图片详情
—
Skip to content

chem_data.py ​

返回章节 · 配套代码说明

复制以下内容并保存为 chem_data.py,使用方法见配套代码说明。

python
# ===========================================================================
# 代码清单 2-4  化学数据集与化学式解析(模块 chem_data)
# ===========================================================================
"""
本章用到的化学数据集。除特别注明者外,数值取自常用教科书与手册中的公认值。

包含:
    1. ELEMENTS        主族 + 部分过渡金属的元素性质表(真实数据)
    2. parse_formula   化学式解析器(支持括号与下标,把 "Al2O3" 变成字典)
    3. describe_formula 把化学式变成一组"化学上有意义"的描述符
    4. BINARY_GAPS     73 种无机固体(含 Si、Ge 两种单质)的带隙(eV,室温附近)
    5. ALKANE_BP       正构烷烃的沸点(°C,1 atm,真实数据)
    6. ARRHENIUS_N2O5  N2O5 分解的速率常数(真实数据)
    7. WATER_VAPOR     水的饱和蒸气压(真实数据,用于 Clausius-Clapeyron)
    8. HAMMETT_SIGMA   取代基常数 σ(定义值)

说明:带隙数值取自文献中反复引用的室温值,不同实验方法、晶型或样品
之间会有零点几 eV 的差异。本章的结论对这一量级的不确定性不敏感,
但读者不应把这些数字当作具有三四位有效数字的精确值。
"""

import math


# ---------------------------------------------------------------------------
# (1) 元素性质表
# ---------------------------------------------------------------------------
# 字段:原子序数 Z、相对原子质量 M、共价半径 r/pm、
#       Pauling 电负性 χ、价层主量子数 n、第一电离能 IE/(kJ/mol)、熔点 T_m/K
# 电负性取 Pauling 标度;共价半径取 Cordero 等 (2008) 的推荐值;
# 对 Mn、Fe 等存在高/低自旋两套半径的元素,这里取低自旋值。

ELEMENTS = {
    #  符号:   (Z,      M,       r,    χ,    n,   IE,   T_m)
    "H":  (1,   1.008,   31,  2.20, 1, 1312,   14.01),
    "Li": (3,   6.94,   128,  0.98, 2,  520,  453.65),
    "Be": (4,   9.012,   96,  1.57, 2,  899, 1560.0),
    "B":  (5,  10.81,    84,  2.04, 2,  801, 2349.0),
    "C":  (6,  12.011,   76,  2.55, 2, 1086, 3823.0),
    "N":  (7,  14.007,   71,  3.04, 2, 1402,   63.15),
    "O":  (8,  15.999,   66,  3.44, 2, 1314,   54.36),
    "F":  (9,  18.998,   57,  3.98, 2, 1681,   53.53),
    "Na": (11, 22.990,  166,  0.93, 3,  496,  370.94),
    "Mg": (12, 24.305,  141,  1.31, 3,  738,  923.0),
    "Al": (13, 26.982,  121,  1.61, 3,  578,  933.47),
    "Si": (14, 28.085,  111,  1.90, 3,  786, 1687.0),
    "P":  (15, 30.974,  107,  2.19, 3, 1012,  317.3),
    "S":  (16, 32.06,   105,  2.58, 3, 1000,  388.36),
    "Cl": (17, 35.45,   102,  3.16, 3, 1251,  171.6),
    "K":  (19, 39.098,  203,  0.82, 4,  419,  336.53),
    "Ca": (20, 40.078,  176,  1.00, 4,  590, 1115.0),
    "Ti": (22, 47.867,  160,  1.54, 4,  658, 1941.0),
    "V":  (23, 50.942,  153,  1.63, 4,  650, 2183.0),
    "Cr": (24, 51.996,  139,  1.66, 4,  653, 2180.0),
    "Mn": (25, 54.938,  139,  1.55, 4,  717, 1519.0),
    "Fe": (26, 55.845,  132,  1.83, 4,  762, 1811.0),
    "Co": (27, 58.933,  126,  1.88, 4,  760, 1768.0),
    "Ni": (28, 58.693,  124,  1.91, 4,  737, 1728.0),
    "Cu": (29, 63.546,  132,  1.90, 4,  745, 1357.77),
    "Zn": (30, 65.38,   122,  1.65, 4,  906,  692.68),
    "Ga": (31, 69.723,  122,  1.81, 4,  579,  302.91),
    "Ge": (32, 72.630,  120,  2.01, 4,  762, 1211.4),
    "As": (33, 74.922,  119,  2.18, 4,  947, 1090.0),
    "Se": (34, 78.971,  120,  2.55, 4,  941,  494.0),
    "Br": (35, 79.904,  120,  2.96, 4, 1140,  265.8),
    "Rb": (37, 85.468,  220,  0.82, 5,  403,  312.46),
    "Sr": (38, 87.62,   195,  0.95, 5,  549, 1050.0),
    "Zr": (40, 91.224,  175,  1.33, 5,  640, 2128.0),
    "Mo": (42, 95.95,   154,  2.16, 5,  684, 2896.0),
    "Ag": (47, 107.868, 145,  1.93, 5,  731, 1234.93),
    "Cd": (48, 112.414, 144,  1.69, 5,  868,  594.22),
    "In": (49, 114.818, 142,  1.78, 5,  558,  429.75),
    "Sn": (50, 118.710, 139,  1.96, 5,  709,  505.08),
    "Sb": (51, 121.760, 139,  2.05, 5,  834,  903.78),
    "Te": (52, 127.60,  138,  2.10, 5,  869,  722.66),
    "I":  (53, 126.904, 139,  2.66, 5, 1008,  386.85),
    "Cs": (55, 132.905, 244,  0.79, 6,  376,  301.59),
    "Ba": (56, 137.327, 215,  0.89, 6,  503, 1000.0),
    "Hf": (72, 178.49,  175,  1.30, 6,  659, 2506.0),
    "W":  (74, 183.84,  162,  2.36, 6,  770, 3695.0),
    "Pb": (82, 207.2,   146,  2.33, 6,  716,  600.61),
    "Bi": (83, 208.980, 148,  2.02, 6,  703,  544.7),
}


def element_property(symbol, name):
    """按字段名取元素性质。name ∈ {Z, M, r, chi, n, IE, Tm}。"""
    idx = {"Z": 0, "M": 1, "r": 2, "chi": 3, "n": 4, "IE": 5, "Tm": 6}[name]
    return ELEMENTS[symbol][idx]


# ---------------------------------------------------------------------------
# (2) 化学式解析器
# ---------------------------------------------------------------------------

def parse_formula(formula):
    """把化学式解析成 {元素: 个数}。支持下标数字、嵌套括号。

    parse_formula("Al2O3")    -> {"Al": 2, "O": 3}
    parse_formula("Ca(OH)2")  -> {"Ca": 1, "O": 2, "H": 2}
    parse_formula("(NH4)2SO4")-> {"N": 2, "H": 8, "S": 1, "O": 4}

    它完成了化学数据处理的第一步:
    把"字符串形式的化学式"翻译成"可以加权求和的元素计数"。
    """
    # 先做分词:元素符号 / 数字 / 括号
    tokens = []
    i = 0
    while i < len(formula):
        ch = formula[i]
        if ch == "(":
            tokens.append("(")
            i += 1
        elif ch == ")":
            tokens.append(")")
            i += 1
        elif ch.isupper():
            sym = ch
            i += 1
            while i < len(formula) and formula[i].islower():
                sym += formula[i]
                i += 1
            if sym not in ELEMENTS:
                raise ValueError("未知元素符号: %r(来自化学式 %r)" % (sym, formula))
            tokens.append(sym)
        elif ch.isdigit():
            num = ""
            while i < len(formula) and formula[i].isdigit():
                num += formula[i]
                i += 1
            tokens.append(int(num))
        else:
            raise ValueError("化学式 %r 中出现了无法识别的字符 %r" % (formula, ch))

    # 用栈处理括号
    def parse_group(pos, depth):
        counts = {}
        while pos < len(tokens):
            tok = tokens[pos]
            if tok == ")":
                if depth == 0:
                    raise ValueError("化学式 %r 中的括号不匹配" % formula)
                return counts, pos + 1
            if tok == "(":
                inner, pos = parse_group(pos + 1, depth + 1)
                # 括号后面可能跟一个数字
                mult = 1
                if pos < len(tokens) and isinstance(tokens[pos], int):
                    mult = tokens[pos]
                    pos += 1
                for k, v in inner.items():
                    counts[k] = counts.get(k, 0) + v * mult
                continue
            # 元素符号
            sym = tok
            pos += 1
            mult = 1
            if pos < len(tokens) and isinstance(tokens[pos], int):
                mult = tokens[pos]
                pos += 1
            counts[sym] = counts.get(sym, 0) + mult
        if depth != 0:
            raise ValueError("化学式 %r 中的括号不匹配" % formula)
        return counts, pos

    counts, _ = parse_group(0, 0)
    return counts


def formula_weight(counts):
    """摩尔质量 /(g/mol)。"""
    return sum(element_property(sym, "M") * n for sym, n in counts.items())


def describe_formula(formula):
    """把化学式变成一组化学上有意义的描述符。

    返回 dict,含:
        d_chi   电负性差(最大值 - 最小值)
        chi_bar 化学计量加权的平均电负性
        r_bar   化学计量加权的平均共价半径 /pm
        n_bar   化学计量加权的平均主量子数
        M       摩尔质量 /(g/mol)
        n_atoms 分子式中的原子总数
        d_ie    第一电离能之差 /(kJ/mol)

    注意 d_chi 用的是"极差"而不是"两种元素之差",
    这样三组分化合物也能算。代价是丢掉了"谁是阳离子"的信息,
    这是描述符设计中常见的取舍(见正文第 2.5.2 节)。
    """
    counts = parse_formula(formula)
    total = sum(counts.values())
    chis = [(element_property(s, "chi"), n) for s, n in counts.items()]
    rs = [(element_property(s, "r"), n) for s, n in counts.items()]
    ns = [(element_property(s, "n"), n) for s, n in counts.items()]
    ies = [(element_property(s, "IE"), n) for s, n in counts.items()]

    chi_vals = [c for c, _ in chis]
    ie_vals = [e for e, _ in ies]

    return {
        "d_chi": max(chi_vals) - min(chi_vals),
        "chi_bar": sum(c * n for c, n in chis) / total,
        "r_bar": sum(r * n for r, n in rs) / total,
        "n_bar": sum(nn * n for nn, n in ns) / total,
        "M": formula_weight(counts),
        "n_atoms": total,
        "d_ie": (max(ie_vals) - min(ie_vals)) / 1000.0,   # 转成 MJ/mol 量级
    }


# ---------------------------------------------------------------------------
# (3) 化合物带隙
# ---------------------------------------------------------------------------
# (化学式, 带隙 / eV, 类别)
# 数值为室温附近的文献常见值。HgTe、HgSe 这类能带反转的半金属没有收录,
# 因为对它们而言"带隙"这一标签本身需要另行定义。

BINARY_GAPS = [
    # 单质与 IV-IV
    ("Si",   1.12, "IV"),
    ("Ge",   0.67, "IV"),
    ("SiC",  2.36, "IV-IV"),
    # III-V
    ("BN",   6.40, "III-V"),
    ("AlN",  6.20, "III-V"),
    ("GaN",  3.44, "III-V"),
    ("InN",  0.64, "III-V"),
    ("BP",   2.00, "III-V"),
    ("AlP",  2.45, "III-V"),
    ("GaP",  2.26, "III-V"),
    ("InP",  1.35, "III-V"),
    ("AlAs", 2.16, "III-V"),
    ("GaAs", 1.42, "III-V"),
    ("InAs", 0.35, "III-V"),
    ("AlSb", 1.58, "III-V"),
    ("GaSb", 0.73, "III-V"),
    ("InSb", 0.17, "III-V"),
    # II-VI
    ("BeO",  10.60, "II-VI"),
    ("BeS",   5.50, "II-VI"),
    ("MgO",   7.80, "II-VI"),
    ("MgS",   4.90, "II-VI"),
    ("MgSe",  4.00, "II-VI"),
    ("MgTe",  3.60, "II-VI"),
    ("ZnO",   3.37, "II-VI"),
    ("ZnS",   3.68, "II-VI"),
    ("ZnSe",  2.70, "II-VI"),
    ("ZnTe",  2.26, "II-VI"),
    ("CdO",   2.20, "II-VI"),
    ("CdS",   2.42, "II-VI"),
    ("CdSe",  1.74, "II-VI"),
    ("CdTe",  1.50, "II-VI"),
    # I-VII 碱金属卤化物
    ("LiF",  13.60, "I-VII"),
    ("NaF",  11.50, "I-VII"),
    ("KF",   10.90, "I-VII"),
    ("RbF",  10.40, "I-VII"),
    ("CsF",   9.90, "I-VII"),
    ("LiCl",  9.40, "I-VII"),
    ("NaCl",  8.50, "I-VII"),
    ("KCl",   8.70, "I-VII"),
    ("RbCl",  8.30, "I-VII"),
    ("CsCl",  8.00, "I-VII"),
    ("LiBr",  8.10, "I-VII"),
    ("NaBr",  7.10, "I-VII"),
    ("KBr",   7.60, "I-VII"),
    ("RbBr",  7.20, "I-VII"),
    ("CsBr",  7.30, "I-VII"),
    ("LiI",   6.10, "I-VII"),
    ("NaI",   5.90, "I-VII"),
    ("KI",    6.20, "I-VII"),
    ("RbI",   5.90, "I-VII"),
    ("CsI",   6.20, "I-VII"),
    # 氧化物
    ("TiO2",  3.20, "oxide"),
    ("SnO2",  3.60, "oxide"),
    ("SiO2",  9.00, "oxide"),
    ("Al2O3", 8.80, "oxide"),
    ("ZrO2",  5.80, "oxide"),
    ("HfO2",  5.80, "oxide"),
    ("V2O5",  2.30, "oxide"),
    ("WO3",   2.70, "oxide"),
    ("MoO3",  3.00, "oxide"),
    ("Fe2O3", 2.20, "oxide"),
    ("Cu2O",  2.10, "oxide"),
    # 3d 过渡金属一氧化物:d 电子强关联体系,后文残差分析的重点
    ("MnO",   3.60, "TMO"),
    ("FeO",   2.40, "TMO"),
    ("CoO",   2.60, "TMO"),
    ("NiO",   3.70, "TMO"),
    ("CuO",   1.20, "TMO"),
    # 硫族化物
    ("PbS",   0.41, "chalcogenide"),
    ("PbSe",  0.27, "chalcogenide"),
    ("PbTe",  0.31, "chalcogenide"),
    ("MoS2",  1.23, "chalcogenide"),
    ("WS2",   1.35, "chalcogenide"),
    ("Ag2S",  1.00, "chalcogenide"),
]


def build_gap_dataset(feature_names=("d_chi", "chi_bar", "r_bar", "M")):
    """构造带隙数据集。

    返回 (formulas, X, y, groups)
        formulas : 化学式列表
        X        : 特征矩阵(列的顺序与 feature_names 一致)
        y        : 带隙 / eV
        groups   : 每个样本所属的"体系"标签,用于按组划分训练/测试集
    """
    formulas, X, y, groups = [], [], [], []
    for formula, gap, category in BINARY_GAPS:
        feats = describe_formula(formula)
        formulas.append(formula)
        X.append([feats[name] for name in feature_names])
        y.append(gap)
        groups.append(category)
    return formulas, X, y, groups


def build_gap_dataset_by_category(categories,
                                  feature_names=("d_chi", "chi_bar", "r_bar", "M")):
    """只取指定类别的子集。例如排除过渡金属氧化物,看模型能好多少。"""
    formulas, X, y, groups = [], [], [], []
    for formula, gap, category in BINARY_GAPS:
        if category not in categories:
            continue
        feats = describe_formula(formula)
        formulas.append(formula)
        X.append([feats[name] for name in feature_names])
        y.append(gap)
        groups.append(category)
    return formulas, X, y, groups


# ---------------------------------------------------------------------------
# (4) 正构烷烃沸点(1 atm)
# ---------------------------------------------------------------------------

ALKANE_BP = [
    # (碳数 n, 名称, 沸点 /°C)
    (1,  "甲烷 methane",            -161.5),
    (2,  "乙烷 ethane",              -88.6),
    (3,  "丙烷 propane",             -42.1),
    (4,  "正丁烷 n-butane",           -0.5),
    (5,  "正戊烷 n-pentane",          36.1),
    (6,  "正己烷 n-hexane",           68.7),
    (7,  "正庚烷 n-heptane",          98.4),
    (8,  "正辛烷 n-octane",          125.7),
    (9,  "正壬烷 n-nonane",          150.8),
    (10, "正癸烷 n-decane",          174.1),
    (11, "正十一烷 n-undecane",       195.9),
    (12, "正十二烷 n-dodecane",       216.3),
    (13, "正十三烷 n-tridecane",      235.4),
    (14, "正十四烷 n-tetradecane",    253.5),
    (15, "正十五烷 n-pentadecane",    270.6),
    (16, "正十六烷 n-hexadecane",     286.8),
    (17, "正十七烷 n-heptadecane",    302.0),
    (18, "正十八烷 n-octadecane",     316.3),
    (19, "正十九烷 n-nonadecane",     329.8),
    (20, "正二十烷 n-eicosane",       342.7),
]


def build_alkane_dataset():
    """返回 (n_list, names, bp_list)。X 就是碳数 n 本身,一维特征。"""
    ns = [row[0] for row in ALKANE_BP]
    names = [row[1] for row in ALKANE_BP]
    bps = [row[2] for row in ALKANE_BP]
    return ns, names, bps


# ---------------------------------------------------------------------------
# (5) Arrhenius:N2O5 气相分解 2 N2O5 -> 4 NO2 + O2
# ---------------------------------------------------------------------------

ARRHENIUS_N2O5 = [
    # (T /K, k /(1/s), 出处标注)
    (298.15, 3.46e-5, "教科书常见值"),
    (308.15, 1.35e-4, "教科书常见值"),
    (318.15, 4.98e-4, "教科书常见值"),
    (328.15, 1.50e-3, "教科书常见值"),
    (338.15, 4.87e-3, "教科书常见值"),
]

# 教材中对同一组数据给出的活化能,作为对照值
ARRHENIUS_N2O5_LITERATURE_EA = 103.6   # kJ/mol


def build_arrhenius_dataset():
    """返回 (T, k, lnk, inv_T, inv_T_milli)。

    inv_T_milli 是 1000/T,这是化学教材里画 Arrhenius 图的习惯做法。
    正文第 2.5.1 节会说明:从数值优化角度看,这只是"缩放"而没有"中心化"。
    """
    Ts = [row[0] for row in ARRHENIUS_N2O5]
    ks = [row[1] for row in ARRHENIUS_N2O5]
    lnk = [math.log(k) for k in ks]
    inv_T = [1.0 / T for T in Ts]
    inv_T_milli = [1000.0 / T for T in Ts]
    return Ts, ks, lnk, inv_T, inv_T_milli


# ---------------------------------------------------------------------------
# (6) Clausius-Clapeyron:水的饱和蒸气压
# ---------------------------------------------------------------------------

WATER_VAPOR = [
    # (T /°C, 饱和蒸气压 /kPa)
    (0,   0.611),
    (10,  1.228),
    (20,  2.339),
    (25,  3.169),
    (30,  4.246),
    (40,  7.384),
    (50, 12.352),
    (60, 19.947),
    (70, 31.200),
    (80, 47.400),
    (90, 70.200),
    (100, 101.325),
]


def build_water_vapor_dataset():
    """返回 (T_K, P_kPa, lnP, 1000/T)。用于演示 Clausius-Clapeyron 线性拟合。"""
    Ts = [c + 273.15 for c, _ in WATER_VAPOR]
    Ps = [p for _, p in WATER_VAPOR]
    lnP = [math.log(p) for p in Ps]
    inv_T_milli = [1000.0 / T for T in Ts]
    return Ts, Ps, lnP, inv_T_milli


# ---------------------------------------------------------------------------
# (7) Hammett 取代基常数 σ(数值取自 Hansch, Leo & Taft, 1991)
# ---------------------------------------------------------------------------
# 定义:25 °C 水中取代苯甲酸的电离,log(K/K0) = ρσ,且按定义取 ρ = 1.000。
# 换言之,σ 本身就是由一个斜率 ρ = 1 的线性关系定义出来的标度。

HAMMETT_SIGMA = [
    # (取代基, σ_meta, σ_para)
    ("H",       0.00,  0.00),
    ("CH3",    -0.07, -0.17),
    ("CH2CH3", -0.07, -0.15),
    ("OCH3",    0.12, -0.27),
    ("OH",      0.12, -0.37),
    ("NH2",    -0.16, -0.66),
    ("N(CH3)2", -0.15, -0.83),
    ("F",       0.34,  0.06),
    ("Cl",      0.37,  0.23),
    ("Br",      0.39,  0.23),
    ("I",       0.35,  0.18),
    ("CH2Cl",   0.11,  0.11),
    ("CN",      0.56,  0.66),
    ("NO2",     0.71,  0.78),
    ("CF3",     0.43,  0.54),
    ("CHO",     0.36,  0.42),
    ("COCH3",   0.38,  0.50),
    ("COOH",    0.37,  0.45),
    ("SO2NH2",  0.46,  0.57),
    ("NH3+",    0.86,  0.60),
    ("C6H5",    0.06, -0.01),
]


def build_hammett_dataset():
    """返回 (substituents, sigma_meta, sigma_para)。"""
    subs = [row[0] for row in HAMMETT_SIGMA]
    meta = [row[1] for row in HAMMETT_SIGMA]
    para = [row[2] for row in HAMMETT_SIGMA]
    return subs, meta, para

《原子智能》· 纸质出版预备版 · PolyAI Team 著