为什么要做这个项目Why this project exists
XMM-Newton 的 EPIC 相机（MOS1、MOS2、pn）在观测微弱弥散发射（CGM、halo、ISM、ICM）时，背景不确定度往往主导误差预算。ParetoGTICal 是父项目 DiffuseXrayCalibration 的一个独立校准子项目：它的任务是把"好时间段 (good time interval, GTI) 选择"这个取舍问题，形式化为帕累托优化，并用仪器荧光线做门控，防止"挑噪声"伪装成"删耀斑"。
When XMM-Newton's EPIC cameras (MOS1, MOS2, pn) observe faint diffuse emission (CGM, halo, ISM, ICM), the background uncertainty often dominates the error budget. ParetoGTICal is an independent calibration sub-project of DiffuseXrayCalibration: it formalises the GTI-selection trade-off as a Pareto optimisation, and uses the instrumental fluorescence lines as a gate to stop "cherry-picking noise" from masquerading as "cutting flares".
- XMM-Newton 与 EPIC 背景XMM-Newton and the EPIC background
- 为什么弥散发射最怕背景Why diffuse emission fears the background most
- 背景的成分The background components
- 标准耀斑筛选（ESAS 2.filter）及其代价The standard flare screen (ESAS 2.filter) and its cost
- 为什么这是个帕累托问题Why this is a Pareto problem
- 时间选择永远删不掉稳态软质子底Time selection can never remove a steady soft-proton floor
1. XMM-Newton 与 EPIC 背景1. XMM-Newton and the EPIC background
XMM-Newton 是欧洲航天局的 X 射线天文卫星，有三台 EPIC（European Photon Imaging Camera）相机：两台 MOS（金属-氧化物-半导体 CCD）和一台 pn CCD。它们共用望远镜，但探测器不同，背景表现也不同。任何观测里，除了天体光子，探测器还会记录本底：仪器荧光线（Al-K 1.4879 keV、Si-K 1.7466 keV 等）、宇宙 X 射线背景（CXB）、静止粒子背景（quiescent particle background, QPB）、软质子耀斑（soft-proton flares, SP）以及可能的稳态软质子底。对于点源，背景是可以局部建模扣除的；但对于弥散发射，背景填满整个视场，任何建模偏差都会直接投射到天空亮度上，成为系统误差的主导项。
XMM-Newton is ESA's X-ray observatory with three EPIC (European Photon Imaging Camera) detectors: two MOS (metal-oxide-semiconductor CCD) cameras and one pn CCD. They share the telescope but have different detectors and different background behaviour. Every observation records, besides celestial photons, an instrumental background: fluorescence lines (Al-K 1.4879 keV, Si-K 1.7466 keV), the cosmic X-ray background (CXB), the quiescent particle background (QPB), soft-proton flares (SP), and possibly a steady soft-proton floor. For point sources the background can be modelled locally and subtracted; for diffuse emission the background fills the whole field of view, and any modelling bias projects directly onto the sky surface brightness, becoming the dominant systematic error.
2. 为什么弥散发射最怕背景2. Why diffuse emission fears the background most
点源分析里，背景通常只占源区域的一小部分，且可以从无源区域测得；弥散发射分析里，源本身就是延展的，背景无处不在，且往往比信号还亮（特别是软 X 波段）。1% 的背景建模偏差，对点源可能只是通量的 1% 偏差；对弥散发射，却可能直接变成 10% 甚至更大的表面亮度系统偏差，因为背景与信号在空间上高度重叠。因此，父项目的核心测试是"同一天区在不同 ObsID、不同离轴角、不同 SP/QPB 状态下，给出相同的天空光谱"—— 这要求对背景有极严的校准。
In point-source analysis the background typically occupies a small fraction of the source region and can be measured in source-free areas; in diffuse emission the source itself is extended, the background is everywhere, and often brighter than the signal (especially in the soft X-ray band). A 1% background modelling bias might be a 1% flux bias for a point source, but for diffuse emission it can become a 10% or larger systematic bias in surface brightness, because the background and the signal spatially overlap almost perfectly. That is why the parent project's core test is "the same sky area seen in different ObsIDs, at different off-axis angles, in different SP/QPB states, must yield the same sky spectrum" — which demands an extremely tight background calibration.
3. 背景的成分3. The background components
| 成分Component | 角落（corner）有无In corners | 时间行为Time behaviour | 光谱/空间处理Spectral/spatial treatment |
|---|---|---|---|
| QPB/仪器粒子QPB / instrumental particles | 有present | 可漂移，但角落能描can drift, but traced by the corners | QPB 光谱/线；探测器空间QPB spectrum/lines; detector-space |
| 天体光子Celestial photons | 无absent | 单次观测内通常稳定，但 SWCX 可变normally stable within one observation, but SWCX can vary | 晕影+RMF×ARF 折叠天空模型vignetted, RMF×ARF folded sky model |
| 软质子Soft protons | FOV 相对角落的超额FOV excess over the corners | 可能耀斑，也可能在 GTI 内残留为稳态may flare, or may remain as a stable residual within GTIs | 探测器空间平坦响应 BPL;别叫天体通量detector-space flat-response BPL; do not call it an astrophysical flux |
时间变化本身不证明是 SP:可变 SWCX 是光子反例。候选 SP 信号需要 FOV/角落对比、探测器空间光谱行为、重复指向天空测试三者一致。
Time variability alone does not prove SP: variable SWCX is a photon counterexample. A candidate SP signal requires the FOV/corner contrast, detector-space spectral behaviour, and the matched-aperture repeated-pointing sky test to agree.
4. 标准耀斑筛选（ESAS 2.filter）及其代价4. The standard flare screen (ESAS 2.filter) and its cost
ESAS 标准流程的 2.filter 用整个探测器的高能（2.5–12 keV）计数率光变，做迭代均值/直方图切割，再用 BTI 守卫扩展边缘。它对强耀斑很有效，但也有代价：在一些观测里它丢掉 20% 甚至更多的曝光，而那些被删的时间里，不一定全是软质子——也可能是统计涨落、稳态粒子背景漂移，甚至微弱天体源变化。而且它不看角落，无法区分"FOV 里的超额"与"角落描出的粒子水平"。这就是为什么我们要问：能不能在"保留更多曝光"和"残留污染更少"之间找到更好的平衡？
The ESAS standard 2.filter uses the whole-detector high-energy (2.5–12 keV) count-rate light curve, applies an iterative mean/histogram cut, and expands the edges with a BTI guard. It works well for strong flares, but at a cost: in some observations it discards 20% or more of the exposure, and the discarded time is not guaranteed to be pure soft protons — it can be statistical fluctuations, steady particle-background drifts, or even faint variable celestial sources. And it never looks at the corners, so it cannot separate "FOV excess" from "corner-traced particle level". That is why we ask: can we find a better balance between "keep more exposure" and "leave less contamination"?
5. 为什么这是个帕累托问题5. Why this is a Pareto problem
在单次观测内，每个候选 GTI 给出两个坐标：保留活时间比例 f_T 和连续谱约减因子 R = C_all/C_K。R 越大说明删掉的时间里连续谱越强（删得越"有价值"）；f_T 越大说明保留的数据越多。这两个方向是对立的：想要 R 大就得多删时间，f_T 就小。所有候选在 (f_T, R) 平面上构成一个点云，帕累托前沿就是那些"没有别的候选同时在两个坐标上都超过它"的点。前沿上的每个点都是一种合理的取舍；不在前沿的点是被支配的（既删得多又删得不值）。这就是帕累托优化的几何意义。
Within one observation, every candidate GTI yields two coordinates: the retained live-time fraction f_T and the continuum reduction factor R = C_all/C_K. A larger R means the discarded time carried more continuum (the cut was more "worthwhile"); a larger f_T means more data retained. These two directions pull against each other: to get a higher R you must cut more time, so f_T drops. All candidates form a cloud in the (f_T, R) plane; the Pareto front is the set of points that are not simultaneously beaten in both coordinates by any other candidate. Every point on the front is a defensible trade-off; points off the front are dominated (they lose more exposure for less continuum gain, or vice versa). That is the geometric meaning of Pareto optimisation here.
6. 时间选择永远删不掉稳态软质子底6. Time selection can never remove a steady soft-proton floor
这是本项目的一个硬边界。如果软质子在整个观测里是恒定的（没有耀斑，只是一个稳定的额外连续谱），它在每个时间段的强度都一样，GTI 选也好、不选也好，它都在。R = C_all/C_K 里的 C_all 和 C_K 都包含这个底，所以 R 根本看不见它。GTI 成功不能等于"删掉了稳态软质子底"。GTI 只能删掉时变的部分；稳态底必须靠光谱建模（父项目的响应折叠拟合）来处理。这也是为什么下游成功判据（D002 第 3 步）必须由父项目的响应折叠拟合给出。
This is a hard boundary of the project. If soft protons are constant throughout an observation (no flares, just a steady extra continuum), they have the same strength in every time bin; whether the GTI keeps or rejects a bin, the steady component is always there. Both C_all and C_K in R = C_all/C_K contain that floor, so R is blind to it. GTI success cannot mean "the steady soft-proton floor was removed". A GTI can only remove the time-variable part; the steady floor must be handled by spectral modelling (the parent project's response-folded fit). That is why the downstream success criterion (D002 step 3) must come from the parent project's response-folded fit.