关于「东西方原典考辨:「履霜,坚冰至」与费希尔—蒂皮特的极值分布」,《极值尾部与退单准备金》给出的定量结论是什么?
结论:第七章考辨《周易·坤》初六的「履霜,坚冰至」(重尾上越过阈值后再翻倍的概率约五分之一)与费希尔—蒂皮特 1928 年的极值分布(正态最大值在十万个样本时与极限仍差 0.0109)。
数据口径:书中的订单、退单、请求量与参数均为示意;全部计算由书中程序生成;准备金与服务器冗余的讨论属于非个人化经营与财务方法,不构成理财、会计、税务或投资建议,实际金额与生产变更须由经营者人工决定。 第一章显示,在形状约 0.4 的示意订单尾部上,真实千分之一分位是正态估计的 2.57 倍,万分之一分位是 5.56 倍;样本上的正态估计中位数只有真实值的 0.363,最大的 20 笔订单占去离差平方和的 0.353。
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: 0.0109
| Kolmogorov gap between max-of-n and its limit | |||||
| parent -> limit | 10 | 100 | 1000 | 10000 | 100000 |
| exponential -> Gumbel | 0.0280 | 0.0027 | 0.0003 | 0.0000 | 0.0000 |
| Burr XII -> Frechet 2.5 | 0.0364 | 0.0050 | 0.0009 | 0.0001 | 0.0000 |
| uniform -> Weibull | 0.0280 | 0.0027 | 0.0003 | 0.0000 | 0.0000 |
| normal -> Gumbel | 0.0522 | 0.0272 | 0.0182 | 0.0137 | 0.0109 |
查看完整验算程序
# Chapter 7. p72: Fisher and Tippett's three limits, and how slowly the
# normal gets there.
# Model: exact law of the largest of n draws, F(a_n x + b_n) ** n, set
# against its limit (Fisher and Tippett 1928): exponential parent ->
# Gumbel exp(-exp(-x)); the illustrative Burr XII order law (b = 200,
# c = 2, k = 1.25) -> Frechet exp(-x ** -2.5); uniform -> reversed
# Weibull exp(x) for x < 0; standard normal -> Gumbel, with b_n the
# 1 - 1/n quantile and a_n = 1 / (n phi(b_n)). We print the largest
# gap (Kolmogorov distance) on a fine grid for n = 10 to 100,000.
import math
from statistics import NormalDist
ND = NormalDist()
B, C, K = 200.0, 2.0, 1.25
def burr_q(p):
return B * ((1.0 - p) ** (-1.0 / K) - 1.0) ** (1.0 / C)
def burr_logcdf(x):
return math.log1p(-(1.0 + (x / B) ** C) ** (-K))
def norm_logcdf(x):
return math.log1p(-0.5 * math.erfc(x / math.sqrt(2.0)))
def gap(n, logcdf, limit, grid):
d = 0.0
for x in grid:
lc = logcdf(n, x)
exact = math.exp(n * lc) if lc > -1e300 else 0.0
d = max(d, abs(exact - limit(x)))
return d
g_gum = [-3.0 + 0.0065 * i for i in range(2001)]
g_fre = [0.05 * math.exp(0.0035 * i) for i in range(2001)]
g_wei = [-8.0 + 0.004 * i for i in range(2000)]
def exp_lc(n, x):
y = x + math.log(n)
return math.log1p(-math.exp(-y)) if y > 0 else -1e301
def burr_lc(n, x):
return burr_logcdf(burr_q(1.0 - 1.0 / n) * x)
def unif_lc(n, x):
y = 1.0 + x / n
return math.log(y) if y > 0 else -1e301
def norm_lc(n, x):
bn = ND.inv_cdf(1.0 - 1.0 / n)
an = 1.0 / (n * ND.pdf(bn))
return norm_logcdf(an * x + bn)
cases = (("exponential -> Gumbel", exp_lc,
lambda x: math.exp(-math.exp(-x)), g_gum),
("Burr XII -> Frechet 2.5", burr_lc,
lambda x: math.exp(-x ** -2.5), g_fre),
("uniform -> Weibull", unif_lc, math.exp, g_wei),
("normal -> Gumbel", norm_lc,
lambda x: math.exp(-math.exp(-x)), g_gum))
NS = (10, 100, 1000, 10000, 100000)
print("Kolmogorov gap between max-of-n and its limit")
print(" parent -> limit " + "".join("%9d" % n for n in NS))
for name, lc, lim, grid in cases:
print(" %-24s" % name + "".join("%9.4f" % gap(n, lc, lim, grid)
for n in NS))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《极值理论广义帕累托尾部、超级鲸鱼客单上限与极端退单准备金母典》第 7 章「东西方原典考辨:「履霜,坚冰至」与费希尔—蒂皮特的极值分布」(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-244 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026244,
author = {Fan, Yafu},
title = {Extreme-Value Theory and Generalised Pareto Tails: Whale-Order Ceilings and Reserves for Extreme Chargebacks},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}