Loops, Unitarity Cuts, and Counterterms

Gravitational loops contain both ultraviolet-sensitive local terms and universal long-distance terms. Unitarity cuts determine the discontinuities associated with on-shell intermediate states; dispersion then fixes logarithms and other nonanalytic momentum dependence up to subtraction polynomials. Those polynomials are precisely where local counterterms and unknown short-distance coefficients enter. The separation explains how a nonrenormalizable theory can still make parameter-free infrared predictions.

Required background. Flat-Spacetime Quantum Field Theory supplies loop analyticity and cutting rules. Weak-Field Gravity supplies the perturbative regime. Gravitational Actions and Boundary Terms supplies the local operator basis.

Helpful background. Einstein Equations supplies classical matching, while Quantum Information supplies unitarity and intermediate-state language.

Unitarity as a construction rule

Write $S=1+iT$. The identity $S^\dagger S=1$ implies

$$-i(T-T^\dagger)=T^\dagger T.$$

Between specified external states, the discontinuity of a loop amplitude across a physical branch cut equals a sum over products of lower-order on-shell amplitudes, integrated over Lorentz-invariant phase space. Schematically,

$$\operatorname{Disc}\mathcal M^{(1)} =i\sum_{\mathrm{states}} \operatorname{LIPS}\!\left[ \mathcal M_L^{(0)}\mathcal M_R^{(0)*} \right].$$

The sum includes only physical intermediate helicities in an on-shell formulation. In a covariant diagram calculation, ghosts and unphysical modes reproduce the same gauge-invariant result after cancellations. Cutkosky’s theorem provides the diagrammatic replacement of cut propagators by on-shell delta functions Cutkosky 1960, §§2–4.

A two-particle cut fixes the branch discontinuity but not every term in the amplitude. Dispersion reconstruction permits subtraction polynomials, and purely rational pieces can lack a four-dimensional branch cut. One therefore combines cuts with factorization, symmetries, dimensional regularization, or direct evaluation.

Local versus nonanalytic momentum dependence

A local contact vertex produces a polynomial in external momenta after expansion about flat space. Renormalization changes its coefficient and scheme. By contrast, massless propagation produces structures such as

$$\log\!\left(\frac{q^2-i0}{\mu^2}\right), \qquad \frac{1}{\sqrt{q^2-i0}},$$

whose branch points cannot be canceled by a local counterterm. The reference mass $\mu$ makes the logarithm dimensionless. Both functions use their principal branches, with the cut along negative real $q^2$: physical timelike thresholds have $q^2\lt0$, while a static transfer has $q^2=\mathbf q^2\gt0$ and the displayed functions are real. Their coefficients follow from the low-energy spectrum and couplings at the chosen EFT order. At higher orders this can include matched higher-derivative vertices inside light-particle loops; nonanalytic does not mean independent of every Wilson coefficient at all orders.

For a static three-dimensional transform, a polynomial in $\mathbf q^2$ gives delta functions and derivatives supported at the source. Nonanalytic terms generate power-law tails: $1/\lvert\mathbf q\rvert$ produces an inverse-square behavior and $\log\mathbf q^2$ an inverse-cube behavior, up to normalization and contact terms. Donoghue uses this analytic distinction to extract long-distance gravitational quantum corrections without knowing the ultraviolet completion Donoghue 1994, §§II–V.

“Universal” is conditional. Adding another massless particle changes the cut and its coefficient. Changing a heavy ultraviolet theory changes only analytic terms at momentum far below the heavy mass, after matching. The observable must also be defined carefully: a coordinate-space potential can change under field redefinitions, while a complete scattering amplitude is invariant.

Ultraviolet poles and counterterms

Dimensional regularization separates a loop amplitude into poles in $D-4$, logarithms of the renormalization scale, and finite terms. Local ultraviolet poles are canceled by curvature operators allowed by diffeomorphism invariance. The renormalized coefficient runs so that the full amplitude is scale independent.

At one loop, candidate curvature-squared operators appear. Pure four-dimensional gravity has a special on-shell cancellation, but matter and off-shell quantities generally require them. At two loops, pure gravity requires a curvature-cubic counterterm. These facts alter the subtraction polynomial; they do not erase the calculable nonanalytic part coming from massless cuts.

The method has a clear order: choose external states and a gauge-invariant amplitude; enumerate physical cuts; construct the cut integrand from trees; reduce it in a declared regulator; restore rational terms and local ambiguities; renormalize against a complete operator basis; and verify factorization, crossing, and scale independence. Skipping the operator-basis step can make a scheme-dependent local term look predictive.

Adverse checks and failure regimes

Change the covariant gauge, loop-momentum parametrization, and integral basis: the final on-shell discontinuity must remain fixed. Change the renormalization scheme: logarithms and local coefficients may be repartitioned, but the full amplitude and long-distance nonanalytic coefficient must agree. Add a small mass to an internal particle: the branch point moves and the large-distance tail becomes exponentially suppressed, showing that masslessness was essential.

The low-energy expansion fails when $GE^2$ approaches unity, a neglected threshold is crossed, large logarithms require resummation, or forward and soft singularities demand an inclusive observable. A cut calculation also assumes an ordinary asymptotic scattering problem; cosmological or finite-boundary settings need different observables and contour data.

Exercises

Exercise. Suppose a small-transfer amplitude contains

$$\mathcal M(q)=a+bq^2+cq^2\log\!\left(\frac{q^2-i0}{\mu^2}\right) +\frac{d}{\sqrt{q^2-i0}}.$$

Which coefficients can local counterterms change, and which terms necessarily encode long-range propagation? Explain what happens if the particle responsible for the branch cut is given a nonzero mass.

Solution

Local operators generate polynomials, so finite matching or a scheme change can shift $a$ and $b$. They cannot remove the branch point in the logarithm or square root; therefore $c$ and $d$ are fixed by the declared low-energy fields and couplings once the observable is normalized. The nonanalytic terms produce power-law spatial tails, whereas $a$ and $bq^2$ give contact terms and derivatives. Giving the cut particle a mass moves the branch point to its threshold. At distances much larger than the inverse mass, its contribution is exponentially suppressed and can be expanded into local operators.

References

• Cutkosky, Richard E. “Singularities and Discontinuities of Feynman Amplitudes.” Journal of Mathematical Physics 1 (1960): 429–433. DOI.
• Donoghue, John F. “General Relativity as an Effective Field Theory: The Leading Quantum Corrections.” Physical Review D 50 (1994): 3874–3888. DOI.