Skip to content

[P2 theory candidate] Mesoscopic two-cloud root and capillary sign law - #533

Draft
LightChainr wants to merge 34 commits into
research/results-first-synthesis-20260831from
analysis/p337-mesoscopic-two-cloud-20260901
Draft

[P2 theory candidate] Mesoscopic two-cloud root and capillary sign law#533
LightChainr wants to merge 34 commits into
research/results-first-synthesis-20260831from
analysis/p337-mesoscopic-two-cloud-20260901

Conversation

@LightChainr

@LightChainr LightChainr commented Sep 1, 2026

Copy link
Copy Markdown
Owner

维护分类(2026-09-01,head 5aa929a6ef4ff92d71f6f36097b763c05ad71ab8): 本 PR 保持 priority:P2、整体 C0,不可整体升级#542 保持 open,不可关闭。可保留为 C1 的仅有:relaxed all-w Catalan–Toeplitz 单粒子恒等式;普通简单 NN 轮廓的 EW/WE cut-edge retrace 禁止规则;一个合法的 nonlocal one-west occupied-corner 反例;以及在“物理壁修正已具有一致指数可和的双-gap 核”额外假设下推出 pq 双零的条件引理。尚未闭合的 C0 接口包括 physical beta-cloud/B_wallw>=2 与 nonlocal physical endpoint transfer、rank1 与 rank0/2 的共同坐标、second thermal/root/original-U、以及 uniform contour/sector dependencies;旧 a*s、控制字符 CR 和 O(Na^4) 量级缺陷仍在。该 PR 不进入当前执行队列,不授权随机生产、合并,且不得从 axis/beta gate 外推 unrestricted square-lattice/full model。详见本 PR 维护复核评论与 #542

结果

本 PR 在 Draft #509 上继续固定闭合源 Sstar 的理论主线,处理此前未覆盖的介观联合极限:

N=25k^2,  m→∞,  N/m^4→0,  但 N/m^2 可以→∞。

新的两相图像不是 empty/full,而是:

  • rank0:均值 N/m^2 可发散的稀疏黑粒子云;
  • rank2:带消失白孔云的近满占背景。

两侧 reference gases 为

R0=(1+h/m^2)^N,
R2=(h+m^-4)^N,

故精确两云中心是 h_bar=1+m^-2。首个 connected pressure 不抵消:rank0 每站点一个 elementary occupied face,系数 c0=1;rank2 每站点两个 adjacent-vacancy edges,系数 c2=2。因此 pooled Sstar root 得到参数无关展开

log h_root = m^-2 -(1/2)m^-4 -(2/3)m^-6 +o(m^-6).

完整证明窗口

N/m^4→0 及两个既定 N25 quotient 上,增长 cutoff 保留旧 exp[A_N] 因子并落在 systole 以下,得到:

Z0=(1+h/m^2)^N[1+O(N/m^4+N^-M)],
Z2/h^N=1+N h^-1m^-4+O(N/m^6+(N/m^4)^2+N^-M),
Z1/(Z0+Z2)=O(N^-M).

相应地,K|r=0Poisson(N/m^2) 的 TV 距离趋零,即使均值发散;Pr(full|r=2)=1-N/m^4+o(N/m^4)。pooled root 上两 sector 各占 1/2+o(1),原 within-geometry denominator 仍为宏观量,既有 rank1 bound 因而继续压低原 U。

同面积几何的 leading one-hole 项精确相同,所以新 shape 预测为

(Z2_f/h^N)/(Z2_s/h^N)=1+O(N/m^6+(N/m^4)^2).

新的毛细窗口

介观两云定理还暴露了一个此前没有分离出的界面尺度。轴向一条边界只在相邻两行切换时,其精确相对生成函数为

F_L(m)=sum_j C(L,2j)m^(-2j)
      =[(1+m^-1)^L+(1-m^-1)^L]/2.

因此 L/m→cF_L→cosh(c)。继续完成全部 leading simple annuli、双边界 hard core 及两侧局部气体后,得到完整闭式:

G(c)=I0(2c)^2,
G(c,u)=I0(2c)^2 sinh(u/2)/(u/2),
Phi(c)=[I0(2c)^2-I1(2c)^2]/Delta.

这里 exterior black singleton 与 interior white singleton-hole 在精确根 h*=1+m^-2 上具有同一 base:1+m^-2 h*=h*+m^-4。所以 width tilt 和 bridge-area tilt 精确取消;original-U 的 surviving mark 位于 width 1/L-1 endpoint,其严格非交叉双桥核是 Karlin–McGregor determinant J1=I0²-I1²

这给出四个直接结论:

  • bulk root 的 N/m^4→0 定理不受影响;
  • 固定-L 系数的充分门由 absolute bound 的 L^2/m→0 锐化为 L/m→0,整个旧的 L≪m≲L^2 缺口已关闭;
  • 对每个有限 c=L/mI0(2c)>I1(2c)>0,故 Phi(c)>0,original U 保持负号;
  • black-only 截断曾产生 c≈3.930888 的候选零点;补入实际合法、activity 为 m^-4h^-1 的内部白孔后,该零点精确消失。纠错链保留在笔记中。

这里 lambda=N/m^2=(L/m)^2 同时是黑粒子云均值和毛细粗糙度的平方;两类现象不是两个无关参数。

非紧毛细尾与第一个 crossover

进一步把单边界更新为精确 directed kernel

t_r(z)=sum_d r^|d| z^d=(1-r^2)/[(1-rz)(1-r/z)],
K_n=[z^n]t_r(z)^L,   J_(1,L)=K_0^2-K_1^2,
r=m^-1.

其高阶跳跃有精确 compound-Poisson 分解。最初的 absolute comparison 由于忘掉 endpoint 的两个 Dirichlet/Vandermonde 零点,给出了较强的临时门

alpha=(L/m)(L/m^2)=L^2/m^3.

它首先将 bounded-c 定律延拓到

L^(2/3) << m << L,
J_(1,L) ~ exp(4L/m)/[8 pi (L/m)^2],
U/A_N ~ -exp(4L/m)m^(-(2L-1))/(8 pi Delta) < 0.

但这个 alpha 门现在已被严格取代。所有含 w≥0 个 west steps 的 relaxed words 可一次闭式重求和:

W_(L,w)(z)=m^(-2w) C(L+2w,w)
 [1-m^-1(z+z^-1)]^(-(L+2w+1)),
R_L(z)=D_L(z) M_L(m^-2/g_m(z)^2).

这是一个正的解析 Catalan–Toeplitz 乘子,不含 inverse-Vandermonde 极点。物理 self-avoidance 只删除这个正系数 majorant 中的非法字;固定半径 collar 的双 sine transform 仍保留两个 endpoint 零点。因此全部 reversal sector 的真正充分门是

beta=L/m^2 -> 0,
|Z_1c-J_run|/J_run=O(beta).

这也解决了所有先前 hairpin 候选:物理 local hairpin 有 kappa_hp(∞)=0;补集 Ward 恒等式虽留下正 boundary flux,placement average 仍额外给 1/c。最短合法 packet 是四个 7-edge L-triomino,Delta g=6,original-source 残差只有 -8m^-2+O(m^-4)

借用宏观 rough bridge 竖边的非局部 west-step 也没有逃逸:它消去 borrowed history 后的 exact gap multiplier 仍指数可和。真实 occupied-corner w=1 类因 west edge 两侧必须是同号竖步,实际 activity 只有 epsilon=L/m^4;去除共同 bulk dressing 后的 Dirichlet 修正为 -4/m^3+o(m^-3)

最后,rough collar 的真实大小由竖 cut-edge 数 S 控制,E_end S=4c-2+O(c^-1),不是额外的 sqrt(c) 状态。首个几何相关 cloud/source 余项仅 beta/m;contractible bulk pressure 的 beta^2 项是等面积几何的共同 thermal reparametrization。额外 essential carrier、横向绕行、其他同调类与 tilted companion 都还有 exp[-Theta(L log m)] barrier。所以完整 axis original-U 定律现在是

m->infinity, L/m^2->0:
U/A_N=-(L^2/Delta)m^(-(2L+1))
      [I0(2L/m)^2-I1(2L/m)^2][1+o(1)] < 0.

即符号律已从 m≫L^(2/3) 再扩到 m≫L^(1/2),不再留一个未定 capillary crossover。

固定 m 还出现一个不同、不能与上述联合极限混淆的精确阈值。directed 正子族的指数基底为

q_dir(m)=[(m+1)/(m(m-1))]^2,
m_dir=1+sqrt(2).

m<m_dir 时,这个正的 rank-one 子族已经越过直条纹的名义 m^(-2L) barrier;在临界点

m_dir^(-2L) J_(1,L) ~ (3-2sqrt(2))/(2 pi) L^-2.

这严格否定“直条纹成本 + 次指数简并”式的正界面张力论证,但不宣称完整 fixed-m 相变;m>m_dir 也只说明该 directed 子族受抑。

倾斜 quotient 的最短类也已精确分类:两条 primitive class 长度均为 7k,每类 thin-cycle 数为 (25k/7)C(7k,3k),完整 minimum class 由 compatible noncrossing resolved annuli 组成。即使取最坏的平方熵上界,额外 m^-4k barrier 仍获胜,所以 unresolved scaling function 是轴向界面问题,倾斜最短类不能在 leading exponential order 翻转它。

更宽窗口

若只看局部两云 chart,第一处未重求和作用在 N/m^6,而不是 N/m^4。新的 axis 定理走了另一条更强的路:beta=L/m^2→0 自动给 N/m^4=beta^2→0,同时用局部 pressure 因子化和 exp[-Theta(L log m)] 拓扑 barrier 补齐了 axis/tilted 序列的长轮廓界。这不意味着任意 N/m^6→0 路径都已被定理覆盖。

文件

  • notes/closed-source-mesoscopic-two-cloud-synthesis.md:统一结论与停止边界;
  • notes/closed-source-mesoscopic-black-gas.md:rank0 相对 partition 与 TV 定理;
  • notes/closed-source-mesoscopic-full-phase.md:rank2 纯化、根与 shape ratio;
  • notes/closed-source-mesoscopic-root-chart.md:两云 chart、power gates、c2-c0=1
  • notes/closed-source-capillary-window.md:精确 cosh(L/m) 子族与新 scaling window;
  • notes/closed-source-axis-capillary-uniformity.mdL^2/m→0 的 signed derivative 充分定理;
  • notes/closed-source-tilted-shortest-winding-class.md:倾斜最短类的精确计数和 barrier/entropy 比较。
  • notes/closed-source-axis-capillary-transfer-scaling.md:两云纠错链与 Bessel determinant;
  • notes/closed-source-axis-signed-interface-transfer.md:补集配对、L/m 锐化和 signed theorem;
  • notes/closed-source-axis-capillary-complete-transfer.md:完整 unmarked/area transfer、overhang 与 endpoint 控制。
  • notes/closed-source-axis-kernel-uniform-capillary-tail.md:精确 compound-Poisson kernel、非紧 Bessel tail 与 fixed-m directed 阈值;
  • notes/closed-source-axis-growing-capillary-window.mdm≫L^(2/3) 的完整两边界充分门与 finite-gap 状态;
  • notes/closed-source-axis-capillary-unbounded-error.md:误差分解、single-hairpin crossover kernel 与 fixed-m 局部 CLT。
  • notes/closed-source-single-hairpin-endpoint-zero.md:物理 local packet 保留 Dirichlet 零点的一般判据;
  • notes/closed-source-single-hairpin-ward-identity.md:正 boundary flux 与 placement/bulk 归一化的 Ward 分解;
  • notes/closed-source-axis-short-hairpin-packet.md:四个 L-triomino 局部字、Delta g=6-8a 热残差。
  • notes/closed-source-delocalized-west-step-kernel.md:全 w≥0 Catalan–Toeplitz 重求和与 beta gate;
  • notes/closed-source-delocalized-borrowed-side-counterexample.md:无统一 local-ear 定理的反例与 span tail;
  • notes/closed-source-axis-one-west-borrowed-bridge.md:真实 w=1L/m^4 权重和负 Dirichlet 修正;
  • notes/closed-source-axis-one-carrier-beta-gate.md:完整 reversal/self-avoidance/fixed-collar 的 L/m^2 定理;
  • notes/closed-source-axis-beta-cloud-closure.md:rough collar、contractible pressure 与 source 余项;
  • notes/closed-source-axis-sector-closure-gate.md:extra carriers、homology、tilted companion 与 sector odds 闭合。

执行与边界

  • 三个独立子任务分别处理稀相、满相/root 与界面几何,主任务再整合;
  • 只做精确组合/渐近推导,并用 managed Python/SymPy 核对 Möbius chart 与三阶级数;
  • git diff --check 通过;控制字符扫描未通过,notes/closed-source-mesoscopic-black-gas.md 仍含一个 carriage return;
  • 0 新枚举、0 MC、0 云作业、0 参数扫描;
  • 不声称 fixed-m 相变、H4/Jordan 身份或新的独立证据票;
  • stacked base 已同步到 Draft research: 决策停线、临界空间求和与下一理论门 #50912974db4,包含新的 critical raw-pair summability theorem 与 guarded claim ledger;
  • 该 summability theorem 控制固定 p_c 上的 raw pair kernel,不自动控制 thermal derivative,因此本 PR 没有把它当成 moving-root 结论。

当前 head:5aa929a6

@LightChainr LightChainr changed the title [P1 theory] Mesoscopic two-cloud coexistence and three-term root expansion [P2 theory candidate] Mesoscopic two-cloud coexistence and root expansion Sep 1, 2026
@LightChainr

LightChainr commented Sep 1, 2026

Copy link
Copy Markdown
Owner Author

维护复核更新:5aa929a6 仍仅局部 C1,完整 physical root/collar 仍为 C0

复核固定到 head 5aa929a6ef4ff92d71f6f36097b763c05ad71ab8。本 PR 保持 P2 theory candidate / C0 overall不可整体升级#542 保持 P2 / open不可关闭。不进入 current claim,不授权生产或合并。

5aa 最新裁决

可保留为 C1 的仅有:

  • relaxed all-w Catalan–Toeplitz 单粒子恒等式;
  • 普通简单 NN 轮廓的 EW/WE cut-edge retrace 禁止规则;
  • 一个合法的 nonlocal one-west occupied-corner 反例;
  • 若额外假设 physical wall correction 已形成一致指数可和的双-gap 核,则其 sine transform 保留 pq 双零的条件引理。

仍为 C0 的缺口:

  • physical beta-cloud/B_wall
  • w>=2 与 nonlocal physical endpoint transfer;
  • rank1 carrier 与 rank0/2 是否共享同一 thermal/root coordinate;
  • second thermal moment、共同 root 与 original-U slope;
  • uniform contour、extra-carrier、homology、tilted/sector dependencies。

旧缺陷也仍在:a*s 尺度遗漏、closed-source-mesoscopic-black-gas.md 的 CR 控制字符,以及旧 O(Na^4) 量级陈述。现有 axis/beta gate 不得表述为 unrestricted square-lattice/full-model 定理。


以下为 b49e2a21 审计历史(原样保留)

可保留的 C1

  • closed-source-axis-kernel-uniform-capillary-tail.md 的 arbitrary-run directed kernel、K_0^2-K_1^2>0 与 fixed-m directed threshold m_dir=1+sqrt(2),只适用于 directed subfamily。
  • closed-source-single-hairpin-endpoint-zero.md:178-196 是条件算子引理:只有在 B=P_-BP_-、stripped gap/collar kernel 一致可和且 placement 含 1/L 时,才得 kappa_hp(infinity)=0
  • closed-source-axis-short-hairpin-packet.md 的最短四个 L-triomino words、Delta g=6 和局部 collar residual -8a+O(a^2) 是有限局部计算。

仍阻断 full-lattice / moving-root / original-U 定理

  1. closed-source-axis-capillary-unbounded-error.md:241-258 仍把 relaxed-word bound 直接提升到 two-gas decorations 和 full original-U;同一 restricted denominator 下的 physical collar、二阶热矩、moving root、U slope 与 w>=2 余项没有证明。
  2. finite-a Ward 识别内部不一致:Ward note 说 h=1+a 时互补 packet 等权,但 packet note 给 w_-/w_+=(1+a)^2/(1+a+a^2) != 1。所以 Ward coefficient 目前只能是 canonical/leading-core 恒等式。
  3. ancestry 只部分修复:eed2190c12974db4 已进入;2690f665762dbaf4410015f5 仍不是祖先,链接的 closed-source-poisson-double-scaling.md 仍不在 tree。
  4. root chart 虽显示 a*s,仍把 exp(a*s) 归入 o(tau),但没有 Na -> infinity 时该 bounded-s 声明不一致;root-local s=O(tau) 的另证尚未写出。
  5. uniform connected/full-lattice remainder 未闭合;closed-source-mesoscopic-black-gas.md:170 仍含 carriage return;signed-interface note 仍写 O(Na^4),相邻孔的正确 Mayer 阶是 O(Na^3)=O(N/m^6)

#542 的处置

#542 保持 P2,只承接一般 physical encoding/collar、thermal/root、multi-west-run 与 nonlocal west-step。它记录的 bare single-W 必要约束和 restricted directed-core 热标记界尚未进入本 PR,也不等于 physical original-U 定理。允许的是解析证明和有界反例搜索;没有新 MC、GPU、云生产或 N25 扩展授权。

@LightChainr LightChainr changed the title [P2 theory candidate] Mesoscopic two-cloud coexistence and root expansion [P1 theory] Mesoscopic two-cloud root and capillary sign law Sep 1, 2026

Copy link
Copy Markdown
Owner Author

容器独立复核:局部系数得到支持,但不替代 uniform remainder;面积标记需要单独控制

复核固定到 df480bb12417fcdc7add7703ce448d8495fc5045,已阅读维护复核。保留本PR的 P2 theory candidate 状态;没有重开被撤回的 growing-cutoff rate 质疑,没有合并/rebase或修改ledger。

1. 旧完整N25表上的独立有限级数

逐字使用 a70eeff09f51ce2fa0fea5ae637e9191efbf2e1f:results/p337-closed-source-finite-coupling/{axis,tilted}.csv;SHA256分别为 2d23fecc98d276d9ad15ad1867199cd308f0570cb5040ef94eb6b923b4c53458225031e612929ed922ba75c55e76703d59990f5283e7ac39b94f022841798da5。每K二项计数和总2^25均通过。

用Fraction对z=1/m解log h,pooled方程保留各几何独立分母,递推到z^18并检查残差各系数严格为零。有限N25结果是

log h_pool=z^2-z^4/2-2z^6/3-5z^8/4-z^10+O(z^12),
log h_axis-log h_tilt=-(2/5)z^10+O(z^12).

前三项独立支持本PR的局部代数,但不认证随N增长的余项。第四项-5/4也只在本次明确的有限表/局部计数范围内作为结果,不直接升级为联合极限定理。

在共同h_bar=1+z²而非各自根处:

log(Z0_axis/Z0_tilt)=-10z^10+O(z^12),
log(Z2_axis/Z2_tilt)=-50z^12+O(z^14).

另用独立lifted-BFS穷举两几何的固定5-vacancy层,共106,260个配置,全部(g,q)计数与旧表一致,含后一个形状系数来源。115位Decimal在m=8,16,32,64,128复算;m64还以210次有理二分和交错log级数给出单点根/余项包围。不是新全量N25枚举或独立统计票。

按a=m^-2,表上逐相的connected pressure/site到a^4为

r0/N = h^4 a^3 +(2h^6-4h^5-h^4)a^4+O_N(a^5),
r2/N = 2h^-2 a^3 +(h^-4+6h^-3-2h^-2)a^4+O_N(a^5).

这些有限系数及1..6个格点polyomino检查不填补维护评论指出的uniform cluster lemma。

2. capillary权重趋近不等于热jet趋近

对已声明两高度物理子族,去掉面积中点公共h幂,令X=sum(sigma_i-1/2):

T(eta)=[[exp(-eta/2),1/m],[1/m,exp(eta/2)]], eta=log h,
Z_L(eta)=Tr T(eta)^L/2,
Var_0(X)=Lm/4*tanh[L*atanh(1/m)].

L=2..12、m=2/3/5/10共44组Fraction直接环枚举通过(32,752序列)。chi=L/m固定时,

Z_L(u/L)/Z_L(0) -> cosh(sqrt(chi²+u²/4))/cosh(chi).

所以在eta=log(1+m^-2)处,虽然Z_L(eta)/Z_L(0)->1,仍有

partial_eta logZ_L -> chi*tanh(chi)/4.

这定位了尚需控制的面积标记。还检查了实际互补条带对子族:未归一化capillary mass有非恒定dressing,但(h^K+h^(N-K))/(1+h^N)在h=1的热导数逐对严格为零。因此从正子族F_L非恒定,不能单凭这一点推出完整原U的Phi非恒定或某个符号;不否定这些结论可能由完整带标记计算建立。

后续

已按用户授权将具体带面积标记双边界transfer任务写入 #534;另一个不同模型的空间方法pilot为 #535。资源建议16CPU/32GiB,不申请GPU,均保留明确验收和停止条件。

本轮隔离包15/15专项tests通过。代码、有限有理证书及运行日志在本对话 matching_one_followup_20260901.zip,尚未写入Git;本评论给出完整可复算的核心公式和输入来源,不冒称全仓CI或统一定理验证。

Copy link
Copy Markdown
Owner Author

条件根误差证书:把尚缺的相内余项写成可验收的不等式

本轮在容器用符号级数和有理界核对了一个可直接复用的根定位接口;它不补齐本 PR 的 connected-cluster remainder,也不提升当前 P2 状态。已读维护复核及新 #534,不重复分发毛细 transfer。

a=m^-2, x=log h,保留精确 reference odds:

F(x)=log(e^x+a^2)-log(1+a e^x),
F(log(1+a))=0,
F'(x)=h(1-a^3)/[(h+a^2)(1+a h)].

0<a<=1/16, 1/2<=h<=2 时,逐因子有理界给

F'(x) >= 455/2052 > 1/5.

若能在包含实际 pooled root 与下述 reference root 的窗口上,对每个几何真正证明

ell_g(x):=log(Z2_g/Z0_g)
        =N[F(x)+a^3]+R_g(x),
max_g sup_x |R_g(x)| <= epsilon_N,

x_daggerF(x_dagger)+a^3=0,则任一在此窗口中的等正权 pooled root 满足

|x_root-x_dagger| <= 5 epsilon_N/N.

原因是 sign q_g = sign ell_g:在 pooled q=0 处,不能所有 ell_g 同为严格正或同为严格负,故 |N(F+a^3)|<=epsilon_N。此根定位步骤本身不需要先把 rank1 质量设零;但 rank1 和热标记仍影响原 U 的斜率分母,不能由此略去它们。

形式级数独立核对:

x_dagger=a-a^2/2-2a^3/3+O(a^4),
F(a-a^2/2-2a^3/3)+a^3=5a^4/4+O(a^5).

因此识别 -2/3 的足够输入正是 epsilon_N=o(N a^3);若有 epsilon_N=O(N a^4),便直接得到 O(a^4) 的根误差。仅有 o(1) partition remainder 不足以解析该系数。

建议把余项主任务表述为上面的实际受限 log-odds 一致界,明确窗口、几何与 normalization;同总 pressure、未标记 capillary 权重、有限阶局部对象计数都不能替代它。根窗口覆盖/存在性与完整 U 的热分母下界仍须分别提供,不能把 reference F' 下界当作真实模型热斜率下界。

本轮还新建 #536,处理另一指定模型的 canonical Q1 热协方差双 pivotal 通道;它不是本 PR 的 fixed-m 或 mesoscopic theorem 输入。两线不互借符号或复杂度结论。

@LightChainr LightChainr added the priority:P2 Deferred research or on-demand support; no default new compute allocation. label Sep 1, 2026
@LightChainr LightChainr changed the title [P1 theory] Mesoscopic two-cloud root and capillary sign law [P2 theory candidate] Mesoscopic two-cloud root and capillary sign law Sep 1, 2026
@LightChainr

Copy link
Copy Markdown
Owner Author

2026-09-01 science card: unbounded capillary tail

PR #533 now extends the bounded-c=L/m sign law into a genuinely noncompact window.

New exact kernel

For r=m^-1, the directed one-boundary transfer is

t_r(z)=sum_d r^|d|z^d=(1-r^2)/[(1-rz)(1-r/z)],
K_n=[z^n]t_r(z)^L,
J_(1,L)=K_0^2-K_1^2>0.

The higher jumps admit an exact compound-Poisson decomposition. Because the endpoint determinant is only a 1/c fraction of the bulk two-bridge kernel, the cancellation-sensitive remainder is

alpha=(L/m)(L/m^2)=L^2/m^3.

Hence, throughout L^(2/3)<<m<<L,

J_(1,L) ~ exp(4L/m)/[8 pi (L/m)^2],
U/A_N ~ -exp(4L/m)m^(-(2L-1))/(8 pi Delta) < 0.

This closes a substantial part of the previous c->infinity boundary. The sign is not an artifact of the asymptotic bound: J_(1,L)>0 exactly at every finite L in the directed kernel.

First remaining mechanism, now one-dimensional in hypothesis space

Long directed vertical runs preserve the determinant. Bare repeated width-one encounters are already summed by J_1. The first unsummed packet capable of competing at alpha=O(1) is one horizontal reversal / hairpin together with any cloud-collar contact pinned to that turn.

At fixed alpha, multiple hairpins remain lower order. The remaining question is therefore one finite-gap matrix element, the stripped single-hairpin endpoint coefficient kappa_hp(infinity):

  • kappa_hp(infinity)=0: the current sufficient gate is not sharp and the law should extend to the bulk condition L/m^2->0;
  • kappa_hp(infinity)!=0: it produces the first genuine unbounded-capillary crossover function.

This is now the next theory target; it replaces open-ended contour cataloguing.

Separate fixed-m discovery

The exact positive directed subfamily has exponential base

q_dir(m)=[(m+1)/(m(m-1))]^2

and the threshold

m_dir=1+sqrt(2).

At the threshold,

m_dir^(-2L)J_(1,L) ~ (3-2sqrt(2))/(2 pi)L^-2.

For m<m_dir, this positive rank-one subfamily already proliferates relative to the nominal straight-interface barrier. This kills a Peierls argument based only on straight cost plus subexponential degeneracy. It is deliberately not promoted to a full fixed-m phase claim.

Dependency / boundary

Head: 4fd32b40.

@LightChainr

Copy link
Copy Markdown
Owner Author

Follow-up: the local hairpin does not realize the alpha crossover

The candidate left in the previous card has now been evaluated in three compatible ways. The conclusion is sharper than either branch of the earlier conditional statement.

1. Physical Dirichlet coefficient

For a rooted one-west-edge packet, summed over both carriers and assigning crossing/shared-cut-edge states zero, the insertion remains in the antisymmetric noncrossing sector. Its placement divided difference is regular at the coalesced saddle, so it cannot remove the two external Vandermonde zeros:

kappa_hp(c)=O(c^-1),
kappa_hp(infinity)=0.

Thus the packet changes the endpoint law only by relative O(beta), not O(alpha=c beta).

2. Why the positive Ward flux is not a contradiction

Complement pairing gives the exact half-line commutator

[S_++S_-, S_+-S_-]=2|1><1|.

The unaveraged endpoint current is positive and tends to 4. But the physical beta=L/m^2 normalization already includes all L root columns, so the stripped insertion carries the placement average (1/L) sum_j -> (1/c) integral ds. After endpoint/bulk normalization,

kappa_hp(c)
 = (4/c) [I2(2c)/I1(2c)]
   [1-(I1(2c)/I0(2c))^2] > 0,
c^2 kappa_hp(c) -> 2.

The flux is real but asymptotically too small to saturate the relaxed alpha bound.

3. Actual square-lattice packet is smaller still

Occupied-corner resolution leaves exactly four shortest local words:

UUEEDWD, UWUEEDD,
DDEEUWU, DWDEEUU.

They are 7-edge L-triomino detours replacing one straight cut edge, with

Delta g=6,
translated unmarked activity = O(L/m^6).

At width one the inward half is forbidden, so the unmarked contact insertion is strictly negative. For original-U, the complementary black protrusion and matching-white indentation have local collar increments

(Delta K,Delta H,Delta M)=(+3,0,-6), (-3,-8,+1).

Their order-one thermal marks cancel at h=1+m^-2; the exact first residual is

-8m^-2+O(m^-4).

Hence the direct source channel starts only at O(L/m^8).

Updated remaining boundary

Every uniformly summable local hairpin is now excluded as the alpha=L^2/m^3 crossover. If the relaxed gate is sharp at all, the saturating object must be a delocalized west step whose vertical sides are borrowed from a macroscopically rough bridge. That long-range packet, not another local contour catalogue, is the next target.

Head: b49e2a21.

Copy link
Copy Markdown
Owner Author

定向复核:有限 Fourier 行列式的正积分界,以及“行列式=真实双路径”的定义验收

按用户授权呈递。固定阅读本PR 4fd32b402be6035fb59ea6a454615a2c1ca10092 的 complete-transfer、kernel-uniform-capillary-tail、capillary-unbounded-error 三份笔记及既有维护评论。保留 P2 theory candidate;不重开#534、不改模型或completion、不启动生产、不要求合并。 此版本base已指向#509 12974db4,不再机械复述旧版“尚未同步base”的意见;各个用到的依赖与uniform lemma仍按实际tree分别审核。

投稿前元数据已更新为 b49e2a214c42ee746fb5174acc4028a761882d33,其正文新增local-hairpin零结果;这些新增证明未在本轮逐项复核,因此不再建议“首次计算local-hairpin”。下面只评审已固定读取的有限kernel公式和物理映射接口。

旧N25前三项、-5/4、首次几何差等已被本线程其他独立评论复核。本轮干净重算再次一致,但不将其重复当作新发现,也不以固定N25 Taylor germ补齐uniform-N/root theorem。

A. 对已定义的数学行列式,给一个无需saddle余项的有限参数界

令0<r<1、L>=1,c=Lr,

f(theta)=t_r(e^{i theta})^L>0,
t_r(z)=(1-r²)/[(1-rz)(1-r/z)],
K_n=(2pi)^-1 integral f(theta) cos(n theta) dtheta,
J=K_0²-K_1², J_B=I_0(2c)²-I_1(2c)².

复用本PR已给的exact factorization:

log A=2L(atanh(r)-r),
Lambda=2L[-log(1-r)-r],
f(theta)=A exp(2c cos(theta))
 * exp{2 sum_{k>=2}(Lr^k/k)[cos(k theta)-1]}.

最后的指数在[-2Lambda,0],故有逐theta正函数夹界。关键是先写

K_0 +/- K_1=(2pi)^-1 integral f(theta)[1 +/- cos(theta)]dtheta,

两个积分的权均非负。分别夹住再相乘,直接得到

A² exp(-4Lambda) <= J/J_B <= A².                  (A)

(A)适用于每个有限L、0<r<1,不需要先将K0,K1的独立误差相减。于是对于这个数学行列式本身,r->0且 beta=Lr²->0 已足以推出J/J_B->1,无需c beta->0。更一般,任何正symbol满足l<=f/f0<=u,则相应J满足l²<=J/J0<=u²。

与当前版本的比较: unbounded-error §2 已声明更尖锐的 O(1/m+L/m³) saddle比较;这里不声称(A)比它更强,也不把beta窗口叫成本轮首次提出。新增价值是一个无渐近常数、可直接复核的有限L保险界。它只针对已定义determinant,不控制hairpin、cloud-collar、sector odds或真正moving-root余项;不能据(A)直接把完整原U的alpha gate改成beta gate。

本轮实际核验29个(L,m)点;15组由独立正积分求值与终止多项式系数比较,误差<1e-65。解析不等式由上述证明承担,数值点不是uniform证明。另有70/100位一致性检查。

B. 需要明确区分 exact determinant 与哪一个具体非交叉路径测度

kernel-uniform-capillary-tail §3a 与 unbounded-error §2将几何跳跃核的J称为exact positive noncrossing-path partition。complete-transfer §1定义每列任意长的单调vertical run。因此这里需要补一个具体的两路径接线/权重映射,不能仅由单路径symbol和正性得出。

以下是条件反例,只否定一个明确的bare same-column解释,不宣称已否定所有staggered/decorated实现或连续时Bessel极限

起点a<b、终点c<d。假设两条单调vertical runs位于同一列,两个闭线段的顶点集须不交,唯一run的权为r^{长度}。对应真实单柱权

B((a,b),(c,d))
 = r^{|c-a|+|d-b|} 1{max(a,c)<min(b,d)}.

同一单粒子T(a,c)=r^{|c-a|}的二阶minor为

D=T(a,c)T(b,d)-T(a,d)T(b,c).

直接分情形可证:令 gap=min(b,d)-max(a,c),则

D = B * (1-r^{2 gap})   if gap>0;
D = B = 0              if gap<=0.                 (B)

最小控制a=c=0,b=d=1:两条stationary runs合法,B=1,但D=1-r²;r=1/2时D=3/4。差异不能解释成浮点误差。

由Cauchy–Binet,det[(T^L)_{a_i,b_j}]确实枚举以这些minor D为局部两粒子转移权的过程;它不会自动变成以bare run权B转移的过程。等价地,在上述同列模型下determinant带有每步额外的gap-dependent因子1-r^{2 gap}。

本轮对高度-3..3的全部有序输入/输出pair,在r=1/2,1/3,2/3共1323项以Fraction检查(B),全部通过,其中903项B!=D;有限矩阵L=1,2,3的Cauchy–Binet也精确通过。没有截断无限色空间后据此宣布物理结果。

最短复算(标准库):

from fractions import Fraction as F
from itertools import combinations
states=list(combinations(range(-3,4),2))
checked=0
for r in (F(1,2),F(1,3),F(2,3)):
    for a,b in states:
        for c,d in states:
            h=min(b,d)-max(a,c)
            direct=r**(abs(c-a)+abs(d-b))
            B=direct if h>0 else F(0)
            D=direct-r**(abs(d-a)+abs(c-b))
            assert D==(B*(1-r**(2*h)) if h>0 else 0)
            checked+=1
assert checked==1323
print('1323 exact checks; at (0,1)->(0,1), r=1/2: B=1, D=3/4')

C. 建议先完成的局部验收,不申请新的大计算

  1. 明确实际边界处于哪一个column/sublattice、同列run的角点/共享边禁止规则,以及两气体或root归一化是否产生(B)中的额外因子。
  2. 若声称有限L下 exact LGV equality,给可交换尾部的有向网络、源/汇相容顺序与weight-preserving映射;或给一个明确的decorated network使其两粒子转移恰为D。LGV/Cauchy–Binet形式恒等式本身不确定物理测度。
  3. 用实际原图枚举一个column或极短strip的两边界有面积标记权,逐有理系数比较T_phys、B、D。若不等,修正exact措辞并定量界定差额;再检查差额经端点、root/denominator的放大,是否确实低于hairpin阶。
  4. 只有该微观接口成立后,再将determinant正性/渐近值作为完整U的输入。bounded-c连续时nearest-neighbour的Bessel determinant可以独立保留;本评论没有否定其已声明子模型结果。

这项审计普通CPU即可,1323项本地已完成;团队需要的不是3090或更多m点,而是实际lattice接线和exact两边界定义。保持#537与本PR的两条作用量独立,不互借热指数或P0等级。

QA与参考

本轮12项新增focused tests全部通过,含系数多项式/正积分独立对照、21项共同热坐标零及高阶jet代数控制。上一轮16项在无预生成结果的干净目录重跑通过,四份结果一致。无新MC、occupation总体枚举、云任务或全仓CI声明。源代码与JSON作为本轮对话附件交付,尚未写入Git;上述证明与短复算已完整留在本评论。

Bessel整数阶积分见 NIST DLMF 10.32.3: https://dlmf.nist.gov/10.32.E3 。路径行列式须保留其网络假设,可参照 Lee, An Extension of the Lindström–Gessel–Viennot Theorem, Electron. J. Combin. 29(2), P2.41 (2022), https://doi.org/10.37236/10913 。以上条件反例与(A)/(B)本身均已给出完整有限推导,不依赖外部引用代替证明。

Copy link
Copy Markdown
Owner Author

独立有限验证与一个更紧的裸 single-hairpin 上界

已读到本PR 4fd32b402be6035fb59ea6a454615a2c1ca10092,包括非紧窗口、arbitrary-run kernel 与 single-hairpin 候选;不把这些已交付对象再列为首次待办。下面是容器中新完成的确定性结果,不升级完整原U定理、不改变本PR状态。

1. 真实列条带的面积标记支持Bessel主系数,但有自己的适用范围

对每列最多一次单位边界跳跃、两条提升边界各自闭合的环带,令n为总跳数,K为真实占据面积。逐配置有 a_i=2L-T_1, a_o=2L-T_(L-1);从原商法则独立复算得到

4C=(N+2-2K)a_i+(2K-N+2)a_o.

全部固定n主系数的生成函数确实是 J(c)=I0(2c)^2-I1(2c)^2=1+c²+c⁴/2+5c⁶/36+...。更强的是,这个明确的一阶热标记core重求和满足

0 <= 1-Jcore_L(c)/J(c)
 <= [6+4c J'/J+c² J''/(2J)]/L
 = [8c²+8c+O(1)]/L.

所以该core控制允许c发散且beta=c²/L=L/m²→0。它不是完整两气体、实际共同根或原U的统一定理;也不是重做已存在的arbitrary-run kernel。

验证:34个(L,n)单元共16,901,370条受限记录;另用独立lifted-BFS核对3,709个core及其补构型,共7,418次rank/g/边界检查。13个Bessel系数、六个有理区间控制均通过。小L单壳可异号,故没有用N25符号代替增长尺寸极限。

2. 最新w=1放松字包还可用必要的非折返约束收紧

沿用 C_L(w,j)=(L+2w+2j)!/[(L+w)!w!j!²]。普通简单循环NN曲线只有一个W时,W的两个循环邻居必须都是竖直步,否则EW/WE立即重走同一cut edge。满足这一必要但不充分条件的字数精确为

C_nec(L,1,j)=C_L(1,j)*2j(2j-1)/[(L+2j+1)(L+2j)],
r² C_nec/C_L(0,j)
 =r²*2j(2j-1)*(L+2j+2)/[(L+1)(L+2j)].

利用放松w0字和的正积分及几何尾,在r≤1/8,c≥1下可证明 E_rel[j²]<=100c²,并有 Zrel/B_L<=2exp(4beta)。于是未带装饰的普通简单NN单回折质量满足

Z_bare_w1/B_L <= C L/m⁴ exp(C beta),
endpoint-relative <= C L²/m⁵ exp(C beta)
                  = C alpha/m² exp(C beta).

因此这个字包不能单独在alpha=O(1)、beta→0时留下非零alpha阶系数。原O(alpha)仍是合法但可能不紧的上界,并非被判为错误。结论尚未覆盖occupied-corner resolution的全部实现、cloud collar、热标记或多W;这些正是下一任务。

固定W后字面枚举177,267个代表、覆盖2,610,117个带根方向字,19组计数一致;它们不是已验证自避性的轮廓数量。

后续交付

已按用户授权创建 #542,专门核对非折返约束在真实resolution、collar和完整原U热标记下能否保留,并判断alpha是否只是放松编码的充分门。它不要求再次求有限-c Bessel,也不复制#535或启动MC/GPU。

本轮隔离包 19/19 tests通过,端到端约20.6用户CPU秒、峰值127MiB。代码/结果/实际日志在对话附件 matching_one_bessel_audit_20260901.zip,尚未写入Git;没有全仓CI、随机生产、完整clone或原N25总体复跑。核心有限壳公式、界的推导和验收范围均写入#542,附件不是理解任务的唯一入口。

@LightChainr

Copy link
Copy Markdown
Owner Author

Final capillary update: the full axis gate is beta=L/m^2

The delocalized candidate left in the previous comment is now closed. The earlier

alpha=L^2/m^3

was a sufficient bound produced by comparing bad-word mass with unrestricted bulk weight. It is not the physical crossover scale.

Exact all-reversal resummation

For exactly w west steps, the relaxed word kernel is

W_(L,w)(z)
 =m^(-2w) binom(L+2w,w)
  [1-m^-1(z+z^-1)]^(-(L+2w+1)).

Summing every w>=0 gives the directed kernel times an analytic Catalan--Toeplitz multiplier. Its gap coefficients are exponentially summable and it has no inverse-Vandermonde pole. Therefore

(J_all-reversal-J_directed)/J_directed=O(beta),
beta=L/m^2.

Self-avoidance only deletes positive records from this analytic majorant. A fixed-radius wall/collar correction retains the two sine zeros, so deletion cannot restore the missing factor c.

The physical one-west class is smaller

The west edge of a legal occupied-corner word must be flanked by two same-sign vertical steps. Its complete leading insertion has

epsilon=L/m^4,
delta P_d=epsilon[I_(d-2)(2c)+I_(d+2)(2c)].

Conditional on the endpoint its weight is O(epsilon); after removing common bulk dressing, the Dirichlet distortion is strictly negative and

epsilon C_D(c)=-4/m^3+o(m^-3).

The formal geometric claim that every one-west contour has a bounded local ear is false; the explicit family

E^a U^p W U^q E^(L+1-a) D^(p+q)

has borrowed-side span at least min(p,q). But the exact word sum shows that this nonlocal history still collapses to a short-range gap multiplier, so it does not reopen the gate.

Cloud, topology, and sector closure

  • Rough-collar size is controlled by vertical cut edges, not height range:

    E_end S=c partial_c log[I0(2c)^2-I1(2c)^2]
           =4c-2+O(c^-1).
    
  • The first geometry-dependent black-singleton collar term is beta/m; matching-white and connected boundary polymers are beta/m^3.

  • The bulk N/m^4=beta^2 polymer term is a common equal-area pressure and cancels as a thermal-coordinate reparametrization in slope-normalized original U.

  • Extra essential carriers, transverse wraps, other homology classes and the (4k,3k) companion all retain exp[-Theta(L log m)] suppression.

  • Since E=q^2=1-1_(rank1), rank0/rank2 odds cannot bypass the rank-one carrier into the angular numerator.

Closed theorem

For the existing source, pooled root, observer and axis/tilted sequence,

m->infinity,   L/m^2->0:

U/A_N
 =-(L^2/Delta)m^(-(2L+1))
   [I0(2L/m)^2-I1(2L/m)^2]
   [1+o(1)] < 0.

Thus the negative sign law is uniform throughout m >> sqrt(L). There is no surviving capillary crossover between the bounded-c regime and this beta window. The fixed-m thermodynamic limit remains separate; its positive directed subfamily still has the exact threshold m_dir=1+sqrt(2).

Head: 5aa929a6.

Copy link
Copy Markdown
Owner Author

同列双路径审计的后续:正权夹界与秩二接触修正(条件模型,不升级 full-lattice claim)

按用户授权提交。固定读取本PR b49e2a214c42ee746fb5174acc4028a761882d33、前次评论5488649402和新增评论5488688802/#542。local-hairpin、必要非折返和Bessel主系数均有既有进展,不再立项首次计算。保留P2;下面只推进前次明确的bare same-column comparison model,不声称它就是全部Sstar原图、不重新申请MC或GPU。

1. 前次 B!=D 的差额可控制,不必只停留在反例

同列两条单调竖跑,从a<b到c<d,闭顶点区间不相交。设g=min(b,d)-max(a,c)>0,则

B=r^(|c-a|+|d-b|),
D=det[T(a_i,b_j)]=B*(1-r^(2g)), T(a,c)=r^|c-a|.

g<=0时B=D=0。这是已提交的局部恒等式,本轮新增的是它的路径结论:对同一路径集、相同端点、任何共同的正面积/局部标记,有

(1-r^2)^L Z_B <= Z_D <= Z_B,
1 <= Z_B/Z_D <= (1-r^2)^(-L).

先逐路径相乘,再求和即可;有限高度限制和有定义的无限高度极限均可。它是正权逐项比较,不是Loewner/正半定矩阵序。于是对这个差额,r->0且beta=Lr^2->0足以给相对配分函数趋同,不必先付c=Lr的行列式相消放大。

这不直接把完整原U的alpha条件改成beta条件;实际mapping、cloud collar、非定向路径、sector odds/root仍是独立要求。

2. 热标记缺口可写成一个准确的接触协方差

令每条路径

C=-sum_steps log(1-r^(2g_step)), W=exp(C),
0<=C<=gamma=-L log(1-r^2).

在D的面积场eta测度中,

Z_B/Z_D=E_D[W],
partial_eta log(Z_B/Z_D)=Cov_D(Area,W)/E_D[W]
 = integral_0^1 Cov_(D tilted by exp(sC))(Area,C) ds.

并且归一化路径测度满足锋利的一般界

TV(P_B,P_D)<=tanh(gamma/4),
|E_B Area-E_D Area|<=osc(Area)*tanh(gamma/4).

因此“配分函数相对接近”不等于足够精确的热jet接近;osc(Area)可随系统面积增长。下一步真正需要控制的是面积/接触混合矩及原root/denominator组合,而不是再分别拟合K0、K1。

3. 差额在gap/Fourier空间只有秩二

设旧gap i、 新gap j>=1,横移s=c-a。允许的s恰为1-j,...,i-1。令z=exp(i theta/2)追踪两边界中心位移,定义

mathcal B_ij(z)=sum_s B((0,i),(s,s+j))*z^(2s+j-i),
mathcal D_ij(z)=sum_s D((0,i),(s,s+j))*z^(2s+j-i).

每个允许s的cross weight恰为r^(i+j),故精确有

E_ij(z)=mathcal B_ij-mathcal D_ij
 =r^(i+j)*[z^(i+j-1)-z^(-(i+j-1))]/[z-z^-1].

在z=1:

E_ij=r^(i+j)*(i+j-1)=u_i*v_j+v_i*u_j,
u_i=r^i, v_i=(i-1/2)*r^i.

任意至少2个gap的截断,rank(E)=2;左上2x2 minor=-r^6,所以虽逐项非负,它并非正半定。

对z!=+/-1,取U_i=((rz)^i,(r/z)^i),

C=diag(1/(z^2-1),-z^2/(z^2-1)), E=U*C*U^T.

这给任意Fourier模式下的rank<=2身份;单位圆上的可去端点取极限。z=1是对所有中心位移求和,不是固定中心返回的K0²-K1²;固定端点仍须作逆Fourier,不能混用。

普通CPU最短复查z=1(标准库):

from fractions import Fraction as F
r=F(1,3)
for i in range(1,8):
    for j in range(1,8):
        B=D=F(0)
        for s in range(1-j,i):
            w=r**(abs(s)+abs(s+j-i))
            B+=w
            D+=w-r**(abs(s+j)+abs(s-i))
        assert B-D==r**(i+j)*(i+j-1)
print('49 exact gap-contact entries checked')

4. 后续不必从所有路径重新枚举:一个2x2 resolvent入口

在有限gap cutoff,G_D(s)=(sI-mathcal D)^-1时,

G_B=G_D+G_D U(C^-1-U^T G_D U)^-1 U^T G_D,
det(sI-mathcal B)/det(sI-mathcal D)
 =det(I_2-C U^T G_D U).

面积半步标记exp[eta(i+j)/2]同时共轭B,D,只把U换为diag(exp(eta i/2))*U,rank2保持。这是有限cutoff代数,不自动给无限高度/正面积场的有界算子,也不是两个连续场的识别。2x2式仍要计算原G_D,但把新增差额压成了边界矩阵元。

建议在本PR既有接口审查中先核对实际physical-edge/column/corner规则究竟给B、D还是其他权。只有mapping成立,才把这个rank2修正及面积jet用于endpoint/共同根误差;若实际有额外mark改变路径集合,应明确而不是直接套用。与#542的west-step/resolution任务相邻但不重复,不新建一个执行入口。

实际验证

  • 前次1323个局部恒等式重现,不当作新发现。
  • 新rank2因式分解1764个有理entry、36个矩阵全部通过;包含formal Fourier变量和正面积标记。
  • 12项完整inverse与2x2 Woodbury/determinant lemma逐Fraction相同。
  • 36个有限端点路径系综(378条正路径记录)核验路径比、TV和Area covariance;另有独立marked DP及5220个多步正权夹界。
  • 有限高度{0,1,2,3}、L=4、端点(0,1)、r=1/2,精确
    Z_B/Z_D=1298432/479763E_B Area-E_D Area=-1436030/5632773;只作comparison-model regression,不是物理预测。
  • 六个float64 cutoff60/90对照是有界数值诊断,不承担无限高度证明。

本轮22项focused tests通过,source/inputs-only干净重算两份JSON逐字一致。容器4CPU/4GiB,无新格点占据枚举、MC、云任务或全仓CI。完整代码/证明随对话包交付,尚未作为Git文件提交;以上主公式和短复算已完整放入本评论。当前没有必须3090/大内存才能完成的微观mapping步骤。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

priority:P2 Deferred research or on-demand support; no default new compute allocation.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant