Controlled Approximations and Correspondence Limits

A candidate quantum-gravity construction becomes physically interpretable only when it recovers established observables in an overlap regime with controlled errors. Merely taking a symbol called $\hbar$ to zero, a lattice spacing to zero, or a quantum number to infinity is not enough: one must specify the states, observables, boundary data, order of limits, and size of the first omitted term. This page gives a practical method for making that correspondence test quantitative.

Required background. Einstein Equations supplies the classical target equations. Quantum Mechanics supplies states, expectation values, and fluctuations.

Helpful background. Flat-Spacetime Quantum Field Theory supplies loop and momentum expansions, while Quantum Information clarifies which reduced states or algebras define operational observables.

A correspondence limit is a map

Let a microscopic theory depend on control parameters $\lambda$ and define an observable $O_{\mathrm{micro}}(\lambda)$. A correspondence claim requires a map $\Phi_\lambda$ from microscopic states and observables to those of the target theory and a norm or statistical distance in which

$$\left\lVert \Phi_\lambda[O_{\mathrm{micro}}]-O_{\mathrm{target}}\right\rVert \le C\,\epsilon(\lambda)^p, \qquad \epsilon(\lambda)\longrightarrow0.$$

The exponent $p$, coefficient domain, and uniformity matter. Pointwise convergence of a two-point function at fixed separation need not control its coincidence limit. Convergence of homogeneous expectation values need not control perturbations. A finite transition amplitude need not imply that the constraint algebra closes.

The target also changes with scale. At distances much larger than all Compton wavelengths and source sizes, classical general relativity may be the correct target. With weak metric fluctuations and external momenta $E\ll M_{\mathrm{Pl}}$, gravitational effective field theory is the target. For quantum matter with negligible metric fluctuations but appreciable mean stress energy, the semiclassical Einstein equation may be appropriate. A single construction should recover compatible answers in overlap regions rather than assigning one unexplained limit to all three.

Choose independent control parameters

Several small ratios are often present simultaneously:

$$\epsilon_E=\frac{E}{M_{\mathrm{Pl}}},\qquad \epsilon_R=\ell_{\mathrm{Pl}}^2\sqrt{\lvert R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma}\rvert},\qquad \epsilon_{\mathrm{loop}}\sim\frac{\kappa^2 E^2}{16\pi^2}\sim GE^2,$$

together with a state-fluctuation ratio. Here $M_{\mathrm{Pl}}=G^{-1/2}$ is the unreduced Planck mass and $\ell_{\mathrm{Pl}}^2=G$. The displayed scalar curvature diagnostic is useful only where representative: vanishing scalar invariants do not exclude tidal fields, so a specified observer and probe can require additional frame components.

To define a fluctuation, choose a real smooth compactly supported tensor test function $f^{\mu\nu}$ in the measurement region and form the Hermitian smeared stress component

$$T[f]=\int d^4x\,\sqrt{-g}\,f^{\mu\nu}(x)\hat T_{\mu\nu}^{\mathrm{ren}}(x).$$

For a state with finite smeared variance and nonzero mean, one diagnostic is

$$\epsilon_T^2= \frac{\left\langle\bigl(T[f]-\langle T[f]\rangle\bigr)^2\right\rangle} {\lvert\langle T[f]\rangle\rvert^2}.$$

This compares one measurement with its own mean; it is not a tensor divided componentwise by another tensor. If the mean vanishes, the ratio is undefined and one must compare the absolute induced metric fluctuations or detector noise with a physical tolerance. Even a small ratio is not a universal semiclassical-validity theorem: the metric response and its stability must be checked. Low curvature does not guarantee small stress-tensor fluctuations; low external energy does not control a process whose state probes a trans-Planckian invariant.

In low-energy quantum gravity, locality and diffeomorphism invariance organize an action in increasing derivatives. Loop effects and higher-derivative operators can then be power counted in $GE^2$ once the process and renormalization prescription are declared Donoghue 2023, §§2–4. The first omitted operator or loop order supplies an error estimate. Unknown local coefficients affect analytic momentum dependence; long-distance nonanalytic terms can be predictions of the low-energy degrees of freedom.

A four-stage test

Use the following sequence for a continuum, semiclassical, or large-quantum-number claim.

Stage	Required declaration	Adverse check
Define	Microscopic variables, state or history sum, constraints, regulator, boundary data	Change gauge or parametrization and verify the observable is unchanged
Scale	Dimensionless control parameters and correlated order of limits	Reverse noncommuting limits or enlarge the retained domain
Match	A target observable and normalization, not only equations of motion	Match a second observable not used to tune coefficients
Bound	First omitted correction and a quantitative residual	Vary regulator, truncation, resolution, or fit window

For a discretized theory, “send the lattice spacing to zero” is incomplete unless a physical length is held fixed, a renormalized trajectory is identified, and observables converge under refinement. For a symmetry-reduced model, agreement within the reduced sector is not evidence that discarded inhomogeneous modes are small. For a large-spin or coherent-state limit, small relative fluctuations must accompany agreement of expectation values; otherwise the nominal classical geometry is not sharply realized.

Independent Schwarzschild check

Consider the exterior of a Schwarzschild source of mass $M$. The invariant curvature scale is measured by the Kretschmann scalar

$$K=R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma} =\frac{48G^2M^2}{r^6}.$$

The local derivative-expansion parameter is therefore

$$\epsilon_R=\ell_{\mathrm{Pl}}^2\sqrt K =4\sqrt3\,\frac{G^2M}{r^3}.$$

At the horizon $r=2GM$, this becomes $\epsilon_R=\sqrt3/(2GM^2)$, which is small for $M\gg M_{\mathrm{Pl}}$. A macroscopic horizon can thus be deep in the low-curvature regime despite its causal importance. Far away, $\epsilon_R\propto r^{-3}$ gives the expected infrared improvement. Near $r\sim(G^2M)^{1/3}$ the curvature expansion fails even if a coordinate component of the metric looks innocuous.

An independent scaling route reaches the same conclusion: the tidal acceleration across a proper separation $L$ is $\Delta a\sim GM L/r^3$. Comparing the tidal gradient $\Delta a/L$ with $\ell_{\mathrm{Pl}}^{-2}$ gives the same combination $G^2M/r^3$ up to an order-one tensor factor. The two routes test dimensions and radial scaling without relying on a coordinate singularity.

This check also illustrates what it does not establish. Small $\epsilon_R$ validates a local curvature expansion; it does not ensure that a chosen quantum state is regular, that horizon entanglement factorizes, or that an evaporation calculation remains adiabatic for all times.

Failure modes and reporting standard

A correspondence statement is weak if coefficients are retuned at every resolution, if only a kinematical spectrum is compared, or if the limit eliminates the observable being claimed. Nonuniform limits are especially dangerous near horizons, phase transitions, and long times: an error $O(\epsilon t)$ becomes order unity when $t\sim\epsilon^{-1}$.

Report the approximation as a tuple: degrees of freedom; state; observable; background or phase; boundary conditions; gauge or constraint treatment; regulator and renormalization; small parameters; retained order; first omitted term; and adverse checks. This makes it possible to distinguish a controlled recovery of GR from a suggestive numerical resemblance.

Exercises

Exercise. A calculation predicts a dimensionless observable

$$O(E)=O_0+c_1GE^2+c_2G^2E^4+O(G^3E^6).$$

Assume $\lvert c_1\rvert\sim\lvert c_2\rvert\sim1$. At what energy ratio is truncation after the $c_1$ term parametrically controlled, and what fractional truncation uncertainty should be quoted when $GE^2=10^{-3}$ and $O_0$ is nonzero and order unity?

Solution

The expansion parameter is $\epsilon=GE^2=(E/M_{\mathrm{Pl}})^2$. Retaining the $c_1$ term is controlled for $E/M_{\mathrm{Pl}}\ll1$. The first omitted absolute contribution is of order $\epsilon^2=10^{-6}$. Because $O_0$ is order unity and nonzero, the fractional uncertainty in the full observable is also of order $10^{-6}$, subject to order-one coefficients. Quoting $10^{-3}$ would confuse the retained correction with its truncation error; quoting no uncertainty would ignore the first omitted operator.

References

• Donoghue, John F. “Quantum General Relativity and Effective Field Theory.” 2023. arXiv:2211.09902, §§2–4.