Strong Coupling and Predictive Cutoffs

Strong coupling begins when interactions of canonically normalized fluctuations become as important as their quadratic dynamics. A predictive cutoff is the lowest scale at which the chosen EFT loses controlled accuracy, whether through strong coupling, omitted heavy states, a ghost pole, rapid background variation, or failure of its derivative expansion. These scales can depend on the background and branch. A large classical field therefore need not imply uncontrolled fluctuations, while a small-looking covariant coefficient can become dangerous after a vanishing kinetic term is canonically normalized.

Required background. Variational and Symplectic Geometry supplies the reduced physical variables and kinetic form, while Ordinary Differential Equations supplies the scale and asymptotic comparisons used for background profiles.

Helpful background. Einstein Equations provide the controlled GR limit, Gravitational Actions and Boundary Terms fix the action and admissible data, and Formulations of General Relativity clarify which gauge-fixed helicity variables are canonically normalized.

Canonical normalization sets the interaction scale

Suppose one reduced mode has quadratic action

$$S_2=\frac12\int d^4x\,Z\left[\dot q^2-c_s^2(\nabla q)^2\right],$$

with slowly varying $Z\gt0$. The canonical field is $q_c=Z^{1/2}q$. An interaction written schematically as

$$\mathcal L_n=\frac{c_n}{\Lambda^{d_n-4}}\mathcal O_{d_n}(q)$$

must be rewritten in $q_c$ before its size is judged. If $\mathcal O_{d_n}$ contains $n$ powers of $q$, its coefficient acquires $Z^{-n/2}$. The dimensionless expansion parameter at characteristic energy $E$ is then built from $c_n$, $E/\Lambda$, $Z$, the sound speeds, and any background tensors. Strong coupling occurs when at least one such parameter reaches order unity.

This calculation belongs to a specified action expanded about a specified background, in a chosen gauge-fixed formulation and solution branch. It is valid only inside the EFT regime used to truncate the action. Canonical normalization can change between branches even when the covariant operator coefficients are identical.

The cubic Galileon in vacuum

With the site signature, take the decoupling-limit scalar convention

$$\mathcal L=-\frac12(\partial\pi)^2 -\frac{1}{2\Lambda_3^3}(\partial\pi)^2\Box\pi +\frac{\pi T}{M_{\mathrm{Pl}}}.$$

The cubic operator has mass dimension seven, so its coefficient is suppressed by $\Lambda_3^3$. Around the trivial vacuum, a mode with energy $E$ probes the dimensionless derivative ratio $(E/\Lambda_3)^3$ at a cubic vertex. Vacuum perturbation theory therefore ceases to be parametrically ordered near $E\sim\Lambda_3$. This is the familiar low strong-coupling scale of the helicity-zero sector described in de Rham 2014, § 8.3 and § 10.

The conclusion is not “the theory predicts nothing at distances shorter than $\Lambda_3^{-1}$.” A sourced background may reorganize perturbation theory. What fails is the expansion around the wrong vacuum. One must solve for the nonlinear background first and then normalize the fluctuations on that branch.

Background kinetic redressing

Write $\pi=\bar\pi+\varphi$. To quadratic order, integrations by parts give

$$\mathcal L_{\varphi}^{(2)} =-\frac12Z^{\mu\nu}(\bar\pi)\partial_\mu\varphi\partial_\nu\varphi,$$

with

$$Z^{\mu\nu}=\eta^{\mu\nu} +\frac{2}{\Lambda_3^3} \left[(\Box\bar\pi)\eta^{\mu\nu} -\partial^\mu\partial^\nu\bar\pi\right]$$

for this normalization. Because $Z^{\mu\nu}$ is a Lorentzian principal tensor, health does not mean that all four of its coordinate eigenvalues are positive. In a local frame with $Z^{0i}=0$, define the time-kinetic coefficient $K_t=-Z^{00}$. A healthy hyperbolic branch requires $K_t\gt0$ and the spatial gradient form $Z^{ij}$ to have the corresponding positive sign; with time–space mixing present, one must inspect the full characteristic polynomial and reduced Hamiltonian instead. On a stable Vainshtein branch, the correctly signed temporal and spatial coefficients can have large magnitudes. If an approximately isotropic estimate gives $Z^{\mu\nu}\simeq Z\eta^{\mu\nu}$ with $Z\gg1$, then $\varphi_c\simeq Z^{1/2}\varphi$, and the cubic fluctuation coefficient becomes

$$\frac{1}{\Lambda_3^3Z^{3/2}} =\frac{1}{\Lambda_{\mathrm{eff}}^3}, \qquad \Lambda_{\mathrm{eff}}\simeq Z^{1/2}\Lambda_3.$$

This kinetic redressing raises the local fluctuation interaction scale. The estimate is schematic when temporal and spatial coefficients differ: each vertex must be normalized with the actual anisotropic quadratic form, and derivatives of $Z^{\mu\nu}$ introduce additional adiabaticity scales. A wrong relative sign can instead produce a ghost, a gradient instability, or loss of hyperbolicity; it is not useful redressing.

For a point source of mass $M$, balancing the linear and cubic terms gives the parametric Vainshtein radius

$$r_V^3\sim\frac{M}{M_{\mathrm{Pl}}\Lambda_3^3}.$$

Inside $r_V$, the background is classically nonlinear, yet fluctuations can be more weakly coupled after redressing. The Galileon operators and their special derivative structure are classified in Nicolis, Rattazzi, and Trincherini 2009, § 2.2.

Classical nonlinearity is not quantum loss of control

Three expansions must be kept separate. The first is the ratio of nonlinear to linear terms in the background equation; it is deliberately large in a screened region. The second is the interaction expansion for canonically normalized fluctuations; redressing can make it small. The third is the Wilsonian derivative expansion controlling operators omitted from the EFT. A large $Z$ does not automatically suppress every higher operator generated by an unknown completion.

The Galileon nonrenormalization structure protects the special operators against certain loop corrections, but it does not prove a UV completion or make arbitrary matter couplings harmless. The upper EFT cutoff is bounded by the redressed interaction scale, heavy-state thresholds, and the scale at which omitted operators compete. Background variation has a different role: its inverse length or time is a lower frequency bound for the local frozen-background approximation, and must itself lie below the EFT cutoff. Slow modes can remain calculable by evolving the varying background explicitly; failure of WKB is not by itself failure of the EFT.

One practical criterion is to require simultaneously

$$\frac{E}{\Lambda_{\mathrm{eff}}}\ll1, \qquad \frac{|\partial Z|}{E|Z|}\ll1, \qquad \left|\frac{\Delta\mathcal L_{\mathrm{omitted}}}{\mathcal L_{\mathrm{kept}}}\right|\ll1.$$

Thus a local scattering estimate needs a nonempty interval $L_{\mathrm{bg}}^{-1}\ll E\ll\Lambda_{\mathrm{EFT}}$. A physical instability is checked separately against the duration of the process; it is not a scale that can always be repaired by reducing an ultraviolet cutoff.

The first controls fluctuation vertices, the second justifies a local canonical basis, and the third controls EFT truncation. Passing only one does not establish predictivity.

The relevant energy is process dependent. For scattering it is set by external momenta; for a binary it may be the orbital frequency, inverse separation, or inverse source size; for a static screened profile it includes radial gradients. Curvature supplies additional invariants such as $|R_{\mu\nu\rho\sigma}|^{1/2}$. A trustworthy claim lists all independent dimensionless ratios and uses the largest one. Testing only frequency can miss a spatial-gradient or curvature cutoff even in a slowly varying configuration.

Uncertainty should grow as the cutoff is approached. If the first omitted operator is known, its estimated contribution provides a truncation error. If its coefficient is unknown, power counting still gives a parametric error band. Reporting a sharp prediction at $E\simeq\Lambda_{\mathrm{EFT}}$ without such an uncertainty is inconsistent with the EFT premise.

Independent checks and failure limits

Dimensional analysis checks the derivation: $\Lambda_3$ and $\Lambda_{\mathrm{eff}}$ have mass dimension one, while $r_V$ has length dimension one. In the vacuum limit $\partial^2\bar\pi\to0$, $Z^{\mu\nu}\to\eta^{\mu\nu}$ and $\Lambda_{\mathrm{eff}}\to\Lambda_3$. If a proposed formula fails this limit, its canonical normalization is inconsistent.

Next reduce the principal quadratic form to a local canonical frame and verify the healthy Lorentzian signature, the positive Hamiltonian time-kinetic sign, and real characteristic speeds on the whole branch. Finally vary the matching radius or background-gradient scale: a prediction that changes at the same order as the retained effect has reached its uncertainty boundary. A vanishing time-kinetic or spatial principal coefficient is especially dangerous because canonical normalization can enhance rather than suppress interactions and the characteristic rank can change.

This page establishes the theory’s calculable regime. Empirical likelihoods and numerical bounds belong to Tests of Gravity; stable low-energy quantum general relativity belongs to Quantum Gravity; generic renormalization methods receive only the bounded bridge needed to define the cutoff.

Common pitfalls

Calling every nonlinear background strongly coupled. Screening can require large classical nonlinearities while the redressed fluctuations remain perturbative.

Quoting the vacuum scale on another branch. Canonical normalization depends on the background kinetic matrix, so the fluctuation scale must be recomputed.

Equating a raised interaction scale with the EFT cutoff. Omitted operators, new states, or rapid background variation may impose a lower limit.

Exercises

In an approximately isotropic region with $Z=100$, estimate the redressed cubic scale and state the checks still needed before using it as the predictive cutoff.

Solution

The canonical field is ten times the unnormalized fluctuation, so $\Lambda_{\mathrm{eff}}=Z^{1/2}\Lambda_3=10\Lambda_3$. This raises the cubic fluctuation scale only. One must still verify the positive time-kinetic sign, correctly signed spatial gradient form and real characteristics, slow variation of $Z$, suppression of every omitted operator, and absence of a lower heavy-state or instability threshold.

Recover the scaling of $r_V$ by equating a linear radial term of order $\bar\pi'/r$ with a cubic term of order $(\bar\pi'/r)^2/\Lambda_3^3$ sourced by $M/M_{\mathrm{Pl}}$.

Solution

The integrated source fixes the linear scaling $\bar\pi'/r\sim M/(M_{\mathrm{Pl}}r^3)$. The cubic term becomes comparable when this quantity is of order $\Lambda_3^3$, giving $r_V^3\sim M/(M_{\mathrm{Pl}}\Lambda_3^3)$. Numerical factors depend on the action normalization, but the dimensions and parameter scaling are fixed.

References

• de Rham, Claudia. 2014. “Massive Gravity.” Living Reviews in Relativity 17, 7. DOI; arXiv:1401.4173.
• Nicolis, Alberto, Riccardo Rattazzi, and Enrico Trincherini. 2009. “The Galileon as a Local Modification of Gravity.” Physical Review D 79, 064036. DOI; arXiv:0811.2197.