CGMStacking 工作台CGMStacking workbench

eROSITA 叠加（stacking）—— Zhang et al. 2024（SDSS 中央星系、银河系质量档）与 He & Li 2026（474 颗 L* 星系，15–50 Mpc）——在银河系质量星系周围 10–50 kpc 处测到热星系周介质（circumgalactic medium, CGM）的 X 射线辐射；而我们对手头邻近侧向旋涡星系的 XMM-Newton 逐星分辨测量（分档 4 星系、Vc 标度）只给出 SB/prediction \approx 0.17 \pm 0.10，低了约 5 倍。CGMStacking 要回答的是：是哪一步数据处理差别造成了这个矛盾？做法是把 XMM 档案按 eROSITA 的方式叠加、把我们的逐星测量用同一套差分估计量组合，然后一次只切换一个处理步骤。2026-10-21 起，Chandra 档案作为第二台仪器并行进入（更好的 PSF、更深的点源剔除），并提出新假设：旋转速度（Vrot）比恒星质量（M*）更能预测热 CGM。eROSITA stacks — Zhang et al. 2024 (SDSS central galaxies, Milky-Way-mass bin) and He & Li 2026 (474 L* galaxies at 15–50 Mpc) — report hot circumgalactic-medium (CGM) X-ray emission at 10–50 kpc around Milky-Way-mass galaxies, while our resolved, galaxy-by-galaxy XMM-Newton measurements of nearby edge-on spirals (4 galaxies in the combination, Vc scaling) give SB/prediction \approx 0.17 \pm 0.10, about five times lower. CGMStacking asks: which processing step accounts for this tension? The method is to stack the XMM archive the eROSITA way, combine our per-galaxy measurements with the same differential estimator, and then switch one processing step at a time. Since 2026-10-21 the Chandra archive enters in parallel as a second instrument (better PSF, deeper point-source removal), with a new hypothesis: rotation velocity (Vrot) predicts the hot CGM better than stellar mass (M*).

来源：cognition-os/00_AI_BRIEFING.md、cognition-os/02_CURRENT_QUESTION.md。Source: cognition-os/00_AI_BRIEFING.md, cognition-os/02_CURRENT_QUESTION.md.

目标Objectives

问题Question	状态Status	当前最好数字Best current number	趋向Trend
主问题：eROSITA 式叠加与我们分辨测量之间 \approx 5 倍（10–20 kpc SB/prediction \approx 0.17）由哪一步处理差别造成？Main: which processing difference accounts for the factor \approx 5 between an eROSITA-style stack and our resolved measurements (SB/prediction \approx 0.17 at 10–20 kpc)?	进行中Provisional	分辨叠加 0.17 \pm 0.10（4 星系，Vc 标度；系统误差未传播）。候选原因 6 项，均未检验。Resolved combination 0.17 \pm 0.10 (4 galaxies, Vc scaling; systematics not propagated). Six candidate causes, none tested. 来源：cognition-os/03_TRUST_TABLE.md T02; I001Source: cognition-os/03_TRUST_TABLE.md T02; I001	持平No significant difference
子问题 A：档案 Chandra 叠加（Zhang 式远距离 M* 选样）能否在 50–250 kpc 测到热 CGM？深点源剔除改变什么？Sub-A: can an archival Chandra stack (Zhang-like distant M*-selected sample) measure the hot CGM at 50–250 kpc, and what does deep point-source removal change?	待定Pending	覆盖普查：Chandra 可覆盖约 3.7\times10^{3} 颗 SDSS 银河系质量星系（去掉星系团目标场后约 1.8\times10^{3}）到 150 kpc；S/N 预报 P1 运行中。Census: Chandra covers ~3.7\times10^{3} SDSS MW-mass galaxies (~1.8\times10^{3} outside cluster fields) to 150 kpc; the S/N forecast P1 is running. 来源：I007 (c); RESEARCH_PLAN P1 (pending)Source: I007 (c); RESEARCH_PLAN P1 (pending)	推进中Advancing
子问题 B：旋转速度（Vrot）是否比恒星质量（M）更好地预测热 CGM？Sub-B: does rotation velocity (Vrot) predict the hot CGM better than stellar mass (M)?	待定Pending	普查：直接 Vrot 星表都在 z \lesssim 0.06，Vrot 选样的 Chandra 叠加在 150–250 kpc 只有几十颗；eRASS1 公共数据是第一个检验场（P2 运行中）。Census: direct-Vrot catalogues reach only z \lesssim 0.06; a Vrot-selected Chandra stack has only tens of galaxies at 150–250 kpc; eRASS1 public data is the first test bed (P2 running). 来源：I007 (d), (e); RESEARCH_PLAN P2 (pending)Source: I007 (d), (e); RESEARCH_PLAN P2 (pending)	推进中Advancing

未解决的问题Open problems

问题Problem	为何重要Why it matters	当前最佳假设Best hypothesis	下一个检验Next test	负责人Owner	状态Status
哪一步切换把分辨值推到叠加值，移动多少？Which step moves the resolved value to the stacked value, and by how much?	这是主问题的本体；不分解就无法知道叠加是否偏亮、还是我们偏暗。This is the main question itself; without the decomposition we cannot know whether the stack is biased bright or we are biased low.	盘光被当作 CGM 计入（已被 He & Li 方位角检验削弱，未排除）；Zhang 内区轮廓是去卷积模型；平均 vs 中位数；点源深度。Disk light counted as CGM (weakened by He & Li's azimuthal test, not excluded); Zhang's inner profile is a deconvolved model; mean vs median; point-source depth.	在 AO-26 星系上一次切换一步（环含盘 \(\to\) eRASS 深度点源掩模 \(\to\) eROSITA PSF 卷积 \(\to\) 模型背景 \(\to\) kT \(\to\) 平均方式）。Switch one step at a time on the AO-26 galaxies (annulus incl. disk \(\to\) eRASS-depth mask \(\to\) eROSITA PSF convolution \(\to\) model background \(\to\) kT \(\to\) averaging).	本项目（I001 实验 1–2）this project (I001 steps 1–2)	待定Pending
仪器制造的"晕"有多大（零信号叠加）？How large is the instrument-made "halo" (null-signal stack)?	每个 XMM 目标都落在 MOS CCD1 上；探测器偏置会在叠加里相干累积，jackknife 看不见。Every XMM target sits on MOS CCD1; a detector offset adds coherently in a stack and jackknife does not see it.	pn \approx 0（六个零场 −0.02 \pm 0.06）；MOS 偏置随场而变（九个零场 −0.19 \pm 0.06；常规场 −0.37 \pm 0.08，硬超额场 +0.06 \pm 0.10），原因未定。pn \approx 0 (six nulls −0.02 \pm 0.06); the MOS offset is field-dependent (nine nulls −0.19 \pm 0.06; normal fields −0.37 \pm 0.08, hard-excess +0.06 \pm 0.10), cause open.	固定 CXB 天空；或 NGC 5746 多历元检验（同一天空、不同 QPB/SP）。Fixed-CXB sky; or the NGC 5746 multi-epoch test (same sky, different QPB/SP).	本项目（08_RUNNING §7）this project (08_RUNNING §7)	进行中Provisional
剔除 AGN 与星暴目标后，XMM 档案样本还能用吗？Is the XMM archival sample usable once AGN and starburst targets are separated?	档案偏向亮源；不清洗就无法做无偏叠加。The archive is biased bright; without cleaning no unbiased stack is possible.	15–50 Mpc 内只剩约 15–20 个干净 L* 旋涡（约 1 Ms）。Only about 15–20 clean L* spirals at 15–50 Mpc (\approx 1 Ms) remain.	按 kpc-在-场内设计重选（ADAP 共识表 + 干净曝光筛选）。Re-select with the kpc-in-field design on the ADAP consensus list plus a clean-exposure screen.	本项目（I003）this project (I003)	进行中Provisional
Chandra 能加什么（亚角秒 PSF、更深点源剔除、更小视场）？What does Chandra add (sub-arcsec PSF, deeper point-source removal, smaller field)?	它是第二台仪器，也是点源深度检验的唯一途径。It is the second instrument and the only route to the point-source-depth test.	普查说 Chandra 适合 Zhang 式远距离 M* 选样与点源深度检验，不适合 Vrot 选样叠加。The census says Chandra suits Zhang-like distant M* samples and the point-source-depth test, not a Vrot-selected stack.	P1 S/N 预报与 P3 现实选样；P4 点源深度设计。P1 S/N forecast and P3 realistic sample; P4 point-source-depth design.	P1/P3/P4（运行中）P1/P3/P4 (running)	待定Pending
Vrot 与 M* 谁是更好的 CGM 预测量？Is Vrot or M* the better CGM predictor?	这是 2026-10-21 的新假设，决定后续选样与标度关系。This is the new 2026-10-21 hypothesis and decides the future selection and scaling.	尚无数据；PI 假设 Vrot 更好。No data yet; the PI hypothesises Vrot is better.	eRASS1 公共数据 + HyperLEDA/ALFALFA/FASHI Vrot 星表（P2）；文献 S2。eRASS1 public data with HyperLEDA/ALFALFA/FASHI Vrot catalogues (P2); literature S2.	P2 / S2（运行中）P2 / S2 (running)	待定Pending

来源：cognition-os/02_CURRENT_QUESTION.md；cognition-os/03_TRUST_TABLE.md；cognition-os/insights/I001、I003、I004、I006、I007；cognition-os/RESEARCH_PLAN.md。Source: cognition-os/02_CURRENT_QUESTION.md, cognition-os/03_TRUST_TABLE.md, cognition-os/insights/I001, I003, I004, I006, I007, cognition-os/RESEARCH_PLAN.md.

骨架：项目各部分如何连接Skeleton: how the parts connect

两条证据（eROSITA 叠加 vs 我们分辨的 XMM 测量）汇入矛盾；Chandra 与覆盖普查从旁约束；分解实验是终点。Two evidence lines (eROSITA stacks vs our resolved XMM measurements) feed the tension; Chandra and the coverage census constrain from the side; the decomposition is the goal.

来源：cognition-os/05_PROJECT_DAG.md（mermaid 图改画为 SVG）。Source: cognition-os/05_PROJECT_DAG.md (mermaid graph redrawn as SVG).

未来 1–4 周计划Plan for the next 1–4 weeks

等 P1/P3/P4 报告并分诊（Chandra S/N 预报、现实选样、点源深度设计）——若 50–150 kpc S/N \ge 3 可行，规划约 20 个场的 Chandra 还原试点。Wait for the P1/P3/P4 reports and triage them (Chandra S/N forecast, realistic sample, point-source-depth design) — if S/N \ge 3 at 50–150 kpc is reachable with realistic cuts, plan a Chandra reduction pilot on ~20 fields.
等 P2/S2 报告：eRASS1 公共数据能否裁决 Vrot vs M*；若能，运行该叠加作为 PI 假设的第一个检验。Wait for the P2/S2 reports: can eRASS1 public data decide Vrot vs M*; if yes, run that stack as the first test of the PI hypothesis.
红队 RT-01（普查）与 RT-02（0.17 \pm 0.10 这个数）分诊后更新信任表与目标表。After the red teams RT-01 (census) and RT-02 (the 0.17 \pm 0.10 number) are triaged, update the trust table and the objectives.
XMM 零场方向（固定 CXB 稳定天空；NGC 5746 多历元）保持停泊，直到 DiffuseXrayCalibration 校准完成。The XMM null-field directions (fixed-CXB stable sky; NGC 5746 multi-epoch) stay parked until DiffuseXrayCalibration is done.

来源：cognition-os/08_RUNNING.md §7；cognition-os/RESEARCH_PLAN.md。Source: cognition-os/08_RUNNING.md §7, cognition-os/RESEARCH_PLAN.md.

本次更新What's new

2026-10-21 05:02 EDT（D007）：Chandra 试点与 Vrot 假设进入范围；普查 I007 完成；研究任务 P1–P4、检索 S1–S2、红队 RT-01/RT-02 与本手册（CGM-H001）派出。所有新任务的结果待定。2026-10-21 05:02 EDT (D007): the Chandra pilot and the Vrot hypothesis entered scope; census I007 done; research tasks P1–P4, searches S1–S2, red teams RT-01/RT-02 and this handbook (CGM-H001) dispatched. All their results are pending.
2026-10-17（D005/D006）：稳定天空重拟合给出 MOS 常规场偏置 −0.27 \pm 0.08（之前 −0.37 \pm 0.08），裁决为"不确定"。2026-10-17 (D005/D006): the stable-sky refit gives a normal-field MOS offset of −0.27 \pm 0.08 (was −0.37 \pm 0.08); verdict ambiguous.
2026-10-16（D003/D004）：pn 泄漏假设在预注册阈值处被拒绝（MOS-only −0.35 \pm 0.11 vs 联合 −0.43 \pm 0.11）。2026-10-16 (D003/D004): the pn-leak hypothesis was rejected at the pre-registered threshold (MOS-only −0.35 \pm 0.11 vs joint −0.43 \pm 0.11).