Renormalization Group and Local Couplings

On a curved background, renormalization-group flow is most cleanly defined by comparing local covariant theories under a constant rescaling of the metric. The comparison produces a local finite-renormalization map, generally containing curvature terms and derivatives of spacetime-dependent couplings. An ordinary beta function for a constant coupling is a projection of this richer object, not its definition.

Required background. Flat-Spacetime Quantum Field Theory supplies conventional running couplings and perturbative counterterms; Partial Differential Equations and Green Functions provides compact-source and support reasoning.

Helpful background. Causality and Global Structure establishes the causal regions on which local interactions are compared; Quantum Fields in Curved Spacetime supplies locally covariant fields, Hadamard subtraction, and prescribed-background scaling.

The local bridge is to regard each coupling as a smooth external source. Compact support makes every relative S-matrix meaningful without turning an interaction on forever. Equality of two sources on a neighborhood of an observable is enough to compare their local algebras, so no global Fourier transform or translation symmetry is assumed.

Local couplings retain spacetime information

For a real scalar field in four dimensions, write a representative interaction as

$$V_f=-\int_M d\mathrm{vol}_g\, f(x)\frac{\lambda}{4!}\phi^4(x), \qquad f\in C_0^\infty(M).$$

The minus sign is the action convention for the stable potential $+\lambda\phi^4/4!$ with $\lambda\gt0$; it agrees with $S(V)=\mathcal T e^{iV}$ and the vertex $-i\lambda$. The source $f$ is not merely an infrared convenience. Treating it as a genuine external field exposes which finite redefinitions are local: at a point $x$, a counterterm may depend on finitely many derivatives of $f$, on the metric and curvature jets, and on the field jet at $x$, but not on the values of $f$ at distant points.

Suppose two admissible prescriptions for time-ordered products define local S-matrices $S$ and $\widehat S$. Their difference can be encoded by a Stückelberg–Petermann map $Z$ on local functionals,

$$\widehat S(F)=S\bigl(Z(F)\bigr), \qquad Z(0)=0,\qquad Z^{(1)}(0)=\mathrm{id}.$$

The map is a formal power series and satisfies locality/additivity, covariance, reality, and the appropriate field-independence conditions. It describes a general finite change of prescription. Calling $Z$ “the beta function” would discard both its nonlinear functional character and its dependence on local source jets.

Metric scaling selects a cocycle

Let $g\mapsto \rho^2g$ with $\rho>0$, and scale the mass as $m\mapsto \rho^{-1}m$ and a scalar field of engineering dimension one as $\phi\mapsto\rho^{-1}\phi$. Transport the time-ordered products on $(M,\rho^2g)$ back to $(M,g)$ by the corresponding scaling isomorphism $\sigma_\rho$. The transported prescription and the original prescription obey the same axioms, so a finite local map $Z_\rho$ relates them:

$$\sigma_\rho^{-1}S_{\rho^2g}(\sigma_\rho F)=S_g\bigl(Z_\rho(F)\bigr).$$

Applying first $\rho'$ and then $\rho$ yields a cocycle relation

$$Z_{\rho\rho'}=Z_\rho\circ\mathfrak s_\rho(Z_{\rho'}),$$

where $\mathfrak s_\rho$ denotes the induced scaling action on finite-renormalization maps. It becomes an ordinary one-parameter group only when that action is trivial in the chosen coordinates on theory space. This distinction matters in curved spacetime because dimensionful parameters and curvature couplings themselves scale.

Hollands and Wald construct the interacting algebra locally, compare renormalization prescriptions, and derive the scaling isomorphism and induced interaction shift in Hollands–Wald 2003, §§3.1–3.2 and §4.1. Their result is coefficientwise in formal perturbation theory.

Deriving the local scale generator

Differentiate the cocycle at unit scale:

$$B(F)=\left.\rho\frac{d}{d\rho}Z_\rho(F)\right|_{\rho=1}.$$

Locality and dimension restrict $B(V_f)$. Schematically, for the scalar example,

$$\begin{aligned} B(V_f)=\int_M d\mathrm{vol}_g\bigg[& b_4(\lambda f)\frac{\phi^4}{4!} +b_z(\lambda f)\frac{(\nabla\phi)^2}{2}\\ &+b_m(\lambda f)\frac{m^2\phi^2}{2} +b_\xi(\lambda f)\frac{R\phi^2}{2}\\ &+\mathcal C(\lambda f,\nabla(\lambda f),\ldots;\phi,g)\bigg]. \end{aligned}$$

The last term includes derivative-of-source operators, which can themselves contain fields, as well as identity-valued curvature terms. The coefficient functions depend on the local coupling $\lambda f$, not generically on $\lambda$ times a single $f$. For example, the one-loop four-point divergence joins two quartic vertices and is proportional to $\lambda^2f(x)^2\phi^4(x)$. Replacing this by $\lambda^2f(x)\phi^4(x)$ is valid only inside a region where $f=1$. Integrations by parts can move derivatives among the source and the fields, so an operator basis must be fixed before numerical coefficients are compared.

For the interacting algebra, the finite scale change can equivalently be written as a shift of the interaction Lagrangian by

$$\delta\mathcal L_I(\rho)=-\frac12\left[ \delta z(\rho)(\nabla\phi)^2+\delta m^2(\rho)\phi^2 +\delta\xi(\rho)R\phi^2+\frac{\delta\lambda(\rho)}{12}\phi^4 \right]d\mathrm{vol}_g,$$

Here the $\delta$ coefficients denote increments of the parameters in the mostly-plus action, hence the overall minus sign. Identity-valued terms and local source derivatives have been suppressed in this constant-source display. The associated algebras are related by a $*$-isomorphism. This is the curved-background renormalization-group statement, including field-strength and curvature-coupling flow.

From the local cocycle to a flat beta function

To recover the familiar constant-coupling quantity, take a flat region and choose $f$ equal to one on a neighborhood of the local observables. There $R=0$ and derivatives of $f$ vanish. Reading the local $-\phi^4/4!$ coefficient gives the flow of $\lambda$, conventionally written in terms of the momentum scale as

$$\beta_\lambda(\lambda)=\mu\frac{d\lambda}{d\mu}.$$

With the standard four-dimensional normalization, the one-loop flat result is

$$\beta_\lambda=\frac{3\lambda^2}{16\pi^2}+O(\lambda^3).$$

The factor three counts the crossed one-loop four-point channels. This ultraviolet coefficient can be determined locally; it does not require a global adiabatic limit or a scattering matrix. There is also a sign check: $\mu$ is an inverse length, whereas $g\mapsto\rho^2g$ increases lengths. Thus $d\lambda/d\ln\rho=-\beta_\lambda$ after the flat identification. A raw coefficient in $B(V_f)$ also carries the sign of the displayed action interaction and must be translated before it is identified with $d\lambda/d\ln\rho$.

Extending this local comparison to a global constant-coupling state or S-matrix is a separate question: in a massless theory the required adiabatic limit may fail. Boundaries and curved backgrounds can preserve derivative-of-source or geometric terms, and observables may mix with curvature operators. The local cocycle remains meaningful when a single constant-coupling beta function is incomplete.

Redundant directions and fixed points

Flow in the coefficients of a Lagrangian is not automatically physical flow. A local field redefinition can move parameters without changing the interacting algebra. Hollands and Wald therefore separate essential from inessential parameters. For their scalar model, Hollands–Wald 2003, §4.1, Theorem 4.1 constructs a family of interaction Lagrangians and a $*$-isomorphism implementing the corresponding field rescaling.

A fixed point is consequently a zero of beta functions transverse to redundant field-redefinition directions, not necessarily a point where every displayed coefficient stops changing. This qualification is already necessary in flat space and becomes more visible when $R\phi^2$, field-strength, and mass terms participate in the curved-space flow.

The theorem assumes local covariant time ordering, smooth compactly supported couplings, a globally hyperbolic prescribed metric, and formal perturbation theory. It does not establish convergence, a global interacting vacuum, or metric backreaction. Nor does it say that all spacetime-dependent coupling profiles represent new physical parameters; some differences are switching artifacts or redundant directions.

Solved diagnostic: length scale versus momentum scale

Suppose the flat-space momentum-scale equation is $\mu\,d\lambda/d\mu=3\lambda^2/(16\pi^2)$ at one loop. If $\rho$ rescales all lengths upward and $\mu=\mu_0/\rho$, find $d\lambda/d\ln\rho$ and explain which additional terms must vanish before this is a legitimate reduction of the local curved-space flow.

Solution

Because $\ln\mu=\ln\mu_0-\ln\rho$, the chain rule gives $d\lambda/d\ln\rho=-3\lambda^2/(16\pi^2)$ to this order. The observation region must lie where $f=1$, derivatives of the source must vanish, and curvature terms must be absent or separately projected. These conditions determine the local ultraviolet coefficient. Existence of a global adiabatic limit is needed only for an additional global constant-coupling construction, not for extracting that coefficient. Away from the constant-source region, powers such as $f^2$ and derivative-of-source operators cannot be omitted.

References

• Brunetti, Romeo, and Klaus Fredenhagen. “Microlocal Analysis and Interacting Quantum Field Theories: Renormalization on Physical Backgrounds.” Communications in Mathematical Physics 208 (2000): 623–661. DOI.
• Hollands, Stefan, and Robert M. Wald. “On the Renormalization Group in Curved Spacetime.” Communications in Mathematical Physics 237 (2003): 123–160. DOI.