[LINK] "Does Gliese 581g exist?"

[rfmcdonald]

Bother. Bad Astronomy's Phil Plait reports on the subject.

Back in September 2010, astronomers announced the discovery of a remarkable and exciting planet: it was three times our mass (high, but far closer to Earth conditions than the super-Jupiters usually found) and orbiting in the "Goldilocks zone" of its star… which meant that it could possibly have liquid water on its surface! This achingly earth-like planet made a major buzz, and in fact I used its characteristics to estimate that there could be billions of Earthlike planets in our galaxy.

But there’s just one small, really eensy-teensy problem: the planet may not exist. But it also might. Maybe.

[. . .]

Almost immediately, the planet was called into doubt; the Swiss team re-examined their data and could not be absolutely certain that Gliese 581 g was there, but still gave it a thumbs-up at the 90+% level. That’s not too bad.

Interestingly, not too long after the announcement I was at a meeting with several astronomers, and one noted that Vogt’s team made a big assumption: all the planet orbits were circular. If in fact one of the planets had an elliptical orbit it could set up a false-positive, making it look like another planet was there when it wasn’t. According to Vogt this turns out not to be the case; I contacted him and he let me know that orbital ellipticity was one of the characteristics they modeled as a variable. In other words, their computer model made no assumptions about orbit shape, but in fact the best fits in the end were circular orbits.

Still and all, there have been some questions about the planet’s existence, and I’ve been holding back from posting until something happened. Well, something did: Philip Gregory, an emeritus astronomer with the University of British Columbia, has analyzed both data sets using sophisticated statistical techniques, and he concluded that Gliese 581 g almost certainly wasn’t real. In fact, he says the odds of it being a false alarm are 99.9978%!

Let me be up front with you: I don’t know. Gregory analyzed the data using Bayesian analysis, a method of looking at the statistical certainty of a set of observations. This is fiendishly complex in practice and to be honest is not something I’m familiar with. However, in his paper, Gregory himself claims that Vogt and Butler underestimated the amount of noise in their data. Vogt disputes this, saying that Gregory adds noise to their data rather arbitrarily. I’ll admit that it seemed odd to me that Gregory would add noise the way he did, but again I’m no expert.

Vogt also notes that how you run the computer model will change whether or not you find the planet. This part interests me, because I’ve run into similar situations myself. If you tell your computer that one of the planets (in this case, Gliese 581 d) has a highly elliptical orbit, then Gliese 581 g disappears: when you calculate the statistics, it’s far more probable that the planet does not exist. But if you keep Gleise 581 d’s orbit circular, Gliese 581 g can be seen in the data. These two different assumptions lead to two different solutions, where one has Gliese 581 g in it and the other doesn’t.

Plait's conclusion? He just can't judge which is the more likely outcome. The mass of controversy and accusations and counter-accusations isn't helping things, either.

Go, read.