DIRECT IMAGING OF A COLD JOVIAN EXOPLANET IN ORBIT AROUND THE SUN-LIKE STAR GJ 504

The rotation of a star slows as the star ages and the stellar wind carries away angular momentum. Gyrochronology, which estimates a star's age from its rotation rate and B − V color (i.e., mass), is generally considered to be the most direct, and possibly the most accurate, method for inferring stellar age (MH08). Chromospheric and X-ray activity are powered by the stellar dynamo, which is closely related to the star's rotation rate. Indeed, chromospheric age estimators often use stellar activity to infer the rotation period, which is then used to infer an age (MH08). The rotation period of GJ 504 has been directly measured to be 3.328 days (Messina et al. 2003) and 3.33 days (Donahue et al. 1996). We adopt the average of these measurements, 3.329 days, as the star's rotation period. The possible error range of <0.01 days is small enough not to affect the uncertainty of the age estimate.

The gyrochronology can be applied to I-sequence stars whose dynamo mechanism is presumed to originate at the interface between the convective and radiative zones in the stellar interior (Barnes 2007). In contrast, this method cannot be applied to C-sequence stars that are fully convective and may thus have a different dynamo mechanism that prevents efficient spin-down. We can judge that GJ 504 is an I-sequence star using two different sets of criteria specified in the literature (Barnes 2007; Dupuy et al. 2009): one in which we relate the age-normalized rotation of GJ 504 to other stars of similar B − V colors, and the other in which the X-ray luminosity and Rossby number are compared to other chromospherically active stars. Both of these methods confirm that GJ 504 is consistent with an I-sequence star. This conclusion agrees with the previous classification (Barnes 2007) of GJ 504 as an I-sequence star.

To rigorously establish the age of GJ 504, we use three gyrochronology relations independently proposed (MH08; Barnes 2007; Meibom et al. 2009). All the relations determine a stellar age as a function of rotation period and B − V color. Using the rotation period (3.329 days) and B − V color (0.585 ± 0.007) of GJ 504, we derive ages of 140–180 (MH08), 100–140 (Barnes 2007), and 110–230 Myr (Meibom et al. 2009), respectively. The error ranges for the age estimates were derived based on all the possible sources of error: the observed scatter in the period–age relation (0.05 dex; MH08) or errors of coefficients in the period–age function (Barnes 2007; Meibom et al. 2009), and the error of the B − V color. Conservatively adopting all these ranges,²⁸ we estimate the age of GJ 504 to be 160$^{+70}_{-60}$ Myr.

In order to verify the adequacy of GK 504's age assessed from the empirical relations of gyrochronology (see above), we directly compare the rotation period and B − V color of GJ 504 to those of stars in young clusters (see Figure 2). The rotation data for stellar members in Pleiades (130 Myr; Hartman et al. 2010), M35 (150 Myr; Meibom et al. 2009), M34 (220 Myr; Meibom et al. 2011), M11 (230 Myr; Messina et al. 2010), M37 (550 Myr; Hartman et al. 2009), Hyades (630 Myr; Radick et al. 1987, 1995; Delorme et al. 2011), and Praesepe (600 Myr; Delorme et al. 2011) can be used for this purpose. As shown in Figure 2, the stars with similar B − V colors to GJ 504 in 130–230 Myr old clusters have periods that are the most consistent with that of GJ 504, and the rotation periods in the older 500–600 Myr cluster are significantly longer,²⁹ thus verifying the age range of GJ 504 derived above. Additionally, combining the rotation data of stars in M11 with those in M34, Messina et al. (2010) derive a median rotation period of G-type stars in these clusters of 4.8 days, again supporting the age estimate of 160$^{+70}_{-60}$ Myr for the G-type star GJ 504 (3.329 day period).

Zoom In Zoom Out Reset image size

Since gyrochronology is a largely empirical method, there could in principle be uncertainties involved in the methodology that are not accounted for in the calibrations. For instance, if we hypothesize that the calibration could be metallicity-dependent, this would lead to systematic errors for targets with significantly different metallicity from the calibration clusters. For the case of GJ 504, it has a non-extreme value in this regard, and hence it is unlikely to be subject to such uncertainties. Another possible systematic uncertainty in gyrochronology involves the fact that rotation is easier to detect in rapidly rotating systems, which could result in a bias in the empirical gyrochronology relation. If this were a significant effect, it would cause the apparent mean rotation in a cluster of a given age to appear more rapid than it is in reality. Hence, age estimates for individual stars based on a relation calibrated against such clusters would result in an overestimation (i.e., an individual star would appear older than it really is). Given that no other age indicators imply a significantly younger age for GJ 504 this is also unlikely to be a real issue, but nonetheless, such uncertainties are important to keep in mind in the age analysis.

Stellar chromospheric activity, as traced by Ca ii H and K emission, is a good indicator of stellar age. The strength of Ca ii H and K emission is parameterized by $R^{\prime }_{\rm {HK}}$, roughly the power in the H and K lines normalized to that in the underlying photospheric continuum. A relation between $R^{\prime }_{\rm {HK}}$, B − V color, and age has recently been derived for solar-type stars (MH08). For GJ 504, we adopt the value of log $R^{\prime }_{\rm {HK}} = -4.45$ measured at Mt. Wilson Observatory with a temporal baseline of 30 yr (Radick et al. 1998), the longest such baseline in the literature. In order to minimize the uncertainty in $R^{\prime }_{\rm {HK}}$, the value for which the star has been monitored for the longest possible time should be adopted (see MH08), which also protects against accidentally adopting an outlier in the activity variation.

This activity level may be used to infer a Rossby number, the ratio of the stellar rotation period to the timescale of convective overturn, of 0.60 ± 0.10 following MH08.³⁰ We use this result,³¹ together with a B − V color of 0.585 ± 0.007 (van Leeuwen 2007), to estimate an age of 330 ± 100 Myr. We further add a 0.2 dex scatter in this relation, estimated from clusters³² (MH08). We assume this scatter to be uncorrelated with other sources of error. As a result, we obtain a chromospheric age estimate of 330 ± 180 Myr.

Other studies using data from Mt. Wilson Observatory have measured $R^{\prime }_{\rm {HK}}$ values of −4.443 (Baliunas et al. 1996) and −4.486 (Lockwood et al. 2007), which differ slightly from our adopted value of −4.45. The chromospheric activity level was also measured to be $R^{\prime }_{\rm {HK}}=-4.44$ using 14 yr of observations at Fairborn and Lowell Observatories (Hall et al. 2009). Many of these studies' observational baselines overlap with the source of our adopted $R^{\prime }_{\rm {HK}}$ value (Radick et al. 1998). Adopting the mean of these measurements would have little effect on our chromospheric age of 330 ± 180 Myr.

A primary uncertainty in age determination from chromospheric activity is long-term activity cycles, which may easily exceed the timescales over which the activity is measured, and thus may potentially provide a systematic error if the activity is measured at an extreme but is taken to represent the mean activity for the star. In addition, as in the case of gyrochronology, the stellar activity also depends on the metallicity, possibly contributing to the uncertainty. Since activity as an age estimator is thought to be closely related to gyrochronology through the stellar dynamo, the possibility of measuring both rotation (which is robust against activity cycles) and activity in the same star as for the case of GJ 504 can alleviate this uncertainty.

Young solar-type stars are active X-ray emitters. As for Ca ii H and K, a relation has been derived (MH08) between stellar age, color, and X-ray luminosity. The best estimate of the X-ray luminosity of GJ 504 comes from the ROSAT all-sky survey of nearby stars (Hünsch et al. 1999), which has provided a ratio of X-ray to bolometric power log L_X/L_bol = −4.42 (Messina et al. 2003). This X-ray activity may be used to infer a Rossby number of 0.54 ± 0.25 (MH08) and lead to the age estimate of 90–530 Myr for GJ 504. Because the temporal baseline for the X-ray measurement is much shorter than for the multi-decade measurements of $R^{\prime }_{\rm {HK}}$, the inferred Rossby number is much more uncertain. Therefore, we do not adopt this estimate based on the X-ray emission, although it is consistent with the estimate from other methods.

The strength of lithium absorption lines declines as a star burns its primordial supply of lithium, providing another indicator of stellar youth (Soderblom 2010). However, many effects, such as variations in the primordial supply of lithium, the extreme sensitivity of early lithium burning to the stellar accretion history (Baraffe & Chabrier 2010), and the importance of non-local thermodynamic equilibrium abundance corrections (Takeda & Kawanomoto 2005), make it difficult to obtain a reliable age estimate from lithium abundances. Therefore, we do not employ lithium as an indicator to date GJ 504, but it can provide a clue to the age of GJ 504 as discussed below. The equivalent width of Li in GJ 504 has been measured to be 0.08 Å, implying a lithium abundance of log n(Li) = log (N(Li)/N(H)) + 12 = 2.9 (Takeda & Kawanomoto 2005). Extensive observations of well-dated open clusters (Sestito & Randich 2005) enable a lithium age estimate for GJ 504. At the effective temperature of a G0 star (∼6000 K), these cluster estimates can be used to infer an age range of approximately 30–500 Myr, fully consistent with our gyrochronological and chromospheric age estimates.

GJ 504 is not currently identified as a member of any clusters or moving groups (MGs), making it impossible to use population-based age estimators. The kinematic velocity of GJ 504 (U = −38.6, V = 1.5, W = −18.0; Mishenina et al. 2004) is consistent with thin disk stars, thus providing no tight constraint on the age, but remaining consistent with the young age provided by gyrochronology and chromospheric activity.

Since stars evolve along mass tracks in an H-R diagram with age, a common age determination method is based on isochronal analysis. This method can provide reasonable estimates for old stars that have evolved significantly off the main sequence and very young stars that are still undergoing rapid contraction. However, it is generally not useful for stars near the zero-age main sequence (ZAMS), around which stellar evolution proceeds slowly (e.g., Soderblom 2010). Indeed, generic isochronal estimates in the literature are known to give erroneous ages for young stars by up to two orders of magnitudes.

As illustrative examples, we consider the cases of HD 152555, HIP 10679, and HD 206860. These are all early G-type (G0–G2) stars which are members of young MGs; hence they are similar types of stars to GJ 504 with well-determined young ages, and thus can be used for comparison to isochronal ages of e.g., Valenti & Fischer (2005). HD 152555 is a member of the AB Dor MG (e.g., Torres et al. 2008) with an age of 50–100 Myr, whereas Valenti & Fischer (2005) give an age of 4.0 Gyr. HIP 10679 is a member of the β Pic MG (e.g., Torres et al. 2008) with an age of ∼10 Myr, whereas Valenti & Fischer (2005) give an age of 3.0 Gyr. Finally, HD 206860 is a member of the Her-Lya MG (e.g., López-Santiago et al. 2006) with an age of ∼200 Myr, whereas Valenti & Fischer (2005) give an age of 3.1 Gyr.

In light of these large discrepancies, it is unsurprising that literature estimates for the isochronal age of GJ 504 vary strongly from hundreds of Myrs to several Gyrs (e.g., Valenti & Fischer 2005; Takeda et al. 2007; Holmberg et al. 2009; da Silva et al. 2012). This implies that isochronal analysis should simply be disregarded for a star such as GJ 504, compared to the more reliable age estimators listed above. Nonetheless, since some of the isochronal literature estimates provide Gyr-range ages that are seemingly inconsistent with the younger age range we have adopted for GJ 504, we discuss their application to this star in more detail below.

An important limiting factor in isochrone analysis for age determination of an isolated main-sequence star such as GJ 504 is the sensitivity to uncertainties in input stellar parameters. This is in particular true for properties such as effective temperature and surface gravity, where the uncertainties are model-dependent and often tend to be underestimated in the literature. As an illustration, we consider the isochronal analysis in da Silva et al. (2012) based on the Yonsei–Yale isochrones (Yi et al. 2001), which gives an age of ∼3 Gyr, and appears to be inconsistent with ages of ⩽500 Myr within their quoted error bars. We have reproduced the analysis of da Silva et al. (2012) and confirmed these results with the same methodology and input observables. This is plotted in Figure 3. Also plotted in the same figure is the exact same analysis, but using the temperature estimates of GJ 504 from Valenti & Fischer (2005) instead; note that all other values are kept as in da Silva et al. (2012). This comparison immediately illuminates one of the considerable uncertainties related to isochronal analysis, which is uncertainties in the temperature scaling. The two temperature estimates in Valenti & Fischer (2005) and da Silva et al. (2012) are inconsistent within their quoted error bars, implying that neither error estimate should be trusted when translating this into an error on the age from the isochronal comparison. Moreover, we find that while the temperature estimate of da Silva et al. (2012) gives the aforementioned age of ∼3 Gyr and is inconsistent with ⩽500 Myr ages, the temperature estimate of Valenti & Fischer (2005) is fully consistent with these younger ages, and inconsistent with the ∼3 Gyr age.

Zoom In Zoom Out Reset image size

It is also important to note that due to pre-main-sequence evolution, there is some overlap between the isochrones of stars arriving at the main sequence, and those leaving it. For instance, the 100 Myr isochrone overlaps with the 1 Gyr isochrone in the relevant temperature range. Hence, while the age is quoted as ∼1.4 Gyr in Valenti & Fischer (2005) which is reproduced in Figure 3, it can be equally well interpreted as being consistent with our adopted age range, and in fact preferentially supporting the youngest ages.

Furthermore, while the temperature and luminosity in da Silva et al. (2012) gives a certain age from comparison with the Yonsei–Yale isochrones, this does not account for the relatively high surface gravity they quote. We can see this in the second panel of Figure 3, where again we perform an isochronal analysis, but in surface gravity versus temperature as opposed to luminosity versus temperature. Here, regardless of which temperature value is used, the data are consistent with a ⩽500 Myr age, but inconsistent with a ∼3 Gyr age.

It is instructive to think of these discrepancies in terms of the radius of the star. If we adopt the da Silva et al. (2012) temperature, then the luminosity is high for its mass, implying a large radius. On the other hand, the surface gravity is also high for its mass, implying a small radius. Hence, there is a tension in the radius between the luminosity and surface gravity if the da Silva et al. (2012) temperature is adopted, such that no self-consistent picture can be reached by either increasing or decreasing the stellar radius. By contrast, if we adopt the Valenti & Fischer (2005) temperature, then this tension disappears, and a self-consistent picture can be reached between the temperature, luminosity, and surface gravity—and simultaneously, we obtain a predicted age which is consistent with the estimates given by the age estimates in the previous sections. This could therefore be interpreted as supporting the higher temperature of Valenti & Fischer (2005) and thus a younger age.

Hence, we note that the isochronal age is readily consistent with gyrochronology and activity, using values for temperature, luminosity, and surface gravity published in the literature. This is also consistent with the arguably most detailed Bayesian isochronal analysis of Takeda et al. (2007), which was a study meant to improve on the analysis of Valenti & Fischer (2005) using the same input data, and which gives an age limit of <760 Myr. Nonetheless, the above discussions imply that it is not feasible to provide age constraints from isochronal dating that are as meaningful nor as stringent as those of gyrochronology and chromospheric activity. Thus, we do not include it in our adopted age range, but merely conclude that it is broadly consistent with such an age.

Finally, it can be noted that the existence of a star with such a short rotation period as GJ 504 (∼3 days) would be unprecedented at the Gyr scale ages predicted by some isochrone analyses, such as da Silva et al. (2012). Indeed, although Reiners & Giampapa (2009) found an unusually rapid rotator in the ∼4 Gyr old M67 cluster, it has a rotation of approximately half the solar rotation rate (P_☉ ∼ 30 days), which is significantly slower than the rotation of GJ 504. Aside from known close spectroscopic binaries, it is the largest known outlier at such an age. No indications of chromospheric activity corresponding to such rapid rotation have been found among 60 stars in the M67 cluster (Giampapa et al. 2006), which supports the notion that rapid rotators like GJ 504 must be exceedingly rare at ages of a few Gyr.

As discussed above, we adopt the age estimated with gyrochronology as the most likely age (160$^{+70}_{-60}$ Myr) for GJ 504. However, we conservatively adopt an age of 160$^{+350}_{-60}$ Myr, with chromospheric activity providing the upper bound and gyrochronology providing the lower bound. Age indicators based on the lithium abundance, X-ray luminosity, kinematics, and isochrone technique are less precise than gyrochronology and chromospheric activity: in addition, for the Li-age estimate it is more difficult to quantify the uncertainty. We therefore do not attempt to combine those with the age range estimated from the gyrochronological and chromospheric activity techniques, although their estimates are consistent with our adopted age of GJ 504. Future asteroseismology studies, or detailed kinematic studies identifying GJ 504 as a member of an MG, may be able to refine our age estimate. The estimated ages are summarized in Table 1 and Figure 1.

For the HiCIAO data reduction, we first removed stripe patterns (see Suzuki et al. 2010) emerged on each data frame, and subsequently performed the flat-fielding. After correcting the image distortion, we registered each data set to the centroid of GJ 504 A. For the HiCIAO observations, which used an opaque 04 diameter occulting mask, we estimated the relative frame centers by calculating cross-correlations of the frames as a function of positional offset. The AO188 and ADC kept the PSF extremely stable and well centered. For confirmation, we fit linear and quadratic functions to the frame-to-frame positional offsets and estimated the positional drifts that remained after AO/ADC correction. The systematic drifts were less than 10 mas for all observations except K_s band (17 mas), with residual rms scatters of ∼2–10 mas. These random positional jitters can be attributed to our registration inaccuracies.

We calibrated the absolute centers using a sequence of unsaturated, unmasked exposures taken with an ND filter. For the H-band data, the positional reference was acquired right after the end of the masked sequence. In the cases of the J- and K_s-band observations, we obtained the positional calibration data in a similar way as for the H-band observations, but with the calibration observations interspersed through the science sequence. The positional errors for the unsaturated data ranged from ∼2 to ∼5 mas, where the lower quality PSFs for the H-band data obtained in 2011 March observations or the K_s-band data in 2012 February caused the largest errors. The unsaturated data used a combination of ND filters and field lens that were different from the main science data, possibly introducing a small systematic offset in position. To measure offsets introduced by the insertion of ND filters, we obtained images of a pinhole mask at the focal plane of HiCIAO, with and without ND filters, and measured the offsets. To determine offsets introduced by the use of a different field lens, we examined stars in images of the M5 globular cluster, observed both with and without the mask field lens. We measured these offsets to be small, <∼3 mas. The values of the offsets introduced by the mask field lens were confirmed with three binaries in the SEEDS samples. We did not attempt to remove the offsets, but we have instead considered both of the offsets in the measurements of positions by adding systematic errors of 3 mas to the image registration uncertainties. In total, for H-band observations, image registration conservatively contributed ∼12 mas to our reported positional uncertainties for GJ 504b in March and ∼6 mas in May. Those uncertainties are ∼10 and ∼6 mas in 2012 February (K_s) and April (J), respectively.

We used the Locally Optimized Combination of Images (LOCI) algorithm (Lafrenière et al. 2007) to reduce the HiCIAO data. LOCI constructs a model PSF for each frame using the other frames in that data set as references. The resulting PSF is locally a linear combination of the reference images, with the coefficients determined using a least-squares minimization. To avoid subtracting astrophysical sources, LOCI only uses reference frames with a minimum level of field rotation. For our reductions of GJ 504, we required at least 1.2 (=N_σ) × FWHM/R (× 180/π deg) of field rotation, where the FWHM of PSFs on HiCIAO images are ∼50–60 mas and R is the radial distance from GJ 504. As a result, any physical source was displaced by at least 1.2 FWHM between the working frame and the reference frames. We investigated if the S/N was improved by applying N_σ = 1.5 to the reductions but the S/Ns remained almost the same or were slightly reduced compared with the cases for N_σ = 1.2, and we thus carry out all analyses based on the images reduced with N_σ = 1.2.

For the H-band observations, our analysis of the data from March 26 detected a faint companion at an S/N of 5, where we calculated the noise as the standard deviation of the intensity in concentric annuli about GJ 504. Because of the longer integration time and better observing conditions for the observations in 2011 May, we detected GJ 504b with S/N = 9.0. We detected no other companion candidates (S/N ⩽ 5) in either observation. GJ 504b is clearly confirmed with S/N of 18.4 at J-band, while the S/N of the K_s-band observations is 4.6.

The J, H, and K_s-band images for GJ 504b are shown in Figure 4. Also, the high-contrast two-color composite images is shown in Figure 5, in which GJ 504b is clearly visible 248 northwest of GJ 504.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For the IRCS data, we first corrected flat-fielding. Subsequently, we subtracted the thermal background and then used an algorithm (Galicher et al. 2011) similar to LOCI in order to further reduce the background. We estimated the background value at each pixel and in each frame by dithering the images and using a linear combination of the frames in which the central source, GJ 504, is sufficiently displaced. As in LOCI, we computed the coefficients for this linear combination using a least-squares fit. Eventually, we subtracted the resulting estimate of the local background whose variation is a function of time. As in HiCIAO reductions, we subsequently corrected the distortion of IRCS images.

The IRCS L'-band data were taken without a mask, and were saturated out to a radius of about 3–4 pixels (∼60–80 mas, 0.6–0.8 FWHM, where FWHM is the full width at half-maximum of the PSF). We first performed relative registration by cross-correlating these saturated images with one another. We then centroided a set of unsaturated reference images, and cross-correlated the mean of these unsaturated PSFs with each saturated frame to measure the positional offset. Finally, we applied this average offset to each of the saturated frames. The rms scatter of the offsets was ∼6 mas for both observations in 2011 August, and ∼4 mas for 2012 May. We adopt these scatters as the registration uncertainties in L'-band data.

We then combined these registered frames. We detected GJ 504b with an S/N of 3 and 4.3 in the August 12 and August 15 data, respectively. Because the data for 2012 May 25 were obtained with ADI, we could use LOCI for the PSF subtraction before stacking the registered data: the LOCI reduction parameters were set to be the same as the case of HiCIAO. In the resulting image, the companion was clearly detected with an S/N of 8.0 (see Figure 4).

In the CA model, the core of a giant planet accretes planetesimals and grows to a critical mass (∼10 Earth masses), at which point the core begins to rapidly accrete gas directly from the protoplanetary disk. It is difficult for the CA model to explain the formation of giant planets in situ beyond 30 AU. A recent study (Dodson-Robinson et al. 2009) suggested that even a locally stable (but globally unstable) disk with an order of magnitude higher mass than the minimum mass solar nebula (Hayashi 1981) around a G-type star like GJ 504 was unable, in simulations, to form giant planets beyond 30 AU within the lifetime of the protoplanetary disk (Dodson-Robinson et al. 2009). Therefore, the presence of GJ 504b, as well as other directly imaged planets such as HR 8799 bcde, remains a particular challenge for the conventional CA theory. In this implication, we should note such numerical investigations as conduced by Dodson-Robinson et al. (2009) have typically neglected some theoretical processes that may enable the cores of giant planets to grow in the shorter timescale: effective damping of small planetesimal's eccentricity due to gas drag (Rafikov 2004) or efficient capture of planetesimals due to atmospheric drag (Inaba & Ikoma 2003) may accelerate the accumulations of planetesimals in such the outer region, so that the in situ formation of GJ 504b may become possible (Rafikov 2011). However, even the models incorporating these processes may not necessarily be promising for producing a planet more massive than the critical mass beyond 30 AU at least for disks that are not particularly massive, as demonstrated by the simulations of Kobayashi et al. (2011).

Even if the in situ formation of a giant planet is difficult in such outer regions, models incorporating planet scattering can account for the origin of GJ 504b. If multiple planets are formed by CA in the inner regions, one or more of them could be scattered outward due to dynamical interactions among the planets (Nagasawa et al. 2008) or between the planets and the protoplanetary disk (Paardekooper et al. 2010; Crida et al. 2009). GJ 504b may have been pushed outward by as-yet undiscovered companions. Multiple giant planets may be common in metal-rich protoplanetary disks (Fischer & Valenti 2005), because the high metallicity may enhance solid materials and planetesimals in their protoplanetary disks, resulting in rapid growth of planetary cores. Eventually, outward migration may frequently occur in the metal-rich systems. Hence, the relatively high metallicity of GJ 504 system (see Table. 1) may be a preferential property to explain the origin of GJ 504b under the outward migration hypothesis.

In contrast to CA, the GI model can form massive planets in situ at large radii. In this model, a massive protoplanetary disk becomes gravitationally unstable in its outer regions (>∼50 AU) as the cooling time becomes short relative to the local dynamical timescale (Durisen et al. 2007). The outer disk then fragments, collapsing directly into one or more giant planets. The instability can also occur between 20 and 50 AU, with continuous inflows of gas into the magnetically inactive region inducing fragmentation on a timescale of ∼10³ yr (Machida et al. 2011). However, the planet would have to escape the very rapid inward migration expected in a turbulent disk (Baruteau et al. 2011). In addition, if multiple massive planets are formed in the GI disk, they would still be scattered and eventually ejected from the systems, as in the case of CA. Therefore, GJ 504b would be a planet surviving against falling onto the central star or ejections form the system, if it was formed through the GI process. We should note that fragmentation could also be suppressed in a metal-rich disk like that expected for GJ 504; such a disk would cool less efficiently due to the higher opacities (Cai et al. 2006). Also, the significant gas accretion onto a clump formed according to the GI process may prevent the planet from remaining less massive (Zhu et al. 2012; Boley et al. 2010), disallowing the occurrence of a low-mass planet like GJ 504b.

Based on the available data, we cannot conclusively determine which process led to the formation of GJ 504b. However, it is interesting to note that GJ 504b is the first wide-orbit giant planet detected around a demonstrably metal-rich star. As discussed above, the high metallicity may enhance massive planetary cores, while preventing the effective cooling of a protoplanetary disk, and thus be more hospitable for CA formation than GI formation. Thus, given these measurable properties, planets such as GJ 504b may provide important clues for understanding the formation and evolution of giant planets.

For clarifying the origin of GJ 504b, the further observations may be crucial. Intriguingly, the CA model suggests the presence of unseen massive planets at smaller angular separations (Nagasawa et al. 2008; Chatterjee et al. 2008), making GJ 504 an outstanding candidate for follow-up observations by current and future instruments. N-body simulations for the dynamically unstable gas-free planetary system predict that such inner companions should orbit at a few AU from the host³⁵ and have masses equal or greater than the corresponding outer planets (Nagasawa et al. 2008; Chatterjee et al. 2008). Therefore, the inner counterpart for GJ 504b would have to be more massive than ∼4 M_Jup if it exists, although we note that radial-velocity observations have found few planets in this mass range at semimajor axes of a few AU (Mayor et al. 2011) and the remaining gasses in the disk may affect the scattering of planets, changing the configurations of planets predicted by the simulations for the gas-free cases (Moeckel & Armitage 2012).

Also, further monitoring of the orbit may be important, since scattered planets preferentially have high eccentricities (Chatterjee et al. 2008). Furthermore, modeling the atmosphere of GJ 504b may provide another piece of evidence to test whether it was produced by the CA process. The enhancement of heavy elements in the planetary atmosphere is a natural outcome of the CA process (Chabrier et al. 2007). It is possible to examine the abundances of such elements in the spectral energy distribution of GJ 504b through the use of atmospheric models. For this purpose, multiple wavelength photometric observations as well as the spectroscopy can play an important role.

Further observations of the GJ 504 system will constrain the origin of GJ 504b and may lead to a better understanding of the origin of planetary systems, including our solar system. Future instruments with direct (e.g., GPI, SPHERE, SCExAO) or indirect (e.g., IRD; Tamura et al. 2012) techniques that can explore more inner region may be able to find a promising evidence to clarify the origin of GJ 504b.