Performance measurement of HARPO: a Time Projection Chamber as a gamma-ray telescope and polarimeter

TL;DR

This paper evaluates a gas TPC's performance as a gamma-ray telescope and polarimeter, demonstrating its ability to measure polarization and angular resolution in the low-energy pair creation regime, with implications for gamma-ray astronomy.

Contribution

It provides the first measurement of gamma-ray polarization asymmetry in the sub-GeV range using a gas TPC, validating its potential for future polarised gamma-ray observations.

Findings

Excellent agreement between Monte Carlo simulations and experimental data on angular resolution.

First measurement of gamma-ray polarization asymmetry in the sub-GeV range.

Demonstrated the detector's capability to measure polarization in the pair creation regime.

Abstract

We analyse the performance of a gas time projection chamber (TPC) as a high-performance gamma-ray telescope and polarimeter in the e$^+$e$^-$ pair creation regime. We use data collected at a gamma-ray beam of known polarisation. The TPC provides two orthogonal projections $(x,z)$ and $(y,z)$ of the tracks induced by each conversion in the gas volume. We use a simple vertex finder in which vertices and pseudo-tracks exiting from them are identified. We study the various contributions to the single-photon angular resolution using Monte Carlo simulations and compare them with the experimental data and find that they are in excellent agreement. The distribution of the azimutal angle of pair conversions shows a bias due to the non-cylindrical-symmetric structure of the detector. This bias would average out for a long duration exposure on a space mission, but for this pencil-beam characterisation we have ensured its accurate simulation by a double systematics control scheme, data taking with the detector rotated at several angles with respect to the beam polarisation direction and systematics control with a non-polarised beam. We measure, for the first time, the polarisation asymmetry of a linearly polarised gamma-ray beam in the low energy pair creation regime. This sub-GeV energy range is critical for cosmic sources as their spectra are power laws which fall quickly as a function of increasing energy. This work could pave the way to extending polarised gamma-ray astronomy beyond the MeV energy regime.

Equations

$${\theta }_{+-}=\arccos {\vec{u}}_{+}\cdot {\vec{u}}_{-}.$$

$${\theta }_{+-}>\left(0.05+\frac{0.5\mathrm{M}\mathrm{eV}}{E_{\gamma }}\right)\mathrm{rad}.$$

$${\theta }_{\mathrm{pair}}=\arccos ({\vec{u}}_{\mathrm{pair}}\cdot {\vec{u}}_{\gamma }),$$

$${\sigma }_{\det ,{\mathrm{e}}^{\pm }}\approx {\left(\frac{p_{\pm }}{125\mathrm{k}\mathrm{eV}/c}\right)}^{-0.64}.$$

$${\varphi }_{\pm }=\arctan \frac{u_{\pm }^X}{u_{\pm }^Y},$$

$${\sigma }_{\varphi }\approx \frac{{\sigma }_{\theta ,e^{+}}\oplus {\sigma }_{\theta ,e^{+}}}{{\hat{\theta }}_{+-}}=\frac{\sqrt{2}{\sigma }_{\det ,{\mathrm{e}}^{\pm }}}{{\hat{\theta }}_{+-}}$$

Full text

We study the various contributions to the single-photon angular resolution using Monte Carlo simulations, compare them with the experimental data and find that they are in excellent agreement. The distribution of the azimuthal angle of pair conversions shows a bias due to the non-cylindrical-symmetric structure of the detector. This bias would average out for a long duration exposure on a space mission, but for this pencil-beam characterisation we have ensured its accurate simulation by a double systematics-control scheme, data taking with the detector rotated at several angles with respect to the beam polarisation direction and systematics control with a non-polarised beam.

We measure, for the first time, the polarisation asymmetry of a linearly polarised gamma-ray beam in the low energy pair-creation regime. This sub-GeV energy range is critical for cosmic sources as their spectra are power laws which fall quickly as a function of increasing energy.

This work could pave the way to extending polarised gamma-ray astronomy beyond the MeV energy regime.

1 Introduction

A number of groups are developing pair-conversion detector technologies alternative to the tungsten-converter / thin-sensitive-layer stacks of the COS-B / EGRET / AGILE / Fermi-LAT series, to improve the single-photon angular resolution. Presently, observers are almost blind in the 1-100  M eV  energy range, mainly due to the degradation of the angular resolution of e+e- pair telescopes at low energies: to a large extent, the sensitivity-gap problem is an angular-resolution issue [1].

We have shown [2] that gaseous detectors, such as TPCs (time projection chambers), can enable an improvement of up to one order of magnitude in the single-photon angular resolution (0.5 °at 100  M eV ) with respect to the Fermi-LAT (5 °at 100  M eV ), a factor of three better than what can be expected for silicon detectors (1.0-1.5 °@ 100  M eV ). With such a good angular resolution, and despite a lower sensitive mass, a TPC can close the sensitivity gap at the level of $10^{-6}$  M eV / c m ^ 2 s ) between 3 and 300  M eV . In addition, the single-track angular resolution is so good that the linear polarisation fraction can be measured.

We first give a brief overview of the experimental configuration, as well as the simulation of the detector and the event reconstruction. We then describe the analysis procedure, and in particular the event selection. Finally, we show the measured performance of the detector, in terms of angular resolution and polarimetry. The difficulties encountered and the potential for improvement are discussed.

2 $\gamma$-ray astronomy and polarimetry with a TPC

TPCs are simple and robust particle detectors widely used in high-energy physics [3]. A volume of matter is immersed into an electric field, so that the ionisation electrons produced by the passage of high-energy charged particles drift and are collected on an anode plane. The anode is segmented so as to provide a 2D image of the electrons raining on it as a function of drift time. The measurement of the drift time provides the third coordinate. In our case, the field is uniform, so that the electron trajectories are straight lines and the electron drift velocity is constant and uniform. Noble gases (mainly helium to xenon) are very convenient as they allow free electrons to drift freely over long distances.

In our case the TPC is used as an active target, that is at the same time the converter in which the $\gamma$-ray converts and the tracker in which the two lepton trajectories are measured, a situation that induces conflicting constraints: for a given volume one would want to increase the matter density and the noble gas atomic number $Z$, so as to increase the telescope effective area (Fig. 7 of Ref. [2]) but in so doing, the single-track angular resolution and therefore the single-photon angular resolution would degrade (Fig. 6 of Ref. [2]). For a liquid xenon TPC, for example, the angular resolution would show no improvement with respect to that of the Fermi-LAT. When the effective area and the angular resolution are combined, the point-like source sensitivity turns out to be barely affected by the gas choice (Fig. 8 right of Ref. [2], estimated for a given gas mass of $10\,\mathrm{k}\mathrm{g}$).

When one considers in addition polarimetry, the measurement of the linear polarisation fraction $P$ and angle $\phi_{0}$ of the incoming radiation, the issue is the measurement of the azimuthal angle of the final state leptons before multiple scattering in the tracker blurs that information: the use of a liquid or solid TPC would need the tracking of sub-millimeter-long track segments, which is out of reach: we are bound to use a gas TPC (section 5 of Ref. [4]). Here again we see the same conflicting effects at work. Increasing $Z$ and / or the pressure increases the $\gamma$-ray statistics but also degrades the dilution factor due to multiple scattering (Fig. 26 of Ref. [4]). For a $1\,\mathrm{m}^{3}$ $5\,\bbar$ argon detector exposed for one year, full time with a perfect efficiency, the precision of the measurement of $P$ for a bright source such as the Crab pulsar is expected to be of $\approx 1.4\%$ (including experimental cuts), that is a $5\,\sigma$ MDP (minimum detectable polarisation) of $\approx 7\%$.

A multi-atomic component (means here $n>2$) added to the gas absorbs the U.V. photons created in the amplification process before they can reach the cathode and induce deleterious or even lethal sustained discharges, hence the name “quencher”. Also electrons accelerated during their drift lose energy by inelastic collisions with the quencher molecules, they cool down, which mitigates the diffusion strongly. A large choice of possible quenchers is available, including the alcanes [3].

In our case of a $2.1\,\bbar$ argon-based gas mixture, the active-target radiation length $X_{0}$ is of about $56\,\mathrm{m}$, so the probability of the pair conversion of a given photon crossing the detector is low. The effective area is proportional to the gas mass, while for a thick detector it is proportional to the geometrical surface. The probability of photon “loss” by Compton scattering is small too and the various possible processes are not competing with each other: the pair/Compton cross section ratio is irrelevant here, in contrast to thick detectors.

Astronomers also need to measure the energy of the collected photons, that is, here, the momentum of each track. This can be performed by a number of techniques including calorimetry, magnetic spectrometry, transition radiation detection, all of which would present a challenge to the mass budget on a space mission for the (multi)-cubic-meter detector that would make the desired effective area possible. An other method uses the track-momentum dependence of multiple scattering to obtain a measurement of the momentum from an analysis of the angle deflections in the TPC itself [5]. A pending question was the optimisation of the longitudinal segmentation pitch over which these deflections are computed (section 6 of Ref. [2]). An optimal treatment has been obtained recently, by a Bayesian analysis of the filtering innovations of a series of Kalman filters applied to the track [6]: meaningful results can be obtained with a gas TPC below $100\,\mathrm{M}\mathrm{eV}/c$ (Fig. 11 of Ref. [6]).

3 Experimental setup

The HARPO (Hermetic ARgon POlarimeter) detector [7] is a demonstrator of the performance of a TPC for measuring polarised $\gamma$ rays. It was designed for a validation on the ground in a photon beam. The most critical constraints related to space operation were taken into account, such as the reduced number of electronic channels and long-term gas-quality preservation [8]. It comprises a (30  c m )3 cubic TPC, designed to use a noble gas mixture from 1 to 4 \bbar. The present work uses an Ar:isobutane (95:5) gas mixture at 2.1 \bbar. A drift cage provides a 220  V / c m  drift field. The electrons produced by the ionisation of the gas drift along the electric field toward the readout plane at a constant velocity $v_{\rm drift}\approx 3.3\,\mathrm{c}\mathrm{m}\mathrm{/}\mathrm{\SIUnitSymbolMicro}\mathrm{s}$. The readout plane is equipped with two Gas Electron Multipliers (GEMs) [9] and one Micromesh Gas Structure (Micromegas) [10] to multiply the electrons. The amplified electrons’ signal is collected by two sets of perpendicular strips at a pitch of $1\,\mathrm{m}\mathrm{m}$ (regular strips in the $X$-direction, and pads connected together by an underlying strip in the $Y$-direction, see Fig. 4 of Ref. [11]). The signals are read out and digitised with a set of AFTER chips [12] and the associated Front End Cards (FECs).

Even though a cubic structure cannot be expected to behave as a fully isotropic detector, efforts have been made to have the longitudinal properties of the TPC (along $Z$) similar to the transverse ones ($X,Y$). The time sampling was set to $30\,\mathrm{n}\mathrm{s}$ so that, given the electron drift velocity, the TPC longitudinal sampling was close to the transverse (strip) pitch. The main residual difference are the (longitudinal) shaping of the electronics, of $100\,\mathrm{n}\mathrm{s}$ RMS, and an un-anticipated saturation of the electronics preamplifier that affected one channel (strip) independently from the others.

The transverse diffusion coefficient for that gas at that pressure was of about $380\,\mathrm{\SIUnitSymbolMicro}\mathrm{m}/\sqrt{\mathrm{c}\mathrm{m}}$, which makes the pitch size not far from optimal over most of the drift length range (see Fig. 7 of Ref. [13]); the longitudinal coefficient was of about $220\,\mathrm{\SIUnitSymbolMicro}\mathrm{m}/\sqrt{\mathrm{c}\mathrm{m}}$.

The HARPO TPC was set up in the NewSUBARU polarised photon beam line [14] in November 2014. The photon beam is produced by Laser Compton Scattering (LCS) of an optical laser on a high energy (0.6-1.5  G eV ) electron beam. A lay-out of the experiment can be found in Fig. 2 right of Ref. [15]. Using lasers of various wavelengths and different electron beam energies [16], 13 photon energies from 1.74  M eV  to 74.3  M eV  were obtained. A graphical representation of the $\gamma$-ray energies for which we did take data, as a function of laser wavelength and electron beam energy, can be found in Fig. 5 left of Ref. [15].

The Compton edge of the laser inverse Compton scattering, that is, the highest part of the $\gamma$-ray energy spectrum, was selected by collimation on axis, a lead brick with a $4\,\mathrm{m}\mathrm{m}$-diameter hole located $24\,\mathrm{m}$ downstream of the laser-electron interaction point, defining a $83\,\mathrm{\SIUnitSymbolMicro}\mathrm{rad}$ half-aperture divergence beam. After collimation, the polarization of the laser beam is almost entirely transferred to the $\gamma$-ray beam (Eq. (39) and Fig. 7 of Ref. [17]). For the collisions of a $0.9824\,\mathrm{G}\mathrm{eV}$ electron beam with a $1.54\,\mathrm{\SIUnitSymbolMicro}\mathrm{m}$ Erbium laser beam, for example, the $\gamma$-ray beam energy and the polarisation transfer varied from $11.8\,\mathrm{M}\mathrm{eV}$ and unity on axis to $11.4\,\mathrm{M}\mathrm{eV}$ and 0.999 close to the collimator jaw, respectively.

In order to mitigate systematic effects due to the geometry of the detector, the detector was rotated around the beam axis to 4 different angular positions (-45, 0, 45 and 90  °). Finally, for some configurations (in particular at low energy), data were also taken with randomly polarised photons as a reference.

A trigger system specific to the beam configuration was built using the signals from scintillators, from the micromegas and from the laser [18]. A study of basic event characteristics for various trigger configuration showed that photon signals were recorded with about 50% efficiency, while about 99% of the background was rejected [19]. For that data set [19], we were logging data at about $65\,\mathrm{Hz}$ ($52\,\mathrm{Hz}$ of signal, $13\,\mathrm{Hz}$ of background), that is with a deadtime of 10%, given the digitisation time of $1.67\,\mathrm{m}\mathrm{s}$, while the incident background rate was of $\approx 1.8\,\mathrm{k}\mathrm{Hz}$.

A specific event reconstruction procedure for pair conversion events was developed [20]. The reconstruction is focused on the local properties of vertices. It does not include any tracking, and does not give any global event information. We show in the following how this limitation is bypassed using the characteristics of the beam configuration.

4 Data selection

Data were taken with 13 values of photon energy from 1.74  M eV  to 74.3  M eV , but not all of them are usable for the analysis. There are two main difficulties:

• •

Low energy points (below 4  M eV ) were obtained using a CO2 laser without pulsing, and with a high intensity beam. This created a lot of “pile-up” events, where several interactions occurred in the detector within the 15  µ s  readout window. Besides, these low energy data are largely dominated by Compton scattering. Data below 4  M eV  were therefore discarded.

• •

The pre-amplifier in the AFTER electronics was found to saturate when a large charge accumulates on a single channel over a few dozen µ s . This induced a loss of signal when tracks were aligned with the drift direction $z$ of the TPC. This signal loss cannot be corrected, and creates a systematic bias in the data. A simple model was used in the simulation which is sufficient to accurately reproduce the systematic bias in the data up to 20  M eV . We do not show polarimetry measurements above this energy.

5 Event selection

The reconstruction [20] relies on local event geometry in 2 dimensions, without tracking. The reconstructed 2D vertices can have one or two associated pseudo-tracks, to accommodate the possibility of overlapping tracks. If both projections of the same vertex have two associated pseudo-tracks, there is a two-fold ambiguity. This ambiguity is resolved by using the ionisation fluctuation along the particle trajectories. The two 2D projections are then combined to obtain a 3D picture.

A reconstructed vertex carries two pieces of information:

• •

the vertex position $\vec{x}_{v}$;

• •

the direction $\vec{u}_{\pm}$ of the associated particles, which are described arbitrarily as electron (“-”) or positron (“+”). The sign of the charge is not necessary for either gamma-ray astronomy or gamma-ray polarimetry.

The opening angle $\theta_{+-}$ is defined as:

[TABLE]

A vertex can have either one or two associated particles. A vertex with only one associated particle is given an opening angle $\theta_{+-}=0$.

There are several reconstructed vertices for each event (in particular entry and exit points are considered as “single particle vertices”). Further topological information is needed to separate real vertices from background. Using the specific configuration of the photon beam, vertices are rejected that are away from the beam or close to the walls of the TPC. Figure 1 shows the space distribution of the reconstructed vertices in the TPC. The beam region is visible as a central zone with a higher density. The vertices around are rejected.

Vertices in the beam region have a high probability to have originated from an interaction of a gamma ray with the gas. These interactions are Compton scattering, pair production in the field of a nucleus (“pair”) or in the field of an electron (“triplet”). They are distinguished by the opening angle $\theta_{+-}$. Figure 2 shows examples of recorded events of each type.

For Compton scattering (single track), events can be misreconstructed as pair production (two track) events with small $\theta_{+-}$. Figure 3 shows the distribution of $\theta_{+-}$ for 4.68  M eV  and 11.8  M eV  beam energies, and compares them to simulations, decomposed into Compton scattering, pair and triplet events. The angular distribution is well described by the simulation. The proportion of Compton and pair events is however not correctly reproduced by the simulation at low energy, suggesting that the trigger efficiency is different for these two categories of events.

An energy-dependent cut on the opening angle $\theta_{+-}$ is used to select the pair events:

[TABLE]

Figure 4 shows the effect of the cut on the opening angle $\theta_{+-}$ on the different simulation samples (Compton, pair and triplet), and for the real data. There is a good agreement between data and simulation. The contamination of the pair production events by Compton scattering events is only of a few percent. Above 40  M eV , the opening angle gets smaller, and it is difficult to distinguish pair events from Compton scattering events.

The selection gives us high purity (over 90%) pair event samples above 4  M eV . At low energy, the stronger cuts applied to remove Compton events reduce the reconstruction efficiency. At high energy, the efficiency is affected by the low opening angles of the conversion events, which cannot be recognised as pairs. In spite of these difficulties, simulations show that the vertices are reconstructed with an efficiency higher than 90% over the whole spectrum presented here.

6 Simulation of the HARPO detector

The cubic geometry of the detector and the configuration of the readout scheme introduce a systematic bias to the polarisation measurement which cannot be addressed analytically. It is therefore necessary to have an accurate simulation of the TPC. We developed a complete simulation to describe the response of the HARPO detector. It contains three main components:

• •

An event generator describes the conversion of photons in the gas [4, 23]. It provides the energy momentum of the electron-positron pair.

• •

The interaction of the electron and positron pair with the gas is simulated using Geant4 [21]. It provides the ionisation electrons in the gas volume.

• •

The processes and the geometry of the TPC are described with a custom software [20] that provides a signal map similar to that of the real data.

The first two components have been validated in [22] and [23] respectively. The last one was developed specifically for HARPO. The description of the TPC includes electron drift, diffusion and amplification in the gas, the readout space and time response, and the signal digitisation, including known electronics saturation effects.

This simulation was thoroughly validated using a tight selection of cosmic rays. All of the simulation parameters were calibrated against data [24]. Figure 5 shows an example of the comparison between cosmic-ray data and the simulation of the raw-charge read out for each channel and time bin. This distribution is affected by most of the effects mentioned above. There is an excellent agreement between data and simulation.

7 Angular resolution

From a reconstructed vertex, the corresponding photon direction is estimated as $\vec{u}_{\rm pair}=\vec{u}_{+}+\vec{u}_{-}$. The residual angle $\theta_{\rm pair}$ is then:

[TABLE]

where $\vec{u}_{\gamma}$ is the beam direction. After applying the vertex selection described in Sect. 5, the distribution of the residual angle $\theta_{\rm pair}$ is obtained for each configuration of the same energy, polarisation and TPC orientation. Such a distribution for 11.8  M eV  photons is shown in Fig. 6.

The angular resolution is dominated by three main effects:

• •

The momentum of the recoil nucleus is not measured. The corresponding contribution is denoted $\sigma_{\rm recoil}$.

• •

The magnitude of the momentum of the two particles is not measured. The corresponding contribution is denoted $\sigma_{p}$.

• •

The detector has a finite angular resolution for single charged particles. The corresponding contribution is denoted $\sigma_{\rm det,\gamma}$.

Figure 6 shows the simulated distributions after neglecting each of these effects. The full simulation (“Sim”) gives the resolution $\sigma_{\rm recoil}\oplus\sigma_{p}\oplus\sigma_{\rm det,\gamma}$. Using the true track directions, the detector resolution is neglected (“MC truth”), and the resolution is $\sigma_{\rm recoil}\oplus\sigma_{p}$. Using the full 4-vector information, the physical limit is reached (“optimal”), and the resolution is $\sigma_{\rm recoil}$.

The angular resolution $\sigma_{\theta,68\%}$ is defined as the 68 % containment angle (i.e. the angle such that 68 % of the events have a smaller residual angle). Figure 7 shows the variation of the resolution with energy. The beam data and the detector simulation are consistent. The measured resolution is better by at least a factor of two than what is obtained by the Fermi-LAT, even with a tracking-less reconstruction. A large contribution to the resolution comes from the lack of momentum information.

The three main components of the resolution ($\sigma_{\rm recoil}$, $\sigma_{p}$ and $\sigma_{\rm det,\gamma}$) are extracted from the simulation. The results are shown in Fig. 8. Below 10  M eV , the main contribution comes from the nucleus recoil. The two other effects give a similar contribution. The beam and detector geometry introduces a systematic bias at low energy (below 5  M eV ), so that the final resolution is not the quadratic sum of the components. The two tracks in a pair are correlated, so that the actual resolution for single tracks is not relevant in this context. An effective angular resolution for single tracks can be defined as $\sigma_{\rm det,e^{\pm}}=\sigma_{\rm det,\gamma}/\sqrt{2}$. It can be approximated (see Fig. 8) as:

[TABLE]

8 Polarimetry

Following [25], an optimal estimate of the polarisation asymmetry is given by the distribution of $\phi_{+-}=(\phi_{+}+\phi_{-})/2$, where $\phi_{\pm}$ is the azimuthal angle of the particle ($+$, positron; $-$, electron) with regard to the beam axis:

[TABLE]

where $X,Y$ are the coordinates in a plane perpendicular to the beam direction, and $X$ is the direction of the beam polarisation.

The distribution of $\phi_{+-}$ is obtained for each configuration of same energy, polarisation and TPC orientation. Figure 9 shows an example of the distributions for 11.8  M eV  photons in each of the configurations. There are large systematic effects due to the cubic geometry of the detector and the fixed direction of the photons.

The geometry effects are cancelled out by taking the ratio of the polarised and unpolarised beam data. Figure 10 shows the result for the four different orientations of the detector around the beam axis. The systematic bias is further reduced by combining the data with different orientations, resulting in Fig 11. Since the unpolarised data are not available for every configuration, the simulation is used to correct the systematic bias. Figure 11, bottom, shows the ratio of real polarised beam data with a simulated unpolarised beam. In each case, the distribution is fitted with the expected function $1+A\cos{2(\phi_{+-}-\phi_{0})}$, where $A$ is the measured polarisation asymmetry.

The above results are strongly influenced by the fixed configuration of the photon beam. In the case of a space telescope, the systematic bias would be very different. Figure 12 shows the azimuthal angle distribution for simulated isotropic 11.8  M eV  photons, converting uniformly inside the detector. This represents a simple model for a long duration exposure in a space mission. The reconstructed azimuthal angle is uniform for an unpolarised source, and shows the expected modulation when the photons are polarised. In that case, no correction is applied. The measured amplitude $A$ of the polarisation asymmetry in this case cannot be directly compared with what is measured in beam data. The amplitude $A$ depends on the fiducial cuts chosen, because the angular resolution depends on the length of the particle trajectories inside the detector. An optimisation of these cuts is necessary to assess the polarimetry potential in such a configuration.

Figure 13 shows the measured polarisation asymmetry $A$ obtained in each of the following cases:

• •

ratio of polarised data over unpolarised data;

• •

ratio of polarised simulation over unpolarised simulation;

• •

ratio of polarised data over unpolarised simulation;

• •

ratio of polarised simulation over unpolarised data (as a validation of the method).

The optimal value of the polarisation asymmetry $A$ from QED is calculated using an exact event generator [4, 25]. In addition to the full simulation represented in Fig. 11, we estimate the contribution of the single-track angular resolution, alone, to the dilution of the polarisation asymmetry. The finite resolution on the azimuthal angle $\sigma_{\phi}$ dilutes this asymmetry by a factor $D=e^{-2\sigma^{2}_{\phi}}$. Approximating the opening angle of the pair by its most probable value $\hat{\theta}_{+-}\approx E_{0}/E$ (with $E_{0}=$1.6  M eV ) [26] gives:

[TABLE]

where $\sigma_{\rm det,e^{\pm}}$ is the effective angular resolution of the detector (Eq. 4). This gives an expected value for the measured polarisation asymmetry, which is shown as the green dashed line and the open stars in Fig. 13. The measured polarisation asymmetry is found to be consistent with the QED limit, taking into account the detector’s angular resolution.

9 Discussion

The data shown here were obtained in a photon beam with high flux, which is not representative of the situation for a telescope for cosmic gamma rays. This high event rate (up to 15  k Hz , localised on the beam region of a few m m ^ 2 ) created extra difficulties related, in particular, to the saturation of the readout electronics. In a space telescope, background tracks will be uniformly distributed in the gas volume, at a rate of about 20  k Hz / m ^ 2 . The event reconstruction could not be completed and optimised due to the limited resources of the project. The results shown here should therefore be seen as a lower limit to the capabilities of the detector.

The detector performance can be further improved in several ways:

• •

reduction of the electronics saturation, by using a more adapted dynamic range;

• •

improvement of the vertexing algorithm, using more adapted peak finding method;

• •

introduction of a tracking algorithm, for event categorisation and possible resolution improvement;

• •

introduction of a track-momentum estimation using multiple scattering[6].

All of these methods should improve the resolution of the detector, without requiring any hardware modifications.

The introduction of extra detectors to get acurate momentum measurement should of course be considered. Conventional solutions such as a calorimeter or a magnet are however made very difficult by the weight constraints of a space telescope.

10 Conclusion

We have built a TPC for gamma-ray detection and polarimetry and we have successfully operated it in a polarised photon beam between 1.74  M eV  and 74  M eV . We have developed a simple but effective event reconstruction for pair production, and a detailed and accurate detector simulation. Measurements of the angular resolution are consistent with simulations and offer an improvement of at least a factor of two over the Fermi-LAT. This resolution can be further improved with simple modifications of the hardware and software. We show the first measurements of the polarisation of pair-production photons below 50  M eV .

The behaviour of the detector is very well reproduced by the simulation. We can now confidently use the simulation to investigate improvements, and to estimate the performance of such a detector for cosmic gamma-ray measurement in the high atmosphere or in space.

This work was funded by the French National Research Agency (ANR-13-BS05-0002) and was performed by using NewSUBARU-GACKO (Gamma Collaboration Hutch of Konan University).