\section{Radial channels: bypass, local synthesis, and gain limits} \label{cn:section} The large multiplicity $2\ell+1$ of a spherical partial wave and the unbounded radial Bragg amplification in Proposition~\ref{cx:bragg} suggest a concrete possible counterexample. This section tests that proposal directly at the physical section. The resulting obstruction does not assume a field-energy bound or a bound on the exterior support. It applies to a single angular order in a homogeneous core, including arbitrary coherent combinations of all azimuthal indices and both electric and magnetic multipoles. Mixed angular orders and a structured core remain outside its scope. The distinction between field amplitude and transported power is central to this test. A high-order spherical mode can have many angular patterns and a large field in the core, but neither fact says that all of those patterns carry power through $D$ without a comparable return outside it. We first isolate that return using symmetry, then examine a single spherical order. The later synthesis and gain estimates show how far one can go by adjusting the radial medium; they do not by themselves supply a counted channel space. \subsection{Inversion parity and the threshold regime} \begin{proposition}[An odd plane flux must return outside the section] \label{cn:odd-return} Let $s\in L^1(\mathbb R^2)$ be a real normal power density, and suppose $s(2c-x)=-s(x)$ almost everywhere for some point $c$ in the section plane. For any measurable section $D$, \begin{equation}\label{cn:odd-return-bound} \int_{\mathbb R^2\setminus D}|s|\,dA \ \geq\ \left|\int_Ds\,dA\right|. \end{equation} Consequently a nonzero counted input with this symmetry requires $\delta\geq\eta\tau$. This conclusion does not require that $D$ be centered at $c$. \end{proposition} \begin{proof} The intersection $D\cap(2c-D)$ is invariant under $x\mapsto2c-x$, so its signed integral is zero. The reflection of $D\setminus(2c-D)$ is contained in the complement of $D$, and has the opposite signed integral. Taking its absolute integral proves \eqref{cn:odd-return-bound}. The counted-input inequalities then give $\delta\|a\|^2\geq P_D(a)\geq\eta\tau\|a\|^2$. \end{proof} \begin{remark}[Application to a Maxwell parity sector] \label{cn:parity} In a medium invariant under inversion about $c$, a field satisfying $E(2c-x)=\sigma E(x)$, $\sigma\in\{1,-1\}$, satisfies $H(2c-x)=-\sigma H(x)$ by the first curl equation. Its real Poynting vector is therefore odd. If $c$ lies in the section plane and the required traces are integrable, Proposition~\ref{cn:odd-return} applies. In particular, a coherent sum of electric-type multipoles of one angular order has one parity, as does a coherent sum of magnetic-type multipoles of that order. Merely using their azimuthal degeneracy does not give passage in the regime $\delta<\eta\tau$. If the square is centered at $c$, its signed passage is exactly zero. Mixing the two parities is necessary for this centered construction; it is not sufficient, as the next theorem shows. \end{remark} \subsection{Both polarizations of a regular spherical partial wave} Fix a homogeneous ball centered at the center of $D$, with real permittivity $n^2\eps_0$ and permeability $\mu_0$. Write $k=nk_0$ and $Z=\sqrt{\mu_0/(n^2\eps_0)}$. Let $\nu=\ell(\ell+1)$, $\ell\geq1$, and let $Y_{\ell m}$ be any orthonormal complex spherical-harmonic basis. For coefficient vectors $A,B\in\mathbb C^{2\ell+1}$ define \begin{equation}\label{cn:angular-fields} Y_A=\sum_m A_mY_{\ell m},\quad \Psi_A=\nu^{-1/2}\nabla_{\mathbb S^2}Y_A,\quad X_A=\widehat r\times\Psi_A, \end{equation} and similarly for $B$. With $j=j_\ell(kr)$, put \begin{equation}\label{cn:radial-fields} u=rj,\qquad a_\ell(r)=\frac{\sqrt\nu\,j}{kr},\qquad b_\ell(r)=\frac{u'}{kr}. \end{equation} The notation $a_\ell,b_\ell$ in this section denotes radial functions, not external incoming amplitudes. The general regular field at this one angular order can be written \begin{align} E&=jX_A-a_\ell Y_B\widehat r-b_\ell\Psi_B,\notag\\ H&=-\frac{i}{Z} (-a_\ell Y_A\widehat r-b_\ell\Psi_A+jX_B). \label{cn:multipoles} \end{align} The usual two regular spherical Maxwell waves give this representation: if $M_A=jX_A$ and $N_A=k^{-1}\nabla\times M_A$, direct spherical differentiation gives $N_A=-a_\ell Y_A\widehat r-b_\ell\Psi_A$, $\nabla\times N_A=kM_A$, and hence both Maxwell curl equations hold in \eqref{cn:multipoles}. Regularity excludes the singular spherical Bessel solution. All possible complex coefficients, including those generated by a distant resonant dielectric, are allowed. \begin{lemma}[The plane flux separates into opposite parities] \label{cn:flux-split} On the equatorial plane $z=0$, a field \eqref{cn:multipoles} has \begin{equation}\label{cn:flux-decomposition} S_z(r,\varphi)=F_\ell(r)O(\varphi)+G_\ell(r)C(\varphi), \end{equation} where the real angular functions satisfy $O(\varphi+\pi)=-O(\varphi)$ and $C(\varphi+\pi)=C(\varphi)$, and \begin{equation}\label{cn:radial-flux-factors} F_\ell=\frac{a_\ell j}{2Z} =\frac{\sqrt\nu\,u^2}{2Zkr^3},\qquad G_\ell=\frac{a_\ell b_\ell}{2Z} =\frac{\sqrt\nu\,u u'}{2Zk^2r^3}. \end{equation} Explicitly, with all angular expressions evaluated at $\theta=\pi/2$, \begin{align} O&=-\operatorname{Im} (\overline{Y_A}\Psi_{A,\theta} +\overline{Y_B}\Psi_{B,\theta}),\notag\\ C&=\operatorname{Im} (Y_B\overline{X_{A,\theta}} -\overline{Y_A}X_{B,\theta}). \label{cn:angular-flux-factors} \end{align} \end{lemma} \begin{proof} At the equator $\widehat z=-\widehat\theta$. In $E\times\overline H$, the tangential terms arise only by crossing one radial and one tangential component. Substituting \eqref{cn:multipoles}, and using $X\times\widehat r=\Psi$, $\Psi\times\widehat r=-X$, $\widehat r\times X=-\Psi$, gives \eqref{cn:flux-decomposition}--\eqref{cn:angular-flux-factors}. For completeness, the tangential vector inside the factor $i/Z$ is \begin{equation}\label{cn:cross-product-check} a_\ell j(Y_B\overline{\Psi_B}-\overline{Y_A}\Psi_A) +a_\ell b_\ell (Y_B\overline{X_A}-\overline{Y_A}X_B). \end{equation} Taking one half its real part after multiplication by $i/Z$, then changing from the $\theta$ component to $z$, gives the stated signs. The spherical-harmonic inversion identity gives $Y_A(\pi/2,\varphi+\pi)=(-1)^\ell Y_A(\pi/2,\varphi)$, whereas its $\theta$ derivative gains the opposite sign. Since $X_\theta=-\nu^{-1/2}\partial_\varphi Y$ at the equator, $X_\theta$ gains the same sign as $Y$. These facts prove the two angular parity assertions for arbitrary coherent coefficients. \end{proof} \begin{lemma}[Radial growth below the angular turning scale] \label{cn:radial-growth} Let $R_2>0$ and assume $\nu>k^2R_2^2$. Define \begin{equation}\label{cn:exponent} \gamma=\frac{1+\sqrt{1+4(\nu-k^2R_2^2)}}2>1. \end{equation} Then the regular real solution $u=rj_\ell(kr)$ is positive with $u'>0$ for $00$. The coefficient of $u$ is positive throughout this interval, so a first zero of $u$ or $u'$ is impossible. For $w=ru'/u$ the equation is \begin{equation}\label{cn:riccati} rw'=\nu-k^2r^2+w-w^2. \end{equation} At $w=\gamma$ its right side is $k^2(R_2^2-r^2)\geq0$, and at $w=\ell+1$ it is $-k^2r^2<0$. The initial expansion has $w=\ell+1-k^2r^2/(2\ell+3)+O(r^4)$, so $w$ begins in the interval $[\gamma,\ell+1]$ and cannot exit it. A first crossing at the lower endpoint can equivalently be excluded by the scalar comparison equation, including the equality case at $R_2$. Integrating $u'/u\geq\gamma/r$ proves the ratio bound. \end{proof} \begin{theorem}[High angular degeneracy cannot evade absolute bypass in a homogeneous core] \label{cn:radial-bypass} Let $R=L/\sqrt2$ be the circumradius of the fixed square $D$, fix $1k^2t^2R^2$, let $\gamma$ be \eqref{cn:exponent} with $R_2=tR$, and set \begin{equation}\label{cn:bypass-constant} B_\ell(s,t)=\frac{\gamma}{\ell+1} \bigl(t^{2\gamma-2}-s^{2\gamma-2}\bigr). \end{equation} Then its physical flux obeys the local, amplitude-independent estimate \begin{equation}\label{cn:bypass-bound} \int_{sR0$ on $(0,tR]$. The square is centrally symmetric, so the integral of the odd term $F_\ell O$ in \eqref{cn:flux-decomposition} over $D$ is zero. Writing the square in polar coordinates and using its circumradius, \begin{equation}\label{cn:inside-angular-bound} |P_D|\leq \left(\int_0^R G_\ell(r)r\,dr\right) \left(\int_0^{2\pi}|C(\varphi)|\,d\varphi\right). \end{equation} On the full annulus, opposite angles have fluxes $FO+GC$ and $-FO+GC$. The triangle inequality $|x+y|+|-x+y|\geq2|y|$ gives \begin{equation}\label{cn:outside-angular-bound} \int_{sR0$. There is a finite angular-order cutoff beyond which no nonzero counted input can have a pure $\ell$ core field. This remains true if arbitrary passive dielectric layers outside the core make either or both polarization amplitudes arbitrarily large. In particular increasing $2\ell+1$ at a single resonance does not by itself yield an unbounded counted subspace through this fixed section. \end{corollary} \begin{proof} The constant $B_\ell(s,t)$ tends to infinity, whereas $\delta/(\eta\tau)$ is fixed. All exterior amplification is already included in the arbitrary coefficients $A,B$ and cancels in the ratio. \end{proof} \begin{remark}[Scope and the remaining constructive alternatives] \label{cn:scope} This is a genuine Maxwell exclusion for the stated spherical route, not a uniform theorem for Conjecture~\ref{pr:channel-target}. A coherent mixture of different angular orders has additional interference terms with different radial factors; the proof does not bound them. Nor does it cover a dielectric structured throughout the core, a spherical center moving away with angular order, or a plane that does not pass through that center. Such changes are allowed by the unrestricted target and require a new argument. The parameter regime $\delta\geq\eta\tau$ permits the return-flow mechanism in Proposition~\ref{cn:odd-return}; it does not prove that an arbitrarily large physical counted subspace exists. A large bypass tolerance, a choice of the full lossless scattering map as $T$, or a single circulating high-$Q$ field is not by itself a quantifier counterexample. The throughput threshold still requires independent physical inputs, and passage must hold for every coherent vector. No admissible all-device counterexample for either threshold regime is inferred here. \end{remark} \subsection{An actual local inverse-synthesis step for mixed angular orders} The preceding obstruction does not make a mixed-order proposal impossible. Conversely, saying that radial layers can independently tune every partial wave is an extra claim. The following local version of that claim can be proved for finite ordinary dielectrics, including control of the interior amplitude rather than only the scattering phase. \begin{theorem}[Local independent control of finitely many complex core amplitudes] \label{cn:local-synthesis} Fix $k_0>0$, $\varepsilon_+>1$, a radius $R_0>0$, an annulus $0R_0 \end{cases} \end{equation} has the following property for all sufficiently small real $\theta\in\mathbb R^{2m}$. For each $l=l_i$, fix the leading coefficient at the origin of the regular real TE radial solution $u_l$ independently of $\theta$. In the vacuum exterior write \begin{equation}\label{cn:incoming-coefficient} u_l(r;\theta)=A_l(\theta)\,\xi_l^-(k_0r) +\overline{A_l(\theta)}\,\xi_l^+(k_0r), \qquad \xi_l^\pm(x)=x h_l^{(2/1)}(x). \end{equation} Here $\xi_l^-$ uses $h_l^{(2)}$ and $\xi_l^+$ uses $h_l^{(1)}$. Each $A_l$ is nonzero. The map \begin{equation}\label{cn:log-controls} \theta\longmapsto \left(\operatorname{Re}\log\frac{A_{l_i}(\theta)}{A_{l_i}(0)}, \operatorname{Im}\log\frac{A_{l_i}(\theta)}{A_{l_i}(0)} \right)_{i=1}^m \end{equation} is a real-analytic local diffeomorphism at zero. Thus every sufficiently small prescribed change of all $m$ core gains and all $m$ scattering phases is realized by a physical radial permittivity in the same fixed interval $[1,\varepsilon_+]$. Here is one explicit, retained neighborhood certificate. Denote the map in \eqref{cn:log-controls} by $\mathcal F$, let $\sigma$ be the smallest singular value of $J=D\mathcal F(0)$, and choose a parameter ball of radius $\rho$ on which \begin{equation}\label{cn:control-radius} \rho\left(\sum_j\|h_j\|_\infty^2\right)^{1/2} <\min(n_b^2-1,\varepsilon_+-n_b^2),\qquad \sup_{|\theta|\leq\rho}\|D\mathcal F(\theta)-J\| \leq\sigma/2. \end{equation} The logarithms are taken on their continuous branches near zero. Then every target vector of Euclidean norm less than $\sigma\rho/2$ is attained in this parameter ball. No lower bound on $\sigma\rho$ uniform in $m$ is asserted. \end{theorem} \begin{proof} The TE equation, its interface conditions, and its regular normalization are \begin{equation}\label{cn:controlled-ode} u_l''+\left(k_0^2\varepsilon_\theta(r) -\frac{l(l+1)}{r^2}\right)u_l=0, \qquad u_l,u_l'\text{ continuous}. \end{equation} Since every control vanishes near zero, the regular initial data can be set at any fixed radius below $r_-$. Standard successive integration of this first-order system on a finite interval gives real-analytic dependence on the finite parameters. The two vacuum Hankel solutions have Wronskian $2ik_0$, with the convention $W(f,g)=fg'-f'g$. The real nonzero regular solution has conjugate exterior coefficients, so $A_l=0$ would force the solution to vanish outside and then everywhere. This proves its nonvanishing and the local analytic logarithms. Let $z_l(r)$ be the outgoing radial solution, equal to $\xi_l^+(k_0r)$ outside $R_0$, continued inward through the base material at $\theta=0$. A real variation $h(r)$ changes the regular solution by $\dot u_l$ satisfying \begin{equation}\label{cn:variation-ode} \dot u_l''+\left(k_0^2\varepsilon_0-l(l+1)/r^2\right)\dot u_l =-k_0^2h u_l, \qquad \dot u_l=\dot u_l'=0\quad(r0$. The inverse-function theorem proves the qualitative local assertion. For the retained certificate, write $\mathcal F(\theta)=J\theta+\mathcal R(\theta)$. The second condition in \eqref{cn:control-radius} makes $\mathcal R$ Lipschitz with constant $\sigma/2$ on the parameter ball and $\mathcal R(0)=0$. For a target $y$ with $|y|<\sigma\rho/2$, the map $\theta\mapsto J^{-1}(y-\mathcal R(\theta))$ is a contraction of that ball into itself. Its fixed point realizes $y$. The first condition in \eqref{cn:control-radius} keeps the permittivity strictly within its prescribed bounds. Finally, for the angular normalization $\int_{S^2}|X_{lm}|^2\,dS=1$, an incoming TE radial amplitude $A$ carries physical power $|A|^2/(2Z_0)$ in the vacuum exterior. Thus the field produced by unit incoming power has regular-core amplitude $\sqrt{2Z_0}/A_l$ in the normalization above, while its outgoing-to-incoming coefficient is $\overline{A_l}/A_l$. The real and imaginary logarithms controlled in \eqref{cn:log-controls} are precisely the logarithmic core gain with a minus sign, and minus one half the scattering phase. The vacuum power convention does not change with the controls. \end{proof} \begin{remark}[The local synthesis range is not an arbitrary matching network] \label{cn:local-synthesis-scope} Theorem~\ref{cn:local-synthesis} supplies actual radial Maxwell material controls for a finite set of angular orders. It neither makes $A_l$ arbitrarily small simultaneously nor estimates the absolute flux on the whole section plane. The certified target neighborhood may shrink with the number and size of the angular orders. Iterating local invertibility without a continuation estimate would not prove global reachability within the fixed permittivity bounds. Independent tuning of finitely many scattering phases alone would be still weaker, because it would not control the gain represented by $|A_l|^{-1}$. \end{remark} \begin{corollary}[Local control of both polarizations at separated orders] \label{cn:both-synthesis} If the selected positive angular orders additionally satisfy $|l_i-l_j|\geq3$ for $i\ne j$, there exist $4m$ smooth real radial controls in the same prescribed annulus that independently vary, locally, the complex regular-core incoming coefficients of both TE and TM polarizations at all selected orders. All material and physical normalization conclusions of Theorem~\ref{cn:local-synthesis} remain valid, with a $4m$-dimensional Jacobian and its own retained neighborhood constants. \end{corollary} \begin{proof} For the TM polarization choose the magnetic radial amplitude $u$. Its real equation and transmission conditions are \begin{equation}\label{cn:tm-ode} (\varepsilon^{-1}u')' +\left(k_0^2-\frac{l(l+1)}{\varepsilon r^2}\right)u=0, \qquad u,\ \varepsilon^{-1}u'\text{ continuous}. \end{equation} It has the same regular/irregular basis inside a constant-index core and the same vacuum radial Hankel basis outside. If $z_l$ is its continued outgoing solution, differentiating the weighted Wronskian $\varepsilon^{-1}W(\dot u_l,z_l)$ and integrating by parts gives \begin{equation}\label{cn:tm-amplitude-derivative} D\log A_l^{\rm TM}(0)[h] =\frac{1}{2ik_0 A_l^{\rm TM}(0)n_b^4} \int h(r)\left(u_l'z_l' +\frac{l(l+1)}{r^2}u_lz_l\right)\,dr. \end{equation} Indeed $\dot p=-h/n_b^4$ for $p=\varepsilon^{-1}$, and the integrated weighted-Wronskian derivative is $-\int\dot p(u_l'z_l'+l(l+1)u_lz_l/r^2)$. The endpoint terms vanish since the control has compact support inside the annulus. The exterior weighted Wronskian is again $2ik_0\dot A_l^{\rm TM}$. Inside the sphere write $u_l=c_l\psi_l(\kappa r)$ and $z_l=\alpha_l\psi_l(\kappa r)+\beta_l\zeta_l(\kappa r)$. The weighted outgoing Wronskian gives $\operatorname{Im}(\alpha_l\overline{\beta_l})=-n_b\ne0$. Consequently the real and imaginary derivative kernels are an invertible real change of basis from \begin{equation}\label{cn:tm-kernels} (\psi_l')^2+\frac{l(l+1)}{x^2}\psi_l^2, \qquad \psi_l'\zeta_l'+\frac{l(l+1)}{x^2}\psi_l\zeta_l, \quad x=\kappa r. \end{equation} The common factor $\kappa^2$ is immaterial for independence. The elementary recurrences $\psi_{l-1}=\psi_l'+l\psi_l/x$ and $\psi_{l+1}=(l+1)\psi_l/x-\psi_l'$, and the identical recurrences for $\zeta_l$, imply \begin{align} (\psi_l')^2+\frac{l(l+1)}{x^2}\psi_l^2 &=\frac{l+1}{2l+1}\psi_{l-1}^2 +\frac{l}{2l+1}\psi_{l+1}^2,\notag\\ \psi_l'\zeta_l'+\frac{l(l+1)}{x^2}\psi_l\zeta_l &=\frac{l+1}{2l+1}\psi_{l-1}\zeta_{l-1} +\frac{l}{2l+1}\psi_{l+1}\zeta_{l+1}. \label{cn:tm-adjacent-identities} \end{align} The independence proof for the product kernels in the preceding theorem also includes order zero. The triples $\{l_i-1,l_i,l_i+1\}$ are disjoint by hypothesis. The TE pair uses the middle index of each triple; the TM pair is a nonzero combination of its two unused neighboring indices, separately in the even and odd kernel families. All $4m$ derivative functions are therefore independent. The same evaluation-point, smooth-bump, and inverse-function argument now proves the corollary. For the magnetic normalization of TM waves, incoming power is $Z_0|A|^2/2$, another fixed vacuum factor independent of the controls. \end{proof} \begin{proposition}[Fixed-annulus controls have a finite gain ceiling] \label{cn:fixed-control-ceiling} Keep $r_->0$, $R_0$, the interior material on $r0. \end{equation} \end{proposition} \begin{proof} The states are continuous across every radial interface. Away from interfaces their first-order systems are respectively \begin{equation}\label{cn:state-generators} v'=k_0\begin{pmatrix}0&1\\ l(l+1)/(k_0^2r^2)-\varepsilon&0\end{pmatrix}v, \qquad v'=k_0\begin{pmatrix}0&\varepsilon\\ l(l+1)/(k_0^2\varepsilon r^2)-1&0\end{pmatrix}v. \end{equation} For a real matrix with off-diagonal entries $a,b$ and zero diagonal, the Euclidean operator norm is $\max(|a|,|b|)$. Both generators are therefore bounded by $C_l/(R_0-r_-)$ on the interval. Gronwall applied also to reverse propagation gives $|v(R_0)|\geq e^{-C_l}|v(r_-)|$. The initial state is fixed and nonzero by uniqueness of the regular radial solution. On the vacuum side of the final interface, \eqref{cn:incoming-coefficient} gives $v(R_0)=2V_l(\operatorname{Re}A_l,\operatorname{Im}A_l)^T$. The Hankel Wronskian makes $V_l$ invertible, and its norm is finite. This proves \eqref{cn:amplitude-ceiling}. \end{proof} \begin{remark}[Focused global continuation attempt and its exact limits] \label{cn:global-attempt} The local controls can leave every selected scattering phase fixed while changing the selected core gains: prescribe zero imaginary coordinates in \eqref{cn:log-controls}. Thus phase matching by itself does not determine the internal response. But the controls cannot be iterated inside their fixed bounded annulus to obtain arbitrarily large gains; Proposition~\ref{cn:fixed-control-ceiling} gives an explicit obstruction to that proposed iteration, even with arbitrarily many increasingly thin layers. Allowing the controlled radius to increase removes this particular ceiling, as the frozen unrestricted target permits. It also adds an uncontrolled portion of the physical section plane. Endpoint radial transfer data $A_l$ do not estimate its absolute normal flux. The single-order estimate in Theorem~\ref{cn:radial-bypass} gives a concrete case in which the exterior plane cost grows rather than disappears under amplification. Theorem~\ref{cc:global-control} below supplies approximate global reachability for any finite set of TE orders with increasing support, including independent core gains up to a common phase. It does not supply simultaneous whole-plane bypass control. Corollary~\ref{cb:finitecore} further bounds the counted dimension in a sufficiently large vacuum core for every fixed finite angular band, regardless of gain. The unrestricted finite-channel statement is still unresolved by this route. \end{remark}