PhysOrgwhere they wrote a story called "

*String Theory Gets a Boost*" about a preprint that was recently accepted to

Phys. Rev. Lett.even though careful readers of the arXiv have known it nearly for a year as

astro-ph/0702223.Quite a long time for the "speedy" PRL. In the paper that has collected 25 citations so far, Neil Bevis, Mark Hindmarsh, Martin Kunz, and Jon Urrestilla statistically investigate the cosmic microwave background. They try to parameterize it by two models. One of them is based on ordinary inflation - what matters is the scale-invariant spectrum - with an adjustable power law tilt. The other has cosmic strings included.

By looking at the l=10 spherical harmonics, they argue that the relative contribution f_{10} of the cosmic strings is optimized for fitting the data at f_{10}=0.11 plus minus 0.05. So it is not zero and the strength of this statement is approximately two sigma. Well, that's not a terribly strong signal but it is justifiably enough for some people to find it intriguing.

I am somewhat skeptical about this kind of an argument because it reminds me of various "proofs" of anthropogenic global warming: you can't match the curves with the first naive natural model you write down and if you add men to the naive model, you do better. Well, it is not too surprising. Two-sigma signals are guaranteed to be almost everywhere and a model with additional parameters - (almost) whatever they are - is guaranteed to fit the data more accurately than a more robust and simple model. Of course, if this were a 5-sigma signal, I would be more afraid to make such a statement but with 2 sigma, I have enough courage to do it. ;-)

The work is surely interesting but the results so far are uncertain enough to allow me to stick to my subjective and purely theoretical 15% probability estimate that cosmic strings exist and will be reliably observed (or produced) by 2100.

What I would find more convincing would be if a cosmic-string model were able to fit the data better than a cosmic-string-free model with the same number of parameters. For example, if you showed that a model with a fraction of cosmic strings and a fixed tilt is more accurate than a model with an adjustable tilt and its time derivative (or scale derivative) or whatever new additional but "conventional" parameter is useful to reduce the errors.

Couldn't this become a standard technique - in all scientific disciplines - to decide about the relevance of a very new effect previously unused to match the data?

## No comments:

## Post a Comment