5D Library › 六维超立方体游走、卦变马尔可夫链与模运算星宫晶格 · English

第六章 三方四正是十二宫环上的卷积

把「三方四正」写成卷积

判断一个宫位时,紫微斗数看的不只是本宫,还有它的对宫与两个三合宫——合称「三方四正」。若宫位 ii 上星曜的数值记为 viv_i,对本宫、对宫、三合宫分别赋予权重 w0,w6,w4,w8w_0, w_6, w_4, w_8,则宫位 ii 的综合分数是

si=∑jwjv(i+j)mod⁡12,s_i = \sum_{j} w_j\, v_{(i + j) \bmod 12},

这正是 ℤ12\mathbb{Z}_{12} 上的循环卷积。循环卷积被离散傅里叶变换对角化:第 kk 个谐波被乘以 W(k)=∑jwje−2πijk/12W(k) = \sum_j w_j e^{-2\pi i jk/12}。凡是 W(k)=0W(k) = 0 的谐波,其中的信息在综合分数里完全看不见。

推导

等权三合的零点。 若只取本宫与两个三合宫、权重相等,则 W(k)=1+e−2πik/3+e−4πik/3W(k) = 1 + e^{-2\pi i k/3} + e^{-4\pi i k/3},当 kk 不是 3 的倍数时为零。也就是说,三合正三角形在 ℤ12\mathbb{Z}_{12} 中是 3 阶子群 {0,4,8}\{0, 4, 8\},对它求和只能看见周期为 4 的模式。

推导

# Chapter 6. "Three sides and four squares" as a convolution on Z12.
# To judge a palace i the reading combines the palace itself, its
# opposite i + 6 and its two trines i + 4 and i + 8 - the triangle
# {0, 4, 8} is the subgroup of order 3 in Z12. Any weighting of these
# four places is a circular convolution s = w * v of the star values v
# with a kernel w on Z12, and is diagonalised by the discrete Fourier
# transform: harmonic k is multiplied by W(k) = sum_j w_j exp(-2 pi i jk/12).
# Where W(k) = 0 the information in harmonic k is lost.
# Part 2: the flowing month. With year branch Y, birth month m, birth hour
# h, one common rule puts the first month's palace ("dou jun") at
# Y - (m - 1) + h and month k a further k - 1 palaces on, so the month-k
# palace is [Y - (m - 1) + h] + (k - 1) = Y - m + h + k (mod 12).
# With the natal Life palace L = 1 + m - h (p10)
# this gives  month palace + L = Y + k + 1  (mod 12), checked for all
# 12^4 combinations.
import cmath

def spectrum(w):
    out = []
    for k in range(7):
        z = sum(w.get(j, 0.0) * cmath.exp(-2j * cmath.pi * j * k / 12)
                for j in range(12))
        out.append(abs(z))
    return out

if __name__ == "__main__":
    kernels = {"self 1, opposite 0.8, trines 0.5": {0: 1, 6: 0.8, 4: 0.5,
                                                    8: 0.5},
               "self and trines only, equal": {0: 1, 4: 1, 8: 1},
               "self and opposite only, equal": {0: 1, 6: 1}}
    print("   kernel                              |W(k)| for k = 0 .. 6")
    for name, w in kernels.items():
        sp = spectrum(w)
        lost = sum((1 if k in (0, 6) else 2) for k, v in enumerate(sp)
                   if v < 1e-9)
        print("   {:34s}  ".format(name) + " ".join("{:4.2f}".format(v)
                                                    for v in sp))
        print("   {:34s}  harmonics lost: {} of 12".format("", lost))
    ok = all((y - m + h + k) % 12 == (y + k + 1 - (1 + m - h)) % 12
             for y in range(12) for m in range(1, 13)
             for h in range(12) for k in range(1, 13))
    print("\n   month palace + natal Life palace = Y + k + 1 (mod 12) in"
          " all cases: {}".format(ok))
    print("\nReading: weighting the triangle equally keeps only the")
    print("harmonics that are multiples of 3 - two thirds of the pattern")
    print("around the circle cannot be seen at all. Adding the opposite")
    print("palace with a different weight restores every harmonic. The")
    print("flowing-month palace is the natal Life palace reflected (L to")
    print("-L) and shifted by the year and the month.")
#    kernel                              |W(k)| for k = 0 .. 6
#    self 1, opposite 0.8, trines 0.5    2.80 0.30 1.30 1.20 1.30 0.30 2.80
#                                        harmonics lost: 0 of 12
#    self and trines only, equal         3.00 0.00 0.00 3.00 0.00 0.00 3.00
#                                        harmonics lost: 8 of 12
#    self and opposite only, equal       2.00 0.00 2.00 0.00 2.00 0.00 2.00
#                                        harmonics lost: 6 of 12
#
#    month palace + natal Life palace = Y + k + 1 (mod 12) in all cases: True
#
# Reading: weighting the triangle equally keeps only the
# harmonics that are multiples of 3 - two thirds of the pattern
# around the circle cannot be seen at all. Adding the opposite
# palace with a different weight restores every harmonic. The
# flowing-month palace is the natal Life palace reflected (L to
# -L) and shifted by the year and the month.

第一种权重(本宫 1、对宫 0.8、三合各 0.5)的谱在所有谐波上都不为零,没有丢失信息。等权的本宫加三合只保留 k=0,3,6k = 0, 3, 6 三个谐波,12 个谐波中丢失 8 个——宫位上三分之二的变化模式,在综合分数里根本看不见。等权的本宫加对宫丢失 6 个。只要对宫的权重与三合不同,全部谐波就都恢复了。

三种汇总权重在 \mathbb{Z}_{12} 各谐波上的放大倍数 |W(k)|。等权三合在 k 不是 3 的倍数处为零。

程序的最后一行检查流月的规则。一种常见的起流月方法(「斗君」)是:以流年地支 YY 所在宫为流年命宫,从此逆数到生月、再顺数到生时为正月,此后顺排。这样正月在 Y−(m−1)+hY-(m-1)+h,第 kk 月再顺数 k−1k-1 宫,流月命宫为 [Y−(m−1)+h]+(k−1)≡Y−m+h+k(mod⁡12)[Y-(m-1)+h]+(k-1)\equiv Y-m+h+k \pmod{12},与本命命宫 LL 合起来满足 流月命宫+L≡Y+k+1(mod⁡12)\text{流月命宫} + L \equiv Y + k + 1 \pmod{12}——流月命宫是本命命宫的反射(L↦−LL \mapsto -L)再加上年与月的平移。各家起流月的方法不尽相同;本书采用上面这一种,程序验证的也是这一种。

一个按事项评分的算法

一个「求财、问婚、涉讼、出行」的定量评分算法,可以写成:确定流月命宫,找到事项对应的宫位(求财看财帛),在它的三方四正上把星曜的数值按卷积加起来。算法本身是精确的;不精确的是输入——每颗星的数值。第十三段程序比较两套都说得通的赋值:A 是一般的吉凶尺度,B 是偏重财务的尺度(提高武曲、贪狼,放宽七杀、破军,降低天同)。

推导

# Chapter 6. A numerical score for "seeking wealth this month" - and how
# much it depends on the conventions put into it.
# The score of a matter is a weighted sum of star values over the "three
# sides and four squares" of the matter's palace (kernel of p12: self 1,
# opposite 0.8, trines 0.5). For wealth the matter's palace is the Wealth
# palace, four places back from the flowing-month Life palace. Star
# values are a convention. Two plausible conventions are compared:
#   A  a general good/bad scale
#   B  a wealth-specific scale (Wu Qu and Tan Lang raised, Qi Sha and
#      Po Jun softened, Tian Tong lowered)
# Over all 12 main-star layouts (p11) x 12 flowing-month Life palaces,
# the program scores the 144 cases under each convention and reports
# their Spearman rank correlation, the share of pairs of cases the two
# conventions order differently, and the overlap of their top 10 cases.
import math
from itertools import combinations

ZI = {"Zi Wei": 0, "Tian Ji": -1, "Tai Yang": -3, "Wu Qu": -4,
      "Tian Tong": -5, "Lian Zhen": -8}
FU = {"Tian Fu": 0, "Tai Yin": 1, "Tan Lang": 2, "Ju Men": 3,
      "Tian Xiang": 4, "Tian Liang": 5, "Qi Sha": 6, "Po Jun": 10}
A = {"Zi Wei": 2, "Tian Fu": 2, "Tai Yang": 1, "Tai Yin": 1,
     "Tian Tong": 1, "Tian Liang": 1, "Tian Xiang": 1, "Wu Qu": 1,
     "Tian Ji": 0.5, "Tan Lang": 0, "Lian Zhen": -0.5, "Ju Men": -1,
     "Qi Sha": -1, "Po Jun": -1}
B = dict(A, **{"Wu Qu": 2, "Tan Lang": 1, "Qi Sha": 0, "Po Jun": -0.5,
               "Tian Tong": 0})
KERNEL = {0: 1.0, 6: 0.8, 4: 0.5, 8: 0.5}

def layout(z):
    f = (4 - z) % 12
    out = {s: (z + o) % 12 for s, o in ZI.items()}
    out.update({s: (f + o) % 12 for s, o in FU.items()})
    return out

def score(z, life, values):
    stars = layout(z)
    wealth = (life - 4) % 12
    # rounded, so that equal scores compare equal whatever the order of
    # the terms
    terms = [KERNEL[off] * values[s] for s, pos in stars.items()
             for off in KERNEL if pos == (wealth + off) % 12]
    return round(math.fsum(terms), 9)

def ranks(xs):
    order = sorted(range(len(xs)), key=lambda i: xs[i])
    r = [0.0] * len(xs)
    i = 0
    while i < len(order):
        j = i
        while j + 1 < len(order) and xs[order[j + 1]] == xs[order[i]]:
            j += 1
        for k in range(i, j + 1):
            r[order[k]] = (i + j) / 2 + 1
        i = j + 1
    return r

def pearson(x, y):
    n = len(x)
    mx, my = math.fsum(x) / n, math.fsum(y) / n
    sxy = math.fsum((a - mx) * (b - my) for a, b in zip(x, y))
    sxx = math.fsum((a - mx) ** 2 for a in x)
    syy = math.fsum((b - my) ** 2 for b in y)
    return sxy / (sxx * syy) ** 0.5

if __name__ == "__main__":
    cases = [(z, life) for z in range(12) for life in range(12)]
    sa = [score(z, l, A) for z, l in cases]
    sb = [score(z, l, B) for z, l in cases]
    rho = pearson(ranks(sa), ranks(sb))
    disagree = sum(1 for i, j in combinations(range(len(cases)), 2)
                   if (sa[i] - sa[j]) * (sb[i] - sb[j]) < 0)
    pairs = len(cases) * (len(cases) - 1) // 2
    top_a = set(sorted(range(len(cases)), key=lambda i: -sa[i])[:10])
    top_b = set(sorted(range(len(cases)), key=lambda i: -sb[i])[:10])
    print("   cases scored: {}; distinct scores A {}, B {}".format(
        len(cases), len(set(sa)), len(set(sb))))
    print("   Spearman rank correlation A vs B: {:.3f}".format(rho))
    print("   pairs of cases ordered oppositely: {:.3f}".format(
        disagree / pairs))
    print("   top-10 cases shared: {} of 10".format(len(top_a & top_b)))
    print("\nReading: two conventions that a practitioner could defend")
    print("agree broadly but not in detail: they order a sizeable share of")
    print("pairs of months oppositely and share only part of their 'best")
    print("months'. A score is exactly as definite as its conventions.")
#    cases scored: 144; distinct scores A 43, B 41
#    Spearman rank correlation A vs B: 0.606
#    pairs of cases ordered oppositely: 0.264
#    top-10 cases shared: 6 of 10
#
# Reading: two conventions that a practitioner could defend
# agree broadly but not in detail: they order a sizeable share of
# pairs of months oppositely and share only part of their 'best
# months'. A score is exactly as definite as its conventions.

在 12 种主星排布与 12 个流月命宫组成的 144 种情形中,两套赋值的排序秩相关只有 0.606;0.264 的情形对被两套赋值排成相反的顺序;各自的「最佳十个月」只有 6 个重合。两套赋值都有理由,却对「哪个月求财最好」给出相当不同的答案。

注

对「流月流日定量评分」的含义。 卷积与谐波分析说明了评分算法的结构,但分数的意义完全来自星曜赋值与汇总权重这两组约定。一个评分系统应当公开它的赋值与权重,并说明换一套合理的约定后结果会变多少;否则分数的精确只是表面的。第八章的检验要求在看到任何结果之前固定这些约定。

小结

三方四正的汇总是 ℤ12\mathbb{Z}_{12} 上的循环卷积,由离散傅里叶变换对角化。等权的三合正三角形是 3 阶子群上的求和,丢失 12 个谐波中的 8 个;加上权重不同的对宫后信息全部保留。流月命宫是本命命宫的反射加平移。一个按事项评分的算法在结构上是精确的,但它的结果取决于星曜赋值:两套都说得通的赋值,排序秩相关只有 0.606,最佳月份只有 6 个重合。

Fan, Y. (2026). Six-Dimensional Hypercube Walks, Hexagram Mutation Markov Chains and Modular Star-Palace Lattices: Boolean Cube Spectral Gaps, Yarrow-versus-Coin Probability Biases and Twelve-Palace Dihedral Projections [六维超立方体游走、卦变马尔可夫链与模运算星宫晶格] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090176 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).