Draft version June 4, 2019 Typeset using LATEX twocolumn style in AASTeX62

Not a simple relationship between Neptune’s migration speed and Kuiper belt inclination excitation

Kathryn Volk1 and Renu Malhotra1

1Lunar and Planetary Laboratory, The University of Arizona, 1629 E University Blvd, Tucson, AZ 85721

ABSTRACT We present numerical simulations of giant planet migration in our solar system and examine how the speed of planetary migration affects inclinations in the resulting population of small bodies (test particles) scattered outward and subsequently captured into Neptune’s 3:2 mean motion resonance (the Plutinos) as well as the hot classical Kuiper belt population. We do not find a consistent relationship between the degree of test particle inclination excitation and e-folding planet migration timescales in the range 5 − 50 Myr. Our results present a counter-example to Nesvorný(2015)’s finding that the Plutino and hot classical inclinations showed a marked increase with increasing e-folding timescales for Neptune’s migration. We argue that these differing results are likely due to differing secular architectures of the giant planets during and after migration. Small changes in the planets’ initial conditions and differences in the numerical implementation of planet migration can result in different amplitudes of the planets’ inclination secular modes, and this can lead to different final inclination distributions for test particles in the simulations. We conclude that the observed large inclination dispersion of Kuiper belt objects does not require Neptune’s migration to be slow; planetary migration with e-folding timescales of 5, 10, 30, and 50 Myr can all yield inclination dispersions similar to the observed Plutino and hot classical populations, with no correlation between the degree of inclination excitation and migration speed.

Keywords: Kuiper belt: general

1. INTRODUCTION cycle. Inclination excitation of KBOs has also been ex- The orbital planes of Kuiper belt objects (KBOs) are amined in the context of a giant planet instability during widely dispersed, and dynamical sub-classes of KBOs which Neptune may have been scattered out close to its have measurably different inclination distributions (see, current semi-major axis but on a somewhat eccentric or- e.g., Gulbis et al. 2010; Petit et al. 2011; Gladman et al. bit and subsequently experienced only a short-distance 2012; Petit et al. 2017). The large inclinations of some migration and eccentricity damping (e.g., Tsiganis et al. observed Kuiper belt objects have prompted several the- 2005). Levison et al.(2008) found that some simula- oretical studies of inclination excitation during plane- tions of such giant planet instability produced large in- tary migration. Malhotra(1995) suggested that slower clinations in the resulting Kuiper belt population while migration of Neptune correlated with higher inclina- others did not. tion excitation due to argument-of-perihelion libration In a recent numerical study, Nesvorný(2015) found within mean motion resonances. Gomes(2003) detailed that slower migration timescales for Neptune led to arXiv:1906.00023v1 [astro-ph.EP] 31 May 2019 how a high-inclination population could be produced larger inclinations in the hot classical population and in during migration as scattered KBOs were temporar- the Plutino population (objects in Neptune’s 3:2 mean ily captured into Neptune’s mean motion resonances, motion resonance), concluding models in which Neptune their inclinations excited by the secular Kozai-Lidov cy- migrated from an initial semimajor axis of aN,0 . 25 au cles, then became resonance drop-outs during the low- to its current orbit on timescales τa & 10 Myr pro- eccentricity/high-inclination phase of the Kozai-Lidov vided the best match to observed inclination distribu- tions. In this scenario, the present-day hot classical and resonant populations consist of objects originating Corresponding author: Kathryn Volk in dynamically cold orbits interior to Neptune’s current [email protected] semi-major axis (∼ 30 au) that were scattered and dis- persed by the outward migration of Neptune. During 2 this outward migration, the scattered objects have a those of Nesvorný(2015). We summarize and conclude propensity to stick to Neptune’s migrating mean motion in Section4. resonances and undergo secular cycling of eccentricity and inclination within those resonances, as described by 2. NUMERICAL SIMULATIONS OF PLANETARY Gomes(2003); most are lost from the Kuiper belt dur- MIGRATION ing this process, but some end up being implanted into In order for planetary migration simulations to con- long-term stable orbits in the hot classical region or in tain enough particles to sufficiently explore the orbital Neptune’s exterior mean motion resonances. Nesvorný architecture of the resulting Kuiper belt populations, (2015) argued that slower migration timescales allowed many simplifying assumptions must be made. In Sec- more time for both encounters with Neptune and secular tion 2.1 we describe some of the challenges of simulat- effects within mean motion resonances to excite parti- ing planetary migration, outlining our approach to the cle inclinations, possibly accounting for the link between problem and comparing that to those used in the liter- migration timescale and more widely dispersed final in- ature. Section 2.2 provides a detailed description of our clinations in those simulations. numerical simulations and initial conditions. In this paper, we describe our effort to understand the dynamical mechanisms underlying the inclination 2.1. Background and Choice of Migration Scheme excitation that occurs during planetary migration us- In the absence of computational limitations, one ing simulations broadly similar to those of Nesvorný would simulate the planetesimal driven outward mi- (2015). To our surprise, we find that the degree of in- gration of Neptune to its current orbit by modeling the clination excitation of the Plutinos and the hot classical full gravitational interaction between the planets and Kuiper belt is not simply monotonically dependent on an initial population of massive proto-Kuiper belt ob- Neptune’s migration timescale, but rather appears to jects in an N-body simulation. In such a simulation, the depend sensitively on the strength of the secular incli- semimajor axis, eccentricity and inclination evolution nation modes of the planets during migration (which for Neptune (and the other planets) would occur self- in turn depends sensitively on the orbital evolution of consistently as the planets interact with a large number the planets). While the simplified migration simulations of self-gravitating objects with a realistic mass distribu- presented here are not meant to mimic the full dynami- tion. However, this approach is not currently feasible. cal history of the outer solar system, they are useful for Full N-body simulations of the planet migration process demonstrating that the long-timescale inclination exci- are typically limited to interactions between the planets tation mechanisms associated with scattering and mean and a moderate population of non-self-interacting plan- motion resonance sticking are not always the dominant etesimals. The planetesimal population is usually much source of inclination excitation during Neptune’s migra- fewer in number and thus comprised of objects that are tion. We instead find that significant secular inclination individually more massive than expected in the condi- excitation is possible on timescales that are short com- tions in the real early solar system; see, for example, the pared to typical migration timescales, especially in the N-body work in the original Nice-model papers, where vicinity of the ν secular inclination resonance. 18 the proto-Kuiper belt is represented by ∼ 1000 objects This means that the widely dispersed inclinations in with individual masses ranging from a few Pluto masses the Kuiper belt cannot be used to definitively argue to a few Lunar masses (Tsiganis et al. 2005). The neglect for slow Neptune migration. An improved, more de- of self-gravity between the planetesimals in such popula- tailed understanding of how inclinations are excited dur- tions has, at least, been shown to not have a significant ing planetary migration is necessary in order to use the impact on the final orbits of the planets (Fan & Batygin inclinations to constrain the history of the outer solar 2017). However, the planets in these simulations experi- system. This is particularly important because inclina- ence a smaller number of stronger interactions with the tion distributions are often better observationally con- planetesimals to arrive at their final orbits rather than a strained than other orbital parameters. large number of weaker interactions, which will impact The rest of this paper is organized as follows. In Sec- the time-evolution of their orbits. The computational tion2, we describe our simulations and results for planet challenges of such simulations are further compounded migration timescales spanning an order of magnitude in by their chaotic nature (although we note below that e-folding timescales (τ = 5 − 50 Myr) for the plan- a even simplified simulations suffer from this problem); a ets’ semi-major axis evolution. In Section3, we offer large number of sets of initial conditions must be tried an analysis of the discrepancies between our results and before an acceptable outcome (where the giant planets end on orbits similar to their current ones) can be found migration timescales and Kuiper belt inclinations 3

(as discussed in, e.g., Nesvorný 2011; Nesvorný & Mor- sired final orbits. As we discuss further in Section3, bidelli 2012; Gomes et al. 2018). Even when successful it is unclear how these different choices about whether simulation initial conditions are found, there are often and how to implement planetary eccentricity/inclination too few of the planetesimals surviving to the end in damping affect the simulation outcomes for the test par- the simulation to have a statistically useful sample for ticles. comparison with observations of Kuiper belt objects, es- In order to limit the number of free parameters in our pecially for comparisons to the dynamical sub-classes of migration simulations, we chose to find initial conditions the Kuiper belt (though for large suites of simulations, that avoid overly exciting the planets’ eccentricities and comparisons are sometimes possible, see, e.g., Gomes inclinations, eliminating the need for eccentricity and et al. 2018). inclination damping. As described in more detail in These computational limitations mean that in sim- Section 2.2, we assign the planets’ initial eccentricities ulations intended to explore the origins of the Kuiper and inclinations to be similar to their current observed belt dynamical structures, the planetesimal-driven mi- values, run a large set of trial simulations, then choose gration of the planets is often instead modeled by using the planetary initial conditions that result in satisfac- prescribed extra forces on the planets to cause their or- tory final planetary orbits. (We do not directly sim- bital migration, allowing the dynamical evolution of the ulate scenarios in which Neptune’s initial eccentricity Kuiper belt to be tracked with large numbers of massless and inclination is excited to large values by close en- test particles (see, e.g., Hahn & Malhotra 2005; Levi- counters with other planets.) This is sufficient for the son et al. 2008; Nesvorný 2015; Kaib & Sheppard 2016). purposes of our study because the correlation between This “short cut" allows better statistics for the orbital migration timescale and inclination dispersion found by distribution of a final Kuiper belt model. The drawback Nesvorný(2015) did not depend on the initial assumed of this approach is that different choices can be made as eccentricity or inclination for Neptune. For simplicity, to how migration is numerically implemented, includ- we also do not consider additional complications, such ing whether and how the planets’ orbital inclinations as the smoothness or graininess of Neptune’s migration and eccentricities are damped during their migration. (as investigated by, e.g., Nesvorný & Vokrouhlický 2016; For example, smooth planet migration can be modeled Kaib & Sheppard 2016), which would depend on the most simply by applying a smoothly declining torque mass distribution of the planetesimals driving the plan- to each planet such that its semimajor axis approaches ets’ migration; we simply implement smooth semimajor its current observed value with the desired e-folding mi- axis migration in the same way as Hahn & Malhotra gration timescale; this is the approach taken in many (2005). Our conclusion that the secular architecture of studies (e.g. Malhotra 1993; Gomes 2000; Chiang & Jor- the planetary system can strongly affect the inclination dan 2002; Hahn & Malhotra 2005; Brasser et al. 2009; distributions in the final Kuiper belt populations should Dawson & Murray-Clay 2012; Nesvorný 2015). For suit- hold even for grainy migration, although the efficiency of able initial conditions, the planets in such simulations trapping test particles into the Plutino and hot classical can maintain eccentricities and inclinations similar to populations could differ. their currently observed ones throughout the duration of the simulation. In contrast, works such as Levison 2.2. Our Numerical Simulations et al.(2008) include eccentricity damping forces in or- We performed our suite of planetary migration sim- der to model a post-instability giant planet system such ulations using the integrator within the as proposed in Tsiganis et al.(2005), wherein Neptune hermes re- software package (the initial release of which starts with a large eccentricity that damps down as it bound is described in Rein 2012). This integrator utilizes a migrates several au to approximately its current orbit. Wisdom-Holman scheme (Wisdom & Holman 1991) re- Other works, such as Dawson & Murray-Clay(2012) ferred to as the routine (described in Rein and Wolff et al.(2012), also choose to include eccentric- whfast & Tamayo 2015) for the majority of the simulation’s ity damping, but use a different numerical implementa- timesteps and switches to the adaptive stepsize tion. Whereas Levison et al.(2008) implement an extra ias15 routine (based on Everhart 1985) to resolve close en- force based on expected friction in nebular gas, Wolff counters between objects in the simulation (described in et al.(2012) use analytical orbit perturbation equations Rein & Spiegel 2015). At each timestep, we implement to construct an extra force that yields the desired eccen- a user-defined force that causes the planets’ semimajor tricity damping behavior. Nesvorný(2015) implemented axes to approach their current values with a prescribed both eccentricity and inclination damping parameter- e-folding timescale (Hahn & Malhotra 2005). As de- ized by e-folding timescales to evolve to the planets’ de- scribed below, we chose initial conditions for which no 4 close encounters occur between the planets, so their evo- a challenge when running simulations on computer clus- lution is entirely calculated with whfast. We chose to ters with multiple node architectures; it also means that use rebound (instead of one of the other hybrid numer- numerical experiments cannot be tested with additional ical integration schemes in common use for solar system test particles or for extended integration lengths if the dynamics, such as Mercury (Chambers 1999) or SWIFT original machine is no longer available. The ability to (Duncan & Levison 1997)) because rebound uniquely remove machine- and compiler-dependent round off er- allows for bit-wise reproducible results regardless of the ror as a source of chaos in the numerical integrator (as choice of compiler or the operating system of the com- done by Rein & Tamayo 2017) is particularly helpful for puter on which the simulations are run when the same planet migration simulations. initial inputs are used (as described in Rein & Tamayo We investigated four migration e-folding timescales: 2017). To correct the unphysical situation of the planets’ τa = 5 Myr, 10 Myr, 30 Myr, and 50 Myr. For each evolution being affected by close encounters with mass- timescale, we ran a test suite of ∼ 500 simulations of less test particles, we made a slight modification to the just the four migrating giant planets. Jupiter and Sat- hermes routine to allow bitwise reproducible histories of urn were initialized on orbits with their current eccen- the planets’ orbits: after a call to the adaptive step size tricities and inclinations but semimajor axes of 5.4 au routine to resolve a close encounter between a planet and and 8.8 au, respectively. Uranus and Neptune were ini- a test particle, the integration for the massive bodies (in tialized on orbits of semimajor axes randomly chosen our case, the giant planets) resumes from the previous from the ranges 16.4 ± 0.035 au and 24 ± 0.035 au, re- primary timestep rather than resuming from the end- spectively; their inclinations were set to their current point of the close encounter; this modification amounts values, and their eccentricities were set randomly in the to discarding the planetary positions and velocities cal- range e = 0.005±0.002. We integrated each set of initial culated during the close encounter and instead smoothly conditions to t = 700 Myr and calculated the average continues the integration with the values calculated from semimajor axis value and the eccentricity ranges of all whfast. This modification ensures that the planets’ 4 planets over the last 100 Myr of the simulation. For orbital histories are calculated entirely with the widely each migration timescale, we then selected a set of initial used and tested Wisdom-Holman routine and are thus conditions that resulted in a planetary system that best unaffected by the presence or absence of test particles matched the current solar system in terms of semimajor in the simulation. We note that the rmvs3 routine in axis ratios and eccentricity ranges. The final eccentric- SWIFT requires a similar modification to ensure that ity and inclination ranges of the planets in these simu- interactions with massless test particles do not result in lations (taken over the last 100 Myr of the simulations) different planet outcomes. are shown in the top row of panels of Figure1. The fully bitwise reproducible planetary histories that For each combination of migration timescale and plan- result from this modified version of rebound signifi- ets’ initial conditions, we then re-integrated the planets cantly simplify the problem of simulating planetary mi- along with 3.5 × 105 massless test particles. The test gration. As noted above in Section 2.1, it can be chal- particles were given initial semimajor axes in the range lenging to find initial conditions for the planets that 24.5−30 au (from just outside Neptune’s initial orbit to result in post-migration orbits similar to their present just inside its current orbit); their eccentricities and in- ones; this is true in both full N-body migration sim- clinations were randomly drawn from a Raleigh distribu- ulations and simplified parameterized migration simu- tion of width 0.025 (where the units are in radians for the lations. We find that, especially for slow migration, inclinations). These test particle initial conditions are the planets often cross mutual mean motion resonances very similar to those used in Nesvorný(2015). The plan- slowly enough for eccentricities to be significantly ex- ets and test particles were integrated to t = 700 Myr. cited; this can sometimes lead to close-encounters be- This 700 Myr integration length was chosen as a compro- tween Uranus and Saturn or Uranus and Neptune, which mise between conserving computational time and reach- further excite the system. Because the exact excita- ing a post-migration final orbital distribution of test tion that results from resonances depends on the phases particles that are stable on ∼ gigayear timescales for of the planets when they encounter them, even tiny comparison with the present-day observed Kuiper belt. changes in the initial conditions can lead to wildly differ- (Our simulations required ∼ 2.5 × 104 cpu hours per ent outcomes. With integrators such as SWIFT, we have migration timescale; a total of ∼ 105 cpu hours were even found that running identical initial conditions on used in this study on the Ocelote cluster maintained by machines with different processors can produce different UA Research Computing High Performance Computing results, even if the same compiler is used. This presents at the University of Arizona.) It is pertinent to note migration timescales and Kuiper belt inclinations 5

1.8 2.5 2 2 N 1.6 τa = 5 Myr, A N τa = 10 Myr, A τa = 30 Myr, A τa = 50 Myr, A 2 1.4 N N 1.5 1.2 1.5 1.5 1 U U U S 1 U 1 U 0.8 S S S U S S S 1 U 0.6 N U S N N N inclination(deg) 0.5 0.4 J 0.5 0.5 J JJ JJ J 0.2 J 0 0 0 0 0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1

1.2 1.2 1.2 1.2

1 U 1 U 1 U 1 U S S S S 0.8 0.8 0.8 0.8 τa = 5 Myr, B τa = 10 Myr, B N τa = 30 Myr, B τa = 50 Myr, B N N N U N 0.6 0.6 0.6 0.6 N U N U N S 0.4 S 0.4 S 0.4 0.4 J J J S J

inclination(deg) U 0.2 0.2 0.2 0.2 J J J J 0 0 0 0 0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1 eccentricity eccentricity eccentricity eccentricity

Figure 1. The eccentricity and inclination ranges of the giant planets over the last 100 Myr in our migration simulations (green bars, labeled by planet) compared to those for the real solar system over the same timescale (black bars, labeled by planet). The top row of panels shows the ranges for our ‘A’ simulations (in which Neptune’s inclination is slightly larger than in the real solar system); the bottom row of panels shows the ‘B’ simulations (in which the planets’ inclinations are slightly lower than in the real solar system). All inclinations are measured relative to the plane defined by the planets’ total angular momentum vector (i.e., the invariable plane). that . 15% of the Plutino population is expected to end of these four simulations described above, Neptune’s have leaked out of Neptune’s 3:2 resonance in the past inclination is in the range 1.3◦–2.0◦, slightly larger than ∼ 3.5 gigayears, and that this loss is not strongly incli- in the real solar system (top row of panels in Figure1). nation dependent (Nesvorný & Roig 2000; Tiscareno & This results in more power being associated with Nep- Malhotra 2009). Therefore, our simulation length is suf- tune’s dominant inclination secular frequency than in ficient to test how the planet migration rate affects the fi- the real solar system. This realization led us to perform nal Plutino inclination distribution, the dynamical class an additional set of migration simulations for each mi- of KBOs for which Nesvorný 2015 found the strongest gration timescale. The initial conditions for the planets dependence on migration rate. The 700 Myr timescale is in this new set of simulations were chosen the same way also sufficient to investigate the inclination distribution as described above, except that we reduced the initial of the hot classical population because the inclination mutual inclinations of the giant planets to try to re- distribution of this population does not evolve signifi- duce Neptune’s final inclination to more closely match cantly on long timescales (e.g., Volk & Malhotra 2011). the real solar system. After running a suite of ∼ 500 The only region of the classical belt that has strongly planets-only migration simulations, we selected initial inclination-dependent stability is the inner region near conditions that resulted in a better match for Neptune’s the ν18 secular resonance (e.g., Kuchner et al. 2002); final inclination; this also resulted in slightly smaller in- in this region, the instability timescale is much shorter clinations for other planets compared to the real solar than 700 Myr. system. The final eccentricity and inclination ranges As discussed later in Section3, our simulations sug- of the planets in these simulations (taken over the last gest that the strengths of the giant planets’ inclination 100 Myr of the simulations) are shown in the bottom secular modes during and after migration have a signifi- row of panels in Figure1. These planetary initial con- cant influence on the inclination distribution of the post- ditions were then re-run with 1.5 − 2.5 × 105 massless migration Kuiper belt. When selecting the planets’ ini- test particles with initial conditions as described above. tial conditions for the set of simulations described above, The second set of simulations are labeled ‘B’, while the we had focused on matching the final eccentricities of the first set of simulations are labeled ‘A’. planets rather than on matching the inclinations. At the 2.3. Simulation Results 6

At the end of each migration simulation, we identified in capture efficiencies 2-5 times lower than for the longer the test particles representative of the Plutino popula- migration timescales. tion by comparing test particles’ time averaged semima- We note that the capture efficiencies into the Plutino jor axes for the last 50 Myr of the simulation to the and hot classical regions depend on many factors, in- expected semimajor axis for the resonance. We also cluding some not explicitly investigated here. Resonant identified the population of test particles in the "hot capture efficiencies depend not only on Neptune’s mi- classical" region, defined as those with semi-major axes gration speed but also the eccentricities of the test par- 40 < a < 48 au and perihelion distances above 35 au ticles (e.g., Hahn & Malhotra 2005) and the smooth- and not obviously in resonance with Neptune (consistent ness of migration (e.g., Nesvorný & Vokrouhlický 2016; with Nesvorný 2015); we discarded test particles with Kaib & Sheppard 2016); the smoothness of migration average semimajor axes indicative of libration within the can also affect the capture efficiency into the hot classi- 8:5, 5:3, 7:4, and 2:1 mean motion resonances from this cal population. For the hot classical population, Daw- hot classical population. The capture efficiencies and son & Murray-Clay(2012) found that Neptune’s apsidal median inclinations for the resulting Plutino and hot precession rate also affects the probability of capturing classical populations from each simulation are given in scattered test particles onto stable orbits. All of these Table1. factors mean that the capture efficiency can vary widely even for a single planetary migration timescale, as seen 2.3.1. Capture efficiencies for the hot classical and Plutino in Table1. The relative capture efficiency between the populations Plutinos and hot classical population also vary widely; our two different τ = 30 Myr simulations produce ra- We find rather small capture efficiencies for the hot a tios of Plutinos to hot classical KBOs of 6 and 1. Thus classical region, similar to those found by Nesvorný outcomes from single sets of initial conditions, such as (2015). For example, with a τ = 30 Myr migra- a capture efficiency, population ratios, and orbital param- tion timescale and a similar initial Neptune orbit, eters, can be difficult to generalize. Nesvorný(2015) found a hot classical capture efficiency Our simulations do confirm that a scattering origin of ∼ 2 × 10−4 compared to our efficiencies of ∼ 6× 10−4 for the resonant populations is much less efficient than and ∼ 2.5 × 10−4 for our two τ = 30 Myr simula- a sweep up from a dynamically cold source considered in tions. Note, however, that the Nesvorný(2015) capture classical resonance sweeping scenarios (e.g., Malhotra efficiencies reflect 4 Gyr of evolution rather than our 1995; Hahn & Malhotra 2005). We list our capture effi- 700 Myr. The results of our two sets of simulations (‘A’ ciencies in Table1 as both a check for consistency with and ‘B’ in Tables1) show that the capture efficiency similar migration simulations in the literature and to in the hot classical region varies significantly for the emphasize the large number of test particles needed in same migration timescale with different initial condi- simulations to capture even moderate numbers in the tions. For fixed migration timescales, the efficiencies hot classical and Plutino populations. vary by a factor of 1.5-5, while across all simulations (and migration timescales) the efficiency only varies by a 2.3.2. Inclination distributions in the hot classical and factor of 8. This shows that capture efficiencies depend Plutino populations on more than just the migration timescale and that the Figure2 shows the inclination distributions of the final timescale dependence is weak, or at least not so strong Plutino populations for our four migration timescales; as to overwhelm these other dependencies. the distributions from the ‘A’ simulations (in which Nep- The Plutino capture efficiencies in our simulation are tune’s inclination is slightly larger than in the real solar also broadly similar to those of Nesvorný(2015), al- system) are shown in the left panels, and the distribu- though ours are slightly lower when accounting for the tions from the ‘B’ simulations (in which the planets have shorter total integration time. For τa = 10 Myr and slightly smaller inclinations) are shown in the right pan- τa = 30 Myr, Nesvorný(2015) described capture ef- els. For each simulation, we have measured the inclina- ficiencies in the range ∼ 0.5 − 1 × 10−3 compared to tions of the test particles relative to the plane defined by efficiencies in the range ∼ 0.5 − 1.5 × 10−3 in our similar the total angular momentum of the planets. As noted simulations. While Nesvorný(2015) found that slower above, the resonance capture efficiency is quite low, re- migration timescales decreased the efficiency of capture sulting in a somewhat noisy inclination distribution for into the 3:2 resonance (for simulations with Neptune each individual simulation despite the large initial num- starting near ∼ 24 au, like in our simulations), we find ber of test particles. In the τa = 5 Myr simulations, a weak trend in the opposite direction; for both our ‘A’ at the end (t = 700 Myr) we had only ∼ 70 Plutinos and ‘B’ simulations, the τa = 5 Myr simulations resulted in both the ‘A’ and ‘B’ simulations; each of the longer migration timescales and Kuiper belt inclinations 7

Table 1. Summary of simulation results

τa simulation median iHC fHC median i3:2 f3:2 50 Myr A 26◦ ∼ 5 × 10−4 13◦ ∼ 1 × 10−3 50 Myr B 17◦ ∼ 1 × 10−4 10◦ ∼ 1.5 × 10−3 30 Myr A 27◦ ∼ 6 × 10−4 13◦ ∼ 6 × 10−4 30 Myr B 18◦ ∼ 2.5 × 10−4 9◦ ∼ 1.5 × 10−3 10 Myr A 20◦ ∼ 8 × 10−4 18◦ ∼ 5 × 10−4 10 Myr B 13◦ ∼ 5.5 × 10−4 8◦ ∼ 1 × 10−3 5 Myr A 21◦ ∼ 7 × 10−4 10◦ ∼ 2 × 10−4 5 Myr B 18◦ ∼ 3 × 10−4 7◦ ∼ 2.5 × 10−4 Note— The median inclination and fraction of initial test particles in the 3:2 and hot classical populations (f3:2 and fHC ) at the end of the 700 Myr simulations; all inclinations are measured relative to the plane defined by the planets’ total angular momentum. As described in Section2, the simulations labeled ‘A’ resulted in Neptune having an inclination slightly larger than in the current solar system; the planets inclinations in the simulations labeled ‘B’ are slightly lower than in the current solar system (see also Figure1).

migration timescale simulations ended with ∼ 140 − 500 curves in Figure2) for comparison to our simulation re- Plutinos. While the inclination distributions are slightly sults; these are from Volk et al.(2016) who modeled the noisy, our results are sufficient to observe a counter ex- observed set of Plutinos discovered in the first quarter of ample to Nesvorný(2015)’s result that the peak in the the Outer Solar System Origins Survey (Bannister et al. Plutino inclination distribution increased markedly with 2016) and determined that their intrinsic inclination dis- increasing migration timescale, shifting from ∼ 10◦ for tribution could be acceptably modeled by a function, ◦ 2 2 ◦ τa = 10 Myr to ∼ 20 for τa = 30 Myr (see his Figure 9). f(i) ∝ sin(i) exp (i /2σi ), with parameter σi ≈ 9 − 13 We find no such trend in our simulations. Our ‘A’ and (consistent with previous estimates, e.g., Gladman et al. ‘B’ simulations with migration timescale τa = 10 Myr 2012; Gulbis et al. 2010). We see that the results of our yield both the highest median Plutino inclination and ‘A’ simulations are generally similar to the estimated a near-tie for lowest median inclination. Our ‘B’ simu- intrinsic inclination distributions while our ‘B’ simula- lations have a slight trend of increasing median Plutino tions result in Plutinos with typical inclinations slightly inclinations with increasing migration timescale, but it is smaller that the observations suggest. very weak with median inclinations only increasing from Figure3 shows the inclination distribution for the test ◦ ◦ 7 to 10 over an order of magnitude in τa; additionally, particles in the hot classical region at the end of the sim- the difference between the inclination distributions for ulations for each simulated migration timescale. Our τa = 10 Myr and τa = 30 Myr ‘B’ simulations is not hot classical test particles have inclination distributions statistically significant. Overall our simulations show that are slightly broader and extend to higher inclina- that there is as much variation in the particle inclina- tions than our Plutino distributions. We again note that tion distributions when comparing multiple runs with there is not a clear trend between peak inclination and the same migration timescale but slightly different ini- migration timescale. The inclination distribution for our tial conditions of the planets as when comparing runs τa = 30 Myr migration timescale simulation ‘A’ peaks in with different migration timescales. the range ∼ 20 − 30◦, which is consistent with Nesvorný In the bottom panels of Figure2, we plot smoothed (2015)’s result (his Figure 9); our τa = 30 Myr migration inclination distributions using a normalized kernel den- timescale simulation ‘B’ peaks at slightly lower inclina- sity estimate for each simulated set of Plutinos. (Note tion, but is of similar width. For comparison, we show that any physical inclination distribution should go to Petit et al.(2017)’s estimate for the intrinsic inclination zero at i = 0, which is not the case for these smoothed distribution of the hot population in the bottom panels distributions; these smoothed distributions are shown of Figure3. All of our simulations produce fairly broad primarily to more easily compare the simulations.) We inclination distributions of the hot classical population, also plot observationally derived estimates of the intrin- similar to the estimated intrinsic distribution. sic inclination distribution of the Plutinos (dotted black 8

Plutino inclination distributions (A simulations) Plutino inclination distributions (B simulations)

90 5 Myr 90 5 Myr 80 10Myr 80 10Myr 70 30 Myr 70 30 Myr 60 50 Myr 60 50 Myr 50 50 40 40 30 30 20 20

test particles per bin per testparticles 10 bin per testparticles 10 0 0 0 10 20 30 40 50 0 10 20 30 40 50 inclination (deg) inclination (deg) 0.08 0.08 5 Myr 5 Myr 0.07 0.07 10Myr 10Myr 0.06 30 Myr 0.06 30 Myr 0.05 50 Myr 0.05 50 Myr observational observational 0.04 constraints 0.04 constraints 0.03 0.03 0.02 0.02 0.01 0.01 0.00 0.00 smoothed normalized density normalized smoothed 0 10 20 30 40 50 density normalized smoothed 0 10 20 30 40 50 inclination (deg) inclination (deg)

Figure 2. Top panels: Histograms of the final inclinations (at t = 700 Myr) of test particles captured in Neptune’s 3:2 resonance for our four different migration timescales in simulation set ‘A’ (top left) and simulation set ‘B’ (top right). Bottom panels: normalized and smoothed kernel density estimates for the same inclination distributions; also plotted with black dashed lines are the 95% confidence limits on the Plutino population’s intrinsic inclination distribution, assuming it has the form −i2 f(i) ∝ sin(i) exp 2 (Volk et al. 2016). These smoothed distributions are shown to better illustrate the differences between the 2σi simulations. We compared our simulated inclination distributions ‘B’ simulations, but the other timescales are not dis- to each other using the Anderson-Darling test1. For the tinguishable from each other at 95% confidence. The Plutinos shown in Figure2, most of the distributions τa = 5 Myr ‘B’ and τa = 10 Myr ‘B’ hot classical in- are statistically distinguishable at 95% confidence ex- clination distributions are statistically different from all cept when comparing the τa = 5 Myr ‘A’ simulation to of the ‘A’ hot classical distributions; the τa = 30 Myr either the ‘A’ or ‘B’ τa = 30 Myr simulations and when ‘B’ and τa = 50 Myr ‘B’ inclination distributions are comparing the τa = 10 Myr ‘B’ and τa = 30 Myr ‘B’ not distinguishable from the ‘A’ simulations because few simulations. For the hot classical inclination distribu- test particles were captured into the hot classical region. tions shown in Figure3, the ‘A’ simulations are distinct Considering that most of our simulations produced sta- from each other with the exception of the τa = 30 Myr tistically distinct inclination distributions, particularly and τa = 50 Myr distributions. There are fewer cap- in the final Plutino populations, we should have eas- tured hot classical test particles in the ‘B’ simulations ily identified the strong trend seen in Nesvorný(2015)’s (due partly to lower capture efficiencies and partly due simulations if that trend were present in our simulations. to smaller numbers of initial test particles), so these dis- tributions are not as distinct. The τa = 10 Myr ‘B’ 3. DISCUSSION simulation is statistically distinguishable from the other Our simulation results are in some conflict with Nesvorný(2015)’s. To identify the cause(s) of the dis- 1 We use the version of this test described by NIST: https: crepancies, we first note some important differences //www.itl.nist.gov/div898/handbook/eda/section3/eda35e.htm, between the simulated migration scenarios. Nesvorný also described in Appendix 1 of Volk et al.(2016). (2015) forced only Neptune to migrate in the simu- lations (keeping the orbits of the other giant planets migration timescales and Kuiper belt inclinations 9

Hot classical inclination distributions (A simulations) Hot classical inclination distributions (B simulations)

70 30 5 Myr 5 Myr 60 10Myr 25 10Myr 50 30 Myr 30 Myr 50 Myr 20 50 Myr 40 15 30 10 20 5

test particles per bin per testparticles 10 bin per testparticles 0 0 0 10 20 30 40 50 60 0 10 20 30 40 50 60 inclination (deg) inclination (deg)

0.05 5 Myr 0.05 5 Myr 10Myr 10Myr 0.04 30 Myr 0.04 30 Myr 50 Myr 50 Myr 0.03 obs. 0.03 obs. estimate estimate 0.02 0.02

0.01 0.01

0.00 0.00 smoothed normalized density normalized smoothed 0 10 20 30 40 50 60 density normalized smoothed 0 10 20 30 40 50 60 inclination (deg) inclination (deg)

Figure 3. Top panels: Histograms of the final inclinations (at t = 700 Myr) of test particles in the hot classical region for our four different migration timescales in simulation set ‘A’ (top left) and simulation set ‘B’ (top right); note that the top panels have different y-axis ranges. Bottom panels: Normalized and smoothed kernel density estimates for the same inclination distributions; also plotted with black dashed lines are observationally derived best-fit functions (sin i times a Gaussian) for the dynamically hot population (Petit et al. 2017). fixed), while we have all four giant planets migrating. each planet perturb each other, causing quasi-periodic Another major difference is that Nesvorný(2015) em- variation of the planets’ eccentricity vectors and incli- ployed an eccentricity and inclination damping scheme nation vectors while the planets’ semimajor axes remain for Neptune (in addition to the torque to migrate its fixed. To linear order, the evolution of the planets’ or- semi-major axis), while we include only the torque re- bital planes and their orbital eccentricities are indepen- quired to migrate the planets in semimajor axis. These dent. Since we are investigating the inclination distri- two differences lead to some differences in the secular butions of the Plutino and hot classical populations, we architecture of the simulated planetary system, both focus here on the linear secular theory of the planets’ during and after migration, which are likely to be im- inclinations. Each planet’s orbital plane is described by portant for shaping the inclination distributions of the an inclination vector (sin i sin Ω, sin i cos Ω). The equa- hot classical and Plutino populations. tions of motion for the linear secular perturbations of In the current solar system, the evolution of the four the planets allow a solution in which the time evolu- giant planets’ orbital planes (i.e., their orbital inclina- tion of the planets’ inclination vectors can be described tions, i, and longitudes of ascending node, Ω) can be de- by a sum over the system’s inclination eigenvectors and scribed fairly accurately using linear secular theory (see, eigenfrequencies. The solution for our four giant planets e.g., Murray & Dermott 1999). Briefly, the premise of yields three non-zero eigen-modes called the secular in- linear secular theory is that the planets’ mutual grav- clination modes. In the notation of Murray & Dermott itational perturbations can be modeled as though the (1999), these are the f6, f7, and f8 modes; in the cur- mass of each planet were spread out in a ring along rent solar system they correspond to nodal regression its eccentric, inclined orbit; this amounts to assuming periods of ∼ 0.05, ∼ 0.45, and ∼ 1.9 Myr, respectively. that the perturbations between the planets can be av- The frequency of each of these modes (to lowest order) eraged over orbital timescales. The rings representing depends only on the masses and semimajor axes of the 10 planets. Because our simulations are designed to have tecture of our simulated planetary systems might be af- the planets end very near their current semimajor axes, fecting the test particle inclinations in our simulations. the mode frequencies at the end of each simulation are We can use the theory to calculate the modes and mode nearly identical and match the mode frequencies of the amplitudes of the planets in our simulations to see how current solar system quite well. For each giant planet, they vary between simulations and how they compare the relative contribution of these three modes to the to the current solar system. However, there are several time-evolution of its inclination vector (i.e., the mode caveats that we need to acknowledge before discussing amplitudes calculated from the system’s eigenvectors) this analysis. First is that a linear secular description is determined by initial conditions. We can calculate of the planets’ interactions does not include the effects the mode amplitudes in the current solar system using of the extra migration forces in the simulation. At each the observed planetary inclinations. The combination of time point being considered, we are effectively treating mode frequencies and mode amplitudes is what we refer the system as having "frozen" semi-major axes at each to as the secular architecture of the giant planets. point in time. In our slower migration cases, the plan- Linear secular theory can also be used to describe how ets’ orbits are changing in semimajor axis on timescales the inclinations of massless test particles are affected by longer than the outer planets’ secular timescales which the planets. A test particle’s inclination vector can be are . 2 Myr; for the shorter migration timescales, the described as a sum of two components: a free inclination evolution toward the end of migration is slow enough to and a forced inclination, determined by initial conditions be much slower than these secular timescales, but this and the secular architecture of the giant planets. The separation of migration timescale and secular timescale amplitude of the forced inclination depends on the pre- is not so large at the beginning of the simulations. We cession rate of the test particle’s free inclination and the also recognize that linear secular theory is not a good ap- planets’ secular modes and mode amplitudes. The free proximation of the planets’ evolution if they approach precession rate is determined by the averaged perturba- and/or cross mutual mean motion resonances as they tions from the giant planets and depends only on the migrate. In such cases, the assumption that we can av- planets’ masses and semimajor axes combined with the erage over orbital timescales breaks down. test particle’s semimajor axis. The amplitude of the test The evolution of test particles in the simulations are particle’s forced inclination is determined by a weighted not expected to be dominated by secular perturbations, sum over each of the planetary secular modes; the con- especially at the beginning of migration. The orbit aver- tribution from each mode is proportional to the mode aging assumption is not valid when test particles are on amplitudes (calculated form the perturbations amongst planet-crossing orbits. Because all of the test particles the planets) and inversely proportional to the difference must be scattered outward to end up in the Plutino or between the test particle’s free precession rate and the hot classical regions, they are clearly subject to pertur- mode frequencies. Very large forced inclinations occur bations that are not secular in nature. However, their where the free precession frequency matches one of the orbital inclinations between or after scattering events secular mode frequencies. The locations in test particle will still be influenced by secular forcing. In particular, semimajor axes where this occurs are referred to as sec- the semimajor axis of the 3:2 resonance is close to the ular resonances (see Knezevic et al. 1991 for a map of semimajor axis of the ν18 inclination secular resonance secular resonances in the solar system). where significant secular forcing can occur on timescales In the current solar system, the free precession peri- comparable to the mode period, so the inclinations of ods of test particles in the outer solar system range from the Plutino population could be particularly sensitive to ∼ 0.25 Myr at 20 au (just outside Uranus’s orbit) to variations in the secular architecture of the planets. For ∼ 5 Myr at 50 au (the outer edge of the classical Kuiper example, Figure4 shows the inclination and semimajor belt). Thus, only the f7 and f8 contribute significantly axis evolution for a test particle at the beginning of the to the forced inclination of a test particle in the Kuiper τa = 50 Myr ‘A’ simulation. In the first million years, belt region. The influence of the f8 mode is particularly the particle experiences a few degrees of inclination ex- apparent in the current Kuiper belt; the secular reso- citation as it scatters with Neptune. Then the particle nance associated with the f8 mode, referred to as the is scattered to a semimajor axis just beyond the 3:2 res- ν18 secular resonance, is located in the a = 40 − 42 au onance, which is near the estimated location of the ν18; range (see, e.g., Chiang & Choi 2008), helping to define this results in the test particle’s inclination increasing the inner edge of the classical Kuiper belt. by ∼ 10◦ over a ∼ 1 Myr period. Neptune’s continued We can use a linear secular analysis of our migration migration then results in the 3:2 resonance catching up simulations to help understand how the secular archi- with the test particle and capturing it into resonance, migration timescales and Kuiper belt inclinations 11 locking it into its higher inclination orbit. Thus, despite good description of Neptune’s inclination evolution; if the limitations, a linear secular analysis of the simula- Neptune’s simulated inclination vector had additional tions can yield useful insights into the resulting test par- frequencies with significant amplitudes besides the three ticle distributions and illustrate the potential for secular predicted by the simplified theory, the mode amplitudes forcing to be an important source of inclination excita- recalculated at each time-point in the simulation would tion. not match. Figure5 shows, in green, the same mode amplitudes for Neptune from two of our migration sim- ulations. The larger variations of these mode amplitudes 35 in the migration simulations compared to the simulation 3:2 30 of the non-migrating current solar system reflects the capture in the 3:2 fact that linear secular theory (as described above) does

a (au) 25 Neptune not account for mean motion resonances that can pro- 20 duce significant perturbations as the planets migrate. 20 Nevertheless, for each simulation, the dominant mode 15 in Neptune’s inclination evolution (the f8 mode) is rel- 10 atively stable, especially at the end of the simulations when migration has finished. 5 As Figure5 demonstrates, our simulations lead to a

inclination (deg) 0 0 1 2 3 4 5 6 final secular architecture of the giant planets that is not time (Myr) a perfect simile of the real solar system. In our ‘A’ simu- lations, the amplitude of f mode is typically larger than Figure 4. Example of secular inclination excitation for a 8 in the current solar system, even when the power asso- test particle in the τa = 50 Myr ‘A’ simulation. The top panel shows the semimajor axis of the test particle (black ciated with the other modes is fairly well matched (as dots) compared to the location of Neptune (gray line) and the in the left panel of Figure5). Because the f8 is the pri- 3:2 resonance (green line); the bottom panel shows the test mary mode associated with Neptune’s inclination, this particle’s inclination. The gray shaded region indicates the results in the simulated Neptune having an inclination time period where the test particle’s evolution is dominated slightly too large compared to the real solar system (see by scattering events with Neptune. The red shaded region Figure1). The power associated with the f mode in our indicates the time period where a marked secular increase 8 in the test particle’s inclination occurs. The vertical dashed ‘B’ simulations better matches that of the real solar sys- line indicates the time where the particle is captured in the tem (corresponding to a better match with Neptune’s 3:2 resonance. real inclination), although the other mode amplitudes are typically slightly smaller (as in the right panel of Figure5). We use linear secular theory to estimate the ampli- This very simple analysis of the f mode amplitudes tudes of the three secular modes in our simulations and 8 in our simulations hints at a plausible explanation for to compare them to the mode amplitudes in the real so- the trends in the inclination distributions in Figure2. lar system. We do this by taking snapshots of our sim- For each simulation, we calculate the average amplitude ulated systems at various points. These points in time of the f mode in Neptune’s orbit at the end of the define the semimajor axes of the planets (the masses of 8 simulation and ratio it to the same amplitude for the the planets are fixed to their observed values throughout current solar system. We plot these mode amplitude ra- the simulations) from which the mode frequencies can tios and the inclinations of the Plutino and hot classical be calculated. Then, just as the observed inclinations of populations in Figure6, observing the following points. the planets are used to calculate the mode amplitudes First, the ‘A’ simulations have mode amplitudes that are in the real solar system, we use the instantaneous incli- ∼ 2 − 4 times larger than the ‘B’ simulations. Second, nation vectors of the planets at these times in the sim- the ‘A’ simulation Plutino populations (left panel of Fig- ulation to determine the appropriate mode amplitudes. ure6) also have higher inclinations than the Plutinos in For comparison, we performed the same calculations for the ‘B’ simulations. The inclination widths, defined as a 700 Myr simulation of the current solar system’s giant the middle 50% of the inclination distribution, in the ‘A’ planets. The black lines in Figure5 show the calculated simulation Plutinos are ∼ 2 times larger than for the ‘B’ amplitudes of the f , f , and f in Neptune’s inclination 6 7 8 simulation Plutinos; the median inclinations in the ‘A’ evolution in this simulation of the current solar system. simulations are also up to ∼ 3 times larger than in the The fact that the calculated mode amplitudes are rel- ‘B’ simulations. atively flat shows that linear secular theory provides a 12

τa = 10 Myr A τa = 5 Myr B 10-1 10-1 migration simulation f8 solar system f8 solar system 10-2 10-2 migration simulation f7 f7 10-3 10-3

f f 10-4 6 10-4 6

10-5 10-5 1x106 1x107 1x108 1x106 1x107 1x108 scaled mode strength for Neptune time (years) scaled mode strength for Neptune time (years)

Figure 5. Estimated amplitudes for the three inclination secular modes in Neptune’s orbital inclination for the τa = 10 Myr ‘A’ simulation (green, left panel) and the τa = 5 Myr ‘B’ simulation (green, right panel) compared to the same mode strengths for the current solar system (black). These trends are much weaker in the hot classical pop- are likely to have spent some time at nearby semimajor ulation (right panel of Figure6). This could be partly axes and could thus be influenced. We visually inspected due to the smaller number statistics in the ‘B’ simula- the semimajor axis and inclination evolution of 50 test tions. It is also possible that the effects of the secu- particles from each simulation that ended up in our final lar modes are less pronounced because the hot classical Plutino populations to estimate the relative frequency test particles have higher inclinations than the Plutino of secular inclination excitation associated with the ν18 test particles; perhaps because these test particles have resonance. In all but the τa = 5 Myr ‘B’ simulation, ap- to be scattered out further than the Plutino test parti- proximately 50-70% of the test particles showed signifi- cles, their inclinations are more affected by these random cant changes in inclination while at constant semimajor scattering events than by secular effects. axes just beyond the 3:2 resonance; in the τa = 5 Myr For the Plutino population at least, the correlation be- ‘B’ simulation, which resulted in the fewest high inclina- tween the f8 mode amplitude and median inclinations tion Plutinos, this fraction was ∼ 30%. For comparison, (as well as inclination distribution width) in our simu- the percentage of test particles showing evidence of incli- lations is much clearer than any trend with migration nation excitation due to close encounters with Neptune timescale. We checked for similar trends with the f7 (i.e., discrete changes in inclination correlated with dis- mode amplitude, but none was apparent. crete changes in semimajor axis) was also ∼ 50 − 70%. A link between the f8 mode amplitude and inclination The changes in inclination at constant semimajor axes excitation in the Plutino population provides a plausible near the estimated location of the ν18 resonance were explanation for the discrepancy between our simulation smooth and in many cases roughly sinusoidal, which is results and those of Nesvorný(2015). If Neptune’s in- consistent with the change being driven by secular forc- clination is damped during planetary migration, as was ing; some test particles exhibited very rapid inclination done in Nesvorný(2015)’s simulations, this would also increases, consistent with being close to the center of the damp the amplitude of the f8 secular mode, reducing the secular resonance. We note that for test particles under- secularly forced inclinations of test particles in the outer going sinusoidal inclination changes, the net change in solar system. This could be particularly important for average inclination depends on the phase of the inclina- test particles passing through the ν18 secular resonance, tion cycle upon being captured into the 3:2 resonance, where maximum forced inclinations due to the f8 occur. and some test particles had a net decrease in final in- In the current solar system, the location of the ν18 reso- clination. It is clear that the f8 mode can contribute nance is just beyond the 3:2 resonance, and based on the significantly to inclination excitation, and the relative linear secular calculations this is also the case during all importance of scattering events and secular evolution but the very beginning of our migration simulations. So in the excitation of inclinations depends on the mode it is likely that many test particles that are scattered out strength. A strong or weak f8 mode during migration past the 3:2 and are then picked up in the resonance as could either enhance or reduce the inclination excitation Neptune migrates experience some inclination excitation these test particles experience before ending up in the due to this resonance with the f8 mode, as shown in Fig- final Plutino population. We also note that the mode ure4. Even particles that more directly stick to the 3:2 amplitudes could affect the so-called ‘Kozai’ resonance migration timescales and Kuiper belt inclinations 13

30 Myr 30 Myr 3 3 10 Myr 10 Myr 2.5 50 Myr 2.5 simulation set A 50 Myr simulation set A 2 5 Myr 2 5 Myr

mode amplitude 1.5 mode amplitude 1.5 8 8 simulation set B 30 Myr 30 Myr 1 1 5 Myr 50 Myr simulation set B 50 Myr 10 Myr 10 Myr 5 Myr

relative f 0.5 relative f 0.5 0 5 10 15 20 25 0 5 10 15 20 25 30 35 median plutino inclination (deg) median hot classical inclination (deg)

Figure 6. The ratio of the estimated amplitude of the f8 inclination mode in Neptune’s orbit at the end of the simulations to its amplitude in the current solar system vs. median inclination in the final Plutino (left panel) and hot classical populations (right panel) in the simulations. The horizontal bars represent the 25-75% inclination range in each population’s cumulative inclination distribution. The ‘A’ simulations are shown in purple and the ‘B’ simulations in black (labeled by τa). within the 3:2 mean motion resonance. This resonance the f8 mode amplitude and frequency is the most plau- is characterized by the libration of a Plutino’s argument sible explanation for the dramatically different trends in of perihelion and causes coupled opposite-phase varia- our simulated Plutino populations compared to those in tions in eccentricity and inclination (e.g. Milani et al. Nesvorný(2015). 1989). Kozai libration is likely to be affected by changes The giant planets’ secular architecture should also af- in the secular architecture of the planets because sta- fect the inclination distribution of the hot classical pop- tionary values of the argument of perihelion correspond ulation, however it is possible that they are more in- to a match between the precession rate of the longitude fluenced by encounters with Neptune than by secular of perihelion and the regression rate of the longitude inclination forcing. Nesvorný(2015) found that it took of ascending node. We leave an assessment of how the of order a few tens of Myr for the inclinations of the mode amplitudes affect the Kozai resonance and its cor- population of Neptune-crossing test particles with semi- responding inclination and eccentricity variations within major axes in the classical belt range (the source pop- the Plutino population for future work. ulation for the final implanted hot classical orbits) to The particular implementation of Neptune’s inclina- become significantly dispersed as a result of encounters tion damping in the migration simulations could have with Neptune (see his Figure 12); this led to the finding different effects on Neptune’s nodal rates in addition to that hot classicals implanted late during the migration influencing the mode amplitudes. (Nesvorný(2015) does process would have higher inclinations and that slower not describe the specific scheme used in their simula- migration would lead to higher inclinations in the final tions.) In addition to affecting the amplitude of the f8 hot classical population. In our simulations, we find a mode by damping Neptune’s inclination, the frequency nearly identical trend to that found by Nesvorný(2015) of the mode itself could change, changing the locations of in the inclination distribution of the hot classical source secular resonances that have a strong influence on parti- region with time, but we do not find a strong trend be- cle inclinations. Fixing the other giant planets’ semima- tween the final hot classical inclinations and migration jor axes and migrating only Neptune would also affect speed. The two longest migration timescale ‘A’ sim- the frequency and amplitude of the f8 mode (and also ulations do have the broadest hot classical inclination secular resonance locations) in Nesvorný(2015)’s sim- distributions, but the τa = 5 Myr ‘B’ simulation had ulations compared to ours. Nesvorný(2015) did per- the highest median inclination hot classical population form some integrations where all four giant planets mi- of the ‘B’ simulations. Additionally, in the τa = 5 Myr grated and did not find evidence that this affected the ‘A’ simulation, the final hot classical inclination distri- correlation between inclinations and migration timescale bution is broader than that of the source region after in those simulations. However these additional simula- ∼ 3 e-folding timescales, after which very little migra- tions are not discussed in detail; it is possible that the tion occurs. This implies that while longer migration initial and final conditions for these simulations differ timescales can allow for more inclination excitation of from our simulations, resulting in different secular ar- the source population (due to more Neptune encounters, chitectures. Thus we find that a significant difference in as suggested by Nesvorný 2015), there are additional 14 mechanisms for exciting inclinations even with shorter ets as well as the presence or absence of eccentricity migration timescales. It seems plausible that secular damping will likely affect simulation outcomes and fi- excitation of inclinations can contribute to creating a nal secular architectures. Previous studies have noted broad hot classical inclination distribution. that the final giant planet secular architecture in plane- In this work we have only investigated the inclina- tary migration simulations is very sensitive to the plan- tion distributions that result from different migration ets’ initial conditions and to whether planets encounter scenarios in order to explain why strong trends between mean motion resonances or undergo mutual scattering inclinations and migration speed seen in one set of simu- events; this makes it difficult to perfectly reproduce the lations (Nesvorný 2015) can be absent in others. As dis- solar system’s observed secular architecture in migra- cussed by Nesvorný(2015), the observational estimates tion simulations (e.g., Morbidelli et al. 2009; Batygin of inclination distributions for different Kuiper belt pop- & Brown 2010; Nesvorný & Morbidelli 2012). Many ulations are often easier to obtain than other orbital pa- previous investigations have focused on the eccentric- rameter distributions or estimates of their total popula- ity secular modes of Jupiter and Saturn because these tions; thus, understanding how inclination distributions modes have a strong influence on the stability of the can be used to constrain the history of the outer solar terrestrial planets during giant planet migration (e.g., system is of particular importance. However, many of Brasser et al. 2009; Agnor & Lin 2012). For the Kuiper the same points made above about the secular architec- belt, both Neptune’s eccentricity and inclination secu- ture of the planetary system also apply to the eccentric- lar modes are important in sculpting the final orbital ity distribution of Kuiper belt sub-populations. Works distributions. Given how difficult it is to reproduce the such as Batygin et al.(2011) and Dawson & Murray- current secular structure of the solar system at the end Clay(2012) have shown that the low eccentricities of the of migration simulations, the origin of Neptune’s secu- cold classical Kuiper belt objects can be used to place lar modes should be investigated in more detail in future constraints on Neptune’s eccentricity and apsidal pre- work. cession rate because these affect the amplitude of the Finally, we note that the simplifying assumption of a secular variations in eccentricity experienced by small massless test particle disk also affects the secular evo- bodies in the Kuiper belt. We have not investigated the lution of objects in these simulations. The eccentrici- evolution of the cold classical Kuiper belt in our simula- ties and inclinations of the massive pre-migration Kuiper tions because there are a number of structures in their belt could differ depending on how their self-gravity af- orbital distribution (such as the over-density of objects fects the evolution of both the planets and the KBOs in the cold classical “kernel” at 44.5 au; Petit et al. 2011) (see, .e.g., Hahn 2003; Reyes-Ruiz et al. 2015). While re- that are not reproduced by the kind of simplified migra- cent work by Fan & Batygin(2017) shows that including tion scenarios investigated here and in Nesvorný(2015). self-gravity between the massive planetesimals in Nice- However, we note that our simulations are consistent model like simulations does not appear to change the with preserving the low eccentricities of the classical belt simulation outcomes for the planets or the overall incli- because we chose initial conditions such that Neptune’s nation distribution for the resulting Kuiper belt parti- eccentricity remained small throughout migration. The cles, the details of the secular architecture and the lo- low inclinations of the cold classical belt are also likely cations of secular resonances could be affected; thus the to be preserved in our migration simulations, including inclinations of specific dynamical populations, such as those with enhanced f8 mode strengths, because the lo- the Plutino and hot classical populations, could be af- cation of the Kuiper belt’s inclination secular resonance fected by the massive pre-migration planetesimal disk. remains close to the 3:2 resonance throughout the sim- ulation; we repeated our taua = 50 Myr ‘A’ simulation 4. SUMMARY AND CONCLUSIONS with a disk of test particles representing the cold classi- We have performed simplified simulations of giant cal belt and found that they did not experience signifi- planet migration to investigate the relationship between cant inclination or eccentricity excitation. migration speed and inclination excitation in the Plutino For the Plutino and hot classical populations, the loca- and hot classical Kuiper belt populations. As with all tion and strength of the ν eccentricity secular resonance 8 such investigations, these simulations do not represent associated with Neptune’s dominant eccentricity secular the full, detailed dynamical history of the outer so- mode will likely affect eccentricities similarly to how the lar system, but instead allow us to better understand f mode and its ν secular resonance appears to af- 8 18 the relationship between migration speed and the in- fect inclinations. Just as with inclination secular modes clination distributions of the Plutino and hot classical and amplitudes, the initial relative positions of the plan- Kuiper belt populations. We do not reproduce Nesvorný migration timescales and Kuiper belt inclinations 15

(2015)’s finding that slower migration speeds lead to inclination excitation and migration speed. Slow plane- more widely dispersed inclinations in these populations, tary migration is not necessarily required to explain the instead finding no clear relationship between planet mi- large inclinations. gration speed and inclination excitation for e-folding mi- We therefore conclude that constraints on the speed gration timescales of τa = 5, 10, 30, and 50 Myr. All of of planet migration must be sought in features other these migration timescales can yield inclination distri- than the Kuiper belt inclination distribution. Some ex- butions of these populations that are broadly consistent amples of potentially useful features have already been with current observations. For the Plutinos, we find that discussed in the literature. These include Kaib & Shep- the degree of inclination excitation in our simulated pop- pard(2016)’s suggestion that the orbital distribution of ulations is correlated with the amplitude of the f8 incli- high-perihelion objects dropped from Neptune’s 3:1 res- nation secular mode of the giant planets; this mode am- onance could be strongly dependent on Neptune’s mi- plitude is sensitive to the simulated planets’ initial con- gration speed. The population ratios of objects captured ditions. Our simulated hot classical population shows in the different libration islands of Neptune’s 2:1 mean only a very weak correlation between inclination excita- motion resonance has also been found to potentially de- tion and the mode amplitude, perhaps indicating that pend on migration speed (Murray-Clay & Chiang 2005). scattering events are more important than secular effects Murray-Clay & Schlichting(2011) suggest that the frac- for this population. Our simulations are broadly similar tion of binaries in different subpopulations of the Kuiper to those of Nesvorný(2015), but differences in the nu- belt might relate to migration speed. While the speed merical implementation of planetary migration as well of migration might not be imprinted in the inclination as small differences in planetary initial conditions have distributions, other aspects of migration might be. This a significant impact on the secular architecture of the work suggests that even small changes in the secular ar- simulated systems. This plausibly leads to the different chitecture of the planets can lead to significant changes results regarding whether planet migration speed signif- in the inclination distributions of the Plutino popula- icantly controls inclination excitation in these Kuiper tion. Future studies to better understand how the plan- belt populations. The choice to migrate only Neptune ets’ secular modes change during migration and how this or all four giant planets, or to damp or not damp plan- affects Kuiper belt populations could lead to new con- etary inclinations, affects the amplitudes of the inclina- straints on the migration history of the outer solar sys- tion secular modes as well as the locations of secular tem. resonances and their corresponding large forced incli- nations in the Kuiper belt; the numerically simulated We thank the referee for helpful comments that im- Kuiper belt orbital distributions can thus be very sensi- proved the manuscript. We acknowledge funding from tive to the simplifications in numerical implementation NASA (grants NNX14AG93G and 80NSSC19K0785) of planet migration. Our simulations indicate that plan- and NSF (grants AST-1312498 and AST-1824869). An etary migration with e-folding timescales of 5 Myr, 10 allocation of computer time from the UA Research Com- myr, 30 myr and 50 myr, can all yield inclination disper- puting High Performance Computing (HPC) at the Uni- sions similar to the observed Plutino and hot classical versity of Arizona is gratefully acknowledged. populations, with no correlation between the degree of Software: rebound

REFERENCES

Agnor, C. B., & Lin, D. N. C. 2012, ApJ, 745, 143 Chiang, E. I., & Jordan, A. B. 2002, AJ, 124, 3430 Bannister, M. T., Kavelaars, J. J., Petit, J.-M., et al. 2016, Dawson, R. I., & Murray-Clay, R. 2012, ApJ, 750, 43 The Astronomical Journal, 152, 1 Duncan, M. J., & Levison, H. F. 1997, Science, 276, 1670 Batygin, K., & Brown, M. E. 2010, ApJ, 716, 1323 Everhart, E. 1985, in Dynamics of Comets: Their Origin Batygin, K., Brown, M. E., & Fraser, W. C. 2011, ApJ, and Evolution, Proceedings of IAU Colloq. 83, held in 738, 13 Rome, Italy, June 11-15, 1984. Edited by Andrea Carusi Brasser, R., Morbidelli, A., Gomes, R., Tsiganis, K., & and Giovanni B. Valsecchi. Dordrecht: Reidel, Levison, H. F. 2009, A&A, 507, 1053 Chambers, J. E. 1999, Mon. Not. of the Royal Astron. Soc., Astrophysics and Space Science Library. Volume 115, 304, 793 1985, p.185, ed. A. Carusi & G. B. Valsecchi, 185 Chiang, E., & Choi, H. 2008, AJ, 136, 350 Fan, S., & Batygin, K. 2017, ApJ, 851, L37 16

Gladman, B., Lawler, S. M., Petit, J.-M., et al. 2012, AJ, Murray-Clay, R. A., & Schlichting, H. E. 2011, ApJ, 730, 144, 23 132 Gomes, R., Nesvorný, D., Morbidelli, A., Deienno, R., & Nesvorný, D. 2011, ApJL, 742, L22 Nogueira, E. 2018, Icarus, 306, 319 —. 2015, AJ, 150, 73 Gomes, R. S. 2000, AJ, 120, 2695 Nesvorný, D., & Morbidelli, A. 2012, AJ, 144, 117 —. 2003, Icarus, 161, 404 Nesvorný, D., & Roig, F. 2000, Icarus, 148, 282 Gulbis, A. A. S., Elliot, J. L., Adams, E. R., et al. 2010, Nesvorný, D., & Vokrouhlický, D. 2016, ApJ, 825, 94 AJ, 140, 350 Petit, J.-M., Kavelaars, J. J., Gladman, B. J., et al. 2011, Hahn, J. M. 2003, ApJ, 595, 531 AJ, 142, 131 Hahn, J. M., & Malhotra, R. 2005, AJ, 130, 2392 —. 2017, AJ, 153, 236 Kaib, N. A., & Sheppard, S. S. 2016, AJ, 152, 133 Rein, H. 2012, MNRAS, 427, L21 Knezevic, Z., Milani, A., Farinella, P., Froeschle, C., & Froeschle, C. 1991, Icarus, 93, 316 Rein, H., & Spiegel, D. S. 2015, MNRAS, 446, 1424 Kuchner, M. J., Brown, M. E., & Holman, M. 2002, AJ, Rein, H., & Tamayo, D. 2015, MNRAS, 452, 376 124, 1221 —. 2017, MNRAS, 467, 2377 Levison, H. F., Morbidelli, A., Vanlaerhoven, C., Gomes, Reyes-Ruiz, M., Aceves, H., & Chavez, C. E. 2015, The R., & Tsiganis, K. 2008, Icarus, 196, 258 Astrophysical Journal, 804, 91 Malhotra, R. 1993, Nature, 365, 819 Tiscareno, M. S., & Malhotra, R. 2009, AJ, 138, 827 —. 1995, AJ, 110, 420 Tsiganis, K., Gomes, R., Morbidelli, A., & Levison, H. F. Milani, A., Nobili, A. M., & Carpino, M. 1989, Icarus, 82, 2005, Nature, 435, 459 200 Volk, K., & Malhotra, R. 2011, ApJ, 736, 11 Morbidelli, A., Brasser, R., Tsiganis, K., Gomes, R., & Volk, K., Murray-Clay, R., Gladman, B., et al. 2016, AJ, Levison, H. F. 2009, A&A, 507, 1041 152, 23 Murray, C. D., & Dermott, S. F. 1999, Solar system Wisdom, J., & Holman, M. 1991, AJ, 102, 1528 dynamics, 1st edn. (New York, New York: Cambridge Wolff, S., Dawson, R. I., & Murray-Clay, R. A. 2012, ApJ, University Press) 746, 171 Murray-Clay, R. A., & Chiang, E. I. 2005, ApJ, 619, 623