5D Library › 双人相位耦合、双线性相容张量与时空网格博弈 · English

第二章 锁相、阿诺德舌与「谁迁就谁」

两个节律

第一章的评分是静态的。关系却在时间里展开:两个人各有自己的节律——作息、情绪的起伏、工作与休息的周期——相处时彼此牵引。描述两个节律相互牵引的最简模型,是相位振子:每个人只用一个不断转动的相位 θ\theta 来代表,自身的转速为 ω\omega,再加上对方的牵引。把这套数学用于两个人,是一种类比;本章研究的是类比的数学本身。

推导

非对称的阿德勒—藏本方程。 θ̇A=ωA+kABsin⁡(θB−θA),θ̇B=ωB+kBAsin⁡(θA−θB).\dot\theta_A = \omega_A + k_{AB}\sin(\theta_B – \theta_A), \qquad \dot\theta_B = \omega_B + k_{BA}\sin(\theta_A – \theta_B). kABk_{AB} 是 A 跟随 B 的强度,kBAk_{BA} 是 B 跟随 A 的强度,两者可以不同。令相位差 ϕ=θB−θA\phi = \theta_B – \theta_A,得到 ϕ̇=Δω−(kAB+kBA)sin⁡ϕ,Δω=ωB−ωA.\dot\phi = \Delta\omega – (k_{AB} + k_{BA})\sin\phi, \qquad \Delta\omega = \omega_B – \omega_A . ϕ\phi 停在一个定值(锁相)当且仅当 |Δω|≤kAB+kBA|\Delta\omega| \le k_{AB} + k_{BA}。锁相后两人以共同的频率 Ω=kBAωA+kABωBkAB+kBA\Omega = \frac{k_{BA}\,\omega_A + k_{AB}\,\omega_B}{k_{AB} + k_{BA}} 转动;不能锁相时,ϕ\phi 以拍频 Δω2−(kAB+kBA)2\sqrt{\Delta\omega^2 – (k_{AB}+k_{BA})^2} 不断滑移。

这个方程是阿德勒为注入锁定的电子振荡器写下的 (Adler 1946),也是藏本模型在两个振子时的特例 (Kuramoto 1984; Strogatz 2000)。第 44 本讨论过许多振子的同步相变;这里只有两个人,但允许两个方向的耦合不同。

推导

# Chapter 2. Two phase oscillators with one-sided coupling (Adler / Kuramoto).
#   dthA/dt = wA + kAB sin(thB - thA)      kAB: how strongly A follows B
#   dthB/dt = wB + kBA sin(thA - thB)      kBA: how strongly B follows A
# With phi = thB - thA:  dphi/dt = (wB - wA) - (kAB + kBA) sin(phi).
# Locking (phi constant) needs |wB - wA| <= kAB + kBA: the width of the
# 1:1 tongue depends only on the SUM of the couplings. When locked both
# turn at Omega = (kBA wA + kAB wB) / (kAB + kBA): the partner who follows
# less sets the common rhythm. When not locked, phi slips at the beat rate
# sqrt(dw^2 - (kAB + kBA)^2). Euler steps of 0.001 over 400 time units.
import math

def run(wa, wb, kab, kba, t_end=400.0, dt=0.001):
    a, b = 0.0, 1.0
    n = int(t_end / dt)
    half = n // 2
    for i in range(n):
        if i == half:
            a0, b0 = a, b
        da = wa + kab * math.sin(b - a)
        db = wb + kba * math.sin(a - b)
        a += da * dt
        b += db * dt
    span = (n - half) * dt
    return (a - a0) / span, (b - b0) / span

if __name__ == "__main__":
    wa, wb = 1.0, 1.3
    print("   wA = 1.0, wB = 1.3, detuning 0.3")
    print("   kAB    kBA    sum   lock   freq A   freq B   predicted")
    for kab, kba in ((0.25, 0.25), (0.45, 0.05), (0.05, 0.45),
                     (0.10, 0.10), (0.18, 0.02)):
        fa, fb = run(wa, wb, kab, kba)
        k = kab + kba
        if abs(wb - wa) <= k:
            pred = "Omega {:.4f}".format((kba * wa + kab * wb) / k)
            lock = "yes"
        else:
            beat = math.sqrt((wb - wa) ** 2 - k ** 2)
            pred = "beat  {:.4f} (gap {:.4f})".format(beat, fb - fa)
            lock = "no"
        print("   {:4.2f}   {:4.2f}   {:4.2f}   {:4s}   {:6.4f}   {:6.4f}"
              "   {}".format(kab, kba, k, lock, fa, fb, pred))
    print("\nReading: (0.45, 0.05) and (0.05, 0.45) lock equally well, but")
    print("in the first A moves almost all the way to B's rhythm and in the")
    print("second B moves to A's. With a sum below 0.3 the pair cannot lock,")
    print("and the measured frequency gap is close to the predicted beat")
    print("rate (the averaging window holds a whole number of slips only")
    print("approximately).")
#    wA = 1.0, wB = 1.3, detuning 0.3
#    kAB    kBA    sum   lock   freq A   freq B   predicted
#    0.25   0.25   0.50   yes    1.1500   1.1500   Omega 1.1500
#    0.45   0.05   0.50   yes    1.2700   1.2700   Omega 1.2700
#    0.05   0.45   0.50   yes    1.0300   1.0300   Omega 1.0300
#    0.10   0.10   0.20   no     1.0392   1.2608   beat  0.2236 (gap 0.2216)
#    0.18   0.02   0.20   no     1.0706   1.2922   beat  0.2236 (gap 0.2216)
#
# Reading: (0.45, 0.05) and (0.05, 0.45) lock equally well, but
# in the first A moves almost all the way to B's rhythm and in the
# second B moves to A's. With a sum below 0.3 the pair cannot lock,
# and the measured frequency gap is close to the predicted beat
# rate (the averaging window holds a whole number of slips only
# approximately).

前三行的耦合之和都是 0.50,大于频差 0.3,所以都锁相。区别在于锁在哪里:对称耦合时共同频率是两者的平均 1.1500;A 强烈跟随 B 时(0.45 对 0.05),共同频率为 1.2700,几乎就是 B 自己的 1.3;反过来(0.05 对 0.45)为 1.0300,几乎就是 A 的 1.0。后两行的耦合之和只有 0.20,无论怎样分配都不能锁相;测得的频率差 0.2216 接近预言的拍频 0.2236(有限的平均时间窗只近似包含整数次滑移)。

两个结论都直接来自方程。第一,锁相的难易只取决于耦合之和:把同样的总耦合在两人之间怎样分配,锁相舌的宽度都不变。第二,不对称决定谁迁就谁:跟随得少的一方决定了共同的节律。用第一章的语言说,耦合的对称部分决定能否同步,反对称部分决定同步的代价由谁承担。

其他节律比:阿诺德舌

两个节律不一定一比一地锁定。一方每转两圈、另一方转一圈,也可以稳定下来。标准的模型是正弦圆映射:每个周期受对方一次强度为 KK 的牵引,

θn+1=θn+W−K2πsin⁡(2πθn).\theta_{n+1} = \theta_n + W – \frac{K}{2\pi}\sin(2\pi\theta_n).

旋转数 ρ=lim⁡(θn−θ0)/n\rho = \lim (\theta_n – \theta_0)/n 是两个节律实际实现的比。在某个区间的 WW 上 ρ\rho 恒等于一个分数 p/qp/q,就称为 p:qp{:}q 锁定;这些区间在 (W,K)(W, K) 平面上张开成舌头的形状,称为阿诺德舌 (Pikovsky et al. 2001)。

推导

# Chapter 2. Locking at other ratios: the sine circle map.
#   th(n+1) = th(n) + W - (K / 2 pi) sin(2 pi th(n))
# is the standard model of a rhythm W driven once per cycle by a partner of
# strength K. The rotation number rho = lim (th(n) - th(0)) / n is the
# realised ratio of the two rhythms. Where rho is a fixed fraction p/q over
# an interval of W, the pair is locked p:q - an Arnold tongue. Exactly, the
# 0:1 tongue is |W| <= K / (2 pi), of width K / pi. A p:q lock is detected
# as an attracting cycle: after 3,000 steps, th(n + q) - th(n) = p to 1e-9.
# The program finds a W inside each tongue (tongues bend away from p/q as K
# grows), measures the width by bisection, and the share of W in
# [0, 1) locked to some p/q with q <= 8 (1,000 values of W).
import math

TWO_PI = 2 * math.pi

def settle(w, k, burn=3000):
    th, c = 0.0, k / TWO_PI
    for _ in range(burn):
        th += w - c * math.sin(TWO_PI * th)
    return th

def lock_ratio(w, k, qmax):
    th, c = settle(w, k), k / TWO_PI
    start = th
    for q in range(1, qmax + 1):
        th += w - c * math.sin(TWO_PI * th)
        jump = th - start
        if abs(jump - round(jump)) < 1e-9:
            return round(jump), q
    return None

def locked_at(w, k, p, q):
    r = lock_ratio(w, k, q)
    return r is not None and r[0] * q == p * r[1]

def tongue_width(k, p, q):
    # tongues bend away from p/q as K grows: first find a point inside
    grid = [p / q + (i - 200) / 400 / q for i in range(401)]
    inside_pts = [w for w in grid if locked_at(w, k, p, q)]
    if not inside_pts:
        return 0.0
    centre = inside_pts[len(inside_pts) // 2]
    edges = []
    for direction in (-1, 1):
        inside, outside = centre, centre + direction * 0.5 / q
        for _ in range(26):
            mid = (inside + outside) / 2
            if locked_at(mid, k, p, q):
                inside = mid
            else:
                outside = mid
        edges.append(inside)
    return edges[1] - edges[0]

def locked_share(k, steps=1000, qmax=8):
    return sum(lock_ratio(i / steps, k, qmax) is not None
               for i in range(steps)) / steps

if __name__ == "__main__":
    print("   K     0:1 width  exact K/pi  1:2 width  1:3 width  2:5 width"
          "  locked")
    for k in (0.5, 0.9, 1.0):
        print("   {:3.1f}   {:8.5f}   {:8.5f}   {:8.5f}   {:8.5f}   {:8.5f}"
              "   {:5.3f}".format(
                  k, tongue_width(k, 0, 1), k / math.pi,
                  tongue_width(k, 1, 2), tongue_width(k, 1, 3),
                  tongue_width(k, 2, 5), locked_share(k)))
    print("\nReading: the 0:1 width matches K/pi. Tongues of higher order")
    print("are far narrower (the 2:5 tongue at K = 0.5 is narrower than the")
    print("search grid and reads 0); all widen as K grows. At K = 1 more")
    print("than half of the line is locked to a ratio with denominator <= 8.")
#    K     0:1 width  exact K/pi  1:2 width  1:3 width  2:5 width  locked
#    0.5    0.15915    0.15915    0.01949    0.00422    0.00000   0.188
#    0.9    0.28648    0.28648    0.06067    0.02286    0.00626   0.466
#    1.0    0.31831    0.31831    0.07396    0.03056    0.01022   0.578
#
# Reading: the 0:1 width matches K/pi. Tongues of higher order
# are far narrower (the 2:5 tongue at K = 0.5 is narrower than the
# search grid and reads 0); all widen as K grows. At K = 1 more
# than half of the line is locked to a ratio with denominator <= 8.

0:10{:}1 舌的宽度精确等于 K/πK/\pi(程序测得 0.15915、0.28648、0.31831,与公式一致)。阶数越高的舌越窄:K=0.5K = 0.5 时 1:21{:}2 舌宽 0.01949,1:31{:}3 舌只有 0.00422,2:52{:}5 舌窄于搜索网格,读作 0。所有的舌都随 KK 变宽;到 K=1K = 1 时,1:21{:}2 舌宽 0.07396,分母不超过 8 的锁定区间合起来占了 WW 轴的 0.578。

正弦圆映射的阿诺德舌:每个色点是一个 (W, K) 组合在分母 q \le 5 的某个节律比上锁定。舌从 K = 0 处的有理数点张开,随耦合增强而变宽。

证据

临界耦合处的「魔鬼阶梯」。 延森、巴克与博尔证明,在 K=1K = 1 时锁定区间的总测度为 1:未锁定的 WW 构成一个测度为零的分形集,其分形维数严格介于 0 与 1 之间,而且这个数值对一大类圆映射是普适的 (Jensen et al. 1983, 1984)。程序中 0.578 的数字只计入分母不超过 8 的舌,其余的测度分布在无穷多条更窄的舌上。

注

这对「节律相配」的说法意味着什么。 在这个模型里,节律比越简单(1:1、1:2),锁定所需的频率匹配越宽松;比越复杂,越需要精确的匹配或很强的耦合。这是数学性质,不是关于任何两个人的结论。真实的人有多个节律,耦合随时间变化,并且受到第四章讨论的噪声干扰。

小结

两个相位振子的非对称耦合可以精确求解:能否锁相只取决于两个耦合强度之和,不对称只决定共同节律偏向哪一方——跟随得少的一方决定节律。不能锁相时,相位差以可以预言的拍频滑移。其他节律比的锁定由正弦圆映射的阿诺德舌描述:0:10{:}1 舌宽精确等于 K/πK/\pi,高阶舌窄得多,所有舌随耦合增强而变宽,在临界耦合处锁定区间几乎覆盖整条轴。

参考文献

Adler, Robert. 1946. “A Study of Locking Phenomena in Oscillators.” Proceedings of the IRE 34 (6): 351–57.
Jensen, Mogens Høgh, Per Bak, and Tomas Bohr. 1983. “Complete Devil’s Staircase, Fractal Dimension, and Universality of Mode-Locking Structure in the Circle Map.” Physical Review Letters 50 (21): 1637–39.
Jensen, Mogens Høgh, Per Bak, and Tomas Bohr. 1984. “Transition to Chaos by Interaction of Resonances in Dissipative Systems. I. Circle Maps.” Physical Review A 30 (4): 1960–69.
Kuramoto, Yoshiki. 1984. Chemical Oscillations, Waves, and Turbulence. Springer.
Pikovsky, Arkady, Michael Rosenblum, and Jürgen Kurths. 2001. Synchronization: A Universal Concept in Nonlinear Sciences. Cambridge University Press.
Strogatz, Steven H. 2000. “From Kuramoto to Crawford: Exploring the Onset of Synchronization in Populations of Coupled Oscillators.” Physica D 143 (1–4): 1–20.

Fan, Y. (2026). Dyadic Phase Coupling, Bilinear Compatibility Tensors and Spatiotemporal Grid Games: Asymmetric Kuramoto Locking, Stopping-Time Separation Hazards and Magic-Square Permutation Operators [双人相位耦合、双线性相容张量与时空网格博弈] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090172 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).