When spectra are too bright for coi-correction, by example

Coincidence loss corrections become unreliable when the count rates are too high. The first reason for this is that the observed count rate per frame (per coi-area) edges ever closer to 1 for incident rates larger than about 3 counts/frame. As a result, the error in the incident count rate derived from the observed count rate becomes larger. The error depends on the exposure time, or rather, the number of times the detector is read (frames) in one exposure. Statistically, a certain number of these frames will not have a photon measurement in a chosen coi-area, and the error will depend on the statistics of that number of frames in the bright limit.

For example, if we have 10,000 frames for an exposure and a spectrum with an observed count rate per frame of 0.97, then 3%, or 300 counts in our chosen coi-area will have no photon. The error is approximately the square root of that, so 17.3, leading to a measurement error of about 6%. If we would do a shorter exposure, say of only 1,000 frames, the statistical error jumps to 18%.

In this way, the statistical error limits the rate to which we can sensibly correct coincidence loss. The method of correction has been worked out for point sources, but is equally applicable to that in spectra, but replacing the circular coi-area for point sources with the rectangular one in spectra.

For bright sources with count rates per frame (per coi-area) exceeding 6, the probability that the pulse-height distribution of the photon splashes in a frame hits the limit in neighbouring pixels appears to be the cause of a decrease in the observed rate in counts per frame. We would expect it to be close to 0.997 counts per frame from the statistical nature of coincidence loss alone. This effect has not been modeled in detail yet, so there is a slight possibility that this also affects the maximum observed rate per frame at lower incident rates. For the version 2.x.x calibration it has been assumed that the observed rate below an incident count rate of 4 counts per frame is unaffected.

The appearance of the spectrum with increasing incident count rate per coi area is an indication of the brightness that changes from a smooth appearance to a heavily patterned appearance due to the approximate centroiding used on-board the UVOT, which fails gradually for increasing coincidence loss and brightness. To illustrate this we show some figures below.