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_Euclid_preparation._Detecting_globular_clusters_in.txt	In order to better understand the number of GCs that will be actually detectable, we used simulated Euclid VIS images (Eu- clid Collaboration: Serrano et al. 2024) and we inserted local- Universe galaxies and GCs in them. We then ran those images through the Euclid Consortium pipeline processing functions that combine four VIS exposures to create a mosaic, subtract extended background emission, and produce catalogues. The pipeline’s individual mosaic images are smaller than a VIS field of view (they are square with a side of 32′) and have pixels of 0 .′′1 size.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Our added objects are positioned in individual VIS exposures such that they fit within one mosaic. We simulated four configu- rations (labelled FoV1 to FoV4), in which we added galaxies and their corresponding GCs at several different distances: 5 Mpc, 10 Mpc, and 20 Mpc for FoV1 to FoV3, and a combination of distances between 40 Mpc and 60 Mpc in FoV4 (with a few ex- ceptions at 20 Mpc, as specified below).
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The injection of globular clusters and galaxies into simulated VIS exposures is done through stamps generated for a given RA and DEC location on the different CCDs by the GalSim Python package (Rowe et al. 2015). It enables the convolution of in- jected objects by the VIS PSF and the addition of Poisson noise to these stamps. The native GalSim package contains a Sérsic function used for galaxies detailed later. For the EGCs, we have written a custom function to allow GalSim to use the King model (King 1962).
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 8. An example of the input galaxies and GCs that were added to simulated Euclid pointings, which already contained foreground stars and background galaxies. The host galaxies are represented by surface- brightness contours marking the half-light radius, µ = 23 mag arcsec−2, and µ = 29.5 mag arcsec−2. The artificial GCs are distributed in a grid pattern, coloured according to their magnitudes. This image is FoV1 for which the distance adopted for the added objects is 5 Mpc. The coordi- nates are shown in units of 0 .′′1 pixels and the colour bar is the apparent IE magnitude of the injected GCs.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	All added galaxies were given a Sérsic profile with an index of n = 2, an ellipticity of 0.3, an effective radius of reff = 1 kpc, = −18.5 mag, but projected to and an absolute magnitude of MIE the required distance. In FoV4, 29 of the standard galaxies were replaced with modelled ultra-diffuse galaxies, with MIE = −14 and reff = 2 kpc, placed at a distance of 20 Mpc. As the appar- ent sizes of the galaxies shrink with distance, the simulations of further distances include more galaxies overall.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	On top of these galaxies we add GCs with varying magni- tudes and colours, all represented with King models. Their IE magnitudes are chosen such that they cover the GC luminosity function in steps of 0.5 mag, within the range detectable in Eu- clid VIS images at a given distance. The range of absolute and apparent magnitudes for each distance are listed in Table 1.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	We adopted a typical GC core radius of rc=1.2 pc and a King concentration index of rt/rc = 1.4, where rt is the tidal radius where the density falls to zero for GCs (McLaughlin & van der Marel 2005). In all simulations except the one at 5 Mpc one, we also included larger size GCs with core radii of 4 pc. The host galaxy properties and the GC ranges are summarised in Table 1 and the arrangement for FoV1 is shown as an example in Fig. 8.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The most straightforward way of identifying GCs is not to ana- lyze the vast data set of VIS images that will be taken by Euclid, but rather to use the data catalogue that will be created by the Euclid analysis pipeline automatically. This pipeline-catalogue contains the merged information for all extracted sources from both VIS and NISP. As such, it contains fluxes in the IE, YE, JE, and HE bands, positions, morphological information such as their compactness, central surface brightness, or their likelihood to be a star or a galaxy. In this catalogue a large quantity of informa- tion on the detected sources is already present. This pipeline1 was also run on our simulated VIS data after the addition of the galaxies and GCs.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Using these output catalogues for artificial GCs will give us an idea of how such objects compare to other sources in the field, and thus what purity and completeness we can expect of a GC candidate catalogue derived from this source catalogue. In addi- tion this gives us the opportunity to compare how accurately the pipeline recovers the known input properties of the GCs.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	To this aim we match the input catalogue of artificial-GC po- sitions with the output pipeline-catalogue. We use a small match- ing radius of 0 .′′1 to avoid random field contamination. The aver- age distance of our matched positions is 0 .′′05, corresponding to half a pixel in VIS and thus indicating that we indeed retrieved the input GC sources. Overall we find that the automatic Euclid pipeline was able to detect 18 340 GCs compared to the 24 154 artificial GCs that were injected, for an overall recovery fraction of 76%.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Notes. Column 1 recalls the labels of our four simulations. Column 2 is the distance at which the added galaxies were placed and Col. 3 specifies whether these galaxies are standard (see text) or the UDG model. Column 4 gives the number of host galaxies at this distance, and the Col. 5 the total number of GCs distributed across all host galaxies. Columns 6 and 7 are the magnitude range of added GCs in absolute and apparent IE magnitudes, respectively.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	by spectroscopy. However, the recovery fractions indicate how many we can expect in the catalogue to begin with, based on which we generate a GC candidate list.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Figure 9 shows how the GC detectability depends on the local surface brightness at the GC location and the input IE magnitude of the GCs. The local surface brightness at the position of each inserted GC is derived by calculating the surface brightness in an annulus between r = 2′′ and r = 3′′. The error bars shown are calculated as the 95% binomial Wilson confidence intervals.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The Euclid pipeline automatically removes foreground galaxy light if it spans more than one CCD. This is the case for FoV1 and FoV2 and thus here the calculated surface brightness is the background of the images plus the residual galaxy surface brightness from this removal. In the two most distant simula- tions, the galaxies are smaller and the pipeline does not attempt to remove the light and thus the calculated surface brightness includes the background plus the galaxy.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 9. Fraction of added GCs that have been detected by the Euclid photometric pipeline as a function of the local residual surface bright- ness at the location of the GC. The four different symbols are denoting the different configurations of the simulations.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The fraction of recovered GCs is mostly below 20% for a surface brightness brighter than around 25 mag arcsec−2 and then increases sharply to an 80–90% recovery fraction when the lo- cal surface brightness drops below that. This surface brightness threshold is the same for each simulation and thus independent of distance. For the most distant simulation FoV4 we find a spike for the recovery fraction at 21 mag arcsec−2, which is likely due to the detection of the inserted nuclear clusters in these simula- tions.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The surface-brightness of 25 mag arcsec−2 at which the de- tectability is very high corresponds to about 4 times the effec- tive radius of the simulated galaxy in FoV3 and FoV4, where the pipeline does not automatically subtract the galaxy light. In the two closest simulations FoV1 and FoV2 this limit is closer to the core of the galaxy as the pipeline removes galaxy light from foreground galaxies that span at least an entire detector.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	We conclude that host galaxy light subtraction is required for the detection of GCs close the the host centre. For nearby (< 20 Mpc) galaxies, the Euclid pipeline does this in an auto- mated, albeit not perfect manner, whereas for galaxies at larger distances this will have to be done separately. Only for ultra-faint dwarfs is this effect negligible, as their central surface brightness levels are not much larger than 25 mag arcsec−2.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Finally, it is worth noticing that in FoV4 the recovery fraction for the GCs placed on low surface-brightness galaxies is 96%. Euclid will excel at detecting the GCs of dwarf galaxies. We refer to Saifollahi et al. (2024) and Marleau et al. (2024) for first results in this area, based on early observations of both the Fornax and the Perseus galaxy clusters.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 10. Left column: Comparison of input and output IE magnitudes of the simulated GCs. The distributions of measured magnitudes from the Euclid photometry pipeline are plotted against the input values, with the mean and the extremes marked with horizontal bars. From top to bottom we show the results for simulated FoV 1, 2, 3, and 4, which correspond to increasing galaxy distance. The solid line marks the 1:1 relation. Right column: Histograms of the differences between the input and output IE magnitudes for the GCs for the same four FoV.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	accurately the recovered magnitude compares to the input mag- nitude. We show the comparison of input VIS magnitude against the recovered VIS aperture photometry in the left column of Fig. 10. The aperture diameter used for the ouput VIS photome- try is twice the FWHM of the PSF. We use aperture photometry instead of PSF photometry as our mock GCs are substantially more extended than the PSF. This figure shows only those ob- jects that had a match in the output data and does not account for those that were not detected by the photometric pipeline. Clus- ters inserted exactly at the centre of a galaxy were also removed, as their recovered magnitudes are inevitably too bright.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	than the input value for the faintest GCs in the two more dis- tant simulations due to injected GCs approaching the detection threshold at 26 mag. Thus, close to the detection threshold the retrieved magnitudes are likely biased due to contamination by noise and neighbouring objects.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	An important difference between the nearby simulations at 5 and 10 Mpc and the more remote ones, is that the nearby host galaxies span more than one VIS-detector chip, and some of the emission in their outer parts is removed by the background- subtraction algorithm of the default pipeline. For the two more distant simulations the size of the simulated host galaxies is smaller, so that their light remains present in the mosaics. This is likely a major source of added uncertainty in the retrieval of GC magnitudes for the two simulations at larger distances.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	In this section, we use the simulated VIS mosaics exactly as they are returned by the pipeline. We could subtract the galaxy light for the smaller galaxies, yet this type of manual intervention into the imaging data will require a very large computational cost due to the vast data size of the full EWS. Thus we want to validate our GC detection accuracy on standard Euclid data products. However, in the future it might be useful to remove the host galaxy light in an additional step for selected sections of the full Euclid survey that are of particular interest to EGCs, such as in galaxy clusters.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The larger the distance, the harder it becomes to distinguish the image of a GC from that of a point source, which may be a star or a small background object. We will thus fall back to deriving a catalogue of candidates for subsequent follow-up. Catalogue column combinations that inform indirectly about source com- pactness make it possible to reduce the number of candidates drastically (Cantiello et al. 2018). For example in Voggel et al. (2020) the astrometric excess noise tabulated in Gaia catalogues was combined with magnitude around Centaurus A, and a source catalogue of several million point-like sources was reduced to several hundred good candidates, which then facilitated a much more focused follow-up. In the central 30 kpc of Cen A, the spec- troscopic success rate of that selection was 68% and outside of 30 kpc it dropped to 8% as in the sparse very outer regions real GCs are more rare (Hughes et al. 2021).
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	We test the purity and completeness of a GC catalogue based on simple selection criteria from the Euclid photometric cata- logue that is provided with each observation. As this is an ide- alised framework based on simulations with limited GC sizes and magnitudes, and only a single type of host galaxy, we as- sume that in reality the foreground/background separation from the GCs will be somewhat more difficult than what we test here. Nevertheless, this offers us a first estimate of the expected con- tamination by fore- and background objects.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 12. Cut-out of the Euclid VIS image focused on NGC 1374. The circles mark the location of the GC candidates in this region.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 11. Maximum surface brightness of all sources in the Euclid output catalogue of each of our four simulated tiles as a function of the IE mag- nitude. All sources are plotted as the blue density distribution and the GCs that we added to the simulations are shown as the orange density distribution.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	ness of GCs is in theory lower than those of stars, but higher than for galaxies. The figure lets us evaluate how much that effect is present in simulated Euclid data.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Figure 11 shows that the GCs fall into a specific parameter space that is offset from the locus of galaxies and stars. Stars form a tight sequence above the GCs location as they have a higher surface brightness compared to GCs, and galaxies form a cloud of data points at lower surface brightness. This separa- tion of objects is clearest for the closest FoV1 at 5 Mpc and with increasing distance in FoV4 the GCs do not separate out eas- ily from background galaxies anymore. Their location in terms of surface brightness has strong overlap with galaxies and stars, and thus more parameters are needed to ensure an efficient pre- selection of GC candidates.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	We can test the combination of selection methods further using Euclid early-release observations (ERO) of the Fornax galaxy cluster (Saifollahi et al. 2024; Cuillandre et al. 2024), which have been acquired recently. The Fornax cluster was chosen be- cause, being located at a distance of about 20 Mpc, it is one of the nearest galaxy clusters to the Milky Way, and it contains a large number of GC-hosting galaxies. Two VIS observations with an exposure time of 560 s each were taken, centred on posi- tion RA=3◦ 36′ 8 .′′759 Dec=−35◦ 16′ 0 .′′38, and were co-added and processed with an ERO-specific pipeline, which is half of the expected observing time of the EWS where four exposures are the standard. A cut out of the image focussed on NGC 1374 is shown in Fig. 12. The co-added full image is shown in the bottom panel of Fig. 15.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Follow-up observations of previous ground- and space-based surveys of Fornax have established a set of spectroscopically confirmed GCs (Saifollahi et al. 2021; Chaturvedi et al. 2022), that we can use to train an automatic classification algorithm. We initially intended to use Euclid-pipeline catalogues for the For- nax VIS data to ensure comparability with the results from the simulations, but this was not possible so early after launch with data taken in a sequence that does not correspond to the standard sequence of the future EWS. We thus ran Source Extractor (Bertin & Arnouts 1996) on the VIS images independently of the pipeline. The resulting catalogue is called the‘ERO-Fornax catalogue’ hereafter.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	When matching the known GCs with the ERO-Fornax cata- logue, we found a handful of sources with ellipticity e > 0.6 and FWHM > 0.75 pixels in the VIS images. Typical GCs are round and do not have large half-light radii. We clipped our sample at 3σ above the mean in FWHM and ellipticity, which resulted in a sample of 181 GCs with ellipticities smaller than 0.38 and FWHM < 0.64 pixels. Upon visual inspection, we found that the rejected sources were affected by blending with faint nearby sources, although this did not trigger the blending flag in Source Extractor as these blending sources were too close and faint to be picked up as secondary sources. In two of these objects they were close to the centre of larger galaxies, which caused the faulty ellipticities. Sources that are blended will be challenging cases to identify as bona fide GCs with such automated classifi- cation methods.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 14. Non-normalised distribution of log likelihoods of extracted sources: known GCs (orange) and all sources (blue). The vast major- ity of the source catalogue falls in the smallest bin of probability, which includes all log-likelihoods below 40 in the units of this figure.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	decided to limit our data set for classification to IE ≤ 24 because that is the magnitude limit of our training sample.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	We train a Gaussian mixture model on the known GCs in the five-dimensional parameter space defined above. The results of this training is shown as ellipses in Fig. 13. The red and yel- low ellipses mark the levels that include 50%, 90%, and 95% of all known GCs in the Gaussian mixture output. This Gaussian model, which is essentially a description of the region of five- dimensional parameter space accessible to GCs, is then applied to the whole photometric data set of 32 554 sources brighter than 24 mag whose nature is unknown. Sources near the centre of the Gaussian model are intrinsically more likely to be GCs than sources near the outer contours. The corresponding likelihood distribution of all sources is shown in Fig. 14. At the cost of lower completeness, one could consider additional truncations based on the preferential locations of non-GC populations, but we focus on completeness here.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 13. Parameters of detected sources against VIS IE magnitude. Panel a) is the concentration index C4−8, which is the difference in magnitude between the aperture with a radius of 4 and 8 pixels. The parameter shown in panel b) is the FWHM of the sources in arcsec, panel c) is the ellipticity and panel d) is the maximum surface brightness. All four pa- rameters are derived from the ERO-Fornax catalogue and are intrinsic size parameters deconvolved with the PSF. Known GCs are shown as red symbols. The blue shaded density distribution represents all photo- metric sources in the Fornax FoV. The yellow, orange, red and maroon ellipses mark the levels that include 95%, 90%, 68%, and 50% of all known GCs in the Gaussian mixture output.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The vast majority of the catalogue sources have very small likelihood values and are thus easily excluded as GCs. The tail at high likelihoods for the full catalogue is where the most likely GCs are located, with properties similar to those of known ob- jects. We make a cut at a log-likelihood of 8.5 (in the units of Fig. 14) for objects that we consider good candidates. That value corresponds to the ellipse that includes 90% of all known GCs (second ellipse from the outside in Fig. 13) assuring a good com- pleteness of our candidate sample. This results in 1541 sources, out of 32 363 in total, that we consider as good GC candidates. The known objects have been removed from the source cata- logue so these are all new candidates. While we do expect a sub- stantial contamination of faint galaxies and stars, the decrease of our original source catalogue by 95% is the key for an ef- ficient selection of GC candidates solely from the photometric catalogues of the full EWS.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The spatial distribution of the 1541 high-likelihood GC can- didates is shown in the top panel of Fig. 15. The colour and con- tours in the top panel are the normalised kernel density estimate for the GCs using a Gaussian kernel.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Fig. 15. Top panel: Positions of the 1541 high likelihood GC candi- dates are shown as black data points and location of the main galaxies is shown as white crosses. The colour and contour shows the normalised kernel density estimate of GC density. Bottom panel: The Euclid image from the VIS instrument in the IE band. The white locations are where there are gaps in the data.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	The above analysis shows the efficiency of the training method in a field with known GCs. For nearby galaxies and clus- ters this will enable us to derive a candidate catalogue with a high completeness based solely on catalogue data.
Processed_Euclid_preparation._Detecting_globular_clusters_in.txt	Such a predictive modelling has the advantage that it is com- putationally cheap and fast, thus enabling an efficient selection of GC candidates for further study. We do expect significant con- tamination from both foreground and background sources, but the removal of typically more than 90% of contaminants will en- able us to make a very efficient follow-up while keeping a large completeness of GCs in our sample.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	noise down to time scales of 80 ns. We observe a signiﬁcant inverse correlation between pulse peak and width. Thus, the average pulse proﬁle produced by selecting for large pulse peaks is narrower than the standard average proﬁle. We ﬁnd no evidence for “diﬀractive” quantization eﬀects in the individual pulse arrival times or amplitudes as have been reported for this pulsar at lower radio frequency using coarser time resolution (Ables et al. 1997). Overall, we ﬁnd that the single pulse properties of PSR J0437−4715 are similar to those of the common slow-rotating pulsars, even though this pulsar’s magnetosphere and surface magnetic ﬁeld are several orders of magnitude smaller than those of the general population. The pulsar radio emission mechanism must therefore be insensitive to these fundamental neutron star properties.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Single pulse studies of millisecond pulsars are considerably more diﬃcult to perform than are those of slow pulsars. Faster data rates are needed to study millisecond pulsars with comparable pulse phase resolution, and ﬁner radio frequency resolution is required to minimize the eﬀect of interstellar dispersion. Also, interstellar scattering time scales comparable to the pulse duration render studying individual pulse morphologies impossible, while steep millisecond pulsar spectra preclude observations of suﬃcient sensitivity at higher radio frequencies where scattering is less important.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Yet single pulse studies of millisecond pulsars are highly desirable for two reasons. First, the origin of the radio emission that has made isolated neutron stars famous is, even 30 years after their discovery, still a mystery. The high brightness temperatures (∼ 1030K) associated with the radio emission point to coherent processes which are poorly understood even under less exotic conditions (Melrose 1996). Previous observations of slow pulsars have not constrained the emission mechanism suﬃciently; the study of radio emission properties of millisecond pulsars may provide important new clues. Millisecond pulsars, because of their fast spin periods, have much smaller light-cylinder radii, and hence magnetospheres, than slow pulsars. They also have lower surface magnetic ﬁeld strengths than the general pulsar population (most likely resulting from their having been “recycled” by a binary companion through the accretion of mass and angular momentum). Were the radio emission mechanism at all dependent on such properties, millisecond pulsars should have diﬀerent radio properties than the slower-spinning general population. The second reason single pulse studies of millisecond pulsars are important is that millisecond pulsar timing is well- known to be an unparalleled source of precision astrometric and astrophysical information. Among factors possibly limiting timing precision is the stability of the average proﬁle, which depends on the properties of single pulses.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Only recently has a systematic study of single pulses from millisecond pulsars become possible, largely due to improving computational and data recording technologies. To date, the only such investigation has been for the 1.5 ms pulsar PSR B1937+21 (Wolszczan, Cordes, & Stinebring 1984, Sallmen & Backer 1995, Backer 1995, Cognard et al. 1996). Interestingly, PSR B1937+21 exhibits giant radio pulses like those seen elsewhere only in the Crab pulsar. With the single pulse properties of only one millisecond pulsar having been studied in any detail, the question of whether all millisecond pulsars show similar properties naturally arises. Unfortunately, PSR B1937+21 suﬀers interstellar scattering at time scales comparable to the duration of a single pulse at the radio frequencies at which it has been observed, rendering detailed study of individual pulse morphologies diﬃcult.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Here we report on high-time-resolution single pulse studies of a second millisecond pulsar, PSR J0437−4715. This pulsar’s large ﬂux density and low dispersion measure (DM), and the corresponding scarcity of line-of-sight scattering material, render it an obvious target for single pulse work. Some single pulse investigations of PSR J0437−4715 have been reported (Johnston et al. 1993, Ables et al. 1997) but none have had suﬃcient time resolution to resolve most individual pulses. Using a fast recording device and powerful supercomputers, we have been able to resolve all pulses in our data, the narrowest being ∼10 µs.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	All data reported on here were obtained at the 64-m radio telescope of Parkes Observatory in New South Wales, Australia on 24 and 25 July, 1995 at a central radio frequency of 1380 MHz. Observations were made using a cryogenically-cooled, dual-channel system which received orthogo- nal linear polarizations. The signal path from the receiver was as follows. A local oscillator signal, locked to the Parkes Observatory frequency standard, was ampliﬁed and then mixed with the in- coming RF signal from the receiver and low-pass ﬁltered to form a single-sideband IF. These IF signals (one for each polarization) were relayed to the control room, where they were ampliﬁed and mixed with a second LO. This second mixer operated as a baseband quadrature mixer, with the in-phase and quadrature-phase output signals low-pass ﬁltered to 25 MHz. Thus, a total bandwidth of 50 MHz at center frequency 1380 MHz was available for recording. The four analog signals from the complex downconverter were then digitized with 2-bit resolution and recorded at the Nyquist rate (400 Mbit s−1) on a digital tape recorder (Datatape LP-400) along with timing information synchronized to the Parkes Observatory clock. The output bus operates at 50 Msample s−1, and is 8 bits wide: 2 digitizer bits × 2 polarizations × 2 signal phases. Thus, a single 13 min observa- tion resulted in 39 Gbytes of raw voltage data. Detailed information about the baseband mixing, digitizing, and data recording system is reported by Jenet et al. (1997).
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	are summarized in Table 1. A calibration scan was done prior to Observations 1 and 3 by moving the telescope oﬀ-source and pulsing a noise diode source at a ﬁxed frequency. The amplitude of the calibration source, and hence the absolute ﬂux scale in Janskys for our pulsar observations were determined using observations of the bright radio continuum source Hydra.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Storage of raw voltage data permits great ﬂexibility in data analysis, but also requires sig- niﬁcant computational power. Most data reduction was therefore done on the massively-parallel Caltech 512-node Intel Paragon XPS L38 supercomputer which has peak computation rate of 38.4 Gﬂop s−1, as well as on the 32-node Intel Paragon XPS A4 supercomputer which has 4.3 Gﬂop s−1 peak computation rate. Data from the Datatape recorder were read into a 0.5 Gbyte Datatape variable rate buﬀer, and then onto the supercomputer ﬁle systems via a high-speed HIPPI network. Most analysis was done using specialized parallel code written in C++, with the NX message pass- ing interface. For details about the hardware and analysis software tools, see Jenet et al. (1997).
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Once on the parallel computers, the raw data is unpacked into ﬂoating point numbers and the interstellar dispersion handled by one or both of two techniques. The ﬁrst, and the less com- putationally intensive, uses a software “incoherent ﬁlter-bank,” which simulates the output of the conventional hardware ﬁlter-bank spectrometer, namely samples of power in many individual nar- row frequency channels. The channelized data are subsequently added with appropriate time delays to achieve a dedispersed time series having time resolution limited by residual dispersion within channels. The second, more computationally-demanding dedispersion method does a direct Fourier deconvolution of the interstellar medium transfer function, i.e. “coherent dedispersion” (Hankins & Rickett 1975). For PSR J0437−4715, it is easy to show that 128 ﬁlterbank frequency channels are needed to minimize single-channel dispersion and time resolution. This results in a time resolution of 2.56 µs and a dispersion smearing of approximately 3.26 µs. In principle, coherent dedispersion can yield a time resolution equal to the inverse of twice the bandwidth, however in practice one is limited by the precision with which the DM is known. For PSR J0437−4715, the uncertainty on the published DM (Sandhu et al. 1997) implies that the true time resolution of our coherently dedispersed time series is no better than ∼200 ns; we conservatively chose to average our coherently dedispersed time series to a time resolution of 320 ns. Higher time resolutions, when necessary, were achieved by coherently dedispersing a smaller bandwidth, obtained using a coarse-resolution soft- ware ﬁlter bank. For example, we achieved 80 ns resolution by coherently dedispersing a 6.25 MHz band. It is important to note that multi-frequency DM measurements are very accurate, but not necessarily precise. Therefore, such DM values may be of limited use in single pulse studies.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	The presence of large amplitude signals in the 2-bit quantized voltage data will introduce unwanted artifacts in the ﬁnal dedispersed time series if proper care is not taken. For bright pulsars like PSR J0437−4715, appropriate corrections are crucial for proper pulse morphology analyses. Two major quantization eﬀects have been identiﬁed: power underestimation and power scattering. We have minimized the power underestimation eﬀects at the data unpacking stage by dynamically adjusting the assigned voltage levels. The eﬀects of the scattered power have only been corrected in the incoherently dedispersed data. A more detailed discussion of these quantization eﬀects and the algorithms used to correct them may be found elsewhere (Jenet 1997, in preparation).
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Thus, we produced a (coherently or incoherently) dedispersed time series for each of the two polarizations. The mean value calculated over a megasample was then subtracted in each polariza- tion, and a gain factor was applied to convert the raw data values to Janskys. The channels were then added to yield the total intensity, or Stokes I. Faraday rotation in the interstellar medium causes a time delay between opposite circular polarizations, as well as a rotation of the angle of linear polarization across the band, however given the low DM and rotation measure (Navarro et al. 1997), both eﬀects are negligible here.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	The known pulsar ephemeris (Sandhu et al. 1997; provided for convenience in Table 2) was used with the TEMPO software package (Taylor & Weisberg 1989) to calculate the expected topocentric pulse period and the pulse phase once every 671 ms. The average pulse proﬁle, shown in Figure 1, was obtained by dedispersing and folding the data in Observation 3, assuming a dispersion measure (DM) of 2.64515 pc cm−3. In the Figure there are 2048 bins across the period, and remaining dispersion smearing due to the ﬁnite size of the simulated ﬁlters is 3.26 µs, just larger than one bin. The pulse morphology is identical to that seen by other groups (e.g. Manchester & Johnston 1995, Navarro et al. 1997) except for the presence of a broad, shallow “dip” in the baseline. A spectral analysis shows that the dip is not dispersed. Hence it is an instrumental artifact. Since each pulse of emission last only about 400 µs in our 50 MHz bandpass, the dip, which is about 2 ms away from the main peak, should not aﬀect the remainder of the analysis. We suspect that this dip is the result of some non-linear process occurring in the receiver chain when the pulse is outside of our bandpass although we have not yet seen this eﬀect in other pulsars observed with the same receiver chain.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Figure 2 is a greyscale plot of 12.5-s of data phase aligned with the pulse peak. Only 368 µs on either side of the main peak is shown. Data used for this plot are incoherently dedispersed, corresponding to a time of resolution 2.56 µs with 3.26 µs DM smearing (as for Fig. 1). No evidence for drifting subpulses can been seen in this ﬁgure. Figure 2 does show regions where the pulse is absent, but we found that averaging as little as ten of these regions reveals the presence of a pulse. Hence, we have no evidence for pulse nulling. The average ﬂuctuation spectrum shown in Figure 3 displays a broad feature indicating a correlation or “memory” between pulses. The ﬂuctuation spectrum is calculated by ﬁrst extracting a time series of intensities at a ﬁxed pulse phase and then Fourier transforming this time series (Backer 1973). In order to reduce the variance in the spectrum, 392 consecutive 256 point spectra were averaged together. The ﬂuctuation spectrum shown in Figure 3 was calculated at zero pulse phase. Structure in the ﬂuctuation spectrum is common among slow pulsars (Manchester & Taylor 1977, Backer 1973). Figure 4 shows a “rogues It gallery” of several phase-aligned sub-pulses from the same data set that produced Figure 2. is clear from Figures 2 and 4 that, as is typical of pulsars, the average proﬁle is a sum of many individual subpulses, which, although forming a relatively stable average proﬁle, individually exhibit a variety of morphologies.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	where a = 4.51 ± 0.01, b = −0.0262 ± 0.0002 Jy−1, Ppeak is the peak power in Jy, and N is the number of pulses. However extrapolating this expression to a pulse peak of 205 Jy suggests only 0.14 pulses at this height should have been observed; that one at 205 Jy was seen suggests the distribution may have a high-peak tail that longer integrations might make observable.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	coherent signal: although they would make a noise signal appear more like a gaussian noise signal, they would not turn a coherent signal into a random noise signal. The envelopes superimposed on the two linear polarizations represent the 98% conﬁdence levels for χ2 1 statistics. If the statistics of the voltage data were given by a gaussian distribution, we would expect the power to be given by a χ2 1 distribution. Using 25 data points to either side of the point in question, we calculated the mean power and used this to ﬁnd the value of the power such that the probability of being less then that value is 0.98. As can be seen from this ﬁgure, the statistics of the emission are consistent with modulated noise. It is also clear from Figure 6 that this single pulse is substantially linearly polarized. The largest ten pulses in our data all show similar behaviour.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	The same subpulse is shown in Figure 7 in each of eight radio frequency sub-bands. It is clear that this pulse is broad-band, extending at least over 50 MHz bandwidth. The scintillation bandwidth at this frequency is much larger than 50 MHz so is not relevant. The sub-band structure is consistent with a gaussian noise process.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	The narrow range of pulse energies exhibited in Figure 8, for a pulsar having very diﬀerent single sub-pulse morphologies, suggests there should be a correlation between pulse peak and width. This correlation is shown in Figure 9. In the plot, only statistically signiﬁcant pulses are included, which explains the lower cutoﬀ in peak ﬂux at the level of the receiver noise.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	timing precision. However, at least in the case of PSR J0437−4715, the steep spectrum (Fig. 5, Eq. 1) of pulse amplitudes precludes timing resolution improvement.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	At least eight bright pulsars display structure in their emission on a time scale shorter than that of the widths of the average pulse proﬁle and individual subpulses (Manchester & Taylor 1977). Such “microstructure” has not previously been detectable in millisecond pulsars both because of insuﬃcient time resolution of observations, and because of DM-smearing and multi-path scattering. Our data on PSR J0437−4715 present the ﬁrst opportunity to detect microstructure in a millisecond pulsar. Along with classical microstructure we can also look for evidence of preferred time scales in the emission that may be due to the presence of primary emitters. The noiselike statistics of pulsar emission along with the high brightness temperatures lead one to postulate the existence of many coherent emitters, or primary emitters, adding up incoherently to form the observed emission (Melrose 1996).
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	We searched for preferred time scales by performing careful autocorrelation function (ACF) analyses of the incoherently and coherently dedispersed data. Preferred time scales would present themselves as rapid changes in the slope of the ACF (i.e. a “break”) and/or local minima and maxima in the ACF. In the incoherently dedispersed data, the time resolution is limited to ∼3 µs; we ﬁnd no evidence for microstructure in the ∼120,000 incoherently dedispersed pulses. In order to search with higher time resolution, ACFs were produced for ∼14,000 consecutive coherently- dedispersed pulses, as well as ∼1000 of the largest amplitude coherently dedispersed pulses. Note that the eﬀective time resolution of the coherently dedispersed data is limited to ∼200 ns by DM uncertainties and non-ideal ﬁlter response. No structure was seen in the ACF (See Figure 11). It is possible that the microstructures could be individually narrow band. Hence, larger bandwidths may wash out any preferred time scales. We coherently dedispersed several sub-bands ranging in width from 6.25 MHz to 781 KHz and centered at 1380 MHz. Again, we found no evidence for preferred time scales. Note that the highest time resolution achieved was 80 ns(=1/2 × 6.25 × 106).
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	pattern in the pulsar’s emission. They emphasized that their results were for the largest amplitude pulses only, which they assumed to be of diﬀerent origin than the emission resulting in the standard proﬁle. Their results predict that with higher time resolution observations, such as those discussed in this paper, the average proﬁle obtained by folding the largest amplitude pulses should show the same fringe pattern.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Figure 10, in which we have folded only the highest amplitude pulses, shows no evidence for any structure, apart from an asymmetric shoulder on the leading edge, which was seen by Ables et al. (1997). We have repeated this analysis by varying the integration time and pulse threshold, and ﬁnd similar results. We note that our fractional bandwidth is the same as that of Ables et al. (1997), hence the eﬀects of the ﬁnite bandwidth should be identical. If their apparent fringes were due to a coherent diﬀractive radiation pattern, the fringe spacing should scale with wavelength; thus we would have expected fringe spacings of ∼4.7 µs, much larger than the 320 ns time resolution of Figure 10. Given the relative sensitivities of our observations and theirs together with the pulsar spectral index, we should have been able to observe their “spikes” with great ease. Thus, we ﬁnd no evidence to suggest coherent radiation patterns exist.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	It is possible to reconcile the disagreement by postulating that the “spikes” have a diﬀerent radio spectrum than does the emission forming the average proﬁle. Ultimately, high time resolution observations at 326.5 MHz will settle this issue.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	We have presented the ﬁrst detailed single pulse study of a millisecond pulsar in which suﬃcient time resolution was available to resolve single pulses as they would be seen in the pulsar vicinity. The similarity of the single pulse properties to those of normal slow pulsars is remarkable given the dramatically reduced magnetospheric volume and magnetic ﬁeld strength of PSR J0437−4715. Indeed, without being told the absolute sample rate, it is unlikely that one could distinguish between this being a millisecond or slow pulsar. To summarize, our observations of PSR J0437−4715 have: resolved individual pulses, and shown that they have a wide variety of morphologies, with multiple sub-pulse components not uncommon; shown that individual pulses are in general broadband; shown that individual pulses can have high linear polarization; provided no evidence for giant pulses as observed in the Crab pulsar and PSR B1937+21, nor for pulse nulling or drifting subpulse phenomena; found structure in the intensity ﬂuctuation spectrum; revealed a correlation of pulse peak with pulse width so that the average proﬁle formed from only the highest amplitude pulses is much narrower than the conventional average proﬁle; not shown any evidence for micro-structure or preferred time scales ≥80 ns; shown that the emission is consistent with an amplitude modulated noise model; provided no evidence to support the claim made by Ables et al. (1997) of the detection of coherent radiation patterns.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 1.— Average pulse proﬁle obtained by folding data from Observation 3 (see Table 1). There are 2048 bins across the pulse proﬁle; dispersion smearing in the ﬁnite-width simulated ﬁlters is 3.26 µs, just larger than one bin. The small “dip” seen near phase 0.9 is an instrumental eﬀect.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 2.— Grey scale plot showing single pulses of PSR J0437−4715 during a single ∼13-s observa- tion. Note that only the central ∼0.7 ms are shown; the average proﬁle formed by these pulses is shown at the top.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 3.— The average intensity ﬂuctuation spectrum calculated for 100,352 pulses at zero pulse phase (see text). The power density spectrum is normalized by the zero frequency power. The frequency axis is normalized by the pulsar frequency.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 4.— A random sampling of phase-aligned individual bright pulses from PSR J0437−4715, using a software ﬁlter bank. The time resolution in this plot is 2.56 µs, with 3.26 µs of DM smearing.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 5.— Histogram of peak pulse amplitudes in Jy from Observation 3 (see Table 1), dedispersed using a software ﬁlter bank. The time resolution was 2.56 µs. Rates for two peak values (1 per second and 1 per minute) are indicated by dashed lines. The low ﬂux-density cutoﬀ is due to an imposed threshold criterion.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 6.— The largest amplitude pulse in our data. The time resolution is intrinsically 10ns but is increased to 200 ns by DM uncertainties and non-ideal ﬁlter response. The dashed line represents the 98% conﬁdence level (see text).
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 7.— The largest amplitude pulse in our data, in each of eight frequency sub-bands. Here the data were coherently dedispersed, then subjected to an eight channel software ﬁlterbank at zero DM. The lowest frequency channel , the nyquist channel, appears to have a lower signal-to-noise ratio because its statistics are intrinsically diﬀerent from the other channels.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 8.— Histogram of pulse energies for Observation 3, including statistically signiﬁcant pulses only. The vertical dashed line is the mean pulse energy, which falls below the peak of the distribu- tion, since the mean pulse is below the noise level. The largest pulse energy we observed was 4.4 times the mean.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 9.— Pulse peak ﬂux versus pulse width. Only statistically signiﬁcant pulses were included in this plot, hence the lower cutoﬀ near the receiver noise temperature. Contours in the plot are 50, 100, 150 and 205 pulses.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 10.— Average proﬁle obtained by folding the 500 single pulses having the highest peak amplitudes in a ∼90 s data span. The width at half maximum here is 75 µs.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	Fig. 11.— Autocorrelation function for ∼14,000 consecutive coherently dedispersed pulses. The inset is the same data plotted on a smaller scale. Since no sharp changes in the slope or local maxima or minima can be seen, there is no evidence for any preferred time scales other then the subpulse width.
Processed_Radio_Pulse_Properties_of_the_Millisecond_Pulsar_P.txt	R. A. (J2000) . . . . . . . . . . . . . . . . . . . . . . Dec. (J2000) . . . . . . . . . . . . . . . . . . . . . . . µα cos δ (mas y−1) . . . . . . . . . . . . . . . . . . µδ (mas y−1) . . . . . . . . . . . . . . . . . . . . . . . . Annual parallax (mas) . . . . . . . . . . . . . . Period, P (ms) . . . . . . . . . . . . . . . . . . . . . Period derivative ˙P (10−20) . . . . . . . . . Epoch of period and position (MJD) Dispersion Measure (cm−3 pc) . . . . . . Binary period, Pb (d) . . . . . . . . . . . . . . . x = apsin i (lt-s) . . . . . . . . . . . . . . . . . . . . Eccentricity . . . . . . . . . . . . . . . . . . . . . . . . Long. of periastron, ω (◦) . . . . . . . . . . Epoch of periastron (MJD) . . . . . . . . . ˙x (10−12lt − s s−1) . . . . . . . . . . . . . . . . . Timing data span (MJD) . . . . . . . . . . .
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	We demonstrate the numerical superiority of the interaction-round-a-face (IRF) density-matrix renormalization-group (DMRG) method applied to SU (2) invariant quantum spin chains over the conventional DMRG. The ground state energy den- sities and the gap energies of both S = 1 and S = 2 spin chains can be calculated using the IRF-DMRG without extensive computations. We have also studied the eﬀect of tuning boundary interaction Jend at both ends of the chain from the IRF view point. It is clearly observed that the magnon distribution is uniform when the best Jend is chosen.
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	where J is a positive coupling constant and Si denotes spin operator at i-th site. After Haldane conjectured the existence of ﬁnite excitation gap for integer spin chains and the absence for half-integer ones [3], many eﬀorts have been performed to conﬁrm this conjecture. We focus on numerical analysis of the AFH model in the following.
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	A simple but concrete way to obtain the gap energy of the spin-S AFH chain in the thermodynamic limit is to numerically calculate the excitation energies for ﬁnite size systems, and to perform the ﬁnite size scaling. Conventionally Lanczos method is applied to the Hamiltonian (Eq. (1)) expressed as a large sparse matrix in order to obtain lower lying states and their energies. The major shortcoming of this way is that the Lanczos diagonalization requires huge numerical resources. This is because the matrix dimension of the N-site system is of the order of (2S + 1)N , that rapidly increases with both N and S. The largest system size N which is manageable with current computer is around 36 (for S = 1/2) even now.
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	The Density-matrix renormalization-group (DMRG) invented by White [1,2] is a novel and powerful computational method, which overcomes the limitation of the manageable system size in the Lanczos diagonalization. It enables us to obtain physical observables such as ground state energy, correlation functions, etc., of large scale systems very accurately without performing extensive com- putations. It is not overwhelming to say that the DMRG is one of the standard tools to study low dimensional quantum systems. DMRG is a real-space renor- malization method, which divides the whole system into the left and the right blocks, and reduces the degree of freedom of each block using the eigenvalues of the reduced density matrix (DM) as probabilistic measures to choose the relevant block-spin states. The number of the block spin state kept under the RG transformation is conventionally represented by the letter m. It is impor- tant to choose suﬃciently large m in order to keep numerical precision. The computational cost in DMRG is, roughly speaking, proportional to m3, which is not heavily dependent on size N.
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	When applying DMRG to the spin-S AFH chain speciﬁed by Eq. (1), the block spin states have total spin L up to nS, where n is a non-negative integer and depends on the number of spins in the block, we have to keep suﬃcient numbers of the block-spin state up to a certain total-spin L. This is problematic for large S chains, because the system has spin rotational symmetry; the eigenvalues of the DM are (2L + 1)-fold degenerated, since a block spin state with spin L belongs to 2L + 1 spin multiplet. Consequently the necessary block spin states (= m) increases rapidly with S.
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	resented as a quantum limit of the interaction round a face (IRF) model [7]. Sierra and Nishino [4] have modiﬁed DMRG algorithm for the S = 1/2 and 1 AFH models in the IRF representation (rep), where their numerical proce- dure is called ‘IRF-DMRG’. Wada [6] has further developed IRF-DMRG for the cases S = 1 and S = 2. It is easily understood that the the computational cost of IRF-DMRG is much lower than that of the conventional DMRG, since the dimension of the associated Hamiltonian matrix in IRF rep is much smaller than that in conventional (= vertex) rep.
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	Now we demonstrate the numerical eﬃciency of IRF-DMRG when it is applied to AFH model for higher spin (S ≥ 1) chains. In order to treat large S systems, Wada [6] has developed a way of obtaining the IRF expression for the Heisenberg interaction for S ≥ 1. In the IRF formulation, the spin excited states can be targeted by putting an additional auxiliary (or ghost) spin to the spin chain; more details are shown in Ref. [6].
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	Before showing the calculated results we stress that all the numerical data are obtained using personal computer with the moderate power (SPECfp95 ∼ 25). This is because the Hilbert space dimension in the IRF rep is much smaller than that in the conventional Sz basis rep. For example, in the case of spin-2 spin chain, to keep mIRF ∼ 90 states in the IRF rep is equivalent to keep msz ∼ 450 in the conventional rep. We keep 100 states at most, in the IRF rep. The ground state energy density e0 and the estimated gap energy ∆ for S = 1 are e0 = −1.401484039J and ∆ = 0.4104J, and those for S = 2 are e0 = −4.76124816J and ∆ = 0.088J [6].
Processed_Interaction-round-a-face_density-matrix_renormaliz.txt	Having described the computational beneﬁts of IRF formulation, now let us observe the boundary eﬀect from the IRF view point. Since DMRG (including the IRF-DMRG) chieﬂy treats systems with open boundary, it is important to decrease the boundary eﬀect and obtain the properties of bulk state.
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	Abstract. Recently, there appears lots of papers on the possibility of light Dark Matter (DM) in MeV and sub-GeV scale. Until now, only INTEGRAL and COMPTEL provided experimental data of 511keV of galactic center, and two spectra of Galactic Diffuse MeV gammas (GDMG)、 and COMPTEL provided the Cosmic Background MeV gammas (CBMG) for wide sky for indirect detection of light DMs. However except 511keV, those spectra for diffuse gammas included large statistical and systematic errors in spite of 10 years observation, since both two instruments suffered from severe background radiation in space. In 2018 April, we (SMILE- project in Comic-ray Group of Kyoto University) have observed MeV gamma rays for whole southern sky by Electron Tracking Compton Camera (ETCC) using JAXA balloon at Australia during one-day. (SMILE2+ Project) By measuring all parameters of Compton scattering in every gamma, ETCC has achieved for the first time to obtain the complete direction of MeV gammas as same as optical telescopes, and also to distinguish signal gammas from huge background gammas in space clearly. In this observation, ETCC with a large Field of View of 3sr observed MeV gammas from 3/5 of all sky including galactic centre, a half disk, crab, and most of CBMG By reconstructing the Compton process, we successfully obtained pure comic gammas by reducing background by more 2 orders, which is clearly certificated by the clear enhancement of detected gamma flux with ~30% during galactic center passing through the Field of View, which is consistent with the ratio of CBMG and GDMG. Now 511keV gammas GDMG are preliminarily detected with ~5 and >10 respectively around Galactic Centre. Also we have obtained near 105events of CBMG in with quite low background of only a few 10% in total CBMG events. Thus we obtained good data for both with high statistics and very low systematics even one day observation.
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	the direction of gamma completely as a point but a circle including its true direction as shown in Figures 1a and b Thus, CC cannot resolve the equation of the process completely, and it hardly reduce the backgrounds [4].
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	SMILE-2+ (SM2+) depicted in Figure.1a was improved from SM-2 for an enough sensitivity estimated based on this PSF for 511keV line gammas around Galactic center [7]. An effective area of SM2+ was increased 2 times at 511keV and >10 times above 1 MeV by pressuring Argon gas at 2atm and the use of a double thick GSO in the bottom Pixel Scintillation Arrays (PSA). Also since a fully contained electron in TPC restrict the energy range of gammas less 1MeV(low energy events), all PSAs were set in the TPC gas vessel to catch higher electrons energy Compton penetrating from the gas TPC as shown in Figures 1a and 1c, which enables us to detect 1~7MeV gammas (high energy events).
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	Figure2 (left) measured event topology in TPC for Compton scattering, pair creation, cosmic-ray and shower (right) the variation of the spectrum by applying event reduction cuts, and finally background were reduced by 2 orders of magnitude.
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	whole southern sky by the JAXA balloon launching at Alice Springs of Australia in 2018 April 7~8. The balloon was successfully floated at the altitude of 37~39km during 30hours, in which we observed the southern sky including Galactic Center a half of Galactic disk, Crab, Cen-A and Sun as planned. Although analysis is under way, we successfully obtained pure comic gammas of 105events by reducing background by more 2 orders, as shown in Figure.2 right. Figure 2 left indicates the ability of the clear event topology measured by gas TPC for the identification of Compton Scattering, and Figure.2 right shows the variation of the spectra of events passing through the noise reduction cuts. The amount of final events is consistent with that estimated from SM1 results, and the dependence of the number of gammas on the altitude matches with the previous data well. Those strongly supports the signal to background ratio is larger than 1, which is also clearly certificated by the obvious enhancements of detected gamma flux with ~30% during Galactic Center passing through the Field of View (FOV), for both low and high energy events as shown in Figure.1c. This ratio is consistent with that previously reported data, and as a result, the final data contains only a few 10% of background in the final event data. Thus, we have achieved the noise-free observation of MeV gamma-rays for the first time in the world. Until now, all observations of MEV gammas done before SM2+ includes about 2 order higher background than the signal.
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	Figure3: Light curve of final event after applying all cuts for low energy (upper) and high energy events (lower) within the zenith of 60o.
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	Thus, we have achieved the proper imaging spectroscopic observation of MeV gammas as same as that in X-ray and GeV gamma telescopes, and this success is certainly a keystone of the beginning of a new era for MeV gamma-ray astronomy.
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	References [1] Strong A.W. et al. 1999, Astrophys. Lett. Commun., 39, 209 [2]Bouchet L. et al., 2008 ApJ 679, 1315 [3] Schönfelder V. et al., A&ASS Vol. 143, 145-179 (2000) [4]Tanimori T. et al. 2017 Scientific Reports 7, 41511 [5]Takada A. et al. 2011, ApJ. 733, 13 [6]Tanimori T. et al. 2015, ApJ 810 [7] Takada A. et al. 2017, Proc. SPIE 9905, Space Teles. And Instrum., 99052M [8] Strong A.W. et al. 2000 ApJ 537, 763 [9] Porter T.A., Jóhannesson G., and Moskalenko I.V., 2017, ApJ, 846:67 [10] Strong, A. W. 2010 Proc. ICATPP Conference on Cosmic Rays for Particle and Astroparticle Physics, Villa Olmo, Oct 2010; page 473 ...
Processed_MeV_Gamma-ray_imaging_spectroscopic_observation_fo.txt	Figure6: Expected gamma-ray intensity maps of astronomical MeV (let) and GeV (center) [9]. And gammas from DM (right) used by Fermi group.
Processed_Recognizing_a_Spatial_Extreme_dependence_structure.txt	CNN model to predict extremes of climate events, such as tropical cyclones, atmospheric rivers and weather fronts. Lin et al.
Processed_Recognizing_a_Spatial_Extreme_dependence_structure.txt	Range Weather Forecasts ECMWF. The second case study is the rainfall amount recorded in the east coast of Australia.
Processed_Recognizing_a_Spatial_Extreme_dependence_structure.txt	Let us give a survey on spatial extreme models ad tail dependence functions, see Coles et al. (1999) for more details.
Processed_Recognizing_a_Spatial_Extreme_dependence_structure.txt	where 𝐹 is the marginal distribution function of 𝑋. If 𝜒(ℎ) = 0, the pair (𝑋(𝑠 + ℎ), 𝑋(𝑠)) is asymptotically independent (AI).

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.