Вы находитесь на странице: 1из 18

[245]

Croatian Journal of Philosophy


Vol. V, No. 14, 2005
Searching for WIMPs
ROBERT G. HUDSON
Department of Philosophy,
University of Saskatchewan
The WIMP (weakly interacting dark matter) is currently the leading
candidate for what is thought to be dark matter, the cosmological material
claimed to make up almost 99% of the matter of the universe and which is
indiscernible by means of electromagnetic radiation. There are many
research groups dedicated to experimentally isolating WIMPs, and in this
paper we describe the work of three of these groups, the Saclay group, DAMA
and UKDM. This exploration into the recent history of astroparticle physics
serves to illuminate two philosophical issues. First, is confirmatory evidence
more compelling if it coordinates results gleaned from independent
experimental investigations? And secondly, in justifying experimental
conclusions, how strong must this justification be? Are the high standards
set by philosophers, in the spirit of Descartes, relevant to experimental
research?
1. Introduction
Contemporary astrophysicists currently maintain that upwards to 99% of
all matter in the universe is dark, by which they mean matter that is
non-luminous or unable to be detected by means of light or any other
form of electromagnetic radiation (see Delehanty [2005]). There are vari-
ous reasons why astrophysicists believe in the existence of dark matter:
some reasons are observational, dealing with observations of the rotation
curves of spiral galaxies, and some are more theoretical, involving consid-
erations of large scale structure formation, that is, details concerning how
the universe was initially created following the Big Bang. These reasons
are weighty enough that the existence of dark matter is taken as an estab-
lished fact.
As regards the more precise nature of dark matter, there are various
theoretical candidates. Without doubt, the main theoretical candidate for
dark matter is the WIMP (weakly interacting dark matter), claimed to be
a neutralino, a special particle theoretically predicted in the supersym-
metric extension of the standard model of particle physics. In this paper,
our concern will not be the theoretical nature of WIMPs; rather we start
from the recognition that, for most astroparticle physicists, WIMPs exist,
246 R. G. Hudson, Searching for WIMPs
and that the task for most of these physicists is to find them. To this end,
a number of high-profile experiments have been set up to search for WIMPs.
Our goal will be to examine these experiments with the purpose of
establishing two philosophically significant points. First of all we con-
sider the issue whether an experimental result is better confirmed if it is
derived by means of two or more independent experimental routes; in
other words, are experimental results better if they are robust? There
have been many philosophers who have argued for such a thesis. How-
ever, I think it is telling for the cogency of this thesis if we find scientists
explicitly disavowing its significance. If philosophers purport that robust-
ness is a good scientific practice, yet we find publicly accredited scientists
explicitly disavowing robustness, then we should be prepared to question
the methodological value of this epistemic strategy.
Secondly, epistemologists often debate the question of how strong jus-
tification needs to be in order for one to know a claim. For instance some
epistemologists, most notably Descartes, are certaintists: they claim we
should seek justifications that guarantee with practically logical force the
truth of the claim being considered. At the very least, the justification of a
claim should make this claim more likely than it was anticipated to be
beforehand. But what we find with the experimental research I will be
describing is that the relevant claims being considered are justified in an
extraordinarily weak fashion; experimenters are seriously considering
claims which, from an objective view of the evidence, one should perhaps
view as refuted. There are various ways of interpreting this scenario, and
I close the paper by canvassing some of these interpretations.
Our strategy in approaching these issues will be the following. After
briefly reviewing three of the main research groups in WIMP detection,
we begin by describing how one of these groups (the Saclay group) arrives
at WIMP detection rates, and then pause to consider the issue of robust-
ness. We then turn to an examination of the other groups, DAMA and
UKDM, and close with a discussion of the surprising weakness of the jus-
tificatory evidence they use in support of WIMP detection rates.
2. The Research Groups
The first dark matter research group we consider is the Saclay group,
whose members are closely affiliated with the EDELWEISS (Exprience
pour DEtecter Les Wimps En SIte Souterrain) group. Saclay (and EDEL-
WEISS) operates in the underground Laboratoire Souterrain de Modane
in France. The usual approach of EDELWEISS is to deploy cryogenic heat
and ionization detectors, but our focus will be the scintillation experi-
ments performed by the Saclay group in the late 1990s (Gerbier et al
[1999]). We then examine the DAMA (DArk MAtter) group who works in
the underground Gran Sasso National Laboratory in Italy. This group is
famous for having positively identified a WIMP signal by means of an
annual modulation experiment, an identification that has been the sub-
ject of much controversy in the astroparticle physics community. Our con-
[2005]
Access via CEEOL NL Germany
R. G. Hudson, Searching for WIMPs 247
cern here will not be with this controversial work but with DAMAs pre-
liminary experimental work using scintillators performed in the mid 1990s
(see Bernabei et al [1996]). Finally we briefly consider experimental work
done in 1994 by the UKDM (United Kingdom Dark Matter) group involv-
ing scintillation experiments performed in the underground Boulby mine
in England (Smith et al [1996]).
As indicated, all three of these groups are deploying scintillation ex-
periments. But what is a scintillator? A scintillator is a crystal which emits
flashes of light when it interacts with cosmic particles (such as muons),
radiation (such as gamma and beta rays), ambient neutrons and kindred
objects. The sort of crystal used by all three groups is a NaI(T1) (that is,
thallium-activated or thallium-doped sodium iodide). What is it like to
view a scintillator? One might imagine, as an experimenter, staring at a
translucent rock and seeing little light flashes, each indicating some sort
of particle interaction. These experiments are performed in mines so as to
reduce the impact of various undesirable cosmogenic material, specifi-
cally muons, that might mimic WIMP events. Overall, the fundamental
idea is that WIMPs, as theoretically understood, are claimed to interact
with scintillating crystals in a different way than the dominant source of
background noise, that is, ambient gamma and beta radiation. Specifi-
cally, incident WIMPs induce nuclear recoils in the detector material
whereas gamma and beta radiation induce electron recoils (also called
Compton events). This difference provides us with a way of experimen-
tally isolating incident WIMPs, so long as we can shield the detector crys-
tal from other sources of nuclear recoils, such as fast neutrons.
The problem is that it is not easy to make this discrimination between
nuclear and electron recoils. One might imagine the associated flashes of
light having different characteristic wavelengths or different characteris-
tic intensities. As it happens, the method our three groups use to accom-
plish this discrimination between nuclear and electron recoils is by means
of pulse shape discrimination. We describe this method of discrimina-
tion next.
3. Pulse Shape Discrimination Analysis
When a particle bombards an NaI detector, the detector material scintil-
lates, releasing a light pulse. A different sort of light pulse is emitted de-
pending on whether it was produced by a nuclear recoil or an electron
recoil. This difference is captured by the time profile of the pulse.
The time profile of a pulse is identified by means of a photo-multiplier
connected to a Lecroy transient digitizer. A photo-multiplier, while view-
ing a scintillating crystal, retrieves and records light pulse information as
a timed assembly of photo-electrons. The digitizer then displays this se-
ries as a visual representation. To give a flavour of what is seen with a
digitizer, consider Figure 1 which presents the time profile of a single
NaI(Tl) light pulse possessing the small energy of 4 keV pulse occurring
over a 450 ns time span:
248 R. G. Hudson, Searching for WIMPs
Figure 1: Time profile of a single 4 keV NaI(Tl) pulse. From Gerbier
et al [1999], 290.
Depending on the source of the pulsee.g., from a nuclear recoil, an elec-
tron recoil, or background noisesuch time profiles will differ. The goal
for researchers is to find a way to distinguish such profiles as regards
their source. Here, their strategy is to use what is known as a time decay
constant. To get a handle on the notion of a time decay constant, con-
sider the smoothed curve in Figure 2 which might be produced if one
were to smoothly connect the peaks of the pulses in Figure 1. For ease of
exposition, we have simplified the values on the x-axis (time) and y-axis
(height, or pulse amplitude), and have assumed that the peaks are found
at four points, <1,1>, <2,3>, <3,2> and <4,1>.
Figure 2: An imagined data plot with the x-axis signifying time (t
i
)
values and the y-axis signifying height (pulse amplitude, a
i
) values.
R. G. Hudson, Searching for WIMPs 249
The time decay constant is mathematically defined as follows:
<t> = ( t
i
a
i
)/ a
i
where a
i
is the amplitude of the graph at time t
i
. With our four data points,
we can calculate <t> fairly easily:
<t> = ((1 1) + (2 3) + (3 2) + (4 1)) / (1 + 3 + 2+ 1) = 2.4
Intuitively, we find <t> to be the unique value designating the point where
the area under the curve is split in half; it thus shares a conceptual affin-
ity to the half-life characteristic of radioactive substances. For the dark
matter researchers we are considering, the <t> value is highly impor-
tant. For instance, according to Gerbier et al (representing Saclay), <t>
reduces the whole information of the pulse ([1999], 290).
Now, we would anticipate that the decay constants for individual pulses,
whether for nuclear recoils or Compton events, would vary somewhat due
to unanticipated experimental influences. This variation is recorded in
Figure 3.
Figure 3: The number of pulses possessing a characteristic time
constant value (data for the 1214 keV energy range). The <t> value
of 190 ns (the vertical border to the shaded area) indicates where the
two curves are maximally different. From Gerbier et al [1999], 291.
Note that Na recoil pulses stand for nuclear recoils, those recoils we
anticipate to result from WIMP interactions. We find after a number of
calibration trials in which nuclear recoils are artificially induced (by bom-
barding the crystal with neutrons) that such recoils characteristically ex-
hibit time constants of about 200220 ns. For artificially generated elec-
tron (or Compton) recoils (produced by bombarding the crystal with
gamma rays) the average time constant value occurs at about 250260 ns,
as seen in Figure 3.
With the calibration data recorded in Figure 3 at hand, Gerbier et al
[1999] now generate the experimental data. This involves simply letting
the scintillation crystal passively record particle interactions, while view-
ing it with photomultipliers and transforming the result with a digitizer
into a visualizable display. Now, in an experimental run, we anticipate
250 R. G. Hudson, Searching for WIMPs
both nuclear and electron recoils to be present in the data. As such, the
strategy is to compare this data with the calibrated nuclear recoil and
Compton pulse shapes to determine the extent to which the data exhibits
the presence of each kind of recoil.
To see how this works, consider Figure 4 which displays the results of
an experimental run for pulses in the 1015 keV range.
Figure 4: The number of experimentally generated individual pulses
corresponding to various time constant values in the 1015 keV range.
From Gerbier et al [1999], 297.
How do we interpret what is going on here? Recall from Figure 3 that
very few nuclear recoils exhibit time constants above 270 ns. However,
this is the peak of the data in Figure 4. Arguably, then, the data above
270 ns results solely from Compton events, that is, electron recoils. Thus,
the drawn Compton curve exhibits what the data would be like if there
were only Compton events to consider.
But now note that the data below <t> = 270 ns do not coincide ex-
actly with the Compton curve. Apparently, then, there are events here
that are not electron recoils. To find out what else is occurring, Gerbier et
al subtract the Compton curve from the data and generate the smaller,
black histogram shape shown in the graph. The events displayed by the
black histogram are events in excess of what is likely a Compton event.
What could these events be? Surprisingly, these excess events exhibit a
distribution with a time constant value smaller than that anticipated with
nuclear recoils. Hence, Gerbier et al [1999] postulate the existence of a
new population of events, called U events (for Unknown events), which
have decay times smaller than nuclear recoils (297). Later speculation
identifies them as surface interactions that mimic nuclear recoils. In
any event, Gerbier et al decide to regard them as nuclear recoils since
they seek to generate an upper limit on (that is, a maximum rate of)
nuclear recoil/WIMP events.
R. G. Hudson, Searching for WIMPs 251
At this stage, Gerbier et al construct an estimator ([1999], 298). For a
set of experimentally generated data events, let
= the fraction of events exhibiting decay times characteristic of
calibrated nuclear recoils.
(1 ) = the fraction of events exhibiting decay times characteristic
of calibrated (i.e., Compton) events.
R = the calibrated nuclear recoil curve.
C = the calibrated Compton curve.
D = the data set.
Gerbier et al then construct the following estimator:
Hyp(R, C) = R + (1 ) C
Hyp(R, C) is essentially a conjecture about the proportion of nuclear ver-
sus electron recoils on finds in the data. To illustrate how it is used, exam-
ine Figure 5:
Figure 5:
2
analysis (goodness-of-fit) between data results and hy-
pothetical proportions of nuclear recoils (R) and Compton events
(C). From Gerbier et al [1999], 298.
In Figure 5, focus on the curve labelled D & Hyp(R,C). This gives us the
closeness of fit of the data D with a range of hypotheses, from = 0 to
= 0.6. The smaller the value of
2
, the closer the fit of data D with
Hyp(R,C). We can find this point on the graph at about = 0.2. Thus, the
data are most compatible with a hypothesized nuclear (i.e., WIMP) recoil
fraction of 0.2. Utilizing a 90% confidence level, we bear off this mini-
mum, and in seeking the maximum recoil fraction with a 90% confidence
level we arrive at a recoil fraction of .28, here considering recoils with an
energy of 8 to 10 keV.
Gerbier et al perform similar calculations with other energy ranges
and yield the results shown in Table 1.
252 R. G. Hudson, Searching for WIMPs
___________________________________________________________________
Experiment LSM LSM DAMA UKDM
Exposure (kg d) 805 805 4123 1003
Energy interval only stat stat+syst only stat only stat
___________________________________________________________________
46 keV 0.10 0.35 0.120 0.18
68 keV 0.07 0.29 0.080 0.08
810 keV 0.06 0.28 0.036 0.06
1012 keV 0.05 0.16 0.022 0.04
1214 keV 0.04 0.14 0.017 0.03
1416 keV 0.03 0.12 0.012 0.03
1620 keV 0.02 0.12 0.012 0.03
___________________________________________________________________
Table 1: Best fit nuclear recoil fractions in different energy ranges
in different experiments. From Gerbier et al [1999], 300.
Gerbier et als results are labeled LSM (for Laboratoire Souterrain de
Modane). The column LSM (stat + syst) displays results in which Gerbier
et al have included the U population as WIMP candidates (syst stands
for systematic error). LSM (only stat) excludes the U population. Clearly,
if we include members of the U population as WIMP events, the recoil
fraction exhibiting WIMPs increases. Gerbiers preferred calculation of
the nuclear recoil fraction includes the U population, and we see in the
table the value .28 for the 810 keV energy range. From Table 1 we also
see analogous pulse shape discrimination results from DAMA (Bernabei
et al [1996]) and UKDM (Smith et al [1996]).
A key thing to note about these results is that the data from DAMA
and UKDM are more in line with the column dubbed LSM (only stat). So,
if one were an advocate of robustness, that is, if one believed we should be
convinced about the accuracy of results if these results are produced by
independent routes of experimental inquiry, then it seems one should be
prompted to support the results listed under LSM (only stat). For there
are various characteristic differences between the experimental procedures
used by DAMA, UKDM and Saclay groupto wit, these experiments are
performed in different mines in different parts of the world in different
ambient environments. Still, they all produce roughly the same values for
nuclear recoil fractions. Thus, by robustness, we have an argument on
behalf of the claim that these values are more accurate. However, Gerbier
et al explicitly disavow this conclusion; they do not consider the (only
stat) results to be accurate. As they comment, we cannot exclude that
nuclear recoils are hidden in the [subtracted] data ([1999], 299). Thus,
the surprising convergence of results just noted holds, for them, no weight,
and they base their eventual conclusions on what they take to be the more
reliable (stat + syst) results. This provides them, they believe, with the
maximum event rate which cannot be excluded as being WIMP interac-
tions (ibid.). We have here tacit denial of the value of robustness by an
accredited experimental group.
R. G. Hudson, Searching for WIMPs 253
This denial is shared by DAMA, who make explicit their disquiet with
robustness. They comment (referring here to exclusion plots which can
be generated using the recoil fraction data just cited):
let us remark that the safest strategy is to compare results on exclusion plot
and modulation obtained within the same experiment. In particular, the
comparison of exclusion plots obtained by different experiments requires a
consistent use of astrophysical (local density, velocities) and nuclear physics
(matrix elements, spin factors, form factors) parameters. Also the instru-
mental effects (energy threshold, noise rejection capability, detector resolu-
tions and quenching factors) have to be always adequately introduced. More-
over, for different target-detectors further uncertainties could also arise
because of the needed rescaling from the cross section of the different tar-
get-nuclei to
P
(the WIMP-proton elastic cross-section) and because of pos-
sible different unknown or underestimated systematic errors (Bernabei et
al [1998], 196)
The import of this passage is pretty clear: different experimental setups
have too many unknown variables to count on them being accurate in any
simple sense. A full determination of the accuracy of an experiment re-
quires that we take into account these variables. There is no easy way out
of this by means of robustness reasoning. This is significant because many
philosophers have advocated the epistemic significance of robustness (Whe-
well, Pierce, Wimsatt, Hacking, Trout, Culp, to name a fewsee Staley
[2004], 467). But if scientists are motivated to deny the epistemic value of
robustness, then I think we should be wary of its value as well.
4. DAMAs Work
In considering the work by the Saclay group, recall that they looked at the
distribution of <t> values amongst individual pulses (Figure 3). By con-
trast, the DAMA group (and the UKDM group) summarize the informa-
tion contained in a large number of individual pulses in terms of a refer-
ence pulse, where this pulse stands for a particular recoil energy range
and a particular kind of recoil, nuclear or electron (see Bernabei et al
[1996], 761) That is, reference pulses summarize the information gleaned
from a number of individual time profiles for pulses in the same energy
range and produced under similar conditions. Here, one might imagine a
reference pulse as produced by summarizing in one curve a large number
of individual time profiles (such as we displayed in Figure 1 above). Fig-
ure 6 exhibits reference pulses for both nuclear (i.e., Na) and electron
(i.e., Compton) recoils.
254 R. G. Hudson, Searching for WIMPs
Figure 6: Reference pulses for electron and sodium (nuclear) re-
coils in the 1214 keV energy range. From Gerbier [1999], 290.
The data for these reference pulses are generated using calibration runs.
In calibration runs, Na recoils are (artificially) produced by directing a
neutron beam at the detector crystal, and electron (Compton) recoils are
(artificially) produced by exposing the detector crystal to gamma rays.
The time profiles for the resultant scintillations are recorded and then
summarized in a reference pulse. As we see from Figure 6, the time pro-
files of electron and nuclear recoils differthe Na recoil pulse reaches a
peak more quickly than the Compton pulse and subsequently descends
more rapidly. As such, its time decay constant is shorter than the correla-
tive decay constant for the Compton curve. As WIMPs are believed to
induce nuclear recoils, this potentially provides a way to identify incident
WIMPs if we find the decay constants of experimental reference pulses to
be shifted towards the value of the decay constant for a calibrated nuclear
recoil reference pulse.
In service of making this determination, Bernabei et al [1996] experi-
mentally determine distinctive <t> values for nuclear recoils (<t>
recoil
),
Compton events (<t>
comp
) and data (experimental) events (<t>
data
). To
improve on the accuracy of their work, they use a number of WIMP detec-
tors, nine 9.70 kg NaI(T1) crystals and four 7.05 kg NaI(T1) crystals for a
total of thirteen detectors. For each detector we have characteristic val-
ues for <t>
recoil
, <t>
comp
and <t>
data
corresponding to a series of energy
ranges which are typically two keV in extent, ranging from 4 to 20 keV.
What we find is that, when DAMA interpret their experimental re-
sults, they are very liberal in what they take to be results that support or
are even consistent with their theoretical presumptions about the exist-
ence of WIMPs. To illustrate, consider the graphical summary of their
results displayed in Figure 7, comparing for each detector and each en-
ergy range the relevant <t>
comp
and <t>
data
values.
R. G. Hudson, Searching for WIMPs 255
Figure 7: Comparison of <t>
data
and <t>
comp
values for all 13 DAMA
crystals. The line with slope 1 is the hypothesis of perfect compat-
ibility. From Bernabei et al [1996], 764.
At face value, the <t>
data
and <t>
comp
values seem perfectly compatible
with one another within statistical error: the data can be accounted for in
terms of electron recoils, and no mention need be made of nuclear recoils,
that is, WIMP events. But DAMA is undaunted: they proceed to a more
mathematical interpretation of the data. Much as Gerbier et al did, they
seek a precise estimate of the appropriate recoil fraction, , given the
data. Thus, where
= the fraction of events exhibiting decay times characteristic of
calibrated nuclear recoils,
DAMA (Bernabei et al [1996], 764) show that
= 1 <t>
data
<t>
comp
/ 1 <t>
recoil
<t>
comp
Without stopping to consider how this result is calculated, we can appre-
ciate its significance: given the decay constant values for (calibrated)
Compton events, (calibrated) nuclear recoils and data events, we can cal-
culate the nuclear recoil fraction in the generated experimental data, ,
and thus the corresponding fraction of events due to incident WIMPs.
The values DAMA arrive at as results for are found in Table 2.
256 R. G. Hudson, Searching for WIMPs
___________________________________________________________________

___________________________________________________________________
46 bin 68 bin 810 bin l012 bin 1214 bin 1416 bin 1620 bin
___________________________________________________________________
C1 0.29 0.43 0.32 0.23 0.26 0.22 0.024 0.056 0.009 0.039 0.028 0.033 0.027 0.021
C2 0.16 0.26 0.22 0.16 0.26 0.21 0.029 0.045 0.09 0.13 0.022 0.027 0.014 0.018
C3 0.08 0.27 0.321 0.17 0.083 0.071 0.054 0.050 0.024 0.036 0.019 0.031 0.015 0.021
C4 0.16 0.30 0.62 0.42 0.17 0.21 0.079 0.048 0.054 0.034 0.015 0.030 0.041 0.019
C5 0.19 0.31 0.03 0.17 0.002 0.074 0.037 0.050 0.08 0.13 0.058 0.031 0.02 1 0.020
C6 0.89 0.68 0.28 0.19 0.110 0.080 0.050 0.056 0.069 0.039 0.030 0.034 0.019 0.022
C7 0.04 0.25 0.44 0.42 0.26 0.21 0.10 0.17 0.10 0.13 0.049 0.027 0.10 0.11
C8 0.09 0.25 0.000 0.16 0.016 0.070 0.013 0.051 0.004 0.037 0.053 0.032 0.023 0.021
C9 0.30 0.39 0.33 0.22 0.042 0.092 0.13 .18 0.081 0.044 0.12 0.13 0.044 0.025
C10 0.16 0.56 0.21 0.31 0.05 0.12 0.071 0.080 0.090 0.054 0.050 0.047 0.08 0.11
C11 0.31 0.35 0.56 0.45 0.073 0.088 0.018 0.061 0.058 0.042 0.046 0.037 0.027 0.024
C12 0.5 1 0.62 0.63 0.34 0.16 0.14 0.056 0.089 0.031 0.060 0.050 0.052 0.009 0.033
C13 0.81 0.83 0.30 0.38 0.13 0.14 0.161 0.092 0.13 0.15 0.11 0.14 0.07 0.12
Table 2: Recoil fraction, a, for each crystal as function of the en-
ergy bin (or range) in keV; ClC9 are the 9.7 kg detectors, C10-Cl3
are 7.05 kg detectors. From Bernabei [1996], 764.
There is no need to study these values intently, for we can observe some-
thing about them that is obvious and significant. As Bernabei et al [1996]
remark, [these values] are compatible with zero within two standard
deviations (746). Indeed, a number of these values are below 0! This re-
sult is truly startling. Not only is the data weak on behalf of the existence
of WIMPsit is worse than weak, it is disconfirmatory. We find a similar,
surprising phenomenon in DAMAs calculations of the weighted mean
values of , <>, shown in Table 3.
___________________________________________________________________
Energy bin combined upper
(keV) limit (90% C.L.)
___________________________________________________________________
24 -1.42 0.74 0.95
46 -0.162 0.095 0.12
68 -0.091 0.062 0.080
810 -0.028 0.028 0.036
1012 0.030 0.017 0.022
1214 0.023 0.013 0.017
1416 0.0146 0.0097 0.012
1620 0.0005 0.0068 0.0087
___________________________________________________________________
Table 3: Weighted mean of the values for each energy bin. From
Bernabei [1996], 764.
Again DAMA note: these values are compatible with zero within two
[standard deviations] (764). So why does anyone think there are WIMPs
here?
R. G. Hudson, Searching for WIMPs 257
A key part of their analysis of the data is to calculate the combined
upper limits on recoil fractions at a 90% confidence level, seen in the last
column of Table 3. With regard to these values, note that they are identi-
cal to the DAMA (only stat) column in (Gerbier et als) Table 1 (except
for the last row). However, Gerbier et al dismiss the value of these results,
despite their convergence with the UKDM and LSM (only stat) results.
Their reason for this dismissal is precisely the compatibility of these re-
sults with a null result, that is, with the compete absence of a WIMP
signal. They comment,
[DAMAs] conclusion is that the data shapes are fully reproduced by [a]
pure Compton population at all energies, which is incompatible with our
results. (Gerbier et al [1999], 299).
If the data can be fully accounted for by Compton events (i.e., electron
recoils), then there is no need to suppose the existence of incident WIMPs.
Presumably, then, Gerbier et al are expressing some puzzlement here
about why DAMA supposes there to be a WIMP signal in their data.
5. UKDMs Work
Similar to the work of the DAMA group just discussed, the goal of the
UKDM group (Smith et al [1996]) is to determine the nuclear recoil frac-
tion, , in experimental data using reference pulses. On their determina-
tion (and equivalently to DAMAs determination),
a = <t>
comp
<t>
data
/ <t>
comp
<t>
recoil
On the basis of their experiments, which use only one crystal detector,
they retrieve the results graphed in Figure 8.
Figure 8: UKDMs observed nuclear recoil fraction, (in units of
dru, differential rate units). From Smith et al [1996], 304.
258 R. G. Hudson, Searching for WIMPs
In Figure 8, graph (a) is the total observed rate of events and graph (b)
is the total observed rate after subtracting artifactual fast noise pulses
(see Smith et al [1996], 304). Graph (c), 2 standard deviation limits on
nuclear recoil events, is the key graph for us; it is analogous to the values
DAMA cites for <> in Table 3. From inspecting graph (c), one thing is
clear: UKDMs results, similar to DAMAs, are statistically compatible
with the absence of nuclear recoils. As Smith et al [1996] comment:
the points shown correspond to 2 values to show more clearly the consis-
tency with zero signal within the calculated errors. (305).
Smith et al [1996] also note that the apparent, significantly positive re-
sult in the 20 keV range can be discarded since the proportion of elastic,
nuclear recoils should diminish with increasing energy. Thus, just as with
DAMAs results, we are left puzzled as to why Smith et al believe there is
a WIMP signal here.
Of value to us is Smith et als explanation for how, on the basis of their
data, they arrive at values for combined upper limits on recoil fractions at
a 90% confidence level. This explanation correlatively clarifies how DAMA
arrive at the values they record in the last column of Table 3. As Smith et
al comment,
in the absence of systematic errors it would be appropriate to estimate a
90% confidence upper limit for a positive signal by a Bayesean [sic] prescrip-
tion, which ensures that the confidence limit remains above zero and ~
[i.e., close to one standard deviation] even when the central value fluctuates
to negative values. However, since this could overstate the limits if the low
energy negative offsets contain a systematic error, we adopt the more con-
servative approach of assuming the data set to be consistent with zero sig-
nal, and define a 90% confidence limit as 1.3 for each energy range.
([1996], 306)
What are Smith et al suggesting here? My sense of [ensuring] that the
confidence limit remains above zero and [assuming] the data set to be
consistent with zero is that we should not set a nuclear recoil rate below
zero, as this is physically infeasible, but rather we should set this rate to
zero and then set an upper rate, within statistical error, at 1.3X the stan-
dard deviation for each energy bin in order to derive the 90% confidence
limit. So, in this regard, look once more at Table 3 above, and note the
errors (expressed as standard deviations) associated with the <> val-
ues:
0.74, 0.095, 0.062, 0.028, 0.017, 0.013, 0.0097, 0.0068.
As suggested, we set the WIMP interaction rates to zero and then multi-
ply each of these errors by 1.28 (the more precise value than 1.3) to arrive
at the combined upper limit (90% CL), or the values in the third column
of Table 3:
0.95, 0.12, 0.080, 0.036, 0.022, 0.017, 0.012, 0.0087.
R. G. Hudson, Searching for WIMPs 259
These values correspond (almost) exactly to those found in Table 1 above
under the heading DAMA 4123 only stat. Performing similar calcula-
tions UKDM arrive at the results shown in Table 4, where the 90% con-
fidence limit is found in the row discrimination gain factor.
___________________________________________________________________
Energy range (keV) 4-5 57 710 1013 1316 1619 1922 2225
___________________________________________________________________
Corrected rate (dru) 3.75 2.00 1.50 1.95 2.45 2.80 2.95 2.80
Discriminated rate (dru mean) 0.46 0.14 0.05 0.09 0.02 0.15 0.19 0.02
Total error 1.3
c
(dru, Eq. {13}) 0.67 0.15 0.09 0.08 0.08 0.08 0.08 0.07
Discrimination gain factor 0.18 0.08 0.06 0.04 0.03 0.03 0.03 0.025
___________________________________________________________________
Table 4: Upper limits on nuclear recoil rates (see row discrimina-
tion gain factor). From Smith et al [1996], 306.
The discrimination gain factor values are reproduced in Gerbier et als
Table 1 under the heading UKDM 1003 only stat column.
What are we to make of this line of reasoning which asserts a maxi-
mum proportion of recoil events in a scintillation crystal as WIMP inter-
action events? That is, what does it mean to set an upper rate on a kind of
event, where the observed rate is set at zero and the upper rate is due
strictly to statistical error? Is this not an absurd way to reason? Imagine,
by analogy, researchers observing cancer rates in a population of smokers
with the goal of determining whether there is a correlation between smok-
ing and cancer. For example, suppose that the normal cancer rate is 10
2 per 1,000, and that researchers find that cancer rates for smokers are 9
2 per 1,000. Would we have found evidence for the claim that smoking
is correlated with cancer? Quite the reverse! In fact there seems to be a
negative correlation. Now imagine our scientists make the following claim:
on the basis of their new evidence, they reason that the upper limit on a
correlation between smoking and cancer is an extra cancer victim per
1,000 people; that is, the positive correlation between smoking and cancer
is not any greater than this. Is this a reasonable conclusion? Perhaps, but
why would they wish to make this claim at all given that we have re-
trieved evidence that disconfirms the existence of such a positive correla-
tion? The observed facts do not seem to be motivating their conclusion.
Something else must be motivating them.
My view is that we have precisely an analogous situation with the WIMP
detection research we are considering. DAMA and UKDM are asserting a
view that lacks evidential motivation, that there is a WIMP incident rate
that deserves consideration. I conjecture, then, that there is an alternate
explanation for why these researchers feel assured of a WIMP signal in
the experimental data. Let us explore what this alternative explanation
might be.
260 R. G. Hudson, Searching for WIMPs
6. Why does anyone believe there is a WIMP signal here?
I advance four possible explanations for why DAMA and UKDM identify
a WIMP signal in the data.
First of all, the research being performed here is frontier research, in
the sense coined by Cole [1992]. It is research at the leading edge of scien-
tific investigation that explores a completely novel realm of physics. Thus,
there are few, if any, methodological precedents for this research, and so
there are no specific, guiding standards. Rather, all we have are some
general standards concerning the probity of evidence, and as I have shown
these general standards seem to be flouted. But one might suggest that
such abrogation is necessary given that we have very few ideas about what
is occurring in this area of physics, and so we need to take some interpre-
tive risks. After all, what alternatives do we have in this research area? If
there are no sure ways to proceed in understanding the phenomena at
hand, then there are only unsure, tentative ways, and thus we must be
very lenient in what we take as probative evidence.
A second explanation for the willingness of dark matter researchers to
discuss WIMP incidence rates despite the weakness of their data is that
such researchers are working under the realist assumption that there is
dark matter in the universe which takes the form of WIMPs. The theo-
retical nature of WIMPs is outlined in the supersymmetric extension of
the standard model of particle physics and, so understood, WIMPs are
claimed to exist on the basis of various key theoretical arguments (which
we do not review here). These theoretical commitments, despite their specu-
lative nature, are strong enough to drive WIMP detection research and to
compel researchers to assert the evidential significance of their retrieved
results in terms of specifying limits on WIMP interaction rates. Here I
would suggest that there are other areas of science (and of common sense)
where this sort of reasoning applies and is quite cogent. For instance, if
one is having trouble starting ones car, one usually assumes there is a
cause for this flaw. Moreover, one is justified in believing in the existence
of this cause, even if there is no (other) evidence for it, that is, even if all
ones investigations (e.g., checking the fuel level, the ignition, the bat-
tery, and so on) do not reveal it. For people standardly work under the
realist assumption that automobile flaws have physical causes, and this
assumption is not undermined by an inability to isolate these causes. A
similar situation, I claim, pertains to WIMP research: the realist intui-
tions of experimenters regarding WIMPs is unaffected by a dearth of posi-
tive evidence on behalf of WIMPs since these experimenters possess a
prior, theoretical conviction about the existence of WIMPs.
A third explanation for why WIMP researchers suggest the justifica-
tory relevance of weak data is that there has been a significant financial
and scholarly investment made by these researchers and their host scien-
tific institutions in designing, setting up and performing these experi-
ments. There are many, many scientists working on these projects, sta-
tioned in expensive laboratories throughout the world. Moreover, WIMP
R. G. Hudson, Searching for WIMPs 261
research is a field with profound theoretical importance, both for the study
of the creation of the universe (big bang nucleosynthesis) and for investi-
gation into the nature of matter, that is, whether matter consists of (non-
baryonic) WIMPs. With such big investments, with the presence of many
established journals devoted to research on these topics, and with the deep
intellectual desire to understand the fundamental nature of reality, it fol-
lows that any evidence, however weak, will be sought after and nurtured.
In such a case, because of these incentives, any reason to abandon or dis-
miss WIMP research must be quite strong, pragmatically speaking. After
all, what research would take its place that is more secure while just as
important for future, fundamental knowledge?
My final explanation for why WIMP researchers allow the weakness of
the adduced evidence to pass muster is their particular epistemology, one
that allows good evidence and justification to be fairly weak. That is, these
researchers lack the reservations that are characteristic of traditional
philosophers who are trained to keep an extraordinarily skeptical eye. In
particular, one of the classics of Western philosophy is Descartes Media-
tions which counsels us to withhold justificatory assent in the case where
we lack absolute certainty. Our certainty, Descartes claims, must be enough
to confute the existence of a malicious god intent on deceiving humans.
With such high standards, philosophers are usually at a loss to find a case
where we can know things in even mundane, everyday circumstances. So
what hope is there for philosophers to find adequate justification in scien-
tific research? Not much at all. However, WIMP researchers, qua scien-
tists, are unconvinced by (if they have even considered) the sorts of epis-
temological concerns entertained by philosophers. They ignore the tradi-
tional Cartesian worries about standards of evidence, and they ignore fa-
miliar sorts of concerns raised in philosophy of science, such as (Hempelian)
paradoxes of confirmation and theoretical underdetermina-tion problems.
Such abstract and very hypothetical concerns bear no weight in WIMP
inquiry and, as a result, WIMP research is able to progress unburdened
by these presumed epistemic risks.
1
References
Bernabei, R., et al [1996], New limits on WIMP search with large-mass low-
radioactivity NaI(Tl) set-up at Gran Sasso, Physics Letters B 389, 757
766.
Bernabei, R., et al [1998], Searching for WIMPs by the Annual Modulation
Signature, Physics Letters B 424, 195201.
Cole, Stephen [1992], Making Science: Between Nature and Society (Cambridge,
Mass.: Harvard University Press).
1
I would like to thank the Social Science and Humanities Research Council of Canada
for financial assistance, Shayne Kulbacki for a number of invaluable conversations on
the topic of dark matter, and the audience at the 2005 Dubrovnik Philosophy of Science
conference for many perceptive comments.
262 R. G. Hudson, Searching for WIMPs
Delehanty, Marc [2005], Space Science SectionDark Matter, web resource
located at http://www.astronomytoday.com/cosmology/darkmatter.html, ac-
cessed May 23, 2005.
Gerbier, G., et al [1999], Pulse shape discrimination and dark matter search
with NaI(Tl) scintillator, Astroparticle Physics 11, 287302.
Smith, P. F., et al [1996], Improved dark matter limits from pulse shape
discrimination in a low background sodium iodide detector at the Boulby
mine, Physics Letters B 379, 299308.
Staley, Kent W. [2004], Robust

Evidence

and

Secure

Evidence

Claims, Phi-
losophy of Science 71, 467488.

Вам также может понравиться