One-Loop Renormalization of Gravity

One-loop gravity illustrates both sides of EFT renormalization. Four-dimensional pure Einstein gravity has a special on-shell cancellation at this order, yet generic off-shell quantities, matter couplings, a cosmological background, and higher loops require new covariant counterterms. The cancellation is real but narrow; EFT predictivity comes from power counting and renormalized coefficients, not from pretending the operator tower is absent.

Required background. Flat-Spacetime Quantum Field Theory supplies dimensional regularization and renormalization. Gravitational Actions and Boundary Terms supplies the covariant operator basis.

Helpful background. Einstein Equations supplies on-shell conditions, Quantum Information supplies unitarity language, and Quantum Gravity supplies the EFT regime.

The one-loop divergence

Quantize metric fluctuations about a smooth background, fix a covariant gauge, include Faddeev–Popov ghosts, and regulate in $D=4-2\epsilon$ with $\epsilon\gt0$. In the convention where the pole below is the divergent contribution to the one-loop effective Lagrangian, pure gravity gives

$$\mathcal L^{(1)}_{\mathrm{pole}} =+\frac{\sqrt{-g}}{16\pi^2\epsilon} \left(\frac{1}{120}R^2+\frac{7}{20}R_{\mu\nu}R^{\mu\nu}\right).$$

The sign reverses for the counterterm added to the bare action. ’t Hooft and Veltman define $1/\varepsilon_{\mathrm{TV}}=1/[8\pi^2(D-4)]$ and print the cancelling counter-Lagrangian. Since $D-4=-2\epsilon$, their expression is

$$\mathcal L_{\mathrm{ct}} =-\frac{\sqrt{-g}}{16\pi^2\epsilon} \left(\frac{1}{120}R^2+\frac{7}{20}R_{\mu\nu}R^{\mu\nu}\right) =-\mathcal L^{(1)}_{\mathrm{pole}}.$$

This translation follows their regulator definition and pure-gravity result ’t Hooft and Veltman 1974, Eqs. (3.3) and (5.24), pp. 73, 87. Using $D=4-\epsilon$ instead would put $1/(8\pi^2\epsilon)$ in the pole magnitude; mixing that dimension convention with $1/(16\pi^2\epsilon)$ creates a factor-of-two error. Individual off-shell coefficients can change under field redefinitions and gauge choices; the need for a complete local basis and the on-shell conclusion do not.

For vacuum Einstein solutions with zero cosmological constant,

$$R_{\mu\nu}=0, \qquad R=0,$$

so the displayed divergence vanishes on shell. A possible Riemann-squared term can be exchanged for Ricci terms plus the four-dimensional Euler density, whose bulk variation is topological on a closed manifold with the proper boundary qualification.

Why the cancellation is limited

The comparison is best made by domain.

Calculation	One-loop conclusion	Limitation
Pure gravity, flat-space on-shell amplitudes	Curvature-squared ultraviolet divergence vanishes	Special to four-dimensional vacuum Einstein shell at one loop
Pure gravity, off-shell effective action	Curvature-squared counterterms appear	Coefficients can be gauge and field-parametrization dependent; off-shell action is not itself an observable
Gravity with matter	Matter and mixed curvature operators renormalize	Coefficients depend on field content, masses, and nonminimal couplings
Nonzero cosmological constant or boundary	Additional invariants and boundary terms contribute	Ricci-flat simplifications no longer apply
Pure gravity at two loops	A nonzero curvature-cubic on-shell divergence occurs Goroff and Sagnotti 1985, pp. 81–86	Requires a new six-derivative coefficient; the raw dimensional-regulator pole coefficient has evanescent-operator qualifications

Thus one-loop ultraviolet finiteness of a restricted on-shell amplitude does not imply perturbative renormalizability; massless soft-graviton infrared divergences still require an inclusive observable or appropriate dressed states. The next local operator order must also be tested, and the two-loop ultraviolet result is nonzero. Evanescent operators and dimensional-regularization choices can shift the printed $R^3$ pole coefficient without changing the renormalized four-dimensional amplitude’s scale dependence Bern et al. 2017, §§I–II. A six-derivative coupling remains necessary. The controlled interpretation is a low-energy EFT Donoghue 2023, §§4–5.

Renormalized coefficients

Include

$$\mathcal L_{4\partial}=\sqrt{-g} \left[c_1(\mu)R^2+c_2(\mu)R_{\mu\nu}R^{\mu\nu}\right].$$

Bare coefficients contain counterterm poles. Writing the displayed loop-pole coefficients as $b_1=1/120$ and $b_2=7/20$ gives

$$c_i^{\mathrm{bare}}= \mu^{-2\epsilon}\left[c_i(\mu)+\delta c_i\right], \qquad \delta c_i=-\frac{b_i}{16\pi^2\epsilon}.$$

The minus sign cancels the positive loop pole defined above. If one instead defines the displayed object as the counter-Lagrangian, both named signs reverse together. The running of $c_i(\mu)$ cancels explicit $\log\mu$ terms in amplitudes. Finite parts are set by measurement or matching and are not predicted by Einstein gravity alone.

At any fixed order in $E/\Lambda$, only finitely many $c_i$ enter. Once they are known, the EFT predicts all other observables through that order, including nonanalytic terms that no local coefficient can change. This is the operational meaning of order-by-order renormalizability.

Field-redefinition check

Consider a local redefinition

$$g_{\mu\nu}\longrightarrow g_{\mu\nu} +aR_{\mu\nu}+b g_{\mu\nu}R.$$

To first order in $a,b$, the change in the Einstein action is the leading Einstein equation contracted with the metric shift, plus a boundary term. It therefore moves $R_{\mu\nu}R^{\mu\nu}$ and $R^2$ coefficients without changing an on-shell S-matrix. This independently explains why the pure Ricci-flat divergence is removable.

The check fails if one uses the leading vacuum equation in a problem with matter, keeps off-shell Green functions as the claimed observable, ignores transformed matter operators, or neglects a physical boundary term. Field redefinitions preserve observables only when the entire calculation is transformed consistently.

Exercises

Exercise. Evaluate the displayed one-loop divergence on a four-dimensional Ricci-flat background. Does the result prove that pure Einstein gravity is finite at all loop orders? Explain the EFT conclusion.

Solution

Ricci flatness sets $R_{\mu\nu}=0$ and hence $R=0$, so both displayed invariants vanish. Any remaining curvature-squared bulk term is related to them by the Euler density and does not change the closed-manifold on-shell amplitude. This proves only the special one-loop on-shell cancellation. At two loops a curvature-cubic invariant is allowed and has a nonzero divergence. EFT includes its coefficient, so low-energy predictions remain controlled with additional powers of $GE^2$ rather than becoming parameter free at every order.

References

• ’t Hooft, Gerard, and Martinus Veltman. “One-Loop Divergencies in the Theory of Gravitation.” Annales de l’Institut Henri Poincaré A 20 (1974): 69–94. Open scan.
• Bern, Zvi, Huan-Hang Chi, Lance Dixon, and Alex Edison. “Two-Loop Renormalization of Quantum Gravity Simplified.” Physical Review D 95 (2017): 046013. DOI. Open manuscript.
• Donoghue, John F. “Quantum General Relativity and Effective Field Theory.” 2023. arXiv:2211.09902.
• Goroff, Marc H., and Augusto Sagnotti. “Quantum Gravity at Two Loops.” Physics Letters B 160 (1985): 81–86. DOI.