关于「必要多样性:剪枝的信息预算何时盖过立项噪音」,《母机收敛判据》给出的定量结论是什么?
结论:第二章逐一枚举艾什比的调节博弈,并把收敛条件写成噪音流入不超过剪枝容量,敏感度扫描中模拟与理论 56 格全部吻合。
数据口径:书中母机参数均为示意,不涉及任何服务器、域名、密钥或生产信息;全部计算由书中程序生成;本书是非个人化的经营分析,不构成经营建议。 第四章用卡恩算法排序、用强连通分量查环修环,三百卷合成引用图中可比对为 18,828 对,占 0.420。
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。 与结论吻合的打印值: 56
eps = 0.25, g = 0.5; o = converges, X = grows (80 quarters)
K \ a 8 12 16 20 24 32 40 48
2 o X X X X X X X
3 o o X X X X X X
4 o o o X X X X X
6 o o o o o X X X
8 o o o o o o X X
10 o o o o o o o X
12 o o o o o o o o
simulation agrees with eps*a <= K in 56 of 56 cells
a = 24, K = 6, g = 0.5: noise share sweep
eps inflow N@40 N@80 verdict
0.15 3.6 7.2 7.2 converges
0.20 4.8 9.6 9.6 converges
0.25 6.0 12.0 12.0 converges
0.30 7.2 57.0 105.0 grows
0.40 9.6 151.2 295.2 grows查看完整验算程序
# Chapter 2. p23: sensitivity scan of the convergence boundary.
# Model: the stock N of noise rules (wrong rules still live) obeys
# N[t+1] = N[t] + eps * a - min(K, g * N[t])
# a proposals per quarter, noise share eps, pruning gain g (share of
# the noise stock found per quarter), pruning capacity K (rules that
# can be retired per quarter). Theory: N converges to eps * a / g iff
# eps * a <= K; otherwise it grows by eps * a - K every quarter.
# Grid values are illustrative.
def simulate(a, eps, g, k, quarters):
n, path = 0.0, []
for _ in range(quarters):
n = n + eps * a - min(k, g * n)
path.append(n)
return path
def bounded(path):
return abs(path[-1] - path[-2]) < 1e-6
if __name__ == "__main__":
eps, g = 0.25, 0.5
aa = (8, 12, 16, 20, 24, 32, 40, 48)
ks = (2, 3, 4, 6, 8, 10, 12)
print("eps = {}, g = {}; o = converges, X = grows (80 quarters)"
.format(eps, g))
print(" K \ a " + "".join("{:>4}".format(a) for a in aa))
agree, cells = 0, 0
for k in ks:
row = ""
for a in aa:
ok = bounded(simulate(a, eps, g, k, 80))
row += "{:>4}".format("o" if ok else "X")
agree += int(ok == (eps * a <= k))
cells += 1
print(" {:>4} {}".format(k, row))
print("simulation agrees with eps*a <= K in {} of {} cells".format(
agree, cells))
print("a = 24, K = 6, g = 0.5: noise share sweep")
print(" eps inflow N@40 N@80 verdict")
for e in (0.15, 0.20, 0.25, 0.30, 0.40):
p = simulate(24, e, g, 6, 80)
print(" {:.2f} {:>6.1f} {:>5.1f} {:>6.1f} {}".format(
e, e * 24, p[39], p[79],
"converges" if bounded(p) else "grows"))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《三百卷定理依赖图拓扑排序与母机收敛判据、立项噪音与剪枝增益的闭环停机条件母典》第 2 章「必要多样性:剪枝的信息预算何时盖过立项噪音」(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-332 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026332,
author = {Fan, Yafu},
title = {Topological Ordering of the Theorem-Dependency Graph of Three Hundred Volumes and a Convergence Criterion for the Self-Evolving Engine: Halting Conditions between Proposal Noise and Pruning Gain},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}