feat(all): NaN 伪收敛否决与 nl_tight 能量回退、emflux 检验 4π 修正、nst 行宽与种子边界修复、gfATO 谱线表接通与网格加密 9216 点、阶段分项统计与跳板机部署

物理修复(400 失败点归因,见 docs/failed400_nan_pseudo_convergence_2026_08_17.md):
- runner: 阶段 converged 后复查 fort.7,含 NaN/Inf 即否决(fort.9 全零伪收敛,
  曾致 261 点误跳过 nl_direct 回退);否决阶段不产出种子,阻断污染传播
- runner: nl_tight 回退——仅能量边际失败时以 CHMAX 收紧 10× 从自身模型续迭代,
  残差降幅 ~10×;execute_tlusty_stage 抽取供主链与回退共用
- conv_check: fort.14 为 Eddington 通量 Hλ,积分需乘 4π 再比 σTeff⁴
  (旧版 ratio 稳定 0.0796=1/4π,全点系统性假阳性)+ 回归测试
- nst_writer: 单行超 80 字符被 TLUSTY 静默截断,IFALI/JALI/TRAD 等从未生效;
  按 75 字符自动换行
- seed_finder: Teff 容忍度改含边界 <=,相邻 5000K 档恢复互为种子 + 回归测试

谱线表与网格:
- 默认线表 gfVIS99 → gfATO(全波段 18-23000Å),TaskSpec.linelist 支持工作流
  级覆盖,节点按需下载(进程互斥锁防并发重复下载 238MB)
- sdB_cno Teff 加密至 5000K 步长,432 → 9216 点;tlusty/synspec 静态二进制更新

统计与部署:
- grid 汇总改按 tlusty_status/synspec_status 分项计数,新增 tlusty_failed/
  synspec_failed/synspec_pending,前端详情页双视图适配
- deploy/fetch_results 支持跳板机 ProxyJump 与 SSH 主连接复用,fetch 新增 --force;
- Docker 构建支持 CARGO_MIRROR/USE_MIRRORS 国内镜像参数;移除 tools/ 拷贝
- 新增 tlusty-synspec-test skill 与 6 篇根因分析/验证文档
This commit is contained in:
fmq
2026-08-17 23:55:26 +08:00
parent 43b82b1ae2
commit b058e66722
54 changed files with 5624 additions and 347 deletions
@@ -0,0 +1,65 @@
# {test_id} / {type} / {grid} / {content}
## 目的
一句话说清这次测试想验证什么。例:复现生产 t60000 冷启动 nl 阶段发散,对比 baseline 与 itek0/iacc0/dpsilg15 修复组合。
## 配置
- 二进制: `dcts/assets/tlusty_static`(或 `synspec_static`
- 网格: `t{teff}_g{logg}_he{loghe}`(如 `t60000_g5.0_he-2`Teff=60000K, logg=5.0, He/H=1e-2
- 链: 冷启动 `lte→nc→nl` / 种子 `seed_nc→nl` / SYNSPEC
- 变体: 与 baseline 不同的关键参数(`ITEK`/`IACC`/`DPSILG`/`ORELAX`/`NITER`
## 输入
- inputs/ 生成方式: 代码生成(`gen_input5.rs`/`nst_writer.rs`)/ 模板修改(留 diff)/ 复制已验证输入
- 关键参数: 例 nc NITER=10, nl NITER=100, DPSILG=1.5
## 执行
- 命令: `bash scripts/run.sh`
- 各阶段: lte(rc=0) → nc(rc=0) → nl(rc=?)
- 耗时: 例 lte 8s + nc 312s + nl 421s
## 结果(TLUSTY
| 阶段 | 迭代 | 末 max_relc | STOP | NaN | 温度结构 |
|---|---|---|---|---|---|
| lte | 0 | — | — | 0 | — |
| nc | 10 | 1.4e0 | 无 | 0 | 物理 |
| nl | 19 | 1.2e16 | STOP iter19 | 0 | 物理(停止时) |
> max_relc 来自 `check_fort9.py`;温度结构来自 `check_temperature.py`DCTS 权威判据)。
## 结果(SYNSPEC,若适用)
| 指标 | 值 |
|---|---|
| 采样点数 | N |
| 波长范围 | ALAM0 ALAST Å |
| NaN/Inf/坏行 | 0 |
| 分段 NaN (EUV/UV/可见/NIR) | 各段计数 |
## 审查结论
- **数值**: ❌/✅(max_relc 是否 < chmax=0.001;有无 STOP/雪崩)
- **物理**: ❌/✅(温度结构三项:表层 < 3×Teff、各深度 ∈ [10,1e8]K、无 NaN
- **结论**: 一句话。**区分数值收敛与物理收敛**——只有后者才是真正成功。
例:
- 数值: ❌ nc 雪崩(9.7→2.3e15), nl STOP(1.23e16)
- 物理: ✅ 停止时温度结构物理(表层 0.85×Teff,无 NaN
- 结论: 精确复现生产失败;修复组合(itek0/iacc0/dpsilg15)使 nl max_relc 降至 48.6 但仍 STOP,未达数值收敛。
## 复现
```bash
cd dcts/test/{test_id}/{type}/{grid}/{content}
bash scripts/run.sh
# 事后审查(TLUSTY:
python3 ../../../../../../.agents/skills/tlusty-synspec-test/scripts/check_fort9.py outputs/nl.9 --teff 60000
python3 ../../../../../../.agents/skills/tlusty-synspec-test/scripts/check_temperature.py outputs/nl.7 --teff 60000
```
## 产物清单(outputs/
| 文件 | 内容 |
|---|---|
| `lte.6/.7/.9/.14` | lte 阶段产物 |
| `nc.6/.7/.9/.14` | nc 阶段产物 |
| `nl.6/.7/.9/.14` | nl 阶段产物(最终模型 nl.7) |
| `{name}.spec/.cont/.iden` | SYNSPEC 光谱产物(若适用) |
@@ -0,0 +1,73 @@
#!/usr/bin/env bash
# ============================================================
# 冷启动链执行模板 (lte → nc → nl) — 由 new_test.sh 拷为 scripts/run.sh
# 规范: docs/testing_workflow_2026_08_11.md §四;references/tlusty-stages.md
# 用法: bash scripts/run.sh [timeout_per_stage]
# ============================================================
set -u
TEST_DIR="$(cd "$(dirname "$0")/.." && pwd)"
TIMEOUT="${1:-1200}" # 每阶段超时(秒)
BIN="/home/fmq/program/tlusty/tl208-s54/dcts/assets/tlusty_static"
RUNTIME_DATA="/home/fmq/program/tlusty/tl208-s54/dcts/assets/data"
cd "$TEST_DIR" || { echo "无法进入 $TEST_DIR"; exit 1; }
# ---------- 执行前检查 ----------
echo "=== [检查] 输入文件 ==="
for f in lte.5 nc.5 nl.5 lte.nst nc.nst nl.nst; do
[ -f "inputs/$f" ] || { echo "❌ 缺输入: inputs/$f"; exit 1; }
done
echo " lte.5 L2: $(sed -n '2p' inputs/lte.5) (应含 T T)"
echo " nc.5 L2: $(sed -n '2p' inputs/nc.5) (应含 F F)"
echo " nl.5 L2: $(sed -n '2p' inputs/nl.5) (应含 F F)"
for s in lte nc nl; do echo " $s.nst L1: $(head -1 inputs/$s.nst)"; done
# 冷启动链不应有 fort.8
[ -f "inputs/fort.8" ] && { echo "❌ 冷启动链不应有 inputs/fort.8lte 阶段 ltgray=T 会删它)"; exit 1; }
mkdir -p run outputs
ln -sfn "$RUNTIME_DATA" run/data
echo " data -> $(readlink run/data)"
# ---------- 准备 run/ ----------
echo "=== [准备] run/ 工作目录 ==="
cp inputs/*.5 inputs/*.nst run/ 2>/dev/null
rm -f run/fort.6 run/fort.7 run/fort.8 run/fort.9 run/fort.14
# ---------- 阶段执行(每阶段产物按阶段重命名) ----------
run_stage() {
local stage="$1"
echo "=== [执行] $stage (timeout ${TIMEOUT}s) ==="
cp "run/$stage.nst" run/nst 2>/dev/null || { echo "❌ 无 run/$stage.nst"; exit 1; }
# 冷启动 lte: 删 fort.8 (ltgray=T 强制灰大气重建)nc/nl: 保留上阶段 fort.7→fort.8
if [ "$stage" = "lte" ]; then rm -f run/fort.8; fi
cd run
timeout "$TIMEOUT" "$BIN" < "$stage.5" > fort.6 2> fort.14
local rc=$?
cd ..
# rc=124 = timeout 杀死(GNU timeout)。结果可能未完成/被截断,写标记供事后审查注意。
if [ "$rc" -eq 124 ]; then
echo " ⚠⚠ $stage TIMEOUT (rc=124, 超过 ${TIMEOUT}s 被 kill —— 迭代可能未完成,产物被截断)"
touch "$TEST_DIR/outputs/${stage}.TIMEOUT"
fi
cp run/fort.6 "$stage.6" 2>/dev/null
[ -f run/fort.7 ] && cp run/fort.7 "$stage.7"
[ -f run/fort.9 ] && cp run/fort.9 "$stage.9"
cp run/fort.14 "$stage.14" 2>/dev/null
[ -f run/fort.7 ] && cp run/fort.7 run/fort.8 # 下阶段种子
echo " $stage rc=$rc"
}
echo "[$(date +%H:%M:%S)] $TEST_DIR 开始"
run_stage lte
run_stage nc
run_stage nl
echo "[$(date +%H:%M:%S)] $TEST_DIR 完成"
# ---------- 归档 ----------
echo "=== [归档] -> outputs/ ==="
for s in lte nc nl; do
[ -f "$s.6" ] && cp "$s.6" "$s.7" "$s.9" "$s.14" outputs/ 2>/dev/null
done
echo "outputs/: $(ls outputs/ | wc -l) 个文件"
echo "=== 完成。请做事后审查(references/physics-checks.md):"
echo " python3 ../../../../../../.agents/skills/tlusty-synspec-test/scripts/check_fort9.py outputs/nl.9"
echo " python3 ../../../../../../.agents/skills/tlusty-synspec-test/scripts/check_temperature.py outputs/nl.7 --teff <Teff>"
@@ -0,0 +1,75 @@
#!/usr/bin/env bash
# ============================================================
# 种子步进链执行模板 (seed_nc [→ nl]) — 由 new_test.sh 拷为 scripts/run.sh
# 规范: docs/testing_workflow_2026_08_11.md §四;references/tlusty-stages.md
# 输入: seed_nc.5 seed_nc.nst fort.8(邻居收敛模型);若跑到 nl 还需 nl.5 nl.nst
# 用法: bash scripts/run.sh [timeout_per_stage]
# ============================================================
set -u
TEST_DIR="$(cd "$(dirname "$0")/.." && pwd)"
TIMEOUT="${1:-1800}" # 种子链 seed_nc 给宽点
BIN="/home/fmq/program/tlusty/tl208-s54/dcts/assets/tlusty_static"
RUNTIME_DATA="/home/fmq/program/tlusty/tl208-s54/dcts/assets/data"
cd "$TEST_DIR" || { echo "无法进入 $TEST_DIR"; exit 1; }
# ---------- 执行前检查 ----------
echo "=== [检查] 输入文件 ==="
for f in seed_nc.5 seed_nc.nst fort.8; do
[ -f "inputs/$f" ] || { echo "❌ 缺输入: inputs/$f"; exit 1; }
done
echo " seed_nc.5 L2: $(sed -n '2p' inputs/seed_nc.5) (应含 F F)"
echo " seed_nc.nst L1: $(head -1 inputs/seed_nc.nst) (应含 NITER=, ICHANG=0, ORELAX=0.3)"
echo " fort.8 NaN行: $(grep -c -iE 'nan|inf|\*{3,}' inputs/fort.8 || true) (应为 0)"
mkdir -p run outputs
ln -sfn "$RUNTIME_DATA" run/data
echo " data -> $(readlink run/data)"
# ---------- 准备 run/ ----------
cp inputs/seed_nc.5 inputs/seed_nc.nst inputs/fort.8 run/ 2>/dev/null
[ -f inputs/nl.5 ] && cp inputs/nl.5 inputs/nl.nst run/ 2>/dev/null
rm -f run/fort.6 run/fort.7 run/fort.8.run run/fort.9 run/fort.14
# seed_nc 阶段种子 = 邻居模型 (inputs/fort.8 已经拷进 run/)
run_stage() {
local stage="$1"
echo "=== [执行] $stage (timeout ${TIMEOUT}s) ==="
cp "run/$stage.nst" run/nst 2>/dev/null || { echo "❌ 无 run/$stage.nst"; exit 1; }
# seed_nc: run/fort.8 已是邻居种子;nl: 上阶段 fort.7→fort.8
if [ "$stage" = "nl" ]; then
[ -f run/seed_nc.7 ] && cp run/seed_nc.7 run/fort.8 || { echo "⚠ 无 seed_nc.7 作 nl 种子"; }
fi
cd run
timeout "$TIMEOUT" "$BIN" < "$stage.5" > fort.6 2> fort.14
local rc=$?
cd ..
# rc=124 = timeout 杀死。种子链 seed_nc 本就慢,更要注意截断。
if [ "$rc" -eq 124 ]; then
echo " ⚠⚠ $stage TIMEOUT (rc=124, 超过 ${TIMEOUT}s 被 kill —— 迭代可能未完成,产物被截断)"
touch "$TEST_DIR/outputs/${stage}.TIMEOUT"
fi
cp run/fort.6 "$stage.6" 2>/dev/null
[ -f run/fort.7 ] && cp run/fort.7 "$stage.7"
[ -f run/fort.9 ] && cp run/fort.9 "$stage.9"
cp run/fort.14 "$stage.14" 2>/dev/null
echo " $stage rc=$rc"
}
echo "[$(date +%H:%M:%S)] $TEST_DIR 开始"
run_stage seed_nc
# nl 仅在 nl.5 存在时执行
if [ -f inputs/nl.5 ]; then
[ -f run/seed_nc.7 ] && cp run/seed_nc.7 run/fort.8
run_stage nl
fi
echo "[$(date +%H:%M:%S)] $TEST_DIR 完成"
# ---------- 归档 ----------
echo "=== [归档] -> outputs/ ==="
for s in seed_nc nl; do
[ -f "$s.6" ] && cp "$s.6" "$s.7" "$s.9" "$s.14" outputs/ 2>/dev/null
done
echo "outputs/: $(ls outputs/ | wc -l) 个文件"
echo "=== 完成。请做事后审查(references/physics-checks.md):"
echo " python3 ../../../../../../.agents/skills/tlusty-synspec-test/scripts/check_fort9.py outputs/seed_nc.9"
echo " python3 ../../../../../../.agents/skills/tlusty-synspec-test/scripts/check_temperature.py outputs/seed_nc.7 --teff <Teff>"
+90
View File
@@ -0,0 +1,90 @@
#!/bin/bash
# ============================================================
# SYNSPEC 执行模板 — 由 new_test.sh 拷为 scripts/run.sh
# 规范: docs/testing_workflow_2026_08_11.md §四;references/synspec-inputs.md
# 输入: fort.5(stdin) fort.55(控制卡) fort.8(收敛大气) + fort.19/data 软链
# 用法: bash scripts/run.sh [timeout_sec]
# 产物: outputs/{name}.spec/.cont/.iden/.lindat + synspec.log
# ============================================================
set -uo pipefail
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
TEST_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
INPUTS="$TEST_DIR/inputs"; RUN="$TEST_DIR/run"; OUTPUTS="$TEST_DIR/outputs"
SYNSPEC="/home/fmq/program/tlusty/tl208-s54/dcts/assets/synspec_static"
RUNTIME_DATA="/home/fmq/program/tlusty/tl208-s54/dcts/assets/data"
TIMEOUT_SEC="${1:-1200}"
OUT_NAME="${OUT_NAME:-spectrum}" # 产物名前缀,可环境变量覆盖
# --- 1. 切 run/,清理上轮 ---
mkdir -p "$RUN" "$OUTPUTS"
cd "$RUN"
rm -f fort.* *.spec *.cont *.iden *.lindat *.log *.err
# --- 2. 复制输入 ---
[ -f "$INPUTS/fort.5" ] || { echo "❌ 缺 inputs/fort.5"; exit 1; }
[ -f "$INPUTS/fort.55" ] || { echo "❌ 缺 inputs/fort.55"; exit 1; }
[ -f "$INPUTS/fort.8" ] || { echo "❌ 缺 inputs/fort.8 (收敛大气)"; exit 1; }
cp "$INPUTS/fort.5" fort.5
cp "$INPUTS/fort.55" fort.55
cp "$INPUTS/fort.8" fort.8
# fort.19 / data: 软链(优先用 inputs 里已有的,否则指 assets
if [ -e "$INPUTS/fort.19" ]; then
ln -sfn "$(readlink -f "$INPUTS/fort.19")" fort.19
else
ln -sfn "$RUNTIME_DATA/gfATO.dat" fort.19 # 全波段默认
fi
ln -sfn "$RUNTIME_DATA" data
# --- 3. 执行前自检 ---
echo "=== 执行前自检 ==="
echo "fort.5 第1行 TEFF/GRAV: $(sed -n '1p' fort.5)"
echo "fort.55 第6行 波长范围: $(sed -n '6p' fort.55)"
echo "fort.8 行数: $(wc -l < fort.8)"
echo "fort.8 NaN/坏行: $(grep -ciE 'nan|inf|\*{3,}' fort.8 || true)"
echo "fort.19 -> $(readlink fort.19)"
echo "data -> $(readlink data)"
echo ""
# --- 4. 执行 SYNSPEC ---
echo "=== 执行 SYNSPEC (timeout ${TIMEOUT_SEC}s) ==="
T0=$(date +%s)
set +e
timeout "$TIMEOUT_SEC" "$SYNSPEC" < fort.5 > synspec.log 2> synspec.err
RC=$?
set -e
T1=$(date +%s)
echo "rc=$RC, elapsed=$((T1-T0))s"
# rc=124 = timeout 杀死。SYNSPEC 大线表全波段易超时,产物可能不完整。
if [ "$RC" -eq 124 ]; then
echo "⚠⚠ SYNSPEC TIMEOUT (rc=124, 超过 ${TIMEOUT_SEC}s 被 kill —— spec 可能不完整/被截断)"
touch "$OUTPUTS/SYNSPEC.TIMEOUT"
fi
# --- 5. 产物重命名 ---
# SYNSPEC: fort.7=spectrum, fort.17=continuum, fort.12=line IDs, fort.11=line data
[ -f fort.7 ] && cp fort.7 "$OUTPUTS/${OUT_NAME}.spec"
[ -f fort.17 ] && cp fort.17 "$OUTPUTS/${OUT_NAME}.cont"
[ -f fort.12 ] && cp fort.12 "$OUTPUTS/${OUT_NAME}.iden"
[ -f fort.11 ] && cp fort.11 "$OUTPUTS/${OUT_NAME}.lindat"
cp synspec.log "$OUTPUTS/synspec.log" 2>/dev/null || true
cp synspec.err "$OUTPUTS/synspec.err" 2>/dev/null || true
# --- 6. 立即数值审查 ---
echo ""
echo "=== 数值审查 ==="
SPEC="$OUTPUTS/${OUT_NAME}.spec"
if [ -f "$SPEC" ]; then
ROWS=$(wc -l < "$SPEC")
FIRST=$(head -1 "$SPEC" | awk '{print $1}')
LAST=$(tail -1 "$SPEC" | awk '{print $1}')
NAN=$(grep -ciE 'nan|inf|\*{3,}' "$SPEC" || true)
echo "spec 行数: $ROWS"
echo "波长范围: $FIRST - $LAST Å"
echo "NaN/Inf/坏行: $NAN $([ "$NAN" = 0 ] && echo '✅' || echo '❌')"
else
echo "!! 未生成 ${OUT_NAME}.specSYNSPEC 失败,rc=$RC"
fi
echo ""
echo "synspec.log 尾 15 行:"
tail -15 "$OUTPUTS/synspec.log" 2>/dev/null || echo "(无 log)"