Secular Evolution of Hierarchical Planetary Systems

Man Hoi Lee and S. J. Peale

Department of Physics, University of California, Santa Barbara, CA 93106

ABSTRACT
arXiv:astro-ph/0304454v1 25 Apr 2003

We investigate the dynamical evolution of coplanar hierarchical two-planet systems

where the ratio of the orbital semimajor axes α = a1 /a2 is small. Hierarchical two-
planet systems are likely to be ubiquitous among extrasolar planetary systems. We
show that the orbital parameters obtained from a multiple Kepler fit to the radial ve-
locity variations of a host star are best interpreted as Jacobi coordinates and that Jacobi
coordinates should be used in any analyses of hierarchical planetary systems. An ap-
proximate theory that can be applied to coplanar hierarchical two-planet systems with a
wide range of masses and orbital eccentricities is the octupole-level secular perturbation
theory, which is based on an expansion to order α3 and orbit-averaging. It reduces the
coplanar problem to one degree of freedom, with e1 (or e2 ) and ̟1 − ̟2 as the relevant
phase-space variables (where e1,2 are the orbital eccentricities of the inner and outer
orbits and ̟1,2 are the longitudes of periapse). The octupole equations show that if
the ratio of the maximum orbital angular momenta, λ = L1 /L2 ≈ (m1 /m2 )α1/2 , for
given semimajor axes is approximately equal to a critical value λcrit , then libration of
̟1 − ̟2 about either 0◦ or 180◦ is almost certain, with possibly large amplitude varia-
tions of both eccentricities. From a study of the HD 168443 and HD 12661 systems and
their variants using both the octupole theory and direct numerical orbit integrations,
we establish that the octupole theory is highly accurate for systems with α . 0.1 and
reasonably accurate even for systems with α as large as 1/3, provided that α is not too
close to a significant mean-motion commensurability or above the stability boundary.
The HD 168443 system is not in a secular resonance and its ̟1 − ̟2 circulates. The
HD 12661 system is the first extrasolar planetary system found to have ̟1 − ̟2 librat-
ing about 180◦ . The secular resonance means that the lines of apsides of the two orbits
are on average anti-aligned, although the amplitude of libration of ̟1 − ̟2 is large.
The libration of ̟1 − ̟2 and the large-amplitude variations of both eccentricities in
the HD 12661 system are consistent with the analytic results on systems with λ ≈ λcrit .
The evolution of the HD 12661 system with the best-fit orbital parameters and sin i = 1
(i is the inclination of the orbital plane from the plane of the sky) is affected by the close
proximity to the 11:2 mean-motion commensurability, but small changes in the orbital
period of the outer planet within the uncertainty can result in configurations that are
not affected by mean-motion commensurabilities. The stability of the HD 12661 system
requires sin i > 0.3.
 –2–

1. INTRODUCTION

Extrasolar planet searches using high-precision radial velocity observations to measure the
reflex motion of the host stars along the line of sight have now yielded over 100 extrasolar planets.1
The discoveries include 10 multiple planet systems with either two or, in the cases of υ And and
55 Cnc, three detected planets. There are also indications that about half of the stars with one
known planet are likely to have additional detectable distant companions (Fischer et al. 2001).
Many of the known multiple planet systems exhibit interesting dynamics. The two planets about
GJ 876 are deep in three orbital resonances at the 2:1 mean-motion commensurability (Laughlin
& Chambers 2001; Lee & Peale 2002), and there are possibly two other systems with mean-motion
resonances: 2:1 for the two planets about HD 829432 (Goździewski & Maciejewski 2001) and 3:1
for the inner two planets about 55 Cnc (Marcy et al. 2002; Lee & Peale 2003). The outer two
planets of the υ And system are apparently locked in a secular resonance with the lines of apsides of
the two orbits being aligned on average (Rivera & Lissauer 2000; Lissauer & Rivera 2001; Chiang,
Tabachnik, & Tremaine 2001), and as we shall demonstrate in this paper, the two planets of the
HD 12661 system are also locked in a secular resonance, but with the lines of apsides being anti-
aligned on average. The resonances are often vital in ensuring the long term stability of the multiple
planet systems.

Although the origin, evolution, and stability of the orbital configurations of multiple planet
systems can usually be analyzed with direct numerical integrations of the full equations of motion,
a theoretical understanding of the dynamics often requires the development and application of
theories with analytic approximations. The approximate theories also allow one to explore the
parameter space much more rapidly than direct numerical integrations in the regions where the
approximations are valid.

Because the orbital eccentricities and inclinations of the planets in the solar system are gen-
erally small while the ratios of the orbital semimajor axes of adjacent planets are generally large
(aj /aj+1 & 0.5 except for the Mars-Jupiter pair), the classical perturbation theory developed for
the solar system is based on an expansion of the disturbing functions that describe the mutual grav-
itational interactions of the planets in powers of the eccentricities and inclinations. In particular,
the Laplace-Lagrange secular solution for the evolution of a two-planet system that is not affected
by mean-motion commensurabilities is based on retaining just the secular terms (i.e., those not
involving the mean longitudes) in the disturbing functions up to second order in the eccentricities
and inclinations (see, e.g., Murray & Dermott 1999). Since high orbital eccentricities are common
among extrasolar planets and the distribution of the orbital eccentricities of the extrasolar planets
in the known multiple planet systems is similar to that of the single planets (Fischer et al. 2003),
the classical Laplace-Lagrange secular perturbation theory is not in general adequate for describing

1
See, e.g., [Link] for a continuously updated catalog.
2
See [Link]
 –3–

the secular evolution of the extrasolar planetary systems that are not affected by mean-motion
commensurabilities.

On the other hand, many of the known multiple planet systems are hierarchical in the sense
that the ratio(s) of the semimajor axes (a1 /a2 for the two-planet systems and a1 /a2 and/or a2 /a3
for the three-planet systems) are small. There are 12 pairs of adjacent planets in the 10 known
multiple planet systems (two of which have three planets). Among the 9 pairs that are not known
or suspected to be in mean-motion resonances, five have aj /aj+1 . 0.1 and all but one (47 UMa)
have aj /aj+1 . 1/3. Thus a secular perturbation theory for a two-planet system that is based on
an expansion in a1 /a2 and valid for high eccentricities should provide an accurate description of
the secular evolution of many extrasolar planetary systems.

It is useful to consider hierarchical two-planet systems in the context of the general hierarchical
triple systems in which a third body orbits an inner binary on a much wider orbit. A hierarchical
triple system can be treated as two binaries on slowly perturbed Kepler orbits by using Jacobi
coordinates, where the position of the secondary of mass m1 of the inner binary is relative to the
primary of mass m0 and the position of the third body of mass m2 is relative to the center of mass
of m0 and m1 . A hierarchical two-planet system is simply a hierarchical triple system with m1 and
m2 much smaller than m0 . In § 2 we derive the orbital parameters in Jacobi coordinates obtained
by the observers from a two (or more generally multiple) Kepler fit to the radial velocity variations
of a host star and show that Jacobi coordinates should be used in any analyses of hierarchical (and
possibly other types of) planetary systems. Star-centered or astrocentric coordinates can introduce
significant high-frequency variations in orbital elements that should be nearly constant on orbital
timescales. The high-frequency variations can then lead to erroneous sensitivity of the evolution of
the orbital elements to the starting epoch.

Secular perturbation theories based on an expansion in α = a1 /a2 have been developed for
hierarchical triple systems. The expansion to order α2 , called the quadrupole approximation, was
developed by Kozai (1962) and Harrington (1968). This quadrupole-level secular perturbation the-
ory is successful in explaining the Kozai mechanism, whereby the perturbations between the inner
binary and the third body can lead to large variations in the eccentricity of the inner binary and
the mutual inclination angle between the inner and outer binaries if the initial mutual inclination is
sufficiently high. However, the quadrupole approximation is not adequate for studying hierarchical
two-planet systems. Although there is no direct information on the mutual inclination angles in
the known hierarchical two-planet systems, the formation of planets from a common disk of ma-
terials surrounding the host star makes nearly coplanar orbits the most probable configuration.
When the orbits are coplanar, the conservation of total angular momentum of the system and the
secularly constant semimajor axes mean that the eccentricities of the two orbits are coupled and
oscillate out of phase. The quadrupole term does not contribute to these eccentricity oscillations
for coplanar orbits. Marchal (1990), Krymolowski & Mazeh (1999), and Ford, Kozinsky, & Rasio
(2000) have extended the approximation to octupole (i.e., α3 ) order. Blaes, Lee, & Socrates (2002)
have recently applied this octupole-level secular perturbation theory, with a sign error corrected
 –4–

(see footnote 5 below) and with modifications to include the effects of general relativistic precession
and gravitational radiation on the inner binary, to study the dynamical evolution of hierarchical
triples of supermassive black holes. Unlike the quadrupole term, the octupole term does produce
eccentricity oscillations for coplanar orbits.

In this paper we use both the octupole-level secular perturbation theory and direct numerical
orbit integrations to investigate the dynamical evolution of coplanar hierarchical two-planet sys-
tems. The applicability of the octupole theory is limited by its use of an expansion in α and an
averaging over the inner and outer orbital motions. In particular, the orbit averaging eliminates
the effects of mean-motion commensurabilities and the possible development of instabilities. We
establish the validity and limits of the octupole theory by comparison with direct numerical orbit
integrations. In § 3 we summarize the derivation of the octupole theory, compare the octupole
theory to the classical Laplace-Lagrange secular solution, and deduce some useful results from the
octupole equations analytically. In § 4 we study the dynamical evolution of the HD 168443 and
HD 12661 systems and their variants. The HD 168443 system is not in a secular resonance and
its secular resonance variable ̟1 − ̟2 circulates, where ̟1,2 are the longitudes of periapse of the
inner and outer orbits, respectively. For the HD 168443 system and a wide variety of systems
with α ≈ 0.1 (including some for which the octupole theory predicts rather unusual dynamical
behaviors), we show that the octupole results are in excellent agreement with the direct integration
results. Direct integrations of two-planet systems similar to HD 168443, but with different initial
a2 (and hence different initial α), are used to show that systems with initial α above a critical
value are generally unstable. As anticipated by the analytic results derived in § 3, the HD 12661
system is in a secular resonance with ̟1 − ̟2 librating about 180◦ , and it shows large amplitude
variations of both eccentricities. We show that the evolution of the HD 12661 system with the
best-fit orbital parameters and sin i = 1 (i is the inclination of the orbital plane from the plane of
the sky) is affected by the proximity to the 11:2 mean-motion commensurability, but that small
changes in the orbital period of the outer planet within the uncertainty can result in configurations
that are not affected by mean-motion commensurabilities. For the latter type of configurations, we
show that the octupole results are in reasonably good agreement with the direct integration results,
even though α(≈ 0.32) is quite large. We also consider the effects of varying the inclination i and
show that the HD 12661 system is unstable if sin i . 0.3. Our conclusions are summarized in § 5.

2. JACOBI ORBITAL PARAMETERS FROM MULTIPLE KEPLER FITS TO

RADIAL VELOCITY OBSERVATIONS

Except for systems such as GJ 876, where the perturbations between the two planets are
significant on orbital timescales and a dynamical fit to the stellar radial velocity variations is
essential (Laughlin & Chambers 2001; Rivera & Lissauer 2001; Nauenberg 2002), it is often adequate
to fit the radial velocity variations of a star with two (or more) planets over the time span of the
available observations by assuming that the planets are on unperturbed Kepler orbits. Many
 –5–

authors have assumed that the orbital parameters obtained from the multiple Kepler fits are in
astrocentric coordinates, but Lissauer & Rivera (2001; see also Note Added in Proof of Rivera
& Lissauer 2000) have pointed out that Jacobi coordinates “better emulate” the assumption of
unperturbed Kepler orbits. In this section we derive explicitly the orbital parameters in Jacobi
coordinates obtained from a two-Kepler fit. It is straightforward to generalize the derivation to
an N -Kepler fit. We show that especially for hierarchical systems such as HD 168443, the use of
astrocentric coordinates can introduce erroneous features in the evolution of the orbital elements.

Let us consider a system consisting of a central star of mass m0 , an inner planet of mass m1 ,
and an outer planet of mass m2 , and use Jacobi coordinates, with r 1 being the position of m1
relative to m0 and r 2 being the position of m2 relative to the center of mass of m0 and m1 . We
shall refer to the orbit of m1 relative to m0 as the inner orbit and the orbit of m2 relative to the
center of mass of m0 and m1 as the outer orbit.

If the inner orbit is an unperturbed Kepler orbit, the line-of-sight (LOS) component of the
velocity of m1 relative to m0 is
h i
V1,r = − ṙ1 sin(ω1 + f1 ) + r1 f˙1 cos(ω1 + f1 ) sin i1
−2πa1
= p [cos(ω1 + f1 ) + e1 cos ω1 ] sin i1 , (1)
P1 1 − e21

where a dot over a symbol denotes d/dt and a1 , e1 , i1 , ω1 , f1 , and

3/2
2πa1
P1 = p (2)
G(m0 + m1 )

are, respectively, the semimajor axis, eccentricity, inclination, argument of periapse, true anomaly,
and period of the inner orbit. In equation (2) G is the gravitational constant. Note that the
reference plane is the plane of the sky, and the planet approaches the observer at the ascending
node, but the radial velocity of an approaching planet is negative. Then the LOS component of
the velocity of m0 relative to the center of mass of m0 and m1 is

′ −m1
V0,r = V1,r = K1 [cos(ω1 + f1 ) + e1 cos ω1 ] , (3)
m0 + m1
where the amplitude
 1/3
2πG m1 sin i1 1
K1 = 2/3
p . (4)
P1 (m0 + m1 ) 1 − e21

Similarly, if the outer orbit is an unperturbed Kepler orbit, the LOS component of the velocity
of m2 relative to the center of mass of m0 and m1 is
−2πa2
V2,r = p [cos(ω2 + f2 ) + e2 cos ω2 ] sin i2 , (5)
P2 1 − e22
 –6–

where a2 , e2 , i2 , ω2 , f2 , and
3/2
2πa2
P2 = p (6)
G(m0 + m1 + m2 )
are, respectively, the semimajor axis, eccentricity, inclination, argument of periapse, true anomaly,
and period of the outer orbit. Then the LOS component of the velocity of the center of mass of m0
and m1 relative to the center of mass of the whole system is

′ −m2
V01,r = V2,r = K2 [cos(ω2 + f2 ) + e2 cos ω2 ] , (7)
m0 + m1 + m2
where the amplitude
 1/3
2πG m2 sin i2 1
K2 = 2/3
p . (8)
P2 (m0 + m1 + m2 ) 1 − e22

Thus, if the orbits of the planets in Jacobi coordinates are unperturbed Kepler orbits, the
radial velocity of the star m0 is
′ ′
Vr = V0,r + V01,r +V
= K1 [cos(ω1 + f1 ) + e1 cos ω1 ] + K2 [cos(ω2 + f2 ) + e2 cos ω2 ] + V, (9)

where V is the LOS velocity of the center of mass of the whole system relative to the observer.
Equation (9) is exactly the formula used by observers in two-Kepler fits, but with the amplitudes
defined in equations (4) and (8) and the orbital periods defined in equations (2) and (6).3 Since the
true anomaly fj depends on Pj , ej , and the time of periapse passage Tperi,j , a two-Kepler fit directly
yields five parameters Pj , Kj , ej , ωj , and Tperi,j for each orbit. Table 1 shows these parameters
for the HD 168443 planets from Marcy et al. (2001) and the HD 12661 planets from D. A. Fischer
(2002, private communication). If the stellar mass m0 is known, equations (2), (4), (6), and (8) can
be used to derive mj and aj for assumed sin ij . (A common but less accurate practice is to derive
mj sin ij and aj by assuming that m1 and m2 are negligible compared to m0 .) Table 1 also shows
mj (in units of Jupiter mass MJ ) and aj for sin ij = 1. Throughout this paper, we refer to the
substellar companions of HD 168443 as planets, but it should be noted that their sin ij = 1 (i.e.,
minimum) masses are about 7.7 and 17MJ , which are near or above the deuterium-burning limit.

The alternative assumption that the orbits of the planets in astrocentric coordinates are unper-
turbed Kepler orbits would also yield equation (9) for the radial velocity of the host star, but with
3/2
Kj = [2πG(m0 +mj )/Pj ]1/3 mj sin ij (m0 +m1 +m2 )−1 (1−e2j )−1/2 and Pj = 2πaj [G(m0 +mj )]−1/2 .
However, as we show next, while the Jacobi orbits of the planets of a hierarchical system are nearly
Keplerian on orbital timescales, the astrocentric orbit of the outer planet can deviate significantly
from a Kepler orbit on orbital timescales.

3
Note that the planetary masses in the equations for the amplitudes and the orbital periods are the physical
masses mj and not the Jacobi masses mj j−1
P Pj
k=0 mk / k=0 mk as stated by Lissauer & Rivera (2001).
 –7–

Table 1. Orbital Parameters of the HD 168443 and HD 12661 Planets

HD 168443 HD 12661

Parameter Inner Outer Inner Outer

P (days) 58.10 1770 263.3 1444.5

K (m s−1 ) 472.7 289 74.4 27.4
e 0.53 0.20 0.35 0.20
ω (deg) 172.9 62.9 292.6 147.0
Tperi (JD) 2450047.58 2450250.6 2449943.7 2449673.9
m (MJ ) 7.73 17.23 2.30 1.57
a (AU) 0.295 2.90 0.823 2.56

Note. — The parameters P , K, e, ω, and Tperi from two-Kepler

fits by Marcy et al. 2001 and D. A. Fischer 2002, private commu-
nication. The parameters m and a are derived for sin i = 1 and
adopted stellar masses of 1.01M⊙ and 1.07M⊙ for HD 168443 and
HD 12661, respectively.
 –8–

Fig. 1.— (a) Variations in the orbital semimajor axes, a1 and a2 , and eccentricities, e1 and e2 , of
the HD 168443 planets with sin i = 1, if we assume that the best-fit orbital parameters obtained
from the two-Kepler fit are in astrocentric coordinates and plot the astrocentric aj and ej . The
solid and dotted lines are from direct numerical orbit integrations starting at Tperi,1 and Tperi,2 of
Table 1, respectively. (b) Same as (a), but we interpret the best-fit orbital parameters obtained
from the two-Kepler fit as orbital parameters in Jacobi coordinates and plot the Jacobi aj and ej .
The solid and dotted lines are almost indistinguishable. The dashed lines in the lower two panels
of (b) are from the octupole-level secular perturbation theory.

Figure 1a shows the variations in the semimajor axes and eccentricities of the HD 168443
planets if we assume that the orbital parameters obtained from the two-Kepler fit are in astrocentric
coordinates and plot the astrocentric aj and ej . The planets are assumed to be on coplanar
orbits with sin i = 1 (note that coplanar orbits have the same inclinations, i1 = i2 = i, and the
same longitudes of ascending node, Ω2 = Ω1 , referenced to the plane of the sky). The solid and
dotted lines are from direct numerical orbit integrations starting at Tperi,1 and Tperi,2 of Table 1,
respectively. The direct integrations were performed using the Wisdom-Holman (1991) integrator
contained in the SWIFT4 software package, with input and output in astrocentric coordinates.
Figure 1b is similar to Figure 1a, but it shows the variations in the Jacobi aj and ej , with the
initial orbital parameters also in Jacobi coordinates. The direct integrations were performed using
the modified Wisdom-Holman integrator described in § 4. In astrocentric coordinates (Fig. 1a), the

4
See [Link]
 –9–

evolution of the orbital elements is sensitive to the starting epoch; and while a1 is nearly constant
and e1 oscillates only on the secular timescale, both a2 and e2 show significant fluctuations on
the orbital timescales (note that we have reduced overcrowding in Fig. 1a by using a sampling
interval of 50 yr, which is long compared to the orbital timescales). Plots of the variations in the
orbital elements of the HD 168443 planets in Marcy et al. (2001), Nagasawa, Lin, & Ida (2003),
and Udry, Mayor, & Queloz (2003, from calculations by W. Benz) show similar behaviors and are
probably due to the use of astrocentric coordinates. In contrast, in Jacobi coordinates (Fig. 1b),
the evolution is not sensitive to the starting epoch (the solid and dotted lines in Fig. 1b are almost
indistinguishable); both a1 and a2 are nearly constant, while both e1 and e2 oscillate only on the
secular timescale, with the maximum in e1 coinciding with the minimum in e2 and vice versa. As
we shall see, these behaviors can be understood with the octupole-level secular perturbation theory
for coplanar hierarchical two-planet systems, which is also based on Jacobi coordinates.

From the facts that the Jacobi aj are nearly constant and that the Jacobi ej oscillate only on
the secular timescale for hierarchical systems, it is easy to estimate the fractional fluctuation in
the astrocentric a2 , which is primarily due to the velocity of the star m0 relative to the center of
mass of m0 and m1 . It is approximately 4(m1 /m0 )[(a2 /a1 )(1 + e1,max )/(1 − e1,max )]1/2 when the
Jacobi e1 is at its maximum e1,max (and e2 is at its minimum, which is assumed to be small and
is neglected). The fluctuations in the astrocentric a2 in Figure 1a are in good agreement with this
estimate. Note that the fractional fluctuation is larger for smaller a1 /a2 , i.e., for systems that are
more hierarchical.

In their study of the three-planet υ And system, Rivera & Lissauer (2000) have also reported
that the use of Jacobi coordinates eliminates the high-frequency variations in the semimajor axes
and eccentricities of the outer two planets and reduces the sensitivity of the evolution to the initial
epoch. It is clear that Jacobi coordinates should be used in the analysis of hierarchical systems
such as HD 168443 (where a1 /a2 ≈ 0.10) and υ And (where a1 /a2 ≈ 0.071 and a2 /a3 ≈ 0.33).
It is likely that Jacobi coordinates should also be used in the analysis of other types of planetary
systems, especially those for which multiple Kepler fits are adequate.

3. OCTUPOLE-LEVEL SECULAR PERTURBATION THEORY

As we mentioned in § 1, an approximate theory that describes the secular evolution of copla-

nar hierarchical two-planet systems in Jacobi coordinates, such as that shown in Figure 1b for the
HD 168443 system, is the octupole-level secular perturbation theory. In this section we summa-
rize the derivation of this octupole theory, compare it to the classical Laplace-Lagrange secular
perturbation theory, and deduce some results from the octupole equations analytically.
 – 10 –

3.1. Equations

As in §2, we consider a system consisting of a central star of mass m0 , an inner planet of mass
m1 , and an outer planet of mass m2 , and use Jacobi coordinates, with r 1 being the position of m1
relative to m0 and r 2 being the position of m2 relative to the center of mass of m0 and m1 . The
Hamiltonian of this system is
   
Gm0 m1 G(m0 + m1 )m2 1 1 1 1
H=− − − Gm0 m2 − − Gm1 m2 − , (10)
2a1 2a2 r02 r2 r12 r2

where aj is the osculating semimajor axis of the jth orbit (with j = 1 and 2 for the inner and outer
orbits, respectively), and rk2 is the distance between mk and m2 . With r1 < r2 , both 1/r02 and
1/r12 can be expanded in powers of r1 /r2 , leading to the expanded Hamiltonian
∞  k  k+1
Gm0 m1 G(m0 + m1 )m2 GX k r1 a2
H=− − − α Mk Pk (cos Φ), (11)
2a1 2a2 a2 a1 r2
k=2

where α = a1 /a2 ,
m0k−1 − (−m1 )k−1
Mk = m0 m1 m2 , (12)
(m0 + m1 )k
Pk is the Legendre polynomial of degree k, and Φ is the angle between r 1 and r 2 . The first
two terms in equations (10) and (11) represent the independent Kepler motions of the inner and
outer orbits, while the remaining terms represent the perturbations to the Kepler motions. For
hierarchical systems with r1 ≪ r2 , the Jacobi decomposition leads to two slowly perturbed Kepler
orbits, even if m1 and m2 are not much smaller than m0 .

In the general case where mutually inclined orbits are allowed, it is convenient to use the
Delaunay variables with the invariable plane as the reference plane. The coordinates are the mean
anomalies lj , the arguments of periapse gj = ωj , and the longitudes of ascending node hj = Ωj ,
and the conjugate momenta are
m0 m1 p
L1 = G(m0 + m1 )a1 , (13)
m0 + m1
(m0 + m1 )m2 p
L2 = G(m0 + m1 + m2 )a2 , (14)
m0 + m1 + m2
q
Gj = Lj 1 − e2j , (15)
Hj = Gj cos ij , (16)

where ej and ij are the eccentricities and inclinations of the orbits. The momenta Lj , Gj , and Hj
are, respectively, the magnitude of the maximum possible angular momentum (if the orbit were
circular), the magnitude of the angular momentum, and the z-component of the angular momentum
of the jth orbit. By expressing cos Φ = r 1 ·r 2 /(r1 r2 ) in terms of the Delaunay variables, it is easy to
show that the Hamiltonian contains h1 and h2 only in the combination h1 − h2 , and h1 − h2 = 180◦
if the invariable plane is used as the reference plane. Constant h1 − h2 and constant total angular
 – 11 –

momentum H1 + H2 mean hj and Hj can be eliminated from the problem — usually referred to
as the elimination of nodes — and the system is reduced from 6 to 4 degrees of freedom (see, e.g.,
Marchal 1990).

Kozai (1962) studied the secular evolution of hierarchical triple systems with m1 ≪ m2 ≪ m0
and e2 = 0 by retaining just the lowest-order (α2 ) quadrupole term in the series in equation (11)
and averaging the Hamiltonian over the inner and outer orbital motions. Harrington (1968) gener-
alized the quadrupole analysis to general hierarchical triple systems. Marchal (1990), Krymolowski
& Mazeh (1999), and Ford et al. (2000) extended the approximation by retaining also the oc-
tupole term of order α3 in the series in equation (11) and derived an octupole-level orbit-averaged
Hamiltonian. The approach used in these studies to average the Hamiltonian is the von Zeipel
method, which involves the determination of a canonical transformation such that the transformed
Hamiltonian is independent of the transformed mean anomalies l1 and l2 , and such that the original
and transformed variables differ only by high frequency terms that are multiplied by powers of the
small parameter α. Since the transformed Hamiltonian is independent of the transformed lj , the
transformed Lj and the corresponding semimajor axes are constant.

In this paper we focus on hierarchical two-planet systems with coplanar orbits, as the formation
of planets from a common disk of materials surrounding the host star makes nearly coplanar orbits
the most probable configuration. In the limit of coplanar orbits (i1 = i2 = 0◦ ), the Delaunay
variables described above are not well-defined, since there are no ascending nodes with respect
to the invariable plane, but the longitudes of periapse ̟j are well-defined. Thus, for coplanar
orbits, it is convenient to use the canonical variables lj , ̟j , Lj , and Gj . In the limit of coplanar
orbits, by noting that the expression for the angle ϕ between the directions of periapse reduces to
cos ϕ = − cos(g1 − g2 ) = cos(g1 − g2 + 180◦ ) = cos(̟1 − ̟2 ) (since h1 − h2 = 180◦ ), the doubly
averaged octupole-level Hamiltonian derived by Marchal (1990), Krymolowski & Mazeh (1999),
and Ford et al. (2000) can be written as5

G2 m30 m31 G2 (m0 + m1 )3 m32

H̄oct = − − −2C2 (2+3e21 )+C3 e1 e2 (4+3e21 ) cos(̟1 − ̟2 ), (17)
2(m0 + m1 )L21 2(m0 + m1 + m2 )L22

where
1 G2 (m0 + m1 )7 m72 L41
C2 = , (18)
16 (m0 + m1 + m2 )3 (m0 m1 )3 L32 G32

15 G2 (m0 + m1 )9 m92 (m0 − m1 ) L61

C3 = , (19)
64 (m0 + m1 + m2 )4 (m0 m1 )5 L32 G52

5
There is a sign error common to both Krymolowski & Mazeh (1999) and Ford et al. (2000). We follow Ford et al.
and denote the coefficients of the quadrupole and octupole terms by C2 and C3 , respectively. Krymolowski & Mazeh
denote these coefficients by C1 and C2 , respectively. In the averaged Hamiltonian and the subsequent equations of
motion of Ford et al., the coefficient C3 should be replaced by −C3 . Similarly, in the averaged Hamiltonian and the
subsequent equations of motion of Krymolowski & Mazeh, the coefficient C2 should be replaced by −C2 .
 – 12 –

and s  2
Gj
ej = 1− . (20)
Lj
As in Ford et al. (2000), we do not include in equation (17) terms of order α7/2 induced by the
canonical transformation of the von Zeipel method, which were included partially by Marchal (1990)
and fully by Krymolowski & Mazeh (1999).

As we mentioned above, the orbit averaging eliminates lj from H̄oct , and hence Lj (and aj ) are
constant. Furthermore, H̄oct contains ̟1 and ̟2 only in the combination ̟1 − ̟2 , and choosing
this difference as a new variable in a canonical transformation simply reveals that the total angular
momentum G1 + G2 is an integral of motion. Thus, in the octupole approximation, the coplanar
problem is reduced to one degree of freedom with the last two terms in equation (17) as an integral
of motion. We can also see from equation (17) that the quadrupole approximation is not adequate
for studying coplanar systems. If we drop the octupole term with coefficient C3 in equation (17), the
Hamiltonian is independent of ̟1 and ̟2 , meaning that both e1 and e2 are constant, which is not
generally true. (For mutually inclined orbits, only e2 is constant in the quadrupole approximation;
Harrington 1968.)

From the equations of motion

d̟j ∂ H̄oct dGj ∂ H̄oct

= and =− , (21)
dt ∂Gj dt ∂̟j

we obtain the following equations for the variation of ej and ̟j :

(1 − e21 )1/2 1 + 43 e21

de1
= −A12 e2 sin(̟1 − ̟2 ), (22)
dt (1 − e22 )5/2

1 + 43 e21


de2
= A21 e1 sin(̟1 − ̟2 ), (23)
dt (1 − e22 )2

(1 − e21 )1/2 e2 (1 − e21 )1/2 1 + 49 e21

 

d̟1
= A11 − A12 cos(̟1 − ̟2 ), (24)
dt (1 − e22 )3/2 e1 (1 − e22 )5/2

1 + 23 e21 e1 (1 + 4e22 ) 1 + 34 e21

 

d̟2
= A22 − A21 cos(̟1 − ̟2 ), (25)
dt (1 − e22 )2 e2 (1 − e22 )3

where the constant coefficients

 5   
4C3 G2 15 m2 m0 − m1
A12 = = n1 α4 , (26a)
L1 L2 16 m0 + m1 m0 + m1
 5  
4C3 G2 15 m0 m1 m0 − m1
A21 = = n2 α3 , (26b)
L2 L2 16 (m0 + m1 )2 m0 + m1
 – 13 –

 3  
12C2 G2 3 m2
A11 = = n1 α3 , (26c)
L1 L2 4 m0 + m1
 3
12C2 G2 3 m0 m1
A22 = = n2 α2 , (26d)
L2 L2 4 (m0 + m1 )2
with α = a1 /a2 and the mean motions
1/2 1/2
n1 = G(m0 + m1 )/a31 n2 = G(m0 + m1 + m2 )/a32
 
and . (27)

Consistent with the fact that the system can be reduced to one degree of freedom, either equation
(22) or (23) is redundant, since e1 and e2 are related by the conservation of total angular momentum,
and equations (24) and (25) can be combined as
"
#
(1 − e21 )1/2 1 + 32 e21
 
d(̟1 − ̟2 ) L1
= A11 −
dt (1 − e22 )3/2 L2 (1 − e22 )2
" 
#
e2 (1 − e21 )1/2 1 + 94 e21 e1 (1 + 4e22 ) 1 + 34 e21

  
L1
−A12 −
e1 (1 − e22 )5/2 L2 e2 (1 − e22 )3

× cos(̟1 − ̟2 ). (28)

If we relax the assumption of coplanar orbits and allow a small mutual inclination angle imu
between the inner and outer orbits, the number of degrees of freedom is increased from one to two,
but an expansion of the orbit-averaged octupole-level Hamiltonian of Marchal (1990), Krymolowski
& Mazeh (1999), and Ford et al. (2000) in powers of imu shows that equations (22)–(25) are only
modified by terms of order i2mu and higher. One of the additional terms for de1 /dt (which has
an octupole term only in the coplanar limit; see eq. [22]) is a quadrupole term with coefficient
C2 i2mu . Thus we expect a small mutual inclination angle imu ≪ α1/2 to have little effect on the
secular evolution of ej and ̟j of a hierarchical system. This is confirmed by direct numerical orbit
integrations in § 4.

It is important to note that the octupole-level secular perturbation theory is derived under
the assumptions that there are no mean-motion commensurabilities and that α (or more precisely
r1 /r2 ) ≪ 1. As we shall see in § 4, close proximity to even a rather high order commensurability
(e.g., 11:2) can affect the evolution of a system. We shall also see in § 4 that although there is a
limit to how large α can be for a system to be stable, the octupole theory can provide a reasonable
description of the secular evolution for α as large as 1/3 if the planets are not too massive.

3.2. Comparison with Classical Laplace-Lagrange Secular Perturbation Theory

As we mentioned in § 1, the classical secular perturbation theory developed for the solar system
is based on an expansion in the eccentricities and inclinations. While it is valid to all orders in
 – 14 –

the ratio of the semimajor axes α, it can be applied only to systems with small planetary masses
on nearly circular and nearly coplanar orbits. In contrast, the octupole-level secular perturbation
theory is based on an expansion in α, and it can be applied to hierarchical systems with a wide
range of masses, eccentricities and, in its most general form, inclinations. We now show that the
octupole equations (22)–(25) agree with the equations for the classical Laplace-Lagrange secular
solution for coplanar systems where the planetary masses, eccentricities, and α are all small.

To the lowest order in the eccentricities, equations (22)–(25) reduce to

de1
= −A12 e2 sin(̟1 − ̟2 ), (29)
dt
de2
= A21 e1 sin(̟1 − ̟2 ), (30)
dt  
d̟1 e2
= A11 − A12 cos(̟1 − ̟2 ), (31)
dt e1
 
d̟2 e1
= A22 − A21 cos(̟1 − ̟2 ), (32)
dt e2

where A12 = (15/16)(m2 /m0 )n1 α4 , A21 = (15/16)(m1 /m0 )n2 α3 , A11 = (3/4)(m2 /m0 )n1 α3 , and
A22 = (3/4)(m1 /m0 )n2 α2 in the limit m1 , m2 ≪ m0 .

The classical Laplace-Lagrange secular solution is based on retaining just the secular terms in
the disturbing functions up to second order in the eccentricities and inclinations. For a coplanar
planetary system, the equations for the variation of ej and ̟j (see, e.g., Murray & Dermott 1999)
are of the same form as equations (29)–(32), but with the Ajk replaced by

1 m2 (2)
A′12 = n1 α2 b3/2 (α), (33a)
4 m0 + m1
1 m1 (2)
A′21 = n2 αb (α), (33b)
4 m0 + m2 3/2
1 m2 (1)
A′11 = n1 α2 b3/2 (α), (33c)
4 m0 + m1
1 m1 (1)
A′22 = n2 αb (α), (33d)
4 m0 + m2 3/2
(j)
where b3/2 (α) is the Laplace coefficient. (It should be noted that the classical theory uses astro-
centric coordinates instead of Jacobi coordinates, but that the distinction between the two sets
(2)
of coordinates vanishes in the limit m1 /m0 ≪ α1/2 /4; see §2.) Since b3/2 (α) = (15/4)α2 and
(1)
b3/2 (α) = 3α to the lowest order in α, the classical and octupole secular perturbation equations are
identical in the limit of small planetary masses, eccentricities, and α.
 – 15 –

3.3. Some Analytic Results

Although it does not appear that the octupole-level secular perturbation equations can be
solved analytically, some useful results can be deduced analytically. We begin by rewriting equations
(22), (23), and (28) as

(1 − e21 )1/2 1 + 43 e21

de1
= −βe2 sin(̟1 − ̟2 ), (34)
dτ (1 − e22 )5/2

1 + 43 e21


de2
= βλe1 sin(̟1 − ̟2 ), (35)
dτ (1 − e22 )2
"
#
d(̟1 − ̟2 ) (1 − e21 )1/2 1 + 32 e21
= −λ
dτ (1 − e22 )3/2 (1 − e22 )2
" 
#
e2 (1 − e21 )1/2 1 + 94 e21 e1 (1 + 4e22 ) 1 + 34 e21

 
−β −λ
e1 (1 − e22 )5/2 e2 (1 − e22 )3

× cos(̟1 − ̟2 ), (36)

where
 
A12 A21 5 m0 − m1
β = = = α, (37)
A11 A22 4 m0 + m1
A22 A21 L1
λ = = = , (38)
A11 A12 L2
τ = t/te , (39)

and  
1 4 m0 + m1 1
te = = . (40)
A11 3α3 m2 n1
Recall that either e1 or e2 can be eliminated from the above equations by the conservation of total
angular momentum, which can be written in dimensionless form as γ = (G1 + G2 )/(L1 + L2 ) =
[λ(1 − e21 )1/2 + (1 − e22 )1/2 ]/(λ + 1) = constant.

We can see from equations (34)–(36) that coplanar hierarchical systems with the same β and
λ (which are constant) and the same initial e1 , e2 , and ̟1 − ̟2 have the same trajectory in the
phase-space diagram of e1 (or e2 ) versus ̟1 − ̟2 and can differ only in the period of eccentricity
oscillations. In particular, in the limit m1 , m2 ≪ m0 , since β ≈ (5/4)α, λ ≈ (m1 /m2 )α1/2 , and
te ≈ 4(m0 /m2 )/(3α3 n1 ), systems with the same α, m1 /m2 , and initial e1 , e2 and ̟1 − ̟2 , but
different m0 , m2 and/or a1 , differ only in the period of eccentricity oscillations, which is proportional
to (m0 /m2 )/n1 . As we discussed in § 2, a two-Kepler fit yields ej , ωj , and in the limit m1 , m2 ≪ m0 ,
mj sin ij and aj , but not the inclinations ij of the orbital planes to the plane of the sky. (Hereafter,
all inclinations are relative to the plane of the sky and not to the invariable plane.) If we assume
 – 16 –

that the orbits are coplanar, the above results mean that the trajectory in e1 (or e2 ) versus ̟1 − ̟2
and the amplitudes of eccentricity oscillations should be independent of the unknown inclination i
of both orbits, and that the period of eccentricity oscillations should be proportional to sin i.

The phase-space structure of the secular evolution of coplanar hierarchical two-planet systems
can be understood by plotting the trajectories of systems with the same β, λ, and γ in a diagram
of e1 (or e2 ) versus ̟1 − ̟2 (see, e.g., Figs. 2 and 6 below). The fixed points in the phase-space
diagram can be found by solving de1 /dτ = de2 /dτ = d(̟1 − ̟2 )/dτ = 0. It is clear from equations
(34) and (35) that de1 /dτ = de2 /dτ = 0 requires sin(̟1 − ̟2 ) = 0 or ̟1 − ̟2 = 0◦ or 180◦ . As
we shall see, there are usually two libration islands, one about a fixed point at ̟1 − ̟2 = 0◦ and
another about a fixed point at ̟1 − ̟2 = 180◦ ; additional fixed points are possible if the total
angular momentum is low.

At ̟1 − ̟2 ≈ 90◦ and 270◦ , since cos(̟1 − ̟2 ) ≈ 0, equation (36) reduces to d(̟1 −̟2 )/dτ ≈
(1 − e21 )1/2 /(1 − e22 )3/2 − λ(1 + 3e21 /2)/(1 − e22 )2 . Therefore, to the lowest order in the eccentricities,
if λ ≈ 1, d(̟1 − ̟2 )/dτ ≈ 0 and ̟1 − ̟2 should be nearly constant while both e1 and e2 change.
According to equations (34) and (35), e1 should be decreasing (from near its maximum possible
value for the given total angular momentum to near zero) and e2 should be increasing (from near
zero to near its maximum possible value for the given total angular momentum) at ̟1 − ̟2 ≈ 90◦ ,
where sin(̟1 − ̟2 ) ≈ 1, and vice versa at ̟1 − ̟2 ≈ 270◦ , where sin(̟1 − ̟2 ) ≈ −1. These
behaviors are most consistent with a phase space that is dominated by two large libration islands,
one about a fixed point at ̟1 − ̟2 = 0◦ and another about a fixed point at ̟1 − ̟2 = 180◦ .
Note also that for a given total angular momentum the maximum possible values of e1 and e2
are comparable if λ ≈ 1. Finite eccentricities change the condition for d(̟1 − ̟2 )/dτ ≈ 0 at
̟1 − ̟2 ≈ 90◦ and 270◦ to λ ≈ (1 − e21 )1/2 (1 − e22 )1/2 /(1 + 3e21 /2), which is slightly less than 1
for moderate eccentricities. To estimate how much smaller than 1 the critical value of λ is for a
given dimensionless total angular momentum γ, we substitute into the above condition the value
of the eccentricities when they are equal [e1 = e2 = (1 − γ 2 )1/2 ] and obtain λcrit = 2γ 2 /(5 − 3γ 2 ).
Note that λcrit = 1 is recovered for γ = 1. Therefore, large libration islands and large amplitude
variations of both ej are likely if λ ≈ λcrit = 2γ 2 /(5 − 3γ 2 ).

4. NUMERICAL RESULTS

In this section we study the dynamical evolution of the HD 168443 and HD 12661 systems and
their variants and demonstrate the validity and limits of the octupole-level secular perturbation
theory by comparison with direct numerical orbit integrations. Except for the direct integrations
discussed in § 4.3, the planets are assumed to be on coplanar orbits. Since the octupole-level
secular perturbation equations do not involve high frequency terms, numerical integrations of these
equations are rapid. We integrated equations (35) and (36), with e1 found from conservation of
total angular momentum, using a Bulirsch-Stoer integrator. The direct numerical orbit integrations
were performed using a modified version of the Wisdom-Holman (1991) integrator contained in the
 – 17 –

SWIFT software package. In addition to changing the input and output to Jacobi orbital elements,
we divide the Hamiltonian into a part that describes the Kepler motions of the inner and outer
orbits and a part that describes the perturbations to the Kepler motions using equation (10) instead
of the division used by Wisdom & Holman (1991), which moves the term Gm1 m2 /r2 from the
perturbation Hamiltonian to the Kepler Hamiltonian of the outer orbit. This modified integrator
can handle hierarchical systems where m1 and m2 are not small.

4.1. HD 168443

The best-fit orbital parameters of the HD 168443 planets from Marcy et al. (2001) and the
inferred planetary masses and semimajor axes for sin i = 1 are listed in Table 1. This system
has α = a1 /a2 = 0.102, β = 5(m0 − m1 )α/[4(m0 + m1 )] = 0.126, λ = L1 /L2 = 0.143, and
γ = (G1 + G2 )/(L1 + L2 ) = 0.963. Figure 1b shows the variations in the Jacobi aj and ej , with the
solid and dotted lines (almost indistinguishable) from direct numerical orbit integrations starting
at Tperi,1 and Tperi,2 of Table 1, respectively, and the dashed lines in the lower two panels from
the octupole-level secular perturbation theory. The trajectories in Figure 2 marked by the squares
(which denote the current parameters of the HD 168443 system) are the trajectories of the sin i = 1
HD 168443 system in the phase-space diagrams of e1 versus ̟1 − ̟2 (upper panels) and e2 versus
̟1 − ̟2 (lower panels), with the left and right panels showing the results from direct integrations
and the octupole theory, respectively (see below for a discussion of the other trajectories in Fig. 2).
The direct integration results agree with the octupole theory in that the evolution is not sensitive
to the starting epoch and that a1 and a2 are nearly constant. In addition, the trajectories in the
diagram of e1 (or e2 ) versus ̟1 − ̟2 from the direct integrations are in excellent agreement with
that from the octupole theory. The only noticeable difference is that the period of eccentricity
oscillations predicted by the octupole theory is about 3% longer that that found in the direct
integrations (≈ 1.8 × 104 yr). Krymolowski & Mazeh (1999) have reported that the oscillation
period predicted by the octupole theory can be improved by the inclusion of the terms of order
α7/2 induced by the canonical transformation of the von Zeipel method and neglected by us.

To study the effects of sin i, we have performed two direct integrations of the coplanar HD
168443 system with sin i = 0.4, one starting at Tperi,1 and the other Tperi,2 . (Although the two-
Kepler fit to the radial velocity observations does not yield sin ij , the lack of evidence for stellar
wobble in the Hipparcos astrometric data limits sin i2 & 0.4 for the outer planet of HD 168443;
Marcy et al. 2001.) In both cases, the amplitudes of eccentricity oscillations and the trajectory
in e1 (or e2 ) versus ̟1 − ̟2 are almost identical to those shown in Figures 1b and 2 for sin i = 1,
and the factor by which the eccentricity oscillation period shortens agrees with sin i = 0.4 to better
than ±0.001. These results are in good agreement with the analytic results derived in § 3.3 in the
limit m1 , m2 ≪ m0 , even though m2 is almost 44MJ (and m2 /m0 ≈ 0.042) when sin i = 0.4.

In addition to the trajectories of the sin i = 1 HD 168443 system, Figure 2 also shows the
trajectories of systems with the same masses, initial semimajor axes, and total angular momentum
 – 18 –

Fig. 2.— Trajectories in the phase-space diagrams of e1 vs. ̟1 − ̟2 (upper panels) and e2 vs.
̟1 − ̟2 (lower panels) for two-planet systems with the same masses, initial semimajor axes, and
total angular momentum as the sin i = 1 HD 168443 system. The left and right panels show
the results from direct numerical orbit integrations and the octupole-level secular perturbation
theory, respectively. The trajectories through the squares are those of the HD 168443 system,
with the squares showing the current parameters of the system. The initial conditions for the
other trajectories are ̟1 − ̟2 = 0◦ and e1 = 0.06, 0.08, 0.10, 0.15, 0.25, 0.40, 0.65, or 0.70,
or ̟1 − ̟2 = 180◦ and e1 = 0.706 or 0.7072, with e2 being determined from the total angular
momentum. All the direct integrations (except that for the HD 168443 system) start with the
outer planet at apoapse and the inner planet at opposition.

as the sin i = 1 HD 168443 system. (Recall that the octupole results are also valid for other
systems with the same β, λ, and γ.) There is excellent agreement between the direct-integration
and octupole results in all cases. The HD 168443 system is not in a secular resonance and its
̟1 − ̟2 circulates. Indeed it is far from the relatively small libration islands about the fixed
points at (̟1 − ̟2 , e1 ) = (0◦ , 0.046) and (180◦ , 0.702) in the phase-space diagram of e1 versus
̟1 − ̟2 . The small libration islands and modest eccentricity variations in Figure 2 are consistent
with the fact that these systems have λ = 0.143 far from λcrit = 0.836 for the dimensionless total
angular momentum γ = 0.963 of these systems.

The above e1 values for the fixed points were obtained from the octupole theory as the roots
of d(̟1 − ̟2 )/dτ = 0 at ̟1 − ̟2 = 0◦ and 180◦ , with d(̟1 − ̟2 )/dτ from equation (36) and e2
 – 19 –

Fig. 3.— Values of e1 for the fixed points at ̟1 − ̟2 = 0◦ (left panels) and 180◦ (right panel) in
the phase-space diagram of e1 vs. ̟1 − ̟2 (like Fig. 2) as a function of the dimensionless total
angular momentum γ = (G1 + G2 )/(L1 + L2 ), as determined by the octupole theory for systems
with the same β(= 0.126) and λ(= 0.143) as the sin i = 1 HD 168443 system. The dotted lines
indicate the dimensionless total angular momentum γ = 0.963 of the sin i = 1 HD 168443 system,
and the dashed lines indicate the maximum possible e1 as a function of γ. See text for the meaning
of the square in the upper left panel.

from the conservation of total angular momentum. As we discussed in § 3.3, the octupole theory
predicts that the fixed points must be at ̟1 − ̟2 = 0◦ or 180◦ . To demonstrate how the octupole
theory can be used to explore the parameter space rapidly, we have used the same root finding
procedure to determine the number and positions of fixed points as a function of the dimensionless
total angular momentum γ = (G1 + G2 )/(L1 + L2 ) for systems with the same β and λ as the
sin i = 1 HD 168443 system. The e1 values for the fixed points at ̟1 − ̟2 = 0◦ and 180◦ as a
function of γ are shown in the left and right panels of Figure 3, respectively. For all values of γ
shown in Figure 3, there is an elliptic fixed point at ̟1 − ̟2 = 0◦ with relatively small e1 (lower
left panel) and an elliptic fixed point at ̟1 − ̟2 = 180◦ with e1 close to the maximum possible
value indicated by the dashed line (right panel). These two fixed points are the ones seen above
for systems with the same γ(= 0.963; dotted lines in Fig. 3) as the sin i = 1 HD 168443 system.

For 0.872 . γ . 0.8818, there are two additional fixed points at ̟1 − ̟2 = 0◦ with e1
very close to 1 (upper left panel of Fig. 3). These additional fixed points emerge at γ ≈ 0.8818
with the same e1 value (square in the upper left panel of Fig. 3), and the one with e1 value that
 – 20 –

increases (decreases) with decreasing γ is an elliptic (hyperbolic) fixed point. Direct numerical
orbit integrations confirm that there are indeed stable librations about an elliptic fixed point close
to the additional one predicted by the octupole theory. For example, for γ = 0.88, the octupole
theory predicts that the additional elliptic fixed point is at e1 = 0.9948 (and e2 = 0.1302). A direct
integration of a system with these values of e1 , e2 , and ̟1 − ̟2 as initial conditions (and with the
masses and semimajor axes of the sin i = 1 HD 168443 system and mean anomalies l1 = 0◦ and
l2 = 180◦ ) shows that this system has ̟1 − ̟2 librating about 0◦ with an amplitude of about 6◦
and e1 varying between 0.9946 and 0.9969. It should be noted that the treatment of this system
with e1 ≥ 0.9946 and a1 ≈ 0.3 AU as point masses interacting via Newtonian gravity is inadequate
since the periapse distance of the inner planet is in fact less than the stellar radius. However,
based on the octupole theory, we expect variants of this system with much larger a1 (so that the
periapse distance of the inner planet is well outside the stellar radius) but the same α to show
similar libration behaviors. On the other hand, the inner planet could still come sufficiently close
to the star for general relativistic precession and tidal effects to be important. For γ . 0.872, the
additional elliptic fixed point vanishes, leaving only the hyperbolic fixed point (upper left panel of
Fig. 3). Both octupole and direct-integration calculations show that systems with γ . 0.872, initial
̟1 − ̟2 = 0◦ and initial e1 above the hyperbolic fixed point have e1 increasing to unity in a finite
time. Again, we expect variants of these systems with larger a1 (so that the periapse distance of
the inner planet is well outside the stellar radius) but the same α to show similar increase in e1 .
However, before e1 reaches unity, either the effects of tides and general relativistic precession would
eventually stop the increase in e1 , or the inner planet would collide with the star.

4.2. HD 12661

The best-fit orbital parameters of the HD 12661 planets from D. A. Fischer (2002, private
communication) and the inferred planetary masses and semimajor axes for sin i = 1 are listed
in Table 1. Figure 4 shows the variations in the Jacobi aj and ej , with the solid and dotted
lines from direct numerical orbit integrations starting at Tperi,1 and Tperi,2 of Table 1, respectively,
and the dashed lines in the lower two panels from the octupole-level secular perturbation theory.
The direct integration results are not consistent with the secular theory in that the evolution of
the orbital elements is sensitive to the starting epoch. It turns out that a system with exactly
the best-fit orbital parameters is very close to the 11:2 mean-motion commensurability (P2 /P1 =
5.486 = 0.9975 × 11/2). The direct integration starting at Tperi,1 (solid lines in Fig. 4) shows
irregular fluctuations in aj and ej (note, e.g., the irregular jumps in the mean values of a1 and a2
at successive minima of e2 ) and is most likely chaotic. The direct integration starting at Tperi,2
(dotted lines in Fig. 4) does not show any obvious irregular jumps in aj or ej , and may be either
regular or very weakly chaotic. But its smaller amplitudes of eccentricity variations are due to the
11:2 commensurability. The chaos in one (and possibly both) of these calculations is due to the
overlap of the resonances at the 11:2 commensurability (see Holman & Murray 1996 and Murray
& Holman 1997 for a similar situation in the planar elliptic restricted three-body problem). For
 – 21 –

Fig. 4.— Variations in the Jacobi semimajor axes, a1 and a2 , and eccentricities, e1 and e2 , of
the sin i = 1 HD 12661 system with the best-fit orbital parameters listed in Table 1. The solid
and dotted lines are from direct integrations starting at Tperi,1 and Tperi,2 of Table 1, respectively,
and the dashed lines in the lower two panels are from the octupole theory. This system with
initial P2 /P1 = 0.9975 × 11/2 is affected by the 11:2 mean-motion commensurability, and the direct
integration results are sensitive to the starting epoch. The direct integration starting at Tperi,1
shows irregular fluctuations in aj and ej (note, e.g., the irregular jumps in the mean values of a1
and a2 at successive minima of e2 ) and is most likely chaotic. The most noticeable effect of the 11:2
commensurability on the direct integration starting at Tperi,2 is the reduction in the amplitudes of
the eccentricity variations.

both of the direct integrations shown in Figure 4, we have examined the 10 eccentricity-type mean-
motion resonance variables at the 11:2 commensurability and confirmed that some of them alternate
between circulation and libration. We have extended the direct integrations shown in Figure 4 and
found that both are stable for at least 106 yr, but we cannot rule out the possibility that the chaos
would lead to instability on longer timescales.

The orbital period P2 (and the other orbital parameters) of the outer planet of HD 12661 are not
currently known to high precision, because the time span of the available observations is comparable
to P2 . To study the effects of varying P2 , we have performed two sets of direct integrations with
different initial P2 , one starting at Tperi,1 and the other at Tperi,2 . The initial P2 were chosen such
that (P2 /P1 )/(11/2) = 0.98, 0.99, 1.01, 1.02, and 1.03. The initial values of the other orbital
parameters that can be obtained from the two-Kepler fit (P1 , K1,2 , e1,2 , ω1,2 , and Tperi,1,2 ) were
 – 22 –

Fig. 5.— Same as Fig. 4, but for the sin i = 1 HD 12661 system with an initial P2 such that
P2 /P1 = 0.99 × 11/2. This system shows regular secular evolution, with the direct integration
results insensitive to the starting epoch.

fixed at the values listed in Table 1, and the planetary masses and initial semimajor axes were
derived assuming that sin i = 1. We find that the cases with initial (P2 /P1 )/(11/2) = 1.01 are also
affected by the 11:2 mean-motion commensurability while those with initial (P2 /P1 )/(11/2) = 0.98,
0.99, and 1.02 show regular secular evolution. Although the variations in aj for the cases with initial
P2 /P1 = 1.03 × 11/2 = 0.9997 × 17/3 indicate that these cases are probably affected by the 17:3
commensurability, the variations in ej and the trajectories in the diagram of e1 versus ̟1 − ̟2 are
qualitatively indistinguishable from those showing regular secular evolution (for at least 2 × 105 yr).

In Figures 5 and 6 we show examples of the cases with regular secular evolution. Figure 5 shows
the variations in aj and ej for the sin i = 1 HD 12661 system with P2 /P1 = 0.99 × 11/2. The solid
and dotted lines are from the direct integrations starting at Tperi,1 and Tperi,2 , respectively, and the
dashed lines are from the octupole-level secular perturbation theory. The trajectories of this system
in the phase-space diagrams of e1 versus ̟1 − ̟2 and e2 versus ̟1 − ̟2 are the trajectories marked
by the squares (which denote the current parameters of the HD 12661 system) in Figure 6. The
direct integration results are not sensitive to the starting epoch. The amplitudes of eccentricity
oscillations and the trajectories in the diagram of e1 (or e2 ) versus ̟1 − ̟2 predicted by the
octupole theory are in reasonably good agreement with those from direct integrations, even though
α(= 0.323) is quite large. The main error of the octupole theory is in the period of eccentricity
oscillations, with the predicted period (≈ 2.1 × 104 yr) about 75% longer than that found in the
 – 23 –

Fig. 6.— Same as Fig. 2, but for two-planet systems with the same masses, initial semimajor axes,
and total angular momentum as the sin i = 1 HD 12661 system with initial P2 /P1 = 0.99 × 11/2.
The phase space of these systems with λ(≈ 0.83) almost identical to λcrit (≈ 0.82) is dominated
by large libration islands about ̟1 − ̟2 = 0◦ and 180◦ , and libration in the secular resonance is
almost the only possibility. The trajectories through the squares are those of the HD 12661 system,
with the squares showing the current parameters of the system. The initial conditions for the other
trajectories are ̟1 − ̟2 = 0◦ and e2 = 0.01, 0.05, 0.10, 0.165, or 0.24, or ̟1 − ̟2 = 180◦ and
e2 = 0.01, 0.05, 0.10 or 0.24, with e1 being determined from the total angular momentum. All the
direct integrations (except that for the HD 12661 system) start with the outer planet at apoapse
and the inner planet at opposition.

direct integrations (≈ 1.2 × 104 yr).

Figures 5 and 6 show some interesting properties of the HD 12661 system that are shared
by all of the cases studied above with different initial P2 , including those affected by the close
proximity to a mean-motion commensurability. The HD 12661 system is in a secular resonance
with ̟1 − ̟2 librating about 180◦ . This means that the lines of apsides of the two orbits are
on average anti-aligned. However, the amplitude of libration of ̟1 − ̟2 is large, and the orbital
eccentricities of both planets exhibit large-amplitude variations. For the direct integrations shown
in Figures 5 and 6, the amplitude of libration of ̟1 − ̟2 is 56◦ , e1 varies between 0.09 and 0.37,
and e2 varies between 0.17 and 0.37. HD 12661 is the first extrasolar planetary system found to
have ̟1 − ̟2 librating about 180◦ . The outer two planets of the υ And system (Rivera & Lissauer
 – 24 –

2000; Lissauer & Rivera 2001; Chiang et al. 2001) and the two planets about GJ 876 (Laughlin &
Chambers 2001; Lee & Peale 2002) are also likely in the secular resonance involving ̟1 − ̟2 , but
in both of these cases ̟1 − ̟2 librates about 0◦ . (In the case of GJ 876, the secular resonance is
associated with the simultaneous librations of both lowest order mean-motion resonance variables
at the 2:1 commensurability.)

The libration of ̟1 − ̟2 and the large-amplitude variations of both ej in the HD 12661 system
were anticipated by the analytic results derived in § 3.3. As we discussed in § 3.3, the octupole
theory predicts that large libration islands and large-amplitude variations of both ej are likely if
λ = L1 /L2 ≈ λcrit = 2γ 2 /(5 − 3γ 2 ). The HD 12661 system has λ ≈ 0.83, which is almost identical
to λcrit ≈ 0.82 for the dimensionless total angular momentum γ ≈ 0.96 of this system. In addition
to the trajectories of the sin i = 1 HD 12661 system with P2 /P1 = 0.99 × 11/2, Figure 6 also
shows the trajectories of systems with the same masses, initial semimajor axes, and total angular
momentum. The direct-integration results (left panels) are in reasonably good agreement with the
octupole results (right panels) and confirm that the phase space is dominated by large libration
islands about ̟1 − ̟2 = 0◦ and 180◦ . The only discrepancy is a narrow region of circulation
represented by a single trajectory in the direct-integration results. Thus libration of ̟1 − ̟2 is
almost the only possibility. We can see in Figure 6 that e1 decreases from near its maximum
possible value to near zero and e2 increases from near zero to near its maximum possible value at
̟1 − ̟2 ≈ 90◦ , and vice versa at ̟1 − ̟2 ≈ 270◦ , again in agreement with the analysis in § 3.3.

Kiseleva-Eggleton et al. (2002) have recently reported that the sin i = 1 HD 12661 system
with the best-fit orbital parameters in Table 1 is regular based on the MEGNO (Mean Exponential
Growth of Nearby Orbits) technique. However, they assumed that the best-fit orbital parameters
are in astrocentric coordinates and showed results from only one (unspecified) starting epoch. We
have repeated the direct integrations shown in Figure 4, but with the best-fit orbital parameters
assumed to be in astrocentric coordinates. Because of the larger fluctuations in the astrocentric
a2 , the average ratio of the orbital periods is more sensitive to the starting epoch. The integration
starting at Tperi,1 turns out to be sufficiently far from the 11:2 commensurability that it shows
regular secular evolution similar to those in Figure 5. The integration starting at Tperi,2 is qualita-
tively similar to the integration in Jacobi coordinates starting at the same epoch (dotted lines in
Figure 4) and is thus either regular or very weakly chaotic. These results again demonstrate the
importance of interpreting the orbital parameters from multiple Kepler fits as orbital parameters
in Jacobi coordinates.

The best-fit orbital parameters of the HD 12661 planets in Table 1 were provided by D. A. Fis-
cher (2002, private communication), and some of them differ slightly from those published later
in Fischer et al. (2003). Adopting the parameters in Fischer et al. (2003) does not qualitatively
change the results presented here, although the smaller ̟1 − ̟2 for the parameters in Fischer et al.
(2003) does lead to slightly larger amplitude of libration of ̟1 − ̟2 and slightly larger variations
in ej .
 – 25 –

After this paper was completed, we learned of contemporaneous work on the HD 12661 system
by Goździewski (2003) and Goździewski & Maciejewski (2003). Goździewski (2003) has examined
a wide region of the parameter space near the HD 12661 best-fit solution in Table 1, using as-
trocentric coordinates and a combination of the MEGNO technique and direct integrations. His
numerical results also show the proximity of the best-fit solution to the 11:2 commensurability and
the robustness of the libration of ̟1 − ̟2 . Goździewski & Maciejewski (2003) have subsequently
reported a best-fit solution to the data set published by Fischer et al. (2003) with P2 ≈ 1660 days
or P2 /P1 ≈ 6.3. This illustrates the uncertainties mentioned above in the orbital parameters of
the outer planet due to the time span of the available observations being comparable to P2 . Their
best-fit solution also shows libration of ̟1 − ̟2 about 180◦ and large-amplitude variations of both
ej .

Mayor et al. (2003) have reported the detection of two planets around the star HD 83443.
There is considerable doubt about the existence of the outer planet in this system (Butler et al.
2002), but it is interesting to note that the system as reported by Mayor et al. shares some of
the properties of the HD 12661 system. The best-fit orbital periods of the HD 83443 planets are
nearly in the ratio 10:1, and direct integrations of this system (neglecting the effects of tides and
general relativistic precession) show that its evolution is also affected by the close proximity to the
mean-motion commensurability. The HD 83443 system also shows large-amplitude variations of
both eccentricities, although its ̟1 − ̟2 is circulating. The value of λ(≈ 0.96) of the HD 83443
system is not as close to its value of λcrit (≈ 0.79) as in the HD 12661 system, and there is a larger
region of circulation in the phase space of systems with the same masses, initial semimajor axes,
and total angular momentum as the HD 83443 system. It should be noted that the inner planet of
the HD 83443 system is sufficiently close to the star (P ≈ 3 days) that tides and general relativistic
precession can significantly affect the dynamics of the system (Wu & Goldreich 2002).

4.3. Small Mutual Inclination

To study the effects of a small mutual inclination angle imu between the orbits, we have repeated
the direct integrations of the HD 168443 and HD 12661 systems shown in Figures 1b and 5 with
initial imu = 1◦ or 2◦ . We assumed that the intersection of the orbital planes is initially along the
line of sight, so that, initially, sin i1 = sin i2 = 1 and the mutual inclination is the difference in
the longitudes of ascending node referenced to the plane of the sky: imu = Ω2 − Ω1 . The results
are nearly identical to those with coplanar orbits, confirming the analysis in § 3.1 based on an
expansion of the more general form of the octupole theory in powers of imu .
 – 26 –

4.4. Stability

The applicability of the octupole theory is limited by its use of an expansion in r1 /r2 and orbit
averaging. In particular, the orbit averaging eliminates the effects of mean-motion commensurabil-
ities and the possible development of instabilities. The direct integrations reported in § 4.1 show
that the coplanar HD 168443 system is stable for sin i & 0.4, and Marcy et al. (2001) have found
that this system is stable for even smaller sin i. The stability of the HD 168443 system is consis-
tent with the empirical stability criteria for hierarchical triple systems derived most recently by
Eggleton & Kiseleva (1995) and Mardling & Aarseth (2001). For example, the sin i = 1 HD 168443
system has α = a1 /a2 = 0.102, while the stability criterion of Eggleton & Kiseleva requires ini-
tial a2 (1 − e2 )/[a1 (1 + e1 )] & 1.65 or α . 0.316 and that of Mardling & Aarseth requires initial
a2 (1 − e2 )/a1 & 3.17 or α . 0.252. The criterion of Eggleton & Kiseleva was tested over a wide
range of mass ratios (including ones similar to those of the HD 168443 system) and is expected
to be reliable to about 20%, while the criterion of Mardling & Aarseth does not have explicit de-
pendence on m1 /m0 and may not be very accurate for m1 /m0 significantly different from unity.
For a more detailed comparison with the above stability criteria, we have performed a set of direct
integrations with different initial a2 : a2 = 1.299–0.738 AU or α = 0.227–0.400. The other initial
conditions are identical to those of the sin i = 1 calculation shown as the solid lines in Figure 1b.
The evolutions of α are shown in Figure 7. Except for the case with initial α = 0.333, which is
apparently stabilized by the 5:1 mean-motion commensurability, there is a clear boundary between
stable and unstable configurations at initial α ≈ 0.30. We do not expect this stability boundary to
move down significantly if we increase the integration length, since 105 yr span many eccentricity
oscillation cycles (if the orbits are stable) and, as we can see in Figure 7, instabilities usually man-
ifest themselves on much shorter timescales. The stability criterion of Eggleton & Kiseleva is a few
percent larger than the stability boundary determined here, consistent with the ≈ 20% accuracy
of the former. Even the stability criterion of Mardling & Aarseth is only 16% smaller than the
stability boundary determined here (see, however, the next paragraph). Finally, we note that some
of the configurations in Figure 7 that are stable for at least 105 yr appear to be chaotic (e.g., the
case with initial α = 1/4). As in the case of the HD 12661 system, the chaos is due to the close
proximity to a high-order mean-motion commensurability (8:1 for the case with initial α = 1/4).

In § 4.2 we have considered the coplanar HD 12661 system with sin i = 1 (i.e., minimum
planetary masses) only. Unlike the HD 168443 system, which has α = a1 /a2 ∼ 0.1 and is stable
for any reasonable value of sin i, the HD 12661 system has α ∼ 0.32 (which is close to the stability
boundary for the systems shown in Fig. 7 with the planetary masses of the sin i = 1 HD 168443
system) and could be unstable for moderately small sin i. To study the effects of sin i, we have
performed two sets of direct integrations with different sin i. The initial values of Pj , Kj , ej , ωj ,
and Tperi,j are the same as those of the HD 12661 system with P2 /P1 = 0.99 × 11/2 shown in
Figure 5, and the planetary masses and initial semimajor axes were derived for the assumed sin i.
The evolutions of α are shown in Figure 8 for sin i = 0.1, 0.2, . . . , 1. The left and right panels are
from the direct integrations starting at Tperi,1 and Tperi,2 , respectively. The dashed line in each
 – 27 –

Fig. 7.— Evolutions of a1 /a2 = α for a set of direct integrations with different initial a2 : a2 = 1.299–
0.738 AU or α = 0.227–0.400. The other initial conditions are identical to those of the sin i = 1
HD 168443 calculation shown as the solid lines in Fig. 1b. The dashed line in each panel indicates the
initial α. Some of the cases that are stable for at least 105 yr are affected by the close proximity to
a high-order mean-motion commensurability (e.g., 8:1 for the case with initial α = 1/4) and appear
to be chaotic. The case with initial α = 0.333 is apparently stabilized by the 5:1 commensurability.

panel indicates the location of the 11:2 mean-motion commensurability. The influence of the 11:2
commensurability increases with decreasing sin i (which is expected since the width of a resonance
increases with increasing planetary masses), but the cases with sin i > 0.3 are stable (for at least
2 × 105 yr). All of the cases with sin i . 0.3 are unstable. Are the stability criteria of Eggleton
& Kiseleva (1995) and Mardling & Aarseth (2001) consistent with this transition from stable to
unstable configurations at sin i ≈ 0.3? For the configurations shown in Figure 8, which have the
same initial P1 and P2 , the initial α decreases slightly from 0.323 to 0.322 as sin i decreases from
1 to 0.1 (and m1 and m2 increase; see eqs. [2] and [6]). For comparison, the stability boundary
predicted by the criterion of Eggleton & Kiseleva decreases from initial α = 0.444 to 0.348 as sin i
decreases from 1 to 0.1, with initial α = 0.397 at sin i = 0.3. This stability boundary is consistent
with the transition at sin i ≈ 0.3 if we simply move it down by 23%, which is compatible with
the ≈ 20% accuracy of the stability criterion of Eggleton & Kiseleva. On the other hand, the
stability boundary predicted by the criterion of Mardling & Aarseth changes only slightly from
 – 28 –

Fig. 8.— Evolutions of a1 /a2 = α for the HD 12661 system with P2 /P1 = 0.99×11/2 and sin i = 0.1,
0.2, . . . , 1. The initial values of Pj , Kj , ej , ωj , and Tperi,j are the same as those of the system shown
in Fig. 5, and the planetary masses and initial semimajor axes were derived for the assumed sin i.
The left and right panels are from direct integrations starting at Tperi,1 and Tperi,2 , respectively.
The dashed line in each panel indicates the location of the 11:2 mean-motion commensurability.

initial α = 0.254 to 0.253 as sin i decreases from 1 to 0.1. Thus, this stability boundary is not
consistent with the transition at sin i ≈ 0.3, unless we allow adjustments that change with sin i.
This is not surprising since the stability criterion of Mardling & Aarseth does not have explicit
dependence on m1 /m0 and is not expected to be very accurate for m1 /m0 significantly different
from unity.

5. CONCLUSIONS

We have investigated the dynamical evolution of coplanar hierarchical two-planet systems

where the ratio of the orbital semimajor axes α = a1 /a2 is small. Hierarchical two-planet systems
are common among the known multiple planet systems and are likely to be common in the overall
population of extrasolar planetary systems. We began by showing that the orbital parameters
obtained from a two (or more generally multiple) Kepler fit to the radial velocity variations of a
host star are best interpreted as Jacobi coordinates and that these coordinates should be used in any
 – 29 –

analyses of hierarchical (and possibly other types of) planetary systems. We used the HD 168443
system as an example to show that the use of astrocentric coordinates can introduce significant
high-frequency variations in orbital elements that should be nearly constant on orbital timescales,
leading to erroneous sensitivity of the evolution of the orbital elements on the starting epoch.

An approximate theory that can be applied to hierarchical two-planet systems with a wide
range of masses, orbital eccentricities, and inclinations is the octupole-level secular perturbation
theory, which was developed for general hierarchical triple systems by Marchal (1990), Krymolowski
& Mazeh (1999), and Ford et al. (2000). The octupole theory is based on an expansion to order
α3 and orbit-averaging. We showed that the octupole approximation reduces the coplanar problem
to one degree of freedom, with e1 (or e2 ) and ̟1 − ̟2 as the relevant phase-space variables. An
analysis of the octupole equations yielded the following results: (1) the scaling properties of the
equations imply that the amplitudes of eccentricity oscillations and the trajectory in the phase-space
diagram of e1 (or e2 ) versus ̟1 − ̟2 should be independent of the inclination i of both orbits from
the plane of the sky (as long as m1 , m2 ≪ m0 ) and that the period of eccentricity oscillations should
be proportional to sin i; (2) the fixed points in the phase-space diagram must be at ̟1 − ̟2 = 0◦ or
180◦ ; and (3) if the ratio of the maximum orbital angular momenta, λ = L1 /L2 ≈ (m1 /m2 )α1/2 , for
given semimajor axes is approximately equal to λcrit = 2γ 2 /(5−3γ 2 ), where γ = (G1 +G2 )/(L1 +L2 )
is the dimensionless total angular momentum, then libration of ̟1 − ̟2 is almost certain, with
possibly large amplitude variations of both eccentricities.

We used both the octupole-level secular perturbation theory and direct numerical orbit in-
tegrations to study the dynamical evolution of the HD 168443 and HD 12661 systems and their
variants. The HD 168443 system is not in a secular resonance and its ̟1 − ̟2 circulates. For the
family of systems with the same masses, initial semimajor axes, and total angular momentum as
the sin i = 1 HD 168443 system, there are two relatively small libration islands in the phase-space
diagram of e1 (or e2 ) versus ̟1 − ̟2 , one about a fixed point at ̟1 − ̟2 = 0◦ and the other
about a fixed point at ̟1 − ̟2 = 180◦ , and the trajectory of the HD 168443 system is far from
both. In all cases, the octupole results are in excellent agreement with the direct-integration results.
Direct integrations of the HD 168443 system with sin i = 0.4 also confirmed the analytic octupole
results on the effects of sin i. To demonstrate how the octupole theory can be used to explore the
parameter space rapidly, we used the octupole theory to determine the number and positions of
fixed points as a function of the total angular momentum for systems with the same masses and
semimajor axes as the sin i = 1 HD 168443 system. For sufficiently low total angular momentum,
the existence of fixed point(s) in addition to those appropriate for the total angular momentum of
the HD 168443 system leads to systems with unusual behaviors such as e1 increasing to unity in
a finite time or libration of ̟1 − ̟2 about 0◦ with e1 very close to 1. These results were again
confirmed by direct integrations. Direct integrations of systems similar to the sin i = 1 HD 168443
system, but with different initial a2 , showed that these systems with relatively massive planets
[(m1 + m2 )/m0 ≈ 0.024] are unstable if the initial α & 0.30.

The HD 12661 system is the first extrasolar planetary system found to have ̟1 − ̟2 librating
 – 30 –

about 180◦ . The secular resonance means that the lines of apsides of the two orbits are on average
anti-aligned, although the amplitude of libration of ̟1 − ̟2 is large. The libration of ̟1 − ̟2
and the large-amplitude variations of both orbital eccentricities in the HD 12661 system, which
has λ ≈ 0.83 and λcrit ≈ 0.82, are consistent with the analytic results on systems with λ ≈
λcrit . Direct integrations showed that the evolution of the sin i = 1 HD 12661 system with the
best-fit orbital parameters (P2 /P1 = 0.9975 × 11/2) is affected by the close proximity to the
11:2 mean-motion commensurability, but that small changes in the orbital period of the outer
planet within the uncertainty can result in configurations that are not affected by mean-motion
commensurabilities. For the family of systems with the same masses, initial semimajor axes, and
total angular momentum as the sin i = 1 HD 12661 system with P2 /P1 = 0.99× 11/2, which are not
affected by mean-motion commensurabilities, the octupole results are in reasonably good agreement
with the direct-integration results, even though α(≈ 0.32) is quite large. The phase space of this
family of systems with λ ≈ λcrit is dominated by large libration islands about ̟1 − ̟2 = 0◦ and
180◦ , which again confirmed the analytic results on systems with λ ≈ λcrit , and libration in the
secular resonance is almost the only possibility. Finally, we showed that the HD 12661 system is
unstable if sin i . 0.3 [or (m1 + m2 )/m0 & 0.012].

The comparisons with direct numerical orbit integrations established that the octupole-level
secular perturbation theory is highly accurate for coplanar hierarchical two-planet systems with
α . 0.1 and reasonably accurate even for systems with α as large as 1/3, provided that α is
not too close to a significant mean-motion commensurability or above the stability boundary. We
have focused our study in this paper on coplanar systems in which the star and the planets can
be treated as point masses interacting via Newtonian gravity, where a small mutual inclination
between the orbits is shown to have little effect on the secular evolution. The more general form
of the octupole theory is applicable to systems with arbitrary mutual inclinations. If the inner or
both planets are very close to the star, tidal interactions and general relativistic precession can be
important (see, e.g., Wu & Goldreich 2002). Nagasawa et al. (2003) have also discussed the effects
of additional orbital precessions due to interactions with the protoplanetary disk during the epoch
of disk depletion. It is straightforward to modify the octupole theory to include any of these effects
(see, e.g., Blaes et al. 2002 for relativistic precession).

We thank Debra Fischer for furnishing the orbital parameters of the HD 12661 system before
publication. We also thank O. Blaes, G. W. Marcy, N. Murray, E. J. Rivera, and Y. Wu for infor-
mative discussions and an anonymous referee for comments that greatly improved the presentation.
This research was supported in part by NASA grants NAG5 11666 and NAG5 12087.
 – 31 –

REFERENCES

Blaes, O., Lee, M. H., & Socrates, A. 2002, ApJ, 578, 775

Butler, R. P., Marcy, G. W., Vogt, S. S., Tinney, C. G., Jones, H. R. A., McCarthy, C., Penny, A.
J., Apps, K., & Carter, B. D. 2002, ApJ, 578, 565

Chiang, E. I., Tabachnik, S., & Tremaine, S. 2001, AJ, 122, 1607

Eggleton, P., & Kiseleva, L. 1995, ApJ, 455, 640

Fischer, D. A., Marcy, G. W., Butler, R. P., Vogt, S. S., Frink, S., & Apps, K. 2001, ApJ, 551,
1107

Fischer, D. A., Marcy, G. W., Butler, R. P., Vogt, S. S., Henry, G. W., Pourbaix, D., Walp, B.,
Misch, A. A., & Wright, J. 2003, ApJ, 586, 1394

Ford, E. B., Kozinsky, B., & Rasio, F. A. 2000, ApJ, 535, 385

Goździewski, K. 2003, A&A, 398, 1151

Goździewski, K., & Maciejewski, A. J. 2001, ApJ, 563, L81

Goździewski, K., & Maciejewski, A. J. 2003, ApJ, 586, L153

Harrington, R. S. 1968, AJ, 73, 190

Holman, M. J., & Murray, N. W. 1996, AJ, 112, 1278

Kiseleva-Eggleton, L., Bois, E., Rambaux, N., Dvorak, R. 2002, ApJ, 578, L145

Kozai, Y. 1962, AJ, 67, 591

Krymolowski, Y., & Mazeh, T. 1999, MNRAS, 304, 720

Laughlin, G., & Chambers, J. E. 2001, ApJ, 551, L109

Lee, M. H., & Peale, S. J. 2002, ApJ, 567, 596

Lee, M. H., & Peale, S. J. 2003, in Scientific Frontiers in Research on Extrasolar Planets, ed. D.
Deming & S. Seager (San Francisco: ASP), in press (astro-ph/0209176)

Lissauer, J. J., & Rivera, E. J. 2001, ApJ, 554, 1141

Marchal, C. 1990, The Three-Body Problem (Amsterdam: Elsevier)

Marcy, G. W., Butler, R. P., Vogt, S. S., Liu, M. C., Laughlin, G., Apps, K., Graham, J. R., Lloyd,
J., Luhman, K. L., & Jayawardhana, R. 2001, ApJ, 555, 418
 – 32 –

Marcy, G. W., Butler, R. P., Fischer, D. A., Laughlin, G., Vogt, S. S., Henry, G. W., Pourbaix, D.
2002, ApJ, 581, 1375

Mardling, R. A., & Aarseth, S. J. 2001, MNRAS, 321, 398

Mayor, M., Naef, D., Pepe, F., Queloz, D., Santos, N. C., Udry, S., & Burnet, M. 2003, in Plan-
etary Systems in the Universe: Observation, Formation and Evolution, ed. A. J. Penny, P.
Artymowicz, A.-M. Lagrange, & S. S. Russell (San Francisco: ASP), in press

Murray, C. D., & Dermott, S. F. 1999, Solar System Dynamics (Cambridge: Cambridge Univ.
Press)

Murray, N., & Holman, M. 1997, AJ, 114, 1246

Nagasawa, M., Lin, D. N. C., & Ida, S. 2003, ApJ, 586, 1374

Nauenberg, M. 2002, ApJ, 568, 369

Rivera, E. J., & Lissauer, J. J. 2000, ApJ, 530, 454

Rivera, E. J., & Lissauer, J. J. 2001, ApJ, 558, 392

Udry, S., Mayor, M., & Queloz, D. 2003, in Planetary Systems in the Universe: Observation,
Formation and Evolution, ed. A. J. Penny, P. Artymowicz, A.-M. Lagrange, & S. S. Russell
(San Francisco: ASP), in press

Wisdom, J., & Holman, M. 1991, AJ, 102, 1528

Wu, Y., & Goldreich, P. 2002, ApJ, 564, 1024

This preprint was prepared with the AAS LATEX macros v5.0.