- Results pages
- -Creation algorithm
- -Caveats and checks
- When are products built?
- Usage policy
- Change log
The Burst Analyser presents BAT and XRT flux light curves of GRBs observed with Swift. Spectral evolution is accounted for in the conversion from count rates to flux, making these light curves better probes of theory than count-rate curves. Where both BAT and XRT data are available for a burst, a joint BAT-XRT plot is also produced. Count-rate light curves are not presented here, since they are already available elsewhere for BAT and XRT. For BAT, a fluxed light curve with no spectral evolution is also created; for XRT this is already available from the XRT light curve repository.
The light curves are presented in three ways: 0.3—10 keV flux, 15—50 keV flux and the flux density at 10 keV. All of these are in units of unabsorbed flux. Since the XRT has a bandpass of 0.3—10 keV and BAT covers 15—150 keV most of the light curves involve extrapolation of the flux assuming a given spectral shape. Details are given in the algorithm section and some warnings are given in the caveats. It is strongly recommended that users read these sections before using the Burst Analyser for the first time.
To view the light curves first choose a GRB from the index page. Once you have chosen a burst you will be presented with the results page. Each results page contains a Data Overview section followed by up to four light curve sections: BAT-XRT combined data, BAT data, XRT data and BAT data where spectral evolution has not been considered.
An example Data Overview section is below. This section contains some general information applicable to all of the data on the page, and some useful links. Below the title of the burst the most recent ObsID used in creating the plots is given, and the version of that ObsID. This can be compared to the Quick-look site to determine whether the page has been automatically updated with the most recent data delivery yet. After links to a few related pages and links to the individual light-curve section are given links to a few useful files: BAT/XRT spectral evolution data, and a tar archive containing all of the files for this GRB.
The BAT and XRT spectral evolution data files are ASCII files (formatted for use with QDP) giving the hardness ratio light curve of the GRB in BAT or XRT, and this hardness ratio expressed as a counts-to-flux conversion factor or Photon index. The files contain 18 columns, six properties which each have a value, positive error and negative error. These properties are: time, hardness ratio, 0.3—10 keV flux conversion factor, 15—50 keV flux conversion factor, photon index (Γ), and flux density at 10 keV conversion factor. The All files for this object link supplied a tar file, whose contents are detailed on the TAR file page.
Following these links there may appear a warning if the BAT spectrum required a cut-off power law (see caveats for details). Then the absorption model used in the flux calculations is also given, and where it was determined from.
The light curve section begins with two links. Reliability checks summarises the sanity checks users should apply to the light curves. Show hardness ratio opens a new panel showing the measured hardness ratio of the GRB the second shows the hardness ratio measured by the two instruments [for BAT this is the (25—50 keV)/(15—25 keV) ratio; for XRT (1.5—10 keV)/(0.3-1.5 keV)]. The hardness ratio can be shown either as the measured hardness ratio, or as the hardness ratio translated into spectral photon index or counts-to-flux conversion factor (see creation algorithm).
The links are followed by up to three light curves: by default only the flux density at 10 keV is shown, however the options panel allows users to also show the 0.3—10 keV and 15—50 keV light curves. Each light curve is preceded by a link to rescale the plot, and each light curve section contains a light curve options panel. These are discussed next.
For each of the four light curve sections a single light curve will be shown by default: flux density at 10 keV, with a subplot showing the evolution of Γ (the spectral photon index). Users can change the subplot, can choose whether the 0.3—10 keV and 15—50 keV light curves are shown as well (or instead) and, for BAT light curves, can choose from a range of available binning criteria.
This is done using the selection panel which appears to the right of the first light curve in the section. An example panel is shown to the left, and contains up to three sections, described below.
For full details of how the light curves are created, users are directed to Evans et al. (2010, A&A, submitted). A summary is given here. The process of creating the fluxed light curves can be split into three phases: building the count-rate light curves and hardness ratios; determining the conversion factors; creating the fluxed light curves.
The XRT count-rate light curves and hardness ratios are taken directly from the XRT light curve repository. See Evans et al. (2007, 2009) for details. The hardness ratio is rebinned to contain at least 20 photons per band, per bin. This helps to avoid low significance fluctuations in hardness from creating artificial features in the light curves.
For BAT, first the
script (part of the Swift software) is
executed. The value of T90 and the time range this covers is noted. A 4-ms-binned light
curve is compiled using the
batbinevt tool, with the detector mask and total event list
batgrbproduct. Signal-to-noise (SNR) binned light curves are then built from
this 4-ms light curve for SNR 4,5,6 and 7 using a custom script which bins the
highest-significance bins first so as to maximise both SNR and time resolution. No bin is allowed
to exceed 40-s in duration. If a bin reaches 40-s and is still not at the SNR threshold, it will
be written regardless. If it has SNR<2 it will be preceeded with an !. This means that, while
the bin is still present in the data file which can be viewed and downloaded, it is not shown in
the online plots. A second set of SNR-binned light curves are created where only data from T>0
are included. We create the second set for use with the BAT-XRT combined data (which is plotted in
log-log space and so cannot include times at or before the trigger) because this ensures that the binning
is optimised for the data which are displayed, which would not be true if we simply ignore the bins at T<0
from the full light curve.
In addition to the SNR-binned light curves, 15-150 keV light curves with fixed-duration bins
are created using
batbinevt, for bin sizes of 4 ms, 64 ms, 1 s and 10 s†. These light curves do not cover the
entire event list (since this would produce many bins composed entirely of noise), but the time range given by:
T90,start−1.05T90 ≤ t ≤ T90,end+1.05T90
mark the time region over which T90 was accumulated, according to
A hardness ratio time series is created in a similar manner to the SNR-binned light curves. 15—25 keV
and 25—50 keV 4-ms binned light curves are created from the total BAT event list using
batbinevt. These are then passed to the SNR-binning code which requires SNR≥5 in
each band in order to create a bin (the maximum bin size is again 40 s.) If fewer than 3 bins with
SNR≥3 were created the target SNR is reduced to 4 and the hardness ratio is rebuilt; and again
with a target SNR of 3 if necessary. Even if this does not produce more than 2 "good"
bins the SNR is not further reduced.
† Not all of these bin durations are used for every GRB. Bin durations which are longer than T90 are not created. Similarly a 4-ms binned light curve is not created if T90 is longer than 10 s.
In an ideal world, to track spectral evolution and thus create a time-dependent counts-to-flux
conversion factor, one would create and fit a spectrum for each light curve bin. In reality this
is impossible since there are insufficient photons per bin to create a usable spectrum. Instead we
assume a spectral shape and use the hardness ratio to track the evolution of that spectrum. The
spectral shape used is either a power-law or a cut-off power-law. The choice of which to use is
made by fitting the BAT T90 spectrum (from
batgrbproduct). If the cut-off
power-law model improved χ2 by at least 9 (a 3-σ improvement) the cut-off
power law is used (for BAT and XRT), otherwise a power-law is adopted. Note that, using a single hardness ratio we
cannot track the evolution of both power-law index and cut-off energy, therefore if a cut-off
model is required the cut-off energy is frozen at the value obtained from the fit to the
T90 spectrum. See caveat 2 for more details.
As well as a spectral shape, an absorption model is needed‡. If there are
XRT data for the GRB the absorption is taken from this. As Butler
& Kocevski (2007) demonstrated that absorption measures
calculated from early-time XRT spectra can be incorrect, if there are
at least 200 X-ray photons detected at t>4000 s, an X-ray spectrum
is constructed from data after this time, and the absorption is
determined from an absorbed power-law model which is automatically
fitted to this spectrum. Otherwise the absorption from the spectrum on
the XRT spectrum
repository is used. This absorption is made of two components, a
Galactic absorber and an intrinsic absorber. If the redshift of the
GRB is known, the intrinsic component is given this redshift. If no
XRT data exist for the burst, the Galactic column at the BAT position
is obtained from the
and the intrinsic component is given NH=1022
cm-2 at z=2.3; these are typical values for XRT spectra
et al. 2009). The absorption model used, and whence it was
determined, is shown on the results web
pages immediately before the light curve plots begin.
The power-law or cut-off power-law spectral model, absorption and instrumental response
matrices are then loaded into
xspec and the
photon index (Γ) is stepped from -1 to +5 in steps of 0.1 (note:
N(E)∝E-Γ) For each Γ value, the hardness ratio, count-rate, flux
(0.3—10 keV and 15—50 keV) and flux density§ are
calculated. This yields a lookup table which translates hardness ratio to Γ and to
counts-to-flux(-density) conversion factor. Using this table each point in the hardness ratio is
then converted into Γ and the necessary conversion factors, by interpolating this lookup
table. For BAT data there may be some times where the hardness ratio was negative (when the burst
is faint and no SNR≥3 hardness bin could be made). These cannot be converted to conversion
factors, so are skipped.
For the non-spectrally evolving BAT fluxed light curve the process is much simpler. The
best-fitting T90 spectrum is loaded into
xspec and the count-rate, fluxes
and flux density are calculated. These give the time-averaged counts-to-flux(-density) conversion
factors. Note that no XRT non-spectrally evolving light curve is created as part of this project
since this is already supplied by the XRT light curve
‡ The light curves are all given as intrinsic, i.e. unabsorbed, flux (or flux density).
§ The normalisation parameter of the power-law and
cut-off power-law models in
xspec is defined as the flux
density at 1 keV, in units of photons keV-1 cm-2
s-1, which is thus converted to Jy by multiplying by
0.000662, and can be extrapolated to the flux density at 10 keV
because the spectral model is known.
The final stage involves taking each bin in the count-rate light curves, interpolating the "hardness-ratio-as-conversion-factors" data produced in the previous step to determine the conversion factor at the centre of the bin, and then multiplying the count-rate by this value. The uncertainty in the conversion factor is not propagated into the final flux error bar (discussed below). Where there are gaps in the BAT "hardness-ratio-as-conversion-factors" data due to negative hardness ratios the conversion factor is interpolated across this gap. Due to the significant spectral evolution seen during the prompt emission of a GRB, we do not extrapolate the BAT "hardness-ratio-as-conversion-factors" to times before (after) the first (last) bins with a positive hardness ratio. However for XRT, because of the way the XRT light curves are created it is only ever the final bins which may not have corresponding hardness ratio information. Since XRT data do not generally show spectral evolution after the first few hundred seconds, the final measured hardness ratio value (and hence conversion factors) is applied to all light curve bins which occur after the hardness ratio finishes.
Although every effort has been made to ensure the reliability of these light curves, blind usage is not recommended. Users are strongly encouraged to read the note about error bars and to ask the five reliability questions.
As discussed in the algorithm section above, the uncertainty in counts-to-flux error is not propagated into the final fluxed light curve. This is because the effect is at least partially systematic and propagating this uncertainty has the effect of 'washing out' genuine light curve features. However, it is important to check the uncertainty in the conversion factor, especially where surprising results are seen. If the Conversion Factor or Γ subplots show large errors (see products pages for how to view these) the final flux value is also much less precise than suggested by its error bar.
Related to the above caveat: because negative hardness ratios are impossible in reality (they arise in the data due to Poisson noise) when converting the hardness ratio to Γ and counts-to-flux conversion factor the negative error on the hardness ratio is necessarily truncated to give HRmin=0. This means that the negative errors on Γ and the conversion factor are also truncated and thus unreliable. This can usually be readily determined by eye because the errors on these properties are highly asymmetric. Alternatively, users can check the hardness ratio: if the minimum value is zero, the uncertainty in Γ/conversion factor has been truncated.
In order to convert the count-rate light curves into flux we need some spectral information. As described above we do this using hardness ratios as we do not have enough data to create accurate spectra for each light curve bin. However, this approach implicitly assumes a spectral shape (either a power-law or cut-off power law). For the light curves where the flux is given within the band pass of the instrument (i.e. the BAT 15-50 keV and XRT 0.3-10 keV light curves), this assumption is usually supportable; however when extrapolating the flux outside the observed band, if the spectral shape is not that assumed, the results may be incorrect. This is why the default light curve on the web pages is the flux density at 10 keV; this value allows us to directly compare BAT and XRT data, but minimises the extrapolation.
The five questions below can be used to identify cases where the fluxed light curves may be unreliable. More details, and what to do in each case, follow the list.
A fundamental assumption used to create the fluxed light curves is that the spectral slope is the same in the BAT (15-50 keV) and XRT (0.3-10 keV) bandpasses at a given time. If this is not true, the extrapolated fluxes may be inaccurate. For example, consider the case where the true spectral shape is a broken power-law with a break between the XRT and BAT bands, and a steeper index above the break. The BAT count-rate will be converted to 0.3-10 keV flux using an erroneously soft spectrum and will thus overestimate the flux. Similarly the XRT count-rate will be converted to 15-50 keV count-rate using a spectral index which is too hard and again will overestimate the flux.
This circumstance is readily identified by looking at the Γ values which accompany the light curves. In the example to the right, GRB 060526, the BAT Γ values at 200-400 s are systematically higher (softer) than the XRT values. The 0.3-10 keV light curve and 15-50 keV light curve show the effects discussed above. However as the break must lie somewhere between the bands, the effect of this inaccuracy on the flux density at 10 keV is minimal and indeed the BAT and XRT flux value here agree well.
In summary: If the Γ values are discontinuous between BAT and XRT the flux value outside of the instruments' bandpasses are unreliable. The flux density at 10 keV is the most immune to this problem.
The BAT spectrum cannot always be adequately described as a power-law. In the event that a cut-off power-law (CPL) fit improves χ2 by at least 9 (i.e. the CPL fit offers a 3-σ improvement over the power-law fit) this model will be used to generate the counts-to-flux conversion factors (for both BAT and XRT). However, to track the evolution of both the photon index and the cut-off energy is not possible; this requires two hardness ratios and there is usually insufficient signal-to-noise to generate such data. Further, if the cut-off energy is towards the edge or outside of the bandpass covered by BAT or XRT a strong degeneracy between this energy and the photon index arises. Consequently, when a CPL model is required by the BAT spectrum, the cut-off energy is frozen at the value determined from the T90 spectrum. In reality we expect the cut-off energy to be evolving with time, thus fixing the parameter introduces an extra source of uncertainty in the counts-to-flux conversion which is not reflected in the errors. As with discontinuous Γ values, this effect is likely to be strongest when considering the flux extrapolated outside of the instruments' own bandpasses. Although experience shows that the effect of this is minimal, users should be aware of this, and be cautious ascribing weight to outlying data points where a CPL spectrum has been used. If a CPL spectrum was used, this is reported prominently on the results page, with a sentence such as the following given before any of the light curves:
The BAT spectrum was fitted with a cut-off power-law. The cutoff energy was 93.9 keV and was not allowed to evolve. See caveats for details.
In summary: If a cut-off power-law spectral fit was required by BAT, the uncertainty in counts-to-flux conversion may be underestimated, especially when extrapolating outside the instruments' bandpasses.
As described above, the Γ values given for each light-curve bin are not determined from spectral fits, but from the hardness ratio. Extreme values (outside the range ~0-4) are not generally seen in fitted spectra (see The Swift Data Table and The XRT catalogue paper) and thus may indicate low signal hardness ratios. These should be accompanied by large error bars, and the associated warnings should be taken into account. Note further that the look-up table from hardness ratio to Γ/conversion factor is only determined for -1 ≤ Γ ≤ 5. Thus Γ values outside this range have been extrapolated beyond the range over which the relationship is calibrated. Again, this should be reflected in large uncertainties in Γ. In general, where the Γ values are outside the 0—4 range (and certainly outside the -1—5 range) the flux points should be viewed with caution.
In summary: Any flux points with corresponding Γ values outside the range 0 ≤ Γ ≤ 4 should be viewed with caution.
This topic is discussed in detail in the discussion of error bars.
By default, BAT points with 'bad' conversion factors are not shown in the online plots. A 'bad' conversion factor is defined as either being consistent with zero, or having an error larger than two orders of magnitude (positive or negative). Such points are still present in the data file; these points are indicated by a !# at the start of the line. Selecting the Show 'bad' points option causes these data points to be included in the plot.
In summary: Errors on Γ and conversion factor have not been propagated to the flux light curves. Where the hardness ratio is consistent with 0 the negative error on Γ and conversion factors are underestimates.
Spectral evolution is accounted for by tracking the hardness ratio evolution, interpolating between measured values to the times of light-curve bins. In fainter bursts there may be only one or two hardness ratio bins. In such cases we cannot track the evolution and the spectrum will tend towards the non-evolving flux spectrum. This can be seen from the Γ or conversion factor sub-plots, since a single hardness ratio point will result in each bin in the sub-plot being the same. This does not give an inaccuracy in the fluxed light curve (provided the errors on the conversion factor are considered), but it does mean that the effects of spectral evolution are not visible in the light curve but rather hidden in the uncertainty in the conversion factor.
In summary: If the Γ or conversion factors are not varying there was minimal hardness ratio information, so the effects of spectral evolution are not visible in the fluxed light curve.
As with all automated products supplied by the UKSSDC, these light curves are built and updated as soon as data become
BAT and XRT data do not necessarily arrive at the same time, the BAT and XRT portions of the Burst
Analyser products can be built independently. This may mean that initially for a newly-detected
burst there may be only a BAT or XRT light curve. The building of the BAT products occurs when the
BAT data first arrive, or the
ONTIME of the BAT data on the quick-look site increases compared to that
available when the light curves were last built. For the XRT data, when the light curve repository builds or updates the
count-rate light curve it sets a flag which causes the Burst Analyser to build or update the flux
◊ Building or updating of light curves is triggered by a
cron job which
runs every ten minutes.
There are no restrictions on the use of these products, however we do request acknowledgement. Where data from the Burst Analyser are used, please cite the paper Evans et al. (2010, A&A, submitted). If you use the count-rate light curves to which the Burst Analyser provides links, please note they they also require acknowledgement: for BAT data please acknowledge Taka Sakamoto and Scott D. Barthelmy, for XRT data please cite Evans et al. (2009, MNRAS, 397, 1177) and Evans et al. (2007, A&A, 469, 379).
Also, wherever these data are used please include the following text in the acknowledgements: This work made use of data supplied by the UK Swift Science Data Centre at the University of Leicester.
Any changes made to the software after the paper is published are shown here.