Introduction
Low-mass X-ray binaries (LMXBs) that contain a weakly magnetized neutron star are divided into atoll and Z sources according to the tracks they trace in X-ray colour–colour diagrams and the variability that goes with each part of the track ((Hasinger & van der Klis 1989)). Their power spectra show quasi-periodic oscillations (QPOs) over a wide range of frequencies. In 1996 the Rossi X-ray Timing Explorer (RXTE) found QPOs near 1 kHz in the Z source Sco X-1 ([59 ] ) and in the atoll sources 4U 1728−34 ((Strohmayer et al. 1996)) and 4U 1608−52 ((Berger et al. 1996)). These kHz QPOs usually appear in pairs, the lower and upper kHz QPO, with frequencies between a few hundred hertz and about 1.3 kHz, and both move together as the source moves along its track ([57 , 58 ] ). During thermonuclear bursts several of the same sources show nearly coherent burst oscillations, which give the spin frequency of the star, for example 363 Hz in 4U 1728−34 ((Strohmayer et al. 1996)) and 581 Hz in 4U 1636−53 ((Strohmayer & Markwardt 2002)). At lower frequencies, Z sources show the horizontal-branch oscillation (HBO) at 15–60 Hz, and atoll sources show QPOs and noise components with similar properties ([58 ] ). The frequencies of the low-frequency features are correlated with those of the kHz QPOs over a wide range, and similar correlations are found in black-hole binaries ((Psaltis et al. 1999)). Since RXTE, kHz QPOs have also been observed with the Large Area X-ray Proportional Counter (LAXPC) on AstroSat ((Chauhan et al. 2017; Anand et al. 2026)). For 4U 1728−34, Anand et al. (2024) detected 13 simultaneous triplets of a low-frequency QPO and twin kHz QPOs in a single observation, with the frequency of the lower kHz QPO measured to about 1 Hz. The kHz QPO frequencies are close to orbital frequencies a few gravitational radii from a neutron star, and most models connect them with the motion of gas in the inner accretion flow. In beat-frequency models the upper kHz QPO is an orbital frequency near the inner edge of the disc and the lower one is its beat with the stellar spin; such models were first proposed for the HBO ((Alpar & Shaham 1985; Lamb et al. 1985)) and later developed for the kHz QPOs as the sonic-point model ((Miller et al. 1998)). The separation of the two kHz QPOs, however, changes with their frequencies and is not simply related to the spin ((van der Klis 2006). Resonance models relate the two peaks to a nonlinear coupling of epicyclic oscillations of the flow (Abramowicz & Kluźniak 2001), and other models use oscillation modes of thick tori (Matuszková et al. 2024). The relativistic precession model (RPM) of Stella & Vietri (1998, 1999) identifies the upper kHz QPO with the orbital frequency, the lower kHz QPO with the periastron precession frequency, and the low-frequency QPO with the nodal precession frequency of a slightly eccentric and tilted orbit, all at the same radius. To match the observed low frequencies, the last identification was later made with twice the nodal frequency (Stella et al. 1999). The RPM uses only test-particle frequencies and has few free parameters. It has been used to estimate neutron-star masses (Török et al. 2010, 2012; du Buisson et al. 2019; Anand et al. 2024) and, for black holes, masses and spins (Motta et al. 2014, 2022). For the low-frequency QPOs of black-hole binaries, Lense–Thirring precession of a hot inner flow is a widely discussed interpretation (Ingram et al. 2009; Ingram & Motta 2019). The model also has known difficulties. The RPM masses inferred for several sources are high, about 2 M⊙ _\odot ⊙ or more, and depend strongly on the assumed spin (Török et al. 2010, 2012). In a study of 13 sources with known spin, van Doesburgh & van der Klis [56 ] found that the low-frequency QPO frequency rises more steeply with the upper kHz QPO frequency than the RPM predicts, and that it shows no clear dependence on the spin.
Of the three RPM frequencies, the nodal precession frequency is the most sensitive to the spacetime of the star. It contains the Lense–Thirring term, proportional to the dimensionless spin j = c J / G M 2 j = cJ/GM^2 j = c J / G M 2 , and a retrograde term caused by the oblateness of the star, proportional to the quadrupole q = c 4 Q / G 2 M 3 q = c^4Q/G^2M^3 q = c 4 Q / G 2 M 3 . Morsink & Stella (1999) computed these frequencies for numerical models of rapidly rotating neutron stars and showed that near the star the two terms are comparable. For a Kerr black hole q = − j 2 q = -j^2 q = − j 2 . For a neutron star the reduced quadrupole Q ˉ = − q / j 2 \bar{Q} = -q/j^2 Q ˉ = − q / j 2 is several times larger, and its value depends on the equation of state (EOS) and the mass (Laarakkers & Poisson 1999; Pappas & Apostolatos 2012; Yagi & Yunes 2013). The orbital frequencies around neutron stars have therefore been computed in spacetimes with independent multipoles. Pappas (2012) showed how the multipole moments could be read from QPO frequencies in the RPM, Pachón et al. (2012) compared the precession frequencies of axisymmetric exterior solutions with those of numerical stars, and Maselli et al. (2020) fitted kHz QPOs of three atoll sources, including 4U 1728–34, with an accurate analytic metric of rotating neutron stars.
The Hartle–Thorne (HT) metric (Hartle 1967; Hartle & Thorne 1968) is the simplest such spacetime. It describes a rigidly rotating star to second order in its angular velocity, its exterior depends only on ( M , J , Q ) (M, J, Q) ( M , J , Q ) , and it agrees well with numerical models for the spin frequencies of most known pulsars (Berti et al. 2005; Urbancová et al. 2019). Boshkayev et al. (2014, 2015) derived the epicyclic frequencies of the HT metric and used the kHz QPOs of atoll and Z sources to infer M M M , J J J and Q Q Q as free parameters. More recently they applied the RPM in HT and other spacetimes to eight neutron-star LMXBs (Boshkayev et al. 2024, 2026). Urbancová et al. (2019) gave a systematic study of the orbital and epicyclic frequencies of the HT metric for realistic EOSs, using the universal relations between I I I , Q Q Q and the stellar parameters (Yagi & Yunes 2013). The same group has studied the oscillations and precession of fluid tori in this metric (Matuszková et al. 2024) and has shown that, because of the competition between the Lense–Thirring and the quadrupole terms, the nodal frequency at a given radius has a maximum as a function of j j j , so that slowly and rapidly rotating stars can show the same precession frequency (Török et al. 2025). On the observational side, Anand et al. (2024) used the RPM in the Kerr metric to obtain M = 1.92 ± 0.01 M ⊙ M = 1.92 \pm0.01\,M_\odot M = 1.92 ± 0.01 M ⊙ for 4U 1728–34 and a moment of inertia of about 2 × 10 45 g c m 2 2 \times10^{45}\,\mathrm{g\,cm^2} 2 × 1 0 45 g c m 2 , and estimated that a self-consistent neutron-star metric would change these values by about 1 per cent and 5 per cent.
In most of these studies the spacetime is Kerr, or j j j and q q q are fitted as independent parameters, or the low-frequency QPO is left out. Studies with EOS-consistent stars, such as Urbančová et al. (2019), have mainly addressed the twin kHz QPOs. Yet the EOS fixes j j j and q q q for a star of given mass and spin frequency, and for the atoll sources the spin frequency is known from burst oscillations. Independent EOS constraints now come from the masses of the heaviest pulsars (Fonseca et al. 2021) and from NICER pulse-profile modelling (Riley et al. 2021; Miller et al. 2021). In this paper we compute the spacetime and the orbital frequencies together from the EOS. For four EOSs we solve the HT equations, including the spin-induced change of mass and the deformed surface, and obtain ( j , q , R e q ) (j, q, R_{\rm eq}) ( j , q , R eq ) for any mass and spin. We compute the orbital and epicyclic frequencies from the full exterior metric without expanding in j j j or q q q . We then integrate eccentric, tilted geodesics around one of these stars and compute the power spectrum of a hot spot moving on them, to see how the three RPM frequencies appear in the flux. Finally, we use the RPM to infer the masses and spins of the atoll sources 4U 1636–53 and 4U 1728–34 and of the Z sources GX 5–1 and GX 340+0. We address three questions. (i) How does the quadrupole of the star change the precession frequencies compared with a black hole of the same spin? (ii) Do the precession frequencies of real orbits appear in the flux of matter moving on them? (iii) What masses and spins do the observed QPOs imply, and can the HBOs of the Z sources be nodal precession?
This is the first paper of a series. Here we treat the stars in the slow-rotation approximation, which gives the multipoles and the frequencies in closed form and makes their dependence on j j j and Q ˉ \bar{Q} Q ˉ transparent. Rapidly rotating stars, computed with numerical spacetimes, will be treated in a companion paper (Paper II), which will also test the accuracy of the approximation used here. § 2 summarises the HT stars and § 3 the orbital frequencies and the RPM. § 4 presents the stellar and orbital properties for the four EOSs, and § 5 the particle orbits and their power spectra. § 6 confronts the model with the observed QPOs, and § 7 discusses the results. The appendices give the HT equations and the derivation of the epicyclic frequencies. We use G = c = 1 G = c = 1 G = c = 1 except where frequencies are quoted in Hz.
Slowly rotating stars in the Hartle–Thorne approximation
Consider a star of central energy density ϵ c \epsilon_c ϵ c that rotates rigidly with angular velocity Ω = 2 π ν \Omega= 2\pi\nu Ω = 2 π ν . Hartle (1967) expanded the metric to second order in Ω \Omega Ω about the nonrotating Tolman–Oppenheimer–Volkoff (TOV) solution,
d s 2 = − e ν [ 1 + 2 ( h 0 + h 2 P 2 ) ] d t 2 + e λ [ 1 + 2 ( m 0 + m 2 P 2 ) r − 2 m ] d r 2 + r 2 [ 1 + 2 k 2 P 2 ] [ d θ 2 + sin 2 θ ( d ϕ − ω d t ) 2 ] , (1) \begin{aligned}
ds^2={}&-e^\nu\left[1+2(h_0+h_2P_2)\right]dt^2\\
&+e^\lambda\left[1+\frac{2(m_0+m_2P_2)}{r-2m}\right]dr^2\\
&+r^2\left[1+2k_2P_2\right]\left[d\theta^2+\sin^2\theta(d\phi-\omega\,dt)^2\right], \tag*{(1)}
\end{aligned} d s 2 = − e ν [ 1 + 2 ( h 0 + h 2 P 2 ) ] d t 2 + e λ [ 1 + r − 2 m 2 ( m 0 + m 2 P 2 ) ] d r 2 + r 2 [ 1 + 2 k 2 P 2 ] [ d θ 2 + sin 2 θ ( d ϕ − ω d t ) 2 ] , (1)
where P 2 = P 2 ( cos θ ) P_2=P_2(\cos\theta) P 2 = P 2 ( cos θ ) is the Legendre polynomial, e λ = ( 1 − 2 m / r ) − 1 e^\lambda=(1-2m/r)^{-1} e λ = ( 1 − 2 m / r ) − 1 , and ν ( r ) \nu(r) ν ( r ) , m ( r ) m(r) m ( r ) are the TOV potential and mass function. The function ω ( r ) \omega(r) ω ( r ) is of order Ω \Omega Ω , and h 0 h_0 h 0 , m 0 m_0 m 0 , h 2 h_2 h 2 , m 2 m_2 m 2 , k 2 k_2 k 2 are of order Ω 2 \Omega^2 Ω 2 . We first solve the TOV equations and then three perturbation problems in turn. The equations are collected in Appendix § 12 ; here we summarise what each one gives.
The first problem (ℓ = 1 \ell=1 ℓ = 1 ) describes frame dragging. The angular velocity of the fluid relative to the local inertial frames, ϖ = Ω − ω \varpi=\Omega-\omega ϖ = Ω − ω , obeys a second-order equation that is regular at the centre. Matching to the exterior solution ω = 2 J / r 3 \omega=2J/r^3 ω = 2 J / r 3 gives the angular momentum J J J and the moment of inertia I = J / Ω I=J/\Omega I = J /Ω .
The second problem (ℓ = 0 \ell=0 ℓ = 0 ) describes the spherical deformation. The functions m 0 m_0 m 0 and p 0 ∗ p_0^* p 0 ∗ give the change of the gravitational mass at fixed central density, δ M = m 0 ( R ) + J 2 / R 3 \delta M=m_0(R)+J^2/R^3 δ M = m 0 ( R ) + J 2 / R 3 , where R R R is the TOV radius. The mass that enters the exterior metric and all orbital frequencies is the total mass M = M T O V + δ M M=M_{\mathrm{TOV}}+\delta M M = M TOV + δ M . All masses quoted in this paper are total masses of rotating stars.
The third problem (ℓ = 2 \ell=2 ℓ = 2 ) describes the quadrupolar deformation. The functions h 2 h_2 h 2 and v 2 = h 2 + k 2 v_2=h_2+k_2 v 2 = h 2 + k 2 obey two coupled first-order equations. Matching them to the exterior solution fixes a constant K K K and the quadrupole moment
Q = − ( J 2 M + 8 5 K M 3 ) . (2) Q=-\left(\frac{J^2}{M}+\frac{8}{5}KM^3\right). \tag*{(2)} Q = − ( M J 2 + 5 8 K M 3 ) . (2)
We use the Geroch–Hansen sign convention, in which an oblate star has Q < 0 Q<0 Q < 0 and the Kerr metric has Q = − J 2 / M Q=-J^2/M Q = − J 2 / M . The dimensionless spin and quadrupole are j = J / M 2 j=J/M^2 j = J / M 2 and q = Q / M 3 q=Q/M^3 q = Q / M 3 . We also use the reduced quadrupole
Q ˉ = − q j 2 = − Q M J 2 , (3) \bar{Q}=-\frac{q}{j^2}=-\frac{QM}{J^2}, \tag*{(3)} Q ˉ = − j 2 q = − J 2 QM , (3)
which equals unity for a Kerr black hole and lies between about 1.5 and 12 for the neutron stars considered here (Figure 1 ). In the Hartle–Thorne (HT) approximation J ∝ Ω J\propto\Omega J ∝ Ω and Q , δ M ∝ Ω 2 Q,\delta M\propto\Omega^2 Q , δ M ∝ Ω 2 at fixed central density, so to leading order Q ˉ \bar{Q} Q ˉ depends on the EOS and the mass but not on the spin.
Figure 1. Stellar properties for the four EOSs. (a) Mass–radius relations of nonrotating stars. (b) Dimensionless spin j = c J / G M 2 j=cJ/GM^2 j = c J / G M 2 at ν s = 363 \nu_s=363 ν s = 363 Hz (thick) and 581 Hz (thin). (c) Reduced quadrupole Q ˉ = − q / j 2 \bar{Q}=-q/j^2 Q ˉ = − q / j 2 at 363 Hz; the dotted line is the Kerr value. (d) ISCO radius (thick) and equatorial radius (thin) in units of M M M at 581 Hz. The disc ends at the larger of the two.
The surface of the rotating star is r ( θ ) = R + ξ 0 + ξ 2 P 2 ( cos θ ) r(\theta)=R+\xi_0+\xi_2P_2(\cos\theta) r ( θ ) = R + ξ 0 + ξ 2 P 2 ( cos θ ) , with ξ 0 \xi_0 ξ 0 and ξ 2 \xi_2 ξ 2 from the ℓ = 0 \ell=0 ℓ = 0 and ℓ = 2 \ell=2 ℓ = 2 pressure perturbations (Appendix § 12 ). The equatorial coordinate radius R e q = R + ξ 0 − ξ 2 / 2 R_{\mathrm{eq}}=R+\xi_0-\xi_2/2 R eq = R + ξ 0 − ξ 2 /2 is the radius at which an accretion disc would touch the star, in the same coordinate as the exterior metric. The circumferential equatorial radius is R e q , c i r c = R e q ( 1 − k 2 / 2 ) R_{\mathrm{eq,circ}}=R_{\mathrm{eq}}(1-k_2/2) R eq , circ = R eq ( 1 − k 2 /2 ) , which is the quantity measured by pulse-profile modelling.
Outside the star the metric functions are known in closed form (Hartle & Thorne 1968). With ξ = r / M − 1 \xi=r/M-1 ξ = r / M − 1 and the associated Legendre functions of the second kind Q 2 1 Q_2^1 Q 2 1 , Q 2 2 Q_2^2 Q 2 2 (Appendix B),
ω = 2 J r 3 , h 0 = J 2 r 3 ( r − 2 M ) , m 0 = − J 2 r 3 , \omega=\frac{2J}{r^3},\qquad h_0=\frac{J^2}{r^3(r-2M)},\qquad m_0=-\frac{J^2}{r^3}, ω = r 3 2 J , h 0 = r 3 ( r − 2 M ) J 2 , m 0 = − r 3 J 2 ,
h 2 = J 2 M r 3 ( 1 + M r ) + K Q 2 2 ( ξ ) , h_2=\frac{J^2}{Mr^3}\left(1+\frac{M}{r}\right)+KQ_2^2(\xi), h 2 = M r 3 J 2 ( 1 + r M ) + K Q 2 2 ( ξ ) ,
v 2 = − J 2 r 4 + K 2 M r ( r − 2 M ) Q 2 1 ( ξ ) . (4) v_2=-\frac{J^2}{r^4}+K\frac{2M}{\sqrt{r(r-2M)}}Q_2^1(\xi). \tag*{(4)} v 2 = − r 4 J 2 + K r ( r − 2 M ) 2 M Q 2 1 ( ξ ) . (4)
with m 2 = ( r − 2 M ) ( − h 2 + 6 J 2 / r 4 ) m_2=(r-2M)(-h_2+6J^2/r^4) m 2 = ( r − 2 M ) ( − h 2 + 6 J 2 / r 4 ) and k 2 = v 2 − h 2 k_2=v_2-h_2 k 2 = v 2 − h 2 . The metric depends on ( M , J , Q ) (M,J,Q) ( M , J , Q ) only, so for given ( M , j , q ) (M,j,q) ( M , j , q ) the exterior and all orbital frequencies are fixed, whatever the interior. The EOS enters through the map ( M , ν s ) ↦ ( j , q , R e q ) (M,\nu_s)\mapsto(j,q,R_{\mathrm{eq}}) ( M , ν s ) ↦ ( j , q , R eq ) . Because the metric is correct to O ( Ω 2 ) \mathcal{O}(\Omega^2) O ( Ω 2 ) , setting q = − j 2 q=-j^2 q = − j 2 reproduces the Kerr metric to O ( j 2 ) \mathcal{O}(j^2) O ( j 2 ) , which is a useful check of the frequencies (Appendix C).
Orbital frequencies and the relativistic precession model
Circular orbits and epicyclic frequencies
Consider a test particle on a prograde circular orbit of radius r r r in the equatorial plane of a stationary, axisymmetric, reflection-symmetric metric with components g t t g_{tt} g tt , g t ϕ g_{t\phi} g tϕ , g ϕ ϕ g_{\phi\phi} g ϕϕ , g r r g_{rr} g r r , g θ θ g_{\theta\theta} g θ θ . The coordinate angular velocity Ω ϕ = d ϕ / d t \Omega_\phi=d\phi/dt Ω ϕ = d ϕ / d t follows from the radial geodesic equation,
Ω ϕ = − g t ϕ , r + g t ϕ , r 2 − g t t , r g ϕ ϕ , r g ϕ ϕ , r , (5) \Omega_\phi=\frac{-g_{t\phi,r}+\sqrt{g_{t\phi,r}^2-g_{tt,r}g_{\phi\phi,r}}}{g_{\phi\phi,r}}, \tag*{(5)} Ω ϕ = g ϕϕ , r − g tϕ , r + g tϕ , r 2 − g tt , r g ϕϕ , r , (5)
and the normalisation of the four-velocity gives u t = [ − ( g t t + 2 Ω ϕ g t ϕ + Ω ϕ 2 g ϕ ϕ ) ] − 1 / 2 u^t=[-(g_{tt}+2\Omega_\phi g_{t\phi}+\Omega_\phi^2g_{\phi\phi})]^{-1/2} u t = [ − ( g tt + 2 Ω ϕ g tϕ + Ω ϕ 2 g ϕϕ ) ] − 1/2 , the energy E = − ( g t t + g t ϕ Ω ϕ ) u t E=-(g_{tt}+g_{t\phi}\Omega_\phi)u^t E = − ( g tt + g tϕ Ω ϕ ) u t and the angular momentum L = ( g t ϕ + g ϕ ϕ Ω ϕ ) u t L=(g_{t\phi}+g_{\phi\phi}\Omega_\phi)u^t L = ( g tϕ + g ϕϕ Ω ϕ ) u t per unit mass. For fixed E E E and L L L the normalisation u ⋅ u = − 1 u\cdot u=-1 u ⋅ u = − 1 reads
g r r r ˙ 2 + g θ θ θ ˙ 2 = − 1 − H ( r , θ ) , g_{rr}\dot r^2+g_{\theta\theta}\dot\theta^2=-1-\mathcal{H}(r,\theta), g r r r ˙ 2 + g θ θ θ ˙ 2 = − 1 − H ( r , θ ) ,
H = g ϕ ϕ E 2 + 2 g t ϕ E L + g t t L 2 g t t g ϕ ϕ − g t ϕ 2 , (6) \mathcal{H}=\frac{g_{\phi\phi}E^2+2g_{t\phi}EL+g_{tt}L^2}{g_{tt}g_{\phi\phi}-g_{t\phi}^2}, \tag*{(6)} H = g tt g ϕϕ − g tϕ 2 g ϕϕ E 2 + 2 g tϕ E L + g tt L 2 , (6)
where the dot denotes the derivative with respect to proper time. Linearising about the circular orbit gives harmonic oscillations in r r r and θ \theta θ with proper-time frequencies κ r 2 = H , r r / 2 g r r \kappa_r^2=\mathcal{H}_{,rr}/2g_{rr} κ r 2 = H , r r /2 g r r and κ θ 2 = H , θ θ / 2 g θ θ \kappa_\theta^2=\mathcal{H}_{,\theta\theta}/2g_{\theta\theta} κ θ 2 = H , θ θ /2 g θ θ (Appendix C). The frequencies seen by a distant observer are
ν ϕ = Ω ϕ 2 π , ν r = κ r 2 π u t , ν θ = κ θ 2 π u t . (7) \nu_\phi=\frac{\Omega_\phi}{2\pi},\qquad\nu_r=\frac{\kappa_r}{2\pi u^t},\qquad\nu_\theta=\frac{\kappa_\theta}{2\pi u^t}. \tag*{(7)} ν ϕ = 2 π Ω ϕ , ν r = 2 π u t κ r , ν θ = 2 π u t κ θ . (7)
The innermost stable circular orbit (ISCO) is the outermost zero of κ r 2 \kappa_r^2 κ r 2 . The disc ends at r i n = max ( R e q , r I S C O ) r_{\mathrm{in}}=\max(R_{\mathrm{eq}},r_{\mathrm{ISCO}}) r in = max ( R eq , r ISCO ) , that is, at the ISCO or at the stellar surface, whichever lies farther out. We evaluate equations (5) –(7) for the full HT metric of equation (4) , with fourth-order finite differences. We do not expand in j j j or q q q , so the frequencies are those of the metric in which we integrate the particle orbits (§ 5 ).
Periastron and nodal precession
The differences
ν p e r = ν ϕ − ν r , ν n o d = ν ϕ − ν θ (8) \nu_{\mathrm{per}}=\nu_\phi-\nu_r,\qquad\nu_{\mathrm{nod}}=\nu_\phi-\nu_\theta \tag*{(8)} ν per = ν ϕ − ν r , ν nod = ν ϕ − ν θ (8)
are the periastron and nodal precession frequencies of a slightly eccentric, slightly tilted orbit. Their weak-field structure explains most of what follows. To leading order in each multipole (Appendix D),
ν r 2 ν ϕ 2 ≃ 1 − 6 x + 8 j x 3 / 2 + 3 q x 2 , ν θ 2 ν ϕ 2 ≃ 1 − 4 j x 3 / 2 − 3 q x 2 (9) \frac{\nu_{r}^{2}}{\nu_{\phi}^{2}} \simeq1-\frac{6}{x}+\frac{8j}{x^{3/2}}+\frac{3q}{x^{2}}, \\
\frac{\nu_{\theta}^{2}}{\nu_{\phi}^{2}} \simeq1-\frac{4j}{x^{3/2}}-\frac{3q}{x^{2}} \tag*{(9)} ν ϕ 2 ν r 2 ≃ 1 − x 6 + x 3/2 8 j + x 2 3 q , ν ϕ 2 ν θ 2 ≃ 1 − x 3/2 4 j − x 2 3 q (9)
with x = r / M x=r/M x = r / M . For q = − j 2 q=-j^{2} q = − j 2 these reduce to the familiar Kerr expressions (Okazaki et al. 1987; Stella & Vietri 1998), and for j = 0 j=0 j = 0 they give the Newtonian precession in the field of a mass quadrupole. The nodal precession is then
ν n o d ≃ ν L T + ν Q , ν L T = ω 2 π = G J π c 2 r 3 , ν Q = 3 2 q x 2 ν ϕ . (10) \nu_{\mathrm{nod}} \simeq\nu_{\mathrm{LT}}+\nu_{Q}, \qquad
\nu_{\mathrm{LT}}=\frac{\omega}{2\pi}=\frac{GJ}{\pi c^{2}r^{3}}, \qquad
\nu_{Q}=\frac{3}{2}\frac{q}{x^{2}}\nu_{\phi}. \tag*{(10)} ν nod ≃ ν LT + ν Q , ν LT = 2 π ω = π c 2 r 3 G J , ν Q = 2 3 x 2 q ν ϕ . (10)
The first term is the Lense–Thirring precession, equal to the frame-dragging angular velocity of zero-angular-momentum observers. The second is the classical precession driven by the oblateness of the star. It is retrograde (q < 0 q<0 q < 0 ) and grows as ν s 2 \nu_{s}^{2} ν s 2 , while ν L T \nu_{\mathrm{LT}} ν LT grows as ν s \nu_{s} ν s . Their ratio is
ν Q ν L T = − 3 4 Q ˉ j x . (11) \frac{\nu_{Q}}{\nu_{\mathrm{LT}}}=-\frac{3}{4}\frac{\bar{Q}j}{\sqrt{x}}. \tag*{(11)} ν LT ν Q = − 4 3 x Q ˉ j . (11)
which is about −0.1 to −0.3 near the ISCO for the 1.9 M ⊙ M_{\odot} M ⊙ stars of § 4 at ν s = 363 \nu_{s}=363 ν s = 363 –581 Hz. The quadrupole is therefore not a small correction to the nodal frequency, and replacing the star by a Kerr metric with the same j j j (Q ˉ = 1 \bar{Q}=1 Q ˉ = 1 ) overestimates ν n o d \nu_{\mathrm{nod}} ν nod . The periastron precession is dominated by the 6 / x 6/x 6/ x term and is much less sensitive to j j j and q q q .
The relativistic precession model
In the RPM (Stella & Vietri 1998, 1999; Stella et al. 1999) the three QPOs of a source come from the same radius,
ν U = ν ϕ , ν L = ν p e r , ν L F = 2 ν n o d , (12) \nu_{\mathrm{U}}=\nu_{\phi}, \qquad\nu_{\mathrm{L}}=\nu_{\mathrm{per}}, \qquad\nu_{\mathrm{LF}}=2\nu_{\mathrm{nod}}, \tag*{(12)} ν U = ν ϕ , ν L = ν per , ν LF = 2 ν nod , (12)
where ν U \nu_{\mathrm{U}} ν U and ν L \nu_{\mathrm{L}} ν L are the upper and lower kHz QPOs. The low-frequency QPO ν L F \nu_{\mathrm{LF}} ν LF is the 20–50 Hz QPO of atoll sources or the horizontal-branch oscillation (HBO) of Z sources. The factor 2 in the last relation was introduced by Stella et al. (1999) because the nodal frequency alone falls short of the observed low frequencies. It can be justified if the modulation has two maxima per precession cycle. Sets of QPOs observed at different times come from different radii; once ( M , j , q ) (M,j,q) ( M , j , q ) are fixed, this radius is the only freedom of the model. § 5 shows how the three frequencies appear in the flux of matter on a real orbit.
The pair ( ν U , ν L ) (\nu_{\mathrm{U}},\nu_{\mathrm{L}}) ( ν U , ν L ) mainly constrains the mass, because ν p e r / ν ϕ \nu_{\mathrm{per}}/\nu_{\phi} ν per / ν ϕ depends mainly on x x x , and ν ϕ \nu_{\phi} ν ϕ then fixes M M M . The low-frequency QPO adds information on the combination j ( 1 − 3 4 Q ˉ j / x ) j(1-\frac{3}{4}\bar{Q}j/\sqrt{x}) j ( 1 − 4 3 Q ˉ j / x ) [equations (10) –(11) ]. With a known spin frequency, j = 2 π ν s I / M 2 j=2\pi\nu_{s}I/M^{2} j = 2 π ν s I / M 2 , so the low-frequency QPO measures a combination of the moment of inertia and the quadrupole. This is where an EOS-consistent spacetime matters.
Equations of state and stellar properties
Equations of state
We use four EOSs that give radii of 10.5–13.3 km at 1.4 M ⊙ M_{\odot} M ⊙ and maximum masses of 2.0–2.15 M ⊙ M_{\odot} M ⊙ . WFF1 (Wiringa et al. 1988) is a microscopic variational EOS (AV14 two-nucleon plus Urbana VII three-nucleon interactions), matched to the Negele–Vautherin crust; we use the table distributed with the RNS code (Stergioulas & Friedman 1995). It is soft at intermediate densities, with R 1.4 = 10.5 R_{1.4}=10.5 R 1.4 = 10.5 km, but still reaches M m a x = 2.15 M ⊙ M_{\mathrm{max}}=2.15 M_{\odot} M max = 2.15 M ⊙ . Its radius lies below the range inferred from NICER data (Miller et al. 2021), and we keep it as a limiting compact case. G3 (Kumar et al. 2017), FSUGarnet (Chen & Piekarewicz 2015) and IOPB-I (Kumar et al. 2018) are relativistic mean-field (RMF) EOSs calibrated to finite nuclei. We use the unified versions of Parmar et al. (2022), in which the inner crust is computed with the same functional as the core and the outer crust from measured and Hartree–Fock–Bogoliubov nuclear masses; the tables are available at https://github.com/hcdas/Unified_eos. They give R 1.4 = 12.6 R_{1.4}=12.6 R 1.4 = 12.6 , 13.1 and 13.3 km and M m a x = 2.00 M_{\mathrm{max}}=2.00 M max = 2.00 , 2.07 and 2.15 M ⊙ M_{\odot} M ⊙ . All four reach about 2 M ⊙ M_{\odot} M ⊙ , as required by the heaviest radio pulsars (Fonseca et al. 2021), but none reaches 2.2 M ⊙ M_{\odot} M ⊙ without rotation. Only the relation ε ( P ) \varepsilon(P) ε ( P ) enters the structure equations.
Spin, quadrupole and the inner edge of the disc
Figure 1 shows the properties that enter the orbital frequencies. At fixed mass and spin, j j j grows with the stellar radius, because j = 2 π ν s I / M 2 j=2\pi\nu_{s}I/M^{2} j = 2 π ν s I / M 2 and I ∼ M R 2 I\sim MR^{2} I ∼ M R 2 . At 1.9 M ⊙ M_{\odot} M ⊙ and 363 Hz, j j j ranges from 0.130 (WFF1) to 0.179 (IOPB-I). The reduced quadrupole Q ˉ \bar{Q} Q ˉ falls with compactness, from about 12 at 1 M ⊙ M_{\odot} M ⊙ for the RMF EOSs to 2–3.6 at 1.9 M ⊙ M_{\odot} M ⊙ , and it is always larger than the Kerr value Q ˉ = 1 \bar{Q}=1 Q ˉ = 1 . At 1.9 M ⊙ M_{\odot} M ⊙ the quadrupole q = − Q ˉ j 2 q=-\bar{Q}j^{2} q = − Q ˉ j 2 is thus 2–4 times larger in magnitude than for a black hole with the same j j j . When I ˉ = I / M 0 3 \bar{I}=I/M_{0}^{3} I ˉ = I / M 0 3 and Q ˉ = − Q M 0 / J 2 \bar{Q}=-QM_{0}/J^{2} Q ˉ = − Q M 0 / J 2 are formed with the nonrotating mass M 0 M_{0} M 0 , our stars follow the EOS-insensitive I ˉ \bar{I} I ˉ –Q ˉ \bar{Q} Q ˉ relation of Yunes (2013) to 0.5–1.3 per cent, which is a check of the ℓ = 1 \ell=1 ℓ = 1 and ℓ = 2 \ell=2 ℓ = 2 solutions. In the exterior metric, and everywhere else in this paper, j j j and q q q are normalised by the total mass M M M ; the two normalisations differ at order δ M / M \delta M/M δ M / M .
The rotation increases the mass at fixed central density. For a star of total mass 1.4 M ⊙ M_{\odot} M ⊙ , δ M / M = 1 \delta M/M=1 δ M / M = 1 –2 per cent at 363 Hz and 3–6 per cent at 581 Hz (WFF1 to IOPB-I), and about half of that at 1.9 M ⊙ M_{\odot} M ⊙ . The maximum mass grows by only 0.3–0.8 per cent at 363 Hz (to 2.15, 2.01, 2.08 and 2.17 M ⊙ M_{\odot} M ⊙ ), because compact stars are less deformed. At a given total mass a rotating star therefore has a lower central density than a nonrotating one, and a larger radius. This matters because the fits place several sources close to the maximum mass.
The inner edge of the disc is the ISCO when the star is compact enough. In Figure 1 (d) the ISCO lies at x ≃ 5.3 x\simeq5.3 x ≃ 5.3 –5.8 and depends weakly on the EOS through j j j and q q q . The equatorial radius R e q R_{\mathrm{eq}} R eq falls steeply with mass. At 581 Hz the star extends beyond the ISCO below M = 1.32 M ⊙ M=1.32 M_{\odot} M = 1.32 M ⊙ (WFF1), 1.59 (G3), 1.67 (FSUGarnet) and 1.70 M ⊙ M_{\odot} M ⊙ (IOPB-I). In those cases the disc reaches the stellar surface and the highest orbital frequency is the Keplerian frequency at R e q R_{\mathrm{eq}} R eq . For the masses preferred by the fits below, the inner edge is always the ISCO.
Table 1. Stellar properties for the four EOSs. M max M_{\max} M m a x is the maximum mass of nonrotating stars, R 1.4 R_{1.4} R 1.4 and R 1.9 R_{1.9} R 1.9 are nonrotating radii, and I ˉ 1.4 = I / M 3 \bar{I}_{1.4}=I/M^3 I ˉ 1.4 = I / M 3 is given at 1.4 M ⊙ 1.4M_\odot 1.4 M ⊙ . I I I , j j j , Q ˉ = − q / j 2 \bar{Q}=-q/j^2 Q ˉ = − q / j 2 and R e q / M R_{\mathrm{eq}}/M R eq / M are given at M = 1.9 M ⊙ M=1.9M_\odot M = 1.9 M ⊙ and ν s = 363 \nu_s=363 ν s = 363 Hz.
Orbital frequencies
Figure 2 shows the frequencies of § 3 for an IOPB-I star of 1.922 M ⊙ 1.922M_\odot 1.922 M ⊙ at 363 Hz (j = 0.178 j=0.178 j = 0.178 , Q ˉ = 3.47 \bar{Q}=3.47 Q ˉ = 3.47 ). This is the preferred model for 4U 1728−34 (§ 6 ). The ISCO is at 15.65 km (x = 5.51 x=5.51 x = 5.51 ), 2.4 km outside the stellar equator, where ν ϕ = 1285 \nu_\phi=1285 ν ϕ = 1285 Hz. The vertical frequency ν θ \nu_\theta ν θ is within 2 per cent of ν ϕ \nu_\phi ν ϕ . The radial frequency ν r \nu_r ν r has a maximum of 414 Hz at x ≈ 7.4 x\approx7.4 x ≈ 7.4 and vanishes at the ISCO, so ν p e r = ν L \nu_{\mathrm{per}}=\nu_{\mathrm{L}} ν per = ν L rises from 0 at large radii to ν ϕ \nu_\phi ν ϕ at the ISCO. Panel (b) shows 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod together with the Lense–Thirring term 2 ν L T 2\nu_{\mathrm{LT}} 2 ν LT , the quadrupole term 2 ν Q 2\nu_Q 2 ν Q and their weak-field sum, equation (10) . Near the ISCO the exact nodal frequency is 28 per cent below the Lense–Thirring term. The weak-field sum recovers most of this difference. It overestimates the exact value by 12 per cent at the ISCO and by 3 per cent at x = 10 x=10 x = 10 ; the rest is due to higher-order terms in 1 / x 1/x 1/ x .
Figure 2. Orbital frequencies around an IOPB-I star of 1.922 M ⊙ 1.922\,M_\odot 1.922 M ⊙ spinning at 363 Hz. (a) Azimuthal, vertical and radial epicyclic frequencies and the periastron precession frequency (the RPM ν U \nu_{\mathrm{U}} ν U and ν L \nu_{\mathrm{L}} ν L ). (b) Twice the nodal precession frequency (the RPM ν L F \nu_{\mathrm{LF}} ν LF ), the Lense–Thirring term 2 ν L T = ω / π 2\nu_{\mathrm{LT}}=\omega/\pi 2 ν LT = ω / π , the (retrograde) quadrupole term − 2 ν Q -2\nu_{\mathrm{Q}} − 2 ν Q and the weak-field sum, equation (10) . The dotted vertical line is the ISCO and the shaded band is the interior of the star.
Figure 3 (a) compares the HT nodal frequency with that of a Kerr metric with the same M M M and j j j . At 1.9 M ⊙ 1.9M_\odot 1.9 M ⊙ and 363 Hz the HT value is lower by 3–7 per cent (WFF1), 7–14 per cent (G3), 10–22 per cent (FSUGarnet) and 11–25 per cent (IOPB-I) between x = 14 x=14 x = 14 and the ISCO. The deficit grows with the radius of the star, because both j j j and Q ˉ \bar{Q} Q ˉ grow. Panel (b) shows a less familiar consequence. At a fixed upper kHz QPO frequency (ν U = 900 \nu_{\mathrm{U}}=900 ν U = 900 Hz) the nodal frequency first grows linearly with the spin frequency, as ν L T ∝ ν s \nu_{\mathrm{LT}}\propto\nu_s ν LT ∝ ν s , but the quadrupole term grows as ν s 2 \nu_s^2 ν s 2 . For the stiffer RMF EOSs the two balance at 600–650 Hz, and 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod reaches a maximum of 32–34 Hz. For the more compact G3 and WFF1 stars it is still rising at 750 Hz, the highest spin shown, where it reaches 38–40 Hz. A fast spin alone therefore cannot produce a high nodal frequency. Török et al. (2025) found the same behaviour at fixed radius: the precession frequency has a maximum as a function of j j j . The height of the maximum depends on the mass and, through I I I and Q ˉ \bar{Q} Q ˉ , on the EOS. It decides the interpretation of the Z sources in § 6 .
Figure 3. (a) Ratio of the nodal precession frequency in the HT metric to that in a Kerr metric with the same M M M and j j j , for M = 1.9 M ⊙ M=1.9M_\odot M = 1.9 M ⊙ and ν s = 363 \nu_{\mathrm{s}}=363 ν s = 363 Hz. (b) 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod at the radius where ν ϕ = 900 \nu_\phi=900 ν ϕ = 900 Hz as a function of the spin frequency, for M = 1.9 M ⊙ M=1.9M_\odot M = 1.9 M ⊙ . The spin-induced quadrupole limits the nodal frequency that a star can produce.
Particle orbits and QPO peaks
The epicyclic frequencies of § 3 describe small oscillations about a circular orbit. In this section we follow a full geodesic of finite amplitude and ask how its frequencies would appear in the X-ray flux.
An eccentric, tilted orbit
We take the IOPB-I star of Figure 2 (M = 1.922 M ⊙ M=1.922M_\odot M = 1.922 M ⊙ , ν s = 363 \nu_s=363 ν s = 363 Hz, j = 0.178 j=0.178 j = 0.178 , Q ˉ = 3.47 \bar{Q}=3.47 Q ˉ = 3.47 , R e q = 13.21 R_{\mathrm{eq}}=13.21 R eq = 13.21 km, ISCO at 15.65 km), which is the preferred model for 4U 1728−34 (§ 6 ). We integrate the geodesic equation x ¨ μ = − Γ α β μ x ˙ α x ˙ β \ddot{x}^\mu=-\Gamma^\mu_{\alpha\beta}\dot{x}^\alpha\dot{x}^\beta x ¨ μ = − Γ α β μ x ˙ α x ˙ β in the full HT metric with an eighth-order Runge–Kutta method (relative tolerance 10 − 10 10^{-10} 1 0 − 10 ); the equations and initial data are given in Appendix E. The orbit starts on the equator at r 0 = 17.39 r_0=17.39 r 0 = 17.39 km (x = 6.13 x=6.13 x = 6.13 ), where the circular orbital frequency is ν ϕ = 1098.4 \nu_\phi=1098.4 ν ϕ = 1098.4 Hz, equal to the upper kHz QPO of one triplet of 4U 1728−34 (Anand et al. 2024). The initial angular velocity is that of the circular orbit, tilted by 8° out of the equatorial plane, and a radial velocity of 1 per cent of the orbital speed is added. We integrate over 0.5 s of coordinate time, or 549 orbits. Along the orbit ∣ u ⋅ u + 1 ∣ ≲ 3 × 10 − 11 |u\cdot u+1|\lesssim3\times10^{-11} ∣ u ⋅ u + 1∣ ≲ 3 × 1 0 − 11 , and the energy and the axial angular momentum are conserved to 2 × 10 − 12 2\times10^{-12} 2 × 1 0 − 12 and 10 − 10 10^{-10} 1 0 − 10 .
Figure 4 (a) shows the first 20 ms of the orbit. The radius oscillates between 16.82 and 17.94 km with the radial epicyclic period 1 / ν r = 2.90 1/\nu_r=2.90 1/ ν r = 2.90 ms [panel (b)], and the particle crosses the equatorial plane twice per vertical period 1 / ν θ = 0.927 1/\nu_\theta=0.927 1/ ν θ = 0.927 ms, reaching ∣ z ∣ = 2.5 |z|=2.5 ∣ z ∣ = 2.5 km [panel (c)]. The two periods differ from the orbital period 1 / ν ϕ = 0.910 1/\nu_\phi=0.910 1/ ν ϕ = 0.910 ms, and the orbit is therefore not closed. At this radius ν r ≈ ν ϕ / 3 \nu_r\approx\nu_\phi/3 ν r ≈ ν ϕ /3 , so the particle makes about three turns per radial oscillation, and the pericentre moves forward from one radial period to the next. Its longitude ϖ \varpi ϖ grows at the rate ν p e r = ν ϕ − ν r \nu_{\mathrm{per}}=\nu_\phi-\nu_r ν per = ν ϕ − ν r [panel (d)]. The line of nodes, where the orbit crosses the equatorial plane upward, rotates in the prograde direction at ν n o d = ν ϕ − ν θ \nu_{\mathrm{nod}}=\nu_\phi-\nu_\theta ν nod = ν ϕ − ν θ [panel (e)]. In 20 ms the orbital plane turns by about 140°, which is visible in panel (a) as a wobbling ring. This is the precession of the orbital plane by frame dragging (at ν L T = 25.9 \nu_{\mathrm{LT}=25.9} ν LT = 25.9 Hz here), reduced by the torque of the stellar oblateness (ν Q = − 4.8 \nu_{\mathrm{Q}}=-4.8 ν Q = − 4.8 Hz in the weak-field estimate).
Figure 4. Geodesic around an IOPB-I star of 1.922 M ⊙ M_\odot M ⊙ spinning at 363 Hz, started at r 0 = 17.39 r_0 = 17.39 r 0 = 17.39 km with an inclination of 8 ∘ 8^\circ 8 ∘ and a radial velocity of 1 per cent of the orbital speed. (a) First 20 ms of the orbit in the coordinates x = r sin θ cos ϕ x = r \sin\theta\cos\phi x = r sin θ cos ϕ , y = r sin θ sin ϕ y = r \sin\theta\sin\phi y = r sin θ sin ϕ , z = r cos θ z = r \cos\theta z = r cos θ (true scale); the colour gives the time. The grey sphere has the equatorial radius of the star and the dashed circle is the ISCO. (b) Radial and (c) vertical oscillation; the dashed line in (b) is r 0 r_0 r 0 . (d) Longitude of the pericentre, measured at successive pericentres (circles), and the line of slope ν p e r \nu_{\rm per} ν per predicted by the epicyclic frequencies. (e) Longitude of the ascending node (solid) and the line of slope ν n o d \nu_{\rm nod} ν nod (dashed) over 0.5 s.
Measured directly from the orbit, the pericentre advances at ν p e r = 760.9 \nu_{\mathrm{per}}=760.9 ν per = 760.9 Hz and the node at ν n o d = 19.51 \nu_{\mathrm{nod}}=19.51 ν nod = 19.51 Hz. The epicyclic values at r 0 r_0 r 0 are 753.9 and 19.31 Hz. The differences of 1 per cent are finite-amplitude effects. For a radial velocity of 0.3 per cent of the orbital speed and an inclination of 0.4° the same comparison at six radii between 17.5 and 47 km agrees to 0.01–0.02 per cent, and doubling the amplitudes multiplies the differences by four, as expected for second-order corrections. The epicyclic frequencies of the HT metric are therefore the precession frequencies of actual orbits of small amplitude.
Light curve and power spectrum of a hot spot
Suppose that a small, isotropically emitting spot of gas moves on this orbit. A distant observer in the direction n ^ \hat{n} n ^ receives photons with the redshift factor
g = ν o b s ν e m = 1 u t ( 1 − n ^ ⋅ v ) . (13) g=\frac{\nu_{\mathrm{obs}}}{\nu_{\mathrm{em}}}=\frac{1}{u^t(1-\hat{n}\cdot\mathbf{v})}. \tag*{(13)} g = ν em ν obs = u t ( 1 − n ^ ⋅ v ) 1 . (13)
where v = d x / d t \mathbf{v}=d\mathbf{x}/dt v = d x / d t is the coordinate velocity of the spot. The factor u t u^t u t contains the gravitational and transverse Doppler shifts and 1 − n ^ ⋅ v 1-\hat{n}\cdot\mathbf{v} 1 − n ^ ⋅ v , the longitudinal Doppler shift. Because I ν / ν 3 I_\nu/\nu^3 I ν / ν 3 is invariant, the bolometric flux of a small spot scales as F ∝ g 4 F \propto g^4 F ∝ g 4 . We neglect light bending, which changes the shape of the pulse but not its periodicities, and include the light-travel time across the orbit, t o b s = t − n ^ ⋅ x t_{\rm obs} = t - \hat{\boldsymbol{n}} \cdot\boldsymbol{x} t obs = t − n ^ ⋅ x . The observer is inclined by 60 ∘ 60^\circ 6 0 ∘ to the spin axis. Appendix E derives the redshift factor and shows which combinations of the orbital frequencies the flux contains.
The light curve [Figure 5 (a)] is dominated by Doppler boosting: the flux rises to 4.5 times its mean when the spot approaches the observer and falls to 0.18 times the mean when it recedes, once per orbit. The power spectrum of 0.5 s of data [Figure 5 (b)] has its strongest peak at the orbital frequency and its harmonics. Below these it shows a series of weaker peaks, at 19.8, 340, 762 and 1082 Hz (frequency resolution 2 Hz), which are ν n o d \nu_{\rm nod} ν nod , ν r \nu_r ν r , ν p e r \nu_{\rm per} ν per and ν θ \nu_\theta ν θ of the orbit. Their origin is kinematic. The line-of-sight velocity varies at ν ϕ \nu_\phi ν ϕ , and the radial epicycle modulates its amplitude at ν r \nu_r ν r , which produces sidebands at ν ϕ ± ν r \nu_\phi\pm\nu_r ν ϕ ± ν r . The lower sideband is the periastron precession frequency. The vertical motion adds a term at ν θ \nu_\theta ν θ , of amplitude proportional to sin ι cos i \sin\iota\cos i sin ι cos i , where ι \iota ι is the orbital inclination and i i i that of the observer. Because F ∝ g 4 F \propto g^4 F ∝ g 4 is a nonlinear function of the velocity, the orbital and vertical terms beat, and the difference frequency ν ϕ − ν θ \nu_\phi- \nu_\theta ν ϕ − ν θ is the nodal precession frequency. The flux of matter on a precessing orbit therefore contains the three frequencies of the RPM.
Figure 5. Hot spot on the orbit of Figure 4 , seen at an inclination of 60 ∘ 60^\circ 6 0 ∘ . (a) Bolometric flux, F ∝ g 4 F \propto g^4 F ∝ g 4 , over the first 6 ms. (b) Power spectrum of 0.5 s of the light curve, normalised to its maximum. Dashed lines are the epicyclic frequencies at r 0 r_0 r 0 : ν n o d \nu_{\rm nod} ν nod , ν r \nu_r ν r , ν p e r \nu_{\rm per} ν per , ν θ \nu_\theta ν θ and ν ϕ \nu_\phi ν ϕ ; the dotted line is 2 ν n o d 2\nu_{\rm nod} 2 ν nod . The shaded bands are the low-frequency QPO (LF) and the lower (L) and upper (U) kHz QPOs of the 4U 1728−34 triplet with ν U = 1098.4 \nu_U = 1098.4 ν U = 1098.4 Hz measured by Anand et al. (2024) (widths: 1 σ 1\sigma 1 σ errors, at least 1.2 per cent).
The shaded bands in Figure 5 (b) are the observed triplet of 4U 1728−34 whose upper kHz QPO fixed the radius. The periastron frequency lies 2.6 per cent (epicyclic value, 754 Hz) to 3.6 per cent (peak, 762 Hz) above the observed lower kHz QPO (734.9 ± 1.2 Hz). Part of this is the finite amplitude of the orbit, and part reflects the fit of this star to the whole data set (§ 6 ). The observed low-frequency QPO (38.6 ± 1.7 Hz) coincides with 2 ν n o d 2\nu_{\rm nod} 2 ν nod , as assumed in the RPM. The hot spot, however, puts most of its low-frequency power at ν n o d \nu_{\rm nod} ν nod itself, and its peak at 2 ν n o d 2\nu_{\rm nod} 2 ν nod is weaker by almost three orders of magnitude. The factor of 2 of the RPM therefore calls for an emitting structure that modulates the flux twice per precession cycle, as assumed by Stella et al. (1999), rather than a single orbiting spot.
The amplitudes of the peaks also show the limits of the picture. Relative to the orbital peak the precession peaks are 10 − 3 10^{-3} 1 0 − 3 –10 − 2 10^{-2} 1 0 − 2 in power, whereas the observed lower and upper kHz QPOs have comparable amplitudes. The hot spot shows which frequencies an orbit can imprint on the flux, not how strongly a real accretion flow does so.
Masses and spins from the observed QPOs
Data and fitting
We use four published data sets (Table 2 ). The atoll sources have spin frequencies known from burst oscillations, 581 Hz for 4U 1636−53 (Strohmayer & Markwardt (2002)) and 363 Hz for 4U 1728−34 (Strohmayer et al. (1996)). For 4U 1636−53 we use the 28 pairs of twin kHz QPOs of Zhang et al. (2017) from RXTE data, as tabulated by Shi et al. (2024). For 4U 1728−34 we use the 16 sets of simultaneous QPOs detected by Anat et al. (2024) in AstroSat/LAXPC data; 13 are complete triplets of a low-frequency QPO and twin kHz QPOs. The lower kHz QPO is measured to about 1 Hz, which makes this the most precise data set. The Z sources have simultaneous HBOs and twin kHz QPOs but unknown spins. For GX
Table 2. QPO data. N N N is the number of QPO sets; ranges are in Hz. The low-frequency QPO of 4U 1728−34 and the HBOs of the Z sources are listed under ν L F \nu_{\rm LF} ν LF . Sources of the data: 4U 1636−53 (Zhang et al. 2017; Shi et al. 2024), 4U 1728−34 (Anand et al. 2024), GX 5−1 (Jonker et al. 2002) and GX 340+0 (Jonker et al. 2000); spin frequencies from Strohmayer & Markwardt (2002); Strohmayer et al. (1996).
5−1 we pair the HBO and kHz QPO frequencies of the same selections on the Z track measured by Jonker et al. (2002), and for GX 340+0 we use the nine selections of Jonker et al. (2000) in which both kHz peaks have free frequencies. Asymmetric errors were symmetrised.
For a given EOS, mass and spin frequency the star fixes ( j , q , R e q ) (j, q, R_{\rm eq}) ( j , q , R eq ) , and the RPM gives ν L \nu_L ν L and ν L F \nu_{\rm LF} ν LF as functions of ν U \nu_U ν U [equation (12) ], parametrised by the radius between the inner edge of the disc and 25 M 25M 25 M . The radius of each observed set of QPOs is unknown, so we marginalise over it with a uniform weight, which can be done exactly (Appendix F).
We use Gaussian errors, to which we add in quadrature an intrinsic scatter proportional to the frequency, one for the kHz QPOs and one for the low-frequency QPO, because the data scatter about any RPM curve by more than their errors. The scatters are free parameters. The mass has a uniform prior on 1.0–2.6 M ⊙ M_\odot M ⊙ . The spin frequency is fixed for the atoll sources and uniform on 200–750 Hz for the Z sources. To fit without an EOS we replace the star by free parameters ( M , j , Q ˉ ) (M, j, \bar{Q}) ( M , j , Q ˉ ) with uniform priors. The posteriors are computed on grids and, for the fits of 4U 1728−34 and the Z sources without an EOS, by nested sampling (Ashton et al. 2019; Speagle 2020). Table 3 lists the results, and Figure 6 compares the best-fitting curves with the data of the atoll sources.
Table 3. Masses and spins from the QPOs (medians and 68 per cent intervals). For the atoll sources ν s \nu_s ν s is fixed by the burst oscillations; for the Z sources it is a free parameter. j j j , Q ˉ = − q / j 2 \bar{Q} = -q/j^2 Q ˉ = − q / j 2 and I I I follow from M M M , ν s \nu_s ν s and the EOS. 4U 1636−53: kHz QPOs only; the other sources: kHz QPOs and the low-frequency QPO or HBO, identified with 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod .
Figure 6. Atoll sources: data and best-fitting RPM curves for the four EOSs (best-fit masses in M ⊙ M_{\odot} M ⊙ in the legends). (a) 4U 1636−53, kHz QPOs of Zhang et al. (2017) as tabulated by Shi et al. (2024). (b) 4U 1728−34, kHz QPOs, and (c) its low-frequency QPO against the lower kHz QPO, from AstroSat/LAXPC data of Anand et al. (2024). Error bars are 1 σ 1\sigma 1 σ measurement errors.
4U 1636-53
The twin kHz QPOs can be described with every EOS, but only with an intrinsic scatter of 4–5 per cent, or 40–50 Hz at ν U ≈ 1000 \nu_U \approx1000 ν U ≈ 1000 Hz: the data scatter about every RPM curve [Figure 6 (a)]. The masses are high, 2.02–2.15 M ⊙ M_\odot M ⊙ , and lie within 0.06 M ⊙ M_\odot M ⊙ of the maximum masses of the stars at 581 Hz (2.03, 2.11, 2.16 and 2.19 M ⊙ M_\odot M ⊙ for G3, FSUGarnet, WFF1 and IOPB-I). Two effects combine. The RPM mass of this source is known to be close to 2 M ⊙ M_\odot M ⊙ (Török et al. 2012). In addition, at 581 Hz the spin of the star is large, j = 0.19 j = 0.19 j = 0.19 –0.25, and the mass inferred from kHz QPOs grows with j j j : a fit without an EOS prefers M = 1.8 9 − 0.05 + 0.08 M ⊙ M = 1.89^{+0.08}_{-0.05} M_\odot M = 1.8 9 − 0.05 + 0.08 M ⊙ with a small spin, j = 0.0 5 − 0.04 + 0.07 j = 0.05^{+0.07}_{-0.04} j = 0.0 5 − 0.04 + 0.07 . Given the large scatter, these data do not discriminate between the EOSs.
4U 1728-34
The kHz QPOs alone are fitted with a scatter below 0.6 per cent and give M = 1.88 M = 1.88 M = 1.88 –1.92 M ⊙ M_\odot M ⊙ , depending on the EOS. They fix the mass for a given j j j , but their evidence is the same for all four EOSs to within Δ ln Z = 0.4 \Delta\ln Z = 0.4 Δ ln Z = 0.4 . The low-frequency QPO, identified with 2 ν n o d 2\nu_{\rm nod} 2 ν nod , separates them [Figure 6 (c)]. At the fitted masses (M = 1.877 M = 1.877 M = 1.877 –1.922 M ⊙ M_\odot M ⊙ ) the four EOSs give I = 1.78 I = 1.78 I = 1.78 , 2.12, 2.42 and 2.53 × 10 45 \times10^{45} × 1 0 45 g cm2 ^2 2 (WFF1, G3, FSUGarnet, IOPB-I), and the nodal frequency grows with I I I through j = 2 π ν s I / M 2 j = 2\pi\nu_s I/M^2 j = 2 π ν s I / M 2 . IOPB-I and FSUGarnet reproduce the low-frequency QPO with a scatter of 5–6 per cent, comparable to its measurement errors, while G3 and WFF1 predict nodal frequencies that are systematically too low and need 10 per cent and 17 per cent. The Bayesian evidence accordingly favours IOPB-I and FSUGarnet, by Δ ln Z ∼ 7 \Delta\ln Z \sim7 Δ ln Z ∼ 7 –8 over G3 and 16–18 over WFF1; the difference between IOPB-I and FSUGarnet (Δ ln Z = 1.3 \Delta\ln Z = 1.3 Δ ln Z = 1.3 ) is not significant. The preferred star has M = 1.922 ± 0.010 M ⊙ M = 1.922 \pm0.010\,M_\odot M = 1.922 ± 0.010 M ⊙ , j = 0.178 j = 0.178 j = 0.178 and Q ˉ = 3.47 \bar{Q} = 3.47 Q ˉ = 3.47 .
Figure 7 shows the fit without an EOS. The QPOs give M = 1.92 0 − 0.012 + 0.014 M ⊙ M = 1.920^{+0.014}_{-0.012}\,M_\odot M = 1.92 0 − 0.012 + 0.014 M ⊙ , j = 0.16 2 − 0.011 + 0.023 j = 0.162^{+0.023}_{-0.011} j = 0.16 2 − 0.011 + 0.023 and Q ˉ = 2. 2 − 0.8 + 1.1 \bar{Q} = 2.2^{+1.1}_{-0.8} Q ˉ = 2. 2 − 0.8 + 1.1 , and with the known spin frequency I = 2.3 0 − 0.17 + 0.35 × 10 45 g c m 2 I = 2.30^{+0.35}_{-0.17} \times10^{45}\ \mathrm{g\,cm^2} I = 2.3 0 − 0.17 + 0.35 × 1 0 45 g c m 2 . The mass agrees with the EOS fits. The spin and the quadrupole are strongly correlated (correlation coefficient 0.94): the posterior is a narrow ridge along which a larger quadrupole cancels part of a larger Lense–Thirring precession [equation (11) ]. The one-dimensional ranges of j j j and I I I contain the stiff EOSs, but the neutron-star sequences lie above the ridge. At j = 0.171 j = 0.171 j = 0.171 and 0.178 0.178 0.178 the QPOs prefer Q ˉ = 2.7 ± 0.15 \bar{Q} = 2.7 \pm0.15 Q ˉ = 2.7 ± 0.15 and 3.0 ± 0.13 3.0 \pm0.13 3.0 ± 0.13 , while FSUGarnet and IOPB-I have Q ˉ = 3.28 \bar{Q} = 3.28 Q ˉ = 3.28 and 3.47 3.47 3.47 . A neutron star with the required moment of inertia is thus 15–20 per cent more oblate than the QPOs prefer. For G3 and WFF1 the offset is larger. The EOS fits absorb this tension in the 5–6 per cent scatter of the low-frequency QPO, which the fit without an EOS does not need (0.7 per cent).
Figure 7. Posterior of the fit to 4U 1728−34 without an EOS: mass, spin, reduced quadrupole and the moment of inertia implied by the spin frequency of 363 Hz. Shaded regions contain 68 per cent and 95 per cent of the posterior; the titles give medians and 68 per cent intervals. Lines are the neutron-star sequences of the four EOSs at 363 Hz.
The quadrupole matters for this inference. At the radius where ν ϕ = 1098.4 \nu_\phi= 1098.4 ν ϕ = 1098.4 Hz, the preferred IOPB-I star has ν n o d = 19.31 \nu_{\mathrm{nod}} = 19.31 ν nod = 19.31 Hz. A Kerr spacetime with the same mass and the same j j j would give 24.6 Hz. To produce 19.31 Hz the Kerr spacetime needs j = 0.138 j = 0.138 j = 0.138 , 22 per cent less, and with the known spin frequency this corresponds to I = 1.97 × 10 45 g c m 2 I = 1.97 \times10^{45}\ \mathrm{g\,cm^2} I = 1.97 × 1 0 45 g c m 2 instead of 2.53. The Kerr value is close to the I = 2 × 10 45 g c m 2 I = 2 \times10^{45}\ \mathrm{g\,cm^2} I = 2 × 1 0 45 g c m 2 that Anand et al. (2024) obtained from these data in the Kerr metric. Their estimate that a self-consistent metric would change I I I by about 5 per cent is, however, much smaller than the difference found here. An analysis in the Kerr metric would therefore select a softer EOS, close to G3. The size of this bias is set by the ratio of equation (11) near the ISCO, that is, by j j j and Q ˉ \bar{Q} Q ˉ of the star, and not by details of the data.
Z sources
We first fit the kHz QPOs alone. In GX 5−1 and GX 340+0 they extend to much lower frequencies than in the atoll sources (ν L ≈ 160 \nu_L \approx160 ν L ≈ 160 –200 Hz at ν U ≈ 480 \nu_U \approx480 ν U ≈ 480 –540 Hz). In the RPM a ratio ν L / ν U ≈ 0.33 \nu_L/\nu_U \approx0.33 ν L / ν U ≈ 0.33 places the emission near x ≈ 10 x \approx10 x ≈ 10 , and the observed ν U \nu_U ν U then requires a high mass. The kHz-only fits give M = 1.97 M = 1.97 M = 1.97 –2.14 M ⊙ M_\odot M ⊙ with a scatter of 3–7 per cent and do not constrain the spin. The quadrupole nodal frequency is far below the HBO: for IOPB-I, 2 ν n o d = 6 2\nu_{\mathrm{nod}} = 6 2 ν nod = 6 , 12 and 15 Hz at ν L = 200 \nu_L = 200 ν L = 200 , 400 and 500 Hz, against observed HBOs of 18–48 Hz (dotted lines in Figure 8 (b) and 8(d)).
Figure 8. Z sources with the best-fitting RPM curves for the four EOSs (best-fit mass in M⊙ _\odot ⊙ and spin frequency in the legends). (a), (b) GX 5−1, data of Jonker et al. (2002); (c), (d) GX 340+0, data of Jonker et al. (2000). Panels (a) and (c) show the kHz QPOs, panels (b) and (d) the HBOs. The dotted grey line is 2 ν n o d 2\nu_{\rm nod} 2 ν nod predicted by the fit of IOPB-I to the kHz QPOs alone.
We then add the HBO. To raise 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod to the observed HBO frequencies the fits push the spin to the upper edge of the prior, ν s = 731 \nu_s = 731 ν s = 731 –744 Hz for all EOSs, and the mass to within 0.01–0.05 M ⊙ M_\odot M ⊙ of the maximum mass at that spin (Table 3 ). The HBO is then reproduced, but the kHz QPOs are not. At ν U ≳ 700 \nu_U \gtrsim700 ν U ≳ 700 Hz the observed ν L \nu_L ν L lies up to 100 Hz above every curve, and the kHz scatter grows to 5–9 per cent. IOPB-I and FSUGarnet, the stars with the largest moments of inertia, come closest.
The reason is the saturation of the nodal frequency shown in Figure 3 (b). Raising the spin increases the Lense–Thirring term, but the quadrupole term grows faster. For EOSs with M max ≈ 2.0 M_{\max} \approx2.0 M m a x ≈ 2.0 –2.2 M ⊙ M_\odot M ⊙ no spin below 750 Hz produces nodal frequencies as high as the HBOs at the radii fixed by the kHz QPOs. The fit without an EOS confirms this. It requires M = 2.49 ± 0.05 M ⊙ M = 2.49 \pm0.05\,M_\odot M = 2.49 ± 0.05 M ⊙ and j = 0.27 ± 0.02 j = 0.27 \pm0.02 j = 0.27 ± 0.02 for GX 5−1 and M = 2.4 7 − 0.06 + 0.06 M ⊙ M = 2.47^{+0.06}_{-0.06}\,M_\odot M = 2.4 7 − 0.06 + 0.06 M ⊙ and j = 0.34 ± 0.03 j = 0.34 \pm0.03 j = 0.34 ± 0.03 for GX 340+0, with a kHz scatter of only 3 per cent. These masses exceed the maximum mass of every EOS in our set and the masses of the heaviest known neutron stars (Fonseca et al. (2021)). Within the RPM and the slow-rotation approximation, the HBOs of these Z sources can be twice the nodal precession frequency at the radius of the kHz QPOs only if the star is heavier than any neutron star measured so far.
Discussion
What the QPOs measure
In the RPM the kHz QPOs constrain mainly a combination of M M M and j j j . For a known spin frequency the EOS fixes j j j , so the kHz QPOs give an EOS-dependent mass. The low-frequency QPO adds the nodal precession, which depends on j j j and q q q at the same order. Through j = 2 π ν s I / M 2 j = 2\pi\nu_{\rm s}I/M^2 j = 2 π ν s I / M 2 and Q ˉ ( I ) \bar{Q}(I) Q ˉ ( I ) it measures the moment of inertia, and the quadrupole has to be that of the same star. For 4U 1728−34 the EOS-consistent analysis favours I ≃ 2.4 I \simeq2.4 I ≃ 2.4 –2.5 × 10 45 g c m 2 2.5 \times10^{45}\ {\rm g\,cm^2} 2.5 × 1 0 45 g c m 2 at M ≃ 1.92 M ⊙ M \simeq1.92 M_\odot M ≃ 1.92 M ⊙ , that is, the stiffer EOSs with R 1.4 ≃ 13 R_{1.4} \simeq13 R 1.4 ≃ 13 km. A Kerr spacetime with the same nodal frequency gives an I I I lower by 22 per cent. This bias applies to any EOS inference that uses the RPM nodal frequency in a Kerr spacetime or with q q q treated as independent of j j j . Anand et al. (2024) analysed the same data with the RPM in the Kerr metric; if the RPM holds, moments of inertia derived from such analyses should be revised upward by a similar amount.
Limits of the model
All of these statements hold within the RPM, and the model is under strain. The fits need intrinsic scatters below 0.6 per cent (kHz QPOs of 4U 1728−34), 4–6 per cent (4U 1636−53 and the low-frequency QPO of 4U 1728−34) and 5–9 per cent (Z sources). Only the first is comparable to the measurement errors. For 4U 1636−53 the masses lie at the maximum masses of all four EOSs. For the Z sources no EOS-consistent star reproduces the HBOs as nodal precession. The preference for the stiffer EOSs in 4U 1728−34 is therefore conditional on the RPM identification of the low-frequency QPO, and it should be read as a demonstration of how much the quadrupole matters rather than as an EOS measurement.
Beyond test particles
The RPM treats the QPOs as test-particle frequencies. § 5 shows that orbits of finite amplitude precess at the epicyclic rates and that a hot spot on such an orbit produces peaks at these frequencies. It also shows two points that the model leaves open: the hot spot puts its low-frequency power at ν n o d \nu_{\mathrm{nod}} ν nod , not at 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod , and the precession peaks are much weaker than the orbital peak. Finite amplitudes shift the frequencies by about 1 per cent for the orbit of Figure 4 (radial amplitude 3 per cent, inclination 8 ∘ 8^\circ 8 ∘ ), and the shift grows as the square of the amplitude; for somewhat larger orbits it becomes comparable to the scatters above. Pressure, viscosity, radiation drag and magnetic stresses in the flow shift the frequencies further. With ν L F = ν n o d \nu_{\mathrm{LF}}=\nu_{\mathrm{nod}} ν LF = ν nod instead of 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod , the low-frequency QPOs of all sources would be even harder to reach.
Slow rotation and data
The HT metric is correct to second order in the spin. For q = − j 2 q=-j^2 q = − j 2 its epicyclic ratios differ from those of the exact Kerr metric at the same orbital frequency by at most 6 × 10 − 5 6\times10^{-5} 6 × 1 0 − 5 at j = 0.1 j=0.1 j = 0.1 and 3 × 10 − 4 3\times10^{-4} 3 × 1 0 − 4 at j = 0.2 j=0.2 j = 0.2 for x ≥ 6 x\geq6 x ≥ 6 . Taken as an estimate of the neglected terms, this shifts 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod by about 0.1 Hz for 4U 1728−34, far below the differences between the EOSs and the 1.7 Hz error of the low-frequency QPO. Comparisons with numerical models (Berti et al. 2005) show that the HT approximation gives the ISCO and the orbital frequencies to about 1 per cent for the spin frequencies of most known millisecond pulsars. The higher multipoles and the stellar shape lose accuracy as the spin approaches the mass-shedding limit, so the fits of 4U 1636−53 at 581 Hz and especially of the Z sources at 700–750 Hz should be repeated with numerically exact rotating models (Stergioulas & Friedman 1995; Morsink & Stella 1999). The saturation of the nodal frequency that excludes the HBOs is driven by the quadrupole growing faster than ν s \nu_s ν s . Higher-order rotation terms make the stars more oblate, so we do not expect them to reverse this conclusion; we will test this with numerically exact models in Paper II.
The QPO frequencies were taken from published tables. They come from two instruments, RXTE and AstroSat, and from different analysis methods. For GX 5−1 the kHz selections overlap, so neighbouring sets are not strictly independent, and the uncertainties of the fitted parameters of this source are somewhat underestimated.
Conclusions
We have studied the relativistic precession model in the Hartle–Thorne spacetimes of slowly rotating neutron stars built from four EOSs. For each star the mass, spin frequency and EOS fix j j j , q q q and the inner edge of the disc, and the orbital, epicyclic and precession frequencies follow from the full exterior metric. We have followed test particles on eccentric, tilted orbits around such a star and confronted the model with the QPOs of two atoll and two Z sources. Our main conclusions are as follows.
The quadrupole of the star controls the nodal precession. At M ≃ 1.9 M ⊙ M\simeq1.9\,M_\odot M ≃ 1.9 M ⊙ the reduced quadrupole of the four EOSs is Q ˉ = 2.1 \bar{Q}=2.1 Q ˉ = 2.1 –3.6, so the quadrupole is 2–4 times that of a Kerr black hole with the same j j j . Its retrograde contribution lowers the nodal frequency near the ISCO by 7–25 per cent relative to the Kerr value at 363 Hz, and it grows faster with the spin than the Lense–Thirring term. At fixed orbital frequency the nodal frequency therefore has a maximum as a function of the spin, 2 ν n o d ≈ 32 2\nu_{\mathrm{nod}}\approx32 2 ν nod ≈ 32 –34 Hz at ν U = 900 \nu_U=900 ν U = 900 Hz for 1.9 M ⊙ 1.9\,M_\odot 1.9 M ⊙ stars of the stiff EOSs, and a fast spin alone cannot produce a high nodal frequency.
Orbits of finite amplitude precess at the epicyclic rates, and these rates appear in the flux. An eccentric, tilted geodesic around a 1.92 M ⊙ 1.92\,M_\odot 1.92 M ⊙ star spinning at 363 Hz advances its pericentre at ν ϕ − ν r \nu_\phi-\nu_r ν ϕ − ν r and turns its orbital plane at ν ϕ − ν θ \nu_\phi-\nu_\theta ν ϕ − ν θ , within 1 per cent of the epicyclic values for a radial amplitude of 3 per cent and an inclination of 8 ∘ 8^\circ 8 ∘ . The Doppler-boosted flux of a hot spot on this orbit has its peaks at ν n o d \nu_{\mathrm{nod}} ν nod , ν r \nu_r ν r , ν r \nu_r ν r , ν θ \nu_\theta ν θ and ν ϕ \nu_\phi ν ϕ . Its low-frequency power, however, lies at ν n o d \nu_{\mathrm{nod}} ν nod , and the peak at 2 ν n o d 2\nu_{\mathrm{nod}} 2 ν nod , the frequency that the RPM identifies with the low-frequency QPO, is weaker by almost three orders of magnitude.
For 4U 1728−34 the low-frequency QPO measures the moment of inertia. The kHz QPOs alone give M = 1.88 M=1.88 M = 1.88 –1.92 M ⊙ 1.92\,M_\odot 1.92 M ⊙ for all four EOSs. The low-frequency QPO then selects the stars with large moments of inertia: the preferred IOPB-1 star has M = 1.922 ± 0.010 M ⊙ M=1.922\pm0.010\,M_\odot M = 1.922 ± 0.010 M ⊙ , j = 0.178 j=0.178 j = 0.178 and I = 2.5 × 10 45 g c m 2 I=2.5\times10^{45}\ {\rm g\,cm^2} I = 2.5 × 1 0 45 g c m 2 , and FSU Garnet is close. Without an EOS the QPOs give M = 1.92 0 − 0.012 + 0.014 M ⊙ M=1.920^{+0.014}_{-0.012}\,M_\odot M = 1.92 0 − 0.012 + 0.014 M ⊙ , j = 0.16 2 − 0.011 + 0.023 j=0.162^{+0.023}_{-0.011} j = 0.16 2 − 0.011 + 0.023 and Q ˉ = 2. 2 − 0.8 + 1.1 \bar{Q}=2.2^{+1.1}_{-0.8} Q ˉ = 2. 2 − 0.8 + 1.1 . Along the narrow j j j –Q ˉ \bar{Q} Q ˉ ridge allowed by the data the preferred quadrupole is 15–20 per cent smaller than that of a neutron star with the required moment of inertia, a tension that the EOS fits absorb in a 5–6 per cent scatter of the low-frequency QPO.
A Kerr spacetime biases the moment of inertia. To reproduce the same nodal frequency at the same orbital frequency, a Kerr spacetime needs a spin 22 per cent smaller than the neutron star, and therefore gives a moment of inertia 22 per cent too small (1.97 1.97 1.97 instead of 2.53 × 10 45 g c m 2 2.53\times10^{45}\ {\rm g\,cm^2} 2.53 × 1 0 45 g c m 2 for 4U 1728−34). This is much larger than the 5 per cent estimated for these data by Amand et al. (2024), and it would move an EOS inference from the stiff EOSs towards softer ones such as G3.
4U 1636−53 is described only with a large scatter. Its kHz QPOs require an intrinsic scatter of 4–5 per cent and masses of 2.02–2.15 M ⊙ \,M_\odot M ⊙ , close to the maximum masses of all four EOSs at 581 Hz, and they do not discriminate between the EOSs.
For these EOSs the HBOs of the Z sources cannot be nodal precession. For GX 5−1 and GX 340+0 no EOS-consistent star reproduces the HBOs as twice the nodal frequency at the radius of the kHz QPOs. The fits are driven to spins of 730–745 Hz and to the maximum mass and still need a kHz scatter of 5–9 per cent. A fit without an EOS requires M ≈ 2.5 M ⊙ M\approx2.5\,M_\odot M ≈ 2.5 M ⊙ , above the maximum mass of every EOS in our set.
Taken together, these results show that an EOS-consistent quadrupole has to be included whenever the nodal precession frequency is used to infer the spin or the moment of inertia of a neutron star. They also show that the RPM is under strain: it needs intrinsic scatters of several per cent, high masses, a factor of two in the nodal frequency that a single orbiting spot does not produce, and it fails for the HBOs of the Z sources. The preference for stiff EOSs in 4U 1728−34 is therefore conditional on the RPM identification of the low-frequency QPO. Three steps would sharpen these conclusions: numerically exact rotating stars for spin frequencies above about 500 Hz, which we will present in Paper II; more QPO triplets with low-frequency QPOs measured to better than 1 per cent, which would show whether the 5 per cent residuals are random or systematic; and joint fits with independent mass–radius measurements of neutron stars.
Acknowledgements
I thank Ranjeev Misra for introducing me to this problem and for discussions on the relativistic precession model and QPOs. I also thank H. C. Das for providing the unified equation-of-state tables of the G3, FSUGarnet and IOPB-I models (Parmar et al. 2022), publicly available at https://github.com/hcdas/Unified_eos , and N. Stergioulas for making the RNS code and its equation-of-state tables publicly available (Stergioulas & Friedman 1995).
Data availability
The QPO frequencies used in this work are taken from the published literature cited in Table 2 . The code that computes the stars, the frequencies, the orbits and their spectra, and the fits, together with the QPO data, the stellar models, the frequency tables and the fit results, is available at https://github.com/AnkitKhunt/slow-rotation-NS-QPO .
References
Rotating stars to second order in the angular velocity
We use units with G = c = 1 G=c=1 G = c = 1 and follow Hartle (1967). Throughout this appendix ν \nu ν denotes the metric function with e ν = − g t t e^\nu=-g_{tt} e ν = − g tt of the nonrotating star, P 2 = P 2 ( cos θ ) = 1 2 ( 3 cos 2 θ − 1 ) P_2=P_2(\cos\theta)=\frac{1}{2}(3\cos^2\theta-1) P 2 = P 2 ( cos θ ) = 2 1 ( 3 cos 2 θ − 1 ) , and a prime denotes d / d r d/dr d / d r .
Metric, fluid and ordering
The metric of a rigidly rotating star is written in the form of equation (1). The rotation is slow in the sense that Ω 2 R 3 / M ≪ 1 \Omega^2 R^3/M\ll1 Ω 2 R 3 / M ≪ 1 ; we keep terms to second order in Ω \Omega Ω . The frame-dragging angular velocity ω ( r ) \omega(r) ω ( r ) is of first order, and the functions h 0 h_0 h 0 , h 2 h_2 h 2 , m 0 m_0 m 0 , m 2 m_2 m 2 and k 2 k_2 k 2 are of second order. Terms of first order would change sign with Ω \Omega Ω and cannot appear in the diagonal components, and the θ \theta θ dependence at second order is that of P 0 P_0 P 0 and P 2 P_2 P 2 only, because the centrifugal potential is Ω 2 r 2 sin 2 θ = 2 3 r 2 ( 1 − P 2 ) \Omega^2 r^2\sin^2\theta=\frac{2}{3}r^2(1-P_2) Ω 2 r 2 sin 2 θ = 3 2 r 2 ( 1 − P 2 ) .
The fluid is a perfect fluid with energy density ε \varepsilon ε and pressure P P P , T μ ν = ( ϵ + P ) u μ u ν + P g μ ν T^{\mu\nu} = (\epsilon+ P)u^\mu u^\nu+ P g^{\mu\nu} T μν = ( ϵ + P ) u μ u ν + P g μν , rotating rigidly, u μ = u t ( 1 , 0 , 0 , Ω ) u^\mu= u^t(1,0,0,\Omega) u μ = u t ( 1 , 0 , 0 , Ω ) . To second order,
g t t + 2 Ω g t ϕ + Ω 2 g ϕ ϕ = − e ν ( 1 + 2 h 0 + 2 h 2 P 2 ) + r 2 sin 2 θ ϖ 2 . (14) g_{tt} + 2\Omega g_{t\phi} + \Omega^2 g_{\phi\phi} = -e^\nu(1 + 2h_0 + 2h_2P_2) + r^2\sin^2\theta\,\varpi^2. \tag*{(14)} g tt + 2Ω g tϕ + Ω 2 g ϕϕ = − e ν ( 1 + 2 h 0 + 2 h 2 P 2 ) + r 2 sin 2 θ ϖ 2 . (14)
with ϖ = Ω − ω \varpi= \Omega- \omega ϖ = Ω − ω the angular velocity of the fluid relative to the local inertial frames. The normalisation u ⋅ u = − 1 u \cdot u = -1 u ⋅ u = − 1 then gives
u t = e − ν / 2 [ 1 − h 0 − h 2 P 2 + 1 2 e − ν r 2 sin 2 θ ϖ 2 ] . (15) u^t = e^{-\nu/2}\left[1-h_0-h_2P_2+\frac{1}{2}e^{-\nu}r^2\sin^2\theta\,\varpi^2\right]. \tag*{(15)} u t = e − ν /2 [ 1 − h 0 − h 2 P 2 + 2 1 e − ν r 2 sin 2 θ ϖ 2 ] . (15)
In the frame comoving with the fluid the pressure at a given coordinate point changes by ( ϵ + P ) p ∗ (\epsilon+P)p^* ( ϵ + P ) p ∗ with p ∗ = p 0 ∗ + p 2 ∗ P 2 p^* = p^*_0 + p^*_2P_2 p ∗ = p 0 ∗ + p 2 ∗ P 2 , and the energy density by ( d ϵ / d P ) ( ϵ + P ) p ∗ (d\epsilon/dP)(\epsilon+P)p^* ( d ϵ / d P ) ( ϵ + P ) p ∗ , since the EOS is barotropic.
Nonrotating background
At zeroth order the field equations reduce to the TOV equations,
m ′ = 4 π r 2 ϵ , P ′ = − ( ϵ + P ) m + 4 π r 3 P r ( r − 2 m ) , m' = 4\pi r^2\epsilon,\qquad P' = -(\epsilon+P)\frac{m+4\pi r^3P}{r(r-2m)}, m ′ = 4 π r 2 ϵ , P ′ = − ( ϵ + P ) r ( r − 2 m ) m + 4 π r 3 P ,
ν ′ = 2 ( m + 4 π r 3 P ) r ( r − 2 m ) , e λ = ( 1 − 2 m r ) − 1 . (16) \nu' = \frac{2(m+4\pi r^3P)}{r(r-2m)},\qquad e^\lambda= \left(1-\frac{2m}{r}\right)^{-1}. \tag*{(16)} ν ′ = r ( r − 2 m ) 2 ( m + 4 π r 3 P ) , e λ = ( 1 − r 2 m ) − 1 . (16)
They are integrated from a regular centre, m = 0 m=0 m = 0 , P = P c P=P_c P = P c , to the surface r = R r=R r = R where P = 0 P=0 P = 0 . The constant in ν \nu ν is fixed by e ν ( R ) = 1 − 2 M T O V / R e^{\nu(R)}=1-2M_{\mathrm{TOV}}/R e ν ( R ) = 1 − 2 M TOV / R , which matches the interior to the exterior Schwarzschild metric. Two combinations will be used repeatedly. Adding the t t tt tt and r r rr r r field equations gives
ν ′ + λ ′ = 8 π r ( ϵ + P ) e λ , (17) \nu' + \lambda' = 8\pi r(\epsilon+P)e^\lambda, \tag*{(17)} ν ′ + λ ′ = 8 π r ( ϵ + P ) e λ , (17)
and near the centre, where m ≃ 4 3 π ϵ c r 3 m \simeq\frac{4}{3}\pi\epsilon_c r^3 m ≃ 3 4 π ϵ c r 3 ,
ν ′ ≃ 8 π 3 ( ϵ c + 3 P c ) r . (18) \nu' \simeq\frac{8\pi}{3}(\epsilon_c+3P_c)r. \tag*{(18)} ν ′ ≃ 3 8 π ( ϵ c + 3 P c ) r . (18)
Hydrostatic equilibrium of the rotating fluid
For a stationary flow along the Killing vector ξ = ∂ t + Ω ∂ ϕ \xi=\partial_t+\Omega\partial_\phi ξ = ∂ t + Ω ∂ ϕ , u μ = u t ξ μ u^\mu=u^t\xi^\mu u μ = u t ξ μ with u t = ( − ξ ⋅ ξ ) − 1 / 2 u^t=(-\xi\cdot\xi)^{-1/2} u t = ( − ξ ⋅ ξ ) − 1/2 , and the four-acceleration is a μ = − ∂ μ ln u t a_\mu=-\partial_\mu\ln u^t a μ = − ∂ μ ln u t . The relativistic Euler equation, ( ϵ + P ) a μ = − ∂ μ P (\epsilon+P)a_\mu=-\partial_\mu P ( ϵ + P ) a μ = − ∂ μ P for the spatial components, becomes
∂ i P ϵ + P = ∂ i ln u t , (19) \frac{\partial_i P}{\epsilon+P}=\partial_i\ln u^t, \tag*{(19)} ϵ + P ∂ i P = ∂ i ln u t , (19)
so that for a barotropic fluid
∫ P d P ϵ + P − ln u t = c o n s t . (20) \int^P\frac{dP}{\epsilon+P}-\ln u^t=\mathrm{const}. \tag*{(20)} ∫ P ϵ + P d P − ln u t = const . (20)
At zeroth order equation (20) is the TOV equation for P ′ P' P ′ written as P ′ / ( ϵ + P ) = − ν ′ / 2 P'/(\epsilon+P)=-\nu'/2 P ′ / ( ϵ + P ) = − ν ′ /2 . At second order the first term changes by δ P / ( ϵ + P ) = p ∗ \delta P/(\epsilon+P)=p^* δ P / ( ϵ + P ) = p ∗ , and the second by δ ln u t \delta\ln u^t δ ln u t from equation (15) . With sin 2 θ = 2 3 ( 1 − P 2 ) \sin^2\theta=\frac{2}{3}(1-P_2) sin 2 θ = 3 2 ( 1 − P 2 ) the P 0 P_0 P 0 and P 2 P_2 P 2 parts of equation (20) give
p 0 ∗ = − h 0 + 1 3 e − ν r 2 ϖ 2 + h 0 c , (21) p^*_0=-h_0+\frac{1}{3}e^{-\nu}r^2\varpi^2+h_{0c}, \tag*{(21)} p 0 ∗ = − h 0 + 3 1 e − ν r 2 ϖ 2 + h 0 c , (21)
p 2 ∗ = − h 2 − 1 3 e − ν r 2 ϖ 2 , (22) p^*_2=-h_2-\frac{1}{3}e^{-\nu}r^2\varpi^2, \tag*{(22)} p 2 ∗ = − h 2 − 3 1 e − ν r 2 ϖ 2 , (22)
where h 0 c h_{0c} h 0 c is a constant. These two relations replace the perturbed Euler equations in what follows.
Frame dragging
At first order in Ω \Omega Ω the only nontrivial field equation is the t ϕ t\phi tϕ component. Its source is T t ϕ = ( ϵ + P ) u t u ϕ = ( ϵ + P ) e − ν r 2 sin 2 θ ϖ T^\phi_{\ t}=(\epsilon+P)u^t u_\phi=(\epsilon+P)e^{-\nu}r^2\sin^2\theta\,\varpi T t ϕ = ( ϵ + P ) u t u ϕ = ( ϵ + P ) e − ν r 2 sin 2 θ ϖ . After the angular dependence is separated, regularity at the centre and asymptotic flatness leave only the ℓ = 1 \ell=1 ℓ = 1 part, so that ω = ω ( r ) \omega=\omega(r) ω = ω ( r ) , and the equation is (Hartle (1967))
1 r 4 d d r ( r 4 ȷ ˉ d ϖ d r ) + 4 r d ȷ ˉ d r ϖ = 0 , ȷ ˉ = e − ( ν + λ ) / 2 . (23) \frac{1}{r^4}\frac{d}{dr}\left(r^4\bar{\jmath}\frac{d\varpi}{dr}\right)+\frac{4}{r}\frac{d\bar{\jmath}}{dr}\varpi=0,\qquad\bar{\jmath}=e^{-(\nu+\lambda)/2}. \tag*{(23)} r 4 1 d r d ( r 4 ˉ d r d ϖ ) + r 4 d r d ˉ ϖ = 0 , ˉ = e − ( ν + λ ) /2 . (23)
By equation (17) ,
d ȷ ˉ d r = − 1 2 ( ν ′ + λ ′ ) ȷ ˉ = − 4 π r ( ϵ + P ) e λ ȷ ˉ , (24) \frac{d\bar{\jmath}}{dr}=-\frac{1}{2}(\nu'+\lambda')\bar{\jmath}=-4\pi r(\epsilon+P)e^\lambda\bar{\jmath}, \tag*{(24)} d r d ˉ = − 2 1 ( ν ′ + λ ′ ) ˉ = − 4 π r ( ϵ + P ) e λ ˉ , (24)
so the second term of equation (23) equals − 16 π ( ϵ + P ) e ( λ − ν ) / 2 ϖ -16\pi(\epsilon+P)e^{(\lambda-\nu)/2}\varpi − 16 π ( ϵ + P ) e ( λ − ν ) /2 ϖ . It is proportional to the angular momentum density of the fluid and vanishes outside the star.
Near the centre, ȷ ˉ ≃ ȷ ˉ c \bar{\jmath}\simeq\bar{\jmath}_c ˉ ≃ ˉ c and e λ ≃ 1 e^\lambda\simeq1 e λ ≃ 1 , and equation (23) reads ϖ ′ ′ + 4 ϖ ′ / r = 16 π ( ϵ c + P c ) ϖ \varpi''+4\varpi'/r=16\pi(\epsilon_c+P_c)\varpi ϖ ′′ + 4 ϖ ′ / r = 16 π ( ϵ c + P c ) ϖ . The regular solution is
ϖ = ϖ c [ 1 + 8 π 5 ( ϵ c + P c ) r 2 + O ( r 4 ) ] , (25) \varpi=\varpi_c\left[1+\frac{8\pi}{5}(\epsilon_c+P_c)r^2+O(r^4)\right], \tag*{(25)} ϖ = ϖ c [ 1 + 5 8 π ( ϵ c + P c ) r 2 + O ( r 4 ) ] , (25)
with ϖ c \varpi_c ϖ c arbitrary. Because equation (23) is linear and homogeneous, we integrate it with ϖ c = 1 \varpi_c=1 ϖ c = 1 and rescale the solution at the end.
Outside the star ȷ ˉ = 1 \bar{\jmath}=1 ˉ = 1 and ( r 4 ϖ ′ ) ′ = 0 (r^4\varpi')'=0 ( r 4 ϖ ′ ) ′ = 0 , so ϖ = C 1 − C 2 / r 3 \varpi=C_1-C_2/r^3 ϖ = C 1 − C 2 / r 3 . Asymptotic flatness requires ω → 0 \omega\to0 ω → 0 , hence C 1 = Ω C_1=\Omega C 1 = Ω , and comparison of g t ϕ = − ω r 2 sin 2 θ g_{t\phi}=-\omega r^2\sin^2\theta g tϕ = − ω r 2 sin 2 θ with the gravitomagnetic field of a body of angular momentum J J J , g t ϕ → − 2 J sin 2 θ / r g_{t\phi}\to-2J\sin^2\theta/r g tϕ → − 2 J sin 2 θ / r , gives C 2 = 2 J C_2=2J C 2 = 2 J :
ω = 2 J r 3 , ϖ = Ω − 2 J r 3 ( r > R ) . (26) \omega=\frac{2J}{r^3},\qquad\varpi=\Omega-\frac{2J}{r^3}\qquad(r>R). \tag*{(26)} ω = r 3 2 J , ϖ = Ω − r 3 2 J ( r > R ) . (26)
Since ϵ \epsilon ϵ and P P P are finite at the surface, ϖ \varpi ϖ and ϖ ′ \varpi' ϖ ′ are continuous there. The exterior solution then gives
J = 1 6 R 4 ϖ ′ ( R ) , Ω = ϖ ( R ) + 1 3 R ϖ ′ ( R ) , I = J Ω . (27) J=\frac{1}{6}R^4\varpi'(R),\qquad\Omega=\varpi(R)+\frac{1}{3}R\varpi'(R),\qquad I=\frac{J}{\Omega}. \tag*{(27)} J = 6 1 R 4 ϖ ′ ( R ) , Ω = ϖ ( R ) + 3 1 R ϖ ′ ( R ) , I = Ω J . (27)
The second relation fixes the scale of the interior solution for a given Ω \Omega Ω . Integrating equation (23) from the centre to R R R gives the equivalent volume integral
J = 8 π 3 ∫ 0 R r 4 ( ϵ + P ) e ( λ − ν ) / 2 ϖ d r , (28) J=\frac{8\pi}{3}\int_0^R r^4(\epsilon+P)e^{(\lambda-\nu)/2}\varpi\,dr, \tag*{(28)} J = 3 8 π ∫ 0 R r 4 ( ϵ + P ) e ( λ − ν ) /2 ϖ d r , (28)
which reduces to the Newtonian J = Ω 8 π 3 ∫ ρ r 4 d r J=\Omega\frac{8\pi}{3}\int\rho r^4dr J = Ω 3 8 π ∫ ρ r 4 d r for weak fields. For the models of this paper equations (27) and (28) agree to better than 10 − 6 10^{-6} 1 0 − 6 . Since ϖ ∝ Ω \varpi\propto\Omega ϖ ∝ Ω at fixed central density, J ∝ Ω J\propto\Omega J ∝ Ω and I I I is independent of Ω \Omega Ω at this order.
The P 0 P_0 P 0 parts of the t t tt tt and r r rr r r field equations at second order involve m 0 m_0 m 0 , h 0 h_0 h 0 and p 0 ∗ p^*_0 p 0 ∗ . Eliminating h 0 h_0 h 0 with equation (21) , Hartle (1967) obtained
d m 0 d r = 4 π r 2 d ϵ d P ( ϵ + P ) p 0 ∗ + 1 12 r 4 ϖ 2 − 1 3 r 3 d ȷ ˉ 2 d r ϖ 2 , (29) \frac{dm_0}{dr}=4\pi r^2\frac{d\epsilon}{dP}(\epsilon+P)p^*_0+\frac{1}{12}r^4\varpi^2-\frac{1}{3}r^3\frac{d\bar{\jmath}^2}{dr}\varpi^2, \tag*{(29)} d r d m 0 = 4 π r 2 d P d ϵ ( ϵ + P ) p 0 ∗ + 12 1 r 4 ϖ 2 − 3 1 r 3 d r d ˉ 2 ϖ 2 , (29)
d p 0 ∗ d r = − m 0 1 + 8 π r 2 P ( r − 2 m ) 2 − 4 π ( ϵ + P ) r 2 r − 2 m p 0 ∗ + r 4 ȷ ˉ 2 ϖ 2 12 ( r − 2 m ) + 1 3 d d r [ r 3 ȷ ˉ 2 ϖ 2 r − 2 m ] . (30) \frac{dp^*_0}{dr}=-m_0\frac{1+8\pi r^2P}{(r-2m)^2}-\frac{4\pi(\epsilon+P)r^2}{r-2m}p^*_0+\frac{r^4\bar{\jmath}^2\varpi^2}{12(r-2m)}+\frac{1}{3}\frac{d}{dr}\left[\frac{r^3\bar{\jmath}^2\varpi^2}{r-2m}\right]. \tag*{(30)} d r d p 0 ∗ = − m 0 ( r − 2 m ) 2 1 + 8 π r 2 P − r − 2 m 4 π ( ϵ + P ) r 2 p 0 ∗ + 12 ( r − 2 m ) r 4 ˉ 2 ϖ 2 + 3 1 d r d [ r − 2 m r 3 ˉ 2 ϖ 2 ] . (30)
The first is the t t tt tt equation: the change of the enclosed mass has a contribution from the change of the density and two from the rotational energy of fluid and field. The terms in ω ˉ \bar{\omega} ω ˉ act as sources, so m 0 m_0 m 0 and p 0 ∗ p_0^* p 0 ∗ are of order Ω 2 \Omega^2 Ω 2 . For a sequence at fixed central density, m 0 = p 0 ∗ = 0 m_0=p_0^*=0 m 0 = p 0 ∗ = 0 at r = 0 r=0 r = 0 .
Outside the star j ˉ = 1 \bar{j}=1 j ˉ = 1 and ω ˉ ′ = 6 J / r 4 \bar{\omega}'=6J/r^4 ω ˉ ′ = 6 J / r 4 , and equation (29) becomes m 0 ′ = 3 J 2 / r 4 m_0'=3J^2/r^4 m 0 ′ = 3 J 2 / r 4 , with solution
m 0 = δ M − J 2 r 3 , m_0=\delta M-\frac{J^2}{r^3}, m 0 = δ M − r 3 J 2 ,
h 0 = − δ M r − 2 M + J 2 r 3 ( r − 2 M ) ( r > R ) , (31) h_0=-\frac{\delta M}{r-2M}+\frac{J^2}{r^3(r-2M)}\qquad(r>R), \tag*{(31)} h 0 = − r − 2 M δ M + r 3 ( r − 2 M ) J 2 ( r > R ) , (31)
where the form of h 0 h_0 h 0 follows from the r r rr r r equation. At large r r r , − g t t = ( 1 − 2 M / r ) ( 1 + 2 h 0 ) = 1 − 2 ( M + δ M ) / r + O ( r − 3 ) -g_{tt}=(1-2M/r)(1+2h_0)=1-2(M+\delta M)/r+O(r^{-3}) − g tt = ( 1 − 2 M / r ) ( 1 + 2 h 0 ) = 1 − 2 ( M + δ M ) / r + O ( r − 3 ) and g r r = 1 + 2 ( M + δ M ) / r + … g_{rr}=1+2(M+\delta M)/r+\ldots g r r = 1 + 2 ( M + δ M ) / r + … , so δ M \delta M δ M is the change of the gravitational mass. Continuity of m 0 m_0 m 0 at R R R gives
δ M = m 0 ( R ) + J 2 R 3 , (32) \delta M=m_0(R)+\frac{J^2}{R^3}, \tag*{(32)} δ M = m 0 ( R ) + R 3 J 2 , (32)
and the total mass is M = M T O V + δ M M=M_{\mathrm{TOV}}+\delta M M = M TOV + δ M . In the exterior metric of equation (4) the constant δ M \delta M δ M has been absorbed into M M M , so that m 0 = − J 2 / r 3 m_0=-J^2/r^3 m 0 = − J 2 / r 3 and h 0 = J 2 / [ r 3 ( r − 2 M ) ] h_0=J^2/[r^3(r-2M)] h 0 = J 2 / [ r 3 ( r − 2 M )] there. The constant h 0 c h_{0c} h 0 c of equation (21) follows from the continuity of h 0 h_0 h 0 , but it is not needed for any quantity used in this paper.
The P 2 P_2 P 2 parts of the field equations involve h 2 h_2 h 2 , m 2 m_2 m 2 and k 2 k_2 k 2 . The θ θ − ϕ ϕ \theta\theta-\phi\phi θ θ − ϕϕ component is algebraic and expresses m 2 m_2 m 2 through h 2 h_2 h 2 and the rotation; outside the star it gives m 2 = ( r − 2 M ) ( − h 2 + 6 J 2 / r 4 ) m_2=(r-2M)(-h_2+6J^2/r^4) m 2 = ( r − 2 M ) ( − h 2 + 6 J 2 / r 4 ) . With v 2 ≡ h 2 + k 2 v_2\equiv h_2+k_2 v 2 ≡ h 2 + k 2 , the remaining components give two first-order equations (Hartle 1967),
d v 2 d r = − ν ′ h 2 + ( 1 r + ν ′ 2 ) [ 1 6 r 4 ω ˉ ′ 2 − 1 3 r 3 d j ˉ 2 d r ω ˉ 2 ] , (33) \frac{dv_2}{dr}=-\nu' h_2+\left(\frac{1}{r}+\frac{\nu'}{2}\right)\left[\frac{1}{6}r^4\bar{\omega}'^2-\frac{1}{3}r^3\frac{d\bar{j}^2}{dr}\bar{\omega}^2\right], \tag*{(33)} d r d v 2 = − ν ′ h 2 + ( r 1 + 2 ν ′ ) [ 6 1 r 4 ω ˉ ′2 − 3 1 r 3 d r d j ˉ 2 ω ˉ 2 ] , (33)
d h 2 d r = − ν ′ h 2 + r h 2 ( r − 2 m ) ν ′ [ 8 π ( ϵ + P ) − 4 m r 3 ] − 4 v 2 r ( r − 2 m ) ν ′ + 1 6 [ r ν ′ 2 − 1 ( r − 2 m ) ν ′ ] r 3 ω ˉ ′ 2 − 1 3 [ r ν ′ 2 + 1 ( r − 2 m ) ν ′ ] r 2 d j ˉ 2 d r ω ˉ 2 . (34) \begin{aligned}
\frac{dh_2}{dr}={}&-\nu' h_2+\frac{r h_2}{(r-2m)\nu'}\left[8\pi(\epsilon+P)-\frac{4m}{r^3}\right]\\
&-\frac{4v_2}{r(r-2m)\nu'}+\frac{1}{6}\left[\frac{r\nu'}{2}-\frac{1}{(r-2m)\nu'}\right]r^3\bar{\omega}'^2\\
&-\frac{1}{3}\left[\frac{r\nu'}{2}+\frac{1}{(r-2m)\nu'}\right]r^2\frac{d\bar{j}^2}{dr}\bar{\omega}^2. \tag*{(34)}
\end{aligned} d r d h 2 = − ν ′ h 2 + ( r − 2 m ) ν ′ r h 2 [ 8 π ( ϵ + P ) − r 3 4 m ] − r ( r − 2 m ) ν ′ 4 v 2 + 6 1 [ 2 r ν ′ − ( r − 2 m ) ν ′ 1 ] r 3 ω ˉ ′2 − 3 1 [ 2 r ν ′ + ( r − 2 m ) ν ′ 1 ] r 2 d r d j ˉ 2 ω ˉ 2 . (34)
At the centre, regularity requires h 2 ≃ A r 2 h_2\simeq Ar^2 h 2 ≃ A r 2 and v 2 ≃ B r 4 v_2\simeq Br^4 v 2 ≃ B r 4 . The rotation terms are of higher order in r r r , and equation (33) with equation (18) gives
B = − 2 π 3 ( ϵ c + 3 P c ) A . (35) B=-\frac{2\pi}{3}(\epsilon_c+3P_c)A. \tag*{(35)} B = − 3 2 π ( ϵ c + 3 P c ) A . (35)
The general regular solution is therefore h 2 = h 2 P + A h 2 H h_2=h_2^{\mathrm{P}}+Ah_2^{\mathrm{H}} h 2 = h 2 P + A h 2 H , v 2 = v 2 P + A v 2 H v_2=v_2^{\mathrm{P}}+Av_2^{\mathrm{H}} v 2 = v 2 P + A v 2 H , where the particular solution ( h 2 P , v 2 P ) (h_2^{\mathrm{P}},v_2^{\mathrm{P}}) ( h 2 P , v 2 P ) solves equations (33) –(34) with zero initial values, and the homogeneous solution ( h 2 H , v 2 H ) (h_2^{\mathrm{H}},v_2^{\mathrm{H}}) ( h 2 H , v 2 H ) solves them without the ω ˉ \bar{\omega} ω ˉ terms, starting from h 2 H = r 2 h_2^{\mathrm{H}}=r^2 h 2 H = r 2 , v 2 H = − 2 π 3 ( ϵ c + 3 P c ) r 4 v_2^{\mathrm{H}}=-\frac{2\pi}{3}(\epsilon_c+3P_c)r^4 v 2 H = − 3 2 π ( ϵ c + 3 P c ) r 4 .
The exterior solution (Appendix B) has one free constant K K K . Continuity of h 2 h_2 h 2 and v 2 v_2 v 2 at r = R r=R r = R ,
h 2 P ( R ) + A h 2 H ( R ) = J 2 M R 3 ( 1 + M R ) + K Q 2 2 ( ξ R ) , h_2^{\mathrm{P}}(R)+Ah_2^{\mathrm{H}}(R)=\frac{J^2}{MR^3}\left(1+\frac{M}{R}\right)+KQ_2^2(\xi_R), h 2 P ( R ) + A h 2 H ( R ) = M R 3 J 2 ( 1 + R M ) + K Q 2 2 ( ξ R ) ,
v 2 P ( R ) + A v 2 H ( R ) = − J 2 R 4 + K 2 M R ( R − 2 M ) Q 2 2 ( ξ R ) , (36) v_2^{\mathrm{P}}(R)+Av_2^{\mathrm{H}}(R)=-\frac{J^2}{R^4}+K\frac{2M}{\sqrt{R(R-2M)}}Q_2^2(\xi_R), \tag*{(36)} v 2 P ( R ) + A v 2 H ( R ) = − R 4 J 2 + K R ( R − 2 M ) 2 M Q 2 2 ( ξ R ) , (36)
with ξ R = R / M − 1 \xi_R=R/M-1 ξ R = R / M − 1 , is a linear system for A A A and K K K . The quadrupole moment is then given by equation (2) (Appendix B2). Because all source terms are quadratic in ω ˉ ∝ Ω \bar{\omega}\propto\Omega ω ˉ ∝ Ω , h 2 h_2 h 2 , v 2 v_2 v 2 , K K K and Q Q Q are proportional to Ω 2 \Omega^2 Ω 2 at fixed central density.
Surface of the rotating star
A fluid element on the surface of the nonrotating star is displaced to r = R + ξ ( θ ) r=R+\xi(\theta) r = R + ξ ( θ ) , where the total pressure vanishes: P ( R + ξ ) + ( ϵ + P ) p ∗ = 0 P(R+\xi)+(\epsilon+P)p^*=0 P ( R + ξ ) + ( ϵ + P ) p ∗ = 0 . To first order in ξ \xi ξ ,
ξ = − ( ϵ + P ) p ∗ P ′ = p ∗ r ( r − 2 m ) m + 4 π r 3 P , (37) \xi=-\frac{(\epsilon+P)p^*}{P'}=p^*\frac{r(r-2m)}{m+4\pi r^3P}, \tag*{(37)} ξ = − P ′ ( ϵ + P ) p ∗ = p ∗ m + 4 π r 3 P r ( r − 2 m ) , (37)
which at the surface, where P = 0 P=0 P = 0 and m = M m=M m = M , is ξ = p ∗ ( R ) R ( R − 2 M ) / M \xi=p^*(R)R(R-2M)/M ξ = p ∗ ( R ) R ( R − 2 M ) / M . With p ∗ = p 0 ∗ + p 2 ∗ P 2 p^*=p_0^*+p_2^*P_2 p ∗ = p 0 ∗ + p 2 ∗ P 2 , the surface is r ( θ ) = R + ξ 0 + ξ 2 P 2 ( cos θ ) r(\theta)=R+\xi_0+\xi_2P_2(\cos\theta) r ( θ ) = R + ξ 0 + ξ 2 P 2 ( cos θ ) , with ξ 0 \xi_0 ξ 0 from p 0 ∗ ( R ) p_0^*(R) p 0 ∗ ( R ) and ξ 2 \xi_2 ξ 2 from equation (22) at r = R r=R r = R . At the equator and the pole (P 2 = − 1 2 P_2=-\frac{1}{2} P 2 = − 2 1 and 1 1 1 ),
R e q = R + ξ 0 − 1 2 ξ 2 , R p o l = R + ξ 0 + ξ 2 . (38) R_{\mathrm{eq}}=R+\xi_0-\frac{1}{2}\xi_2,\qquad R_{\mathrm{pol}}=R+\xi_0+\xi_2. \tag*{(38)} R eq = R + ξ 0 − 2 1 ξ 2 , R pol = R + ξ 0 + ξ 2 . (38)
The circumference of the equator is 2 π g ϕ ϕ 2\pi\sqrt{g_{\phi\phi}} 2 π g ϕϕ at θ = π / 2 \theta=\pi/2 θ = π /2 , and from equation (1)
R e q , c i r c = R e q [ 1 + k 2 ( R ) P 2 ( 0 ) ] = R e q [ 1 − 1 2 k 2 ( R ) ] . (39) R_{\mathrm{eq,circ}}=R_{\mathrm{eq}}\left[1+k_2(R)P_2(0)\right]=R_{\mathrm{eq}}\left[1-\frac{1}{2}k_2(R)\right]. \tag*{(39)} R eq , circ = R eq [ 1 + k 2 ( R ) P 2 ( 0 ) ] = R eq [ 1 − 2 1 k 2 ( R ) ] . (39)
The coordinate radius R e q R_{\mathrm{eq}} R eq is the inner edge of a disc that touches the star, in the same coordinate as the exterior metric used for the orbits.
Stars of given mass and spin
For a target total mass M M M and spin frequency ν s \nu_s ν s , we solve M T O V ( ϵ c ) + δ M ( ϵ c , ν s ) = M M_{\mathrm{TOV}}(\epsilon_c)+\delta M(\epsilon_c,\nu_s)=M M TOV ( ϵ c ) + δ M ( ϵ c , ν s ) = M on the stable branch by a secant iteration in ln ϵ c \ln\epsilon_c ln ϵ c . For the fits we compute one model at each of 64 central densities, from 0.9 M ⊙ 0.9\,M_\odot 0.9 M ⊙ to the maximum mass, at a reference spin frequency ν r e f \nu_{\mathrm{ref}} ν ref , and use the scalings of this appendix at fixed ϵ c \epsilon_c ϵ c ,
J ∝ ν s , δ M , Q , ξ 0 , ξ 2 ∝ ν s 2 , (40) J\propto\nu_s,\qquad\delta M,\ Q,\ \xi_0,\ \xi_2\propto\nu_s^2, \tag*{(40)} J ∝ ν s , δ M , Q , ξ 0 , ξ 2 ∝ ν s 2 , (40)
to obtain all quantities at any ν s \nu_s ν s , after which M ( ϵ c , ν s ) M(\epsilon_c,\nu_s) M ( ϵ c , ν s ) is inverted by interpolation. These scalings are exact in the HT approximation.
Exterior spacetime and multipole moments
Vacuum solutions
Outside the star ϵ = P = 0 \epsilon=P=0 ϵ = P = 0 , m = M m=M m = M and ν ′ = 2 M / [ r ( r − 2 M ) ] \nu'=2M/[r(r-2M)] ν ′ = 2 M / [ r ( r − 2 M )] . The frame dragging and the monopole deformation were solved above. For the quadrupole sector the sources in equations (33) –(34) reduce to the ω ˉ ′ 2 = 36 J 2 / r 8 \bar{\omega}'^2=36J^2/r^8 ω ˉ ′2 = 36 J 2 / r 8 terms. A particular solution is
h 2 P = J 2 M r 3 ( 1 + M r ) , v 2 P = − J 2 r 4 , (41) h_2^{\mathrm{P}}=\frac{J^2}{Mr^3}\left(1+\frac{M}{r}\right),\qquad v_2^{\mathrm{P}}=-\frac{J^2}{r^4}, \tag*{(41)} h 2 P = M r 3 J 2 ( 1 + r M ) , v 2 P = − r 4 J 2 , (41)
as direct substitution shows. For the homogeneous part we eliminate v 2 v_2 v 2 : equation (34) gives v 2 v_2 v 2 in terms of h 2 h_2 h 2 and h 2 ′ h_2' h 2 ′ , and inserting it into equation (33) gives a second-order equation for h 2 h_2 h 2 . In the variable ξ = r / M − 1 \xi=r/M-1 ξ = r / M − 1 it is
( 1 − ξ 2 ) d 2 h 2 d ξ 2 − 2 ξ d h 2 d ξ + [ 6 − 4 1 − ξ 2 ] h 2 = 0. (42) (1-\xi^2)\frac{d^2h_2}{d\xi^2}-2\xi\frac{dh_2}{d\xi}+\left[6-\frac{4}{1-\xi^2}\right]h_2=0. \tag*{(42)} ( 1 − ξ 2 ) d ξ 2 d 2 h 2 − 2 ξ d ξ d h 2 + [ 6 − 1 − ξ 2 4 ] h 2 = 0. (42)
the associated Legendre equation of degree 2 and order 2. Its solutions are P 2 2 ( ξ ) ∝ ξ 2 − 1 P_2^2(\xi) \propto\xi^2-1 P 2 2 ( ξ ) ∝ ξ 2 − 1 , which grows at large r r r and is excluded by asymptotic flatness, and the function of the second kind Q 2 2 ( ξ ) Q_2^2(\xi) Q 2 2 ( ξ ) . Integrating v 2 ′ = − ν ′ h 2 v_2'=-\nu'h_2 v 2 ′ = − ν ′ h 2 then gives the accompanying v 2 v_2 v 2 . The result is equation (4) ,
h 2 = J 2 M r 3 ( 1 + M r ) + K Q 2 2 ( ξ ) , h_2=\frac{J^2}{Mr^3}\left(1+\frac{M}{r}\right)+KQ_2^2(\xi), h 2 = M r 3 J 2 ( 1 + r M ) + K Q 2 2 ( ξ ) ,
v 2 = − J 2 r 4 + K 2 M r ( r − 2 M ) Q 2 1 ( ξ ) , (43) v_2=-\frac{J^2}{r^4}+K\frac{2M}{\sqrt{r(r-2M)}}Q_2^1(\xi), \tag*{(43)} v 2 = − r 4 J 2 + K r ( r − 2 M ) 2 M Q 2 1 ( ξ ) , (43)
with
Q 1 2 ( ξ ) = ξ 2 − 1 [ 3 ξ 2 − 2 ξ 2 − 1 − 3 2 ξ ln ξ + 1 ξ − 1 ] , Q_1^2(\xi)=\sqrt{\xi^2-1}\left[\frac{3\xi^2-2}{\xi^2-1}-\frac{3}{2}\xi\ln\frac{\xi+1}{\xi-1}\right], Q 1 2 ( ξ ) = ξ 2 − 1 [ ξ 2 − 1 3 ξ 2 − 2 − 2 3 ξ ln ξ − 1 ξ + 1 ] ,
Q 2 2 ( ξ ) = 3 2 ( ξ 2 − 1 ) ln ξ + 1 ξ − 1 − 3 ξ 3 − 5 ξ ξ 2 − 1 , (44) Q_2^2(\xi)=\frac{3}{2}(\xi^2-1)\ln\frac{\xi+1}{\xi-1}-\frac{3\xi^3-5\xi}{\xi^2-1}, \tag*{(44)} Q 2 2 ( ξ ) = 2 3 ( ξ 2 − 1 ) ln ξ − 1 ξ + 1 − ξ 2 − 1 3 ξ 3 − 5 ξ , (44)
for ξ > 1 \xi>1 ξ > 1 , in the sign convention of Hartle & Thorne (1968), in which both are positive. For large r r r , with s = M / r s=M/r s = M / r ,
Q 2 2 = 8 5 s 3 + 24 5 s 4 + O ( s 5 ) , Q 2 1 = 2 5 s 3 + 6 5 s 4 + O ( s 5 ) . (45) Q_2^2=\frac{8}{5}s^3+\frac{24}{5}s^4+O(s^5),\qquad Q_2^1=\frac{2}{5}s^3+\frac{6}{5}s^4+O(s^5). \tag*{(45)} Q 2 2 = 5 8 s 3 + 5 24 s 4 + O ( s 5 ) , Q 2 1 = 5 2 s 3 + 5 6 s 4 + O ( s 5 ) . (45)
We have checked by symbolic computation that these functions satisfy equations (A20)–(A21) in vacuum, and that the change of variable turns them into equation (42) . Because the closed forms lose precision through cancellation at large ξ \xi ξ , we evaluate them by the convergent series in ξ − 2 \xi^{-2} ξ − 2 for ξ ≥ 3 \xi\geq3 ξ ≥ 3 .
Multipole moments
By equation (45) , at large r r r
h 2 ≃ ( J 2 M + 8 5 K M 3 ) 1 r 3 ≡ Q H T r 3 , (46) h_2\simeq\left(\frac{J^2}{M}+\frac{8}{5}KM^3\right)\frac{1}{r^3}\equiv\frac{Q_{\mathrm{HT}}}{r^3}, \tag*{(46)} h 2 ≃ ( M J 2 + 5 8 K M 3 ) r 3 1 ≡ r 3 Q HT , (46)
and
− g t t = ( 1 − 2 M r ) ( 1 + 2 h 2 P 2 ) + ⋯ = 1 − 2 M r + 2 Q H T r 3 P 2 + O ( r − 4 ) . (47) \begin{aligned}
-g_{tt}=\left(1-\frac{2M}{r}\right)(1+2h_2P_2)+\cdots\\
&=1-\frac{2M}{r}+\frac{2Q_{\mathrm{HT}}}{r^3}P_2+O(r^{-4}). \tag*{(47)}
\end{aligned} − g tt = ( 1 − r 2 M ) ( 1 + 2 h 2 P 2 ) + ⋯ = 1 − r 2 M + r 3 2 Q HT P 2 + O ( r − 4 ) . (47)
For a Newtonian body with quadrupole moment Q Q Q the potential is Φ = − M / r − Q P 2 / r 3 \Phi=-M/r-QP_2/r^3 Φ = − M / r − Q P 2 / r 3 , and − g t t ≃ 1 + 2 Φ -g_{tt}\simeq1+2\Phi − g tt ≃ 1 + 2Φ . An oblate body has more mass near the equator, where P 2 = − 1 / 2 P_2=-1/2 P 2 = − 1/2 , so its potential is deeper there and Q < 0 Q<0 Q < 0 ; this is the Geroch–Hansen sign convention, in which the Kerr metric has Q = − J 2 / M Q=-J^2/M Q = − J 2 / M . Comparison gives Q = − Q H T Q=-Q_{\mathrm{HT}} Q = − Q HT , which is equation (2) . The spin J J J was identified in the same way from g t ϕ g_{t\phi} g tϕ . The dimensionless moments are j = J / M 2 j=J/M^2 j = J / M 2 and q = Q / M 3 q=Q/M^3 q = Q / M 3 .
Kerr limit
For a Kerr black hole of mass M M M and specific angular momentum a = J / M a=J/M a = J / M , the quadrupole is Q = − M a 2 = − J 2 / M Q=-Ma^2=-J^2/M Q = − M a 2 = − J 2 / M , so Q H T = J 2 / M Q_{\mathrm{HT}}=J^2/M Q HT = J 2 / M and K = 0 K=0 K = 0 . The HT exterior with K = 0 K=0 K = 0 must then be the Kerr metric expanded to second order in a a a . Writing the Boyer–Lindquist coordinates ( r B L , θ B L ) (r_{\mathrm{BL}},\theta_{\mathrm{BL}}) ( r BL , θ BL ) as
r B L = r − a 2 2 r [ ( 1 + 2 M r ) ( 1 − M r ) − cos 2 θ ( 1 − 2 M r ) ( 1 + 3 M r ) ] , r_{\mathrm{BL}}=r-\frac{a^2}{2r}\left[\left(1+\frac{2M}{r}\right)\left(1-\frac{M}{r}\right)-\cos^2\theta\left(1-\frac{2M}{r}\right)\left(1+\frac{3M}{r}\right)\right], r BL = r − 2 r a 2 [ ( 1 + r 2 M ) ( 1 − r M ) − cos 2 θ ( 1 − r 2 M ) ( 1 + r 3 M ) ] ,
θ B L = θ − a 2 2 r 2 ( 1 + 2 M r ) sin θ cos θ , (48) \theta_{\mathrm{BL}}=\theta-\frac{a^2}{2r^2}\left(1+\frac{2M}{r}\right)\sin\theta\cos\theta, \tag*{(48)} θ BL = θ − 2 r 2 a 2 ( 1 + r 2 M ) sin θ cos θ , (48)
Hartle & Thorne (1968), we have verified symbolically that every component of the Kerr metric, including the induced g r θ g_{r\theta} g r θ , agrees with equation (49) to O ( a 2 ) O(a^2) O ( a 2 ) . This is the analytic basis of the statement that the HT frequencies with q = − j 2 q=-j^2 q = − j 2 reproduce the Kerr frequencies to O ( j 2 ) O(j^2) O ( j 2 ) .
Metric components used for the orbits
Collecting the exterior functions, and absorbing δ M \delta M δ M into M M M ,
g t t = − ( 1 − 2 M r ) [ 1 + 2 h 2 P 2 + 2 J 2 r 4 2 cos 2 θ − 1 1 − 2 M / r ] , g_{tt}=-\left(1-\frac{2M}{r}\right)\left[1+2h_2P_2+\frac{2J^2}{r^4}\frac{2\cos^2\theta-1}{1-2M/r}\right], g tt = − ( 1 − r 2 M ) [ 1 + 2 h 2 P 2 + r 4 2 J 2 1 − 2 M / r 2 cos 2 θ − 1 ] ,
g t ϕ = − 2 J r sin 2 θ , g θ θ = r 2 ( 1 + 2 k 2 P 2 ) , g_{t\phi}=-\frac{2J}{r}\sin^2\theta,\qquad g_{\theta\theta}=r^2(1+2k_2P_2), g tϕ = − r 2 J sin 2 θ , g θ θ = r 2 ( 1 + 2 k 2 P 2 ) ,
g ϕ ϕ = r 2 sin 2 θ ( 1 + 2 k 2 P 2 ) , g_{\phi\phi}=r^2\sin^2\theta(1+2k_2P_2), g ϕϕ = r 2 sin 2 θ ( 1 + 2 k 2 P 2 ) ,
g r r = ( 1 − 2 M r ) − 1 [ 1 − 2 ( h 2 − 6 J 2 r 4 ) P 2 − 2 J 2 r 4 ( 1 − 2 M / r ) ] . (49) g_{rr}=\left(1-\frac{2M}{r}\right)^{-1}\left[1-2\left(h_2-\frac{6J^2}{r^4}\right)P_2-\frac{2J^2}{r^4(1-2M/r)}\right]. \tag*{(49)} g r r = ( 1 − r 2 M ) − 1 [ 1 − 2 ( h 2 − r 4 6 J 2 ) P 2 − r 4 ( 1 − 2 M / r ) 2 J 2 ] . (49)
with k 2 = v 2 − h 2 k_2=v_2-h_2 k 2 = v 2 − h 2 . The J 2 J^2 J 2 terms of g t t g_{tt} g tt collect h 0 h_0 h 0 and the ω 2 r 2 sin 2 θ \omega^2r^2\sin^2\theta ω 2 r 2 sin 2 θ term of ( d ϕ − ω d t ) 2 (d\phi-\omega\,dt)^2 ( d ϕ − ω d t ) 2 ; the J 2 J^2 J 2 terms of g r r g_{rr} g r r collect m 0 m_0 m 0 and m 2 m_2 m 2 . The metric depends on ( M , J , Q ) (M,J,Q) ( M , J , Q ) only.
Circular orbits and epicyclic frequencies
Conserved quantities and the effective potential
Consider a timelike geodesic in a stationary, axisymmetric metric that is symmetric under reflection in the equatorial plane, with nonzero components g t t g_{tt} g tt , g t ϕ g_{t\phi} g tϕ , g ϕ ϕ g_{\phi\phi} g ϕϕ , g r r g_{rr} g r r and g θ θ g_{\theta\theta} g θ θ depending on ( r , θ ) (r,\theta) ( r , θ ) . With L = 1 2 g μ ν x ˙ μ x ˙ ν \mathcal{L}=\frac{1}{2}g_{\mu\nu}\dot{x}^{\mu}\dot{x}^{\nu} L = 2 1 g μν x ˙ μ x ˙ ν and a dot denoting d / d τ d/d\tau d / d τ , the coordinates t t t and ϕ \phi ϕ are cyclic, and the specific energy and angular momentum are
E = − u t = − ( g t t t ˙ + g t ϕ ϕ ˙ ) , L = u ϕ = g t ϕ t ˙ + g ϕ ϕ ϕ ˙ (50) E=-u_t=-(g_{tt}\dot{t}+g_{t\phi}\dot{\phi}),\qquad L=u_\phi=g_{t\phi}\dot{t}+g_{\phi\phi}\dot{\phi} \tag*{(50)} E = − u t = − ( g tt t ˙ + g tϕ ϕ ˙ ) , L = u ϕ = g tϕ t ˙ + g ϕϕ ϕ ˙ (50)
are conserved. Inverting the ( t , ϕ ) (t,\phi) ( t , ϕ ) block, with D = g t t g ϕ ϕ − g t ϕ 2 < 0 D=g_{tt}g_{\phi\phi}-g_{t\phi}^2<0 D = g tt g ϕϕ − g tϕ 2 < 0 ,
t ˙ = g ϕ ϕ E + g t ϕ L − D , ϕ ˙ = − g t ϕ E + g t t L − D , (51) \dot{t}=\frac{g_{\phi\phi}E+g_{t\phi}L}{-D},\qquad\dot{\phi}=-\frac{g_{t\phi}E+g_{tt}L}{-D}, \tag*{(51)} t ˙ = − D g ϕϕ E + g tϕ L , ϕ ˙ = − − D g tϕ E + g tt L , (51)
and the ( t , ϕ ) (t,\phi) ( t , ϕ ) part of g μ ν u μ u ν g^{\mu\nu}u_\mu u_\nu g μν u μ u ν is
H ( r , θ ; E , L ) = g ϕ ϕ E 2 + 2 g t ϕ E L + g t t L 2 D . (52) \mathcal{H}(r,\theta;E,L)=\frac{g_{\phi\phi}E^2+2g_{t\phi}EL+g_{tt}L^2}{D}. \tag*{(52)} H ( r , θ ; E , L ) = D g ϕϕ E 2 + 2 g tϕ E L + g tt L 2 . (52)
The normalisation u ⋅ u = − 1 u\cdot u=-1 u ⋅ u = − 1 becomes
g r r r ˙ 2 + g θ θ θ ˙ 2 = − 1 − H ( r , θ ; E , L ) ≡ V ( r , θ ) , (53) g_{rr}\dot{r}^2+g_{\theta\theta}\dot{\theta}^2=-1-\mathcal{H}(r,\theta;E,L)\equiv V(r,\theta), \tag*{(53)} g r r r ˙ 2 + g θ θ θ ˙ 2 = − 1 − H ( r , θ ; E , L ) ≡ V ( r , θ ) , (53)
so motion is allowed where V ≥ 0 V\geq0 V ≥ 0 . Eliminating the cyclic coordinates by a Routh transformation gives the reduced Lagrangian R = 1 2 ( g r r r ˙ 2 + g θ θ θ ˙ 2 ) − 1 2 H \mathcal{R}=\frac{1}{2}(g_{rr}\dot{r}^2+g_{\theta\theta}\dot{\theta}^2)-\frac{1}{2}\mathcal{H} R = 2 1 ( g r r r ˙ 2 + g θ θ θ ˙ 2 ) − 2 1 H , whose Euler–Lagrange equations are
d d τ ( g r r r ˙ ) = 1 2 ∂ r g r r r ˙ 2 + 1 2 ∂ r g θ θ θ ˙ 2 − 1 2 ∂ r H , \frac{d}{d\tau}(g_{rr}\dot{r})=\frac{1}{2}\partial_rg_{rr}\dot{r}^2+\frac{1}{2}\partial_rg_{\theta\theta}\dot{\theta}^2-\frac{1}{2}\partial_r\mathcal{H}, d τ d ( g r r r ˙ ) = 2 1 ∂ r g r r r ˙ 2 + 2 1 ∂ r g θ θ θ ˙ 2 − 2 1 ∂ r H ,
d d τ ( g θ θ θ ˙ ) = 1 2 ∂ θ g r r r ˙ 2 + 1 2 ∂ θ g θ θ θ ˙ 2 − 1 2 ∂ θ H . (54) \frac{d}{d\tau}(g_{\theta\theta}\dot{\theta})=\frac{1}{2}\partial_\theta g_{rr}\dot{r}^2+\frac{1}{2}\partial_\theta g_{\theta\theta}\dot{\theta}^2-\frac{1}{2}\partial_\theta\mathcal{H}. \tag*{(54)} d τ d ( g θ θ θ ˙ ) = 2 1 ∂ θ g r r r ˙ 2 + 2 1 ∂ θ g θ θ θ ˙ 2 − 2 1 ∂ θ H . (54)
Circular equatorial orbits
On the equator ∂ θ H = 0 \partial_\theta\mathcal{H}=0 ∂ θ H = 0 by reflection symmetry. A circular orbit of radius r r r has r ˙ = θ ˙ = 0 \dot{r}=\dot{\theta}=0 r ˙ = θ ˙ = 0 for all τ \tau τ , which by equation (54) requires ∂ r H = 0 \partial_r\mathcal{H}=0 ∂ r H = 0 , and by equation (53) V = 0 V=0 V = 0 . Equivalently, the radial component of the geodesic equation, Γ μ ν r u μ u ν = − 1 2 g r r ∂ r g μ ν u μ u ν = 0 \Gamma^r_{\mu\nu}u^\mu u^\nu=-\frac{1}{2}g^{rr}\partial_r g_{\mu\nu}u^\mu u^\nu=0 Γ μν r u μ u ν = − 2 1 g r r ∂ r g μν u μ u ν = 0 , with u ϕ = Ω ϕ u t u^\phi=\Omega_\phi u^t u ϕ = Ω ϕ u t , gives
∂ r g t t + 2 Ω ϕ ∂ r g t ϕ + Ω ϕ 2 ∂ r g ϕ ϕ = 0. (55) \partial_r g_{tt}+2\Omega_\phi\partial_r g_{t\phi}+\Omega_\phi^2\partial_r g_{\phi\phi}=0. \tag*{(55)} ∂ r g tt + 2 Ω ϕ ∂ r g tϕ + Ω ϕ 2 ∂ r g ϕϕ = 0. (55)
Its prograde root is equation (5) . The normalisation then gives
u t = [ − ( g t t + 2 Ω ϕ g t ϕ + Ω ϕ 2 g ϕ ϕ ) ] − 1 / 2 , (56) u^t=\left[-\left(g_{tt}+2\Omega_\phi g_{t\phi}+\Omega_\phi^2g_{\phi\phi}\right)\right]^{-1/2}, \tag*{(56)} u t = [ − ( g tt + 2 Ω ϕ g tϕ + Ω ϕ 2 g ϕϕ ) ] − 1/2 , (56)
and E = − ( g t t + g t ϕ Ω ϕ ) u t E=-(g_{tt}+g_{t\phi}\Omega_\phi)u^t E = − ( g tt + g tϕ Ω ϕ ) u t , L = ( g t ϕ + g ϕ ϕ Ω ϕ ) u t L=(g_{t\phi}+g_{\phi\phi}\Omega_\phi)u^t L = ( g tϕ + g ϕϕ Ω ϕ ) u t . For the Schwarzschild metric these give Ω ϕ 2 = M / r 3 \Omega_\phi^2=M/r^3 Ω ϕ 2 = M / r 3 , and for Kerr Ω ϕ = M 1 / 2 / ( r 3 / 2 + a M 1 / 2 ) \Omega_\phi=M^{1/2}/(r^{3/2}+aM^{1/2}) Ω ϕ = M 1/2 / ( r 3/2 + a M 1/2 ) .
Small oscillations
Let r = r 0 + δ r r=r_0+\delta r r = r 0 + δ r and θ = π / 2 + δ θ \theta=\pi/2+\delta\theta θ = π /2 + δ θ with E E E and L L L those of the circular orbit at r 0 r_0 r 0 . The terms in r ˙ 2 \dot{r}^2 r ˙ 2 and θ ˙ 2 \dot{\theta}^2 θ ˙ 2 in equation (54) are of second order, ∂ r H \partial_r\mathcal{H} ∂ r H and ∂ θ H \partial_\theta\mathcal{H} ∂ θ H vanish on the circular orbit, and ∂ r ∂ θ H = 0 \partial_r\partial_\theta\mathcal{H}=0 ∂ r ∂ θ H = 0 on the equator by reflection symmetry. To first order,
g r r δ r ¨ = − 1 2 H , r r δ r , g θ θ δ θ ¨ = − 1 2 H , θ θ δ θ , (57) g_{rr}\,\delta\ddot{r}=-\frac{1}{2}\mathcal{H}_{,rr}\,\delta r,\qquad
g_{\theta\theta}\,\delta\ddot{\theta}=-\frac{1}{2}\mathcal{H}_{,\theta\theta}\,\delta\theta, \tag*{(57)} g r r δ r ¨ = − 2 1 H , r r δ r , g θ θ δ θ ¨ = − 2 1 H , θ θ δ θ , (57)
two uncoupled harmonic oscillators with proper-time angular frequencies
κ r 2 = H , r r 2 g r r , κ θ 2 = H , θ θ 2 g θ θ , (58) \kappa_r^2=\frac{\mathcal{H}_{,rr}}{2g_{rr}},\qquad
\kappa_\theta^2=\frac{\mathcal{H}_{,\theta\theta}}{2g_{\theta\theta}}, \tag*{(58)} κ r 2 = 2 g r r H , r r , κ θ 2 = 2 g θ θ H , θ θ , (58)
evaluated at ( r 0 , π / 2 ) (r_0,\pi/2) ( r 0 , π /2 ) . A distant observer measures time t t t , and d t / d τ = u t dt/d\tau=u^t d t / d τ = u t , which gives equation (7) . For the Schwarzschild metric κ r 2 / ( u t ) 2 = Ω ϕ 2 ( 1 − 6 M / r ) \kappa_r^2/(u^t)^2=\Omega_\phi^2(1-6M/r) κ r 2 / ( u t ) 2 = Ω ϕ 2 ( 1 − 6 M / r ) and κ θ / u t = Ω ϕ \kappa_\theta/u^t=\Omega_\phi κ θ / u t = Ω ϕ , so the vertical frequency equals the orbital frequency and orbits do not precess nodally. In our implementation the second derivatives are taken by fourth-order finite differences, which reproduce these Schwarzschild results to ∼ 10 − 10 \sim10^{-10} ∼ 1 0 − 10 .
Innermost stable circular orbit
A circular orbit is stable against radial perturbations if κ r 2 > 0 \kappa_r^2>0 κ r 2 > 0 . As r r r decreases, κ r 2 \kappa_r^2 κ r 2 changes sign at the ISCO, where also d E / d r = d L / d r = 0 dE/dr=dL/dr=0 d E / d r = d L / d r = 0 . We take the ISCO as the outermost zero of κ r 2 \kappa_r^2 κ r 2 and the inner edge of the disc as r i n = max ( R e q , r I S C O ) r_{\rm in}=\max(R_{\rm eq},r_{\rm ISCO}) r in = max ( R eq , r ISCO ) .
Weak-field expansions
To interpret the results we expand the frequencies for large x = r / M x=r/M x = r / M , keeping for each multipole only its leading term. At this order the contributions of M M M , J J J and Q Q Q add linearly. For the mass and the spin we use the Kerr metric, for which in Boyer–Lindquist coordinates (Okazaki et al. 1987)
Ω r 2 Ω ϕ 2 = 1 − 6 x + 8 j x 3 / 2 − 3 j 2 x 2 , Ω θ 2 Ω ϕ 2 = 1 − 4 j x 3 / 2 + 3 j 2 x 2 , (59) \frac{\Omega_r^2}{\Omega_\phi^2}=1-\frac{6}{x}+\frac{8j}{x^{3/2}}-\frac{3j^2}{x^2},\qquad
\frac{\Omega_\theta^2}{\Omega_\phi^2}=1-\frac{4j}{x^{3/2}}+\frac{3j^2}{x^2}, \tag*{(59)} Ω ϕ 2 Ω r 2 = 1 − x 6 + x 3/2 8 j − x 2 3 j 2 , Ω ϕ 2 Ω θ 2 = 1 − x 3/2 4 j + x 2 3 j 2 , (59)
where Ω r , θ = κ r , θ / u t \Omega_{r,\theta}=\kappa_{r,\theta}/u^t Ω r , θ = κ r , θ / u t . The terms linear in j j j come from the gravitomagnetic potential g t ϕ = − 2 J sin 2 θ / r g_{t\phi}=-2J\sin^2\theta/r g tϕ = − 2 J sin 2 θ / r and hold for any spinning body; the j 2 j^2 j 2 terms are those of the Kerr quadrupole q = − j 2 q=-j^2 q = − j 2 .
For the quadrupole, a Newtonian body with potential Φ = − M / r − Q P 2 ( cos θ ) / r 3 \Phi=-M/r-QP_2(\cos\theta)/r^3 Φ = − M / r − Q P 2 ( cos θ ) / r 3 has, on the equator,
Ω ϕ 2 = 1 r ∂ Φ ∂ r = M r 3 − 3 Q 2 r 5 , \Omega_\phi^2=\frac{1}{r}\frac{\partial\Phi}{\partial r}=\frac{M}{r^3}-\frac{3Q}{2r^5}, Ω ϕ 2 = r 1 ∂ r ∂ Φ = r 3 M − 2 r 5 3 Q ,
κ r 2 = 1 r d d r ( r 4 Ω ϕ 2 ) = M r 3 + 3 Q 2 r 5 , \kappa_r^2=\frac{1}{r}\frac{d}{dr}\left(r^4\Omega_\phi^2\right)=\frac{M}{r^3}+\frac{3Q}{2r^5}, κ r 2 = r 1 d r d ( r 4 Ω ϕ 2 ) = r 3 M + 2 r 5 3 Q ,
κ θ 2 = 1 r ∂ Φ ∂ r + 1 r 2 ∂ 2 Φ ∂ θ 2 = Ω ϕ 2 − 3 Q r 5 , (60) \kappa_\theta^2=\frac{1}{r}\frac{\partial\Phi}{\partial r}+\frac{1}{r^2}\frac{\partial^2\Phi}{\partial\theta^2}=\Omega_\phi^2-\frac{3Q}{r^5}, \tag*{(60)} κ θ 2 = r 1 ∂ r ∂ Φ + r 2 1 ∂ θ 2 ∂ 2 Φ = Ω ϕ 2 − r 5 3 Q , (60)
using ∂ θ 2 P 2 = 3 \partial_\theta^2P_2=3 ∂ θ 2 P 2 = 3 at θ = π / 2 \theta=\pi/2 θ = π /2 . To first order in q = Q / M 3 q=Q/M^3 q = Q / M 3 ,
κ r 2 Ω ϕ 2 = 1 + 3 q x 2 , κ θ 2 Ω ϕ 2 = 1 − 3 q x 2 . (61) \frac{\kappa_r^2}{\Omega_\phi^2}=1+\frac{3q}{x^2},\qquad
\frac{\kappa_\theta^2}{\Omega_\phi^2}=1-\frac{3q}{x^2}. \tag*{(61)} Ω ϕ 2 κ r 2 = 1 + x 2 3 q , Ω ϕ 2 κ θ 2 = 1 − x 2 3 q . (61)
For q = − j 2 q=-j^2 q = − j 2 these are exactly the j 2 j^2 j 2 terms of the Kerr expressions, so the two coefficients continue into equation (9) , valid for independent j j j and q q q at this order. Taking square roots gives the precession frequencies,
ν p e r ≃ ν ϕ ( 3 x + 9 2 x 2 − 4 j x 3 / 2 − 3 q 2 x 2 + ⋯ ) , (62) \nu_{\rm per}\simeq\nu_\phi\left(\frac{3}{x}+\frac{9}{2x^2}-\frac{4j}{x^{3/2}}-\frac{3q}{2x^2}+\cdots\right), \tag*{(62)} ν per ≃ ν ϕ ( x 3 + 2 x 2 9 − x 3/2 4 j − 2 x 2 3 q + ⋯ ) , (62)
ν n o d ≃ ν ϕ ( 2 j x 3 / 2 + 3 q 2 x 2 + ⋯ ) . (63) \nu_{\rm nod}\simeq\nu_\phi\left(\frac{2j}{x^{3/2}}+\frac{3q}{2x^2}+\cdots\right). \tag*{(63)} ν nod ≃ ν ϕ ( x 3/2 2 j + 2 x 2 3 q + ⋯ ) . (63)
With ν ϕ ≃ ( 2 π M ) − 1 x − 3 / 2 \nu_\phi\simeq(2\pi M)^{-1}x^{-3/2} ν ϕ ≃ ( 2 π M ) − 1 x − 3/2 at leading order, the spin term of ν n o d \nu_{\rm nod} ν nod is j / ( π M x 3 ) = J / ( π r 3 ) = ω / 2 π j/(\pi Mx^3)=J/(\pi r^3)=\omega/2\pi j / ( π M x 3 ) = J / ( π r 3 ) = ω /2 π , the Lense–Thirring frequency of equation (10) . The quadrupole term is the classical precession of an orbit around an oblate body, retrograde for q < 0 q<0 q < 0 . Their ratio, with q = − Q ˉ j 2 q=-\bar{Q}j^2 q = − Q ˉ j 2 , is equation (11) . The expansions are used only for interpretation; all fits use the exact frequencies of Appendix C. Near the ISCO they are accurate to about 10 per cent.
Geodesics, hot spot and light curve
Geodesic equations and initial data
We integrate x ˙ μ = u μ \dot{x}^\mu=u^\mu x ˙ μ = u μ and u ˙ μ = − Γ α β μ u α u β \dot{u}^\mu=-\Gamma^\mu_{\alpha\beta}u^\alpha u^\beta u ˙ μ = − Γ α β μ u α u β in the metric of equation (49) , with Christoffel symbols computed from the analytic metric by fourth-order finite differences. The orbit starts at ( r 0 , π / 2 , 0 ) (r_0,\pi/2,0) ( r 0 , π /2 , 0 ) with coordinate velocities
d ϕ d t = Ω ϕ ( r 0 ) cos ι , d θ d t = Ω ϕ ( r 0 ) sin ι , \frac{d\phi}{dt}=\Omega_\phi(r_0)\cos\iota,\qquad
\frac{d\theta}{dt}=\Omega_\phi(r_0)\sin\iota, d t d ϕ = Ω ϕ ( r 0 ) cos ι , d t d θ = Ω ϕ ( r 0 ) sin ι ,
d r d t = e Ω ϕ ( r 0 ) r 0 , (64) \frac{dr}{dt}=e\,\Omega_\phi(r_0)r_0, \tag*{(64)} d t d r = e Ω ϕ ( r 0 ) r 0 , (64)
where ι \iota ι is the tilt of the orbital plane and e e e is the radial kick, and u t u^t u t follows from u ⋅ u = − 1 u\cdot u=-1 u ⋅ u = − 1 . For e = ι = 0 e=\iota=0 e = ι = 0 this is the circular orbit of Appendix C; for small e e e and ι \iota ι the orbit oscillates about r 0 r_0 r 0 and θ = π / 2 \theta=\pi/2 θ = π /2 with the epicyclic frequencies at r 0 r_0 r 0 . The conserved E E E and L z L_z L z and the norm u ⋅ u u\cdot u u ⋅ u are monitored along the orbit (§ 5 ).
Measuring the precession
From ( r , θ , ϕ ) (r,\theta,\phi) ( r , θ , ϕ ) we form flat-space Cartesian coordinates x x x and velocities v = d x / d t v=dx/dt v = d x / d t , and the orbital angular momentum direction l ∝ x × v \boldsymbol{l}\propto\boldsymbol{x}\times\boldsymbol{v} l ∝ x × v . The ascending node lies along z ^ × l \hat{z}\times\boldsymbol{l} z ^ × l , so its longitude is Ω N = arctan ( l x , − l y ) \Omega_N=\arctan(l_x,-l_y) Ω N = arctan ( l x , − l y ) , and ν n o d = Ω N / 2 π \nu_{\rm nod}=\Omega_N/2\pi ν nod = Ω N /2 π . The pericentres are the local minima of r ( t ) r(t) r ( t ) ; the longitude of the pericentre advances at the rate 2 π ( ν ϕ − ν r ) = 2 π ν p e r 2\pi(\nu_\phi-\nu_r)=2\pi\nu_{\rm per} 2 π ( ν ϕ − ν r ) = 2 π ν per . Both rates are obtained by linear fits to the angles as functions of time.
Redshift and flux of a hot spot
We treat the emitter as a point source on the orbit that emits isotropically in its rest frame, and the photons as moving on straight lines in the direction n ^ = ( sin i , 0 , cos i ) \hat{\boldsymbol{n}}=(\sin i,0,\cos i) n ^ = ( sin i , 0 , cos i ) of a distant observer at inclination i i i . This neglects light bending and the dragging of photons, which change the amplitudes of the modulation but not its frequencies. A photon received at infinity with unit energy has k t = − 1 k_t=-1 k t = − 1 and spatial components k j = n j k_j=n_j k j = n j in this approximation. The energy measured by the emitter is − k μ u μ = u t − n j u j = u t ( 1 − n ^ ⋅ v ) -k_\mu u^\mu=u^t-n_j u^j=u^t(1-\hat{\boldsymbol{n}}\cdot\boldsymbol{v}) − k μ u μ = u t − n j u j = u t ( 1 − n ^ ⋅ v ) , so the ratio of observed to emitted photon energy is
g = 1 u t ( 1 − n ^ ⋅ v ) . (65) g=\frac{1}{u^t(1-\hat{\boldsymbol{n}}\cdot\boldsymbol{v})}. \tag*{(65)} g = u t ( 1 − n ^ ⋅ v ) 1 . (65)
The factor 1 / u t 1/u^t 1/ u t contains the gravitational and transverse Doppler shifts, and ( 1 − n ^ ⋅ v ) − 1 (1-\hat{\boldsymbol{n}}\cdot\boldsymbol{v})^{-1} ( 1 − n ^ ⋅ v ) − 1 the longitudinal Doppler shift. Because I ν / ν 3 I_\nu/\nu^3 I ν / ν 3 is invariant along rays, the bolometric intensity scales as g 4 g^4 g 4 , and so does the bolometric flux of a compact source, F ∝ g 4 F\propto g^4 F ∝ g 4 , as for a relativistically moving source in flat space. A photon emitted at coordinate time t t t from position x \boldsymbol{x} x arrives at
t o b s = t − n ^ ⋅ x (66) t_{\rm obs}=t-\hat{\boldsymbol{n}}\cdot\boldsymbol{x} \tag*{(66)} t obs = t − n ^ ⋅ x (66)
up to a constant (Roemer delay). We resample F ( t o b s ) F(t_{\rm obs}) F ( t obs ) on a uniform grid and compute its power spectrum with a Hann window.
Which frequencies appear
The origin of the peaks in Figure 5 follows from an expansion in the small quantities e e e and ι \iota ι . For the tilted orbit the vertical displacement is z ≃ r ι sin ( 2 π ν θ t ) z\simeq r\iota\sin(2\pi\nu_\theta t) z ≃ r ι sin ( 2 π ν θ t ) , because the angle between the particle and the precessing node advances at 2 π ( ν ϕ − ν n o d ) = 2 π ν θ 2\pi(\nu_\phi-\nu_{\rm nod})=2\pi\nu_\theta 2 π ( ν ϕ − ν nod ) = 2 π ν θ . With β = n ^ ⋅ v = v x sin i + v z cos i \beta=\hat{\boldsymbol{n}}\cdot\boldsymbol{v}=v_x\sin i+v_z\cos i β = n ^ ⋅ v = v x sin i + v z cos i ,
v x ≃ − r Ω ϕ sin ( 2 π ν ϕ t ) , v z ≃ r ι Ω θ cos ( 2 π ν θ t ) , v_x\simeq-r\Omega_\phi\sin(2\pi\nu_\phi t),\qquad v_z\simeq r\iota\Omega_\theta\cos(2\pi\nu_\theta t), v x ≃ − r Ω ϕ sin ( 2 π ν ϕ t ) , v z ≃ r ι Ω θ cos ( 2 π ν θ t ) ,
F ∝ ( u t ) − 4 ( 1 + 4 β + 10 β 2 + 20 β 3 + 35 β 4 + ⋯ ) . (67) F\propto(u^t)^{-4}\left(1+4\beta+10\beta^2+20\beta^3+35\beta^4+\cdots\right). \tag*{(67)} F ∝ ( u t ) − 4 ( 1 + 4 β + 10 β 2 + 20 β 3 + 35 β 4 + ⋯ ) . (67)
The term linear in β \beta β gives peaks at ν ϕ \nu_\phi ν ϕ , with amplitude ∝ r Ω ϕ sin i \propto r\Omega_\phi\sin i ∝ r Ω ϕ sin i , and at ν θ \nu_\theta ν θ , with amplitude ∝ ι r Ω θ cos i \propto\iota r\Omega_\theta\cos i ∝ ιr Ω θ cos i . The term 10 β 2 10\beta^2 10 β 2 contains 2 v x v z sin i cos i 2v_xv_z\sin i\cos i 2 v x v z sin i cos i , whose product of oscillations at ν ϕ \nu_\phi ν ϕ and ν θ \nu_\theta ν θ has a component at ν ϕ − ν θ = ν n o d \nu_\phi-\nu_\theta=\nu_{\rm nod} ν ϕ − ν θ = ν nod , with amplitude ∝ ι ( r Ω ϕ ) 2 sin 2 i \propto\iota(r\Omega_\phi)^2\sin2i ∝ ι ( r Ω ϕ ) 2 sin 2 i . The radial epicycle, r = r 0 [ 1 + O ( e ) cos ( 2 π ν r t ) ] r=r_0[1+O(e)\cos(2\pi\nu_r t)] r = r 0 [ 1 + O ( e ) cos ( 2 π ν r t )] , modulates u t u^t u t and ∣ v ∣ |\boldsymbol{v}| ∣ v ∣ at ν r \nu_r ν r and, multiplied by the orbital term, gives sidebands at ν ϕ ± ν r \nu_\phi\pm\nu_r ν ϕ ± ν r ; the lower sideband is ν p e r \nu_{\rm per} ν per . A component at 2 ν n o d 2\nu_{\rm nod} 2 ν nod needs v z 2 v_z^2 v z 2 , which appears first in the β 4 \beta^4 β 4 term with amplitude ∝ ι 2 ( r Ω ϕ ) 2 sin 2 2 i \propto\iota^2(r\Omega_\phi)^2\sin^2 2i ∝ ι 2 ( r Ω ϕ ) 2 sin 2 2 i . It is smaller than the ν n o d \nu_{\rm nod} ν nod term by a further factor of order ι ( r Ω ϕ ) 2 \iota(r\Omega_\phi)^2 ι ( r Ω ϕ ) 2 , which is why a single orbiting spot produces a weak peak at 2 ν n o d 2\nu_{\rm nod} 2 ν nod . The Roemer delay modulates the phase at the same frequencies and does not add new ones. Since r Ω ϕ ≃ 0.4 r\Omega_\phi\simeq0.4 r Ω ϕ ≃ 0.4 for the orbit of § 5 , the series is not rapidly convergent, and this argument predicts which frequencies appear and how their amplitudes scale, not their exact values.
Likelihood and radius marginalisation
Likelihood
A source has n n n sets of simultaneous QPOs. Set i i i contains measured frequencies d i k d_{ik} d ik with errors σ i k \sigma_{ik} σ ik in the channels k ∈ { U , L , L F } k\in\{U,L,LF\} k ∈ { U , L , L F } that were observed. For given model parameters θ \boldsymbol{\theta} θ (for example M M M and ν s \nu_s ν s with an EOS, or M M M , j j j and Q ˉ \bar{Q} Q ˉ without one), the RPM predicts frequencies m k ( x ; θ ) m_k(\boldsymbol{x};\boldsymbol{\theta}) m k ( x ; θ ) along the radius x = r / M x=r/M x = r / M (equation (12) ). The radius of each set is unknown and is marginalised with a uniform prior on [ x i n , x m a x ] [x_{\rm in},x_{\rm max}] [ x in , x max ] , with x i n = r i n / M x_{\rm in}=r_{\rm in}/M x in = r in / M and x m a x = 25 x_{\rm max}=25 x max = 25 :
L ( θ , η ) = ∏ i = 1 n 1 x m a x − x i n ∫ x i n x m a x d x ∏ k 1 2 π s i k 2 exp [ − ( d i k − m k ( x ; θ ) ) 2 2 s i k 2 ] . (68) \mathcal{L}(\boldsymbol{\theta},\boldsymbol{\eta})=\prod_{i=1}^{n}\frac{1}{x_{\rm max}-x_{\rm in}}\int_{x_{\rm in}}^{x_{\rm max}}dx\prod_k\frac{1}{\sqrt{2\pi s_{ik}^2}}\exp\left[-\frac{(d_{ik}-m_k(\boldsymbol{x};\boldsymbol{\theta}))^2}{2s_{ik}^2}\right]. \tag*{(68)} L ( θ , η ) = i = 1 ∏ n x max − x in 1 ∫ x in x max d x k ∏ 2 π s ik 2 1 exp [ − 2 s ik 2 ( d ik − m k ( x ; θ ) ) 2 ] . (68)
The variance s i k 2 = σ i k 2 + ( η k d i k ) 2 s_{ik}^2=\sigma_{ik}^2+(\eta_kd_{ik})^2 s ik 2 = σ ik 2 + ( η k d ik ) 2 includes a fractional intrinsic scatter, which absorbs the QPO widths and any inadequacy of the model. In the two-scatter model η U = η L = η k H z \eta_U=\eta_L=\eta_{k{\rm Hz}} η U = η L = η k Hz and η L F \eta_{LF} η L F is separate; in the one-scatter model all channels share one η \eta η .
Exact marginalisation over the radius
The model curves are tabulated on a grid x 0 < x 1 < ⋯ < x K x_0<x_1<\cdots<x_K x 0 < x 1 < ⋯ < x K and interpolated linearly. On the interval [ x a , x a + 1 ] [x_a,x_{a+1}] [ x a , x a + 1 ] , with t = x − x a ∈ [ 0 , h a ] t=x-x_a\in[0,h_a] t = x − x a ∈ [ 0 , h a ] , each curve is m k = m k , a + b k t m_k=m_{k,a}+b_kt m k = m k , a + b k t . With residuals ρ k = d i k − m k , a \rho_k=d_{ik}-m_{k,a} ρ k = d ik − m k , a and weights w k = s i k − 2 w_k=s_{ik}^{-2} w k = s ik − 2 ,
A = ∑ k w k b k 2 , B = ∑ k w k b k ρ k , C = ∑ k w k ρ k 2 . (69) A=\sum_k w_kb_k^2,\qquad B=\sum_k w_kb_k\rho_k,\qquad C=\sum_k w_k\rho_k^2. \tag*{(69)} A = k ∑ w k b k 2 , B = k ∑ w k b k ρ k , C = k ∑ w k ρ k 2 . (69)
Completing the square,
∫ 0 h a e − 1 2 ( A t 2 − 2 B t + C ) d t = 2 π A e − 1 2 ( C − B 2 / A ) [ Φ ( u a + A h a ) − Φ ( u a ) ] , (70) \int_0^{h_a}e^{-\frac{1}{2}(At^2-2Bt+C)}\,dt=\sqrt{\frac{2\pi}{A}}e^{-\frac{1}{2}(C-B^2/A)}\left[\Phi(u_a+\sqrt{A}h_a)-\Phi(u_a)\right], \tag*{(70)} ∫ 0 h a e − 2 1 ( A t 2 − 2 B t + C ) d t = A 2 π e − 2 1 ( C − B 2 / A ) [ Φ ( u a + A h a ) − Φ ( u a ) ] , (70)
with u a = − B / A u_a=-B/\sqrt{A} u a = − B / A and Φ \Phi Φ the standard normal distribution function. The integral over x x x is the sum over intervals. This is exact for the interpolated curves and remains accurate when the measurement errors are smaller than the change of the model between grid points, as for the lower kHz QPO of 4U 1728−34 (σ ≃ 1 \sigma\simeq1 σ ≃ 1 Hz); averaging the integrand over grid points would then be undersampled. We have checked the closed form against direct quadrature to 10 − 18 10^{-18} 1 0 − 18 in ln L \ln\mathcal{L} ln L .
Priors, evidence and derived quantities
With an EOS, the mass has a uniform prior on 1.0 1.0 1.0 –2.6 M ⊙ 2.6\,M_\odot 2.6 M ⊙ ; the spin frequency is fixed at the burst-oscillation value for the atoll sources and uniform on 200–750 Hz for the Z sources. Without an EOS, j j j and Q ˉ \bar{Q} Q ˉ have uniform priors on 0–0.45 and 1–12, and the mass prior is uniform up to 2.6 M ⊙ M_\odot M ⊙ for the atoll sources and 3.2 M ⊙ M_\odot M ⊙ for the Z sources. The scatters have log-uniform priors on 10 − 4 10^{-4} 1 0 − 4 –0.2. The posterior is p ( θ , η ∣ d ) = L π / Z p(\boldsymbol{\theta},\boldsymbol{\eta}\mid d)=\mathcal{L}\pi/Z p ( θ , η ∣ d ) = L π / Z with the evidence Z = ∫ L π d η d θ Z=\int\mathcal{L}\pi\,d\boldsymbol{\eta}\,d\boldsymbol{\theta} Z = ∫ L π d η d θ . For the fits with an EOS we evaluate Z Z Z and the posterior by quadrature on a grid in ( M , ν s , η ) (M,\nu_s,\eta) ( M , ν s , η ) ; for the fits without an EOS we use nested sampling (Ashton et al. 2019; Speagle 2020). We quote medians and 68 per cent intervals of the marginal posteriors. The derived quantities follow from the star: with an EOS, ( M , ν s ) ↦ ( j , q , I ) (M,\nu_s)\mapsto(j,q,I) ( M , ν s ) ↦ ( j , q , I ) by Appendix § 12 ; without one, J = j M 2 / ( 2 π ν s ) J=jM^2/(2\pi\nu_s) J = j M 2 / ( 2 π ν s ) in units with G = c = 1 G=c=1 G = c = 1 , that is I = j G M 2 / ( 2 π ν s ) I=jGM^2/(2\pi\nu_s) I = j G M 2 / ( 2 π ν s ) .
This paper has been typeset from a TEX/LATEX file prepared by the author.