is greater then 1000 times the vapour of the liquid,
so ∆v = v g − vl ' v g ' ρ− 1
g . Let’s
the volume occupated by the liquid, so ∆V > 0 and so ∆S > 0 which
use the Clasius-Clapeyron equation
is coherent with what we already know.
∂P ∆s L Pa
It is possible to follow the same steps for the solid-liquid transition = = ' 3.6 103
∂T ∆v T∆v K
and in general you can get the same results because ∆V > 0, but As the first exercise
there are situations in which ∆V < 0, for example the freezing water ∆P
∆T = ' 18◦C
and in this case you have to conclude that ∂P
∂T

And so the boiling temperature on the

∆S
 
∂P
>0⇒ <0 Everest is Tb ' 80◦C.
∂T coex. ∆V

Order parameter, classification of phase transitions and critical ex-

ponent

In order to distinguished the critical points let’s define the order parameter
as a macroscopic variable O such that It will better discusses the case T = TC .
(
6= 0 T < TC
O=
= 0 T > TC

O is related to the symmetry of the system and when the symme-

try breaks O became 6= 0.
~ as an order
Figure 9: Example of M
parameter in an uniaxial magnetic
system. Observe that before the critical
temperature M 6= 0, then for every
temperature T > TC M = 0.
18

For example an order parameter for the magnetic system could be

M~ (figure 9), for liquid-gas transition could ∆ρ (Experimental data on
∆ρ could be provide by 1 ) and it could be also a tensor depending of
the symmetry.
What about the response functions at critical temperature? 1
V.S. Vorob’ev. The law of corresponding
states for the entropy of rare gases.
It is to study easy it for a magnetic systems. Since at T = TC , December 2003
∂H
→ 0 you get
∂M
∂M
χM = →∞ Figure 10: Graph of H versus M for an
∂H uniaxial magnetic system.
Substantially at the critical temperature, since H displays a flex
respect to M χ M → ∞. Similar situation is the one of the pressure,
that displays a flex respect to the volume and so
 
∂V
kT ∝ →∞
∂P T
Let’s provide some simple classifications of phase transition The thermodynamical approach is
specific for thermodynamics, in fact
Thermodynamical : This classification distinguishes between the there is latten heat if one of the variable
is the entropy, but in general systems
transition that develop latent heat, called 1th order transitions and
do not necessarily depends by the
the others that do not develop latten heat; entropy.
The Ehrenfest’s classification is
Ehrenfest : A phase transition is of order n if all n − 1 derivatives are more complete, but do not keep in
continuous but nth derivative displays a finite discontinuity; consideration that sometimes it could
been a divergence instead of a jump.
Modern : It is called 1th order transitions that transitions that display There is not transition that display
a finite discontinuity in the first derivative and it is called continu- divergences on the first derivatives.
ous all the other transitions, so that a critical point is a continuous
phase transition.

We have seen that the behavior of an observable near a critical tem-

perature could be different depending on the situations. Let’s con-
sider the variable
T − TC
t=
TC
which represent a sort of undimensional distance from the critical
point. Let’s define a critical exponent λ associated to an observable
F (t) the quantity
ln | F (t)|
λ± = lim
t →0± ln |t|
or equivalently the asymptotic behavior of F (t) close enough at TC is Or more elegantly let’s use the expan-
sion
| F (t)| ' |t|λ± 
| F (t)| = A|t|λ± 1 + btλ1 + · · ·


Let’s show some examples in thermodynamics with λ1 , · · · > 0.

exponent β : Referring to figure 9 you can define β such that

| M| ' (−t) β t → 0−
 19

It has no sense to consider t → 0+ because we would be wonder-

ing how the curve approach 0 from 0.

∂M
χM = →∞
∂H
exponent γ : According to figure 11 let’s define
Figure 11: Reppresentation of χ M
t →0− t →0+
χ M ' − t − γ− χ M ' t γ+ varying T near the critical point TC .

but it is possible to show that γ+ = γ− =: γ and so

χ M ' | t |−γ γ>0

near the critical point.

exponent α in the same way we construct the γ exponent we can

construct the α exponent

C H ' | t |−α

exponent δ We do not necessarily use t as the parameter, for example

we can use M as the parameter for H so we can define
H →0 1
H ' Mδ sign( M ) | M| ' H δ

These exponents show a particular propriety: they do not depend

on the details of the system. In order to understand let’s consider
the Guggenheim experiment in figure 12. In the picture for different
T ρ
material we put on y−axies and on x −axies . The conclusion Figure 12: The Guggenheim experi-
TC ρC ment’s results.
is that every material have the same behavior and the data collapses In order to have more details we
in an unique curve (for all material) where suggest to read
E. A. Guggenheim. The principle of
1 corresponding states. December 2004
ρ L − ρG ' (−t) β
β=
3
Another example of universality is reported in 2 , a graph which
synthetize the situation is reported in figure 13.
This phenomena is not clear at this level, but it could be under-
stood by the formalism of renormalization groups!
Let’s introduce some inequalities between critical exponents;
considering the relation for uniaxial magnetic system
 2
T ∂M
cH = + cM Figure 13: Another example of the
χT ∂T universality
H   is reported in this graph
T ρ
and since at thermal stability c M ≥ 0 one can get , . As you can see all data
TC ρC
 2 collapses in an unique curve indepen-
T ∂M dentely on the microscopic detail (in
cH ≥ this case the material).
χT ∂T H 2
V.S. Vorob’ev. The law of correspond-
ing states for the entropy of rare gases.
December 2003
20

This inequality is true for all T, in particular for T → TC− .Recalling

the definitions of the critical exponents one can conclude that two
positive amplitudes B, B0 exist, such that We have to recall that close enough the
critical temperature
h i2
( TC − T ) β−1 CH ' (−t)−α χ T ' (−t)−γ M ' (−t) β
B ( TC − T )−α ≥ B0 T and so when we derive M in T
( TC − T )γ
∂M
' (−t) β−1
Rearranging the inequality and performing the limit T → TC ∂T

B0
lim ( TC − T )2−α−2β−γ ≥ T>0
T → TC B

In order to obtain a quantity that does not go to zero when you

perform the limit T → TC it is necessary to request that the exponent
of TC − T is grater than zero. In formula this is the Rushbnocke
inequality:

2 − α − 2β − γ ≥ 0 ⇒ α + 2β + γ ≥ 2

In similar way, from the convexity of A( T, V ) in T and V one can

show the Griffith inequality

α + β (1 + δ ) ≥ 2

In conclusion this inequalities show that the critical exponents are not
totally independent!
Recap of theory of ensamble

We have already recall the thermodynamical potentials. Let’s recall

some basic knowledge from the theory of ensamble. This is a recap, for more detail it is
Let Ω( E, V, N ) be the number of microstates with energy E and suggested
K. Huang. Statistical Mechanics. Wiley,
V, N fixed. Let’s assume the equal probability rule: if the system is second edition, 2008
isolated and at equilibrium with energy E it visits each microstates,
with energy E, with equal probability. As a consequence, according
to the probability theory rules, the probability of finding the system
in a configuration C is
1
P(C) =
Ω( E, V, N )

This is the probability of the microcanonical ensamble . In this

ensamble it is possible to find one of the most important equations in
statistical mechanics Where k B is the Boltzmann constant
k B = 1.38064852 10−23 m2 kg2 s−1 K−1
 
S( E, V, N )
Ω( E, V, N ) ∝ exp ⇒ S = k B ln(Ω)
kB
This equation tells that if you have the number of microstates you can
get the system’s entropy and, by the entropy all the other information
about the system.
Let’s suppose that the system is now in contact with a heat bath
such that the number of particles of the bath NB is grater that the Figure 14: Representation of the system
in an external bath.
number of particles of the system NS and the temperature TB of the
bath is constant. It’s necessary to assume that there are weak interac-
tions between B and S and that the bath is an isolated system. This
assumption is translated in mathematical terms by the conservation
of energy
ET = ES + EB = constant
The system can exchange energy with the bath at constant tempera-
ture. With some calculations you could get We defined the Boltzmann factor
  1
E(C) β=
P(C) ∝ exp − = exp (− βE(C)) kB T
kB T
P(C) is the probability of finding the system in configuration C of
the canonical ensamble . The probability P(C), in order to be a Sometimes it could be useful to remark
that this is a canonical isothermal
ensamble.
22

probability, has to been normalized. It should be easy to check that

the correct normalization is

Q( T, V, N ) = ∑ e−βE(C)
C

Q is called partition function . Notice that you still counting the

number of microstates, but this time you have to weigh the mi-
crostates with the exp(− βE) factor. Notice also that

Q= ∑ e−βE Ω(E, V, N )
E

and this can remember a Fourier transformation and so it’s possible

to consider the partition function of canonical ensamble as a Laplace
transformation of the partition function of the microcanonical en-
samble. We started with a microcanonical ensable and we keep the
temperature fixed, this should remember the Helmortz free energy
which, in fact, is provided by This should remind what we done for
the thermodynamical potentials.
A( T, V, N ) = −k B T ln ( Q( T, V, N ))

Now let’s change the volume of the system considering an isothermal

and isobaric ensamble. The two equations that you have are It could be not easy to imagine a system
in which could change the volume
ET = ES + EB = constant VT = VS + VB = constant of the system but the total volume
does not change, in fact there are some
surface effect that disappear when we
Also in this case it is possible to obtain the probability of finding the consider the limit VB → ∞.
system in the configuration C

P(C) ∝ exp (− β( E(C) − PV (C)))

And so the partition function is

∆( T, P, N ) = ∑ exp (− β(E(C) − PV (C)))

C

This ensamble is called Gibbs ensable .

Notice that also in this case it is possible to recognize a Fourier
transformation, in fact

∆( T, P, N ) = ∑ e−βPV Q(T, V, N )
V

As you could aspect, taking the logarithm of the partition function

you will find the Gibbs free energy

G = k B T ln (∆( T, P, N ))

This two examples show that at the Legendre transformations of the

potentials correspond the Laplace transformations of the partition
functions.
 23

The last ensable we want to introduce is called Gran canonical ensamble

and it is a canonical ensamble in which we assume that could be pos-
sible an exchange of particles between the system and the bath. The
probability of finding the system in a configuration C is

1
P(C) = exp (− βE(C) + βµN (C))
Ω( T, V, µ)

The partition function is Also in this case we could recognize a

Fourier transformation
Ω( T, V, N ) = ∑ ∑ e−β(E(C)−µN ) = ∑ Z N Q(T, V, N )
N C N

where
Z = e βµ
is the fugacity.
It shouldn’t be a surprise that the gran canonical potential is given
by
Ω = k B T ln (Ω( T, V, N ))
Let’s discuss the Gibbs ensamble for an uniaxial magnetic system.
E and M can fluctuate but the temperature is kept constant. Since
− P → H and V → M you easily check that

∆( T, H, N ) = ∑ e−β(E(C)− HM(C)) = ∑ e−βE+βHM Ω(E, M, N )

C E,M

It could be interesting to known what append if we think of a con-

tinuous set of configurations like the classical fluids. A configuration
in this case is given when we choose (~ri , ~pi )i=1,··· ,N and this is a mi-
crostate. The partition function is

1
Z
Q( T, V, B) = ∑ e−βE(C) = h3N N!
e− βH d3 p1 · · · d3 p N d3 r1 · · · d3 r N
C

Where H = H ({~r1 , · · · ,~r N , ~p1 , · · · , ~p N }) is the Hamiltonian of the

system.
A fluid can usually changes his volume so the native ensable is
Z ∞  
1
Z
∆( T, P, N ) = dV e− βPV 3N e− βH d3 p1 · · · d3 p N d3 r1 · · · d3 r N =
0 h N!
Z ∞
= dV e− βPV Q( T, V, N )
0

This formula is quite similar to the Fourier transformation.

Statistical mechanics and phase transitions

The connection between statistical mechanics and thermodynamics is

in the thermodynamic limit. The limit that send the number of
The question is: how can we get the divergence, that phase transi- particles to infinity.

tions display in thermodynamic, in statistical mechanics?

In order to answer to this question let’s consider a system Ω which
has a volume V (Ω) with a boundary ∂V (Ω). One can approximate
the volume using a typical length L

V (Ω) ' Ld

where d is the dimension of Ω. In principle Ω could be a continuous The dimension d is the dimension of the
space or a discrete space. system and not the dimension of the
place where the system is embedded in.
Let’s consider a finite system, this means that there are a finite
set of possible configurations; Let C be a configuration and {ci }i
the parameters that identify this configuration. One can write the
Hamiltonian
HΩ (C) = − ∑ k n Θn
n
where k n are the coupling constant which usually (but not always)
are intensive variable and Θn is a combination of the microscopical
degrees of freedom. In quantum mechanics Θn could be
The Hamiltonian has to be coherent with the symmetries of the local operators.

system, in particular every term k n Θn has to be coherent with the

symmetries.
In order to better understand the concept let’s introduce it for
the magnetic systems. It should be known that a system of N (Ω)
particles is described by the spins of the particles, so that {~Si }i is a
Figure 15: An example of a configura-
general configuration. tion for a 1−dimensional lattice model
A lattice model for a magnetic system is provided by figure 15. for a magnetic system. In every site of
Notice that in this case the dimension d is the dimension in with the the lattice there is a particle with its
spin represented by a vector (in this
spins lives and since they are 3 dimensional vectors d = 3. On the picture a 2−dimension vector, but it
other hand the system is enbedded in a one dimensional space. could be a 3−dimensional vector).
Some possible Θ are
1
Θ1 (C) = ∑ ~Si Θ2 (C) =
2∑
~Si · ~S j
i ij
26

The first term is not invariant under rotations and so there must be
a k1 such that the product k1 Θ1 is invariant under rotations. Θ2 is For instance a possible choice could be
invariant under rotation. We will use the trace to indicate the sum (or k 1 Θ1 = ∑ ~Si · H
~i
i
the integral) over all possible configurations
Tr → ∑
C
In this case
Tr = ∑∑··· ∑ ∑
S1 S2 S N −1 S N
This notation is useful to define the partition function
 
QΩ = Tr e− βHΩ (C)
Let’s introduce a fluid system to give another example. One should
know that a configuration is given when {(~xi , ~pi )}i is known. The
simplest Θ one can introduce is
!
~p2i
Θ1 = ∑ + Ui (~xi )
i
2mi
This is a one body potential: there is not any kind of interaction
between the particles. In order to introduce an interaction one can
add a term like
1
Θ2 = ∑ U2 (|~xi − ~x2 |)
2 i,j
The gran canonical partition function is
Z = Tr (exp (− βHΩ − βµN ))
Where the trace in this case is
Nmax N
1 dd x i dd p i
Z
Tr = ∑ N! ∏ hdN
N =0 i =1
In general one can compute the potential starting from the partition One can compute the gran canonical
function, for example one can compute the free energy potential
FΩ = −k B T ln Z
FΩ ( T, {k n }) = −k B T ln QΩ
Since the number of particle is finite this is a finite free energy.
Since FΩ is extensive one can suppose that there is a term propor-
tional to the volume, so that
 
FΩ ( T, {k n }) = V (Ω) f b ( T, {k n }) + S(Ω) f s ( T, {k n }) + O Ld−2

Where V (Ω) is the volume and S(Ω) is the boundary surface. Let’s
define the bulk free energy In the same way

FΩ ( T, {k n }) FΩ ( T, {k n }) − V (Ω) f b ( T, {k n })
f b ( T, {k n }) = lim f s ( T, {k n }) = lim
S(Ω)→∞ S(Ω)
V (Ω)→∞ V (Ω)
if the limit exists.
if the limit exists. It could be very difficult to show that this limit
exists and so usually we assume that it exists.
 27

Phase transitions and thermodynamical limit

Notice that when you compute the limit V (Ω) → ∞ you have to
impose that N (Ω) → ∞ otherwise the density

N (Ω)
ρ=
V (Ω)
goes to zero and the system is not interesting. So when one compute
the limit for V → ∞ have to impose also that ρ = const.
Since the system is finite the partition function

QN = ∑ e−βHΩ (C)
C

is a finite sum of exponential terms ans so it is an analytic function.

This means that it does not display singularities.
In conclusion before performing the thermodynamic limit the
system has no singularities and so no phase transitions, but after the
thermodynamic limit one can have singularities!
We have seen that the response functions display a divergence
in the critical point. One can show that this divergence is the result
of some microscopic fluctuations. Let’s consider a magnetic system.
There are two way to face the same problem: compute the thermic
fluctuation or perturb the system with an external filed and see how
the system response. The second way is more physical and so we
choose it. In Gibbs ensamble − HM is the work done by the system
against the field H.
Z [ T, {k n }, H ] = Tr (exp (− βH(C) + βM (C) H )) =
= ∑ ∑ e− βE+ βHM Ω( E, M)
M E

where H is the external magnetic field.

Since the average of an observable is defined by Let’s remember that
1
P(C) = exp (− βH(C) + βHM (C))
hOi = Tr (O(C) P(C)) Z

the average of M is
∂ 1
h Mi = ln ( Z ) = Tr ( M(C) exp (− βH(C) + βHM(C)))
∂( βH ) Z
So the magnetic susceptibility is by definition

∂ h Mi
χT = =
∂H
β   β   2
= Tr M2 (C)e− βH(C)+ βHM(C) − 2 Tr M(C)e− βH(C)+ βHM(C) =
Z Z
1 D 2 E 
= M − h M i2
kB T
Let’s introduce a magnetization density m such that

Figure 16: The magnetization density

can be interpret as a local average of M.
28

Z
M (C) = dd r m (~r )

Using this definition

Z
dd r dd r 0 m (~r ) m ~r 0 − hm (~r )i m ~r 0






k B Tχ T =

If the system is invariant under translations the density of magneti-

zation is constant for every ~r. Let’s define the connected correlation
function
G ~r −~r 0 = m (~r ) m ~r 0





and

Gc ~r −~r 0 = (m (~r ) − hm (~r )i) m ~r 0 − m ~r 0 = G ~r −~r 0 − m2




Using this two definition

Z   Z
dd r dd r 0 G ~r −~r 0 − m2 = dd r dd r 0 Gc ~r −~r 0



k B Tχ T =

This is the proof of the fluctuation-dissipation relation

1
Z
dd r dd r 0 Gc ~r −~r 0


χT =
kB T

Using the invariance under translation one can also find the propor-
tional constant between the χ T and the volume V (Ω)

1
Z
χ T = V (Ω) dd r Gc (~r )
kB T

If the system is also isotropic the Gc depends only by the absolute

value of ∆~r and if we consider a correlation length ξ one can assume
that
|∆~r |
 
Gc (|∆~r |) ∝ exp −
ξ
One can show that Z ∞ Explicit calculation:
− ξx
x2 e dx = 2ξ 3 Z ∞
− ξx x
−
∞ Z ∞
−x
x2 e dx = x2 
(− e ξ  − 2 (−ξ )
ξ)  xe ξ dx =

0 0
0  0
So that we can estimate the intagral in the fluctuation-dissipation ∞
 
− ξx  −x
Z
 + ξ e ξ = 2ξ 3
= 2ξ  −ξxe


 0
relation
k B Tχ T ∝ Vξ 3
For T → TC you have that χ T/V → ∞ and so ξ → ∞. This is an im-
portant result: the typical length of interaction display a divergence
in correspondence of a critical point.
One can think to define a critical temperature as the temperature
at which the typical length of interaction display a singularity. This
definition do not depend by thermodynamics and so it could be a
 29

good definition. Notice that since the ξ goes to infinity the particles
are all interacting and so the mycroscopic details of the system are
not so important. This is the deep meaning of universality. It is
possible to write Also in this case in principle there are
ξ ∝ t−ν ν+ and ν− the first for the behavior f
ξ coming from TC− and the second for
Also the behavior of Gc could be ruled by a critical exponent η de- ξ coming from TC+ . We will show that
ν+ = ν− = ν.
fined (because of some historical reasons) by
1
Gc (∆~r ) ∝
k∆~r kd−2+η
Since there is no length of correlation (because ξ goes to infinity)
there is correlation in every scale.
One can test experimentally this phenomena. The most evidence
is the critial opalescence: the study of the diffusion of the light by More information and details about
a fluid. Many experiments show that a transparent liquid became critical opalescence is provide by
S. M. Shapiro and H. Z. Cummins.
"milky" when the temperature approaches the critical temperature. Critial opalescence in quartz. October
The physical interpretation of this phenomena is that the length of 1968
correlation of the liquid is much smaller that the wavelength of the
light (λ ' 500 nm), but when the temperature goes to the critical
temperature the correlation length became of the same order of the
wavelenght of light and so one can observe the diffusion of light (the
liquid became "milky").
The main difference between statistical mechanics and thermo-
dynamics is the number of particles considered. The statistical me-
chanics works with a finite number of particle in a finite volume
. Finite number of particles until you
Let’s study the critical phenomena with a finite size system. For perform the thermodynamic limit.

instance let’s consider a fluid: one can use the distance between the
liquid droplets as a typical correlation length ξ. Obviously ξ = ξ ( T )
and in particular when T → TC Remember that
T − TC
ξ ' | t |−ν t=
TC

Let’s assume that our system has a typical size of L, so that ξ ≤ L.

There could not be any singularities and so one can not reach the real
critical temperature, however it is possible to calculate how far we are
from TC . If ξ 0 = 1 nm and

ξ = ξ 0 t−ν for T → TC

one can calculata that

 − 1
ξ ν
t'
ξ0

− ν −1
For normal fluids L ' 10 cm and so t ' 107 and experimen-
tally ν ' 0.5 so t ' 10−14 . So one can reach temperature very close to
the critical temperature.
30

The problem of the limit is important also in simulations, in fact it

is impossible to simulate with an infinity (or very large) number of
particles. An example of what happens is provided in figure 17.
Also in simulations one can see that if the system has a finite
number of particles there can not be any singularities.
Figure 17: Example of simulations
around the critical temperature with
N (Ω) < ∞ (in a simulation it is
impossible to reach N → ∞).
Models

There are two type of possible models. The first is the detailed mod-
els in which one can put a lot of microscopic information about the
system. Usually this models are not solvable (at least analitically).
The second type of model is the "coarse-grained" in which there are
only the fundamental informations and neglect many details using
only the dimensionality, the symmetries and the range of interaction
(with few parameters).
For instance one can study the DNA with a atomic description
and simulation (detailed model) or one can think of DNA as a finite
number of monomers that interact each others.
In order to study the phase transition phenomena we choose the
"coarse-grained" model infact we have already seen that close enough
to the critical points the physics does not depend by microscopic
details.
The most important and famous model is the Ising model .
Let’s discuss it for unixial magnetic systems: for every site of a
lattice, with N (Ω) sites where L(Ω) is the linear size, put a spin Si .
This spin can assume only the value of ±1. A configuration is shown Figure 18: A configuration in the ising
model for an unixial magnetic system.
in figure 18. The number of possible configurations is 2 N (Ω) .
The 1−dimensional Ising model does
Since if the temperature is too large the only important energy is not display any phase transition, but
k B T it is important to assume that the temperature is small enough the two dimensional does it. The
solution of the two dimensional Ising
such that the system could be sensitive to a magnetic field Hi . Let’s model wa found by Onsager in 1944. To
introduce also a interaction term Jij which represent the interaction more detail we suggest
between the i −site and the j−site. The Hamiltonian is Lars Onsager. A two-dimensional
model with an order-disorder transition.
N (Ω) N (Ω) N (Ω) October 1944
1
HΩ ({Si }) = − ∑ Hi Si −
2 ∑ ∑ Jij Si S j
i =1 i =1 j =1

In principle there could be other terms but the Ising model does not
take care of them. One can compute the partition function For example there could be a term like

− ∑ Γijk Si S j Sk
 
ZΩ ( T, { Hi }, { ji }) = Tr e− βHΩ ({Si }) Tr = ∑ ∑ ··· ∑ i,j,k
S1 =±1 S2 =±2 S N (Ω) =±1
but, as we will see, it does not satisfy
And so the potential is the symmetry of the system.


FΩ T, { Hi }, { Jij } = −k B T ln Z
32

One can show that if

∑ | Jij | < ∞
i6= j

then

1
f b T, { Hi }, { Jij } = lim FΩ
N →∞ N

exists. This means that if one assume that

 −σ
Jij = A ~ri −~r j 

the f b exists if σ > d. Notice that if the interaction is dipolar,

This show that if the system has d = 1 then f b can not exists and and so ' r −3 and the system is a
3−dimensional system there could be
in fact it is possible to show that the one dimensional Ising model do some problems.
not display any phase transitions.
Let’s assume that f b exists and
(
J the distance between i and j is 1 step
Hi = H ∀i Jij =
0 otherwise

The Hamiltonian became

−HΩ = J ∑ Si S j + H ∑ Si
hi,ji i

where
∑ =∑ ∑
hi,ji i j = i ±1

This means that the partition function is

  

Z= ∑ exp  β  J ∑ Si Sj + H ∑ Si 
{ Si } hi,i i

Let’s define the magnetization per spin that is

1
hmi =
N ∑ h Si i
i

One can check that Where

1 ∂FΩ F = −k B T ln ZΩ
hmi = −
N ∂H
hmi is an order parameter of the model.
This model has a Z2 symmetry. In fact the Hamiltonian is Z2
invariant
HΩ ( H, J {Si }) = HΩ (− H, J {−Si })
One can show that for every scalar function φ ({Si }):

∑ φ ({Si }) = ∑ φ ({−Si })
{Si =±1} {Si =±1}
 33

We will see that this symmetry at the critical temperature will be

spontaneous broken.
Let’s see what happens at the partition function: In the step with the ! we have used that
exp (− βHΩ (− H, J, {Si }))
ZΩ (− H, J, T ) = ∑ exp (− βHΩ (− H, J, {Si })) =
is a function of {Si } and so
{Si =±1}
! exp (− βHΩ (− H, J, {Si })) = exp (− βHΩ (− H, J, {−Si }))
= ∑ exp (− βHΩ (− H, J, {−Si })) =
{Si =±1}

= ∑ exp (− βH( H, J, {Si })) = ZΩ ( H, J, T )

{Si =±1}

The conclusion is that the partition function reflects the Z2 symmetry

ZΩ (− H, J, T ) = ZΩ ( H, J, T )

and in particular
f b ( H, J, T ) = f b (− H, J, T )

And so also the thermodynamic proprieties are Z2 invariant.

Let’s compute the hm( H )i:

1 ∂FΩ
hm( H )i = − (H) =
N (Ω) ∂H
1 ∂FΩ (− H )
=− =
N (Ω) ∂H
1 ∂FΩ (− H )
= = −hm(− H )i
N (Ω) ∂(− H )

And so
hm( H )i = −hm(− H )i ∀H
In particular this equation is true for H = 0

hm(0)i = −hm(0)i hm(0)i = 0

Since hm(0)i is an order parameter and it is everywhere null there

are no phase transitions.
Notice that in this calculation we have not performed the thermo-
dynamic limit (or if you prefer we have performed it at the end of the
calculations). We will see that the thermodynamic limit is the key to
have phase transitions, after performing the limit the Z2 symmetry
will be broken and hmi will show a phase transition.

1-dimensional Ising model

Let’s see in particular the 1−dimensional Ising model. There are

two possible formalism, the first formalism is the recursive method
and the second is the transfer matrix method. We quickly show
34

the first method and then we enter in the details of the transfer
matrix method which are more general, it is used also in numerically
simulations, and it is also common in other parts of physics (for
example in particles physics).
Let’s assume a Ising linear chain as in figure 18, and let’s assume
that the first and the last site are independent. This assumption
is called free boundary condition. Since the spin of every site can
assume only the value of ±1 there are 2 N possible configurations.
The interaction is contained in the Hamiltonian
N −1
H=J ∑ Si Si + 1
i =1

And defining k = βJ and h = βH one can write the partition We have assumed that H = 0 and so
function h = 0.
!
N −1
ZN = ∑ ∑ ··· ∑ exp k ∑ Si Si + 1 =
S1 =±1 S2 =±1 S N =±1 i =1

= ∑ ∑ ··· ∑ ek(S1 S2 +S2 S3 +···+S N −1 S N )

S1 =±1 S2 =±1 S N =±1

Now let’s add a site in the lattice in which there is a new spin S N +1 .
The partition function became In the step with ! it is used the fact that

∑ ekS N S N +1 = ekS N + e−kS N =

Z N +1 ( T ) = ∑ ∑ ··· ∑ eKS1 S2 +KS2 S3 +···+KS N −1 S N +KS N S N +1 = S N +1 =±1
S1 =±1 S2 =±1 S N =±1
= 2 cosh(kS N )
= ∑ ∑ ··· ∑ eKS1 S2 +KS2 S3 +···+KS N −1 S N ekS N S N +1 = and since S N = ±1
S1 =±1 S2 =±1 S N =±1
∑ ekS N S N +1 = 2 cosh (k )
= ∑ ZN ekS N S N +1 = S N +1 =±1
N +1
!
= 2ZN ( T ) cosh (k)

In this steps is fundamental the role of the free boundary condition.

You can think at ZN as Z1 which it has been added N − 1 sites

ZN ( T ) = ZN −1 ( T ) 2 cosh(k ) = ZN −2 ( T ) 22 cosh2 (k) = · · · =

= Z1 ( T ) 2 N −1 cosh N −1 (k)

Since
Z1 ( T ) = ∑ 1=2
S1 =±1

one can conclude that

ZN ( T ) = 2 N cosh N −1 (k)

It is so possible to find the free energy of the system

   
J
FN ( T ) = −k B T ln( ZN ( T )) = −k − BT ln 2 + ( N − 1) ln 2 cosh
KB T
 35

And so the bulk free energy is

  
1 J
f b ( T ) = lim FN ( T ) = −k B T ln 2 cosh
N →∞ N kB T

In order to find the magnetization, is useful to recall that

1
m = h Si i =
ZN ∑ Si e − β H
{Si =±1}

In fact since H is fixed to be zero it has no sense to derive F in H.

Let’s use the identity In order to prove this identity we have
to put Si = ±1 and Si+1 = ±1. So for
Si = ±1 and Si+1 = ±1
ekSi Si+1 = cosh(k) + Si Si+1 sinh(k)
ek = cosh k + sinh k
in the partition function for Si = ±1 and Si+1 = ∓1
e−k = cosh(k ) − sinh(k )
" #
ZN (k) = ∑ exp k ∑ Si Si+1 =
{Si =±1} i
N −1
= cosh N −1 (k) ∑ ∏ (1 + Si Si+1 tanh (k))
{Si =±1} i =1

Let’s study the generic term

(tanh k) M Si1 Si2 · · · Si M+1

Since the sum is over Si = ±1 for all i, all terms have a opposite term
that cancel it. In order to understand it you can think at all Si fixed
except one S j . There is the case S j = 1 and S j = −1 in the sum and
since the other Si are fixed this two terms cancel each other out. The
only terms that survives is for M = 0, in fact it is

∑ 1 = 2N
{Si =±1}

and so one can re-prove, in a different way, that

ZN ( T ) = 2 N (cosh(k )) N −1

But now let’s study

S J (tanh(k )) M Si1 Si2 · · · Si M+1

When one sum over all Si = ±1, again all the terms for M 6= 0 is null
for the same reason, but now since we have add S j also for M = 0
this terms is zero. This term is contained in the formula for m, infact
N −1
1 1
m = hS j i =
ZN ∑ Sj exp [− βH] = ZN ∑ Sj ∏ (1 + Si Si+1 tanh (k)) = 0
{S} {S} i =1
36

The magnetization is an order parameter for this system and since it

is null everywhere there are not phase transitions.
Let’s study again this system, but this time with the transfer
matrices.
This time let’s think at the system as a circle with N sites such that
the ( N − 1)th site coincide with the first site (S N +1 = S1 ) as in figure
19. This condition is called periodic boundary condition. The system
Figure 19: A configuration in the Ising
is described by a Hamiltonian model for an unixial magnetic system
with periodic boundary condition.
N N
− βH({S}) = k ∑ Si Si+1 + h ∑ Si
i =1 i =1

and so the partition function is

h h
ZN (k, h) = ∑ ∑ ··· ∑ ekS1 S2 + 2 (S1 +S2 ) · ekS2 S3 + 2 (S2 +S3 ) · · ·
S1 =±1 S2 =±1 S N =±1
h h
· · · ekSN −1 SN + 2 (SN −1 +SN ) · ekSN S1 + 2 (SN +S1 )
Where there is a product of N exponential terms. Each of this expo-
nential represent the interaction between the ith site and the (i + 1)th
site and one can see at it as a sort of local information. Noticed that
each of this exponential can be written as an element of a matrix. In
fact, defining
0 h 0
ekSS + 2 (S+S ) = hS|T|S0 i
one can see that
N h
N
ZN = ∑ ∏ ekSi Si+1 + 2 (Si +Si+1 ) = ∑ ∏ h Si | T | Si + 1 i
{ S } i =1 { S } i =1

Is simple to find explicitly the matrix T. In fact

h+1|T| + 1i = ek+h h−1|T| − 1i = ek−h h+1|T| − 1i = h−1|T| + 1i = e−k

And so ! By construction the matrix is symmet-

ek+h e−k ric!
T=
e−k ek−h

This is the transition matrix of the one dimensional Ising model.

In this formalism
 E 1  E 0
 (+)  (−)
Si = Si =
0 1
and the corresponding bra
D    D   
(+)  (−) 
Si  = 1 ? 0 Si = 0 1?

It is possible to check the orthogonality

  ! !
1  0   1 0 1 0
∑ | Si i h Si | = 0 0 1 + 1 1 0 = 0 0
+
0 0
=1
S =±1
i
 37

This condition has also a physical consequence, in fact

1 1 1

ZN (k, b) = ∑ hS1 | T|S2 i hS2 |T|S3 i hS3 | · · · |S N i hS N |T |S1 i =

z }| { z }| { z }| {

{S}
 
= ∑ hS1 | T N |S1 i = Tr T N
S1 =±1

With simple algebra one can show that the trace is the product of
the eigenvalue . Let λ+ be the larger eigenvalue of T and λ− be the In fact, let’s find the diagonal matrix
smallest. It should be clear that TD = P−1 TP

N N We know that P exist because T is a

ZN (k, h) = λ+ + λ− symmetric real metric.
 
N times
Moreover in general (also for n−dimensional transfer matrices) let   z }| {
N
Tr T = Tr TTT · · · T =
 
λ+ > λ− the two larger eigenvalues, one can show that
N N
+ ∑ λiN
 
N times
ZN = λ+ + λ−
 z −1 −1
}|
−1
{
−1 
PP TPP T · · · PP TPP  =
i = Tr 

So that, the bulk free energy is  

N −1
= Tr PTD P =
1
f b = −k B T lim ln ZN = !

N −1
  
N
N →∞ N = Tr TD P P = Tr TD
 N !  N !
1 λ− λj In the step ! we used the propriety
= −k B T lim N
ln λ+ 1+ +∑ = Tr( ABC ) = Tr (CAB).
N →∞ N λ+ λ+
j

= −k B T ln λ+

where we used the fact that λ+ > λ− > λi . The conclusion is that

f b (k, h) = −k b T ln λ+

The power of the transfer matrices is in its utility in different contests.

For instance if the Hamiltonian is

H = k 1 ∑ Si Si + 1 + k 2 ∑ Si s i + 1 Si + 1 Si + 3
i i

one can use a transfer matrix defined by

h S i S i + 1 | T | S i + 2 S i + 3 i = e k 2 Si Si + 1 Si + 2 Si + 3

Or one can think at an osservable that can assume 9 values Si =

1, 2, 3, 4, 5, 6, 7, 8, 9 and so the transfer matrix is a 9 × 9 matrix.
For example the diluted Ising model is a model for magnetic
system in which Si = ±1, 0, and the transfer matrix is a 3 × 3 matrix.
In general T is a (n + 2) × (n + 2) square matrix. Suppose that Si
assumes n + 2 possible values. The normalization is

∑ | Si i h Si | = 1
Si
38

At each site we can associate a diagonal matrix

Si = ∑ | Si i Si h Si |
i

Where Si are the eigenvalues. Let {λ+ , λ− , λ1 , · · · , λn } be the eigen-

value of T. So that in a appropriate basis {ti }i {ti } is the basis of eigenvector of T.

TD = PTP−1 = ∑ | ti i λi h ti |
i

Where |ti i is the eigenvector associated to λi . The partition function

is Let’s notice that the partition function
n depends only by the eigenvalue of T.
N
ZN = λ+ N
+ λ− + ∑ λ Nj
j =1

And so
1
f b = −k B T lim ln ZN = −k B T ln λ+
N →∞ N
Exists a useful theorem witch can be used in this context: the Perron-
Frobenius theorem. Let A be an n × n matrix (finite n < ∞). Let
all the element of A be positive. The theorem say that the largest
eigenvalue λ+ satisfies the following condition

• λ + ∈ R+ ;

• λ+ has no degeneracy;

• λ+ is an analytic function of all parameter of A.

The transfer matrix satisfy the hypothesis of the theorem and so λ+ is

an analytic function of the parameter and so also the bulk free energy
is analytic and so it can no display singularities. In other words there
are no phase transitions!
In this calculation we have considered a finite range of interaction,
so until the range is finite there cannot be phase transitions.
Let’s calculate the two point connected correlation function

Γ R = hS1 SR i − hS1 ihSR i

When Γ R is known one can compute the limit R → ∞ and, since ξ is

a typical correlation length it is expected
 
R
Γ R ' exp −
ξ

Using this definition

 
1
ξ −1 = lim − ln Γ R
R→∞ R

So the first goal is to compute hS1 SR i N By definition Where N indicates that you have to
compute it for a system with N finite.
Recalling
n +2
T= ∑ | ti i λi h ti |
i =1

and
Si = ∑ | Si i Si h Si |
Si
 39

1
h S1 S R i =
ZB ∑ S1 S R e − β H =
{S}
1
=
ZN ∑ S1 h S1 | T | S2 i h S2 | T | S3 i · · · h S R − 1 | T | S R i S R ·
{Si =±1}

· h S R | T | S R + 1 i · · · h S N | T | S1 i =
1
=
ZN ∑ S1 h S1 | T R − 1 | S R i S R h S R | T N − R + 1 | S1 i =
S1 ,SR
n +2
1
=
ZN ∑ ∑ S1 hS1 |ti iλiR−1 hti |SR iSR hSR |t j iλ N
j
− R +1
h t j | S1 i =
S1 ,SR i,j=1
n +2
1
∑ t j  S1 |ti i λiR−1 hti | SR t j λ N

   − R +1
= j
ZN i,j=1

So recalling that
n +2
ZN = ∑ λkN
k =1

the result is
R −1
n +2
  − R +1
= 1 t j S1 | t i i λ i h ti | S R  t j λ N


∑i,j 
j
h S1 S R i N = +2 N
∑nk= 1 λk

N , so that
Let’s multiply and divided by λ+ λ+ is the largest eigenvalue, this step is
useful to compute the limit R → ∞.

n +2
  R −1    λ j  N − R +1
S S


∑i,j
λi
t
=1 j 1 i
 | t i λ+ h t |
i R tj

λ+
h S1 S R i N =  N
+2 λ k
∑nk= 1 λ+

Let’s compute the N → ∞ limit:

hS1 SR i = lim hS1 SR i N = ht+ | S1 |t+ i ht+ | SR |t+ i +

R→∞
λ i R −1
 
+ ∑ ht+ | S1 |ti i hti | SR |t+ i
i6=+
λ+

Similar calculations can be done for hS1 ihSR i. Let’s compute the
generic term This time is useful to recall also that

1 ∑ | Si i h Si | = 1
h Sk i =
ZB ∑ Sk e − β H = Si

{S}
1
=
ZN ∑ h S1 | T | S2 i h S2 | T | S3 i · · · h S k − 1 | T | S k i S k ·
{Si =±1}

· h S k | T | S k + 1 i · · · h S N | T | S1 i =
1
=
ZN ∑ h S1 | T k − 1 | S k i S k h S k | T N − k + 1 | S1 i =
S1 ,Sk
40

And using the propriety as before

n +2
1
h Sk i =
ZN ∑ ∑ hS1 |ti iλik−1 hti |Sk iSk hSk |t j iλ Nj −k+1 ht j |S1 i =
S1 ,Sk i,j=1
n +2
1
∑ δij λik−1 hti | Sk t j λ N
  − k +1
= j
ZN i,j=1

As a conserquence

∑in=+12 λiN hti | Sk |ti i

h Sk i =
∑k λkN
As before
λiN
∑in=+12 λ+N h t i | Sk | t i i
h Sk i =  N
∑k
λk
λ+

And in conclusion
h Sk i = h t + | Sk | t + i
and so
hS1 ihSR i = ht+ | S1 |t+ i ht+ | SR |t+ i
So that   R −1 This is the correct correlation function
λi of two sites at distance of R sites in the
ΓR = ∑ λ+
h t + | S1 | t i i h t i | S R | t + i lattice.
i6=+
This function depends only by T. In fact there are only its eigenvalue
λi and its eigenvector |ti i.
It’s time to compute the typical correlation length
 
1
ξ −1 = lim − ln Γ R
R→∞ R−1
The sum over all eigenvalue in the R → ∞ limit are not relevant, in
fact the largest term (which dominate) is the terms such that i = −,
and so
λ − R −1
  !
−1 1
ξ = − lim ln ht+ | S1 |t− i ht− | SR |t+ i =
R→∞ R − 1 λ+
 
λ− 1
= − ln − lim ln (ht+ | S1 |t− i ht− | SR |t+ i)
λ+ R→∞ R − 1
The second term is vanishing. In fact the logarithm do not depend by
R. At the end one can conclude that
 
λ+
ξ −1 = ln
λ−

Let’s apply this general formula in the Ising model. The eigenvalue
of the transfer matrix are
q
λ± = ek cosh h ± e2k sinh2 (h) + e−2k
 41

And so the bulk free energy is Is simple to check that if h → 0 the

  result is similar to the result obtained
with the recursive method.
q
f b (k, h) = −k B Tk − k B T ln cosh(h) + sinh2 (h) + e−4k

There are some limits that is necessary to check. The first limit is for
T → 0. Doing this limit it is important to keep fixed H and J and
send k, h → ∞:
fb ' − J − H T→0
The fundamental result is that f b is constant when the temperature
approach the zero.
The second important limit is T → ∞ (k → 0 and h → 0):

f b ' − J − k B T ln 2 T→∞

We have already shown that the physics of the system is stored in the
order parameter. Let’s compute m

∂ fb ∂f sinh(h)
m=− = −β b = q
∂H ∂h
sinh2 (h) + e−4k

If you fix T > 0 and let H → 0 (or h → 0) you simply have that
m(h) → 0 for all T > 0 and so there is no spontaneous magnetization
in the one dimensional Ising model.
Otherwise there could be a ferromagnetic phase at T = 0. In fact One can compute also the limits for
T → 0 and T → ∞ of χ T . In the limit
∂m ∂m cosh(h) T → ∞ one can find the Curie law
χT = =β = q
∂H ∂h χT ' β T→∞
1 + e4k sinh (h) sinh2 (h) + e−4k
2

For the T → 0 one can compute that

Notice that at T → 0 the χ T diverges.
χ T ' βe2βJ T→0

Classical Heisemberg chain

Let’s now discuss a quantum model in which the Hamiltonian is

Ĥ = − Js ∑ Ŝi · Ŝi
hi,ji

The main difference between a classical system is that the spin opera-
tor do not commute. Since one want to solve the problem in any case From classical mechanics you should
let’s consider the case in which Ŝ → ∞. Let’s define recall that
 
h i
(k) ( j) (l )
Ŝi , Ŝi = iekjl Ŝi
Ŝ
ŝi =  i
Ŝ

Now the ŝi operator commute, in fact

(l ) 1
h i
(k) ( j)
ŝi , ŝi = iekjl ŝi   → 0
Ŝi 
42

 2
Defining J = Js Ŝi  the Hamilltonian is

H = − J ∑~si ·~si+1
i

So now the system seems a classical system. Let’s assume that ~si is a
versor and so it represents a point in a unitary sphere. The partition Notice for the future that there is a O(3)
function of the system is symmetry.

N Z
dΩi − βH
ZN = ∑e − βH
=∏
4π
e
{ si } i =1

where is used the spherical coordinates. Let’s use the transfer matrix
method to re-write the partition function.
 
ZN (k) = Tr T N

where
h~si | T |~si+1 i = ek~si ·~si+1
Since
T= ∑ t j
 

λj tj 
j

one can write

f j (~sm ) f j? (~sm+1 )

z }| {z
}| {
ek~si ·~si+1 = λ j ~sm |t j ~t j |sm+1 = ∑ λ j f j (~sm ) f j? (~sm+1 )
j

Notice that the left member in this equation is quite similar to a plane
wave Let’s use the hat for the versors.
ei~q·~r = eiqrq̂·r̂
in which
q̂ → ~sm r̂ → ~sm+1 iqr → k ⇒ qr → −ik
One famous result is that a generic plane wave can be written in
terms of Bessel function and spherical harmonics: This calculation should be already done
in some quantum mechanics course, if
∞ l it is not yet done we suggest
ei~q·~r = 4π ∑ ∑ ?
(i )l jl (qr )YlM (q̂)YlM (r̂ ) Haber. Expansion of plane waves
l =0 M=−l in spherical harmonics. URL http:
//scipp.ucsc.edu/~haber/ph215/
So by analogy one can conclude that PlaneWaveExpansion.pdf

∞ l
ek~sm ·~sm+1 = 4π ∑ ∑ ?
(i )l jl (−ik)YlM (~sm )YlM (~sm+1 )
l =0 M =−l

And matching this formula with the previous result one conclude
that
λi = λl (k ) = 4π (i )l jj (−ik )
 43

So the eigenvalue do not depends by m, but only by l. The largest

eigenvalue is
sin(k )
λ+ = λ0 (k) = 4π
k
And the next
 
cosh(k) sinh(k)
λ− = λ1 (k) = 4πi −
k k2

So now you can compute the bulk free energy.

In any case the Perron-Frobenius theorem do not let phase transi-
tion exist.
So one have to violate the hypothesis of the theorem. The simplest
way to do that is to take a Tij = 0 so that the transfer matrix is not
positive. An example is the Zipper model for DNA distribution.
This model is thought by Kittel to explain the denaturalization of
DNA.

Zipper Model

The DNA is composed by two strengths, each of one we can think

done by N bond. We know from biology that for T → 0 the stregths
tends to separate. This process is called DNA denaturalization. The
Zipper model, discussed the first time by Kittel in 1969, tries to
describe this process. This is only an exmaple to understand
In order to simplify the model one need some assumption. The how to have a phase transition. To have
a more details about Zipper model the
first assumption that Kittel propose is to let the ith bond open only if original article is
the (i + 1)th bound is opened. C. Kittel. Phase transition of a
Let’s consider the last bound always close. An example of a con- molecular zipper. 1969

figuration is provided by figure 20. Let’s take as an assumption the

fact that when a bond is broken the system increase its entropy. The
energy that the system receives when one bond is broken is e0 . Let’s
define for each bond a spin such that

S=0 if the bond is closed

S = 1, 2, · · · , G if the bond is opened Figure 20: A configuration in Zipper

model.
So that there are G possible value for every opened bond. This
assumptions are sufficient to write explicitly the partition function Let’s notice that, if there are k opened
bond there are G k configuration and the
N −1 N −1 energy of each configuration, and so the
ZN = ∑ G k e− βke0 = ∑ ekβ(ST −e0 ) Hamiltonian, is ke0 .
k =0 k =0

where S is the entropy

S = k B ln G
Let’s define the variable x such that It’s useful to recognize the geometric
series
N −1
1 − xN
∑ xk = 1 − x
k =0
44

N −1
1 − xN
ZN = ∑ xk =
1−x
x = Ge− βe0
k =0

This equation shows that there is a pole in the partition function. One
can compute the partition function

1 − xN
 
FN = −k B T ln ZN = −k B T ln
1−x

And with this definition one can define the order parameter of the
system
∑ kx k d Nx N x
hki N = k k = x ln Zn = N −
∑k x dx x −1 x−1
which represent physically the average number of opened bonds.
The explicit shape of this formula is provided by figure 21. Since
there is a flex at x = 1 one can except a singularity in the next
Figure 21: Behavior of the order param-
derivatives, and so it is. Let’s now compute the bulk free energy, it eter of the system.
can be useful re-write x = 1 + e with small e so that

1 − xN 1 − (1 + e ) N
   
ln ZN ( x ) = ln = ln =
1−x −e
N ( N −1) 2 N ( N −1)( N −2) 3
!
e 1 1 − 1 − eN + 2 e + 6 e +···
' ln =
−e
N−1 ( N − 1)( N − 2) 2
 
' ln N + ln 1 + e+ e +··· =
2 6
N 1 Ne N 2 e2
' ln N + + +···
2 24
The bulk free energy is
 
1 Ne
f b = lim ln N + +···
N →∞ N 2

And in similar way

Ne ( Ne)2
 
N
hki N ' 1+ − +···
2 6 360

And in particular
hki N 1
→ N→∞
N 2
The next derivative is This singularity is quite similar to the
magnetic susceptibility!
1 dh k i N N N 3 e2
Y= ' −
N de 12 240
which diverges when e = 0 and N → ∞ (linearly in N). So we have
discovered a phase transition when
e0
1 = x = Ge− βe0 ⇒ TC =
k B ln G
 45

So that if G = 1 there are no phase transitions, but if G > 1 there is

because TC 6= 0.
It is possible to re-discuss the zipper model with the transfer
matrix method. In particular one can define that if the spin is zero
then the energy is zero, but when the spin is not zero then

S i − 1 6 = 0 ⇒ E = e0 Si−2 = 0 ⇒ E = e0 + V0

This is the mathematical form for our assumptions where we intro-

duce an interaction term V0 . In a compact formula



E(Si , Si−1 ) = e0 + V0 δSi−1 ,0 1 − δSi ,0

So that the Hamiltonian is

N −2
H N = e0 1 − δS1 ,0 + ∑ e0 + V0 δSi−1 ,0 1 − δSi ,0



i =2

And notice that by construction the first row of the transfer matrix
is zero when V → ∞. Since there is a row in which all the elements
are zero the hypothesis that Tij > 0 for all i, j is not true and so the
Perron-Frobenious theorem does not allow to derive any conclusion:
in general there can be some phase transition and, as we had already
seen, in this model there is one.
Dimensions, symmetries and range of interaction

Let’s discuss the role of some important physical quantities. In

the last chapter we have seen that if the hypothesis of the Perron-
Frobenius theorem are true then there cannot exist phase transitions.
The fisrt way to violate this hypothesis is to have a non-positive
transfer matrix as in the Zipper model. On the other hand one can
try to solve some problems using the Ising model in D dimensions
with D > 1. The 3−dimensional Ising model has no analytic solution
yet. But the 2−dimensional Ising model have a solution.

Role of dimension

Let’s now discuss the 2−dimensional Ising model: the goal of this
computations is to show how the dimension of the system can play a
role in phase transition, so in the calculations is important to observe
how the transfer matrix change.
Let’s consider the 2−dimensional lattice as in figure 22. Now the
Figure 22: The two dimensional lattice
spins are labeled by two indexes m and n. In particular m indicate the in which the Ising model lives.
column and the n the row. So that the spins now are

Sm,n = ±1

The periodic boundary condition now action twice: first one have to
assume that S M+1,n = S1,n and then Sm,N +1 = Sm,1 . As in the one Geometrically the boundary condition
dimensional model means that the lattice is a torus, in fact
the first condition means that one have
to "join" the begin and the end of the
− β H = k ∑ Si S j + h ∑ Si lattice and if one try to figure out this
hi,ji i procedure one understand that the
result is a cylinder. In the same way
but this time there are two indexes "closing" the cylinder one obtains a
torus.
N N M N
− βH = k ∑ ∑ (Sn,m Sn,m+1 + Sn,m Sn+1,m ) + h ∑ ∑ Sn,m =
n =1 m =1 m =1 n =1
M
= ∑ ( E2 (µm , µm+1 ) + E1 (µm ))
m =1

Where Notice that

µm = {S1,m , S2,m , · · · , Sn,m }
48

N N N
E1 (µm ) = k ∑ Sn,m Sn+1,m + h ∑ Sn,m E2 (µm , µm+1 ) = k ∑ Sn,m Sn,m+1
n =1 n =1 m =1

In this way one can recognize a formula that is similar to the one
dimensional case. The one body interaction is re-writable as
1
( E (µ ) + E1 (µ2 ))
2 1 1
Let’s use this to write the partition function
M  !
1
ZN,M = ∑ ∏ exp E2 (µm , µm+1 ) + 2 (E1 (µm ) + E1 (µm+1 )) =
{ µi } m =1

= ∑ h µ1 | T | µ2 i h µ2 | T | µ3 i · · · h µ M | T | µ1 i
{ µi }

A typical term is
 
1
hµm | T |µm+1 i = exp E2 (µm , µm+1 ) + ( E1 (µm ) + E1 (µm+1 ))
2

The important thing to notice is that T is a 2 N × 2 N matrix, so that

in the limit N → ∞ is an infinite matrix and so the Perron-Frobenius
theorem does not work As in the one dimensional case The fact that the Perron-Frobenius
  theorem does not work.
Z M,N = Tr T M

Now the goal is to find {λ+ , λ− , λ1 , · · · , λ2N −2 }. Remember that

The partition function λ + > λ − > λi

ZN M = ∑

M
λj (N)
j=±,1,···

As in the one dimensional case

1
f b = −k B T limln ZN M =
N,M →∞ N M
 
1 
M M
= −k B T lim lim ln λ+ + λ− + λ1M + · · · =
N →∞ M→∞ N M
 ! !!
λ− M λ1 M
  
1 M
= −k B T lim lim ln λ+ 1 + + +··· =
N →∞ M→∞ N M λ+ λ+
1
= −k B T lim ln λ+ ( N )
N →∞ N
With a lot of calculation that we will not do one can find
!
2N −1
N 1
2 i∑
λ+ ( N ) = (2 sinh(2k)) 2 exp ψi ( N )
=1

where  
iπ
cosh(ψi ) = coth 2k cosh(2k) − cos
N
 49

Let N goes to infinity. The sum became an integral and the conclu-
sion is that f b ( T, H = 0) is not analytic at T = TC where
 
2 2J J
2 tanh = 1 ⇒ TC ' 2.265...
k B TC KB
So one can see a phase transition at this temperature. Young in 1952
show also that
 !

 1− 1
  T < TC
m= sinh k2JT
 B
0 T > TC


1
From this results one can conclude that β =. One can also compute
8
that    
T
CH ∝ A − ln 1 − +B
TC
So that the specific heat displays a logarithm divergence for the
2−dimensional Ising model. So that

C ' t−α ⇒ α = 0

Is important to notice that the fact that the lattice is bi-dimensional

have as a physical result the presence of a phase transitions. This
means that the dimensionality plays a fundamental role in statistical
mechanics.
Another way to see how important is the dimension is the Energy-
Entropy competition argument. The concept is very simple: low
energy state can be stable with respect to thermal fluctuations but
fluctuation will destroy the long range order. More precisely, since

dF = dU − TdS = 0

dU and TdS are in a sort of "competition". Let’s perturbe the system

near an equilibrium state.
In the (one dimensional) Ising model special equilibrium states are
Figure 23: The two ground states in the
the ground state which are the states in which the spins are all +1 or one-dimensional Ising model.
all −1.
The energy of the ground state in which the spins are +1 is EG =
− JN and the entropy is SG = 0 because there are only one possible Notice that the ground state are the
configuration. more "ordinate" states, in fact SG = 0.

A perturbation can be visualized as a wall such that before the

wall the spins are all +1 and after the wall the spins are all −1. This
wall is called domain wall . The differences in energy is given by Figure 24: A domain wall in the one-
E G dimensional Ising model.
∆E = E − E = − J ( N − 1) + J − JN = 2J
Where E E is the energy of the excited
state.
If one perturb with M domain walls the energy variation is

∆E M = 2MJ
50

Let’s rewrite M as
M = xN

so that if x = 0 the system is in a ground state, if x 6= 0 it is not. For

M domain walls the variation in entropy is more complicated. The
number of configurations is
 
N N!
=
M M!( N − M )!

and so  
N
∆S = S M = k B ln
xN
The free energy variation is Where we used the Stirling approxima-
tion
ln k! = k ln k − k
 
N
∆FN ( x ) = ∆E M − T∆S M = 2MJ − k B T ln = that is valid when K → ∞.
Nx
' N [2Jx + k B T ( x ln x + (1 − x ) ln (1 − x ))]

The new equilibrium is at

∂∆F 1
=0⇒x=
∂x 1 + e2βJ
Since x 6= 0 for every T 6= 0 then exists a finite number of domain
walls and so there are no order phases for T 6= 0 and so no phase
transitions. This because U ∝ x and S ∝ x ln x so that the entropy is
always "more important" than U and so there is no "competition".
In D > 1 the domain walls can be a straight line or a more com-
plicated path. Now the calculations are more complicated and we do Figure 25: Examples of domain walls in
two-dimensional Ising model.
not do it, but the results are that

∆E ∝ 2JL D−1 S∝L

So that in 2 dimension both energy and entropy are proportional to L

and so, since
∆F = ∆E − T∆S

there is a possibility that at a certain temperature ∆F change its sign,

or in other words there is the possibilities of phase transitions.
This argument is not so rigorous, to do it better one can think at
the Peierls argument. The idea of this rigorous argument is that for N We provide only the idea, for more
finite and Z2 symmetry we known details .
Robert B. Griffiths. Peierls proof
of spontaneous magnetization in a
hmi N = 0 ∀N two-dimensional ising ferromagnet. 1964

On the other hand

lim hmi N → 0 T < TC 6= 0

N →∞
 51

So that the symmetry has to be broken in some way. One way to

show it is to use a small field to perturb and then one can take the
limit H → 0 (so that h → 0). So that

hmi = lim lim hmihN

h →0+ N → ∞

An one conclude that the symmetry has to be broken. Another way

to do that, which is the Periles idea, is to use a boundary condition.
For example one can require that at the boundary of the system the
spins are +1. So that

h N+ i − h N− i h N− i
hmi N = = 1−2
N N
In order to show that the symmetry is broken one can show that In particular one can show that

h N− i x2 2 − x
h N− i 1 ≤
≤ −e ∀N N 36 (1 − x )2
N 2
where
infact x = qe−4Jβ
hmi N ≥ 2e ∀N
This method in which we don’t enter give a critical temperature very
similar to the real value. This method use both the dimension and
the symmetries of the system, and in fact the symmetries play a deep
role in phase transitions.

Role of symmetries

We have already seen some Hamiltonian like (for instance when

H = 0)
1
H = − ∑ Jij σi σj
2 i,j

which is Z2 symmetric which is generated by two matrices (1, η )

such that
1σi = σi = −ησi η 2 = 12 = 1
Z2 is not the only possibility for a discrete symmetry. For example
the Potts model assume that the spins can assume values for 1 to 9
and the Hamiltonian is

H p = − ∑ Jij δσi σj
i< j

The Potts model is invariant under permutations of the index, so that

the group of symmetry of the Potts model is the permutation group
(9−dimensional permutation group).
What if one consider a model in which the symmetry is continu-
ous? Let’s find out it in the specific case of the XY model . In this
Figure 26: A spin in XY model takes
values in S1 .
52

model the spins are unitary vector , so one can think at them as
variables that take values in the circle S1 .  
In mathematical form ~Si  = 1.
 
Let’s identify an angle θ that parametrizes S1 , sine the absolute
value of ~Si is 1 this angle represent a spin and in fact it is possible to
re-write the Hamiltonian in terms of angles

H = − ∑ Jij ~Si · ~S j = − ∑ Jij cos (θi − θ j )

i< j i< j

Since in the Hamiltonian only differences between the angles appear

if one perform a rotation of an angle α, θ → θ + α, the Hamiltonian
remain the same. So this model is O(2) invariant. In general if the
spins take values in Sn then the system is O(n + 1) invariant, for this
reason this models are also called O(n) models.
O(2) is a continuous group, in fact, while Z2 flips the spins, O(2) Figure 27: A flip of the spin in the XY
rotate it in a diluted way. Let’s consider a interaction acting only if model can be reach with a "diluted"
transformation.
j = i ± 1 with a typical length L → ∞. For simplicity let’s assume that
Jij = − J for every i, j (such that j = i ± 1). If the differences between
the angles are small enough The differences is small when the
system is "ordered" which means for
1 small temperature. For this reason
cos(θi − θ j ) = 1 + (θi − θ j )2 + h.o. this is also called low temperature
2 approximation.
The Hamiltonian can be re-written as
 
1
H ' J ∑ 1+

2
θ − θj
hi,ji
2 i

But now notice that θi − θ j ∝ ∂ x θ if i and j are neighbors in the x

direction. So The Hamiltonian can be written as
J
Z
H = E0 + dd r (grad θ )2
2
Comparing the previous equation one find that

E0 = 2JN

Since the Hamiltonian is known one can compute the partition

function Z  Z  The notation Z
βJ Dθ
Z = e− βE0 D θ exp − dd r (grad θ )2
2 is used because θ is not a variable but a
functional, and so D θ is the integration
The energy that depends by the angles is called slitness energy and it
over all the functional θ.
is This ES is the analog to ∆E we defined
1
Z
ES = J (grad θ )2 dd r in the discrete case, in fact conceptually
2 we are following the same steps (but in
the continuous case).
There is nothing to do if one does not explicit the θ, a good assump-
tion is
2πnx
θ (~r ) = θ ( x ) =
L
 53

where n is an integer. We assumed that θ depends only by x so the

energy contribution of the other direction is a L D−1 factor, so
2
J D −1 L
 
d 2πnx
Z
ES = L dx =
2 0 dx L
 
J 2πn
Z
= L D−1 dx = 2π 2 n2 L D−2
2 L

So the energy goes, in the O(n) models, as L D−2 while we have seen
that for the Z2 symmetry it goes as L D−1 . In XY model the entropy
always wins the energy-entropy competition. If D ≤ 2 there are no
ordered phases and so no critical points.
However there is the possibility to have some more complicated θ
and, perhaps, phase transitions. The idea is to minimize the action,
solving the Euler-Lagrangiana equation that in this case is simply the
Laplace equation
∆θ = 0
And we know that θ =constant are not the only solutions.
The non-trivial solutions depends on the boundary condition and
in general are called vortex. They satisfy
I
grad θ d~l = 2πn Figure 28: Representation of a vortex
(n = 1).
where n is called charge of the vortex. One can compute the energy of the
Sometimes a vortex and a anti-vortex are generated in the same vortex assuming θ (~r ) = θ (r ), so that

position and the phase is so ordered, but at a certain temperature 2πn = 2πr grad θ (r )
the two separate. This is a sort of phase transition called Kosterlitz- and so
Thauless transition. Jn2
Z 2π Z L
1 L
Es = dθ rdr = πn2 J ln
So now the role of symmetries and dimension are discussed, what 2 0 a r2 a

about the role of range interaction? that show a logarithmic divergent.

Role of range interaction

Let’s come back to the Ising model, and let’s assume that
J
Jij = 
~ri −~r j α


If 1 ≤ α ≤ 2 there are phases with a long interaction range order, for

temperature 0 ≤ T ≤ TC this is true also in D = 1.
The Ising with this Jij shows phase transitions, and this is not a
problem for the Perrot-Frobenius theorem, in fact the transfer matrix
is infinite. If α > 2 the system goes back to the Ising in which the
only interaction concern neighbors spins. α < 1 is not considered because the
Let’s now discuss the Ising model with an infinite range. This the- thermodynamic limits does not exist for
this values of α.
ory is also called Hamiltonian mean field theory. The Hamiltonian
54

is
J0
−H =
2 ∑ Si S j + H ∑ Si
i,j i

The first problem one have to face is the fact that the first term of
this Hamiltonian is of the second order in N, so it seems that the
thermodynamic limits does not exists. In order to avoid this problem
let’s define
J0
J=
N
and let’s consider the Hamiltonian
J
2∑
−H = Si S j + H ∑ Si
i,j i

The partition function is

!
βJ0
ZN ( J, H ) = ∑ exp 2N ∑ Si Sj + βH ∑ Si
{S} i,j i

But !2
∑ Si S j = ∑ Si
i,j i

and so
 !2 
 
k k 2
ZN = ∑ exp  2N ∑ Si + h ∑ Si  = ∑ exp 2N
x + hx
{S} i i {S}

where
x= ∑ Si
i

Now let’s use the Hubber-Stratonovitch identity In order to prove it let’s complete the
square
r Z ∞
kx2
   
Nk Nk 2 Nk 2 Nk  x 2 kx2
exp = exp − y + kxy dy Re k > 0 − y + kxy = − y− +
2N 2π −∞ 2 2 2 N 2N
so
to re-write the partition function r
Nk
Z ∞ 
Nk 2

exp − y + kxy dy =
2π −∞ 2
r Z ∞
Nk Nk y2 kx2
Z ∞
∑ e(h+ky) ∑i Si
2
e−
Nk
Z (k, h) = 2 Q(y) dy Q(y) = = e 2N e− 2 (y− Nx ) =
2π −∞ {S}
−∞
Z ∞
kx2 Nk z2
= e 2N e− 2 dz =
−∞
The Q(y) is r
kx2 π2
= e 2N
N Nk
Q(y) = ∑ e(h+ky) ∑i Si = ∏ ∑ exp ((h + ky) Si ) =
{S} i =1 Si =±1
N
= ∏ 2 cosh(h + ky) = (2 cosh(h + ky)) N
i =1
 55

In conclusion the partition function can be written as

Nk ∞ N L(k,h,y)
r
k
Z
ZN (k, h) = e dy L(k, h, y) = ln (2 cosh(h + ky)) − y2
2π −∞ 2
This is the simplest form of the partition function that one can find.
The idea now is to approximate the integral as the maximum of the
integrand: the more the function is spiked, the better the approxima-
tion is. This approximation is called Saddle point approximation .
Using it the partition function became
r i r Nk
Nk h
N L(k,h,y)
ZN (k, h) ' max e = e N L(k,hy)
2π y 2π

where y is the y that maximize L. In the limit of N → ∞ the saddle

point approximation is exact, so the result in the thermodynamic
limit is not an approximation, let’s compute the bulk free energy

ln( ZN )
f b (k, h) = −k B T lim = −k B T L (k, hy)
N →∞ N
The equilibrium condition is that the derivative in y of the bulk free
energy is zero, so This equation is not analytical solvable.

∂L
= 0 ⇒ y = tanh (ky + h)
∂y

What is y? To answer at this question let’s compute the magnetiza-

tion
∂ fb 1 ∂ ln ZN
m=− = lim =
H N →∞ βN ∂H
∂L (k, h, y)
= = tanh (ky + h) = y
∂h
So the y is the order parameter in this case the magnetization.
The equation for y became now an equation for m which is called
self consistent equation :

m = tanh(km + h)

The bulk free energy is

f b (k, h) = −k B T L (k, h, y) =
= −k B T L(k, h, m) =
k
= −k B T ln (2 cosh(km + h)) − m2
2
Mean Field Theory

This theory is an approximate theory which allows to find out im-

portant results. The more famous mean field theory are the Weiss’
theories for magnetic systems, the Vann der Wall’s theory for fluids
and the Flory’s theory for polymers. Unfortunately we will not discuss
The idea is to neglect the fluctuations or equivalently one can as- the Flory theory. For more detail we
suggest
sume the probabilistic independence of the degrees of freedom. Let’s Somendra M Bhattacharjee. Flory
discuss the Mean field theory for the Ising model. The Hamiltonian is theory for polymers. 2013
always We assume an uniform external field H.
1
H({S}) = − ∑ Jij Si S j − H ∑ Si
2 i,j i

The partition function is

ZN = ∑ e− βH
{S}

and since H is uniform

h Si i = h S i = m ∀i

Let’s re-write in a smart way the product Si S j

Si S j = (Si − m + m)(S j − m + m) = (Si − m) S j − m + m2 + m(Si − m) + m(S j − m)

The first term is the fluctuation one and so, in the mean field theory
one can neglect it, so
1 1   1  
∑
2 i,j
Jij Si S j ' ∑ m2 + m(Si − m) + m(S j − m) = ∑ −m2 + m(Si + S j )
2 i,j 2 i,j

Notice that now this term is linear both in Si and S j . Let’s define
J i = ∑ j Jij so that

1 2
H= m ∑ J i − m ∑ J i Si − H ∑ Si
2 i i i

If J i does not depends on i one can define a J equal for all i, so that
1 2
m N J − mJ + H ∑ Si


H=
2 i
58

Let’s assume that the interaction takes place only if the two sites are
neighbors, so that
1
2∑
Jij Si S j → ∑ Si S j Jij
i,j hi,ji

Defining The sum

1 ∑
∑ Jij = z Ĵi ⇒ Ĵi =
z ∑ Ji,j j∈n.n. o f i
j∈n.n. o f i j∈n.n. o f i is the version of
where z is the number of neighbor bonds, one can re-write the Hamil- ∑
hi,ji
tonian as
1
H = m2 Nz Ĵ − mz Ĵ + H ∑ Si when i is fixed.


2 i

And so the partition function is

m2
ZN = e− Nβ Ĵz 2
∑ eβmz Ĵ ∑i Si +βH ∑i Si =
{S}
!N
2
− Nβ Ĵz m2
=e ∑ e β(mz Ĵ + H )S =
S=±1
m2
= e− Nβ Ĵz 2 N cosh N βmz Ĵ + βH


2

And the bulk free energy It can be useful to consider some

dimensionless variables
FN 1 FN kB T H
f = lim = lim (−k B T ln ZN ) = f? = θ= H? =
N →∞N N →∞ N Nz Ĵ z Ĵ z Ĵ
1 and using them
= Ĵzm2 − k B T ln cosh( βmz Ĵ + βH ) − k B T ln 2


2 1 2

m + H?

f? = m − θ ln − θ ln 2
The order parameter, which is the magnetization, can be computed 2 θ

∂f β

m=− =
tanh βmz Ĵ + βH
∂H β


And so the self-consistent equation is In some books mz Ĵ is called molecular
field.



m = tanh β mz Ĵ + H

and the equilibrium value of m has to satisfy it. Let’s assume H = 0.

The equation for m cannot be solved analytically, but it could be
visualised in a graph (figure 29).
What we discover is that for βz Ĵ < 1 there is only one solution
Figure 29: Representation of the
which is m = 0 and it means that there are no phase transitions. behavior of the solution of the self
However when βz Ĵ > 1 the solutions are three (m = 0, m = ±m0 ) consistent equation: notice that if
that collapse in one solution when βz Ĵ → 1. βm Ĵz < 1 there is only one solution but
if βm Ĵz > 1 there are three solutions.
This argument prove that the critical temperature is

z Ĵ
β( TC )z Ĵ = 1 ⇒ TC =
kB
 59

This temperature contains all the information (dimension in z, sym-

metry in Ĵ and z and range of interaction in Ĵ).
When the purpose is to compute in an analytic way it’s useful to
expand the free energy in m ' 0, so that
A 2 B 4
f b = const. + m + m + h.o.
2 4
First of all notice that f b is even, and this is due to the fact that the
system is Z2 symmetric. The calculation show that

4
2 Ĵz
B=β A = Ĵz(1 − βz Ĵ )
3
So B is always positive but it is not the same for A that can change
its sign. Obviously A = 0 is the condition to have a critical point, in
fact
A = 0 ⇒ βz Ĵ = 1
Notice that f (|mo |) = f (−|m0 |) so this phase transition is of the first Figure 30: Behavior of f b respect to m in
order, in fact the phases corresponding on the two values (±m0 ) can the case in which A > 0 and A < 0.
coexist.
In this approximation one can compute all the critical exponent.
From the equilibrium condition
∂f
=0
∂m
one can derive

Am0 + Bm30 = mo ( A + Bm20 ) = 0

And using the correct expression for A and B

k B TC 1
( T − TC ) m0 + Bm30 = 0 ⇒ mo ∝ ( TC − T ) 2
T
and by definition Recalling:
1 m ∝ tβ
β=
2
In a similar way
m3 m5
β Ĵzm + H = tanh−1 (m) = m + + + h.0
3 5
And so Recalling:
H ' m3 ⇒ δ = 3 H ∝ tδ

One can find out α, γ, · · · and all the critical exponent. The results of
the mean field theory for the Ising model are

1
β= α=0 δ=3 γ=1
2
This results are not exact if one compare them with the experimental
results. However this theory explain the qualitative behavior of the
system and provide some quantitative (but not exact) results.
60

Mean field variational method

The idea now is to find a ground state energy E0 such that

Eα = hψα | Ĥ |ψα i ≥ E0

The goal is to minimize ψα to get E0 . The Hartree method consist in

minimize a ψα which can be written as The notation (1) means that that state is
a one-particle state corresponding to the
N kth state.
∏ ψα
(1)
ψα = (~rk , k)
k =1

So the wave function is a product of single particle’s wave functions.

We will use two results

• Let φ be a random variable,then

D E
e−λϕ ≥ eλh ϕi P ∀P
P

• Let ρ( ϕ) be a probability distribution, then

e− βFN = ZN = Tr e− βH =
  D E
= Tr ρ( ϕ)e− βH−ln (ρ( ϕ)) = e− βH−ln ρ
ρ

This two proprieties lets conclude that

D E
e− βFN = e− βH−ln ρ ≥ e βhHiρ −hln ρiρ
ρ

so that

FN ≤ hHiρ + k B T hln ρiρ = Tr(ρH) + k B T Tr(ρ ln ρ) =: Fρ

Now the problem is to choose a density ρ that minimize Fρ . In fact

the probability density that minimize Fρ gives the equilibrium vari-
ables. There are many approaches to do that: first of all let’s use the
mean field approximation This formula means that we can
factorize at probability level.
ρ(φ1 , · · · , φN ) = ∏ ρ(1) (φα ) = ∏ ρα
α α

Let’s compute the second term of Fρ :

!!
Tr(ρ ln ρ) = Tr ∏ ρα ∑ ln ρβ = ∑ Trα (ρα ln ρα )
α β α

Where Trα means that the trace is over a single density ρα which is
fixed. Let’s conclude that

Fρ = hHiρ + k B T ∑ Trα (ρα ln ρα )

α
 61

Now the different approaches follow different step: the fist method
is to use hφα iρα , which is the local order parameter, to parametrize
ρα . Let’s use it for the magnetic system, this approximation is called
Bragg-Williams approximations. For magnetic systems φα 7→ Si =
±1 and so hφα i → hSi i = mi . The probability ρi = ρ(1) (Si ) is the
probability that Si is +1 and that Si is −1, in general one can write it
as
ρi = ai (1 − δSi ,1 ) + bi δSi ,1

where ai is the probability that Si is −1 and bi the probability that Si

is 1. The two conditions are This are two condition and so the ρi
can depends at least on two parameter.

∑ ρ i ( Si ) = 1 Tr(Si ρi ) = mi In this case it is true, but what if

Si = ±1, 0? In this case one can separate
Si =±1
the probability that Si = +1 and
consider Si = −1 or 0 as an unique
This two conditions can be written as event such that the probability density
ρi depends only on two parameters.
a i + bi = 1 bi − a i = m i

and solving this two equations

1 − mi 1 + mi
ai = bi =
2 2
The probability density is now

1 − mi
1 + mi
ρ i ( Si ) = 1 − δSi ,1 + δSi ,1
2 2
Only to check that all is correct let’s compute hSi iρ : one could find
mi . In fact for a generic function g(Si ) one
1 + mi 1 − mi can compute
h Si i = 1+ (−1) = mi
2 2 (i )
h g(Si )i = TrS =±1 ( g(Si )ρi ) =
i
Now one can simply compute the terms in the Fρ . The first is 
1 − mi
1 + mi

= ∑ g ( Si ) 1 − δSi ,1 + δSi ,1 =
S =±1
2 2
∑ hSi Sj i − ∑ Hi hSi i =
i
hH({Si })iρ = − J 1 + mi 1 − mi
hi,ji i = g (1) + g(−1)
2 2
= −J ∑ hSi ihSj i − ∑ Hi hSi i =
hi,ji i

= −J ∑ mi m j − ∑ Hi mi
hi,ji i

The second term is

hln ρi = Tr(ρ ln ρ) = ∑ Tr(ρ j ln ρ j ) =

j
1 − mi 1 − mi
    
1 + mi 1 + mi
=∑ ln + ln
i
2 2 2 2

Putting all together Recall that this result is a mean field

approximation.
62

F =−J ∑ mi m j − ∑ Hi mi +
hi,ji i

1 − mi 1 − mi
    
1 + mi 1 + mi
+ +k B T ∑ ln + ln
i
2 2 2 2

In order to find the equilibrium simply minimize F in mi , so that

 
∂F kB T 1 + mi
= 0 ⇒ 0 = − J ∑ m j − Hi + ln
∂mi J ∈n.n. of i
2 mi

Using the trick as before (zmi = ∑ J ∈n.n. of i m j ) one can can find And using
 
1 1+x
mi = tanh ( βJzmi + βHi ) tanh−1 x = ln
2 1−x

This is again a self-condition equation.

The second method consist in use ρi as a variable and derive
directly the potential in ρi . This is called Blume-Emery-Griffith
model (BEG model) for diluted Ising Si = ±1, 0. In this case one can
see a new type of critical point which is a tricritical point : while
a critical point is the end of a first order phase transition line, the
tricrical point is the end of a line of critical points.
The diluted Ising model can be mapped to the He4 − He3 mixture
undergoing fluid-superfluid transition, which is a mixture of bosons
(He4 ) and fermions (He3 ). What we see is that at a critical point of Explicitly to map the diluted Ising
2.17 K the He4 became superfluid (zero viscosity liquid). model simply think that Si = ±1 is the
He4 and Si = 0 is the He3 .
This transition is called lambda transition because of the shape of
the C respect to the temperature.
If one insert some He3 , for example in He4 , with a concentration
of x, the critical temperature Tλ depends on x. What experimentally
one can prove is that at xt ' 0.67 the λ transition became a first order
phase transitions.
Let’s go back to the BEG model to explain this. Let’s use hSi i = mi .
Notice that hSi2 i ∝ n4 which is the concentration of He4 , and so Figure 31: Behavior of C respect to T in
a λ transition.
x = n3 = 1 − hSi2 i. Let’s define a parameter ∆ which is proportional to
the difference µ3 − µ4 . The Hamiltonian of this system is

H = −J ∑ Si Sj + ∆ ∑ Si2 + ∆N
hi,ji i

And the Gibbs potential is

1
2  
G = − N Jz TrSi (ρi Si ) + N∆ TrSi ρi Si2 − N∆ − Nk B T TrSi (ρi ln ρi )
2
Now simply consider the derivative Explicitly

∂G MF ∂
2
TrSi (ρi Si ) = 2 Tr(ρi Si )Si
=0 ∂ρi
∂ρi
The solution for ρi is

ρ(1) (Si ) = A−1 e β(zJmSi −∆Si )

2
 63

And the correct A come from the condition that

TrSi =0,±1 ρi = 1 ⇒ A = 1 + 2e− β∆ cosh( βzmJ )

And simple calculations show that x = A−1 so

G z
g= = Jm2 − ∆ − k B T ln A
N 2
In order to study this solution let’s expand for m  1:
c
g( T, A, m) = a0 ( T, A) + am2 + bm4 + m6 + h.o.
6
where     Notice that we always stop the expan-
zJ zJ δ sion when a coefficient is always grater
a= 1− b = k 1− c>0 the 0.
2 k B Tδ 3
1
a0 is only a constant and δ = 1 + e β∆ . The λ transition is defined by
2
zJ ZJ
a=0⇒δ= ⇒ Tλ = = Tλ ( x = 0)(1 − x )
k B Tλ kB δ
But now when a = 0 the sign of g is not fixed but depends on b, so
that the tricritical point is defined by

a=b=0⇒δ=3

And so a simple estimation for the concentration of He3 is

δt − 1 2
xt ' =
δt 3
And the experimental result is xt ' 0.67, so the tricritial point is
given by the mean field theory with a good precision.
Another approach consist in using a generalization of the Hubber-
Stratonovitch identity and the saddle point approximation. Let’s see
it in the Ising model:
1
2∑
− βH = k ij Si S j + ∑ hi Si
ij i

and let’s call k ij = βJij and hi = βHi so that the partition function can
be written as
ZN = ∑ exp (− βH)
{S}
Let’s focus the attenction on the term
1
e 2 ∑i,j kij Si S j

This can be rewritten in term of a generalization of the Hubber-

Stratonovitch identity
1 !
e− 2 xi Aij xi + xi bi
Z N
1 1 b A −1 b
∏ dxi √
2π
= √
det A
e 2 i ij j
i =1
64

Now if one identify bi 7→ Si and xi 7→ φi one can write Is useful the equation
1
det k−1 N 1
∑i,j φi (k−1 )ij φj +φi Si det k−1 =
Z
1
e 2 ∑ij Si kij S j =
(2π ) N R N i =1
∏ dφi e 2 det k

And so the partition function is

 
N
1 − 21 ∑i,j φi ( k −1 )ij φj
Z
ZN = p
det k(2π ) N
∏ dφi e
R N i =1
∑ e∑i (φi +hi )Si  =
{S}
!
N
1 − 12 ∑i,j φi (k−1 )ij φj
Z
= p
det k(2π ) N
∏ dφi
R N i =1
e ∏ 2 cosh(φi + hi ) =
i
N  
1 − 12 ∑i,j φi (k−1 )ij φj +∑i ln(2 cosh(φi +hi ))
Z
= p
det k(2π ) N
∏ dφi
R N i =1
e =

N
1
Z
= p
det k(2π ) N
∏ e−L(φi ) dφi
R N i =1

Where
1  
L= ∑
2 i,j
φi k−1 φj − ∑ ln (2 cosh(φi + hi ))
ij
i

Using the saddle point approximation The minimum of L correspond to the

maximum of e−L
0
ZN ' A N e−L(φ1 ) L(φi0 ) = min L(φi )
φi

Where A N is only a constant.

Now the goal is to find φi0 , to do that simply

∂L   
∂φi φi =φ0
= 0 ⇒ ∑ k −1
kj
− tanh(φk0 + hk ) = 0
i j

Let’s multiply this equation by a term k ik and sum over all k:

!
 
∑ kik ∑ k −1
φj − ∑ k ik tanh(φk + hk ) = 0
0
kj
k j k

This means
φi0 = ∑ kij tanh(φ0j + hk )
j

The free energy is

FN = k B T L(φi0 )
But what is φi0 ? To answer let’s compute the local magnetization
∂FN ∂F
mk = − = − β N = tanh(φk0 + hk )
∂Hk ∂hk
Using it one can obtain
!
φi0 = ∑ kij m j ⇒ mi = tanh ∑ kij m j + hi
i,j j
 65

This is the general self-consistent equation. In the case in which

Hi = H (so that mi = m for all i) and ∑ j k ij → kz one can show that We have already seen it in the previous
chapter.

m = tanh(mJz + h)

So let’s re-write the free energy in the case in which hi = 0 (no

external fields), the first term is

kB T    
2 ∑ φi0 k −1
ij
φ0j − ∑ ln 2 cosh(φi0 ) =
i,j i
! !
k T  
=− B
2 ∑ ∑ kik1 mk1 k −1
ij
∑ k jk2 mk2 =
i,j k1 k2
kB T
= ∑ kik1 mk1 k−ij 1 k jk2 mk2 =
∑ i,k1 ,k2 j
kB T
=
2 ∑ kij mi m j
i,j

The second term is To show this recall that

1 1−x
tanh−1 ( x ) =
 
ln =t
   2 2 1+x
∑ ln 2 cosh tanh−1 (mi ) = ∑ ln  q 
so that
i i 1 − m2i
2 cosh t = et + e−t =
r r
Finally the free energy is 1+x 1−x
= + =
1−x 1+x
kB T kB T 2
s s
FN =
2 ∑ kij mi m j + 2 ∑ ln q =
1 − x2
(1 − x ) 2
+
1 − x2
(1 + x )2
=
i,j i 1 − m2i
2
= √
1 − x2
Van der Walls theory

Let’s now apply the mean field theory to fluid system (not ideal
fluids). Let’s simply introduce
!
Z N N
Q N (T ) =
V
∏ d3 ri exp − βφ ({~ri }) − β ∑ ψext (~ri )
i =1 i =1

where ψext is an external field and φ is the interaction between parti-

cles. Let’s assume, in the mean field approximation, that φ does not
depend on the position of each particles, but only on the position one
compute it
φ = φ (~r )

So that, if ψext = 0 and the fluid is isotropic (spatially) The isotropy means that Q (~r ) = Q(r ).

Z N
Q N (T ) = d3 r exp − βφ(r )
V

Figure 32: Example of interaction.

66

The simplest version of interaction term is provided in figure 32.

Considering this interaction
  N
4 3
3 πr0
z}|{
V − Vexc  e− βu 
 
Q( T ) = 
  

And so
FN ( T ) = − Nk B T (ln (V − Vexc ) − βu)
the higher the density is the higher the interaction is. Since the
interaction has to be attractive a good approximation for u is

aN
u=− a ∈ R+
B
And since Vexc = bN, where b is the volume of a single particle, one
can simply conclude that
 2
∂FN Nk B T N
PN = − = −a
∂V V − bN V

This is the Van der Walls equation of state.

V P
Usually it is written in terms of v = N and p = N so that
Figure 33: Representation of p respect
k T a to v in the three cases in which T < TC ,
p= B − 2 T = TC and T > TC .
v−b v

Considering figure 33, one can see that for T < TC the p constant has
three solution of the equation of state. This three solution merge to
one single solution when we approach the critical temperature. In
fact the critical temperature show a flex

∂p ∂2 p a 8a
= 2 = 0 ⇒ vC = 3b, pC = 2
k B TC =
∂v ∂v 27b 27b
Notice that exist a combination of pC , vC and k B TC that does not
depend by the microscopic detail (which are represented by a and b)
which is
pC vC 3
=
k B TC 8
This quantity is universal, and notice that defining

p v T
π= ν= τ=
pC vC TC

the equation of state does not depend on a and b and so it does not
depend by microscopic details
 
3
π + 2 (3ν − 1) = 8τ
ν
 67

It seems all good but there is a problem. There is a region (for T <
TC ) in which ∂π
∂v > 0. This means that when one increase the volume
the pressure increase which has no physical meaning, and in fact the
real curve is different. To solve this problem let’s recall that
F + PV S V
µ= dµ = − dT + dP = dg
N N N
So an isothermal path is described by
V
dµ = dP
N
but at the coexistence dµ = 0 so that
Z P
G V 1 PG
Z
0= dP = VdP Figure 34: Representation the Maxwell
PL N N PL construction, notice that the green areas
So that a constant value we can replace instead of the curve PX has to have to be the same.

satisfy
Z PX Z PX
dp dv = dp dv
PG PL

This construction is called Maxwell construction. A graphical inter-

pretation is provided in 34.
Let’s now compute the critical exponent of the Van der Wall’s
theory. The first is the beta exponent: at coexistence
8τ 3 3τ 3
π= − = −
3vl − 1 vl 3v g − 1 v g
Where the subscript g is refered to the gas phase and l to the liquid
one. Solving for τ this equation
(3vl − 1)(3v g − 1)(vl + v g )
τ=
8v2g v2l
when τ goes to 1 the system goes to a critical point and so vl , v g goes
to 1 by definition. So let’s define δ = v g − vl and so
s
9 2 T 9 2 TC − T
τ ∝ 1− δ ⇒ −1 ∝ δ ⇒ δ ∝
16 TC 16 Tc

1
from where one can read β = .
2
Let’s now compute the δ exponent. At critical pressure
∂2 P 
 
∂P 
= =0
∂V VC ∂V 2 VC
and so a simple Taylor expansion
∂3 P 

3
P = PC + (V − VC ) + h.o.
∂V 2 VC

and from this equation δ = 3 . Recall: experimentally β ' 0.32 and

δ ' 4.8. The differences between this
value and the value we find are due to
the mean field theory approximation.
68

Virial expansion

Another approach is the theory of weakly interacting fluids. The idea

is to consider a sort of perturbation theory: in the first approximation
one can think that the particles of a gas does not interact between
them , the next step of approximation is the weakly interacting gas in But they have to interact with the
which N particles, confined in a Ω region of volume V interact with a boundary reservoir, otherwise the
 
pressure is not well defined.
potential φ = φ ~ri −~r j  . So the Hamiltonian is Let’s consider a isotropic system, so the
potential depends only on the modulus
~pi
HΩ ({~r }) = ∑ + ∑ φ ~ri −~r j  of ~ri −~r j .
 

i
2m i,j>i

The partition function in the canonical ensamble is

1
ZΩ ( N, V, T ) = Q N (V, T )
N!Λ3N
Let’s modify Q N ( N, T ) starting from the ideal gas case For an ideal gas
QN = V N
Q N (V, T ) ' V N χ( N, V, T )
so the dependence of T is only in the Λ.
and so
FN = Fideal + k B T ln χ
Notice that if the density ρ is small then the gas is diluted and the
interaction between particles are negligible, if ρ is high the interaction
between particles became more important, in some sense one can
consider a sort of expansion in ρ
ideal gas
z }| { z not ideal terms
P
}| {
2 3
= ρ + B2 ( T )ρ + B3 ( T )ρ + h.o.
kB T

This is called viral expansion and the coefficients Bk are called

viral coefficients. Let’s now use the Wann der Valls to find explicit
formulas for Bk , using the fact that ρ = N
V is small We used the expansion of
 
 −1 2 N
1−b
   
P N N a N a V
= 1−b − ' ρ + ρ2 b − + ρ3 b2 + ρ4 b2 + h.o.
kB T V V kB T V kB T

So the virial coefficients are

a
B2VdW ( T ) = b − B3VdW = b2
kB T

The Boyle’s temperature is the temperature such that B2 ( TB ) = 0 in

Wann der Wall’s case
a
TB =  TC
kB b
How can we justify all this with a microscopic model? First let’s
write For simplicity
 

φ ~ri −~r j  = φij
 69

Z N
Q N ( T, V ) = ∏ d3 rk e−β ∑i>j φij
i =1

The Mayer function is defined by

f (r ) = e− βφ(r) − 1

The assumption is that βφij << 1 (notice that this quantity is undi-
mensional) and so f (r )  1. Let’s write a generic termThe terms that Again for simplicity
appears after the first one can be find as a special expansion called  

f ij = f ~ri −~r j 
cluster expansion. But let’s neglect this terms.
We will see something more at the end
! of the chapter.
e− β ∑i ∑ j>i φij = ∏ ∏(1 − fij ) =
i j >i

= 1 + ∑ ∑ f ij + ∑ f ij f kl + h.o.
i j >i i,k >i,l,j>i

Using only the linear term

Z
! Z
QN ' ∏ d3 r i 1 + ∑ ∑ f ij ' V N + V N −2 ∑ d3 ri d3 r j f ij
i i j >i i,j>i

But
Z Z 
Z 3 Z 3 Z
d3 ri d3 r j f ~ri −~r j  = d3 r f (|~r |)

d ri d r f (|~r |) = V
V V V V

And by let’s define this quantity −2VB2 , so that

1
Z
B2 = − d3 r f (|~r |)
2 V

This is exactly the B2 in the virial expansion. To see it let’s compute

' N2
z }| {
Q N (V, T ) = V N − V N −1 N ( N − 1) B2

And so the partition function is

VN N2
  
ZN (V, T ) = 1− B2 ( T ) + h.o.
N!Λ3N V

As a consequence

N2
 
FN = Fideal − K B T ln 1 − B2 ( T ) + h.o.
V

So the pressure is

∂FN
PN = − = ρk B T (1 + ρB2 ( T ) + h.o.)
∂V
70

and using the fact that (1 − x )−1 = 1 + x + h.o.

P 1
= (1 + ρB2 + · · · ) ' ' 1 + B2 ρ + B22 ρ2 + B22 ρ3
ρk B T 1 − B2 ρ

And so one can approximate the terms Bk with k > 2 using only B2 .
For example let’s compute B2 for some potentials. Let’s assume a
potential as in figure 35. This is a potential for an hard sphere gas. So Figure 35: Representation of a potential.
(
0 r<σ
e− βφ =
1 r>σ

This means that

1 1 2
Z Z  
B2 = − f (r ) d3 r = − e− βφ(r=−1 d3 r = πσ3
2 2 V 3
Notice that this result does not depend on the temperature. The Hard
sphere equation of state is
 
PV 2 N
= N 1 + πσ3
kB T 3 V

Another very common potential is the Lennard-Jones potential

  
σ 12  σ 6 Figure 36: Graphical representation of
φ(r ) = 4e −
r r Lennard-Jones potential.
Z ∞   12 6
 This potential has a quantum origin, for
− 4e ( σ ) −( σr )
B2 = −2π r2 e k B T r more detail
0 Masahiro Yamamoto. Origin of
the atttractive part of lennard-jones
This is not simple to compute: it has to be computed in series expan- potential. URL http://www.chem.
sion. Let’s consider the following change of variables konan-u.ac.jp/PCSI/web_material/LJ.
pdf
r kB T
x= τ=
σ e
A simple integration by parts

A
z }| { Z     
2 3 4 2 12 6 4 12 6
B2 = π σ x − exp − − dx
3 τ x1 2 x6 τ x1 2 x6

Expanding in series Where Γ is the Gamma Euler function

Z ∞
   2n+1 Γ(z) = x z−1 e− x dx
∞
2n − 1

1 1 4 0
B2 (τ ) = −2A0 ∑ Γ
n =0
4n! 4 τ

Now let’s say something to the higher order in the cluster expansion
which is the expansion of
!
∏ ∏(1 − fij )
i j >i
 71

The problem is that it groups terms quite different from one another,
for instance f 12 f 23 and f 12 f 34 .
Indeed we can represent this term in a diagrammatic way (figure
37). This approach recalls the Feynaman integral and in fact it is
Figure 37: Graphical representation of
similar. higher order terms (some examples).
The first term is provided in figure 38, we have already compute
it and it contribute with −2VB2 . Then there is a term f ij f kl with a
molteplicity
N ( N − 1) ( N − 1)( N − 3)
1
2 2
and the corresponding integral is
Z 4
f 12 f 34 ∏ dd r i
i =1 Figure 38: Graphical representation of
the first term.
this term involves four particles and in particular it is the square
of the first term and so the contribution is 4V 2 B22 . The third term is
equal to the second but this time j = k. The molteplicity is
N ( N − 1)( N − 2)
3
3!
The integral became
Z Z 
d d d d
f 12 f 23 d r1 d r2 d r3 = 4V f (r )d r = 4VB22

Another interesting diagram is provided in figure 39. This integral

involves three particles and by definition it is
Z 3
f 12 f 23 f 31 ∏ dd ri = 3!V ( B3 − 2B22 ) Figure 39: An example of possible
i =1 diagram: notice that this is a loop
diagram.
such that the configurational partition function is
N ( N − 1) N ( N − 1)( N − 2)( N − 3) 2 N ( N − 1)( N − 3)
 
N
Q N ( T, V ) ' V 1+ B2 + B2 + B3
V 2V 2 V2
In a different way one can think that
Z
Q N ( T, V ) = ∑ ∏ f kl d3N r
diagram kl

Let’s consider only connected diagrams for i sites. In other words

there are i particles from a system of N points and I consider all the
possible ways I can connect these i points. I define the contribute
of the sum of this connected diagram (when i is fixed) as (i!VBi )mi
where mi are the number of clusters of size i. One can show that the
degeneracy of this clusters is
N!
mi !(i!)mi
and so
N!
Q N (V, T ) = ∑ ∏ mi !(i!)mi (i!VBi )mi
{ mi } i
Landau Theory

A different approach which is equivalent to the mean field is the Lan-

dau theory of phase transition. We will see it for uniform systems.
Substantially Landau proposed a phenomenological approach
based on simple assumption: It can be interesting to study this
directly on Landau papers
• The existence of an order parameter η such that D. Ter Haar. Collected papers of L.D.
( Landau. 1965
0 T≥T For magnetic system this is m, for gas
η=
6= 0 T < T ∆ρ.

• The existence of an analytic function which is called Landau free energy

L(η ).
• L(η ) has to satisfy the symmetry of the system.
• The equilibrium states of the system correspond to the absolute
minima of L(η ).

This is assumption are enough to describe phase transitions. Since

L(η ) is analytic one can expand it

L(η ) = a0 + a1 η + a2 η 2 + a3 η 3 + a4 η 4 + h.o.

This function has to satisfy the symmetries of the system. Let’s

neglect the first term a0 : this is only a constant to one can redefine
L → L(η ) − a0 . If the temperature is T > T then ηeq = 0 which means

∂L
(0) = 0
∂η
But since
∂L ∂L
= a1 + 2a2 η + · · · ⇒ 0 = (0) = a1
∂η ∂η
and so a1 = 0. So the Landau free energy is

L(η ) = a2 η 2 + a3 η 3 + a4 η 4 + · · ·

In order to see the powerful of this theory let’s apply it to the Ising
model. The Z2 symmetry means that the Landau free energy has to
be an even function Since
L(η ) = L(−η )
74

L = a2 η 2 + a4 η 4 + h.o.

When we stop the expansion? The answer is simple: one has to stop
the expansion when the Landau free energy describe our system. We
will see with this concrete example (Ising model) that this means that
we can stop the expansion when ak > 0.
This because if the other coefficient can be change the sign one can
put all zero and then the only term that remain in the expansion is
ak η k . If ak < 0 the minima of L is ∞ which has no physical meaning.
Let’s rewrite a2 ( T ) and a4 ( T ) as an expansion in t = T − T
T , but let’s
assume that a4 > 0 so We named some coefficient a and b only
( to simplicity.
a2 ( T ) = a2? + t 2a
a4 ( T ) = 4b > 0

a2? is only a translation so let’s neglect it. For Ising model the Landau But this translation change a bit the
free energy is solutions, fortunately not qualitatively.

a b
L = tη 2 + η 4
2 4
This function contain all the physical information and a and b contain
the microscopic detail. The equilibrium phases are

∂L   
= 0 ⇒ η at + bη 2 = 0
∂η η =η


The solution are r

− at
η=0 η± = ±
b
Notice that η± exist (at least real) only for T < T so the only
solution for T > T is η = 0. This is coherent with want we already
know.
If T < T there are three solutions, a simple plot (provided in figure
Figure 40: Behavior of L respect to η in
40) shows that the two minima are η± and in particular L(η+ ) = the case of T > T, in which only one
(η− ) that means that we are in presence of a coexistence of two solution (η = 0) is present, and T < T
in which three solution can be seen.
phases.
Let’s now introduce a magnetic field. To introduce it one can
simply Legendre transform the Landau free energy

a 2 b 4
L= tη + η − hη
2 4

Now one can compute the critical exponents. The β exponent can be Defined by
simply find by the solutions η ∝ tβ t → 0−
r
− at 1 1
η=± ∝ t2 ⇒ β =
b 2

The exponent alpha requires more work. There are two cases to Recall that
C ∝ t−α h = 0, |t| → 0
 75

study: the case in which t > 0 i which η = 0, L = 0 and so α = 0 and

the case in which t < 0 in which
We are proving that for the Ising
− a2 t2 α+ = α− = α
( η± ) =
4b
By definition
∂2 L a2
C = −T = T
∂T 2 2bT
2

If T → T (t → 0)
a2
C=
2b
which is a constant (in t) and so α = 0 .
The δ exponent can be read out by the equation of state which is Which is defined by
h ∝ ηδ T→T
∂L
= 0 ⇒ h = atη + bη 3
∂L
If t = 0 the equation of state became h = bη 3 which means δ = 3 .
The last is the γ exponent Since

∂η
χR =
∂h
and deriving the equation of state in h
1
1 + atχ T + b3η 2 χ T ⇒ χ T =
at + 3bη 2
Figure 41: Representation of the
− at
if h = 0 and η 2 = one can simply find behavior of L. Notice that the presence
b of a field does not consent to have a
coexistence of phases.
χ T ∝ t −1 ⇒ γ = 1 Recall

Let’s now figure out what if we use a order parameter which is not a χ T ∝ | t |−γ h = 0, t → 0

scalar, for instance let’s consider the O(n) model in which the order
parameter are vectors. In this case the Landau free energy has to be a
scalar and so one has to be consider the terms ~η · ~η and so or the Gibbs version
a b
at b LG (~η ) = t~η ·~η + t (~η · ~η )2 + h.o. −~h ·~η
L (~η ) = ~η · ~η + t (~η · ~η )2 + h.o. 2 4
2 4
One can also consider some system in which the order parameter is a
tensor . For instance the liquid crystals, for
Can the Landau theory predict also first order phase transitions? more information we suggest
Denis Andrienko. Introduction to
Yes it can, let’s consider for example a system in which liquid crystals. 2018
b
LG = atη 2 − ωη 3 + η 4 − hη
2
The equation of state is
 
h = η 2at − 3ωη + 2bη 2
76

the solution η = 0 is the disordered phase, the solutions Where

3ω
c=
r 4b
1  p  at
η± = 2ω ± 9ω 2 − 16atb2 = c ± c2 −
4b b
Notice that this solutions are well defined (real) only if

c2 b
t≤ = t??
a
So if t > t?? the only solution is η = 0. if t < t?? there are several
possibilities: depend on the value of L(η± ) respect to L(0), in fact if
L(η+ ) is grater the L(0) = 0 then the absolute minima is η = 0 and
η+ is only a local minima and so is a stable state only if the energy η+ has to be a maxima.
is less then a given amout of energy (L(η− ) − L(η+ )). This is called
metastable state. If L(0) = 0 = (η+ ) there is the coexistence of two
states and if L(0) = 0 > (η+ ) then the metastable state is η = 0 and
the absolute minima is η+ .
Also for this systems one can easily compute the critical exponent
directly by the Landau theory.
Also the tricritical points are explained by Landau theory: let’s
consider a Landau free energy

a 2 b 4 c 6
L= η + η + η + h.o.
2 4 6
where c > 0 but a, b can change their sign. The tricritical point is
simply defined by the condition a = b = 0.
Let’s study in more detail the BEG model. The Landau free energy
is
a( T, ∆) 2 b( T, ∆) 4 c( T, ∆) 6
L(η, T, ∆) = η + η + η + h.o.
2 4 6
where c( T, ∆) > 0. As before in order to introduce an external field
one can simply Legendre transform the Landau free energy

LG = L − hη

The equilibrium states can be computed by

∂L ∂2 L
=0 >0
∂η ∂η 2

Two phases can coexist if they obey to the last equations and The two states has to be both equilib-
rium states.
L(η1 ) = L(η2 )

One can study mathematically this system and the results are

• When ∆ < ∆C if T decreases also a T decreases and so the critical

points T = TC (∆) are such that a( T, ∆) = 0. This is the usually
symmetry breaking.
 77

• When ∆ > ∆C if T decreases b became zero "before" then a. So

the interesting region is a > 0 and b < 0. The first order phase
transition curve is
3b2
a=
16c
The equations a( T, ∆) = b( T, ∆) = 0 define the tricritical point. Observe that two equations define
the tricritical point and both a and b
Notice that Landau theory and mean field theory produce the same depends on two variables T and ∆, so
one can find ∆t and Tt .
results. The reason at the moment can be not clear, but we will see in
the next chapter.

Figure 42: Representation of the phase

diagram in function of a and b.
Ginzburg-Landau theory

The assumption of a mean field theory is that

hSi S j i ' hSi ihS j i

or equivalently
ρ ( S1 , · · · , S N ) = ∏ ρ (1 ) ( Si )
Si

The natural questions is: when a mean field is a reasonable approxi-

mation?
There are many way to do that: for a microscopic point of view
one can consider the number of bond and when this number in-
creases one can think to a mean field as a good approximations.
Let’s consider an heuristic argumentdue to Ginzburg. One can try to make it more formally,
Let’s consider a system in which the rage of interactions is ξ: the but it works.

energy due to the fluctuation is ' k B T so that in d−dimensional

space the fluctuations provide a free energy

kB T
∆Ff l '
ξd

Since a mean field theory is valid a good approximation if ∆FH  F

and in proximity of a critical point ξ ∝ |t|−ν F ' |t|2−α In order to show the second relation
let’s consider
k C TC |t|νd  TC |t|2−α ⇒ |t|νd−α−2  1 ∂2 F T ∂2 F
CH ' − T =− ” 2
∂T 2 TC ∂t
is the condition we wont. This condition is true if and only if and since CH ∝ |t|−α when t → 0 one
can integrate the latter equation and
2−α find the behavior of F when t → 0.
νd + α − 2 > 0 ⇒ d > = dup
ν

where dup is the upper critical dimension and represent the lower
dimension at which the mean field is a good approximation. This
equations is called Ginzburg criterium (or Ginzburg argument).
For instance for Ising model ν = 1/2 and α = 0 and so dip = 4,
however for a tricritical model dup = 3 From χ T = |t|−γ one can find So the mean field is a good approxima-
that tion for study the tricritical phenomena
2β + γ but it is not a good approximation for
d> = dup the critical ones.
ν
80

But since the upper critical dimension has to be the same

2 − α = 2β + γ

which is the Rushbrooke’s inequality that has become an equality.

What if the mean field is not a good approximation? Let’s try to
study beyond the mean field theory. In order to do that let’s use a
coarse graining procedure: we are not interesting in the microscopic
details of the systems (because the critical phenomena does not
depend on them) so we write the partition function "integrating out"
all the microscopic details
  Z ?
Z = Tr e− βH({S}) = dm(r ) ∑ e− βH(S)

where the ∑? is compatible with a given m(r ) is well defined if the

microscopic length a is much less then the typical length of the lattice
l (let’s think to a two-dimensional lattice) which is much less then the Figure 43: Representation of the
situation.
typical length of interaction ξ which has to be less then the typical
size of the system L. In this case
?
∑ e−βH(S) = e−βHe f f [m(r)]
where He f f [m(r )] is an effective Hamiltonian which is mathemati-
cally a functional. This functional use as an domain some functions
defined in a given subset of the lattice (referring to figure ??) labeled
with ~r: S ∈ ~r means that the spin I’m consider-
1
Nl S∑
ml (~r ) = Si ing is in the subset of the lattice labeled
with ~r.
∈~r

Where Nl = (l/a)d . In Fourier-space one can define the cutoff

wavelength which is always less then the absolute value of ~q

|~q| > Λ = l −1

Since |~q| is never zero there is no ultraviolet divergence.

The partition function can be re-written as a functional integral in
ml (~r ) Z
Zl = D [ml (~r )] e− βH[m(~r)]

To go on we have to provide an explicit formula for H [m (~r )]. There

are two important terms: the first is the energy for a uniform system,
and so one can simply use the Landau free energy, so that

b
βH[m] = atm2 + m4 + h.o.
2
Notice that since l/L  1 one can perform the continuous limit
substituting To do it formally you can think to the
Riemann theory of integration.
 81

1
Z
∑→ ld
dd r
~r

so that   where
b a b
Z
d
βH[m] ' d r atm (~r ) + m4 (~r ) 2 a=
ld
b=
ld
2
The second term is the variation of energy that has to be consid- Figure 44: When we pass from a cell to
ered when one pass through a surface that divided two boxes in the a neighbor cell we have to pay a cost in
term of energy: this is the second term.
lattice (as in figure 44). Let’s assume that this term is
One can put also higher order terms.

k
− βH[m] = − ∑ (ml (~r + ~µ) − ml (~r ))2
µ 2

And in the continuous limit Where k = k/l d−2

2
ml (~r + ~µ) − ml (~r )

k 1
Z
∑
2 ~r,~µ
(ml (~r + ~µ) − ml (~r ))2 = d−2
l
∑ k
l
dd r =
~µ
1
Z
=
l d −2
∑ k (grad ml )2 dd r =
~µ
k
Z
= (grad ml )2 dd r
2

This two terms has to be summed and so the Ginzburg-Landau theory

has a partition function as
  
b k
Z Z
ZGL = D [m (~r )] exp − d r atm + m4 + (grad m)2
d 2
2 2

In general this partition function can be very complex, but if m(~r ) =

m0 constant for every position ~r:

ZGL = e−V f L (m0 )

where f L (m0 ) is the Landau free energy. If one introduce a magneti-

zation
  
b k
Z Z
ZGL = D [m (~r )] exp − d r atm + m4 + (grad m)2 − hm
d 2
2 2

Our goal now is to solve it, at least in some approximations.

Let’s introduce the functional derivative One can show that
δ f (~r )
Z [h(~r ) + eδ (~r −~r 0 )] − Z [h(~r )] = δ ~r −~r 0


δZ [h(~r )] δ f (δr 0 )
= lim
δh(~r 0 ) e →0 e

This formalism lets write the generalization of the magnetization

δF δ ln Z
hm (~r )i = − = −k B T
δh(~r ) δh(~r )
82

And similarly
δ2 F 2
−1 δ ln Z [ h ]
χ T (~r,~r 0 ) = = β−1 Gl ~r,~r 0



= β
δh(~r )δh(~r 0 ) δh(~r )δh(~r 0 )
This is also the generalization of the fluctuation dissipation theorem.
The next step is to approximate the partition function: the zero
order is the saddle point approximation. This approximation does
not consider any fluctuation and so it is equivalent to the mean field.
Let’s define
Z Z
L[m, h] = dd r ( βH[m] − hm) = dd r L(m, grad m)

we want to minimize it , so that let’ define a m0 (~r ) such that Minimize it to maximize the exponen-
tial integrate in the partition function.
min L[m(~r )] = L[m0 (~r )]
m(~r )

A necessary condition to have a minimum is δL = 0, so let’s mini-

mize it . So We proceed as in mathematical physics
Z or analytical mechanical lectures: this
δL = δ dd r L = is very similar to the minimal action
Ω principle.
Z Z
= dd r L(m + δm + grad m + δ(grad m)) − dd r L(m, grad m) = The boundary term is assumed to be
Ω vanishing. This because we assume
L L that the variation δm vanishing at the
Z  
∂ ∂
= dd r δm + δ(grad m) = boundary!
Ω ∂m ∂(grad m)
∂L ∂L
Z  
= dd r δm + grad(δm) =
Ω ∂m ∂(grad m)
 
∂L ∂L ∂L 
Z     Z 
d
= d r δm − grad δm +  m δm · ~n dS =

Ω ∂m ∂(grad m) grad
∂
∂Ω
∂L ∂L
Z  
= dd r − grad δm =
Ω ∂m ∂ ( grad m)

So the condition δL = 0 is equivalent to the Euler-Lagrange equations

∂L ∂L
− grad =0
∂m ∂(grad m)

Let’s replace L = βH. The Euler-Lagrange equation is First of all one needs the variation of
the gradient, but is simple because
h (~r ) = −k 4 mo (~r ) + 2atm0 (~r ) + 2bm30 (~r ) grad(m + δm) = grad m + 4m δm + h.o.

If one assume that the system is uniform, which means h(~r ) = h and And this means
m(~r ) = m0 δ grad m = 4m δm
h = 2atm0 + 2bm30
This is the Landau theory! This prove that Landau theory is the
saddle point approximation (zero order) of the Ginzburg-Landau
theory, and so it is a mean field theory. For a non uniform system
 
b 4 k
Z
d 2 2
F [m, h] = d r atm0 (~r ) + m0 (~r ) + (grad m0 )
2 2
 83

Let’s compute the correlation function. To do that one can simply

derive the equation of state so that

δ δ  
0
h(~r ) = 0
−k 4 m0 (~r ) + 2atm0 (~r ) + 2bm30 (~r )
δh(~r ) δh(~r )

and is simple to obtain

 
−k 4 +2at + 6bm20 (~r ) χ(~r −~r 0 ) = δ ~r −~r 0

And so Recalling that

  Gc (~r −~r 0 ) = k B Tχ T (~r −~r 0 )
β−1 −k 4 +2at + 6bm20 (~r ) χ(~r −~r 0 ) = δ ~r −~r 0

This mathematically is the fundamental equation for the differential

operator that acts on Gc . In order to solve it we have to distinguish
two cases: for T > TC m0 = 0 and so Where r
k
ξ > (t) =
  k T 2at
−2
− 4 +ξ > (t) Gc (~r −~r ) = B δ(~r −~r 0 )
0
k
q
But if T < TC m = ± − atb and the equation became Where r
−k
ξ < (t) =
  4at
−2
− 4 +ξ < Gc (~r −~r 0 ) = δ(~r −~r 0 )

So this two equations are mathematically the same equation and

one can solve it using the Fourier theory. In fact one can Fourier
transform both the member in the equation In this equation ξ can be ξ < or ξ > .

e( q ) = k B T
G
1
k q 2 − ξ −2

Let’s use ~x = ~r −~r 0 , so that

dD q ei~q·~x
Z
G (~x ) =
(2π ) D q2 + ξ −2

In three dimension we get

d3 q ei~q·~x
Z
G (~x ) = =
(2π ) q + ξ −2
3 2

4π ∞Z
q2
Z
= dq eiq|~x| cos(θ ) d(cos θ ) =
(2π )3 0 q2 + ξ −2
Z ∞
2 q sin (q | x |)
= dq =
(2π ) |~x | −∞ q2 + ξ −2
3
"I #
2 zeiz|~x|
= Im dz =
(2π )2 |~x | z 2 + ξ −2
"I #
2 zeiz|~x|
= Im dz
(2π )2 |~x | (z + iξ −1 ) (z − iξ −1 )
84

Using the residue theorem one can get that the solution is

1 −|~x|/ξ
G (|~x |) = e
2π

and so ξ is the range of interaction. Since ξ ∝ t−1/2 the critical

1
exponent ν = and η = 0 if the system is three-dimensional. Recalling that
2
Now we have all the critical exponent in the Landau theory, and Gc (|~x |) ' |~x |2−d−η

we know that this is only a mean field theory. What if one goes when T → TC .
beyond? The zero order is the saddle point approximation that we
have already seen, the first order is vanishing because we want to
minimize and so the first derivative has to be zero, the second order
is not zero. Let’s consider

m(~r ) = m0 (~r ) + δm(~r )

and one can simply compute that For simplicity we impose that h = 0.
 
2 b 4
βH[m] = V atm0 + m0
2
 
k b
Z  
+ dd r (grad δm)2 + at + 3bm20 (δm)2 + 2bm0 (δm)3 + (δm)4
2 2

For T > TC , since m0 = 0

 
k b
Z
βH > (δm) = dd r (grad δm)2 + at(δm)2 + (δm)4
2 2

It si very difficult to integrate this but two terms over three are Recall that
quadratic and those therm are computable using gaussian integrals.
Z
Z(T ) = D[δm]e− βH[m]
So in the first approximation we can omit the term (δm)4 and con-
sider what is called gaussian model or gaussian approximation
that means Where the term es.p. is the term coming
from the saddle point approximation.
 Z  
k
Z
Z gaus ( T ) = es.p. D[δm] exp − dd r (grad δm)2 + at(δm)2
2

This is a simple approximation that can be solved in the Fourier

space. Let’s assume

1
Z
~ ~
 
δm = ϕ =
V ∑ eik·~r ϕ~k ϕ~k = e−ik·~r ϕ ~k
~k

Let’s recall that This because when we pass from a

1 i~k (~r −~r 0 )
V ∑e = δ(~r −~r ) 0 discrete system to a continuous one
~k V
Z
∑→ (2π ) D
d3~k
~k
and Z
~ ~0
dD~x ei(k−k )·~r = δ~k,~k0 and so
1 1 V
Z
~ 0 0
∑ eik·(~r−~r ) → dD keik(~r−~r ) = δ( D) ~r −~r 0


V ~k
V (2π ) D

And in the limit r → ∞ this became

 
(2π ) D δ ~k −~k0
 85

and so one can conclude that

 
Vδ~k,~k0 → (2π ) D δ( D) ~k −~k0 r→∞

Since the field ϕ has to be real we have also to impose that ϕ~k =
ϕ? ~ . Moreover we have already notice that there are no ultraviolet
−k   π
divergences; this in Fourier space means that ~k  ≤ ≤ Λ. We know
 
a
that Z Z
∏ d Re( ϕ~k ) d Im( ϕ~k )



D [ ϕ (~r )] →
|~k|<Λ
But since ϕ is real the term in which appear ~k and the term asso-
ciated to −~k are not independent. To solve this over-counting let’s
divided by a factor of 2 or, equivalently, let’s define
0
∏= ∏
0≤|~k |≤Λ

So the partition function is

Z 0 gauss ( ϕ )
Zgauss = e− βF = ∏ dϕ~k eiβH ~k

If T > TC then m0 = 0 and so Similar steps can be followed for

! T > TC but in this case
k
Z r
βHgauss = dD~r (grad ϕ)2 + atϕ2 m0 = −
at
2 b

and so, by a simple Fourier transform

1   1   
∑
2
βHgauss = 2at + kk2 ϕ~k ϕ~?k = l 2at + kk2  ϕ~k 
2V ~k
2V

Notice that we obtained a quadratic term in ϕk that means that the

modes are independent: this is due to the gaussian approximation.
The partition function became We have defined
1  
0 Z A= 2at + kk2
1  
2V
Zgauss = ∏ dϕ~k exp −
2V
2at + kk2 | ϕ~k |2 =
0 Z 2 2
=∏ dx dy e− A( x +y )
 
0  
π 1 2πV
=∏ = exp  ∑ ln 
A 2 ~ 2at + kk 2
|k |<Λ

And so the free energy is

 
KB T 2πV
Fgauss = −
2 ∑ ln
2at + kk2
|~k|<Λ
86

For T < TC one can get

 
1 2πV
Fgauss = βA0 V − k B T
2 ∑ ln
2at + kk2
|~k|<Λ

Let’s now compute the α exponent to compare the result with the
mean field theory. The specific heat is

∂2 F T2k B a 1 TkB a 1
C = −T = ∑ 2 − ∑

∂T 2 V VTC2
2
VTCT
 
|~k|<Λ 2at + kk2 |~k|<Λ 2at + kk2

For t → 0+ the sums became integrals and so the first term can be Figure 45: Behavior of the specific heat
written as in the mean field theory and using the
gaussian approximation.
Z Λ 2 Using
dD k ξ k− D ξΛ dD y

1 1
Z
r
I1 = D
= 2 D 2 k
0 ( 2π ) 2at + kk 2 k 0 ( 2π ) (1 + y2 ) ξ (t) =
2at
and then using a change of variables
But ξΛ goes to infinity when t goes to 0+ , so what happen to the ~y = ξ~k.
integral? The integrand behaves as Observe that
dD y ' y D−1 dy
y D −1
2
(1 + y2 )
What means that if D > 1 and y → 0 then the integral is zero, but if
y → ∞ then our integral is asymptotically equal to q D−4 . And so the
integral converges only if 5 − D > 1 which means D < 4 . For D > 4 In fact the integral is proportional to
the behavior of the first term is ξ 4− D . Recall that
Z ∞
( dx
ξ 4− D ' t − ν (4− D ) D < 4 Λ xγ
finite D>4 converges if and only if γ > 1.

The second term diverges if D < 2 ans so it is less relevant of the first
term. In conclusion
(
t − ν (4− D ) D < 4
C∝
finite D>4

And this means that for D < 4 α = ν(4 − D ). This result is different
from the result obtained from the mean field. This is the effect of
fluctuations. However, as we had already notice, the modes are
independent and this is due to the gaussian approximation. Higher
order approximations produce the dependence between the modes.
Let’s study the ν exponent. Since
1  ~ ~0 0 1 ~
 
ϕ (~r ) ϕ ~r 0 = 2 ∑ ϕ̃~k ϕ̃~k0 ei(k·~r+k ·~r ) = 2 ∑ eik·~x G ~k




V ~ ~0 V ~0
k,k k

where   1
G ~k =
c ( k 2 + ξ −2 )
 87

This result is the same result obtained for the mean field theory and
so the ν exponent is the same. This is due to the fact that the modes
are not independent and so one have to use at least the quadratic
term. Since analytically it is very difficult to do one can use the
theory of perturbation where the perturbation is given by the term

b 4
ϕ
2
in the Hamiltonian. So for small enough b one can conclude that To have a different point of view on the
relationship between the mean field
eH+bU ' eH (1 + bU ) theory and the Ginzburg-Landau theory
we suggest
Als-Nielsen and R. J. Birgeneau.
where eH is the gaussian approximation theory and the perturbation Mean field theory, the ginzburg crite-
is contained in the term bU. Unfortunately b is not always a small rion, and marginal dimensionality of
parameter. phasetransitions. 1977
Scaling theory and renormalization group

In the previous chapters we introduced the universality of the critical

exponents as a phenomenological fact we have to take in account.
Now let’s study the universality from a formal point of view. The
first attempt of explain universality was provided by Widom and he
connected this physical phenomena to the theory of homogeneous
functions. Let’s do a little digression on the homogeneous functions.
By definition f (r ) is an homogeneous function if ∀λ ∈ R

f (λr ) = g(λ) f (r )

If you know the value of the function in a point and the function g(λ)
you know the function in every point, in fact if f (r0 ) is known then
 
r
f (r ) = g f (r0 )
r0
Some mathematical results on homogeneous function are the fact that
the only homogeneous function are monomers and if g(λ) = λ P then
f (λr ) = λ P f (r ).
When the we try to generalize the homogeneous function we have
to take in account that every variable has a proper index, so that in
two dimension there exists two real number a, b such that The case f λ a x, λb y = λ P f ( x, y) is

contained in the general case, in fact

let’s define λ p = s such that
 
f λ a x, λb y = λ f ( x, y)  
a b
f s p x, s p y = s f ( x, y)
This is the most general form of an homogeneous function.
1
Let’s choose λ = y− b in an generalized homogeneous function
! !
x − 1b x f (x)
f a ,1 = y f ( x, y) ⇒ f a ,1 = 1
y b y b yb
Let’s come back to the physics. The Windom’s hypothesis is that the
bulk free energy is the sum of a analytic term and a singular term

f ( T, H ) = f a ( T, H ) + f s ( T, H )

and the singular component is a homogeneous function

f s (λ p1 t, λ p2 h) = λ f s (t, h) ∀λ ∈ R
90

− 1
As we done before we can choose λ = h p2 such that
 
1 t
f s (t, h) = h p2 f s , 1
h∆
p1
where it is defined the gap exponent ∆ = .
p2
What is important is that given p1 and p2 we can compute all
the critical exponent . The beta exponent can be calculated simply This is the first time that we connect
deriving the definition of f s , in fact all the critical exponent in only two
exponent.
∂ f s p1 ∂ fs
λ p2 (λ t, λ p2 h) = λ (t, h)
∂h ∂h
But notice that the derivatives of the bulk free energy in the magneti-
zation (or in general the order parameter) m and so This means that also the magnetization
is an homogeneous function.
λ p2 ms (λ p1 t, λ p2 h) = λms (t, h)
1− p2
p1
Using h = 0 and choosing λ = (−t) ms (−1, 0). In other words

1 − p2
β=
p1

The delta exponent can be calculated in a similar way simply using

the fact that
1− p2
ms (0, h) = λ p2 −1 ms (0, λ p2 h) ⇒ ms (0, h) = h p2
ms (0, 1)

and so
p2
δ=
1 − p2

So given β and δ one have also p1 and p2 The gap exponent is done In fact
by p1 =
1
p2 =
δ
p2 β ( δ + 1) δ+1
∆= = βδ
p1

From the specific heat one can get α = 2 − p1−1 and so α + β(δ + 1) = 2
which is the Griffith equality. On the other hand

2p2 − 1
γ= ⇒ α+β+γ = 2
p1

which is the Rushbrooke’s inequality that has become an equality.

Using this relations one can conclude that
 
2− α h
f s (t, h) ' t f s 1, ∆
t

Which we already know. The next step is to explain the Widsom’s

hypothesis from a microscopic point of view. In some sense we have
 91

to "integrate out" all the microscopic detail to obtain p1 and p2 . This

work was done by Kadanoff : the main idea is to start from a system To see the original article
perform a coarse grained but with the assumption that the Hamilto- Leo P. Kadanoff. Operator algebra
and the determination of critical indices.
nian of the original system is equal in form of the Hamiltonian of the 1968
coarse grained system.
So let’s start from a system which is microscopically described by
the Hamiltonian
βH = k ∑ σi σj + h ∑ σi
h,ji i

Where σi = ±1 are the spins.

Let’s perform a coarse grained approximation (as in figure 46). For
each cell of the new system we can associate a number which is again Figure 46: Graphical representation of a
course grained system.
±1 and it can be called superspin Si :

1 1 1
Si =
|ml | l D ∑ σi ml =
ml ∑ σi
i ∈cell i ∈cell

Now Kadanoff proposed two assumption

first assumption : The Hamiltonian of the coarse grained system is

equal in form to the Hamiltonian of the microscopic system. This
N
means that if Nl = D and ξ l = xl then
l
hl
z }| {
h ∑ σi = h ∑ ∑ σi = h ∑ | m l | l D Si = h | m l | l D ∑ S I
i cells i ∈cell cells I

and this means that hl = h|m L |l D . Since Hl is equal in form to H, In a very similar way one can compute
the statistic is the same, that means that the partition function is also k l as a function of k.

the same
Zl = Z ⇒ Nl f s (tl , hl ) = N f (t, h)
that means that the bulk free energy is an extensive variable: this
is a prove of coherency of our theory

f s (tl , hl ) = l D f s (t, h)

This equation is similar to the Widom’s hypothesis, but it miss

something.
second assumption : In order to obtain the Widom’s theory let’s
assume that (
tl = tl yt
hl = hl yh

The exponents yt and yh are called scaling exponents . With this

assumption the last equation became the Widom’s hypothesis

f s (t, h) = l − D f s (tl yt , hl yh )
92

So now we have a microscopic interpretation of the scaling theory ad

− y1
so we can again choose l = |t| t t obtain
y
 
D − yh D
f s (t, h) = |t| yt f s 1, h|t| t ⇒ 2−α =
yt

But now we cal also compute the correlation function Let’s use the capital index to the coarse
grained system and the lowercase
G I J = hS I S J i − hS I ihS J i = letters for the microscopical system.

1
= 2(y − D) 2D ∑ ∑ hσi σj i − hσi ihσj i =
 
e h e i∈ I j∈ J

= e−2(yh − D) Gij

This means that

1
ν= ⇒ 2( D − y t ) = D − 2 + ν
yt

An so we have a new equation This equation connect some thermo-

dynamic quantities (the first member)
2 − α = νD with some system quantities (the
second term)

This equation is called hyperscaling relation . The next step is to

construct a formal way to describe all this steps. We have to explain

• How to compute yt and yh ;

• Where singularities come from?

• How can we explain universality?

Renormalization group

This theory is proposed by Wilson . Let’s use [k ] to indicate the vector Wilson won the Nobel prize for this
~k; let’s assume that a  la  ξ < L. What we have done in Kadanoff theory.

theory is to re-define the coupling constant h → hl and t → tl using a

coarse grained procedure, let’s call Rl this transformation:
 0
k = Rl [k ] l>1

By assumption Rl is analytic (it comes from a finite integration). But

this does not explain from where the singularities come. Rl form a
semi-group where the composition of Rl1 and Rl2 si This cannot be a group since l > 1
and so there cannot exist the inverse
transformation.
R l2 ◦ R l1 = R l1 · l2

The Kadanoff main assumption was the fact that the Hamiltonian of
the coarse grained system is the same, in form, to the Hamiltonian
of the old system. This is a strong hypothesis and we don’t need
 93

it, the only assumption we need for the Hamiltonian is that the
Hamiltonian after a transformation preserve the same symmetries.
However this is not enough but we can assume that the statistic is
invariant, so that
ZN 0 k0 = ZN [k]
 

It is very easy to show that this two assumption generalize the

Kadanoff theory. There is an important question to answer: Where In some sense the Kadanoff theory is a
singularities come from? Every scaling step is analytic but we need special case of this new theory.

infinity steps and so, in general, the results it is no more analytic; this
means that at the end we can get singularities.
Notice that the use of Rl means that we are moving in the space of
the models (fixing the symmetries) ; in this sapce we are interesting We will see it again with the prolifera-
in the fixed point which means the point ~k? such that tion problem.

[k? ] = Rl [k? ]

For this point

ξ [k? ]
ξ [k? ] =
l
There are only two possible cases: ξ [k? ] = 0 and this is called trivial
fixed point or ξ [k? ] → ∞ which is called critical fixed point.
Every fixed point has its own basis of attraction which are the set
B [k? ] = {[k]} such that

(n) n→∞
Rl [k ] → [k? ]

This set is very important since we know that all the point in the
basis of attraction of a critical fixed point have ξ [k ] = ∞. So in some We can show this simply recalling that
sense B [k? ] is the space of critical models. This is the deep meaning h i
ξ [ k ] = · · · = l n ξ k (n) → ∞
of universality since all the model which has the same symmetries h i
and are critical models are in this set and so they are associated on since ξ k(n) → ξ [k? ] = ∞.

the same critical fixed point. The set of critical models is also called
critical manifold.
Let’s perturb the system near the critical point Notice that we can expand since we
known that Rl is analytic.
∂~k j
!
 
k0j = (Rl ) ji ~k? + δ~k = k?j + ∑ δki + h.o.
i
i ∂~k i ~k?

And so
∂~k j
!
~0
δk = Πδ~k Πij =
∂~k i ~k?

Let’s suppose that Π is symmetric From the equation In general Π is not symmetric and so
we have to distinguish in left and right
(σ) eigenvalue since Π is not diagonaliz-
Πl~e(σ) = λl ~e(σ) Πl Πl 0 = Πl ·l 0 able. This problem can be solved use
the complex number but for simplicity
let’s assume that Π is symmetric.
94

one can prove that

(σ) (σ) (σ) (σ)

λl λl 0 = λl ·l 0 ⇒ λl = l yσ

Let’s now figure out how δ~k behaves under Rl : since

δ~k = ∑ a(σ)~eσ
σ

then

δ~k0 = Πδ~k = ∑ a(σ) Π~e(σ) =

σ

=∑ (σ) (σ) (σ)

a λl ~e = ∑ a(σ)~e(σ)
σ σ

So there are two possible cases: λ(σ) > 1 and so a > a that means
that the solution escape from the fixed point or λ < 1 that means
0
a <  a which means that the solution goes to the fixed point. If
 (σ) 
λ  > 1 then yσ > 0 and this is called relevant eigenvalue . If
 
 (σ) 
λ  < 1 then yσ < 0 and this is called irrelevant eigenvalue . The
 
last case if the case in which λ(σ)  = 1 then yσ = 0 and this is called
 

marginal eigenvalue .
In order to study the scaling exponent let’s assume that there are
only k1 ( T ) = T and k2 ( T ) = H and let’s assume for a magnetic
system that for a certain fixed point ξ ( T ? , H ? ) = ∞ and H ? = 0, let’s
linearize the matrix π such that Where
 0  t T − T?
t 
(t) (h) t= h=H
= diag λ , λ T
h0 l l h

But ξ 0 = ξ/l and so when h = 0

 
ξ (t, 0) = l h ξ l hyt t, 0

where l is arbitrary. So we conclude that if b is a constant scaled

temperature
 − 1
b yt 1
ξ (t) = ξ (b) ⇒ ν =
t yt

and so Notice that now we have an explicit

1 (t) formula for the critical exponent.
yt =
ln λl
l
And form the free energy one can prove that

1 (h)
yh = ln λl
l
Now let’s apply it in the one-dimensional Ising model using l = 2
that means that we want to integrated out half of the spin.A visual Figure 47: Graphical representation of
the renormalization with l = 2.
This because if we integrated out half of
the spin N → N/2 and so l = 2.
 95

representation is provided in 47. Let’s called Si the spins we want

to integrated out and σi the spins we want to keep. Let’s call W the
weight such that We introduced a constant term ĝ
because we will see that it appear, this
ĥ is a first example of proliferation that
W (σi σj ) = − ĝ − (σi + σj ) − Ĵσi σj we will explain better later. z in the
z
usual number of bond.
This means that The step ! is an assumption!
 

Z= ∑ exp  ∑ W (σi , σj ) =
σ hi,ji
!
N/2
!
= ∑ ∑ exp ∑ (W (σi , Si ) + W (Si , σi+1 )) =
σ Si i =1
exp W 0 (σi ,σi+1 )
N/2 z
=∑ ∏
}| {
(W (σi , Si ) + W (Si , σi+1 )) =
σi i =1
!
= ∑ exp ∑ W 0 (σi , σi+1 )
σi i

Where
h0
W 0 (σi , σi+1 ) = g0 + (σ + σi+1 ) + k0 σi σi+1
2 i
So the assumption is that Let’s define
σi +σi +1 σi +Si Si +σi +1 x = ek y = eh z = e g
z̃ỹ 2 x̃ σi σi+1 = ∑ z2 y 2 x σi Si y 2 x Si σi+1 0 0 0
Si =±1 x̃ = ek ỹ = eh z̃ = e g

There are 4 possibility depending on the sign of σi and σi+1 . In

particular when both of them are +1 then
 
z̃ỹ x̃ = z2 y x2 y + x −2 y−1

When both of them are −1 we have

 
z̃ỹ x̃ = z2 y−1 x −2 y + xy−1

and when one of them is +1 and the other is −1 (in both cases) one
have  
z̃ x̃ −1 = z2 y + y−1
This are three independent equation for the three independent vari-
ables ( g0 , k0 , h0 ). Integrating out z and using h = 0 which means y = 1
we can prove that
x 2 + x −2
ỹ2 = = 1 ⇒ h0 = h = 0
x 2 + x −2
That means that in absence of an initial external field we cannot
obtain an external field. The last equation is

2
x 2 + x −2 1
4
x̃ = ⇒ k0 = ln (cosh (2k))
4 2
96

This is the equation that define our transformation. Notice that Since
0
ek + 1 = 2 cosh2 (k)
 
k0 = tanh−1 tanh2 (k) and
0
e2k − 1 = 2 sinh2 (k )

And if y = tanh(k) then y0 = y2 . This exponent is related to l.

Let’s find the fixed point. y ∈ [0, 1] and so the two fixed point
are 0+ and 1− . y = 1− is an unstable fixed point since if one take a
(n)
y0 < 1 then R2 (y0 ) → 0+ while y = 0+ is stable for the same reason.
The phase diagram of the system is provided in figure 48. Figure 48: Phase diagram of the one-
dimensional Ising model.
As we have already notice there is a problem called proliferation
problem.
Let’s discuss it in D = 2, as in figure 49 when we done a coarse
grained approximation some new term in the Hamiltonian appear
and so, even if in the original Hamiltonian there are only interaction
between neighbors spins, in the new Hamiltonian we can see new
terms. To solve this problem we have to introduce new variables from
the starting point (eventually fixed to zero in the starting Hamilto-
nian).
Since in principle we have to perform infinitely steps, infinitely
Figure 49: Two dimensional Ising
terms can appear in our Hamiltonian, to avoid it one has to approxi- model: coarse grained system. Notice
mate at a certain step. Lets’ study the two-dimensional Ising model that now there could appear new inter-
action we are called Q−interactions.
neglecting the Q term (see figure 49). The equation for h = 0 is

3
k0 = ln (cosh(4k))
8
There are three fixed point: k? = 0, k? = ∞ and k = k C ' 0.50689. Recall that the Onsanger solution has
The phase diagram is show in figure 50 . Notice that before k C k C ' 0.4406.

the system in the paramagnetic phase, after it the system is in the

ferromagnetic phase; this show that k C is the critical point.
Let’s compute the critical exponent yt , since

dk0 

ln yt Figure 50: Phase diagram of the two-
λt = ⇒ yt = √ ' 1.070
dk k=kC ln 2 dimensional Ising model.

and so the ν exponent is ν = y−

t
1
' 0.93 and the alpha exponent
α ' 0.1313.
Before conclude this chapter we want to introduce a geometrical
approach to the renormalization groups which used the transfer
matrix T. Since
    N 
2
ZN = Tr T N = Tr T2

And this means that the transfer matrix became

Tl = Tl
Spontaneous symmetry breaking

In this chapter we want to introduce the spontaneous symmetry

breaking, in particular we want to introduce this concept in the
physics of fundamental interactions, in particular the Higgs boson.
When we talk about broken symmetry we refer to a Hamiltonian
of two (or more) terms such that one is not invariant under a group
G while the other is. So in general the total Hamiltonian is invariant
under a subgroup of G . For instance in the Ising-model Other important broken symmetry
are the atom in presence of an electric
H=J ∑ Si Sj + ∑ Hi Si or magnetic field (which break the
symmetry).
hi,ji i

the Z2 symmetry is broken by the second term. In this example the

symmetry is explicitly broken, but we want to study the spontaneous
symmetry breaking. In this case the Hamiltonian is symmetric but
the variables used to describe the system became asymmetric. Let’s
consider the discrete Z2 symmetry of an Ising-Model (H = 0). Let’s
recall that in the saddle point approximation the equation of state is

− 4 φ + r0 φ + u0 φ3 = h

If h does not depend on ~x the solution is uniform grad φ = 0. The

saddle point is equivalent to find the uniform value of φ0 which is an
extrema point of the potential

1 u0
V (φ) = r0 φ2 + φ4 − hφ
2 4
For h = 0  
V 0 = r0 + u0 φ2 φ = 0

And so for T > TC the unique solution is φ0 = 0 but for T < TC there
are two possible solutions Notice that the two solution are related
by the transformation Z2 : φ → −φ
and since the two states have a lower
r
r0
φ0 = ± − symmetry than state φ0 = 0 if the
u0 thermal fluctuation are strong enough
the symmetry is preserved since
If the thermal fluctuation is not too high, the physical state are the hφi = 0.
fluctuation around the minima φ0 and so φ → φ0 + δφ. So δφ now
98

is the new variable, bt the presence of a non vanishing φ0 breaks the

symmetry. One can re-write the Hamiltonian to see
Let’s now study a continuous symmetry, some example can be the it.

superfluids or the superconductor in which a complex field

φ1 + iφ2
φ= √
2
has a non trivial vacuum. In particle physics we can study the action

1 r u
Z
S[φ] = d4 x L( ϕ) L = − ∂µ φ∂µ φ? − 0 φφ? − 0 (φφ? )2
2 2 4
√
This action describe a complex scalar field with a mass m = r0 . This
action has a global U (1) symmetry The potential is Which means

φ → eiθ φ
r0  2 2 u 
0
2
V (φ1 , φ2 ) = φ1 + φ22 + φ12 + φ22 where θ ∈ R (does not depend on x).
2 4
and it is a mexican hat potential.
The ground state are
r0
φ12 + φ22 = v2 = −
u0
So we have a circle of ground states and we can choose one of them.
Let’s choose for instance φ1 = v and φ2 = 0 so that the new fields are

φ1 = v + δφ1 φ2 = δφ2

Let’s rewrite the Lagrangian neglecting the terms in v, we obtain

Figure 51: A mexican hat potential.
1
2 1
2
L =− ∂µ δφ1 − ∂µ δφ2 − λv2 δφ12 +
2 2
  λ 2
− vλδφ1 δφ12 + δφ22 − δφ12 + δφ22
4
Notice that the mass of δφ1 is 2λv2 . However the Lagrangian is not
invariant under δφ1 → −δφ1 . Notice that the field δφ2 has no mass. Someone usually speak about hidden
This particles is called Goldstone boson and the existence of such symmetry because the Lagrangian is
still invariant under the transformation
a boson is predicted by the Goldstone theorem which tells that in in φ1 and φ2 but not in δφ1 and δφ2 , and
every spontaneous symmetry breaking of a continuous symmetry so in some sense the U (1) symmetry
is still there but hidden in the new
exists a particle with no mass. In general the new Lagrangian is variables.
invariant under a subgroup P of the initial one G and so there are
dim G − dim P Goldstone bosons.
This is the essential point of the Higgs mechanism . In order To more information we suggest
to satisfy the local U (1) symmetry we have to promote the usual Peter Higgs. Prehistory of the higgs
boson. 2007
derivative to a covariant derivative

∂µ → Dµ = ∂µ + iqAµ

The Lagrangian in Where

Fµν = ∂µ Aν − ∂ν Aµ
and
m2 ? λ
V= φφ + (φφ? )2
2 4
 99

1
?
L = − Fµν F µν + Dµ φ ( D µ φ) − V (φ, φ? )
4
If m2 < 0 we can introduce a non trivial vacuum state φ0 =
√
−m2 /λ = v and so one can choose φ1 = v and φ2 = 0. In the
new Lagrangian two new terms appear which are λv2 δφ1 which tells
√
that the mass of δφ1 = v 2λ and q2 v2 Aµ Aµ that means that the
√
boson Aµ has a mass 2qv. One can also use a gauge transformation
on Aµ such that
1
Aµ → Aµ + ∂µ δφ2
qv
So that the field φ2 is eaten by the vector boson Aµ since the new
Lagrangian is

1 1
2
L = − Fµν F µν + ∂µ δφ1 − λv2 δφ1 + q2 v2 Aµ Aµ + h.o.
4 2
This Lagrangian density predict some new phenomena we can
observe!! Notice that we have simple re-distribute the degrees of
freedom, in fact we started from 2 + 2 = 4 (two massive scalar fields
plus a massless vector boson) and we obtained 1 + 3 = 4 (one massive
scalar field and one massive vector boson). For more detail on this topics we
One can apply it also in the non-abelian gauge group: the idea suggest
Michele Maggiore. A Modern Introduc-
is the same but since the group now is not abelian (as U (1)) the tion to Quantum Field Theory
representation has dimension > 1 and there are some technical
complications, but we had experimentally prove that this calculations
predict the correct quantities.
Exercises

Exercise. Consider the following phenomenological equation of state for non

ideal gas
p N − Na
= e k B TV
kB T V − Nb
• Find the critical point ( pC , vC , TC ) as a function of the parameters a, b.
(Hint: write first the equation in terms of the specific volume v);

• By using virial expansion, compare (in terms of the 2nd and 3rd virial
coefficients and Boyle’s temperature) this equation of state with the van
der Walls equation of state.

Solution. We know that in the critical point the pressure p shows a flex
respect to the (specific) volume v. So our conditions are Since v = V/N th equation of state is
1 βa
∂p ∂2 p p= e− v
β (v − b)
=0 =0
∂v ∂v2
Performing the calculation

∂p 1 βa a βa p pβa
=− 2
e− v + 2
e− v = − + 2 =0
∂v β(v − b) (v − b)v v−b v

And so
1 βa
= 2
v−b v
The second derivative is
βa βa βa βa
∂2 p 2e− v 3ae− v 2ae− v a2 βe− v
= − − + =0
∂v2 β ( v − b )3 ( v − b )2 v2 ( v − b ) v3 ( v − b ) v

Using the equation shown before

a3 β2 2a2 β aβ
= − ⇒v=−
v6 v5 2
And using again the fact that

1 βa
= 2
v−b v
102

One can find

2 1 4b 3a
= ⇒β= ⇒ TC =
βa − aβ + 2b 3a 4bk B

Since now we have β we can compute also v which is

2
vC = − b
3

And using the equation of state we can get also the pressure

9 a 2
pC = e
25 b2

Now we have to compute the virial expansion in terms of ρ = V/N,

so
 
ρ 1
βρ = e− βρa = ρ(1 + ρb − ρ2 b2 ) 1 − βρa + β2 ρ2 a2 + O(ρ4 ) =
1 − ρb 2
1
= ρ − βρ2 a + β2 ρ3 a2 + ρ2 b − ρ3 βab − ρ3 b2 O(ρ4 ) =
2
= ρ + B2 ( T )ρ2 + B3 ( T )ρ3 + O(ρ4 )

where
1 2 2
B2 = b − βa B3 = β a − βab − b2
2
Observe that B2 is equal to the vann der Walls’ one and so the Boyle’s
temperature is the same of the Walls’ one: Recalling that B2 ( TB ) = 0.

a
TB =
kB b

The terms B3 is not equal to the vann der Walls’ B3 and so the expan-
sion is equal up to ρ3 terms.

Exercise. A spin half Ising model with three spin interactions on a triangu-
lar lattice has Hamiltonian

H = −J ∑ Si S j S k
i,j,k ∈∆

where the sum is over all nearest neighbor triangles on an infinite triangular
lattice ad the interaction is ferromagnetic (J > 0).

• Using a leading order expansion in the flcutuations find the mean field
Hamiltonian for this problem;

• Using the mean field Hamiltonian and assuming an homogeneous state

where mi = m, find an expression for the the mean field Helmholtz free
energy and the mean field equation for the problem;
 103

• Taking J = 1, skecth the bejaviour of the solutions to the mean field

equation as a function of T. Is the order parameter continuous or discon-
tinuous as T → TC ?

Solution. First of all let’s compute

Si S j S k = ( Si − m + m ) S j − m + m ( S k − m + m ) =
= m3 + m2 (Si − m) + (S j − m) + (Sk − m) + O((Si − m)2 ) =

= −2m3 + m2 (Si + S j + Sk ) + O((Si − m)2 )

So the Hamiltonian is (we have already use mi = m)

H = −2Jm3 ∑ 1 + m2 J ∑ ( Si + S j + S k )
i,j,k∈∆ i,j,k ∈∆

Now let’s keep free the index i. When we choose one i we have to
sum over z = 6 triangles (one can simply do a little draw of the
situation). So
N
∑ 1= ∑ z = zN
i,j,k ∈∆ i =1

Similar steps can be followed for the second term but this time

∑ Si = z ∑ Si
i,j,k ∈∆ i

and so
−H = −2m3 zN J + 3m2 Jz ∑ Si
i

Let’s compute the partition function It is important to recall that this time
the spin is ±1/2 and not ±1 as usual.
N
∑ ∑ ∏ e3βm JzSi =
3 zN J 2
ZN = e− βH = e−2βm
{Si =±1/2} {Si =±1/2} i =1
N    N
 3
=e −2βm3 zN J
∑ e 3βm2 JzS
=e −2βm3 zN J
2 cosh
2
βm2 Jz
S=±1/2

Now one can compute also the Helmholtz free energy

    
3
FN = −k B T ln ZN = −K B TN −2βm3 zJ + ln cosh βm2 Jz + ln 2
2

The equilibrium states are given by the stationary point of the free
energy and so
 
∂FN 3 3 2
= 0 ⇒ m = tanh m Jzβ or m=0
∂m 2 2
 
3 3 2
Since there is an m2 the derivative of y = tanh m Jzβ com-
2 2
puted in m = 0 has to be zero and so if some other intersection with
y = m appears m has to be finite.
104

Exercise. Consider the following Landau free energy

F (η, χ) = atη 2 + bη 4 + cχ2 + gη 2 χ

where t = T − TC and a,b,c,g are positive constant.

• Minimize the free energy with respect to χ for fixed η and determine χ?
as a function of η;

• Obtain the effective free energy for η. What is the new renormalised value
b of b ?

• Which is the behavior of b as a function of g ?

• Discuss the phase transition as a function of g.

Solution. By a simply calculation

∂F c 2gη 2
= χ + gη 2 = 0 ⇒ χ? =
∂η 2 c

So the effective free energy is

g2
 
2
F (η ) = atη + b − η4
4c

and so
g2
 
b= b−
4c

The new b, quadratic in g, is the renormalized parameter, in fact we

have integrated out the variable χ. The equilibrium phase are such
that
g2
  
∂F at 1
=0⇒η + b − η2 =0
∂η 2 4 4c
which means η = 0 or
s
8ac
η=± −
4cb − g2

But the two solution η 6= 0 exist only if g2 > 4cb which means
√
g > 2 cb = g? . So for g < g? there is no phase transition since
there is only one equilibrium phase for η = 0. If g > g? there could
be a phase transition between the two phase η 6= 0 and if g = ? the
equation became
at
η =0
2
which means η = 0 and so there cannot be phase transitions.
 105

Exercise. Given the Landau free energy

r 2 u 4
L(m, h) = m + m − hm
2 4
show that, the magnetic suscettivity χ

∂hmi 
χ=
∂h h=0

satisfies the relation

 χ( TC + ∆T ) 
 
 χ( T − ∆T )  = 2
 
C

Solution. Let’s find the equation of state

∂L
= 0 ⇒ h = rm + um3
∂m
Deriving both the member in h one can simply find

1
1 = rχ + 3um2 χ ⇒ χ =
r + 3um2

Since m( TC + ∆T ) = 0 one have

1
χ( TC + ∆T ) =
r
while in TC − ∆T one can get, with h = 0
r
m(r + um2 ) = 0 ⇒ m2 = −
u
and so
1 1
χ( TC + ∆T ) = =
r − 3r −2r
and so
 χ( TC + ∆T ) 
 
 χ( T − ∆T )  = 2
 
C

Exercise. Let us consider a 1D system of N spins si with si = ±1. Suppose

that this system is subject to an alternating magnetic field. The Hamiltonian
of the system is given by

N  
H = h ∑ 1 + 2(−)i Si
i =1

Compute

• The partition function of the system

• The free energy of the system at temperature T

• The magnetization of the i-esim spin si ;

106

• Does this model display a phase transition in the limit of h → 0 ? Argue

your answer.
Solution. To compute the partition function let’s assume first N odd. In
this case
N/2 N/2
− βH = ∑ ( βh)Si + ∑ (−3βh)Si
i =1 i =1
Now let’s define ho = βh and he = −3βh such that
N/2 N/2
− βH = ∑ (ho + he )Si = −2βh ∑ Si
i =1 i =1

And so
N/2
ZN = ∑ e− βH = ∑ ∏ e−2βhSi = 2N/2 coshN/2 (2βh)
{Si =±1} {Si =±1} i =1

But if we consider a even N:

( N +1)/2 ( N −1)/2
− βH = ∑ ( βh)Si + ∑ (−3βh)Si
i =1 i =1

which means
( N −1)/2 2 ( N −1)/2 2 ( N −3)/2
− βH = ∑ (ho + he )Si + ∑ ho Si = −2βh ∑ Si + 2βh ∑ Si = −2βh ∑ Si
i =1 i =1 i =1 i =1 i =1

And so
( N −3)/2
ZN = ∑ e− βH = ∑ ∏ e−2βhSi = 2( N −3)/2 cosh( N −3)/2 (2βh)
{Si =±1} {Si =±1} i =1

The free energy of the system is simply

FN = −k B T ln ZN

And we have to separate the two cases: if N is odd

k B TN
FN = − ln (2 cosh (2βh))
2
while if N is even
k B T ( N − 3)
FN = − ln (2 cosh (2βh))
2
For the magnetization if i is odd Notice that the result does not depend
on the parity of N, in reality it depend
∂ ln ZN 2k TN 2 sinh(2βh) 2 tanh(2βh) on the limit in which N → ∞.
mi = = B

=
∂( βh) k B
 TN 2
cosh 2βh ln ( 2 cosh ( 2βh )) ln ( 2 cosh ( 2βh ))
If i is even
∂ ln ZN 2k  TN
sinh(2βh)
2 2 tanh(2βh)

mi = − =− B  =−
∂(3βh) 3k B TN

 2 cosh 2βh ln (2 cosh(2βh)) 3 ln (2 cosh(2βh))
In the limit h → 0 one can simply get mi = 0 (both for i odd and
even). This means that the order parameter is zero for all tempera-
ture and so there cannot exist a phase transition.
 107

Exercise. A Ising model with four spin interactions on a square lattice has
the Hamiltonian
H = − J ∑ Si S j S k S l
i,j,k,l ∈

where the sum is over all nearest neighbour unit squares of a square lattice
and J > 0 (ferromagnetic interaction).

• Find the mean field Hamiltonian of this problem;

• Using the mean field Hamiltonian and assuming a homogeneous state

(mi = m, ∀i), find the expression for the mean field free energy and the
self-consistent equation of the order parameter;

• Assuming J = 1, expand the limiting free energy density f and discuss

its behaviour with respect to the standard Ising mean field where the
spin-spin interaction is only between bonds.

Solution. Let’s write, using m = mi for all i,

Si S j Sk Sl = (Si − m + m)(S j − m + m)(Sk − m + m)(Sl − m + m) =

  
' m((Si − m) + (S j − m)) + m2 m((Sk − m) + (Sl − m)) + m2 =
 
' m 2 m ( Si − m ) + ( S j − m ) + ( S k − m ) + ( S l − m ) + m 2 =

= m2 (−3m2 + m(Si + S j + Sl + Sl )) =
= −3m4 + m3 (Si + S j + Sk + Sl )

And so
− βH = 3βJNzm4 − 4m3 βJz ∑ Si
i

Where z is the number of possible square when we fixed i, which

means z = 4.The partition function is
N
∑ ∑ ∏ e−4m βJzSi =
4 3
ZN = e− βH = e3βJNzm
{Si =±1} {Si =±1} i =1
3βJNzm4 N N 2
=e 2 cosh (4m βJz)

So the self consistence equation can be computed as

4 3 βJzS
e3βJNzm ∑{Si =±1} ∏iN=1 Si e−4m i
m= 4 3
= tanh(4m2 βJz)
e3βJNzm ∑{Si =±1} ∏iN=1 e−4m βJzSi

and so it is
m = tanh(4m2 βJz)
The bulk free energy

k B T ln ZN  
f b = − lim = −3Jzm4 − k B T ln cosh(4m2 βJz) − k B T ln 2
N →∞ N
108

Using z = 4 and J = 1 one have

 
f B = −12m4 − k B T ln cosh(16m2 β) − k B T ln 2

The approximation for small m bring us to

(16m2 β)2 (16m2 β)4

f B ' −k B T ln 2 − 12m4 − k B T + kB T
2 24
The difference between the Ising model in which the interactions are
only between bonds is that now the first, non constant, term in the
bulk free energy is of the order m4 and not m2 .
Bibliography

Als-Nielsen and R. J. Birgeneau. Mean field theory, the ginzburg

criterion, and marginal dimensionality of phasetransitions. 1977.

Denis Andrienko. Introduction to liquid crystals. 2018.

Somendra M Bhattacharjee. Flory theory for polymers. 2013.

Callen and Herbert. Thermodynamics and an Introduction to Thermo-

statistics. Graphics Press, LLC, second edition, 1985.

E. Fermi. Termodinamica. November 1977.

Robert B. Griffiths. Peierls proof of spontaneous magnetization in a

two-dimensional ising ferromagnet. 1964.

E. A. Guggenheim. The principle of corresponding states. December

2004.

D. Ter Haar. Collected papers of L.D. Landau. 1965.

Haber. Expansion of plane waves in spherical harmonics. URL

http://scipp.ucsc.edu/~haber/ph215/PlaneWaveExpansion.pdf.

Peter Higgs. Prehistory of the higgs boson. 2007.

K. Huang. Statistical Mechanics. Wiley, second edition, 2008.

Leo P. Kadanoff. Operator algebra and the determination of critical

indices. 1968.

C. Kittel. Phase transition of a molecular zipper. 1969.

Michele Maggiore. A Modern Introduction to Quantum Field Theory.

Lars Onsager. A two-dimensional model with an order-disorder

transition. October 1944.

S. M. Shapiro and H. Z. Cummins. Critial opalescence in quartz.

October 1968.
110

V.S. Vorob’ev. The law of corresponding states for the entropy of rare
gases. December 2003.

Masahiro Yamamoto. Origin of the atttractive part of lennard-

jones potential. URL http://www.chem.konan-u.ac.jp/PCSI/web_
material/LJ.pdf.
Index

Adiabatic walls, 7 finite free energy, 26 Kosterlitz-Thauless transition, 53

finite system, 25
basis of attraction, 93 fixed point, 93 lambda transition, 62
Blume-Emery-Griffith model, 62 flexible wall, 7 Landau free energy, 73
Boltzmann factor, 21 fluctuation-dissipation relation, 28 laten heat, 16
Boyle’s temperature, 68 free boundary condition, 34 Legendre transformation, 10
Bragg-Williams, 61 fugacity, 23 Lennard-Jones potential, 70
bulk free energy, 26
gap exponent, 90 magnetic susceptibility, 13
canonical ensamble, 21 gaussian approximation, 84 marginal eigenvalue, 94
charge of the vortex, 53 gaussian model, 84 maximum entropy postulate, 9
Clausius- Clapeyron equation, 17 Gibbs ensable, 22 Maxwell construction, 67
cluster expansion, 69 Gibbs Potential, 11 Maxwell relations, 12
coarse graining procedure, 80 Gibbs-Duhem relation, 10 Mayer function, 69
connected correlation function, 28 Ginzburg argument, 79 metastable state, 76
coupling constant, 25 Ginzburg criterium, 79 mexican hat potential, 98
critial opalescence, 29 Ginzburg-Landau theory, 81 microcanonical ensamble, 21
critical exponent, 18 Goldstone boson, 98 molar heat capacity, 12
critical fixed point, 93 Gran canonical ensamble , 23
critical manifold, 93 Gran canonical potential, 11 order parameter, 17
Curie law, 41 Griffith equality, 90
cutoff wavelength, 80 Griffith inequality, 20 paramagnetic system, 16
partition function, 22
diathermic wall, 7 Pascal law, 10
diluted Ising model, 37 Hamiltonian mean field theory, 53 periodic boundary condition, 36
DNA denaturalization, 43 Hartree method, 60 permeable wall, 7
domain wall, 49 Helmholtz free energy, 10 Perron-Frobenius theorem, 38
Hentalpy, 11 Potts model, 51
Energy-Entropy competition argu- hidden symmetry, 98 proliferation problem, 96
ment, 49 Higgs boson, 97
entropy, 8 homogeneous function, 89 quasi-static transformation, 7
equal probability rule, 21 Hubber-Stratonovitch identity, 54
equation of states, 9 hyperscaling relation, 92 relevant eigenvalue, 94
equilibrium states, 7 Rigid walls, 7
Euler equation, 11 impermeable wall, 7 Rushbnocke inequality, 20
Euler-Lagrange equations, 82 intensive variable, 9
extensive variable, 7 irrelevant eigenvalue, 94 Saddle point approximation, 55
Ising model, 31 scaling exponents, 91
ferromagnetic system, 16 isothermal compressibility, 12 self consistent equation, 55
112

self-consistent equation, 58 transition matrix, 36 viral coefficients, 68

semi-group, 92 tricritical point, 62 viral expansion, 68
slitness energy, 52 triple point, 15
spontaneous symmetry breaking, 97 trivial fixed point, 93
superspin, 91 two point connected correlation XY model, 51
function, 38
thermal expansion, 12
thermodynamical potentials, 10 upper critical dimension, 79 Zipper model, 43