\section{Two comparisons for the common radial value}\label{sec:comparison} Assume $\alpha>0$. The radial rows give a lower bound for the common value $C$. A trial configuration in consecutive cells of widths proportional to the masses gives an upper bound. Sections~\ref{sec:source-low}--\ref{sec:native-low} compare these bounds after controlling the mass ratio. For $\alpha\geq4$, Section~\ref{sq:section} pairs the same lower bound with a finite-edge trial upper bound. The proofs below two begin with consecutive force rows and Section~\ref{sec:pressure} also uses the remainders in the cell comparison. Put $P=\alpha+1$, $\nu=1/P$, $w_i=m_i/M$, and $u_i=nw_i$. We distinguish the uniform and mass-weighted expectations \[ \E_Uf=\frac1n\sum_i f_i,\qquad \E_wf=\sum_iw_if_i=\E_U(uf). \] In particular $\E_Uu=1$ and $\E_w(1/u)=1$. The common scale factor is $Q_R=(2R)^{-\alpha}$. \subsection{A continuous extension of the regular-polygon row} \begin{lemma}\label{lem:gamma} The even density \[ \mu_\alpha(\xi)=\frac{2^{\alpha-1}}{\pi\Gamma(\alpha)} |\Gamma(\alpha/2+\ii\xi)|^2 \] is a probability density, and \begin{equation}\label{eq:gamma-mgf} \int_{\R}e^{z\xi}\mu_\alpha(\xi)\dd\xi =\cos(z/2)^{-\alpha}\qquad(|\Re z|<\pi), \end{equation} with the analytic branch taking the value one at zero. \end{lemma} \begin{proof} Let $s=\alpha/2$. The substitution $x=e^t$ in Euler's beta integral gives \[ \int_{\R}\frac{e^{st}}{(1+e^t)^{2s}}e^{\ii\xi t}\dd t =\frac{\Gamma(s+\ii\xi)\Gamma(s-\ii\xi)}{\Gamma(2s)}. \] Fourier inversion, applied to the smooth exponentially decaying function on the left, gives \[ \int_{\R}e^{\ii v\xi}|\Gamma(s+\ii\xi)|^2\dd\xi =\frac{2\pi\Gamma(2s)}{[2\cosh(v/2)]^{2s}} \qquad(v\in\R). \] For completeness, the needed decay also follows directly from the beta integral. Choose an integer $k>s$. Taking absolute values in $B(s+\ii\xi,k-s)$ yields \[ \frac{|\Gamma(s+\ii\xi)|}{|\Gamma(k+\ii\xi)|} \le\frac{\Gamma(s)}{\Gamma(k)}. \] The recurrence and reflection formulas give \[ |\Gamma(k+\ii\xi)|^2 =\frac{\pi\xi}{\sinh(\pi\xi)} \prod_{j=1}^{k-1}(j^2+\xi^2). \] The quotient at zero is interpreted continuously. Thus $\mu_\alpha$ is bounded by a polynomial times $e^{-\pi|\xi|}$. Its transform is holomorphic on $|\Re z|<\pi$, with all derivatives obtained under the integral. The displayed Fourier identity and analytic continuation prove \eqref{eq:gamma-mgf}. Setting $z=0$ proves normalization. \end{proof} For real $N>1$ define \begin{equation}\label{eq:analytic-row} S_\alpha(N)=\int_{\R} \frac{\sinh((N-1)\pi\xi/N)}{\sinh(\pi\xi/N)} \mu_\alpha(\xi)\dd\xi. \end{equation} This is positive. Its integrand at $\xi=0$ has the value $N-1$. At infinity its growth is strictly smaller than $e^{\pi|\xi|}$, so Lemma~\ref{lem:gamma} proves convergence. For integer $n\ge2$ the quotient expands as $\sum_{k=1}^{n-1}e^{(1-2k/n)\pi\xi}$. Using $\cos((1-2k/n)\pi/2)=\sin(\pi k/n)$ in \eqref{eq:gamma-mgf} gives \begin{equation}\label{eq:integer-row} S_\alpha(n)=\sum_{k=1}^{n-1}\sin(\pi k/n)^{-\alpha}. \end{equation} The antipodal rank, when present, appears once. We will also need a representation that keeps the subtracted constant visible. Put \[ \nu_\alpha(\xi)=\frac{\sinh(\pi\xi)|\Gamma(\alpha/2+\ii\xi)|^2} {\pi\Gamma(\alpha)},\qquad B_\alpha(N)=2^{\alpha+1}\int_0^\infty \frac{\nu_\alpha(\xi)}{e^{2\pi\xi/N}-1}\dd\xi \quad(N>0). \] Then \begin{equation}\label{eq:bose-row} S_\alpha(N)=B_\alpha(N)-B_\alpha(1). \end{equation} Indeed, with $t=\pi\xi$, \[ 2\sinh t\left(\frac1{e^{2t/N}-1}-\frac1{e^{2t}-1}\right) =\frac{\sinh(t-t/N)}{\sinh(t/N)}, \] and evenness gives the formula. The function $\nu_\alpha$ is $O(\xi)$ at zero and has at most polynomial growth at infinity. These bounds, together with the exponential denominator, justify convergence and count differentiation locally on $N>0$. Define the two mass-dependent quantities \begin{align} B_{\mathrm{root}}&=Q_RS_\alpha(n)\frac Mn(\E_Uu^\nu)^P,\label{eq:root-value}\\ V_{\mathrm{log}}&=Q_R\exp\{\E_w\log[m_iS_\alpha(1/w_i)]\}.\label{eq:log-value} \end{align} Our goal in this section is $B_{\mathrm{root}}\le C\le V_{\mathrm{log}}$. \subsection{The lower comparison} \begin{lemma}[One chord support]\label{lem:holder} For any finite probability law let $t,s\in(0,2\pi)$, $A>0$, $c=\E A$, $r=\sin(s/2)/\sin(t/2)$ and $b=(s-t)\cot(t/2)/2$. With $P=1+\alpha$ put $D(x,c)=x^P-c^P-Pc^{P-1}(x-c)$. Then \[ \E(A^P r^{-\alpha})-c^P+\alpha c^P\E b =\E\{rD(A/r,c)+\alpha c^P(1+b-r)\}\ge0 . \] \end{lemma} \begin{proof} Expand using $\E A=c$. The two remainders are nonnegative by the tangents to the convex power and the concave sine, respectively; the sine remainder vanishes only at $s=t$. \end{proof} \begin{proposition}\label{prop:lower} Every configuration under consideration satisfies $B_{\mathrm{root}}\le C$. \end{proposition} \begin{proof} At forward rank $k$, take $t=2\pi k/n$, $s=\theta_{i+k}-\theta_i$ in the cyclic lift, $A=m_i^\nu$ under the uniform index $i$. Each gap occurs $k$ times, so $\E_U s=t$ and $\E_U b=0$. Multiply the lemma by $Q_R\sin(t/2)^{-\alpha}$ and sum all nonzero ranks. Each term with mass $m_i$ belongs to the radial row at $i+k$, so the left side before subtraction is the average of those rows. The lower bound is $Q_RS_\alpha(n)(\E_U m^\nu)^P =B_{\mathrm{root}}$ by \eqref{eq:integer-row}. \end{proof} \subsection{An exact cell identity} Partition the angular circle into consecutive cells of widths $2\pi x_i$, where $x_i>0$ and $\sum_i x_i=1$, and let $\phi_i$ be their midpoints. For $\xi\ge0$ put \[ K_\xi(t)=\cosh(\xi(\pi-t)),\quad 0\le t\le2\pi,\qquad a_i=\frac{\sinh(\pi\xi x_i)}{\pi\xi},\quad b_i=\frac{\sinh(\pi\xi(1-x_i))}{\sinh(\pi\xi x_i)}. \] At zero set $a_i=x_i$, $b_i=(1-x_i)/x_i$. Write $K_{ij}=K_\xi(\phi_j-\phi_i)$ with angular differences reduced modulo $2\pi$; this is symmetric. \begin{lemma}\label{lem:cell} For arbitrary positive trial masses $v_i$, \begin{align} \Delta_{\mathrm{cell}} &:=\sum_iv_i^2S_\alpha(1/x_i) -\sum_{i\ne j}v_iv_j\sin((\phi_j-\phi_i)/2)^{-\alpha}\nonumber\\ &=2\int_0^\infty\sum_{i0$. The integral of $K_\xi(t-\phi_i)$ over cell $j\ne i$ is $2\pi a_jK_{ij}$. The integral over the whole circle is $2\sinh(\pi\xi)/\xi$. Subtracting the own-cell integral leaves $2\sinh(\pi\xi(1-x_i))/\xi$. Thus \[ \sum_{j\ne i}K_{ij}a_j=a_ib_i. \] Expansion of squares now gives \[ \sum_i b_iv_i^2-\sum_{i\ne j}K_{ij}v_iv_j =\sum_{i0$ off uniformity in Section~\ref{sec:native-low}. For $\alpha\ge4$, Section~\ref{sq:section} instead compares $B_{\rm root}$ with the finite-edge upper $U$.