Summary

During the recent/ongoing recalibration of the LETG/HRC-S Effective Area, Nick and Jeremy found a faster-than-trend decrease in the EA at some (especially the longest) wavelengths covered by HZ43 spectra. Brad noted that the affected wavelengths correspond to regions of the HRC-S where detector gain is lowest and suggested that as gain continues to decrease, a significant and increasing fraction of events are now falling below threshold and being lost. Further investigation confirmed this idea and showed that a substantial range of the LETG/HRC-S is experiencing detectable excess QE loss and that the rate of loss is increasing. The critical point seems to be when the mean PHA falls below ~40. All figures below are from Nick unless otherwise noted.

General Trend of Decreasing QE

Figure 1: Data points represent the average flux at all wavelengths covered by LETG/HRC-S spectra of HZ43 (i.e., there is an implicit assumption that the trend of decreasing QE is "gray"). Jul 2008 was chosen as the reference point for normalization of the HRC-S QE corrections. After the best-fit linear corrections, some points (around 32, 37, 43, and 79 months) are significantly below trend for unknown reasons. The latest points (after 130 months) are also below trend, at least partially because of the excess QE loss discussed here. Observations later than month 100 occurred after the latest HRC-S gain calibration.

Excess QE Loss Correlated with Low Gain

Figure 2: Mean SAMP (which became the new PHA in 2008) versus wavelength, using HZ43 and PKS 2155 data. Mean SAMP is calculated using 5|95% truncation of background-subtracted data. Excess QE loss (see Figs. 3 and 4) is seen in regions with low mean SAMP (i.e., low detector gain). (This is Fig. 11 from the 2008 Gain Map report.)

The plots below are similar to Fig. 1, but have been broken into separate curves for different wavelength bands. The curves have been "globally renormalized" by dividing by the red curve in the bottom panel of Fig. 1 to remove the all-wavelength deviations from linear decline (especially for points 3-5 and 11), and then dividing each wavelength's points by the statistically weighted average of the 2nd through 10th points (which are assumed to be free of excess QE losses) so that things will average to 1 in the absence of excess QE losses. The first observation (ObsID59) is excluded from that average because of relatively rapid gain change at early times. The net result of this renormalization is to better reveal deviations of QE within specific wavelength ranges from the broad-band QE changes.

Note that the excess QE decrease at some wavelengths makes the global renormalization factor too large for wavelength ranges that are not experiencing excess QE loss; the curves for the latter wavelengths therefore trend toward values higher than 1 as time goes on. The net excess QE loss seen for λ=150-160 Å is therefore 1.02-0.93 or ~9% in the latest (March 2011) observation.

Figure 3. Renormalized QE versus time, with representative error bars on a few curves. Statistics at negative short wavelenths (roughly -60 to -90 Å) are not good and it is hard to draw conclusions. Excess QE losses at large positive wavelengths (> +140 Å) are clear; 130-140 Å also shows some deviations relative to the unaffected wavelengths.

Figure 4. Same as Fig. 3 but versus mean background-subtracted PHA instead of time. QE appears to increase its rate of decline rapidly when mean PHA falls below ~40.

Loss of Low-Channel Events in PI Pulse Height Distributions

Fig. 5 shows PI PHDs over time for a few wavelength ranges that show clear excess QE losses (λ = +140 to +150, +150 to +160, and +160 to +170 Å) or possible excesses (λ = -60 to -70, -70 to -80, and -80 to -90 Å). The wavelength range -150 to -160 Å has not displayed excess QE loss and was chosen for comparison. The PHDs for ObsID 13025 are systematically low for unknown reasons; those for ObsID 5957 are somewhat higher than average, as noted during the 2008 gain map recalibration. As can be seen, most of the wavelength ranges shown appear to be losing an increasing fraction of their lowest-channel events as time goes on.

Figure 5: PI pulse height distributions (PHDs) over time for various wavelength ranges, in approximate order of the degree of low-channel event loss (starting with ~no above-trend loss on top).

	-150:-160 Å (pdf)
-80:-90 Å (pdf)	-70:-80 Å (pdf)	-60:-70 Å (pdf)
140:150 Å (pdf)	150:160 Å (pdf)	160:170 Å (pdf)

Figure 6 shows integrated count fractions (normalized to total BG-subtracted counts) versus PI. The spread between curves is a measure of whether and how fast the HRC-S gain decrease is leading to low-amplitude event losses. Note that the widest spreads are seen in detector regions with the lowest gains (see Fig. 2), e.g., λ>130 Å. Although statistics are too poor to show significant excess QE loss around -70 Å in Figs. 3 and 4, these plots indicate that it is probably occuring there, too. Another advantage is that trends in event loss can be seen even when relative QE is ~impossible to measure directly because of source variability (such as for PKS 2155); the region around -30 Å needs to be looked at, since gain is low there. (I think that if we rescale these curves to the number of counts in the PI range where event loss should be negligible--say, PI above 50 or 60--then we can obtain a pretty good estimate of the QE loss as f(t,λ). The equation for the normalized number of counts in a given PI bin in ObsID "obs" would be N_obs^norm(PI) =N_obs(PI)*{N_obs59(PI>60)/N_obs59(all PI)}/N_obs(PI>60).

(pdf)
Figure 6: Integrated count fraction versus PI over time (17 HZ43 observations using the same color scheme as in Fig. 5), for 20 wavelength ranges.

HZ43 Visibility

Lately we've been making HZ43 observations once a year around March. If we need to observe more frequently that might be a little tricky.