Approach to the separatrix with eccentric orbits

Guillaume Lhost⋆ , Geoffrey Compère⊙

⋆
Physique de l’Univers, Champs et Gravitation, Université de Mons – UMONS,
Place du Parc 20, 7000 Mons, Belgium
⊙
Université Libre de Bruxelles – International Solvay Institutes – BLU-ULB space center,
CP 231, B-1050 Brussels, Belgium
arXiv:2412.04249v1 [gr-qc] 5 Dec 2024

E-mail: guillaume.lhost@umons.ac.be, geoffrey.compere@ulb.be

Abstract: Eccentric binary compact mergers are prime targets of current and future
gravitational wave observatories. In the small mass ratio expansion, post-adiabatic inspirals
have been modeled up to the separatrix, where first-principle modeling currently ends. In
this paper, we derive the analytic late time solution to the adiabatic inspiral in terms of
self-force coefficients at the separatrix. We identify the role of the Lambert W−1 function
as a key mathematical ingredient in the approach to the separatrix.
Contents

1 Introduction 2

2 Review of eccentric orbits around Schwarzschild 3

2.1 Geodesic Equations 3
2.2 Quasi-Keplerian parametrization of bound orbits 4
2.3 Fundamental Boyer-Lindquist frequencies 5
2.4 Osculating Equations 6
2.5 Adiabatic equations 7
2.6 Adiabatic equations: expansion in Fourier modes 9

3 Analytic expansion towards the separatrix 10

3.1 Expansion of elliptic integrals 11
3.2 Asymptotic adiabatic equations 12
3.3 Asymptotic analytical solutions 15
3.4 Consequences 17

4 Conclusion and outlook 22

A Osculating equations 23

B Limits of elliptic integrals 24

B.1 Conventions 24
B.2 Reformulation of some elliptic functions 25
B.3 Asymptotic expansions 26

C Coefficients B and A 28

D Subleading orders of the asymptotic adiabatic equations and solutions 29

E Lambert W function 30

–1–
1 Introduction

Solving the two-body problem in General Relativity has never been as pressing as today,
given the breakthrough of gravitational wave observations of compact binary mergers [1]
and prospect of upcoming ET, LISA and Cosmic Explorer observatories. One method
to address the two-body problem is the small mass expansion or self-force framework [2].
Based on the mass ratio to separate timescales [3], adiabatic waveforms have now been
numerically generated during the inspiral phase for generic orbits [4], which is the first step
towards the much larger program to model with a numerically efficient scheme the full pa-
rameter space of post-adiabatic generic inspiral-to-plunge systems including, in particular,
eccentricity.
Eccentric orbits have been modeled within the self-force framework [5–7]. Numerical
evaluations of the self-force quantities have been performed [8–10] and orbits have been
numerically generated with increasingly faster evaluation schemes [11–19]. Only the inspiral
motion has been computed so far from first principles within the self-force formalism in
the eccentric case, though several hybrid or approximation schemes have been formulated
[20–27]. In the mass ratio expansion, adiabatic and post-adiabatic inspirals become ill-
defined at the separatrix (for a thorough description, see [28]). Formulating a consistent
dynamics around the separatrix is challenging. For the non-eccentric case, the motion
around the innermost stable circular orbit has been understood in a consistent expansion
scheme [29–31] following the seminal work of [32–34].
In this paper, we aim to get one step closer to the analytic understanding of the motion
close to the separatrix, in order to build a first-principle scheme for a transition-to-plunge
expansion which would generalize the quasi-circular case [29, 30]. The main aim of this
paper is to analytically solve for the late time dynamics of the adiabatic inspiral with
eccentricity in terms of self-force coefficients at the separatrix. Along the way, we will
reproduce and extend former analytical work for this system [5, 6].
The rest of the paper is organized as follows. We first review the description of eccentric
orbits around the Schwarzschild black hole in Section 2. We develop the analytic solution
close to the separatrix in Section 3. We conclude in Section 4. Technical expressions are
relegated to the appendices. The notebooks with a summary of technical expressions as
well as partial proofs is provided on this Github repository for the ease of reproductibility.

–2–
2 Review of eccentric orbits around Schwarzschild

In this paper, the main objective is to derive the analytic solutions of the adiabatic equa-
tions of motion of the eccentric equatorial inspiral asymptotically reaching the separatrix
in a Schwarzschild background. Before starting the involved computation of the corrections
due to the mass of the secondary, we set our notation by first reviewing the motion of a
test-particle that spans a bound orbit around a massive primary black hole. In this section,
we recall the geodesic equations around a Schwarzschild black hole located at the center of
the coordinate system. The conserved quantities and the Kepler-like parametrization are
discussed. We use the conventions of [2]. From now on, we set the spin of the primary to
zero and the geodesic motion to be equatorial.

2.1 Geodesic Equations

Consider a primary massive spherically symmetric compact object at the center of the
dimensionless Boyer-Lindquist coordinates {t, r, θ, ϕ}. The spacetime metric of such system
is given, by the Schwarzschild metric
   −1 !
2 2
ds2 = M 2 − 1− dt2 + 1 − dr2 + r2 dθ2 + r2 sin2 θ dϕ2 , (2.1)
r r

where M is the mass of the primary compact object in geometric units. The standard
dimensionful Boyer-Linquist coordinates are {M t, M r, θ, ϕ}. A test-body that moves in
this metric is located at zpµ = (t, r, θ, ϕ) and has a velocity given by the quadrivector uµ =
(ṫ, ṙ, θ̇, ϕ̇) where the operator ” ˙ ” denotes a derivative with respect to the (dimensionless)
proper time per unit primary mass τ . We denote the conserved dimensionless specific
energy E := −M −2 ut and the dimensionless specific azimuthal angular momentum as
L = M −2 uϕ . We use the primary mass to adimensionalize all quantities. We set the Carter
constant to zero to describe the equatorial motion. The geodesic equations, together with
the mass-shell constraint uα gαβ uβ = −1, are given by
 2
4 dr
r = R(r) := E 2 r4 − (L2 + r2 )(r2 − 2r),
dτ
dt r2 E
r2 = ftgeo (r) := , (2.2)
dτ 1 − 2r
dϕ
r2 = fϕgeo := L.
dτ
The quantities ftgeo (r), fϕgeo are the “instantaneous frequencies of motion”, which we denote
with the letter f as in the review [2].

–3–
2.2 Quasi-Keplerian parametrization of bound orbits

We only consider bound orbits that oscillate between the periapsis rp and the apoapsis ra .
The (dimensionless) semi-latus rectum p and eccentricity e ∈ [0; 1[ are defined as
p p
rp := ; ra := . (2.3)
1+e 1−e
The radial potential can be reexpressed in terms of these roots as

R(r) = (E 2 − 1) r (r − rp )(r − ra )(r − r3 ), (2.4)

where the third root is

2p
r3 =
. (2.5)
p−4
The specific energy and angular momentum can be written as functions of e and p as
[5]
(p − 2)2 − 4e2 p2
E2 = 2
, L2 = . (2.6)
p(p − 3 − e ) p − 3 − e2
Bound orbits exist above the separatrix characterised by p = 6 + 2e. At this specific point,
E and L take the values
√
2 2
E⋆ (e) = √ , (2.7)
9 − e2
(6 + 2e)
L⋆ (e) = √ . (2.8)
3 + 2e − e2
Whenever p > 6 + 2e, that is r3 < rp ≤ ra , the test body’s radius oscillates between ra and
rp . This fast motion, in complement with the fast azimuthal motion, can be parametrized
by the relativistic anomaly ψr defined by
p
r(ψr ) = . (2.9)
1 + e cos(ψr )
Thus the geodesic motion for r is equivalent to the geodesic equation for ψr that reads
p
2 dψr p2 p − 6 − 2e cos(ψr )
r = p =: frgeo (ψr ). (2.10)
dτ 3 2
p (p − 3 − e )
The advantage of this parametrization is that the growth of ψr (τ ) is monotonous, which
allows for a simple numerical solution starting from an initial ψr0 . This is an improvement
over the evolution of the radius, which has turning points where ṙ = 0. The time and
azimuthal phases are then solved as
ψr (τ )
dψr′
Z
ϕ(τ ) = ϕ0 + L geo , (2.11)
ψr0 fr (ψr′ )
f geo [r(ψ ′ )]
Z ψr (τ )
t(τ ) = t0 + dψr′ t geo ′ r . (2.12)
ψr0 fr (ψr )

–4–
2.3 Fundamental Boyer-Lindquist frequencies

Since the motion of interest is periodic, we can use the action-angle variables as phase
space variables to describe the motion. Here, we denote by φα the set of angle variables
{φt = t, φϕ , φr }. We warn the reader that Greek indices label both spacetime components
and components of the angle variables, depending upon the variable being used, as should
be clear from the context. They are linear in time, and their linear growth is dictated as
dφα
= Ωα , (2.13)
dt
Υα
where Ωα =: are the dimensionless Boyer-Lindquist fundamental frequencies expressed
Υt
1
RΛ geo
in terms of the fundamental Mino frequencies Υα := limΛ→∞ 2Λ −Λ dλfα where Mino
time λ is related to proper time τ using dλ = r−2 dτ . The explicit analytic form of these
functions can be found in the Black Hole Perturbation Toolkit (BHPT) following [35].
They are pure functions of eccentricity and semi-latus rectum; in particular Ωt = 1.
In terms of the angle variables, any function ξ[r(t)] can be expanded as a Fourier series
X
ξ[r(t)] = ξ(k) e−ikφr (t) . (2.14)
k∈Z

For a generic non-resonant orbit, ξ(0) is the average of ξ over a period which we also denote
as ξ(0) := ⟨ξ⟩. The generic mode ξ(k) can be written as an integral over the phase ψr (see
e.g. [2])
2π
ftgeo [r(ψr )] ikφr (ψr )
Z
ikφr (ψr ) Ωr
ξ(k) = ⟨ξ[r(ψr )]e ⟩ := e ξ[r(ψr )] dψr , (2.15)
2π 0 frgeo (ψr )
after using Eq. (2.9). The function φr (ψr ) is known analytically in terms of elliptic integrals
after solving for dφr
dψr = ftgeo Ωr /frgeo . We can therefore rewrite the integral as
Z 2π
ikφr (ψr ) 1
ξ(k) = ⟨ξ[r(ψr )]e ⟩ := eikφr (ψr ) ξ[r(ψr )] dφr . (2.16)
2π 0

As a direct consequence of these definitions, we have ⟨ dψ dϕ Ωr

H
dt ⟩ = Ωr , ⟨ dt ⟩ =
r
2π dϕ = Ωϕ
where the last integral is over an orbital period.
We can formally express the solution to the relativistic anomaly ψr in the geodesic
case in terms of the angle variable φr as

ψrgeo (φr ) = φr + ∆ψr (φr ) − ∆ψr (0) (2.17)

where
X −1  dψr 
∆ψr (φr ) = e−ikφr . (2.18)
ikΩr dt (k)
k̸=0

–5–
  
dψr
We did not derive the analytic expression for the Fourier coefficients dt , see [36] for
(k)
relevant recent work. Instead, the Mino angle variables qα are defined as

dqα
= Υα . (2.19)
dλ
The geodesic relativistic anomaly can be expressed in terms of the Mino angle variable as

ψrgeo (qr ) = qr + ∆ψr (qr ) − ∆ψr (0) (2.20)

where
X −1  dψr 
∆ψr (qr ) = e−ikqr . (2.21)
ikΥr dt q(k)
k̸=0

The k Fourier mode of a function ξ(ψr ) is now defined as

Z 2π
ikqr (ψr ) 1
ξq(k) = ⟨ξ(ψr )e ⟩ := eikqr (ψr ) ξ(ψr ) dqr . (2.22)
2π 0

2.4 Osculating Equations

Osculating methods describe accelerating motion of point particles around a given back-
ground. They are based on the following observation. At each proper time τ , the true world-
line z µ (τ ) is tangent to a geodesic zG (τ ; I A (τ )) of orbital elements I A (τ ) = {p, e, ψr0 , t0 , ϕ0 }
which evolve with proper time. Here, ψr0 , t0 , ϕ0 are the initial relativistic anomaly, time
and orbital phase, respectively. In the inspiral phase, the evolution timescale is the ra-
diation reaction timescale proportional to 1/ε where ε = mp /M is the initial small bi-
nary mass ratio. However, it is computationally simpler to switch to the orbital elements
I A (τ ) = {p, e, ψr , t, ϕ} where ψr , t, ϕ are the evolving relativistic anomaly, time, and orbital
phases. The instantaneously tangential geodesics are the osculating orbits. Following [6]
the osculating equations read

dp
= Fp (p, e, ψr ), (2.23a)
dt
de
= Fe (p, e, ψr ), (2.23b)
dt
dψr
= fr (p, e, ψr ) + δfr (p, e, ψr ), (2.23c)
dt
dϕ
= fϕ (p, e, ψr ), (2.23d)
dt
where Fp = O(ε), Fe = O(ε) and δfr = O(ε). The explicit values of the functions are given
in Appendix A. In terms of previously defined quantities we have fr = frgeo (ψr )/ftgeo (r(ψr ))

–6–
where the dependency in (p, e) was previously understood and now explicit. It is relevant
to note that in the approach to the separatrix
Fp 2(3 + e)
lim = . (2.24)
p→6+2e Fe e−1
We denote the self-force components per unit of the secondary mass mp as

d2 xµ α
µ dx dx
β
fµ = + Γ αβ . (2.25)
dτ 2 dτ dτ
The self-force components appear linearly in Eqs. (2.23a), (2.23b) and in the second term
on the right-hand side of Eq. (2.23c). Thanks to the mass-shell constraint u2 = −1,
the self-force obeys f α uα = 0. Thanks to this constraint, complementary to the equatorial
plane constraint, we can express the right-hand side of Eq. (2.23) as the linear combination
of two components of the self-force. We choose to combine the components f r and f ϕ and
write

Fp (p, e, ψr ) = Fp,ϕ (p, e, ψr )f ϕ (p, e, ψr ) + Fp,r (p, e, ψr )f r (p, e, ψr ), (2.26a)

Fe (p, e, ψr ) = Fe,ϕ (p, e, ψr )f ϕ (p, e, ψr ) + Fe,r (p, e, ψr )f r (p, e, ψr ), (2.26b)
δfr (p, e, ψr ) = δfr,ϕ (p, e, ψr )f ϕ (p, e, ψr ) + δfr,r (p, e, ψr )f r (p, e, ψr ). (2.26c)

We choose to parameterize the radius, and thus the osculating equations, with the
relativistic anomaly ψr . Instead, we could have used φr as an angle parameter, which
would avoid the use of Eq. (2.23c). The calculation of the average (2.15) applied to the
osculating equations would also be much simpler. However, using the φr parametrization
must come with knowledge of the geodesic expression r(φr ), which we do not know ana-
lytically, see however the recent work [36]. This is the reason why we decided to stick with
the parametrization ψr . The price to pay is the need to use the Jacobian in the average
(2.15).

2.5 Adiabatic equations

We now follow the multiscale expansion framework [2, 3]. We will rederive the adiabatic
equations based on the first order differential equations (2.23).
Let us denote by pi the set of the two orbital elements {p, e}. These variables can
be treated as independent slow variables in the adiabatic expansion. The angle variables
φα = {t, φr , φϕ } are the fast variables. We expand the self-force as

f α = εf(1)
α
(t̃) + ε2 f(2)
α
(t̃) + O(ε3 ), (2.27)

–7–
where the slow time is t̃ := ε t. We adopt the following adiabatic expansion
1 h (0) i
φα (t̃, ε) = φα (t̃) + ε φ(1)
α ( t̃) + O(ε 2
) ,
ε (2.28)
pi (t̃, ε) = pi(0) (t̃) + ε pi(1) (t̃) + O(ε2 ).

The fast variables satisfy

dφα 1
= Ω(0) (t̃) + ε0 Ω(1)
α (t̃) + O(ε). (2.29)
dt̃ ε α
The analysis of the adiabatic expansion [3] leads to the derivation of the 0PA equations.
They read
(0)
dφα
= Ωα (pj(0) (t̃)),
dt̃ (2.30)
dpi(0)
= Γ(1)i (pj(0) (t̃)),
dt̃
where Ωα (pj(0) ) are the geodesic formulae of the Boyer-Lindquist fundamental frequencies
evaluated as a function of the 0PA expressions of the slow variables. In particular, Ωt = 1,
(0)
which is consistent with the fact that φt = t̃. We have Ω(0) (t̃) = Ω(pi(0) (t̃)). We also have
defined * +
Γ(1)i (pj(0) ) := Fp (pj(0) , ψr ) α
, (2.31)
f α →f(1)

where the average over an orbit is given by Eq. (2.15). These functions only depend
upon the adiabatic slow variables. At 0PA order, the evolution of the slow variables is
given by the average of the osculating equations where the self-force term is replaced by
its leading O(ε) order. Moreover, we define the time reversal as f µ (ψr ) 7→ ϵµ f µ (−ψr ),
ϵµ = (−1, 1, 1, −1) and the dissipative and conservative parts of the self-force as

µ 1 1 µ µ
fdiss = f µ (ψr ) − ϵ f (−ψr ), (2.32)
2 2
µ 1 1 µ µ
fcons = f µ (ψr ) + ϵ f (−ψr ). (2.33)
2 2
At 0PA order, only the dissipative part of the self-force contributes, and we have [2, 3]
* +
Γ(1)i (pj(0) ) := Fp (pj(0) , ψr ) α
. (2.34)
f α →fdiss(1)

It is interesting to note that the formula (2.34) admits a nice quasicircular limit when
we set e(0) = 0 before computing the average. In fact, in this limit, we obtain φϕ = ϕ,
de(0)
dt̃
= 0 and
3/2
dp(0) 2(p(0) − 3)2 p(0)
= f ϕ (p(0) ). (2.35)
dt̃ p(0) − 6

–8–
This formula exactly matches with the Schwarzschild case [30, 31, 37]. In the quasi-circular
limit, p(0) is interpreted as the reduced radius of the shrinking orbital circle. When the
secondary approaches theq ISCO, the time-domain evolution of its radius is asymptotically
given by p(0) = 6 + 67/4 −f∗ϕ (t⋆ − t) for t < t⋆ , where t⋆ is the time at which the ISCO
is crossed and f∗ϕ < 0 is the ϕ component of the self-force evaluated at the ISCO.

2.6 Adiabatic equations: expansion in Fourier modes

In this section, we compute the driving force Γ(1)i of the orbital elements at adiabatic order
(2.34). We assume that the self-force is periodic in ψr and that it admits a Fourier series
expansion such that we can write
µ
X
f µ (pi , ψr ) = f(k) (pi ) e−ikψr , (2.36)
k∈Z
µ 1
R 2π
where f(k) = 2π 0 f µ (pi , ψr ) eikψr dψr . Inserting this expansion into Eq. (2.26), the right-
hand sides of the osculating equations for the orbital elements pi now contain a linear sum
of self-force Fourier modes. The average, and thus the 0PA equations, can be computed.
These equations are
dp(0) X 
= Γϕ(1)p(k) (pj(0) ) + Γr(1)p(k) (pj(0) ) ,
dt̃ k∈Z
(2.37)
de(0) X 
= Γϕ(1)e(k) (pj(0) ) + Γr(1)e(k) (pj(0) ) ,
dt̃ k∈Z

where D E
Γϕ(1)p(k) = Fp,ϕ (pj(0) , ψr ) e−ikψr f(1)(k)
ϕ
(pj(0) ) ,
D E
Γr(1)p(k) = Fp,r (pj(0) , ψr ) e−ikψr f(1)(k)
r
(pj(0) ) ,
D E (2.38)
Γϕ(1)e(k) = Fe,ϕ (pj(0) , ψr ) e−ikψr f(1)(k)
ϕ
(pj(0) ) ,
D E
Γr(1)e(k) = Fe,r (pj(0) , ψr ) e−ikψr f(1)(k)
r
(pj(0) ) .
It turns out that these averages can be analytically computed for any value of the integer k.
We shall not display the expressions Γϕ(1)p(k) and Γϕ(1)e(k) because they are lengthy. However,
the interested reader can consult the notebook associated to this paper on Github where
the adiabatic equations are explicitly written. It is however easier to compute Γr(1)p(k) and
Γr(1)e(k) . Their analytical expressions are
(0) 3 2
Ωr 2 e(0) p(0) (3 + e(0) − p(0) )
Γr(1)p(k) = − r
K(e(0) ; k) f(1)(k) (p(0) , e(0) ),
2π 4e2(0) − (p(0) − 6)2
(0) 2 2 2
(2.39)
Ωr (3 + e(0) − p(0) )(6 + 2e(0) − p(0) )p(0)
Γr(1)e(k) =− r
K(e(0) ; k) f(1)(k) (p(0) , e(0) ),
2π 4e2(0) − (p(0) − 6)2

–9–
 R 2π sin(ψ) R 2π sin(ψ) sin(k ψ)
where K(e(0) ; k) := 0 (1+e(0) cos(ψ))2
e−ikψ dψ = −i 0 (1+e(0) cos(ψ))2
dψ can be computed
analytically and does not diverge at the separatrix for any value of k. To do so, we first
write sin(k ψ) as given by the formula [38]
   
k k
sin(k ψ) = k cosk−1 (ψ) sin(ψ) −   cosk−3 (ψ) sin3 (ψ) +   cosk−5 (ψ) sin5 (ψ) − . . .
3 5
(2.40)
Then, thanks to the change of variables z := tan( ψ2 ) and trigonometric reductions, it is
possible to write the integrand as the fraction of polynomials of z which can be explicitly
integrated. In particular, K(e(0) ; 0) = 0. Therefore, the average radial self-force does not
contribute to the driving force at adiabatic order.
The post-adiabatic equations can be derived following the Hinderer-Flanagan analysis
[3] using the fast variable ψr . However, since fr depends upon ψr , the post-adiabatic
equations are coupled between Fourier modes k and are not therefore easily solvable. We
shall not develop them here.

3 Analytic expansion towards the separatrix

In this section, we asymptotically develop the expansion of the equations (2.37) in the
approach towards the separatrix. For that purpose, we define the parameter δ at adiabatic
order, which identically vanishes at the separatrix, as

δ ≡ p(0) − (6 + 2e(0) ). (3.1)

We will study the limit δ → 0+ . In that limit, we expect the eccentricity to asymptote to
a finite value e(0) → e⋆ , which is interpreted as the eccentricity at the separatrix crossing.
We have immediately
dδ X  ϕ 
= Γ(1)δ(k) (δ, e(0) ) + Γr(1)δ(k) (δ, e(0) ) , (3.2)
dt̃ k∈Z
where
Γϕ(1)δ(k) = Γϕ(1)p(k) − 2Γϕ(1)e(k) ,
(3.3)
Γr(1)δ(k) = Γr(1)p(k) − 2Γr(1)e(k) ,
have been rewritten in terms of δ and e(0) .
µ
We now assume that the self-force f(1) as a function of the energy E and angular
momentum L is continuous and differentiable at the separatrix. The relationship between
(E, L) and (e, p) (understood as (e(0) , p(0) )) is given in Eq. (2.6). The Jacobian is singular
at the separatrix. Given the ratio dp/de at the separatrix (2.24), we have
µ e(0) − 1 µ
∂δ f(1) (δ, e(0) )|δ=0+ = ∂e(0) f(1) (δ, e(0) )|δ=0+ . (3.4)
8

– 10 –
3.1 Expansion of elliptic integrals

We now aim to find the leading order solution in the δ → 0 expansion. The major diffi-
culty that we encounter is the following. The orbit average quantities Γϕ(1)p(k) and Γϕ(1)e(k)
 
2e(2+2e+δ) 4e
depend upon complete elliptic integrals of the third kind such as Π (1+e)(4e+δ) , 4e+δ and
 
16e 4e
Π (4+δ)(4e+δ) , 4e+δ (here we substituted e(0) by e in order to simplify the notation). The
asymptotic development of these functions as δ → 0 is not documented in standard text-
books of special functions. We are however able to prove that, at leading order in δ, and
as long as 1 ≥ e > 0,
 
2e(2 + 2e + δ) 4e
Π , =
(1 + e)(4e + δ) 4e + δ
p !
1 2 2 e(e + 1)  2e(3 + 2e) 4e  p
p × Π , + (3 + 2e)(1 + 4e) Λ(e)
δ (3 + 2e)(1 + 4e) (1 + e)(1 + 4e) 1 + 4e

1 + log(64e) − log(δ) 1 + 3e
+ + √ p
4 2 2 e(1 + e)(3 + 2e)(1 + 4e)
"   #
2e(3 + 2e) 4e p
× Π , + (3 + 2e)(1 + 4e) Λ(e) + O(δ),
(1 + e)(1 + 4e) 1 + 4e
 
16e 4e
Π , =
(4 + δ)(4e + δ) 4e + δ
p 
16e 4e
 √ 
1 2 5e (1 + 4e)(5 + 4e) Π 5+20e , 1+4e + 5 e (1 + 4e)Θ(e)
√
δ 5 1 + e (1 + 4e)
1−e log(64e) − log(δ) (1 + e + e2 )Θ(e)
+ + + √
4(1 + e) 4 4 e (1 + e)3/2
 p
16e 4e
Π 5+20e , 1+4e (1 + e)(1 + 4e)(5 + 4e)  
+ √ 5e1/2 + 9e3/2 + 9e5/2 + 4e7/2 + O(δ),
4 5 e(1 + e)2 (1 + 4e)(5 + 4e)
(3.5)
where we have defined the functions
 
4e
Z 1 (1 + e) K 4e+x
Λ(e) ≡ − √ dx,
0 (2 + 2e + x)3/2 4e + x
    (3.6)
4e 4e
Z 1 (4e + x)E 4e+x − 2(2 + 2e + x)K 4e+x
Θ(e) ≡ p dx.
0 2 (4e + x)(4 + x)(4 + 4e + x)

These are pure functions of the eccentricity which are well defined throughout the entire
range 0 ≤ e ≤ 1 and admit finite e = 0 and e = 1 values. The zero eccentricity values are

– 11 –
Λ(0) = − 2√π 3 and Θ(0) = − π2 . The e = 1 values are
     
1 3 3 4 4 1 4
Λ(1) = π 2 F1 , ; 2; − 1 = −1 + E − K ≈ 0.2730,
25 2 2 5 5 5 5
 
Z 1 E 4 − 2K 4
  (3.7)
4+x 4+x
Θ(1) = √ dx ≈ −0.7315,
0 2 8+x
where 2 F1 (a, b, c, z) is the Gaussian hypergeometric function (see [39] for elements on
hypergeometric functions). We note that the two expansions truncated at subleading order
(3.5) become better fits of the exact functions at high values of the eccentricity. We detail
in Appendix B the derivation of these series expansion as well as further relevant identities
and plots.
Once the elliptic integrals are replaced by Eq. (3.5), we can proceed to the δ-expansion
of the adiabatic equations and derive their behaviour as the inspiral approaches the sepa-
ratrix. This is the content of the next section.

3.2 Asymptotic adiabatic equations

Let us now present the resulting asymptotic expansion of the adiabatic equations in the
limit δ → 0+ . We could use δ as a substitute for the slow time t̃. In other words, we could
define the functions t̃ = t̃(δ) and e(0) (δ) = e(0) (t̃(δ)). Given Eq. (2.24), we have at the
separatrix
de(0) (t̃)
dt̃
de(0) (δ) e⋆ − 1
= = , (3.8)
dδ(t̃)
δ=0 dδ δ=0 8
dt̃
where e⋆ is the value of the eccentricity at the separatrix. This tells us that the eccentricity
e⋆ −1
e(0) viewed as a function of δ has e(0) (0) = e⋆ and e′(0) (0) = 8 . With this at hand, we
are ready to derive the expansions of the 0PA equations.

Fast angle variables We first expand the evolution of the angle variables given in Eq.
(2.30). This amounts to compute the asymptotic behavior of the Boyer-Lindquist geodesic
(0) (0)
frequencies Ωr and Ωϕ . At leading order, we find
(0)
dφr (e⋆ (1 + e⋆ ))3/2 π 1
Ω(0)
r = =− 2
+ o(δ 0 ),
dt̃ 2(3 + e⋆ ) e⋆ log(δ) + B(e⋆ )
(0)
(3.9)
dφϕ 1 + e⋆ 3/2 log(δ) − log(64e⋆ )
 
(0) e⋆
Ωϕ = = √ + o(δ 0 ).
dt̃ 2 2 3 + e⋆ e⋆ log(δ) + B(e⋆ )
The function B(e) is real in the range 0 ≤ e < 1 and admits two roots at e = 0 and
e ≈ 0.01433. It is positive in the range 0 < e < 0.01433 and negative otherwise. In the
low eccentricity limit we have B(e) = −e log(64e) + O(e2 ). In the high eccentricity limit,

– 12 –
we have that B(e) = − √π2 (1 − e)−3/2 + O((1 − e)−1/2 ). The complete expression and plots
of B can be found in Appendix C. Given the denominator e⋆ log(δ) + B(e⋆ ), the domain of
validity of this expansion is 0 < δ < e−B(e⋆ )/e⋆ , which reduces to 0 < δ < 64e⋆ in the small
eccentricity limit. The expansion therefore is not valid in the strict zero eccentricity limit1 .
The next-to-leading order of (3.9) is computed and can be found in the Mathematica
notebook of this paper.
At finite e⋆ , the limit δ → 0+ of the equations gives
 3/2
(0) 1 1 + e⋆
Ω(0)
r⋆ =0 , Ωϕ⋆ = √ . (3.10)
2 2 3 + e⋆
(0) 1
These latter equations have a zero eccentricity limit where Ωϕ⋆ = √
6 6
, as the quasi-circular
case.
(0) (0)
Given the known asymptotic behaviour of the fundamental frequencies Ωr and Ωϕ we
can derive the instantaneous number of whirls the secondary body achieves during the
late inspiral. These cycles, first discussed in [40], are mostly occurring as a whirl motion
around the peripasis, where the secondary body is the closest to the primary black hole.
This instantaneous number of whirling cycles is given by
(0)
Ωϕ
r
ϕ(ψr = 2π) − ϕ(ψr = 0) 3 + e⋆ log(64e⋆ ) − log(δ)
Nw := = (0) = √ + o(δ 0 ). (3.11)
2π Ωr e⋆ 2π

This expression explicitly shows that the instantaneous number of whirls increases in a
logarithmic manner as we come closer and closer to the end of the inspiral. This confirms
the computation produced in [40]. The graphic representation of Nw is given in Fig. 1.

Slow orbital variables The equations that drive the evolution of δ and e(0) read

dδ A(e⋆ )
= + O(δ 0 ), (3.12a)
dt̃ δ (e⋆ log(δ) + B(e⋆ ))
de(0) e⋆ − 1 A(e⋆ )
= + O(δ 0 ). (3.12b)
dt̃ 8 δ (e⋆ log(δ) + B(e⋆ ))
 
P r ϕ
The constant A can be decomposed as A = k∈Z Ar(k) (e⋆ )f(1)(k)⋆ + Aϕ(k) (e⋆ )f(1)(k)⋆ .
We checked explicitly this structure for the Fourier modes ranging from k = −2 to k = 2
and we expect that it holds for any k ∈ Z. It turns out that Ar(0) = 0, Ar(k) = −Ar(−k)
and Aϕ(k) = Aϕ(−k) . The explicit expressions of Ar(k) , Aϕ(0) , Aϕ(±1) and Aϕ(±2) are given

1
The small eccentricity limit could be studied instead by performing a post-adiabatic expansion of the
low eccentricity expansion of the osculating equations formulated in the appropriate (α, β) formulation [6].
Such equations, which could be compared with the quasi-circular case, are not studied here.

– 13 –
Figure 1. Plots of Nw as a function of δ. The left panel shows the case where the eccentricity
is set to e = 0.3, while the right panel corresponds to e = 0.9. These results can be directly
compared with [40], where a similar plot is presented. As expected, the instantaneous number of
whirls, Nw , increases sharply as the system approaches the separatrix, diverging to infinity at δ = 0.
Specifically, for δ = 0.01, we find Nw (0.01) = 5.645 for e = 0.3 and Nw (0.01) = 4.058 for e = 0.9.

in Appendix C. In these equations, the B(e⋆ ) factor arises from the asymptotic behavior of
Ωr . The A factor arises from the asymptotic expansion of Fp −2Fe and Fe . The subleading
orders of Eq. (3.12a) and (3.12b) are explicitly given in Appendix D. The relations (3.12a)
and (3.12b) approximate well the 0PA equations for small δ; the plots in Fig. 2 compare
the approximations with the analytical formulae for δ < 0.4.

Rewriting of the fast angle variables The differential equations for the fast variables
(0)
dφα
(3.9) can be written as dδ as well. Thanks to (3.12a), we have
(0)
dφr (e⋆ (1 + e⋆ ))3/2 π
=− δ + o(δ),
dδ 2 (3 + e⋆ )2 A(e⋆ )
(0)
(3.13)
dφϕ 1 + e⋆ 3/2 e⋆ (log(64) + log(e⋆ ) − log(δ)) δ
 
1
=− √ + o(δ).
dδ 2 2 3 + e⋆ A(e⋆ )

We can also develop an adiabatic equation involving δ and the anomalous phase ψr while
considering ψr (δ). For that purpose, we consider Eq. (3.12a) and the 0PA term of Eq.
(2.23c). This gives
√
dψr e⋆ (1 + e⋆ cos(ψr ))2 (−2 − e⋆ + e⋆ cos(ψr ))(e⋆ log(δ) + B(e⋆ )) sin( ψ2r )
= −δ √ + o (δ) .
dδ 4 ε 1 + e⋆ (3 + e⋆ )2 A(e⋆ )
(3.14)

– 14 –
Figure 2. Comparison of the complete 0PA equations (solid lines) with their asymptotic expan-
dδ de(0)
sions (dashed lines). The quantities dt̃ 0PA
(left panel) and dt̃ 0PA
(right panel) show a strong
agreement with their respective asymptotic expansions for δ < 0.4. In generating these plots, the
eccentricity e(0) was set to 0.2673, with self-force data at the separatrix for this value of e(0) kindly
provided by Maarten van de Meent. For δ = 0.4, the relative mismatches between the numerical
de(0) dδ
and asymptotic values are approximately 3% for dt̃
and 0.3% for dt̃
.

3.3 Asymptotic analytical solutions

It is possible to solve the asymptotic equations analytically thanks, notably, to the real
branch k = −1 of the Lambert W function W−1 (·).
First of all, we focus on the solution of Eq. (3.12a) for δ(t̃). The solution is
2B(e⋆ )
 
1
W−1 − 4A e e⋆ −1 (t̃⋆ −t̃)
1
− B(e ⋆) 2 e⋆
δ(t̃) = e 2 e⋆ e , (3.15)

and is valid for t̃ < t̃⋆ , where t̃⋆ is the separatrix crossing time. In this solution, W−1 (z)
is the −1 type real branch of the W Lambert function defined as being the reciprocal of
z = W (z) eW (z) . Elements on this type of function can be found in the Appendix E. The
function δ(t̃) is monotonously decreasing with slow time and reaches 0 at t̃ = t̃⋆ . We can

– 15 –
invert Eq. (3.15) and obtain
  
e⋆ 2B(e⋆ )
t̃(δ) = t̃⋆ + δ 2 − 1 + log(δ)
4A(e⋆ ) e⋆
8 9Eδ (e⋆ ) − 3Fδ (e⋆ ) + 2Gδ (e⋆ ) − 3(e⋆ − 1) (e⋆ − 3B(e⋆ ))A′ (e⋆ ) + A(e⋆ )(−1 + 3B ′ (e⋆ ))



3
+δ −
216A2 (e⋆ )
−24Fδ (e⋆ ) + 16Gδ (e⋆ ) + 3(e⋆ − 1) A(e⋆ ) − e⋆ A′ (e⋆ )


Gδ (e⋆ ) 2
+ log(δ) − log (δ)
72A2 (e⋆ ) 3 A2 (e⋆ )
+ O(δ 4 log3 δ). (3.16)

Here we used the subleading orders of the equations derived in Appendix D to compute
the residual of the leading order solution.
The solution of Eq. (D.2) can be found. In terms of δ the eccentricity is given by
e⋆ − 1 8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ ) 2
e(0) (δ) = e⋆ + δ+ δ log(δ) + O(δ 2 ). (3.17)
8 16 e⋆ A(e⋆ )
This relation tells us that the eccentricity increases as the secondary black hole approaches
the separatrix. Such behavior in the strong-field regime is consistent with expectations, as
it his described in [14, 17, 41].
After analysis, we conjecture that the coefficient appearing in (3.17) can be written in
terms of the self-force evaluated at the periapsis,
√
2 2 e⋆ (3 + e⋆ )3/2 (3 − e⋆ )2 ϕ
8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ ) = √ f(1) (p⋆ , e⋆ , 0). (3.18)
1 + e⋆
We proved that this identity holds when including all Fourier modes k = −2, −1, 0, 1, 2 and
we conjecture that it holds more generally.
Finally, the solutions of the leading order differential equations for the angle variable
(3.13) read as

(e⋆ (e⋆ + 1))3/2 π 2

φ(0)
r (δ) = φr⋆ − δ
4(3 + e⋆ )2 A(e⋆ )
√ 
3/2 5/2 7/2

1 + e⋆ 3 e⋆ + 4e⋆ + e⋆ πGδ (e⋆ )
+ δ 3 log(δ) + O(δ 3 ),
6e⋆ (3 + e⋆ )3 A2 (e⋆ )
(3.19)
1 + e⋆ 3/2 e⋆ 1 + log 4096 e2⋆ − 2 log(δ) 2
 


(0) 1
φϕ (δ) = φϕ⋆ − √ δ
2 2 3 + e⋆ 4 A(e⋆ )
1 + e⋆ 3/2
 
G (e )
− √δ ⋆ δ 3 log2 (δ) + O(δ 3 log(δ)),
3 + e⋆ 2
6 2 A (e⋆ )
where φϕ⋆ and φr⋆ are the angle variables at the separatrix.
Furthermore, we can expand δ(t̃) and e(0) (δ(t̃)) into Eq. (2.23c). No pole arises in this
expansion, and the leading term reduces to Eq. (2.23c) with e(0) replaced by e⋆ and δ set

– 16 –
to 0. Then, we can directly integrate and obtain

ψr = ψr⋆ + fr (6 + 2e⋆ , e∗ , ψr⋆ )(t − t⋆ ) + O(t − t⋆ )2 , (3.20)

where ψr⋆ is the value of ψr at the separatrix. This result is the first order Taylor devel-
opment of the anomalous phase around its value at the separatrix. We can also express ψr
as a function of δ after using Eq. (3.16).

3.4 Consequences

Up to this point, we have derived the adiabatic equations of motion near the last stable
orbit and obtained their corresponding solutions. This information improves our under-
standing of the dynamics. For example, it allows us to determine how the energy and
angular momentum evolve before crossing the separatrix, but also how the periapsis and
the apoapsis behave asymptotically. Ultimately, we have all the tools required to compute
the late evolution of the radius and the dynamics of the late inspiral in the (E, L) and
(p, e) phase spaces.

Energy and momentum The (adimensionalized) energy and angular momentum are
functions of the orbital elements e and δ as given in Eq. (2.6). In the approach to the
separatrix, E(δ) and L(δ) behave as

8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ )

E(δ) = E⋆ (e⋆ ) − √ p δ 2 log(δ) + O(δ 2 ), (3.21)
4 2 (−3 + e⋆ )(3 + e⋆ ) 9 − e2⋆ A(e⋆ )
8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ ) 2
L(δ) = L⋆ (e⋆ ) +
3/2 δ log(δ) + O(δ 2 ), (3.22)
2 3 + 2e⋆ − e⋆ 2 A(e⋆ )

where the E⋆ and L⋆ (their value at the separatrix) are given by Eq. (2.7) and (2.8)
with e replaced by e⋆ . The quantities Ge (e⋆ ) and Gδ (e⋆ ) appear at subleading order in
the expansion close to the separatrix. They are defined in Appendix D. We deduce in
particular that
√
dE (−3 + e⋆ )e⋆ 1 + e⋆ A(e⋆ )
= Ωϕ⋆ + √ √
+ O(δ). (3.23)
dL 16 2 3 + e⋆ 8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ ) log(δ)

Recall that Ωϕ⋆ is given by the second equality in Eq. (3.10). This relation can be seen as
a generalization of what is known for the quasicircular case; in the late inspiral, the rates of
change of E and L are directly given by the value of the azimuthal fundamental frequency
dE 1 dL
at the separatrix. In particular, the limit e⋆ → 0 provides dδ = √
6 6 dδ
+ O(δ).

– 17 –
Critical function As defined in [41], the so-called critical function is a function that
compares the evolution of the semi-latus rectum to that of the eccentricity. It is defined by
d log(e)
C(p, e) := . (3.24)
d log(p)
The computation of this function is immediate. We obtain


1 − e⋆ 4 8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ )
C(δ, e) = − + δ log(δ) + O(δ), (3.25)
e⋆ e2⋆ (3 + e⋆ )A(e⋆ )
where the higher order in δ is available in the notebook. This result reproduces what has
been computed in [41]. In this paper, this computation was used to highlight the fact
that, as 0 < e⋆ < 1, the critical function remains negative regardless of the value of the
eccentricity at the separatrix. This implies that, as long as the semi-latus rectum decreases,
the eccentricity increases at the late inspiral near the separatrix. This is a characteristic
behavior of the strong-field regime for eccentric orbits, and we have demonstrated this
more explicitly through Eq. (3.17).
Using the conjecture (3.18), the formula (3.25) can be rewritten as
√ 2
√ ϕ
1 − e⋆ 8 2 (−3 + e⋆ ) 3 + e⋆ f(1) (p⋆ , e⋆ , 0)
C(δ, e) = − + √ δ log(δ), (3.26)
e⋆ e⋆ 1 + e⋆ A(e⋆ )
in terms of the self-force evaluated at the periapsis.
p
Apoapsis, periapsis and radius The apoapsis ra := 1−e slowly decreases as we ap-
proach the separatrix. More precisely, the solution (3.17) provides


(6 + 2e⋆ ) 8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ ) 2
ra (δ) = + δ log(δ) + O(δ 2 ). (3.27)
1 − e⋆ 2(e⋆ − 1)2 e⋆ A(e⋆ )
p
This can be compared with the periapsis rp := which instead behaves as
1+e


(6 + 2e⋆ ) (3 + e⋆ ) 8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ ) 2
rp (δ) = + δ− δ log(δ) + O(δ 2 ). (3.28)
1 + e⋆ 2(1 + e⋆ )2 4e⋆ (1 + e⋆ )2 A(e⋆ )
The distance between the two roots of the radial potential grows during the late inspiral.
This explains the increase in eccentricity in the approach to the separatrix.
The late-time evolution of the radius combines two distinct motions: fast oscillations
between the two roots ra and rp and slow changes due to the gradual decrease of δ and
increase of e. The formulae (3.17), (3.15) and (3.20) together with Eq. (3.16) allow to
compute r(δ) which admits the expansion
p⋆ (3 + e⋆ )(1 + cos(ψr⋆ )) 2
r(δ) = ⋆
+
2 δ + O(δ log(δ)). (3.29)
1 + e⋆ cos ψr 4 1 + e⋆ cos(ψr⋆ )
A plot of r(δ) is shown in Fig. 3.

– 18 –
Figure 3. Evolution of the radius (solid line) as a function of δ. The anomalous phase at the
separatrix has been chosen to be ψr⋆ = 2.4 and the final eccentricity is e⋆ = 0.2672. The dashed
colored lines indicate the instantaneous locations of the periapsis and apoapsis, which evolve at
orders O(δ) and O(δ 2 ), respectively. Here, the mass ratio is ε = 10−4 .

Redshift Expanding Eq. (2.2) we obtain

 
ψr⋆
dt (3 + e⋆ ) cos2 2 2
= γ⋆ − √ p
2 δ + O(δ log(δ)), (3.30)
dτ 2
2 9 − e⋆ 2 + e⋆ − e⋆ cos(ψr⋆ )
√
E⋆ 2 2 (3+e⋆ )
where γ⋆ := 1− 2M
= √
is the redshift at the separatrix. Using the
r⋆ 9−e2⋆ 2+e⋆ −e⋆ cos(ψr⋆ )
explicit solution (3.15) we obtain
5/2 p p
γ⋆ (1 + cos(ψr⋆ ))(2L1 + log(−L1 )) A(e⋆ ) e⋆ (3 − e⋆ )
t = t⋆ − γ⋆ (τ⋆ − τ ) + √ √ (τ⋆ − τ )3/2 ,
24 2 e⋆ 3 + e⋆ (−L1 )3/2
(3.31)
where we have defined
 
−1+ 2B(e ⋆)
4e e ⋆ A(e⋆ )
L1 := log  γ⋆ (τ⋆ − τ ) . (3.32)
e⋆

– 19 –
Energy and angular momentum Combining this result with Eqs. (3.21), (3.22) and
(3.16) allows to write the energy and angular momentum as a function of τ

E(τ ) = E⋆ (e⋆ ) + (τ⋆ − τ ) (3.33)

(
8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ )
× √ p
2 2 (e⋆ − 3)e⋆ (3 + e⋆ ) 9 − e2⋆ γ⋆
1
+ √ p
8 2 (e⋆ − 3)e2⋆ (1 + e⋆ )(3 + e⋆ ) 9 − e2⋆ γ⋆ log(τ⋆ − τ )
"
× e⋆ 3 − 6e⋆ + e2⋆ + 2e3⋆ A(e⋆ ) − 8(1 + e⋆ )

#)


− 8e⋆ Fe (e⋆ ) + (e⋆ − 1)e⋆ Fδ (e⋆ ) + 2B(e⋆ ) 8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ )
 
(τ⋆ − τ )
+O ,
log2 (τ⋆ − τ )
L(τ ) = L⋆ (e⋆ ) + (τ⋆ − τ ) (3.34)
(
8Ge (e⋆ ) + (1 − e⋆ )Gδ (e⋆ ) −1
× −
3/2 +
3/2
e⋆ 3 + 2e⋆ − e2⋆ γ⋆ 4 e2⋆ 3 + 2e⋆ − e2⋆ γ⋆ log(τ⋆ − τ )
"
h 
× e⋆ (3 + e2⋆ )A(e⋆ ) − 8 − 8e⋆ Fe (e⋆ ) + (e⋆ − 1)e⋆ Fδ (e⋆ )
#)  

i (τ − τ⋆ )
+ 2B(e⋆ ) 8Ge (e⋆ ) + Gδ (e⋆ ) − e⋆ Gδ (e⋆ ) +O .
log2 (τ⋆ − τ )

Motion in phase space The motions in the (e, p) and (E, L) phase spaces display a
strong contrast. In the (e, p) phase space, Eq. (3.25) implies that the approach to the
separatrix is along a line that crosses the separatrix line p⋆ = 6 + 2e⋆ . Instead in the (E, L)
phase space, Eq. (3.23) implies that the approach to the separatrix becomes asymptotically
tangential to the the separatrix line (E⋆ (e⋆ ), L⋆ (e⋆ )) given by Eqs. (2.7)-(2.8). Such a
distinct behavior is clearly displayed on Figure 4.

– 20 –
Figure 4. Late inspiral in phase space. On the left panel, the motion in the (L, E) phase space
(dashed line) becomes increasingly tangent to the separatrix (solid line), eventually gazing it. On
the right panel, the same dynamics is shown but in the (p, e) phase space. There, the secondary
approaches the separatrix (solid line) head-on.

– 21 –
4 Conclusion and outlook

We derived analytically the solution to the adiabatic equations for the forced geodesic
motion with eccentricity around the Schwarzschild black hole in the approach to the sep-
aratrix, in terms of self-force coefficients evaluated at the separatrix. We found that the
explicit solution is expressed in terms of Lambert W−1 function, which appears as a key
mathematical feature of this dynamical system.
It is known since the work [42] that the evolution in the phase space (p, e) is such that
the separatrix line p = 6 + 2e is crossed head on. Here, we obtained analytically, after
solving for the motion at subleading order in the deviation from the separatrix, that the
evolution in the phase space (E, L) is such that the separatrix line is crossed asymptotically
tangentially, see Figure 4. Since the energy and angular momentum are well-defined beyond
the separatrix, this suggests to formulate a transition-to-plunge regime where (E, L) are
defined as perturbations of their values at the separatrix crossing. Our result in terms of
the proper time expansion is
!
log
(0) E⋆
E = E⋆ + ε Ωϕ⋆ κ⋆ (τ⋆ − τ ) 1 + + O(log−2 (τ⋆ − τ )) , (4.1a)
log(τ⋆ − τ )
!
Llog
⋆
L = L⋆ + εκ⋆ (τ⋆ − τ ) 1 + + O(log−2 (τ⋆ − τ )) , (4.1b)
log(τ⋆ − τ )
(0)
where Ωϕ⋆ is given in Eq. (3.10) and κ⋆ = κ⋆ (e⋆ , ψr⋆ ) depends upon the relativistic anomaly
ψr⋆ at separatrix crossing through its dependency in the redshift, see Eq. (3.33). If we
admit our conjecture (3.18), the leading coefficient κ⋆ is linear in the ϕ component of the
self-force at the periapsis. The expansion (4.1) admits a circular limit which is compatible
with the results derived in [29]. This suggests to formulate a transition-to-plunge regime
based on the critical analytical solutions around the separatrix values of the energy and
angular momentum [43, 44]2 . The conjecture (3.18) points to the role of the periapsis for
the transition-to-plunge regime at an arbitrary relativistic anomaly ψr⋆ at the separatrix.
For a generic eccentric orbit, the transition-to-plunge regime will depend upon the
eccentricity e⋆ and the relativistic anomaly ψr⋆ at separatrix crossing, which looses its
intuitive meaning after crossing since the periapsis ceases to exist. Contrary to the quasi-
circular case, the radius evolves both under the orbital and the radiation reaction timescales.
A formulation using angle-action variables is therefore necessary in order to formulate a
multiscale expansion scheme. We foresee that the recently obtained analytical maps be-
tween quasi-Keplerian angles and Mino time action angles [36] will be instrumental for that
2
Note that critical solutions also exist in the scattering range E > 1 [45].

– 22 –
purpose. Any transition-to-plunge expansion scheme should reproduce at early times the
late time evolution of the inspiral phase that we derived. We leave the development of a
transition-to-plunge expansion scheme to further work.

Acknowledgments

We would like to thank Leor Barack and Maarten van de Meent for interesting discussions
and Maarten van de Meent for providing self-force data. G.C. is Research Director of the
F.R.S.-FNRS.

A Osculating equations

The osculating equations provide the equations of motion of an accelerated point particle
by considering self-force effects. Following the notation introduced in Sec. 2.4, the func-
tions that define the osculating equations for a particle orbiting around a Schwarzschild
background in the equatorial plane read

Fp (p, e, ψr ) = Fp,ϕ (p, e, ψr )f ϕ (p, e, ψr ) + Fp,r (p, e, ψr )f r (p, e, ψr ), (A.1)

Fe (p, e, ψr ) = Fe,ϕ (p, e, ψr )f ϕ (p, e, ψr ) + Fe,r (p, e, ψr )f r (p, e, ψr ), (A.2)
p
(e cos(ψr ) + 1)2 −2e cos(ψr ) + p − 6 (−2e cos(ψr ) + p − 2)
fr (p, e, ψr ) = p , (A.3)
p2 (p − 2)2 − 4e2
δfr (p, e, ψr ) = δfr,ϕ (p, e, ψr )f ϕ (p, e, ψr ) + δfr,r (p, e, ψr )f r (p, e, ψr ), (A.4)
p(e cos(ψr ) + 1)2 (−2e cos(ψr ) + p − 2)
fϕ (p, e, ψr ) = p , (A.5)
p5 ((p − 2)2 − 4e2 )

where
p3/2 −3 − e2 + p (−2 + p − 2e cos(ψr ))


Fp,ϕ (p, e, ψr ) = p
(−4e2 + (−6 + p)2 ) −4e2 + (−2 + p)2 (1 + e cos(ψr ))2
h
× 36 + 6e2 − 18p − e2 p + 2p2 + e 12 + 3e2 − 4p cos(ψr )


i
− e2 (−6 + p) cos(2ψr ) + e3 cos(3ψr ) , (A.6)

p 
2e 3 + e2 − p p −6 + p − 2e cos(ψr ) (2 − p + 2e cos(ψr )) sin(ψr )
Fp,r (p, e, ψr ) = − p ,
(−4e2 + (−6 + p)2 ) −4e2 + (−2 + p)2
(A.7)

– 23 –
and
p −e2 + p − 3 (−2e cos(ψr ) + p − 2)


Fe,ϕ (p, e, ψr ) = p
2 ((p − 6)2 − 4e2 ) p ((p − 2)2 − 4e2 ) (e cos(ψr ) + 1)2
h
× e e 2e2 − p + 6 cos(3ψr ) − (p − 6) 2e2 − p + 6 cos(2ψr )

− 2e2 p + 20e2 + 3p2 − 32p + 60

i
+ 6e4 + e2 (42 − 11p) + 4 p2 − 9p + 18 cos(ψr ) ,



(A.8)
p
e2 − p + 3 2e2 − p + 6 sin(ψr ) −2e cos(ψr ) + p − 6 (−2e cos(ψr ) + p − 2)



Fe,r (p, e, ψr ) = p .
((p − 6)2 − 4e2 ) (p − 2)2 − 4e2
(A.9)

Moreover
p −e2 + p − 3 (−2e cos(ψr ) + p − 2)


δfr,ϕ = p
e (4e2 − (p − 6)2 ) p ((p − 2)2 − 4e2 ) (e cos(ψr ) + 1)2
× sin(ψr ) −e 4e2 − (p − 6)2 cos(ψr ) − (p − 6) e2 cos(2ψr ) + e2 − 2p + 6 ,




(A.10)
p
e2 − p + 3


−2e cos(ψr ) + p − 6 (2e cos(ψr ) − p + 2)(2e + (p − 6) cos(ψr ))
δfr,r =− p .
e (4e2 − (p − 6)2 ) (p − 2)2 − 4e2
(A.11)

B Limits of elliptic integrals

In this appendix, we provide a summary of the asymptotic limits of elliptic integrals that are
required to analytically describe the approach to the separatrix. The method of derivation
of the asymptotic expressions of these functions given in Eq. (3.19) is described.

B.1 Conventions

In this work, several elliptic integrals are used. First, using the conventions of Mathematica,
the incomplete integrals of the first and second kind are respectively
Z ϕ
dx
F (ϕ, m) := p , (B.1)
0 1 − m sin2 (x)

and Z ϕq
E(ϕ, m) := 1 − m sin2 (x) dx. (B.2)
0
In particular, we define the complete integrals of the second kind by
π  π 
K(m) := F ,m , E(m) := E ,m . (B.3)
2 2

– 24 –
The complete integrals of the third kind Π(n|m) have an integral representation as
Z π
2 dx
Π(n, m) := p . (B.4)
0 (1 − n sin2 (x)) 1 − m sin2 (x)

B.2 Reformulation of some elliptic functions

In this section, we focus on the asymptotic behaviour of Π(n|m) in the special case for
which n → 1 and m → 1 simultaneously. In textbooks of mathematical functions, the
asymptotic series of this function are only known when either m ̸= 1 is fixed and n 7→ 1,
or, when n ̸= 1 is fixed and m 7→ 1. Here, we take advantage of the fact that n and m are
functions of δ and are equal to 1 when δ = 0. This means, that we should be able to express
Π(n(δ)|m(δ)) as a pure function of δ (that we call Π̄(δ)) whose asymptotic expansion for
δ → 0 is feasible.
For that purpose, recall that Π(n|m) is defined as being the solution of the two partial
differential equations

∂2Π ∂3Π
 
∂Π 1
= 3Π + 4 (n(2 − 6m) + m(11m − 7))) − 8 m(n − m)(m − 1) ,
∂m 6(2 + n − 7m) ∂m2 ∂m3
∂2Π ∂3Π
 
∂Π 1 2
= −2 Π + (−13n − 3m + 8n(m + 1)) − 2 n (n − 1) (n − m) .
∂n 16n − 4(m + 1) ∂n2 ∂n3
(B.5)
Considering that the function Π̄ depends on δ via m(δ) and n(δ), we can compute its
derivative with respect to δ. Thanks to the chain rule

dΠ̄ ∂Π dm ∂Π dn
= + , (B.6)
dδ ∂m dδ ∂n dδ
∂Π dΠ̄ ∂Π
we can express ∂n as a function of dδ and ∂m which is function of its higher derivatives
via (B.5). These latter can be replaced by the analytical formulae for the derivatives of Π
with respect to m and n. This involves several elliptic K, E and Π functions.
By doing so, we conclude that Π̄(δ) is the solution of a first order differential equation

0 = 2n(δ)3 (−1 + m(δ)) Π̄′ (δ) + K(m(δ)) − Π̄(δ) (−1 + m(δ)) m(δ) n′ (δ)

 
2 2 ′ ′ ′ ′


− n(δ) 2 (−1 + m(δ) ) Π̄ (δ) − Π̄(δ) (−1 + m(δ)) n (δ) − m (δ) + E(m(δ)) m (δ)
 
+ n(δ) 2 m(δ)2 Π̄′ (δ) + (K(m(δ)) − E(m(δ))) n′ (δ) + E(m(δ)) − Π̄(δ) m′ (δ)


 
+ m(δ) −2 Π̄′ (δ) + (E(m(δ)) − K(m(δ))) n′ (δ) + Π̄(δ) m′ (δ) ,


(B.7)
where a prime over a function denotes the derivatives with respect to δ.

– 25 –
 The solution can be found up to a constant that can be fixed by imposing the condition
Π̄(1) = Π(n(1), m(1)) for any eccentricity.
The results for the elliptic integrals of interest are
  p
2e(2 + 2e + δ) 4e (2 + 2e + δ)(4e + δ)
Π , = √
(1 + e)(4e + δ) 4e + δ δ 8e2 + 14e + 3
   p 
2e(3 + 2e) 4e 2
× Π , + 8e + 14e + 3 Λ(δ; e) ,
(1 + e)(1 + 4e) 1 + 4e
  r
16e 4e 1 16e
Π , = δ+
(4 + δ)(4e + δ) 4e + δ 5δ 4 + 4e + δ
r   !
20 16e 4e
× 5+ Π , + 5Θ(δ; e) ,
1 + 4e 5 + 20e 1 + 4e
(B.8)
where
 
4e
Z δ (1 + e)K 4e+x
Λ(δ; e) = λ(x; e) dx, λ(x; e) := √ ,
1 (2 + 2e + x)3/2 4e + x
   
4e 4e
Z δ −(4e + x)E 4e+x + 2(2 + 2e + x)K 4e+x
Θ(δ; e) = θ(x; e) dx, θ(x; e) := p .
1 2 (4e + x)(4 + x)(4 + 4e + x)
(B.9)

B.3 Asymptotic expansions

The expansion of (B.8) when δ → 0 can be computed at any order, as long as we use
another trick at the level of the functions Λ(δ; e) and Θ(δ; e). This is necessary since the
functions λ and θ diverge at x = 0, but can still be integrated from 1 to 0 (this is a property
we find, for instance, for the function log(x)).
Let us call ξ(x) a generic function of x which diverges as x → 0 but whose integral on
the domain [0 ; 1] exists in this limit. The derivatives of this function diverge as well, but
xn ξ (n) admits a well defined limit as x → 0 for any integer n > 0, where ξ (n) is the nth
dn n+1 ξ (n+1) is well defined when x → 0.


derivative of ξ. Also, ξ must be such that dx n x

The functions xn ξ (n) (x) can be integrated on the range [0; 1] too. These properties are the
same than those carried by λ(x; e) and θ(x; e).
The function ξ(x) and its derivative admit asymptotic expansions around x = 0 which can
be reduced to
ξ(x) = ζ0 + ζ1 log(x) + [ζ2 + ζ3 log(x)] x + O(x2 log x),
(B.10)
ζ̂−1 h i
ξ ′ (x) = + ζ̂0 + ζ̂1 log(x) + ζ̂2 + ζ̂3 log(x) x + O(x2 log x).
x
Rδ
We want to develop the expansion of I(δ) := 1 ξ(x) dx for δ → 0 up to any order in δ. For

– 26 –
that purpose, integrating by part m times allows to compute the order δ m . For example,
if we do not integrate by part, we obtain the leading order of I via

I(δ) = I1 + O(δ), (B.11)

where Z 0
I1 = ξ(x) dx. (B.12)
1
Rδ d
[x ξ(x)] − x ξ ′ (x) dx. This


The next order of I is obtained by writing it as I(δ) = 1 dx
leads us to the expansion
I(δ) = I1 + I2 (δ) δ + O(δ 2 ), (B.13)

where I2 (δ) = ζ0 − ζ̂−1 + ζ1 log(δ).

The next order of I can be obtained as well. It is based on the identity

1 d2  2  d x2 ′′
ξ(x) = − x ξ(x) + [2x ξ(x)] + ξ (x). (B.14)
2 dx2 dx 2
With this in mind, we can calculate I and obtain

I(δ) = I1 + I2 (δ) δ + I3 (δ) δ 2 + O(δ 3 ), (B.15)

   h 2 i
where I3 (δ) = (ζ2 + ζ3 log(δ))− 21 d x
ζ̂0 + ζ̂1 log(δ) +limx→0 dx ξ ′′ (x) . Similarly, and
4
using identities such as (B.14) but with higher derivatives, we can manage to develop the
higher order terms of the function I(δ).
Applying this strategy to (B.9) allows us to write

1 + log(64 e) − log(δ)
Λ(δ; e) = Λ(e) + √ p δ + O(δ 2 ),
8 2 e(1 + e)
(B.16)
1 − e + (1 + e) (log(64 e) − log(δ))
Θ(δ; e) = Θ(e) + p δ + O(δ 2 ),
8 e(1 + e)

where Z 0
Λ(e) = λ(x; e) dx,
1
Z 0
(B.17)
Θ(e) = θ(x; e) dx.
1
Plugging this latter series expansions into eq. (B.8), we reproduce the expansions (3.5).

– 27 –
C Coefficients B and A

The coefficient B(e) reads as

√ √
3/2
2 2 e3/2 1 + e 11 − 3e2


e(1 + e)
B(e) = Λ(e) + Θ(e)
(3 + e)2 (−1 + e2 ) (3 + e)2
e
+ p
5(3 + e) (1 + e)(3 + 2e)(1 + 4e) (−1 + e2 )
2
(
√ √
 
2e(3 + 2e) 4e
× 110 2 e (1 + e)Π ,
(1 + e)(1 + 4e) 1 + 4e
(C.1)
√ p   16e 

3 4 4e
+ 5 e(3 + 2e)(5 + 4e) − 1 − 2e + 2e + e Π ,
5 + 20e 1 + 4e
h √ 
2e(3 + 2e) 4e

5/2
− 5(1 + e) 6 2 e Π ,
1 + 5e + 4e2 1 + 4e
)
p
i
+ (3 + e) (1 + e)(3 + 2e)(1 + 4e) − 4e + (−1 + e)(3 + e) log(64e) .

The function is plotted in Fig. C.

Figure 5. Left: Coefficient B(e) as a function of e. Its roots are at e = 0 and e ≈ 0.01433. We
have lime7→1 B(e) = −∞. This coefficient is positive for small non-zero eccentricities.

The term of coefficient A proportional to the modes of the self-force, from modes

– 28 –
k = −2 to k = 2, read
√ p "
4(−3 + e) e (1 + e)(3 + e) √ √
2e 1 − e 51 + 2e + 19e2 − 12e3


Aϕ(0) (e) = 7/2
3(1 − e)
#
√
 r
2 2

e
− 3(1 + e) − 7 − 7e + 4e arcsin 2 ,
1+e
"
4(−3 + e) (1 + e)(3 + e) √ √
p
2 3


Aϕ(1) (e) = √ 2e 1 − e 3 − 20e − 73e + 30e
3 e (1 − e)7/2
#
√
 r
e
+ 3(1 + e)2 −1 + e − 18e2 + 8e3 arcsin


2 ,
1+e
(C.2)
= Aϕ(−1) (e),
p
4(−3 + e) (1 + e)(3 + e)
Aϕ(2) (e) =
3(1 − e)7/2 e3/2
"
√ √
2e 1 − e − 114 + 124e + 203e2 − 194e3 − 19e4 + 60e5


×
#
√
 r
e
+ 3(1 + e)2 (38 − 130e + 153e2 − 55e3 + 4e4 ) arcsin 2 ,
1+e

= Aϕ(−2) (e).
The coefficients Ar(k) can be written in a general way, for any k. We have

Ar(k) = 2e3/2 (e − 3)(1 + e)5/2 K(e; −k) − K(e; k) = −Ar(−k) .

(C.3)

D Subleading orders of the asymptotic adiabatic equations and solutions

This appendix provides the subleading orders of the asymptotic equations governing the
evolution of the slow variables and their asymptotic solutions.
The asymptotic adiabatic equations that drive the evolution of δ and e(0) read as
dδ A(e⋆ ) 1
= + (D.1a)
dt̃ δ (e⋆ log(δ) + B(e⋆ )) (e⋆ log(δ) + B(e⋆ ))2

e⋆ − 1 
B(e⋆ )A′ (e⋆ ) − A(e⋆ )B ′ (e⋆ )

× Eδ (e⋆ ) +
8
  
e⋆ − 1 ′
e⋆ A (e⋆ ) − A(e⋆ ) + Fδ (e⋆ ) log(δ) + Gδ (e⋆ ) log (δ) + o(δ 0 ),
2


+
8
de(0) e⋆ − 1 A(e⋆ ) 1
= + (D.1b)
dt̃ 8 δ (e⋆ log(δ) + B(e⋆ )) (e⋆ log(δ) + B(e⋆ ))2

e⋆ − 1 h
i
× Ee (e⋆ ) + (e⋆ − 1)B(e⋆ )A′ (e⋆ ) + A(e⋆ ) B(e⋆ ) − (e⋆ − 1)B ′ (e⋆ )
64
  
e⋆ − 1 ′
A(e⋆ ) + (e⋆ − 1) e⋆ A (e⋆ ) log(δ) + Ge (e⋆ ) log (δ) + o(δ 0 ).
2


+ Fe (e⋆ ) +
64

– 29 –
 P
The constants {Ci } := {A, Eδ , Ee , Fδ , Fe , Gδ , Ge } are written as Ci = k Ci,k where Ci,k are
real functions of e⋆ and of the values of the k Fourier modes of the self-force at the sepa-
ratrix. The explicit expressions for these constants are available on this Github repository.
The equations (D.1) have been expanded up to the first subleading order in δ. This makes
de(0)
it possible to derive the expression of dδ up to subleading order:

de(0) e⋆ − 1 δ
= +
dδ 8 64 A(e⋆ ) (e⋆ log(δ) + B(e⋆ ))2
(  
e⋆ − 1
× (e⋆ − 1)A(e⋆ )B(e⋆ ) + 64 Ee (e⋆ ) − Eδ (e⋆ )
8
 i (D.2)
h e⋆ − 1
+ (e⋆ − 1) e⋆ A(e⋆ ) + 64 Fe (e⋆ ) − Fδ (e⋆ ) log(δ)
8
  )
e⋆ − 1
+ 64 Ge (e⋆ ) − Gδ (e⋆ ) log2 (δ) + o(δ).
8

The solution to this equation is

 
e⋆ − 1 e⋆ − 1 2 1 2
e(0) (δ) = e⋆ + δ+ δ − e⋆ δ − 16 e⋆ Fe (e⋆ ) + 2(e⋆ − 1)e⋆ Fδ (e⋆ )
8 128 32 e3⋆ A(e⋆ )

+ (2e⋆ log(δ) − e⋆ − 2B(e⋆ ))(−8 Ge (e⋆ ) + (e⋆ − 1)Gδ (e⋆ ))
   
− 2B(e ⋆) B(e⋆ )
+ 4e e ⋆ Ei 2 log(δ) + − 8 e2⋆ Ee (e⋆ ) + (e⋆ − 1)e2⋆ Eδ (e⋆ )
e⋆



+ B(e⋆ ) 8 e⋆ Fe (e⋆ ) − (e⋆ − 1)e⋆ Fδ (e⋆ ) + B(e⋆ ) (−8 Ge (e⋆ ) + (e⋆ − 1)Gδ (e⋆ )) ,
(D.3)
et
Rz
where Ei is the exponential integral function defined as Ei(z) = −∞ t dt. This asymptotic
solution provides the subleading orders of Eq. (3.17) given in the body of the paper. In
particular, we see from this expression that the residual of the leading solution (3.17) is
O(δ 2 log(δ)).

E Lambert W function

This section provides elements on the Lambert W function that are useful for the main
text. In this section, the symbol e always denotes Napier’s constant, not the eccentricity.
The Lambert W function is the multivalued inverse of the function w → w ew :

W (z) = w ⇔ z = w ew . (E.1)

The distinct branches of this function are denoted with an integer k. Among these branches,
two are real, the branches k = 0 and k = −1, which interest us most. On the one hand,

– 30 –
Figure 6. The two real branches of the W Lambert function in the domain z ∈ [− 1e , 0[. In black,
W0 (z) has the image set [−1 , 0[. In red, W−1 (z) has the image set [−1, −∞[. It is monotonically
decreasing with z and asymptotes to −∞ as z → 0.

the branch k = 0 provides real values of the Lambert function for any z ≥ −e−1 . On the
other hand, the real branch k = −1 is defined for −e−1 ≤ z < 0 and can be seen as an
extension of the branch 0 on the real axis. The plots of the two real branches of the W
function are given in Figure 6.
In particular, we have the identities

Wk (−e−1 ) = −1, k = 0, −1,

W0 (0) = 0,
W−1 (z) → −∞ as z → 0, (E.2)
W0 (e) = 1,
Wk (z)
e−Wk (z) = if z ̸= 0.
z
The W function is a solution to the differential equation

df (z) 1
= f (z) . (E.3)
dz e (1 + f (z))

The principal branch of W is analytic at 0 and it can be shown [46] that the series expansion
around z = 0 is ∞
X (−n)n−1
W0 (z) = zn. (E.4)
n!
n=1

– 31 –
Figure 7. Solid line: the function W−1 (z). Dashed line: the asymptotic expansion (E.5) (truncated
to the order shown) of W−1 (z).

The other real branch, which is the one relevant for the late time dynamics of the inspiral,
is expanded as z 7→ 0 as

L2 L2 (−2 + L2 ) L2 (6 − 9L2 + 2L22 )

W−1 (z) = L1 − L2 + + +
L1 2L21 6L31
 5 ! (E.5)
L2 (−12 + 36L2 − 22L22 + 3L32 ) L2
+ +O ,
12L41 L1

where L1 := log(−z) and L2 := log(− log(−z)). This expansion is numerically accurate,

as it can be seen on Figure 7, which compares the function W−1 with its (truncated)
approximation (E.5).

– 32 –
References

[1] LIGO Scientific, Virgo collaboration, Observation of Gravitational Waves from a Binary
Black Hole Merger, Phys. Rev. Lett. 116 (2016) 061102 [1602.03837].

[2] A. Pound and B. Wardell, Black hole perturbation theory and gravitational self-force,
2101.04592.

[3] T. Hinderer and E.E. Flanagan, Two timescale analysis of extreme mass ratio inspirals in
Kerr. I. Orbital Motion, Phys. Rev. D78 (2008) 064028 [0805.3337].

[4] S.A. Hughes, N. Warburton, G. Khanna, A.J.K. Chua and M.L. Katz, Adiabatic waveforms
for extreme mass-ratio inspirals via multivoice decomposition in time and frequency, Phys.
Rev. D 103 (2021) 104014 [2102.02713].

[5] C. Cutler, D. Kennefick and E. Poisson, Gravitational radiation reaction for bound motion
around a Schwarzschild black hole, Phys. Rev. D 50 (1994) 3816.

[6] A. Pound and E. Poisson, Osculating orbits in Schwarzschild spacetime, with an application
to extreme mass-ratio inspirals, Phys. Rev. D 77 (2008) 044013 [0708.3033].

[7] J.R. Gair, E.E. Flanagan, S. Drasco, T. Hinderer and S. Babak, Forced motion near black
holes, Phys. Rev. D 83 (2011) 044037 [1012.5111].

[8] L. Barack and N. Sago, Gravitational self-force on a particle in eccentric orbit around a
Schwarzschild black hole, Phys. Rev. D 81 (2010) 084021 [1002.2386].

[9] L. Barack, T. Damour and N. Sago, Precession effect of the gravitational self-force in a
Schwarzschild spacetime and the effective one-body formalism, Phys. Rev. D 82 (2010)
084036 [1008.0935].

[10] L. Barack and N. Sago, Beyond the geodesic approximation: conservative effects of the
gravitational self-force in eccentric orbits around a Schwarzschild black hole, Phys. Rev. D 83
(2011) 084023 [1101.3331].

[11] S. Drasco and S.A. Hughes, Gravitational wave snapshots of generic extreme mass ratio
inspirals, Phys. Rev. D 73 (2006) 024027 [gr-qc/0509101].

[12] S. Akcay, A Fast Frequency-Domain Algorithm for Gravitational Self-Force: I. Circular

Orbits in Schwarzschild Spacetime, Phys. Rev. D 83 (2011) 124026 [1012.5860].

[13] N. Warburton, S. Akcay, L. Barack, J.R. Gair and N. Sago, Evolution of inspiral orbits
around a Schwarzschild black hole, Phys. Rev. D 85 (2012) 061501 [1111.6908].

[14] M. Van De Meent and N. Warburton, Fast Self-forced Inspirals, Class. Quant. Grav. 35
(2018) 144003 [1802.05281].

[15] R. Fujita and M. Shibata, Extreme mass ratio inspirals on the equatorial plane in the
adiabatic order, Phys. Rev. D 102 (2020) 064005 [2008.13554].

– 33 –
[16] M.L. Katz, A.J.K. Chua, L. Speri, N. Warburton and S.A. Hughes, Fast
extreme-mass-ratio-inspiral waveforms: New tools for millihertz gravitational-wave data
analysis, Phys. Rev. D 104 (2021) 064047 [2104.04582].

[17] P. Lynch, M. van de Meent and N. Warburton, Eccentric self-forced inspirals into a rotating
black hole, Class. Quant. Grav. 39 (2022) 145004 [2112.05651].

[18] B. Leather and N. Warburton, Applying the effective-source approach to frequency-domain

self-force calculations for eccentric orbits, Phys. Rev. D 108 (2023) 084045 [2306.17221].

[19] L. Speri, M.L. Katz, A.J.K. Chua, S.A. Hughes, N. Warburton, J.E. Thompson et al., Fast
and Fourier: Extreme Mass Ratio Inspiral Waveforms in the Frequency Domain,
2307.12585.

[20] S. Albanesi, A. Placidi, A. Nagar, M. Orselli and S. Bernuzzi, New avenue for accurate
analytical waveforms and fluxes for eccentric compact binaries, Phys. Rev. D 105 (2022)
L121503 [2203.16286].

[21] S. Albanesi, A. Nagar and S. Bernuzzi, Effective one-body model for extreme-mass-ratio
spinning binaries on eccentric equatorial orbits: Testing radiation reaction and waveform,
Phys. Rev. D 104 (2021) 024067 [2104.10559].

[22] D.R. Becker and S.A. Hughes, Transition from adiabatic inspiral to plunge for eccentric
binaries, 2410.09160.

[23] P.A. Sundararajan, The Transition from adiabatic inspiral to geodesic plunge for a compact
object around a massive Kerr black hole: Generic orbits, Phys. Rev. D77 (2008) 124050
[0803.4482].

[24] R.W. O’Shaughnessy, Transition from inspiral to plunge for eccentric equatorial Kerr orbits,
Phys. Rev. D 67 (2003) 044004 [gr-qc/0211023].

[25] S. Albanesi, S. Bernuzzi, T. Damour, A. Nagar and A. Placidi, Faithful effective-one-body

waveform of small-mass-ratio coalescing black hole binaries: The eccentric, nonspinning case,
Phys. Rev. D 108 (2023) 084037 [2305.19336].

[26] M. De Amicis, S. Albanesi and G. Carullo, Inspiral-inherited ringdown tails, Phys. Rev. D
110 (2024) 104005 [2406.17018].

[27] T. Islam, G. Faggioli, G. Khanna, S.E. Field, M. van de Meent and A. Buonanno,
Phenomenology and origin of late-time tails in eccentric binary black hole mergers,
2407.04682.

[28] L.C. Stein and N. Warburton, Location of the last stable orbit in Kerr spacetime, Phys. Rev.
D 101 (2020) 064007 [1912.07609].

[29] G. Compère and L. Küchler, Self-consistent adiabatic inspiral and transition motion, Phys.
Rev. Lett. 126 (2021) 241106 [2102.12747].

– 34 –
[30] G. Compère and L. Küchler, Asymptotically matched quasi-circular inspiral and
transition-to-plunge in the small mass ratio expansion, SciPost Phys. 13 (2022) 043
[2112.02114].

[31] L. Küchler, G. Compère, L. Durkan and A. Pound, Self-force framework for

transition-to-plunge waveforms, 2405.00170.

[32] A. Ori and K.S. Thorne, The Transition from inspiral to plunge for a compact body in a
circular equatorial orbit around a massive, spinning black hole, Phys. Rev. D62 (2000)
124022 [gr-qc/0003032].

[33] A. Buonanno and T. Damour, Transition from inspiral to plunge in binary black hole
coalescences, Phys. Rev. D62 (2000) 064015 [gr-qc/0001013].

[34] M. Kesden, Transition from adiabatic inspiral to plunge into a spinning black hole, Phys.
Rev. D83 (2011) 104011 [1101.3749].

[35] R. Fujita and W. Hikida, Analytical solutions of bound timelike geodesic orbits in Kerr
spacetime, Class. Quant. Grav. 26 (2009) 135002 [0906.1420].

[36] P. Lynch and O. Burke, A note on the conversion of orbital angles for extreme mass ratio
inspirals, 2411.04955.

[37] J. Miller and A. Pound, Two-timescale evolution of extreme-mass-ratio inspirals: waveform

generation scheme for quasicircular orbits in Schwarzschild spacetime, Phys. Rev. D 103
(2021) 064048 [2006.11263].

[38] G. I.S. and R. I.M., 1 - elementary functions, in Table of Integrals, Series, and Products
(Eighth Edition), D. Zwillinger, V. Moll, I. Gradshteyn and I. Ryzhik, eds., (Boston),
pp. 25–62, Academic Press (2014), DOI.

[39] F. Olver, D. Lozier, R. Boisvert and C. Clark, The NIST Handbook of Mathematical
Functions, Cambridge University Press, New York, NY (2010-05-12 00:05:00, 2010).

[40] K. Glampedakis and D. Kennefick, Zoom and whirl: Eccentric equatorial orbits around
spinning black holes and their evolution under gravitational radiation reaction, Phys. Rev. D
66 (2002) 044002 [gr-qc/0203086].

[41] C. Cutler, D. Kennefick and E. Poisson, Gravitational radiation reaction for bound motion
around a schwarzschild black hole, Phys. Rev. D 50 (1994) 3816.

[42] D. Kennefick, Stability under radiation reaction of circular equatorial orbits around Kerr
black holes, Phys. Rev. D58 (1998) 064012 [gr-qc/9805102].

[43] A. Mummery and S. Balbus, Inspirals from the Innermost Stable Circular Orbit of Kerr
Black Holes: Exact Solutions and Universal Radial Flow, Phys. Rev. Lett. 129 (2022) 161101
[2209.03579].

[44] A. Mummery and S. Balbus, Complete characterization of the orbital shapes of the

– 35 –
 noncircular Kerr geodesic solutions with circular orbit constants of motion, Phys. Rev. D
107 (2023) 124058 [2302.01159].

[45] C. Gundlach, S. Akcay, L. Barack and A. Nagar, Critical phenomena at the threshold of
immediate merger in binary black hole systems: the extreme mass ratio case, Phys. Rev. D
86 (2012) 084022 [1207.5167].

[46] R. Corless, G. Gonnet, D. Hare, D. Jeffrey and D. Knuth, On the lambert w function,
Advances in Computational Mathematics 5 (1996) 329.

– 36 –