5D Library › 四柱排盘的历法计算与规则引擎 · English

第七章 《渊海子平》《三命通会》与《滴天髓》

为什么在这里读这些原典

前六章的规则都有文本来源:以日干为主、十神与藏干来自宋代以后的子平书,起运「三日为一岁」与人元司令的天数见于明代的汇编,旺衰的判断在《滴天髓》中被推为全部命理的关键。本章回到这些文本,核对前面的模型所依据的说法,并标明何处由阅读变成了诠释。

《渊海子平》与以日为主

《渊海子平》题宋徐升编,托名徐子平之法,以日干为「我」,看年、月、时的干支与我的关系 (Xu, n.d.)。第四章的十神就是这一「以日为主」的形式化:一旦确定日干,其余九个天干各对应一个十神,这正是 ℤ5×ℤ2\mathbb{Z}_5 \times \mathbb{Z}_2 上以日干为原点的坐标。原书的十神名目与推法今天各本大体一致,本书只用它们的结构,不涉及其中的断语。

《三命通会》:人元司令与起运

明万民英的《三命通会》是子平术的大型汇编 (Wan, n.d.)。它与《渊海子平》都载有各月「人元司令」的天数,把一个月的三十天分给几个天干。第四章的权重 S3 用的是其中一种通行版本。

推导

# Chapter 7. Two classical tables read as data.
# (1) "Days in command" (one commonly printed version, as in p09): each
# month's 30 days are shared among stems. We check the sums, compare the
# stems with the hidden-stem list, and look at the proportions.
# (2) The luck-cycle rule "three days make one year": its usual glosses,
# one day = four months and one double-hour = ten days, assume a year of
# 360 days (12 months of 30). We check them against 360 and 365.2422.
BRANCHES = ("zi", "chou", "yin", "mao", "chen", "si", "wu", "wei", "shen",
            "you", "xu", "hai")
STEMS = ("jia", "yi", "bing", "ding", "wu", "ji", "geng", "xin", "ren",
         "gui")
HIDDEN = ((9,), (5, 9, 7), (0, 2, 4), (1,), (4, 1, 9), (2, 4, 6), (3, 5),
          (5, 3, 1), (6, 8, 4), (7,), (4, 7, 3), (8, 0))
DAYS = (((8, 10), (9, 20)), ((9, 9), (7, 3), (5, 18)),
        ((4, 7), (2, 7), (0, 16)), ((0, 10), (1, 20)),
        ((1, 9), (9, 3), (4, 18)), ((4, 5), (6, 9), (2, 16)),
        ((2, 10), (5, 9), (3, 11)), ((3, 9), (1, 3), (5, 18)),
        ((4, 10), (8, 3), (6, 17)), ((6, 10), (7, 20)),
        ((7, 9), (3, 3), (4, 18)), ((4, 7), (0, 5), (8, 18)))

if __name__ == "__main__":
    print("   branch  days (stem:days)              only in days"
          "  only in list")
    for b, name in enumerate(BRANCHES):
        stems = [s for s, _ in DAYS[b]]
        extra = [STEMS[s] for s in stems if s not in HIDDEN[b]]
        miss = [STEMS[s] for s in HIDDEN[b] if s not in stems]
        cells = " ".join("{}:{}".format(STEMS[s], d) for s, d in DAYS[b])
        print("   {:6s}  {:28s}  {:12s}  {}".format(
            name, cells, " ".join(extra) or "-", " ".join(miss) or "-"))
    print("   every month sums to 30 days: {}".format(
        all(sum(d for _, d in row) == 30 for row in DAYS)))
    tomb = [sorted(d for _, d in DAYS[b]) for b in (1, 4, 7, 10)]
    print("   the four storage branches (chou chen wei xu) split 30 as {}"
          .format(tomb[0]))
    print("   in all four: {}".format(all(t == [3, 9, 18] for t in tomb)))
    for year in (360, 365.2422):
        print("   year of {:8.4f} days: 1 day -> {:6.2f} days of age,"
              .format(year, year / 3))
        print("                            1 double-hour -> {:5.2f}".format(
            year / 3 / 12))
    print("\nReading: the days-in-command table sums correctly but does not")
    print("match the hidden-stem list in five branches; the storage branches")
    print("split their 30 days 18 / 9 / 3, the same 0.6 / 0.3 / 0.1 as S2.")
    print("The age glosses of the luck-cycle rule assume a 360-day year.")
#    branch  days (stem:days)              only in days  only in list
#    zi      ren:10 gui:20                 ren           -
#    chou    gui:9 xin:3 ji:18             -             -
#    yin     wu:7 bing:7 jia:16            -             -
#    mao     jia:10 yi:20                  jia           -
#    chen    yi:9 gui:3 wu:18              -             -
#    si      wu:5 geng:9 bing:16           -             -
#    wu      bing:10 ji:9 ding:11          bing          -
#    wei     ding:9 yi:3 ji:18             -             -
#    shen    wu:10 ren:3 geng:17           -             -
#    you     geng:10 xin:20                geng          -
#    xu      xin:9 ding:3 wu:18            -             -
#    hai     wu:7 jia:5 ren:18             wu            -
#    every month sums to 30 days: True
#    the four storage branches (chou chen wei xu) split 30 as [3, 9, 18]
#    in all four: True
#    year of 360.0000 days: 1 day -> 120.00 days of age,
#                             1 double-hour -> 10.00
#    year of 365.2422 days: 1 day -> 121.75 days of age,
#                             1 double-hour -> 10.15
#
# Reading: the days-in-command table sums correctly but does not
# match the hidden-stem list in five branches; the storage branches
# split their 30 days 18 / 9 / 3, the same 0.6 / 0.3 / 0.1 as S2.
# The age glosses of the luck-cycle rule assume a 360-day year.

这张表每个月都恰好分足三十天,但它列出的天干与通行的藏干表在五个地支上不一致:子、卯、午、酉、亥。四个墓库之支(丑、辰、未、戌)都按 18、9、3 天分配,正是 S2 的 0.6、0.3、0.1;这说明后来软件常用的比例,很可能是从人元司令天数折算来的。

起运的规则是三天折一岁;通行的推算口诀又说一天折四个月、一个时辰折十天。程序说明这两条注解以一年 360 天(十二个月、每月三十天)为前提:按 360 天,一天折 120 天、一个时辰折 10.00 天;按回归年 365.2422 天,则分别为 121.75 天与 10.15 天。差别很小,但它说明口诀是按传统的整月计算的,而不是按天文年。

注

版本核对。 上面的人元司令天数是一种常见的版本,各本在申、亥等月的分配上不尽相同;起运的折算口诀也以通行的转述为准。本书采用这一版本;依其他版本时,只需替换程序中对应的表格。

《滴天髓》:衰旺之真机

原典

《滴天髓·衰旺》

能知衰旺之真機,其於三命之奧,思過半矣。

(Jing Tu (attributed) et al., n.d.)

《滴天髓》旧题京图撰、刘基注,清代任铁樵增注为《滴天髓阐微》,是子平术中最受推重的理论书之一。它把衰旺放在全部判断的中心。第五章的结果从另一面印证了这句话的分量:旺衰确实是整个判断中最关键、也最难统一的一步;三套说得通的参数两两只有中等的一致。

假说 · 尚未验证

作者的诠释(尚未验证): 《滴天髓》强调「真机」,而不给出数值的规则,可以读作对旺衰不能被一组固定参数完全刻画的承认。什么会证明这一读法不当:若能找到《滴天髓》或其注本中明确给出数值权重与阈值的段落,并且历代注家对这些数值一致,这一读法就不成立。

小结

《渊海子平》以日为主,十神是以日干为原点的 ℤ5×ℤ2\mathbb{Z}_5 \times \mathbb{Z}_2 坐标。《三命通会》等书的人元司令天数每月分足三十天,但与藏干表在五个地支上不一致;墓库之支的 18、9、3 天就是软件常用的 0.6、0.3、0.1。起运口诀以 360 日为一年。《滴天髓》把衰旺视为命理的关键,而第五章显示它也正是最依赖约定的一步。

参考文献

Jing Tu (attributed), Liu Ji (attributed commentary), and Tieqiao Ren. n.d. Ditian Sui Chanwei [the Ditian Sui Elucidated]. In Chinese; Qing Dynasty Commentary by Ren Tieqiao.
Wan, Minying. n.d. Sanming Tonghui [Comprehensive Collection on the Three Fates]. In Chinese; Ming Dynasty.
Xu, Sheng. n.d. Yuanhai Ziping [the Sea of Ziping]. In Chinese; Song Dynasty, Later Expanded Editions.

Fan, Y. (2026). Four-Pillar Calendrical Computation and Formal Rule Engines: Julian-Day Sexagenary Cycles, Midnight-Boundary Disagreements, Birth-Time Sensitivity Bounds and Weighted-Graph Strength Scoring [四柱排盘的历法计算与规则引擎] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090180 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).