No.
The CDMS experiment today reported the observation of two events, with an expected background of .8 events (I gather this is a 1.5 sigma result, but there is no arXiv preprint yet). Based on this, the Guardian reports that “Hunt may well be over for mysterious and invisible substance that accounts for three-quarters of mass of universe” and Science News has Experiment Detects Particles of Dark Matter, Maybe. Science News quotes Craig Hogan as saying the results are “potentially very exciting”, and the Guardian has “If they have a real signal, it’s a seriously big deal.” Unfortunately they don’t have a real signal, so it’s not a seriously big deal.
I just noticed that the New York Times is also covering this, but more soberly, describing the results as “faint hints”. They do quote Gordy Kane, who describes the mood at the KITP in Santa Barbara as “a high level of serious hysteria”, which he then embodies by claiming “It seems likely it is dark matter detection, but no proof.”
For those unfamiliar with the terminology experimentalists use to characterize signals of various statistical significance, here’s a summary:
5 sigma: discovery
3 sigma: observation
1.5 sigma: noise
Update: Scientific American gets it right here.
Update: The paper is here. It includes the information that “Reducing the revised expected surface-event background to 0.4 events would remove both candidates.” There really literally is no signal here.
Update: Ethan Siegel’s blog posting about this explains the appropriate scientific response to the news that two events were observed when .8 were expected:
Well, La-dee Frickin’ Dah!
Adrian Cho at Science has an excellent piece about the story: Wimpy Evidence for Dark Matter Particles. He quotes two experimentalists who explain the significance of this (Richard Gaitskell: “Nobody should be attempting to say that this is evidence” for dark matter, and Edward Thorndike: “Absolutely not” an observation of dark matter). There’s also this comment from theorist Joseph Lykken:
Even so, Joseph Lykken, a theorist at Fermilab, says he’s relieved that CDMS has finally seen something. WIMPs are predicted to exist by theories involving a principle called supersymmetry, which posits a heavy partner for every particle currently known. Had CDMS continued to see nothing, the results would have undermined those theories. So seeing something is better than seeing nothing, Lykken says.
Lykken seems to be ignoring the fact that the new CDMS results, two events and all, rule out yet more of the supersymmetry parameter space. For an explanation of this, written when the last CDMS results came out, already causing problems for supersymmetry, see Tommaso Dorigo’s posting SUSY more unlikely by the new CDMS II results.
Actually, I think the correct answer to this question is ‘maybe’. The events could be from dark matter, it’s just not possible to say this definitively. However, considering the null results up to this point, these results are tantalizing and of interest. We’ll have to wait a short while for better statistics, but it’s very encouraging to be this close.
umm, ” … three quarters of mass of universe … ” ?
That IS a sigma 5 observation the Guardian is making for the dark matter proportion in the universe isn’t it ?
p.s. – Sorry about overstepping bounds in a previous post, Dr. Woit.
Thanks
I guess observation was ruled out as soon as Nature denied they would publish anything by CDMS today.
Oh well, the excitement is over. I had actually hoped for a bit more.
Peter, do you know what caused these impatient scientists to publish a paper with such a low significance value?
Dear Martin:
In experimental physics, you publish what you have. If a search comes out empty, you set new limits on detection, if you have some partial information, it could be noise or real signal. Either way, it must be reported. The grant agencies process can not let people wait for discovery before an announcement.
And by the way, they were not impatient. The data is from a run starting on 2007, and they waited two years to see the events in the corresponding window where detection was possible. They did a blind analysis: they set up the complete analysis of systematics etc on the control region before opening the data in the discovery region. As far as I can tell they did a great job and presented it the right way.
Martin,
David B. is right, this is not a case either of impatient scientists, or of a paper that shouldn’t have been published. Null results and somewhat better bounds on dark matter are very much worthwhile science.
The only odd thing going on here is anyone portraying this as something other than a null result.
Not quite a true null result… a true null result would be another zero, like (more or less) their previous 3 results.
There are two good events in their signal box, which was defined blindly on calibration data. About a 23% chance the background fluctuated up. That summarizes the situation better than more qualitative things like `discovery’ `observation’ etc which is a terminology that has only started being used in the past few years, and are by no means universally accepted in particle physics.
Obviously rumors leaked out that there were good events in the signal region this time. Given the importance of this experiment and result, it is easy to understand the excitement. Frankly, this excitement was more justified than all the excitement during the past 30 years about SUSY and string theory; at least this was a real experiment, carefully done.
The paper is scientifically perfectly correct. Don’t blame the authors for the excitement… the excitement arose naturally and actually is a sign that people still want real data and not mere mathematical virtuosity.
SpearMarkII,
Actually this is a null result. The two events go away if you tighten the cuts to the point where you don’t expect any events (i.e. from .8 to .4 expected events).
I’m all in favor of excitement over experimental results, be they negative or positive, and happy to have helped spread the unfounded rumors that led to some of this excitement. But when an experiment finds no evidence for something, having people going on about how it might really be something is not a good idea. Same as when a theoretical idea doesn’t work out and having it promoted as still a possibility isn’t good either…
Pingback: Jodi Cooley del CDMS: observados dos eventos tipo WIMP de 100 GeV con certeza al 77% « Francis (th)E mule Science's News
So if more events are seen at the same point, then this will constitute evidence in favor of a dark matter particle?
In the hypothetical situation where the results are not statistical noise, will having seen this negligible number of events now– and thus “knowing where to look”– aid CDMS or LHC researchers in any way by allowing them to conduct their experiment in a way that will ensure further signals at that point are not overlooked? (Alternately, is there risk that there will be an effect where since CDMS or LHC researchers “know to look there”, they will be subconsciously more likely to commit experimental error that would reinforce the idea there’s a dark matter particle there?)
Even though the balance of the evidence is against the detection being real, is there value in theorists checking now to see which models are compatible with a dark matter particle at that mass, in case more events are seen?
I am excited and I can’t help it. I’m excited people are interested in real physics. I’m excited we are debating real experimental results instead of multiverses and dreamworlds. I read three papers on multiverses this week and all three proposed different mechanisms and not any of them had any way to test their conjectures. Hats off to the CDMS team for conducting real science and reporting their results honestly.
Coin,
There’s not much point in running the same apparatus much longer, you’d have to do this for quite a few years to get significantly better statistics than they already have. Better to put the time into building a more sensitive detector, and that’s what they are doing (“SuperCDMS”).
I don’t think the observation of these two background events helps one know what a real dark matter event would look like. All it does is tell you more about what the difficult to handle background looks like. Theorists have studied intensively the question of which models are compatible with the null results of this experiment and others. For more, see the old Tommaso Dorigo posting I linked to.
Peter, hard to say. 2 events when 0.4 is the expected background is 2.3 sigma or so (single sided). Doesn’t quite qualify as null. This experiment is in the uncomfortable in-between region, and folks want to editorialize one way or another. Rick Gaitskell is on a competing experiment (LUX), and Ed Thorndike at a LUX institution (Rochester). Lykken is at a CDMS institution. They all fall into line. You, Peter, are averse to excitement in the string theory community.
SpearMarkII,
You have the numbers wrong, look at page 4 of the paper. The two events correspond to cuts with expected background of .8 +-.1 +- .2. They go away and there are zero events if the cuts are tightened and the expected background is reduced to .4
SpearMarkTheSecond,
The CDMS paper says
“the probability of observing two or more background events is
23%”
In this case the distribution is not Gaussian so it doesn’t strictly make sense to talk about sigma. But it is interesting to compare to the canonical Gaussian experiment of an assumed mean and standard deviation, sigma.
Then one says, OK if this was the true distribution and we took a sample, what would be the probability of drawing a sample with a value a greater distance from the mean than the observed one. If the probability is small, this leads you to start doubting the model. In this gaussian case, if you have drawn a sample 1.2 sigma away from the mean then there would be about a 23% probability of drawing a sample a further distance from the mean. So as a useful shorthand you can call a 23% chance of your data coming from background events as 1.2 sigma result. You may think its better to just stick to probabilities and that is fine but when you start talking about 3 sigma you are already talking about a probability of about 0.0026 which is starting to get a bit hard to visualize and once you start getting down to such small values you probably have a lot of data which means, by the central limit theorem, your distribution is probably pretty close to Gaussian anyway.
So you may say “well 23% thats less than 50% so it is a hint”. But bear in mind there are lots of experiments looking for any one thing, so if you have 20 experiments, one of them is going to get a 2 sigma (5%) result on average. Also, it is generally not possible to eliminate systematics to beyond about a sigma so a two sigma result could easily be a one sigma result. So I think it really is worth having some reasonably strict criteria to when you are going to start taking an experimental detection seriously. I think a lot of theorists are too eager for new results though and that’s why you get some fairly unwise comments in cases like this.
Peter, I was responding to your statement that at 0.4 expected backround, the two events are cut. They are just barely cut, so if the cut is at 0.4001 or so, two events remain. Had the cut been placed there, this could have been a >2 sigma result.
The cut variations are all unfair, however, Peter, mine just above and yours. Your argument that tightening the cuts cut the events and therefore it is a null result is as bogus as my response that had the cuts been tightened to a point just before the two were cut it would have been >2 sigma.
My original point is what I stick too: it is not a null result (because 2 events in the signal region were seen) but not a significant result because 23% chance of background fluctuating. All the rhetoric of words is BS here. From my perspective, this one experiment is more valuable than all the theory of the last 20 years, however.
Physicsphilie, of course, but everyone translates from confidence levels to sigma because that is established shorthand, whether gaussians apply or not. And I think the real reason the field has come to demand 5 sigma etc is that we don’t really trust our confidence interval calculations. In reality 3 sigma is just fine if we trusted the underlying understanding of the statistics. That we demand 5 sigma is a technique like in the very old days `multiplying by pi’ to get a more conservative limit.
SpearMarkII,
Where in the paper does it say that the two events remain at an expected background of .4001? You seem to just be making up numbers.
One other number that is in the paper is what happens if you loosen the cuts: no new event appear until an expected background of 1.7.
These numbers all correspond to exactly the sort of thing you expect to see if there is no real signal.
A statistical significance of 23% is very little, to be sure, and one could not call this “evidence” for a signal. Even if one supposes that the experimenters were conservative in they way they interpret their results (ie, with systematics, background estimates, etc.), this result is far from what one could designate to be a signal.
Are one or two of those two events actual dark matter scattering events? There is no way to tell at present. The only way to make progress is to take more data! Happily for all of us, the Xenon experiment will have relevant data rather soon, so we don’t have to wait until SuperCDMS is ready and able to record new data.
So – if some of us think this is a null result, while others want to see a signal – what are the *predictions* for Xenon? 😉
Michael,
Good point, someone should start circulating rumors about Xenon100, which may be reporting results in “early 2010”, and is supposed to be an order of magnitude more sensitive than CDMS.
I don’t know if they’ll see anything, but, if they’re 10 times more sensitive than CDMS, my prediction is that they won’t see the 20 events you’d expect if the two CDMS events are real…
From the Xenon100 experiment page:
“A WIMP search in the current XENON100 will be first performed in 50 kg fiducial target with 40 live-days of exposure, to reach a WIMP-nucleon spin-independent cross section at 6 x 10-45 cm2 for 100 GeV/c2 WIMPs”
Is the expectation of a dark-matter candidate mass at 100 GeV/c2 based on cosmological estimates of “missing mass”? Also, is there a theoretical expectation for the cross section and is it possible that it’s lower than what we think? If so, is there any way to know if we will ever detect a WIMP?
Unrelated to the post, but I think you’d find this comic funny:
http://xkcd.com/171/
I just read the CDMSII abstract and answered one of the questions I posted previously on cross-section:
The upper limit (based on current and previous data) for cross-section is 3.8 x 10 ^-44 cm2 for a WIMP mass of 70 GeV/c2. Is there any chance that the dark matter candidate has a mass greater than 100 GeV/c2 or is the lifetime to short to interact with other particles?
Pingback: CDMS’ Endresultat: Wie die Dunkle Materie wieder nicht entdeckt wurde « Skyweek Zwei Punkt Null
teaser: the mass can be both lower or higher, and the cross section can vary by orders of magnitude depending on what model for dark matter you’re considering. Look at Fig 4 in the new CDMS paper. The colored area represents different parameters in the MSSM, so even within the MSSM the WIMP can have very different properties depending on in what part of parameter space you are.
That the events are cut very suddenly at an expected background of 0.4 is in Lauren Hsu’s talk at FNAL, see http://cdms.berkeley.edu/hsu_091217_FNAL.pdf , page 41. Still, the point is that the cut they used was determined blind and represents the fair value. Peter, your observations about loosening or tightening are hindsight and statistically biased.
Forgive me if I question your expertise, Peter, concerning what to expect about the behavior of a real signal. I doubt you’ve ever had to stand up and announce a new experimental result before a skeptical audience. Certainly your assertion that I make up numbers belies your noobility.
Michael… follow the impurities in Xenon100’s LXe… doubt they’re gonna get the sensitivity they advertise.
And if all the money spent on string theorists salaries and grants over the past 20 years had been spent instead on initiatives like CDMS, Xenon, LUX, Coupp, etc, science would have advanced one heck of a lot further in this area…
SpearMarkII,
I have no idea who you are, although am quite willing to believe your expertise in this area is much greater than mine.
On the other hand, I think I have more than enough expertise to understand the numbers put out by this experiment. I don’t question the accuracy of the numbers or the huge effort and expertise they represent. It’s quite clear though what it means when an experiment looks for a signal that will revolutionize physics, expects one background event, instead sees two, and the two events go away when the cuts are modestly tightened (I don’t see how the graph you point to has both events going away exactly at .4, but, whatever….).
They had a few little detectors? Why didn’t they just have bigger detectors, and more of them?
point,
money.
reminds me of a fellow grad student
TAing a lab class. one of his student
did an experiment with two data points
and wrote something like
‘this proves the linear relationship’
One event is cut at an expected background of about 0.35 and the second at an expected background of about 0.18, according to Lauren Hsu’s plot. I suppose the neutron background should be added in, which probably raise those to about 0.43 and 0.26. Had CDMS chosen a tighter cut prior to unblinding, they easily could have ended up with a >2 sigma result, but not a >3 sigma result. Or, if they had chosen an expected background of 0.1, they’d have had a null result. They chose what they chose based on an optimization of their sensitivity that was done with no knowledge of their actual signal events… shown in some of the other CDMS talks this week. Very fair and very correct.
Then the unblinding of this CDMS exposure returned a result between null and significant. We all used to discuss, in the pre-blinding days, how results like this simply wouldn’t get published… there was a strong publishing bias favoring true null results or truly significant results. Luckily blinding has helped allow the community to accept results in this in-between region.
Sure some people got too excited. Now other people are getting too cynical. I think the right way is to say, well, there is a 23% or so chance the background fluctuated, and so the CDMS results are interesting but not conclusive and not null either.
Dear Peter,
amongst all the ‘hype bashing’ you do in this blog, i find the bashing of this particulat CDMS result as the most necessary piece. really.
it might seem as legit advertisind of the CDMS people to blow their result slightly out of proportion. well, this is how we (the trained scientists) react to it. 1.5 sigma is nothing. you don’t have to say much more.
but the general public: oh my! you know, they trust us (scientists). and if these people go around telling their story about how they found ‘dark matter’ but there is a 25% chance that is was not dark matter, then guess how an average intelligent person reacts to that: oh, 75% chance that it was DM. and you know what: they are right. 1.5 sigma taken literally mean there is a 3:1 chance that this is the discovery of the century. i would get freakin’ excited by that, too if i didn’t know better.
if you think i am nitpicking or so, i am not. really. i spent a good part of the weekend trying to explain to laypeople why they should not get excited about a 75% chance of DM detection. and yo know what: almost nobody understood. and you know what’s even worse: the people supporting me in this discussion were the ‘einstien was wrong’ nutcases and the ‘all research is bad’ types. the laypeople seriously interested in physics didn’t follow me at all. and why? because they rightly invoked that these CDMS people know much better than i what they were doing and if they claim 75% chance, then this is so.
i am so depressed. seriously. i mean, what have we come to? where is the rigour of past days, where you would have to fight a steep uphill battle against an extremely conservative established opinion that you could only win with the correct theory as your companion.
i have somewhat accepted that nowadays you can claim multiverses, backward causation and all that experimentally unsupported nonsense if you include the small cautionary note that it is speculative. but it seems that this constant flow of the ‘excitement’ drug to the public has spoiled their perception so much that even hard-fact data based experimentalists can’t resist the urge of pumping up their non-results instead of humbly reporting that they set better limits on DM. don’t they see that in the long run this undermines the credibility of science? aren’t they afraid about the long term effects of their claims, once DM limits have moved by a factor 10 and shown their results to be a fluctuation at the 5-sigma level? what do they say to joe the plumber then when he asks them about the 75% certainty they claimed?
i really hoped that experimentalists by their very nature would be cautious against hypes. i am really depressed about this.
point says: They had a few little detectors? Why didn’t they just have bigger detectors, and more of them?
Woit says: money.
point: Hmmm, so I wonder what their approach would be to try to attract more money.
SpearMarkII,
Your interpretation of that plot from the talk is in direct contradiction with the unambiguous statement in the paper on page 4 which I’ve quoted here earlier and pointed out to you:
“Reducing the revised expected surface-event background to 0.4 events would remove both candidates while reducing the WIMP exposure by 28%.”
Whatever detector one builds, it will always be `little’ by the standards of the next generation.
I imagine that CDMS-II was not little when it was designed.
As I understand CDMS-I was at Stanford (‘very little’?)
One demonstrates that the design and experimental concept are fundamentally sound (fundamentally WIMPY?)
and proposes a next-generation experiment.
Chris, as far as I am aware the CDMS have not been hyping their result, that’s been done by others not related to the experiment.
SpearMarktheSecond, usually in statistics you have the null hypothesis and the alternative hypothesis. You assume the null hypothesis and if your data are sufficiently badly fit you can interpret this as evidence for the alternative hypothesis. So in the CDMS case the null hypothesis is that there was no detectable dark matter (ie either dark matter does not have the properties to be detected by this experiment or there simply was no dark matter passing through this experiment) and the alternative hypothesis was that there was detectable dark matter. Assuming the null hypotheses one gets a 1.2 sigma result. Such a fit to the null hypothesis is not considered bad, the usual convention being that a 5 sigma or greater result is needed to refute the null hypothesis. Perhaps if it was 3 sigma I think you would be justified to refer to this as “CDMS results are interesting but not conclusive”. I think it is more accurate to say the CDMS results are consistent with the null hypothesis that dark matter does not have the properties to be detected by this experiment.
No Peter, it is an issue of significant figures (0.4 was quoted, not 0.40 or 0.400, meaning 0.3 to 0.5), consistent with not giving ultra-precise information so as to give the illusion of false precision and tuning of the cuts.
Physicsphile, sure, you’re right, CDMS could not reject their null hypothesis. We can agree to disagree about whether 2 events in the signal box is interesting. Perhaps after Xenon, LUX, and Coupp have reported we’ll know (assuming they can solve all their purity and discrimination issues)… all sorts of early observations are kept track of for years in the lore of experimental particle physics.
Is the present non-detection of CDM at various detectors a problem for DM models, based on expected flux, probability, distribution, sensitivity?
Dan, the answer is ‘No’. Even in the somewhat limited supersymmetric DM model with neutralino LSPs the new limit from CDMS hardly makes a dent in the allowed parameter space. It only really starts to do so if the limit get about 1000x stronger..but that will be years from now. In the meantime the LHC will have a lot to say about most models with DM partcle candidates & is more likely to compress their parameter spaces if nothing is seen.
thanks for replying. It sounds like LHC timetable won’t be collecting statistically meaningful data for another 6-10 years out.
Until that time, isn’t there time to build several thousands of these detectors?
Would increasing the size of the detectors, or increasing the number of detectors around the world, improve sensitivity? Would say having a null result from a couple thousand of these detectors all around the world, shielded from cosmic rays, neutrons, ground radioactivity, constrain it?
dan,
I think the LHC time-table is shorter than this, with them collecting a significant amount of data at or near their design energy in 2012.
The cost of reproducing these detectors by the thousands, as well as installing and operating them around the world would be prohibitive. New, better detectors are being planned and built now, but the way they achieve higher sensitivity is by making the detector larger, not building a lot of smaller ones. Given the lead time necessary to design and build these things, the competition for the LHC is going to be those that are already being built and close to or already in operation.
PW, thanks for replying. I was also thinking in terms of luminosity and collecting enough collision events to be statistically meaningful, say 3 bar. LHC by 2012 will have collection enough data by 2012 to discover DM and/or Higgs?
I understand that building larger detectors improves sensitivities, and I understand it is expensive and time consuming. What I had in mind is that other countries would pay for it, i.e 10+ built in Japan, Korea, China, India, Germany Britian, Italy, France, Russia, Mexico, Australia, would be paid by those respective countries.
Is there a reason that specifically the DM candidate neutralino has not been found in Tevatron? Is it that Tev lacks energy or luminosity?
dan,
I think those countries aren’t interested in spending more on this. Britain is planning on cutting spending on this kind of research, not increasing it.
The luminosity and energy you need to discover the neutralino depends on your model (supersymmetric theories have lots of extra undetermined parameters, and for different parameters you get different experimntal signatures). From what I can tell, immediately after the CDMS result was released, phenomenologists started writing papers calculating what the experimental signature would be for their favorite models, putting out almost half a dozen papers/day on the arXiv this week. It’s a complicated business, consult those papers for the latest news on this topic.
Already 9 papers on the arxiv on the cdms results…
Hey Woit,
Get back to work!
I’m kidding, and I’m too ignorant to comment on the subject matter here, but I enjoy reading NEW.
Best Wishes and I hope you aren’t troubled by illness or sadness.