LArGe - Active background suppression using argon scintillation for the GERDA $0\nu\beta\beta$-experiment

LArGe is a GERDA low-background test facility to study novel background suppression methods in a low-background environment, for future application in the GERDA experiment. Similar to GERDA, LArGe operates bare germanium detectors submersed into liquid argon (1 m$^3$, 1.4 tons), which in addition is instrumented with photomultipliers to detect argon scintillation light. The scintillation signals are used in anti-coincidence with the germanium detectors to effectively suppress background events that deposit energy in the liquid argon. The background suppression efficiency was studied in combination with a pulse shape discrimination (PSD) technique using a BEGe detector for various sources, which represent characteristic backgrounds to GERDA. Suppression factors of a few times $10^3$ have been achieved. First background data of LArGe with a coaxial HPGe detector (without PSD) yield a background index of (0.12$-$4.6)$\cdot 10^{-2}$ cts/(keV$\cdot$kg$\cdot$y) (90% C.L.), which is at the level of GERDA Phase I. Furthermore, for the first time we monitor the natural $^{42}$Ar abundance (parallel to GERDA), and have indication for the $2\nu\beta\beta$-decay in natural germanium. These results show the effectivity of an active liquid argon veto in an ultra-low background environment. As a consequence, the implementation of a liquid argon veto in GERDA Phase II is pursued.


Introduction
GERDA is an experiment to search for the neutrinoless double beta (0νβ β ) decay of 76 Ge. Bare high-purity germanium (HPGe) detectors enriched in 76 Ge, which serve both as source and detector for the 0νβ β -decays, are submersed in liquid argon (LAr). The LAr serves as a high purity shield against external radiation, and as a coolant for the HPGe detectors. The searched 0νβ β -signal is a sharp peak in the energy spectrum at Q β β = 2039 keV which is caused by the sum energy of the two beta particles in a single HPGe detector. Details of the experimental setup and performance are summarised in [1]. The GERDA experiment follows a staged approach: Phase I has been recently completed after acquiring an exposure of 21.6 kg·yr and a background count rate at Q β β of 1·10 −2 cts/(keV·kg·yr) after pulse shape analysis [2,3]. No signal was observed and a limit for the half-life of T 0ν 1/2 > 2.1 · 10 25 yr (90% C.L.) was derived [4]. Phase II is currently under preparation: the goal is to explore half-live values in the range of 10 26 yr by further reducing the background by one order of magnitude to ≤10 −3 cts/(keV·kg·yr), and by collecting an exposure of up to 100 kg·yr quasi background free.

Background suppression in GERDA Phase II
To reach this demanding background count rate, several experimental modifications with respect to Phase I are being implemented: the most important are (1) the additional deployment of approximately 20 kg novel thick-window Broad-Energy Germanium (BEGe) detectors with highly efficient pulse shape discrimination (PSD) performance [5,6], and (2) the implementation of a sensor system to detect the liquid argon scintillation light in anti-coincidence with the germanium detectors for background suppression, first shown in references [7,8].
LArGe, the Liquid Argon Germanium test facility of GERDA, was constructed to study these novel active background suppression methods in a low-background environment [9,10]. Similar to GERDA, bare Ge-detectors are operated in LArGe in 1 m 3 (1.4 tons) of liquid argon, which in addition is instrumented with photomultiplier tubes (PMT). The setup is located underground at 3800 m w.e. at the Laboratori Nazionali del Gran Sasso (LNGS), and is taking data since May 2010. The data presented in this paper demonstrates, that the argon scintillation veto technique works very efficiently both alone, and in combination with PSD applied to BEGe detector signals.

Concept of liquid argon scintillation veto
It is well known that liquid argon emits scintillation light in response to ionizing radiation. Up to approximately 40000 photons, peaked at a wavelength of 128 nm (XUV photons), are emitted per 1 MeV beta/gamma energy deposition [11]. Background events often have energy deposition outside the Ge-detector in the surrounding medium, in our case the scintillating LAr. Conversely, 0νβ β -events are confined to the Ge-detector, so that no scintillation light is triggered. An observation of the light is therefore a good indicator for a background event, and can be used to veto the coincident Ge-signal. For that purpose, the LAr must be instrumented to detect the light. In case of LArGe the light is shifted in its wavelength to match the sensitive range of the PMTs and is guided to the PMTs with mirror foil on the inner cryostat wall.

Detector design
An illustration of the LArGe setup is shown in Fig. 1. A vacuum insulated copper cryostat at the center can hold 1000 l (1.4 t) of LAr. Nine 8" ETL 9357 photomultiplier tubes are immersed into the LAr from the top. The inner cryostat walls are lined with VM2000 radiant mirror foil 1 . PMTs and mirror foil are coated with TPB/polysterene wavelength shifter. The cryostat is surrounded by a passive shield against external background. A double chamber lock system on the top serves as an access port for the deployment of Ge-detectors and internal radioactive sources for calibration. The cryogenic infrastructure, a slow control system, and the DAQ are located adjacent to this setup. 1

Shielding
The graded shield of increasing radiopurity is designed to have the gamma background dominated by the innermost layer of 15 cm electrolytic copper. This is followed by 10 cm of low-activity lead and 23 cm of steel. The outermost layer consists of 20 cm polyethylene to attenuate neutrons. Together, all three layers attenuate an external gamma from the 2615 keV line of 208 Tl to 5·10 −8 of the initial flux. The purest shielding is provided by >40 cm LAr inside the cryostat. At the inside of the cryostat, the PMT glass and the VM2000 mirror foil yield the highest radioimpurities. To maintain the purity concept, the distance of the PMTs to the Ge-detectors is chosen as large as 90 cm. For the mirror foil, the low area density leads to a surface activity below that of the copper. The radiopurity screening results of these LArGe components are given in Table 1. Throughout the measurements presented here, a part of the steel and polyethylene shield was not completed yet.

Cryostat & cryogenics
The main body of the double-wall cryostat is made from electrolytic copper. It has an inner diameter of 90 cm and a height of 210 cm. Heat loss is primarily prevented by the insulation vacuum of 10 −5 to 10 −6 mbar. The uppermost 40 cm collar and bellow are made from stainless steel to minimize the thermal conductivity to the top. The cryostat is closed by a 38 mm thick flange from electrolytic copper. Infrared shields from thin copper foil are mounted both inside the insulation vacuum, and below the top flange. A PMT support structure of copper and PTFE rests on a rim at the copper-steel transition. The filling level of the LAr is adjusted slightly above the PMTs' equator, leaving the upper part of the cryostat in the gas phase of argon. A strong temperature gradient builds up from the copper at LAr temperature, across the stainless steel to the top flange, which itself stays above the freezing point. The total heat load is ∼90 W. An active cooling system cools the cryostat by evaporating liquid nitrogen (LN 2 ) in an integrated cooling spiral in the steel collar. The innermost infrared shield (3 mm thick) is thermally coupled to this heat exchanger to prevent heat lost through radiation. All relevant cryogenic parameters are compiled in a slow control system, which regulates the cooling power via the flow of LN 2 . During normal operation a LN 2 flow of 2.5 kg/h is sufficient to reduce the LAr loss to zero, which allows a continuous operation of LArGe. The working pressure of the cryostat is kept at 30−70 mbar overpressure, to prevent gaseous impurities from the outside to enter the LAr.
The filling procedure for the cryostat requires special precaution to prevent contamination of the LAr. The measurements presented in this article were performed in LAr 5.5 (purity 99.9995%). Against traces of humidity several pumping-flushing cycles were performed with gaseous argon, while the cryostat was heated to >40 • C. To prevent radioactive background from radon, the argon was filled through an active-charcoal trap (602 g of CarboAct), followed by a PTFE particle filter.

Lock & source insertion system
The lock on top of the LArGe setup serves for the deployment of Ge-detector strings and internal radioactive wire sources into the cryostat. So far, only one detector has been inserted at a time. The Ge-detectors are mounted to lowmass copper holders in a separate cleanbench. They are transferred to the main lock using a transportation container, keeping them in gaseous nitrogen atmosphere at all times. The Internal 'close-by' wire sources are inserted directly adjacent to the Ge-diodes in the argon (distance d ≈ 7 cm). External sources are deployed through access ports in the shield to the outer wall of the cryostat (d ≈ 50 cm). The vertical position of all sources matches the center of the Ge-detector.

Light instrumentation
The light read-out is done using nine ETL 2 9357 photomultiplier tubes. The 8" (200 mm) diameter end window with a low resistance bialkali photocathode is sensitive to wavelengths from 275−630 nm 3 . The peak quantum efficiency of the PMTs is 18% at 370 nm. Since the glass of the end window is not transparent for the 128 nm scintillation light, it must be covered with wavelength shifter (WLS). A picture of the PMTs and mirror foil in the setup is shown in Fig. 2.
The PMTs are equipped with a custom made voltage divider with a wide dynamic range from 2 mV to 4 V. Clean pulse shapes are obtained by operating with negative HV on the photocathode. The voltage divider is based on a 0.5 mm thin CuFlon R printed circuit board with components selected for low mass and radiopurity.
The wavelength shifter used to coat Mirror foil and PMTs consists of 10% fluorescent dye (tetraphenyl butadiene) embedded into a polymer matrix of purified polystyrene. Both substances are dissolved in toluene. A coating thickness of 1−4 µm is choosen as a compromise between shifting efficiency and mechanical stability in cryogenic liquid. The coated foil has a specular reflectivity of ∼95% at the peak fluorescence wavelength around 420 nm.
A characteristic quantity of the setup is the photoelectron (pe) yield Y . For the measurements discussed here we had Y ≈ 0.05 pe/keV, which corresponds to an energy threshold of 1/Y ≈ 20 keV for a single photoelectron. Two reasons for this low Y are: (1) only 5 out of 9 PMTs were operational at low temperature, (2) trace contaminations of the argon can strongly quench the scintillation and significantly shorten the attenuation length of the 128 nm photons. An indicator for the light quenching is the lifetime τ of the triplet states of the argon excimers, which can be measured from recorded scintillation waveforms. During data taking τ was monitored in the range between 450 to 550 ns, compared to τ = 1590 ns [14] in clean argon.

Germanium detectors
A modified thick-window broad-energy germanium (BEGe) detector has been used to carry out the suppression-efficiency measurements described in section 3. It is a p-type diode of 878 g by Canberra Semiconductors, N.V. Olen/Belgium. The depletion voltage is +4 kV [15]. A small p+ contact leads to a strong weighting field close to the read-out electrode, which allows good pulse shape discrimination. A prototype of the GERDA Phase I multi-channel charge-sensitive preamplifier (CC2) [16] is used and mounted to the lowmass copper-holder, in which the diode is submersed 'naked' into LAr. A detector resolution of 1.99 keV FWHM at 1332 keV is achieved in this setup, compared to 1.63 keV in a vacuum cryostat with the same detector [5].
For the background measurements described in section 4 the BEGe detector is replaced by the coaxial p-type HPGediode GTF44. In contrast to BEGe it has a low intrinsic background ( 60 Co, 68 Ge) and a high mass (2465 g), whereas pulse shape discrimination is inferior and not applied here. The detector has been modified by the manufacturer Canberra for the bare operation in LAr [17]. It is equipped with a low-background version of the CC2 charge sensitive preamplifier. Within an investigation program of 42 Ar background the diode has been encapsuled in a grounded Faraday cage made from thin layers of PTFE and copper.

Combined Ge-detector and PMT readout & waveform processing
A block diagram of DAQ and front-end electronics is shown in Fig. 3. The Ge-detector is supplied with bias HV (Iseg NHQ 225M NIM). A pulser signal (Ortec Mod. 448 NIM) can be fed into a test input of the preamplifier. The output signal is amplified without shaping by a custom-made linear amplifier, and fed into a FADC (Struck SIS3301 VME; 8 channels, 14-bit, 105 MS/s). The PMT HV is supplied by Iseg NHQ 204M/225M. The signals are amplified by a factor ten (Phillips Scientific Mod. 776 NIM) and merged in a linear fan-in (LeCroy Mod. 428F NIM). The resulting analogsum is amplified in another custom-made analog shaper (NIM) with a shaping constant of a few 10 ns, to match the dynamic range and sampling rate of the subsequent FADC. The FADC is internally triggered on the Ge-signal, and simultaneously records Ge-and PMT waveforms of 40 µs tracelength with 100 MHz sampling rate. The FADC aquisition is controlled by a custom-made software by MIZZI Computer Software GmbH [18].
The offline analysis of the digitized Ge-waveforms is performed with the software framework GELATIO [19]. The deposited energy in the Ge-detectors is reconstructed by applying an approximated Gaussian digital filter. Events generated by discharges or due to electromagnetic noise are rejected by a set of quality cuts following the procedure described in reference [20]. The PMT waveforms are analysed without filtering or quality cuts. Merely the baseline and veto condition are determined.

LAr scintillation veto cut
The veto condition is fulfilled when one or more photoelectrons are detected in a 5 µs window around the Ge-trigger. Threshold and window size are optimized to maximize the product of suppression factor SF at Q β β and the veto acceptance ε acc . The threshold is slightly above noise at 5σ of the baseline spread, corresponding to ∼20% of the average single photoelectron amplitude. The veto acceptance is the complementary probability for an event being vetoed by random coincidences (p rc ), ε acc = 1 − p rc . It is measured by applying the veto cut on pulser signals or single full energy peaks. The random coincidence probability can also be estimated via p rc ≈ ν PMT trig · ∆t, using the PMT trigger rate and the veto window size. It turns out that all methods yield consistent results. Veto acceptance values in measurements with different sources are listed in Table 2.
The suppression factor is the ratio of events in the unsuppressed (N 0 ) versus the suppressed (N S ) spectrum. To make SF independend of the source strength in the measurement it is weighted by ε acc , hence SF = ε acc · N 0 /N S . Suppression factors for the ROI are determined in a 70 keV window around Q β β . Uncertainties are calculated according to Poisson counting statistics.

Pulse shape discrimination
The objective of PSD is to distinguish the single site events (SSE) of the β β -decay from multi site events (MSE) of common gamma-background with multiple interaction vertices within the Ge-diode. SSE and MSE of the same energy (E) can be distinguished by the amplitude (A) of their current pulse. The cut parameter A/E has been established [5]. The PSD cut is calibrated to 90% acceptance on the double escape peak (DEP) of the 2615 keV 208 Tl line, which by its nature is dominantly of SSE character. As discussed in [2], in a BEGe detector the DEP is a good proxy for 0νβ β and their acceptances agree within about 1%. Uncertainties of PSD suppression factors include both statistical and systematic uncertainties.
In addition, the combined suppression of LAr veto and PSD is determined from the spectra. Due to the strong sup-pression of close-by 60 Co and 228 Th the analysis window for these sources is extended from 70 keV width to 200 keV, excluding the single escape peak of 208 Tl at 2104 keV.

Measurement of suppression factors
Energy spectra of various sources in close-by and external position (see 2.1.3) were recorded with the BEGe detector. The sources represent characteristic background contributions in GERDA. While from external sources only gammas can enter the cryostat, close-by sources are encapsuled in ceramics (D = 1 mm) and thin steel (R = 0.25 − 0.5 mm), thus allow some high-energy beta particles to enter the LAr as well. The source activities are chosen to balance high signal rates in the Ge-detector with random coincidences (2.4.2) -see Table 2. Without a source the pulser acceptance is 97.3%, and the PMT trigger rate (∼5 kHz) is dominated by dark noise and the decay of 39 Ar (1.4 kBq).
The LAr veto and PSD cuts are applied to each measurement in the whole energy range. While the achieved suppression factor in the ROI is the ultimately relevant number, other energy regions of distinct gamma lines illustrate the different and complementary suppression mechanism underlying PSD and LAr veto.

Th-228 suppression
Even after careful material screening and selection, 228 Th and its progenies from the natural decay chain are present at trace level in the construction materials of GERDA. Background from sources close-by the germanium originates from detector holders, cables and front-end electronics [3]. These components are immersed in the liquid argon and are referred to as 'close-by sources'. Conversely, 'external sources' are located in the cryostat, its neck and the photomultipliers of the LAr instrumentation itself. The corresponding energy spectra of a close-by and external 228 Th source are shown in Fig. 4.
Both spectra are dominated by the 2615 keV gamma line of 208 Tl and its pendants, the single-and double escape peak (SEP at 2104 keV, DEP at 1593 keV). Since the double escape peak is dominantely of SSE nature, and as such used to calibrate the PSD acceptance, it remains practically unsuppressed by PSD (Fig. 5). On the other hand, the two 511 keV annihilation gammas trigger the LAr veto reliably such that the peak vanishes. In contrast to the DEP, the neighbouring 1621 keV full energy peak from the 212 Bi decay is affected nearly opposite by both cuts: all the energy of this line is deposited in the germanium detector, leaving none for the surrounding LAr to create scintillation. Moreover, the transition is not part of a gamma cascade, thus no additional energy deposition in the LAr is occuring. Hence, the LAr veto does not come in. The suppression factor of PSD is about ten for both source positions. The situation is the same for PSD on the 2615 keV gamma line. However, the LAr veto can also suppress this peak, because the gamma is emitted in a cascade preceded by other gammas (coincident gammas), which can themselves trigger the LAr veto. The suppression is much stronger for the close-by source (SF = 47) than for the external (SF = 1.3), because in the former case coincident gammas have little chance to escape from the active LAr volume. This instance can be exploited to identify the location of a 228 Th background source via the LAr suppression factors.
The ROI of the 0νβ β -decay is dominated by a flat Compton region of 208 Tl before and after the cuts in both spectra. The suppression factors of the LAr veto cleary differ for the close-by (1180 ± 250) and external (25 ± 1) sources, while beeing quasi independent of the source location for PSD (2.4 ± 0.1 and 2.8 ± 0.1 respectively -see Table 3). The LAr suppression is strongly enhanced in the Compton region of the 2615 keV line, as compared to the full energy peak itself: since a fraction of about 2 MeV is deposited in the Ge-detectors, an excess energy of ∼600 keV is deposited in the liquid argon in its vicinity, providing an additional handle for the LAr veto to act upon. The suppression factors of the combined LAr veto and PSD cuts are 5200 ± 1300 (close-by) and 129 ± 15 (external), thus providing a strong suppression on all 228 Th background sources. Since only 5 counts survive the combined suppression of the close-by source, the analysis window was extended from 70 keV to 200 keV width around Q β β . Still, the number of counts after the cut (=15) dominate the uncertainty of the suppression factor. 226 Ra is a long lived progeny of 238 U in the natural decay chain. Similar to 228 Th, it is abundant in trace amounts in all construction materials. 226 Ra decays to 222 Rn, which is a radioactive noble gas that diffuses into the LAr of the GERDA cryostat, and is a potential source of background. Energy spectra were recorded with LArGe for a close-by and external 226 Ra source, both of which are shown in Fig. 6.

Ra-226 suppression
The predominant isotope in the region at Q β β is 214 Bi, with several gamma lines up to 2448 keV. The suppression of gamma peaks follows the same logic as showcased for 228 Th: single lines not beeing affected by the LAr veto (e.g. 1764 keV, 2204 keV, 2448 keV), as opposed to lines emitted as part of a gamma cascade and therefore in coincidence ROI are 4.6 ± 0.2 (close-by) and 3.2 ± 0.2 (external), and about four for PSD (see Table 3). The LAr suppression is much inferior compared to 228 Th for mainly two reasons: (1) all gamma lines with sufficient energy to create Compton events at Q β β are single, hence depriving the LAr veto of a possibility to veto on a coincident gamma. The lack of coincident gammas also makes the veto less dependend on the source position. And (2) the gammas have only little energy exceeding Q β β , thus only few light is created to trigger the veto. Despite the individual suppression of LAr veto and PSD beeing moderate, their combination again provides a suppression well beyond one order of magnitude.

Co-60 suppression
Cosmogenic 60 Co is formed in Ge-detectors and their copper holders during production above ground. While 60 Co in the detectors can create background directly via the beta particle, 60 Co background from the holders relies on gammas: only two gamma lines with significant branching ratio are emitted after the decay of 60 Co. Since their energies of 1173 keV and 1332 keV are below Q β β , they can create background events only via summation. As summation strongly depends on the solid angle and the angular correlation of the gammas, only 60 Co sources close-by the Gedetectors are of concern for the GERDA background. The energy spectrum of close-by 60 Co as measured in LArGe is shown in Fig. 7.
The energy region above the two gamma lines is dominated by the summation spectrum, which expands up to the summation peak at 2505 keV 4 . Similar to 214 Bi, LAr suppression in the ROI happens mainly via the gamma energy exceeding Q β β , which in case of 60 Co is almost 500 keV. This higher energy and multiplicity reduces the chance for it beeing deposited in dead volume rather than active argon, hence leading to a superior suppression factor of 27±2 compared to 226 Ra ( Table 3). The PSD cut works very efficiently with a suppression of 76 ± 9, as by construction summation events are MSE. Again, the combined cut can reject background by three orders of magnitude.

Conclusion on suppression factors
The suppression factors of the different sources in the ROI of the 0νβ β -decay are summarized in Table 3. These numbers sketch the order of magnitude by which active background rejection in GERDA may be achievable, indicating that the goal for Phase II, namely to reduce the background by one order of magnitude, is in reach. The large variation of the suppression factors is consistant with our understanding of the underlying physics, as described in the previous sections.
The combination of LAr veto and PSD proofs to be more powerful than would be expected from independent cuts: the mean average of the combined suppression in the ROI is enhanced by a factor 1.84 ± 0.17, compared to the product of the individual suppression factors of PSD and LAr veto. This means that event classes leading to rejection by one or the other cut are anti-correlated. For example, a Compton event at Q β β from close-by 208 Tl leaves the 2.  gamma to deposit ∼600 keV outside the Ge-detector, alongside its 583 keV coincident partner. However, if the event at Q β β results from the summation of the two gammas (a multisite event likely to be rejected by PSD), a single gamma of higher energy ∼1.2 MeV can leave the detector, and is more likely to escape the active LAr volume than two energetically lower gammas. Hence, this event class is anticorrelated in LAr veto and PSD. An analogue analysis of full energy peaks of the investigated sources returns an average 'enhancement' of 1.007 ± 0.015. This is consistent, since no (anti-)correlated event topologies are expected here. Comparison of the experimental LAr veto suppression factors with full MC simulations are in reasonable agreement and will be reported elsewhere.

Background measurements in natural germanium
LArGe has been designed to demonstrate the applicability of the LAr veto in an ultra low-background environment. For that purpose a measurement with a semi-coaxial detector with natural isotopic composition (GTF44, see 2.3) and improved radiopure front-end electronics was conducted. The LAr veto was operated under the same conditions as previously described, except for an exchange of the LAr by high purity argon 6.0. The veto acceptance of 91% is determined by the pulser. The detector resolution is 3.5 keV at 1332 keV, no PSD was applied.

Background components
The full energy spectrum with an exposure of 116 kg·d (life time 47.05 days) is shown in Fig. 8. The dominant background source above 1.5 MeV is the 2615 keV line of 208 Tl and its Compton continuum. The suppression factor of the full energy peak is 1.52 ± 0.62, which is in agreement with the corresponding value 1.28 ± 0.01 obtained in the external source measurement. Hence, the suppression factor points towards a distant origin of this background, presumably the PMTs. Other prominent background sources are 40 K (1461 keV) and 214 Bi (1764 keV and 2204 keV). Their lines appear only in the vetoed spectrum: while the continuous Compton background is vetoed effectively, single full energy peaks are rejected only by random coincidences. Cosmogenic 58 Co is found at 811 keV, likely sitting in the activated copper of the detector encapsulation. At low energies the spectrum is dominated by the 39 Ar beta spectrum (Q β -value 565 keV) and accompanying Bremsstrahlungs photons.

2νβ β contribution
The strong suppression in the vetoed background spectrum makes it possible to observe the 2νβ β spectrum even so the detector is made from non-enriched natural germanium. The prediction of the 2νβ β spectrum based on [21,22] for this detector infers that 69 out of 135 observed counts in the continuum from 500 keV (above dominant bremsstrahlung from 39 Ar) to 2100 keV (above the 2νβ β endpoint) are expected to stem from 2νβ β decays.

42 K abundance
A unique background to GERDA and LArGe is that of 42   3525 keV, T 1/2 = 12.6 h) is a β -emitting progeny of 42 Ar (Q β = 599 keV, T 1/2 = 32.9 yr) with traces expected in natural argon. In LArGe we observe 7 counts in the interval (1523−1527) keV around the peak, out of which (1.35±0.27) counts are expected from Compton background. The probability 5 to observe ≥7 events from this background is 0.08%. Neglecting possible inhomogeneities of the 42 K spatial distribution, the number of counts observed in the 1525 keV peak corresponds to an abundance 42 Ar/ nat Ar of about 2 · 10 −21 g/g [10]. Along with GERDA [3], the data presented here is the first positive detection of natural 42 Ar.

Background Index
The inlay of Fig. 8 shows the ROI of the 0νβ β -decay. In a large energy window of 300 keV centered around Q β β only one event survives the LAr veto cut. Depending on the chosen width for that region, the achieved background index after veto is about 10 −2 cts/(keV·kg·yr) -see Table 4. The lower limits cover a background index of 10 −2 cts/(keV·kg·yr), which is the design goal of LArGe and GERDA Phase I. The 90% confidence intervals are determined for 'the mean of a Poisson variable in the absence of background' using frequentists statistics according to [23]. The ratio of counts be- 5 The probability is calculated from a gaussian distributed background g(λ |µ, σ ) and poisson distributed counts p(n|λ ), using prob(n ≥ 7) = ∑ ∞ n=7 ∞ 0 g(λ |µ, σ )p(n|λ )dλ .
fore and after the LAr veto yield a background suppression by one order of magnitude or more.

Conclusion
The LArGe test facility has demonstrated the great potential of an active liquid argon veto for the suppression of residual background signals which deposit part of their energy in LAr. It is the first time bare Ge-detectors are operated in a low-background environment with 1 m 3 of instrumented LAr. The background suppression efficiency has been studied in combination with pulse shape discrimination (PSD) of a BEGe detector. Suppression factors have been measured for several sources ( 60 Co, 226 Ra, 228 Th) representing characteristic background sources to GERDA in different locations (close-by and external). The strongest suppression factors were obtained for combined LAr veto and PSD for close-by 228 Th (SF ≈ 5200) and 60 Co (SF ≈ 3900). The combined suppression of LAr veto and PSD is mutually enhanced. The particular response of the different suppression methods is a useful tool to understand the location of different backgrounds even in the case of low counting statistics. In a low background measurement without PSD, the LAr veto helped to achieve an excellent background index of (0.12−4.6)·10 −2 cts/(keV·kg·y) (90% C.L.). The confidence interval coincides with the background level of GERDA Phase I <10 −2 cts/(keV·kg·y), despite LArGe being a much more compact setup. LArGe has the sensitivity to measure the natural abundance of 42 Ar and the 2νβ β -decay in non-enriched germanium. As a consequence of these results, an active liquid argon veto has been developed for GERDA and will be used in Phase II of the experiment.