一面镜子只照一遍;把 Theta 提升首尾串成一条长廊,每多一面镜子就多一层“导数”。问题是这条链能走多远、走到哪里会被成本吃光。本页把这笔账画成两个场景,共用一条时间轴:链长预算 b = m/√X。本页是数值演示与示意模型,不是证明。
bigCellMinors;把一面“镜子”倒回去只差一个标量 |𝔽|,是 final_gaussOp,读数在 𝔽₅ 上逐项验算。shortVector:L 里有非零向量不长于 √rk·covol1/rk = √7·e−5/7 ≈ 1.295。muMaxLeMin:若一切非零向量都不短于 r,则 μmax ≤ −ln r + ½ ln 7;所以 μmax = 1.8 时,任何大于 √7·e−1.8 ≈ 0.437 的 r 都做不到这一点,必有更短的非零向量。屋顶越平,两个半径越接近。书中:第四章 · Theta 提升链 · 跷跷板法则与大胞腔 · 第五章 · 格子的屋顶 · 跨特征胶合
One mirror shows you once; link the theta lifts into a corridor and every extra mirror buys one more layer of "derivative". The question is how far the chain can go before its cost eats the gain. The page draws that account as two scenes on one rail: the chain budget b = m/√X. The page is a numerical illustration with an illustrative model, not a proof.
bigCellMinors; undoing one "mirror" costs only the scalar |𝔽|, which is final_gaussOp, checked term by term on 𝔽₅ in the readouts.shortVector: L has a nonzero vector no longer than √rk·covol1/rk = √7·e−5/7 ≈ 1.295. muMaxLeMin: if every nonzero vector is at least r long, then μmax ≤ −ln r + ½ ln 7; so with μmax = 1.8 no r above √7·e−1.8 ≈ 0.437 can do that, and some nonzero vector is shorter. The flatter the roof, the closer the two radii.In the book: Chapter 4 · the theta-lift chain · the seesaw and the big cell · Chapter 5 · the roof of a lattice · gluing across characters
final_gaussOpbigCellMinorsshortVectormuMaxLeMin