6.3 Critical exponents

Definition of critical exponents

Critical exponents describe how physical quantities diverge or vanish as a system approaches a continuous (second-order) phase transition. Near the critical point, observables follow power-law behavior controlled by a small set of exponents. The remarkable fact is that these exponents don't depend on microscopic details of the system. Instead, they depend only on the system's dimensionality and the symmetry of its order parameter, which is why very different physical systems can share the same critical exponents.

This property, called universality, means you can classify all continuous phase transitions into a small number of universality classes. A ferromagnet and a liquid-gas system can have identical critical exponents if they share the same dimension and order parameter symmetry.

Significance in phase transitions

• Critical exponents quantify how observables diverge (like susceptibility blowing up) or vanish (like the order parameter going to zero) as you approach the critical temperature $T_c$
• They reveal the underlying symmetries and dimensionality that govern the transition
• Because of universality, you can predict a system's critical behavior without knowing its microscopic Hamiltonian in full detail
• They provide the basis for classifying transitions into universality classes

Universality of critical exponents

Two systems belong to the same universality class if they share the same spatial dimensionality $d$ and order parameter symmetry (e.g., scalar, planar, or spherical). Within a class, all critical exponents are identical. This is why the 3D Ising model describes both uniaxial ferromagnets and liquid-gas transitions near their respective critical points.

Universality is not just a theoretical convenience. It means that studying a simple model (like the Ising model on a lattice) gives you quantitative predictions for real, complicated materials.

Types of critical exponents

Each critical exponent governs the power-law behavior of a specific observable as the reduced temperature $t = (T - T_c)/T_c \to 0$. Here are the standard six exponents.

Order parameter exponent $\beta$

The order parameter (magnetization $M$ in a magnet, density difference $\rho_l - \rho_g$ in a fluid) vanishes continuously as you approach $T_c$ from below:

$M \sim (-t)^{\beta}, \quad t \to 0^-$

The notation $\beta$ should not be confused with inverse temperature $1/k_BT$. For the 3D Ising class, $\beta \approx 0.326$.

Susceptibility exponent $\gamma$

The susceptibility $\chi$ (the response of the order parameter to its conjugate field) diverges on both sides of $T_c$:

$\chi \sim |t|^{-\gamma}$

This divergence reflects the enormous fluctuations in the order parameter near criticality. For the 3D Ising class, $\gamma \approx 1.237$.

Correlation length exponent $\nu$

The correlation length $\xi$, which sets the spatial scale over which fluctuations are correlated, diverges as:

$\xi \sim |t|^{-\nu}$

At $T_c$ itself, $\xi \to \infty$, meaning the system has no characteristic length scale. This divergence underlies the scale invariance that makes renormalization group methods so effective. For the 3D Ising class, $\nu \approx 0.630$.

Specific heat exponent $\alpha$

The specific heat $C$ near $T_c$ behaves as:

$C \sim |t|^{-\alpha}$

The exponent $\alpha$ can be positive (true divergence), zero (logarithmic divergence, as in the 2D Ising model), or negative (a finite cusp rather than a divergence). For the 3D Ising class, $\alpha \approx 0.110$.

Additional exponents: $\delta$ and $\eta$

Two more exponents complete the standard set:

• $\delta$ governs the critical isotherm, the relationship between the order parameter and the conjugate field $h$ exactly at $T_c$: $h \sim |M|^{\delta} \text{sgn}(M)$
• $\eta$ is the anomalous dimension, describing how the correlation function decays at criticality: $G(r) \sim r^{-(d-2+\eta)}$ at $T = T_c$

Scaling relations

The six standard exponents ($\alpha, \beta, \gamma, \delta, \nu, \eta$) are not all independent. They are connected by exact scaling relations, so only two are truly independent (for a given universality class).

Widom scaling

Widom's scaling hypothesis states that the singular part of the free energy is a generalized homogeneous function of its arguments. This leads to the relation:

$\gamma = \beta(\delta - 1)$

It also implies that the equation of state near $T_c$ can be written in terms of a single universal scaling function, collapsing data from different temperatures onto one curve.

Rushbrooke relation

This connects the thermal exponents:

$\alpha + 2\beta + \gamma = 2$

Originally derived as an inequality ($\geq 2$) from thermodynamic stability, it becomes an exact equality when the scaling hypothesis holds. It serves as a useful consistency check: if you measure $\alpha$, $\beta$, and $\gamma$ independently and they don't satisfy this relation, something is off.

Fisher's relation

This links the correlation exponents to the susceptibility exponent:

$\gamma = \nu(2 - \eta)$

It connects the divergence of susceptibility to the spatial structure of correlations. Since $\eta$ is typically small (e.g., $\eta \approx 0.036$ for 3D Ising), $\gamma \approx 2\nu$ is often a reasonable first approximation.

Josephson (hyperscaling) relation

One more relation worth knowing ties the exponents to the spatial dimension $d$:

$2 - \alpha = d\nu$

This is called a hyperscaling relation because it explicitly involves $d$. It holds below the upper critical dimension $d_c$ ($d_c = 4$ for standard $\phi^4$ theory). Above $d_c$, mean field exponents take over and hyperscaling breaks down.

Calculation methods

Mean field theory

Mean field theory replaces fluctuating interactions with an effective average field acting on each particle. It's the simplest approach and gives a universal set of exponents:

$\alpha = 0, \quad \beta = frac{1}{2}, \quad \gamma = 1, \quad \delta = 3, \quad \nu = \frac{1}{2}, \quad \eta = 0$

These values are exact for dimensions $d \geq 4$ (the upper critical dimension). For $d < 4$, mean field theory neglects critical fluctuations and gives quantitatively wrong exponents. The Landau theory of phase transitions is the prototypical mean field approach.

Renormalization group approach

The renormalization group (RG) is the most powerful framework for computing critical exponents. The core idea:

1. Start with a microscopic Hamiltonian defined at some short-distance cutoff
2. Integrate out (coarse-grain) the shortest-wavelength degrees of freedom
3. Rescale lengths to restore the original cutoff
4. Repeat, tracking how coupling constants flow under this transformation
5. Critical behavior is controlled by the fixed points of this flow; critical exponents are determined by the eigenvalues of the linearized RG transformation around the fixed point

The $\epsilon$-expansion (Wilson and Fisher) sets $\epsilon = 4 - d$ and computes exponents as a perturbative series in $\epsilon$. For $d = 3$, $\epsilon = 1$, and the series must be resummed carefully, but the results agree well with experiments and simulations.

Monte Carlo simulations

Monte Carlo methods generate statistical samples of configurations using algorithms like Metropolis or Wolff cluster updates. To extract critical exponents:

1. Simulate the system at several temperatures near $T_c$ for multiple system sizes $L$
2. Measure observables (magnetization, susceptibility, etc.) as functions of $T$ and $L$
3. Apply finite-size scaling: near $T_c$, observables depend on the ratio $L/\xi$, so data for different $L$ collapse onto universal curves when plotted against $L^{1/\nu}t$
4. Extract exponents from the collapse or from the $L$-dependence of observables at $T_c$

This approach can handle systems where analytic methods are intractable and provides high-precision numerical estimates of critical exponents.

Experimental measurements

Scattering techniques

Scattering experiments (X-ray, neutron, light) directly probe the correlation function $G(r)$ through its Fourier transform, the structure factor $S(q)$. Near $T_c$, the Ornstein-Zernike form gives:

$S(q) \sim \frac{1}{q^{2-\eta}(\xi^{-2} + q^2)}$

From the $q$-dependence you can extract $\eta$, and from the temperature dependence of the peak width you get $\nu$. Small-angle neutron scattering (SANS) is particularly useful for measuring correlation lengths in magnetic and polymer systems. Critical opalescence in fluids is literally visible-light scattering from diverging density fluctuations.

Thermodynamic measurements

Bulk measurements of specific heat, susceptibility, and compressibility give $\alpha$, $\gamma$, and related exponents directly. Precise temperature control is essential because the power-law regime is often narrow ($|t| < 10^{-2}$ or less). Techniques include:

• AC and adiabatic calorimetry for specific heat ($\alpha$)
• SQUID magnetometry for magnetic susceptibility ($\gamma$)
• PVT measurements near the liquid-gas critical point for compressibility

Extracting clean exponents requires careful treatment of corrections to scaling and background contributions.

Critical exponents in different systems

Magnetic systems

Magnetic systems are the classic testing ground for critical exponent theory. The universality class depends on the spin dimensionality $n$:

• $n = 1$ (Ising): Uniaxial ferromagnets like $\text{FeF}_2$. Discrete up/down symmetry.
• $n = 2$ (XY): Planar magnets, superfluid helium. Continuous $U(1)$ symmetry.
• $n = 3$ (Heisenberg): Isotropic ferromagnets like EuO. Full $O(3)$ rotational symmetry.

Each class has distinct exponents. For example, $\beta$ increases from about 0.326 (Ising) to 0.345 (XY) to 0.366 (Heisenberg) in 3D.

Liquid-gas transitions

The liquid-gas critical point (e.g., water at 647 K and 22.1 MPa) belongs to the 3D Ising universality class. The order parameter is the density difference $\rho_l - \rho_g$, and the conjugate field is the chemical potential deviation from its critical value. Critical opalescence, where the fluid becomes milky white, occurs because density fluctuations at all length scales scatter visible light.

Percolation phenomena

Percolation is a geometric phase transition: as you randomly occupy sites (or bonds) on a lattice with probability $p$, a spanning cluster first appears at the percolation threshold $p_c$. Near $p_c$, quantities like the cluster size distribution and the probability of belonging to the infinite cluster follow power laws with their own set of critical exponents. These exponents depend on $d$ but are distinct from thermal (Ising/XY/Heisenberg) exponents because percolation has no energy or temperature.

Universality classes

Ising model class ($n = 1$)

Systems with a scalar order parameter and discrete $\mathbb{Z}_2$ symmetry. Includes uniaxial magnets, binary alloys (order-disorder transitions), and the liquid-gas critical point. In 3D:

$\beta \approx 0.326, \quad \gamma \approx 1.237, \quad \nu \approx 0.630, \quad \alpha \approx 0.110$

The 2D Ising model is exactly solvable (Onsager), giving $\beta = 1/8$, $\gamma = 7/4$, $\nu = 1$, $\alpha = 0$ (logarithmic divergence).

XY model class ($n = 2$)

Systems with a two-component order parameter and $U(1)$ (planar rotational) symmetry. Relevant for superfluid helium, planar magnets, and superconducting films. In 2D, the XY model does not have a conventional phase transition but instead exhibits the Kosterlitz-Thouless (KT) transition, where correlations change from exponential to algebraic decay. The KT transition is driven by the unbinding of topological defects (vortex-antivortex pairs) and has no standard power-law exponents in the usual sense.

In 3D, the XY class has a conventional critical point with exponents distinct from Ising, e.g., $\nu \approx 0.672$.

Heisenberg model class ($n = 3$)

Systems with a three-component order parameter and full $O(3)$ rotational symmetry. Applies to isotropic ferromagnets and antiferromagnets. In 3D: $\beta \approx 0.366$, $\nu \approx 0.711$. The higher symmetry of the order parameter leads to stronger fluctuation effects compared to the Ising case.

Critical phenomena beyond equilibrium

Dynamic critical exponents

Near $T_c$, the relaxation time $\tau$ of the system diverges as:

$\tau \sim \xi^z \sim |t|^{-z\nu}$

where $z$ is the dynamic critical exponent. This divergence is called critical slowing down: the system takes longer and longer to equilibrate as it approaches the critical point. The value of $z$ depends not only on the universality class but also on the dynamics (whether the order parameter is conserved or not). For the 3D Ising model with non-conserved dynamics (Model A), $z \approx 2.02$.

Non-equilibrium phase transitions

Systems driven out of equilibrium can exhibit their own continuous phase transitions with distinct critical exponents. Directed percolation is the most studied non-equilibrium universality class, describing transitions between absorbing and active states (relevant to epidemic spreading, catalytic reactions). Self-organized criticality describes systems that tune themselves to a critical point without external parameter adjustment (sandpile models). These non-equilibrium transitions generally belong to different universality classes than their equilibrium counterparts.

Applications of critical exponents

Materials science

Critical exponents guide the understanding of phase transitions in technologically important materials. In superconductors, the transition to the superconducting state near $T_c$ involves critical fluctuations described by XY-class exponents (for conventional superconductors). In multiferroics and complex oxides, competing order parameters can lead to multicritical points with their own exponent values. Knowledge of the universality class helps predict material behavior without exhaustive characterization.

Biological systems

Concepts from critical phenomena appear in biological contexts. Liquid-liquid phase separation in cells (formation of membraneless organelles) involves critical-point-like behavior. Some researchers have proposed that neural networks in the brain operate near a critical point, where the dynamic range and information processing capacity are maximized. These applications are active research areas where the universality framework is being tested and extended.

Financial markets

Statistical physicists have applied power-law analysis to financial data, finding that stock price fluctuations and market crash precursors sometimes exhibit scaling behavior reminiscent of critical phenomena. These analogies are suggestive but should be treated with caution: financial systems lack the well-defined Hamiltonians and equilibrium conditions that make universality rigorous in physics.

Limitations and challenges

Finite-size effects

Real systems and simulations are always finite, so the correlation length $\xi$ can never truly diverge. When $\xi$ becomes comparable to the system size $L$, the observed behavior deviates from the infinite-system power laws. Finite-size scaling theory handles this systematically: observables are expressed as functions of $L/\xi$, and data from multiple system sizes are used to extrapolate to $L \to \infty$. Without this analysis, measured exponents can be significantly off.

Crossover phenomena

When a system has competing interactions or symmetries, it may show behavior characteristic of one universality class far from $T_c$ and cross over to another class very close to $T_c$. For example, a weakly anisotropic magnet might look Heisenberg-like at moderate $|t|$ but cross over to Ising behavior at very small $|t|$. The crossover region complicates the extraction of true asymptotic exponents and requires careful analysis, often using crossover scaling functions, to disentangle the contributions.