\section{The interval \texorpdfstring{$0<\alpha\leq1$}{00$, $L>0$ and $s>0$, define \begin{align} w_{p,L}(s)&=\frac{2^{p-1}\pi^{p-2}L^{1-p}}{\Gamma(p)} \sinh(Ls/2)\left|\Gamma\left(\frac p2+\frac{Ls}{2\pi}\ii\right)\right|^2, \label{low:density}\\ k(t)&=\int_0^\infty w_{p,L}(s)e^{-st}\dd s,\qquad U(t)=\int_0^\infty s w_{p,L}(s)e^{-st}\dd s\quad(t>0). \label{low:kernel} \end{align} The parameters of $k$ and $U$ will be suppressed when fixed. The density is strictly positive. Stirling's formula on vertical lines gives $w_{p,L}(s)=O(s^{p-1})$ at infinity, while $w_{p,L}(s)=O(s)$ at zero. Consequently the integrals and all their derivatives converge locally uniformly for $t>0$, and $U=-k'$. \begin{lemma}[Exact periodization]\label{low:periodization} For $00$ the self-image sum \begin{equation}\label{low:self} S=2\sum_{j=1}^\infty k(jL), \end{equation} satisfies $00$, the geometric series is \[ \sum_{j\in\mathbb Z}e^{-s|t+jL|} =\frac{e^{-st}+e^{-s(L-t)}}{1-e^{-Ls}},\qquad \sinh(Ls/2)\frac{e^{-st}+e^{-s(L-t)}}{1-e^{-Ls}} =\cosh\bigl(s(L/2-t)\bigr). \] Tonelli's theorem, followed by $u=Ls/(2\pi)$ and $y=\pi-2\pi t/L$ in \eqref{low:coshmoment}, proves \eqref{low:periodicidentity}. The $O(s)$ bound at zero implies $k(t)=O(t^{-2})$ for large $t$, so \eqref{low:self} converges. For $00$, \begin{equation}\label{low:positiveintegral} W_p(u)=\frac{2^{p-2}u}{\Gamma(b)}\E\int_0^\infty \frac{r^a}{(r+Y)^{3/2}}e^{-u^2/(4(r+Y))}\dd r. \end{equation} Consequently $W_p(u)/u$ has strictly negative derivative, and \begin{equation}\label{low:gaussiancumulative} \int_0^T uW_p(u)\dd u<\frac{T^{p+1}}{p+1}\quad(T>0). \end{equation} For $1c$, couple the first $N$ summands to obtain $Y\ge G_N/(4\pi^2N^2)$ with $G_N\sim\operatorname{Gamma}(Np,1)$, so $\E Y^{-c}<\infty$. The Euler product gives $\E e^{-t^2Y}=[t/(2\sinh(t/2))]^p$. Put $h_p(t)=(2\sinh(t/2))^{-p}$ and $c_p=\Gamma(1-p)\cos(\pi p/2)$, with $c_1=\pi/2$. Beta substitution and reflection first give, for $00$. Tonelli, $Y>0$, and the pure formula therefore give \eqref{low:gaussiancumulative} directly. This argument uses only positive integrands, also when $p<1$. If $10$. Comparing with $Y=0$ puts the ratio in $(0,1)$. Every integrand, extended by zero below its lower endpoint, is nondecreasing in $u$ and has positive derivative on $B>Y/u^2$. Fatou applied to nonnegative forward difference quotients gives a strictly positive derivative of the smooth Gamma ratio. \end{proof} \begin{lemma}[Two density orders]\label{low:orders} For $00). \end{equation} The latter inequality is strict when $p<1$. \end{lemma} \begin{proof} Put $\eta=2\pi/L$. Direct scaling gives $w_{p,L}(s)=\eta^{p-1}W_p(s/\eta)/\Gamma(p)$. Lemma~\ref{low:gaussianorders} gives both assertions for $p<2$, and its zero-end bound gives $w(s)\le Cs$. At $p=2$ use $w_{2,L}(s)=s$ directly. \end{proof} \begin{lemma}[Decreasing tests]\label{low:test} Let $00$, \begin{equation}\label{low:ABratio} 0s$. By Lemma~\ref{low:orders}, $w(s)/s$ is also nonincreasing. Therefore, for $u,v>0$, the product \[ \{P_A(u)P_B(v)-P_A(v)P_B(u)\} \{w(u)v-w(v)u\} \] is nonnegative. Multiply it by $u^jv^je^{-t(u+v)}$ and integrate in both variables. Symmetrizing gives, for every integer $j\geq0$, \begin{equation}\label{low:momentcomparison} \frac{\int_0^\infty s^{j+1}P_A(s)e^{-st}\dd s} {\int_0^\infty s^{j+1}P_B(s)e^{-st}\dd s} \leq \frac{\int_0^\infty s^jw(s)P_A(s)e^{-st}\dd s} {\int_0^\infty s^jw(s)P_B(s)e^{-st}\dd s}. \end{equation} For $j=0$, the left ratio is $r_0(t)$, by $\int_0^\infty s^je^{-ts}\dd s=j!t^{-j-1}$. This proves \eqref{low:ABratio}. For positivity and convexity of $A-B/4$, use \eqref{low:momentcomparison} for $j=0,1,2$. The corresponding reference differences $\int s^{j+1}(P_A-P_B/4)e^{-st}\dd s$ are \[ \begin{array}{c|c} j&\text{difference, with }y=t-1\geq0\\\hline 0&(9y^2+17y+6)/(4t^3(t+1)^3)\\ 1&(18y^3+56y^2+51y+10)/(2t^4(t+1)^4)\\ 2&(90y^4+390y^3+585y^2+315y+18)/(2t^5(t+1)^5). \end{array} \] Every numerator is positive. Derivatives of the Laplace transforms insert $-s$, so the case $j=2$ proves strict convexity. For $B-A/4$, the positivity of every moment follows directly from $P_B>P_A>0$. Finally $w(s)\leq Cs$ and $00$, with $K(x)=O(x^{-3})$. Let $t_00$, and masses $00$, each assigned weight $s$. It is an auxiliary point sequence, not a new physical configuration. Put $H_K(t,s)=s\sum_{j\geq0}K(t+js)$ and $T_K(t,s)=s\sum_{j\geq1}K(t+js)$. Convex secants show that $H_K(t,s)$ is nondecreasing in $s$ when $t\geq d$, and $T_K(t,s)$ is nonincreasing whenever $t+s\geq d$. Here is a finite justification that also allows corners of $K$. A finite convex function is locally absolutely continuous in the interior of its domain and differentiable almost everywhere. On a compact interval of positive step sizes, each finite sum below is absolutely continuous away from a possible endpoint $t+js=d$. For almost every step size, put $v_j=K(t+js)$ and $z_j=K'(t+js)$ for $j\geq1$. Set $z_0=0$: the value $v_0=K(t)$ is constant in the step size. Convexity gives $v_{j+1}-v_j\leq sz_{j+1}$ and $sz_{j+1}\leq v_{j+2}-v_{j+1}$. Summing these after multiplication by $j+1$ yields \[ \sum_{j=0}^N(v_j+jsz_j)\geq(N+1)v_N\geq0, \quad \sum_{j=1}^N(v_j+jsz_j)\leq Nv_{N+1}. \] The second upper bound tends locally uniformly to zero by cubic decay. Integrate these almost-everywhere inequalities between two interior step sizes and pass to the limit in the finite sums. Cubic decay gives locally uniform convergence. Continuity then includes any endpoint where a sample equals $d$. This proves both monotonicities, including the boundary function $T_K(0,s)$ for $s\geq d$. The identity $H_K(t,s)=sK(t)+T_K(t,s)$ now gives \[ \min(s,h)K(t)+H_K(t+h,h)\leq H_K(t,s). \] For $s\leq h$ this is tail monotonicity; for $h\leq s$ it is head monotonicity. Apply it successively with $(t,s,h)=(t_j,s_j,s_{j+1})$. The nonnegative remainders telescope and give $\sum q_jK(t_j)\leq H_K(t_0,s_0)=T_K(0,s_0)\leq T_K(0,d)$. For a positive Laplace density the head and tail monotonicities are strict: this follows by integrating the strict scalar monotonicities of $s/(1-e^{-us})$ and $s/(e^{us}-1)$ for $u>0$. If equality holds, no nonnegative local deficit in the telescoping inequality can be positive. The boundary strictness first gives $s_0=1$; the strict one-step inequality gives $s_{j+1}=s_j$, and positivity of $K(t_j)$ gives $q_j=\min(s_j,s_{j+1})=1$. This proves the equality claim. \end{proof} \begin{lemma}[The force balances]\label{low:balances} With the incident normalization above, put $D=k(1)+S$. Then \begin{align} \sum_jq_jA(t_j)+\sum_jr_jB(v_j)&=U(1)+(1-b)D,\label{low:balanceA}\\ \sum_jq_jB(t_j)+\sum_jr_jA(v_j)&=bU(1)-(1-b)D.\label{low:balanceB} \end{align} For $00 \end{equation} and \begin{equation}\label{low:circledefect} 0\leq F_{p,L}(b):= \sum_{j=1}^\infty\{A(j)-B(j)/4\} +b\sum_{j=1}^\infty\{B(jb)-A(jb)/4\} -(1-b/4)U(1)-\tfrac54(1-b)D. \end{equation} \end{lemma} \begin{proof} At a body position $x$, sum the image potential over all other atoms and their translates, including the nonzero translates of $x$. By \eqref{low:periodicidentity} the result is the scaled radial row plus $m_x S$, where $m_x$ is the mass at $x$. Differentiating the chord potential gives the tangential equation; the derivatives from nonzero translates of $x$ cancel in pairs. Subtracting the radial row at $1$ from the row at $0$ gives \begin{equation}\label{low:radialdiff} \sum_jq_j C_1(t_j)-\sum_jr_j C_1(v_j)=(1-b)(k(1)+S). \end{equation} The two zero tangential rows at $0$ and $1$ give, respectively, \begin{equation}\label{low:tangentrows} \sum q_jU(t_j)-\sum r_jU(v_j+1)=bU(1),\qquad \sum r_jU(v_j)-\sum q_jU(t_j+1)=U(1). \end{equation} Combining these three identities proves the asserted balances. The ratio $r_0$ is strictly increasing, lies in $(0,1)$, and every $v_j\geq b$. Lemma~\ref{low:kernelcomparison} therefore gives \[ r_0(b)\{\sum q_jA(t_j)+\sum r_jB(v_j)\} <\sum q_jB(t_j)+\sum r_jA(v_j). \] The strict part follows already from $A0$, this implies $b>r_0(b)$ and \[ (1-b)(1+r_0(b))<\frac{U(1)}D(b-r_0(b)) \leq p(b-r_0(b)), \] where the last step is \eqref{low:radialslope}. Multiplication by the positive denominator $(b+1)(3b+2)$ gives exactly \eqref{low:cubic}. Finally subtract one quarter of \eqref{low:balanceB} from \eqref{low:balanceA} and apply Lemma~\ref{low:packing} to the left ray with $K=A-B/4$, $d=1$, and the right ray with $K=B-A/4$, $d=b$. Lemma~\ref{low:kernelcomparison} verifies all its hypotheses. The result is \eqref{low:circledefect}. \end{proof} \subsection{Transfer to the line and the scalar exclusion} For $s,b>0$ write \begin{align*} T_b(s)&=\frac{b}{e^{bs}-1},\quad P(s)=P_B(s)-\tfrac14P_A(s),\quad c_b=\tfrac54(1-b),\\ E_b(s)&=P(s)\{T_b(s)-T_1(s)\} -(1-b)(s/4+5/4)e^{-s},\\ \mathcal L_p(b)&=\int_0^\infty s^{p-1}E_b(s)\dd s. \end{align*} These are the full infinite sampling expressions, rather than truncated approximations. \begin{lemma}[The finite-period comparison]\label{low:comparison} For $00. \] The numerator in braces has value and first derivative zero at zero, and second derivative $xe^x>0$. Also $0\leq-h'(x)\leq1$, the upper bound following from $e^x\geq1+x$. Thus \[ \Delta_b(s)=\int_b^1-h'(sx)\dd x \] lies in $[0,1-b]$ and is nonincreasing in $s$. Differentiating $P(s)/s$ shows that it is nondecreasing and belongs to $[0,1]$: its derivative has the sign of $5/4-(5/4+5s/4+s^2/4)e^{-s}$, which is nonnegative by $e^s\geq1+s+s^2/2$. The identity \[ f_b(s):=\frac{c_b+E_b(s)}s =(1-b)-\frac{P(s)}s\{(1-b)-\Delta_b(s)\} \] therefore proves that $f_b$ is nonnegative, nonincreasing and bounded by $1-b$. Apply Lemma~\ref{low:test} to $e^{-\tau s}f_b(s)$, $\tau>0$. Its product with the signed first density is integrable, and the result is \[ 0\leq c_b\{\tau^{-p}-k(\tau)\} +\int_0^\infty(\rho_p-w_{p,L})E_b(s)e^{-\tau s}\dd s. \] Let $\tau\downarrow0$. Equation~\eqref{low:abel} and dominated convergence yield $0\leq c_bS+\int(\rho_p-w)E_b$. Combine this with \eqref{low:defectintegral} to obtain \eqref{low:transfer}. \end{proof} \begin{lemma}[The scalar sign]\label{low:scalarsign} For $00$: indeed $R_*'=25/(s+5)^2-e^{-s}>0$, since $1+s/58/27$ and $R_*(5)>3/2$, so its value at $5$ exceeds $1/3$. There is exactly one $c\in(0,5)$ where $J(c)=0$, with $J<0$ before it and $J>0$ after it. Here is a fully rational certificate for the sign at $p=1$. Define \[ f(t)=\frac{3t-1}{4t^2(t+1)^2},\qquad g(t)=\frac{3t+4}{4t^2(t+1)^2}. \] Termwise integration of the geometric series, or elementary integration of the Laplace transforms of $P_A,P_B$, gives \[ \mathcal L_1(b)=\sum_{k=1}^\infty\{f(k)+b g(kb)\}-9/4+3b/2. \] For $t>0$, \[ \frac1{t^3}-f(t)=\frac{t^2+9t+4}{4t^3(t+1)^2}>0,\qquad \frac1{t^3}-g(t)=\frac{(t+2)^2}{4t^3(t+1)^2}>0. \] Therefore the tail after $N$ terms is at most $(1+b^{-2})/(2N^2)$, by comparison with $\int_N^\infty x^{-3}\dd x$. At $b=2/3$, $N=12$, exact rational addition gives \[ \sum_{k=1}^{12}\{f(k)+\tfrac23g(2k/3)\} +\frac{13}{1152}-\frac54 =-\frac{2804360137027488583}{2150082517456954080000}<0. \] Thus $\int_0^\infty e^{-s}J(s)\dd s<0$. For $p\leq1$, the weight $s^{p-1}$ is nonincreasing. On both sides of the crossing $c$ this gives \[ s^{p-1}e^{-s}J(s)\leq c^{p-1}e^{-s}J(s). \] The integrals converge, so $\mathcal L_p(2/3)<0$. Finally $b\mapsto T_b(s)=h(bs)/s$ is convex, by the displayed formula for $h''$. The coefficient $P(s)$ is positive since $P_B>P_A>0$. Consequently $b\mapsto s^{p-1}E_b(s)$ is convex. Write $b=\tau(2/3)+(1-\tau)1$ with $\tau=3(1-b)>0$. Since $E_1=0$, pointwise convexity and integration give $\mathcal L_p(b)\leq\tau\mathcal L_p(2/3)<0$. \end{proof} \begin{theorem}\label{low:rigidity} Under the hypotheses of Theorem~\ref{thm:main}, if $0<\alpha\leq1$, the masses are equal and the sites form a regular polygon. \end{theorem} \begin{proof} Use the attained incident normalization and take $p=\alpha$. The cubic inequality \eqref{low:cubic} implies $b>2/3$. Indeed \[ Q_1(b)-Q_p(b)=(1-p)(3b^3+2b^2+b)\geq0,\qquad Q_1(b)=(b-2/3)(9b^2+8b+7/3)-4/9. \] If $b<1$, Lemmas~\ref{low:balances}, \ref{low:comparison} and \ref{low:scalarsign} give $0\leq F_{p,L}(b)<0$. Thus $b=1$. We use the undivided balances to recover equality, without dividing by $1-b$. The identity \[ \sum_{j=1}^\infty\{A(j)+B(j)\} =\sum_{j=1}^\infty\{U(j)-U(j+1)\}=U(1) \] and \eqref{low:balanceA} show that the separate nonnegative packing deficits for the left $A$ ray and the right $B$ ray both vanish. These kernels have strictly positive Laplace densities, so the equality part of Lemma~\ref{low:packing} forces every atom on the repeated left ray to have mass one and consecutive gap one. Every original atom occurs there. Hence all original masses are equal; equivalently the normalized word has all masses and gaps one and its actual period is $L=n$. Corollary~\ref{cor:equal} finishes the proof. \end{proof}