Thermodynamics · Advanced Topics

The Renormalization Group

The renormalization group (RG) is a framework for understanding how physics changes with scale. It explains why phase transitions exhibit universal behavior independent of microscopic details, and connects condensed matter physics to quantum field theory through a profound mathematical analogy.

PrerequisitesPhase transitions (Ch. PT) · Statistical mechanics (Ch. S) · Fourier analysis (Ch. F) · Group theory (Ch. GT)
Learning Goals
  • Explain why different systems near a continuous phase transition share identical critical exponents and belong to the same universality class.
  • Perform a real-space RG decimation on the 1D Ising model and show that K flows to zero at any finite temperature, confirming the absence of a phase transition.
  • Write the Landau-Ginzburg-Wilson ϕ4\phi^{4} functional and derive the one-loop RG flow equations for r and u near d=4d = 4.
  • Classify RG perturbations as relevant, irrelevant, or marginal, and use this classification to derive all critical exponents from two independent eigenvalues yty_{t} and yhy_{h}.
  • Verify the scaling relations α+2β+γ=2\alpha + 2\beta + \gamma = 2 and dν=2−αd\nu = 2 - \alpha using the 3D Ising critical exponents.

RG.1 Scale and Universality

Near a continuous phase transition (T→TcT \to T_c), the correlation length diverges:ξ∼∣T−Tc∣−ν→∞\xi \sim |T - T_c|^{-\nu} \to \infty. At TcT_c, the system looks the same at all length scales — it is scale invariant. This is why different systems (magnets, liquids, superconductors) show identical critical exponents — the same universality class.

Critical exponents (defined near TcT_c, h = 0 unless noted):

ξ∼∣t∣−ν\xi \sim |t|^{-\nu},   C∼∣t∣−αC \sim |t|^{-\alpha},   m∼∣t∣βm \sim |t|^{\beta} (t=(T−Tc)/Tct = (T - T_c)/T_c, h = 0),   χ∼∣t∣−γ\chi \sim |t|^{-\gamma},   m∼h1/δm \sim h^{1/\delta} (T=TcT = T_c).

Mean-field values: α = 0, β = 1/2, γ = 1, δ = 3, ν = 1/2. Ising 3D (Wilson-Fisher): β ≈ 0.326, γ ≈ 1.237, ν ≈ 0.630. These differ from mean-field because of fluctuations.

RG.2 Real-Space Renormalization Group

The RG procedure: coarse-grain the system by a factor b, then rescale to restore the original lattice spacing. Under this transformation, the coupling constants of the Hamiltonian flow:

K′=R(K)(RG transformation: couplings K→K′ after coarse-graining by b)K' = R(K) \qquad \text{(RG transformation: couplings } K \to K' \text{ after coarse-graining by } b\text{)}(RG.1)

Fixed points K* satisfy R(K*) = K* — scale-invariant systems. The nature of the fixed point determines the phase:

K = 0 (high T): high-T fixed point — disordered phase. K = ∞ (low T): low-T fixed point — ordered phase. K = K* (critical): unstable fixed point — the phase transition.

Example RG.1 — 1D Ising Model — Exact RG

Apply the RG to the 1D Ising model H=−K∑isisi+1H = -K \sum_{i} s_{i} s_{i+1} by summing out every other spin.

Partition function:Z=∑{s}eK∑isisi+1Z = \sum_{\{s\}} e^{K \sum_i s_i s_{i+1}}. Integrate out even spins s2,s4,⋯(b=2s_{2}, s_{4}, \cdots (b = 2 decimation).
Sums2Sum s_{2}:For each s2s_{2}: ∑s2=±1eKs1s2+Ks2s3=2cosh⁡(K(s1+s3))\sum_{s_2 = \pm 1} e^{K s_1 s_2 + K s_2 s_3} = 2\cosh\bigl(K(s_1 + s_3)\bigr). This depends on s1s3s_{1}s_{3} only. Write as eK′s1s3+conste^{K' s_1 s_3 + \text{const}}.
New coupling:Comparing: 2cosh⁡(2K)=AeK′2 \cosh(2K) = A e^{K'} (if s1=s3=+1).2cosh⁡(0)=Ae(−K′s_{1}=s_{3}=+1). 2 \cosh(0) = A e^(-K') (if s1=−s3s_{1}=-s_{3}). Dividing: tanh⁡(K′)=tanh⁡2(K)\tanh(K') = \tanh^{2}(K). So K' = arctanh(tanh⁡2(K))<K\tanh^{2}(K)) < K.
Fixed points:K′=KK' = K only for K=0K = 0 (trivial) or K=∞K = \infty (trivial). No finite fixed point →no\to no phase transition in 1D Ising at finite T! RG confirms the exact result (Peierls argument).
Physical:K→0K \to 0 under RG: the system always flows to disorder at any finite T. Long-range order requires T=0T = 0. The 2D Ising model has a nontrivial fixed point (Kc=ln⁡(1+2)/2)—K_{c} = \ln(1+\sqrt 2)/2) — gives the phase transition.

RG.3 The Wilson-Fisher Fixed Point

Kadanoff and Wilson reformulated RG in momentum space. Start with the Landau free energy functional (the Wilson-Fisher φ⁴ theory):

F[ϕ]=∫ddr[12(∇ϕ)2+12rϕ2+uϕ4](Landau-Ginzburg-Wilson functional)F[\phi] = \int d^d r \left[\tfrac{1}{2}(\nabla\phi)^2 + \tfrac{1}{2} r \phi^2 + u \phi^4\right] \qquad \text{(Landau-Ginzburg-Wilson functional)}(RG.2)

RG in momentum space: integrate out modes with |k| between Λ/b and Λ (shell), then rescale k → bk to restore cutoff Λ. The couplings flow:

drdl=2r+12uKdΛdr+Λ2(flow of mass parameter)\frac{dr}{dl} = 2r + \frac{12u K_d \Lambda^d}{r + \Lambda^2} \qquad \text{(flow of mass parameter)}(RG.3)
dudl=(4−d)u−36u2KdΛd(r+Λ2)2(flow of coupling)\frac{du}{dl} = (4-d)u - \frac{36u^2 K_d \Lambda^d}{(r + \Lambda^2)^2} \qquad \text{(flow of coupling)}(RG.4)

where l = ln(b). The crucial observation: the coefficient of u in Eq. RG.4 is (4−d). In d = 4: u is marginal (flows logarithmically). In d < 4: u is relevant — interactions matter. In d > 4: u is irrelevant — mean-field is exact.

The Wilson-Fisher fixed point at d = 4−ε (ε expansion):u∗=ε36Kd+O(ε2)u^* = \dfrac{\varepsilon}{36 K_d} + O(\varepsilon^2). The critical exponents to first order in ε:

ν=12+ε12+O(ε2)η=0+O(ε2)(Ising class, ε=4−d)\nu = \tfrac{1}{2} + \frac{\varepsilon}{12} + O(\varepsilon^2) \qquad \eta = 0 + O(\varepsilon^2) \qquad \text{(Ising class, } \varepsilon = 4 - d\text{)}(RG.5)

At ε = 1 (d = 3): ν ≈ 0.583 (actual 0.630) — good first approximation. Higher orders: ν = 0.630 at 5-loop order, matching experiments and Monte Carlo to 4 significant figures.

RG.4 Scaling and Universality from Fixed Points

At a fixed point K*, linearize the RG transformation: δK′=M δK\delta K' = M\, \delta K where M=dR/dK∣K∗M = \left.dR/dK\right|_{K^*}. The eigenvalues Λi=byi\Lambda_i = b^{y_i} classify the perturbations:

Relevant: yi>0 (Λi>1)y_i > 0\ (\Lambda_i > 1) — grows under RG, takes system away from fixed point. Controls the distance from TcT_c and h.

Irrelevant: yi<0 (Λi<1)y_i < 0\ (\Lambda_i < 1)— shrinks under RG. Microscopic details that don't affect critical behavior. This is WHY universality exists!

Marginal: yi=0y_i = 0 — needs higher-order analysis. Leads to logarithmic corrections (2D Ising at TcT_c, or QCD asymptotic freedom).

Theorem RG.1 — Scaling Hypothesis and Exponent Relations
At a fixed point with two relevant eigenvalues yty_t (temperature) and yhy_h (field), the free energy obeys a homogeneous scaling form:f(t,h)=b−df(bytt, byhh)for any bf(t, h) = b^{-d} f\bigl(b^{y_t} t,\ b^{y_h} h\bigr) \quad \text{for any } bThis leads to exact exponent relations — only two exponents are independent: α+2β+γ=2\alpha + 2\beta + \gamma = 2 (Rushbrooke), dν=2−αd\nu = 2 - \alpha (hyperscaling), γ=ν(2−η)\gamma = \nu(2 - \eta). All critical exponents of a universality class follow from two numbers (yt,yhy_t, y_h).

RG.5 Applications Beyond Phase Transitions

The RG idea extends far beyond condensed matter:

Quantum field theory: the beta function β(g) = μ dg/dμ is the RG flow of coupling g with energy scale μ. Asymptotic freedom (QCD) = g → 0 under RG to high energy. The Landau pole in QED = g → ∞ (UV problem).

Chaotic systems: Feigenbaum constants (δ = 4.669..., α = 2.502...) in the period-doubling route to chaos — universal regardless of the map. Explained by a fixed point of the RG acting on maps.

Turbulence: Kolmogorov scaling E(k)∝k−5/3E(k) \propto k^{-5/3} can be understood as a fixed point of the RG for the Navier-Stokes equation (though a rigorous derivation remains incomplete).

Disordered systems: random systems have infinite-randomness fixed points where quantum fluctuations are amplified — describes random quantum spin chains, many-body localization.

Definition RG.2 — Common Traps
  • RG transformations change descriptions, not the underlying system: coarse-graining tracks which couplings matter at large scale.
  • Relevant does not mean important in everyday language: it means the perturbation grows under repeated rescaling.
  • Universality ignores irrelevant details: lattice structure and microscopic chemistry often wash out near a fixed point.
  • Mean-field breaks below the upper critical dimension: fluctuations change exponents in d < 4 for the Ising class.
Exercises — RG.1–RG.5 Renormalization Group
1.
Verify the scaling relations α+2β+γ=2\alpha + 2\beta + \gamma = 2 and dν=2−αd\nu = 2 - \alpha using the 3D Ising exponents α=0.110,β=0.326,γ=1.237,ν=0.630\alpha = 0.110, \beta = 0.326, \gamma = 1.237, \nu = 0.630. What is the upper critical dimension?
(upper critical dimension)
Straightforward
2.
Apply the Migdal-Kadanoff real-space RG to the 2D Ising model with b=2b=2. Find the approximate KcK_{c} and compare to the exact Onsager value Kc=ln⁡(1+2)/2K_{c} = \ln(1+\sqrt 2)/2.
Intermediate
3.Describe the Berezinskii-Kosterlitz-Thouless (BKT) transition using the RG. What is the role of vortices and what is the universal jump in the superfluid density?
Intermediate
4.Derive the Wilson-Fisher fixed point in the ε=4−d\varepsilon = 4-d expansion for the ϕ4\phi^{4} theory. What are the flow equations for r and u to one loop, and what is u*?
Challenging
Key Takeaways
  • RG: coarse-grain by b, rescale. Couplings flow K→K′K \to K'. Fixed points → phases; unstable fixed point → phase transition.
  • Relevant perturbations (y>0): control critical behavior. Irrelevant (y<0): explains universality.
  • 1D Ising: RG flows to disorder at any T→noT \to no transition. 2D: nontrivial fixed point at KcK_{c}.
  • Wilson-Fisher: ϕ4\phi^{4} theory near d=4.u∗=ε/36d=4. u* = \varepsilon/36, exponents in ε expansion. Ising 3D at ε=1\varepsilon=1.
  • Exponent relations: α+2β+γ=2,dν=2−α,γ=ν(2−η)\alpha+2\beta+\gamma=2, d\nu=2-\alpha, \gamma=\nu(2-\eta). Only two independent exponents.
  • RG explains QFT running couplings, Feigenbaum universality in chaos, Kolmogorov turbulence scaling.