Wilsonian and Functional Renormalization

Wilsonian renormalization changes resolution by integrating out or suppressing fluctuations, so the action itself moves through the space of symmetry-allowed interactions. Functional renormalization expresses that motion as an exact differential equation for an entire functional. These ideas explain why apparently irrelevant operators are generated, how fixed points organize long-distance behavior, and why a finite ansatz remains an approximation even when its parent flow equation is exact.

This chapter follows three related but distinct constructions: a finite momentum-shell step, Polchinski’s flow for a Wilsonian interaction action, and Wetterich’s flow for an effective average action. It then develops the projection and symmetry checks required to turn a functional equation into a defensible calculation. Model-specific phase diagrams, lattice blocking algorithms, and theorem-level constructions remain with their neighboring volumes.

Flowing functionals across resolution scales

Let a microscopic Euclidean scalar theory be regulated at a high scale $\Lambda_{\rm UV}$ . Split its field into slow and fast components and define a Wilsonian action by

e^{-S_k[\phi_<]} \equiv \int\mathcal D\phi_>\, e^{-S_{\Lambda_{\rm UV}}[\phi_<+\phi_>]},

where the fast integration covers modes between $k$ and $\Lambda_{\rm UV}$ in a declared blocking prescription. Lowering $k$ integrates out more fluctuations. The result is generally not the original action with only a few running constants:

S_k[\phi] = \sum_i g_i(k)\,\mathcal O_i[\phi].

All quasi-local operators compatible with the symmetries can be induced. The coefficients $g_i(k)$ are coordinates in theory space; a finite truncation selects a submanifold or projection, not the full flow. Wilson and Kogut develop blocking, rescaling, fixed points, and continuum trajectories as one connected construction in Wilson and Kogut 1974, §§ 1.1 and 11–12, pp. 78–83 and 152–176.

There are two functional objects in this chapter:

Object	What it is	Regulator role	Endpoint statement
Wilsonian action $S_k[\phi]$	Action for modes retained below a coarse-graining scale after faster modes have been integrated out	A UV covariance or blocking kernel determines which modes move between integrated and retained sectors	At the starting scale it matches declared microscopic data; lowering $k$ successively incorporates higher-momentum fluctuations
Effective average action $\Gamma_k[\varphi]$	Modified Legendre transform of a source functional with low-momentum modes suppressed	An additive infrared kernel $R_k$ gives modes with $p^2\ll k^2$ an effective gap	Under suitable limits, $\Gamma_{k\to0}$ becomes the full 1PI effective action

The objects are related, but they are not interchangeable. $S_k$ organizes the action of fields still to be integrated; $\Gamma_k$ is a scale-dependent 1PI generator. Their kernels, boundary data, field variables, and endpoint limits must be named before formulas are compared.

This is also different from the subtraction-scale flow of Renormalization-Group Equations and Running. A Callan–Symanzik equation varies $\mu$ while preserving the same bare theory and complete prediction. A Wilsonian step changes the resolution and hence the action used for the remaining modes. The two descriptions reproduce compatible physical scaling when constructed consistently, but $k$ is not merely another symbol for $\mu$ .

Check your preparation

This diagnostic chooses a route; it is not a scored assessment.

Can you perform this task?	Ready: enter here	Unsure: repair
Split a Gaussian functional integral into independent momentum sectors and integrate one sector	Wilsonian coarse graining	Review Gaussian Fields and Sources.
Distinguish a regulator from a physical cutoff and from a subtraction scale	Wilsonian coarse graining	Review Regulators, Cutoffs, and Continuum Limits and the scale-operation map.
Use a cumulant expansion and rescale momenta, coordinates, and fields without mixing dimensionful and dimensionless couplings	Momentum-shell integration	Review Gaussian Vectors, Processes, Random Distributions, and Wick Structure.
Take two functional derivatives and interpret a trace over momentum, internal, and statistics indices	Polchinski equation	First project a functional identity onto its two- and four-point vertices.
Construct a Legendre transform and relate its Hessian to the connected two-point function	Wetterich equation	Review The 1PI Effective Action and Mean-Field Equations.
State which operators a finite ansatz omits and how a projection extracts each retained coupling	Truncations and projections	Work through the exact-flow pages before interpreting a projected beta function.
Write the Ward or Slavnov–Taylor identity that a regulator modifies	Symmetry and error control	Review Coupling to Background Gauge Fields and Bundles or Slavnov–Taylor and Zinn-Justin Identities.

Readers interested only in scalar flows can take the first five rows. Gauge-theory applications require the symmetry branch before a regulator-dependent result can be interpreted physically.

Choose a route

In the route column, ⇒ marks a hard dependency and → a useful continuation.

Goal	Route	Required output
Understand what “integrating out modes” means	Coarse graining	A blocking map with cutoff, remaining field, induced operators, and invariant long-distance data identified
Derive perturbative shell beta functions	Coarse graining ⇒ shell integration and rescaling	A finite shell step with cumulants, rescalings, dimensionless couplings, and semigroup check
Obtain a differential equation for a Wilson action	Core route through shell integration ⇒ Polchinski equation	A cutoff-independent partition-function identity and a projection showing what the functional equation retains
Flow a scale-dependent 1PI action	Coarse graining + 1PI preparation ⇒ Wetterich equation	A modified Legendre transform, inverse-Hessian trace, UV initial condition, and $k\to0$ endpoint
Turn an exact flow into numbers	Wetterich equation ⇒ truncations and projections	An ansatz, projection rule, closure assumption, nested enlargement, regulator scan, and independent benchmark
Make a symmetry-sensitive claim	Core route through truncations ⇒ symmetry and error control	A modified identity, endpoint-restoration test, and decomposed numerical, regulator, projection, and omitted-operator errors
Test the workflow computationally	Wetterich equation → zero-dimensional benchmark	Compare exact quadrature with matched nested projections and more than one regulator

The Polchinski and Wetterich branches can be read in either order after coarse graining. Their crosswalk is clearest once their functionals and kernels have been written side by side.

The six pages in order

Wilsonian Coarse Graining and Theory Space defines a Wilson action, blocking map, quasi-locality, semigroup composition, dimensionless coordinates, and the distinction between exact and projected trajectories.
Momentum-Shell Integration and Rescaling performs a finite scalar shell step. It separates Gaussian integration, cumulants, coordinate and field rescaling, induced operators, and the leading mass and quartic flows near four dimensions.
The Polchinski Exact RG Equation replaces infinitesimal shell bookkeeping by a smooth-cutoff functional identity. It derives the classical product and quantum trace terms and projects them onto a bounded vertex sector.
Effective Average Actions and the Wetterich Equation introduces an infrared regulator and modified Legendre transform. It derives the exact inverse-Hessian trace and tests its local-potential projection against a bounded scalar example.
Functional-RG Truncations and Projection Methods compares derivative, vertex, polynomial, and grid expansions. It makes closure rules, projection points, field reparameterizations, regulator choices, and nested-convergence claims explicit.
Symmetry, Regulator Dependence, and Functional-RG Error Control derives the consequences of regulator-broken identities and assembles an error statement that does not mistake regulator variation for complete uncertainty.

The first two pages establish the physical operation. The next two derive exact functional equations for different objects. The last two control the approximation and symmetry evidence needed for applications.

Scale direction and endpoint convention

This chapter uses

t\equiv\ln\frac{k}{\Lambda_{\rm UV}}, \qquad \partial_t=k\partial_k.

With fixed $\Lambda_{\rm UV}$ , lowering $k$ toward the infrared makes $t$ decrease. When an infrared-directed parameter is more readable, define

\ell \equiv \ln\frac{\Lambda_{\rm UV}}{k} =-t, \qquad \partial_\ell=-\partial_t.

Every plotted arrow and stability statement must say whether it follows increasing $t$ or increasing $\ell$ . The next chapter reserves “relevant” and “irrelevant” for directions defined relative to a fixed point and a declared flow direction.

The endpoint card is:

Construction	Starting data	Flow direction used here	Endpoint qualification
Finite shell blocking	$S_{\Lambda_{\rm UV}}$ and a mode split	Lower $k$ or increase $\ell$	Compose shell maps only while the derivative or locality expansion remains controlled
Polchinski flow	Smooth UV covariance and Wilson interaction action	Integrate the differential identity from microscopic to lower $k$	Partition-function invariance is exact; a vertex or derivative projection is not
Wetterich flow	$\Gamma_{k_{\rm UV}}$ matched to microscopic data and an IR regulator $R_k$	Lower $k$ until $R_k\to0$	$\Gamma_{k\to0}=\Gamma$ only if the regulator limit, initial matching, and solution are controlled

A large starting $k$ in the effective-average-action construction suppresses fluctuations; it does not mean that $\Gamma_k$ is automatically equal to a chosen classical action. Equality is an initial-condition approximation whose cutoff corrections must be stated.

Two exact equations, two different functionals

For a smooth Wilsonian covariance $C_k$ , write the interaction part as $S_k^{\rm I}[\phi]$ . One common sign convention for Polchinski’s equation has the schematic structure

-\partial_t S_k^{\rm I} = \frac12 \frac{\delta S_k^{\rm I}}{\delta\phi} \cdot\dot C_k\cdot \frac{\delta S_k^{\rm I}}{\delta\phi} - \frac12 \operatorname{Tr} \left( \dot C_k \frac{\delta^2S_k^{\rm I}}{\delta\phi\,\delta\phi} \right),

where the dot denotes the derivative of the chosen covariance with respect to the declared flow parameter. Kernel and sign conventions vary; the page derives them from the regulated partition function. The first term joins two interaction vertices, while the second contracts two legs of one vertex. Polchinski’s original construction shows how cutoff independence yields a differential flow for an effective Lagrangian and supports perturbative renormalizability Polchinski 1984, §§ 2–4, pp. 271–293.

For the effective average action, add

\Delta S_k[\phi] = \frac12\phi\cdot R_k\cdot\phi

before taking a modified Legendre transform. With $t=\ln k/\Lambda_{\rm UV}$ , the bosonic Wetterich equation is

\boxed{ \partial_t\Gamma_k[\varphi] = \frac12 \operatorname{Tr} \left[ \left( \Gamma_k^{(2)}[\varphi]+R_k \right)^{-1} \partial_tR_k \right]. }

Fermions and ghosts require a supertrace with the appropriate statistics signs. The equation looks one-loop because it contains one trace, but it is exact: the inverse propagator uses the full running Hessian $\Gamma_k^{(2)}$ . Replacing that functional Hessian by one from a finite ansatz is the approximation. Wetterich derives the exact scale-dependent effective-action equation and its endpoints in Wetterich 1993, pp. 90–94.

These displayed equations are orientation formulas. Their pages fix the covariance, field normalization, statistics, measure, endpoints, and derivative signs before using them.

Exact parent, approximate projection

An exact functional equation contains infinitely many coupled momentum-dependent vertices. A calculation chooses a representation such as

\Gamma_k[\varphi] = \int d^dx \left[ U_k(\rho) +\frac12Z_k(\rho)(\partial_\mu\varphi)^2 +Y_k(\rho)(\partial_\mu\rho)^2 +\cdots \right], \qquad \rho=\frac12\varphi^2.

Keeping only $U_k$ is a local-potential approximation. Expanding $U_k$ as a finite polynomial, sampling it on a field grid, evaluating vertices at one momentum configuration, or holding $Z_k$ constant adds further choices. None follows from the word “exact.”

Every projected result in this chapter records:

Component	Question that must be answered
Regulator	What shape function and normalization define $R_k$ , and are the required UV and IR limits satisfied?
Ansatz	Which field, derivative, vertex, and momentum structures are retained or forbidden?
Projection	At which field value and momentum configuration is each running parameter extracted?
Closure	How are vertices or operators outside the ansatz treated when they appear on the right-hand side?
Symmetry	Which Ward, modified Ward, or Slavnov–Taylor identity is exact at finite $k$ , and what must be restored at the endpoint?
Numerics	What discretization, tolerance, Hessian-domain check, and stability test control the integration?
Benchmark	Which exact limit, perturbative coefficient, alternative formulation, or independent quadrature tests the result?
Enlargement	What changes when the polynomial order, grid, derivative order, vertex sector, projection point, or regulator family is varied?

Regulator variation explores one direction in approximation space. A stationary result under a regulator parameter can be encouraging, but the principle of minimal sensitivity is a diagnostic, not a proof of convergence. Dupuis and collaborators survey the derivative, vertex, and related approximation schemes and emphasize the need to control their domain in Dupuis et al. 2021, § 2, arXiv:2006.04853.

The scalar thread and its computational check

A $Z_2$ -symmetric scalar theory connects the chapter:

shell integration generates shifts of the vacuum, mass, quartic, and higher operators;
rescaling exposes their canonical contributions and loop corrections;
the Polchinski equation organizes the same contractions in a smooth-cutoff Wilson action;
the Wetterich equation flows an effective potential and its derivative corrections;
polynomial and grid projections test whether the chosen representation controls the potential; and
the endpoint result is compared across regulators and against an independent benchmark.

A reproducible calculation uses a zero-dimensional $Z_2$ integral for the decisive comparison. Exact quadrature supplies the reference free energy. Wetterich flows with matched ultraviolet data, a running vacuum term, two regulator choices, and nested polynomial projections expose five common failures: calling a truncation exact, omitting the vacuum term, mismatching initial data, crossing a Hessian singularity, and reporting more digits than the combined error permits.

Zero dimensions do not test momentum dependence or field-theory universality. They do test the Legendre transform, regulator endpoints, normalization, closure, projection, numerical integration, and error decomposition without an unknown exact answer.

Limits of the chapter’s methods

Stop when the operator set does not close. A shell step or functional trace that generates omitted symmetry-allowed structures has left the proposed truncation. Projecting them away is an approximation that must be varied or bounded.

Stop at a Hessian pole. If $\Gamma_k^{(2)}+R_k$ develops a zero eigenvalue outside a controlled convexity limit, the inverse trace is singular. Changing solvers does not repair an invalid domain.

Stop when symmetry restoration is untested. A regulator can modify a Ward or Slavnov–Taylor identity. A physical endpoint claim requires the modified identity at finite $k$ and restoration or controlled breaking as $k\to0$ .

Stop before interpreting regulator spread as a confidence interval. Regulator, ansatz, projection, numerical, and input errors probe different directions and can be correlated. A small regulator spread does not bound omitted operators.

Stop before calling a projected fixed point a complete theory. The next chapter tests stability, universality, and continuum evidence; conformal data and subject-specific phases require their own analyses.

Review the chapter

Capability	Prompt	Successful response and repair
Object identification	Compare $S_k$ , $\Gamma_k$ , and a renormalized 1PI action $\Gamma(\mu)$ .	State the field content, regulator, integrated or suppressed modes, boundary data, and endpoint for each. Repair in coarse graining.
Shell composition	Apply two thin shell steps to a scalar action.	Show mode integration plus rescaling, include induced operators, and recover the semigroup structure through the retained order. Repair in shell integration.
Functional derivation	Explain both terms in Polchinski’s equation.	Derive them by differentiating the Gaussian covariance and integrating by parts in field space. Repair in Polchinski flow.
Legendre-flow derivation	Explain why the Wetterich trace is exact despite its one-loop form.	Use the modified Legendre transform and full running Hessian, with regulator subtraction and endpoint stated. Repair in Wetterich flow.
Projection	Extract a quartic flow from a local potential.	Declare the field point, normalization, closure, and terms discarded before reading off the coefficient. Repair in truncations.
Error claim	A result changes little between two regulators. What follows?	Report that variation as one diagnostic, then add nested ansatz, projection, symmetry, numerical, and benchmark checks. Repair in symmetry and error control.

Continue from functional flow

Continue to Fixed Points, Universality, and Continuum Limits to define scaling directions, critical surfaces, crossover, and evidence-qualified continuum limits.
Continue to Conformal Field Theory and the Bootstrap for fixed-point operator data and conformal constraints.
Continue to Many-Body Quantum Matter for phase-specific functional-RG applications.
Continue to Lattice and Hamiltonian Field Theory for executable real-space or lattice blocking and nonperturbative step scaling.
Continue to Mathematical Quantum Field Theory for theorem-first exact-RG and continuum-trajectory results.
Return to Renormalization-Group Equations and Running when the problem is subtraction-scale evolution at fixed bare data, or to the Volume 5 overview to choose another scale operation.

References

Dupuis, Nicolas, Léonie Canet, Astrid Eichhorn, Walter Metzner, Jan M. Pawlowski, Michel Tissier, and Nicolás Wschebor. “The Nonperturbative Functional Renormalization Group and Its Applications.” Physics Reports 910 (2021): 1–114. DOI. Open PDF.
Polchinski, Joseph. “Renormalization and Effective Lagrangians.” Nuclear Physics B 231 (1984): 269–295. DOI.
Wetterich, Christof. “Exact Evolution Equation for the Effective Potential.” Physics Letters B 301 (1993): 90–94. DOI. Open PDF.
Wilson, Kenneth G., and John Kogut. “The Renormalization Group and the $\epsilon$ Expansion.” Physics Reports 12 (1974): 75–199. DOI.