\documentclass[11pt]{amsart} \usepackage{mathtools} \usepackage[hidelinks]{hyperref} \hypersetup{ pdftitle={A counterexample to the Barker--Larman problem in dimension three}, pdfauthor={Sihan Chen}, pdfkeywords={Barker--Larman problem, geometric tomography, Legendre polynomials} } \numberwithin{equation}{section} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\R}{\mathbb R} \newcommand{\Lip}{\operatorname{Lip}} \newcommand{\av}{\operatorname{av}} \newcommand{\vol}{\operatorname{vol}} \newcommand{\area}{\operatorname{area}} \newcommand{\dd}{\,\mathrm d} \allowdisplaybreaks[1] \raggedbottom \setlength{\emergencystretch}{1em} \title[The Barker--Larman problem in dimension three]{A counterexample to the Barker--Larman problem in dimension three} \author{Sihan Chen} \address{Chongqing Bashu Science City Secondary School: ChongQing, CN} \email{3565586261@qq.com} \subjclass[2020]{Primary 52A20, 44A12. Secondary 33C45, 11J86} \keywords{Barker--Larman problem, parallel sections, Legendre polynomials, geometric tomography, small divisors} \begin{document} \begin{abstract} We construct an origin-symmetric convex body of revolution in \(\R^3\), with real-analytic boundary and positive principal curvatures, whose planar sections tangent to a concentric inner ball all have the same area. The body is nonspherical and has the same section data as a centered ball. The construction bifurcates from the unit ball at height \(1/\sqrt{3}\), where degree two is the only even Legendre resonance. We construct approximate solutions to arbitrary order and correct them by a Newton iteration on nested closed parameter sets. A finite-order reduction of the linearized section operator and measure estimates for its small divisors provide the inverse bounds needed for convergence. \end{abstract} \maketitle \section{Introduction and main results}\label{sec:intro} Write \(B\) for the closed unit ball of \(\R^3\) and set \(H(u,h)=\{z\in\R^3:z\cdot u=h\}\) for \(u\in S^2\), \(h>0\). The planes \(H(u,h)\) are precisely the supporting planes of \(hB\). Barker and Larman \cite[Conjecture~2, p.~81]{BL} asked whether a convex body containing a prescribed inner body in its interior is determined by the section volumes on the supporting hyperplanes of the inner body. We consider the spherical-inner-body case. \begin{theorem}\label{thm:BL} For every \(t>0\), there exist two distinct origin-symmetric convex bodies \(K,L\subset\R^3\) with real-analytic boundaries and positive principal curvatures, both strictly containing \(tB\), such that \[ \area(K\cap H(u,t))=\area(L\cap H(u,t)) \qquad\text{for every }u\in S^2. \] The body \(L\) is a centered ball and \(K\) is an origin-symmetric, nonspherical body of revolution. Such pairs can be chosen with radial functions arbitrarily close in any prescribed \(C^m\) norm. They satisfy \(\vol(K)>\vol(L)\). \end{theorem} Yaskin proved uniqueness when the two outer bodies are polytopes \cite[Theorem~2.1]{Yaskin}. Matthews \cite[Theorem~1.1]{Matthews} extended planar polygon reconstruction to sections supporting a piecewise-analytic hedgehog. In dimension three, Yaskin and Yaskina obtained uniqueness with a ball as the comparison body under an equal-volume assumption \cite{YY}. The volume ordering in Theorem~\ref{thm:BL} follows from a strict moment inequality. Kurusa and \'Odor \cite[Theorem~5.3]{KurusaOdor} give rigidity when section data and cap volumes agree with those of a centered ball. Makai and Martini \cite[Theorem~1]{MakaiMartini} recover the first radial variation of a smooth outward deformation from the asymptotics of sections tangent to the body being deformed. Here the cutting sphere stays strictly inside both outer bodies, at a radius asymptotic to \(1/\sqrt3\) times the comparison radius. Fish, Ryabogin, Sun and Yaskin \cite[Theorem~6.2]{FRSY} prove uniqueness for fixed bodies when their section functions agree on infinitely many heights with an accumulation point inside the common radial interval. Theorem~\ref{thm:BL} concerns one prescribed height for each pair of bodies. Morales-Amaya \cite[Theorem~1]{Morales} studies the shape of each section. The hypothesis there requires centrally symmetric sections tangent to a suitable inner ball which excludes the center of the outer body. Here the inner ball is concentric and the data are the areas of the sections. Haddad and Ryabogin \cite[Theorem~1]{HR} proved a rigidity result in dimension four for an arithmetically specified set of constant section volumes. Haddad \cite[Theorem~1]{Haddad} constructed four-dimensional examples with section volumes constant to arbitrarily high finite order. Knowledge Sense subsequently gave exact constructions in all dimensions \(d\ge4\) \cite[Theorems~1.1--1.3]{KSExact}, and proved in a companion preprint \cite[Theorem~1.1]{KSFormal} that Haddad's specified four-dimensional series has zero radius of convergence. Part~II of \cite{KSExact} treats the degree-two resonance for \(d\ge5\) using dimension-dependent normalizations and estimates for the effective diagonals. In dimension three the section derivative has Legendre multipliers at the ball. We prove isolation at \(1/\sqrt3\) and use its exact weighted section operator to obtain an order-eight remainder after finite diagonal reduction. The existence proof uses chord coordinates, finite-order diagonal reduction, and parameter exclusion. The strategy follows the approach of de Groot and Price \cite[Sections~3--5]{DGP} for the floating-body equation. Here we establish the estimates for the section-area operator and compute the critical equations for the degree-two resonance. The parameter \(a\) ranges over compact sets of asymptotically full relative measure. For each \(a\) in such a set the resulting boundary is real analytic in the spatial variables. \begin{theorem}[Volume-normalized solutions]\label{thm:main} Let \(P_2(s)=(3s^2-1)/2\), let \(x\) be the meridian angle measured from the axis of revolution, and put \[ h_* =\frac1{\sqrt3},\quad a_* =\arccos h_*,\quad h_1=\frac1{7\sqrt3},\quad a_1=\frac{3}{28\sqrt2}. \] For every sufficiently small \(\tau>0\) there is a compact set \(G_\tau\subset I_\tau=[a_*+\tau,a_*+2\tau]\) with \(|I_\tau\setminus G_\tau|=o(\tau)\) as \(\tau\downarrow0\). For each \(a\in G_\tau\) there exist \(h_a,\nu_a>0\) and an origin-symmetric, nonspherical body of revolution \(K_a\), with real-analytic boundary and positive principal curvatures, such that \[ \begin{gathered} \vol(K_a)=4\pi/3,\qquad h_aB\subset\operatorname{int}K_a,\\ \area(K_a\cap H(u,h_a))=\nu_a\qquad(u\in S^2). \end{gathered} \] The radial functions, expressed in the meridian angle, are analytic on a common positive strip. On a possibly smaller fixed strip, uniformly over \(a\in G_\tau\), they satisfy \begin{equation}\label{eq:main-asym} \begin{aligned} \rho_a(x)&=1+\varepsilon_aP_2(\cos x)+O(\tau^2),\\ h_a&=h_*+h_1\varepsilon_a+O(\tau^2),\\ \varepsilon_a&=(a-a_*)/a_1+O(\tau^2). \end{aligned} \end{equation} Here \(\varepsilon_a\) is the degree-two Legendre coefficient of \(\rho_a-1\). The parameter \(a\) is half the increment between meridional endpoints of a section in the coordinates of Section~\ref{sec:chords}. \end{theorem} The proof first constructs a polynomial seed and computes the drift of \(a\) from \(a_*\). In chord coordinates the linearization splits into a normal operator of weighted spherical section type, a tangential difference operator, and two scalar equations. Two oscillatory factors of the normal operator are kept exact in the finite-order reduction. The resulting diagonal can be controlled on large closed parameter sets, and its inverse composed with the remainder is small. Section~\ref{sec:iteration} gives the Newton iteration, including the strip allocation and the total measure discarded. The comparison with a ball is proved in Section~\ref{sec:comparison}. \section{The degree-two Legendre resonance}\label{sec:arithmetic} Write $P_l(1)=1$ for the Legendre normalization, and put \[ h_*=3^{-1/2},\qquad a_*=\arccos h_*,\qquad \sigma_*=\sin a_*=\sqrt{2/3}. \] We use Holt's change of variables \cite{Holt} and the degree interpolation approach of Kadets and Mangoubi \cite[Sections~2 and~4]{KM}. The calculation at the specific value $h_*$ determines the zeros in each of four residue classes. Deng's preprint \cite[Theorem~1.1]{Deng} claims the irreducibility of all even Legendre polynomials over \(\mathbb Q\), which would also give the isolation below. The special-height calculation supplies the arithmetic input directly. \begin{lemma}\label{lem:isolation} For every integer $m\ge0$, \[ P_{2m}(1/\sqrt3)=0\quad\Longleftrightarrow\quad m=1. \] Moreover $a_*/\pi$ is irrational, and there exist fixed constants $\gamma_*>0$ and $\tau_*<\infty$ such that \[ |\sin(ja_*)|\ge\gamma_*(1+|j|)^{-\tau_*} \qquad(0\ne j\in\mathbb Z). \] \end{lemma} \begin{proof} Define \[ H_m(z)=\sum_{k=0}^m\frac{\binom mk^2}{\binom{2k}k}z^k. \] Holt's identity \cite{Holt}, in the normalization of \cite[Section~2.2, (3)--(4)]{KM}, gives \[ H_m(z)=\binom{2m}m^{-1}(z-4)^m P_{2m}\!\left(\sqrt{\frac z{z-4}}\right). \] There is no choice of square root in this identity, since $P_{2m}$ is even. In particular \begin{equation}\label{eq:Holt-special} H_m(-2)=\frac{(-6)^m}{\binom{2m}m}P_{2m}(1/\sqrt3). \end{equation} Let $v$ be the $2$-adic Gauss valuation of a polynomial in $u$, normalized by $v(2)=1$, and let $s_2(k)$ count the binary digits equal to one in $k$. For $r=0,1,2,3$ consider \[ F_r(u)=\sum_{k\ge0}T_{r,k}(u),\qquad T_{r,k}(u)=\frac{(-2)^k}{\binom{2k}k}\binom{r+4u}{k}^{\!2}. \] For $k\ge1$, counting the factors divisible by $2$ and by $4$ in $\prod_{j=0}^{k-1}(r+4u-j)$ gives \[ v(T_{r,k})=2S_r(k)-k+s_2(k), \] where \[ \begin{array}{ll} S_0(k)=\lceil k/2\rceil+\lceil k/4\rceil,& S_1(k)=\lfloor k/2\rfloor+\lfloor(k+2)/4\rfloor,\\ S_2(k)=\lceil k/2\rceil+\lfloor(k+1)/4\rfloor,& S_3(k)=\lfloor k/2\rfloor+\lfloor k/4\rfloor. \end{array} \] Indeed a linear factor $4u+c$ has Gauss valuation $\min(2,v_2(c))$, also when $c=0$, while $v_2(k!)=k-s_2(k)$ and $v_2\binom{2k}k=s_2(k)$. For $k=4q+d$ with $0\le d\le 3$, the valuation is $2q+s_2(k)$ plus the entry in the following table. \[ \begin{array}{c|rrrr} r\backslash d&0&1&2&3\\ \hline 0&0&3&2&3\\ 1&0&-1&2&1\\ 2&0&1&0&3\\ 3&0&-1&0&-1 \end{array} \] In particular, the valuations tend to $+\infty$ at least linearly in $k$. Thus each series converges in the Tate algebra $\mathbb Q_2\langle u\rangle$. For a nonnegative integer $j$, it specializes to $F_r(j)=H_{r+4j}(-2)$, since the binomial coefficients truncate. We now determine the zeros on $\mathbb Z_2$ by explicit units. For $r=0$, all terms with $k\ge1$ have valuation at least three. For $r=2$ they have valuation at least one. Consequently \[ F_0\equiv1\pmod8,\qquad F_2\equiv1\pmod2 \quad\hbox{in }\mathbb Z_2\langle u\rangle. \] For $r=3$, the terms with $k\ge4$ have valuation at least three, and $1+T_{3,1}=1-(3+4u)^2$ is divisible by eight. Directly, \[ T_{3,2}+T_{3,3} =\frac4{45}(3+4u)^2(1+2u)^2(7-4u-8u^2). \] It follows that $F_3/4\equiv1\pmod2$. These three functions therefore have no zeros in $\mathbb Z_2$. Finally $1+T_{1,1}=-8u-16u^2$. Each $T_{1,k}$ for $k\ge2$ is divisible by $u^2$. For $k=2,3,4,5$ the quotients $T_{1,k}/(8u^2)$ are respectively \[ \frac{(4u+1)^2}{3},\qquad -\frac{(4u-1)^2(4u+1)^2}{45}, \] \[ \frac{(2u-1)^2(4u-1)^2(4u+1)^2}{315},\qquad -\frac{(2u-1)^2(4u-3)^2(4u-1)^2(4u+1)^2}{14175}. \] All four are integral and congruent to one modulo two. For $k\ge6$, $v(T_{1,k})\ge5$. Dividing by $u^2$ preserves the Gauss valuation. Hence \[ F_1(u)=-8u+16u^2G(u),\qquad G\in\mathbb Z_2\langle u\rangle. \] The factor $1-2uG(u)$ is a unit at every $u\in\mathbb Z_2$, so $F_1$ has exactly the zero $u=0$ there. Combining the four residue classes with \eqref{eq:Holt-special} proves the isolation. For the last assertion, \[ \zeta=e^{2ia_*}=\frac{-1+2i\sqrt2}{3},\qquad \zeta+\zeta^{-1}=-\frac23. \] A root of unity and its inverse are algebraic integers, so their sum cannot be the nonintegral rational number $-2/3$. Hence $a_*/\pi$ is irrational. Choose $q_j\in\mathbb Z$ so that $|ja_*-q_j\pi|\le\pi/2$. With the fixed logarithms $\log\zeta=2ia_*$ and $\log(-1)=i\pi$, the nonzero linear form $j\log\zeta-2q_j\log(-1)$ has coefficients of size $O(1+|j|)$. Matveev's bound \cite[Corollary~2.3]{Matveev} gives a polynomial lower bound for its absolute value. The inequality $|\sin y|\ge 2|y|/\pi$ on $[-\pi/2,\pi/2]$ transfers this bound to $|\sin(ja_*)|$. \end{proof} \section{Formal solutions and the parameter drift}\label{sec:formal} A zonal radial function is written as \(\rho(\omega)=\rho(\omega_1)\) when its argument is a cosine. Write \(A_\rho(u,h)\) for the section area of the radial body. We also write \(P_l\) for the zonal function \(\omega\mapsto P_l(\omega_1)\) or its angular expression \(P_l(\cos x)\). Denote by \([g]_l\) the coefficient of \(P_l\) in a zonal function \(g\), and by \(\av g\) the average of a periodic function in its angle variable. Set \[ f(x)=P_2(\cos x),\qquad \mu=\av f=\frac14,\qquad f^2=\frac15 P_0+\frac27 P_2+\frac{18}{35}P_4. \] The last identity is in the Legendre basis. In particular, \([f]_0=0\) and \([f^2]_0=1/5\). For \(h\) near \(h_*\) define \[ (S_h g)(u)=\int_{S^2}\delta(u\cdot\omega-h)g(\omega)\dd\omega, \qquad S_h P_l=s_l(h)P_l,\qquad s_l(h)=2\pi P_l(h). \] Here \(\delta\) is the Dirac distribution and \(\dd\omega\) is surface measure. The multiplier follows from the Funk--Hecke formula \cite[Theorem~1.2.9]{DaiXu}. In dimension three the coordinate \(t=u\cdot\omega\) gives \(\dd\omega=\dd t\,\dd\varphi\), so the multiplier for a continuous zonal kernel \(p(t)\) is \(2\pi\int_{-1}^1p(t)P_l(t)\dd t\). Approximating \(\delta(t-h)\) gives the displayed multiplier. In particular \(s_2'(h_*)=2\pi\sqrt3\). \begin{lemma}\label{lem:formal} There is a unique formal, even zonal radial series with \([\rho-1]_2=\varepsilon\), beginning at \(\rho=1+\varepsilon f+O(\varepsilon^2)\), such that \(\vol(K_\rho)=4\pi/3\) and \(A_\rho(u,h)=\nu\) as formal series, with \(h(0)=h_*\) and \(\nu(0)=\pi(1-h_*^2)\). All radial coefficients are polynomials in the polar cosine. It satisfies \begin{equation}\label{eq:firstterms} \begin{aligned} h&=h_*+h_1\varepsilon+O(\varepsilon^2),\\ [\rho-1]_0&=-\tfrac15\varepsilon^2+O(\varepsilon^3),\\ [\rho-1]_4&=\tfrac{27}{245}\varepsilon^2+O(\varepsilon^3). \end{aligned} \end{equation} It has a formal meridional reparametrization \(x\mapsto x+\varepsilon w_1(x)+O(\varepsilon^2)\) by odd \(\pi\)-periodic trigonometric polynomials, and a formal parameter \(a\), for which every chord with endpoints labeled \(x\pm a\) has distance \(h\) from the origin. Their first terms are \begin{equation}\label{eq:adrift} a=a_*+a_1\varepsilon+O(\varepsilon^2),\qquad w_1(x)=-\frac{3}{16}\sin(2x). \end{equation} For any positive integer \(N\), truncation and formal inversion of \eqref{eq:adrift} give meridian, height and area seeds depending polynomially on \(a-a_*\), with residual \(O((a-a_*)^{N+1})\) in the chord equations of Section~\ref{sec:chords}. \end{lemma} \begin{proof} The radial formula for the section area, interpreted by coarea, is \begin{equation}\label{eq:radial-delta} A_\rho(u,h)= \int_{S^2}\int_0^{\rho(\omega)} r^2\delta(r\,\omega\cdot u-h)\dd r\dd\omega. \end{equation} The section of the reference sphere is transverse since \(h_*<1\). The boundary of nearby sections can be solved for in polar coordinates centered at \(hu\) in the section plane. For a unit \(v\perp u\), solve \[ \sqrt{h^2+r^2}-\rho\bigl((hu+rv)/\sqrt{h^2+r^2}\bigr)=0 . \] At the ball the derivative with respect to the planar radius is \(r>0\) at \(h^2+r^2=1\). Thus derivatives of \eqref{eq:radial-delta} at the ball may be computed by distributional differentiation of the moving radial upper limit. If \(t=\omega\cdot u\), one differentiation gives \(S_h\). For two directions \(g_1,g_2\), the kernel at the second differentiation is \(2\delta(t-h)+t\delta'(t-h)\). Using \((t-h)\delta'(t-h)=-\delta(t-h)\) gives \begin{equation}\label{eq:second-var} D_\rho^2 A_{1}(u,h)[g_1,g_2] =(S_h-hS_h')(g_1g_2)(u). \end{equation} Here and below a prime on \(S_h\) denotes differentiation in the physical height \(h\). At any finite higher order the kernel in \eqref{eq:radial-delta} is a finite linear combination of derivatives \(\partial_h^j\delta(t-h)\) with scalar coefficients depending on \(h\). This follows by repeatedly differentiating in \(r\) and using \((t-h)\delta^{(j)}(t-h)=-j\delta^{(j-1)}(t-h)\). The spherical multipliers therefore preserve polynomial inputs. Subtract the spherical mean of the section equation and write \(\rho=1+\varepsilon f+\varepsilon^2g+\cdots\) and \(h=h_*+h_1\varepsilon+\cdots\). The order-two equation is the nonconstant part of \begin{equation}\label{eq:quadratic} S_{h_*}g+\frac12(S_{h_*}-h_*S'_{h_*})(f^2) +h_1 S'_{h_*}f=0 . \end{equation} Its \(P_2\) coefficient is \(s_2'(h_*)(h_1-h_*/7)\). The volume condition \(\frac13\int_{S^2}\rho^3\dd\omega=4\pi/3\) gives \([g]_0=-[f^2]_0=-1/5\). In degree four, inserting \(P_4(h_*)=-7/18\) and \(h_*P'_4(h_*)=-5/9\) in \eqref{eq:quadratic} gives \([g]_4=27/245\). For order \(j\ge2\), the new radial coefficient \(g_j\) enters the nonconstant equation by \(S_{h_*}g_j\). The as yet undetermined height coefficient of order \(j-1\) enters by a scalar multiple of \(S_{h_*}'f\). The height coefficient of order \(j\) contributes only to the constant section coefficient. Lemma~\ref{lem:isolation} solves each noncritical even Legendre equation. The critical equation determines the height coefficient of order \(j-1\) since \(s'_2(h_*)\ne0\). Volume determines \([g_j]_0\), while \([g_j]_2=0\) fixes the chosen amplitude. Once the radial and height series are known, the spherical mean gives the area series \(\nu\). This proves the induction. Put \(e_r(x)=(\cos x,\sin x)\), \(e_\theta(x)=(-\sin x,\cos x)\), and let \(T_ag(x)=g(x+a)\). For the reference chord at \(x\), the normal is \(e_r(x)\) and the midpoint is \(h_*e_r(x)\). Its first-order distance variation is the average of the normal components of the two endpoint variations. The radial variations give \(\tfrac{h_*}{2}(T_{a_*}+T_{-a_*})f\). A tangent vector at \(x+a_*\) has scalar component \(-\sigma_*\) along \(e_r(x)\), and the tangent vector at \(x-a_*\) has component \(\sigma_*\). The change in the two labels is \(\pm a_1\varepsilon\). The distance equation at order one is consequently \[ \frac{h_*}{2}(T_{a_*}+T_{-a_*})f -\frac{\sigma_*}{2}(T_{a_*}-T_{-a_*})w_1 -\sigma_*a_1=h_1 . \] Its mean gives \(a_1=(h_*\mu-h_1)/\sigma_*=3/(28\sqrt2)\). For the mode \(\cos(2x)\), the equation uses \(f=1/4+(3/4)\cos(2x)\), \(\cos(2a_*)=-1/3\), and \((T_{a_*}-T_{-a_*})\sin(2x)=2\sin(2a_*)\cos(2x)\). It gives the stated \(w_1\). At each order the operator on the new reparametrization and on the new angle coefficient is the same linear difference operator and constant term above. The forcing is a finite even, \(\pi\)-periodic Fourier sum because the distance is obtained from the endpoints by dot products, a square root with nonzero reference value, and division by the chord length. The arithmetic lemma inverts the difference operator from odd functions onto the even functions of zero mean, frequency by frequency. The constant equation gives the angle coefficient. Since \(a_1>0\), the formal inversion is possible. Taylor truncation gives the last assertion. At fixed order the analytic functions in the chord equations have a genuine Taylor expansion near the reference configuration, as also shown by the integral formulas in Section~\ref{sec:chords}. \end{proof} \section{Chord coordinates and the section derivative}\label{sec:chords} \subsection{Coordinates and analytic norms} We use a plane through the axis of revolution for meridians. Let \(\gamma(x)=(X(x),Y(x))\) be close to \(e_r(x)\), with \(X\) even, \(Y\) odd, and \(\gamma(x+\pi)=-\gamma(x)\). On \((0,\pi)\) one has \(Y>0\) for meridians sufficiently close in \(C^1\). Rotating \((X,Y)\) about the first axis gives the radial body \(K_\gamma\). Real meridians close enough in \(C^2\) give convex bodies with positive principal curvatures, as verified at the end of Section~\ref{sec:iteration}. In a fixed neighborhood of the reference meridian use the definitions \begin{align} \gamma_\pm(x)&=\gamma(x\pm a),& L&=|\gamma_+-\gamma_-|,& n&=\frac{((\gamma_+-\gamma_-)_2,-(\gamma_+-\gamma_-)_1)}{L}, \label{eq:chord-geometry}\\ q&=n\cdot\gamma_+=n\cdot\gamma_-,& S&=|\gamma'|,& N&=(Y',-X')/S . \notag \end{align} Square roots in the complex domain use the analytic branches at the reference values \(L=2\sin a\), \(S=1\). Write \(n=(\cos\theta,\sin\theta)\) with \(\theta-x\) small, and put \(n^\perp=(-\sin\theta,\cos\theta)\). Thus \(\gamma_+-\gamma_-=L n^\perp\). The section is taken with the three-dimensional normal \((n_1,n_2,0)\). Let \(A_\gamma(x)\) be its area. The equation for \(z=(\gamma,h,\nu)\) is \begin{equation}\label{eq:F} \begin{aligned} F(z;a)&=(F_q,F_A,F_W)=(q-h,A_\gamma-\nu,W(\gamma)-4\pi/3),\\ W(\gamma)&=-\pi\int_0^\pi Y^2X'\dd u . \end{aligned} \end{equation} We write displacements as \(\delta\gamma=\psi N+w\gamma'\), with \(\psi\) even and \(w\) odd, both \(\pi\)-periodic. The volume derivative is \begin{equation}\label{eq:vol-der} \delta W(\psi)=2\pi\int_0^\pi\psi(u)S(u)Y(u)\dd u. \end{equation} Indeed a tangential displacement has zero normal speed, and the area element of the surface of revolution is \(S Y\dd u\dd\varphi\). Translations of the chord labels keep \(a\) fixed during differentiation in \(z\). The equations of the preceding section use \(a\) as the external parameter. Fix a small positive upper strip width \(r_0\), and an integer \(s_0\ge2\). For a complex, \(2\pi\)-periodic function \(g(x)=\sum_m g_m e^{imx}\) put \[ \|g\|_{r,s}=\sum_m e^{r|m|}(1+|m|)^s|g_m|,\qquad 0q\}} (z\cdot n-q)(n\cdot N)\dd S,\\ C_\gamma(g)&=\int_{\partial K_\gamma\cap\{z\cdot n>q\}}g\dd S,\\ q A_\gamma&=C_\gamma(z\cdot N)-3V_\gamma. \end{align*} In the first identity apply the divergence theorem to \((z\cdot n-q)n\). The planar face gives zero flux. For the last identity apply it to the field \(z\), whose flux on the planar face of the cap is \(-qA_\gamma\). In these formulas the revolved unit normal at a boundary point is \((N_1,N_2\eta_{\rm old},N_2\eta_{\perp,\rm old})\). In particular \(z\cdot N=XN_1+YN_2\). Set \(t=s-\xi y\). The transformed expressions use \[ J=e-ot,\qquad n\cdot N=n_1N_1+ \frac{n_2N_2}{\sin x\sin u}\,\frac{et-H^2o}{J}. \] The quotient \(n_2N_2/(\sin x\sin u)\) is analytic by parity. The cap restriction is the fixed spherical cap \(s>h_0\). Here is a direct bound for the cap integrals on angular strips. The reference density after the substitution is \(b_0\sqrt\chi/J\), where \(b_0=SY/\sin u\). For each real \(s\in[h_0,1]\) the quantities \(t=s-\xi y\), \(H^2=(1-\xi^2)(1-y^2)\) are bounded on a fixed pair of complex strips. The denominator \(J\) is uniformly close to one there. Hence the transformed amplitudes, including the factor \(\chi(s-h_0)/J\) in \(V_\gamma\), have double Chebyshev series with coefficient estimates uniform in \(s\). Their shape derivatives are obtained by the product and quotient rules. For a term \(g_{pq}(s)T_p(\xi)T_q(y)\) in such a series, the cap integral is \(T_p(\xi)\) times the zonal transform of \(T_q(y)\) with kernel \(1_{[h_0,1]}(s)g_{pq}(s)\). On degree \(l\le q\) its multiplier is \(2\pi\int_{h_0}^1g_{pq}(s)P_l(s)\dd s\). Its size is at most \(C\|g_{pq}\|_\infty\). A parameter difference of the multiplier is bounded by the difference of the amplitudes and the length of the interval between the two lower limits times their supremum. The same reasoning for a fixed spherical section uses the value at \(h_0\), multiplied by \(2\pi P_l(h_0)\), and a fixed number of derivatives in \(s\) or \(h_0\) costs a fixed polynomial in \(1+l\). Converting between Chebyshev and Legendre coefficients costs only powers of the degrees as established above. Thus the sum of the transformed integrals on the angular strip has bounds with only polynomial losses in the strip gap. As a consequence there exist fixed \(C,P\) such that \begin{equation}\label{eq:Taylor} \|F(z+v)-F(z)-D_zF(z)v\|_{r-\delta}^{\#} \le C\delta^{-P}(\|v\|_r^{\#})^2 . \end{equation} This holds with the displayed geometric denominators bounded away from their zeros on the intermediate strips. One can obtain the estimate by taking the second derivative along \(z+tv\) and integrating with weight \(1-t\). For the parameter difference, take the difference of these second derivatives at the two parameters. The estimate is bounded by the two terms in which one \(v\) is differenced and by the third shape derivative times the difference of the base meridians, together with the translated-function and cap-limit differences above. The base need only have bounded coefficientwise family norm. A finite division of the gap \(\delta\) among the estimates gives a fixed exponent \(P\). The same proof gives a local increment estimate for \(b_\gamma\) in the double Chebyshev norm. When estimates are centered at an already controlled tuple of quotients and other geometric factors, their increments are bounded by \(C\delta^{-P}\|\Delta\gamma\|_r^{\#}\) on the smaller strip. For an inverse or square root this follows by expanding about the old value under the condition \(\delta^{-P}\|\Delta\gamma\|_r^{\#}\) sufficiently small. This local form will be used for shrinking strips. The physical height derivative \(A_h\), at fixed normal, has the same bounds. Indeed translating the entire body in the direction \(n\) changes the effective height from \(q\) to \(q-t\). Its normal speed is \(n\cdot N\), giving \begin{equation}\label{eq:height-derivative} A_h(x)=-\int_{\partial K_\gamma}(n\cdot N) \delta(z\cdot n-q)\dd S . \end{equation} The transformed amplitude contains the analytic expression for \(n\cdot N\) above, evaluated at \(s=h_0\). In particular \(A_h\) can be computed without taking a derivative in \(a\). \subsection{Splitting the linearization} \begin{lemma}\label{lem:approx} Set \(q_\theta=q'/\theta'\), \(t_\pm=n^\perp\cdot\gamma_\pm-q_\theta\), and \(c_q=\theta't_+t_-/L\). For a displacement \(\delta\gamma=\psi N+w\gamma'\) put \[ \zeta_\psi=-\frac{t_-}{L}n\cdot N_+\psi_+ +\frac{t_+}{L}n\cdot N_-\psi_- . \] After subtracting \(A_h\,D_zF_q\) from the section row, the derivative is the sum of \begin{equation}\label{eq:approx} \mathcal L_z(\psi,w,\delta h,\delta\nu) =\bigl(\zeta_\psi+c_q(T_a-T_{-a})w-\delta h, \mathcal T_\gamma\psi+A_h\delta h-\delta\nu, \delta W(\psi)\bigr) \end{equation} and the error \begin{equation}\label{eq:residual-error} \left(\frac{F_q'}{\theta'}\delta\theta, \frac{F_A'-A_hF_q'}{\theta'}\delta\theta,0\right). \end{equation} At the circle \(c_q=-\sin a/2\) and \(\zeta_\psi=\tfrac{\cos a}{2}(T_a+T_{-a})\psi\). The angle variation satisfies \(\delta\theta=n\cdot(\delta\gamma_--\delta\gamma_+)/L\). \end{lemma} \begin{proof} The equations \(n\cdot\gamma_\pm=q\) give \[ \delta q-q_\theta\delta\theta =n\cdot\delta\gamma_\pm+t_\pm\delta\theta,\qquad n\cdot\gamma_\pm'=-\theta't_\pm . \] Taking the difference of the two equations and using \(t_+-t_-=L\) gives the formula for \(\delta\theta\). Eliminating \(\delta\theta\) from their common value gives \(\delta q-q_\theta\delta\theta=\zeta_\psi+ c_q(T_a-T_{-a})w\). With height and normal varied independently the section derivative is \(\mathcal T_\gamma\psi+A_h\delta q+A_\theta\delta\theta\). Differentiating the section area with respect to \(x\) gives \(A_\theta=(A_\gamma'-A_hq')/\theta'\). Subtracting the row \(A_hD_zF_q\) gives the formulas. The circle has \(\theta'=1\), \(t_\pm=\pm\sin a\), and \(n\cdot N_\pm=\cos a\), proving the reference values. \end{proof} \section{Finite-order diagonal reduction}\label{sec:reduction} \subsection{Matrices on the Legendre coefficients} We use the Legendre coefficient model \[ \mathcal H P_l=e^{ik_lx},\qquad k_l=l+\tfrac12,\qquad De^{ik_lx}=k_le^{ik_lx}. \] The index set is initially \(l\ge0\). A periodic multiplier \(b(x)=\sum_m b_m e^{imx}\) gives the Toeplitz matrix \((T_b)_{jl}=b_{j-l}\) on this half-lattice. The notation \(T_a\) with scalar angle \(a\) still denotes translation on periodic functions. For a matrix \(R\), define \begin{equation}\label{eq:matrixnorm} |R|_{r,s;N} =\sup_{l\ge0}(1+l)^N\sum_{j\ge0} e^{r|j-l|}(1+|j-l|)^s|R_{jl}|, \qquad N\ge0 . \end{equation} Finite norm for \(N=0\) implies boundedness on the Legendre coefficient norm \(\|\cdot\|^L_{r,s}\). For every \(\delta>0\) with \(r-\delta>0\), \begin{align} |AB|_{r,s;N}&\le |A|_{r,s;0}|B|_{r,s;N},\\ |BA|_{r-\delta,s;N} &\le C_N\delta^{-N}|B|_{r,s;N}|A|_{r,s;0} \qquad(0<\delta\le1). \label{eq:right-product} \end{align} Indeed, in the second product the intermediate column index \(i\) of \(B\) can be replaced by the input index \(l\) using \((1+l)/(1+i)\le 1+|l-i|\). Absorb the resulting factor \((1+|i-l|)^N\) using \(e^{-\delta|i-l|}\). The first product leaves the input index unchanged. In particular, \(|\cdot|_{r,s;0}\) is submultiplicative. Each assertion also holds with the coefficientwise family norm. Throughout this section the exponents in losses can depend on the chosen finite orders and on \(s_0\). \begin{lemma}[Polynomial multiplication and section asymptotics]\label{lem:calculus} For a polynomial \(B_0\) of degree \(p\), on input indices \(l\ge p\) the exact multiplication formula is \begin{equation}\label{eq:exact-multiplication} \mathcal H M_{B_0(\cos x)}\mathcal H^{-1} =\sum_{j=0}^{p}\binom{-1/2}{j} (-iM_{\sin x}D^{-1})^jM_{B_0^{(j)}(\cos x)} . \end{equation} Here \(M_b\) means multiplication in the displayed Fourier representation. Lower columns are handled by the actual polynomial multiplication matrix on the half-lattice. For an analytic amplitude \(b(\xi,y)\) with a double Chebyshev expansion, let \(\mathcal T_b\) be the operator in \eqref{eq:Tgamma} with amplitude \(b\), and put \[ \begin{gathered} V_\pm e^{ik_lx}=e^{\pm ik_la} e^{ik_lx},\qquad A(a)=\sqrt{2\pi/\sin a},\\ B_b=A(a)^{-1}\mathcal H\mathcal T_b\mathcal H^{-1}D^{1/2}. \end{gathered} \] For any fixed \(J\ge2\), this matrix has an expansion \begin{equation}\label{eq:branch-exp} B_b=\sum_{\sigma=\pm1}\sum_{j=0}^{J-1} T_{b_{\sigma,j}}D^{-j}V_\sigma+R_J. \end{equation} The leading profiles are \begin{equation}\label{eq:principal} b_{\sigma,0}(x)=e^{-\sigma i\pi/4} b(\cos x,\cos(x+\sigma a)). \end{equation} Every profile and the remainder, the latter measured with \(N=J-1\) in \eqref{eq:matrixnorm}, satisfy linear bounds in the double analytic norm of \(b\) with fixed polynomial strip losses, including in the coefficientwise family norms. The same expansion and remainder estimates hold for finite products involving Toeplitz matrices and fixed nonpositive powers of \(D\), with the phase in a product kept exact. \end{lemma} \begin{proof} For the algebraic identity, use the usual Gegenbauer polynomials \(C_l^\lambda\) and write \(\mathcal H_\lambda C_l^\lambda=e^{i(l+\lambda)x}\). Off the lower boundary the recurrence in \cite[Section~18.9]{DLMF} gives \[ \mathcal H_\lambda M_{\cos x}\mathcal H_\lambda^{-1} =M_{\cos x}-i(\lambda-1)M_{\sin x}D^{-1}. \] In fact the coefficients of the upward and downward transfers are \((k-\lambda+1)/(2k)\) and \((k+\lambda-1)/(2k)\) with \(k=l+\lambda\). For positive integers \(m\), use \[ (C_l^m)'=2m C_{l-1}^{m+1},\qquad C_l^m=\frac{m}{l+m}(C_l^{m+1}-C_{l-2}^{m+1}). \] These are the derivative and parameter change identities in \cite[Section~18.9]{DLMF}. For \(l\ge1\), converting differentiation in the polynomial variable by \(\mathcal H_{m+1}\) gives \(2m\mathcal H_m\). The second identity gives \(\mathcal H_{m+1}\mathcal H_m^{-1}=2imM_{\sin x}D^{-1}\). Write \(A_B^m=\mathcal H_m M_B\mathcal H_m^{-1}\). Apply the product rule to the derivative of \(Bf\) and these two conversion formulas to get, for sufficiently high inputs, \[ A_B^{m+1}=A_B^m-iM_{\sin x}D^{-1} A_{B'}^m . \] At \(m=1\) the matrix for multiplication by the coordinate is ordinary Toeplitz multiplication away from the lower boundary, so the same holds for a polynomial on sufficiently high inputs. Induction with Pascal's identity gives the right side of \eqref{eq:exact-multiplication} with \(\binom{m-1}{j}\). For each transfer, both sides are polynomials in \(\lambda-1\) with coefficients rational in \(k\), by the displayed three-term matrix and the finite formula on the right. At any fixed positive integer \(\lambda\), equality holds for infinitely many \(k\), so the rational functions agree. Equality at all positive integers \(\lambda\) identifies their polynomial coefficients. At \(\lambda=1/2\) the three-term matrix applies on every path from \(l\ge p\), since the path has length at most \(p\). This gives the exact identity there and \(C_l^{1/2}=P_l\). We detail the estimates needed to use this identity for analytic amplitudes. For the Chebyshev polynomial \(T_p\), the Markov inequality on \([-1,1]\), applied repeatedly, gives \(\|T_p^{(j)}\|_\infty\le p^{2j}\) for \(j\ge1\). The Fourier sum of \(T_p^{(j)}(\cos x)\) is bounded by \((2p+1)p^{2j}\) and has bandwidth at most \(p-j\). For \(p\le c\sqrt{k_l}\) with a sufficiently small constant \(c>0\), every frequency in \eqref{eq:exact-multiplication} is at least \(k_l/2\). In the column sum with the weights of \eqref{eq:matrixnorm} at \(N=0\), the terms with \(j\ge J\) are bounded by \begin{equation}\label{eq:markov-tail} C_J e^{rp}(1+p)^{s+1} k_l^{-J}(1+p)^{2J}. \end{equation} Indeed the absolute value of \(\binom{-1/2}{j}\) is at most one, and the remaining geometric series has ratio at most \(2p^2/k_l\le 1/2\). For the terms \(jp . \] The lower-bandwidth bound follows by moving \(T_p\) to the other side of the inner product. Thus the column sum has a polynomial bound in \(1+p\), times \(e^{rp}\), uniformly in \(l\). A truncated multiplication matrix has the same type of bound, using \(k_l\ge1/2\). If \(p>c\sqrt{k_l}\), then \(k_l^N\le C_N(1+p)^{2N}\) for any fixed \(N\). After a strip gap \(\delta\), all the additional powers are absorbed using \(\sup_{p\ge0}e^{-\delta p}(1+p)^d\le C_d\delta^{-d}\). In particular the columns where a projected product meets the lower boundary are controlled by this estimate. For the middle factor of the section matrix, the Legendre asymptotics in \cite[Section~18.15(iii), (18.15.12)--(18.15.13)]{DLMF} give, for \(k=l+1/2\), \begin{equation}\label{eq:legendre-asym} d_l^*(a):=A(a)^{-1}2\pi k^{1/2}P_l(\cos a) =\sum_{\sigma=\pm1} e^{\sigma ika} \sum_{j=0}^{J-1}p_{\sigma,j}(a)k^{-j} +r_J(k,a). \end{equation} Here \(p_{\sigma,0}=e^{-\sigma i\pi/4}\) and \(p_{-,j}=\overline{p_{+,j}}\). The remainder satisfies \(|r_J|\le C_J k^{-J}\) and \(|\partial_a r_J|\le C_J k^{1-J}\) uniformly on a fixed compact interval about \(a_*\). For the expansion, in the \(m\)-th term of the cited formula write \[ (-1)^l\binom{m-1/2}{l} =\frac{\Gamma(l+1/2-m)} {\Gamma(1/2-m)\Gamma(l+1)}. \] The fixed-shift Stirling expansion \cite[Section~5.11(iii)]{DLMF} of the ratio has leading power \(l^{-m-1/2}\). The factor \((-1)^l\) cancels that of the cosine in (18.15.13), leaving phases \(\pm k_l a\) and smooth coefficients. Expanding the powers of \(l=k_l-1/2\) gives \eqref{eq:legendre-asym}. To obtain the derivative statement from the undifferentiated expansion, one can first use one additional order at \(l\) and \(l-1\) in \[ \frac{\dd}{\dd a}P_l(\cos a) =\frac{l}{\sin a}\bigl(\cos a,P_l(\cos a)-P_{l-1}(\cos a)\bigr). \] There is also a direct remainder argument. Denote the finite expansion in \eqref{eq:legendre-asym} by \(F_J\). The recurrence for the derivative gives an expansion for \((d_l^*)'-F_J'\) with error \(O(k^{1-J})\), whose possible terms of larger size are finite sums of smooth coefficients times \(k^p e^{\sigma ika}\). Integrate this difference on \([a,a+b/k]\) for a fixed \(01-J\), the leading terms in the integral are \(k^{p-1}e^{\sigma ika}\) times their coefficients and \((e^{\sigma ib}-1)/(\sigma i)\). Use two different choices of \(b\) in \((0,\pi)\) to obtain an invertible two-by-two matrix. Letting \(k\to\infty\) forces both coefficients to vanish. Continuing in decreasing powers proves the derivative estimate on an interior compact interval. The leading factor \(A(a)\) follows from the cosine term of the standard expansion. For an amplitude term \(b_{pq}T_p(\xi)T_q(y)\) the exact operator is \[ b_{pq}M_{T_p} S_{\cos a}M_{T_q}. \] With \(p+q\le c\sqrt{k_l}\), insert the multiplication expansions on both sides of \eqref{eq:legendre-asym}. If the right multiplication transfers by \(\nu\), the phase is exactly \(e^{\sigma i(k_l+\nu)a}\), and the power normalization includes \((k_l/(k_l+\nu))^{1/2}\). Expand this ratio and all the inverse powers of \(k_l+\nu\) in powers of \(k_l^{-1}\). The resulting leading right multiplier is \(T_q(\cos(x+\sigma a))\). The nonleading terms and their remainders are bounded by the appropriate power of \(k_l^{-1}\) times the weight \(e^{r(p+q)}(1+p+q)^d\). Differencing an external phase costs at most one power of \(k_l+|\nu|\). All profile differences have polynomial transfer bounds. If \(p+q>c\sqrt{k_l}\), use the exact multiplication bounds, the bounded normalized diagonal \(d_l^*\), and \((k_l/(k_l+\nu))^{1/2}\le C(1+p+q)^{1/2}\). Here one may trade \(k_l^N\) for \(C_N(1+p+q)^{2N}\). The same estimate applies to each truncated term. Summing against the double Chebyshev coefficients with one strip gap gives \eqref{eq:branch-exp} and \eqref{eq:principal}, with \(N=J-1\) covering the loss of one input power in the family norms. For use in the finite recursion, the composition estimate can also be seen directly from one matrix entry. For a product \((T_bD^{-j}V_\sigma)(T_cD^{-m}V_\kappa)\), allow \(V_0=I\). The contribution of transfers \(\nu\) on the right and \(\mu\) on the left to column \(l\) has coefficient \begin{equation}\label{eq:compose} b_\mu c_\nu (k_l+\nu)^{-j} k_l^{-m} e^{i\sigma\nu a}e^{i(\sigma+\kappa)k_la}. \end{equation} It is present when the intermediate and final indices belong to the half-lattice. For \(|\nu|\tau_*+1\) and sufficiently small fixed \(\gamma_0>0\). The operator \(T_a-T_{-a}\) on mean-zero, \(\pi\)-periodic functions has multiplier \(2i\sin(ma)\). Its inverse costs a fixed number of powers of \(1+|m|\). Its parameter differences cost at most \(C(1+|m|)^{2\tau_h+1}\), using the product of the two lower bounds for the denominators. We use the inverse between the opposite reflection parities whenever the data have a given parity. \begin{proposition}[Finite reduction]\label{prop:normalform} For \(a\in\mathcal G_h\) and \(b=b_\gamma\) near the reference amplitude of Lemma~\ref{lem:fixed-cut}, there are invertible order-zero matrices \(T_w,T_{1/e},Q,E\), each close to \(I\), a real diagonal \(\mathcal D\) and a matrix \(R\) such that \begin{equation}\label{eq:normalform} T_{1/e}\,B_b\,T_w Q=E\mathcal D+R . \end{equation} They preserve the real, even Legendre coefficient sequences. At \(\gamma=e_r\), these four matrices are \(I\), \(R=0\), and the diagonal entries are \(d_l^*(a)\) from \eqref{eq:legendre-asym}. On a smaller strip \(R\) has finite norm in \eqref{eq:matrixnorm} with \(N=8\). The construction uses a fixed number of operations on the coefficients. Starting with bounded old reduction data on a strip of width \(r\), a meridian increment of norm \(\eta\) there changes the data by at most \(C\delta^{-P}\eta\) on width \(r-\delta\), provided \(\delta^{-P}\eta\) is small and the old data satisfy the nonvanishing margins. The data include all finite profiles, the matrices in \eqref{eq:normalform}, and the order-eight remainder. These bounds hold for \(\#\) norms. For all sufficiently large \(l\) the entries have the exact form \begin{equation}\label{eq:branches} d_l=e^{ik_la}M_+(k_l,a)+e^{-ik_la}M_-(k_l,a), \qquad M_- =\overline{M_+}. \end{equation} The same increment bound holds for \(\sup_{k} k|\Delta M_+(k,a)|_G^\#\). \end{proposition} \begin{proof} Put \(\beta_\sigma(x)=b(\cos x,\cos(x+\sigma a))\). By Lemma~\ref{lem:fixed-cut} these functions are positive on the real line, \(\pi\)-periodic, and \(\beta_-(x)=\beta_+(-x)\). Solve \[ \begin{gathered} (T_a-T_{-a})v=\log\beta_- -\log\beta_+,\qquad \av v=0,\\ w=\exp(v),\qquad e=\beta_+T_aw=\beta_-T_{-a}w . \end{gathered} \] The right side is odd, so \(v\) is even. Uniqueness, reality and the reflection relation show that \(w,e\) are real and even. They are close to one on a smaller strip, where their Toeplitz matrices and \(T_{1/e}\) are invertible by a Neumann series. The leading profiles of \(C_b=T_{1/e}B_b T_w\) are therefore the constants \(c_\pm=e^{\mp i\pi/4}\). This follows from the order-zero case of \eqref{eq:compose}. At the reference amplitude \(b=1\), \(C_b\) is the exact diagonal \(d_l^*\). We perform the recursion through degree \(J-1\) in \(D^{-1}\), where \(J=16\) suffices. At step \(j\ge1\) write \(Q_{j\). Set \begin{align*} m_{\sigma,j}&=\av F_{\sigma,j}, \qquad \av q_j=0,\\ (T_a-T_{-a})q_j &=-F_{+,j}+F_{-,j}+\av(F_{+,j}-F_{-,j}),\\ e_j&=F_{+,j}+T_aq_j-m_{+,j}. \end{align*} These equations cancel the profiles at order \(j\) in both branches. At the circle \(q_j=e_j=0\) for all \(j\), and \(c_\sigma m_{\sigma,j}^*=p_{\sigma,j}\). For the exact diagonal retain the spherical eigenvalue and add only the changes computed in the recursion: \begin{equation}\label{eq:exact-d} d_l=d_l^*+ \sum_{\sigma=\pm1}c_\sigma e^{\sigma ik_la} \sum_{j=1}^{J-1}(m_{\sigma,j}-m_{\sigma,j}^*)k_l^{-j}. \end{equation} This definition applies to all \(l\), including the low modes. For large \(k_l\), split the real reference remainder \(r_J(k_l,a)\) of \eqref{eq:legendre-asym} by assigning \(r_J(k_l,a)e^{-ik_la}/2\) and its conjugate to the two branches. This gives \eqref{eq:branches}, with reference positive branch \(M_+^*=e^{-i\pi/4}+O(k^{-1})\) and \(\partial_a M_+^*=O(k^{-1})\). The new branch terms begin at \(k^{-1}\). Reality here is conjugation of the coefficients at the same transfer: \(F_{-,j,\nu}=\overline{F_{+,j,\nu}}\). This relation is preserved by \eqref{eq:compose}, with the real initial matrix and the two opposite phases. The difference in the equation for \(q_j\) is purely imaginary at each nonzero transfer and division uses \(2i\sin(\nu a)\). Thus the coefficients of \(q_j\) are real and the two formulas for \(e_j\) give conjugate coefficients, so that its coefficients are real too. The joint sign symmetry of \(b\) implies that only even transfers occur. In particular the inversion at every step uses exactly \eqref{eq:homogeneous}. Define \(R\) by \eqref{eq:normalform}. The canceled coefficients and Lemma~\ref{lem:calculus} give the order-eight bound, including the low columns. To see the local difference estimate, expand the logs and exponentials about the old bounded profiles using the algebra inequality. Apply the homogeneous inverse once to their increment. For each following recursion the coefficient operations are the finite convolutions of \eqref{eq:compose}, derivatives, translations and the homogeneous inverse. A difference in a product is a sum of products with one differenced factor. Apply the strip estimates to these factors, using the old norms and the smallness condition for the increments. For the matrix remainders, use the actual remainders in the high- and low-transfer estimates of Lemma~\ref{lem:calculus}. Each is multilinear in its profiles before taking the norms. This also controls the change in \(R\) after cancellation: one subtracts the retained coefficients before estimating the difference so that the first omitted input power is retained. The change introduced by replacing the truncated reference diagonal with the exact one is \(E\) times a diagonal remainder of order at least \(J-1\) in the family norm. The first inequality in \eqref{eq:right-product} preserves that order. Finite initial columns are estimated by the same polynomial bounds on matrix entries. Taking the same fraction of the gap for each of the finitely many operations gives one fixed exponent \(P\). All profiles, the exact diagonal and the remainders are defined by coefficient formulas independent of the strip used for estimation. This is why the increment estimate can be centered at old data when the width is decreased. \end{proof} If the perturbative data of Proposition~\ref{prop:normalform} are \(O^{\#}(\tau)\), then for all sufficiently large \(l\), \begin{equation}\label{eq:phase} \begin{gathered} d_l=2A_l\cos(k_la+\phi_l),\qquad 0\tau_*+1\). For the second assertion extend the real \(\phi_l\) linearly across complementary intervals of \(G\), and constantly to the endpoints of \(I_\tau\) if needed. The Lipschitz constant is still at most \(C/k_l\). Thus \(k_la+\phi_l(a)\) is increasing, with difference quotients in \([k_l/2,2k_l]\) for large \(l\). The phase range meets at most \(C(k_l\tau+1)\) neighborhoods of zeros of cosine. By the lower bound on \(A_l\), each has phase length at most \(C\Gamma(1+l)^{-2}\), hence parameter length at most \(C\Gamma k_l^{-1}(1+l)^{-2}\). This proves \eqref{eq:one-measure}. Only the real phase is extended for this estimate. \end{proof} \subsection{The iteration} \begin{proof}[Proof of Theorem~\ref{thm:main}] We first fix the arithmetic constants in \eqref{eq:homogeneous}, the finite reduction order, the index \(s_0\), and a small positive width \(r_0'\) with a larger initial strip available. The first formal perturbation in \eqref{eq:adrift} has size \(O^\#(\tau)\). The fixed-gap difference estimates in Sections~\ref{sec:chords} and~\ref{sec:reduction}, and Lemma~\ref{lem:schur}, give a bound \(C_0\tau\) for the perturbed data on width \(r_0'\). The data kept at a step consist of the geometric denominators and factors in Section~\ref{sec:chords}, including \(A_h\), the double amplitude, the coefficients and matrices of the finite reduction, and the Schur rows, columns and remainder. In particular we include \(B_b\) in its order-zero matrix norm. Its increment bound is the order-zero case of the exact multiplier estimates in Lemma~\ref{lem:calculus}. It therefore gives an estimate at the working width, while \(D^{-1/2}\) and the changes of coefficient basis give the stated strip estimates for \(\mathcal T_\gamma\) in angular coordinates. The reference values are taken at the circle with the same \(a\). The bound \(C_0\tau\) also holds for higher-order formal seeds for sufficiently small \(\tau\), since their remaining terms have size \(O^\#(\tau^2)\) on the larger initial strip, for any fixed seed order. Increase \(C_0\) once to include the leading meridian perturbation. Fix \(M\) as in Proposition~\ref{prop:normal-inverse} to handle the bound \(2C_0\tau\). Fix the strip losses and the large exponent used below, then choose the order \(N\) of the formal seed. Finally shrink \(\tau\). Let \(z_0=(\gamma_0,h_0^{\mathrm{seed}},\nu_0)\) be this seed, expressed as a polynomial in \(a-a_*\). By Lemma~\ref{lem:formal}, its residual on width \(r_0'\) is \(O^\#(\tau^{N+1})\). Indeed a Taylor remainder of order \(N+1\) on a fixed neighborhood has a first parameter derivative of order \(N\), which is multiplied by \(\tau\) in the family norm. The seed is in the regime \eqref{eq:wedge} with \(\varepsilon=(a-a_*)/a_1\). For \(j\ge0\) set \begin{equation}\label{eq:thresholds} \Gamma_j=M\tau(1+2^{-j}). \end{equation} On \(\mathcal G_h\) form the seed diagonal \(d_{0,l}\) and define the compact set \[ G_0=\{a\in\mathcal G_h: |d_{0,l}(a)|\ge \Gamma_0(1+l)^{-2}\text{ for all even }l\ge4\}. \] Every fixed even \(l\ge4\) is safe for all sufficiently small \(\tau\), by Lemma~\ref{lem:isolation} and \eqref{eq:exact-d}. For the larger indices, the right side of \eqref{eq:one-measure}, divided by \(\tau\), is bounded by \(CM((1+l)^{-2}+(1+l)^{-3})\). This sequence is summable. Combining the fixed-index observation, this tail estimate, and Lemma~\ref{lem:measure} gives \begin{equation}\label{eq:seed-measure} |I_\tau\setminus G_0|=o(\tau). \end{equation} Take \(\delta_j=d/(j+1)^2\), with \(d>0\) so small that \(4\sum_j\delta_j4\) be fixed and large. Since \(\log B_j\le C'+P\log(1/\tau)+2P\log(j+1)\), \[ \sum_{j=0}^\infty 2^{-j-1}\log B_j \le C''+P\log(1/\tau),\qquad \sup_j 2^{-j}\log B_j\le C''+P\log(1/\tau). \] Taking logarithms in \eqref{eq:Newton}, or replacing \(e_j\) by its positive recursive upper bound when it vanishes, gives \[ \log e_j\le 2^j\left(\log e_0+ \sum_{m0\) and \(X'/\sin x<0\) by parity, division at the simple zeros and closeness to the circle. The signed meridian curvature \((X'Y''-Y'X'')/S^3\) is positive by \(C^2\)-closeness. Thus the radius \(Y\) as a function of the axial coordinate \(X\) is concave: its second derivative is \((X'Y''-Y'X'')/(X')^3<0\). The corresponding family of disks of concave radius defines a convex body by the triangle inequality. At each pole \(Y\) has a simple zero and \(X\) is even about the pole, so locally \(X\) is analytic in \(Y^2\). Rotation yields a real-analytic surface there. The rotational principal curvature is \(-X'/(SY)>0\), with a positive limit at each pole by the removable ratios just described. The principal curvatures remain positive on the entire surface. The parity of \(\gamma\) gives origin symmetry. The normal angle \(\theta\) has derivative close to one and covers all meridional directions. Rotation about the axis therefore gives every unit normal in \(S^2\). Since \(q=h_a\) and \(A_\gamma=\nu_a\), all sections in the theorem have the asserted area. The reference value of the area is \(\pi(1-h_*^2)>0\). The total correction is \(O(\tau^\kappa)\), so the first-order expansions of the seed give \eqref{eq:main-asym}, including the indicated Legendre normalization of \(\varepsilon_a\). The nonzero degree-two coefficient and origin symmetry imply the body is nonspherical. Finally \(h_a\to h_*<1\) and \(\rho_a\to1\) uniformly, giving strict containment of \(h_aB\). \end{proof} \section{Comparison with a ball}\label{sec:comparison} Use \(\langle g\rangle=(4\pi)^{-1}\int_{S^2}g\dd\omega\) for the spherical mean. For a body with radial function \(\rho>h>0\), averaging the parallel section areas gives \begin{equation}\label{eq:mean-section} \frac{1}{4\pi}\int_{S^2} A_\rho(u,h)\dd u =\pi\bigl(\langle\rho^2\rangle-h^2\bigr). \end{equation} This is the three-dimensional case of \cite[Lemma~4.2, (4.4)]{KurusaOdor}. To see the normalization directly, integrate the kernel in \eqref{eq:radial-delta} over the normals \(u\). For \(r>h\), this integral is \(2\pi/r\) by the coordinate \(t=u\cdot\omega\), and for \(rh_a\). Since \(\rho_a\) is nonconstant, the strict convexity of \(s\mapsto s^{3/2}\) gives \[ R_a^3=\langle\rho_a^2\rangle^{3/2} <\langle\rho_a^3\rangle=1. \] Thus \(\vol(R_aB)<4\pi/3=\vol(K_a)\). Dilate both bodies by \(t/h_a\). Their areas in corresponding sections scale by \((t/h_a)^2\), and the cutting sphere becomes \(tB\). The dilation preserves the strict containment, regularity and volume ordering. Because \(\rho_a\to1\) on a common positive strip, their radial functions approach each other in every fixed \(C^m\) norm after dilation. \end{proof} The leading size of the volume discrepancy follows from the same identity. Write \(\rho_a=1+\varepsilon_a f+g\) with \(g=O(\varepsilon_a^2)\). Expanding \(\langle\rho_a^3\rangle=1\) gives \(\langle g\rangle=-\varepsilon_a^2/5+O(\varepsilon_a^3)\). It follows from \eqref{eq:mean-section} that \begin{equation}\label{eq:area-defect} \begin{aligned} \nu_a&=\pi(1-h_a^2)-\frac{\pi}{5}\varepsilon_a^2 +O(\varepsilon_a^3),\\ R_a&=1-\frac{1}{10}\varepsilon_a^2+O(\varepsilon_a^3). \end{aligned} \end{equation} \section*{AI use declaration} OpenAI's ChatGPT and Codex were used to develop and draft the argument and to revise the manuscript. Their mathematical contributions include the specialization of the Legendre root calculation in Section~\ref{sec:arithmetic}, the section-area construction and its linearization in Sections~\ref{sec:formal}--\ref{sec:chords}, and the finite-order reduction, inverse estimates and Newton argument in Sections~\ref{sec:reduction}--\ref{sec:iteration}. They also assisted in checking the calculations and references in the present revision. % Keep the author address and email on the final bibliography page. \enlargethispage{3\baselineskip} \begin{thebibliography}{99} \bibitem{BL} J.~A.~Barker and D.~G.~Larman, \emph{Determination of convex bodies by certain sets of sectional volumes}, Discrete Math. \textbf{241} (2001), 79--96. \href{https://doi.org/10.1016/S0012-365X(01)00111-X}{DOI}. \bibitem{DaiXu} F.~Dai and Y.~Xu, \emph{Approximation theory and harmonic analysis on spheres and balls}, Springer Monogr. Math., Springer, New York, 2013. \href{https://doi.org/10.1007/978-1-4614-6660-4}{DOI}. \bibitem{DGP} T.~de Groot and L.~Price, \emph{Analytic symmetric floating bodies}, preprint, 2026, \href{https://hexagonmath.org/2610.00629v1}{Hexagon:2610.00629v1}. \bibitem{Deng} Z.~Deng, \emph{A proof of the irreducibility conjecture for Legendre polynomials}, preprint, 2026, \href{https://arxiv.org/abs/2609.22336v2}{arXiv:2609.22336v2}. \bibitem{FRSY} A.~Fish, D.~Ryabogin, W.~R.~Sun, and V.~Yaskin, \emph{On determination of convex bodies by their section functions}, preprint, 2026, \href{https://arxiv.org/abs/2610.10229v1}{arXiv:2610.10229v1}. \bibitem{Haddad} J.~Haddad, \emph{An approximate counterexample to the Barker--Larman problem in dimension \(4\)}, preprint, 2026, \href{https://arxiv.org/abs/2609.10257v2}{arXiv:2609.10257v2}. \bibitem{HR} J.~Haddad and D.~Ryabogin, \emph{On convex bodies with constant non-central sections}, preprint, 2026, \href{https://arxiv.org/abs/2605.00299v1}{arXiv:2605.00299v1}. \bibitem{Holt} J.~B.~Holt, \emph{The irreducibility of Legendre's polynomials}, Proc. Lond. Math. Soc. (2) \textbf{11} (1913), 351--356. \href{https://doi.org/10.1112/plms/s2-11.1.351}{DOI}. \bibitem{KM} B.~Kadets and D.~Mangoubi, \emph{Root sharing and \(2\)-adic interpolation for Legendre polynomials}, preprint, 2026, \href{https://arxiv.org/abs/2611.09312v1}{arXiv:2611.09312v1}. \bibitem{KSExact} Knowledge Sense, \emph{Exact counterexamples to the Barker--Larman problem in every dimension at least four}, preprint, October~12, 2026. \href{https://github.com/KnowledgeSense/math/blob/main/preprints/Exact-Counterexamples-to-the-Barker-Larman-Problem-in-Every-Dimension-at-Least-Four-October-12-2026/article.pdf}{Manuscript}. \bibitem{KSFormal} Knowledge Sense, \emph{The formal Barker--Larman branch in dimension four has zero radius of convergence}, preprint, October~12, 2026. \href{https://github.com/KnowledgeSense/math/blob/main/preprints/The-Formal-Barker-Larman-Branch-in-Dimension-Four-Has-Zero-Radius-of-Convergence-October-12-2026/article.pdf}{Manuscript}. \bibitem{KurusaOdor} \'A.~Kurusa and T.~\'Odor, \emph{Characterizations of balls by sections and caps}, Beitr. Algebra Geom. \textbf{56} (2015), 459--471. \href{https://doi.org/10.1007/s13366-014-0203-9}{DOI}. \bibitem{MakaiMartini} E.~Makai, Jr. and H.~Martini, \emph{Unique local determination of convex bodies}, Acta Math. Hungar. \textbf{150} (2016), 176--193. \href{https://doi.org/10.1007/s10474-016-0640-z}{DOI}. \bibitem{Matthews} B.~Matthews, \emph{Hedgehog reconstruction of polygons: Non-central sections and slabs}, Canad. Math. Bull. (2026), published online, 1--13. \href{https://doi.org/10.4153/S0008439526102355}{DOI}. \bibitem{Matveev} E.~M.~Matveev, \emph{An explicit lower bound for a homogeneous rational linear form in the logarithms of algebraic numbers. II}, Izv. Math. \textbf{64} (2000), 1217--1269. \href{https://doi.org/10.1070/IM2000v064n06ABEH000314}{DOI}. \bibitem{Morales} E.~Morales-Amaya, \emph{Convex bodies with centrally symmetric sections}, preprint, 2026, \href{https://arxiv.org/abs/2307.07624v3}{arXiv:2307.07624v3}. \bibitem{DLMF} F.~W.~J.~Olver, A.~B.~Olde Daalhuis, D.~W.~Lozier, B.~I.~Schneider, R.~F.~Boisvert, C.~W.~Clark, B.~R.~Miller, B.~V.~Saunders, H.~S.~Cohl, and M.~A.~McClain (eds.), \emph{NIST Digital Library of Mathematical Functions}, National Institute of Standards and Technology, Release~1.2.8, 2026. \url{https://dlmf.nist.gov/}. \bibitem{Yaskin} V.~Yaskin, \emph{Unique determination of convex polytopes by non-central sections}, Math. Ann. \textbf{349} (2011), 647--655. \href{https://doi.org/10.1007/s00208-010-0538-y}{DOI}. \bibitem{YY} V.~Yaskin and M.~Yaskina, \emph{Thick sections of convex bodies}, Adv. Appl. Math. \textbf{71} (2015), 174--189. \href{https://doi.org/10.1016/j.aam.2015.09.010}{DOI}. \end{thebibliography} \end{document}