1: \documentclass{elsart}
2: \journal{}
3:
4: \usepackage[latin1]{inputenc}
5: \usepackage{longtable}
6: \usepackage[dvips]{graphicx}
7: \usepackage{epsfig}
8: \usepackage{amssymb}
9: \usepackage{amsfonts}
10: \usepackage{amsmath}
11: \usepackage{mathrsfs}
12: \usepackage{hhline}
13: \usepackage{array}
14: \usepackage{graphicx}
15: \usepackage{dcolumn}
16: \usepackage{bm}
17: \usepackage{lscape}
18:
19: \begin{document}
20: \begin{frontmatter}
21:
22: \title{Test of pulse shape analysis
23: using single Compton scattering events}
24: %
25: \author{I.~Abt},
26: \author{A.~Caldwell},
27: \author{K.~Kr\"oninger$^{**}$},
28: \author{J.~Liu},
29: \author{X.~Liu$^*$},
30: \author{B.~Majorovits}
31: %
32: \address{Max-Planck-Institut f\"ur Physik, M\"unchen, Germany}
33: %
34: \corauth[cor]{email: {\tt xliu@mppmu.mpg.de},
35: Tel. +49-(0)89-32354-337}
36: %
37: \corauth[cor]{Current address: II. Physikalisches Institut, Universit\"at G\"ottingen, Germany}
38: %
39: \begin{abstract}
40: %
41: Compton scattering is one of the dominant interaction
42: processes in germanium for photons with an energy of around two MeV.
43: If a photon scatters only once inside a germanium detector,
44: the resulting event contains only one electron which normally
45: deposits its energy within a $mm$ range.
46: Such events are similar
47: to $^{76}$Ge neutrinoless double beta-decay (0$\nu$$\beta\beta$) events
48: with just two electrons
49: in the final state.
50: Other photon interactions like pair production or
51: multiple scattering can result in events
52: composed of separated energy deposits.
53: % which in turn tag the
54: %the event as photon induced.
55: One method to identify the
56: multiple energy deposits is the
57: use of timing information contained in
58: the electrical response of a detector or a segment of a detector.
59:
60: The procedures developed to separate single- and multiple-site
61: events~\cite{cite:siegfried_dep_psa}
62: are tested with specially selected event samples
63: provided by
64: an 18-fold segmented prototype germanium detector for Phase II of
65: the GERmanium Detector Array, GERDA~\cite{cite:gerda_white_paper}.
66: The single Compton scattering, i.e. single-site,
67: events are tagged by coincidently
68: detecting the scattered photon with a second detector
69: positioned at a defined angle.
70: A neural network is trained to separate such events from
71: events which come from multi-site dominated samples.
72: Identification efficiencies of $\approx$~80\%
73: are achieved for both single- and multi-site events.
74: %
75: \end{abstract}
76: %
77: \begin{keyword}
78: double beta decay, germanium detectors, pulse shape analysis
79: \PACS 23.40.-s \sep 14.60Pq \sep 29.40.-n
80: \end{keyword}
81: %
82: \end{frontmatter}
83:
84: %\newpage
85: %\tableofcontents
86:
87: \newpage
88:
89: % --------------------------------------------------------
90: % introduction
91: % --------------------------------------------------------
92: \section{Introduction}
93: \label{section:introduction}
94:
95:
96: Photons of energies around 2~MeV
97: have a high probability to interact in Germanium
98: through Compton scattering. The mean free
99: path of the process is a couple of centimeters.
100: If a photon Compton scatters only
101: once inside a germanium detector, the recoiling
102: electron deposits its energy most likely within a 1~mm range,
103: resulting in a so-called single-site event (SSE).
104: If, in contrast, a photon interacts through pair production
105: or scatters multiple times,
106: energy can be deposited at different locations separated
107: by typically a few centimeters,
108: resulting in a so-called multi-site event
109: (MSE).
110: The charge carriers created by the energy
111: deposition in the germanium detector
112: drift towards the anode and cathode of the detector.
113: While the charge amplitude of the induced pulse
114: is determined by the number of carriers
115: (thus by the energy deposited), the pulse shape
116: is determined by the location(s) of the energy deposition(s)
117: and thus the charge drifting times.
118: MSEs are expected to have more involved
119: pulse shapes than SSEs, and thus, pulse shape analysis (PSA)
120: can be used to separate the two classes of
121: events~\cite{cite:siegfried_dep_psa,cite:first_scs,cite:HdMo,cite:HdMo_dep,cite:igex,cite:dep_part_majorana}.
122:
123:
124: One application of PSA is the background rejection
125: in experiments searching for neutrinoless double-beta
126: decay (0$\nu$$\beta\beta$)
127: in $^{76}$Ge-enriched detectors,
128: such as the GERDA experiment~\cite{cite:gerda_white_paper,cite:siegfried_mc}.
129: The expected 0$\nu$$\beta\beta$ signal events
130: have two electrons in the final state
131: with a total energy
132: of 2.039~MeV. These are mostly SSEs.
133: A large fraction of the expected background events
134: are induced by external photons
135: with energy depositions around the Q-value.
136: These events are expected to be predominantly MSEs
137: which can be rejected by PSA.
138:
139:
140: In order to study and improve the performance of PSA,
141: SSE- and MSE-dominant data samples have to be collected
142: independently of the pulse shape.
143: In this paper a method to collect single Compton scattering
144: events (SCS) as an SSE-dominant sample is
145: investigated in more detail.
146: The energy of the scattered photon in an SCS event can be calculated
147: given the incoming photon energy and the scattering angle.
148: Therefore, SCS events can be collected
149: by positioning a second germanium detector
150: at a specific angle with respect to the first detector
151: and using it to tag escaped
152: photons with the correct energy~\cite{cite:first_scs}.
153: If the incoming photon
154: has an energy of 2.614~MeV
155: as emmitted by a $^{208}$Tl
156: source, a photon Compton scattered at 72$\textdegree$
157: has an energy of 575~keV.
158: This signature is used to tag
159: the single recoiling electron
160: inside the first germanium detector.
161: The energy in the event is
162: equal to the germanium 0$\nu$$\beta\beta$ Q-value.
163: The location of the energy deposition of the electron
164: within the detector volume
165: is controlled by positioning
166: the source and the second detector correspondingly.
167:
168:
169: Another common method to collect an SSE-dominant sample
170: is to select the double-escape
171: events (DEP)~\cite{cite:siegfried_dep_psa,cite:HdMo,cite:HdMo_dep,cite:dep_part_majorana}.
172: The incoming photon interacts with the germanium detector
173: through pair production and
174: the two 511~keV photons from the positron annihilation
175: escape the detector without further interaction.
176: The electron and positron mostly deposit their energies
177: very locally and result in an SSE.
178: Another useful sample contains
179: so-called single-escape
180: events (SEP) where only one 511~keV photon escapes.
181: The other photon mostly deposits its energy
182: at locations different from those of the
183: electron and positron.
184: Thus, SEP events provide an MSE-dominant sample
185: with energy deposition close to the 0$\nu$$\beta\beta$ Q-value.
186:
187:
188: However, the DEP events are not a perfect test sample for
189: the expected 0$\nu$$\beta\beta$ events.
190: If the two photons escape the detector,
191: the interaction point is more likely
192: close to the detector surface as compared to SCS events.
193: 0$\nu$$\beta\beta$ events, on the other hand,
194: are distributed evenly within the detector volume.
195: In addition, DEP and 0$\nu$$\beta\beta$ events have different energies.
196: A DEP event induced by a 2.6~MeV photon from a $^{208}$Tl source
197: has an energy of 1.59~MeV,
198: quite different from the 0$\nu$$\beta\beta$ Q-value.
199: In these respects studies with SCS samples suffer less
200: from systematic effects.
201:
202:
203: The experimental setup and the data collection
204: are described in chapter~\ref{chapter:experiment}.
205: The Monte Carlo simulation is also included in this chapter.
206: It is used to verify that the collected SCS samples are SSE-dominated.
207: In chapter~\ref{chapter:psa}
208: a PSA package based on an artificial neural network (ANN) is presented.
209: The training methods are described and the results given.
210:
211:
212: %\input{introduction/introduction}
213: %\clearpage
214: % --------------------------------------------------------
215: % experiment set and data selection
216: % --------------------------------------------------------
217: \section{Experimental setup, data selection and MC simulation}
218: \label{chapter:experiment}
219:
220:
221:
222:
223: %In this chapter, first the experimental setup
224: %is introduced. The collected data samples
225: %are then presented, followed by the
226: %description of the Monte Carlo.
227:
228:
229: \subsection{Experimental setup}
230: \label{section:setup}
231:
232:
233: The experimental setup is
234: illustrated in Figure~\ref{fig:setup}.
235: \begin{figure}[th!]
236: \center
237: \includegraphics[scale=0.65]{coin_setup.eps}
238: \caption{Schematic of the experimental setup with the 18-fold segmented
239: germanium detector as target and the REGe detector to tag photons
240: at 72$\textdegree$.
241: The dotted lines illustrate the segment boundaries.}
242: \label{fig:setup}
243: \end{figure}
244: The segmented germanium detector under study is a prototype detector
245: for Phase-II of the GERDA experiment~\cite{cite:gerda_white_paper}.
246: The true coaxial 18-fold segmented $n$-type HPGe
247: dectector has a weight of 1.63~kg
248: and the dimensions are
249: 69.8~mm height and 75.0~mm diameter;
250: the inner hole has a diameter of 10.0~mm.
251: The segmentation scheme is 3-fold along
252: the vertical axis and 6-fold in the azimuthal angle.
253: (see Figure~\ref{fig:setup}).
254: Signals from the 18 segments and the core
255: of the detector are amplified by charge sensitive pre-amplifiers and
256: read out by a Pixie4 DAQ system~\cite{cite:pixie4_daq}
257: with 14-bit ADC's at a sampling rate of 75~MHz.
258: The resolution (FWHM) of the core is $\approx$3.5~keV at 1.3~MeV
259: and those of the segments are between 2.5 and 4.0~keV.
260: A time resolution of roughly 10~ns can be achieved with
261: the sampling rate used.
262: This corresponds to a position resolution
263: of $\approx$1~mm inside the detector volume.\footnote{
264: The typical drift velocity of the charge
265: carriers inside a germanium detector is
266: $\approx$1~cm per 100~ns.}
267: More information about the segmented detector
268: and the DAQ system can be found
269: in~\cite{cite:siegfried_characterization}.
270:
271:
272: A 100~kBq $^{228}$Th source is positioned
273: at a distance of 23$\pm$1~cm from the center of the segmented detector
274: and faces the center point of segment 14, as illustrated
275: in Figure~\ref{fig:setup}.
276: A second non-segmented and well-type germanium detector,
277: a Canberra REversed Germanium detector (REGe)~\cite{cite:rege},
278: is positioned at the same height
279: with the closed end facing the segmented germanium detector.
280: The distance from the closed end surface to the center of the
281: segmented detector is 23$\pm$1~cm.
282: The REGe crystal is 60~mm in height and 65~mm in diameter.
283: It has a resolution (FWHM) of 2.3~keV at 1.3~MeV.
284: It is used to tag the photons scattered mostly in segment~14.
285: The geometrical acceptance of the
286: REGe~detector results in recorded SCS events with
287: scattering angles
288: between $\approx$~65$\textdegree$ and $\approx$~80$\textdegree$
289: corresponding to energy depositions in the segmented detector
290: between $\approx$~1940~keV and $\approx$~2110~keV.
291: The precision of the alignment of the REGe detector with respect to
292: the $^{228}$Th source and the segmented detector is $\approx$~5$\textdegree$.
293:
294:
295: The energy thresholds for all channels are set to 100~keV.
296: A coincidence trigger is required between the
297: core of the segmented detector and the REGe
298: with a coincident time window of 500~ns.
299: Due to a technical limitation of the coincidence trigger of the DAQ system,
300: only four channels could be read out.
301: Thus, for each coincidence trigger,
302: only the energies of the core ($E_{Core}$),
303: segment 14 ($E_{Seg14}$), segment 17 ($E_{Seg17}$)
304: (below segment 14, as illustrated in Figure~\ref{fig:setup})
305: and the REGe ($E_{REGe}$) were recorded.
306: 300 time samples were taken for each pulse for all 4 channels.
307: This corresponds to a time window of 4~$\mu$s
308: including 1~$\mu$s before the arrival of the trigger.
309: In this analysis, however, only the core pulses are
310: used for the PSA.
311:
312:
313: The actual coincidence trigger rate was $\approx$~12~Hz.
314: The independent trigger rates of the segmented detector and
315: of the REGe detector were both $\approx$2000~Hz.
316: This results in an accidental coincidence rate of $\approx$2~Hz.
317: The coincidence trigger rate
318: without the $^{228}$Th source is $<$0.1~Hz.
319: Therefore, without further cuts,
320: $\approx$~20\% of all events
321: are expected to originate from accidental coincidences.\footnote{
322: The fraction is expected to differ for different energy ranges,
323: as the trigger rate varies.}
324: However, the fraction of accidental coincidence events
325: among the selected SCS events is negligible,
326: as discussed in the next section.
327:
328:
329: \subsection{Event selection}
330:
331:
332: In total 360~000 coincident events were collected.
333: Four different data samples are selected:
334:
335: \begin{itemize}
336:
337: \item $\Gamma_{SCS}$ : Single-Compton-Scattering (SCS) events
338:
339: \hspace{0.7cm} $|E_{Core}+E_{REGe} - 2614.5 | < $ 5.0~keV
340:
341: $and$ $1940<E_{Core}<2090$ keV
342:
343:
344: %$and$ $E_{REGe}$$<$580.2 or $E_{REGe}$$>$586.2 keV
345: $and$ $|E_{REGe}-583.2 | > $ 3.0~keV
346:
347: \item $\Gamma_{2.6}$ ~~: events with the 2.6~MeV
348: photon fully absorbed in the segmented detector
349:
350: \hspace{0.7cm} $|E_{Core} - 2614.5 | < $ 5.0~keV
351:
352: \item $\Gamma_{DEP}$ : DEP events
353:
354: \hspace{0.7cm} $|E_{Core} - 1592.5 |< $ 5.0 keV (Two 511~keV photons escape.)
355:
356: \item $\Gamma_{SEP}$ : SEP events
357:
358: \hspace{0.7cm} $|E_{Core} - 2103.5 |< $ 5.0 keV (One 511~keV photon escapes.)
359:
360: \end{itemize}
361:
362: The $\Gamma_{SCS}$~sample is selected through three cuts.
363: The allowed window of $\pm$\,5~keV of the sum energy of both detectors
364: around 2614.5~keV covers about three times the combined
365: energy resolution (3$\sigma$) of the detectors.
366: The geometrical acceptance for SCS~events extends to 2110~keV, but
367: SEP~events would contaminate the sample, as they have
368: a core energy of $E_{Core}$~=~2103.5~keV in this setup.
369: They are excluded by removing events with the
370: core energy of the segmented detector above~2090~keV.
371: The $^{208}$Tl decay also produces 583.2~keV photons with
372: a branching ratio of 84.5\%.
373: To avoid coincidences orginating from
374: these photons an energy window of $|E_{REGe}-583.2 | <$~3.0~keV
375: is excluded.
376:
377:
378: The single-segment events are selected from
379: each data sample by additionally requiring
380: \begin{itemize}
381:
382: \item single-segment requirement:
383:
384: $|E_{seg14}-E_{Core}|<5.0$~keV $or$ $|E_{seg17}-E_{Core}|<5.0$~keV
385:
386: \end{itemize}
387: The single-segment event samples are noted as
388: $\Gamma_{SCS}^{S}$, $\Gamma_{2.6}^{S}$, $\Gamma_{DEP}^{S}$ and
389: $\Gamma_{SEP}^{S}$, respectively.
390:
391:
392: The coincidence trigger is only relevant for
393: the $\Gamma_{SCS}$ sample.
394: However, the other samples are selected
395: out of the collected coincident events
396: to ensure the same experimental conditions.
397: In principle the REGe detector could also
398: be used to tag 511~keV photons
399: for events in the $\Gamma_{SEP}$ and $\Gamma_{DEP}$ samples.
400: However, the statistics available is not sufficient.
401:
402:
403: The distribution of the energy of the core, $E_{Core}$,
404: of all coincident events is shown
405: in Figure~\ref{fig:energy_data_mc}a.
406: \begin{figure}[th!]
407: \center
408: \includegraphics[scale=0.75]{paper_energy_data_mc.eps}
409: \caption{$E_{Core}$ distributions a) for all coincident events
410: b) for events with $E_{Core}$+$E_{REGe}$=(2614~$\pm$~5)~keV.
411: The 8 selected samples are indicated.
412: The predicted distributions from the Monte Carlo are shown as well.}
413: \label{fig:energy_data_mc}
414: \end{figure}
415: The $E_{Core}$ distribution of all single-segment coincident events
416: is shown in the same plot. Also shown are the simulated spectra
417: which will be discussed in the next section.
418: Figure~\ref{fig:energy_data_mc}b
419: shows the $E_{Core}$ distribution
420: for all coincident events with
421: $E_{Core}+E_{REGe} = (2614.5 \pm 5.0)$~keV.
422: The arrows indicate the $E_{Core}$ range
423: corresponding to the acceptance angles for
424: the $\Gamma_{SCS}$ sample.
425: \begin{table}[th!]
426: \center
427: \caption{The numbers of events in all data samples are
428: presented in the first row.
429: For the $\Gamma_{SCS}$ sample,
430: $f_{c}$ in the second row
431: corresponds to the fraction of events
432: with $|\Delta T|>$~107~ns.
433: For the $\Gamma_{2.6}$, $\Gamma_{DEP}$ and $\Gamma_{SEP}$,
434: it corresponds to the fraction of events in
435: the central peaks of the $|\Delta T|$ distributions.
436: The ratios of event numbers for data and MC
437: are given in the third row
438: with statistical errors only.}
439: \begin{tabular}{|c|c|c|c|c||c|c|c|c|}
440: \hline
441: sample & $\Gamma_{SCS}$ & $\Gamma_{2.6}$
442: & $\Gamma_{DEP}$ & $\Gamma_{SEP}$
443: & $\Gamma_{SCS}^{S}$ & $\Gamma_{2.6}^{S}$
444: & $\Gamma_{DEP}^{S}$ & $\Gamma_{SEP}^{S}$ \\ \hline
445: \#events & 6,716 & 25,780 & 6,898 & 10,093
446: & 642 & 1,131 & 1,059 & 411 \\ \hline
447: $f_{c}$ [\%] & $>$99
448: & 78$_{\pm1}$
449: & 87$_{\pm1}$
450: & 85$_{\pm1}$
451: & 97$_{\pm4}$
452: & 78$_{\pm2}$
453: & 87$_{\pm3}$
454: & 82$_{\pm4}$
455: \\ \hline
456: %\# data/MC & 0.97 & 1.52 & 1.26 & 1.26
457: % & 1.13 & 1.42 & 1.29 & 1.37 \\ \hline
458: \# MC/data [\%] & 103$_{\pm1}$ & 66$_{\pm1}$
459: & 80$_{\pm1}$ & 79$_{\pm1}$
460: & 88$_{\pm3}$ & 70$_{\pm2}$
461: & 78$_{\pm2}$ & 73$_{\pm4}$ \\ \hline
462: \end{tabular}
463: \label{table:number_of_events}
464: \end{table}
465:
466:
467: The DEP, SEP and 2.6~MeV peaks are all
468: prominant in Figure~\ref{fig:energy_data_mc}a.
469: Only the SEP peak is also prominant
470: in Figure~\ref{fig:energy_data_mc}b.
471: The 511~keV annihilation photon that escapes the segmented detector
472: is fully absorbed by the REGe in these events.
473: The DEP~peak disappears because
474: the two 511~keV photons are emitted
475: back to back and only one of the two photons
476: can be tagged by the REGe detector.
477: The numbers of events in all samples are given in the
478: first row of Table~\ref{table:number_of_events}.
479:
480:
481: The time between the arrival of the core trigger ($T_{Core}$)
482: and the REGe trigger ($T_{REGe}$),
483: $\Delta T$ = $T_{Core}$-$T_{REGe}$, is shown
484: in Figure~\ref{fig:triggertime_diff}.
485: \begin{figure}[th!]
486: \center
487: \includegraphics[scale=0.5]{paper_triggertime_diff.eps}
488: \caption{$T_{Core}$-$T_{REGe}$ distributions.
489: The unit of the $x$ axis is the sampling clock.}
490: \label{fig:triggertime_diff}
491: \end{figure}
492: The $\Delta T$ distribution of the $\Gamma_{SCS}$ events
493: has a mean value of -9.4~ns with a RMS of 12.4~ns.
494: %The fraction of events with $|\Delta T|$ larger
495: %than 6 sampling clocks (80~ns), 1-$f_{c}$, is $\approx$~3.5\%.
496: Only one event falls outside the Gaussian peak
497: ($|\Delta T|$ more than 8$\times$13.3=107~ns).
498: This confirms that events in the $\Gamma_{SCS}$ sample
499: are predominantly induced by 2614~keV photons
500: from the $^{208}$Tl decay
501: and the fraction of accidental coincidences is negligible
502: at the 10$^{-4}$ level.
503:
504:
505: The $\Delta T$ distributions of
506: the $\Gamma_{2.6}$, $\Gamma_{DEP}$ and $\Gamma_{SEP}$
507: samples are also shown in
508: Figure~\ref{fig:triggertime_diff}.
509: These $\Delta T$ distributions
510: are composed of ``signal'' peaks at $\Delta T$$\approx$0
511: and flat distributions of accidental coincidences.
512: The ``signal'' events in the $\Gamma_{DEP}$ and $\Gamma_{SEP}$ samples
513: register the 2.6~MeV photon in the segmented detector
514: through pair production
515: with one annihilation photon reaching the REGe detector.
516: The ``signal'' events in the $\Gamma_{2.6}$ sample
517: have another photon from the same $^{208}$Tl decay registered
518: in the REGe.
519: The numbers of accidental coincidence events can be
520: calculated by fitting the $\Delta T$ distributions
521: with $|\Delta T|>80$~ns with a constant function.
522: The fractions of ``signal'' events
523: after subtracting the accidental coincidence events
524: are indicated by $f_{c}$ and
525: given in Table~\ref{table:number_of_events}.
526: The fractions of accidental coincidence events
527: (1-$f_{c}$)
528: agree with the rough estimate of $\approx$20\%
529: from the trigger rates, as explained
530: in Section~\ref{section:setup}.
531: Notice, that most accidental coincidence events
532: in the $\Gamma_{DEP}$, $\Gamma_{SEP}$ and $\Gamma_{2.6}$ samples
533: can be treated as events triggered with only the core of the
534: segmented detector and they
535: are actually classified correctly.
536: This was concluded in~\cite{cite:siegfried_dep_psa} where a
537: detailed study of core~only triggered events was presented.
538:
539:
540: The $\Gamma_{2.6}$, $\Gamma_{DEP}$ and $\Gamma_{SEP}$
541: samples have wider $\Delta T$ distributions
542: than the $\Gamma_{SCS}$ sample.
543: This is an artefact of the fixed 100~keV energy threshold applied
544: to the REGe detector. As the overall rise-time of a pulse,
545: see Figure~\ref{fig:pulse_example}a, does not depend on the energy,
546: the time at which a fixed threshold is reached does.
547: The $\Gamma_{2.6}$, $\Gamma_{DEP}$ and $\Gamma_{SEP}$
548: samples are selected
549: without any cut on $E_{REGe}$. This results in
550: much wider spreads in
551: $E_{REGe}$ and thus in wider
552: $\Delta T$ distributions.
553:
554:
555:
556: \subsection{MC simulation}
557: \label{section:mc}
558:
559:
560: The GEANT4 based Monte Carlo package MaGe~\cite{cite:MaGe}
561: is used to simulate the setup.
562: In order to speed up the computation
563: only the $^{208}$Tl decay is simulated and not
564: the complete decay chain of the $^{228}$Th source.
565: The energies as deposited in the germanium
566: detectors are smeared event by event according
567: to the detector resolutions.
568: The same energy thresholds and the coincidence
569: trigger as for the measured data are applied to the simulated events.
570: The MC is normalized to the data
571: by counting the number of events within
572: the energy region of $E_{Core}+E_{REGe}$=2614$\pm$5~keV,
573: since events satisfying this requirement
574: are almost exclusively induced by the 2614~keV photon
575: from the $^{208}$Tl decay (see previous section).
576:
577:
578: The simulated distributions of $E_{Core}$ are shown in
579: Figure~\ref{fig:energy_data_mc}a and b.
580: The same selection cuts as required for the 8 data samples
581: are applied to the MC events.
582: The data to MC ratios
583: are given in Table~\ref{table:number_of_events}.
584: They agree with the fractions of
585: events with true coincident triggers ($f_{c}$)
586: within $\approx$10\%.
587: The overall excess of data of $\approx$20\%
588: for all but the SCS samples
589: agrees well with the accidental coincidence rate.
590:
591:
592: \subsection{Distinction between MSE and SSE in MC}
593:
594:
595: The variable $R_{90}$ is defined as
596: the radius of the volume that contains
597: 90\% of the total energy deposition in a germanium detector.
598: It is used to study the size of the volume
599: within which the energy is distributed.
600: Details are described in~\cite{cite:siegfried_mc}.
601: The distributions of $R_{90}$ as calculated using MC information
602: are shown in Figure~\ref{fig:r90} for
603: the 8 selected samples.
604: \begin{figure}[th!]
605: \center
606: \includegraphics[scale=0.65]{paper_r90.eps}
607: \caption{
608: R$_{90}$ distributions of the 8 selected samples.
609: The 3 dotted vertical lines on each plot
610: indicate the $R_{90}$ values of 1, 2 and 3~mm.
611: }
612: \label{fig:r90}
613: \end{figure}
614: Events from $\Gamma_{DEP}$ and $\Gamma_{SCS}$ samples
615: mostly have much smaller $R_{90}$ than those from
616: $\Gamma_{SEP}$ and $\Gamma_{2.6}$ samples.
617: $\Gamma_{SCS}$ events have slightly larger $R_{90}$ than
618: $\Gamma_{DEP}$ events
619: due to the higher energy of the recoiling electron.
620:
621:
622: A fraction of the SCS events
623: have relatively
624: large $R_{90}$ ($>$~2~mm).
625: In most of these events the 2.6~MeV
626: photon Compton scatters several
627: times inside the segmented detector
628: before reaching the REGe detector.
629: They still survive the $\Gamma_{SCS}$ cuts due to the relatively
630: large geometrical acceptance of the REGe detector.
631: Events with $R_{90}$$>$2~mm
632: in the $\Gamma_{DEP}$ sample originate from
633: photons not interacting with the detector through pair production,
634: but through multiple Compton scattering, and still depositing
635: the same amount of energy as in DEP events.
636: These events are significantly reduced
637: by applying a single-segment cut, as shown in Figure~\ref{fig:r90}.
638: The fraction of events from the $\Gamma_{2.6}$
639: and $\Gamma_{SEP}$ samples with $R_{90}$$<$2~mm
640: have the high energy photon depositing energy very locally.
641: These fractions of events increase after applying
642: a single-segment cut.
643:
644:
645: The ``position resolution'' of
646: the DAQ is $\approx$~1~mm,
647: as explained in Section~\ref{section:setup}.
648: However, a conservative cut of $R_{90}$$<$2~mm is used to
649: distinguish SSEs from MSEs~\cite{cite:siegfried_dep_psa}.
650: The fractions of SSEs ($f_{SSE}$) in each sample
651: are listed in Table~\ref{table:fraction_of_sse}.
652: The errors on $f_{SSE}$ are estimated by
653: varying the $R_{90}$ cut value between 1 and 3~mm.
654: \begin{table}[th!]
655: \center
656: \caption{Fractions $f_{SSE}$ of
657: events with $R_{90}$$<$2$mm$ in each sample.
658: }
659: \begin{tabular}{|c|c|c|c|c|}
660: \hline
661: sample & $\Gamma_{SCS}$ & $\Gamma_{2.6}$
662: & $\Gamma_{DEP}$ & $\Gamma_{SEP}$ \\ \hline
663: $f_{SSE}$ &
664: % following are results from r90r
665: % 79 $^{+ 4}_{- 5}$ \% &
666: % 31 $^{+ 11}_{- 15}$ \% &
667: % 91 $^{+ 1}_{- 2}$ \% &
668: % 30 $^{+ 9}_{- 10}$ \%
669: % following are results from r90
670: 72 $^{+ 3}_{- 6}$ \% &
671: 10 $^{+ 6}_{- 7}$ \% &
672: 88 $^{+ 1}_{- 2}$ \% &
673: 15 $^{+ 3}_{- 3}$ \%
674: \\ \hline \hline
675: sample & $\Gamma_{SCS}^{S}$ & $\Gamma_{2.6}^{S}$
676: & $\Gamma_{DEP}^{S}$ & $\Gamma_{SEP}^{S}$ \\ \hline
677: $f_{SSE}$ &
678: % following are results from r90r
679: % 93 $^{+ 1}_{- 1}$ \% &
680: % 52 $^{+ 16}_{- 22}$ \% &
681: % 96 $^{+ 1}_{- 1}$ \% &
682: % 57 $^{+ 14}_{- 14}$ \%
683: % following are results from r90
684: 92 $^{+ 1}_{- 3}$ \% &
685: 26 $^{+ 12}_{- 15}$ \% &
686: 96 $^{+ 1}_{- 1}$ \% &
687: 31 $^{+ 6}_{- 5}$ \%
688: \\ \hline
689: \end{tabular}
690: \label{table:fraction_of_sse}
691: \end{table}
692: $\Gamma_{SCS}$ has a smaller fraction of SSEs
693: than $\Gamma_{DEP}$,
694: due to the relatively large selection window.
695: The $f_{SSE}$ fractions for the $\Gamma^S$ samples are larger
696: than for the $\Gamma$ samples, since the single-segment cut
697: already removes most MSE events.
698:
699:
700: If only the segmented detector is used for triggering,
701: $f_{SSE}$~$\approx$78\% for the $\Gamma_{DEP}$ sample,
702: and $\approx$12\% for the $\Gamma_{2.6}$ sample~\cite{cite:siegfried_dep_psa}
703: (89~\% and 30\% for $\Gamma_{DEP}^S$ and
704: $\Gamma_{2.6}^S$ samples, respectively).
705: These values are similar to the ones for coincident events.
706: Therefore, even though accidental coincidences are not
707: simulated by the MC, the $f_{SSE}$ values as presented
708: in Table~\ref{table:fraction_of_sse} can be used to
709: evaluate the data samples
710: collected with the coincidence trigger.
711:
712:
713: If the estimated 1~mm position resolution
714: can be achieved through PSA,
715: the SSEs from each sample should be correctly identified.
716: The PSA procedure is described in the following section.
717:
718:
719: %\input{experiment/experiment}
720: %\clearpage
721: % --------------------------------------------------------
722: % PSA
723: % --------------------------------------------------------
724: \section{Pulse shape analysis}
725: \label{chapter:psa}
726:
727:
728:
729:
730: The same Artificial Neural Network (ANN) package
731: as used in~\cite{cite:siegfried_dep_psa}
732: is used here to perform the pulse shape analysis.
733: The ANN is trained with
734: an SSE sample against an MSE sample.
735: In~\cite{cite:siegfried_dep_psa}
736: $\Gamma_{DEP}$ (without coincidence trigger)
737: was used as the SSE--dominant sample
738: and events in the 1620~keV line
739: (with the 1620~keV photon from $^{212}$Bi decay
740: fully absorbed in the segmented detector)
741: as the MSE--dominant sample.
742: The trained ANN was able to identify
743: both SSE and MSE events with $\approx$~85\% efficiencies.
744:
745:
746: In this study, a similar analysis is performed.
747: The ANN is trained
748: with the $\Gamma_{DEP}$ sample (SSE--dominant)
749: against the $\Gamma_{SEP}$ sample (MSE--dominant).
750: The trained ANN is used to
751: verify that the collected events
752: in the $\Gamma_{SCS}$ sample are SSE--dominant.
753: The results are shown in section~\ref{sec:psa_verification}
754: after a general description in section~\ref{sec:ann_and_r90}.
755:
756:
757: In a second analysis the ANN is trained with
758: the $\Gamma_{SCS}$ against the $\Gamma_{2.6}$ sample.
759: It is shown in section~\ref{sec:psa_results}
760: that the results are consistent.
761:
762:
763: \subsection{General features of the ANN}
764: \label{sec:ann_and_r90}
765:
766:
767: The core pulse of the segmented detector
768: of a typical $\Gamma_{DEP}$ event
769: is shown in Figure~\ref{fig:pulse_example}a.
770: \begin{figure}[th!]
771: \center
772: \includegraphics[scale=0.65]{paper_pulses.eps}
773: \caption{a) Core pulse of one DEP event,
774: $t_{10}$, $t_{50}$ and $t_{90}$
775: are indicated by arrows.
776: b) distributions of the rise time $T_r$=$t_{90}-t_{10}$
777: for the 4 samples under consideration.}
778: \label{fig:pulse_example}
779: \end{figure}
780: The rising part of the pulse contains information
781: about the event structure
782: as explained in Section~\ref{section:introduction}.
783: The time $t_{50}$ is
784: defined as the time at which the pulse has reached
785: 50\% of its maximum\footnote{
786: Pedestals are subtracted by using the information
787: during the 1~$\mu$s interval before the trigger.
788: }.
789: The 20~values before and the~20
790: after~$t_{50}$ are used for PSA.
791: Thus, the selection of the 40 values is
792: independent of the absolute amplitude of the pulse and
793: thus independent of the energy.
794:
795:
796: $t_{10}$ and $t_{90}$ are defined as
797: the times when the pulse
798: reaches 10\% and 90\% of its maximum, respectively.
799: The distributions of the pulse rise time, $T_{r}$=$t_{90}$-$t_{10}$,
800: are shown in Figure~\ref{fig:pulse_example}b.
801: $T_r$ is fully covered by the 40 values
802: which cover a time window of 533~$ns$.
803: The dominance of long risetimes in the $\Gamma_{DEP}$
804: sample reflects the dominance of events close to the
805: detector surface.
806:
807:
808: The ANN package
809: as used here
810: has 40 input neurons for
811: the 40 pulse values.
812: It has two hidden layers with 8 and 2 neurons each
813: and 1 output neuron.
814: The ANN is trained such that
815: a large ANN output ($NN_{out}$) indicates
816: that the event is SSE--like
817: and a small $NN_{out}$ indicates that it is MSE--like.
818:
819:
820: Since both $NN_{out}$ and $R_{90}$ are related
821: to the size of the energy deposition in the detector,
822: a correlation between
823: $R_{90}$ and $NN_{out}$ is predicted.
824: On average events with small $R_{90}$ should have
825: large $NN_{out}$ and vice versa.
826: It is clear that $R_{90}$ is not the
827: only variable that determines
828: the pulse shape.
829: Other, second order effects like the drift anisotropies caused
830: by the crystal structure and inhomogenious doping concentrations
831: also modify the pulse shapes.
832: Therefore, a 100\% correlation between
833: $NN_{out}$ and $R_{90}$ is not expected.
834: The details of this correlation can only be studied with
835: a detailed pulse shape simulation which
836: is beyond the scope of this paper.
837:
838:
839: \subsection{Verification of ANN training with single Compton scattering
840: events}
841: \label{sec:psa_verification}
842:
843:
844: The ANN is trained with the $\Gamma_{DEP}$ sample
845: as SSE--dominant (signal--like)
846: and the $\Gamma_{SEP}$ sample as MSE--dominant (background--like).
847: The training takes 300 iterations.\footnote{
848: The ANN trained with 500 iterations gives similar results.}
849: The trained ANN is then applied to
850: all $\Gamma_{SCS}$ and $\Gamma_{2.6}$ events.
851: It should correctly identify them as single--site and the
852: multi--site events.
853: The $NN_{out}$ distributions for all 4 samples are shown
854: in Figure~\ref{fig:psa_dep_vs_sep}a.
855:
856: \begin{figure}[th!]
857: \center
858: \includegraphics[scale=0.75]{paper_nn_final_version.eps}
859: \caption{
860: a-c) Results of the ANN analysis, if the ANN is trained
861: with the $\Gamma_{DEP}$ sample as SSE--dominant
862: and the $\Gamma_{SEP}$ sample as MSE--dominant:
863: a) $NN_{out}$ distributions for all four samples.
864: b) $\eta^{ANN}_{SSE}$ and $\eta^{ANN}_{MSE}$ vs. $NN_{out}^{CUT}$.
865: The fitted slope $a$, see text, is shown as well.
866: Errors are taken from the MINUIT fit.
867: c) $\epsilon^{ANN}$ vs. $f_{SSE}$;
868: $\epsilon^{ANN}$
869: values correspond to the value of $NN_{out}^{CUT}$
870: giving the maximum fitted slope $a$.
871: Also given are results for the single segment samples indicated
872: by $^S$~(open points).
873: d)
874: Cross--check with the ANN trained
875: with the $\Gamma_{SCS}$ sample as SSE--dominant
876: and the $\Gamma_{2.6}$ sample as MSE--dominant:
877: $\epsilon^{ANN}$ vs. $f_{SSE}$. Results are again
878: also given for single segment events.
879: See text for details.
880: }
881: \label{fig:psa_dep_vs_sep}
882: \end{figure}
883:
884:
885: The $\Gamma_{SCS}$~events have in average larger $NN_{out}$ values
886: than the $\Gamma_{2.6}$ events. The peaks of the distributions
887: are well separated.
888: However, while the distribution for $\Gamma_{2.6}$~events is
889: quite similar to the one for $\Gamma_{SEP}$~events,
890: the distribution for the $\Gamma_{SCS}$~events looks different
891: from the one for $\Gamma_{DEP}$~events.
892: A shift of the peak is expected from the MC simulation, since there is
893: a higher percentage of
894: $\Gamma_{SCS}$~events
895: with $R_{90}$~values above 2~mm indicating an MSE--like structure
896: of the events, see Figure~\ref{fig:r90}.
897: The $\Gamma_{DEP}$ distribution in addition features
898: a plateau towards high $NN_{out}$ values.
899: This is probably an artefact of the spatial distribution of
900: the events which are predominantly close to the surface
901: which also influenced the ANN training.
902:
903:
904: The classification of events using the distributions
905: depicted in Figure~\ref{fig:psa_dep_vs_sep}a
906: is based on a cut in $NN_{out}$, $NN_{out}^{CUT}$.
907: An event is classified as
908: SSE--like, if $NN_{out}$$>$ $NN_{out}^{CUT}$,
909: or MSE--like, if $NN_{out}$$<$$NN_{out}^{CUT}$.
910: For a given value of $NN_{out}^{CUT}$,
911: the survival efficiency for any data sample,
912: $\epsilon^{ANN}$, is defined as the fraction of
913: events in that sample that are
914: identified by the ANN as SSE--like events.
915:
916:
917: The probabilities to correctly identify
918: SSE-- and MSE--like events,
919: $\eta^{ANN}_{SSE}$ and $\eta^{ANN}_{MSE}$,
920: are calculated using the Monte Carlo predictions for the
921: purities $f_{SSE}$ of the samples used,
922: see Table~\ref{table:fraction_of_sse},
923: and using the measured $\epsilon^{ANN}$ for the data samples.
924: A linear dependence
925: $\epsilon^{ANN}=a\times f_{SSE}+b$
926: is assumed.
927: For a given $NN_{out}^{CUT}$,
928: the values for $\epsilon^{ANN}$ are calculated for all samples,
929: a linear fit is performed to obtain the slope
930: and the line is extrapolated to
931: $f_{SSE}$=1 to obtain
932: $\eta^{ANN}_{SSE}$. It is extrapolated to
933: $f_{SSE}=0$ to determine 1-$\eta^{ANN}_{MSE}$
934: (see Figures~\ref{fig:psa_dep_vs_sep}c and d for two fits).
935: The fit procedure takes errors into account.
936: The errors on $f_{SSE}$ are
937: listed in Table~\ref{table:fraction_of_sse}
938: and those on $\epsilon^{ANN}$ are
939: statistical only.
940: The resulting $\eta^{ANN}_{SSE}$ and $\eta^{ANN}_{MSE}$
941: as a function of $NN_{out}^{CUT}$ are shown in
942: Figure~\ref{fig:psa_dep_vs_sep}b.
943: The fitted slope~$a$ is shown in Figure~\ref{fig:psa_dep_vs_sep}b
944: as a function of $NN_{out}^{CUT}$ as well.
945: A clear maximum for $a$ is visible.
946:
947:
948: The correlations between the values of $\epsilon^{ANN}$
949: and $f_{SSE}$ are shown in
950: Figure~\ref{fig:psa_dep_vs_sep}c
951: for the value of $NN_{out}^{CUT}$ which maximizes the
952: slope $a$.
953: The slope $a$
954: does not approach the ideal value of~1,
955: indicating that $f_{SSE}$ and $NN_{out}$ are not fully correlated.
956: This is expected as the predictions for $f_{SSE}$
957: are entirely based on the simple variable~$R_{90}$
958: as discussed in Section~\ref{sec:ann_and_r90}.
959: The results for the single segment samples are also shown.
960: They were subjected to the identical analysis using
961: the equivalent samples for training.
962: The results of the fits are indicated for both single segment and
963: unrestricted event samples.
964:
965:
966: The results for $\eta^{ANN}_{SSE}$ and $\eta^{ANN}_{MSE}$
967: are given in the first two rows of Table~\ref{table:psa_results}
968: with errors deduced from the fits.
969: The ANN can correctly
970: identify both SSE and MSE events at the 75~\% to 80~\% level.
971: The results for the single segment data sets
972: are similar to ones for the unrestricted samples.
973: These results agree in general with the values of $\approx$85~\%
974: as achieved in~\cite{cite:siegfried_dep_psa}.
975: %This is most probably due to
976: %the larger statistics
977: %the $\Gamma_{2.6}$ sample being used
978: %as the MSE--enriched training sample.
979: %It has a smaller contamination of
980: %SSE--like events than the $\Gamma_{SEP}$~sample.
981:
982: The compatability of the points with the linear fits in
983: Figure~\ref{fig:psa_dep_vs_sep}c
984: leads to the conclusion that the
985: SSE--like events in the $\Gamma_{SCS}$~sample are identified
986: with about the same efficiency as in the other samples.
987: This is the most important result of this study indicating
988: that tagged SCS~events can indeed be used to further study
989: pulse-shapes in more detail.
990:
991:
992: \begin{table}[th!]
993: \center
994: \begin{tabular}{|c|c||c|c|}
995: \hline
996: \multicolumn{2}{|c||}{ANN Training} & \multicolumn{2}{c|}{Analysis} \\ \hline\hline
997: SSE-dominant & MSE-dominant & $\eta_{SSE}^{ANN}$ & $\eta_{MSE}^{ANN}$ \\ \hline
998: $\Gamma_{DEP}$ & $\Gamma_{SEP}$ & $74.1_{\pm 2.7}$ \% & $78.3_{\pm 2.8}$ \% \\ \hline
999: $\Gamma_{DEP}^S$ & $\Gamma_{SEP}^S$ & $79.1_{\pm 7.2}$ \% & $74.3_{\pm 6.8}$ \% \\ \hline
1000: $\Gamma_{SCS}$ & $\Gamma_{2.6}$ & $69.0_{\pm 2.1}$ \% & $81.5_{\pm 2.5}$ \% \\ \hline
1001: $\Gamma_{SCS}^S$ & $\Gamma_{2.6}^S$ & $70.2_{\pm 4.3}$ \% & $84.2_{\pm 5.1}$ \% \\ \hline
1002: \end{tabular}
1003: \caption{$\eta^{ANN}_{SSE}$ and $\eta^{ANN}_{MSE}$
1004: with the ANN trained with various SSE-dominant samples
1005: against various MSE-dominant samples.}
1006: \label{table:psa_results}
1007: \end{table}
1008:
1009:
1010: \subsection{Cross-check using SCS~events for ANN training}
1011: \label{sec:psa_results}
1012:
1013:
1014: The same procedure as described in the previous section is repeated with
1015: the ANN trained using the $\Gamma_{SCS}$ ($\Gamma_{SCS}^S$)
1016: as the SSE-dominant
1017: and the $\Gamma_{2.6}$ ($\Gamma_{2.6}^S$) as the MSE-dominant samples.
1018: The values of $\epsilon^{ANN}$ versus $f_{SSE}$ corresponding
1019: to the maximum slope are shown
1020: in Figure~\ref{fig:psa_dep_vs_sep}d.
1021:
1022: The resulting identification probabilities
1023: $\eta^{ANN}_{SSE}$ and $\eta^{ANN}_{MSE}$ are given
1024: in the last two rows of Table~\ref{table:psa_results}.
1025: The ANN can correctly
1026: identify SSE--like events at the 70~\% and MSE--like events
1027: at the 80~\% level.
1028: This confirms again that the selected
1029: SCS~samples are enriched in SSE--like events and
1030: can be used to train the ANN package.
1031:
1032:
1033:
1034: %\input{psa/psa}
1035: %\clearpage
1036: % --------------------------------------------------------
1037: % conclusion
1038: % --------------------------------------------------------
1039: \section{Conclusions and outlook}
1040: \label{chapter:conclusion}
1041:
1042:
1043:
1044: Events with photons Compton scattering only once inside
1045: a germanium detector, SCS events,
1046: can be selected by tagging the scattered photon with
1047: a second germanium detector.
1048: The pulse shapes of these events can be studied and used to
1049: test methods that distinguish between
1050: single-site and multi-site events.
1051:
1052:
1053: In order to collect SCS events
1054: and perform pulse shape analysis,
1055: an 18-fold segmented prototype detector
1056: for the Phase-II of the GERDA experiment
1057: was positioned in front of a $^{228}$Th source.
1058: A second germanium detector was positioned
1059: to record the escaped photons at 72$\textdegree$,
1060: corresponding to 2040~keV energy deposit
1061: in the segmented detector, close to
1062: the $Q$-value of the 0$\nu$$\beta\beta$ decay of $^{76}$Ge.
1063:
1064:
1065: According to the MC simulation
1066: $\approx$72~\% of the collected SCS events
1067: are true SSE events.
1068: The SSE-dominance is verified
1069: by an artifical neural network
1070: (ANN) trained in an independent way.
1071: These SCS events are then themselves used to train the pulse shape analysis
1072: package and thus the trained PSA is able to identify single-
1073: and multi-site events with efficiencies at the $\approx$80\% level.
1074:
1075: Future studies can improve in two ways.
1076: The fraction of SSE events in the collected SCS sample
1077: can be increased by further improving the tagging method.
1078: For example, the whole experimental setup can be shielded
1079: from external photons and
1080: collimators can be positioned between the two detectors.
1081: The Monte Carlo predictions can also be improved.
1082: Currently they are based on the size of the energy
1083: deposits only.
1084: Better predictions
1085: require a detailed pulse shape simulation
1086: which is currently being developed for the detectors under study.
1087:
1088:
1089:
1090: %\input{conclusion/conclusion}
1091: %\clearpage
1092:
1093: \addcontentsline{toc}{section}{Bibliography}
1094: %
1095: \begin{thebibliography}{99}
1096: %
1097: \bibitem{cite:siegfried_dep_psa}
1098: I.~Abt {\it et al.}, arXiv:0704.3016,
1099: submitted to EJC, to be published.
1100:
1101: \bibitem{cite:gerda_white_paper}
1102: S. Sch\"onert {\it et al.} [GERDA Collaboration],
1103: Nucl. Phys. Proc. Suppl. {\bf 145} (2005) 242.
1104:
1105: \bibitem{cite:first_scs}
1106: F. Petry, {\it et al.},Nucl. Instr. and Meth.
1107: {\bf A 332} (1993) 107.
1108:
1109: \bibitem{cite:HdMo}
1110: J. Hellmig and H.V. Klapdor-Kleingrothaus,
1111: Nucl. Instr. and Meth.
1112: {\bf A 455} (2000) 638.
1113:
1114: \bibitem{cite:HdMo_dep}
1115: B. Majorovits and H.V. Klapdor-Kleingrothaus,
1116: Eur.Phys. J. {\bf A 6} (1999) 463.
1117:
1118: \bibitem{cite:igex}
1119: D. Gonz\~alez, {\it et al.}, Nucl. Instr. and Meth.
1120: {\bf A 515} (2003) 634.
1121:
1122: \bibitem{cite:dep_part_majorana}
1123: S.R. Elliott, V.M. Gehman, K. Kazkaz, D-M. Mei, A.R. Yong,
1124: Nucl. Instr. and Meth. {\bf A 558} (2006) 504.
1125:
1126: \bibitem{cite:siegfried_mc}
1127: I.~Abt {\it et al.}, Nucl. Instr. and Meth.
1128: {\bf A 570} (2007) 479-486.
1129:
1130: \bibitem{cite:pixie4_daq}
1131: User's Manual Digital Gamma Finder (DGF) PIXIE-4,
1132: XIA LLC, {\it http://www.xia.com}
1133:
1134:
1135: \bibitem{cite:siegfried_characterization}
1136: I.~Abt {\it et al.}, Nucl. Instr. and Meth. {\bf A 577} (2007) 574
1137:
1138:
1139: \bibitem{cite:siegfried_photon_identification}
1140: I.~Abt {\it et al.}, nucl-ex/0701005, submitted to
1141: NIM, to be published.
1142:
1143:
1144: \bibitem{cite:rege}
1145: Canberra Reverse-Electrode Coaxial Ge Detector,\\
1146: {\it http://www.canberra.com/Products/494.asp}
1147:
1148: \bibitem{cite:MaGe}
1149: M.~Bauer {\it et al.},
1150: Journal of Physics, Conf. Series. {\bf 39} (2006) 362.
1151:
1152: %
1153: \end{thebibliography}
1154:
1155:
1156:
1157: \end{document}
1158: