maxwells_daemon: (Default)
[personal profile] maxwells_daemon
Today was an exciting day. Finally we can see our our new Higgs search results public, and we can see what CMS have seen and compare. As everyone keeps emphasising, we haven't discovered the Higgs. ATLAS has seen a fairly large "excess" and CMS have a couple of moderate excesses, one of which matches ours at a mass of 126 GeV.

You might want to stop reading here, as it's all science (too much science for LJ?) from here on in...

The two most telling plots are these:



which give the significance (actually (im)probability that there is no Higgs at each mass) of the ATLAS and CMS results, separated into different channels (ways the Higgs can decay) and combined (in black) within each experiment. Those plots are the ones to look at to answer the question "did we see anything?", so probably most interesting at this stage (nothing to do with the fact that I worked most on this one for ATLAS, though I was a small cog in an enormous machine).

If this is a hint of the Higgs, then ATLAS was rather lucky to get such a large signal (though it is still not so large to be suspicious). If this is a statistical fluctuation, then ATLAS was really unlucky (as our spokesperson put it "if it's background, it will be really difficult to kill"). CMS's smaller signal was neither lucky or unlucky.

These numbers have to be corrected for the "look-elsewhere effect" (the fact that we are looking for the Higgs at many different masses and if we look often enough we are bound to see a statistical fluctuation) - unfortunately this was not included in all the internet rumours that prefigured the announcements, so today may have been a bit of a let-down for some people. There is actually a philosophical problem with that correction: should we count everywhere we looked, or just the area that we are still looking (and haven't excluded). ATLAS and CMS actually took different options in their conclusions (probably because it doesn't make much difference for ATLAS, so they can be conservative, while CMS have nothing very exciting to say if they follow us).

So ATLAS/CMS see a significance of 2.5/1.9 sigma (0.6%/3% probability that this is just chance, not a Higgs) or 2.3/0.6 sigma (1%/27%) for a Higgs at around 124/126 GeV, depending on how you calculate the look-elsewhere effect. It is quite incorrect to try to combine these numbers without detailed study (several of the uncertainties are correlated between ATLAS and CMS - it took months for ATLAS and CMS to combine their previous Higgs results), but that hasn't stopped the blogosphere. If you were to naively combine the larger set of numbers you reach 3.1 sigma (0.1%), which in our field is sufficient to claim "evidence" (still well short of 5 sigma required for an "observation"). But I couldn't possibly comment.

For me, the fact that we see an excess at the same place in so many channels (3 channels in ATLAS, 1-3 in CMS, not to mention various sub-channels), makes me quite hopeful that this is something real. Each channel looks for different things (and the two experiments have different detectors and analysis techniques), so it is unlikely to be a mistake. That means it can probably only be a fluke if this isn't really a Higgs. Next year we hope to have lots more data (and at a higher energy), so we should be able to pin it down soon.

Date: 2011-12-13 09:47 pm (UTC)
From: [identity profile] parrot-knight.livejournal.com
Fascinating - but what does GeV stand for, and why measure probability in 'sigma'?

Date: 2011-12-13 10:19 pm (UTC)
From: [identity profile] e-pepys.livejournal.com
Sorry for the jargon. I sort of forget what is jargon, but also want to give a flavour of what we are talking about without converting to millions and millions.

GeV stands for "giga electron volts", but its just a convenient unit for particle masses (actually it's a unit of energy, but E=mc2 and all that - if I were being formal, I'd say "GeV/c2"). For scale, a proton has a mass of about 1 GeV.

"sigma" is a measure of probability that doesn't need many zeros for small values (eg. 5 sigma is a probability of 3x10-7). It is also convenient for combining probabilities (like I certainly wouldn't do with ATLAS and CMS results) and estimating fluctuations. "1 sigma", "2 sigma", etc have also become common benchmarks in measuring significances.

(Formally "sigma" is the symbol for the "standard deviation" measure of the width of a Normal distribution (bell curve). Many of the probabilities are distributed like that (in fact part of my job was to check this for our measurements), but equivalent Normal sigmas have come to be used regardless of the distribution.)

Date: 2011-12-13 10:26 pm (UTC)
From: [identity profile] e-pepys.livejournal.com
Oh look, CMS used "GeV/c2" on the axis of their plot, while we just said "GeV". Maybe we should fix that informality in the paper!

Date: 2011-12-14 09:35 am (UTC)
From: [identity profile] zengineer.livejournal.com
It looks like the experiments are seeing something.

Atlas shows two overlapping peaks in the H-4l and the H-yy (sorry for the lack of Greek) channels. Would you expect to see a peak in the H-lvlv channel if it were the Higgs? Is this lack negative or neutral information?

CMS doesn't look that supportive to me. The H-yy channel shows a peak matching the Atlas result but also shows a peak around 137. The H-ZZ-4l shows a different peak at 118 and the other channels show no peak (as above is this neutral or negative?).

The trouble with sigma values is that is presumes a distribution (often binomial) and the statistical signal errors (rather than background) may be systematic in a non-binomial way depending on the process that creates them. I am not saying this invalidates the statistics as you have to use some method and in part the use of 5 sigma rather than 3 sigma gets around this.

Good point to release data and hopefully more information will either strengthen each peak or lead the peaks to converge. It gives the theorists pointers as to what they need to predict which is both good and bad.

Date: 2011-12-14 06:13 pm (UTC)
From: [identity profile] e-pepys.livejournal.com
Good points. You honed in on some of the crucial questions.

These p-value plots are designed to show how significant any deviation we are seeing is from no-Higgs, not how much our observations are incompatible with there being a Higgs. That's better seen with the yellow band CLs plots in the papers, but then you have to look at lots of plots if you want to see the individual contributions. As often with statistics, if you ask a different question, you should expect a different answer. Still, the ATLAS p-value plot on the left shows the expected sensitivity (if there really were a Higgs at each mass) with dotted lines and that can give you an idea.

For the H→lνlν (which CMS call H→WW, since each W decays to lν here), you can see that observation roughly matches expectation. This channel does not show a nice peak because we can't measure the neutrinos (Greek letter ν), but we would expect a broad excess if there were a Higgs around this mass - just as we do see. But it can't distinguish 126 from 119 GeV.

CMS's other measurements (apart from γγ, lνlν, and 4l) aren't so sensitive in this region, so it isn't surprising that they don't have dips in them. They had made an impressive range of measurements with the full data sample, while ATLAS have only updated their analysis for the most sensitive channels. CMS's excess at 119 GeV should, by rights, also be seen in γγ - and it clearly isn't. That is why they don't place much emphasis on 119 GeV. On the other hand, the peak at 124 GeV is compatible with what they see in all channels (4l is smaller than expected, but not significantly smaller).

Concerning sigma values (or let me say "σ", since I see that LJ can cope with Greek), we just quote them as a unit of probability, as if they did refer to a Normal distribution, even when that is not true and they aren't the same as standard deviation. Of course that means that one's intuition might be wrong on some questions (combining them, how they might vary with more data, etc), but mostly the Central Limit Theorem applies and we are OK.

Date: 2011-12-15 09:10 am (UTC)
From: [identity profile] zengineer.livejournal.com
Thanks. I get some explanation at work but generally it either assumes you are already working at CERN as a particle physicist or you work in the finance office without much in between.

Profile

maxwells_daemon: (Default)
maxwells_daemon

April 2013

S M T W T F S
 1 23456
78910111213
14151617181920
21222324252627
282930    

Style Credit

Expand Cut Tags

No cut tags
Page generated Sep. 20th, 2017 12:44 pm
Powered by Dreamwidth Studios