Statistical physics:

phase transitions
Denis Grebenkov
CNRS – Ecole Polytechnique
[email protected]

September 2022
 Scope
1. Motivation, qualitative description, Ising model

2. Mean field approximation and critical exponents

3. (Ginzburg-)Landau theory of phase transitions

 Reminder: The Ising model
We consider a simplified model with N sites occupied by
atoms with spins taking only two values: 𝑠𝑖 = ±1
The energy of a given configuration is 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑗
exp(−𝛽𝐸) 𝑖 𝑖∼𝑗
Its probability is 𝑃( 𝑠1 , … 𝑠𝑁 ) = 𝑍(𝛽) Interactions with Inter-spin
magnetic field interactions
𝑀
Measurable quantity is the magnetization: 𝑀 = ෍ 𝑠𝑖 𝑚=
𝑖 𝑁

It is “enough” to compute the partition function 𝑍 𝛽 =෍ exp(−𝛽𝐸)

{𝑠1 ,…,𝑠𝑁 }
This is yet unsolved problem in 3D!

We search for an approximate solution of this N-body problem

 Mean field approximation (MFA)
Basic idea: an infinite system of a single particle in
interacting particles an effective medium

𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗 = − ෍ 𝑠𝑖 𝐵 + 𝐽 ෍
𝑖 𝑗 𝑖
𝑠𝑗 = − ෍ 𝑠𝑖 𝐵 + 𝐽 ෍
𝑖 𝑗 𝑖
𝑠𝑗 + (𝑠𝑗 − 𝑠𝑗 ) !
mean fluctuation
local field

𝐸 𝑚 = − ෍ (𝐵 + ℎ𝑖𝑚 )𝑠𝑖 ℎ𝑖𝑚 = 𝐽 ෍ 〈𝑠𝑗 〉 is the mean local field

𝑖 𝑗(𝑖)

Σ 𝑠1,…,𝑠𝑁 𝑋 𝑠1 , … , 𝑠𝑁 exp(𝛽Σ𝑖 𝑠𝑖 𝐵 + ℎ𝑖𝑚 ) Fully decoupled, as for

𝑋 = independent particles!
Σ 𝑠1,…,𝑠𝑁 exp(𝛽Σ𝑖 𝑠𝑖 𝐵 + ℎ𝑖𝑚 )
Σ𝑠1 exp(𝛽𝑠1 𝐵 + ℎ1𝑚 ) … Σ𝑠𝑁 exp(𝛽𝑠𝑁 𝐵 + ℎ𝑁𝑚
) 𝑋 𝑠1 , … , 𝑠𝑁
=
Σ𝑠1 exp(𝛽𝑠1 𝐵 + ℎ1𝑚 ) … Σ𝑠𝑁 exp(𝛽𝑠𝑁 𝐵 + ℎ𝑁
𝑚
)
(see slide 36)
 Mean field approximation (MFA)
Basic idea: an infinite system of a single particle in
interacting particles an effective medium
number of
For translationally invariant systems: 𝑠𝑖 = 𝑚 neighbors

ℎ𝑖𝑚 = 𝐽 ෍ 〈𝑠𝑗 〉 ℎ𝑖𝑚 = 𝑧𝐽〈𝑚〉

𝑗(𝑖)
exp(𝛽(𝐵 + 𝑧𝐽 𝑚 ) σ𝑖 𝑠𝑖 )
𝐸𝑚 = −(𝐵 + 𝑧𝐽 𝑚 ) ෍ 𝑠𝑖 𝑚
𝑃 ( 𝑠1 , … 𝑠𝑁 ) =
𝑖 𝑍 𝑚 (𝛽)
However, the Boltzmann’s weights depend on 〈𝑚〉 , Self-consistent (or
which in turn should be determined with that weights closure) equation
Σ 𝑠1,…,𝑠𝑁 𝑠1 exp 𝛽 𝐵 + 𝑧𝐽 𝑚 (𝑠1 + ⋯ + 𝑠𝑁 ) Σ 𝑠1 𝑠1 exp 𝛽 𝐵 + 𝑧𝐽 𝑚 𝑠1
𝑚 = =
Σ 𝑠1,…,𝑠𝑁 exp 𝛽 𝐵 + 𝑧𝐽 𝑚 (𝑠1 + ⋯ + 𝑠𝑁 ) Σ 𝑠1 exp 𝛽 𝐵 + 𝑧𝐽 𝑚 𝑠1
(+1) exp 𝛽 𝐵 + 𝑧𝐽 𝑚 (+1) + (−1) exp 𝛽 𝐵 + 𝑧𝐽 𝑚 (−1)
= = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )
exp 𝛽 𝐵 + 𝑧𝐽 𝑚 (+1) + exp 𝛽 𝐵 + 𝑧𝐽 𝑚 (−1)
 Mean field approximation (MFA) 1

+𝑚𝑠
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )

tanh(𝛽𝑧𝐽 𝑚 )
0.5

Let B = 0: 〈𝑚〉 = tanh(𝛽𝑧𝐽 𝑚 )

-0.5

−𝑚𝑠
The slope of tanh(𝛽𝑧𝐽 𝑚 ) near 0 distinguishes: -1
-2 -1 0 1 2
𝑚
1) If 𝛽𝑧𝐽 ≤ 1 , then the only solution is 𝑚 = 0
𝑇 ≥ 𝑇𝑐 = 𝑧𝐽/𝑘𝐵 Disordered (paramagnetic) phase 1

2) If 𝛽𝑧𝐽 > 1 , then there are three solutions 0.5 Disordered

𝑇 < 𝑇𝑐 = 𝑧𝐽/𝑘𝐵 𝑚 = 0, −𝑚𝑠 , +𝑚𝑠 Ordered phase
0
The solution 𝑚 = 0 is unstable (Peierls argument)! phase
Phase
-0.5
Ordered (ferromagnetic) phase transition
Symmetry breaking: an arbitrarily small field B>0 “selects” -10 0.5 1 1.5 2
the solution 𝑚 = +𝑚𝑠 to minimize the free energy
 Validity of the MFA: Ginzburg’s criterion
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )

The accuracy of the MFA increases with the number of neighbors 𝑧 = 2𝑑

𝐸 = −𝐽 ෍ 𝑠𝑖 𝑠𝑗 = − ෍ 𝑠𝑖 𝐽෍ 𝑠𝑗 𝐸 𝑚 = − ෍ 𝑠𝑖 ℎ𝑖𝑚
𝑖∼𝑗 𝑖 𝑗 𝑖 𝑖

𝐽∗
local field ℎ𝑖 𝑠𝑗 = ℎ𝑖𝑚 + 𝑧 σ𝑗 𝑖 (𝑠𝑗 − 𝑠𝑗 )
𝑂( 𝑧)
The MFA is thus exact in the limit 𝑑 → ∞
Moreover, if we are interested in the critical behavior (i.e., at 𝑇~𝑇𝑐 ), then the
MFA yields exact results for 𝑑 ≥ 𝑑𝑐 = 4 (Ginzburg’s criterion) (see below)

In contrast, in dimensions d=2,3, the MFA predictions are qualitatively correct

but inaccurate, whereas for d=1, the MFA wrongly predicts phase transitions.
 Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point
Let us investigate the behavior of several quantities:
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )
Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0)
𝐵
Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵)
𝜕〈𝑚〉
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ 𝑇𝑐 𝑇
𝜕𝐵 𝐵=0
𝜕〈𝐸〉
Specific heat near Tc 𝐶 𝑇 = ቚ
𝜕𝑇 𝐵=0
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗

How to get them? Exact solution, approximation solution, renormalization group,

1D/2D MFA (see TD) simulation, experiment
 Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point
Let us investigate the behavior of several quantities:
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )
Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0) 1
𝑚 𝑇, 0 ∝ 𝑇𝑐 − 𝑇 𝛽′
Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵) 0.5
𝜕〈𝑚〉
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ 0
𝜕𝐵 𝐵=0
𝜕〈𝐸〉 -0.5
Specific heat near Tc 𝐶 𝑇 = ቚ
𝜕𝑇 𝐵=0 -1
0 0.5 1 1.5 2
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗

How to get them? Exact solution, approximation solution, renormalization group,

1D/2D MFA (see TD) simulation, experiment
 Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point.
Let us investigate the behavior of several quantities:
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )
Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0) 1

Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵)

𝜕〈𝑚〉
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ 0.5
𝜕𝐵 𝐵=0
𝜕〈𝐸〉 𝑚 𝑇𝑐 , 𝐵 ∝ 𝐵1/𝛿
Specific heat near Tc 𝐶 𝑇 = ቚ
𝜕𝑇 𝐵=0 0
0 0.5 1 1.5 2
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗

How to get them? Exact solution, approximation solution, renormalization group,

1D/2D MFA (see TD) simulation, experiment
 Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point.
Let us investigate the behavior of several quantities:
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )
Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0) 150

Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵)

100
𝜕〈𝑚〉
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ
𝜕𝐵 𝐵=0 50
𝜕〈𝐸〉 𝜒 ∝ 𝑇𝑐 − 𝑇 −𝛾− 𝜒 ∝ 𝑇 − 𝑇𝑐 −𝛾+
Specific heat near Tc 𝐶 𝑇 = ቚ
𝜕𝑇 𝐵=0 0
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗 0 0.5 1 1.5 2

How to get them? Exact solution, approximation solution, renormalization group,

1D/2D MFA (see TD) simulation, experiment
 Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point.
Let us investigate the behavior of several quantities:
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )
Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0) 1.5
𝐶(𝑇𝑐− )
Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵)
1
𝜕〈𝑚〉 𝐶 𝑇 ∝ 𝑇𝑐 − 𝑇 −𝛼
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ
𝜕𝐵 𝐵=0 0.5
𝜕〈𝐸〉
Specific heat near Tc 𝐶 𝑇 = ቚ 𝐶(𝑇𝑐+ )
𝜕𝑇 𝐵=0 0
0 0.5 1 1.5 2
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗

How to get them? Exact solution, approximation solution, renormalization group,

1D/2D MFA (see TD) simulation, experiment
 𝑇Τ𝑇𝑐 = 0.97 𝑇Τ𝑇𝑐 = 0.99
Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point.
Let us investigate the behavior of several quantities: 𝑇Τ𝑇𝑐 = 1 𝑇Τ𝑇𝑐 = 1.01

Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0)

Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵)
𝜕〈𝑚〉
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ
𝜕𝐵 𝐵=0
𝜕〈𝐸〉 𝑇Τ𝑇𝑐 = 1.06 𝑇Τ𝑇𝑐 = 1.15
Specific heat near Tc 𝐶 𝑇 = ቚ
𝜕𝑇 𝐵=0
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗
𝑟𝑖𝑗
near Tc 𝐶𝑖𝑗 ∝ exp −
𝜉 𝑇
 Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point.
Let us investigate the behavior of several quantities:
〈𝑚〉 = tanh(𝛽𝐵 + 𝛽𝑧𝐽 𝑚 )
Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0)
[SHOW VIDEO ]
Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵)
𝜕〈𝑚〉 In 1D (see TD1)
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ −1
𝜕𝐵 𝐵=0 𝜉 𝑇 =
𝜕〈𝐸〉 ln tanh(𝐽/(𝑘𝐵 𝑇))
Specific heat near Tc 𝐶 𝑇 = ቚ
𝜕𝑇 𝐵=0 𝜉 𝑇 ≈ 𝑒 2𝐽/(𝑘𝐵 𝑇) 𝑇 → 0
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗
𝑟𝑖𝑗 However, generally, 𝜉 𝑇 ∝ 𝑇 − 𝑇𝑐 −𝜈±
near Tc 𝐶𝑖𝑗 ∝ exp −
𝜉 𝑇 𝑑−2+𝜂
at Tc 𝜉 𝑇𝑐 = ∞ 𝐶 𝑖𝑗 ∝ 1/𝑟𝑖𝑗 Self-similarity!
 Critical exponents
When T crosses Tc, the system undergoes drastic changes in its macroscopic state
One may thus expect that most quantities are not analytic near the critical point.
Let us investigate the behavior of several quantities:
𝛽′ 𝛽 ′
Magnetization near Tc at B=0 〈𝑚〉(𝑇, 0) 𝑚 𝑇, 0 ∝ 𝑇𝑐 − 𝑇 𝑇 < 𝑇𝑐

Magnetization at Tc for B0 〈𝑚〉(𝑇𝑐 , 𝐵) 𝑚 𝑇𝑐 , 𝐵 ∝ 𝐵1/𝛿 𝛿

𝜕〈𝑚〉 𝛾± 𝛾±
Magnetic susceptibility near Tc 𝜒 𝑇 = ቚ 𝜒 ∝ 𝑇 − 𝑇 𝑐
𝜕𝐵 𝐵=0
𝜕〈𝐸〉 −𝛼 𝛼𝑇
Specific heat near Tc 𝐶 𝑇 = ቚ 𝐶 𝑇 ∝ 𝑇𝑐 − 𝑇 𝑇 < 𝑐
𝜕𝑇 𝐵=0 𝜈±
Correlations near or at Tc 𝐶𝑖𝑗 = 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗
𝑟𝑖𝑗 However, generally, 𝜉 𝑇 ∝ 𝑇 − 𝑇𝑐𝜂−𝜈±
near Tc 𝐶𝑖𝑗 ∝ exp −
𝜉 𝑇 𝑑−2+𝜂
at Tc 𝜉 𝑇 =∞ 𝐶 𝑖𝑗 ∝ 1/𝑟𝑖𝑗
 Critical exponents
Ising model 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗

𝛼 𝛽′ 𝛾± 𝛿 𝜈± 𝜂 𝑠𝑖 = ±1
 Critical exponents
Ising model 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗

𝛼 𝛽′ 𝛾± 𝛿 𝜈± 𝜂 𝑠𝑖 = ±1
Exact solution d=2 ln(𝑇𝑐 − 𝑇) 1/8 7/4 15 1 1/4
 Critical exponents
Ising model 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗

𝛼 𝛽′ 𝛾± 𝛿 𝜈± 𝜂 𝑠𝑖 = ±1
Exact solution d=2 ln(𝑇𝑐 − 𝑇) 1/8 7/4 15 1 1/4

Simulations d=3 0.110 0.326 1.237 4.800 0.630 0.036

 Critical exponents
Ising model 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗

𝛼 𝛽′ 𝛾± 𝛿 𝜈± 𝜂 𝑠𝑖 = ±1
Exact solution d=2 ln(𝑇𝑐 − 𝑇) 1/8 7/4 15 1 1/4

Simulations d=3 0.110 0.326 1.237 4.800 0.630 0.036

MFA (see TD) d4 0 1/2 1 3 1/2 0

We see that the exponents depend on space dimensionality (connectivity)

What are the other factors that may change these exponents? Symmetries!
 Critical exponents
Ising model 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗

𝛼 𝛽′ 𝛾± 𝛿 𝜈± 𝜂 𝑠𝑖 = ±1
Exact solution d=2 ln(𝑇𝑐 − 𝑇) 1/8 7/4 15 1 1/4

Simulations d=3 0.110 0.326 1.237 4.800 0.630 0.036

Heisenberg d=3 -0.12 0.366 1.395 0.707 0.035
MFA (see TD) d4 0 1/2 1 3 1/2 0 𝑠Ԧ𝑖 ∈ 𝐑3
𝐸 = −𝐵 ෍ 𝑠Ԧ𝑖 − 𝐽 ෍ 𝑠Ԧ𝑖 ⋅ 𝑠Ԧ𝑗
𝑖 𝑖∼𝑗
We see that the exponents depend on space dimensionality (connectivity)
What are the other factors that may change these exponents? Symmetries!
 Critical exponents
Ising model 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗

𝛼 𝛽′ 𝛾± 𝛿 𝜈± 𝜂 𝑠𝑖 = ±1
Exact solution d=2 ln(𝑇𝑐 − 𝑇) 1/8 7/4 15 1 1/4
3-states Potts d=2 1/3 1/9 13/9 14 5/6 4/15 𝑠𝑖 = 1,2,3
Simulations d=3 0.110 0.326 1.237 4.800 0.630 0.036
Heisenberg d=3 -0.12 0.366 1.395 0.707 0.035
MFA (see TD) d4 0 1/2 1 3 1/2 0

𝐸 = −𝐵 ෍ 𝑠Ԧ𝑖 − 𝐽 ෍ 𝑠Ԧ𝑖 ⋅ 𝑠Ԧ𝑗

𝑖 𝑖∼𝑗
We see that the exponents depend on space dimensionality (connectivity)
What are the other factors that may change these exponents? Symmetries!

Critical exponents are (believed to be) universal, i.e. they do not depend on
the details of the physical system, but only on some of its general features
 Critical exponents
Ising model 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗

𝛼
Simple cubic
𝛽′ Face-centered
𝛾± 𝛿
cubic
𝜈± 𝜂
Magnetite Fe3O4
𝑠𝑖 = ±1
Exact solution d=2 ln(𝑇𝑐 − 𝑇) 1/8 7/4 15 1 1/4
3-states Potts d=2 1/3 1/9 13/9 14 5/6 4/15 𝑠𝑖 = 1,2,3
Simulations d=3 0.110 0.326 1.237 4.800 0.630 0.036
Heisenberg d=3 -0.12 0.366 1.395 0.707 0.035
MFA (see TD) d4 0 1/2 1 3 1/2 0
The same universality class𝐸 = −𝐵 ෍ 𝑠Ԧ − 𝐽 ෍ 𝑠Ԧ𝑖 ⋅ 𝑠Ԧ𝑗
𝑖
𝑖 𝑖∼𝑗
We see that the exponents depend on space dimensionality (connectivity)
What are the other factors that may change these exponents? Symmetries!

Critical exponents are (believed to be) universal, i.e. they do not depend on
the details of the physical system, but only on some of its general features
 Scaling relations
Are the exponents of a given model inter-related?

There are 4 independent scaling relations that determine all exponents from two
2−𝛼
For instance, 𝛼 and 𝛽 determine 𝛾 = 2 − 𝛼 − 2𝛽 𝛿= −1
𝛽
2 − 𝛼 − 2𝛽
𝜈 = (2 − 𝛼)/𝑑 𝜂 =2− 𝑑
2−𝛼
According to a pair 𝛼, 𝛽 , various models can be separated into universality classes
More rigorously, a universality class is a collection of mathematical models which
share a single scale invariant limit under the process of renormalization group flow

Attention! For MFA, the exponents are exact for any 𝑑 ≥ 4 so that the last two
relations are only valid at the critical dimension dc=4

G. Ódor, Universality classes in nonequilibrium lattice systems, Rev. Mod. Phys. 76, 663 (2004)
 Universality
Ferro/paramagnetic Liquid/gas transition
transition for spins {𝑠𝑖 } 𝑛𝑖 = (𝑠𝑖 + 1)/2

Magnetization 𝑚 𝑇, 0 ∝ 𝑇𝑐 − 𝑇 𝛽′ Difference in densities 𝜌𝑙 − 𝜌𝑔

Difference in pressures or
Magnetic field 𝑚 𝑇𝑐 , 𝐵 ∝ 𝐵1/𝛿 in chemical potentials
𝛾± Compressibility
Magnetic susceptibility 𝜒 ∝ 𝑇 − 𝑇𝑐
𝐶 𝑇 ∝ 𝑇𝑐 − 𝑇 −𝛼
Specific heat
−𝜈±
Correlation length 𝜉 𝑇 ∝ 𝑇 − 𝑇𝑐
𝑑−2+𝜂
Correlations at Tc 𝐶𝑖𝑗 (𝑇𝑐 ) ∝ 1/𝑟𝑖𝑗

More generally, one can speak of an “order parameter” (see lecture 3)

 Universality
Guggenheim’s experiment:
liquid-gaseous phase coexistence Critical point
curves for different compounds
𝛽′
𝑇 1/𝛽′ 𝛽′ ≈ 0.33
|𝜌 − 𝜌𝑐 | ∝ 𝑇𝑐 − 𝑇 ∝ 1 − 𝜌 − 𝜌𝑐 𝑇
𝑇𝑐 𝑇𝑐
𝑇𝑐 ,K
While these systems are microscopically different 44.8
150.7
from each other (monoatomic and diatomic 209.4
289.8
gases, and even CH4), their critical behaviors are ...
remarkably similar and universally described by 𝛽′ 3D Ising model: 𝛽′ ≈ 0.326
𝜌
Moreover, the same critical exponents 𝜌𝑐
appears in completely different systems, e.g.,
Heller and Benedek, Phys. Rev. Lett. 13, 253 (1962)
for the magnetization of MnFe2 and for the Thompson and Rice, JACS 86, 3547 (1964)
phase separation in a mixture of CCl4 and C7F16 See https://en.wikitolearn.org/Course:Statistical_Mechanics
E.A. Guggenheim, The Principle of Corresponding states, J. Chem. Phys. 13, 253-261 (1945)
 Origin of Universality
Microscopic system Coarse-grained picture

Physics at a microscopic
scale a is specific
𝑎 𝑎
If 𝜉 ≫ 𝑎 , these features
are averaged out
𝜉 𝜉
A unique effective large-
scale theory for various
systems

This is the starting point of the renormalization group approach

K. G. Wilson, Problems in Physics with Many Scales of Length, Scient. Am. (August 1979) , p. 158
 Origin of Universality
𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑗
Microscopic system 𝑖 𝑖∼𝑗

Physics at a microscopic
scale a is specific

If 𝜉 ≫ 𝑎 , these features 𝑠𝑖′ = sign ෍ 𝑠𝑗

are averaged out 𝑗∈𝐵𝑖

A unique effective large-

𝜉
scale theory for various
systems

This is the starting point of the renormalization group approach

L. P. Kadanoff, Scaling laws for Ising models near Tc, Physics Physique Fizika. 2 (6), 263 (1966).
 Origin of Universality
𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑗
Microscopic system 𝑖 𝑖∼𝑗

Physics at a microscopic 𝐸′ = −𝐵′ ෍ 𝑠𝑖′ − 𝐽′ ෍ 𝑠𝑖′ 𝑠𝑗′

𝑖 𝑖∼𝑗
scale a is specific

If 𝜉 ≫ 𝑎 , these features 𝑠𝑖′ = sign ෍ 𝑠𝑗

are averaged out 𝑗∈𝐵𝑖

A unique effective large- 𝐽′ 𝐵, 𝐽, 𝑎

𝜉
scale theory for various 𝐵′(𝐵, 𝐽, 𝑎)
systems
Repeat...
This is the starting point of the renormalization group approach

L. P. Kadanoff, Scaling laws for Ising models near Tc, Physics Physique Fizika. 2 (6), 263 (1966).
 Classification of phase transitions
Ehrenfest classification is based on free energy behavior
Free energy is continuous at the critical point: 𝐹1 𝑇𝑐 = 𝐹2 (𝑇𝑐 )

First-order transition Second-order transition n-order transition

𝐹′ is discontinuous 𝐹′′ is discontinuous 𝐹 (𝑛) is discontinuous

𝜕𝐹 𝜕(−𝑘𝐵 𝑇 ln 𝑍) 𝜕(−𝑘𝐵 𝑇 ln 𝑍) 𝐹 𝜕 ln 𝑍 𝐹 − 𝐸
= = −𝑘𝐵 ln 𝑍 − 𝑘𝐵 𝑇 = + = = −𝑆
𝜕𝑇 𝜕𝑇 1 𝑇 𝑇 𝜕𝛽 𝑇
𝜕
𝑘𝐵 𝛽
𝜕𝐹 𝜕(−𝑘𝐵 𝑇 ln 𝑍)
= = 〈𝑠1 + ⋯ + 𝑠𝑁 〉 = 𝑀
𝜕𝐵 𝜕𝐵
𝜕2𝐹 𝜕 𝐹− 𝐸 1 𝜕(𝐹 − 𝐸 ) 𝐹 − 𝐸 1𝜕 𝐸 𝐶(𝑇)
2
= = − 2
=− =−
𝜕𝑇 𝜕𝑇 𝑇 𝑇 𝜕𝑇 𝑇 𝑇 𝜕𝑇 𝑇
−𝛼
𝐶 𝑇 ∝ 𝑇𝑐 − 𝑇
Solid/liquid, liquid/gas Ferro/para, superconductivity, superfluidity
 Classification of phase transitions
Ehrenfest classification is based on free energy behavior
Free energy is continuous at the critical point: 𝐹1 𝑇𝑐 = 𝐹2 (𝑇𝑐 )

First-order transition Second-order transition n-order transition

𝐹′ is discontinuous 𝐹′′ is discontinuous 𝐹 (𝑛) is discontinuous

At the critical point, a system

either absorbs or releases a
fixed amount of energy

Δ𝐻vap = 40.7 kJ/mol

Δ𝐻fus = 6.01 kJ/mol
1 kg Ice ≈ 55.5 mol → 333 kJ
Solid/liquid, liquid/gas
~2 min in 2 kW boiler
 Classification of phase transitions
Ehrenfest classification is based on free energy behavior
Free energy is continuous at the critical point: 𝐹1 𝑇𝑐 = 𝐹2 (𝑇𝑐 )

First-order transition Second-order transition n-order transition

𝐹′ is discontinuous 𝐹′′ is discontinuous 𝐹 (𝑛) is discontinuous

At the critical point, a system

either absorbs or releases a
fixed amount of energy

coexistence of
three phases 2nd order
transition

Solid/liquid, liquid/gas
 Classification of phase transitions
Ehrenfest classification is based on free energy behavior
Free energy is continuous at the critical point: 𝐹1 𝑇𝑐 = 𝐹2 (𝑇𝑐 )

First-order transition Second-order transition n-order transition

𝐹′ is discontinuous 𝐹′′ is discontinuous 𝐹 (𝑛) is discontinuous

At the critical point, a system Continuous transitions are characterized by a divergent

either absorbs or releases a susceptibility, an infinite correlation length, and a
fixed amount of energy power law decay of correlations near criticality

But there are more sophisticated cases, e.g., liquid-glass

transition, which is a dynamic, out-of-equilibrium phenomenon

Solid/liquid, liquid/gas Ferro/para, superconductivity, superfluidity

 Validity of the MFA: Ginzburg’s criterion
𝜕𝑍 𝜕 ln 𝑍
The MFA neglects fluctuations between spins: 𝑠𝑖 𝑠𝑗 ≈ 〈𝑠𝑖 〉〈𝑠𝑗 〉 𝑚 = =
𝑍 𝛽𝜕𝐵 𝛽𝜕𝐵
To estimate the error of such approximation, 1
𝐶𝑖,𝑗 = 𝑠𝑖 𝑠𝑗 − 〈𝑠𝑖 〉〈𝑠𝑗 〉 〈𝑚〉 = 𝑁 ෍ 〈𝑠𝑖 〉
one needs to study (two-point) correlator 𝑖
exp(−𝛽𝐸)
Consider inhomogeneous field 𝐵𝑖 : 𝐸 = −𝐽 ෍ 𝑠𝑖 𝑠𝑗 − ෍ 𝑠𝑖 𝐵𝑖 𝑃( 𝑠1 , … 𝑠𝑁 ) =
𝑖∼𝑗 𝑖 𝑍(𝛽)
𝜕𝑍 𝜕 ln 𝑍 𝜕2𝑍 𝜕 2 ln 𝑍
𝑠𝑖 = = 〈𝑠𝑖 𝑠𝑗 〉 = 𝐶𝑖𝑗 = 2
𝑍 𝛽𝜕𝐵𝑖 𝛽𝜕𝐵𝑖 𝑍 𝛽2 𝜕𝐵𝑖 𝜕𝐵𝑗 𝛽 𝜕𝐵𝑖 𝜕𝐵𝑗
𝜕〈𝑀〉 𝜕〈𝑠𝑖 〉 For a translationally invariant system with 𝐵𝑗 = 𝐵 :
=෍ = ෍ 𝛽 𝐶𝑖𝑗
𝜕𝐵𝑗 𝑖 𝜕𝐵𝑗 𝑖 𝜕〈𝑀〉 𝜕〈𝑀〉 𝜕𝐵𝑗
𝜒= =෍ = ෍ 𝛽 𝐶𝑖𝑗
𝜕𝐵 𝑗 𝜕𝐵𝑗 𝜕𝐵 𝑖,𝑗
1
Fluctuation-dissipation theorem
 Validity of the MFA: Ginzburg’s criterion
The MFA neglects fluctuations between spins: 𝑠𝑖 𝑠𝑗 ≈ 〈𝑠𝑖 〉〈𝑠𝑗 〉 𝜒 = ෍ 𝛽 𝐶𝑖𝑗
𝑖,𝑗
To estimate the error of such approximation,
𝐶𝑖,𝑗 = 𝑠𝑖 𝑠𝑗 − 〈𝑠𝑖 〉〈𝑠𝑗 〉
one needs to study (two-point) correlator
Near the critical point, 𝑇 < 𝑇𝐶 , there are domains of correlated spins of size 𝜉(𝑇)
𝑠𝑖 𝑠𝑗 − 〈𝑠𝑖 〉〈𝑠𝑗 〉
What is the error for a pair of spins? 𝐸𝑖,𝑗 = ≪1
〈𝑠𝑖 〉〈𝑠𝑗 〉
σ𝑟𝑖,𝑗 <𝜉 𝑠𝑖 𝑠𝑗 − 〈𝑠𝑖 〉〈𝑠𝑗 〉
What is the error within such a domain? 𝐸𝑡𝑜𝑡 = ≪1
𝜒 −𝛾
σ 𝑟𝑖,𝑗 <𝜉 〈𝑠𝑖 〉〈𝑠𝑗 〉
σ𝑟𝑖𝑗 <𝜉( 𝑠𝑖 𝑠𝑗 − 𝑠𝑖 𝑠𝑗 ) ≈ ∝ 𝑇𝑐 − 𝑇
𝛽
𝑇𝑐 − 𝑇 −𝛾 ≪ 𝑇𝑐 − 𝑇 2𝛽−𝑑𝜈
෍ 𝑠𝑖 𝑠𝑗 ≈ 𝑚 2 𝑉𝑑 𝜉 ∝ 𝑇𝑐 − 𝑇 2𝛽 𝜉 𝑑 ∝ 𝑇𝑐 − 𝑇 2𝛽−𝑑𝜈 −𝛾−2𝛽+𝑑𝜈
𝑟𝑖𝑗 <𝜉 𝑇𝑐 − 𝑇 ≪1
𝛾 + 2𝛽 Ginzburg’s criterion
−𝛾 − 2𝛽 + 𝑑𝜈 > 0 𝑑 > 𝑑𝑐 =
𝜈 For Ising model: dc = 4
 Summary of the lecture
MFA is a very powerful technique to study phase transitions and critical phenomena

It yields exact results for Ising model in d4 and qualitative behavior in d=2 and 3

Near and at the critical point, most quantities exhibit power laws

Critical exponents are inter-related, only two are independent

Critical exponents allow one to attribute models/experiments into universality classes

Universality and self-similarity are crucial notions in statistical physics

 Mean field approximation (MFA)
Basic idea: an infinite system of a single particle in
interacting particles an effective medium

𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍
𝑖 𝑖∼𝑗
𝑠𝑖 𝑠𝑗 = − ෍ 𝑠𝑖 𝐵 + 𝐽 ෍
𝑖 𝑗 𝑖
𝑠𝑗 = − ෍ 𝑠𝑖 𝐵 + 𝐽 ෍
𝑖 𝑗 𝑖
𝑠𝑗 + (𝑠𝑗 − 𝑠𝑗 ) !
mean fluctuation
local field
1
෍ 𝑠𝑖 𝑠𝑗 = ෍ 𝑠𝑖 ෍ 𝑠𝑗
𝑖∼𝑗 2 𝑖 𝑗(𝑖)
Sum over all “links” Sum over neighbors of i

𝐸 = −𝐽 ෍ 𝑠𝑖 𝑠𝑗 = −𝐽 ෍ 𝑠𝑖 − 𝑠𝑖 𝑠𝑗 − 𝑠𝑗 + 𝑠𝑖 𝑠𝑗 + 𝑠𝑗 𝑠𝑖 − 〈𝑠𝑖 〉〈𝑠𝑗 〉
𝑖∼𝑗 𝑖∼𝑗
constant
𝐽 𝐽
≈− ෍ ෍ 𝑠𝑖 𝑠𝑗 + 𝑠𝑗 𝑠𝑖 − 𝑠𝑖 𝑠𝑗 = −෍ 𝑠𝑖 ℎ𝑖𝑚 + ෍ 〈𝑠𝑖 〉 ෍ 〈𝑠𝑗 〉
2 𝑖 𝑗 𝑖 𝑖 2 𝑖 𝑗(𝑖)