WorksPhysics and Astronomy
Metric reconstruction in a Bondi-like gauge
Abstract
Metric perturbations obtained from a Hertz potential in an ingoing radiation gauge (IRG) generically grow in an outgoing coordinate at future null infinity, and are therefore not asymptotically flat in the Bondi sense. Hollands and Toomani showed that this growth can be removed, within the Green–Hollands–Zimmerman (GHZ) framework, by an IRG-compatible residual'' gauge transformation together with a carefully chosen choice of the undetermined integration constants'' of the Hertz potential. In this work we turn that proof into a practical algorithm. We show that this entire completion is generated by a single ``Held Master Scalar'' (HMS), which can be thought of as a function on future null infinity. The HMS is obtained from the leading asymptotic piece of the -Weyl scalar through a Teukolsky–Starobinsky inversion. We give explicit formulas for the four integration constants of the Hertz potential associated with the -Weyl scalar, and for the residual gauge vector in terms of the HMS, thereby bringing the metric perturbation into a Bondi-like gauge. In Schwarzschild the entire construction is diagonal in -modes and every completion field is available in closed form; in Kerr the construction is tri-diagonal. We have numerically implemented the full scheme in and in -modes, including the corrector piece of the GHZ framework. We illustrate our scheme numerically for a point-particle on a circular orbit in a puncture scheme in Schwarzschild and in Kerr, by demonstrating the Bondi-like fall-off. We also find agreement with the known Detweiler redshift computations in Schwarzschild and Kerr. Our method directly gives the Bondi news, mass-, and angular momentum aspects, for which we give exemplary results in Kerr.
Introduction
Large-mass-ratio inspirals of a comparatively small companion into a black hole present interesting conceptual challenges, such as the problem of scaling solutions of the Einstein equations, described from a physical perspective in [1], and more recently [2] in a limited scenario permitting complete mathematical control. Because such systems are among the principal targets of the Laser Interferometer Space Antenna [3] there is also a very strong practical motivation for developing accurate models capable of describing the dynamical evolution of the binary, and the emitted gravitational waves, over a very large number of orbits. The large mass ratios of up to expected in practice [4] are so far inaccessible to fully numerical schemes [5], in particular since a quite large number of such simulations corresponding to varying orbital- and black hole parameters would be needed for wave-form modeling [6, 7].
The current state of the art are instead perturbative, semi-analytical methods within the self-force scheme, see [8, 9] as general references. In this scheme, the near field of the small object is typically isolated at various orders in perturbation theory—in practice, second order is considered necessary [10]. The residual field, i.e., the gravitational perturbation field with the small object field extracted, represents the large-scale “trend” of the gravitational field as opposed to the—often infinite-local perturbation, and is thought of as responsible for the self-force on the particle. This self-force gives rise to a gradual change of the orbital parameters, which must be updated throughout the dynamical evolution. Since the gradual change is very slow compared to the time-scale of the gravitational waves (≳ orbital period), it is possible to separate the problem into an evolution equation for the orbital parameters governed by a slow time, and perturbative calculations, at various orders, in a fixed Kerr background geometry and fixed orbital parameters, governed by a fast time [8, 10, 11].
In this paper, we focus on this second problem, which is the main input into the first. In each perturbative order, our problem, therefore, is basically to find a solution, , to the linearized Einstein equation on a fixed Kerr background, where is the linearized Einstein operator1, and is the stress tensor at the respective order. In case represents the residual field at that order, will include, besides the point-particle stress tensor associated with the smaller object itself, also pieces arising from the lower order residual metric perturbations, and with the subtraction pieces of the near field (“punctures” [13]), up to the same order. The latter are determined entirely locally [14, 15, 16, 17] and, while complicated, do not pose an essential problem.
There are many different approaches to solving in practice. A common benchmark criterion is that one would like an in a well-behaved gauge2. Here, “well behaved” means e.g., that one would not like the metric to contain artificial singularities, or a non-decaying behavior at null-infinity, non-smooth behavior at the horizon, etc. In particular, it is highly preferable to find in a Bondi-like gauge, such that asymptotic quantities like BMS-charges, memory, or fluxes are readily identifiable, or such that the contribution of to the stress tensor at the next—in practice second—perturbation order is manageable.
In the present paper, we focus on the metric reconstruction method, originally due to [18, 19] in the vacuum () case, but mention e.g., the recently developed Lorenz gauge method [20, 21] or [22] as interesting alternatives. Even though, as our application, we will consider a concrete puncture scheme yielding the residual part of the metric perturbation from an “effective” stress energy tensor at first order, and for a circular equatorial orbit of Kerr, our scheme is general and conceptually independent of such restrictions as long as has sufficient, and standard for radiating sources, decay (6) at future null infinity.
The broad aim of reconstruction methods is to obtain the metric perturbation from the spin Weyl scalar , which obeys Teukolsky’s decoupled separable equation for each mode [23]. Such mode decompositions are particularly suitable for systems with periodic sources representing orbiting point-like companions. From , one then obtains a so-called Hertz potential, , by a well-known and easy inversion [24], and then the metric perturbation as , where is a tensor-valued “reconstruction operator” [19, 25]. Actually, this scheme by itself is inconsistent for generic non-trivial sources . But as Green–Hollands–Zimmerman (GHZ) showed [26], it can be adapted by adding a correction piece, determined by transport equations, to . A practical way to combine the GHZ scheme with the “no-string-gauge” methods by [13, 27], for determining the residual metric perturbation, was proposed later in [28], and implemented for a circular orbit in Schwarzschild in [29]. While the approach [28] avoids a stringlike gauge singularities emanating from the source, it does not by itself produce a metric perturbation in an asymptotically flat gauge.
While far from obvious, it is physically plausible that actually has an asymptotically flat gauge representative. However, even if it does (which we assume is the case and provide evidence for3) it is non-trivial to actually find this gauge representative, see [30, 31] for recent
works investigating and using asymptotically flat gauges in a self-force setting. In this paper, we instead follow a recent extension of the GHZ metric reconstruction method due to [12], which delivers a gauge vector field (and at the same time a particular choice of the non-unique Hertz-potential ) achieving a Bondi-like gauge. The total metric perturbation is hence
wherein is the Kerr metric, and a metric perturbation corresponding to an infinitesimal change4 of the Kerr parameters . The reconstructed piece and the corrector by design are in ingoing radiation gauge (IRG)—meaning a vanishing contraction with the outgoing principal null direction —while may be presented in IRG by a gauge choice [12 App. C]. Because is a “residual” i.e., IRG compatible gauge vector field [32], the entire expression (1) therefore is in IRG. Our Bondi-like gauge means this, plus the remaining four decay conditions (7) on the null-tetrad components of .
Building on [12], we will present a remarkably simple algorithm for obtaining both , and the particular choice of at the same time. This algorithm is as follows:
Find the leading order asymptotic piece of the “0” perturbed-Weyl-scalar component (the piece), which is used to
Find a certain “Held Master Scalar” (HMS) , where indicates a quantity that is intrinsic to future null infinity, or more pragmatically, a function of in an outgoing Eddington-like coordinate system. is determined from by an angular Teukolsky-Starobinsky relation.
Apply certain explicit geometrical differential operators to our HMS , to obtain both the leading pieces (the coefficients of the -pieces of , which are undetermined from the inversion relation linking and ), and the components of the residual gauge vector field .
Since the corrector can be found by three simple integrable [28] transport equations [26] with explicit Green’s function [33] if we want, since it is known how to solve Teukolsky’s equation for with retarded boundary conditions, and since it is known [24] how to solve the inversion problem , the problem of finding as a solution to in reconstructed form and Bondi-like form near future null infinity is solved by (i)—(iii).
We can consider both a so-called “-mode” scheme (factoring out only the -dependence of the solutions), or a full “-mode” scheme (factoring out also the spheroidal harmonics) for (i)—(iii). In general, our implementation architecture supports both schemes, though we mostly describe the -mode scheme in Kerr. As an exception from this, we use an -mode scheme for solving the Teukolsky equation for . But we note that also this step could be based entirely on an -mode scheme once is given as a numerical solution on an -grid covering the exterior of Kerr, e.g., using the very recently developed spectral method by [34]. As a consistency check of our scheme, we numerically check that is in a Bondi-like gauge, and we compute the Detweiler redshift quantity.
Expectedly, the -mode scheme is particularly straightforward in Schwarzschild-spacetime, but we also outline the necessary changes in Kerr. These are mainly due to the fact that the angular GHP-like operators do not act as simple raising or lowering operators, unlike Schwarzschild [35]. We have successfully performed consistency checks of some of our main numerical findings between the – and mode schemes.
The concrete proof-of-concept type application of our general method that we consider is to find the first order residual metric perturbation within a puncture scheme [13], for circular equatorial orbits in Kerr. Here, the concrete problem to be solved is , where is an effective source taking into account the original point particle source,
( is the unit tangent) and additionally a stress tensor corresponding to the singular field near the body, called a “puncture”, . In formulas,
In concretely constructing in our setting, we partly rely on previous non-trivial works [16, 29]. The full metric perturbation at first order would then be , where would e.g., be used to compute the self-force near the particle. Although we could do this, we will instead mostly concentrate on the asymptotic region where the technical advance of our new method plays out.
We finally note that, if one can solve the non-trivial problem of constructing at higher orders in perturbation theory, which includes the problem of punctures at higher orders [9], our method and -mode implementation pipeline would directly generalize to this case: We would again find using a some Teukolsky solver, then fit the inversion relation to get , then use our integration implementation to find , and then go through the same steps (i)—(iii), which does not use particular features about the orbit or perturbation order.
Conventions for Kerr background and differential operators
Our conventions regarding the Kerr-metric, the Newman-Penrose (NP) [35] – and Geroch-Penrose-Held (GHP) formalisms [36], are the same as e.g., in [26], [12], where one can also find reviews of these formalisms. In particular, the signature of the metric is . As in these references, we will actually mainly use Held’s modified GHP calculus [37, 38], in the same conventions as stated in [12]. The reader is not expected to have any familiarity with these formalisms since for our concrete application, it is only necessary to recall the definitions of certain coefficient functions and differential operators.
These are stated in the Kinnersley frame; see, e.g., [26], [12], and Boyer-Lindquist coordinates called , or in outgoing Eddington coordinates called for Kerr, which are related by
where . In these conventions, the coefficients that we require, are
, , , , and a certain geometric basis of directional derivative operators, see Table I.
| Held quantity | Value | Spin weight |
| \thorn | ||
| \thorn' | ||
Table I. Summary of Held’s GHP quantities in outgoing Eddington coordinates and the Kinnersley frame (taken from [26]). On a mode with dependence (or with in Boyer–Lindquist coordinates), one puts and . and are thereby equal to the spin-raising and spin-lowering operators when acting on spin-weighted spherical harmonics; see, e.g., [28 App. C].
Held’s -notation: As a matter of notation, in the Held-formalism, a on a quantity intrinsically means a GHP scalar that is annihilated by the GHP-operator ; pragmatically this just means that it is independent of in Eddington coordinates in our conventions. By construction [37, 38], any Held operator- or NP coefficient in Table I preserves this property, unlike the usual GHP operators.
The quantity is the spin weight of a GHP scalar or operator [36]. It must be taken into account for every quantity these operators are acting on, and a GHP operator with the indicated weight will shift these by the amount associated with the operator. A nice feature of the above Held operators– and scalars is that they lead to manifestly gauge invariant equations, which thereby could be converted readily from the Kinnersley frame to any other one by a simple local boost/rotation factor associated with the NP tetrad change, see, e.g., [12]. This would be practically useful if we wanted to convert our abstract formulas e.g., to the Hartle-Hawking– or Carter frames, and is also of course geometrically much more satisfying.
The asymptotically flat reconstruction scheme
Here we describe our asymptotically flat reconstruction scheme for obtaining (1). It is extracted from [12, Thm. 4] and its derivation in [12 App. J]. The theorem applies rigorously to (smooth) solutions to the linearized Einstein equations (EE) , under the following
Assumptions:
The NP-components of the stress energy tensor in our Kinnersley frame has the standard5 behavior
towards , i.e., as , at fixed . The conditions are equivalent to the Bondi coordinate-type conditions given e.g., in [39].
There exists some gauge such that the NP components of have asymptotic expansions with decay orders
as , at fixed . These conditions would be satisfied if one were to assume that there exists a Bondi-gauge in the usual (coordinate) sense for , see [12 App. D]. In an IRG, all NP components involving at least one “” would vanish.
There exist such that, subtracting a corresponding zero mode in IRG satisfying (7) [28] (see, e.g., [12 Eq. C.1] for a concrete formula)
the leading expansion coefficients in the asymptotic expansion all go to zero as .
At the Bondi news tensor satisfies the usual Price decay law [40] as .
Theorem: [12 Thm. 4] There exists a gauge in which takes the form (1), and such that it is in addition in a Bondi-like gauge.
Bondi-like gauge: This means that is in IRG, i.e. , and that the remaining four asymptotic behaviors in (7) hold for the non-zero NP components. This gauge could easily be converted to an actual Bondi gauge in the traditional (coordinate) sense [12 App. D], but for the purposes of extracting quantities associated with gravitational radiation, or for using to build the source at the next perturbative order, the Bondi-like gauge appears just as good if not better. Of course in practice, one does not know from the outset, so one does not know whether conditions 2)–4) on the metric perturbation hold unless one can appeal to abstract mathematical theorems to this effect! We are unfortunately not aware of such abstract theorems in the specific self-force settings below, mainly due to the fact that the particle exists forever. We therefore need to assume that the conclusions hold. Then, we can extract the following prescription from [12 Thm. 4], to actually find the pieces , , , in the decomposition (1) for .
Algorithm: Step 1) The perturbed Teukolsky scalar6
of spin-weight , is determined through Teukolsky’s equation [23],
where is Teukolsky’s source operator, and is Teukolsky’s wave operator, see App. § 2.0.2 for their concrete expressions in terms of our operators in Table I. We assume retarded boundary conditions7. Then by the peeling theorem8
Step 2) The Hertz potential is obtained from the well-known [41] inversion relation9
We then have a unique minimal solution, i.e., such that . However, this minimal solution is not the one appearing in the asymptotically flat IRG metric (1)10. Instead, we have to add to the minimal solution specific terms called completion pieces :
These terms are growing as for , and this is actually essential: their contribution under the reconstruction operator combines with , in (1), furthermore leading to a delicate cancelation of various subleading terms and leaving in the end a perturbation with Bondi-like falloff at future null infinity [12 Thm. 4]. For this, a specific choice of the growing terms must of course be made; in fact, by working through the proof of [12 App. J], one can see that the entire completion is generated by a single Held Master Scalar (HMS) of spin weight , defined by
The four Held integration constants appearing in (13) are determined from the HMS by the triangular system
noting that in modes.
Step 3) We build —which give the residual gauge vector field [32], by differentiation formulas found, e.g., in [12 App. E]—from
from the HMS , noting again that in modes. In fact, for (1), we only need to evaluate not itself but , which is determined directly by through the formulas in App. § 2.0.1.
Step 4) We determine from the GHZ transport equations by a backward integration in from , see, e.g., [28 App. H], and we determine the zero mode by Abbott–Deser type integrals from , see e.g., [28]. These last steps are actually even easier in our model problem as we will see, since the stress tensor is non-zero only locally near the particle.
We thereby have all pieces in (1) giving us the metric perturbation in a Bondi-like gauge in IRG. For completeness, the spin weights of every object appearing in the construction are collected in Table II.
| Spin weight | ||||||
Table II. Spin weights of the GHP and reconstruction fields entering the adjusted reconstruction. The second row is the sign appearing in the conjugation rule for mode sum coefficients (23).
Circular orbit in Schwarzschild
Here we outline our scheme for a point particle on a circular orbit in Schwarzschild. Due to the high symmetry, it is convenient to expand any spin- quantity in this spacetime in modes as
| Spin weight | |||||||
Table III. Spin weights of the metric perturbation.
where our conventions for the spin-weighted spherical harmonics are the same as in [12]. For generic orbits, possible frequencies are the harmonics of the basic orbital frequencies labeled by three integers. For a circular orbit as considered here, this reduces to , with , where is the radius of the orbit.
Following [16, 17], a puncture for such an orbit is obtained by the method described in Sec. V A. Then the NP components of the effective stress energy tensor (3) are formed using this puncture and the stress energy tensor of the particle (2). Using a version of the semi-analytic tools developed in [16, 17], see Sec. V A, one then obtains the -mode expansion (23) of the corresponding effective Teukolsky source in (10). Its expansion coefficients are numerically given as an array on a discrete set of -values per mode. The order of the puncture governs the fall-off of the mode amplitudes in .
With these at hand, one obtains the mode expansion coefficients of the perturbed Weyl-scalar (9) by solving the mode version of Teukolsky’s equation (10), i.e., the radial Teukolsky equation for the modes coefficients (23) of , with the effective Teukolsky source, and with purely downgoing (at the horizon ) and purely outgoing (at future null infinity ) boundary conditions. We have numerically checked that this satisfies peeling (11), and thereby we can extract the mode-sum coefficients of the leading coefficient,
by fitting against . Recall that is the Eddington coordinate, and that quantities with a by construction have no -dependence. Either by the modal inversion scheme [24], or by simply integrating four times against beginning from (in practice an -value much larger than both and the particle’s orbital radius ), we get the minimal solution of the inversion relation (12) for the Hertz-potential, in -modes.
Next, we solve the -mode version of our HMS (15). In doing so, we use that in modes, and that the combination of angular operators in (15) is precisely the combination appearing in the Teukolsky-Starobinsky identities [41, 42]—where both observations generalize to Kerr. The -mode coefficients of the HMS are thereby found to be:
where is a Teukolsky-Starobinsky constant11. Its value in Schwarzschild is .
Using the ladder operator properties of Held’s angular derivatives in terms of spin-weighted harmonics recalled e.g., in [28 App. C], the four Held completion pieces (15) in the Hertz potential are similarly found to be
The residual gauge vector ’s building blocks (16) needed to compute for the formulas in App. A
are
where “+ c.c.” denotes the mode-space real part, i.e. the addition of inside the same prefactor. We also defined and .
In our inward integration scheme, the corrector , obtained via the GHZ transport equations, vanishes to the right of the interval around the particle where the puncture is windowed. Thus, outside this window, the reconstructed residual metric (1) is in this case simply
The NP components of are obtained by applying the formulas of App. A, which drastically simplify for due to , mode by mode to (21) in automated form using the spin-raising and lowering properties [35] of Held’s angular derivatives as given in Table I. is the zero mode in our Bondi-like gauge, see e.g., [12 Eq. C.2], which is determined in the present case by , , with the particle’s mass and the particle’s orbital angular momentum.
Since we have the full corrector , we of course have the entire metric perturbation (1) in a mode-decomposed form with coefficients dependent on all the way from the horizon to infinity. We will use this to compute the Detweiler redshift at the particle in Sec. V. When we compute asymptotic quantities, see Sec. V for numerical results (in Kerr), we do not need in the present first order model application of our scheme. On the other hand, at second order, the effective source no longer is supported only near the particle, so the second order corrector is supported all the way from the horizon to infinity and would be needed for the computation of asymptotic quantities as well.
Circular orbit in Kerr
As our main example, we derive the first order metric perturbation in our asymptotically flat IRG for a circular orbit in Kerr. We could proceed by a mode expansion similar to the Schwarzschild case, where spin-weighted spheroidal functions would now appear. While such an “-mode” scheme is possible, we here describe an “-mode” scheme, which seems more promising in general for inclined, elliptical orbits as needed in reality. However, we will briefly describe some changes necessary in order to generalize the -mode scheme to Kerr in Sec. V G, as could be useful for moderately inclined and elliptical orbits.
In the -mode scheme, quantities are generally expanded as
as opposed to (17). Consequently, the mode amplitudes in a sense already include an infinite -summation. The downside is that, of course, decoupled ordinary differential equations (ODEs) such as the radial and angular Teukolsky equation now become partial differential equations (PDEs) in . The modes therefore have to be stored as an array of number on a 2-dimensional -grid, and the numerical solution is considerably less straightforward. Fortunately, it turns out that the only PDE in our scheme is the Teukolsky equation (10). On the other hand, the GHZ transport equations for the non-zero corrector components remain ODEs in (for each fixed -value), and the equation for the HMS is an ODE in . Furthermore, the inversions needed to find the completion pieces from the HMS in Eqs. (15), (16) remain algebraic since on -modes.
Comoving punctures
The puncture is a local, analytically constructed approximation to the singular part of the metric perturbation near the particle’s worldline, and it is only ever needed in a neighborhood thereof cut out by a window step-function in . We use the high-order construction initiated for a circular orbit in Schwarzschild and a scalar toy problem in [16], and later generalized to circular orbits in Kerr in the spin-2 case by [17].12 It is based on a coordinate system centered on the particle rather than Boyer–Lindquist coordinates. Let be the worldline with proper time and four-velocity , and let be an orthonormal tetrad along which is parallel transported, i.e. and whose spatial triad is Fermi–Walker transported. Local spatial coordinates are obtained by shooting geodesics orthogonal to and then Lie-dragging the resulting labels along the congruence,
so that in a whole neighborhood of in the coordinates , and the particle sits at for all .
The payoff of (36) is that comoving stationarity becomes triviality in coordinates. If a tensor field is stationary in the comoving frame, , then its tetrad components obey , and the field is a function of the spatial comoving coordinates alone. The local singular field, of the point particle, and the particle’s stress tensor, [2], are of exactly this type. Its dependence on is fixed by subjecting its local expansion in to the EEs , thereby pushing the information about the particular orbit into the expansion coefficients. This expansion has the general form in terms of frame index components for our NP-tetrad [16, 17],
where where the constant coefficients are determined by the EEs. The order of the puncture dictates the number of expansion terms to be computed but it is always finite13. This general structure survives for a generic bound orbit, but we now specialize to the case of a circular orbit, recording the minimum needed to follow the -mode reduction; the construction is given in full in [43], which builds on [16], but we give a brief outline in App. C as forward guidance.
In order to decompose into -modes (35), we require the transformation from body-fixed coordinates to Boyer–Lindquist coordinates in order to integrate (37) against . To write these down for our circular orbit, we define , which is a corotating phase, and we introduce
together with a comoving time that is a fixed linear combination of and and employ the Mathematica [44] package [45] for coordinate based tensor manipulations. The constants above are fixed by the orbital radius , as , and
Quantities may be pulled back from body-fixed to Boyer–Lindquist coordinates by these formulas.
Two reductions bring (37) into a form whose -modes can be evaluated in closed form. First, powers of beyond the first are traded for powers of , which leaves at most one explicit factor of .
Second, that remaining factor is replaced by the single-valued angular variable , using and truncating at the order of the puncture. The result is
now with , so that only four types of angular integral can occur.
Indeed, since , the -mode amplitudes of (40) can be considered as integrals over at fixed . Since with is linear in , every such integral reduces to a single master function of the form
now with , through the dictionary
The entire -mode puncture is therefore assembled from associated Legendre functions of a single argument, with no numerical angular quadrature [16]. This is what makes the puncture side of the -mode scheme cheap enough to be recomputed at every radial grid point. For generic orbits, part of this structure usefully survives, see App. C.
Effective source
Once we have as in (40), we multiply it by a radial window function supported on some suitable (which could be varied for numerical verification), and afterwards apply the linearized Einstein-operator, . This gives us the effective source [3], in which the -function localized piece corresponding to the particle is cancelled analytically. Since the procedure has already been implemented e.g., by [29], we will not describe it in detail here. After this, we apply Teukolsky’s source operator to get the source in Teukolsky’s equation , see App. B.
Integration of GHZ equations for corrector
The corrector does not enter the construction of the HMS and completion pieces require for our Bondi-like gauge, but it is required in the full metric perturbation (1).
In IRG, the corrector has three non-vanishing NP components, . They are found by integrating the GHZ transport equations for them form a triangular system of radial transport equations along the outgoing principal null congruence [26], see e.g., [12 Sec. 3.1.3]. These can be solved by integration [28 App. H]. In principle, in our general scheme [12 Sec. 3.1], see Sec. III, we integrate the GHZ transport equations inward from with vanishing data there. In the application considered, the effective source (3) has compact radial support in the worldtube by construction, so vanishes identically from the outer edge of the worldtube out to . In that whole region the metric is given by (22) with no corrector at all14.
The -mode reduction of the GHZ transport equations is again the substitution (23), after which each level is an ordinary integration in at fixed . The two-dimensional problem therefore decouples into independent radial problems, and this—not the Teukolsky equation—is the sense in which the corrector is the easy part of the -mode scheme. Two of the three levels require a double integration in ; the third only a single integration.
Each radial problem is discretized by Chebyshev collocation on two patches meeting at the particle, with interface conditions imposed on the value and the derivative, so that the non-smoothness of the windowed effective source at is never differentiated across.
Weyl scalar and Hertz potential
This is the one genuinely hard step of the -mode scheme, and the only place where a PDE has to be solved. The Teukolsky wave operator and source (obtained from ) for take a compact form in the Held-type operators and coefficients of tbl:1, which we record in App. B, see (B1), (B1). In outgoing Eddington coordinates, in which and , the resulting equation is, schematically
with coefficients , , that are obtained by substituting the values of the Held quantities in tbl:1 into the formulas of App. B. These are operator-valued coefficient functions of : only is a plain function of , while contains and separately, and contains them only through the combination . Note also that carries no radial derivatives.
In the -mode reduction we should replace and in (31). The mixed derivative and the term inside then become algebraic, and
Because and still carry by tbl:1, (32) remains a PDE in at each .
We solve it with retarded15 boundary conditions. In the future, we intend to use a multidomain spectral method [34, 46], with domain boundaries placed at the edges of the worldtube so that the windowed effective source is smooth on each domain. However, in the present analysis, we use standard -mode Teukolsky solvers in this step [47, 48, 49]. We then verify a posteriori that the solution satisfies peeling (11), so that may be read off as the coefficient of .
By the inversion relation (12), the Hertz potential is obtained by integrating inward in four times—in modes, at is assembled from at by the conjugation rule quoted with Table II—starting from a radius much larger than both and taking vanishing data there. This selects precisely the minimal solution , and the growing in pieces that are left undetermined by (12) are exactly the Held integration constants supplied by the completion scheme of Sec. III, where we should concretely substitute the entries of tbl:1 into (15) with , .
HMS and residual gauge vector
For this completion, and for the residual gauge vector in (1), we next require the HMS , determined by (14). In -modes, this is an 8th order ODE in . If we do not want to perform an -mode reduction, which would render this equation algebraic by (19), we need to solve this ODE. Intrinsically, the HMS is smooth, i.e., it is smooth in the sense of a mathematical object called a section in a “GHP/Held-bundle”, see, e.g., [12 Sec. 2.1] for explanation. With our conventions of coordinates and NP tetrads, this corresponds to certain boundary conditions at the poles. We solve our ODE by collocation on Legendre-Gauss–Lobatto nodes in , subject to these boundary conditions. Two ingredients make this practical.
The first is pole factorization. For a spin-weight- GHP/Held quantity at azimuthal number intrinsic smoothness translates into a certain (singular) endpoint behavior in our NP-frame. It is convenient to factor this behavior out and work throughout with the regular reduced amplitude defined by
In this representation Held’s angular derivatives (see Table I) act as first-order operators with polynomial coefficients, of the general form
and similarly for with coefficients , and the opposite sign on the undifferentiated term. Here and are polynomials of degree at most one in , fixed by the signs of , and is the Legendre differentiation matrix representing . Because the removed factor in (48) vanishes at the poles, no quantity in the chain ever has to be evaluated at a coordinate singularity.16
The second is that (15) is solved as a dense linear system rather than integrated. The left-hand side is assembled by eight successive applications of (49), four of and four of ; applying this composite operator to each collocation basis vector in turn gives the matrix column by column (here is the number of collocation points). The system is row- and column-equilibrated, factorized by full-pivot LU (a method that decomposes a matrix into a product of a lower triangular matrix and an upper triangular matrix to enhance numerical stability), and refined iteratively; the relative residual and the condition number of the equilibrated matrix are returned alongside the solution and are the primary diagnostics of the step.
With in hand, the four Held integration constants (15) and the components of (16) are formed on the same grid by further applications of (49). Since with the growing completion pieces is a cubic polynomial in (5) and is second order in -derivatives, each NP component of the adjusted residual field perturbation is returned as
with the two Laurent polynomial coefficients stored as arrays over the nodes, separately for the reconstructed piece, the gauge piece, and their sum. The Bondi falloff is then simply the statement that at every collocation node for the , and NP-tetrad components17, which can be checked pointwise without any mode sum.
Numerical results for
We begin this section with benchmarks of the -mode chain. So that every intermediate object can be compared against a “closed-form” reference, the runs reported here were first performed at : the operators, the grid and the reconstruction are the Kerr ones described above, but the background is Schwarzschild, where the sums are available in closed form, see Sec. IV. All figures use a circular geodesic at with , fitted data summed through , and Legendre–Gauss–Lobatto (LGL) nodes in .
Finding the HMS . Figure 1 compares the collocation solution of the -mode scheme of § 2 E for the HMS with the -scheme in Sec. IV (-summed HMS) for . The maximum relative complex HMS errors are , and , with relative operator-equation residuals for the analytic HMS of , and . The errors fall steeply with , as expected from the pole factorization (48): the residual function becomes smoother as grows, so the spectral representation converges faster. The equilibrated condition number of the discrete operator falls correspondingly, from at to at , which is why the equilibrated full-pivot solve with iterative refinement is used rather than a plain factorization. This validates both the -summed input and the -mode seed inversion.
Figure 1. Validation of the -mode HMS for . Top row: the modulus of on the LGL grid (circles), obtained by solving (34) in , against the analytic Schwarzschild sum (solid). Bottom row: the pointwise complex error relative to . The quoted operator residual is the relative residual of the reduced angular equation when the analytic form is substituted, and measures the discretization rather than the solve.
Cancellation and Bondi falloff. The central claim of Sec. III is that the linear in and constant asymptotic expansion orders carried separately by and cancel in the sum (34), leaving a perturbation with Bondi-like falloff (here hatted indices indicate NP-frame components). Figure 2 exhibits this cancellation in the -mode scheme. Across 1254 nonstationary radiative mode components through , the largest relative cancellation error in the growing and constant Laurent coefficients is , i.e. at the level of double precision. Direct finite-radius sums at for show , and approaching between and , with fitted large-radius slopes between and . The residual drift from is the expected contamination from the subleading and terms at finite radius, and shrinks as the fitting window is pushed outward.

Figure 2. mode scheme reference for the adjustment (22), built from the closed-form Schwarzschild components summed at fixed through , evaluated at , . Each panel shows the reconstructed piece , the gauge piece , their sum, and the fitted slope coefficient. We clearly see that the individually growing pieces cancel to leave a tail.
-mode scheme. Figure 3 repeats the radial test with the full -mode pipeline in place of the modal reference: the and adjustment scalars are obtained by collocation, the reflected partner being the conjugate of the reduced adjustment scalars since the angular harmonics already supply the odd- reflection phase, and all subsequent angular derivatives are taken spectrally. At the interior node the six plotted components again have large-radius slopes between and . Across all interior nodes the largest relative difference from the modal reference of Figure 2 is for and for , while the largest relative residual in the growing and constant Laurent coefficients is . The cancellation is therefore not an artifact of the mode sum: it survives the spectral realization of the Held operators, which is the only form

Figure 3. The same six components as Figure 2, constructed instead from the -mode scheme on the grid and evaluated at the interior node . The reconstruction forms through , applies the spectral Held angular derivatives for and , and sums the complex Laurent coefficients directly at each radius, without ever projecting back onto individual modes.
available in Kerr.
Detweiler redshift. This is among the simplest observables in the self-force context at linear order. Denoting the (non-normalized) circular orbit’s tangent by , the first order gauge invariant redshift observable relative to a distant observer tangent to is [8, 50]
in a gauge such that . It is invariant under gauge transformations such that [50 App. B]. In particular, this holds for each mode of our gauge vector field . In fact, using only the modal time and angular dependence, one can easily see that each mode contributes
which clearly vanishes as . However, this is not manifestly obvious when we determine through the formulas in App. § 2.0.1. But Figure 4 confirms this: over 70 modes the maximum helical value is , at roundoff, while ranges over some twelve decades in the same modes. Detuning the frequency by raises the maximum to , confirming that the null result stems from , and not an accidental cancellation or a quantity that was never computed.

Figure 4. Gauge vector contribution to in (1) for the Kerr () circular orbit (), over 70 modes with and . Left: the worldline contraction , which is generically nonzero. Center: the helical pullback evaluated on shell at . Right is the control quantity , with detuned. Color scales are and differ between panels.
The remaining non-gauge pieces alone in in our decomposition (1) therefore give the Detweiler redshift. The values we obtained for are: () and (), in agreement with the no-string GHZ field [29] and a traditional radiation-gauge (CCK) metric-reconstruction calculation [51]. In Figure 5 we show the approach of the partial -sums in the calculation of to the reference value: namely Eq. (1) times when decomposing the terms “” into -modes and summing first over , then over and then adding a numerically-fitted tail for .

Figure 5. Partial -sums in the calculation of for (left) and (right), : is Eq. (1) when decomposing the terms “” into -modes and summing first over , then over and then adding a numerically-fitted tail for . The horizontal dashed lines are the reference values () and (). The dashed orange curve in the left plot corresponds to the calculation using the CCK residual modes (up to it lies under the Bondi-like curve: absolute difference ).
Asymptotic quantities in Kerr. We have also validated the cancellation at non-decay orders , between the reconstructed piece and the gauge piece in the full metric perturbation (1) as in Kerr. This is shown in Figure 6.

Figure 6. The NP-components of the adjusted metric perturbation in the asymptotic region , constructed from the -mode scheme on the grid and evaluated at the interior node for a moderately spinning Kerr black hole (). The top row again shows the cancellation between the reconstructed piece and the gauge piece, leaving only the subleading decay. The second row shows the extrapolation of the -pieces, which can be seen as observables for gravitational radiation. Shown are absolute values.
For example, as shown in [12 App. D], the Bondi mass aspect is given up to normalization by in outgoing Eddington coordinates, as could be inferred for a particular angle and for the mode contribution in the first panel and the second row of Figure 6. Likewise, gives the “” entries of the Bondi-news tensor, whereas gives the “” entries for each particular mode with dependence , . These quantities hence directly follow from the third panel in the second row of Figure 6.
The Bondi angular momentum aspect can be retrieved from at the order by formulas given in [12, App. D]. Since the leading order is , as follows from the second panel in the second row of Figure 6, another step is necessary to dig out the this subleading piece. It is shown in Figure 7. That plot suggests also the absence of any terms of the form in the asymptotic expansion.18
Figure 7. Subleading piece of , as required for the Bondi angular momentum aspect.
Remarks on -mode scheme in Kerr
For moderately inclined and eccentric orbits an -mode treatment remains attractive, since every step of our scheme other than the Teukolsky solver, and the GHZ transport equations for the corrector [26] stays algebraic. Furthermore, the GHZ transport equations can be solved by -integration per -mode [28 App. H], which is no substantial difference to solving them per angle in the -mode scheme, and thus would be expected to be of a comparable computational cost. We summarize here what changes in the remaining steps of our scheme relative to the Schwarzschild case of Sec. IV.
A generic bound geodesic has three fundamental frequencies as we recall in App. § 2.0.3 § 3.1.
Spin-weighted quantities , such as the Weyl scalar , are correspondingly expanded in terms of their harmonics, , and in spin-weighted spheroidal eigenfunctions as
in place of (17), where , and . For , the mode amplitudes are in fact the quantities returned by most production Teukolsky frequency domain codes and the asymptotic amplitude is the leading piece of this quantity (which can be viewed as an asymptotic quantity at if we want).
The associated modal HMS is produced algebraically as in Schwarzschild by (19) in the spheroidal basis, where is now the Teukolsky-Starobinsky constant for Kerr, see e.g., [52 Lem. 2.7]:
where is the the angular eigenvalue of the spheroidal harmonics in the conventions of [52 Eqs. 2.3, 2.1]. On the other hand, Held’s angular derivatives, which are required in our scheme, act algebraically on spin-weighted spherical harmonics, see e.g., [28 App. C]. If we want to avoid giving up on the expansion in spheroidal harmonics and keep the Held-operators algebraic, then we have to use a well-known expansion
see, e.g., [53, 54]. The mixing coefficients are computed once per mode and cached, since every Held chain downstream reuses them. We denote the half-width at which the expansion is truncated as ,
In Kerr, the Held operators and (Table I) are not simply spin-raising/lowering operators but acquire off-diagonal pieces proportional to , see e.g., [28 App. C]. Furthermore, the multiplication operators , and —which vanish identically when —are themselves off-diagonal. However, all five, when considered as matrix operators, are tridiagonal in at fixed . Since these operators are linear and , they involve in fact only “nearest neighbor” couplings, such as
with
and similarly for . Hence corresponds to a banded tri-diagonal matrix and to a banded tri-diagonal matrix , respectively. Then , etc. The case is degenerate and must be considered separately.
The multiplication operator (5) needed for the completion pieces is likewise represented by the tri-diagonal matrix
The mode amplitudes of the completion piece of the Hertz potential (13) are, when considered as a vector in at fixed ,
Since each of these five angular operators is tridiagonal, a chain of of them is banded with half-bandwidth . The longest chain is the appearing in , and (15d) adds at most three further neighbours, so at output multipole requires the HMS out to , and the HMS in turn requires spheroidal amplitudes out to , where is the half-width at which (39a) is truncated. The cost of the completion is then per , which is still negligible when compared with the radial Teukolsky solve that produce its input.
The upshot is that nothing structural in the improved reconstruction scheme is lost with an scheme in going to Kerr and to generic orbits. The map from the Teukolsky amplitudes to the completed Hertz amplitudes remains purely algebraic, mode by mode, with no further radial or angular ODEs to solve beyond those already required for itself. What is diagonal in for now is merely banded via two independent mechanisms, the spheroidal-to-spherical transformation (39a), which couples a window of half-width , and the five Held operators which are tridiagonal in at fixed . Both truncations are numerically under control: the mixing coefficients decay rapidly in at fixed , so that a modest suffices and the fixed buffer inherited from the longest Held chain is exact rather than approximate. As an implementation, one simply solves the radial Teukolsky equation out to and monitors the convergence of under increasing .
Acknowledgements: We thank Patrick Bourg for discussions and explanations of the implementation of (17). We thank Adam Pound, Andrew Spiers, and Barry Wardell for discussions.
AI assistance declaration: During the preparation of this work, the authors used Claude’s Opus (Anthropic) to assist with numerical investigations of the Detweiler redshift quantity in Sec. V F. Specifically, Claude was prompted to apply existing code referred to in App. D to produce numerical results and to generate the corresponding Figure 5. In addition, we used Claude’s Opus for generating documentation, as well as plotting and unit testing scripts. All scientific interpretation of the results, as well as the writing of the manuscript, were carried out
solely by the authors. The authors take full responsibility for the content of this publication.
Lie derivative of along
Since is by construction compatible with the traceless IRG, we have and . The remaining, non-zero, NP components are [32]
and
The remaining components follow by conjugation, and , which in mode space are
The relative minus sign between the two is the arising when conjugating a spin-weighted harmonic.
Teukolsky operators in Held form
The Teukolsky wave operator for the 0-perturbed Weyl scalar, denoted in this paper, appears in (10), which
which we rewrite as , where is Teukolsky’s source (10). Especially when working with -modes, it is convenient to express these quantities in terms of (5) and Held’s coefficients and operators from Table I, because these are related to spin-raising and -lowering operators, see e.g., [28 App. C].
and the source is
Important structural relations in the Held formalism are
used e.g., to rewrite (B1) in equivalent forms.
Remarks on punctures for generic orbits
What is lost, and what is not
In this appendix, we outline how one could generalize the methods by [17] used for the punctures in this work, to more general orbits. First, for a generic orbit, the comoving chart (24) conditions are untouched.
Second, the singular field—and therefore neither the residual field—is no longer exactly stationary in the comoving coordinates; i.e.,
so that the NP tetrad components, labeled by hatted indices , retain a dependence on comoving time, where agrees with proper time on the worldline; .
Third, the map between the comoving coordinates and the background Boyer–Lindquist coordinates , in terms of which the modes of the reconstruction algorithm are defined, ceases to be linear with constant coefficients and becomes quasi-periodic.
Orbital phases for generic bound orbits
Generic geodesics satisfy the radial and polar equations,
coupled only through the common factor . Mino time decouples them into independent periodic sectors with torus frequencies . See e.g., [9], for a detailed review of these and other concepts of orbital motion in Kerr and for further references. The associated phases
advance uniformly in Mino time, not in proper time. The - and -geodesic equations have right-hand sides that are sums of a function of alone and a function of alone, and integrate to a secular piece plus an oscillatory piece,
where and , have zero average over the torus. The frequencies conjugate to Boyer–Lindquist time, which are the ones entering the mode decomposition, are the ratios of the secular rates,
where
with are the harmonics of the orbital frequencies appearing in the mode-decompositions.
The local expansion on the torus
The comoving coordinates (26) are built from , (the particle’s initial radius and latitude) now depend on the phases,
still with the particle at . Here , with the corotating phase now defined as
Unlike for circular orbits, is generally no longer linear in due to the oscillatory part of (C4) that this introduces.
The monomial structure (25) is unchanged. What changes is that the coefficients, together with (27), become functions on the orbital two-torus,
and etc. Each is then expanded as a Fourier series on the torus,
The master integral
For a circular equatorial orbit is an even function of , which reduces every integral to the single master function (29). The evenness is due to the discrete isometry , and composing it with a time translation gives an isometry that maps the circular orbit to itself.
For a generic orbit this reflection maps the orbit to its time reverse; the invariance is lost and an odd term appears. The most general dependence a single azimuthal angle can generate is then
Since with and , this is again a single cosine, merely displaced: the point of closest approach on the circle sits at rather than at . The substitution in the defining integral then costs only a phase,
so that (29) continues to do all the work.
From -modes to frequency modes
The frequency modes are assembled in two steps. First, at fixed orbital phase, the local azimuthal transform which is the only place (C12) and the dictionary (30) enter. Second, the transform over the torus. Using (C4) and (C5),
with , since . Averaging over and changing variables to the Mino phases gives
Each factor has a definite role. The phase is related to (C4), reflecting the fact that the azimuth against itself oscillates; corrects the mismatch between the orbital clock and Boyer–Lindquist time; performs the harmonic extraction; and is the Jacobian of the change from the -average that defines the Fourier coefficient to the uniform average over the Mino phases.
In the circular limit , , and is independent of ; the integral collapses to and the single helical frequency is recovered. The generic-orbit algorithm is therefore the circular one of Sec. V A, evaluated on a grid in and followed by (C14).
Reference map to numerical code
For the reader interested in using our open source code, we have written a comprehensive README file which indicates which header files, classes, and functions in the code are associated with generating the main quantities in each section, as well as a general description for its use.
References
References
- [1]S. E. Gralla and R. M. Wald, Class. Quant. Grav. 25, 205009 (2008), [Erratum: Class.Quant.Grav. 28, 159501 (2011)], arXiv:0806.3293 [gr-qc].arxiv.org/abs/0806.3293
- [2]P. Hintz, Gluing small black holes along timelike geodesics III: construction of true solutions and extreme mass ratio mergers (2024), arXiv:2408.06715 [gr-qc].arxiv.org/abs/2408.06715
- [3]P. Amaro-Seoane, S. Aoudia, S. Babak, P. Binetruy, E. Berti, et al., GW Notes 6, 4 (2013), arXiv:1201.3621 [astro-ph.CO].
- [4]S. Babak, J. Gair, A. Sesana, E. Barausse, C. F. Sopuerta, C. P. L. Berry, E. Berti, P. Amaro-Seoane, A. Petiteau, and A. Klein, Phys. Rev. D 95, 103012 (2017), arXiv:1703.09722 [gr-qc].
- [5]C. O. Lousto and Y. Zlochower, Phys. Rev. Lett. 106, 041101 (2011), arXiv:1009.0292 [gr-qc].
- [6]M. L. Katz, A. J. K. Chua, L. Speri, N. Warburton, and S. A. Hughes, Phys. Rev. D 104, 064047 (2021), arXiv:2104.04582 [gr-qc].
- [7]A. J. K. Chua, M. L. Katz, N. Warburton, and S. A. Hughes, Phys. Rev. Lett. 126, 051102 (2021), arXiv:2008.06071 [gr-qc].
- [8]L. M. Barack and A. Pound, Rept. Prog. Phys. 82, 016904 (2019), arXiv:1805.10385 [gr-qc].
- [9]A. Pound and B. Wardell, Black hole perturbation theory and gravitational self-force (2021), arXiv:2101.04592 [gr-qc].DOI
- [10]T. Hinderer and E. E. Flanagan, Phys. Rev. D 78, 064028 (2008), arXiv:0805.3337 [gr-qc].
- [11]B. Wardell, A. Pound, N. Warburton, J. Miller, L. Durkan, and A. Le Tiec, Phys. Rev. Lett. 130, 241402 (2023), arXiv:2112.12265 [gr-qc].
- [12]S. Hollands and A. Toomani, Class. Quant. Grav. 43, 055001 (2026), arXiv:2405.18604 [gr-qc].
- [13]A. Pound, C. Merlin, and L. Barack, Phys. Rev. D 89, 024009 (2014), arXiv:1310.1513 [gr-qc].
- [14]A. Heffernan, A. Ottewill, and B. Wardell, Phys. Rev. D 89, 024030 (2014), arXiv:1211.6446 [gr-qc].
- [15]A. Heffernan, A. Ottewill, and B. Wardell, Phys. Rev. D 86, 104023 (2012), arXiv:1004.0794 [gr-qc].
- [16]P. Bourg, A. Pound, S. D. Upton, and R. Panosso Macedo, Phys. Rev. D 110, 084007 (2024), arXiv:2404.10082 [gr-qc].
- [17]B. Leather et al., Kerr metric perturbations for second-order self-force theory (2026), arXiv:2611.03888 [gr-qc].arxiv.org/abs/2611.03888
- [18]J. M. Cohen and L. S. Kegeles, Phys. Rev. D 10, 1070 (1974).
- [19]P. L. Chrzanowski, Phys. Rev. D 11, 2042 (1975).
- [20]S. R. Dolan, L. Durkan, C. Kavanagh, and B. Wardell, Class. Quant. Grav. 41, 155011 (2024), arXiv:2306.16459 [gr-qc].
- [21]S. R. Dolan, N. Warburton, A. J. Harte, A. Le Tiec, B. Wardell, and L. Barack, Phys. Rev. D 89, 064011 (2014), arXiv:1312.0775 [gr-qc].
- [22]D. Li and N. Yunes, Metric Reconstruction for Generic Black-Hole Perturbations (2026), arXiv:2605.11080 [gr-qc].arxiv.org/abs/2605.11080
- [23]S. A. Teukolsky, Astrophys. J. 185, 635 (1973).
- [24]A. Ori, Phys. Rev. D 67, 124010 (2003), arXiv:gr-qc/0207045.
- [25]J. M. Cohen and R. M. Wald, J. Math. Phys. 13, 543 (1972).
- [26]S. R. Green, S. Hollands, and P. Zimmerman, Class. Quant. Grav. 37, 075001 (2020), arXiv:1908.09095 [gr-qc].DOI
- [27]C. Merlin, A. Ori, L. Barack, A. Pound, and M. van de Meent, Phys. Rev. D 94, 104066 (2016), arXiv:1609.01227 [gr-qc].DOI
- [28]V. Toomani, P. Zimmerman, A. Spiers, S. Hollands, A. Pound, and S. R. Green, Class. Quant. Grav. 39, 015019 (2022), arXiv:2108.04273 [gr-qc].DOI
- [29]P. Bourg, B. Leather, M. Casals, A. Pound, and B. Wardell, Phys. Rev. D 110, 044007 (2024), arXiv:2403.132634 [gr-qc].
- [30]B. Leather, A. Spiers, A. Pound, S. D. Upton, B. Wardell, L. Durkan, and N. Warburton, Second-order Teukolsky calculations for nons pinning, quasicircular binaries (2026), arXiv:2609.08995 [gr-qc].arxiv.org/abs/2609.08995
- [31]A. Spiers, A. Pound, and J. Moxon, The Bondi–Sachs gauge, BMS frames, and memory in black hole perturbation theory (2026), arXiv:2606.24816 [gr-qc].arxiv.org/abs/2606.24816
- [32]L. R. Price, K. Shankar, and B. F. Whiting, Class. Quant. Grav. 24, 2367 (2007), arXiv:gr-qc/0611070 [gr-qc].DOI
- [33]M. Casals, S. Hollands, A. Pound, and V. Toomani, Class. Quant. Grav. 41, 215010 (2024), arXiv:2402.15468 [gr-qc].DOI
- [34]B. Cownden, R. Panosso Macedo, C. Pantelidou, S. D. Upton, and B. Wardell, Multi-domain spectral method for Kerr spacetimes: a hyperboloidal Teukolsky m-mode solver for self-force computations (2026), arXiv:2611.03870 [gr-qc].arxiv.org/abs/2611.03870
- [35]S. Chandrasekhar, The Mathematical Theory of Black Holes (Clarendon Press, Oxford, 1983).
- [36]R. P. Geroch, A. Held, and R. Penrose, J. Math. Phys. 14, 874 (1973).
- [37]A. Held, Commun. Math. Phys. 37, 311 (1974).
- [38]A. Held, Commun. Math. Phys. 44, 211 (1975).
- [39]É. É. Flanagan and D. A. Nichols, Phys. Rev. D 95, 044002 (2017), [Erratum: Phys.Rev.D 108, 069902 (2023)], arXiv:1510.03386 [hep-th].DOI
- [40]R. H. Price, Phys. Rev. D 5, 2439 (1972).
- [41]S. A. Teukolsky and W. H. Press, Astrophys. J. 193, 443 (1974).
- [42]A. A. Starobinsky and S. M. Churilov, Zh. Eksp. Teor. Fiz. 65, 3 (1973), english translation: Sov. Phys. JETP 38, 1–5 (1973).
- [43]P. Zimmerman, unpublished notes.
- [44]Wolfram Research, Inc., Mathematica, version 14.2 (2025), champaign, IL.
- [45]D. Yllanes and J. M. Martín-García, xCoba: General component tensor computer algebra, https://www.xact.es/xCoba/, part of the xAct suite.
- [46]R. Panosso Macedo, P. Bourg, A. Pound, and S. D. Upton, Phys. Rev. D 110, 084008 (2024), arXiv:2404.10083 [gr-qc].DOI
- [47]B. Wardell, N. Warburton, L. Durkan, K. Cunningham, B. Leather, Z. Nasipak, T. Torres, et al., Teukolsky: A Mathematica package for computing solutions to the Teukolsky equation (2025), part of the Black Hole Perturbation Toolkit. This work makes use of the Black Hole Perturbation Toolkit.
- [48]R. K. L. Lo, Phys. Rev. D 110, 124070 (2024), arXiv:2306.16469 [gr-qc].DOI
- [49]Y. Yin, R. K. L. Lo, and X. Chen, Phys. Rev. D 113, 124007 (2026), arXiv:2511.08673 [gr-qc].DOI
- [50]S. L. Detweiler, Phys. Rev. D 77, 124026 (2008), arXiv:0804.3529 [gr-qc].DOI
- [51]M. van de Meent and A. G. Shah, Phys. Rev. D 92, 064025 (2015), arXiv:1506.04755 [gr-qc].DOI
- [52]M. Casals and R. Teixeira da Costa, Classical and Quantum Gravity 38, 165016 (2021), arXiv:2102.06734 [gr-qc].DOI
- [53]E. Berti, V. Cardoso, and M. Casals, Phys. Rev. D 73, 024013 (2006), erratum: Phys. Rev. D 73, 109902 (2006), arXiv:gr-qc/0511111 [gr-qc].arxiv.org/abs/gr-qc/0511111
- [54]C. Kavanagh, A. C. Ottewill, and B. Wardell, Physical Review D 93, 124038 (2016).

