My research projects Exoplanet detection from radial-velocity measurements.

Menu Start Research Publications Statistics playground Teaching CV Contact

Background Since the first discovery of an exoplanet in the mid 1990s, several hundred exoplanets have been discovered. There are several different methods to detect such exoplanets, but the vast majority of them has been discovered from radial-velocity measurements of their host star. The search for exoplanets - eventually the search for Earth-like planets potentially hosting life - is of great interest not only for astronomers but also for the general public. What is the problem? Given radial-velocity measurements taken at different times for the same star, one has to infer the right number of exoplanets present in the observed system. That is a case for model comparison. Unfortunately, inappropriate methods are regularly used to estimate the number of exoplanets. One example is reduced chi-square, which only holds for purely linear models. Furthermore, any type of p-values (e.g. from FAPs or KS-tests) are also inappropriate. These methods try to give an absolute answer to the question: "Does the model match the data?" Unfortunately, this is an ill-posed question, since its answer is always "no". There is always a discrepancy between model and data! Therefore, these methods tend to systematically favour complex models over simple models, irrespective of which model is the better one. In fact, these methods will eventually reject any model, if ever more data becomes available. This effect is well known in the statistics community! Consequently, proclaimed exoplanet detections based on these methods are always suspicious and in general not trustworthy. An example: Gliese 581 Gliese 581 is a nearby star hosting a number of exoplanets. The system recently attracted a lot of interest after the detection of a potentially habitable planet was announced. Before September 2009, there were roughly 100-150 observations of Gliese 581, which seemed to suggest the existence of 4 exoplanets. Then, Vogt et al. (2010) published additional observations such that the data set increased to 240. Correspondingly, the number of suggested exoplanets increased from 4 to 6, though this is still under hot debate. In the light of the aforementioned methodological effect, two things are now obvious: The score of exoplanets may have increased because of inappropriate methodology. More data cannot settle this dispute. Using these methods, the number of exoplanets will not converge but will increase without limit, if the number of observations increases without limit. Instead of more data, other methods are required. What am I doing? The crucial point is that the aforementioned effect - though well known in the statistics community - appears to be completely unheard of in the astronomy community. Therefore, I am currently working on simulations designed to demonstrate the devastating impact of this effect in this astrophysical problem. My goal is to make this demonstration as clear as possible but simultaneously as simple as possible.