\subsection{Moment estimates for \texorpdfstring{$2/\sqrt3\le\lambda\le707/500$}{2/sqrt(3) <= lambda <= 707/500}}\label{sec:near-region} We now prove $D_\lambda>0$ for \begin{equation}\label{eq:near-domain} \frac2{\sqrt3}\le\lambda\le\frac{707}{500},\qquad \lambdau^2$, and Lemma~\ref{lem:canonical-moments} gives \begin{equation}\label{eq:near-moment-domain} u0$, \[ \frac1{1-e^{-s}}<\frac1s+\frac12+\frac{s}{12}. \] Indeed, after multiplication by the positive denominator, the difference is \[ (e^s-1)(1-s/2+s^2/12)-s =\sum_{n\ge5}\frac{(n-3)(n-4)}{12n!}s^n>0. \] The function $f(x)=\lambda^{x-1}/(\lambda^x-d)^2$ is the positive exponential mixture \[ f(x)=\lambda^{-1}\sum_{j\ge0}(j+1)d^j e^{-(j+1)x\log\lambda}. \] Tonelli's theorem and the preceding one-variable inequality give \[ \sum_{n\ge1}f(n)<\int_1^\infty f(x)\dd x+\tfrac12 f(1)-\tfrac1{12}f'(1). \] Evaluation of these three terms, using $c+h=c/(1-u)$, yields \[ \kappa<\frac c{\lambda\log\lambda}+\frac{1-u}{2} +\frac{\log\lambda}{12c}(\lambda+1-2\lambda u)(1-u). \] The positive series for $2\operatorname{arctanh}(c/(\lambda+1))$ shows \[ \ell_L<\log\lambda<\ell_U<\frac{c(\lambda+1)}{2\lambda}. \] Since $\lambda+1-2\lambda u>0$, this proves $\kappa\frac2{\lambda+1}+\frac{1-\lambda^{-2}}2>1 \] on the same interval. This proves the asserted range of $K_{\rm cap}$. Divide each moment $a_s$ by the corresponding term of $1,u,\rho\beta_2,\rho\beta_3,\ldots$. The ratios equal one through index two and decrease thereafter. The successive backward ratios of the comparison sequence are $1/u$, $1/z$, and then $1/r_s$, in decreasing order. The covariance argument used in Lemma~\ref{lem:harmonic-speed}, first for finite sums and then by absolute convergence, gives $Q/T>Q_2(\kappa)/T_2(\kappa)$. The formulas follow from \[ \sum_{n\ge1}\lambda^{n-1}\beta_n^2=\frac{u^2\kappa}{1-u}, \qquad \sum_{n\ge1}\lambda^{n-1}\beta_n\beta_{n-1}=\frac u{1-u}. \] The second identity telescopes, since $(\lambda^{n-1}-d)^{-1}-(\lambda^n-d)^{-1} =c\lambda^{n-1}/[(\lambda^{n-1}-d)(\lambda^n-d)]$. For completeness, the denominators needed for this comparison can be checked before taking any quotient. Set \[ R=1+\frac{A^2(K_{\rm cap}-1+u)z^2}{1-u},\qquad Q=2+2\lambda uz+\frac{2Auz^2}{1-u}. \] Both $Q-R$ and $Q-(\lambda+1)uR$ are positive for $0\le z\le1/A$. If their quadratic coefficient is nonnegative, every coefficient is nonnegative and the constant is positive. Otherwise they are concave, and their endpoint values are positive: at $z=1/A$ they are respectively $(2-K_{\rm cap})/(1-u)$ and $[2-(\lambda+1)uK_{\rm cap}]/(1-u)$. The range of $K_{\rm cap}$ proves these signs. Increasing $\kappa$ to $K_{\rm cap}$ increases the speed quotient, so $v\lambda^2u$ and $a(v)=\alpha$, so $v0. \end{split} \] The first summands are nonnegative because $A_i$ increases with $x_i$. The second sum is positive when there are at least two children. Every absolute integrand is bounded by a constant times $\nu$, so finite mean justifies conditioning and summation. Substitution of $F\overset{\mathrm d}=\lambda U$ gives \[ \E[Fp(F)] =\frac{\lambda cu}{1-z} \left[z-(\lambda+1-\lambda z)\frac{a_3}{u}\right]. \] Its strict positivity is precisely~\eqref{eq:near-three-moment-bound}. \end{proof} To deduce positivity of the moment differences $\Delta_s^{\mathrm{ass}}$ below from~\eqref{eq:near-three-moment-bound}, we need the following bounds on $K_{\rm cap}$. \begin{lemma}[Upper bounds on $K_{\rm cap}$]\label{lem:near-cap-arithmetic} Let $2/\sqrt3\le\lambda\le707/500$, $00$, and $K_{\rm cap}<\overline K0,\quad J(3/4)<0,\\ f(1/2)>2,\quad -181000,\quad f'(3/5)>0,\quad f'(3/4)>0. \end{gathered} \] The supplement verifies the signs of these Bernstein coefficients and evaluations. Their denominators are products of positive integers and powers of $\sqrt3$. \end{lemma} \begin{proof} Since $a_6<0$, $J'$ decreases to its positive value at $3/4$. Thus $J$ increases and is negative throughout $[0,3/4]$. It follows that $f'''=6a_3+uJ<0$. In particular, $f''>1000$ on $[0,3/5]$, whence \[ f(u)>2-18|u-1/2|+500(u-1/2)^2\ge919/500. \] On $[3/5,3/4]$, $f'$ is concave with positive endpoint values, hence $f'>0$. Thus $f>919/500$ there too. Positivity of the Bernstein coefficients gives $F_B(\lambda,u)\ge f(u)>0$. The denominator in \[ K_B-\overline K=\frac{F_B}{72\lambda(\lambda R_B+uBP)} \] is positive: indeed \[ R_B=(\lambda+1)^2(1-u)[A\lambda^2(1+u)-(1-\lambda u)^2]+A(2\lambda+1)>0. \] The comparisons $K_{\rm cap}<\overline K$ follow from the two finite identities \[ (1-g^2/3)(1+g^2/3+g^4/5)-1=\frac{g^4(4-3g^2)}{45}>0, \] \[ 1+\frac{g^2}{3(1-g^2)}-\left(1+\frac{g^2}3+\frac{g^4}5+ \frac{g^6}{7(1-g^2)}\right) =\frac{g^4(2/15+2g^2/35)}{1-g^2}>0, \qquad g=\frac c{\lambda+1}. \] Finally, \[ \frac{\lambda A}{B}-\frac{\lambda R_B}{BP} =\frac{\lambda(\lambda+1)^2(1-u)(1-\lambda u)^2}{BP}>0. \] Since $2x/(x+u)$ increases in $x>0$, this gives the last comparison. \end{proof} \begin{lemma}[Positivity of the moment differences]\label{lem:near-condition-f} For $(w,a)$ in~\eqref{eq:near-assigned-pair}, let \[\begin{gathered} Q_a(U)=-\frac wc+\left[1+\frac{(\lambda+1)w}{c}\right]U -\left(a+\frac{\lambda w}{c}\right)U^2,\\ \Delta_s^{\mathrm{ass}}=\E[U^{s-1}Q_a(U)]-(1-a)a_s. \end{gathered}\] Then $\Delta_s^{\mathrm{ass}}>0$ for every real $s\ge2$. \end{lemma} \begin{proof} Retain $R,Q$ from the denominator check in Lemma~\ref{lem:near-prefix}, and put $D_f=Q-R$, $D_w=Q-(\lambda+1)uR$. The only remaining scalar inequality is \begin{equation}\label{eq:near-W} W:=\lambda D_f-uR(\lambda+1-\lambda z)>0. \end{equation} Since $a<1$ and $w\frac{\lambda u}{1+cu}. \] If $u\le5/8$, solving $W>0$ for $K_{\rm cap}$ gives $K_{\rm cap}\frac{15}{8}. \] The last difference has numerator $72\lambda^3+24\lambda^2-119\lambda+24$, positive at $\lambda=1$ and increasing for $\lambda\ge1$. The cubic is consequently less than \[ -2\left\{\lambda^2+u^2[\lambda^3-2\lambda^2-2\lambda-1+(15/8)\lambda u]\right\}<0. \] The expression in braces increases with $\lambda\ge9/8$ and decreases with $0\le u\le5/8$; its value at $(9/8,5/8)$ is $643/8192$. Thus $K_{\rm cal}$ decreases, and its value at $1/A$ is $2\lambda A/(\lambda A+uB)$. Lemma~\ref{lem:near-cap-arithmetic} proves~\eqref{eq:near-W}. If $u\ge5/8$, the identity \[ 2R-K_{\rm cap}Q =2(1-Az)[(1-K_{\rm cap})+(A-(\lambda+1)K_{\rm cap})z]<0 \] gives $R/Q0$, since $3\cdot16476^2-28378^2=9064844$. The linear coefficient is positive by grouping its negative terms with $8564\lambda^4$. The cubic coefficient is negative, the quartic coefficient is $2\lambda(\lambda^2+10\lambda+1)>0$, and the quadratic coefficient plus one eighth of the cubic coefficient is \[ \frac{\lambda^4(400-16\lambda)+389\lambda^3+258\lambda^2+1525\lambda+384}{16}>0. \] This proves the quartic's sign, and hence~\eqref{eq:near-W}, also in this case. Now let $\theta=a_3/a_2$. The formula for $a$ and~\eqref{eq:near-three-moment-bound} give \[ \frac{\Delta_2^{\mathrm{ass}}}{a_2} =a(1-\theta)+\frac wc(\lambda+1-\lambda\theta-1/z) >\frac{(1-z)W}{(\lambda+1-\lambda z)zD_w}>0. \] The direction is correct because the coefficient of $\theta$ is $-(a+\lambda w/c)=-D_f/(zD_w)<0$. Finally $Q_a(y)-(1-a)y$ is a concave quadratic, negative at zero and positive at $1/\lambda$, where its value is $ac/\lambda^2$. It changes sign from negative to positive at its unique zero $\xi\in(0,1/\lambda)$. For every $s\ge2$, \[ \Delta_s^{\mathrm{ass}}-\xi^{s-2}\Delta_2^{\mathrm{ass}} =\E\{U[U^{s-2}-\xi^{s-2}][Q_a(U)-(1-a)U]\}\ge0. \] Thus the bound holds for every real exponent $s\ge2$. \end{proof} \subsubsection{A lower bound before truncating the series} We bound each summand with index $s\ge2$ from below by a positive quantity. This allows us to keep only the first four terms in the subsequent estimate. \begin{lemma}[A series lower bound for $D_\lambda$]\label{lem:near-absorption} For the pair $(w,a)$ in~\eqref{eq:near-assigned-pair} and every integer $s\ge2$, put \[ e_s=1-\lambda r_s,\qquad L_s=\frac1\lambda-r_{s-1},\qquad B_s=\frac{1+w}{\lambda}e_s^2+2(1-a)(r_s-r_{s-1}),\qquad s\ge2. \] Then $B_s,L_s>0$, and \begin{equation}\label{eq:near-root-floor} D_\lambda> \frac{(1+w)(u-\lambda a_2)^2-2(1-a)u(u-\lambda a_2)}\lambda +\sum_{s\ge2}s\lambda^{s-1} [a_s^2B_s+2a_sL_s\Delta_s^{\mathrm{ass}}]. \end{equation} The series converges absolutely. \end{lemma} \begin{proof} Use the integral representation of the moment ratios in Lemma~\ref{lem:canonical-marker} and the series expansion in Lemma~\ref{lem:mellin-floor}. With \[\begin{gathered} \delta_s=a_sr_s-a_{s+1},\qquad \varepsilon_s=a_s-\lambda a_{s+1},\\ \Theta_s=\frac c{\lambda\log\lambda}\beta_s[-\rho'(s)],\quad C_s=\frac c{\lambda\log\lambda}\frac{1-1/(2s)}{r_s}, \end{gathered}\] they give $\delta_s>0$, $\Theta_s>C_s\delta_s$, and \[ \gamma_{s-1}=a_s(r_s-r_{s-1})-\delta_s+\Theta_s, \qquad \gamma_{s-1}+\varepsilon_s/\lambda=a_sL_s+\Theta_s>a_sL_s. \] We claim that \begin{equation}\label{eq:near-age-sign} e_s+(1-\lambda^2u)(C_s-1)>0\qquad(s\ge2). \end{equation} To prove it, set $N_s=cZ_s=\lambda^s(1-u)-1+\lambda u$. The recursion $N_{s+1}=\lambda N_s+c(1-\lambda u)$ and $N_1=c$ give \[ e_s\ge\frac{1-\lambda^2u}{\sum_{j=0}^s\lambda^j}. \] Indeed, $1-\lambda u\ge1-\lambda^2u$ and $N_{s+1}\le\lambda^{s+1}-1=c\sum_{j=0}^s\lambda^j$. Convexity of $1/x$ on $[1,\lambda]$ gives $c/(\lambda\log\lambda)>2/(\lambda+1)$, while $r_s<1/\lambda$. Finally \[ s(\lambda+1)-[s-(s-1)\lambda]\sum_{j=0}^s\lambda^j =\lambda\left[(s-1)+c\sum_{j=1}^{s-1}j\lambda^j\right]>0. \] These three inequalities imply~\eqref{eq:near-age-sign}. If $C_s\ge1$, then $e_s>0$ gives the next inequality directly. If $C_s<1$, use $a>\lambda^2u$ in the same bound. In both cases \[ (1+w)e_s+(1-a)(C_s-1)>0. \] Writing $\xi_s=\Theta_s-C_s\delta_s>0$, expansion now gives \[ \begin{split} \frac{1+w}{\lambda}\varepsilon_s^2+2(1-a)a_s\gamma_{s-1} ={}&a_s^2B_s+2a_s\delta_s[(1+w)e_s+(1-a)(C_s-1)]\\ &+(1+w)\lambda\delta_s^2+2(1-a)a_s\xi_s>a_s^2B_s. \end{split} \] Here $r_s-r_{s-1}>0$ follows from $Z_s^2-Z_{s-1}Z_{s+1}=(1-u)\lambda^{s-1}(1-\lambda u)>0$. To identify the term with $s=1$ and bound those with $s\ge2$, write the coefficient $J_{s-1}$ as \[ J_{s-1}=2ct_{s-1}+c[(1+w)\varepsilon_s-2(1-a)a_s], \qquad t_{s-1}=(1-a)a_s+\Delta_s^{\mathrm{ass}}. \] For $s\ge2$ the term in the series for $\widetilde F(w)$, before multiplication by $s\lambda^{s-1}$, is therefore \[ \frac{1+w}{\lambda}\varepsilon_s^2+2(1-a)a_s\gamma_{s-1} +2(\gamma_{s-1}+\varepsilon_s/\lambda)\Delta_s^{\mathrm{ass}}. \] At $s=1$, $t_0=0$, and the summand is the first term of~\eqref{eq:near-root-floor}. For $s\ge2$, Lemma~\ref{lem:near-condition-f} gives $\Delta_s^{\mathrm{ass}}>0$, so we may bound its coefficient from below by $2a_sL_s$. Lemma~\ref{lem:assigned-floor} gives $D_\lambda\ge F_*=\widetilde F(v)>\widetilde F(w)$, proving the strict inequality. All these rearrangements can first be made in finite sums. At fixed $\lambda>1$, $a_s\le\lambda^{-s}$ and $\beta_s\le C_\lambda\lambda^{-s}$. Moreover $\E[U^s|\log U|]\le C_\lambda\lambda^{-s}$ for all sufficiently large $s$, by maximizing $x^s|\log x|$ on $[0,1/\lambda]$. The same bounds hold for $\Theta_s,\gamma_{s-1},\Delta_s^{\mathrm{ass}}$ up to a fixed factor, and $t_{s-1}=O_\lambda(\lambda^{-s})$. Every term above is thus bounded by $C_\lambda s\lambda^{-s}$. Dominated summation proves the asserted convergence and all passages to the infinite sum. \end{proof} \begin{lemma}[A lower bound using the first four terms]\label{lem:near-four-age} Let $T=1-\sqrt{1-\rho}$ and $f_s=T^{s-1}[s-(s-1)T]$. Define \[ M_s=a(1-r_s)+\frac wc\left[\lambda+1-\lambda r_s- \frac1{r_{s-1}f_s}\right],\qquad s\ge2, \] and \begin{equation}\label{eq:near-S4} \frac{S_4}{u^2}= \frac{(1+w)(1-\lambda\rho/A)^2-2(1-a)(1-\lambda\rho/A)}\lambda +\sum_{s=2}^4\frac{s\lambda^{s-1}f_s^2}{Z_s^2}[B_s+2L_sM_s]. \end{equation} Then $D_\lambda>S_4$. \end{lemma} \begin{proof} The integral representation of $\rho_s$ with respect to $\sigma$ gives \[ \rho_s=\E_\sigma[t^{s-1}(s-(s-1)t)],\qquad \E_\sigma(1-t)^2=1-\rho. \] As a function of $x=(1-t)^2$, the first integrand is $(1-\sqrt x)^{s-1}[1+(s-1)\sqrt x]$. Its derivative is $-s(s-1)(1-\sqrt x)^{s-2}/2$, an increasing function for integer $s\ge2$. Jensen's inequality therefore gives $a_s\ge\beta_sf_s$, with equality at $s=1,2$. Also $a_{s-1}\le\beta_{s-1}$ and $a_{s+1}/a_sM_s$. Lemma~\ref{lem:near-condition-f} gives $\Delta_s^{\mathrm{ass}}>0$. For $s\ge2$ the expression in square brackets in~\eqref{eq:near-root-floor} is therefore bounded below by \[ (\beta_sf_s)^2[B_s+2L_s\max\{0,M_s\}]. \] Since $B_s,L_s>0$, the bracket is positive when we use $a_s\ge\beta_sf_s$. We can now drop the terms with $s\ge5$ and replace $\max\{0,M_s\}$ by $M_s$ for $s=2,3,4$. This gives~\eqref{eq:near-S4}. \end{proof} \subsubsection{Positivity of \texorpdfstring{$S_4$ for $00$ by estimating~\eqref{eq:near-S4} on a larger set of scalar parameters. The comparison $D_\lambda>S_4$ from Lemma~\ref{lem:near-four-age} uses the moments of $U$ under the given escape law. Temporarily replace $K_{\rm cap}$ in~\eqref{eq:near-assigned-pair} by a scalar $K$. In the variables $\lambda,u,\rho$, put \begin{equation}\label{eq:near-finite-calibration} \begin{gathered} R=1-u+\rho^2(K-1+u),\qquad Q=A(1-u)+\lambda u\rho(1-u)+u\rho^2,\\ D_K=2Q-(\lambda+1)uAR,\qquad A_0^\mathrm{num}=[1-(\lambda+1)u]A+\lambda u\rho,\\ w=\frac{cuAR}{D_K},\qquad a=\frac{A(D_K-RA_0^\mathrm{num})}{\rho D_K},\qquad \Phi(K)=\rho D_K S_4(K)/u^2. \end{gathered} \end{equation} Define the explicit quadratic bound \begin{equation}\label{eq:near-Kb} K_b=\frac{\lambda+1}{2\lambda}+\frac{1-u}{2} +\frac{(\lambda+1)(\lambda+1-2\lambda u)(1-u)}{24\lambda}. \end{equation} The inequalities for $\ell_L,\ell_U$ give $K_{\rm cap}\Phi(K_b). \] \end{lemma} \begin{proof} First $(1+\lambda u)K_b<2$. Indeed this product increases with $u$ on $[0,1/2]$, as follows by differentiating its displayed cubic; its derivative is at least $(-2\lambda^3+9\lambda^2-2\lambda-1)/(24\lambda)>0$. At $u=1/2$ the remaining difference is \[ 2-(1+\lambda/2)K_b(\lambda,1/2) =\frac{6+c(19-37c)}{96\lambda}>0. \] Direct subtraction now gives \begin{equation}\label{eq:near-positive-D} \frac{D_{K_b}}{A(1-u)}-R =(1-\rho)\left[(1-\lambda u)(1+\rho)+\frac{2\lambda u\rho}{A}\right] +\frac{2-(1+\lambda u)K_b}{1-u}\rho^2>0. \end{equation} It also proves positivity at the smaller $K_{\rm cap}$. Here is the full dependence on $K$. Put $E=1-\lambda\rho/A$ and $W_s=s\lambda^{s-1}/Z_s^2$. The cancellation $r_s-r_{s-1}=L_s(1-r_s)$ in~\eqref{eq:near-S4} gives \[ S_4/u^2=F_0+aF_a+wF_w, \] where \[ \begin{split} F_0&=(E^2-2E)/\lambda+ \sum_{s=2}^4W_sf_s^2[e_s^2/\lambda+2(r_s-r_{s-1})],\\ F_a&=2E/\lambda,\\ F_w&=E^2/\lambda+ \sum_{s=2}^4W_s\left\{f_s^2\left[\frac{e_s^2}{\lambda} +\frac{2L_s}{c}(\lambda+1-\lambda r_s)\right] -\frac{2L_sf_s}{cr_{s-1}}\right\}. \end{split} \] Set $\widehat C=\rho F_0+AF_a$. Substitution of~\eqref{eq:near-finite-calibration} gives \[ \Phi(K)=2Q\widehat C-AR\widehat H, \qquad \widehat H=(\lambda+1)u\widehat C+A_0^\mathrm{num}F_a-cu\rho F_w. \] Its sign is exposed by the exact cancellation \[ \widehat H=\frac{2E}{\lambda}\left(A-\frac{\lambda u\rho^2}{A}\right) +2u\rho\sum_{s=2}^4\frac{W_sL_sf_s}{r_{s-1}} [1-\lambda r_{s-1}r_s^2f_s]>0. \] Here $00$. Thus $\partial_K\Phi=-A\rho^2\widehat H<0$, proving the claim. Only the scalar argument $K$ of $\Phi$ varies in this comparison. \end{proof} Henceforth $K=K_b$ and $\eta=K_b-1+u$. Define \begin{equation}\label{eq:near-complete-finite-profile} \begin{gathered} \Phi_R(\rho)=\frac2\lambda\left(1-\frac{\lambda\rho}{A}\right) \left\{Q\left(2A-\rho-\frac{\lambda\rho^2}{A}\right) -AR\left(A-\frac{\lambda u\rho^2}{A}\right)\right\},\\ \omega_s=\frac{s\lambda^{s-2}(1-\lambda u)}{Z_s^3},\qquad V_s=\frac{\omega_suAR}{r_{s-1}},\\ U_s=\omega_s[Q(2-r_s-\lambda r_s^2)+uAR\lambda r_s^2],\qquad s=2,3,4. \end{gathered} \end{equation} All $U_s,V_s$ are positive. Substituting the formulas for $a,w$ in~\eqref{eq:near-S4} and multiplying both sides by $\rho D_{K_b}$ gives \begin{equation}\label{eq:near-complete-Phi} \Phi(K_b)=\Phi_R+2\rho\sum_{s=2}^4(U_sf_s^2-V_sf_s). \end{equation} Here $\Phi_R$ comes from the term with $s=1$. Since $\rho^2-f_3=T^2(1-T)^2$ and $f_3^2-\rho f_4=T^4(1-T)^2$, it also gives \begin{equation}\label{eq:near-merge} \begin{split} \Phi(K_b)=\Phi_R+2\rho\{&(U_2-V_3)\rho^2-V_2\rho +(\rho U_3-V_4)f_4+U_4f_4^2\\ &+T^2(1-T)^2[V_3+U_3T^2]\}. \end{split} \end{equation} \begin{lemma}[Bounds on $U_2-V_3$ and $\rho U_3-V_4$]\label{lem:near-age-comparisons} In the present range, \[ U_2-V_3>\frac{(1-\lambda u)Q}{A^3}>0. \] For $1/2\le\rho\le1$, one also has $\rho U_3-V_4>0$. \end{lemma} \begin{proof} For the first claim, let $x=uAR/Q$. Equation~\eqref{eq:near-positive-D} implies $x<2u/(1+\lambda u)\le40/63$. The identities in~\eqref{eq:near-profile-defs} give \[ U_2-V_3=\frac{2(1-\lambda u)Q}{A^3} [2-r_2-\lambda r_2^2(1+x/2)]. \] The derivative $\partial_u r_2=\lambda^2/Z_3^2$ is positive. At $u=1/2$, direct multiplication gives $r_2<5/7$ and $\lambda r_2^2<7/12$. For the latter sign, the cleared difference is \[ 7(\lambda^2+\lambda+2)^2-12\lambda(\lambda+2)^2 =4-12c+35c^2+30c^3+7c^4>0, \] because $35c^2-12c+4=35(c-6/35)^2+104/35$. For the former, $5\lambda^2-2\lambda-4>0$ at $23/20$ and increases. The bracket is therefore larger than $2-5/7-(7/12)(1+20/63)=391/756>1/2$. For the second claim, $r_3$ increases with $u$, and at $u=1/2$ its value is $(\lambda^2+\lambda+2)/(\lambda^3+\lambda^2+\lambda+2)$. Thus $r_3<3/4$ and $\lambda r_3^2<2/3$: the first cleared difference is $3\lambda^3-\lambda^2-\lambda-2>0$, and the second is \[ 2+29c^2+47c^3+35c^4+13c^5+2c^6>0. \] Consequently $2-r_3-\lambda r_3^2>7/12$. The ratio $K_b/(1-u)$ decreases as $1-u$ increases, giving \begin{equation}\label{eq:near-K-ratio} \frac{K_b}{1-u}\le\frac{31}{12},\qquad \frac R{1-u}\le1+\frac{19}{12}\rho^2. \end{equation} At $\rho=1/2$, \[ \frac{uAR}{Q}\le\frac{67}{16}\frac{u(1-u)}{3-2u}\le\frac{67}{128}. \] The last inequality is $(2u-1)(4u-3)\ge0$. The exact comparison is \[ \rho U_3-V_4=\omega_3\left\{\rho Q(2-r_3-\lambda r_3^2) -uA\lambda r_3^2R(4/3-\rho)\right\}. \] At $1/2$, the expression in braces divided by $Q$ exceeds $7/24-(5/6)(67/128)(2/3)=1/1152$. As $\rho$ increases, $\rho Q$ increases, whereas \[ \frac1{1-u}\frac\dd{\dd\rho}[R(4/3-\rho)] =-1+\frac\eta{1-u}(8\rho/3-3\rho^2)\le-11/144<0. \] Indeed the quadratic is at most $7/12$ on $[1/2,1]$, and $0<\eta/(1-u)\le19/12$. This proves the claim throughout that interval. \end{proof} \begin{lemma}[The range $0<\rho\le2/3$]\label{lem:near-low-rho} One has $\Phi(K_b)>0$ whenever $0<\rho\le2/3$. \end{lemma} \begin{proof} First suppose $\rho\le1/2$. The negative terms in~\eqref{eq:near-complete-Phi} are bounded using $f_s\le\rho^{s-1}$ and \begin{equation}\label{eq:near-price-constants} \frac{W_2L_2}{r_1}\le\frac12,\qquad \frac{W_3L_3}{r_2}\le\frac14,\qquad \frac{W_4L_4}{r_3}\le\frac4{25}. \end{equation} Here $W_s=s\lambda^{s-1}/Z_s^2$. To verify all three bounds, put $k=1-\lambda u$ only in this calculation. Then $A\ge1+k$ and $A\ge2-k$. The first bound is $2k/A^2\le1/2$. For the next two, \[ Z_3^2-\lambda(A+k)^2=c(\lambda A^2-k^2)>0, \qquad Z_4-\lambda(A+2k)=c(\lambda A-k)>0. \] They reduce to $3k/[A(A+k)^2]\le1/4$ and $4k/[(A+k)(A+2k)^2]\le4/25$. For $k\le1/2$, use $A\ge2-k$ and monotonicity of the resulting one-variable fractions. For $k\ge1/2$, use $A\ge1+k$ and \[ (1+2k)^2(1+k)-12k=(2k-1)(2k^2+5k-1), \] \[ (1+3k)^2(1+2k)-25k=(2k-1)(9k^2+15k-1). \] For the term $\Phi_R$ we have \[ \frac{\Phi_R}{A(1-u)}\ge \frac2\lambda\left[\frac{29}{48}(A-\lambda\rho)-\rho+ \frac{\lambda^2\rho^3}{A^2}\right]. \] This follows from $Q\ge A(1-u)$, $R/(1-u)\le67/48$, and the identity $(1-\lambda\rho/A)(AD-\rho-\lambda\rho^2/A) =D(A-\lambda\rho)-\rho+\lambda^2\rho^3/A^2$. The bracket decreases with $\rho\le1/2$ and, at $\rho=1/2$, increases with $A\ge1+\lambda/2$. Hence \[ \frac{\Phi_R}{A(1-u)}\ge\frac5{24\lambda}+\frac\lambda{(\lambda+2)^2} \ge\frac{461}{1764}. \] The last function decreases on $[1,3/2]$, as is seen by differentiation. By~\eqref{eq:near-price-constants}, the sum of the absolute values of the negative terms satisfies \[ \frac{2\rho\sum_{s=2}^4V_sf_s}{A(1-u)} \le\frac{67}{48}\left(\frac18+\frac1{32}+\frac1{100}\right) =\frac{8911}{38400}. \] Deleting only the positive squares in~\eqref{eq:near-complete-Phi} gives $\Phi/[A(1-u)]>1/40$. Now suppose $1/2\le\rho\le2/3$. In~\eqref{eq:near-merge}, Lemma~\ref{lem:near-age-comparisons} gives $U_2-V_3>0$ and $\rho U_3-V_4>0$. The only negative term left is $-2\rho^2V_2$, with \[ \frac{2\rho^2V_2}{A(1-u)}<\frac{322}{1215}, \] because \[ \frac1{\lambda^2(\lambda+1)}- 2u\frac{W_2L_2}{r_1} =\frac{[\lambda-(2\lambda+1)(1-\lambda u)]^2} {\lambda^2(\lambda+1)A^2}\ge0, \] $1/[\lambda^2(\lambda+1)]<7/20$, and $R/(1-u)\le46/27$. To bound $\Phi_R$, we retain the positive terms of $Q$ that depend on $u$. The explicit formulas give \[ \frac Q{A(1-u)}\ge1+\frac u{1-u}\frac{\rho(1+2\rho)}3, \qquad \frac R{1-u}\le1+\rho^2\left(\frac{17}{24}+\frac u{1-u}\right). \] After substitution in~\eqref{eq:near-complete-finite-profile}, the coefficient of $u/(1-u)$ inside the braces for $\Phi_R/[A(1-u)]$ is at least \[ \frac\rho3\{A(2+\rho)-\rho(1+2\rho)(1+\rho)\}>\frac7{20}. \] To check the last bound, set $A=63/40$, where the expression is concave in $\rho\in[1/2,2/3]$ and has endpoint values $13/32$ and $434/1215$. Both exceed $7/20$. Since $E\ge3/7$ and $2/\lambda\ge4/3$, this part contributes at least $u/[5(1-u)]$. For the other part, use \[ g(A,\rho)=\frac{37}{54}(A-\lambda\rho)-\rho+\frac{\lambda^2\rho^3}{A^2}. \] Its $\rho$ derivative is negative and its $A$ derivative exceeds $3/8$. It follows that \[ \frac{\Phi_R}{A(1-u)}\ge \frac1{27\lambda}-\frac{37}{162}+\frac{64\lambda}{27(\lambda+2)^2} +\frac34(1/2-u)+\frac15\frac u{1-u}. \] The first three terms increase with $\lambda\in[1,3/2]$ and are at least $35/486$. For the monotonicity, $\lambda^2(2-\lambda)/(\lambda+2)^3$ decreases to $9/343$, so its contribution to the derivative is still larger than the negative term. Finally, with $x=1/2-u$, \[ \frac34(1/2-u)+\frac15\frac u{1-u}-\frac{199}{1000} =\frac{(3/2)(x-2/125)^2+77/125000}{1+2x}>0. \] Thus $\Phi/[A(1-u)]>35/486+199/1000-322/1215 =1457/243000>1/200$. \end{proof} \subsubsection{Bounds for \texorpdfstring{$2/3\le\rho\le5/6$}{2/3 <= rho <= 5/6}} Define the finite functions \begin{equation}\label{eq:near-intermediate-functions} F(\rho,\lambda,u)=\Phi_R(\rho)-\frac{4\rho^2u(1-\lambda u)R}{A}, \qquad g(\rho,\lambda,u)=\frac{(1-\lambda u)Q}{A^3}. \end{equation} Equation~\eqref{eq:near-merge} and Lemma~\ref{lem:near-age-comparisons} imply \begin{equation}\label{eq:near-intermediate-lower} \Phi(K_b)>F(\rho,\lambda,u)+2\rho^3g(\rho,\lambda,u) \qquad(1/2\le\rho<1). \end{equation} \begin{lemma}[Bounds for $F$ and $g$ at $\rho=3/4$ and $\rho=5/6$]\label{lem:near-intermediate-arithmetic} On $23/20\le\lambda\le3/2$, $0\le u\le1/2$, the explicitly defined rational functions in~\eqref{eq:near-intermediate-functions} satisfy \[ -\partial_uF(3/4,\lambda,u)>0,\quad -\partial_uF(5/6,\lambda,u)>0,\quad -\partial_ug(3/4,\lambda,u)>0. \] Their boundary values satisfy \[ F(3/4,\lambda,1/2)>1/8,\qquad F(5/6,\lambda,1/4)>0, \] \[ F(5/6,\lambda,1/2)+\frac{27}{32}g(3/4,\lambda,1/2)>1/100. \] These are finite polynomial signs: the supplement forms each displayed difference from~\eqref{eq:near-intermediate-functions}, clears its positive powers of $\lambda$ and $A$, maps the stated rational rectangle to $[0,1]^2$, and lists its complete Bernstein coefficients. The exact positive factors, degrees and coefficient transform are recorded in Lemma~\ref{lem:near-finite-certificate}; no numerical evaluation is used to infer an interval sign. \end{lemma} \begin{lemma}[The range $2/3\le\rho\le5/6$]\label{lem:near-intermediate-rho} One has $\Phi(K_b)>0$ throughout this range. \end{lemma} \begin{proof} We first show that the expression in braces in $\Phi_R$ decreases with $\rho$. The explicit quadratic $K_b$ gives \begin{equation}\label{eq:near-eta-bounds} 0<\eta<3/4,\qquad 10$. Its maximum endpoint value is $(\lambda+1)(\lambda^2+2\lambda+13)/(24\lambda) \le(35-3\lambda)/24<4/3$. The bracket in $\Phi_R$ has exactly the five $\rho$ coefficients \[ \begin{split} &A^2(1-u),\quad A(1-u)(2\lambda u-1),\quad 2Au-\lambda(1-u)-A^2\eta,\\ &-u[1+\lambda^2(1-u)/A],\qquad (\lambda u/A)(A\eta-1). \end{split} \] Its derivative is negative on $[2/3,5/6]$. Indeed the quadratic coefficient is less than $-\lambda(1-u)\le-\lambda/2$. If the linear coefficient is positive, it is less than $2/3$, whereas twice $\rho$ times the quadratic coefficient is at most $-23/30$. The cubic and quartic contributions to the derivative sum to \[ u\rho^2\{-3[1+\lambda^2(1-u)/A]+4\lambda\rho(\eta-1/A)\}\le0, \] using $\lambda\rho\le5/4$, $1/A\ge2/5$, and~\eqref{eq:near-eta-bounds}. The first pair is already strictly negative, including $u=0$. For $2/3\le\rho\le3/4$, Lemma~\ref{lem:near-intermediate-arithmetic} gives $F(3/4,\lambda,u)>1/8$. Since $F\le\Phi_R$, the bracket in $\Phi_R$ is positive at $3/4$ and therefore positive to its left. Both that bracket and $1-\lambda\rho/A$ decrease, so $\Phi_R$ decreases. The term $4\rho^2u(1-\lambda u)R/A$ subtracted from $\Phi_R$ increases, so $F$ decreases as well. Thus $\Phi>F(\rho)\ge F(3/4)>1/8$. For $3/4\le\rho\le5/6$, the same finite lemma gives \[ F(5/6,\lambda,u)+\frac{27}{32}g(3/4,\lambda,u)>1/100. \] We also need $\Phi_R(5/6)>0$. If $u\le1/4$, this follows from $F(5/6,\lambda,1/4)>0$ and the $u$ derivative. If $u\ge1/4$, concavity of $uA(1-u)$ and its two endpoint values give $uA(1-u)>1/3$. Since $R(5/6)\ge1-u$ and $Q(3/4)\le Q(1)=A$, \[ \frac{4(5/6)^2u(1-\lambda u)R(5/6)}A >\frac{27}{32}g(3/4). \] After division by $2(1-\lambda u)/A^3$, this follows from $25/54>27/64$. The preceding mixed bound now gives $\Phi_R(5/6)>0$. As before, $\Phi_R(5/6)>0$ implies that $F$ decreases with $\rho$ on $[3/4,5/6]$. The function $g$ increases with $\rho$, since $Q$ does. Equation~\eqref{eq:near-intermediate-lower} therefore gives \[ \Phi(K_b)>F(5/6)+\frac{27}{32}g(3/4)>1/100. \] \end{proof} \subsubsection{Bounds for \texorpdfstring{$5/6\le\rho<1$}{5/6 <= rho < 1}} For $5/6\le\rho<1$, we need a lower bound that also applies when the quadratic term in $1-T$ is negative. We estimate it together with the value of $\Phi(K_b)$ at $\rho=1$ and the cubic and quartic terms in $1-T$. Put $n_s=s(s-1)/2$ and $B_s^{\rm age}=2U_s-V_s$. The following are finite polynomials in $\rho$: \begin{equation}\label{eq:near-core-defs} \begin{gathered} \Psi(\rho)=\Phi_R(\rho)+2\rho\sum_{s=2}^4(U_s-V_s),\qquad p=\Psi(1),\\ L_{\rm core}(\rho)=\frac{\Psi(\rho)-p}{1-\rho} -2\rho\sum_{s=2}^4n_s B_s^{\rm age}(\rho) =\sum_{j=0}^4c_j\rho^j,\qquad C_0=L_{\rm core}(1),\\ H(\rho)=-c_2-(1+\rho)c_3-(1+\rho+\rho^2)c_4. \end{gathered} \end{equation} The quotient is taken by polynomial division, including at $\rho=1$. In particular, \begin{equation}\label{eq:near-core-chord} L_{\rm core}(\rho)=(1-\rho)L_{\rm core}(0)+\rho C_0+ \rho(1-\rho)H(\rho). \end{equation} Define the three further rational functions \begin{equation}\label{eq:near-recovery-defs} \begin{gathered} b_{\rm env}=\frac53\left[2B_3^{\rm age}(5/6)+\frac{27}{4}B_4^{\rm age}(5/6)\right],\\ j_{\rm env}=\frac53\left[U_2(5/6)+\frac{169}{36}U_3(5/6) +\frac{2601}{256}U_4(5/6)\right],\\ K_{5/6}=L_{\rm core}(0)-C_0+\frac56H(5/6),\\ J_1=\frac{25p+C_0}{(1-\lambda u)A},\qquad J_2=4b_{\rm env}+3C_0,\qquad J_3=K_{5/6}+j_{\rm env}+4C_0. \end{gathered} \end{equation} Here $U_s$ and $V_s$ for $s=2,3,4$ are evaluated at the same $K_b$ and $u$. The following Bernstein coefficient bounds also give the inequalities of Lemma~\ref{lem:near-intermediate-arithmetic}. To clear the denominators, write \[ D_{\rm profile}=A^3Z_3^3Z_4^3Z_5^2. \] For the two $J_1$ boundary derivatives, use \[ \begin{split} D_0&=12\lambda^3 Z_3(\lambda,0)^4Z_4(\lambda,0)^4Z_5(\lambda,0)^3,\\ D_{1/2}&=24\lambda^3(2-\lambda)^2(\lambda+2)^5 (\lambda^2+\lambda+2)^4(\lambda^3+\lambda^2+\lambda+2)^4 (\lambda^4+\lambda^3+\lambda^2+\lambda+2)^3. \end{split} \] Every one of these factors is strictly positive on the intervals below. \begin{lemma}[Finite coefficient signs]\label{lem:near-finite-certificate} For each row of the following table, let $P$ be the function in its first column multiplied by the positive denominator in its second column. Then $P$ is a polynomial of the indicated bidegree, and every Bernstein coefficient of $P$ on the indicated rectangle $[a,b]\times[0,1/2]$ is strictly positive. The functions $F,g$ are exactly~\eqref{eq:near-intermediate-functions}, and $p,c_j,J_i$ are exactly~\eqref{eq:near-core-defs}--\eqref{eq:near-recovery-defs}. \begin{center}\small \setlength{\tabcolsep}{3pt} \begin{tabular}{@{}llll@{}} Function & Positive denominator & Bidegree & $[a,b]$\\ \hline $-\partial_uF(3/4)$ & $4096\lambda^2A^3$ & $(7,6)$ & $[23/20,3/2]$\\ $F(3/4,\lambda,1/2)-1/8$ & $8192\lambda^2(\lambda+2)^2$ & $(5,0)$ & $[23/20,3/2]$\\ $-\partial_uF(5/6)$ & $93312\lambda^2A^3$ & $(7,6)$ & $[23/20,3/2]$\\ $-\partial_ug(3/4)$ & $16A^4$ & $(3,2)$ & $[23/20,3/2]$\\ $F(5/6,\lambda,1/4)$ & $497664\lambda^2(3\lambda+4)^2$ & $(6,0)$ & $[23/20,3/2]$\\ $\begin{gathered}F(5/6,\lambda,1/2)\\+\frac{27}{32}g(3/4,\lambda,1/2)-\frac1{100}\end{gathered}$ & $4665600\lambda^2(\lambda+2)^3$ & $(6,0)$ & $[23/20,3/2]$\\ $-c_2$ & $12\lambda D_{\rm profile}$ & $(29,14)$ & $[23/20,10/7]$\\ $-(c_3+4c_4)/u$ & $12\lambda D_{\rm profile}$ & $(28,13)$ & $[23/20,10/7]$\\ $-\partial_u^2J_1$ & $6\lambda(1-\lambda u)^3A^6Z_3^5Z_4^5Z_5^4$ & $(51,23)$ & $[23/20,707/500]$\\ $\partial_\lambda J_1(\lambda,0)-1/4$ & $D_0$ & $(35,0)$ & $[23/20,707/500]$\\ $\partial_\lambda J_1(\lambda,1/2)$ & $D_{1/2}$ & $(40,0)$ & $[23/20,707/500]$\\ $J_2$ & $216\lambda^2D_{\rm profile}$ & $(30,15)$ & $[23/20,3/2]$\\ $J_3-(1-\lambda u)/20$ & $2488320\lambda^2D_{\rm profile}$ & $(30,15)$ & $[23/20,10/7]$\\ $p-1/1000$ & $3000\lambda^2D_{\rm profile}/A$ & $(29,14)$ & $[15/13,707/500]$ \end{tabular} \end{center} The division by $u$ in the eighth row is exact polynomial cancellation before setting $u=0$. The two additional rational evaluations are \[ J_1(23/20,0)>57/100, \qquad J_1(23/20,1/2)>21/25. \] More explicitly, the finite coefficient obligation in each row is as follows. If \[ P(a+(b-a)x,y/2)=\sum_{k=0}^M\sum_{l=0}^Np_{kl}x^ky^l, \] then for every $0\le i\le M$, $0\le j\le N$, \begin{equation}\label{eq:near-Bernstein-obligation} \sum_{k=0}^i\sum_{l=0}^j p_{kl}\frac{\binom{i}{k}}{\binom{M}{k}} \frac{\binom{j}{l}}{\binom{N}{l}}>0. \end{equation} The supplement supplies the full coefficients and the exact rational verification of these finite inequalities. \end{lemma} \begin{proof} Only finite addition, multiplication and division by positive integers are needed to generate~\eqref{eq:near-Bernstein-obligation} from the displayed definitions. The complete generated data are provided in the supplement. Once these finite signs are checked, the passage to the entire rectangle is immediate from \[ P(a+(b-a)x,y/2)=\sum_{i=0}^M\sum_{j=0}^N b_{ij} \binom Mi x^i(1-x)^{M-i}\binom Nj y^j(1-y)^{N-j}. \] The basis functions are nonnegative and sum to one, including on the boundary. Positivity of all $b_{ij}$ proves every claimed sign. All denominators listed in the table are positive; thus no multiplication changes an inequality's direction. \end{proof} \begin{lemma}[The range $5/6\le\rho<1$]\label{lem:near-high-rho} One has $\Phi(K_b)>0$ throughout this range. \end{lemma} \begin{proof} For $s=2,3,4$ and $u>0$, all three coefficients of $B_s^{\rm age}$ as a polynomial in $\rho$ are positive. Dividing them by $\omega_s>0$ gives \[ \begin{split} &[\rho^0]:\ A(1-u)[2(2-r_s-\lambda r_s^2)+u(2\lambda r_s^2-1/r_{s-1})],\\ &[\rho^1]:\ 2\lambda u(1-u)(2-r_s-\lambda r_s^2),\\ &[\rho^2]:\ u[2(2-r_s-\lambda r_s^2)+A\eta(2\lambda r_s^2-1/r_{s-1})]. \end{split} \] Here $r_2<5/7$, $r_3<3/4$, $r_4<5/6$, $1/r_2\le19/10$, $1/r_3\le65/38$, and $r_s>1/2$. These bounds follow from monotonicity in $u$ and $\lambda$ and the finite recursion for $Z_s$; for $r_4<5/6$, the endpoint difference is $c(5\lambda^3+4\lambda^2+3\lambda+2)>0$. The constant brackets are respectively larger than $1/7,1/20,7/114$. The linear coefficients are positive. For $s=3,4$, if $2\lambda r_s^2-1/r_{s-1}<0$, then~\eqref{eq:near-eta-bounds} bounds the brackets in the quadratic coefficients below by $2/15$ and $1/19$, respectively. If $2\lambda r_s^2-1/r_{s-1}\ge0$, both terms in the bracket are nonnegative and the first is positive. At $s=2$, put $\overline H=(35-3\lambda)/24$. Since $A\eta\le\overline H$, its bracket is at least \[ f(A)=4-2r_2+2\lambda(\overline H-1)r_2^2-\overline H A, \qquad r_2=\frac A{(\lambda+1)A-\lambda}. \] This function decreases with $A$, since $f'(A)\le2\lambda/Z_3^2-\overline H<3/4-61/48<0$. At $A=\lambda+1$, its numerator after multiplication by $24(1+\lambda+\lambda^2)^2$ is \[ 3c^6-8c^5-85c^4-206c^3-169c^2+14c+64>0. \] Indeed $00$. This proves the claimed signs of the coefficients of $B_s^{\rm age}$. The coefficient of $\rho^5$ in $\Phi_R$ gives \[ c_4=-[\rho^5]\Phi_R=\frac{2\lambda u}{A^2}(A\eta-1)>0. \] Lemma~\ref{lem:near-finite-certificate} gives $c_2<0$ and $c_3+4c_4<0$. Therefore $H$ in~\eqref{eq:near-core-defs} is positive and increases on $[0,1]$. By~\eqref{eq:near-core-chord}, the chord joining the endpoint values of $L_{\rm core}$ is a lower bound for $L_{\rm core}$. Let $x=1-T$, so $\rho=1-x^2$ and $0\le x\le1/\sqrt6<5/12$. Directly, \[ f_2=1-x^2,\quad f_3=1-x^2(3-2x),\quad f_4=1-x^2(6-8x+3x^2). \] Substitution in the complete expression~\eqref{eq:near-complete-Phi} gives the exact identity \begin{equation}\label{eq:near-recovery-identity} \Phi=p+x^2[L_{\rm core}(\rho)+xB_*+x^2J_*], \end{equation} where \[ B_*=2\rho[2B_3^{\rm age}+(8-3x)B_4^{\rm age}], \] \[ J_*=2\rho[U_2+U_3(3-2x)^2+U_4(6-8x+3x^2)^2]. \] Their coefficients are positive. Since $\rho\ge5/6$ and $x\le5/12$, they satisfy $B_*\ge b_{\rm env}$ and $J_*\ge j_{\rm env}$. Combining~\eqref{eq:near-core-chord} and the increase of $H$ gives \begin{equation}\label{eq:near-quartic} \Phi\ge p+C_0x^2+b_{\rm env}x^3+(K_{5/6}+j_{\rm env})x^4. \end{equation} The finite signs now have three distinct roles. First $p>1/1000$; its rectangle begins at $15/13<2/\sqrt3$. Second, $J_1$ is concave in $u$, and both boundary functions increase with $\lambda$. Its two rational endpoint values therefore imply \[ J_1(\lambda,u)\ge(1-2u)J_1(\lambda,0)+2uJ_1(\lambda,1/2)>57/100. \] In particular $25p+C_0>0$. Third, $J_2>0$ and $J_3>(1-\lambda u)/20>0$ give \[ b_{\rm env}>-3C_0/4,\qquad K_{5/6}+j_{\rm env}>-4C_0. \] The complete bound~\eqref{eq:near-quartic} is consequently larger than \[ p+C_0(x^2-3x^3/4-4x^4). \] On $0\le x\le1/\sqrt6$, set $G(x)=x^2-3x^3/4-4x^4$. Then \[ 0\le G(x)<1/25,\qquad 1-25G(x)\ge37639/640000. \] Here $G\ge0$ follows from $x<33/80$ and $1-3(33/80)/4-4(33/80)^2=1/100$. For the other bound, $1-25G$ decreases up to $1/4$ and has value $31/256$ there. Above $1/4$ its second derivative exceeds $50$. Its value and first derivative at $3/10$ are $53/800$ and $69/80$; the quadratic Taylor lower bound has minimum $37639/640000$. This proves the two assertions without an extremum assumption for $\Phi$. If $C_0\ge0$, the preceding bound is at least $p$. If $C_0<0$, $C_0>-25p$ gives \[ \Phi>p[1-25G(x)]>\frac{37639}{640000000}>0. \] This proves the remaining moment range. \end{proof} Lemmas~\ref{lem:near-low-rho}, \ref{lem:near-intermediate-rho} and~\ref{lem:near-high-rho} cover $0<\rho<1$. Lemma~\ref{lem:near-K-affinity}, the positivity of its denominator, and Lemma~\ref{lem:near-four-age} therefore prove $D_\lambda>0$ whenever $u\le1/2$ in~\eqref{eq:near-domain}. \subsubsection{The first two terms for \texorpdfstring{$1/2u^2H_2, \quad H_2=\frac{(1+w)(1-\lambda z)^2-2(1-a)(1-\lambda z)}\lambda +2\lambda z^2[B_2+2L_2M_2], \end{equation} where \[ M_2=a(1-r_2)+\frac wc(\lambda+1-\lambda r_2-1/z). \] All discarded terms, those with $s\ge3$, are positive. The term with $s=1$ and the value $a=a(w)$ are the same as in the full series. For the polynomial calculations below we use $k$, previously $c/(\lambda\log\lambda)$, for the coordinate $1-\lambda u$. Thus \[ A=\lambda+k,\qquad B=\lambda^2+(\lambda+1)k, \qquad u=(1-k)/\lambda. \] Put \begin{equation}\label{eq:near-high-denominator} E=\frac{K_{\rm cap}-1+u}{1-u},\qquad D_\rho=2-(\lambda+1)u+\frac{2\lambda u}{A}\rho +\left[\frac{2u}{A(1-u)}-(\lambda+1)uE\right]\rho^2. \end{equation} Then $w=cu(1+E\rho^2)/D_\rho$. The denominator $D_\rho$ is positive on $uA\le\rho\le1$. More precisely, with \[ E_+=\frac{2/(1+\lambda u)-(1-u)}{1-u}, \] one has \begin{equation}\label{eq:near-high-denominator-gap} D_\rho-(1-u)(1+E\rho^2) =(1-\rho)\left[k(1+\rho)+\frac{2\lambda u\rho}{A}\right] +(1+\lambda u)(E_+-E)\rho^2>0. \end{equation} Thus $01/2$ and $\alpha<1$. Differentiating $w$ with respect to $\rho$ gives \[ w_\rho=\frac{cu}{D_\rho^2} \left[-\frac{2\lambda u}{A}+4\left(E-\frac{u}{A(1-u)}\right)\rho +\frac{2\lambda uE}{A}\rho^2\right]>0. \] Indeed $E\ge u/(1-u)$ and $\rho\ge uA>1/2$ give the positive lower bracket $(2\lambda u/A)[-1+2\rho+u\rho^2/(1-u)]$. Writing $J=1/u-(\lambda+1)+\lambda z$, we have $a=(1-wJ/c)/z$. Here $J\ge(1-u)k/u>0$ and \[ a+\lambda w/c=\frac{1+(w/c)(\lambda+1-1/u)}z>0. \] Consequently \[ a_\rho=-\frac{a+\lambda w/c+AJw_\rho/c}{\rho}<0. \] At $\rho=1$ a direct substitution gives \[ a(1)-1=\frac{(1-u)[2\lambda-(\lambda+1)K_{\rm cap}]} {2-(\lambda+1)uK_{\rm cap}}. \] Since $a<\alpha<1$ and $a$ decreases with $\rho$, we must have $K_{\rm cap}>2\lambda/(\lambda+1)$. On the other hand $K_{\rm cap}$ decreases with $u>1/2$, and the logarithmic bounds in~\eqref{eq:near-cap} give \[ K_{\rm cap}<\frac{25(\lambda+1)}{48\lambda}+\frac14. \] For $\lambda\ge11/8$ this is smaller than $2\lambda/(\lambda+1)$, since \[ 59\lambda^2-62\lambda-25 =59(\lambda-11/8)^2+\frac{401}{4}(\lambda-11/8)+\frac{83}{64}>0. \] Thus $u>1/2$ and $\alpha<1$ imply $\lambda<11/8$. It remains to prove $H_2>0$ for $2/\sqrt3\le\lambda\le11/8$. Collecting the terms in~\eqref{eq:near-H2} according to their dependence on $w$ gives \[ H_2=H_0(z)+wH_v(z), \] where \begin{equation}\label{eq:near-H0-Hv} \begin{split} H_0(z)&=\frac2{\lambda z}-2-\frac1\lambda +z^2\left[\lambda+\frac{2k^2}{B^2} +\frac{4\lambda^2(1-u)k}{AB}\right],\\ H_v(z)&=\frac{(1-\lambda z)^2}{\lambda} -\frac{2(1-\lambda z)J}{\lambda cz}+\frac{2z^2k^2}{B^2} +\frac{4kz^2}{Ac}(\lambda+1-\lambda A/B-1/z). \end{split} \end{equation} In particular $N(\rho)=\rho D_\rho H_2(\rho)$ is a polynomial of degree five. Its coefficients $N=\sum_{j=0}^5n_j\rho^j$ are \begin{equation}\label{eq:near-N-coefficients} \begin{gathered} P=\lambda+2k^2/B^2+4\lambda^2(1-u)k/(AB),\\ C_2=\lambda+2k^2/B^2+4k(\lambda+1-\lambda A/B)/(Ac),\\ W=\lambda-2k/B+2k^2/B^2,\qquad S_0=[2-(\lambda+1)u]P+cuC_2,\\ n_0=2A/\lambda,\qquad n_1=4u-2-2/\lambda,\\ n_2=-2AE/\lambda+4u/[\lambda(1-u)]-4uB/A^2,\\ n_3=2E+S_0/A^2-2u(2\lambda+1)/[\lambda A(1-u)],\\ n_4=(2u/A^3)[\lambda P+EA(A-2k)],\qquad n_5=(2u/A^2)[P/(A(1-u))-EW]. \end{gathered} \end{equation} These are obtained by multiplying the quadratic $D_\rho$ by $\rho H_0$ and the quadratic $cu(1+E\rho^2)$ by $\rho H_v$; thus the displayed six coefficients include all terms. \begin{lemma}[Positivity of $N'''(\rho)$]\label{lem:near-high-curvature} For $2/\sqrt3\le\lambda\le11/8$ and~\eqref{eq:near-high-u-domain}, $N'''(\rho)>0$ on $uA\le\rho\le1$. \end{lemma} \begin{proof} First $n_4>0$, since $A-2k=c+\lambda u>0$. Also $00. \] Indeed $u\le3/4$, $1/4\le1-u\le1/2$, $A\le27/16$, $B\le337/128$, $G>3/2$, and $\lambda^2\ge4/3$. The first term in braces is at least $(384/337)(113/81)>385/256$, whereas the second is at most $385/256$. Since $E0$. Using $P,C_2\ge\lambda$, $S_0\ge2\lambda(1-u)$, and $\rho\ge uA$ in~\eqref{eq:near-N-coefficients}, we obtain \begin{equation}\label{eq:near-third-derivative-bound} N'''>\frac{12u}{1-u}\frac{R_0}{\lambda A} +\frac{12\lambda(1-u+4\lambda u^2)}{A^2}, \end{equation} where \[ R_0=\lambda^2-\lambda-1+\lambda^2(4u^3-u)+4\lambda cu^2. \] In obtaining this bound, replace $E$ by $u/(1-u)$ before combining it with the negative term: its coefficient is $12[1+4u^2(A-2k)/A]>0$. The sign of $R_0$ is not used in this replacement. The function $R_0$ increases with both variables, since \[ (R_0)_u=\lambda^2(12u^2-1)+8\lambda cu>0, \] \[ (R_0)_\lambda=2\lambda(1-u+4u^2+4u^3)-(1+4u^2) \ge1-2u+4u^2+8u^3\ge2. \] On $1/2\le u\le3/5$, $R_0\ge(5-4\sqrt3)/3>-2/3$, $u/(1-u)\le3/2$ and $\lambda A>5/3$. The first term of~\eqref{eq:near-third-derivative-bound} is therefore greater than $-36/5$; the second is at least $12(23/20)(23/20+2/5)/(27/16)^2>36/5$. On $3/5\le u\le2/3$, use $R_0\ge-5749/25000>-1/4$, $u/(1-u)\le2$ and $\lambda A>3/2$. The two terms are respectively greater than $-4$ and $4$. Finally, on $u\ge2/3$, $R_0\ge(275-150\sqrt3)/81>0$. This proves the result in every case. \end{proof} The next lemma gives the signs needed to bound $N'(1)$ and $N(1)$. It verifies inequalities on a rectangle in $\lambda,k$, with $u=(1-k)/\lambda$, which also includes parameter values outside~\eqref{eq:near-high-u-domain}. \begin{lemma}[Polynomial inequalities for $N'(1)$ and $N(1)$]\label{lem:near-high-arithmetic} Let $23/20\le\lambda\le11/8$, $0\le k\le1/2$, $u=(1-k)/\lambda$, $A=\lambda+k$, $B=\lambda^2+(\lambda+1)k$, and let $n_j(E)$ be exactly~\eqref{eq:near-N-coefficients}. Define \[ Y(E)=\sum_{j=1}^5j n_j(E),\qquad D=\lambda A^4(c+k)B^2>0, \] \[ P_-=-\tfrac12D\,Y\left(\frac u{1-u}\right),\qquad P_+=-\tfrac12(2-k)D\,Y\left(\frac{2/(2-k)-(1-u)}{1-u}\right). \] These are polynomials in $(\lambda,k)$ of bidegrees $(10,7)$ and $(10,8)$, respectively, and both are strictly positive. Also define \begin{equation}\label{eq:near-upper-S} \begin{split} S(K)={}&k(1-uK)(1+2\lambda/B^2)-A(1-u)[(\lambda+1)K-2\lambda]\\ &+\frac{2\lambda^2(1-u)}{AB}[2\lambda-(\lambda+1)uKA]. \end{split} \end{equation} The polynomial $24\lambda^2AB^2S(K_b)$ has bidegree $(10,7)$ and is strictly positive on the same rectangle. The exact finite obligations are positivity of all Bernstein coefficients of $P_-$ and $24\lambda^2AB^2S(K_b)$ on that rectangle, and positivity of all Bernstein coefficients of $P_+$ on the three rectangles \[ [23/20,101/80]\times[0,1/4],\quad [23/20,101/80]\times[1/4,1/2],\quad [101/80,11/8]\times[0,1/2]. \] The coefficients are generated by~\eqref{eq:near-Bernstein-obligation} with the corresponding affine substitution in each coordinate. They and their exact verification are in the supplement. Every clearing factor here is positive, including $2-k$, $c+k$, and $1-u=(c+k)/\lambda$. \end{lemma} Because $Y$ is affine in $E$, Lemma~\ref{lem:near-high-arithmetic} and $u/(1-u)0, \] \[ H_h(u)=\frac k\lambda\left\{(1+h)k(1+2\lambda u^2/B^2)+2c +\frac{4\lambda u^2[\lambda c-(\lambda^2+1)k]}{AB}\right\}. \] We verify the latter sign. The positive function \[ R(\lambda,k)=\frac{\lambda^2+1}{\lambda(\lambda+k)[\lambda^2+(\lambda+1)k]} \] decreases with $\lambda$: its logarithmic derivative is \[ \frac{2\lambda}{\lambda^2+1}-\frac1\lambda-\frac1{\lambda+k} -\frac{2\lambda+k}{\lambda^2+(\lambda+1)k}<0, \] since $\lambda k<1$. Also $h=c(1-k)/(c+k)$ increases with $c$. Dropping two positive terms gives \[ \frac{\lambda H_h(u)}k> 2c+(1+h)k-4k(1-k)^2R(\lambda,k) \] \[ \ge\frac3{10}+\left[1+\frac{(3/20)(1-k)}{3/20+k}\right]k -4k(1-k)^2R(23/20,k). \] After multiplication by the positive denominator $(23/20)(23/20+k)[(23/20)^2+(43/20)k](3/20+k)$, the last expression equals \[ Q(k)=\frac{2518569}{32000000}-\frac{29541}{200000}k -\frac{7316903}{3200000}k^2+\frac{1790369}{80000}k^3-\frac{57507}{8000}k^4. \] For $0\le k\le1/2$, coefficient comparison and $k^4\le k^3/2$ give \[ Q(k)>\frac3{40}-\frac3{20}k-\frac52k^2+18k^3 =18(k+1/9)(k-1/8)^2+\frac{11}{160}k+\frac7{160}>0. \] Hence $H_h(u)>0$. Equation~\eqref{eq:near-high-denominator-gap} gives $00$. At the upper endpoint, direct substitution in~\eqref{eq:near-H2} gives \[ N(1)=\frac{2k}{\lambda A^2(1-u)}S(K_{\rm cap}), \] with $S$ from~\eqref{eq:near-upper-S}. Its coefficient of $K$ is negative. More explicitly, if $\delta=(\lambda+1)K-2\lambda$, then $S=P-\delta Q$ with \[ P=k\left[\frac{c+2k}{\lambda+1}(1+2\lambda/B^2) +\frac{4\lambda^3(1-u)^2}{AB}\right]>0, \] \[ Q=A(1-u)+\frac{uk}{\lambda+1}(1+2\lambda/B^2) +\frac{2\lambda^2u(1-u)}B>0. \] The specific logarithmic bound $K_{\rm cap}S(K_b)>0$, the latter sign being Lemma~\ref{lem:near-high-arithmetic}. Thus $N(1)>0$. Equation~\eqref{eq:near-high-endpoint-reduction} proves $H_2>0$ and, by~\eqref{eq:near-H2}, $D_\lambda>0$ for $1/21/2$ cover $0<\alpha<1$ under~\eqref{eq:near-domain}. Lemma~\ref{lem:calibration} covers $\alpha=1$. Thus $D_\lambda>0$ throughout~\eqref{eq:near-domain}, and Lemma~\ref{lem:full-response} gives $v'(\lambda)<0$ there.