Datasets:

SouthernCrossAI
/

Arxiv

Dataset card Data Studio Files Files and versions Community

Split (1)

train · 132k rows

Filename stringlengths 22 64	Paragraph stringlengths 8 5.57k
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	Four of our sample, GRBs 051022, 060814, 050223 and 100621A, have been identiﬁed as dark bursts in earlier work. These are deﬁned as sources which are sub-luminous in the optical relative to the X-ray, typically with an optical-to-X- ray spectral slope βOX < 0.5. This spectral index is beyond those normally allowed in GRB ﬁreball models, and is most commonly attributed to the presence of dust in the GRB host galaxy. Indeed, the majority of these dark bursts have been shown to be associated with high optical extinctions (Perley et al. 2013). Studies in the optical and near-infared have suggested that the hosts of such bursts are typically more massive and chemically evolved than those of other GRBs, with higher star formation rates (e.g. Perley et al. 2013; Kr¨uhler et al. 2011). However, as already discussed, dust obscuration can make intepretation of galaxy spectral energy distributions challenging. Given the diﬃculties asso- ciated with faint optical emission, radio continuum obser- vations have the potential to improve our understanding of the host galaxies of these sources (Zauderer et al. 2013).
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	Two of our candidates, GRBs 051022 and 060814 are included in the dark burst sample of Perley et al. (2013). Another, GRB 050223, does not satisfy the strict criteria for βOX, with the strongest available constraint of βOX < 0.8, but based on its host galaxy properties Pellizza et al. (2006) claim this source as part of the dark sample (although Page et al. 2005, ﬁnd no evidence for excess extinction in the X-ray emission from the burst). A high extinction (AV =3.8) and low βOX = 0.278 has also been reported for a fourth source, GRB 100621A (Kr¨uhler et al. 2011; Melandri et al. 2012).
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	Radio constraints on the host galaxies of an exclusively ‘dark’ sample of 15 hosts, observed from the VLA, have re- cently been presented by Perley & Perley (2013). Only one source, GRB 051022, overlaps with the sample presented here, with the majority of targets in the Perley & Per- ley study lying at higher redshift than those in our sam- ple. Four of their targets were detected: 3 host galaxies at SFR> 800 M⊙ yr−1 and GRB 051022 at 74 ± 20 M⊙ yr−1, consistent with our limit given in table 1. Two further host galaxies (GRBs 060202 and 090417B) were constrained to have radio star formation rates under 120 M⊙ yr−1, while the remaining, higher redshift targets, have less constrain- ing upper limits on their star formation rates spanning from < 325 to a relatively weak < 4140 M⊙ yr−1.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	We detect the hosts of dark GRBs 050223 and 100621A in our 5.5 GHz observations, giving radio-derived star forma- tion rates of ∼ 90 − 100 M⊙ yr−1 – signiﬁcantly higher than the majority of GRB hosts. Given the likely high redshift of GRB 060814 (see section 2.1), the star formation rate con- straint on this source is rather weak and we exclude it from further discussion.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	be harbouring star formation rates of several hundred so- lar masses per year, and yet remain undetected. As with GRBs 060814, we exclude these from further analysis.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	To evaluate the probability of this detection rate occur- ing by chance, assuming that dark burst hosts are drawn from the same underlying star formation rate distribution as the non-dark burst sample, we consider a bootstrap re- sampling of the full existing, non-dark burst, dataset. The observations presented here and in Micha lowski et al. (2012) provide detections or limits of < 120 M⊙ yr−1 or better on 28 individual non-dark sources. We make the conservative assumption that we have barely missed detecting the ma- jority of sources (i.e. that their 2 σ limits are representative of their true star formation rates), and randomly draw sub- samples of nine objects from this compilation to mimic the dark burst sample, repeating the process 107 times. We ﬁnd that in only 1.7% of cases do six or more randomly sampled objects in a sample of nine have star formation rates exceed- ing 60 M⊙ yr−1. As such, it is highly unlikely that the dark busts we have identiﬁed above, are drawn from same under- lying population as the non-dark bursts. The deviation of dark burst host star formation rates from those of the non- dark GRBs, is signiﬁcant at the 2 σ level (subject, of course, to the vagaries of low number statistics) and merits further investigation.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	This may suggest that while, as Perley & Perley (2013) and others have concluded, dark GRB hosts are typically forming stars at a lower rate than submillimeter galaxies, they are nonetheless biased towards systems with higher star formation rates than seen in the general, less extinguished long GRB host population.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	z ∼ 0.86. These detections correspond to star formation rates that are substantially lower (by an order of magni- tude) than current GRB host limits. They also show some evidence for a trend towards higher star formation rates at higher redshift.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	This paper is based in part on data obtained at the Aus- tralia Telescope Compact Array associated with programme C2544. The Australia Telescope Compact Array is part of the Australia Telescope National Facility which is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	Also based in part on observations taken at the NRAO Karl G. Jansky Very Large Array (VLA), associated with programme 12A-279. The National Radio Astronomy Ob- servatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Uni- versities, Inc.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	This work made use of data supplied by the UK Swift Science Data Centre at the University of Leicester (Evans et al. 2009). We also made use of Ned Wright’s very useful cosmology calculator (Wright 2006).
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	AJL and ERS acknowledge support from the UK Sci- ence and Technology Facilities Council, under the Warwick Astrophysics consolidated grant ST/L000733/1 and PATT- linked travel support grant. We also acknowledge the in- ﬂuence of suggestions from the anonymous referee on the evolution of this paper.
Processed_Radio_Observations_of_GRB_Host_Galaxies.txt	As discussed in section 5.2, we extract radio ﬂuxes at the locations of core-collapse supernovae originally iden- tiﬁed as part of the GOODS campaign (Giavalisco et al. 2004). Sources identiﬁed during the 2002-2003 observ- ing campaign were catalogued, classiﬁed and reported by Strolger et al. (2004), listed both by ‘nickname’ and IAU approved transient name. Sources from 2004-2005 were clas- siﬁed by and discussed in Dahlen, Strolger, & Riess (2008) and Dahlen et al. (2012). Formal IAU naming was not sought for these later objects. Spectroscopic redshifts were obtained either from literature values for the supernova host galaxy or through Keck Observatory spectroscopy pri- marily originating from or compiled by the Team Keck Treasury Redshift Survey (Wirth et al. 2004). Target clas- siﬁcations, locations and redshifts for all these sources were made available electronically in 2008, as described in Dahlen, Strolger, & Riess (2008), and we make use of this publically released catalogue, and speciﬁcally those super- novae classiﬁed as core collapse and within the coverage re- gion of the available radio imaging.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Nulling pulsars, ﬁrst identiﬁed by Backer (1970b), sponta- neously and suddenly (typically within one rotation) cease to emit at a detectable level. After an interval ranging from a few rotations (e.g. Wang, Manchester & Johnston 2007) to a few years, e.g. PSR J1841 600 days, Camilo et al. 2012), emission resumes just as suddenly. This wide range of timescales, most of which are signiﬁcantly longer than those natural to a pulsar magnetosphere, is diﬃcult to explain through a single mechanism.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Because radio emission arises from coherent plasma processes, short nulls have been interpreted as ﬂuctuations in the plasma state that brieﬂy disrupt or alter the re- quired coherency. In other pulsars—particularly those with drifting subpulses—the nulling phenomenon may be re- lated to the geometry of a drifting “carousel” of sparks (Deshpande & Rankin 1999; Rankin & Wright 2008) whose drift rate depends on the local charge density above the po- lar cap. For example, Unwin et al. (1978) determined the subpulse drift rate of PSR B0809+74 drops to zero during nulls, while Herﬁndal & Rankin (2007) identiﬁed subpulse periodicity persisting through nulls in PSR B1133+16, sug- gesting these short nulls are simply gaps in the drifting beam pattern.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	On the other hand, the pulsars that null on long 30 days, Kramer et al. timescales, e.g. PSR B1931+24 ( 600 days, Camilo et al. 2012), 2006), PSR J1841 and PSR J1832+0029 ( 600–800 days, Lorimer et al. 2012) are intermittent in the sense that the currents and particle acceleration powering radio emission may cease, or at the very least change suﬃciently to steer the radio beam away from the earth (e.g. Timokhin 2010). The key piece of evi- dence comes from the measurement of the spindown rate ( ˙ν) of each state, which decreases by a factor of 1.5–2.5 when the pulsar is nulling, indicating a dramatic reconﬁguration of the magnetospheric currents and, consequently, torque.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	between metastable magneto- sphere/spindown states have been identiﬁed by Lyne et al. (2010), who discovered a quasiperiodic modulation of the spindown rate in a sample of pulsars monitored over several decades. Moreover, they found that in some pulsars, switches between ˙ν levels were accompanied by changes in pulse proﬁle (i.e. mode changes, Backer 1970a), thus linking emission properties to magnetosphere state. The stochastic switches between ˙ν states introduce red noise into timing residuals relative to a spindown model with a single ˙ν, suggesting that such state switching may be of fundamental importance to the timing noise phenomenon and thus to pulsar timing arrays.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	magnetospheres. However, because switches happen rarely and rapidly, it is diﬃcult to catch these pulsars “in the act” and gain insight into both what these states physically rep- resent and how the switching is accomplished. Pulsars with intermittency of hours likely also represent magnetospheric switching, though the intermittent durations are too short to measure independent values of ˙ν. Such pulsars can be targeted with observations to resolve all relevant timescales and may thus yield clues to the switching process. In a clear example of the power of such systems, Hermsen et al. (2013) discovered that the X-ray emission from PSR B0943+10 is correlated with radio mode changes, implying rapid cooling of the neutron star polar cap after the pulsar switches from its radio-faint to its radio-bright mode.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Right ascensiona , R.A. (J2000.0) . . . . . Declinationa , decl. (J2000) . . . . . . . . . . . Position epoch (MJD) . . . . . . . . . . . . . . . . Frequency, ν (Hz) . . . . . . . . . . . . . . . . . . . . . Frequency derivative, ˙ν (10−15 Hz/s) . . Frequency epoch (MJD) . . . . . . . . . . . . . . b c (Myr) . . . . . . . . Characteristic age, τc Spindown luminosity, ˙Eb d (erg s−1) . . . Dipole magnetic ﬁeld, Be (1012 G). . . . . DM Distance (NE2001, kpc) . . . . . . . . . . Flux density, 732 MHz (mJy) . . . . . . . . . Flux density, 1369 MHz (mJy) . . . . . . . . Flux density, 3094 MHz (mJy) . . . . . . . . b (mJy) Flux density, 1400 MHz, S1400 Spectral index, α . . . . . . . . . . . . . . . . . . . . . Dispersion measure, DM (pc cm−3) . . . Rotation measure, RM (rad m−2) . . . . . Scattering timescale, τsc,732 (ms) . . . . . . Scattering timescale, τsc,1369 (ms) . . . . . b (ms). . . . . Scattering timescale τsc,1000 Inactive fraction (%) . . . . . . . . . . . . . . . . . . Active nulling fraction (%) . . . . . . . . . . . Mean active state (AP) duration (s) . . Mean inactive state (IP) duration (s) .
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Note. — Numbers in parentheses give the uncertainty on the terminal signiﬁcant ﬁgure(s). The formal uncertainties on ﬂux densities and timescales are smaller than systematic uncertainties, which we estimate to be about 10%.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	data obtained are used for both educational and scientiﬁc purposes (Hobbs et al. 2009). During two-hour observing sessions, students work in small groups and select suitable pulsars from the program catalogue, observing each pulsar for a few minutes. As the project has been running since De- cember 2007, the data set provides an excellent set of snap- shot observations for a number of pulsars. Because its inter- 4054 mittent nature engages student interest, PSR J1717 is typically observed in any session it is visible, resulting in 85 observations as of April 2014.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Figure 1. Residuals of 20 cm pulse times of arrival in fractions of a neutron star rotation for the best-ﬁt timing model.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	We primarily processed the data using the PSRCHIVE software suite (Hotan, van Straten & Manchester 2004). We mitigated aliased signals and narrowband RFI by, respec- tively, excising channels within 5% of the band edge and those with a level substantially above a median-smoothed bandpass. To convert the observed intensity to absolute ﬂux density, we made use of observations of the radio galaxy 3C 218 (Hydra A; Baars et al. 1977) taken every few weeks to support the Parkes Pulsar Timing Array project (Manchester et al. 2013) to determine the system equivalent ﬂux density of the calibration noise source and consequently an absolute ﬂux scale for each Stokes parameter.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	All data recorded through the PULSE@Parkes pro- gram become public immediately and are available through the CSIRO Data Archive Portal1 (Hobbs et al. 2011). Data taken under other programs, including those described be- low, are generally available after an 18-month proprietary period.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	observa- data with Combining the Tempo2 package tions, we (Hobbs, Edwards & Manchester 2006) to obtain a phase- connected timing solution extending over six years. The residuals to a simple ﬁt for the frequency (ν) and its derivative ( ˙ν) show modest timing noise and appear in Figure 1. This is the ﬁrst published measurement of ˙ν, from which several important properties, listed in Table 1, can be estimated.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	of baseline and (2) ﬁtting an analytic template to the proﬁle and inferring the ﬂux from the template scale.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Table 2. State switches from active to inactive and vice versa observed, as well as total time recorded in each state, in PULSE@Parkes and archival data.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	PULSE@Parkes observations. Further, as there is a natu- ral bias among both high schools students and astronomers to extend (curtail) observations of the pulsar when it is ac- tive (inactive), a simple tally of the total time the pulsar is observed to be bright is biased.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	On the other hand, switches are random, and, if the process is memoryless (Poisson), switch rates are unaﬀected by observational bias. Moreover, since switches are rare for this pulsar, we expect to see no more than one state switch in a given sub-integration. Thus, to estimate switch rates from our fold-mode data, which has typical time resolution of 30 s, we classify by inspection each sub-integration into one of four classes: IP IP . We then simply count the instances of each class and tally the total time spent in each state in Table 2. Averaged over all PULSE@Parkes and archival observations, we ﬁnd a rate of IP → ± 10−4 Hz. That is, the typical time the pulsar spends in 1.6) the active (inactive) state is 1100 800 s), and 15%. Although limited by the the inactive fraction is 80 small number of observed switches, we note that these rates appear to be consistent from year to year. We also note that the nulling fraction obtained by simply tallying time in each state is 77%, lower than, but consistent with, the estimate from switch counting.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Figure 3. Pulse proﬁles at (top to bottom) 10 cm, 20 cm, and 50 cm. Total intensity (Stokes I) is the dominant black trace. A small amount of circular polarization (Stokes V) appears in dashed blue at 10 cm and 20 cm, as does a similar level of lin- ear polarization (solid red). The inset in the top panel shows a zoomed view of the proﬁle, in which appears a hint of a leading component.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Figure 4. Stokes I proﬁles at 20 cm (grey, sharp) and 50 cm (grey, long tailed) along with models of the scattered proﬁle (black) as described in the text. The scale is arbitrary.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	take search-mode data, no fold-mode data were obtained. We reduced all search-mode data to 1024-bin single pulse proﬁles with dspsr (van Straten & Bailes 2011).
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Concurrent with the 20 cm Parkes observations on Day 1, we also observed PSR J1717 4054 with the six-element ATCA interferometer for 5.5 hr. We recorded 512 channels of data over a total bandwidth of 2048 MHz centred at 2102 MHz using the Compact Array Broadband Backend in pulsar- binning mode (Wilson et al. 2011), which produces visibil- ities in 32 phase bins each 10 s correlator cycle. Primary, bandpass, and ﬂux calibration were achieved using the ra- dio galaxy PKS 1934 638, which has little time-dependent ﬂux variation. Secondary phase calibration and gain calibra- 397, tion were conducted with the radio galaxy PKS 1714 and we used self-calibration to correct for phase variations of the antennas during the observation.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	The observations covered the ﬁnal three APs of Day 1 (see Figure 5). To measure the pulsar ﬂux, we ﬁrst sub- tracted the oﬀ-pulse visibilties from the on-pulse visibilties, removing the non-pulsed ﬂux in the ﬁeld. We inverted these visibilities to form a dirty image and subsequently cleaned the image using a multi-frequency algorithm.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	We measured the ﬂux density across the entire band and also in interference-free sub-bands, and we ﬁt our broad- band observations to derive a spectral index of 0.08, consistent with those obtained with Parkes data. We esti- 0.1 mJy at 1369 MHz, mate the AP ﬂux density to be 4.7 about 10% lower than the Parkes measurement. Since the ATCA ﬂux measurements include nulls within the AP, while the Parkes measurements do not, we expect the former to be lower by about 10%. Further, the Parkes and ATCA ﬂux scales are derived from two diﬀerent sources (Hydra A and 4054 was about 500′′ from PKS 1934 the phase centre of the ATCA observations, making the ﬂux measurement particularly dependent on the primary beam model for the receiving system.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Using the ATCA images, we searched for but detected nei- ther extended nor pointlike emission in the oﬀ-pulse phase bins. The former could arise, e.g., from a pulsar wind nebula. These typically form around pulsars with large energy out- puts (strong winds) and/or those with high space velocities embedded in dense environments (Gaensler & Slane 2006). Although PSR J1717 4054 sits in the Galactic plane and its velocity is unknown, it has a relatively low spin-down luminosity, and our nondetection is unsurprising.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	where A is the material albedo, fg is a beaming function, Ma,−4 = Ma/10−4M⊕ is the mass of the asteroid belt, r10 = 1010r cm is the radius of the asteroid belt, and ˜Ra,2 is the root mean square (r.m.s.) radius of the asteroids. The beaming factor fg is the unknown ratio of ﬂuence beamed in the direction of the asteroid belt to the ratio beamed in the direction of Earth. By imaging the oﬀ-pulse bins both during APs and over the entire observation, we place an AP limit of 120 µJy beam−1 and 33 µJy beam−1 overall. The former limit is about 10% of the pulsed ﬂux and does not constrain the properties of any circumpulsar material. On the other hand, if the pulsar’s intermitency is caused by a modest shift of the pulsar beam away from the earth while the disk il- lumination remains unchanged, the latter limit applies and begins to probe the predictions of Eq. 1.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	To obtain high S/N pulse proﬁles, appearing in Figure 3, we co-added the APs throughout the long tracks, using the fold-mode data available at 20 cm and the folded search- mode data for 10 cm and 50 cm.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	At 10 cm, the proﬁle is a narrow, approximately gaus- sian pulse. Linear polarization is present at low levels, while the modest circular polarization appears to ﬂip handedness in the pulse center. The polarization position angle has a classic “S” swing, but the pulse is too narrow to constrain the magnetic inclination. There is some evidence for a lead- ing component of comparable width to the dominant, visi- ble peak. The 20 cm proﬁle is similar to the 10 cm proﬁle, save for the appearance of a scattering tail. By dividing the 20 cm and 10 cm bands into eight sub-bands and minimizing DM ν −2, we determined the dispersion the residuals to δt 0.1 pc cm−3. We note that this measure (DM) to be 306.9 is not necessarily the absolute DM, as the ﬁt is covariant with proﬁle evolution and an unknown time oﬀset between the two bands.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	we obtained a reasonable representation of the lower fre- quency data (Figure 4). We ﬁnd values of τsc,732 = 60 ms and τsc,1369 = 7 ms at 732 MHz and 1369 MHz. We estimate a systematic uncertainty due to approximating the intrinsic ν −β proﬁle with the 10 cm proﬁle of about 10%. If τsc describes the relation of scattering time and frequency, we ﬁnd β = 0.4, somewhat less than the canonical value 4. Extrapolating the 732 MHz measurement to 1 GHz, β = we obtain τsc = 21 ms. We note that the NE2001 model of the electron distribution in the Galaxy (Cordes & Lazio 2002) predicts a τsc of only 1.0 ms at 1 GHz, indicating sub- stantial scattering structure in excess of the model predic- tion along the line of sight.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	pulse shape, the mean proﬁle was generally an acceptable description of the single pulse emission. We estimated the baseline from the oﬀpulse mean, and the phase was known from the timing solution, leaving the signal strength, s, of the template in each pulse as the only free parameter. We adopted twice the log likelihood ratio for s = ˆs, the best-ﬁt value, versus s = 0 as our test statistic (TS), setting TS = 0 when ˆs <= 0. In the absence of a signal, TS = 0 half of the time, while the remaining TS > 0 values follow a χ2 distri- bution with one degree of freedom. Then, √TS = σ, i.e. the chance probability to observe TS > σ is the tail probability of a standard normal distribution integrated from Z = σ.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	We examined the 20 cm fold-mode data for the presence of pulsed emission while the pulsar was in the inactive state. After excising APs and sub-integrations contaminated by impulsive broadband RFI, we retained 6.6 hr which we reduced to a 128-bin Stokes I proﬁle with an r.m.s. of 14 µJy. The excellent agreement with the predicted 180 radiometer-noise limit of 169 µJy, indicates the absence of both a pulsed signal and any substantial RFI. Using the mean 20 cm proﬁle as a template, we obtain a 2σ upper limit on emission from the inactive state of 3.9 µJy, less than 0.1% of the mean AP ﬂux.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	We likewise searched the ATCA images for on-pulse emission during IPs, and we set an upper limit of 12 µJy, consistent with that above. To detect emission with a diﬀer- ent pulse proﬁle, we also formed the diﬀerence between the two halves of the pulse-phase bins, but we again found no evidence for pulsed emission.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	During the 20 cm observations of Day 1, the pulsar switched on for ﬁve discrete APs with durations given in Figure 6; see also Figure 5. Including only complete active and inactive intervals, the mean AP lasts 1347 rotations, or 1196 s, while the mean IP lasts 4645 rotations, or 4124 s, consistent with 2.4, albeit with much poorer statistics. our estimates in § Since the bounding nulls must be at least as long as the observed values, including these intervals increases the mean null duration to >4729 s.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	On Day 2 (10 cm and 50 cm observations), we observed only two full APs (durations of 450 s and 2220 s; see Fig- ure 5) and one partial AP at the onset of the observation (>870 s). The IPs within the observation were much longer (9960 s and 16,980 s) than those of Day 1. The IP/AP duty cycle observed during the long track observations disfavour a memoryless activation process, as there appear to be no short IPs.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	The distribution of null and non-null durations within the APs is shown in Figure 7. The distribution of non-null dura- tions is approximately exponential, indicating that the pul- sar does not “remember” how long it has been shining since the previous null. The slight excess of short “on” states rel- ative to the model may stem from incorrect classiﬁcation of a pulse as a null (though see below).
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	The distribution of nulls, on the other hand, is poorly ﬁt by an exponential, and appears to be bimodal. Short nulls of a few rotations (Type I as classiﬁed by Backer 1970b) signiﬁcantly outnumber the longer (Type II) nulls extending for tens of rotations. Since the TS distributions of null and non-null pulses are not perfectly separated, some of the Type I nulls may simply be faint pulses. To check this, we co-added the 52 single-pulse nulls and found TS= 3.58σ, indicating that a few of the nulls may be pulses, but that the majority are bona ﬁde nulls. A similar analysis of 16 two-pulse nulls yields TS= 0, while co-adding Type II nulls with durations of > 5 rotations yields TS= 3.1, indicating the long nulls may include a few faint pulses.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Figure 5. The phase-resolved intensity (arbtirary units) obtained during the two days of long-track observations, demonstrating the stochastic nature of the intermittency.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Figure 6. √TS for 20 cm single pulse proﬁles. “ON” (green) and “OFF” (salmon) are classiﬁed according to the algorithm described in the text. Values with √TS < 0.1 are set to 0.1 σ for display purposes and appear at the bottom of the ﬁgure. Pulses aﬀected by the presence of the calibrator are coloured in blue, while those ﬂagged as suﬀering from impulsive RFI are drawn in red. The solid black line is drawn at 3σ, the dashed lines at 2σ and 4σ. In rotations, the ﬁve active states measure 2363, 1436, 1352, 263, and 1320, the intervening inactive states 2172, 7019, 2285, and 7105, and the bookending inactive states >6320 and >7059.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	bution (r.m.s. 2.05 mJy), the observed distribution is a good proxy for the intrinsic distribution of single pulse ﬂuxes.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Figure 8. Distribution of observed single pulse ﬂux densities; those identiﬁed as nulls are shown in grey, while those identiﬁed as non-nulls are drawn in black. The mean ﬂux, 5.2 mJy, is indicated by the dashed vertical line.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	single pulse ﬂuxes. However, the PSD of each of the APs was consistent with white noise, and we conclude there is no mea- surable correlation between pulses. Likewise, we looked for any correlation between null duration and the ﬂux of the pre- ceeding/following pulses, e.g. as in that of PSR B1944+17 (e.g. Deich et al. 1986). There is no evidence of correlation, save perhaps that the ﬁrst pulse of an AP tends to be weaker than average.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	In summary, we ﬁnd PSR J1717 4054 is well described as a four-state system: an “active” state (0) with nulls at two discrete timescales (1 and 2) and no evidence of complex pulse substructure or memory across nulls; and an “inac- tive state” (3) whose emission, if present, must be at least 1,000 times fainter than that of state 0. It remains unclear if any of the null states (1, 2, and 3) correspond to the same magnetospheric conﬁguration. The stability of the pulsar’s ﬂux density and proﬁle over timescales of years, as well as the absence of pecularities in its timing solution, suggest the pulsar is stable when averaged over many state transitions, i.e. more than a few days.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	Although we cannot directly measure the diﬀerence in spindown rate between APs and IPs, we can estimate the contribution of such switching to the r.m.s. of the TOA residuals to our timing solution (Figure 1). Following Equa- tion 12 of Cordes (2013) and using the measured state dura- tions of 1100 s (active) and 4300 s (inactive) and the corre- sponding state probabilities 0.2 and 0.8, the estimated con- tribution σ 16 ms, 9, somewhat larger than values mea- this implies δ ˙ν/ ˙ν sured from intermittent pulsars (1.5–2.5), implying other sources of timing noise are also important, though Timokhin (2010) suggests that relative spindown rates may depend sensitively on small changes in magnetosphere geometry.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	tern. This may be the case for PSR J1717 4054, though we have no additional evidence through subpulse structure. Fi- nally, it is also worth pointing out PSR J1502–5653, a scaled 4054; its ˙E and τc agree to Doppelg¨anger of PSR J1717 − within a factor of two, and its NF is similarly high, 93% (Wang, Manchester & Johnston 2007). It, too, displays a pattern of IP/nulling AP, save with IP and AP durations scaled down by a factor of 10, making it a tempting target for long-track study.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	In conclusion, we now have a detailed picture of PSR J1717 4054 on all timescales of interest and have also measured a panoply of important properties. Substantial advance—e.g., identifying pulse substructure—must likely await the the large collecting area of the SKA, though some interim progress might be made with, e.g., ultrawide band- width feeds, or with additional long-track observations. The identiﬁcation of nulling at three timescales (few pulses, tens of pulses, and thousands of pulses) is challenging to inter- pret in any single picture, and we hope these observations will stimulate the introduction of new physical models for state switching.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	We thank the anonymous referee and the journal editors, whose eﬀort and input improved this paper.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	The Parkes radio telescope and Australia Telescope Compact Array are part of the Australia Telescope, which is funded by the Commonwealth Government for operation as a National Facility managed by CSIRO.
Processed_The_Three_Discrete_Nulling_Timescales_of_PSR_J1717.txt	PULSE@Parkes is funded by CSIRO Astronomy and Space Science. We are grateful to the Australia-Japan Foundation, the Victorian Space Science Education Centre, SPICE at the University of Western Australia, the Astro- physics groups at the University of Cardiﬀ and University of Oxford, ASTRON, NAOJ Mizusawa VLBI Observatory, Koriyama Space Park, Penrith Anglican College, and the University of Brownsville, Texas for support in funding and hosting sessions. The authors would like to acknowledge the more than 1,100 students from 93 schools who have taken part in sessions and contributed to the data gathering since the program’s inception in December 2007, as well as the CSIRO Astronomy and Space Science staﬀ, visitors, and co- supervised PhD students whose volunteer participation has enriched the program.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Recently companies started to embrace Machine Learning (ML) techniques for introducing a series of advanced functionality in their products such as personal- isation of the user experience, improved search, content recommendation and automation. The technical challenges for tackling these problems are heavily researched in literature. A less studied area is a pragmatic approach to the role of humans in a complex modern industrial environment where ML based systems are devel- oped. Key stakeholders affect the system from inception and up to operation and maintenance. Product managers want to embed “smart” expe- riences for their users and drive the decisions on what should be built next; software engineers are challenged to build or utilise ML software tools that require skills that are well outside of their comfort zone; legal and risk departments may inﬂuence design choices and data access; operations teams are requested to maintain ML systems which are non-stationary in their nature and change behaviour over time; and ﬁnally ML practitioners should communicate with all these stakeholders to successfully build a reliable system. This paper discusses some of the challenges we faced in Atlassian as we started investing more in the ML space.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	offers software companies some new beneﬁts such as the access to detailed behavioural data of their users. Storage and computation costs have dropped signiﬁcantly allowing the analysis of these vast data sources and leading to data- driven decision making processes that further improves the products.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Recently, software companies started to experience the next big paradigm shift in their operation. The bar for utilising ML techniques has been lowered signiﬁcantly. Even advanced modelling techniques have started to be- come a commodity. As discussed companies already hold a plethora of data about their users which can be used as training data for ML algorithms. Furthermore, because products live in the cloud it is feasible to constantly im- prove their products on the ﬂy and thus increasing the ap- petite for experimentation. However, we are also experi- encing a lot of hype driven mainly by the media, consul- tants and vendors. This becomes more challenging given that companies typically lack the scarce relevant talent re- quired to drive them successfully in this new era.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	In a more pragmatic approach Data Scientists, deﬁned here as ML practitioners, unfortunately are a scarce resource ab- sorbed by the few big orgs that pioneer the ﬁeld. Thus non- experts are getting involved in the development of the ML system and this creates a less compelling environment in most cases in the rest of the industry. It is a learning pro- cess for most parties involved in the process.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Table I summarises the key stakeholders and the challenges they face for the different phases of the development of a ML based product feature. The challenges listed here are those beyond the typical expected daily tasks of ML experts such as cleaning data, preprocessing, training and evaluating models until some acceptable level of perfor- mance is achieved. Note that the proposed bucketing of roles and challenges is not absolute: different companies — and even different projects within the same company — may encounter a different set of challenges or experience an active involvement of key stakeholders across multiple phases of the project.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Proceedings of the Human in the Loop Machine Learning Work- shop, International Conference on Machine Learning, Sydney, Australia, 2017. Copyright 2017 by the author(s).
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	the potential impact of an ML system and ultimately make a decision on whether the idea deserves funding and the al- location of resources (both human and computational). In the execution phase data-scientists will come up with some early exploration of the potential in the form of a prototype. An experiment could provide some quantitative validation of the system. If all goes well then a development team needs to polish and “productionise” the prototype, while the data engineers team needs to secure the stability and availability of the required data sources. Practice shows that the productionasition of prototypes may seem like the last 10% of the project but can actually take 90% of the time without some provisions. After the feature is com- plete it is surfaced to end-users. After that point typically Site Reliability Engineering (SRE) team will monitor its performance.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	This paper discusses some of the challenges we faced in At- lassian, the role of the different stakeholders and the chal- lenges we encountered as we started investing more in ML based solutions.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	need to decide whether the idea should get the required funding. To make that decision two main factors need to be considered: the size of the opportunity and the feasibil- ity of the idea.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	The size of the opportunity is an estimation of how many users will get value from the new system. Is it the entire user base or a small segment of it? For example, a content recommendation system provides value only if the volume of content is large enough so that simple browsing is not an adequate solution or if the pace of creation of new con- tent would require the constant attention of users. Product managers would need some help to evaluate the magnitude of the impact that such a system would have on their users. This should be compared to the value provided from exist- ing functionality. In the case of the content recommenda- tion system, that would be the current browsing or search capabilities of the product. In summary, the new ML sys- tem should add value on top of that provided by the current functionality and provide that value to an adequate number of end users.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Another factor to be considered is the feasibility of the idea, given limited resources in time, people and data. This re- quires some feedback from ML practitioners. The ques- tions to be answered include: what is the right metric we try to optimise for? Are the data and infrastructure required for operation currently available? How conﬁdent are we that a ML system can be built with some minimum level of performance in key metrics like mean square error, or pre- cision and recall? Currently there is not much research that helps practitioners estimate the anticipated performance of the system given the data available, in terms of samples, features and some methodology. They need to rely on rules of thumb, their experience and previous efforts in similar systems to make an estimation. Some follow up questions include how easy it would be to maintain such a system, and whether there are any privacy concerns. While answer- ing all these questions requires subject matter expertise, they are outside the typical training of a ML practitioner on building a learning system.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	A very sensitive area that affects decisions on the feasibil- ity or the design of a ML system are concerns around the privacy of users and their data. The legal framework within which ML systems operate is changing fast as governments try to regulate the access of companies to the private data of their citizens. Legal and risk departments of companies are getting involved and the design of ML based systems is severely affected in an effort to provide compliance with the privacy policy of the organisation and the legal frame- work. A new European regulation, namely the “General Data Protection Regulation” (GDPR) that will take effect in May 2018, introduces new requirements for companies and strict rules on how user data can be used (Bryce Goodman, 2016). There are still quite a few unknowns and blurry ar- eas, however the new challenges for ML practitioners are to balance company policies and procedures, comply with the new upcoming legislation, respect of user privacy, and at the same time design automation and product functionality that requires access to the raw data. Some hard questions include how to debug a ML system at scale when there is no access to the raw data generated by users? Can users be proﬁled based on their sex, age, experience, their role? If not, can these attributes indirectly be inferred from other known variables and fed into a ML system? The building of a ML system is a challenging experience considering that such systems are the central places where data from dif- ferent data sources and potentially different users are inte- grated, in non-interpretable ways, to offer the desired func- tionality.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	puts, cleaning up and preprocessing the data. Sometimes a solution may require further post processing and ad-hoc rules for edge cases.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	When a prototype is ready it will be deployed to some end users. Only then some problems will be revealed. For ex- ample, maybe the system does not achieve the minimum level of required performance; or maybe the target concept is not learnable for all users and all the different ways they use the product; or maybe more data is required. Problems like these may be hard to detect when building a proto- type and before deploying to real users. While this risk is well understood by ML practitioners and they can pro- vision some work arounds, it is a quite a novel experience for software engineers that are accustomed to working with deterministic systems.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Another class of common problems is around the scalabil- ity challenges of traditional ML techniques. While an in- dividual ML system may be reasonably fast to train and in making predictions, it may actually be quite slow over- all when we consider the number of times it needs to be deployed. Suddenly even a few-minutes training time be- comes a blocker if the training must be repeated for mil- lions of users. Techniques that assume unbounded re- sources, for example the construction of a “full” dictionary of term frequencies, or techniques that scale linearly with the number of samples suddenly appear as quite slow per- formers.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Obviously ML practitioners are the most qualiﬁed people to create ML solutions. However, their skills are a scarce resource for most development teams. A typical engineer- ing team does not have the experience to deal with this set of problems (Zinkevich, 2016). Some solutions include the renting of external services or the use of consultants. How- ever, external consultants can not support and maintain the system after its delivery, as discussed further in the next section.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Even simpler problems that require the creation of back- end business services suffer from unique challenges. For example, in our company there was a need to automate the monitoring of some business metrics and provide au- tomated alerts for anomalies in the data. The development team tasked to solve the issue explored the options to either rent or build a relevant service. Both choices had a caveat though: there should be some human feedback to tune and maintain the system, speciﬁcally a process to notify the sys- tem of false positives and false negatives. From the devel- opment team’s perspective that looked like a minor step in the process, compared to the challenge of creating the al- gorithm that detects the anomalies. On the other hand, the data scientists that were pulled in for domain expertise all reached the consensus that building such a system would be easy, if not trivial, but the tuning phase would be the real challenge and the time consuming step. They advised that the system will require full re-training, for example in cases of data loss/delays, or annotation of data points that represent one-off well understood events, and in cases of more permanent changes a mechanism for the system to become less sensitive until a new baseline is established.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	The maintenance and operation of a ML based system is a signiﬁcant challenge. The system faces not only the chal- lenges of any other software system, but even more that relate to its non-deterministic nature that changes depend- ing on the data input. If the system is poorly designed, then its operation is potentially harder than the actual build- ing of the system. An implicit assumption when building an ML system is that all the underlying data sources will be available, stable and with similar statistics as those en- countered in the training phase. Even if one data source becomes unavailable the whole system’s performance may degrade to zero. In practice, the assumption of a stationary environment is problematic, especially when the system is designed to consume data sources that are typically outside of the control of the team that built it. While to some extent the missing data may be considered as input noise from a system designer’s perspective, in practice this “noise” may sustain for a signiﬁcant amount of time, saturating online learning systems. The ML practitioners end up trading off the performance of the system for stability and reliability (Sculley et al., 2015).
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	The development of an operational ML system requires the involvement of many different stakeholders. These people may be experts in their respective ﬁelds and they all need to cooperate to successfully productionise a ML system. The ML practitioner needs to provide input for a series of deci- sions outside her strict domain of expertise and beyond the typical task of training a ML system. The development of the system requires taking into account factors that balance the performance with the scalability and the cost of long term maintenance of the system.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Author is grateful to Arik Friedman, Daniel Louw and Her- cules Konstantopoulos for discussions and feedback; and the anonymous reviewers for their constructive comments.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Breck, Eric, Cai, Shanqing, Nielsen, Eric, Salib, Michael, and Sculley, D. What´s your ML test score? A rubric for ML production systems. In Reliable Machine Learning in the Wild - NIPS 2016 Workshop (2016), 2016.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	In practice, the operation of an ML system requires con- stant monitoring of some key performance metrics. Moni- toring the system performance requires also training devel- opment operations on what signals need to be monitored and what needs to be done when things go wrong. Some best practices for monitoring such systems are discussed in (Breck et al., 2016). Sometimes the system performance may degrade to a level that makes it unusable. A solution in those cases would be to use feature ﬂags, a software devel- opment methodology that allows turning off or swapping backend services on the ﬂy. This way the system may fail over to some basic functionality until the problem is solved.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Bryce Goodman, Seth Flaxman. European union regula- tions on algorithmic decision-making and a “right to ex- In Workshop on Human Interpretability in planation”. Machine Learning at the International Conference on Machine Learning, 2016.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	Sculley, D, Holt, Gary, Golovin, Daniel, Davydov, Eu- gene, Phillips, Todd, Ebner, Dietmar, Chaudhary, Vinay, Young, Michael, Crespo, Jean-Franc¸ois, and Dennison, Dan. Hidden technical debt in machine learning systems. In Advances in Neural Information Processing Systems, pp. 2503–2511, 2015.
Processed_Beyond_the_technical_challenges_for_deploying_Mach.txt	A relevant policy typically found in a software company is summarised in the phrase “you built it, you maintain it”. This policy renders development teams responsible for the proper operational status of the services and systems they build. They can not build something suboptimal and ex- pect the operations and support teams to maintain it. In the case of the ML systems that policy means that the data scientists need to create stable systems with provisions for missing or noisy data sources. Otherwise they will end up continuously maintaining and tackling problems one after the other, focusing all their time in one single system. The problem becomes worst if “edge cases” need to be tackled separately and are ignored at design. In a system that is used by millions of users the edge cases are actually ex- pected to be encountered quite often.
Processed_Screening_in_two-dimensional_gauge_theories.txt	We analyze the problem of screening in 1+1 dimensional gauge theories. Using QED2 as a warm- up for the non-abelian models we show the mechanism of the string breaking, in particular the vanishing overlap of the Wilson loops to the broken-string ground state that has been conjectured in higher-dimensional analyses. We attempt to extend our analysis to non-integer charges in the quenched and unquenched cases, in pursuit of the numerical check of a renowned result for the string tension between arbitrarily-charged fermions in the massive Schwinger model.
Processed_Screening_in_two-dimensional_gauge_theories.txt	c(cid:13) Copyright owned by the author(s) under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike Licence.
Processed_Screening_in_two-dimensional_gauge_theories.txt	Two-dimensional U(1) gauge theory with fermions (QED2) has long been a test-bed for con- cepts relating to four-dimensional gauge theories. Being extremely simple compared to QCD it captures some of its crucial non-perturbative features such as conﬁnement and string breaking. Although over the years there has been a plethora of numerical studies of QED2 using various methods such as discrete light cone quantization, hamiltonian lattice ﬁeld theory, euclidean LFT (see Refs. [1, 2, 3] and references therein), this system still attracts a lot of attention, see e.g. Refs. [2, 4].
Processed_Screening_in_two-dimensional_gauge_theories.txt	As far as the pure gauge theory (Quantum Maxwell Dynamics, QMD2) is concerned, when formulated in R2 it turns out to be trivial and results in a linear conﬁning potential for the probe charges [5]. On a torus the situation is altered – due to the fact that one cannot gauge away the ﬁelds along closed contours there exists a single space-independent quantum degree of freedom and the spectrum of the theory yields the Manton’s model of QMD2 on a circle [6] in the continuum limit [7].
Processed_Screening_in_two-dimensional_gauge_theories.txt	Non-abelian two-dimensional models, which are the ultimate aim of our investigation [10] are even more interesting. There are theoretical predictions for the spectrum of QCD2 with fundamen- tal fermions [11] and the large-N limit of the theory was analytically solved by ’t Hooft [12]. The practicality of this solution is however limited by the fact that the fundamental matter is quenched in the large-N limit, thus even more interesting are the large-N limits of theories with two-index representation matter where the fermion dynamics plays an important role.
Processed_Screening_in_two-dimensional_gauge_theories.txt	For example, the adjoint fermions in 1+1 dimensions were analyzed both theoretically, giving encouraging results [13], and numerically by DLCQ techniques [14]. However, no lattice analysis has been performed until recently.
Processed_Screening_in_two-dimensional_gauge_theories.txt	In this letter we report the results of our study of QED2, treated as a warm-up for the non- abelian models. In Section 2 we study the string breaking in QED2 using Wilson loops and in Section 3 we sum up our attempts to deﬁne operators carrying fractional charges and to calculate the string tension with them, which would lead to a lattice veriﬁcation of Eq. (1.1).
Processed_Screening_in_two-dimensional_gauge_theories.txt	Figure 1: Energies and overlaps to the Wilson loop of the ground state and ﬁrst excited state, V = 24 × 24, g−2 = 1.5, κ = 0.245. The red points of in Fig. (a), inserted for comparison, are the numerical calculation of the (analytically known) pure gauge result.
Processed_Screening_in_two-dimensional_gauge_theories.txt	Monte Carlo simulations. We use Wilson fermions and generate the conﬁgurations using the Ra- tional Hybrid Monte Carlo algorithm. As a test of the possibilities of the simulation we analyze the string breaking using Wilson loops.
Processed_Screening_in_two-dimensional_gauge_theories.txt	It is well known that QED2 resembles QCD4 in this respect. It exhibits a conﬁning linear potential in the pure gauge case (as does pure gauge lattice QCD4 [15]) and charge screening at large distances, interpreted as string breaking by quark-antiquark pair [9, 16]. The string breaking is very hard to observe in the lattice QCD using Wilson loops alone – a poor overlap of the Wilson loop to the broken-string ground state was postulated and a larger set of operators had to be used to deliver a ﬁrm observation [17].
Processed_Screening_in_two-dimensional_gauge_theories.txt	extracting the ground state and the ﬁrst excited state from the Wilson loop together with the cor- responding overlaps. Sample results are presented in Fig. 1. It is natural to interpret Fig. 1(a) as a conﬁned state and a broken-string state, where the latter becomes the ground state around R (cid:39) 4−5, with quantum repulsion (i.e. mixing) of states observed. Perhaps the most instructive result of this exercise it Fig. 1(b) where one can see how the Wilson loop clearly prefers the conﬁned state, explaining the difﬁculty of observing the string breaking using solely this operator.
Processed_Screening_in_two-dimensional_gauge_theories.txt	In this section we analyze the possibility of introducing operators carrying arbitrary real charge Qext to verify the equation (1.1) (from now on, to be concise, we refer to Qext simply as Q). This is a relatively unexplored topic on the lattice. One notable exception is Ref. [18] presenting a hamiltonian lattice analysis where real charges were introduced by means of constant electric ﬁeld.
Processed_Screening_in_two-dimensional_gauge_theories.txt	While the two coincide for integer charges, they can give vastly different results for the non-integer case when the complex logarithms used to deﬁne the non-integer powers fall on different branches. The numerical results presented in this work are obtained using the ﬁrst deﬁnition (3.1). We will brieﬂy discuss the different results given by the second method later in this section.
Processed_Screening_in_two-dimensional_gauge_theories.txt	To ﬁnd the string tension dependence on Q in the massive Schwinger model we have performed similar analysis as for the “ordinary” Wilson loops in the previous section. For every real Q we found that up to additional perimeter terms, that do not inﬂuence the string tension, the Wilson loops with charge Q behave in a very similar manner to the ones with the integer charge closest to Q – thus for every charge analyzed the data is consistent with σQ = 0.
Processed_Screening_in_two-dimensional_gauge_theories.txt	To understand this result we went to an even simpler theory i.e. the pure gauge model (QMD2). There we know the exact results for the string tension of integer charges both on inﬁnite [5] and ﬁnite lattices [7]. In the continuum limit one might also expect the well-known continuum result σQ ∼ Q2/2 for all real charges.
Processed_Screening_in_two-dimensional_gauge_theories.txt	The ansatz assumes that the ﬁnite size effects for the integer charges are negligible. In fact, using the results of Ref. [7] one can include them, leading to an exact result [19].

Subsets and Splits