The primary science objective of HPF is to discover and confirm low-mass exoplanets in or near the habitable zones of nearby M dwarf stars. The technical challenge of such an endeavor is significant; we must be able to measure sub picometer-level shifts in the spectral lines of our stellar targets, and calibrate out any instrumental effects that would create these shifts. In other words, we are eliminating the “instrumental measurement noise” that prevents less precise spectrographs from detecting Earthlike planets. But not all measurement noise comes from the instrument! We have known for centuries that the Sun is not a perfect, stable object. Sunspots are almost always visible on the solar surface, and the Sun’s 11-year activity cycle has been documented continuously since Galileo’s sunspot observations in the early 1600s. These features–known collectively as “solar activity”–are caused by magnetic fields within the Sun, and can create signals or noise in radial velocity data. At the high levels of velocity precision achieved by HPF and other planet-hunting spectrographs, the RV data may be completely dominated by stellar magnetic activity! In this post, we will explore this problem in detail, using our recent research into the high-profile Gliese 581 planet system as an example.
The Stellar Activity Challenge
Studies of other stars have shown the magnetic phenomena observed in the Sun (spots, prominences, flares, etc.) to be quite typical. There are a number of observational signatures of stellar activity. Measurements of stellar brightness–otherwise known as photometry–reveal small, often semi-periodic variations in brightness as starspots rotate in and out of view. Using spectroscopy, we can monitor the strength of certain stellar emission or absorption lines, which vary as magnetic fields stimulate emission from the atomic species that produce those lines. In addition to the starspot modulation, which changes on the timescale of a star’s rotation, we often observe long-term variability similar to the Sun’s 11-year activity cycle. Astronomers have identified activity cycles as short as about a year, and others that last 20 years or more.
Each type of stellar magnetic activity can alter RV measurements in ways that can either mimic or hide the signal of an exoplanet. As a star rotates, its hemispheres are Doppler shifted as portions of the stellar surface rotate towards or away from our line of sight. In the absence of spots or active regions, these Doppler shifts remain mostly constant (but heed granulation induced noise!) and do not impede planet detection. If, however, one or more spots rotate across the star, it alters the stellar lines as it covers regions of the Doppler-shifted surface. Check out Svetlana Berdyugina’s research page for a more in-depth discussion of this phenomenon. Activity cycles, flares, and large-scale plasma flows can similarly alter stellar line shapes or positions, creating RV signals many times larger than those of Earth-mass planets. As astronomical spectrographs reach velocity precisions of a meter per second and below, these activity-induced RV shifts will begin to represent the primary obstacle for finding and confirming small exoplanets!
A rotating starspot and its effect on the shape of a stellar absorption line (animation courtesy Svetlana Berdyugina).
Exoplanet scientists have long been aware of the possibility that stellar activity can influence RV measurements. It is common practice whenever a new exoplanet discovery is announced for astronomers to check whether the planet signal might actually be caused by activity. A number of methods are used to do so, but the most common is to measure the emission of calcium at the centers of a pair of absorption lines (known as the calcium H and K lines) at the blue edge of the visible spectrum.
The calcium H and K lines have proven quite effective for vetting exoplanet signals around stars like the Sun, but are more difficult to use for the small, cool M dwarf stars that HPF will target. Because their temperatures are so much lower than the Sun, M dwarfs emit very little light at blue wavelengths where the calcium lines lie. This adds measurement noise, and leads to less reliable detections of the magnetic activity signals that might impede planet detection. We have been experimenting with using redder absorption lines such as the sodium D and hydrogen alpha lines for M stars, and are also considering new analysis techniques to disentangle Doppler signals from planets and their host stars.
As a fun illustration of the techniques we are exploring, check out the photograph of the Sun below. Half of the picture is taken with a filter centered on the calcium K line, while the other half is seen through a filter at the hydrogen alpha line. This stunning image was made specially for us by astrophotographer extraordinaire Alan Friedman; we strongly encourage you to visit his website and admire the rest of his fantastic work!
The Sun as seen through calcium K (blue) and hydrogen alpha (red) filters (image courtesy Alan Friedman).
About Gliese 581
Gliese 581 (or GJ 581), an M dwarf about 20 light-years away, might be one of the most well known stars in the history of exoplanets. Beginning in 2005, astronomers using the HARPS spectrograph at the European Southern Observatory (ESO) in Chile discovered a series of progressively smaller planets around the star, eventually finding evidence for four low-mass planets. In 2009, they concluded the outermost of these planets–named GJ 581d–orbited at the outer edge of the star’s habitable zone. With a minimum mass of just 6 times the mass of the Earth, GJ 581d has been considered by many to be the first truly habitable exoplanet ever discovered.
In 2010, another group of astronomers using the HIRES spectrograph at Hawaii’s Keck Observatory announced the discovery of two additional planets (581f and 581g) around Gliese 581. Planet g was believed to have a minimum mass of just 3 times the Earth’s mass and orbit in the very middle of GJ 581′s habitable zone. However, the statistical significance of these planets’ detections was almost immediately questioned by several members of the astronomical community, most notably by the HARPS team. At just over 1 meter per second, the Doppler signals of these two candidate planets lie near the limits of what current technology can measure, and their significance depends greatly on the techniques used to analyze the data. While the discovery of planet f was eventually retracted by the HIRES team, the existence of the tantalizing planet g has thus far remained the subject of intense debate. Planet d has more recently become part of the dispute as well, with one study concluding its detection was less solid than previously believed.
As we studied stellar activity in M stars, we saw evidence that activity might be affecting the radial velocities of Gliese 581. In hopes of potentially settling the debate over planet g, we downloaded the publicly available spectra from HARPS and HIRES and dug into the problem a bit further. Let’s look at the results:
Activity on Gliese 581
Stellar activity has been examined for GJ 581 before, but not much turned up. The calcium emission and X-ray luminosity of the star suggest it is one of the oldest, least active M dwarfs in the HARPS sample. Its brightness is very constant as well, leading previous studies to conclude that the observed velocity signals must be due to planets, since any starspots present must be too small to create a false alarm.
HPF team member and current Penn State Center for Exoplanets and Habitable Worlds postdoctoral fellow Paul Robertson devoted part of his PhD thesis to studying the use of the sodium D and hydrogen alpha ( or H-alpha) absorption lines to characterize activity in M stars. At the time, the HARPS and HIRES spectra used to identify the GJ 581 planets were not publicly available, but data from the HRS spectrograph at HET suggested stellar activity might be influencing the RVs of the star. The data from HRS even suggested activity might have been responsible for the spurious detection of the now-withdrawn planet f in the system. So with the HARPS/HIRES data in hand, we decided to look at the sodium and lines again to see if there were any more surprises lurking, as well as to learn what activity indicators we should consider for HPF
Right away, we learned something new about Gliese 581: its rotation period. Astronomers typically measure a star’s rotation period by measuring the time required for magnetic features such as spots to rotate across the surface, but for GJ 581 its spots are hard to detect with ground-based photometry. Evidently, though, there is some form of magnetic activity that stimulates and sodium emission, because we got a very clear detection of the stellar rotation in both lines. GJ 581 completes a single rotation in 130 days, which is about 5 times as long as the Sun takes. Since the Sun is also about three times as large in diameter, this means GJ 581 is rotating much slower than the Sun, further confirming its status as a stellar senior citizen, since stars rotate more slowly as they age.
A closer look at the line reveals that the measured radial velocities of Gliese 581 are in fact influenced by stellar activity. During periods of low H-alpha emission, the measured RVs tend to be higher than the RVs from periods of high H-alpha emission. This observed anticorrelation between RV and activity suggests that the Doppler shift of the star depends on how active the star is when the measurement is taken.
So what is the net effect of this activity-RV dependence? Is it hiding planets? Is it creating false planet signals? To find out, we designed a technique to correct the RV measurements for the effects of activity. Very simply, we modeled the functional dependence of velocity on the activity, and then subtracted that model from the data. You can get all the gory physics and math in our research article if you are interested in the details. We then started the analysis of the planetary system from scratch with the corrected data.
The signal of the system’s largest planet–the Neptune-mass GJ 581b–was essentially unaffected by our activity correction. This was to be expected; the signal of this planet dominates the variability of the RV data with or without the stellar activity. On the other hand, the signals of the super-Earth planets c and e became much stronger after we applied our correction. For planet e, which is to date the second smallest exoplanet ever discovered with RV, its “false alarm probability” (the odds that a planet’s signal is an illusion created by noise) decreased by a factor of 800! This is tremendously exciting, as it proves the magnetic activity of cool stars can be corrected to reveal the planets hiding beneath the stellar noise.
Unfortunately, this story does not have such a happy ending for the habitable zone planets around Gliese 581. While the signals of GJ 581′s inner planets all increased after the activity correction, the signal of planet d was reduced to the level of measurement noise, which proves it was an activity signal all along. In light of our determination of the star’s rotation period, this does not come as a great surprise. The orbital period of “planet d” is 66 days, or about half the period of the stellar rotation. Rotating starspots produce RV signals at the rotation period and its integer fractions (that is, half the rotation period, one third, one fourth, etc…), as shown by Isabella Boisse’s starspot modeling code SOAP. If you are feeling experimental, you can head to the SOAP website and create your own starspot-induced RV signals using the web-based interface!
Orbital architecture of the Gliese 581. The orbits of the inner planets are shown as solid lines, while those of the false planet signals created by activity are given as dashed lines. The stellar habitable zone is shaded green.
With an orbital period of 33 days, the controversial “planet g” also lies at an integer ratio of the stellar rotation period. Sure enough, no sign of g remains after our activity correction, revealing that it too was an artifact of magnetic activity. While this outcome is certainly disappointing for anyone hoping to find signs of life in the GJ 581 system, it is heartening to finally put the confusion and dispute surrounding this system to rest.
But wait…how can planets d and g be caused by starspots if previous observations showed there was not enough spot coverage on the star to create these signals? This is an interesting question that illustrates how our knowledge of stellar physics is still a work in progress. While a lot of previous research focuses on how magnetic activity can affect the light coming from a star (that is, create dark starspots and bright filaments), less attention has been devoted to how magnetism influences the motion of the gas in the stellar atmosphere. While we are actually not entirely sure yet how the magnetic activity of Gliese 581 created the observed RV signals, our results–and a few other observations of M dwarf stars–suggest magnetic activity in M dwarfs may alter the convective motion of the atmosphere without creating dark spots. This effect can create RV signals in the same way as normal starspots. In the case of GJ 581, the activity created the tantalizing signals of two potentially habitable planets.
The evolution of Gliese 581 and its known planets in period-eccentricity space as a function of time. Blue indicates detections of candidate planets in or near the habitable zone, where liquid water could exist. Orange indicates detections in the too-hot region that is too close to the star. Green indicates detections in the too-cold region farther away from the star and outside the habitable zone. The size of each planet corresponds to its minimum mass. Some simplifications have been made for illustrative purposes. For a complete history of the publications relevant to GJ 581 and its planets see the refereed literature.
Credit: This animation was adapted from a d3 visualization by Mike Bostok.”
What does this all mean?
For Gliese 581, it means we have only found three planets in the system, and probably none of them are currently hosting life. But for HPF, the result is much more exciting and interesting. We want to find planets in the habitable zones of stars like GJ 581, and we are particularly interested in planets close to the mass of the Earth. Many of the stars we will target are younger and more active than GJ 581, so it is likely that magnetic activity will have similarly insidious effects on our RV measurements. Until now, it was unclear whether this activity could be reliably corrected to reveal these high-value exoplanets. As the example of GJ 581 shows, a careful analysis and correction of the activity can both boost the signals of real planets and eliminate false positives. We have a great deal of confidence moving forward that the techniques we have developed here will enable us to find real, habitable exoplanets with HPF and the next generation of upcoming spectrographs!
1. Among planets listed in the Exoplanet Orbit Database as of July 2, 2014