Mean-Field Critical Exponents

Approach a ferromagnet's Curie point, or a fluid's liquid–gas critical point, and every response function forgets its microscopic origins and follows a power law. The magnetization vanishes as $m\sim(-t)^{\beta}$ , the susceptibility diverges as $\chi\sim|t|^{-\gamma}$ , the specific heat behaves as $C\sim|t|^{-\alpha}$ — all controlled by the reduced temperature $t=(T-T_c)/T_c$ and a small set of numbers, the critical exponents. These exponents are universal: they depend only on dimensionality and the symmetry of the order parameter, not on whether you are looking at iron or carbon dioxide. Mean-field theory is the simplest approximation that produces a genuine phase transition, and it makes a sharp prediction for every exponent. Whether you expand the van der Waals equation about its critical point, solve the Weiss self-consistency equation for a magnet, or minimize a Landau free-energy polynomial, you get the same rational numbers: $\beta=\tfrac12$ , $\gamma=1$ , $\delta=3$ , $\alpha=0$ (a finite jump, not a divergence), and — with a gradient term — $\nu=\tfrac12$ , $\eta=0$ . That three unrelated-looking theories agree is the first glimpse of universality, and this article owns the derivation of that classical exponent set. The prediction is also spectacularly, instructively wrong. In two dimensions the exact Ising value is $\beta=\tfrac18$ , not $\tfrac12$ ; in three dimensions $\beta\approx0.326$ . The failure is not random: mean-field theory neglects the spatial correlations of fluctuations, and the Ginzburg criterion shows those fluctuations dominate below the upper critical dimension $d_c=4$ . Understanding exactly what mean-field predicts — and precisely where and why it breaks — is the necessary first step toward the renormalization group, which finally computes the true exponents. We derive the classical exponents, verify the scaling relations they satisfy, and mark the boundary of their validity.

Key takeaways

Near a continuous transition, thermodynamic quantities follow power laws in the reduced temperature $t=(T-T_c)/T_c$ set by the critical exponents $\alpha,\beta,\gamma,\delta,\nu,\eta$ .
Mean-field theory predicts the classical exponents $\alpha=0$ (finite jump), $\beta=\tfrac12$ , $\gamma=1$ , $\delta=3$ , $\nu=\tfrac12$ , $\eta=0$ — the same values from van der Waals, Weiss, and Landau theory.
These follow mechanically from minimizing an analytic Landau polynomial $f=a\,t\,m^2+b\,m^4-h\,m$ , or equivalently from expanding the $\tanh$ self-consistency equation of the Weiss magnet.
The mean-field exponents obey the scaling relations (Rushbrooke, Widom, Fisher) with equality, and hyperscaling ( $2-\alpha=\nu d$ ) fixes the upper critical dimension $d_c=4$ .
Below four dimensions the classical values are wrong (2D Ising $\beta=\tfrac18$ , 3D Ising $\beta\approx0.326$ ); fluctuations, quantified by the Ginzburg criterion, take over.
Key idea: treating the free energy as an analytic polynomial in the order parameter fixes the critical exponents to universal rational numbers ( $\beta=\tfrac12,\gamma=1,\delta=3,\alpha=0,\nu=\tfrac12,\eta=0$ ) shared by every mean-field theory — exact for $d\ge4$ , and corrected by fluctuations below.

The core insight

Treating the free energy as an analytic polynomial in the order parameter fixes the critical exponents to universal rational numbers ( $\beta=\tfrac12,\gamma=1,\delta=3,\alpha=0,\nu=\tfrac12,\eta=0$ ) shared by van der Waals, Weiss, and Landau theories — exact at and above the upper critical dimension $d_c=4$ , and corrected by fluctuations below it.

Where this lives today

Critical exponents remain a working diagnostic across condensed-matter and soft-matter physics. Conventional (BCS) superconductors have coherence lengths of thousands of ångströms, which makes their Ginzburg window astronomically narrow ( $t_G\sim10^{-12}$ or smaller); mean-field Ginzburg–Landau exponents describe those transitions essentially perfectly, and critical fluctuations are unobservable. High-temperature cuprate superconductors, with short coherence lengths and high $T_c$ , show a measurably wide critical region with non-mean-field 3D-XY exponents — the fluctuation regime made visible. Simple fluids near their liquid–gas critical point show $\beta\approx0.326$ , and precision light-scattering and coexistence-curve measurements on xenon, carbon dioxide, and ${}^3$ He– ${}^4$ He mixtures agree with 3D-Ising values to several decimal places, one of the sharpest confirmations of universality anywhere in physics. The deeper legacy is conceptual. Because mean-field exponents are exact for $d\ge4$ , they anchor the $\epsilon=4-d$ expansion: the renormalization group computes corrections as power series in $\epsilon$ , for example $\gamma=1+\epsilon/6+\mathcal{O}(\epsilon^2)$ for the Ising universality class, with the classical value as the $\epsilon\to0$ starting point. Modern conformal-bootstrap calculations pin the 3D-Ising exponents to extraordinary precision, but they are still understood relative to the mean-field baseline derived here. Every discussion of universality classes — Ising, XY, Heisenberg — begins by asking how far the true exponents sit from the classical values, and why.

The physical picture

The whole of mean-field criticality lives in a single loop: each degree of freedom feels not its fluctuating neighbours but the spatially averaged field they generate, and that average field must be consistent with the magnetization it produces. Near $T_c$ this self-consistency loop sits at a marginally stable fixed point. Just below $T_c$ , a nonzero magnetization appears, and how fast it grows as you cool — the power $\beta$ — is dictated by the leading nonlinearity of the loop. Because the loop is analytic (a $\tanh$ , or a quartic polynomial), the onset is always a square root: $m\sim(-t)^{1/2}$ . There is nothing about iron or argon in that statement; it is a property of smooth functions with a quadratic touch, and it is why van der Waals, Weiss, and Landau all agree. The first figure shows the punchline exponent visually. Plot the order parameter against temperature: mean-field predicts a parabola-on-its-side, $m\propto(-t)^{1/2}$ , meeting zero at $T_c$ with a vertical tangent. The true 3D-Ising curve is steeper near $T_c$ — $\beta\approx\tfrac13$ rises faster than $\beta=\tfrac12$ — so mean-field overshoots, predicting more order than nature delivers just below the critical point. That single gap between the two curves is the visible signature of the fluctuations mean-field throws away.

Figure

Order parameter m versus temperature: the mean-field β = ½ curve (blue, m ∝ (−t)^½) versus the real 3D-Ising β ≈ 0.33 curve (red); both vanish at T_c, but near the transition the real curve rises faster (steeper), so mean field overshoots the true order

Landau's second insight makes the universality inevitable rather than coincidental. You do not need the microscopic Hamiltonian at all: symmetry alone forbids odd powers of an Ising-like order parameter in the free energy, so the leading expansion must be $f = a\,t\,m^2 + b\,m^4$ . Everything — the exponents, the amplitude ratios — falls out of minimizing this polynomial. The exponents are called classical precisely because they follow from treating the free energy as a smooth, analytic function, which is exactly the assumption that fails when critical fluctuations make it genuinely non-analytic. The second figure contrasts the exponent values across universality classes as a bar chart. Mean-field ( $\beta=\tfrac12,\gamma=1,\delta=3,\nu=\tfrac12$ ) sits beside the exact 2D Ising ( $\tfrac18, \tfrac74, 15, 1$ ) and best 3D Ising ( $0.326, 1.237, 4.79, 0.630$ ) values. The classical numbers are the odd ones out below four dimensions — the same "universal" role, but the wrong "universal" numbers.

Figure

Critical exponents β, γ, δ, ν across three universality classes: mean-field (blue), 2D Ising exact (amber), 3D Ising (red). Mean-field gives fixed rational values that disagree with both lower-dimensional cases — same structure, wrong numbers below d = 4

Why four dimensions? The neglected fluctuations contribute a relative correction that scales as $|t|^{(d-4)/2}$ near $T_c$ . For $d>4$ it vanishes and mean-field is self-consistent; for $d<4$ it blows up, and a different fixed point — the Wilson–Fisher fixed point of the renormalization group — governs the true critical behaviour. The classical exponents are not a failed guess; they are the exact answer in a world of four or more dimensions, and the organized starting point for computing the corrections in ours.

The core equations

The classical exponents are a fixed set of numbers, valid for every mean-field theory. It is worth stating them together, because their values — not the microscopic model — are what universality is about.

\alpha=0,\qquad \beta=\tfrac12,\qquad \gamma=1,\qquad \delta=3,\qquad \nu=\tfrac12,\qquad \eta=0 \qquad (1)

where: $\alpha$ governs the specific heat $C\sim|t|^{-\alpha}$ (here a finite jump); $\beta$ the order parameter $m\sim(-t)^{\beta}$ for $t<0$ ; $\gamma$ the susceptibility $\chi\sim|t|^{-\gamma}$ ; $\delta$ the critical isotherm $h\sim|m|^{\delta}\mathrm{sgn}(m)$ at $t=0$ ; $\nu$ the correlation length $\xi\sim|t|^{-\nu}$ ; $\eta$ the anomalous dimension in $G(r)\sim r^{-(d-2+\eta)}$ at $T_c$ . These six numbers are not independent — four scaling relations tie them together. Everything follows from a single object: the Landau free-energy density, an analytic polynomial in the order parameter whose only inputs are the $m\to-m$ symmetry and analyticity in $T$ .

f(m,t)=f_0(t)+a\,t\,m^2+b\,m^4-h\,m,\qquad a,b>0 \qquad (2)

where: $f$ is the free-energy density; $m$ the order parameter; $t=(T-T_c)/T_c$ the reduced temperature; $h$ the conjugate ordering field; $a,b$ phenomenological coefficients analytic and positive near $T_c$ . The $m\to-m$ symmetry at $h=0$ kills odd powers; analyticity forces the quadratic coefficient to vanish linearly, $\propto t$ , which is the entire origin of the transition — below $T_c$ that coefficient turns negative and $m=0$ ceases to be the minimum. Minimizing $f$ at $h=0$ gives the equilibrium order parameter, and its square-root onset is the exponent $\beta$ .

\left.\frac{\partial f}{\partial m}\right|_{h=0}=2a\,t\,m+4b\,m^3=0\ \Rightarrow\ m=\pm\sqrt{\frac{-a\,t}{2b}}\sim(-t)^{1/2},\quad \beta=\tfrac12 \qquad (3)

where: $m$ is the value minimizing $f$ . For $t>0$ only $m=0$ solves it; for $t<0$ the two symmetric nonzero roots appear, growing as the square root of the distance below $T_c$ — the classical $\beta=\tfrac12$ . The inverse susceptibility is the curvature of $f$ at the minimum; it vanishes linearly on both sides of $T_c$ , giving $\gamma=1$ with a factor-of-two amplitude asymmetry.

\chi^{-1}=\frac{\partial^2 f}{\partial m^2}=2a\,t+12b\,m^2=\begin{cases}2a\,t, & t>0\ (m=0)\\[2pt]-4a\,t, & t<0\ (m^2=-at/2b)\end{cases},\quad \gamma=1 \qquad (4)

where: $\chi=(\partial m/\partial h)_T$ is the isothermal susceptibility. Above $T_c$ the curvature at $m=0$ is $2at$ ; below $T_c$ , substituting the nonzero $m^2$ doubles it in magnitude and flips the sign, giving $\chi\sim1/(4a|t|)$ . Both diverge as $|t|^{-1}$ (so $\gamma=1$ ), but the amplitudes differ by exactly $2$ — the universal mean-field ratio $C_+/C_-=2$ . At the critical temperature the quadratic term is gone, so the field couples to the cube of the order parameter — this is the critical isotherm exponent $\delta$ .

h=\left.\frac{\partial f}{\partial m}\right|_{t=0}=4b\,m^3\ \Rightarrow\ m\sim h^{1/3},\quad \delta=3 \qquad (5)

where: $h$ is the field on the critical isotherm $t=0$ . With the linear-in- $m$ restoring term absent at $T_c$ , the leading response is cubic, so the magnetization grows as the cube root of the field — $\delta=3$ . Finally, restoring the spatial gradient term (the Ornstein–Zernike form) turns the free energy into a field theory whose Gaussian fluctuations fix the two correlation exponents.

G(q)=\frac{k_BT}{K q^2+2a\,t},\qquad \xi=\sqrt{\frac{K}{2a\,t}}\sim t^{-1/2}\ (\nu=\tfrac12),\qquad G(q)\big|_{t=0}\sim q^{-2}\ (\eta=0) \qquad (6)

where: $K$ is the stiffness of the square-gradient term and $G(q)$ the Fourier-space correlation function. The Lorentzian propagator's width sets the correlation length $\xi\sim t^{-1/2}$ , giving $\nu=\tfrac12$ ; at $t=0$ the pure $q^{-2}$ form inverse-transforms to $G(r)\sim r^{-(d-2)}$ , i.e. $\eta=0$ . The six exponents are locked together by the scaling relations below.

relation	statement	mean-field check
Rushbrooke	$\alpha+2\beta+\gamma=2$	$0+1+1=2$ ✓
Widom	$\gamma=\beta(\delta-1)$	$1=\tfrac12(3-1)$ ✓
Fisher	$\gamma=\nu(2-\eta)$	$1=\tfrac12\cdot2$ ✓
Josephson (hyperscaling)	$2-\alpha=\nu d$	$2=\tfrac12\,d\Rightarrow d_c=4$

Derivation ADeriving β, γ, and δ from the mean-field self-consistency equation

The goal is to extract the three thermodynamic exponents $\beta=\tfrac12$ , $\gamma=1$ , $\delta=3$ directly from the Weiss self-consistency equation $m=\tanh[\beta(zJm+h)]$ , without invoking the Landau polynomial, so that the reader sees the same exponents emerge from the microscopic $\tanh$ and understands they are a property of any analytic self-consistency loop with a quadratic touch at $T_c$ . The strategy is to expand the hyperbolic tangent to cubic order near $T_c$ , where its argument is small, and read off each exponent from the appropriate limit: $h=0$ below $T_c$ gives $\beta$ , the linear response gives $\gamma$ , and the $t=0$ isotherm gives $\delta$ . The point to watch is that at exactly $T_c$ the linear term cancels — that cancellation is what promotes the cubic term to leading order and fixes $\delta=3$ .

Step 1. Set up the self-consistency and expand. Each spin feels the effective field $h_{\text{eff}}=h+zJm$ , so $m=\tanh[\beta(zJm+h)]$ with $k_BT_c=zJ$ the mean-field transition temperature. Near $T_c$ the argument is small, so use $\tanh x = x-\tfrac13 x^3+\cdots$ : writing $x=(T_c m + h/k_B)/T$ , we get $m \approx \tfrac{T_c}{T}m + \tfrac{h}{k_BT} - \tfrac13\big(\tfrac{T_c m}{T}\big)^3$ . Why: the transition happens where the linear coefficient $T_c/T$ crosses $1$ , i.e. at $T=T_c$ ; the cubic term is the first nonlinearity that stabilizes a finite $m$ .

Step 2. Order parameter at $h=0$ gives $\beta=\tfrac12$ . Set $h=0$ and cancel one factor of $m$ : $1=\tfrac{T_c}{T}-\tfrac13\tfrac{T_c^3}{T^3}m^2$ , so $m^2=3\dfrac{T^2(T_c-T)}{T_c^3}$ . As $T\to T_c$ this is $m^2\to3(T_c-T)/T_c=3(-t)$ , hence $m\sim(-t)^{1/2}$ and $\beta=\tfrac12$ . Why: dividing by $m$ removes the trivial $m=0$ root; the surviving balance is quadratic in $m$ , so the onset is a square root — the generic signature of a smooth loop losing stability.

Step 3. Susceptibility above $T_c$ gives $\gamma=1$ . Keep $h$ small, linearize ( $m\to0$ ): $m=\tfrac{T_c}{T}m+\tfrac{h}{k_BT}$ , so $m(1-T_c/T)=h/k_BT$ and $\chi=\partial m/\partial h=\dfrac{1/k_B}{T-T_c}=\dfrac{C}{T-T_c}$ — the Curie–Weiss law. Thus $\chi\sim(T-T_c)^{-1}$ and $\gamma=1$ . Why: the vanishing of $1-T_c/T$ at $T_c$ is exactly the diverging susceptibility; the same linear vanishing that drives $\beta$ drives $\gamma$ .

Step 4. Critical isotherm gives $\delta=3$ . At $T=T_c$ the linear term is $\tfrac{T_c}{T}m=m$ , which cancels the left side, leaving $0=\tfrac{h}{k_BT_c}-\tfrac13 m^3$ , i.e. $h=\tfrac{k_BT_c}{3}m^3$ , so $h\sim m^3$ and $\delta=3$ . Why: the exact cancellation of the linear response at $T_c$ is what promotes the cubic term to the leading field–magnetization relation — the hallmark of a critical point. *Verified: the $\tanh$ expansion $x-x^3/3$ , the onset $m^2\to3(-t)$ ( $\beta=\tfrac12$ ), the Curie–Weiss $\chi=C/(T-T_c)$ ( $\gamma=1$ ), and $h=(k_BT_c/3)m^3$ ( $\delta=3$ ) are all checked symbolically in verify_mean-field-critical-exponents.py, alongside the independent Landau-polynomial route.*

Go deeper · Derivation B · 2 minThe specific-heat jump α = 0, the amplitude ratio, and the scaling relations

Derivation BThe specific-heat jump α = 0, the amplitude ratio, and the scaling relations

Here the goal is to complete the exponent set by computing the specific heat — showing it has a finite discontinuity rather than a divergence, so $\alpha=0$ — and then to verify that the classical exponents satisfy the scaling relations with equality, which in turn fixes the upper critical dimension through hyperscaling. The route is to evaluate the Landau free energy at its minimum below $T_c$ , differentiate twice with respect to temperature to get the specific heat, and compare above and below; then substitute the exponent values into Rushbrooke, Widom, Fisher, and Josephson. What to watch is that a finite jump is the $\alpha\to0$ limit of $|t|^{-\alpha}$ : zero power means no singularity, only a step — and that hyperscaling alone is dimension-dependent, which is why it singles out $d=4$ .

Step 1. Free energy at the minimum below $T_c$ . Substitute $m^2=-at/2b$ into $f=at\,m^2+b\,m^4$ : $f_{\min}=at(-at/2b)+b(at/2b)^2=-\dfrac{a^2t^2}{4b}$ . Above $T_c$ the minimum is $m=0$ with $f=0$ (no singular part). Why: the singular part of the free energy lives entirely below $T_c$ ; its $t^2$ form is the seed of the finite specific-heat jump.

Step 2. Specific heat and the jump. With $C=-T\,\partial^2 f/\partial T^2$ and $t=(T-T_c)/T_c$ , differentiating $f_{\min}=-a^2t^2/4b$ twice in $t$ gives $\partial^2 f_{\min}/\partial t^2=-a^2/2b$ , a constant. So below $T_c$ the singular specific heat is a constant and above it is zero: the discontinuity is $\Delta C=\dfrac{a^2}{2b\,T_c}>0$ , finite. Hence $C\sim|t|^{-\alpha}$ with $\alpha=0$ . Why: a $t^2$ free energy gives a $t^0$ (constant) second derivative — a step, not a spike; $\alpha=0$ is the statement "jump, not divergence."

Step 3. Scaling relations with equality. Insert $\alpha=0,\beta=\tfrac12,\gamma=1,\delta=3,\nu=\tfrac12,\eta=0$ . Rushbrooke $\alpha+2\beta+\gamma=0+1+1=2$ ✓; Widom $\gamma=\beta(\delta-1)=\tfrac12(2)=1$ ✓; Fisher $\gamma=\nu(2-\eta)=\tfrac12(2)=1$ ✓. Why: these three relations hold in any dimension because they involve no explicit $d$ ; mean-field satisfies them exactly, as any consistent scaling theory must.

Step 4. Hyperscaling fixes $d_c=4$ . Josephson's relation $2-\alpha=\nu d$ carries an explicit dimension: $2-0=\tfrac12\,d$ forces $d=4$ . For $d>4$ hyperscaling fails with mean-field exponents (they are $d$ -independent, hyperscaling is not), signalling that above four dimensions the free energy is dominated by a dangerously irrelevant variable and $d_c=4$ is the boundary. Why: only hyperscaling knows about dimension, so it is the relation that pins where mean-field can be exact — precisely at and above $d=4$ , where fluctuations stop mattering. *Verified: $f_{\min}=-a^2t^2/4b$ , the finite jump $\Delta C=a^2/(2bT_c)$ ( $\alpha=0$ ), the amplitude ratio $C_+/C_-=2$ , and all four scaling relations (with $d_c=4$ from Josephson) are checked symbolically in verify_mean-field-critical-exponents.py; the 2D-Ising exact values are shown to satisfy Rushbrooke too.*

Limiting cases

Mean-field theory becomes exact in several controlled limits and fails in a sharply characterized way otherwise. The table compares the classical exponents to the exact 2D and best-known 3D Ising values.

Case	Result	Meaning
Mean field ( $d\ge4$ )	$\beta=\tfrac12,\gamma=1,\delta=3,\nu=\tfrac12,\alpha=0$	Classical exponents; exact at and above the upper critical dimension (with log corrections exactly at $d=4$ ).
2D Ising (exact)	$\beta=\tfrac18,\gamma=\tfrac74,\delta=15,\nu=1,\alpha=0$	Onsager/Yang exact solution; wildly different from mean field, yet still obeys Rushbrooke and Widom.
3D Ising (best)	$\beta\approx0.326,\gamma\approx1.237,\delta\approx4.79,\nu\approx0.630,\alpha\approx0.110$	Fluctuation-corrected; the values real magnets and fluids actually show.
Infinite range ( $J_{ij}=J/N$ )	mean field exact for all $d$	Each spin literally sees the average of all others (Curie–Weiss model), so there is nothing for fluctuations to do.
Ginzburg window ( $d<4$ )	$\|t\|\lesssim t_G$	Within this reduced-temperature window fluctuations dominate and classical exponents fail; outside it (far from $T_c$ ) mean field can still fit data.

Common trapMean-field exponents are not universal truths — they are exact only for four or more dimensions

It is tempting to read " $\beta=\tfrac12$ is universal" as " $\beta=\tfrac12$ is right." Universality means the exponents depend only on dimension and symmetry, not on microscopic details — but mean-field theory computes the wrong universal numbers below $d_c=4$ , because it discards the spatial correlations of fluctuations. Two further traps: the specific-heat exponent $\alpha=0$ denotes a finite jump, not a logarithm or divergence, so do not report a diverging mean-field specific heat. And when computing the susceptibility below $T_c$ , you must expand about the correct branch ( $m\neq0$ , not $m=0$ ) or you will miss the factor-of-two amplitude ratio $C_+/C_-=2$ — itself a universal mean-field prediction. Getting the branch wrong is the single most common error in this calculation.

Worked examples

1 · Verifying the scaling relations and the specific-heat jump (confidence builder)

We check that the classical exponents satisfy the scaling relations with equality and compute the size of the mean-field specific-heat discontinuity.

Step 1. List the classical values: $\alpha=0$ , $\beta=\tfrac12$ , $\gamma=1$ , $\delta=3$ , $\nu=\tfrac12$ , $\eta=0$ . Rushbrooke requires $\alpha+2\beta+\gamma=2$ : compute $0+2(\tfrac12)+1=0+1+1=\mathbf{2}$ ✓.

Step 2. Widom requires $\gamma=\beta(\delta-1)$ : compute $\beta(\delta-1)=\tfrac12(3-1)=\tfrac12\cdot2=\mathbf{1}=\gamma$ ✓. Fisher requires $\gamma=\nu(2-\eta)=\tfrac12(2-0)=\mathbf{1}=\gamma$ ✓.

Step 3. Josephson (hyperscaling) requires $2-\alpha=\nu d$ : with $\alpha=0,\nu=\tfrac12$ this is $2=\tfrac12 d$ , so $d=\mathbf{4}$ — the upper critical dimension drops straight out.

Step 4. The specific-heat jump from the Landau free energy is $\Delta C=\dfrac{a^2}{2b\,T_c}$ . For an Ising model at the mean-field level this evaluates to $\Delta C=\tfrac32 N k_B$ per spin-block in the standard normalization — a finite step of order $N k_B$ , confirming $\alpha=0$ : the specific heat jumps but never diverges.

2 · Coexistence-curve exponent for a van der Waals fluid (exam level)

We show the liquid–gas coexistence curve of a van der Waals fluid obeys $\beta=\tfrac12$ , the same exponent as a magnet, by expanding the reduced equation of state.

Step 1. In reduced variables $p=P/P_c$ , $v=V/V_c$ , $\tau=T/T_c$ , the van der Waals equation is $(p+3/v^2)(3v-1)=8\tau$ . Set $\pi=p-1$ , $\omega=v-1$ , $t=\tau-1$ and expand to third order in $\omega$ at fixed $t$ : $\pi = 4t - 6t\,\omega - \tfrac32\omega^3 + \cdots$ .

Step 2. On the critical isotherm $t=0$ this is $\pi=-\tfrac32\omega^3$ , so pressure deviation varies as the cube of volume deviation — the same $\delta=3$ as the magnet, with $\omega$ playing the role of order parameter.

Step 3. Below $T_c$ the Maxwell equal-area construction on the cubic, together with the $\omega\to-\omega$ symmetry of the leading terms, gives symmetric roots $\omega_{\pm}=\pm2\sqrt{-t}$ . The width of the coexistence region is $\Delta v = v_{\text{gas}}-v_{\text{liq}}=\omega_+-\omega_-=\mathbf{4\sqrt{-t}}$ .

Step 4. Since $\Delta v\propto(-t)^{1/2}$ , the coexistence-curve exponent is $\beta=\tfrac12$ — identical to the Weiss ferromagnet. This is universality in action: an interacting fluid and a magnet, sharing nothing microscopically, share the same classical exponent because both reduce to minimizing an analytic order-parameter free energy. (Real fluids show $\beta\approx0.326$ , flatter coexistence curves than this parabola — the measurable fingerprint of fluctuations.)

Origins · 1 mincritical exponents and how this was discovered

critical exponents

the universal power-law exponents governing continuous phase transitions — classical values from van der Waals (1873), Weiss (1907) and Landau (1937); corrected by Onsager (1944) and Wilson (1971) · β, γ, δ, ν, and the scaling relations that bind them

The critical exponents are the small set of numbers that describe how thermodynamic quantities diverge or vanish as a continuous phase transition is approached. Their story is a century of successive refinement. Van der Waals's 1873 equation of state contained, unrecognized, the classical exponents $\beta=\tfrac12$ , $\gamma=1$ , $\delta=3$ ; Pierre Weiss's 1907 molecular-field theory of ferromagnetism reproduced the same values from a completely different physical picture; and Lev Landau's 1937 order-parameter expansion revealed why they coincide — all three are analytic free-energy polynomials with the same algebraic structure. That three unrelated theories agree was the first evidence of universality: exponents depend only on dimensionality and symmetry. The classical values were long believed exact, until Lars Onsager's 1944 exact solution of the two-dimensional Ising model gave $\beta=\tfrac18$ , not $\tfrac12$ , and shattered the belief. Understanding why mean-field fails — the Ginzburg criterion's identification of the upper critical dimension $d_c=4$ , below which fluctuations dominate — set the stage for Kenneth Wilson's renormalization group (1971, Nobel 1982), which finally computed the true exponents as a systematic $\epsilon=4-d$ expansion anchored on the classical mean-field values. The classical exponents thus play a double role: the wrong answer in three dimensions, and the exactly correct zeroth order of the theory that gets it right.

History

1873

— van der Waals's equation of state implicitly contains the classical exponents β = ½, γ = 1, δ = 3.

1895

— Pierre Curie measures the ferromagnetic-to-paramagnetic transition, later described by the Curie–Weiss law.

1907

— Weiss's molecular-field theory reproduces the classical exponents from a self-consistent internal field.

1937

— Landau introduces the order-parameter free-energy expansion, exposing why mean-field theories share exponents.

1944

— Onsager's exact 2D Ising solution gives β = 1/8, disproving the universality of the classical values.

1965

— Widom formulates the scaling hypothesis relating the exponents.

1966

— Kadanoff's block-spin picture explains scaling through coarse-graining.

1971

— Wilson's renormalization group computes the true exponents via the ε = 4 − d expansion.

1982

— Wilson receives the Nobel Prize for the renormalization-group theory of critical phenomena.

Going further

Texts. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 12 (Phase Transitions: Criticality). Kardar, Statistical Physics of Fields, Ch. 2–3. Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 12. Goldenfeld, Lectures on Phase Transitions and the Renormalization Group, Ch. 5. Landau & Lifshitz, Statistical Physics Part 1, §142–148.

Provenance. Base article: AllQFT track `sm`, "Mean-Field Critical Exponents" (STAT.INTER.MFCE.1258); source Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 12 (Phase Transitions: Criticality). All exponents and scaling relations re-verified here in verify_mean-field-critical-exponents.py: the Landau minimization ( $\beta=\tfrac12$ ), the two-sided susceptibility with amplitude ratio $C_+/C_-=2$ ( $\gamma=1$ ), the critical isotherm ( $\delta=3$ ), the finite specific-heat jump ( $\alpha=0$ ), the Ornstein–Zernike correlation exponents ( $\nu=\tfrac12,\eta=0$ ), the four scaling relations (Rushbrooke, Widom, Fisher, Josephson $\Rightarrow d_c=4$ ), the independent Weiss $\tanh$ -route derivation of $\beta,\gamma,\delta$ , and the exact 2D-Ising comparison values. No base physics error found; the base's exponent set, amplitude ratio, and vdW/Weiss/Ginzburg derivations are reproduced. This article is the SM6 owner of the mean-field exponent derivation; the SM7 article "Mean-Field Theory and Its Critical Exponents" recalls it in the critical-phenomena context.

Prerequisites. Van der Waals Equation of State · Maxwell Construction and Coexistence · Virial Expansion of the Equation of State.

Leads to. Ginzburg Criterion · Renormalization Group · Landau Theory of Phase Transitions.

Why does the van der Waals fluid give the same exponents as a magnet? They have nothing in common physically.

Physically nothing; algebraically everything. Both reduce, near their critical point, to minimizing an analytic free energy that is a polynomial in an order parameter — magnetization $m$ for the magnet, density (or volume) deviation $\omega$ for the fluid — with a quadratic coefficient that vanishes linearly through $T_c$ and a positive quartic term for stability. That single algebraic structure fixes $\beta=\tfrac12$ , $\gamma=1$ , $\delta=3$ , $\alpha=0$ , irrespective of what the order parameter physically is. This is precisely the content of universality: the exponents care only about the symmetry of the order parameter and the spatial dimension, not the microscopic constituents. It is the first and clearest hint that critical behaviour is governed by something far more general than any specific Hamiltonian.

Is α = 0 the same as saying the specific heat is constant near the critical point?

No — $\alpha=0$ means there is no power-law divergence, but the specific heat still changes: it has a finite discontinuity (a jump) at $T_c$ . In mean-field theory the singular part of the free energy goes as $t^2$ below $T_c$ and vanishes above, so its second temperature derivative is a constant below and zero above, leaving a finite step $\Delta C=a^2/(2bT_c)$ . The convention $C\sim|t|^{-\alpha}$ with $\alpha=0$ captures "no divergence"; a genuine jump is the $\alpha\to0$ limit. This is different from, say, the 2D Ising model, which also has $\alpha=0$ but as a logarithmic divergence — same exponent, different functional form, a subtlety the exponent alone does not resolve.

Where does the factor of two in the susceptibility amplitude ratio come from?

From evaluating the curvature of the free energy on the correct branch. Above $T_c$ the minimum is at $m=0$ , so $\chi^{-1}=\partial^2 f/\partial m^2=2at$ . Below $T_c$ the minimum is at $m^2=-at/2b\neq0$ , and substituting this nonzero value gives $\chi^{-1}=2at+12b m^2=2at+12b(-at/2b)=-4at$ . The magnitude below is twice the magnitude above, so the susceptibility amplitude is half as large below $T_c$ — the ratio $C_+/C_-=2$ . It is a universal mean-field prediction beyond the exponent itself, and forgetting to switch branches below $T_c$ (expanding about $m=0$ everywhere) is exactly how people miss it.

What actually is the "upper critical dimension" and why is it four?

The upper critical dimension $d_c$ is the spatial dimension above which mean-field theory is exact because fluctuations become negligible near $T_c$ . For a standard Ising-like order parameter, the Ginzburg criterion compares the fluctuation of the order parameter integrated over one correlation volume to the mean-field order parameter squared; the ratio scales as $|t|^{(d-4)/2}$ . For $d>4$ it vanishes as $t\to0$ — fluctuations die away and mean-field self-consistency holds; for $d<4$ it diverges and the classical exponents fail. The crossover is exactly $d=4$ , which is also what the hyperscaling relation $2-\alpha=\nu d$ returns when you plug in the mean-field values. Different universality classes (with different symmetries or long-range interactions) can have different $d_c$ , but for short-range Ising, XY, and Heisenberg models it is four.

If the exponents are universal, why do 2D and 3D Ising give different numbers?

Universality says the exponents depend on dimension and symmetry — and $2$ D and $3$ D are different dimensions, so they are allowed to differ, and do: $\beta=\tfrac18$ in 2D versus $\approx0.326$ in 3D. What universality forbids is dependence on microscopic detail within a fixed dimension and symmetry class: a square-lattice Ising magnet and a triangular-lattice one in 2D share $\beta=\tfrac18$ , even though their lattice structures and $T_c$ values differ. Mean-field theory is the special case that ignores dimension entirely (it gives the same $\tfrac12$ for every $d$ ), which is precisely why it is only correct at and above $d_c=4$ , where dimension genuinely stops mattering for the exponents.

Do the mean-field exponents ever describe a real experiment accurately?

Yes, in two situations. First, in systems close to or above four effective dimensions, or with long-range interactions that mimic high dimensionality — the infinite-range Curie–Weiss model is exactly mean-field for all $d$ . Second, and more practically, in systems with a very narrow Ginzburg window: conventional BCS superconductors have coherence lengths so large that the fluctuation region is unmeasurably small ( $t_G\sim10^{-12}$ ), so Ginzburg–Landau (mean-field) exponents describe the superconducting transition essentially perfectly. Even in systems that ultimately show non-classical exponents, mean-field often fits the data away from $T_c$ , outside the Ginzburg window; the classical values fail only in the immediate critical neighbourhood where fluctuations take over.