RG Flow in Coupling Space

The flow picture is now the everyday language of critical phenomena across physics. Functional renormalization group methods integrate the coupling-space flow numerically for problems from ultracold atomic gases to the strong-interaction phase diagram of quantum chromodynamics, tracking the flow of many (often thousands of) effective couplings simultaneously. In cosmology the same flow logic underlies the running of couplings from the early universe to today, and the search for an ultraviolet fixed point drives the asymptotic-safety program for quantum gravity. Closer to the laboratory, the crossover geometry described here explains why experimentally extracted effective critical exponents in real magnets and fluids can appear to drift as $T_c$ is approached: the system is following a trajectory that lingers near a mean-field-like fixed point before the true asymptotic fixed point captures it, exactly the scenario the crossover scaling function describes.

Picture the space of all symmetry-allowed couplings as a landscape, and draw the coarse-graining operation as a field of arrows. Repeatedly coarse-graining a system slides its representative point along those arrows. Almost every starting point drains into a trivial sink — a pure phase where the correlation length has collapsed to zero. Only points balanced precisely on a ridge flow instead toward a saddle, and that ridge is the critical surface: the set of Hamiltonians whose correlation length is infinite.

Distance travelled along the flow corresponds to length scale. One step coarse-grains by a factor $b$; $n$ steps reach scale $b^n$. The correlation length, measured in the coarsened lattice, shrinks by $b$ each step, so any finite correlation length must eventually flow to zero — a sink. Only where $\xi=\infty$ can a trajectory come to rest at a nontrivial fixed point without shrinking away, which is why criticality and scale invariance are the same statement. The saddle, not the starting point, governs the long-distance physics — this is the picture-level content of universality. A uniaxial magnet and a liquid near its boiling point have utterly different microscopic Hamiltonians, but if both sit on the critical surface of the same fixed point, they flow to it and inherit its exponents. Their microscopic differences are directions that decay along the flow, surviving only as small corrections. And because the same diagram, drawn with different dimensionality, can move which fixed point is the stable critical one, watching the flow as $d$ crosses four dimensions is the clearest way to see why an upper critical dimension exists at all.

The flow is generated by a single map that advances one coarse-graining step. Everything else is the geometry of iterating it.

where: $K=(K_1,K_2,\dots)$ — a point in coupling space, the full set of symmetry-allowed couplings; $R_b$ — the coarse-grain-and-rescale transformation with linear factor $b>1$; $K'$ — the renormalized couplings after one step. Reading it physically: the sequence $K,R_b(K),R_b^2(K),\dots$ is the trajectory whose shape encodes the phase diagram. The maps satisfy $R_bR_{b'}=R_{bb'}$ for $b,b'>1$, but they generally have no inverse, so they form a semigroup rather than a group — the defining structural fact of the RG.

where: $\circ$ — composition of successive steps. Reading it physically: coarse-graining by $b$ then $b'$ rescales lengths by $bb'$ in total, so the maps compose with multiplying scale factors. Because each step sums over (discards) short-wavelength degrees of freedom, no inverse $R_b^{-1}$ exists; the structure is a semigroup, and the flow has a definite arrow toward long distances.

The correlation length is measured in the coarsened units, so it shrinks by $b$ at every step — the fact that pins fixed points and drives finite trajectories to sinks.

where: $\xi$ — correlation length in lattice units; $b$ — rescaling factor. Reading it physically: one step reports every length in units expanded by $b$, so $\xi$ is quoted as $\xi/b$; iterating the RG drives the correlation length measured in coarse-grained lattice units toward zero for any phase with finite physical correlation length (a sink) while $\xi=\infty$ is the only value that can survive to a nontrivial fixed point. The fixed-point condition and its consequence $\xi^*\in\{0,\infty\}$ are established in the companion article on fixed points; the flow simply carries $\xi$ downhill toward the sinks. The critical surface is defined by that surviving value of the correlation length, and its dimensionality is set by the number of relevant directions.

where: codim — codimension, the number of directions transverse to the surface. Reading it physically: the critical surface is the stable manifold of the critical fixed point, consisting of trajectories that flow into it under repeated coarse-graining; you leave it only along relevant directions, so the count of those directions is exactly how many parameters you must tune to sit on the surface. For the ordinary critical point there are two — reduced temperature $t$ and field $h$ — so the critical surface has codimension two. When two relevant directions have different rates, a trajectory can be governed by one before the other, producing crossover with a characteristic exponent.

where: $u_1,u_2$ — two relevant scaling fields with rates $y_1>y_2>0$; $\phi$ — crossover exponent; $\Phi$ — a scaling function. Reading it physically: the dimensionless argument $u_2/|u_1|^{\phi/y_1}$ decides whether the system still shows the first fixed point's behaviour or has crossed over to the asymptotic one; along the critical surface ($u_1=0$) the crossover length scales as $\xi_\times\sim u_2^{-1/y_2}$, while more generally crossover is controlled by the scaling variable $u_2/|u_1|^{\phi/y_1}$.

The flow simplifies in several controlled regimes.

Example 1 — Crossover exponent and the invariant ratio. A fixed point has a strongly relevant field $u_1$ with $y_1=1.0$ and a weakly relevant crossover field $u_2$ with $y_2=0.4$, starting at $u_1=0.03$, $u_2=0.05$. Find the crossover exponent and confirm the invariant ratio is conserved after twenty coarse-graining steps of $b=2$.

Step 1. The crossover exponent is $\phi=y_2/y_1=0.4/1.0=\mathbf{0.4}$.

Step 2. The RG-invariant combination is $u_2/u_1^{\phi}=0.05/0.03^{0.4}$. Now $0.03^{0.4}=e^{0.4\ln0.03}=e^{0.4\times(-3.507)}=e^{-1.403}=0.2459$, so the ratio is $0.05/0.2459=\mathbf{0.2034}$.

Step 3. After twenty steps of $b=2$: $u_1\to2^{20\times1.0}\times0.03$ and $u_2\to2^{20\times0.4}\times0.05$. The ratio becomes $u_2/u_1^{0.4}=(2^{8}\times0.05)/(2^{20}\times0.03)^{0.4}=(2^{8}\times0.05)/(2^{8}\times0.03^{0.4})=0.05/0.03^{0.4}=\mathbf{0.2034}$ — identical to Step 2. The ratio is conserved along the flow, as the derivation requires; the factor $2^8$ cancels exactly because $y_2=\phi\,y_1$.

Example 2 — How fast does microscopic detail wash out? An irrelevant scaling field with $y=-0.8$ starts at $u=1$. Using coarse-graining steps of $b=2$, how many steps are needed before it falls below one hundredth of its initial value, and what is its size then?

Step 1. Require $2^{yn}=2^{-0.8n}\le0.01$, i.e. $-0.8n\ln2\le\ln0.01$, giving $n\ge\ln0.01/(-0.8\ln2)=(-4.605)/(-0.5545)=8.30$. Rounding up, $\mathbf{n=9}$ steps.

Step 2. After nine steps the field is $u_9=2^{-0.8\times9}=2^{-7.2}=\mathbf{0.0068}$, indeed below $0.01$. The microscopic difference has shrunk to under one percent of its starting size — the concrete meaning of "irrelevant details wash out," and the reason two very different materials on the same critical surface end up sharing exponents. (The slowest such field sets the leading correction to scaling; for the 3D Ising class its rate is $\omega\approx0.83$.)

The fixed points that anchor the flow — the sinks and the critical saddle, and the proof that $\xi^*\in\{0,\infty\}$ — are established in Fixed Points of the RG. The scaling fields $u_i$ and their multiplicative rates $y_i$, used above without re-derivation, come from Linearization About a Fixed Point and Eigenvalue Spectrum. The classification of directions into relevant, irrelevant, and marginal is the subject of Relevant, Irrelevant, and Marginal Operators, and the conversion of the relevant eigenvalues $y_t,y_h$ into the numbers $\nu,\beta,\gamma,\delta$ is done in Critical Exponents from the RG. This article supplies the flow geometry those results live on; the quoted 3D Ising exponents ($\beta\approx0.326$, $\nu\approx0.630$, $\gamma\approx1.237$, $\omega\approx0.83$) are standard high-precision values.

References. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 14 · Kardar, Statistical Physics of Fields, Ch. 4–5 · Goldenfeld, Lectures on Phase Transitions and the Renormalization Group, Ch. 9–10 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 12 · Wilson & Kogut, Phys. Rep. 12, 75 (1974) · Cardy, Scaling and Renormalization in Statistical Physics, Ch. 3. The scaling-field structure $u_i'=b^{y_i}u_i$ is established in the linearization article, and the exponent formulas $\nu=1/y_t$, $\beta=(d-y_h)/y_t$ are derived in the critical-exponents article — they are cited here, not re-derived. Quoted 3D Ising numbers ($\beta\approx0.326$, $\nu\approx0.630$, $\gamma\approx1.237$, $\omega\approx0.83$) are standard high-precision values (Pelissetto–Vicari, Phys. Rep. 368, 549 (2002)); the upper critical dimension $d_u=4$ and $\epsilon=4-d$ are textbook.

Builds on: The Renormalization Group Idea: Coarse-Graining and Rescaling · Block-Spin (Real-Space) Renormalization Group · Decimation and the One-Dimensional Ising RG · Fixed Points of the RG. Leads to: Linearization About a Fixed Point and Eigenvalue Spectrum · Relevant, Irrelevant, and Marginal Operators · Critical Exponents from the RG · Scaling Fields and Scaling Variables · Wilson–Fisher Fixed Point · Gaussian Fixed Point and Upper Critical Dimension.