\section{Growing radial support: simultaneous gain and its flux cancellation} \label{cg:section} This section carries a growing-support construction through the actual radial Maxwell equations. It proves simultaneous large gains for any prescribed finite band of partial waves. It also proves why this particular construction does not give many power channels: the gains of opposite electric inversion parities are reciprocal, whereas signed flux through a centered section couples those opposite parities. The resulting channel bound is independent of the mirror length and radius within the prescribed band. It is not a bound for arbitrary dielectric geometry or for partial waves outside that band. The useful comparison is between two effects of the same layers: they can amplify certain field components, but they amplify their partners by the reciprocal factor. The flux couples those partners, so large internal fields need not yield a correspondingly large passage operator. We first bound the vacuum reference flux, construct the layers, and then evaluate their effect on the flux form in the chosen finite band. \subsection{A vacuum flux trace estimate with physical normalization} \begin{lemma}[Trace-class vacuum section flux] \label{cg:vacuum-trace} Let $\mathcal H=L^2_t(S^2)$ be the power-normalized incoming vacuum space. For the regular vacuum Maxwell fields generated by \eqref{hs:three-illumination}, let $Q_D^0$ be the Hermitian signed flux operator through a bounded planar section of area $A$. Then \begin{equation}\label{cg:trace-bound} \|Q_D^0\|_1\leq\frac{2k_0^2A}{\pi}. \end{equation} If $D$ is centrally symmetric about the origin, the positive part has \begin{equation}\label{cg:positive-trace} \operatorname{Tr}(Q_D^0)_+\leq\frac{k_0^2A}{\pi}. \end{equation} Consequently a subspace $V$ with $\langle b,Q_D^0b\rangle\geq t\|b\|^2$ for all $b\in V$, $t>0$, has dimension at most $k_0^2A/(\pi t)$ in the symmetric case. \end{lemma} \begin{proof} Denote the two restriction maps to $L^2(D;\mathbb C^3)$ by $\mathsf E$ and $\mathsf H$. The physical normalization is \begin{align} \mathsf E g(x)&=\frac{k_0\sqrt{Z_0}}{\sqrt2\,\pi} \int_{S^2}g(d)e^{ik_0d\cdot x}\,dS(d),\notag\\ \mathsf H g(x)&=\frac{k_0}{\sqrt{2Z_0}\,\pi} \int_{S^2}d\times g(d)e^{ik_0d\cdot x}\,dS(d). \label{cg:herglotz} \end{align} The incoming far-field amplitude is $\sqrt{2Z_0}\,g(-\widehat x)/r$ up to a unit-modulus phase, so $\|g\|^2$ is incoming power. At each direction the tangential projection has trace two, and cross product by $d$ is an isometry on its tangential range. Direct integration of the squared kernels therefore gives \begin{equation}\label{cg:hilbert-schmidt} \|\mathsf E\|_{\rm HS}^2=\frac{4k_0^2Z_0A}{\pi},\qquad \|\mathsf H\|_{\rm HS}^2=\frac{4k_0^2A}{\pi Z_0}. \end{equation} For $Jv=v\times\widehat z$, $\|J\|\leq1$, and the flux operator is \begin{equation}\label{cg:flux-factor} Q_D^0=\tfrac14 (\mathsf E^*J\mathsf H+\mathsf H^*J^*\mathsf E). \end{equation} The product of two Hilbert--Schmidt operators is trace class, with trace norm bounded by the product of their Hilbert--Schmidt norms. The triangle inequality proves \eqref{cg:trace-bound}. The electric inversion map $E(x)\mapsto E(-x)$ acts unitarily on the incoming space. Its two eigenspaces are the electric parity sectors. For a same-parity field the magnetic field has the opposite parity by the curl equation, so its Poynting density is odd and its flux through centered $D$ vanishes. Polarization of this identity makes both diagonal blocks of $Q_D^0$ zero. The parity involution therefore conjugates $Q_D^0$ to $-Q_D^0$. Its positive and negative spectra have equal multiplicities, and their trace norms are equal. This proves \eqref{cg:positive-trace}. The compact selfadjoint min--max principle, followed by summing positive eigenvalues at least $t$, proves the dimension assertion. \end{proof} \subsection{An explicit finite ordinary-dielectric stack} Fix $1x_1$, and in a mirror interval with reference index $n\in\{1,q\}$ it is \begin{equation}\label{cg:material} \varepsilon(x)=n^2+2/x^2. \end{equation} This is a compact, real, scalar, nonmagnetic dielectric with finitely many smooth radial pieces. If $x_0\geq\sqrt{2/(\varepsilon_+-q^2)}$, it obeys $1\leq\varepsilon\leq\varepsilon_+$ everywhere. Here are explicit finite-band radius requirements. Write \begin{equation}\label{cg:hankel-constants} p_l(x)=1+\sum_{s=1}^l a_{ls}x^{-s},\qquad S_l=\sum_{s=1}^l|a_{ls}|,\quad T_l=\sum_{s=1}^l s|a_{ls}|,\quad K_L=1+\max_{1\leq l\leq L}(2S_l+T_l), \end{equation} where $a_{ls}$ are the finite Hankel coefficients in \eqref{ms:hankel-polynomials}, and set $D_L=L(L+1)+2$. In addition to the material bound, require \begin{equation}\label{cg:radius-choice} x_0\geq1,\qquad x_0\geq 8K_L q^{2M},\qquad x_0^2\geq32 D_Ld_Mq^{4M+4}. \end{equation} An integer multiple of $\pi$ satisfying these finitely many lower bounds always exists. The growing radius is an allowed resource of this construction and is not hidden in a fixed-support claim. \begin{theorem}[Simultaneous reciprocal gain of all prescribed partial waves] \label{cg:simultaneous-gain} For the dielectric \eqref{cg:material} with \eqref{cg:radius-choice}, let $1\leq l\leq L$ and consider either polarization. Normalize its real regular radial solution in the vacuum core to $u_l(x)=\psi_l(x)=xj_l(x)$, and write outside the device \begin{equation}\label{cg:outside-coefficient} u_l(x)=A_l\xi_l^-(x)+\overline{A_l}\xi_l^+(x). \end{equation} The coefficient is nonzero. For TE use the electric radial amplitude, and for TM the magnetic radial amplitude. Uniformly over the prescribed band and all azimuthal indices, \begin{equation}\label{cg:amplitude-bounds} \frac{3}{16\sqrt2}\,q^{\sigma M} \leq |A_l|\leq\frac{5\sqrt2}{4}\,q^{\sigma M}, \end{equation} where $\sigma=-1$ for TE odd $l$ and TM even $l$, and $\sigma=+1$ for TE even $l$ and TM odd $l$. Relative to the free-space regular solution with the same physical incoming amplitude, the core multiplier is $g_l=(2A_l)^{-1}$. Thus the electric-odd parity sector has gain $q^M$ times a complex factor whose modulus lies between $2/(5\sqrt2)$ and $8\sqrt2/3$; the electric-even sector has gain $q^{-M}$ times such a factor. \end{theorem} \begin{proof} For TE take the dimensionless state $v=(u,u_x')^T$, and for TM take $v=(u,\varepsilon^{-1}u_x')^T$. These states are continuous at every layer interface, and satisfy, respectively, \begin{equation}\label{cg:generators} v'=\begin{pmatrix}0&1\\l(l+1)/x^2-\varepsilon&0\end{pmatrix}v, \qquad v'=\begin{pmatrix}0&\varepsilon\\l(l+1)/(\varepsilon x^2)-1&0\end{pmatrix}v. \end{equation} Their reference generators on an index-$n$ interval are obtained by discarding the $x^{-2}$ terms and setting $\varepsilon=n^2$. The exact quarter-period reference matrices are \begin{equation}\label{cg:quarter-matrices} T_n^{\rm TE}=\begin{pmatrix}0&1/n\\-n&0\end{pmatrix},\qquad T_n^{\rm TM}=\begin{pmatrix}0&n\\-1/n&0\end{pmatrix}. \end{equation} The full reference products, in physical layer order, are \begin{equation}\label{cg:reference-products} U_0^{\rm TE}(x_1,x_0)=(-1)^M\operatorname{diag}(q^{-M},q^M), \qquad U_0^{\rm TM}(x_1,x_0)=(-1)^M\operatorname{diag}(q^M,q^{-M}). \end{equation} Within an individual reference interval, each propagator and its inverse has norm at most $q$. Combining complete period products with the at most two partial intervals gives the safe bounds \begin{equation}\label{cg:prefix-bounds} \|U_0(x,x_0)\|,\ \|U_0(x,x_0)^{-1}\|\leq q^{M+2} \quad(x_0\leq x\leq x_1). \end{equation} For either polarization the difference of the actual and reference generators has norm at most $D_L/x_0^2$. For TE it is the single entry $(l(l+1)-2)/x^2$. For TM its entries are $2/x^2$ and $l(l+1)/(\varepsilon x^2)$, with $\varepsilon\geq1$. Writing the actual propagator as $U=U_0W$ gives a Volterra equation whose coefficient norm integrates to at most \begin{equation}\label{cg:beta} \beta=\frac{D_Ld_Mq^{2M+4}}{x_0^2} \leq\frac{q^{-2M}}{32}. \end{equation} Gronwall, followed by the exact endpoint norm in \eqref{cg:reference-products}, gives \begin{equation}\label{cg:transfer-error} \|U(x_1,x_0)-U_0(x_1,x_0)\| \leq q^M(e^\beta-1)\leq2q^M\beta. \end{equation} This estimate retains the possible growth of the reference transfer matrix; no small generator error is identified with a small endpoint error without this amplification factor. The finite Hankel formulas give, for $x\geq1$, \begin{equation}\label{cg:bessel-errors} |\psi_l(x)-\sin(x-l\pi/2)| +|\psi_l'(x)-\cos(x-l\pi/2)|\leq K_L/x. \end{equation} They also give $|\xi_l^+(x)|,|\xi_l^{+\prime}(x)|\leq2$ for $x\geq K_L$. At the multiple $x_0$ of $\pi$, the ideal initial state is a signed first coordinate vector for odd $l$ and a signed second coordinate vector for even $l$. Its error has norm at most $K_L/x_0\leq q^{-2M}/8$, and the actual initial state has norm at most two. Combining this with \eqref{cg:transfer-error} shows that the actual final state differs from its ideal value in \eqref{cg:reference-products} by at most \begin{equation}\label{cg:state-error} q^M\frac{K_L}{x_0}+4q^M\beta\leq\frac14q^{-M}. \end{equation} Therefore, with $\sigma$ as in the theorem, \begin{equation}\label{cg:state-size} \tfrac34q^{\sigma M}\leq|v(x_1)| \leq\tfrac54q^{\sigma M}. \end{equation} In the exterior both state conventions equal $(u,u')$. Since $W(\xi_l^-,\xi_l^+)=2i$, the exterior coefficient satisfies $A_l=(u\xi_l^{+\prime}-u'\xi_l^+)/(2i)$. The Hankel bounds and the conjugate coefficient representation give both inequalities \begin{equation}\label{cg:coefficient-state} |v(x_1)|/(4\sqrt2)\leq|A_l|\leq\sqrt2|v(x_1)|. \end{equation} This proves \eqref{cg:amplitude-bounds}. Nonvanishing also follows directly from uniqueness: a zero incoming coefficient of a real regular solution would make both exterior coefficients zero. For a vacuum regular wave $\psi_l=(\xi_l^-+\xi_l^+)/2$, so its incoming coefficient is $1/2$. Fixed vacuum power normalization therefore gives the exact relative core multiplier $1/(2A_l)$ for either polarization. Finally the electric parity is $(-1)^l$ for TE and $(-1)^{l+1}$ for TM. This identifies the amplified and suppressed sectors as stated. \end{proof} \subsection{What happens to physical passage in this entire finite band} \begin{theorem}[An area bound survives the common-stack gains] \label{cg:stack-count} Let $D=(-L_D/2,L_D/2)^2\times\{0\}$, with area $A=L_D^2$, be inside the vacuum core of the preceding construction. Let $\mathcal H_L$ contain all power-normalized incoming TE and TM partial waves with $1\leq l\leq L$ and all their azimuthal indices. For any physical selected-output contraction $T$ of this device, every counted subspace $W\subset\mathcal H_L$ satisfies \begin{equation}\label{cg:stack-dimension} \dim W\leq\frac{128}{9\pi}\,\frac{k_0^2A}{\eta\tau}. \end{equation} The bound is independent of $L,M,x_0$, the local resonant field norm, and the exterior support size, provided the explicit radius conditions are met. Its proof uses no bypass hypothesis, so it applies in particular whenever the frozen absolute-bypass condition also holds. It does not assert a bound for inputs in angular orders above $L$. \end{theorem} \begin{proof} Identify the finite physical incoming basis with the corresponding orthonormal vacuum basis. On the core, every actual field equals the regular vacuum field with its coefficient in each mode multiplied by $g_l$. Let $P_-$ and $P_+$ be the electric-odd and electric-even parity projections. Theorem~\ref{cg:simultaneous-gain} writes this diagonal multiplier exactly as \begin{equation}\label{cg:gain-factor} G=(q^MP_-+q^{-M}P_+)B, \qquad \|B\|\leq C:=8\sqrt2/3, \end{equation} where every diagonal entry of $B$ is nonzero. Its complex phases are included in $B$; no common phase is assumed. Both diagonal parity blocks of the vacuum section-flux operator vanish, by the inversion argument in Lemma~\ref{cg:vacuum-trace}. The two real gain factors in \eqref{cg:gain-factor} therefore cancel in each cross block, giving the exact identity \begin{equation}\label{cg:flux-cancellation} Q_D^{\rm device}\big|_{\mathcal H_L} =G^*Q_D^0G=B^*Q_D^0B. \end{equation} For counted inputs, $\langle a,Q_D^{\rm device}a\rangle \geq\eta\tau\|a\|^2$. Hence $V=BW$ has the same dimension as $W$ and obeys \begin{equation} \langle b,Q_D^0b\rangle \geq\frac{\eta\tau}{C^2}\|b\|^2\quad(b\in V). \end{equation} Apply \eqref{cg:positive-trace} and $C^2=128/9$. This proves \eqref{cg:stack-dimension} for all coherent inputs, without replacing the device transmission by an isolated link. \end{proof} \begin{theorem}[Arbitrary-rank physical field amplification with zero plane flux] \label{cg:resource-counteraudit} Fix $N\geq1$, a ball $B_R$ centered on the section plane, the same frequency and material upper bound. There is a sequence of the finite ordinary dielectrics above and an $N$-dimensional physically power-normalized input subspace $U_N$ such that the full scattering map is unitary on $U_N$, every coherent field has $S_z=0$ almost everywhere on the entire plane, and \begin{equation}\label{cg:resource-growth} \int_{B_R}|E_a|^2\,dx \ \geq\ \frac{16Z_0}{25}q^{2M} \min_{1\leq j\leq N} \int_0^R|\psi_{2j-1}(k_0r)|^2\,dr\ \|a\|^2 \qquad(a\in U_N). \end{equation} The positive constant after $q^{2M}$ is independent of $M$. Thus every unit vector in this physical subspace has diverging local field energy while full-map throughput is one and absolute bypass is zero. Its passage through $D$ is also zero, so it is not a counted subspace for any $\eta>0$. \end{theorem} \begin{proof} Take $L=2N-1$ and the TE modes $(l,m_{\mathrm{az}})=(2j-1,0)$, $1\leq j\leq N$. Choose $x_0$ also greater than $k_0R$ for each $M$. Radial real dielectrics preserve these orthogonal channels. In each channel the outgoing-to-incoming coefficient is $\overline{A_l}/A_l$, of modulus one, so their full physical scattering map is unitary. On the equatorial plane, $Y_{l0}=0$ for odd $l$ and $\partial_\varphi Y_{l0}=0$. A TE electric field there is purely azimuthal, whereas its magnetic field is normal to the plane. These component properties persist for every radius, in every layer, and for every coherent sum of the selected channels. Their normal Poynting component is consequently zero everywhere except possible interface circles, a set of plane measure zero. A free unit-power TE partial wave has electric radial coefficient $2\sqrt{2Z_0}\,\psi_l(k_0r)/r$ with unit-normalized $X_{l0}$. The actual coefficient is multiplied by $g_l$. The lower bound in Theorem~\ref{cg:simultaneous-gain} gives $|g_l|\geq2q^M/(5\sqrt2)$ for these odd orders. Angular orthogonality on $B_R$ removes all cross terms and proves \eqref{cg:resource-growth}. Each radial integral is strictly positive because the regular Bessel solution is not identically zero. The frozen passage inequality would require $0=P_D(a)\geq\eta\|Ta\|^2>0$ for a nonzero efficient input; thus no such input is counted. \end{proof} \begin{remark}[What the growing-support test establishes] \label{cg:scope} The simultaneous gain theorem closes a specific global synthesis step for actual finite dielectrics: there is no assumed coupled-mode model, no independent lumped loads, and no unproved extension of local Cauchy data. It also shows that an estimate of local field norm from passivity, throughput, and bypass alone cannot be uniform, even on an arbitrarily large fixed-dimensional physical subspace. The positive-passage condition remains essential. For this common radial stack its required opposite-parity product cancels the large gain, leading to the rigorous bandwise area bound \eqref{cg:stack-dimension}. To escape that result by angular-selective inner layers one would have to change the parity-dependent radial gain pattern while retaining full-plane bypass control. The local inverse-synthesis theorem does not prove this nonlocal statement. Neither the construction nor its bandwise exclusion settles the unrestricted channel conjecture. \end{remark}