small planet, about 3 times the mass of the Earth, orbiting within the habitable zone of a red dwarf star a mere 20 light years from us, has been the closest thing yet to a world that we might recognize as a true family member. Announced last year by Vogt et al. it emerged from exquisite spectroscopy data taken over periods of 11 and 4 years with two specialized instruments. It was a hugely exciting discovery, something that we'd all been anticipating as the first of many more such worlds to come.
Soon after this result then further independent analysis by Pepe et al. using additional data seemed to suggest that little GL 581g and its relatively weak signature in the Doppler shifted spectra of this star might just be a figment of overambitious data modeling. More recently an even more sophisticated analysis by Gregory, employing a Bayesian, or probabilistic, approach has also suggested a low likelihood that GL 581g exists. Instead of 6 inner planets orbiting this star there may only be 5, with the small rocky world gone in a puff of statistics. Intriguingly Gregory's analysis also suggests that one of the two datasets, taken with a different spectrograph, may have some as of yet poorly understood systematic problems and a worse accuracy than assumed. This can throw the complex modeling of multiple planets off the scent.
As always. more and better data will ultimately resolve the issue. GL 581g may or may not be gone. Regardless, it is still only a matter of time before similar, real, planets show up elsewhere. This whole event does however raise some very interesting issues that are well worth a little examination. Finding planets with a time-series of ultra-high precision Doppler spectroscopy is not easy. The influence of every planet in a system is simultaneously present, shifting the center-of-mass of the system and therefore the instantaneous orbital velocity of the star about that point. One planet in a circular orbit causes a nice simple sinusoidal variation in the observed stellar velocity. One planet in an elliptical orbit causes a skewed, distorted sinusoidal variation. Two planets cause a superimposed set of variations, with different periods and phases and skewness. By the time there are 5 or 6 planets then the shape of this stellar velocity curve is determined by more than two dozen parameters. Data is never perfect, noise and systematics abound, finding the best model fit to reveal the planets is a very, very significant computational challenge. Whatever objects you think are there must also obey Kepler's laws and form a dynamically stable system. Testing all of that requires care, patience, and even a degree of luck.
The interesting thing is that as our data get better, increasingly sensitive and with longer and longer coverage, the more we are going to run into these issues. We want to find multiple planets in systems. We want to find the low mass guys. But at the same time the more complex the parameter space becomes the more 'local minima' will exist in probability space - the more potentially deceiving 'best fits' that lure us into thinking we have new planets. Over time the sheer bulk of data will offer a resolution, and our tools of analysis will get further refined. However, the next year or two may well be a little bumpy. It should be fun.