Performance, operation and detector studies with the ATLAS Resistive Plate Chambers

Resistive Plate Chambers provide the barrel region of the ATLAS detector with an independent muon trigger and a two-coordinate measurement. The chambers, arranged in three concentric double layers, are operated in a strong magnetic toroidal field and cover a surface area of about 4000 m2. During 2011 the LHC has provided proton-proton collisions at 7 TeV in the center-of-mass frame with a steady increase in instantaneous luminosity, summing up to about 5 fb−1. The operational experience for this running period is presented along with studies of the detector performance as a function of luminosity, environmental conditions and working point settings. Non-event based information including in particular the large number of gas gap currents, individually monitored with nA accuracy, have been used to study the detector behavior with growing luminosity and beam currents. These data are shown to provide, when calibrated, an independent luminosity measurement and a crucial handle for understanding the ATLAS backgrounds well beyond the scope of muon triggering and detection. The measurements presented here allow to plan a strategy for the data taking in the next years and make some predictions about the detector performance at higher luminosities. They also improve the knowledge on RPC detector physics.


Introduction
In 2011, the bulk of the LHC program provided proton-proton collisions at a center of mass energy of 7 TeV with instantaneous luminosities up to 3.65 × 10 33 cm −2 s −1 , totaling an integrated luminosity of 5.6 fb −1 . In addition, the last month of the 2011 running also provided 166 µb −1 of lead-lead nuclei collisions. This document presents the ATLAS Resistive Plate Chambers (RPCs) performance and operational data taking experience during this period, and studies done with data coming from the RPC Detector Control and monitoring System (DCS) during the operation with colliding beams. The main focus is on extracting fundamental knowledge from the present experience which will be useful to extrapolate the behavior at higher LHC luminosity from all the relevant points of view: detector and trigger performance, structural and ageing issues, service refurbishment, upgrade studies. The detector performance is presented in section 3, where also results on the timing and trigger performance are shown. The dynamic stabilization of the working point with respect to the environmental conditions, as described in section 4, effectively improves the detector performance. The careful control of the gas system also enabled to reliably use the detector current as background and luminosity monitor as described in sections 5 and 6. The high granularity of the information has also been used to evaluate the distribution maps of the cavern backgrounds -1 - coming from the p-p interactions, allowing a comparison with the available simulation and an improvement of its description. The accurate and granular knowledge of the beam induced counting rate (through currents) and its linear relation with the LHC luminosity have several applications such as: pointing out locations where shielding could be improved, optimizing the gas flow with respect to the background intensity, setting up a precise monitoring system based on the ratios of local quantities to the instantaneous luminosity.

The ATLAS resistive plate chambers
RPCs [1] provide the barrel region of the ATLAS detector [2] with an independent muon trigger and a two-coordinate measurement. They cover a surface area of about 4000 m 2 and are operated in a strong magnetic, toroidal field. The chambers, (see figure 1) are arranged in three concentric double layers called Middle Confirm, Middle Pivot, Outer Confirm. Each layer is organized in 16 sectors along the azimuthal coordinate (φ ). To facilitate the overlap between sectors, the even ones are slightly smaller than the odd sectors, thus we refer to large (L) and small (S) chambers, respectively. A low-p T (< 10 GeV) trigger requires a projective coincidence between hits in the pivot and the middle confirm layer, while high-p T triggers require hits also in the outer confirm layer.
An ATLAS RPC chamber is made of two detector layers, each consisting of a gas gap coupled to a 2D readout system, providing both η and φ coordinates. A gas gap is made of two 2 mm thick phenolic-melaminic resistive electrode plates, kept at constant distance by insulating spacers. The electrodes delimit a 2 mm wide gas volume which is filled with a mixture of C 2 H 2 F 4 (94.7%)-C 4 H 10 (5%)-SF 6 (0.3%). The external surface of the plates is coated with a thin layer of graphite paint to allow a uniform distribution of the high voltage across the plates. The smoothness of the inner surfaces is enhanced by means of a thin layer of linseed oil. The high voltage (HV) working point is chosen to be 9.6 kV at a temperature of 24 • C and a pressure of 970 mbar. In these conditions, the RPCs work in saturated avalanche mode. The discharge electrons drift in the gas -2 -2013 JINST 8 P02020  Figure 2. (a) The spatial coincidence between η and φ RPC pivot strips generating a High p T trigger threshold in terms of pseudo-rapidity and azimuthal angle coordinates, respectively. (b) The spatial coincidence between η and φ strips of the outer confirm double layer. Shown are data from the Muon Physics stream taken between March and May 2011. and the signal, induced on pick-up copper strips, is read out via capacitive coupling, and detected by the front-end. For a minimum ionizing particle, a prompt charge of about 1 pC is induced on the pick-up strips, delivering in the gas an average total charge of 30 pC. Read-out strips have a typical width of ∼ 30 mm and are grouped in two (η and φ ) read-out panels with strips orthogonal to each other. Custom front-end electronics amplifies, discriminates and converts the detector signals to ECL standard. These signals are trasferred to the on-detector trigger electronics [3] which, by requiring appropriate coincidences in the η and φ detector layers, provides ATLAS with a Level 1 trigger (L1) decision along with the detector data for accepted events.

Detector status and performance
The 2011 data taking was very successful for the RPC system which has been running with a fraction of active readout channels of 97% (out of a total of 370 k) and an average active trigger area of (99 ∼ 99.5)% [4]. The number of gas volumes either disconnected or operated at non-nominal conditions has been below 2%, with 47 (out of 3592) being off due to broken gas inlets and 23 being kept at lower voltage for recovery due to problems caused by insufficient gas refresh rate, giving rise to anomalous operating currents. The physics data taking was stable with around 99% of the collected data flagged as good for physics analyses. Figure 2 shows in strip units the typical detector coverage during the 2011 proton run. The plots reflect the detector chamber structure and show a uniform coverage both for the trigger and the detector hits. The empty regions around φ strip 1400 and the two surrounding sectors come from the different design of the ATLAS detector in the lower sectors where the infrastructure to sustain the whole detector and the access paths for maintenance are located. Figure 3(a) presents the efficiency per gas volume and compares the performance of 2011 with the one in 2010. As can be seen an important improvement was achieved in 2011 coming mainly from: • the dynamic adjustment of the high voltage working point as a function of the local environmental conditions (temperature and atmospheric pressure); • the general detector consolidation; • the precise timing calibration. On the working point adjustment more details are given in section 4. The plot in figure 3(b) shows the RPC timing performance. The profile is sharp and centered in the middle of the readout window indicating that the RPCs can be used as a precise timing system [5]. Figure 4(a) shows the trigger efficiency as a function of the muon transverse momentum p T . Six different muon triggers with increasing momentum threshold are displayed. The plateau efficiency is around 80% for low-p T triggers thresholds (4, 6, 10 GeV); this value is expected because it includes also the detector geometric acceptance. For the high-p T triggers (p T > 11, 15, 20 GeV) the efficiency plateau is around 70% since a further coincidence in the outer plane is required, thus reducing the geometric acceptance. The effect of the detector acceptance is clearly seen when plotting the efficiency as a function of φ as seen in figure 4(b). The region with the two major dips corresponds to the lower ATLAS sectors where the support of the whole detector is located and where some chambers are missing to allow access inside the detector. In addition to this, the structure with small even and large odd sectors following the 8 large toroidal coils over φ can be also seen.
During 2011 the bulk of the proton collisions have been taken with 50 ns bunch spacing following the fact that the LHC was able to achieve higher luminosity and beam stability with respect to the running with 25 ns bunch spacing. Running with higher bunch spacing means that concurrently at equal luminosity the pileup increases and effects on detector efficiency, or linearity and saturation effects may start to play a role. Figure 5 shows the RPC trigger performance as a function of the average number of interactions per bunch crossing, µ. The data correspond to the period of highest luminosity with an average pileup < µ > of 15. No significant dependence on the RPC trigger efficiency is observed.  The L1 Muon Barrel trigger efficiency for the low-p T threshold MU10 and the high-p T threshold MU11 with respect to offline reconstructed combined muon, selected with p T > 15 GeV, as a function of φ . The efficiency has been determined with a tag and probe method using di-muon events. The data used correspond to a total integrated luminosity of 380 pb −1 .  Figure 5. The L1 muon barrel trigger efficiency, for the 2-station threshold MU10 (a) and for the 3-station threshold MU11 (b), with respect to offline reconstructed combined muon as a function of the muon p T . The efficiency curves are shown for different ranges of the average number of interactions per bunch crossing, µ. The efficiency has been determined with a tag and probe method using di-muon events. The data used correspond to total integrated luminosity of 380 pb −1 .

Detector and DCS operation
The DCS is in charge of safely operating and monitoring the detector power system including the detector HV and LV supply. In the RPC DCS [6], a large number of settings (DAC ∼ 4000) and monitoring (ADC ∼ 6500) channels has been integrated into the system to optimize the detector performance and allows a fine monitoring at the level of the single individual gas gap (∼ 3600). The remaining ADC channels are used to monitor with high granularity the current drawn by the -5 -front end electronics and RPC gas and environmental sensors (temperature, atmospheric pressure, relative humidity and gas flow). The ability to control by tuning thresholds, and monitor the current of each RPC gap has shown to be very powerful for the detector operation both for tracing problems and fine tuning the detector. This is particularly important as the RPC performance and ageing is strongly related to the environmental parameters, namely the temperature (T ), the atmospheric pressure (P), and the relative humidity.
The RPCs benefited in the 2011 run from the introduction of several automatic monitoring and control tools to simplify the detector operation and optimize the data taking conditions. The number of shifters was reduced to a single person taking care of the whole ATLAS muon system which includes also trigger chambers in the endcap region and precision chambers.
One improvement was to add to the RPC DCS the full automatic control of the HV settings. These are automatically adjusted • to follow the different LHC beam phases from injection, stable beams, to the final dump; • to compensate for the local changes of the environmental conditions (mainly local temperature and atmospheric pressure); • to automatically check the individual gas gaps currents, and to recalibrate at each end of fill the pedestals; • to provide an online measurement of the cavern background rates and an instantaneous luminosity.

HV working point correction
The gas gain, the noise rate and the dark current of a chamber depend on the environmental parameters following the formula: where V appl is the applied voltage, T , P are the environment measurements and T 0 , P 0 , V eff are the reference environmental values and HV settings. The HV correction factor ρ can be expressed in a modified form: where the introduced α and β factors express the individual correction due to the atmospheric pressure and the temperature, respectively. In the RPCs, the atmospheric pressure measurement along with the data of about 300 temperature sensors located on the chambers across the whole system are used to adjust the nearly 280 HV channels in total. In order to avoid wrong or overcorrected HV set-points, the data from sensors of neighboring chambers are used only if within a range of validity and then combined. Following the limited granularities of the HV channels and of the temperature sensors compared to the number of gas gaps only a coarse correction averaged over a larger volume can be achieved. Conservative settings, which were confirmed from previous studies during detector tests [7] and from data with beam in 2009-2010 [8] were used: a factor α = 0.8 for the atmospheric pressure term and β = 0.5 for the temperature. Furthermore, for detector safety, a limit to the HV (9500 V) for chambers with temperature above 26 • C was set along -6 -2013 JINST 8 P02020 with independent procedures steadily monitoring the gas flow and the gas gap currents which would lower the corresponding HV set point in case of abnormal readout values. During nominal running, the environmental correction updates every few minutes, refreshing the working points of the HV channels which are adjusted with the lowest ramp up/down speed. The correction is automatically disabled for the periods of no beam or during run transition. An example of the HV working point correction as a function of time is shown in figure 6. From the beginning of 2011, the correction described has been stably used allowing a general improvement in the detector performance and efficiency.

Detector monitoring and online measurements
The size and high granularity of the information read out and archived by the DCS is a valuable source of data for detector physics. This information correlated with the detector readout and layer redundancies has allowed to monitor and intervene in case of problems, for instance in case of chambers with high current or insufficient gas flow. Furthermore, the gas gap currents of the RPCs, measured by the DCS with a sensitivity of 2 nA, allow a precise estimation of cavern background and in general beam induced effects. The monitored currents and pedestals, the environmental variables and the beam information are used to estimate the average radiation induced counting rate per surface unit to study beam background and activation effects and their relation with the integrated and instantaneous luminosity, as demonstrated in sections 5 and 6. Part of these features are implemented online by the DCS, allowing instantaneous monitoring and publishing of the background distribution. Figure 7 shows the online distributions over the longitudinal (z) and azimuthal (φ ) coordinates of the beam induced detector currents as measured and normalized by the detector surface for the three double layers of RPC chambers.
-7 - These measurements and the studies presented in the next sections, have suggested to adjust the gas flow in the shutdown after the 2011 run. A new scheme with individually tuned impedances as function of the chamber volume and the expected integrated radiation was installed across the whole detector [9].

Cavern background measurement through RPC gap currents
The detection of an ionizing particle is associated with the production of an electron avalanche in the gas. The total charge delivered has to be compensated by the power supply with a time constant of about 20 ms driven by the total capacitance and resistance of the system [10]. The average value of the detector current, cleaned up from all detector systematic contributions, represents the average charge per second delivered due to the incident particle rate. Knowing the average charge per particle delivered in the gas < Q tot >, the gap current I gap and its area A gap , the average particle rate R incident on each gap can be extracted as: The gap current measurement sensitivity is driven by the precision of the volt-amperometric method used to measure the detector current flowing on a 100 kΩ resistor in parallel to a 10 nF capacitor, placed in series on the HV return wire before joining with the common detector ground. The voltage drop across the resistor is read with CAEN A-3801, a 128 channels ADC module having a sensitivity of 0.2 mV which corresponds to 2 nA, equivalent to about 30 Hz/m 2 of particle rate.
-8 - The instrument can average a programmable number, up to 250, of 1 ms samples, within a readout cycle of about 1 s. Thus the measurement duty cycle is very high compared to the hit counting method which has a duty cycle of about 4 × 10 −5 as given by the readout window of 200 ns for the RPCs, multiplied by the data acquisition rate of the order of 200 Hz. This means that the time necessary to collect an equivalent sample is about 6000 times faster, allowing in principle sensitive measurement at very low luminosity (L = 6 × 10 28 cm −2 s −1 is the lowest measured value). To enhance the statistics and lower the noise, the DCS acquires samples at maximum speed and keeps the average of the current measurements of the last 30 s (being this time interval tunable). As explained above, the working point correction performed by the DCS removes the environmental induced current variations. The gas gaps installed in the ATLAS barrel region are organized in doublets of identical gaps. A chamber can contain up to 4 doublets coupled side by side in φ or in z direction, or both. The statistical combination of all the gaps provides the best possible estimate of the average background rate in the RPC area.
A conversion factor of 30 pC/count at the standard working point of 9600 V, (P = 975 mbar and T = 24 • C) has been estimated for photons at the CERN GIF (Gamma Irradiation Facility) with a 137 Cs source [11] and with a 60 Co source in laboratory tests [12]. Other cases such as neutrons, are neglected in this analysis. This approximation is justified by the fact that on one hand in the barrel the neutron hits are negligible with respect to the photons due to the very low sensitivity [13], while on the other hand a highly saturated regime should strongly suppress the signal charge dynamical range [14].

Offline gap currents analysis
The raw data from the RPC current measurements are reprocessed offline to improve the quality of the measurement cleaning up the noisy channels and improving the statistical error. To cleanup the noise, the facing gaps of the same doublet (the minimal RPC subset measuring the background in 3D) are compared. Having the same surface and position, these are very correlated and expected to measure the same background intensity. The average RPC working current without beam is of the order of 100 nA/m 2 and is subtracted using the calibration data after each beam dump. The best value for the doublet is determined comparing the currents of the facing gaps (see at figure 8). If the difference of the two currents is less than 20%, the average current will be taken as representative of the doublet; otherwise, the lowest of the two is taken. Other checks are applied to avoid fake values due to disconnected gaps or readout failures or known noisy cases. The rate in Hz/cm 2 is obtained by using the 30 pC/count conversion factor as explained above.
The map shown in figure 9 represents the rate values as a function of doublet position within the ATLAS barrel in cylindrical coordinates. The radius corresponds to the chamber layer distance from the beam axis (e.g. middle and outer layer stations); the longitudinal coordinate z is given by the RPC unit identifier along the beam axis (values are given relatively to the interaction point considering the clockwise circulating beam, named beam 1); the azimuthal coordinate φ is given by the sector and semi sector identifiers running clockwise around beam 1. These maps show counting rate in Hz/cm 2 in each bin, the color scale, given to help the plot understanding is generated by a series of proportional shades of color normalized to the maximum and minimum rate values. The z and φ projective averages are also provided, as well as the global average at the bottom right corner -9 - of each table. In general, we observe that the rate tends to increase at higher |z| for the same radius as expected due to the higher value of η. Moreover, a systematic rate difference is seen between odd and even sectors due to the effect of the toroid coils shadowing the even sectors and acting as absorber material. Going more in details two hot regions are visible, in particular in the innermost chamber layer for chamber id along z = ±5. This is due to the known unshielded crack present until 2011 between the calorimeter and the small wheel shielding, called J-disk as described in [17].
-10 - Figure 10. Typical distribution of background abundance as a function of particle type and energy in the muon spectrometer region as predicted in a FLUKA-based simulation.
For the 2012 run, the shielding has been improved to suppress the rate excess. The lower ATLAS sectors (12, 13 and 14) show a reduced counting rate due to the presence of the ATLAS feet steel supports.

Comparison with other measurements in ATLAS
Cavern background refers to the diffuse background from p-p collisions which fills the ATLAS detector cavern. It mostly consists of neutral, low energy and long lived particles. Figure 10 shows the expected background abundance in different areas of the ATLAS muon spectrometer, as -11 -2013 JINST 8 P02020 a function of particle type and energy. This prediction was obtained by a FLUKA based simulation for 7 TeV + 7 TeV p-p collisions. The two most abundant particle types are photons and neutrons. The cavern background study is necessary to quantify the background levels and how these might evolve as a function of the LHC luminosity and beam energies and to verify the Monte Carlo (MC) prediction done before the LHC start with very large uncertainties. This is particularly crucial to check the present detector reliability and to ensure in advance that the upgrades of detectors and shielding have the appropriate performance. Figure 1 shows a view of the ATLAS detector with particular focus on the Muon Detectors which include, besides the RPC spectrometer, also Monitored Drift Tubes (MDT) over the whole rapidity coverage, and Cathode Strip Chambers (CSC) and Thin Gap Chambers (TGC) in the forward region. A combined measurement using data from the Muon system is in preparation [17].
The cavern background is measured by all of the muon detectors to guarantee the maximum possible coverage in terms of solid angle and cavern volume. This allows to study the behavior of the different detector technologies, their sensitivity and systematic errors, while being exposed to radiation. Moreover, the cavern background has been simulated with FLUGG [18], providing the expected rates for each detector type and region. In the barrel region (|η| < 1.05), the comparison between different detectors is ensured by the MDTs and RPCs which are bundled into muon stations everywhere, except in the innermost region. The barrel is characterized by a lower background intensity with respect to the forward region (endcap). The background is largely dominated by photon hits, and has negligible beam halo contamination, and thus allows very clean measurements. The endcap region (1.05 < η < 2.4) is covered by TGCs and MDT chambers installed on separated layers at different distances from the interaction point. In the innermost region and closest to the beam axis 1.9 < η < 2.7 the CSC is the only muon detector installed, and therefore crucial for the background study in this most critical area.
Two principal, complementary measurement strategies have been pursued, based on hits collected through the DAQ and detector currents collected through DCS, respectively. In the first case hits from all the muon chambers are read out via the event data acquisition path. To minimize the bias from the trigger the data come from the minimum bias or random filled trigger stream. The result is sensitive to the pileup effect due to the bunch structure of the beam and to the long tail of the time spectrum of the hits correlated to the bunch crossing. The HV current measurements integrate the signal over a period much longer than the LHC revolution frequency, and are therefore insensitive to the bunch structure and the trigger. In general, the charge signal amplitude depends on the primary ionization which in its turn depends on the particle type. This uncertainty has to be taken in to account through appropriate conversion factors. Table 1 summarizes the principal features of the different methods and detectors potentiality.

Data and Monte Carlo simulation comparison
A FLUGG-based application simulates individual proton-proton collisions whose daughter particles are followed through the ATLAS detector materials. As they enter the logical volumes of the muon detectors, they are scored and converted to particle fluxes according to the respective sensitivity functions which depend on the particle type and energy. More details on the FLUGG-based simulation can be found in [18]. The comparison of RPC data with FLUGG predictions is performed by converting the simulated counts in each scoring volume to an equivalent rate at a given luminosity using the following definition of the MC hit rate [Hz/cm 2 ]: where L corresponds to the instantaneous luminosity at which the data have been sampled, n hit,MC the number of hits on a given surface A, n events,MC the total number of simulated events, ε a multiplication factor to artificially increase the detector sensitivity, thus enhancing the MC statistics, and  (BML). Looking at the φ projection, it is clear that the even sectors show an overall effect of lower counting with respect to the simulation at all z values, an effect which is more pronounced in the outer layer chambers; this overall effect is systematically concentrated in the lowermost sectors. The discrepancy can be separated in two factors: an overall scaling factor error which may be attributed to a conversion factor and/or to the sensitivity function hardwired in the simulation; a local discrepancy due to inaccuracy in material description or measurement errors. An intercalibration measurement to fix the ratio photon sensitivity/charge per count independently will be needed to disentangle the two effects. In both cases we must also consider a bias introduced by the approximations of neglecting the heavy ionizing particles such as the neutrons, which have a non-uniform distribution. This effect would show up as an excess of counts at locations where there is a higher concentration of neutron hits (typically at higher η). A better knowledge of the neutron distribution and detector response function is needed to go beyond the present level of comprehension at this point. The systematic effect on the differences shown in the distributions of figures 11 and 12 seems to be related to passive material which is more concentrated in the ATLAS barrel feet area.

Luminosity and detector activation
The RPC detector currents during normal running show a clear correlation with the LHC operation. This information can be used as a monitor to estimate the instantaneous radiation or to spot any problematic behavior related to bad gas supply or anomalous detector ageing. The correlation of the gap currents with the instantaneous luminosity has been studied aiming at a measurement of the latter with the RPCs. Considering that the RPCs are a few meters away from the beam pipe, the measurement should be insensitive to beam halo background and also to temperature and atmospheric pressure changes due to the automatic adjustment of the high voltage working point. The most straightforward quantity to estimate the luminosity can be obtained by combining all RPC gas gap currents properly pedestal subtracted and normalized to their active surface: where I j are the gap currents, I j 0 the current pedestals taken at same detector voltage with no beam in the machine and A j are the active gap areas. Those few chambers showing a problematic trend due to detector, gas flow or electronics problems are manually flagged and removed from the computation.
For safe operation, the RPCs are automatically set at one of two High Voltage (HV) configurations depending on the LHC beam conditions. These are: • the READY state, with the detector at full voltage/gain/efficiency (∼ 9600 V); the RPC are required to reach that configuration as soon as the LHC declares stable beams; • the STANDBY state, with the detector at lower voltage/gain/efficiency (∼ 9000 V); this configuration is requested whenever the LHC is injecting, ramping or squeezing the beams.
-15 - In order to be less sensitive to changes in the detector settings or the LHC luminosity first measurements were performed for the following two well-defined conditions: • at the beginning of the fill, when the beams start colliding but before the stable beams declaration, in order to avoid the detector ramp up to READY state; • at the end of the fill, when the beams are dumped and before the ramp down of the detector to STANDBY state.
The RPCs HV were at a fixed value during these beam transition phases which made it possible to estimate the contribution to the chamber current due to the beam collisions. In the first case, the detector was in STANDBY state; in the second case in READY state. The typical behavior of the RPCs can be observed by directly looking at a DCS plot (see figure 13), where the current follows the ATLAS instantaneous luminosity. A first positive step of the current can be seen at the beginning of the fill, when the detector was still at lower voltages (9000 V). At the end of the fill a negative step is also visible and occurs when the beams are dumped. Once the fill is over, a residual current (indicated as "Pedestal" in figure 13) is present, corresponding to the detector dark current at 9600 V. The observed current differences ∆I RPC for the two conditions were then associated to a luminosity difference ∆L . Figure 14 shows the instantaneous luminosity versus the RPCs current for the two voltage gains. The data, obtained in several fills, show a clear linear relation between current and luminosity. The slopes of the two curves are different due to the different detector gains.

Measurement technique and results
One of the main systematic effects is the strong influence of the environmental parameters on the detector behavior. A dynamic adjustment of the high voltage working point as a function -16 -  of the local environmental conditions (temperature and atmospheric pressure) was necessary for a stable operation. For technical reasons it was only possible to apply this correction when the detector was at full voltage (9600 V). In this case the RPCs current was expected to be strictly correlated to the beam collisions and not influenced by other factors. Consequently the analysis efforts have been concentrated on the extraction of the corrected values of current at beam dumps. Figure 15 shows a collection of several measurements of average current density (pedestal sub--17 - stracted) versus instantaneous luminosity at beam dump. The measurements refer to 2010 and 2011 data and span over a range of more than 4 decades. The data fit a straight line with a slope of 0.312 ± 0.001 nA m −2 /10 30 cm −2 s −1 and a negligible intercept. This result shows that the ratio between the ATLAS luminosity and the RPCs current (∆I) is remarkably constant. Thus, demonstrating that an independent online luminosity measurement is possible once calibrated. Data from 2010 were taken without the environmental correction from the DCS active and an equivalent correction was applied offline on the data.
As a byproduct of this analysis, it was possible to extrapolate the fit to the LHC design luminosity of L = 10 34 cm −2 s −1 . From the extracted value of the current, an average counting rate of ∼ 10 Hz/cm 2 was predicted, using a 30 pC/count conversion factor. This result together with several 2-dimensional maps of the current have been extensively used for ATLAS upgrade studies, in order to have a reliable estimation and description of the hit rate for higher LHC luminosity. Looking in more detail at the current trends after beam dump, an interesting effect of the ATLAS cavern activation has been observed. A very important effect that had to be understood, quantified and constantly monitored. This observation was possible even if the time needed by the DCS to update and calculate the average value of the currents was not optimal. Approximately every 30 seconds an independent data set is provided, thus implying a step in the RPCs current 30-40 seconds after a beam dump. Rather than dropping to the expected detector dark current (∼ 100 nA/m 2 ), the trends were instead described by an exponential decay function.
The lifetime (τ) and the intensity (A 0 ) of the exponential decay function have been extracted, analysing data for several fills with different instantaneous luminosities at beam dump. In order to have a reasonable amount of data to fit, the detector was kept regularly at full voltages for 20 minutes after each dump. Figure 16 shows the results for a sub-sample of LHC fills where the RPCs current is shown as a function of time after beam dump for instantaneous luminosities ranging from 0.6 to 3.1 · 10 33 cm −2 s −1 . The results of the fits are summarized in table 2.
The decay rate is almost independent from the instantaneous luminosity and has an estimated average lifetime of < τ >= (234 ± 1) s. This result is similar to the one found by the other ATLAS -18 - Table 2. The distributions displayed in figure 16 have been fit with an exponential decay function y = A 0 * exp(−t/τ) with < τ >= (234 ± 1) s.

Instantaneous Luminosity
3  figure 17, the total current, the current induced by activation and their ratio (activation current/total current) are shown as a function of the instantaneous luminosity at beam dump. The activation current (A 0 ) appears to depend linearly on the instantaneous luminosity. The ratio has been calculated and found to be almost constant with luminosity. The average value of the ratio is (4.1 ± 0.1)%.

Comparison with ATLAS luminosity measurements and future prospects
The ratio between the ATLAS luminosity [19] and the current in the RPCs is remarkably constant as shown in figure 15. Before providing an independent luminosity measurement some control -19 - Figure 18. Percentage differences between the ratio R(t) and the reference R(t 0 )(red square) calculated at beam dump for fills with a different number of colliding bunches. checks were necessary. As a first step, it was decided to monitor through the DCS the online ratio: ∆I RPC (t) (6.2) in order to see if any systematic deviation was present during the LHC fill. No major effects were observed. A further systematic check was the extraction of the ratio R(t) at the beam dump for several fills where the conditions of the RPCs detector or the LHC bunch configuration were different. Some examples of modified conditions are listed here: • before and after the LHC technical stops, when major interventions on the detector occurred; • different gap selection (used by the averaging algorithm); • different LHC bunch configuration. Figure 18 shows a quantitative comparisons between the reference ratio R(t 0 ), extracted from a calibration run (at beam dump) and the ratios calculated for several fills with different number of colliding bunches during the 2011 data taking. It can be noticed that a maximum deviation of ±1.5% was found; for the other systematic checks, similar or smaller fluctuations have been observed. The agreement and the stability of the result can be considered satisfactory as can be seen also for a single luminosity fill in figure 19. In order to quote the RPCs luminosity for 2012, a new reference ratio R(t 1 ) will be used to calibrate the current with the ATLAS online luminosity measurement. This is mainly due to the changes in the ATLAS shielding and LHC beam energies with respect to 2011. Once this ratio is fixed, it will be possible to provide an independent measurement based on the following formula: L RPC = R(t 1 ) · ∆I RPC (6.3) During the whole 2012 run, RPCs are providing ATLAS with an independent luminosity measurement that is used to monitor and understand potential discrepancies. However, more detailed studies regarding the stability of the measurement will be needed to further reduce the systematic -20 - Figure 19. ATLAS and RPC Online instantaneous luminosity. The ratio between the two quantities is remarkably constant allowing, once calibrated, for an independent online luminosity measurement and studies on cavern activation background at beam dump.
uncertainty from the actual ±1.5% to a value below ±1%. The fact that the activation current was found to scale linearly with the luminosity and that the activation time is much shorter with respect to the fill length, means that it will not be necessary to add a further systematic to the measurement.

Conclusions
The ATLAS RPCs have worked very well in 2011, delivering good trigger and data for physics. The detector redundancies along with the extensive large monitoring capabilities allowed to quantify the linear correlation between the LHC luminosity and the RPCs average current over four orders of magnitude. A precise estimation of the cavern activation has been extracted and found to be constant as a function of the luminosity. A map of the cavern background as measured by the RPCs has been extracted and compared to the available Monte Carlo simulation. The discrepancies found have allowed a better estimation and an improvement of the available simulation. The encouraging results obtained will be verified with the 2012 data taking, where RPCs are measuring the online luminosity together with the other ATLAS measurements. The running of an extended system as the ATLAS RPC has provided a unique opportunity to improve the understanding of the physics of the detectors beyond that obtained from previous studies of aging and performance generally limited by the small number of detectors under test. The experience gained in the detector operation will help to plan the data taking for the next years and the system upgrades in view of the future high luminosity runs at the LHC.