Mode Stability of Schwarzschild

Mode stability of Schwarzschild says that no nonzero separated radiative perturbation grows exponentially while satisfying the future-horizon and future-null-infinity boundary conditions. For Regge–Wheeler and Zerilli modes, the proof is a short positive-potential integral identity. Its conclusion is important but narrow: it excludes unstable separated modes; it does not by itself bound arbitrary initial-data solutions or prove their decay.

Required background. PDEs and Green Functions for Gravitation supplies spectral boundary conditions, integration by parts, and energy estimates.

Helpful background. Black-Hole Geometry and Mechanics fixes the Schwarzschild exterior; Formulations of General Relativity separates constraints and evolution; Mathematical Relativity supplies the theorem/estimate distinction; Black-Hole Perturbation Theory supplies the master equations.

Schwarzschild mode-stability statement

Fix $M\gt0$, write the mass length as $m=GM$, and take one gravitational harmonic $\ell\ge2$ in the vacuum exterior $r\gt2m$. The tortoise coordinate and its derivative are

$$x=r_*=r+2m\log\left(\frac r{2m}-1\right), \qquad D=\frac{d}{dx}=f\frac{d}{dr},\qquad f=1-\frac{2m}{r}.$$

Thus $x$ covers the whole real line: the horizon is at $-\infty$ and spatial infinity at $+\infty$. For either parity, separation gives

$$\Psi(t,x)=e^{-i\omega t}\psi(x),\qquad H_\pm\psi=\omega^2\psi,\qquad H_\pm=-D^2+V_\pm.$$

Primes below also mean $D$. The odd Regge–Wheeler potential and even Zerilli potential are

$$\begin{aligned} V_-&=V_{\mathrm{RW}} =f\left[\frac{\ell(\ell+1)}{r^2}-\frac{6m}{r^3}\right],\\ V_+&=V_{\mathrm Z}=\frac{2fP(r)}{r^3(\lambda r+3m)^2},\\ P(r)&=\lambda^2(\lambda+1)r^3+3\lambda^2mr^2 +9\lambda m^2r+9m^3,\\ \lambda&=\frac{(\ell-1)(\ell+2)}2\ge2. \end{aligned}$$

These agree with the original radial equations of Regge and Wheeler 1957, Eqs. (23)–(26), p. 1066 and Zerilli 1970, Eqs. (2)–(5), p. 738. The earlier Regge–Wheeler and Zerilli pages derive the metric-to-master reductions. Their constant master normalizations do not change this homogeneous radial argument.

Both potentials are strictly positive at every finite exterior radius. For odd parity, $\ell(\ell+1)r-6m\ge6(r-m)\gt0$; for even parity, every term of $P$ and the denominator is positive. Both potentials tend to zero at the ends, with fixed-$m,\ell$ estimates

$$V_\pm=O(e^{x/(2m)})\quad(x\to-\infty),\qquad V_\pm=O(x^{-2})\quad(x\to+\infty).$$

In particular they are real, bounded and integrable on the tortoise line.

Theorem. No nonzero smooth solution of either radial equation has $\operatorname{Im}\omega\gt0$ and is simultaneously ingoing at the future horizon and outgoing at future null infinity, with the endpoint branches defined below. The proof treats the two potentials separately, without transferring boundary conditions through a parity transformation.

Physical growing-mode boundary conditions

Write $\omega=\omega_R+i\beta$, with $\beta\gt0$. The time factor grows as $e^{\beta t}$. The corresponding radial factors decay at the physical ends. The null coordinates explain the opposite radial signs:

Physical endpoint	Null coordinate	Full wave factor	Radial magnitude
Ingoing at the future horizon, $x\to-\infty$	$v=t+x$	$e^{-i\omega v}$	$e^{\beta x}\to0$
Outgoing at future null infinity, $x\to+\infty$	$u=t-x$	$e^{-i\omega u}$	$e^{-\beta x}\to0$

Let $F$ and $G$ be the separately unit-normalized endpoint solutions for the chosen potential. For fixed $\omega$ in the open upper half-plane they obey

$$\begin{aligned} F&=e^{-i\omega x}\bigl[1+O(e^{x/(2m)})\bigr],\\ F'&=e^{-i\omega x}\bigl[-i\omega+O(e^{x/(2m)})\bigr] && (x\to-\infty),\\ G&=e^{i\omega x}\bigl[1+O(x^{-1})\bigr],\\ G'&=e^{i\omega x}\bigl[i\omega+O(x^{-1})\bigr] && (x\to+\infty). \end{aligned}$$

A nonzero physical mode would satisfy $\psi=A_HF=A_\infty G$, with two nonzero complex amplitudes. Their ratio is determined by the equation; there is only one arbitrary overall normalization. Requiring both amplitudes to equal one would impose an unintended extra condition.

Why the derivative estimates hold

One must not differentiate a bare asymptotic equivalence. Instead put $U_H=e^{i\omega x}F$ and $U_\infty=e^{-i\omega x}G$. Variation of constants gives the normalized Volterra equations

$$\begin{aligned} U_H(x)&=1+\int_{-\infty}^x \frac{e^{2i\omega(x-y)}-1}{2i\omega}V(y)U_H(y)\,dy,\\ U_\infty(x)&=1+\int_x^\infty \frac{e^{2i\omega(y-x)}-1}{2i\omega}V(y)U_\infty(y)\,dy. \end{aligned}$$

Here $V$ means the selected $V_-$ or $V_+$. In both integrals the exponential has magnitude at most one. Successive substitution converges on either endpoint half-line: its $n$th term is bounded by the $n$th power of an integrable potential tail, divided by $n!$, with constants depending on $\omega$. Thus $U-1$ has the potential-tail bound. Differentiating the integral is legitimate by the same integrable bound; its moving-endpoint term vanishes because the kernel is zero at $y=x$. This supplies the displayed derivative estimates. Continue each solution to finite $x$ by ordinary differential-equation uniqueness.

This is the construction in Donninger, Schlag, and Soffer 2011, arXiv v2, Lemma 3.1 and Eq. (8), pp. 10–11, PDF, reflected for the horizon endpoint. Their Regge–Wheeler argument uses precisely the endpoint decay shared here by the Zerilli potential. These are fixed-frequency estimates, not uniform bounds as $\beta\downarrow0$ or $\omega\to0$.

For the hypothetical physical mode, both $\psi$ and $\psi'$ are therefore square-integrable, and $\overline\psi\psi'\to0$ at both ends. These are the specific facts needed for integration by parts.

Positive-potential integral argument

Multiply the radial equation by $\overline\psi$ on a finite interval $[a,b]$. Before removing either endpoint, retain the boundary term:

$$\int_a^b\bigl(|\psi'|^2+V|\psi|^2\bigr)\,dx -[\overline\psi\psi']_a^b =\omega^2\int_a^b|\psi|^2\,dx.$$

The endpoint estimates justify taking $a\to-\infty$ and $b\to+\infty$. Define the norm and quadratic form

$$I=\int_{\mathbb R}|\psi|^2\,dx,\qquad Q=\int_{\mathbb R}\bigl(|\psi'|^2+V|\psi|^2\bigr)\,dx.$$

Both are finite, $Q\ge0$, and $I\gt0$ for a nonzero mode. The identity is $Q=\omega^2I$. Its imaginary part reads

$$0=\operatorname{Im}(\omega^2)I=2\omega_R\beta I.$$

Since $\beta I\gt0$, necessarily $\omega_R=0$. Set $\omega=i\sigma$, $\sigma=\beta\gt0$. The real part now gives

$$0=Q+\sigma^2I =\int_{\mathbb R} \bigl(|\psi'|^2+V|\psi|^2+\sigma^2|\psi|^2\bigr)\,dx.$$

Every term is nonnegative, whereas $\sigma^2I$ is strictly positive for a nonzero mode. This contradiction proves the theorem.

In operator language, an unstable mode would need a negative eigenvalue $\omega^2=-\sigma^2$ of a nonnegative quadratic form. The elementary proof has already established the required domain and boundary facts; the word “positive” alone would not do so. It also explains why a different master potential with a negative region needs a separate argument, such as a positive deformation or a boundary-compatible transformation. A negative region invalidates this particular inference; it does not by itself prove instability. The smooth negative-well exercise below shows what can go wrong.

Zero frequency and static boundary conditions

The frequency $\omega=0$ is a threshold, not an exponentially growing mode. The preceding proof cannot be reused by setting $\sigma=0$ and silently keeping its square-integrability assumption. A horizon-regular static master usually approaches a constant as $x\to-\infty$, which is not in $L^2(dx)$ unless that constant vanishes.

There is nevertheless a stronger useful statement: a solution of $H_\pm\psi=0$ bounded at both ends must vanish. This includes the isolated static problem with a regular horizon and a decaying outer master field. Here is a proof that does not assume a finite $I$.

The equation is $\psi''=V\psi$. Since $V$ is integrable and $\psi$ bounded, $\psi''$ is integrable, so $\psi'$ has a finite limit at each end. A nonzero limit would make $\psi$ grow linearly in $x$, contradicting boundedness. Both limits are therefore zero, and

$$\psi'(x)=\int_{-\infty}^xV(y)\psi(y)\,dy =-\int_x^\infty V(y)\psi(y)\,dy.$$

The first representation gives $\psi'=O(e^{x/(2m)})$ near the horizon; the second gives $\psi'=O(x^{-1})$ at infinity. Thus $\overline\psi\psi'\to0$ at both ends, $\psi'$ is square-integrable, and $V|\psi|^2$ is integrable. The finite-cut identity now yields $Q=0$ without ever forming $0\times I$. Strict positivity of $V$ in the exterior forces $\psi=0$.

A nonzero bounded static solution would be a zero-energy resonance in the full-line convention of Donninger, Schlag, and Soffer 2011, arXiv v2, Definition 6.1 and Lemma 6.3, pp. 24–25, PDF. Their explicit Regge–Wheeler result agrees with this argument; the proof just given also covers Zerilli.

This is not a ban on local stationary $\ell\ge2$ solutions. The regular horizon branch can grow at infinity when an external tidal field is allowed. The quadrupole exercise gives an exact example. Nor should a static metric be discarded merely because a time-differentiated odd master vanishes: use the complete odd invariant and the stationary reconstruction described in Metric Reconstruction and Radiative Observables.

Algebraically special frequencies

The Chandrasekhar transformation has an inverse factor $\omega^2+\alpha^2$, where

$$\alpha=\frac{\lambda(\lambda+1)}{3m}\gt0.$$

It fails at both $\omega=+i\alpha$ and $\omega=-i\alpha$. Failure of an inverse map is not evidence for a physical mode. Its exact kernels are

$$\begin{aligned} \chi_-&=\left(\lambda+\frac{3m}{r}\right)e^{-\alpha x}, & H_-\chi_-&=-\alpha^2\chi_-,\\ \chi_+&=\chi_-^{-1}, & H_+\chi_+&=-\alpha^2\chi_+. \end{aligned}$$

The prefactor of $\chi_-$ approaches positive nonzero constants at both ends. Hence $\chi_-$ grows at the horizon and decays at infinity, while $\chi_+$ decays at the horizon and grows at infinity. Neither is square-integrable on the whole line. At the upper root $+i\alpha$, physical boundary conditions require decay at both ends, so neither kernel is a counterexample. The direct positive-potential proof excludes every other candidate solution there as well; it never divides by the inverse factor.

The usual algebraically special frequency is the lower root $\omega_*=-i\alpha$; for $\ell=2$, it is $-2i/m$. Its time dependence is damped, so it is outside the theorem. Its quasinormal and total-transmission classification requires analytically continued endpoint solutions, not just their leading exponentials: the horizon normalization is exceptional. The exact kernels and this distinction are treated in Maassen van den Brink 2000, arXiv v1, Eqs. (1.5)–(1.7), p. 2, § II, pp. 3–4, § III, pp. 5–6, and §§ V.B–C, p. 7, PDF. The linked transformation page works through that continued classification. Nothing about it weakens the upper-half-plane exclusion proved here.

Exceptional modes, gauge, and parameters

The radiative spin-two master variables used here are undefined for $\ell\lt2$, not zero fields that may be substituted into these formulas. The low-multipole equations must instead be solved in their actual harmonic sectors. Modulo unrestricted smooth coordinate changes, smooth vacuum solutions contain an even monopole mass shift and three odd dipole spin shifts, all constant in time; the even dipole has no intrinsic field. Restricting coordinate changes at infinity can leave momentum and mass-dipole frame data. Such frame data are not an additional radiative instability.

Within $\ell\ge2$, vanishing complete gauge invariants can leave a coordinate perturbation. Its representative need not be stationary or bounded in a poorly chosen gauge. A growing pure-gauge component or a singular coordinate representative therefore does not establish physical growth. Conversely, subtracting mass and spin without checking the allowed gauge and boundary class is not a complete metric stability proof.

For smooth asymptotically flat characteristic seed data with the specified finite weighted norms, Dafermos, Holzegel, and Rodnianski prove boundedness in an initial-data-normalized gauge and inverse-polynomial decay in a further horizon-normalized gauge, after separating a stationary linearized Kerr solution. Their result requires the reconstruction and evolution estimates absent from this short mode argument Dafermos, Holzegel, and Rodnianski 2019, §§ 10.3.2–10.4, Theorems 3–4 and Corollary 10.3, pp. 111–116. The next comparison in Boundedness and Decay of Linear Fields spells out these hypotheses and keeps scalar and gravitational results apart.

What mode stability does not imply

The theorem excludes exponentially growing separated modes. It does not alone give a bound for arbitrary initial-data evolution, a decay exponent, a quantitative resolvent estimate near the real axis, or completeness of quasinormal modes. For the latter questions one must control frequency limits and inverse transforms, not merely rule out a homogeneous solution at each fixed frequency.

For a simple logical comparison, the nonzero Jordan block $A=\left(\begin{smallmatrix}0&1\\0&0\end{smallmatrix}\right)$ has only the eigenvalue zero, yet $e^{tA}=I+tA$ grows linearly. This is not a model of the Schwarzschild master operator; it shows why a generic spectral exclusion is weaker than an evolution estimate. The positive master energy supplies a separate evolution argument in its stated norm. Decay needs additional control of the horizon redshift, trapping and low frequencies. Schwarzschild has no rotational superradiant band; the extra difficulty of Kerr belongs in Whiting Mode Stability of Kerr.

The static exclusion also does not remove the low-frequency branch structure responsible for late-time tails. Absence of a bounded homogeneous threshold solution is not analyticity of the full resolvent through that threshold. No decay rate follows from the contradiction above. Full nonlinear evolution belongs in Black-Hole Dynamics, and broader proof architecture in Mathematical Relativity.

Exercises

A decaying cusp

For $V=0$ and $\sigma\gt0$, consider $\psi=e^{-\sigma|x|}$ with $\omega=i\sigma$. It decays with the required sign at each endpoint. Why is it not a counterexample? Compute its distributional equation and the positive integral in the contradiction.

Solution

The derivative jumps by $[\psi']_0=-2\sigma$ while $\psi$ is continuous. Consequently

$$(-D^2+\sigma^2)\psi=2\sigma\delta(x),\qquad \int_{\mathbb R}\bigl(|\psi'|^2+\sigma^2|\psi|^2\bigr)\,dx=2\sigma.$$

It solves the homogeneous equation away from zero, but has a point source at zero and is not a smooth full-line mode. If one integrates separately on the two half-lines, the internal boundary term supplies exactly $2\sigma$. Endpoint decay is necessary for the proof, not sufficient to solve the homogeneous equation.

A smooth negative well

Let $k\gt0$, $V=-2k^2\operatorname{sech}^2(kx)$, and $\psi=\operatorname{sech}(kx)$. Show that $H\psi=-k^2\psi$ and compute the norm, derivative integral and potential integral. Which hypothesis of the Schwarzschild proof fails?

Solution

Differentiation gives

$$\psi''=k^2\psi-2k^2\psi^3,\qquad (-D^2+V)\psi=-k^2\psi.$$

The smooth field $e^{kt}\psi$ is a genuine growing solution for this toy wave equation. Its endpoints satisfy the required decaying branches at $\omega=ik$. Substitution $z=\tanh(kx)$ yields

$$\begin{aligned} I&=\frac2k,\qquad \int_{\mathbb R}|\psi'|^2\,dx=\frac{2k}3,\\ \int_{\mathbb R}V|\psi|^2\,dx&=-\frac{8k}3,\qquad Q=-2k=-k^2I. \end{aligned}$$

There is no contradiction: $Q$ is negative. This is a different full-line potential, not a Schwarzschild instability. Nor does the example prove that every potential with a negative region is unstable; one must examine its actual quadratic form and boundary problem.

A regular static solution with an outer boundary term

For the quadrupole odd equation, take a real constant $c\ne0$ and $\psi_0=cr^3$. Show that $H_-\psi_0=0$ and that it is regular at the horizon. Evaluate its quadratic form from the horizon to a finite radius $R\gt2m$. Why does the bounded static theorem not apply?

Solution

For $\ell=2$, $V_-=f(6/r^2-6m/r^3)$ and $D\psi_0=3cfr^2$. Differentiating once more gives $D^2\psi_0=6cf(r-m)=V_-\psi_0$. At the horizon, $\psi_0\to8cm^3$ and $D\psi_0\to0$. The lower boundary term therefore vanishes. At $r=R$ the finite-cut identity gives

$$\int_{-\infty}^{x(R)} \bigl(|D\psi_0|^2+V_-|\psi_0|^2\bigr)\,dx =\psi_0(R)D\psi_0(R)=3c^2f(R)R^5.$$

This grows without bound as $R\to\infty$. The horizon-regular solution is not bounded at the other end and is not an isolated asymptotically flat static perturbation. It illustrates the growing external-field branch. For its role as a complete odd invariant, rather than the vanishing time-differentiated odd variable, see the stationary reconstruction exercise.

References

• Dafermos, M., Holzegel, G., and Rodnianski, I. (2019). “The linear stability of the Schwarzschild solution to gravitational perturbations.” Acta Mathematica 222, 1–214. DOI.
• Donninger, R., Schlag, W., and Soffer, A. (2011). “A proof of Price’s law on Schwarzschild black hole manifolds for all angular momenta.” Advances in Mathematics 226, 484–540. DOI. Open PDF, arXiv:0908.4292v2, 12 July 2010.
• Maassen van den Brink, A. (2000). “Analytic treatment of black-hole gravitational waves at the algebraically special frequency.” Physical Review D 62, 064009. DOI. Open PDF, arXiv:gr-qc/0001032v1.
• Regge, T., and Wheeler, J. A. (1957). “Stability of a Schwarzschild Singularity.” Physical Review 108, 1063–1069. DOI.
• Zerilli, F. J. (1970). “Effective potential for even-parity Regge–Wheeler gravitational perturbation equations.” Physical Review Letters 24, 737–738. DOI.