Fermi Gamma-ray Space Telescope

Caveats About Analyzing LAT Pass 7 Reprocessed (P7REP) Data

A detailed discussion of the performance of the Large Area Telescope Pass 7 data is available on arXiv. The same methods were used to characterize the performance of the LAT with Pass 7 Reprocessed (P7REP) data.

These caveats are relevant for the P7REP version of the photon dataset. They are an updated version of previous sets of caveats for analysis of Pass7 (P7_V6) and Pass 6 (P6_V3 and P6_V11) photon selections and Instrument Response Functions (IRFs).

Event Selection

Prescriptions for event selection for analysis of Fermi-LAT data are provided in the data preparation section of the Cicerone.

The Fermi-LAT performance associated with the released P7REP IRFs is documented on the LAT Performance Page. These IRFs were derived using Monte Carlo generated samples of photons between 18 MeV and 1.78 TeV, with overlays of periodic trigger events to realistically simulate the effects of residual signals from off-time or non-triggering particles. However, the validity of these IRFs when performing analysis of LAT data, in terms of usable event classes and energy range, is determined by the caveats discussed below.

  • In doing cross-comparisons, the Fermi-LAT team found a small bias in the unbinned likelihood calculations compared to the binned analysis. Simulations have confirmed that the Test Statistic and flux values are overestimated by the unbinned analysis. The magnitude of the bias depends on the local conditions for the source, with sources having a low source-to-background ratio (e.g. in the Galactic plane) showing a larger effect, especially over long integrations. Although this bias has little effect on most results, a new version of the Science Tools has been developed to remove the inconsistency. This issue is solved in current Science Tools version (v9r31p1 or later).
  • The "P7REP_TRANSIENT" event class (bit 0 of EVENT_CLASS, evclass=0 in gtselect) has large non-photon backgrounds and is only appropriate for the study of short transient events with a duration of less than 200s. Beyond these timescales the level on non-photon backgrounds can compromise the quality of the data analysis.
  • We recommend using the "P7REP_SOURCE" (bit 2 of EVENT_CLASS, evclass=2 in gtselect) photons for all point source analyses as well as for the analysis of bright diffuse sources. When integrating over large regions of the sky, and particularly at high energies, the residual non-photon contamination in this event class may result in spurious spectral features, and so we recommend using only the "P7REP_CLEAN" (bit 3 of EVENT_CLASS, evclass=3 in gtselect) photons for studies of diffuse emission at these high energies.
  • Data below 100 MeV present challenges in spectral analysis because of the rapid change of effective area with energy, the residual uncertainty in the instrument response and the energy dispersion of the instrument, which is not treated by the Science Tools. Therefore, when using the "P7SOURCE" and "P7CLEAN" selections we recommend starting at 100 MeV for the most robust results. Inclusion of data outside the recommended energy range can result in the creation of spurious spectral features under the current IRFs.

Systematic effects and uncertainties

  • The systematic uncertainties in the effective area are evaluated by comparing the efficiencies of analysis cuts for data and simulation of observations of Vela and the limb of the Earth, among other consistency checks. (See Sec.5.3.1 of Ackermann et al. 2012, ApJS 203, 4 for details.) These studies suggest a 10% systematic uncertainty below 100 MeV, decreasing linearly in Log(E) to 5% in the range between 316 MeV and 10 GeV and increasing linearly in Log(E) up to 15% at 1 TeV.
    The Vela pulsar does not provide sufficient statistics to measure the systematic error above 10 GeV. However, studies using the photons from the limb of the Earth confirm a 10% systematic error from 10 GeV up to 100 GeV. Furthermore, the limited statistics we do have suggest that the 10% error extrapolates unchanged to hundreds of GeV.
  • We tested the self-consistency of the P7REP_SOURCE_V15 IRFs to estimate the accuracy of the Monte Carlo simulations by comparing the fraction of events in various subsamples for data and simulations. We find that the P7REP_SOURCE_V15 IRFs are self-consistent to better than 2% from 100 MeV to 3 GeV, to better than 5% up to 100 GeV, and to better than 10% up to 1 TeV. We note, however, that these tests do not rule out any uncorrelated global-biases that might have been introduced in earlier event trigger, reconstruction or selection algorithm stages.
  • The above uncertainties apply to the absolute effective area and to artificially induced spectral features with a width of a half decade or more in energy. A dedicated search for spectral features with a relative width of 15% (about 1/16 of a decade, matching the instrument energy resolution) found that the point to point correlation in the effective area is very large, so the systematic uncertainty is closer to 2% for for spectral features with relative widths of 15% and rises as the width of the feature increases.
  • While the current Instrument Response Functions describe the average effect on the acceptance due to the pile-ups and chance coincidences in the detector, residual effects remain on short time scales. More details can be found in Post-launch performance of the Fermi Large Area Telescope.
    When the spacecraft is in orbital regions with high particle backgrounds, the increased levels of activity in the LAT reduce the photon selection efficiency, especially in the lower parts of the LAT energy range.
    Variability studies and studies involving data taken over time periods of less than one orbit should directly account for these effects. Studies comparing sources in different regions of the sky should allow for slight differences in mean efficiency during the time periods when the various sources were visible.
    These effects have been corrected to first order in the Pass 7 and later IRFs. We estimate the residual effect to be on the order of 1-2% of the effective area.
  • Along similar lines, the effective area tables are averaged over the azimuthal angle in the LAT instrument reference frame. For time scales of more than a few days this averaging is accurate to better than 1-2%. However, for shorter timescales it is important to use the optional phi-dependent extensions to the livetime when calculating exposures.
  • The updated instrument calibrations used to produce the P7REP data release improve the LAT PSF above 3 GeV, resulting in a better overall agreement between the Monte Carlo (MC) PSF model and the angular distributions of gamma-ray point sources. However the MC PSF was still found to slightly underestimate the PSF width above 3 GeV. The LAT team has derived a new on-orbit PSF for P7REP data (included in the P7REP_V15 IRFs) by rescaling the MC PSF model to match the angular distribution of Vela below 10 GeV and the stacked distribution of bright, high-latitude blazars above 10 GeV. Whereas the on-orbit PSF model delivered with the Pass7 IRFs did not contain a description of the PSF dependence on inclination angle (theta), the P7REP_V15 PSF model preserves the theta-dependence of the MC PSF and thus should provide a better representation of the PSF on short timescales.
  • Because the LAT field of view is very large, it is likely that somewhat different biases are relevant in different parts of the field of view. When coupled with the sky survey observational strategy of Fermi, this can cause apparent variability when studying particular sources. The magnitude of such effects depends on the timescale of the study. For example, the 2FGL catalog analysis found the pulsar Geminga to be variable at the 2% level on one-month time scales; an effect which is likely instrumental. Furthermore, dedicated variability studies have seen evidence of 10% variability on the 55-day time scale corresponding to the precession of the LAT orbital plane relative to the Earth's rotational axis. When doing precision variability and time-domain studies, the LAT team recommends using sources at similar declinations to control for these instrument-induced effects.
  • The absolute energy scale for the LAT is determined with an uncertainty of +2% -5% For additional details, see In-flight measurement of the absolute energy scale of the Fermi Large Area Telescope (Ackermann, M. et al. 2011 APh, 35, 364).
  • The LIVETIME column in the photon FITS file should be ignored. It was intended to contain the livetime accumulated since the start of the current run (data-taking interval, typically ~1 orbit in duration), but this not technically feasible. The livetime value resets frequently and this column should be ignored. The LIVETIME accumulations in the spacecraft data files are correct and complete.

Diffuse Models

  • Please see the diffuse model page to access the currently-recommended model of Galactic diffuse emission and the corresponding spectrum of the isotropic emission including residual background from cosmic rays misclassified as gamma rays. Documentation is provided here for the most recent version.
  • The model of Galactic diffuse gamma-ray emission is defined between 50 MeV and 600 GeV. To derive the isotropic spectrum at higher energies, we extrapolated the Galactic diffuse emission model. Analyses at energies >100 GeV are likely to be limited primarily by photon statistics, but the accuracy of the modeling of diffuse gamma-ray emission at those high energies is also reduced.
  • All the released diffuse models were derived for point sources and compact extended sources studies only, and are not suited for studies of extended sources and/or large-scale diffuse emissions. There is no standard background model for extended diffuse analyses, nor a generic treatment of extended sources and diffuse extended emissions. The approach for such studies depends on the aim and the objective of the study and should be evaluated, treated and tested on a case by case basis.
  • Each diffuse model should be used with the corresponding Event Selection and IRF. For example using P6 background models with P7 dataset or P7 background models with P6 dataset is strongly discouraged, because each model was fitted to the specific Event Selection and IRF. Ideally the resulting diffuse emission model is independent of IRFs used to derive it, but in practice the model will have built into it the low-level systematics associated with our imperfect knowledge of the IRFs and the approximations that we make regarding neglecting the finite energy resolution. However, as the diffuse emission is bright over most regions of the sky, these small imperfections may result in large systematic effects for sources. For this reason we do not recommend using diffuse emission models for analyses with event classes for which their validity has not been tested and established.
    Furthermore, the templates for the isotropic spectrum include misclassified cosmic rays. The probability of such misclassifications differs between event classes as well as between front-converting and back-converting events. Therefore, it is important to use the isotropic template corresponding to the data selection. This applies both in terms of the event class and conversion class, i.e., if P7SOURCE front-converting events are selected, then the corresponding isotropic template for P7SOURCE front-converting events should be used. If both front-converting and back-converting events are selected then the combined template for front+back events is appropriate.

GRB analysis

  • It is recommended to use the "P7REP_TRANSIENT" class to maximize the statistics for spectral analysis of GRB prompt emission since it is more or less background free due to the short duration of the GRB prompt emission. When using the "P7REP_TRANSIENT" class, it is recommended to not use data below 100 MeV for spectral analysis since the instrument response is not validated in the lower energy band.
  • It is recommended to use the "P7REP_SOURCE" class to search for GRB afterglows since the "P7REP_TRANSIENT" class only adds lower energy photons which many not help given the increased backgrounds.
  • It is recommended to use data above 500 MeV to localize the GRB since low energy events suffers from the so called "Fish eye" effect where the reconstructed direction points toward smaller inclination angles due to selection bias. It is also recommended to use the "P7REP_SOURCE" class for GRB localization since the "P7REP_TRANSIENT" class does not improve statistics at the high energy band.

LAT Low Energy (LLE) data product

The LAT Low Energy analysis (LLE) is a new type of analysis developed by the Fermi-LAT and Fermi-GBM teams for increasing the effective area of the Large Area Telescope at low energy, and it is suitable for studying transient phenomena, such as Gamma-Ray Bursts and Solar Flares. The LLE analysis filters event data with a very loose event selection, requiring only minimal information, such as the existence of a reconstructed direction. The reconstructed direction is used to select events that are compatible with a certain location in the sky, using information on the Point Spread Function to increase the signal-to-noise ratio. The instrument response is calculated with a dedicated Monte Carlo simulation that uses satellite pointing information and the celestial location of the source. The released FITS files contain the data (in various formats) and the instrument response.

The LLE data is a data product that differs from the standard LAT data products, and goes with some caveats and considerations:

  • The LLE data selection and response depend on the input localization of the GRB. The default localization is taken from the Fermi GBM Trigger Catalog. Although the procedure to obtain LLE data and response is fully automatic, the update of the GBM trigger catalog is not, and this might introduce a latency in the delivery of the data products with the optimized location.
  • The Monte Carlo used to generate the response covers an energy range between 10 MeV and 100 GeV. At low energy (<100 MeV) the effect of the energy dispersion can be significant, and, with the current analysis, we discourage any spectral analysis below 30 MeV.
  • Above few hundreds MeV (depending on the off axis angle of the event) the signal-to-noise ratio for standard data (TRANSIENT, SOURCE and cleaner event classes) is higher than for LLE data. In case of bright events with emission above few hundred MeV we suggest using standard LAT event data at high energy and LLE data at low energy.
  • The background in LLE is mainly driven by residual particle events and soft gamma-ray events in particular coming from the bright limb of the Earth. During Autonomous Repoint Requests (ARR), a significant fraction of the Earth limb enters the LAT field of view and an increase of the event rate is clearly visible in LLE data. We urge users to exercise care in the treatment of the Earth limb.
  • Data and response are, by construction, related to each other. In case of updates, we recommend updating both the response and the data, making sure that the version number is the same.

LAT Monitored Source List

  • These flux estimates do not include systematic uncertainties and reflect the recommended analysis and calibrations at the time the data were acquired. Therefore, use of these data as absolute flux measurements for constraining models or for comparison to other data is strongly discouraged. In addition to overall normalization uncertainties, source fluxes may have variations of up to 10%, particularly earlier in the mission, because of uncorrected dependencies of the gamma-ray detection efficiency on variations of the particle background over the orbit. See the caveats on variability studies mentioned above. Please note that these results have been produced using a variety of instrument response functions and calibrations.

Periodicity artifacts in LAT time-series analysis

Last updated by: Elizabeth Ferrara 6/5/2014