Gravitational Vertices and Power Counting

Einstein gravity contains infinitely many interaction vertices but only two derivatives at each vertex. Because $\kappa$ has mass dimension $-1$ in four dimensions, every additional loop raises the momentum order and requires new higher-derivative operators. That makes the theory nonrenormalizable in the old finite-coupling sense, yet systematically predictive as an EFT when $GE^2\ll1$.

Required background. Flat-Spacetime Quantum Field Theory supplies loop counting and counterterms. Weak-Field Gravity supplies the metric expansion. Gravitational Actions and Boundary Terms supplies the Einstein–Hilbert Lagrangian.

Helpful background. Einstein Equations supplies the classical limit, while Quantum Information supplies unitarity language.

The infinite vertex tower

With $g_{\mu\nu}=\eta_{\mu\nu}+\kappa h_{\mu\nu}$ and $\kappa^2=32\pi G$, the Einstein–Hilbert density is

$$\mathcal L_{\mathrm{EH}}=\frac{2}{\kappa^2}\sqrt{-g}\,R.$$

Expanding the inverse metric, determinant, and curvature gives schematically

$$\mathcal L_{\mathrm{EH}} \sim(\partial h)^2 +\kappa h(\partial h)^2 +\kappa^2h^2(\partial h)^2+\cdots.$$

After integrations by parts, every Einstein vertex has two derivatives. An $n$-graviton vertex therefore scales as

$$V_n\sim\kappa^{n-2}p^2.$$

The tensor expression is lengthy, but its coupling and momentum powers are fixed before any index algebra. Gauge fixing and ghosts add vertices with the same derivative counting needed for Ward identities.

Topological loop counting

Consider a connected graph with $L$ loops, $I$ internal graviton lines, $V$ Einstein vertices, and $E$ external gravitons. Four-dimensional loop measures contribute $p^{4L}$, propagators contribute $p^{-2I}$, and vertices contribute $p^{2V}$. The superficial momentum degree is

$$d_p=4L+2V-2I.$$

Using $L=I-V+1$ gives

$$d_p=2+2L.$$

The coupling power follows from $\sum_v n_v=2I+E$:

$$\sum_v(n_v-2)=E-2+2L.$$

Hence a generic $E$-graviton contribution scales as

$$\mathcal M_E^{(L)} \sim\kappa^{E-2}p^2 \left(\frac{\kappa^2p^2}{16\pi^2}\right)^L,$$

up to dimensionless functions, infrared singularities, and symmetry cancellations. With $\kappa^2=32\pi G$, the loop parameter is $\kappa^2E^2/(16\pi^2)\sim GE^2$; the proportionality contains convention-dependent numerical factors.

Several immediate cases make the formula concrete. At tree level, every multiplicity carries two net powers of momentum and $\kappa^{E-2}$, even though individual diagrams contain different numbers of cubic and higher vertices. At one loop, the generic local part carries four powers of momentum and two additional powers of $\kappa$. At two loops it carries six powers of momentum and four additional powers of $\kappa$. The topology identities guarantee that changing a quartic vertex into two cubic vertices joined by a propagator does not change the overall order.

When several invariants are present, $p$ is shorthand rather than a single number. A scattering amplitude depends on $s$, $t$, masses, and possibly background curvature. Power counting requires every independent dimensionless ratio in the claimed regime to be small, and forward limits can enhance propagator poles even when $Gs$ is small. The derivative count organizes local ultraviolet terms; it does not replace a separate infrared analysis of soft or collinear regions.

Counterterms by derivative order

A local ultraviolet divergence at $L$ loops has $2L+2$ derivatives. Diffeomorphism invariance packages it into curvature tensors and covariant derivatives. At one loop, the candidates include curvature-squared terms; in four-dimensional pure gravity, field redefinitions, the Euler density, and the vacuum Einstein equation make the on-shell divergence vanish, but matter or off-shell backgrounds remove that special cancellation ’t Hooft and Veltman 1974, §§3–6.

At two loops, six-derivative curvature-cubic terms are allowed. A nonzero on-shell divergence proportional to a cubic Riemann contraction was found explicitly Goroff and Sagnotti 1985, pp. 81–86. Its coefficient cannot be absorbed into $G$ or the cosmological constant. This is the decisive perturbative nonrenormalizability result for pure Einstein gravity.

In EFT, the response is not to discard the theory. Include every local diffeomorphism-invariant operator,

$$\mathcal L_{\mathrm{EFT}}= \frac{2}{\kappa^2}\sqrt{-g}\,R +\sqrt{-g}\left(c_1R^2+c_2R_{\mu\nu}R^{\mu\nu}+\cdots\right),$$

renormalize its coefficient, and truncate at a declared accuracy. At low momentum, only finitely many operators contribute at each order.

Checks, cancellations, and limitations

Superficial degree is an upper bound. Gauge symmetry, on-shell identities, topology, helicity, and supersymmetry can cancel candidate terms. A cancellation at one loop does not override the operator analysis at the next loop. Conversely, a divergence in an off-shell Green function can be moved by a field redefinition and need not represent a new on-shell observable.

The counting can be checked dimensionally. In four dimensions, $[\kappa]=-1$ and an $E$-point amplitude has mass dimension $4-E$ after the momentum-conserving delta function is removed. The expression $\kappa^{E-2+2L}p^{2+2L}$ has exactly that dimension. In the infrared limit $p\to0$, higher loops are suppressed; at $p\sim1/\sqrt G$, the hierarchy fails.

Matter lines change graph bookkeeping but not the EFT principle. Their masses and couplings introduce additional ratios, and heavy fields can be matched onto local curvature operators. Massless fields also produce nonanalytic momentum dependence that cannot be represented by local counterterms; those terms will be central to long-distance predictions.

Exercises

Exercise. A connected four-graviton graph uses only Einstein vertices and has two loops. Determine its overall power of $\kappa$, its superficial momentum degree, and the derivative order of the counterterm it can require.

Solution

Set $E=4$ and $L=2$. The coupling exponent is

$$E-2+2L=4-2+4=6,$$

so the graph scales as $\kappa^6$. Its superficial momentum degree is $2+2L=6$. A local divergence can therefore require a six-derivative counterterm, represented schematically by curvature-cubic operators because each curvature carries two derivatives. Special helicity or on-shell cancellations may reduce a particular amplitude, but power counting permits this counterterm class.

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.
• Goroff, Marc H., and Augusto Sagnotti. “Quantum Gravity at Two Loops.” Physics Letters B 160 (1985): 81–86. DOI.