Energy Estimates and Test-Field Stability

Stability of a test field is proved by controlling an energy, not by observing that a separated potential looks positive. For a massless scalar on a subextremal Schwarzschild exterior, a composite vector-field argument gives a uniformly bounded nondegenerate energy and an integrated local-energy estimate. The stationary Killing current supplies conservation, a redshift multiplier restores control at the horizon, and a radial Morawetz multiplier measures dispersion while degenerating at trapped null geodesics. On Kerr, the ergoregion prevents the stationary energy alone from being positive, so frequency and superradiant structure must also be controlled.

Required background. PDEs and Green Functions for Gravitation supplies energy methods, well-posedness, Sobolev norms, and characteristic boundary fluxes.

Helpful background. Black-Hole Geometry and Mechanics supplies Killing horizons, surface gravity, and ergoregions; Formulations of General Relativity supplies hyperbolicity and multiplier identities; Frames, Coframes, and Anholonomy supplies regular horizon frames.

Scalar stress–energy and multiplier identity

Let $\phi$ be a real solution of $\Box_g\phi=0$ on the four-dimensional Schwarzschild domain of outer communications, with $m=GM\gt0$. Its stress–energy is

$$T_{\mu\nu}[\phi] =\nabla_\mu\phi\nabla_\nu\phi -\frac12g_{\mu\nu}\nabla^\alpha\phi\nabla_\alpha\phi, \qquad \nabla^\mu T_{\mu\nu}=0.$$

For any vector field $X$, define

$$J^X_\mu=T_{\mu\nu}X^\nu, \qquad \pi^X_{\mu\nu}=\frac12\mathcal L_Xg_{\mu\nu}, \qquad K^X=T^{\mu\nu}\pi^X_{\mu\nu}.$$

Direct differentiation gives the fundamental identity

$$\nabla^\mu J^X_\mu=K^X.$$

Let $\Sigma_\tau$ be the stationary, horizon-crossing spacelike slices ending at spatial infinity specified below, with future unit normal $n_{\Sigma_\tau}$. Define the energy and horizon flux by

$$\begin{aligned} E_X(\tau)&=\int_{\Sigma_\tau}J^X_\mu n_{\Sigma_\tau}^{\mu}\,d\Sigma,\\ \mathcal F_H^X(0,\tau)&= \int_{\mathcal H^+(0,\tau)}J^X_\mu T^\mu\,(2m)^2\,dv\,d\Omega. \end{aligned}$$

Here $v=t+r_*$ is advanced time, $dr_*/dr=(1-2m/r)^{-1}$, $T=\partial_t=\partial_v$, and $\mathcal H^+(0,\tau)$ is the horizon portion between the two slices. The generator $T$ and measure $dv\,dA$ are paired: rescaling the generator requires the inverse change of its parameter. For future causal $X$, both densities are nonnegative by the scalar stress tensor’s dominant energy property.

On the exterior slab $\mathcal D_{0,\tau}$, compactly supported data and finite propagation give the exact identity

$$E_X(\tau)+\mathcal F_H^X(0,\tau) +\int_{\mathcal D_{0,\tau}}K^X\,dV_g=E_X(0).$$

The bulk integral enters with a plus sign on the left; $K^X$ need not be nonnegative for an arbitrary multiplier. If the slab is truncated at a timelike cylinder $r=R$, also put the outward energy flux $\mathcal F_R^X=-\int_{r=R}J^X_\mu s^\mu\,dB$ on the left, where $s$ is the outward spacelike unit normal and the slice integrals are truncated too. This cylinder flux can have either sign. For compact data its limit vanishes as $R\to\infty$ at fixed $\tau$. These signs agree with Dafermos, Rodnianski, and Shlapentokh-Rothman 2014, v2, § 2.3.2, Eqs. (16)–(17), p. 14, PDF.

For the Killing field $X=T$, $K^T=0$. Thus $E_T(\tau)+\mathcal F_H^T(0,\tau)$ is conserved; the exterior slice energy alone can decrease as energy enters the hole. This is the Schwarzschild balance in Dafermos and Rodnianski 2008, § 3.2.2, Eq. (8), p. 30, PDF.

Why horizon energy loses a derivative

Coercivity means a positive lower bound on every derivative being estimated, measured in a regular frame. Positivity without that lower bound is weaker. One can see the distinction without solving the wave equation. In ingoing coordinates, put $F=1-2m/r$ and write

$$ds^2=-F\,dv^2+2\,dv\,dr+r^2d\Omega^2, \qquad L=\partial_v+\frac F2\partial_r, \qquad \underline L=-\partial_r.$$

The two null vectors are future directed and satisfy $g(L,\underline L)=-1$. The local observer $n=(L+\underline L)/\sqrt2$ is unit timelike. Set $A=L\phi$, $B=\underline L\phi$, and $Q=|\nabla_{S_r^2}\phi|^2$, where the sphere has radius $r$. Since $T=L+(F/2)\underline L$, direct contraction of the stress tensor gives

$$\begin{aligned} J^T_\mu n^\mu &=\frac1{\sqrt2}\left[A^2+\frac F2B^2 +\frac{1+F/2}{2}Q\right],\\ T_{\mu\nu}n^\mu n^\nu&=\frac12(A^2+B^2+Q). \end{aligned}$$

At $r=2m$, the coefficient of $B^2$ in the Killing density vanishes. There can be an arbitrarily large transverse derivative with no corresponding increase in that density. The regular observer’s density retains $B^2/2$. The horizon Killing flux is more degenerate still: $J^T_\mu T^\mu=(T\phi)^2=A^2$ there. This is an algebraic obstruction to a particular estimate, not an example of a growing solution. The observer $n$ used here illustrates the density; it is not automatically the redshift multiplier whose deformation tensor has the required sign.

Uniform boundedness and local dispersion

For a concrete statement, take smooth initial data $(\phi,n_{\Sigma_0}\phi)$ of compact support on a spherical, horizon-crossing $\Sigma_0$ that avoids the past horizon and agrees with $t=0$ for sufficiently large $r$. Let $\Sigma_\tau$ be its translates by the Killing flow. Choose the slice to be $v-r=\text{constant}$ near the future horizon, with a smooth spacelike transition to constant $t$ before $r=3m$. Thus its geometry and lapse do not change with $\tau$. The solution is the Cauchy evolution in the future of $\Sigma_0$; the future horizon is an outflow boundary, not a place to impose additional incoming data. The field equation is the massless case developed on Scalar Fields and Effective Potentials.

There is a smooth, stationary, future timelike redshift multiplier $N$, equal to $T$ sufficiently far from the horizon, and a constant $C$ independent of $\tau$ and the data, such that

$$E_N(\tau)+\mathcal F_H^N(0,\tau)\le C E_N(0), \qquad \tau\ge0.$$

The $N$ density controls all first derivatives in regular frames uniformly up to the horizon. Stationarity and uniform comparison with the slice normal are essential: an arbitrarily time-dependent timelike multiplier or foliation would not give a uniform constant. The displayed inequality combines the horizon and slice bounds in Dafermos, Rodnianski, and Shlapentokh-Rothman 2014, v2, Theorem 3.1, Eqs. (21), (23)–(24), pp. 15–16, PDF, specialized to Schwarzschild, with the constant enlarged to bound the sum.

Integrated local energy is a different assertion. Let $Z$ be the coordinate radial derivative at fixed slice time: it is regular at the horizon and agrees with $\partial_r$ at fixed Schwarzschild $t$ near trapping and farther out. Put $\zeta(r)=(1-3m/r)^2$. For any fixed finite $R\gt3m$, a consequence of the same theorem is

$$\begin{aligned} \int_{\mathcal D_{0,\infty}\cap\{2m\le r\le R\}} \bigl[&|Z\phi|^2+\zeta(r) \bigl(|T\phi|^2+|\nabla_{S_r^2}\phi|^2\bigr)\\ &+m^{-2}|\phi|^2\bigr]dV_g \le C_R\,m\,E_N(0). \end{aligned}$$

Here $dV_g$ is the spacetime volume measure, and $C_R$ depends on $R/m$ and the fixed slice shape, not on time or the solution. This is a finite-radius consequence, not an estimate uniform as $R\to\infty$: take the Schwarzschild parameter range $a_0=0$ and any fixed $\delta\gt0$ in Eq. (20), restrict to this region, bound its positive radial weights above and below, and restore the length scale $m$. The regular radial derivative and Schwarzschild trapping factor are specified in Dafermos, Rodnianski, and Shlapentokh-Rothman 2014, v2, §§ 2.2.3–2.2.6, pp. 12–13, and Theorem 3.1, Eq. (20), p. 15, PDF. The source’s hybrid radial vector $\widetilde Z$ and our $Z$ differ by $b(r)T$ with a smooth bounded $b$ that vanishes near $3m$. Therefore $b^2\le C_R\zeta$ on the compact radial region, and its extra time-derivative term is absorbed by the displayed trapping weight. The admissible stationary slice change is covered by § 3.3, p. 17, of the same source. The derivative $Z$ remains controlled at $r=3m$, but the time and angular terms degenerate there. The displayed degenerate estimate uses the initial first-derivative energy, not an unspecified higher norm.

Compact support also fixes the otherwise invisible constant part of $\phi$. For a more general finite-energy class with asymptotic constant $\phi_\infty$, the zeroth-order term must be $|\phi-\phi_\infty|^2$, as in Eq. (20). A nonzero constant solution has zero stress-energy and would immediately contradict an estimate of $|\phi|^2$ by derivative energy alone.

The finite spacelike slab has no segment $\mathscr I^+(0,\tau)$: both slices end at spatial infinity. A bound on the flux through the whole future null infinity is a separate statement, Eq. (22) of the same theorem. To discuss flux between finite null-infinity cuts, choose hyperboloidal slices ending on those cuts; the change of foliation is explicit in Dafermos, Rodnianski, and Shlapentokh-Rothman 2014, v2, § 3.3, p. 17, PDF.

This result establishes boundedness and integrated dispersion for this scalar problem. It does not state a sharp Price-law exponent, and it does not yet concern Maxwell, Dirac, Proca, or gravitational perturbations, each of which has a different stress tensor or constraint structure.

Proof architecture: redshift and trapping

The argument has four linked parts.

Stationary conservation. The $T$ current gives an exact energy balance and controls the exterior away from the horizon. Its failure is localized: near $r=2m$, $T$ is null rather than timelike.

Redshift coercivity. Choose $N=T+Y$ near the horizon, where $Y$ points transversely through it and is cut off to zero farther out. For a nondegenerate horizon with surface gravity $\kappa\gt0$, its deformation term satisfies a local inequality of the form

$$K^N\ge b\,J^N_\mu N^\mu, \qquad b\gt0,$$

in a sufficiently small horizon neighborhood. Its extension to $T$ creates cutoff errors in a compact overlap region; positivity is not asserted there. Combining the integrated identity with the exterior estimates controls those errors. Positive surface gravity gives the favorable transverse term, while compact horizon sections and the stationary construction make its constant uniform. The construction and proof are given in Dafermos and Rodnianski 2008, § 7.1, Theorem 7.1, pp. 86–87, PDF.

Morawetz estimate. A radial multiplier $X=f_X(r)\partial_{r_*}$, supplemented by a lower-order current, is chosen so that the modified bulk term controls spacetime derivatives. The coefficient must change behavior around the photon sphere. At the same initial derivative order it cannot remain strictly coercive on all derivatives at $r=3m$, because geometric-optics wave packets can linger near the trapped null orbit. A Hardy inequality controls the remaining zeroth-order terms. The orbit underlying this obstruction is derived on Geometric-Optics and Eikonal Propagation.

Asymptotic hierarchy and commutation. Weighted outgoing multipliers control energy transport toward null infinity. Commuting with $T$ and angular momenta, then applying Sobolev inequalities, converts energy decay into pointwise estimates. Boundedness alone does not imply pointwise decay; these extra estimates are indispensable.

The regions overlap; these are cooperating estimates, not disjoint zones.

Ingredient	Main contribution	What it does not supply alone
Killing current $T$	Exact flux balance and positive exterior energy	Uniform transverse control at the horizon
Redshift current $N$	Regular first-derivative control near the horizon	Dispersion through the trapped region
Radial Morawetz current	Time-integrated local energy	Nondegenerate trapping control at the same derivative order
Outgoing weighted currents and commutation	Decay on suitable slices and pointwise estimates	A conclusion without additional weighted data bounds

Rotation, ergoregions, and boundaries

On Kerr, $T=\partial_t$ becomes spacelike in the ergoregion, so $J^T_\mu n^\mu$ is not positive definite. The horizon generator $T+\Omega_H\Phi$ is null on the horizon but is not globally timelike to infinity. There is therefore no single globally causal stationary Killing multiplier whose conservation proves coercive boundedness for arbitrary modes.

Here the defect is a negative direction, not just a missing square. At a point choose an orthonormal frame with future unit vector $n$ and $T=\alpha n+\beta e_1$, where $\alpha\gt0$ and $|\beta|\gt\alpha$. For a scalar gradient with $n\phi=u$, $e_1\phi=w$, and no other components,

$$J^T_\mu n^\mu=\frac\alpha2(u^2+w^2)+\beta uw.$$

Taking $u=1$ and $w=-\operatorname{sgn}\beta$ gives $\alpha-|\beta|\lt0$. Thus conservation of this quadratic form cannot bound the positive local observer energy. It still does not prove that a solution grows: superradiant energy exchange and global boundary conditions matter.

For every fixed subextremal range $|a|\le a_0\lt GM$, the scalar problem has uniform energy boundedness and integrated local-energy decay for sufficiently regular solutions with bounded initial energy, subtracting the asymptotic constant in zeroth-order terms. Polynomial energy decay on hyperboloidal slices and pointwise decay additionally require sufficiently regular, localized data with finite higher-order weighted energy. The proof combines physical-space currents with separation, quantitative mode stability, control of superradiant frequencies, and the geometry of frequency-dependent trapping Dafermos, Rodnianski, and Shlapentokh-Rothman 2014, v2, Theorem 3.1, pp. 15–16, and Corollary 3.1, p. 17, PDF. Positivity of a one-dimensional potential is neither necessary nor sufficient for this statement.

Boundary conditions can also change the conclusion. A reflecting timelike boundary may return flux instead of allowing it to disperse. A massive field may possess long-lived quasi-bound states. A horizonless spacetime with an ergoregion lacks the absorbing horizon term that helps close the black-hole energy balance. Every stability claim must therefore state its field, background, data class, boundary conditions, regularity, and norm.

Limitations and discriminating counterexamples

The hypotheses expose several sharp limitations.

• At extremality, $\kappa=0$, so the nondegenerate redshift estimate fails. One cannot obtain the same horizon control by continuity in surface gravity; conserved horizon quantities and transverse growth require a different analysis.
• Stable trapping prevents a uniform strong decay estimate of the same form. Even unstable trapping forces degeneration or derivative loss: high-frequency Gaussian beams concentrated near the photon sphere are the counterexample to an everywhere-coercive Morawetz bulk term.
• An indefinite $T$ energy in an ergoregion does not itself prove instability, but it invalidates the naive conservation-plus-positivity proof.
• Linear test-field boundedness does not imply stability of the metric, nonlinear stability, or control of backreaction. Those require constraint-compatible gravitational variables and nonlinear estimates.

Quantum-state particle interpretation is likewise beyond this classical method. An energy estimate controls solutions of a classical PDE; it does not select a vacuum or compute occupation numbers in QFT in Curved Spacetime.

Exercises

Using the regular null frame above, take a pointwise derivative assignment $A=0$, $Q=0$, $B=1$. Calculate the ratio of Killing energy density to the local observer energy density. Can a uniform positive lower bound survive as $r\to2m$? Explain what a redshift multiplier must add, and why the answer does not exhibit an unstable solution.

Solution

The two densities are $F/(2\sqrt2)$ and $1/2$, so their ratio is $F/\sqrt2\to0$. No lower bound by a positive constant independent of $r$ is possible. Scaling $B$ changes both quadratic forms but cannot repair the vanishing ratio. The assignment concerns an admissible derivative at a point, not a claim about a solution’s late-time behavior.

Choose a future timelike vector $N=T+Y$ near the horizon, with $Y$ regular and transverse, and match it to $T$ outside a small neighborhood. Compute $K^N=T^{\mu\nu}\pi^N_{\mu\nu}$. Positive surface gravity permits the coefficients to be chosen so that the bulk term controls the full nondegenerate $N$ energy near the horizon, up to errors already controlled by the $T$ energy. Integrate the $N$ identity, add the conserved $T$ estimate away from the horizon, and absorb the overlap errors. A separate radial Morawetz current is still required for integrated decay and must degenerate at $r=3m$.

For an extremal horizon, $\kappa=0$, and this redshift positivity is lost. Reusing the same inequality with a constant independent of $\kappa$ is false. One must formulate a degenerate estimate and account for horizon conservation laws or derivative growth; the subextremal proof only establishes the result under its stated $\kappa\gt0$ hypothesis.

A proposed local-energy estimate bounds a spacetime integral containing $m^{-2}|\phi|^2$ by $C m E_N(0)$ for every smooth finite-energy solution, without a condition at infinity. Give an exact counterexample and two valid ways to repair the statement.

Solution

Set $\phi=c_0\ne0$ everywhere. It solves $\Box_g\phi=0$ and has $T_{\mu\nu}=0$, hence $E_N(0)=0$. The proposed bulk is positive on any nonempty finite spacetime region and diverges if integrated over all future times at fixed radius. Require data with zero asymptotic constant, such as the compact data in the stated theorem, or estimate $|\phi-\phi_\infty|^2$ in a finite-energy class with a defined asymptotic constant. Extra commuted derivative energies alone cannot detect $c_0$.

References

• Dafermos, Mihalis, and Igor Rodnianski. “Lectures on Black Holes and Linear Waves.” In Evolution Equations, Clay Mathematics Proceedings 17, 97–205. Providence: American Mathematical Society, 2013. Stable preprint record. Open PDF, 2008 lecture edition; the section and page locators above refer to this edition.
• Dafermos, Mihalis, Igor Rodnianski, and Yakov Shlapentokh-Rothman. “Decay for Solutions of the Wave Equation on Kerr Exterior Spacetimes III: The Full Subextremal Case $|a|\lt M$.” Annals of Mathematics 183 (2016): 787–913. DOI. Open PDF, 2014 preprint v2; the page locators above refer to this version.