M104 R4–7 MOS2 Fe-L Color Map Pilot

Executive verdict

This map is valuable, but its value is as an independent morphology cross-check and contamination diagnostic, not as a temperature map replacing the full spectral fit.

The primary product strictly retains point sources and uses

\[ R_{\rm FeL}= \frac{C(0.875-1.05\ {\rm keV})} {C(0.70-0.875\ {\rm keV})}, \]

The discrete event PI strictly uses low 700 <= PI < 875 (integer PI 700–874) and high 875 <= PI < 1050 (integer PI 875–1049); the matched exact-band QPB, MOS2-native soft protons, and response-folded sky background are then subtracted in counts space, and smoothed with exactly the same 60 arcsec Gaussian kernel for low/high. The results show:

Inside R4–R7, the high Wind S ratio and lower Disk W ratio persist across 60 and 90 arcsec and the point-source included and masked controls; this is large-scale morphology worth continued spectral-fit testing.
With point sources retained, the eight-sector image ratio against the S1–S7 full-fit kT gives Spearman \(\rho=0.750\) (\(p=0.052\)); after applying the existing ESAS source mask it is \(\rho=0.786\) (\(p=0.036\)). Only 7 sectors are comparable; the p-values must not be taken as independent-discovery significance.
60 vs 90 arcsec morphology is highly stable: \(\rho=0.951\), median \(|\Delta R|=0.040\). The 30 arcsec version has too small an effective area and unstable detail, unsuited to the main conclusion.
The integrated ratio of Wind S (S7) is \(1.366\pm16\%\); the point-source-masked control still gives \(1.284\pm23\%\). Its full spectral fit is \(kT=0.981\) keV, but that is a boundary-limited result, so it can only be called a consistent candidate signal.
SP is below 1% of the R0–15 raw counts in both bands; QPB and sky background are the dominant backgrounds. Projecting the MOS1-derived common-SP stress seed directly onto MOS2 would over-subtract by 17–22×, and it must not serve as the main SP image normalization.

MOS2 background-corrected Fe-L low/high rate, point-source-included color ratio and statistical fractional error — The primary background-subtracted Fe-L color map. Point sources retained; orange ellipses mark the existing ESAS source masks.

Component imaging

Each band first constructs

\[ N_i(x,y)=C_i-Q_i-P_i-B_{{\rm sky},i}-L_i, \]

then

\[ S_i(x,y)=\frac{K\ast N_i}{K\ast E_i}, \qquad R_{\rm FeL}=\frac{S_H}{S_L}. \]

Here \(K\) is exactly the same for both bands and all components.

QPB

mos-spectra mask=0 generates full-FOV, point-source-included science products.
mos_back elow/ehigh generates detector-coordinate QPB images at the integer boundaries 700–874 and 875–1049, matching the half-open event selection.
rot-im-det-sky mode=1 rotates the QPB onto sky WCS.
QPB is not a mirror-vignetted sky photon; it is subtracted in counts space first, then divided by the photon exposure map.
Authoritative-grid full-map counts: low 1,602.69, high 1,567.99.

Residual soft protons

proton spectrumcontrol=2 uses the dedicated MOS2 SP detector template; rot-im-det-sky mode=2 rotates it to sky.
The Sherpa model multiplies the SP surface norm by the sky area, so the bnorm passed to proton must be \(334.9896\times3.6361\times10^{-6}=1.21805\times10^{-3}\); the surface norm must not be passed directly.
The generated template is then normalized to the R0–15 response-folded native-MOS2 counts: low scale=1.3065, high scale=1.3347.
The final full-map SP counts are only 21.21 and 21.67, with little effect on the main morphology.

Vignetted sky background

Local Bubble, Galactic foreground, and CXB are treated with sky-uniform surface brightness:

\[ B_{{\rm sky},i}(x,y)=\beta_iE_i(x,y). \]

\(\beta_i\) is set by the R0–15 real-MOS2 response-folded model counts and the BACKSCAL area. The full-map counts are low 1,314.96, high 984.25.

R0–15 spectral closure

Band	Raw PHA	QPB	M104 APEC	Sky	Residual lines	Native SP	Model/raw closure
0.70–0.875 keV	2,188	902.29	415.22	861.86	13.89	13.51	+0.9%
0.875–1.05 keV	1,941	878.95	374.43	645.10	21.94	13.84	−0.4%

This shows the native MOS2 background topology closes in both bands; the MOS1 common-SP stress seed does not.

Four control panels comparing raw, QPB-subtracted, full-background-subtracted and common-soft-proton stress Fe-L ratios — Sequential comparison of raw, QPB-only, full background subtraction, and the MOS1 common-SP stress.

Smoothing-scale stability

Running adapt separately for low/high and then dividing is not allowed, because the two bands would acquire different spatial transfer functions. This pilot fixes one shared Gaussian kernel and checks 30, 60, 90 arcsec.

Shared sigma	Comparison	Spearman \(\rho\)	Median \(\|\Delta R\|\)	95th percentile \(\|\Delta R\|\)
30 vs 60 arcsec	R4–R7 common-valid pixels	0.791	0.093	0.294
30 vs 90 arcsec	R4–R7 common-valid pixels	0.669	0.114	0.378
60 vs 90 arcsec	R4–R7 common-valid pixels	0.951	0.039	0.126

60 arcsec is therefore the primary version balancing spatial resolution and statistical stability; only large-scale structure appearing at both 60/90 deserves further interpretation.

Common-kernel stability comparison for 30, 60 and 90 arcsec smoothing — Shared-kernel check at 30, 60, 90 arcsec. The 30 arcsec effective area is small; the 60 and 90 arcsec large-scale morphology agrees.

Point-source control

The primary product retains point sources as required. To judge whether the large-scale color is entirely point-source driven, a companion control was made: the same 102-region ESAS cheese mask applied to low/high counts, QPB, SP, sky, and exposure, then the same 60 arcsec kernel.

6,745 pixels masked inside R4–R7.
The included and masked ratios over common-valid pixels give \(\rho=0.857\), median \(|\Delta R|=0.054\).
The high Wind S ratio and lower Disk W ratio persist; many small-scale hotspots and central structure disappear.
The masked control is not the primary product; it only shows the large-scale ordering is not entirely caused by point sources.

Point-source-included and point-source-masked Fe-L color comparison with uncertainty and exclusion mask — The point-source-included primary map vs the masked control; the bottom row gives the masked error and the actual source mask.

R4–R7 sector cross-check

The ratios below are obtained by integrating net counts and exposure directly within each 45-degree sector, not by averaging smoothed ratio pixels.

Sector	Role	Full ratio	Statistical frac. error	Point-masked ratio	Point-masked frac. error	Full-fit kT (keV)
S1	Disk W	0.749	13.8%	0.788	21.0%	0.698
S2	NW transition	0.721	22.4%	0.688	26.7%	0.807
S3	Wind N	0.781	47.4%	0.623	77.6%	0.707
S4	NE transition	0.996	21.2%	0.932	24.5%	0.911 boundary-limited
S5	Disk E	0.846	17.4%	0.809	21.8%	0.808
S6	SE transition	0.761	21.4%	0.663	25.5%	0.771
S7	Wind S	1.366	16.4%	1.284	23.4%	0.981 boundary-limited
S8	SW transition	0.898	11.6%	0.743	19.3%	unavailable

S3's fractional error in the point-source-masked control reaches 77%, unusable for morphology. S7's high ratio is the most stable, but its spectral fit is also near the high-temperature boundary, so it still needs normal/confidence and multi-observation/multi-camera verification.

Sector-integrated Fe-L ratios under different background and point-source treatments compared with spectral-fit temperatures — Image-integrated sector ratios, background controls, and the full-fit kT comparison.

Caveats

With point sources retained, the ratio mixes thermal gas, AGN/XRB power-laws, and PSF wings; no local \(kT\) interpretation holds.
Even point-masked, the ratio→kT mapping still depends on NH, abundance, APEC/AtomDB, and response, and the background posterior has not yet been propagated per-pixel by Monte Carlo.
The current fractional-error map propagates only the raw-count Poisson term; QPB corner normalization, FWC finite counts, sky-fit posterior, SP-template, and residual-line systematics must be added in the formal version.
This is display-level background-corrected morphology, not a joint-Poisson posterior inference map; the low-band \(Z\ge3\) profile-likelihood gate, background nuisance draws, independent-beam FDR, and before/after GTI time split are not yet done, so there is no pixel-level detection significance.
The R4–R7 sector comparison and the spectral fit use the same observation and cannot be viewed as fully independent validation.
The full-fit kT of S4 and S7 is boundary-limited; S8 has no qualified spectrum.
The conclusions only support a "Wind S hard/high-Fe-L-ratio candidate" and the large-scale color asymmetry, not a measured 2-D temperature field.

Reproducibility issue solved

The old project wrapper fkeypar on this machine reads only the FITS primary HDU, while mos-spectra reads LIVETI0n from the EVENTS extension; the result was a division by zero on CCD2. Two parallel tasks also shared the same /tmp/fkeypar_last_value.txt, a race condition. This pilot uses a project-local, per-workdir fkeypar/pget compatibility layer; the SAS installation was not modified.

Discrete PI half-open correction

The first version used the inclusive semantics of SAS PI in [lo:hi], so 50 events at PI=875 entered both low/high and 29 events at PI=1050 fell beyond the nominal upper limit. That version is archived and superseded. The current version uses 8430 low events and 7698 high events, with matching QPB/SP images regenerated.

The R4–R7 common-valid ratio morphology before vs after the correction remains highly consistent: \(\rho=0.9978\), median \(|\Delta R|=0.0058\), 95th percentile \(|\Delta R|=0.0232\); the largest sector full-ratio change is 0.01895. The science morphology does not substantively flip, but strict band provenance requires the current half-open release.

Independent Claude Code Opus audit

Claude Code Opus 4.8, high effort, ran two read-only independent audits of the historical 2-sector, R4–R7 4-sector, axis-centered 8-sector/7-fit, and the current band-ratio. The second round read back the post-half-open-rebuild FITS headers, HDU counts, QPB/SP hashes, sector CSV, and focused tests directly.

Current band-ratio release: PASS, but limited to the corrected color-morphology cross-check; it is not a 2-D temperature or significance map.
8-sector spectral publication: REQUEST_CHANGES. The current S4/S7 are limited by kT.max=1.0, and the release is still FAST triage without final normal/2D confidence.
The 4-sector gives Wind N/S=0.780/0.760 keV while the 8-sector gives 0.707/0.981 keV — the N–S ordering flips with binning. A quantitative "Wind S hotter by 0.27 keV" conclusion therefore cannot be published.
The next step must refit 4/8-sectors under a unified MOS2-only contract with identical filter/background/grouping/free-state, raise the kT ceiling to at least 1.5 keV, and run final confidence.