Polchinski RG Equation

• Polchinski-type RG equations are exact Wilsonian formulations that feature a unique combination of quadratic and second-order functional derivative terms to ensure invariance of the partition function.
• They recast the renormalisation group flow into diffusion, stochastic, and transport frameworks, thereby providing clear probabilistic and physical interpretations.
• They support diverse reformulations—including Hamiltonian, holographic, and BV approaches—and enable robust fixed-point analyses and observable S-matrix generation.

Polchinski-type renormalisation group equations are exact Wilsonian flow equations for a scale-dependent action, effective interaction, or related functional. Their defining structural feature is the coexistence of a second-order functional derivative term and a quadratic term in first functional derivatives, arranged so that changing the cutoff leaves the partition function, or equivalently the low-energy physics, invariant. In generalized formulations the flow is written as a total functional derivative acting on $e^{-S_\Lambda[\phi]}$, while in probability-functional form it becomes a convection–diffusion equation in field space (Matsumoto et al., 2020, Cotler et al., 2022). Later work recast the same structure in Hamiltonian, holographic, Batalin–Vilkovisky, stochastic, transport, and $S$-matrix-generating languages (Akhmedov et al., 2010, Zucchini, 2017, Cotler et al., 2022, Freidel et al., 29 Aug 2025).

1. Functional form and Wilsonian meaning

A broad Wilsonian form is the exact flow for a scalar Wilsonian action $S_\Lambda[\phi]$ with cutoff function $C_\Lambda(p^2)$ and quadratic seed action $\hat S_\Lambda[\phi]$,

$$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$

In this Morris-type generalization, the seed action encodes the coarse-graining scheme, and quasi-locality is imposed by requiring a derivative expansion for $\hat S_\Lambda$ (Matsumoto et al., 2020).

A standard scalar specialization uses a smooth cutoff $K_\Lambda(p^2)$ in the Gaussian part of the action and yields Polchinski’s flow for the interaction functional,

$$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$

Equivalently,

$$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$

The second form makes explicit the functional-Laplacian character of the flow and its interpretation as diffusion in field space restricted to modes near $S$0 (Cotler et al., 2022).

In matrix scalar theory the same logic starts from cutoff independence of

$S$1

and produces a Polchinski equation for the interaction part $S$2 involving $S$3, functional derivatives with respect to matrix Fourier modes $S$4, and the usual quadratic-plus-second-derivative structure (Akhmedov et al., 2010). In all these formulations the flow implements infinitesimal shell integration while keeping the Wilsonian partition function invariant.

At the level of the normalized probability functional,

$S$5

Polchinski flow can be rewritten as a convection–diffusion equation,

$S$6

with a drift term proportional to $S$7 (Cotler et al., 2022). This suggests a direct probabilistic interpretation of exact Wilsonian coarse graining.

2. Scheme dependence, anomalous dimension, and exact RG families

A central refinement is the explicit incorporation of the anomalous dimension $S$8 into the exact flow. In the Wilsonian interaction representation $S$9, Osborn and Twigg obtain

$S_\Lambda[\phi]$0

with $S_\Lambda[\phi]$1 (Osborn et al., 2011). In this framework $S_\Lambda[\phi]$2 is introduced as a free parameter reflecting the freedom of exact RG equations up to contributions which vanish in the functional integral. Its exact value is not fixed kinematically; it is fixed by the requirement that there should exist a well defined non trivial limit at an IR fixed point (Osborn et al., 2011).

The same analysis identifies the determination of $S_\Lambda[\phi]$3 with the existence of an exact marginal operator. In the fixed-point theory $S_\Lambda[\phi]$4, an exact zero-mode redundant operator is constructed explicitly,

$S_\Lambda[\phi]$5

and it generates infinitesimal field rescalings at the fixed point (Osborn et al., 2011). In this sense, the marginal redundant direction organizes the line of equivalent fixed points associated with field normalization.

The Wilsonian equation and the Wetterich equation are related by a modified Legendre transform. Osborn and Twigg derive a 1PI flow

$S_\Lambda[\phi]$6

which is the Wetterich equation in dimensionless variables with the explicit $S_\Lambda[\phi]$7-term retained (Osborn et al., 2011). A related comparison appears in the 4PI truncation analysis, where the flow equations for $S_\Lambda[\phi]$8 are described as the 1PI, Wetterich-type representation of a Polchinski-like RG, with the standard mapping supplied by a modified Legendre transform (Carrington, 2013).

This family structure is also visible in generalized seed-action schemes. For arbitrary quasi-local $S_\Lambda[\phi]$9 and arbitrary quadratic seed $C_\Lambda(p^2)$0,

$C_\Lambda(p^2)$1

the exact RG still has Polchinski-type structure, but the detailed realization of diffusion, drift, and field rescaling depends on the pair $C_\Lambda(p^2)$2 (Matsumoto et al., 2020). This suggests that “Polchinski-type” denotes a structural class rather than a single unique equation.

3. Hamiltonian and holographic reformulations

In large-$C_\Lambda(p^2)$3 matrix scalar theory, the Polchinski equation can be reduced to Hamiltonian evolution in one higher dimension. For the subsector generated by single-trace operators $C_\Lambda(p^2)$4, the variables

$C_\Lambda(p^2)$5

form canonical pairs, and the RG flow becomes

$C_\Lambda(p^2)$6

The Hamiltonian is known exactly in this subsector,

$C_\Lambda(p^2)$7

and the RG time $C_\Lambda(p^2)$8 is defined from $C_\Lambda(p^2)$9 and the shell derivative of the regulated propagator (Akhmedov et al., 2010). Large-$\hat S_\Lambda[\phi]$0 factorization is what closes the flow on this canonical phase space.

A distinct geometric rewriting appears for the Wilson–Polchinski exact RG of free Majorana vector models in $\hat S_\Lambda[\phi]$1. There the bilocal sources for quadratic singlet operators are reorganized as a connection $\hat S_\Lambda[\phi]$2 and a section $\hat S_\Lambda[\phi]$3 on a jet bundle over a $\hat S_\Lambda[\phi]$4-dimensional RG space. The RG equations take the covariant form

$\hat S_\Lambda[\phi]$5

$\hat S_\Lambda[\phi]$6

so the beta functions become components of a bulk curvature (Leigh et al., 2014). In this construction a particular flat connection $\hat S_\Lambda[\phi]$7 realizes $\hat S_\Lambda[\phi]$8 geometry, and the passage to the corresponding principal bundle yields a structure strikingly similar to Vasiliev theory: the horizontal part of the connection is Vasiliev’s higher-spin connection, while the vertical part, interpreted as a Faddeev–Popov ghost, corresponds to the $\hat S_\Lambda[\phi]$9-field (Leigh et al., 2014).

These Hamiltonian and geometric reformulations do not change the Wilsonian content of the flow. They reorganize it. A plausible implication is that Polchinski-type equations provide a common language for Wilsonian RG, radial Hamilton–Jacobi evolution, and higher-spin gauge structure whenever the operator algebra closes sufficiently well.

4. BV, diffusion, stochastic, and transport perspectives

Within the Batalin–Vilkovisky formalism, the exact RG can be formulated as a flow between scale-dependent BV manifolds $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$0 with scale-dependent bracket, Laplacian, and effective action $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$1. Zucchini extends the scale line $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$2 to the shifted tangent bundle $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$3, introduces an odd partner $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$4, and shows that RG supersymmetry constrains the infinitesimal BV RGE to a Polchinski-type form,

$$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$5

The first two terms reproduce the characteristic second-order-plus-quadratic structure, while the remaining terms play the role of seed or inhomogeneous contributions (Zucchini, 2017). In the free $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$6 model this reduces to a seed-free Polchinski form exactly.

A different but closely related viewpoint identifies exact RG with generalized diffusion. For arbitrary cutoff function $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$7 and arbitrary quadratic seed $$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$8, the diffused field satisfies

$$-\Lambda \frac{\partial}{\partial\Lambda} e^{-S_\Lambda[\phi]} = \int_p C_\Lambda(p^2)\, \frac{\delta}{\delta\phi(p)}\left( \frac{\delta \hat S_\Lambda[\phi]}{\delta\phi(-p)} - \frac{1}{2}\frac{\delta}{\delta\phi(-p)} \right)e^{-S_\Lambda[\phi]}.$$9

with solution

$\hat S_\Lambda$0

and the correlation functions obey the exact identity

$\hat S_\Lambda$1

with $\hat S_\Lambda$2 (Matsumoto et al., 2020). In a specific scheme this reduces to the Sonoda–Suzuki relation for the ordinary heat equation.

A related gradient-flow construction replaces the bare action in the flow equation by the scale-dependent effective action $\hat S_\Lambda$3 and obtains

$\hat S_\Lambda$4

Abe and Fukuma argue that this can be regarded as an RG equation if one makes a field-variable transformation at every step such that the kinetic term is kept to take the canonical form (Abe et al., 2018).

The stochastic interpretation makes the Hamilton–Jacobi structure even more explicit. In the renormalised-potential language

$\hat S_\Lambda$5

the Polchinski semigroup $\hat S_\Lambda$6, backward stochastic differential equation, stochastic localisation, Föllmer process, Boué–Dupuis variational formula, and transport-of-measure viewpoint are all organized by the same scale-dependent potential $\hat S_\Lambda$7 (Bauerschmidt et al., 2023). Finally, in the optimal-transport formulation the entire Wegner–Morris class, including Polchinski’s choice, becomes

$\hat S_\Lambda$8

the Wasserstein-$\hat S_\Lambda$9 gradient flow of a field-theoretic relative entropy (Cotler et al., 2022). This yields a regularized relative entropy monotone along the RG flow.

5. Truncations, derivative structures, and closure schemes

Approximation theory for Polchinski-type equations is dominated by the problem of controlling infinitely many derivative and vertex couplings. One route is the usual derivative expansion. Another is the covariant Hamiltonian reformulation in terms of momentum fields of increasing rank. Starting from a local action

$K_\Lambda(p^2)$0

where $K_\Lambda(p^2)$1 denotes arbitrary spacetime derivatives, the generalized Legendre transform introduces tensor momenta $K_\Lambda(p^2)$2 and a Hamiltonian density $K_\Lambda(p^2)$3 (Zambelli, 2015). The expansion is then organized by the rank of the retained momenta fields rather than by the derivative order in the Lagrangian. Its first order, one next to the local potential approximation, is regulator-independent and already includes infinitely many derivative interactions (Zambelli, 2015).

At rank one, rotational symmetry reduces the dependence to

$K_\Lambda(p^2)$4

and the exact RG becomes a second-order nonlinear PDE in $K_\Lambda(p^2)$5 (Zambelli, 2015). Further truncating to

$K_\Lambda(p^2)$6

gives an alternative to the first order of the derivative expansion. In three dimensions this scheme yields

$K_\Lambda(p^2)$7

for the Ising universality class (Zambelli, 2015). The result is numerically close to high-temperature, Monte Carlo, and conformal bootstrap values quoted in the same work.

A different closure strategy uses $K_\Lambda(p^2)$8PI effective actions. In scalar $K_\Lambda(p^2)$9 theory, the 4PI effective action defines Bethe–Salpeter equations for 4- and 6-point functions. When these are inserted into the exact RG hierarchy, the truncated flows of the 2- and 4-point functions become total derivatives in the flow parameter and are equivalent to the 4PI equations of motion (Carrington, 2013). Although that analysis is carried out in a Wetterich representation, the paper states that the 4PI-based truncation can be understood as a non-trivial closure of the Polchinski-type hierarchy (Carrington, 2013).

The gradient-flow-based exact RG also admits an LPA and $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$0-expansion. In $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$1, Abe and Fukuma show that the eigenvalues of the linearized RG transformation around both the Gaussian and the Wilson–Fisher fixed points are reproduced to the order of $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$2 (Abe et al., 2018). This suggests that heat-kernel realizations of Polchinski-type flows can retain standard universal data even when their kernel differs from the usual $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$3 structure.

6. Fixed points, correlation functions, and observable-oriented flows

Polchinski-type equations are particularly effective near fixed points. In the fermionic setting, the scale-dependent interaction $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$4 satisfies

$$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$5

with an infinitesimal propagator $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$6 obeying a scaling law that fixes

$$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$7

For $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$8 and $$- \Lambda \frac{\partial S_{\text{int},\Lambda}[\phi]}{\partial \Lambda} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \Bigg[ \frac{\delta^2 S_{\text{int},\Lambda}}{\delta \phi(p)\,\delta \phi(-p)} - \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(p)} \frac{\delta S_{\text{int},\Lambda}}{\delta \phi(-p)} \Bigg].$$9, the quartic interaction scales as $$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$0 and is relevant, while the sextic term is irrelevant when $$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$1. The rigorous fixed-point analysis yields a weakly-interacting non-Gaussian fixed point with

$$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$2

more precisely

$$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$3

(Greenblatt, 2024). This is a controlled continuous-RG construction of a fermionic Wilson–Fisher-type fixed point.

In the renormalisation of composite-operator correlators, the Polchinski equation can be used geometrically on theory space. In a Wilsonian framework with local couplings, normal coordinates are defined so that the beta vector takes Poincaré or Poincaré–Dulac normal form near a fixed point. Normal correlators are then functional derivatives with respect to these parameters, and the renormalised correlators are given by the continuum limit of correlators associated to a cutoff-dependent parametrisation (Lizana et al., 2017). In a class of minimal subtraction schemes, the renormalised correlators are exactly equal to normal correlators evaluated at a finite cutoff (Lizana et al., 2017). This identifies a precise relation between Wilsonian flow, operator mixing, and standard renormalised correlation functions in scale-invariant theories.

A further development moves from off-shell generating functionals toward observables. For the $$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$4-matrix generating functional

$$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$5

the flow equation

$$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$6

has the same polynomial, Polchinski-type character as the Schwinger-functional equation, but it is built to generate scattering amplitudes more directly (Freidel et al., 29 Aug 2025). Compared to the Wetterich equation, the on-shell condition is simplified to the classical free equation

$$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$7

rather than a quantum equation of motion (Freidel et al., 29 Aug 2025). This suggests an observable-oriented exact RG in which polynomial flow structure and direct amplitude access coexist.

Across these applications, the unifying content of a Polchinski-type renormalisation group equation is not a particular regulator or truncation, but an exact Wilsonian flow with quadratic-plus-second-order functional structure, flexible enough to support local-coupling geometry, large-$$- \Lambda \frac{\partial}{\partial \Lambda} e^{-S_{\text{int},\Lambda}[\phi]} = \frac{1}{2} \int d^d p \,(2\pi)^d\, (p^2 + m^2)^{-1}\, \Lambda \frac{\partial K_\Lambda(p^2)}{\partial \Lambda}\, \frac{\delta^2}{\delta \phi(p)\,\delta \phi(-p)} \, e^{-S_{\text{int},\Lambda}[\phi]}.$$8 Hamiltonian dynamics, BRST/BV extensions, stochastic and transport formulations, nontrivial fixed-point constructions, and observable-generating generalizations.