1.

Introduction
The asteroids forming the Main Belt between Mars and Jupiter
are remnants of the bricks that accreted to form the planets.
Prints of the events that occurred in the early Solar System are
still present in the distribution of their orbit, size, and compo-
sition (DeMeo & Carry, 2014; Morbidelli et al., 2015; Clement
et al., 2020). The current population, however, differs from its
pristine distribution. Giga-years of collisions have fragmented
bodies and created clumps of objects called families (Hirayama,
1918; Zappala et al., 1990; Milani et al., 2014). All dynamical
structures are furthermore secularly spreading through the non-
gravitational Yarkovsky effect (Bottke et al., 2001).
The Yarkosky effect results from the delayed re-emission of
the Solar incident flux and depends on many physical and surface
properties such as diameter, albedo, density, obliquity, and rota-
tion period (Vokrouhlický et al., 2015). Some properties, such as
the diameter, are available for hundreds of thousands of asteroids
thanks to mid-infrared (MIR) surveys such as IRAS, AKARI, or
WISE (Tedesco et al., 2002; Usui et al., 2011; Masiero et al.,
2011). Others, such as the density or thermal inertia, are much
less constrained and available for only a tiny fraction of asteroids
(see Berthier et al., 2023, for a recent compilation).
Spin properties (rotation period and coordinates) typically
require numerous photometric measurements over a long pe-
? The catalog is available at the CDS via anonymous ftp to
cdsarc.u-strasbg.fr (130.79.128.5) or via http://cdsarc.
u-strasbg.fr/viz-bin/cat/J/A+A/XXX/xxx
riod of time, covering several apparitions ( ˇDurech et al., 2015).
These measurements can be dense-in-time time series, here-
after referred to as "light curves" (Kaasalainen et al., 2001),
or they can be photometry sparse-in-time, collected by surveys
(Kaasalainen, 2004). However, owing to the potential multiple
period aliases from sparse data, a light curve often helps to un-
ambiguously determine the rotation period ( ˇDurech et al., 2015).
Surveys aimed at discovering and characterizing exoplanets
via the transit method offer a tremendous amount of time series
over wide fields in which asteroids can be searched for (Berthier
et al., 2016; Grice et al., 2017). Recently, Pál et al. (2020) ex-
tracted light curves for 9912 Solar system objects (SSOs) from
the Transiting Exoplanet Survey Satellite (TESS) first data re-
lease (DR1). The released catalog of asteroid light curves and
periods is dubbed TSSYS-DR1. Pál et al. (2020) determined the
rotation period of 9912 asteroids by fitting second order Fourier
series. However, in some cases, such as asteroids (118) Peitho,
(511) Davida, and (775) Lumiere among many others, it is not
enough to describe accurately the light curve as second-order in
the Fourier Series (Scheirich & Pravec, 2009), leading to poten-
tially erroneous period determinations.
In this work, we aim here to analyse the large sample of
light curves released by Pál et al. (2020) with a robust period-
determination algorithm. We present examples demonstrating
that a Fourier series of the second order is often not sufficient
to describe asteroid light curves. The article is organized as fol-
lows. In Section 2, we describe how we model the light curve of
asteroids. In Section 3, we explain how we selected the optimum
Article number, page 1 of 8
arXiv:2501.07189v1 [astro-ph.EP] 13 Jan 2025
A&A proofs: manuscript no. main
Fig. 1. Top. Distribution of the maximal, minimal, and range of phase
angles. Bottom. Range of phase as a function of the minimum phase
angle.
solution among different potentially degenerated solutions. We
then present the results of the analysis of the TESS light curves in
Section 4. We validate these results in Section 5 and we discuss
their implications in Section 6. Finally, we give our conclusions
in Section 7.
2. Modeling asteroid light curves
The apparent visual magnitude of an asteroid depends on the
geometry of observation and can be described as (Bowell et al.,
1989):
V = H + 5 log10 r∆ − 2.5 log10 φ(α) + g(t), (1)
where H is the absolute magnitude of the asteroid, r and ∆ are
the heliocentric distance and range to the observer (in au), g(t) is
a periodic function related to the asteroid shape in rotation, and
the phase function φ(α) describes how brightness evolves with
the phase angle α (the Sun-asteroid-observer angle; see Bowell
et al., 1989; Muinonen et al., 2010, for the different functions
in usage in the community). The present study focuses on the
determination of the period of the g(t) function.
2.1. Reduction to unit distances
To find the rotation period of an asteroid, we first take into ac-
count the change of visual magnitude occasioned by the chang-
ing distances due to the asteroid and Earth orbital motions. For
each observation, i, we compute:
V′
i = Vi − 5 log10 ri∆i. (2)
which depends only on the asteroid absolute magnitude, rota-
tion, and phase function. We also correct the timing of the i-th
observation by:
t = t′ − c
∆i
, (3)
where c is the speed of light, t′ is the recorded observation epoch,
and t is the actual epoch for the observed visual magnitude.
Before moving into further analysis, we exclude the observa-
tions deemed to be unreliable. We exclude all observations with
0 2 4 6 8 10 12
Time (JD - 2458437.98416639)
4.85
4.90
4.95
5.00
5.05
5.10
TESS magnitude
Data (572) Outliers (29) Fit
0.0 0.2 0.4 0.6 0.8 1.0
Rotation phase (7.210h)
4.85
4.90
4.95
5.00
5.05
5.10
TESS magnitude
Data (572) Outliers (29) Fit
5 10 15 20
Frequency (cycles/day)
0.01
0.02
0.03
0.04
σ Solution
Periodogram
Significance threshold
Fig. 2. Example of light curve fit for asteroid (3) Juno. Top. TSSYS-DR1
data as function of time. Observations are plotted as dots, color-coded
by epoch. Crosses represent data rejected by the fitting procedure (see
text). The line represents the best fit. Middle. Same as above, folded
over a rotation period (7.2100.001
−0.001). Bottom. Periodogram showing the
residuals of the fit as a function of rotation frequency (number of cy-
cles per day). The horizontal dashed line correspond to the significance
threshold (sigma-level, see text).
flags different from 0 (see Table 1 in Pál et al., 2020, for a full
description of the flags). Observations with flag values 128 and
16384 (manual exclusion for anomaly or because outlier) were
only excluded from the first iteration of fitting and may be in-
cluded in the next iterations according to a 3 σ rule.
2.2. Fourier fitting and phase function
The reduced V′ magnitude fluctuates because of the phase func-
tion φ(α) and the periodic function g(t), with a period equals to
the asteroid’s synodic rotation period. The function g(t) is de-
composed into a Fourier series:
g(t) =
k∑
j=1
A j cos(2π f jt) + Bj sin(2π f jt), (4)
where f is the frequency of rotation (cycles per day), t is time (in
days), k is the maximal number of Fourier series, and {A j, Bj} the
amplitude coefficients at each frequency. In the present analysis,
we consider values of k up to k = 10.
Article number, page 2 of 8
D. E. Vavilov and B. Carry: Rotation periods of asteroids from light curves of TESS
data
The phase angles of asteroids observed by TESS typically
span a small range or a few degrees only, around 3◦ of phase,
with minimum values of typically 3◦ and maximum values
around 8◦ (Figure 1). The lack of coverage of the opposition
surge implies that the system of equations is ill-conditioned for
finding the parameters of the usual phase functions H, G (Bow-
ell et al., 1989) or H, G1, G2 (Muinonen et al., 2010). We refer
to Mahlke et al. (2021) for a discussion of phase coverage. We
hence follow here the approach by Pál et al. (2020) of describing
the phase function as a second-degree polynomial:
φ(α) = c0 + c1α + c2α2.
This approach precludes the determination of the absolute
magnitude, H, of an asteroid, but that is not our goal. On the
other hand, the fitting is accurate and well conditioned. We limit
here the possible range of values for c1 and c2 to ascertain the
physical results (i.e., increasing reduced magnitude with phase
angle). We used the H, G phase function (Bowell et al., 1989) as
a reference and assumed that the parameter G cannot take values
outside the given range (−0.25, 0.95). The H-G function cannot
be properly approximated by a second degree polynomial. That
is why we find an approximation on the arc of 1◦. For each angle
α from {1◦, 2◦, 3◦, ..., 120◦} we construct two phase curves with
(Gmax = 0.95 and Gmin = −0.25). We then fit parameters c0, c1,
and c2 on the arc (α − 0.5◦, α + 0.5◦). The maximal and minimal
values for c1 and c2 found this way for a particular phase angle
are then used as boundaries for the fit of TESS data. We should
note here that the free term c0 encompasses H, which is removed
from the fitting procedure.
For a given frequency, f , we found the parameters c0, c1,
c2, {A j, Bj
}
j=1,k via a least-squares minimisation. We took the
weights of observations into account according to chapter VIII
in Linnik (1961). We also excluded observations differing from
the fit by more than 3 σ iteratively until the procedure converged.
We put “soft” limitations on the parameters c1 and c2 in
the least-squares fitting procedure, so that these values cannot
be substantially outside the interval [cmax
j , cmin
j ]. We added the
following two equations to the system of conditional equations
(Gubanov, 1997, p. 73):
c j = (cmax
j (α) + cmin
j (α))/2, j = 1, 2, (5)
where α is a minimal phase angle in the observation data
of the asteroid. These equations are added with weights of
1 /(cmax
j (α) − cmin
j (α))/2 . The aim of it is to guide the fitting pro-
cedure toward mean values of c1 and c2 of (cmax
j (α) + cmin
j (α))/2,
with standard deviations of (cmax
j (α) − cmin
j (α))/2. We present
in Figure 2 an example of the fitting procedure, on asteroid
(3) Juno.
3. Selection of the best model
To find the rotational period, we sampled different frequencies,
f , and for each of them we fit via least-squares routine the 2k +
3 parameters c0, c1, c2, {A j, Bj
}
j=1,k. We did not simply select
the solution with the lowest residuals; rather, we imposed the
following criteria to be fulfilled.
3.1. Defining the frequency range
First, at least two full periods of the asteroid rotation must have
been observed. Shorter coverage could indeed result in spurious
period determination. We thus set a minimal frequency:
fmin = 2
tmax − tmin
, (6)
where tmax and tmin are the epochs of the last and first observa-
tion, respectively. Owing to the TESS observing strategy, each
sector is observed for approximately a month, namely: fmin >
0.05 days−1
Furthermore, TESS full-frame images are taken every 30
min approximately (Pál et al., 2020). The maximal possible fre-
quency is thus analogous to a Nyquist-limit, adapted to the case
of irregular sampling (Eyer & Bartholdi, 1999): where p is the
largest value, such that each ti can be written as ti = t0 + ni p,
for integers, ni; then the Nyquist frequency is fNy = 1/(2p). Of
course, the equality ti = t0 + ni p cannot be fulfilled precisely,
so we allowed for a 2% accuracy for this equation. For most as-
teroids in TSSYS-DR1, the Nyquist frequency is approximately
24 days−1, implying a minimal rotational period of about 1 hour
(however, in some cases, fNy reached about 48 days−1). We set
the right end of the frequency interval to 24 days−1, thus: a min-
imal rotation period of 1 hour.
We note that the maximal frequency decreases with the num-
ber of Fourier series (i.e., greater k). The k series contains a
term in cos(2π f kt) which period is 1/k f , and hence the maxi-
mal frequency is fNy/k. We consider 40,001 possible frequen-
cies on the interval between fmin and min( fNy, 24) for k = 1. For
k > 1, we keep the frequency steps and the interval becomes
[ fmin, min( fNy/k, 24)].
3.2. Choosing the optimum fit
For each number of Fourier series, k, we first found the best fit
according to our criteria (Equation 7). We then compared fits
obtained with different k to select the final solution.
The main criterion is the minimal value of standard deviation
of one observation σ, which is a classical unbiased estimate of a
observational dispersion (Linnik, 1961):
σ2 = Nobs
∑ j p j
∑i pi(Oi − Ci)2
(Nincl − npar) , (7)
where Oi and Ci are i-th observation and computed magnitude, pi
is the weight of i-th observation (taken as 1/σ2
i ), Nobs is the num-
ber of observations, Nincl is the number of observations included
in the fit, and npar is the number of fitted parameters (2k + 3).
We note that the number of observations included in the
fit can slightly differ for each frequency. The factor Nobs in
Nobs
/∑ j p j is required to bring the value σ to the observational
error with mean weight. This does not change the result but helps
us understand the accuracy of the fit. We chose the model with
the lowest value of σ as solution for this number of Fourier se-
ries.
If the shape of an asteroid is well-described by a tri-axial el-
lipsoid, the light-curve is expected to display two local maxima
and two local minima. However, the number of local maxima
from fitting can be up to k. For instance, eight local minima can
appear with eight Fourier series. This point is critical as many
objects present two solutions for the frequency, only differing by
a factor of two. There are cases where the two solutions are in
Article number, page 3 of 8
A&A proofs: manuscript no. main
0 2 4 6 8
Time (JD - 2458382.04080062)
5.9
6.0
6.1
TESS magnitude
(511) Davida
0 5 10 15 20 25
Time (JD - 2458543.24578116)
11.0
11.1
11.2
11.3
11.4 (699) Hela
0 2 4 6 8 10
Time (JD - 2458382.18951445)
14.75
15.00
15.25
15.50
15.75 (543339) 2014 AX12
0.0 0.2 0.4 0.6 0.8 1.0
Rotation phase
5.9
6.0
6.1
TESS magnitude
0.0 0.2 0.4 0.6 0.8 1.0
Rotation phase
11.0
11.1
11.2
11.3
11.4
0.0 0.2 0.4 0.6 0.8 1.0
Rotation phase
14.75
15.00
15.25
15.50
15.75
5 10 15 20
Frequency (cycles/day)
0.02
0.04
0.06
0.08
0.10
Sigma
Period: 5.130+0.001
−0.001 h
5 10 15 20
Frequency (cycles/day)
0.040
0.045
0.050
0.055
Period: 3.397+0.001
−0.002 h
5 10 15 20
Frequency (cycles/day)
0.120
0.125
0.130
0.135
Period: 74.355+163.351
−72.355 h
Fig. 3. Illustrations of the advantages of the method (symbols are the same as Fig.
2Example of light curve fit for asteroid (3) Juno. Top.
TSSYS-DR1 data as function of time. Observations are plotted as dots, color-coded
by epoch. Crosses represent data rejected by the fitting procedure
(see text). The line represents the best fit. Middle. Same as above, folded over a
rotation period (7.2100.001
−0.001). Bottom. Periodogram showing the
residuals of the fit as a function of rotation frequency (number of cycles per
day). The horizontal dashed line correspond to the significance
threshold (sigma-level, see text).figure.2). Left: Clear period determination, but
it would be (erroneously) twice smaller if only the second order
of the Fourier series was used. Center: Two solutions with acceptable sigma values
but corresponding to the same period – if the number of local
maxima is taken into account. Right: All solutions below the significance level,
thus, the uncertainty interval is 2–237 h, i.e., the period is not
determined.
reality associated with the same period (e.g., Figure 3), particu-
larly if one of the solutions refers to a light curve with one local
maxima only.
We repeated the above procedure for each number of Fourier
series, k, from 1 to 10. For each k, we have a single candidate
for the final model and final period. To choose among these, we
used the F-test and computed it for each pair of models. The F-
test tells us whether the difference between two dispersions is
significant or not. The F-test gives us the probability (p-value)
for the first model being better than the second one. In total,
we obtained 10 × 9/2 = 45 results for the F-tests. Following the
Occam’s razor rule, we chose the model with smallest possible
number of Fourier series. We chose the smallest k for which no
other model is significantly better, namely, such a k value that
all the p-values of F-tests of k and any j are less than 95%
For this chosen model, we computed its sigma-level (F-
test, p-value 90%). With the F-test, we computed pvalue =
Ftest(σ1, σ2, n1, n2). However, we can also solve the reverse
problem: for a given σ1, the aim is to find the σ2 of the sec-
ond model, so that the p-value is 90%. In this case, we used the
same number of used observations (n1 = n2). This is how we
define the sigma-level. All the models with σ values lower than
this threshold are also possible, with their associated rotation pe-
riod. As a result, we provide a range of valid rotation period and,
whenever applicable, all the other potential periods that are not
in the computed interval (i.e., generally a multiple of 2 of the
period). In the case of a high number of possible periods, we
provide the whole interval for all, which means that the result is
imprecise.
Also, we checked whether using a different number of
Fourier series can lead to any plausible solutions (the σ value
of which is lower than for the chosen one). In general, this is the
case when for a higher order of Fourier series, the frequency can
only be low (so the period is high) and the period is doubled or
even tripled. These periods might be realistic (i.e., the σ is still
lower than for chosen solution), but still unlikely.
Article number, page 4 of 8
D. E. Vavilov and B. Carry: Rotation periods of asteroids from light curves of TESS
data
0 10 20 30 40 50
Period relative precision (%)
1
10
100
1000
Count
100 101 102 103
Period (h)
10−3
10−2
10−1
100
101
102
103
Period uncertainty (h)
Valid (4829)
Rejected (5054)
1:1 1:10 1:1001:1 1:10 1:100
0.0
0.5
1.0
Fraction (valid)
Fig. 4. Comparison of the uncertainty on the periods with the periods.
4. Results
From the sample of 9912 asteroids with TESS light curves, we
found a period for all, with 4839 (48.8%) for which the period
has been determined with a 33% relative error (i.e., the uncer-
tainty on the period is at most a third of the period). Hereafter,
we consider as “valid” the periods determined with this level of
precision and “reject” the others. While there is a long tail of less
precise determinations (Figure 4), the determinations are more
precise than a percent for 3042 (30.7%) asteroids and 10% for
4521 (45.6%) asteroids.
We present in Figure 5 the distribution of the validated and
rejected solutions as a function of the apparent magnitude of the
asteroids during TESS observations and the amplitude of their
light curves. There is a clear trend with rejected solutions being
mostly for the faintest targets with low amplitude light curves.
Hence, inaccurate solutions are due to a level of noise that is too
high compared to the signal, as expected.
Finally, 4366 (44.0%) asteroids have a unique, non-
ambiguous, rotation period. The remaining 492 (5.0%) asteroids
have a single other possible period in 277 (56.3%) cases, and
several ambiguous periods in 215 (43.7%) cases. The distribu-
tion of these degenerated solution is not random but corresponds
to period aliases, as visible in Figure 6. All the data are available
as electronic format on the CDS.
5. Validation
We assessed the quality of the periods determined here with three
comparisons: the original TSSYS-DR1 results of Pál et al. (2020),
the rotation periods on an independent extraction of TESS pho-
tometry by McNeill et al. (2023), and other values from the lit-
erature (providing periods from altogether different data sets).
As visible in Figure 7, there is a significant spread among peri-
ods when comparing the present determinations with those from
Pál et al. (2020). It, however, mainly concerns rejected periods.
Among the validated solutions (4829, i.e., 48.8% of the sam-
0
500
1000
6 8 10 12 14 16 18
Mean apparent magnitude
0.0
0.2
0.4
0.6
0.8
1.0
Light curve amplitude (mag)
Valid (4829)
Rejected (5054)
500 1000
0.5
1
0.5 1
Fig. 5. Distribution of the validated and rejected samples against light
curve amplitude and apparent magnitude. The black lines represent the
fraction of rejected asteroids in the entire sample.
0 1 2 3 4 5
Ratio of periods
0
200
400
600
800
1000
1200
1400
Count
2:1 3:1 4:11:2 3:2 5:2
Fig. 6. Ratio of periods between ambiguous solutions. The vertical
dashed lines indicate the ratio of integers.
ple), the comparison is much better and 78.5% of the solutions
agree within 1%. Another 3.6% correspond to period aliases
(half or double period). The comparison with McNeill et al.
(2023) presents a larger spread, even in the validated solutions
(3601, i.e., 40.0% of the common sample). Overall, 63.1% of the
solutions agree within one percent and an additional 2.9% cor-
respond to period aliases. While based on similar original TESS
data, the two studies extracted the photometry and determined
the period using different methods, explaining the differences
observed here. In their study, McNeill et al. (2023) actually re-
ported numerous cases of disagreement with Pál et al. (2020),
while about 80% of solutions agreed. For instance, while mainly
periods reported by Pál et al. (2020) correspond to either half
or twice the period reported here (Figure 7), those reported by
McNeill et al. (2023) almost never correspond to half the period
reported here.
We then made a comparison with periods reported by other
authors on different data sets (excluding TESS). We thus com-
piled the rotation period for each asteroid in our sample, using
the ssoCard of SsODNet1 through its rocks2 interface (Berthier
et al., 2023). We focus on period determinations with a quality
flag of 3, that is, those that are deemed definitive, following the
criterion by LCDB (Warner et al., 2009). As visible on Figure 7,
1 https://ssp.imcce.fr/webservices/ssodnet/
2 https://rocks.readthedocs.io/en/latest/
Article number, page 5 of 8
A&A proofs: manuscript no. main
1 101 102
Period (h)
100
101
102
103
Comparison period (h)
Pál et al. (2020)
Reject (5050)
Valid (4788)
1 101 102
Period (h)
McNeill et al. (2023)
Reject (3008)
Valid (3601)
1 101 102 103
Period (h)
Literature
Reject (131)
Valid (1483)
1:1
Aliases
1:1
Aliases
Fig. 7. Comparison of rotation periods found here with those derived by Pál et al.
(2020) (9912, left) by McNeill et al. (2023) (6635, center) and
gathered from the literature (1483 periods for 670 asteroids, right). The black
line represents full agreement and the dashed gray lines the aliases
for double and half periods.
there is globally a better agreement with 1483 period determina-
tions (for 670 unique asteroids). We find 91.0% agreeing within
one percent, with additional 0.9% corresponding to aliases. This
highlights the robustness of the method used here to select the
period among the different possible solution (Section 3).
There are three main advantages of the technique presented
in the present study. First, we considered several number of
Fourier series, while also taking into account the number of local
maxima (there should be at least two local maxima, otherwise we
would be doubling the period). We assume that the light curve
is a result of the shape features. In Figure 3 one can see the re-
sult for asteroid (511) Davida. The period is clearly determined
and in agreement with many previous studies from ground-based
light curve observations (e.g., De Angelis, 1995; Torppa et al.,
2003; Cellino et al., 2019; Vernazza et al., 2021); however, at
least the third order of a Fourier series is required. The result for
the second order is twice smaller and coincides with Pál et al.
(2020). Our explanation for this is that the light curve is quite π-
periodic, on the one hand, and it does not resemble a cosine, on
the other hand. Therefore the second-order Fourier series with
twice smaller period can fit the contour of the light curve much
better, but once we take third-order Fourier series, the difference
between the two peaks of the light curve becomes important.
Second, we checked the number of local extrema. We illus-
trate this second advantage with asteroid (699) Heia in Figure 3.
The lower part of the figure shows that there is two possible so-
lutions with radically different frequencies. However, the second
frequency of 14.13 cycles/day provides only one local maxima
and hence its period should be multiplied by 2, yielding the same
fit to the data.
Finally, we did not just provide the chosen solution, but all
the possible solutions. Sometimes there can be two or three pos-
sible solutions (periods for which σ is not significantly higher).
These periods are also possible and we report their values. For
some asteroids, the data are not accurate enough to obtain reli-
able results and the range of possible rotational periods ends up
including the entire possible interval: from 2 h to 360 h. This in-
terval helps us understand the quality and reliability of the result.
In the case of asteroid 2014 AX12 (Figure 3), the uncertainty
in the photometry is so high that almost any period can be fit-
ted. The uncertainty interval for this asteroid is [2 h, 237 h]. This
basically means that the result should not be considered, even
though there is a formal solution associated with the minimal
value of σ.
6. Discussion
We present in Figure 8 how periods derived here are distributed
against the diameter measurements. We also present data com-
piled from the literature (retrieved from SsODNet, Berthier et al.,
2023). The limits imposed by TESS observations are clearly vis-
ible. First, the cadence of exposures (30 min) and length of ob-
servations (27 days) limit the range of periods that can be de-
termined (Section 3), between 1 h and 27 days. Second, there is
a clear drop in the amount of solutions below a diameter of 2–
3 km. This diameter roughly corresponds to an apparent magni-
tude of 18 in the asteroid belt, which is indeed the peak in the
magnitude distribution in TSSYS-DR1 data set (Pál et al., 2020).
This distribution, combined with the trend for faintest asteroids
to have the highest rejection rate (Figure 5), explains the drop of
solutions below 2–3 km in diameter.
The lower limit of 1 h is not reached by any of our solu-
tions. As also visible in the data from the literature, there is a
clear boundary at 2.2 h which is refereed to as the “spin bar-
rier” and corresponds to the critical rotation period at which self
gravity and centrifugal acceleration are balanced (Pravec & Har-
ris, 2000). The situation is different for the upper limit, with
longer rotation periods reported in the literature from long-term
campaigns of observation or archival data (e.g., Waszczak et al.,
2015; Erasmus et al., 2021; Marciniak et al., 2021). We note the
presence of a valley around 50–100 h separating the bulk of fast
rotators from the slower ones, with an apparent dependence on
diameter. This low-density region was already visible in the ro-
tation periods determined from K2 by Kalup et al. (2021) and
from Gaia by ˇDurech & Hanuš (2023).
We present in Figure 9 the distribution of the amplitude of
the light curves as a function of the asteroid diameter. We also
present the asteroids reported in the LCDB (Warner et al., 2009)
for comparison. Most asteroids have a diameter between 3 and
20 km. There is a clear trend of higher light-curve amplitudes to-
ward smaller diameters, revealed by the running average. Such a
trend was already visible in the results of the Palomar Transient
Factory on asteroids (Waszczak et al., 2015). The amplitude of
the light curve is directly related to the change of the projected
area of the shape on the plane of the sky. Smaller amplitudes are
thus indicative of rounder shapes. The trend here is a clear signa-
ture of asteroids being less spherical at smaller diameters. Such
a trend has been reported from the results of 3D shape modeling
of the 42 of the largest (diameter above 100 km) main belt as-
teroids, with an increase of asphericity toward smaller diameters
Article number, page 6 of 8
D. E. Vavilov and B. Carry: Rotation periods of asteroids from light curves of TESS
data
10−1 100 101 102 103
Diameter / km
10−1
100
101
102
103
Period / h
1h
14d
Literature (33496) Valid (4829)
Fig. 8. Distribution of the rotation period as function of the asteroid
diameter, compared with the literature. The two horizontal dashed lines
represent the minimum and maximum periods that can be determined
here (Sect. 3Selection of the best modelsection.3).
100 101 102
Diameter (km)
0.0
0.2
0.4
0.6
0.8
1.0
Light curve amplitude (mag) LCDB
Valid
Running mean
Fig. 9. Distribution of the light-curve amplitude as a function of asteroid
diameter. The asteroids in LCDB (Warner et al., 2009) are plotted for
reference, as well as a running mean (red line) and standard deviation
(shaded area).
(Vernazza et al., 2021). The regular increase in amplitude below
the 100 km reveals that the trend continues down to diameters as
small as 2 km (size at which the present data set is limited).
We compare the light-curve amplitude to the rotation period
in Figure 10, focusing on the shortest rotation periods. The pe-
riod distribution is not random, with a limit at the shortest peri-
ods dependent on the amplitude of the light curve. As a guide-
line, we present the theoretically largest possible amplitude for
bodies held together by self-gravity only (taken from, Waszczak
et al., 2015) for three reference densities: 1000, 2000, and 3000
kg·m−3. There is a clear difference between asteroids in a broad
“S” complex (including the S, Q, A, V, E classes) and those in a
broad “C” complex (C, Ch, B, D, P, Z classes, see Mahlke et al.,
2022, for a rationale on this grouping), as expected from the
higher density of the first group. The fraction of S-like asteroids
is higher below the spin limit of 1000 kg·m−3, revealing the lower
density of C-like asteroids (below or slightly above 1000 kg·m−3)
compared to S-like asteroids, closer to 2000 kg·m−3.
2345678
Rotation period (h)
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Lightcurve amplitude (mag)
63 ± 1% 79 ± 2% 70 ± 7%
Complex
All
S (1470)
C (792)
Density
3 kg.m−3
2 kg.m−3
1 kg.m−3
Density
3 kg.m−3
2 kg.m−3
1 kg.m−3
Fig. 10. Distribution of amplitude against rotation period. The gray lines
represent the fastest spin for bodies held together solely by self-gravity
from Waszczak et al. (2015). The fraction of S-types among all asteroids
is indicated between those curves.
7. Conclusion
We present a new approach of determining the rotational period
of asteroids from optical light curves. The key aspects of the
approach are:
– Fit of Fourier series to the data by the least-squares method.
– Selection of the model with the lowest weighted root-mean
square residuals.
– Selection of the lowest possible order of Fourier series, fol-
lowing the principle of "the less the better.”
– Check with the F-test if higher orders Fourier series are nec-
essary.
– Test the number of local maxima of the function. Multiply
the period by two if only one local maxima is present (we
assume that the shape mostly produces the light curve)
– Report the determined rotation period with its uncertainties.
This includes the confidence interval of the period and the
possible alternative rotational periods (ambiguous solutions
with similar residuals, generally alises of the solution). If
there are too many alternative periods, an interval encom-
passing all these periods is reported.
This approach was used to compute the rotational periods
of asteroids observed by Transiting Exoplanet Survey Satellite
(TESS) reported by Pál et al. (2020). This dataset has observa-
tions of 9912 asteroids. We determined the period of 4521 aster-
oids with an accuracy better than 10%.
Comparison of our results with (Pál et al., 2020) shows
78.5% of a full agreement. For some of the asteroids, we iden-
tified the advantage of our technique, in particular, when fitting
higher order of Fourier series and checking for the number of
local maxima. The method we propose here is robust and can be
applied to any dataset of dense light curves of asteroids. Thus,
we plan to use it on available asteroid datasets in a future work.
Acknowledgements. B. Carry acknowledges support by the French ANR, project
T-ERC SolidRock (ANR-20-ERC8-0003). This paper includes data collected by
the TESS mission. Funding for the TESS mission is provided by the NASA Ex-
plorer Program. We did an extensive use of the VO tools TOPCAT (Taylor, 2005)
and SsODNet (Berthier et al., 2023). Thanks to all the developers and maintain-
ers.
References
Berthier, J., Carry, B., Mahlke, M., & Normand, J. 2023, A&A, 671, A151
Article number, page 7 of 8
A&A proofs: manuscript no. main
Berthier, J., Carry, B., Vachier, F., Eggl, S., & Santerne, A. 2016, MNRAS, 458,
3394
Bottke, W. F., Vokrouhlický, D., Broz, M., Nesvorný, D., & Morbidelli, A. 2001,
Science, 294, 1693
Bowell, E., Hapke, B., Domingue, D., et al. 1989, in Asteroids II, ed. R. P.
Binzel,
T. Gehrels, & M. S. Matthews, 524–556
Cellino, A., Hestroffer, D., Lu, X. P., Muinonen, K., & Tanga, P. 2019, A&A,
631, A67
Clement, M. S., Morbidelli, A., Raymond, S. N., & Kaib, N. A. 2020, MNRAS,
492, L56
De Angelis, G. 1995, Planet. Space Sci., 43, 649
DeMeo, F. E. & Carry, B. 2014, Nature, 505, 629
ˇDurech, J., Carry, B., Delbo, M., Kaasalainen, M., & Viikinkoski, M. 2015, As-
teroid Models from Multiple Data Sources (Univ. Arizona Press), 183–202
Erasmus, N., Kramer, D., McNeill, A., et al. 2021, MNRAS, 506, 3872
Eyer, L. & Bartholdi, P. 1999, A&AS, 135, 1
Grice, J., Snodgrass, C., Green, S., Parley, N., & Carry, B. 2017, Asteroids,
Comets, and Meteors: ACM 2017
Gubanov, V. C. 1997, Obobshchennyj metod naimen’shih kvadratov. Teoriya i
primeneniya v astrometrii. (in Russian. Generalized method of least squares.
Theory and application in astrometry.) (Nauka)
Hirayama, K. 1918, AJ, 31, 185
Kaasalainen, M. 2004, A&A, 422, L39
Kaasalainen, M., Torppa, J., & Muinonen, K. 2001, Icarus, 153, 37
Kalup, C. E., Molnár, L., Kiss, C., et al. 2021, ApJS, 254, 7
Linnik, Y. V. 1961, Method of Least Squares and Principles of Theory of Obser-
vations (translated from Russian by R. C. Elandt), ed. N. L. Johnson (Perga-
mon Press)
Mahlke, M., Carry, B., & Denneau, L. 2021, Icarus, 354, 114094
Mahlke, M., Carry, B., & Mattei, P. A. 2022, A&A, 665, A26
Marciniak, A., ˇDurech, J., Alí-Lagoa, V., et al. 2021, A&A, 654, A87
Masiero, J. R., Mainzer, A. K., Grav, T., et al. 2011, ApJ, 741, 68
McNeill, A., Gowanlock, M., Mommert, M., et al. 2023, AJ, 166, 152
Milani, A., Cellino, A., Kneževi´c, Z., et al. 2014, Icarus, 239, 46
Morbidelli, A., Walsh, K. J., O’Brien, D. P., Minton, D. A., & Bottke, W. F. 2015,
The Dynamical Evolution of the Asteroid Belt, ed. P. Michel, F. DeMeo, &
W. F. Bottke, 493–507
Muinonen, K., Belskaya, I. N., Cellino, A., et al. 2010, Icarus, 209, 542
Pál, A., Szakáts, R., Kiss, C., et al. 2020, ApJS, 247, 26
Pravec, P. & Harris, A. W. 2000, Icarus, 148, 12
Scheirich, P. & Pravec, P. 2009, Icarus, 200, 531
Taylor, M. B. 2005, in Astronomical Society of the Pacific Conference Se-
ries, Vol. 347, Astronomical Data Analysis Software and Systems XIV, ed.
P. Shopbell, M. Britton, & R. Ebert, 29
Tedesco, E. F., Noah, P. V., Noah, M. C., & Price, S. D. 2002, AJ, 123, 1056
Torppa, J., Kaasalainen, M., Michałowski, T., et al. 2003, Icarus, 164, 346
Usui, F., Kuroda, D., Müller, T. G., et al. 2011, Publications of the Astronomical
Society of Japan, 63, 1117
ˇDurech, J. & Hanuš, J. 2023, A&A, 675, A24
Vernazza, P., Ferrais, M., Jorda, L., et al. 2021, A&A, 654, A56
Vokrouhlický, D., Bottke, W. F., Chesley, S. R., Scheeres, D. J., & Statler, T. S.
2015, The Yarkovsky and YORP Effects, ed. P. Michel, F. DeMeo, & W. F.
Bottke, 509–531
Warner, B. D., Harris, A. W., & Pravec, P. 2009, Icarus, 202, 134
Waszczak, A., Chang, C.-K., Ofek, E. O., et al. 2015, AJ, 150, 75
Zappala, V., Cellino, A., Farinella, P., & Kneževi´c, Z. 1990, AJ, 100, 2030