WorksPhysics and Astronomy
Maximum entropy reconstruction of the black hole greybody factors from its Korteweg-de Vries integrals
Abstract
The dynamics of perturbed non-rotating black holes (BHs) can be described by master wave equations with a potential barrier. Recent work has shown that these equations possess an infinite number of symmetries generated by the Korteweg-de Vries (KdV) hierarchy, leading to conserved quantities (KdV integrals) that fully determine the BH greybody factors via a moment problem. In this paper, we introduce and validate a new method to solve this moment problem using the maximum entropy principle. The method reconstructs, at each order of truncation, the least-biased transmission probability compatible with a finite set of KdV integrals, guaranteeing positivity by construction. We first validate the technique against the exactly solvable Pöschl-Teller potential. We then apply it to the Regge-Wheeler potential governing axial perturbations of a Schwarzschild BH, where no exact solution is available; the resulting reconstruction agrees closely with the fourth-order Wentzel-Kramers-Brillouin (WKB) approximation across several multipoles. This is the first application of the maximum entropy method to the reconstruction of BH greybody factors, demonstrating that a modest number of KdV integrals is sufficient to recover them accurately.
Introduction
Black holes (BHs) [1] are one of the most radical predictions of General Relativity [2, 3, 4, 5]. Decades of theoretical and observational work have established BHs as central objects in astrophysics, cosmology, and fundamental physics. In general relativity (GR), two conjectures hold up the standard picture of how BHs form and what they look like. Gravitational collapse generically produces singularities [6] and the weak cosmic censorship conjecture states that, for generic initial data, such singularities are hidden behind event horizons [7, 8]. On the other hand, the no-hair conjecture [9, 10] states that the resulting stationary BH is characterized by just three parameters: mass, intrinsic angular momentum (spin), and electric charge. That is, it belongs to the Kerr-Newman family [11, 12]. The latter is supported by the uniqueness theorems for stationary, asymptotically flat electrovacuum BHs [10, 13, 14, 15, 16, 17, 18], by the decay of perturbations in collapse [19, 20], and by the nonlinear stability of slowly rotating Kerr BHs [21, 22], but it does not extend to arbitrary matter or radiation [10, 23]. The charge is generally regarded as negligible for astrophysical BHs: any significant net charge is expected to be rapidly neutralized by selective accretion from the surrounding plasma [24, 25], and a rotating BH immersed in a magnetic field only acquires the tiny electric charge [8]. Observations of BH candidates are accumulating from various sources, including X-rays [26] and gravitational waves [27, 28, 29], with future detectors promising many more candidates with improved sensitivities [30, 31, 32]. There are physical phenomena in the cosmos involving BHs, like scattering processes of different matter fields and quasinormal mode (QNM) oscillations, which are of particular relevance. Scattering theory for BHs [33, 34] provides a powerful tool to investigate the nature of the scatterer by collecting asymptotic data. When the scattered particles or waves only induce small deviations from the BH background geometry, BH perturbation theory (BHPT) is the appropriate framework to study them. As found in pioneering works [35, 36, 37, 38, 39, 40, 41], and generalized to higher dimensions [42, 43], it turns out that the perturbations can be described by a set of master wave-like equations with a potential barrier. Actually, all the physical observables can be obtained from the solution, the master wave functions. On the other hand, in the frequency domain, the master equations become time-independent Schrödinger equations, where the energy would correspond to minus square of the frequency of the corresponding mode.
In this context, the greybody factors are key physical quantities as they represent the transmission (and reflection) probabilities associated with the BH potential barrier (the S-matrix for BH scattering), encoding deviations from the black-body spectrum caused by the interaction with the background potential [44]. The poles of the analytic extension of the transmission coefficient correspond to the BH QNMs [45, 46, 47, 48], which are crucial in gravitational wave astronomy as they may contain the fingerprints of new physics [49, 50, 51, 52, 53]. There is an extensive literature on calculations of greybody factors, including low-frequency [54, 55, 56, 57] and high-frequency [58, 59, 60] limits, different Wentzel–Kramers–Brillouin (WKB) approximations [61, 62, 63, 64, 65, 66], bounding of the Bogoliubov coefficients [67, 68], and monodromy methods [69, 70, 71].
Recent work [72, 73, 74, 75] has revealed a rich structure in the space of master equations and functions describing Schwarzschild BH perturbations. All possible master equations are connected by Darboux transformations (DTs) (see also [76, 77]), establishing their physical equivalence. This leads, in particular, to the isospectrality of the QNM spectrum, in the sense that even- and odd-parity modes share the same spectrum. Additionally, these equations admit an infinite number of symmetries generated by the Korteweg-de Vries (KdV) hierarchy, connecting in this way with the world of integrable systems [78], which lead to an infinite number of conserved quantities (KdV integrals) that are invariant under DTs [73]. Remarkably, these KdV integrals fully determine the BH greybody factors via a moment problem [74, 79]. The KdV integrals turn out to be, up to a rescaling, the moments of a distribution function built out of the logarithm of the transmission probability. In [79], a method to invert the BH moment problem to recover the transmission coefficient in terms of the KdV integrals was introduced. The method uses Padé approximants and the Laplace transform. Another application of the KdV integrals is the study of stability properties of the greybody factors and QNM spectrum [80].
Executive Summary. In this paper, we develop and validate a new method to solve the BH moment problem using the maximum entropy principle [81, 82, 83, 84]. The method reconstructs, at each order of truncation, the least-biased probability distribution compatible with a given set of moments, guaranteeing positivity at every order. We first validate the technique against the exactly solvable Pöschl-Teller potential [85], where we can compare with the exact greybody factor. We then apply it to the Regge-Wheeler potential [35] governing axial perturbations of a Schwarzschild BH, comparing our results with the WKB approximation. We use two different approaches, one based on integration on the original infinite frequency domain and, another one based on a compactification method that transform the integrals on an infinite domain to a finite one, improving in this way the performance of the method. This is the first application of the maximum entropy method to the reconstruction of BH greybody factors, demonstrating that a modest number of KdV integrals is sufficient to recover them accurately.
Structure of the Paper. In Section II we review BHPT for the Schwarzschild case, introducing the master equations and the role of DTs. In Section III we review the KdV symmetries and the moment problem formulation. In Section IV we summarize the mathematical aspects of the BH moment problem, recall the Padé-based inversion of Ref. [79], and introduce the dimensionless variables used in the rest of the paper. In Section V we introduce the maximum entropy method and its numerical implementation. In Section VI we validate the method against the Pöschl-Teller potential and identify the limitations of the implementation. In Section VII we address these limitations by reformulating the problem on a compact interval, as a Hausdorff moment problem, and compare both formulations using the Pöschl-Teller potential. In Section VIII we apply the method to the Regge-Wheeler potential, choosing the formulation and the number of moments according to the multipole . We conclude in Section IX with a discussion of our results and future directions.
Throughout this paper, unless otherwise stated, we use geometric units with .
BH Perturbation Theory and Master Equations
The study of perturbations of non-rotating BHs started with the seminal work of Regge and Wheeler [35]. The background spacetime is described by the Schwarzschild metric [86]:
where we use a hat to denote quantities defined in the Schwarzschild background spacetime. The metric is completely determined by the function
with being the Schwarzschild radius, and denotes the line element of the 2-sphere (). In standard spherical coordinates its line element is: .
We can introduce metric perturbations of the Schwarzschild spacetime, , satisfying , via
Taking into account the spherical symmetry of the background, we can expand the perturbations in terms of scalar, vector and tensor spherical harmonics, separating in this way the dependence on from the angular dependence. Moreover, it is well-known that the perturbative Einstein equations decouple for each harmonic and parity4. Then, for each mode we can decouple the Einstein perturbative equations in terms of master functions (combinations of the metric perturbations and their first-order derivatives), which satisfy master wave equations of the form ²:
where is the tortoise coordinate, defined by , is the BH potential barrier, and are the even/odd master functions, which are gauge-invariant (do not depend on the choice of perturbative gauge). For odd-parity (axial) perturbations, the standard potential considered is the Regge-Wheeler potential [35] (see also [72, 75]):
while for even-parity (polar) perturbations, it is the Zerilli potential [39]:
with .
Recent work [72] has shown that the space of possible master functions and equations is much larger than previously known. Indeed, by assuming that the master functions are linear combinations of the metric perturbations and its first-order derivatives, in [72] was found that the space of master functions/equations [of pairs ] contains two branches: the standard branch, containing the Regge-Wheeler and Zerilli potentials, and the Darboux branch, containing an infinite family of new potentials. All these master equations are connected by DTs [73], which are isospectral transformations preserving the reflection and transmission coefficients. It is also possible to reconstruct the metric perturbations from the master functions in all cases [75].
The DT connects elements of the space of master functions and equations (the master equations are determined by the shape of the potential). Then, we say that two pairs and are connected by a DT, when they are related by the following expressions:
where the DT generating function, , has to satisfy a consistency in the form of a Riccati equation
with an arbitrary constant. The DT preserves the frequency spectrum, as it can be seen by working in the frequency domain by introducing the separation , being the frequency, which leads to a time-independent Schrödinger equation
The physical equivalence of all these possible descriptions of the metric perturbations means that the scattering data, in our case the greybody factors as there are no bound states, and the QNM spectrum are the same regardless of which potential is used. This redundancy, what we call Darboux covariance [73], is at the heart of the developments that follow.
KdV Symmetries and the BH Moment Problem
The master equations describing BH perturbations possess additional symmetries beyond DTs. These symmetries are generated by the flow of the KdV equation [87]
which is one of the best well-known examples of a completely integrable equation (see [88, 89, 90, 91, 92] for diverse literature on integrable systems), a non-linear partial differential equation in this case. Associated to the KdV equation, there is an associated infinite hierarchy of KdV equations that, considered as deformations of our time-independent master equation [Eq. (9)], turn out to be isospectral deformations: they preserve the transmission coefficient as well as the QNM frequencies [73]. The parameter represents an auxiliary KdV time parameterizing a flow through the space of potentials, distinct from the physical time of the gravitational perturbations.
Let us now consider a scattering process (as in [74]; see [88, 93, 94, 95, 96] for a detailed treatments of scattering processes), described by our master equations, in which an incident wave comes from the BH horizon, interacts with the BH potential barrier, and part is transmitted and part is reflected. The solution of the time-independent Schrödinger equation (9) for this physical situation can be written in the following form
where we have introduced the dependence on the KdV time to account for the dependence of the different objects on the KdV deformation of the master equation. The complex coefficients and are usually known as the Bogoliubov coefficients and they determine completely the transmission/reflection coefficients (and hence the scattering matrix) as follows
The transmission and reflection probabilities, i.e. the greybody factors, are simply given by the square of the modulus of the corresponding coefficients
which, for real , satisfy
One can show that under the deformations induced by the KdV flow, the Bogoliubov coefficients evolve as [97]
which means that we have a conservation law, namely
The modulus of is also preserved by the KdV flow. Similar evolution laws, can be obtained for the reflection and transmission coefficients.
The integrability of the KdV hierarchy is related to the presence of an infinite sequence of conserved quantities, the KdV integrals [98, 99, 100], which in our case can be derived from the conservation law of Eq. (16). They characterize the large-frequency asymptotics of the scattering wave function (see [88, 93, 94, 95, 96] for detailed developments). Writing the solution of the Schrödinger equation (as including the KdV flow time ):
so that in the half-plane, , we have
and in consequence,
The differential phase function satisfies the following complex Riccati equation
Expanding in inverse powers of (see [99]),
and introducing it in Eq. (20) leads to a recurrence for the coefficients :
The KdV integrals are conserved with respect to the KdV flow by virtue of Eq. (16), that is, they are independent of the KdV time . They are given by:
All even-indexed integrals vanish (), while the odd ones are the true conserved quantities.
The KdV integrals are invariant under DTs [73], meaning they take the same value for every potential in the Schwarzschild perturbation theory equivalence class. Or, in other words, for any element of the space of master equations and functions determined in [72]:
Using the analytic properties of in the upper half -plane (), one can obtain the trace identities [from the diagonal of the resolvent of the master equation (9)] [74, 99]:
It is straightforward to see that this infinite set of equations define what in mathematics is known as a classical moment problem [101, 102, 103, 104, 105, 106, 107] (see [108, 109, 110] for modern accounts). Introducing the positive definite distribution function
and the moments
the trace identities take a more compact form which clearly corresponds to a moment problem:
In this form, this corresponds to a (symmetric ) Hamburger moment problem . The goal is to provide the sequence of even moments and, from Eq. (29), to recover the distribution , from which we obtain the greybody factor . We have numerically checked [79] that the BH moment problem [as defined by Eq. (29) with the definitions of Eqs. (27) and (28)] is determinate, meaning a unique solution exists [74, 110] (see also [111]).
A final remark worth mentioning here concerns the complete integrability of the KdV equation. It was shown in [99] that there is an infinite-dimensional Hamiltonian system associated (see, e.g. [112]) for which a set of action-angle variables, say , exist. In our case, the action variable is associated to the distribution of the BH moment problem [Eq. (27)] while the angle variable is given by the argument of the Bogoliubov coefficient , that is: , and .
Solving the BH moment problem
In [74] there is a broad discussion of basic mathematical aspects of the BH moment problem. Here, we summarize the main points. First of all, there is a one-to-one relation between the moment problem and the problem of finding the analytic function in defined as the Stieltjes transform of the distribution :
Then, has the following asymptotic expansion in terms of the moments
for any range of angles with , can be seen as a generating function for the moments. This shows that the distribution is indeed uniquely determined by , by using the Stieltjes-Perron inversion [113] (see also [108, 110, 114]), which relates the probability distribution with the branch cut discontinuity of the Stieltjes transform across the real axis:
The functions defined by the Stieltjes transform [Eq. (30)] have the Herglotz property [115], . In our case, one can show that [74]
which defines a Stieltjes transform and the Stieltjes-Perron inversion becomes
The practical implementation of this formula is not straightforward as it has been discussed in [74]. Then, we need to resort to approximate computational methods, which in practice means that we are going to use a finite number of moments in order to compute the transmission coefficient.
finite number of moments in order to compute the transmission coefficient. Solving the moment problem from a finite set of moments is an ill-posed inverse problem: Infinitely many distributions are compatible with a given finite moment sequence. Also, it is not easy to find robust computational methods to carry out the inversion. In Ref. [79], we exploited the connection between the moments and the moment generating function. For the Stieltjes formulation of the moment problem, the moment generating function is defined as the Laplace transform of the distribution ,
where we have introduced the following definitions:
where is a dimensionless variable and for the Pöschl-Teller potential and for the Regge-Wheeler one. The asymptotic expansion of the moment generating function for large is directly given by the moments:
Considering a finite set of moments, Padé approximants for the truncated asymptotic series of were constructed in Ref. [79]. For Stieltjes series, diagonal and subdiagonal Padé approximants are known to converge to the Stieltjes function [Eq. (33)], which is only valid in the upper half -plane. Padé approximants of the Stieltjes series of Eq. (31) reproduce the real line branch-cut evident from Eq. (33). Given the close connection between both the Stieltjes and the moment generating functions with the distribution , we would also expect that Padé approximants to the second converge, as it happens in the approximations shown in Ref. [79] and in others quoted there. The next important step in Ref. [79] is to decompose the Padé approximant into partial fractions, for which the inverse Laplace transform can be computed. The relevant expressions for the approximate computation of can be seen in Ref. [79].
The Padé-based approximation has the advantage of being computationally very efficient and often yields accurate results with only a few moments. However, it may produce poles with positive real parts that must be discarded, and it does not automatically guarantee positivity of the reconstructed distribution at every truncation order (the reconstructed can locally exceed unity if the exponential argument becomes positive). In practice, these issues are mild for the BH potentials as it was shown in [79], and the method compares well with WKB results.
Before moving to the description of a new method to solve the truncated moment problem, it is important to discuss several general questions that are going to be useful for what comes in the next section. First of all, the physical dimensions (units) of the different quantities involved, taking into account that we are using natural geometric units and hence, all the quantities can be expressed as a power of the length unit, say . Let us start from the BH potential, which is the object that defines the physical framework. It is clear from the definitions see Eqs. (5) and (6) that it has inverse square length dimensions, i.e.
Then, we can see that the dimensions of the KdV densities described in Eq. (23) are:
which means that the dimensions of the KdV integrals in Eq. (24) are:
where we have use the definition of the moments in Eq. (28). Finally, the frequency, , and the density, , have obvious dimensions:
We can introduce dimensionless quantities by using , which was introduced in Eq. (36). In this way, we can introduce dimensionless moments and frequency as follows:
A New Method: Maximum Entropy Reconstruction
A plausible alternative to the Padé-based method described above is the maximum entropy (MaxEnt) method [81, 83]. Mathematical aspects of the method, like existence, convergence and numerical implementations have been discussed, for instance in [116, 117, 118, 119]. The MaxEnt method has been used previously for the moment problem in other physical contexts, among others: Electronic densities of states [120, 121, 122, 123]; reconstruction of parton and baryon-number distributions from moments and cumulants [124]; polymer distribution functions [125]; invariant densities of chaotic maps [126, 127]; entropy-based moment closures in kinetic theory [128, 129]; etc.
The MaxEnt method [81] selects the least-biased distribution consistent with the known moments by maximizing the Shannon entropy
Consider the classical moment problem for a positive density on ,
The MaxEnt method maximizes the functional
where the are Lagrange multipliers. Since only a finite number of moments is known, we can only reconstruct an approximation to . The functional variation with respect to yields
To write in terms of the remaining multipliers, we have to normalize all the moments by , so that the zeroth moment equals one.
Let us now turn to our moment problem. Since the distribution in Eq. (29) can be taken to be even, we fold the problem onto the half-line by introducing
and hence, we have the following Stieltjes moment problem:
Normalizing by , the moment problem becomes
where and , so that . These equations are the constraints for the maximization of the entropy, introduced through Lagrange multipliers , one for each of the retained moments. The functional to be extremized is
and its stationary point is
Since , must be normalized to one, which fixes in terms of the other multipliers:
The multipliers () are found by minimizing the dual potential
is convex in the multipliers, and its gradient is
where the average is taken with , so its stationary points coincide with the solutions of the moment constraints. We minimize using the L-BFGS-B algorithm [130, 131], a limited-memory quasi-Newton method with constraints.
The highest-order multiplier is constrained to be strictly positive, which guarantees the convergence of the integral defining . Since this integral extends to infinity, in practice we truncate it at a finite value of , chosen large enough for the integrand to be negligible beyond it. In addition, we clip the exponent inside the integrand to avoid numerical overflow.8
Before normalizing and minimizing, we rescale the moments, since they differ from each other by orders of magnitude. We introduce the scale
that is, it has the units of , so that
is a dimensionless variable. The rescaled moments and multipliers are
By construction, , and by the log-convexity of the moments all the others lie in . The normalized constraints are: , which enter in the expression of . Since , rescaling and normalizing commute. The minimization of is carried out entirely with respect to the variable .
Once the rescaled multipliers are found, follows from the normalization condition:
so that the rescaled density, normalized to one in the variable , is
The rescaling is undone only when reconstructing the greybody factor. Since , the density in the original variable is obtained by evaluating at and dividing by (the Jacobian of the change of variable):
Finally, we have that the greybody factor from Eq. (27) is
Equivalently, one can obtain , Eq. (51), in terms of the original multipliers, using Eq. (57).
The MaxEnt method has several advantages to deal with the moment problem: (i) it guarantees positivity of the reconstructed distribution at every order of truncation; (ii) the reconstruction is an entire function of , with no spurious poles; (iii) the method provides a systematic way to incorporate additional moments as they become available.
Validation of the MaxEnt Method: The Pöschl-Teller Potential
Before applying the MaxEnt method to the BH potential barriers, we are going to validate it against the Pöschl-Teller potential [85], one of the few exactly solvable potentials on the real line:
where . It is important to mention that the Pöschl-Teller potential has been frequently used as an approximation to the actual BH potential. The main reason is that the Pöschl-Teller potential, as it has been written in Eq. (62) with , is also a positive potential barrier. While the BH and Pöschl-Teller potentials decay exponentially as we approach the horizon (), the main difference between them lies in the behaviour at spatial infinity (), where the Pöschl-Teller potential also decays exponentially while the BH potentials decay as a power law, specifically: . That is, the behavior at spatial infinity is dominated by the centrifugal barrier.
Taking into account the similar qualitative behaviour between the Pöschl-Teller potential and the BH potential, the first one has been used in the literature [132, 133] as an approximation to the BH potential barrier. To that end, the first step is to identify the coordinate in Eq. (62) with the tortoise coordinate of the Schwarzschild BH. Then, the value of the constant is chosen to be the location of the maximum of the BH potential and hence, is identified with the maximum value of the BH potential, that is, and . Finally, the constant in Eq. (62) is related to the BH potential curvature via:
The exact transmission coefficient for the Pöschl-Teller potential can be found analytically. The expression is:
Since , the associated distribution is also symmetric, , and we can restrict ourselves to positive frequencies, as in Eq. (48). For the Pöschl-Teller potential the natural length scale is , so the dimensionless variables of Eq. (42) are
which satisfy . For convenience, we have also introduced the following definitions:
Then, the Schrödinger equation, Eq. (9), becomes , and the potential and the greybody factor take the form
where we have used in Eq. (63). The shift does not affect the scattering data, since the potential only depends on .
Finally, the dimensionless moments are the ones from Eq. (42). Multiplying Eq. (48) by and using that the density is dimensionless, the moment problem becomes
The first KdV integrals for the Pöschl-Teller potential are given in Appendix A. Using these exact values for the moments we have carried out the inversion of the moment problem as described in § 1.4.
Figure 1 shows the MaxEnt reconstruction, , for different values of the Pöschl-Teller constant and the number of moments considered, . The reconstruction captures the shape of even for small values of . However, the number of moments required to obtain a faithful reconstruction depends strongly on . For , the curves with are already very close to the exact result (the curves for and are indistinguishable). For and , by contrast, the low-order reconstructions () exhibit an overly gradual transition, requiring at least moments (even moments for ) to accurately capture the correct steepness, position, and width of the step.

Figure 1. Greybody factors for the Pöschl-Teller potential as functions of the dimensionless frequency , obtained from the MaxEnt reconstruction described in § 1.4, and compared with the exact solution [Eq. (63)], for (top), (middle), and (bottom). The subscript indicates the number of moments used in the reconstruction.
A surprising feature of Figure 1 is the fact that increasing the number of moments does not seem to improve the reconstruction. For , the curves with are clearly worse: they undershoot the exact result around the step and overshoot it beyond. For and the best agreement is reached at intermediate (– and –, respectively). Beyond that, the reconstruction becomes steeper than the exact transmission coefficient and it is displaced with respect to it, towards smaller for and , with a displacement that grows with (for the displacement of the high- curves is in the opposite direction, towards larger ). Hence, the best reconstruction is obtained at an intermediate number of moments, and adding further moments degrades rather than refines the result.
The results above rely on several truncations that have no counterpart in the exact problem: the upper limit in the integral defining , the clipping of the exponent, and the bounds imposed on the multipliers during the minimization. It is therefore difficult to separate the limitations of the MaxEnt method itself from those introduced by our numerical implementation. In the next section, we reformulate the problem on a compact interval, which removes the need for the truncations of integrals and allows us to approach the exact result as closely as the method permits.
Improving the BH MaxEnt Method: Compactification
One of the main difficulties involved in inverting the BH moment problem is that the integrals defining the moments have an infinite range of integration, both in the physical and in the frequency domains. In the case of the physical domain, it is not a problem as we have already shown that the KdV integrals for the potentials under consideration can be obtained analytically. In the case of the frequency domain, we have been dealing either with a Hamburger or with a Stieltjes moment problem. In both cases we have unbounded integration domains. Given that we have to truncate the integration domain in the numerical implementations of our algorithm, this can induce errors that will depend on the way we truncate. The best we can expect is that the contributions from the part of the domain that has been stripped away do not contribute much to the momenta. However, it is also possible that this truncation may affect the convergence of the maximization process of the MaxEnt method, in the sense that the algorithm can at some point be driven to favor terms whose contribution is canceled by the truncation.
cation.
In the case of Pöschl-Teller potential, we can deal better with this problem as it decays exponentially towards the two ends of the domain. As we have already mentioned, the situation is different for a BH potential, for instance for the Regge-Wheeler potential, as it only decays exponentially towards the horizon, but it decays as a negative power of (or ) towards spatial infinity, where the centrifugal barrier dominates, i.e. . In this sense, it would be very convenient to transform the integration domain to a finite interval, that is, to convert our Hamburger/Stieltjes moment problem into a Hausdorff moment problem, or to a similar formulation. And this is what we explore in this section.
Our starting point is the Hamburger moment problem:
where , the even moments are connected with the BH KdV integrals by Eq. (28), and the (probability) density function by Eq. (27). Recall that the KdV integrals, , are conserved quantities of the KdV flow which can be computed from the KdV densities, , through Eq. (24). The KdV densities are obtained by means of the recurrence given in Eq. (23). Their analytic expressions for the Pöschl–Teller and Regge–Wheeler potentials are given in Appendices A and B, respectively.
We have seen before how to introduce dimensionless variables and parameters [see Eq. (42)] and how to restrict the integration to positive frequencies [see Eq. (48)]. This leads to the formulation of the moment problem in Eq. (68), which is very similar to a Stieltjes moment problem. To bring it to that form, we change variable in the dimensionless frequency,
which gives the Stieltjes moment problem
where [see Eq. (47)]
that is, we have an Stieltjes moment problem with dimensionless moments [see Eq. (42)].
The key point to transition towards a compactification that produces integrals on a compact interval is a change of integration variable. We use a relatively simple one,
which maps into . The moments are then given by
In practice, to solve numerically the moment problem we can only use a finite number of moments, parametrized by , i.e. . Then Eq. (74) can be rewritten as
where
The factor in front of is related to the appearance of the Bernstein polynomials in the integrals of our moment problem [Eq. (75)]. The Bernstein polynomials are typically defined as
Given the connection of the Bernstein polynomials with the binomial formula
and since in our case and , so that , we have
Multiplying Eq. (75) by and summing over gives the identity
We normalize the moments and the distribution by
which satisfy
Since the new moments, , are positive we must have
The resulting generalized moment problem is
The entropy functional from the MaxEnt method that needs to be applied to this form of the moment problem is
and the solution is
Notice that now cannot be expressed in terms of the remaining Lagrange multipliers, so we have to treat all of them at once. Instead of , we minimize the dual potential
The moment problem defined by Eq. (84), with moments that have the property in Eq. (83), looks very appealing to be solved by the MaxEnt method [see Eqs. (85) and (86)]. However, the fact that the normalized moments are bound to live in the interval does not guarantee that the numerical algorithms to obtain the MaxEnt solution are going to converge. Indeed, we have observed in our numerical experiments, that despite the moments belong to the interval they can still differ by orders of magnitude. Therefore, as it is, this generalized formulation of the moment problem is not amenable for numerical computations.
To revert the situation, following what we did in previous sections, we introduce a rescaling of the moments. The rescaling factor is defined as follows
where we have used Eq. (81). This factor is dimensionless as the moments we are dealing with are also dimensionless. In this sense, the factor we have just introduced is going to play the same roles as the scale in § 1.4, with the difference that was not dimensionless but has the units of frequency. In practice, we implement now the rescaling as a simplified Möbius transformation of the variable that we used for the compactification of the frequency domain. That is, we introduce
and the inverse transformation is, as expected, another Möbius transformation:
where is the dimensionless frequency introduced in Eq. (42). Given that , the transformation of Eq. (89), and its inverse of Eq. (90), are mappings of this interval onto itself. Other relevant properties of the Möbius map are:
where
Now, we can introduce rescaled moments as
However, the Möbius map does not preserve the Bernstein basis. From their definition in Eq. (77), we find that
from where, using the property of Eq. (79), we have
This is the key identity to rewrite the momentum constraints of Eq. (84). Indeed, using Eqs. (91) and (94) we can write
Now, we can introduce the rescaled distribution as follows:
with
so that the new rescaled distribution is normalized:
We can introduce another set of moments that add to unity. Indeed,
Then, after the rescaling with the factor we have arrived at the same type of generalized moment problem as in Eq. (84):
In addition, by construction, we have
so that the new moments are no going to be separated by orders of magnitude. The numerical procedure to invert this generalized moment problem starts with the computation of the dimensionless raw moments . Then, we normalize them by and we have the moments [Eq. (81)]. Before minimizing with the L-BFGS-B method to obtain the Lagrange multipliers, we must rescale the problem. The minimization of is carried out entirely with respect to the variable . Once we get the rescaled multipliers , we have that the rescaled density is:
The rescaling is undone when reconstructing the transmission coefficient, as it needs to be expressed in the original variable :
with
where
Figure 2 shows the reconstruction of for the Pöschl–Teller potential with the compactified frequency domain, for the same values of as in Figure 1. For every and the reconstruction breaks down at , where jumps to 1 instead of vanishing as the exact does. This is a direct consequence of the change of variables: after the compactification, the exponent in Eq. (104) carries an overall factor of , so that for any value of the multipliers. We call breakdown region the narrow neighborhood of the origin in which $T_N departs from the (exponentially small) exact value, dropping from over a finite interval. Defining its edge as the point at which falls below , we find that the region extends up to – for all except , with only a weak dependence on and . For it is wider, reaching –.

Figure 2. Greybody factors for the Pöschl-Teller potential as functions of the dimensionless frequency , obtained from the MaxEnt reconstruction presented in § 1.6 and compared with the exact solution for the following values of : (top), (middle), and (bottom). The subscript indicates the number of moments used in the reconstruction.
Away from the origin, the quality of the reconstruction depends strongly on , which sets the position of the step, . For the reconstruction with compactification is not the best option: the step is not well separated from the origin, and a small residual deviation remains at for all . The reconstruction without compactification (Figure 1) reaches a better accuracy already with –. For larger values of ( and ) the step lies far from the origin and the reconstruction improves systematically as increases.
The compactified formulation also removes several numerical ingredients required in § 1.4: the integrals run over the finite interval , and neither a clipping of the exponent nor bounds on the multipliers are needed. In the formulation without compactification, beyond a certain number of moments ( for , for and for ) the highest-order multiplier, which must be positive to guarantee the convergence of , ends up at its lower bound. The minimization then stops on that bound rather than at a stationary point, so the highest moments are not actually matched. This suggests that the non-monotonic behavior found there is mainly related to this constraint and to the numerical solution of the dual problem, and not to the information contained in the moments themselves. The truncation of the frequency domain and the clipping of the exponent may also contribute.
To quantify the accuracy of the reconstruction, we compare with the exact Pöschl-Teller transmission probability, Eq. (67). The point-wise error,
measures the local deviation of the reconstruction with moments from the exact result at each frequency. The integrated error,
is the area under and summarises the accuracy over the whole frequency range in a single number for each . We take , so that the integration window contains the step and the beginning of the plateau. For the reconstruction without compactification , while for the one with compactification , which excludes the breakdown region at the origin.
Figure 3 shows the point-wise error (first two rows) and the integrated error (last two rows) with respect to the exact . Every column corresponds to a value of (the ones we have already presented).

Figure 3. Errors of the MaxEnt reconstruction of the Pöschl-Teller greybody factor. Each column correspond to a different value of : (left), (middle) and (right). From top to bottom, the rows show: the point-wise error on a logarithmic scale of base 10 for the standard reconstruction; and the same quantity for the reconstruction using the compactification. The integrated error as a function of the number of moments for the standard reconstruction; and the same quantity for the reconstruction with compactification.
In the first row, we see the point-wise error for the standard reconstruction (without the compactification). The error is smallest for low and grows beyond it. For , the best curve is ; for the error grows with , both near the step and at small . The same happens for , where is best, and for with . The integrated error in the third row confirms this: has a minimum at (), (, although are indistinguishable at the scale of the figure) and (), and then grows, although not monotonically (for the value at is slightly below the one at , and for the error at is comparable to the one at ). In the second row, we show the point-wise error for the reconstruction using the compactification. Here, for all and , but the error then drops by many orders of magnitude, since both and are exponentially small there. Away from the origin the trend is the opposite to the case without compactification: the error decreases with . We see in the fourth row that decreases, up to small non-monotonic bumps. Since the error decreases essentially monotonically, can be chosen just by checking the convergence of the reconstruction, without knowing the exact result.
A complementary diagnostic, which is useful when no exact solution is available, as in the case of the BH potentials, is the successive difference between reconstructions,
together with its integrated version,
with the same and as in Eq. (108).
Figure 4 shows both quantities for in the same way as in Figure 3. The successive differences from Eq. (109), in the first row, do not decay with , so the sequence does not converge. The integrated differences from Eq. (110), in the third row, give the same picture: for they grow and saturate at , while for and they decrease up to and then fluctuate without any systematic decay.

Figure 4. Errors between consecutive MaxEnt reconstructions. Columns correspond to the following values of : (left), (middle) and (right). The rows show, from top to bottom, the successive difference between reconstructions on a logarithmic scale of base 10 for the standard reconstruction; the same quantity for the reconstruction using the compactification; the integrated version as a function of the number of moments for the standard reconstruction; and the same quantity for the reconstruction using the compactification.
Only two cases show an almost vanishing : for () and for (). By construction of Eq. (110), this means that and , respectively, which is why these curves overlap in Figure 1. Such coincidences are not evidence of convergence: a small successive difference is necessary but not sufficient, since it only indicates that the added moment barely modifies the reconstruction. The decisive quantity is the distance to the exact result, , shown in the third row of Figure 3. After an initial decrease at low , it does not decrease any further: for it grows monotonically for , and for and it grows or fluctuates beyond . Adding moments therefore does not bring closer to the exact greybody factor.
Regarding the reconstruction using the compactification, in the second row of Figure 4 we can see that for all , since all share the same value at the origin, where the approach that uses the compactification breaks down. Immediately after, shows a narrow peak and then drops by many orders of magnitude. Excepting for the origin, and in contrast to the standard reconstruction, decreases with . Likewise, in the fourth row decreases with , with only minor non-monotonic fluctuations (except for , where they are more pronounced). Since adding moments no longer changes the reconstruction, we stop at .
tbl:I compares, for the formulations with and without compactification, the minimum over of the integrated error with respect to the exact result. We quantify the improvement as
For and , the integrated error of the best compactified reconstruction is 54% and 80% lower, respectively, than that of the best standard one. In addition, in the compactified formulation the error decreases systematically with , so a suitable can be chosen without knowing the exact result. In the standard formulation, by contrast, the error is non-monotonic in , and the optimal value can only be identified by comparing with , which is not available in practice. This supports the use of the compactification for these values of .
For the situation is different. The compactified reconstruction with still gives a lower integrated error, but the gain is smaller (38%). The region , which is excluded from the integral for all values of , is not negligible here: the transition of lies close to the origin (), so the breakdown region and the spurious oscillations of affect a sizeable part of the relevant range, whereas for and the transition is far from the origin. Since the improvement relies on discarding a region that matters for , we prefer the standard reconstruction with in this case.
These results demonstrate that the MaxEnt method, applied to a modest number of KdV integrals, reconstructs the Pöschl-Teller greybody factor with good accuracy. The form without compactification is the best option when only a few moments are available or when is small, but its accuracy deteriorates when more moments are added and it depends on a number of ad hoc truncations. The form with compactification is clean and converges systematically for large , but it requires more moments and it cannot reproduce the region around . This behaviour will also be reflected in the following section, when we apply the method to the BH potentials, where no exact solution is available.
Greybody Factors for the Schwarzschild BH
We now apply the MaxEnt method to the Regge-Wheeler BH potential in Eq. (5), which describes the response of the BH to axial perturbations of a Schwarzschild BH. Unlike the Pöschl-Teller case, no closed-form expression for the greybody factor are available nor expected. The accuracy of the MaxEnt reconstruction is assessed by comparison with results of the WKB approximation [61, 62, 63].
Using a dimensionless radial variable, , the Regge-Wheeler potential takes the simpler form
The expressions for the first KdV integrals for this BH potential are given in Appendix B.
Figure 5 compares the MaxEnt reconstructions with the fourth-order WKB approximation for . For the compactified formulation we take in all cases, since the error decreases approximately monotonically with , as found for the Pöschl-Teller potential. In the standard formulation the behavior is non-monotonic in , so the optimal choice is not obvious a priori. For transitions far from the origin, is already a good choice, as in the Pöschl-Teller benchmark for and . The case requires more care: its transition lies close to the origin, as for , where the benchmark favors a smaller () and the standard error worsens as moments are added. However, the reconstructions with and turn out to be practically identical: their integrated difference, computed excluding the region as in the Pöschl–Teller case, is of order . We therefore use for as well, as for the other multipoles.

Figure 5. Greybody factors for the Regge-Wheeler potential as functions of the dimensionless frequency for (top), (middle) and (bottom). The results of the MaxEnt reconstructions without compactification (, orange) and with compactification (, green, dashed) are compared with the fourth-order WKB approximation.
For and the step lies far from the origin, and the compactified curves are nearly indistinguishable from WKB at the scale of the figure, describing the central part of the transition better than the standard ones. For neither reconstruction is uniformly closer to WKB. The standard reconstruction coincides with WKB at small and for , but is shifted towards larger in the transition region. The compactified one follows WKB more closely in the rising part of the transition, but it slightly underestimates for – and approaches more slowly; its integrated difference with respect to WKB is only somewhat lower (19%) than that of the standard one.
In all cases the compactified curves jump to at , as discussed in § 1.6. Since WKB is itself an approximation, the residual differences cannot be attributed solely to the reconstruction.
It is worth remembering that by virtue of the Darboux covariance, these results extend to any admissible BH potential, in particular for the Zerilli potential of the polar sector of the Schwarzschild metric perturbations.
The MaxEnt method offers several advantages over other approaches. Unlike Padé approximants [79], which introduce poles that require physical interpretation, the MaxEnt reconstruction is an entire function of with no spurious singularities. Unlike the WKB approximation, which becomes increasingly complicated at higher orders, the MaxEnt method can be systematically improved by including additional moments within the limits set by the convergence of the dual problem, with the structure of the calculation remaining unchanged.
Conclusions and Discussion
In this paper, we have developed and validated a new method for computing BH greybody factors using the MaxEnt principle. The method builds on recent work showing that the KdV integrals of the BH potential fully determine the greybody factors via a classical moment problem [72, 73, 74]. The MaxEnt method selects the least-biased distribution compatible with a finite set of moments, guaranteeing positivity at every truncation order [81].
We first validated the method against the exactly solvable Pöschl-Teller potential, where we used two formulations of the moment problem: the original one on the semi-infinite domain and a reformulation that compactifies the frequency domain onto , leaving a moment problem similar to a Hausdorff moment problem but not identical. The resulting method, instead of using moments defined with the powers of the frequency, , uses the Bernstein polynomials, which appear in a natural way from the way in which we carry out the compactification. The two formulations are complementary. The standard one is the best option for small or when only a few
moments are available: its integrated error with respect to the exact result reaches a minimum at , 6 and 5 for , 5 and 10, respectively (Table I), and grows beyond that point. Its accuracy, however, deteriorates when more moments are added. Our results also suggest that the standard method is limited by the numerical solution of the dual problem: starting at for , for and for , the highest-order multiplier, which must be positive, ends up at its lower bound. The minimization then stops on that bound instead of at a stationary point, so the highest moments are not matched. The truncation of the infinite frequency domain and the clipping of the exponent may also contribute. The formulation that uses the compactification converges systematically with for large . The price is that it cannot reproduce a tiny region around the origin of the frequency, where jumps to 1 for any value of the Lagrange multipliers used in the MaxEnt method. This appears to be a structural consequence of the compactification, not a failure of the minimization procedure.
| % of Improvement | |||||
| 1 | 4 | 10 | 37.53% | ||
| 5 | 6 | 10 | 54.38% | ||
| 10 | 5 | 10 | 79.81% |
Table I. Minimum integrated error associated with the plots shown Figure 3 for the MaxEnt reconstruction without (third row) and with (fourth row) compactification. The improvement is calculated using Eq. (111).
We then applied the method to the Regge–Wheeler potential governing axial perturbations of a Schwarzschild BH, for . The compactified reconstruction with agrees closely with the fourth-order WKB approximation away from the breakdown region, while the standard one with agrees somewhat less well. For , whose transition lies close to the origin, neither formulation is uniformly closer to WKB.
This is the first application of the MaxEnt method to the reconstruction of BH greybody factors, and it demonstrates that a modest number of KdV integrals is sufficient to recover them accurately.
The MaxEnt method has several advantages for this problem: (i) it guarantees the positivity of the reconstructed distribution at every truncation order; (ii) the reconstruction is an entire function of , with no spurious poles; (iii) additional moments can be incorporated in a straightforward way; and (iv) the dual problem is convex, so standard optimizers like L-BFGS-B can be used, although convergence at high requires care.
The method can be extended to other situations of physical interest, such as (i) perturbations of rotating (Kerr) BHs, using the Teukolsky formalism [134, 135]; (ii) perturbations of exotic compact objects that mimic BHs; (iii) perturbations of BHs in alternative theories of gravity; (iv) perturbations of BHs in higher dimensions; and (v) scalar, vector, and neutrino perturbations, in addition to gravitational ones.
The connection between KdV integrals, DTs, and greybody factors may also provide new insights into the structure of BH perturbation theory and its relationship to integrable systems. The fact that the KdV integrals are invariant under DTs suggests a deep connection between the isospectral symmetries of the Schrödinger equation and the physical scattering data.
In summary, the MaxEnt method provides a systematic and computationally inexpensive route to BH greybody factors from KdV integrals, with positivity guaranteed at every truncation order. Its accuracy is, however, limited by the numerical solution of the dual problem. In the semi-infinite formulation the highest-order multiplier reaches its lower bound from (), () and (); the minimization then stops on the bound and the highest moments are not matched. The compactified formulation remains stable up to the highest order considered (), but it cannot reproduce a narrow region around the origin, whose width depends on and (– in our runs), and its error decreases with only up to small non-monotonic fluctuations. In practice, a modest number of moments, chosen according to the formulation and the potential, yields accurate reconstructions.
Acknowledgments
CFS is supported by contracts PID2019-106515GB-I00/AEI/10.13039/501100011033 and PID2025-173920NB-C21/AEI/10.13039/501100011033 (Spanish Ministry of Science and Innovation) and 2025 SGR 00191 (AGAUR, Generalitat de Catalunya). The work of ML is supported by the F.R.S.-FNRS Belgium through convention T. 44516.26 and CDR grant J.0006.26. This work was also partially supported by the program Unidad de Excelencia María de Maeztu CEX2020-001058-M (Spanish Ministry of Science and Innovation). We have used the Computer Algebra System Maxima [136] and the SciPy library [137] for the numerical computations in this paper.
KdV Integrals for the Pöschl-Teller Potential
For the reader’s convenience, we reproduce here the first non-vanishing KdV integrals of the Pöschl-Teller potential (62), originally given in Appendix B of Ref. [80], so that the paper is self-contained. In dimensionless form (with ), they read:
KdV Integrals for the Regge-Wheeler Potential
For the reader’s convenience, we reproduce here the first non-vanishing KdV integrals of the Regge-Wheeler potential (5), originally computed in Appendix E of Ref. [79], so that the paper is self-contained. They are given in units , in terms of .
Details of the Numerical Implementation
The MaxEnt reconstruction of the greybody factors from the BH moment problem has been implemented in Python. We used SciPy [137] for the L-BFGS-B minimization and the numerical quadrature, NumPy for array operations, and Maxima [136] for the symbolic computation of the KdV integrals.
Before the minimization, the moments were rescaled as in Eq. (57) for the case without compactification and as in Eq. (93) for the case with compactification to improve the numerical conditioning. The dual potential (53) and (87) were then minimized with L-BFGS-B, using a maximum of 10000 iterations, together with a relative tolerance of on the decrease of the objective function (ftol), and a tolerance of on the projected gradient (gtol). For the case without compactification, a bound on the highest-order multiplier was also set.
The integral defining the partition function (52) was evaluated with adaptive Gauss–Kronrod quadrature, truncated at a cutoff in the rescaled variable. The cutoff must be large enough for the integrand to be negligible beyond it, and since this depends on the potential, we took for the Pöschl–Teller potential (for all ) and , 5 and 7 for the Regge–Wheeler potential with , 5 and 10, respectively. Finally, the exponent in the integrand was clipped to to prevent numerical overflow during the optimization.
References
References
- [1]R. Ruffini and J. A. Wheeler, Introducing the black hole, Phys. Today 24, 30 (1971).
- [2]S. W. Hawking and G. F. R. Ellis, The Large Scale Structure of Space-Time, Cambridge Monographs on Mathematical Physics (Cambridge University Press, 2011).
- [3]R. Wald, General Relativity (The University of Chicago Press, Chicago, 1984).
- [4]I. D. Novikov and V. P. Frolov, Physics of Black Holes (Kluwer Academic, Dordrecht, Netherlands, 1989).
- [5]S. Chandrasekhar, The Mathematical Theory of Black Holes (Oxford University Press, New York, 1992).
- [6]R. Penrose, Gravitational collapse and space-time singularities, Phys. Rev. Lett. 14, 57 (1965).
- [7]R. Penrose, Gravitational collapse: The role of general relativity, Riv. Nuovo Cim. 1, 252 (1969).
- [8]R. M. Wald, Black hole in a uniform magnetic field, Phys. Rev. D 10, 1680 (1974).
- [9]J. D. Bekenstein, Black hole hair: 25 - years after, in 2nd International Sakharov Conference on Physics (1996) pp. 216–219, arXiv:gr-qc/9605059.
- [10]P. T. Chrusciel, J. Lopes Costa, and M. Heusler, Stationary Black Holes: Uniqueness and Beyond, Living Rev. Rel. 15, 7 (2012), arXiv:1205.6112 [gr-qc].DOI
- [11]R. P. Kerr, Gravitational field of a spinning mass as an example of algebraically special metrics, Phys. Rev. Lett. 11, 237 (1963).
- [12]E. T. Newman, E. Couch, K. Chinnapared, A. Exton, A. Prakash, and R. Torrence, Metric of a rotating, charged mass, J. Math. Phys. 6, 918 (1965).
- [13]W. Israel, Event horizons in static vacuum space-times, Phys. Rev. 164, 1776 (1967).DOI
- [14]W. Israel, Event horizons in static electrovac space-times, Commun. Math. Phys. 8, 245 (1968).DOI
- [15]B. Carter, Axisymmetric black hole has only two degrees of freedom, Phys. Rev. Lett. 26, 331 (1971).
- [16]S. W. Hawking, Black holes in general relativity, Commun. Math. Phys. 25, 152 (1972).
- [17]D. C. Robinson, Uniqueness of the Kerr black hole, Phys. Rev. Lett. 34, 905 (1975).
- [18]P. O. Mazur, Proof of uniqueness of the Kerr-Newman black hole solution, J. Phys. A 15, 3173 (1982).
- [19]R. H. Price, Nonspherical perturbations of relativistic gravitational collapse. 1. Scalar and gravitational perturbations, Phys. Rev. D 5, 2419 (1972).
- [20]R. H. Price, Nonspherical Perturbations of Relativistic Gravitational Collapse. II. Integer-Spin, Zero-Rest-Mass Fields, Phys. Rev. D 5, 2439 (1972).DOI
- [21]S. Klainerman and J. Szeftel, Kerr stability for small angular momentum, Pure Appl. Math. Quart. 19, 791 (2023), arXiv:2104.11857 [math.AP].arxiv.org/abs/2104.11857
- [22]E. Giorgi, S. Klainerman, and J. Szeftel, Wave equations estimates and the nonlinear stability of slowly rotating Kerr black holes, Pure Appl. Math. Quart. 20, 2865 (2024), arXiv:2205.14808 [math.AP].arxiv.org/abs/2205.14808
- [23]C. A. R. Herdeiro and E. Radu, Asymptotically flat black holes with scalar hair: a review, Int. J. Mod. Phys. D 24, 1542014 (2015), arXiv:1504.08209 [gr-qc].arxiv.org/abs/1504.08209
- [24]D. M. Eardley and W. H. Press, Astrophysical processes near black holes, Ann. Rev. Astron. Astrophys. 13, 381 (1975).
- [25]M. Zajáček and A. Tursunov, Electric charge of black holes: Is it really always negligible?, The Observatory 139, 231 (2019), arXiv:1904.04654 [astro-ph.GA].arxiv.org/abs/1904.04654
- [26]J. M. Corral-Santana, J. Casares, T. Muñoz-Darias, F. E. Bauer, I. G. Martínez-Pais, and D. M. Russell, BlackCAT: A catalogue of stellar-mass black holes in X-ray transients, Astron. Astrophys. 587, A61 (2016), arXiv:1510.08869 [astro-ph.HE].
- [27]B. P. Abbott et al. (LIGO Scientific, Virgo), GWTC-1: A gravitational-wave transient catalog of compact binary mergers observed by LIGO and Virgo during the first and second observing runs, Phys. Rev. X 9, 031040 (2019), arXiv:1811.12907 [astro-ph.HE].DOI
- [28]R. Abbott et al. (LIGO Scientific, VIRGO), GWTC-2.1: Deep extended catalog of compact binary coalescences observed by LIGO and Virgo during the first half of the third observing run, Phys. Rev. D 109, 022001 (2024), arXiv:2108.01045 [gr-qc].arxiv.org/abs/2108.01045
- [29]R. Abbott et al. (KAGRA, VIRGO, LIGO Scientific), GWTC-3: Compact binary coalescences observed by LIGO and Virgo during the second part of the third observing run, Phys. Rev. X 13, 041039 (2023), arXiv:2111.03606 [gr-qc].DOI
- [30]B. Sathyaprakash et al., Scientific Objectives of Einstein Telescope, Class. Quant. Grav. 29, 124013 (2012), [Erratum: Class.Quant.Grav. 30, 079501 (2013)], arXiv:1206.0331 [gr-qc].DOI
- [31]M. Evans et al., A horizon study for Cosmic Explorer: Science, observatories, and community (2021), arXiv:2109.09882 [astro-ph.IM].arxiv.org/abs/2109.09882
- [32]P. Amaro-Seoane et al. (LISA), Laser Interferometer Space Antenna (2017), arXiv:1702.00786 [astro-ph.IM].
- [33]J. A. H. Futterman, F. A. Handler, and R. A. Matzner, Scattering from Black Holes (Cambridge University Press, 2012).
- [34]N. Andersson and B. P. Jensen, Scattering by black holes. Chapter 01 (2000), arXiv:gr-qc/0011025.DOI
- [35]T. Regge and J. A. Wheeler, Stability of a Schwarzschild singularity, Phys. Rev. 108, 1063 (1957).
- [36]C. T. Cunningham, R. H. Price, and V. Moncrief, Radiation from collapsing relativistic stars. I - Linearized odd-parity radiation, Astrophys. J. 224, 643 (1978).DOI
- [37]C. T. Cunningham, R. H. Price, and V. Moncrief, Radiation from collapsing relativistic stars. II. Linearized even parity radiation, Astrophys. J. 230, 870 (1979).DOI
- [38]C. T. Cunningham, R. H. Price, and V. Moncrief, Radiation from collapsing relativistic stars. III - Second order perturbations of collapse with rotation, Astrophys. J. 236, 674 (1980).DOI
- [39]F. J. Zerilli, Gravitational Field of a Particle Falling in a Schwarzschild Geometry Analyzed in Tensor Harmonics, Phys. Rev. D 2, 2141 (1970).DOI
- [40]F. J. Zerilli, Effective potential for even parity Regge-Wheeler gravitational perturbation equations, Phys. Rev. Lett. 24, 737 (1970).DOI
- [41]V. Moncrief, Gravitational perturbations of spherically symmetric systems. I. The exterior problem, Ann. Phys. (N.Y.) 88, 323 (1974).
- [42]A. Ishibashi and H. Kodama, Stability of higher dimensional Schwarzschild black holes, Prog. Theor. Phys. 110, 901 (2003), arXiv:hep-th/0305185.DOI
- [43]H. Kodama and A. Ishibashi, Master equations for perturbations of generalized static black holes with charge in higher dimensions, Prog. Theor. Phys. 111, 29 (2004), arXiv:hep-th/0308128.DOI
- [44]S. W. Hawking, Particle Creation by Black Holes, Commun. Math. Phys. 43, 199 (1975), [Erratum: Commun. Math. Phys. 46, 206 (1976)].
- [45]K. D. Kokkotas and B. G. Schmidt, Quasinormal modes of stars and black holes, Living Rev. Rel. 2, 2 (1999), arXiv:gr-qc/9909058.DOI
- [46]H.-P. Nollert, Quasinormal modes: the characteristic ‘sound’ of black holes and neutron stars, Class. Quant. Grav. 16, 159 (1999).
- [47]E. Berti, V. Cardoso, and A. O. Starinets, Quasinormal modes of black holes and black branes, Class. Quant. Grav. 26, 163001 (2009), arXiv:0905.2975 [gr-qc].DOI
- [48]R. Konoplya and A. Zhidenko, Quasinormal modes of black holes: From astrophysics to string theory, Rev. Mod. Phys. 83, 793 (2011), arXiv:1102.4014 [gr-qc].DOI
- [49]L. Barack et al., Black holes, gravitational waves and fundamental physics: a roadmap, Class. Quant. Grav. 36, 143001 (2019), arXiv:1806.05195 [gr-qc].DOI
- [50]S. E. Perkins, N. Yunes, and E. Berti, Probing Fundamental Physics with Gravitational Waves: The Next Generation, Phys. Rev. D 103, 044024 (2021), arXiv:2010.09010 [gr-qc].DOI
- [51]E. Barausse et al., Prospects for fundamental physics with LISA, Gen. Rel. Grav. 52, 81 (2020), arXiv:2001.09793 [gr-qc].DOI
- [52]K. G. Arun et al. (LISA), New horizons for fundamental physics with LISA, Living Rev. Rel. 25, 4 (2022), arXiv:2205.01597 [gr-qc].DOI
- [53]A. Abac et al. (ET), The science of the Einstein Telescope, JCAP 03 (03), 081, arXiv:2503.12263 [gr-qc].
- [54]D. N. Page, Particle Emission Rates from a Black Hole: Massless Particles from an Uncharged, Nonrotating Hole, Phys. Rev. D 13, 198 (1976).
- [55]W. G. Unruh, Absorption Cross-Section of Small Black Holes, Phys. Rev. D 14, 3251 (1976).
- [56]A. A. Starobinskii, Amplification of waves reflected from a rotating ”black hole”, Sov. Phys. JETP 37, 28 (1973).
- [57]A. A. Starobinskii and S. M. Churilov, Amplification of electromagnetic and gravitational waves scattered by a rotating "black hole", Sov. Phys. JETP 65, 1 (1974).
- [58]N. G. Sanchez, Scattering of scalar waves from a Schwarzschild black hole, J. Math. Phys. 17, 688 (1976).
- [59]N. G. Sanchez, The Wave Scattering Theory and the Absorption Problem for a Black Hole, Phys. Rev. D 16, 937 (1977).DOI
- [60]N. G. Sanchez, Elastic Scattering of Waves by a Black Hole, Phys. Rev. D 18, 1798 (1978).
- [61]B. F. Schutz and C. M. Will, Black Hole Normal Modes: A Semianalytic Approach, Astrophys. J. Lett. 291, L33 (1985).
- [62]S. Iyer and C. M. Will, Black Hole Normal Modes: A WKB Approach. 1. Foundations and Application of a Higher Order WKB Analysis of Potential Barrier Scattering, Phys. Rev. D 35, 3621 (1987).
- [63]S. Iyer, Black Hole Normal Modes: A WKB Approach. 2. Schwarzschild Black Holes, Phys. Rev. D 35, 3632 (1987).
- [64]R. A. Konoplya, Quasinormal behavior of the d-dimensional Schwarzschild black hole and higher order WKB approach, Phys. Rev. D 68, 024018 (2003), arXiv:gr-qc/0303052.
- [65]J. Matyjasek and M. Opala, Quasinormal modes of black holes. The improved semianalytic approach, Phys. Rev. D 96, 024011 (2017), arXiv:1704.00361 [gr-qc].DOI
- [66]R. A. Konoplya, A. Zhidenko, and A. F. Zinhailo, Higher order WKB formula for quasinormal modes and grey-body factors: recipes for quick and accurate calculations, Class. Quant. Grav. 36, 155002 (2019), arXiv:1904.10333 [gr-qc].DOI
- [67]M. Visser, Some general bounds for 1-D scattering, Phys. Rev. A 59, 427 (1999), arXiv:quant-ph/9901030.DOI
- [68]P. Boonserm and M. Visser, Bounding the greybody factors for Schwarzschild black holes, Phys. Rev. D 78, 101502 (2008), arXiv:0806.2209 [gr-qc].DOI
- [69]A. Neitzke, Greybody factors at large imaginary frequencies (2003), arXiv:hep-th/0304080.DOI
- [70]A. Castro, J. M. Lapan, A. Maloney, and M. J. Rodriguez, Black hole scattering from monodromy, Class. Quant. Grav. 30, 165005 (2013), arXiv:1304.3781 [hep-th].DOI
- [71]T. Harmark, J. Natario, and R. Schiappa, Greybody Factors for d-Dimensional Black Holes, Adv. Theor. Math. Phys. 14, 727 (2010), arXiv:0708.0017 [hep-th].arxiv.org/abs/0708.0017
- [72]M. Lenzi and C. F. Sopuerta, Master functions and equations for perturbations of vacuum spherically symmetric spacetimes, Phys. Rev. D 104, 084053 (2021), arXiv:2108.08668 [gr-qc].DOI
- [73]M. Lenzi and C. F. Sopuerta, Darboux covariance: A hidden symmetry of perturbed Schwarzschild black holes, Phys. Rev. D 104, 124068 (2021), arXiv:2109.00503 [gr-qc].DOI
- [74]M. Lenzi and C. F. Sopuerta, Black hole greybody factors from Korteweg–de Vries integrals: Theory, Phys. Rev. D 107, 014010 (2023), arXiv:2212.03732 [gr-qc].DOI
- [75]M. Lenzi and C. F. Sopuerta, Gauge-independent metric reconstruction of perturbations of vacuum spherically-symmetric spacetimes, Phys. Rev. D 109, 084030 (2024), arXiv:2402.10004 [gr-qc].arxiv.org/abs/2402.10004
- [76]M. Lenzi, G. A. Mena Marugán, A. Mínguez-Sánchez, and C. F. Sopuerta, Master variables and Darboux symmetry for axial perturbations of the exterior and interior of black hole spacetimes, Phys. Rev. D 113, 064039 (2026), arXiv:2512.10664 [gr-qc].DOI
- [77]M. Lenzi, G. A. Mena Marugán, A. Mínguez-Sánchez, and C. F. Sopuerta, Hamiltonian formalism, master functions and Darboux transformations for perturbed (interiors and exteriors of) nonrotating black holes, Gen. Rel. Grav. 58, 56 (2026), arXiv:2605.19829 [gr-qc].DOI
- [78]J. L. Jaramillo, M. Lenzi, and C. F. Sopuerta, Integrability in perturbed black holes: Background hidden structures, Phys. Rev. D 110, 104049 (2024), arXiv:2407.14196 [gr-qc].DOI
- [79]M. Lenzi and C. F. Sopuerta, Black hole greybody factors from Korteweg–de Vries integrals: Computation, Phys. Rev. D 107, 084039 (2023), arXiv:2301.01096 [gr-qc].DOI
- [80]M. Lenzi, A. M. Agudo, and C. F. Sopuerta, Korteweg-de Vries integrals for modified black hole potentials: instabilities and other questions, JCAP 2025 (09), 021, arXiv:2503.09918 [gr-qc].
- [81]L. R. Mead and N. Papanicolaou, Maximum entropy in the problem of moments, J. Math. Phys. 25, 2404 (1984), https://doi.org/10.1063/1.526446.
- [82]C. E. Shannon, A Mathematical Theory of Communica- tion. The Bell System Technical Journal 27, 379 (1948).
- [83]E. T. Jaynes, Information theory and statistical mechanics, Phys. Rev. 106, 620 (1957).
- [84]E. T. Jaynes, Information Theory and Statistical Mechanics. II, Phys. Rev. 108, 171 (1957).
- [85]Pöschl, G. and Teller, E., Bemerkungen zur Quantenmechanik des anharmonischen Oszillators, Z. Phys. 83, 143 (1933).DOI
- [86]K. Schwarzschild, On the gravitational field of a mass point according to Einstein’s theory, Sitzungsber. Preuss. Akad. Wiss. Berlin (Math. Phys.) 1916, 189 (1916), arXiv:physics/9905030 [physics].
- [87]D. J. Korteweg and G. de Vries, On the change of form of long waves advancing in a rectangular canal, and on a new type of long stationary waves, Philos. Mag. 39, 422 (1895).
- [88]M. J. Ablowitz and H. Segur, Solitons and the Inverse Scattering Transform (Society for Industrial and Applied Mathematics, Philadelphia, 1981).
- [89]V. E. Zakharov, ed., What is integrability? (Springer Berlin, Heidelberg, 1991).
- [90]M. J. Ablowitz and P. A. Clarkson, Solitons, Nonlinear Evolution Equations and Inverse Scattering, London Mathematical Society Lecture Note Series, Vol. 149 (Cambridge University Press, 1991).
- [91]M. A. Semenov-Tian-Shansky, Integrable Systems and Factorization Problems, arX iv e-prints , nlin/0209057 (2002), arXiv:nlin/0209057 [nlin.SI].
- [92]O. Babelon, D. Bernard, and M. Talon, Introduction to Classical Integrable Systems, Cambridge Monographs on Mathematical Physics (Cambridge University Press, 2003).
- [93]L. D. Faddeev, The Inverse problem in the quantum theory of scattering, Usp. Mat. Nauk 14, 57 (1959).
- [94]L. D. Faddeev, Inverse problem of quantum scattering theory. II, J. Sov. Math. 5, 334 (1976).
- [95]P. Deift and E. Trubowitz, Inverse scattering on the line, Commun. Pure Appl. Math. 32, 121 (1979).
- [96]S. Novikov, S. V. Manakov, L. P. Pitaevsky, and V. E. Zakharov, Theory of Solitons. The Inverse Scattering Method (Consultants Bureau, New York, 1984).
- [97]C. S. Gardner, J. M. Greene, M. D. Kruskal, and R. M. Miura, Method for solving the Korteweg-de Vries equation, Phys. Rev. Lett. 19, 1095 (1967).
- [98]R. M. Miura, C. S. Gardner, and M. D. Kruskal, Korteweg-de Vries Equation and Generalizations. II. Existence of Conservation Laws and Constants of Motion, J. Math. Phys. 9, 1204 (1968).DOI
- [99]V. E. Zakharov and L. D. Faddeev, Korteweg-de Vries equation: A completely integrable Hamiltonian system, Functional Analysis and Its Applications 5, 280 (1971).
- [100]P. D. Lax, Integrals of Nonlinear Equations of Evolution and Solitary Waves, Commun. Pure Appl. Math. 21, 467 (1968).
- [101]H. Hamburger, On an Extension of the Stieltjes Moment Problem (Part I), Mathematische Annalen 81, 235 (1920).
- [102]F. Hausdorff, Summation Methods and Moment Sequences. I, Mathematische Zeitschrift 9, 74 (1921).
- [103]H. Hamburger, On an Extension of the Stieltjes Moment Problem (Part II), Mathematische Annalen 82, 120 (1921).
- [104]H. Hamburger, On an Extension of the Stieltjes Moment Problem (Part III), Mathematische Annalen 82, 168 (1921).
- [105]M. Riesz, Sur le problème des moments. Première note, Ark. Mat. Astron. Fys. 16, 1 (1922).
- [106]M. Riesz, Sur le problème des moments. Deuxième note, Ark. Mat. Astron. Fys. 16, 1 (1922).
- [107]M. Riesz, Sur le problème des moments. Troisième note, Ark. Mat. Astron. Fys. 17, 1 (1923).
- [108]N. Akhiezer, The Classical Moment Problem and Some Related Questions in Analysis, University mathematical monographs (Oliver & Boyd, 1965).
- [109]T. H. Kjeldsen, The early history of the moment problem, Historia Mathematica 20, 19 (1993).
- [110]K. Schmüdgen, The Moment Problem, Graduate Texts in Mathematics (Springer International Publishing, 2017).
- [111]G. A. Baker and P. Graves-Morris, Padé Approximants, 2nd ed., Encyclopedia of Mathematics and its Applications (Cambridge University Press, 1996).
- [112]V. I. Arnold, Mathematical Methods of Classical Mechanics, Graduate Texts in Mathematics, Vol. 60 (Springer-Verlag, New York, 1989).
- [113]T. J. Stieltjes, Recherches sur les fractions continues, Annales de la Faculté des sciences de Toulouse : Mathématiques 8, J1 (1894).
- [114]J. A. Shohat and J. D. Tamarkin, The problem of moments (American Mathematical Society, Providence, Rhode Island, 1943).
- [115]C. Bender and S. Orszag, Advanced Mathematical Methods for Scientists and Engineers (McGraw-Hill, 1978).
- [116]J. M. Borwein and A. S. Lewis, Convergence of best entropy estimates, SIAM J. Optim. 1, 191 (1991).
- [117]A. Tagliani, Hausdorff moment problem and maximum entropy: A unified approach, Applied Mathematics and Computation 105, 291 (1999).DOI
- [118]R. V. Abramov, The multidimensional maximum entropy moment problem: A review of numerical methods, Commun. Math. Sci. 8, 377 (2010).
- [119]P. Biswas and S. R. Elliott, A swarm intelligence approach to density function reconstruction from moments using entropy optimization, Proc. Natl. Acad. Sci. USA 123, e2523574123 (2026).DOI
- [120]I. Turek, A maximum-entropy approach to the density of states within the recursion method, J. Phys. C: Solid State Phys. 21, 3251–3260 (1988).DOI
- [121]D. A. Drabold and O. F. Sankey, Maximum entropy approach for linear scaling in the electronic structure problem, Phys. Rev. Lett. 70, 3631 (1993).
- [122]R. N. Silver and H. Röder, Calculation of densities of states and spectral functions by Chebyshev recursion and maximum entropy, Phys. Rev. E 56, 4822 (1997).DOI
- [123]K. Bandyopadhyay, A. K. Bhattacharya, P. Biswas, and D. A. Drabold, Maximum entropy and the problem of moments: A stable algorithm, Phys. Rev. E 71, 057701 (2005), arXiv:cond-mat/0412717.DOI
- [124]S. Zhang, X. Wang, T. Lin, and L. Chang, Reconstructing parton distribution function based on maximum entropy method, Chin. Phys. C 48, 033106 (2024), arXiv:2309.01417.DOI
- [125]D. Poland, Maximum-entropy representation of non-Gaussian polymer distribution functions, J. Chem. Phys. 102, 2604 (1995).DOI
- [126]J. Ding and L. R. Mead, Maximum entropy approximation for Lyapunov exponents of chaotic maps, J. Math. Phys. 43, 2518 (2002).
- [127]P. Biswas, H. Shimoyama, and L. R. Mead, Lyapunov exponent and natural invariant density determination of chaotic maps: An iterative maximum entropy ansatz, J. Phys. A: Math. Theor. 43, 125103 (2010), arXiv:0910.4561.DOI
- [128]C. D. Levermore, Moment closure hierarchies for kinetic theories, J. Stat. Phys. 83, 1021 (1996).
- [129]C. D. Hauck, C. D. Levermore, and A. L. Tits, Convex duality and entropy-based moment closures: Characterizing degenerate densities, SIAM J. Control Optim. 47, 1977 (2008).
- [130]R. H. Byrd, P. Lu, and J. Nocedal, A Limited Memory Algorithm for Bound Constrained Optimization, SIAM Journal on Scientific and Statistical Computing 16, 1190 (1995).
- [131]C. Zhu, R. H. Byrd, and J. Nocedal, L-BFGS-B: Algorithm 778: L-BFGS-B, FORTRAN routines for large scale bound constrained optimization, ACM Transac-
- [132]V. Ferrari and B. Mashhoon, Oscillations of a black hole, Phys. Rev. Lett. 52, 1361 (1984).
- [133]V. Ferrari and B. Mashhoon, New approach to the quasinormal modes of a black hole, Phys. Rev. D 30, 295 (1984).
- [134]S. A. Teukolsky, Rotating Black Holes: Separable Wave Equations for Gravitational and Electromagnetic Perturbations, Phys. Rev. Lett. 29, 1114 (1972).
- [135]S. A. Teukolsky, Perturbations of a rotating black hole. 1. Fundamental equations for gravitational electromagnetic and neutrino field perturbations, Astrophys. J. 185, 635 (1973).
- [136]Maxima, Maxima, a computer algebra system, version 5.46.0 (2022).
- [137]P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, and et al., SciPy 1.0: fundamental algorithms for scientific computing in Python, Nature Methods 17, 261 (2020).