\section{The interval \texorpdfstring{$1\leq\alpha\leq9/5$}{1 <= alpha <= 9/5}}\label{sec:middle}
We continue to use the attained incident normalization and the kernels
$k,U,w_{p,L}$ of Section~\ref{sec:low}. This section uses two further
bodies of the original configuration. Proposition~\ref{prop:small} has already settled
$n\leq4$, so it remains to consider $n\geq5$ and $p=\alpha$ in
$[1,9/5]$.
The force rows give two inequalities constraining the third and fourth
masses. We use them to bound an expression that the first two rows
require to be nonnegative. For the gap $u\leq7/8$ we bound the third
mass by one root and compare the resulting mesh along a family of
kernels. For $u>7/8$ we use both mass constraints and compare kernels
at fixed geometry.
\subsection{Force inequalities at the next two bodies}
Suppose the configuration is not already equal-mass regular. In cyclic
order write four consecutive atoms as
\begin{equation}\label{mid:chart}
(0,1),\quad(1,b),\quad(1+u,m),\quad(1+u+v,r).
\end{equation}
Lemma~\ref{edge:low} gives $b>2/3$. The mass caps give $b\leq\min(1,u)$, $m\leq\min(u,v)$ and $r\leq v$.
In fact $u<1$. To see this, if $u\geq1$, pack the left $A$ ray at unit
spacing. On the right, first pack with head $H_B(u,m)$, increase the step
from $m$ to $u$, then use the decreasing boundary tail to bound it by
$T_B(0,1)$. The notation $H_K,T_K$ is that of
Lemma~\ref{low:packing}; their strict sampling properties hold for $A,B$
because their Laplace densities are positive. The first force balance
\eqref{low:balanceA} and $T_A(0,1)+T_B(0,1)=U(1)$ now force $b=1$.
The undivided equality argument in Theorem~\ref{low:rigidity}, which used
only $0
0.
\end{equation}
For $h>0$ introduce the positive Peano kernels
\[
A_h(t)=k(t)-k(t+h)-hU(t+h),\qquad
B_h(t)=hU(t)-k(t)+k(t+h).
\]
Their Laplace densities are $w(s)a_0(hs)$ and $w(s)b_0(hs)$, where
\[
a_0(x)=1-(1+x)e^{-x},\qquad b_0(x)=x-1+e^{-x}.
\]
Both are strictly positive for $x>0$. Therefore all positive-kernel
packing and strict head/tail comparisons in Lemma~\ref{low:packing}
apply at every positive starting point.
\begin{lemma}[Two necessary force inequalities]\label{mid:source}
Set $D_u=k(u)+S$, $M_u=uU(u)+D_u$, and
\begin{align}
q(v,m,r)&=T_{B_u}(0,1)+uD_u-mM_u+H_{A_u}(v,r),\label{mid:q}\\
H(v,r;m)&=uB_v(u)+T_{B_v}(u,1)+T_{A_v}(0,r)
-vrU(v)+(m-r)(k(v)+S).\label{mid:H}
\end{align}
At the actual configuration parameters \eqref{mid:chart},
$q(v,m,r)\geq0$ and $H(v,r;m)\geq0$.
\end{lemma}
\begin{proof}
The second Peano balance across a gap $h$, with left and right endpoint
masses $a,c$, is
\[
\text{left }B_h\text{ ray}+\text{right }A_h\text{ ray}
=hcU(h)-(a-c)(k(h)+S).
\]
Indeed, with $C_h(t)=k(t)-k(t+h)$, the radial difference is
$\text{left }C_h\text{ ray}-\text{right }C_h\text{ ray}
=(a-c)(k(h)+S)$ by the calculation in Lemma~\ref{low:balances}.
Subtract this from $h$ times the left tangential row,
$\text{left }U\text{ ray}-\text{right }U(\cdot+h)\text{ ray}=cU(h)$.
The coefficients on the two rays are then $B_h$ and $A_h$.
Apply it across the gap $u$. The left ray begins with the unit atom at
distance one and is bounded by $T_{B_u}(0,1)$. The right ray begins with
mass $r$ at distance $v$ and is bounded by $H_{A_u}(v,r)$.
Replacing $b$ by its upper bound $u$ increases the resulting expression
by $(u-b)D_u\geq0$, giving \eqref{mid:q}.
Apply the same identity across the gap $v$. On its left, retain the atom
of mass $b$ at distance $u$, bound $b$ by $u$, and pack all more remote
atoms by $T_{B_v}(u,1)$. On its right, packing gives $T_{A_v}(0,r)$.
These are exactly the four terms in \eqref{mid:H}, so $H\geq0$.
All repeated copies, including self images, occur in these rays; the
single term $S$ in each difference is the correction for the omitted
physical diagonal.
\end{proof}
For the following source estimates we allow the continuous family
\begin{equation}\label{mid:law}
w_\theta=(1-\theta)\rho_p+\theta w_{p,L},\qquad
S_\theta=\theta S,\qquad0\leq\theta\leq1,
\end{equation}
and use it to evaluate $k,U,A_h,B_h,q,H$ at any fixed arguments.
The configuration inequalities apply at $\theta=1$.
For $1\leq p<2$, $w_\theta(s)/s$ is
nonincreasing: the line term is $s^{p-2}/\Gamma(p)$ and the image term
has this property by Lemma~\ref{low:orders}.
Write
\[
\beta_h(s)=\frac{h}{e^{hs}-1},\quad
T_u(s)=e^{-us}(u+\beta_1(s)).
\]
Here $h_0(x)=x/(1-e^{-x})$ as in Lemma~\ref{mid:commonbose}.
\subsection{Signed sampling and positive parts}
For later use we record explicitly the properties of $x/(e^x-1)$ needed for
signed sampling. Put $h(x)=x/(e^x-1)$,
\[
C(x)=1-h_0'(x)=-h'(x)=\frac{(x-1)e^x+1}{(e^x-1)^2},\qquad E(x)=1-C(x).
\]
For $x>0$, $00$, and $\log C$ is concave.
The first two signs follow from $e^x>1+x$ and the formula for $h''$
in Lemma~\ref{low:comparison}. To check the last two without an
uncontrolled series remainder, direct differentiation gives
\[
C''=\frac{e^x V(x)}{(e^x-1)^4},\quad
V(x)=(x-3)e^{2x}+4xe^x+x+3,
\]
and
\[
C''C-(C')^2=-\frac{e^x J(x)}{(e^x-1)^6},\quad
J(x)=e^{3x}+(-2x^2+3x-5)e^{2x}+(-2x+7)e^x-x-3.
\]
The values of $V,V',V'',V'''$ at zero vanish, while
\[
V^{(4)}(x)=16(x-1)e^{2x}+(4x+16)e^x>0,
\]
because $(1-x)e^x<1$. The values of $J$ through its fourth derivative
at zero also vanish, and
\[
J^{(5)}(x)=243e^{3x}-(64x^2+224x+240)e^{2x}-(2x+3)e^x>0.
\]
For the last inequality divide by $e^{2x}$, use $e^{-x}\leq1$ and
$e^x\geq1+x+x^2/2$; what remains is at least
$17x+(115/2)x^2>0$. Repeated integration proves $V,J>0$.
It follows that $C(x)/x$ and $E(x)/x$ are decreasing. For $E$, use
concavity and $E(0+)=1/2$ to get $xE'(x)\leq E(x)$.
Also, for every $q\geq1$, $C(qx)/C(x)$ is decreasing: differentiate its
logarithm, use the negative nonincreasing logarithmic derivative of $C$,
and multiply its value at $qx$ by $q\geq1$.
We shall repeatedly use a simple signed comparison. If an integrable
function $f$ is nonnegative before a single crossing $s_0>0$ and
nonpositive after it, $\int f>0$, and $g>0$ is decreasing, then
\begin{equation}\label{mid:onecut}
\int_0^\infty f(s)g(s)\dd s
\geq g(s_0)\int_0^\infty f(s)\dd s>0,
\end{equation}
provided the product is integrable. This is a pointwise comparison on
the two sides of the crossing, so it applies to the entire signed tail.
The force-row combination below needs coefficients with $a+z=u$.
The bound $a\leq6/25$ ensures the left sampling signs in
Lemma~\ref{mid:signedpacking}. We use the following choice throughout:
\[
\begin{gathered}a(u)=\frac6{25}u(2-u),\quad z(u)=u-a(u),\\
\ell(s)=1-e^{-s}-a(u)s-z(u)se^{-s},\\
g_u(s)=z(u)+a(u)e^{-s}-\frac{1-e^{-s}}s.\end{gathered}
\]
Define the signed left and right kernels
\begin{align}
F_u(t)&=k(t)-k(t+1)-a(u)U(t)-z(u)U(t+1),\label{mid:FG}\\
G_u(t)&=z(u)U(t)+a(u)U(t+1)-k(t)+k(t+1),\nonumber\\
J_u(t)&=\max(0,G_u(t)).\nonumber
\end{align}
Their densities are respectively $w_\theta(s)\ell(s)$ and
$w_\theta(s)s g_u(s)$.
\begin{lemma}[The signed left ray and positive right part]
\label{mid:signedpacking}
For $00.0079.
\end{align*}
The first is elementary exponential integration. For the second,
integrate $-h'$ by parts and expand $1/(e^s-1)$ into its positive
geometric series; the boundary terms vanish. It is affine in $a,z$.
Its minimum on $0\leq a\leq6/25$, $0\leq z\leq1-a$ occurs at
$(a,z)=(6/25,19/25)$; the signs of the coefficients follow, for example,
from $1.2<\zeta(3)<1.203$ and $1.082<\zeta(4)<1.083$.
At that point the lower bound is
$-2(1.203)+(114/25)(1.082)-63/25=0.00792$.
All four zeta bounds follow from ten terms and neighboring integral
tails. Multiplication by the decreasing ratio $w_\theta(s)/s$ in
\eqref{mid:onecut} makes both corresponding actual-law moments positive.
For $t\geq1$, apply \eqref{mid:onecut} to the first actual moment and
the decreasing multipliers
\[
e^{-(t-1)s}/s,\qquad
e^{-(t-1)s}E(qs)/s,\qquad
e^{-(t-1)s}C(qs)/s.
\]
This proves $F_u(t)>0$ and the required signs of the head and tail
derivatives: the head and tail factors $q/(1-e^{-qs})$ and
$q/(e^{qs}-1)$ have derivatives $E(qs)$ and $-C(qs)$, respectively.
For the boundary derivative at $q\geq1$, use the second actual moment
and the decreasing ratio $C(qs)/C(s)$.
The densities have a polynomial bound and all the exponential factors
have positive distance; dominated convergence justifies the derivatives.
The one-step telescope in Lemma~\ref{low:packing} uses exactly these
sampling monotonicities, so proves the claimed signed left bound even
though no positive Laplace density for $F_u$ was assumed.
For the right kernel, $g_u$ is strictly increasing, since
\[
s^2g_u'(s)=1-(1+s+as^2)e^{-s}>0
\]
when $a<1/2$. It starts at $u-1\leq0$ and ends at $z>0$.
Therefore its density has at most one crossing, from negative to
positive. Whenever $G_u(t)\geq0$, multiplication by the increasing
positive functions $s$ and $s^2$ shows that its first and second
Laplace moments are positive. Hence $G_u'(t)<0$ and $G_u''(t)>0$ on
its positive part. For $u<1$, the small-frequency limit $g_u(0+)=u-1$
shows $G_u(t)<0$ eventually, whereas the positive high-frequency tail
makes $G_u(t)>0$ for sufficiently small $t$. The same one-crossing
comparison shows that once $G_u$ has become nonpositive it cannot
become positive again. Thus its positive part is decreasing and convex;
at its zero the derivative jumps upwards to zero. For $u=1$ its density
is positive, giving the same conclusion directly.
The decay needed for packing concerns $J_u$, not the signed kernel
$G_u$. If $u<1$, the crossing just proved makes $J_u$ identically zero
beyond a finite point. If $u=1$, then $g_u(s)=O(s)$ as $s\downarrow0$.
Since $00$.
For an outer geometry $v\geq u$, define
\[
F=\int_0^\infty w_\theta(s)b_0(us)
(\beta_u(s)-\beta_1(s))\dd s>0,
\qquad m_R=u-F/M_u.
\]
The force equations at the bodies imply $m\leq m_R\int_1^\infty U(t)\dd t+\tfrac12U(1)
=k(1)+\tfrac12U(1)$, by strict trapezoidal convexity.
Since $a\leq6/25<1/2$ and $S>0$, the base is strictly negative.
If $G_u(u)\leq0$, decreasing $J_u$ vanishes throughout $[u,\infty)$,
contradicting \eqref{mid:physicalreceiver}.
For an outer source, $v\geq u$ and $r\leq v$ give
$H_{A_u}(v,r)\leq H_{A_u}(v,v)=T_{A_u}(0,v)
\leq T_{A_u}(0,u)$. Insert this in $q\geq0$.
The identity $T_{A_u}(0,u)+T_{B_u}(0,u)=u^2U(u)$ then gives
$mM_u\leq uM_u-F$, so $m\leq m_R$.
The same positive-part packing gives
$H_{J_u}(u+v,r)\leq H_{J_u}(u+v,v)=T_{J_u}(u,v)
\leq T_{J_u}(u,u)$. Since $G_u(u)>0$, the bound on $m$ proves
\eqref{mid:outermesh}.
\end{proof}
The geometric series turns \eqref{mid:H} into the exact integral
\begin{equation}\label{mid:Hdensity}
H(v,r;m)=\int_0^\infty w_\theta(s)
\{b_0(vs)T_u(s)+a_0(vs)\beta_r(s)
-vrse^{-vs}+(m-r)e^{-vs}\}\dd s+(m-r)S_\theta.
\end{equation}
\begin{lemma}[The fourth comparison root]\label{mid:fourthroot}
For $00.
\]
Since $u<1$, the strict step comparison gives
$T_u(s)0.
\]
Since $y\leq x$, use $T_u<\beta_v$, increasing $h_0$ and decreasing $C$
to obtain
\[
H_v+H_m-2D<\int_0^\infty w_\theta(s)N(vs)\dd s-S_\theta,
\quad
N(x)=e^{-x}[-1-2x+xh_0(x)]-2a_0(x)C(x).
\]
We verify its sign on the entire frequency ray. Clearing its positive
denominator gives
\[
e^{2x}(1-e^{-x})^2N(x)
=(x^2-4x+1)e^x+x^2+4x-2+e^{-x}.
\]
The coefficients of degrees zero and one vanish; degrees two, three and
four are $-1,-1,-1/12$, degree five is zero, and for every $j\geq6$
the coefficient multiplied by $j!$ is
$j^2-5j+1+(-1)^j>0$. After division by $x^4$, the negative part is
$-x^{-2}-x^{-1}-1/12$ and the positive part is strictly increasing.
Thus $N$ has exactly one crossing, from negative to positive. Expanding
the geometric denominators, integrating their exponential terms and
collecting the absolutely convergent series gives
\[
\int_0^\infty xN(x)\dd x=1+4\zeta(3)-6\zeta(4)<-0.68.
\]
The last bound follows from $\zeta(3)<1.203$ and $\zeta(4)>1.082$;
these follow by summing the first ten positive terms and bounding the
remaining decreasing series by the two neighboring integral tails.
At a negative-to-positive crossing, multiplication by any nonnegative
decreasing function can only decrease the integral relative to its
value at the crossing. Apply this to $w_\theta(x/v)/(x/v)$.
The complete $w_\theta$ integral of $N(vs)$ is therefore negative.
This proves $0<(H_v+H_m)/D<2$, namely \eqref{mid:diagonalspeed}.
For a fixed $m$, direct differentiation gives
\[
H_m=k(v)+S_\theta,\qquad
H_v(v,r;m)=H_v(v,r;v)+(v-m)U(v).
\]
Here $H_v(v,r;v)$ is a partial derivative evaluated at $m=v$,
not the diagonal total derivative $H_v+H_m$.
The diagonal estimate above holds at every $r\leq v$, not just a root.
Also $vU(v)\leq2k(v)$: integrate the derivative of
$s w_\theta(s)e^{-vs}$ and use $s w_\theta'\leq w_\theta$;
the boundary terms vanish at zero and infinity.
Consequently, whenever $m/v\geq1/2$,
\[
H_v(v,r;m)<2D-(k(v)+S_\theta)+(v-m)U(v)\leq2D.
\]
Its positivity follows directly from the complete density
\[
H_v=\int w_\theta(s)s\{(1-e^{-vs})T_u(s)
+e^{-vs}[v h_0(rs)-m]\}\dd s>0.
\]
Thus $0<\partial_vR=H_v/D<2$ on this domain.
\end{proof}