Did the Higgs Signal Fade?

When the Lepton-Photon conference started 10 days ago there was a report in the Guardian that the signal for the Higgs boson reported at the earlier Europhysics conference had “faded”. They even put figures on it saying that the excess observed by ATLAS had decreased from 2.8 to less than two and in CMS from 2.8 to 2.3. The message was echoed in other papers who picked up the story and was also reported at the conference by the collaborations themselves with CMS saying that “Excess in the low mass range seems to persist but with reduced significance.”

The cause of this change was said to be the addition of some new data into the analysis, but I think this has to be looked at in more detail, so I have been doing some more combinations of the LHC data and am now working from the individual decay channel plots. The first clue that the story is not quite as straightforward as it seems comes when you look at what was said by CMS about the WW decay channels at the conferences. here first is the slide from Europhysics.

You may need to click to see full-sized. This slide shows that there are two distinct analysis methods available, “Cut based” and “MVA based”. The MVA gives a much better result as shown by the lower expected CLs line. In fact it is about as good as twice as much data. You will also notice that the excess from the MVA analysis was bigger which is what you would expect if the signal is real. Indeed the MVA analysis was the one used in the final CMS combination for Europhysics.

Now look at what they said at Lepton-Photon.

This shows just the Cut-based analysis with a note that the MVA-based result is coming soon! They have used 1.5/fb compared with 1.09/fb at EPS but remember that the MVA method is as good as twice as much data, so in fact the data used for WW at EPS was better and they took a backward step. The WW channel dominates the plot over the crucial range where the biggest excess was observed. You can even see directly that the expected CLs line went up higher in the LP plot compared to EPS, so really they took a step backward. A fading excess is therefore exactly what we should expect.

To see how much the excess actually changed we can reconstruct the EPS plot using the cut-based data and do the same thing for LP data. This is what we get


What we find is that there has been a small decrease in the excess in places, but not by much.

The natural thing to do next is to construct the plot using the MVA based data from EPS for the WW channel with the other channel data from LP. Since the EPS data was better for WW we should get the best possible results this way.

This brings back the broad excess previously seen with almost 3 sigma significance at 140 GeV. So what we can now say is that the observed decrease in the excess for CMS was mostly due to a change in the analysis rather than a statistical fluctuation as implied.

What about the ATLAS data? They were reported to have an even bigger decrease in the excess from 2.8 to less than 2 sigma. Here are my plots reconstructed from individual channel data.

There is some decrease in the excess but not as much as advertised. In fact the signal does not appear to have been as strong as originally claimed in the first place. Of course my combinations may not be as accurate as the official ones, but at least I can be sure the analysis has not changed, just the data.

Conclusion: The CMS excess did not fade at all, the difference was due to a change in the analysis from Cut-based to MVA-based for the dominant WW channel. The ATLAS combinations when reconstructed consistently only show a small decrease in the excesses. Not the large decrease advertised. Higgs boson hints are still alive.

11 Responses to Did the Higgs Signal Fade?

  1. JollyJoker says:

    How sure are you that the MVA data is “better” rather than simlpy wrong? If the MVA excess was real, adding more data should give you more signal even in the cut-based analysis. Now the cut based CMS EPS -> CMS LP change is next to nothing, which is weird in itself unless I overestimate how much a 1 -> 1,7 /fb increase in the WW channel would affect the results.

    I’m still confused.

    • Philip Gibbs says:

      The addtional data in the cut based CMS plots moves everything down a little. I think that is a fair result. You might expect that the excess could have got a little bigger, but the stats can’t be in your favour all the time. Point is that it was not a big drop in the excess.

      I cant be sure that the MVA result is better but I have to assume that CMS know what they are doing and that what they say should be correct. They show a lower value for the expected CLs line with MVA over Cut, so this tells us that the MVA should be better. Of course they could be wrong.

      • JollyJoker says:

        Is it possible that the Look Elsewhere Effect is magnified in the MVA result, giving a de facto higher chance of an x sigma deviation? I might just be reading too much into the fact that we had a bump that was larger in MVA but seems to have weakened in the newer cut-based curve.

        I should probably have chosen my words better in the previous post; I didn’t mean “wrong” to imply it couldn’t be random fluctuation, just as “not a real signal”.

  2. David George says:

    Dear Phil,

    I am not a physicist but am following this story with interest. On the “Quantum Diaries” website, Corinne Mills has a post

    http://www.quantumdiaries.org/2011/08/28/lost-in-transition-2/

    which says,

    “In the results shown at EPS, both the ATLAS and CMS Higgs searches reported limits that were not as good as expected for low masses (130 GeV < m(H) < 160 GeV or so). The excess of events driving the degradation in the limits was mainly in the WW → lνlν final state."

    My first question is, what is a "degradation"? It doesn't sound very good. If you read the article, does it give you a clue what is going on?

    David George

    • Philip Gibbs says:

      The choice of language is missleading. What they are saying is that they did not exclude as much of the mass range as expected if there is no Higgs boson. The reason for it may be because there is a Higgs boson in the range. So it is a good thing, not a bad thing.

      In my next post I will show what happens when you construct the plots with everything except the WW channels.

  3. [...] the previous post I explained that the CMS Higgs plots shown at Lepton-Photon were not as good as they could have [...]

  4. Ian Sample says:

    Hi Phil,
    First thing to say is thanks for your regular updates on the Higgs searches – I follow them and the comments they generate with huge interest. I am still at the slightly-informed-lay-person level on this, so can’t do the kinds of reconstructions and analyses you’ve posted here.
    I wanted to say that the figures I quoted in the Guardian as evidence for fading excesses were read over the phone to me and I suspect were fairly rough estimates. Really good to see your thoughts here on what may have given rise to those changes. Keep it coming!

    • Philip Gibbs says:

      Ian, thanks for the inside info. Your article was fine in that it reflected perfectly what the experimenters reported. Somehow they missed that there had been a change in the analysis. I think that is just one of the consequences of the rapid pace at which they are working. It was only after some questions in the comments on these posts that I started to look into it in more detail. There will be lots more twists and turns in this story before it concludes i am sure.

  5. Joe Khan says:

    I’ve just started getting into this area of physics.. Do you think higgs boson is actually flawed and will the LHC prove that this is a matter of fact or not?

    • Philip Gibbs says:

      I think that is unlikely given both the current experimental results and the theoretical background. There is always a small chance however.

  6. Valinda says:

    Please teach the rest of these internet hooaligns how to write and research!

Follow

Get every new post delivered to your Inbox.

Join 276 other followers

%d bloggers like this: