XARTATOMS · M104 热晕 X 射线分析手册XARTATOMS · M104 hot-halo X-ray analysis handbook

1. 机器与角色

1. Machines and their roles

2. 数据、代码与结果的位置

2. Where data, code and results live

项目根:    /u/Lang3/program/XARTATOMS
XMM ODF:   data/ → /u/Lang3/Data/XARTATOMS/XMM/ODF
           0900170101 = 中心指向; 0900170501/0601 = 偏置(R15-45); 0900170701 = 背景观测
提取产品:  data/0900170101/rpc/4.background/bkg_mos*_*/ (obj.pi, back.pi, .arf, .rmf)
Chandra:   M104_Chandra_ACIS/repro/9532_repro/, 9533_repro/ (174 ks 合计)
拟合脚本:  joint_spectrum_fitting_2T_basedon_region_v18.py (基线)
拟合结果:  joint_spectrum_fitting_2T_basedon_region_v18/region*_*
轮廓表:    v18_profile_analysis/v18_complete_parameters.csv
           figure/radial_thermo_profiles_v18/*.csv
联合管线:  chandra_xmm_joint_20261001/ (products/, figures/, RESUME_CHECKPOINT.md)
认知系统:  cognition-os/ (00_AI_BRIEFING, 02_CURRENT_QUESTION, 03_TRUST_TABLE,
           04_DECISION_LOG, PITFALLS, experiments/, insights/)

3. 可复制粘贴的命令（每条都验证过）

3. Copy-paste commands (each verified)

3.1 初始化环境（每次开新终端）

3.1 Initialise the environment (each new terminal)

export HEADAS=/home/huangrui/software/heasoft-6.33.2/x86_64-pc-linux-gnu-libc2.35
source $HEADAS/headas-init.sh
source /home/huangrui/software/miniconda3/envs/ciao-4.18/bin/ciao.sh
cd /u/Lang3/program/XARTATOMS

验证：`python --version` 应打印 Python 3.12.12；`python -c "import sherpa; print(sherpa.__version__)"` 应打印 4.18（在 CIAO 环境里）。注意：CIAO 的 ciao.sh 路径以实际安装为准（v17 脚本内置了同样的加载逻辑，见 joint_spectrum_fitting_2T_basedon_region_v18.py 头部）。

Verify: `python --version` should print Python 3.12.12; `python -c "import sherpa; print(sherpa.__version__)"` should print 4.18 (inside the CIAO env). Note the ciao.sh path depends on the installation (the v17+ scripts embed the same loading logic; see the head of joint_spectrum_fitting_2T_basedon_region_v18.py).

3.2 恢复一个已保存的 v18 拟合会话并打印 kT 与区间

3.2 Restore a saved v18 fit session and print kT with its interval

cd /u/Lang3/program/XARTATOMS
python inspect_fitting_file.py   # 模板：restore 一个 .fitting 会话并列出参数

inspect_fitting_file.py 展示了 `ui.restore(session_file)` 的用法。要打印某个区域的 kT 与 68% 区间，直接读 conf 输出文件（更快、不需要 Sherpa）：

inspect_fitting_file.py shows how `ui.restore(session_file)` works. To print a region's kT with its 68% interval, read the conf output file directly (faster, no Sherpa needed):

cd /u/Lang3/program/XARTATOMS
column -t -s $'\t' joint_spectrum_fitting_2T_basedon_region_v18/regionR2-4_Angle1_SRCspec_1_SrcApec.txt

实测输出（2026-10-15 运行）：kT = 0.7174567，Lower_Bound = 0.6880459，Upper_Bound = 0.7476932（keV）。来源：该文件。

Measured output (run 2026-10-15): kT = 0.7174567, Lower_Bound = 0.6880459, Upper_Bound = 0.7476932 (keV). Source: that file.

3.3 跑一个区域的 v18 拟合（FAST 模式）

3.3 Run the v18 fit for one region (FAST mode)

cd /u/Lang3/program/XARTATOMS
FAST_MODE=1 ipython -- joint_spectrum_fitting_2T_basedon_region_v18.py regionR2-4_Angle1

或用现成脚本 `./run_v18_fit.sh regionR2-4_Angle1`（内容一致）。FAST_MODE 跳过 conf 与绘图（v18 源码第 53–55 行注释；T51：每区约 3 分钟）。区域名来自 `ls data/0900170101/rpc/*-*reg` 对应的名字。

Or use the ready script `./run_v18_fit.sh regionR2-4_Angle1` (identical content). FAST_MODE skips conf and plotting (source comments at lines 53–55 of the v18 script; T51: ~3 min per region). Region names follow `ls data/0900170101/rpc/*-*reg`.

3.4 画 v18 径向 kT profile

3.4 Plot the v18 radial kT profile

cd /u/Lang3/program/XARTATOMS
python plot_temperature_profile_v18.py

该脚本从 `joint_spectrum_fitting_2T_basedon_region_v18/*_SRCspec_1_SrcApec.txt` 读 kT 与 conf 边界（方法 1），从 `*_reg_proj_data.csv` 读 2D 投影（方法 2），输出 `temperature_profile_comparison_v18.png`（脚本第 202–204 行）。更快的方式：直接读现成表 `figure/radial_thermo_profiles_v18/radial_kT_profile_v18_sector_mean.csv`。

The script reads kT and conf bounds from `joint_spectrum_fitting_2T_basedon_region_v18/*_SRCspec_1_SrcApec.txt` (method 1) and the 2D projection from `*_reg_proj_data.csv` (method 2), writing `temperature_profile_comparison_v18.png` (script lines 202–204). Faster: read the ready table `figure/radial_thermo_profiles_v18/radial_kT_profile_v18_sector_mean.csv`.

3.5 列出 Chandra/XMM 联合状态表

3.5 List the Chandra/XMM joint status table

cd /u/Lang3/program/XARTATOMS
/home/huangrui/software/miniconda3/envs/ciao-4.18/bin/python - <<'EOF'
import csv
rows = list(csv.DictReader(open("chandra_xmm_joint_20261001/products/full_temperature_comparison.csv")))
for r in rows:
    print(f"{r['method']:>3} {r['region']:<12} kT={r['kT_kev'] or 'NA':<9} {r['status']}")
EOF

20 行 = 5 方法 × 4 区域。状态列：CLOSED_68_PROFILE_INTERVAL / OPEN_… / MULTIMODAL_… / UNCONSTRAINED_… / PENDING。8 个非 PENDING 单元即 8/20 PASS（2026-10-05 状态）。注：CSV 的 intervals 列内嵌逗号，shell 的 `column -s,` 会对不齐，所以用上面的 python 片段。

20 rows = 5 methods × 4 regions. The status column: CLOSED_68_PROFILE_INTERVAL / OPEN_… / MULTIMODAL_… / UNCONSTRAINED_… / PENDING. The 8 non-PENDING cells are the 8/20 PASS (state of 2026-10-05). Note: the CSV's intervals column embeds commas, so shell `column -s,` misaligns; use the python snippet above.

3.6 临时文件放哪

3.6 Where temp files go

export TMPDIR=/u/Lang2/ai_tmp/<你的名字>   # 永远不要写 /tmp, /var/tmp 或 $HOME

4. 新人必须遵守的规则

4. Rules a newcomer must keep

• 临时文件只写 `/u/Lang2/ai_tmp/`，永远不写 `/tmp`、`/var/tmp` 或 `$HOME`。
• Temp files go only to `/u/Lang2/ai_tmp/`, never `/tmp`, `/var/tmp` or `$HOME`.
• 每条科学结论必须带标签：Trusted / Provisional / Questioned / Deprecated（定义见 cognition-os/SCHEMA.md）。
• Every scientific statement carries a label: Trusted / Provisional / Questioned / Deprecated (definitions in cognition-os/SCHEMA.md).
• 温度梯度主张需要两个独立方法确认才能从 Questioned 升到 Provisional。
• A temperature-gradient claim needs two independent methods before upgrading from Questioned to Provisional.
• D45：不要重启 P1 反馈链。P0 已 NO-GO；不要再跑 exact-footprint pn0/MOS P1 或 post-failure CXB/SP 控制分支。
• D45: do not restart the P1 feedback chain. P0 returned NO-GO; do not run the exact-footprint pn0/MOS P1 or post-failure CXB/SP control branches.
• 只用验证过的 DS9 fk5 区域文件；pie() 仅在 R4–7 生产扇区合同（D30）下允许。
• Use only verified DS9 fk5 region files; pie() is allowed only under the R4–7 production sector contract (D30).
• 参数边界不是误差棒；区间触界要写明（boundary-limited）。
• Parameter bounds are not error bars; intervals hitting a bound must be labelled boundary-limited.
• 几何（区域定义 + overlay 图）必须先过人眼审核，再提取谱（D21/D24/D33）。
• Geometry (region definitions + overlay images) must pass human visual review before spectrum extraction (D21/D24/D33).
• 不要重装/重启 SAS 生产：全局 fkeypar/pget 包装器污染问题未关闭（D43/D44）。
• Do not reinstall/restart SAS production: the global fkeypar/pget wrapper contamination is not closed (D43/D44).

5. 下一步：编号计划

5. Next steps: the numbered plan

顺序遵循 cognition-os/02_CURRENT_QUESTION.md（2026-10-14/15 节）：Q-A 先于 Q-D 温度图。以下按 2026-10-15 04:30Z 的最新状态更新。

The order follows cognition-os/02_CURRENT_QUESTION.md (2026-10-14/15 sections): Q-A precedes the Q-D kT map. Updated to the 2026-10-15 04:30Z state.

第 1 步 — 读已完成 job 的报告（半天）Step 1 — Read the finished job reports (half a day)

• 目标：吸收 Q-A/Q-C/Q-B/GTI/mask 的首批结论。输入：`.agent/jobs/2026-10-14-{fit-reliability-audit,sb-profile-13arcmin,cluster-13arcmin-test,chandra-xmm-joint-status,gti-lightcurve-audit,source-mask-design}/REPORT.*.md` 与 cognition-os 的 I065/I066/B020/E034/E035/T148–T153/D46/D47。方法：通读 + 摘录结论与标签。成功标准：能复述每个 job 的主要发现及其信任标签。先读：`cognition-os/sessions/2026-10-14.md`（注意 02:40Z 撤回）。
• Goal: absorb the first-wave conclusions of Q-A/Q-C/Q-B/GTI/mask. Inputs: the REPORT.*.md files of the six finished jobs (2026-10-14: fit-reliability-audit, sb-profile-13arcmin, cluster-13arcmin-test, chandra-xmm-joint-status, gti-lightcurve-audit, source-mask-design) and cognition-os I065/I066/B020/E034/E035/T148–T153/D46/D47. Method: read and extract conclusions with their trust labels. Success: you can restate each job's main finding and its label. Read first: `cognition-os/sessions/2026-10-14.md` (note the 02:40Z retraction).

第 2 步 — Q-A 全网格注入–恢复（1–2 周，需加速）Step 2 — The Q-A full-grid injection–recovery (1–2 weeks, needs acceleration)

• 目标：认证一个参考温度方法。输入：试点已证明流程可行（R2-4 Angle1，50 实现 × 2 策略：偏差 +0.0046/−0.0027 keV，条件覆盖率 0.76/0.80 — T153）。方法：扩展到三个代表区（强 R2-4、弱 R8-10/R10-13、FoV 边缘 R11-13_Angle2）× kT ∈ {0.5, 0.7, 0.9}；QPB 观测值按其误差模型独立抽样（不能把噪声 QPB 抽样当独立观测）；加背景错设臂（SP ×1.3、CXB ×1.1）；加 MOS1-free vs MOS1-with-extra-SP 对比（I066）。成功：某策略偏差≈0 且覆盖率≈0.68；失败：全部有偏 → 先修背景模型。先读：`.agent/jobs/2026-10-14-fit-reliability-audit/out/sol/pilot/` 与其 validation_protocol。
• Goal: certify one reference temperature method. Inputs: the pilot proved the workflow (R2-4 Angle1, 50 realizations × 2 strategies: biases +0.0046/−0.0027 keV, conditional coverages 0.76/0.80 — T153). Method: extend to three representative regions (strong R2-4, weak R8-10/R10-13, FoV-edge R11-13_Angle2) × kT ∈ {0.5, 0.7, 0.9} keV; sample the observed QPB estimate independently from its error model (a noisy QPB draw is not an independent observation); add a background-misspecification arm (SP ×1.3, CXB ×1.1); add MOS1-free vs MOS1-with-extra-SP tests (I066). Success: one strategy with bias ≈ 0 and coverage ≈ 0.68; failure: all biased → fix the background model first. Read first: `.agent/jobs/2026-10-14-fit-reliability-audit/out/sol/pilot/` and its validation protocol.

第 3 步 — 外环谱拟合方法（Q-C 遗留，2–4 天）Step 3 — The outer-ring spectral method (Q-C remainder, 2–4 days)

• 目标：让 10–14′（FoV 边缘）的谱点可用。13′ 假象已定性（I065），剩下的是方法问题。输入：v18 会话 + E034/E035 的诊断。方法：更宽 kT 范围（≥1.5–2 keV，去掉 1 keV 顶）+ profile-likelihood 误差，或绑定/正则化径向模型，或改用偏置指向（0900170501/0601）的谱。成功：外环 kT 有非触界的 68% 区间；失败：区间仍开放 → 外环标灰。先读：`cognition-os/insights/I065_*.md`（ implication 段）。
• Goal: make the 10–14′ (FoV-edge) spectral points usable. The 13′ artefact is characterised (I065); what remains is a method question. Inputs: the v18 sessions + the E034/E035 diagnoses. Method: a wider kT range (≥1.5–2 keV, removing the 1 keV ceiling) + profile-likelihood errors, or a tied/regularised radial model, or the offset pointings (0900170501/0601). Success: outer-ring kT with a non-boundary 68% interval; failure: interval still open → grey out the outer ring. Read first: `cognition-os/insights/I065_*.md` (the implication paragraph).

第 4 步 — Chandra+XMM 逐仪器 profile（Q-B 新路线，1 周）Step 4 — The Chandra+XMM per-instrument profile (Q-B new route, 1 week)

• 目标：10′ 内的联合温度 profile（D46）。输入：job `2026-10-15-chandra-xmm-perinst-profile`（运行中）+ v2 mask + 旧管线诊断（T152：J1/J2 双分支 = 共享 kT + 各仪器自由 norm 的模型假象）。方法：每仪器各自全环覆盖；面亮度跨仪器联结（交叉标定 0.7–1.3 自由）；kT 0.2–3 keV；profile 网格 ≥40 点；报告 linked 与 unlinked 两种。成功：R2-4/R4-6 联合 68% 区间闭合且与 v18 相容；失败：区间开放/双分支 → 报上限。先读：`.agent/jobs/2026-10-15-chandra-xmm-perinst-profile/BRIEF.md`。
• Goal: the joint kT profile within 10′ (D46). Inputs: job `2026-10-15-chandra-xmm-perinst-profile` (running) + the v2 mask + the old-pipeline diagnosis (T152: the J1/J2 bimodality is a model artefact of shared kT + free per-instrument norms). Method: each instrument its own full-annulus coverage; SB linked across instruments (free cross-calibration 0.7–1.3); kT 0.2–3 keV; profile grids ≥40 points; report both linked and unlinked variants. Success: closed joint 68% intervals for R2-4/R4-6 compatible with v18; failure: open/bimodal intervals → report upper limits. Read first: `.agent/jobs/2026-10-15-chandra-xmm-perinst-profile/BRIEF.md`.

第 5 步 — 2D 强度图（Q-D 前半，1 周）Step 5 — The 2D intensity map (first half of Q-D, 1 week)

• 目标：0.4–1.25 keV 背景扣除马赛克 + 与 v18 谱轮廓交叉核对。输入：job `2d-intensity-map`（运行中）的产品清单 + 现有 ESAS 图像/QPB/SP 产品 + mask v2。方法：Σ(counts − QPB − SP)/Σexposure；点源用 mask v2；曝光 <10% 中位数的像素遮蔽；星系团残余按 E035 建模而不是只遮蔽。成功：SB_image(r)/SB_spectral(r) 随半径近似常数（注意：谱轮廓必须用裸 SB，B020）；失败：比例漂移 → 先解决背景/残余。先读：`.agent/jobs/2026-10-14-2d-intensity-map/BRIEF.md`。
• Goal: a 0.4–1.25 keV background-subtracted mosaic cross-checked against the v18 spectral profile. Inputs: the running job `2d-intensity-map`'s inventory + existing ESAS image/QPB/SP products + mask v2. Method: Σ(counts − QPB − SP)/Σexposure; point sources with mask v2; mask pixels with exposure <10% of the median; model the cluster residual per E035 rather than masking alone. Success: SB_image(r)/SB_spectral(r) roughly constant with radius (note: the spectral profile must use the bare SB, B020); failure: the ratio drifts → fix the background/residual first. Read first: `.agent/jobs/2026-10-14-2d-intensity-map/BRIEF.md`.

第 6 步 — 2D 温度图（Q-D 后半，等第 2 步）Step 6 — 2D temperature map (second half of Q-D, waits for Step 2)

• 目标：按 20% 规则的试点温度图。输入：Step 2 选定的方法 + E030/E031 分区方案。方法：先四类试点（盘 W 强区、北侧弱区、共同支持内 halo、外低曝光区），每区真实谱验收（68% 单区间、两侧 ≤20%、95% 无竞争分支）。成功：至少盘 W 试点通过；失败：北区仍 31–33% → 灰标，不插值。先读：`analysis_plans/m104_temperature_area_20261005/temperature_map_binning_recipe_zh.md`。
• Goal: a pilot temperature map under the 20% rule. Inputs: the method chosen in Step 2 + the E030/E031 binning design. Method: four pilot classes first (disk-W strong, north weak, common-support inner halo, outer low-exposure), each accepted by real spectra (single 68% interval, both sides ≤20%, no competing branch at 95%). Success: at least the disk-W pilot passes; failure: the north still 31–33% → grey it out, do not interpolate. Read first: `analysis_plans/m104_temperature_area_20261005/temperature_map_binning_recipe_zh.md`.