---
title: Effective Average Action Overview
url: https://www.emergentmind.com/topics/effective-average-action
type: topic
---

# Effective Average Action Overview

The effective average action (EAA), usually denoted \(\Gamma_k\) or, in some applications, \(\Gamma_A\), is the scale-dependent effective action of the functional renormalization group. It is defined by adding an infrared regulator term to the action and performing a modified Legendre transform, so that \(\Gamma_k\) interpolates between the bare or classical action in the ultraviolet and the usual quantum effective action in the infrared. In the 1PI or “Wetterich” formulation, it is the scale-dependent generator of irreducible vertices and the central coarse-grained functional for exact renormalization-group flow equations [1108.1813] [1505.03119].

## 1. Definition and exact flow

A standard definition starts from the cutoff-modified generating functional
\[
e^{W_k[J]} \equiv \int \mathcal{D}\chi\, e^{-S-\Delta S_k + J\cdot \chi},
\qquad
\Delta S_k=\frac{1}{2}\int \chi\, R_k\, \chi,
\]
with a regulator kernel \(R_k\) chosen so that low-momentum modes are suppressed while high-momentum modes are essentially unaffected. In the usual formulation, \(R_k(z)\to 0\) for \(z\gg k^2\), \(R_k(z)\) gives a large mass-like suppression to low-momentum modes \(z\ll k^2\), and \(R_k(z)\to 0\) as \(k\to 0\). The EAA is then defined by subtracting the cutoff term after the Legendre transform,
\[
\Gamma_k \equiv \tilde\Gamma_k-\Delta S_k.
\]
In wetting theory the same construction appears in the notation
\[
\Gamma_A[l] = -W_A[h]+\int d^{d-1}x\, h(x)l(x)-\Delta H_A[l],
\]
with \(\Gamma_{A_0}[l]=H[l]\) at the ultraviolet scale and \(\Gamma_{A\to 0}[l]\) equal to the full Gibbs free energy or effective action [1108.1813] [1505.03119].

The exact flow is the Wetterich equation,
\[
\partial_t \Gamma_k = \frac{1}{2}\,\mathrm{STr}\!\left[\left(\Gamma_k^{(2)}+R_k\right)^{-1}\partial_t R_k\right],
\qquad t=\log(k/k_0),
\]
or, in the notation of the wetting paper,
\[
\partial_A \Gamma_A[l] =\frac{1}{2}\,\mathrm{Tr}\left[ \left(\Gamma_A^{(2)}[l]+R_A\right)^{-1} \partial_A R_A \right].
\]
This equation is exact and nonperturbative. Its one-loop form is deceptive: the nonperturbative content sits in the full inverse propagator \(\Gamma_k^{(2)}+R_k\). The ordinary effective action is recovered in the strict infrared limit, \(\Gamma_{k\to 0}=\Gamma_0\) [1108.1813] [1505.03119].

## 2. Truncations, locality, and momentum dependence

Practical calculations require a truncation of theory space. A standard derivative expansion writes
\[
\Gamma_k[\varphi] = \int_x \left[ U_k(\rho) + \frac{1}{2} Z_k(\rho)(\nabla\varphi)^2 + \cdots \right],
\]
while the wetting analysis uses
\[
\Gamma_A[l]=\int d^{d-1}x\left\{
U_A(l)+\frac{1}{2}Z_A(l)(\nabla l)^2+\cdots
\right\}.
\]
The simplest step is the local potential approximation or its close variants. In the wetting problem, the key approximation is \(Z_A(l)=\sigma\) for all \(A\), with higher gradient terms dropped, which reduces the flow to an equation for the effective potential \(U_A(l)\). The derivation makes the involved approximations transparent and is argued to be especially mild because in wetting the anomalous dimension is absent, \(\eta=0\) [1108.1813].

Beyond a strict derivative expansion, several works keep full momentum dependence in selected sectors. The nonlocal potential approximation for the \(O(N)\) model retains a full local potential \(U_\Lambda\), a fully momentum-dependent quadratic term \(\sigma_\Lambda(k)\), and a momentum-dependent nonlocal density-density interaction \(u_\Lambda(k)\), with all approximations done directly in the effective average action rather than in the flow equations of irreducible vertices. For crystalline phantom membranes, the nonperturbative renormalization group is formulated with a nonlocal effective average action whose running coupling functions \(\tilde\kappa_\Lambda(q)\), \(\tilde\mu_\Lambda(q)\), and \(\tilde\lambda_\Lambda(q)\) are functions of momentum, specifically to capture anomalous elasticity and crossover structure that a derivative expansion would miss [1206.6121] [1012.0313].

This architecture makes the EAA unusually flexible. It can support local truncations, nonlocal form factors, single-field truncations, and bi-field truncations, depending on the problem. A plausible implication is that the EAA formalism is best viewed not as one approximation scheme but as a regulated functional framework within which several approximation strategies coexist.

## 3. Nonlocal form factors and reconstruction of effective actions

A major use of the EAA is the reconstruction of known nonlocal effective actions by integrating the flow from the ultraviolet to the infrared. For a minimally coupled scalar field on a two-dimensional curved space, one may truncate the EAA as
\[
\Gamma_k[0,g] = \int d^2x\,\sqrt{g}\left( a_k+b_k R + R\,c_k(\Delta)\,R \right)+O(R^3),
\]
where the form factor \(c_k(\Delta)\) keeps the full nonlocal momentum dependence of the curvature-squared sector. Integrating the flow with \(c_\Lambda(x)=0\) yields
\[
c_0(x)=-\frac{1}{96\pi x},
\]
and therefore
\[
\Gamma[0,g] = -\frac{1}{96\pi}\int d^2x\,\sqrt{g}\,R\,\frac{1}{\Delta}\,R,
\]
which is exactly the Polyakov effective action [1004.2171].

In low-energy quantum gravity the same strategy is applied to a curvature expansion with nonlocal form factors,
\[
\bar{\Gamma}_{k}[g]=\int \mathrm{d}^{d}x\sqrt{g}\left[ \frac{1}{16\pi G_{k}}(2\Lambda_{k}-R) +R\,F_{1,k}(-\square)\,R +R_{\mu\nu}\,F_{2,k}(-\square)\,R^{\mu\nu} \right] +O(\mathcal{R}^{3}).
\]
Restricting to the one-loop flow and using the non-local heat kernel expansion, the infrared limit reproduces the familiar nonlocal one-loop effective field theory action, including logarithmic terms such as \(R\log(-\square/k_0^2)R\) and \(R_{\mu\nu}\log(-\square/k_0^2)R^{\mu\nu}\) [1006.3808].

The same computational logic extends to ordinary quantum field theory. The functional renormalization group can be used to reproduce the four point scattering amplitude in a real scalar field theory with quartic potential, the pion chiral lagrangian, the vacuum polarization of QED and of Yang–Mills theory, and two point functions for scalars and gravitons in the effective field theory of scalar fields minimally coupled to gravity. In these examples the ordinary effective action \(\Gamma_0\) is obtained by integrating the flow equation from an ultraviolet scale \(k=\Lambda\) down to \(k=0\), and the non-local heat kernel coefficients are the technical device that retains full momentum dependence rather than only local couplings [1505.03119].

## 4. Background fields, covariance, and gauge dependence

In gauge theories the EAA is often formulated through the background field method. One introduces a background field \(B\), adds a regulator term \(S_k(\phi,B)\), performs the Legendre transform, and defines the background effective action by setting the mean quantum fields to zero,
\[
\Gamma_k(B)=\Gamma_k(\Phi=0,B).
\]
With regulators chosen as functions of background-covariant operators, the background average effective action is invariant under background transformations, and the regulator action can be made background gauge invariant for a wide class of regulator functions. In the language of the background effective average action, this is the distinction between manifest background gauge invariance and the more delicate question of gauge-fixing dependence [1905.08296].

The explicit flow hierarchy for proper vertices of the background effective average action shows why the background-field dependence of the cutoff must be handled carefully. Differentiation with respect to background fields generates extra terms involving derivatives of the cutoff kernel, \(R_k^{(m)}[\bar A]\), and these terms are essential for preserving covariance of the background flow. This representation supports single- or bi-field truncations of local or non-local character and can be projected directly in momentum space, without relying only on a heat-kernel expansion [1304.2059].

A separate line of work addresses covariance and background independence in scalar field theories. There the modified splitting Ward identity can remain exactly solvable if the infrared cutoff is chosen so that
\[
\nabla_i R_{mn} = 0,
\]
allowing the scale-dependent effective average action to depend on the single total field rather than separately on background and fluctuation fields. In that setting, the flow remains manifestly covariant and truncations can be made single-field rather than double-field [1607.07074].

Gauge dependence remains a central limitation. The regulator term breaks BRST symmetry at finite \(k\), the standard Ward identity is modified, and preserving background gauge invariance does not eliminate gauge-fixing dependence on-shell. One analysis concludes that the dependence of the gauge fixing remains on-shell even when the symmetry of the average effective action is maintained for a wide class of regulator functions, while another states that at any scale of IR cutoff the effective average action depends on gauges, making impossible physical interpretation of all obtained results in that form of the method [1905.08296] [2004.00317].

## 5. Fixed points, conformal symmetry, and composite operators

At criticality the EAA admits fixed-point equations that are the Legendre-transformed counterparts of Wilsonian exact renormalization-group equations. Starting from a modified Polchinski equation, Morris’ fixed-point equation for the effective average action can be derived, and with it an explicit expression for the line of equivalent fixed-points associated with every critical fixed-point. Equivalent fixed points are related by quasi-local field redefinitions and correspond to redundant directions rather than distinct physics [1010.1530].

The fixed-point structure can be sharpened from scale invariance to full conformal invariance. A Legendre transform of the conformal fixed-point equation gives an unintegrated equation in the EAA framework that encodes the full conformal group, not merely the RG fixed-point condition. In this construction the scale equation and its special conformal partner combine into a single conformal fixed-point equation for \(\Gamma\), with explicit regulator dependence through \(\Gamma^{(2)}+R\) [1605.01729].

Scaling can also be formulated through a second evolution equation in the normalization scale \(\mu\), fully analogous to the Callan–Symanzik equation. In that approach the EAA satisfies
\[
\mu\frac{d}{d\mu}\Gamma_k[\varphi]=0
\]
along a fixed RG trajectory once the compensating changes of couplings and field normalization are included. Composite operators are introduced by adding sources \(\varepsilon\), and the inserted operator is represented by
\[
[O_k]\equiv \frac{\delta}{\delta\varepsilon}\Gamma_k[\varphi,\varepsilon].
\]
Their scaling dimensions follow from operator mixing matrices and anomalous-dimension matrices; in fixed-point regime they are eigenvalues of the appropriate mixing operator. In the local potential approximation, field-dependent composite operators without derivative mixing reduce to the same eigenvalue problem as the linearized flow of the potential [1603.07250].

## 6. Wetting transitions as a representative application

A particularly transparent application is the functional renormalization-group theory of wetting transitions. The starting point is the capillary-wave Hamiltonian
\[
H[l]=\int d^{d-1}x\left\{\frac{\sigma}{2}\,(\nabla l)^2+V(l)\right\},
\]
where \(l(\mathbf{x})\) is the local interface height above a planar wall, \(\sigma\) is the interfacial stiffness, and \(V(l)\) is the effective interfacial potential. Using the exact flow of the effective average action and the local truncation
\[
\Gamma_A[l]=\int d^{d-1}x\left\{
U_A(l)+\frac{1}{2}Z_A(l)(\nabla l)^2+\cdots
\right\},
\qquad
Z_A(l)=\sigma,
\]
the flow reduces to an equation for \(U_A(l)\). With a sharp cutoff, the momentum integral can be carried out explicitly and the continuous limit of the Lipowsky–Fisher nonlinear RG is recovered; expanding the logarithm further yields the older linear RG of Fisher and Huse [1108.1813].

This derivation is important because it makes every approximation explicit: locality of the effective action, retention of the local potential, neglect of stiffness renormalization, and omission of higher gradients. It also isolates a nontrivial scheme issue. The capillary parameter \(\omega\), which controls the linear RG theory of critical wetting, is scheme dependent below \(d=3\) because the coefficient of the curvature term depends on the regulator shape. At \(d=3\), however, the regulator dependence disappears: the relevant regulator integral reduces to a universal boundary term, and \(\omega\) is robust against scheme variation [1108.1813].

## 7. Terminological scope and common confusions

The expression “effective average action” is specific to the renormalization-group coarse-grained action \(\Gamma_k\). It should not be conflated with “Average Action Efficiency” (AAE), a distinct quantity defined for self-organizing stochastic systems by
\[
\alpha(t)=\frac{\eta}{\langle I\rangle_t},
\]
or, empirically,
\[
\alpha_{\mathrm{emp}}(t)=\frac{\eta\,\phi(t)}{Q(t)}.
\]
AAE is a variational, dynamical metric for open stochastic systems and not a field-theoretic effective action [2507.02209].

Nor is every “effective action” an effective average action. In interacting Bose–Einstein condensates, for example, the relevant object is a thermodynamic functional \(\Gamma[\vec\Psi,\hat G]\) obtained by a double Legendre transform with respect to the condensate and the Green’s function, while the finite-temperature QED effective action in a time-dependent electric field is a one-loop nonequilibrium in/out effective action for pair production. Both are effective actions, but neither is the FRG average effective action governed by the Wetterich equation [1405.5291] [1006.0774].

This distinction matters because the term “average” in the EAA is technical: it refers to the suppression of momentum modes below the running scale \(k\). The EAA is therefore best identified by its defining ingredients—a regulator term, a modified Legendre transform, and an exact flow equation—rather than by the generic phrase “effective action.”

Source: https://www.emergentmind.com/topics/effective-average-action