Are there any differences between "standard" and SAS-generated EPIC response matrices?

There is, however, a case where the usage of SAS-generated response matrices is largely to be preferred. If one has a spectrum with very good statistics in an emission line with centroid energy larger than 2.5 keV (e.g.: the Iron K-alpha line), one should be careful about using the "standard" matrices. These are binned at en energy resolution of 50 eV. XSPEC does not perform any interpolations in the energy space, and therefore the line energy can be obtained with an accuracy ~25 eV. A more accurate determination of the line energy can be obtained by running arfgen and rmfgen to create more finely spaced matrices:

rmfgen spectrumset=${myspectrum} rmfset=${myrmf} withenergybins=yes
nenergybins=7475 energymin=0.05 energymax=15.0

arfgen spectrumset=${myspectrum} arfset=${myarf} withrmfset=yes
rmfset=${myrmf}

The redistribution matrix, binned at 2eV, takes about 25 MB of disk space. On the other hand, the ARF is small.