Fluctuations Beyond Mean Field: The Ginzburg Criterion and Ginzburg-Landau Theory

Intuition Beginner

Mean-field theory works by averaging everything out. It treats each region of the system as if it has the same value of the order parameter (magnetisation, density, etc.) as every other region. But real systems fluctuate. Near a critical point, those fluctuations become enormous — they grow in both magnitude and spatial extent.

Think of a magnet near its critical temperature $T_c$. In mean-field theory, every small region has exactly the average magnetisation $m$. In reality, each small patch can have a magnetisation that differs from $m$ by a fluctuation. Far from $T_c$, these fluctuations are tiny and mean-field theory is an excellent approximation. But as you approach $T_c$, two things happen simultaneously:

1. The fluctuations grow in magnitude — patches deviate more and more from the average.
2. The fluctuations grow in spatial extent — correlated regions expand to enormous size (this is the correlation length $\xi$ diverging).

At some point close enough to $T_c$, the fluctuations overwhelm the average. Mean-field theory says the magnetisation is $m\sim \sqrt{T_c-T}$, but the fluctuation in $m$ within a correlated patch can be larger than $m$ itself. When the fluctuation is larger than the signal, the theory that ignored fluctuations is in trouble.

The Ginzburg criterion is the quantitative test for when this happens. It compares the energy in fluctuations to the mean-field energy. If fluctuations are small, mean-field theory is reliable. If fluctuations are large, mean-field theory breaks down and you need a more sophisticated approach (the renormalization group).

The Ginzburg-Landau theory extends mean-field theory by allowing the order parameter to vary in space. Instead of a single number $m$, you have a field $m(\mathbf{x})$ that can fluctuate from point to point. The theory assigns an energy cost to these spatial variations through a gradient term $\kappa |\text{grad}m{|}^2$ — smooth variations are cheap, sharp variations are expensive. This single ingredient is enough to define a correlation length, compute fluctuation magnitudes, and derive the Ginzburg criterion.

The punchline: mean-field theory is self-consistent (and hence gives correct critical exponents) only in spatial dimensions $d>4$. For $d=3$ (the physical world), mean-field theory always fails sufficiently close to $T_c$. The region where it fails — the Ginzburg temperature window — can be vanishingly narrow for some materials (like conventional superconductors, where it is microkelvins wide) or wide for others (like liquid helium, where it is several kelvin). This explains why mean-field theory works well for some phase transitions and fails catastrophically for others.

Visual Beginner

The correlation length $\xi$ sets the size of correlated patches of the order parameter. Far from $T_c$, $\xi$ is microscopic (a few lattice spacings) and fluctuations are small. As $T\to T_c$, $\xi$ diverges and correlated patches grow to macroscopic size.

Worked example Beginner

Consider a 3D Ising magnet near $T_c$. Mean-field theory predicts the magnetisation $m\approx \sqrt{3(T_c-T)/T_c}$ for $T$ slightly below $T_c$. The correlation length is $\xi \approx a/\sqrt{|T_c-T|/T_c}$ where $a$ is the lattice spacing.

At $T=0.99T_c$ (1% below $T_c$): $m\approx \sqrt{3\times 0.01}\approx 0.17$. The correlation length is $\xi \approx 10a$. Fluctuations within a correlated volume (${\xi }^3\approx 1000a^3$, containing about 1000 spins) have magnitude $\delta m\sim 1/\sqrt{1000}\approx 0.03$. Since $0.03\ll 0.17$, fluctuations are small and mean-field theory is reliable.

At $T=0.9999T_c$ (0.01% below $T_c$): $m\approx \sqrt{3\times 0.0001}\approx 0.017$. The correlation length is $\xi \approx 100a$. A correlated volume contains about $10^6$ spins, so $\delta m\sim 1/\sqrt{10^6}=0.001$. Now $\delta m/m=0.001/0.017\approx 0.06$ — still small, but growing.

In 3D, the fluctuation-to-signal ratio grows as $T\to T_c$, but it grows slowly. For most materials, the Ginzburg temperature window is extremely narrow, which is why mean-field theory works surprisingly well for practical purposes even though it is fundamentally incorrect in three dimensions.

Check your understanding Beginner

Formal definition Intermediate+

The Ginzburg-Landau free energy functional

The Ginzburg-Landau free energy (sometimes called the Landau-Ginzburg Hamiltonian or the coarse-grained effective Hamiltonian) for a scalar order parameter field $m(\mathbf{x})$ in $d$ dimensions is

$$\boxed{F[m]=\int d^{d}x\left[a(T-T_c)m(\mathbf{x}{)}^2+bm(\mathbf{x}{)}^4+\kappa |\text{grad}m(\mathbf{x}){|}^2-hm(\mathbf{x})\right],}$$

where $a>0$, $b>0$, $\kappa >0$ are phenomenological parameters, $T$ is the temperature, $T_c$ is the mean-field critical temperature, and $h$ is the external field conjugate to the order parameter. This functional encodes three physical ingredients:

1. Bulk free energy: $a(T-T_c)m^2+b{m}^4$. This is the Landau expansion from 11.06.03. The quadratic coefficient changes sign at $T_c$, driving the phase transition. The quartic term stabilises the free energy.

2. Gradient penalty: $\kappa |\text{grad}m{|}^2$. Spatial variations in the order parameter cost energy proportional to the square of the gradient. This is the stiffness of the order-parameter field — it penalises rapid spatial changes and favours uniform order. The coefficient $\kappa$ has dimensions of (length)${}^2$ in the free energy density.

3. External field coupling: $-hm$. The external field $h$ couples linearly to the order parameter, exactly as in the mean-field theory.

The functional $F[m]$ is to be minimised over all possible spatial profiles $m(\mathbf{x})$ to find the equilibrium state. In mean-field theory 11.06.02, $m$ is assumed uniform, the gradient term vanishes, and the problem reduces to minimising the bulk Landau free energy. The Ginzburg-Landau theory goes further by allowing spatial variation and treating the fluctuations thermally.

Correlation length

The gradient term defines a natural length scale. Consider the quadratic part of $F$ (valid for $T>T_c$ where $m$ is small): $F\approx \int d^{d}x[a(T-T_c)m^2+\kappa |\text{grad}m{|}^2]$. The Euler-Lagrange equation for this functional is

$$-\kappa {\nabla }^{2}m+2a(T-T_c)m=h(\mathbf{x}).$$

Setting $h=0$ and looking for the decay of correlations, the Green's function satisfies

$$(-\kappa {\nabla }^2+2a|T-T_c|)G(\mathbf{r})=\delta (\mathbf{r}),$$

with solution $G(r)\sim r^{-(d-2)/2}{e}^{-r/\xi }$ (for $d>2$), where the correlation length is

$$\boxed{\xi =\sqrt{\frac{\kappa }{2a|T-T_c|}}\sim |T-T_c{|}^{-1/2}.}$$

This is the mean-field correlation length exponent $\nu =1/2$. As $T\to T_c$, $\xi \to \infty$ and correlations become long-ranged. The correlation length sets the size of the largest fluctuating domains.

Fluctuation-dissipation theorem for the order parameter

The thermal average of the square of the order-parameter fluctuation within a correlation volume is given by the fluctuation-dissipation theorem:

$$⟨(\delta m{)}^2⟩=k_{B}T\cdot \chi /{\xi }^d,$$

where $\chi$ is the susceptibility and ${\xi }^d$ is the volume of one correlated domain. The susceptibility in mean-field theory is $\chi \sim |T-T_c{|}^{-1}$ (exponent $\gamma =1$), so

$$⟨(\delta m{)}^2⟩\sim k_{B}T\cdot \frac{|T-T_c{|}^{-1}}{|T-T_c{|}^{-d/2}}=k_{B}T\cdot |T-T_c{|}^{(d-2)/2}.$$

The Ginzburg criterion

The Ginzburg criterion states that mean-field theory is self-consistent when the fluctuation of the order parameter within a correlation volume is much smaller than the square of the mean-field order parameter:

$$\boxed{⟨(\delta m{)}^2⟩\ll m_{\mathrm{mf}}^2.}$$

Substituting: $k_{B}T|T-T_c{|}^{(d-2)/2}\ll |T-T_c|$, since $m_{\mathrm{mf}}^2\sim |T-T_c|$ (with $\beta =1/2$). This simplifies to

$$|T-T_c{|}^{(4-d)/2}\gg \text{const}.$$

This is the Ginzburg criterion in $d$ dimensions.

Consequences:

• For $d>4$: the exponent $(4-d)/2<0$, so $|T-T_c{|}^{(4-d)/2}\to \infty$ as $T\to T_c$. The criterion is always satisfied near $T_c$. Mean-field theory is exact at the critical point for $d>4$.

• For $d<4$: the exponent $(4-d)/2>0$, so $|T-T_c{|}^{(4-d)/2}\to 0$ as $T\to T_c$. The criterion is always violated sufficiently close to $T_c$. Mean-field theory always fails near the critical point for $d<4$.

• For $d=4$: the exponent is zero, and the criterion becomes $|T-T_c{|}^0=1$, which is marginal. Mean-field theory holds but with logarithmic corrections: $m\sim |T-T_c{|}^{1/2}|\ln |T-T_c|{|}^{1/3}$.

The dimension $d_c=4$ is the upper critical dimension for the Ising universality class (and for any theory with a scalar order parameter and a ${\varphi }^4$ interaction).

Ginzburg temperature

The Ginzburg temperature $t_G=|T_G-T_c|/T_c$ is the reduced temperature at which the Ginzburg criterion is marginally violated. It defines the width of the critical region:

$$t_G\sim {\left(\frac{k_B^2}{a^2{\kappa }^d}\right)}^{2/(4-d)}.$$

For $T$ satisfying $|T-T_c|/T_c\gg t_G$, mean-field theory works. For $|T-T_c|/T_c\lesssim t_G$, mean-field theory fails and fluctuations dominate.

In conventional (low-$T_c$) superconductors, $t_G\sim 10^{-14}$ — the Ginzburg window is immeasurably narrow, and mean-field theory (BCS theory) works essentially perfectly. In the superfluid transition of liquid ${}^4$He, $t_G\sim 10^{-2}$ — the critical region is several millikelvin wide and fluctuations are experimentally observable. In structural phase transitions, $t_G\sim 1$ — fluctuations dominate the entire transition, and mean-field theory is qualitatively wrong.

Key derivation: The Ginzburg criterion from the Ginzburg-Landau functional Intermediate+

Proposition. The Ginzburg criterion for the validity of mean-field theory in $d$ dimensions is

$$\frac{k_B{T}_c}{\Delta C{\xi }_0^d}\cdot t^{(4-d)/2}\ll 1,$$

where $t=|T-T_c|/T_c$, $\Delta C$ is the mean-field specific heat jump, and ${\xi }_0=\sqrt{\kappa /(2a{T}_c)}$ is the bare correlation length. Mean-field theory is self-consistent when this condition holds.

Proof. The mean-field order parameter is $m_{\mathrm{mf}}^2=a(T_c-T)/(2b)$ for $T<T_c$, so $m_{\mathrm{mf}}^2\sim |T-T_c|$.

The fluctuation contribution is computed from the Gaussian approximation to the Ginzburg-Landau functional. Expanding around the mean-field solution $m_{\mathrm{mf}}$ in Fourier modes: write $m(\mathbf{x})=m_{\mathrm{mf}}+{\sum }_{\mathbf{q}}{\varphi }_{\mathbf{q}}{e}^{i\mathbf{q}\cdot \mathbf{x}}$. The fluctuation part of $F$ to quadratic order is

$$F_{\mathrm{fl}}=\frac{1}{2}\sum _{\mathbf{q}}\left[2a(T-T_c)+\kappa q^2\right]|{\varphi }_{\mathbf{q}}{|}^2+\text{higher-order terms}.$$

For $T>T_c$ (where $m_{\mathrm{mf}}=0$ and the expansion is simplest), each Fourier mode has variance

$$⟨|{\varphi }_{\mathbf{q}}{|}^2⟩=\frac{k_{B}T}{2a(T-T_c)+\kappa q^2}=\frac{k_{B}T}{\kappa (q^2+{\xi }^{-2})},$$

where $\xi =\sqrt{\kappa /(2a(T-T_c))}$. The total mean-square fluctuation of the order parameter in real space is

$$⟨(\delta m{)}^2⟩=\int \frac{d^{d}q}{(2\pi {)}^d}\frac{k_{B}T}{\kappa (q^2+{\xi }^{-2})}.$$

The integral is dominated by modes with $q\lesssim {\xi }^{-1}$ (long-wavelength fluctuations). Estimating:

$$⟨(\delta m{)}^2⟩\sim \frac{k_{B}T}{\kappa {\xi }^{-2}}\cdot \frac{{\xi }^{-d}}{(2\pi {)}^d}\cdot {\xi }^{-d}\cdot {\xi }^d\sim \frac{k_{B}T}{\kappa }{\xi }^{2-d}.$$

More carefully, in $d$ dimensions: ${\int }_0^{{\xi }^{-1}}{q}^{d-1}dq/(q^2+{\xi }^{-2})\sim {\xi }^{2-d}$ for $d<4$. So

$$⟨(\delta m{)}^2⟩\sim k_{B}T\cdot {\xi }^{2-d}/\kappa \sim |T-T_c{|}^{(d-2)/2}.$$

The Ginzburg criterion $⟨(\delta m{)}^2⟩\ll m_{\mathrm{mf}}^2$ becomes

$$|T-T_c{|}^{(d-2)/2}\ll |T-T_c|,$$

which gives $|T-T_c{|}^{(4-d)/2}\gg \text{const}$. This is the advertised result. $\square$

Bridge. The Ginzburg criterion tells us that mean-field theory fails below $d=4$, but not what replaces it. The renormalization group [11.07.01, 11.07.02] provides the resolution: it treats the Ginzburg-Landau functional as a starting point and systematically integrates out fluctuations mode by mode, tracing the flow of the parameters $a$, $b$, $\kappa$ as the length scale changes. The Wilson-Fisher fixed point of this flow gives the corrected critical exponents for $d<4$. The connection to Landau theory 11.06.03 is that the Ginzburg-Landau functional is Landau theory with a gradient term added; the gradient term is what makes fluctuations computable and what makes the renormalization group possible.

Exercises Intermediate+

Lean formalization Intermediate+

The Ginzburg-Landau theory is a variational calculus problem: minimise a functional $F[m]=\int d^{d}x[\alpha m^2+\beta m^4+\kappa |\nabla m{|}^2]$ over a space of functions. The mathematical prerequisites are within Mathlib's scope (Fréchet derivatives, Sobolev spaces, Gaussian measures on function spaces), but the physical content requires domain-specific definitions: the order parameter as a measurable function satisfying symmetry constraints, the correlation length as a spectral property of the linearised Euler-Lagrange equation, and the Ginzburg criterion as an inequality involving the fluctuation variance and the mean-field order parameter. The Gaussian integral over Fourier modes is a multivariate Gaussian computation that Mathlib could support, but the passage from the continuum Ginzburg-Landau functional to the lattice model, and the extraction of the upper critical dimension from the divergence structure of the fluctuation integral, require a framework that does not exist.

Advanced results Master

Epsilon expansion and the Wilson-Fisher fixed point

The Ginzburg criterion shows that mean-field theory fails for $d<4$, but it does not provide the corrected critical exponents. Wilson and Fisher (1972) introduced the epsilon expansion: treat the dimensionality as a continuous parameter and expand around the upper critical dimension $d_c=4$ in powers of $ϵ=4-d$.

The idea is to write the Ginzburg-Landau Hamiltonian as

$$\mathcal{H}[\varphi ]=\int d^{d}x\left[\frac{1}{2}|\nabla \varphi {|}^2+\frac{r}{2}{\varphi }^2+\frac{u}{4!}{\varphi }^4\right],$$

where $r\sim T-T_c$ and $u>0$ is the quartic coupling. At $d=4$ ($ϵ=0$), the Gaussian fixed point ($u^{\ast }=0$) controls the critical behaviour and mean-field exponents are correct.

For $d<4$, the quartic coupling $u$ is a relevant perturbation at the Gaussian fixed point: it grows under RG transformations. But there is a new fixed point — the Wilson-Fisher fixed point — at $u^{\ast }=O(ϵ)$. Computing perturbatively:

$$u^{\ast }=\frac{16{\pi }^2}{n+8}ϵ+O({ϵ}^2),$$ $$\nu =\frac{1}{2}+\frac{n+2}{4(n+8)}ϵ+O({ϵ}^2),$$ $$\eta =\frac{n+2}{2(n+8{)}^2}{ϵ}^2+O({ϵ}^3),$$

where $n$ is the number of components of the order parameter ($n=1$ for Ising, $n=2$ for XY, $n=3$ for Heisenberg). At $ϵ=1$ (i.e., $d=3$) with $n=1$: $\nu \approx 1/2+1/12\approx 0.583$ (to be compared with the exact value $\nu \approx 0.630$; higher-order terms in $ϵ$ improve the agreement). The epsilon expansion is an asymptotic series — it does not converge, but it can be resummed to give accurate results.

Lower critical dimension and the Mermin-Wagner theorem

The lower critical dimension $d_l$ is the dimension below which no phase transition occurs at any finite temperature. For the Ising universality class ($Z_2$ symmetry, discrete order parameter): $d_l=1$. For the $O(n)$ universality class with $n\ge 2$ (continuous symmetry): $d_l=2$.

The Mermin-Wagner theorem (1966) proves that continuous symmetries cannot be spontaneously broken in $d\le 2$ at finite temperature for systems with short-range interactions. The proof uses the fluctuation-dissipation theorem: the mean-square fluctuation of the order parameter diverges for $d\le 2$ because the density of long-wavelength Goldstone modes grows faster than their energy cost. Physically, spin waves — long-wavelength twists of the order parameter — cost vanishingly little energy in low dimensions and proliferate at any finite temperature, destroying long-range order.

The lower critical dimension provides a complementary boundary to the upper critical dimension: mean-field theory is exact for $d\ge d_c=4$, and no phase transition occurs for $d\le d_l$. The interesting physics lives in the window $d_l<d<d_c$, where phase transitions exist but mean-field theory is inadequate. For the Ising model ($Z_2$), this window is $1<d<4$; for the Heisenberg model ($O(3)$), it is $2<d<4$.

Amplitude ratios and universal quantities

While critical exponents are the most familiar universal quantities, amplitude ratios provide additional universal signatures that are sensitive to the universality class. Examples include:

• $A_{+}/A_{-}$: ratio of the specific heat amplitude above and below $T_c$.
• ${\xi }_{+}/{\xi }_{-}$: ratio of correlation length amplitudes.
• $R_{\chi }=C_{+}{\xi }_{+}^d/(k_B{\chi }_{+})$: a dimensionless combination of the specific heat, correlation length, and susceptibility amplitudes.

Mean-field theory predicts specific values for these ratios (e.g., $A_{+}/A_{-}=1$ for the mean-field specific heat, since the jump is symmetric). The true values differ from mean-field predictions for $d<4$ and depend on the universality class. For the 3D Ising model: $A_{+}/A_{-}\approx 0.52$, ${\xi }_{+}/{\xi }_{-}\approx 0.52$. These ratios are universal in the same sense as critical exponents — they do not depend on microscopic details, only on $d$ and $n$.

Connection to superconductivity

The Ginzburg-Landau theory was originally developed by Ginzburg and Landau (1950) for superconductivity, where the order parameter is the complex superconducting wave function $\psi (\mathbf{x})$ (a two-component real vector, so $n=2$). The functional is

$$F[\psi ]=\int d^{d}x\left[\alpha (T-T_c)|\psi {|}^2+\frac{\beta }{2}|\psi {|}^4+\frac{1}{2m^{\ast }}{\left|(-i\hslash \nabla -e^{\ast }\mathbf{A})\psi \right|}^2+\frac{|\mathbf{B}{|}^2}{2{\mu }_0}\right],$$

where $\mathbf{A}$ is the vector potential and $\mathbf{B}=\nabla \times \mathbf{A}$ is the magnetic field. The coupling to the gauge field $\mathbf{A}$ is the key new ingredient: it makes the superconducting order parameter a charged field (the $O(2)$ symmetry is promoted to a $U(1)$ gauge symmetry). This leads to the Abrikosov vortex lattice, type-I/type-II distinction, and the quantisation of magnetic flux in units of $h/(2e)$. The Ginzburg-Landau parameter ${\kappa }_{\mathrm{GL}}=\lambda /\xi$ (ratio of penetration depth to coherence length) distinguishes type-I (${\kappa }_{\mathrm{GL}}<1/\sqrt{2}$) from type-II (${\kappa }_{\mathrm{GL}}>1/\sqrt{2}$) superconductors.

For conventional superconductors, the Ginzburg temperature is $t_G\sim 10^{-14}$ — the mean-field Ginzburg-Landau theory is essentially exact for all experimentally accessible temperatures. For high-$T_c$ cuprate superconductors, $t_G$ can be much larger (up to $10^{-2}$), and fluctuation effects are observable experimentally.

Synthesis. The Ginzburg criterion identifies the boundary between mean-field and fluctuation-dominated regimes, establishing $d_c=4$ as the upper critical dimension. The Ginzburg-Landau functional provides the starting point for the renormalization group 11.07.01, which resolves the fluctuation problem by integrating out modes systematically. The Wilson-Fisher fixed point [11.07.01, 11.07.02] replaces the Gaussian fixed point for $d<4$, yielding corrected critical exponents through the epsilon expansion. The lower critical dimension $d_l$ (proved by Mermin-Wagner for continuous symmetries) complements $d_c$ by bounding the regime where phase transitions exist at all. Putting these together with Landau theory 11.06.03 and mean-field theory 11.06.02, the Ginzburg-Landau programme provides the complete framework: Landau theory gives the universal mean-field starting point, the Ginzburg-Landau functional adds spatial structure, the Ginzburg criterion diagnoses when mean-field theory fails, and the renormalization group cures the failure.

Full proof set Master

Proposition. The Ginzburg criterion for the Ising universality class in d dimensions gives the upper critical dimension $d_c=4$, and the Ginzburg temperature is $t_G\sim (k_B/(\Delta C{\xi }_0^d){)}^{2/(4-d)}$.

Proof. The mean-field order parameter is $m_{\mathrm{mf}}^2=a(T_c-T)/(2b)\sim |T-T_c|$.

The mean-square fluctuation in a correlation volume ${\xi }^d$ is computed from the Gaussian Ginzburg-Landau functional. In Fourier space, each mode has variance

$$⟨|{\varphi }_q{|}^2⟩=\frac{k_{B}T}{2a|T-T_c|+\kappa q^2}.$$

The real-space variance is

$$⟨(\delta m{)}^2⟩=\int \frac{d^{d}q}{(2\pi {)}^d}\frac{k_{B}T}{\kappa (q^2+{\xi }^{-2})},$$

where $\xi =\sqrt{\kappa /(2a|T-T_c|)}$. Converting to $k=q\xi$:

$$⟨(\delta m{)}^2⟩=\frac{k_{B}T}{\kappa {\xi }^d}\int \frac{d^{d}k}{(2\pi {)}^d}\frac{1}{k^2+1}=\frac{k_{B}T}{\kappa }{\xi }^{2-d}\cdot I_d,$$

where $I_d=\int d^{d}k/((2\pi {)}^d(k^2+1))$ is a dimensionless integral that converges for $d<2$ and diverges logarithmically for $d=2$. For $2<d<4$, $I_d$ is finite.

The Ginzburg criterion is $⟨(\delta m{)}^2⟩\ll m_{\mathrm{mf}}^2$:

$$\frac{k_{B}T}{\kappa }{\xi }^{2-d}\ll |T-T_c|.$$

Substituting $\xi ={\xi }_0|T-T_c{|}^{-1/2}$ with ${\xi }_0=\sqrt{\kappa /(2a{T}_c)}$:

$$\frac{k_B{T}_c}{\kappa }{\xi }_0^{2-d}|T-T_c{|}^{(d-2)/2}\ll |T-T_c|.$$ $$|T-T_c{|}^{(4-d)/2}\gg \frac{k_B{T}_c}{\kappa }{\xi }_0^{2-d}.$$

Converting to the reduced temperature $t=|T-T_c|/T_c$ and using $\Delta C=T_c{a}^2/(2b)$, ${\xi }_0=\sqrt{\kappa /(2a{T}_c)}$, and $b$ from the mean-field jump:

$$t_G={\left(\frac{k_B}{\Delta C{\xi }_0^d}\right)}^{2/(4-d)}.$$

For $d>4$: the exponent $(4-d)/2<0$, so the left side diverges as $t\to 0$ — the criterion is always satisfied. For $d<4$: the exponent is positive and the criterion fails for $t<t_G$. For $d=4$: logarithmic corrections replace the power law. $\square$

Proposition. The Gaussian fluctuation correction to the free energy diverges for $d\ge 4$, and the quartic coupling $u$ is relevant at the Gaussian fixed point for $d<4$.

Proof. The Gaussian fluctuation free energy is

$$F_{\mathrm{fl}}=\frac{k_{B}T}{2}\sum _q\ln (2a|T-T_c|+\kappa q^2).$$

Converting to an integral and scaling $q=k/\xi$:

$$F_{\mathrm{fl}}\sim \frac{k_{B}T}{2}{\xi }^{-d}\int d^{d}k\ln (\kappa (k^2+1)/{\xi }^2).$$

The integral diverges in the UV as $\int dk{k}^{d-1}\ln k^2$, which diverges for $d\ge 2$. This UV divergence is regularised by the lattice cutoff $\Lambda \sim 1/a$. The UV-divergent part is an analytic function of $T$ and does not affect critical behaviour. The IR-divergent part (from $k\to 0$) is $\sim {\xi }^{-d}\sim |T-T_c{|}^{d/2}$, which diverges for all $d$ but is more singular than the mean-field free energy only for $d\ge 4$.

For the RG relevance of $u$: the scaling dimension of the quartic coupling at the Gaussian fixed point is $[u]=4-d$ (obtained by dimensional analysis of the Ginzburg-Landau functional). For $d<4$: $[u]>0$, so $u$ is relevant and grows under RG flow — the Gaussian fixed point is unstable. For $d>4$: $[u]<0$, so $u$ is irrelevant and flows to zero — the Gaussian fixed point is stable and mean-field theory is exact. For $d=4$: $[u]=0$, marginal — the flow is logarithmic and leads to the Wilson-Fisher fixed point at $u^{\ast }=O(ϵ)$. $\square$

Connections Master

• 11.06.02 Mean-field theory 11.06.02 provides the predictions (critical exponents, order parameter, susceptibility) that the Ginzburg criterion tests for self-consistency. The Ginzburg-Landau functional reduces to the mean-field Landau free energy when the order parameter is uniform.
• 11.06.03 Landau theory 11.06.03 is the starting point; the Ginzburg-Landau functional adds the gradient penalty term $\kappa |\nabla m{|}^2$ that enables the study of spatial correlations and fluctuations. Without this term, there is no correlation length and no Ginzburg criterion.
• 11.07.01 The renormalization group 11.07.01 is the theory that resolves the breakdown of mean-field theory identified by the Ginzburg criterion. The Wilson-Fisher fixed point replaces the Gaussian fixed point for $d<4$.
• 11.07.02 Block-spin renormalization 11.07.02 applied to the Ginzburg-Landau Hamiltonian demonstrates the RG flow from the Gaussian to the Wilson-Fisher fixed point, providing a concrete realisation of the epsilon expansion.
• 11.06.01 The Ising model 11.06.01 is the microscopic theory; the Ginzburg-Landau functional is its coarse-grained (long-wavelength) description. The Ginzburg criterion applied to the Ising model shows that mean-field exponents are exact for $d\ge 4$ but modified for $d<4$.

Historical and philosophical context Master

Vitaly Ginzburg and Lev Landau introduced the gradient-coupled free energy functional in 1950 to describe superconductivity, recognising that the order parameter (the superconducting wave function) could vary in space. The functional form — quadratic temperature dependence near $T_c$, quartic stabilisation, and a gradient penalty — was motivated by phenomenological arguments about the symmetry and analyticity of the free energy. Ginzburg (1960) and, independently, Levanyuk (1959) derived the criterion that bears their names by estimating when the fluctuation contribution to the specific heat exceeds the mean-field jump, thereby identifying the temperature window near $T_c$ where mean-field theory fails.

The Ginzburg criterion revealed a deep fact: the validity of mean-field theory depends on the spatial dimension. In high dimensions ($d>4$), the "effective coordination number" is large enough that the mean-field approximation becomes exact. In low dimensions ($d<4$), fluctuations dominate and a new theoretical framework is needed. This dimensional threshold — the upper critical dimension $d_c=4$ — became the organising principle for the renormalization group, which was developed by Wilson and Fisher (1972) specifically to handle the $d<4$ regime.

Wilson and Fisher's key insight was to treat $d$ as a continuous variable and expand around $d_c=4$ in powers of $ϵ=4-d$. This epsilon expansion computes critical exponents perturbatively, yielding results that agree with experiment and simulation to remarkable accuracy when resummed. The Wilson-Fisher fixed point — the non-Gaussian zero of the RG beta function — replaces the Gaussian fixed point as the controller of critical behaviour for $d<4$.

The lower critical dimension has a separate history. Peierls (1936) argued informally that the 1D Ising model has no finite-$T$ transition. The rigorous result for continuous symmetries — the Mermin-Wagner theorem (1966) — proved that Goldstone modes destroy long-range order in $d\le 2$ for systems with short-range interactions and continuous symmetry. Hohenberg (1967) proved an analogous result for superfluids. Together, the upper and lower critical dimensions define the regime $d_l<d<d_c$ where phase transitions exist but are dominated by fluctuations, and where the renormalization group is essential.

The Ginzburg-Landau functional also has deep connections to quantum field theory. Written in Euclidean signature, it is the scalar ${\varphi }^4$ field theory — one of the simplest interacting quantum field theories. The renormalization group flow of the ${\varphi }^4$ coupling is the same calculation whether the context is critical phenomena or particle physics. This equivalence, recognised by Wilson and others in the early 1970s, unified the study of phase transitions and quantum field theory, establishing the renormalization group as a universal tool.

Bibliography Master

@article{ginzburg1960,
  author = {Ginzburg, V. L.},
  title = {Some remarks on phase transitions of the 2nd kind and the microscopic theory of ferroelectrics},
  journal = {Sov. Phys. Solid State},
  volume = {2},
  pages = {1824--1834},
  year = {1960}
}

@article{levanyuk1959,
  author = {Levanyuk, A. P.},
  title = {Contribution to the theory of light scattering near the second-order phase transition point},
  journal = {Sov. Phys. JETP},
  volume = {36},
  pages = {571--576},
  year = {1959}
}

@article{wilsonfisher1972,
  author = {Wilson, K. G. and Fisher, M. E.},
  title = {Critical exponents in 3.99 dimensions},
  journal = {Phys. Rev. Lett.},
  volume = {28},
  pages = {240--243},
  year = {1972}
}

@article{merminwagner1966,
  author = {Mermin, N. D. and Wagner, H.},
  title = {Absence of ferromagnetism or antiferromagnetism in one- or two-dimensional isotropic {H}eisenberg models},
  journal = {Phys. Rev. Lett.},
  volume = {17},
  pages = {1133--1136},
  year = {1966}
}

@article{ginzburglandau1950,
  author = {Ginzburg, V. L. and Landau, L. D.},
  title = {On the theory of superconductivity},
  journal = {Zh. Eksp. Teor. Fiz.},
  volume = {20},
  pages = {1064--1082},
  year = {1950}
}

@article{hohenberg1967,
  author = {Hohenberg, P. C.},
  title = {Existence of long-range order in one and two dimensions},
  journal = {Phys. Rev.},
  volume = {158},
  pages = {383--386},
  year = {1967}
}

@book{kardar2007,
  author = {Kardar, Mehran},
  title = {Statistical Physics of Fields},
  publisher = {Cambridge University Press},
  year = {2007}
}

@book{pathria-beale2011,
  author = {Pathria, R. K. and Beale, Paul D.},
  title = {Statistical Mechanics},
  edition = {3rd},
  publisher = {Academic Press},
  year = {2011}
}

@book{goldenfeld1992,
  author = {Goldenfeld, Nigel},
  title = {Lectures on Phase Transitions and the Renormalization Group},
  publisher = {Addison-Wesley},
  year = {1992}
}