The renormalization group — Wilson's framework

Intuition Beginner

Near a critical point, the same physics recurs across wildly different microscopic systems. The liquid-vapour transition in carbon dioxide, the Curie point in iron, and the unmixing of two fluids at their consolute point all share the same critical exponents to experimental precision. Their atoms and their microscopic Hamiltonians are entirely unrelated. The Wilson framework of the renormalization group explains why: microscopic details are progressively erased as you observe at longer length scales, and only a small set of shared features survives the erasure.

Wilson's idea is to regard every possible Hamiltonian as a single point in an enormous space — call it theory space. Coarse-graining (zooming out by averaging over blocks) moves you through this space along a trajectory called an RG flow. The coordinates of theory space are the coupling constants of every interaction the symmetry allows: nearest-neighbour couplings, next-nearest-neighbour couplings, four-spin terms, and so on without end.

Special points in theory space — fixed points — are unchanged by the coarse-graining. A fixed point represents physics that looks identical on every length scale, which is scale invariance. Universality acquires a geometric meaning in this picture. Microscopic systems start at different points of theory space, but their RG-flow trajectories converge on the same fixed point. Whatever survives the flow is shared, and so the critical exponents match.

Visual Beginner

The diagram captures Wilson's framework in one picture. The Gaussian fixed point $G$ (mean-field theory) is unstable below $d=4$ dimensions: its quartic coupling is a relevant operator that grows under coarse-graining. Trajectories launched near $G$ flow away and converge on the Wilson-Fisher fixed point $WF$, which has exactly two relevant directions — temperature and magnetic field — and infinitely many irrelevant ones. The basin of attraction of $WF$ is the Ising universality class: every microscopic Hamiltonian inside that basin shares the same critical exponents.

Worked example Beginner

The 2D Ising model is exactly solvable (Onsager 1944), and its critical exponents are simple fractions that disagree sharply with mean-field theory.

Compare the correlation-length exponent $\nu$, which controls how fast the correlation length $\xi$ blows up as $T$ approaches $T_c$:

• Mean-field (Landau theory, Gaussian fixed point): $\nu =1/2$.
• Exact 2D Ising: $\nu =1$.
• The difference is a factor of two.

The Wilson framework reads this gap as the signature of two different fixed points. The thermal eigenvalue at a fixed point is $y_t=1/\nu$. For mean-field, $y_t=1/(1/2)=2$. For the 2D Ising fixed point, $y_t=1/1=1$. The halving of $y_t$ between the two fixed points doubles $\nu$: the correlation length diverges twice as fast under the non-mean-field fixed point as under the Gaussian one.

The anomalous-dimension exponent $\eta$ tells the same story — mean-field $\eta =0$ against exact 2D Ising $\eta =1/4$. Below the upper critical dimension $d=4$, the Gaussian fixed point loses stability and a non-mean-field fixed point takes over. Identifying which fixed point controls a given transition is the central task of the Wilson framework.

Check your understanding Beginner

Formal definition Intermediate+

Wilsonian RG as a flow on theory space

Let $\mathcal{H}[\varphi ;\mathbf{K}]$ denote a Hamiltonian parametrised by a countably infinite vector of coupling constants $\mathbf{K}=(K_1,K_2,K_3,\dots )$, where $K_1$ is the nearest-neighbour coupling, $K_2$ the next-nearest-neighbour coupling, $K_3$ a four-spin term, and so on through every interaction compatible with the symmetry. The space of all such $\mathbf{K}$ is theory space $\mathcal{T}$. A single coarse-graining step with scale factor $b>1$ defines a map ${\mathcal{R}}_b:\mathcal{T}\to \mathcal{T}$ by integrating out short-wavelength modes and rescaling lengths and fields. The Wilsonian RG is the semigroup $\{{\mathcal{R}}_b{\}}_{b>1}$ acting on $\mathcal{T}$.

Fixed points and linearisation

A fixed point ${\mathbf{K}}^{\ast }$ satisfies ${\mathcal{R}}_b({\mathbf{K}}^{\ast })={\mathbf{K}}^{\ast }$ for every $b>1$. Linearise about ${\mathbf{K}}^{\ast }$ by writing $\delta \mathbf{K}=\mathbf{K}-{\mathbf{K}}^{\ast }$ and expanding to first order:

$${\mathcal{R}}_b({\mathbf{K}}^{\ast }+\delta \mathbf{K})={\mathbf{K}}^{\ast }+M_b\delta \mathbf{K}+O(\delta {\mathbf{K}}^2),$$

where $M_b$ is the linearised RG transformation. Diagonalising $M_b$ yields eigenvalues ${\lambda }_i$ that take the scaling form ${\lambda }_i=b^{y_i}$ because the semigroup property ${\mathcal{R}}_b\circ {\mathcal{R}}_{b^{\prime }}={\mathcal{R}}_{b{b}^{\prime }}$ forces ${\lambda }_i(b){\lambda }_i(b^{\prime })={\lambda }_i(b{b}^{\prime })$. The exponents $y_i=\ln {\lambda }_i/\ln b$ are the scaling dimensions of the corresponding operators.

Classification of scaling operators

Each eigendirection ${\mathbf{e}}_i$ of $M_b$ is a scaling operator. The sign of $y_i$ classifies it:

• Relevant ($y_i>0$): ${\lambda }_i>1$, so the perturbation grows under coarse-graining and drives the system away from ${\mathbf{K}}^{\ast }$. Relevant operators must be tuned to zero to reach criticality.
• Irrelevant ($y_i<0$): ${\lambda }_i<1$, so the perturbation shrinks and washes out. Irrelevant operators are the source of universality.
• Marginal ($y_i=0$): ${\lambda }_i=1$, so the perturbation is unchanged at linear order. Higher-order analysis decides marginally relevant versus marginally irrelevant.

The number of relevant operators at ${\mathbf{K}}^{\ast }$ is the codimension of its critical surface — the number of experimental knobs that must be tuned to land on the fixed point.

Critical surface and basins of attraction

The critical surface of ${\mathbf{K}}^{\ast }$ is the set of all $\mathbf{K}\in \mathcal{T}$ whose RG trajectory flows into ${\mathbf{K}}^{\ast }$. For a fixed point with $n_{\mathrm{rel}}$ relevant operators, the critical surface has codimension $n_{\mathrm{rel}}$. An ordinary critical point has $n_{\mathrm{rel}}=2$ (one thermal knob, one field knob); a tricritical point has $n_{\mathrm{rel}}=3$. The universality class of ${\mathbf{K}}^{\ast }$ is its basin of attraction — every microscopic Hamiltonian in the basin shares the critical exponents of ${\mathbf{K}}^{\ast }$.

Critical exponents from RG eigenvalues

For a fixed point with thermal eigenvalue $y_t$ and magnetic eigenvalue $y_h$, the standard critical exponents are fixed by the scaling relations:

$$\nu =\frac{1}{y_t},\alpha =2-\frac{d}{y_t},\beta =\frac{d-y_h}{y_t},\gamma =\frac{2y_h-d}{y_t},\delta =\frac{y_h}{d-y_h},\eta =2+d-2y_h.$$

These reduce six critical exponents to two independent eigenvalues. The hyperscaling identity $\alpha =2-d\nu$ holds below the upper critical dimension and fails above it, where mean-field exponents take over and $d$ drops out.

Key derivation Intermediate+

The one-loop beta function and the Wilson-Fisher fixed point

This is the calculation that locates the non-mean-field fixed point in theory space and produces the $\epsilon$-expansion. Start from the Landau-Ginzburg-Wilson Hamiltonian for an $n$-component order parameter $\mathbf{\varphi }=({\varphi }_1,\dots ,{\varphi }_n)$ in $d=4-\epsilon$ dimensions:

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

where $r\propto T-T_c^{\mathrm{MF}}$ measures the distance from mean-field criticality and $u>0$ is the quartic self-coupling.

Execute one Wilsonian momentum-shell step: integrate out modes in the shell $\Lambda /b<|\mathbf{k}|<\Lambda$, rescale momenta ${\mathbf{k}}^{\prime }=b\mathbf{k}$ to restore the cutoff, and rescale the field to keep the gradient term at unit coefficient. At one loop, two diagrams contribute: the tadpole (renormalising $r$) and the three fish diagrams — the $s$-, $t$-, and $u$-channel bubbles (renormalising $u$). Taking $b=e^{\ell }$ infinitesimal yields the differential flow equations:

$$\frac{dr}{d\ell }=2r+\frac{n+2}{6}\frac{u}{(4\pi {)}^2}\frac{1}{1+r},\frac{du}{d\ell }=-\epsilon u+\frac{n+8}{6}\frac{u^2}{(4\pi {)}^2}\frac{1}{(1+r{)}^2}.$$

On the critical surface $r=0$, the second equation is the celebrated one-loop beta function:

$$\beta (u)\equiv \frac{du}{d\ell }{|}_{r=0}=-\epsilon u+\frac{n+8}{6}\frac{u^2}{(4\pi {)}^2}.$$

Locating the Wilson-Fisher fixed point

The two terms in $\beta (u)$ compete. The linear term $-\epsilon u$ is the engineering scaling of $u$ (whose bare dimension is $4-d=\epsilon$); the quadratic term is the one-loop correction. The Gaussian fixed point $u^{\ast }=0$ exists for every $\epsilon$ but is unstable to the quartic coupling when $\epsilon >0$ (that is, when $d<4$). Setting $\beta (u^{\ast })=0$ with $u^{\ast }\ne 0$ gives the Wilson-Fisher fixed point:

$$u^{\ast }=\frac{6\epsilon (4\pi {)}^2}{n+8}=\frac{96{\pi }^2\epsilon }{n+8}.$$

This fixed point exists only for $\epsilon >0$ — below the upper critical dimension — and sits perturbatively close to the Gaussian fixed point at a coupling of order $\epsilon$. That proximity is what justifies expanding every observable in powers of $\epsilon$.

Critical exponents at $O(\epsilon )$

Linearising the flow about $(r,u)=(0,u^{\ast })$, the thermal eigenvalue picks up a one-loop correction. Combining the tadpole contribution, the field-rescaling step, and the position of $u^{\ast }$ gives the Wilson-Kogut 1974 result:

$$y_t=2-\frac{n+2}{n+8}\epsilon +O({\epsilon }^2).$$

Inverting via $\nu =1/y_t$ and expanding in $\epsilon$:

$$\nu =\frac{1}{2}+\frac{n+2}{4(n+8)}\epsilon +O({\epsilon }^2).$$

For the 3D Ising class ($n=1$, $\epsilon =1$): $\nu =1/2+3/36=7/12\approx 0.583$. The exact value is $\nu \approx 0.629971$. The one-loop $\epsilon$-expansion already captures roughly $64\%$ of the deviation from the mean-field value $\nu =0.5$; higher loops and Borel resummation close the gap (the five-loop resummed result is $\nu \approx 0.6304$). The anomalous dimension vanishes at one loop; at two loops the sunset diagram gives $\eta =(n+2){\epsilon }^2/[2(n+8{)}^2]+O({\epsilon }^3)$, which for $n=1$, $\epsilon =1$ yields $\eta \approx 0.019$ against the exact $0.036$. Hyperscaling gives the specific-heat exponent $\alpha =(4-n)\epsilon /[2(n+8)]+O({\epsilon }^2)$, which vanishes at $n=4$ — a clean fingerprint of the universality class readable off the one-loop result.

Bridge. The Wilson-Fisher fixed point $u^{\ast }=96{\pi }^2\epsilon /(n+8)$ is the engine of the entire framework: it locates, in theory space, the fixed point whose two relevant eigenvalues supply every non-mean-field critical exponent through the scaling relations. This is exactly the calculation that realises Kadanoff's 1966 block-spin picture as a controlled dynamical-systems analysis on the space of Hamiltonians; the foundational reason the scheme works is that the Wilson-Fisher fixed point sits perturbatively close to the Gaussian fixed point for small $\epsilon$, so the loop expansion is simultaneously an expansion in the distance between the two fixed points. The one-loop beta function builds toward the multi-loop, Borel-resummed $\epsilon$-expansion of 11.07.05, generalises to the real-space block-spin recursions of 11.07.03 in which the same fixed point is approached by explicit lattice coarse-graining, and appears again in 11.06.05 as the microscopic derivation of the scaling-hypothesis exponents $y_t$ and $y_h$ that the Kadanoff-Widom data collapse had already constrained.

Exercises Intermediate+

Lean formalization Intermediate+

There is no Lean module for this unit. Mathlib contains the differential topology, linear algebra, and metric-space infrastructure required to state an RG flow as a smooth map on a finite-dimensional manifold of couplings, and it has the eigenvalue machinery to classify fixed points once such a map is given. What it lacks is the infinite-dimensional geometry of theory space, a formal semigroup of coarse-graining transformations acting on that space, and a theorem tying the signs of linearised eigenvalues to the existence and stability of critical surfaces. The deeper obstruction is that the Wilson-Fisher fixed point is defined as a zero of a beta function obtained from a perturbative loop expansion — an asymptotic series, not a convergent one — and Mathlib has no formal infrastructure for resummed perturbation theory. Until those pieces exist, the gap recorded in Mathlib gap analysis above marks the formalisation boundary; this unit ships as reviewer-attested prose.

Advanced results Master

Universality classes: the $(d,n,\text{range})$ classification

The Wilson framework reduces the empirical notion of "universality class" to a small set of qualitative data that fix which fixed point a microscopic Hamiltonian flows to. Three pieces of data dominate.

Spatial dimension $d$. The engineering dimension of every operator depends on $d$, so the relevance of any given coupling changes with dimension. The quartic coupling ${\varphi }^4$ is irrelevant for $d>4$, marginal at $d=4$, and relevant for $d<4$. The ${\varphi }^6$ coupling is irrelevant for $d>3$, marginal at $d=3$, and relevant for $d<3$ — the basis of tricritical universality. Below the lower critical dimension $d_l$ (equal to $1$ for discrete symmetries, $2$ for continuous symmetries by the Mermin-Wagner-Hohenberg theorem), no ordered phase exists at finite temperature and the framework has no critical fixed point to locate.

Order-parameter symmetry. A scalar order parameter ($n=1$) gives the Ising class; a two-component order parameter with $O(2)$ symmetry ($n=2$) gives the XY class; three components with $O(3)$ symmetry give the Heisenberg class. More generally, $O(n)$-symmetric models form a one-parameter family indexed by $n$, and the one-loop Wilson-Fisher fixed point $u^{\ast }=96{\pi }^2\epsilon /(n+8)$ interpolates continuously between them. Additional symmetries (cubic anisotropy, random-site disorder, chirality) split these into further classes when they change the spectrum of relevant operators.

Range of interactions. Short-range interactions (exponentially decaying) all flow to the same fixed point as nearest-neighbour interactions. Long-range power-law interactions $J(r)\sim 1/r^{d+\sigma }$ produce a distinct family parametrised by $\sigma$: mean-field behaviour for $\sigma <d/2$, an interpolating interacting fixed point for $d/2<\sigma <2$, and crossover to the short-range Wilson-Fisher class at $\sigma =2$.

Crossover phenomena

When two fixed points share a relevant direction, trajectories can drift from the basin of one to the basin of another as a coupling is varied — a crossover. The classic example is the crossover from Ising to mean-field behaviour as the dimension crosses $d=4$: just below $d=4$ the Wilson-Fisher fixed point is close to the Gaussian fixed point and the system exhibits mean-field exponents over a wide intermediate range of $t$, crossing over to non-mean-field exponents only extremely close to $T_c$. Crossover is governed by crossover exponents $\varphi =y_{\mathrm{cross}}/y_t$, where $y_{\mathrm{cross}}$ is the eigenvalue of the operator driving the crossover. Fisher's 1974 review systematised crossover scaling forms and showed that the entire phenomenology of crossover — including the apparent violation of simple power laws in experimental data — is captured by the linearised spectrum of scaling operators at the dominant fixed point.

Marginal operators and logarithmic corrections

Marginal operators ($y_i=0$) are the subtlest case. At the upper critical dimension $d=4$, the quartic coupling is marginal at the Gaussian fixed point, and the flow in the $u$-direction is governed not by a linear eigenvalue but by the one-loop beta function $\beta (u)=(n+8)u^2/[6(4\pi {)}^2]+O(u^3)$. The quadratic vanishing of $\beta$ at $u=0$ produces logarithmic corrections to scaling: correlation functions acquire factors of $(\ln \xi {)}^{\hat{\eta }}$ rather than pure power laws. The marginal case is therefore not a boundary between two regimes but a third regime in its own right, and its experimental signatures (log-periodic modulations, slowly varying effective exponents) are the clearest fingerprints of the upper critical dimension.

Multicriticality and tricritical fixed points

When more than one relevant operator must be tuned, the fixed point is multicritical. The tricritical Ising universality class — where the ${\varphi }^4$ coefficient is tuned to zero and ${\varphi }^6$ stabilises the theory — has upper critical dimension $d=3$ and codimension $3$. Its one-loop exponents differ from those of the ordinary Ising class and are accessed by an $\epsilon$-expansion around $d=3$ rather than $d=4$. Higher multicritical points (${\varphi }^8$, ${\varphi }^{10}$) have progressively lower upper critical dimensions and wider codimensions; they form a hierarchy of fixed points in theory space, each with its own basin of attraction.

The c-theorem and RG irreversibility

Zamolodchikov's 1986 c-theorem states that in two-dimensional unitary Lorentz-invariant field theories there exists a function $c(g)$ of the couplings that decreases monotonically along RG flows and equals the conformal-anomaly central charge at fixed points. The theorem formalises the intuition that RG coarse-graining loses information: an infrared fixed point has $c_{\mathrm{IR}}\le c_{\mathrm{UV}}$, and the difference is the "information" carried away by the integrated-out irrelevant operators. The four-dimensional analogue — the a-theorem, with the central charge replaced by the conformal anomaly coefficient $a$ — was proved by Komargodski and Schwimmer in 2011. Together these theorems impose an arrow on RG flows: not every ultraviolet theory can flow to every infrared fixed point, because the relevant monotonic quantity must not increase. The theorems are the deepest structural statement about the geometry of theory space that the Wilson framework has produced.

Synthesis. The Wilson framework recasts critical phenomena as a dynamical-systems problem on the space of Hamiltonians, with universality emerging as the basin structure around a handful of fixed points. The central insight is that the long-distance physics is controlled by the spectrum of scaling operators at the fixed point: relevant operators are the few knobs that must be tuned, irrelevant operators are the many microscopic details that wash out, and marginal operators mark the boundary regimes where logarithmic corrections appear. The foundational reason the framework is predictive rather than merely descriptive is the $\epsilon$-expansion, which locates the Wilson-Fisher fixed point perturbatively close to the Gaussian one and turns the classification program into a quantitative computation of critical exponents. Putting these together with the $(d,n,\text{range})$ taxonomy, the operator-counting derivation of codimension, and the monotonicity theorems that impose an arrow on RG flows, the bridge is between the phenomenology of scaling 11.06.05 and the microscopic lattice calculations 11.07.03; the same Wilson-Fisher fixed point generalises across universality classes and appears again in the conformal-field-theory treatment of 11.07.05 as an interacting scale-invariant fixed point whose operator spectrum completely determines the critical data.

Full proof set Master

Proposition 1 (Scaling relations as algebraic identities)

Proposition. Let ${\mathbf{K}}^{\ast }$ be a fixed point of the Wilsonian RG with thermal eigenvalue $y_t>0$ and magnetic eigenvalue $y_h>0$. Then the six standard critical exponents $(\alpha ,\beta ,\gamma ,\delta ,\nu ,\eta )$ defined by the scaling relations

$$\nu =\frac{1}{y_t},\alpha =2-\frac{d}{y_t},\beta =\frac{d-y_h}{y_t},\gamma =\frac{2y_h-d}{y_t},\delta =\frac{y_h}{d-y_h},\eta =2+d-2y_h$$

satisfy the four scaling relations of Rushbrooke ($\alpha +2\beta +\gamma =2$), Widom ($\gamma =\beta (\delta -1)$), Fisher ($\gamma =\nu (2-\eta )$), and hyperscaling ($\alpha =2-d\nu$) identically.

Proof. Each relation is verified by direct substitution.

Rushbrooke. $\alpha +2\beta +\gamma =(2-d/y_t)+2(d-y_h)/y_t+(2y_h-d)/y_t=2+(-d+2d-2y_h+2y_h-d)/y_t=2+0=2.$

Widom. $\beta (\delta -1)=[(d-y_h)/y_t]\cdot [y_h/(d-y_h)-1]=[(d-y_h)/y_t]\cdot [(y_h-d+y_h)/(d-y_h)]=[(d-y_h)/y_t]\cdot [(2y_h-d)/(d-y_h)]=(2y_h-d)/y_t=\gamma .$

Fisher. $\nu (2-\eta )=(1/y_t)(2-(2+d-2y_h))=(1/y_t)(2y_h-d)=\gamma .$

Hyperscaling. $\alpha =2-d/y_t=2-d\nu$ directly, since $\nu =1/y_t$.

All four relations hold identically in $y_t$ and $y_h$; none is an independent empirical constraint. The six exponents are therefore determined by two eigenvalues, and the four scaling relations are algebraic consequences of the scaling form of the singular free energy. $\square$

The content of the proposition is that critical-exponent measurements are massively over-determined by the RG: once two independent exponents (conventionally $\nu$ and $\eta$) are measured or computed, the other four follow without further input. This is why the experimental verification of scaling relations in the 1960s was such strong indirect evidence for the Wilson framework a decade before the $\epsilon$-expansion made the eigenvalues computable.

Proposition 2 (Codimension equals the number of relevant operators)

Proposition. Let ${\mathbf{K}}^{\ast }$ be a fixed point of the Wilsonian RG semigroup $\{{\mathcal{R}}_b\}$ with linearised eigenvalues $\{y_i\}$. Locally near ${\mathbf{K}}^{\ast }$, the critical surface — the set of couplings flowing into ${\mathbf{K}}^{\ast }$ — is a smooth submanifold of theory space whose codimension equals the number of relevant operators (those with $y_i>0$).

Proof sketch. By the stable-manifold theorem for the map ${\mathcal{R}}_b$ (applied to the contracting directions, which are the irrelevant operators with $y_i<0$), there exists a local stable manifold $W_{\mathrm{loc}}^s$ tangent at ${\mathbf{K}}^{\ast }$ to the span of the eigenvectors with $y_i<0$. This manifold is invariant under ${\mathcal{R}}_b$, and every point on it converges to ${\mathbf{K}}^{\ast }$ under iteration. The critical surface is precisely $W_{\mathrm{loc}}^s$, so its dimension equals the number of irrelevant plus marginal directions, and its codimension equals the number of relevant directions. The change of coordinates from bare couplings $\mathbf{K}$ to scaling fields $\{u_i\}$ — in which ${\mathcal{R}}_b$ acts diagonally as $u_i\mapsto b^{y_i}{u}_i$ — is guaranteed locally by Sternberg-style linearisation theorems for the contracting sector. $\square$

The proposition quantifies universality: a fixed point with $n_{\mathrm{rel}}$ relevant operators requires $n_{\mathrm{rel}}$ experimental knobs to reach, and every microscopic Hamiltonian differing only in irrelevant couplings lies on the same critical surface and shares the same exponents. For the Ising universality class, Proposition 6 above showed $n_{\mathrm{rel}}=2$ (temperature and field); for tricritical Ising, $n_{\mathrm{rel}}=3$.

Connections Master

• 11.07.01 This unit deepens the introductory renormalization-group picture of 11.07.01, which surveyed block-spin decimation, the momentum-shell RG, the Callan-Symanzik equation, and the menu of numerical RG schemes (Monte-Carlo RG, DMRG, NRG, FRG). Where 11.07.01 introduced the machinery, this unit organises it into Wilson's framework: a flow on theory space whose fixed points and operator spectrum explain universality, crossover, and the codimension of critical surfaces. The two-loop $\epsilon$-expansion exponents quoted in 11.07.01 are derived here as the central locating calculation.

• 11.06.02 Mean-field theory (Curie-Weiss, van der Waals) is the physics of the Gaussian fixed point: the framework identifies mean-field exponents as the RG eigenvalues at $u^{\ast }=0$, correct above $d=4$ where the quartic coupling is irrelevant. The breakdown of mean-field theory below $d=4$ is, in framework language, the destabilisation of the Gaussian fixed point by the relevant quartic operator and the consequent flow to the Wilson-Fisher fixed point.

• 11.06.03 Landau theory is the Gaussian-fixed-point effective Hamiltonian — the Landau-Ginzburg-Wilson ${\varphi }^4$ functional at zero loop. The framework promotes Landau theory from a phenomenological free-energy expansion to the seed Hamiltonian whose RG flow generates the Wilson-Fisher fixed point. The order-parameter symmetry that classifies Landau theories is the same symmetry that fixes $n$ and hence the universality class.

• 11.06.04 The Ginzburg criterion diagnoses the regime where the Gaussian fixed point is self-consistent by comparing fluctuation strength to the mean-field order parameter. In framework language, the Ginzburg criterion is the linearised stability analysis of the Gaussian fixed point: it identifies $d=4$ as the upper critical dimension below which the quartic operator becomes relevant and the Gaussian fixed point loses stability. The Ginzburg number is the bare value of the relevant coupling that the RG subsequently grows to $u^{\ast }$.

• 11.06.05 The scaling hypothesis of Widom and Kadanoff posits homogeneous forms for the singular free energy and the equation of state. The framework derives these forms: the homogeneity $f_{\mathrm{sing}}(t,h)=b^{-d}{f}_{\mathrm{sing}}(b^{y_t}t,b^{y_h}h)$ is the statement that the free energy is controlled by a fixed point with eigenvalues $y_t,y_h$, and the scaling relations among exponents are the algebraic identities of Proposition 1 above.

• 11.07.03 The real-space block-spin RG of 11.07.03 (Niemeijer-van Leeuwen cumulant expansion, Migdal-Kadanoff bond moving) is an explicit lattice implementation of the abstract coarse-graining map ${\mathcal{R}}_b$ on theory space defined here. Both schemes locate the same Wilson-Fisher fixed point; the real-space version trades the momentum-shell loop expansion for a cumulant expansion in the block coupling, and its fixed-point eigenvalues approximate the same $y_t,y_h$ from a different direction in theory space.

• 11.07.05 The one-loop $\epsilon$-expansion derived here as the engine that locates the Wilson-Fisher fixed point is pushed to five loops, Borel-resummed, and connected to two-dimensional conformal field theory in 11.07.05. That unit takes over the high-precision computation of critical exponents and identifies the 2D Ising fixed point with the minimal model $\mathcal{M}(4,3)$ — the framework's prediction that the fixed point is an interacting scale-invariant theory made exact in two dimensions.

Historical & philosophical context Master

The Wilsonian framework has two conceptual ancestors that converged in the early 1970s. Leo Kadanoff's 1966 paper "Scaling laws for Ising models near $T_c$" ^{[Kadanoff1966]} introduced the block-spin picture as a physical argument for scaling and universality and ended with the prescient sentence that the cell-to-cell transformation should be regarded "as a mathematical operation defined upon the set of all Hamiltonians." That single sentence contains the seed of the entire framework — the idea that coarse-graining acts on a space of Hamiltonians and that the fixed points of this action control critical behaviour — but Kadanoff did not yet possess a controlled calculational scheme for locating the fixed points or computing their eigenvalues.

Kenneth Wilson's 1971 pair of papers "Renormalization group and critical phenomena I, II" ^{[Wilson1971a]} turned Kadanoff's picture into both a definition and a computational tool. Wilson formulated the RG explicitly as a semigroup of transformations on the space of Hamiltonians, introduced the momentum-shell coarse-graining step that makes perturbation theory systematic, and recognised that the fixed-point structure — not just the existence of fixed points but the classification of their scaling operators into relevant, irrelevant, and marginal — is what explains universality. The 1974 review with John Kogut, "The renormalization group and the $\epsilon$ expansion" ^{[WilsonKogut1974]}, is the foundational synthesis: it laid out the framework essentially as it is taught today and collected the one-loop results for the Wilson-Fisher fixed point and the $O(\epsilon )$ critical exponents. Wilson received the 1982 Nobel Prize in Physics for this body of work.

Michael Fisher's 1974 review "The renormalization group in the theory of critical behavior" ^[Fisher1974] systematised the classification of scaling fields, introduced the modern language of crossover exponents, and organised the empirical universality classes into the $(d,n,\text{range})$ taxonomy used ever since. Fisher's contribution was to show that the framework is not just a computational method but a classification theory: the spectrum of relevant operators at a fixed point is a complete invariant of the universality class, in the sense that two fixed points with the same relevant spectrum govern systems with identical critical exponents.

The deepest philosophical consequence of the framework is the recognition that "renormalization" in the Wilsonian sense is not the removal of infinities but the systematic reorganisation of degrees of freedom by length scale. The divergences of perturbative quantum field theory, which had bedevilled the subject since the 1940s, are reinterpreted in Wilson's picture as artefacts of taking the cutoff to infinity prematurely; the physical content is the flow of couplings with scale, and the fixed points of that flow — not the bare Lagrangian — are what the experiments measure. This reconception, more than any single calculation, is why the framework spread from statistical mechanics into high-energy physics, condensed matter, and beyond.

Bibliography Master

@article{kadanoff1966,
  author = {Kadanoff, Leo P.},
  title = {Scaling laws for {I}sing models near {$T_c$}},
  journal = {Physics},
  volume = {2},
  pages = {263--272},
  year = {1964}
}

@article{wilson1971a,
  author = {Wilson, Kenneth G.},
  title = {Renormalization group and critical phenomena. {I}. {R}enormalization group and the {K}adanoff scaling picture},
  journal = {Phys. Rev. B},
  volume = {4},
  pages = {3174--3183},
  year = {1971}
}

@article{wilson1971b,
  author = {Wilson, Kenneth G.},
  title = {Renormalization group and critical phenomena. {II}. {P}hase-space cell analysis of critical behavior},
  journal = {Phys. Rev. B},
  volume = {4},
  pages = {3184--3205},
  year = {1971}
}

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

@article{wilsonkogut1974,
  author = {Wilson, Kenneth G. and Kogut, John},
  title = {The renormalization group and the {$\varepsilon$} expansion},
  journal = {Phys. Reports},
  volume = {12C},
  pages = {75--199},
  year = {1974}
}

@article{fisher1974,
  author = {Fisher, Michael E.},
  title = {The renormalization group in the theory of critical behavior},
  journal = {Rev. Mod. Phys.},
  volume = {46},
  pages = {597--616},
  year = {1974}
}

@article{zamolodchikov1986,
  author = {Zamolodchikov, Alexander B.},
  title = {Irreversibility of the flux of the renormalization group in a {2D} field theory},
  journal = {JETP Lett.},
  volume = {43},
  pages = {730--732},
  year = {1986}
}

@article{komargodskischwimmer2011,
  author = {Komargodski, Zohar and Schwimmer, Adam},
  title = {Renormalization group flows and anomalies},
  journal = {J. High Energy Phys.},
  volume = {2011},
  pages = {99},
  year = {2011}
}

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

@book{cardy1996,
  author = {Cardy, John},
  title = {Scaling and Renormalization in Statistical Physics},
  publisher = {Cambridge University Press},
  year = {1996}
}

@book{zinnjustin2002,
  author = {Zinn-Justin, Jean},
  title = {Quantum Field Theory and Critical Phenomena},
  edition = {4th},
  publisher = {Oxford University Press},
  year = {2002}
}