Renormalization Ambiguities and Gravitational Counterterms

A renormalized stress tensor is not a unique tensor-valued function of a state and metric. Two prescriptions satisfying locality, covariance, conservation, the same field equation, and the regularity and scaling axioms may differ by a state-independent conserved curvature tensor. For a scalar field in four dimensions, these assumptions restrict the finite freedom to redefinitions of the cosmological, Newton, and curvature-squared couplings in the gravitational action.

Required background. Stress–Energy and Conservation provides metric variation and conserved Hilbert tensors. Curvature supplies the curvature invariants and four-dimensional identities used in the ambiguity basis.

Helpful background. Flat-Spacetime Quantum Field Theory provides ordinary coupling renormalization. Partial Differential Equations and Green Functions supplies the local parametrix viewpoint. Causality and Global Structure develops the background category. Quantum Fields in Curved Spacetime provides the common Hadamard prescription whose finite freedom is being compared.

The ambiguity as an action variation

For any scalar curvature density $F$, define the conserved bulk tensor

$$\mathcal H_{ab}[F] =-\frac{2}{\sqrt{-g}}\frac{\delta}{\delta g^{ab}} \int_M d^4x\,\sqrt{-g}\,F.$$

The sign matches the Hilbert stress-tensor convention on this page. Diffeomorphism invariance makes $\nabla^a\mathcal H_{ab}=0$ when boundary terms vanish under the allowed compact metric variations.

Require the usual almost-homogeneous scaling and regular dependence on $m^2$ through $m^2=0$, as well as local covariance and conservation. These assumptions exclude arbitrary inverse-mass towers of higher-curvature tensors; covariance and conservation alone would not do so. The difference between two four-dimensional scalar-field prescriptions then has the form

$$\Delta\langle T_{ab}\rangle =A\,m^4g_{ab}+B\,m^2G_{ab} +C\,\mathcal H_{ab}[R^2] +D\,\mathcal H_{ab}[R_{cd}R^{cd}],$$

up to a change of basis and dimension-specific identities. The constants $A,B,C,D$ belong to the renormalization prescription, not to the state. A Riemann-squared term can replace one quadratic basis element, but in four dimensions the Euler combination relates the three curvature-squared variations. Décanini and Folacci 2008, §§II–IV develop the arbitrary-dimensional Hadamard stress tensor and the corresponding counterterm classification.

Which gravitational coupling absorbs which term

Write a local gravitational action schematically as

$$S_{\mathrm{grav}} =\int_M d^4x\,\sqrt{-g} \left[ -\rho_\Lambda+\frac{R}{16\pi G} +\alpha R^2+\beta R_{ab}R^{ab} \right].$$

Changing $\rho_\Lambda$ shifts the metric term, changing $G^{-1}$ shifts the Einstein tensor, and changing $\alpha,\beta$ shifts the curvature-squared tensors. Thus moving a finite local term from $\langle T_{ab}\rangle$ to the left side of a semiclassical metric equation is a redefinition of a gravitational coupling. Only the sum of the two sides, with a declared convention and measured renormalized couplings, has invariant predictive content.

The comparison has three useful axes:

Stress-tensor ambiguity	Action coupling	Discriminating feature
$m^4g_{ab}$	cosmological term	survives on flat spacetime unless fixed by vacuum normalization
$m^2G_{ab}$	inverse Newton coupling	has two derivatives of the metric and vanishes on Ricci-flat backgrounds
curvature-squared variation	higher-curvature couplings	has four metric derivatives and affects the scheme-dependent $\Box R$ trace term

The engineering dimensions match: every entry has mass dimension four, the dimension of stress–energy in four spacetime dimensions. Conservation follows from action variation rather than from a component-by-component cancellation.

Four-dimensional identities and scheme dependence

The Euler density

$$E_4=R_{abcd}R^{abcd}-4R_{ab}R^{ab}+R^2$$

has a topological integral on a closed four-manifold. Its bulk metric variation therefore supplies no independent local ambiguity, although boundaries require additional terms and spoil a naive dismissal. Similarly, $\int\sqrt{-g}\,\Box R$ is a boundary term under the compact-support assumptions used here. These facts reduce the bulk quadratic basis; they do not say the invariants vanish pointwise.

A finite $R^2$ counterterm changes the trace by a multiple of $\Box R$. This is why the total-derivative part of the four-dimensional trace anomaly is scheme dependent. It cannot shift the Euler or Weyl-squared anomaly coefficients in the same way. The general local-covariance theorem constraining these redefinitions is Hollands and Wald 2001, §5.1, Theorem 5.1, and §5.2, PDF.

The sign and normalization can be checked by varying this one term explicitly. With our definition of $\mathcal H_{ab}$,

$$\begin{aligned} \mathcal H_{ab}[R^2] ={}&g_{ab}R^2-4RR_{ab} +4\nabla_a\nabla_bR-4g_{ab}\Box R,\\ g^{ab}\mathcal H_{ab}[R^2]&=-12\Box R. \end{aligned}$$

The contracted Bianchi identity and $\Box\nabla_bR-\nabla_b\Box R=R_b{}^c\nabla_cR$ make its divergence vanish. This independently checks both the trace shift and the derivative signs. On a four-dimensional maximally symmetric metric, $R_{ab}=Rg_{ab}/4$ and $\nabla_aR=0$, so this entire tensor vanishes; nonzero curvature alone is insufficient to measure its coefficient.

State dependence is a different axis. If two Hadamard states are compared using one prescription, every tensor in the displayed ambiguity cancels. A state-dependent flux, thermal density, or polarization cannot be absorbed into fixed gravitational couplings merely because it is finite and conserved.

Checks and boundaries of the comparison

An ambiguity basis should pass conservation, dimension, and independence checks. Vary its defining actions to establish conservation; verify mass dimensions without using inverse curvature scales; then apply four-dimensional Gauss–Bonnet identities to remove redundancy. On a maximally symmetric background, all conserved local rank-two tensors collapse to multiples of $g_{ab}$, so that geometry cannot determine all four coefficients independently. Less symmetric backgrounds are needed to discriminate the basis.

The classification assumes a smooth manifold without untreated boundaries and a fixed field content. Boundary counterterms, defects, gauge fields, and other dimensions enlarge or change the basis. Nor does this fixed-background classification solve the metric response: choosing renormalized couplings and evolving a self-consistent geometry belongs to semiclassical gravity.

Exercise

Why can a measurement of $\langle T_{ab}\rangle$ in Minkowski vacuum fix the coefficient of $m^4g_{ab}$ but not the coefficient multiplying $\mathcal H_{ab}[R^2]$?

Solution

Minkowski spacetime has vanishing curvature, so the curvature-squared variation vanishes there. Requiring the Minkowski vacuum stress tensor to be zero can therefore fix the surviving constant metric term, subject to the declared cosmological normalization, but it supplies no information about $C$. Determining a curvature-squared coefficient requires data or a normalization condition on a curved background where its tensor is nonzero and distinguishable from the other conserved terms.

References

• Décanini, Yves, and Antoine Folacci. “Hadamard Renormalization of the Stress-Energy Tensor for a Quantized Scalar Field in a General Spacetime of Arbitrary Dimension.” Physical Review D 78 (2008): 044025. DOI. Open PDF.
• Hollands, Stefan, and Robert M. Wald. “Local Wick Polynomials and Time Ordered Products of Quantum Fields in Curved Spacetime.” Communications in Mathematical Physics 223 (2001): 289–326. DOI. Open PDF.
• Wald, Robert M. “Trace Anomaly of a Conformally Invariant Quantum Field in Curved Spacetime.” Physical Review D 17 (1978): 1477–1484. DOI.