LAMOST telescope reveals that Neptunian cousins of hot Jupiters are mostly single offspring of stars that are rich in heavy elements

Significance

Hot Jupiters are Jupiter-size planets at £1/10 of the Sun–Earth distance, and even though they were the first exoplanet population discovered around sun-like stars, their origins still remain elusive. Using data from NASA’s Kepler satellite and China’s Large Sky Area Multi-Object Fiber Spectroscopic Telescope, we discover a population of close-in Neptune-size planets (called “Hoptunes”) that share key similarities with hot Jupiters. Like hot Jupiters, Hoptunes prefer to reside around stars with higher metal abundance than the Sun. Nearly half of the Kepler planets are discovered in systems with multiple transiting planets, but both hot Jupiters and Hoptunes are preferentially found in single-transiting planet systems. The “kinship” between hot Jupiters and Hoptunes suggests likely common origins and offers fresh clues into the formation of these exotic close-in planets.

Abstract

We discover a population of short-period, Neptune-size planets sharing key similarities with hot Jupiters: both populations are preferentially hosted by metal-rich stars, and both are preferentially found in Kepler systems with single-transiting planets. We use accurate Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) Data Release 4 (DR4) stellar parameters for main-sequence stars to study the distributions of short-period $(1\mathrm{d}<P<10\mathrm{d})$ Kepler planets as a function of host star metallicity. The radius distribution of planets around metal-rich stars is more “puffed up” compared with that around metal-poor hosts. In two period–radius regimes, planets preferentially reside around metal-rich stars, while there are hardly any planets around metal-poor stars. One is the well-known hot Jupiters, and the other one is a population of Neptune-size planets ($2R_{\oplus }\lesssim R_p\lesssim 6R_{\oplus }$), dubbed “Hoptunes.” Also like hot Jupiters, Hoptunes occur more frequently in systems with single-transiting planets although the fraction of Hoptunes occurring in multiples is larger than that of hot Jupiters. About $1\%$ of solar-type stars host Hoptunes, and the frequencies of Hoptunes and hot Jupiters increase with consistent trends as a function of [Fe/H]. In the planet radius distribution, hot Jupiters and Hoptunes are separated by a “valley” at approximately Saturn size (in the range of $6R_{\oplus }\lesssim R_p\lesssim 10R_{\oplus }$), and this “hot-Saturn valley” represents approximately an order-of-magnitude decrease in planet frequency compared with hot Jupiters and Hoptunes. The empirical “kinship” between Hoptunes and hot Jupiters suggests likely common processes (migration and/or formation) responsible for their existence.

More than two decades after the first surprising discovery (1), hot Jupiters still remain one of the most hotly studied exoplanet populations. Observationally, they never seem to fail to yield new surprises. It was realized early on that their host stars were predominantly more metal rich than the Sun (2), and their frequency was later found to strongly correlate with host [Fe/H] (3, 4). Lately, Kepler data show that they stand out as a distinctly “lonely” population for the dearth of other planets on nearby orbits in their systems (5–7). Despite the plethora of observational findings, we still do not know their origins with certainty—we do not know how they migrate to their present close-in orbits $(P\lesssim 10\text{d})$ (review in ref. 8) or whether they have migrated at all (9, 10).

More clues about their origins may come from examining them in the context of planet distributions and the dependence of such distributions on host environment: How unique are the conditions forming hot Jupiters? And do hot Jupiters have “relatives,” which share similarities in their planetary and host-star properties?

With thousands of planets discovered from monitoring $\sim \mathrm{200,000}$ target stars, the Kepler mission has unprecedented potential to study planet distributions over a wide range of parameter space and their possible links to stellar properties (11–17). However, making any reliable statistical inference with a large Kepler sample is seriously limited by the lack of accurate stellar parameters for the majority of the targets. For most of the Kepler targets, stellar parameters are available only via the Kepler Input Catalog (KIC) (18), whose [Fe/H] and $\log g$ measurements are known to have large uncertainties and serious systematic errors (19–22). Significant efforts have been put into characterizing planet hosts by using high-resolution spectroscopy (22–24) or extracting asteroseismic parameters (25). But the accurate parameters of the underlying parent sample (with and without detected planets) are poorly known, which presents a major uncertainty in Kepler planet statistics (relevant discussions in refs. 12 and 15) and also makes it difficult to reliably derive planet distributions as a function of stellar parameters.

With 4,000 fibers and a $5^{\circ }$ diameter field of view, the 4-m Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) (a.k.a. Goushoujing Telescope) (26–29) is uniquely positioned to perform a systematic spectroscopic survey of Kepler target stars. The “LAMOST–Kepler project” (30, 31) attempts to observe all Kepler target stars with no preference for known planet hosts. It forms a large and unbiased sample to perform statistical inference on planet distributions and correlations with host properties (19). For main-sequence stars, the stellar parameters inferred from the official LAMOST Stellar Parameter Pipeline (LASP) (32) are demonstrated to be accurate [typical uncertainties: $\sigma T_{\text{eff}}=100$ K, $\sigma \log g=0.1-0.15$ decimal exponent (dex, referring to a factor of ten), $\sigma [\text{Fe}/\text{H}]=0.1$ dex] from comparisons with high-resolution spectroscopic and asteroseismic parameters (section 1 of supporting information in ref. 21 and also refs. 19 and 33). The accurate LASP stellar parameters make possible the study of the metallicity distribution of Kepler targets (19) and the eccentricity distribution of Kepler planets (21). Stellar parameters have also been extracted from LAMOST spectra using pipelines other than the official LASP, such as LSP3 (34) and ROTFIT (35). Using ROTFIT parameters, ref. 36 found that on average the hosts for hot Earth-size planets have supersolar metallicity.

We study host metallicity distribution for Kepler planets in the period range of $1\text{d}<P<10\text{d}$, using the LAMOST/LASP stellar parameters (32) from Data Release 4 (DR4) (dr4.lamost.org), which contains data for $\sim \mathrm{60,000}$ main-sequence stars with spectroscopic types AFGK Kepler stars ($\sim 30\%$ of all Kepler targets). Previous studies (37, 38) identified a “desert” of short-period Neptunes ($P=2-4\text{d}$ and it may extend to $5-10\text{d}$) by studying samples from radial-velocity (RV) surveys and mixed ground-based and Kepler transit searches. We benefit from the homogeneity of the Kepler sample with well-quantified detection efficiencies as well as accurate stellar and planetary parameters due to LAMOST, and this work is focused on the effects of metallicity.

We discover a “cousin” population of hot Jupiters: Like hot Jupiters, these short-period Neptune-size planets ($2R_{\oplus }\lesssim R_p\lesssim 6R_{\oplus }$), dubbed “Hoptunes,” are predominantly hosted by metal-rich stars, and they also reside more often in single-transiting than in multiple-transiting planetary systems. In the radius distribution the populations of hot Jupiters and Hoptunes frame a “valley” at approximately Saturn size (in the range of $6R_{\oplus }\lesssim R_p\lesssim 10R_{\oplus }$), and in this “hot-Saturn valley” there is a significant deficit in planet frequency.

Sample Selection

We select our stellar sample using the official LASP (32) parameters from the LAMOST DR4 AFGK stellar catalog. The stars satisfy $4,700\text{K}<T_{\text{eff}}<6,500\text{K}$ and $\log g>4.0$ so that our sample consists primarily of solar-type main-sequence stars. They also cover a range of stellar parameters for which we have comparison stars available with accurate high-resolution spectroscopic parameters (19, 21). Our stellar sample consists of 30,727 stars in total. Stellar masses and radii are derived by isochrone fitting as described in ref. 21. We then cross-match with the catalog of Kepler planet candidates and remove the known false positives and those with large false-positive probabilities (FPPs). Stellar parameters derived from LAMOST/LASP are applied to derive the planet properties such as planet radius $R_p$, which has a typical error of about $15\%$. We focus on those with orbital period $1\text{d}<P<10\text{d}$ and planet radius $1R_{\oplus }<R_p<20R_{\oplus }$. The sample has 295 planets in 256 systems. A total of 151 planets are in systems that contain single Kepler transiting planets, and 144 planets are in multiple-transiting planet systems where the other transiting planets in the systems are not necessarily in the abovementioned period and radius ranges. In Supporting Information, Selection of Stellar and Planet Samples, we provide more details on sample selection, and we also show that the main results discussed below are not sensitive to the exact choices of sample selection. The LAMOST stellar and planet parameters of our sample are given in Table S1 and the LAMOST DR3 stellar parameters for the Kepler field are given in Table S2.

Analysis and Results

We divide the stellar sample into two metallicity subsamples: metal rich ([Fe/H] $\ge$ 0) and metal poor ([Fe/H] $<$ 0). Fig. 1 shows the distributions of period and planet radius for the planets that belong to the two subsamples, respectively.

Fig. 1.

Period–radius distribution for short-period Kepler planet candidates hosted by metal-rich (Top) and metal-poor (Bottom) stars. Planets in single- and multiple-transiting planetary systems are plotted in blue and red circles, respectively. The numbers of planets and stars (including nontransiting targets) are shown on the top of each panel. The dark green horizontal line ($R_p=10R_{\oplus }$) denotes the empirical lower boundary of hot Jupiter, and the magenta lines denote the empirical boundaries of Hoptunes (Analysis and Results, iii) Hoptunes and Hot Jupiters Are Separated by a Hot-Saturn Valley). Note that the typical uncertainty in $R_p$ is about $15\%$.

In the metal-poor subsample, except for a handful of objects, almost all of the planets lie in the bottom part of the period–radius plane: For those with $1\text{d}<P<3\text{d}$, their radii are smaller than $\approx 2R_{\oplus }$, and for those with $3\text{d}<P<10\text{d}$, the upper boundary in radius grows from $\approx 2R_{\oplus }$ to $\approx 4R_{\oplus }$ with increasing period. This boundary is approximately illustrated as a magenta polyline connecting $(P,R_p)$ = $(\mathrm{1,2})$ with $(\mathrm{3,2})$ with $(\mathrm{10,4})$ on the period–radius plane. At $P>3$ d, this boundary is similar to the lower boundary of the desert described in ref. 38 (Fig. S4). In the metal-rich subsample, about three-quarters of the planets are concentrated below the magenta polyline. The remaining approximately one-quarter of all planets in the metal-rich subsample reside in the parameter space that is sparsely populated with planets for the metal-poor subsample. One population of planets hosted by these metal-rich stars has radii larger than $\sim 10R_{\oplus }$ (dark green solid line in Fig. 1), and they are the hot Jupiters. The other population hosted by metal-rich stars has radii smaller than $\sim 6R_{\oplus }$ while larger than those indicated by the bottom magenta polyline in Fig. 1. Their sizes $(2-6R_{\oplus })$ are close to that of Neptune at about $4R_{\oplus }$, but we do not know whether all of them are physically Neptune-like planets. Given the uncertainties in their physical states (Neptunes, Super-Earths, mini-Neptunes, or other possibilities), we choose to dub this population Hoptunes in the text. This name reflects our current level of understanding of this population—without mass measurements, we do not know whether they are physically Neptunes, and so far, they can be isolated only in a specific regime in the period–radius plane as found in this study.

Next we discuss three main features in the distributions, including two striking similarities between Hoptunes and hot Jupiters and a significant valley separating them in the radius distribution.

i) Hoptunes and Hot Jupiters Are Both Preferentially Hosted by Metal-Rich Stars, and for Both Populations the Planet Frequencies Correlate with the Host [Fe/H].

The contrast between metal-rich and metal-poor stars in hosting hot Jupiters and Hoptunes is statistically significant. In the metal-rich subsample, there are 17 hot Jupiters and 24 Hoptunes, which are ${9.1}_{-2.2}^{+2.8}\%$ and ${12.9}_{-2.6}^{+3.2}\%$ of all planets in that subsample, respectively, with the uncertainties inferred from the Poisson distribution. If the metal-poor subsample had similar relative fractions of these two populations, we would expect to see ${10.0}_{-2.4}^{+3.0}$ hot Jupiters and ${14.1}_{-2.9}^{+3.5}$ Hoptunes, but there are only one hot Jupiter and two Hoptunes detected. Fig. 2 shows the cumulative fraction of various planet hosts and the stellar sample as a function of [Fe/H]. The strong dependence of hot Jupiters’ occurrence on host-star metallicity is well known, and such a dependence is clearly seen from our sample. Compared with the stellar sample as well as “other hot planets” (i.e., not hot Jupiters or Hoptunes), the hot Jupiters are preferentially hosted by metal-rich stars. Using the two-sample Kolmogorov–Smirnov (K-S) tests, the [Fe/H] distributions of hot Jupiters are inconsistent with drawing from the same sample for the stellar sample and other hot planets (at P values of $3\times {10}^{-5}$ and $3\times {10}^{-4}$, respectively). Hoptunes have similarly strong preference for metal-rich hosts, and their distributions are inconsistent with the stellar sample and other hot planets with even higher statistical significance (at P-values of $4\times {10}^{-4}$ and $3\times {10}^{-4}$, respectively). The two-sample K-S test on the cumulative distributions of hot Jupiters and Hoptunes results a P value of $45\%$ so we cannot reject the null hypothesis that the host metallicities of hot Jupiters and Hoptunes are drawn from the same distribution. The distribution of other hot planets, dominated by Earth-size planets ($1R_{\oplus }<R_p<2R_{\oplus }$), is also more skewed (although in a degree significantly less than hot Jupiters and Hoptunes) toward higher metallicities compared with the stellar sample (the K-S test P value $=0.002$), which is qualitatively consistent with the conclusion by ref. 36 that there is a preference of metal-rich hosts for hot Earths. Later in this section we derive the intrinsic frequencies by taking into account the incompleteness of Kepler.

Fig. 2.

Cumulative fractions as a function of host [Fe/H]. Hoptunes (magenta) and hot Jupiters (green) have similar distributions, and using the two-sample K-S test, the two samples are drawn from the same underlying distribution with a probability of $45\%$. In contrast, the distributions for both Hoptunes and hot Jupiters are different from those of the stellar sample (blue) and other planets (black) at high statistical significance.

ii) Hoptunes and Hot Jupiters Both Tend to Preferentially Exist in Kepler’s Single-Transiting Planetary Systems.

A distinguishing feature of hot Jupiters is that they tend not to have neighboring planets in close-by orbits (5–7). All of the hot Jupiters in our sample are single-transiting planet systems, and the majority of the Hoptunes are in Kepler singles ($73\pm 9\%$ with the uncertainty estimated from binomial distribution). In contrast, in the regimes under the broken polyline of Fig. 1, slightly less than half of planets ($45\pm 3\%$) are in single-transiting planetary systems—note that the single fractions are similar for both metal-poor and metal-rich subsamples, ∼$46\pm 4\%$ for metal poor and $43\pm 5\%$ for metal rich. Therefore, like hot Jupiters, Hoptunes are also preferentially in singles compared with other hot planets. The single fraction of Hoptunes is smaller than that of hot Jupiters in our sample. We use a likelihood analysis based on binomial distribution to test the statistical significance of this difference, and the null hypothesis that they have the same single fraction is ruled out at $2.9\sigma$ significance (Supporting Information, Statistical Significance of a Larger Single-Planet Fraction of Hot Jupiters than That of Hoptunes for details).

iii) Hoptunes and Hot Jupiters Are Separated by a Hot-Saturn Valley.

As can be seen in Fig. 1, there is a deficit of planets near Saturn size (about 6–10 $R_{\oplus }$) in the planet-radius distributions between the Hoptunes and hot Jupiters, and we refer to this deficit as the hot-Saturn valley. The hot-Saturn valley can be clearly seen in the cumulative distribution of planet radius for planets with $R_p>4R_{\oplus }$ from our sample (Fig. 3, Top). The cumulative distribution has two clear breaks occurring at $\sim 6.5R_{\oplus }$ and $\sim 10R_{\oplus }$, and between these breaks, planets occur significantly less frequently compared with other ranges of radii. In Fig. 3, Middle, we show a histogram of observed radius distribution with 12 uniformly spaced bins in logarithm. The four bins closest to $8R_{\oplus }$ have 1 planet where the four bins on the left (Hoptunes) have 10 planets and the four bins on the right (hot Jupiters) have 16 planets. If the significance of the valley is assessed by the difference between adjacent sets of bins, we can adopt the Skellam distribution to calculate the probability assuming a null hypothesis of all bins having the same expected numbers of planets. Comparing 10 or 16 planets in the neighboring sets of bins with 1 planet in the central bins, we find that the probability of the valley due to random fluctuations is about $1\%$. The existence of such a valley with relatively sharp boundaries also suggests that the (unknown) effects of blending due to unresolved background binaries are unlikely to be significant in blurring the planet-radius distribution.

Fig. 3.

The hot-Saturn valley revealed from the radius distribution of planets in our sample. (Top) Cumulative distribution of planets with radii larger than 4 $R_{\oplus }$. (Middle) Number of detected planets in radius bins with equal size in logarithm. (Bottom) Intrinsic planet frequencies as a function of planetary radius.

Finally we consider the incompleteness of the Kepler survey and calculate the intrinsic planet frequency for our sample. We take the survey selection, which is the incompleteness due to the survey detection thresholds, and the transit geometric bias into account. To calculate the incompleteness, we use the code supplied by ref. 15 based on the detection efficiency characterization method of ref. 39 to calculate the correction factor per bin in the parameter space, and we apply the LAMOST stellar parameters in the calculations.

To better examine the dependence of the planet frequency on metallicity, we divide the sample according to [Fe/H] into three subsamples—“supersolar” ([Fe/H] $\ge$ 0.1), “solar” ($-0.1<$ [Fe/H] $<0.1$), and subsolar ([Fe/H] $\le -0.1$). The planet period–radius distributions for three metallicity subsamples are shown in Fig. 4, Left. In the subsolar subsample (Fig. 4, Bottom Left), there are only a handful of hot Jupiters and Hoptunes. Then in the solar subsample (Fig. 4, Middle Left), there are several hot Jupiters, and a few Hoptunes start to emerge close to the lower boundary (the magenta polyline). Finally, in the supersolar subsample (Fig. 4, Top Left), there is a dramatic increase in numbers for both hot Jupiters and Hoptunes. A noticeable trend is that the lower boundary of the planet valley evolves from $\approx 6R_{\oplus }$ (at around Saturn size) in the supersolar subsample to lower values, close to $\approx 3-4R_{\oplus }$ (at around Neptune size) in the solar subsample. Fig. 4, Right shows the intrinsic frequencies of hot Jupiters and Hoptunes after correcting the survey incompleteness and geometric biases. The frequencies of Hoptunes and hot Jupiters increase dramatically (by a factor of $\sim 10$ from subsolar to supersolar) with [Fe/H]. Such a trend for hot Jupiter hosts is well known (e.g., ref. 40 and references therein), and the trend for Hoptune hosts is remarkably similar. The frequencies of Hoptunes are similar (within a factor of $\sim 2$) to those of hot Jupiters for all subsamples. Hosts of Hoptunes and hot Jupiters have similarly strong preferences for host metallicities higher than those of the Sun (supersolar vs. solar).

Fig. 4.

The dependence of planet distribution (Left) and intrinsic frequency (Right) on stellar metallicity ([Fe/H]), using three metallicity subsamples. The frequencies of hot Jupiters and Hoptunes have similar trends as a function of metallicity—for both populations, they increase by a factor of $\sim 10$ from the “subsolar” metallicity regime ([Fe/H]$\le -0.1$) to the supersolar regime ([Fe/H]$\ge 0.1$), and the frequencies of Hoptunes are similar (within a factor of $\sim 2$) to those of hot Jupiters for all subsamples.

Fig. 3, Bottom shows the intrinsic frequencies of planets within the $\log (R_p)$ bins spanning $0.2$ dex around $\sim 5R_{\oplus },\sim 8R_{\oplus }$, and $\sim 12R_{\oplus }$ in our sample. The averaged frequencies of the planets in these subsamples are ${0.23}_{-0.07}^{+0.10}\%$, ${0.02}_{-0.02}^{+0.05}\%$, and ${0.36}_{-0.09}^{+0.10}\%$ at approximately Neptune size, Saturn size, and Jupiter size, respectively. Thus, the hot-Saturn valley represents an approximately one order-of-magnitude depression in planet frequency as a function of planet radius. The averaged planet frequencies $\text{d}{N}_p/\text{d}\log (R_p)\text{d}\log (P)$ for hot Jupiters and Hoptunes in these subsamples are consistent within the uncertainties.

Discussion and Conclusion

Our findings benefit from the homogeneity of the Kepler sample as well as accurate planet radii and host metallicities due to the high accuracy of LAMOST stellar parameters. Refs. 37 and 38 found a desert of short-period Neptunian planets, using mixed samples of planets discovered from various surveys. The upper boundary of the desert defined by ref. 38 is similar to that of the hot-Saturn valley we identify in this work (Fig. S4). While this upper boundary from ref. 38 is a diagonal line (Fig. S4) on the period–radius plane based on a large number of planets mostly contributed from ground-based transits, our sample is too small to clearly determine a slope. Their lower boundary resembles the lower boundary of Hoptunes, especially for $P>3\text{d}$. According to our results, their desert appears to encompass both the hot-Saturn valley and the Hoptunes. Inside their desert, we find that Hoptunes have a similar averaged frequency of $\sim 1\%$ to that of hot Jupiters. In comparison, the hot-Saturn valley has an order-of-magnitude deficit in planet frequency. These features appear to be largely “washed out” in ref. 38. This is mainly due to the large uncertainties and systematics biases in $\log$(g) measurements of the KIC (e.g., figure S4 of ref. 21), and the resulting large errors in planet radii fill some objects in the valley (Fig. S3, Right).

The similarities in host metallicity, intrinsic frequency, and preference for single-transiting planetary systems suggest a close link between hot Jupiters and Hoptunes in their migration and/or formation processes.

The correlation between the intrinsic frequency of short-period Jupiters and stellar metallicity has been well established (3, 4). According to core-accretion models, the total masses of building-block planetesimals and/or embryos are proportional to those of the heavy elements in the host stars (41). One possible interpretation of the metallicity correlation for short-period Jupiters is that the metal-rich environment provides more building blocks to form massive planetary cores ($\sim$10 $M_{\oplus }$) before the gas disk dissipates, which are crucial for gas accretion to form giant planets like hot Jupiters (42). However, such an interpretation may not be well suited for the metallicity correlation of Hoptunes, since for most Hoptunes, especially those smaller ones with radius less than 4$R_{\oplus }$, they do not need massive cores to accrete as much gas as needed for forming Jupiters. Another possibility is that metallicity may play an important role to trigger/amplify certain migration mechanisms for hot Jupiters (43, 44). Such mechanisms should then similarly operate for Hoptunes and also preferentially produce single-transiting planetary systems. Note that we find at $2.9\sigma$ significance that the fraction of single-transiting planet systems of hot Jupiters is higher than that of Hoptunes, indicating that whatever process removes the “brothers” of hot Jupiters likely operates less efficiently for Hoptunes. The in situ formation mechanisms (45) are also subject to these constraints.

The radius range of the hot-Saturn valley ($R_p$ $\sim 6-10R_{\oplus }$) roughly corresponds to the mass domain (between $\sim 10-30$ and $\sim 100-200$ $M_{\oplus }$) of the “planet desert” expected from core accretion simulations of planet formation (46, 47). However, these predictions apply to $a=0.2-3$ astronomical units (AU) while the planets in our sample have $a<0.1\text{AU}$.

When studying the radius distribution for short-period planets, it is important to consider the effects of planetary inflation (e.g., ref. 48 and references therein) and/or evaporation (49–51). Ref. 49 suggests that hot Neptunes may originate from evaporation of hot Jupiters, and thus they may share common origins and evolution history (although note that ref. 49 has been contradicted by some follow-up theoretical studies such as ref. 52). This hypothesis is consistent with the similarities between hot Jupiters and Hoptunes found in this work. In addition, it may be interesting to test whether mechanisms such as photoevaporation can be responsible in shaping features such as the sharp lower boundary for Hoptunes (the magenta polyline in Fig. 1) in the period–radius distribution and also explain how the sharpness of this boundary varies with host metallicity. For instance, the consequence of photoevaporation may depend on planet core mass (50, 51), and one may speculate that core mass distribution can differ according to host [Fe/H] thus can potentially play a role in forming the [Fe/H]-dependent planet distribution found here. RV follow-ups of Hoptunes can provide the crucial mass measurements to reveal their physical states and test such scenarios.

The “kinship” between hot Jupiters and Hoptunes as well as the hot-Saturn valley separating these two cousins offers unique clues and constraints for the formation and migration of short-period planets. Future surveys and missions, particularly Transiting Exoplanet Survey Satellite (TESS), are expected to detect many more short-period planets and explore these features in greater detail.