Momentum-shell (Wilson) renormalization group

Intuition Beginner

Imagine watching a magnet near its critical temperature through a camera that you can zoom out, one small step at a time. Each time you zoom out, fine details blur together and only the broad patterns of order survive. The momentum-shell renormalization group is a precise recipe for that zoom. It removes the finest-wavelength wiggles first, then shrinks the picture back to its original size so you can compare before and after.

The payoff is that the rules of the system change a little with every zoom step. A few of those rules grow, a few shrink, and a few stay put. The ones that stay put describe the system exactly at its critical point. This is why very different materials can behave identically at criticality: they all zoom toward the same unchanging set of rules.

The whole method splits into two moves you repeat over and over: average out the short-wavelength fluctuations, then rescale lengths so the smallest surviving wiggle looks the same size as before.

Visual Beginner

The left band shows all the wavelengths present in the field; the thin outer ring is the shortest-wavelength shell that gets averaged away in one step. The right band shows the same field after rescaling, so the picture looks the same size again but with slightly altered rules.

The picture emphasizes that one renormalization step is a thin shell removed plus a uniform shrink, not a sudden jump.

Worked example Beginner

Track one number: the strength $u$ of the interaction that makes the field prefer to settle near two values rather than wander freely. Suppose one zoom step changes $u$ by the simple rule "new $u$ equals old $u$ plus a bonus of $0.10$ times old $u$, minus a penalty of $9$ times old $u$ squared."

Start at $u=0.01$. The bonus is $0.10\times 0.01=0.001$. The penalty is $9\times (0.01{)}^2=0.0009$. New $u$ is about $0.01+0.001-0.0009=0.0101$. The strength grew a little.

Start instead at $u=0.02$. Bonus is $0.002$, penalty is $9\times 0.0004=0.0036$. New $u$ is about $0.0184$. Now it shrank.

Somewhere between these two starting points the bonus and penalty cancel and $u$ stops moving. That balance point is the fixed value: here about $u=0.10/9\approx 0.011$. What this tells us: a growth term and a shrink term can balance at one special coupling, and that balance is the critical fixed point.

Check your understanding Beginner

Formal definition Intermediate+

Work with a single real scalar field $\varphi (\mathbf{x})$ in $d$ Euclidean dimensions, governed by the Landau-Ginzburg-Wilson weight $e^{-\mathcal{H}[\varphi ]}$ with

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

where $t$ measures the distance from the mean-field critical temperature and $u>0$ is the quartic coupling. A sharp ultraviolet cutoff $\Lambda$ restricts the Fourier modes to $|\mathbf{q}|<\Lambda$; this is the regulator that makes every integral below finite. The conventions follow Kardar ^{[Kardar Ch. 5]}. The Gaussian piece is the free theory of 08.06.01.

Split the field by wavenumber: $\varphi ={\varphi }^{<}+{\varphi }^{>}$, where ${\varphi }^{<}$ keeps modes with $|\mathbf{q}|<\Lambda /b$ and ${\varphi }^{>}$ keeps the momentum shell $\Lambda /b<|\mathbf{q}|<\Lambda$, for a rescaling factor $b=e^{\ell }>1$. One renormalization-group step is the composition of three operations.

Coarse-grain. Integrate out the shell field ${\varphi }^{>}$, producing an effective weight for the slow field,

$$e^{-{\mathcal{H}}_b[{\varphi }^{<}]}=\int \mathcal{D}{\varphi }^{>}{e}^{-\mathcal{H}[{\varphi }^{<}+{\varphi }^{>}]}.$$

Rescale. Restore the original cutoff by setting $\mathbf{x}=b{\mathbf{x}}^{\prime }$, equivalently ${\mathbf{q}}^{\prime }=b\mathbf{q}$, so that $|{\mathbf{q}}^{\prime }|<\Lambda$ again.

Renormalize the field. Define ${\varphi }^{\prime }({\mathbf{x}}^{\prime })={\zeta }^{-1}{\varphi }^{<}(b{\mathbf{x}}^{\prime })$ and fix $\zeta$ by demanding that the gradient term keep unit coefficient. The map $(t,u)\mapsto (t^{\prime },u^{\prime })$ is the renormalization-group transformation $R_b$.

The fixed point $(t^{\star },u^{\star })$ satisfies $R_b(t^{\star },u^{\star })=(t^{\star },u^{\star })$ for all $b$. Linearising $R_b$ about a fixed point yields scaling eigenvalues $b^{y_i}$; the exponents $y_i$ classify each coupling as relevant ($y_i>0$), irrelevant ($y_i<0$), or marginal ($y_i=0$).

Key theorem with proof Intermediate+

Theorem (one-loop recursion relations). For the ${\varphi }^4$ theory above with an $n$-component field $\mathbf{\varphi }$ and an infinitesimal step $b=e^{\delta \ell }$, the couplings obey, to one loop in $u$,

$$\frac{dt}{d\ell }=2t+(n+2)4C\frac{u}{1+t},\frac{du}{d\ell }=(4-d)u-(n+8)4C\frac{u^2}{(1+t{)}^2},$$

where $C=K_d{\Lambda }^d/(2{\Lambda }^2)$ collects the angular factor $K_d=S_{d-1}/(2\pi {)}^d$ from the shell integration. Near the fixed point ($t\to 0$) these reduce to $dt/d\ell =2t+(n+2)4Cu$ and $du/d\ell =ϵu-(n+8)4C{u}^2$ with $ϵ=4-d$.

Proof. Write $\mathcal{H}={\mathcal{H}}_0[{\varphi }^{<}]+{\mathcal{H}}_0[{\varphi }^{>}]+\mathcal{U}[{\varphi }^{<},{\varphi }^{>}]$, where ${\mathcal{H}}_0$ is the Gaussian part and $\mathcal{U}$ collects the ${\varphi }^4$ terms coupling slow and fast modes. Integrating out ${\varphi }^{>}$ gives a cumulant expansion of the effective weight,

$${\mathcal{H}}_b[{\varphi }^{<}]={\mathcal{H}}_0[{\varphi }^{<}]+⟨\mathcal{U}{⟩}_{>}-\frac{1}{2}(⟨{\mathcal{U}}^2{⟩}_{>}-⟨\mathcal{U}{⟩}_{>}^2)+\cdots ,$$

with $⟨\cdot {⟩}_{>}$ the Gaussian average over the shell. The shell propagator is $⟨{\varphi }_i^{>}(\mathbf{q}){\varphi }_j^{>}(-\mathbf{q}){⟩}_0={\delta }_{ij}/(q^2+t)$ supported on $\Lambda /b<|\mathbf{q}|<\Lambda$.

The first-order term $⟨\mathcal{U}{⟩}_{>}$ contracts two of the four legs of each ${\varphi }^4$ vertex across the shell. For an $O(n)$-symmetric quartic $u(\mathbf{\varphi }\cdot \mathbf{\varphi }{)}^2$ the combinatorial weight of self-contracting one pair is $(n+2)$, and the contracted pair contributes the shell integral

$${\int }_{\Lambda /b}^{\Lambda }\frac{d^{d}q}{(2\pi {)}^d}\frac{1}{q^2+t}=\frac{K_d{\Lambda }^d}{{\Lambda }^2+t}\delta \ell (1+O(\delta \ell )),$$

shifting the mass term by $\delta t=(n+2)4u{K}_d{\Lambda }^d/({\Lambda }^2+t)\delta \ell$. Setting $\Lambda =1$ and writing $C=K_d/2$ gives the loop coefficient stated.

The second-order cumulant generates the leading correction to $u$. Contracting two vertices by a pair of shell propagators leaves a four-slow-leg vertex; the $O(n)$ combinatorics of the distinct pairings sum to $(n+8)$, and the two propagators contribute $[K_d{\Lambda }^d/({\Lambda }^2+t{)}^2]\delta \ell$. The induced shift is $\delta u=-(n+8)4u^2{K}_d{\Lambda }^d/({\Lambda }^2+t{)}^2\delta \ell$, the sign reflecting that the connected second cumulant lowers the effective coupling.

Finally apply rescaling and field renormalization. Under $\mathbf{q}\to b\mathbf{q}$ the engineering dimensions are $[t]=2$ and $[u]=4-d$ at the Gaussian fixed point, contributing the linear pieces $2t$ and $(4-d)u$. Adding the engineering and loop contributions and dividing by $\delta \ell$ yields the stated differential recursion relations ^{[Wilson 1971]}. $\square$

Bridge. This recursion is exactly the engine that 08.04.02 asserts but does not turn. The foundational reason the Wilson-Fisher value $u^{\star }=ϵ/[(n+8)4C]$ exists is visible right here: the linear term $ϵu$ and the quadratic loop term balance, and this is the central insight that builds toward the exponent computation in the Advanced section. The mass-flow equation generalises the naive scaling dimension of $t$ by a fluctuation correction, and that correction is what putting these two equations together converts into a corrected critical exponent $\nu$. The linearised eigenvalues of this same map appear again in 08.05.01, where the two independent exponents $y_t$ and $y_h$ are read off; the bridge is that scaling laws there are nothing but the eigenvalues of the map proved here. This identifies criticality with the fixed point of $R_b$, and it is dual to the beta-function picture of 08.04.03 under $\beta (u)=-du/d\ell$.

Exercises Intermediate+

Advanced results Master

The fixed-point structure organizes the whole flow. Two fixed points coexist in the $(t,u)$ plane: the Gaussian point $(0,0)$ and the Wilson-Fisher point $(t^{\star },u^{\star })$ with $u^{\star }=ϵ/[(n+8)4C]$ and $t^{\star }=-(n+2)2C{u}^{\star }=O(ϵ)$. For $ϵ>0$ the Gaussian point has two relevant directions ($t$ and $u$), so it cannot control criticality; the Wilson-Fisher point has a single relevant direction ($t$), making it the generic critical fixed point reached by tuning one parameter, the temperature. This single-relevant-direction count is the renormalization-group explanation of universality stated in 08.04.02.

Linearising $R_b$ at the Wilson-Fisher point gives the two scaling eigenvalues. The coupling direction carries eigenvalue $y_u=-ϵ<0$ (irrelevant, computed in Exercise 4), confirming infrared attractivity. The thermal direction carries $y_t=2-\frac{n+2}{n+8}ϵ$, and the correlation-length exponent follows from $\nu =1/y_t$:

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

A parallel computation with a symmetry-breaking field $h$ gives the magnetic eigenvalue $y_h=\frac{1}{2}(d+2-\eta )=3-\frac{1}{2}ϵ+O({ϵ}^2)$. With $y_t$ and $y_h$ in hand, every other exponent follows from the scaling relations of 08.05.01: $\alpha =2-d\nu$, $\beta =(d-y_h)\nu$, $\gamma =(2y_h-d)\nu$, and $\delta =y_h/(d-y_h)$. The exponent identities are therefore not independent postulates but consequences of two renormalization-group eigenvalues.

The anomalous dimension enters only at second order. The one-loop tadpole that corrects the propagator is momentum-independent, so it renormalizes $t$ but not the coefficient of $q^2$; hence $\eta =O({ϵ}^2)$, with the explicit two-loop value $\eta =(n+2){ϵ}^2/[2(n+8{)}^2]$ ^{[Wilson-Fisher 1972]}. This is why mean-field theory, which sets $\eta =0$ and $\nu =\frac{1}{2}$, is corrected gently rather than catastrophically just below four dimensions: the corrections are organized as a power series in the small parameter $ϵ=4-d$.

Synthesis. Putting these together, the momentum-shell renormalization group is the foundational reason the disparate facts of critical phenomena form one structure. The central insight is that integrating out a shell and rescaling defines a flow whose fixed points are scale-invariant theories; this is exactly the mechanism that converts the Gaussian theory of 08.06.01 into the interacting Wilson-Fisher theory below four dimensions. The single relevant direction at that fixed point is dual to the experimental fact of a one-parameter critical surface, and it generalises Kadanoff's heuristic block-spin picture 08.04.04 into a controlled computation. The two eigenvalues $y_t$ and $y_h$ identify criticality with a linear map, and this is exactly what makes the scaling relations of 08.05.01 theorems rather than fits; the $ϵ$-expansion builds toward the modern conformal-bootstrap determinations that appear again in the conformal-criticality units, and the whole flow is dual to the beta-function geometry of 08.04.03 under $\beta (u)=-du/d\ell$.

Full proof set Master

Proposition (existence and order of the interacting fixed point). For $0<ϵ\ll 1$ and $n\ge 1$, the one-loop recursion $du/d\ell =ϵu-(n+8)4C{u}^2$ has exactly one fixed point with $u>0$, located at $u^{\star }=ϵ/[(n+8)4C]=O(ϵ)$, and it is infrared-stable in the $u$ direction.

The right-hand side $f(u)=ϵu-(n+8)4C{u}^2=u[ϵ-(n+8)4Cu]$ vanishes precisely at $u=0$ and $u=u^{\star }:=ϵ/[(n+8)4C]$. Since $ϵ>0$ and $C>0$, the second root is positive, and it is the unique positive zero because $f$ is a quadratic with leading coefficient $-(n+8)4C<0$. Differentiating, $f^{\prime }(u)=ϵ-2(n+8)4Cu$, so $f^{\prime }(0)=ϵ>0$ (the Gaussian point is unstable) and $f^{\prime }(u^{\star })=ϵ-2ϵ=-ϵ<0$ (the Wilson-Fisher point is stable). Because $u^{\star }\propto ϵ$, the coupling at the fixed point is small for small $ϵ$, which is what justifies truncating the cumulant expansion at one loop: the neglected $O(u^3)$ terms are $O({ϵ}^3)$ relative to the retained $O({ϵ}^2)$ balance. $\square$

Proposition (thermal eigenvalue and correlation-length exponent). Linearising the coupled recursion about $(t^{\star },u^{\star })$ to first order in $ϵ$, the thermal eigenvalue is $y_t=2-\frac{n+2}{n+8}ϵ$, and hence $\nu =1/y_t=\frac{1}{2}+\frac{n+2}{4(n+8)}ϵ+O({ϵ}^2)$.

Write $t=t^{\star }+\delta t$, $u=u^{\star }+\delta u$. The mass recursion $dt/d\ell =2t+(n+2)4Cu/(1+t)$ linearises to

$$\frac{d\delta t}{d\ell }=[2-(n+2)4C\frac{u^{\star }}{(1+t^{\star }{)}^2}]\delta t+\frac{(n+2)4C}{1+t^{\star }}\delta u.$$

The eigenvalue along the relevant (thermal) direction is the coefficient of $\delta t$ after accounting for the irrelevant $\delta u$ direction, which decays as $e^{-ϵ\ell }$ and so does not shift the leading eigenvalue. Substituting $u^{\star }=ϵ/[(n+8)4C]$ and $t^{\star }=O(ϵ)$ and keeping first order in $ϵ$,

$$y_t=2-(n+2)4C\cdot \frac{ϵ}{(n+8)4C}=2-\frac{n+2}{n+8}ϵ.$$

Inverting, $\nu =y_t^{-1}=[2-\frac{n+2}{n+8}ϵ{]}^{-1}=\frac{1}{2}(1+\frac{n+2}{2(n+8)}ϵ)+O({ϵ}^2)=\frac{1}{2}+\frac{n+2}{4(n+8)}ϵ+O({ϵ}^2)$. $\square$

Connections Master

This unit derives the one-loop recursion whose fixed point 08.04.02 states but leaves unproved; the Wilson-Fisher value $u^{\star }$ and the exponent $\nu$ computed here are precisely the quantities asserted there in its synthesis.

The flow equation $du/d\ell$ is the negative of the beta function of 08.04.03: with $\beta (u)=-du/d\ell$, the infrared-stable zero $u^{\star }$ is a zero of $\beta$ with positive slope, matching the fixed-point classification given there.

The two scaling eigenvalues $y_t$ and $y_h$ extracted from the linearised map feed directly into the critical-exponent machinery of 08.05.01; the scaling relations are read off as algebraic consequences of these eigenvalues rather than postulated.

The Gaussian fixed point that the flow departs from is the free field theory of 08.06.01, and the coarse-grain-then-rescale step generalises the discrete block-spin transformation of 08.04.04 into a continuous momentum-space operation built on the real-space framework of 08.04.01.

Historical & philosophical context Master

The momentum-shell construction is due to Kenneth Wilson, who in two 1971 papers recast the renormalization group as an explicit coarse-graining flow in field space rather than the multiplicative reparametrization of high-energy physics ^{[Wilson 1971]}. Wilson's move was to treat the cutoff as physical and to track how effective couplings change as short-wavelength modes are removed, converting Kadanoff's block-spin intuition into a calculable transformation. The decisive quantitative result came in 1972, when Wilson and Fisher observed that dimension $d$ could be treated as a continuous parameter and that the interacting fixed point sits at a coupling of order $ϵ=4-d$, opening a controlled perturbative expansion around four dimensions ^{[Wilson-Fisher 1972]}. The comprehensive account, including the diagrammatics summarized here, appeared in the Wilson-Kogut review ^{[Wilson-Kogut 1974]}. Wilson received the 1982 Nobel Prize for this reformulation, which dissolved a long-standing puzzle: why microscopically unrelated systems share critical exponents. The philosophical shift is that universality is a statement about flow in the space of theories, not about any microscopic Hamiltonian.

Bibliography Master

@article{Wilson1971RG1,
  author  = {Wilson, Kenneth G.},
  title   = {Renormalization Group and Critical Phenomena. I. Renormalization Group and the Kadanoff Scaling Picture},
  journal = {Physical Review B},
  volume  = {4},
  year    = {1971},
  pages   = {3174--3183}
}

@article{Wilson1971RG2,
  author  = {Wilson, Kenneth G.},
  title   = {Renormalization Group and Critical Phenomena. II. Phase-Space Cell Analysis of Critical Behavior},
  journal = {Physical Review B},
  volume  = {4},
  year    = {1971},
  pages   = {3184--3205}
}

@article{WilsonFisher1972,
  author  = {Wilson, Kenneth G. and Fisher, Michael E.},
  title   = {Critical Exponents in 3.99 Dimensions},
  journal = {Physical Review Letters},
  volume  = {28},
  year    = {1972},
  pages   = {240--243}
}

@article{WilsonKogut1974RG,
  author  = {Wilson, Kenneth G. and Kogut, John},
  title   = {The Renormalization Group and the epsilon Expansion},
  journal = {Physics Reports},
  volume  = {12},
  year    = {1974},
  pages   = {75--199}
}

@book{Kardar2007Fields,
  author    = {Kardar, Mehran},
  title     = {Statistical Physics of Fields},
  publisher = {Cambridge University Press},
  year      = {2007}
}