Wednesday 7 September 2011

Cresstfallen

Until yesterday CRESST was a sort of a legend: everybody heard of them but nobody ever saw them. That is to say, the CRESST excess has been informally discussed for a long time. Moreover, the events from one of the modules displaying an excess of events in the oxygen band have been shown at conferences since more than a year. However we were in the dark about the significance of the excess, backgrounds and systematic effects. Now CRESST has finally come out with a paper that spells out the excess and provides interesting details.

CRESST is in a way the fanciest of all dark matter experiments. Located in the Gran Sasso underground laboratory, it uses the target of CaWO4 crystals cooled down to 10 miliKelvins. When a particle scatters inside the target the deposited energy is converted into phonons and scintillation light, both of which can be detected. The light-to-phonon ratio helps discriminating the dark matter signal from backgrounds, for example electrons and photons produce mostly light. Furthermore, that ratio depends on the atom of the crystal molecule on which the scattering occurred: it is largest for oxygen, intermediate for calcium, and smallest for tungsten. This leads to characteristic bands in the light yield vs. recoil energy plane that you can see in the plot above showing events from one of the eight CRESST modules used in this analysis. These bands provide another handle on the signal, as heavy dark matter would show up mostly via scattering on tungsten, while the light one would pop up in the oxygen band.

At the same time CRESST is paying a price for their innovative technology, as they have to deal with incalculable and sometimes unexpected backgrounds. Apart from the usual neutron background and the leakage of e/γ events into the signal region they had to face α particles and Pb atoms emitted from the clamps holding the crystals, not to mention the exhaust fumes from the nearby DAMA detector. Some of these backgrounds will be reduced in future runs, but for the moment CRESST needs to estimate their contribution in the signal region using sideband analysis. Having done so, CRESST finds that a fraction (slightly less than a half) of the 67 events in the signal region cannot be understood in terms of the known backgrounds. Therefore they study the likelihood of the background plus dark matter signal hypothesis assuming vanilla elastic scattering of dark matter on the target. Here is their result for the preferred mass and cross section of the dark matter particle:
The likelihood function has 2 minima corresponding to 4.7 and 4.2 sigma rejection of the background-only hypothesis. We can safely forget about the deeper one: for these parameters Xenon100, CDMS and Edelweiss would see an elephant in their data. The shallower minimum, where the preferred dark matter mass is 9-15 GeV, also seems excluded by orders of magnitude. This one however lies in the tantalizing proximity to the CoGeNT and DAMA preferred region; actually the mass region (though not the cross section) perfectly agrees with the DAMA low-mass region. Some argue that CDMS and Xenon collaboration grossly overestimate their sensitivity near the threshold. This may be imagined in the case of 5-7 GeV dark matter, in which case combining experimental and astrophysical uncertainties with some good will and the presumption of innocence one can try to argue that the CoGeNT signal is marginally consistent with the Xenon and CDMS exclusion limits. On the other hand, 10 GeV dark matter would produce observable signals further away from the threshold of these 2 experiments, and it's unlikely it could escape their attention. Therefore, given CRESST is facing pesky backgrounds very similar to the suspected signal (both in spectral shape and the order of magnitude), the hypothesis of unknown and/or underestimated backgrounds faking the signal is currently the most probable one.

Summarizing, the new CRESST results are welcome and illuminating but they do not change significantly the landscape of dark matter searches. Clearly, experiment is closing in on IDM; what is not clear is whether that stands for Inelastic or Italian Dark Matter ;-)

See also Lubos, Matt, and again Matt.

21 comments:

Robert L. Oldershaw said...

Sigh. Pseudo-reality prevails.

Anonymous said...

I like the second iDM interpretation ;)

Anonymous said...

Jester, your blog is most enjoyable when you adhere to areas of your expertise. Can you quantify your final statement about rates in XENON and CDMS from a 10 GeV WIMP within the CRESST region? (some of your readers can).

Anonymous said...

Forgive me; "Pb particles"??

Anonymous said...

Anon2, if you are implying that you are better able to "quantify" this statement than Jester, why don't you share for the benefit of the rest of us?

Anon3, Pb = lead

Anonymous said...

Yes I'm aware of the chemical symbol for lead. But why would lead atoms be a problem if found in the clamps holding a detector like this together (and how/why are they "emitted"). It's not exactly a volatile metal like Hg and it's not radioactive. I don't see how it poses a problem, which it presumably must... somehow.

Phil said...

There's quite a bit of discussion of the problem with lead nuclei in the paper, but this is the gist of it:

"The most important isotope in this context is 210 Po, a decay product of the gas 222 Rn. It can be present on or
implanted in the surfaces of the detectors and surrounding material. The 210 Po nuclei decay to 206 Pb, giving a
5.3 MeV α-particle and a 103 keV recoiling lead nucleus. It can happen that the lead nucleus hits the target crystal and deposits its energy there, while the α-particle escapes. Due to its low quenching factor, the lead nucleus can often not be distinguished from a tungsten recoil and thus can
mimic a WIMP interaction."

Jester said...

Anon#2, take Xenon100. A 10 GeV dark matter particle arriving with v \sim 500km/s produces a maximum recoil of \sim 3.5keV. That 1) is already in the region where Leff is measured and 2) should produce almost \sim 1.5 PEs on average. Hence I reckon they would have seen it no matter what. On the other hand, for 5 GeV dark matter the recoil energy is 4 times smaller and there is more wiggle room, for example the limits rely on extrapolation of Leff to lower energies...

Andrew Oh-Willeke said...

10 GeV (or for that matter the entire range from about 7 GeV to 40 GeV, which includes both M1 and M2) is a real desert when it comes to any well motivated hypothetical high energy physics particles - most dark matter candidates are either lighter or heavier. It is too heavy for any plausible composite candidates (even glueballs are too light), much lighter than a top quark or Higgs boson, and well outside the exclusion range established by ATLAS, LEP and others for SM4 or MSSM or other not so minimal SUSY particles (Lubos has argued for a photino or bino, but not very forcefully or specifically.)

So far as I can discern, the only well motivated hypothetical particle that might exist in this mass range and yet still have escaped detection would be a right handed neutrino (or its antiparticle), which should have a cross section of interaction of something on the order of 5*10^-44 (E/[1 MeV])^2 cm^2 (not in quite the same units as CRESST used), just as a neutrino does but is not theoretically constrained to be so light. A right handed neutrino, if it oscillates like ordinary neutrinos do, could also explain the presence of more than one mass peak. And, since it does not interact with the weak force and lacks an electromagnetic charge, it would escape detection in precision electroweak measurements, even as missing energy (which is pretty much all accounted for in those tests).

Astronomical observations through 2007, have also established that the cross section of interaction for dark matter must be "0.5-1.25 cm2 g-1", although I admit to an inabilty to convert those units to the units to the units used by CRESST in its chart.

The proposed combination of mass and cross section for dark matter in this study does not overlap at the 2 sigma level with several of the studies already conducted.

Despite a nominal 4 sigma significance level claimed for this study, it is hard to give this study much credit. Systemic error is patently much greater than claimed, probably mostly due to the messy background.

Kea said...

So perhaps it is time to consider explanations other than an WIMP aether for the three positive results that we now have with keV recoils.

Anonymous said...

Andrew, you can find easily DM candidates in this range of mass. The definition of "motivation" in the theoretical world is very subjective. But scalar DM or vectorial like or hidden fermions with extra U(1) are quite motivated by any top-down approaches can can easily fit with DAMA/CoGENT/CRESST-like results.

NW said...

I'm not sure that I agree that they've really cleared it up. They've quantified their backgrounds better, explained their statistics, but most of this has already been seen. In particular, while they tell us what the accepted events look like in E and LY, they don't in the 2-D plane. This would be helpful for three reasons

1) the M1 point is mostly Ca and W recoils (8% O). Ca has pretty low light, so an obvious question is: do we really need O recoils to explain this? Or is that something that comes out of a fit? This isn't an indictment, but something we should probably see.

2) S/B ~ 1 here. That means that it would be useful to see the B that didn't make it into the S region. That would also help reinforce.

3) One can go below the threshold (which they set due to e-recoil leakage) to see if the rates rise exponentially. One can't use that range for a positive signal, but it's certainly worth knowing if it's consistent.

It's definitely an interesting result, but hopefully this is the beginning and not the final say on the data. CoGeNT has been a real trendsetter in providing access to their data after having done their own analysis first. Hopefully, we'll see more of that. (Not down to the noise level, where theorists have little to offer, but at the analysis and interpretation level.)

JK said...

Also interesting in this context is the CoGeNT talk at TAUP: http://taup2011.mpp.mpg.de/php/downloadPresentationFile.php?type=presentation&sessionid=5&presentationid=280

According to that talk part of the CoGeNT excess is probably due to surface backgrounds. Effectively, this moves the CoGeNT-allowed region to higher masses and smaller cross sections.

Also, Collar & company have re-measured quenching factors in NaI(Tl) (DAMA's target material). According to their results, which are in some tension with older measurements, the quenching factor for Na recoils is *smaller* than what was assumed by DAMA. This would move the DAMA-allowed region to the right.

So, the picture of DM direct detection has changed quite a bit this week, but it looks still as confusing as ever.

Andrew Oh-Willeke said...

"The definition of "motivation" in the theoretical world is very subjective."

What I mean by well motivated is that: (1) a professionally formulated extension of the Standard Model that has not already been ruled out be experiment predicts that (2) a particle of this mass, without electrical charge, that does not have weak force interactions that could have been detected to date exists (since any thing with that mass that can decay from a W or Z in that mass range would have been discoverd by now), and (3) that the extension of the Standard Model serves some purpose to explain experimental data in addition to this particular group of dark matter direct detection experiments, or provides deeper understanding and explanation of Standard Model phenomena that have been observed (e.g. addressing the hierarchy problem or the strong CP problem).

The biggest issue is that weak force interactions have been studied so carefully and not revealed anything in this mass range when they should have if something existing in this mass range due to the democratic nature of weak force decays. Few professionally formulated extensions of the Standard Model have many massive particles that are exempt from weak force interactions.

Anonymous said...

@NW:
>In particular, while they tell us
>what the accepted events look
>like in E and LY, they don't in
>the 2-D plane. This would be
>helpful for three reasons
The events can be both contributed to oxygen and calcium, since the acceptance bands for both nuclei overlap heavily in the region of interest.

>1) the M1 point is mostly Ca and
>W recoils (8% O). Ca has pretty
>low light, so an obvious question
>is: do we really need O recoils
>to explain this? Or is that
>something that comes out of a
>fit? This isn't an indictment,
>but something we should probably
>see.
It does come out of a fit. You know how your considered backgrounds behave in terms of spectral shape, light yield and multiplicity. When looking at figure 11 you see the spectral shape attributed to lead recoils, neutron background and alpha events. The spectral shape does not match the observed events. Gamma leakage, on the other hand, does. But gammas do have a much higher light yield (as seen in figure 12), which is of course included in the maximum likelihood fit.
The result means, that the resolution of the light detector is not sufficient to attribute the events mainly to oxygen or mainly to calcium, giving the two convergence points of the fit.

>2) S/B ~ 1 here. That means that
>it would be useful to see the B
>that didn't make it into the S
>region. That would also help
>reinforce.
You see that in several figures. The spectral distribution of alpha recoils is discussed in section 4.2, an exemplary figure (fig.6) shows the alpha band and allows you even to estimate the events "by hand".
This is done for every type of background, so I am not sure what you mean.
Generally, the significance with which to reject the background hypothesis is based strongly on the spectral shape of the individual background and not simply on statistics assumptions.

Andrew Oh-Willeke said...

Scalar dark matter presumes that "the dark matter consists of an ultralight particle with a mass of 1.1 × 10−23 eV" (and hence beyond detection by DAMA/COGENT/CRESS which are built to find WIMPs. Hypothetical axions are similarly light with "mass in the range from 10−6 to 1 eV/c2." It is easy enough to add a U(1) to a model but you need some sort of reason to do that.

NW said...

@Anonymous:
>The events can be both contributed to oxygen and >calcium, since the acceptance bands for both nuclei >overlap heavily in the region of interest.

This is true at low enough energies and less true at higher energies. My point was more the one below:

>>1) the M1 point is mostly Ca and
>>W recoils (8% O). Ca has pretty
>>low light, so an obvious question
>>is: do we really need O recoils
>>to explain this? Or is that
>>something that comes out of a
>>fit? This isn't an indictment,
>>but something we should probably
>>see.
>It does come out of a fit. You know how your
>considered backgrounds behave in terms of spectral
>shape, light yield and multiplicity. When looking at
>figure 11 you see the spectral shape attributed to lead
>recoils, neutron background and alpha events. The
>spectral shape does not match the observed events.
>Gamma leakage, on the other hand, does. But
>gammas do have a much higher light yield (as seen in
>figure 12), which is of course included in the
>maximum likelihood fit.

>The result means, that the resolution of the light
>detector is not sufficient to attribute the events
>mainly to oxygen or mainly to calcium, giving the two
>convergence points of the fit.

This somewhat misses my point. I understand that the results of the M1 and M2 points come out of a fit (obviously). But it is not inconceivable that in some detectors with small enough alpha backgrounds, there would be some events that simply had to be O recoils. Or, more likely, which statistically implied that there should very likely be O recoils. Or, the pull for an O component could be very weak.

Fig 11 is the total accepted event energy spectrum. Fig 12 is the total accepted event light yield distribution. I am interested in the 2D distribution simultaneously. Perhaps it would not be useful, but in particular Fig 12 has lost some information because presumably the width of the alpha/O/Ca/W bands is different for different detectors.

>>2) S/B ~ 1 here. That means that
>>it would be useful to see the B
>>that didn't make it into the S
>>region. That would also help
>>reinforce.
>You see that in several figures. The spectral
>distribution of alpha recoils is discussed in section
>4.2, an exemplary figure (fig.6) shows the alpha band
>and allows you even to estimate the events "by hand".
>This is done for every type of background, so I am not
>sure what you mean.
>Generally, the significance with which to reject the
>background hypothesis is based strongly on the
>spectral shape of the individual background and not
>simply on statistics assumptions.

This is precisely the point: fig 6 is an exemplary figure for one detector. Fig 8 is another. But there are 6 other detectors, and it would be useful to see the distributions of backgrounds in those. In particular, if one wanted to relax the assumptions about the modeling of those backgrounds and see what effects that has, this would be necessary.

Like I said, this is not an indictment of the analysis. I don't have any issue with what has been done. My principle issue is that I cannot (even approximately) reproduce the result myself with the information available, or ask simple additional questions. I do truly believe it appears there is an excess to be understood, but especially given the apparent tension with other experiments, as a theorist, I cannot even begin to study compatibility of model variations. And, of course, this is what I want to do.

My understanding is that there will likely be more information to come, and I think once that comes, it will be much easier to see whether the "tension" with other experiments as it is described in the paper is reconcilable.

Robert L. Oldershaw said...

Xenon100 has a relevant new preprint posted to arxiv.org on 9/8/11. Cresstfallen, indeed.

Will "WIMPs" ever come in from the cold?

Certainly! The same day that the greased pig "Higgsy" flies into an extra-dimension riding a Randall-Sundrum graviton and escorted by assorted sparticles and magnetic monopoles.

Gotta love that postmodern pseudo-reality!

Anonymous said...

@Robert Oldershaw: No, Xenon100 collaboration hasn't released any arXiv preprint since July 11 this year.

JL said...

One question for Fig.6. I believe the horizontal axis is the recoil energy keVnr. But to get this, they have to use their measured light output devided by the Quenching Factor.

Since Calcium and Oxygen have overlap in LY and they don't know whether it is a Ca or O collision. Which QF to use makes a big difference. For the signal near the 12KeV in the Fig.6, if we change the QF from Ca to O, will move them to the left which make them disappear from the signal region.

NW said...

@JL CRESST has independent light and heat outputs. Most of the energy is deposited in heat (phonons), which changes the resistance of the tungsten film on the crystal, which they can measure. The energy resolution of CRESST is quite good.