What is the author's name?
请输入页面作者的名字。
Incorrect / 不正确
Not a label for "12 bad observations", but a per-case diagnosis of the current line-anchored selection rule: which cases have non-shared instrumental lines at the mode, which have a scan range that does not cover the mode, and which still show signs of an insufficient continuum model.
It is a model comparison of whether the instrumental line needs to be fit separately in the low/high spectra, not a goodness of the whole spectrum, nor a probability of soft proton or "the probability that the instrumental line really changed".
Δχ²=χ²shared−χ²freep=P(χ²₂≥Δχ² | Mshared)The current screening rule takes p≥0.05 as the condition that "the shared line of this trial is not rejected". p=0.106 means that after releasing the two line areas this time, Δχ²=4.48; under the shared-line model, an improvement this large or larger has about a 10.6% chance of occurring, so we cannot reject sharing, but this is by no means "proof of equality". p=0.018 means Δχ²=7.98; under the same convention it is below 0.05, so the current model rejects sharing, but by no means "only a 1.8% probability that the instrumental line really changed".
The two component+residual figures below visualize exactly this computation: solid-line/open residuals are the free-line model, dashed/solid residuals are the shared-line model; the χ² difference between them is the source of p.
The "range" here is the continuous trials on the rate axis that pass the current line-tie rule, not the 1–2 keV energy fit range; all trials use the same fixed 1–2 keV, 20-eV PI grid.
| Quantity | actual value for 0800730101 MOS2 | what it determines |
|---|---|---|
| actual rate scan | 1.63–2.39 c/s, one trial per 0.01 c/s | Re-partitions low/high time samples per trial and performs a joint spectral fit. |
| broadband histogram mode | 2.383 c/s; nearest tested point 2.38 c/s | The current old rule requires this point to lie inside an eligible rate segment. |
| all eligible rate segments | 1.78–1.79, 1.91–2.18, 2.23–2.30, 2.32–2.35 c/s | eligible only means p≥0.05 and low/high partitions each at least 10% of the coarse support. It is not "only these rates can make the spectral fit converge". |
| segment nearest the mode | 2.32–2.35 c/s, 4 grid trials, 0.033 c/s from the mode | It is indeed very narrow on the rate axis and does not contain 2.383; it therefore triggers the mode-containment veto. This is exactly the rule fragility we need to review/improve. |
p and goodness-of-fit are different questions.At the mode-nearest trial=2.38, the forced shared line has χ²ν=1.019 (94 dof) and the free-line model χ²ν=0.920 (92 dof); neither is "the whole spectrum obviously failing to fit". But freeing the two line areas lowers χ² by a further Δχ²=11.178. The current code converts this gain through the asymptotic χ²₂ distribution of 2 extra parameters to p=0.00374, so it only says: under the current narrowband model and this approximation, the constraint "the two spectra's line areas must be equal" is not supported by the data. It does not measure overall goodness, and is not an SP probability, a physical line-change probability, or a posterior probability; this p threshold has not been Monte-Carlo calibrated and cannot stand alone as a physical conclusion.
The upper figure is the complete 51-s broad lightcurve within the base GTI (y-axis fixed 0–5 c/s; higher flares shown as top-of-plot triangles). Orange points are the low centre samples within the 2.09±0.162 c/s window, blue points are the high samples within the same coarse support, grey points are outside the coarse support. Bottom-left uses Al as the centre criterion (thick solid line, left axis) with Si only as a faint-dashed right-axis consistency reference; bottom-right is the histogram of the improved 51-s decision stream, marking the decided mean rate and the candidate GTI rate window corresponding to that centre.
Delta chi2=0.025 (p=0.988), where Al high−low=+0.024σ, Si high−low=−0.156σ, and low/high exposure is 8.238/11.155 ks. It is determined by the joint consistency of the two lines, not the exact geometric intersection of one line area. The histogram shading is the candidate rate window 1.928–2.252 c/s; no new final GTI is currently written from it, so it must not be misread as a completed new filter product. Stable SP also cannot be removed or quantified from this figure alone. Click to enlarge.Each point is the actual trial nearest the broadband histogram mode. The x-axis is the log likelihood-ratio p-value of the shared Al/Si area relative to the free-area model. The red dashed line is p=0.05; its left side means the current narrowband model prefers independent line areas.
The x-axis is the absolute distance from the broadband mode to the nearest line-supported interval. A small distance does not prove a GTI is acceptable; it shows that treating the histogram mode as a hard containment gate is very sensitive to the finite grid/bin resolution.
Each row is normalized to that observation's own p1–p75 scan range. Green bars = continuous trials passing shared line + p≥0.05 and partition balance; red line = broadband histogram mode. A red line beyond the right edge means the mode is not in the scan at all.
This is the complete actual trial profile, not just the final p-value. The y-axis is the line-tie p (log); the red dashed line is 0.05, the cyan bottom bars are eligible trials, and the vertical red line is the broadband mode. Note: every trial redefines a set of symmetric low/high time samples; adjacent 0.01 c/s trials are not continuous fine-tuning of the same batch of times.
| case | primary category | mode / trial | tie p | small partition | nearest supported-region distance | free χ²ν | Al / Si difference |
|---|
Expand any case to see the full values, all supported intervals, and the native QDP lightcurve. That QDP figure is for observing the standard-side time selection; it is not this page's broad 0.2–10 keV line-anchored decision rate, nor an SP flux figure.
NO_MODE_SUPPORTED_RANGE only means: the global mode of the broadband histogram does not fall inside a continuous, line-consistent, low/high-balanced trial interval. It does not mean the observation has no usable GTI, does not mean the spectral fit is necessarily bad, and does not mean SP was or was not identified.
readings from actual trialsfixed 20-eV PI gridGehrels χ²
Under the current 1–2 keV model with fixed centroid/width and one power-law continuum per spectrum, a better fit with free line areas does not imply the fluorescent-line physical intensity necessarily changed. After allowing a common Al/Si PI shift (±20 eV) per spectrum, the fitted shift is only about 0–1.3 eV, yet still worse than the free-area model by Δχ²=4.8–15.8; a simple centroid/gain drift is not a sufficient explanation. The free-model χ²ν of 0301 and 0901 remains high, and continuum residuals and controlled shape sensitivity must continue to be checked.
This page publishes only derived diagnostic values; no event, GTI, PHA, RMF, or ARF is published or rewritten, and no new spectral or SP physical conclusion is produced.