\documentclass[11pt]{article} \usepackage[margin=1in]{geometry} \usepackage{amsmath,amssymb,amsthm,mathtools} \usepackage{enumitem} \usepackage[hidelinks]{hyperref} \numberwithin{equation}{section} \theoremstyle{plain} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \newtheorem{derivation}[theorem]{Derivation} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \newtheorem{construction}[theorem]{Construction} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\fg}{\mathfrak{g}} \newcommand{\Tr}{\operatorname{Tr}} \DeclareMathOperator{\Real}{Re} \DeclareMathOperator{\Var}{Var} \title{The Assumption Package of the Yang--Mills Construction:\\ Formal Definitions and Derivations\\ \large Companion to \emph{From Four-Dimensional Wilson Lattice Gauge Theory to Quantum Yang--Mills Theory, with a Conditional Mass-Gap Theorem}} \author{Chunping Zeng} \date{} \begin{document} \maketitle \begin{abstract} This companion extracts, from the Lean statement glossary of the validated formalization (Appendix~A of the validation document \cite{YMRepo}), the \emph{mathematical definitions} behind each external input of the construction---(RG1)--(RG5), (UV1), (PH1)--(PH3), (AN1), (Y1)--(Y4), (B1), (C1)---and records, for those inputs that are derivable from more primitive data, the derivations in full: the one-loop recursion of the marginal coupling, the growth bound of the inverse RG transport, the one-loop beta-function constant $b_0=\frac{11C_{\fg}}{48\pi^2}$ from the loop sum, and the short-distance reduction of the trace two-point function. Statements already proved in the main text are not reproved; they are cited by number. \end{abstract} \section*{Conventions} The main text is cited as [M]: Lemma~5.8, Theorem~9.9, etc. Everything defined here is a definition extracted from the formalization unless a derivation is explicitly given; nothing here upgrades an assumption to a theorem. The one-loop constant $b_0=\frac{11C_{\fg}}{48\pi^2}$ is fixed throughout. \section{The renormalization-group package}\label{sec:rg-defs} \subsection{(RG1): symmetry data}\label{subsec:rg1} \begin{definition}[Lattice symmetry group] $\Gamma_{\mathrm{lat}}$ is the finite group of the hypercubic lattice generated by translations by one lattice unit, the signed coordinate permutations (hyperoctahedral group, order $2^4\cdot4!=384$), and the reflection $\theta:(x_0,\vec x)\mapsto(-x_0,\vec x)$. \end{definition} \begin{construction}[Equivariant blocking kernel] A blocking kernel $C_{\rho}$ coarsening level $m$ (spacing $a(m)$) to level $m-1$ (spacing $a(m-1)=b_ma(m)$) is a Markov kernel on field configurations assigning to each coarse edge a local average of its $b_m$-dilated preimage with kernel profile $\rho$ of compact support. Symmetry-preserving gauge fixing (averaging over the coset of a fixed axial gauge; the invariance is the finite-group averaging of [M, Lemma~5.2]) makes $C_{\rho}$ $\Gamma_{\mathrm{lat}}$-equivariant. \end{construction} \subsection{(RG2): one-step estimates}\label{subsec:rg2} \begin{definition}[Effective action decomposition] At level $m$ the effective action has the form \[ S_m(U)=\beta_m\sum_{\square}\Big(1-\frac1m\Real\Tr U_{\square}\Big)+J_m(U), \] where $\beta_m$ is the marginal (plaquette) coefficient and $J_m$ lies in the finite-dimensional space of irrelevant local terms (plaquette correlators of diameter $\ge2$), normed by $\|J_m\|_m=\sum_{\tau}|c_{\tau}|b_m^{-\mathrm{diam}(\tau)}$. \end{definition} \begin{definition}[One-step estimates] With $\varrho\in(0,1)$, $K>0$, $\beta_\ast>0$ and the dominance $2b_0\log3>K/\beta_\ast$: \[ \|J_{m+1}\|_{m+1}\le\varrho\|J_m\|_m+K\beta_m^{-1},\qquad \big|\beta_{m+1}-\beta_m-2b_0\log b_m\big|\le K\beta_m^{-1},\qquad \beta_m\ge\beta_\ast , \] uniformly in the volume index $p$. \end{definition} \begin{derivation}[The marginal recursion] The second estimate is the discretization of the one-loop beta function. With $g^2=1/\beta$, \[ \frac{\mathrm d}{\mathrm d\log\mu}\,g^{-2}(\mu) =-\frac{2\beta(g)}{g^3}=2b_0+O(g^2) \] ([M, Lemma~9.9]). Integrating over one blocking step $\mu_{m-1}=\mu_m/b_m$ gives \[ \beta_{m+1}-\beta_m-2b_0\log b_m =\int_{\log\mu_{m-1}}^{\log\mu_m}\!\!O(g^2(\mu))\,\mathrm d\log\mu . \] The two-loop term $\beta(g)=-b_0g^3+\beta_2g^5+\cdots$ contributes $O(g_m^2)=O(\beta_m^{-1})$ to the integral, since $\log b_m$ is schedule-controlled ($b_m=\lceil(\log\hat\beta_m)^2\rceil$ type growth, validation document Appendix~D) so that $\int O(g^2)\le K\beta_m^{-1}$. The same schedule control gives the weighted Ces\`aro form of the remainder recorded in [M, Lemma~5.8].\qed \end{derivation} \begin{remark}[Status] The contraction of irrelevant directions is a property of the blocked polymer gas (kernel-averaging); it is certified in the repository (YM.RG.Program, YM.RG.Calendar.Fits). No derivation from bare definitions is given here. \end{remark} \subsection{(RG3): the insertion readout}\label{subsec:rg3} \begin{definition}[Transport map and readout] For a degree-$d$ label with coefficient vector $c$ at level $N$, the RG transport $T_{N\to r}$ is the linear map of coefficient vectors induced by applying the blocking kernel $N-r$ times to the insertion [M, eq.~(4.7)]. The readout $\mathcal R(\Pi)=c|_{r(d)}$ is evaluation of the transported coefficient at the reference level $r(d)\le N$. \end{definition} \begin{derivation}[Growth of the inverse] Each blocking step acts on a degree-$d$ insertion with a norm factor at most $b_m^{d}$ (dilation of the support pattern); hence \[ \|T_{N\to r}^{-1}\|\le\prod_{m=r+1}^{N}b_m^{\,d}=D_N^{\,d}=a_N^{-d}, \] using $D_N=\prod_{m2\,\operatorname{diam}(K)\ \text{and no periodic image of }K\text{ meets }K\}, \] so that for every $p\ge p_0(K)$ the support of $f$ lies strictly inside $((-L_p/2,L_p/2)^4)^n$ and avoids all periodic images. This is the same threshold as in [M, Assumption~6.1 (UV1)]. \begin{lemma}[Riemann-sum localization] \label{lem:riemann} Let $f\in\mathcal T_n$, $K=\operatorname{supp}f$, and $p\ge p_0(K)$. Then for every level $N$, with $|K|$ the $4n$-dimensional Lebesgue measure of $K$, \begin{equation} a_N^{4n}\sum_{x_1,\dots,x_n\in\Lambda_{N,p}} \bigl|f(a_Nx_1,\dots,a_Nx_n)\bigr| \;\le\; \|f\|_{L^1(K)}+c_n a_N\|\nabla f\|_\infty |K| +O\!\left(a_N\|\nabla f\|_\infty|\partial K|\right), \tag{C.1} \end{equation} where $c_n=\sqrt{4n}/2$. In particular the right-hand side is independent of $L_p$. \end{lemma} \begin{proof} On each cell $\mathcal C_{N,x}$ of side $a_N$ centered at $a_Nx$ one has $|f(a_Nx)-\bar f_{N,x}|\le c_n a_N\|\nabla f\|_\infty$. Only cells meeting $K$ contribute; their number is $|K|a_N^{-4n}+O(|\partial K|a_N^{-4n+1})$. Multiplying by $a_N^{4n}$ and using $a_N^{4n}\sum_{x}\bar f_{N,x}=\int_K f=\|f\|_{L^1(K)}$ gives (C.1). For $p\ge p_0(K)$ the set of contributing cells is exactly the same for every admissible volume, hence no $L_p$ appears. \end{proof} \begin{remark}[Numerical check of (C.1)] With $\varphi(x)=\prod_i(1-x_i^2)_+$, $n=2$, $\int f=(4/3)^8=9.98872123$: the left side of (C.1) is identical in all three volumes $L_p=8,32,256$ once $L_p/2>1$ (values $5.96046$, $8.80632$, $9.68081$, $9.91095$, $9.96923$ at $a_N=1/2,1/4,1/8,1/16,1/32$), with error $\propto a_N$ and independent of $L_p$. The global bound $L_p^{4n}=L_p^8$ overestimates by about $10^{19}$ at $L_p=256$. \end{remark} \begin{lemma}[$p$-independence of the cumulant constant] \label{lem:B-p-indep} In the proof of [M, Lemma~5.13], the constant $B$ bounding the level-dependent part of the first cumulant by $B\|J_N\|_N$ may be taken of the form $B=B(v,f,K)$, independent of $N$ and of $p$ for all $p\ge p_0(K)$. \end{lemma} \begin{proof} We inspect each source of $B$: \begin{enumerate} \item[(i)] By (RG3) the RG transport from level $N$ back to $r(d)$ is invertible and independent of $N$ and $p$ ([M, Lemma~5.13]); its inverse obeys $\|T_{N\to r}^{-1}\|\le D_N^d=a_N^{-d}$ by the Derivation of \S\ref{subsec:rg3}, with $p$-independence supplied by [M, Lemma~4.15(4)] (shared kinetic paths). \item[(ii)] $\|J_N\|_N\le\varrho^N\|J_0\|_0+K/(\beta_\ast(1-\varrho))$ by [M, Lemma~5.10], which records ``all constants are $p$-independent by (RG2)''. \item[(iii)] (RG5) already carries the quantifier $\sup_{N,p}\sup_{x\in K}|\partial^\alpha_xK_{N,p}(v;x)|\le M_{\alpha,K,v}$. \item[(iv)] The only remaining possible $p$-dependence is whether $\operatorname{supp}f$ fits inside the finite-volume box; [M, Definition~3.1] turns this into the finitely many exceptions $p0 . \] This is [M, Lemma~5.10] and is $p$-uniform. The contraction rate $\varrho$ itself is a certificate of the blocked polymer gas (\texttt{YM.RG.Program}, \texttt{YM.RG.Calendar.Fits}); Remark~1.6 of this companion states that no derivation from bare definitions is given. \textbf{(B) Mechanism behind (RG5)/(UV1).} Reflection positivity (E2) $\Rightarrow$ OS form $\Rightarrow$ Cauchy--Schwarz plus permutation symmetry (E3) $\Rightarrow$ $|S_p(v;x)|\le\prod_i\langle A_i,A_i\rangle^{1/2}$ $\Rightarrow$ chessboard bounds by finite-volume two-point functions $\Rightarrow$ local boundedness by (RG5); equicontinuity follows from analyticity at separated points and the Cauchy integral formula on a polydisc of radius $\delta$, giving $\omega_K(r)=CM_Kr/\delta$. This expands [M, Lemma~6.2]; (UV1) remains an external input. As noted in [M, Lemma~6.6], these bounds are local on compact sets and therefore cannot supply the $s\to\infty$ rate required by (MG). \textbf{(D) Audit statement of (C1).} The adapter is a bijection on sites, links and plaquettes of $\Lambda_{N,p}=(\mathbb Z/(L_pD_N)\mathbb Z)^4$; normalized Haar measures agree pointwise; the conversion $\beta_{\mathrm{submission}}=m\beta_{\mathrm{lgt}}$ equates the normalized finite laws; all finite moments then agree. By [M, Proposition~4.16] the internal proof of the adapter \emph{is} input (C1) (record only, enters no derivation), so no equivalence of the two independent formalizations is proved here. \section*{Boundary statement} \addcontentsline{toc}{section}{Boundary statement} All statements of this companion belong to the ultraviolet/definition layer. The counterexample family of [M, Proposition~10.7] satisfies (AN1) E0--E4, vacuum uniqueness, (B1), (Y1), (Y2), the short-distance constant [M, Theorem~9.9], the OPE data, \emph{and} the volume-uniform (RG4$'$) of \S\ref{sec:rg4-uniform}; its gap is still $0$. Hence deriving A--D in full leaves the registered input set compatible with both gap and no-gap. The remaining new mathematics is unchanged: a volume-uniform log-Sobolev inequality with $C_{\mathrm{LS}}\sim1/a_N$, a dynamical-to-Hamiltonian connection theorem with the $a_N$ factor, and a multiscale UV-summability/nonperturbative mass-generation identity (the missing dilation Ward / Callan--Symanzik intermediate term of [M, Proposition~10.9]). The mass gap is not proved. \begin{thebibliography}{3} \bibitem{YMRepo} Yang--Mills Shepherding Committee, \emph{Validation of the OpenAI Yang--Mills construction}, release of November 10, 2026, repository \texttt{yang-mills-group/YM} (Appendix~A: Lean statement glossary). \bibitem{M} C.~Zeng, \emph{From Four-Dimensional Wilson Lattice Gauge Theory to Quantum Yang--Mills Theory, with a Conditional Mass-Gap Theorem}, companion submission, 2026. \bibitem{LGT} Lattice gauge theory formalization, \texttt{mrdouglasny/lgt}, pinned revision. \end{thebibliography} \end{document}