As described in a recent STAN*, the COS FUV detector is beginning to exhibit gain sag. One consequence of this effect is the mis-registration of photon events in the cross-dispersion (XD) or Y direction, commonly referred to as Y walk. While Y walk does not adversely affect science data, it can reduce the accuracy of target acquisitions obtained in dispersed light. If the target is centered in the aperture, but the Y walk shifts its spectrum in the XD direction, then the ACQ/PEAKXD algorithm will miscalculate its centroid and move the target away from the aperture center.

We have considered several schemes for modifying the ACQ/PEAKXD algorithm to correct for this effect. The simplest turns out to be the following:

- Ignore the data from segment B when computing the spectral centroid.
- Add 2 pixels to the computed centroid to account for Y walk on segment A.

This algorithm has two properties that make it particularly appealing:

- At present, we require a S/N of 40 in the combined A+B data to compute an accurate centroid. Simulations indicate that an accurate centroid can be obtained with a S/N of 20 to 25 using segment-A data alone. (The XD distribution of A+B counts is broader than that from segment A alone, so more counts are required to compute its centroid.) According to the COS ETC, integration times long enough to achieve S/N = 40 in A+B yield S/N > 25 on segment A, so no change to exposure times is required for most targets.
- These changes will enable all FUV gratings to be effectively used for spectroscopic target acquisitions and will go into effect on April 18, 2011. Cycle 18 programs have been checked and modified as necessary to function with this change.

For Cycle 19, version 19.2 of the COS ETC will compute the S/N for segments A and B separately for dispersed-light target acquisitions. In the mean time, here is a simple algorithm for estimating the S/N of a PEAKXD acquisition:

- Perform an ETC calculation for your acquisition.
- Note the value labeled "Count rate segment A," which represents the entire detector, a region larger than the checkbox used to compute the centroid.
- The dark current over the entire detector is 38 counts/sec, while that in the checkbox is 2 counts/sec. The sky is even fainter on segment A (assuming the use of G130M), so we neglect both contributions.
- The signal-to-noise ratio in the checkbox on segment A is thus
S/N = SQRT[ ((Count rate segment A) - 38) * EXPTIME ]

If the S/N of your segment-A spectrum is greater than 25, then your ACQ/PEAKXD should perform swimmingly.

* http://www.stsci.edu/hst/cos/documents/newsletters/cos_stis_newsletters/stan1012.html