Sie sind auf Seite 1von 9

MNRAS (2017) doi:10.

1093/mnras/stx1418
Advance Access publication 2017 June 26

Limitations of backward integration method for asteroid family age


estimation

Viktor Radovic
Department of Astronomy, Faculty of Mathematics, University of Belgrade, Studentski trg 16, 11000 Belgrade, Serbia

Accepted 2017 June 6. Received 2017 May 27; in original form 2017 April 9

ABSTRACT
Determining the age of an asteroid family is important as it gives us a better understanding
of the dynamics, formation and collisional evolution of a family. So far, a few methods for
determining the age of a family have been developed. The most accurate one is probably the
backward integration method (BIM) that works very well for young families. In this paper, we
try to study its characteristics and limitations in more detail using a fictional asteroid family.
The analysis is performed with two numerical packages: ORBFIT and MERCURY. We studied the
clustering of the secular angles  and  and obtained linear relationship between the depth of
the clustering and the age of the family. Our results suggest that the BIM could be successfully
applied only to families not older than 18 Myr.
Key words: minor planets, asteroids: general.

the proper semimajor axis (ap ). At the beginning of the simulation,


1 I N T RO D U C T I O N
the random walkers are distributed in the region that was occupied
Asteroid families were first discovered by Hirayama in 1918, and by the family members immediately after the formation. The age of
have been intensively studied since then due to their importance the family is obtained when a certain per cent of random walkers
to understand the Solar system evolution, asteroid composition, leave an ellipse determined by the present size of the family. MCC
dynamics and collision modelling. They are formed when a collision was used for obtaining the ages of various families (e.g. Tsiganis,
breaks a parent body apart into smaller fragments (Zappala et al. Knezevic & Varvoglis 2007; Novakovic 2010; Novakovic, Tsiganis
1984). Asteroid families are identified as a clusters of asteroids in & Knezevic 2010). MCC could be applied if there is a presence of
the space of proper elements (Milani & Knezevic 1990; Knezevic & chaos in the region occupied by the family.
Milani 2000): proper semimajor axis ap , proper eccentricity ep and For the old families (older than 50 Myr), methods that are based
sine of proper inclination sin (Ip ). Proper elements are used for on Yarkovsky/YORP effects are the most widely used. They can be
family identification because they are quasi-integrals of motion, divided into methods that use a Monte Carlo approach and methods
and hence more constant over time than instantaneous (osculating) that are using a V shape.
elements. For family identification, hierarchical clustering method A method that uses Monte Carlo approach to obtain estimation
(HCM; Zappala et al. 1990) is usually used. The HCM identifies of an asteroid family age was pioneered by Vokrouhlicky et al.
an asteroid as a part of a family if its distance from the closest (2006a,b) who found the ages for several asteroid families. Also,
neighbour is smaller than an adopted cut-off distance. Until now, besides age, this approach allows us to obtain ejection velocities of
several family classifications have been published (Zappala et al. the fragments at the instant of the family formation. To determine the
1990; Nesvorny et al. 2005; Novakovic, Cellino & Knezevic 2011; age using this method, fictitious distributions of asteroids are created
Carruba et al. 2013; Milani et al. 2014; Nesvorny, Broz & Carruba using different values of ejection velocity and evolved under the
2015; Milani et al. 2017). influence of Yarkovsky/YORP effect. For each fictitious asteroids,
One of the main problems related to the research of asteroid fam- the C-parameter is calculated using the following relationship:
ilies is estimating their ages, and several methods that addressed
0.2H = log10 (a/C), (1)
this problem have been developed. The most frequently used meth-
ods are the backward integration method (BIM), chaotic chronology where H is the absolute magnitude and a is the semimajor axis. The
and methods based on Yarkovsky/Yarkovsky-OKeefe-Radzievskii- distribution of the C-parameter of a real family and of a simulated
Paddack (YORP) effects (V-shape criterion, V-shape fitting and particles is compared using a 2 -like variable C (Vokrouhlicky
Monte Carlo methods). et al. 2006a,b), whose minimum value is associated with the best-
The method of chaotic chronology (MCC) is based on simulating fitting solution. That solution gives us the age of the family. This
the evolution of the diffusion coefficients, which are functions of method was modified to account for the stochastic YORP effect
and also for past changes of solar luminosity. Using the modified
method, the ages of one of the oldest families in the main belt were
 E-mail: rviktor@matf.bg.ac.rs calculated (see Bottke et al. 2015; Carruba et al. 2015, 2016).

C 2017 The Author

Published by Oxford University Press on behalf of the Royal Astronomical Society


2 V. Radovic
The V shape is apparent when absolute magnitude (or inverse Nesvorny et al. (2002) first determined the age of the Karin fam-
diameter) is plotted against proper semimajor axis. Two processes ily using BIM and found it to be 5.8 0.2 Myr old. Later, Nesvorny
are the reason behind the formation of a V shape. The first process & Bottke (2004) improved this estimation by accounting for
is due to different ejection velocities between smaller and bigger Yarkovsky drift and obtained the age of 5.75 0.05 Myr. Carruba
fragments, which leads to their different locations from the family et al. (2016) revised the Karin family membership using a new
centre (smaller ones tend to be ejected at higher speeds and therefore orbital catalogue and obtained a more accurate solution (5.746
are located further from family centre, while the larger ones are 0.01). Another interesting example is the Veritas family whose age
ejected at slower speeds and are located closer to the centre; see is also determined using BIM, and it is estimated to 8.3 0.5 Myr
e.g. Cellino et al. 1999; Carruba & Nesvorny 2016). The second, (see e.g. Nesvorny et al. 2003; Tsiganis et al. 2007; Carruba et al.
more dominant process, is the non-gravitational Yarkovsky effect 2017). Also, ages for the very young asteroid families (with Tage <
(Farinella & Vokrouhlicky 1999; Vokrouhlicky, Farinella & Bottke 1 Myr) are determined using BIM: (14627) Emilkowalski, (16598)
2000b; Vokrouhlicky, Milani & Chesley 2000a; Vokrouhlicky et al. 1992 YC2, (21509) Lucascavin (Nesvorny & Vokrouhlicky 2006),
2006a,b). The Yarkovsky is a thermal effect that could lead to (2384) Schulhof (Vokrouhlicky & Nesvorny 2011), (1270) Datura
changes in the semimajor axis of an asteroid. The rate of change (Nesvorny, Vokrouhlicky & Bottke 2006; Vokrouhlicky et al. 2009,
of the semimajor axis in time caused by the Yarkovsky effect is 2017) and (18777) Hobson (Rosaev & Plavalova 2017). The ages
proportional to 1/D, where D is the diameter of an asteroid. of some families determined using BIM are shown in Table 1.
Due to the Yarkovsky effect, the spread of the members of an It is interesting to find how the minimum of clustering of the
asteroid family in ap increases over time. By calculating this spread secular angles depends on the age of the family. Obviously, tight
from the centre of the family ac (usually coincide with the largest clustering occurs for the very young families, while for the older
fragment in the family) to the border of the V shape, the age of families dispersion is much larger (see Table 1). This indicates that
the family could be obtained (see e.g. Nesvorny et al. 2003; Bottke, BIM could only be used for the young families for which it is
Vokrouhlicky & Nesvorny 2007). Spoto, Milani & Knezevic (2015) still possible to see the clustering. In this respect, some authors
used modified method where they fitted both sides of a V shape suggested that BIM can be applied for the families not older than
in the proper semimajor axis and inverse diameter plane in order 20 Myr (Nesvorny et al. 2003; Novakovic 2010). Up until now, there
to determine their slopes from which the age is calculated. They have not been any published papers dealing with the limitations and
computed the ages for 37 asteroid families. the accuracy of BIM. This is the main motivation for this paper,
Methods based on the V shape are the most frequently used, in which we will present the analysis of BIM using two different
but they have a large source of inaccuracy because of the numerical integrators on a simulated family. The relationship be-
used Yarkovsky calibration. Uncertainties in inferred ages due to tween depth of the clustering of secular angles and integration time
Yarkovsky calibrations are estimated to be between 20 per cent and is analysed. Also, the comparison between the results obtained with
30 per cent (Spoto et al. 2015). two integrators is studied and discussed.
A possible problem with applying this method to small families
is the presence of interlopers among them. Their presence could
1.1 Backward integration method
influence the accuracy of obtained conjunctions of secular angles,
Probably, the BIM is the most accurate one, and unlike the methods and therefore it is important to reduce the number of interlopers (see
based on Yarkovsky/YORP effects could be applied to younger for e.g. Novakovic et al. 2011; Milani et al. 2014; Radovic et al.
families. 2017).
Immediately after the family-forming event, the remaining de- The paper is organized as follows: In Section 2, we briefly present
bris accumulate due to mutual gravitational interaction and from it, the theory behind numerical integrators as well as the basic prop-
larger objects (rubble-piles) are formed. When those mutual gravi- erties of the considered packages for numerical integrations. The
tational influences cease to be important, objects start moving on a methodology and the characteristics of a test family that is used in
very close heliocentric orbits, which means that they have nearly the simulations are presented in Section 3. In Section 4, the results and
same values for orbital angles (ascending node , longitude of per- the analysis of obtained data with different integrators are shown.
ihelion  and mean longitude ). These angles become dispersed Discussion of the results and conclusions are given in Section 5.
on an orbit in short period of time by different influences (mainly
planetary perturbation). Besides planetary perturbation, other ef-
fects have a significant influence on dispersion of the secular an-
gles. Yarkovsky effect and close encounters with massive asteroids 2 N U M E R I C A L I N T E G R AT I O N
could notably change a possible clustering of the secular angles
(Carruba, Nesvorny & Vokrouhlicky 2016; Carruba, Vokrouhlicky The N-body problem is a problem in which the motions of a group
& Nesvorny 2017). of celestial objects interacting with each other due to mutual grav-
A potential clustering of is not expected to be found if a family itational attraction are determined. Solving this problem has been
is older than a few tens of thousand years because fragments will be- motivated by the desire to understand the motions of the objects
come fully dispersed on an orbit relatively fast (T = 30030 000 yr). in the Solar system. The N-body problem plays an important part
Clustering of  and  is not expected to be seen for families older in understanding and analysing the dynamics of globular clusters,
than 1 Myr. galaxies and their groupings.
The idea behind the BIM is to numerically integrate the orbits Newtons universal law of gravity describes the force acting be-
of a dynamically stable family members into the past, in order to tween two points, and it could be generalized to a system of any
obtain the clustering of secular angles ( and  ). This clustering N-bodies. The equations of motion of the planets may be written
of the secular angles should occur around the moment of family as
formation. The instant of this clustering should correspond to the m0 + m1
age of the family. q = k 2 q + P, (2)
3
Limitations of backward integration method 3
Table 1. The ages of the families that were determined using BIM. Only families older than 1 Myr are shown, except for the Hobson
family that is used for demonstration purposes. Tage is the determined age of the family in Myr, and m and  m are the depths of
the clustering of secular angles  and  , respectively. Values with were not given in the considered papers.

Family Tage (Myr) m ( )  m ( ) References

(18777) Hobson 0.365 67 1 1 Rosaev & Plavalova (2017)


P/2012 F5 (Gibbs) 1.5 0.1 7 7 Novakovic et al. (2014)
(5438) Lorre 1.9 0.3 14 24 Novakovic et al. (2012b)
(832) Karin 5.75 0.05 35 35 Nesvorny et al. (2002), Nesvorny & Bottke (2004)
Carruba et al. (2016)
(778) Theobalda 6.2 31 58 Novakovic (2010)
P/2006 VW139 7.5 0.3 32 55 Novakovic, Hsieh & Cellino (2012a)
(490) Veritas 8.3 0.5 31 Nesvorny et al. (2003), Tsiganis et al. (2007)
(656) Beagle 10 Nesvorny et al. (2008)

where q is the relative distance in a heliocentric coordinate system, predictor which performs most of the propagation. The predictor
k denotes the Gaussian gravitational constant, m0 is the mass of has a constant step-size h and uses only one evaluation of the right-
the central body, m1 is the mass of the body, is a scalar distance hand side of the equations of motion for each step. The step h is
between bodies and P is the perturbing function. Analytical and chosen so that the truncation error is minimized (Milani & Nobili
numerical methods are a possible ways of solving the equations of 1988). Output from ORBIT9 may include information about close
motion. As it is not possible to solve the N-body problem using ana- encounters, and besides gravitational influences, it is possible to
lytical approach (except for some special cases), numerical method include the Yarkovsky drift.
are usually used. The ORBFIT has additional programs for determination of the or-
With a numerical integrator, it is possible to follow the system bits of Solar system bodies and could compute their positions on
development by calculating its evolution step by step. Due to the the celestial sphere. Also, ORBFIT has tools for the computation of
development in computer science, especially in CPU processing Lyapunov exponents and proper elements (analytic and synthetic).
power, numerical integrations have become very popular and are
widely used for solving N-body problems. Large-scale numerical
experiments are possible today, and various numerical algorithms 2.2 MERCURY
have been developed (see e.g. Eggl & Dvorak 2010).
The MERCURY package (Chambers & Migliorini 1997) includes
Numerical algorithms are often divided in two groups: explicit
different algorithms, and it is written using the FORTRAN 77 pro-
and implicit methods. The explicit methods calculate the state of
gramming language. The MERCURY package is publicly available
a system at a later time from the state of the system at the current
at https://github.com/4xxi/mercury. MERCURY has the follow-
time, while implicit methods find a solution by solving an equation
ing N-body algorithms: second-order mixed-variable symplec-
involving both the current state of the system and the later one.
tic integrator, general BulirshStoer integrator, conservative
Hence, if s(n) is the current state of the system at time t = nh, the
BulirshStoer, Everharts RA15 (RADAU) algorithm and Hybrid
next state at later time s(n + 1) is calculated as
symplectic/BulirshStoer integrator.
(i) for the explicit as s(n + 1) = s(n) + hF(s(n)), For the purpose of our analysis, we chose the explicit Hybrid
(ii) for the implicit as s(n + 1) = s(n) + hF(s(n + 1)), symplectic integrator (Chambers 1999) that could compute close
encounters. Symplectic integrators (for complete theory of sym-
where F is a known function and h is a chosen time-step. The
plectic integrators, see e.g. Sanz-Serna 1991; Yoshida 1993) do not
main advantage of the explicit methods is that they are very easy to
have a long-term build-up of energy error, and they are significantly
program, because at each time-step one calculation is performed.
faster with problems in which most of the mass is contained in a
Implicit methods are much more difficult as they can involve many
single body. The idea behind symplectic integrators is to split the
iterations per time-step. However, explicit methods have a tendency
Hamiltonian (H) of a system into pieces, each of which can be
to be unstable, and require very small time-steps, while the implicit
solved on its own. Symplectic integrator use a fixed time-step that
are stable and could use larger time-steps.
makes calculating close encounter difficult due to the necessity of
In this work, two numerical integrators from ORBFIT and MERCURY
changing the step size. One of the possible solutions is to use a
packages are used. In the following subsections, we briefly present
smaller step size or to split the perturbing terms, and to set a differ-
the characteristics of the chosen integrators, and also the properties
ent step size for each term (SYMBA integrator; Duncan, Levison &
of the packages ORBFIT and MERCURY.
Lee 1998). Chambers (1999) described an alternative solution a
hybrid integrator that has symplectic and non-symplectic compo-
2.1 ORBFIT nents. That alternative solution keeps the best properties of both
symplectic and non-symplectic integrators.
The ORBFIT is a free software written in FORTRAN 95, which can be
downloaded at http://adams.dm.unipi.it/orbfit. ORBFIT allows one to
compute the orbits of asteroids and to numerically simulate those
3 M E T H O D O L O G Y A N D DATA
orbits, as well. The implemented numerical integrator for the pur-
pose of long-term numerical integrations is ORBIT9 (Milani & To study the limitations of the BIM, first a fictional family is going
Nobili 1988). ORBIT9 uses a symplectic single-step method (im- to be created based on the Veritas asteroid family. In this respect,
plicit RungeKutta Gauss) as a starter, and an explicit multi-step all adopted values of the fictitious family are the same as of Veritas.
4 V. Radovic

Figure 1. Distribution of the test particles in the ae (top panel) and ai Figure 2. Dynamical structure of the test particles. LCEs are multiplied by
(bottom panel) planes. The equivelocity ellipses are obtained using Gauss the factor of 106 . The LCEs are inversely proportional to Lyapunov time
equations. (Tlyap ). The green crosses represent stable particles (Tlyap 105 yr), while
the black circles represent unstable ones, which are not considered in the
further analysis with BIM.

using both integrators (ORBIT9 and Hybrid symplectic integrator),1


Families are formed from the disruption of the target asteroid
while Yarkovsky is included only with Orb9. Simulations without
during a high-velocity impact. Escape speed (vesc ) is much smaller
Yarkovsky are supposed to demonstrate how the results are sensitive
than orbital speed (vorb 1520 km s1 ), and, consequentially, all
on the chosen integrator, while those with Yarkovsky drift included
the orbits of family members are similar immediately after their
are supposed to test the efficiency of the BIM for the more realistic
creation. To create the initial orbital distribution of the simulated
case.
family members, we have used Gauss equations. These equations
For the estimation of the Yarkovsky drift for the asteroid, the
describe changes in the osculating elements due to the components
diameter of which is 1 km, we used [da/dt] 5 104 au Myr1
of the ejection velocity:
(see Vokrouhlicky et al. 2015, and references within). For each test
2 particle, a random value in the interval ([da/dt], +[da/dt]) is
a/a = [(1 + e cos f )VT + (e sin f )VR ], assigned and divided by the asteroid diameter in order to represent
na(1 e2 )1/2
its Yarkovsky drift. In order to obtain asteroids diameters, we used
(1 e2 )1/2  e + 2 cos f + e cos2 f 
e = VT + (sin f )VR , absolute magnitudes H of the Veritas family in the interval [13 mag,
na 1 + e cos f 15 mag ]. Corresponding diameters are calculated using the equation
(1 e ) cos( + f )
2 1/2
(Bowell et al. 1989)
i = VW , (3)
na 1 + e cos f 10H /5
D [km] = 1329 , (4)
where a, e, i, f and are the semimajor axis, eccentricity, inclination, pv
true anomaly and argument of perihelion, respectively; na is the where pv is the geometric albedo. The Veritas family is a C-type
mean orbital velocity; and VT , VR and VW are the components family; so for the geometric albedo, an average value for the C-type
of ejection velocity along the direction of the motion, the radial is used: pv = 0.06 (DeMeo & Carry 2013).
direction and perpendicular to the orbital plane, respectively. The From the forward integration, we calculated Lyapunov character-
previously mentioned angles f and are the angles at the instant of istic exponents (LCEs, see Fig. 2). An LCE is inversely proportional
family creation, and their values are generally unknown but could be to Lyapunov time (Tlyap ), and it is used as an indicator of chaos in
estimated. By knowing these angles, we should be able to reproduce the considered system. As our test family corresponds to the lo-
distribution of the fragments after the collision. The curves obtained cation of the Veritas family, both of their dynamics are the same.
by describing the relationship e versus a and i versus a are called The Veritas family has been studied in various works (see e.g.
equivelocity curves for some ejection velocity V. Knezevic & Pavlovic 2002; Nesvorny et al. 2003; Tsiganis et al.
In our analysis, we created equivelocity curves for a fictional 2007; Novakovic et al. 2010; Carruba et al. 2017). The most impor-
asteroid family using V = 35 m s1 , f = 30 and = 330 . For tant role for the dynamics in this region is played by two three-body
the coordinates of the centre of the ellipse in the space of orbital mean motion resonances (MMR): 5J2S2A and 3J+3S2A
elements (a, e, sin i), we have (3.17, 0.067, 0.16), respectively. The (Fig. 2). The main chaotic zone in this region is around 5J2S2A
centre corresponds to the location of the Veritas family in the outer MMR (a 3.174 au), and particles in this resonance have a very
part of the main belt. The distribution of the 500 test particles is short Lyapunov time (Tlyap 10 kyr).
shown in Fig. 1. The region that our test particles occupy (and Veritas also) is not
To simulate the evolution of the test family, orbits of the family a dynamically stable part of the main belt. The motion of many
members are numerically integrated 20 Myr forward in time from particles is chaotic due to important three-body MMR (especially
the family formation, with gravitational model, which includes all
Solar system planets except Mercury as perturbing bodies. These
forward integrations are performed with and without Yarkovsky 1 From now on, we will refer to the ORBIT9 integrator as Orb9 and to the

drift included. Without Yarkovsky, the integrations are performed Hybrid symplectic integrator from the MERCURY package as HSM.
Limitations of backward integration method 5
Table 2. For each point in time after the family formation, determined ages are shown using the clustering of secular angles  and  . Results
are shown for each numerical integrator (Orb9 and HSM). m and  m are the minimum values of  and  , respectively.

Time Orb9 HSM


   
(Myr) Age (Myr) m ( ) Age (Myr)  m ( ) Age (Myr) m ( ) Age (Myr)  m (deg)

2 2.001 1.93 1.999 12.90 1.999 3.86 1.998 46.91


4 3.996 4.41 3.999 21.79 3.998 5.80 3.998 47.81
6 5.994 6.50 5.999 33.53 5.998 36.97 5.998 63.11
8 7.992 12.17 7.999 43.18 7.998 9.68 7.997 36.92
10 10.010 13.66 9.998 46.57 9.997 39.39 9.994 62.53
12 11.975 20.06 11.997 51.11 11.988 28.49 11.991 64.29
14 13.980 23.50 13.996 54.40 13.970 44.59 14.000 64.35
16 15.962 21.84 16.000 54.51 15.991 43.80 15.998 66.57
18 18.010 32.09 17.996 71.34 18.053 62.19 18.050 72.34
20 19.987 29.56 19.997 68.24 20.009 58.38 19.998 66.69

5JS2SA) and nearby 2:1 MMR with Jupiter. As this region For every backward integration, the average of differences in node
is highly chaotic, it is possible to use the method of the chaotic longitude ( ) and longitude of perihelion ( ) is calculated
chronology as well as BIM in order to obtain the age of a family. For using the following equations:
instance, Nesvorny et al. (2003) and Carruba et al. (2017) used BIM
1 
n
on stable particles of the Veritas family and obtained its age (8.3
 = (i j ), (5)
0.5 Myr and 8.23+0.37
0.31 , respectively), while Tsiganis et al. (2007) and n i=1
Novakovic et al. (2010) calculated the age with chaotic chronology j =i+1

(8.7 1.2 Myr). Both results are in very good agreement; hence,
1 
n
both methods could be applied to obtain the age of a family in this  = (i j ), (6)
region. But, it should be noted that BIM can only be applied to n i=1
j =i+1
the stable asteroids, while chaotic chronology is applicable to the
unstable ones. where n is the total number of values of  and  . To compare the
As the application of BIM is limited only to the objects on stable difference between integrators, osculating elements are used. But
orbits, therefore only particles with Tlyap 105 yr are considered. in the case where Yarkovsky drift is included, mean elements are
Of the 500 test particles, 291 particles have requested Lyapunov used, as they could significantly improve the obtained results.
time. From these differences, conjunctions of  and  are determined
Forward integrations give us time evolution of orbital elements as well as their maximum depth (m and  m ). To determine the age
for each test particle, and these data are used to test the limits of of the family, we have used quadratic function to fit the peaks where
BIM. Backward integrations are performed from different points in clustering is the most prominent. Obtained results when Yarkovsky
time after the formation of a family, starting from 2 Myr with the drift is not included are given in Table 2.
step of 2 Myr, until it finally reaches 20 Myr. As output from each From these results (Table 2), it is easily seen that the clustering
integration, osculating elements are obtained. To determine mean of  is not as deep as that of  for both integrators. This differ-
elements,2 we used a digital filter that allows us removal of the ence could be explained by the fact that the gradient of dg/da at
signal with periods up to 300 yr3 (Carpino, Milani & Nobili 1987; 3.153.18 au is considerably large due to nearby 2/1 MMR with
Knezevic & Milani 2000). Jupiter (Nesvorny et al. 2003). The ages of the simulated family are
For each backward integration, we tried to determine the clus- determined with large precision for each point in time using both
tering time of secular angles  and  , from which the age of the integrators (the largest deviation from nominal age is 50129 yr).
family is obtained. For the backward integration, Yarkovsky drift For example, plots for the backward integrations performed
was not used because the exact drift speed for the individual object 10 Myr after the family was formed can be seen in Fig. 3. For
is not known in the case of real families. Clone-based approach Orb9 integrator, maximum depths of  and  are 13. 06 and 46. 57,
could be used for the families whose Yarkovsky drift is not known, respectively, while their values for HSM are 39. 39 and 62. 53. In
but it is beyond the scope of this work. both cases, value of  m is significantly larger than the value of m .
Also, the clustering of both secular angles is deeper for Orb9 than
4 A N A LY S I S A N D R E S U LT S for the HSM integrator. The determined ages are very precise in
both cases, with the maximum deviation being 10 000 yr, which is
In order to test the limits of the BIM, we integrated the orbits of almost negligible.
a test family (see Section 3) 20 Myr forward in time using both Since we have the results obtained by using both integrators,
integrators (Orb9 and HSM). Then, from defined points in time, the it gives us an opportunity to compare them. For each integrator,
backward integration was performed with each integrator. the error is calculated as the difference between the determined
age and the expected one. Obtained results are given in Table 3.
2 Mean elements are elements freed from the short periodic perturbations. The largest difference between these integrators is 50 000 years
3 It is possible to obtain better result with filtering method developed by for the backward integration up to 18 Myr in the past. Hence, both
Carruba (2010), which allows removal of frequencies with period smaller integrators have a good accuracy in determining the age of the
than 105 yr. family.
6 V. Radovic

Figure 3. The average differences in the mean longitude of ascending node ( ), longitude of perihelion ( ) and DCF between  and  .
The results are shown for backward integrations performed 10 Myr after the family formation using both integrators (Orb9 and HSM). The most prominent
clustering is visible at 10 Myr.

Table 3. Differences between the age estimates and the true age of the family between the Orb9 and HSM
integrators. Columns are as follows: #1 time after the family formation; #2 and #3 deviation from the nominal
age and the age determined using clustering of  with Orb9 and HSM, respectively; #4 difference between error
with  for each integrator; #5 and #6 deviation from the nominal age and the age determined using clustering
of  with Orb9 and HSM, respectively; #7 difference between error with  for each integrator.

Time (Myr)  [106 (Myr)]  [106 (Myr)]


Orb9 MH  Orb9 MH 

2 714.56 1319.61 605.05 876.03 2058.31 1182.28


4 3494.32 1576.48 1917.84 1357.621 1670.55 312.93
6 6318.09 884.08 5434.01 384.93 14 771.21 14 386.28
8 8192.54 1902.91 6289.63 1300.28 2597.91 1297.63
10 10 213.06 2773.18 7439.88 1531.08 5293.20 3762.12
12 25 103.38 12 389.47 12 713.91 3096.62 8958.27 5861.65
14 19 526.5 30 412.23 10 885.73 4203.88 535.10 3668.78
16 37 612.24 9012.66 28 599.58 270.20 1491.44 1221.24
18 10 119.89 52 554.42 42 434.53 4293.25 50 129.30 45 836.05
20 12 466.81 9486.79 2980.02 3368.16 2365.35 1002.81

In order to determine the limits of BIM for Orb9, the linear the age is given with linear relationship ax + b, with coefficients
dependence between the age of the family and maximum depth a = 1.695, b = 2.079, while the coefficients for  m are a = 2.994,
of secular angles m and  m is going to be found. Their lin- b = 12.823. A similar idea is performed for HSM integrator (Fig. 5).
ear dependence is shown in Fig. 4. Correlation between m and Linear coefficient for m are a = 3.071, b = 0.465 and for  m are
Limitations of backward integration method 7
Table 4. Determined ages where the highest
correlation between  and  exists
for both integrators (Orb9 and HSM). Corre-
lation between them is calculated using DCF.

Time Orb9 HSM


(Myr) (Myr) (Myr)

2 1.999 1.999
4 3.998 3.997
6 5.999 5.996
8 7.998 7.996
10 9.997 9.995
12 11.997 11.991
14 13.996 14.001
16 15.998 16.003
18 17.995 18.070
20 19.995 19.993

a = 1.289, b = 44.602. Due to the larger dispersion of angles with


the HSM integrator, error of the fit used for HSM is significantly
larger than it is for Orb9.
For Orb9 integrator, the conjunction of  m reaches 90 for the
integration time of 25 Myr. However, based on the obtained linear
fit, we estimate that m reaches 90 for 50 Myr integration time.
But in practice these results cannot be reached due to large noise
and a possibility of other clustering with the same depths (see e.g.
Novakovic 2010; Novakovic et al. 2012a). Therefore, clustering can
Figure 4. Linear dependence between the maximum depth of clustering of be effectively determined up to 70 . Hence, for Orb9 the limit for
secular angles m (top panel) and  m (bottom panel) and integration time. finding significant clustering is up to 20 Myr for  m , and for m it
Results are obtained using Orb9 integrator from ORBFIT package.
goes up to 40 Myr. For the HSM integrator, the minimum depth for
both m and  could be determined for up to 25 Myr of integration
time. We have to keep in mind that these results are obtained without
including the Yarkovsky drift; therefore, we expect a lower upper
limit for integration time in a more realistic case.
Around the time of the creation of a family, both secular angles
have to cluster around the same value. Therefore, it is desirable to
test whether their clustering is not some random fluctuation, but
rather an indication that the considered objects share a common
origin. This is the motivation for using discrete correlation function
(DCF, Edelson & Krolik 1988) to study the correlation between
clustering of  and  . In our case, DCF for the arrays of (a =
 ) and (b =  ) is given by

1 (ai a)(bj b)
DCFij = , (7)
n a b

where a, b are the mean values for both arrays; a , b represent


their standard deviations. The value of DCF is between 1 and 1,
where 0 means that there is no correlation between the data sets.
Positive or negative value of DCF means positive (increasing) or
negative (decreasing) correlation. The determined ages when the
highest correlation between  and  exists are shown in
Table 4. This data suggests that the clustering that we used in our
work is statistically significant.

4.1 The role of Yarkovsky drift

Figure 5. Linear dependence between the maximum depth of clustering of In order to test the more realistic case, our following simulation
secular angles m (top panel) and  m (bottom panel) and integration time. is performed with Yarkovsky effect included in the model. In the
Results are obtained using HSM integrator from MERCURY package. forward integration, Yarkovsky effect is included, but it is excluded
in the backward integrations because it is generally unknown.
8 V. Radovic
Carruba et al. 2017). The inability to find convergence of  is
mainly due to presence of the nearby 2:1 resonance with Jupiter as
well as encounters with (1) Ceres and (10) Hygiea. Carruba et al.
(2017) managed to minimize the dispersion of  and to obtain its
clustering by choosing an appropriate 2 variable (see equation 5
in their work).
For other families, we calculated lower values for the minimum
of clustering of secular angles and that could be explained as be-
ing a consequence of different locations and dynamics between our
test family and considered families. In addition to dynamical influ-
ences, close encounters with massive asteroids could significantly
increase the convergence of secular angles.4 And finally, as we used
larger number of family members it could explain difference in re-
sults. As an example, the minimum clustering of m for P/2006
VW139 is 24.2 from our model, while Novakovic et al. (2012a)
obtained 32 .

5 CONCLUSIONS
In this work, we presented the following results:
(i) We performed forward integration in order to simulate the
evolution of fictional family in time. Forward integrations are per-
formed using two different integrators: Orbit9 from ORBFIT and Hy-
brid symplectic integrator from MERCURY package (integrations are
performed without the Yarkovsky drift). Additional forward inte-
gration with Orbit9 is performed with the Yarkovsky drift included
Figure 6. Linear dependence between the maximum depth of clustering in order to simulate more realistic situation in the main belt.
of secular angles m (top panel) and  m (bottom panel) and integration (ii) The backward integrations are applied to our fictional family
time. Results are obtained using Orb9 integrator where Yarkovsky drift was but with different points in time from the family creation. With
included in forward integrations. this approach, we managed to determine the depth of clustering of
secular angles  and  . We used these clustering to obtain the age
The same procedure for the age determination is applied as in the of the family and to find a correlation between these depth and the
case without the Yarkovsky effect included. The linear dependence age. To test the significance of clustering between  and  , DCF
between the obtained ages and angles m and  m is given in Fig. 6. is used.
Linear coefficients for fitting between m and integration time are
a = 4.914, b = 12.261 and for  m are a = 10.706, b = 23.778. The main result of this work is finding the upper limit in applying
From these results, conjunction of longitude of perihelion starts the BIM. Based on our results, if Yarkovsky drift is included we
at minimum value of 40 , for integration time of 2 Myr and goes could find significant clustering of secular angle  for the families
above 80 just 6 Myr after the family formation. Above this value, not older than 18 Myr. The clustering of  is harder to distinguish,
it is almost impossible to reliably distinguish its clustering. For m , but DCF can be used to confirm its statistical significance. Our
the situation is better, and clustering goes to 80 at 20 Myr after the results are consistent with the results concerning other families
family formation. DCF of this data has a significant correlation for where BIM was applied. Both HSM and Orb9 integrator showed
each time span, and can be used to show statistical significance of good accuracy, but Orb9 has a deeper clustering of secular angles
the clustering. But, we have to note that the values of DCF are a than HSM integrator.
little above zero, and that DCF has a large fluctuations in this case. The BIM is the most accurate method for asteroid family age
From these results, we could conclude that with the Yarkovsky determination, but from other papers and from our simulations it
effect included, m reaches 70 around 18 Myr, while  is always is evident that BIM is only useful for very young families. Other
above this value, except for the integration time less than 4 Myr. limitation of BIM is that its application is limited to dynamically
Therefore, significant clustering could be found for an integration stable bodies. Using BIM to the families in the chaotic regions
time up to 18 Myr for m . Due to the problems with distinguish- could increase the dispersion of secular angles, especially  . Close
ing clustering of  m , DCF can be used to confirm its statistical encounters can significantly influence the convergence of the lon-
significance. gitude of perihelion as it was shown in various cases. For the Karin
As our fictional family corresponds to the location of the Veritas cluster, close encounters increased its dispersion by 4 (Carruba
family, it is interesting to compare the obtained limits to the results et al. 2016), but in the case of Veritas family convergence of  is
obtained for Veritas family by other authors. The age of the Veritas completely destroyed (Carruba et al. 2017).
family is around 8.3 Myr, hence we should expect to find the fol-
lowing minima of clustering: m = 28.5 ,  m > 80 . In Table 1,
we see that our results match the results obtained by other authors. 4 Carruba et al. (2016) discussed the influence of close encounters with
The difference between the clustering of  is around 3 , while the Ceres for the convergence of secular angels of the Karin cluster. They
clustering of  cannot be distinguished, which was also suggested found that close encounters can increase the dispersion of the angles by
in the papers concerning the Veritas family (Nesvorny et al. 2003; approximately 4 .
Limitations of backward integration method 9
From our results, sufficiently significant clustering of secular Knezevic Z., Pavlovic R., 2002, Earth Moon Planets, 88, 155
angles could be reliably determined for the integration time up Milani A., Knezevic Z., 1990, Celest. Mech. Dyn. Astron., 49, 347
to 18 Myr. By using Yarkovsky clones and accounting for close Milani A., Nobili A. M., 1988, Celest. Mech., 43, 1
encounters with massive asteroids in the backward integrations, it Milani A., Cellino A., Knezevic Z., Novakovic B., Spoto F., Paolicchi P.,
2014, Icarus, 239, 46
is probably possible to achieve better results.
Milani A., Knezevic Z., Spoto F., Cellino A., Novakovic B., Tsirvoulis G.,
Other integrators were not used in this work, and neither were
2017, Icarus, 288, 240
other possible locations within the main belt, except the one which Nesvorny D., Bottke W. F., 2004, Icarus, 170, 324
corresponds to the location of the Veritas family. If different lo- Nesvorny D., Vokrouhlicky D., 2006, AJ, 132, 1950
cations were used, that could lead to different dynamics of a test Nesvorny D., Bottke W. F., Jr, Dones L., Levison H. F., 2002, Nature, 417,
family, and it may give slightly different results. For the future work, 720
it should be interesting to test how different location within the main Nesvorny D., Bottke W. F., Levison H. F., Dones L., 2003, ApJ, 591, 486
belt affect the determined limits of BIM. Nesvorny D., Jedicke R., Whiteley R. J., Ivezic Z., 2005, Icarus, 173, 132
Nesvorny D., Vokrouhlicky D., Bottke W. F., 2006, Science, 312, 1490
Nesvorny D., Bottke W. F., Vokrouhlicky D., Sykes M., Lien D. J.,
AC K N OW L E D G E M E N T S Stansberry J., 2008, ApJ, 679, L143
Nesvorny D., Broz M., Carruba V., 2015, in Michel P., DeMeo F.E., Bottke
I would like to thank Prof. Valerio Carruba, the referee, for his
W.F., eds, Asteroids IV. Univ. Arizona Press, Tucson, AZ, p. 297
valuable comments and suggestions, which significantly improved Novakovic B., 2010, MNRAS, 407, 1477
this manuscript. Also, I am grateful to Bojan Novakovic for valuable Novakovic B., Tsiganis K., Knezevic Z., 2010, MNRAS, 402, 1263
help during the work on this paper. Novakovic B., Cellino A., Knezevic Z., 2011, Icarus, 216, 69
Novakovic B., Hsieh H. H., Cellino A., 2012a, MNRAS, 424, 1432
Novakovic B., DellOro A., Cellino A., Knezevic Z., 2012b, MNRAS, 425,
REFERENCES 338
Bottke W. F., Vokrouhlicky D., Nesvorny D., 2007, Nature, 449, 48 Novakovic B., Hsieh H. H., Cellino A., Micheli M., Pedani M., 2014, Icarus,
Bottke W. F. et al., 2015, Icarus, 247, 191 231, 300
Bowell E., Hapke B., Domingue D., Lumme K., Peltoniemi J., Harris A. Radovic V., Novakovic B., Carruba V., Marceta D., 2017, MNRAS,
W., 1989, in Binzel R. P., Gehrels T., Matthews M. S., eds, Asteroids II. 470, 576
Univ. Arizona Press, Tucson, AZ, p. 524 Rosaev A., Plavalova E., 2017, Icarus, 282, 326
Carpino M., Milani A., Nobili A. M., 1987, A&A, 181, 182 Sanz-Serna J. M., 1991, Acta Num., 243
Carruba V., 2010, MNRAS, 408, 580 Spoto F., Milani A., Knezevic Z., 2015, Icarus, 257, 275
Carruba V., Nesvorny D., 2016, MNRAS, 457, 1332 Tsiganis K., Knezevic Z., Varvoglis H., 2007, Icarus, 186, 484
Carruba V., Domingos R. C., Nesvorny D., Roig F., Huaman M. E., Souami Vokrouhlicky D., Nesvorny D., 2011, AJ, 142, 26
D., 2013, MNRAS, 433, 2075 Vokrouhlicky D., Milani A., Chesley S. R., 2000a, Icarus, 148, 118
Carruba V., Nesvorny D., Aljbaae S., Huaman M. E., 2015, MNRAS, 451, Vokrouhlicky D., Farinella P., Bottke W. F., 2000b, Icarus, 148, 147
244 Vokrouhlicky D., Broz M., Morbidelli A., Bottke W. F., Nesvorny D.,
Carruba V., Nesvorny D., Aljbaae S., Domingos R. C., Huaman M., 2016, Lazzaro D., Rivkin A. S., 2006a, Icarus, 182, 92
MNRAS, 458, 3731 Vokrouhlicky D., Broz M., Bottke W. F., Nesvorny D., Morbidelli A., 2006b,
Carruba V., Nesvorny D., Vokrouhlicky D., 2016, AJ, 151, 164 Icarus, 182, 118
Carruba V., Vokrouhlicky D., Nesvorny D., 2017, MNRAS, 469, 4400 Vokrouhlicky D. et al., 2009, A&A, 507, 495
Cellino A., Michel P., Tanga P., Zappala V., Paolicchi P., DellOro A., 1999, Vokrouhlicky D., Bottke W. F., Chesley S. R., Scheeres D. J., Statler T. S.,
Icarus, 141, 79 2015, in Michel P., DeMeo F. E., Bottke W. F., eds, Asteroids IV. Univ.
Chambers J. E., 1999, MNRAS, 304, 793 Arizona Press, Tucson, AZ, p. 509
Chambers J. E., Migliorini F., 1997, DPS meeting #29, 27.06 Vokrouhlicky D. et al., 2017, A&A, 598, A91
DeMeo F. E., Carry B., 2013, Icarus, 226, 723 Yoshida H., 1993, Celest. Mech. Dyn. Astron., 56, 27
Duncan M. J., Levison H. F., Lee M. H., 1998, AJ, 116, 2067 Zappala V., Farinella P., Knezevic Z., Paolicchi P., 1984, Icarus, 59, 261
Edelson R. A., Krolik J. H., 1988, ApJ, 333, 646 Zappala V., Cellino A., Farinella P., Knezevic Z., 1990, AJ, 100, 2030
Eggl S., Dvorak R., 2010, in Souchay J., Dvorak R., eds, Lecture Notes in
Physics, Vol. 790, Dynamics of Small Solar system Bodies and Exo-
planets. Springer, Berlin, p. 431
Farinella P., Vokrouhlicky D., 1999, Science, 283, 1507
Knezevic Z., Milani A., 2000, Celest. Mech. Dyn. Astron., 78, 17 This paper has been typeset from a TEX/LATEX file prepared by the author.

Das könnte Ihnen auch gefallen