Introduction
In a binary black hole (BBH) with spins misaligned with the orbital angular momentum, spin-orbit and spin-spin couplings make the orbital angular momentum L \mathbf{L} L and the two spins S 1 \mathbf{S}_1 S 1 and S 2 \mathbf{S}_2 S 2 precess and nutate, which modulates the emitted gravitational waves (GWs) [1]. Spins are also of astrophysical interest, since their orientations depend on how the binary was formed, and measuring them with GWs helps discriminate between formation channels [3 ] [2].
In the post-Newtonian (PN) regime, the orbital period of a spinning BBH is much shorter than its precession timescale, which is itself much shorter than the radiation-reaction (RR) timescale [4 , 5 ] . This hierarchy allows one to average the precession equations over a Newtonian orbit. The resulting 2PN orbit-averaged (OA) precession equations were introduced by Schnittman [6 ] and extended by Racine to include the quadrupole-monopole couplings [7 ] . Racine also found a new constant of motion (COM), later called the projected effective spin χ e f f \chi_{\mathrm{eff}} χ eff [4 ] .
The OA equations have since been one of the main frameworks for studying spin evolution. Schnittman found that they admit spin-orbit resonances, in which L \mathbf{L} L , S 1 \mathbf{S}_1 S 1 , and S 2 \mathbf{S}_2 S 2 are coplanar and precess jointly about the total angular momentum J \mathbf{J} J [6 ] . Bogdanović et al. [8 ] found that, for circular BBHs with identical masses, spin magnitudes, and initial separations, evolved with spin-independent RR, an initially isotropic distribution of spin directions remains isotropic during the inspiral. Gerosa et al. [3 ] showed that anisotropic distributions can instead become locked near resonant planes.
On the analytical side, the conservation of χ e f f \chi_{\mathrm{eff}} χ eff has been the key to understanding the OA system. It reduces the precession to the motion of a single variable in an effective potential, which led to closed-form solutions [4 , 9 , 10 , 11 ] . The stability of the aligned-spin configurations relies on χ e f f \chi_{\mathrm{eff}} χ eff as well. The up-down configuration was shown to become unstable in a finite range of orbital separations, first from the shape of the effective potentials [12 ] , and later by describing small perturbations around it as a harmonic oscillator [13 ] . Secular precession equations for eccentric orbits and arbitrary spin-induced quadrupole moments were derived in Ref. [14 ] , and the linear stability of the aligned configurations was analyzed in Ref. [15 ] .
The OA system has also been studied with the tools of celestial mechanics. Correia [16 ] 4 orbit averaged the full Hamiltonian, obtained the OA equations of motion (EOMs) from it through the Poisson brackets of the angular momenta, noted that the system is integrable, and argued that the spin-orbit resonances are Cassini states rather than resonances. An analogous problem, the precession of a planet with a satellite, involves three precessing angular momenta and a Hamiltonian quadratic in their mutual cosines. Boué and Laskar [17 ] showed that it is integrable and obtained its nutation and precession frequencies by quadratures.
In this article, we revisit the 2PN OA system from the point of view of Hamiltonian mechanics. We show why the orbit average of the full PN Hamiltonian generates the OA EOMs, which is not immediate for L \mathbf{L} L , and we derive the same Hamiltonian independently from the OA EOMs alone. We relate the three COMs in involution of the OA system to those of the full 2PN system, compute the action variables with the fictitious variables introduced in Ref. [18 ] , and then obtain the frequencies and angle variables. The frequencies confirm that the spin-orbit resonances are not resonances in the usual sense.
We then propose a criterion for the breakdown of the integrability conditions, expressed through the Jacobian between the COMs and the action variables. Like the method of fictitious variables, it is a general tool of Hamiltonian mechanics, not tied to PN theory. We illustrate it on simple systems. For the OA system, it recovers the up-down instability and the separations at which it starts and ends. Since the criterion depends on the orbit only through the magnitude of L \mathbf{L} L , it gives these separations explicitly for eccentric orbits, for which the onset is set by a ( 1 − e 2 ) a(1-e^2) a ( 1 − e 2 ) rather than by the semi-major axis a a a alone. The linear stability analysis of Ref. [15 ] , which holds for eccentric orbits, also depends on the orbit only through L \mathbf{L} L , but does not give the thresholds L ± L_{\pm} L ± explicitly.
The Hamiltonian structure also has astrophysical consequences. We show that it explains the preservation of isotropy observed in Ref. [8 ] , through an argument close to Liouville’s volume theorem. The argument holds when all members of the ensemble share the same parameters and RR does not depend on the spins. It also predicts when isotropy can be lost, for instance when the masses of the members are correlated with the direction of L \mathbf{L} L , which we confirm numerically.
The methods developed here also apply to OA binary neutron stars (BNSs). Their quadrupole-monopole terms differ from the black-hole values [19 ] , and χ e f f \chi_{\mathrm{eff}} χ eff is no longer conserved [20 ] . The OA BNS system nevertheless remains integrable, and in Sec. VIII we use its OA Hamiltonian to generalize χ e f f \chi_{\mathrm{eff}} χ eff to BNSs, in closed form and for eccentric orbits.
The paper is organized as follows. § 2 introduces the system and its OA EOMs. Section III discusses its Hamiltonian structure and derives the OA Hamiltonian. Section IV discusses its integrability and computes its action variables, and § 3 its frequencies and angle variables. Section VI presents the criterion for the breakdown of the integrability conditions and its applications. Section VII studies the consequences of the Hamiltonian structure for ensembles of BBHs, Sec. VIII extends part of these results to BNSs, and § 3.5 concludes.
Setup
Notation and conventions
We consider a spinning and eccentric BBH with masses m 1 > m 2 m_1 > m_2 m 1 > m 2 . The system has a total mass m = m 1 + m 2 m = m_1 + m_2 m = m 1 + m 2 , a reduced mass μ = m 1 m 2 / m \mu= m_1m_2/m μ = m 1 m 2 / m , and a symmetric mass ratio ν = μ / m \nu= \mu/m ν = μ / m . We set G = 1 G = 1 G = 1 and keep c c c explicit as a PN bookkeeping parameter.
Throughout this work, three-dimensional vectors are denoted by boldface letters (e.g., V \mathbf{V} V ), their Euclidean norms by the corresponding unbolded letter, V ≡ ∥ V ∥ V \equiv\lVert\mathbf{V}\rVert V ≡ ∥ V ∥ , and unit vectors by a hat, V ^ ≡ V / V \hat{\mathbf{V}} \equiv\mathbf{V}/V V ^ ≡ V / V . t t t denotes the coordinate time and overdots denote d / d t d/dt d / d t .
We work in the center-of-mass frame, where P 1 = − P 2 ≡ P \mathbf{P}_1 = -\mathbf{P}_2 \equiv\mathbf{P} P 1 = − P 2 ≡ P , with relative separation vector R = R 1 − R 2 \mathbf{R} = \mathbf{R}_1 - \mathbf{R}_2 R = R 1 − R 2 , and unit vector n = R / R \mathbf{n} = \mathbf{R}/R n = R / R . The orbital angular momentum is L = R × P \mathbf{L} = \mathbf{R} \times\mathbf{P} L = R × P , the spins of the two black holes are S 1 \mathbf{S}_1 S 1 and S 2 \mathbf{S}_2 S 2 , and the total angular momentum is J = L + S 1 + S 2 \mathbf{J} = \mathbf{L} + \mathbf{S}_1 + \mathbf{S}_2 J = L + S 1 + S 2 . We refer to the three vectors ( L , S 1 , S 2 ) (\mathbf{L}, \mathbf{S}_1, \mathbf{S}_2) ( L , S 1 , S 2 ) as the triad, and to a given set of their values, i.e., a point of the orbit-averaged phase space, as a configuration.
We use two mass-weighted spin combinations S e f f ≡ σ 1 S 1 + σ 2 S 2 \mathbf{S}_{\mathrm{eff}} \equiv\sigma_1\mathbf{S}_1 + \sigma_2\mathbf{S}_2 S eff ≡ σ 1 S 1 + σ 2 S 2 and S 0 ≡ Σ 1 S 1 + Σ 2 S 2 \mathbf{S}_0 \equiv\Sigma_1\mathbf{S}_1 + \Sigma_2\mathbf{S}_2 S 0 ≡ Σ 1 S 1 + Σ 2 S 2 , where σ a ≡ 1 + 3 m b / 4 m a \sigma_a \equiv1 + 3m_b/4m_a σ a ≡ 1 + 3 m b /4 m a and Σ a ≡ m / m a \Sigma_a \equiv m/m_a Σ a ≡ m / m a , with ( a , b ) ∈ { ( 1 , 2 ) , ( 2 , 1 ) } (a,b) \in\{(1,2),(2,1)\} ( a , b ) ∈ {( 1 , 2 ) , ( 2 , 1 )} . S e f f \mathbf{S}_{\mathrm{eff}} S eff features in the spin-orbit coupling terms while S 0 \mathbf{S}_0 S 0 features in the spin-spin coupling terms. κ 1 \kappa_1 κ 1 , κ 2 \kappa_2 κ 2 , and γ \gamma γ denote the angles between ( L , S 1 ) (\mathbf{L}, \mathbf{S}_1) ( L , S 1 ) , ( L , S 2 ) (\mathbf{L}, \mathbf{S}_2) ( L , S 2 ) , and ( S 1 , S 2 ) (\mathbf{S}_1, \mathbf{S}_2) ( S 1 , S 2 ) , whereas Δ Φ \Delta\Phi ΔΦ is the angle between the projections of the two spins on the plane perpendicular to L \mathbf{L} L .
The system
The two BHs move on a bound orbit, and their dynamics is governed by a Hamiltonian expressed in the Arnowitt-Deser-Misner (ADM) coordinates [21 ] with the Newton-Wigner-Pryce (NWP) spin supplementary condition (SSC) [22 , 23 ] , which fixes the center-of-mass worldline of each spinning body. The Hamiltonian reads [24 , 25 , 26 , 27 ]
H = H N + 1 c 2 H 1 P N + 1 c 2 H S O + 1 c 4 H 2 P N + 1 c 2 H S S , (1) H = H_N + \frac{1}{c^2}H_{\mathrm{1PN}} + \frac{1}{c^2}H_{\mathrm{SO}} + \frac{1}{c^4}H_{\mathrm{2PN}} + \frac{1}{c^2}H_{\mathrm{SS}}, \tag*{(1)} H = H N + c 2 1 H 1PN + c 2 1 H SO + c 4 1 H 2PN + c 2 1 H SS , (1)
where we display the spin-dependent terms
H S O = 2 R 3 L ⋅ S e f f , (2) H_{\mathrm{SO}} = \frac{2}{R^3}\mathbf{L}\cdot\mathbf{S}_{\mathrm{eff}}, \tag*{(2)} H SO = R 3 2 L ⋅ S eff , (2)
H S S = ν 2 R 3 [ 3 ( n ⋅ S 0 ) 2 − S 0 2 ] . (3) H_{\mathrm{SS}} = \frac{\nu}{2R^3}\left[3(\mathbf{n}\cdot\mathbf{S}_0)^2-S_0^2\right]. \tag*{(3)} H SS = 2 R 3 ν [ 3 ( n ⋅ S 0 ) 2 − S 0 2 ] . (3)
With S a = m a 2 χ a / c \mathbf{S}_a=m_a^2\boldsymbol{\chi}_a/c S a = m a 2 χ a / c , H S O H_{\mathrm{SO}} H SO and H S S H_{\mathrm{SS}} H SS enter at 1.5PN and 2PN order, respectively. The purely orbital terms H N H_N H N , H 1 P N H_{\mathrm{1PN}} H 1PN , and H 2 P N H_{\mathrm{2PN}} H 2PN can be found in Eqs. (2.2) of Ref. [11 ] . The form of H S S H_{\mathrm{SS}} H SS in terms of the single combination S 0 \mathbf{S}_0 S 0 includes the S 1 S 2 S_1S_2 S 1 S 2 couplings as well as the self-spin (S 1 2 S_1^2 S 1 2 , S 2 2 S_2^2 S 2 2 ) couplings. The evolution of any function f f f of the phase-space variables is given by
d f d t = { f , H } , (4) \frac{df}{dt}=\{f,H\}, \tag*{(4)} d t df = { f , H } , (4)
with the fundamental canonical Poisson brackets (PBs)
{ R i , P j } = δ i j , (5) \{R_i,P_j\}=\delta_{ij}, \tag*{(5)} { R i , P j } = δ ij , (5)
{ S a i , S b j } = δ a b ϵ i j k S a k , a , b ∈ { 1 , 2 } , (6) \{S_{ai},S_{bj}\}=\delta_{ab}\epsilon_{ij}^{\ \ k}S_{ak},\qquad a,b\in\{1,2\}, \tag*{(6)} { S ai , S bj } = δ ab ϵ ij k S ak , a , b ∈ { 1 , 2 } , (6)
combined with the usual PB rules of anti-symmetry and chain differentiation [28 ] (with v l v_l v l representing the phase space variables),
{ f , g } = − { g , f } , (7) \{f,g\}=-\{g,f\}, \tag*{(7)} { f , g } = − { g , f } , (7)
{ f , g ( v l ) } = { f , v l } ∂ g ∂ v l , (8) \{f,g(v_l)\}=\{f,v_l\}\frac{\partial g}{\partial v_l}, \tag*{(8)} { f , g ( v l )} = { f , v l } ∂ v l ∂ g , (8)
and any PB that is not covered by Eqs. (2.4)–(2.5) simply vanishes. If { f , g } = 0 \{f,g\}=0 { f , g } = 0 , then we say that f f f and g g g commute, or they are in involution.
One can obtain the EOMs for the three angular momenta directly in terms of physical variables5 6
d S a d t = 1 c 2 R 3 { 2 σ a L + ν Σ a [ 3 ( n ⋅ S 0 ) n − S 0 ] } × S a , (9) \frac{d\mathbf{S}_a}{dt}=\frac{1}{c^2R^3}\left\{2\sigma_a\mathbf{L}+\nu\Sigma_a\left[3(\mathbf{n}\cdot\mathbf{S}_0)\mathbf{n}-\mathbf{S}_0\right]\right\}\times\mathbf{S}_a, \tag*{(9)} d t d S a = c 2 R 3 1 { 2 σ a L + ν Σ a [ 3 ( n ⋅ S 0 ) n − S 0 ] } × S a , (9)
d L d t = − 1 c 2 R 3 { 2 L × S e f f + 3 ν ( n ⋅ S 0 ) ( n × S 0 ) } . (10) \frac{d\mathbf{L}}{dt}=-\frac{1}{c^2R^3}\left\{2\mathbf{L}\times\mathbf{S}_{\mathrm{eff}}+3\nu(\mathbf{n}\cdot\mathbf{S}_0)(\mathbf{n}\times\mathbf{S}_0)\right\}. \tag*{(10)} d t d L = − c 2 R 3 1 { 2 L × S eff + 3 ν ( n ⋅ S 0 ) ( n × S 0 ) } . (10)
The so-called 2PN orbit-averaged (OA) EOMs for the three angular momenta S 1 \mathbf{S}_1 S 1 , S 2 \mathbf{S}_2 S 2 , and L \mathbf{L} L were introduced by Schnittman [6 ] and have been the subject of many analyses [4 , 7 ] . They are obtained by averaging Eq. (2.6) over a Newtonian orbit, keeping the angular momenta ( L , S a ) (\mathbf{L}, \mathbf{S}_a) ( L , S a ) fixed, using [7 ]
⟨ 1 R 3 ⟩ ≡ 1 D 3 = 1 ( a 1 − e 2 ) 3 , ⟨ n i n j R 3 ⟩ = 1 2 D 3 ( δ i j − L ^ i L ^ j ) (11) \left\langle\frac{1}{R^3} \right\rangle\equiv\frac{1}{D^3} = \frac{1}{\left(a\sqrt{1-e^2}\right)^3}, \qquad\left\langle\frac{n^i n^j}{R^3} \right\rangle= \frac{1}{2D^3}\left(\delta^{ij}-\hat{L}^i\hat{L}^j\right) \tag*{(11)} ⟨ R 3 1 ⟩ ≡ D 3 1 = ( a 1 − e 2 ) 3 1 , ⟨ R 3 n i n j ⟩ = 2 D 3 1 ( δ ij − L ^ i L ^ j ) (11)
where D D D , a a a and e e e are respectively the semi-minor axis, the semi-major axis and the eccentricity of the associated Newtonian ellipse. The OA spin dynamics therefore depend on the shape of the ellipse only through D D D . The averaging procedure is justified because the spins precess on a timescale ∼ ( c / v ) 2 T o r b \sim(c/v)^2 T_{\mathrm{orb}} ∼ ( c / v ) 2 T orb and the orbital angular momentum on ∼ ( c / v ) 3 T o r b \sim(c/v)^3 T_{\mathrm{orb}} ∼ ( c / v ) 3 T orb , both much longer than the orbital period. The resulting OA EOMs can be written as6 7
⟨ d S a d t ⟩ = 1 c 2 D 3 { 2 σ a L + ν Σ 2 [ S 0 − 3 ( L ^ ⋅ S 0 ) L ^ ] } × S a (12) \left\langle\frac{d\mathbf{S}_a}{dt} \right\rangle= \frac{1}{c^2D^3}\left\{2\sigma_a\mathbf{L}+\frac{\nu\Sigma}{2}\left[\mathbf{S}_0-3(\hat{\mathbf{L}}\cdot\mathbf{S}_0)\hat{\mathbf{L}}\right]\right\}\times\mathbf{S}_a \tag*{(12)} ⟨ d t d S a ⟩ = c 2 D 3 1 { 2 σ a L + 2 ν Σ [ S 0 − 3 ( L ^ ⋅ S 0 ) L ^ ] } × S a (12)
⟨ d L d t ⟩ = 1 c 2 D 3 { 2 S e f f − 3 ν 2 ( L ⋅ S 0 ) L 2 S 0 } × L (13) \left\langle\frac{d\mathbf{L}}{dt} \right\rangle= \frac{1}{c^2D^3}\left\{2\mathbf{S}_{\mathrm{eff}}-\frac{3\nu}{2}\frac{(\mathbf{L}\cdot\mathbf{S}_0)}{L^2}\mathbf{S}_0\right\}\times\mathbf{L} \tag*{(13)} ⟨ d t d L ⟩ = c 2 D 3 1 { 2 S eff − 2 3 ν L 2 ( L ⋅ S 0 ) S 0 } × L (13)
One can check directly that Eqs. (2.8) conserve J \mathbf{J} J , L \mathbf{L} L , S 1 \mathbf{S}_1 S 1 , and S 2 \mathbf{S}_2 S 2 . In what follows, we omit the angular brackets on the OA time derivatives for brevity.
Hamiltonian nature of the system
Hamiltonian of the orbit-averaged system
We begin this section with the statement that the 2PN OA spinning BBH system is Hamiltonian, i.e., there exists a Hamiltonian H ˉ \bar{H} H ˉ with a set of fundamental PBs such that the EOMs (2.8) follow from d f / d t = { f , H ˉ } df/dt=\{f,\bar{H}\} df / d t = { f , H ˉ } . This Hamiltonian is
H ˉ = 1 c 2 D 3 [ 2 L ⋅ S e f f + 1 2 S 1 ⋅ S 2 − 3 ν 4 ( L ^ ⋅ S 0 ) 2 ] (14) \bar{H}=\frac{1}{c^2D^3}\left[2\mathbf{L}\cdot\mathbf{S}_{\mathrm{eff}}+\frac{1}{2}\mathbf{S}_1\cdot\mathbf{S}_2-\frac{3\nu}{4}(\hat{\mathbf{L}}\cdot\mathbf{S}_0)^2\right] \tag*{(14)} H ˉ = c 2 D 3 1 [ 2 L ⋅ S eff + 2 1 S 1 ⋅ S 2 − 4 3 ν ( L ^ ⋅ S 0 ) 2 ] (14)
together with the fundamental SO(3) PBs
{ S a i , S b j } = δ a b ϵ i j k S a k , { L i , L j } = ϵ i j k L k , { L i , S a j } = 0. (15) \{S_{ai},S_{bj}\}=\delta_{ab}\epsilon_{ijk}S_{ak},\qquad\{L_i,L_j\}=\epsilon_{ijk}L_k,\qquad\{L_i,S_{aj}\}=0. \tag*{(15)} { S ai , S bj } = δ ab ϵ ij k S ak , { L i , L j } = ϵ ij k L k , { L i , S aj } = 0. (15)
For any V ∈ { L , S 1 , S 2 } \mathbf{V}\in\{\mathbf{L},\mathbf{S}_1,\mathbf{S}_2\} V ∈ { L , S 1 , S 2 } , Eqs. (2.5) and (15) let us write
d V d t = { V , H ˉ } = ∂ H ˉ ∂ V × V . (16) \frac{d\mathbf{V}}{dt}=\{\mathbf{V},\bar{H}\}=\frac{\partial\bar{H}}{\partial\mathbf{V}}\times\mathbf{V}. \tag*{(16)} d t d V = { V , H ˉ } = ∂ V ∂ H ˉ × V . (16)
This directly reproduces Eqs. (2.8).
For the full non-OA system, L = R × P \mathbf{L}=\mathbf{R}\times\mathbf{P} L = R × P is built from the canonical position and momenta variables, and the canonical PBs (2.4) imply { L i , L j } = ϵ i j k L k \{L_i,L_j\}=\epsilon_{ijk}L_k { L i , L j } = ϵ ij k L k , the same algebra as the spins.
In the OA system, R \mathbf{R} R and P \mathbf{P} P have been averaged out and only L \mathbf{L} L remains; we therefore take Eqs. (15) , inherited from the full system, as the fundamental brackets. L 2 L^2 L 2 , S 1 2 S_1^2 S 1 2 , and S 2 2 S_2^2 S 2 2 commute with every function of ( L , S 1 , S 2 ) (\mathbf{L},\mathbf{S}_1,\mathbf{S}_2) ( L , S 1 , S 2 ) , and they are called Casimir invariants. Thus they are conserved and the OA phase space is the product of the spherical manifolds for S 1 \mathbf{S}_1 S 1 , S 2 \mathbf{S}_2 S 2 and L \mathbf{L} L : S 2 ( S 1 ) × S 2 ( S 2 ) × S 2 ( L ) \mathbb{S}^2(S_1) \times\mathbb{S}^2(S_2) \times\mathbb{S}^2(L) S 2 ( S 1 ) × S 2 ( S 2 ) × S 2 ( L ) , with L L L , S 1 S_1 S 1 , and S 2 S_2 S 2 as fixed parameters.
However, we note that the variables ( L , S 1 , S 2 ) (\mathbf{L},\mathbf{S}_1,\mathbf{S}_2) ( L , S 1 , S 2 ) are not canonical. The canonical pairs are ( ϕ S 1 , S 1 z ) (\phi_{S_1}, S_{1z}) ( ϕ S 1 , S 1 z ) , ( ϕ S 2 , S 2 z ) (\phi_{S_2}, S_{2z}) ( ϕ S 2 , S 2 z ) , and ( ϕ L , L z ) (\phi_L, L_z) ( ϕ L , L z ) , where ϕ V \phi_{\mathbf{V}} ϕ V is the azimuthal angle corresponding to V \mathbf{V} V about a fixed axis z ^ \hat{\mathbf{z}} z ^ , chosen not parallel to J \mathbf{J} J so that J 2 J^2 J 2 and J z J_z J z are independent. One can verify that Eqs. (15) imply [18 ]
{ ϕ S 1 , S 1 z } = { ϕ S 2 , S 2 z } = { ϕ L , L z } = 1 , (17) \{\phi_{S_1},S_{1z}\}=\{\phi_{S_2},S_{2z}\}=\{\phi_L,L_z\}=1, \tag*{(17)} { ϕ S 1 , S 1 z } = { ϕ S 2 , S 2 z } = { ϕ L , L z } = 1 , (17)
and all other brackets among them vanish (apart from the ones connected via anti-symmetry), thereby supporting the canonical nature of these coordinates. This Hamiltonian and these brackets were obtained in Ref. [16 ] by orbit averaging the full Hamiltonian. In the rest of this section, we check that the OA EOMs are Hamiltonian without assuming prior knowledge of H ˉ \bar{H} H ˉ (Sec. III B), then show why orbit averaging the full Hamiltonian gives the OA EOMs, and finally derive H ˉ \bar{H} H ˉ from the OA EOMs alone (Sec. III C).
Testing the Hamiltonian nature of the system
As per Ref. [29 ] , a system is Hamiltonian if and only if for any two functions f f f and g g g of the phase space variables V \mathbf{V} V ,
d d t { f , g } − { f ˙ , g } − { f , g ˙ } = 0 , (18) \frac{d}{dt}\{f,g\}-\{\dot{f},g\}-\{f,\dot{g}\}=0, \tag*{(18)} d t d { f , g } − { f ˙ , g } − { f , g ˙ } = 0 , (18)
holds true. Eq. (2.5) lets us write
d d t { f , g } − { f ˙ , g } − { f , g ˙ } = ∂ i f ∂ j g [ d d t { V i , V j } − { V ˙ i , V j } − { V i , V ˙ j } ] , (19) \frac{d}{dt}\{f,g\}-\{\dot{f},g\}-\{f,\dot{g}\}=\partial_i f\,\partial_j g\left[\frac{d}{dt}\{V^i,V^j\}-\{\dot{V}^i,V^j\}-\{V^i,\dot{V}^j\}\right], \tag*{(19)} d t d { f , g } − { f ˙ , g } − { f , g ˙ } = ∂ i f ∂ j g [ d t d { V i , V j } − { V ˙ i , V j } − { V i , V ˙ j } ] , (19)
which implies that the condition
d d t { V i , V j } − { V ˙ i , V j } − { V i , V ˙ j } = 0 , (20) \frac{d}{dt}\{V^i,V^j\}-\{\dot{V}^i,V^j\}-\{V^i,\dot{V}^j\}=0, \tag*{(20)} d t d { V i , V j } − { V ˙ i , V j } − { V i , V ˙ j } = 0 , (20)
alone guarantees the Hamiltonian nature of a system. It can be checked that the system of Eq. (2.8) with PBs given by Eq. (15) indeed satisfies Eq. (20) . This confirms, without using the explicit form of H ˉ \bar{H} H ˉ , that the OA EOMs (2.8) correspond to a Hamiltonian. We now show how this Hamiltonian is obtained.
Derivation of the orbit-averaged Hamiltonian
We obtain H ˉ \bar{H} H ˉ in two ways, (a) by orbit averaging the full Hamiltonian H H H , and (b) from an ansatz built from the OA EOMs alone.
Derivation via direct orbit averaging
If the full non-OA Hamiltonian (presented in Eq. (1) ) is known, one can derive H ˉ \bar{H} H ˉ directly. We first remark that the purely orbital terms H N H_{\mathrm{N}} H N , H 1 P N H_{\mathrm{1PN}} H 1PN , and H 2 P N H_{\mathrm{2PN}} H 2PN commute with ( L , S 1 , S 2 ) (\mathbf{L}, \mathbf{S}_1, \mathbf{S}_2) ( L , S 1 , S 2 ) and do not contribute to Eqs. (2.6). Averaging the remaining terms H S O H_{\mathrm{SO}} H SO and H S S H_{\mathrm{SS}} H SS over R \mathbf{R} R (including n \mathbf{n} n ) with Eqs. (11) , while holding ( L , S 1 , S 2 ) (\mathbf{L}, \mathbf{S}_1, \mathbf{S}_2) ( L , S 1 , S 2 ) fixed, gives
⟨ H ⟩ = H ˉ + 1 4 c 2 D 3 [ m 2 m 1 S 1 2 + m 1 m 2 S 2 2 ] . (21) \langle H\rangle= \bar{H} + \frac{1}{4c^2D^3}\left[\frac{m_2}{m_1}S_1^2+\frac{m_1}{m_2}S_2^2\right]. \tag*{(21)} ⟨ H ⟩ = H ˉ + 4 c 2 D 3 1 [ m 1 m 2 S 1 2 + m 2 m 1 S 2 2 ] . (21)
Holding the angular momenta fixed is justified because they evolve on timescales much longer than T o r b T_{\mathrm{orb}} T orb (§ 2 B). The Casimir terms in square brackets in Eq. (21) commute with all functions in phase space of the OA system and therefore do not contribute to the dynamics. They can thus be omitted. One can check that ⟨ H ⟩ \langle H\rangle ⟨ H ⟩ generates the OA EOMs (2.8) using Eq. (4) . However it is instructive to see how it comes about using simple arguments, as we do below, first for S a \mathbf{S}_a S a and then for L \mathbf{L} L .
Spins. The right-hand sides of Eqs. (11) depend on L ^ \hat{\mathbf{L}} L ^ but not on S a \mathbf{S}_a S a , and S a \mathbf{S}_a S a is fixed while orbit averaging. Averaging therefore commutes with ∂ / ∂ S a \partial/\partial\mathbf{S}_a ∂ / ∂ S a , and since ⟨ H ⟩ = H ˉ \langle H\rangle= \bar{H} ⟨ H ⟩ = H ˉ up to the omitted Casimir terms,
⟨ d S a d t ⟩ = ⟨ ∂ H ∂ S a ⟩ × S a = ∂ H ˉ ∂ S a × S a = { S a , H ˉ } . (22) \left\langle\frac{\mathrm{d}\mathbf{S}_a}{\mathrm{d}t}\right\rangle= \left\langle\frac{\partial H}{\partial\mathbf{S}_a}\right\rangle\times\mathbf{S}_a = \frac{\partial\bar{H}}{\partial\mathbf{S}_a}\times\mathbf{S}_a = \{\mathbf{S}_a,\bar{H}\}. \tag*{(22)} ⟨ d t d S a ⟩ = ⟨ ∂ S a ∂ H ⟩ × S a = ∂ S a ∂ H ˉ × S a = { S a , H ˉ } . (22)
Orbital angular momentum. The above argument does not apply to L \mathbf{L} L , because the averages (11) depend on L \mathbf{L} L , so averaging does not commute with ∂ / ∂ L \partial/\partial\mathbf{L} ∂ / ∂ L . We use instead the conservation of the total angular momentum J = L + S 1 + S 2 \mathbf{J}=\mathbf{L}+\mathbf{S}_1+\mathbf{S}_2 J = L + S 1 + S 2 in both systems. Only the status of L \mathbf{L} L differs. In the full system, L = R × P \mathbf{L}=\mathbf{R}\times\mathbf{P} L = R × P is a function of the canonical variables, with the PBs (2.4), from which { L i , L j } = ϵ i j k L k \{L_i,L_j\}=\epsilon_{ijk}L_k { L i , L j } = ϵ ij k L k follows. In the OA system, L \mathbf{L} L is itself a phase-space variable, with the PBs (15) . In both cases, for any fixed unit vector e ^ \hat{\mathbf{e}} e ^ , the flow generated by e ^ ⋅ J \hat{\mathbf{e}}\cdot\mathbf{J} e ^ ⋅ J ,
d V d λ e ^ ⋅ J = { V , e ^ ⋅ J } = e ^ × V , V ∈ { R , P , S a } (full) , V ∈ { L , S a } (OA) , (23) \frac{\mathrm{d}\mathbf{V}}{\mathrm{d}\lambda_{\hat{\mathbf{e}}\cdot\mathbf{J}}}=\{\mathbf{V},\hat{\mathbf{e}}\cdot\mathbf{J}\}=\hat{\mathbf{e}}\times\mathbf{V},\qquad\mathbf{V}\in\{\mathbf{R},\mathbf{P},\mathbf{S}_a\}\ \text{(full)},\qquad\mathbf{V}\in\{\mathbf{L},\mathbf{S}_a\}\ \text{(OA)}, \tag*{(23)} d λ e ^ ⋅ J d V = { V , e ^ ⋅ J } = e ^ × V , V ∈ { R , P , S a } (full) , V ∈ { L , S a } (OA) , (23)
rigidly rotates all vectors about e ^ \hat{\mathbf{e}} e ^ [29 , 30 ] . H H H and H ˉ \bar{H} H ˉ depend on these vectors only through scalar products and magnitudes, with no fixed external direction. They are therefore invariant under these rotations, { H , J } = { H ˉ , J } = 0 \{H,\mathbf{J}\}=\{\bar{H},\mathbf{J}\}=0 { H , J } = { H ˉ , J } = 0 , and J \mathbf{J} J is conserved in both systems. It follows that
⟨ d L d t ⟩ = − ∑ a ⟨ d S a d t ⟩ = − ∑ a { S a , H ˉ } = { L , H ˉ } , (24) \left\langle\frac{\mathrm{d}\mathbf{L}}{\mathrm{d}t}\right\rangle=-\sum_a\left\langle\frac{\mathrm{d}\mathbf{S}_a}{\mathrm{d}t}\right\rangle=-\sum_a\{\mathbf{S}_a,\bar{H}\}=\{\mathbf{L},\bar{H}\}, \tag*{(24)} ⟨ d t d L ⟩ = − a ∑ ⟨ d t d S a ⟩ = − a ∑ { S a , H ˉ } = { L , H ˉ } , (24)
where the first equality uses the conservation of J \mathbf{J} J in the full non-OA system, the second uses the averaged spin EOM Eq. (22) , and the third uses the conservation of J \mathbf{J} J in the OA system.
Derivation via ansatz matching
Alternatively, H ˉ \bar{H} H ˉ can be reconstructed from the OA EOMs (2.8) alone without the knowledge of the full Hamiltonian. We take the most general scalar built from ( L , S 1 , S 2 ) (\mathbf{L},\mathbf{S}_1,\mathbf{S}_2) ( L , S 1 , S 2 ) at spin-orbit and spin-spin order that is invariant under rotations and time reversal. Since time reversal flips all angular momenta, it excludes the triple product L ⋅ ( S 1 × S 2 ) \mathbf{L}\cdot(\mathbf{S}_1\times\mathbf{S}_2) L ⋅ ( S 1 × S 2 ) . The ansatz reads
H ˉ = 1 c 2 D 3 [ k 1 L ⋅ S 1 + k 2 S 1 ⋅ S 2 + k 3 L ⋅ S 2 + k 4 ( L ⋅ S 1 ) 2 L 2 + k 5 ( L ⋅ S 1 ) ( L ⋅ S 2 ) L 2 + k 6 ( L ⋅ S 2 ) 2 L 2 ] . (25) \bar{H}=\frac{1}{c^{2}D^{3}}\left[k_{1}\mathbf{L}\cdot\mathbf{S}_{1}+k_{2}\mathbf{S}_{1}\cdot\mathbf{S}_{2}+k_{3}\mathbf{L}\cdot\mathbf{S}_{2}+k_{4}\frac{(\mathbf{L}\cdot\mathbf{S}_{1})^{2}}{L^{2}}+k_{5}\frac{(\mathbf{L}\cdot\mathbf{S}_{1})(\mathbf{L}\cdot\mathbf{S}_{2})}{L^{2}}+k_{6}\frac{(\mathbf{L}\cdot\mathbf{S}_{2})^{2}}{L^{2}}\right]. \tag*{(25)} H ˉ = c 2 D 3 1 [ k 1 L ⋅ S 1 + k 2 S 1 ⋅ S 2 + k 3 L ⋅ S 2 + k 4 L 2 ( L ⋅ S 1 ) 2 + k 5 L 2 ( L ⋅ S 1 ) ( L ⋅ S 2 ) + k 6 L 2 ( L ⋅ S 2 ) 2 ] . (25)
Being built from scalar products only, it is invariant under the rotations generated by J \mathbf{J} J , so { J , H ˉ } = 0 \{\mathbf{J},\bar{H}\}=0 { J , H ˉ } = 0 . Requiring that Eq. (16) reproduces Eqs. (2.8) gives the unique solution
k 1 = 2 σ 1 , k 2 = 1 2 , k 3 = 2 σ 2 , k 4 = − 3 4 m 2 m 1 , k 5 = − 3 2 , k 6 = − 3 4 m 1 m 2 , (26) k_{1}=2\sigma_{1},\qquad k_{2}=\frac{1}{2},\qquad k_{3}=2\sigma_{2},\qquad k_{4}=-\frac{3}{4}\frac{m_{2}}{m_{1}},\qquad k_{5}=-\frac{3}{2},\qquad k_{6}=-\frac{3}{4}\frac{m_{1}}{m_{2}}, \tag*{(26)} k 1 = 2 σ 1 , k 2 = 2 1 , k 3 = 2 σ 2 , k 4 = − 4 3 m 1 m 2 , k 5 = − 2 3 , k 6 = − 4 3 m 2 m 1 , (26)
i.e., Eq. (14) . H ˉ \bar{H} H ˉ is fixed up to an additive term which could be a function of the Casimirs L 2 L^{2} L 2 , S 1 2 S_{1}^{2} S 1 2 and S 2 2 S_{2}^{2} S 2 2 ; this term has vanishing PBs with all variables and does not affect the EOMs.
Integrability and action variables
The setup
On the 6-dimensional standard phase space (SPS) S 2 ( S 1 ) × S 2 ( S 2 ) × S 2 ( L ) S^{2}(\mathbf{S}_{1})\times S^{2}(\mathbf{S}_{2})\times S^{2}(\mathbf{L}) S 2 ( S 1 ) × S 2 ( S 2 ) × S 2 ( L ) , where the Casimirs L L L , S 1 S_{1} S 1 , and S 2 S_{2} S 2 are fixed, the OA EOMs admit the three COMs
C = ( J 2 , J z , Λ ) , (27) \mathbf{C}=\left(J^{2},J_{z},\Lambda\right), \tag*{(27)} C = ( J 2 , J z , Λ ) , (27)
with J ≡ ∥ L + S 1 + S 2 ∥ J\equiv\lVert\mathbf{L}+\mathbf{S}_{1}+\mathbf{S}_{2}\rVert J ≡ ∥ L + S 1 + S 2 ∥ and Λ = L ⋅ S 0 \Lambda=\mathbf{L}\cdot\mathbf{S}_{0} Λ = L ⋅ S 0 . We use J 2 J^{2} J 2 rather than J J J as a COM, since J 2 J^{2} J 2 is a smooth function on the whole phase space, whereas J J J is not differentiable at J = 0 J=0 J = 0 (see Sec. VI). J 2 J^{2} J 2 and J z J_{z} J z are conserved since { H ˉ , J } = 0 \{\bar{H},J\}=0 { H ˉ , J } = 0 (Sec. III C). For Λ \Lambda Λ , we use J 2 = L 2 + S 1 2 + S 2 2 + 2 L ⋅ ( S 1 + S 2 ) + 2 S 1 ⋅ S 2 J^{2}=L^{2}+S_{1}^{2}+S_{2}^{2}+2\mathbf{L}\cdot(\mathbf{S}_{1}+\mathbf{S}_{2})+2\mathbf{S}_{1}\cdot\mathbf{S}_{2} J 2 = L 2 + S 1 2 + S 2 2 + 2 L ⋅ ( S 1 + S 2 ) + 2 S 1 ⋅ S 2 to rewrite H ˉ \bar{H} H ˉ of Eq. (14) in terms of J 2 J^{2} J 2 , Λ \Lambda Λ , and the Casimirs only
H ˉ = 1 4 c 2 D 3 [ J 2 − L 2 − S 1 2 − S 2 2 + 6 Λ − 3 ν Λ 2 L 2 ] . (28) \bar{H}=\frac{1}{4c^{2}D^{3}}\left[J^{2}-L^{2}-S_{1}^{2}-S_{2}^{2}+6\Lambda-\frac{3\nu\Lambda^{2}}{L^{2}}\right]. \tag*{(28)} H ˉ = 4 c 2 D 3 1 [ J 2 − L 2 − S 1 2 − S 2 2 + 6Λ − L 2 3 ν Λ 2 ] . (28)
Since Λ \Lambda Λ and J 2 J^{2} J 2 are scalars, they commute with every component of J \mathbf{J} J , the generator of rotations [Eq. (23) ], so that J 2 J^{2} J 2 , J z J_{z} J z and Λ \Lambda Λ are in involution. Eq. (28) then gives { Λ , H ˉ } = { Λ , J 2 } ∂ H ˉ / ∂ J 2 = 0 \{\Lambda,\bar{H}\}=\{\Lambda,J^{2}\}\,\partial\bar{H}/\partial J^{2}=0 { Λ , H ˉ } = { Λ , J 2 } ∂ H ˉ / ∂ J 2 = 0 , so Λ \Lambda Λ is conserved. Hence H ˉ \bar{H} H ˉ is not an independent COM and is not included in C \mathbf{C} C . The three COMs in C \mathbf{C} C are also independent, d C 1 ∧ d C 2 ∧ d C 3 ≠ 0 dC_{1}\wedge dC_{2}\wedge dC_{3}\ne0 d C 1 ∧ d C 2 ∧ d C 3 = 0 , except at some special configurations like the (anti-)aligned configurations, spin–orbit resonances and certain other points (as discussed in Sec. VI).8 The 2PN OA system is therefore integrable [29 , 30 ] , as noted in Ref. [16 ] , where an equivalent form of Eq. (28) appears. Integrability also extends to the case m 1 = m 2 m_1=m_2 m 1 = m 2 , where Λ = Σ ⋅ ( S 1 + S 2 ) \Lambda=\boldsymbol{\Sigma}\cdot(\mathbf{S}_1+\mathbf{S}_2) Λ = Σ ⋅ ( S 1 + S 2 ) , with Σ 1 = Σ 2 ≡ Σ \Sigma_1=\Sigma_2\equiv\Sigma Σ 1 = Σ 2 ≡ Σ , remains independent of J 2 J^2 J 2 and J z J_z J z . Only the explicit expression of the third action variable (Sec. IV E 2) degenerates in this limit.
Connection between the orbit-averaged and the full 2PN constants of motion
A natural question is how the OA COMs relate to those of the full (non-OA) 2PN system. We intuitively expect that orbit averaging maps the COMs of the full system to those of the OA system. Besides H H H and J \mathbf{J} J , the full 2PN system possesses two more COMs [31 ]
L ~ 2 = L 2 + δ L 2 (29) \widetilde{L}^{2}=L^{2}+\delta L^{2} \tag*{(29)} L 2 = L 2 + δ L 2 (29)
S ~ e f f ⋅ L = S e f f ⋅ L + 1 4 S 1 ⋅ S 2 + δ 2 ( S e f f ⋅ L ) , (30) \widetilde{\mathbf{S}}_{\mathrm{eff}}\cdot\mathbf{L}=\mathbf{S}_{\mathrm{eff}}\cdot\mathbf{L}+\frac{1}{4}\mathbf{S}_{1}\cdot\mathbf{S}_{2}+\delta_{2}(\mathbf{S}_{\mathrm{eff}}\cdot\mathbf{L}), \tag*{(30)} S eff ⋅ L = S eff ⋅ L + 4 1 S 1 ⋅ S 2 + δ 2 ( S eff ⋅ L ) , (30)
where the PN orders are indicated relative to the leading term. The OA Hamiltonian (Eq. (14) ) contains only the leading spin-orbit term and the spin-spin term, the latter being smaller by a relative 0.5PN factor. The OA COMs can therefore match the full COMs only up to relative 0.5PN order, and δ L 2 \delta L^{2} δ L 2 and δ 2 \delta_{2} δ 2 must be dropped. Doing so gives:
L ~ 2 → L 2 \widetilde{L}^{2}\to L^{2} L 2 → L 2 , which is a Casimir of the OA system and is therefore conserved;
S ~ e f f ⋅ L → Λ e f f ≡ S e f f ⋅ L + S 1 ⋅ S 2 / 4 \widetilde{\mathbf{S}}_{\mathrm{eff}}\cdot\mathbf{L}\to\Lambda_{\mathrm{eff}}\equiv\mathbf{S}_{\mathrm{eff}}\cdot\mathbf{L}+\mathbf{S}_{1}\cdot\mathbf{S}_{2}/4 S eff ⋅ L → Λ eff ≡ S eff ⋅ L + S 1 ⋅ S 2 /4 , which is a COM of the OA system, since
Λ e f f = 1 2 c 2 D 3 H ‾ + 3 ν Λ 2 8 L 2 . (31) \Lambda_{\mathrm{eff}}=\frac{1}{2}c^{2}D^{3}\overline{H}+\frac{3\nu\Lambda^{2}}{8L^{2}}. \tag*{(31)} Λ eff = 2 1 c 2 D 3 H + 8 L 2 3 ν Λ 2 . (31)
Each COM of the full system thus corresponds to a COM of the OA system:
( H , J , L ~ 2 , S ~ e f f ⋅ L ) ⟶ ( H ‾ , J , L 2 , Λ e f f ) . (32) \left(H,\mathbf{J},\widetilde{L}^{2},\widetilde{\mathbf{S}}_{\mathrm{eff}}\cdot\mathbf{L}\right)\longrightarrow\left(\overline{H},\mathbf{J},L^{2},\Lambda_{\mathrm{eff}}\right). \tag*{(32)} ( H , J , L 2 , S eff ⋅ L ) ⟶ ( H , J , L 2 , Λ eff ) . (32)
Moreover, k = 1 / 4 k=1/4 k = 1/4 is the only coefficient for which S e f f ⋅ L + k S 1 ⋅ S 2 \mathbf{S}_{\mathrm{eff}}\cdot\mathbf{L}+k\mathbf{S}_{1}\cdot\mathbf{S}_{2} S eff ⋅ L + k S 1 ⋅ S 2 is conserved by the OA EOMs ((2.8)). This provides an independent consistency check between the full COM of Ref. [31 ] and the OA Hamiltonian ((14) ). Finally, Λ \Lambda Λ and Λ e f f \Lambda_{\mathrm{eff}} Λ eff are related by
Λ = 4 3 Λ e f f − 1 6 ( J 2 − L 2 − S 1 2 − S 2 2 ) . (33) \Lambda=\frac{4}{3}\Lambda_{\mathrm{eff}}-\frac{1}{6}\left(J^{2}-L^{2}-S_{1}^{2}-S_{2}^{2}\right). \tag*{(33)} Λ = 3 4 Λ eff − 6 1 ( J 2 − L 2 − S 1 2 − S 2 2 ) . (33)
Both are valid choices for the third COM for the OA system. We use Λ \Lambda Λ because its flow equation is simpler, which simplifies the computation of the action variables in Sec. IV E 2.
Defining canonical coordinates and action variables
The OA system is integrable, and its 6-dimensional phase space S 2 ( S 1 ) × S 2 ( S 2 ) × S 2 ( L ) S^2(S_1) \times S^2(S_2) \times S^2(L) S 2 ( S 1 ) × S 2 ( S 2 ) × S 2 ( L ) is compact. By the Liouville–Arnold theorem [30 ] , at fixed L L L , S 1 S_1 S 1 , and S 2 S_2 S 2 , each connected component of a regular level set, on which the COMs are independent,
C = c o n s t , (34) C = \mathrm{const}, \tag*{(34)} C = const , (34)
is then a 3-torus.
The torus is built from the flows of the COMs. The flow generated by a phase-space function h h h is encapsulated by the following evolution equation [29 , 30 ]
d g d λ h = { g , h } , (35) \frac{dg}{d\lambda_h} = \{g,h\}, \tag*{(35)} d λ h d g = { g , h } , (35)
for any phase-space function g g g , and λ h \lambda_h λ h is the flow parameter. When g g g and h h h are in involution, d g / d λ h = 0 dg/d\lambda_h = 0 d g / d λ h = 0 . Thus both g g g and h h h remain constant along the flow of the other, and these flows stay on the level set C = c o n s t C = \mathrm{const} C = const . Since H ˉ \bar{H} H ˉ is a function of the COMs, which are in mutual involution, [Eq. (28) ], the same is true for the time evolution flow,
d g d t = d g d λ H ˉ = { g , H ˉ } , (36) \frac{dg}{dt} = \frac{dg}{d\lambda_{\bar{H}}} = \{g,\bar{H}\}, \tag*{(36)} d t d g = d λ H ˉ d g = { g , H ˉ } , (36)
and we say that the torus is invariant. Since our COMs are in involution, their flows commute (Sec. 39 of [30 ] ), and they can be applied in any order without leaving the torus.
For an integrable system with n n n canonical pairs ( q i , p i ) (q_i,p_i) ( q i , p i ) , the action variables are defined as
J k = 1 2 π ∮ C k ∑ i = 1 n p i d q i , k = 1 , … , n , (37) \mathcal{J}_k = \frac{1}{2\pi}\oint_{\mathcal{C}_k}\sum_{i=1}^{n} p_i\,dq_i,\qquad k=1,\ldots,n, \tag*{(37)} J k = 2 π 1 ∮ C k i = 1 ∑ n p i d q i , k = 1 , … , n , (37)
where the homotopy classes of the loops C k \mathcal{C}_k C k generate the fundamental group of the torus. J k \mathcal{J}_k J k does not depend on the choice of loop within its class (Proposition 11.2 of Ref. [32 ] ). Here n = 3 n=3 n = 3 , and with the canonical pairs ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) of Eq. (17) ,
J k = 1 2 π ∮ C k ∑ V V z d ϕ V , V ∈ { L , S 1 , S 2 } . (38) \mathcal{J}_k = \frac{1}{2\pi}\oint_{\mathcal{C}_k}\sum_V V_z\,d\phi_V,\qquad V\in\{L,S_1,S_2\}. \tag*{(38)} J k = 2 π 1 ∮ C k V ∑ V z d ϕ V , V ∈ { L , S 1 , S 2 } . (38)
Two of these loops come directly from the flows of J z J_z J z and J 2 J^2 J 2 , which rotate the whole system rigidly about z ^ \hat{\mathbf{z}} z ^ and J ^ \hat{\mathbf{J}} J ^ [Eq. (23) ]. The third one is more complicated and involves the flow of Λ \Lambda Λ (Sec. IV E 2).
Since ϕ V \phi_V ϕ V is undefined at the poles V ∥ ± z ^ \mathbf{V}\parallel\pm\hat{\mathbf{z}} V ∥ ± z ^ , a deformation of a loop that crosses a pole shifts J k \mathcal{J}_k J k by ± V \pm V ± V (App. A). The actions are thus defined up to n L L + n S 1 S 1 + n S 2 S 2 n_L L+n_{S_1}S_1+n_{S_2}S_2 n L L + n S 1 S 1 + n S 2 S 2 , with integer n n n 's, which gives again action-angle variables (Proposition 11.3 of Ref. [32 ] ).
First two action variables
The flow generated by J 2 J^2 J 2 follows from Eq. (23) with e ^ = J ^ \hat{\mathbf{e}}=\hat{\mathbf{J}} e ^ = J ^ and the chain rule
d V d λ J 2 = { V , J 2 } = 2 J J ^ × V , V ∈ { L , S 1 , S 2 } . (39) \frac{\mathrm{d}\mathbf{V}}{\mathrm{d}\lambda_{J^2}}=\{\mathbf{V},J^2\}=2J\hat{\mathbf{J}}\times\mathbf{V},\qquad\mathbf{V}\in\{\mathbf{L},\mathbf{S}_1,\mathbf{S}_2\}. \tag*{(39)} d λ J 2 d V = { V , J 2 } = 2 J J ^ × V , V ∈ { L , S 1 , S 2 } . (39)
All three vectors thus rotate rigidly about the fixed J ^ \hat{\mathbf{J}} J ^ at angular velocity 2 J 2J 2 J , and the configuration returns to itself after λ J 2 = π / J \lambda_{J^2}=\pi/J λ J 2 = π / J . This closed loop gives the first action
J 1 = 1 2 π ∮ J 2 -flow ( L z d ϕ L + S 1 z d ϕ S 1 + S 2 z d ϕ S 2 ) . (40) \mathcal{J}_1=\frac{1}{2\pi}\oint_{J^2\text{-flow}}\left(L_z\,\mathrm{d}\phi_L+S_{1z}\,\mathrm{d}\phi_{S_1}+S_{2z}\,\mathrm{d}\phi_{S_2}\right). \tag*{(40)} J 1 = 2 π 1 ∮ J 2 -flow ( L z d ϕ L + S 1 z d ϕ S 1 + S 2 z d ϕ S 2 ) . (40)
Along this loop, each vector V ∈ { L , S 1 , S 2 } \mathbf{V}\in\{\mathbf{L},\mathbf{S}_1,\mathbf{S}_2\} V ∈ { L , S 1 , S 2 } rotates counterclockwise (when seen from outside the sphere) about J ^ \hat{\mathbf{J}} J ^ at fixed magnitude, so its tip traces a circle on the sphere of radius V V V [Fig. 1(a)]. This circle divides the sphere into two parts. We call the part that is towards J ^ \hat{\mathbf{J}} J ^ , i.e., the part containing the point V J ^ V\hat{\mathbf{J}} V J ^ , the cap . The area of the cap is 2 π V ( V − V ⋅ J ^ ) 2\pi V(V-\mathbf{V}\cdot\hat{\mathbf{J}}) 2 π V ( V − V ⋅ J ^ ) . Therefore a cap defined as such can be bigger than a hemisphere. The evaluation of ∮ V z d ϕ V \oint V_z\,\mathrm{d}\phi_V ∮ V z d ϕ V depends on the position of the poles relative to this cap, because ϕ V \phi_V ϕ V is measured about the z z z -axis and is undefined at the poles. If the north pole lies inside the cap and the south pole outside, i.e., if − z ^ ⋅ J ^ < V ^ ⋅ J ^ < z ^ ⋅ J ^ -\hat{\mathbf{z}}\cdot\hat{\mathbf{J}}<\hat{\mathbf{V}}\cdot\hat{\mathbf{J}}<\hat{\mathbf{z}}\cdot\hat{\mathbf{J}} − z ^ ⋅ J ^ < V ^ ⋅ J ^ < z ^ ⋅ J ^ , the loop goes once around the z z z -axis and ϕ V \phi_V ϕ V increases by 2 π 2\pi 2 π along it. We treat this case here, as in all panels of Figure 1 , and show in App. A that the three other cases lead to the same action, up to n L L + n S 1 S 1 + n S 2 S 2 n_L L+n_{S_1}S_1+n_{S_2}S_2 n L L + n S 1 S 1 + n S 2 S 2 , with n V ∈ Z n_V\in\mathbb{Z} n V ∈ Z .
Figure 1. Computation of ∮ V z d ϕ V \oint V_z\,\mathrm{d}\phi_V ∮ V z d ϕ V for V ∈ { L , S 1 , S 2 } \mathbf{V}\in\{\mathbf{L},\mathbf{S}_1,\mathbf{S}_2\} V ∈ { L , S 1 , S 2 } , when the north pole lies inside the cap bounded by the loop and the south pole outside (case shown in all three panels). (a). Under the flow generated by J 2 J^2 J 2 , the tip of S 1 \mathbf{S}_1 S 1 traces a circle (green) about J \mathbf{J} J . (b). Closed contour: the loop (green), completed by a meridian arc, the equator traversed backward and the same meridian arc (blue). The blue arcs do not contribute, so the loop and the contour give the same integral. (c). The same contour in the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane. Since the loop goes once around the z z z -axis, ϕ V \phi_V ϕ V runs from 0 to 2 π 2\pi 2 π along it. The shaded area equals ∮ V z d ϕ V \oint V_z\,\mathrm{d}\phi_V ∮ V z d ϕ V . In (b) and (c), the two meridian arcs are drawn slightly apart for clarity.
To evaluate ∮ V z d ϕ V \oint V_z\,d\phi_V ∮ V z d ϕ V as an area ∫ d ϕ V d V z \int d\phi_V\,dV_z ∫ d ϕ V d V z with Stokes’ theorem, the loop must bound a region on which ϕ V \phi_V ϕ V is defined everywhere. Neither region bounded by the loop qualifies, since the cap contains the north pole and its complement the south pole. We therefore close the loop C a C_a C a of Figure 1 (a) into the contour C b C_b C b of Figure 1 (b), adding a meridian arc down to the equator, the equator traversed backward, and the same meridian arc back up, with the two meridians infinitesimally close. The new contour encloses no pole, and the added arcs do not contribute, since d ϕ V = 0 d\phi_V=0 d ϕ V = 0 on a meridian and V z = 0 V_z=0 V z = 0 on the equator. Therefore
∮ C a V z d ϕ V = ∮ C b V z d ϕ V . (41) \oint_{C_a} V_z\,d\phi_V=\oint_{C_b} V_z\,d\phi_V. \tag*{(41)} ∮ C a V z d ϕ V = ∮ C b V z d ϕ V . (41)
In the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane, the loop of Figure 1 (a) becomes a curve V z ( ϕ V ) V_z(\phi_V) V z ( ϕ V ) with ϕ V \phi_V ϕ V running from 0 to 2 π 2\pi 2 π , and ∫ V z d ϕ V \int V_z\,d\phi_V ∫ V z d ϕ V is the area under this curve [Figure 1 (c)]. Since the area element on the sphere of radius V V V is V d ϕ V d V z V\,d\phi_V\,dV_z V d ϕ V d V z , this area is 1 / V 1/V 1/ V times the corresponding area on the sphere, the region between the equator and the loop. It is therefore the area of the upper hemisphere minus that of the cap, divided by V V V .9
∮ C a V z d ϕ V = 2 π V 2 − 2 π V ( V − V ⋅ J ^ ) V = 2 π V ⋅ J ^ . (42) \oint_{C_a} V_z\,d\phi_V=\frac{2\pi V^2-2\pi V\left(V-\mathbf{V}\cdot\hat{\mathbf{J}}\right)}{V}=2\pi\mathbf{V}\cdot\hat{\mathbf{J}}. \tag*{(42)} ∮ C a V z d ϕ V = V 2 π V 2 − 2 π V ( V − V ⋅ J ^ ) = 2 π V ⋅ J ^ . (42)
Inserting Eq. (42) into Eq. (40) gives
J 1 = L ⋅ J ^ + S 1 ⋅ J ^ + S 2 ⋅ J ^ = J . (43) \mathcal{J}_1=\mathbf{L}\cdot\hat{\mathbf{J}}+\mathbf{S}_1\cdot\hat{\mathbf{J}}+\mathbf{S}_2\cdot\hat{\mathbf{J}}=J. \tag*{(43)} J 1 = L ⋅ J ^ + S 1 ⋅ J ^ + S 2 ⋅ J ^ = J . (43)
The other cases occur in practice. For instance, when L ≫ S 1 , S 2 L\gg S_1,S_2 L ≫ S 1 , S 2 , L \mathbf{L} L is close to J \mathbf{J} J and its cap is small, so that it generally contains neither pole. Going from the case when the loop contains a pole to the one where it does not requires a pole to cross the loop, which we discuss in Sec. IV C. In App. A, we show that, for any position of the poles (with respect to the loop of Figure 1 (a)), ∮ V z d ϕ V = 2 π [ V ⋅ J ^ + ( n N + n S − 1 ) V ] \oint V_z\,d\phi_V=2\pi[\mathbf{V}\cdot\hat{\mathbf{J}}+(n_N+n_S-1)V] ∮ V z d ϕ V = 2 π [ V ⋅ J ^ + ( n N + n S − 1 ) V ] , where n N n_N n N (n S n_S n S ) is 1 if the north (south) pole lies inside the cap and 0 otherwise. The other cases thus shift J 1 \mathcal{J}_1 J 1 by an integer combination of L L L , S 1 S_1 S 1 , and S 2 S_2 S 2 . So Eq. (43) holds for any orientation of the z z z -axis, up to integer multiples of L L L , S 1 S_1 S 1 and S 2 S_2 S 2 .
The second action follows from the flow of J z J_z J z , which by Eq. (23) with e ^ = z ^ \hat{\mathbf{e}}=\hat{\mathbf{z}} e ^ = z ^ rotates all vectors rigidly about the z z z -axis, d V / d λ J z = z ^ × V d\mathbf{V}/d\lambda_{J_z}=\hat{\mathbf{z}}\times\mathbf{V} d V / d λ J z = z ^ × V . Along this loop, each tip traces a circle about the z z z -axis itself, so V z V_z V z is constant and ϕ V \phi_V ϕ V increases by 2 π 2\pi 2 π . Here we have ( n N , n S ) = ( 1 , 0 ) (n_N,n_S)=(1,0) ( n N , n S ) = ( 1 , 0 ) . Each term then contributes 2 π V z 2\pi V_z 2 π V z , and
J 2 = L z + S 1 z + S 2 z = J z . (44) \mathcal{J}_2=L_z+S_{1z}+S_{2z}=J_z. \tag*{(44)} J 2 = L z + S 1 z + S 2 z = J z . (44)
The third action variable
Introducing fictitious variables
The first two actions follow from rigid rotations about fixed axes, J ^ \hat{\mathbf{J}} J ^ and z ^ \hat{\mathbf{z}} z ^ , along which each integral reduces to an area on a sphere. The loop defining the third action is instead built from the Λ \Lambda Λ -flow, which is not a rigid rotation. As we will see in Sec. IV E 2, L \mathbf{L} L precesses about S 0 \mathbf{S}_0 S 0 , and each S a \mathbf{S}_a S a about L \mathbf{L} L , so each vector rotates about an axis that itself moves. Along such a motion, V z V_z V z and ϕ V \phi_V ϕ V vary in a complicated way, and we have not found a direct way to compute ∮ V z d ϕ V \oint V_z\,\mathrm{d}\phi_V ∮ V z d ϕ V , as for the previous two actions. The same difficulty arises in Ref. [18 ] ,10 where it is bypassed by introducing fictitious variables. We follow this method and refer to Ref. [18 ] for details.
The method rests on the following observation. When V \mathbf{V} V is rotated by a small angle d α \mathrm{d}\alpha d α about z ^ \hat{\mathbf{z}} z ^ , ϕ V \phi_V ϕ V increases by d α \mathrm{d}\alpha d α , and the integrand is V z d ϕ V = ( z ^ ⋅ V ) d α V_z\,\mathrm{d}\phi_V=(\hat{\mathbf{z}}\cdot\mathbf{V})\,\mathrm{d}\alpha V z d ϕ V = ( z ^ ⋅ V ) d α . In words, the integrand is the component of V \mathbf{V} V along the rotation axis times the rotation angle. For a rotation about any other axis, V z d ϕ V V_z\,\mathrm{d}\phi_V V z d ϕ V no longer takes this form. If instead V \mathbf{V} V is written as V = r × p \mathbf{V}=\mathbf{r}\times\mathbf{p} V = r × p , with canonical pairs ( r , p ) (\mathbf{r},\mathbf{p}) ( r , p ) , the integrand of Eq. (37) becomes p ⋅ d r \mathbf{p}\cdot\mathrm{d}\mathbf{r} p ⋅ d r , which takes this form for every axis. Indeed, a rotation by d α \mathrm{d}\alpha d α about an arbitrary axis n ^ \hat{\mathbf{n}} n ^ gives d r = d α n ^ × r \mathrm{d}\mathbf{r}=\mathrm{d}\alpha\,\hat{\mathbf{n}}\times\mathbf{r} d r = d α n ^ × r , so that
p ⋅ d r = ( n ^ ⋅ V ) d α . (45) \mathbf{p}\cdot\mathrm{d}\mathbf{r}=(\hat{\mathbf{n}}\cdot\mathbf{V})\,\mathrm{d}\alpha. \tag*{(45)} p ⋅ d r = ( n ^ ⋅ V ) d α . (45)
This form makes the action computation feasible; see Eqs. (28)–(33) of Ref. [18 ] .
For each of the angular momenta V ∈ { L , S 1 , S 2 } \mathbf{V}\in\{\mathbf{L},\mathbf{S}_1,\mathbf{S}_2\} V ∈ { L , S 1 , S 2 } , which we call the physical variables in contrast to the fictitious ones, we introduce a fictitious position r V \mathbf{r}_V r V and a fictitious momentum p V \mathbf{p}_V p V such that
V = r V × p V , (46) \mathbf{V}=\mathbf{r}_V\times\mathbf{p}_V , \tag*{(46)} V = r V × p V , (46)
with the canonical PBs
{ r V i , p V ′ j } = δ V V ′ δ i j , (47) \{r_{V i},p_{V' j}\}=\delta_{V V'}\delta_{ij}, \tag*{(47)} { r V i , p V ′ j } = δ V V ′ δ ij , (47)
and all other PBs between these variables being zero apart from the ones connected via anti-symmetry.11 These nine canonical pairs span an 18-dimensional extended phase space (EPS), as opposed to the standard phase space (SPS) of Sec. III A. With Eq. (46) , the PBs (47) of the fictitious variables imply the SO(3) PBs (15) of the physical variables. Any function of the physical variables, in particular H ˉ \bar{H} H ˉ , therefore has the same dynamics in the SPS and in the EPS. In the EPS, the action integrand is ∑ V p V ⋅ d r V \sum_{\mathbf{V}}\mathbf{p}_V\cdot\mathrm{d}\mathbf{r}_V ∑ V p V ⋅ d r V .
Along the flow of any function F ( L , S 1 , S 2 ) F(\mathbf{L},\mathbf{S}_1,\mathbf{S}_2) F ( L , S 1 , S 2 ) , the PBs (47) and the chain rule applied to Eq. (46) give
d r V d λ F = ∂ F ∂ V × r V , d p V d λ F = ∂ F ∂ V × p V . (48) \frac{\mathrm{d}\mathbf{r}_V}{\mathrm{d}\lambda_F}=\frac{\partial F}{\partial\mathbf{V}}\times\mathbf{r}_V,\qquad
\frac{\mathrm{d}\mathbf{p}_V}{\mathrm{d}\lambda_F}=\frac{\partial F}{\partial\mathbf{V}}\times\mathbf{p}_V . \tag*{(48)} d λ F d r V = ∂ V ∂ F × r V , d λ F d p V = ∂ V ∂ F × p V . (48)
The fictitious vectors r V \mathbf{r}_V r V and p V \mathbf{p}_V p V thus obey the same equation as V \mathbf{V} V [Eq. (16) ]: at each instant, r V \mathbf{r}_V r V , p V \mathbf{p}_V p V , and V \mathbf{V} V undergo the same rotation, with angular velocity ∂ F / ∂ V \partial F/\partial\mathbf{V} ∂ F / ∂ V . Their lengths and mutual angles are therefore preserved, and r V \mathbf{r}_V r V and p V \mathbf{p}_V p V remain in the plane orthogonal to V \mathbf{V} V . Applying Eq. (45) to each pair, with n ^ d α = ( ∂ F / ∂ V ) d λ F \hat{\mathbf{n}}\,d\alpha= \left(\partial F/\partial\mathbf{V}\right)d\lambda_F n ^ d α = ( ∂ F / ∂ V ) d λ F , gives
∑ V p V ⋅ d r V = ∑ V V ⋅ ∂ F ∂ V d λ F . (49) \sum_{\mathbf{V}} \mathbf{p}_{\mathbf{V}} \cdot d\mathbf{r}_{\mathbf{V}} = \sum_{\mathbf{V}} \mathbf{V} \cdot\frac{\partial F}{\partial\mathbf{V}} d\lambda_F. \tag*{(49)} V ∑ p V ⋅ d r V = V ∑ V ⋅ ∂ V ∂ F d λ F . (49)
For F = J 2 F=J^2 F = J 2 , ∂ F / ∂ V = 2 J \partial F/\partial\mathbf{V}=2\mathbf{J} ∂ F / ∂ V = 2 J for each V \mathbf{V} V , and the sum in Eq. (49) is ∑ V V ⋅ 2 J = 2 J 2 \sum_{\mathbf{V}}\mathbf{V}\cdot2\mathbf{J}=2J^2 ∑ V V ⋅ 2 J = 2 J 2 . Since the J 2 J^2 J 2 -loop closes after λ J 2 = π / J \lambda_{J^2}=\pi/J λ J 2 = π / J , we obtain 2 π J 1 = 2 J 2 π / J 2\pi\mathcal{J}_1=2J^2\pi/J 2 π J 1 = 2 J 2 π / J , i.e., J 1 = J \mathcal{J}_1=J J 1 = J [Eq. (43) ], which provides a check of the method.
Three points are worth mentioning. First, the fictitious variables are not unique. Indeed, for a given V \mathbf{V} V , any pair ( r V , p V ) (\mathbf{r}_{\mathbf{V}},\mathbf{p}_{\mathbf{V}}) ( r V , p V ) lying in the plane orthogonal to V \mathbf{V} V and satisfying r V × p V = V \mathbf{r}_{\mathbf{V}}\times\mathbf{p}_{\mathbf{V}}=\mathbf{V} r V × p V = V is admissible. This freedom affects the actions only through an integer combination of the Casimirs, n L L + n S 1 S 1 + n S 2 S 2 n_L L+n_{S_1}S_1+n_{S_2}S_2 n L L + n S 1 S 1 + n S 2 S 2 , as shown below [Eq. (52) ].
Second, the EPS and SPS integrands differ, because the EPS integrand p V ⋅ d r V \mathbf{p}_{\mathbf{V}}\cdot d\mathbf{r}_{\mathbf{V}} p V ⋅ d r V is sensitive to a motion that the physical variables cannot see. Indeed, by the cross product structure of Eq. (16) , the component of ∂ F / ∂ V \partial F/\partial\mathbf{V} ∂ F / ∂ V along V \mathbf{V} V leaves V \mathbf{V} V fixed. By Eq. (48) , however, it rotates r V \mathbf{r}_{\mathbf{V}} r V and p V \mathbf{p}_{\mathbf{V}} p V (which are orthogonal to V \mathbf{V} V ) about V \mathbf{V} V . Using Eq. (45) , such a rotation by d α d\alpha d α gives p V ⋅ d r V = V d α \mathbf{p}_{\mathbf{V}}\cdot d\mathbf{r}_{\mathbf{V}}=V\,d\alpha p V ⋅ d r V = V d α , whereas V z d ϕ V = 0 V_z\,d\phi_{\mathbf{V}}=0 V z d ϕ V = 0 . To separate this rotation from the motion of V \mathbf{V} V , we measure the orientation of the pair by the angle ψ V \psi_{\mathbf{V}} ψ V such that
r ^ V = cos ψ V e ^ θ + sin ψ V e ^ ϕ , (50) \hat{\mathbf{r}}_{\mathbf{V}}=\cos\psi_{\mathbf{V}}\,\hat{\mathbf{e}}_\theta+\sin\psi_{\mathbf{V}}\,\hat{\mathbf{e}}_\phi, \tag*{(50)} r ^ V = cos ψ V e ^ θ + sin ψ V e ^ ϕ , (50)
where e ^ θ \hat{\mathbf{e}}_\theta e ^ θ and e ^ ϕ \hat{\mathbf{e}}_\phi e ^ ϕ are unit polar and azimuthal vectors corresponding to the polar and azimuthal angles of V \mathbf{V} V . Along any flow of the form of Eq. (48) , we show in App. A that
p V ⋅ d r V = V z d ϕ V + V d ψ V . (51) \mathbf{p}_{\mathbf{V}}\cdot d\mathbf{r}_{\mathbf{V}}=V_z\,d\phi_{\mathbf{V}}+V\,d\psi_{\mathbf{V}}. \tag*{(51)} p V ⋅ d r V = V z d ϕ V + V d ψ V . (51)
Along a loop that is closed in the EPS, V \mathbf{V} V , r V \mathbf{r}_{\mathbf{V}} r V , and p V \mathbf{p}_{\mathbf{V}} p V all return to their initial values, so ψ V \psi_{\mathbf{V}} ψ V changes by 2 π n V 2\pi n_{\mathbf{V}} 2 π n V , with n V n_{\mathbf{V}} n V an integer, and
∮ p V ⋅ d r V = ∮ V z d ϕ V + 2 π n V V . (52) \oint\mathbf{p}_{\mathbf{V}}\cdot d\mathbf{r}_{\mathbf{V}}=\oint V_z\,d\phi_{\mathbf{V}}+2\pi n_{\mathbf{V}}V. \tag*{(52)} ∮ p V ⋅ d r V = ∮ V z d ϕ V + 2 π n V V . (52)
Summing over V \mathbf{V} V , the action computed in the EPS equals the SPS action up to n L L + n S 1 S 1 + n S 2 S 2 n_L L+n_{S_1}S_1+n_{S_2}S_2 n L L + n S 1 S 1 + n S 2 S 2 , as mentioned above in Sec. IV C.
Third, the EPS is integrable as well, so the loops used there lie on tori, as in the SPS. Since the EPS is 18-dimensional, this requires nine independent COMs in involution. Six of them are J 2 J^2 J 2 , J z J_z J z , Λ \Lambda Λ , L 2 L^2 L 2 , S 1 2 S_1^2 S 1 2 , and S 2 2 S_2^2 S 2 2 , where the last three are Casimirs in the SPS but ordinary COMs in the EPS.12 These six COMs are functions of the physical variables only, and they leave the shape of each pair ( r V , p V ) (\mathbf{r}_{\mathbf{V}},\mathbf{p}_{\mathbf{V}}) ( r V , p V ) free, e.g., the length r V r_{\mathbf{V}} r V and the product r V ⋅ p V \mathbf{r}_{\mathbf{V}}\cdot\mathbf{p}_{\mathbf{V}} r V ⋅ p V , so one more COM per pair is needed. Along the flow of any function of the physical variables, r V \mathbf{r}_{\mathbf{V}} r V and p V \mathbf{p}_{\mathbf{V}} p V only rotate rigidly [Eq. (48) ], so their lengths are unchanged. The three quantities r V 2 + p V 2 r_{\mathbf{V}}^2+p_{\mathbf{V}}^2 r V 2 + p V 2 therefore commute with the six COMs above, and with each other since they involve different pairs, and they are independent of them. Fixing r V 2 + p V 2 r_V^2 + p_V^2 r V 2 + p V 2 bounds r V r_V r V and p V p_V p V , so the level sets are compact and, by the Liouville–Arnold theorem, are tori.13
The third action variable
To compute J 3 \mathcal{J}_3 J 3 , we need a third closed loop on the 3-torus of the SPS, i.e., on the level set of ( J 2 , J z , Λ ) (J^2, J_z, \Lambda) ( J 2 , J z , Λ ) , independent of the J 2 J^2 J 2 - and J z J_z J z -loops. We build it from flows in the EPS, where it is easier to evaluate, and Eq. (52) relates the result to the SPS action.
Since J 2 J^2 J 2 and J z J_z J z generate rigid rotations, only the Λ \Lambda Λ -flow changes the mutual angles of the triad ( L , S 1 , S 2 ) (\mathbf{L}, \mathbf{S}_1, \mathbf{S}_2) ( L , S 1 , S 2 ) , so a loop containing a Λ \Lambda Λ -flow segment is independent of the first two. Such a segment does not close by itself, and we complete it with the flows of J 2 J^2 J 2 , L 2 L^2 L 2 , S 1 2 S_1^2 S 1 2 , and S 2 2 S_2^2 S 2 2 . All these COMs are in involution, so their flows commute and keep the system on the same level set, and they can therefore be applied one after the other.
This construction follows Ref. [18 ] , where the fifth action of the non-OA 1.5PN system, the analog of our J 3 \mathcal{J}_3 J 3 , is built from the flow of L ⋅ S e f f \mathbf{L} \cdot\mathbf{S}_{\rm eff} L ⋅ S eff . Since Λ = L ⋅ S 0 \Lambda= \mathbf{L} \cdot\mathbf{S}_0 Λ = L ⋅ S 0 can be obtained from L ⋅ S e f f \mathbf{L} \cdot\mathbf{S}_{\rm eff} L ⋅ S eff by the replacement σ a → Σ a \sigma_a \to\Sigma_a σ a → Σ a (§ 2 A), the two flows have the same structure. From Eqs. (35) and (15) , the Λ \Lambda Λ -flow of the physical vectors reads
d L d λ Λ = S 0 × L , (53) \frac{d\mathbf{L}}{d\lambda_\Lambda} = \mathbf{S}_0 \times\mathbf{L}, \tag*{(53)} d λ Λ d L = S 0 × L , (53)
d S a d λ Λ = Σ a L × S a , (54) \frac{d\mathbf{S}_a}{d\lambda_\Lambda} = \Sigma_a \mathbf{L} \times\mathbf{S}_a, \tag*{(54)} d λ Λ d S a = Σ a L × S a , (54)
and, by Eq. (48) , each fictitious pair rotates in the same way as its physical vector, i.e., ( r L , p L ) (\mathbf{r}_L,\mathbf{p}_L) ( r L , p L ) about S 0 \mathbf{S}_0 S 0 , and ( r S a , p S a ) (\mathbf{r}_{S_a},\mathbf{p}_{S_a}) ( r S a , p S a ) about Σ a L \Sigma_a\mathbf{L} Σ a L . Apart from the replacement σ a → Σ a \sigma_a \to\Sigma_a σ a → Σ a , these flows are given by Eq. (29) of Ref. [18 ] , and the lengthy calculations of that reference simply carry over. We therefore only give the main steps and the final result.
The idea is to leave a point and come back to it using only these flows. This takes three steps
(i) Under the Λ \Lambda Λ -flow, the shape of the triad goes through one full cycle after a flow amount Δ λ Λ \Delta\lambda_\Lambda Δ λ Λ , so that the mutual scalar products and the triple product of ( L , S 1 , S 2 ) (\mathbf{L},\mathbf{S}_1,\mathbf{S}_2) ( L , S 1 , S 2 ) are back to their initial values. The final triad is therefore a rigid rotation of the initial one. Since Λ \Lambda Λ commutes with all components of J \mathbf{J} J , the Λ \Lambda Λ -flow leaves J \mathbf{J} J fixed, and this rotation is about J ^ \hat{\mathbf{J}} J ^ , by some angle Δ Φ L \Delta\Phi_L Δ Φ L .
(ii) The J 2 J^2 J 2 -flow rotates all vectors about J ^ \hat{\mathbf{J}} J ^ at angular velocity 2 J 2J 2 J . It brings the physical vectors back to their initial position after the flow amount
Δ λ J 2 = − Δ Φ L 2 J . (55) \Delta\lambda_{J^2} = -\frac{\Delta\Phi_L}{2J}. \tag*{(55)} Δ λ J 2 = − 2 J Δ Φ L . (55)
(iii) Each fictitious pair is still rotated about V \mathbf{V} V by some angle Δ ψ V \Delta\psi_V Δ ψ V . The V 2 V^2 V 2 -flow turns it back at angular velocity 2 V 2V 2 V without moving the physical vectors, after the flow amount
Δ λ V 2 = − Δ ψ V 2 V . (56) \Delta\lambda_{V^2}=-\frac{\Delta\psi_V}{2V}. \tag*{(56)} Δ λ V 2 = − 2 V Δ ψ V . (56)
The loop is then closed in the EPS.
The third action is the sum of the contributions of the five segments. Along the Λ \Lambda Λ -segment, the integrand of the action, computed with Eq. (49) for F = Λ F=\Lambda F = Λ , is constant,
L ⋅ S 0 + ∑ a S a ⋅ L = 2 Λ , (57) \mathbf{L}\cdot\mathbf{S}_0+\sum_a\mathbf{S}_a\cdot\mathbf{L}=2\Lambda, \tag*{(57)} L ⋅ S 0 + a ∑ S a ⋅ L = 2Λ , (57)
and this segment contributes 2 Λ Δ λ Λ 2\Lambda\,\Delta\lambda_\Lambda 2Λ Δ λ Λ to 2 π J 3 2\pi\mathcal{J}_3 2 π J 3 . The other generators F ∈ { J 2 , L 2 , S 1 2 , S 2 2 } F\in\{J^2,L^2,S_1^2,S_2^2\} F ∈ { J 2 , L 2 , S 1 2 , S 2 2 } are also quadratic, and each contributes 2 F Δ λ F 2F\,\Delta\lambda_F 2 F Δ λ F in the same way.14 Hence, as in eq:27 and (35) of Ref. [18 ] ,
J 3 = 1 π ( Λ Δ λ Λ + J 2 Δ λ J 2 + L 2 Δ λ L 2 + S 1 2 Δ λ S 1 2 + S 2 2 Δ λ S 2 2 ) . (58) \mathcal{J}_3=\frac{1}{\pi}\left(\Lambda\,\Delta\lambda_\Lambda+J^2\,\Delta\lambda_{J^2}+L^2\,\Delta\lambda_{L^2}+S_1^2\,\Delta\lambda_{S_1^2}+S_2^2\,\Delta\lambda_{S_2^2}\right). \tag*{(58)} J 3 = π 1 ( Λ Δ λ Λ + J 2 Δ λ J 2 + L 2 Δ λ L 2 + S 1 2 Δ λ S 1 2 + S 2 2 Δ λ S 2 2 ) . (58)
The L 2 L^2 L 2 - and S a 2 S_a^2 S a 2 -segments deserve a comment, since they leave the physical vectors fixed. Each of them rotates a fictitious pair about its physical vector, and therefore contributes only through the term V d ψ V V\,\mathrm{d}\psi_V V d ψ V of Eq. (51) . Their role is to undo the rotation of each pair about V \mathbf{V} V accumulated along the Λ \Lambda Λ - and J 2 J^2 J 2 -segments, so that ψ V \psi_V ψ V returns to its initial value up to whole turns. Equation (58) , obtained in the EPS, therefore equals the action defined in the SPS by Eq. (38) , up to n L L + n S 1 S 1 + n S 2 S 2 n_L L+n_{S_1}S_1+n_{S_2}S_2 n L L + n S 1 S 1 + n S 2 S 2 with integer n V n_V n V 's [Eq. (52) ].
The evaluation of J 3 \mathcal{J}_3 J 3 thus reduces to the computation of the five flow amounts, derived in App. C. At fixed J 2 J^2 J 2 , Λ \Lambda Λ , and Casimirs, the mutual scalar products of the triad are affine in the single variable f = S 1 ⋅ S 2 / ( Σ 1 − Σ 2 ) f=\mathbf{S}_1\cdot\mathbf{S}_2/(\Sigma_1-\Sigma_2) f = S 1 ⋅ S 2 / ( Σ 1 − Σ 2 ) , which oscillates between two roots f 1 < f 2 f_1<f_2 f 1 < f 2 of a cubic polynomial. Δ λ Λ \Delta\lambda_\Lambda Δ λ Λ is the period of this oscillation, given by a complete elliptic integral of the first kind [Eq. (138) ]. The other four amounts follow from the angles by which L \mathbf{L} L and the fictitious pairs rotate during one period, and involve complete elliptic integrals of the third kind [Eqs. (144) –(C22)].
These expressions degenerate for m 1 = m 2 m_1=m_2 m 1 = m 2 . In this limit, Σ 1 = Σ 2 \Sigma_1=\Sigma_2 Σ 1 = Σ 2 , and both spins precess about L \mathbf{L} L at the same rate under the Λ \Lambda Λ -flow [Eq. (54) ], so that S 1 ⋅ S 2 \mathbf{S}_1\cdot\mathbf{S}_2 S 1 ⋅ S 2 is constant along it. The variable f f f used in App. C [Eq. (122) ] is then undefined, and the leading coefficient of its cubic [Eq. (129) ] vanishes. The system remains integrable, but this limit requires a different parametrization of the mutual angles, which we do not pursue here.
Frequencies and angle variables
Computing the frequencies
In action-angle variables ( J i , θ i ) (\mathcal{J}_i,\theta_i) ( J i , θ i ) , the equations of motion read J ˙ i = 0 \dot{\mathcal{J}}_i=0 J ˙ i = 0 and θ ˙ i = ω i ≡ ∂ H ˉ / ∂ J i \dot{\theta}_i=\omega_i\equiv\partial\bar{H}/\partial\mathcal{J}_i θ ˙ i = ω i ≡ ∂ H ˉ / ∂ J i , so that the motion on each torus is characterized by three frequencies, instead of five for the non-OA 1.5PN system of [18 ] . By Eq. (28) , H ˉ \bar{H} H ˉ depends on the COMs C = ( J 2 , J z , Λ ) \mathbf{C}=(J^2,J_z,\Lambda) C = ( J 2 , J z , Λ ) only through J 2 J^2 J 2 and Λ \Lambda Λ , with the Casimirs L L L , S 1 S_1 S 1 , and S 2 S_2 S 2 as fixed parameters. The frequencies can be computed from the chain rule
ω i = ∑ j = 1 3 ∂ H ˉ ∂ C j ∂ C j ∂ J i . (59) \omega_i=\sum_{j=1}^{3}\frac{\partial\bar{H}}{\partial C_j}\frac{\partial C_j}{\partial\mathcal{J}_i}. \tag*{(59)} ω i = j = 1 ∑ 3 ∂ C j ∂ H ˉ ∂ J i ∂ C j . (59)
The first factor in the sum, ∂ H ˉ / ∂ C j \partial\bar{H}/\partial C_j ∂ H ˉ / ∂ C j , is evaluated by taking the partial derivatives of the Hamiltonian in Eq. (28) , yielding
∂ H ˉ ∂ J 2 = 1 4 c 2 D 3 , ∂ H ˉ ∂ J z = 0 , ∂ H ˉ ∂ Λ = 3 2 c 2 D 3 ( 1 − ν Λ L 2 ) . (60) \frac{\partial\bar{H}}{\partial J^2}=\frac{1}{4c^2D^3},\qquad\frac{\partial\bar{H}}{\partial J_z}=0,\qquad\frac{\partial\bar{H}}{\partial\Lambda}=\frac{3}{2c^2D^3}\left(1-\frac{\nu\Lambda}{L^2}\right). \tag*{(60)} ∂ J 2 ∂ H ˉ = 4 c 2 D 3 1 , ∂ J z ∂ H ˉ = 0 , ∂ Λ ∂ H ˉ = 2 c 2 D 3 3 ( 1 − L 2 ν Λ ) . (60)
The second factor requires the inverse of the Jacobian M i j = ∂ J i / ∂ C j \mathcal{M}_{ij}=\partial\mathcal{J}_i/\partial C_j M ij = ∂ J i / ∂ C j . Since J 1 = J = ( J 2 ) 1 / 2 \mathcal{J}_1=J=(J^2)^{1/2} J 1 = J = ( J 2 ) 1/2 , J 2 = J z \mathcal{J}_2=J_z J 2 = J z , and among the COMs, J 3 \mathcal{J}_3 J 3 depends only on J 2 J^2 J 2 and Λ \Lambda Λ [Eq. (58) ], M \mathcal{M} M is lower triangular
M = ∂ J ∂ C = ( 1 2 J 0 0 0 1 0 ∂ J 3 ∂ J 2 0 ∂ J 3 ∂ Λ ) , det M = 1 2 J ∂ J 3 ∂ Λ . (61) \mathcal{M}=\frac{\partial\mathcal{J}}{\partial\mathbf{C}}=\begin{pmatrix}\frac{1}{2J} & 0 & 0 \\ 0 & 1 & 0 \\ \frac{\partial\mathcal{J}_3}{\partial J^2} & 0 & \frac{\partial\mathcal{J}_3}{\partial\Lambda}\end{pmatrix},\qquad\det\mathcal{M}=\frac{1}{2J}\frac{\partial\mathcal{J}_3}{\partial\Lambda}. \tag*{(61)} M = ∂ C ∂ J = 2 J 1 0 ∂ J 2 ∂ J 3 0 1 0 0 0 ∂ Λ ∂ J 3 , det M = 2 J 1 ∂ Λ ∂ J 3 . (61)
It is invertible whenever J ≠ 0 J\ne0 J = 0 and ∂ J 3 / ∂ Λ \partial\mathcal{J}_3/\partial\Lambda ∂ J 3 / ∂ Λ is finite and non-zero. Its inverse is then
M − 1 = ∂ C ∂ J = ( 2 J 0 0 0 1 0 − 2 J ∂ J 3 / ∂ J 2 ∂ J 3 / ∂ Λ 0 ( ∂ J 3 ∂ Λ ) − 1 ) , (62) \mathcal{M}^{-1}=\frac{\partial\mathbf{C}}{\partial\mathcal{J}}=\begin{pmatrix}2J & 0 & 0 \\ 0 & 1 & 0 \\ -2J\frac{\partial\mathcal{J}_3/\partial J^2}{\partial\mathcal{J}_3/\partial\Lambda} & 0 & \left(\frac{\partial\mathcal{J}_3}{\partial\Lambda}\right)^{-1}\end{pmatrix}, \tag*{(62)} M − 1 = ∂ J ∂ C = 2 J 0 − 2 J ∂ J 3 / ∂ Λ ∂ J 3 / ∂ J 2 0 1 0 0 0 ( ∂ Λ ∂ J 3 ) − 1 , (62)
so that Eq. (59) gives the three frequencies of the system,
ω 1 = 2 J ( ∂ H ˉ ∂ J 2 − ∂ H ˉ ∂ Λ ∂ J 3 / ∂ J 2 ∂ J 3 / ∂ Λ ) , (63) \omega_1=2J\left(\frac{\partial\bar{H}}{\partial J^2}-\frac{\partial\bar{H}}{\partial\Lambda}\frac{\partial\mathcal{J}_3/\partial J^2}{\partial\mathcal{J}_3/\partial\Lambda}\right), \tag*{(63)} ω 1 = 2 J ( ∂ J 2 ∂ H ˉ − ∂ Λ ∂ H ˉ ∂ J 3 / ∂ Λ ∂ J 3 / ∂ J 2 ) , (63)
ω 2 = 0 , (64) \omega_2=0, \tag*{(64)} ω 2 = 0 , (64)
ω 3 = ∂ H ˉ ∂ Λ ( ∂ J 3 ∂ Λ ) − 1 . (65) \omega_3=\frac{\partial\bar{H}}{\partial\Lambda}\left(\frac{\partial\mathcal{J}_3}{\partial\Lambda}\right)^{-1}. \tag*{(65)} ω 3 = ∂ Λ ∂ H ˉ ( ∂ Λ ∂ J 3 ) − 1 . (65)
The derivatives of J 3 \mathcal{J}_3 J 3 can be obtained without explicitly differentiating the elliptic integrals of App. C. We instead use the structure of the loop of Sec. IV E 2: each of its segments is generated by a function F ∈ { Λ , J 2 , L 2 , S a 2 , S 2 2 } F \in\{\Lambda, J^2, L^2, S_a^2, S_2^2\} F ∈ { Λ , J 2 , L 2 , S a 2 , S 2 2 } over a flow amount Δ λ F \Delta\lambda_F Δ λ F . When J 2 J^2 J 2 and Λ \Lambda Λ are varied independently, at fixed J z J_z J z and Casimirs, the loop moves to a neighboring torus, and each segment contributes Δ λ F δ F \Delta\lambda_F\,\delta F Δ λ F δ F to the first-order variation of the action (see Appendix B for a proof for the case when F = Λ F=\Lambda F = Λ is varied)
δ ( 2 π J 3 ) = ∑ F Δ λ F δ F . (66) \delta(2\pi\mathcal{J}_3)=\sum_F\Delta\lambda_F\,\delta F. \tag*{(66)} δ ( 2 π J 3 ) = F ∑ Δ λ F δ F . (66)
At fixed Casimirs, δ ( L 2 ) = δ ( S a 2 ) = 0 \delta(L^2)=\delta(S_a^2)=0 δ ( L 2 ) = δ ( S a 2 ) = 0 , so that only the Λ \Lambda Λ - and J 2 J^2 J 2 -segments contribute. Since J 3 \mathcal{J}_3 J 3 is a function of J 2 J^2 J 2 and Λ \Lambda Λ , we also have δ J 3 = ( ∂ J 3 / ∂ J 2 ) δ ( J 2 ) + ( ∂ J 3 / ∂ Λ ) δ Λ \delta\mathcal{J}_3=(\partial\mathcal{J}_3/\partial J^2)\,\delta(J^2)+(\partial\mathcal{J}_3/\partial\Lambda)\,\delta\Lambda δ J 3 = ( ∂ J 3 / ∂ J 2 ) δ ( J 2 ) + ( ∂ J 3 / ∂ Λ ) δ Λ . Identifying the coefficients of δ ( J 2 ) \delta(J^2) δ ( J 2 ) and δ Λ \delta\Lambda δ Λ in Eq. (66) gives
∂ J 3 ∂ Λ ∣ J 2 = Δ λ Λ 2 π , ∂ J 3 ∂ J 2 ∣ Λ = Δ λ J 2 2 π . (67) \left.\frac{\partial\mathcal{J}_3}{\partial\Lambda}\right|_{J^2}=\frac{\Delta\lambda_\Lambda}{2\pi},\qquad\left.\frac{\partial\mathcal{J}_3}{\partial J^2}\right|_{\Lambda}=\frac{\Delta\lambda_{J^2}}{2\pi}. \tag*{(67)} ∂ Λ ∂ J 3 J 2 = 2 π Δ λ Λ , ∂ J 2 ∂ J 3 Λ = 2 π Δ λ J 2 . (67)
Inserting Eq. (67) into Eqs. (5.5a) and (65) , with Δ λ J 2 = − Δ Φ L / ( 2 J ) \Delta\lambda_{J^2}=-\Delta\Phi_L/(2J) Δ λ J 2 = − Δ Φ L / ( 2 J ) [Eq. (4.29)], gives
ω 1 = 2 J ∂ H ˉ ∂ J 2 + Δ Φ L Δ λ Λ ∂ H ˉ ∂ Λ , ω 3 = 2 π Δ λ Λ ∂ H ˉ ∂ Λ , (68) \omega_1=2J\frac{\partial\bar{H}}{\partial J^2}+\frac{\Delta\Phi_L}{\Delta\lambda_\Lambda}\frac{\partial\bar{H}}{\partial\Lambda},\qquad\omega_3=\frac{2\pi}{\Delta\lambda_\Lambda}\frac{\partial\bar{H}}{\partial\Lambda}, \tag*{(68)} ω 1 = 2 J ∂ J 2 ∂ H ˉ + Δ λ Λ Δ Φ L ∂ Λ ∂ H ˉ , ω 3 = Δ λ Λ 2 π ∂ Λ ∂ H ˉ , (68)
with ∂ H ˉ / ∂ J 2 \partial\bar{H}/\partial J^2 ∂ H ˉ / ∂ J 2 and ∂ H ˉ / ∂ Λ \partial\bar{H}/\partial\Lambda ∂ H ˉ / ∂ Λ given by Eq. (60) , and Δ λ Λ \Delta\lambda_\Lambda Δ λ Λ and Δ Φ L \Delta\Phi_L Δ Φ L by Eqs. (138) and (C18).
These frequencies have a direct physical meaning. Since H ˉ = H ˉ ( J 2 , Λ ) \bar{H}=\bar{H}(J^2,\Lambda) H ˉ = H ˉ ( J 2 , Λ ) , the time derivative of any function g g g of the physical variables is
d g d t = { g , H ˉ } = ∂ H ˉ ∂ J 2 { g , J 2 } + ∂ H ˉ ∂ Λ { g , Λ } , (69) \frac{dg}{dt}=\{g,\bar{H}\}=\frac{\partial\bar{H}}{\partial J^2}\{g,J^2\}+\frac{\partial\bar{H}}{\partial\Lambda}\{g,\Lambda\}, \tag*{(69)} d t d g = { g , H ˉ } = ∂ J 2 ∂ H ˉ { g , J 2 } + ∂ Λ ∂ H ˉ { g , Λ } , (69)
with coefficients ∂ H ˉ / ∂ J 2 \partial\bar{H}/\partial J^2 ∂ H ˉ / ∂ J 2 and ∂ H ˉ / ∂ Λ \partial\bar{H}/\partial\Lambda ∂ H ˉ / ∂ Λ [Eq. (5.2)] that are constant along the motion, since they depend only on the COMs J 2 J^2 J 2 and Λ \Lambda Λ . For a mutual angle g g g of the triad, { g , J 2 } = 0 \{g,J^2\}=0 { g , J 2 } = 0 , since J 2 J^2 J 2 generates rigid rotations, and Eq. (69) reduces to
d g d t = ∂ H ˉ ∂ Λ { g , Λ } = ∂ H ˉ ∂ Λ d g d λ Λ . (70) \frac{dg}{dt}=\frac{\partial\bar{H}}{\partial\Lambda}\{g,\Lambda\}=\frac{\partial\bar{H}}{\partial\Lambda}\frac{dg}{d\lambda_\Lambda}. \tag*{(70)} d t d g = ∂ Λ ∂ H ˉ { g , Λ } = ∂ Λ ∂ H ˉ d λ Λ d g . (70)
Time and the Λ \Lambda Λ -flow parameter are thus proportional, λ Λ = ( ∂ H ˉ / ∂ Λ ) t \lambda_\Lambda=(\partial\bar{H}/\partial\Lambda)t λ Λ = ( ∂ H ˉ / ∂ Λ ) t . Since g g g returns to its initial value after Δ λ Λ \Delta\lambda_\Lambda Δ λ Λ , it is periodic in time with period T 3 = Δ λ Λ / ( ∂ H ˉ / ∂ Λ ) = 2 π / ω 3 T_3=\Delta\lambda_\Lambda/(\partial\bar{H}/\partial\Lambda)=2\pi/\omega_3 T 3 = Δ λ Λ / ( ∂ H ˉ / ∂ Λ ) = 2 π / ω 3 , so that ω 3 \omega_3 ω 3 is the nutation frequency.
To interpret ω 1 \omega_1 ω 1 , consider one nutation period T 3 T_3 T 3 . By Eq. (69) , the motion combines the Λ \Lambda Λ -flow, with λ Λ = ( ∂ H ˉ / ∂ Λ ) t \lambda_\Lambda=(\partial\bar{H}/\partial\Lambda)t λ Λ = ( ∂ H ˉ / ∂ Λ ) t , and the J 2 J^2 J 2 -flow, with λ J 2 = ( ∂ H ˉ / ∂ J 2 ) t \lambda_{J^2}=(\partial\bar{H}/\partial J^2)t λ J 2 = ( ∂ H ˉ / ∂ J 2 ) t . Since these two flows commute, their effects over T 3 T_3 T 3 add (App. § 2.3.4 ). During T 3 T_3 T 3 , λ Λ \lambda_\Lambda λ Λ increases by Δ λ Λ \Delta\lambda_\Lambda Δ λ Λ , so that the Λ \Lambda Λ -flow restores the mutual angles and rotates the configuration about J ^ \hat{\mathbf{J}} J ^ by Δ Φ L \Delta\Phi_L Δ Φ L , as we saw earlier in Sec. IV E 2. Meanwhile, λ J 2 \lambda_{J^2} λ J 2 increases by ( ∂ H ˉ / ∂ J 2 ) T 3 (\partial\bar{H}/\partial J^2)T_3 ( ∂ H ˉ / ∂ J 2 ) T 3 , and the J 2 J^2 J 2 -flow, a rotation about J ^ \hat{\mathbf{J}} J ^ at angular velocity 2 J 2J 2 J , rotates the configuration by 2 J ( ∂ H ˉ / ∂ J 2 ) T 3 2J(\partial\bar{H}/\partial J^2)T_3 2 J ( ∂ H ˉ / ∂ J 2 ) T 3 . After each nutation period, the three angular momenta are therefore rotated about J \mathbf{J} J by Δ Φ L + 2 J ( ∂ H ˉ / ∂ J 2 ) T 3 \Delta\Phi_L+2J(\partial\bar{H}/\partial J^2)T_3 Δ Φ L + 2 J ( ∂ H ˉ / ∂ J 2 ) T 3 . Dividing by T 3 T_3 T 3 gives ω 1 \omega_1 ω 1 of Eq. (68) , which is thus the precession frequency about J \mathbf{J} J averaged over one nutation period. Finally, ω 2 = 0 \omega_2=0 ω 2 = 0 [Eq. (5.5b)], since H ˉ \bar{H} H ˉ does not depend on J z J_z J z , which reflects that J \mathbf{J} J is fixed in space. The same decomposition, a uniform precession about the total angular momentum combined with a periodic nutation, holds for the analogous planet-satellite problem [17 ] .
Once we account for the scaled (reduced) length and time variables employed in Ref. [11 ] , our ω 3 \omega_3 ω 3 and ω 1 \omega_1 ω 1 coincide with their nutation and precession frequencies ω n u t \omega_{\mathrm{nut}} ω nut and ω p r e c \omega_{\mathrm{prec}} ω prec , respectively [Eqs. (59) and (60) ], obtained from their time-domain solution of the same OA EOMs.15
Since the frequencies are known in closed form, one can look for resonances, i.e., n ⋅ ω = 0 \mathbf{n} \cdot\boldsymbol{\omega}=0 n ⋅ ω = 0 for some nonzero n ∈ Z 3 \mathbf{n}\in\mathbb{Z}^3 n ∈ Z 3 . Since ω 2 = 0 \omega_2=0 ω 2 = 0 , this involves only the ratio
ω 1 ω 3 = Δ Φ L 2 π + J K ( k ) 3 π Ω Λ ( 1 − ν Λ / L 2 ) (71) \frac{\omega_1}{\omega_3}=\frac{\Delta\Phi_L}{2\pi}+\frac{J K(k)}{3\pi\Omega_\Lambda(1-\nu\Lambda/L^2)} \tag*{(71)} ω 3 ω 1 = 2 π Δ Φ L + 3 π Ω Λ ( 1 − ν Λ/ L 2 ) J K ( k ) (71)
where Δ Φ L \Delta\Phi_L Δ Φ L is the rotation of L \mathbf{L} L about J ^ \hat{\mathbf{J}} J ^ during one nutation cycle. The spin-orbit resonances of Ref. [6 ] , at which f 1 = f 2 f_1=f_2 f 1 = f 2 , are not resonances in this sense [35 ] , as pointed out by Correia [16 ] , who identified them with Cassini states. At fixed masses, spins, and L L L , the condition f 1 = f 2 f_1=f_2 f 1 = f 2 is a single equation on ( J , Λ ) (J,\Lambda) ( J , Λ ) , so the spin-orbit resonances form a one-parameter family, the boundary of the region in Figure 2 . The ratio (71) varies continuously along this family, so f 1 = f 2 f_1=f_2 f 1 = f 2 does not make it rational. The condition instead sets to zero the nutation amplitude f 2 − f 1 f_2-f_1 f 2 − f 1 in Eq. (134) , so that the shape of the triad is frozen and the triad precesses rigidly about J \mathbf{J} J .
Figure 2. Physical region of the ( J , Λ ) (J,\Lambda) ( J , Λ ) plane for m 1 = 1 m_1=1 m 1 = 1 , m 2 = 0.6 m_2=0.6 m 2 = 0.6 , and χ 1 = χ 2 = 0.8 \chi_1=\chi_2=0.8 χ 1 = χ 2 = 0.8 , for which ( L − , L + ) = ( 0.061 , 3.78 ) (L_-,L_+)=(0.061,3.78) ( L − , L + ) = ( 0.061 , 3.78 ) . The color gives log 10 ( 1 − k 2 ) \log_{10}(1-k^2) log 10 ( 1 − k 2 ) , with 1 − k 2 = ( f 3 − f 2 ) / ( f 3 − f 1 ) 1-k^2=(f_3-f_2)/(f_3-f_1) 1 − k 2 = ( f 3 − f 2 ) / ( f 3 − f 1 ) . The boundary corresponds to the spin-orbit resonances, f 1 = f 2 f_1=f_2 f 1 = f 2 . The white points are the stable aligned configurations, which are corners of the region, and the red point is the up-down configuration. (a) L = 1 L=1 L = 1 and a ( 1 − e 2 ) ≃ 2.8 m a(1-e^2)\simeq2.8m a ( 1 − e 2 ) ≃ 2.8 m . The up-down configuration is unstable and lies inside the region, where 1 − k 2 1-k^2 1 − k 2 vanishes. The additional corner at J = 0 J=0 J = 0 is not an aligned configuration but corresponds to L + S 1 + S 2 = 0 L+\mathbf{S}_1+\mathbf{S}_2=0 L + S 1 + S 2 = 0 , which requires L ≤ S 1 + S 2 L\leq S_1+S_2 L ≤ S 1 + S 2 . (b) L = 15 L=15 L = 15 and a ( 1 − e 2 ) ≃ 625 m a(1-e^2)\simeq625m a ( 1 − e 2 ) ≃ 625 m . The up-down configuration is stable and is a corner of the region, J = 0 J=0 J = 0 is not reached, and 1 − k 2 ≥ 0.98 1-k^2\geq0.98 1 − k 2 ≥ 0.98 everywhere.
The angle variables and action-angle-based solution
In terms of action-angle variables, the motion is simple: the actions are constant and the angles advance linearly, θ i ( t ) = θ i ( 0 ) + ω i t \theta_i(t)=\theta_i(0)+\omega_i t θ i ( t ) = θ i ( 0 ) + ω i t [29 ] . To use this solution, we need the map between the angles and the configuration ( L , S 1 , S 2 ) (\mathbf{L},\mathbf{S}_1,\mathbf{S}_2) ( L , S 1 , S 2 ) on a given torus. Section VI B of Ref. [18 ] constructs the map from the angles to the configuration, by flowing from a fiducial point by amounts set by the angles. We follow this construction, and also derive the inverse map, which was not given there. It expresses the angles of any configuration explicitly in terms of ( L , S 1 , S 2 ) (\mathbf{L},\mathbf{S}_1,\mathbf{S}_2) ( L , S 1 , S 2 ) [Eq. (5.14)] and gives them a physical interpretation.
Fixed values of C C C and of the Casimirs define an invariant 3-torus. We locate a configuration on it by three geometric quantities. The first is the shape of the triad. It depends only on the mutual scalar products, which are affine in f f f , as it can be seen from Eq. (C3). Thus the shape of the triad is fixed by f ∈ [ f 1 , f 2 ] f\in[f_1,f_2] f ∈ [ f 1 , f 2 ] together with the handedness of the triad, s ≡ sgn [ L ⋅ ( S 1 × S 2 ) ] s\equiv\operatorname{sgn}[\mathbf{L}\cdot(\mathbf{S}_1\times\mathbf{S}_2)] s ≡ sgn [ L ⋅ ( S 1 × S 2 )] . The second is the azimuth ϕ J \phi_J ϕ J of J ^ \hat{\mathbf{J}} J ^ about z ^ \hat{\mathbf{z}} z ^ . The third is the orientation of the triad about J ^ \hat{\mathbf{J}} J ^ , given by the azimuth Φ L \Phi_L Φ L of L \mathbf{L} L about J ^ \hat{\mathbf{J}} J ^ , measured counterclockwise from the line of nodes z ^ × J ^ \hat{\mathbf{z}}\times\hat{\mathbf{J}} z ^ × J ^ . The capital letter distinguishes it from the canonical azimuth ϕ L \phi_L ϕ L of L \mathbf{L} L about z ^ \hat{\mathbf{z}} z ^ (Sec. III).
We assign θ = 0 \boldsymbol{\theta}=\mathbf{0} θ = 0 to the fiducial configuration f = f 1 f=f_1 f = f 1 , Φ L = 0 \Phi_L=0 Φ L = 0 , ϕ J = 0 \phi_J=0 ϕ J = 0 , a turning point of the nutation. Any configuration on the torus is reached from it by flowing under Λ \Lambda Λ , J 2 J^2 J 2 , and J z J_z J z by some amounts λ Λ \lambda_\Lambda λ Λ , λ J 2 \lambda_{J^2} λ J 2 , and λ J z \lambda_{J_z} λ J z , in any order since these flows commute. Each flow changes only some of the coordinates ( f , s , ϕ J , Φ L ) (f,s,\phi_J,\Phi_L) ( f , s , ϕ J , Φ L ) , as summarized in Table I . In particular, the J z J_z J z -flow does not evolve Φ L \Phi_L Φ L , since the line of nodes rotates with the triad. The Λ \Lambda Λ -flow changes Φ L \Phi_L Φ L as it deforms the triad (App. C 2). The flow amounts can therefore be read off one at a time: λ Λ \lambda_\Lambda λ Λ from f f f and s s s , then λ J z \lambda_{J_z} λ J z from ϕ J \phi_J ϕ J , and finally λ J 2 \lambda_{J^2} λ J 2 from what remains of Φ L \Phi_L Φ L .
Table I. Effect of each flow on the coordinates ( f , s , ϕ J , Φ L ) (f,s,\phi_J,\Phi_L) ( f , s , ϕ J , Φ L ) of a configuration. A bullet marks the coordinates changed by the flow.
Under the Λ \Lambda Λ -flow (App. C 1), f f f increases from f 1 f_1 f 1 to f 2 f_2 f 2 over the first half of the nutation cycle and comes back to f 1 f_1 f 1 over the second half. Since d f / d λ Λ = L ⋅ ( S 1 × S 2 ) df/d\lambda_\Lambda=\mathbf{L}\cdot(\mathbf{S}_1\times\mathbf{S}_2) df / d λ Λ = L ⋅ ( S 1 × S 2 ) , the sign s s s tells us which half a configuration belongs to, so f f f and s s s together fix λ Λ \lambda_\Lambda λ Λ . At the same time, the azimuth of L \mathbf{L} L about J ^ \hat{\mathbf{J}} J ^ goes from 0 0 0 to Φ L Λ ( λ Λ ) \Phi_L^\Lambda(\lambda_\Lambda) Φ L Λ ( λ Λ ) , given in Eq. (140) , and reaches Φ L Λ ( Δ λ Λ ) = Δ Φ L \Phi_L^\Lambda(\Delta\lambda_\Lambda)=\Delta\Phi_L Φ L Λ ( Δ λ Λ ) = Δ Φ L after one period, the rotation angle of step (i) in Sec. IV E 2.
The two other flows are rigid rotations, at unit angular velocity for the J z J_z J z -flow and at angular velocity 2 J 2J 2 J for the J 2 J^2 J 2 -flow. Only the J z J_z J z -flow moves J ^ \hat{\mathbf{J}} J ^ , so λ J z = ϕ J \lambda_{J_z}=\phi_J λ J z = ϕ J . The azimuth Φ L \Phi_L Φ L , on the other hand, receives Φ L Λ ( λ Λ ) \Phi_L^\Lambda(\lambda_\Lambda) Φ L Λ ( λ Λ ) from the Λ \Lambda Λ -flow and 2 J λ J 2 2J\lambda_{J^2} 2 J λ J 2 from the J 2 J^2 J 2 -flow. The flow amounts that lead from the fiducial configuration to ( f , s , ϕ J , Φ L ) (f,s,\phi_J,\Phi_L) ( f , s , ϕ J , Φ L ) are therefore
λ Λ = U ( f , s ) Ω Λ , λ J z = ϕ J , λ J 2 = Φ L − Φ L Λ ( λ Λ ) 2 J , (72) \lambda_\Lambda=\frac{U(f,s)}{\Omega_\Lambda},\qquad\lambda_{J_z}=\phi_J,\qquad\lambda_{J^2}=\frac{\Phi_L-\Phi_L^\Lambda(\lambda_\Lambda)}{2J}, \tag*{(72)} λ Λ = Ω Λ U ( f , s ) , λ J z = ϕ J , λ J 2 = 2 J Φ L − Φ L Λ ( λ Λ ) , (72)
where the first relation is Eq. (139) , with U U U and Ω Λ \Omega_\Lambda Ω Λ defined in App. C 1.
It remains to convert these flow amounts into angles. The flow of an action J j \mathcal{J}_j J j shifts θ j \theta_j θ j only, since d θ i / d λ J j = { θ i , J j } = δ i j d\theta_i/d\lambda_{\mathcal{J}_j}=\{\theta_i,\mathcal{J}_j\}=\delta_{ij} d θ i / d λ J j = { θ i , J j } = δ ij . The configuration with angles θ \boldsymbol{\theta} θ is therefore reached from the fiducial one by flowing under each J i \mathcal{J}_i J i by θ i \theta_i θ i , in any order (Sec. 39 E of Ref. [30 ] ). Since J i \mathcal{J}_i J i is a function of the COMs, flowing under J i \mathcal{J}_i J i by θ i \theta_i θ i is equivalent to flowing under each C j C_j C j by ( ∂ J i / ∂ C j ) θ i (\partial\mathcal{J}_i/\partial C_j)\theta_i ( ∂ J i / ∂ C j ) θ i (App. § 2.3.4 ). Collecting the contributions of the three actions, the flow amount under C j C_j C j is λ C j = ∑ i ( ∂ J i / ∂ C j ) θ i \lambda_{C_j}=\sum_i(\partial\mathcal{J}_i/\partial C_j)\theta_i λ C j = ∑ i ( ∂ J i / ∂ C j ) θ i 16 , and with the help of the Jacobian (Eq. (61) ) we get
λ J 2 = θ 1 2 J + ∂ J 3 ∂ J 2 θ 3 , λ J z = θ 2 , λ Λ = ∂ J 3 ∂ Λ θ 3 . (73) \lambda_{J^2}=\frac{\theta_1}{2J}+\frac{\partial\mathcal{J}_3}{\partial J^2}\theta_3,\qquad\lambda_{J_z}=\theta_2,\qquad\lambda_\Lambda=\frac{\partial\mathcal{J}_3}{\partial\Lambda}\theta_3. \tag*{(73)} λ J 2 = 2 J θ 1 + ∂ J 2 ∂ J 3 θ 3 , λ J z = θ 2 , λ Λ = ∂ Λ ∂ J 3 θ 3 . (73)
Inverting these relations with Eqs. (55) and (67) , and inserting the flow amounts of Eq. (72) , we obtain the angles of any configuration on the torus,
θ 1 = Φ L − [ Φ L Λ ( λ Λ ( f , s ) ) − Δ Φ L Δ λ Λ λ Λ ( f , s ) ] , (74) \theta_1=\Phi_L-\left[\Phi_L^\Lambda\bigl(\lambda_\Lambda(f,s)\bigr)-\frac{\Delta\Phi_L}{\Delta\lambda_\Lambda}\lambda_\Lambda(f,s)\right], \tag*{(74)} θ 1 = Φ L − [ Φ L Λ ( λ Λ ( f , s ) ) − Δ λ Λ Δ Φ L λ Λ ( f , s ) ] , (74)
θ 2 = ϕ J , (75) \theta_{2} = \phi_{J}, \tag*{(75)} θ 2 = ϕ J , (75)
θ 3 = 2 π λ Λ ( f , s ) Δ λ Λ , (76) \theta_{3} = 2\pi\frac{\lambda_{\Lambda}(f,s)}{\Delta\lambda_{\Lambda}}, \tag*{(76)} θ 3 = 2 π Δ λ Λ λ Λ ( f , s ) , (76)
with λ Λ ( f , s ) \lambda_{\Lambda}(f,s) λ Λ ( f , s ) and Φ L Λ \Phi_{L}^{\Lambda} Φ L Λ given by Eqs. (139) and (140) .
These angles also have a physical interpretation. Only θ 2 \theta_{2} θ 2 is a geometric angle and it coincides with the azimuth ϕ J \phi_{J} ϕ J of J ^ \hat{\mathbf{J}} J ^ , which is constant since ω 2 = 0 \omega_{2}=0 ω 2 = 0 . The angle θ 3 \theta_{3} θ 3 is the nutation phase, with θ 3 / 2 π \theta_{3}/2\pi θ 3 /2 π being the fraction of the nutation cycle completed since f = f 1 f=f_{1} f = f 1 . The angle θ 1 \theta_{1} θ 1 differs from the azimuth Φ L \Phi_{L} Φ L by the bracket, a periodic function of the triad shape, with period Δ λ Λ \Delta\lambda_{\Lambda} Δ λ Λ in λ Λ \lambda_{\Lambda} λ Λ . For the motion that starts at the fiducial configuration,
Φ L ( t ) = ω 1 t + [ Φ L Λ ( λ Λ ) − Δ Φ L Δ λ Λ λ Λ ] , (77) \Phi_{L}(t)=\omega_{1}t+\left[\Phi_{L}^{\Lambda}(\lambda_{\Lambda})-\frac{\Delta\Phi_{L}}{\Delta\lambda_{\Lambda}}\lambda_{\Lambda}\right], \tag*{(77)} Φ L ( t ) = ω 1 t + [ Φ L Λ ( λ Λ ) − Δ λ Λ Δ Φ L λ Λ ] , (77)
which is the sum of a uniform precession at frequency ω 1 \omega_{1} ω 1 and of an oscillation with the nutation period. The angle θ 1 = ω 1 t \theta_{1}=\omega_{1}t θ 1 = ω 1 t is thus the precession of L \mathbf{L} L about J ^ \hat{\mathbf{J}} J ^ once the nutation is removed.
On a regular torus, the map ( L , S 1 , S 2 ) → θ (\mathbf{L},\mathbf{S}_{1},\mathbf{S}_{2})\to\boldsymbol{\theta} ( L , S 1 , S 2 ) → θ of Eq. (5.14) is singular only when L ∥ ± J ^ \mathbf{L}\parallel\pm\hat{\mathbf{J}} L ∥ ± J ^ , where Φ L \Phi_{L} Φ L is undefined.
Violation of integrability conditions and separatrices
Separatrices
Informally, a separatrix is a level set of the COMs that separates the phase space into two regions with qualitatively different kinds of trajectories [36 ] . The level set of the pendulum through its unstable equilibrium separates libration from rotation, and the parabolic orbit under Newtonian gravitation separates bound from unbound orbits. In these simple systems, with one degree of freedom or one independent frequency, the separatrix is easy to find. The goal of this section is to propose a criterion that finds it in general. We also include level sets where the nature of the motion changes abruptly, without separating the phase space, for example, an isolated point of the ( C 1 , C 2 ) (C_{1},C_{2}) ( C 1 , C 2 ) plane. We call all of them separatrix-like. These level sets are not tori.
Traditionally, separatrices are associated with hyperbolic fixed points or with a vanishing frequency [37 , 38 ] . A hyperbolic fixed point, however, is a single point of a level set, and its analysis only describes the motion close to it. A frequency is defined on the whole level set, but it follows from the actions, which are themselves singular there. We look instead for a criterion on the level set as a whole, written directly in terms of the COMs. Our starting point is that a breakdown of the Liouville–Arnold integrability conditions can lead to separatrix-like level sets, where the topology of the level sets changes (see Sec. 50 of Ref. [30]). In the next subsection, we give a criterion that implies such a breakdown.
A mathematical criterion
Consider an integrable Hamiltonian system with 2 n 2n 2 n phase-space variables ( p , q ) (\mathbf{p},\mathbf{q}) ( p , q ) and n n n independent COMs C \mathbf{C} C in involution, whose level sets are compact and connected in some region of phase space. There, the system has action-angle variables ( J , θ ) (\mathcal{J},\boldsymbol{\theta}) ( J , θ ) . We propose the following criterion: if
lim C → C ∗ det ( ∂ C ∂ J ) = 0 , (78) \lim_{\mathbf{C}\to\mathbf{C}_*}\det\left(\frac{\partial\mathbf{C}}{\partial\mathcal{J}}\right)=0, \tag*{(78)} C → C ∗ lim det ( ∂ J ∂ C ) = 0 , (78)
then integrability breaks down on the level set C = C ∗ \mathbf{C}=\mathbf{C}_* C = C ∗ (App. § 2.3.5 ). We take the limit because the actions need not be defined on C ∗ \mathbf{C}_* C ∗ itself. In practice, upon obtaining det ( ∂ C / ∂ J ) \det(\partial\mathbf{C}/\partial\mathcal{J}) det ( ∂ C / ∂ J ) as an explicit function of C \mathbf{C} C , C ∗ \mathbf{C}_* C ∗ is found by setting it to zero.
Since independence of COMs is equivalent to d C 1 ∧ d C 2 ∧ ⋯ ∧ d C n ≠ 0 \mathrm{d}C_1\wedge\mathrm{d}C_2\wedge\cdots\wedge\mathrm{d}C_n\ne0 d C 1 ∧ d C 2 ∧ ⋯ ∧ d C n = 0 , we confine ourselves to the COMs C \mathbf{C} C whose differentials do not diverge at finite values of ( p , q ) (\mathbf{p},\mathbf{q}) ( p , q ) , for example C = ( J 2 , J z , Λ ) \mathbf{C}=(\mathcal{J}^2,J_z,\Lambda) C = ( J 2 , J z , Λ ) for the OA spinning BBH system. In general, action variables do not qualify.17 That is why we do not set C = J \mathbf{C}=\mathcal{J} C = J in the criterion det ( ∂ C / ∂ J ) = 0 \det(\partial\mathbf{C}/\partial\mathcal{J})=0 det ( ∂ C / ∂ J ) = 0 ; doing so yields an unrealizable condition 1 = 0 1=0 1 = 0 anyway.
Vanishing frequencies, which are usually associated with separatrices, imply the criterion. Indeed, by the chain rule of Eq. (59) ,
ω i = ∑ j ∂ C j ∂ J i ∂ H ˉ ∂ C j . (79) \omega_i=\sum_j\frac{\partial C_j}{\partial\mathcal{J}_i}\frac{\partial\bar{H}}{\partial C_j}. \tag*{(79)} ω i = j ∑ ∂ J i ∂ C j ∂ C j ∂ H ˉ . (79)
If all the frequencies go to zero in the limit C → C ∗ \mathbf{C}\to\mathbf{C}_* C → C ∗ , while the ∂ H ˉ / ∂ C j \partial\bar{H}/\partial C_j ∂ H ˉ / ∂ C j remain finite and not all zero, then the matrix ∂ C j / ∂ J i \partial C_j/\partial\mathcal{J}_i ∂ C j / ∂ J i becomes singular, which is Eq. (78) . Finally, the converse of the criterion does not hold, as we will see for the pendulum in Sec. VI C 1.
Applications to various systems
Pendulum
For a pendulum of mass m m m and length l l l in a uniform gravitational field g g g , with q q q the angle from the downward vertical and p q p_q p q its conjugate momentum, the Hamiltonian is
H = p q 2 2 m l 2 − m g l cos q . (80) H=\frac{p_q^2}{2ml^2}-mgl\cos q. \tag*{(80)} H = 2 m l 2 p q 2 − m g l cos q . (80)
For a libration orbit, ∣ H ∣ < m g l |H|<mgl ∣ H ∣ < m g l , the action variable J = ( 2 π ) − 1 ∮ p q d q \mathcal{J}=(2\pi)^{-1}\oint p_q\,\mathrm{d}q J = ( 2 π ) − 1 ∮ p q d q is [37 ]
J = 8 π m l g l [ E ( κ ) − ( 1 − κ 2 ) K ( κ ) ] , 2 κ 2 ≡ 1 + H m g l , (81) \mathcal{J}=\frac{8}{\pi}ml\sqrt{gl}\left[E(\kappa)-(1-\kappa^2)K(\kappa)\right],\qquad2\kappa^2\equiv1+\frac{H}{mgl}, \tag*{(81)} J = π 8 m l g l [ E ( κ ) − ( 1 − κ 2 ) K ( κ ) ] , 2 κ 2 ≡ 1 + m g l H , (81)
where K K K is the complete elliptic integral of the first kind defined in Eq. (137) , and
E ( k ) = ∫ 0 π / 2 1 − k 2 sin 2 ϑ d ϑ (82) E(k) = \int_{0}^{\pi/2} \sqrt{1-k^2\sin^2\vartheta}\,\mathrm{d}\vartheta \tag*{(82)} E ( k ) = ∫ 0 π /2 1 − k 2 sin 2 ϑ d ϑ (82)
is the complete elliptic integral of the second kind. With C = H \mathbf{C}=H C = H , the criterion reduces to the vanishing of the frequency ω = d H / d J = π g / l / [ 2 K ( κ ) ] \omega= \mathrm{d}H/\mathrm{d}\mathcal{J} = \pi\sqrt{g/l}/[2K(\kappa)] ω = d H / d J = π g / l / [ 2 K ( κ )] , which tends to zero as κ → 1 \kappa\to1 κ → 1 , i.e., H → m g l H \to mgl H → m g l . This is the level set through the upright unstable equilibrium, the separatrix between libration and rotation.
The integrability conditions also break down at the stable equilibrium, H = − m g l H=-mgl H = − m g l , where the level set shrinks from a closed curve to a point. Yet the frequency does not vanish there, since ω → g / l \omega\to\sqrt{g/l} ω → g / l . This is no cause for worry since in Sec. VI B, we claimed that Eq. (78) implies a breakdown of integrability conditions, and not the converse. Near this point, the action shrinks to zero together with the level set, so H + m g l H+mgl H + m g l and J \mathcal{J} J vanish together and their ratio ω \omega ω stays finite. At the separatrix, on the contrary, d J / d H \mathrm{d}\mathcal{J}/\mathrm{d}H d J / d H diverges. The criterion therefore captures the breakdowns where the actions become singular with respect to the COMs, which typically correspond to unstable equilibria.
Orbits under Newtonian gravitation
For bound orbits under Newtonian gravitation, with total mass m m m and reduced mass μ \mu μ , the Hamiltonian depends on the radial action J r = ( 2 π ) − 1 ∮ p r d r J_r=(2\pi)^{-1}\oint p_r\,\mathrm{d}r J r = ( 2 π ) − 1 ∮ p r d r and on L L L only through their sum [39 ] ,
H = − m 2 μ 3 2 ( J r + L ) 2 . (83) H=-\frac{m^2\mu^3}{2(J_r+L)^2}. \tag*{(83)} H = − 2 ( J r + L ) 2 m 2 μ 3 . (83)
The radial and azimuthal frequencies are thus degenerate, ∂ H / ∂ J r = ∂ H / ∂ L \partial H/\partial J_r=\partial H/\partial L ∂ H / ∂ J r = ∂ H / ∂ L , which is why bound orbits close. With C = ( H , L 2 , L z ) \mathbf{C}=(H,L^2,L_z) C = ( H , L 2 , L z ) and J = ( J r , L , L z ) \mathcal{J}=(J_r,L,L_z) J = ( J r , L , L z ) , the criterion reduces to the vanishing of the radial frequency ∂ H / ∂ J r = m 2 μ 3 / ( J r + L ) 3 = ( − 2 H ) 3 / 2 / ( m μ 3 / 2 ) \partial H/\partial J_r=m^2\mu^3/(J_r+L)^3=(-2H)^{3/2}/(m\mu^{3/2}) ∂ H / ∂ J r = m 2 μ 3 / ( J r + L ) 3 = ( − 2 H ) 3/2 / ( m μ 3/2 ) , which tends to zero as H → 0 − H\to0^- H → 0 − . This is the parabolic orbit, which separates the bound elliptic orbits from the unbound hyperbolic ones. There, all the frequencies vanish while ∂ H / ∂ C \partial H/\partial\mathbf{C} ∂ H / ∂ C remains finite, as discussed at the end of Sec. VI B.
Orbit-averaged spinning BBH system
We now apply the criterion to the OA spinning BBH system. Recall that, at fixed ( J , Λ ) (\mathcal{J},\Lambda) ( J , Λ ) , the shape of the triad is determined by f f f , defined in Eq. (C3) by S 1 ⋅ S 2 = ( Σ 1 − Σ 2 ) f \mathbf{S}_1\cdot\mathbf{S}_2=(\Sigma_1-\Sigma_2)f S 1 ⋅ S 2 = ( Σ 1 − Σ 2 ) f . Along the Λ \Lambda Λ -flow, f f f oscillates between the two smallest roots, f 1 f_1 f 1 and f 2 f_2 f 2 , of the cubic P P P of Eq. (129) , whose roots are f 1 ≤ f 2 ≤ f 3 f_1\le f_2\le f_3 f 1 ≤ f 2 ≤ f 3 , as
f ( λ Λ ) = f 1 + ( f 2 − f 1 ) sn 2 ( Ω Λ λ Λ , k ) , Ω Λ = 1 2 a 3 ( f 3 − f 1 ) , k 2 = f 2 − f 1 f 3 − f 1 , (84) f(\lambda_\Lambda)=f_1+(f_2-f_1)\operatorname{sn}^2(\Omega_\Lambda\lambda_\Lambda,k),\qquad\Omega_\Lambda=\frac{1}{2}\sqrt{a_3(f_3-f_1)},\qquad k^2=\frac{f_2-f_1}{f_3-f_1}, \tag*{(84)} f ( λ Λ ) = f 1 + ( f 2 − f 1 ) sn 2 ( Ω Λ λ Λ , k ) , Ω Λ = 2 1 a 3 ( f 3 − f 1 ) , k 2 = f 3 − f 1 f 2 − f 1 , (84)
where sn \operatorname{sn} sn is the Jacobi elliptic sine. These roots also govern the physical motion. Since H ˉ = H ˉ ( J 2 , Λ ) \bar{H}=\bar{H}(\mathcal{J}^2,\Lambda) H ˉ = H ˉ ( J 2 , Λ ) [Eq. (28) ] and f f f , a scalar product, is invariant under the rotations generated by J \mathbf{J} J , so that { f , J 2 } = 0 \{f,J^2\}=0 { f , J 2 } = 0 , Eq. (69) gives
d f d t = ∂ H ˉ ∂ Λ d f d λ Λ . (85) \frac{df}{dt}=\frac{\partial\bar{H}}{\partial\Lambda}\frac{df}{d\lambda_{\Lambda}}. \tag*{(85)} d t df = ∂ Λ ∂ H ˉ d λ Λ df . (85)
On a level set, ∂ H ˉ / ∂ Λ \partial\bar{H}/\partial\Lambda ∂ H ˉ / ∂ Λ is constant, so the time evolution of f f f is the Λ \Lambda Λ -flow of Eq. (84) with λ Λ = ( ∂ H ˉ / ∂ Λ ) t \lambda_{\Lambda}=(\partial\bar{H}/\partial\Lambda)t λ Λ = ( ∂ H ˉ / ∂ Λ ) t . In real time too, f f f therefore oscillates between the turning points f 1 f_1 f 1 and f 2 f_2 f 2 , with the nutation frequency ω 3 \omega_3 ω 3 of Eq. (68) .
With C = ( J 2 , J z , Λ ) \mathbf{C}=(J^2,J_z,\Lambda) C = ( J 2 , J z , Λ ) and J = ( J , J z , J 3 ) \mathcal{J}=(J,J_z,\mathcal{J}_3) J = ( J , J z , J 3 ) , the criterion reduces, by Eqs. (62) , (67) , and (138) , to
det ( ∂ C ∂ J ) = 2 π J Ω Λ K ( k ) → 0. (86) \det\left(\frac{\partial\mathbf{C}}{\partial\mathcal{J}}\right)=\frac{2\pi J\Omega_{\Lambda}}{K(k)}\to0. \tag*{(86)} det ( ∂ J ∂ C ) = K ( k ) 2 π J Ω Λ → 0. (86)
Since K ( k ) ≥ π / 2 K(k)\geq\pi/2 K ( k ) ≥ π /2 , this happens when one of the following conditions holds: (i) K ( k ) K(k) K ( k ) diverges, (ii) Ω Λ = 0 \Omega_{\Lambda}=0 Ω Λ = 0 , or (iii) J = 0 J=0 J = 0 . We discuss them in turn.
Condition (i) means k → 1 k\to1 k → 1 , i.e., by Eq. (84) , that the two largest roots coalesce, f 1 < f 2 = f 3 f_1<f_2=f_3 f 1 < f 2 = f 3 . Since a 3 > 0 a_3>0 a 3 > 0 , condition (ii) means f 3 = f 1 f_3=f_1 f 3 = f 1 , i.e., that all three roots coincide, f 1 = f 2 = f 3 f_1=f_2=f_3 f 1 = f 2 = f 3 . In both cases, the nutation frequency ω 3 = π Ω Λ ( ∂ H ˉ / ∂ Λ ) / K ( k ) \omega_3=\pi\Omega_{\Lambda}(\partial\bar{H}/\partial\Lambda)/K(k) ω 3 = π Ω Λ ( ∂ H ˉ / ∂ Λ ) / K ( k ) of Eq. (68) vanishes, i.e., the nutation period 2 π / ω 3 2\pi/\omega_3 2 π / ω 3 diverges. In case (i), the divergence is logarithmic, since K ( k ) ≃ ln ( 4 / 1 − k 2 ) K(k)\simeq\ln(4/\sqrt{1-k^2}) K ( k ) ≃ ln ( 4/ 1 − k 2 ) as k → 1 k\to1 k → 1 , and it is the divergence of the precession period τ \tau τ found in [12 ] , which we call the nutation period.
Conditions (i) and (ii) correspond to a multiple root f ∗ f_* f ∗ of P P P , double in case (i) and triple in case (ii). For any multiple root, the configuration with f = f ∗ f=f_* f = f ∗ is an equilibrium of the shape of the triad. Since P ( f ∗ ) = 0 P(f_*)=0 P ( f ∗ ) = 0 , Eq. (129) gives d f / d λ Λ = 0 df/d\lambda_{\Lambda}=0 df / d λ Λ = 0 at f ∗ f_* f ∗ , and differentiating it gives d 2 f / d λ Λ 2 = P ′ ( f ) / 2 d^2f/d\lambda_{\Lambda}^2=P'(f)/2 d 2 f / d λ Λ 2 = P ′ ( f ) /2 , which also vanishes at f ∗ f_* f ∗ , so f f f remains equal to f ∗ f_* f ∗ . Its stability follows from the continuity of the roots of P P P with respect to ( J , Λ ) (J,\Lambda) ( J , Λ ) . A small perturbation of the configuration moves it to a nearby level set, on which f f f oscillates between the two smallest roots of the perturbed cubic. If f ∗ = f 1 = f 2 < f 3 f_*=f_1=f_2<f_3 f ∗ = f 1 = f 2 < f 3 , these roots are both close to f ∗ f_* f ∗ , and f f f stays close to f ∗ f_* f ∗ , so the equilibrium is stable. These equilibria are the spin-orbit resonances [4 , 6 ] , or Cassini states [16 ] , which the criterion does not detect, since K ( 0 ) = π / 2 K(0)=\pi/2 K ( 0 ) = π /2 keeps the determinant finite. If f ∗ = f 2 = f 3 f_*=f_2=f_3 f ∗ = f 2 = f 3 , the two smallest roots of the perturbed cubic are instead close to f 1 f_1 f 1 and f ∗ f_* f ∗ . However small the perturbation, f f f then oscillates over almost the whole interval [ f 1 , f ∗ ] [f_1,f_*] [ f 1 , f ∗ ] , and thus moves away from f ∗ f_* f ∗ by about f ∗ − f 1 f_*-f_1 f ∗ − f 1 , so the equilibrium is unstable.
We now apply these results to the four aligned configurations, in which S 1 \mathbf{S}_1 S 1 and S 2 \mathbf{S}_2 S 2 are each parallel or antiparallel to L \mathbf{L} L [12 , 13 ] . We first show that each of them corresponds to a double root of P P P . Since the three vectors are collinear, all the cross products in the Λ \Lambda Λ -flow, Eqs. (53) and (54) , vanish, so an aligned configuration is a fixed point of the flow. Then f f f is constant, d f / d λ Λ = d 2 f / d λ Λ 2 = 0 df/d\lambda_{\Lambda}=d^2f/d\lambda_{\Lambda}^2=0 df / d λ Λ = d 2 f / d λ Λ 2 = 0 , and Eq. (129) gives P ( f ∗ ) = P ′ ( f ∗ ) = 0 P(f_*)=P'(f_*)=0 P ( f ∗ ) = P ′ ( f ∗ ) = 0 : the value f ∗ f_* f ∗ of f f f at an aligned configuration is a double root of P P P . Which kind of double root it is follows from the third root, f r d f_{\mathrm{rd}} f rd , given by the sum of the roots, 2 f ∗ + f r d = − a 2 / a 3 2f_*+f_{\mathrm{rd}}=-a_2/a_3 2 f ∗ + f rd = − a 2 / a 3 . For the up-down configuration, with S 1 \mathbf{S}_1 S 1 along L \mathbf{L} L and S 2 \mathbf{S}_2 S 2 opposite to it, S 1 ⋅ S 2 = − S 1 S 2 \mathbf{S}_1\cdot\mathbf{S}_2=-S_1S_2 S 1 ⋅ S 2 = − S 1 S 2 , so that, by Eq. (122) , f ∗ = − S 1 S 2 / ( Σ 1 − Σ 2 ) f_*=-S_1S_2/(\Sigma_1-\Sigma_2) f ∗ = − S 1 S 2 / ( Σ 1 − Σ 2 ) , on the level set
( J , Λ ) = ( ∣ L + S 1 − S 2 ∣ , L ( Σ 1 S 1 − Σ 2 S 2 ) ) . (87) (J,\Lambda)=\left(\left|L+S_1-S_2\right|,L(\Sigma_1S_1-\Sigma_2S_2)\right). \tag*{(87)} ( J , Λ ) = ( ∣ L + S 1 − S 2 ∣ , L ( Σ 1 S 1 − Σ 2 S 2 ) ) . (87)
Inserting Eq. (87) into Eqs. (126) and (C8), we find
f r d − f ∗ = m 1 − m 2 2 m ( L − L + ) ( L − L − ) , L ± = m 1 m 2 m 1 2 − m 2 2 ( Σ 1 S 1 ± Σ 2 S 2 ) 2 (88) f_{\mathrm{rd}}-f_*=\frac{m_1-m_2}{2m}(L-L_+)(L-L_-),\qquad L_\pm=\frac{m_1m_2}{m_1^2-m_2^2}\left(\sqrt{\Sigma_1S_1}\pm\sqrt{\Sigma_2S_2}\right)^2 \tag*{(88)} f rd − f ∗ = 2 m m 1 − m 2 ( L − L + ) ( L − L − ) , L ± = m 1 2 − m 2 2 m 1 m 2 ( Σ 1 S 1 ± Σ 2 S 2 ) 2 (88)
with 0 ≤ L − < L + 0\leq L_-<L_+ 0 ≤ L − < L + for m 1 > m 2 m_1>m_2 m 1 > m 2 . For L − < L < L + L_-<L<L_+ L − < L < L + , the third root lies below f ∗ f_* f ∗ , so that f r d = f 1 < f ∗ = f 2 = f 3 f_{\mathrm{rd}}=f_1<f_*=f_2=f_3 f rd = f 1 < f ∗ = f 2 = f 3 : condition (i) holds, and the up-down configuration is unstable. The other configurations of this level set, with f 1 ≤ f < f ∗ f_1\leq f<f_* f 1 ≤ f < f ∗ , are not aligned, but approach the up-down configuration as λ Λ → ± ∞ \lambda_\Lambda\to\pm\infty λ Λ → ± ∞ , in agreement with the divergence of K ( k ) K(k) K ( k ) . At L = L ± L=L_\pm L = L ± , f r d = f ∗ f_{\mathrm{rd}}=f_* f rd = f ∗ is a triple root: condition (ii) holds, and the up-down configuration changes stability. For L < L − L<L_- L < L − or L > L + L>L_+ L > L + , f ∗ = f 1 = f 2 < f r d = f 3 f_*=f_1=f_2<f_{\mathrm{rd}}=f_3 f ∗ = f 1 = f 2 < f rd = f 3 , and it is stable. The same computation for the up-up, down-down, and down-up configurations gives f r d > f ∗ f_{\mathrm{rd}}>f_* f rd > f ∗ for all L L L , so their double root is always f 1 = f 2 < f 3 f_1=f_2<f_3 f 1 = f 2 < f 3 , and they are stable, in agreement with Refs. [12 , 13 ] .
At Newtonian order, L = μ m a ( 1 − e 2 ) L=\mu\sqrt{ma(1-e^2)} L = μ ma ( 1 − e 2 ) . With S a = χ a m a 2 S_a=\chi_am_a^2 S a = χ a m a 2 and q = m 2 / m 1 q=m_2/m_1 q = m 2 / m 1 , Eq. (88) becomes L ± = μ m ( χ 1 ± q χ 2 ) 2 / ( 1 − q ) L_\pm=\mu m(\sqrt{\chi_1}\pm\sqrt{q\chi_2})^2/(1-q) L ± = μ m ( χ 1 ± q χ 2 ) 2 / ( 1 − q ) , and the up-down configuration of an eccentric binary is unstable for
p u d − < a ( 1 − e 2 ) < p u d + , p u d ± ≡ L ± 2 μ 2 m = ( χ 1 ± q χ 2 ) 4 ( 1 − q ) 2 m (89) p_{\mathrm{ud}-}<a(1-e^2)<p_{\mathrm{ud}+},\qquad p_{\mathrm{ud}\pm}\equiv\frac{L_\pm^2}{\mu^2m}=\frac{(\sqrt{\chi_1}\pm\sqrt{q\chi_2})^4}{(1-q)^2}m \tag*{(89)} p ud − < a ( 1 − e 2 ) < p ud + , p ud ± ≡ μ 2 m L ± 2 = ( 1 − q ) 2 ( χ 1 ± q χ 2 ) 4 m (89)
and changes stability at a ( 1 − e 2 ) = p u d ± a(1-e^2)=p_{\mathrm{ud}\pm} a ( 1 − e 2 ) = p ud ± . At fixed eccentricity, the unstable range of the semi-major axis is p u d − / ( 1 − e 2 ) < a < p u d + / ( 1 − e 2 ) p_{\mathrm{ud}-}/(1-e^2)<a<p_{\mathrm{ud}+}/(1-e^2) p ud − / ( 1 − e 2 ) < a < p ud + / ( 1 − e 2 ) . For circular orbits, e = 0 e=0 e = 0 and a = r a=r a = r , and Eq. (89) reduces to r u d − < r < r u d + r_{\mathrm{ud}-}<r<r_{\mathrm{ud}+} r ud − < r < r ud + , with
r u d ± = ( χ 1 ± q χ 2 ) 4 ( 1 − q ) 2 m (90) r_{\mathrm{ud}\pm}=\frac{(\sqrt{\chi_1}\pm\sqrt{q\chi_2})^4}{(1-q)^2}m \tag*{(90)} r ud ± = ( 1 − q ) 2 ( χ 1 ± q χ 2 ) 4 m (90)
which is the instability range found in Refs. [12 , 13 ] . In practice, r u d − r_{\mathrm{ud}-} r ud − lies below the domain of validity of the PN approximation, and only r u d + r_{\mathrm{ud}+} r ud + is relevant.
Condition (iii), J = 0 J=0 J = 0 , is of a different nature. It requires L + S 1 + S 2 = 0 \mathbf{L}+\mathbf{S}_1+\mathbf{S}_2=0 L + S 1 + S 2 = 0 , meaning that the three angular momenta form a closed triangle. It can happen only if ∣ S 1 − S 2 ∣ ≤ L ≤ S 1 + S 2 \lvert S_1-S_2\rvert\leq L\leq S_1+S_2 ∣ S 1 − S 2 ∣ ≤ L ≤ S 1 + S 2 . This level set is the corner at J = 0 J=0 J = 0 of the region in Figure 2 (a). Squaring S 1 + S 2 = − L \mathbf{S}_1+\mathbf{S}_2=-\mathbf{L} S 1 + S 2 = − L gives S 1 ⋅ S 2 = ( L 2 − S 1 2 − S 2 2 ) / 2 \mathbf{S}_1\cdot\mathbf{S}_2=(L^2-S_1^2-S_2^2)/2 S 1 ⋅ S 2 = ( L 2 − S 1 2 − S 2 2 ) /2 , and similarly for the other two dot products. The shape of the triangle is therefore fixed by the magnitudes alone, f f f takes a single value, so that f 1 = f 2 f_1=f_2 f 1 = f 2 , and the shape is stable. The determinant vanishes for a different reason. The gradient of J 2 J^2 J 2 is proportional to the vector J \mathbf{J} J , and thus vanishes at J = 0 J=0 J = 0 , so the COMs are no longer independent. Accordingly, the level set consists of all rotations of a single triangle, which is not a torus. The criterion therefore detects a degeneracy in the orientation of the triad, not an instability of its shape. Moreover, the condition L ≤ S 1 + S 2 L\leq S_1+S_2 L ≤ S 1 + S 2 corresponds to separations of a few m m m , outside the domain of validity of the PN approximation.
Figure 2 shows the physical region of the ( J , Λ ) (J,\Lambda) ( J , Λ ) plane, that is, the pairs ( J , Λ ) (J,\Lambda) ( J , Λ ) reached by at least one configuration of the triad, at fixed masses and magnitudes. We compute it by reconstructing, at each ( J , Λ ) (J,\Lambda) ( J , Λ ) and f f f , the dot products of L \mathbf{L} L , S 1 \mathbf{S}_1 S 1 , and S 2 \mathbf{S}_2 S 2 , and requiring their Gram matrix to be positive semidefinite, so that a triad with these dot products exists. Up to the rescaling χ e f f ≡ c Λ / ( L m 2 ) \chi_{\rm eff}\equiv c\Lambda/(Lm^2) χ eff ≡ c Λ/ ( L m 2 ) , this is the ( J , χ e f f ) (J,\chi_{\rm eff}) ( J , χ eff ) plane of Ref. [4 ] , with the spin-orbit resonances, f 1 = f 2 f_1=f_2 f 1 = f 2 , on its boundary. What is new here is the color, which shows 1 − k 2 1-k^2 1 − k 2 . Since 1 − k 2 1-k^2 1 − k 2 vanishes when f 2 = f 3 f_2=f_3 f 2 = f 3 , it measures how close each level set is to condition (i). For L − < L < L + L_-<L<L_+ L − < L < L + , the up-down configuration lies inside the physical region, and 1 − k 2 1-k^2 1 − k 2 reaches its minimum, zero, there. For L > L + L>L_+ L > L + , the up-down configuration is a corner of the region, and the instability disappears. Far above the threshold, as in panel (b), 1 − k 2 1-k^2 1 − k 2 stays close to unity everywhere, so no level set is near condition (i).
For L − < L < L + L_-<L<L_+ L − < L < L + , the motion on the level set of the up-down configuration is not periodic. Every other configuration on it tends to the up-down configuration, in both the infinite past and the infinite future, as on the separatrix of the pendulum. Unlike the separatrix of the pendulum, however, this level set is a single point of the ( J , Λ ) (J,\Lambda) ( J , Λ ) plane and does not split it into two regions. It is therefore separatrix-like, in the sense of Sec. VI. At L = L ± L=L_\pm L = L ± , the triple root marks the change of stability of this configuration. The other level set detected by the criterion, J = 0 J=0 J = 0 (as per Eq. (86) ), is stable and reflects a degeneracy in the orientation of the triad.
a. Onset of the up-down instability during the inspiral. For circular orbits, the onset of the up-down instability during the inspiral was studied numerically in Refs. [12 , 13 , 40 ] . We test here the eccentric threshold of Eq. (89) , which predicts that the onset is set by L L L , i.e., by a ( 1 − e 2 ) a(1-e^2) a ( 1 − e 2 ) , and not by a a a alone. We evolve two ensembles of N = 3000 N=3000 N = 3000 BBHs, with q = 0.8 q=0.8 q = 0.8 and χ 1 = χ 2 = 0.8 \chi_1=\chi_2=0.8 χ 1 = χ 2 = 0.8 , starting close to the up-down configuration, with misalignments of S ^ 1 \hat{\mathbf S}_1 S ^ 1 from L ^ \hat{\mathbf L} L ^ and of S ^ 2 \hat{\mathbf S}_2 S ^ 2 from − L ^ -\hat{\mathbf L} − L ^ drawn from a Gaussian distribution of width 3 ∘ 3^\circ 3 ∘ . The first ensemble starts on a circular orbit at a 0 = 500 m a_0 = 500m a 0 = 500 m , and the second on an eccentric orbit with e 0 = 0.9 e_0 = 0.9 e 0 = 0.9 at a 0 = 1580 m a_0 = 1580m a 0 = 1580 m . Both start above the threshold and are evolved down to a = 15 m a = 15m a = 15 m , with the OA precession, while Peters’ equations (7.1) [41 ] drive the inspiral. Above the threshold, the spins oscillate around the up-down configuration. To measure the amplitude of this nutation independently of its phase, we interrupt the inspiral at 150 stages, equally spaced in log a \log a log a . At each stage, a copy of the ensemble is evolved with the conservative OA dynamics only, over about three precession periods at fixed orbit, and we record for each member the largest value of 1 − cos κ 1 1 - \cos\kappa_1 1 − cos κ 1 . The inspiral then continues from the state before this conservative evolution.
Figure 3 shows this amplitude as a function of L / L + L/L_+ L / L + . For both ensembles, it remains at the level of the initial spread as long as L > L + L > L_+ L > L + , then grows by two orders of magnitude between L ≃ 1.1 L + L \simeq1.1L_+ L ≃ 1.1 L + and L ≃ 0.9 L + L \simeq0.9L_+ L ≃ 0.9 L + , as the spins move away from the up-down configuration. The two curves coincide, although the eccentric ensemble reaches the threshold at e ≃ 0.55 e \simeq0.55 e ≃ 0.55 and a ≃ 295 m a \simeq295m a ≃ 295 m and the circular one at a = r u d + = 206 m a = r_{\mathrm{ud}+} = 206m a = r ud + = 206 m . The onset is thus set by a ( 1 − e 2 ) a(1-e^2) a ( 1 − e 2 ) , as predicted, whereas a threshold on a a a alone would place it much later in the inspiral of eccentric binaries.
Figure 3. Onset of the up-down instability during the inspiral, for two ensembles of 3000 BBHs (q = 0.8 q = 0.8 q = 0.8 , χ 1 = χ 2 = 0.8 \chi_1 = \chi_2 = 0.8 χ 1 = χ 2 = 0.8 ) starting close to the up-down configuration, with Gaussian misalignments of width 3 ∘ 3^\circ 3 ∘ , on a circular orbit with a 0 = 500 m a_0 = 500m a 0 = 500 m (solid) and an eccentric orbit with e 0 = 0.9 e_0 = 0.9 e 0 = 0.9 and a 0 = 1580 m a_0 = 1580m a 0 = 1580 m (dashed), evolved with spin-independent radiation reaction. The maximum of 1 − cos κ 1 1 - \cos\kappa_1 1 − cos κ 1 is taken over about three precession periods at fixed orbit. Lines show the median over the ensemble and bands the 16–84% range. The bottom axis gives L / L + L/L_+ L / L + , the top axis a ( 1 − e 2 ) / m a(1-e^2)/m a ( 1 − e 2 ) / m , and the vertical line marks L = L + L = L_+ L = L + . The two curves collapse at L = L + L = L_+ L = L + , which verifies that the onset of the instability is set by a ( 1 − e 2 ) a(1-e^2) a ( 1 − e 2 ) , in Eq. (89) .
In summary, the determinant in the criterion of Eq. (78) vanishes at the up-down configuration for L − < L < L + L_- < L < L_+ L − < L < L + , which coincides with the instability range found in Refs. [12 , 13 ] , and it accounts for the logarithmic divergence of the nutation period, called precession period in Refs. [4 , 12 ] . Combined with the root structure of P P P , it identifies the instability with the coalescence of the two largest roots, f 2 = f 3 f_2=f_3 f 2 = f 3 , and its threshold with the triple root, f 1 = f 2 = f 3 f_1=f_2=f_3 f 1 = f 2 = f 3 . Since the criterion depends on the orbit only through L L L , this result extends to eccentric orbits, for which the instability is set by a ( 1 − e 2 ) a(1-e^2) a ( 1 − e 2 ) rather than by a a a , consistent with Ref. [15 ] and as confirmed numerically in Figure 3 .
Non-orbit-averaged spinning BBH system at 1.5PN order
Finally, we apply the criterion of Eq. (78) to the non-OA spinning BBH system. Unlike for the 2PN OA system, its action variables are known only at 1.5PN order [18 , 31 ] . Following Ref. [18 ] , we take C = ( J 2 , J z , L 2 , H , S e f f ⋅ L ) \mathcal{C}=(J^2,J_z,L^2,H,\mathbf{S}_{\rm eff}\cdot\mathbf{L}) C = ( J 2 , J z , L 2 , H , S eff ⋅ L ) , i.e., the smooth versions of the COMs of Ref. [18 ] , and J = ( J , J z , L , J 4 , J 5 ) \mathcal{J}=(J,J_z,L,\mathcal{J}_4,\mathcal{J}_5) J = ( J , J z , L , J 4 , J 5 ) , where J 4 \mathcal{J}_4 J 4 is the radial action of Ref. [31 ] and J 5 \mathcal{J}_5 J 5 is the spin action of Ref. [18 ] . Since J 4 \mathcal{J}_4 J 4 depends only on ( H , L 2 , S e f f ⋅ L ) (H,L^2,\mathbf{S}_{\rm eff}\cdot\mathbf{L}) ( H , L 2 , S eff ⋅ L ) and J 5 \mathcal{J}_5 J 5 only on ( J 2 , L 2 , S e f f ⋅ L ) (J^2,L^2,\mathbf{S}_{\rm eff}\cdot\mathbf{L}) ( J 2 , L 2 , S eff ⋅ L ) , the Jacobian ∂ J / ∂ C \partial\mathcal{J}/\partial\mathcal{C} ∂ J / ∂ C , given in Eq. (55) of Ref. [18 ] with ( J , L ) (J,L) ( J , L ) in place of ( J 2 , L 2 ) (J^2,L^2) ( J 2 , L 2 ) , is block triangular, and the criterion factorizes,
det ( ∂ C ∂ J ) = 4 J L ( ∂ J 4 ∂ H ) − 1 ( ∂ J 5 ∂ ( S e f f ⋅ L ) ) − 1 → 0. (91) \det\left(\frac{\partial\mathcal{C}}{\partial\mathcal{J}}\right)=4JL\left(\frac{\partial\mathcal{J}_4}{\partial H}\right)^{-1}\left(\frac{\partial\mathcal{J}_5}{\partial(\mathbf{S}_{\rm eff}\cdot\mathbf{L})}\right)^{-1}\to0. \tag*{(91)} det ( ∂ J ∂ C ) = 4 J L ( ∂ H ∂ J 4 ) − 1 ( ∂ ( S eff ⋅ L ) ∂ J 5 ) − 1 → 0. (91)
The factor 4 J L 4JL 4 J L comes from ∂ J 2 / ∂ J = 2 J \partial J^2/\partial J=2J ∂ J 2 / ∂ J = 2 J and ∂ L 2 / ∂ L = 2 L \partial L^2/\partial L=2L ∂ L 2 / ∂ L = 2 L . Apart from J = 0 J=0 J = 0 and L = 0 L=0 L = 0 , the criterion is met as soon as one of the two derivatives diverges. For L = 0 L=0 L = 0 , the bodies move along a straight line, so the orbital plane is undefined, as is the orientation of the triad for J = 0 J=0 J = 0 .
The first factor is the radial frequency, ω 4 = ( ∂ J 4 / ∂ H ) − 1 \omega_4=(\partial\mathcal{J}_4/\partial H)^{-1} ω 4 = ( ∂ J 4 / ∂ H ) − 1 . From Eq. (38) of Ref. [31 ] ,
∂ J 4 ∂ H = m μ 3 / 2 ( − 2 H ) 3 / 2 − m c 2 μ 1 / 2 ( ν − 15 ) 2 32 − H , (92) \frac{\partial\mathcal{J}_4}{\partial H}=\frac{m\mu^{3/2}}{(-2H)^{3/2}}-\frac{m}{c^2}\frac{\mu^{1/2}(\nu-15)}{2\sqrt{32}\sqrt{-H}}, \tag*{(92)} ∂ H ∂ J 4 = ( − 2 H ) 3/2 m μ 3/2 − c 2 m 2 32 − H μ 1/2 ( ν − 15 ) , (92)
which diverges as H → 0 − H\to0^- H → 0 − . This is the PN counterpart of the Newtonian parabolic orbit. The radial period diverges, and the level set H = 0 H=0 H = 0 separates the bound orbits from the unbound ones.
The second factor is the counterpart of Eq. (86) . As discussed in Sec. IV E 2, J 5 \mathcal{J}_5 J 5 depends on S e f f ⋅ L \mathbf{S}_{\rm eff}\cdot\mathbf{L} S eff ⋅ L in the same way as J 3 \mathcal{J}_3 J 3 depends on Λ \Lambda Λ , and the corresponding cubic is obtained from Eq. (129) by replacing Σ a \boldsymbol{\Sigma}_a Σ a with σ a \boldsymbol{\sigma}_a σ a . The analysis of Sec. VI C 3 then carries over. For J ≠ 0 J\ne0 J = 0 , the criterion is met when the two largest roots of this cubic coalesce, which again happens at the up-down configuration. Repeating the computation of Eq. (88) with σ a \boldsymbol{\sigma}_a σ a , we find that this configuration is unstable for L − ′ < L < L + ′ L'_-<L<L'_+ L − ′ < L < L + ′ , with
L ± ′ = 4 m 1 m 2 3 ( m 1 2 − m 2 2 ) ( σ 1 S 1 ± σ 2 S 2 ) 2 . (93) L'_\pm=\frac{4m_1m_2}{3(m_1^2-m_2^2)}\left(\sqrt{\sigma_1S_1}\pm\sqrt{\sigma_2S_2}\right)^2. \tag*{(93)} L ± ′ = 3 ( m 1 2 − m 2 2 ) 4 m 1 m 2 ( σ 1 S 1 ± σ 2 S 2 ) 2 . (93)
For the system of Figure 2 , the absence of the 2PN spin-spin coupling increases the separation at which the instability starts by about 37% compared with the 2PN OA system. The same criterion thus captures both the bound-unbound transition of the orbit and the instability of the spins.
a. Remark. For all the systems of this section, the criterion selects the separatrix-like level sets, together with the level set J = 0 J=0 J = 0 of the spinning BBH system, which is stable but where the orientation of the triad is undefined. It is a sufficient condition for the breakdown of integrability, but not a necessary one, since it misses the stable equilibrium of the pendulum and the spin-orbit resonances. Its main limitation is that it requires the actions as functions of the COMs, which are in general hard to obtain. Another point to note is that for the systems considered here, these level sets could also have been found by other means, by elementary calculations for the pendulum and the Newtonian two-body problem, and from the stability analysis of Refs. [12 , 13 ] for the spinning BBH system. The merit of the criterion is to provide a single condition covering all these cases, justified by the breakdown of the Liouville–Arnold integrability conditions (App. § 2.3.5 ). Also, it gives the entire C level set in terms of C without needing to explore the individual equilibrium points on this level set, where the integrability conditions actually break down.18
Hamiltonian mechanics and isotropic spins
The background
The evolution of an ensemble of spin directions during the inspiral has been studied numerically before. Schnittman [6 ] derived OA equations for the precession, which contain the spin-orbit and spin-spin couplings, and evolved the orbit with Peters’ equations [41 ] , which change only the magnitude of L \mathbf{L} L . Bogdanović et al. [8 ] integrated these equations for ensembles of circular BBHs with identical masses, spin magnitudes and initial relative separation and found that an initially isotropic distribution of spin directions remains isotropic during the inspiral. Gerosa et al. [3 ] discussed how this result is maintained during the inspiral, but its origin has not been explained. Lousto et al. [42 ] evolved similar ensembles with the non-OA PN equations of motion, with conservative terms up to 3PN and the spin-dependent RR force of Ref. [43 ] , which tilts L ^ \hat{\mathbf{L}} L ^ . They found instead a small bias of the spins toward an anti-alignment with L ^ \hat{\mathbf{L}} L ^ . Correia [16 ] showed that the Cassini states move during the inspiral, which modifies the dynamics around them, and argued that this is why the final distribution of the spins may differ from the initial one.
An ensemble of BBHs is isotropic if ( L ^ , S ^ 1 , S ^ 2 ) (\hat{\mathbf{L}},\hat{\mathbf{S}}_1,\hat{\mathbf{S}}_2) ( L ^ , S ^ 1 , S ^ 2 ) is uniformly distributed on S 2 × S 2 × S 2 S^2 \times S^2 \times S^2 S 2 × S 2 × S 2 . We show that the preservation of isotropy follows from the Hamiltonian structure of the OA precession, provided that all the members share the same parameters and that RR does not act on the directions of the angular momenta. This covers the conservative regime and spin-independent RR, for circular as well as eccentric orbits. The equations of Ref. [6 ] satisfy both conditions, explaining the result of Ref. [8 ] . The RR force used in Ref. [42 ] violates both conditions, since the energy and angular momentum fluxes depend on the spin orientations, and part of the radiated angular momentum is not along L ^ \hat{\mathbf{L}} L ^ .
Our derivation is close to the proof of Liouville’s volume theorem, and follows the presentation of Ref. [44 ] . We derive the result in Sec. VIIB and test it numerically in Sec. VIIC.
Preservation of isotropy
In this subsection, the member BBHs of our ensemble evolve under either just the OA EOMs (2.8) in the conservative regime where L L L and D D D are constant, or together with spin-independent RR, where L L L and D D D decrease. Spin-dependent RR is discussed at the end of this subsection and in Sec. VII C. The derivation has four steps. We first show that all members share the same parameters at all times, and therefore follow the same equations of motion for the directions of their angular momenta. We then choose coordinates in which an isotropic distribution corresponds to a uniform density. Next, we show that the flow generated by these equations preserves volume in these coordinates, whether L L L and D D D are constant or decreasing. Finally, our extension of Liouville’s theorem implies that the density is constant along each trajectory, so an initially uniform density remains uniform.
a. Common parameters. Consider an ensemble of spinning BBHs whose members share the same masses m a m_a m a , spin magnitudes S a S_a S a , initial semi-major axis a 0 a_0 a 0 and eccentricity e 0 e_0 e 0 , but have different orientations of L \mathbf{L} L , S 1 \mathbf{S}_1 S 1 , and S 2 \mathbf{S}_2 S 2 . In the conservative regime, a a a and e e e are constants. With leading-order RR, they evolve according to Peters’ equations [41 ]
d a d t = − 64 5 c 5 ν m 3 a 3 ( 1 − e 2 ) 7 / 2 ( 1 + 73 24 e 2 + 37 96 e 4 ) (94) \frac{da}{dt}=-\frac{64}{5c^5}\frac{\nu m^3}{a^3(1-e^2)^{7/2}}\left(1+\frac{73}{24}e^2+\frac{37}{96}e^4\right) \tag*{(94)} d t d a = − 5 c 5 64 a 3 ( 1 − e 2 ) 7/2 ν m 3 ( 1 + 24 73 e 2 + 96 37 e 4 ) (94)
d e d t = − 304 15 c 5 ν m 3 e a 4 ( 1 − e 2 ) 5 / 2 ( 1 + 121 304 e 2 ) . (95) \frac{de}{dt}=-\frac{304}{15c^5}\frac{\nu m^3 e}{a^4(1-e^2)^{5/2}}\left(1+\frac{121}{304}e^2\right). \tag*{(95)} d t d e = − 15 c 5 304 a 4 ( 1 − e 2 ) 5/2 ν m 3 e ( 1 + 304 121 e 2 ) . (95)
These depend solely on a a a , e e e and the masses, thus a ( t ) a(t) a ( t ) and e ( t ) e(t) e ( t ) remain common for all members, as do the values of the orbital angular momentum and the semi-minor axis, since
L ( t ) = μ m a ( t ) [ 1 − e ( t ) 2 ] , D ( t ) = a ( t ) 1 − e ( t ) 2 . (96) L(t)=\mu\sqrt{m\,a(t)\left[1-e(t)^2\right]},\qquad D(t)=a(t)\sqrt{1-e(t)^2}. \tag*{(96)} L ( t ) = μ m a ( t ) [ 1 − e ( t ) 2 ] , D ( t ) = a ( t ) 1 − e ( t ) 2 . (96)
The same holds for any spin-independent corrections to RR. At leading order, RR does not act on the spins, and the averaged flux of angular momentum is along L ^ \hat{\mathbf{L}} L ^ [41 ] . It therefore only adds to d L / d t d\mathbf{L}/dt d L / d t a term along L ^ \hat{\mathbf{L}} L ^ , which changes only the magnitude of L \mathbf{L} L , while the precession term of Eq. (2.8b) is orthogonal to L \mathbf{L} L and changes only its direction. The spins thus still obey Eqs. (2.8), and L ^ \hat{\mathbf{L}} L ^ obeys a slightly modified form of Eq. (2.8b)
d L ^ d t = 1 c 2 D 3 { 2 S e f f − 3 ν 2 ( L ^ ⋅ S 0 ) L S 0 } × L ^ , (97) \frac{d\hat{\mathbf{L}}}{dt}=\frac{1}{c^2D^3}\left\{2\mathbf{S}_{\mathrm{eff}}-\frac{3\nu}{2}\frac{(\hat{\mathbf{L}}\cdot\mathbf{S}_0)}{L}\mathbf{S}_0\right\}\times\hat{\mathbf{L}}, \tag*{(97)} d t d L ^ = c 2 D 3 1 { 2 S eff − 2 3 ν L ( L ^ ⋅ S 0 ) S 0 } × L ^ , (97)
with the common L ( t ) L(t) L ( t ) and D ( t ) D(t) D ( t ) of Eq. (96) .
b. Coordinates and density. In the dissipative regime, the phase space S 2 ( L ) × S 2 ( S 1 ) × S 2 ( S 2 ) \mathbb{S}^2(\mathbf{L})\times\mathbb{S}^2(\mathbf{S}_1)\times\mathbb{S}^2(\mathbf{S}_2) S 2 ( L ) × S 2 ( S 1 ) × S 2 ( S 2 ) changes continuously with time as the radius L ( t ) L(t) L ( t ) of the first sphere shrinks due to GW emission. In both regimes, we therefore work with S 2 ( L ^ ) × S 2 ( S 1 ) × S 2 ( S 2 ) \mathbb{S}^2(\hat{\mathbf{L}})\times\mathbb{S}^2(\mathbf{S}_1)\times\mathbb{S}^2(\mathbf{S}_2) S 2 ( L ^ ) × S 2 ( S 1 ) × S 2 ( S 2 ) , whose spheres have constant radii. On each sphere, we use cylindrical coordinates defined by the azimuth around a fixed axis z ^ \hat{\mathbf{z}} z ^ and the component along it, ( ϕ L , L ^ z ) (\phi_L,\hat{L}_z) ( ϕ L , L ^ z ) for L ^ \hat{\mathbf{L}} L ^ and ( ϕ S a , S a z ) (\phi_{S_a},S_{az}) ( ϕ S a , S a z ) for the spins. For a vector of magnitude V V V , d ϕ V d V z = V d Ω V d\phi_V\,dV_z=V\,d\Omega_V d ϕ V d V z = V d Ω V , with d Ω V d\Omega_V d Ω V the solid angle, so a distribution uniform in ( ϕ V , V z ) (\phi_V, V_z) ( ϕ V , V z ) is isotropic.19 We describe the ensemble by its density ρ ( ϕ L , L ^ z , ϕ S 1 , S 1 z , ϕ S 2 , S 2 z , t ) \rho(\phi_L, \hat{L}_z, \phi_{S_1}, S_{1z}, \phi_{S_2}, S_{2z}, t) ρ ( ϕ L , L ^ z , ϕ S 1 , S 1 z , ϕ S 2 , S 2 z , t ) , defined by
d N = ρ d ϕ L d L ^ z d ϕ S 1 d S 1 z d ϕ S 2 d S 2 z , (98) dN = \rho\,d\phi_L\,d\hat{L}_z\,d\phi_{S_1}\,dS_{1z}\,d\phi_{S_2}\,dS_{2z}, \tag*{(98)} d N = ρ d ϕ L d L ^ z d ϕ S 1 d S 1 z d ϕ S 2 d S 2 z , (98)
where d N dN d N is the number of members in a cell of phase space at time t t t . The ensemble is isotropic at time t t t if ρ \rho ρ does not depend on the coordinates. Our aim is to show that if this property holds at t = 0 t = 0 t = 0 , it holds at all later times.
c. Equations of motion . In these coordinates, Eqs. (2.8) for the spins and Eq. (97) for L ^ \hat{\mathbf{L}} L ^ read
d ϕ L d t = 1 L ∂ H ˉ ∂ L ^ z , d L ^ z d t = − 1 L ∂ H ˉ ∂ ϕ L , d ϕ S a d t = ∂ H ˉ ∂ S a z , d S a z d t = − ∂ H ˉ ∂ ϕ S a . (99) \begin{aligned}
\frac{d\phi_L}{dt} &= \frac{1}{L}\frac{\partial\bar{H}}{\partial\hat{L}_z}, &
\frac{d\hat{L}_z}{dt} &= -\frac{1}{L}\frac{\partial\bar{H}}{\partial\phi_L},\\
\frac{d\phi_{S_a}}{dt} &= \frac{\partial\bar{H}}{\partial S_{az}}, &
\frac{dS_{az}}{dt} &= -\frac{\partial\bar{H}}{\partial\phi_{S_a}}. \tag*{(99)}
\end{aligned} d t d ϕ L d t d ϕ S a = L 1 ∂ L ^ z ∂ H ˉ , = ∂ S a z ∂ H ˉ , d t d L ^ z d t d S a z = − L 1 ∂ ϕ L ∂ H ˉ , = − ∂ ϕ S a ∂ H ˉ . (99)
where H ˉ \bar{H} H ˉ is seen as a function of the six coordinates and of the parameters L L L , D D D , S a S_a S a , and m a m_a m a , which are held fixed in the partial derivatives. The magnitude L L L enters at two places, through H ˉ \bar{H} H ˉ and through the prefactor 1 / L 1/L 1/ L in Eqs. (99) . Together with Eqs. (7.1) and (96) , which give L ( t ) L(t) L ( t ) and D ( t ) D(t) D ( t ) , these equations form the complete system of evolution for ( S 1 , S 2 , L ) (\mathbf{S}_1,\mathbf{S}_2,\mathbf{L}) ( S 1 , S 2 , L ) in both regimes. RR therefore enters only through these two functions of time. In the conservative regime, where L L L is constant, Eqs. (99) are Hamilton’s equations with canonical pairs ( ϕ L , L z ) (\phi_L, L_z) ( ϕ L , L z ) and ( ϕ S a , S a z ) (\phi_{S_a}, S_{az}) ( ϕ S a , S a z ) . When L L L varies, the evolution is no longer Hamiltonian with H ˉ \bar{H} H ˉ and the PBs of Eq. (15) since, with these PBs, any Hamiltonian implies d L / d t = 0 dL/dt = 0 d L / d t = 0 (L 2 L^2 L 2 is a Casimir), which does not hold with RR. This is why we cannot apply Liouville’s theorem directly. Our derivation, however, relies only on the conservation of phase-space volume, which still holds, as we now show.
d. Volume preservation . Since all members share the same parameters, their velocity in phase space depends only on their position and on time, and defines a velocity field v = ( ϕ ˙ L , L ^ ˙ z , ϕ ˙ S 1 , S ˙ 1 z , ϕ ˙ S 2 , S ˙ 2 z ) \mathbf{v} = (\dot{\phi}_L, \dot{\hat{L}}_z, \dot{\phi}_{S_1}, \dot{S}_{1z}, \dot{\phi}_{S_2}, \dot{S}_{2z}) v = ( ϕ ˙ L , L ^ ˙ z , ϕ ˙ S 1 , S ˙ 1 z , ϕ ˙ S 2 , S ˙ 2 z ) . In these coordinates, the divergence takes its Cartesian form, and Eqs. (99) give
∇ ⋅ v = 1 L ( ∂ 2 H ˉ ∂ ϕ L ∂ L ^ z − ∂ 2 H ˉ ∂ L ^ z ∂ ϕ L ) + ∑ a ( ∂ 2 H ˉ ∂ ϕ S a ∂ S a z − ∂ 2 H ˉ ∂ S a z ∂ ϕ S a ) = 0. (100) \nabla\cdot\mathbf{v} = \frac{1}{L}\left(\frac{\partial^2\bar{H}}{\partial\phi_L\,\partial\hat{L}_z} - \frac{\partial^2\bar{H}}{\partial\hat{L}_z\,\partial\phi_L}\right) + \sum_a\left(\frac{\partial^2\bar{H}}{\partial\phi_{S_a}\,\partial S_{az}} - \frac{\partial^2\bar{H}}{\partial S_{az}\,\partial\phi_{S_a}}\right) = 0. \tag*{(100)} ∇ ⋅ v = L 1 ( ∂ ϕ L ∂ L ^ z ∂ 2 H ˉ − ∂ L ^ z ∂ ϕ L ∂ 2 H ˉ ) + a ∑ ( ∂ ϕ S a ∂ S a z ∂ 2 H ˉ − ∂ S a z ∂ ϕ S a ∂ 2 H ˉ ) = 0. (100)
For each pair of coordinates, the two mixed derivatives of H ˉ \bar{H} H ˉ cancel. For the pair ( ϕ L , L ^ z ) (\phi_L, \hat{L}_z) ( ϕ L , L ^ z ) , this requires 1 / L 1/L 1/ L to factor out, which holds because L L L depends only on time and not on the coordinates. Since a small volume δ V \delta V δ V carried by the flow changes as d ( δ V ) / d t = ( ∇ ⋅ v ) δ V d(\delta V)/dt = (\nabla\cdot\mathbf{v})\,\delta V d ( δ V ) / d t = ( ∇ ⋅ v ) δ V , the flow preserves volume.
e. Extension of Liouville’s theorem . The ensembles satisfy the following two conditions: (i) the members of the ensemble are neither created nor destroyed, and (ii) all members share the same parameters ( m a , S a , L ( t ) , D ( t ) ) (m_a, S_a, L(t), D(t)) ( m a , S a , L ( t ) , D ( t )) at any given time thereby resulting in a single-valued velocity field on the phase space. The consequence is that ρ \rho ρ obeys the continuity equation ∂ t ρ + ∇ ⋅ ( ρ v ) = 0 \partial_t \rho+ \nabla\cdot(\rho\mathbf{v}) = 0 ∂ t ρ + ∇ ⋅ ( ρ v ) = 0 , which with Eq. (100) , reduces to
d ρ d t = ∂ ρ ∂ t + v ⋅ ∇ ρ = 0 , (101) \frac{d\rho}{dt} = \frac{\partial\rho}{\partial t} + \mathbf{v} \cdot\nabla\rho= 0, \tag*{(101)} d t d ρ = ∂ t ∂ ρ + v ⋅ ∇ ρ = 0 , (101)
where d / d t d/dt d / d t is the material or Lagrangian derivative along the trajectory of a member. This extends Liouville’s theorem to our non-Hamiltonian flow. The density is constant along each trajectory, so an initially uniform density remains uniform. An initially isotropic ensemble thus remains isotropic, in the conservative regime and with spin-independent RR, for any eccentricity. The circular ensembles of Ref. [8 ] correspond to e = 0 e = 0 e = 0 with leading-order RR.
f. Conditions. The argument relies on two conditions. First, all members share the same parameters, m a m_a m a , S a \mathbf{S}_a S a , L ( t ) L(t) L ( t ) , and D ( t ) D(t) D ( t ) . Otherwise, the velocity of a member would depend on its parameters as well as on its position, thus making the velocity field multi-valued at a given point in the phase space, leading to non-applicability of the continuity equation. Second, in the evolution considered here, RR acts only on L L L and D D D , and the directions of L \mathbf{L} L and S a \mathbf{S}_a S a evolve under Eqs. (99) .
Spin-dependent RR (see, e.g., Eq. (17) of Ref. [3 ] and Refs. [45 , 46 , 47 , 48 ] ) violates both conditions. Its spin-orbit and spin-spin contributions to the fluxes depend on L ^ ⋅ S a \hat{\mathbf{L}} \cdot\mathbf{S}_a L ^ ⋅ S a , so members with different orientations acquire different L ( t ) L(t) L ( t ) and D ( t ) D(t) D ( t ) . The radiated angular momentum is also no longer along L ^ \hat{\mathbf{L}} L ^ , so RR tilts L ^ \hat{\mathbf{L}} L ^ [1] and adds a term to Eq. (97) . Isotropy is then no longer protected. The violation is suppressed by the relative PN order of these terms, and we show in Sec. VII C that it lowers the mean of the effective spin χ e f f = c Λ / ( L m 2 ) = c L ^ ⋅ S 0 / m 2 \chi_{\rm eff} = c\Lambda/(Lm^2) = c\hat{\mathbf{L}} \cdot\mathbf{S}_0/m^2 χ eff = c Λ/ ( L m 2 ) = c L ^ ⋅ S 0 / m 2 .
When the masses, the initial semi-major axis and eccentricity, or the spin magnitudes vary between members, the first condition fails, but the argument applies to each subset of members with identical parameters. Within each subset, if the initial directions of the angular momenta are isotropic and independent of these parameters, the subset starts isotropic and remains isotropic, and so does the whole ensemble. If some parameters are correlated with the directions, however, the subsets may not be isotropic at the start, even if the whole ensemble is, and isotropy can then be lost. Both cases are tested in Sec. VII C.
Numerical check
We evolve four ensembles of N = 10 4 N = 10^4 N = 1 0 4 BBHs with χ 1 = χ 2 = 0.8 \chi_1 = \chi_2 = 0.8 χ 1 = χ 2 = 0.8 and isotropic initial directions, all starting at a 0 = 780 m a_0 = 780m a 0 = 780 m . They differ in the mass ratio and initial eccentricity of their members, given in Table II . Each ensemble is evolved either in the conservative regime over one inspiral time, or down to a = 15 m a = 15m a = 15 m , with the OA precession and RR. Spin-independent RR is given by Eqs. (7.1). Spin-dependent RR adds the LO spin-orbit and spin-spin contributions to the fluxes and the torque on L ^ \hat{\mathbf{L}} L ^ [45 , 46 , 47 , 48 ] . For the isotropic ensemble, the RR runs start from the final state of the conservative run, so that the members have first precessed over many cycles and RR acts on a generic state of the precessing ensemble, and not on the sampled initial directions.
Table II. Mass ratio q q q and initial eccentricity e 0 e_{0} e 0 of the members of each ensemble. In the correlated ensembles, they depend on the hemisphere of S ^ 1 \hat{\mathbf{S}}_{1} S ^ 1 relative to L ^ \hat{\mathbf{L}} L ^ .
Table III gives the mean cosines of the final angles, and the mean change of the effective spin χ e f f = c L ^ ⋅ S 0 / m 2 \chi_{\rm eff} = c\hat{\mathbf{L}} \cdot\mathbf{S}_0/m^2 χ eff = c L ^ ⋅ S 0 / m 2 , computed binary by binary. The theoretical mean of the cosines of κ 1 \kappa_1 κ 1 , κ 2 \kappa_2 κ 2 , γ \gamma γ and Δ Φ \Delta\Phi ΔΦ vanishes for an isotropic distribution. For an isotropic ensemble, cos κ 1 \cos\kappa_1 cos κ 1 , cos κ 2 \cos\kappa_2 cos κ 2 , and cos γ \cos\gamma cos γ are uniform on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] , with variance 1 / 3 1/3 1/3 , and Δ Φ \Delta\Phi ΔΦ is uniform on [ 0 , 2 π ) [0,2\pi) [ 0 , 2 π ) , so
Table III. All numbers are in units of 10 − 3 10^{-3} 1 0 − 3 . Mean cosines of the final angles and mean change of the effective spin, Δ χ e f f \Delta\chi_{\mathrm{eff}} Δ χ eff , for ensembles of N = 10 4 N = 10^{4} N = 1 0 4 BBHs with isotropic initial directions. The last row gives the theoretical standard deviations for mean cosines. For each binary, the change of χ e f f \chi_{\mathrm{eff}} χ eff , between initial and final state, vanishes to numerical precision (∼ 10 − 10 \sim10^{-10} ∼ 1 0 − 10 ) except with spin-dependent RR, where we give ⟨ Δ χ e f f ⟩ \langle\Delta\chi_{\mathrm{eff}}\rangle ⟨ Δ χ eff ⟩ along with its error. Rows 1, 2, 6 and 7 indicate isotropy preservation (all mean cosines lie within 2 σ 2\sigma 2 σ of zero). Rows 3, 4 and 5 indicate isotropy breaking: row 5 at about 10 σ 10\sigma 10 σ , row 4 marginally at about 3 σ 3\sigma 3 σ , and row 3 through ⟨ Δ χ e f f ⟩ \langle\Delta\chi_{\mathrm{eff}}\rangle ⟨ Δ χ eff ⟩ only. All of these are in line with the theoretical predictions. Rows 8 and 9, however, show no evidence of isotropy breaking, although the theory no longer predicts isotropy preservation there.
cos Δ Φ \cos\Delta\Phi cos ΔΦ has variance 1 / 2 1/2 1/2 . The last row gives the resulting theoretical standard deviation of each mean as a point estimate, σ = 1 / ( 3 N ) \sigma= \sqrt{1/(3N)} σ = 1/ ( 3 N ) and 1 / ( 2 N ) \sqrt{1/(2N)} 1/ ( 2 N ) , respectively; see Chapter 31 of [49 ] .
When both conditions of Sec. VII B hold, all mean cosines are compatible with zero, within 2 σ 2\sigma 2 σ . This is the case for the isotropic ensemble, in the conservative regime and with spin-independent RR, and for the ensemble with random q q q , where for the latter, the argument applies to each of its two subsets (Rows 1, 2, 6, and 7 of Table III ). The correlation between q q q and the directions of the angular momenta breaks isotropy (Rows 4 and 5). This is clear in the spin-independent RR case, at about 10 σ 10\sigma 10 σ , where the spins tend to align with each other, and marginal, at about 3 σ 3\sigma 3 σ , in the conservative case. The correlation between L L L and the directions of the angular momenta does not exhibit a breaking of isotropy, although isotropy is no longer protected by our theoretical derivation (Rows 8 and 9).
With spin-dependent RR, the mean cosines are also compatible with zero, but isotropy is broken. Since χ e f f \chi_{\mathrm{eff}} χ eff is linear in cos κ 1 \cos\kappa_1 cos κ 1 and cos κ 2 \cos\kappa_2 cos κ 2 , it has a zero mean in an isotropic ensemble whose masses are independent of the directions (Row 3). It is conserved for each binary by the OA precession and spin-independent RR, as confirmed by Table III , but not by the torque on L ^ \hat{\mathbf L} L ^ caused by the spin-dependent RR. Its mean then changes by ( − 2.86 ± 0.02 ) × 10 − 3 (-2.86 \pm0.02) \times10^{-3} ( − 2.86 ± 0.02 ) × 1 0 − 3 , and by ( − 2.87 ± 0.02 ) × 10 − 3 (-2.87 \pm0.02) \times10^{-3} ( − 2.87 ± 0.02 ) × 1 0 − 3 for a second realization of the ensemble with different random initial directions. Here the error (0.02) for the sample mean (-2.86) was computed by dividing the sample standard deviation by N \sqrt{N} N [49 ] . Since the ensemble starts isotropic and the final mean of χ e f f \chi_{\mathrm{eff}} χ eff is nonzero, the final ensemble is not isotropic anymore. The sign agrees with the bias toward anti-alignment found by Lousto et al. [42 ] .
Effective spin of binary neutron stars
Hamiltonian and integrability
For a neutron star (NS), the spin-induced quadrupole moment differs from its Kerr value. It is parametrized by a constant C Q a C_{Q_a} C Q a for each body, with C Q a = 1 C_{Q_a}=1 C Q a = 1 for a BH and C Q a > 1 C_{Q_a}>1 C Q a > 1 for a NS [50 ] . Only the self-spin terms of H S S H_{\mathrm{SS}} H SS change, each being multiplied by C Q a C_{Q_a} C Q a [19 ] . Averaging them with Eqs. (11) and dropping the Casimir terms gives
H ˉ B N S = H ˉ − 3 ν 4 c 2 D 3 ∑ a ( C Q a − 1 ) Σ a 2 ( L ^ ⋅ S a ) 2 . (102) \bar{H}^{\mathrm{BNS}}=\bar{H}-\frac{3\nu}{4c^2D^3}\sum_a(C_{Q_a}-1)\Sigma_a^2(\hat{\mathbf L}\cdot\mathbf S_a)^2. \tag*{(102)} H ˉ BNS = H ˉ − 4 c 2 D 3 3 ν a ∑ ( C Q a − 1 ) Σ a 2 ( L ^ ⋅ S a ) 2 . (102)
The arguments of Sec. III C apply equally to BNSs, so H ˉ B N S \bar{H}^{\mathrm{BNS}} H ˉ BNS generates the OA BNS EOMs through Eq. (16) , with the PBs (15) . Since H ˉ B N S \bar{H}^{\mathrm{BNS}} H ˉ BNS is invariant under rotations, ( J 2 , J z , H ˉ B N S ) (J^2,J_z,\bar{H}^{\mathrm{BNS}}) ( J 2 , J z , H ˉ BNS ) are three COMs in involution, and the OA BNS system is integrable, by the same argument as for BBHs [16 , 17 ] . These results hold for any C Q a C_{Q_a} C Q a , and therefore also for NS-BH binaries.
Λ \Lambda Λ , and therefore χ e f f = c Λ / ( m 2 L ) = ( m 1 χ 1 + m 2 χ 2 ) ⋅ L ^ / m \chi_{\mathrm{eff}}=c\Lambda/(m^2L)=(m_1\boldsymbol{\chi}_1+m_2\boldsymbol{\chi}_2)\cdot\hat{\mathbf L}/m χ eff = c Λ/ ( m 2 L ) = ( m 1 χ 1 + m 2 χ 2 ) ⋅ L ^ / m , is no longer conserved [14 , 20 ] . Since { Λ , H ˉ } = 0 \{\Lambda,\bar{H}\}=0 { Λ , H ˉ } = 0 and { L ⋅ S 1 , L ⋅ S 2 } = L ⋅ ( S 1 × S 2 ) \{\mathbf L\cdot\mathbf S_1,\mathbf L\cdot\mathbf S_2\}=\mathbf L\cdot(\mathbf S_1\times\mathbf S_2) { L ⋅ S 1 , L ⋅ S 2 } = L ⋅ ( S 1 × S 2 ) , only the last term of Eq. (102) contributes, and
d Λ d t = 3 2 c 2 D 3 L 2 [ ( C Q 1 − 1 ) Σ 1 L ⋅ S 1 − ( C Q 2 − 1 ) Σ 2 L ⋅ S 2 ] L ⋅ ( S 1 × S 2 ) . (103) \frac{d\Lambda}{dt}=\frac{3}{2c^2D^3L^2}\left[(C_{Q_1}-1)\Sigma_1\mathbf L\cdot\mathbf S_1-(C_{Q_2}-1)\Sigma_2\mathbf L\cdot\mathbf S_2\right]\mathbf L\cdot(\mathbf S_1\times\mathbf S_2). \tag*{(103)} d t d Λ = 2 c 2 D 3 L 2 3 [ ( C Q 1 − 1 ) Σ 1 L ⋅ S 1 − ( C Q 2 − 1 ) Σ 2 L ⋅ S 2 ] L ⋅ ( S 1 × S 2 ) . (103)
This vanishes for BBHs, and for BNSs when the triad is coplanar. In general, χ e f f \chi_{\mathrm{eff}} χ eff oscillates on the precession timescale.
Conserved effective spin
As in Sec. IV, we can rewrite H ˉ B N S \bar{H}^{\mathrm{BNS}} H ˉ BNS of Eq. (102) in terms of J 2 J^2 J 2 , the Casimirs, and the projections L ⋅ S a \mathbf L\cdot\mathbf S_a L ⋅ S a only,
4 c 2 D 3 H ˉ B N S = J 2 − L 2 − S 1 2 − S 2 2 + 4 K , (104) 4c^2D^3\bar{H}^{\mathrm{BNS}}=J^2-L^2-S_1^2-S_2^2+4\mathcal{K}, \tag*{(104)} 4 c 2 D 3 H ˉ BNS = J 2 − L 2 − S 1 2 − S 2 2 + 4 K , (104)
with
K ≡ 3 2 Λ − 3 ν 4 L 2 [ Λ 2 + ∑ a ( C Q a − 1 ) Σ a 2 ( L ⋅ S a ) 2 ] . (105) \mathcal{K} \equiv\frac{3}{2}\Lambda-\frac{3\nu}{4L^2}\left[\Lambda^2+\sum_a(C_{Qa}-1)\Sigma_a^2(\mathbf{L}\cdot\mathbf{S}_a)^2\right]. \tag*{(105)} K ≡ 2 3 Λ − 4 L 2 3 ν [ Λ 2 + a ∑ ( C Q a − 1 ) Σ a 2 ( L ⋅ S a ) 2 ] . (105)
Since H ˉ B N S \bar{H}^{\mathrm{BNS}} H ˉ BNS and J 2 J^2 J 2 are conserved, so is K \mathcal{K} K . For BBHs, K = 3 Λ / 2 − 3 ν Λ 2 / ( 4 L 2 ) \mathcal{K}=3\Lambda/2-3\nu\Lambda^2/(4L^2) K = 3Λ/2 − 3 ν Λ 2 / ( 4 L 2 ) , and we recover Eq. (28) . For BNSs, the additional terms in ( L ⋅ S a ) 2 (\mathbf{L}\cdot\mathbf{S}_a)^2 ( L ⋅ S a ) 2 make K \mathcal{K} K no longer a function of Λ \Lambda Λ alone, and in general no linear combination of L ⋅ S 1 \mathbf{L}\cdot\mathbf{S}_1 L ⋅ S 1 and L ⋅ S 2 \mathbf{L}\cdot\mathbf{S}_2 L ⋅ S 2 is conserved.
To obtain an effective spin that reduces to χ e f f \chi_{\mathrm{eff}} χ eff , we invert the BBH relation, i.e., we define Λ B N S \Lambda^{\mathrm{BNS}} Λ BNS by K = 3 Λ B N S / 2 − 3 ν ( Λ B N S ) 2 / ( 4 L 2 ) \mathcal{K}=3\Lambda^{\mathrm{BNS}}/2-3\nu(\Lambda^{\mathrm{BNS}})^2/(4L^2) K = 3 Λ BNS /2 − 3 ν ( Λ BNS ) 2 / ( 4 L 2 ) . Of the two roots of this quadratic equation, we keep the one that reduces to Λ \Lambda Λ for C Q 1 = C Q 2 = 1 C_{Q1}=C_{Q2}=1 C Q 1 = C Q 2 = 1 ,
Λ B N S = L 2 ν ( 1 − 1 − 4 ν K 3 L 2 ) , (106) \Lambda^{\mathrm{BNS}}=\frac{L^2}{\nu}\left(1-\sqrt{1-\frac{4\nu\mathcal{K}}{3L^2}}\right), \tag*{(106)} Λ BNS = ν L 2 ( 1 − 1 − 3 L 2 4 ν K ) , (106)
the other one, which exceeds L 2 / ν L^2/\nu L 2 / ν , lies outside the PN regime. In terms of the spins, with l ≡ c L / ( μ m ) l\equiv cL/(\mu m) l ≡ c L / ( μ m ) , this gives
χ e f f B N S ≡ c Λ B N S m 2 L = l − ( l − χ e f f ) 2 + ∑ a ( C Q a − 1 ) m a 2 m 2 ( L ^ ⋅ χ a ) 2 . (107) \chi_{\mathrm{eff}}^{\mathrm{BNS}}\equiv\frac{c\Lambda^{\mathrm{BNS}}}{m^2L}=l-\sqrt{(l-\chi_{\mathrm{eff}})^2+\sum_a(C_{Qa}-1)\frac{m_a^2}{m^2}(\hat{\mathbf{L}}\cdot\boldsymbol{\chi}_a)^2}. \tag*{(107)} χ eff BNS ≡ m 2 L c Λ BNS = l − ( l − χ eff ) 2 + a ∑ ( C Q a − 1 ) m 2 m a 2 ( L ^ ⋅ χ a ) 2 . (107)
χ e f f B N S \chi_{\mathrm{eff}}^{\mathrm{BNS}} χ eff BNS is thus not an independent COM, but a function of H ˉ B N S \bar{H}^{\mathrm{BNS}} H ˉ BNS and J 2 J^2 J 2 . It is exact in C Q a − 1 C_{Qa}-1 C Q a − 1 , holds for eccentric orbits through l = c a ( 1 − e 2 ) / m l=c\sqrt{a(1-e^2)}/m l = c a ( 1 − e 2 ) / m at Newtonian order, and satisfies χ e f f B N S ≤ χ e f f \chi_{\mathrm{eff}}^{\mathrm{BNS}}\leq\chi_{\mathrm{eff}} χ eff BNS ≤ χ eff , with equality when the spins of the bodies with C Q a > 1 C_{Qa}>1 C Q a > 1 lie in the orbital plane. Unlike χ e f f \chi_{\mathrm{eff}} χ eff for BBHs [7 ] , it depends on l l l , and is therefore not conserved even under spin-independent RR. Since the OA BNS system has only three independent COMs, the conserved quantity of Eq. (40) of Ref. [20 ] , obtained there for circular orbits as a constant of integration, is a function of J 2 J^2 J 2 and χ e f f B N S \chi_{\mathrm{eff}}^{\mathrm{BNS}} χ eff BNS .
Toward action-angle variables
For the action variables, K \mathcal{K} K is the natural third COM. Since Λ B N S \Lambda^{\mathrm{BNS}} Λ BNS is a function of K \mathcal{K} K , both generate the same loops, but K \mathcal{K} K is a polynomial in L ⋅ S a \mathbf{L}\cdot\mathbf{S}_a L ⋅ S a and H ˉ B N S \bar{H}^{\mathrm{BNS}} H ˉ BNS is affine in it, with ∂ H ˉ B N S / ∂ K = 1 / ( c 2 D 3 ) \partial\bar{H}^{\mathrm{BNS}}/\partial\mathcal{K}=1/(c^2D^3) ∂ H ˉ BNS / ∂ K = 1/ ( c 2 D 3 ) . As in Eq. (70) , the mutual angles of the triad therefore evolve along the K \mathcal{K} K -flow with λ K = t / ( c 2 D 3 ) \lambda_{\mathcal{K}}=t/(c^2D^3) λ K = t / ( c 2 D 3 ) , and the nutation period is c 2 D 3 Δ λ K c^2D^3\Delta\lambda_{\mathcal{K}} c 2 D 3 Δ λ K , where Δ λ K \Delta\lambda_{\mathcal{K}} Δ λ K is the period of the K \mathcal{K} K -flow. We do not compute the BNS actions here. The construction of Sec. IV E 2 only uses flows generated by COMs and the homogeneity of these COMs. Since K \mathcal{K} K , like Λ \Lambda Λ , is homogeneous of degree two, the third action would take the form of Eq. (58) , with Λ Δ λ Λ \Lambda\Delta\lambda_{\Lambda} ΛΔ λ Λ replaced by K Δ λ K \mathcal{K}\Delta\lambda_{\mathcal{K}} K Δ λ K . The difficulty lies in the flow amounts. For BBHs, fixing J 2 J^2 J 2 and Λ \Lambda Λ makes all the scalar products of the triad affine in the single variable f f f . For BNSs, K \mathcal{K} K is quadratic in L ⋅ S a \mathbf{L}\cdot\mathbf{S}_a L ⋅ S a , the scalar products are no longer affine in a single variable, and the elliptic integrals of App. C no longer apply. We leave the computation of the flow amounts, and hence of the BNS actions, frequencies, and angles, to future work.
Concluding remarks
Scope of the present work
Most analytical results obtained so far for the OA spinning BBH system rely on the conservation of the projected effective spin [4 , 7 , 9 , 10 , 12 , 13 ] . In this work, we studied this system from the point of view of Hamiltonian mechanics. Our main results are the following:
Hamiltonian nature: In Sec. III, we confirmed that the 2PN OA system is Hamiltonian [16 ] , with the SO(3) Poisson brackets of L \mathbf{L} L , S 1 \mathbf{S}_1 S 1 , and S 2 \mathbf{S}_2 S 2 . We derived its Hamiltonian both by orbit averaging the full 2PN Hamiltonian, which requires the conservation of J \mathbf{J} J for L \mathbf{L} L , and from an ansatz built from the OA EOMs alone.
Integrability and action-angle variables: In Secs. IV and § 3 , we showed that the OA system is integrable, as noted in Ref. [16 ] , and computed its action variables, frequencies, and angle variables, which give an action-angle-based solution. The actions were obtained with the fictitious variables of Ref. [18 ] , a second application of this method. We also obtained four further results. First, we related the OA COMs to the COMs of the full non-OA system at 2PN. Second, we showed that the actions computed in the EPS and in the SPS differ only by integer multiples of the Casimirs. These two results clarify points discussed in Refs. [31 , 18 ] , respectively. Third, we computed the frequencies without differentiating the elliptic integrals. Fourth, these frequencies confirm that the spin-orbit resonances are not resonances in the usual sense, as argued in Ref. [16 ] .
Breakdown of integrability and separatrices: In Sec. VI, we proposed a criterion for the breakdown of the integrability conditions, based on the Jacobian between the COMs and the actions. It applies to any integrable Hamiltonian system. For the pendulum and the Kepler problem, it recovers the standard separatrices. For the OA BBH system, it recovers the up-down instability, the logarithmic divergence of its nutation period, and the thresholds L ± L_{\pm} L ± [12 ] . It also gives these thresholds explicitly for eccentric orbits, for which the onset is set by a ( 1 − e 2 ) a(1-e^2) a ( 1 − e 2 ) , consistent with the linear analysis of Ref. [15 ] , and we verified this numerically. For the non-OA 1.5PN system, it gives both the separatrix between bound and unbound orbits and the up-down instability.
Ensembles of binaries: In Sec. VII, we showed that the preservation of isotropy observed in Ref. [8 ] follows from the Hamiltonian nature of the OA system. Our derivation predicts that isotropy is also preserved for eccentric orbits, and that it can be lost when the parameters of the members are correlated with the directions of their angular momenta, or when RR depends on the spins. We verified these predictions numerically.
Binary neutron stars: In Sec. VIII, we used the conserved OA Hamiltonian of BNSs to generalize χ e f f \chi_{\mathrm{eff}} χ eff to BNSs. The result is exact in the quadrupole parameters and holds for eccentric orbits. We also identified K \mathcal{K} K as the natural third COM for their action-angle variables.
Future perspectives
For OA BNSs, the next step is to compute the action-angle variables from the flow of K \mathcal{K} K (Sec. VIII). Alternatively, since the quadrupole terms are of relative 0.5PN order with respect to the leading spin-orbit coupling, approximate BNS action-angle variables can be constructed from the BBH ones with canonical perturbation theory [29 ] . The criterion of Sec. VI would then give their separatrices, complementing the linear stability analysis of Ref. [15]. Since χ e f f B N S \chi_{\rm eff}^{\rm BNS} χ eff BNS depends on L L L , it is not conserved even under spin-independent RR, unlike χ e f f \chi_{\rm eff} χ eff for BBHs, and its evolution during the inspiral also remains to be studied. Finally, since our criterion is sufficient but not necessary (Sec. VI C 4), whether it captures all unstable equilibria remains an open question.
Acknowledgments
We would like to thank Laura Bernard and José Tomás Gálvez Ghersi for motivating us to pursue this work. We thank Leo Stein, Samuel Lisi, Guillaume Faye, Paul Ramond, Soichiro Isoyama, Thibault Damour, Alan Weinstein, and Davide Gerosa for fruitful discussions. S.T. was supported by the PSL Postdoctoral Fellowship and UTEC Internal Research Grant during the course of this work.
Loop integrals and the poles of the azimuthal coordinates
The azimuth ϕ V \phi_V ϕ V of V ∈ { L , S 1 , S 2 } \mathbf{V} \in\{\mathbf{L},\mathbf{S}_1,\mathbf{S}_2\} V ∈ { L , S 1 , S 2 } about z ^ \hat{\mathbf{z}} z ^ is undefined at the poles V ∥ ± z ^ \mathbf{V} \parallel\pm\hat{\mathbf{z}} V ∥ ± z ^ . In this appendix, we evaluate the integrals of Eq. (40) for any orientation of the z z z -axis, derive the relation (51) between the EPS and SPS integrands, and show that the resulting additive constants do not affect the dynamics. We first fix the sign convention of Stokes’ theorem.
Sign convention with Stokes' theorem
With the x x x and y y y axes pointing right and up, let a loop in the ( x , y ) (x,y) ( x , y ) plane enclose a region of area A > 0 A>0 A > 0 . Since ∮ d ( x y ) = 0 \oint d(xy)=0 ∮ d ( x y ) = 0 , we have
∮ y d x = − ∮ x d y = { − A , loop traversed counterclockwise , + A , loop traversed clockwise . (108) \oint y\,dx=-\oint x\,dy=
\begin{cases}
-A, & \text{loop traversed counterclockwise},\\
+A, & \text{loop traversed clockwise}.
\end{cases} \tag*{(108)} ∮ y d x = − ∮ x d y = { − A , + A , loop traversed counterclockwise , loop traversed clockwise . (108)
J2-loop for any position of the poles
To evaluate ∮ V z d ϕ V \oint V_z\,d\phi_V ∮ V z d ϕ V along the J 2 J^2 J 2 -loop, we use Stokes’ theorem in the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane. ∮ V z d ϕ V \oint V_z\,d\phi_V ∮ V z d ϕ V equals the area enclosed by the contour, counted positively for a clockwise contour and negatively for a counterclockwise one, and this area is $1/V$ times the corresponding area on the sphere. Stokes’ theorem requires the integrand to be defined on the contour and everywhere inside it, so the enclosed region must contain no pole. In the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane, $V_z ranges from − V -V − V to V V V , so that the north and south poles become the top and bottom edges, V z = ± V V_z=\pm V V z = ± V (Figure 4 ). Seen from outside the sphere, the loop runs counterclockwise around the cap, of area 2 π V ( V − V ⋅ J ^ ) 2\pi V(V-\mathbf{V}\cdot\hat{\mathbf{J}}) 2 π V ( V − V ⋅ J ^ ) , and its image in the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane runs in the same sense. Recall that we gave a specific definition of cap in Sec. IV D. Four cases arise, depending on whether the north and south poles lie inside the cap (n N = 1 n_N=1 n N = 1 and n S = 1 n_S=1 n S = 1 , respectively):
Figure 4. The J 2 J^2 J 2 -loop (green) in the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane, for the four positions of the poles relative to the cap (hatched). The north and south poles are the top and bottom edges, and the dashed line is the equator. The shaded region contains no pole and is the one to which Stokes’ theorem is applied. In (i) and (ii), the loop is closed by a meridian arc and the equator (blue), as in Figure 1 (c); in (iii) and (iv), the loop is itself a closed contour.
(i) North pole only (n N = 1 n_N=1 n N = 1 , n S = 0 n_S=0 n S = 0 ) [Figure 4 (i)]. This is the case of Sec. IV D, where closing the loop with a meridian arc and the equator gives ∮ V z d ϕ V = 2 π V ⋅ J ^ \oint V_z\,d\phi_V=2\pi\mathbf{V}\cdot\hat{\mathbf{J}} ∮ V z d ϕ V = 2 π V ⋅ J ^ [Eq. (42) ].
(ii) South pole only (n N = 0 n_N=0 n N = 0 , n S = 1 n_S=1 n S = 1 ) [Figure 4 (ii)]. Since the cap contains the south pole but not the north pole, J ^ z < 0 \hat{J}_z<0 J ^ z < 0 , and the loop goes once around the z z z -axis with ϕ V \phi_V ϕ V decreasing by 2 π 2\pi 2 π . Closing it with a meridian arc and the equator, as in case (i), and using Eq. (108) , we again obtain ∮ V z d ϕ V = 2 π V ⋅ J ^ \oint V_z\,d\phi_V=2\pi\mathbf{V}\cdot\hat{\mathbf{J}} ∮ V z d ϕ V = 2 π V ⋅ J ^ .
(iii) No pole (n N = n S = 0 n_N=n_S=0 n N = n S = 0 ) [Figure 4 (iii)]. The loop does not go around the z z z -axis, so the net change of ϕ V \phi_V ϕ V along it is zero, instead of ± 2 π \pm2\pi ± 2 π in cases (i) and (ii). In the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane, the loop is an oval that touches neither the top nor the bottom edge, and the cap is the inside of the oval. The oval thus runs counterclockwise in the ( ϕ V , V z ) (\phi_V,V_z) ( ϕ V , V z ) plane, and ∮ V z d ϕ V = − 2 π V ( V − V ⋅ J ^ ) / V = − 2 π ( V − V ⋅ J ^ ) \oint V_z\,d\phi_V=-2\pi V(V-\mathbf{V}\cdot\hat{\mathbf{J}})/V=-2\pi(V-\mathbf{V}\cdot\hat{\mathbf{J}}) ∮ V z d ϕ V = − 2 π V ( V − V ⋅ J ^ ) / V = − 2 π ( V − V ⋅ J ^ ) . This is negative as per Eq. (108) .
(iv) Both poles (n N = n S = 1 n_N=n_S=1 n N = n S = 1 ) [Figure 4 (iv)]. The loop is again an oval, but the cap now touches both the top and the bottom edge, so it lies outside the oval. The inside of the oval is the complementary cap, centered on − J ^ -\hat{\mathbf{J}} − J ^ , which contains no pole and has area 4 π V 2 − 2 π V ( V − V ⋅ J ^ ) 4\pi V^2-2\pi V(V-\mathbf{V}\cdot\hat{\mathbf{J}}) 4 π V 2 − 2 π V ( V − V ⋅ J ^ ) . Seen from outside the sphere, a loop running counterclockwise around the cap runs clockwise around the complementary cap. The oval thus runs clockwise, and ∮ V z d ϕ V = [ 4 π V 2 − 2 π V ( V − V ⋅ J ^ ) ] / V = 2 π ( V ⋅ J ^ + V ) > 0 \oint V_z\,d\phi_V=[4\pi V^2-2\pi V(V-\mathbf{V}\cdot\hat{\mathbf{J}})]/V=2\pi(\mathbf{V}\cdot\hat{\mathbf{J}}+V)>0 ∮ V z d ϕ V = [ 4 π V 2 − 2 π V ( V − V ⋅ J ^ )] / V = 2 π ( V ⋅ J ^ + V ) > 0 .
The four cases combine into
∮ l o o p V z d ϕ V = 2 π [ V ⋅ J ^ + ( n N + n S − 1 ) V ] . (109) \oint_{\mathrm{loop}} V_z\,d\phi_V = 2\pi\left[\mathbf{V}\cdot\hat{\mathbf{J}}+(n_N+n_S-1)V\right]. \tag*{(109)} ∮ loop V z d ϕ V = 2 π [ V ⋅ J ^ + ( n N + n S − 1 ) V ] . (109)
The integral is therefore invariant under rotations of the coordinate axes, as long as the number of poles inside the cap remains the same. When a pole lies exactly on the loop, ϕ V \phi_V ϕ V is undefined at one point of it. This happens only when a pole crosses the loop, i.e., in the transition between two cases, where n N n_N n N or n S n_S n S changes by one and the integral jumps by 2 π V 2\pi V 2 π V . Such orientations of z ^ \hat{\mathbf{z}} z ^ form a set of measure zero and are excluded. The same holds for any loop on the invariant torus: a deformation that crosses a pole shifts the contribution of V V V to the action by ± V \pm V ± V .
EPS and SPS integrands
Let θ V \theta_V θ V and ϕ V \phi_V ϕ V be the polar and azimuthal angles of V \mathbf{V} V about z ^ \hat{\mathbf{z}} z ^ , so that V z = V cos θ V V_z=V\cos\theta_V V z = V cos θ V , and let
e ^ θ = ( cos θ V cos ϕ V , cos θ V sin ϕ V , − sin θ V ) , e ^ ϕ = ( − sin ϕ V , cos ϕ V , 0 ) (110) \hat{\mathbf{e}}_\theta=(\cos\theta_V\cos\phi_V,\ \cos\theta_V\sin\phi_V,\ -\sin\theta_V),\qquad\hat{\mathbf{e}}_\phi=(-\sin\phi_V,\ \cos\phi_V,\ 0) \tag*{(110)} e ^ θ = ( cos θ V cos ϕ V , cos θ V sin ϕ V , − sin θ V ) , e ^ ϕ = ( − sin ϕ V , cos ϕ V , 0 ) (110)
be the spherical unit vectors at V ^ \hat{\mathbf{V}} V ^ : e ^ θ \hat{\mathbf{e}}_\theta e ^ θ is tangent to the meridian and points toward the south pole, and e ^ θ × e ^ ϕ = V ^ \hat{\mathbf{e}}_\theta\times\hat{\mathbf{e}}_\phi=\hat{\mathbf{V}} e ^ θ × e ^ ϕ = V ^ . Since r V ⊥ V \mathbf{r}_V\perp\mathbf{V} r V ⊥ V , we define ψ V \psi_V ψ V , the angle of r V \mathbf{r}_V r V about V \mathbf{V} V such that
r ^ V = cos ψ V e ^ θ + sin ψ V e ^ ϕ . (111) \hat{\mathbf{r}}_V=\cos\psi_V\,\hat{\mathbf{e}}_\theta+\sin\psi_V\,\hat{\mathbf{e}}_\phi. \tag*{(111)} r ^ V = cos ψ V e ^ θ + sin ψ V e ^ ϕ . (111)
We compute in two ways the angle V ^ ⋅ ( r ^ V × d r ^ V ) \hat{\mathbf{V}}\cdot(\hat{\mathbf{r}}_V\times d\hat{\mathbf{r}}_V) V ^ ⋅ ( r ^ V × d r ^ V ) by which r ^ V \hat{\mathbf{r}}_V r ^ V turns about V ^ \hat{\mathbf{V}} V ^ along a flow of the form (48) .
From the flow. Along the flow, d r ^ V = ( ∂ F / ∂ V ) × r ^ V d λ F d\hat{\mathbf{r}}_V=(\partial F/\partial\mathbf{V})\times\hat{\mathbf{r}}_V\,d\lambda_F d r ^ V = ( ∂ F / ∂ V ) × r ^ V d λ F , so that r ^ V × d r ^ V = [ ∂ F / ∂ V − r ^ V ( r ^ V ⋅ ∂ F / ∂ V ) ] d λ F \hat{\mathbf{r}}_V\times d\hat{\mathbf{r}}_V=[\partial F/\partial\mathbf{V}-\hat{\mathbf{r}}_V(\hat{\mathbf{r}}_V\cdot\partial F/\partial\mathbf{V})]d\lambda_F r ^ V × d r ^ V = [ ∂ F / ∂ V − r ^ V ( r ^ V ⋅ ∂ F / ∂ V )] d λ F . Since r ^ V ⊥ V ^ \hat{\mathbf{r}}_V\perp\hat{\mathbf{V}} r ^ V ⊥ V ^ , only the first term survives the projection on V ^ \hat{\mathbf{V}} V ^ , and Eq. (45) gives
p V ⋅ d r V = V ⋅ ∂ F ∂ V d λ F = V V ^ ⋅ ( r ^ V × d r ^ V ) . (112) \mathbf{p}_V\cdot d\mathbf{r}_V=\mathbf{V}\cdot\frac{\partial F}{\partial\mathbf{V}}\,d\lambda_F=V\hat{\mathbf{V}}\cdot(\hat{\mathbf{r}}_V\times d\hat{\mathbf{r}}_V). \tag*{(112)} p V ⋅ d r V = V ⋅ ∂ V ∂ F d λ F = V V ^ ⋅ ( r ^ V × d r ^ V ) . (112)
From the frame. Differentiating Eq. (111) and using V ^ ⋅ ( r ^ V × d r ^ V ) = ( V ^ × r ^ V ) ⋅ d r ^ V \hat{\mathbf{V}}\cdot(\hat{\mathbf{r}}_V\times d\hat{\mathbf{r}}_V)=(\hat{\mathbf{V}}\times\hat{\mathbf{r}}_V)\cdot d\hat{\mathbf{r}}_V V ^ ⋅ ( r ^ V × d r ^ V ) = ( V ^ × r ^ V ) ⋅ d r ^ V , with V ^ × r ^ V = − sin ψ V e ^ θ + cos ψ V e ^ ϕ \hat{\mathbf{V}}\times\hat{\mathbf{r}}_V=-\sin\psi_V\,\hat{\mathbf{e}}_\theta+\cos\psi_V\,\hat{\mathbf{e}}_\phi V ^ × r ^ V = − sin ψ V e ^ θ + cos ψ V e ^ ϕ , we obtain
V ^ ⋅ ( r ^ V × d r ^ V ) = d ψ V + cos 2 ψ V e ^ ϕ ⋅ d e ^ θ − sin 2 ψ V e ^ θ ⋅ d e ^ ϕ = d ψ V + e ^ ϕ ⋅ d e ^ θ , (113) \hat{\mathbf{V}}\cdot(\hat{\mathbf{r}}_V\times d\hat{\mathbf{r}}_V)=d\psi_V+\cos^2\psi_V\,\hat{\mathbf{e}}_\phi\cdot d\hat{\mathbf{e}}_\theta-\sin^2\psi_V\,\hat{\mathbf{e}}_\theta\cdot d\hat{\mathbf{e}}_\phi=d\psi_V+\hat{\mathbf{e}}_\phi\cdot d\hat{\mathbf{e}}_\theta, \tag*{(113)} V ^ ⋅ ( r ^ V × d r ^ V ) = d ψ V + cos 2 ψ V e ^ ϕ ⋅ d e ^ θ − sin 2 ψ V e ^ θ ⋅ d e ^ ϕ = d ψ V + e ^ ϕ ⋅ d e ^ θ , (113)
where we used e ^ θ ⋅ d e ^ θ = e ^ ϕ ⋅ d e ^ ϕ = 0 \hat{\mathbf{e}}_\theta\cdot d\hat{\mathbf{e}}_\theta=\hat{\mathbf{e}}_\phi\cdot d\hat{\mathbf{e}}_\phi=0 e ^ θ ⋅ d e ^ θ = e ^ ϕ ⋅ d e ^ ϕ = 0 and e ^ θ ⋅ d e ^ ϕ = − e ^ ϕ ⋅ d e ^ θ \hat{\mathbf{e}}_\theta\cdot d\hat{\mathbf{e}}_\phi=-\hat{\mathbf{e}}_\phi\cdot d\hat{\mathbf{e}}_\theta e ^ θ ⋅ d e ^ ϕ = − e ^ ϕ ⋅ d e ^ θ . The first term is the rotation of r ^ V \hat{\mathbf{r}}_V r ^ V relative to the frame, and the second is the rotation of the frame itself about V ^ \hat{\mathbf{V}} V ^ . From the components of e ^ θ \hat{\mathbf{e}}_\theta e ^ θ and e ^ ϕ \hat{\mathbf{e}}_\phi e ^ ϕ , e ^ ϕ ⋅ d e ^ θ = cos θ V d ϕ V \hat{\mathbf{e}}_\phi\cdot d\hat{\mathbf{e}}_\theta=\cos\theta_V\,d\phi_V e ^ ϕ ⋅ d e ^ θ = cos θ V d ϕ V , we have
V ^ ⋅ ( r ^ V × d r ^ V ) = d ψ V + cos θ V d ϕ V . (114) \hat{\mathbf{V}}\cdot(\hat{\mathbf{r}}_V\times d\hat{\mathbf{r}}_V)=d\psi_V+\cos\theta_V\,d\phi_V. \tag*{(114)} V ^ ⋅ ( r ^ V × d r ^ V ) = d ψ V + cos θ V d ϕ V . (114)
Inserting Eq. (114) into Eq. (112) , we finally have Eq. (51) ,
p V ⋅ d r V = V z d ϕ V + V d ψ V . (115) \mathbf{p}_V\cdot d\mathbf{r}_V=V_z\,d\phi_V+V\,d\psi_V. \tag*{(115)} p V ⋅ d r V = V z d ϕ V + V d ψ V . (115)
The first term is the SPS integrand, due to the motion of V \mathbf{V} V and the second is due to the rotation of the pair about V \mathbf{V} V , which the SPS integrand does not see. Like ϕ V \phi_V ϕ V , the angle ψ V \psi_V ψ V is undefined at the poles.
Additive constants
By Eqs. (109) and (52) , the actions are defined up to constants n L L + n S 1 S 1 + n S 2 S 2 n_L L+n_{S_1}S_1+n_{S_2}S_2 n L L + n S 1 S 1 + n S 2 S 2 with integer n V n_V n V ’s. These are combinations of the Casimirs, which are fixed parameters on the SPS. The integers can jump only where a loop crosses a pole, and are constant on the tori between such crossings. On these tori, the actions are thus shifted by constants, which leaves the angle variables and the frequencies unchanged, since both involve only derivatives with respect to the COMs at fixed Casimirs (see Proposition 11.3 of Ref. [32 ] ).
Derivative of the third action with respect to Λ \Lambda Λ
We derive Eqs. (67) . Denote the canonical pairs of the EPS collectively by ( q i , p i ) (q^i,p_i) ( q i , p i ) , with a sum over repeated i i i implied. We vary Λ \Lambda Λ at fixed J 2 J^2 J 2 , J z J_z J z , and Casimirs. Each value of Λ \Lambda Λ labels a torus, and on it the loop of Sec. IV E 2, which we parametrize by τ ∈ [ 0 , 2 π ) \tau\in[0,2\pi) τ ∈ [ 0 , 2 π ) . The loop is then described by q i ( τ , Λ ) q^i(\tau,\Lambda) q i ( τ , Λ ) and p i ( τ , Λ ) p_i(\tau,\Lambda) p i ( τ , Λ ) . Since the interval of integration in
2 π J 3 = ∮ p i d q i = ∫ 0 2 π p i ∂ q i ∂ τ d τ (116) 2\pi\mathcal{J}_3=\oint p_i\,dq^i=\int_0^{2\pi}p_i\frac{\partial q^i}{\partial\tau}\,d\tau \tag*{(116)} 2 π J 3 = ∮ p i d q i = ∫ 0 2 π p i ∂ τ ∂ q i d τ (116)
does not depend on Λ \Lambda Λ , the derivative goes under the integral sign
2 π ∂ J 3 ∂ Λ = ∫ 0 2 π ( ∂ p i ∂ Λ ∂ q i ∂ τ + p i ∂ 2 q i ∂ Λ ∂ τ ) d τ . (117) 2\pi\frac{\partial\mathcal{J}_3}{\partial\Lambda}=\int_0^{2\pi}\left(\frac{\partial p_i}{\partial\Lambda}\frac{\partial q^i}{\partial\tau}+p_i\frac{\partial^2q^i}{\partial\Lambda\partial\tau}\right)d\tau. \tag*{(117)} 2 π ∂ Λ ∂ J 3 = ∫ 0 2 π ( ∂ Λ ∂ p i ∂ τ ∂ q i + p i ∂ Λ ∂ τ ∂ 2 q i ) d τ . (117)
Integrating the second term by parts, with the boundary term vanishing because the loop is closed for every Λ \Lambda Λ , gives
2 π ∂ J 3 ∂ Λ = ∫ 0 2 π ( ∂ p i ∂ Λ ∂ q i ∂ τ − ∂ q i ∂ Λ ∂ p i ∂ τ ) d τ . (118) 2\pi\frac{\partial\mathcal{J}_3}{\partial\Lambda}=\int_0^{2\pi}\left(\frac{\partial p_i}{\partial\Lambda}\frac{\partial q^i}{\partial\tau}-\frac{\partial q^i}{\partial\Lambda}\frac{\partial p_i}{\partial\tau}\right)d\tau. \tag*{(118)} 2 π ∂ Λ ∂ J 3 = ∫ 0 2 π ( ∂ Λ ∂ p i ∂ τ ∂ q i − ∂ Λ ∂ q i ∂ τ ∂ p i ) d τ . (118)
On the segment of the loop generated by F ∈ { Λ , J 2 , L 2 , S 1 2 , S 2 2 } F\in\{\Lambda,J^2,L^2,S_1^2,S_2^2\} F ∈ { Λ , J 2 , L 2 , S 1 2 , S 2 2 } , Eq. (35) gives ∂ q i / ∂ τ = ( ∂ F / ∂ p i ) d λ F / d τ \partial q^i/\partial\tau=(\partial F/\partial p_i)\,d\lambda_F/d\tau ∂ q i / ∂ τ = ( ∂ F / ∂ p i ) d λ F / d τ and ∂ p i / ∂ τ = − ( ∂ F / ∂ q i ) d λ F / d τ \partial p_i/\partial\tau=-(\partial F/\partial q^i)\,d\lambda_F/d\tau ∂ p i / ∂ τ = − ( ∂ F / ∂ q i ) d λ F / d τ , and by the chain rule the integrand of Eq. (118) becomes
( ∂ F ∂ p i ∂ p i ∂ Λ + ∂ F ∂ q i ∂ q i ∂ Λ ) d λ F d τ = ∂ F ( q ( τ , Λ ) , p ( τ , Λ ) ) ∂ Λ d λ F d τ . (119) \left(\frac{\partial F}{\partial p_i}\frac{\partial p_i}{\partial\Lambda}+\frac{\partial F}{\partial q^i}\frac{\partial q^i}{\partial\Lambda}\right)\frac{d\lambda_F}{d\tau}
=\frac{\partial F(q(\tau,\Lambda),p(\tau,\Lambda))}{\partial\Lambda}\frac{d\lambda_F}{d\tau}. \tag*{(119)} ( ∂ p i ∂ F ∂ Λ ∂ p i + ∂ q i ∂ F ∂ Λ ∂ q i ) d τ d λ F = ∂ Λ ∂ F ( q ( τ , Λ ) , p ( τ , Λ )) d τ d λ F . (119)
Along the loop, F F F equals its constant value on the torus. Hence the derivative on the right-hand side is 1 for F = Λ F=\Lambda F = Λ , and 0 for the other four, which are held fixed. Only the Λ \Lambda Λ -segment contributes, and integrating d λ Λ / d τ d\lambda_\Lambda/d\tau d λ Λ / d τ over it gives Δ λ Λ \Delta\lambda_\Lambda Δ λ Λ , so that ∂ J 3 / ∂ Λ ∣ J 2 = \left.\partial\mathcal{J}_3/\partial\Lambda\right|_{J^2}= ∂ J 3 / ∂ Λ ∣ J 2 = Δ λ Λ / ( 2 π ) \Delta\lambda_{\Lambda}/(2\pi) Δ λ Λ / ( 2 π ) . Varying instead J 2 J^2 J 2 at fixed Λ \Lambda Λ , only the J 2 J^2 J 2 -segment contributes, which gives
∂ J 3 ∂ J 2 ∣ Λ = Δ λ J 2 2 π . \left.\frac{\partial J_3}{\partial J^2}\right|_{\Lambda}=\frac{\Delta\lambda_{J^2}}{2\pi}. ∂ J 2 ∂ J 3 Λ = 2 π Δ λ J 2 .
The Lambda-flow and the flow amounts
In this appendix, we give the explicit solution of the flow generated by Λ \Lambda Λ and the five flow amounts Δ λ Λ \Delta\lambda_{\Lambda} Δ λ Λ , Δ λ J 2 \Delta\lambda_{J^2} Δ λ J 2 , Δ λ L 2 \Delta\lambda_{L^2} Δ λ L 2 , Δ λ S 1 2 \Delta\lambda_{S_1^2} Δ λ S 1 2 , and Δ λ S 2 2 \Delta\lambda_{S_2^2} Δ λ S 2 2 that close the loop of Sec. IV E 2 and enter the third action through Eq. (58) . The Λ \Lambda Λ -flow, Eqs. (53) and (54) , coincides with the flow generated by L ⋅ S e f f \mathbf{L}\cdot\mathbf{S}_{\rm eff} L ⋅ S eff in the non-OA 1.5PN system of Ref. [18] under the replacement σ a → Σ a \sigma_a\to\Sigma_a σ a → Σ a , i.e., S e f f → S 0 \mathbf{S}_{\rm eff}\to\mathbf{S}_0 S eff → S 0 and L ⋅ S e f f → Λ \mathbf{L}\cdot\mathbf{S}_{\rm eff}\to\Lambda L ⋅ S eff → Λ . The derivations of that reference therefore carry over directly, and we only adapt the results to the 2PN OA problem. We treat in turn the mutual angles of the triad, which yield Δ λ Λ \Delta\lambda_{\Lambda} Δ λ Λ (App. C 1), the orientation of L \mathbf{L} L about J ^ \hat{\mathbf{J}} J ^ , which yields Δ λ J 2 \Delta\lambda_{J^2} Δ λ J 2 (App. C 2), and the rotation of the fictitious pairs about their physical vectors, which yields Δ λ L 2 \Delta\lambda_{L^2} Δ λ L 2 , Δ λ S 1 2 \Delta\lambda_{S_1^2} Δ λ S 1 2 , and Δ λ S 2 2 \Delta\lambda_{S_2^2} Δ λ S 2 2 (App. C 3).
Mutual angles and Delta lambda Lambda
Along the Λ \Lambda Λ -flow, the Casimirs and the COMs J 2 J^2 J 2 and Λ \Lambda Λ are fixed. The three scalar products of the triad are then related by
Σ 1 L ⋅ S 1 + Σ 2 L ⋅ S 2 = Λ , (120) \Sigma_1\mathbf{L}\cdot\mathbf{S}_1+\Sigma_2\mathbf{L}\cdot\mathbf{S}_2=\Lambda, \tag*{(120)} Σ 1 L ⋅ S 1 + Σ 2 L ⋅ S 2 = Λ , (120)
L ⋅ S 1 + L ⋅ S 2 + S 1 ⋅ S 2 = 1 2 ( J 2 − L 2 − S 1 2 − S 2 2 ) , (121) \mathbf{L}\cdot\mathbf{S}_1+\mathbf{L}\cdot\mathbf{S}_2+\mathbf{S}_1\cdot\mathbf{S}_2=\frac{1}{2}\left(J^2-L^2-S_1^2-S_2^2\right), \tag*{(121)} L ⋅ S 1 + L ⋅ S 2 + S 1 ⋅ S 2 = 2 1 ( J 2 − L 2 − S 1 2 − S 2 2 ) , (121)
so that only one of them is free. We choose
f ≡ S 1 ⋅ S 2 Σ 1 − Σ 2 , (122) f\equiv\frac{\mathbf{S}_1\cdot\mathbf{S}_2}{\Sigma_1-\Sigma_2}, \tag*{(122)} f ≡ Σ 1 − Σ 2 S 1 ⋅ S 2 , (122)
in terms of which Eqs. (C1) give
S 1 ⋅ S 2 = ( Σ 1 − Σ 2 ) f , (123) \mathbf{S}_1\cdot\mathbf{S}_2=(\Sigma_1-\Sigma_2)f, \tag*{(123)} S 1 ⋅ S 2 = ( Σ 1 − Σ 2 ) f , (123)
L ⋅ S 1 = Σ 2 ( f − Δ 1 ) , (124) \mathbf{L}\cdot\mathbf{S}_1=\Sigma_2(f-\Delta_1), \tag*{(124)} L ⋅ S 1 = Σ 2 ( f − Δ 1 ) , (124)
L ⋅ S 2 = Σ 1 ( Δ 2 − f ) , (125) \mathbf{L}\cdot\mathbf{S}_2=\Sigma_1(\Delta_2-f), \tag*{(125)} L ⋅ S 2 = Σ 1 ( Δ 2 − f ) , (125)
with the constants
Δ a = 1 Σ 1 − Σ 2 [ 1 2 ( J 2 − L 2 − S 1 2 − S 2 2 ) − Λ Σ b ] , ( a , b ) ∈ { ( 1 , 2 ) , ( 2 , 1 ) } . (126) \Delta_a=\frac{1}{\Sigma_1-\Sigma_2}\left[\frac{1}{2}\left(J^2-L^2-S_1^2-S_2^2\right)-\frac{\Lambda}{\Sigma_b}\right],\qquad(a,b)\in\{(1,2),(2,1)\}. \tag*{(126)} Δ a = Σ 1 − Σ 2 1 [ 2 1 ( J 2 − L 2 − S 1 2 − S 2 2 ) − Σ b Λ ] , ( a , b ) ∈ {( 1 , 2 ) , ( 2 , 1 )} . (126)
Using Eq. (54) , we obtain
d f d λ Λ = L ⋅ ( S 1 × S 2 ) . (127) \frac{df}{d\lambda_{\Lambda}}=\mathbf{L}\cdot(\mathbf{S}_1\times\mathbf{S}_2). \tag*{(127)} d λ Λ df = L ⋅ ( S 1 × S 2 ) . (127)
The square of the triple product is fixed by the magnitudes and the scalar products,
[ L ⋅ ( S 1 × S 2 ) ] 2 = L 2 S 1 2 S 2 2 + 2 ( L ⋅ S 1 ) ( L ⋅ S 2 ) ( S 1 ⋅ S 2 ) − L 2 ( S 1 ⋅ S 2 ) 2 − S 1 2 ( L ⋅ S 2 ) 2 − S 2 2 ( L ⋅ S 1 ) 2 . (128) \left[\mathbf{L}\cdot(\mathbf{S}_1\times\mathbf{S}_2)\right]^2 = L^2 S_1^2 S_2^2+2(\mathbf{L}\cdot\mathbf{S}_1)(\mathbf{L}\cdot\mathbf{S}_2)(\mathbf{S}_1\cdot\mathbf{S}_2)-L^2(\mathbf{S}_1\cdot\mathbf{S}_2)^2-S_1^2(\mathbf{L}\cdot\mathbf{S}_2)^2-S_2^2(\mathbf{L}\cdot\mathbf{S}_1)^2. \tag*{(128)} [ L ⋅ ( S 1 × S 2 ) ] 2 = L 2 S 1 2 S 2 2 + 2 ( L ⋅ S 1 ) ( L ⋅ S 2 ) ( S 1 ⋅ S 2 ) − L 2 ( S 1 ⋅ S 2 ) 2 − S 1 2 ( L ⋅ S 2 ) 2 − S 2 2 ( L ⋅ S 1 ) 2 . (128)
Inserting Eq. (C3), we obtain a cubic polynomial in f f f ,
( d f d λ Λ ) 2 = P ( f ) ≡ a 3 f 3 + a 2 f 2 + a 1 f + a 0 , (129) \left(\frac{df}{d\lambda_\Lambda}\right)^2=P(f)\equiv a_3f^3+a_2f^2+a_1f+a_0, \tag*{(129)} ( d λ Λ df ) 2 = P ( f ) ≡ a 3 f 3 + a 2 f 2 + a 1 f + a 0 , (129)
with
a 3 = 2 Σ 1 Σ 2 ( Σ 2 − Σ 1 ) , (130) a_3=2\Sigma_1\Sigma_2(\Sigma_2-\Sigma_1), \tag*{(130)} a 3 = 2 Σ 1 Σ 2 ( Σ 2 − Σ 1 ) , (130)
a 2 = 2 Σ 1 Σ 2 ( Σ 1 − Σ 2 ) ( Δ 1 + Δ 2 ) − L 2 ( Σ 1 − Σ 2 ) 2 − Σ 1 2 S 1 2 − Σ 2 2 S 2 2 , (131) a_2=2\Sigma_1\Sigma_2(\Sigma_1-\Sigma_2)(\Delta_1+\Delta_2)-L^2(\Sigma_1-\Sigma_2)^2-\Sigma_1^2S_1^2-\Sigma_2^2S_2^2, \tag*{(131)} a 2 = 2 Σ 1 Σ 2 ( Σ 1 − Σ 2 ) ( Δ 1 + Δ 2 ) − L 2 ( Σ 1 − Σ 2 ) 2 − Σ 1 2 S 1 2 − Σ 2 2 S 2 2 , (131)
a 1 = 2 [ Σ 1 2 S 1 2 Δ 2 + Σ 2 2 S 2 2 Δ 1 + Σ 1 Σ 2 ( Σ 2 − Σ 1 ) Δ 1 Δ 2 ] , (132) a_1=2\left[\Sigma_1^2S_1^2\Delta_2+\Sigma_2^2S_2^2\Delta_1+\Sigma_1\Sigma_2(\Sigma_2-\Sigma_1)\Delta_1\Delta_2\right], \tag*{(132)} a 1 = 2 [ Σ 1 2 S 1 2 Δ 2 + Σ 2 2 S 2 2 Δ 1 + Σ 1 Σ 2 ( Σ 2 − Σ 1 ) Δ 1 Δ 2 ] , (132)
a 0 = L 2 S 1 2 S 2 2 − Σ 1 2 S 1 2 Δ 2 2 − Σ 2 2 S 2 2 Δ 1 2 . (133) a_0=L^2S_1^2S_2^2-\Sigma_1^2S_1^2\Delta_2^2-\Sigma_2^2S_2^2\Delta_1^2. \tag*{(133)} a 0 = L 2 S 1 2 S 2 2 − Σ 1 2 S 1 2 Δ 2 2 − Σ 2 2 S 2 2 Δ 1 2 . (133)
For m 1 > m 2 m_1>m_2 m 1 > m 2 , a 3 > 0 a_3>0 a 3 > 0 , and P P P has three real roots f 1 < f 2 < f 3 f_1<f_2<f_3 f 1 < f 2 < f 3 . Since P ( f ) ≥ 0 P(f)\geq0 P ( f ) ≥ 0 along the flow, f f f oscillates between the turning points f 1 f_1 f 1 and f 2 f_2 f 2 , where the triple product vanishes and the triad is coplanar.
We measure λ Λ \lambda_\Lambda λ Λ from the fiducial configuration of § 3 B, a turning point with f = f 1 f=f_1 f = f 1 . The solution of Eq. (129) then reads
f ( λ Λ ) = f 1 + ( f 2 − f 1 ) sn 2 ( Ω Λ λ Λ , k ) , (134) f(\lambda_\Lambda)=f_1+(f_2-f_1)\operatorname{sn}^2(\Omega_\Lambda\lambda_\Lambda,k), \tag*{(134)} f ( λ Λ ) = f 1 + ( f 2 − f 1 ) sn 2 ( Ω Λ λ Λ , k ) , (134)
with
Ω Λ ≡ 1 2 a 3 ( f 3 − f 1 ) , k 2 ≡ f 2 − f 1 f 3 − f 1 . (135) \Omega_\Lambda\equiv\frac{1}{2}\sqrt{a_3(f_3-f_1)},\qquad k^2\equiv\frac{f_2-f_1}{f_3-f_1}. \tag*{(135)} Ω Λ ≡ 2 1 a 3 ( f 3 − f 1 ) , k 2 ≡ f 3 − f 1 f 2 − f 1 . (135)
In Eq. (134) , sn \operatorname{sn} sn is the Jacobi elliptic sine and am \operatorname{am} am the Jacobi amplitude,
sn ( u , k ) = sin am ( u , k ) , am ( F ( φ , k ) , k ) = φ , (136) \operatorname{sn}(u,k)=\sin\operatorname{am}(u,k),\qquad\operatorname{am}(F(\varphi,k),k)=\varphi, \tag*{(136)} sn ( u , k ) = sin am ( u , k ) , am ( F ( φ , k ) , k ) = φ , (136)
where am \operatorname{am} am inverts, with respect to φ \varphi φ , the incomplete elliptic integral of the first kind
F ( φ , k ) = ∫ 0 φ d ϑ 1 − k 2 sin 2 ϑ , K ( k ) = F ( π / 2 , k ) , (137) F(\varphi,k)=\int_0^\varphi\frac{d\vartheta}{\sqrt{1-k^2\sin^2\vartheta}},\qquad K(k)=F(\pi/2,k), \tag*{(137)} F ( φ , k ) = ∫ 0 φ 1 − k 2 sin 2 ϑ d ϑ , K ( k ) = F ( π /2 , k ) , (137)
and K K K is the complete elliptic integral of the first kind. Since sn 2 ( u , k ) \operatorname{sn}^2(u,k) sn 2 ( u , k ) has period 2 K ( k ) 2K(k) 2 K ( k ) , the mutual angles return to their initial values after
Δ λ Λ = 2 K ( k ) Ω Λ . (138) \Delta\lambda_\Lambda=\frac{2K(k)}{\Omega_\Lambda}. \tag*{(138)} Δ λ Λ = Ω Λ 2 K ( k ) . (138)
By Eq. (127) , f f f increases during the first half of this period and decreases during the second, according to the sign s s s of L ⋅ ( S 1 × S 2 ) \mathbf{L}\cdot(\mathbf{S}_1\times\mathbf{S}_2) L ⋅ ( S 1 × S 2 ) . Inverting Eq. (134) , a configuration with given f f f and s s s is reached after
λ Λ = U ( f , s ) Ω Λ , U ( f , s ) ≡ { F ( φ ( f ) , k ) , s ≥ 0 , 2 K ( k ) − F ( φ ( f ) , k ) , s < 0 , (139) \lambda_{\Lambda}=\frac{U(f,s)}{\Omega_{\Lambda}},\qquad U(f,s)\equiv\begin{cases}F(\varphi(f),k),&s\ge0,\\2K(k)-F(\varphi(f),k),&s<0,\end{cases} \tag*{(139)} λ Λ = Ω Λ U ( f , s ) , U ( f , s ) ≡ { F ( φ ( f ) , k ) , 2 K ( k ) − F ( φ ( f ) , k ) , s ≥ 0 , s < 0 , (139)
with φ ( f ) ≡ arcsin ( f − f 1 ) / ( f 2 − f 1 ) \varphi(f)\equiv\arcsin\sqrt{(f-f_1)/(f_2-f_1)} φ ( f ) ≡ arcsin ( f − f 1 ) / ( f 2 − f 1 ) .
Orientation of L and Delta lambda J2
After a Λ \Lambda Λ -flow by Δ λ Λ \Delta\lambda_{\Lambda} Δ λ Λ , the triad recovers its shape but has rotated about J ^ \hat{\mathbf{J}} J ^ . To obtain this rotation, we follow the azimuth Φ L \Phi_L Φ L of L \mathbf{L} L about J ^ \hat{\mathbf{J}} J ^ , measured as in § 3 B. From the fiducial configuration, where f = f 1 f=f_1 f = f 1 and Φ L = 0 \Phi_L=0 Φ L = 0 , a Λ \Lambda Λ -flow by λ Λ \lambda_{\Lambda} λ Λ brings it to [18 ]
Φ L Λ ( λ Λ ) = 1 Ω Λ ∑ i = 1 2 B i L D i L Π ( α i L 2 , am ( Ω Λ λ Λ , k ) , k ) (140) \Phi_L^{\Lambda}(\lambda_{\Lambda})=\frac{1}{\Omega_{\Lambda}}\sum_{i=1}^{2}\frac{B_{iL}}{D_{iL}}\Pi\left(\alpha_{iL}^{2},\operatorname{am}(\Omega_{\Lambda}\lambda_{\Lambda},k),k\right) \tag*{(140)} Φ L Λ ( λ Λ ) = Ω Λ 1 i = 1 ∑ 2 D i L B i L Π ( α i L 2 , am ( Ω Λ λ Λ , k ) , k ) (140)
with
B 1 , 2 L = 1 2 { J [ Λ + L 2 ( Σ 1 + Σ 2 ) ] ± L [ Λ + L 2 ( Σ 1 + Σ 2 ) + Σ 1 S 1 2 + Σ 2 S 2 2 + ( Σ 1 + Σ 2 ) ( Δ 2 Σ 1 − Δ 1 Σ 2 ) ] } (141) B_{1,2\,L}=\frac{1}{2}\left\{J\left[\Lambda+L^{2}(\Sigma_1+\Sigma_2)\right]\pm L\left[\Lambda+L^{2}(\Sigma_1+\Sigma_2)+\Sigma_1S_1^{2}+\Sigma_2S_2^{2}+(\Sigma_1+\Sigma_2)(\Delta_2\Sigma_1-\Delta_1\Sigma_2)\right]\right\} \tag*{(141)} B 1 , 2 L = 2 1 { J [ Λ + L 2 ( Σ 1 + Σ 2 ) ] ± L [ Λ + L 2 ( Σ 1 + Σ 2 ) + Σ 1 S 1 2 + Σ 2 S 2 2 + ( Σ 1 + Σ 2 ) ( Δ 2 Σ 1 − Δ 1 Σ 2 ) ] } (141)
D 1 , 2 L = L ( L ± J ) + Δ 2 Σ 1 − Δ 1 Σ 2 − ( Σ 1 − Σ 2 ) f 1 , α i L 2 = ( Σ 1 − Σ 2 ) ( f 2 − f 1 ) D i L , (142) D_{1,2\,L}=L(L\pm J)+\Delta_2\Sigma_1-\Delta_1\Sigma_2-(\Sigma_1-\Sigma_2)f_1,\qquad\alpha_{iL}^{2}=\frac{(\Sigma_1-\Sigma_2)(f_2-f_1)}{D_{iL}}, \tag*{(142)} D 1 , 2 L = L ( L ± J ) + Δ 2 Σ 1 − Δ 1 Σ 2 − ( Σ 1 − Σ 2 ) f 1 , α i L 2 = D i L ( Σ 1 − Σ 2 ) ( f 2 − f 1 ) , (142)
where the upper (lower) signs refer to i = 1 i=1 i = 1 (i = 2 i=2 i = 2 ), and where
Π ( n , φ , k ) = ∫ 0 φ d ϑ ( 1 − n sin 2 ϑ ) 1 − k 2 sin 2 ϑ , Π ( n , k ) = Π ( n , π / 2 , k ) (143) \Pi(n,\varphi,k)=\int_{0}^{\varphi}\frac{\mathrm{d}\vartheta}{(1-n\sin^{2}\vartheta)\sqrt{1-k^{2}\sin^{2}\vartheta}},\qquad\Pi(n,k)=\Pi(n,\pi/2,k) \tag*{(143)} Π ( n , φ , k ) = ∫ 0 φ ( 1 − n sin 2 ϑ ) 1 − k 2 sin 2 ϑ d ϑ , Π ( n , k ) = Π ( n , π /2 , k ) (143)
are the incomplete and complete elliptic integrals of the third kind.
Since J \mathbf{J} J is fixed, f f f , s s s , and Φ L \Phi_L Φ L determine the triad entirely. Indeed, the scalar products [Eq. (C3)], and hence J ⋅ L = L 2 + L ⋅ S 1 + L ⋅ S 2 \mathbf{J}\cdot\mathbf{L}=L^{2}+\mathbf{L}\cdot\mathbf{S}_1+\mathbf{L}\cdot\mathbf{S}_2 J ⋅ L = L 2 + L ⋅ S 1 + L ⋅ S 2 , are affine in f f f , and s s s fixes the handedness.
After one period Δ λ Λ \Delta\lambda_{\Lambda} Δ λ Λ , Ω Λ λ Λ \Omega_{\Lambda}\lambda_{\Lambda} Ω Λ λ Λ has increased by 2 K 2K 2 K and am \operatorname{am} am by π \pi π , so that L \mathbf{L} L has rotated about J ^ \hat{\mathbf{J}} J ^ by
Δ Φ L = Φ L Λ ( Δ λ Λ ) = 2 Ω Λ ∑ i = 1 2 B i L D i L Π ( α i L 2 , k ) . (144) \Delta\Phi_L=\Phi_L^{\Lambda}(\Delta\lambda_{\Lambda})=\frac{2}{\Omega_{\Lambda}}\sum_{i=1}^{2}\frac{B_{iL}}{D_{iL}}\Pi(\alpha_{iL}^{2},k). \tag*{(144)} Δ Φ L = Φ L Λ ( Δ λ Λ ) = Ω Λ 2 i = 1 ∑ 2 D i L B i L Π ( α i L 2 , k ) . (144)
Since the J 2 J^{2} J 2 -flow rotates all vectors rigidly about J ^ \hat{\mathbf{J}} J ^ by an angle 2 J λ J 2 2J\lambda_{J^{2}} 2 J λ J 2 , the flow amount that brings them back to their initial orientation is
Δ λ J 2 = − Δ Φ L 2 J . (145) \Delta\lambda_{J^{2}}=-\frac{\Delta\Phi_L}{2J}. \tag*{(145)} Δ λ J 2 = − 2 J Δ Φ L . (145)
Rotation of the fictitious pairs and Δ λ V 2 \Delta\lambda_{V^2} Δ λ V 2
After the Λ \Lambda Λ - and J 2 J^2 J 2 -segments, the physical vectors are back to their initial values, but each fictitious pair ( r V , p V ) (\mathbf{r}_V,\mathbf{p}_V) ( r V , p V ) has rotated about its physical vector V \mathbf{V} V by an angle Δ ψ V \Delta\psi_V Δ ψ V , with ψ V \psi_V ψ V defined as in Eq. (111) . Since the V 2 V^2 V 2 -flow leaves the physical vectors fixed and rotates the pair about V \mathbf{V} V by an angle 2 V λ V 2 2V\lambda_{V^2} 2 V λ V 2 [Eq. (48) ], the flow amounts that undo these rotations are
Δ λ L 2 = − Δ ψ L 2 L , Δ λ S 1 2 = − Δ ψ S 1 2 S 1 , Δ λ S 2 2 = − Δ ψ S 2 2 S 2 , (146) \Delta\lambda_{L^2}=-\frac{\Delta\psi_L}{2L},\qquad\Delta\lambda_{S_1^2}=-\frac{\Delta\psi_{S_1}}{2S_1},\qquad\Delta\lambda_{S_2^2}=-\frac{\Delta\psi_{S_2}}{2S_2}, \tag*{(146)} Δ λ L 2 = − 2 L Δ ψ L , Δ λ S 1 2 = − 2 S 1 Δ ψ S 1 , Δ λ S 2 2 = − 2 S 2 Δ ψ S 2 , (146)
with [18 ]
Δ ψ L = 2 Ω Λ [ B 1 L D 1 L Π ( α 1 L 2 , k ) − B 2 L D 2 L Π ( α 2 L 2 , k ) ] − L ( Σ 1 + Σ 2 ) Δ λ Λ , (147) \Delta\psi_L=\frac{2}{\Omega_\Lambda}\left[\frac{B_{1L}}{D_{1L}}\Pi(\alpha_{1L}^2,k)-\frac{B_{2L}}{D_{2L}}\Pi(\alpha_{2L}^2,k)\right]-L(\Sigma_1+\Sigma_2)\Delta\lambda_\Lambda, \tag*{(147)} Δ ψ L = Ω Λ 2 [ D 1 L B 1 L Π ( α 1 L 2 , k ) − D 2 L B 2 L Π ( α 2 L 2 , k ) ] − L ( Σ 1 + Σ 2 ) Δ λ Λ , (147)
Δ ψ S a = 2 Ω Λ [ B 1 S a D 1 S a Π ( α 1 S a 2 , k ) − B 2 S a D 2 S a Π ( α 2 S a 2 , k ) ] − S a ( Σ a − Σ b ) Δ λ Λ , b ≠ a , (148) \Delta\psi_{S_a}=\frac{2}{\Omega_\Lambda}\left[\frac{B_{1S_a}}{D_{1S_a}}\Pi(\alpha_{1S_a}^2,k)-\frac{B_{2S_a}}{D_{2S_a}}\Pi(\alpha_{2S_a}^2,k)\right]-S_a(\Sigma_a-\Sigma_b)\Delta\lambda_\Lambda,\qquad b\ne a, \tag*{(148)} Δ ψ S a = Ω Λ 2 [ D 1 S a B 1 S a Π ( α 1 S a 2 , k ) − D 2 S a B 2 S a Π ( α 2 S a 2 , k ) ] − S a ( Σ a − Σ b ) Δ λ Λ , b = a , (148)
where B i L B_{iL} B i L , D i L D_{iL} D i L , and α i L 2 \alpha_{iL}^2 α i L 2 are given by Eq. (C16), and
B 1 , 2 S 1 = 1 2 { J [ S 1 2 ( Σ 1 − 2 Σ 2 ) − Δ 1 Σ 2 ( Σ 1 − Σ 2 ) ] ± S 1 [ L 2 Σ 1 − J 2 Σ 2 + S 1 2 ( Σ 1 − Σ 2 ) + Δ 2 Σ 1 2 + Δ 1 Σ 2 2 − 2 Δ 1 Σ 1 Σ 2 ] } , (149) B_{1,2\,S_1}=\frac{1}{2}\left\{J\left[S_1^2(\Sigma_1-2\Sigma_2)-\Delta_1\Sigma_2(\Sigma_1-\Sigma_2)\right]\pm S_1\left[L^2\Sigma_1-J^2\Sigma_2+S_1^2(\Sigma_1-\Sigma_2)+\Delta_2\Sigma_1^2+\Delta_1\Sigma_2^2-2\Delta_1\Sigma_1\Sigma_2\right]\right\}, \tag*{(149)} B 1 , 2 S 1 = 2 1 { J [ S 1 2 ( Σ 1 − 2 Σ 2 ) − Δ 1 Σ 2 ( Σ 1 − Σ 2 ) ] ± S 1 [ L 2 Σ 1 − J 2 Σ 2 + S 1 2 ( Σ 1 − Σ 2 ) + Δ 2 Σ 1 2 + Δ 1 Σ 2 2 − 2 Δ 1 Σ 1 Σ 2 ] } , (149)
D 1 , 2 S 1 = S 1 ( S 1 ± J ) − Δ 1 Σ 2 + Σ 1 f 1 , (150) D_{1,2\,S_1}=S_1(S_1\pm J)-\Delta_1\Sigma_2+\Sigma_1f_1, \tag*{(150)} D 1 , 2 S 1 = S 1 ( S 1 ± J ) − Δ 1 Σ 2 + Σ 1 f 1 , (150)
α i S 1 2 = − Σ 1 ( f 2 − f 1 ) D i S 1 , (151) \alpha_{iS_1}^2=-\frac{\Sigma_1(f_2-f_1)}{D_{iS_1}}, \tag*{(151)} α i S 1 2 = − D i S 1 Σ 1 ( f 2 − f 1 ) , (151)
with the same sign convention as in Eq. (C16). The coefficients B i S 2 B_{iS_2} B i S 2 , D i S 2 D_{iS_2} D i S 2 , and α i S 2 2 \alpha_{iS_2}^2 α i S 2 2 follow from B i S 1 B_{iS_1} B i S 1 , D i S 1 D_{iS_1} D i S 1 , and α i S 1 2 \alpha_{iS_1}^2 α i S 1 2 by exchanging the two bodies, S 1 ↔ S 2 S_1\leftrightarrow S_2 S 1 ↔ S 2 and Σ 1 ↔ Σ 2 \Sigma_1\leftrightarrow\Sigma_2 Σ 1 ↔ Σ 2 , at fixed i i i . Since Σ 1 − Σ 2 \Sigma_1-\Sigma_2 Σ 1 − Σ 2 then changes sign, so do f f f and Δ a \Delta_a Δ a [Eqs. (122) and (126) ], and the exchange must be supplemented by ( f 1 , f 2 ) → ( − f 1 , − f 2 ) (f_1,f_2)\to(-f_1,-f_2) ( f 1 , f 2 ) → ( − f 1 , − f 2 ) and ( Δ 1 , Δ 2 ) → ( − Δ 2 , − Δ 1 ) (\Delta_1,\Delta_2)\to(-\Delta_2,-\Delta_1) ( Δ 1 , Δ 2 ) → ( − Δ 2 , − Δ 1 ) .
The angles Δ ψ V \Delta\psi_V Δ ψ V are defined modulo 2 π 2\pi 2 π . A shift by 2 π 2\pi 2 π changes J 3 J_3 J 3 by ± V \pm V ± V , as discussed in Sec. IV E 2, which does not affect the dynamics since L L L , S 1 S_1 S 1 , and S 2 S_2 S 2 are Casimirs.
Flow generated by a function of commuting constants of motion
Consider n n n COMs C 1 , … , C n C_1,\ldots,C_n C 1 , … , C n in involution, i.e., such that { C i , C j } = 0 \{C_i,C_j\}=0 { C i , C j } = 0 for all i , j ∈ { 1 , … , n } i,j\in\{1,\ldots,n\} i , j ∈ { 1 , … , n } . Under the flow of C i C_i C i , a phase-space function g g g evolves as d g / d λ C i = L i g dg/d\lambda_{C_i}=\mathcal{L}_i g d g / d λ C i = L i g , with L i ≡ { ⋅ , C i } \mathcal{L}_i\equiv\{\cdot,C_i\} L i ≡ { ⋅ , C i } . The Jacobi identity gives
L i L j g − L j L i g = { g , { C j , C i } } = 0 , (152) \mathcal{L}_i\mathcal{L}_j g-\mathcal{L}_j\mathcal{L}_i g=\{g,\{C_j,C_i\}\}=0, \tag*{(152)} L i L j g − L j L i g = { g , { C j , C i }} = 0 , (152)
so the operators L i \mathcal{L}_i L i commute. For constant coefficients c i c_i c i , the flow of C = ∑ i c i C i C=\sum_i c_i C_i C = ∑ i c i C i by λ C \lambda_C λ C is then
g ( λ C ) = exp ( λ C ∑ i c i L i ) g ( 0 ) = ∏ i exp ( c i λ C L i ) g ( 0 ) , (153) g(\lambda_C)=\exp\left(\lambda_C\sum_i c_i\mathcal{L}_i\right)g(0)=\prod_i\exp(c_i\lambda_C\mathcal{L}_i)g(0), \tag*{(153)} g ( λ C ) = exp ( λ C i ∑ c i L i ) g ( 0 ) = i ∏ exp ( c i λ C L i ) g ( 0 ) , (153)
where g ( 0 ) g(0) g ( 0 ) is the initial value of g g g and each exponential is a power series in its operator. The second equality holds because the operators L i \mathcal{L}_i L i commute, so that the exponential of their sum factorizes into a product of exponentials. The factor exp ( c i λ C L i ) \exp(c_i\lambda_C\mathcal{L}_i) exp ( c i λ C L i ) is the flow of C i C_i C i by c i λ C c_i\lambda_C c i λ C . Flowing under C C C by λ C \lambda_C λ C is thus the same as flowing under each C i C_i C i by c i λ C c_i\lambda_C c i λ C , in any order. See Sec. 5.3.4 of Ref. [29 ] or Sec. 9.6 of Ref. [39 ] for more details.
This extends to any function G ( C 1 , … , C n ) G(C_1,\ldots,C_n) G ( C 1 , … , C n ) . Indeed, { g , G } = ∑ i ( ∂ G / ∂ C i ) { g , C i } \{g,G\}=\sum_i(\partial G/\partial C_i)\{g,C_i\} { g , G } = ∑ i ( ∂ G / ∂ C i ) { g , C i } , and the coefficients ∂ G / ∂ C i \partial G/\partial C_i ∂ G / ∂ C i do not change along the flow. Flowing under G G G by λ G \lambda_G λ G is therefore the same as flowing under each C i C_i C i by ( ∂ G / ∂ C i ) λ G (\partial G/\partial C_i)\lambda_G ( ∂ G / ∂ C i ) λ G . We use this with G = H ˉ G=\bar{H} G = H ˉ in § 3 A and with G = J 1 , J 2 , J 3 G=J_1,J_2,J_3 G = J 1 , J 2 , J 3 in § 3 B.
Justification of the criterion for the breakdown of integrability conditions
We first briefly review the Liouville–Arnold theorem, and then justify the criterion.
Background on the Liouville-Arnold theorem
Consider a Hamiltonian system with n n n degrees of freedom, on a 2 n 2n 2 n -dimensional symplectic manifold S \mathcal{S} S with symplectic form ∑ i = 1 n d p i ∧ d q i \sum_{i=1}^{n}dp_i\wedge dq_i ∑ i = 1 n d p i ∧ d q i and canonical coordinates ( p i , q i ) (p_i,q_i) ( p i , q i ) . Now assume the following integrability conditions:
There exist n n n functions C = ( C 1 , … , C n ) \mathbf{C}=(C_1,\ldots,C_n) C = ( C 1 , … , C n ) in mutual involution, { C i , C j } = 0 \{C_i,C_j\}=0 { C i , C j } = 0 , whose level set T \mathcal{T} T , defined by C = c o n s t \mathbf{C}=\mathrm{const} C = const , is compact and connected.
The C i C_i C i ’s are independent on T \mathcal{T} T . This is equivalent to the linear independence of their differentials, d C 1 ∧ d C 2 ∧ ⋯ ∧ d C n ≠ 0 dC_1\wedge dC_2\wedge\cdots\wedge dC_n\ne0 d C 1 ∧ d C 2 ∧ ⋯ ∧ d C n = 0 .
Then as per the Liouville-Arnold theorem [30 , 32 ] , T \mathcal{T} T is diffeomorphic to a torus, and in the neighborhood of T \mathcal{T} T , there exist action-angle coordinates ( J i , θ i ) (J_i,\theta_i) ( J i , θ i ) , i.e.,
∑ i d p i ∧ d q i = ∑ i d J i ∧ d θ i . (154) \sum_i dp_i\wedge dq_i=\sum_i dJ_i\wedge d\theta_i. \tag*{(154)} i ∑ d p i ∧ d q i = i ∑ d J i ∧ d θ i . (154)
For an integrable Hamiltonian system, the Hamiltonian is one of the C i C_i C i ’s, or a function of them. The C i C_i C i ’s then commute with it and are therefore COMs.
a. Preferred sets of COMs. As discussed in Sec. VI B, we use COMs whose differentials remain finite at finite values of ( p i , q i ) (p_i,q_i) ( p i , q i ) , such as ( J 2 , J z , Λ ) (J^2,J_z,\Lambda) ( J 2 , J z , Λ ) for the OA spinning BBH system, and denote them by C \mathbf{C} C .
Justification of the criterion
We proceed in two steps. We first show that the condition of Eq. (78) does not depend on the choice of action-angle variables, and then that it implies the breakdown of at least one integrability condition.
Step 1. The condition of Eq. (78) cannot be removed by a change of action-angle variables. Starting from a point on the level set C = C ∗ \mathbf{C} = \mathbf{C}_{*} C = C ∗ satisfying Eq. (78) , we define a curve γ : [ 0 , 1 ] → S \gamma: [0,1] \to\mathcal{S} γ : [ 0 , 1 ] → S with γ ( 0 ) \gamma(0) γ ( 0 ) on this level set and γ ( p ) \gamma(p) γ ( p ) , p ∈ ( 0 , 1 ] p \in(0,1] p ∈ ( 0 , 1 ] off the level set, such that for points γ ( p ) \gamma(p) γ ( p ) with p ∈ ( 0 , 1 ] p \in(0,1] p ∈ ( 0 , 1 ] the integrability conditions hold. Implicit is the assumption that such a curve can indeed be defined.
The level set of C \mathbf{C} C through γ ( p ) \gamma(p) γ ( p ) for each p ∈ ( 0 , 1 ] p \in(0,1] p ∈ ( 0 , 1 ] is then a torus T p \mathcal{T}_{p} T p with action-angle variables, and det ( ∂ C / ∂ J ) \det(\partial\mathbf{C}/\partial\mathbf{J}) det ( ∂ C / ∂ J ) varies smoothly with p p p . We also denote by T ∗ \mathcal{T}_{*} T ∗ the level set C = C ∗ \mathbf{C} = \mathbf{C}_{*} C = C ∗ through γ ( 0 ) \gamma(0) γ ( 0 ) . Hence
lim p → 0 C ( γ ( p ) ) = C ∗ , (155) \lim_{p\to0}\mathbf{C}(\gamma(p))=\mathbf{C}_{*}, \tag*{(155)} p → 0 lim C ( γ ( p )) = C ∗ , (155)
lim p → 0 T p = T ∗ . (156) \lim_{p\to0}\mathcal{T}_{p}=\mathcal{T}_{*}. \tag*{(156)} p → 0 lim T p = T ∗ . (156)
Action-angle variables are not unique, so one may ask whether the vanishing determinant in Eq. (78) is merely an artifact of a poor choice of actions. Suppose that there exist other action-angle variables ( J ~ , θ ~ ) (\widetilde{\mathbf{J}},\widetilde{\boldsymbol{\theta}}) ( J , θ ) on the tori T p \mathcal{T}_{p} T p for which det ( ∂ C / ∂ J ~ ) \det(\partial\mathbf{C}/\partial\widetilde{\mathbf{J}}) det ( ∂ C / ∂ J ) remains finite and non-zero as p → 0 p \to0 p → 0 . Then, by the chain rule,
det ( ∂ C ∂ J ) = det ( ∂ C ∂ J ~ ) det ( ∂ J ~ ∂ J ) , (157) \det\left(\frac{\partial\mathbf{C}}{\partial\mathbf{J}}\right)
=
\det\left(\frac{\partial\mathbf{C}}{\partial\widetilde{\mathbf{J}}}\right)
\det\left(\frac{\partial\widetilde{\mathbf{J}}}{\partial\mathbf{J}}\right), \tag*{(157)} det ( ∂ J ∂ C ) = det ( ∂ J ∂ C ) det ( ∂ J ∂ J ) , (157)
so that Eq. (78) would require
lim p → 0 det ( ∂ J ~ ∂ J ) = 0. (158) \lim_{p\to0}\det\left(\frac{\partial\widetilde{\mathbf{J}}}{\partial\mathbf{J}}\right)=0. \tag*{(158)} p → 0 lim det ( ∂ J ∂ J ) = 0. (158)
We now show that this is impossible by contradiction. For p > 0 p > 0 p > 0 , both sets of actions are invertible functions of C \mathbf{C} C only. Hence J ~ = J ~ ( J ) \widetilde{\mathbf{J}}=\widetilde{\mathbf{J}}(\mathbf{J}) J = J ( J ) , and the Jacobian of the transformation ( J , θ ) → ( J ~ , θ ~ ) (\mathbf{J},\boldsymbol{\theta}) \to(\widetilde{\mathbf{J}},\widetilde{\boldsymbol{\theta}}) ( J , θ ) → ( J , θ ) is lower block-triangular,
M = ( ∂ J ~ ∂ J 0 ∂ θ ~ ∂ J ∂ θ ~ ∂ θ ) . (159) \mathcal{M}=
\begin{pmatrix}
\frac{\partial\widetilde{\mathbf{J}}}{\partial\mathbf{J}} & 0\\[6pt]
\frac{\partial\widetilde{\boldsymbol{\theta}}}{\partial\mathbf{J}} & \frac{\partial\widetilde{\boldsymbol{\theta}}}{\partial\boldsymbol{\theta}}
\end{pmatrix}. \tag*{(159)} M = ∂ J ∂ J ∂ J ∂ θ 0 ∂ θ ∂ θ . (159)
Since the transformation is canonical, its Jacobian has unit determinant, and the block-triangular structure gives
1 = det M = det ( ∂ J ~ ∂ J ) det ( ∂ θ ~ ∂ θ ) , (160) 1=\det\mathcal{M}
=\det\left(\frac{\partial\widetilde{\mathbf{J}}}{\partial\mathbf{J}}\right)
\det\left(\frac{\partial\widetilde{\boldsymbol{\theta}}}{\partial\boldsymbol{\theta}}\right), \tag*{(160)} 1 = det M = det ( ∂ J ∂ J ) det ( ∂ θ ∂ θ ) , (160)
⇒ det ( ∂ θ ~ ∂ θ ) = [ det ( ∂ J ~ ∂ J ) ] − 1 . (161) \Rightarrow\quad\det\left(\frac{\partial\tilde{\theta}}{\partial\theta}\right)=\left[\det\left(\frac{\partial\tilde{\mathcal{J}}}{\partial\mathcal{J}}\right)\right]^{-1}. \tag*{(161)} ⇒ det ( ∂ θ ∂ θ ~ ) = [ det ( ∂ J ∂ J ~ ) ] − 1 . (161)
By Eq. (158) , the angular Jacobian would then diverge as p → 0 p \to0 p → 0 .
The key point is that this divergence is in contradiction with our assumption that both θ \theta θ and θ ~ \tilde{\theta} θ ~ are angle coordinates on the same torus T p \mathcal{T}_p T p , each covering it once with range [ 0 , 2 π ) n [0,2\pi)^n [ 0 , 2 π ) n . The total angular volume of the torus, computed in either set of angles, must therefore be ( 2 π ) n (2\pi)^n ( 2 π ) n . Changing variables from θ ~ \tilde{\theta} θ ~ to θ \theta θ , and noting from Eq. (E7) that det ( ∂ θ ~ / ∂ θ ) \det(\partial\tilde{\theta}/\partial\theta) det ( ∂ θ ~ / ∂ θ ) depends on the actions only, and is hence constant on T p \mathcal{T}_p T p , we obtain
( 2 π ) n = ∮ T p d n θ ~ = ∮ T p ∣ det ( ∂ θ ~ ∂ θ ) ∣ d n θ = ∣ det ( ∂ θ ~ ∂ θ ) ∣ ∮ T p d n θ = ( 2 π ) n ∣ det ( ∂ θ ~ ∂ θ ) ∣ , (162) (2\pi)^n=\oint_{\mathcal{T}_p}d^n\tilde{\theta}=\oint_{\mathcal{T}_p}\left|\det\left(\frac{\partial\tilde{\theta}}{\partial\theta}\right)\right|d^n\theta=\left|\det\left(\frac{\partial\tilde{\theta}}{\partial\theta}\right)\right|\oint_{\mathcal{T}_p}d^n\theta=(2\pi)^n\left|\det\left(\frac{\partial\tilde{\theta}}{\partial\theta}\right)\right|, \tag*{(162)} ( 2 π ) n = ∮ T p d n θ ~ = ∮ T p det ( ∂ θ ∂ θ ~ ) d n θ = det ( ∂ θ ∂ θ ~ ) ∮ T p d n θ = ( 2 π ) n det ( ∂ θ ∂ θ ~ ) , (162)
for every p > 0 p>0 p > 0 . Hence ∣ det ( ∂ θ ~ / ∂ θ ) ∣ = 1 \left|\det(\partial\tilde{\theta}/\partial\theta)\right|=1 det ( ∂ θ ~ / ∂ θ ) = 1 on every torus T p \mathcal{T}_p T p , and the angular Jacobian cannot diverge as p → 0 p\to0 p → 0 . This contradicts Eqs. (158) and (E7). So our supposition that the vanishing of det ( ∂ C / ∂ J ) \det(\partial\mathcal{C}/\partial\mathcal{J}) det ( ∂ C / ∂ J ) can be removed by a change of action-angle variables must be false.
As a by-product, Eqs. (E7) and (162) show that any two sets of action-angle variables on the same torus satisfy
∣ det ( ∂ J ~ ∂ J ) ∣ = 1. (163) \left|\det\left(\frac{\partial\tilde{\mathcal{J}}}{\partial\mathcal{J}}\right)\right|=1. \tag*{(163)} det ( ∂ J ∂ J ~ ) = 1. (163)
Step 2. Breakdown of the integrability conditions. Let ( J , θ ) (\mathcal{J},\theta) ( J , θ ) be any set of action-angle variables on the tori T p \mathcal{T}_p T p . For all p > 0 p>0 p > 0 ,
d C 1 ∧ ⋯ ∧ d C n = det ( ∂ C ∂ J ) d J 1 ∧ ⋯ ∧ d J n , (164) d\mathcal{C}_1\wedge\cdots\wedge d\mathcal{C}_n=\det\left(\frac{\partial\mathcal{C}}{\partial\mathcal{J}}\right)d\mathcal{J}_1\wedge\cdots\wedge d\mathcal{J}_n, \tag*{(164)} d C 1 ∧ ⋯ ∧ d C n = det ( ∂ J ∂ C ) d J 1 ∧ ⋯ ∧ d J n , (164)
where both wedge products are expanded in the original canonical coordinates ( q i , p i ) (q_i,p_i) ( q i , p i ) . By Eq. (78) and Step 1, the determinant tends to zero as p → 0 p\to0 p → 0 for every such choice, so at least one of the following holds.
(i) d J 1 ∧ ⋯ ∧ d J n d\mathcal{J}_1\wedge\cdots\wedge d\mathcal{J}_n d J 1 ∧ ⋯ ∧ d J n does not diverge as p → 0 p\to0 p → 0 . Then Eq. (164) forces d C 1 ∧ ⋯ ∧ d C n → 0 d\mathcal{C}_1\wedge\cdots\wedge d\mathcal{C}_n\to0 d C 1 ∧ ⋯ ∧ d C n → 0 . Since the COMs are smooth, d C 1 ∧ ⋯ ∧ d C n = 0 d\mathcal{C}_1\wedge\cdots\wedge d\mathcal{C}_n=0 d C 1 ∧ ⋯ ∧ d C n = 0 somewhere on T ∗ \mathcal{T}^* T ∗ , so the second integrability condition is violated.
(ii) d J 1 ∧ ⋯ ∧ d J n d\mathcal{J}_1\wedge\cdots\wedge d\mathcal{J}_n d J 1 ∧ ⋯ ∧ d J n diverges as p → 0 p\to0 p → 0 . Then at least one d J i d\mathcal{J}_i d J i diverges, so Eq. (154) cannot hold on T ∗ \mathcal{T}^* T ∗ . Since, by the Liouville–Arnold theorem, Eq. (154) is an implication of the integrability conditions, at least one of these conditions must fail on T ∗ \mathcal{T}^* T ∗ .