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

Bibliography
L. Landau & E. Lifshitz, Statistical Physics (Pergamon Press, 1958)
H. E. Stanley, Introduction to Phase Transitions and Critical Phenomena (Oxford University Press, 1971)
R. Feynman, Statistical Physics (CRC Press, 1972)
R. J. Baxter, Exactly Solved Models in Statistical Mechanics (1982)
G. Parisi, Statistical Field Theory (Addison-Wesley, 1988)
J. Zinn-Justin, Quantum field theory and critical phenomena (Oxford, Clarendon Press, 2002)
 Motivation

Solid/liquid/gas states of matter Magnetic materials

Liquid crystals

How minor variations of external

parameters can drastically change
the macroscopic state of the system?
Panic effects
 Motivation
Microscopic world Avogadro number Macroscopic world
𝑁𝐴 ≈ 6 ⋅ 1023 mol−1
1 mol of water = 18 gr
size of water molecule 1000
3 𝐴 = 3 ⋅ 10−10 m 1 liter of water = ⋅ 𝑁𝐴 ≈ 3 ⋅ 1025
18

Statistical physics provides efficient tools to describe

macroscopic systems from laws for microscopic constituents,
i.e., a passage from micro to macro scales

Local interactions can induce correlations that play the crucial role!
 Reminder
Microcanonical ensemble Canonical ensemble Grand canonical ensemble
Thermal bath Thermal bath with
with fixed T fixed T and 𝜇
𝛽 = 𝑘 1𝑇
𝐵
1
𝛽=𝑘
𝐵𝑇

Isolated system with Closed system with fixed Open system with fixed
fixed energy E and N mean energy 〈𝐸〉 and N mean energy 〈𝐸〉 and 〈𝑁〉
1
if 𝐸𝑚 =𝐸
exp(−𝛽𝐸𝑚 ) Boltzmann’s exp(−𝛽(𝐸𝑚 − 𝜇𝑁𝑚 ))
𝑃𝑚 = 𝑊(𝐸) 𝑃𝑚 = 𝑃𝑚 =
𝑍(𝛽) distribution 𝑍𝑔 (𝛽, 𝜇)
0, otherwise
−𝛽𝐸 𝑍 𝛽, 𝜇 = Σ exp −𝛽(𝐸 −𝜇𝑁 )
𝑆 = −𝑘𝐵 Σ𝑚 𝑃𝑚 ln(𝑃𝑚 ) = 𝑘𝐵 ln 𝑊 𝐸 𝑍 𝛽 = Σ𝑚 exp −𝛽𝐸 𝑚 = Σ𝐸 𝑊 𝐸 𝑒 𝑔 𝑚 𝑚 𝑚

𝑘𝐵 = 1.380649 ⋅ 10−23 𝐽/𝐾 𝐹 = −𝑘𝐵 𝑇 ln 𝑍 = 〈𝐸〉 − 𝑇𝑆 𝐴 = −𝑘𝐵 𝑇 ln 𝑍𝑔 = 〈𝐸〉 − 𝑇𝑆 − 𝜇𝑁

Entropy is maximized by 𝐸 = −𝜕 𝜕𝛽ln 𝑍
𝑆 = −𝜕𝐹𝜕𝑇
Grand potential is
this distribution Free energy is minimized by this law minimized by this law
 Ising model
Quantum mechanics describes microscopic interactions 𝐵
between two electrons of neighboring atoms

We consider a simplified model with N sites occupied by

atoms with spins taking only two values: 𝑠𝑖 = ±1
The space of microstates is composed of 2𝑁 configurations 𝑁𝐴 ≈ 6 ⋅ 1023 mol−1

The energy of a given configuration is 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑗 1 mol of Fe = 26 gr

exp(−𝛽𝐸) 𝑖 𝑖∼𝑗 1 kg Fe = 2.3 ⋅ 1025 atoms
Its probability is 𝑃( 𝑠1 , … 𝑠𝑁 ) = 𝑍(𝛽) Interactions with Inter-spin 2.3⋅1025 6.9⋅1024
magnetic field interactions 2 ∼ 10

𝑀 1
Measurable quantity is the magnetization: 𝑀 = ෍ 𝑠𝑖 𝑚 = = 𝑚 +𝑂
𝑖 𝑁 𝑁
It is “enough” to compute the partition function 𝑍 𝛽 =෍ exp(−𝛽𝐸)
{𝑠1 ,…,𝑠𝑁 }
 Competition between energy and entropy
At equilibrium, the system minimizes the free energy: 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑗
𝑖 𝑖∼𝑗
𝐹 = −𝑘𝐵 𝑇 ln 𝑍 = 〈𝐸〉 − 𝑇𝑆 exp(−𝛽𝐸)
𝑃( 𝑠1 , … 𝑠𝑁 ) =
𝑍(𝛽)
Let set B=0 and consider two limits of low and high temperature
At T=0, only two microstates contribute: +1, +1 … , +1 and {−1, −1, … − 1} Small 〈𝐸〉,
Ordered phase 𝑚 = ±1 small 𝑆
At 𝑇 → ∞ , the energy does not matter, and the maximal entropy minimizes F
𝑚 = 𝑁1 (𝑁+ − 𝑁− ) 𝑁
𝑁± = 2 (1 ± 𝑚) 𝑆 = 𝑘𝐵 ln 𝑊𝑚 = 𝑘𝐵 ln
𝑁 Large 〈𝐸〉,
𝑁 = 𝑁+ + 𝑁− 0.8 𝑁+ large 𝑆
𝑆 ≈ −𝑁 1+𝑚2
ln 1+𝑚
2
+ 1−𝑚
2
ln 1−𝑚
2 0.6

The entropy is maximal at 𝑚 = 0 Disordered phase 0.4

Competition between energy and entropy 0.2

𝑚
is at the origin of phase transitions 0
-1 -0.5 0 0.5 1
 Competition between energy and entropy
At equilibrium, the system minimizes the free energy: 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑗
𝑖 𝑖∼𝑗
𝐹 = −𝑘𝐵 𝑇 ln 𝑍 = 〈𝐸〉 − 𝑇𝑆 exp(−𝛽𝐸)
𝑃( 𝑠1 , … 𝑠𝑁 ) =
𝑍(𝛽)
Let set B=0 and consider two
Canlimits
exist of low and
a critical high temperature
temperature 𝑇𝑐 > 0 ?
Ordered
At T=0, only two microstates contribute: +1, +1 … , +1 and {−1, −1, … − 1}
phase Small 〈𝐸〉,
Ordered phase 𝑚 = ±1 small 𝑆
At 𝑇 → ∞ , the energy does not matter, and the maximal
Ordered entropy minimizes F
Disordered
𝑚 = 𝑁1 (𝑁+ − 𝑁− ) phase ?
𝑁
𝑁± = 2 (1 ± 𝑚)
phase
𝑆 = 𝑘𝐵 ln 𝑊𝑚 = 𝑘𝐵 ln
𝑁 Large 〈𝐸〉,
𝑁 = 𝑁+ + 𝑁−0 0.8 𝑁+ large 𝑆
𝑇
1+𝑚 1+𝑚 1−𝑚 1−𝑚 𝑇
𝑆 ≈ −𝑁 2 ln 2 + 2 ln 2
𝑐
0.6

The entropy is maximal 𝑚 = an

at exist
Does Disordered
0 ordered phase
phase for 0 ≤ 𝑇 ≤ 𝑇𝑐 ? 0.4

Competition between energy and entropy 0.2

𝑚
is at the origin of phase transitions 0
-1 -0.5 0 0.5 1
 Do phase transitions exist at a finite T?
Peierls argument (1936) in favor of phase transitions at a finite T 𝐸 = −𝐽 ෍ 𝑠𝑖 𝑠𝑗
𝑖∼𝑗
𝑇=0 𝑇>0

𝐿 = 12

Energy change: Δ𝐸 = 2𝐽𝐿 L is the number of pairs of neighboring sites (+,-),

i.e., the length of such an island
Entropy change: Δ𝑆 = 𝑘𝐵 ln 𝑊𝐿 How many configurations with the same L?
𝑊𝐿 ≈ 4 ⋅ 3 ⋅ … ⋅ 3 ≈ 3𝐿
𝑊𝐿 ≈ 𝑐 𝐿 2<𝑐<3
Free energy change: Δ𝐹 = 2𝐽𝐿 − 𝑇𝑘𝐵 𝐿 ln 𝑐 = 𝐿(2𝐽 − 𝑘𝐵 𝑇 ln 𝑐)
 Do phase transitions exist at a finite T?
Peierls argument (1936) in favor of phase transitions at a finite T 𝐸 = −𝐽 ෍ 𝑠𝑖 𝑠𝑗
𝑖∼𝑗
𝑇=0 𝑇>0

𝐿 = 12

Δ𝐹 = 2𝐽𝐿 − 𝑇𝑘𝐵 𝐿 ln 𝑐 = 𝐿(2𝐽 − 𝑘𝐵 𝑇 ln 𝑐)

Δ𝐹 > 0 Δ𝐹 = 0 Δ𝐹 < 0
0 𝑇
fluctuations are 2𝐽 fluctuations are
unfavorable 𝑇𝑐 = favorable
𝑘𝐵 ln 𝑐
Ordered phase Critical point: Disordered phase
(ferromagnetic) Possibility of arbitrarily (paramagnetic)
Kramers & Wannier (1941): large fluctuations with
𝑐 = 1 + 2 ≈ 2.414 no effect on F
 Do phase transitions exist at a finite T?
Peierls argument (1936) in favor of phase transitions at a finite T 𝐸 = −𝐽 ෍ 𝑠𝑖 𝑠𝑗
𝑖∼𝑗
𝑇=0 𝑇>0

𝐿 = 12

The argument holds in higher dimensions: L is replaced by Ld-1, and c depends on d

What about d=1? 𝑇=0 𝐸 = 𝐸𝐹 , 𝑀 = 𝑁
1 spin flipped 𝐸 = 𝐸𝐹 + 4𝐽, 𝑀 = 𝑁 − 2
N/2 spins flipped 𝐸 = 𝐸𝐹 + 4𝐽, 𝑀 = 0
Flipping a large number of spins does not cost energy. As a consequence, there is no
ordered phase at any T>0 No phase transition at a finite temperature in 1D
 Do phase transitions exist at a finite T?
Argument against phase transitions at a finite T in any dimension
1 1 Σ{𝑠1,…,𝑠𝑁 } 𝑠𝑖 exp(𝛽𝐽Σ𝑖∼𝑗 𝑠𝑖 𝑠𝑗 )
〈𝑚〉 = ෍ 〈𝑠𝑖 〉 = ෍ =0 by symmetry
𝑁 𝑖 𝑁 𝑖 Σ {𝑠1 ,…,𝑠𝑁 } exp(𝛽𝐽Σ 𝑠 𝑠
𝑖∼𝑗 𝑖 𝑗 )
In fact, each configuration 𝑠1 , … 𝑠𝑁 has the same contribution as −𝑠1 , … −𝑠𝑁
because the energy is invariant under the inversion of all spins (as B=0)
Thus, for any finite N, 〈𝑚〉 = 0 Disordered phase
𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑗
How to resolve this paradox? 𝑖 𝑖∼𝑗
Let us include a small magnetic field B>0 that breaks the symmetry 𝑠 → −𝑠
It gives the energy reduction −2𝐵𝑁𝑚 for the configuration with 𝑚 > 0
Even for small B, if BN remains large, the configuration with 𝑚 > 0 Symmetry breaking
is much more probable and thus there is no cancellation of two terms by minor fluctuations
1 1
Strictly speaking, 〈𝑚〉 = lim lim ෍ 〈𝑠𝑖 〉 whereas lim lim ෍ 〈𝑠𝑖 〉 = 0
𝐵→0 𝑁→∞ 𝑁 𝑖 𝑁→∞ 𝐵→0 𝑁 𝑖
 Exact solutions of the Ising model
𝑁 𝑁−1
1D case (Ising, 1925) 𝐸 = −𝐵 ෍ 𝑠𝑖 − 𝐽 ෍ 𝑠𝑖 𝑠𝑖+1
𝑠1 𝑠2 𝑠𝑁 𝑖=1 𝑖=1
𝐹
Exact solution 𝑓 = lim = −𝑘𝐵 𝑇 ln 𝑒 𝛽𝐽 cosh 𝛽𝐵 + 𝑒 2𝛽𝐽 sinh2 𝛽𝐵 + 𝑒 −2𝛽𝐽
𝑁→∞ 𝑁

At B=0: 𝑓 = −𝑘𝐵 𝑇 ln 𝑒 𝛽𝐽 + 𝑒 −𝛽𝐽 (see TD)

2D case (Onsager 1944)
2𝜋 2𝜋
1
At B=0: −𝛽𝑓 = ln 2 + 2 න 𝑑𝜃1 න 𝑑𝜃2 ln cosh2 2𝛽𝐽 −sinh 2𝛽𝐽 (cos 𝜃1 + cos 𝜃2 )
8𝜋 0 0
1
2𝐽 1 8
Announced by Onsager in 1948,
𝑇𝑐 = 〈𝑚〉 = 1 − proved by Yang in 1951
𝑘𝐵 ln(1 + 2) sinh4 2𝛽𝐽
No analytic solution is known for B0
One of the most challenging models in statistical physics.
3D case
It is believed that no analytic solution is possible...
B. McCoy & T. T. Wu, The Two-Dimensional Ising Model (HUP, 1973), 418 pages!
 Related problems 𝑈(𝑟) = 4𝜀 𝜎 12
𝑟 − 𝜎 6
𝑟
The ideal gas model ignores interactions between atoms 4
Lennard-Jones potential

strong
2
repulsion weak
attraction
Two neutral atoms A charge fluctuation The dipole induces a 0

creates a dipole dipole on the other atom

0 1 2 𝑟/𝜎 3 4
The Ising model is equivalent to a gas lattice model
Let 𝑛𝑖 molecules occupy the ith site of a lattice
Short-range repulsion = exclusion effect: 𝑛𝑖 = 0 or 1
Longer-range attraction: 𝐸attr = −𝜀 ෍ 𝑛𝑖 𝑛𝑗
𝑖∼𝑗
Site occupation cost: 𝐸cost = −𝜇 ෍ 𝑛𝑖
𝑖
𝑠𝑖 + 1 𝜀 𝑧𝜀 𝜇
Setting 𝑛𝑖 = yields 𝐸gas = − ෍ 𝑠𝑖 𝑠𝑗 − + ෍ 𝑠𝑖 + const
2 4 𝑖~𝑗 4 2 𝑖
𝑠𝑖 = ±1 𝐽 𝐵 The Ising model!
 Related problems
The Ising model is generic and finds applications beyond physics
Example: model of market agents (traders) or voting model
Consider N agents who take binary decisions (+1 and -1) at each time t,
e.g., 𝑠𝑖 𝑡 = +1 for buying and 𝑠𝑖 𝑡 = −1 selling an asset at t
Let the decision depend on the average opinion of their colleagues: ℎ = ෍ 𝑠𝑗 (𝑡 − 1)
𝑗(𝑖)
sign ℎ with probability 𝑝 = exp(𝛽 exp(𝛽 ℎ)
ℎ )+exp(−𝛽 ℎ )
Probabilistic choice: 𝑠𝑖 𝑡 =
−sign ℎ with probability 1−𝑝
Here 𝛽 quantifies the propensity of the agent: large 𝛽 𝑝≈1 〈𝑠𝑖 𝑡 〉 ≠ 0
small 𝛽 𝑝 ≈ 1/2 〈𝑠𝑖 𝑡 〉 ≈ 0
This dynamics converges to a Boltzmann’s steady state, in which the
probability of a given configuration {𝑠1 , … , 𝑠𝑁 } is 𝑍1 exp(𝛽Σ𝑖∼𝑗 𝑠𝑖 𝑠𝑗 )
There exists the critical value 𝛽𝑐 depending on the connection network
 Related problems
Initial condition: 𝑠𝑖 0 = 1
𝛽 = 0.7 1
𝛽 = 0.5 𝑚(𝑡) = ෍ 𝑠𝑖 (𝑡) average decision
𝑁 𝑖

𝛽𝑐 = Arctanh 2 − 1 ≈ 0.44

𝛽 > 𝛽𝑐 “ordered” phase when agents

𝛽 = 0.3 follow the average opinion
𝛽 = 0.1 Financial crisis (panic)!
𝛽 < 𝛽𝑐 “disordered” phase when
agents take decisions more
independently
(Simulation and image credits to G. Montambaux & R. Monasson) “normal” market conditions
 Generalizations of the Ising model
From lattices to arbitrary graphs (e.g., social networks or lattice quantum gravity)

𝑠𝑖 = ±1 𝑠𝑖 ∈ 𝑅𝑛 , 𝑠𝑖 = 1 n=2 XY model
n=3 Heisenberg model
𝐸 = −𝐵 ෍ 𝑠Ԧ𝑖 − 𝐽 ෍ 𝑠Ԧ𝑖 ⋅ 𝑠Ԧ𝑗
𝑖 𝑖∼𝑗

𝑠𝑖 ∈ {1, … , 𝑞} 1, if 𝑎 = 𝑏
Potts model 𝐸 = −𝐽 ෍ 𝛿(𝑠𝑖 , 𝑠𝑗 ) 𝛿(𝑎, 𝑏) =
𝑖∼𝑗 0, if 𝑎 ≠ 𝑏
A formal limit 𝑞 → 1 yields the results for percolation models
filling sites (or bonds) of a lattice
Independently with 𝑝
There is no interaction but
there is still a phase transition 𝑇→𝑝
R. J. Baxter, Exactly Solved Models in Statistical Mechanics (1982)
 Summary of the lecture
More is different1: (infinitely) large systems are (much) richer than their constituents

Competition between energy and entropy is the origin of phase transitions

The Ising model is a simple model of ferromagnets, liquids, and far beyond

The Ising model is exactly solvable in 1D and partly in 2D

Phase transition at a finite temperature exists for this model in dimensions d2

P. W. Anderson, More Is Different, Science 177, 393-396 (1972)