Perturbative Divergences of Quantum Gravity

Perturbative quantum gravity is nonrenormalizable in the traditional finite-coupling sense, yet it is predictive as an effective field theory. The crucial distinction is between a divergence removable by field redefinitions or lower-order equations of motion and a new on-shell invariant whose coefficient must be measured. In four-dimensional pure Einstein gravity with zero cosmological constant, the first genuine on-shell ultraviolet counterterm appears at two loops.

Required background. Flat-Spacetime Quantum Field Theory supplies dimensional regularization, loop counting, and effective operators; Gravitational Actions and Boundary Terms supplies the variational problem and the equivalence of terms proportional to the classical equations of motion.

Helpful background. Einstein Equations supplies the vacuum on-shell condition; Quantum Information helps separate physical states from gauge redundancy; and Quantum Gravity provides the comparison between effective predictions and ultraviolet completion.

Loop counting and the counterterm basis

Expand $g_{\mu\nu}=\bar g_{\mu\nu}+\kappa h_{\mu\nu}$ with $\kappa^2=32\pi G$. A connected $L$-loop amplitude in pure gravity has a superficial momentum order corresponding to $2L+2$ derivatives. Diffeomorphism invariance then requires local divergent terms built from curvature and covariant derivatives. Schematically,

$$\Gamma_{\mathrm{div}}^{(L)}\sim \frac{G^{L-1}}{\epsilon}\int d^4x\sqrt{-g}\; \sum_i c_i^{(L)}\,\mathcal O_i^{(2L+2)}, \qquad \epsilon=4-d.$$

The coefficients in this schematic formula are dimensionless. The derivative count follows directly from $D=4L-2I+2V$ and $L=I-V+1$, where $I$ and $V$ count internal propagators and two-derivative vertices: $D=2L+2$. The factor $G^{L-1}$ then supplies the mass dimension needed for a dimensionless action. Counting allowed terms does not prove that their coefficients are nonzero.

At one loop the candidates have four derivatives: $R^2$, $R_{\mu\nu}R^{\mu\nu}$, and $R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma}$. In four dimensions their Euler-density combination is topological up to a boundary term. On a vacuum Einstein background, $R_{\mu\nu}=0$, the remaining Ricci terms vanish or can be removed by a local metric redefinition. This is the celebrated on-shell cancellation found by ’t Hooft and Veltman 1974, pp. 69–72. It is a special property of pure gravity: generic matter loops produce curvature-squared divergences, and off-shell effective actions remain gauge- and parametrization-dependent.

The two-loop obstruction

At two loops the six-derivative basis contains a cubic Riemann invariant that survives the vacuum equations,

$$\mathcal O_{R^3} =R_{\mu\nu}{}^{\rho\sigma} R_{\rho\sigma}{}^{\alpha\beta} R_{\alpha\beta}{}^{\mu\nu}.$$

Goroff and Sagnotti computed a nonzero pole multiplying this structure Goroff and Sagnotti 1985, pp. 81–84. Unlike the one-loop Ricci terms, $\mathcal O_{R^3}$ is neither a four-dimensional topological density nor proportional to the lowest-order vacuum equations. Its coefficient therefore cannot be absorbed solely into $G$ and $\Lambda$, nor eliminated from on-shell scattering by a field redefinition. A new Wilson coefficient is required.

This establishes perturbative nonrenormalizability: continuing to higher loops demands an expanding tower of independent operators. It does not establish that quantum general relativity lacks meaning below the Planck scale. In effective field theory one writes every symmetry-allowed term,

$$S_{\mathrm{EFT}}=S_{\mathrm{EH}}+\int d^4x\sqrt{-g} \left(c_1R^2+c_2R_{\mu\nu}R^{\mu\nu}+d_1G R^3+\cdots\right),$$

renormalizes the coefficients at a scale $\mu$, and truncates according to $GE^2\ll1$. The analytic pieces depend on the $c_i$, while nonanalytic loop terms such as $Gq^2\ln[(q^2-i0)/\mu^2]$ encode parameter-free long-distance propagation Donoghue 2023, §§ 2–4. Here the principal logarithm has its cut on $q^2\lt0$, as required by the $(-,+,+,+)$ convention; static momentum transfer has $q^2=\lvert\mathbf q\rvert^2\gt0$.

What the comparison actually says

The relevant comparison axes are loop order, on-shell status, and field content. “Finite at one loop” refers only to pure Einstein gravity evaluated on vacuum solutions in four dimensions. It does not include external matter, a cosmological constant without further qualifications, off-shell Green functions, or higher loops. Conversely, “nonrenormalizable” means that infinitely many Wilson coefficients are needed for arbitrarily high energy accuracy; it does not mean that every low-energy amplitude is incalculable.

An independent check comes from dimensions. Since $[R]=2$ and $[G]=-2$, the Einstein term scales as $R/G$; a one-loop $R^2$ term is dimension four, and a two-loop term must carry $G R^3$ to give a dimension-four Lagrangian density. The sequence matches the $2L+2$ derivative count. A second check is the equivalence theorem: counterterms proportional to $R_{\mu\nu}$ can be shifted by $g_{\mu\nu}\mapsto g_{\mu\nu}+aR_{\mu\nu}+bg_{\mu\nu}R$, whereas the on-shell Weyl-cubic component remains.

Common pitfalls

Confusing an on-shell cancellation with ultraviolet completion. A vanished one-loop S-matrix pole is one order in one theory. The two-loop invariant answers the renormalizability question.

Treating a regulator pole as an observable. The pole determines running and counterterms; a physical amplitude combines it with renormalized coefficients and finite loop pieces, and must be evaluated in a stated kinematic regime.

Exercise

Show by dimensional analysis that an $L$-loop pure-gravity counterterm in four dimensions has $2L+2$ derivatives. Then identify the curvature power expected when no covariant derivatives act explicitly.

Solution

Each loop brings a factor of $\kappa^2\sim G$, of mass dimension $-2$. The action is dimensionless, so a local Lagrangian density multiplying $G^{L-1}$ must have dimension $4+2(L-1)=2L+2$. A curvature tensor carries two derivatives and mass dimension two, hence a derivative-free curvature monomial has $n=L+1$ factors: $R^{L+1}$. Thus $L=1$ permits $R^2$, while $L=2$ permits $G R^3$. Symmetry and identities decide which contractions survive; the counting alone does not prove a nonzero coefficient.

References

• ’t Hooft, Gerard, and Martinus Veltman (1974), “One-Loop Divergencies in the Theory of Gravitation,” Annales de l’Institut Henri Poincaré A 20, 69–94, Numdam open article.
• Donoghue, John F. (2023), “Quantum General Relativity and Effective Field Theory,” in Handbook of Quantum Gravity, arXiv:2211.09902.
• Goroff, Marc H., and Augusto Sagnotti (1985), “Quantum Gravity at Two Loops,” Physics Letters B 160, 81–86, DOI: 10.1016/0370-2693(85)91470-4.