The Unofficial Higgs Combination Tool has now been updated with all the new Higgs plots released in the last few days, including the Tevatron updates and the new 8 TeV data from the LHC. There will probably be more to add on 7th and 9th July from ICHEP. Feel free to play around with it.
At the CERN press conference yesterday the Director General Dr Rolf Heuer warned journalists about unofficial combinations. What he said exactly was at follows (It is 26:50 in if you are looking on the recording):
“The fact that they [CMS and ATLAS] have not yet combined their results today is that they did not have enough time. We should have shifted the Melbourne conference by 2 weeks or 3 weeks or 4 weeks but that was not possible. You have to stay tuned until at some time they combine their results. Whatever combination you get beforehand is unauthorised and is certainly not valid because you have to take into account the different correlations, one has to be very careful.”
I agree with what he says. The unofficial combinations you find on this blog are approximate and unofficial and should be used with caution. I have always made that clear. It is not just the correlations that are neglected. The quick combination method assumes that the statistical errors have a flat normal distribution and that is not quite correct. The detector collaborations don’t provide detailed likelihood data to outsiders so this is the best I can do. Luckily all statistical errors tend towards the normal Gaussian as the quantity of data increases (central limit theorem) and in most cases there is enough data for the results to be good, with a few exceptions.
Whether the combinations are “valid” or not depends on what you are using them for. I don’t consider them valid for writing up published results of any kind, but they are good enough as a rough guide to theorists looking for possible signals in the data and there is nothing wrong with showing them at conferences as some eminent theorists have already done, provided they come with appropriate caveats.
I have previously shown some comparisons between official combinations and my unofficial ones to show how accurate they can be (or not). I think it is worth doing a few more now using some of the recent results where the amount of data has increased. In all the plots below the red line is the official result and the black is the unofficial. First up is the latest version of the Tevatron combination compared with an unofficial combination of the updated Dzero and the latest CDF plot that was updated in March. You can click on the plots to get a larger version.
The combinations across all channels have always worked quite well because they use lots of data. The last time that the LHC provided an official combination for ATLAS + CMS was in November when there was only 2.3/fb. here is how it looked next to the unofficial combination that I had done 10 weeks earlier.
Notice here how the accuracy gets worse at higher energies where there is less data available. Heuer seemed to be implying that there should be another combination due out soon. If so it will be interesting to see if the comparison improves as I would expect.
The combinations for single channels have been less successful in the past, but now they are improving. Here is a reconstruction of the ATLAS combination for 7 TeV + 8 TeV data in the diphoton channel
But the results don’t always come out so well even now. The 4 lepton channel uses very few events in both the signal and the background. Here is the result of a similar combination (Update: There was an error in the digitisation that I now fixed and it is not so bad now)
The combination across ATLAS and CMS should be better because it involves twice as much data. They should also have twice as much again by the end of the year so by then combination should work OK even in this channel.
If you want to try more the Higgs combination tool is easy to use and free.
Update: I said that I dont think these combination methods should be used in published papers but other theorists are apparently not as reticent. arXiv:1207.1347 is one example of paper showing a combined signal plot as well as combined channel values and other fits. There conclusion is that everything fits the standard model except that the diphoton rate is 2.5 sigma too high, in agreement with my figure.