tests/test_arrhenius.py

"""湿度修正 Arrhenius 与温湿度双因素建模的单元测试（任务 13）。

覆盖需求：
- 10.1：提供多温度且含 RH 时支持 Extended Arrhenius 双因素模型
  （lnk = lnA − Ea/RT + B·RH）拟合。
- 10.2：数据不足以支撑双因素模型时优雅降级为标准 Arrhenius / 单点假设，
  并声明所用模型与原因。
- 10.3：Ea 采用默认假设值（83.14 kJ/mol）时明确标注其为假设值而非实测值。
- 10.4：仅 2 个温度点时计算实测 Ea 但声明双温点回归的统计可信度局限。
- 10.5：Arrhenius / Extended Arrhenius 计算为纯 Python（compute 阶段），不依赖 LLM。

导入路径由 ``tests/conftest.py`` 与仓库根 ``conftest.py`` 统一设置，
故可直接导入顶层 ``skills`` 包。
"""

from __future__ import annotations

import inspect
import math
import sys

import numpy as np
import pytest

from skills.stability.arrhenius import (
    R,
    EA_DEFAULT,
    MODEL_EXTENDED,
    MODEL_STANDARD,
    MODEL_ASSUMED,
    ArrheniusResult,
    analyze_arrhenius,
    fit_standard_arrhenius,
    fit_moisture_arrhenius,
)


# ---------------------------------------------------------------------------
# 合成数据辅助：由已知模型生成 k 值，便于检验参数回收
# ---------------------------------------------------------------------------

def _k_standard(temps_K, ln_A, Ea):
    """标准 Arrhenius：k = exp(lnA − Ea/RT)。"""
    temps = np.asarray(temps_K, dtype=float)
    return np.exp(ln_A - Ea / (R * temps)).tolist()


def _k_extended(temps_K, rh, ln_A, Ea, B):
    """双因素：k = exp(lnA − Ea/RT + B·RH)。"""
    temps = np.asarray(temps_K, dtype=float)
    rh = np.asarray(rh, dtype=float)
    return np.exp(ln_A - Ea / (R * temps) + B * rh).tolist()


def _collinear_rh(temps_K):
    """构造与 1/T 严格仿射相关的 RH 序列（使设计矩阵秩不足）。

    令 RH = a + b·(1/T)，则 [1, 1/T, RH] 三列线性相关，rank < 3。
    选取 a/b 使生成的 RH 落在合理的非负范围内。
    """
    inv = 1.0 / np.asarray(temps_K, dtype=float)
    inv_lo, inv_hi = inv.min(), inv.max()
    # 把 [inv_lo, inv_hi] 线性映射到 [20, 60] %RH。
    if math.isclose(inv_hi, inv_lo):
        return (np.full_like(inv, 40.0)).tolist()
    b = (60.0 - 20.0) / (inv_hi - inv_lo)
    a = 20.0 - b * inv_lo
    return (a + b * inv).tolist()


# 常用真值。
LN_A_TRUE = 30.0
EA_TRUE = 80000.0      # 80 kJ/mol
B_TRUE = 0.04          # 每 %RH 的 lnk 增量

T40, T50, T60 = 313.15, 323.15, 333.15


# ---------------------------------------------------------------------------
# 需求 10.1：双因素（Extended Arrhenius）拟合
# ---------------------------------------------------------------------------

def test_moisture_arrhenius_recovers_parameters():
    """无噪声合成数据下，双因素拟合应精确回收 lnA / Ea / B。"""
    temps = [T40, T40, T60, T60]
    rh = [25.0, 60.0, 25.0, 60.0]
    ks = _k_extended(temps, rh, LN_A_TRUE, EA_TRUE, B_TRUE)

    res = fit_moisture_arrhenius(temps, rh, ks)

    assert res.model == MODEL_EXTENDED
    assert res.is_moisture_model is True
    assert res.B == pytest.approx(B_TRUE, rel=1e-6)
    assert res.Ea == pytest.approx(EA_TRUE, rel=1e-6)
    assert res.ln_A == pytest.approx(LN_A_TRUE, rel=1e-6)
    assert res.ea_is_assumed is False
    assert res.n_temperatures == 2
    assert res.n_rh_levels == 2
    assert res.n_points == 4
    # n=4, 3 参数 → df=1 → 有自由度 → 无统计局限标记。
    assert res.statistical_limitation is False


def test_moisture_arrhenius_with_degrees_of_freedom():
    """更多数据点（n > 3）时双因素模型具自由度，R² 有意义且无统计局限标记。"""
    temps = [T40, T40, T50, T60, T60, T50]
    rh = [25.0, 60.0, 40.0, 25.0, 60.0, 75.0]
    ks = _k_extended(temps, rh, LN_A_TRUE, EA_TRUE, B_TRUE)

    res = fit_moisture_arrhenius(temps, rh, ks)

    assert res.model == MODEL_EXTENDED
    assert res.statistical_limitation is False
    assert res.R2 == pytest.approx(1.0, abs=1e-9)  # 无噪声 → 完美拟合
    assert res.B == pytest.approx(B_TRUE, rel=1e-6)


def test_analyze_selects_extended_when_sufficient():
    """analyze_arrhenius 在温度/湿度水平与点数充足时自动选择双因素模型。"""
    temps = [T40, T40, T60, T60, T50]
    rh = [25.0, 60.0, 25.0, 60.0, 75.0]
    ks = _k_extended(temps, rh, LN_A_TRUE, EA_TRUE, B_TRUE)

    res = analyze_arrhenius(temps, ks, rh_values=rh)

    assert res.model == MODEL_EXTENDED
    assert res.degraded is False
    assert res.B is not None


def test_moisture_arrhenius_rejects_collinear_design():
    """温度与湿度共线（无法分离效应）时双因素拟合抛错以触发降级。"""
    # RH 与 1/T 严格仿射相关 → 设计矩阵 [1, 1/T, RH] 秩不足。
    temps = [T40, T50, T60, T40]
    rh = _collinear_rh(temps)
    ks = _k_extended(temps, rh, LN_A_TRUE, EA_TRUE, B_TRUE)
    with pytest.raises(ValueError):
        fit_moisture_arrhenius(temps, rh, ks)


# ---------------------------------------------------------------------------
# 需求 10.2：降级路径（标准 Arrhenius / 假设值）并声明
# ---------------------------------------------------------------------------

def test_degrade_to_standard_when_rh_levels_insufficient():
    """提供 RH 但仅单一湿度水平 → 降级标准 Arrhenius 并声明原因。"""
    temps = [T40, T50, T60]
    rh = [60.0, 60.0, 60.0]  # 仅 1 个 RH 水平
    ks = _k_standard(temps, LN_A_TRUE, EA_TRUE)

    res = analyze_arrhenius(temps, ks, rh_values=rh)

    assert res.model == MODEL_STANDARD
    assert res.degraded is True
    assert res.B is None
    assert "湿度水平不足" in res.degrade_reason
    # 声明中应说明降级与原因（需求 10.2）。
    assert any("降级" in d for d in res.declarations)


def test_degrade_to_standard_when_points_insufficient():
    """RH 水平足够但点数 < 4（不足以支撑三参数）→ 降级标准 Arrhenius。"""
    temps = [T40, T60, T40]
    rh = [25.0, 60.0, 60.0]  # 2 温度水平、2 RH 水平，但仅 3 点
    ks = _k_standard(temps, LN_A_TRUE, EA_TRUE)

    res = analyze_arrhenius(temps, ks, rh_values=rh)

    assert res.model == MODEL_STANDARD
    assert res.degraded is True
    assert "数据点不足" in res.degrade_reason


def test_degrade_to_standard_on_failed_extended_fit():
    """双因素水平/点数表面充足但设计共线时，analyze 捕获异常并降级标准模型。"""
    # 4 点、表面 2 温度 2 RH，但 RH 与 1/T 共线 → lstsq 秩不足触发降级。
    temps = [T40, T60, T40, T60]
    rh = [25.0, 75.0, 25.0, 75.0]  # RH 与温度一一绑定 → 共线
    ks = _k_standard(temps, LN_A_TRUE, EA_TRUE)

    res = analyze_arrhenius(temps, ks, rh_values=rh)

    assert res.model == MODEL_STANDARD
    assert res.degraded is True
    assert res.B is None


def test_standard_arrhenius_recovers_ea():
    """无 RH 的标准 Arrhenius 应回收实测 Ea，且不标记为假设值。"""
    temps = [T40, T50, T60]
    ks = _k_standard(temps, LN_A_TRUE, EA_TRUE)

    res = analyze_arrhenius(temps, ks)

    assert res.model == MODEL_STANDARD
    assert res.degraded is False
    assert res.Ea == pytest.approx(EA_TRUE, rel=1e-6)
    assert res.ln_A == pytest.approx(LN_A_TRUE, rel=1e-6)
    assert res.ea_is_assumed is False
    assert res.statistical_limitation is False  # n=3 > 2


# ---------------------------------------------------------------------------
# 需求 10.3：Ea 默认假设值标注
# ---------------------------------------------------------------------------

def test_single_temperature_falls_back_to_assumed_ea():
    """仅单一温度水平 → 采用默认假设 Ea 并明确标注为假设值。"""
    temps = [T40, T40, T40]  # 仅 1 个温度水平
    ks = [1e-4, 1.1e-4, 0.9e-4]

    res = analyze_arrhenius(temps, ks)

    assert res.model == MODEL_ASSUMED
    assert res.ea_is_assumed is True
    assert res.degraded is True
    assert res.Ea == pytest.approx(EA_DEFAULT, rel=1e-12)
    assert res.ea_kj_per_mol == pytest.approx(83.14, abs=1e-2)
    assert res.statistical_limitation is True
    # 声明须明确标注「假设」与默认值（需求 10.3）。
    assert any("假设" in d for d in res.declarations)


def test_assumed_ea_respects_custom_default():
    """可传入自定义默认 Ea；仍标注为假设值。"""
    temps = [T50]
    ks = [2e-4]
    custom = 90000.0

    res = analyze_arrhenius(temps, ks, ea_default=custom)

    assert res.model == MODEL_ASSUMED
    assert res.Ea == pytest.approx(custom, rel=1e-12)
    assert res.ea_is_assumed is True


# ---------------------------------------------------------------------------
# 需求 10.4：仅 2 个温度点 → 实测 Ea + 统计局限声明
# ---------------------------------------------------------------------------

def test_two_temperature_points_compute_measured_ea_with_limitation():
    """2 温度点：计算实测 Ea（非假设），但声明无自由度的统计局限。"""
    temps = [T40, T60]
    ks = _k_standard(temps, LN_A_TRUE, EA_TRUE)

    res = analyze_arrhenius(temps, ks)

    assert res.model == MODEL_STANDARD
    assert res.ea_is_assumed is False                      # 实测，非假设
    assert res.Ea == pytest.approx(EA_TRUE, rel=1e-6)      # 双点精确解
    assert res.statistical_limitation is True              # 无自由度
    assert res.n_points == 2
    # 声明须提到统计可信度局限 / 无自由度（需求 10.4）。
    assert any(
        ("自由度" in d) or ("局限" in d) or ("可信度" in d)
        for d in res.declarations
    )


# ---------------------------------------------------------------------------
# 预测与加速因子的物理一致性
# ---------------------------------------------------------------------------

def test_predict_k_matches_known_model():
    """拟合后预测的 k 应与生成数据的真值模型一致。"""
    temps = [T40, T50, T60]
    ks = _k_standard(temps, LN_A_TRUE, EA_TRUE)
    res = fit_standard_arrhenius(temps, ks)

    expected = math.exp(LN_A_TRUE - EA_TRUE / (R * T50))
    assert res.predict_k(T50) == pytest.approx(expected, rel=1e-6)


def test_acceleration_factor_consistent_with_arrhenius():
    """标准模型的加速因子应等于 exp[(Ea/R)(1/T_long − 1/T_acc)]。"""
    temps = [T40, T50, T60]
    ks = _k_standard(temps, LN_A_TRUE, EA_TRUE)
    res = fit_standard_arrhenius(temps, ks)

    af = res.acceleration_factor(T_accelerated=313.15, T_longterm=298.15)
    expected = math.exp((EA_TRUE / R) * (1 / 298.15 - 1 / 313.15))
    assert af == pytest.approx(expected, rel=1e-6)
    assert af > 1.0  # 加速条件速率更高


def test_moisture_model_predict_includes_rh_term():
    """双因素模型预测应随 RH 升高而增大（B > 0 时）。"""
    temps = [T40, T40, T60, T60]
    rh = [25.0, 60.0, 25.0, 60.0]
    ks = _k_extended(temps, rh, LN_A_TRUE, EA_TRUE, B_TRUE)
    res = fit_moisture_arrhenius(temps, rh, ks)

    k_low_rh = res.predict_k(T50, rh_percent=20.0)
    k_high_rh = res.predict_k(T50, rh_percent=80.0)
    assert k_high_rh > k_low_rh


# ---------------------------------------------------------------------------
# 输入校验
# ---------------------------------------------------------------------------

def test_rejects_nonpositive_k():
    with pytest.raises(ValueError):
        analyze_arrhenius([T40, T50], [0.0, 1e-4])


def test_rejects_nonpositive_temperature():
    with pytest.raises(ValueError):
        analyze_arrhenius([0.0, T50], [1e-4, 2e-4])


def test_rejects_mismatched_lengths():
    with pytest.raises(ValueError):
        analyze_arrhenius([T40, T50], [1e-4])


def test_rejects_negative_rh():
    with pytest.raises(ValueError):
        analyze_arrhenius([T40, T50], [1e-4, 2e-4], rh_values=[-5.0, 60.0])


def test_rejects_empty_input():
    with pytest.raises(ValueError):
        analyze_arrhenius([], [])


# ---------------------------------------------------------------------------
# 需求 10.5：纯 Python，compute 阶段不依赖 LLM（design Property 1）
# ---------------------------------------------------------------------------

def test_compute_functions_take_no_service_or_llm_handle():
    """Arrhenius 计算函数签名不得包含 svc / llm 等服务句柄参数。"""
    for fn in (analyze_arrhenius, fit_standard_arrhenius, fit_moisture_arrhenius):
        params = set(inspect.signature(fn).parameters)
        assert "svc" not in params
        assert "llm" not in params
        assert "services" not in params


def test_module_does_not_import_llm():
    """模块源码不得引入任何 LLM / 服务依赖（架构级阻断幻觉计算）。"""
    import skills.stability.arrhenius as arr_mod

    src = inspect.getsource(arr_mod).lower()
    for forbidden in ("import openai", "llm_service", "model_invoker", "llm_providers"):
        assert forbidden not in src


if __name__ == "__main__":  # pragma: no cover
    sys.exit(pytest.main([__file__, "-v"]))


# ---------------------------------------------------------------------------
# 活化能不确定性（SE / 95% CI）与前瞻性 ASAP 设计建议（专家反馈）
# ---------------------------------------------------------------------------

def test_ea_uncertainty_not_estimable_for_two_temperatures():
    """仅 2 个温度点：Ea 可算但标准误/置信区间不可估计（df=0），且明确声明。"""
    res = analyze_arrhenius([298.15, 313.15], [0.0146, 0.1167])
    assert res.model == MODEL_STANDARD
    assert res.Ea_se is None
    assert res.Ea_ci_lower is None and res.Ea_ci_upper is None
    assert res.ea_se_kj_per_mol is None
    assert res.ea_ci_kj_per_mol is None
    # 声明中应提示无法估计不确定性并建议增加温度点。
    joined = " ".join(res.declarations)
    assert "无自由度" in joined or "置信区间" in joined


def test_ea_uncertainty_estimable_for_three_temperatures():
    """≥3 个温度点：给出 Ea 标准误与 95% 置信区间（J/mol 与 kJ/mol 一致）。"""
    res = analyze_arrhenius([298.15, 313.15, 323.15], [0.0146, 0.1167, 0.35])
    assert res.model == MODEL_STANDARD
    assert res.Ea_se is not None and res.Ea_se > 0
    assert res.Ea_ci_lower is not None and res.Ea_ci_upper is not None
    assert res.Ea_ci_lower < res.Ea < res.Ea_ci_upper
    # 3 个不同温度水平 → 非"温度水平受限"。
    assert res.temperature_levels_limited is False
    # kJ 换算与 J 一致。
    assert res.ea_se_kj_per_mol == pytest.approx(res.Ea_se / 1000.0)
    lo, hi = res.ea_ci_kj_per_mol
    assert lo == pytest.approx(res.Ea_ci_lower / 1000.0)
    assert hi == pytest.approx(res.Ea_ci_upper / 1000.0)


def test_two_temperature_levels_with_replicates_flags_limitation():
    """关键科学局限：多批次重复使 df≥1 可算出 CI，但仅 2 个**温度水平**时，
    该 CI 仅反映重复散度、Arrhenius 线性不可检验，须标记 temperature_levels_limited
    并在声明中警示 Ea 不确定性可能被低估。"""
    # 3 批次 × 2 温度水平（25°C / 40°C）= 6 个点，df=4。
    temps = [298.15, 298.15, 298.15, 313.15, 313.15, 313.15]
    ks = [0.0146, 0.0155, 0.0162, 0.1167, 0.122, 0.131]
    res = analyze_arrhenius(temps, ks)

    assert res.model == MODEL_STANDARD
    assert res.n_temperatures == 2
    assert res.n_points == 6
    # CI 数学上可算（df≥1）……
    assert res.Ea_se is not None and res.Ea_ci_lower is not None
    # ……但温度水平受限标记为真，且声明含低估/线性不可检验警示。
    assert res.temperature_levels_limited is True
    joined = " ".join(res.declarations)
    assert "温度水平" in joined
    assert ("低估" in joined) and ("线性" in joined)
    # 仍给出补充第 3 温度水平的前瞻性建议。
    assert any("温度" in s for s in res.design_suggestions)


def test_three_temperature_levels_not_limited():
    """3 个不同温度水平：temperature_levels_limited 为 False（线性可检验）。"""
    temps = [298.15, 313.15, 323.15, 298.15, 313.15, 323.15]
    ks = [0.0146, 0.1167, 0.35, 0.0152, 0.121, 0.36]
    res = analyze_arrhenius(temps, ks)
    assert res.n_temperatures == 3
    assert res.temperature_levels_limited is False


def test_design_suggestions_present_when_degraded():
    """传统 ICH 双温双湿（温湿共线）无法解析 B 时，应给出前瞻性 ASAP 设计建议。"""
    res = analyze_arrhenius([298.15, 313.15], [0.0146, 0.1167], rh_values=[60.0, 75.0])
    assert res.design_suggestions, "降级场景应给出前瞻性实验设计建议"
    joined = " ".join(res.design_suggestions)
    # 应建议补充温度点与温湿度交叉点（ASAP）。
    assert "温度" in joined
    assert "交叉" in joined or "ASAP" in joined


def test_design_suggestions_are_pure_python_no_llm():
    """设计建议为确定性纯 Python 产物：同输入两次调用结果一致。"""
    a = analyze_arrhenius([298.15, 313.15], [0.0146, 0.1167], rh_values=[60.0, 75.0])
    b = analyze_arrhenius([298.15, 313.15], [0.0146, 0.1167], rh_values=[60.0, 75.0])
    assert a.design_suggestions == b.design_suggestions