关于「信息论下限:一个报告值几比特,一道题值几比特」,《出生时刻误差与交界翻转》给出的定量结论是什么?

结论:第四、五章用熵、切尔诺夫信息与期望信息增益给出题数下限:正确率八成的题要 7 道,六成的要 67 道。

数据口径:本书只分析排盘规则的结构、误差传播与概率,不做任何个人占断,不宣称任何体系有预测效力;书中参数为示意,全部计算由书中程序生成。 第一章说明,示意模型中凭记忆报告的时刻有 0.2782 恰好落在时辰分界上,不取整时只有 0.0083。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 67

abI(bits)C(bits)n for error <= 0.051/I
0.950.050.71361.198011.4
0.900.100.53100.737031.9
0.800.200.27810.321973.6
0.700.300.11870.1258178.4
0.600.400.02900.02946734.4
0.550.450.00720.0072269138.4
0.300.100.04680.04884021.4
0.500.500.00000.0000never (no information)
查看完整验算程序
# Chapter 4. p42: a past-event question is a noisy channel.
# Two candidate double hours A and B, equal prior. A question is
# answered "yes" with probability a under A and b under B (both
# illustrative; no published study supplies such numbers). Per
# question we print the mutual information I (bits) between the
# candidate and the answer, the Chernoff information C (bits; the
# exponent of the Bayes error), and the exact smallest number n of
# independent questions for which the Bayes error with equal priors,
#   0.5 * sum_k min(Bin(k; n, a), Bin(k; n, b)),
# is at most 0.05. A perfect question has I = 1 and needs n = 1.
import math


def h2(p):
    if p <= 0.0 or p >= 1.0:
        return 0.0
    return -p * math.log2(p) - (1 - p) * math.log2(1 - p)


def mutual(a, b):
    return h2(0.5 * (a + b)) - 0.5 * (h2(a) + h2(b))


def chernoff(a, b):
    def f(lam):
        return math.log2(a ** lam * b ** (1 - lam)
                         + (1 - a) ** lam * (1 - b) ** (1 - lam))
    lo, hi = 0.0, 1.0
    for _ in range(100):
        m1, m2 = lo + (hi - lo) / 3, hi - (hi - lo) / 3
        if f(m1) < f(m2):
            hi = m2
        else:
            lo = m1
    return max(0.0, -f(0.5 * (lo + hi)))


def bayes_error(n, a, b):
    terms = [min(math.comb(n, k) * a ** k * (1 - a) ** (n - k),
                 math.comb(n, k) * b ** k * (1 - b) ** (n - k))
             for k in range(n + 1)]
    return 0.5 * math.fsum(terms)


def n_needed(a, b, target=0.05):
    n = 1
    while bayes_error(n, a, b) > target + 1e-12:
        n += 1
    return n


PAIRS = [(0.95, 0.05), (0.90, 0.10), (0.80, 0.20), (0.70, 0.30),
         (0.60, 0.40), (0.55, 0.45), (0.30, 0.10), (0.50, 0.50)]
print("  a     b     I(bits)  C(bits)   n for error <= 0.05  1/I")
for a, b in PAIRS:
    i = mutual(a, b)
    if i == 0.0:
        print("%.2f  %.2f   %7.4f  %7.4f   never (no information)"
              % (a, b, i, chernoff(a, b)))
        continue
    print("%.2f  %.2f   %7.4f  %7.4f   %6d %15.1f"
          % (a, b, i, chernoff(a, b), n_needed(a, b), 1.0 / i))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《出生时刻记忆误差传播与鲁棒定盘、交界时辰盘面翻转概率与最少前事反推题数母典》第 4 章「信息论下限:一个报告值几比特,一道题值几比特」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-284 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026284,
  author    = {Fan, Yafu},
  title     = {Error Propagation from Remembered Birth Times and Robust Chart Fixing: Flip Probabilities at Double-Hour Boundaries and the Fewest Past-Event Questions},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论