jump to navigation

Your opinion on a suggestive excess of same-sign lepton pairs March 17, 2008

Posted by dorigo in games, news, physics, science.

A hot new result from CDF has emerged in a global analysis of Run II data by an automated search called VISTA-SLEUTH. I intend to discuss it tomorrow, but I thought I would in the meantime post just a simple, impressive plot here, together with a test. Mind you, I will be probably criticized for taking the plot out of context, but this is indeed a test -and the plot is goddamn blessed.

The idea is to ask you, dear reader, whether the study of that plot alone – which shows the transverse momentum distribution of muon tracks in events with a muon and an electron having the same electric charge found by CDF in 2 inverse femtobarns of Tevatron Run II data- would allow you to conclude that either:

  1. The Standard Model is dead, and this is the first firm evidence of the creation of some new exotic doubly-charged particle
  2. There is a mistake in the shape of the background
  3. There is a funny fluctuation in the number of high-Pt muon events seen by CDF containing an additional electron candidate with the same charge sign
  4. There is a systematic underestimation of the probability of assigning the wrong charge to electrons

A sucked mint will be sent via flying pidgeon to the first commenter indicating which of the above four interpretations is considered the most likely.

PS: since the plot is a .gif file and it shows poorly above, and I am too lazy to fish out a better one tonight, here are the entries in the data (black points with almost invisible error bars) in each of the bins: 0, 2, 5, 4, 4, 7, 7, 7, 2, 4, 2, plus a single outlier far away.


1. Alejandro Rivero - March 17, 2008

case (4) would have terrible political implications inside CDF staff, shouldn’t it? And in the same way, but perhaps minor, (2). So I vote for (3).

2. Anonymous - March 17, 2008

3., but is there an invariant mass plot of the above? How about e-mu-?

3. L. Zoel - March 17, 2008

1,3 seem rather improbable and 2 is easy to check.

4. Kea - March 17, 2008

A sucked mint? Gee. I don’t see how (3) can possibly be wrong – it’s so fluffy – there’s a fluctuation – well, yeah, there is a fluctuation. Arivero is probably closer to you than me, so I’ll save your pidgeon the trouble. Anyway, LOL.

5. Kea - March 17, 2008

Mind you, it would be kind of cool if you killed off the standard model the same month that JPL killed gravity.

6. carlbrannen - March 17, 2008

I’d vote for “3” but I don’t see anything funny here. This isn’t much of a signal, just the usual noise, I think, and that’s before systematics.

7. Stefan - March 17, 2008

It looks like the pentaquark evidence 😉

Cheers, Stefan

8. dorigo - March 17, 2008

Dear all,

I am quite surprised and pleased that most of you shrug shoulders at the sight of the plot above. You appear to be more skeptical than myself – and I am off the typical scale of particle experimentalists.

Indeed, I must confess that when I first saw this plot I jumped on my chair. If this was the result of a blind search for doubly charged weakly -decaying particles, I would in fact tend to put my money on option 1! Look at the entries and how they are distributed: in the low-Pt region, data and background agree – a sign that the Z->tau tau background, the one in lighter shading of pink which is the largest expected contribution at low Pt, is computed correctly. On the other hand, one certainly does not expect muons from Z->tau tau to have 40 GeV of Pt easily – tau decays to muons come together with two additional neutrinos, and the latter carry away on average two-thirds of the tau momentum. Instead, the data has an excess right where the contribution from a direct weak boson decay is largest: between 30 and 50 GeV. You can see it by comparing to the Z->mu mu background (white histogram in the plot). But if the normalization of Z->tau tau is correct, the Z->mu mu one cannot be screwed by a factor four (which would bring the total background to agree with the data points in the region where a discrepancy is evident).

The above argument can be complemented with the observation that the QCD background – the one which may carry the largest systematic uncertainty – is expected to contribute 19.5% (see legend) and also at low Pt. So really, if the excess in the data is systematic, it has to come from a quite different source.

Now, Carl says he does not see much of a signal. But if we do some math we could prove the contrary: there are 30 events above 30 GeV, with a total background probably not larger than about 11. Throw in a 30% systematics on background normalization, and you would still get a (30-11)/sqrt(3.3^2+11)=4 sigma discrepancy.
Further, the excess looks really “signal-like”: a wide bump in a transverse momentum distribution…

Now, I said above “IF this was the result of a blind search”, and that is the key to interpreting this effect. For the plot is a result in a mass of hundreds of plots tested for a discrepancy – that is what automated searches do. I will describe the matter in detail later today in a post about VISTA-SLEUTH, but here I just note that indeed, the effect is not statistically compelling – sizably less than 3-sigma, once one takes the “trial factor” in consideration.

A few further notes on the options I asked you to choose from.

1) If five-sigma evidences are not even enough to call off the SM these days, we have to agree that our a priori degree of belief in the breakdown of the SM is at a level of 10^-7 or below.
2) A mistake is always possible, but this is a blessed plot and CDF produced in its history more than 400 papers and more than 10000 blessed plots (my estimate). Among those, I do not recall a mistake materializing 4-sigma peaks anywhere. Flukes, yes. But real mistakes, later retracted, no. So the degree of belief in option 2 is at the level of 10^-4 or below.
3) If you did not know this was one among many results of an automated search, you could estimate the probability of a fluctuation at about 3.5sigma, or less than 10^-3. It is however larger than that, as I explained above.
4) Assigning wrong charge to a high-Pt electron is possible – the mechanism consists in an electron radiating a hard photon, which in turn converts into an electron-positron pair of which the positron is the highest momentum particle. You track the positron and determine a positive charge, but the process originally created a negative charge at the start. These are called “tridents”. There are other mechanisms by which one can mistake the charge of an electron, but the very fact that these things are not uncommon makes a underestimation of the effect quite improbable: CDF does have a peak of Z->ee events with same-sign electrons, and can thus estimate very precisely how likely it is to mistake the charge in a high Pt electron track. I would assign the likelihood of such a blatant mistake in the rate of charge swaps at less than 10^-4.

All in all, you can see that we are bound to conclude that the most likely cause of the excess in the plot above is indeed a statistical fluctuation. The sucked mint is on its way, Alejandro!

Cheers all,

9. Alejandro Rivero - March 17, 2008

No merit at all :-D- , I put no physics into the argument. But a double charged higgs is still a candidate in some models.

10. goffredo - March 17, 2008

I’d say 2 (background) and then maybe 3 (fluctuation).
I am out of HEP for some time but backgrounds and systematics are the real pains in the arse, not fluctuations (That is unless one has found a golden event = “smoking gun” signal). How does one know the background estimates are correct? Convincing others that you know the backgrounds, by listing and quantifying each, is the main fight in getting any analysis blessed. It is also the hardest to explain to non-scientists.

11. carlbrannen - March 18, 2008

The problem with calculating the 4 sigma extent is that you don’t know in advance which bins to look in. Particle experiments generate charts like this that turn into nothing all the time.

It’s sort of a shame that the easy stuff has been picked out already. Oh for the days when “one positron” was an unexpected signal of new physics.

12. dorigo - March 18, 2008

Hi Jeff, sure – unless you study the full gory details of the analysis (which is not even always possible, since collaborations screen what is made public of an analysis and what isn’t) you cannot convince yourself that backgrounds were estimated correctly, let alone convincing others.

I think in the case of CDF, which published 414 papers in its long history, one can rely both on past documentation made available through those publications, and on the feeling that usually results they put out are correct and error bars are conservative.


13. dorigo - March 18, 2008

Carl, sure – that is the reason why analyses such as the one described here include a “trial factor”. Looking at a specific plot, one is led to estimate the fluctuation as more significant than it is. In the acse above, it is absolutely legit however to pick a minimum Pt and estimate the significance of cutting above that value, since this is exactly what the automated search does. The figure one derives (“4-sigma”) is then to be reweighted by the trials factor.

About the good-old days: I think it is both an effect of the fact that the easy stuff was still there to find, and the fact that nowadays we are much more skeptical, having seen 5-sigma effects consistently disappear and the SM rule for thirty-something years…


Sorry comments are closed for this entry

%d bloggers like this: