Mean-Field Theory and Its Critical Exponents

Mean-field theory is the simplest quantitative description of a continuous phase transition: replace the fluctuating environment of each degree of freedom by its average, expand the free energy in powers of the order parameter (Landau theory), and minimize. From that single algebraic exercise emerge a complete set of critical exponents — α=0, β=½, γ=1, δ=3, ν=½, η=0 — that are dimension-independent. This article views those exponents through the lens of critical phenomena: not as an isolated calculation, but as the exponents of a specific fixed point, the Gaussian one, which governs criticality above the upper critical dimension $d_c=4$ . The microscopic derivation of the mean-field exponents — the Weiss self-consistent field, the Landau expansion, the amplitude ratios — is owned by the companion article Mean-Field Theory and Its Critical Exponents (its full derivation of the set β=½, γ=1, δ=3, α=0, ν=½, η=0, the amplitude ratio $C_+/C_-=2$ , and the scaling relations is the reference); here we recall that result by name and ask a different question: when are these exponents right, and why are they wrong in low dimensions? The answer is the upper critical dimension. Mean-field theory neglects fluctuations of the order parameter. Whether that neglect is justified is decided by comparing the size of fluctuations inside a correlation volume to the order parameter itself — the Ginzburg criterion. For short-range O(n) systems this comparison singles out $d_c=4$ : above four dimensions fluctuations are subdominant, the Gaussian fixed point is stable, and the mean-field exponents are exact; below four they diverge faster than the mean and take over, driving the system to a non-trivial Wilson–Fisher fixed point with different exponents. This is why mean-field theory is simultaneously indispensable and quantitatively wrong. It is the correct leading description in high dimensions, the starting point for every fluctuation expansion (the ε=4−d expansion is built on it), and a spectacular failure in $d=2$ , where the exact Ising β=1/8 is four times smaller than the mean-field ½. Understanding the mean-field exponents as the exponents of the d≥4 fixed point is the bridge from Landau theory to the renormalization group.

Key takeaways

Mean-field (Landau) theory gives a complete, dimension-independent set of critical exponents: α=0, β=½, γ=1, δ=3, ν=½, η=0.
The microscopic derivation (Weiss field, Landau expansion, self-consistency, amplitude ratios) is owned by the companion article Mean-Field Theory and Its Critical Exponents; here they are recalled by name and re-framed as the exponents of the Gaussian fixed point.
These exponents are exact above the upper critical dimension $d_c=4$ for short-range O(n) systems, and increasingly wrong below it (2D Ising β=1/8 vs mean-field ½ — a factor of four).
Whether mean-field applies is decided by the Ginzburg criterion, which compares fluctuations in a correlation volume to the order parameter; the crossover defines $d_c$ .
Mean-field exponents satisfy the thermodynamic scaling relations (Rushbrooke, Widom) exactly but hyperscaling $2−α=dν$ only at $d=4$ — the tell of the upper critical dimension.
Key idea: the mean-field exponents α=0, β=½, γ=1, δ=3, ν=½, η=0 are the exponents of the Gaussian fixed point; they are exact for d≥d_c=4 and fail below it because order-parameter fluctuations, which mean-field neglects, become dominant.

The core insight

The mean-field exponents α=0, β=½, γ=1, δ=3, ν=½, η=0 are the exponents of the Gaussian fixed point. They are exact above the upper critical dimension $d_c=4$ (short-range O(n)), where order-parameter fluctuations are subdominant, and they fail below it, where fluctuations dominate and drive the system to the Wilson–Fisher fixed point.

Where this lives today

The upper critical dimension is not an abstraction: it decides which theory a real system obeys. Superconductors are close to mean-field because their long coherence length makes the critical region (the Ginzburg window) extraordinarily narrow — the specific-heat jump at $T_c$ is essentially the mean-field discontinuity, and fluctuation corrections are visible only in high- $T_c$ cuprates with short coherence lengths. Conversely, membrane and polymer criticality in effectively low dimensions is strongly non-mean-field. The concept also organizes modern field theory: any O(n) φ⁴ theory is "mean-field plus ε-expansion," and $d_c=4$ is why four-dimensional field theories are renormalizable and marginal. The companion article on the Ginzburg criterion develops the crossover in detail.

The physical picture

Mean-field theory silences the noise. Each spin, instead of consulting its few fluctuating neighbours, feels the population average — and the population average obeys a smooth, dimension-blind equation. That smoothness is the source of both the theory's clarity and its failure: by throwing away the geometry of correlated fluctuating regions, it throws away the only ingredient that could carry dimensional information. The punchline is the exponent comparison: the mean-field β=½ parabola sits far above the true low-dimensional curves.

Figure

Punchline: spontaneous magnetization m ~ (−t)^β near T_c. Mean-field β=½ (blue) lies far above the exact 2D Ising β=1/8 (red) and 3D Ising β≈0.326 (green) — mean-field overestimates the order

Why does the theory become exact at high dimension? Each spin on a $d$ -dimensional hypercubic lattice has $z=2d$ neighbours. As $d$ grows a spin couples to more and more others, and by the law of large numbers the field it feels has vanishing relative fluctuation — the average really is the whole story. Landau theory makes this transparent: expand $f=\tfrac12 at\,m^2+\tfrac14 b\,m^4-hm$ , minimize, and the singular behaviour follows from algebra that knows nothing about $d$ .

Figure

The Ginzburg criterion sets d_c=4: fluctuations of the order parameter in a correlation volume ξᵈ, relative to m², scale as |t|^{(d−4)/2}. For d>4 they vanish at T_c (mean-field valid); for d<4 they diverge (mean-field fails); d=4 is marginal

The breakdown is diagnosed precisely by comparing the size of order-parameter fluctuations inside a correlation volume to the order parameter itself. That comparison — the Ginzburg criterion — yields $|t|^{(d-4)/2}$ : it vanishes as $t\to0$ for $d>4$ (mean-field self-consistent), is marginal at $d=4$ , and diverges for $d<4$ (mean-field self-inconsistent). Four dimensions is the watershed.

The core equations

The mean-field exponents follow from minimizing the Landau free energy. We state the results here and recall their microscopic derivation from the companion article.

f(m)=\tfrac12 a\,t\,m^2+\tfrac14 b\,m^4-hm,\qquad t=\frac{T-T_c}{T_c},\quad a,b>0 \qquad (1)

where: $f$ — free-energy density expanded in the order parameter $m$ ; $a,b$ — positive phenomenological constants; $t$ — reduced temperature. Physical reading: the coefficient $at$ of $m^2$ changes sign at $T_c$ ; below it the minimum moves off zero. Minimizing at $h=0$ gives the order parameter and its exponent β; a second field derivative gives the susceptibility and γ.

m_0=\left(\frac{a|t|}{b}\right)^{1/2}\!\!\propto|t|^{1/2}\ (t<0)\ \Rightarrow\ \beta=\tfrac12,\qquad \chi\propto|t|^{-1}\ \Rightarrow\ \gamma=1 \qquad (2)

where: $m_0$ — spontaneous magnetization; χ — isothermal susceptibility. Physical reading: the parabolic minimum of the quartic well gives the square-root onset; the curvature at the minimum gives the inverse-linear susceptibility. On the critical isotherm and in the specific heat, the remaining exponents appear; the whole set is dimension-independent.

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

where: δ — critical isotherm $m\sim h^{1/3}$ ; ν — Ornstein–Zernike correlation length $\xi\sim|t|^{-1/2}$ ; η — anomalous dimension, zero for the Gaussian theory. Physical reading: these are the exponents of the Gaussian fixed point, exact for $d\ge4$ . The Gaussian-fixed-point eigenvalues make the $d$ -dependence explicit — the key to understanding why four dimensions is special.

y_t=2,\quad y_h=\frac{d+2}{2}\ \Rightarrow\ \gamma=1,\ \nu=\tfrac12,\ \eta=0\ (\text{all }d);\quad \alpha=2-\tfrac{d}{2},\ \beta=\tfrac{d}{4}-\tfrac12,\ \delta=\tfrac{d+2}{d-2} \qquad (4)

where: $y_t,y_h$ — thermal and magnetic RG eigenvalues at the Gaussian fixed point. Physical reading: γ, ν, η are $d$ -independent, but α, β, δ hit the classic mean-field values (0, ½, 3) exactly at $d=4$ — mean-field is literally the $d=4$ Gaussian theory.

Derivation cards

Derivation AMean-field exponents by minimizing the Landau free energy

The goal is to obtain β, γ, δ, α directly from the quartic Landau free energy, confirming the dimension-independent mean-field set. The full microscopic route — Weiss molecular field, single-site self-consistency $m=\tanh(\beta(Jzm+h))$ , and the Taylor expansion that produces the Landau coefficients — is derived at depth in the companion article Mean-Field Theory and Its Critical Exponents; this card carries out the minimization once, at the level of $f(m)$ , to fix the exponents this article re-frames.

Step 1. At $h=0$ , extremize: $\partial f/\partial m=atm+bm^3=0$ , so either $m=0$ (stable for $t>0$ ) or $m_0^2=-at/b=a|t|/b$ (stable for $t<0$ ). Hence $m_0\propto|t|^{1/2}$ and $\beta=\tfrac12$ . Why the sign flip matters: the quadratic coefficient $at$ passing through zero is what turns the single well ( $t>0$ ) into a double well ( $t<0$ ), the mechanism of the transition.

Step 2. Susceptibility: from $atm+bm^3=h$ , differentiate to get $\chi=\partial m/\partial h=1/(at+3bm^2)$ . Above $T_c$ ( $m=0$ ): $\chi=1/(at)\propto|t|^{-1}$ ; below ( $m^2=a|t|/b$ ): $\chi=1/(2a|t|)\propto|t|^{-1}$ . So $\gamma=1$ , with amplitude ratio $\Gamma_+/\Gamma_-=2$ .

Step 3. Critical isotherm ( $t=0$ ): $bm^3=h$ gives $m\sim h^{1/3}$ , so $\delta=3$ .

Step 4. Specific heat: the minimized free energy below $T_c$ is $f(m_0)=-a^2t^2/(4b)$ , so $\partial^2 f/\partial t^2=-a^2/(2b)$ is a finite jump (not a divergence). Hence α=0 (discontinuity), and ν=½, η=0 come from the Gaussian correlation function. Verified: `verify_aq-sm-mean-field-critical-exponents-sm7.py` reproduces $m_0^2=-at/b$ , $\chi$ above/below, $\Gamma_+/\Gamma_-=2$ , the $-a^2t^2/(4b)$ free energy, and the full exponent set symbolically.

Go deeper · Derivation B · 1 minThe upper critical dimension from the Ginzburg criterion

Derivation BThe upper critical dimension from the Ginzburg criterion

The goal is to derive $d_c=4$ — the dimension above which mean-field is exact — by comparing order-parameter fluctuations in a correlation volume to the order parameter squared. This is the self-consistency test of mean-field theory: the theory is trustworthy only when the fluctuations it neglected are indeed small. The card shows that the answer is set by dimensional counting alone, independent of microscopic detail.

Step 1. The order-parameter fluctuation in a correlation volume is $\langle\delta m^2\rangle_\xi\sim k_BT\,G(0\!\to\!\xi)/\xi^d$ , and using the Ornstein–Zernike correlation function integrated to ξ this scales as $\xi^{2-d}$ (for $d<4$ ), i.e. $\sim|t|^{-\nu(2-d)}=|t|^{(d-2)/2}$ with mean-field ν=½.

Step 2. The order parameter squared is $m_0^2\propto|t|^{2\beta}=|t|$ with mean-field β=½.

Step 3. Their ratio — the Ginzburg ratio — is $\langle\delta m^2\rangle_\xi/m_0^2\propto|t|^{(d-2)/2}/|t|=|t|^{(d-4)/2}$ . Why this is the criterion: mean-field is self-consistent only if the neglected fluctuations are small compared to the mean, i.e. if the ratio stays finite (or vanishes) as $t\to0$ .

Step 4. As $t\to0$ : for $d>4$ the ratio $\to0$ (fluctuations negligible, mean-field exact); for $d=4$ it is marginal (logarithmic corrections); for $d<4$ it diverges (fluctuations dominate, mean-field fails). Hence $\boxed{d_c=4}$ for short-range O(n) systems. Verified: the exponent $(d-4)/2$ and its sign at $d=3,4,5$ are checked in `verify_aq-sm-mean-field-critical-exponents-sm7.py`; below $d_c$ the true β (0.326 in 3D, 0.125 in 2D) is smaller than the mean-field ½, confirming mean-field overestimates the order.

Limiting cases

Case	Result	Meaning
$d>d_c=4$ (short-range O(n))	mean-field exponents exact	Fluctuations vanish at $T_c$ ; Gaussian fixed point stable; α=0, β=½, γ=1, δ=3, ν=½, η=0 hold exactly.
$d=d_c=4$	mean-field with log corrections	Marginal case; exponents are mean-field but modified by logarithms (e.g. $\chi\sim\|t\|^{-1}\|\ln\|t\|\|^{1/3}$ ).
3D Ising ( $d=3$ )	β≈0.326, γ≈1.237, ν≈0.630	Mean-field β=½ overestimates by ~50%; the Wilson–Fisher fixed point governs.
2D Ising ( $d=2$ )	β=1/8, γ=7/4, ν=1	Mean-field β=½ is four times too large — the most dramatic failure; fluctuations dominate.
Infinite-range / Curie–Weiss	mean-field exact for any $d$	Every spin couples to all others, so the average field has no fluctuation — mean-field is the exact solution.

Worked examples

Example 1 — Mean-field $T_c$ and exchange constant for iron. Mean-field predicts $k_BT_c=Jz$ . Iron is body-centred cubic with coordination $z=8$ and Curie temperature $T_c=1043$ K. Estimate the exchange constant.

Step 1. $J/k_B=T_c/z=1043/8=\mathbf{130.4\ K}$ , i.e. $J\approx1.80\times10^{-21}$ J ≈ 11.2 meV.

Step 2. This is the mean-field estimate; the true exchange (from spin-wave and neutron data) is smaller because mean-field overestimates $T_c$ — fluctuations reduce the ordering temperature below the mean-field value. The lesson: mean-field gives the scale of $J$ correctly (tens of meV) but the numerical factor is off by the same fluctuation physics that spoils the exponents.

Example 2 — How wrong is mean-field β in two dimensions? Compare the mean-field β to the exact 2D Ising value.

Step 1. Mean-field β=½=0.500; exact 2D Ising β=1/8=0.125 (Onsager/Yang).

Step 2. Ratio $\beta_{\rm MF}/\beta_{\rm exact}=0.5/0.125=\mathbf{4}$ . Mean-field overestimates the exponent by a factor of four, meaning it predicts the order parameter rising far too steeply below $T_c$ — a qualitative as well as quantitative failure, precisely because $d=2$ is well below $d_c=4$ and fluctuations dominate.

Origins · 1 minupper critical dimension and how this was discovered

upper critical dimension

concept — d_c=4 for short-range O(n) systems (Ginzburg 1960; Wilson–Fisher 1972) · The upper critical dimension is the spatial dimension above which mean-field critical exponents become exact because order-parameter fluctuations are subdominant. For short-range O(n) models it is four; below it the Wilson–Fisher fixed point governs non-classical exponents, computed in the ε=4−d expansion. The concept converts the vague "mean-field neglects fluctuations" into a sharp, dimension-counting criterion.

History

1907

— Weiss introduces the molecular (mean) field, the first mean-field theory of ferromagnetism

1937

— Landau expands the free energy in the order parameter, systematizing mean-field exponents

1944

— Onsager's exact 2D Ising solution gives β=1/8, contradicting mean-field ½

1960

— Ginzburg formulates the criterion for the validity of mean-field theory

1972

— Wilson and Fisher identify d_c=4 and compute exponents below it via the ε-expansion

1982

— Wilson receives the Nobel Prize for the renormalization group

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 12–13 · Kardar, Statistical Physics of Fields, Ch. 2–3 (Landau theory, Ginzburg criterion) · Goldenfeld, Lectures on Phase Transitions and the Renormalization Group, Ch. 5, 8 · Sethna, Statistical Mechanics, Ch. 12 · Stanley, Introduction to Phase Transitions and Critical Phenomena, Ch. 6, 12 · Ginzburg, Sov. Phys. Solid State 2, 1824 (1960); Wilson & Fisher, Phys. Rev. Lett. 28, 240 (1972).

Provenance. The mean-field exponent set and its microscopic derivation (Weiss field, Landau expansion, amplitude ratio $C_+/C_-=2$ ) are owned by the companion article Mean-Field Theory and Its Critical Exponents (STAT.INTER.MFCE.1258), whose served content states β=½, γ=1, δ=3, α=0, ν=½, η=0 and the scaling relations; this article recalls those results by name and re-derives only the minimization needed to fix them, then adds the Ginzburg-criterion route to $d_c=4$ . The Ginzburg ratio $|t|^{(d-4)/2}$ follows Kardar SPF Ch. 3 and Goldenfeld Ch. 5. Reference exponents (3D Ising β=0.326, 2D Ising β=1/8) as in the universality references. No nonstandard numbers introduced.

Builds on: Order Parameter · Continuous Phase Transitions · Landau Theory of Phase Transitions · Critical Exponents. Leads to: Scaling Hypothesis and Scaling Relations · Universality and Universality Classes · Correlation Function · Spontaneous Symmetry Breaking.

How is this article different from Mean-Field Theory and Its Critical Exponents?

That companion article (1258) owns the microscopic derivation: the Weiss molecular field, the self-consistency equation $m=\tanh(\beta(Jzm+h))$ , the Landau expansion from the single-site partition function, the full exponent set with two independent routes, the amplitude ratios, and all four scaling relations. This SM7 article recalls that result by name and asks the critical-phenomena question: these exponents are the exponents of the Gaussian fixed point, exact above $d_c=4$ and wrong below it. So the split is derivation (there) versus placement in the RG picture and the upper critical dimension (here).

Why is the specific-heat exponent α=0 rather than a divergence?

In mean-field theory the free energy below $T_c$ is $f=-a^2t^2/(4b)$ , a smooth quadratic in $t$ . Its second temperature derivative is a constant, $-a^2/(2b)$ , so the specific heat has a finite jump at $T_c$ (the difference between the ordered and disordered branches) rather than a power-law divergence. A finite discontinuity corresponds to α=0. This is different from the 2D Ising α=0, which is a logarithmic divergence — both are labelled α=0 but mean the "borderline" cases of different kinds (jump vs log).

Is mean-field ever exact in a real, finite-dimensional system?

Yes, in two situations. First, for infinite-range (Curie–Weiss) interactions where every spin couples equally to all others, mean-field is the exact solution in any dimension because the field a spin feels has no fluctuation. Second, in more than four dimensions for short-range interactions, where the Ginzburg criterion shows fluctuations vanish at $T_c$ . Real three-dimensional systems with short-range forces are never exactly mean-field, but some (like conventional superconductors) are extremely close because their critical region is vanishingly narrow.

Does mean-field get T_c right even if the exponents are wrong?

Only approximately. Mean-field predicts $k_BT_c=Jz$ , which captures the correct scale (set by the exchange energy and the number of neighbours) but overestimates the actual $T_c$ because fluctuations, which lower the ordering temperature, are neglected. For the 3D Ising model mean-field gives $k_BT_c/J=6$ (z=6) versus the true ≈4.51 — about 33% too high. So mean-field is a decent order-of-magnitude estimator of $T_c$ but a poor predictor of the critical behaviour near it.

What does the Ginzburg criterion actually compare?

It compares two things evaluated in a single correlation volume $\xi^d$ : the typical fluctuation of the order parameter, $\langle\delta m^2\rangle_\xi$ , and the mean order parameter squared, $m_0^2$ . Mean-field theory assumed the fluctuations are negligible; the criterion checks whether that is self-consistent. The ratio scales as $|t|^{(d-4)/2}$ , so it stays small (mean-field valid) for $d>4$ and blows up (mean-field invalid) for $d<4$ as you approach $T_c$ . The dimension where it is marginal, $d=4$ , is the upper critical dimension.

Why do superconductors look mean-field but magnets don't?

The width of the critical region — the range of $|t|$ over which fluctuations matter — is set by the Ginzburg number, which depends on the correlation-length amplitude. Conventional superconductors have a very long coherence length (hundreds of nanometres), making their correlation volume huge and their Ginzburg window extraordinarily narrow (as small as $|t|\sim10^{-14}$ ), so experiments essentially always see mean-field behaviour. Magnets have atomic-scale correlation lengths, a wide critical region, and therefore show the true non-mean-field exponents. Same physics, different amplitudes.