The Feasibility of Directly Imaging Nearby Cold Jovian Planets with MIRI/JWST

Direct imaging is a powerful technique for exoplanet detection and characterization. Planets with wide orbital separations are difficult to detect with time-series methods such as transit photometry and radial velocity measurements. However, direct imaging can detect planets with sufficient brightness and orbital separations more quickly than other methods and, depending on the capabilities of the instrument, over more of the planet’s orbit. Unfortunately, imaging planets requires very high contrast imaging capabilities, which make direct imaging very challenging for all but the most favorable targets. Ground-based, adaptive optics–equipped, large-diameter telescopes and high-contrast coronagraphs are the main drivers of direct-imaging exoplanet detection, accounting for most of the directly imaged exoplanets currently known. These include notable discoveries such as the four planets orbiting HR 8799 (Marois et al. 2008, 2010), Beta Pictoris b (Lagrange et al. 2009), HD 95086 b (Rameau et al. 2013a, 2013b), HIP 65426 b (Chauvin et al. 2017), 51 Eridani b (Macintosh et al. 2015), and the first directly imaged protoplanet, PDS 70 b (Keppler et al. 2018). Space-based direct imaging has also been successfully demonstrated; planets have been directly imaged with the Spitzer Space Telescope (HN Pegasi b, FU Tauri b, WD 086-661 b; Luhman et al. 2007, 2009, 2011) and the Hubble Space Telescope (CXHR 73 b, Fomalhaut b, 2MASS J04414489+2301513 b; Luhman et al. 2006; Kalas et al. 2008; Todorov et al. 2010). As of 2019 March, where mass and semimajor axis data are available in the Exoplanet Archive (Akeson et al. 2013), all directly imaged exoplanets are more massive than Jupiter, almost all are on orbits wider than 10 au, and almost all are young planets that still retain much of the internal heat from their formation.

Current ground-based high-contrast imaging instruments have poor sensitivity at the wavelengths needed to probe older, cooler, fainter planets, and as such, they are limited to detecting planetary-mass companions around younger stars where the planets are warmer and brighter at shorter wavelengths. These younger host stars are typically found further away, and when combined with the sensitivity limits, they generally lead to detections of wider-orbit, more massive planets. Existing space-based imagers are similarly insensitive to cooler targets and more limited to the red optical/near-infrared bands. As such, upcoming space-based mid-infrared direct imaging has the potential to focus on older and closer stars and allow the detection of cooler, fainter, lower-mass planetary companions at smaller projected separations.

The upcoming James Webb Space Telescope (JWST) will be highly capable of high-contrast imaging. Three instruments on the spacecraft are imaging-capable: the Near-infrared Camera (NIRCam), the Near-infrared Imager and Slitless Spectrograph (NIRISS), and the Mid-infrared Instrument (MIRI). NIRCam and MIRI are equipped with coronagraphs, while NIRISS is capable of high-contrast imaging through its Aperture Masking Interferometry imaging mode.

1.1. Planned JWST Observations

1.2. Demographics of Giant Exoplanets around Low-mass Stars

While there is strong evidence that giant planets are rarer than smaller planets (Burke et al. 2015) and less common around M dwarfs than FGK dwarfs (Mulders et al. 2015), many giant planets around M dwarfs have been found. In Figure 1, we show known exoplanets in the left panel and planets orbiting cool stars in the right panel, demonstrating the sample of giant planets around cool stars. Some notable Jupiter-sized planets orbiting M dwarfs are Kepler-45 b (Johnson et al. 2012), HATS-6 b (Hartman et al. 2015), NGTS-1 b (Bayliss et al. 2018), and HATS-71 b (Bakos et al. 2018). Radial velocity observations have detected over a dozen giant planets around M dwarfs (Butler et al. 2006; Johnson et al. 2007, 2010, 2012; Bailey et al. 2009; Apps et al. 2010; Haghighipour et al. 2010; Howard et al. 2010; Forveille et al. 2011; Robertson et al. 2013; Hartman et al. 2015; Bryan et al. 2016; Sahlmann et al. 2016; Bakos et al. 2018; Bayliss et al. 2018; Trifonov et al. 2018). Microlensing observations are typically sensitive to planets orbiting between 1 and 10 au from their host stars, and due to the statistical dominance of M dwarfs in the galaxy, most microlensing-detected planets orbit low-mass stars (Zhu et al. 2017). Early results from microlensing surveys, such as Cassan et al. (2012), found a giant planet occurrence rate of for planets with masses 0.3–10 M_J at distances between 0.5 and 10 au. More recent microlensing results further reveal that giant planets are relatively common, although the planet mass function is steep (Shvartzvald et al. 2016; Suzuki et al. 2016; Mróz et al. 2017), with many more Neptune-mass planets than Jupiter-mass planets and tentative signs of a break between the ice giant (∼15 M_⊕) and super-Earth (∼5 M_⊕) mass regimes (Suzuki et al. 2016; Udalski et al. 2018).

Zoom In Zoom Out Reset image size

Combining RV and direct-imaging observations, Montet et al. (2014) estimated that 6.5% ± 3.0% of M dwarfs host a 1–13 M_J planet between 0 and 20 au, while Clanton & Gaudi (2016) and Meyer et al. (2018) found results consistent with previous microlensing studies, and that giant planets are more common at wider separations around low-mass stars. These results paint a similar picture for M dwarfs as for FGK dwarfs, where of all types of stars host giant planets between 3 and 7 au (Wittenmyer et al. 2016). Beyond about 10 au, planets are scarce; Lannier et al. (2016) found that the planetary-mass companion frequency was between 8 and 400 au, which is consistent with the findings from Bowler (2016) that <3.9% of M stars host a 5–13 M_J planet between 30 and 300 au. While giant planets are relatively uncommon around M dwarfs, they do exist a few au from their stars at an occurrence rate in the range of 5%–20%, which motivates us to search for companions, and a sufficiently sensitive imaging survey of the nearest M dwarfs to Earth has the potential to yield a handful of such planets.

1.3. The MIRI Instrument and Direct Imaging

Our primary interest is detecting field-aged planets around the nearest stars. These planets are likely to have temperatures of 100–300 K, similar to the giant planets in the solar system, corresponding to peak blackbody wavelengths of 10–29 μm. The MIRI instrument on JWST covers 4.9–28.8 μm (see Figure 2 and Rieke et al. 2015), which is well suited to our parameter space of interest. MIRI has four coronagraphs: three are four-quadrant phase masks optimized for use at 10.65, 11.4, and 15.5 μm and have small inner working angles (IWAs) of 033, 036, and 049 (equal to the 1 λ/D limit), respectively. The fourth is a traditional Lyot spot for use at 23 μm, which we do not consider due to its hard stop and large IWA of 216 (Boccaletti et al. 2015).

For the three four-quadrant phase mask coronagraphs, the IWA is defined as the 50% transmission radius. Our simulations (see Section 3) reveal that for particularly suitable exoplanets at projected separations inside the IWA, the signal-to-noise ratio (S/N) could indicate a photometric detection, although the planet will not be spatially separated from the host star. Further discussion regarding the feasibility of sub-IWA planet detections is provided in Appendix A.

Table 1 outlines the performance of the three coronagraphs. For the 10.65 μm coronagraph with an IWA of 033, a planet orbiting at 1 au falls within the IWA at host star distances beyond 3.03 pc. However, even planets 5 au from their host stars will be within the IWA of the instrument if they are further than 15 pc.

Table 1.  Limiting Target Distances for Each MIRI Four-quadrant Phase Mask Coronagraph

Filter	IWA	1 au	2 au	5 au
f1065c	033	3.03 pc	6.06 pc	15.15 pc
f1140c	036	2.7 pc	5.5 pc	13.8 pc
f1550c	049	2.04 pc	4.08 pc	10.2 pc

Note. The system distances indicate where the projected separation is equivalent to the IWA for each filter.

Download table as:  ASCIITypeset image

Direct-imaging measurements require the detection of faint planetary sources around much brighter stars. While previous studies typically concentrated on planets that still retain heat from their formation, the goal of this investigation is to concentrate on cold planets. This naturally leads us to nearby and intrinsically dim stars where the projected separation is widest and the planet-to-star brightness ratio is most favorable.

We selected a population of stars from the RECONS list of the 100 nearest star systems to Earth.⁸ Of the nearest M dwarf stellar systems, we excluded those in tight binaries due to the observational and dynamical complications that would produce. Proxima Centauri was included despite its presence as the third component of the Alpha Centauri system because of its extremely wide separation from Alpha Centauri A and B. Although Proxima Centauri is not known to have any giant exoplanets and is unlikely to have any undiscovered wide-orbit giant exoplanets, it makes for a good test calculation. We selected 27 stars in the RECONS list that fit our criteria. These stars and their properties are listed in Table 2.

To use these 27 stars as simulated exoplanet hosts, we needed to compile their stellar parameters. The parameters were found, where possible, by using the default parameters on SIMBAD (Wenger et al. 2000) and NASA’s Exoplanet Archive (Akeson et al. 2013). When the parameters were not found in SIMBAD or the Exoplanet Archive, they were found from individual references in the literature or estimated from comparisons to other similar stars in our sample. Although not every parameter was found, comparisons with similar systems show that the calculations from those systems are similar enough to not invalidate the general results. Table 2 shows the full set of stellar parameters we used to create our sample of targets.

Zoom In Zoom Out Reset image size

We used NASA Goddard’s Planetary Spectrum Generator (PSG; Villanueva et al. 2018), an online radiative-transfer tool, to synthesize planetary spectra given our various stellar, planetary, and observational parameters. We modeled our simulated planets as mature Jovian-type planets much like Jupiter and Saturn in our own solar system. The PSG’s built-in atmospheric templates are available for many solar system objects, such as Jupiter and Saturn, as well as a number of generic giant gaseous exoplanet models. Observations of Jupiter and Saturn show variations in cloud cover and atmospheric scattering that have significant effects on planetary infrared emission. Assuming that clouds uniformly block emission from lower layers in the planetary atmosphere, the projected area of the planet’s disk shows differences in brightness depending on whether or not cloud bands are present. Figure 3 shows the effect of varying cloud bands on Jupiter’s brightness in 4.8 μm as seen from Gemini North using the Near-infrared Imager instrument. Since PSG’s atmospheric templates produce one-dimensional planetary spectra, clouds are represented as a single layer across the entirety of the spectrum, attenuating it as appropriate for the particular molecular species present in the cloud. To emulate the variations in cloud cover that would partially lower the aggregate brightness of the planet’s disk, we produced three model spectra: (1) with attenuation due to clouds and hazes, (2) with no clouds or hazes, and (3) a combined model simulating partial cloud cover. We assumed a somewhat conservative estimate that a Jovian-type planet is likely to have ∼90% cloud cover over its disk at any point in time. As such, our combined spectrum is a linear sum made up of 10% of the cloud-free model and 90% of the cloudy model. This produced a combined spectrum representing a planet with an intermediate brightness to the full- and no-cloud planets. Figure 4 shows the synthesized planetary spectra used in this work, as well as overlaid MIRI coronagraphic bandpasses. Given the planetary spectra of Saturn and Jupiter, MIRI’s bandpasses from 10 to 16 μm give us the best chance of detecting planets where they are most luminous.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

2.1. JWST Exposure Simulations

A suite of software has been developed by the Space Telescope Science Institute for planning observations with JWST. The JWST Exposure Time Calculator (ETC) is available as both a web interface and a Python library, Pandeia (Pontoppidan et al. 2016), enabling simulated observations from any of JWST’s observing modes. Pandeia incorporates the most current performance estimates of the JWST instruments, including MIRI, as described in the JWST MIRI special issue of the Publications of the Astronomical Society of the Pacific, of which the relevant papers to this work are Boccaletti et al. (2015) and Rieke et al. (2015). In the course of this investigation, we use the Pandeia/ETC MIRI coronagraphic observing mode. For the simulations performed here, the Pandeia coronagraphic imaging mode supported a specific subset of coronagraphic operations, outlined in the JWST User Documentation (STScI 2016). The relevant specifications are as follows.

1. 1.  
   Pandeia supports only point-spread function (PSF) reference differential imaging (STScI 2017a).
2. 2.  
   The PSF reference observations center the reference star in the coronagraphic mask (STScI 2017a).
3. 3.  
   For PSF reference subtraction, the user must specify a PSF reference star close in magnitude and stellar type to the science target (STScI 2017b).
4. 4.  
   Pandeia assumes no difference in the thermal/dynamical state of the observatory between PSF reference observations and science observations (STScI 2017b).

Without data from the observatory as it performs in flight, the simulations performed here provide upper limits on planet detectability. Once JWST is launched and commissioning/ERS observations have been conducted, there will be sufficient information to update the ETC and other JWST proposal tools with detailed capabilities to simulate actual observatory and instrument performance.⁹

Our goal is to sample a wide potential parameter space of giant nearby exoplanets. Our full set of simulations of a planetary system consists of simulations performed at three orbital distances (1, 2, and 5 au) in three separate filters (10.65, 11.4, and 15.5 μm) for three different atmospheric configurations (cloud-free, patchy cloud, and full cloud cover) and bracketed across a range of exposure times (898–22,470 s). Although we use solar system Jupiter and Saturn spectra for our planetary atmosphere models, our orbital modeling is idealized to represent the best-case configurations for direct imaging. Our planets were simulated to be on circular orbits, with no inclination to the line of sight, and always situated at the maximum projected separation from the host star (phase 90°). This allows us to directly correlate the orbital distance of our synthetic targets with projected angular separation. As coronagraphic performance strongly depends on the angular separation of targets and their host stars, tying orbital distance and angular separation is a useful strategy for investigating favorable planetary targets.

To perform simulations on a full set of planets, expanding along each desired parameter and investigating such planets around a large sample of stellar systems, we elected to use Pandeia. This allowed us to programmatically construct large sets of inputs and parallelize the execution of those simulations. In total, we performed over 17,000 individual simulations (see Figure 5), so having access to such parallelized simulations was critical.

Zoom In Zoom Out Reset image size

The basic Pandeia input is a JavaScript Object Notation (JSON) file that defines a hierarchical data structure containing information describing the state of the telescope, astronomical scene being observed, observing strategy, astronomical background, and other relevant observational data. This hierarchical structure simplifies the process of accessing and modifying data in a particular input file by allowing each field to be directly referenced and changed. If each stellar system is treated as a tree-based structure, a depth-first traversal of nodes in the tree produces a single sequence of parameters, and a path down to each leaf represents a unique, fully populated input file. Therefore, looping through a list of every star system to be observed can produce a full set of ETC inputs. A section of such an input tree, describing a single star’s worth of parameters, is shown in Figure 5.

Pandeia requires that astronomical sources be described by either built-in spectral models or user-defined spectra. Our stellar sources can be appropriately described using the available PHOENIX stellar models, which Pandeia uses as inputs to the WebbPSF tool to create a stellar PSF library, and for the planetary sources we used in the previously created PSG spectra. These spectra can be directly placed into the corresponding Pandeia input file for use in simulating a JWST observation. Once the input file is complete, Pandeia’s perform_calculation function takes the input file, runs the observation simulation, and returns a new JSON object containing the results of the calculation, including total exposure time and S/N.

The GSFC’s Goddard Private Cloud computing resource was used to perform the large set of simulations. Because each simulation is independent of the others, no complicated communication or synchronization is required, and the simplest solution is to run a number of multithreaded virtual machines concurrently to perform as many simulations as possible. All of the inputs were created, populated, and stored in a single backing database. Each virtual machine would check the database for an unstarted simulation, pull out that specific input structure, and then start a simulation using the selected input. When a simulation finished execution, the machine would insert the calculated S/N and the exposure time back into the database for that specific simulation. Each simulation could then be referenced by a unique identifier to immediately find the results.

We ran our simulations for each star/planet/orbit/filter combination and created an individual S/N–exposure time plot. We investigated exposure time by trying to minimize saturation in each up-the-ramp exposure and then bracketing between a number of exposures. This yielded a minimum integration time of 898.8 s and a maximum of 6.24 hr. For each combination, we plotted three S/N–exposure time curves, one for each atmospheric configuration. The S/N–exposure time curves were found by interpolating between the discrete points using a quadratic approximation of the S/N behavior. We classed a planet as detectable if the S/N was above 4 at the maximum integration duration of 6.24 hr. Figures 6 and 7 show an example selection of results for one star in our sample, Proxima Centauri.

Zoom In Zoom Out Reset image size

Proxima Centauri is the closest star to Earth and on the lower end of the temperature range of our sample. As such, the sample planets around Proxima Centauri are the highest-S/N detections we have. Figure 6 shows Jupiters around Proxima Centauri, and we find that only the 1 au planets take any significant time to be able to detect. Our worst-case observation is Jupiter at 1 au in the 10.65 μm filter, and it is detectable for all cloud configurations within about 3.5 hr. Figure 7 shows that for Saturns, things are somewhat more challenging, with only the brightest planets detectable within our maximum exposure time of 6.24 hr. No Saturn is actually detectable at all atmospheric configurations at our maximum exposure time; the 6.24 hr, 15.5 μm, cloudy Saturn misses a 4 S/N detection by 0.05.

Zoom In Zoom Out Reset image size

Figures 8 and 9 show how planet detectability changes with system distance. Each point plotted in these two figures represents the max-S/N longest-duration exposure (6.24 hr) observation for that specific planet. Open circles represent planets within the IWA for that particular coronagraph. Ordinarily, it would be difficult to make any sort of determination about potential detections within the IWA. However, some of our sub-IWA simulated planets produce detections near S/N = 10, particularly for Jupiters in 15.50 μm at 1 and 2 au. In Figures 8 and 9, drops in S/N due to the IWA can be clearly seen.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Each star in our sample has at least one detectable planet configuration, for a total of 300 detectable planets. Of these planets, 222 are Jupiters, and 78 are Saturns. Seventy-seven planets were detectable in the 10.65 μm filter, 88 in 11.4 μm, and 135 in 15.5 μm. There are 199 cloudless planets, 66 have patchy clouds, and 35 are fully clouded. Fifty-six planets are detectable at 1 au, 87 are detectable at 2 au, and 157 are detectable at 5 au. For the most favorable system configuration (cloud-free Jupiters at 5 au in 15.5 μm), we can detect at least one planet within our maximum 6.24 hr exposure time around every star in our sample. As shown in Figures 8 and 9, system distance is the most fundamental factor affecting planet detectability. We also find that some simulated planets may be detectable inside the IWAs of the MIRI coronagraphs. More details are provided in Appendix A.

As shown in Figure 8, MIRI can be used to detect completely cloud-covered Jupiters around stars within 3 pc, partially cloudy planets around stars within 5 pc, and cloud-free well beyond 6.4 pc. Saturns are more challenging, as we are unable to detect any fully clouded Saturns and can only detect partially clouded Saturns around Proxima Centauri. However, cloud-free Saturns can be detected around stars out to 4 and 5 pc. Because our simulations only investigated planets within 5 au of their host stars, these results present upper bounds on the exposure times needed to detect planets further than 5 au. Sections 4 and 5 of Boccaletti et al. (2015) show the contrast curves and simulated system performance (Figures 9 and 10), as well as some simulated MIRI coronagraphic observations of Jovian-type exoplanets and the HR 8799 system. As the simulated exoplanetary targets (especially in the HR 8799 case) move further away from the center of the coronagraphic image, they become more detectable, as they are less attenuated by the coronagraph. Following this, if we observe a system and target a particular planet, say, a Jupiter at 1 au, we are necessarily sensitive to potentially undiscovered outer companions, provided their projected separation from the star is large enough. This full parameter space where JWST MIRI is sensitive to more massive and widely separated planets is shown as the dashed box in Figure 1. More detailed maximum S/N–exposure time information is available in Table 4 in Appendix B, and the full maximum S/N–exposure timetable is available in machine-readable format.

Zoom In Zoom Out Reset image size

4.1. Known Potential Targets: GJ 832 b

One of the stars in our sample population, GJ 832, is known to have a Jovian exoplanet (Bailey et al. 2009) with a minimum mass of ∼0.68 M_J at an orbital distance of ∼3.56 au (Wittenmyer et al. 2014). The planet GJ 832 b was discovered through radial velocity measurements taken over a 10 yr period and is not known to transit its host star. Given its minimum mass and wide maximum angular separation (∼069), it should be a tempting target for direct detection efforts like ours. The inclination of the system is unknown, but by assuming 45° inclination and estimating the system properties (i.e., nearly circular, with projected orbital radius ∼069), we derive a minimum separation close to the projected boundaries of the IWA of the 15 μm coronagraph. Detecting GJ 832 b with this instrument would therefore be challenging or impossible for about half of GJ 832 b’s ∼3600 day orbit. However, given the planet only has a mass lower bound of M sin(i), if its true orbital inclination is close to 45°, its true mass could be closer to ∼1 M_J. To some extent, this would make GJ 832 b a slightly better imaging target for the portions of the orbit where it is observable.

We simulated the detectability of this system following the process used for the rest of our simulations. We produced a set of PSG spectra (shown in Figure 10) and Pandeia inputs using the known parameters of the planet and substituted modeled planetary parameters for GJ 832 b where needed. Since we only have an estimated mass for GJ 832 b, we used the forecaster Python package (Chen & Kipping 2017) to estimate the radius of the planet. For each mass (0.68 and 1.0 M_J), the mean predicted radii were both ∼1.25 R_J. Using Jupiter-type atmospheres again, we created spectra for both masses, placed them into the relevant Pandeia inputs, and simulated JWST observations using MIRI in the 15.5 μm filter.

As in previous simulations, we found that our potential detections are split between the brighter and dimmer atmospheric configurations. We found that for both the 0.68 and 1 M_J cases, we were only able to detect the planets if they did not have significant cloud cover. Due to the significant gaps in our understanding of the system, we feel comfortable in designating GJ 832 b a potentially detectable planet in the 15.5 μm filter. Figure 11 shows S/N versus exposure time curves for our GJ 832 b simulations.

Zoom In Zoom Out Reset image size

As of now, GJ 832 b is the only known potential target in our sample of nearby M stars. However, given the estimates on the occurrence rates of Jovian-type planets around M dwarfs, there may be another undiscovered planet or two within our stellar sample. Even if this is not the case, potentially imaging GJ 832 b would be an unprecedented feat, which could lead to a new era of detecting and potentially characterizing long-period giant exoplanets.

4.1.1. MCMC Orbital Modeling

Since GJ 832 b is only known from radial velocity data, and those data are only complete up until 2010, we decided to predict when the planet would be observable in the future by JWST. We used the Monte Carlo simulation tools in the exoplanet (Foreman-Mackey 2018) Python package to model the orbit of GJ 832 b given some of the previously determined orbital parameters from past work. Projecting the orbit into the future, we found that GJ 832 b would be most observable near 2025 September (JD 2,460,930). The results of the orbital modeling are shown in Figure 12, along with the predicted maximum visibility windows. However, this analysis is not quite sufficient, since potential exoplanetary targets are liable to be outside JWST’s visibility zones for some portion of the year. Nearby stars may also affect the quality of JWST observations. Therefore, we also evaluated GJ 832 b’s visibility and potential contamination using the STScI’s Exoplanet Characterization Toolkit’s Contamination and Visibility Calculator (ExoCTK). According to the ExoCTK output, the system is observable without background stellar contamination from May to July between JWST observatory V3 position angles of about 240° and 290°. There is also a period of visibility from 2025 September to November at lower position angles of 30°–70°, but that window has moderate background contamination. Comparing this with STScI’s JWST Coronagraph Visibility Tool, we find that GJ 832 would be visible in quadrants 1, 2, and 4 of MIRI’s 15.5 μm coronagraph across the feasible range of position angles. Ideally, we would also observe this system in 2024 December (JD 2,460,661) and 2026 June (JD 2,461,199) to further constrain the planet’s orbit with additional observations bracketing the point of maximum projected separation, about a year before and after. Due to the restrictions from the spacecraft’s orbit and stellar background, we instead suggest that the three characterizing observations be made near 2024, 2025, and 2026 June (JD 2,460,469, 2,460,834, and 2,461,199).

Zoom In Zoom Out Reset image size

4.2. Synergy with Gaia

4.3. Example Observing Programs

We present an investigation into the feasibility of JWST’s MIRI for the direct imaging of cold Jovian-type planets around nearby M dwarfs. Any planets directly imaged in this sample would be the nearest to Earth, as well as the lowest-mass and closest-orbiting of any giant planets yet directly imaged. MIRI’s sensitivity to these 0.3–1 M_Jup planets at projected separations of 1–5 au also implies that it would be sensitive to similar (or more massive) planets at wider separations where there are more favorable contrast ratios (see Figure 1). In addition, more detailed simulations of the known planet GJ 832 b reveal that it could potentially be recovered with MIRI imaging. It would be the second-closest and likely the least massive exoplanet yet to be imaged.

In order to maximize the chances of such detections, we suggest that future observing programs utilize MIRI’s 15.5 μm coronagraphic filter. Although the 15.5 μm coronagraph has the largest IWA of MIRI’s four-quadrant phase mask coronagraphs, the slightly reduced sample size is worth the increase in S/N associated with searching for planets with peak emission nearer 15 μm. However, if a planet does lie within the coronagraph’s IWA, it may still be detectable, provided it is sufficiently bright (see Appendix A). This allows for some flexibility in the sample of stellar systems used as potential targets, especially if any new long-period Jovian-mass planets are discovered before JWST’s launch or early in the mission. Given that in most cases, we can detect our desired planets within 6.24 hr of observing time (excluding telescope operations overhead), a full survey of all 27 stars in our sample would be possible in under 170 hr of observing time for one filter.

As this work has demonstrated the feasibility of using the MIRI coronagraphic imaging modes to detect cold Jovian planets around some of the nearest M dwarfs, it naturally motivates future work on a full survey simulation. Such an expanded simulation would take into account an input planet population rooted in measured occurrence rates, more varied planet parameters, more accurate stellar spectra and PSF modeling, randomized orbits, and detailed estimates of telescope and coronagraphic inefficiencies. The goal would be to obtain realistic completeness estimates and investigate the full sensitivity parameter space for a larger stellar host sample.

In the years after launch, JWST coronagraphic imaging will also be supplemented by the Wide-Field Infrared Survey Telescope (WFIRST), whose microlensing survey will be extremely sensitive to Jovian-mass and smaller planets at intermediate orbital distances of 1–10 au (see Figure 9 in Penny et al. 2019). It will also be capable of high-contrast coronagraphic imaging from 575 to 825 nm at IWAs down to 015 (Mennesson et al. 2018). Jovian planet statistics from the microlensing sample could inform future JWST imaging campaigns, and WFIRST direct imaging will usher in a new era of exoplanet characterization through reflected-light spectroscopy.

Further in the future, the Astro2020 Decadal Survey large mission concept proposals contain a number of observatory concepts capable of high-contrast coronagraphy and direct imaging. The Large UV/Optical/IR Surveyor (LUVOIR) and the Habitable Exoplanet Observatory (HabEx) are intended to directly image differing populations of exoplanets in varied UV, optical, and infrared bands. LUVOIR’s Extreme Coronagraph for Living Planetary Systems (ECLIPS) will be capable of coronagraphic imaging in the near-UV, optical, and near-IR (0.2–2 μm; The LUVOIR Study Team 2019). HabEx will be capable of coronagraphy in the optical and near-IR (0.45–1.8 μm; The HabEx Study Team 2019). These missions are intended to discover and characterize exo-Earth-type planets and represent the next great frontier in space-based direct imaging.

The authors would like to thank Michael Moore and the rest of the Goddard Private Cloud team for their support and assistance. This research has made use of NASA’s Astrophysics Data System Bibliographic Services; the NASA Exoplanet Archive, which is operated by the California Institute of Technology, under contract with the National Aeronautics and Space Administration under the Exoplanet Exploration Program; the SIMBAD database, operated at CDS, Strasbourg, France; and the RECONS 10 Parsec Sample, compiled by Georgia State University, Atlanta, GA, USA (http://www.astro.gsu.edu/RECONS/index.htm). This work was supported by the Sellers Exoplanet Environments Collaboration (SEEC) at NASA’s Goddard Space Flight Center. We would like to thank Michael McElwain for enlightening discussions that greatly improved this manuscript. We also thank the anonymous reviewer for helpful feedback.

This research was conducted on the historical territory of the Piscataway people.

Facility: Exoplanet Archive. -

Software: Matplotlib (Hunter 2007), NumPy (Oliphant 2006), Pandeia (Pontoppidan et al. 2016), forecaster (Chen & Kipping 2017), exoplanet (Foreman-Mackey 2018), Astropy (Astropy Collaboration et al. 2013, 2018), Theano (Theano Development Team 2016), PyMC3 (Salvatier et al. 2016).

While the IWA is defined as the 50% transmission radius of the MIRI coronagraphs, the shape of the radial transmission curves inside the IWA determines whether sub-IWA imaging is feasible at all. Figure 6 in Boccaletti et al. (2015) shows the radial transmission curves for both the four-quadrant phase mask and Lyot MIRI coronagraphs. For the Lyot coronagraph, the radial transmission curve falls extremely steeply from near 90% transmission at ∼4λ/D, to 50% transmission at ∼3λ/D, 10% at ∼2.75λ/D, and 5% at ∼2λ/D. As the IWA of the Lyot coronagraph is 216, this makes recovery of targets effectively impossible within ∼2″ of the center of the image. For the four-quadrant phase mask coronagraphs, the radial transmission curve is not as steep as the Lyot coronagraphs. Here 100% relative transmission occurs at ∼3λ/D, 50% at ∼1λ/D, and 20% at ∼0.5λ/D. Given the sub-IWA transmission curves, we adopt the 0.5λ/D 20% transmission limit as an inner bound for where sub-IWA detections are possible. For each of the 10.65, 11.4, and 15.5 μm coronagraphs, this 0.5λ/D limit corresponds to 0165, 018, and 0245, respectively. Given that we have simulated cold gas giants at relatively small projected separations (1, 2, and 5 au) around the nearest stars, some are likely to fall within the IWAs of the coronagraphs. We expect that planets in this regime will be detectable provided they are bright enough to overcome the sub-IWA attenuation by the coronagraphs.

In addition, because the PSF of the planetary companion is likely to be entirely encompassed by the stellar host’s PSF, signal from the planet can only be recovered when the PSF subtraction does not entirely zero out the pixels containing the image of the planet. As such, in the course of such sub-IWA observations, care must be taken to prevent the camera from being saturated in both the target image and the PSF reference image. Otherwise, the standard coronagraphic imaging process and reference PSF subtraction can be used to recover sub-IWA planets, provided they are bright enough to overcome the steep attenuation of the coronagraph inside this region.

In our suite of simulations, 33 unique planet configurations are detectable inside the IWA of the MIRI coronagraphs. These are all Jupiter-type planets with no clouds and were only observable with the 15.5 μm coronagraph. Since our planetary atmospheres are brightest at 15.5 μm without clouds, we do expect that these planets would be most detectable at this wavelength. Also, as the 15.5 μm coronagraph has the widest IWA, we do expect that there would be a certain range of stellar host distances where planets with smaller orbital distances would lie within the IWA and the wider orbital distances may be outside the IWA. In these cases, planets are detectable around 23 of our stellar hosts, ranging from 2.97 to 6.4 pc away.

Further coronagraphic imaging techniques, like small-grid dithering (Lajoie et al. 2016), are available to optimize the PSF subtraction. In situations where the PSF reference target acquisition is uncertain, the small-grid dither technique can significantly improve PSF reference subtraction performance. This adds some time expense to JWST coronagraphic operations, as the reference star needs to be more thoroughly observed. The JWST commissioning and preliminary science observations may show that these techniques will be necessary for imaging challenging coronagraphic targets, and sub-IWA targets in particular. Although JWST in-flight pointing and thermal performance is not yet known, current performance requirements indicate that some of the more marginally detected targets, especially the aforementioned 33 sub-IWA planets, may be rendered undetectable. Future work will be needed to better quantify this and should incorporate the updated PanCAKE package and less optimistic image post-processing, such as the inclusion of pointing drift between science targets and PSF references.

Table 4 below shows planet S/N at maximum exposure time for planets around Proxima Centauri.