Higher Derivatives, Runaways, and Order Reduction

Curvature-squared counterterms make the local semiclassical equation fourth order, but a truncated effective equation must not automatically be treated as an exact fundamental theory. Its additional high-frequency solutions are often nonanalytic in the small expansion parameter and lie at the cutoff. Order reduction removes those spurious branches while preserving the low-frequency solution through the retained order. The operation is controlled only when the correction is perturbative and its first omitted term remains small.

Required background. Einstein Equations supplies the lower-order equations used in reduction, and Gravitational Actions and Boundary Terms supplies the higher-curvature origin and boundary data.

Helpful background. Quantum Fields in Curved Spacetime supplies the state-dependent nonlocal terms that accompany local derivatives, while Black-Hole Geometry and Mechanics supplies later horizon applications.

A toy higher-derivative equation

Consider a real oscillator with $\omega\gt 0$ and $\epsilon\gt 0$, where $[\epsilon]=\mathrm{time}^2$:

$$\ddot q+\omega^2q=\epsilon q^{(4)}, \qquad \epsilon\omega^2\ll1.$$

With $q\propto e^{st}$, the exact characteristic equation is

$$\epsilon s^4-s^2-\omega^2=0, \qquad s^2_\pm=\frac{1\pm\sqrt{1+4\epsilon\omega^2}}{2\epsilon}.$$

The minus branch is analytic as $\epsilon\to0$:

$$s_-^2=-\omega^2+\epsilon\omega^4+O(\epsilon^2).$$

The plus branch behaves as $s_+^2=\epsilon^{-1}+\omega^2+\cdots$ and contains exponentially growing and decaying solutions on the timescale $\sqrt\epsilon$. It has no smooth limit as the correction is removed. Treating the truncated equation as exact has enlarged the solution space beyond the expansion from which it was derived.

The order-reduction operation

Use the zeroth-order equation $\ddot q=-\omega^2q+O(\epsilon)$ inside the term already multiplied by $\epsilon$. Differentiating twice gives $q^{(4)}=\omega^4q+O(\epsilon)$, hence

$$\ddot q+(\omega^2-\epsilon\omega^4)q=O(\epsilon^2).$$

This second-order equation reproduces $s_-^2$ through $O(\epsilon)$ and needs only the classical pair $(q,\dot q)$ as initial data. Here powers of $\epsilon$ mean an expansion at fixed $\omega$; the dimensionless parameter is $\epsilon\omega^2$. It does not approximate the runaway branch, because that branch varies on the scale at which the derivative expansion fails.

For an explicit remainder check, insert $q_{\mathrm{red}}=\cos(\Omega t)$ with $\Omega^2=\omega^2-\epsilon\omega^4$ into the original equation. The residual is

$$\ddot q_{\mathrm{red}}+\omega^2q_{\mathrm{red}} -\epsilon q_{\mathrm{red}}^{(4)} =\left(2\epsilon^2\omega^6-\epsilon^3\omega^8\right)q_{\mathrm{red}}.$$

This verifies the retained order directly. It does not bound the solution error for arbitrarily long times: the omitted frequency correction is $O(\epsilon^2\omega^5)$, so its phase error accumulates with time.

Parker and Simon formulate this perturbative constraint systematically for semiclassical gravity and work through Friedmann, conformally flat, and general examples Parker and Simon 1993, §§ 2–6 and Appendix A, PDF. Their conclusion is equality through the retained perturbative order, not an assertion that an unknown exact theory contains no additional degrees of freedom.

From the toy model to gravity

Varying $R^2$ and $C^2$ produces terms with up to four derivatives of the metric. Schematically,

$$G_{\mu\nu}=8\pi GT_{\mu\nu} +\epsilon H_{\mu\nu}[g,\nabla g,\ldots,\nabla^4g] +O(\epsilon^2).$$

Order reduction substitutes the leading Einstein equation and its derivatives into the $\epsilon H_{\mu\nu}$ term. The resulting tensor remains covariant and conserved through $O(\epsilon)$ when the substitution is performed consistently. One must reduce the constraints and evolution equations together; reducing only selected components can violate the Bianchi identity.

State-dependent response is generally nonlocal, so it cannot always be replaced by a finite derivative polynomial. A causal retarded kernel may be kept explicitly, with its poles and branch cuts defined by the initial-value prescription. Anderson, Molina-París, and Mottola compare order reduction with contour treatment of the response equation in Anderson, Molina-París, and Mottola 2003, §§ III–IV, PDF.

Branch and validity tests

Four checks distinguish a physical low-energy branch from a truncation artifact:

1. Analyticity: the solution approaches a classical solution as $\epsilon\to0$.
2. Frequency: all retained poles satisfy $\lvert\omega_{\mathrm{mode}}\rvert\ll\epsilon^{-1/2}$ or the appropriate cutoff.
3. Remainder: substituting the reduced solution in the unreduced equation leaves an $O(\epsilon^2)$ residual over the claimed time interval.
4. Constraints: the reduced stress and geometric terms obey conservation to the same order.

Secular growth can invalidate the residual estimate at late times even when each instantaneous correction is small. Order reduction also cannot decide the fate of a genuine low-frequency instability that survives as $\epsilon\to0$; deleting such a pole would alter rather than truncate the theory.

Hu and Verdaguer discuss the Planck-frequency runaway issue and its treatment in Hu and Verdaguer 2008, §§ 3.3 and 6.5.3. Their flat-space example illustrates why stability must be stated together with the prescription used for the high-frequency branch.

Interpretation and limits

Order reduction is best understood as perturbative matching. For the same physical low-energy branch and the same initial, state, and boundary data, consistent reduction or contour implementations may disagree by $O(\epsilon^2)$ without contradicting the retained calculation. An $O(\epsilon)$ disagreement for a below-cutoff observable then signals an inconsistent implementation or a failed equivalence assumption. Different physical boundary prescriptions can instead select different low-energy solutions; they are not required to agree.

The method does not repair Planckian curvature, a non-Hadamard state, large stress fluctuations, or a nonlocal memory effect that cannot be expanded. Fixed-background calculation of the quantum kernel remains separate, and a decision about genuine ultraviolet gravitational modes belongs to Quantum Gravity rather than to the truncated semiclassical equation.

Exercise

Expand both roots $s_\pm^2$ of the toy equation for small positive $\epsilon$ and verify that the order-reduced equation reproduces only the analytic branch through first order.

Solution

Using $\sqrt{1+4\epsilon\omega^2}=1+2\epsilon\omega^2-2\epsilon^2\omega^4+O(\epsilon^3)$ gives $s_-^2=-\omega^2+\epsilon\omega^4+O(\epsilon^2)$ and $s_+^2=\epsilon^{-1}+\omega^2-\epsilon\omega^4+O(\epsilon^2)$. The reduced equation has $s^2=-(\omega^2-\epsilon\omega^4)$, exactly the first expansion through $O(\epsilon)$. The $s_+$ branch diverges as $\epsilon\to0$ and lies at the derivative-expansion cutoff, so no perturbative second-order equation should reproduce it.

References

• Anderson, P. R., Molina-París, C., and Mottola, E. 2003. “Linear Response, Validity of Semi-Classical Gravity, and the Stability of Flat Space.” Physical Review D 67, 024026. Open PDF.
• Hu, B. L., and Verdaguer, E. 2008. “Stochastic Gravity: Theory and Applications.” Living Reviews in Relativity 11, 3. DOI.
• Parker, L., and Simon, J. Z. 1993. “Einstein Equation with Quantum Corrections Reduced to Second Order.” Physical Review D 47, 1339–1355. Open PDF.