\documentclass[11pt,a4paper]{article} \usepackage[T1]{fontenc} \usepackage{lmodern} \usepackage[margin=30mm]{geometry} \usepackage{graphicx} \usepackage{amsmath,amssymb,amsthm} \usepackage[hidelinks]{hyperref} \input{macros} \hypersetup{ pdftitle={Sup-norm stability radii for a nodal face of the B3 spherical discriminant}, pdfauthor={Oleksiy Babanskyy}, pdfsubject={Nodal graph stability in the full real spherical polynomial band} } \setlength{\emergencystretch}{1em} \title{Sup-norm stability radii for a nodal face\\of the $B_3$ spherical discriminant} \author{Oleksiy Babanskyy} \date{} \begin{document} \maketitle \begin{abstract} How large can a real perturbation be before a nodal line ceases to be a single graph in a fixed spherical sector? We study a truncated face of the degree-nine $B_3$ discriminant, allowing all 100 real spherical polynomial modes of degree at most nine and measuring the supremum norm on the entire closed tube. The failure radius equals the minimum cost of cancelling the field and its normal derivative. We bound it on the full parameter domain and determine it exactly in two sufficient regimes, using an explicit atomic mixed-jet certificate on the sphere. In these exact regimes the unique critical field is a scaled Chebyshev polynomial, and the whole closed radius ball still preserves the graph, although the critical zero level is singular; every larger amplification of that field fails. Two continuous examples separate the limits of a support criterion from those of the radius formula. Below the radius we obtain quantitative graph bounds. In the exact regimes a smooth family approaching the critical field from inside the ball has uniformly convergent graphs and terminal slopes diverging as $\varepsilon^{-1/3}$ at the minimizing terminals. Persistence and $C^1$ conditioning therefore separate. \end{abstract} \section{Introduction}\label{sec:introduction} The zero set of a signed field describes a geometric pattern. Its stability depends on the permitted errors, the norm used to measure them and the property one wants to preserve. Here the property is particularly concrete: within a prescribed spherical tube, the entire zero set must remain one interior $C^1$ graph over the angular interval. The perturbation is measured only by its values, but belongs to a fixed finite-dimensional band. This restriction allows a positive stability radius; arbitrary smooth errors with small values need not preserve the graph. Our target is the product of the nine unit-normal mirror forms of $B_3$, restricted to the unit sphere. We truncate one of its nodal faces before its mirror intersections and work in a closed tube with two normal sides and two angular terminal fibres. Perturbations use all real spherical modes of degree at most nine, without symmetry constraints. A smooth graph is permitted even when the field defining its zero level has a vanishing gradient. This distinction is decisive at the critical radius. The analysis has three parts. First, the graph-failure radius reduces to a variational cost for a value and an oriented normal derivative. For this target the normal-side escape cost is strictly larger, and the same minimum is obtained by setting both jet components to zero. Second, a moving atomic formula on the sphere gives a sharp lower bound in two geometric regimes, identifies every minimizing point and the unique minimizing field, and proves safety of the entire closed ball. Third, the margin below this radius controls graph position and slope, but an explicit interior approach to the critical field yields unbounded terminal slope despite uniform convergence of the graphs. These are geometric assertions about signed fields; no plate response, particle transport or material deposition is assumed or certified. \subsection{Relation to approximation and discriminant distances} Atomic representations of functionals on finite-dimensional subspaces of $C(K)$ and complementary sign saturation are classical; a useful account is \cite[Section 4.2]{Dressler}. That work fixes the sum of the highest-degree coefficients and constructs explicit alternating Chebyshev atoms on diagonally determined sets \cite[Theorem 4.8]{Dressler}. For prescribed nodes, \cite[Theorem 9]{DavydovSchaback} relates exact formulas with $\ell^1$ weights to dual $\ell^\infty$ bounds on sampled data. These principles motivate the certificate here. Its specific ingredients are a moving spherical support for a mixed value--normal-derivative functional, a rational interpolation cancellation on the complete band, and saturation in the continuous tube norm. A single-functional formula does not imply reconstruction of every field from its atoms, nor is a minimal number of atoms asserted. The complementary fibre bound uses the real trigonometric value--derivative inequality of Erd\'elyi \cite[Theorem 2.1]{Erdelyi}. We state the imported inequality and check its interval and degree hypotheses in Section~\ref{sec:barrier}. The remaining finite-dimensional separation, support and rigidity arguments are proved in the article. Distances to singularity discriminants provide a related viewpoint with different norms and events. Raffalli computes the Bombieri distance for homogeneous forms on the whole sphere and gives a closest contact perturbation at a minimizing point \cite[Theorems 5.1 and 5.4]{Raffalli}. Kozhasov, Mourrain and Parusi\'nski express higher discriminant distances through Taylor jets in the Bombieri--Weyl geometry \cite[Proposition 3.2, equations (3.8)--(3.9)]{KozhasovMourrainParusinski}; their Theorem 5.3 identifies a Chebyshev extremizer among real-rooted binary forms with fixed norm. Our space contains every degree from zero through nine and uses a supremum on a truncated tube. The global $B_3$ target already has singular mirror intersections. Consequently, the whole-sphere discriminant distance does not supply the positive local radius studied here. The ordinary singularity event uses the value and full tangential gradient; our jet event uses the value and one normal derivative, with escape through the normal sides checked separately. Lerario and Stecconi obtain isotopy under a strict distance from nontransversality in a differentiable norm and describe the scalar $C^1$ distance on a closed disk using the value together with the full gradient in the interior and its projection tangent to the boundary on the boundary \cite[Lemma 20 and Proposition 31]{LerarioStecconi}. Their boundary transversality and isotopy of pairs differ from the fixed-fibre graph property on our tube with angular terminals and corners. In the two exact regimes, the graph at our sharp threshold remains smooth even though the defining level is singular. The comparison identifies the role of the band, norm, domain and preserved property; it does not transfer a conclusion between these settings. \subsection{Organization} Section~\ref{sec:problem} fixes the problem and states the main results. Sections~\ref{sec:variational} and \ref{sec:dualandbounds} reduce graph failure to a jet cost and give general bounds. Section~\ref{sec:atomic} constructs the certificate; Section~\ref{sec:exact} proves rigidity and the exact regimes. Section~\ref{sec:examples} gives the two delimiting examples. Sections~\ref{sec:quantitative} and \ref{sec:limits} treat conditioning and the scope boundaries. Section~\ref{sec:discussion} summarizes the conclusions and the autonomous computational companion. Appendix~\ref{app:harmonics} justifies the full spherical-band description. \section{Setting and main results}\label{sec:problem} \subsection{Target, coordinates, and the closed tube} Let $\Sph=\{(x,y,z)\in\R^3:x^2+y^2+z^2=1\}$. On this unit sphere, the target is the product of nine unit-normal linear forms, with multiplicative factor one: \begin{equation}\label{eq:target} f(x,y,z)=\frac{xyz(x^2-y^2)(x^2-z^2)(y^2-z^2)}{8}. \end{equation} This convention does not set the supremum of $f$ to one. The factor $1/8$ is part of the normalization: this is the product of the nine unit-normal linear forms \[ x,\ y,\ z,\quad \frac{x-y}{\sqrt2},\ \frac{x+y}{\sqrt2},\quad \frac{x-z}{\sqrt2},\ \frac{x+z}{\sqrt2},\quad \frac{y-z}{\sqrt2},\ \frac{y+z}{\sqrt2}. \] We consider the open nodal face $z=0$, $x>y>0$ and then truncate it. Fix \begin{equation}\label{eq:domain} 0<\eta<\omega<\frac{\pi}{4},\qquad I=[\eta,\omega],\qquad \dd=\arctan\frac{\sin\eta\cos\eta}{4},\qquad \aeta=\sin\dd. \end{equation} Dependence on the fixed upper endpoint $\omega$ is suppressed in the notation. Define \begin{equation}\label{eq:chart} X(\theta,r)=(\cos r\cos\theta,\cos r\sin\theta,\sin r),\qquad D_\eta=I\times[-\dd,\dd],\qquad T_\eta=X(D_\eta). \end{equation} Here $r$ is signed geodesic distance from the equator and increases toward positive $z$. Since $\dd<\pi/2$ and $0<\theta<\pi/4$, the chart is injective: its inverse on the tube is $r=\arcsin z$, $\theta=\arctan(y/x)$. Its metric is $dr^2+\cos^2r\,d\theta^2$. We distinguish the \emph{normal sides} $r=\pm\dd$ from the two \emph{angular terminal fibres} $\theta=\eta,\omega$. All four are included in the tube. \subsection{Perturbation class and graph property} We use the full real space \begin{equation}\label{eq:space} \V=\{p|_{\Sph}:p\in\R[x,y,z],\ \deg p\le9\} =\bigoplus_{\ell=0}^{9}\cH_\ell(\Sph;\R), \qquad \dim\V=100. \end{equation} Here $\cH_\ell$ consists of restrictions of homogeneous harmonic polynomials of degree $\ell$. The equality and dimension in \eqref{eq:space} are justified in Section~\ref{sec:finite} and Appendix~\ref{app:harmonics}. There is no symmetry constraint on the perturbation, and it need not lie in the degree-nine eigenspace alone. For $e\in\V$, write \[ F=f\circ X,\qquad E=e\circ X,\qquad G_e=F+E,\qquad \normT e=\max_{p\in T_\eta}|e(p)|. \] The maximum is over the \emph{entire closed tube}; it is not a maximum over sampled points or over a single fibre. This quantity is a norm on $\V$ by Lemma~\ref{lem:norm}. \begin{definition}[The property to be preserved]\label{def:graph} The property $\PP(e)$ holds if there is a function $h\in C^1(I)$, with $-\dd1$. \end{theorem} The general conclusions follow from Propositions~\ref{prop:exactjet}, \ref{prop:jetonly} and \ref{prop:strongfloor}. The exact regimes and threshold assertions are proved in Section~\ref{sec:sharp}. The two hypotheses are sufficient criteria for the radius formula; neither is asserted to describe the maximal possible geometry. Example~\ref{ex:M} satisfies M while violating the uniform support condition L. Example~\ref{ex:F} has $\rho_\eta0, \qquad \|h\|_\infty\le\frac{2m}{\beta}. \] It also bounds the slope using a finite derivative operator norm on the fixed band and tube. In L or M, $d_\eta=A_*$ is explicit. Theorem~\ref{thm:degeneration} shows a different phenomenon at equality: fields $e_\varepsilon$ strictly inside the ball converge smoothly to $e_C$, their graphs converge uniformly to zero, yet $|h_\varepsilon'(\theta_*)|$ diverges like $\varepsilon^{-1/3}$ at each minimizing terminal. The critical graph itself is smooth. \section{Variational characterization of graph failure}\label{sec:variational} \subsection{Geometry and signs on the whole tube}\label{sec:signs} We first establish the target estimates, including the entire boundary of the tube. Set \[ c=\cos\theta,\quad s=\sin\theta,\quad v=c^2s^2,\quad C(\theta)=\frac{cs(c^2-s^2)}8>0,\quad t=\tan r,\quad q=t^2. \] Substitution into \eqref{eq:target}, using $c^2+s^2=1$, gives \begin{equation}\label{eq:F} F(\theta,r)=C(\theta)\frac{t(v-q+q^2)}{(1+q)^{9/2}}. \end{equation} In particular, $F$ is odd in $r$ and $F(\theta,0)=0$. Define \begin{equation}\label{eq:B} B(\theta)=\partial_rF(\theta,0) =C(\theta)v =\frac{\cos^3\theta\sin^3\theta\cos2\theta}{8}>0. \end{equation} \begin{lemma}[Positive normal derivative]\label{lem:monotone} On the entire closed rectangle $D_\eta$, \begin{equation}\label{eq:monotone} \partial_rF(\theta,r)>\frac{B(\theta)}2>0. \end{equation} Consequently $j_\eta>0$, the unperturbed zero set in the tube is exactly $r=0$, and $F(\theta,-\dd)=-F(\theta,\dd)$. \end{lemma} \begin{proof} Differentiating \eqref{eq:F} with $dt/dr=1+t^2$, without division by $t$, yields \begin{equation}\label{eq:Q} \partial_rF=C(\theta)\frac{Q(v,q)}{(1+q)^{9/2}},\qquad Q(v,q)=v-(3+8v)q+11q^2-4q^3. \end{equation} Since $v=\tfrac14\sin^2(2\theta)$ increases on $(0,\pi/4)$, \[ 0\le q\le\tan^2\dd=\frac{\sin^2\eta\cos^2\eta}{16} \le\frac v{16}\le\frac1{64}. \] The inequalities $3+8v\le5$ and $q^2(11-4q)\ge0$ imply $Q(v,q)\ge v-5q\ge11v/16$. Also \[ (1+q)^{9/2}\le(65/64)^{9/2}<(65/64)^5<11/8, \] where the last inequality follows from $8\cdot65^5=9282325000<11811160064=11\cdot64^5$. Combining these bounds proves \eqref{eq:monotone}. Oddness was already obtained from \eqref{eq:F}. Integrating the positive derivative from zero to $\dd$ gives \[ F(\theta,\dd)>\tfrac12\dd B(\theta)>0. \] Continuity on the compact interval $I$ gives a positive minimum in \eqref{eq:j}. Strict increase and the zero at $r=0$ give the remaining claims. \end{proof} \begin{corollary}[Signs below the lateral cost]\label{cor:signs} If $\normT e0$; by oddness of $F$, the left-hand value is at most $-j_\eta+\normT e<0$. No symmetry of $e$ is used. \end{proof} \subsection{The full finite-dimensional space and the minimum of the jet cost} \label{sec:finite} \begin{proposition}[A complete set of 100 coordinates]\label{prop:basis} Each $e\in\V$ has a unique representation on the sphere of the form \begin{equation}\label{eq:basis} e(x,y,z)=A(y,z)+xB(y,z),\qquad \deg A\le9,\quad\deg B\le8. \end{equation} The corresponding 55 monomials of $A$ and 45 monomials of $B$ form a basis of $\V$. \end{proposition} \begin{proof} In any ambient polynomial of degree at most nine, repeatedly replace $x^2$ by $1-y^2-z^2$. This preserves its restriction to the sphere, does not increase degree, and produces \eqref{eq:basis}. For uniqueness, suppose the expression in \eqref{eq:basis} vanishes on the sphere. At every point with $y^2+z^2<1$, evaluate it at the two values $x=\pm\sqrt{1-y^2-z^2}$. Adding and subtracting gives $A(y,z)=B(y,z)=0$ on the open disk. A polynomial vanishing on an open set is identically zero: one may fix one coordinate on a small rectangle and apply the one-variable zero theorem successively to its coefficients. Thus all coefficients vanish. There are $(9+1)(9+2)/2=55$ monomials of degree at most nine in two variables and $(8+1)(8+2)/2=45$ of degree at most eight. \end{proof} \begin{lemma}[The tube norm and compact sublevels]\label{lem:norm} For fixed parameters \eqref{eq:domain}, $\normT\cdot$ is a norm on $\V$, and every set $\{e:\normT e\le M\}$ with $M<\infty$ is compact in coefficient coordinates. \end{lemma} \begin{proof} Only positive definiteness needs proof. The tube contains a nonempty open subset $U$ of the sphere. If $e$ vanishes on $T_\eta$, it vanishes on $U$. Its restriction to a great circle is a trigonometric polynomial of degree at most nine: each ambient coordinate is linear in the sine and cosine of the circle parameter. Such a polynomial that vanishes on an open arc vanishes on the whole circle, since multiplication by $\exp(9it)$ makes it an ordinary polynomial in $\exp(it)$ with infinitely many roots. For any $q\in\Sph$, choose a point of $U$ not parallel to $q$ and the great circle through them. That circle contains an open arc in $U$, so $e(q)=0$. Hence $e$ is the zero function on the sphere. For compactness, let $b_1,\ldots,b_{100}$ be the basis from Proposition~\ref{prop:basis}. Evaluations at points of $T_\eta$ span the dual of $\V$. Otherwise a nonzero field would be annihilated by every such evaluation, contradicting positive definiteness. Select 100 linearly independent evaluations. The matrix $M_{ij}=b_j(p_i)$ is invertible, and $\normT e\le M_0$ bounds every coordinate of $(e(p_i))_{i=1}^{100}$ by $M_0$. It therefore bounds the coefficient vector by application of $M^{-1}$. Finally, \[ \bigl|\normT{e_c}-\normT{e_{c'}}\bigr| \le\normT{e_c-e_{c'}} \le\sum_{j=1}^{100}|c_j-c'_j|\,\normT{b_j}, \] so the sublevel set is closed. The result follows from compactness of closed bounded sets in $\R^{100}$. \end{proof} The evaluation points above are a compactness argument, not a numerical norming mesh or an algorithm for computing the sup. The bound on the inverse matrix may depend on $\eta$ and $\omega$; no uniform bound as the tube degenerates is asserted. \begin{proposition}[Attainment of the oriented-jet cost]\label{prop:d} The infimum in \eqref{eq:d} is attained by some pair $(e_*,(\theta_*,r_*))$, and $d_\eta>0$. \end{proposition} \begin{proof} The feasible set is nonempty: $e=-f\in\V$ makes $G_e$ and $\partial_rG_e$ identically zero. Set $M_0=\normT f$. It suffices to minimize over pairs in \[ \{e\in\V:\normT e\le M_0\}\times D_\eta \] satisfying the two jet conditions. This product is compact by Lemma~\ref{lem:norm}. The conditions $G_e=0$ and $\partial_rG_e\le0$ are closed, because value and derivative depend continuously on coefficients and point. The norm is continuous, so it attains a minimum on this nonempty closed subset. Candidates of norm larger than $M_0$ cannot improve it, as the candidate $-f$ already has norm $M_0$. If the minimum were zero, positive definiteness would force $e_*=0$. But $\partial_rF>0$ everywhere on $D_\eta$, which is incompatible with the derivative constraint. Thus the minimum is strictly positive. \end{proof} \subsection{From oriented fibre zeros to a \texorpdfstring{$C^1$}{C1} graph}\label{sec:graph} Injectivity of $X$ implies that one interior zero on each fibre gives a set-theoretic graph. It also gives a continuous graph in this setting: if $\theta_n\to\theta$, any subsequential limit of the roots lies in $[-\dd,\dd]$, is a root at $\theta$ by continuity, and must equal its unique root. This compactness argument does not supply a derivative. The stronger hypothesis below supplies it, without assuming that $G_e$ is monotone away from its zeros. \begin{lemma}[Fibre-to-graph lemma]\label{lem:graph} Suppose \eqref{eq:signs} holds and \begin{equation}\label{eq:positivezeros} G_e(\theta,r)=0\quad\Longrightarrow\quad \partial_rG_e(\theta,r)>0\qquad ((\theta,r)\in D_\eta). \end{equation} Then $\PP(e)$ holds, including both angular terminal fibres. \end{lemma} \begin{proof} Fix $\theta$. The side signs give an interior zero by the intermediate value theorem. Every zero is simple by \eqref{eq:positivezeros}, hence isolated. The zero set of this fibre is compact and is disjoint from the normal endpoints. It is finite: an infinite set would have an accumulation point that is also a zero, contradicting simplicity there. Every zero crosses from negative to positive as $r$ increases. If two consecutive zeros existed, the function would be positive just after the first and negative just before the second; continuity would give another zero between them. This contradicts consecutiveness. Thus there is exactly one zero, denoted $h(\theta)$. For completeness, the local scalar implicit-function argument can be made directly. At a zero $(\theta_0,r_0)$, continuity of the derivative gives a small open rectangle on which $\partial_rG_e$ is bounded below by a positive number. At $\theta_0$ choose two nearby normal values around $r_0$ with opposite signs. Those signs persist on a smaller open interval of $\theta$ values. The intermediate value theorem and strict monotonicity on that rectangle give a unique local root. The preceding compactness argument gives its continuity. Applying the mean value theorem separately in the two variables to \[ G_e(\theta+t,h(\theta+t))-G_e(\theta,h(\theta))=0 \] and letting $t\to0$ then gives \begin{equation}\label{eq:hprime} h'(\theta)=-\frac{\partial_\theta G_e(\theta,h(\theta))} {\partial_rG_e(\theta,h(\theta))}. \end{equation} The denominator stays positive locally, so this derivative is continuous. Their restrictions to $I$ agree wherever they overlap, by fibrewise uniqueness. The same argument applies at $\theta=\eta$ and $\theta=\omega$: the field is a global polynomial restriction and the chart is defined on an open neighbourhood of the closed rectangle. Apply the local argument there and then restrict to $I$. This supplies the required one-sided endpoint derivatives. The side signs exclude zeros on $r=\pm\dd$, and fibrewise uniqueness accounts for all zeros in the tube. \end{proof} \begin{corollary}[A necessary event below $j_\eta$]\label{cor:event} If $\normT ed_\eta, \] which is impossible. Put $z_* = \sin r_*$ and, for $\varepsilon>0$, define the global field \begin{equation}\label{eq:correction} e_\varepsilon(x,y,z)=e_*(x,y,z)-\varepsilon(z-z_*). \end{equation} The correction is in $\mathcal V_1=\mathcal H_0\oplus\mathcal H_1$ and so remains in the original admissible space. At the selected point, \begin{equation}\label{eq:negative} G_{e_\varepsilon}(\theta_*,r_*)=0,\qquad \partial_rG_{e_\varepsilon}(\theta_*,r_*) =\partial_rG_{e_*}(\theta_*,r_*)-\varepsilon\cos r_*<0. \end{equation} The value is unchanged, and $\cos r_*>0$ on this tube. On the \emph{whole tube}, $|z|\le\aeta$ and $|z_*|<\aeta$, so \begin{equation}\label{eq:costcorrection} \normT{e_\varepsilon} \le d_\eta+\varepsilon(\aeta+|z_*|) \le d_\eta+2\aeta\varepsilon. \end{equation} Choose \begin{equation}\label{eq:eps} 0<\varepsilon<\frac{j_\eta-d_\eta}{2\aeta}. \end{equation} The field $e_\varepsilon$ still has the strict side signs \eqref{eq:signs}. By the negative derivative in \eqref{eq:negative}, there is a small $s>0$ for which \[ -\dd0, \quad G_{e_\varepsilon}(\theta_*,r_*+s)<0. \] Together with the negative sign at $-\dd$ and positive sign at $\dd$, the intermediate value theorem gives one zero in $(-\dd,r_*-s)$ and another in $(r_*+s,\dd)$. Along with $r_*$, these are three distinct interior zeros on the same fibre. Thus $\PP(e_\varepsilon)$ genuinely fails. The argument is unchanged if $\theta_*$ is either angular endpoint, since that entire fibre is included in $D_\eta$. Letting $\varepsilon\downarrow0$ in \eqref{eq:costcorrection} yields $\rho_\eta\le d_\eta$. If instead $d_\eta\ge j_\eta$, the lateral witness already suffices. Combining these facts with \eqref{eq:lower} proves Theorem~\ref{thm:main}. In particular, the proof never replaces an admissible global field by a freely chosen one-variable perturbation of a single fibre. \subsection{Equality, attainment, and the limits of the conclusion} \label{sec:threshold} \begin{proposition}[What happens at the threshold]\label{prop:attainment} The reduction has the following consequences. \begin{enumerate} \item Every field of norm strictly less than $\rho_\eta$ satisfies $\PP$. \item If $j_\eta\le d_\eta$, the constant field $-j_\eta$ realizes a genuine failure of norm $\rho_\eta=j_\eta$. \item If $d_\eta0$. For $0s$. The feasible multiplier set is compact: the linear map $(\lambda,\mu)\mapsto\lambda v_p+\mu\ell_p$ is injective, so its dual norm bounds the two coefficients. Letting $s\uparrow D(p)$ and taking a convergent subsequence proves the reverse inequality to \eqref{eq:weakdual} and attainment of the maximum. The two inequalities in \eqref{eq:weakdual} must then both be equalities, giving the displayed conclusions. \end{proof} An exact norm for a constructed functional need not make its objective the maximum in \eqref{eq:dualmax}. The certificates below therefore give lower bounds, except at the central points where a feasible field also attains the bound. \subsubsection{Exact and oriented jets} \begin{proposition}[Active derivatives and normal-side costs]\label{prop:exactjet} If $|r|<\delta$, then $D(\theta,r)>|F(\theta,r)|$. Every optimal dual pair in \eqref{eq:dualmax} has $\mu>0$, and every optimal primal field has $E_r=-F_r$. At either normal side, $D(\theta,\pm\delta)=|F(\theta,\pm\delta)|$. There an optimal field with $E_r=-F_r$ exists, although other optimal fields have $E_r<-F_r$ and every optimal oriented-dual pair has $\mu=0$. Consequently, on the entire closed tube, \begin{equation}\label{eq:exactjetcost} D(p)=\min\{\normT e:e\in\V,\ G_e(p)=0,\ \partial_rG_e(p)=0\}. \end{equation} \end{proposition} \begin{proof} At an interior normal point, a feasible field of norm $|F(p)|$ would attain a maximum of its absolute value along the fibre there, so $E_r(p)=0$, contradicting $E_r(p)\le-F_r(p)<0$. If $F(p)=0$, norm zero would force $e=0$, with the same contradiction. Since $\|v_p\|_*=1$ (constant fields attain this value), a dual optimum with $\mu=0$ could achieve at most $|F(p)|$. Thus $\mu>0$, and Proposition~\ref{prop:dual} gives the derivative equality for every primal optimum. These arguments include angular terminals: only interiority along the normal coordinate is used. For the sides use the exact target factors derived in Section~\ref{sec:barrier}, equation~\eqref{eq:RS}: \[ R_\theta(q)=(1-q)^4-\frac{q(1-q)^2(1-2q)}{v_\theta}, \qquad S_\theta=R_\theta+2qR_\theta'. \] The derivative of $q(1-q)^2(1-2q)$ is $(1-q)(1-7q+8q^2)>0$ for $0\le q\le a^2<1/65$. Hence $R_\theta'<0$ and \eqref{eq:RSbounds} gives $00, \] \[ 5c_2-7c_3q+9c_4q^2\ge30+20/v_\theta-(28+35/v_\theta)/65>0. \] Moreover $c_1a^2<4/65+1/16=129/1040<1/8$. Consequently \begin{equation}\label{eq:RSbounds} R_\theta(a^2t^2)\ge1-t^2/8,\qquad S_\theta(a^2t^2)\ge1-3t^2/8>0\quad(|t|\le1). \end{equation} These estimates require no extra small-angle assumption or restriction on perturbation coefficients. Put \begin{equation}\label{eq:Atheta} A_\theta=\frac{aB(\theta)}9. \end{equation} \subsubsection{A trigonometric inequality and its application} For a real trigonometric polynomial $Q$ of degree at most a positive integer $n$, and $00$, giving \begin{equation}\label{eq:barrierK} D(\theta,r)\ge K(\theta,r):=\sqrt{F(\theta,r)^2+ F_r(\theta,r)^2/\beta_{9,\delta}(r)^2} \quad(|r|<\delta). \end{equation} At $r=\pm\delta$ use instead $D\ge|F|$, from the value constraint. Defining $K=|F|$ there makes this bound continuous. In particular $K(\theta,0)=A_\theta/[2\cos(\delta/2)]A_\theta\quad\hbox{if }|t|\ge1/10. \end{equation} \end{proposition} \begin{proof} Set $c=\sqrt{1-a^2u}$ and $c_\delta=\cos\delta$. In the interior, \[ \frac{c^2}{a^2} \frac{\sin^2(\delta/2)-\sin^2(r/2)}{\cos^2(r/2)} =\frac{c^2(1-u)}{(1+c)(c+c_\delta)} \ge\frac{16}{65}(1-u), \] because $c^2\ge c_\delta^2>64/65$ and the denominator is at most $4$. Equations~\eqref{eq:RS}--\eqref{eq:barrierK} give \eqref{eq:Qbarrier}. At the normal sides $u=1$, its second term is zero and the separate value bound gives the same conclusion. On $[0,1]$ the first summand of $\mathcal Q$ has derivative $81(1-u/8)(1-3u/8)\ge2835/64$. The derivative of the second is bounded below by $-(16/65)(1+3/4)=-28/65$. Hence $\mathcal Q'>0$, and $\mathcal Q(0)=16/65$. Also \[ \mathcal Q(1/100)> \frac{81}{100}\left(\frac{99}{100}\right)^2+ \frac{16}{65}\left(\frac{99}{100}\right)^3 =\frac{67127049}{65000000}>1. \] These facts prove both assertions, including the angular terminals and the normal sides. \end{proof} \subsubsection{The universal radius and a sharper floor} The function $B$ has derivative \begin{equation}\label{eq:Bprime} B'(\theta)=\frac{\sin^2(2\theta)}{32} \bigl(3-4\sin^2(2\theta)\bigr). \end{equation} It increases strictly up to $\pi/6$ and decreases strictly afterwards. Thus $B\ge\beta$ on $I$, and its minimum is reached only at those endpoints which realize $\beta$. \begin{proposition}[The normal-side event is never first]\label{prop:jetonly} On the whole original domain, $\rho_\eta=d_\eta\le A_*A_*$. Theorem~\ref{thm:main} proves the assertion about $\rho_\eta$. A point on a normal side costs at least $j_\eta$ by its value constraint, so cannot occur in a minimizing pair. \end{proof} \begin{proposition}[A stronger all-point floor]\label{prop:strongfloor} Put $g=[2(1+\cos\delta)]^{-1}$. For every point of the closed tube, \begin{equation}\label{eq:strongfloor} \frac{D(\theta,r)^2}{A_\theta^2}\ge g+61t^2,\qquad D(\theta,r)\ge\frac{A_\theta}{2\cos(\delta/2)}. \end{equation} The second inequality is strict when $r\ne0$. In particular $A_*/[2\cos(\delta/2)]\le\rho_\eta\le A_*$. \end{proposition} \begin{proof} For $u=t^2$ and $c=\sqrt{1-a^2u}$, keep the exact derivative factor from the preceding proof instead of replacing it by a constant: \[ \frac{c^2(1-u)}{(1+c)(c+\cos\delta)} \ge g(1-a^2u)(1-u). \] The denominator is at most $2(1+\cos\delta)$. Applying the target bounds gives \[ \frac{D^2}{A_\theta^2}\ge 81u(1-u/8)^2+ g(1-a^2u)(1-u)(1-3u/8)^2. \] At $u=1$ the derivative term is zero and the value constraint gives the same inequality. Since $(1-u/8)^2\ge49/64$ and $\prod_i(1-x_i)\ge1-\sum_i x_i$ for $x_i\in[0,1]$, the last display is at least \[ g+\left(\frac{3969}{64}-g(a^2+7/4)\right)u\ge g+61u. \] Indeed $g<1/2$ and $a^2+7/4<2$ make the coefficient greater than $61$. Finally $\sqrt g=1/[2\cos(\delta/2)]$ and $A_\theta\ge A_*$, while Proposition~\ref{prop:jetonly} supplies the upper bound and the identity of costs. \end{proof} \begin{corollary}[Localization and reflection of minimizing points]\label{cor:location} Every minimizing pair has $|t|<1/10$ and $B(\theta)\le2\cos(\delta/2)\beta$, with the latter inequality strict if $t\ne0$. Moreover $D(\theta,-r)=D(\theta,r)$. \end{corollary} \begin{proof} The outer bound \eqref{eq:barrierconsequences} and $d_\eta\le A_*$ exclude $|t|\ge1/10$. The sharper floor gives the angular restriction. The transformation $e(x,y,z)\mapsto-e(x,y,-z)$ preserves the space and tube norm. At the reflected point it reverses the value and preserves the normal derivative, exactly as required by the oddness of $F$ and evenness of $F_r$. Applying it twice proves the equality of costs. \end{proof} In particular the sharper all-point floor complements, rather than replaces, the $\mathcal Q$ barrier needed at $|t|=1/10$. \section{The spherical atomic certificate}\label{sec:atomic} \subsection{An atomic certificate on all 100 modes}\label{sec:lift} The existence of a norm-preserving atomic formula for a functional on a finite-dimensional subspace of $C(K)$, and the extremal sign conditions at its atoms, are classical \cite[Theorem 4.4, Corollary 4.5 and Proposition 4.7]{Dressler}. At prescribed nodes, the relation between exact formulas and minimal weight norms is described by \cite[Section 3, Theorem 9]{DavydovSchaback}. Representing one functional does not reconstruct every field from those evaluations. Here the work is to exhibit supported points and weights, prove the identity on the complete spherical space and attain the continuous tube norm explicitly. \subsubsection{A common geometric condition} Fix $\tau=1/10$ and $r_t=\arcsin(at)$, $c_t=\sqrt{1-a^2t^2}$. At an angular position $\theta$, choose a nonzero signed angle $\alpha$ such that the closed segment between $\theta$ and $\theta+\alpha$ lies in $I$. Assume \begin{equation}\label{eq:supportcondition} \sin(|\alpha|/2)\ge W(a),\qquad W(a)=\frac{11a}{10+a^2}. \end{equation} In particular $|\alpha|\le L<\pi/4$. For each fixed $\theta,t,\alpha$, with $|t|\le\tau$, define \begin{equation}\label{eq:bq} \begin{gathered} w=e^{i\alpha}-1,\quad h=|w|^2=4\sin^2(\alpha/2),\qquad \\ b_t(s)=\frac{1-a^2ts}{c_t},\quad q_t(s)=\frac{a^2(s-t)^2}{c_t^2h}\quad(-1\le s\le1). \end{gathered} \end{equation} All derivatives in $s$ below hold $\theta,t,\alpha$ fixed. In particular they do not differentiate a choice of angular side with respect to its centre. The positive affine function $b_t$ is tangent to $c(s)=\sqrt{1-a^2s^2}$ at $s=t$: \begin{equation}\label{eq:tangent} b_t(t)=c_t,\quad b_t'(t)=c'(t)=-a^2t/c_t,\quad b_t(s)^2-c(s)^2=\frac{a^2(s-t)^2}{c_t^2}. \end{equation} The following exact maximum determines reality of the intersections: \begin{equation}\label{eq:maxdefect} \max_{|s|\le1,\ |t|\le\tau} \frac{a|s-t|}{1-a^2ts}=W(a). \end{equation} Indeed, for fixed $t$, the quotient without absolute value has derivative $(1-a^2t^2)/(1-a^2ts)^2>0$ with respect to $s$. The maximum of the absolute value is at $s=\pm1$. For $t\ge0$, the difference between the two endpoint ratios, after removing $a$, is $2t(1-a^2)/(1-a^4t^2)\ge0$, selecting $s=-1$. By reflection, the maximum at fixed $t$ is $a(1+|t|)/(1+a^2|t|)$, strictly increasing in $|t|$. Equality in \eqref{eq:maxdefect} occurs only at $(s,t)=(-1,\tau),(1,-\tau)$. \subsubsection{Points in the tube and conditional weights} Let \begin{equation}\label{eq:roots} S_\pm(s)=\frac{b_t(s)\pm\sqrt{b_t(s)^2-4q_t(s)}}2. \end{equation} Equations~\eqref{eq:supportcondition} and \eqref{eq:maxdefect} imply that these are real and belong to $[0,b_t(s)]$, since \[ \frac{4q_t(s)}{b_t(s)^2} =\frac{1}{\sin^2(\alpha/2)} \left(\frac{a|s-t|}{1-a^2ts}\right)^2\le1. \] Use rotated coordinates $u=x\cos\theta+y\sin\theta$, $v=-x\sin\theta+y\cos\theta$ and define points $Q_{s,\pm}$ by \begin{equation}\label{eq:atoms} u+iv=b_t(s)+S_\pm(s)w,\qquad z=as. \end{equation} The quadratic equation for $S$ and \eqref{eq:tangent} give $|u+iv|^2=1-a^2s^2$. Thus these are points on the original sphere. For $0\le\xi\le1$ the argument of $1+\xi(e^{i\alpha}-1)$ lies between $0$ and $\alpha$: its real part is positive and \[ \frac{\operatorname{Im}}{\operatorname{Re}} =\frac{\xi\sin\alpha}{1-\xi+\xi\cos\alpha} \] is monotone with the sign of $\alpha$. Taking $\xi=S_\pm/b_t$ proves $Q_{s,\pm}\in T_\eta$, even when $\theta$ is a terminal. The auxiliary endpoints $b_t$ and $b_te^{i\alpha}$ themselves need not lie on the sphere. Define the probability measure \begin{equation}\label{eq:conditional} \pi_{t,s}=\frac{S_+(s)}{b_t(s)}\delta_{Q_{s,-}} +\frac{S_-(s)}{b_t(s)}\delta_{Q_{s,+}}. \end{equation} At $s=t$, the roots are $0$ and $b_t(t)$, so the weights are $1$ and $0$. For $s\ne t$ both weights are positive. At a repeated root the two half-weights are added at their common point. \subsubsection{Moment recurrence and both data of the jet} For the distribution on the coordinate $S$ underlying \eqref{eq:conditional}, its moments satisfy \begin{equation}\label{eq:moments} m_0=1,\quad m_1=2q_t/b_t,\qquad m_n=b_tm_{n-1}-q_tm_{n-2}\quad(n\ge2). \end{equation} This follows from $S^2-b_tS+q_t=0$ and remains true at repeated roots. Set $k_n=b_tm_n$. Then \[ k_0=b_t,\quad k_1=2q_t,\quad k_n=b_tk_{n-1}-q_tk_{n-2},\qquad \deg_s k_n\le n+1. \] The degree bound follows by induction from $\deg b_t\le1$ and $\deg q_t\le2$. Hence \begin{equation}\label{eq:rationalmoment} \begin{aligned} R_m(t,s):=\int(u+iv)^m\,d\pi_{t,s} &=\sum_{n=0}^{m}\binom mn b_t(s)^{m-n}w^n m_n(s)\\ &=\frac{P_{m+1}(s)}{b_t(s)},\qquad \deg P_{m+1}\le m+1. \end{aligned} \end{equation} Because $q_t(t)=q_t'(t)=0$, \eqref{eq:moments} also gives $m_n(t)=m_n'(t)=0$ for $n\ge1$. Thus \begin{equation}\label{eq:momentjet} R_m(t,t)=c_t^m,\qquad \partial_sR_m(t,t)=m c_t^{m-1}c'(t). \end{equation} For completeness, the real and imaginary parts of \begin{equation}\label{eq:fourierbasis} z^k(u+iv)^m,\qquad k+m\le9, \end{equation} with only the real part when $m=0$, form a basis of $\V$. There are $10+2(9+8+\cdots+1)=100$ such functions. They are independent: at fixed $z\in(-1,1)$, put $u+iv=\sqrt{1-z^2}e^{i\phi}$. Fourier uniqueness in $\phi$ forces each polynomial coefficient in $z$ to vanish; its nonzero spherical factor $(1-z^2)^{m/2}$ can be divided out. Varying $z$ shows that all coefficients vanish. Proposition~\ref{prop:basis} supplies the dimension, so independence proves completeness. For every real field $e\in\V$, equations \eqref{eq:rationalmoment}--\eqref{eq:momentjet} imply \begin{equation}\label{eq:fulljet} g_e(s):=\int e\,d\pi_{t,s}=\frac{P_e(s)}{b_t(s)},\quad \deg P_e\le10,\qquad g_e(t)=E(\theta,r_t),\quad g_e'(t)=\frac{a}{c_t}E_r(\theta,r_t). \end{equation} The factor $z^k=a^ks^k$ contributes its ordinary product-rule derivative, so the statement includes every $k,m$ in \eqref{eq:fourierbasis}. This reproduces the two original jet data. It does not assert that an arbitrary field or its average is a free polynomial of degree nine. The degree-ten numerator is retained and handled next. \subsubsection{Rational interpolation and signs} Let $P=T_9$ and $U=U_8$, with their explicit forms \begin{equation}\label{eq:chebyshev} \begin{split} P(s)&=9s-120s^3+432s^5-576s^7+256s^9,\\ U(s)&=1-40s^2+240s^4-448s^6+256s^8,\qquad P'=9U,\\ N(s)&=(s^2-1)U(s),\qquad s_j=\cos(j\pi/9),\quad \sigma_j=(-1)^j\quad(0\le j\le9). \end{split} \end{equation} The elementary identity $P(\cos\phi)=\cos(9\phi)$ follows by expanding the cosine recurrence. It gives $P(s_j)=\sigma_j$ and $\max_{[-1,1]}|P|=1$. The ten $s_j$ are the simple roots of $N$. Since $N$ has leading coefficient $256$ and the roots are decreasing, $\operatorname{sign}N'(s_j)=\sigma_j$. For $|t|\le\tau$, $U(t)\ge3/5-7/15625>1/2$, by discarding the positive terms in its polynomial. Thus $N(t)<0$. Also $\min_j|s_j|=\sin(\pi/18)>1/9$ by strict concavity of sine, so $|s_j-t|>1/90$. Define \begin{equation}\label{eq:lambdamass} \lambda(t)=-\frac{N'(t)}{N(t)}-\frac{a^2t}{1-a^2t^2}, \qquad M(t)=P'(t)+\lambda(t)P(t). \end{equation} Writing $\kappa=b_t'(t)/b_t(t)=-a^2t/(1-a^2t^2)$, the functional $J_tQ=Q'(t)+\lambda(t)Q(t)$ satisfies $J_t(N/b_t)=0$. For the cardinal polynomials $L_j(s)=N(s)/[(s-s_j)N'(s_j)]$, every polynomial $V$ of degree at most ten has $V-\sum_jV(s_j)L_j$ equal to a constant times $N$. Divide this identity by $b_t$ and apply $J_t$ to obtain \begin{equation}\label{eq:rationalinterp} J_t(V/b_t)=\sum_{j=0}^{9}d_j(t)(V/b_t)(s_j),\qquad d_j(t)=\frac{b_t(s_j)}{c_t} \frac{-N(t)}{(t-s_j)^2N'(s_j)}. \end{equation} The expression for $d_j$ follows by differentiating $b_t(s_j)L_j(s)/b_t(s)$; the $N'$ and $b_t'$ terms cancel using the specified sign of $\kappa$. All denominators are nonzero and \begin{equation}\label{eq:signweights} \sigma_jd_j(t)>0. \end{equation} \begin{theorem}[Full-space atomic representation]\label{thm:certificate} Under \eqref{eq:supportcondition}, for every $|t|\le1/10$ and every $e\in\V$, the functional \begin{equation}\label{eq:originalJ} J_{\theta,t}(e)=\frac{a}{c_t}E_r(\theta,r_t)+ \lambda(t)E(\theta,r_t) \end{equation} has the representation \begin{equation}\label{eq:atomicJ} J_{\theta,t}(e)=\sum_{j=0}^{9}d_j(t) \left[\frac{S_+(s_j)}{b_t(s_j)}e(Q_{s_j,-})+ \frac{S_-(s_j)}{b_t(s_j)}e(Q_{s_j,+})\right]. \end{equation} Its exact norm is \begin{equation}\label{eq:exactnorm} \|J_{\theta,t}\|_*=M(t)=\sum_{j=0}^{9}|d_j(t)|>0. \end{equation} The normalized original multipliers are $\lambda_r=\lambda(t)/M(t)$ and $\mu_r=a/(c_tM(t))>0$. \end{theorem} \begin{proof} Apply \eqref{eq:rationalinterp} to the degree-ten numerator in \eqref{eq:fulljet}. The two matching data there identify its left-hand side with \eqref{eq:originalJ}. At the contact nodes $q_t(s_j)>0$, so both conditional weights are positive. The signed measure in \eqref{eq:atomicJ} has total variation at most $\sum_j|d_j|$. All its points lie in $T_\eta$, so this bounds the dual norm. The global polynomial $p_*(x,y,z)=P(z/a)$ belongs to $\V$ and has $\normT{p_*}=1$. At both points of level $s_j$ it equals $\sigma_j$. Therefore \[ J_{\theta,t}(p_*)=\sum_jd_j\sigma_j= \sum_j|d_j|=P'(t)+\lambda(t)P(t)=M(t)>0. \] This proves saturation and hence \eqref{eq:exactnorm}. Points at different levels cannot coincide. If two points at the same level coincide, their weights have the same sign and add, preserving total variation and saturation. The chain factor $a/c_t$ is positive, giving the asserted sign of $\mu_r$ for both signs of $t$. \end{proof} The proof verifies the identity on a basis of the full 100-dimensional real space and uses the original tube norm. A positive combination of \emph{signed contact evaluations} is not a functional positive on every nonnegative field. \subsubsection{Tangency, aggregation, and terminal positions} If \eqref{eq:supportcondition} is strict, all twenty contact points are distinct. On its equality boundary, \eqref{eq:maxdefect} shows that a discriminant vanishes only for $(s,t)=(-1,\tau)$ or $(1,-\tau)$. At that level $S_+=S_-=b_t/2$, so the two half-weights form one atom. There are exactly nineteen atoms when $|t|=\tau$, and twenty when $|t|<\tau$. This statement applies at each angular centre whose chosen side has equality, including a terminal; it is not restricted to the angular midpoint. Although the individual square roots may fail to be differentiable in parameters at tangency, the moment recurrence and rational expressions are regular because $b_t>0$. Jet matching differentiates those moments, not the individual roots. At $t=0$, separately, $\lambda(0)=0$, $M(0)=9$ and $\mu_r=a/9$; the value constraint remains present. \subsection{The objective estimate on the original geometry}\label{sec:objective} We now prove the scalar estimate on the full parameter domain. Its geometric premise is precisely the support condition needed for Theorem~\ref{thm:certificate}; the scalar estimates themselves hold for every original geometry. \begin{lemma}[Uniform objective margin]\label{lem:objective} For $|t|\le1/10$, let $R=R_\theta(a^2t^2)$ and $S=S_\theta(a^2t^2)$. With \eqref{eq:lambdamass}, $01+\frac{111}{8}t^2,&0<|t|\le1/10. \end{cases} \end{equation} Here $\Delta_\theta(t)=9(S+\lambda(t)tR)-M(t)>333t^2$ off centre. \end{lemma} \begin{proof} Put $x=t^2\le1/100$. The explicit polynomials give, for $t\ne0$, \[ \begin{gathered} \tfrac122-\frac1{64}=\frac{127}{64}>0. \] Also \[ 0<-tU'(t)=x(80-960x+2688x^2-2048x^3)<81x, \] since $80+2688/10000<81$. It follows that $0<\lambda(t)t<3x+162x=165x$, and hence \[ 037x, \] \[ 9R-P/t\ge x(951/8-108/25-4/15625)>114x. \] The final strict inequalities follow respectively from $9/40>4/15625$ and $111/200>4/15625$. Since $\lambda t>0$, \[ \Delta_\theta(t)=9(S-U)+(\lambda t)(9R-P/t)>333x. \] Dividing by $0A_\theta(1+111t^2/8) \quad(0<|t|\le1/10),\qquad D(\theta,0)=A_\theta. \end{equation} \end{corollary} \begin{proof} For every feasible field, \eqref{eq:exactnorm} and the positivity of $a/c_t$ imply \[ \normT e\ge-\frac{J_{\theta,t}(e)}{M(t)} \ge\frac{(a/c_t)F_r+\lambda(t)F}{M(t)} =\frac{aB(\theta)(S+\lambda(t)tR)}{M(t)}. \] Take the infimum and use Lemma~\ref{lem:objective}. The value constraint is an equality, so its transfer permits either sign of $\lambda$. The inequality uses the positive derivative multiplier; $\lambda t>0$ was used in the scalar objective estimate. At $t=0$ the lower bound is attained by the global field $e=-A_\theta P(z/a)$: its value is zero, its normal derivative is $-B(\theta)$, and its tube norm is $A_\theta$. This proves the pointwise central equality, without classifying all its minimizers. \end{proof} \section{Rigidity and exact stability regimes}\label{sec:exact} \subsection{Central rigidity and two support regimes}\label{sec:regimes} \subsubsection{Central rigidity, including its angular boundary}\label{sec:centralrigidity} The central rigidity argument includes an available side of length exactly $2\delta$ by distinguishing the auxiliary chord from the locations of its actual contact atoms. \begin{lemma}[A closed angular-side condition gives central rigidity]\label{lem:rigidity} Fix $\theta_0\in I$. If one of its available angular sides has length $b_*\ge2\delta$, then \[ D(\theta_0,0)=A_{\theta_0},\qquad e=-A_{\theta_0}p_*\quad\text{is its unique minimizer in }\V. \] \end{lemma} \begin{proof} Write $\varepsilon\in\{-1,1\}$ for the direction of the available side and $\ell=|\alpha|$ for an auxiliary chord angle. At $t=0$ the kernel has \[ b_0(s)=1,\qquad q_0(s)=\frac{a^2s^2}{4\sin^2(\ell/2)},\qquad S_\pm(s)=\frac{1\pm\sqrt{1-4q_0(s)}}2. \] Reality for every $|s|\le1$ needs only $\ell\ge2\delta$. The coordinates $z=as$ and $1+S_\pm(e^{i\varepsilon\ell}-1)$ give points on the sphere, since their horizontal squared modulus is $1-a^2s^2$. At $\ell=2\delta$ the ten contact levels $s_j=\cos(j\pi/9)$ are all nonzero, hence $00$, even when the auxiliary chord endpoint lies beyond that side. The roots remain real for all $|s|\le1$. The moment and interpolation identities are identities for global polynomial fields on the sphere; their algebra does not require the intermediate, noncontact points to lie in the tube. Only the atoms used to bound the functional by the tube norm must lie there, as just proved. For completeness, at $t=0$ the conditional moments obey $m_0=1$, $m_1=2q_0$, and $m_n=m_{n-1}-q_0m_{n-2}$. Thus $\deg m_n\le n+1$, and $m_n(0)=m_n'(0)=0$ for $n\ge1$. On every Fourier--$z$ basis element the averaged function has degree at most ten and derivative $aE_r(\theta_0,0)$ at zero. Its interpolation remainder at the ten levels is a multiple of the even polynomial $N(s)=(s^2-1)U_8(s)$, whose derivative at zero vanishes. The same coefficients \[ d_j=\frac{-N(0)}{s_j^2N'(s_j)},\qquad \operatorname{sgn}d_j=(-1)^j \] therefore represent $a\ell_{(\theta_0,0)}$ for all the chord angles just described. The two conditional weights are positive, and evaluation on $p_*=T_9(z/a)$ shows that the mass and norm are exactly $\sum_j|d_j|=T_9'(0)=9$. This proves the lower cost $A_{\theta_0}$; the field $-A_{\theta_0}p_*$ attains it with value zero. Normalize any minimizer by $h=-e/A_{\theta_0}$. Then $\normT h=1$ and $a\,\partial_r(h\circ X)(\theta_0,0)\ge9$ by feasibility; the norm of the functional forces equality. In each representation every signed contact value equals one, since their positive weighted sum equals its full mass. Put $z_j=as_j$, $R_j=\sqrt{1-z_j^2}$. The farther contact has angular offset $\varepsilon\Psi_j(\ell)$, where \[ \Psi_j(\ell)=\frac{\ell}{2} +\arccos\frac{\cos(\ell/2)}{R_j},\qquad \Psi_j'(\ell)=\frac12+ \frac{\sin(\ell/2)}{2\sqrt{\sin^2(\ell/2)-z_j^2}}>0. \] This follows by rotating the chord by $-\varepsilon\ell/2$; its real part becomes $\cos(\ell/2)$. For $\ell>2\delta$ in the above interval every farther contact traces an open angular arc within the tube. Thus $h-p_*$ vanishes on an arc at each of the ten heights $z_j$. At fixed height it is a trigonometric polynomial of degree at most nine, so it vanishes on the entire latitude. In \eqref{eq:fourierbasis} the coefficient at frequency $m$ has a polynomial factor of degree at most $9-m$, times the nonzero factor $(1-z^2)^{m/2}$. Ten distinct roots force every coefficient to vanish. Hence $h=p_*$ on the sphere. \end{proof} \subsubsection{A uniform side throughout the sector} If \begin{equation}\label{eq:Lcondition} \sin(L/4)\ge W(a), \end{equation} choose $\alpha=L/2$ for $\theta\le(\eta+\omega)/2$ and $\alpha=-L/2$ otherwise. Every chosen segment lies in $I$ and \eqref{eq:supportcondition} holds. Thus the central bound \eqref{eq:pointlower} applies at every angle, and the barrier \eqref{eq:barrierconsequences} covers $|t|\ge1/10$. Both bounds apply at $|t|=1/10$. The side selection may jump at the angular midpoint, but both choices represent the same functional and exact norm; no globally continuous selection of atoms is asserted. Since $B$ has its minimum on $I$ at an endpoint, these two overlapping bounds imply the global lower bound \begin{equation}\label{eq:Llower} d_\eta\ge\frac a9\min\{B(\eta),B(\omega)\}. \end{equation} The single-maximum property of $B$ follows from the derivative computed immediately before Proposition~\ref{prop:jetonly}. Condition~\eqref{eq:Lcondition} is necessary and sufficient for reality of all intersections of this particular uniform-side construction: if it fails, $(s,t)=(-1,1/10)$ has negative discriminant for each chosen side. This does not exclude other certificates or imply a smaller nodal-stability radius. Moreover $W(a)>a=\sin\delta$, so \eqref{eq:Lcondition} implies $L>4\delta$, even on its equality boundary. In particular Lemma~\ref{lem:rigidity} applies at every centre, using a side of length at least $L/2>2\delta$. \subsubsection{A local side and a closed target margin} Let $\eta<\theta_c<\omega$ and suppose \begin{equation}\label{eq:Mcondition} \sin((\omega-\theta_c)/2)\ge W(a),\qquad \min\{B(\theta_c),B(\omega)\}\ge2\cos(\delta/2)B(\eta). \end{equation} For $\theta\in[\eta,\theta_c]$ choose $\alpha=\omega-\theta$. The first condition qualifies the atomic construction on this entire zone, including its cutoff, both strip edges and support tangencies. The outer barrier covers $|t|\ge1/10$. The second condition supplies the target margin on the complementary angular zone, with its closed boundary treated separately below. In the open second zone $(\theta_c,\omega)$ strict unimodality gives $B(\theta)>\min\{B(\theta_c),B(\omega)\}$. Proposition~\ref{prop:strongfloor} therefore gives $D>A_\eta$. At either endpoint $\theta=\theta_c,\omega$ the strict off-centre floor gives the same conclusion for $r\ne0$. For the two remaining centres note that $W(a)>a=\sin\delta$, so $\omega-\theta_c>2\delta$. From $\theta_c$ the side toward $\omega$ is long enough for Lemma~\ref{lem:rigidity}; from $\omega$ the side toward $\eta$ has length $L>2\delta$. It follows that $D(\theta_c,0)=A_{\theta_c}>A_\eta$ and $D(\omega,0)=A_\omega>A_\eta$, since $2\cos(\delta/2)>1$. We have proved, including the equality in the target margin, \begin{equation}\label{eq:Mright} D(\theta,r)>A_\eta \quad(\theta_c\le\theta\le\omega,\ |r|\le\delta). \end{equation} Thus no atom construction is needed to cover the open second zone; the two endpoint centres use only the central cost lemma. A strict lower bound of $\sqrt{65}B(\eta)/4$ or a lower bound of $2B(\eta)$ for $\min\{B(\theta_c),B(\omega)\}$ each implies the target-margin condition in \eqref{eq:Mcondition}. On $[\eta,\theta_c]$ strict unimodality and $B(\theta_c)>B(\eta)$ give $B(\theta)>B(\eta)$ for $\eta<\theta\le\theta_c$. The central-strip certificate and outer barrier then show \begin{equation}\label{eq:Mlower} d_\eta\ge A_\eta. \end{equation} The closed regions overlap at $\theta_c$ and the normal bounds overlap at $|t|=1/10$; no unexamined seam remains. The support premise already implies $L>2\delta$ and does not require $L>4\delta$. \begin{figure}[tbp] \centering \includegraphics[width=\linewidth]{figures/tube_support.pdf} \caption{The tube and an explicit supported jet, in coordinate charts. Here $\eta=\arctan(1/100)$, $\omega=\arctan(1/2)$. (a) The full angular interval and signed normal distance, with independent axis scales; this is not an isometric development of the sphere. (b) The twenty signed contacts of Theorem~\ref{thm:certificate} at $\theta_0=\eta$, $t=1/10$, using $\alpha=3\delta$, in coordinates $((\theta-\eta)/\delta,\sin r/a)$. The signs label the functional weights. The support count is not asserted minimal, and the drawing is not a norm certificate.} \label{fig:tube} \end{figure} The plotted case is qualified without numerical sampling: $\eta<1/100$ and $\omega>2/5$ imply $L>39/100$, whence $\sin(L/4)>9/100$, whereas $W(a)<11\eta/40<11/4000$. Also $\delta<1/8$ gives $\sin(3\delta/2)>(11/10)\delta>W(a)$ and $3\deltaA_\theta$ whenever $r\ne0$. At $r=0$, every angle has an available side of length at least $L/2>2\delta$, so Lemma~\ref{lem:rigidity} gives $D(\theta,0)=A_\theta$ with the unique field $-A_\theta T_9(z/a)$. Since $B$ has no interior minimum, the minimizing points are exactly the terminal centres that realize $\beta$. The minimum cost is $A_*$ and all those points have the same unique minimizing field $e_C$. This includes a tie between the two terminal fibres. Under M, \eqref{eq:Mright} excludes the entire second zone. In the first zone the same two bounds exclude $r\ne0$, and $B(\theta)>B(\eta)$ for $\theta>\eta$ excludes the other centres. Lemma~\ref{lem:rigidity} applies at $\eta$, giving the only minimizing pair $((\eta,0),e_C)$ and cost $A_\eta=A_*$. In both cases Proposition~\ref{prop:jetonly} identifies this cost with $\rho_\eta$. \subsubsection{The critical graph and the closed ball} \begin{proposition}[The critical field and every amplification]\label{prop:criticalgraph} For every original geometry the only zeros of $f+e_C$ in the tube are $r=0$. At any terminal where $B=\beta$, the defining level has zero spherical gradient there. For every $\gamma>1$, $f+\gamma e_C$ has at least three distinct interior zeros on each such terminal fibre. \end{proposition} \begin{proof} The identity \[ 1-\frac{T_9(t)}{9t}=t^2H(t^2),\qquad H(u)=\frac{40}{3}-48u+64u^2-\frac{256}{9}u^3 \] extends continuously to $t=0$. Its derivative satisfies $H'(u)=-(256/3)(u-3/4)^2\le0$, so $H(u)\ge H(1)=8/9$ for $0\le u\le1$. Using $B\ge\beta$ and the positive target bound yields, for $r>0$ and $t=\sin r/a$, \begin{equation}\label{eq:criticalsign} \frac{F+E_C}{\sin r} =B(\theta)R_\theta(a^2t^2) -\beta\frac{T_9(t)}{9t} \ge\frac{55}{72}\beta t^2>0. \end{equation} Oddness gives the opposite sign for $r<0$. Along the centre, the angular derivative is zero and the normal derivative is $B(\theta)-\beta$, proving the gradient assertion. On a terminal with $B=\beta$, the field $F+\gamma E_C$ vanishes at the centre with derivative $(1-\gamma)\beta<0$. It is negative just to its right. At $t_0=\sin(\pi/9)\in(0,1)$, the first positive zero of $T_9$, its value equals the strictly positive $F$. The intermediate value theorem gives another positive zero strictly between the centre and $t_0$. Oddness gives a negative zero as well. This argument places no upper bound on $\gamma$ and does not use normal-side signs. \end{proof} \begin{proof}[Completion of Theorem~\ref{thm:radius}] The general bounds and the exact-jet identity were proved in Propositions~\ref{prop:jetonly}, \ref{prop:strongfloor} and \ref{prop:exactjet}. The preceding classification proves the exact L/M radii and all their minimizing pairs. If a field of norm at most $A_*1$; as $\gamma$ decreases to one their costs approach $A_*$. Thus the failure infimum is not attained in these regimes. \end{proof} \begin{figure}[tbp] \centering \includegraphics[width=\linewidth]{figures/fibre_threshold.pdf} \caption{The terminal fibre $\theta=\eta$ for the same geometry as Figure~\ref{fig:tube}. The horizontal coordinate is $t=\sin r/a$; values are divided by $a\beta$. Curves show the target $F$ and the perturbed fields $F+\gamma E_C$ with $\gamma=1,2$. The second panel enlarges the central interval. The guides $t=\pm\sin(\pi/9)$ are zeros of $T_9$, used as sign witnesses in Proposition~\ref{prop:criticalgraph}; they are not the outer roots of the amplified field. The curves illustrate the continuous proof; the conclusion of at least three zeros follows from the proposition.} \label{fig:fibre} \end{figure} The critical graph is smooth although its defining level is singular at the minimizing terminals. The attainment of the jet minimum is therefore compatible with nonattainment of actual graph failure. Outside the proven exact regimes, $e_C$ still preserves the graph and its amplifications still fail, but $A_*$ need not be the true failure radius. \section{Two examples separating a criterion from a formula}\label{sec:examples} \subsection{A valid radius beyond the uniform support condition} \begin{example}[An adapted sector beyond L]\label{ex:M} Let \begin{equation}\label{eq:Mexample} \eta=\arctan(1/100),\qquad L=\frac{21}{5}\delta,\qquad \omega=\eta+L,\qquad \theta_c=\eta+\frac32\delta. \end{equation} Then \eqref{eq:Lcondition} fails but \eqref{eq:Mcondition} holds. Hence $\rho_\eta=A_\eta$ with the unique critical field and safe closed ball of Theorem~\ref{thm:radius}. \end{example} \begin{proof} Here $\tan\delta=25/10001$, so \[ \eta<1/100,\qquad \delta<1/400,\qquad \omega<41/2000<\pi/6. \] In particular the original domain holds. Since $\arctan x\ge x/(1+x^2)$ for $x>0$ (integrate the decreasing function $(1+s^2)^{-1}$), \[ \delta>\frac{25/10001}{1+(25/10001)^2} >\frac{3}{1250}>\frac{6\eta}{25}. \] Thus $\theta_c>34\eta/25>4\eta/3$. Also $a>(99/100)\delta$ and $a^2<1/10$, using $\sin x\ge x-x^3/6$ and $\delta<1/400$. Therefore \[ \sin(L/4)<\frac{21}{20}\delta <\frac{1089}{1010}\delta <\frac{11a}{10+a^2}=W(a), \] which violates the uniform support test, without saying anything yet about the radius. For the adapted side, $\omega-\theta_c=27\delta/10$. On $0\le u\le27\delta/20$, $\cos u>9/10$, giving \[ W(a)<\frac{11}{10}\delta <\frac{243}{200}\delta <\sin(27\delta/20). \] This proves the first M premise. For $\theta\in[\theta_c,\omega]$, $2\theta<1/20$. The elementary sine and cosine estimates give \[ \frac{\sin(2\theta)}{2\theta}>\frac{2399}{2400}, \qquad \cos(2\theta)>\frac{799}{800}. \] Since $(2399/2400)^3>799/800$ and $(799/800)^2>399/400$, while $B(\eta)<\eta^3/8$, \[ \frac{B(\theta)}{B(\eta)} >\left(\frac{\theta}{\eta}\right)^3\frac{399}{400} >\left(\frac43\right)^3\frac{399}{400} =\frac{532}{225}>2>2\cos(\delta/2). \] This verifies the closed M margin, with room to spare, at both endpoints and on the whole second zone. The stated theorem applies. \end{proof} \subsection{A narrow sector where the formula is false} The next construction changes the geometry, not the perturbation space or its continuous norm. Its degree-six correction exploits the small variation of a field between very close terminal fibres. \begin{example}[A strict upper bound below the central candidate]\label{ex:F} Let \begin{equation}\label{eq:Fexample} \tan\omega=\frac12,\qquad \tan\eta=\frac{499999999}{10^9}. \end{equation} There is an explicit $e_*\in\V$ and an explicit positive $U$ such that \[ G_{e_*}(\eta,0)=\partial_rG_{e_*}(\eta,0)=0,\qquad \normT{e_*}\le U0. \] The symbol $\nu$ is distinct from the jet minimum $d_\eta$. Define the global real polynomial of degree six \begin{equation}\label{eq:Fglobal} H(x,y,z)=z\prod_{i=0}^4(x\cos\eta+y\sin\eta-u_i). \end{equation} It is odd in $z$. On the terminal fibre let $H_0(z)=z\prod_{i=0}^4(\sqrt{1-z^2}-u_i)$. Then $H_0'(0)=\nu$. Put $\Delta(z)=\prod_{i=0}^4(\sqrt{1-z^2}+u_i)\ge(2b)^5$ for $|z|\le a$. Factoring the five differences, and using the eight roots of the Chebyshev derivative, gives \[ \mathcal T'(z)=\frac{2304}{a^9} \prod_{i=1}^4(z^2-z_i^2),\qquad H_0(z)=\frac{a^9 z(a^2-z^2)\mathcal T'(z)} {2304\Delta(z)}. \] Here derivatives are with respect to $z$. The Chebyshev identity $1-\mathcal T(z)^2=(a^2-z^2)\mathcal T'(z)^2/81$ and $|z|\sqrt{a^2-z^2}\le a^2/2$ imply \begin{equation}\label{eq:Fterminalbound} |H_0(z)|\le C\sqrt{1-\mathcal T(z)^2},\qquad C=\frac{a^{11}}{512(2b)^5}. \end{equation} Consequently, for $\epsilon>0$, $|\mathcal T+\epsilon H_0|\le\sqrt{1+\epsilon^2C^2}$. This last inequality follows from the Cauchy--Schwarz inequality in two dimensions, including the contact points. At a general point of the tube, $x\cos\eta+y\sin\eta=\sqrt{1-z^2}\cos(\theta-\eta)\in[0,1]$. On $[0,1]$ the derivative of $\prod_{i=0}^4(u-u_i)$ has absolute value at most five, since each factor has absolute value at most one. The mean value theorem therefore gives \begin{equation}\label{eq:Ftubebound} |H-H_0|\le h,\qquad h=5a(1-\cos L)\le\frac52aL^2 \end{equation} on the entire closed tube. Suppose for the moment that $L^2<2\nu/45$. Then the constants \[ g=\frac{a\nu}{9}-h>0,\qquad \epsilon=\frac{g}{C^2},\qquad K=\frac9a+\epsilon\nu \] are positive. Define \begin{equation}\label{eq:Fwitness} e_*=-\frac{B(\eta)}K(\mathcal T(z)+\epsilon H(x,y,z)). \end{equation} Since $z_r=1$ at the centre, this field has the stated exact jet at $(\eta,0)$. Equations~\eqref{eq:Fterminalbound} and \eqref{eq:Ftubebound}, together with $\sqrt{1+x}\le1+x/2$ for $x\ge0$, give \begin{align} \normT{e_*} &\le U:=\frac{B(\eta)}K[1+\epsilon(h+g/2)] \notag\\ &= A_\eta\left[1-\frac{\epsilon g} {2(1+\epsilon a\nu/9)}\right] \frac1{16}, \] because $\alpha>3/8$ and its displayed denominator is less than six. The double-angle identity gives $\cos(\pi/9)\cos(2\pi/9)\cos(4\pi/9)=1/8$; with $\cos(\pi/3)=1/2$ this yields $\prod_{i=0}^4s_i=1/16$. Hence \[ \nu\ge\prod_{i=0}^4\frac{a^2s_i^2}{2} =\frac{a^{10}}{8192}>2^{-53},\qquad L^2\le10^{-18}<\frac{2^{-52}}{45}<\frac{2\nu}{45}, \] where the middle comparison is the integer inequality $45\cdot2^{52}<10^{18}$. This proves the required hypothesis. Finally $\omega=\arctan(1/2)<\pi/6$, so $\beta=B(\eta)$. For any $1<\gamma0. \end{equation} Furthermore, \begin{equation}\label{eq:graphbounds} \|h\|_\infty\le\frac{2m}{\beta},\qquad h'=-\frac{\partial_\theta G_e}{\partial_rG_e}\bigg|_{r=h}, \qquad \|h'\|_\infty\le\frac{m(L_F+M_\theta)}{\kappa}. \end{equation} The derivatives at angular terminals are one-sided. \end{proposition} \begin{proof} Since $m0$. The global affine correction \[ q_p(x,y,z)=-\frac{g}{\cos r}(z-\sin r) \] belongs to the degree-at-most-one subspace of $\V$. It preserves the zero and cancels its normal derivative. Its exact norm is \[ \normT{q_p}=\frac{g}{\cos r}(a+|\sin r|). \] Thus $D(p)\le\normT{e+q_p}\le m+\normT{q_p}$, which proves the pointwise inequality and, using $D(p)\ge d_\eta$, the uniform one. At the graph $|F(\theta,h)|=|E(\theta,h)|\le m$. Integrating $\partial_rF>\beta/2$ from zero to $h$ yields the position bound. The implicit function theorem and \eqref{eq:normalmargin} give the derivative identity. Lemma~\ref{lem:relative} bounds $|\partial_\theta F|$ there by $L_Fm$, and \eqref{eq:Mtheta} bounds $|\partial_\theta E|$ by $M_\theta m$. At a terminal the field extends to an open chart neighborhood and the same theorem gives a local smooth graph; restricting to $I$ justifies the stated one-sided derivative. \end{proof} This proof uses positivity only at the nodes, not monotonicity of the perturbed field throughout the tube. Outside L and M, $d_\eta$ is still the variational value, not an evaluated number. Within either exact regime one can replace it by $A_*$. The bound on slope grows like the inverse of the available margin; no optimality of this general estimate is asserted. \begin{corollary}[Closed additive budget in an exact regime] \label{cor:budget} Assume L or M, $0\le m0. \] Then $B'(\theta_*)\ne0$ and, as $\varepsilon\downarrow0$, \begin{equation}\label{eq:slopedegeneration} h_\varepsilon'(\theta_*)= \frac{aB'(\theta_*)(18C_*)^{1/3}}{3\beta C_*} \varepsilon^{-1/3}\bigl(1+O(\varepsilon^{1/3})\bigr). \end{equation} The derivative is in increasing $\theta$, with its one-sided meaning at a terminal. In particular the graph map from the closed $A_*$ ball, with the tube norm, to $C^1(I)$ is discontinuous at $e_C$, even along fields strictly inside the ball. \end{theorem} \begin{proof} The entire interval $[-1,1]$ is attained by $t=z/a$ on each fibre. Since $T_9$ attains both $-1$ and $1$, the range of $e_\varepsilon/A_*$ is $[-1+3\varepsilon/2,1-\varepsilon/2]$. For $0<\varepsilon<1$ its largest absolute value is $1-\varepsilon/2$. Also $(e_\varepsilon-e_C)/(\varepsilon A_*)=T_9(z/a)+1/2$ has norm $3/2$. This proves \eqref{eq:familynorms}. The difference is $\varepsilon$ times a fixed smooth global field, which proves the stated $C^k$ convergence. Let $G_C=F+e_C\circ X$ and $t_\varepsilon(\theta)=\sin h_\varepsilon(\theta)/a$. Theorem~\ref{thm:radius} and Proposition~\ref{prop:margin} give the unique graphs and their positive normal derivatives, including the terminal interpretation of the implicit function theorem. The critical sign bound \eqref{eq:criticalsign} and oddness give \[ |G_C(\theta,r)|\ge\frac{55}{72}a\beta|t|^3. \] At a zero of $G_{e_\varepsilon}$, on the other hand, $|G_C|=\varepsilon A_*|T_9(t_\varepsilon)+1/2| \le3\varepsilon A_*/2$. Thus $|t_\varepsilon|^3\le12\varepsilon/55$ uniformly on $I$. Since $|\sin h|\ge\cos\delta\,|h|$ for $|h|\le\delta$, \eqref{eq:graphconvergence} follows. The only critical point of $B$ in $(0,\pi/4)$ is its strict maximum $\pi/6$, by \eqref{eq:Bprime}. It cannot minimize on the nondegenerate interval $I$; consequently $B'(\theta_*)\ne0$ at each minimizing terminal. From the coefficient bound behind \eqref{eq:RSbounds}, $a^2c_1(\theta_*)<1/8$, so $C_*>0$. Before restricting to a graph, the polynomial formula \eqref{eq:RS} reads \[ F(\theta,r)=aB(\theta)t \bigl(1-a^2c_1(\theta)t^2+O(t^4)\bigr),\qquad T_9(t)=9t-120t^3+O(t^5). \] The remainders here come from polynomials in $t$ with smooth coefficients on the fixed interval $I$. They may therefore be differentiated as polynomial expressions before the zero is inserted. At $\theta=\theta_*$ the cancellation $B(\theta_*)=\beta$ yields \begin{equation}\label{eq:cubicexpansion} \frac{G_{e_\varepsilon}(\theta_*,r)}{a\beta} =C_*t^3+\frac{\varepsilon}{18} +\varepsilon t+O(t^5+\varepsilon|t|^3). \end{equation} At $t=t_\varepsilon(\theta_*)$, the previously proved $|t|=O(\varepsilon^{1/3})$ now gives \[ t^3=-\frac{\varepsilon}{18C_*}+O(\varepsilon^{4/3}), \qquad t=-\left(\frac{\varepsilon}{18C_*}\right)^{1/3} \bigl(1+O(\varepsilon^{1/3})\bigr). \] The perturbation is independent of $\theta$. Differentiating the field formulas first, with $\partial_rt=\cos r/a$, gives \begin{align} \partial_\theta G_{e_\varepsilon}(\theta_*,r) &=aB'(\theta_*)t+O(t^3),\label{eq:thetadegenerate}\\ \partial_rG_{e_\varepsilon}(\theta_*,r) &=\beta\cos r\bigl(3C_*t^2+O(t^4+\varepsilon)\bigr). \label{eq:rdegenerate} \end{align} At the zero, the error in the second parentheses is $O(\varepsilon^{1/3})$ relative to $3C_*t^2$; $\cos h_\varepsilon=1+O(\varepsilon^{2/3})$. Substituting these estimates into $h'=-G_\theta/G_r$ proves \eqref{eq:slopedegeneration}, including its sign. In a terminal tie it is positive at $\eta$ and negative at $\omega$. Its nonzero coefficient forces $\|h_\varepsilon'\|_\infty\to\infty$, whereas the critical graph is identically zero. This proves the discontinuity into $C^1(I)$ despite \eqref{eq:graphconvergence}. \end{proof} All constants and remainders in this theorem refer to the fixed tube. This is not the limit $\eta\downarrow0$ at fixed $\omega$. Along this path $A_*-\normT{e_\varepsilon}=\varepsilon A_*/2$; the proved $\varepsilon^{-1/3}$ slope divergence neither asserts nor refutes optimality of the general inverse-margin upper bound in \eqref{eq:graphbounds}. \section{Singularity, bandwidth and observable boundaries}\label{sec:limits} \subsection{Two singular events in the closed tube} Let $d_{\mathrm n}$ be the least tube norm of a field $e\in\V$ with $G_e=\partial_rG_e=0$ somewhere in $D_\eta$, and let $d_{\mathrm s}$ be the infimum for a zero of $f+e$ at which the full spherical gradient also vanishes, with the point restricted to $T_\eta$. Then \begin{equation}\label{eq:singulardistances} d_{\mathrm n}=d_\eta,\qquad d_\eta\le d_{\mathrm s}\le A_*; \qquad d_{\mathrm n}=d_{\mathrm s}=A_*\quad\hbox{in L or M}. \end{equation} Indeed, the first identity is Proposition~\ref{prop:exactjet}. A full singular zero is also a feasible normal jet, proving the lower bound. For the upper bound, Proposition~\ref{prop:criticalgraph} gives $G_C(\theta,0)=0$ for every $\theta$ and $\partial_rG_C(\theta_*,0)=B(\theta_*)-\beta=0$ at each minimizing terminal. Hence $\partial_\theta G_C$ vanishes there too. The chart metric $dr^2+\cos^2r\,d\theta^2$ is nonsingular, so these two vanishing components give a zero spherical gradient, including at the boundary of the tube. The same field has norm $A_*$. In L/M both singular events are attained at this norm, while the failure infimum of the graph has the same value and is not attained. No equality $d_{\mathrm s}=d_\eta$ outside these regimes is inferred. If the singular point may lie anywhere on the sphere, the distance is instead zero already for $e=0$. At $(1,0,0)$ the target and its ambient gradient vanish: both factors $y$ and $z$ vanish in the product \eqref{eq:target}. This point is outside the chosen tube. Thus a global discriminant distance is a different question. \subsection{A globally smooth counterexample outside the band} For $\ell>0$ define on the whole sphere \begin{equation}\label{eq:outsideband} e_\ell(x,y,z)=-\frac{3\beta}{2}\ell\sin(z/\ell). \end{equation} These are real smooth global fields with $\|e_\ell\|_{\infty,\Sph}\le3\beta\ell/2\to0$. At a minimizing terminal $\theta_*$, the pullback of $f+e_\ell$ is odd in $r$, vanishes at zero and has derivative $B(\theta_*)-3\beta/2=-\beta/2<0$ there. For sufficiently small $\ell$, $3\beta\ell/20$ and the real temporal field \[ u(x,\tau)=\mathop{\rm Re}\bigl(U(x)e^{i\nu\tau}\bigr), \qquad \nu\ne0. \] Its signed real component at $\tau=0$ is $f$, with the original nodal set. In contrast $|U|=\sqrt{f^2+\epsilon^2}\ge\epsilon$ has no zero. If $T=2\pi/|\nu|$ and $N$ is a positive integer, the RMS over any $N$ complete periods is \[ \left(\frac{1}{NT}\int_{\tau_0}^{\tau_0+NT} u(x,\tau)^2\,d\tau\right)^{1/2} =\frac{|U(x)|}{\sqrt2}>0. \] On any other positive-duration interval the RMS is still positive, although that simple formula need not hold: a nonzero sinusoid cannot vanish identically on such an interval. The nonzero-frequency condition matters; at zero frequency the imaginary part is not observed by the displayed real temporal field. These distinctions concern specified mathematical observables. They do not convert this tube norm into a pressure tolerance, a deposition law or a material response. \section{Discussion and reproducibility}\label{sec:discussion} The variational reduction answers which event controls the local graph: simultaneous cancellation of the value and normal derivative determines the radius throughout the original domain, and normal-side escape costs more. The spherical certificate makes this answer explicit in the uniform and adapted regimes. It also identifies the unique critical field and shows that a minimum-cost singular level can still define a single smooth graph. The genuine graph-failure infimum has the same value but is not attained in these regimes. The two examples delimit that answer in different ways. The adapted sector retains the exact radius beyond the uniform support condition; the narrow sector has a rigorously smaller failure radius than the Chebyshev candidate. Its exact radius is not evaluated here. The general pointwise cost away from the centre and the maximal geometry of validity of the candidate formula remain analytic questions. The conditioning theorem establishes a separate boundary: a fixed positive margin gives graph estimates, whereas convergence to the critical field can produce divergent slopes at the minimizing terminals. The compactness constants in those estimates have no certified numerical values in this work and are not asserted uniform under degeneration of the tube. \subsection{A companion with fixed inputs} The accompanying directory \path{companion/} is an autonomous finite-check package. Its \path{README.md} records the exact input parameters, dependencies, commands and output identities for the adapted-sector and narrow-sector checks. It includes a proof/source map, the selected input records, the check scripts and the figure generator. The two generated PDFs are in \path{paper/figures/} in the \href{https://github.com/aconsciousfractal/B3-Nodal-Stability-Radii}{public repository}. The companion runs independently of the laboratory application and its stored simulations. The checks evaluate prescribed symbolic and rational identities; the normal and optimized Python modes are compared on the same fixed inputs. They are not searches for a support or for a better radius. In particular, selected record checks and finite algebraic identities do not establish the continuous admissibility, signs, rigidity or norm bounds by themselves. Those arguments are contained in the proofs above. The two figures sample explicit formulas for illustration and do not certify a continuous maximum. The companion records the scope of each check separately from the theorem it accompanies. \subsection{Scope of the geometric conclusion} All radius statements concern the real degree-at-most-nine band, the specified normalization and the complete closed tube. They do not describe global nodal topology or guarantee stability for arbitrary smooth errors. A real component, a complex amplitude and a temporal RMS are different observables, as the explicit examples show. Establishing a relation to a physical response would require its own model, admitted inputs and evidence. The present result provides a precise geometric stability question, a finite-dimensional variational characterization, exact regimes and proved obstructions within that question. \paragraph{Acknowledgement.} The preparation of this manuscript and its computational companion used AI assistance. \appendix \section{Why the polynomial space is the full spherical band} \label{app:harmonics} Here is an elementary justification of the equality in \eqref{eq:space}. Write $R^2=x^2+y^2+z^2$. If $p_n$ is homogeneous of degree $n$, a direct product-rule calculation gives, for a homogeneous polynomial $q_m$ of degree $m$, \begin{equation}\label{eq:laplace} \Delta(R^{2k}q_m) =2k(2m+2k+1)R^{2k-2}q_m+R^{2k}\Delta q_m. \end{equation} Put $N=\lfloor n/2\rfloor$, $a_0=1$, and \[ a_k=-\frac{a_{k-1}}{2k(2n-2k+1)}\quad(1\le k\le N),\qquad H_n=\sum_{k=0}^{N}a_kR^{2k}\Delta^k p_n. \] Equation~\eqref{eq:laplace} cancels adjacent terms in $\Delta H_n$; the final derivative $\Delta^{N+1}p_n$ is zero. Thus $H_n$ is homogeneous and harmonic of degree $n$, and $p_n-H_n$ is divisible by $R^2$. Repeating on the quotient expresses $p_n$ as a sum of $R^{2k}$ times homogeneous harmonic polynomials of degrees $n-2k$. On the unit sphere $R=1$, so every ambient polynomial of degree at most nine restricts to the sum of spherical harmonics of degrees at most nine. The reverse inclusion is immediate. The sum is direct. The radial Laplacian formula, obtained by the chain rule in polar coordinates, is \[ \Delta(R^\ell Y)=R^{\ell-2} \bigl(\ell(\ell+1)Y+\Delta_{\Sph}Y\bigr). \] A degree-$\ell$ harmonic restriction therefore has spherical Laplacian eigenvalue $-\ell(\ell+1)$. These eigenvalues are distinct. Applying the product of the other corresponding linear factors in $\Delta_{\Sph}$ to a vanishing sum separates each component and shows that it is zero. Proposition~\ref{prop:basis} then gives the total dimension 100 without assuming a smaller symmetry class. This is a reconstruction of the standard harmonic decomposition, not a novelty claim. \bibliographystyle{plain} {\small\bibliography{references}} \end{document}