Continued-Fraction Quasinormal Modes

Continued fractions turn a black-hole resonance boundary-value problem into a scalar complex root condition. Their power comes from enforcing the minimal solution of a recurrence, not from truncating a power series at a convenient order. A reliable computation must therefore expose the factored endpoint behavior, recurrence coefficients, fraction inversion, tail approximation, root search, and precision sweep. The goal here is the numerical construction and its checks, while the definition and physical meaning of quasinormal modes remain with Black-Hole Perturbation Theory.

Required background. Ordinary Differential Equations supplies Frobenius series, recurrence relations, and complex root finding.

Helpful background. Formulations of General Relativity fixes the background variables, Black-Hole Perturbation Theory develops the resonance problem, and Gravitational Radiation and Waveforms supplies the waveform conventions used to interpret its roots.

From endpoint factors to a recurrence

After removing the ingoing horizon and outgoing-infinity factors, write the regular remainder as a series in a compact radial variable $x$,

$$X(r)=B_H(r,\omega)B_\infty(r,\omega) \sum_{n=0}^{\infty}a_n x^n, \qquad 0\leq x\lt 1.$$

Substitution produces a finite-band recurrence. Gaussian elimination reduces a higher-band relation to

$$\alpha_n a_{n+1}+\beta_n a_n+\gamma_n a_{n-1}=0, \qquad n\geq 1,$$

with the starting equation $\alpha_0a_1+\beta_0a_0=0$. The coefficients depend on $\omega$, the background, spin weight, and angular eigenvalue. Because the desired series must converge at the outer endpoint, its coefficients form the minimal rather than dominant solution. Pincherle’s theorem then yields

$$0=\beta_0-\frac{\alpha_0\gamma_1} {\beta_1-\displaystyle\frac{\alpha_1\gamma_2} {\beta_2-\displaystyle\frac{\alpha_2\gamma_3}{\beta_3-\cdots}}}.$$

Leaver 1985, pp. 286–289, Eqs. (2)–(14) derives this construction with the endpoint factors included. Each implementation must state its mass normalization and Fourier sign before comparing roots.

Evaluate and invert the fraction

For an implementable Schwarzschild example, let $\widehat r=r/(2GM_{\rm phys})$, $\widehat\omega=2GM_{\rm phys}\omega$, $\rho=-i\widehat\omega$, and $x=1-1/\widehat r$. The axial gravitational field has $\ell\ge2$ and $\epsilon=3$. Its factors and coefficients are

$$X=(\widehat r-1)^\rho\widehat r^{-2\rho} e^{-\rho(\widehat r-1)}\sum_{k=0}^{\infty}a_kx^k, \qquad a_0=1,$$ $$\begin{aligned} \alpha_k&=k^2+(2\rho+2)k+2\rho+1,\\ \beta_k&=-2k^2-(8\rho+2)k-8\rho^2-4\rho-\ell(\ell+1)+\epsilon,\\ \gamma_k&=k^2+4\rho k+4\rho^2-\epsilon-1. \end{aligned}$$

Here $k$ labels a series coefficient, not an overtone. All complex powers use the real logarithm of their positive exterior arguments before continuation in frequency. Substitution into the radial equation yields these three coefficients independently of any root search; compare Leaver 1985, § 2, Eqs. 3–8, p. 287. The same formulas give the massless scalar and electromagnetic equations with $\epsilon=-1$ and $0$, respectively, but the quoted frequency below is gravitational.

A transparent finite-depth evaluation starts with $C_{N+1}=0$ and recurses backward,

$$C_k=\frac{\alpha_{k-1}\gamma_k}{\beta_k-C_{k+1}} \quad(k=N,\ldots,1), \qquad F_N(\widehat\omega)=\beta_0-C_1.$$

Zeros of $F_N$ approximate the infinite-fraction roots. A small denominator must be handled by rescaling or an alternative evaluation rather than blindly dividing; it can be a pole of the finite fraction rather than a mode. Compare $N=100,200,400$ and independent working precisions before accepting digits.

A finite truncation with a zero tail is easy but can converge slowly for damped overtones. Evaluate the fraction backward from depth $N$, preferably with a large-$n$ asymptotic estimate for $a_{n+1}/a_n$. Increase $N$ until the root stabilizes, then increase working precision independently. Modified Lentz evaluation can help with near-zero denominators, but it does not replace a truncation study.

An $n$th inversion places the $n$th recurrence denominator at the center, joining a finite upward fraction to an infinite downward fraction. Different inversions have the same exact roots where the algebraic divisions are valid, but condition different overtones differently. At a recurrence coefficient zero or pole, return to the original recurrence before inferring equivalence; multiplying by a vanishing denominator can create a spurious root or remove a special solution. A generic physical root should persist under neighboring nonsingular inversions. Search the complex plane with contours or multiple seeds before Newton polishing, because Newton’s method finds only the basin containing its initial guess.

For Kerr, the angular eigenvalue and radial fraction are coupled. Iterate or solve the two complex conditions simultaneously, recording the angular branch connected to the intended $a\omega\to0$ harmonic. Mislabeling that branch can produce a precisely converged but physically wrong mode. Berti, Cardoso, and Starinets 2009, §§4.6 and 5.3 review the continued-fraction construction and the coupled Kerr problem.

Schwarzschild benchmark

Let $M_{\rm phys}$ be the physical black-hole mass and $M_{\rm geom}=G M_{\rm phys}$ its geometric length. For axial gravitational perturbations with $(\ell,n)=(2,0)$, Leaver uses $2M_{\rm geom}=1$ and reports approximately

$$\omega=0.747343-0.177925i.$$

Restoring the mass convention gives

$$G M_{\rm phys}\omega=M_{\rm geom}\omega \simeq0.3736715-0.0889625i.$$

The conversion is part of the benchmark, not a cosmetic rescaling. Leaver 1985, Table 1, p. 293 supplies the reference spectrum at the displayed precision.

A useful evidence table records depth $N$, inversion index, precision, initial seed, unpolished residual, polished root, and change from the previous run. Test $N$, $2N$, and $4N$ at two or more precisions. Evaluate the original differential equation or a two-sided Wronskian at the final frequency as an independent check; a tiny continued-fraction residual alone can share the same recurrence-algebra error.

The accepted tolerance must be commensurate with the claimed digits. If the last two digits change under the tail model or precision, report fewer digits. Highly damped modes require deeper truncation and may cross recurrence singularities; the fundamental benchmark does not establish uniform reliability for all overtones.

Environment, rerun, and failure modes

Pin the recurrence generator, symbolic simplification version, complex arithmetic library, working precision, tail prescription, truncation sequence, inversion indices, angular convention, and code revision. A clean interface could be qnm-cf --config schwarzschild-s2-l2-n0.toml --output run/; it should emit coefficients, roots, residuals, convergence sweeps, and environment metadata. This describes the record an executable must produce and does not imply that a repository solver ran.

Reject a mode if it disappears under increased depth, moves with inversion, has the wrong damping sign for $e^{-i\omega t}$, changes when the angular branch is tracked continuously, or fails an ODE residual or Wronskian test. Catastrophic cancellation can make a fraction residual look small at machine precision while the root is inaccurate; higher precision and backward-error estimates diagnose that failure. A sequence of roots is not evidence of completeness unless the search region and missed-root test are also specified.

Diagnostic

A run in Leaver’s $2M_{\rm geom}=1$ normalization returns $\omega=0.747343-0.177925i$. A comparison script labels it wrong because it expects $M_{\rm geom}\omega=0.3736715-0.0889625i$. The continued-fraction residual is $10^{-13}$. Which result is correct?

Solution

Both numerical values represent the same dimensionless frequency under different mass units: multiplying the first by $M_{\rm geom}=1/2$ gives the second. The comparison failed because it omitted the normalization conversion. Record $M_{\rm geom}\omega=G M_{\rm phys}\omega$, the Fourier convention, and the definition of the radial coordinate; then verify the converted root with another inversion and an independent differential-equation residual.

References

• Berti, E., Cardoso, V., and Starinets, A. O. (2009). “Quasinormal modes of black holes and black branes.” Classical and Quantum Gravity 26, 163001. arXiv.
• Leaver, E. W. (1985). “An analytic representation for the quasi-normal modes of Kerr black holes.” Proceedings of the Royal Society A 402, 285–298. DOI.