Low Energy Transmission Grating with the ACIS-S detector -- Carbon K edge region

The problem was first noticed in a GO observation (see Figure 1).  This feature was subsequently detected in the calibration observation of 3C 273 (obs ID 1198).  The residuals of the GO target and 3C 273 are compared in Figure 1  The similarity is quite striking and indicates that the difference was intrinsic to the instrument.  [The drop beyond 55 A is due to the event threshold cutting into the PH distribution, so it is not likely to be intrinsic to the source.]  The LETG/HRC-S data from 3C 273 taken immediately prior to the LETG/ACIS-S observation do not show a comparable residual, indicating that the feature is not entirely intrinsic to the source.  See the LETG/HRC spectrum from Jeremy Drake and Deron Pease showing a 2PL model; the (+1 order) residuals do not show a jump near 43 A.  The -1 side is similar.
Figure 1. Comparison of two LETG/ACIS-S observations of AGN.  The residuals give the fractional differences between the simple power law spectral model that was folded through the instrument response and the observed count spectra.  Note the apparent edge at about 43 A, which is near the K edge of carbon.  The gap near 36 A is due to a gap between the S1 and S2 CCDs.  The upper target is a GO target and the lower one is 3C 273 (obs ID 1198).

The data in Figure 1 were analyzed using released, pre-flight calibration products.  More details of the 3C 273 fits are shown in two PS files giving the 0-30 A region and the 30-60 A regions separately for both the plus side and the minus sides.  The spectral fit was a simple power law with a photon index of 1.6 and Galactic absorption of 1.7e20 cm2.  For this column density, the edge at C-K would only be about 10%, substantially greater than the observed edge.  The model of the C-K edge is clearly inadequate in these released calibration files, so calibration files wIth higher spectral resolution were obtained from the ACIS team.  Figure 2 shows the new model of the S3 QE in the C-K region, improved primarily due to the improved resolution of the optical blocking filter (OBF) model.

Figure 2.  New and old ACIS BI quantum efficiency models compared.  The main difference is in the improved resolution of the OBF model.  Note the resonance absorption features at 0.2851 and 0.2866 keV and the EXAFS at E > 0.288 keV.  The new OBF model was supplied by George Chartas at PSU and the ACIS QE models were supplied by Catherine Grant and Mark Bautz of MIT.

In early May, data from a much brighter source, XTE J1118+48, were obtained.  The data will go public in mid-August and will be available for all users.  Because it's so bright at 0.28 keV, the data were not binned, so that the filter model could be examined in detail. Figure 3 shows the C-K region and how the location of the resonance absorption features don't quite match the model.  A shift of 0.5 eV was required to get a decent fit.  In addition to the energy shift, the depth of the resonance features was tested by extrapolating a simple model from lower energies.  The result is that a 13% increase in the optical depth of the resonance features gives an agreement shown in Figure 4.  The filter was decomposed empirically in order to determine the transmissions of each filter component so that the C-K edge could be adjusted independently of the other components in the model.

Figure 4 Figure 4 (portion).  Click on this fragment to get the full version.

The absorption above the edge was fit to a simple model where the opacity decreases as 1/E^9, which is somewhat steeper than expected if due only to ionization of neutral material.  Figure 5 shows the fit of the data to this model and Figure 6 shows that the FI chip shows the same overall level of absorption in the .29-.33 keV range.  Thus, the new absorption component is not limited to the BI chip and if it is due to the OBF or contamination on the OBF, then the new component is rather uniform.

Although the origin of the new absorption component is not known, it appears that it is not associated with a particular CCD type, which don't have carbon in them anyway, so I have constructed a new OBF model based on the original model from George Chartas that includes this new absorption component as well as fixing the energy shift and adjusts the optical depth by 13%, as described earlier.  The IDL code is available as well as the new filter model, which I have pruned to an accuracy of 0.1% using Dan Dewey's IDL pruning routine.  It is compared to the version released by George Chartas in Figure 7.

Caveats and Conclusions

There are a number of possible problems with this analysis that I have tried to address and a few other points bear discussion.
  1. The 0.0005 keV shift that is needed to line up the observed and modelled resonance features may be partially explained by systematic errors in the wavelength scale of the data, rather than the model.  The sense of the correction is that the OBF energies near C-K are reduced -- equivalently, one could increase the energies of the events or decrease their wavelengths.  Although a final conclusion hasn't yet been reached, there is evidence that the LETG/ACIS-S wavelengths are systematically too large by about 0.1(1)%, which would be an error of 0.28eV -- about 50% of the required energy shift.  The cause of the wavelength error has not yet been established firmly.
  2. Pulse height (PH) extraction was quite liberal but should not have a sudden change across the C-K edge, so it is unlikely to cause the apparent absorption.  The pre-flight gains and offsets were applied to the PH data and then the event energies were corrected on a node-by-node basis to provide better agreement with the energy inferred from the dispersion by the LETG.  The PH selections should be photometric at a level of order 1%.
  3. Bad columns are flagged using a 50-pixel running median filter.  There are very few, so the data were eliminated but the exposure function wasn't adjusted to account for them.
  4. If the new absorption feature were due to carbon in a material similar to polyimide, then the additional optical depth, 1.34, can be compared to the optical depth due to the carbon (as decomposed) at 0.2875 keV in the OBF: 1.1.  Taking the fraction of C by mass in polyimide, 0.72, there is an equivalent thickness of 1440 A of carbon in the filter; thus, the new absorption component would require about 1750 A of carbon in a compound such as polyimide, or 2435 A of polyimide.   This is substantially greater than the estimated contamination due to nonvolitile residues estimated by Lorraine Ryan -- 125-250 A -- in an e-mail to Steve O'Dell on 4/27/00.
  5. Intrinsic absorption is not well known but was estimated to be of order 1.0e20 per cm^2, so the jump at the C-K edge that is intrinsic to the source should be of order 4%, based on a cosmic abundance of carbon (3.6e-4) relative to hydrogen.  It is not yet clear whether there are associated resonance features, because some of the carbon along the line of sight would be in grains and even atomic carbon might well have some resonant absorption features.  Current experts (such as Brendan McLaughlin) do expect some resonant edge structure, so I may have included some part of the ISM term into the 13% filter adjustment, which is the equivalent of 187A of carbon alone.  I estimate that ISM with NH of 1e20 would have an equivalent thickness of ~50 A, so ~30% of the adjustment may be due to the ISM.  Better theoretical models of the resonant absorption at C-K may help.  The ISM cannot explain the excess absorption above 0.2867 keV, however, because it would require a x30 overabundance of carbon relative to hydrogen and because of the LETG/HRC-S result.
  6. Background was not subtracted from any of the spectra because the source (XTE J1118) was very bright and the background was, in general, very small.  The background will be important only in the .29 keV region in the +1 spectrum which is detected on a FI detector.  This could explain why the FI seems to deviate from the BI in Figure 6 except that the agreement above 0.31 keV would then be in question.
  7. Even if the 13% thickness adjustment needed for agreement in the resonance absorption features is associated with the excess absorption at energies above 0.2867 keV, then the material has a very different C-K absorption profile than the carbon in polyimide.  The data are consistent with the possibility that there is only continuum absorption and no resonant lines, which seems peculiar because even pure carbon should show the resonance features.  I have heard no plausible explanations for how this can happen.

Update (8/4/00)

I have compared the flight ACIS and HRC filter models using a decomposition of the +1, thin Al HRC filter model provided on the HRC UV ion shield calibration page. I also compared the decompositions with predictions from the relatively new transmission model for the HETGS polyimide provided by Kathy Flanagan (optical-constants.990510.tbl). The results are in one PS file. I assumed that the polyimide thicknesses were 2000 A for the OBF and 2050 A for the HRC filter. For Al thicknesses, I used 1300 A for the OBF and 270 A for the HRC. The HETGS model is normalized to 10 A, according to the header documentation. Briefly, for each of 6 energy bands, there are some discrepancies.
  1. Below the Al-L edge near 0.073 keV, all filter opacities are dominated by C and O absorption which I have modelled using fitting constants in Astrophysical Quantities (4e, p. 109) and the fractional areal densities of C and O expected using the polyimide formula given in the ACIS calibration report v2.20, p. 273: C22 H10 O4 N2. Results are shown in the top two panels of the first figure. The prediction from the HETGS and the OBF agree but are a x1.42 below the opacity observed in the 0.06-0.07 keV region. The prediction of the HRC opacity below the Al-L edge requires a 10% increase.
  2. In the lower two panels of the first figure, the predicted opacity due to polyimide is subtracted from the modelled filter opacities in order to estimate that component due to Al-L opacity alone for both the HRC and ACIS filters. The result is that the energies of the features agree well but that the magnitude of the opacities differ by ~10% in the 0.073-0.30 keV region. No single correction brings these into agreement at the 1% level.
  3. The C-K region is compared in the second figure in the PS file. The HRC and ACIS filter models show good agreement in the energies of the resonance features between 0.285 and 0.290 keV and in the overall shape in the 0.29-0.36 keV region. The magnitude of the opacity differs 25%, however in the resonance features but only 3% in the .305-0.360 keV region. The ACIS filter model shows features in the 0.29-0.30 keV region that the HRC filter does not mimic.
  4. The N K region is compared in the third figure in the PS file. The two filters agree very well on the shape of the N opacity except for two places: 0.395-0.400 keV, where the HRC model shows opacity while the ACIS and HETGS polyimide models do not, and at 0.45 keV where there is a 1% "kink" in the ACIS model that is not observed in the HETGS and HRC polyimide models. The kink occurs at a point where the sampling frequency in the ACIS model changes from 0.2 eV to 0.5 eV, so it appears that two different measurement runs were used. Overall, a 10% increase in the ACIS/HETGS prediction is required to match the HRC model.
  5. The O K region is compared in the fourth figure. The HRC and HETGS models agree on the energies of the resonance feature at 0.531 keV while the ACIS edge has to be shifted by 0.8 eV to match the HRC and HETGS. The overall structure of the opacity is quite similar in all models except for a kink in the HETGS model at about 0.537 keV. If one normalizes the opacity of the resonance features, the HETGS model has to be shifted by x1.3 to match the HRC opacity model but the shift is x0.8 if normalized in the 0.54-0.64 keV region. The HRC and ACIS opacities agree better but require a scaling of 0.75-0.88. Again, there is a 4.5% kink in the ACIS model -- at 0.56 keV in this case -- which can be related to a change in the sampling size of the ACIS model.
  6. The Al K region is compared in the fifth figure. There is an overall energy shift of 8.0 eV required to match the shapes in the 1.575-1.66 keV region. A scaling of 1.13 is also required. The ACIS filter shows a feature that is not present in the HRC model at 1.570 keV.
In summary, one can get a good agreement in the expected Al opacities if the thickness of the Al in the HRC UV ion shield is increased about 10% to 300 A, which is within the Luxel measurement error. The overall shapes of the edges agree well and the energies of the edges agree to 0.2 eV at Al-L, C-K and N-K. There are two cases where an energy shift is required to bring filter models into agreement. There is no simple scaling that brings the edge opacities due to the polyimide components into complete agreement; values between 0.75 (C-K and O-K resonance features) and 1.1 (N-K) were obtained. There seem to be kinks in the ACIS opacity models that appear at energies where the grid spacing changes that may be due to measurement setup calibrations. There are some other odd differences that have no obvious explanation.

In Progress

I am currently in the middle of checking this new filter model against other LETG/ACIS-S data, including 3C 273, a GTO observation of PKS 2155-304 and the GO observation where this effect was originally found.  Preliminary indications are that the adjusted model works well for PKS 2155 but that it slightly overcorrects the data of the GO target and 3C 273.  I am also working on an adjustment to the BI QE model because there is a definite discrepancy of 5-20% between the FI and BI data in the 0.6-1.3 keV band.  This adjustment has been worked out for the XTE J1118 data and seems to work for 3C 273 and PKS 2155-304.

Update -- Feb. 26, 2001

I have determined an empirical correction to the ACIS-S QE model that accounts just for the anomalous C K edge.  Using two observations of the BL Lac object PKS 2155-304, I have the following results:
  1. The strength and shape of the edge are consistent with the results obtained from the observation of XTE J1118+480.  The optical depth at the edge is 1.34, the energy at which the edge appears is about 0.286 keV, and the spectrum recovers as though the opacity decreases as E^(-3.0).
  2. The edge strength has not changed in the six months between the two observations (May 2000 to December 2000).
  3. The edge does not appear to have associated anomalous O K edge absorption.
  4. The two observations took place at slightly different off-axis angles (+1.5' vs -1.5') and the results are consistent.
Figure 8 shows the spectra obtained when using the pre-flight releases of the detector QE.  The absorption edge at C K is clear and present in both observations.  The dotted lines are simple power law fits to all the coarsely binned data.  Figure 9 shows a set of simple fits to these data.  The top panel gives the ratio of the two spectra to the individually fit power law spectra.  The residuals are consistent, so the data were combined using an uncertainty weighting.  The middle panel (shown below) gives the result of combining the observations and shows a simple edge model fit that requires accounting for a systematic error in the spectral slope.  The bottom panel shows the residuals from this fit, which show some systematic deviations that are partly related to the detector QE models.

A portion of Fig. 9, which shows the 0.2-0.4 keV region and the model for the anomalous C-K absorption edge.

The model fit is available as an ascii file that can be used as an auxiliary filter for LETG/ACIS-S QE modelling. The functional form is very smooth except at the edge, so it has been "pruned" using Dan Dewey's IDL software and should still be accurate to 0.1% (which is smaller than the systematic residuals).  We do not yet know if this auxiliary QE model should be used for observations near the center of the ACIS-S array, because all LETG/ACIS-S observations are offset close to the readout rows.

Contact Herman L. Marshall (hermanm@space.mit.edu) for further information.
Last modified: 8/4/00