关于「东西方原典考辨:《管子·国蓄》的敛散与康托罗维奇的解乘数」,《周期配额的路由与倒计时》给出的定量结论是什么?
结论:第七章考辨《管子·国蓄》「岁有凶穰,故谷有贵贱」与康托罗维奇 1939 年的解乘数。
数据口径:示意参数下,线性规划相对先到先用多出 0.1668 到 0.2217,相对按复杂度分档的固定规则只多出 0.41 % 到 5.05 %,大头来自分档本身。 书中参数为示意,不含任何厂商的价格或配额;全部计算由书中程序生成。
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
| draws: lean 11, normal 21, rich 8 | |||
| case | sd(use) | short/lost | wasted/idle |
| grain, granary cap 0.0 | 0.1786 | 0.1100 | 0.0800 |
| grain, granary cap 0.4 | 0.1520 | 0.0700 | 0.0400 |
| grain, granary cap 1.2 | 0.1428 | 0.0600 | 0.0100 |
| quota, backlog 0.0 s 0.0 | 0.1786 | 0.0825 | 0.1100 |
| quota, backlog 0.4 s 0.4 | 0.1520 | 0.0412 | 0.0700 |
| quota, backlog 1.2 s 0.4 | 0.1428 | 0.0103 | 0.0600 |
查看完整验算程序
# Chapter 7. p71: the Guanzi granary next to a quota that cannot keep.
# Model, part 1 (grain): need 1.0 a year; harvests 0.6, 1.0 or 1.4
# (lean, normal, rich) with chances 1/4, 1/2, 1/4, seeded, 40 years. A
# public granary buys the surplus of rich years and sells in lean years
# ("gather when cheap, release when dear"), store capped at C.
# Part 2 (quota): capacity 1.0 a cycle that vanishes at the reset;
# demand 0.6, 1.0 or 1.4 with the same chances. The supply cannot be
# stored, but a share s of the demand can wait: a backlog capped at C
# plays the granary's part with time running the other way. Shares
# are of total need or demand. Parameters illustrative.
import math
import random
import statistics
YEARS = 40
rng = random.Random(20261113)
draws = [rng.choices((0.6, 1.0, 1.4), weights=(1, 2, 1))[0]
for _ in range(YEARS)]
def grain(cap):
stock, short, waste, eat = 0.0, [], [], []
for y in draws:
if y >= 1.0:
put = min(y - 1.0, cap - stock)
stock += put
waste.append(y - 1.0 - put)
eat.append(1.0)
short.append(0.0)
else:
out = min(1.0 - y, stock)
stock -= out
eat.append(y + out)
short.append(1.0 - y - out)
waste.append(0.0)
return eat, math.fsum(short) / YEARS, math.fsum(waste) / YEARS
def quota(cap, s):
back, lost, idle, done = 0.0, [], [], []
for d in draws:
urgent, flex = (1 - s) * d, s * d
u = min(1.0, urgent)
room = 1.0 - u
b = min(room, back)
back -= b
room -= b
f = min(room, flex)
room -= f
keep = min(flex - f, cap - back)
back += keep
lost.append(urgent - u + flex - f - keep)
idle.append(room)
done.append(u + b + f)
tot = math.fsum(draws)
return done, math.fsum(lost) / tot, math.fsum(idle) / YEARS
print("draws: lean %d, normal %d, rich %d" % (
draws.count(0.6), draws.count(1.0), draws.count(1.4)))
print("case sd(use) short/lost wasted/idle")
for cap in (0.0, 0.4, 1.2):
eat, sh, wa = grain(cap)
print("grain, granary cap %.1f %7.4f %10.4f %11.4f" % (
cap, statistics.pstdev(eat), sh, wa))
for cap, s in ((0.0, 0.0), (0.4, 0.4), (1.2, 0.4)):
done, lo, idl = quota(cap, s)
print("quota, backlog %.1f s %.1f %7.4f %10.4f %11.4f" % (
cap, s, statistics.pstdev(done), lo, idl))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《周期配额削峰填谷与跨模型路由、重置倒计时线性规划与重推理轻量模型算力分配母典》第 7 章「东西方原典考辨:《管子·国蓄》的敛散与康托罗维奇的解乘数」(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-270 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026270,
author = {Fan, Yafu},
title = {Peak-Shaving Periodic Quotas and Cross-Model Routing: Linear Programming against Reset Countdowns to Split Work between Heavy-Reasoning and Fast Models},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}