\documentclass[11pt]{article} \usepackage[T1]{fontenc} \usepackage{lmodern} \usepackage{amsmath,amssymb,amsthm,mathtools} \usepackage{booktabs,tabularx,array} \usepackage{graphicx} \usepackage{placeins} \usepackage{enumitem} \usepackage{multicol} \usepackage{microtype} \usepackage{xcolor} \usepackage{hyperref} \usepackage[margin=1in]{geometry} \hypersetup{colorlinks=true,linkcolor=blue!55!black,citecolor=blue!55!black,urlcolor=blue!55!black} \makeatletter \makeatother \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\dd}{\mathrm d} \newcommand{\ii}{\mathrm i} \newcommand{\PhiN}{\Phi_L^{M,\sigma}} \newcommand{\sixj}[6]{\left\{\begin{matrix}#1\\#4\end{matrix}\right\}} \newcommand{\threej}[6]{\left(\begin{matrix}#1\\#4\end{matrix}\right)} \newcommand{\Op}{\mathsf} \makeatletter\g@addto@macro\UrlBreaks{\do\_}\newcommand{\eqalias}[1]{\ltx@label{#1}}\makeatother \newcommand{\SuppSec}[1]{Appendix~\ref{#1}} \title{Semiclassical Asymptotics of Wigner $3j$ and $6j$ Symbols\\at Growing Rank: Phase Horizons and Turning Connections} \hypersetup{pdfauthor={Antony Lewis},pdftitle={Semiclassical asymptotics of Wigner 3j and 6j symbols at growing rank: phase horizons and turning connections}} \newcommand{\Sussex}{Department of Physics \& Astronomy, University of Sussex, Brighton BN1 9QH, UK} \author{\href{https://orcid.org/0000-0001-5927-6667}{Antony Lewis}\\[2pt]{\small\normalfont \Sussex}} \date{} \begin{document} \maketitle \begin{abstract} Wigner $3j$ and $6j$ symbols encode angular-momentum coupling and recoupling. Their large-spin asymptotics connect to tetrahedral geometry, three-dimensional quantum gravity and quantum topology. We study the intermediate regime in which the rank of a two-large/one-small $3j$, one-small $6j$, or one-small $q$-$6j$ symbol grows with the large scale $N$. In Weyl-midpoint coordinates $j+\frac12$, their exact rank recurrences belong to a common class of even centered Jacobi chains. For this class we identify a generic phase-coherence horizon at rank $\asymp N^{2/3}$, where the leading quadratic phase-defect coefficient $\chi$ is nonzero. On compact turning-free corridors, a Hamilton--Jacobi effective-angle pullback approximates the physical solution to $O(N^{-2})$ in natural gauge, uniformly from fixed rank to fixed positive rank ratio; a real discrete-flux amplitude improves the bound to $O(N^{-3})$ without a new phase correction. Bulk zeros are located to $O(N^{-4})$, with an $O(N^{-3})$ Bohr--Sommerfeld law. At simple rank turns with lattice-quantized action we obtain a physically normalized Airy connection; under a forbidden-side bound, its turning-layer bracket is $\operatorname{Ai}(t)+N^{-4/3}\varpi\operatorname{Ai}'(t)+O(N^{-2})$. Exact hyperbolicity proves that bound for all three symbol families and extends the uniform Airy law through a fixed forbidden-side neighbourhood except for an $O(N^{-1}\log N)$ collar at the outer end of that neighbourhood. For $6j$ and $q$-$6j$ the rank momentum is a Euclidean or hyperbolic tetrahedral dihedral angle, and the $q$-$6j$ turning action obeys an exact Schl\"afli-type quantization identity. \end{abstract} \setcounter{tocdepth}{1} \begin{multicols}{2} \footnotesize\tableofcontents \end{multicols} \section{Introduction} Wigner $3j$ and $6j$ symbols describe how angular momenta couple and recouple. They appear throughout atomic, molecular and nuclear physics~\cite{Edmonds1957}, in the harmonic analysis of fields on the sphere such as the cosmic microwave background~\cite{CarronLewis}, and, through the $6j$ symbol and its $U_q(\mathfrak{sl}_2)$ deformation, in spin-network models of three-dimensional quantum gravity~\cite{PonzanoRegge1968,Roberts1999,TaylorWoodward}, and in quantum topology, where the growth of quantum $6j$ symbols at roots of unity is governed by volumes of hyperbolic tetrahedra and feeds the volume conjecture for Turaev--Viro invariants~\cite{Costantino2007,ChenYang2018}. For large quantum numbers their behaviour is geometric: when all six spins of a $6j$ symbol are large it oscillates with the Regge action of a tetrahedron whose edge lengths are $j+\frac12$~\cite{PonzanoRegge1968,Roberts1999}, and when one angular momentum is small it reduces to a rotation matrix element. This paper is about the regime between these two limits, in which the small angular momentum grows with the others, from fixed values up to a fixed fraction of them. In the Edmonds regime, where one angular momentum (the rank, $L$ for $3j$ and $f$ for $6j$) is small compared with the others, Wigner $3j$ and $6j$ symbols reduce to Wigner $d$-functions~\cite{Flude}. McFarlane improved these limits by averaging the large quantum numbers paired by the coupling, choosing the average by matching rank recurrences~\cite{McFarlane1992,McFarlane1993}, and Morehead derived the averaged $3j$ form from the Weyl correspondence with an $O(N^{-2})$ error~\cite{Morehead1996}. Our earlier fixed-rank paper~\cite{LewisFixedRank} made this exact at fixed rank: in midpoint variables built from the $SU(2)$ Weyl lengths $j+\frac12$, where the Casimir $j(j+1)=(j+\frac12)^2-\frac14$ is invariant under the reflection $j+\frac12\mapsto-(j+\frac12)$, the symbols admit all-order expansions in even inverse powers $N^{-2}$ away from caustics. Once the rank grows with $N$, however, the first surviving $O(N^{-2})$ local correction grows quadratically in the rank and accumulates an order-one phase, so every fixed-rank truncation eventually loses phase coherence. Describing what replaces it, and how far the replacement reaches, is the aim of this paper. Two structural facts govern this transition. First, write $q$ for the rescaled rank, $p$ for its conjugate momentum and $h=N^{-1}$ for the lattice step. A centered nearest-neighbour recurrence is then an exact lattice Weyl quantization, \begin{equation} \operatorname{Op}_h^W\!\bigl(a(q)e^{ip}\bigr)u(q)=a(q+h/2)u(q+h), \end{equation} so centering eliminates the subprincipal terms linear in $h$ that would otherwise come from operator ordering. Second, for the Hahn, centered Racah, and centered $q$-Racah recurrences, reflected Casimir factors are exchanged by $h\mapsto-h$, which forces an even Weyl symbol. A coherent interaction-picture increment of size $k^s/N^r$ therefore accumulates a phase of size $K^{s+1}/N^r$ and breaks down at the coherence horizon $K\asymp N^{r/(s+1)}$; for $s=r=2$ this gives the generic bulk $N^{2/3}$ law (where the coefficient $\chi(x)$ of the defect is nonzero). Beyond that horizon the natural object is a discrete Hamilton--Jacobi eikonal rather than a fixed-rank Taylor polynomial. We prove the growing-rank theory once, for a class of recurrences that contains all three families. An \emph{even centered Wigner--Jacobi chain} (Definition~\ref{def:even-centered-chain}) is a three-term rank recurrence in Weyl-midpoint gauge whose coefficients are $h^2$-analytic deformations of the Wigner rank recurrence with integer or half-integer offsets $(a,b)$, together with an $h^2$-analytic starting value (seed). For every chain in this class we prove the following. \begin{enumerate}[label=(\roman*),leftmargin=2.2em,itemsep=1pt] \item \emph{Automatic contact.} At every fixed rank the second-order construction satisfies the exact recurrence up to an $O_k(N^{-4})$ residual: the second-order fixed-rank contact identity holds identically across the class (Proposition~\ref{prop:general-fixed-rank-contact}), with no identity specific to the Hahn, Racah or $q$-$6j$ families needed. \item \emph{The $N^{2/3}$ coherence horizon.} With $\kappa=L+\frac12$ and $x=\cos\theta$, the $N^{-2}$ Taylor series converges for $\kappa^3/N^2$ small (Theorem~\ref{thm:general-mesoscopic}), and at $\kappa\asymp N^{2/3}$ the Wigner phase acquires the order-one shift $-\chi(x)\kappa^3/(3N^2\sin\theta)$ (Theorem~\ref{thm:general-critical}). This is the coherence horizon where $\chi(x)\ne0$. Where $\chi$ vanishes (for Hahn at $x=0$, and for Racah and $q$-$6j$ along a codimension-one strict-triangle locus when the two large pairs are unequal) the shift disappears. \item \emph{Phase-exact all-rank pullback.} The Darboux-renormalized second-order pullback $W^{[2]}_q\,d^L_{ab}(\Theta^{[2]}_q)$, a Wigner function evaluated at a corrected Hamilton--Jacobi angle, approximates the physical solution to $O(N^{-2})$ in natural gauge, uniformly from fixed rank through fixed positive ratio $q=\kappa/N$ (Theorem~\ref{thm:general-second-order}). Its phase is correct to $O(N^{-3})$: up to that order the remainder is a real discrete-flux amplitude factor $1+N^{-2}\alpha(q,x)$, with $\alpha=\delta_{\rm fl}-\mathsf J_2$ explicit and $\delta_{\rm fl}=0$ for Hahn and Racah (Theorem~\ref{thm:general-phase-exactness}). \item \emph{Bulk zeros.} Interior polynomial zeros lie within $O(N^{-4})$ of the pullback zeros and satisfy an $O(N^{-3})$ Bohr--Sommerfeld quantization law (Theorem~\ref{thm:general-bulk-zeros}). \item \emph{Second-order Airy connection.} At a simple rank turn with lattice-quantized turning action, the physical solution is a normalized Airy function. When it grows at most polynomially in $N$ beyond the turn (the forbidden-side hypothesis (R)), its turning-layer bracket is $\operatorname{Ai}(t)+N^{-4/3}\varpi_\tau\operatorname{Ai}'(t)+O(N^{-2})$, with $\varpi_\tau$ given in closed jet and Olver forms. On the part $I'$ of a fixed turning neighbourhood $I$ consisting of the oscillatory side and an initial stretch of the forbidden side, the approximation has Olver coefficient $B_\tau(q)$ and an explicit $O(N^{-2})$ flux-matched amplitude, leaving an $O(N^{-3}\mathsf M/\mathsf E)$ error (Theorem~\ref{thm:general-rank-airy}, Proposition~\ref{prop:airy2-amplitude}). \item \emph{Forbidden-side decay.} For the Hahn, centered Racah, and one-small quantum $6j$ chains, exact hyperbolicity beyond the turn proves recessivity without orthogonality inputs, and extends the uniform $O(N^{-3}\mathsf M/\mathsf E)$ law to the end of $I$ except for a collar of width $O(N^{-1}\log N)$ (Propositions~\ref{prop:exact-hyperbolicity} and~\ref{prop:forbidden-decay}). \end{enumerate} Sections~\ref{sec:growing3j} and~\ref{sec:growing6j} verify the class axioms for the fixed-offset Hahn/$3j$ and centered Racah/$6j$ chains (integer and half-integer offsets) and supply closed actions, $\mathsf J_2(0,x)$, closed Hahn and Racah $\varpi_\tau$, and exact turning-action integers; for Hahn the rank turn at a fixed node coincides with the spatial band edge. Section~\ref{sec:geometry} identifies the Racah action with the Ponzano--Regge action of the principal Euclidean Weyl tetrahedron and with the bispectral generator of the dual $c$-lattice problem. Section~\ref{sec:q6j} brings the one-small $U_q(\mathfrak{sl}_2)$ quantum $6j$ symbol ($q_{\rm def}=e^{1/N}$) into the same framework via the $t$-dependent off-diagonal factor of (E1): its Weyl lengths form a hyperbolic triangle, its rank momentum is a dihedral angle of a hyperbolic tetrahedron, its turning action is quantized exactly by a Schl\"afli-type identity, and the full second-order, phase-exact (with flux defect $\delta_{\rm fl}=-\frac1{12}$ in the normalization of the quantum $6j$ symbol used here), zero, and Airy theorems carry over. Global fixed-ratio Hahn asymptotics are due to Lin and Wong~\cite{LinWong}, and varying-parameter Racah and difference-equation turning-point asymptotics are developed in~\cite{GeronimoSmithWKB,WangWong,GeronimoBrunoVanAssche,WangWongRacah}. What is new here is not the existence of WKB or Airy approximations, but the following: (i) the exact physical Weyl-midpoint gauge and the coherent $N^{2/3}$ phase horizon; (ii) quantitative matching from fixed rank to fixed ratio via automatic second-order contact, yielding an all-rank $O(N^{-2})$ pullback that is phase exact to $O(N^{-3})$; and (iii) physically normalized turning connections with closed second-order Olver coefficients $\varpi_\tau$, $O(N^{-2})$ Airy amplitudes, and intrinsic forbidden-side decay. This recurrence-based analysis complements geometric and integral-representation treatments of Wigner symbols~\cite{PonzanoRegge1968,Roberts1999,ReinschMorehead1999,AquilantiHaggardLittlejohnYu,DupuisLivine2009,LittlejohnYu,BonzomFleury2012} and discrete orthogonal-polynomial methods~\cite{NikiforovSuslovUvarov,Braun1993,BaikKriecherbauerMcLaughlinMiller}. For the $U_q(\mathfrak{sl}_2)$ $6j$ symbol, all-large non-Euclidean tetrahedral asymptotics are due to Taylor and Woodward~\cite{TaylorWoodward} (cf.~\cite{MizoguchiTada} at roots of unity), and symmetric Jacobi formulations of the rank recurrence appear in~\cite{BonzomDupuisGirelli,KhavkineQ6j}. Section~\ref{sec:q6j} treats the complementary one-small growing-rank regime without using~\cite{TaylorWoodward}: we prove the midpoint recurrence from the Kirillov--Reshetikhin sum by a $q$-Zeilberger certificate, evaluate the minimal-rank seed in closed form, establish exact Schl\"afli turning-action quantization, and derive the flux defect $-\frac1{12}$ of the quantum $6j$ normalization and the second-order Airy data. The fixed-rank paper~\cite{LewisFixedRank} treats fixed rank uniformly across the entire spatial lattice, including hard edges and corners, but its fixed-rank $N^{-2r}$ coefficients grow with the rank and break down as the rank grows. The present paper treats growing rank across the turning-free, critical, and simple-turning regimes; the two uniformities act in complementary variables. Only the following inputs from Ref.~\cite{LewisFixedRank} are needed for the main results, all recalled explicitly below: the two exact recurrences and minimal-rank product structure in Section~\ref{sec:setup}, the fixed-rank $N^{-2}$ constant fixing $\delta_{\rm fl}=0$ for $3j$ and $6j$ (Corollaries~\ref{cor:3j-phase-exactness} and~\ref{cor:6j-phase-exactness}), and, in a single geometric remark (Remark~\ref{prop:6j-dual-principal-symbol}), the $c$-lattice Racah operator; the fourth-order comparison of Proposition~\ref{prop:D2D4match} also uses its $N^{-4}$ formula. Sections~\ref{sec:normalformbridge} and~\ref{sec:q6j} are self-contained; Ref.~\cite{LewisFixedRank} enters them only through the Hahn and Racah value $\delta_{\rm fl}=0$ (Remark~\ref{rem:flux-gauss}). Throughout, asymptotic families run along admissible integer/half-integer sequences satisfying the triangle and parity constraints, and analyticity in $h^2$ refers to the branch of the exact finite-product or recurrence expressions through the physical point at fixed continuum parameters. \begin{table}[ht] \centering \small \begin{tabularx}{\textwidth}{>{\raggedright\arraybackslash}p{0.27\textwidth}>{\raggedright\arraybackslash}p{0.25\textwidth}>{\raggedright\arraybackslash}X} \toprule Regime & Scale & Result\\ \midrule Mesoscopic / horizon & $L,f\asymp N^{2/3}$ & convergent $N^{-2}$ series below; first $O(1)$ phase shift at $N^{2/3}$ where $\chi(x)\ne0$\\ Turning-free bulk & $1\le L,f\lesssim N$ ($q,\rho$ bounded) & all-rank pullback, $O(N^{-2})$ natural gauge, phase exact to $O(N^{-3})$; zeros to $O(N^{-4})$\\ Simple turning window & $\zeta_\tau=O(N^{-2/3})$; $q\in I$ up to a log collar (general chain: $q\in I'$) & Airy bracket $\operatorname{Ai}+N^{-4/3}\varpi_\tau\operatorname{Ai}'+O(N^{-2})$; uniform $O(N^{-3}\mathsf M/\mathsf E)$ law on $I$ up to the collar (general chain: on $I'$)\\ \bottomrule \end{tabularx} \caption{Distinct asymptotic regimes across Hahn/$3j$, centered Racah/$6j$, and quantum $6j$ chains. The $N^{2/3}$ rank coherence horizon (generic: where $\chi\ne0$) is distinct from the $N^{-2/3}$ turning-point layer; $I'\subset I$ are the turning neighbourhoods of Theorem~\ref{thm:general-rank-airy}(ii) and Proposition~\ref{prop:forbidden-decay}.} \label{tab:roadmap} \end{table} \FloatBarrier \subsection{Scope and conventions} \label{sec:roadmap} \paragraph{Error and claim conventions.} All-rank bulk theorems bound the \emph{natural-gauge difference} $\sqrt{2k+1}\,|Y_k-U_k^{[2]}|\le CN^{-2}$, improving to $CN^{-3}$ with the real amplitude factor $1+N^{-2}\alpha$ of Theorem~\ref{thm:general-phase-exactness} ($O(N^{-5/2})$ and $O(N^{-7/2})$ unscaled at fixed positive ratio); no relative bound is claimed at zeros of $d^L_{ab}$. Bulk zero estimates are coordinate bounds for simple zeros of the off-lattice polynomial continuation, and Airy bracket errors are additive inside the bracket (or weighted by $\mathsf M/\mathsf E$ in the uniform law). Open problems are collected in Remark~\ref{rem:flux-gauss} and Section~\ref{sec:discussion}. \paragraph{Scope.} \begin{itemize}[nosep] \item The Racah and $q$-$6j$ turning theorems exclude $\tau=0$ (limiting triangles right-angled at the vertex opposite $\alpha$ or $\beta$, resp.\ $\mathsf X$ or $\mathsf Y$, where the rank turn meets the hard edge) and the degenerate corner $A-B=O(1)$, $C=O(B^{1/2})$ (Remark~\ref{rem:racah-collision-excluded}). \item The proportional-index Hahn regime (where minimal rank and caustic move at order $N$), operator-level Egorov/Fourier-integral pullbacks, Christoffel--Darboux kernels, and all-order Ponzano--Regge remainders are not treated. \end{itemize} \paragraph{Organization.} Section~\ref{sec:setup} recalls the exact $3j$ and $6j$ recurrences from Ref.~\cite{LewisFixedRank}. Section~\ref{sec:normalformbridge} defines even centered Wigner--Jacobi chains and proves the general theorems. Sections~\ref{sec:growing3j} and~\ref{sec:growing6j} specialize to Hahn/$3j$ and centered Racah/$6j$, closing with the tetrahedral and bispectral geometry (Section~\ref{sec:geometry}). Section~\ref{sec:q6j} treats the one-small quantum $6j$ chain, Section~\ref{sec:numerics} compares with exact symbols, and Section~\ref{sec:discussion} summarizes open problems and applications. Appendices~\ref{sec:supp-general}--\ref{sec:supp-rank-airy} complete the general proofs, Appendices~\ref{sec:supp-hahn}--\ref{sec:supp-racah} and~\ref{sec:supp-q6j} contain the chain-specific proofs, and Appendix~\ref{sec:supp-verification} records an a~posteriori Jacobi bound. \section{Exact Weyl-midpoint recurrences} \label{sec:setup} We recall from Ref.~\cite{LewisFixedRank}, without proof, the exact Weyl-midpoint recurrences and minimal-rank structure of the $3j$ and centered one-small $6j$ families; centering places every angular momentum $j$ at $j+\frac12$ so that $h=N^{-1}$ enters the recurrence coefficients exclusively through $h^2$. \subsection{The \texorpdfstring{$3j$}{3j} family} Let \begin{equation} j_\mp:=\frac{N-1\mp\sigma}{2},\qquad m_-:=\mu-\frac M2,\qquad m_+:=-\mu-\frac M2,\qquad x:=\frac{2\mu}{N},\qquad h:=N^{-1}, \end{equation} so $(j_++\frac12)+(j_-+\frac12)=N$ and $(j_++\frac12)-(j_-+\frac12)=\sigma$, and define the phase-adjusted normalized coefficient~\cite[Eq.~(17)]{LewisFixedRank} \begin{equation} \Phi_L^{M,\sigma}(x;h) :=(-1)^{2L+j_++m_+}\sqrt N\, \threej{L}{j_-}{j_+}{M}{m_-}{m_+}. \label{eq:PhiDef} \end{equation} We use standard Condon--Shortley/Wigner conventions~\cite{DLMF34}. The rank $L$ and fixed offsets $M,\sigma$ are all integers or all half-integers (in the latter case $N\in\mathbb Z+\tfrac12$); $2L$, $L\pm M$, $L\pm\sigma$, and $j_++m_+$ are always integers. The sign $(-1)^{j_++m_+}$ removes $(-1)^{2j_+}$ along fixed-$x$ sequences, while $(-1)^{2L}$ ensures the fixed-$L$ limit $+d^L_{M\sigma}(\arccos x)$ for half-integer $L$ as well. With $R_k:=\sqrt{(k^2-M^2)(k^2-\sigma^2)}$, the exact rank recurrence is~\cite[Eq.~(18)]{LewisFixedRank} \begin{equation} (2L+1)\bigl[L(L+1)x-M\sigma\bigr]\Phi_L =(L+1)R_L\sqrt{1-L^2h^2}\,\Phi_{L-1} +LR_{L+1}\sqrt{1-(L+1)^2h^2}\,\Phi_{L+1}, \label{eq:3jrank} \end{equation} and the magnetic recurrence gives the exact difference equation~\cite[Eq.~(30)]{LewisFixedRank} \begin{equation} a_h(x+h)\Phi(x+2h)+a_h(x-h)\Phi(x-2h)-2(1-x^2)\Phi(x) +2h^2\bigl[2L(L+1)-M^2-\sigma^2+1\bigr]\Phi(x)=0, \label{eq:3jmag} \end{equation} with $a_h(y)=\sqrt{[(1-y)^2-h^2(\sigma-M)^2][(1+y)^2-h^2(\sigma+M)^2]}$. Every appearance of $N$ is through $h^2$. At minimal rank $L_{\min}=\max(|M|,|\sigma|)$, $\Phi_{L_{\min}}$ is, up to sign and constant, the square root of an explicit finite ratio of centered factors $u_\pm+jh$ ($u_\pm=\frac{1\pm x}{2}$) and $1-j^2h^2$ ($j\in\tfrac12\mathbb Z$)~\cite[Eq.~(27)]{LewisFixedRank}, and at fixed $L$ and interior $x$, $\Phi_L^{M,\sigma}(x;h)=d^L_{M\sigma}(\arccos x)+f_2(x)h^2+f_4(x)h^4+\cdots$ with closed $f_2,f_4$~\cite[Theorems~6.1 and~6.3]{LewisFixedRank}. \subsection{The centered one-small \texorpdfstring{$6j$}{6j} family} Consider \begin{equation} W_f(a,b,c;m,n)= \sixj{a}{b}{c}{b+m}{a+n}{f}, \label{eq:6jfamily} \end{equation} with fixed admissible $m,n\in\tfrac12\mathbb Z$, $m-n\in\mathbb Z$. The Weyl-midpoint lengths ($A$ and $B$ the averages of the Weyl lengths $a+\frac12,a+n+\frac12$ and $b+\frac12,b+m+\frac12$, and $C=c+\frac12$) and their scaling are \begin{equation} A=a+\frac{n+1}{2}=N\alpha,\qquad B=b+\frac{m+1}{2}=N\beta,\qquad C=c+\frac12=N\gamma, \label{eq:ABC} \end{equation} with a strict limiting Euclidean triangle $(\alpha,\beta,\gamma)$. Set $x=(A^2+B^2-C^2)/(2AB)=\cos\theta$ and \begin{equation} \Psi_f(N):=(-1)^\xi 2\sqrt{AB}\,W_f, \qquad \xi:=a+b+c+f+m\in\mathbb Z, \label{eq:6jPhi} \end{equation} in the Edmonds/Condon--Shortley convention for which $\Psi_f(N)\to d^f_{mn}(\theta)$ at fixed $f$~\cite{Flude} (since $d^f_{nm}=(-1)^{m-n}d^f_{mn}$, the kernel of Ref.~\cite{LewisFixedRank} with exponent $a+b+c+f+n$ equals $(-1)^{m-n}\Psi_f$). With \begin{equation} R_k^{(6)}:=\sqrt{(k^2-m^2)(k^2-n^2)},\quad \eta_k:=\sqrt{\Bigl(1-\frac{k^2}{4A^2}\Bigr)\Bigl(1-\frac{k^2}{4B^2}\Bigr)},\quad \delta_f^{(6)}:=\frac{2f(f+1)+1-m^2-n^2}{8AB}, \end{equation} the exact centered Racah recurrence is~\cite[Lemma~7.1]{LewisFixedRank} \begin{equation} (2f+1)\{f(f+1)[x-\delta_f^{(6)}]-mn\}\Psi_f =fR_{f+1}^{(6)}\eta_{f+1}\Psi_{f+1} +(f+1)R_f^{(6)}\eta_f\Psi_{f-1}. \label{eq:6jrec} \end{equation} reducing at $N=\infty$ to the Wigner rank recurrence for $d^f_{mn}(\theta)$, and the minimal-rank value at $f_{\min}=\max(|m|,|n|)$ is a single-term finite product of factorial ratios~\cite[Lemma~8.1 and Eq.~(147)]{LewisFixedRank}. \section{Even centered Wigner--Jacobi chains} \label{sec:normalformbridge} This section states the growing-rank theory once, for an abstract class of recurrences that contains both chains of Section~\ref{sec:setup}. The class is defined by two properties: an exact Weyl-midpoint Jacobi form whose coefficients are analytic and even in $h=N^{-1}$ and reduce at $h=0$ to the Wigner rank recurrence, and an analytic stretched starting vector. The fixed-rank contact identity used by the second-order theorem then holds automatically (Proposition~\ref{prop:general-fixed-rank-contact}). Sections~\ref{sec:growing3j} and~\ref{sec:growing6j} verify these properties for the Hahn/$3j$ and centered Racah/$6j$ chains and supply their explicit data. \subsection{Exact Weyl structure and the class} \label{sec:class} \begin{proposition}[Exact even midpoint structure] \label{prop:exact-weyl-quantization} \label{prop:exact-midpoint-reversibility} \label{prop:exact-recoupling-midpoint} \label{lem:quadratic-lattice-parity} For a finite Fourier symbol $c(q,p;h)=\sum_m c_m(q;h)e^{imp}$ define \begin{equation} (\operatorname{Op}_h^W(c)u)(q)=\sum_m c_m(q+mh/2;h)u(q+mh). \label{eq:lattice-weyl-def} \end{equation} Hence \begin{equation} \operatorname{Op}_h^W(a(q)e^{ip})u(q)=a(q+h/2)u(q+h). \label{eq:weyl-halfstep} \end{equation} Any recurrence \begin{equation} [x-b(q;h^2)]Y(q)=a(q+h/2;h^2)Y(q+h)+a(q-h/2;h^2)Y(q-h) \label{eq:generic-exact-midpoint} \end{equation} is therefore the exact Weyl eigenvalue equation \begin{equation} \operatorname{Op}_h^W\!\left(b(q;h^2)+2a(q;h^2)\cos p\right)Y=xY. \label{eq:exact-weyl-symbol} \end{equation} It is invariant under step reversal $h\mapsto-h$. Where $a(q\pm h/2;h^2)>0$, its frozen transfer $T(q)=\bigl(\begin{smallmatrix}(x-b(q;h^2))/a(q+h/2;h^2)&-a(q-h/2;h^2)/a(q+h/2;h^2)\\1&0\end{smallmatrix}\bigr)$ has determinant $a(q-h/2;h^2)/a(q+h/2;h^2)$, so after the scalar flux normalization (multiplying $T(q)$ by $\sqrt{a(q+h/2;h^2)/a(q-h/2;h^2)}$) it has determinant one. In the elliptic region $|x-b(q;h^2)|<2\sqrt{a(q+h/2;h^2)a(q-h/2;h^2)}$, its eigenvalues are $e^{\pm ip_h}$ with \begin{equation} \boxed{\cos p_h(q,x)=\frac{x-b(q;h^2)}{2\sqrt{a(q+h/2;h^2)a(q-h/2;h^2)}}.} \label{eq:exact-shadow-dispersion} \end{equation} If moreover $a(q;t)$ and $b(q;t)$ are analytic in $(q,t)$ near $t=0$ (as (E1) of Definition~\ref{def:even-centered-chain} gives), then $p_h=p_0+h^2p_2+h^4p_4+\cdots$, convergent on a compact elliptic set with $a_0>0$; evenness of the recurrence in $h$ alone does not give this (for $a(q;t)=1+\sqrt t$, $b=0$, $x=1$ one has $p_h=\pi/3+|h|/\sqrt3+O(h^2)$). For a rank recurrence with lattice $q_k=(k+\tfrac12)h$, the half-rank coordinate is the unique affine coordinate compatible with the quadratic-lattice reflection $s\mapsto-s-1$, since $s(s+1)=(s+\tfrac12)^2-\tfrac14$. \end{proposition} \begin{proof} The Weyl identity is immediate from the Fourier-mode definition. Step reversal exchanges the two half-step hoppings. The relation $k(k+1)=q_k^2h^{-2}-1/4$ places quadratic-lattice diagonal terms at the midpoint. \end{proof} We first fix the Wigner base chain. Let $a,b\in\tfrac12\mathbb Z$ with $a-b\in\mathbb Z$, put $k_{\min}=\max(|a|,|b|)$ and $\kappa=k+\tfrac12$; the ranks are $k\in k_{\min}+\mathbb Z_{\ge0}$, integers or half-integers with $a,b$. In the natural gauge $\widehat d_k:=\sqrt{2k+1}\,d^k_{ab}(\arccos x)$ the Wigner rank recurrence is the Jacobi equation \begin{equation} x\widehat d_k=A_{k+1}^{(0)}\widehat d_{k+1}+A_k^{(0)}\widehat d_{k-1}+B_k^{(0)}\widehat d_k, \qquad A_k^{(0)}=\frac{\sqrt{(k^2-a^2)(k^2-b^2)}}{k\sqrt{4k^2-1}},\quad B_k^{(0)}=\frac{ab}{k(k+1)}, \label{eq:wigner-natural-jacobi} \end{equation} with $A_{k_{\min}}^{(0)}:=0$ (the displayed quotient is $0/0$ when $|a|=|b|=\tfrac12$) and $B_0^{(0)}:=0$ when $a=b=0$; in the exceptional channel $a=b=0$ row $k=0$ of \eqref{eq:wigner-natural-jacobi} is $x\widehat d_0=\tfrac1{\sqrt3}\widehat d_1$, whereas ungauged rank recurrences such as \eqref{eq:3jrank} and \eqref{eq:6jrec} vanish identically at $k=0$ and iterate from $k\ge1$. Since $2A_k^{(0)}=1+O(k^{-2})$ and $B_k^{(0)}=O(k^{-2})$, the Wigner chain is a summable perturbation of the constant elliptic chain. The ordered Condon--Shortley sign and the interior Darboux formula are \begin{equation} \epsilon_{ab}:=(-1)^{\max(b-a,0)},\qquad d^k_{ab}(\vartheta)=\epsilon_{ab}\sqrt{\frac{2}{\pi\kappa\sin\vartheta}} \left[\cos\!\left(\kappa\vartheta+\frac{p_{ab}\pi}{2}-\frac\pi4\right)+O(\kappa^{-1})\right], \label{eq:wigner-darboux-sign} \end{equation} with $p_{ab}:=|a-b|$, uniformly for $\vartheta$ in compact subsets of $(0,\pi)$. \begin{definition}[Even centered Wigner--Jacobi chain] \label{def:even-centered-chain} Let $\mathcal P$ be a compact parameter set and $X\Subset(-1,1)$ compact. A family of real recurrences, indexed by $h=N^{-1}$ along an admissible sequence and by $(\pi,x)\in\mathcal P\times X$, \begin{equation} x\widehat Y_k=A_{k+1}(h)\widehat Y_{k+1}+A_k(h)\widehat Y_{k-1}+B_k(h)\widehat Y_k, \qquad k\ge k_{\min}\ \text{if }(a,b)\ne(0,0),\quad k\ge1\ \text{if }(a,b)=(0,0), \label{eq:class-jacobi} \end{equation} with $A_{k_{\min}}(h)\widehat Y_{k_{\min}-1}:=0$ (in the exceptional channel $a=b=0$, row $k=0$ is assumed only where stated in Theorems~\ref{thm:general-phase-exactness} and~\ref{thm:general-bulk-zeros}), together with a distinguished \emph{physical solution} $\widehat Y_k=\sqrt{2k+1}\,Y_k$, is an \emph{even centered Wigner--Jacobi chain} with offsets $(a,b)$ if: \begin{enumerate}[label=(E\arabic*),leftmargin=3em] \item \emph{Even midpoint coefficients.} There are $s_1>0$, $t_0>0$ and functions $e_\pi(s,t)$, $v_\pi(s,t)$, independent of $x$ and $k$, continuous in $(\pi,s,t)$, analytic in $(s,t)$, and uniformly bounded on $\mathcal P$ times a fixed complex neighbourhood of $[0,s_1]\times\{|t|\le t_0\}$, real for real arguments, with $e_\pi(0,0)=1$, $v_\pi(0,0)=0$ and $e_\pi(s,0)\ge c>0$ on $[0,s_1]$, such that, with $\lambda_\pi(t):=e_\pi(0,t)$, \begin{equation} A_k(h)=A_k^{(0)}\,e_\pi(k^2h^2,h^2),\qquad B_k(h)=\lambda_\pi(h^2)\,B_k^{(0)}+v_\pi(q_k^2,h^2),\qquad q_k=\kappa h, \end{equation} exactly, for all ranks with $q_k^2\le s_1$ (so also $k^2h^2\le s_1$). \item \emph{Analytic stretched seed.} There are a bounded open set $U\supset X$ with $\overline U\subset\mathbb C\setminus\bigl((-\infty,-1]\cup[1,\infty)\bigr)$ and $t_0>0$ such that at the two lowest ranks $k_{\min},k_{\min}+1$ the physical solution is the value at $t=h^2$ of a function $\widehat Y_k(x;t)$ analytic and bounded in $(x,t)$ jointly on $U\times\{|t|\le t_0\}$, uniformly for $\pi\in\mathcal P$, real for real arguments, whose value at $t=0$ is the Wigner value $\widehat d_k(\arccos x)$ (which continues analytically to $U$, being a polynomial in $x$ times $(1-x)^{\mu}(1+x)^{\nu}$ with fixed exponents $\mu,\nu\in\tfrac12\mathbb Z$; when $k_{\min}>0$, or when $a=b=0$ and the row $k=0$ holds, joint analyticity at $k_{\min}$ alone gives $k_{\min}+1$ from \eqref{eq:class-jacobi} after shrinking $t_0$). \end{enumerate} No further hypothesis is needed for the second-order theory: the fixed-rank contact property it uses follows from (E1) (Proposition~\ref{prop:general-fixed-rank-contact}). \end{definition} \begin{remark}[Analyticity of the seed in $x$] \label{rem:E2-x} Joint analyticity in $(x,t)$ is used twice: it makes the fixed-rank datum $c(x)$ of Theorem~\ref{thm:general-phase-exactness}, and with it $\alpha(q,x)$, analytic in $x$, and it supplies the continuation of the exact solution to a complex neighbourhood of $X$ in the proof of Theorem~\ref{thm:general-bulk-zeros}. Analyticity in $t$ alone, uniformly for real $x$, does not give either. For $e_\pi\equiv1$, $v_\pi\equiv0$ (the Wigner chain itself) the solution $\widehat Y_k=(1+tf(x))\,\widehat d_k$ with $f$ smooth and bounded but not analytic solves the recurrence and is analytic in $t$ at fixed real $x$, uniformly on $X$; here $\mathsf J_2\equiv0$ and $c=f$, so $\alpha(q,x)=f(x)$ in Theorem~\ref{thm:general-phase-exactness}. Shrinking $U$ and $t_0$ if necessary, $\widehat d_{k_{\min}}(\arccos x)$ has no zero on $\overline U$ and $|\widehat Y_{k_{\min}}(x;t)-\widehat d_{k_{\min}}(\arccos x)|\le C|t|$ on $U\times\{|t|\le t_0\}$ by Cauchy's estimate, so for $N$ large the lowest-rank seed has no zeros on $U$. The Hahn, Racah and quantum $6j$ seeds are closed products and satisfy the joint form (Propositions~\ref{prop:hahn-class}, \ref{prop:racah-class} and~\ref{prop:q6j-class}). \end{remark} The Hahn and centered Racah chains have $e_\pi(s,t)=e_\pi(s)$ independent of $t$, so $\lambda_\pi\equiv1$; the one-small quantum $6j$ chain of Section~\ref{sec:q6j} does not. The $t$-dependence is needed because the Wigner diagonal $B_k^{(0)}=ab/(k(k+1))$ is deformed there by the factor $\lambda_\pi(h^2)=e_\pi(0,h^2)$ (Remark~\ref{rem:E1-t}). Thus $Y_k(h)$ is analytic in $t=h^2$ at every fixed rank, with $Y_k(0)=d^k_{ab}(\arccos x)$, because the forward hopping $A_{k+1}(h)$ is analytic and nonzero near $t=0$. We write \begin{equation} \begin{gathered} e_\pi(s,t)=1-\varepsilon_1s+\lambda_1t+\cdots,\qquad v_\pi(s,t)=\beta_1s+\beta_2t+\cdots,\\ \chi(x):=x\varepsilon_1-\beta_1,\qquad \eta_\pi(s):=\partial_t\log e_\pi(s,t)\big|_{t=0}, \end{gathered} \label{eq:class-first-coefficients} \end{equation} so $\eta_\pi(0)=\lambda_1$ (dots: terms of second order in $(s,t)$). We write $e_\pi(s):=e_\pi(s,0)$ for the \emph{principal coefficient}. For Hahn and Racah $\lambda_1=0$ and $\eta_\pi\equiv0$. The principal data of the chain are defined as follows. Put $a_0(q):=\tfrac12e_\pi(q^2)=\tfrac12e_\pi(q^2,0)$ and $b_0(q):=v_\pi(q^2,0)$. The principal symbol is $b_0(q)+2a_0(q)\cos p$, so the principal eikonal, discriminant, action, effective angle and transport are \begin{equation} \begin{gathered} \cos p(q,x)=\frac{x-b_0(q)}{e_\pi(q^2)},\qquad D(q,x):=e_\pi(q^2)^2-[x-b_0(q)]^2=4a_0^2\sin^2p,\\ S(q,x):=\int_0^qp(s,x)\,\dd s,\qquad \Theta_q(x):=\frac{S(q,x)}q,\qquad W_q(x):=\left[\frac{\sin^2\Theta_q(x)}{D(q,x)}\right]^{1/4}, \end{gathered} \label{eq:class-principal-data} \end{equation} on the branch $p\to\theta=\arccos x$ as $q\downarrow0$. Since $p$ is analytic in $q^2$, $\Theta_q$ is even and analytic at $q=0$, with \begin{equation} \Theta_q=\theta-\frac{\chi(x)}{3\sin\theta}q^2+O(q^4),\qquad W_q=1+\frac{(\varepsilon_1-x\beta_1)/2-x\chi(x)/6}{1-x^2}\,q^2+O(q^4). \label{eq:class-small-q} \end{equation} A \emph{turning-free corridor} is a compact set $\mathcal C$ of points $(q,x,\pi)$ with $q^2\le s_1-\delta_{\mathcal C}$ for some $\delta_{\mathcal C}>0$ (so that, for large $N$, the neighbouring ranks $k\pm1$ of every corridor rank also satisfy (E1)) and $D>0$ that contains $(r,x,\pi)$ for $0\le r\le q$ whenever it contains $(q,x,\pi)$; on it $\sin p$, $e_\pi$ and $\sin\Theta_q$ are bounded below. The \emph{leading pullback} at rank $k$ is \begin{equation} U_k:=W_{q_k}(x)\,d^k_{ab}\bigl(\Theta_{q_k}(x)\bigr),\qquad \widehat U_k=\sqrt{2k+1}\,U_k. \label{eq:class-leading-pullback} \end{equation} Equation~\eqref{eq:class-small-q} follows from $\cos p=x+\chi q^2+O(q^4)$, $\Theta_q=q^{-1}\int_0^qp$ and $\sin^2\Theta_q/D=(1-x^2-\tfrac23x\chi q^2)/(1-x^2-2(\varepsilon_1-x\beta_1)q^2)+O(q^4)$. \begin{remark}[The $t$-dependence of $e_\pi$] \label{rem:E1-t} Dividing the recurrence by $\lambda_\pi(h^2)$ gives a chain whose off-diagonal coefficient equals the Wigner one at $q=0$, at the shifted spectral parameter $x/\lambda_\pi(h^2)=x-\lambda_1xh^2+\cdots$. Hence the $t$-dependence of $e_\pi$ has two effects, both of order $h^2$ and invisible at leading order: the shift $x\mapsto x-\lambda_1xt$ at fixed rank, and the fixed-ratio term $\eta_\pi(q^2)$ in the order-$h^2$ coefficient of the off-diagonal symbol, \eqref{eq:class-a2b2}. Everything that depends only on the principal data (the principal symbol, $\chi$, the critical phase law, the mesoscopic series, the first-order pullback, the turning-point geometry) is unchanged; the second-order data pick up $\eta_\pi$ and $\lambda_1$, and these terms are displayed where they occur. The $h^4/q^2$ coefficient of $B_k-B_k^{(0)}$ in the quantum $6j$ chain is $mn(m^2+n^2-1)/6$, nonzero when $mn\ne0$, so without the factor $\lambda_\pi$ that chain would not satisfy the analyticity of $v_\pi$. \end{remark} \begin{remark}[Scope of the class] \label{rem:class-scope} \label{prop:master-principal-geometry} The class is defined by the deformation of the Wigner chain, not by a special-function identity. Sections~\ref{sec:growing3j} and~\ref{sec:growing6j} show that the fixed-offset Hahn/$3j$ chain ($e=\sqrt{1-s}$, $v=0$) and the centered Racah/$6j$ chain ($e$ a product of two square roots, $v$ affine) belong to it, and Section~\ref{sec:q6j} that the one-small quantum $6j$ chain does, with $e_\pi(s,t)$ genuinely $t$-dependent. At principal level the Hahn rank symbol is a degeneration of the Racah one (\SuppSec{sec:supp-racah}). \end{remark} \begin{lemma}[Interior Wigner--Darboux expansion] \label{prop:wigner-darboux-first-phase} \label{lem:wigner-darboux-differentiated-closure} Fix $a,b\in\tfrac12\mathbb Z$ with $a-b\in\mathbb Z$ and $J\Subset(0,\pi)$, and put $\kappa=k+\tfrac12$. Uniformly on $J$, \begin{gather} \sqrt{2k+1}\,d^k_{ab}(\vartheta) =(\sin\vartheta)^{-1/2}\sum_{\pm}c_{ab}^{\pm} \exp\!\left[\pm i\left(\kappa\vartheta+ \frac{\mathfrak c_{ab}(\vartheta)}{\kappa}\right)\right]+O_J(k^{-2}), \label{eq:wigner-darboux-first-phase}\\ \mathfrak c_{ab}(\vartheta):=\frac{(a^2+b^2-1/4)\cos\vartheta-2ab}{2\sin\vartheta}. \label{eq:wigner-darboux-phase-coeff} \end{gather} with $c_{ab}^\pm:=\pi^{-1/2}\omega_{ab}^{\pm1}$, $\omega_{ab}:=\epsilon_{ab}e^{i(p_{ab}\pi/2-\pi/4)}$ (so that the leading term is \eqref{eq:wigner-darboux-sign}). More generally, for every $R,s$ there are smooth branch amplitudes $A_{r,ab}^{\sigma}$, independent of $k$, such that \begin{equation} \sqrt{2k+1}\,d^k_{ab}(\vartheta)=\sum_{\sigma=\pm1}e^{i\sigma\kappa\vartheta}\Bigl[\sum_{r=0}^{R-1}\kappa^{-r}A_{r,ab}^{\sigma}(\vartheta)+\mathcal E^{\sigma}_{R,k}(\vartheta)\Bigr], \label{eq:wigner-darboux-diff-closure} \end{equation} where the remainders $\mathcal E^{\pm}_{R,k}$ are carrier-separated amplitude remainders: they stand inside the brackets, beside the carriers $e^{\pm i\kappa\vartheta}$, and \begin{equation} \max_{0\le m\le s}\ \sup_{\vartheta\in J}\bigl|\partial_\vartheta^m\mathcal E^{\pm}_{R,k}(\vartheta)\bigr|\le C_{J,R,s}\,\kappa^{-R}, \label{eq:wigner-darboux-diff-closure-rem} \end{equation} with $C_{J,R,s}$ independent of $k$. The derivatives fall on the amplitude remainders, not on the carriers; a derivative of a carrier costs a factor $\kappa$ and is taken explicitly. \end{lemma} \begin{proof} The expansion \eqref{eq:wigner-darboux-first-phase} is the standard interior Liouville--Green/Jacobi expansion; the first phase coefficient follows by integrating the $\kappa^{-1}$ correction in the Liouville normal form. The carrier-separated differentiated remainder \eqref{eq:wigner-darboux-diff-closure-rem} is proved in \SuppSec{sec:supp-general}: the undifferentiated Bessel-type (Frenzen--Wong/Hankel) expansion carries a remainder bounded only in the supremum norm, and the differential equation $w''+(\kappa^2-V)w=0$ of $w=(\sin\vartheta)^{1/2}\sqrt{2k+1}\,d^k_{ab}$ upgrades it, at the price of one power of $\kappa$ per derivative, which an expansion $s$ orders deeper absorbs. \end{proof} \subsection{Analytic transfer and the convergent mesoscopic series} \label{sec:transfer} The basic tool is the analytic oscillatory transfer bound of Lemma~\ref{lem:transfer} (\SuppSec{sec:supp-general}): if the transfer matrices are analytic in $t$ with order-$t^r$ coefficients $O(k^{2r})$ and the unperturbed fundamental matrix is naturally bounded, then in interaction form each order-$t^r$ insertion has size $O(k^{2r})$, and summing to rank $K$ produces the natural parameter $K^3t$. \begin{theorem}[Convergent mesoscopic even series] \label{thm:general-mesoscopic} Let $Y_L$ be the physical solution of an even centered Wigner--Jacobi chain, and write $Y_L(t)=\sum_{r\ge0}t^ry_{L,2r}(x)$ for its Taylor series at $t=h^2=0$, so that $y_{L,0}=d^L_{ab}(\arccos x)$. There are $\eta_*>0$ and $C<\infty$, independent of $L$, such that the series converges uniformly on $\mathcal P\times X$ whenever $|t|\kappa^3<\eta_*$, $\kappa=L+\tfrac12$, and \begin{equation} \sqrt{L+1}\,\sup_{\mathcal P\times X}|y_{L,2r}|\le C\eta_*^{-r}\kappa^{3r}. \label{eq:general-mesoscopic-coeff} \end{equation} For $0<\varrho<1$ and $|t|\kappa^3\le\varrho\eta_*$, \begin{equation} \boxed{\sqrt{L+1}\sup_{\mathcal P\times X}\left|Y_L(t)-\sum_{r=0}^{R-1}t^ry_{L,2r}\right|\le\frac{C\varrho^R}{1-\varrho}.} \label{eq:general-mesoscopic-tail} \end{equation} In particular, for every $R$ there is $C_R$ with $\sqrt{L+1}\,|Y_L-\sum_{rk_{\min}+1$, iterate $Z_{k+1}=T_k(t)Z_k$ for $k_{\min}+1\le k0$; without such a dichotomy the off-diagonal cancellation need not hold. \begin{theorem}[Critical phase law for even centered chains] \label{thm:general-critical} Let $Y_L$ be the physical solution of an even centered Wigner--Jacobi chain and $\chi$ as in \eqref{eq:class-first-coefficients}. Fix $\Lambda<\infty$. Uniformly on $\mathcal P\times X$, with $x=\cos\theta$, $\kappa=L+\tfrac12$, $L\to\infty$ and $\kappa^3/N^2\le\Lambda$, \begin{equation} \sqrt{\frac{\pi\kappa\sin\theta}{2}}\,Y_L =\cos\!\left(\varphi_{L,ab}(\theta)-\frac{\chi(x)}{3\sin\theta}\frac{\kappa^3}{N^2}\right) +O_\Lambda\!\left(\frac1\kappa+\frac{\kappa^2}{N^2}+\frac{\kappa^5}{N^4}\right), \label{eq:general-critical} \end{equation} where $\varphi_{L,ab}$ is the standard large-rank Wigner-$d$ phase, $\sqrt{\pi\kappa\sin\theta/2}\,d^L_{ab}(\theta)=\cos\varphi_{L,ab}(\theta)+O(\kappa^{-1})$; by \eqref{eq:wigner-darboux-sign}, $\varphi_{L,ab}(\theta)=\kappa\theta+\frac{p_{ab}\pi}2-\frac\pi4$, plus $\pi$ when $\epsilon_{ab}=-1$. In particular, if $\kappa^3/N^2\to\lambda$, the additional phase tends to $-\lambda\chi(x)/(3\sin\theta)$, and $\sqrt{L+1}\,|Y_L-d^L_{ab}(\theta)|=O(\kappa^3/N^2)$ for $L=o(N^{2/3})$. The additional phase is of order one at $\kappa\asymp N^{2/3}$ exactly where $\chi(x)\ne0$; at a point with $\chi(x)=0$ (for Hahn, $x=0$; for Racah, $x=2\alpha\beta/(\alpha^2+\beta^2)$ when $\alpha\ne\beta$; for the quantum $6j$ chain, $x=2\coth\mathsf X\coth\mathsf Y/(\coth^2\mathsf X+\coth^2\mathsf Y)$ when $\mathsf X\ne\mathsf Y$) the leading shift disappears, \eqref{eq:general-critical} then says only that $Y_L$ agrees with the Wigner phase up to the stated error, and the $N^{2/3}$ scale is not a horizon there. \end{theorem} \begin{proof} In the natural gauge the transfer of \eqref{eq:class-jacobi} for $k\ge k_{\min}+1$ is $T_k=\bigl(\begin{smallmatrix}(x-B_k)/A_{k+1}&-A_k/A_{k+1}\\1&0\end{smallmatrix}\bigr)$. By (E1) and \eqref{eq:class-first-coefficients}, $e_\pi((k+1)^2h^2,h^2)^{-1}=1+(\varepsilon_1k^2-\lambda_1)h^2+O(kh^2+k^4h^4)$, $\lambda_\pi(h^2)B_k^{(0)}=B_k^{(0)}+O(h^2k^{-2})$, $v_\pi(q_k^2,h^2)=\beta_1k^2h^2+O(kh^2+h^2+k^4h^4)$, and $(x-B_k^{(0)})/A_{k+1}^{(0)}=2x+O(k^{-2})$; moreover $e_\pi(k^2h^2,h^2)/e_\pi((k+1)^2h^2,h^2)=1+O(kh^2)$ because $\partial_s\log e_\pi(s,t)$ is bounded near $(0,0)$ (the $\lambda_1h^2$ term is $O(h^2)$ in $T_{11}$ and cancels to $O(kh^4)$ in $T_{12}$). Hence \begin{equation} T_k-T_k^{(0)}=2\bigl(x\varepsilon_1-\beta_1\bigr)\frac{k^2}{N^2}E_{11}+O\!\left(\frac{k}{N^2}+\frac{k^4}{N^4}\right), \end{equation} which is \eqref{eq:critical-transfer-assumption} with $\chi=x\varepsilon_1-\beta_1$. By (E2) the physical starting vector at $k_0\ge k_{\min}+1$ differs from the Wigner one by $O(N^{-2})$ in natural norm, and propagating it through the bounded critical transfer costs $O(N^{-2})$. The critical transfer normal form and $\sum_{k_0\le k-1$, accompanied by an oscillatory off-diagonal interaction with bounded partial sums, accumulates the phase $\chi K^{s+1}/((s+1)N^r)$ and therefore has the coherence horizon \begin{equation} \boxed{K\asymp N^{r/(s+1)}}. \label{eq:two-parameter-horizon} \end{equation} This is a heuristic organizing principle; the precise statement used in the proofs is the critical transfer normal form \eqref{eq:critical-transfer-assumption}--\eqref{eq:critical-diagonal-phase}. For an even midpoint chain the first defect has $s=r=2$, giving the $N^{2/3}$ horizon of Theorem~\ref{thm:general-critical} at every $x$ with $\chi(x)\ne0$. After $R$ even principal phase terms have been resummed, an omitted increment $k^{2R+2}/N^{2R+2}$ gives \begin{equation} \boxed{K_R\asymp N^{(2R+2)/(2R+3)}} \label{eq:even-phase-horizon-hierarchy} \end{equation} provided its coefficient is nonzero. The Hamilton--Jacobi action below resums this entire hierarchy. \end{remark} \subsection{Hamilton--Jacobi action and the leading pullback} \label{sec:general-first} Beyond the critical scale the correct object is not a longer fixed-rank Taylor polynomial but the discrete eikonal \eqref{eq:class-principal-data}. Its basic estimates are a turning-free adiabatic basis with fixed-start stability, and the first secular coefficient of the leading pullback (used for the $6j$ bound \eqref{eq:6j-g2-pullback-bound}); the all-rank error of the leading pullback follows from the second-order theory (Corollary~\ref{thm:general-first-order}). \begin{lemma}[Turning-free adiabatic rank basis] \label{lem:general-adiabatic} \label{lem:adiabatic-rank} \label{lem:adiabatic-racah} Let $\mathcal C$ be a compact set of points $(q,x,\pi)$ with $q^20$, put $h=N^{-1}$, $q_k=(k+\tfrac12)h$, and \begin{equation} z_k^\pm=D(q_k,x)^{-1/4}\exp\!\bigl(\pm\ii NS(q_k,x)\bigr). \label{eq:adiabaticbranches} \end{equation} If $K\to\infty$, $K/N\to0$, and $(q_k,x,\pi)\in\mathcal C$ for $K\le k\le L$, then the exact Jacobi system \eqref{eq:class-jacobi} has a fundamental matrix $\mathcal F_k^{(N)}$ satisfying \begin{equation} \mathcal F_k^{(N)}=P_k\left[I+O_{\mathcal C}\!\left(K^{-1}+L/N^2\right)\right], \qquad P_k=\begin{pmatrix}z_k^+&z_k^-\\z_{k-1}^+&z_{k-1}^-\end{pmatrix}, \label{eq:adiabaticfundamental} \end{equation} uniformly for $K\le k\le L$, in $(x,\pi)$, and in the final index $L$ as long as the path remains in $\mathcal C$. \end{lemma} \begin{proof} See \SuppSec{sec:supp-general}. \end{proof} \begin{corollary}[Fixed-start turning-free stability] \label{cor:general-fixed-start} \label{cor:adiabatic-fixed-start-hahn} \label{cor:adiabatic-fixed-start-racah} On a turning-free corridor there is a fixed admissible rank $k_*$ such that, for every fixed $k_0\ge k_*$ and all sufficiently large $N$, one may choose an exact naturally normalized fundamental matrix $G_k^{(N)}$ of \eqref{eq:class-jacobi} with \begin{equation} \sup_{k_0\le k\le L}\left(\|G_k^{(N)}\|+\|(G_k^{(N)})^{-1}\|\right)\le C_{\mathcal C,k_0} \end{equation} whenever the swept path stays in the corridor. \end{corollary} \begin{proof} See \SuppSec{sec:supp-general}. \end{proof} \begin{lemma}[First secular coefficient] \label{lem:general-first-coefficient} With $y_{L,2}$ the $t$-coefficient of Theorem~\ref{thm:general-mesoscopic} and \begin{equation} \mathcal P_2:=\frac{\chi(x)}{3}\,\partial_x+\frac{(\varepsilon_1-x\beta_1)/2-x\chi(x)/6}{1-x^2}, \label{eq:general-P2} \end{equation} one has $\sqrt{L+1}\,|y_{L,2}-\kappa^2\mathcal P_2d^L_{ab}(\arccos x)|\le C(L+1)$ uniformly on $\mathcal P\times X$. The operator $\kappa^2\mathcal P_2$ is the $t$-coefficient of the leading pullback $W_qd^L_{ab}(\Theta_q)$ at fixed rank, by \eqref{eq:class-small-q}. \end{lemma} \begin{proof} This follows from the fixed-rank contact of Proposition~\ref{prop:general-fixed-rank-contact}, whose proof (\SuppSec{sec:supp-general-second}) does not use the present lemma. At fixed rank the corrected pullback is $\widehat U^{[2]}_k=\widehat d_k+t\,\widehat u_k+O_k(t^2)$ with $\widehat u_k=\sqrt{2k+1}\,\mathcal Q_{2,k}d_k$, where \begin{equation*} \mathcal Q_{2,k}-\kappa^2\mathcal P_2=-\sqrt{1-x^2}\,\rho_2(0,x)\,\partial_x+\frac{x\,\rho_2(0,x)}{2\sqrt{1-x^2}}, \end{equation*} an operator with coefficients bounded on $X$ and independent of $k$. Since $\widehat Y$ satisfies the recurrence exactly and $\mathcal R_k(\widehat U^{[2]})$ has no $t$-coefficient, differentiating both at $t=0$ shows that $\delta_k:=\partial_t\widehat Y_k|_{t=0}-\widehat u_k$ solves the Wigner recurrence \eqref{eq:wigner-natural-jacobi} for $k>k_{\min}$; its values at $k_{\min},k_{\min}+1$ are bounded by (E2) and Proposition~\ref{prop:general-second-eikonal-pole}, and the Wigner propagator is uniformly bounded in natural gauge on $X$, so $|\delta_k|\le C$. As $\sqrt{2L+1}\,y_{L,2}=\widehat u_L+\delta_L$, and $\sqrt{2L+1}\,|d^L_{ab}|\le C$, $\sqrt{2L+1}\,|\partial_xd^L_{ab}|\le C\kappa$ on $X$ (Lemma~\ref{lem:wigner-darboux-differentiated-closure}), the claim follows. \end{proof} \begin{remark}[Action-Jacobian zero transport] \label{lem:action-zero-transport} Let $S(r,x)$ be a smooth turning-free action, put $\Theta(r,x)=S(r,x)/r$ and $x_r(x)=\cos\Theta(r,x)$, and define \begin{equation} \varrho_r(x):=-\frac{1}{\pi r}\partial_xS(r,x),\qquad \rho_0(y)=\frac{1}{\pi\sqrt{1-y^2}}. \end{equation} Then wherever $0<\Theta<\pi$, \begin{equation} \boxed{\varrho_r(x)\,dx=\rho_0(x_r)\,dx_r.} \label{eq:generic-action-zero-transport} \end{equation} Indeed $\partial_xx_r=-\sin\Theta\,S_x/r$ and $\rho_0(x_r)=1/(\pi\sin\Theta)$. Thus every effective coordinate obtained as the cosine of an action per unit rank transports the principal zero measure exactly; application-specific work enters only through the action and the finite-$N$ remainder. \end{remark} \subsection{Second Weyl eikonal and the second-order all-rank pullback} \label{sec:general-second} \label{sec:generic-second-weyl} The leading pullback misses an order-$h$ accumulated phase at fixed ratio although the local symbol defect is $O(h^2)$. The correction is the order-$h^2$ Weyl eikonal, renormalized at $q=0$ by the finite-index Darboux phase. \begin{proposition}[Second eikonal for an even midpoint Jacobi chain] \label{prop:generic-second-weyl-eikonal} Assume \eqref{eq:generic-exact-midpoint} with \begin{equation} a(q;h^2)=a_0(q)+h^2a_2(q)+O(h^4),\qquad b(q;h^2)=b_0(q)+h^2b_2(q)+O(h^4), \end{equation} on a compact elliptic set with $a_0>0$. Let $p(q,x)\in(0,\pi)$ and $g(q,x)$ be defined by \begin{equation} x-b_0(q)=2a_0(q)\cos p(q,x),\qquad g(q,x)=-\frac12\log(a_0(q)\sin p(q,x)). \label{eq:generic-principal-phase-transport} \end{equation} Then \begin{equation} e^g\exp\{ih^{-1}(S+h^2S_2)\},\qquad S_q=p, \end{equation} solves the midpoint equation through order $h^2$ if and only if $S_{2,q}=\Sigma_2[a_0,a_2,b_2;p,g]$, where \begin{align} \Sigma_2[a_0,a_2,b_2;p,g] =\frac{1}{2a_0\sin p}\Bigl[{}&a_0(g_q^2+g_{qq})\cos p-a_0g_qp_q\sin p -\frac{a_0}{4}p_q^2\cos p-\frac{a_0}{3}p_{qq}\sin p\\ &+2a_2\cos p+a_{0,q}g_q\cos p-\frac{a_{0,q}}2p_q\sin p +\frac{a_{0,qq}}4\cos p+b_2\Bigr]. \label{eq:generic-second-weyl-density} \end{align} \end{proposition} \begin{proof} Insert the displayed ansatz into \eqref{eq:generic-exact-midpoint}, Taylor-expand the two half-step coefficients and the phase, and use the principal eikonal and flux transport to cancel orders $h^0$ and $h^1$. The real order-$h^2$ scalar coefficient is exactly \eqref{eq:generic-second-weyl-density}; the details are in \SuppSec{sec:supp-general-second}. \end{proof} For an even centered Wigner--Jacobi chain the exact midpoint coefficients are $a(s;h^2)=A_k$ at $s=kh$ and $b(q;h^2)=B_k$ at $q=q_k$. Since $A_k^{(0)}=\tfrac12[(1-a^2h^2/s^2)(1-b^2h^2/s^2)/(1-h^2/(4s^2))]^{1/2}$, $B_k^{(0)}=abh^2/(q^2-h^2/4)$, $e_\pi(s^2,h^2)=e_\pi(s^2)\,(1+h^2\eta_\pi(s^2)+O(h^4))$ and $\lambda_\pi(h^2)=1+O(h^2)$ (which changes $b$ only at order $h^4$), their order-$h^2$ coefficients at fixed positive $q$ are \begin{equation} a_2(q)=a_0(q)\Bigl[\frac{1}{q^2}\Bigl(\frac18-\frac{a^2+b^2}{2}\Bigr)+\eta_\pi(q^2)\Bigr],\qquad b_2(q)=\frac{ab}{q^2}+\partial_tv_\pi(q^2,0). \label{eq:class-a2b2} \end{equation} The new term $\eta_\pi(q^2)$ is regular at $q=0$ and is zero for the Hahn and Racah chains. It adds $\eta_\pi(q^2)\cot p$ to $\Sigma_2$ below. Put $g=-\tfrac12\log(a_0\sin p)=-\tfrac14\log D+\mathrm{const}$ and \begin{equation} \begin{gathered} \Sigma_2:=\Sigma_2[a_0,a_2,b_2;p,g],\qquad \rho_2(q,x):=\Sigma_2(q,x)-\frac{d}{dq}\left(\frac{\mathfrak c_{ab}(\Theta_q(x))}{q}\right),\\ \Gamma_2(q,x):=\int_0^q\rho_2(s,x)\,\dd s,\qquad \Theta_q^{[2]}:=\Theta_q+\frac{h^2}{q}\Gamma_2,\qquad W_q^{[2]}:=\left[\frac{\sin^2\Theta_q^{[2]}}{D(q,x)}\right]^{1/4}, \end{gathered} \label{eq:class-second-data} \end{equation} with $\mathfrak c_{ab}$ from \eqref{eq:wigner-darboux-phase-coeff}. For the Hahn and centered Racah chains $\widetilde\Gamma=\Gamma_2+\mathfrak c_{ab}(\Theta_q)/q$ has a three-term algebraic closed form in $\sqrt D$ (companion supplement~\cite{JointUniformitySupplement}), so $\Theta^{[2]}_q$, and $B_\tau$ in Theorem~\ref{thm:general-rank-airy}, need no quadrature. The amplitude $W^{[2]}_q$ is not an extra fitted function: it keeps the principal complex-branch amplitude equal to $D^{-1/4}$ after the angle is moved. The \emph{corrected pullback} is \begin{equation} U_k^{[2]}:=W_{q_k}^{[2]}(x)\,d^k_{ab}\bigl(\Theta_{q_k}^{[2]}(x)\bigr),\qquad \widehat U_k^{[2]}=\sqrt{2k+1}\,U_k^{[2]}. \label{eq:class-corrected-pullback} \end{equation} \begin{proposition}[Darboux renormalization of the second eikonal] \label{prop:general-second-eikonal-pole} On every compact interior $x$-set, \begin{equation} \Sigma_2(q,x)=-\frac{\mathfrak c_{ab}(\theta)}{q^2}+O(1)\qquad(q\downarrow0),\qquad\theta=\arccos x, \label{eq:general-second-weyl-pole} \end{equation} and $\rho_2$ extends \emph{analytically in $q^2$} to $q=0$. Consequently $\Gamma_2/q$, $\Theta_q^{[2]}$ and $W_q^{[2]}$ are analytic in $q^2$ at $q=0$, and $U_k^{[2]}$ is analytic in $t=h^2$ at every fixed rank, with $U_k^{[2]}(0)=d^k_{ab}(\arccos x)$. \end{proposition} \begin{proof} By (E1), $a_0$, $b_0$, $p$, $g$, $\eta_\pi(q^2)$, $\partial_tv_\pi(q^2,0)$ and $\Theta_q$ are analytic in $q^2$ near $0$, with $a_0=\tfrac12+O(q^2)$, $p=\theta+O(q^2)$ and $g_q=O(q)$. In \eqref{eq:generic-second-weyl-density} only $2a_2\cos p+b_2$ has a pole (the term $\eta_\pi$ of $a_2$ is regular), so \begin{equation} \Sigma_2=\frac{x(1/8-(a^2+b^2)/2)+ab}{q^2\sin\theta}+O(1)=-\frac{\mathfrak c_{ab}(\theta)}{q^2}+O(1). \end{equation} All terms of the density are even meromorphic in $q$ with at worst a $q^{-2}$ pole. Because $\Theta_q=\theta+O(q^2)$ is even analytic, $\frac{d}{dq}(\mathfrak c_{ab}(\Theta_q)/q)=-\mathfrak c_{ab}(\theta)q^{-2}+O(1)$ is even meromorphic with the same sole pole, so the difference is analytic in $q^2$. Then $\Gamma_2$ is odd analytic and $\Gamma_2/q$ is even analytic; at fixed rank $q^2=\kappa^2t$. Integrating the derivative term against $-\mathfrak c_{ab}(\theta)q^{-2}$ also writes $\Gamma_2(q,x)=(\mathfrak c_{ab}(\theta)-\mathfrak c_{ab}(\Theta_q(x)))/q+\int_0^q(\Sigma_2(s,x)+\mathfrak c_{ab}(\theta)s^{-2})\,\dd s$ without nested integration of $\Theta_s$. \end{proof} \begin{proposition}[Automatic second-order fixed-rank contact] \label{prop:general-fixed-rank-contact} For every even centered Wigner--Jacobi chain and every fixed admissible rank $k$, the coefficient of $t$ in the Taylor expansion at $t=h^2=0$ of the natural-gauge residual \begin{equation} \mathcal R_k(\widehat U^{[2]}):=x\widehat U_k^{[2]}-A_{k+1}\widehat U_{k+1}^{[2]}-A_k\widehat U_{k-1}^{[2]}-B_k\widehat U_k^{[2]} \end{equation} vanishes identically on $\mathcal P\times X$; hence $\mathcal R_k(\widehat U^{[2]})=O_k(t^2)$. In the exceptional channel $a=b=0$ this includes $k=0$, with $B_0^{(0)}:=0$. Its $t^0$ coefficient vanishes because $U^{[2]}(0)$ is the Wigner solution. Moreover \begin{equation} \begin{aligned} \rho_2(0,x)=\frac{1}{12(1-x^2)^{3/2}}\Bigl[&\varepsilon_1\bigl(9x-x^3-8(a^2+b^2)x+4abx^2+12ab\bigr)\\ &+\beta_1\bigl(8(a^2+b^2)-16abx-2x^2-6\bigr)+12(\beta_2+\lambda_1x)(1-x^2)\Bigr]. \end{aligned} \label{eq:general-rho20} \end{equation} The $\lambda_1$ term is the shift $\beta_2\to\beta_2+\lambda_1x$ of Remark~\ref{rem:E1-t}; it is zero for Hahn and Racah. \end{proposition} \begin{proof} The closed form of $\rho_2(0,x)$ follows by expanding \eqref{eq:generic-second-weyl-density} at $q=0$; the contact identity reduces, through the Wigner rank recurrence and the ladder identities, to a rational identity in $(a,b,k,x)$. The computation is in \SuppSec{sec:supp-general-second}. \end{proof} \begin{theorem}[Weyl--Darboux midpoint engine] \label{thm:second-weyl-darboux-engine} Let $h=N^{-1}$, $p\in\{3,4\}$, $R:=7(p-1)/(5-p)+1$ ($R=8$ for $p=3$, $R=22$ for $p=4$), and let $U_k(h)$ be a corrected oscillatory pullback for a turning-free midpoint Jacobi chain. Assume: \begin{enumerate} \item for a fixed rank $k_0\ge1$ and all $k\ge k_0$ with $k^3h^2\le\eta_0$, the fixed-rank bound \begin{equation} |\mathcal R_k(\widehat U)|\le C h^4 k^6; \end{equation} \item for $k\ge k_0$ in the turning-free corridor, \begin{equation} |\mathcal R_k(\widehat U)|\le C(h^p+k^{-R}); \end{equation} \item the exact naturally gauged fundamental matrix and its inverse are uniformly bounded for $k_0\le k\le L$. \end{enumerate} Then, uniformly for every admissible terminal rank $L$ in the corridor with $k_0\le L=O(h^{-1})$, \begin{equation} \boxed{ \sup_{k_0\le k\le L} |\widehat Y_k-\widehat U_k|\le C h^{p-1},} \label{eq:abstract-second-order-all-rank} \end{equation} provided the two initial entries differ by $O(h^{p-1})$ at the fixed rank $k_0$. In particular the same bound holds simultaneously at fixed, mesoscopic, and fixed-positive-ratio rank as long as the swept corridor remains turning-free. On any subfamily for which the corrected comparison satisfies $|\widehat U_L|\ge c>0$, the same estimate is equivalently a relative error $\widehat Y_L/\widehat U_L=1+O(h^{p-1})$. \end{theorem} \begin{proof} Let $K_*:=\min\{L,\lfloor h^{-(5-p)/7}\rfloor\}$. Since $K_*^3h^2\le h^{(3p-1)/7}\to0$, the fixed-rank estimate of hypothesis~1 applies up to $K_*$, and $\sum_{k=k_0}^{K_*}h^4k^6=O(h^4K_*^7)=O(h^{p-1})$. If $L>K_*$, the high-rank estimate of hypothesis~2 gives $\sum_{k=K_*}^{L-1}(h^p+k^{-R})=O(Lh^p+K_*^{-(R-1)})=O(h^{p-1})$, since $K_*\asymp h^{-(5-p)/7}$, $(R-1)(5-p)=7(p-1)$, and $L=O(h^{-1})$ (if $L\le K_*$ the second sum is empty). Variation of constants with the uniformly bounded exact fundamental matrix of hypothesis~3 propagates the $O(h^{p-1})$ initial mismatch and these residual sums without amplification, which proves \eqref{eq:abstract-second-order-all-rank} for every terminal rank in the corridor. The split rank and the number of Darboux orders behind hypothesis~2 are set by the target global order, not by a special-function identity: $O(h^4k^6)$ sums to $O(h^4K^7)=O(h^{p-1})$ at $K\asymp h^{-(5-p)/7}$, where $O(k^{-R})$ sums to $O(K^{-(R-1)})$ and $R=8$ ($p=3$) or $R=22$ ($p=4$) is the smallest integer giving $O(h^{p-1})$. \end{proof} \begin{proposition}[The engine hypotheses for even centered chains] \label{prop:general-engine-inputs} \label{prop:hahn-second-engine-inputs} \label{prop:racah-second-engine-inputs} Let an even centered Wigner--Jacobi chain be given, and fix a closed turning-free corridor. For the corrected pullback \eqref{eq:class-corrected-pullback} there exist $k_0,\eta_0,C>0$ such that \begin{align} |\mathcal R_k(\widehat U^{[2]})|&\le Ch^4\kappa^6, &&\kappa^3h^2\le\eta_0,\label{eq:class-engine-small-rank}\\ |\mathcal R_k(\widehat U^{[2]})|&\le C(h^3+k^{-8}), &&k_0\le k\le L-1.\label{eq:class-engine-large-rank} \end{align} The exact naturally gauged fundamental matrix and its inverse are uniformly bounded on the same corridor, and the fixed starting entries of the exact and corrected physical branches differ by $O(h^2)$. \end{proposition} \begin{proof} The fixed-rank analytic bound \eqref{eq:class-engine-small-rank} follows from Proposition~\ref{prop:general-fixed-rank-contact} and Cauchy's estimate on the disk $|t|\le c\kappa^{-3}$, on which the residual is analytic and vanishes to second order at $t=0$. It is also bounded there: $q^2=\kappa^2t=O(\kappa^{-1})$ and $\Theta^{[2]}_q-\theta=O(\kappa^{-1})$, so $\kappa|\Im\Theta^{[2]}_q|=O(1)$, and the interior Wigner--Darboux bounds, which continue to complex angles with $\kappa|\Im\vartheta|=O(1)$ (Lemma~\ref{lem:bulk-zero-continuation}), bound $\sqrt{2k+1}\,d^k_{ab}(\Theta^{[2]}_q)$ uniformly. \SuppSec{sec:supp-general-second} proves the fixed-ratio $O(h^3)$ residual of the corrected pullback (Lemma~\ref{lem:general-cubic-residual}, a direct Taylor expansion of the Darboux branch that uses only (E1), Lemma~\ref{lem:darboux-parity} and Propositions~\ref{prop:generic-second-weyl-eikonal} and~\ref{prop:general-second-eikonal-pole}) and the Darboux--Weyl corner matching that combines it with Proposition~\ref{prop:general-fixed-rank-contact} to yield \eqref{eq:class-engine-large-rank} (Corollary~\ref{cor:general-matched-remainder}). Turning-free stability is Corollary~\ref{cor:general-fixed-start}, and the statement about starting entries follows from (E2) and Proposition~\ref{prop:general-second-eikonal-pole}. \end{proof} \begin{theorem}[Second-order all-rank effective-angle pullback] \label{thm:general-second-order} Let an even centered Wigner--Jacobi chain be given, and fix a closed turning-free corridor $\mathcal C$. Then, for every admissible rank $L$ whose swept path stays in $\mathcal C$, \begin{equation} \boxed{ \sqrt{2L+1}\,\bigl|Y_L-W_q^{[2]}\,d^L_{ab}(\Theta_q^{[2]})\bigr|\le\frac{C_{\mathcal C}}{N^2}, \qquad q=\frac{L+\frac12}{N}.} \label{eq:general-second-order} \end{equation} The constant includes the finitely many admissible ranks below the fixed starting rank of Theorem~\ref{thm:second-weyl-darboux-engine}. At fixed positive ratio this is the absolute error $O(N^{-5/2})$. \end{theorem} \begin{proof} Proposition~\ref{prop:general-engine-inputs} verifies the $p=3$ hypotheses of Theorem~\ref{thm:second-weyl-darboux-engine} for $U_k=U_k^{[2]}$; fixed ranks below its stability threshold are covered by the analytic seed (E2) and Proposition~\ref{prop:general-second-eikonal-pole}. Applying the engine and dividing by $\sqrt{2L+1}$ proves the estimate. \end{proof} \begin{corollary}[First-order all-rank effective-angle pullback] \label{thm:general-first-order} In the setting of Theorem~\ref{thm:general-second-order}, for every admissible rank $L$ whose swept path stays in $\mathcal C$, \begin{equation} \boxed{ \sqrt{2L+1}\,\bigl|Y_L-W_q\,d^L_{ab}(\Theta_q)\bigr|\le C_{\mathcal C}\,\frac{L+1}{N^2}, \qquad q=\frac{L+\frac12}{N}.} \label{eq:general-first-order} \end{equation} At fixed positive ratio this is $O(N^{-1})$ in natural gauge: the leading pullback misses the accumulated phase $h\Gamma_2(q,x)$ (Remark~\ref{rem:fixed-ratio-odd-solution}). \end{corollary} \begin{proof} By Proposition~\ref{prop:general-second-eikonal-pole}, $\rho_2$ is continuous on the closed swept paths in $\mathcal C$, so $\Gamma_2/q$ is bounded on $\mathcal C$, so $\Theta^{[2]}_q-\Theta_q=h^2\Gamma_2/q=O(N^{-2})$ and $W^{[2]}_q-W_q=O(N^{-2})$, and for large $N$ both angles lie in a fixed compact $J\Subset(0,\pi)$. On $J$, $\sqrt{2L+1}\,|d^L_{ab}|\le C$ and $\sqrt{2L+1}\,|\partial_\vartheta d^L_{ab}|\le C\kappa$ (Lemma~\ref{lem:wigner-darboux-differentiated-closure}; the finitely many low ranks are bounded directly). Hence $\sqrt{2L+1}\,|W^{[2]}_qd^L_{ab}(\Theta^{[2]}_q)-W_qd^L_{ab}(\Theta_q)|\le C\kappa N^{-2}$, and Theorem~\ref{thm:general-second-order} gives \eqref{eq:general-first-order}. \end{proof} \begin{remark}[Transport curvature and the first odd fixed-ratio correction] \label{rem:higher-order-pullback} \label{rem:fixed-ratio-odd-solution} The integral of the frozen shadow phase $p_2$ diverges and double-counts the finite-index Darboux phase. Subtracting the pole (Lemma~\ref{lem:general-shadow-pole}) leaves a bounded frozen density, but it tracks only the frozen characteristic phase. The additional terms in \eqref{eq:generic-second-weyl-density} are the Weyl transport-curvature contribution; using $\rho_2$ lowers the local residual by one full power of $h$. The exact centered operator is even in $h$, but a nonzero ratio $q=(L+\tfrac12)h$ couples the rank to $h^{-1}$, so the leading pullback misses the accumulated phase $(L+\tfrac12)(\Theta_q^{[2]}-\Theta_q)=h\,\Gamma_2(q,x)$. Thus the physical fixed-ratio expansion need not be even in $h$, although the operator and every fixed-rank coefficient are; including this odd correction improves the natural-gauge remainder from $O(h)$ to $O(h^2)$ uniformly from fixed rank through fixed ratio. \end{remark} \subsection{Phase exactness} \label{sec:general-phase-exactness} The mechanism of phase exactness is parity: the exact operator is invariant under $h\mapsto-h$ at fixed $q$, and the Wigner--Darboux expansion has an even amplitude and an odd phase in $\kappa^{-1}$. Hence the first residual of the corrected pullback that survives at fixed ratio, at order $h^3$, is of transport type and is removed by a real amplitude factor; no order-$h^2$ phase is produced. \begin{lemma}[Parity of the Wigner--Darboux expansion] \label{lem:darboux-parity} Fix offsets $a,b$, both integers or both half-integers (so $a-b\in\mathbb Z$, as for Wigner $d$), and $J\Subset(0,\pi)$. The branch amplitudes of \eqref{eq:wigner-darboux-diff-closure} can be written \begin{equation} A^{\pm}_{r,ab}(\vartheta)=\omega_{ab}^{\pm1}(\pm i)^r\,\mathfrak a_{r,ab}(\vartheta),\qquad \mathfrak a_{r,ab}\ \text{real},\qquad \mathfrak a_{0,ab}=(\pi\sin\vartheta)^{-1/2}, \label{eq:darboux-parity} \end{equation} with $\omega_{ab}$ as in Lemma~\ref{prop:wigner-darboux-first-phase}. Equivalently, for every $R$, \begin{equation} \sqrt{\tfrac{\pi}{4}\sin\vartheta}\;\sqrt{2k+1}\,d^k_{ab}(\vartheta) =\epsilon_{ab}\,\mathfrak A_{ab}(\vartheta,\kappa)\cos\!\Bigl(\kappa\vartheta+\frac{p_{ab}\pi}2-\frac\pi4+\mathfrak P_{ab}(\vartheta,\kappa)\Bigr)+O_J(\kappa^{-R}), \end{equation} with $\epsilon_{ab}$ as in \eqref{eq:wigner-darboux-sign}. Here $\mathfrak A_{ab}=1+\sum_{1\le j0$. Write $d^k_{ab}$ through the Jacobi polynomial $P_n^{(\alpha,\beta)}$ with $\alpha=p_{ab}=|a-b|$, $\beta=|a+b|$, $n=k-\max(|a|,|b|)$, so that $n+(\alpha+\beta+1)/2=\kappa$. The Bessel-type uniform expansion of Frenzen and Wong, DLMF~18.15.6~\cite{DLMF18,FrenzenWong}, gives \begin{multline} (\sin\tfrac\vartheta2)^{\alpha+\frac12}(\cos\tfrac\vartheta2)^{\beta+\frac12}P_n^{(\alpha,\beta)}(\cos\vartheta)\\ =\frac{\Gamma(n+\alpha+1)}{2^{1/2}n!\,\kappa^\alpha}\Bigl[\vartheta^{1/2}\sum_{m}\frac{\mathsf A_m(\vartheta)J_\alpha(\kappa\vartheta)}{\kappa^{2m}}+\vartheta^{3/2}\sum_m\frac{\mathsf B_m(\vartheta)J_{\alpha+1}(\kappa\vartheta)}{\kappa^{2m+1}}\Bigr] \end{multline} to any finite order, uniformly for $\vartheta\in[0,\pi-\delta]$, with real $\mathsf A_m,\mathsf B_m$ and remainder $O(\kappa^{-R})$ relative to the leading term on $J$. The Wigner normalization contributes $\sqrt{(n+\alpha)!(n+\alpha+\beta)!/(n!(n+\beta)!)}\cdot\kappa^{-\alpha}$ after combining with $\Gamma(n+\alpha+1)/n!$. Each of the factorial quotients $(n+\alpha+\beta)!/n!$ and $(n+\alpha)!/(n+\beta)!$ (or its reciprocal) is a product of linear factors $\kappa+c$ over a set of shifts $c$ symmetric about $0$, hence of the parity of their number, $\alpha+\beta$ and $|\alpha-\beta|$ respectively; these numbers have equal parity (both odd for half-integer offsets), so the product of the two is an even rational function of $\kappa$ with leading term $\kappa^{2\alpha}$; its square root times $\kappa^{-\alpha}$ is a convergent series in $\kappa^{-2}$. Finally the Hankel expansion of $J_\nu(\kappa\vartheta)$ (\cite[10.17.3]{DLMFAiryBessel}) is $\sqrt{2/(\pi\kappa\vartheta)}[\mathcal P_\nu\cos\varpi_\nu-\mathcal Q_\nu\sin\varpi_\nu]$ with $\varpi_\nu=\kappa\vartheta-\nu\pi/2-\pi/4$, $\mathcal P_\nu$ even and $\mathcal Q_\nu$ odd in $\kappa^{-1}$. Since $\varpi_{\alpha+1}=\varpi_\alpha-\pi/2$, collecting terms gives $\sqrt{2k+1}\,d^k_{ab}(\sin\vartheta)^{1/2}=(2/\sqrt\pi)\,[\mathcal C\cos\varpi_\alpha+\mathcal S\sin\varpi_\alpha]$ with $\mathcal C$ even and $\mathcal S$ odd in $\kappa^{-1}$, $\mathcal C=\pm1+O(\kappa^{-2})$; the factor $\sqrt{2\kappa}$ of the natural gauge cancels the $\kappa^{-1/2}$ of the Hankel expansion, so no fractional powers occur. Writing $\mathcal C\cos\varpi+\mathcal S\sin\varpi=\frac12\sum_\pm e^{\pm i\varpi}(\mathcal C\mp i\mathcal S)$ gives \eqref{eq:darboux-parity}, and $\mathfrak A=(\mathcal C^2+\mathcal S^2)^{1/2}$, $\mathfrak P=-\arctan(\mathcal S/\mathcal C)$ give the amplitude--phase form; $\varpi_\alpha$ and $\kappa\vartheta+p_{ab}\pi/2-\pi/4$ differ by $p_{ab}\pi$, a sign, and comparison of the leading terms with \eqref{eq:wigner-darboux-sign} shows that the overall sign is $\epsilon_{ab}$. The coefficient of $\kappa^{-1}$ in $\mathfrak P$ is fixed by \eqref{eq:wigner-darboux-first-phase}. \end{proof} On the continuum variable $q$, with $a(s;h^2)=A^{(0)}(s/h)e_\pi(s^2,h^2)$, $A^{(0)}(s/h)=\frac12[(1-a^2h^2/s^2)(1-b^2h^2/s^2)/(1-h^2/4s^2)]^{1/2}$ and $b(q;h^2)=\lambda_\pi(h^2)abh^2/(q^2-h^2/4)+v_\pi(q^2,h^2)$, the midpoint operator \begin{equation} (\mathcal L_hu)(q)=\bigl(x-b(q;h^2)\bigr)u(q)-a(q+\tfrac h2;h^2)u(q+h)-a(q-\tfrac h2;h^2)u(q-h) \label{eq:midpoint-operator} \end{equation} reproduces the natural-gauge residual on the lattice $q=q_k$, and $\mathcal L_{-h}=\mathcal L_h$ exactly. By Lemma~\ref{lem:darboux-parity} the corrected pullback splits into two Darboux branches, \begin{equation} \widehat U^{[2],\pm}_k:=W^{[2]}_{q_k}(x)\,e^{\pm i\kappa\Theta^{[2]}_{q_k}}\sum_{r0$ at fixed $h$ with $\kappa=q/h$, and put \begin{equation} \mathsf J(s):=\pi\,a(s;h^2)\,\Im\bigl(\overline{u(s-\tfrac h2)}\,u(s+\tfrac h2)\bigr). \end{equation} \begin{enumerate}[label=(\roman*)] \item $\pi\,\Im\bigl(\overline{u}\,\mathcal L_hu\bigr)(q)=-\bigl[\mathsf J(q+\tfrac h2)-\mathsf J(q-\tfrac h2)\bigr]$ exactly. \item $\mathsf J$ is even in $h$, and at fixed $s>0$, $\mathsf J(s)=\frac12+h^2\mathsf J_2(s,x)+O(h^4)$ with $\mathsf J_2$ given by \eqref{eq:phase-exactness-J2}. \item $\mathsf J_2$ extends to an even analytic function of $s$ at $s=0$, with $\mathsf J_2(0,x)$ given by \eqref{eq:phase-exactness-J20} in the notation \eqref{eq:class-first-coefficients}. \item With $C^+_3$, $B_0$ the leading coefficients of $\mathcal L_hu$ and $u$ at fixed ratio (Lemma~\ref{lem:cubic-residual-transport}), $\pi q^{-3}\Im\bigl(\overline{B_0}C^+_3\bigr)=-\partial_q\mathsf J_2(q,x)$. \end{enumerate} \end{lemma} \begin{proof} (i) In $\overline u\,\mathcal L_hu$ the diagonal term $(x-b)|u|^2$ is real, and the two hopping terms are $-a(q\pm\frac h2)\overline{u(q)}u(q\pm h)$; since $\Im(\overline{u(q)}u(q-h))=-\Im(\overline{u(q-h)}u(q))$, their imaginary parts are $-\pi^{-1}\mathsf J(q+\frac h2)$ and $+\pi^{-1}\mathsf J(q-\frac h2)$. (ii) Since $\Theta^{[2]}$ and $D$ depend on $h^2$ only, Lemma~\ref{lem:darboux-parity} gives $u(q;-h)=\omega_{ab}^2\overline{u(q;h)}$ (an identity of the explicit finite branch expressions, with $\kappa=q/h$ of either sign; see the proof of Lemma~\ref{lem:cubic-residual-transport}), and $a$ is even in $h$; substituting gives $\mathsf J(s;-h)=\pi a\,\Im\bigl(u(s+\frac h2;h)\overline{u(s-\frac h2;h)}\bigr)=\mathsf J(s;h)$. By Lemma~\ref{lem:darboux-parity}, up to $O(\kappa^{-R})$, \begin{equation} u=\omega_{ab}\pi^{-1/2}D^{-1/4}\mathfrak A_{ab}(\Theta^{[2]}_q,\kappa)\exp\bigl\{i\bigl(\kappa\Theta^{[2]}_q+\mathfrak P_{ab}(\Theta^{[2]}_q,\kappa)\bigr)\bigr\}, \end{equation} where $\kappa\Theta^{[2]}_q=S/h+h\Gamma_2$, $\mathfrak P_{ab}=h\,\mathfrak c_{ab}(\Theta_q)/q+O(h^3)$ and $\mathfrak A_{ab}=1+h^2\mathfrak A_2(\Theta_q)/q^2+O(h^4)$. The phase increment is therefore $p(s)+h^2(p_{ss}/24+\widetilde\Gamma_s)+O(h^4)$ with $\widetilde\Gamma=\Gamma_2+\mathfrak c_{ab}(\Theta_q)/q$, and $\widetilde\Gamma_s=\Sigma_2$ by \eqref{eq:class-second-data}; the amplitude product is $\pi^{-1}D^{-1/2}[1+h^2(2\mathfrak A_2(\Theta_s)/s^2-(\log D)_{ss}/16)]+O(h^4)$; $a(s;h^2)=\frac12e_\pi(s^2)\bigl(1+h^2(c_{ab}/s^2+\eta_\pi(s^2))\bigr)+O(h^4)$ with $c_{ab}=\frac18-\frac{a^2+b^2}2$; and $D^{1/2}=e_\pi\sin p$. Multiplying out, \begin{equation} 2\mathsf J_2=\frac{c_{ab}+2\mathfrak A_2(\Theta_s)}{s^2}+\cot p\,\Bigl(\Sigma_2+\frac{p_{ss}}{24}\Bigr)-\frac{(\log D)_{ss}}{16}+\eta_\pi(s^2). \end{equation} For the Wigner chain itself ($e_\pi=1$, $v_\pi=0$, so $\eta_\pi=0$) one has $p=\Theta_s=\vartheta$, $D=\sin^2\vartheta$ and $\Sigma_2=-\mathfrak c_{ab}(\vartheta)/s^2$, and $u$ is the $+$ Darboux branch of $\widehat d$, a formal solution of the Wigner recurrence by the two-carrier uniqueness in the proof of Lemma~\ref{lem:darboux-weyl-corner}; by (i) its $\mathsf J$ is independent of $s$, which forces $c_{ab}+2\mathfrak A_2(\vartheta)=\mathfrak c_{ab}(\vartheta)\cot\vartheta$, i.e.\ $\mathfrak A_2=-\frac12\mathfrak c_{ab}'$ (by the identity $c_{ab}-\mathfrak c_{ab}'=\mathfrak c_{ab}\cot$). Substituting this gives \eqref{eq:phase-exactness-J2}. (iii) By Proposition~\ref{prop:general-second-eikonal-pole} every term of \eqref{eq:phase-exactness-J2} is even meromorphic in $s$ with at most a double pole, and the double poles $\mathfrak c_{ab}(\theta)\cot\theta/s^2$ and $\cot\theta\cdot(-\mathfrak c_{ab}(\theta)/s^2)$ cancel. Expanding with \eqref{eq:class-small-q}, $\eta_\pi(0)=\lambda_1$ and $\cot p\to x/\sqrt{1-x^2}$ gives \eqref{eq:phase-exactness-J20}: the new terms are $\lambda_1x\cdot x/(2(1-x^2))$ from the shift $\beta_2\to\beta_2+\lambda_1x$ and $\lambda_1/2$ from $\eta_\pi$, which add to $\lambda_1/(2(1-x^2))$. As for $\rho_2(0,x)$, the higher jets of $e_\pi,v_\pi$ enter only at order $s^2$. (iv) At fixed $q>0$, $\mathcal L_hu=e^{i\Theta_q/z}[h^3q^{-3}C^+_3+O(h^4)]$ (with $z:=h/q=\kappa^{-1}$) and $u=e^{i\Theta_q/z}[B_0+O(h)]$, so the left side of (i) is $h^3\pi q^{-3}\Im(\overline{B_0}C^+_3)+O(h^4)$, while by (ii) the right side is $-h^3\partial_q\mathsf J_2+O(h^5)$. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:general-phase-exactness}] By Lemma~\ref{lem:cubic-residual-transport} (\SuppSec{sec:supp-phase-exactness}) the order-$h^3$ fixed-ratio residual of $\widehat U^{[2],+}$ is $e^{i\Theta_q/z}h^3q^{-3}C^+_3$ with $C^+_3/B_0\in i\mathbb R$, which a real factor $1+h^2\alpha$ cancels exactly when $\partial_q\alpha=\pi q^{-3}\Im(\overline{B_0}C^+_3)$; by Lemma~\ref{lem:pullback-flux}(iv) this is $-\partial_q\mathsf J_2$. The initial value $\alpha(0,x)=c(x)$ matches the fixed-rank Taylor coefficient, an order-four corner bound controls the corrected residual from fixed rank to fixed ratio, and the engine summation with split rank $h^{-1/7}$ gives the $O(h^3)$ bound. The details are in \SuppSec{sec:supp-phase-exactness}. \end{proof} \subsection{Bulk zeros and rank-direction turning points} \label{sec:general-zeros-turns} Bulk zero quantization needs no further hypothesis beyond Theorem~\ref{thm:general-phase-exactness}; the Airy connection adds hypothesis (R) and a lattice quantization of the turning action. \begin{theorem}[Bulk zero quantization] \label{thm:general-bulk-zeros} Let an even centered Wigner--Jacobi chain be given; in the exceptional channel $a=b=0$ assume also the $k=0$ row of Theorem~\ref{thm:general-phase-exactness}. Fix a compact turning-free set of $(q,x,\pi)$ with $q$ bounded away from zero and swept paths in a turning-free corridor $\mathcal C$, and put $\widetilde\Gamma(q,x):=\Gamma_2(q,x)+\mathfrak c_{ab}(\Theta_q(x))/q$. After shrinking the complex neighbourhood $U$ of (E2) if necessary, for all sufficiently large admissible $N$ the lowest-rank seed $G_N(x):=Y_{k_{\min}}(x;h)$ is analytic and zero-free on $U$ (Remark~\ref{rem:E2-x}) and $Y_L(x;h)=G_N(x)P_{L,N}(x)$ with $P_{L,N}$ a polynomial in $x$. Moreover, each zero $\widetilde x_j$ of $x\mapsto d^L_{ab}(\Theta^{[2]}_q(x))$, $q=(L+\tfrac12)/N$, in that set has exactly one zero $x_j$ of $P_{L,N}$ within $O(N^{-4})$ (all zeros of $P_{L,N}$ are real and simple); every zero of $P_{L,N}$ in the set lies within $O(N^{-4})$ of exactly one zero $\widetilde x_j$ (possibly in an $O(N^{-4})$ collar outside the set); and, with $p_{ab}=|a-b|$ and a consistent integer index $\nu_j$, \begin{equation} \boxed{ x_j=\widetilde x_j+O(N^{-4}),\qquad NS(q,x_j)+h\widetilde\Gamma(q,x_j)+\frac{p_{ab}\pi}{2}-\frac\pi4=\left(\nu_j+\frac12\right)\pi+O(N^{-3}).} \label{eq:general-bulk-zeros} \end{equation} Consequently $x_j$ lies within $O(N^{-2})$ of the zero of $d^L_{ab}(\Theta_q(x))$ with the same index, and $NS(q,x_j)+p_{ab}\pi/2-\pi/4=(\nu_j+\frac12)\pi+O(N^{-1})$. Both parts of $h\widetilde\Gamma$ are needed at order $N^{-1}$: the Bohr--Sommerfeld condition with $h\Gamma_2$ but without the Darboux term $h\,\mathfrak c_{ab}(\Theta_q)/q=\mathfrak c_{ab}(\Theta_q)/\kappa$ is in error at order $N^{-1}$ wherever $\mathfrak c_{ab}(\Theta_q)\ne0$. Theorem~\ref{thm:general-second-order} alone, without Theorem~\ref{thm:general-phase-exactness}, gives the same statements with $O(N^{-3})$ and $O(N^{-2})$ (assuming the $k=0$ row when $a=b=0$ for the polynomial factorization). \end{theorem} \begin{proof} On $\mathcal C$, $e_\pi(k^2h^2,h^2)>0$ for $N$ large, so $A_k(h)>0$ for $k_{\min}0\ (0\le q0$, uniform on the family, with $|\widehat Y_k|\le CN^m$ whenever $q_*\le q_k\le q_*+\delta$. \end{enumerate} Put $\widetilde\Gamma=\Gamma_2+\mathfrak c_{ab}(\Theta_q)/q$, let $I=[q_*-\delta_0,q_*+\delta_0]$ be the fixed turning neighbourhood of the canonical Airy basis (\SuppSec{sec:supp-rank-airy}), shrunk if necessary so that $\delta_0\le\delta$, and $I'=\{q\in I:\zeta_\tau(q)\le4^{-2/3}\zeta_\tau(q_*+\delta_0)\}$. There is a real-analytic function $B_\tau(q)=B_\tau(q,x)$ on $I$ with \begin{equation} B_\tau(q)=\frac{\tau\,\widetilde\Gamma(q,x)}{\sqrt{-\zeta_\tau(q)}}-\frac{5}{48\,\zeta_\tau(q)^2}\qquad(q0$), $e_\pi((q_*+\delta)^2)=e_0+e_1\delta+e_2\delta^2+O(\delta^3)$ for the principal coefficient, $v_1=\partial_tv_\pi(q_*^2,0)$ and $\eta_\pi$ from \eqref{eq:class-first-coefficients} ($\eta_\pi\equiv0$ for Hahn and Racah). Uniformly at admissible ranks with $q=\kappa/N\in I'$, on both sides of the turn, \begin{equation} \boxed{\begin{aligned} Y_L={}&\epsilon_{ab}(-1)^{K_\tau}\sqrt{\frac2q}\,N^{-1/3} \left|\frac{\zeta_\tau(q)}{D(q,x)}\right|^{1/4} \\&\times\Bigl[\bigl(1+\mathfrak a_N(q)\bigr)\operatorname{Ai}(t)+N^{-4/3}B_\tau(q)\operatorname{Ai}'(t)+\mathfrak r_N(q)\Bigr],\qquad t=N^{2/3}\zeta_\tau(q), \end{aligned}} \label{eq:general-rank-airy-uniform} \end{equation} where $\mathfrak a_N=N^{-2}\mathfrak a(q)+c_N$ with $\mathfrak a$ real-analytic on $I$ and constants $|c_N|\le CN^{-2}$, and $|\mathfrak r_N(q)|\le CN^{-3}\mathsf M(t)/\mathsf E(t)$, with Olver's modulus and weight functions $\mathsf M,\mathsf E$ of the Airy functions~\cite[Chapter~11]{OlverASF}. Equivalently the bracket is $(1+O(N^{-2}))\operatorname{Ai}(N^{2/3}\widehat\zeta_\tau)+O(N^{-3})\mathsf M/\mathsf E$ with the second-order Langer variable $\widehat\zeta_\tau=\zeta_\tau+N^{-2}B_\tau$. In particular, uniformly for $|N^{2/3}\zeta_\tau(q)|\le A_0$, with the prefactor of \eqref{eq:general-rank-airy}, \begin{equation} Y_L=\epsilon_{ab}(-1)^{K_\tau}\sqrt{\frac2q}\,N^{-1/3} \left|\frac{\zeta_\tau(q)}{D(q,x)}\right|^{1/4} \left[\operatorname{Ai}(t)+N^{-4/3}\varpi_\tau(x)\operatorname{Ai}'(t)+O_{A_0}(N^{-2})\right]. \label{eq:general-rank-airy-second} \end{equation} \end{enumerate} The error terms in \eqref{eq:general-rank-airy} and \eqref{eq:general-rank-airy-second} are additive inside the brackets, and $\mathfrak r_N$ is measured against Olver's envelope; no relative-error assertion is made at zeros of $\operatorname{Ai}$. \end{theorem} \begin{proof} (i) The exact recurrence satisfies the uniform simple-turning hypotheses of Geronimo--Bruno--Van Assche~\cite{GeronimoBrunoVanAssche}, which yield an exact canonical $\operatorname{Ai}/\operatorname{Bi}$ basis in the coordinate $\zeta_\tau$ with the transport $|\zeta_\tau/D|^{1/4}$ and $O(N^{-1})$ envelope error; we use only the non-recessive envelope $(\operatorname{Ai}^2+\operatorname{Bi}^2)^{1/2}$ of~\cite[Theorem~2.6]{Geronimo2009}. The two connection coefficients are fixed by matching a two-component Jacobi vector at a fixed oscillatory rank ratio $q_0q_*$ and $\Upsilon(q):=0$ for $0\le q\le q_*$, $\xi(q)=\frac23N\zeta_\tau(q)^{3/2}$ ($q\ge q_*$), and $I=[q_*-\delta_0,q_*+\delta_0]$ with $q_*+\delta_0$ at a fixed distance from the hard edge. \begin{enumerate}[label=(\alph*),leftmargin=2em,itemsep=0pt] \item $\widetilde Y_k$ is strictly nonzero of one sign for $q_k\ge q_*+C_0N^{-2}$ (up to the hard edge $k_{\max}$, where $\widetilde Y_{k_{\max}+1}:=0$ and so $\rho_{k_{\max}}=0$), and $0<\rho_k\le1$ there for $k0$ with $\rho_{j-1}\le\exp\bigl(-\Upsilon(\sqrt{(q_j^2-D_0/N)_+})\bigr)$ for $q_j\in I$ with $q_j\ge q_*+N^{-2/3}$ and all sufficiently large $N$. Hence $|\widehat Y_k|\le CN^{1/6}e^{-\xi(q_k)}$ there (at least the recessive Airy decay). \item \eqref{eq:general-rank-airy-uniform} holds, with $\mathfrak a_N$ as in Proposition~\ref{prop:airy2-amplitude} and $|\mathfrak r_N|\le CN^{-3}\mathsf M/\mathsf E$, at all lattice points $q\in I$ with $q\le q_*+\delta_0-C_1N^{-1}\log N$; this replaces $I'$ of Theorem~\ref{thm:general-rank-airy}(ii) for these chains (for the $q$-Hahn limit of Remark~\ref{rem:qhahn-limit} only (a) and the ratio bound in (b) are claimed). \end{enumerate} \end{proposition} \begin{proof} See \SuppSec{sec:supp-forbidden-side} (comparison with constant-ratio supersolutions; beyond $I$ only (a)--(b) is available, and the logarithmic collar is where the $\operatorname{Bi}$ component stops being below $N^{-3}\mathsf M/\mathsf E$). \end{proof} \begin{remark}[Flux defect and the Gauss weight; open step] \label{rem:flux-gauss} Since $\alpha=\delta_{\rm fl}-\mathsf J_2$, the identity $\alpha=-\mathsf J_2$ for Hahn and Racah is $\delta_{\rm fl}=0$, proved from the exact fixed-rank $N^{-2}$ coefficients of Ref.~\cite{LewisFixedRank}. For Racah it also follows without that input: the quantum $6j$ value of Corollary~\ref{cor:q6j-consequences}(iii) uses only the closed seed, the flat limit (vi) rescales it to $\delta_{\rm fl}=-\varsigma^2/12$, and as $\varsigma\to0$ the seed and the (E1) data tend to those of the centered Racah chain. Since $\delta_{\rm fl}$ is determined at fixed rank, by the $t$-coefficients of the seed and of the (E1) data (through $c(x)$), and these are analytic in $\varsigma$ at $\varsigma=0$ ($q$-numbers $[n]$ are even analytic in $\varsigma$), the limit may be taken in $\delta_{\rm fl}$, so the Racah value is $\lim_{\varsigma\to0}(-\varsigma^2/12)=0$. An intrinsic route is the Gauss-weight identity $\lambda_j^2w_j=2\,(1+O(h^4))$, where $\lambda_j^2=\sum_k\widehat Y_k(x_j)^2$ is the natural-gauge column norm and $w_j$ is the node gap, equivalent to $\delta_{\rm fl}=0$ once the constant physical-branch flux $\mathsf J_{\rm phys}(x_j)=\frac12+h^2\delta_{\rm fl}+O(h^4)$ satisfies $\mathsf J_{\rm phys}(x_j)=\frac14\lambda_j^2w_j(1+O(h^3))$. The weight identity holds exactly for $3j$ ($\lambda_j^2=N$, $w_j=2/N$) and $6j$ ($\lambda_j^2=4AB/(2C)$, $w_j=C/(AB)$); proving $\mathsf J_{\rm phys}(x_j)=\frac14\lambda_j^2w_j(1+O(h^3))$ across the forbidden zone for a general chain in the class is \emph{open}. For the one-small quantum $6j$ chain the deformed identity $\lambda_j^2\bar w_j=2h/\sinh h$ holds exactly, with $\lambda_j^2=\sum_k\widehat\Psi_k(x_j)^2=[2A][2B]/[2C]$ by $q$-orthogonality and $\bar w_j=|\partial_Cx|=2h\sinh\mathsf Z/(\sinh\mathsf X\sinh\mathsf Y)$ the node density in $C$ (Section~\ref{sec:q6j}); the centred node gap $2\sinh h\,\sinh\mathsf Z/(\sinh\mathsf X\sinh\mathsf Y)$, the analogue of $w_j=C/(AB)$ above, equals $(\sinh h/h)\,\bar w_j$, and with it $\lambda_j^2w_j=2$ exactly. So the weight comparison does not single out a value for the $q$-chain: matching $\mathsf J_{\rm phys}(x_j)$ to $\frac14\lambda_j^2\bar w_j=\frac12\frac{h}{\sinh h}=\frac12-\frac{h^2}{12}+O(h^4)$ reproduces the value $\delta_{\rm fl}=-\frac1{12}$ proved by exact computer algebra from the closed seed, whereas the centred gap would give $0$. The value $-\frac1{12}$ is relative to the normalization \eqref{eq:q6j-Psi}, which already contains the factors $\sqrt{[2k+1]}$ and $\sqrt{[2A][2B]}$: by the rule above, multiplying the physical solution by the constant $(\sinh h/h)^{1/2}=1+\frac{h^2}{12}+O(h^4)$ keeps (E1)--(E2) and replaces $\delta_{\rm fl}$ by $0$. It is thus a property of the normalization \eqref{eq:q6j-Psi} of the $q$-$6j$ symbol, not an obstruction to the class. \end{remark} \begin{remark}[Olver's form] \label{rem:airy2-olver} For $w''=(u^2f+g)w$ with a simple turning point, Olver's uniform expansion is $\operatorname{Ai}(u^{2/3}\zeta)\sum_sA_su^{-2s}+u^{-4/3}\operatorname{Ai}'(u^{2/3}\zeta)\sum_sB_su^{-2s}$, $A_0=1$, where $B_0$ is the solution regular at $\zeta=0$ of $2\zeta B_0'+B_0=\psi$~\cite[Chapter~11]{OlverASF}. The bracket of \eqref{eq:general-rank-airy-uniform} is the discrete analogue through $A_1$ and $B_0$, with $u=N$: $B_\tau$ solves the same transport equation with an explicit lattice function $\psi_\tau$ in place of $\psi$ (Lemma~\ref{lem:airy2-olver}), so $\varpi_\tau=B_\tau(q_*)=\psi_\tau(0)$. By \eqref{eq:general-airy-varpi-olver}, $\psi_\tau(0)$ is Olver's $\psi(0)$ for $f=F$ and $g(q_*)=F_1^2/240+G_\tau$, where $G_\tau$ is the bracketed sum after $F_1^2/240$. The term $F_1^2/240$ is the $O(h^2)$ discreteness remnant of the lattice second difference at the turn, absent for a second-order ODE: for the model chain (constant $a$, linear $F$) with a symbol $T(p)=1+p^2/2+t_2p^4+t_3p^6+\cdots$ in place of $\cosh p$ it equals $F_1^2(4t_2^2-2t_3)$, which is $F_1^2/240$ for $\cosh$ ($t_2=\frac1{24}$, $t_3=\frac1{720}$) and $0$ for the ODE, and for the class the value $\frac1{240}$ is what the exact identity \eqref{eq:general-airy-varpi-olver}${}={}$\eqref{eq:general-airy-varpi-local} gives. It arises because $F=\operatorname{arcosh}^2(\tau u)$ already carries the classical effect of the higher difference terms but not their action on the Airy parametrix at $\zeta=0$, where $\operatorname{Ai}^{(6)}(0)=4\operatorname{Ai}(0)$. It is a potential term, not a phase change: $\varpi_\tau$ is Olver's $\psi(0)$ for $w''=(u^2F+g_\tau)w$ with $g_\tau(q_*)=F_1^2/240+G_\tau$, equivalently for the Langer variable $F+h^2g_\tau$; omitting it would change $\varpi_\tau$ by exactly $-F_1^{4/3}/240$ (for the Hahn chain with $(M,\sigma)=(2,1)$ and $x=\frac35$ this is $-0.030$) and leave the bracket error of order $N^{-4/3}$. The coefficient blows up where the turn meets the zero of $e_\pi$ (Hahn: $x\to0$; Racah: triangles right-angled opposite $\alpha$ or $\beta$, \eqref{eq:racah-varpi}); there a different comparison function is needed (Remark~\ref{rem:racah-collision-excluded}), as for a turning point coalescing with a singularity in Olver's theory. \end{remark} \section{Hahn/\texorpdfstring{$3j$}{3j}: verification and explicit data} \label{sec:growing3j} This section verifies that the fixed-offset $3j$ rank recurrence is an even centered Wigner--Jacobi chain and records its explicit data and the resulting statements for $3j$ symbols. Put $\kappa=L+\frac12$ and $q=\kappa/N$. \begin{proposition}[The Hahn chain is an even centered chain] \label{prop:hahn-class} With $q_k=(k+\tfrac12)h$ and the natural gauge $\widehat\Phi_k=\sqrt{2k+1}\,\Phi_k$, the exact rank recurrence \eqref{eq:3jrank} is \begin{equation} x\widehat\Phi_k=A_{k+1}^{(N)}\widehat\Phi_{k+1}+A_k^{(N)}\widehat\Phi_{k-1}+B_k\widehat\Phi_k, \qquad A_k^{(N)}=\frac{R_k}{k\sqrt{4k^2-1}}\sqrt{1-k^2h^2},\quad B_k=\frac{M\sigma}{k(k+1)}. \label{eq:adiabaticJacobi}\eqalias{eq:AkNexact} \end{equation} It is an even centered Wigner--Jacobi chain with offsets $(a,b)=(M,\sigma)$, on every compact $X\Subset(-1,1)$ and every $s_1<1$, with \begin{equation} e(s)=\sqrt{1-s},\qquad v\equiv0,\qquad \varepsilon_1=\tfrac12,\quad\beta_1=\beta_2=0,\quad\chi(x)=\frac x2 . \label{eq:hahn-class-data} \end{equation} Its principal data are $\cos p=x/\sqrt{1-q^2}$ and $D(q,x)=1-x^2-q^2$. \end{proposition} \begin{proof} Dividing \eqref{eq:3jrank} by $k(k+1)(2k+1)$ and applying the square-root gauge gives the displayed coefficients, with $B_k=B_k^{(0)}$. For (E2), the lowest-rank value is, by \cite[Eq.~(27)]{LewisFixedRank}, a fixed constant times the square root of $\mathfrak C_{L+\sigma}(u_+,h)\,\mathfrak C_{L-\sigma}(u_-,h)/\Pi_L(h^2)$ with $u_\pm=\frac{1\pm x}2$ (for general offsets after the symmetry reduction of Ref.~\cite{LewisFixedRank}, which changes it by a sign independent of $N$). Each $\mathfrak C_r(u,h)$ is a product of the factors $u+qh$ over a symmetric set of $q$, hence a polynomial in $(u,t)$, $t=h^2$, and $\Pi_L$ is a polynomial in $t$ with $\Pi_L(0)=1$. At $t=0$ the radicand is $u_+^{L+\sigma}u_-^{L-\sigma}>0$ for $x\in(-1,1)$ (in the chamber $M\ge|\sigma|$, $L_0=M$, to which the symmetry reduction brings every offset pair; there it is $\widehat d_{L_0}^{\,2}$ up to a constant), so on a complex neighbourhood $U$ of $X$ and a disk $|t|\le t_0$, uniform for $X\Subset(-1,1)$, the square root is analytic and bounded jointly in $(x,t)$, real for real arguments, with value $\widehat d_{L_0}(\arccos x)$ at $t=0$; when $L_0>0$ this also gives $L_0+1$ by (E2). In the exceptional sector $M=\sigma=0$, where the $k=0$ rank equation degenerates to $0=0$, one may start from \begin{equation} \Phi_0=1,\qquad \Phi_1=\frac{x}{\sqrt{1-t}}. \label{eq:3jexceptionalstart} \end{equation} \end{proof} \subsection{Closed Hahn action} \label{sec:wkb3j} \label{sec:critical} Inserting $\Phi=\mathcal A(x)e^{\ii S(x)/h}$ into the magnetic equation \eqref{eq:3jmag} gives at principal order $(1-x^2)\cos(2S_x)-(1-x^2)+2q^2=0$. On the branch reducing to the continuum Wigner phase, \begin{equation} S_x=-p_q(x),\qquad p_q(x)=\arcsin\!\left(\frac{q}{\sqrt{1-x^2}}\right), \label{eq:pq} \end{equation} with oscillatory band $1-x^2-q^2>0$ and turning points $x_\pm=\pm\sqrt{1-q^2}$. Direct integration of $F(q,x)=\int_0^x\arcsin(q/\sqrt{1-u^2})\,\dd u$ gives \begin{equation} F(q,x)=q\arcsin\!\left(\frac{x}{\sqrt{1-q^2}}\right) +x\arcsin\!\left(\frac{q}{\sqrt{1-x^2}}\right) -\arctan\!\left(\frac{xq}{\sqrt{1-x^2-q^2}}\right), \label{eq:Fclosed} \end{equation} and we put \begin{equation} S(q,x)=\frac{\pi q}{2}-F(q,x),\qquad \Theta_q(x)=\frac{S(q,x)}q,\qquad x_q:=\cos\Theta_q,\qquad W_q(x)=\left[\frac{1-x_q^2}{1-x^2-q^2}\right]^{1/4}. \label{eq:Theta} \end{equation} Then $\Theta_q\to\arccos x$ as $q\to0$, $\partial_qS=\arccos(x/\sqrt{1-q^2})=p$ and $D=1-x^2-q^2$, so $S$, $\Theta_q$ and $W_q$ are the generic action data \eqref{eq:class-principal-data} of the Hahn chain, and the transport equation of \eqref{eq:3jmag} gives the same amplitude $(1-x^2-q^2)^{-1/4}$. For small $q$, \begin{equation}\begin{gathered} \Theta_q(x)=\arccos x -\frac{q^2x}{6\sqrt{1-x^2}} -\frac{q^4x(3-2x^2)}{40(1-x^2)^{3/2}} +O_K(q^6),\\ W_q(x)=1+\frac{q^2(3-x^2)}{12(1-x^2)} +\frac{q^4(41x^4-94x^2+225)}{1440(1-x^2)^2}+O_K(q^6).\end{gathered} \label{eq:Thetaexp} \end{equation} \label{sec:secular}% We now relate this to the secular growth of the fixed-rank coefficients. At order $N^{-2r}$ the maximal $L^{3r}$ growth of the fixed-rank coefficients comes from $r$ separate insertions of the first $k^2/N^2$ perturbation, so the leading terms are iterates of a phase displacement. For Hahn the exact fixed-rank coefficients of Ref.~\cite{LewisFixedRank} satisfy $f_2=\kappa^2\mathcal P_2f+O_K(L^{1/2})$ and $f_4=\kappa^4\mathcal P_4f+O_K(L^{7/2})$, $f=d^L_{M\sigma}(\arccos x)$ (Proposition~\ref{prop:D2D4match}), where $\mathcal P_2=\tfrac x6\partial_x+\tfrac{3-x^2}{12(1-x^2)}$ is the operator \eqref{eq:general-P2} with \eqref{eq:hahn-class-data} and \begin{equation} \mathcal P_4=\frac{x^2}{72}\partial_x^2 +\frac{7x(3-2x^2)}{180(1-x^2)}\partial_x +\frac{41x^4-94x^2+225}{1440(1-x^2)^2}; \label{eq:P4} \end{equation} these are exactly the $q^2$ and $q^4$ terms of the pullback $W_qd^L_{M\sigma}(\Theta_q)$ by \eqref{eq:Thetaexp}. The closed action resums the whole hierarchy \eqref{eq:even-phase-horizon-hierarchy} (Remark~\ref{rem:hahn-higher-horizons}). \begin{proposition}[Hahn data] \label{prop:hahn-data} \label{thm:growing3j} \label{thm:3j-convergent-mesoscopic} \label{cor:subcritical3j} \label{thm:critical} \label{cor:critical-limit} \label{thm:effective-all-rank} \label{cor:effective-fixed-ratio} \label{thm:effective-fixed-ratio} \label{prop:3j-second-weyl-eikonal} Fix $M,\sigma$ and a compact $K\Subset(-1,1)$, and let $\Phi_L=\Phi_L^{M,\sigma}$. \begin{enumerate}[label=(\alph*),leftmargin=2em] \item \emph{Mesoscopic and critical regimes.} Write $\Phi_L(x;t^{1/2})=\sum_{r\ge0}t^rf_{2r}^{L}(x)$ with $t=N^{-2}$ and $f_0^L=d^L_{M\sigma}(\arccos x)$. The series converges and obeys the bounds \eqref{eq:general-mesoscopic-coeff}--\eqref{eq:general-mesoscopic-tail} with $Y_L=\Phi_L$; in particular $\sqrt{L+1}\sup_K|\Phi_L-d^L_{M\sigma}(\arccos x)|=O(\kappa^3/N^2)$ for $L=o(N^{2/3})$. Uniformly for physical nodes $x=\cos\theta\in K$ and $\kappa^3/N^2\le\Lambda$, \begin{equation} \sqrt{\frac{\pi\kappa\sin\theta}{2}}\,\Phi_L(x;N^{-1}) =\cos\!\left(\varphi_{L,M,\sigma}(\theta)-\frac{\kappa^3}{6N^2}\cot\theta\right) +O_{K,\Lambda}\!\left(\frac1{\kappa}+\frac{\kappa^2}{N^2}+\frac{\kappa^5}{N^4}\right), \label{eq:criticalphase-quant} \end{equation} with $\varphi_{L,M,\sigma}$ the large-$L$ Wigner-$d$ phase. If $\kappa^3/N^2\to\lambda$ and $x_N\to\cos\theta_0$ with $|x_N-\cos\theta_0|=O(N^{-1})$, the added phase tends to $-\lambda\cot\theta_0/6$. \item \emph{First-order pullback.} For every compact corridor $\mathcal C\Subset\{(q,x):0\le q<1,\ D(q,x)>0\}$ and every admissible $L$ whose swept path lies in $\mathcal C$, the bound \eqref{eq:general-first-order} holds for $\Phi_L-W_q\,d^L_{M\sigma}(\arccos x_q)$ with the closed data \eqref{eq:Fclosed}--\eqref{eq:Theta}. On compact fixed-ratio sets this is $O(N^{-1})$ in natural gauge and $O(N^{-3/2})$ absolutely; the remainder at this order is the omitted phase $h\Gamma_2(q,x)$ of Remark~\ref{rem:fixed-ratio-odd-solution}, and the pullback carries no further $O(1)$ $q$-dependent phase. \item \emph{Second-order data.} The coefficients \eqref{eq:class-a2b2} are \begin{equation} a_0=\tfrac12\sqrt{1-q^2},\quad a_2=\frac{a_0}{q^2}\Bigl(\frac18-\frac{M^2+\sigma^2}{2}\Bigr),\quad b_2=\frac{M\sigma}{q^2}, \label{eq:3j-a0a2b2} \end{equation} and for them the second-order data \eqref{eq:class-second-data} define $\Sigma_2^{(3)}$, $\rho_2^{(3)}$, $\Gamma_2^{(3)}$, $\Theta^{[2]}_q$ and $W^{[2]}_q$, with $W^{[2]}_q=[\sin^2\Theta^{[2]}_q/(1-x^2-q^2)]^{1/4}$; $\rho^{(3)}_2$ is analytic in $q^2$ at $q=0$. \item \emph{Amplitude and Airy data.} \begin{equation} \mathsf J_2(0,x)=-\frac{(8(M^2+\sigma^2)-5)x^2+4M\sigma x(x^2-5)-3}{48(1-x^2)^2}, \label{eq:hahn-J20} \end{equation} and at the rank turn $q_*=\sqrt{1-x^2}$ of a fixed node $x\ne0$, of type $\tau=\operatorname{sgn}x$, \begin{equation} \varpi^{\mathrm H}(x):=\varpi_{\operatorname{sgn}x}(x)=\frac{2^{1/3}\bigl(140(M^2+\sigma^2)x^4-280M\sigma x^3-50x^4+15x^2-9\bigr)}{280\,|x|^{8/3}(1-x^2)^{4/3}}. \label{eq:hahn-varpi} \end{equation} \end{enumerate} \end{proposition} \begin{proof} (a) is Theorems~\ref{thm:general-mesoscopic} and~\ref{thm:general-critical} with Proposition~\ref{prop:hahn-class}: the added phase is $-\chi\kappa^3/(3N^2\sin\theta)=-\kappa^3\cot\theta/(6N^2)$. For the limit, $\theta_N-\theta_0=O(N^{-1})$, so the added phase at $\theta_N$ differs from its value at $\theta_0$ by $O(\kappa^3/N^3)=O(N^{-1})$ in the critical window $\kappa\asymp N^{2/3}$; no off-lattice continuation is needed. (b) is Corollary~\ref{thm:general-first-order} with the identification of the action data above. (c) is Proposition~\ref{prop:general-second-eikonal-pole} with $(a,b)=(M,\sigma)$. (d) is \eqref{eq:phase-exactness-J20} with \eqref{eq:hahn-class-data}, and \eqref{eq:general-airy-varpi-local} with $\tau u(q)=|x|(1-q^2)^{-1/2}$, $e_0=|x|$, $u_1=q_*/x^2$, $v_1=0$ (exact computer algebra, \SuppSec{app:hahn-airy-dictionary}). \end{proof} \begin{remark}[Sharpness of the $N^{2/3}$ coherence scale] For fixed interior $x=\cos\theta\ne0$ the first pullback phase correction from \eqref{eq:Thetaexp} is $\kappa(\Theta_q-\theta)=-\kappa^3\cot\theta/(6N^2)+O(L^5/N^4)$. Thus $L=o(N^{2/3})$ is precisely the regime in which the uncorrected Wigner-$d$ phase perturbation vanishes uniformly on compact sets avoiding $x=0$, and at $L\asymp N^{2/3}$ it has a finite nonzero limit. For $x\ne0$ the principal part $\kappa^2\mathcal P_2f$ of $f_2$ has exact order $L^{5/2}$ (its leading oscillating part is the phase term displayed above, which does not vanish there), so $\sqrt{L+1}\,N^{-2}f_2$ is of exact order $\kappa^3/N^2$ and the exponent $3$ in (a) cannot be lowered. \end{remark} \begin{corollary}[Phase-exact Hahn pullback] \label{cor:3j-phase-exactness} \label{thm:effective-all-rank-second} \label{cor:effective-all-rank-second} \label{cor:effective-fixed-ratio-second} \label{thm:effective-fixed-ratio-second} Fix $M,\sigma$ and a closed corridor $\mathcal C\Subset\{(q,x):0\le q<1,\ 1-x^2-q^2>0\}$. For every admissible rank $L$ whose swept path stays in $\mathcal C$, \begin{equation} \boxed{ \sqrt{2L+1}\left| \Phi_L^{M,\sigma}(x;N^{-1}) -\Bigl(1-\frac{\mathsf J_2(q,x)}{N^2}\Bigr)W_q^{[2]}(x)\,d^L_{M\sigma}(\Theta_q^{[2]}(x))\right| \le \frac{C_{\mathcal C}}{N^3},} \label{eq:3j-phase-exactness} \end{equation} with $\mathsf J_2$ from \eqref{eq:phase-exactness-J2} and \eqref{eq:hahn-J20}; without the factor $1-N^{-2}\mathsf J_2$ the bound is $C_{\mathcal C}N^{-2}$. At fixed positive ratio the absolute errors are $O(N^{-7/2})$ and $O(N^{-5/2})$. \end{corollary} \begin{proof} Theorems~\ref{thm:general-second-order} and~\ref{thm:general-phase-exactness} with Proposition~\ref{prop:hahn-class}; the $k=0$ row of the exceptional channel is \eqref{eq:3jexceptionalstart}. The fixed-rank datum is $c(x)=-\mathsf J_2(0,x)$ by the exact $N^{-2}$ coefficient~\cite[Theorem~6.1]{LewisFixedRank} (\SuppSec{sec:supp-phase-exactness}). \end{proof} \begin{corollary}[Bulk Hahn zero quantization] \label{cor:hahn-bulk-zero-quantization} Fix $M,\sigma$, $00$ there), and it equals $\widehat d_m^{\,2}>0$ at $t=0$ because the limiting triangles are strict, so $\alpha\pm\beta\pm\gamma\neq0$. Hence $\widehat\Psi_m$ is analytic and bounded jointly in $(x,t)$ on $U\times\{|t|\le t_0\}$ for a complex neighbourhood $U$ of $X$, uniformly on the compact family, with value $\widehat d_m$ at $t=0$; the second rank follows from the first by the recurrence at $f=f_{\min}$, where the backward coefficient vanishes and the forward one is analytic and nonzero. In the exceptional channel $m=n=0$ one may take \begin{equation} \Psi_0=1,\qquad \Psi_1=\frac{x-t/(8\alpha\beta)}{\sqrt{(1-t/(4\alpha^2))(1-t/(4\beta^2))}}. \label{eq:6jexceptionalstart} \end{equation} Finally $\eta^2-(x-b)^2=1-x^2-\rho^2(\alpha^2+\beta^2-2\alpha\beta x)/(4\alpha^2\beta^2)$. \end{proof} \subsection{Closed Racah action} \label{sec:6jWKB} \label{sec:6jcritical} The rank action and the effective angle and transport \begin{equation} S_6(\rho)=\int_0^\rho\omega(s)\,\dd s,\qquad \Theta_6(\rho)=\frac{S_6(\rho)}{\rho}, \qquad W_6(\rho)=\left[\frac{\sin^2\Theta_6(\rho)}{D_6(\rho)}\right]^{1/4} \label{eq:S6} \end{equation} are the generic action data \eqref{eq:class-principal-data} of the centered Racah chain. The action is elementary. Put $s_\theta:=\sin\theta$ and \begin{equation} \rho_*:=\frac{2\alpha\beta s_\theta}{\gamma},\qquad \phi:=\arcsin\frac{\rho}{\rho_*},\qquad c_\beta:=\frac{\alpha-\beta x}{\gamma},\qquad c_\alpha:=\frac{\beta-\alpha x}{\gamma}, \end{equation} so that $c_\beta$ and $c_\alpha$ are the cosines of the angles of the triangle $(\alpha,\beta,\gamma)$ opposite $\beta$ and $\alpha$. Integration by parts, followed by $\rho^2=\rho_*^2\sin^2\phi$, gives for $0\le\rho<\rho_*$ the exact closed form \begin{equation} \boxed{ S_6(\rho)=\rho\,\omega(\rho)+2\gamma\phi -2\alpha\operatorname{atan2}(c_\beta\sin\phi,\cos\phi) -2\beta\operatorname{atan2}(c_\alpha\sin\phi,\cos\phi),} \label{eq:S6closed} \end{equation} since the remaining integrand reduces to $-2\gamma+2\alpha c_\beta(1-(1-c_\beta^2)\sin^2\phi)^{-1}+2\beta c_\alpha(1-(1-c_\alpha^2)\sin^2\phi)^{-1}$. The formula holds on every strict triangle and extends continuously to $\rho=\rho_*$ unless the triangle is right-angled at the vertex opposite $\alpha$ or $\beta$, where the turn meets an off-diagonal boundary. \begin{proposition}[Racah data] \label{prop:racah-data} \label{thm:growing6j} \label{cor:6j-convergent-mesoscopic} \label{cor:6jsubcritical} \label{thm:6jcritical} \label{prop:6jaction} \label{thm:6j-effective-all-rank} \label{cor:6j-fixed-ratio} \label{thm:6j-fixed-ratio} \label{prop:6j-second-weyl-eikonal} Fix admissible $m,n$ and a compact family of strict limiting triangles. \begin{enumerate}[label=(\alph*),leftmargin=2em] \item \emph{Mesoscopic and critical regimes.} The Taylor series $\Psi_f(t)=\sum_rt^rg^{(f)}_{2r}(x)$, $g_0^{(f)}=d^f_{mn}(\arccos x)$, converges and obeys \eqref{eq:general-mesoscopic-coeff}--\eqref{eq:general-mesoscopic-tail}; in particular $\sqrt{f+1}\,|\Psi_f(N)-d^f_{mn}(\theta)|=O(\kappa_f^3/N^2)$ for $f=o(N^{2/3})$. As $\rho\to0$, \begin{gather} \omega(\rho)=\theta-\frac{\chi_6(x)}{\sin\theta}\rho^2+O(\rho^4), \label{eq:6jomegaexp}\\ NS_6(\rho)=\kappa_f\theta+\frac{\kappa_f^3}{N^2}C_6(\alpha,\beta,x)+O\Bigl(\frac{\kappa_f^5}{N^4}\Bigr),\qquad C_6:=\frac{1/(12\alpha\beta)-x(\alpha^{-2}+\beta^{-2})/24}{\sin\theta}, \label{eq:6jcriticalphasepred} \end{gather} and uniformly when $\kappa_f^3/N^2\le\Lambda$, \begin{equation} \sqrt{\frac{\pi\kappa_f\sin\theta}{2}}\,\Psi_f(N) =\cos\!\left(\varphi_{f,mn}(\theta)+\frac{\kappa_f^3}{N^2}C_6(\alpha,\beta,x)\right) +O_{\Lambda}\!\left(\frac1{\kappa_f}+\frac{\kappa_f^2}{N^2}+\frac{\kappa_f^5}{N^4}\right). \label{eq:6jcriticaltheorem-quant} \end{equation} \item \emph{First-order pullback.} For a compact strict-triangle corridor whose swept family stays away from $D_6=0$ (hence from the off-diagonal boundaries $\eta=0$, since $\eta^2\ge D_6$), the bound \eqref{eq:general-first-order} holds for $\Psi_f-W_6(\rho)d^f_{mn}(\Theta_6(\rho))$; at fixed ratio this is $O(N^{-1})$ in natural gauge and $O(N^{-3/2})$ absolutely, with no fixed-offset $O(1)$ phase beyond $NS_6(\rho)$. \item \emph{Second-order data.} At fixed $\rho>0$, \begin{equation} a_{6,0}=\frac{\eta}{2},\quad a_{6,2}=\frac{a_{6,0}}{\rho^2}\left(\frac18-\frac{m^2+n^2}{2}\right),\quad b_{6,0}=\frac{\rho^2}{4\alpha\beta},\quad b_{6,2}=\frac{1/2-m^2-n^2}{8\alpha\beta}+\frac{mn}{\rho^2}, \label{eq:6j-a0a2b0b2} \end{equation} and $\Sigma_2^{(6)}$, $\rho_2^{(6)}$, $\Gamma_2^{(6)}=\int_0^\rho\rho_2^{(6)}$, $\Theta_6^{[2]}=\Theta_6+h^2\Gamma_2^{(6)}/\rho$ and $W_6^{[2]}=[\sin^2\Theta_6^{[2]}/D_6]^{1/4}$ are the data \eqref{eq:class-second-data}; $\rho_2^{(6)}$ is analytic in $\rho^2$ at $\rho=0$. \item \emph{Amplitude and Airy data.} $\mathsf J_2(0,x)$ is \eqref{eq:phase-exactness-J20} with \eqref{eq:racah-class-data} and $\beta_2=(\frac12-m^2-n^2)/(8\alpha\beta)$; the Airy coefficient at the rank turn is \eqref{eq:racah-varpi} of Theorem~\ref{thm:racahAiryConnection}. \end{enumerate} \end{proposition} \begin{proof} (a) is Theorems~\ref{thm:general-mesoscopic} and~\ref{thm:general-critical} with Proposition~\ref{prop:racah-class}; the expansion of $\omega$ is $p=\theta-\chi_6q^2/\sin\theta+O(q^4)$, and $-\chi_6/(3\sin\theta)=C_6$. (b) is Corollary~\ref{thm:general-first-order}; the swept family is turning-free by the monotonicity of $D_6$. (c) is Proposition~\ref{prop:general-second-eikonal-pole} with $(a,b)=(m,n)$, and (d) is immediate. \end{proof} In the symmetric channel $m=n=0$, Chen, Ismail and Simeonov proved the Legendre and phase-coherent cosine forms of the Ponzano--Regge asymptotics of the one-small $6j$ symbol $\{a\,b\,e;\,b\,a\,f\}$, with relative error $O(f^2/N)$ plus $O(f^{-3/2})$, for $f=o(N^{1/2})$ \cite[Theorem~2]{ChenIsmailSimeonov}, and they note that Ponzano and Regge's condition $f=O(N)$ is not sufficient for this. Part~(a) of Proposition~\ref{prop:racah-data} extends the phase-coherent Wigner-$d$ window to $f=o(N^{2/3})$, for fixed nonzero offsets as well, for the absolute error in the natural gauge (a relative bound is not claimed near zeros). The coherence-horizon rule of Remark~\ref{rem:even-phase-hierarchy}, itself heuristic, suggests that the next even term of \eqref{eq:6jomegaexp} gives an $f\asymp N^{4/5}$ horizon once the first noncontinuum phase is resummed (Remark~\ref{rem:6j-second-secular-horizon}); this is not proved here. \begin{corollary}[Phase-exact centered-Racah pullback] \label{cor:6j-phase-exactness} \label{thm:6j-effective-all-rank-second} \label{cor:6j-effective-all-rank-second} \label{cor:6j-fixed-ratio-second} \label{thm:6j-fixed-ratio-second} Fix admissible $m,n$ and a closed strict-triangle corridor whose swept rank family stays away from $D_6=0$ (hence from the off-diagonal boundaries $\eta=0$, since $\eta^2\ge D_6$). For every admissible terminal rank $f$ in the corridor, \begin{equation} \boxed{ \sqrt{2f+1}\left| \Psi_f(N)-\Bigl(1-\frac{\mathsf J_2(\rho,x)}{N^2}\Bigr)W_6^{[2]}(\rho)\,d^f_{mn}(\Theta_6^{[2]}(\rho))\right| \le \frac{C}{N^3},} \label{eq:6j-phase-exactness} \end{equation} with $\mathsf J_2$ from \eqref{eq:phase-exactness-J2}; without the factor $1-N^{-2}\mathsf J_2$ the bound is $CN^{-2}$. At fixed ratio the absolute errors are $O(N^{-7/2})$ and $O(N^{-5/2})$. \end{corollary} \begin{proof} Theorems~\ref{thm:general-second-order} and~\ref{thm:general-phase-exactness} with Proposition~\ref{prop:racah-class}; the $k=0$ row of the exceptional channel is \eqref{eq:6jexceptionalstart}. The fixed-rank datum is $c(x)=-\mathsf J_2(0,x)$ by the exact degree~$-2$ coefficient~\cite[Theorem~9.1]{LewisFixedRank} (\SuppSec{sec:supp-phase-exactness}). \end{proof} Littlejohn and Yu proposed a uniform approximation of the $6j$ symbol by a rotation matrix, supported by numerical tests~\cite{LittlejohnYu,Yu2011}. For the centered one-small family, Corollary~\ref{cor:6j-phase-exactness} is a rotation-matrix pullback with a proved error bound away from turning points. \begin{corollary}[Bulk centered-Racah zero quantization] \label{cor:racah-bulk-zero-quantization} Fix $m,n$, $0<\rho_0<\rho_1$, and a compact turning-free fixed-ratio bulk set. For fixed $A,B,f,m,n$ regard the normalized $6j$ kernel through its Racah-polynomial continuation in the quadratic $c$-lattice, and let $x_{j,N}^{\rm R}$ be a zero in the compact set. Let $\widetilde x_{j,N}$ be the zero with the same index of $x\mapsto d^f_{mn}(\Theta_6^{[2]}(\rho,x))$, $\rho=(f+\frac12)/N$ (possibly in an $O(N^{-4})$ collar), $p_+=|m-n|$ and $\widetilde\Gamma^{(6)}:=\Gamma_2^{(6)}+\mathfrak c_{mn}(\Theta_6)/\rho$. Then, with a consistent integer index $\nu_j$, \begin{equation} \boxed{x_{j,N}^{\rm R}=\widetilde x_{j,N}+O(N^{-4}),\qquad NS_6(\rho,x_{j,N}^{\rm R})+h\widetilde\Gamma^{(6)}(\rho,x_{j,N}^{\rm R})+\frac{p_+\pi}{2}-\frac\pi4=\left(\nu_j+\frac12\right)\pi+O(N^{-3}).} \label{eq:racah-zero-displacement} \end{equation} In particular the zero of $d^f_{mn}(\Theta_6(\rho,x))$ with the same index lies within $O(N^{-2})$ of $x^{\rm R}_{j,N}$, and $NS_6(\rho,x^{\rm R}_{j,N})+p_+\pi/2-\pi/4=(\nu_j+\frac12)\pi+O(N^{-1})$. \end{corollary} \begin{proof} By Proposition~\ref{prop:racah-class} and Theorem~\ref{thm:general-bulk-zeros}, the polynomial $P_{f,N}$ is a constant multiple of the Racah polynomial in the quadratic spectral variable, since $x$ is affine in $C^2=c(c+1)+\tfrac14$. The exceptional start \eqref{eq:6jexceptionalstart} satisfies the $k=0$ row with $B_0=h^2/(8\alpha\beta)=v(h^2/4,h^2)$. Theorem~\ref{thm:general-bulk-zeros} with $p_{ab}=|m-n|$ gives the statements. \end{proof} \subsection{Normalized turning connections for the centered Racah chain} \label{sec:racah-airy} Semiclassical (discrete WKB) analysis of angular-momentum recurrences goes back to Neville~\cite{Neville1971} and to Schulten and Gordon~\cite{SchultenGordon}; here the centered Jacobi gauge lets the modern difference-equation Airy theory be normalized directly against the Weyl bulk solution. Retain $\eta$, $D_6$ and $\omega$ from \eqref{eq:D6adiabatic}, and put \begin{equation} u_6(\rho):=\frac{x-\rho^2/(4\alpha\beta)}{\eta(\rho)}. \label{eq:racahAiryU} \end{equation} By the exact simplification \eqref{eq:D6linear}, the unique rank-direction turning radius is \begin{equation} \boxed{\rho_* = \frac{2\alpha\beta\sin\theta}{\gamma}.} \label{eq:racahTurningRadius} \end{equation} We restrict to an interior simple turning point, equivalently to a compact strict-triangle family on which $\eta(\rho_*)\ge c>0$. The characteristic roots collide at $+1$ or $-1$ according as \begin{equation} \tau:=\operatorname{sgn}u_6(\rho_*) =\operatorname{sgn}\!\bigl((\alpha^2-\beta^2)^2-\gamma^4\bigr) \in\{+1,-1\}. \label{eq:racahTurnType} \end{equation} \begin{remark}[Excluded boundary collision] \label{rem:racah-collision-excluded} The simple-turn results below exclude $\tau=0$. By \eqref{eq:racahTurnType}, $\tau=0$ means $\alpha^2-\beta^2=\pm\gamma^2$, so the limiting triangle $(\alpha,\beta,\gamma)$ is right-angled at the vertex opposite $\alpha$ or $\beta$ (a right angle opposite $\gamma$ has $\tau=-1$ and is not excluded); then the formal turning radius meets the hard edge $\rho=2\min(\alpha,\beta)$, where the off-diagonal hopping $\eta$ vanishes. They also exclude the degenerate corner $A-B=O(1)$, $C=O(B^{1/2})$, where the strict limiting-triangle hypothesis fails. There the local comparison model is no longer Airy, and uniform two-column asymptotics through a fixed neighbourhood of $\tau=0$, and through the degenerate corner, are open. Since $\varpi_\tau$ grows like $|\varepsilon|^{-8/3}$ as $\varepsilon\to0$ (\eqref{eq:racah-varpi}), the correction $N^{-4/3}\varpi_\tau$ of the bracket is of size $(N\varepsilon^2)^{-4/3}$: the simple-turn expansion can remain uniform as the angle opposite $\alpha$ or $\beta$ approaches $\pi/2$ only while $N\varepsilon^2\to\infty$. \end{remark} For compact simple-turning families with $\tau\in\{+1,-1\}$ we use the following Airy coordinates. For $\tau=+1$, define the signed Airy coordinate \begin{equation} \zeta_+(\rho)= \begin{cases} -\left[\dfrac32\displaystyle\int_\rho^{\rho_*}\omega(s)\,\dd s\right]^{2/3}, &\rho\le\rho_*,\\[9pt] \left[\dfrac32\displaystyle\int_{\rho_*}^{\rho} \operatorname{arcosh}u_6(s)\,\dd s\right]^{2/3}, &\rho\ge\rho_*. \end{cases} \label{eq:racahZetaPlus} \end{equation} For $\tau=-1$, remove the alternating lattice phase: $\zeta_-$ is defined by \eqref{eq:racahZetaPlus} with $\omega$ replaced by $\pi-\omega$ and $u_6$ by $-u_6$, \begin{equation} \zeta_-(\rho):=\zeta_+(\rho)\big|_{\omega\to\pi-\omega,\ u_6\to-u_6}. \label{eq:racahZetaMinus} \end{equation} Write $\zeta_\tau=\zeta_+$ or $\zeta_-$ according to the turn type. With $d_*:=\gamma^2\rho_*/(2\alpha^2\beta^2)$ and $\eta_*:=\eta(\rho_*)$, \begin{equation} \zeta_\tau(\rho)=(d_*/\eta_*^2)^{1/3}(\rho-\rho_*)+O((\rho-\rho_*)^2), \label{eq:racahZetaLinear} \end{equation} so $|\zeta_\tau/D_6|^{1/4}$ extends continuously and positively through the turning point. The remaining global phase can be evaluated without a Racah-polynomial parameter translation. \begin{lemma}[Flat centered-Racah turning action] \label{lem:racahFlatTurnAction} On the branch $0<\omega<\pi$ with $S_6(0)=0$, \begin{equation} \boxed{ \frac{S_6(\rho_*)}{\pi}= \begin{cases} \gamma-|\alpha-\beta|,&\tau=+1,\\[3pt] \rho_*-(\alpha+\beta-\gamma),&\tau=-1. \end{cases}} \label{eq:racahFlatTurnAction} \end{equation} \end{lemma} \begin{proof} \SuppSec{sec:supp-racah}. \end{proof} To state the physical sign, let $p=|n-m|$. In the same ordered Wigner-$d$ convention, \eqref{eq:wigner-darboux-sign} gives the fixed sign $\epsilon^{(6)}_{mn}=(-1)^{\max(n-m,0)}$ and \begin{equation} d^f_{mn}(\vartheta)=\epsilon^{(6)}_{mn} \sqrt{\frac{2}{\pi F\sin\vartheta}} \left[ \cos\!\left(F\vartheta+\frac{p\pi}{2}-\frac\pi4\right) +O(F^{-1}) \right],\qquad F=f+\tfrac12. \label{eq:racahWignerAirySign} \end{equation} Define \begin{align} K_+&:=C-|A-B|+\frac{p-1}{2}, &&(\tau=+1), \label{eq:racahKplus}\\ K_-&:=F-A-B+C+\frac p2, &&(\tau=-1). \label{eq:racahKminus} \end{align} The triangle parities make $K_\pm$ integers, also for half-integer $m,n$. On the $-1$ branch, $K_-$ changes by one when $f$ changes by one, exactly encoding the alternating-lattice gauge. \begin{theorem}[Weyl-normalized centered Racah Airy connection] \label{thm:racahAiryConnection} Fix admissible $m,n$, $A_0<\infty$, and a compact family of strict limiting triangles for which the turning point \eqref{eq:racahTurningRadius} is interior and simple and the turn type $\tau$ is fixed. Let $F=f+\tfrac12=N\rho$. Uniformly at admissible ranks satisfying \begin{equation} |N^{2/3}\zeta_\tau(\rho)|\le A_0, \end{equation} one has \begin{equation} \boxed{ \Psi_f(N)= \epsilon^{(6)}_{mn}(-1)^{K_\tau} \sqrt{\frac{2}{\rho}}\,N^{-1/3} \left|\frac{\zeta_\tau(\rho)}{D_6(\rho)}\right|^{1/4} \left[ \operatorname{Ai}(t)+N^{-4/3}\varpi_\tau\operatorname{Ai}'(t) +O_{A_0}(N^{-2}) \right],} \label{eq:racahAiryConnection} \end{equation} where $K_\tau=K_+$ on the $+1$ branch and $K_\tau=K_-$ on the $-1$ branch, $t=N^{2/3}\zeta_\tau(\rho)$, and $\varpi_\tau$, the value at $\rho_*$ of \eqref{eq:general-airy-varpi-local} for the data \eqref{eq:racah-class-data}, is \begin{equation} \boxed{\varpi_\tau=\frac{2^{1/3}\bigl[140\,\varepsilon^3\bigl((m^2+n^2)x-2mn\bigr)+\mathcal P(x,s_*)\bigr]}{280\,(\rho_*\sin\theta)^{4/3}\,|\varepsilon|^{8/3}},\qquad s_*:=\frac{\rho_*^2}{4\alpha\beta},\quad\varepsilon:=x-s_*,} \label{eq:racah-varpi} \end{equation} \begin{equation} \mathcal P(x,s_*)=-50x^4+15x^2-9+6s_* x(25x^2-3)+s_*^2(21-149x^2)+36s_*^3x+4s_*^4, \end{equation} with $x=\cos\theta$ and $\tau=\operatorname{sgn}\varepsilon$. Here $s_*=b_0(\rho_*)$ is the diagonal shift at the turn; for the Hahn chain $b_0=0$, and the same formula with $\mathcal P(x,0)$, $\varepsilon=x$, $\rho_*\to q_*=\sin\theta$ and offsets $(M,\sigma)$ in place of $(m,n)$ is \eqref{eq:hahn-varpi}. In terms of the angles $A_\alpha,A_\beta$ of the limiting triangle opposite $\alpha,\beta$, $s_*=b_0(\rho_*)=\sin A_\alpha\sin A_\beta$, $\varepsilon=\tau\eta(\rho_*)=-\cos A_\alpha\cos A_\beta$ and $\rho_*\sin\theta=2\gamma\,s_*$, so $\tau=\operatorname{sgn}\varepsilon$ is \eqref{eq:racahTurnType}, and $\varpi_\tau$ blows up like $|\varepsilon|^{-8/3}$ at the excluded triangles where $A_\alpha$ or $A_\beta$ is $\pi/2$. The uniform form \eqref{eq:general-rank-airy-uniform} also holds on the neighbourhood $I'$ of $\rho_*$ of Theorem~\ref{thm:general-rank-airy}(ii), and, for this chain, to the end of $I$ except for a collar of width $O(N^{-1}\log N)$ (Proposition~\ref{prop:forbidden-decay}(c)). The $O_{A_0}(N^{-2})$ term is additive inside the square brackets, as in Theorem~\ref{thm:hahnAiryConnection}, so the error of $\operatorname{Ai}$ alone is $O_{A_0}(N^{-4/3})$, attained for the window supremum where $\varpi_\tau\ne0$ (and pointwise where also $\operatorname{Ai}'(t)\ne0$). Once the ordered Wigner-$d$ sign $\epsilon^{(6)}_{mn}$ is chosen, the physical centered $6j$ solution continues as the normalized recessive $\operatorname{Ai}$ branch beyond either simple Racah turning point, with no additional connection phase. \end{theorem} \begin{proof} We verify the hypotheses of Theorem~\ref{thm:general-rank-airy}. By \eqref{eq:D6linear}, $D_6>0$ on $0\le\rho<\rho_*$ and $\partial_\rho D_6(\rho_*)=-\gamma^2\rho_*/(2\alpha^2\beta^2)<0$, while $e(\rho_*^2)=\eta(\rho_*)\ge c$. Any $s_1\in(\rho_*^2,4\min(\alpha,\beta)^2)$, uniformly on the compact family, is admissible in Proposition~\ref{prop:racah-class}, so Lemma~\ref{lem:general-airy-applicability} gives the uniform coefficient bounds. The characteristic quotient is $u_6$, so \eqref{eq:racahZetaPlus}--\eqref{eq:racahZetaMinus} are the coordinates \eqref{eq:general-zeta}, and \eqref{eq:wigner-darboux-sign} with $(a,b)=(m,n)$ is \eqref{eq:racahWignerAirySign}. Since $A=N\alpha$, $B=N\beta$, $C=N\gamma$, Lemma~\ref{lem:racahFlatTurnAction} gives $NS_6(\rho_*)=\pi(C-|A-B|)$ for $\tau=+1$ and $NS_6(\rho_*)-\pi N\rho_*=-\pi(A+B-C)$ for $\tau=-1$, so the quantities \eqref{eq:general-turn-quantization} are exactly $K_\pm$ of \eqref{eq:racahKplus}--\eqref{eq:racahKminus}, which are integers. Theorem~\ref{thm:general-rank-airy} gives \eqref{eq:racahAiryConnection} with bracket error $O_{A_0}(N^{-1})$. Condition~(R) of Theorem~\ref{thm:general-rank-airy}(ii) holds because $6j$ orthogonality, $\sum_f(2f+1)(2c+1)W_f^2=1$, gives $|\widehat\Psi_f|=2\sqrt{(2f+1)AB}\,|W_f|\le2\sqrt{AB/(2c+1)}=O(N^{1/2})$ for all $f$; that theorem gives the bracket $\operatorname{Ai}(t)+N^{-4/3}\varpi_\tau\operatorname{Ai}'(t)+O_{A_0}(N^{-2})$ and \eqref{eq:general-rank-airy-uniform}, with $\varpi_\tau$ from \eqref{eq:general-airy-varpi-local}, which for the data \eqref{eq:racah-class-data} is \eqref{eq:racah-varpi} (Lemma~\ref{lem:racah-varpi}). \end{proof} For Racah polynomials, Wang and Wong developed large-degree Racah asymptotics with varying parameters, including outer and oscillatory regimes, and note that their uniform difference-equation technique supplies turning-point formulas without carrying out that specialization~\cite{WangWongRacah}. The present contribution is complementary and needs no translation to a printed Racah-polynomial parameter chamber: the Regge/Weyl centering selects the exact $N^{-2}$-even recurrence, and the evaluated turning action gives the canonical Airy basis the physical $6j$ normalization, branch and lattice parity directly. \subsection{Geometric interpretation} \label{sec:geometry} \label{sec:HJ} \label{sec:6j-dual-HJ} \label{sec:racah-tetrahedral} The results above use only the actions as integrals of the principal rank momentum. Their geometric meaning is recorded here; none of it is used in the proofs. We begin with the bispectral meaning of the Hahn action. For fixed $M,\sigma$ the rank recurrence gives, at principal order, $\cos S_q=x/\sqrt{1-q^2}$, while the magnetic recurrence gives $S_x=-p_q$ with $p_q$ from \eqref{eq:pq}, and $\partial_qS_x=-(1-x^2-q^2)^{-1/2}=\partial_xS_q$. Hence \begin{equation} \omega_0= \arccos\!\left(\frac{x}{\sqrt{1-q^2}}\right)\dd q -\arcsin\!\left(\frac{q}{\sqrt{1-x^2}}\right)\dd x \label{eq:oneformfixed} \end{equation} is closed, its primitive is $S(q,x)$ of \eqref{eq:Theta}, and $\Lambda_S=\{(q,p_q;x,p_x):p_q=S_q,\ p_x=-S_x\}$ is an exact Lagrangian graph on the turning-free bulk, an instance of the bispectral viewpoint of Duistermaat and Gr\"unbaum~\cite{DuistermaatGrunbaum}. The effective angle is the action per unit angular momentum, not an ad hoc coordinate correction. By Remark~\ref{lem:action-zero-transport} the effective coordinates transport the zero measures exactly: with $\rho_q(x)=-S_x/(\pi q)$ and $\varrho^{(6)}_\rho:=-\partial_xS_6/(\pi\rho)$, positive on the turning-free band because $\partial_x\omega=-1/(\eta\sin\omega)$, \begin{equation} \rho_q(x)\,\dd x=\rho_0(x_q)\,\dd x_q,\qquad \varrho^{(6)}_\rho(x)\,\dd x=\rho_0(x_{6,\rho})\,\dd x_{6,\rho},\qquad x_{6,\rho}:=\cos\Theta_6. \label{eq:hahn-zero-transport} \end{equation} These are action-Jacobian identities, not asymptotic matchings of densities, and place the construction within the constrained-equilibrium picture for discrete orthogonal polynomials~\cite{RakhmanovDiscrete,DragnevSaff,KuijlaarsVanAssche,BaikKriecherbauerMcLaughlinMiller}. The Racah action has a tetrahedral realization. Let $F=f+\tfrac12=N\rho$ and consider the principal Weyl tetrahedron with edge array $(A,B,C;B,A,F)$. With the edge $F$ on an axis the transverse radii are $r_A=\sqrt{A^2-F^2/4}$ and $r_B=\sqrt{B^2-F^2/4}$, and if $\Theta_F$ is the dihedral angle at $F$, the opposite edge satisfies $C^2=r_A^2+r_B^2-2r_Ar_B\cos\Theta_F$. So the principal Racah symbol is exactly $\omega(\rho)=\Theta_F$, on the branch with $\Theta_F\to\theta$ as $F\downarrow0$. The Schl\"afli identity, with $\mathcal S_{\rm PR}$ written in terms of interior dihedral angles (with the exterior angles of the usual Ponzano--Regge form it reads $\dd\mathcal S/\dd F=\pi-\Theta_F$ and changes $\mathcal S_{\rm PR}$ by $\pi F$), gives $\dd\mathcal S_{\rm PR}/\dd F=\Theta_F$, so $\mathcal S_{\rm PR}(F)=NS_6(\rho)+\mathrm{const}$ when the other edges are fixed, and $\mathcal V=Fr_Ar_B\sin\omega/6$, so the Ponzano--Regge amplitude, after the normalization $2\sqrt{AB}$ of \eqref{eq:6jPhi}, is \begin{equation} \mathcal A_6(\rho)=\sqrt{\frac{2}{\pi F\,\eta(\rho)\sin\omega(\rho)}}, \label{eq:6jPRamp} \end{equation} which in the channel $m=n=0$ agrees with the amplitude of Proposition~\ref{prop:racah-data}(b) because $\eta\sin\omega=\sqrt{D_6}$. Roberts proved the Ponzano--Regge formula~\cite{PonzanoRegge1968} for uniformly dilated admissible labels~\cite{Roberts1999}; that theorem does not literally cover the fixed-offset centered family, because $O(1)$ edge offsets contribute an $O(1)$ phase, which Proposition~\ref{prop:racah-data}(b) supplies. The integral of $\rho_2^{(6)}$ is the first $h^2$ fixed-offset correction to this rank-direction Regge phase; its invariant tetrahedral form is open (Section~\ref{sec:discussion}). \begin{lemma}[The Racah turning curve is the flat-tetrahedron locus] \label{lem:racah-cayley-menger} Let $\mathrm{CM}$ be the Cayley--Menger matrix of the principal Weyl tetrahedron and $T$ the area of the triangle with sides $(\alpha,\beta,\gamma)$. Then \begin{equation} \boxed{ \det\mathrm{CM} =8N^6\rho^2\alpha^2\beta^2\,D_6(\rho) =2N^6\rho^2\bigl(16T^2-\gamma^2\rho^2\bigr).} \label{eq:racah-cayley-menger} \end{equation} Consequently the Racah turning curve $D_6=0$ is exactly the locus where the principal Weyl tetrahedron is flat; $D_6>0$ is the nondegenerate Euclidean tetrahedron, and for $D_6<0$ the six lengths are not those of any Euclidean tetrahedron. \end{lemma} \begin{proof} \SuppSec{sec:supp-racah}. \end{proof} \begin{remark}[Dual physical-$c$ problem] \label{prop:6j-dual-principal-symbol} The fixed-rank theory constructs the exact self-adjoint Racah operator on the physical $c$-lattice, with eigenvalue $f(f+1)$~\cite[Theorem~8.2]{LewisFixedRank}. Put $\Delta_C(\gamma):=[\gamma^2-(\alpha-\beta)^2][(\alpha+\beta)^2-\gamma^2]$. In the all-large scaling its principal symbol is $\mathcal H_C(\gamma,p_C)=\Delta_C(\gamma)(1-\cos p_C)/(2\gamma^2)$, so the dual Hamilton--Jacobi equation is \begin{equation} \rho^2=\frac{\Delta_C(\gamma)}{2\gamma^2}(1-\cos p_C),\qquad p_C=2\arcsin\!\left(\frac{\rho\gamma}{\sqrt{\Delta_C(\gamma)}}\right). \label{eq:racah-dual-HJ} \end{equation} Since $\Delta_C=4\alpha^2\beta^2(1-x^2)$, the dual oscillatory condition $\Delta_C-\rho^2\gamma^2>0$ is exactly $D_6>0$, and the two momenta satisfy $\partial_\gamma\omega=\partial_\rho p_C=2\gamma/\sqrt{\Delta_C-\rho^2\gamma^2}$. Hence $S_6(\rho,\gamma)$ is a bispectral generating function, \begin{equation} \dd S_6=\omega\,\dd\rho+p_C\,\dd\gamma, \label{eq:racah-bispectral-action} \end{equation} and the rank and physical-edge problems share the principal caustic and the principal action (proof in \SuppSec{sec:supp-racah}). \end{remark} \section{The one-small quantum \texorpdfstring{$6j$}{6j} chain} \label{sec:q6j} The quantum $6j$ symbol of $U_q(\mathfrak{sl}_2)$ with deformation parameter $q_{\rm def}=e^{1/N}$ (distinct from the rank coordinate $q$) has the same one-small rank recurrence as the centered Racah chain, with $q$-numbers $[z]:=\sinh(z/N)/\sinh(1/N)$ in place of integers. This section shows that it is an even centered Wigner--Jacobi chain in the sense of Definition~\ref{def:even-centered-chain} (with $e_\pi$ genuinely $t$-dependent), so that Sections~\ref{sec:general-first}--\ref{sec:general-zeros-turns} apply, and gives the explicit data. The curvature scale is normalized to $1$; a scale $\varsigma$ (that is, $q_{\rm def}=e^{\varsigma/N}$) is obtained by $N\to N/\varsigma$ in the $q$-numbers and the Weyl lengths, and multiplies the flux defect below by $\varsigma^2$. Proofs are in \SuppSec{sec:supp-q6j}; steps that rest on exact computer algebra (rational-function identities with symbolic labels) say so where they occur. We first define the family. With the Kirillov--Reshetikhin $q$-$6j$ symbol~\cite{KirillovReshetikhin1989} take, for fixed admissible $m,n\in\tfrac12\mathbb Z$ with $m-n\in\mathbb Z$, \begin{equation} W_f=\sixj{a}{b}{c}{b+m}{a+n}{f}_{\!q},\qquad A=a+\tfrac{n+1}2=N\alpha,\quad B=b+\tfrac{m+1}2=N\beta,\quad C=c+\tfrac12=N\gamma, \end{equation} with the same labels as in \eqref{eq:6jfamily}--\eqref{eq:ABC}. Put $\mathsf X=2\alpha$, $\mathsf Y=2\beta$, $\mathsf Z=2\gamma$ and assume the strict triangle inequalities $|\mathsf X-\mathsf Y|<\mathsf Z<\mathsf X+\mathsf Y$. Then $(\mathsf X,\mathsf Y,\mathsf Z)$ are the side lengths of a hyperbolic triangle of curvature $-1$, with angle $\theta$ opposite $\mathsf Z$, \begin{equation} \begin{gathered} x=\cos\theta=\frac{\cosh\mathsf X\cosh\mathsf Y-\cosh\mathsf Z}{\sinh\mathsf X\sinh\mathsf Y},\qquad \mu:=\frac{\sinh\mathsf Z}{\sinh\mathsf X\sinh\mathsf Y},\\ \Delta_0:=\coth^2\mathsf X+\coth^2\mathsf Y-2x\coth\mathsf X\coth\mathsf Y=\sin^2\theta+\mu^2 . \end{gathered} \label{eq:q6j-triangle} \end{equation} (The Weyl lengths $\alpha,\beta,\gamma$ play the role of the Racah triangle $(\alpha,\beta,\gamma)$; the flat limit is Corollary~\ref{cor:q6j-consequences}(vi).) The natural-gauge solution is \begin{equation} \widehat\Psi_k:=(-1)^{a+b+c+k+m}\sqrt{[2k+1][2A][2B]}\;W_k,\qquad \Psi_k=\widehat\Psi_k/\sqrt{2k+1}, \label{eq:q6j-Psi} \end{equation} the $q$-analogue of \eqref{eq:6jPhi}, with $\Psi_k\to d^k_{mn}(\theta)$ at $h\to0$. \begin{proposition}[The $q$-$6j$ chain is an even centered chain] \label{prop:q6j-class} \begin{enumerate}[label=(\alph*),leftmargin=2em] \item \emph{Exact Jacobi form.} With $\mathrm{s}_z:=\sinh(z/N)$, \begin{equation} \begin{aligned} x\widehat\Psi_k&=A_{k+1}\widehat\Psi_{k+1}+A_k\widehat\Psi_{k-1}+B_k\widehat\Psi_k,\\ A_k^2&=\frac{4\,\mathrm{s}_{k-n}\mathrm{s}_{k+n}\,\mathrm{s}_{k-m}\mathrm{s}_{k+m}\,\bigl(1-\mathrm{s}_k^2/\sinh^2\mathsf X\bigr)\bigl(1-\mathrm{s}_k^2/\sinh^2\mathsf Y\bigr)}{\mathrm{s}_{2k}^2\,\mathrm{s}_{2k+1}\mathrm{s}_{2k-1}},\\ B_k&=\coth\mathsf X\coth\mathsf Y\Bigl(1-\frac{2\cosh\frac mN\cosh\frac nN}{\cosh\frac{2k+1}N+\cosh\frac1N}\Bigr)+\frac{\mathrm{s}_m\mathrm{s}_n}{\mathrm{s}_k\,\mathrm{s}_{k+1}} . \end{aligned} \label{eq:q6j-AkBk} \end{equation} with $f_{\min}:=\max(|m|,|n|)$ and $A_{f_{\min}}:=0$ (the quotient for $A_k^2$ is $0/0$ when $|m|=|n|=\tfrac12$). For $m=n=0$ the row $k=0$ reads $x\widehat\Psi_0=A_1\widehat\Psi_1+B_0\widehat\Psi_0$ with $B_0=\coth\mathsf X\coth\mathsf Y(1-1/\cosh\frac1N)$ and $\widehat\Psi_0=1$. \item \emph{(E1) with explicit data.} Put $K=\sqrt s$, $H=\sqrt t$, $\Xi(u,w):=(\cosh2u-\cosh2w)/(2(u^2-w^2))$ (an analytic function of $(u^2,w^2)$, $\Xi(0,0)=1$). Then $A_k=A_k^{(0)}e(k^2h^2,h^2)$ and $B_k=\lambda(h^2)B_k^{(0)}+v(q_k^2,h^2)$ exactly, with \begin{equation} \begin{gathered} e(s,t)^2=\frac{\Xi(K,Hn)\,\Xi(K,Hm)\,\bigl(1-\sinh^2K/\sinh^2\mathsf X\bigr)\bigl(1-\sinh^2K/\sinh^2\mathsf Y\bigr)}{\Xi(2K,0)\,\Xi(2K,H)},\\ \lambda(t)=e(0,t)=\frac{\operatorname{shc}(Hm)\operatorname{shc}(Hn)}{\operatorname{shc}(H)},\qquad \operatorname{shc}(u):=\frac{\sinh u}u,\\ v(s,t)=\coth\mathsf X\coth\mathsf Y\Bigl(1-\frac{2\cosh Hm\cosh Hn}{\cosh2K+\cosh H}\Bigr)+\frac{2\sinh Hm\sinh Hn}{\cosh2K-\cosh H}-\lambda(t)\frac{mnH^2}{K^2-H^2/4}, \end{gathered} \label{eq:q6j-e} \end{equation} (the last term of $v$ is $0$ if $mn=0$), jointly analytic in $(s,t)$, any $s_1<4\min(\alpha,\beta)^2$ being admissible. The class data are $e(s)=e(s,0)=\sqrt{(1-\tanh^2q\coth^2\mathsf X)(1-\tanh^2q\coth^2\mathsf Y)}$ and $b_0(q)=\tanh^2q\,\coth\mathsf X\coth\mathsf Y$ ($q^2=s$), and \begin{equation} \begin{gathered} \varepsilon_1=\tfrac12(\coth^2\mathsf X+\coth^2\mathsf Y),\quad \beta_1=\coth\mathsf X\coth\mathsf Y,\quad \lambda_1=\tfrac16(m^2+n^2-1),\\ \beta_2=\tfrac12(\tfrac12-m^2-n^2)\coth\mathsf X\coth\mathsf Y-\tfrac13mn,\quad \chi(x)=\tfrac x2(\coth^2\mathsf X+\coth^2\mathsf Y)-\coth\mathsf X\coth\mathsf Y,\\ \eta(q^2)=\frac1{q^2}\Bigl[\frac{(m^2+n^2)(\sinh^2q-q^2)}{2\sinh^2q}+\frac12\Bigl(\frac{q^2}{\sinh^22q}-\frac14\Bigr)\Bigr], \end{gathered} \label{eq:q6j-data} \end{equation} so that $a_2/a_0=\frac1{2\sinh^22q}-\frac{m^2+n^2}{2\sinh^2q}$ and $\partial_tv(q^2,0)=-\coth\mathsf X\coth\mathsf Y\bigl(\frac{m^2+n^2}{2\cosh^2q}-\frac1{4\cosh^4q}\bigr)+mn\bigl(\frac1{\sinh^2q}-\frac1{q^2}\bigr)$. The principal discriminant is \begin{equation} D(q,x)=\frac{\sin^2\theta-\mu^2\sinh^2q}{\cosh^2q},\qquad \sinh q_*=\frac{\sin\theta}{\mu},\qquad \partial_qD(q_*)=-\frac{2\sin^2\theta}{\tanh q_*\cosh^2q_*}<0 . \label{eq:q6j-D} \end{equation} \item \emph{(E2) and (R).} The two lowest ranks of $\widehat\Psi$ are the values at $t=h^2$ of functions analytic in $(x,t)$ on a complex neighbourhood of $X$ times a disk (even in $h$) that equal $\widehat d_{f_{\min}},\widehat d_{f_{\min}+1}$ at $h=0$. Moreover (R) holds both intrinsically at a simple turn by Proposition~\ref{prop:exact-hyperbolicity} and globally via $\widehat\Psi_k^2\le[2A][2B]/[2C]=O(N)$ for all $k$. \item \emph{Turn.} The turn $q_*$ of \eqref{eq:q6j-D} is simple with $q_*<\min(\mathsf X,\mathsf Y)$ if and only if the triangle is not right-angled at the vertex opposite $\mathsf X$ or $\mathsf Y$. Its type is $\tau=\operatorname{sgn}\varepsilon_*$, \begin{equation} \begin{gathered} \varepsilon_*:=x-b_0(q_*)=\frac{(x\coth\mathsf X-\coth\mathsf Y)(\coth\mathsf X-x\coth\mathsf Y)}{\Delta_0},\\ \tau=-\operatorname{sgn}\bigl[(\cosh\mathsf X-\cosh\mathsf Y\cosh\mathsf Z)(\cosh\mathsf Y-\cosh\mathsf X\cosh\mathsf Z)\bigr], \end{gathered} \label{eq:q6j-tau} \end{equation} and the lattice quantization \eqref{eq:general-turn-quantization} holds exactly: $K_+=C-|A-B|+\tfrac{|m-n|-1}2$ and $K_-=\kappa-(A+B-C)+\tfrac{|m-n|}2$ are integers. \end{enumerate} \end{proposition} Part (a) is the three-term recurrence of the $q$-Racah polynomials in the degree variable (the $q$-analogue of the Schulten--Gordon recurrence~\cite{AskeyWilson1979,KoekoekLeskySwarttouw}) in the Weyl-midpoint gauge. It is proved here from the Kirillov--Reshetikhin sum~\cite{KirillovReshetikhin1989} by a $q$-Zeilberger certificate (after reducing general $(m,n)$ to the fundamental chamber $m\ge|n|$ by the tetrahedral symmetries of the sum): after the gauge change $\sqrt{[2f+1]}\,P(f)$ the recurrence of the sum is an identity of rational functions in the eight symbols $q,q^a,q^b,q^c,q^f,q^m,q^n,q^z$, verified by exact symbolic computation (Lemma~\ref{lem:q6j-recurrence}); only the definition (K1) is needed (with $q$-orthogonality (K2) supplying an independent global bound for (R) and the Gauss-weight identity of Remark~\ref{rem:flux-gauss}). Everything else in the proposition is proved from (a) in \SuppSec{sec:supp-q6j}, including the closed seed, which is a one-term sum. The $q$-deformation also makes $e(0,t)=\lambda(t)\ne1$, whereas $e(0,t)=1$ for the Hahn and Racah chains: without the factor $\lambda$, the $h^4/q^2$ coefficient of $B_k-B_k^{(0)}$ is $mn(m^2+n^2-1)/6\ne0$ and $v$ would not be analytic at $q=0$. \begin{lemma}[Turning action of the $q$-$6j$ chain] \label{lem:q6j-turn-action} With $p(q)=\arccos((x-b_0)/e)$ on the branch $p(0)=\theta$, $q_*$ as in \eqref{eq:q6j-D}, \begin{equation} \boxed{\;\int_0^{q_*}p\,\dd q=\begin{cases}\dfrac\pi2\,\bigl(\mathsf Z-|\mathsf X-\mathsf Y|\bigr),&\tau=+1,\\[6pt] \pi q_*-\dfrac\pi2\,(\mathsf X+\mathsf Y-\mathsf Z),&\tau=-1,\end{cases}} \label{eq:q6j-turn-action} \end{equation} that is $NS(q_*)/\pi=C-|A-B|$ for $\tau=+1$ and $NS(q_*)/\pi-Nq_*=-(A+B-C)$ for $\tau=-1$: the Racah values of Lemma~\ref{lem:racahFlatTurnAction} with the Weyl labels of the quantum symbol in place of the flat ones. \end{lemma} The geometry behind it is that $p(q)$ is the dihedral angle at the edge of length $2q$ of the hyperbolic tetrahedron with the edge array $(A,B,C;B,A,F)$ of Section~\ref{sec:geometry}, $F=f+\frac12$, all lengths multiplied by $2/N$ (hyperbolic edges $2q;\mathsf Z;\mathsf X,\mathsf X;\mathsf Y,\mathsf Y$), by exact symbolic computation: $\cos^2p=C_{34}^2/(C_{33}C_{44})$ for the cofactors of the matrix of cosh's of the edge lengths, the hyperbolic analogue of the principal Racah symbol being a dihedral angle. The proof of \eqref{eq:q6j-turn-action} avoids degenerate tetrahedra: it differentiates the action in $\mathsf Z$ at fixed $\mathsf X,\mathsf Y$, where $\partial_{\mathsf Z}p=\mu/\sqrt D$ turns the integral into $\arcsin(\mu\sinh q/\sin\theta)|_0^{q_*}=\pi/2$, and integrates from the degenerate triangle (\SuppSec{sec:supp-q6j}). This is a Schl\"afli-type identity for the one-parameter family of tetrahedra; the integer $K_\tau$ is the lattice quantization needed by Theorem~\ref{thm:general-rank-airy}. \begin{corollary}[The growing-rank theory for the one-small $q$-$6j$] \label{cor:q6j-consequences} Let $(\alpha,\beta,\gamma)$ range over a compact family of strict triangles, with fixed admissible $m,n\in\tfrac12\mathbb Z$, $m-n\in\mathbb Z$. \begin{enumerate}[label=(\roman*),leftmargin=2em] \item \emph{Horizon.} Theorem~\ref{thm:general-mesoscopic} and Theorem~\ref{thm:general-critical} hold with $\chi$ of \eqref{eq:q6j-data}: $\sqrt{\pi\kappa\sin\theta/2}\,\Psi_L=\cos(\varphi_{L,mn}(\theta)+C_q\kappa^3/N^2)+O(\kappa^{-1}+\kappa^2N^{-2}+\kappa^5N^{-4})$ with $C_q=-\chi(x)/(3\sin\theta)$. The coherence horizon is $L\asymp N^{2/3}$ where $\chi(x)\ne0$, and $C_q\to C_6$ of \eqref{eq:6jcriticalphasepred} in the flat limit. \item \emph{Second-order pullback.} Theorem~\ref{thm:general-second-order} holds for $\Psi_L$, with $\Theta^{[2]}_q$, $W^{[2]}_q$ built from \eqref{eq:q6j-data} and \eqref{eq:class-a2b2}, \eqref{eq:class-second-data}: $\sqrt{2L+1}\,|\Psi_L-W^{[2]}_qd^L_{mn}(\Theta^{[2]}_q)|\le CN^{-2}$ in any turning-free corridor. \item \emph{Phase exactness, with a flux defect.} Theorem~\ref{thm:general-phase-exactness} holds with $\mathsf J_2$ from \eqref{eq:phase-exactness-J2}--\eqref{eq:phase-exactness-J20} and \begin{equation} \boxed{c(x)=-\mathsf J_2(0,x)-\frac1{12},\qquad \delta_{\rm fl}=-\frac1{12},\qquad \alpha(q,x)=-\mathsf J_2(q,x)-\frac1{12}.} \label{eq:q6j-c} \end{equation} This is proved from the closed seed by exact symbolic computation: in the fundamental chamber $m\ge|n|$ at $k=m$ (to which general $(m,n)$ reduce by the symmetries of (K1), which leave $\widehat\Psi_k/\widehat U^{[2]}_k$ and $\mathsf J_2$ invariant) both $[t]\log\widehat\Psi_m$ and $[t]\log\widehat U^{[2]}_m$ are explicit, and their difference plus $\mathsf J_2(0,x)$ is $-\frac1{12}$ identically in $(e^{\mathsf X},e^{\mathsf Y},e^{\mathsf Z},m,n)$; the $k$-independence of $c$ is Proposition~\ref{prop:general-fixed-rank-contact}. The constant $-\frac1{12}$ is the $h^2$-coefficient of $\frac12\,h/\sinh h$ in the exact Gauss-weight identity (Remark~\ref{rem:flux-gauss}). Consequently $\sqrt{2L+1}\,|\Psi_L-(1-N^{-2}(\mathsf J_2+\frac1{12}))W^{[2]}_qd^L_{mn}(\Theta^{[2]}_q)|\le CN^{-3}$. \item \emph{Bulk zeros.} The polynomial $P_{L,N}$ is a constant multiple of the $q$-Racah polynomial in $\cosh\mathsf Z$ ($x$ is affine in $\cosh\mathsf Z$), and Theorem~\ref{thm:general-bulk-zeros} holds: its zeros are within $O(N^{-4})$ of those of $d^L_{mn}(\Theta^{[2]}_q)$ and satisfy the Bohr--Sommerfeld rule \eqref{eq:general-bulk-zeros} with $p_{ab}=|m-n|$ and $S=\int_0^qp$. \item \emph{Turning connection and forbidden-side decay.} If the triangle is not right-angled at the vertex opposite $\mathsf X$ or $\mathsf Y$ and the turn type $\tau$ is fixed on the compact family, Theorem~\ref{thm:general-rank-airy} and Propositions~\ref{prop:exact-hyperbolicity}--\ref{prop:forbidden-decay} hold for $\Psi_L$ with $\epsilon_{mn}=(-1)^{\max(n-m,0)}$, $K_\pm$ and $\tau$ as in Proposition~\ref{prop:q6j-class}(d), $q_*,D$ from \eqref{eq:q6j-D}, $u_1=\sin^2\theta/(\sinh q_*\cosh q_*\,\varepsilon_*^2)$, and $\varpi_\tau$ from \eqref{eq:general-airy-varpi-local}--\eqref{eq:general-airy-varpi-olver} with the data \eqref{eq:q6j-data}, which contain the $\eta$-term. With Proposition~\ref{prop:airy2-amplitude} and $\delta_{\rm fl}=-\frac1{12}$ the uniform bracket \eqref{eq:general-rank-airy-uniform} has the amplitude $1+N^{-2}(-\frac1{12}-\frac12\mathfrak j(q))$ and holds on $I$ except for a collar of width $O(N^{-1}\log N)$ at its forbidden end. \item \emph{Flat limit.} Under $q_{\rm def}=e^{\varsigma/N}$ at step $h=N^{-1}$, put $\mathsf X=2\varsigma\alpha$, $\mathsf Y=2\varsigma\beta$, $\mathsf Z=2\varsigma\gamma$ and $q=\varsigma\rho$ ($\rho=(f+\frac12)/N$), so the (E1) arguments rescale as $(s,t)\mapsto(\varsigma^2\rho^2,\varsigma^2h^2)$, multiplying $\varepsilon_1,\beta_1,\beta_2,\chi,\lambda_1,\eta,\delta_{\rm fl}$ of \eqref{eq:q6j-data} by $\varsigma^2$ and $\varpi_\tau$ by $\varsigma^{4/3}$. As $\varsigma\to0$, $x\to(\alpha^2+\beta^2-\gamma^2)/(2\alpha\beta)$, $q_*/\varsigma\to\rho_*$, $e(\varsigma^2\rho^2)\to e^{(6)}(\rho^2)$, $b_0(\varsigma\rho)\to\rho^2/(4\alpha\beta)$ of \eqref{eq:racah-class-data}, $\varsigma^2(\varepsilon_1,\beta_1,\beta_2,\chi)$ tend to their Racah values, $\varsigma^2(\lambda_1,\eta(\varsigma^2\rho^2),\delta_{\rm fl})\to0$, \eqref{eq:q6j-D} becomes \eqref{eq:D6linear}, \eqref{eq:q6j-turn-action} divided by $\varsigma$ has the form of Lemma~\ref{lem:racahFlatTurnAction} identically at every $\varsigma>0$ (with the turning rank of the quantum chain), and $\varsigma^{4/3}\varpi_\tau\to\varpi_\tau$ of \eqref{eq:racah-varpi}. \end{enumerate} \end{corollary} \section{Comparison with exact symbols} \label{sec:numerics} The proofs are analytic, and the comparisons in this section illustrate the rates; none is used in a proof. Table~\ref{tab:accuracy} compares the main statements with exact $3j$ and $6j$ symbols. Figure~\ref{fig:horizon} uses the exact $3j$ rank recurrence, started from an exact minimal-rank value and run in $30$-digit arithmetic, to compare the fixed-rank Wigner approximation with the leading and second-order pullbacks of Proposition~\ref{prop:hahn-data} and the phase-exact pullback of Corollary~\ref{cor:3j-phase-exactness}. \begin{table}[ht] \centering \small \begin{tabular}{llcccc} \toprule & case & \multicolumn{2}{c}{$N=160$} & \multicolumn{2}{c}{$N=640$}\\ \cmidrule(lr){3-4}\cmidrule(lr){5-6} & & without & with & without & with\\ \midrule \multicolumn{6}{l}{\emph{(a) Fixed-ratio pullback, $\max\sqrt{2L+1}\,|Y_L-U_L|$; $U=U^{[2]}$ without, $(1-N^{-2}\mathsf J_2)U^{[2]}$ with}}\\ & $3j$, $(M,\sigma)=(2,1)$, $x=0.3$, $q\approx0.45$ & $2.3\cdot10^{-5}$ & $3.8\cdot10^{-8}$ & $1.3\cdot10^{-6}$ & $4.3\cdot10^{-10}$\\ & $3j$, $(\frac32,-\frac12)$, $x\approx0.2$, $q\approx0.5$ $^\dagger$ & $2.5\cdot10^{-6}$ & $4.1\cdot10^{-8}$ & $1.5\cdot10^{-7}$ & $6.6\cdot10^{-10}$\\ & $6j$, $(m,n)=(2,1)$, $(\alpha,\beta,\gamma)=(1,\frac65,\frac32)$, $\rho\approx0.9$ & $4.7\cdot10^{-6}$ & $1.3\cdot10^{-7}$ & $3.0\cdot10^{-7}$ & $2.0\cdot10^{-9}$\\ \midrule & & \multicolumn{2}{c}{$N=100$} & \multicolumn{2}{c}{$N=800$}\\ \cmidrule(lr){3-4}\cmidrule(lr){5-6} \multicolumn{6}{l}{\emph{(b) Bulk zero, $|x_j-\widetilde x_j|$ against $d^L(\Theta_q)$ (without) and $d^L(\Theta^{[2]}_q)$ (with), $q=0.4$}}\\ & $3j$, $(2,1)$, $x\approx0.5$ & $6.4\cdot10^{-5}$ & $1.5\cdot10^{-8}$ & $1.0\cdot10^{-6}$ & $3.6\cdot10^{-12}$\\ & $6j$, $(1,0)$, $(\alpha,\beta)=(1,\frac54)$, $x\approx0.3$ & $1.3\cdot10^{-6}$ & $3.7\cdot10^{-11}$ & $2.1\cdot10^{-8}$ & $9.4\cdot10^{-15}$\\ \midrule & & \multicolumn{2}{c}{$N=500$} & \multicolumn{2}{c}{$N=4000$}\\ \cmidrule(lr){3-4}\cmidrule(lr){5-6} \multicolumn{6}{l}{\emph{(c) Airy bracket, $\max_{|t|\le2}|\text{bracket}-\operatorname{Ai}(t)-\text{term}|$; term $=0$ without, $N^{-4/3}\varpi\operatorname{Ai}'(t)$ with}}\\ & $3j$, $(2,1)$, upper turn, $x=0.6$ & $1.9\cdot10^{-4}$ & $3.6\cdot10^{-6}$ & $1.2\cdot10^{-5}$ & $5.5\cdot10^{-8}$\\ & $3j$, $(\frac32,-\frac12)$, lower turn, $x\approx-0.5$ $^\dagger$ & $7.7\cdot10^{-5}$ & $3.9\cdot10^{-6}$ & $5.0\cdot10^{-6}$ & $6.6\cdot10^{-8}$\\ \midrule & & \multicolumn{2}{c}{$N=40$} & \multicolumn{2}{c}{$N=320$}\\ \cmidrule(lr){3-4}\cmidrule(lr){5-6} & $6j$, $(2,1)$, $(1,\frac65,\frac32)$, $\tau=-1$ & $3.2\cdot10^{-3}$ & $2.9\cdot10^{-4}$ & $2.2\cdot10^{-4}$ & $4.9\cdot10^{-6}$\\ & $6j$, $(1,0)$, $(1,2,\frac65)$, $\tau=+1$ & $2.3\cdot10^{-3}$ & $4.1\cdot10^{-4}$ & $1.4\cdot10^{-4}$ & $6.7\cdot10^{-6}$\\ \midrule & & \multicolumn{2}{c}{$N\approx40$} & \multicolumn{2}{c}{$N\approx320$}\\ \cmidrule(lr){3-4}\cmidrule(lr){5-6} \multicolumn{6}{l}{\emph{(d) Uniform law, $\max|\text{bracket}-\operatorname{Ai}-N^{-4/3}B\operatorname{Ai}'|/(\mathsf M/\mathsf E)$; $B=\varpi_\tau$ without, $B_\tau(\rho)$ with}}\\ & $6j$, $(2,1)$, $(1,\frac65,\frac32)$, $\tau=-1$ & $2.2\cdot10^{-3}$ & $2.1\cdot10^{-4}$ & $3.8\cdot10^{-4}$ & $3.3\cdot10^{-6}$\\ & $6j$, $(1,0)$, $(1,2,\frac65)$, $\tau=+1$ & $1.3\cdot10^{-3}$ & $1.8\cdot10^{-4}$ & $1.6\cdot10^{-4}$ & $3.4\cdot10^{-6}$\\ & $6j$, $(\frac12,-\frac12)$, $(1,\frac53,2)$, $\tau=-1$ $^\ddagger$ & $9.9\cdot10^{-4}$ & $1.1\cdot10^{-5}$ & $1.2\cdot10^{-4}$ & $1.9\cdot10^{-7}$\\ & $q$-$6j$, $(2,1)$, $(1,\frac{11}{10},\frac32)$, $\varsigma=0.6$, $\tau=-1$ $^\S$ & $1.1\cdot10^{-3}$ & $2.3\cdot10^{-5}$ & $5.2\cdot10^{-5}$ & $2.2\cdot10^{-8}$\\ & $q$-$6j$, $(3,3)$, $(2,1,\frac{13}{10})$, $\varsigma=0.5$, $\tau=+1$ $^\S$ & $1.1\cdot10^{-3}$ & $2.5\cdot10^{-5}$ & $2.9\cdot10^{-5}$ & $9.7\cdot10^{-9}$\\ \bottomrule \end{tabular} \caption{Errors against exact symbols. (a) Corollaries~\ref{cor:3j-phase-exactness} and~\ref{cor:6j-phase-exactness}: over seven ranks around the stated ratio, natural gauge; a fourfold increase of $N$ lowers the error by about $4^2$ without and $4^3$ with the amplitude factor. (b) Corollaries~\ref{cor:hahn-bulk-zero-quantization} and~\ref{cor:racah-bulk-zero-quantization}: one zero tracked in $N$; rates $N^{-2}$ and $N^{-4}$. (c) Theorems~\ref{thm:hahnAiryConnection} and~\ref{thm:racahAiryConnection}: rates $N^{-4/3}$ and $N^{-2}$. (d) the Airy bracket of \eqref{eq:general-rank-airy-uniform} (for the Racah rows without its $N^{-2}$ amplitude): maximum over the ranks in $\rho\in[\rho_*-\frac14,\rho_*+\frac15]$ (every rank of this window, both sides of the turn), $N=40$ and $320$; observed rates $N^{-1}$ (the ``without'' column, still pre-asymptotic at these $N$ in some rows) and $N^{-2}$. $^\ddagger$~Half-integer $6j$ family, $N\in\frac14+\frac12\mathbb Z$: $N=\frac{159}4$ and $\frac{1281}4$. $^\S$~Quantum $6j$ at curvature scale $\varsigma$ in the rescaled variable $\rho=(f+\frac12)/N$ of Corollary~\ref{cor:q6j-consequences}(vi) (Section~\ref{sec:q6j}): eight equally spaced ranks of the window $[\rho_*-\delta_-,\rho_*+\delta_+]$ with $\delta_-\le0.10$ and $\delta_+=\min(0.08,\,0.2\,d_{\rm edge})$, $d_{\rm edge}$ the distance of $\rho_*$ from the hard edge (so the window stays away from the edge collar), $N=35$ and $315$; the column with $B_\tau$ includes the amplitude $1+N^{-2}(\delta_{\rm fl}-\frac12\mathfrak j)$ with rescaled flux defect $\delta_{\rm fl}=-\varsigma^2/12$ and decays like $N^{-3}$ (log-log slopes $-3.0$ and $-3.5$ over $N=35,125,315,795$; at $N=795$ the entries are $1.9\cdot10^{-9}$ and $5.1\cdot10^{-10}$); omitting $\delta_{\rm fl}$ leaves slopes $-2.1$ and $-2.2$. The $\tau=+1$ turn of the $(3,3)$ row lies at $d_{\rm edge}\approx0.105$ (at $\varsigma=0.5$ every $\tau=+1$ turn lies within about $0.13$ of the hard edge), so its window top is about $3$ ranks from the edge at $N=35$ and about $67$ at $N=795$; the small-$N$ entries of that row are pre-asymptotic. For $3j$ in half-integer channels ($^\dagger$) $N$ is the nearest half-integer above the stated value.} \label{tab:accuracy} \end{table} \begin{figure}[ht] \centering \includegraphics[width=\textwidth]{figures/horizon_error.pdf} \caption{Natural-gauge error $E_L=\sqrt{2L+1}\,|\Phi_L-A_L|$ of an approximation $A_L$ to the $3j$ chain with $(M,\sigma)=(2,1)$ and $x=\cos\theta=0.3$, at every rank $3\le L\le0.65N$. Since $\Phi_L$ oscillates, $E_L$ passes near zero; panels (a) and (b) plot its maximum over logarithmic rank bins (eight per decade). \textbf{(a)} $N=800$. The vertical line marks the rank at which the phase shift $\kappa^3\cot\theta/(6N^2)$, $\kappa=L+\frac12$, of \eqref{eq:criticalphase-quant} reaches one radian; there the fixed-rank approximation $d^L(\theta)$ has an order-one error. The three pullbacks stay accurate up to fixed ratio. \textbf{(b)} The error of $d^L(\theta)$ for $N=100,\dots,1600$ against $s=L/N^{2/3}$: the curves coincide, grow like $s^3$ (dotted) and level off at order one. \textbf{(c)} The maximum of $E_L$ over all ranks is of order one for $d^L(\theta)$ and decays like $N^{-1}$, $N^{-2}$ and $N^{-3}$ for the leading, second-order and phase-exact pullbacks.} \label{fig:horizon} \end{figure} For the quantum $6j$ symbols of Section~\ref{sec:q6j}, consider three families with curvature scale $\varsigma\in\{0.6,0.7,0.5\}$, $(m,n)=(2,1),(1,0),(3,-1)$, hyperbolic triangles $2\varsigma(1,1.1,1.5)$, $2\varsigma(1.5,0.8,1.1)$, $2\varsigma(1.2,1,1.7)$ and $N=25$--$255$. Evaluated in $40$-digit arithmetic from the closed seed and the Jacobi rows, the natural-gauge errors against the exact symbol scale as $N^{-1}$ (leading pullback), $N^{-2}$ (second order) and $N^{-3}$ (phase-exact with \eqref{eq:q6j-c}); dropping $\eta$ restores an $N^{-1}$ error, and replacing $-\frac1{12}$ by $0$ restores an $N^{-2}$ error. The Airy bracket $\operatorname{Ai}+N^{-4/3}\varpi_\tau\operatorname{Ai}'$ has an $O(N^{-2})$ error for both turn types at $(m,n)=(2,1)$, $\tau=-1$ and $(3,3)$, $\tau=+1$, $N\le415$, while omitting the $\eta$-term of $\varpi_\tau$ leaves an $N^{-4/3}$ residual. The $O(\varsigma^2)$ rate of the flat limit of the rescaled class data, of the turn, and of $\varsigma^{4/3}\varpi_\tau$ (Corollary~\ref{cor:q6j-consequences}(vi)) is also observed. Exact computer algebra reproduces the identities used in the proofs: the closed form \eqref{eq:general-rho20} and the contact identity of Proposition~\ref{prop:general-fixed-rank-contact}, symbolically in $(a,b,k)$; the value \eqref{eq:phase-exactness-J20} and the identity $c=-\mathsf J_2(0,\cdot)$ for the Hahn and Racah chains; the local formula \eqref{eq:general-airy-varpi-local} from the formal Airy parametrix, its Olver form \eqref{eq:general-airy-varpi-olver}, and its Hahn and Racah reductions \eqref{eq:hahn-varpi} and \eqref{eq:racah-varpi}; and the secular cancellation behind \eqref{eq:P4}. \section{Discussion} \label{sec:discussion} Weyl-midpoint centering places the fixed-offset $3j$, one-small $6j$, and one-small quantum $6j$ rank recurrences in the single class of even centered Wigner--Jacobi chains. Across this class, the $k^2/N^2$ phase defect sets the generic $N^{2/3}$ coherence horizon (where $\chi\ne0$), the Hamilton--Jacobi action resums the secular series, Darboux renormalization of the second Weyl eikonal produces an all-rank $O(N^{-2})$ pullback that $h$-parity makes phase exact to $O(N^{-3})$ via a real discrete-flux amplitude, bulk zeros follow to $O(N^{-4})$, and simple rank turns admit uniform second-order Airy asymptotics to $O(N^{-3}\mathsf M/\mathsf E)$ with explicit Olver coefficients on a fixed turning neighbourhood $I'$ under the forbidden-side bound (R), and, for all three families, with intrinsic forbidden-side decay extending the law to the end of $I$ up to a logarithmic collar. A separate technical supplement~\cite{JointUniformitySupplement} addresses joint rank and hard-edge uniformity. It combines the physical-lattice edge factors of Ref.~\cite{LewisFixedRank} with the present Hamilton--Jacobi construction and proves, for the fixed-offset Hahn/$3j$ chain, a joint error bound $O\bigl([N(1-|x|)]^{-2}\bigr)$ in the natural gauge weighted by $D(q_L,x)^{1/4}$ for every physical position and all ranks below a fixed turning ratio. It also gives algebraic, fixed-rank and special-corner results for the centered Racah/$6j$ chain; the analogous global growing-rank Racah estimate remains conjectural. The supplement uses the results of Ref.~\cite{LewisFixedRank} and of the present paper, but no theorem in the present paper depends on the supplement. Several natural extensions remain open; none is used in the proofs above: \begin{itemize}[nosep] \item \emph{Boundary collisions.} Uniform two-column asymptotics where the Racah or $q$-$6j$ rank turn meets the hard edge ($\tau=0$, limiting triangles right-angled opposite $\alpha$ or $\beta$, resp.\ $\mathsf X$ or $\mathsf Y$, where, for Racah, $\varpi_\tau$ diverges like $|\varepsilon|^{-8/3}$) and at the degenerate corner $A-B=O(1)$, $C=O(B^{1/2})$. \item \emph{Dual Racah and Regge remainders.} Subprincipal two-direction transport for the dual $c$-lattice problem through the moving caustic, and an invariant tetrahedral formula for $\rho_2^{(6)}$ and higher Ponzano--Regge remainders. \item \emph{Operator-level pullbacks and kernels.} Egorov or Fourier-integral operator formulations of the pullback and Christoffel--Darboux kernel asymptotics. \item \emph{Global order $N^{-4}$.} In the proof of Theorem~\ref{thm:second-weyl-darboux-engine} a fixed-rank residual $O(h^{2m+2}\kappa^{3m+3})$, from contact through order $t^m$, sums to $O(h^{p-1})$ below a growing split rank $K_*$ ($1\ll K_*\ll h^{-2/3}$, needed so the high-rank tail $K_*^{-(R-1)}$ is $O(h^{p-1})$) only when $p\le2m+2$. Second-order contact ($m=1$) therefore stops at $p=4$, the $O(N^{-3})$ of Theorem~\ref{thm:general-phase-exactness}; $O(N^{-4})$ would need a third-order pullback with fixed-rank contact through order $t^2$, a fixed-ratio residual $O(h^5)$ and an $O(h^4)$ start. \item \emph{Proportional-index Hahn and general quantum $6j$.} Connection formulas when minimal rank and caustic move at order $N$; a direct semiclassical Gauss-weight proof of $\delta_{\rm fl}=0$ (Hahn/Racah) and $\delta_{\rm fl}=-\frac1{12}$ ($q$-$6j$, Remark~\ref{rem:flux-gauss}); and root-of-unity / Turaev--Viro regimes~\cite{MizoguchiTada,BonzomDupuisGirelli}. \end{itemize} The results bear on several applications. In full-sky cosmic microwave background bispectrum and lensing estimators, the two-large/one-small $3j$ symbol couples large multipoles $\ell\asymp N$ to a third multipole $L$, where the Regge-symmetric Wigner-$d$ approximation of Carron and Lewis~\cite{CarronLewis} is used. Strongly squeezed configurations ($L$ fixed) are covered by the fixed-rank theory~\cite{LewisFixedRank}; mildly squeezed configurations ($L$ in the hundreds for $\ell$ in the thousands) cross the $L\asymp N^{2/3}$ phase horizon, where Corollary~\ref{cor:3j-phase-exactness} and Theorem~\ref{thm:hahnAiryConnection} replace the fixed-$L$ formula to $O(N^{-3})$ on compact interior sets of $x$ and $O(N^{-2})$ in the band-edge Airy bracket. Likewise, growing-rank one-small $6j$ and $q$-$6j$ pullbacks bridge the Edmonds tensor-operator regime~\cite{Edmonds1957} and all-large spin-network asymptotics~\cite{PonzanoRegge1968,Roberts1999,TaylorWoodward}, evaluating a symbol at a given rank without stepping through all intermediate ranks~\cite{SchultenGordon1975a,JohanssonForssen2016}. \section*{Reproducibility} A companion GitHub repository, \url{https://github.com/cmbant/weyl-midpoint-semiclassics}, contains the technical supplement~\cite{JointUniformitySupplement}, the verification programs behind Section~\ref{sec:numerics}, their captured outputs, and the exact symbolic certificates used in the proofs (including the $q$-Zeilberger certificate of Lemma~\ref{lem:q6j-recurrence} and the rational-function identity of Corollary~\ref{cor:q6j-consequences}(iii)). Symbolic checks certify the specified algebraic identities; finite numerical comparisons do not by themselves prove the uniform asymptotic bounds. \section*{Use of AI} All results, documents and scripts were generated by AI agents (Claude Opus and Sonnet 5.5, OpenAI's GPT 6.1 sol and Gemini 4 Hydrogen/Argon/Carbon) with human supervision and many cross-checks, including exact symbolic and high-precision numerical regression checks of the stated formulas. I thank DeepMind and Anthropic for access to their scientist programs. \section*{Acknowledgements} We acknowledge support by the UK STFC grant ST/T000473/1. \appendix \input{weyl_midpoint_semiclassics_appendices} \begin{thebibliography}{99} \footnotesize\setlength{\itemsep}{1pt}\setlength{\parsep}{0pt} \bibitem{JointUniformitySupplement} A.~Lewis, \emph{Joint Rank--Position Uniformity for Weyl--Midpoint Wigner $3j$ Symbols: A Technical Supplement with Racah $6j$ Extensions}, companion technical supplement to the present paper, with its verification programs in the companion GitHub repository \url{https://github.com/cmbant/weyl-midpoint-semiclassics} (2026). \bibitem{LewisFixedRank} A.~Lewis, \emph{Even Asymptotic Expansions and Uniform Hard-Edge Approximations for Wigner $3j$ and $6j$ Symbols at Fixed Rank}, Hexagon 2611.00533v1 (2026); \url{https://hexagonmath.org/2611.00533v1}. \bibitem{CarronLewis} J.~Carron and A.~Lewis, \emph{Spherical bispectrum expansion and quadratic estimators}, JCAP \textbf{07} (2024) 067. \url{https://doi.org/10.1088/1475-7516/2024/07/067}. \bibitem{McFarlane1992} S.~C.~McFarlane, \emph{Angular momentum and Heisenberg's correspondence principle}, J. Phys. B \textbf{25} (1992), 4045--4057. \bibitem{McFarlane1993} S.~C.~McFarlane, \emph{Angular momentum and Heisenberg's correspondence principle. II. More complex matrix elements}, J. Phys. B \textbf{26} (1993), 1871--1884. \bibitem{Morehead1996} J.~J.~Morehead, \emph{Semiclassical integrable matrix elements}, Phys. Rev. A \textbf{53} (1996), 1285--1294. \bibitem{Neville1971} D.~Neville, \emph{A technique for solving recurrence relations approximately and its application to the 3-J and 6-J symbols}, J. Math. Phys. \textbf{12} (1971), 2438--2453. \bibitem{Edmonds1957} A.~R.~Edmonds, \emph{Angular Momentum in Quantum Mechanics}, Princeton University Press (1957). \bibitem{PonzanoRegge1968} G.~Ponzano and T.~Regge, \emph{Semiclassical limit of Racah coefficients}, in \emph{Spectroscopic and Group Theoretical Methods in Physics}, ed. F.~Bloch, North-Holland, Amsterdam (1968). \bibitem{BonzomFleury2012} V.~Bonzom and P.~Fleury, \emph{Asymptotics of Wigner $3nj$-symbols with small and large angular momenta: an elementary method}, J. Phys. A: Math. Theor. \textbf{45} (2012) 075202; arXiv:1108.1569. \bibitem{Braun1993} P.~A.~Braun, \emph{Discrete semiclassical methods in the theory of Rydberg atoms in external fields}, Rev. Mod. Phys. \textbf{65} (1993) 115--161. \url{https://doi.org/10.1103/RevModPhys.65.115}. \bibitem{SchultenGordon1975a} K.~Schulten and R.~G.~Gordon, \emph{Exact recursive evaluation of $3j$- and $6j$-coefficients for quantum-mechanical coupling of angular momenta}, J. Math. Phys. \textbf{16} (1975), 1961--1970. \bibitem{JohanssonForssen2016} H.~T.~Johansson and C.~Forss\'en, \emph{Fast and accurate evaluation of Wigner $3j$, $6j$, and $9j$ symbols using prime factorization and multiword integer arithmetic}, SIAM J. Sci. Comput. \textbf{38} (2016) A376--A384. \url{https://doi.org/10.1137/15M1021908}. \bibitem{BenettinGiorgilli} G.~Benettin and A.~Giorgilli, \emph{On the Hamiltonian interpolation of near-to-the-identity symplectic mappings with application to symplectic integration algorithms}, J. Stat. Phys. \textbf{74} (1994), 1117--1143. \url{https://doi.org/10.1007/BF02188219}. \bibitem{ReinschMorehead1999} M.~W.~Reinsch and J.~J.~Morehead, \emph{Asymptotics of Clebsch--Gordan coefficients}, J. Math. Phys. \textbf{40} (1999), 4782--4806. \bibitem{DupuisLivine2009} M.~Dupuis and E.~R.~Livine, \emph{Pushing further the asymptotics of the $6j$-symbol}, Phys. Rev. D \textbf{80} (2009), 024035. \bibitem{AquilantiHaggardLittlejohnYu} V.~Aquilanti, H.~M.~Haggard, R.~G.~Littlejohn, and L.~Yu, \emph{Semiclassical analysis of Wigner $3j$-symbol}, J. Phys. A: Math. Theor. \textbf{40} (2007), 5637--5674. \url{https://doi.org/10.1088/1751-8113/40/21/013}. \bibitem{LittlejohnYu} R.~G.~Littlejohn and L.~Yu, \emph{Uniform semiclassical approximation for the Wigner $6j$-symbol in terms of rotation matrices}, J. Phys. Chem. A \textbf{113} (2009), 14904--14922. \url{https://doi.org/10.1021/jp905056y}. \bibitem{Yu2011} L.~Yu, \emph{Uniform approximation from symbol calculus on a spherical phase space}, J. Phys. A: Math. Theor. \textbf{44} (2011), 505202; arXiv:1105.4220. \url{https://doi.org/10.1088/1751-8113/44/50/505202}. \bibitem{BaikKriecherbauerMcLaughlinMiller} J.~Baik, T.~Kriecherbauer, K.~T.-R.~McLaughlin, and P.~D.~Miller, \emph{Discrete Orthogonal Polynomials: Asymptotics and Applications}, Annals of Mathematics Studies \textbf{164}, Princeton University Press, 2007. \bibitem{RakhmanovDiscrete} E.~A.~Rakhmanov, \emph{Equilibrium measure and the distribution of zeros of the extremal polynomials of a discrete variable}, Sb. Math. \textbf{187} (1996), 1213--1228. \url{https://doi.org/10.1070/SM1996v187n08ABEH000153}. \bibitem{DragnevSaff} P.~D.~Dragnev and E.~B.~Saff, \emph{Constrained energy problems with applications to orthogonal polynomials of a discrete variable}, J. Anal. Math. \textbf{72} (1997), 223--259. \url{https://doi.org/10.1007/BF02843160}. \bibitem{DuistermaatGrunbaum} J.~J.~Duistermaat and F.~A.~Gr\"unbaum, \emph{Differential equations in the spectral parameter}, Comm. Math. Phys. \textbf{103} (1986), 177--240. \bibitem{NikiforovSuslovUvarov} A.~F.~Nikiforov, S.~K.~Suslov, and V.~B.~Uvarov, \emph{Classical Orthogonal Polynomials of a Discrete Variable}, Springer Series in Computational Physics, Springer, 1991. \url{https://doi.org/10.1007/978-3-642-74748-9}. \bibitem{DLMF34} NIST Digital Library of Mathematical Functions, \emph{Chapter 34: $3j$, $6j$, $9j$ Symbols}, especially Secs.~34.3--34.5. \url{https://dlmf.nist.gov/34}. \bibitem{DLMF18} NIST Digital Library of Mathematical Functions, \emph{Chapter 18: Orthogonal Polynomials}, especially Secs.~18.15, 18.24, and 18.38. \url{https://dlmf.nist.gov/18}. \bibitem{Flude} J.~P.~M.~Flude, \emph{The Edmonds asymptotic formulae for the $3j$ and $6j$ symbols}, J. Math. Phys. \textbf{39} (1998), 3906--3915; arXiv:physics/9710037. \bibitem{KoornwinderHahn} T.~H.~Koornwinder, \emph{Clebsch--Gordan coefficients for $SU(2)$ and Hahn polynomials}, Nieuw Arch. Wisk. (3) \textbf{29} (1981), 140--155. \bibitem{LinWong} Y.~Lin and R.~Wong, \emph{Global asymptotics of the Hahn polynomials}, Analysis and Applications \textbf{11} (2013), 1350018. \url{https://doi.org/10.1142/S0219530513500188}. \bibitem{ChenIsmailSimeonov} L.-C.~Chen, M.~E.~H.~Ismail, and P.~Simeonov, \emph{Asymptotics of Racah coefficients and polynomials}, J. Phys. A: Math. Gen. \textbf{32} (1999), 537--553. \url{https://doi.org/10.1088/0305-4470/32/3/009}. \bibitem{TaylorWoodward} Y.~U.~Taylor and C.~T.~Woodward, \emph{$6j$ symbols for $U_q(\mathfrak{sl}_2)$ and non-Euclidean tetrahedra}, Selecta Math. (N.S.) \textbf{11} (2006), 539--571. \url{https://doi.org/10.1007/s00029-005-0014-9}. \bibitem{Costantino2007} F.~Costantino, \emph{$6j$ symbols, hyperbolic structures and the volume conjecture}, Geom. Topol. \textbf{11} (2007), 1831--1854. \url{https://doi.org/10.2140/gt.2007.11.1831}. \bibitem{ChenYang2018} Q.~Chen and T.~Yang, \emph{Volume conjectures for the {R}eshetikhin--{T}uraev and the {T}uraev--{V}iro invariants}, Quantum Topol. \textbf{9} (2018), 419--460. \url{https://doi.org/10.4171/QT/111}. \bibitem{MizoguchiTada} S.~Mizoguchi and T.~Tada, \emph{Three-dimensional gravity from the Turaev--Viro invariant}, Phys. Rev. Lett. \textbf{68} (1992), 1795--1798. \url{https://doi.org/10.1103/PhysRevLett.68.1795}. \bibitem{BonzomDupuisGirelli} V.~Bonzom, M.~Dupuis, and F.~Girelli, \emph{Towards the Turaev--Viro amplitudes from a Hamiltonian constraint}, Phys. Rev. D \textbf{90} (2014), 104038. \url{https://doi.org/10.1103/PhysRevD.90.104038}. \bibitem{KirillovReshetikhin1989} A.~N.~Kirillov and N.~Yu.~Reshetikhin, \emph{Representations of the algebra $U_q(\mathfrak{sl}(2))$, $q$-orthogonal polynomials and invariants of links}, in \emph{Infinite-dimensional Lie algebras and groups (Luminy--Marseille, 1988)}, Adv. Ser. Math. Phys. \textbf{7}, World Scientific, Teaneck, NJ, 1989, 285--339. \bibitem{AskeyWilson1979} R.~Askey and J.~Wilson, \emph{A set of orthogonal polynomials that generalize the Racah coefficients or $6$-$j$ symbols}, SIAM J. Math. Anal. \textbf{10} (1979), 1008--1016. \url{https://doi.org/10.1137/0510092}. \bibitem{KoekoekLeskySwarttouw} R.~Koekoek, P.~A.~Lesky, and R.~F.~Swarttouw, \emph{Hypergeometric Orthogonal Polynomials and Their $q$-Analogues}, Springer Monographs in Mathematics, Springer, Berlin, 2010. \url{https://doi.org/10.1007/978-3-642-05014-5}. \bibitem{KhavkineQ6j} I.~Khavkine, \emph{Recurrence relation for the $6j$-symbol of $su_q(2)$ as a symmetric eigenvalue problem}, Int. J. Geom. Methods Mod. Phys. \textbf{12} (2015), 1550117. \url{https://doi.org/10.1142/S0219887815501170}. \bibitem{Roberts1999} J.~Roberts, \emph{Classical $6j$-symbols and the tetrahedron}, Geometry \& Topology \textbf{3} (1999), 21--66. \url{https://doi.org/10.2140/gt.1999.3.21}. \bibitem{SchultenGordon} K.~Schulten and R.~G.~Gordon, \emph{Semiclassical approximations to $3j$- and $6j$-coefficients for quantum-mechanical coupling of angular momenta}, J. Math. Phys. \textbf{16} (1975), 1971--1988. \bibitem{GeronimoBrunoVanAssche} J.~S.~Geronimo, O.~Bruno, and W.~Van Assche, \emph{WKB and turning point theory for second-order difference equations}, in \emph{Spectral Methods for Operators of Mathematical Physics}, Operator Theory: Advances and Applications \textbf{154} (2004), 101--138. \url{https://doi.org/10.1007/978-3-0348-7947-7_7}. \bibitem{GeronimoSmithWKB} J.~S.~Geronimo and D.~T.~Smith, \emph{WKB (Liouville--Green) analysis of second order difference equations and applications}, J. Approx. Theory \textbf{69} (1992), 269--301. \url{https://doi.org/10.1016/0021-9045(92)90003-7}. \bibitem{WangWongRacah} X.-S.~Wang and R.~Wong, \emph{Asymptotics of Racah polynomials with varying parameters}, J. Math. Anal. Appl. \textbf{436} (2016), 1149--1164. \url{https://doi.org/10.1016/j.jmaa.2015.12.035}. \bibitem{BenzaidLutz} Z.~Benzaid and D.~A.~Lutz, \emph{Asymptotic representation of solutions of perturbed systems of linear difference equations}, Stud. Appl. Math. \textbf{77} (1987), 195--221. \url{https://doi.org/10.1002/sapm1987773195}. \bibitem{OlverASF} F.~W.~J.~Olver, \emph{Asymptotics and Special Functions}, Academic Press, New York, 1974; reprinted A K Peters, Wellesley, MA, 1997. \bibitem{WangWong} Z.~Wang and R.~Wong, \emph{Asymptotic expansions for second-order linear difference equations with a turning point}, Numer. Math. \textbf{94} (2003), 147--194. \url{https://doi.org/10.1007/s00211-002-0416-y}. \bibitem{Geronimo2009} J.~S.~Geronimo, \emph{WKB and turning point theory for second order difference equations: external fields and strong asymptotics for orthogonal polynomials}, arXiv:0905.1684 (2009). \bibitem{FrenzenWong} C.~L.~Frenzen and R.~Wong, \emph{A uniform asymptotic expansion of the Jacobi polynomials with error bounds}, Canad. J. Math. \textbf{37} (1985), 979--1007. \url{https://doi.org/10.4153/CJM-1985-053-1}. \bibitem{DLMFAiryBessel} NIST Digital Library of Mathematical Functions, \emph{Chapters 9 and 10: Airy and Related Functions; Bessel Functions}, especially Secs.~9.7--9.8 and 10.17. \url{https://dlmf.nist.gov/9}, \url{https://dlmf.nist.gov/10}. \bibitem{FedotovKlopp} A.~Fedotov and F.~Klopp, \emph{The complex WKB method for difference equations and Airy functions}, SIAM J. Math. Anal. \textbf{51} (2019), 4413--4447. \url{https://doi.org/10.1137/18M1228694}. \bibitem{KuijlaarsVanAssche} A.~B.~J.~Kuijlaars and W.~Van Assche, \emph{The asymptotic zero distribution of orthogonal polynomials with varying recurrence coefficients}, J. Approx. Theory \textbf{99} (1999), 167--197. \url{https://doi.org/10.1006/jath.1999.3316}. \end{thebibliography} \end{document}