\section{An actual radial Green limit with whole-plane flux convergence} \label{rg:section} This section proves a source-response and whole-plane convergence theorem for finite ordinary dielectric stacks. The gain is fixed while the stack radius tends to infinity. A later diagonal choice may select that radius after the gain and the inclusion geometry. This order avoids assuming a uniform resolvent bound over unrelated devices. Earlier transfer estimates control finitely many incident partial waves. Inserted inclusions, however, generate sources whose response contains all angular orders and can affect the plane far from the square. To complete the physical argument we must control that entire source response, including the remote layers, rather than just the selected incident modes. The proof first identifies the limiting source kernel at fixed gain, then bounds the angular tails and the field in the layers, and finally proves convergence of the absolute-flux error over the full measurement plane. \subsection{The common stack and its source coefficient} Fix $1) +\gamma_l\psi_l(x)\psi_l(y). \end{equation} For free outgoing vacuum $\gamma_l=-i$. \end{lemma} \begin{proof} The exterior coefficient vector is $(1,i)^T$, hence the core coefficient vector is $M_l^{-1}(1,i)^T=(d-ib,-c+ia)^T$. Division and $ad-bc=1$ give \eqref{rg:gamma}. The row norm giving $|\alpha_l|$ is at most $\|M_l^{-1}\|$, which equals $\|M_l\|$ for a real determinant-one $2$ by $2$ matrix. Also $|\beta_l|=|M_le_1|\geq\|M_l\|^{-1}$. The exterior incoming coefficient of the origin-normalized regular solution is $(a+ic)/2$, giving the physical multiplier $g_l=1/(a+ic)$ and \eqref{rg:gammabound}. Finally $W(\psi_l,z_l)=\beta_l$, so the Green function $\psi_l(x_<)z_l(x_>)/\beta_l$ has the asserted jump. The weighted TM derivative equals the ordinary derivative in the core. The real part of $\gamma_l$ is not controlled by incoming gain alone; the full transfer is used here. \end{proof} \subsection{Uniform angular tails at fixed gain} \begin{lemma}[Low and intermediate orders]\label{rg:lowmiddle} Fix $0q^2X^2$ let \[ \gamma_*=\frac{1+\sqrt{1+4(\nu-R^2)}}2,\qquad \theta_q=\sqrt{1-q^{-2}}. \] Then the outgoing core load $h=z_l'(R)/z_l(R)$ has $\operatorname{Re}h\leq0$, and \begin{align} |\gamma_l|&\leq \frac{|\chi_l'(R)|}{\psi_l'(R)} +\frac{|\chi_l(R)|}{\psi_l(R)},\label{rg:loadbound}\\ \psi_l(R)&\geq\frac{R^{l+1}}{(2l+1)!!} \exp\left[-\frac{R^2}{2(l+\gamma_*+2)}\right]. \label{rg:psilower} \end{align} For fixed $s0$ for TE, and $\eps_r|p|^2+(\nu/(\eps_rx^2)-1)|u|^2>0$ for TM. At $X$ its value is nonpositive by \eqref{rg:hankelpositive}. Its value at every inner point is therefore nonpositive. The conserved nonzero imaginary Wronskian excludes $u=0$, so $\operatorname{Re}h\leq0$. The regular solution is positive and increasing up to $R$. For $y=x\psi_l'/\psi_l$, the Riccati equation is $xy'=\nu-x^2+y-y^2$. Comparison with its constant lower barrier $\gamma_*$ gives $y\geq\gamma_*$ on $(0,R]$; comparison with $l+1$ gives $y\leq l+1$. Solving the load equation gives \[ \gamma_l=\frac{\chi_l'-h\chi_l}{h\psi_l-\psi_l'}. \] Since $\psi_l,\psi_l'>0$ and $\operatorname{Re}h\leq0$, $|h\psi_l-\psi_l'|^2\geq |h|^2\psi_l^2+(\psi_l')^2$. This proves \eqref{rg:loadbound}. Put $v=\psi_l/x^{l+1}$. Then $(x^{2l+2}v')'=-x^{2l+2}v$. The Riccati bound gives $v(t)/v(x)\leq(x/t)^{l+1-\gamma_*}$ for $00$; for TE the first coefficient is one and $V=\nu/x^2-\eps_r>0$. The proof of Lemma~\ref{rg:barrier} gave $\operatorname{Re}(p\overline u)\leq0$ throughout this stack. Consequently \[ (|u|^2)'=2\eps_r\operatorname{Re}(p\overline u)\leq0,\qquad (|p|^2)'=2V\operatorname{Re}(u\overline p)\leq0 \] for TM, with coefficient one in the first equation for TE. Both states are continuous across every interface. Together with \eqref{rg:psilower} this proves, on the entire annulus, \begin{equation}\label{rg:barrierstates} |f_{lm}|(|u_l|+|p_l|) \leq C\|P\|_2(l+1)^b e^{C_q'R}(S/R)^{l+1}. \end{equation} The sum over $l\geq qR+O(1)$ decays faster than every inverse power of $R$, including arbitrary fixed polynomial weights. No large-order transfer estimate of the form $\exp(Cl^2/R^2)$ is being used. For low orders, \eqref{rg:low} and the bounded partial transfers give $|f_{lm}|(|u_l|+|p_l|)\leq C_A|f_{lm}|$. For intermediate orders the analogous bound is $e^{C_AR}|f_{lm}|$ by \eqref{rg:middle}. The latter weighted sum is $O(e^{-cR\log R})$ by \eqref{rg:moments}, since $l\geq c_0R+O(1)$. This proves the uniform annulus assertion. It also proves the exterior assertion, including all turning points. By \eqref{rg:hankelpositive}, $|\xi_l(x)|$ decreases to one. Moreover \[ (|\xi_l'|^2)'= 2(\nu/x^2-1)\operatorname{Re}(\xi_l\overline{\xi_l'}), \] so $|\xi_l'|$ decreases up to $\sqrt\nu$ and then increases to one. Thus, for all $x\geq X$, \[ |\xi_l(x)|\leq|\xi_l(X)|,\qquad |\xi_l'(x)|\leq\max\{|\xi_l'(X)|,1\} \leq\max\{|\xi_l'(X)|,|\xi_l(X)|\}. \] Multiply by the actual exterior outgoing coefficient. The exterior states are bounded by the annular endpoint states. This proves the lemma without requiring a far-field expansion uniform in uncontrolled angular order. \end{proof} \begin{corollary}[Absolute annulus and exterior flux] \label{rg:remote} Add a fixed finite-band incident family and its radial-background scattering field, with bounded physical incident coefficients. For its sum with the source response of Lemma~\ref{rg:states}, \begin{equation}\label{rg:radialtangential} |E_{\rm tan}|+|H_{\rm tan}|\leq C/r,\qquad |E_{\rm rad}|+|H_{\rm rad}|\leq C/r^2,\qquad r\geq R/k . \end{equation} The labels refer to spheres centered at the origin. The absolute normal plane flux over the stack annulus is $O(d_A/R^2)$ and over the exterior is $O(1/R)$. \end{corollary} \begin{proof} Tangential vector-wave components are constants times $u/r$ or $p/r$, and radial components are constants times $\sqrt\nu\,u/r^2$, with a bounded $\eps_r^{-1}$ factor where appropriate. The angular factors have polynomial sup-norm bounds. Lemma~\ref{rg:states} gives \eqref{rg:radialtangential}. The finitely many background orders satisfy the same bounds by the fixed-order transfer limit in the annulus and bounded incoming and outgoing coefficients in the exterior. On $z=0$ the spherical radial direction lies in the plane. Two sphere-tangential vectors have a radial cross product, whose normal component is zero. Thus only radial--tangential products contribute and $|S_z|\leq C/r^3$. Integrate over the annulus and exterior. Interface circles have plane measure zero. \end{proof} \subsection{The whole-plane physical limit} \begin{theorem}[Source and dielectric whole-plane limit] \label{rg:physical-limit} Fix $A=q^M$, a compact region $K$ at positive distance from $z=0$, and a finite-dimensional finite-band incident family. Use a common phase adjustment so that its radial-background fields converge on compact sets to their limiting core fields. Suppose $P_R\to P$ in $L^2(K)$, uniformly on that finite input family. Form the total fields by adding the actual radial-background response to $P_R$. Define their limiting fields using \eqref{rg:kernellimit} and $P$. Then the normal plane fluxes are integrable and \begin{equation}\label{rg:L1} \sup_{\|c\|=1}\|S_{z,R}(c)-S_{z,\infty}(c)\|_{L^1(\mathbb R^2)} \longrightarrow0 . \end{equation} Furthermore, let $\chi$ be a fixed bounded contrast supported in $K$. If the limiting volume equation \begin{equation}\label{rg:volume} P=\chi(E_\infty^{\rm bg}+{\cal K}_\infty P) \end{equation} has a bounded inverse on $L^2(K)$ (or on its invariant rotation sector), then for all sufficiently large $R$ the actual radial-background volume equation has a bounded inverse, its solutions converge to $P$, and \eqref{rg:L1} holds for the actual scattering fields of the radial layers plus that dielectric perturbation. Their original physical incoming normalization is retained. \end{theorem} \begin{proof} The positive source--plane distance and Theorem~\ref{rg:local} give uniform field convergence on every fixed plane disk, including the magnetic fields obtained by curl. It remains to control the growing vacuum-core part of the plane. Split its reaction series at $\nu=c_0^2R^2$. Low-order coefficients $(\gamma_l+i)f_{lm}$ have uniformly bounded factorial tails by \eqref{rg:low} and \eqref{rg:moments}. The associated Herglotz densities are therefore uniformly bounded in every fixed smooth angular norm, although their cutoff depends on $R$. Their tangential fields are $O(1/r)$ and their radial fields $O(1/r^2)$ uniformly in $R$. For an elementary verification, express the angular integral in polar coordinates about the observation direction and integrate once in $t=\cos\theta$. This gives the first bound. The radial amplitude vanishes at both endpoints by transversality, so a second integration gives the second. The azimuthally averaged smooth amplitude has bounded endpoint derivatives, as follows from its even local Taylor expansion in $\sin\theta$; uniform $C^4$ control suffices. The magnetic angular density is the direction crossed with the electric density, divided by $Z_0$, and has the same properties. The free compact-source response and the fixed finite-band background satisfy these bounds outside a fixed source ball as well. Their combined absolute plane-flux tail beyond radius $B$ is $O(1/B)$, uniformly in $R$. The remaining reaction field is negligible uniformly on $B\leq r\leq R/k$, where $B$ is fixed beyond twice the source ball. In the intermediate range, \eqref{rg:middle} contributes at most $e^{CR}$, the source moments have factorial decay with $l\geq cR$, and the observation factors and first derivatives have polynomial bounds. For example the integral representation of $j_l$ gives $|j_l|,|j_l'|\leq1$; inverse-radius factors are harmless on $r\geq B$. Thus this weighted field sum is $O(e^{-cR\log R})$. For barrier orders, combine \eqref{rg:loadbound} and \eqref{rg:psilower} with the regular upper bounds at $S$ and at any $0B}|S_{z,R}|\,dA\leq C/B. \] The limiting free-plus-Herglotz field has the same tail. Compact convergence, followed by $B\to\infty$, proves \eqref{rg:L1}. The field estimates are linear and uniform on bounded source sets; the quadratic flux inequality then makes the conclusion uniform on every coherent unit input in the fixed finite-dimensional family. For the final assertion, the singular free-volume kernel is identical for all $R$; only its smooth reaction part varies. Theorem~\ref{rg:local} therefore implies \[ \|\chi({\cal K}_R-{\cal K}_\infty)\|_{L^2(K)\to L^2(K)} \longrightarrow0. \] The inverse perturbation identity applied to $I-\chi{\cal K}_\infty$ gives invertibility for large $R$ and convergence of the actual source solutions in $L^2(K)$. The source version of the theorem applies to these solutions. Uniqueness of outgoing scattering for each finite positive lossless dielectric identifies the volume solution with the physical scattering field. \end{proof} \begin{remark}[Quantifiers for the small-inclusion construction] \label{rg:diagonal} For every fixed inclusion size $a$ and fixed gain $B_a$, the preceding theorem is applied first. The radius can then be chosen large enough for any prescribed plane-flux error. There is no claim of a radius bound uniform in $a$. To realize the actual gain $B_a=\sqrt\lambda\,t/a$ in \eqref{cl:scales} using one fixed index $q$, take $a_M=t\sqrt\lambda/q^M$. Then $B_{a_M}=q^M$ exactly and $a_M\downarrow0$. The inverse and remainder theorem \ref{cl:remainder} applies along this sequence. All finite devices still have scalar permittivity between one and $\eps_+$, finite support, and a positive vacuum gap at the section plane for each member. \end{remark}