PSF Issues and Caveats
With the combination of ChaRT and MARX, users may now easily perform detailed simulations of the Chandra PSF. Both of these pieces of software have limitations of which users should be aware; those limitations are described here.
The Chandra Instruments and Calibration page contains the most recent updates about the quality of the mirror (HRMA) calibrations.
As part of efforts to push the spatial resolution of Chandra to the sub-ACIS-pixel regime, we have identified a feature in the Chandra/HRC point spread function within the central arcsecond which may affect high fidelity deconvolutions. There is evidence that the feature is also present in Chandra/ACIS data. The problem does not affect images on scales larger than one arcsecond. Complete details are in the Probing higher resolution: an asymmetry in the Chandra PSF caveat.
The raytraces performed by SAOTrace are designed to precisely model the optics and their support structures, and are based upon mirror metrology, as-built and as-designed drawings of the support structure, and pre-flight tests of the HRMA. The raytraces are deterministic, rather than statistical, in that they follow individual photons through the optical system (see the Chandra optics overview for a description).
The HRMA User's Guide contains a more detailed discussion of HRMA PSF characteristics, including some SAOTrace issues.
The low-energy (< 2keV) PSF core and inner wing region (out to 10 arcseconds) match well with observations (see ARLac memo, Jerius 2002).
We believe that the high energy core should be well modeled.
High energy (greater than 2keV) comparisons of the PSF with in flight data are not yet sufficiently mature enough to draw quantitative conclusions (see the PSF wings presentation), but the models seem to underpredict the flux in the wings (both close to the core and further out) at these energies.
The off-axis PSF is a complicated function of energy and source off-axis and azimuthal position. The gross (and not so gross) structure of the PSF is well modeled. There are differences in the details of the structures in the cores of the PSF, but unless one is doing incredibly detailed off-axis structural analysis (which it is not recommended to do at this point), this is not an important consideration.
The model seems to replicate the observed HRMA vignetting well (to better than 10%). This has been calibrated (in rather large 1keV bins) against observations, see HRMA calibration pages for details.
ChaRT now allows users to supply SAOTrace an aspect solution file to include the effects of dither in their simulation. Users may either select the file from the Chandra archive (by supply the OBS_ID) or may upload the if they have applied an astrometric shift using the wcs_update tool.
However, since wcs_update does not update the values for the quaternions, the astrometric shift is not included in the simulated PSF. Therefore the PSF may be offset from the observed data. Since the astrometric shifts are typically < 1 arcsec, this is generally not a problem for most users.
On 2016-06-10, ChaRT was patched to adjust the quaternion values in the uploaded aspect solution files to match the RA, Dec, & Roll values (which do include the astrometric shift). ChaRT generated ray files after this time will now properly include the astrometric shifts.
It is difficult to quantify errors in individual representations of the PSF, so the generated rays do not include error information. The user is directed to the calibration analyses to estimate their errors.
A single raytrace of the PSF samples only a portion of the possible optical paths in the HRMA, especially when run with the number of photons typical of most Chandra observations. Several realizations (or one with a larger number of photons) may be necessary in order to make a detailed comparison with observations.
Currently the user can either select ACIS-S or ACIS-I chips but not a combination ACIS-S and ACIS-I chips. So for example to simulate the PSF for a source on ACIS-3 when the aimpoint is on ACIS-7, the detector must be set to ACIS-I with a large SIM_Z offset back to ACIS-7 aimpoint.
Aspect reconstruction uncertainties
By default, MARX simulations should be run with the DitherModel parameter set to a value of "INTERNAL" to ensure that sub-pixel event location information is preserved in the final Sky X and Y values stored in the event file. This value is the default. For real Chandra data, the aspect reconstruction process introduces small (~0.35 arcsec for ACIS and 0.18 arcsec for HRC) positional errors in the derived sky positions for events. MARX includes the effects of these errors using the DitherBlur parameter.
The DitherBlur parameter is a statistical term which combines aspect reconstruction accuracy, pixelization by the detector, and pixel randomization ("anti-aliasing"). The default value for this parameter derives from comparison to flight data. For exact projection of a priori known photon positions to the sky without this blurring, the value of DitherBlur can be set to zero. See the MARX manual or the Using MARX to Create an Event File thread for further details.
MARX, specifically the marx2fits program, in version 5 has options to simulate the location of an event detected within an ACIS pixel. These options allow users to simulate the EDSER ACIS subpixel algorithm, the legacy RANDOMIZE setting that applied a uniform +/- 0.5 pixel randomization to an event location, as well as the EXACT algorithm that was used in MARX 4.x.x.
When MARX is used with dithered ChaRT/SAOSac rays, the AspectModel parameter must be set to FILE with the DitherFile set to the aspect solution file used to run the raytrace simulation; using the internal dither model will result in the source visualizing the Lissajous dither pattern on the sky.
The AspectBlur parameter was introduced, replacing MARX 4's DitherBlur parameter, to blur the PSF by the spacecraft's aspect reconstruction uncertainty, while pixel randomization and quantization effects are handled internally by MARX. Presently, there is a discrepency between the size of observed PSFs and simulated PSFs projected with MARX using the default AspectBlur value, and is being investigated by the MARX developers.
Users may experiment with the appropriate values of AspectBlur. We suggest AspectBlur=0.2 [arcsec] for ACIS-I, AspectBlur=0.25 [arcsec] for ACIS-S, and AspectBlur=0.07 [arcsec] for HRC as suitable for typical data, based on a limited set of simulations.
ACIS pileup simulation
If users choose, they may simulate the effects of photon pileup in the ACIS detectors by using the PILEUP tool included as part of the MARX package. This option is not run by default. This tool implements the pileup algorithm developed by John Davis (MIT). This same algorithm has been implemented into the ISIS, Sherpa, and XSPEC spectral fitting packages.
While this implementation of the pileup algorithm emulates most of the qualitative effects of ACIS photon pileup, users should keep in mind that we are still calibrating the procedure. The ACIS pileup model is statistical and is not an a priori photon-silicon interaction model which generates charge clouds and then PHAs per event "island." The model is valid on-axis for point sources for low to moderate pileup. While valid for qualitative predictions of the effects of pileup on the PSF, it has not been verified for image reconstruction. Detailed studies of the effects of pileup on the HRMA PSF including comparisons to actual on-orbit data are still underway. The model is very good for spectral modeling of light to moderately piled point sources. Users should interpret all ChaRT and MARX results including the effects of pileup cautiously.
Further reading about ACIS pileup modeling:
HRC Micro-Channel Plate (MCP) event position
For simulations utilizing the HRC-I or HRC-S, event locations are not calculated from simulated tap voltages, but assume low-level instrumental signatures have been removed (i.e., degapped) and converted to linear detector coordinates. In MARX, the positional uncertainty produced by the HRC detection process is modeled using a simple Gaussian blurring factor. There may be differences in detail from the observed PSF due to uncalibrated non-linearities in the detector.
HRC-I PSF blur factor
The latest HRC-I calibration observation of Ar Lac (~18ks from December 2010; ObsID 10182) was used to determine the blurring factor that needs to be applied by psf_project_ray to the simulated HRMA PSFs to match the ECF of the observation. The recent discovery of the PSF artifact ~0.8" complicates this by adding additional effect on the PSFs.
We find that the resulting best blur is between 0.18" to 0.20" for the inner part of the PSF core, <~3 HRC-I pixels (< 0.4"), with ~50% ECF, at ~0.4". From 3 to 4 HRC-I pixels the PSF is slightly larger (blur of 0.22"), with 70% ECF at ~0.5". Beyond that, from 4 to 10 HRC-I pixels (~ 0.5" to ~1.3"), the sigma of the gaussian blur increases from 0.22" to 0.25". This larger blur may include the effects of the PSF HRMA model discrepancy and of the PSF asymmetry feature-artifact contributing extra counts, especially from ~0.7" to 1".
We conclude that for most of the HRC-I data, a blur sigma of 0.2" is reasonable, given that the number of counts will be much less than in the case of the Ar Lac observation (~100,000) counts, and the variation between 0.18" and 0.22" will be invisible when compared the the noise in the calculated ECFs. However, users should be aware that there is an additional PSF blur beyond ~0.5", possibly due to the PSF HRMA model uncertainty and the PSF artifact.
When comparing simulations with data, users should be cautious in interpreting any extention of HRC sources as astrophysical. There is a blur term in the HRC observations in addition to the errors due to
- the HRMA PSF,
- an intrinsic detector PSF, and
- aspect reconstruction
Note that the first term is modeled by ChaRT and the second and third are modeled by MARX.
The additional blur, which is not modeled by either ChaRT or MARX, is caused by dithering over residual errors in the HRC event detector position reconstruction. Its size depends on how the dither pattern samples detector space but is typically 3-4 HRC pixels (0.4-0.5 arcsec). For more information see the Blur from Residual Errors in HRC Event Position Reconstruction memo.