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

第三章 振幅、「虹吸」与一追一躲

只有相位还不够

第二章的振子只有相位,没有强弱。要讨论「一方增益、一方耗散」,模型必须有一个可以增减的量。最简单的选择是斯图尔特—朗道振子:每个振子是复平面上的一个点 zz,相位是它的辐角,振幅是它的模 |z||z|。单独一个振子会稳定在半径为 μ\sqrt{\mu} 的圆上;耦合把两个点拉向彼此。

推导

非对称扩散耦合的斯图尔特—朗道振子。 żA=(μA+iωA−|zA|2)zA+kAB(zB−zA),\dot z_A = (\mu_A + i\omega_A – |z_A|^2)\, z_A + k_{AB}(z_B – z_A), żB=(μB+iωB−|zB|2)zB+kBA(zA−zB).\dot z_B = (\mu_B + i\omega_B – |z_B|^2)\, z_B + k_{BA}(z_A – z_B). 耦合项改变 |zA|2/2|z_A|^2/2 的速率为 PA=Re(z‾AkAB(zB−zA))P_A = \mathrm{Re}\big(\bar z_A\, k_{AB}(z_B – z_A)\big),称为耦合送给 A 的「功率」。这是模型内部的记账量,不是人身上的能量;把它读作「付出」或「获益」是类比。

第五段程序在七种情形下求解这组方程,报告平均振幅与两个方向的耦合功率。

推导

# Chapter 3. Amplitude as well as phase: two Stuart-Landau oscillators.
#   dzA/dt = (mA + i wA - |zA|^2) zA + kAB (zB - zA)
#   dzB/dt = (mB + i wB - |zB|^2) zB + kBA (zA - zB)
# Alone, each settles on a circle of radius sqrt(m). Coupling is diffusive
# and may be one-sided (kAB != kBA). The term kAB (zB - zA) changes
# |zA|^2 / 2 at the rate P_A = Re(conj(zA) kAB (zB - zA)): the "power" the
# coupling brings to A - a book-keeping quantity of the model, not energy
# in a person. Runge-Kutta 4, step 0.01, 2,000 time units; averages over
# the second half. wA = 0 throughout; dw = wB.
import cmath

def deriv(za, zb, p):
    ma, mb, wa, wb, kab, kba = p
    da = (ma + 1j * wa - abs(za) ** 2) * za + kab * (zb - za)
    db = (mb + 1j * wb - abs(zb) ** 2) * zb + kba * (za - zb)
    return da, db

def run(p, t_end=2000.0, h=0.01):
    za, zb = 0.5 + 0.0j, 0.0 + 0.5j
    n = int(t_end / h)
    ra = rb = pa = pb = 0.0
    cnt = 0
    kab, kba = p[4], p[5]
    for i in range(n):
        k1 = deriv(za, zb, p)
        k2 = deriv(za + h / 2 * k1[0], zb + h / 2 * k1[1], p)
        k3 = deriv(za + h / 2 * k2[0], zb + h / 2 * k2[1], p)
        k4 = deriv(za + h * k3[0], zb + h * k3[1], p)
        za += h / 6 * (k1[0] + 2 * k2[0] + 2 * k3[0] + k4[0])
        zb += h / 6 * (k1[1] + 2 * k2[1] + 2 * k3[1] + k4[1])
        if i >= n // 2:
            ra += abs(za)
            rb += abs(zb)
            pa += (za.conjugate() * kab * (zb - za)).real
            pb += (zb.conjugate() * kba * (za - zb)).real
            cnt += 1
    return ra / cnt, rb / cnt, pa / cnt, pb / cnt

if __name__ == "__main__":
    print("   mB     dw    kAB   kBA    |zA|    |zB|     P_A      P_B")
    cases = ((1.0, 0.5, 0.3, 0.3), (1.0, 0.5, 0.5, 0.1),
             (1.0, 0.5, 0.1, 0.5), (1.0, 4.0, 1.5, 1.5),
             (1.0, 4.0, 2.0, 0.2), (0.25, 0.5, 0.3, 0.3),
             (0.25, 0.5, 0.05, 0.5))
    for mb, dw, kab, kba in cases:
        ra, rb, pa, pb = run((1.0, mb, 0.0, dw, kab, kba))
        print("   {:4.2f}   {:3.1f}   {:4.2f}  {:4.2f}   {:6.4f}  {:6.4f}"
              "  {:+7.4f}  {:+7.4f}".format(mb, dw, kab, kba, ra, rb,
                                            pa + 0.0, pb + 0.0))
    print("\nReading: between equal oscillators (mA = mB = 1) coupling costs")
    print("both sides amplitude, and the follower pays more. With a large")
    print("gap and strong mutual coupling both go to zero (amplitude death).")
    print("Only when the natural amplitudes differ (mB = 0.25) does one side")
    print("gain: the weaker oscillator is pumped up and the stronger pays.")
#    mB     dw    kAB   kBA    |zA|    |zB|     P_A      P_B
#    1.00   0.5   0.30  0.30   0.9305  0.9305  -0.1162  -0.1162
#    1.00   0.5   0.50  0.10   0.9097  0.9781  -0.1427  -0.0414
#    1.00   0.5   0.10  0.50   0.9781  0.9097  -0.0414  -0.1427
#    1.00   4.0   1.50  1.50   0.0000  0.0000  +0.0000  +0.0000
#    1.00   4.0   2.00  0.20   0.4456  0.9103  -0.1591  -0.1419
#    0.25   0.5   0.30  0.30   0.9065  0.5405  -0.1465  +0.0123
#    0.25   0.5   0.05  0.50   0.9874  0.6201  -0.0244  +0.0517
#
# Reading: between equal oscillators (mA = mB = 1) coupling costs
# both sides amplitude, and the follower pays more. With a large
# gap and strong mutual coupling both go to zero (amplitude death).
# Only when the natural amplitudes differ (mB = 0.25) does one side
# gain: the weaker oscillator is pumped up and the stronger pays.

相同的振子之间没有「增益」。 前五行双方的自然振幅相同(μA=μB=1\mu_A = \mu_B = 1)。对称耦合时双方振幅都从 1 降到 0.9305,耦合功率都是 −0.1162:耦合让双方都付出。耦合不对称时,跟随的一方付出更多——A 强烈跟随 B 时(0.50 对 0.10),A 的振幅降到 0.9097、功率 −0.1427,B 只降到 0.9781、功率 −0.0414。两人的功率之和始终为负。

振幅死亡。 第四行频差为 4.0,双方都强烈耦合(1.50 对 1.50),结果两个振子都停了下来,振幅为 0。这是阿伦森、埃尔门特劳特与科佩尔分析过的「振幅死亡」:频率差得太远而又拉得太紧时,两个本来各自振荡的系统互相熄灭 (Aronson et al. 1990)。同样的频差下改成单向的强耦合(2.00 对 0.20),被拉的一方 A 振幅降到 0.4456,B 仍有 0.9103——熄灭的只是跟随者。

只有当双方的自然振幅不同时,才有一方获益。 后两行把 B 的自然振幅设为 0.5(μB=0.25\mu_B = 0.25)。这时 B 的耦合功率为正(+0.0123 与 +0.0517),振幅被抬到 0.5405 与 0.6201;A 付出(−0.1465 与 −0.0244)。在这个模型里,「虹吸」的方向由强弱决定:较弱的一方被较强的一方抬起,较强的一方付出。

非对称程度对振幅与耦合功率的影响(频差 0.5,耦合之和 0.6,双方自然振幅相同)。横轴为 (k_{AB} - k_{BA})/(k_{AB} + k_{BA}):越往右,A 越是单方面跟随。两条功率曲线始终在零以下,虚线是两者之和。

注

这对「寄生关系」的说法意味着什么。 提纲设想的「一方增益、一方耗散」在相同振子的扩散耦合中不会出现:双方都付出,只是多少不同。要得到一方净获益,模型必须加入双方自身强弱的差别。所以在这个模型里,「寄生」不是耦合矩阵不对称的结果,而是双方本来就不相等的结果。这是对一个类比的数学分析,不是对任何真实关系的诊断。

一方追,一方躲

第二章的耦合都是正的:双方都向对方靠拢。耦合也可以是负的——一方向对方靠近,另一方躲开。物理学近年把这种「非互惠」耦合作为一类新的相变来研究:当两个群体一追一躲时,系统可以进入一种持续旋转的「追逐」状态 (Fruchart et al. 2021)。两个相位振子的情形可以精确求解:相位差的方程与第二章相同,仍是 ϕ̇=Δω−(kAB+kBA)sin⁡ϕ\dot\phi = \Delta\omega – (k_{AB} + k_{BA})\sin\phi,只要耦合之和为正并且大于频差,就能锁相,共同频率的公式也不变。不同之处在于,当一个权重为负时,Ω\Omega 不再是 ωA\omega_A 与 ωB\omega_B 的加权平均。

推导

# Chapter 3. When one side is attracted and the other repelled.
# In the phase model of Chapter 2 a coupling may be negative: B moves AWAY
# from A's phase. With phi = thB - thA the pair still obeys
#   dphi/dt = dw - (kAB + kBA) sin(phi),
# so it locks when |dw| <= kAB + kBA (the sum must be positive), and the
# common frequency is again Omega = (kBA wA + kAB wB) / (kAB + kBA). With
# kBA < 0 the weights are no longer both positive: Omega can lie OUTSIDE
# the range [wA, wB] of the natural rhythms - a pursuit that speeds both up.
# Euler steps of 0.001 over 400 time units; wA = 1.0, wB = 1.3.
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)
    for i in range(n):
        if i == n // 2:
            a0, b0 = a, b
        a, b = (a + (wa + kab * math.sin(b - a)) * dt,
                b + (wb + kba * math.sin(a - b)) * dt)
    span = (n - n // 2) * dt
    return (a - a0) / span, (b - b0) / span, (b - a) % (2 * math.pi)

if __name__ == "__main__":
    wa, wb = 1.0, 1.3
    print("   kAB     kBA    sum    freq A   freq B   formula   lag phi")
    for kab, kba in ((0.4, 0.4), (1.0, -0.6), (0.8, -0.45),
                     (-0.3, 0.8), (1.0, -0.9)):
        fa, fb, phi = run(wa, wb, kab, kba)
        k = kab + kba
        form = "{:7.4f}".format((kba * wa + kab * wb) / k) \
            if abs(wb - wa) <= k else "  none "
        print("   {:+4.2f}   {:+5.2f}   {:4.2f}   {:6.4f}   {:6.4f}   {}"
              "   {:5.3f}".format(kab, kba, k, fa, fb, form, phi))
    print("\nReading: with both couplings positive the shared rhythm lies")
    print("between 1.0 and 1.3. When B is repelled (kBA < 0) and A pursues,")
    print("the pair still locks if the sum exceeds the gap 0.3, but the")
    print("common rhythm is faster than either partner alone; when A is")
    print("repelled it is slower. With a sum of 0.1 there is no lock at all.")
#    kAB     kBA    sum    freq A   freq B   formula   lag phi
#    +0.40   +0.40   0.80   1.1500   1.1500    1.1500   0.384
#    +1.00   -0.60   0.40   1.7500   1.7500    1.7500   0.848
#    +0.80   -0.45   0.35   1.6857   1.6857    1.6857   1.030
#    -0.30   +0.80   0.50   0.8200   0.8200    0.8200   0.644
#    +1.00   -0.90   0.10   1.1718   1.4546     none    1.030
#
# Reading: with both couplings positive the shared rhythm lies
# between 1.0 and 1.3. When B is repelled (kBA < 0) and A pursues,
# the pair still locks if the sum exceeds the gap 0.3, but the
# common rhythm is faster than either partner alone; when A is
# repelled it is slower. With a sum of 0.1 there is no lock at all.

双方都靠拢时(0.40 对 0.40),共同节律落在两人之间,为 1.1500。A 追、B 躲时(1.00 对 −0.60,和为 0.40),两人仍然锁相,但共同节律是 1.7500,快于双方各自的 1.0 与 1.3;0.80 对 −0.45 时为 1.6857。反过来 A 躲、B 追时(−0.30 对 0.80),共同节律为 0.8200,慢于双方。耦合之和只有 0.10 时(1.00 对 −0.90),无法锁相,两人各自漂移。

一追一躲的锁定是一种「双方都被带着跑」的状态:共同的节律落在双方的自然节律之外,谁也不能按自己的步调行事。在模型里,这是负权重的必然结果;把它读作某种关系模式,是类比。

小结

加入振幅以后,扩散耦合在相同的振子之间总让双方都付出,跟随的一方付出更多;频差大而耦合强时,双方可以同时熄灭(振幅死亡),单向的强耦合则只熄灭跟随者。只有当双方的自然强弱不同时,才会出现一方净获益——较弱的一方被抬起,较强的一方付出。一追一躲的耦合只要和为正且足够大仍能锁相,但共同节律可以落在双方的自然节律之外:追的一方带着躲的一方一起加速。

参考文献

Aronson, D. G., G. B. Ermentrout, and N. Kopell. 1990. “Amplitude Response of Coupled Oscillators.” Physica D 41 (3): 403–49.
Fruchart, Michel, Ryo Hanai, Peter B. Littlewood, and Vincenzo Vitelli. 2021. “Non-Reciprocal Phase Transitions.” Nature 592: 363–69.

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).