关于「宽恕的安全剂量:宽恕型以牙还牙与入侵窗口」,《依恋博弈与停火协议》给出的定量结论是什么?

结论:宽恕有一个安全窗口。在 eps 为 0.05 的噪音下,宽恕概率应在 0.0437 与 0.2811 之间:太少会被困在回声里,连交替者都守不住;太多会被交替者和无条件背叛者占便宜。噪音越大,窗口越窄,到 eps 为 0.10 时窗口关闭,此时该降低的是误读,而不是调整宽恕。演化模拟复现了「一报还一报开路、宽恕型接班」的四幕,最终占比最大的策略是 (0.988, 0.433)。收益表改变安全剂量:「狠话甜」时上限只有 1/6。下一章要问的是:依恋研究里的焦虑与回避,能不能写成这些参数?

数据口径:书中参数均为示意,全部计算由书中程序生成。 本书不是心理咨询、医疗或法律建议,不宣称任何改变关系或身心状况的效果,也不写任何操控、冷暴力或情感勒索技巧;遇到家庭暴力、威胁、自伤风险或严重心理困扰,请立即联系当地紧急求助渠道与专业机构。

English version

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 标准库与结论吻合的打印值: 0.0437, 0.05, 0.10, 0.2811

eps0.05; GTFT(g) residents against two entrants
g P(CC) self pay self ALLD vs G ALT vs G invaded by
0.0000.250 2.250 1.325 2.453 ALT
0.1000.543 2.668 1.663 2.590 none
0.2000.675 2.790 2.001 2.727 none
0.2500.716 2.822 2.170 2.796 none
0.3000.748 2.847 2.338 2.864 ALT
0.3330.766 2.859 2.451 2.910 ALT
0.4000.794 2.879 2.676 3.002 ALT
0.5000.826 2.901 3.014 3.139 ALLD+ALT
0.7000.868 2.927 3.689 3.413 ALLD+ALT
1.0000.902 2.947 4.702 3.825 ALLD+ALT
largest g safe from ALLD at eps0.05: 0.4646 (noise-free 0.5)
largest g safe from ALT at eps0.05: 0.2811 (noise-free 0.3333)
smallest g safe from ALT at eps0.05: 0.0437 (noise-free 0)
safe forgiveness window by noise (grid of 1/100, then bisection)
eps 0.01: g from 0.008 to 0.325
eps 0.02: g from 0.016 to 0.316
eps 0.05: g from 0.044 to 0.281
eps 0.08: g from 0.087 to 0.227
eps 0.10: empty
查看完整验算程序
# Chapter 2. p21: how much forgiveness, generosity sweep at eps = 0.05.
# Model: the exact chain of p11 with payoffs T=5, R=3, P=1, S=0. GTFT
# with generosity g cooperates after a partner's C and, after a D,
# still cooperates with probability g. For each g we report the payoff
# of GTFT against itself and of two entrants against it: a pure
# defector (ALLD) and an alternator (ALT: C after own D, D after own
# C). An entrant invades when its payoff beats the residents' own.
# Molander's (1985) noise-free bound is g* = min(1 - (T-R)/(R-S),
# (R-P)/(T-P)) = min(1/3, 1/2): ALT sets the first term, ALLD the
# second. g is illustrative.
from fractions import Fraction as F
from functools import reduce
from operator import add

T, R, P, S = 5, 3, 1, 0
PAY = [R, S, T, P]
SWAP = [0, 2, 1, 3]


def fsum_exact(xs):
    return reduce(add, xs, F(0))


def noisy(p, e):
    return (1 - e) * p + e * (1 - p)


def chain(s1, s2, e):
    rows = []
    for k in range(4):
        a = noisy(F(s1[k]), e)
        b = noisy(F(s2[SWAP[k]]), e)
        rows.append([a * b, a * (1 - b), (1 - a) * b, (1 - a) * (1 - b)])
    return rows


def solve(a, y):
    n = len(y)
    m = [row[:] + [y[i]] for i, row in enumerate(a)]
    for k in range(n):
        piv = next(r for r in range(k, n) if m[r][k] != 0)
        m[k], m[piv] = m[piv], m[k]
        for r in range(n):
            if r != k and m[r][k] != 0:
                f = m[r][k] / m[k][k]
                m[r] = [x - f * z for x, z in zip(m[r], m[k])]
    return [m[i][n] / m[i][i] for i in range(n)]


def stationary(mat):
    n = len(mat)
    a = [[mat[j][i] - (1 if i == j else 0) for j in range(n)]
         for i in range(n)]
    a[-1] = [F(1)] * n
    return solve(a, [F(0)] * (n - 1) + [F(1)])


def pay(s1, s2, e):
    pi = stationary(chain(s1, s2, e))
    return fsum_exact(p * v for p, v in zip(pi, PAY)), pi


E = F(1, 20)
ALLD = [0, 0, 0, 0]
ALT = [0, 0, 1, 1]
print("eps = 0.05; GTFT(g) residents against two entrants")
print("    g   P(CC) self  pay self  ALLD vs G  ALT vs G  invaded by")
for g in (F(0), F(1, 10), F(1, 5), F(1, 4), F(3, 10), F(1, 3),
          F(2, 5), F(1, 2), F(7, 10), F(1)):
    rule = [1, g, 1, g]
    ps, pi = pay(rule, rule, E)
    pd, _ = pay(ALLD, rule, E)
    pa, _ = pay(ALT, rule, E)
    who = [n for n, v in (("ALLD", pd), ("ALT", pa)) if v > ps]
    print("%5.3f  %10.3f  %8.3f  %9.3f  %8.3f  %10s"
          % (float(g), float(pi[0]), float(ps), float(pd), float(pa),
             "+".join(who) if who else "none"))


def bound(entrant, lo=F(0), hi=F(1), rising=True):
    # bisect the g at which the entrant stops (or starts) winning
    for _ in range(40):
        mid = (lo + hi) / 2
        rule = [1, mid, 1, mid]
        pe, _ = pay(entrant, rule, E)
        ps, _ = pay(rule, rule, E)
        if (pe > ps) == rising:
            hi = mid
        else:
            lo = mid
    return float(lo)


print("largest g safe from ALLD at eps = 0.05: %.4f (noise-free 0.5)"
      % bound(ALLD))
print("largest g safe from ALT  at eps = 0.05: %.4f (noise-free %.4f)"
      % (bound(ALT, F(1, 5), F(1)), 1 / 3))
print("smallest g safe from ALT at eps = 0.05: %.4f (noise-free 0)"
      % bound(ALT, F(0), F(1, 5), False))


def safe(g, e):
    rule = [1, g, 1, g]
    own = pay(rule, rule, e)[0]
    return pay(ALT, rule, e)[0] <= own and pay(ALLD, rule, e)[0] <= own


print("safe forgiveness window by noise (grid of 1/100, then bisection)")
for e in (F(1, 100), F(1, 50), F(1, 20), F(2, 25), F(1, 10)):
    ok = [i for i in range(101) if safe(F(i, 100), e)]
    if not ok:
        print("  eps %.2f: empty" % float(e))
        continue
    ends = []
    for i, j in ((ok[0], ok[0] - 1), (ok[-1], ok[-1] + 1)):
        lo, hi = F(i, 100), F(j, 100)
        for _ in range(30):
            mid = (lo + hi) / 2
            lo, hi = (mid, hi) if safe(mid, e) else (lo, mid)
        ends.append(float(lo))
    print("  eps %.2f: g from %.3f to %.3f" % (float(e), ends[0], ends[1]))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《亲密关系依恋博弈与冲突停火协议、重复非零和博弈情绪衰减方程与创业伴侣冷战破局母典》第 2 章「宽恕的安全剂量:宽恕型以牙还牙与入侵窗口」(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-285 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026285,
  author    = {Fan, Yafu},
  title     = {Attachment Games and Ceasefire Protocols in Intimate Partnerships: Repeated Non-Zero-Sum Games, Emotion-Decay Equations and Ending Cold Wars between Founder Couples},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论