Comparison between self-force and post-Newtonian dynamics: Beyond circular orbits The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters. Citation Akcay, Sarp, Alexandre Le Tiec, Leor Barack, Norichika Sago, and Niels Warburton. “Comparison Between Self-Force and Post-Newtonian Dynamics: Beyond Circular Orbits.” Phys. Rev. D 91, no. 12 (June 2015). © 2015 American Physical Society As Published http://dx.doi.org/10.1103/PhysRevD.91.124014 Publisher American Physical Society Version Final published version Accessed Thu May 26 12:52:49 EDT 2016 Citable Link http://hdl.handle.net/1721.1/97204 Terms of Use Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. Detailed Terms PHYSICAL REVIEW D 91, 124014 (2015) Comparison between self-force and post-Newtonian dynamics: Beyond circular orbits Sarp Akcay,1 Alexandre Le Tiec,2 Leor Barack,3 Norichika Sago,4 and Niels Warburton5 1 School of Mathematical Sciences and Complex & Adaptive Systems Laboratory, University College Dublin, Belfield, Dublin 4, Ireland 2 Laboratoire Univers et Théories, Observatoire de Paris, CNRS, Université Paris Diderot, 92190 Meudon, France 3 School of Mathematics, University of Southampton, Southampton SO17 1BJ, United Kingdom 4 Faculty of Arts and Science, Kyushu University, Fukuoka 819-0395, Japan 5 MIT Kavli Institute for Astrophysics and Space Research, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA (Received 5 March 2015; published 3 June 2015) The gravitational self-force (GSF) and post-Newtonian (PN) schemes are complementary approximation methods for modeling the dynamics of compact binary systems. Comparison of their results in an overlapping domain of validity provides a crucial test for both methods and can be used to enhance their accuracy, e.g. via the determination of previously unknown PN parameters. Here, for the first time, we extend such comparisons to noncircular orbits—specifically, to a system of two nonspinning objects in a bound (eccentric) orbit. To enable the comparison we use a certain orbital-averaged quantity hUi that generalizes Detweiler’s redshift invariant. The functional relationship hUiðΩr ; Ωϕ Þ, where Ωr and Ωϕ are the frequencies of the radial and azimuthal motions, is an invariant characteristic of the conservative dynamics. We compute hUiðΩr ; Ωϕ Þ numerically through linear order in the mass ratio q, using a GSF code which is based on a frequency-domain treatment of the linearized Einstein equations in the Lorenz gauge. We also derive hUiðΩr ; Ωϕ Þ analytically through 3PN order, for an arbitrary q, using the known near-zone 3PN metric and the generalized quasi-Keplerian representation of the motion. We demonstrate that the OðqÞ piece of the analytical PN prediction is perfectly consistent with the numerical GSF results, and we use the latter to estimate yet unknown pieces of the 4PN expression at OðqÞ. DOI: 10.1103/PhysRevD.91.124014 PACS numbers: 04.70.Bw, 04.25.D-, 04.25.dg, 04.25.Nx I. INTRODUCTION With the Advanced LIGO observatories scheduled to start science runs in 2015 [1], the next few years are likely to see first direct detections of gravitational waves from astrophysical sources. Prime targets are inspiraling and coalescing binary systems of neutron stars and/or black holes, with predicted rates that may be as high as a few dozen per observation year [2]. Theoretical templates of the gravitational waveforms must be developed to enable detection and interpretation of the weak signals [3]. The parameter space of these waveforms is too large for numerical relativity simulations to cover sufficiently well. Instead, the community has been seeking semianalytical models that can be informed by a judiciously chosen set of numerical relativity templates. A leading framework is the effective one-body (EOB) model, where the two-body relativistic dynamics is mapped onto a model of (non) geodesic motion in an effective spacetime [4–7]. EOB waveforms will play a crucial role in searches based on matched filtering, and there is an important need to refine the model, particularly in the strong-field regime [8–10]. 1550-7998=2015=91(12)=124014(32) One avenue of refinement is provided by the gravitational self-force (GSF) method, a perturbative scheme based on an expansion in the mass ratio of the binary [11–13]. The GSF approach is complementary to the postNewtonian (PN) approximation, a weak-field/smallvelocity expansion valid for arbitrary mass ratios [14]. Recently, there has been much activity in attempt to “synergize” the two schemes. The goal of such crosscultural studies is threefold: to test the two independent approximation schemes—GSF and PN—and help delineate their respective domains of validity; to determine yetunknown, high-order expansion terms in both approaches (hence, improving both approximations); and to help calibrate the EOB model across the entire inspiral parameter space. To facilitate such studies requires the identification of concrete gauge-invariant physical quantities that can be computed using both approaches. A first such quantity was identified by Detweiler in 2008 within the GSF framework [15]: the so-called “redshift” variable, defined for strictly circular orbits when dissipation is ignored. The functional relation between the redshift and the orbital frequency is a gauge-invariant diagnostic of the conservative sector of the 124014-1 © 2015 American Physical Society AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) binary dynamics. Detweiler made the first successful comparison with the PN prediction at 2PN order [15]. This comparison was later extended by Blanchet et al. to 3PN order and to even higher orders [16–20]. The calculation of the redshift through linear order in the mass ratio was subsequently confirmed by several other GSF computations in different gauges [21,22], which provided an internal consistency check for the GSF formalism. Soon after, Barack and Sago considered two more such “conservative” invariant quantities, namely, the frequency of the innermost stable circular orbit (ISCO) and the rate of periastron advance [23,24]. These results led to a plethora of comparisons between PN, GSF and numerical relativity [25–30] and the subsequent refinement of EOB theory [31–34]. More recently, the geodetic spin precession along circular orbits was computed by Dolan et al. through linear order in the mass ratio, and the numerical results successfully compared to a 3PNaccurate prediction [35]. The results allowed a numerical prediction of the (hitherto unknown) 4PN expression for the spin precession. This was later confirmed analytically by Bini and Damour [9], who also proceeded to obtain all PN terms up to the 8.5PN order, at linear order in the mass ratio. Dolan et al. [36] then presented a computation of the leading post-geodesic corrections to certain tidal invariants defined along the orbit. The PN series for these tidal invariants were also computed analytically up to 7.5PN order in Ref. [10], still at linear order in the mass ratio. All synergistic work so far has focused on circular orbits, for simplicity. Here, for the first time, we extend this program to orbits of arbitrary eccentricity. There are several reasons to do so. First, eccentricity provides more “handle” on the strong-field dynamics, giving access to new degrees of freedom in the EOB formulation. Second, although most Advanced LIGO binaries would have completely circularized by the time they enter the observable frequency band, there are scenarios where eccentricity effects could become observable and would give access to much interesting physics [37–42]. Third, eccentric inspirals in the extreme-mass-ratio regime will be key sources for a future mHz-band detector in space [43–47]. A gauge-invariant quantity for eccentric orbits, suitable for synergistic studies, was introduced by Barack and Sago in Ref. [24] (henceforth, BS2011). This quantity is a straightforward generalization of Detweiler’s redshift, obtained by averaging the time component of the particle’s four-velocity with respect to proper time over one epicyclic period of the motion. In other words, it is the ratio between the period measured in the coordinate time of a static observer at infinity and the proper-time period. This “averaged redshift,” denoted here hUi, is defined with the dissipative piece of the GSF ignored. The functional relationship between hUi and the two invariant frequencies that characterize the motion is a gauge-invariant diagnostic of the conservative eccentric-orbit dynamics. BS2011 calculated hUi (numerically) through linear order in the mass ratio for a sample of strong-field orbits, but they stopped short of attempting a calculation in a weaker-field regime where a meaningful comparison with PN results might be possible. The method of BS2011, which is based on a time-domain numerical integration of the relevant field equations, was best suited for tackling strong-field orbits, and its performance deteriorated fast with increasing orbital radius because of the longer evolution time required. Here we extend the range of BS2011’s calculation into the weaker-field regime, derive a 3PN-accurate formula for hUi, valid for any mass ratio, and compare between the numerical GSF results and the analytical PN prediction in the small mass-ratio limit. This is the first such comparison for noncircular orbits. It shows a good agreement for large and medium separations, and allows us to assess the performance of the PN expansion all the way down to the innermost stable orbit. Moreover, we are also able, through fits to the numerical GSF data, to extract some information about the 4PN approximation. Our numerical GSF calculation improves on that of BS2011 in both accuracy and weak-field reach. This improvement is achieved in two ways. First, our computation is based on the frequency-domain approach of Akcay et al. [48], in which the field equations are reduced to ordinary differential equations. This offers significant computational saving, particularly at lower eccentricities (e ≲ 0.4). Second, we have found a way to significantly simplify the expression given in BS2011 for hUi as a function of the two orbital frequencies. The new form requires a simpler type of numerical input, which can be obtained at greater accuracy. This paper is organized as follows. In Sec. II we review relevant results for bound motion in Schwarzschild spacetime and for the redshift as defined for circular orbits. We then extend the definition to eccentric orbits and obtain a simple expression for the generalized redshift hUi in terms of calculable perturbative quantities. Section III discusses the numerics and sources of error, and displays a sample of numerical results for hUi. In Sec. IV we perform a detailed derivation of the PN expression for hUi through 3PN order. Our calculations rely crucially on the known 3PN near-zone metric and the 3PN quasi-Keplerian representation of the motion. The numerical GSF and analytical PN results are compared in Sec. V. In Appendix A we establish the equivalence between our simplified formulation of hUi and that of BS2011. Appendix B derives some useful PN formulas valid in the test-mass limit. Table I summarizes some of our notation, for easy reference. In the GSF context, we denote the mass of 124014-2 COMPARISON BETWEEN SELF-FORCE AND POST- … TABLE I. PHYSICAL REVIEW D 91, 124014 (2015) Important symbols. m1 m2 m ¼ m1 þ m2 q ¼ m1 =m2 ν ¼ m1 m2 =m2 Δ ¼ ðm2 − m1 Þ=m Ωr Ωϕ particle’s mass black hole’s mass total mass mass ratio symmetric mass ratio reduced mass difference radial (epicyclic) frequency average azimuthal frequency the background Schwarzschild geometry by m2 and the mass of the orbiting particle by m1, with the assumption that q ≡ m1 =m2 ≪ 1. In the PN context, the two particles of masses m1 and m2 have an arbitrary mass ratio q. We will set G ¼ c ¼ 1, except in Sec. IV where we keep these constants explicit in PN expressions. We use a metric signature ð−; þ; þ; þÞ. where fðrÞ ≡ 1 − 2m2 =r, and V eff ðr; L2 Þ ≡ fðrÞð1 þ L2 =r2 Þ is an effective potential for thepffiffiradial motion. Bound ffi (eccentric) geodesics exist for 2 2=3 < E < 1 and L > pffiffiffi 2 3m2 . For a given bound geodesic, the radial distance rp ðτ0 Þ is confined to a finite range 2m2 < rmin ≤ rp ðτ0 Þ ≤ rmax < ∞ with rmin , rmax denoting periastron and apastron radii, respectively. These two turning-point radii can be mapped bijectively to fE; Lg. Thus the pair frmin ; rmax g can also parameterize the family of bound geodesics. Another such pair is given by the dimensionless “semilatus rectum” p and the “eccentricity” e, defined by p≡ 2rmax rmin ; m2 ðrmax þ rmin Þ A. Bound geodesic orbits in Schwarzschild spacetime We first review relevant results for bound geodesic motion in Schwarzschild spacetime. We consider a test particle of mass m1 moving on a bound (timelike) geodesic orbit in the Schwarzschild spacetime of a black hole of mass m2 . Using Schwarzschild coordinates ft; r; θ; ϕg, we label the position of the particle by xαp ðτ0 Þ ¼ ðtp ðτ0 Þ; rp ðτ0 Þ; θp ðτ0 Þ; ϕp ðτ0 ÞÞ, with four-velocity uα0 ≡ dxαp =dτ0 , where τ0 is a proper-time parameter along the geodesic, and the label ‘0’ indicates normalization with respect to the background (Schwarzschild) metric g0αβ , i.e., g0αβ uα0 uβ0 ¼ −1. Without loss of generality, we confine the motion to lie in the equatorial plane, i.e., θp ¼ π=2, such that uθ0 ¼ 0. We parameterize the geodesics by the two constants of motion: the specific energy E ≡ −u0t and specific angular momentum L ≡ u0ϕ , where u0α ¼ g0αβ uβ0 . The geodesic equation of motion is given by β u0 ∇0β uα0 ¼0, where ∇0β is the covariant derivative compatible with the background metric g0αβ . For the above setup, this gives dtp E ¼ ; dτ0 fðrp Þ ð2:1aÞ dϕp L ¼ ; dτ0 r2p ð2:1bÞ 2 drp ¼ E 2 − V eff ðrp ; L2 Þ; dτ0 rmax − rmin : rmax þ rmin ð2:2Þ These relations can be inverted to yield the Keplerian-like formulas rmax ¼ II. GENERALIZED REDSHIFT: FORMULATION IN SELF-FORCE APPROACH e≡ pm2 ; 1−e rmin ¼ pm2 : 1þe ð2:3Þ We can further express the specific energy and angular momentum in terms of p and e by solving the equation E 2 ¼ V eff ðr; L2 Þ at r ¼ frmin ; rmax g. Using Eqs. (2.3), this yields ðp − 2 − 2eÞðp − 2 þ 2eÞ 1=2 pm2 E¼ ; L ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : 2 pðp − 3 − e Þ p − 3 − e2 ð2:4Þ Following Darwin [49], we parameterize the radial motion using the “relativistic anomaly” χ via rp ðχÞ ¼ pm2 ; 1 þ e cos χ ð2:5Þ where χ ¼ 0 and χ ¼ π correspond to periastron and apastron passages, respectively. Using Eq. (2.1c) with Eq. (2.5), we obtain dτ0 m2 p3=2 ¼ dχ ð1 þ e cos χÞ2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi p − 3 − e2 ; p − 6 − 2e cos χ ð2:6Þ which, with the help of Eqs. (2.1a), (2.1b) and (2.4), also gives ð2:1cÞ 124014-3 dtp m2 p2 ¼ dχ ðp − 2 − 2e cos χÞð1 þ e cos χÞ2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðp − 2 − 2eÞðp − 2 þ 2eÞ ; × p − 6 − 2e cos χ dϕp ¼ dχ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi p : p − 6 − 2e cos χ ð2:7aÞ ð2:7bÞ AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) The functions τ0 ðχÞ, tp ðχÞ and ϕp ðχÞ are all monotonically increasing along the orbit. The radial periods in coordinate and proper times are calculated, respectively, via Z 2π Z 2π dtp dτ0 T r0 ¼ dχ; T r0 ¼ dχ; ð2:8Þ dχ dχ 0 0 and the accumulated azimuthal angle between successive periastron passages is rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z 2π dϕp p 4e Φ0 ¼ ellipK : dχ ¼ 4 p − 6 þ 2e p − 6 þ 2e dχ 0 ð2:9Þ R Here ellipKðkÞ ≡ 0π=2 ð1 − ksin2 θÞ−1=2 dθ is the complete elliptic integral of the first kind and subscripts ‘0’ serve to distinguish the geodesic values T r0 , T r0 and Φ0 from their corresponding GSF-perturbed quantities to be introduced below. For any ðp; eÞ we have Φ0 > 2π; hence, the periastron advances. We can now define the radial and (average) azimuthal frequencies via Ωr ≡ 2π ; T r0 Ωϕ ≡ Φ0 : T r0 ð2:10Þ The pair fΩr ; Ωϕ g provides a gauge-invariant parametrization of eccentric orbits. It should be noted, however, that the mapping between ðp; eÞ and ðΩr ; Ωϕ Þ is not bijective: there exist (infinitely many) pairs of physically distinct geodesics of different fp; eg values but the same set of frequencies. This degeneracy, first noted in BS2011, was thoroughly studied in [50]. The phenomenon is a feature of orbits very close to the innermost stable orbit. Since in this work we focus on less bound orbits (for the purpose of comparison with PN theory), the phenomenon of isofrequency pairing will not be relevant to us. In the parameter space of eccentric geodesics, stable orbits are located in the region given by p > 6 þ 2e. The curve p ¼ 6 þ 2e is called the “separatrix.” Along it both Φ0 and T r0 diverge, but Ωϕ remains finite. This gives rise to the so-called “zoom-whirl” behavior [51], where the orbiting particle zooms in from far away, whirls around the black hole many times, thus accumulating a large azimuthal phase, then zooms back out. In the limit p → 6 þ 2e, the particle sits exactly at the peak of the effective potential and whirls infinitely on an unstable circular geodesic. B. The redshift invariant for circular orbits Now let the particle’s mass m1 be finite but small, i.e., q ≡ m1 =m2 ≪ 1; ð2:11Þ theory, Detweiler and Whiting [52] showed that such a particle follows a geodesic motion in a certain smooth, effective, locally-defined spacetime with metric gαβ ¼ g0αβ þ hRαβ : ð2:12Þ Here, hRαβ is a certain smooth piece of the physical (retarded) metric perturbation produced by the particle. The physical perturbation itself is a solution of the linearized Einstein equation, sourced by the particle’s energy-momentum, with suitable “retarded” boundary conditions. How hRαβ may be computed in practice, on a Schwarzcshild background, is discussed, for example, in Ref. [23]. Within linear perturbation theory, hRαβ may be split into a dissipative piece and a conservative (time-symmetric) piece, and the effects of the two pieces may be considered separately. The conservative part of the perturbation is R;adv R;ret R defined as hR;cons ¼ 12 ðhR;ret αβ αβ þ hαβ Þ, where hαβ ≡ hαβ R;adv and hαβ is a smooth perturbation constructed just like hR;ret but starting with the particle’s “advanced” metric αβ R;cons perturbation. Replacing hR;ret in the effective αβ → hαβ metric (2.12) amounts to “turning off” the dissipation. The resulting equations of motion capture only conservative aspects of the dynamics. In Ref. [15] Detweiler considered a particle in circular geodesic motion in the “conservative” effective spacetime R;cons 0 gcons : αβ ¼ gαβ þ hαβ ð2:13Þ In the absence of dissipation the orbit remains circular, and the spacetime possesses a helical Killing vector field, which, on the orbit, is proportional to the 4-velocity uα ¼ dxα =dτ. We introduce here τ as a proper-time paramα eter along the geodesic in the effective metric gcons αβ , with u α β normalized with respect to that metric, i.e., gcons αβ u u ¼ −1. Thanks to the helical symmetry, all components of the particle’s 4-velocity are invariant under gauge transformations that respect the helical symmetry [21]. Detweiler proposed to use the functional relationship between ut and Ω ≡ uϕ =ut as a gauge-invariant benchmark for the conservative self-force effect beyond the geodesic approximation. The frequency Ω is the circular-orbit reduction of the frequency Ωϕ defined earlier for eccentric orbits. The quantity ut (or rather, its inverse) may be assigned a heuristic meaning of “redshift” (as measured in the smooth metric hR;cons by a static asymptotic observer located along αβ the helical symmetry axis), but it should be remembered that the true redshift, as measured in the physical metric of the particle, is, of course, divergent. Detweiler obtained [15] and consider the effect of self-interaction on the motion through OðqÞ. Within the context of linear perturbation 124014-4 ut ðΩÞ ¼ ut0 ðΩÞ þ qutgsf ðΩÞ; ð2:14Þ COMPARISON BETWEEN SELF-FORCE AND POST- … where and ut0 2=3 −1=2 ¼ ½1 − 3ðm2 ΩÞ PHYSICAL REVIEW D 91, 124014 (2015) is the geodesic limit, 1 qutgsf ¼ ut0 uα uβ hR;cons αβ 2 ð2:15Þ is the OðqÞ correction arising from self-interaction. Note that the correction utgsf is defined for a fixed value of Ω at the background, which ensures its gauge invariance. In Ref. [15] and subsequent work [16,17] (see also [18–20]), Detweiler and collaborators calculated numerically the post-geodesic correction utgsf ðΩÞ, and showed that it is consistent with corresponding PN expressions in an overlapping domain of validity. Detweiler’s numerical results were derived using the Regge-Wheeler gauge. An independent calculation using a direct numerical integration of the Lorenz-gauge form of the perturbation equations later recovered the same invariant relation utgsf ðΩÞ [21]. This comparison highlighted a subtlety in the notion of invariance as applied to utgsf ðΩÞ: the gauge transformation between the Lorenz-gauge metric perturbation and the Regge-Wheeler one does not leave utgsf ðΩÞ invariant, due to a certain minor gauge irregularity of the Lorenz-gauge metric (that was first identified in Ref. [53] and further discussed in [21]). Specifically, the physical metric perturbation does not vanish at infinity when expressed in the Lorenz gauge; see Eq. (2.23) below. While the perturbation remains helically symmetric, the transformation to an “asymptotically flat” gauge like Regge-Wheeler’s (or the harmonic gauge of PN theory), in which Eq. (2.15) applies, has a generator that itself does not have a helical symmetry. As a result, the transformation introduces a correction to utgsf ðΩÞ. Denoting by ĥαβ the Lorenz- gauge metric perturbation, one finds [21] 1 qutgsf ¼ ut0 uα uβ ĥR;cons þ αEðut0 Þ2 : αβ 2 ð2:16Þ The parameter α is extracted from the Lorenz-gauge perturbation as prescribed in Eq. (2.23) below; for a circular orbit it reads α ¼ qðm2 ΩÞ2=3 ut0 . One must be mindful, when working in the Lorenz gauge (as we do here), to take proper account of this gauge irregularity. We shall return to this point in more detail when discussing eccentric orbits. subscripts ‘0’ dropped. The functional relation between these invariant frequencies and any gauge-dependent set of parameters can be written as the sum of a “geodesic” term and a GSF correction; such relations were derived in explicit form in BS2011 but will not be needed here. The GSF-perturbed orbit is a geodesic in the effective R;cons 0 metric gcons , with tangent four-velocity uα αβ ¼ gαβ þ hαβ t normalized in gcons αβ . It is easily checked that u is no longer gauge-invariant in a pointwise sense when the orbit is noncircular. Instead, BS2011 suggested to consider the orbital average Z T r 1 T hUi ≡ hut i ≡ ut dτ ¼ r ; ð2:17Þ Tr 0 Tr where T r is the radial period measured in proper time τ. BS2011 argued that hUi is invariant under gauge transformations that respect the periodicity of the orbit and are well behaved (in a certain sense) at infinity. We may split hUi in the form hUiðΩi Þ ¼ hUi0 ðΩi Þ þ qhUigsf ðΩi Þ; where Ωi ≡ fΩr ; Ωϕ g, hUi0 is the geodesic limit of hUi taken with fixed Ωi , and qhUigsf is the GSF correction, defined for fixed Ωi . The functional relation hUigsf ðΩi Þ is an invariant measure of the GSF effect on the eccentric orbit, and it is the quantity that we will use for our GSF–PN comparison in this paper. The geodesic limit of hUi is given by hUi0 ¼ T r0 ; T r0 ð2:19Þ where the periods T r0 and T r0 may be calculated via (2.8) given the parameters p; e of the geodesic orbit. BS2011 describes a practical method for (numerically) inverting the relations Ωi ðp; eÞ in order to obtain pðΩi Þ and eðΩi Þ. This method may be used in conjunction with Eqs. (2.8) and (2.19) in order to compute hUi0 for given frequencies Ωi . Our goal now is to express hUigsf ðΩi Þ explicitly in terms of calculable perturbative quantities (the metric perturbation and/or the GSF). Since fixing Ωi fixes T r , the only contribution to hUigsf ðΩi Þ comes from the OðqÞ difference T r − T r0 . From the normalizations g0αβ uα0 uβ0 ¼ −1 and Þuα uβ ¼ −1 one obtains ðg0αβ þ hR;cons αβ C. The redshift invariant generalized to eccentric orbits Now consider an eccentric orbit subject to the conservative effect of the GSF. In absence of dissipation, the orbit remains bound and has a constant radial period T r and a constant accumulated azimuthal phase Φ per radial period. Hence, it possesses a well defined pair of frequencies fΩr ; Ωϕ g, defined via Eq. (2.10) with the ð2:18Þ dτ0 1 1 ¼ 1 þ uα0 uβ0 hR;cons ≡ 1 þ hRuu ; αβ 2 2 dτ ð2:20Þ where terms of Oðq2 Þ and higher are omitted. Here we have used uα0 ¼ dxα =dτ0 and uα ¼ dxα =dτ, assigning to each given physical point on the perturbed orbit the same coordinate value it has on the background orbit [this choice 124014-5 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) is consistent with the “fixed frequency” mapping introduced in Eq. (2.18)]. Since the contraction hRuu automatically picks out the conservative piece of hRαβ , the label ‘cons’ becomes redundant and we have dropped it. Neglecting subleading terms in the mass ratio q, we now obtain Z T r dτ 1 T r − T r0 ¼ 1 − 0 dτ ¼ − T r0 hhRuu i; ð2:21Þ 2 dτ 0 Eq. (2.25) reduces to (2.16) in the circular-orbit limit. As in the circular case, the expression for hUigsf involves only the R field along the orbit (and the parameter α), and not the GSF itself. Our result (2.25) is much simpler than the one derived in BS2011 using a different procedure. In that work, certain simplifications that reduce the expression for hUigsf to the form (2.25) have been overlooked. In Appendix A we establish the equivalence between the two results. where the average is taken with respect to τ (or τ0 ) over a radial period. The OðqÞ perturbation of hUi ¼ T r =T r at fixed Ωi therefore reads III. NUMERICAL CALCULATION OF THE GENERALIZED REDSHIFT qhUigsf T 1 ¼ − r0 2 ðT r − T r0 Þ ¼ hUi0 hhRuu i: 2 ðT r0 Þ ð2:22Þ This would be our final result for hUigsf if hRuu were to be calculated in a suitable “asymptotically flat” gauge. Our calculation, however, will be performed in the Lorenz gauge, which suffers from the aforementioned irregularity at infinity. Let us now describe this irregularity more specifically. For either circular or noncircular orbits, the Lorenz-gauge metric component ĥtt tends to a finite nonzero value at r → ∞ (other components are regular). This behavior is due entirely to the static piece of the mass monopole perturbation, and therefore the asymptotic value of ĥtt does not depend on the angular direction even for eccentric orbits; it depends only on the orbital parameters. To remove this gauge artifact, following BS2011 we introduce the normalized time coordinate t ¼ ð1 þ αÞt̂, where t̂ denotes the original Lorenz-gauge time coordinate, and α ¼ αðΩi Þ is given by 1 α ¼ − ĥtt ðr → ∞Þ: 2 A. Details of numerics and sources of error ð2:23Þ This normalization, which amounts to an OðqÞ gauge transformation away from the Lorenz gauge, corrects the asymptotic behavior. Under t̂ → t we have, at leading order, ĥRuu → hRuu þ 2αg0tt ðut0 Þ2 ¼ hRuu − 2αEut0 : ð2:24Þ Thus, to re-express hUigsf in Eq. (2.22) in terms of the Lorenz-gauge perturbation, we need simply replace hhRuu i → hĥRuu i þ 2αEhUi0 . We finally get 1 qhUigsf ¼ hUi0 hĥRuu i þ αEhUi20 : 2 We have used the frequency-domain computational framework of Ref. [48] in order to compute hUigsf for a large sample of orbits, focusing primarily on obtaining weak-field data for PN comparisons. Our calculation is based on Eq. (2.25), which takes as input the regularized Lorenz-gauge metric perturbation evaluated along the orbit (as well as the asymptotic value α, also to be read off the Lorenz-gauge perturbation). Since the GSF correction qhUigsf (defined at fixed frequencies Ωi ) is of OðqÞ, it is sufficient to use as input the metric perturbation calculated along geodesic orbits. For convenience we shall use p; e (as defined in Sec. II A), rather than Ωi , to parameterize these geodesics, and will thus express our results in the form hUigsf ¼ hUigsf ðp; eÞ. It is important to emphasize that our results refer to the GSF correction to the functional relation hUigsf ðΩi Þ defined for fixed invariant frequencies Ωi , even though we use the geodesic parameters p and e as independent variables. These two facts should not be confused. ð2:25Þ Equation (2.25) is one of our main results, giving hUigsf in terms of quantities directly calculable using existing GSF codes: the R field ĥRαβ in the Lorenz gauge, and the corresponding asymptotic parameter α. It is clear that We use the eccentric-orbit GSF code of Ref. [48] to obtain the metric perturbation ĥRαβ ðχÞ along the geodesic orbit. This code employs a frequency-domain approach, coupled with the method of extended homogeneous solutions of Ref. [54], to compute the regularized metric perturbation ĥRαβ . It then outputs ĥRuu ðχÞ at 2400 evenly spaced points along the orbit, and interpolates the numerical data using Mathematica’s Interpolation function. In its default setting, Interpolation fits cubic polynomials between successive data points. Since ĥRuu ðχÞ is very smooth this level of interpolation is sufficient for our purposes. We subsequently calculate the orbital average hĥRuu i using NIntegrate with the appropriate numerical integration options/controls offered by MATHEMATICA. The coefficient α is extracted, using (2.23), from the static monopole piece of the metric perturbation, whose construction is prescribed in App. B of [48]. Since this piece is essentially known analytically (its computation involves the evaluation of a certain orbital integral, easily done with MATHEMATICA at extremely high accuracy), numerical 124014-6 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) error in our calculation of hUigsf comes entirely from the numerical evaluation of ĥRuu ðχÞ. Reference [48] contains a thorough analysis of error sources for ĥRuu ðχÞ and the GSF. Here, we briefly review two dominant sources. Each Fourier mode of our computation has associated ~ r þ mΩ ~ ϕ , where n~ and m ~ are with it a frequency, ω ¼ nΩ integer harmonic numbers. The dominating source of numerical error depends on the value of ω. For modes of sufficiently large frequency (m2 ω ≳ 10−4 ), the dominant error comes from the estimation of the contribution from the tail of uncomputed multipoles of large l values. Typically, we compute the contributions to all the l-modes up to and including l ¼ 15, and estimate the remaining contribution to the mode sum by fitting numerical data to suitable power-law models of the large-l behavior [23,48,55]. This is a relatively well-modeled and well-controlled source of error, and it can be reduced in a straightforward manner using additional computational resources. For modes with small frequencies, m2 ω ≲ 10−4 , a second source of numerical error takes over. This comes from rounding errors introduced when inverting the matrix of amplitude coefficients as part of the procedure for computing inhomogeneous solutions to the Lorenzgauge field equations [48]. When ω is very small, the matrix becomes nearly singular, and its inversion using machine-precision arithmetic introduces large errors. The problematic “nearly-static” modes occur generically in our TABLE II. Numerical data for the GSF contribution hUigsf to the generalized redshift (defined with fixed invariant frequencies Ωi ), for various eccentric geodesic orbits in a Schwarzschild background. The orbital parameters e (eccentricity) and p (semilatus rectum) are defined in Sec. II A. Parenthetical figures indicate estimated error bars on the last displayed decimals. e 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 e 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 e 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 e 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 p ¼ 10 −0.12878023ð4Þ −0.1277540ð3Þ −0.1260434ð2Þ −0.123648ð3Þ −0.120567ð2Þ −0.1168020ð6Þ −0.112352ð2Þ −0.107221ð2Þ p ¼ 15 −0.07751154ð5Þ −0.0768706ð2Þ −0.07580395ð6Þ −0.07431376ð9Þ −0.07240329ð5Þ −0.0700768ð5Þ −0.0673398ð5Þ −0.064199ð1Þ p ¼ 20 p ¼ 25 −0.0556761252ð8Þ −0.05522166ð7Þ −0.05446527ð1Þ −0.05340854ð9Þ −0.0520537ð3Þ −0.05040377ð4Þ −0.0484623ð4Þ −0.0462337ð9Þ −0.0434829334ð1Þ −0.043132423ð1Þ −0.042548963ð6Þ −0.041733648ð3Þ −0.04068802ð6Þ −0.0394141ð2Þ −0.0379143ð1Þ −0.0361916ð2Þ p ¼ 30 p ¼ 35 p ¼ 40 p ¼ 50 −0.0356833158ð1Þ −0.035398479ð5Þ −0.03492427ð6Þ −0.034261480ð8Þ −0.03341121ð1Þ −0.0323749ð2Þ −0.0311543ð2Þ −0.0297515ð6Þ −0.0302606957ð1Þ −0.0300209627ð8Þ −0.029621799ð3Þ −0.029063791ð5Þ −0.028347763ð7Þ −0.0274748ð4Þ −0.0264462ð2Þ −0.0252634ð2Þ −0.0262706836ð5Þ −0.0260637929ð8Þ −0.025719277ð2Þ −0.025237591ð2Þ −0.024619373ð3Þ −0.02386545ð8Þ −0.0229768ð4Þ −0.0219547ð5Þ −0.0207905297ð5Þ −0.0206282073ð6Þ −0.0203578673ð9Þ −0.019979802ð2Þ −0.01949443ð6Þ −0.01890227ð5Þ −0.0182040ð1Þ −0.0174004ð4Þ p ¼ 60 p ¼ 70 p ¼ 80 p ¼ 90 −0.0172030750ð2Þ −0.0170695612ð3Þ −0.016847175ð1Þ −0.016536122ð4Þ −0.01613669ð1Þ −0.01564925ð2Þ −0.01507427ð8Þ −0.0144123ð3Þ −0.0146718447ð3Þ −0.0145584700ð2Þ −0.0143696130ð3Þ −0.0141054255ð4Þ −0.0137661204ð6Þ −0.013351971ð9Þ −0.0128633ð1Þ −0.0123002ð5Þ −0.0127901170ð3Þ −0.0126916092ð2Þ −0.0125275072ð7Þ −0.0122979274ð7Þ −0.012003033ð2Þ −0.011643033ð7Þ −0.01121819ð3Þ −0.0107288ð2Þ −0.0113362814ð8Þ −0.0112491974ð8Þ −0.0111041186ð3Þ −0.0109011371ð5Þ −0.010640382ð2Þ −0.010322020ð3Þ −0.00994625ð3Þ −0.0095133ð2Þ p ¼ 100 p ¼ 110 p ¼ 120 p ¼ 130 −0.0101792669ð2Þ −0.0101012344ð10Þ −0.0099712296ð7Þ −0.0097893274ð4Þ −0.0095556341ð8Þ −0.009270281ð4Þ −0.00893344ð2Þ −0.0085453ð2Þ −0.0092365822ð2Þ −0.0091658975ð1Þ −0.0090481316ð3Þ −0.0088833462ð1Þ −0.0086716263ð10Þ −0.008413085ð2Þ −0.00810786ð1Þ −0.0077561ð2Þ −0.0084537150ð1Þ −0.0083891149ð2Þ −0.0082814820ð1Þ −0.0081308692ð1Þ −0.0079373488ð7Þ −0.007701017ð5Þ −0.00742198ð3Þ −0.0071004ð2Þ −0.0077931956ð9Þ −0.0077337155ð1Þ −0.0076346120ð1Þ −0.0074959280ð1Þ −0.0073177296ð5Þ −0.007100089ð2Þ −0.00684311ð3Þ −0.0065469ð2Þ 124014-7 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) calculation, since, given any orbital parameters, there will ~ and n~ in the Fourier sum for which m2 ω is exist values of m ~ is truncated very small. In practice, the sum over n~ and m once our results reach a desired accuracy. Consequently, the problem is less severe for low-eccentricity orbits, where the effective frequency band is narrow, and more severe at high eccentricity, where the broad frequency band implies a higher chance of encountering nearly static modes. Ultimately, this restricts our calculation to orbits with eccentricities of e ≲ 0.4. Low-ω modes are encountered also when the fundamental frequencies themselves are small, as with weak-field orbits—the main focus of the present work. Our code incorporates several methods for mitigating this small-frequency problem (see Ref. [48] for details), but even with these techniques employed, our current calculation appears limited to orbits with p ≲ 130; at larger p we observe a rapid reduction in accuracy. The issue of nearly static modes has been addressed in a very recent paper by Osburn et al. [56], who proposed additional mitigation methods. These may be used to improve the performance of weak-field calculations in future work. B. Numerical results Table II displays a sample of our numerical results for hUigsf . Parenthetical figures indicate estimated error bars on the last displayed decimals; for instance, −0.0556761ð1Þ stands for −0.0556761 1 × 10−7. Additional data may be made available to interested readers upon request from the authors. Some of the data shown in the table will be plotted in Sec. V, where we discuss the comparison with PN results. As a check of our frequency-domain computation, we compare our results for hUigsf with those obtained in FIG. 1 (color online). Numerical GSF output for hUigsf (black data points) versus analytical PN approximations (solid curves). Each panel shows hUigsf as a function of semilatus rectum p for a fixed eccentricity e. Insets display, on a log-log scale, the relative differences ΔnPN rel ≡ j1 − U nPN =hUigsf j, where U nPN is the PN approximation through nPN order. In both the main plots and the insets, the three curves correspond, top to bottom, to the 1PN, 2PN and 3PN approximations. Solid curves in the insets are the analytical PN residues 1 − U 1PN =U 3PN (upper curve) and 1 − U 2PN =U 3PN (middle curve); for the lower curve we have fitted the simple model 1 − U 3PN =hUigsf ¼ p−4 ðα1 þ α2 ln p þ α3 =pÞ. 124014-8 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) TABLE III. Our frequency-domain numerical results for hUigsf and the corresponding time-domain values from BS2011. Each cell shows our result (top) in comparison to BS2011’s (bottom). The relative disagreement between the two data sets is ∼10−6 , roughly consistent with the magnitude of error bars. As discussed in the text, evidence suggests that our results are accurate to within the error bars given, whereas the magnitude of error in the time-domain data is slightly underestimated in some cases. No time-domain data exist for p > 20 to allow comparison in the weak-field domain. e p ¼ 10 p ¼ 15 p ¼ 20 0.1 Here −0.1277540ð3Þ −0.0768706ð2Þ −0.05522166ð7Þ BS2011 −0.1277554ð7Þ −0.0768709ð1Þ −0.05522177ð4Þ 0.2 −0.123648ð3Þ −0.07431376ð9Þ −0.05340854ð9Þ −0.1236493ð7Þ −0.0743140ð1Þ −0.05340866ð4Þ 0.3 −0.1168020ð6Þ −0.0700768ð5Þ −0.05040377ð4Þ −0.1168034ð6Þ −0.0700771ð1Þ −0.05040388ð4Þ 0.4 −0.107221ð2Þ −0.064199ð1Þ −0.0462337ð9Þ −0.1072221ð5Þ −0.0641991ð1Þ −0.04623383ð4Þ BS2011 using a time-domain method. BS2011 provided a small sample of numerical results in the range p ≤ 20 and e ≤ 0.5. The comparison is shown in Table III. There is evidently a good agreement between the two sets of numerical results, although in some of the entries the values appear not fully consistent given the stated error bars (in all these cases the BS2011 values are smaller than ours). We have strong evidence to suggest that the source of disagreement is a slight underestimation of the magnitude of systematic error in the time-domain analysis of BS2011: We have tested the output of our frequency-domain code against accurate GSF data published in Ref. [56], and against yet unpublished redshift data calculated by van de Meent [57] (using a very different frequencydomain method based on a semianalytical treatment of Teukolsky’s equation [58]). These comparisons strongly favor the frequency-domain data in the table. We have also gone back to the time-domain code of BS2011 and implemented more robust error analysis methods, again suggesting that the error reported in BS2011 is slightly underestimated in some cases. Also evident from the table is the fact that our code’s accuracy starts to degrade for e ¼ 0.4; however, it still matches BS2011’s results to five or six significant digits. No published numerical data exist to allow comparison beyond p ¼ 20. (Reference [56] gives results for e ≤ 0.7 and p ≤ 90, but these are for the GSF components, not for hUigsf .) except that we will consider the orbital average of a quantity that is related to the orbital dynamics of a binary of nonspinning compact objects, modeled as point particles, while Refs. [59,60] calculated the orbital-averaged fluxes of energy and angular momentum radiated at infinity. Furthermore, while these fluxes are invariant under the exchange 1↔2 of the bodies’ labels, and requires knowledge of the gravitational field in the wave zone, the generalized redshift hUi is a property of one particle, whose evaluation involves the near-zone metric. A. Redshift variable in standard harmonic coordinates 1. The regularized 3PN metric Throughout Sec. IV we assume that m1 < m2 , and treat m1 as the “particle” orbiting the “black hole” of mass m2 . The redshift of the particle can be computed from the knowledge of the regularized PN metric gαβ ðy1 Þ ≡ gαβ ðt; y 1 Þ, generated by the two bodies and evaluated at the coordinate location y 1 ðtÞ of the particle, as [16] U≡ We shall now derive the invariant relation hUiðΩr ; Ωϕ Þ within the PN approximation. Our calculations will be similar in spirit to those performed by Arun et al. [59,60], vα1 vβ1 −1=2 ¼ −gαβ ðy1 Þ 2 ; c ð4:1Þ where vα1 ¼ ðc; v1 Þ, with v1 ¼ dy 1 =dt the coordinate velocity of the particle. The generalized redshift will be given by the proper-time average of Eq. (4.1) over one radial period. The regularized PN metric gαβ ðy1 Þ was itself computed up to 2.5PN order, in harmonic coordinates, in Ref. [61]. This calculation was then extended to 3PN order in Ref. [16], partly based on existing computations of the 3PN equations of motion using Hadamard regularization [62] and dimensional regularization [63]. Reference [16] performed two calculations of the 3PN regularized metric, using both Hadamard and dimensional regularizations, obtaining the same metric but expressed in two different harmonic coordinate systems. The two metrics were found to differ by an infinitesimal 3PN coordinate transformation in the “bulk,” i.e., outside the particle’s worldlines, and also by an intrinsic shift of these worldlines. Combining Eqs. (4.2) and (A15) of Ref. [16], the 3PN-accurate expression of the regularized metric reads, in the standard harmonic coordinates corresponding to the use of Hadamard regularization,1 As usual we denote by r12 ¼ jy 1 − y 2 j the coordinate separation, by n12 ¼ ðy 1 − y 2 Þ=r12 the unit direction from particle 2 to particle 1, and by v12 ¼ v1 − v2 the relative velocity, where va ¼ dy a =dt is the 3-velocity of particle a. The Euclidean scalar product between two 3-vectors A and B is denoted ðABÞ. Parentheses around indices are used to indicate symmetrization, i.e., Aði BjÞ ¼ Ai Bj þ Aj Bi . 1 IV. GENERALIZED REDSHIFT: POST-NEWTONIAN CALCULATION ut1 124014-9 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) 2Gm2 Gm2 Gm1 Gm2 Gm2 3 2 2 ðn12 v2 Þ4 − 3ðn12 v2 Þ2 v22 þ 4v42 4v2 − ðn12 v2 Þ − 3 þ 6 g00 ðy1 Þ ¼ −1 þ 2 þ −2 r12 r12 c r12 c4 r12 c r12 4 Gm2 Gm2 Gm1 87 47 55 39 23 þ 3ðn12 v2 Þ2 − v22 þ 2 þ − ðn12 v1 Þ2 þ ðn12 v1 Þðn12 v2 Þ − ðn12 v2 Þ2 − ðv1 v2 Þ þ v21 4 2 4 2 4 r12 r12 r12 47 Gm1 17 Gm2 Gm 5 Gm2 þ 8 2 − ðn12 v2 Þ6 − 5ðn12 v2 Þ2 v42 þ 3ðn12 v2 Þ4 v22 þ 4v62 þ þ v22 − þ ð−4ðn12 v2 Þ4 4 2 r12 r12 r12 c r12 8 Gm1 617 491 225 41 ðn v Þ4 þ ðn12 v1 Þ3 ðn12 v2 Þ − ðn12 v1 Þ2 ðn12 v2 Þ2 þ ðn12 v1 Þðn12 v2 Þ3 − þ 5ðn12 v2 Þ2 v22 − v42 Þ þ 24 12 1 6 4 2 r12 53 79 101 49 27 ðn12 v2 Þ2 ðv1 v2 Þ þ ðv1 v2 Þ2 − ðn12 v1 Þ2 v21 þ ðn12 v2 Þ4 − ðn12 v1 Þ2 ðv1 v2 Þ þ 42ðn12 v1 Þðn12 v2 Þðv1 v2 Þ þ 8 4 4 4 8 23 273 39 305 139 ðn12 v2 Þ2 v21 − 25ðv1 v2 Þv21 þ v41 − ðn12 v1 Þ2 v22 þ ðn12 v1 Þðn12 v2 Þv22 þ ðn12 v1 Þðn12 v2 Þv21 − 4 8 8 8 4 291 77 235 4 G2 m 2 r 32 ðn12 v2 Þ2 v22 − 62ðv1 v2 Þv22 þ v21 v22 þ v2 þ 2 1 ln 12 32ðn12 v12 Þ2 − v212 − 8 2 8 3 r0 r12 r12 6603 584 r12 1881 r 2 2789 364 2 þ ðn12 v1 Þ − ln 0 þ ln 0 − 44 ln 12 þ ðn12 v1 Þðn12 v2 Þ þ ðn12 v2 Þ − 100 5 50 5 20 r1 r1 r01 3053 584 r 1613 364 r 1421 44 r þ ðv1 v2 Þ − − ln 12 þ ln 12 þ ln 12 þ v21 − þ v22 0 0 150 15 100 15 60 3 r1 r1 r01 Gm1 1091 44 r 32 r12 G2 m1 m2 r12 2 − 32 v2 − ln 12 ln þ − þ þ ln 32ðn v Þ 12 12 60 3 3 3 12 r12 r01 r0 r0 r212 109 141 2 197 177 2 2 2 391 213 2 þ ðn12 v1 Þ − − π þ ðn12 v1 Þðn12 v2 Þ − þ π þ ðn12 v2 Þ − π 12 16 6 8 6 16 812 59 2 299 47 2 1097 71 2 Gm1 3571 15 2 44 r 2 2 þ ðv1 v2 Þ − π þ v1 − þ π þ v2 − þ π þ − π − ln 12 − 9 8 18 16 18 16 60 8 3 r12 r01 64 r Gm2 28141 15 2 44 r 32 r12 G2 m22 Gm2 2 2 − π þ ln 12 ln þ ln 12 þ − þ þ −ðn v Þ þ 2v − 2 12 2 2 3 630 8 3 3 r0 r12 r02 r0 r12 r212 þ oðc−8 Þ; ð4:2aÞ 4Gm2 i Gm2 i Gm1 Gm2 Gm1 i Gm1 2 2 i v 2ðn þ 4 v þ v Þ − 4v − 2 þ v þ n f10ðn12 v1 Þ þ 2ðn12 v2 Þg 12 2 2 12 r12 r12 r12 1 r12 c3 r12 2 c5 r12 2 Gm2 Gm2 i 3 Gm1 4 2 2 4 − v − ðn12 v2 Þ þ 4ðn12 v2 Þ v2 − 4v2 þ 48ðn12 v1 Þ2 − 44ðn12 v1 Þðn12 v2 Þ ðn v Þ þ 7 2 r12 12 2 r12 c r12 2 Gm2 Gm2 G2 m1 m2 95 3 2 2 2 2 2 2 þ 10ðn12 v2 Þ þ 40ðv1 v2 Þ − 16v2 − 26v1 þ − π −2ðn12 v2 Þ þ v2 − 2 þ 6 4 r12 r12 r212 G2 m2 1013 24 r 17 43 i Gm1 þ ln 12 þ 2 1 − þ v − ðn12 v1 Þ2 − 15ðn12 v1 Þðn12 v2 Þ þ ðn12 v2 Þ2 þ 3ðv1 v2 Þ 1 75 5 2 2 r01 r12 r12 2 2 2 17 2 15 2 G m1 m2 57 3 2 G m1 1973 24 r12 Gm1 21 i − ln 0 ðn v Þ3 − þ π þ 2 þ n12 þ v1 − v2 þ 2 2 2 4 75 5 2 12 1 r1 r12 r212 r12 43 29 3 1 − ðn12 v1 Þ2 ðn12 v2 Þ − ðn12 v1 Þðn12 v2 Þ2 þ ðn12 v2 Þ3 − 11ðn12 v1 Þðv1 v2 Þ − 19ðn12 v2 Þðv1 v2 Þ þ ðn12 v1 Þv21 2 2 2 2 2 39 41 3 Gm2 Gm2 G m1 m2 51 2 2 2 2 2 þ ðn v Þ þ ðn12 v2 Þv1 þ ðn12 v1 Þv2 þ ðn12 v2 Þv2 þ ðn v Þ 2ðn12 v2 Þ − v2 − 2 2 2 2 2 12 1 r12 12 2 r12 r212 97 9 G2 m2 2398 1088 72 r þ oðc−7 Þ; ðn12 v1 Þ þ ðn12 v2 Þ þ ðn12 v12 Þ ln 12 − ðn12 v2 Þ − π 2 ðn12 v12 Þ þ 2 1 − 2 4 25 25 5 r01 r12 g0i ðy1 Þ ¼ − ð4:2bÞ 124014-10 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) 2Gm Gm Gm1 Gm2 Gm1 Gm2 þ 4vi2 vj2 þ ni12 nj12 −8 gij ðy1 Þ ¼ δij þ 2 2 δij þ 4 2 δij −ðn12 v2 Þ2 þ þ þ r12 r12 r12 r12 c r12 c r12 Gm2 ij 3 Gm2 Gm1 71 47 2 59 Gm1 25 Gm2 4 2 2 2 2 þ 6 δ − ðn12 v12 Þ þ v12 − ðn v Þ − v2 ðn12 v2 Þ − ðn v Þ þ þ 4 12 2 4 4 5 r12 6 r12 r12 12 2 r12 c r12 Gm1 Gm2 Gm1 ði jÞ Gm1 i j Gm1 þ vi2 vj2 4v22 − 2ðn12 v2 Þ2 − 10 þ 24 −16ðn12 v1 Þ2 − v v − 16 v v þ ni12 nj12 r12 r12 r12 1 2 r12 1 1 r12 182 Gm1 Gm2 Gm1 Gm2 Gm1 ði jÞ 2 2 þ −2ðn12 v2 Þ þ 3 þ 40 þ 32ðn12 v1 Þðn12 v2 Þ − 12v12 þ þ2 ðn12 v12 Þn12 v1 5 r12 r12 r12 r12 r12 Gm2 ði jÞ Gm1 þ n12 v2 f−60ðn12 v1 Þ þ 36ðn12 v2 Þg þ 2 ðn v Þ þ oðc−6 Þ: ð4:2cÞ r12 r12 12 2 Since we are considering only the conservative part of the binary dynamics, we did not include in (4.2) the dissipative 2.5PN radiation-reaction terms; these can be found in Eqs. (7.6) of Ref. [61]. Notice the occurrence at 3PN order of some logarithmic terms, containing two constants r01 and r02 (one for each body) that have the dimension of a length. These ultraviolet (UV) regularization parameters come from regularizing the self-field of point particles using the Hadamard regularization of Ref. [62]. The constants r01 and r02 are gauge-dependent, as they can be arbitrarily changed by a coordinate transformation of the bulk metric [16,62] or by some shifts of the worldlines of the particles [63]. The metric coefficient g00 ðy1 Þ also involves a constant r0 that originates from the infrared (IR) regularization of the metric at spatial infinity, as discussed in Ref. [16]. This arbitrary IR scale should also disappear from the final gauge-invariant results. We introduce the expression (4.2) of the regularized 3PN metric into the definition (4.1) of the redshift, and expand in powers of 1=c, keeping all terms up to Oðc−8 Þ. This gives an expression for U as a function of the two masses m1 and m2 , the coordinate separation r12 , and the scalar products ðn12 v1 Þ, ðn12 v2 Þ, ðv1 v1 Þ, ðv1 v2 Þ and ðv2 v2 Þ, as well as the regularization constants r0 , r01 and r02 , in an arbitrary reference frame. The resulting expression is too lengthy to be displayed here. 2. Reduction to the center-of-mass frame We wish to specialize the previous expression to the center-of-mass (c.m.) frame, which is consistently defined at 3PN order by the vanishing of the center-of-mass integral deduced from the 3PN binary equations of motion [64]. This condition yields expressions for the individual positions y a and velocities va relatively to the c.m. in terms of the relative position y ≡ y 1 − y 2 and relative velocity v ≡ v1 − v2 [65]. Since these results play an important role in our algebraic manipulations, we recall here the expressions for the functional relationships y a ½y; v in the harmonic gauge that was used to derive the regularized metric (4.2). Thus, y 1 ¼ ½X2 þ νðX1 − X2 ÞPy þ νðX1 − X2 ÞQv þ oðc−6 Þ; ð4:3aÞ y 2 ¼ ½−X1 þ νðX1 − X2 ÞPy þ νðX1 − X2 ÞQv þ oðc−6 Þ; ð4:3bÞ where ν ≡ m1 m2 =m2 is the symmetric mass ratio and Xa ≡ ma =m, with m ≡ m1 þ m2 the total mass of the binary. The coefficients P and Q depend on the parameters m and ν, the separation r ≡ jyj, the relative velocity squared v2 ≡ ðvvÞ, and the radial velocity r_ ≡ ðnvÞ. They explicitly read [65] 1 v2 Gm 1 Gm 2 19 3 Gm 2 1 3 G2 m 2 7 ν 4 3 3 _ þ − ν þ v þ ν þ r þ ν þ − v − − 2r 8 2 r 8 2 r 8 4 4 2 c2 2 c4 r2 1 5 11 Gm 4 53 15 2 Gm 4 1 5 21 2 Gm 2 2 5 21 11 2 6 2 þ 6 v − ν þ 6ν þ v − 7ν − ν þ r_ − νþ ν þ v r_ − þ ν − ν 16 4 r 16 2 r 16 8 16 r 16 16 2 c 2 2 2 2 3 3 2 Gm 101 33 Gm 7 73 Gm 14351 22 r ν ν þ 2 v2 − − ν þ 3ν2 þ 2 r_ 2 − þ ν þ 4ν2 þ 3 þ ln 00 þ − ; ð4:4aÞ 12 8 3 8 1260 3 r0 8 2 r r r P¼ Q¼− 7 Gm_r Gm_r 2 15 21 19 Gm 235 21 2 5 _ þ ν þ r − ν − þ ν : v − þ 4 c4 8 4 12 24 r 24 4 c6 ð4:4bÞ Again, we did not include the radiation-reaction contributions at 2.5PN order. A logarithmic term contributes at 3PN order in (4.4a); it involves a particular combination r000 of the gauge constants r01 and r02 , that is defined by 124014-11 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) ðX1 − X2 Þ ln r000 ¼ X21 ln r01 − X22 ln r02 : ð4:5Þ By computing the time derivatives of Eqs. (4.3)–(4.4) and by applying, where necessary, an iterative orderreduction of all accelerations by means of the c.m. equations of motion given in Eqs. (3.9)–(3.10) of Ref. [65], we obtain expressions analogous to (4.3)–(4.4) for the particle’s individual velocities va as functions of the relative variables y and v. Replacing the positions and velocities by their c.m. expressions y a ½y; v and va ½y; v yields 3PN-accurate expressions for the scalar products ðn12 v1 Þ, ðn12 v2 Þ, ðv1 v1 Þ, ðv1 v2 Þ, ðv2 v2 Þ as functions of r, r_ and v2 . Finally, the c.m. expression for the redshift U½r; r_ ; v2 in standard harmonic coordinates takes the form U ¼1þ 1 1 1 1 U þ 4 U 1PN þ 6 U 2PN þ 8 U 3PN þ oðc−8 Þ; 2 N c c c c ð4:6Þ where the various PN contributions read 1 Δ ν 2 1 Δ Gm UN ¼ þ − v þ þ ; 4 4 2 2 2 r 2 2 3 3 5 11 2 4 5 5 ν 2 Gm 2 Δ Gm 2 1 Δ Gm þ Δ − ν − Δν þ ν v þ þ Δ − − ν v þ −ν ν r_ þ þ − 2ν ; U1PN ¼ 16 16 8 8 4 4 2 r 2 r 4 4 r2 ð4:7aÞ ð4:7bÞ 5 5 3 19 157 2 75 2 85 3 6 27 27 37 13 5 19 11 Gm 4 þ Δ − ν − Δν þ ν þ Δν − ν v þ þ Δ − ν − Δν þ ν2 − Δν2 þ ν3 v 32 32 2 16 32 32 16 16 16 8 4 16 16 2 r 5 7 11 23 Gm 2 2 23 23 17 59 2 1 2 3 3 G2 m2 2 2 r_ v þ þ Δ − 2ν − Δν þ ν − Δν − ν þ − þ Δ − ν − Δν þ 6ν ν v 8 8 8 8 r 8 8 8 8 4 2 r2 1 Δ 19 55 2 5 2 9 3 G2 m2 2 3 9 9 3 2 Gm 4 r_ þ þ − 5ν þ Δν − ν þ Δν − ν r_ þ − Δ þ ν þ Δν − ν ν 4 4 8 8 4 2 8 16 16 2 r r2 3 3 1 Δ 9 1 G m þ þ − ν − Δν ; ð4:7cÞ 4 4 8 8 r3 U2PN ¼ 35 35 221 705 2 939 2 855 3 665 3 3059 4 8 65 65 49 163 þ Δ − 2ν − Δν þ ν þ Δν − ν − Δν þ ν v þ þ Δ− ν− Δν U3PN ¼ 256 256 128 64 128 32 64 128 32 32 4 16 169 2 45 2 57 3 145 3 255 4 Gm 6 61 19 131 229 251 2 67 2 289 3 þ ν þ Δν þ ν þ Δν − ν v þ − þ Δþ ν− Δν þ ν þ Δν − ν 8 4 8 16 8 r 32 32 32 32 16 4 8 2 2 Gm 2 4 231 231 19 39 297 2 129 2 597 3 55 Gm 4 15 33 9 ×ν r_ v þ þ Δ − ν − Δν þ ν þ Δν − ν − Δν3 þ ν4 − Δ þ Δν v þ r 32 32 2 4 16 16 16 4 32 32 2 r2 87 111 3 Gm 4 2 3 3 323 283 667 2 617 2 725 3 27 3 161 4 − 9ν2 − Δν2 þ ν ν r_ v þ þ Δ − νþ Δν − ν − Δν þ ν − Δν þ ν 16 8 r 8 8 32 32 32 32 16 2 4 G2 m 2 2 2 35 35 39731 71 2 23761 71 2 64933 2 35 2 1 × 2 r_ v þ þ Δþ − þ π νþ − þ π Δν − ν þ Δν þ12ν3 − Δν3 − 2ν4 8 8 2520 64 1260 64 5040 16 2 r 3 3 16 r 11 r 11 r G m 2 5 15 35 þ ð1 − Δ þ ν − ΔνÞνln 0 þ ð−1 − Δ þ 3ν þ ΔνÞνln 0 − νln v þ Δ − ν − Δν 3 r0 3 r1 3 r2 16 32 32 r3 2 2 35 5 15 Gm 6 425 73 177 Δν 2 3 Gm þ ν2 þ Δν2 − ν3 ν r_ þ − − Δþ ν þ − ν2 þ Δν2 − 9ν3 ν 2 r_ 4 16 8 8 r 48 6 16 8 3 2 r 5 5 17917 213 2 5767 213 2 122333 2 7 r 2 3 5 3 4 þ π νþ − π Δν þ ν þ Δν −16ν þ Δν − 14ν þ 16νln þ þ Δ− 4 4 420 64 840 64 1680 16 2 r0 3 3 r r Gm 2 3 3 67853 15 2 _ þ11ð−1 þ Δ − ν þ ΔνÞνln 0 þ11ð1 þ Δ − 3ν − ΔνÞνln 0 þ Δ − þ π ν r þ r1 r2 16 16 5040 32 r3 4 4 20141 15 2 656 2 16 r 11 r 11 r G m þ π Δν − ν þ νln − þ ð−1 þ ΔÞνln 0 þ ð1 þ Δ − 2νÞνln 0 : ð4:7dÞ 5040 32 315 3 r0 3 r1 3 r2 r4 124014-12 COMPARISON BETWEEN SELF-FORCE AND POST- … pffiffiffiffiffiffiffiffiffiffiffiffiffi Here, Δ ≡ ðm2 − m1 Þ=m ¼ X2 − X1 ¼ 1 − 4ν denotes the reduced mass difference, so that the test-mass limit of particle 1 corresponds to ν → 0. Since the redshift (4.1) is a property of particle 1, the expressions (4.7) are not symmetric by exchange 1↔2 of the bodies’ labels. The redshift of particle 2 is simply obtained by setting Δ → −Δ in Eqs. (4.7). As expected, the regularization constants r0 , r01 and r02 that enter the expression (4.2) of the regularized 3PN metric appear in the c.m. expression (4.6)–(4.7) for the redshift. In Sec. IV D we will check that the orbital averaging cancels out the dependance on these arbitrary length scales. PHYSICAL REVIEW D 91, 124014 (2015) U½r; r_ ; v2 in SH coordinates by means of the functional equality U0 ¼ U þ δξ U, where δξ U ¼ − ∂U ∂U ∂U δξ r − δξ r_ − 2 δξ v2 þ Oðξ2 Þ; ∂r ∂ r_ ∂v with δξ r ¼ ðnξÞ þ Oðξ2 Þ; _ þ ðvξÞ − r_ ðnξÞ þ Oðξ2 Þ; δξ r_ ¼ ðnξÞ r r B. Redshift variable in alternative coordinates In the previous section we obtained an expression for the redshift variable in the standard harmonic (SH) coordinate system, namely, the coordinate system in which the 3PN equations of motion were originally derived [62,65]. These coordinates are such that the equations of motion involve some gauge-dependent logarithmic terms at 3PN order. Importantly, these logarithms prevent the use of the 3PN quasi-Keplerian representation of the binary motion (reviewed in Sec. IV C below), thus impeding the averaging of the redshift over an orbit. Therefore, it is useful to have the expression for the redshift in a modified harmonic (MH) coordinate system, without logarithmic terms in the equations of motion, such as the one used in Refs. [59,60]. Alternatively, we shall use ADM-type coordinates, which are also free of such logarithms at 3PN order in the equations of motion. Both the MH coordinates and the ADM coordinates are suitable for a 3PN quasi-Keplerian parametrization of the motion [66]. This will require us to re-express the redshift in terms of the variables r, r_ and v2 in these alternative coordinate systems. 1. Modified harmonic coordinates The trajectories y 0a ðtÞ of the particles in MH coordinates are related to their counterparts y a ðtÞ in SH coordinates by some 3PN shifts ξa ðtÞ of the worldlines induced by a coordinate transformation in the “bulk,” namely, y 0a ¼ y a þ ξa [62]. Therefore, in the c.m. frame, the MH coordinate separation y 0 is related to the SH coordinate separation y through y 0 ¼ y þ ξ, where the relative shift ξ ≡ ξ1 − ξ2 is given by [59] 22 G3 m3 ν r ξðSH→MHÞ ¼ − ln 0 n þ oðc−6 Þ; ð4:8Þ 6 2 3 cr r0 with n ≡ y=r the unit direction pointing from particle 2 to particle 1. Following [59], we introduced the “logarithmic barycenter” r00 of the constants r01 and r02 , (not to be confused with the IR constant r0 ): ln r00 ≡ X1 ln r01 þ X2 ln r02 : ð4:9Þ The expression U 0 ½r; r_ ; v2 for the redshift in MH coordinates can then be deduced from the formula for ð4:10Þ _ þ Oðξ2 Þ: δξ v2 ¼ 2ðvξÞ ð4:11aÞ ð4:11bÞ ð4:11cÞ Since the relative shift (4.8) comes at 3PN order, the nonlinear terms Oðξ2 Þ in Eqs. (4.10) and (4.11) contribute at leading 6PN order, and can thus be neglected. Plugging the expression (4.8) into Eqs. (4.11), we find the explicit expressions 22 G3 m3 ν r ðδξ rÞðSH→MHÞ ¼ − ð4:12aÞ ln 0 ; 6 2 3 cr r0 22 G3 m3 ν r _ ðδξ r_ ÞðSH→MHÞ ¼ − r − 2_ r ln ; ð4:12bÞ 6 3 3 cr r00 ðδξ v2 ÞðSH→MHÞ 44 G3 m3 ν 2 r 2 2 r_ þ ðv − 3_r Þ ln 0 : ¼− 6 3 3 cr r0 ð4:12cÞ In order to compute the change δξ U in the redshift induced by the relative shift (4.8), we only require the Newtonian expression for U½r; r_ ; v2 , which is given by Eq. (4.7). Combined with (4.10) and (4.12), this gives 11 G3 m3 ν ð1 þ Δ − 2νÞ_r2 ðδξ UÞðSH→MHÞ ¼ 3 c 8 r3 2Gm r 2 2 ν ln 0 þ ð1 þ ΔÞðv − 3_r Þ − r r1 þ ð1 þ Δ − 3ν − ΔνÞðv2 − 3_r2 Þ Gm r − þ oðc−8 Þ: ð1 þ Δ − 2νÞ ln 0 r r2 ð4:13Þ Adding the above shift to the formula (4.6)–(4.7) for the redshift in SH coordinates yields the expression for the redshift in MH coordinates. Since U 0 ¼ U þ δξ U is a functional equality, the resulting MH redshift is expressed as a function of the “dummy” variables r, r_ and v2 . Adding together Eqs. (4.7d) and (4.13), we find that the UV regularization constant r02 disappears from the 124014-13 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) expression for U0 ½r; r_ ; v2 in MH coordinates. However the UV and IR constants r01 and r0 remain and enter the result through the logarithmic contributions 11 G3 m3 ν 2 2 Gm v − 3_ r − ½U0 log ¼ 3 c 8 r3 r r 16 r × ð1 − Δ þ 2νÞ ln 0 − ln : ð4:14Þ r1 11 r0 For circular orbits, such that r_ ¼ 0 and v2 ¼ Gm=rþ Oðc−2 Þ, these logarithmic contributions cancel out. We will see that the factor ðv2 − 3_r2 − Gm=rÞ=r3 vanishes when ξðSH→ADMÞ ¼ averaged over one radial period, such that the constants r0 and r01 will cancel out from the final, gauge-invariant result for the orbital-averaged redshift, as expected. 2. ADM-type coordinates Similarly, the individual trajectories y 0a ðtÞ of the particles in ADM coordinates are related to the trajectories y a ðtÞ in SH coordinates by some shifts ξa ðtÞ of the worldlines: y 0a ¼ y a þ ξa [64,67]. In the c.m. frame, the ADM coordinate separation y 0 is related to the SH coordinate separation y through y 0 ¼ y þ ξ, where the relative shift ξ ¼ ξ1 − ξ2 reads [59,65] Gm 5 2 ν 2 Gm 1 9 Gmν 1 11 15 4 2 2 5 _ _ − v − νv r þ 3ν n þ ν_ r v þ þ ν þ r − ν þ − v 8 8 r 4 4 2 8 16 16 c4 c6 2 2 1 5 Gm 2 451 3 Gm 2 161 5 G m 2773 21 2 22 r 4 þ r_ − þ ν − v þ ν þ r_ − ν þ 2 þ π − ln 0 n 16 16 r 48 8 r 48 2 280 32 3 r0 r 21 5 29 Gm 43 2 17 2 þ v − ν þ r_ − þ ν þ þ 5ν r_ v þ oðc−6 Þ; ð4:15Þ 8 4 12 24 r 3 from which the authors of Ref. [59] deduced, using Eqs. (4.11), the transformation of variables that we need to compute the redshift in ADM coordinates2: ðδξ rÞðSH→ADMÞ ¼ ðδξ r_ ÞðSH→ADMÞ Gm 5 2 19 2 Gm 1 Gmν 4 1 11 39 99 2 2 4 23 73 _ _ νv ν_ r þ 3ν þ − ν þ r þ ν þ r − ν v − − þ v 8 r 4 2 8 16 16 48 48 c4 8 c6 2 2 Gm 2 451 3 Gm 2 283 5 Gm 2773 22 r 21 þ − ; ð4:16aÞ v þ ν − r_ þ ν þ 2 þ ln 0 − π 2 r 48 8 r 16 2 280 3 r0 32 r Gm 19 2 19 2 Gm 1 ν Gmν 39 99 163 443 4 2 2 − þ þ 6 r_ v − þ ν þ r_ v − ν ¼ 4 r_ − νv þ ν_r þ 4 4 r 4 2 8 8 24 24 cr cr 23 73 Gm 2 1603 17 Gm 2 1777 131 4 þ r_ − þ ν − v þ ν þ r_ þ ν 12 12 r 48 4 r 48 24 G2 m2 3121 44 r 21 11 þ 2 − ln 0 þ π 2 − ν ; ð4:16bÞ 105 3 r0 16 4 r ðδξ v2 ÞðSH→ADMÞ ¼ Gm 13 4 5 2 2 3 4 Gm 2 1 21 Gm 2 19 _ 1 þ νv ν_ r ν_ r v þ ν − r ν þ v þ þ − 4 2 4 r 2 2 r 2 c4 r Gmν 6 13 31 31 75 3 5 25 Gm 4 9 25 2 4 4 2 6 þ 6 − ν þ r_ v − ν þ r_ − þ ν − v þ ν v − þ ν þ r_ v 4 4 8 8 2 8 8 r 8 4 cr Gm 2 2 131 121 Gm 4 99 259 G2 m2 3839 44 r 21 þ r_ v − þ ν þ r_ − ν þ 2 v2 − þ ln 0 − π 2 þ ν r 8 4 r 4 12 420 3 r0 16 r 2 2 Gm 28807 r 63 13 − 44 ln 0 þ π 2 − ν : ð4:16cÞ þ 2 r_ 2 420 r0 16 2 r The expression U 0 ½r; r_ ; v2 for the redshift in ADM coordinates can then be deduced from the result (4.6)–(4.7) in SH coordinates via the functional equality U0 ¼ U þ δξ U. Using the expressions (4.10) and (4.16), the SH redshift is found to be modified by 2PN and 3PN corrections that read 2 The remainder Oðξ2 Þ in Eqs. (4.10) and (4.11) is of order 4PN, which is still negligible in the transformation to ADM coordinates. 124014-14 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) Gm 13 4 5 2 2 3 4 ¼ 6 ð1 þ Δ − 2νÞν − v þ v r_ þ r_ 16 8 16 cr Gm 2 1 Δ 33 37 21 þ v þ þ ν þ Δν − ν2 r 8 8 16 16 4 Gm 2 1 Δ 11 19 19 G2 m2 1 Δ 3 3 − r_ þ þ ν þ Δν − ν2 − 2 þ þ ν þ Δν r 4 4 16 16 4 8 8 2 2 r Gmν 65 65 161 205 2 þ 8 þ Δ− ν − 6Δν þ ν −v6 32 32 16 16 cr 61 61 297 175 185 2 þ v4 r_ 2 þ Δ− ν− Δν þ ν 32 32 32 32 16 9 15 21 45 2 2 _4 9 þv r þ Δ − ν − Δν þ ν 32 32 8 16 16 5 5 35 25 25 − r_ 6 þ Δ − ν − Δν þ ν2 32 32 32 32 16 Gm 4 661 661 1669 125 11 2 þ r_ þ Δ− νþ Δν − ν r 96 96 96 96 6 2 2 Gm 3 39 33 287 3 4 3 2 2 þ 8 2 v þ Δ − ν − Δν − 18ν − 14Δν þ ν 16 16 16 16 8 cr 3 3 53 29 847 2 361 2 þ v2 r_ 2 − − Δ þ ν þ Δν þ ν þ Δν − 36ν3 8 8 32 32 32 32 Gm 2 5 5 2189 21 2 2329 21 2 5263 21 2 2 þ v þ Δþ − π νþ − π Δν þ þ π ν r 16 16 1120 64 1120 64 1680 32 Δν2 11 r 11 r þ − 8ν3 þ ð1 þ ΔÞν2 ln 0 þ ð1 þ Δ − 3ν − ΔνÞν ln 0 3 r1 3 r2 16 Gm 2 5 5 53099 63 2 50159 63 2 15821 63 2 2 − r_ þ Δ− þ π ν− þ π Δν þ þ π ν r 4 4 3360 64 3360 64 420 32 11 59 r r þ Δν2 − ν3 þ 11ð1 þ ΔÞν2 ln 0 þ 11ð1 þ Δ − 3ν − ΔνÞν ln 0 8 4 r1 r2 G2 m2 1 Δ 2493 21 2 1933 21 2 þ 2 þ π νþ þ π Δν þ 12ν2 − − þ 8 8 560 64 560 64 r 22 r 11 r þ oðc−8 Þ: − ν2 ln 0 − ð1 þ Δ − 2νÞν ln 0 3 r1 3 r2 ðδξ UÞðSH→ADMÞ Although the additional contribution (4.17) in ADM coordinates is more involved than its counterpart (4.13) in MH coordinates, they share the same logarithmic terms. Thus, adding Eqs. (4.7d) and (4.17) we find that the UV regularization constant r02 disappears from the expression for U0 ½r; r_ ; v2 in ADM coordinates, while the constants r0 and r01 remain and enter the final result through the logarithmic terms (4.14). C. The generalized quasi-Keplerian representation Before we discuss the orbital averaging of the redshift in Sec. IV D, we must summarize the 3PN generalized ð4:17Þ quasi-Keplerian (QK) representation of the motion of Memmesheimer et al. [66]. Indeed, since averaging over one radial period is most conveniently performed using an explicit solution of the equations of motion, the generalized QK representation is an essential input for our 3PN calculation. The QK representation was originally introduced by Damour and Deruelle [68] to account for the leading-order 1PN general relativistic effects in the timing formula of the Hulse-Taylor binary pulsar. It was later extended at 2PN order in Refs. [69–71], in ADM coordinates, and more recently at 3PN order [66] in both ADM and harmonic coordinates. 124014-15 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) We first introduce the mean anomaly l ≡ Ωr ðt − tper Þ; ð4:18Þ where tper is the coordinate time at a periastron passage and Ωr ¼ 2π=T r is the radial frequency (also known as the mean motion n), i.e., the frequency associated with the periodicity T r of the radial motion. The mean anomaly simply maps one radial period t ∈ ½tper ; tper þ T r Þ to the trigonometric interval l ∈ ½0; 2πÞ. We then adopt a parametric description of the binary’s motion in polar coordinates, in the c.m. frame, in terms of the eccentric anomaly u ∈ ½0; 2πÞ. At 3PN order, this parametrization reads rðuÞ ¼ ar ð1 − er cos uÞ; ð4:19aÞ lðuÞ ¼ u − et sin u þ f t sin V þ gt ðV − uÞ þ it sin 2V þ ht sin 3V; with βϕ ≡ ½1 − ð1 − e2ϕ Þ1=2 =eϕ . Equations (4.18)–(4.20) provide a 3PN-accurate generalization of the usual Keplerian representation of the Newtonian motion.3 The previous generalized QK representation is complete only once the orbital elements Ωr , K, ar , et , er , eϕ , f t , gt , it , ht , f ϕ , gϕ , iϕ and hϕ have been related to the first integrals of the motion, namely, the binding energy E and the orbital angular momentum J, both per reduced mass μ ¼ m1 m2 =m. Following Ref. [59], we shall instead make use of the convenient, dimensionless, coordinate-invariant quantities4 ε≡− ð4:19bÞ ϕðuÞ ¼ ϕper þ KðV þ f ϕ sin 2V þ gϕ sin 3V þ iϕ sin 4V þ hϕ sin 5VÞ; angle of return to periastron is given by Φ ¼ 2πK (equivalent to ΔΦ ¼ 2πk), and the true anomaly V is defined by βϕ sin u VðuÞ ¼ u þ 2 arctan ; ð4:20Þ 1 − βϕ cos u ð4:19cÞ where ϕper is the value of the orbital phase when t ¼ tper , at a periastron passage, K ≡ 1 þ k is the fractional angle of advance of the periastron per orbital revolution, such that the 2E ; c2 j≡− 2EJ2 ; ðGmÞ2 ð4:21Þ such that ε > 0 and j > 0 for a generic bound eccentric orbit (since E < 0 for such orbits). Notice the PN scalings ε ¼ Oðc−2 Þ and j ¼ Oðc0 Þ. Therefore, we shall consider expansions in powers of the PN parameter ε, with coefficients depending on j and ν. In ADM coordinates, the 3PNaccurate expressions for the orbital elements read [59,66] ε3=2 c3 ε ε2 192 1 þ ½−15 þ ν þ 555 þ 30ν þ 11ν2 þ 1=2 ð−5 þ 2νÞ 8 Gm 128 j 3 ε 5760 þ −29385 − 4995ν − 315ν2 þ 135ν3 þ 1=2 ð17 − 9ν þ 2ν2 Þ 3072 j 16 2 2 3 − 3=2 ð10080 − 13952ν þ 123π ν þ 1440ν Þ þ oðε Þ ; j 3ε ε2 3 15 ADM ð−5 þ 2νÞ þ 2 ð7 − 2νÞ ¼1þ þ K j 4 j j 3 ε 24 1 þ ð5 − 5ν þ 4ν2 Þ − 2 ð10080 − 13952ν þ 123π 2 ν þ 1440ν2 Þ 128 j j 5 2 2 þ 3 ð7392 − 8000ν þ 123π ν þ 336ν Þ þ oðε3 Þ; j Gm ε ε2 1 2 aADM ½−7 þ ν þ ð−68 þ 44νÞ ¼ þ 1 þ 1 þ 10ν þ ν r 4 j 16 c2 ε 3 ε 1 þ 3 − 9ν − 6ν2 þ 3ν3 þ ð864 − 2212ν − 3π 2 ν þ 432ν2 Þ j 192 1 2 2 3 − 2 ð6432 − 13488ν þ 240π ν þ 768ν Þ þ oðε Þ ; j ΩADM ¼ r ð4:22aÞ ð4:22bÞ ð4:22cÞ In the Newtonian limit, ar is the semimajor axis, the three eccentricities coincide (et ¼ er ¼ eϕ ≡ e), an eccentric orbit does not precess (K ¼ 1), and f t ¼ gt ¼ it ¼ ht ¼ f ϕ ¼ gϕ ¼ iϕ ¼ hϕ ¼ 0. 4 For circular orbits, we have the well-known Newtonian limits ε ∼ v2 =c2 ∼ Gm=ðrc2 Þ and j ∼ 1. 3 124014-16 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) ε 1 − j þ ½−8 þ 8ν þ jð17 − 7νÞ 4 2 ε 24 þ 8 þ 4ν þ 20ν2 − 1=2 ð5 − 2νÞ þ 24j1=2 ð5 − 2νÞ 8 j 4 − jð112 − 47ν þ 16ν2 Þ þ ð17 − 11νÞ j 3 ε þ 24ð−2 þ 5νÞð−23 þ 10ν þ 4ν2 Þ þ 15jð528 − 200ν þ 77ν2 − 24ν3 Þ 192 2 − 72j1=2 ð265 − 193ν þ 46ν2 Þ − ð6732 − 12508ν þ 117π 2 ν þ 2004ν2 Þ j 2 þ 1=2 ð16380 − 19964ν þ 123π 2 ν þ 3240ν2 Þ j 2 − 3=2 ð10080 − 13952ν þ 123π 2 ν þ 1440ν2 Þ j 1=2 96 2 2 3 þ 2 ð134 − 281ν þ 5π ν þ 16ν Þ þ oðε Þ ; j ε eADM ¼ 1 − j þ ½24 − 4ν þ 5jð−3 þ νÞ r 4 ε2 8 2 2 þ 52 þ 2ν þ 2ν − jð80 − 55ν þ 4ν Þ þ ð17 − 11νÞ j 8 3 ε þ −768 − 344ν − 6π 2 ν − 216ν2 þ 3jð−1488 þ 1556ν − 319ν2 þ 4ν3 Þ 192 4 − ð588 − 8212ν þ 177π 2 ν þ 480ν2 Þ j 1=2 192 þ 2 ð134 − 281ν þ 5π 2 ν þ 16ν2 Þ þ oðε3 Þ ; j ε ADM eϕ ¼ 1 − j þ ½24 þ jð−15 þ νÞ 4 ε2 1 þ −32 þ 176ν þ 18ν2 − jð160 − 30ν þ 3ν2 Þ þ ð408 − 232ν − 15ν2 Þ j 16 3 ε þ −16032 þ 2764ν þ 3π 2 ν þ 4536ν2 þ 234ν3 − 36jð248 − 80ν þ 13ν2 þ ν3 Þ 384 6 − ð2456 − 26860ν þ 581π 2 ν þ 2689ν2 þ 10ν3 Þ j 1=2 3 2 2 3 3 þ 2 ð27776 − 65436ν þ 1325π ν þ 3440ν − 70ν Þ þ oðε Þ ; j eADM ¼ t f ADM t pffiffiffiffiffiffiffiffiffiffi ε2 1 − j ¼− νð4 þ νÞ 8 j1=2 ε3 j1=2 1 4148 2 2 2 3 p ffiffiffiffiffiffiffiffiffiffi ν þ π ν þ 200ν þ 11ν þ νð−64 − 4ν þ 23ν Þ þ 2 576 − 3 64 1 − j j 1 4232 þ þ oðε3 Þ; −576 þ ν − π 2 ν − 209ν2 − 35ν3 j 3 124014-17 ð4:22dÞ ð4:22eÞ ð4:22fÞ ð4:22gÞ AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) gADM t 3ε2 5 − 2ν ε3 1 þ ¼ ð10080 − 13952ν þ 123π 2 ν þ 1440ν2 Þ 1=2 3=2 2 192 j j 1 2 þ 1=2 ð−3420 þ 1980ν − 648ν Þ þ oðε3 Þ; j ε3 1 − j νð23 þ 12ν þ 6ν2 Þ þ oðε3 Þ; 32 j3=2 13ε3 1 − j 3=2 3 hADM ¼ ν þ oðε3 Þ; t j 192 iADM ¼ t f ADM ¼ ϕ ε2 1 − j νð1 − 3νÞ 8 j2 ε3 4ν 1 þ ð−11 − 40ν þ 24ν2 Þ þ 2 ð−256 þ 1192ν − 49π 2 ν þ 336ν2 − 80ν3 Þ 256 j j 1 2 2 3 þ 3 ð256 − 1076ν þ 49π ν − 384ν − 40ν Þ þ oðε3 Þ; j 3ε2 ν2 ð1 − jÞ3=2 32 j2 pffiffiffiffiffiffiffiffiffiffi ε3 1 − j 1 220 2 2 ν νð9 − 26νÞ þ þ π þ 104ν þ 50ν − j 3 256 j 1 220 2 2 − 2 þ oðε3 Þ; þ π þ 32ν þ 15ν 3 j ð4:22hÞ ð4:22iÞ ð4:22jÞ ð4:22kÞ gADM ¼− ϕ iADM ¼ ϕ ð4:22lÞ ε3 ð1 − jÞ2 νð5 þ 28ν þ 10ν2 Þ þ oðε3 Þ; 128 j3 ð4:22mÞ 5ε3 ν3 ð1 − jÞ5=2 þ oðε3 Þ: 256 j3 ð4:22nÞ hADM ¼ ϕ The eccentricities eADM , eADM and eADM are all such that t r ϕ 2 −2 j ¼ 1 − e þ Oðc Þ at Newtonian order; they start differing from each other at leading 1PN order. The expressions (4.22) are specific to the ADM coordinates. Before we give the corresponding expressions in MH coordinates, let us recall an important point related to the use of gauge-invariant variables. As shown in Ref. [69], the functional forms of Ωr ¼ 2π=T r and K ¼ Φ=ð2πÞ as functions of gauge-invariant variables like ε and j are identical in different coordinate systems. In particular we have the exact same relations in MH coordinates as in ADM coordinates: ΩMH ¼ ΩADM ≡ Ωr ; r r ð4:23aÞ K MH ¼ K ADM ≡ K: ð4:23bÞ We may therefore use any combination of Ωr and K instead of the constants of the motion ε and j to parameterize in a physically meaningful way a given eccentric orbit (assuming a one-to-one relation). Following Ref. [59], we introduce the frequency Ωϕ ≡ KΩr , which is a natural generalization of the circular-orbit frequency Ω,5 and we define the dimensionless coordinate-invariant parameters (remember that k ¼ K − 1) GmΩϕ 2=3 3x x≡ ð4:24Þ ; ι≡ : k c3 The PN parameter x is Oðc−2 Þ, while ι is merely Newtonian at leading order (the relativistic periastron advance first appears at 1PN order). The choice of variables (4.24) is the obvious generalization of the gauge-invariant variable x that is commonly used for Note that Ωϕ coincides with R the average angular frequency of _ ≡ 1 T r ϕðtÞdt. _ the motion: Ωϕ ¼ hϕi Tr 0 124014-18 5 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) circular orbits. It will thus facilitate checking the circularorbit limit. In Sec. IV E, we shall express our final results in terms of either of the two sets of gauge-invariant parameters ðε; jÞ or ðx; ιÞ. To compute the invariant relationships hUiðε; jÞ and hUiðx; ιÞ from the expressions (4.6)–(4.7) and (4.10) for the redshift in MH coordinates, we shall also need expressions for the orbital elements ar , et , er , eϕ , f t , gt , it , ht , f ϕ , gϕ , iϕ and hϕ in these coordinates. They are given by Eqs. (4.22c)–(4.22l), to which we must add the differences [66] Gmε 5 1 1 17 þ ν − ¼ 2 − νþ 8 j 4 4 c Gmε2 ν ν2 1 1 11499 21 2 19 2 þ 2 þ þ − þ − þ π νþ ν 32 32 j 2 560 32 4 c 1 3 14501 21 2 þ 2 þ − π ν − 5ν2 þ oðε2 Þ; 420 16 j 2 ε2 1 17 1 MH ADM p ffiffiffiffiffiffiffiffiffiffi et − et þ ν 1− ¼ j 1−j 4 4 ε3 19 52 225 2 1 29 79039 21 2 201 2 þ pffiffiffiffiffiffiffiffiffiffi − − ν þ ν þ þ − π ν− ν 32 3 32 j 16 1680 16 16 1−j 1 3 14501 21 2 2 þ 2 − þ − þ oðε3 Þ; þ π ν þ 5ν 2 420 16 j ε2 1 73 5 1 1 17 MH ADM er − er þ ν−j ν− þ ν ¼ pffiffiffiffiffiffiffiffiffiffi 8 j 2 2 1−j 2 8 ε3 13 5237 21 2 19 2 143 37 2 þ pffiffiffiffiffiffiffiffiffiffi þ − þ π νþ ν þj − νþ ν 1680 32 16 64 64 1 − j 16 1 13 3667 105 2 51 þ þ − π ν − ν2 j 8 56 32 4 1 14501 21 2 2 þ 2 −3 þ − þ oðε3 Þ; þ π ν þ 10ν 210 8 j ε2 1 71 ν 1 1 141 ADM ¼ pffiffiffiffiffiffiffiffiffiffi eMH − ν þ j þ þ ν − e − ϕ ϕ 4 16 32 j 4 32 1−j 3 ε 13 36511 21 2 1723 2 17 33 2 þ pffiffiffiffiffiffiffiffiffiffi − þ − π ν− ν þj νþ ν 32 8960 128 256 256 256 1−j 1 13 21817 147 2 169 2 þ − þ − þ π νþ ν j 16 480 64 8 1 3 621787 273 2 1789 2 þ 2 þ oðε3 Þ; þ − π ν− ν 13440 128 128 j 2 19ε2 1 − j 1=2 ADM f MH − f ¼ ν t t j 8 ε3 296083 21 2 989 2 361 171 2 þ π νþ ν þj ν− ν þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi −1 þ − 6720 32 64 64 64 jð1 − jÞ 1 276133 21 2 799 2 1þ − π ν− ν þ oðε3 Þ; þ j 6720 32 64 aMH r aADM r − gADM ¼ oðε3 Þ; gMH t t 124014-19 ð4:25aÞ ð4:25bÞ ð4:25cÞ ð4:25dÞ ð4:25eÞ ð4:25fÞ AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) 11ε3 1 − j νð19 − 10νÞ þ oðε3 Þ; ð4:25gÞ 32 j3=2 ε3 1 − j 3=2 MH ADM ht − ht ¼ νð23 − 73νÞ þ oðε3 Þ; ð4:25hÞ j 192 ε2 1 1 ε3 1 1045 99 2 1 5 139633 21 2 117 2 − þ ν− ν þ − þ − þ π νþ ν ð1 þ 18νÞ þ ¼− 32 j 4 3360 16 8 8 j j2 j 32 192 1 3 92307 21 2 351 2 þ 2 þ oðε3 Þ; þ − π ν− ν ð4:25iÞ 2240 16 32 j 2 iMH − iADM ¼ t t ADM f MH ϕ − fϕ gMH ϕ iMH ϕ − iADM ϕ − gADM ϕ pffiffiffiffiffiffiffiffiffiffi ε2 ð1 − jÞ3=2 7 5 1 49709 21 2 445 3 1−j ν − νþ − þ π þ ν ¼ νþε j 128 32 j 13440 128 128 32 j2 1 100783 21 2 847 þ 2 − π − ν þ oðε3 Þ; 26880 128 256 j ð4:25jÞ where l0 ≡ dl=du can be computed from Eqs. (4.19b) and (4.20). We first perform the orbit averaging in MH coordinates. ε3 ð1 − jÞ2 ¼ νð149 − 198νÞ þ oðε3 Þ; 384 j3 ð4:25kÞ 1. Orbital average in MH coordinates ADM hMH ¼ ϕ − hϕ 3 5=2 ε ð1 − jÞ 256 j3 νð1 − 5νÞ þ oðε3 Þ: ð4:25lÞ Notice, in agreement with the comment made earlier in Sec. IV B 2, that the MH coordinates differ from the ADM coordinates at leading 2PN order. D. Orbital average of the redshift We are finally in a position to compute the generalized redshift 1 hUi ≡ Tr Z Tr 0 hUi 1 ¼ Tr Z 0 Tr dt 1 ¼ UðtÞ 2π Z 0 2π ð4:26Þ UðτÞdτ; dl 1 ¼ UðlÞ 2π Z 0 2π l0 ðuÞ du; UðuÞ ð4:27Þ 6 6 4 X X l0 αN ðet ; εÞ ln ð1 − et cos uÞ þ βN ðet ; εÞ : ¼ N U N¼−1 ð1 − et cos uÞ ð1 − et cos uÞN N¼2 ð4:28Þ which coincides with the ratio T r =T r of the coordinate time period T r and the proper time period T r of the radial motion.6 The averaged redshift (4.26) can be written in the convenient alternative forms7 −1 Using the generalized QK representation (4.18)–(4.20), (4.22)–(4.23) and (4.25), the variables r, r_ and v2 ¼ r_ 2 þ r2 ϕ_ 2 that enter the expression (4.6)–(4.7) and (4.13) for the redshift in MH coordinates can be expressed as functions of the binding energy ε, the time eccentricity et ≡ eMH t , and the eccentric anomaly u. The integrand in Eq. (4.27) then reads Beware that, although we are using the same symbol to denote the generalized redshift in Eqs. (2.17) and (4.26), the former definition is restricted to linear order in the mass ratio, while the latter holds for any q. 7 Notice the simple relation hUiτ h1=Uit ¼ 1, where h·iτ (resp. h·it ) denotes an averaging over one radial period with respect to the proper time τ (resp. the coordinate time t). The computation of the coefficients αN and βN is straightforward, but the resulting expressions are too cumbersome to be reported here. The integral in (4.27) is readily performed thanks to the following formulas, which are valid for all integers N ≥ 1: Z 1 2π du I N ðeÞ ≡ 2π 0 ð1 − e cos uÞN ð−ÞN−1 dN−1 1 ; ffiffiffiffiffiffiffiffiffiffiffiffiffiffi p ¼ ð4:29aÞ N−1 ðN − 1Þ! dy y2 − e2 y¼1 Z 1 2π ln ð1 − e cos uÞ log du I N ðeÞ ≡ 2π 0 ð1 − e cos uÞN ð−ÞN−1 dN−1 Yðy; eÞ ¼ ; ð4:29bÞ ðN − 1Þ! dyN−1 y¼1 where 124014-20 COMPARISON BETWEEN SELF-FORCE AND POST- … pffiffiffiffiffiffiffiffiffiffiffiffi2ffi 1 1−e þ1 Yðy; eÞ ≡ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ln 2 y2 − e2 p ffiffiffiffiffiffiffiffiffiffiffiffi ffi 1 − e2 − 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi : þ 2 ln 1 þ y þ y2 − e2 PHYSICAL REVIEW D 91, 124014 (2015) ð4:30Þ MH 2 MH 3 MH 4 4 hUi ¼ 1 þ U MH N ε þ U 1PN ε þ U 2PN ε þ U 3PN ε þ oðε Þ; We note that the logarithmic contributions in (4.28) arise at 3PN order from those terms proportional to ln ðr=r0 Þ and ln ðr=r01 Þ in Eq. (4.14). Indeed, combining Eqs. (4.19a), (4.22d), (4.22e), (4.25b) and (4.25c), one finds r a ln ¼ ln r þ ln ð1 − et cos uÞ þ Oðc−2 Þ; ð4:31aÞ r0 r0 r a ln 0 ¼ ln 0r þ ln ð1 − et cos uÞ þ Oðc−2 Þ: ð4:31bÞ r1 r1 Hence, some coefficients αN in (4.28) depend on the regularization constants r0 and r01 through ln ðar =r0 Þ and ln ðar =r01 Þ. However, when averaged over one radial period, these terms cancel out from the final expression, because they appear only through the vanishing combination 2I 2 ðeÞ − 5I 3 ðeÞ þ 3ð1 − e2 ÞI 4 ðeÞ ¼ 0: Implementing all the above integrations, the expression (4.6)–(4.7) and (4.13) for the redshift in MH coordinates can be averaged over an orbit. Up to 3PN order, the generalized redshift (4.26) then takes the form ð4:33Þ where the PN coefficients depend on thepsymmetric ffiffiffiffiffiffiffiffiffiffiffiffiffi mass ratio ν, the reduced mass difference Δ ¼ 1 − 4ν, and the time eccentricity et in MH coordinates (hence, et ≡ eMH t ). They read 3 3 ν U MH N ¼ þ Δ− ; 4 4 2 3 3 45 9 3 þ 3Δ ν − Δν þ pffiffiffiffiffiffiffiffiffiffiffiffiffi ; U MH 1PN ¼ − − Δ − 4 4 16 16 1 − e2t U MH 2PN ¼ ð4:32Þ The final expression for hUiðε; eMH t Þ is thus free of the regularization constants r0 and r01 . ð4:34aÞ ð4:34bÞ 1 Δ 111 15 75 3 þ þ ν − Δν þ ν2 þ Δν2 8 8 16 16 32 32 21 21 57 15 1 þ Δ þ ν − Δν þ 3ν2 pffiffiffiffiffiffiffiffiffiffiffiffiffi − 2 2 4 4 1 − e2t 41 41 37 37 1 þ Δ − ν − Δν þ 5ν2 ; þ 2 2 4 4 ð1 − e2t Þ3=2 43 119 93 15 45 νþ Δν − ν2 þ Δν2 − ν3 4 16 32 8 64 405 405 1419 525 3 2 15 3 1 2 þ Δþ ν− Δν − 3ν − Δν − ν pffiffiffiffiffiffiffiffiffiffiffiffiffi þ 16 16 16 16 4 4 1 − e2t 1 45 45 1 þ Δ − 9ν − 9Δν þ − ð27 þ 27Δ þ 18ν − 18ΔνÞ 2 2 2 1 − et ð1 − e2t Þ2 1467 1467 3281 287 2 9185 287 2 − Δþ − π νþ − π Δν þ − 8 8 48 256 48 256 4193 41 2 2 415 2 107 3 1 þ − þ π ν − Δν þ ν 48 64 16 4 ð1 − e2t Þ3=2 873 873 861 2 861 2 þ Δ þ −278 þ π ν þ −278 þ π Δν þ 4 4 256 256 695 123 2 2 135 2 51 3 1 þ − π ν þ Δν − ν : 4 64 4 2 ð1 − e2t Þ5=2 ð4:34cÞ U MH 3PN ¼ − ð4:34dÞ For notational simplicity we did not add a label on et to indicate that it is the time eccentricity in MH coordinates. (No such label is required over ε, which is gauge invariant.) This point should be remembered when comparing expressions derived in different gauges, as we shall do next. 2. Orbital average in ADM coordinates We shall now perform an independent calculation in ADM coordinates. We start from the expression for the redshift in ADM coordinates, as given by (4.6)–(4.7) and (4.17), employ the appropriate QK parametrization and perform the orbital 124014-21 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) averaging as outlined above. We find that the form (4.28) is obtained also in the ADM case, with the same coefficients βN but different coefficients αN in general. The result for the generalized redshift in ADM coordinates is of the form 2 ADM 3 ADM 4 4 ε þ U ADM hUi ¼ 1 þ U ADM N 1PN ε þ U 2PN ε þ U 3PN ε þ oðε Þ; ð4:35Þ ), and read where the various coefficients depend on ν, Δ, and the time eccentricity in ADM coordinates (hence, et ≡ eADM t 3 3 ν U ADM ¼ þ Δ− ; N 4 4 2 ð4:36aÞ 3 3 45 9 3 þ 3Δ ν − Δν þ pffiffiffiffiffiffiffiffiffiffiffiffiffi ; U ADM 1PN ¼ − − Δ − 4 4 16 16 1 − e2t ð4:36bÞ 1 Δ 111 15 75 3 U ADM þ ν − Δν þ ν2 þ Δν2 2PN ¼ þ 8 8 16 16 32 32 21 21 57 15 1 þ Δ þ ν − Δν þ 3ν2 pffiffiffiffiffiffiffiffiffiffiffiffiffi − 2 2 4 4 1 − e2t 41 41 37 37 1 þ Δ − ν − Δν þ 5ν2 ; þ 2 2 4 4 ð1 − e2t Þ3=2 43 119 93 15 45 νþ Δν − ν2 þ Δν2 − ν3 4 16 32 8 64 405 405 1419 525 3 2 15 3 1 2 þ Δþ ν− Δν − 3ν − Δν − ν pffiffiffiffiffiffiffiffiffiffiffiffiffi þ 16 16 16 16 4 4 1 − e2t 1 45 45 1 þ Δ − 9ν − 9Δν þ − ð27 þ 27Δ þ 18ν − 18ΔνÞ 2 2 1 − e2t ð1 − e2t Þ2 1461 1461 3893 287 2 9797 287 2 − Δþ − π νþ − π Δν þ − 8 8 48 256 48 256 4193 41 2 2 415 2 107 3 1 þ − þ π ν − Δν þ ν 48 64 16 4 ð1 − e2t Þ3=2 435 435 1163 861 2 1163 861 2 þ Δþ − þ π νþ − þ π Δν þ 2 2 4 256 4 256 695 123 2 2 135 2 51 3 1 þ − π ν þ Δν − ν : 4 64 4 2 ð1 − e2t Þ5=2 ð4:36cÞ U ADM 3PN ¼ − Although the coefficients (4.34) and (4.36) coincide ADM through 2PN order, the 3PN coefficients U MH 3PN and U 3PN are different. A useful internal check of the PN calculations of the generalized redshift in MH and ADM coordinates is the verification that the equality of Eqs. (4.33)–(4.34) and (4.35)–(4.36) holds if and only if the time eccentricities eMH and eADM are related t t by eMH t ¼ eADM t 1− 1 þ 17ν ε2 3 þ Oðε Þ : 1 − ðeADM Þ2 4 t ð4:37Þ This relation is in perfect agreement with what is predicted from using different QK representations of the motion, namely, Eq. (4.22d) together with (4.25b). ð4:36dÞ E. Gauge-invariant formulations To compare the analytical PN predictions with the numerical results of the GSF calculation (Sec. III), it is best to use a coordinate-invariant relationship. We shall thus replace the coordinate-dependant time eccentricity et in favor of the coordinate-invariant angular momentum variable j. Substituting the PN expansion (4.22d) into Eq. (4.36), or alternatively Eqs. (4.22d) and (4.25b) into (4.34), we get hUi ¼ 1 þ U N ε þ U 1PN ε2 þ U 2PN ε3 þ U 3PN ε4 þ oðε4 Þ; ð4:38Þ where 124014-22 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) UN ¼ 3 3 ν þ Δ− ; 4 4 2 3 3 45 9 3 þ 3Δ U 1PN ¼ − − Δ − ν − Δν þ pffiffi ; 4 4 16 16 j 1 Δ 111 15 75 3 U 2PN ¼ þ þ ν − Δν þ ν2 þ Δν2 8 8 16 16 32 32 33 33 135 9 1 2 p ffiffi þ Δþ ν − Δν þ 3ν − 8 8 8 8 j 35 35 25 25 1 þ þ Δ − ν − Δν þ 5ν2 3=2 ; 2 2 4 4 j 43 119 93 15 45 νþ Δν − ν2 þ Δν2 − ν3 4 16 32 8 64 297 297 3819 1509 453 2 459 2 9 3 1 þ Δþ ν− Δν þ ν − Δν − ν pffiffi þ 128 128 64 64 128 128 8 j 9 9 1 − þ Δ þ 27ν − 9Δν 2 2 j 945 945 17 287 2 10063 287 2 − þ Δþ þ π νþ − þ π Δν 16 16 96 256 96 256 4471 41 2 2 65 2 85 3 1 þ − π ν þ Δν − ν 96 64 32 8 j3=2 693 693 875 861 2 875 861 2 þ Δþ − þ π νþ − þ π Δν þ 4 4 4 256 4 256 271 123 2 2 21 2 21 3 1 þ : − π ν þ Δν − ν 2 64 2 2 j5=2 ð4:39aÞ ð4:39bÞ ð4:39cÞ U 3PN ¼ − ð4:39dÞ Since the relationship hUiðε; jÞ is coordinate-invariant, it is physically meaningful. However, the binding energy E and angular momentum J are not easily accessible to perturbative GSF calculations, so a direct comparison is not obvious. Thankfully, Eq. (4.38) can also be expressed using the invariant parameters (4.24) defined with respect to the fundamental frequencies Ωr and Ωϕ . Indeed, inverting the PN expansions (4.22a) and (4.22b) yields 5 ν 2 5 5 ν2 5 − 2ν 5 − 3ν 5 2 − − xþ − ν − þ pffi − þ 2 x ε¼x 1þ 4 12 ι 8 8 24 ι ι ι 185 75 25 2 35 3 105 35 7 1 þ − − ν− ν − ν þ − ν − ν2 pffi 192 64 288 5184 8 6 6 ι 2 15 15 ν 1 95 211 41 2 5 2 1 3 3 þ − þ νþ − þ − π ν − ν 3=2 x þ oðx Þ ; 4 4 8 9 96 2 4 ι ι pffi 27 5 5 35 373 41 2 55 − ν þ νι x þ − þ − π ν − ν2 þ ½5 − 2ν ι j¼ιþ 4 2 12 8 16 128 24 2 35 25 ν 115 665 205 2 15 1 2 þ − þ νþ − − π ν − ν2 x þ oðx2 Þ: ιþ 16 48 16 12 128 8 ι 8 ð4:40aÞ ð4:40bÞ We thus have the leading-order relationships x ¼ ε þ Oðc−2 Þ and ι ¼ j þ Oðc−2 Þ. Introducing the expansions (4.40) into Eq. (4.38)–(4.39), our main PN result reads hUi ¼ 1 þ V N x þ V 1PN x2 þ V 2PN x3 þ V 3PN x4 þ oðx4 Þ; 124014-23 ð4:41Þ AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) where the various PN coefficients, which depend on the variable ι as well as on the particle’s masses, read up to 3PN order VN ¼ V 1PN ¼ 3 3 ν þ Δ− ; 4 4 2 3 3 7 5 ν2 3 þ 3Δ 3 3 1 þ Δ − ν − Δν þ þ pffi − þ Δ−ν ; 16 16 2 8 ι 2 2 ι 24 41 41 3 43 99 5 ν3 − Δ − ν − Δν þ ν2 þ Δν2 þ 32 32 4 16 32 32 48 57 57 3 1 3 3 5 ν2 1 þ þ Δ − 22ν − Δν − 2ν2 pffi − þ Δ − 14ν − Δν þ 8 8 2 ι 4 4 2 6 ι 37 37 5 5 1 15 15 5 1 − þ Δ þ ν þ Δν − 5ν2 3=2 þ þ Δ− ν 2; 8 8 2 2 4 4 2 ι ι ð4:42aÞ ð4:42bÞ V 2PN ¼ − 605 605 385 323 565 2 419 2 895 3 25 35 4 − Δþ νþ Δν þ ν þ Δν − ν þ Δν3 þ ν 256 256 48 128 64 128 864 1728 10368 117 117 89 499 1267 2 91 2 5 3 1 þ Δ− ν− Δν þ ν − Δν þ ν pffi þ 128 128 2 16 96 32 6 ι 3 411 411 63 129 297 2 15 2 ν 1 þ þ Δ− νþ Δν − ν − Δν − 16 16 2 8 16 16 8 ι 1755 1755 4339 41 2 497 41 2 þ − − Δþ − π νþ − π Δν 64 64 48 128 24 128 181 41 2 2 115 2 5 3 1 þ − þ π ν þ Δν þ ν 144 96 16 4 ι3=2 1797 1797 355 123 2 355 123 2 þ Δþ − þ π νþ − þ π Δν þ 128 128 16 128 16 128 1321 123 2 2 99 2 33 3 1 þ − π ν − Δν þ ν : 32 64 32 4 ι5=2 ð4:42cÞ V 3PN ¼ − ð4:42dÞ The noncircular nature of the motion only explicitly enters the result at leading 1PN order via the invariant parameter ι. Since we have the qualitative behavior ι ∼ 1 − e2, this suggests that the effect of the eccentricity on hUi will be moderate (at least in the weak-field regime). 1. Circular-orbit limit Another key check of the results (4.39) and (4.42) is provided by the circular-orbit limit. For such orbits, the two constants of the motion are no longer independent variables. Indeed, the angular momentum variable, say j⊙, is related to the energy ε by the 3PN gauge-invariant expansion [72] 9 ν 81 ν2 2 j⊙ ¼ 1 þ þ εþ − 2ν þ ε 4 4 16 16 945 7699 41 2 ν2 ν3 3 þ − þ π νþ þ þ ε þ oðε3 Þ: 64 192 32 2 64 ð4:43Þ It can be checked that the eccentricities et , er , eϕ all vanish when j is replaced by (4.43) in Eqs. (4.22d)–(4.22f) and (4.25b)–(4.25d). The invariant result (4.38)–(4.39) then reduces to 124014-24 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) 3 3 ν 9 9 45 9 þ Δ− εþ þ Δ − ν − Δν ε2 U⊙ ¼ 1 þ 4 4 2 4 4 16 16 81 81 265 103 139 2 3 2 þ Δ− ν− Δν þ ν þ Δν ε3 þ 8 8 16 16 32 32 891 891 3809 287 2 4945 287 2 þ Δ− − π ν− − π Δν þ 16 16 24 128 48 128 7727 41 2 2 143 2 205 3 4 þ − π ν þ Δν − ν ε þ oðε4 Þ: 96 32 16 64 ð4:44Þ Setting et → 0 in Eq. (4.34) or (4.36) yields the same expression. We then replace the constant of the motion ε in favor of the frequency-related parameter x [recall Eq. (4.24)], using the well-known 3PN-accurate expression for the binding energy as a function of the circular-orbit frequency, namely, [see, e.g., Eq. (232) of Ref. [14]] 3 ν 27 19 ν2 2 ε⊙ ¼ x 1 þ − − xþ − þ ν− x 4 12 8 8 24 675 34445 205 2 155 2 35 3 3 3 þ − π ν− ν − ν x þ oðx Þ : þ − 64 576 96 96 5184 ð4:45Þ Finally, replacing ε in (4.44) using (4.45), we recover the known 3PN result for the circular-orbit redshift (see Eq. (4.10) of Ref. [16]): 3 3 ν 27 27 5 5 ν2 2 þ Δ− xþ þ Δ − ν − Δν þ U⊙ ¼ 1 þ x 4 4 2 16 16 2 8 24 135 135 37 67 115 2 5 ν3 3 2 þ Δ − ν − Δν þ ν þ Δν þ þ x 32 32 4 16 32 32 48 2835 2835 2183 41 2 12199 41 2 þ þ Δ− − π ν− − π Δν 256 256 48 64 384 64 17201 41 2 2 795 2 2827 3 25 35 4 4 3 þ − π ν þ Δν − ν þ Δν þ ν x 576 192 128 864 1728 10368 þ oðx4 Þ: ð4:46Þ Interestingly, at Newtonian order, the averaged redshift hUi along an eccentric orbit has the same functional form as U ⊙ in the case of a circular orbit. This shows that the effect of the eccentricity cancels out at Newtonian order, because of the orbital averaging. Alternatively, we can also combine Eqs. (4.40b), (4.43), (4.45) to obtain the PN expansion of the invariant relation ι⊙ ðxÞ in the circular-orbit limit, namely, 9 7 9 397 41 2 28 2 2 ι⊙ ¼ 1 þ − þ ν x þ − þ − π ν þ ν x þ oðx2 Þ; 2 3 4 12 32 9 ð4:47Þ and introduce this expression into (4.41)–(4.42) to recover (4.46). Our third and last check of the correctness of the formula (4.42) will be to recover the known result in the testparticle limit. 2. Extreme mass-ratio limit The 3PN result (4.41)–(4.42) is valid for any mass ratio q ¼ m1 =m2 . To extract from this result the contribution due to the conservative piece of the GSF, we introduce an alternative set of dimensionless coordinate-invariant parameters, better suited than ðx; ιÞ to the extreme mass-ratio limit q ≪ 1: 124014-25 AKCAY et al. Gm2 Ωϕ 2=3 y≡ ; c3 PHYSICAL REVIEW D 91, 124014 (2015) λ≡ 3y : k ð4:48Þ We substitute the relations x ¼ yð1 þ qÞ2=3 and ι ¼ λð1 þ qÞ2=3 in (4.41)–(4.42), and expand in powers of the mass ratio q, neglecting terms of Oðq3 Þ or higher. The 3PN result for the sum of the test mass, GSF and post-GSF contributions reads hUi ¼ hUi0 þ qhUigsf þ q2 hUip-gsf þ Oðq3 Þ; ð4:49Þ where 3 3 6 3 2 41 57 3 37 15 3 þ pffiffiffi − y þ − þ pffiffiffi − − 3=2 þ 2 y hUi0 ¼ 1 þ y þ 2 8 16 4 λ 2λ 4λ 2λ λ λ 605 117 411 1755 21 1797 20 4 þ pffiffiffi þ − y þ oðy4 Þ; þ − þ þ − 128 64 λ 8λ 32λ3=2 4λ2 64λ5=2 λ3 2 2 14 16 5 5 3 8 293 36 hUigsf ¼ −y − 4 − y − 6 þ pffiffiffi − þ 3=2 þ 2 y þ − pffiffiffi þ λ λ 3 4 λ λ λ λ λ 1789 41 2 1 56 355 123 2 1 40 4 þ − π þ π y þ oðy4 Þ; − þ − þ 24 64 8 64 λ3=2 λ2 λ5=2 3λ3 2 69 29 16 15 5 275 1533 207 hUip-gsf ¼ y þ 4 − y2 þ þ pffiffiffi − þ 3=2 þ 2 y3 þ þ pffiffiffi − λ 8 4 λ λ λ 24 32 λ 4λ λ 6377 41 2 1 56 2031 369 2 1 40 4 þ − y þ oðy4 Þ: þ π − π þ þ − 96 32 16 64 λ3=2 λ2 λ5=2 3λ3 ð4:50aÞ ð4:50bÞ ð4:50cÞ In the test-particle limit q ¼ 0, we recover the 3PN expansion of the fully relativistic result (2.19) for a geodesic orbit, as derived in App. B. The 3PN prediction (4.50c) could be compared with future calculations of the second-order GSF [73–78]. Finally, we may express the result (4.50b) for the 3PN expansion of the GSF contribution to the generalized redshift by means of the usual parametrization of bound timelike geodesic orbits in Schwarzschild in terms of the semilatus rectum p and eccentricity e (see Sec. II A). Substituting for y and λ from Eqs. (B1) into (4.50b), we find pffiffiffiffi je 2je 123 2 pffiffiffiffi 3=2 2 1 hUigsf ¼ − je π 1þ þ ð5 je − 4je þ 9je − 5je Þ 2 þ 95 − 64 p p p 79 41 2 3=2 27 5=2 1 2 3 −3 þ oðp Þ ; ð4:51Þ − 16je þ − þ π je − 16je − je þ 4je 6 64 2 p3 where je ≡ 1 − e2 . For small eccentricities, we may write hUigsf ¼ a þ be2 þ ce4 þ de6 þ Oðe8 Þ; ð4:52Þ where the weak-field expansions of the coefficients aðpÞ, bðpÞ, cðpÞ and dðpÞ read 1 2 5 121 41 2 1 a¼− − 2− 3− − π þ oðp−4 Þ; p p 3 32 p p4 ð4:53aÞ 1 4 7 5 41 2 1 þ π þ oðp−4 Þ; ð4:53bÞ b¼ þ 2þ 3− p p 3 32 p p4 2 1 705 123 2 1 c¼− 2þ 3þ þ oðp−4 Þ; ð4:53cÞ − π 8 256 p 4p p4 5 475 41 2 1 d¼− 3þ − þ π þ oðp−4 Þ; 12 128 2p p4 ð4:53dÞ and higher-order terms in the eccentricity all contribute at leading 2PN order. V. COMPARISON OF POST-NEWTONIAN AND SELF-FORCE RESULTS In Fig. 1 we plot our data for hUigsf as a function of p for a sample e ¼ f0.1; 0.2; 0.3; 0.4g of eccentricities. We show, superposed, the corresponding 1PN, 2PN and 3PN predictions from Eq. (4.51). The insets display the relative differences between the GSF data and the successive PN approximations. We make the following observations: (i) There is an excellent agreement between the numerical GSF results and the analytical PN prediction at 124014-26 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) “large” p, in what should be considered a very strong test of both calculations. This is a first demonstration of such an agreement for noncircular orbits. (ii) The PN series appears to converge uniformly to the GSF result at any p for any fixed e in our survey, at least through 3PN order. (iii) The 3PN formula reproduces the GSF results extremely well even in what might be considered a “strong-field” regime: at p ¼ 10 it does so to within ∼1% for e ¼ 0.1 and to within a few percent for e ¼ 0.4; at p ¼ 20 the agreement is already at the level of one part in a thousand. We can make the comparison more quantitative by attempting to extract the large-p behavior of the numerically computed function hUigsf ðp; eÞ. Our strategy will be to fit the numerical data against the PN model (4.51), leaving the numerical coefficients as unknown fitting parameters, later to be compared with the analytically known values. Given the relative sparseness of data available, we shall not attempt a simultaneous fit over p and e, but rather fit over each of the two dimensions separately, as described below. We will follow a “marginalization” procedure, whereby each of the PN orders is fitted for in turn, assuming the analytic values of all terms at lower PN order. Since the circular limit of hUigsf has been computed previously at great accuracy [17,18,22,34], we are able to accurately “remove” the circular (e-independent) part of hUigsf from the data, fitting only for the e-dependent residue. This should allow to fit the eccentricity-related terms of interest here with greater accuracy. Let us now describe this procedure in more detail. We assume the e-expanded form (4.52) of the full PN expression (4.51). The term aðpÞ is the circular-orbit limit of hUigsf , which has been computed to at least ten significant figures in Refs. [17,18,22,34]. By subtracting off these numerical data from ours, we construct a new data set for the difference ðeÞ hUigsf ≡ hUigsf − aðpÞ ¼ bðpÞe2 þ cðpÞe4 þ : ð5:1Þ We assume that the functions bðpÞ; cðpÞ; … admit expansions in p−1 as in Eqs. (4.53), but pretend that the PN coefficients are unknown: b ¼ p−1 þ b1 p−2 þ b2 p−3 þ ; c ¼ c1 p−2 þ c2 p−3 þ ; ð5:2Þ where subscripts are mnemonics for the PN order at which coefficients occur, and we have fixed the “Newtonian,” 1=p term of bðpÞ at its known value of unity. Our goal is to determine the coefficient bn ; cn ; … from the numerical data ðeÞ for hUigsf . To this end, we first prepare subsets of data where in each subset p is fixed and e varies. We fit each TABLE IV. Best-fit values for the PN coefficients bn and cn [Eqs. (5.1) and (5.2)] as extracted from the numerical data, compared to their known exact values. Parenthetical figures are estimated fitting uncertainties in the last displayed decimals, obtained by varying the value of the truncation index N in the fitting model [e.g., Eq. (5.3) for bðpÞ]. The exact value of b3 is −ð5=3 þ 41π 2 =32Þ. Coefficient b1 b2 b3 c1 Estimate Exact result þ4.0002ð8Þ þ7.02ð3Þ −14.5ð4Þ −2.00ð1Þ þ4 þ7 −14.312… −2 subset with respect to e using the model (5.1), including terms through Oðe6 Þ. This yields three one-dimensional data sets, representing bðpÞ, cðpÞ and dðpÞ. Focusing first on the data set for bðpÞ, we fit it against the PN model bðpÞ ¼ p−1 þ N X p−ðnþ1Þ ðbn þ blog n ln pÞ; ð5:3Þ n¼1 log log in which blog 1 ¼ b2 ¼ b3 ¼ 0, since logarithmic terms are known not to occur before the 4PN order [79,80].8 The truncation order N is left as a control parameter; by varying it we obtain a rough estimate of the numerical uncertainty in the fitted values of the parameters. We apply a marginalization procedure, whereby to determine bn we set all bn0 <n at their known analytic values. We use this procedure to estimate the values of b1 , b2 and b3 , and we later similarly determine c1 . Our results are shown in Table IV, alongside the known analytic values for these parameters. We see a good agreement through 3PN order in the Oðe2 Þ term, and at 1PN order in the Oðe4 Þ term. Unfortunately, the accuracy of our current code (and its limited utility at e ≳ 0.4) does not seem to allow us an accurate extraction of bn≥4 , cn≥2 , or any of the blog n ’s. The reason for this can be appreciated from Fig. 2, where we compare the amplitudes of the 3PN and 4PN terms with the ðeÞ amplitude of numerical noise in our hUigsf data. Note that, while the “signal” from the b3 term lies well above the noise, the c2 signal is buried deep inside it. Since our data is limited to relatively small eccentricities, it is clear why we have less “handle” on the cn [Oðe4 Þ] terms than on the bn [Oðe2 Þ] terms. 8 The form of the circular-orbit limit, in which the PN expansion of hUigsf is known analytically up to a very high order [17–20,81], suggests that the function bðpÞ could also involve powers of ln p. However, those would contribute at even higher orders than we consider here, so we do not include them in the PN model (5.3). 124014-27 AKCAY et al. 10 2 10 4 10 10 10 PHYSICAL REVIEW D 91, 124014 (2015) 0.100 Error Error b3 e2 p 4 c2 e p 4 (b4 blog 4 ln 6 b4 e2 p blog 4 8 b3 e2 p 0.001 3 2 c2 e p 10 5 5 b4 e2 p blog 4 5 10 7 10 9 4 3 2 (b4 blog 4 ln p) e p 5 p) e p ln p e2 p 4 5 5 ln p e2 p 5 10 20 40 60 80 100 120 140 20 p 40 60 80 100 120 140 p FIG. 2 (color online). Absolute magnitude of various PN terms (“signal”) compared to the magnitude of numerical error in the data (“noise”), shown as a function of p for e ¼ 0.1 (left panel) and e ¼ 0.3 (right panel). The red (upper solid) curve shows the 3PN term of the Oðe2 Þ piece of hUigsf , and the blue (lower solid) curve shows the 2PN term of the Oðe4 Þ piece. Comparison with the magnitude of numerical noise (black dots) suggests that b3 should be easily discernible while c2 might not. This is confirmed by attempting to fit the data against PN models, as detailed in the text. The dashed curve estimates the amplitude of the 4PN term of the Oðe2 Þ piece of hUigsf , which is not known analytically. We used here the values b4 ¼ −1500 and blog 4 ¼ 250 chosen from the middle of the estimated range shown in Eqs. (5.4). This 4PN signal appears to lie just over the noise and is detectable. However, as it can be seen from the near overlap of the densely dashed (green) and sparsely dashed (brown) curves, the jb4 j and blog 4 terms become almost equal in magnitude as p . increases, making it very difficult to extract the individual values of b4 and blog 4 Figure 2 also suggests that we might have just enough “signal” coming from the Oðe2 Þ terms at 4PN to allow a rough estimation of the coefficients b4 and blog 4 , which are not known analytically. We have experimented fitting to a large number of models of the form (5.3), where all the analytically known coefficients are prespecified, and varying both the cutoff N and the number of nonzero logarithmic terms. We find that fitting uncertainties are almost as large as the fitted values themselves. However, we are able to confidently constrain the values of b4 and blog to lie within the 4 following ranges: −2000 ≲ b4 ≲ −1000; ð5:4aÞ þ150 ≲ blog 4 ≲ þ350: ð5:4bÞ Future analytic calculations of the 4PN terms may be checked against these predictions. Our current code does not allow the determination of unknown PN coefficients related to eccentricity with any greater accuracy. To improve on our predictions would require (i) to push the reach of the computation to higher eccentricities and larger p, and at the same time (ii) to reduce the numerical error in the calculation of hUigsf . Some improvement may be achieved using the method of Ref. [56], which is a slightly more advanced variant of our method. More significant improvements may have to await the development of eccentric-orbit GSF codes based on the Teukolsky equation [58,82]. We expect such codes to start delivering accurate numerical results in the very near future. ACKNOWLEDGMENTS We are grateful to Maarten van de Meent for providing us with unpublished comparison data generated by a new GSF code now being developed by him. S. A. also thanks Maarten van de Meent and Haris Markakis for many useful discussions. The research leading to these results received funding from the European Research Council under the European Union’s Seventh Framework Programme (FP7/20072013)/ERC Grant Agreement No. 304978. S. A. and L. B. acknowledge additional support from STFC through Grant No. PP/E001025/1. S. A.’s work is further supported by the Irish Research Council, funded under the National Development Plan for Ireland. A. L. T. acknowledges support from a Marie Curie FP7 Integration Grant under the 7th European Union Framework Programme (PCIG13-GA-2013-630210). N. S. acknowledges the support of the Grand-in-Aid for Scientific Research (No. 25800154). N. W. gratefully acknowledges support from a Marie Curie International Outgoing Fellowship (PIOF-GA-2012-627781) and the Irish Research Council, which is funded under the National Development Plan for Ireland. 124014-28 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) APPENDIX A: EQUIVALENCE BETWEEN OUR EXPRESSION FOR hUigsf AND BS2011’S BS2011 give their formula for hUigsf in their Eq. (84). Adjusting notation and rearranging the terms, their expression reads ΔT r qhUigsf ¼ α þ ðhUi0 þ Cr Ωr þ Cϕ Ωϕ Þ T r0 ΔΦ ΔT r − Cϕ Ωϕ − hUi0 ; ðA1Þ Φ0 T r0 where Cr ≡ ∂hUi0 ; ∂Ωr Cϕ ≡ ∂hUi0 ; ∂Ωϕ Cϕ ¼ hUi0 Cr ¼ ½ðE − LΩϕ ÞhUi0 − 1; Ωr ðA3aÞ ðA3bÞ which shall be derived below. Substituting these Cr and Cϕ into Eq. (A1) and using T r0 ¼ T r0 =hUi0 , we obtain qhUigsf ¼ αEhUi20 þ hUi0 ðEΔT r − LΔΦ − ΔT r Þ: T r0 ðA4Þ Now, writing ðdτ=dχÞ2 ¼ −gαβ ðdxα =dχÞðdxβ =dχÞ and perturbing linearly with Δ, holding fixed p, e and χ (hence, also, rp and drp =dχ), as in BS2011, we find 1 dτ Δðdτ=dχÞ ¼ EΔðdt=dχÞ − LΔðdϕ=dχÞ − ĥRuu 0 ; ðA5Þ 2 dχ which, upon integrating over a radial period, gives 1 ΔT r ¼ EΔT r − LΔΦ − T r0 hĥRuu i: 2 hUi−1 0 ¼ E − Ωϕ L − Ωr J r ; ðA6Þ Substituting in Eq. (A4) now reproduces our Eq. (2.25) for hUigsf . It remains to establish the relations (A3a) and (A3b). This can be achieved by manipulating the explicit ellipticintegral representations of Ωϕ ; Ωr and hUi0 , given in BS2011, but this approach involves much ungainly algebra and will not be presented here. A much neater derivation ðA7Þ H R where Jr ≡ ð2πÞ−1 u0r dr ¼ ð2πE 0 Þ−1 0T r ður0 Þ2 dt is the invariant action variable (per mass m1 ) associated with the radial motion [84]. This relation is the Schwarzschild reduction of Eq. (3.4) of Ref. [83]. In addition, we require a relation between the partial derivatives of E, L and J r with respect to Ωi . The necessary relation follows most directly from the general variational formula (“first law”), ðA2Þ and ΔX denotes the GSF correction to a quantity X, holding fixed p and e, rather than the invariant frequencies Ωi . BS2011 give explicit expressions for ΔT r, ΔΦ and ΔT r in terms of GSF quantities, but these will not be needed here. We observe that the expression (A1) is much more complicated than our result, Eq. (2.25). Our goal here is to show that the two expressions are, in fact, identical. To this end, we note the two key relations, LhUi20 ; uses general results derived from the Hamiltonian formulation of geodesic motion in Kerr spacetime [83]. Start by averaging uα0 u0α ¼ −1 with respect to t over a radial period of the geodesic orbit, to obtain δE ¼ Ωϕ δL þ Ωr δJ r ; ðA8Þ established in [83] [this form is the reduction of Eq. (3.5) therein to Schwarzschild spacetime, with a fixed black-hole mass m2 , and with suitable notational adjustments]. Here δE, δL and δJr correspond to an arbitrary variation of a geodesic with frequencies Ωi onto a nearby geodesic. If we regard E, L and Jr as functions of Ωi , we obtain ∂E ∂L ∂J − Ωϕ − Ωr r ¼ 0: ∂Ωi ∂Ωi ∂Ωi ðA9Þ Taking the partial derivative of (A7) with respect to Ωϕ and using (A9) immediately leads to (A3a). Equation (A3b), in turn, is obtained by taking the derivative of (A7) with respect to Ωr , then using Eq. (A9), and finally substituting for J r from (A7). The above establishes the equivalence of our simple expression (2.25) and the BS2011 result (A1). The simplification obtained here owes itself primarily to the two key relations (A3a) and (A3b), which have unfortunately gone unnoticed (by two of us) in BS2011. APPENDIX B: POST-NEWTONIAN EXPANSION OF hUi0 Here we consider a test mass on a bound geodesic orbit around a nonspinning black hole of mass m2 and obtain the PN expansion of the relationship hUi0 ðΩr ; Ωϕ Þ. This calculation provides a powerful check of our PN result (4.41)–(4.42), because it is based on a different formalism, it makes use of an alternative parametrization of the motion, and it is performed using a different coordinate system. Since the relationships (2.7)–(2.10) cannot be inverted analytically to yield the expressions for the parameters p and e as functions of the frequencies Ωr ¼ 2π=T r0 and Ωϕ ¼ Φ0 =T r0 , we shall work perturbatively, expanding all quantities in powers of the small parameter 1=p. From T r0 and Φ0 we define the invariant parameters y ≡ ðm2 Ωϕ Þ2=3 and λ ≡ 3yðΦ0 =2π − 1Þ−1 [recall Eq. (4.48)]. Expanding the formulas (2.7)–(2.10) up to 3PN order, we obtain 124014-29 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) pffiffiffiffi 1 je 2ð1 − je Þ 17 þ ½1 − je þ 5je ½je − je y¼ 1þ p 2 p p2 133 40 3 1 3=2 5=2 2 −3 − 48je − 35je þ 27je þ 25je − je þ þ oðp Þ ; 3 3 p3 11 7 1 75 23 73 2 1 3=2 λ ¼ je 1 − þ j þ − þ je − 5je þ je 4 4 e p 16 8 16 p2 1849 849 45 3=2 17 2 95 5=2 2341 3 1 −3 þ j − je − je þ je − j þ − þ oðp Þ ; 96 64 e 4 16 4 192 e p3 ðB1aÞ ðB1bÞ where we introduced the notation je ≡ 1 − e2 . In the limit of vanishing eccentricity, e → 0, we have the simple relation y ¼ p−1 þ oðp−4 Þ. Actually, we know that for circular orbits the relation y ¼ 1=p holds exactly, such that in Schwarzschild coordinates the semimajor axis coincides with an invariant measure of the orbital radius. [This, however, is no longer true at OðqÞ in the GSF approximation.] For circular orbits, the 3PN-accurate relationship between the invariants y and λ then reads 9 9 27 λ ¼ 1 − y − y2 − y3 þ oðy3 Þ: 2 4 4 ðB2Þ Inverting the relations (B1) yields expressions for the semilatus rectum p and eccentricity e (or equivalently je ¼ 1 − e2 ) as functions of the invariant parameters y and λ. Up to 3PN order, we find 1 y 1 19 9 5 151 2 ¼ 1þ − yþ − þ y p λ 4 4λ 16 16λ 8λ2 65 5 25 1 2255 3 3 p ffiffi ffi y þ oðy Þ ; þ − þ þ − ðB3aÞ 64 4 λ 64λ 4λ2 32λ3 7 11 5 63 13 je ¼ λ 1 þ þ y þ 2 þ pffiffiffi þ − y2 4 4λ λ 16λ 16λ2 5 145 221 95 289 263 3 3 y þ oðy Þ : þ þ þ pffiffiffi þ − þ ðB3bÞ 6 8 λ 32λ 8λ3=2 64λ2 192λ3 We now have all the pieces required to compute the relation hUi0 ðy; λÞ up to the required PN order. The generalized redshift is defined as Z T Z 2π r0 1 1 dt0 dχ −1 hUi0 ≡ ut0 ðτ0 Þdτ0 ¼ ; ðB4Þ T r0 0 T r0 0 dχ ut0 ðχÞ where T r0 is the proper time period of the radial motion. From the expressions (2.4), (2.5), (2.7a) and (2.8), we find pffiffiffiffi 21 pffiffiffiffi je 3 1 55 2 1 3=2 hUi0 ¼ 1 þ þ 6 je − je þ 23 je − 6je − 12je þ je 8 p 16 p 2 p2 249 pffiffiffiffi 75 5=2 525 3 1 3=2 2 −3 je − 24je − 105je þ 12je þ je − j þ þ oðp Þ : ðB5Þ 2 4 128 e p3 Finally, substituting for ðp; eÞ in terms of ðy; λÞ in Eq. (B5), using (B3), we obtain the 3PN-accurate coordinate-invariant relation, 3 3 6 3 2 41 57 3 37 15 3 p ffiffi ffi p ffiffi ffi hUi0 ¼ 1 þ y þ þ − y þ − þ − − y þ 2 8 16 4 λ 2λ 4λ3=2 2λ2 λ λ 605 117 411 1755 21 1797 20 4 y þ oðy4 Þ: þ − þ 2þ − ðB6Þ þ pffiffiffi þ − 3=2 128 64 λ 8λ 32λ 4λ 64λ5=2 λ3 124014-30 COMPARISON BETWEEN SELF-FORCE AND POST- … PHYSICAL REVIEW D 91, 124014 (2015) In the circular-orbit limit, we may introduce the PN expansion (B2) for λðyÞ in (B6), expand in powers of y up to the appropriate PN order, and recover the 3PN expansion of the fully relativistic result U ⊙ ¼ ð1 − 3yÞ−1=2 . Although the result (B6) can in principle be extended up to an arbitrarily high PN order, we only need here the 3PN approximation to the exact result. Comparing with the formula (4.50a) derived from our 3PN calculation valid for any mass ratio, we find perfect agreement. [1] J. Aasi et al., arXiv:1304.0670. [2] J. Abadie et al., Classical Quantum Gravity 27, 173001 (2010). [3] L. S. Finn and D. F. Chernoff, Phys. Rev. D 47, 2198 (1993). [4] A. Buonanno and T. Damour, Phys. Rev. D 59, 084006 (1999). [5] A. Buonanno and T. Damour, Phys. Rev. D 62, 064015 (2000). [6] A. Taracchini et al., Phys. Rev. D 89, 061502(R) (2014). [7] T. Damour and A. Nagar, Phys. Rev. D 90, 044018 (2014). [8] D. Bini and T. Damour, Phys. Rev. D 89, 104047 (2014). [9] D. Bini and T. Damour, Phys. Rev. D 90, 024039 (2014). [10] D. Bini and T. Damour, Phys. Rev. D 90, 124037 (2014). [11] L. Barack, Classical Quantum Gravity 26, 213001 (2009). [12] E. Poisson, A. Pound, and I. Vega, Living Rev. Relativity 14, 7 (2011). [13] J. Thornburg, GW Notes 5, 3 (2011). [14] L. Blanchet, Living Rev. Relativity 17, 2 (2014). [15] S. Detweiler, Phys. Rev. D 77, 124026 (2008). [16] L. Blanchet, S. Detweiler, A. Le Tiec, and B. F. Whiting, Phys. Rev. D 81, 064004 (2010). [17] L. Blanchet, S. Detweiler, A. Le Tiec, and B. F. Whiting, Phys. Rev. D 81, 084033 (2010). [18] A. G. Shah, J. L. Friedman, and B. F. Whiting, Phys. Rev. D 89, 064042 (2014). [19] L. Blanchet, G. Faye, and B. F. Whiting, Phys. Rev. D 89, 064026 (2014). [20] L. Blanchet, G. Faye, and B. F. Whiting, Phys. Rev. D 90, 044017 (2014). [21] N. Sago, L. Barack, and S. Detweiler, Phys. Rev. D 78, 124024 (2008). [22] A. G. Shah, T. S. Keidl, J. L. Friedman, D.-H. Kim, and L. R. Price, Phys. Rev. D 83, 064018 (2011). [23] L. Barack and N. Sago, Phys. Rev. D 81, 084021 (2010). [24] L. Barack and N. Sago, Phys. Rev. D 83, 084023 (2011). [25] M. Favata, Phys. Rev. D 83, 024027 (2011). [26] M. Favata, Phys. Rev. D 83, 024028 (2011). [27] A. Le Tiec, A. H. Mroué, L. Barack, A. Buonanno, H. P. Pfeiffer, N. Sago, and A. Taracchini, Phys. Rev. Lett. 107, 141101 (2011). [28] A. Le Tiec, E. Barausse, and A. Buonanno, Phys. Rev. Lett. 108, 131103 (2012). [29] A. Le Tiec et al., Phys. Rev. D 88, 124027 (2013). [30] A. Le Tiec, Int. J. Mod. Phys. D 23, 1430022 (2014). [31] T. Damour, Phys. Rev. D 81, 024017 (2010). [32] L. Barack, T. Damour, and N. Sago, Phys. Rev. D 82, 084036 (2010). [33] E. Barausse, A. Buonanno, and A. Le Tiec, Phys. Rev. D 85, 064010 (2012). [34] S. Akcay, L. Barack, T. Damour, and N. Sago, Phys. Rev. D 86, 104041 (2012). [35] S. R. Dolan, N. Warburton, A. I. Harte, A. Le Tiec, B. Wardell, and L. Barack, Phys. Rev. D 89, 064011 (2014). [36] S. R. Dolan, P. Nolan, A. C. Ottewill, N. Warburton, and B. Wardell, Phys. Rev. D 91, 023009 (2015). [37] L. Wen, Astrophys. J. 598, 419 (2003). [38] R. M. O’Leary, B. Kocsis, and A. Loeb, Mon. Not. Roy. Astron. Soc. 395, 2127 (2009). [39] T. A. Thompson, Astrophys. J. 741, 82 (2011). [40] F. Antonini and H. B. Perets, Astrophys. J. 757, 27 (2012). [41] B. Kocsis and J. Levin, Phys. Rev. D 85, 123005 (2012). [42] J. Samsing, M. MacLeod, and E. Ramirez-Ruiz, Astrophys. J. 784, 71 (2014). [43] J. R. Gair and E. K. Porter, in 9th LISA Symposium, ASP Conference Series, Vol. 467, edited by G. Auger, P. Binétruy, and E. Plagnol (Astronomical Society of the Pacific, San Fransisco, 2013), p. 173. [44] P. Amaro-Seoane et al., GW Notes 6, 4 (2013). [45] P. Amaro-Seoane et al., arXiv:1305.5720. [46] P. Amaro-Seoane, J. R. Gair, A. Pound, S. A. Hughes, and C. F. Sopuerta, arXiv:1410.0958. [47] E. Barausse et al. arXiv:1410.2907. [48] S. Akcay, N. Warburton, and L. Barack, Phys. Rev. D 88, 104009 (2013). [49] C. Darwin, Proc. R. Soc. A 263, 39 (1961). [50] N. Warburton, L. Barack, and N. Sago, Phys. Rev. D 87, 084012 (2013). [51] C. Cutler, D. Kennefick, and E. Poisson, Phys. Rev. D 50, 3816 (1994). [52] S. Detweiler and B. F. Whiting, Phys. Rev. D 67, 024025 (2003). [53] L. Barack and C. O. Lousto, Phys. Rev. D 72, 104026 (2005). [54] L. Barack, A. Ori, and N. Sago, Phys. Rev. D 78, 084021 (2008). [55] L. Barack and N. Sago, Phys. Rev. D 75, 064021 (2007). [56] T. Osburn, E. Forseth, C. R. Evans, and S. Hopper, Phys. Rev. D 90, 104031 (2014). [57] M. van de Meent (private communication). [58] C. Merlin and A. G. Shah, Phys. Rev. D 91, 024005 (2015). [59] K. G. Arun, L. Blanchet, B. R. Iyer, and M. S. S. Qusailah, Phys. Rev. D 77, 064035 (2008). [60] K. G. Arun, L. Blanchet, B. R. Iyer, and S. Sinha, Phys. Rev. D 80, 124018 (2009). 124014-31 AKCAY et al. PHYSICAL REVIEW D 91, 124014 (2015) [61] L. Blanchet, G. Faye, and B. Ponsot, Phys. Rev. D 58, 124002 (1998). [62] L. Blanchet and G. Faye, Phys. Rev. D 63, 062005 (2001). [63] L. Blanchet, T. Damour, and G. Esposito-Farèse, Phys. Rev. D 69, 124007 (2004). [64] V. C. de Andrade, L. Blanchet, and G. Faye, Classical Quantum Gravity 18, 753 (2001). [65] L. Blanchet and B. R. Iyer, Classical Quantum Gravity 20, 755 (2003). [66] R.-M. Memmesheimer, A. Gopakumar, and G. Schäfer, Phys. Rev. D 70, 104011 (2004). [67] T. Damour, P. Jaranowski, and G. Schäfer, Phys. Rev. D 63, 044021 (2001), 66, 029901(E) (2002). [68] T. Damour and N. Deruelle, Ann. Inst. Henri Poincaré 43, 107 (1985). [69] T. Damour and G. Schäfer, Nuovo Cimento B 101, 127 (1988). [70] G. Schäfer and N. Wex, Phys. Lett. A 174, 196 (1993); 177, 461(E) (1993). [71] N. Wex, Classical Quantum Gravity 12, 983 (1995). [72] T. Damour, P. Jaranowski, and G. Schäfer, Phys. Rev. D 62, 044024 (2000). [73] S. Detweiler, Phys. Rev. D 85, 044048 (2012). [74] S. E. Gralla, Phys. Rev. D 85, 124011 (2012). [75] A. Pound, Phys. Rev. Lett. 109, 051101 (2012). [76] A. Pound, Phys. Rev. D 86, 084019 (2012). [77] A. Pound, Phys. Rev. D 90, 084039 (2014). [78] A. Pound and J. Miller, Phys. Rev. D 89, 104020 (2014). [79] J. L. Anderson, R. E. Kates, L. S. Kegeles, and R. G. Madonna, Phys. Rev. D 25, 2038 (1982). [80] L. Blanchet and T. Damour, Phys. Rev. D 37, 1410 (1988). [81] D. Bini and T. Damour, Phys. Rev. D 91, 064050 (2015). [82] A. Pound, C. Merlin, and L. Barack, Phys. Rev. D 89, 024009 (2014). [83] A. Le Tiec, Classical Quantum Gravity 31, 097001 (2014). [84] W. Schmidt, Classical Quantum Gravity 19, 2743 (2002). 124014-32