Draft version October 20, 2017 Preprint typeset using LATEX style AASTeX6 v. 1.0

HEK VI: ON THE DEARTH OF GALILEAN ANALOGS IN KEPLER, AND THE EXOMOON CANDIDATE KEPLER-1625B I

A. Teachey1, D. M. Kipping1 & A. R. Schmitt2 [email protected]

1Department of Astronomy, Columbia University, 550 W 120th St., New York, NY 10027 2Citizen Science

ABSTRACT Exomoons represent an outstanding challenge in modern astronomy, with the potential to provide rich insights into planet formation theory and habitability. In this work, we stack the phase-folded transits of 284 viable moon hosting Kepler planetary candidates, in order to search for satellites. These planets range from Earth-to-Jupiter sized and from 0.1-to-1.0 AU in separation - so-called “warm” ∼ planets. Our data processing includes two-pass harmonic detrending, transit timing variations, model selection and careful data quality vetting to produce a grand light curve with a r.m.s. of 5.1 ppm. We find that the occurrence rate of Galilean-analog moon systems for planets orbiting between 0.1 ∼ and 1.0 AU can be constrained to be η < 0.38 to 95% confidence for the 284 KOIs considered, with +0.13 a 68.3% confidence interval of η = 0.16−0.10. A single-moon model of variable size and separation locates a slight preference for a population of short-period moons with radii 0.5 R⊕ orbiting at 5-10 ∼ planetary radii. However, we stress that the low Bayes factor of just 2 in this region means it should be treated as no more than a hint at this time. Splitting our data into various physically-motivated subsets reveals no strong signal. The dearth of Galilean-analogs around warm planets places the first strong constraint on exomoon formation models to date. Finally, we report evidence for an exomoon candidate Kepler-1625b I, which we briefly describe ahead of scheduled observations of the target with the Hubble Space Telescope. Keywords: planetary systems — techniques: photometric

1. INTRODUCTION uproot conventional thinking about satellite formation Moons present unique scientific opportunities. In our mechanisms. Galilean-sized moons ( 0.2-0.4 R⊕) are generally Solar System, they offer clues to the mechanisms driving ∼ early and late planet formation, and several of them are thought to be able to form in a variety of ways. For thought to be promising targets in the search for life, the regular satellites of Jupiter, the Galilean moons are as several are rich in volatiles (e.g. Squyres et al. 1983; thought to have condensed out of a circumplanetary Hansen et al. 2006) and possess internal heating mech- disk, akin to planet formation within a protoplanetary anisms (e.g. Morabito et al. 1979; Hansen et al. 2005; disk (Canup & Ward 2002). This process is expected to limit regular satellites to a cumulative mass of [10−4] Sparks et al. 2016). The moons of our Solar System O arXiv:1707.08563v2 [astro-ph.EP] 18 Oct 2017 also demonstrate the great variety of geological features that of the primary (Canup & Ward 2006). Higher mass- that may be found on other terrestrial worlds. ratio moons, such as the Earth’s Moon, are evidently In this new era of exoplanetary science it stands to viable too and may form from catastrophic collisions reason that moons in extrasolar systems, so-called exo- in the first few hundred million years of the solar sys- moons, should tell us a great deal about the common- tem, coalescing from that collision’s debris (e.g. Ida et ality of the processes that shaped our Solar System and al. 1997). Finally, retrograde Triton is hypothesized to may yield just as many surprises as their host planets have originated from a capture event via a binary ex- before them. Just as the study of exoplanets has com- change mechanism (Agnor & Hamilton 2006). Put to- plicated our picture of planetary formation by revealing gether, Galilean-sized satellites appear to have formed (for example) the existence of Hot Jupiters (Mayor & via at least three independent pathways within the So- Queloz 1995) – worlds without Solar System analogs – lar System, and their existence around exoplanets can so too might moons show us what else is possible and therefore be reasonably hypothesized. 2

Galilean-sized exomoons are challenging to detect us- amount of data before and after stacking is the same, ing the transit method1 for a number of reasons. First, with the only difference being that stacking assumes the the transit of a 0.2-0.4 R⊕ moon across a Sun-like star ephemeris of the planet is known to infinite precision. results in a depth of 3-13 ppm, below the typical sensi- Modeling the full unstacked light curve with a model tivity achievable with Kepler (Christiansen et al. 2012). conditioned upon the same ephemeris assumption would Second, the moon signal will almost certainly be found result in identical posteriors and thus no improvement at each epoch in a different location with respect to is achieved for the inference itself. Despite this, stacking the host planet, sometimes occurring before the tran- is attractive because the signal’s coherence means that sit, sometimes after, and at a different projected dis- full light curve modeling is unnecessary in the context of tance from the planet (Kipping 2011). Third, multi- signal detection. Specifically, one may simply evaluate ple moons around a single planet may wash out any the weighted mean centered around the pivot point of transit timing (Sartoretti & Schneider 1999) or dura- the fold to achieve a detection, which is why the pop- tion variations (Kipping 2009a,b). The three-body mo- ular Box Least Squares (BLS) algorithm (Kov´acset al. tion combined with the potential for overlapping disks 2002) is a computationally efficient yet sensitive tool in (syzygies), makes accurate modeling of exomoon transits conventional transit detection. computationally demanding (Kipping 2011). Fourth, Stacking light curves in pursuit of exomoons is compli- the long-period nature of plausible moon hosts means cated by the fact that simply phase-folding light curves that relatively few transits are usually available. In this upon a linear ephemeris will lead to the moon appearing regime, a planet+moon light curve model has sufficient at different phases in each epoch. Despite the fact that flexibility to almost always provide a superior fit to the the moon is not perfectly coherent, it is constrained to limited data (Kipping 2013), thereby necessitating rig- lie within a fraction of the Hill sphere radius (Barnes orous Bayesian approaches to model selection. Despite & O’Brien 2002) and this imparts some quasi-coherent these challenges, the “Hunt for Exomooons with Kepler” properties into the phase-folded light curve. Simon et al. (HEK) project has performed Bayesian photodynami- (2012) were the first to describe this possibility, where cal fits of 60 exoplanets to date (Kipping et al. 2012, they argued that this quasi-coherence will lead to an in- ∼ 2013a,b, 2014, 2015a), with no unambiguous detections crease in the photometric scatter in the temporal region and limits typically hovering around an Earth-mass. surrounding the planetary event - an effect they dubbed In this work, our project pursues a different approach “scatter peak”. A similar idea is discussed in Heller to searching for evidence of moons which focuses on (2014), who instead considered looking for a slight pho- seeking a population of moons around a population of tometric decrease in this temporal region. By consider- planets. Rather than pursuing individual limits which ing the probability density of the moon’s sky-projected are then combined to constrain the population, we here position, an effect dubbed the “orbital sampling effect” approach the problem from a broad statistical perspec- (OSE), Heller(2014) derives a simple formula for pre- tive in order to directly measure the occurrence rate of dicting phase-folded light curve shape in the presence of moons, η . Resulting from this analysis, we identify moons, enabling a simple approach to seeking exomoon a single exomoon$ candidate, Kepler-1625b I. We briefly shadows. describe its detection and vetting ahead of scheduled ob- As with the case of a planetary transit, or indeed any servations of the planet with the Hubble Space Telescope kind of stacking, this approach does not boost SNR in in October 2017. any way. The data volume and quality are the same be- fore and after the stacking. However, unlike the planet 2. STACKING EXOMOONS case, the shape of the phase-folded moon signal repre- 2.1. Phase-Folding sents a washed-out depiction of the individual signals. Accordingly, the subtle individual variations in dura- The work presented here aims to exploit the power of tions, positions and shapes are lost, meaning that stack- stacking in order to search for exomoons. Stacking is a ing imposes a fundamental loss of information content, familiar technique to those studying exoplanet transits, and therefore sensitivity. who typically phase-fold a light curve upon the period. A similar situation occurs when observing planetary For a linear ephemeris, the transits align leading to a transits with long exposures, such as the 30 minute ca- coherent signal. It is important to stress that this act dence (LC) mode of Kepler, causing the shape of the does not improve the signal-to-noise ratio (SNR). The transit to be slightly washed-out, thereby degrading the information content. It is for this reason that short- cadence (SC) Kepler data provides tighter constraints 1 Note that the transit method is by no means the only method sensitive to exomoons; microlensing, for example, is an- on transit times, despite the fact that both see the same other promising avenue (Bennett et al. 2014). SNR transit depth (e.g. see Kipping et al. 2013a and 3

Kipping et al. 2014). ensemble phase folded signals, which in their case was Accordingly, searching for exomoons in phase-folded used to boost the overall SNR. In many ways, this ap- light curves will always be less sensitive than full photo- proach is reminiscent of a hierarchical Bayesian model dynamic fits – although precisely how much has not been (HBM; Hogg et al. 2010) but by stacking the objects formally evaluated and would be sensitive to the specific the identities of each object are sacrificed. While an planet-moon parameters. Despite this, phase-folding is HBM approach would be better suited, in general, di- attractive for its simplicity and as a pure detection tool rect planet stacking is attractive for its simplicity, par- – analogous to BLS for planet hunting. ticularly if the objective is purely to test whether an One particularly attractive feature of phase-folding ex- ensemble signal even exists rather than attempting to omoons is that a sizable fraction of the quasi-coherent perform detailed characterization of said signal. There- signal appears exterior to the planetary transit. As- fore, in the same vein, we decided to try stacking dif- suming the planetary transits are well-aligned and the ferent phase-folded planet signals together, to solve the duration is well-known, one may simply crop the plane- N 1 problem. We highlight that Hippke(2015) inde-  tary transit leaving behind a pure exomoon signal. This pendently arrived at the same idea and published before greatly simplifies the analysis, since the planet proper- our effort, although many differences exist in our actual ties are not covariant with this signal2. For these rea- implementations, as will become clear throughout this sons, we identify this out-of-transit phase-folded moon paper. signal as the target signature in this work. Co-adding different planets decreases the overall noise, since we are extending the data volume upon which our 2.2. Planet-Stacking inferences are conditioned. However, this approach is not guaranteed to increase the SNR, since many of the Unlike a simple planetary phase-fold, the quasi- objects co-added may not even possess moons and thus coherent nature of the phase-folded moon light curve their inclusion only dilutes the overall signal, rather than means that a large number of transits are needed to co-adding to it. Nevertheless, we can quantify the over- produce a predictable signal. At its core, the phase- all signal amplitude as being a combination of the oc- folded moon signal depends upon the law of averages currence rate and the moon radii. Even so, selecting a and thus co-adding relatively few transits can lead to a sample of planets which are expected, a priori, to be phase-folded moon signal which is highly irregular and plausible hosts for large moons will be crucial for max- erratic. Without a characteristic and predictable shape, imizing our chances of a successful detection and corre- it is very difficult to convincingly argue the signal is gen- spondingly deriving meaningful, physically-constraining uinely a moon, rather than some peculiarity of the data upper limits. in those limited co-added events. Indeed, Heller(2014) argue that at least a dozen events are generally needed, such that N 1 and the averaging effect can become 3. TARGET SELECTION  noticeable. 3.1. Automated Target Selection Unfortunately, Kepler’s primary mission lifetime of Not all exoplanets are equally likely, a priori, to yield 4.35 years means that the long-period planets, where an exomoon detection. At the most basic level, two moons are most a priori expected to be viable, were only questions guide our target selection process: 1) what is observed to transit a few times. Only in a small number the largest stable moon plausible around a given planet of cases are there Kepler planets for which their period 2) would this moon be detectable, given the current data is long enough such that moons are dynamically stable in hand? for Gyr and we possess N 1 transit events within the  In this work, we estimate whether a detectable moon is 4.35 years of Kepler observations. This point seemingly plausible following a similar approach to that adopted excludes phase-folding as a viable exomoon approach, in previous HEK papers. Specifically, we employ the except for a few rare cases. Target Selection Automatic (TSA) algorithm described We devised an approach to solve this problem, in- in Kipping et al.(2013a). spired by the work of Sheets & Deming(2014). In that To summarize, the algorithm first estimates a mass work, the authors not only phase-folded each planetary for each Kepler Object of Interest (KOI) using the maxi- light curve but also co-added different planets together. mum likelihood radius reported on the NASA Exoplanet This allowed them to greatly increase the number of Archive(Akeson et al. 2013) and the mass-radius rela- tion defined in Kipping et al.(2013a). This is used to further estimate the extent of the planet’s Hill sphere. 2 Phase variations are not included in our model and thus can- not be covariant, nor should such phase variations persist given Moons are expected to have their lifetime limited by our detrending algorithm will have largely removed them. the time it takes to tidally spin out from the Roche 4

ferential photometric precision (CDPP) of each host star -3.0 ☾☾ (Christiansen et al. 2012), which along with the maxi- 10 ⊕⊕ mum moon radius, allows us to estimate the signal-to- 10-3.5 noise ratio (SNR) expected due to the moon. ♆♆ ♄♄ For the SNR calculation, we estimate the phase- -4.0 / Q 10 averaged signal amplitude using the so-called “orbital 2 k sampling effect” (OSE) described in Heller(2014). 10-4.5 While these expressions formally assume a large num- ber of transits, which is rarely true, they work well as -5.0 ♃♃ 10 log10(k2/Q)=-2.90-0.187(RP/R⊕) an approximation for the signal strength marginalized 0 2 4 6 8 10 12 over the moon’s phase, which is of course unknown to us a priori. The expected OSE flux decrease for the R [R ] P ⊕ out-of-transit data is given in Heller(2014), from which Figure 1. Relationship between the tidal property k2/Q we may integrate over the signal to compute the signal and planet size for the Solar System bodies. When excluding strength of the out-of-transit portion to be Saturn, the remaining four points closely follow a power-law used in this work. 2 p 2 −1 RS  (aSP /RP ) 4 2 cos (2RP /aSP ) S = − − . R? π((aSP /RP ) 2) limit to some critical fraction of the Hill sphere, fR . − H (1) Using the expressions of Barnes & O’Brien(2002), this logic may be inverted to compute the maximum allowed The fraction on the right-hand side varies from about moon mass, MS,max, which can survive for a fiducial a quarter to a third for (aSP /RP ) in the range of 5 to age of t? = 5 Gyr. For TSA, we set f = 0.9309 for the 100 i.e. it is a relatively gentle function. We therefore optimistic case of a retrograde moon (Domingos et al. take the limit of large (aSP /RP ), giving 2006). Note that we assume a single moon for stability estimates and tidal evolution timescales; the presence of 1 R 2 S = S . (2) multiple moons is expected to modify these values. How- π R? ever, since we do not know a priori how many moons The SNR may now be calculated by dividing this by may be present in a given system, it is impossible and the noise expected impractical to apply more sophisticated stability esti- mates at this stage. 2 p r M 1 RS  THill/0.25 B In order to compute S,max, we must adopt a value SNR = (3) for the ratio (k2/Q), which represents the efficiency of π R? CDPP6 P tidal dissipation. Whereas in previous HEK papers we where B is the time baseline of observations, optimisti- 5 simply adopted (k2/Q) = 0.5/10 , here we use an em- cally assumed to be the full Q1-17 baseline for these cal- pirical relation based on the Solar System. As shown culations and CDPP6 is the combined differential photo- in Figure1, empirical estimates of ( k2/Q) for the Moon metric precision on a 6 hour timescale. We may express (Dickey et al. 1994), the Earth (Kozai 1968), Neptune the Hill time, assuming a simple circular orbit approxi- (Trafton 1974), Jupiter (Lainey et al. 2009) and Saturn mation, as (Lainey et al. 2012) follow a power-law distribution ver- sus planetary radii, except for Saturn. In this work, we 1/3  P  MP  invert this empirical relation (ignoring Saturn) to esti- THill = f , (4) 2π 3M? mate (k2/Q) for a given RP . While we don’t claim this to be a fundamentally general law, it at least provides which when substituted in leads to the P terms can- a somewhat more reasonable estimate than the blanket celling out, such that fixed value assumed previously. " # Once the maximum moon mass has been computed, (R /R )2 √2Bf  M 1/6 S P P . it is then converted into a moon radius using the same SNR = 1/6 (5) π 3 √πCDPP6 3M? mass-radius relation as before. Since target selection was conducted early on in the two-year duration of this As discussed later in Section6, we find that the OSE research comprising this paper, it predated the more ro- model overestimates the signal-to-noise for large aSP , bust probabilistic mass-radius relation of Chen & Kip- with numerical experiments showing it is around a fac- ping(2017a), which is why that latter relation was not tor of three too high by the time we hit (aSP /RP ) = 100. used for these calculations. We query the combined dif- We therefore correct the SNR quoted above by dividing 5 by a factor of 3, which together with by the π de- 4. two-pass data processing. ∼ nominator we simply approximate to a factor of 10 ∼ denominator, yielding We explain and discuss these requirements in what follows.

2 " 1/6# (RS/RP ) √2Bf  MP  TTVs SNR . (6) 4.2. 1/6 π M ' 10 3 √ CDPP6 3 ? In order to create an accurate phase-folded light curve of a sequence of planetary transits, it is necessary to en- 3.2. Applying to the Kepler Planetary Candidates sure the transits accurately phase up. In the absence of TSA was first run for this project in November 2014, transit timing variations (TTVs), this is straightforward at which time 7305 KOIs were listed on the NASA Exo- and is a simple linear ephemeris fold. However, the sig- planet Archive. However, 27 were removed due to hav- nal we seek, an exomoon, will always introduce a small ing some incomplete column entries. Of these, 4109 were TTV signal into the data (Sartoretti & Schneider 1999). not classified as a “false-positive” by the NASA Exo- Moreover, TTVs can be caused by other effects, notably planet Archive and thus were considered further. In perturbing planets (Holman & Murray 2005; Agol et al. order to calculate SNR, basic stellar properties are re- 2005) and thus TTVs are observed to be fairly com- quired and so we cross-referenced this list with the Hu- mon (&10%) in Kepler planetary systems (Holczer et ber et al.(2014) catalog, in which we were unable to find al. 2016). Carefully removing these TTVs is crucial in a match for 11 KOIs. They were thus removed giving us creating an accurate phase-folded transit signal. a total of 4098 KOIs which were then put through the One approach might be to take the catalog of known TSA algorithm. TTVs from Holczer et al.(2016) and use these for cor- Due to the ensemble nature of our analysis, the total rections. There are several reasons why this is unsat- SNR is expected to be much greater than that of individ- isfactory for this work though. First, in order to as- ual objects and thus we can afford to use a relatively gen- sess robust confidence limits, we require covariant, joint erous SNR cut. Accordingly, we elected to use SNR> 0.1 posteriors distributions of the transit times and basic and apply the criteria that P < B/4 = 397.39 days (to planet parameters, which were not derived in Holczer et give three transits), leading to a sample of 966 KOIs al.(2016). Second, whenever possible, accurate phase outputted from the TSA procedure. These targets are stacking is aided by first conducting model selection be- visualized in Figure2. tween the TTV and linear ephemeris models, which itself Since this work has taken several years to complete, formally requires computation of the Bayesian evidence some of the objects that were considered exoplanet - again something not derived in Holczer et al.(2016). candidates when we began are now considered false- Third, the TTVs derived in Holczer et al.(2016) were positives. In December 2015, we elected to remove all conditioned on a different data set to that used in this KOIs if they were either classified as “false-positives” at work. More specifically, although both Holczer et al. NEA or the probability of any false-positive scenario was (2016) and this work are based on Kepler photometry, in excess of 1%, as given by the values listed on NEA. our data detrending methods are distinct meaning that This filter removed 292 ( 30%) of the objects origi- ∼ these differences should be expected to affect the TTV nally considered, reducing the number of usable KOIs measurements to some degree. When one is ultimately from 966 to 674. seeking the discovery of a few parts-per-million signal, these conventionally minor issues cannot be ignored and 4. DATA PROCESSING REQUIREMENTS should be expected to influence the results. 4.1. Overview For these reasons, we concluded that creating an ac- The objective of this work is to create a phase-folded curate grand light curve was not possible without first planet-stacked out-of-transit light curve, which may be deriving TTV posteriors ourselves for every system con- used to search for evidence of exomoons. For the sake sidered. of brevity, we will refer to this light curve as the grand light curve, or simply GLC, in what follows. 4.3. Data Quality We identify four unique and critical requirements for There is a unique property of the phase-folded moon realizing this objective, specifically: signal that has strong implications for the data quality requirements, which is not conventionally an issue for 1. removal of TTVs, planet analyses. The GLC signal is a phase-fold of the 2. very high quality light curves, planetary transit, after removing TTVs, and thus at any given instant in phase, the moon actually only induces 3. temporal rescaling, a transit-dip for some fraction, F , of the co-added light 6

● ● ● ● ● ● ● ● 16 ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ●● ●●● ●● ● ● ● ● 25 ● ●● ●● ● ● ● ●○ ● ●○●● ● ] ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●○ ● ● ●○ ● ● ●

⊕ ● ● ● ● ● ○●○● ●● ●●●●● ○● ● ● ●● ●●● ●● ●●●● ●● ● ●●○● ● ●○ ● ● ●●●○ ● ● ● ● ●●● ● ○● ● ● ● ● ● R ● ● ● ● ●● ● 8 ●● ● ●● ● ● ● ● ● ○● ● ●● ○● ● ● [ ● ●● ● ● ●● ● ●○ ●●○● ● ● ● ●○● ● ● ●● ● ● ●● ○● ○● ● ● ● ●○○● ○● ● ● ●● ● ●● ● ● ● ●●○ ● ● ● ● ●● ●● ●●● ● ● ● ●○●○● ● ● P ● ● ● ●● ●○ ●○ ●● ● ● ● ● ● ● ●● ●●● ● ● ● ● ●●● ● ○● ● ●●○ ● ● ● ● ● ●● ●○ ●● ○●○● ○● log(SNR) ● ● ●●● ● ●●●●● ●● ● ● ●● ● ● ●● ●● ● ● ● ●● R ● ● ● ● ●●● ● ●●● ●● ● ●●● ●● ● ● ● ●○● ●○●● ● ● ● ●●●● ● ●● ●●● ● ● ●● ● ● ● ●● ● ● ●● ● ● ●●● ● ●●● ●● ● ● ● ●●● ●●● ●●○ ●○ ● ● ● ● ●●● ●● ● ●●●●●●●● ●● ●● ● ● ●●●● ● ● ● ● ● ● ●●●● ● ●● ●●● ●●●●● ●● ○● ●●○●●●○ ● ● ● ● ● 4 ● ●● ●●● ● ●●● ●●● ●●●● ●● ●●●●● ●○● ○●●○●● ○● ●○● ●●○○● ● ● ● ● ● ● ● ●●●● ●●●● ●● ● ●○●●○●●●● ●●○○● ●○●● ○●●○ ●●● ●● ● ● ●●● ●●●●● ●●● ●●●●●●●●●●●● ●● ●●● ○●●○○●○●○● ●● ○●●○ ●○○●●○●○●○●●●●● ●●● ●● ● ● ●● ● ● ●●●●●●●●●●●●●● ●● ●●●●●●●● ●○●●●●○●●○●●○●●○●○●●● ●●○●●●● ●○ ○●○● ● ●○ ●●○● ● ● ●●●● ●●●● ● ●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●●●●●○●○●●●○●○●●○●● ○●○● ●○●●●○ ●○●●●●○●● ●●●●● ● ● ●●●● ●●●●● ●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●○●●●●●○●●○●○●●○●○●●○ ● ●○●●●○●● ●●○ ○●○●●●○● ●○●○●○● ●○● ●●● ●●● ● ● ●●●●●● ●●●●●●●●●●●●●●●●●○●●●●●●●●●●○●●●●●●○●○●○●●○●○●○●●○○●●○●●○● ●●○●●●○●●● ●●○●○●●●○ ● ●○○●●● ●● ●● ● ●●● ●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●○●●○●●●○●●○●○●●○●●●●○●●○●● ○●○●●○ ●○●●○● ○● ●●○●●○●● ○●●●●● ● ● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●○●○●○●●○○●●○●○○●●●○●○●●○●● ● ●●●○●● ●●● ●●●● ●●●○● ●●● ● ●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●○●●●●●●○●○●●○●●○●○●●○●●○ ●●○○●●●●●○●●○● ●○●●○● ●●● ● ● ● ●● ●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●○●●●○●●○●●○●●●●●●○●●●●● ●●○●●○● ○●●○●○○●●○●●● ○●●●○●●○●● ○● ●● ● ● ● ●● ●● ●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●●●●●●●●●○●●●●○●●●●○●●●●○●○●●○●○●○●●○●○●●●●○● ● ○●●●● ●●● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●●○●●●●●●●●○●●○●●●○●●●○●●●●●●●●●●● ● ● ●○●●●●● ● ● ●● 2 ● ● ●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●○●○●●●○○●●●●○● ●●●●○● ●○●●● ●●●● ●● ●●○● ● ●● ● ● ● ●●●●●●● ● ● ●● ●● ●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○○●○●●●●●●●○●●●●●●● ● ● ● ●●●●● ●● ●● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●●●●●●○●●○●●●●●●●●●● ●● ●● ●○●●○● ●● ●●●● ● ●● ●●● ●●●● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●○●●●●●●○●●●●●●●●●●●○○●●●●● ● ●○●○● ● ●●● ●● ●● ●●● ●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●○●●●●●●●● ●●● ●●● ● ●● ○●●●● ●●● ●● ○● ●● ● ● ●● ●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●○●●●●●● ●○●●○● ●● ● ● ● ●● ●● ●● ● ●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●● ●●●● ● ●●● ●○ ●●● ●○ ●● ●● ●● ● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●○●● ●● ●●●●●●○ ●○●●●●● ● ●● ● ●● ● ● ● ●● ● ●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●● ●●●●●●●●○● ●●● ●●●● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●● ●●●●●●●●●●●● ●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●● ●● ●● ● ● ● ● ●● ●●● ●●●●●● ●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●● ●●● ● ● ● ● ● ● ● ●● ●● ●●●●● ● ●●●●●● ●●●●●●●●●●●●●● ●●● ●●●●●●●●● ●●● ●● ●●● ● ○● ● ● ● ● ● ●●●● ●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●● ● ● ● ● ● ● ● 1 ●● ●●●● ● ●● ●● ●● ●●●●●●●●●●●●●●● ●●● ●●● ● ●● ●● ● ● ● ● ●●● ● ● ●●● ●●●●●●●●●●●●●●●● ● ● ●●●●●● ● ● ●● ●●● ●●●●●●●●●●● ●●●●●●● ●●●●●●●●●●●● ● ●● ● ● ● ● ● ●● ●● ● ●● ●● ●● ●● ●●● ●●● ● ●●● ● ● planetary radius, ● ● ● ●● ●● ● ●●● ● ● ● ●● ●● ● ●●● ●●● ●●●●●●●●●●●● ● ●● ● ● ● ● ●● ● ● ●● ● ● ●●● ● ● ● ● ● ● ● ●● ●●● ●●● ●● ● ● ●● ● ●●● ●●● 1 ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● 0.5 ● ● ● ● ● 1 5 20 100 400 period, P [days]

Figure 2. Location of the KOIs selected to search for exomoons around. The 966 colored points, color-coded by the SNR given in Equation6, tend to be at relatively long orbital periods, where Hill spheres are larger. Points with a solid, black circle around it are the 347 KOIs found to pass our data quality vetting. curves. Geometry demands that this fraction must al- a single grand light curve. This is similar to the co- ways be less than one-half (i.e. F < 0.5) for all phase addition performed for occultations by Sheets & Deming points occurring outside of the planetary transit signal. (2014). In their case, each occultation has a distinct du- This is a key point which has a major implication: me- ration and thus simply co-adding the occultations would dian binning kills the GLC signal. cause the signal to smear out and produce an averaged This is extremely important, because median binning signal distorted from the true morphology. To overcome is a robust point estimate. The forgiving nature of me- this, Sheets & Deming(2014) re-scaled each event by the dian binning means that one can actually do a bad job known duration and then co-added, producing a more of detrending some small fraction of your light curves coherent signal. Just as with the occultations, each GLC (which represent outliers) yet still recover an accurate signal will have a different velocity and impact param- phase-folded signal. However, if one cannot use median eter and thus cause a different duration. However, the binning, then one is forced to use mean-based estimates problem is actually worse since unlike Sheets & Deming which are sensitive to each and every transit co-added. (2014), we don’t know what the true duration of each In this case, even a single inaccurately detrended transit event should be, since the duration is highly sensitive to light curve will contribute to the phase-stacked signal. the semi-major axis of the moon(s), which are of course Once again, since we seek the detection of a signal with not yet discovered. an amplitude of a few parts per million, this cannot be Ultimately, re-scaling will always be flawed since we ignored and demands the highest levels of scrutiny and can’t know the semi-major axis of the moon prior to dis- data quality. covering it. A full hierarchical Bayesian model (HBM) We therefore establish that each and every transit would be an appealing way of approaching this problem, used in our grand light curve must be verified to be of allowing each object to have a unique semi-major axis. very high data quality, which of course greatly increases However, since each planet would not satisfy N 1  the time demands needed to complete such an analysis. transits, the OSE approximation would break down and thus each system would require modeling with a full 4.4. Temporal Rescaling photodynamic simulation, such as that from LUNA (Kip- When we finally arrive at an accurate phase-folded ping 2011). For five years, we in the HEK project have light curve for each planet, they must be combined into 7 been conducting Bayesian regression of individual sys- servables directly to convert from time into a physically tems with LUNA, and the computational demands for motivated dependent variable via: even individual systems are formidable ( 30, 000 CPU ∼ hours per planet). Linking this into a full HBM would v "u ! # u 2π be computationally extremely challenging and was not t0 = tb2 + (a2 b2) sin2 (t τ) 1 /p. (8) a strategy we elected to pursue here. R − P − − Moreover, in this work, we ultimately hoped to find a signal which was visually evident in the final grand light If all of the moons shared the same (aSP /RP ), this curve and thus not conditional upon the inferences of would produce a coherent signal. In reality, we do not an HBM. While this does not maximize the information expect this statement to be true, but moons do appear content of the final data product, we are motivated to in the Solar System to be distributed log-uniformly with follow this philosophy on the basis that the discovery respect to this term (Kane et al. 2013). This approach of any novel phenomenon, which exomoons would rep- means that we could model the resulting grand light resent, requires a much higher confidence than routine curve assuming exomoons followed a formulaic distribu- discoveries (Gould et al. 2004). tion for (aSP /RP ), such as a log-uniform. In order to convert from t t0, we need estimates Accordingly, we seek a method of re-scaling which is → “least-bad” and will maximize the expected signal co- for the impact parameter and scaled semi-major axis. herence even when marginalizing out our ignorance of Since our data is not strictly the same as that used for the moon’s semi-major axis. One approach is that of the inferences quoted elsewhere, a self-consistent analy- Hippke(2015), who re-scaled by the duration of the sis demands we derive these estimates ourselves, which planetary transit. The advantage of this is mostly sim- forms another requirement for our work. plicity, the duration is well-constrained and easy to un- While our conversion equation assumes a circular or- derstand. One downside of this is that even if all the bit, if we fit the data under the same assumption, the moons had the same semi-major axes, they would still relative estimate is at least self-consistent. Further, ec- lead to the grand light curve having a smeared out OSE centric planets have smaller regions of stability for ex- signal, since each system has different barycentric ve- omoons (Domingos et al. 2006) and have likely experi- locity and impact parameter across the star. Another enced scattering which decreases the chances for moons approach, re-scaling by the Hill radii, is not possible further (Gong et al. 2013). Thus, if the planet is ec- since the exoplanet masses are unknown. centric, the incorrect conversion is likely irrelevant since Instead, in this work, we argue a better approach is such planets likely do not contribute OSE-signals into to re-scale the time axis into distance from the planet, the grand light curve anyway. in units of planetary radii. This can be accomplished by considering the original Seager & Mall´en-Ornelas(2003) 4.5. Two-Pass Detrending equation for the duration of a planet, under the assump- In this work, we use the CoFiAM algorithm to detrend tion of circular orbits: the Kepler light curves. We direct the reader to Kipping et al.(2013a) for details on the algorithm, but essentially s ! its goal is to remove long-term trends without introduc- P (1 p)2 b2 ing any power, in a Fourier sense, at periodicities less T 14 = sin−1 ± − . (7) 23 π a2 b2 than the transit duration. This requirement ensures that R − any signals with a time-scale approximately equal to or If we let (1 p) 1 in the above, we recover the less than this duration will not be distorted by the de- ± → transit duration as defined when the center of the planet trending process itself, since a transit can be thought of overlaps with the stellar limb, T˜ (Kipping 2010a). Thus, as a Fourier series with the lowest frequency being that at contact point 1 & 4, we can think of this instant in of the duration (Waldmann 2012). Accordingly, both time as when a shell of radius p centered on the planet the planet and moon transits are preserved, in contrast first starts to induce transit-dip features. By extension, to polynomial-based methods which introduce power at we could adapt (1 p) (1 + t0p) in the above, which all frequencies. ± → would equal the duration of a shell of radius t0p, cen- High frequency noise is not even attempted to be re- tered on the planet, to start/end creating transit-dip moved, but is monitored by measuring the autocorrela- features. In this way, we can think t0p as being the or- tion at the cadence-lag and used to optimize both the bital distance of the moon at the instant in time when harmonic filtering and subsequent identification of “bad the transit begins/ends. Accordingly, t0 represents the transits”, which are typically rejected. planet-moon distance in units of the planetary radius. A disadvantage of CoFiAM is the requirement for a pre- This convenient form allows us to use the transit ob- cise estimate of the time and duration of all transits in 8 the time series. For this reason, it generally is not useful can be achieved. For example, in cases where there is for blind searches for exoplanet transits. However, when plenty of out-of-transit data, but no in-transit, these seeking exomoons this requirement is generally true and would be rejected in pass A (but can be picked up later indeed CoFiAM should be really thought of as an exo- in pass B since such signals may still contain exomoon moon optimized detrending method more than anything transits). Ultimately, defining a clear and independent else. objective for each pass allows us to optimize required As before, transit times and durations are often avail- steps. able for the planets under consideration but those times In what follows, we describe the different stages of and durations were conditioned on a different data de- data analysis which are performed in pass A. trending. In order to make our analysis self-consistent, we must derive these times and durations ourselves. 5.2.2. STAGE A1 Here-in lies a chicken-and-egg problem though, since to The first step is simply to download the Kepler Sim- derive these times we first need detrended data, which ple Aperture Photometry (SAP) for each target. We itself first requires the times. To tackle this, we use wrote a shell script to step through the target list two passes to iterate onto the best solution. The first (see Section3), creating a local directory for each pass uses the literature values for the times and duration KOI. Using a wget script, any and all of the tar- of the transits and then performs CoFiAM. The second get’s LC data is then downloaded from MAST (from pass takes the times and durations inferred using the archive.stsci.edu/pub/kepler/lightcurves/) and saved in first-pass data product, and then performs a new round the appropriate directory. As the data were downloaded of detrending with CoFiAM. This approach ensures both in batches, the Kepler Science Operations Center data self-consistency and reliability in our estimates, and pro- processing pipeline used for these files varied from 9.0.3 vides several opportunities to vet the data quality ensur- (2013-04-18 creation date) to 9.2.23 (2014-11-18 creation ing only the highest quality light curves are used in the date). final analysis. Pre-prepared template detrending scripts are copied into each target directory. As with previous HEK pa- 5. DATA PROCESSING PIPELINE pers, we use CoFiAM to perform our detrending. We di- 5.1. Overview rect the reader to Kipping et al.(2013a) and Kipping et al.(2013b) for details on the algorithm. We stress here As motivated in Section4, we require a joint poste- that the algorithm is designed specifically for the exo- rior distribution for the times and basic transit param- moon problem and requires detailed initial information eters of all planets used for the final analysis. Using a such as planetary periods, transit times and durations two-pass approach to detrending-fitting ensures that our in order to work, which ultimately again explains why inferences are self-consistent and conditioned upon the the two-pass data processing strategy is used. actual data product used in this work. In this section, We use the Kepler SAP time series throughout, since we provide a detailed explanation of the data process- this time series has less chance of having unintended ing steps comprising each pass, which are tailored to artificial signals present than the Pre-search Data Con- the specific and unique goals of the two. An illustrative dition (PDC) time series (Stumpe et al. 2012; Smith overview is provided in Figure3 for reference. et al. 2012), which has been subject to data massaging techniques already. 5.2. Pass A 5.2.1. PASS A GOALS 5.2.3. STAGE A1.5 We first re-assert that the primary goals of pass A We manually inspect an image of each and every tran- are to: 1) derive transit times for each transit 2) de- sit light curve epoch, centred upon the time of transit rive the transit duration for each planetary candidate. minimum expected from a linear ephemeris and includ- These products may then be used to conduct a second ing 0.5 orbital periods of data either side. We never ± detrending later in pass B, since our detrending pro- attempt to stitch different quarters together and instead cedure (CoFiAM) requires the times and durations for simply reject any data which occurs in a different quar- execution. ter to that of the transit epoch under consideration. As These two objectives necessitate detrending of the before, ephemeris parameters are taken from the NASA data first, since we do a covariant detrending plus in- Exoplanet Archive for this task. ference model which require an HBM, which is beyond At each epoch, we identify if any sharp jumps, ex- the scope of this work for reasons described earlier. ponential flux variations or any other anomalous light For pass A, we make all decisions regarding data qual- curve feature exists in the data. This process is per- ity based upon whether we conclude that these two goals formed by one of us with the perspective of whether 9

remove KOI if now NEA disposition remove all KOIs for which largest

674 changed to FP or NEA PFP>0.01 966 stable exomoon has SNR<1 4098

create dir, wget data, A1 convert to ASCII manual task automated algorithm conducted by a HEK coded by a HEK team member team member inspect each transit A1.5 ±0.5P and flag jumps

run CoFiAM on each run CoFiAM on each A2 transit using NEA transit using pass A B2 ephemeris & duration ephemeris & duration

inspect transit images inspect transit images A2.5 and flag pass A and flag pass B B2.5 inadequate transits inadequate transits

remove object if <2 adequate transits remove object if <2 adequate transits 665 13927 epochs 653

collate good transits, collate good transits, 346 segmented 375 segmented create model LIN & TTV create model LIN & TTV A3 2010 distinct fits 2105 distinct fits B3 MultiNest job folders MultiNest job folders

load & execute jobs load & execute jobs A3.5 on PLEIADES, scp on PLEIADES, scp B3.5 results back to home results back to home

choose favored model, choose favored model, 551 cold (83%) infer new ephemeris & 388 cold (59%) infer new ephemeris & 114 hot (17%) B4 A4 duration, setup B1 folders 265 hot (41%) duration

select OOT data, normalize accept epochs accept objects reject epochs with remove KOIs with by OOTs & contamination, -¾ post 284 with Δt<10 d 284 with Zcirc/Zecc>1 353 high autocorrelation 365 Pr(b>1-p)>4.6% convert to residual ppm, 6096 transits 6144 transits 7351 transits 8513 transits rescale times

Figure 3. Schematic diagram of the pipeline used to process the Kepler SAP photometry to ultimately construct the grand light curve. We color code the manual steps in blue and the automated steps in green. The left column broadly describes pass A, and the right column pass B. For details on each step, we direct the reader to the relevant subsection in Section5.

CoFiAM would be able to fit the light curve variations or Anomalous features are flagged by saving the in- not. The aim is to keep a sufficiently long series of data stant in time just before/after the feature, depending to perform a robust detrending, but clip out patterns on whether the feature occurs after/before the time of which may degrade the performance of CoFiAM, with an transit minimum. This process required approximately appreciation for the basis set which CoFiAM employs. 60 hours of human labor in total. We initially pursued a variety of automated metrics We note that provided cotrending basis vectors are for this purpose, such as standard deviation, autocor- derived from the study of common trends between stars relation and linear trends. However, we found that a and, in an ideal world, would provide a perfect removal wide variety of anomalous features survived and thus of instrumental effects. However, they do not remove deceived these simple metrics. Rather than creating an stellar variations, which must also be removed to ap- ever larger battery of metrics, for which still no guar- ply our method. We therefore opt to use CoFiAM, not antee of completeness could be assured, we instead ac- only because it is optimized for the moon problem, but knowledged that the human eye remains an unparalleled also because it accounts for both instrumental and as- tool in quickly identifying anomalies in time series data. trophysical trends in a single step, which reduces the 10 chances of artificially injecting or removing a moon sig- essentially a second-check of the data quality, catching nal. Fewer manipulations of the data are preferable, any missed anomalies from earlier. and by setting a strict frequency limit to protect the In general, these assessments are made by one of us transit Fourier decomposition signal, we ensure that our searching for any visually evident trends which would method does not overfit out small signals of potential significantly impede our ability to fit the light curves to interest. determine T14. A bad transit does not necessarily have a poor DW statistic, although that tends to be a common 5.2.4. STAGE A2 scenario. Because we anticipate a second pass, we can Stage A2 involves the first detrending of the light be generous in considering acceptable data qualities at curves. As mentioned earlier, this is performed automat- this stage. This acceptance level is non-constant, since ically using CoFiAM and the list of anomalous features we try to allow KOIs displaying very frequent bad tran- to ignore (found manually in stage A1.5). While details sits to have at least a few transits which can be used for of CoFiAM can be found in Kipping et al.(2013a,b), we fitting in stage A3. Vice versa, if a KOI has many clean point out some general options selected for the execution transits, we apply more stringent conditions in assessing in this paper. data quality. Finally, we note that assessments are gen- The outlier threshold was set to 3-σ from a 20-point erally based on SNR of the transit, not the raw wiggles moving median. Before detrending, all planetary tran- in the data, but the relative size of those wiggles com- sits are removed with an exclusion window of 0.6T14 for pared to the transit. In cases of very low SNR, where ± all events (half a duration either side plus 20% buffer), the transit is not visible in a single epoch, we work under including the object of interest. the assumption that sometimes data will wiggle up and Each transit is detrended separately using 0.5P of sometimes down, but we must trust that on the average ± the data surrounding each event. If the transit epoch there is power (thus we try to allow for almost anti- has any associated anomalous flux changes, as discussed transit like features in the interests of being unbiased in stage A1.5, then points beyond these times are also and balanced). In such cases, our criteria switches from cropped. trying to make a good measure of the transit duration Periodic functions described by a sum of harmonic to simply avoiding “catastrophic” detrendings. cosines are explored from twice the data baseline down In addition to a “bad-transit flag”, we also use a to twice the transit duration, with a cap of 30 harmonics “sparse flag” for transits where there is insufficient in- (beyond which we tend to encounter numerical instabil- transit data (or none at all). In some rare cases, a third ities). This choice ensures that CoFiAM does not disturb type of flag was used, “missing flag”, where the data are the shape of the planetary transit in a Fourier sense well-detrended, we have good temporal coverage, but a (Waldmann 2012). transit which should be visually obvious (given the tran- Each model is regressed to the data then ranked by sit depth) is missing in the data. We consider these cases the local autocorrelation, as computed using the Durbin to be most likely due to an erroneous transit ephemeris & Watson(1950) statistic (DW) on the timescale of the on the the NASA Exoplanet Archive. LC cadence, with the lowest autocorrelation being fa- If fewer than two good transits remain for a KOI, the vored. The favoured model is then applied to original object is removed from our sample as being a useful time series, re-including the planetary transit of inter- object. In total, this removes 9 KOIs dropping our sam- est. The final light curve is saved with a 6.66T14 gap ple down from 674 to 665 KOIs. It should be noted ± either side (which is the local timescale used for the DW that any rejection of bad data, manual or automated, calculation). injects additional uncertainty into the occurrence rate Finally, an image of every detrended transit is stored of exomoons calculated in this work which is, strictly along with the best DW statistic. speaking, of unquantifiable magnitude. If there is any correlation between unusable transits and the presence 5.2.5. STAGE A2.5 of moons the calculation could be particularly affected, This is the second manual stage in pass A, where we but this is impossible to measure since the bad transits manually identify “bad” transits. In some rare cases, are by definition unusable for exomoon characterization. CoFiAM fails and produces a light curve which cannot be We proceed under the assumption that bad data result used for fitting, due to visually evident trends remaining from instrumental effects that are distributed randomly in the data. For example, if we missed a location of an across the data set. anomalous flux variation in stage A1.5, CoFiAM may be trying to detrend sharp features with a smooth cosine 5.2.6. STAGE A3 function, producing a poorly detrended light curve. By The third-stage is an automated shell script which be- manually going through the light curves in this way, it is gins by stitching the good transits together for each KOI 11 into a single file. Our script then creates two directo- single fit due to the tractable number of epochs available. ries for a linear ephemeris model fit, LIN, and a transit H 5.2.7. STAGE A3.5 timing variation model fit, TTV, to be fed into Multi- H Nest (Feroz & Hobson 2008; Feroz et al. 2009). The Stage A3.5 was primarily performed using NASA’s script then queries the orbital period, P , and time of Pleiades cluster, and essentially involved loading, com- transit minimum, τ, from the NASA Exoplanet Archive piling, executing and then retrieving the over two thou- database to construct priors for these terms. In the case sands light curve regression jobs required. In total, we of model LIN, the prior on both terms is uniform cen- estimate that approximately 100, 000 CPU hours were H ∼ tered on reported the NASA Exoplanet Archive value used during this phase of the analysis and spanned sev- with 1.0 days window. For model TTV, the period is eral months of wall time. ± H treated as a fixed parameter with the individual transit times following a uniform prior centered on the expected 5.2.8. STAGE A4 time of transit for a linear ephemeris with a 1.0 day In the fourth stage, we segue into pass B by laying ± window again. The choice of the 1.0 day window is es- the ground-work needed using the results from our light sentially arbitrary, but assumes that transit timing vari- curve fits. The first task is to decide for each KOI ations larger than one day are highly unlikely in the re- whether it is dynamically “hot” or “cold”, by which we gion between 0.1 and 1.0 AU (c.f. Holczer et al.(2016) mean whether LIN model (cold) or TTV model (hot) H H who find < 0.3% of their sample have TTV amplitudes is preferred. In cases where the fits were completed us- larger than one day). A larger window will unneces- ing a single segment, the evidences from MultiNest sarily increase the time it takes to explore the parame- can be directly used to compute the Bayes factor and ter space, while a smaller window tailored to the linear rigorously assign the preferred model. For segmented ephemerides of each planet could result in a bias against models, direct evidence calculation is not possible since finding large TTVs. the TTV model has multiple copies of the same param- For model LIN, the basic free parameters are P & eters for the transit shape. Instead, we use weighted H τ, two quadratic limb darkening terms (we use the q1 linear regression to find the maximum likelihood linear & q2 prescription from Kipping 2013), the ratio-of-radii, ephemeris through the posterior transit times and then p, the impact parameter, b, and the stellar density, ρ?. inspect the residuals for evidence of TTVs. This is sim- Uniform priors are adopted for all except ρ? which fol- ply done using a p-value test searching for a excessive lows a log-uniform prior. This gives a total of d = 7 free χ2 (cut off used was 3 σ). parameters, which is easily handled by MultiNest. For Formally, model assessments using a p-value are incor- model TTV, the period is treated as fixed, removing rect since they are never actually compared to another H one degree of freedom, but then each transit epoch has model. More precisely, the p-value test is prone to infer- a unique τ parameter (same prior as LIN), giving us ring significant evidence for the alternative hypothesis H d = 6+N free parameters. For d & 20, the performance even in cases where it should not. For example, excess of MultiNest is severely impeded and global fits are noise from other sources or a single poor measurement not possible. could lead to the p-value test favoring the TTV model A common approach is “templating”, where one folds erroneously. Let us consider the effect of this by imag- the transits, creates a template which is then regressed ining a linear ephemeris fit to a set of transits with the to the individual epochs (e.g. Holczer et al. 2016). TTV model. The times of transit and basic transit pa- This approach underestimates measurement uncertain- rameters will all still come out formally correct, just with ties since it ignores the covariance between the tran- inflated uncertainties. Giving each epoch a free transit sit shape parameters and the individual transit times. time is still able to recover the original linear ephemeris Rather than underestimating errors, we prefer to over- solution. Therefore, despite the p-value’s tendency to estimate them and so adopt a different strategy. Instead, overestimate significance, this merely acts to conserva- we split up our light curves in segments of 10 epochs tively inflate our uncertainties and does not formally ∼ each, providing a manageable number of free parameters invalidate our inferences. for each. The downside is that each segment is not able In total, we find 551 of the 665 KOIs are “cold”, with to utilize the information about the global transit shape the remaining 114 being dubbed hot. For comparison learnt from other segments, and so the uncertainties will to later, we point out that one might expect pass B to be larger (but more robust) than templating. Accord- increase the hot fraction due to the improved detrending ingly, in stage A3 our script automatically segments the and thus greater sensitivity to even small TTVs. data up where necessary. Stage 4 ends by duplicating all of the KOI folders into We find a total of 346 out of our sample of 665 KOIs a new directory with a small text file recording the fa- require segmenting, whereas the rest are treated in a vored model and ephemeris parameters. The transit 12 duration is also recorded in this file, where for LIN- ing another round of 100,000 CPU hours and several H favored KOIs is computed directly from the joint poste- months of wall time. riors and for TTV-favored KOIs from a weighted sum H 5.3.4. STAGE B4 of each segment’s marginalized credible interval for the duration. Finally stage B4 repeats stage A4 with the new light curve fits, performing model selection using the same 5.3. Pass B framework described earlier. After completion, we found 5.3.1. PASS B GOALS that the fraction of cold KOIs indeed decreased as ex- Before describing each data processing step for pass pected from 83% to 59%, giving 265 hot KOIs in our B, we first re-assert the objectives, which represents sample. In this work, these TTVs represent purely a the backdrop against which all decisions in pass B are nuisance but we acknowledge that this data set rep- framed. resents a rich and interesting catalog for others in the Ultimately, the data product from pass B should be community. We therefore make all of the transit times, high-quality, cleaned light curves with accurate esti- both for the hot and cold samples, publicly available at mates of the transit times and basic parameters needed github.com/alexteachey/TTV_posteriors. for stacking and re-scaling. Our tolerance for poor- For each KOI, we export the maximum a posteriori quality light curve is necessarily more stringent here, transit fit and the corresponding vector of out-of-transit since unlike pass A, there is second-chance for these light baseline fluxes (OOTs), which are found by linear min- curves and they have to be of sufficient quality for stack- imization of the maximum a posteriori model and the ing by the time pass B is complete. data (this is actually done on the fly during every regres- sion step, following the approach using by Kundurthy et 5.3.2. STAGE B2 & 2.5 al. 2011). These OOTs will be useful later for stacking Downloading the data (stage 1A) and removing jumps the final light curves. (stage A1.5) does not need to be repeated since the 5.4. Post-Processing raw data product is unchanged. Accordingly, we skip straight to stage B2. Mirroring stage A2, we detrend In post-processing, we aim to export a single file for the SAP light curves using CoFiAM but now using the each KOI which contains a phase-folded light curve suit- duration and transit times determined earlier in stage able for planet-stacking. First, the transit times are re- A4 (specifically we use the maximum a posteriori val- moved using the favored model and the maximum a pos- ues). teriori parameter vector. Next, the planetary transit is In stage 2.5 we again inspect these light curves visually removed by excluding all data which falls within the 2 σ for poorly detrended examples and find 12 KOIs end up upper limit of the derived full duration, T14. Each epoch with fewer than two usable transits after this process. is then corrected for any residual DC power detected Removing these objects reduces our sample from 665 to by the OOT vector regressed during Stage B4. Global 653 KOIs. blending factors, as well as quarter-to-quarter aperture flux contamination factors are corrected for following the 5.3.3. STAGE B3 & 3.5 approach described in Kipping & Tinetti(2010). As with stage A3, stage B3 collates the good transits Next, we subtract unity from the normalized fluxes and sets up folders for models LIN and TTV ready and multiply by a million to create a ppm residual light H H for fitting with MultiNest. Once again this results in curve for each object. Finally, the time column is con- just over two thousand distinct jobs to run, with 375 of verted to t0 using Equation (8) and the maximum a the KOIs being segmented. posteriori transit parameters derived from the preferred Unlike stage A, we here allow the light curve model model regressed in Stage B4. to account for any known blending from nearby sources. These are collated from Everett et al.(2015); Kolbl et 5.5. Filtering al.(2015); Adams et al.(2012, 2013); Dressing et al. Before stacking these planets together, we first remove (2014); Law et al.(2014), and we use Kepler-converted KOIs and individual transits which fail to satisfy sev- magnitudes to estimate the appropriate contamination eral criteria. First, we require a 2 σ confidence that the factor for each band. These blending factors are treated planet is not a grazing event, which would mean that as Gaussian priors, with a standard deviation set by the b > (1 p). Grazing events have degenerate planetary − uncertainty on the converted magnitude. In total, 39 of radii and could be far larger in size, potentially even a the targets required a blending term to be included. false-positive. Erring on the side of caution, we remove In stage 3.5, we again load, compile, execute and re- any such KOI which filters out 288 KOIs, leaving us with trieve these runs on the NASA Pleiades cluster, requir- 365 targets. 13

We next test for excessive autocorrelation using the re-scaled times are well-described by a half-normal dis- DW metric. For each transit, we generate 10,000 mock tribution with a standard deviation of 113. We elect realizations where the data are drawn from perfect nor- to remove any points which fall outside of the range mal distributions at the exact same sampling observed t0 > 150, leaving us with 309750 points. in the data and the reported uncertainties. We use these The grand light curve photometry shows no evidence synthetic transits to generate an expected distribution for correlated noise structure, as expected from averag- for the DW metric and flag any transits for which the ing so many independent data sets together. This is real DW metric is more than two standard deviations verified in Figure5, where we plot the root mean square away from our synethetic population. If a KOI has 50% (r.m.s.) of the time series as a function of bin size, which or more of it’s transits flagged in this way, the entire displays excellent agreement with the expected root N KOI is dropped from the sample. This filtering removes scaling. a further 12 KOIs. Dividing the fluxes by their formal reported uncertain- Next, we compare the transit derived stellar density ties, we find that the robust r.m.s. (given by 1.4286 mul- (which assumes a circular orbit) to an independent esti- tiplied by the median absolute deviation) equals 1.09, mate, in order to exploit the photoeccentric effect (Kip- indicating only a small amount of extra noise above the ping et al. 2012; Dawson & Johnson 2012) to infer a reported uncertainties. We re-scaled the errors by this minimum eccentricity, emin, of each KOI (Kipping 2014). factor and then performed 3 σ clipping, removing just We draw a random sample from the transit derived pos- under one percent of the points. The final time series terior found in stage B4 and divide it by a random sam- is found to have a standard deviation of 5.1 ppm when ple drawn from the KOI’s corresponding stellar density binned to a scale of ∆t0 = 0.5. posterior derived in Mathur et al.(2017). This ratio is then converted into a minimum eccentricity using Equa- 6. MODELING tion (39) of Kipping(2014), and the process is repeated 6.1. Choosing a modeling formalism until we have derived 40,000 posterior samples for emin for each KOI. For each KOI, we also construct a prior In addition to manipulating the Kepler data to con- for emin based off the prior used in the transit fits and struct the grand light curve, we also require the ability Mathur et al.(2017) distribution. to model its shape, as a function of various exomoon We next evaluate the Savage-Dickey ratio between the parameters of interest. posterior and the prior to estimate the Bayes factor, There are two possible avenues to modeling the grand Zcirc/Zecc. We find that 284 KOIs have Zcirc/Zecc > 1, light curve. The first is to model the individual systems implying a near-circular orbit, whereas the other 69 then combine them to create an ensemble model, and the KOIs are rejected for further analysis, on the basis second is to use a model describing the ensemble from that eccentric planets likely result from scattering which the outset. The latter approach describes the model pro- would disrupt moon systems (Gong et al. 2013). We plot posed in Heller(2014) and later modified in Heller et al. emin as a function of the Bayes factor in Figure4. (2016), who refer to this model as the “orbital sampling Finally, we elected to remove transits for which we are effect” (OSE). The great advantage of this approach is unable to measure the transit time to within a precision that one can employ analytic expressions described the of 10−0.75 days, chosen to remove unconverged posteriors ensemble signal without having to laboriously simulate given the prior width, which is necessary to ensure we each of the individual systems and then combine later. are able to reasonably phase-fold transits together. This Thus, in principle, the OSE approach has the advantage did not change the number of KOIs from 284, but did of speed and being more straight-forward in application. reduce the number of transits in our sample from 6144 Indeed, this was the model used in Hippke(2015). to 6096. The alternative approach would be to use a detailed “photodynamical” light curve simulator, such as LUNA 5.6. Constructing a grand light curve (Kipping 2011), to predict the light curve of each system With each target now having a fully processed phase- with some trial set of moon parameters and then later folded light curve, we are finally ready to stack differ- combine them to produce a grand model. Photodynam- ent targets together to create a grand light curve. This ics, a test first coined in Carter et al.(2011), refers to stacking can be across all 284 surviving targets, or a sub- a light curve simulator which evolves a planetary sys- set of them, as explored later. Although we describe here tem at each time step and computes the corresponding the case for the complete ensemble, the planet-stacking shadows cast onto the sky-projected stellar disk. Unlike methodology is the same when dealing with subsets. the OSE model of Heller(2014), this model is not spe- Across the 284 KOIs, we have 6096 unique transits cific to phase-folded events but of course can be easily comprising of 364059 photometric measurements. The used to simulate such a case by simply folding the final 14 ��� ��� 1.0 ◆◆◇◇ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ◆◆◇◇ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ◆◆◇◇ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ 0.8��� ◆◆◇◇ ◆◆◇◇ ◆◆◇◇ ↧↧ ��� ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ 0.6��� ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ��� ◆◆◇◇ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧

� ↧↧ ↧↧ � min ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧

e ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ 0.4��� ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ��� ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ◆◆◇◇ ↧↧ ↧↧ ↧↧ ◆◆◇◇ ↧↧ ↧↧ ↧↧ 0.2��� ↧↧ ��� ↧↧ ↧↧ ↧↧ ↧↧ 0.0��� ��� ��10--8� ��10--7� ��10--6� ��10--5� ��10--4� ��10--3� ��10--2� ��10--1� ��10�0 10��0.25���� ��10���0.5 ��100.75���� ��101.0��� ��101.25���� 10��1.5��� �����/���� Zcirc/Zecc �����/����

Figure 4. Minimum eccentricity of 353 KOIs derived using the photoeccentric effect as a function of the Bayes factor for a circular vs eccentric orbit. KOIs to right of unity are depicted as upper limits on eccentricity, whereas we plot 1 σ credible intervals for the others. The 284 KOIs favoring a circular orbit are considered further as viable exomoon candidates in this work.

Heteroscedastic weighting, • Laplace resonances. •

] 100 We briefly describe these three reasons in what follows.

ppm 50 [ 6.1.1. Inter-population variation

r.m.s. The OSE model is derived assuming one co-adds many 10 transits of the same planet-moon system i.e. that the basic parameters of the system are not changing. How- 5 ever, in our case we co-add different systems together 1 10 100 1000 which have distinct planet and moon parameters. For bin size example, in our Galilean moon fits, we assume that the moons have inclinations representative of Io, Europa, Figure 5. RMS of the grand light curve as a function of bin size, demonstrating the expected N −1/2 scaling of white Ganymede & Callisto, but each planet’s moon system noise. will have unique moon inclinations, despite being drawn from a common underlying distribution. While we could co-add many OSE signals together modeled individu- predicted light curve. ally with the corresponding parameters, each OSE curve In general, LUNA provides a physically detailed light would be modeling only a small number of transits and curve simulation, but comes at the expense of greater thus would be formally invalid - since it is by definition computational cost than the simple closed-form expres- an ensemble model. Without detailed investigation, it sions of OSE. For these reasons, if the accuracy of OSE was unclear that one could simply co-add across a popu- is validated, it would be far simpler and thus prefer- lation in this way and recover the correct phase stacked able to employ the OSE formalism for our model fits of signal and thus we preferred to use LUNA which provided the grand light curve. However, after photodynamical an accurate model of the individual events. testing of the OSE predictions and consideration of the 6.1.2. Laplace Resonances specifics of our problem hand, we came to the conclusion that OSE would not be an accurate modeling tool for A subtle and minor point of concern was dealing with our data product. In particular, we argue that three key the Laplace resonance in the OSE framework when mod- barriers prevent us from directly using the OSE models eling Galilean analogs. In LUNA each individual moon to describe our grand light curve: transit is generated and thus we are able to assign rel- ative phases between the satellites such that they re- Inter-population variation, side in not only the correct mean motion resonance but • 15 also share the Laplace resonance in terms of their mu- max RS ● ● ● ● ●● ● ● ●● ● ●●● ● ●●●● ● ● ● ●● Callistos●● ● ● ●●● ●●● ●● ● ● ● ● ● ●● ●● ●● tual phases: π = λIo + 3λEuropa + 2λGanymede. In con- ● ●● ●●● ● ● ●●●● ● ● ● ●● ●●● ● ●● ●● ●● 3000 ● ● ● ● ● ●● ● ● ●● ● ●● ●● ●● ● ● ● ●● ● ● ●●●●●●●●●●●●● ● ● ● ● ●● ● ●● ●●●●●●●● ●●●●●●●● ● ● ● ● ● ●● ●●●●●●●● ●●● ●●●● ●●● ● trast, the OSE framework never models the individual ● ● ● ● ● ● ● ● ● ●● ●● ●●●● ● ● ● ● ●Ganymedes● ● ●● ● ● ● ● ●● ●●●●●●●●●●●●●●●●●●●● ● ●● ●● ● ● ● ●●● ●●●●● ●● ●●●● ●●● ● ● ● ● ● ●●● ● ●●●● ● ● ●● ●● ●●●● ●●●●●●●●●● ● ● ● ● ●●●● ● ● ●●● ● ●●● ●●●●●●● ● ● ●● ● ● ● ●● ● ●●●● ● ●●●● ●●● ●●●●● ● ●● ● ●● ● ● ● ●●● ●● ●● ●●●● ● ●●●● ● ●●●●● ● ●● ●●● events, rather just the ensemble, and we were unable to ● ● ●●● ●♃ ●●●●●●●●●●●●●● ● ●●●●●● ●●●●●●●●●●●●●● ● ● ● ●● ●● ● ●●● ●●●● III●●●● ●●● ● ●● ●●●● ●●●●●● ●●● ●●●●● SP ●● ● ● ●●●●● ● ●●● ● ●●●●●●●●●●● ] ● ●● ● ● ●● ● ● ● ● ●● ● ●● ● Europas● ● ● ● ● ● ●● ●●●●● ●●●●● ●●●●●●●●●●●●●● ●●●●● ●●●●●● ● ● ●● ●● ● ●● ● ● ●● ● ●●●●●●●●● ●●●●●●●● ●●●●●●●●●●●●●●●● ● ● ●● ● ●●●●●● ● ● ●●●●● ●●●●● ●●●●● ● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●● demonstrate that OSE was correctly accounting for such 2500 ● ● ●●●● ● ●●●●●●●● ●●●● ●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●● ● ●● ● ● ●● ●● ●●●●●●●● ● ●● ● ●●●●●●●●●●●●●●● ● ● ● ●●● ●●●●●●●●●●●●●●●●●●●●●● ● ●● ●●●●● ● ● ●● ●●●●●● ●●●● ●●●●●●●●●●●●●●● ● ●●●● ●●●●● ●♃●IV●●●● ●● km ● ● ● ● ●● ● ● ●●●● ● ●● ●● ● ●● ●●●● ●●●●● min a ● ● ● ● ● ● ● ● ● ●●● ● ●●●●●●●●● ● ●●●●●● ●●●●●●●●●●●●●●●●●●●●● ●●●●● ●●● ●●●●●●●● ●●●●●●● [ ● ●●● ● ●●● ● ●● ● ● ●●●●●●●● ●●●●● ●●● ●● ●● ●●●●●● ● ●●● ●● ● ●● ● ●●●●●●●● ●●●●●● ●●●●●●●●●●● ●●●●●●●●●●●●● ●● ● ● ●● ●●●●●●●●● ●●●●●●● a phase lock. ● ● ●●● ●● ● ● ● ●●●●●● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ● ●●●● ●● ●●● ●●● ●●●●● ● ●●●● ●●●●●● ● ●●●● ●●● ●●●●●●●●●●●●●● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ●●●●●●●●●● ●●●●● S ● ●● ● ●●●●● ● ●● ●●●● ● ●●●●● ● ● ● ● ●●●● ●Ios●● ●● ●● ●●●● ●●●●●●●●●●● ●●●●● ●●●●●● ● ●●●●●●●●●●●●●●● ●● ●●●● ●●●●● ●● ● ●● ●●●●●●●● ●●●●●● ●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●●●●● ● ●●●● ●●●●●● ●●●●●●● ●● ● ●● ●●● ●● ● ● ●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●●●● ● ●● ● ● ● ●●●●●● R ●●● ●● ●●●●●●●● ● ●●●● ●●●● ●●●●●●●● ●●●●●● ●●●●● ●●●●●●●● ● ● ● ● ● ●●● ●●●●● ●●●●●●●● ● ● ●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●●●●●● ●● ●● ●● ●●●● ●● ●●●●●●●●●●●●●●●●● ●●● ●●●●●●●●● ●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ● ●● ●●● ●●●●●●●● ●●●● ●●● ●●●●● ●●●●● ●●●●●●●●● ●● ●●●●●●● ●●●●●● ●●●●●●●●● ●●●●● ●● ● ● ●●● ●● ●● ●●●●●●●●●●●●●●●●●standard●●●●●●●●● ●●●●●●●●●●●●●●Kane●●●●●●●●●●● ●et al. ●(2013)●●●●●●●●●●●●●●●●●●●●●●● ●●● ●● ●● ● ● ●●●● 6.1.3. Heteroscedastic weighting ● ●●●●● ●●●●● ●●●●●●● ●●●●●●●●●●●●●●●● ●●●●●●● ●●●●●● ●●●●●●●●●●●●●●● ● ●●● ●●●● ● ●● 2000 ●●●●● ●●●●●●●●●●●●●● ●●●●●●●●● ●●●●●●●●●●●●●●●●power● ● ●law●●●●●●●●●●●●●●●● ●● ● ● ●● ● ●● ●● ●●●●●●●●deviation●●●●●●●●●●● ●●●●●● ●●●●● ●●●●●●●●●●● ●●●● ●●●●●●●●●● ●●●●●●●● ● ● ● ●● ●●● ●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ●●●● ●●●●●●●●●●●●● ● ●● ●● ●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●●●●● ●●●●●●● ●●●●● ●●● ●●● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●● ●●●●● ●●●●● ●● ● ● ● SP ●●● ●●● ♃●●●●I●●●●●●●● ●●● ●●●●●● ●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●● ● ●●●● ● ●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●● ● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●● ●●●●●●●●●●●● ● ●●●●● ●●●●●● ● ● ● ● ● Finally, OSE is an average of light curves, which by ●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●● ● ●●●●●●●●●●●●●●●●●● ●● ●● ●● ●●● ●● ● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●● ●●●●●●●●●●●●● ●● ●●●●●●● ●●● ● ● ●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●●● ●●● ●● ●●●●●●●●●● ●●● ●● ●●●●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ● ●●●●●● ●●●●● ● ●●● ● ● ● ●●●●●●●●●●●●● ●● ● ●● ●●● ● max a ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●● ●● ●● ● ●● ●●● ● ●●●●● ●●●●●●●●●●●●●● ●●● ●●●●●●●●●●●●♃●● ●●●●●●●● ● ● ● ●● ● ● definition means each light curve is given precisely the ●●●●●●●●●●●●●●●●●●●●●●●● ●●● ●●●●II●●●●●●●● ● ●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ● ● ● ●● ● ● ● ● ● ● 1500 ●●●●●●●●●● ●●●●●●●●● ●●●● ● ● ● ● min RS same weight. Second, each light curve contribution is 5 10 20 assumed to be uniformly and densely sampled. Our real a [R ] data products do not satisfy such constraints, since first SP P we co-add the different planets together using weights Figure 6. One thousand randomly generated Galilean- based off the root mean square of the photometric resid- moon system analogs using the method described in the main uals. Second, light curves are not uniformly sampled, text. Each color represents a unique system comprising of featuring data gaps and removed outliers, as well as be- four moons. ing non-uniformly transformed in time via our temporal re-scaling. Since LUNA models individual events, data gaps, integration time effects (Kipping 2010b) and re- is the actual value for Io around Jupiter and [a, b] U scaling are easily accounted for before applying any co- is a uniform distribution from a b. The next → addition, enabling us to ensure our model is representa- moon along is then assumed to lie in a 2:1 resonance, 2/3 tive of the data. such that (aSP /RP )Europa = 2 (aSP /RP )Io, and sim- ilarly for the Ganymede-analog with respect to the 6.2. Photodynamic Look-Up Tables (LUTs) Europa-analog. The semi-major axis of the Callisto- As a result of the myriad of complicated effects analog, which does not reside in the resonance chain, influencing the final model yet the relative low- is then randomly drawn as (aSP /RP )Callisto [3 + ∼ U dimensionality of the model itself, we elected to build aSP /RP )Ganymede, 1.2 27.2], where 27.2 is the actual × a grid of pre-computed models which accounted for all value for Callisto around Jupiter. Any moon systems of the effects described above. For each KOI, we took generated where two moons have semi-major axes within the maximum a posteriori transit parameters from the 3 planetary radii of each other are rejected. planet-only fits (using the favored model) and generated To generate moon radii which were stochastic yet rep- a planet+moon light curve using LUNA with the same resentative, we adopted the radius-separation power- planet parameters but adding in one or more moons. law model of Kane et al.(2013). The authors note The model curve is generated at precisely the same ca- that the radii of moons tend to increase with respect dence as the data used for the planet-stack and accounts to semi-major axis following a power-law model. We for the long-cadence integration time using 1 minute nu- took the four Galilean moons in isolation and cali- merical re-sampling (Kipping 2010b). After all of the brated a least squares power-law model to it, giving KOI model light curves have been computed for a spe- log(R/R⊕) (6.95 + 0.27 log(aSP /RP ), 0.17), where ∼ N cific choice of underlying moon population, they are co- the standard deviation quoted is that resulting from the added with the same weighting used for the real data. residuals of the best-fitting line. To protect against pe- In other words, we inflict precisely the same transfor- culiar draws, we required that the quadrature sum of the mations to the model as we do to the data, to ensure a radii was within 20% of the actual sum for the Galilean like-for-like comparison at the end. system, that the minimum radius moon was at least 80% the radius of Io and the maximum radius moon was no 6.2.1. Galilean Analogs more than 120% the radius of Ganymede. The resulting The moons are generated in two ways. The first covariant distribution is illustrated in Figure6. case was for a Galilean analog. Here, we assume that To go from physical radii, RS, to RS/RP four moons orbit each KOI with properties resembling needed for the modeling (or equivalently RS/R? = those of Io, Europa, Ganymede & Callisto. To in- (RS/RP )(RP /R?)), we used planetary radii derived in ject some stochastic variation between each moon, yet Chen & Kipping(2017b) to make the conversion. We maintain the 1:2:4 resonance of the inner three, we also included the slight TTV and TDV effects induced randomly place each Io-analog to have a semi-major by the Galilean moons by including (MS/MP ), which axis of (aSP /RP ) [0.8 6.1, 1.2 6.1], where 6.1 was computed by using forecaster (Chen & Kipping ∼ U × × 16

2017a) applied to the moon radius to predict a mass, permit very massive moons which would cause notice- and then using the physical planetary masses predicted able TTVs, which would then be subsequently removed in Chen & Kipping(2017b). anyway by our data processing pipeline described ear- The mean longitude of the inner two moons are ran- lier. domly generated uniformly but the third is enforced to When dealing with subsets, we apply the scaling fac- satisfy a Laplacian resonance. The inclination of the tors found earlier with the Galilean-analog experiments, moons are randomly drawn from a Von Mises distribu- since the computation time to create the grid required tion with κ = 42637, which we found to maximize the many weeks. During the actual fits, we used a bi-linear likelihood of a Von Mises distribution conditioned on interpolation of every unique binned photometric data the real Galilean moons. Eccentricities were kept fixed point, conditioned upon the LUT. We also added an ex- at zero and the moons all follow perfect keplerian orbits tra grid point of RS = 0.0 R⊕, corresponding to a flat i.e. we do not model gravitational interactions between line, to provide numerical stability if fits attempted to the moons. compute the likelihood of radii below our RS = 0.2 R⊕ After stacking the resulting model light curves with limit. correct cadence and weightings, we tried varying the 7. ANALYSIS fraction of systems which harbor moons. Treating each system as a Bernoulli experiment with a probability of 7.1. Galilean Global Fits having a moon system given by η, we found varying η We first discuss our results from regressing our pho- was equivalent to simply scaling the η = 1 resulting light todyanmical phase-folded planet-stacked planet+moon curve by the same factor. Having demonstrated this, we light curve models (see Section6) to all 284 KOIs were able to exploit it to aid in later fits. deemed to be of suitable quality for our analysis (see In cases where a subset of systems were modeled, the Section3). As discussed, the fits are conducted for two process described above was repeated creating unique different light curve models, a Galilean-analog and a sin- model light curves for each specific subset. Generally, gle moon. the shape of each resulting light curve were very similar, For the Galilean-analog, the only parameter directly but ended up with different amplitudes as a result of affecting the light curve model is η, the fraction of KOIs the differing weights, data gaps and stellar radii for the which harbor a Galilean-analog. In addition, we added host stars. As noted these factors represent data specific two other free parameters into our fits. The first was to properties for each subset and were saved for later use account for excess photometric scatter, σ, and was sim- with the single moon simulations. ply added in quadrature to our derived uncertainties in the planet-stacked light curve. The second was an offset 6.2.2. Single Moons term, γ, to allow for a re-normalization of the data set. While a single moon is four times quicker to generate While η and γ were assigned uniform priors, σ was as- than four Galilean moons, the Galilean moons follow an sumed to follow a log-uniform prior from 0.1 to 10 ppm. expected distribution in terms of their sizes and orbital The regression was performed using MultiNest with semi-major axes. For a single hypothetical moon, we 1000 live points. The fits were repeated ten times each, have no idea what these properties are a priori and thus from which the posteriors were combined. Since Multi- our grid cannot simply span η, as before, but now must Nest estimates the marginal likelihood, we repeated our also span RS and aSP leading to a three-dimensional fits with η fixed to zero and removed as a free parame- look-up table. Fortunately, the effect of η is a simple ter, giving us a direct estimate of the Bayes factor, BSP , scaling and thus can be applied easily during the fits for the moon model. themselves, yet this still means we need to generate a We find that the null model is slightly favored, with two-dimensional grid of models, rather than just a single log BSP = 0.84, or an odds ratio of 2.3-to-1 prefer- − look-up example in the case of the Galilean-analogs. ence for the null model. The resulting light curve and We setup a logarithmically-spaced grid from RS = model fitting lines are shown in Figure7, and the asso- 0.2 R⊕ to RS = 2.0 R⊕, with 16 unique grid points. For ciated posterior distribution is plotted in Figure8. Our x aSP , our grid is again logarithmic, defined as aSP = 2 results imply that η < 0.38 to 95% confidence for the where x is stepped through from 1 to 7 in 0.1 steps, 284 KOIs considered, with a 68.3% confidence interval +0.13 leading to a total grid size of 976 elements. The moon of η = 0.16−0.10. is treated as being exactly coplanar and circular with random phase and as before we generate unique light 7.2. Single Moon Global Fits curves for each KOI and then co-add with the appropri- As discussed in Section 6.2.2, the single moon case ate weightings to create our final models. For these sim- required on-the-fly interpolation of a look-up table for ulations we set (MS/MP ) = 0, since some simulations the likelihood calls. For this reason, we found it more 17

] 30 20

ppm ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ [ 10 ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪◻▪◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪◻▪◻▪ ◻▪◻▪◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪◻▪◻▪◻▪ ◻▪◻▪◻▪◻▪◻▪◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪◻▪◻▪◻▪◻▪◻▪ ◻▪◻▪◻▪◻▪◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪◻▪◻▪◻▪ ◻▪◻▪◻▪◻▪◻▪◻▪ ◻▪◻▪ ◻▪ 0 ◻▪◻▪ ◻▪◻▪◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪◻▪◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪◻▪ ◻▪ ◻▪◻▪◻▪◻▪ ◻▪◻▪ ◻▪◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪◻▪◻▪ ◻▪◻▪ ◻▪◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ -10 ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ ◻▪ -20

residual -30 0 25 50 75 100 125 150 separation from planet, t' [planetary radii]

Figure 7. Phase-folded planet-stacked light curve of all 284 KOIs deemed to be of acceptable quality. Temporal axis has been re-scaled and binned, with uncertainties shown given by the standard deviations within each bin. Black solid line represents the expected signature if η = 100% of the KOIs had a Galilean-analog moon system. Blue lines show 100 posterior samples from +0.13 our fits, giving η = 0.16−0.10.

+0.36 σ [ppm] = 2.83 0.36 − size in Earth radii, (RS/R⊕). Both were assumed to fol- low log-uniform priors spanning the limits in our LUT. The η term was allowed to follow a log-uniform prior spanning 0.01 to 100, since technically it represents an effective moon in this case and thus should be inter-

+0.13 η = 0.16 0.10 preted as the average number of moons per system. − +0.59 The fits converged to a solution of RS = 0.51−0.23 R⊕ 0.60 +7.6 +0.33 and (aSP /RP ) = 6.3−3.1 for η = 0.43−0.28. This fit does 0.45

η not directly return a marginal likelihood, since MCMC 0.30 was used, nor is the Savage-Dickey ratio suitable given 0.15 +0.31 that three extra covariant free parameters have been γ [ppm] = 1.57 0.31 − − added. However, the kernel-approach shown later re- 0.6 veals that the evidence favoring the single moon fit in

] 1.2 m

p this region is modest at BSP 2. p [ ' γ 1.8 7.3. Single Moon Kernel 2.4 While the single moon fit is useful for identifying the 1.6 2.4 3.2 4.0 2.4 1.8 1.2 0.6 0.15 0.30 0.45 0.60 maximum a posteriori region of parameter space, it does σ [ppm] η γ [ppm] not provide a clear view of the overall likelihood trends Figure 8. Corner plot of the three-parameter joint posterior occurring within the prior volume. To address this, we distribution from our Galilean-analog moon fit. This fit was repeated our fits for a single moon but fixed aSP /RP disfavored over a null fit with a Bayes factor of BSP = 0.43. and RS to a specific choice and just regressed η, σ and γ. Since no interpolation was necessary, it was straight- forward to use MultiNest with η-rescaling on a single practical to use a Markov Chain Monte Carlo (MCMC) interpolated model each time. RS was varied across a algorithm instead of MultiNest. Our regression was grid from 0.02 to 2 Earth radii in 100 log-evenly spaced performed using a simple MCMC that we wrote, which steps. Similarly, aSP /RP was varied from 2 to 100 in used the Metropolis rule for sampling and normal pro- 100 log-even steps. posal functions tuned by hand to give a 50% accep- At each point, we derived a three-dimensional joint ∼ tance rate. Ten independent chains were seeded from posterior distribution and marginalize over σ and γ to random locations within the prior volume, all of which directly measure the occurrence rate of exomoons at converged within 100 steps, and were then allowed to each location. The posterior derived is mathematically propagate for 105 accepted steps, giving 106 posterior equivalent to ∼ samples in total. In addition to the three free parameters used in the ZZ P(η RS, aSP /RP ) = P(η, σ, γ RS, aSP /RP )P(σ)P(γ)dσdγ Galilean fit, we added in a parameter controlling the | | semi-major axis of the moon, (aSP /RP ), and the moon (9) 18

In addition to deriving a posterior at each grid point, however, that more recent theoretical work (Spalding et we also estimate the evidence against the null model, al. 2016) suggests by contrast that satellites closer in allowing us to compute the Bayes factor. In Figure9 to their planet (aSP . 10RP ) are also vulnerable to dy- we plot the Bayes factor and exomoon frequency for the namical moon-loss during migration. It is unclear, then, ensemble as a function of effective moon radius RS and whether one or both of these mechanisms could be at semi-major axis aSP . The Bayes factor (left) indicates play here, and indeed, the strength of these mechanisms whether the moon model is favored over the model with- rely in part of the evolutionary history of each system out a moon. Red represents regions of parameter space which will of course be unique. In any case, our results where the moon model is disfavored, while green regions suggesting a dearth of exomoons at small aP appear to are areas where the moon model is favored, and inten- provide observational support for the findings of both sity represents our degree of confidence in that model Namouni(2010) and Spalding et al.(2016), and more selection. We emphasize to the reader that paying at- broadly, could be evidence of giant planet migration. tention to the contours in this plot is essential for an accurate interpretation; while much of the plot appears 7.5. Subset Fits green, the moon model is in fact only weakly favored In addition to fitting the entire sample with effective (BSP 2) on a small island in parameter space. By moon and Galilean analog GLC models, we also per- ' contrast, large values of RS and aSP are strongly ruled formed GLC fits on a number of physically motivated out (BSP around 0.01). A value of 1 in this plot means subsets. The aim here was to identify whether a certain we can make no statement about one model being a bet- class of planets in the sample preferentially hosts moons ter fit to the data than the other. over another. As such we divided the sample into several The right side of Figure9 should then be read in the equally-populated pairs and fit the GLC model to these context of the left side. The similarity between the con- subsets. These pairs were small/large planets, cold/hot tours on both sides is readily apparent. For large values planets, cold/hot stars, and inner/outer planets. We of RS and aSP we find an exceptionally low occurrence also split the sample into single/multi-planet systems, rate, while at the lower end of these variables the oc- and habitable zone/non-habitable zone planets (most of currence rate shoots up. It is tempting to read this as the latter residing inside the innermost edge of the hab- a moon signal, but in the context of the Bayes factor itable zone). These last two categories, of course, are on left it is clear these occurrence rates are not at all not equally populated. Insolations were taken from the constrained or well supported by the evidence. Only in NASA Exoplanet Archive and anything less than the regions of high confidence (BSP much greater or much maximum (inner-edge) insolation given in Yang et al. less than 1) should the exomoon frequency values be (2014) equation 2 was considered to be in the habitable given much credence. It is worth pointing out, perhaps, zone. that there is little qualitative difference between a very The results for Galilean moon fits can be seen in Fig- low exomoon occurrence rate and a very low value for ure 10 and Table1. The thick black line in the figure rep- BSP . Both are consistent with virtually no signal in this resents the peak posterior value, while the light red lines region of parameter space. represent 50 fair draws from the posterior. From these plots we can make a number of comparisons. Dividing 7.4. Evidence for a Population of Super-Ios? the sample in two by size, we see a marginally higher The island on the left side of Figure9 where BSP > occurrence rate for the larger planets. For planet tem- 1.50 is intriguing, if only marginally significant. We hes- perature we see a higher occurrence rate on the colder itate to make any strong statement about this region of end. We also see a higher exomoon occurrence rate for parameter space, but it is worth pointing out that theo- colder stars, which we can take to mean later-type or retical modeling (e.g. Namouni 2010) suggests that while evolved stars. All of these observations are in line with planets migrating inward will tend to lose their moons in what we might expect a priori. the process due to a shrinking Hill sphere, they are more We see very little difference in the occurrence rate for likely to retain moons orbiting closer to the host planet single- and multi-planet systems. Interestingly, though, (that is, close-in moons tend to survive planetary migra- the inner 50% of planets (those closest to their star) tion to smaller aP ). Recall that we are probing planets show a significantly higher exomoon occurrence rate within roughly 1 AU of their host star, suggesting that a than those farther away, where the maximum a posteri- large fraction of these planets may have migrated from ori value indicates a total non-detection. This is unex- beyond the snow line. The island of modest moon sig- pected, since the Hill sphere shrinks with smaller semi- nal could therefore be evidence (albeit inconclusive) of major axis. Finally, and perhaps most intriguingly, the a population of moons that have survived migration by maximum a posteriori values for habitable zone plan- virtue of their separation from their host planet. Note, ets ran away to the maximum, while non-habitable zone 19

Bayes factor, BSP (log scaling) exomoon frequency, η☾ (log scaling) 0.01 100 0.01 100 ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 2.0■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■1.50■■ ■■ ■■ ■■ ■■ ■■ ■■ 1.00■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.33■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.10■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.01■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■2.0■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■0.33■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.10■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.03■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 1.00■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 3.33■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 1.0■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■1.0■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 10.00■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.5■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■0.5■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ] ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ] ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♃■■ ■■III■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♃■■ ■■III■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⊕ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ VI■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⊕ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ VI■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♃■■ ■■ IV■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♃■■ ■■ IV■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■

R ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ R ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ [ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ [ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ S ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♃■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ S ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♃■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⊕■■ ■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⊕■■ ■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■

R ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ R ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♃■■ ■■ II■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♃■■ ■■ II■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♆■■ ■■I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♆■■ ■■I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.2■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■0.2■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ III■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ III■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ V■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■IV■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ V■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■IV■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ VII■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ VII■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.1■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■0.1■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ moon radius, ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♇■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ moon■■ radius, ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♇■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■⛢■■ II■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■⛢■■ II■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ IV■■ ■■ ■■ ■■ ■■I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ IV■■ ■■ ■■ ■■ ■■I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ III■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ III■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.05■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.05■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■♄■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ II■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ II■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ ■■V■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ⛢■■ ■■V■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♆■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♆■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ VIII■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ VIII■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ ■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ♄■■ ■■ I■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.02■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 0.02■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ ■■ 2 5 10 20 50 1002 5 10 20 50 100

moon separation, aSP [RP] moon separation, aSP [RP]

Figure 9. Left: Heatmap of the Bayes factor BSP as a function of single effective moon radius RS and semi-major axis aSP for the ensemble. Red indicates regions of parameter space where the moon model is disfavored, while green represents regions where the moon model is favored. Greater color intensity corresponds to greater confidence in the selected model. Right: Exomoon frequency in the ensemble as a function of RS and aSP . A collection of solar system moons are plotted for context.

Galilean Analog Subset Fits max max Group η [1 σ] η [2 σ] BSP Group η [1 σ] η [2 σ] BSP Smaller Planets [0.06,$ 0.35] $0.48 0.353 0.035 Larger Planets [0.07,$ 0.44] $0.61 0.448 0.035 ± ± Colder Planets [0.15, 0.68] 0.86 1.003 0.070 Hotter Planets [0.07, 0.39] 0.53 0.426 0.043 ± ± Colder Stars [0.07, 0.33] 0.44 0.639 0.063 Hotter Stars [0.06, 0.47] 0.66 0.411 0.034 ± ± Inner Planets [0.21, 0.64] 0.80 2.564 0.250 Outer Planets [0.03, 0.28] 0.42 0.224 0.018 ± ± Single-Planet Systems [0.08, 0.50] 0.68 0.689 0.066 Multi-Planet Systems [0.07, 0.33] 0.44 0.420 0.038 ± ± Habitable Zone [0.23, 0.88] 0.97 1.679 0.083 Non-habitable Zone [0.08, 0.37] 0.50 0.653 0.066 ± ± Table 1. Table of occurrence rates η for various subsets of the 284 planets examined in this work. Here η represents 1σ credible th interval values from the posterior distributions while 95 pct is the 95 percentile upper limit. BSP is the Bayesian evidence computed by MultiNest. planets have a much lower occurrence rate. This should cause the depth of the moon signal is controlled by the be read with caution, however, considering the size of size of the effective moon. There is a degeneracy be- the error bars in the habitable zone planet case. While tween moon size and occurrence rate, so we model the the comparisons above are made between equally pop- size of the moon as a proxy for occurrence rate. In ulated subsets, there are far more non-habitable zone essence, a smaller effective moon can mean either a) we planets than there are planets in the habitable zone, have a high occurrence rate, with small moons, or b) we making the results in the latter case far more uncertain. have a low occurrence rate of larger moons. The truth, Single, effective moon fits were also performed for of course, is probably somewhere in the middle. these same OSE subsets. Results can be found in Table To characterize whether our fits exclude a null detec- 2. Unlike the case for Galilean analog fits, we cannot tion to high confidence we compute the Savage-Dickey meaningfully quote an occurrence rate in this case be- ratio between a uniform prior and the posterior distri- 20

Single Effective Moon Subset Fits max max Group RS [1 σ] RS [2 σ] BSP Group RS [1 σ] RS [2 σ] BSP Smaller Planets [0.02, 0.36] 0.90 0.75 Larger Planets [0.02, 0.41] 1.18 0.76 Colder Planets [0.02, 0.41] 1.12 0.81 Hotter Planets [0.02, 0.42] 1.03 0.79 Colder Stars [0.02, 0.35] 0.83 0.79 Hotter Stars [0.02, 0.36] 1.07 0.74 Inner Planets [0.03, 0.81] 1.40 1.04 Outer Planets [0.02, 0.29] 0.83 0.69 Single-Planet Systems [0.02, 0.42] 1.05 0.79 Multi-Planet Systems [0.02, 0.34] 0.96 0.72 Habitable Zone [0.03, 1.10] 1.66 1.12 Non-habitable Zone [0.02, 0.33] 0.92 0.74

Table 2. Table of effective moon sizes RS for various subsets of the 284 planets examined in this work, in units of Earth radii. th We present 1σ credible interval values from the posterior distributions while 95 pct is the 95 percentile upper limit. Here BSP is the Savage-Dickey ratio computed from the log(RS ) posteriors.

2 bution for RS = 0. We treat this number as our Bayes bounding a 68.3% confidence interval, a metric which factor. For the colder stars we find BSP = 1.53, for we loosely refer to as “significance” in what follows (al- outer planets BSP = 0.7, large planets BSP = 1.29, though it is best interpreted as a power). In Figure 11, multi-planet systems BSP = 0.77, and habitable zone we show a histogram of the resulting significances for planets BSP = 3.11. These values suggest that there is 353 KOIs, where we have also included the likely ec- only marginal evidence in support of an effective moon centric KOIs to provide a wider sample to assess dis- signal in the cold star, large planet, and habitable zone tribution properties. Inspection of the results reveals planet subsets, while the null hypothesis is favored for a sizable spread centered around zero, as might be ex- the outer planet and multi-planet subsets. pected. However, we note that KOI-5084.01 (Kepler- 1625b) appears quite deviant from the bulk popula- 8. EXOMOON CANDIDATE KEPLER-1625b I tion with a +4.4 σ significance. Similarly, KOI-4202.01 8.1. Individual Fits (Kepler-1567b) is an outlier on the negative scale with a 4.4 σ significance. Thus far, our analysis has taken a population-based − approach to seeking exomoons, in contrast to the con- The negative outlier clearly cannot be a genuine exo- ventional method adopted in previous HEK papers moon and thus we do not consider it further it in what where candidates were interrogated individually. Al- follows. The positive outlier though could be an isolated though a full suite of photodynamical Bayesian fits to candidate missed by the ensemble analyses. Exclud- each planetary candidate is beyond the scope of this ing the two outliers, the histogram shown in Figure 11 work, representing a formidable computational chal- is well-described by a normal distribution centered on lenge, we do here investigate systems individually with zero with a standard deviation of 0.8. If we generate a a simpler model. list of 353 random variates from that distribution and For each KOI, we took the final phase-folded light take the maximum value, +4.4 is highly improbable; we curve and used MultiNest to fit the Heller(2014) OSE used Monte Carlo simulations to estimate the proba- bility, which was found to be 4 10−6. On this basis, model through the data. As discussed earlier, OSE does × have drawbacks, yet it remains a useful and quick tool Kepler-1625b appears quite unexpected and thus worthy for checking for any significant flux decreases surround- of more detailed investigation. ing the phase-folded transits. For each KOI, we fit for an offset term, a photometric jitter, the moon size, and 8.2. Detailed Investigation of Kepler-1625b a semi-major axis, (aSP /RP ) with MultiNest using To investigate further, we performed detailed photo- 1000 live points. In the earlier ensemble fits, we used dynamical fits of Kepler-1625b using the LUNA model and a log-uniform prior on the moon-size (see Section 7.1), MultiNest, in the same manner as that conducted in which means negative moon radii cannot be sampled. In our previous series of HEK papers (e.g. see Kipping et these fits, we wished to allow for negative radius moon, al. 2013b). This enables a rigorous Bayesian model selec- which correspond to inverted transit signals, to provide tion to ensure not only a physically plausible model can insights into possible biases affecting the results. Ac- explain the photometry, but that the moon parameters cordingly, we modified our moon parameter to be the are justified given the extra complexity they introduce. satellite-to-star ratio-of-radii, s, squared, to account for Comparing the evidences between a planet-only and the fact that transit detection bias is approximately pro- planet+moon model revealed that the moon model was 2 portional to s . favored with log BSP = 10.2, or a 4.1 σ preference, con- We evaluated the median s2 value for each KOI from sistent with the level found previously. The light curve the posterior chain and divided by the lower quantile fits are illustrated in Figure 12. 21 40 Colder Stars Outer Planets Colder Planets Larger Planets Habitable Zone Multi-Planet Systems 35 30 25 [planetary radii] 0 t 20 15 separation from planet, 10 5 0 40 Hotter Stars Inner Planets Hotter Planets Smaller Planets Non-Habitable Zone 35 Single-Planet Systems 30 25 [planetary radii] 0 t 20 15 separation from planet, 10 5 0 0 0 0 0 0 0 20 10 10 20 20 10 10 20 20 10 10 20 20 10 10 20 20 10 10 20 20 10 10 20

eiul[ppm] residual [ppm] residual eiul[ppm] residual [ppm] residual [ppm] residual eiul[ppm] residual

Figure 10. Galilean analog GLC plots for a variety of sample subsets. 22

2.4, 4.6, 4.0, 3.1, and 2.7 σ preference for the moon model (comparable to the original 4.1 σ detection when using all of the data). As with all previous moon candidates we attempted Kepler-1625b to rule out all other possible explanations for the sig-

frequency nal. For any candidate moon signal there will be one of two possible explanations: the signal is either an in- strumental artifact or it has a true astrophysical ori- gin (be it an exomoon or something else). To test the -5 -4 -3 -2 -1 0 +1 +2 +3 +4 +5 possibility of an instrumental aberration we performed significance an independent and manual detrending on the Kepler DR25 data (our OSE survey used DR24) using CoFiAM Figure 11. Histogram of the “significance” of an OSE detec- on the PA and PDC data. In both cases the planet-moon tion for several hundred KOIs, the test which revealed the presence of a possible candidate around Kepler-1625b. The model remained the favored hypothesis over the planet- vertical axis scale is linear. only model. In contrast, polynomial-based detrending was found to remove the signal, likely due to very long- ●○ ●○ ●○ timescale nature of the driving event occurring in epoch ●○ ●○ ●○ ●○ ●○ ●○ +1 ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○ ●○●○●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○●○ ●○ three. 0 ●○ ●○●○●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○●○ ●○ ●○●○●○●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○●○ ●○ ●○ ●○●○ -1 ●○ ●○ ●○ ●○ ●○ We also examined data from individual Kepler pix- -2 ●○ els to determine whether there might have been anoma- ●○ ●○ ●○ ●○ ●○ ●○ ●○ -3 ●○ ●○ ●○ ●○ ●○●○●○ ●○ lous behavior in the vicinity of the transit events (as ●○ ●○ ●○ ●○ -4 ●○●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○●○ ●○ occurred for false-positive Kepler-90g.01; Kipping et al. -5 ●○

●○ 2015b), but no unusual spikes or drop outs were detected ●○ ●○ ●○ ●○ +1 ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ in any relevant pixel. We also verified that there were no ●○ ●○●○●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○●○ 0 ●○ ●○●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○●○ ●○ ●○●○ ●○ ●○ ●○ ●○ bad data flags at the time of the transits, which was a ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ -1 ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○●○ ●○ ●○ ●○ ●○ ●○ ●○ source of the false-positive result for the moon candidate -2 ●○ ●○ -3 ●○●○ ●○ around PH2-b (Kipping et al. 2015a). ●○ ●○ ●○ ●○ ●○ ●○ -4 ●○●○●○ ●○●○ If the signal were astrophysical in origin, then there ●○ ●○ ●○●○ ●○●○ ●○●○ ●○ ●○ ●○●○ ●○●○ ●○ -5 ●○ ●○ are several possible hypotheses, including a moon. A ring is not likely since it should produce a coherent sig- ●○ ●○ +1 ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○●○ ●○ ●○●○ ●○ ●○●○●○●○ ●○ ●○●○ ●○ ●○●○ ●○ ●○●○ ●○ ●○ ●○●○●○ ●○ ●○ ●○ ●○●○ nal in all events, which is not seen, unless the ring pre- ●○ ●○●○●○ ●○ ●○●○ ●○ ●○●○ ●○ ●○●○ ●○ ●○ ●○ ●○ ●○●○ ●○ 0 ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○ ●○●○●○ ●○ ●○ ●○ relative flux (parts per thousand) ●○●○ ●○●○ ●○ ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○ cession rate is very fast. Rotating spots on the surface -1 ●○ ●○ ●○ ●○ ●○●○ ●○ ●○ ●○ ●○ ●○ -2 ●○ of the star also affect the light curve, primarily produc- ●○ -3 ●○ ●○ ing a long-term undulation in the data. This long-term ●○ ●○●○ ●○ ●○ ●○●○ -4 ●○ ●○ ●○ ●○●○ ●○●○●○ ●○ ●○ ●○ ●○●○●○ ●○ trend is removed via our detrending (repeated several -5 times independently) and thus the only remaining pos- -36 -30 -24 -18 -12 -6 0 +6 +12 +18 +24 +30 +36 sible starspot-induced signal would be crossing events. time from inferior conjunction [hours] However, starspot crossings (when a transiting planet Figure 12. The three transits of Kepler-1625b observed with occults a dark spot; e.g. see Rabus et al. 2009) cannot Kepler, overlaid with 100 draws from the model posteriors. be responsible for the observed moon-like dips. This is The black line is the maximum a posteriori model. because spot crossings can only occur inside the main planetary transit and can never produce out-of-transit Given the limited numbed of transits available, flux decreases, as seen for Kepler-1625b, purely from a just three, the conventional HEK approach of cross- geometrical argument. If the signal were confirmed then, validating by removing a transit was not tenable and this would leave the exomoon hypothesis as the leading thus we adapted this strategy somewhat in what follows. explanation based on current information. Instead, we performed k-fold cross-validations, where we The quoted stellar properties of Kepler-1625 in the omit one-half of a transit (centered about the time of in- NASA Exoplanet Archive changed significantly from ferior conjunction) and re-fit the remaining data blindly DR24 to DR25, owing to the addition of updated infor- each time. Since there are three transits in the Kepler mation in the latter data release (Mathur et al. 2017). data, this yields six unique ways of conducting the cross- This update pushed the star from a sub-Solar to a super- +0.366 +0.263 validation. In each case, we performed a new blind fit Solar radius (R? = 0.838 1.793 R ), en- −0.079 → −0.488 and found a positive detection consistent with the origi- hanced the metallicity from sub- to super-Solar abun- nal signal in all cases, with Bayes factors indicating 3.9, dances, and lowered the density substantially (ρ? = 23

2.059+0.4626 0.2636+0.3257 g/cm3), indicating that been captured by the planet, a kind of intermediate pro- −1.306 → −0.0768 this star is likely climbing the giant branch. Criti- cess between typical capture scenarios (e.g. Neptune cally, our planet-only and planet+moon fits favor a low and Triton; see Agnor & Hamilton 2006) and the cat- +0.034 3 stellar density of ρ? = 0.387−0.083 g/cm and ρ? = aclysmic impact event that is believed to have formed +0.028 3 ´ 0.405−0.054 g/cm and if the true density were much Earth’s Moon (Cuk & Stewart 2012). In this scenario higher, then Kepler-1625b would need to be either highly a grazing impact might be experienced as a kind of ex- eccentric or blended (Kipping 2014), both of which treme atmospheric drag sufficient to capture the passing would be severely detrimental to the exomoon hypothe- body. Observation of this system might therefore not sis. Determining the true nature of this star is critical as only produce the first unambiguous detection of an ex- it will also dictate the sizes of the planet and moon de- omoon, but could also go a long way in demonstrating rived from the transit depth (which we describe shortly). once again that what we observe in our Solar System is We also attempted to recover a rotation period for the not all that is possible. star (following the methodology described in Torres et al. 2015) but the amplitude of variability appears too 8.3. Validating the Exomoon Candidate Kepler-1625b I small to recover a consistent period across each quar- At this time, we remain cautious about the reality of ter, with best-fitting periods ranging from 4.5 days to this signal, given the relatively small number of transits 21 days. Attempting to regress a coherent signal across available. This is particularly true because the third all quarters gives an amplitude of 66 ppm, and when transit appears to be crucial to the exomoon interpre- performed on each quarter independently, the median tation and can be removed using polynomial-based de- amplitude was 136 ppm. Given the lack of strong evi- trending approaches. Detrending the photometric time dence for rotation, the weak amplitudes in comparison series of long-period transits is more challenging than to the candidate moon transit depth (570 ppm), and the their shorter-period counterparts and it remains wholly arguments made earlier as to why rotational modula- plausible that the signal observed is nothing more than tions are unlikely to be a source of false-positive, we an artifact of our detrending process. We strongly em- deem it unlikely that activity is responsible for the sig- phasize these points and encourage the community to nal observed. not treat this signal as genuine until it can be confirmed. Our photodynamical fits combined with the DR25 Fortunately, our photodynamic moon fit yields a stellar properties indicate that Kepler-1625b is likely testable prediction for the morphology of the next tran- a Jupiter-sized planet with approximately ten times sit event occurring October 2017. With such a long Jupiter’s mass, orbited by a moon roughly the size of event duration from anticipated exomoon ingress to Neptune. We calculate the radii of the planet and egress, the event cannot be observed in its entirety by moon by measuring the depth of the flux dip (∆F/F = any single optical/NIR instrument on the ground south 2 ◦ (Ro/R∗) , where Ro is the radius of the object in ques- of latitude 78 N (north of which the long Arctic night ∼ tion) and we are able to derive a mass based on the has already begun on the date of observation). A space- photodynamical model fit. We note that both the planet based observation is clearly essential to characterize the and the moon show good agreement between mass and system. We have therefore secured HST observations to radius estimates and lie in physically reasonable param- validate the signal during the next transit of the planet, eter space based on the mass-radius forecaster model of and we strongly advocate treating this object as no more Chen & Kipping(2017a). We find the semi-major axis than a candidate at this time, similar to previous moon +2.1 of the moon aS = 19.1−1.9 RP , which is well outside candidates discussed in earlier HEK papers. the Roche limit and comfortably within the Hill sphere for this planet. It is dynamically stable and should not 9. CONCLUSION have spun out / escaped over 5 Gyr (Barnes & O’Brien In this work we have examined 284 Kepler exoplan- 2002). ets (from an original sample of 4098 KOIs) in search While the existence of a Neptune-sized moon has of an exomoon signal in the ensemble. We performed largely not been anticipated in the literature (however, a rigorous multi-stage analysis to select only the high- see Cabrera & Schneider 2007), we cannot readily rule est quality data, measure and correct for TTVs, and out its existence on these grounds. Indeed, the exis- stack a total of 6096 transit events to characterize the tence of Hot-Jupiters was also wholly unexpected prior exomoon population. As a byproduct of our work we to their discovery in the mid-1990s. It seems clear that present new TTV posterior distributions, along with a a moon of this type could not have formed in a circum- handful of stellar properties, and make them available planetary accretion disk akin to that which is thought online to the community. to have formed the regular moons of Jupiter and Sat- Our results place new upper limits on the exomoon urn. It is conceivable however that the moon could have population for planets orbiting within about 1 AU of 24 their host star, upper limits that are remarkably low. at Ames Research Center. We have also analyzed subsets of the ensemble to test This research has made use of the NASA Exoplanet the effect of various data cuts. Our analysis suggests Archive, which is operated by the California Institute that exomoons may be quite rare around planets at of Technology, under contract with the National Aero- small semi-major axes, a finding that supports theo- nautics and Space Administration under the Exoplanet retical work suggesting moons may be lost as planets Exploration Program. migrate inward. On the other hand, if the dearth of ex- This research has made use of the corner.py code by omoons can be read as a reliable indicator of migration, Dan Foreman-Mackey at github.com/dfm/corner.py. our results suggest a large fraction of the planets in the This work made use of the Michael Dodds Computing ensemble have migrated to their present location. Facility, for which we are grateful to Michael Dodds, Finally, we have briefly highlighted our identification Carl Allegretti, David Van Buren, Anthony Grange, of an exomoon candidate in the Kepler-1625 system, Cameron Lehman, Ivan Longland, Dell Lunceford, Gre- for which we have secured a follow-up observation with gor Rothfuss, Matt Salzberg, Richard Sundvall, Gra- HST. This candidate has passed a thorough preliminary ham Symmonds, Kenneth Takigawa, Marion Adam, inspection, but we emphasize again our position that the Dour High Arch, Mike Barrett, Greg Cole, Sheena Kepler data are insufficient to make a conclusive state- Dean, Steven Delong, Robert Goodman, Mark Greene, ment about the existence of this moon. Only after the Stephen Kitt, Robert Leyland, Matthias Meier, Roy HST observation is made should any claim about this Mitsuoka, David Nicholson, Nicole Papas, Steven Pur- moon’s existence be given much credence. cell, Austen Redman, Michael Sheldon, Ronald Sonen- thal, Nicholas Steinbrecher, Corbin Sydney, John Vaj- grt, Louise Valmoria, Hunter Williams, Troy Winarski and Nigel Wright. We thank members of the Cool Worlds Lab for help- ful conversations in preparing this manuscript. Finally, we thank the anonymous referees for their constructive This paper includes data collected by the Kepler mis- comments. DMK acknowledges support from NASA sion. Funding for the Kepler mission is provided by the grant NNX15AF09G (NASA ADAP Program). AT ac- NASA Science Mission directorate. knowledges support from the NSF GRFP grant DGE Resources supporting this work were provided by the 16-44869. NASA High-End Computing (HEC) Program through the NASA Advanced Supercomputing (NAS) Division

REFERENCES Adams, E. R., Ciardi, D. R., Dupree, A. K., Gautier, T. N., III, Domingos, R. C., Winter, O. C. & Yokoyama, T., 2006, MNRAS, Kulesa, C. & McCarthy, D., 2012, AJ, 144, 42 373, 1227 Adams, E. R., Dupree, A. K., Kulesa, C. & McCarthy, D., 2013, Dressing, C. D., Adams, E. R., Dupree, A. K., Kulesa, C. & AJ, 146, 9 McCarthy, D., 2014, AJ, 148, 78 Agnor, C. & Hamilton, D. P., 2006, Nature 441, 192 Durbin, J. & Watson, G. S., 1950, Biometrika, 37, 409 Agol, E., Steffen, J., Sari, R. & Clarkson, W., 2005, MNRAS, Everett, M. E., Barclay, T., Ciardi, D. R., Horch, E. P., Howell, 359, 567 S. B., Crepp, J. R., Silva, D. R., 2015, AJ, 149, 55 Akeson, R. L., Chen, X., Ciardi, D., et al., 2013, PASP, 125, 989 Gong, Y.-X., Zhou, J.-L., Wie, J.-W. & Wu, X.-M., 2013, ApJL, Barnes, J. W. & O’Brien, D. P. 2002, ApJ, 575, 1087 769, L14 Bennett, D. P., Batista, V., Bond, I A., et al. 2014, ApJ, 785, Gould, A., Gaudi, S. B. & Han, C., 2004, arXiv e-prints:0405217 155 Feroz, F. & Hobson, M. P., 2008, MNRAS, 384, 449 Cabrera, J., & Schneider, J. 2007, A&A, 464, 1133 Feroz, F., Hobson, M. P. & Bridges, M. 2009, MNRAS, 398, 1601 Canup, R. M. & Ward, W. R., 2002, ApJ 124, 3404 ., Hendrix, A., Esposito, L., et al. 2005, AGU Fall Meeting Canup, R. M. & Ward, W. R., 2006, Nature 441, 834 Abstracts, Carter, J. A., Fabrycky, D. C.; Ragozzine, D. et al., 2011, Hansen, C. J., Esposito, L., Stewart, A. I. F., et al. 2006, Science, 331, 562 Science, 311, 1422 Chen, J. & Kipping, D. M., 2017a, ApJ, 834, 17 Heller, R., 2014, ApJ, 787, 14 Chen, J. & Kipping, D. M., 2017b, ApJ, submitted Heller, R., Hippke, M. & Jackson, B., 2016, ApJ, 820, 88 (astro-ph:1706.01522) Holman, M. J. & Murray, N. W., 2005, Science, 307, 1288 Christiansen, J. L., Jenkins, J. M., Caldwell, D. A., et al. 2012, Huber, D., Silva A. V., Matthews, J. M. et al., 2014, ApJS, 211, 2 PASP, 124, 1279 Hippke, M., 2015, ApJ, 806, 51 Cuk,´ M. & Stewart, S. T. 2012, Science, 338, 1047 Hogg, D. W., Myers, A. D. & Bovy, J., 2010, ApJ, 725, 2166 Dawson, R. I. & Johnson, J. A., 2012, ApJ, 756, 122 Holczer, T., Mazeh, T., Nachmani, G., et al., 2016, ApJS, 225, 9 Dickey, J. O., Bender, P .L., Faller, J. E., et al., 1994, Science, Ida, S., Canup, R. M., & Stewart, G. R. 1997, Nature, 389, 353 265, 482 Kane, S. R., Hinkel, N. R. & Raymond, S. N., 2013, AJ, 146, 122 25

Kipping, D. M., 2009a, MNRAS, 392, 181 Law, N. M. Morton, T., Baranec, C., Riddle, R., Ravichandran, Kipping, D. M., 2009b, MNRAS, 396, 1797 G., Ziegler, C., Johnson, J. A., Tendulkar, S. P., Bui, K., Kipping, D. M., 2010a, MNRAS, 407, 301 Burse, M. P., Das, H. K., Dekany, R. G., Kulkarni, S., Kipping, D. M., 2010b, MNRAS, 408, 1758 Punnadi, S., Ramaprakash, A. N., 2014, ApJ, 791, 35 Kipping, D. M. & Tinetti, G., 2010, MNRAS, 407, 2589 Mayor, M. & Queloz, D., 1995, Nature, 378, 355 Mathur, S., Huber, D., Batalha, N. M., Ciardi, D. R., Bastien, Kipping, D. M., 2011, MNRAS, 416, 689 F. A., Bieryla, A., Buchhave, L. A., Cochran, W. D., Endl, M., Kipping, D. M., Bakos, G. A.,´ Buchhave, L. A., Nesvorn´y,D. & Esquerdo, G. A., Furlan, E., Howard, A., Howell, S. B., Schmitt, A. R., 2012, ApJ, 750, 115 Isaacson, H., Latham, D. W., MacQueen, P. J. & Silva, D. R., Kipping, D. M., Dunn, W. R., Jasinski, J. M. & Manthri, V. P., 2017, ApJS, 229, 30 2012, MNRAS, 421, 1166 Morabito, L. A., Synnott, S. P., Kupferman, P. N., & Collins, Kipping, D. M., 2013, MNRAS, 435, 2152 S. A. 1979, Science, 204, 972 Kipping, D. M., Hartman, J., Buchhave, L. A., Schmitt, A. R., Namouni, F. 2010, ApJL, 719, L145 Bakos, G. A.,´ Nesvorn´y,D., 2013, ApJ, 770, 101 Rabus, M., Alonso, R., Belmonte, J. A., Deeg, H. J., Gilliland, Kipping, D. M., Forgan, D., Hartman, J., Nesvorn´y,D., Bakos, R. L., Almenara, J. M., Brown, T. M., Charbonneau, D., G. A.,´ Schmitt, A. R., Buchhave, L. A., 2013, ApJ, 777, 17 Mandushev, G. 2009, A&A, 494, 391 Kipping, D. M., 2013, Proc. for the Frank N. Bash Symp. 2013: Sartoretti, P. & Schneider, J., 1999, A&AS, 134, 553 Seager, S., Mall´en-Ornelas,G., 2003, ApJ, 585, 1038 New Horizons in Astronomy, held October 68, 2013 in Austin, Sheets, H. A. & Deming, D., 2014, ApJ, 794, 133 TX Simon, A. E., Szab´o,Gy. M., Kiss, L. L. & Szatm´ary, K., 2012, Kipping, D. M., 2014, MNRAS, 440, 2164 MNRAS, 419, 164 Kipping, D. M., Nesvorn´y,D., Buchhave, L. A., Hartman, J., Smith, J. C., Stumpe, M. C., Van Cleve, J. E., Jenkins, J. M., Bakos, G. A.,´ Schmitt, A. R., 2014, ApJ, 784, 28 Barclay, T. S., Fanelli, M. N., Girouard, F. R., Kolodziejczak, Kipping, D. M., Schmitt, A. R., Huang, X., Torres, G., J. J., McCauliff, S. D., Morris, R. L. & Twicken, J. D., 2012, Nesvorn´y,D., Buchhave, L. A., Hartman, J., Bakos, G. A.,´ PASP, 124, 1000 2015, ApJ, 813, 14 Spalding, C., Batygin, K., & Adams, F. C. 2016, ApJ, 817, 18 Kipping, D. M., Huang, X., Nesvorn´y,D., et al. 2015, ApJL, Squyres, S. W., Reynolds, R. T., & Cassen, P. M. 1983, Nature, 799, L14 301, 225 Kipping, D. M. & Sandford, E., 2016, MNRAS, 463, 1323 Sparks, W. B., Hand, K. P., McGrath, M. A., et al. 2016, ApJ, Kipping, D. M., Cameron, C., Hartman, J. D. et al., 2016, AJ, 829, 121 153, 93 Stumpe, M. C., Smith, J. C., Van Cleve, J. E., Twicken, J. D., Barclay, T. S., Fanelli, M. N., Girouard, F. R., Jenkins, J. M., Kolbl, R., Marcy, G. W., Isaacson, H. & Howard, A. W., 2015, Kolodziejczak, J. J., McCauliff, S. D. & Morris, R. L., 2012, AJ, 149, 18 PASP, 124, 985 Kov´acs,G., Zucker, S., Mazeh, T., 2002, A&A, 391, 369 Torres, G., Kipping, D. M., Fressin, F. et al., 2015, ApJ, 800, 99 Kozai, Y., 1968, PASJ, 20, 24 Trafton, L., 1974, ApJ, 193, 477 Kundurthy, P., Agol, E., Becker, A. C., Barnes, R., Williams, B. Waldmann, I. P., Tinetti, G., Drossart, P., Swain, M. R., Deroo, & Mukadam, A., 2011, ApJ, 731, 123 P. & Griffith, C. A., 2012, ApJ, 744, 35 Lainey, V., Arlot, J.-E., Karatekin, O.,¨ et al., 2009, Nature, 459, Yang, J., Bou´e,G., Fabrycky, D. C., & Abbot, D. S. 2014, ApJL, 957 787, L2 Lainey, V., Karatekin, O.,¨ Desmars, J., et al., 2012, ApJ, 752, 14