Next: A.3 SPECUNIT Field Up: A New Implementation of Previous: Overview

A.2 Implementation

Systematic errors shall be stored in the !aux.syerr field of the sdrs. Using the "XSPEC standard" approach. The systematic error is stored as a fraction of the data value (!aux.dat) for each channel. Thus they are independent of data units - much easier to handle.

When used in fitting, the new error is computed

   err =  sqrt( !aux.err^2  + (!aux.dat * !aux.syerr)^2)

and stored in !aux.err prior to the call to do_foldanal or ffold.

The specunit field is updated to indicate SyErrs applied (see SpecUnits below)

SDR_SUM.PRO has been modified to detect and propagate !aux.syerr using weighted root mean square.

SDR_REBIN.PRO has been modified to detect and rebin !aux.syerr using weighted root mean square.

If systematic errors have not been put in the sdrs, all routines operate as before. FITIT.PRO has a new keyword, /NOSystematics, which will disable the quadrature summing of systematic errors when they are present in the sdrs.

At this time, the actual creation of the systematic errors is an independent process. For AGN work and all of the OSSE products I provide users for XSPEC analysis, a routine, DTW_SYS.PRO, creates systematic errors based on uncertainty in the DTW correction. This provides about a 3at 50 keV which drops to 0.3for papers....

        OSSE systematics. 

    The present data include estimated systematic errors.  These systematics
    were computed from the uncertainties in the low energy calibration and 
    response of the detectors using both in-orbit and prelaunch calibration
    data.  The energy-dependent systematic errors are expressed as an
    uncertainty in the effective area in the OSSE response.  These systematic
    errors were added in quadrature to the statistical errors prior to
    spectral fitting.  They are most important at the lowest energies
    (approximately 3% uncertainty in effective area at 50 keV, decreasing to
    0.3% at 150 keV and above).  In the case of a source as strong as Cyg X-1,
    the systematic errors dominate the statistical errors from a standard two
    week observation out to an energy of about 120 keV.

Other implementations of systematics can be supported to the user's desire. The !aux.syerr field must have the same number of channels as the data field and must be expressed as a fraction of the data field.