5D Library › 百卷知识超图拓扑、多模态个人先验编译与自主知识帝国标度律 · English
第一章 一个方程,许多世界:跨学科同构是什么、不是什么
同一个算子,不同的现象
一百卷书讨论周期、大脑、市场、注意力与组织,却一再遇到同样几块数学。维格纳称数学在自然科学中的效力「不合理地有效」(Wigner 1960);在一个书库内部,这种效力并不那么神秘,因为同样的建模选择被一次又一次地作出。第一块是马尔可夫链:一个只有几种状态、在状态之间以固定机会转移的系统。p01 用同一个函数计算三个示意的链:周期的阶段、顾客的状态与服务器的队列。
推导
# Chapter 1. p01: One operator, three domains. The same code computes
# the long-run distribution of a three-state Markov chain and its mixing
# time (steps until every starting state is within 0.01 of the long run
# in total variation). Illustrative matrices: (i) phases of a cycle
# (expansion, peak, contraction), one step = a month; (ii) a customer
# (active, lapsing, lost), one step = a week; (iii) a server queue
# (idle, busy, overloaded), one step = a second.
DOMAINS = (
("cycle phases", "month",
((0.95, 0.05, 0.00), (0.00, 0.80, 0.20), (0.10, 0.00, 0.90))),
("customer state", "week",
((0.90, 0.10, 0.00), (0.30, 0.50, 0.20), (0.02, 0.00, 0.98))),
("server queue", "second",
((0.60, 0.40, 0.00), (0.30, 0.50, 0.20), (0.00, 0.50, 0.50))),
)
def step(dist, p):
return [sum(dist[i] * p[i][j] for i in range(3)) for j in range(3)]
def long_run(p):
d = [1 / 3] * 3
for _ in range(20000):
d = step(d, p)
return d
def mixing(p, eps=0.01):
pi = long_run(p)
rows = [[1.0 if j == i else 0.0 for j in range(3)] for i in range(3)]
for t in range(1, 100000):
rows = [step(r, p) for r in rows]
if max(0.5 * sum(abs(a - b) for a, b in zip(r, pi))
for r in rows) < eps:
return t
if __name__ == "__main__":
print(" domain unit long-run shares mixing")
for name, unit, p in DOMAINS:
pi = long_run(p)
print(" {:<14} {:<7} {:.3f} {:.3f} {:.3f} {:4d} {}s".format(
name, unit, *pi, mixing(p), unit))
# domain unit long-run shares mixing
# cycle phases month 0.571 0.143 0.286 28 months
# customer state week 0.313 0.063 0.625 73 weeks
# server queue second 0.349 0.465 0.186 7 seconds代码完全相同,答案却不同。周期有 0.571 的月份处于扩张,28 个月后混合;顾客在长期中有 0.625 的周处于流失状态,需要 73 周才「忘记」起点;队列 7 秒就混合。同构共享的是计算的形式,而不是机制,也不是时间尺度。能在领域之间迁移的是方法(怎样计算长期份额与混合时间);不能迁移的是任何一个数字。
两种尺度上的锁相
第二块反复出现的数学,是描述受周期信号驱动的振子的阿德勒方程 (Adler 1946),它也是藏本模型的两振子特例 (Kuramoto 1984)。p02 把它用于以 40 Hz 驱动的 38 Hz 神经节律,以及受昼夜驱动、周期为 24.2 小时的生物钟。
推导
# Chapter 1. p02: Phase locking (Adler's equation) at two scales. The
# phase difference phi between an oscillator and a periodic drive obeys
# d(phi)/dt = dw - K sin(phi): it locks at sin(phi) = dw / K if |dw| <=
# K, and otherwise slips with beat frequency sqrt(dw^2 - K^2). (i) A
# 38 Hz neural rhythm driven at 40 Hz (time in seconds); (ii) a
# circadian clock of 24.2 hours driven by a 24-hour light-dark cycle
# (time in hours). Coupling strengths K are illustrative. The same
# Euler integration checks each case: the drift of phi per unit time
# over the second half of the run.
from math import asin, pi, sin, sqrt
CASES = (
("40 Hz drive, K 15", 2 * pi * (40 - 38), 15.0, 1e-4),
("40 Hz drive, K 8", 2 * pi * (40 - 38), 8.0, 1e-4),
("24 h light, K 0.01", 2 * pi * (1 / 24 - 1 / 24.2), 0.01, 0.05),
("24 h light, K 0.001", 2 * pi * (1 / 24 - 1 / 24.2), 0.001, 0.05),
)
def simulate(dw, k, dt, cycles=200):
phi, t = 0.0, 0.0
t_end = cycles * 2 * pi / dw
mid = None
while t < t_end:
phi += (dw - k * sin(phi)) * dt
t += dt
if mid is None and t >= t_end / 2:
mid = (phi, t)
return (phi - mid[0]) / (t - mid[1])
if __name__ == "__main__":
print(" case dw/K locks lag deg beat "
"simulated")
for name, dw, k, dt in CASES:
r = dw / k
lag = "{:6.1f}".format(asin(r) * 180 / pi) if r <= 1 else " --"
beat = 0.0 if r <= 1 else sqrt(dw * dw - k * k)
print(" {:<20} {:5.2f} {:<5} {} {:8.4f} {:9.4f}".format(
name, r, "yes" if r <= 1 else "no", lag, beat,
simulate(dw, k, dt)))
print(" ratio of the two drive frequencies: {:,.0f}".format(
40 * 3600 * 24))
# case dw/K locks lag deg beat simulated
# 40 Hz drive, K 15 0.84 yes 56.9 0.0000 0.0000
# 40 Hz drive, K 8 1.57 no -- 9.6909 9.6853
# 24 h light, K 0.01 0.22 yes 12.5 0.0000 0.0000
# 24 h light, K 0.001 2.16 no -- 0.0019 0.0019
# ratio of the two drive frequencies: 3,456,000耦合足够强时(失谐与耦合之比为 0.84 与 0.22),两者都锁定,相位滞后分别是 56.9 度与 12.5 度;耦合较弱时(1.57 与 2.16),两者都以公式预测的拍频滑动,模拟结果吻合(第一种情形每秒 9.6853,公式为 9.6909)。两种驱动的频率相差 3,456,000 倍。同构在方程层面是精确的;它并不告诉我们某个具体的大脑或身体处于锁定还是滑动状态,那必须分别测量。
假说 · 尚未验证
作者的诠释(尚未验证): 本书库中至少三分之二的卷,其主要计算可以用第三章的十五个核心原语中的一个或多个,加上至多一个领域专用算子写出。什么会证明它错: 一位独立读者拿到原语清单与各卷程序后,为超过三分之一的卷标注了不止一个领域专用算子。
小结
跨学科同构是共享的计算形式。同样的马尔可夫与锁相代码,可以原封不动地用于时间尺度相差一百万倍以上的领域;每个领域必须自己提供的是参数,而参数从不迁移。
参考文献
Fan, Y. (2026). Hundred-Volume Knowledge Hypergraphs, Multimodal Personal Prior Compilation and Autonomous Empire Scaling: Cross-Disciplinary Isomorphisms, Zero-Marginal-Cost Replication and Century-Scale Knowledge Preservation [百卷知识超图拓扑、多模态个人先验编译与自主知识帝国标度律] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090260 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).