\section{A uniform mass-ratio bound above four}\label{sec:source-high} \begin{theorem}\label{thm:high-source-floor} For $\alpha\ge4$ and $n\ge5$, the original force and centering equations imply $\min_i m_i/\max_i m_i>1/5$. \end{theorem} The minimum-mass site will bound the conserved current from above, and the maximum-mass site will bound it from below. We use both bounds for the mass ratio and then for the count. Denote the common radial row by $C_0>0$ and the largest mass by $m_*$. Replace masses by $m_i/m_*$ and positions by $\ell z_i$, where $\ell^\alpha=C_0/m_*$. The new common row is $C_0/(m_*\ell^\alpha)=1$, and the largest mass is one. Both sides of the force equations scale by $1/(m_*\ell^{\alpha+1})$ when the multiplier is replaced by $\lambda/(m_*\ell^{\alpha+2})$. Centering and distinctness are preserved. We use these normalized variables throughout this section. Write $x=(2R)^{-2}$, $r=4/\alpha$, $q=\min_i m_i$, $U=M-1$, $T=M-q$. Centering at a unit mass gives $U\ge1$. Fix a maximum-mass site $o$. Its contribution to any other radial row gives $d_{oj}>1$. On the other atoms write $y=d_{oj}^{-2}$ and $d\nu=y^{\alpha/2-2}d\mu$, where $\mu$ is the mass measure. Thus $x\le y<1$, $\nu\le\mu$, and $\int y^2d\nu=1$. As $W=\int (y^{\alpha/2})^{1-r}d\mu$, H\"older gives $W\le U^r$. At any site $\sum_{j\ne i}m_j d_{ij}^2=M/(2x)$ by centering. Jensen at the minimum therefore yields $2x\le M/T^{1+2/\alpha}$, and \[ k=Wx^2\le [M/(2T)]^2(U/T)^r. \tag{A} \] \subsection{One two-sided current primitive} Route each pair along its shorter arc of length $l$ with weight $m_im_jK(l)$, $d(l)=\sin(\sqrt x\,l)/\sqrt x$, $K(l)=d(l)^{-\alpha-1}\cos(\sqrt x\,l)$ up to $L=\pi R$, zero thereafter. An antipodal pair has zero current. The crossing sum $J$ is constant since the jump at a label is its zero tangential force; each incident directional sum is at most $J$. From every label the cumulative mass $F(t)$ on either side up to arclength $t$, excluding the marked atom, satisfies $F(t)\le t$: each successive mass is at most one and below its preceding chord to power $\alpha$, hence below its preceding gap. Restarting at the first atom of an arc gives $\mu(I)\le1+|I|$. In particular, if $f\ge0$ has interval superlevel sets on one semicircle, the layer-cake identity gives \[ \int f\,d\mu=\int_0^{\sup f}\mu\{f>s\}\,ds \le\sup f+\int f(l)\,dl. \] This is the one-hump occupation bound used below; its integral is with respect to actual arclength, not the mass measure. At the minimum take the side with nonantipodal radial sum at most $1/2$, put $A_{\min}=(2q)^{-1/\alpha}$ and $z=x/A_{\min}^2$. To bound $z$, keep both adjacent terms in each unit radial row and use the mass floor $q$. If $d_i$ are the adjacent chords, summing over the sites gives $n^{-1}\sum_i d_i^{-\alpha}\le1/(2q)$. Jensen's inequality for the inverse power and then for the chord sine gives \[ (2q)^{1/\alpha}\le n^{-1}\sum_i d_i\le 2R\sin(\pi/n), \qquad z\le\sin^2(\pi/n)\le\sin^2(\pi/5). \] On the chosen side each incident chord has $d^{-1}\le A_{\min}$. Multiplying the radial terms by $qd^{-1}\sqrt{1-xd^2}$ bounds twice the incident current by $qA_{\min}\sqrt{1-z}$. For nonincident crossings let $F_+,F_-$ be the cumulative masses on the two semicircles measured from the minimum site. A crossing pair follows the short path through this site if its length $u+v$ is at most $L$. The total nonincident current is $\iint F_+(u)F_-(v)K''(u+v)\,du\,dv$ in the distributional sense: integrating twice from any atom pair gives its exact kernel. In the interior $K''=\cos(\sqrt x\,l)d^{-\alpha-3} [(\alpha+1)(\alpha+2)-\alpha^2xd^2]\ge0$; the zero extension adds the positive mass $x^{1+\alpha/2}\delta_L$ since $K'(L-)=-x^{1+\alpha/2}$. On $u\le v$ charge only $F_-(v)\le v$ and use symmetry on $v\le u$. The resulting inner primitive for $uL/2$. The primitive includes the boundary mass of $K''$ at $L$. Integrating from a surviving atom at distance $t$ to $L/2$ yields \[ \frac{3}{4\alpha}[d(2t)^{-\alpha}-d(L)^{-\alpha}]+\frac t2K(2t) \] since $K=-(d^{-\alpha})'/\alpha$. Both triangles thus give twice the full nonincident cost at most $\sum_i m_i d(t_i)^{-\alpha} P_\alpha(\theta_i)$, $\theta_i=\sqrt x\,t_i$, where $P_\alpha=0$ for $\theta\ge\pi/4$ and \[ P_\alpha(\theta)=(2\cos\theta)^{-\alpha} \left[\frac{3(1-\sin^\alpha(2\theta))}{2\alpha}+\theta\cot(2\theta)\right]. \] Here $t_i$ is measured from the marked minimum, and the weights $m_i d(t_i)^{-\alpha}$ sum to its unit radial row. Put $w=\tan^2\theta$, $Y=w/(1+w)$ and $b=\sin(2\theta)$. Since $(1-b^\alpha)/\alpha= \int_b^1s^{\alpha-1}ds$ and $2\cos\theta>1$, $P_\alpha\le P_4$. Write $\theta\cot(2\theta)=(1-w)\mathcal A(w)/2$, where $\mathcal A(w)=\int_0^1(1+wt^2)^{-1}dt$. Then $\mathcal A(w)=1-w\int_0^1t^2/(1+wt^2)\,dt\le1-w/(3+9w/5)$ by Jensen under $3t^2dt$. Substitution gives \[ 16(1/16-5Y^2/32-P_\alpha)\ge \frac{(3+w)(5-16w)^2+w^2(82+94w+295w^2+405w^3+80w^4)} {120(1+w)^2(5+3w)}\ge0 . \] Thus twice the nonincident current is at most $1/16$. For the count bound we use $m_i\ge q$ in the unit radial row, so $d(t_i)^\alpha\ge q$ and $Y=xd(t_i)^2\ge z2^{-2/\alpha}\ge z/\sqrt2$. The stronger bound is $e(z)=1/16-5z^2/64$. It also bounds the zero contributions since $z\le\sin^2(\pi/5)$ gives $e(z)>0$. \subsection{One barycenter at the maximum} Use the whole measure $d\sigma=(y-x)d\nu$ there, $D=\int d\sigma>0$, $\beta=D^{-1}\int y\,d\sigma$, $v=x/\beta<1$. Antipodes have zero weight and contribute no current. Then $H=1-k=\int(y+x)d\sigma =D\beta(1+v)$. The two incident directions give $2J\ge\int y^2/\sqrt{y-x}\,d\sigma$. The integrand is convex in $y>x$: at $w=y-x$ its second derivative times $4w^{5/2}$ is $3w^2-2xw+3x^2>0$. Jensen's inequality therefore gives \[ (2J)^2\ge \frac{H^2\beta}{(1-v)(1+v)^2}. \tag{J} \] The comparison atom $\beta$ denotes a mean, not a physical solution. \subsection{The strong one-fifth ratio} If $q\le1/5$, $T\ge U+4/5$ and (A) gives $k\le[(U+1)/(2U+8/5)]^2\le25/81$. At the minimum $qA_{\min}\le (5/2)^{1/4}/5<63/250$, so $2J<629/2000=:B_0$. Set $B=3\beta$. The definition $\int y\,d\sigma=D\beta$ gives the exact signed identity $2D/3=\int(y-x)(1-y/B)d\nu$. Restricting this integral to $x\le y\le B$ can only increase it, since its omitted integrand is negative. On that interval $\nu\le\mu$. On each true semicircle the positive part under $\mu$ has one hump of height at most $(B-x)^2/(4B)$. With $w=\sqrt y$ the arclength element is $|dw|/(w^2\sqrt{1-x/w^2})$; its weighted integral is at most $\int_{\sqrt x}^{\sqrt B}(1-w^2/B)dw$. Thus the preceding arc occupation gives \[ D\le\beta(3-v)^2/4+\sqrt\beta[2\sqrt3-\sqrt v(3-v/3)] . \] The first term after multiplication by $\beta(1+v)$ is at most $3\beta^2$, since $(1+v)(3-v)\le4$ and $3-v\le3$. For the second, $(1+v)(3-v/3)\ge2\sqrt{3v}$ gives \[ (1+v)\{2\sqrt3-\sqrt v(3-v/3)\}\le2\sqrt3<4. \] Thus $H=\beta(1+v)D\le3\beta^2+4\beta^{3/2}$. But $H\ge h=56/81$ and (J) gives $H^2\beta\le(32/27)B_0^24$. \end{lemma} \begin{proof} Suppose now $M\le4$. Strict centering at the maximum gives $1903/1000$, and $\sqrt{1-d}>809/1000$ leaves a positive net slope $26843567/2028000000$. Thus it suffices at fixed $r,z$ to disprove the bound at $q_*=3/4,k_*=2^r z^2$. This is a scalar comparison, not a variation of the configuration. The remaining difference \[ V(r,z)=c(1-2^r z^2)^p-q_*(2/3)^{r/4}\sqrt{1-z}-e(z) \] is separately concave on $0\le r\le1$, $0\le z\le d_0=691/2000>d$. In $r$ the first term's second derivative has the sign of $p2^rz^2-1<0$ when $z>0$, and the negative exponential is concave. In $z$ the first second derivative is $-2Bpc(1-Bz^2)^{p-2}[1-(2p-1)Bz^2]$ for $B=2^r$. Its magnitude exceeds $2p(847/1000)(5/8)$ since $2(2p-1)d_0^2<3/8$. The two positive contributions sum to at most $3/8+5/32$, since $(1-d_0)^3>1/4$. Thus only the four corners remain. At $z=0$ use $c>847/1000$, $(2/3)^{1/4}<904/1000$. At $z=d_0$ also use $\sqrt{1-d_0}<8091/10000$, $(1-d_0^2)^{1/4}>968/1000$ and $(1-2d_0^2)^{1/4}>934/1000$. All follow by positive powers; substitution at all four corners gives $V>1/2500$. This contradiction proves the strict count bound. \end{proof}