General relativity as an effective theory

23.1 A theory can be incomplete without being unreliable#

A laboratory measurement has a finite resolution. At wavelengths much longer than an unknown microscopic scale, we can organize possible gravitational corrections by how small they are at that resolution. This is the effective field theory (EFT) approach to quantum gravity.

An effective theory describes the degrees of freedom accessible at a chosen resolution. It represents unresolved shorter-distance physics through coefficients multiplying allowed local interactions. It does not require us to know every microscopic detail before predicting a long-wavelength experiment.

Think about sound in a solid. At long wavelengths, elasticity uses displacement, density, and a few elastic constants. Its predictions can be excellent without tracking every electron. If you demand wavelengths comparable to atomic separations, the continuum expansion loses its organizing advantage. Gravity’s microscopic completion need not resemble a crystal; the analogy concerns separation of scales, not a claim that spacetime is an atomic material.

The quantum EFT treatment separates calculable long-distance effects from unknown short-distance coefficients. This is the central result of Donoghue’s original work on general relativity as an effective field theory.

23.2 Why unresolved heavy physics becomes derivatives#

Use natural units $c=\hbar=1$ in this subsection. A simple algebraic model explains the expansion. Suppose a heavy field $X$ responds to a source $J$ through

Formally,

This is the geometric series $1/(1-z)=1+z+z^2+\cdots$, now applied to a differential operator. It is useful when the source varies on scales for which the relevant derivatives are small compared with $M_*^2$. At higher frequencies the expansion fails, and the heavy field’s independent dynamics must be restored.

The inverse operator also needs initial or boundary conditions, just as the Green function did in Chapter 18. The displayed series describes the slowly varying sourced response; it does not include an independently excited fast solution of the homogeneous heavy-field equation.

In a quantum calculation, matching means choosing the effective coefficients so that it reproduces the low-energy predictions of a more detailed theory. Perturbative quantum corrections involve integrals over intermediate modes. Some are called loop corrections, after the closed loops in diagrams that organize those integrals; they are not literal particle trajectories. Contributions from arbitrarily large momenta may require a regulator. Renormalization adjusts the coefficients consistently with that regulator, fixing measured quantities so that predictions to the retained order do not depend on this intermediate choice.

For a generally covariant metric theory, local gravitational interactions must be scalar combinations of curvature and covariant derivatives, integrated with the invariant volume. In these units, define the reduced Planck mass by

A schematic local EFT action is

The symbols $\mathcal O_i^{(6)}$ denote local scalar operators of mass dimension six, such as suitable cubic-curvature contractions. $S_{\mathrm{light}}$ contains the light matter fields retained explicitly.

Here mass dimension means the power of mass carried by a quantity’s units when $c=\hbar=1$. Length and time then have inverse-mass units. For the dimensional count, choose length-valued local coordinates and a dimensionless metric. A coordinate then has mass dimension $-1$, a derivative has dimension $+1$, and curvature has dimension $+2$. The measure $d^4x$ has dimension $-4$. Thus $M_{\mathrm{Pl}}^2R$, $R^2$, and $\mathcal O^{(6)}/M_*^2$ all have dimension $+4$, as required for a dimensionless action. The $a_i$ and $b_i$ are dimensionless in this notation. $M_*$ is a heavy-physics or cutoff scale; it need not equal $M_{\mathrm{Pl}}$.

Generic derivative power counting compares curvature-squared terms with the Einstein term at relative order $a_i\mathcal R_*/M_{\mathrm{Pl}}^2$, where $\mathcal R_*$ denotes a characteristic magnitude of curvature components in a physically specified orthonormal frame, with mass dimension two. This estimate organizes possible corrections; specific operators can vanish on particular backgrounds. In four dimensions, constant-coefficient local curvature-squared terms produce no bulk correction when evaluated on a Ricci-flat vacuum solution: variations of $R^2$ and $R_{\mu\nu}R^{\mu\nu}$ vanish there, and the remaining quadratic contraction is related to them by the Gauss–Bonnet combination below. Nonzero Weyl curvature still matters for higher operators and for EFT validity. The actual coefficients determine the suppression scale. “Low energy” is a quantitative hierarchy, not a promise that every coefficient is conveniently small.

The displayed curvature-squared basis is intentionally redundant. In four dimensions the constant-coefficient Gauss-Bonnet combination

does not change local bulk equations under the appropriate variational boundary conditions. A field redefinition changes the variables used to describe the same low-energy configurations. For a small local change $g\mapsto g+\delta g$, the action changes at first order by its field equation contracted with $\delta g$, plus boundary terms. Operators proportional to the leading equations can therefore be exchanged for other terms at the corresponding perturbative order. In a coupled theory, this can move contributions into matter interactions; all fields and observables must be transformed consistently. Counting written terms is not the same as counting measurable new parameters.

Massless quantum fields also produce nonlocal contributions, schematically involving expressions such as $R\log(-\Box/\mu^2)R$. They cannot all be hidden in a finite list of local constants: massless particles propagate over long distances. Here $\mu$ is an arbitrary reference energy, the renormalization scale; its dependence cancels with the corresponding scale dependence of the coefficients in a physical prediction.

The logarithm of an operator can be understood through its modes. In Euclidean coordinates, a mode $e^{i\mathbf k\cdot\mathbf x}$ is an eigenfunction of $-\nabla^2$ with eigenvalue $|\mathbf k|^2$. Acting with $\log(-\nabla^2/\mu^2)$ multiplies that mode by $\log(|\mathbf k|^2/\mu^2)$. This multiplier cannot be represented by a finite polynomial in derivatives. Reassembling the modes produces a response depending on field values across a region, which is the meaning of nonlocal here. The Lorentzian expression needs an additional state and boundary prescription.

23.3 Nonrenormalizable does not mean nonpredictive#

A perturbatively renormalizable theory can absorb ultraviolet divergences into a fixed finite set of couplings at all orders. Einstein gravity, treated as a quantum theory about a suitable background, requires successively higher-order operators. It is not perturbatively renormalizable in that narrow sense.

EFT asks a different question: how many parameters contribute at the accuracy of this experiment? At a fixed order in the low-energy expansion, only finitely many operators contribute. Determine their coefficients by measurement or matching to a more microscopic theory, and the remaining predictions at that order follow.

It resembles approximating a smooth function by a Taylor series. An arbitrary function contains infinitely many coefficients, but a controlled second-order approximation does not require knowing the coefficient of $x^{47}$. The crucial requirement is a valid small expansion parameter and an estimate of neglected terms.

Quantum gravitational loop corrections often carry powers of energy divided by the Planck scale, along with loop factors. In a long-distance problem, a characteristic quantum ratio is

A classical strong-gravity ratio is instead

These are different. Near the horizon of a large black hole, the second can be order unity while the first is tiny. Strong classical gravity is not automatically Planckian quantum gravity. Nor must one expand about flat space to use low-energy reasoning; curved backgrounds can be treated when their physical scales and the quantum state allow a controlled approximation. See Donoghue’s review of quantum GR and its effective-theory limits.

Using a truncated equation. if we truncate an EFT and then solve its higher-derivative equations exactly at arbitrarily high frequency, we may find extra exponentially growing solutions, called runaways, or extra modes with the wrong kinetic-energy sign, called ghosts. That extrapolates the truncated expression beyond the expansion that justified it. Consistent EFT calculations treat higher-order corrections perturbatively. For example, order reduction substitutes the leading equation into higher-derivative correction terms and retains only the desired perturbative order, rather than treating every new high-frequency solution as an independent physical mode. Suitable field redefinitions can serve a related purpose. An extra physical pole genuinely below the proposed cutoff would require reexamining the field content, not dismissing it by slogan. These distinctions are developed in Solomon and Trodden’s research on higher derivatives in EFT.

WORKED EXAMPLE

Make the approximation promise numerical

How accurately can a short derivative expansion reproduce a known low-energy response?

Builds onWave modes Quantum input and units

See the idea

“Small correction” is only useful after naming an expansion parameter and estimating what was omitted. A harmonic source in the heavy-field toy model lets us compare a truncated effective description to an exact particular response. The calculation is classical; it isolates a piece of EFT reasoning without pretending to compute a quantum loop.

Work it out

1. 01
   Choose a mode and a response prescription

   Use c=ℏ=1 and the equation $(M^2-\Box)X=J$. For a spatially uniform source $J=J_0\cos\omega t$, $\Box J=\omega^2J$ in signature $(-+++)$. Choose the time-harmonic particular solution with no homogeneous component, away from $\omega=M$.

   $$X=\frac{J_0}{M^2}\frac1{1-x}\cos\omega t,\qquad x=\omega^2/M^2.$$

   Why this step works Substitution fixes the response amplitude for the specified mode and choice of homogeneous solution.

2. 02
   Keep two terms and compute the remainder

   The two-term amplitude is $1+x$. Subtract it from the exact $1/(1-x)$ and simplify. For $0\le x<1$, dividing by the exact amplitude gives a particularly simple relative error.

   $$\frac1{1-x}-(1+x)=\frac{x^2}{1-x},\qquad \frac{|A_{\rm exact}-A_{\rm two}|}{|A_{\rm exact}|}=x^2.$$

   Why this step works The geometric-series remainder supplies an exact error for this toy mode.

3. 03
   Translate a target accuracy into a scale separation

   To make that relative error below 1%, require $x^2<0.01$, or $\omega/M<\sqrt{0.1}\simeq0.316$. A general EFT may have unknown coefficients, multiple scales, and loop corrections, so this bound belongs to this explicitly solved example.

   $$\omega/M=0.1\quad\Longrightarrow\quad x=0.01,\quad \text{relative error}=10^{-4}.$$

   Why this step works The physical frequency ratio is squared once to form x and again in the two-term error.

Go deeper

Near the pole x=1 this expansion fails; adding a few derivative terms cannot repair the missing heavy degree of freedom. At fixed low energy, increasing the retained order gives a controlled hierarchy for this analytic response. In gravitational EFT, a curvature component scale, derivatives, operator coefficients, and the process being predicted all enter the power count. A term written in an action can be redundant under field redefinition or vanish on a chosen background. Observable matching and a specified operator basis are needed before turning an order estimate into a claimed new gravitational signal.

Further readingJohn Donoghue · General relativity as an effective field theory

Test the idea

FIRST, PREDICT

Can a truncated low-frequency expansion be trusted near the heavy-field resonance just because its first term is simple?

Its failure near the pole makes every low-energy prediction useless.No. Its expansion parameter is no longer small.Yes. Effective means valid at every energy.

Compare the reasoning

Its failure near the pole makes every low-energy prediction useless.

A bounded expansion can be accurate well below its breakdown scale.

No. Its expansion parameter is no longer small.

The exact pole exposes dynamics that the short derivative series cannot capture.

Yes. Effective means valid at every energy.

An effective theory makes predictions within a specified hierarchy of scales.

A hint

Inspect the exact denominator and the size of the omitted term.

NOW CHANGE THE EXAMPLE

In this two-term toy response, $\omega/M=0.2$. Find the relative amplitude error as a decimal.

A hint

Compute $x=(\omega/M)^2$, then $x^2$.

Work through the solution

$x=0.04$ and the relative error is $0.04^2=0.0016$, or 0.16%.

An approximation is a quantitative promise tied to a scale, an observable, and a stated error budget.

23.4 Consistent interactions of a spin-2 field#

There is another way to approach Einstein’s equation. Begin with a free massless spin-2 field $h_{\mu\nu}$ in Minkowski spacetime. Its linear gauge freedom has the form

This removes unphysical components and is related to the two propagating polarizations. Couple $h$ to matter schematically through

Under the gauge transformation, integration by parts makes its variation proportional to $\xi_\nu\partial_\mu T^{\mu\nu}$. Linear gauge consistency therefore wants the source to be conserved.

But once matter interacts with gravity, it exchanges energy and momentum with the gravitational field. Matter stress alone cannot continue to serve as an independently conserved source in the naive flat-background equation. The field must also respond to its own contribution. That changes its dynamics, which changes its contribution again. The nonlinear completion reorganizes this self-coupling into GR under the relevant assumptions.

The logic is powerful: a universally interacting massless spin-2 field cannot consistently behave as if its own interactions are invisible to its source equation. It also explains why simple superposition fails. A primary presentation of the consistency construction is Deser’s “Self-Interaction and Gauge Invariance”.

This is not a theorem that any imaginable spin-2 system must equal pure GR at all energies. The argument relies on assumptions including locality, Lorentz-compatible dynamics, appropriate gauge consistency, field content, and the leading derivative structure. Field redefinitions and stress-tensor improvements affect intermediate expressions. Additional fields, higher-derivative terms, nonlocality, or different backgrounds require separate analysis. A Minkowski-background construction also presupposes an appropriate flat-background limit; it does not determine an arbitrary cosmological constant from nothing.

23.5 The assumptions behind uniqueness#

In four dimensions, the Lovelock classification implies that a natural, symmetric, covariantly divergence-free rank-two tensor built locally from the metric and at most its second derivatives has the Einstein tensor and metric as the available gravitational ingredients, under the theorem’s hypotheses. Consequently, a metric-only second-order field equation of this type takes the Einstein-plus-cosmological form, up to constants. A precise mathematical statement appears in “Lovelock’s theorem revisited”.

Here “natural” means that the construction itself respects smooth coordinate changes, rather than depending on extra coordinate choices. The listed assumptions restrict the available tensors. Add another field, permit higher derivatives, change dimension, or change locality, and the conclusion changes. Thus GR’s distinguished simplicity and EFT’s higher-order corrections are compatible claims. One concerns a restricted class of exact equations; the other organizes small departures when that class is not assumed exact at every scale.

Einstein’s equation is remarkably constrained at its leading level. That makes its success intelligible without making the unfinished parts of physics disappear.

23.6 Vacuum energy: the term that refuses to be a small correction#

The cosmological constant is a special challenge because it multiplies the zero-derivative volume term. It is not automatically suppressed by the long-wavelength expansion that weakens higher-curvature terms.

A Lorentz-invariant vacuum has stress tensor

It has positive energy density if $\epsilon_{\mathrm{vac}}>0$, and pressure $p=-\epsilon_{\mathrm{vac}}$. In Einstein’s equation its effect has exactly the form of a cosmological constant. Schematically, restoring $c$,

with the separate terms understood within a consistent renormalization prescription. Only their physical combination is measurable.

To understand the zero-point estimate, start with a quantum harmonic oscillator of frequency $\omega$. Its lowest energy is $\hbar\omega/2$, rather than zero. One can see the lower bound from the uncertainty relation $\Delta x\Delta p\ge\hbar/2$: the sum $(\Delta p)^2/(2m)+m\omega^2(\Delta x)^2/2$ is at least $\omega\Delta x\Delta p\ge\hbar\omega/2$. A Gaussian state with $\Delta x=\sqrt{\hbar/(2m\omega)}$ and $\Delta p=\sqrt{\hbar m\omega/2}$ attains the bound. This is additional quantum input, not a classical consequence of Einstein’s equation.

A free bosonic field decomposes into harmonic modes, one oscillator per wave vector. In a large periodic box of volume $V=L^3$, allowed wave-vector components are spaced by $2\pi/L$. Dividing the number of modes by volume gives the measure $d^3k/(2\pi)^3$. In natural units the mode frequency is $\omega_{\mathbf k}=\sqrt{|\mathbf k|^2+m^2}$, from the Klein–Gordon dispersion relation in Chapter 13. Adding their lowest energies suggests

A large-momentum cutoff $M_*$ makes this grow roughly as $M_*^4$. You can see the fourth power without doing the integral: the three-dimensional momentum measure contributes three powers, and the high-momentum oscillator energy contributes one more.

That cutoff estimate is not a unique, covariant prediction of the measured cosmological constant. Renormalization, the regulator, masses, interactions, phase transitions, and the gravitational vacuum parameter all matter. The notorious “roughly 120 orders of magnitude” comparison uses a Planck-scale heuristic; it should not be presented as an exact regulator-independent prediction that an experiment simply refuted. Jérôme Martin’s review of the cosmological constant problem.

Nevertheless, the problem survives the correction to the slogan. In the standard cosmological interpretation, the effective dark-energy density corresponds to an energy scale of only a few millielectronvolts raised to the fourth power. An electronvolt is the energy gained by one elementary charge across one volt, $1\,\mathrm{eV}=1.602176634\times10^{-19}\,\mathrm J$; a millielectronvolt is $10^{-3}$ of that. In natural units energy density has energy-to-the-fourth units. Contributions associated with much higher known particle-physics scales naturally dwarf that. Why does the renormalized combination stay so small when such contributions change? This is the radiative-stability question: why does a small measured combination remain small after quantum corrections change its separate contributions? Naturalness asks whether that smallness is protected by a mechanism or requires a fine cancellation among much larger terms.

GR permits a small cosmological constant. It does not explain its observed value, and ordinary EFT bookkeeping does not supply the missing explanation by itself.

23.7 The semiclassical equation and its limits#

An intermediate framework keeps the metric classical while treating matter quantum mechanically:

The higher-curvature terms are included because renormalizing quantum matter on a curved background requires corresponding gravitational couplings. The state-dependent expectation value is a mean stress tensor, not a literal classical list of particles.

This approximation is most credible when the relevant curvature and momenta are below its cutoff, the quantum state is suitable, and neglected metric fluctuations or stress fluctuations do not undermine the mean-field description. Small curvature alone is not a universal certificate of validity. Long evolution, delicate quantum correlations, unusual states, or large fluctuations can raise additional issues.

Hawking radiation inhabits this framework. The endpoint of evaporation and the complete accounting of information generally do not follow just by extending the leading approximation until a black hole becomes arbitrarily small. The approximation must be checked as the mass and curvature evolve.

23.8 What experiments test—and what dark matter and dark energy mean#

An observation does not compare “all of GR” with “all alternatives” in a single stroke. It constrains particular effects over particular scales and source conditions.

For example, the July 2026 LIGO-Virgo-KAGRA GWTC-5.0 analysis compares waveform residuals, polarizations, generation, and remnant properties and reports no overall evidence for physics beyond GR in those tests. That is a strong set of constrained comparisons, with stated statistical and modeling limits. It is not a proof that every possible modification at every scale has vanished. LVK’s primary GWTC-5.0 tests paper.

Dark matter and dark energy also name different explanatory roles. In the usual cosmological model, dark matter behaves approximately as clustering, nearly pressureless matter on large scales. Dark energy denotes a component producing the observed accelerated expansion; a cosmological constant is its simplest standard representation. They are not two names for vacuum energy, nor does either term alone establish that Einstein’s geometric equation is wrong.

The observational inference always depends on a combined model: gravitational laws, visible and invisible sources, their interactions, and initial conditions. A successful alternative must fit the web of measurements together. Matching one galaxy curve or one expansion history is a starting point, not the entire examination.

23.9 Open questions beyond the effective theory#

The unresolved frontier contains concrete questions:

• What microscopic or nonperturbative description remains predictive where the gravitational low-energy expansion fails?
• How do smooth causal geometry and approximately local fields emerge, if they are not fundamental at every scale?
• What counts the black-hole entropy in sufficiently general situations, and how is information represented through formation and evaporation?
• What mechanism, if any, explains the small effective cosmological constant and its stability under quantum corrections?
• Which quantum properties of gravity can be isolated experimentally, rather than inferred solely from a classical gravitational fit?

Different research programs offer different partial answers and controlled special cases. The effective description does not select a unique microscopic theory. That requires additional theoretical consistency and empirical evidence.

The modern achievement is already substantial: the same geometry can be understood as a dynamical constrained system, a local-frame gauge structure, a theory of causal focusing, a thermodynamic participant, and a predictive low-energy quantum field theory. Those are independent pressures on the same equation. A future theory must explain why this structure works so well, as well as where its limits lie.