\section{The small-inclusion limit for an explicit parity reaction kernel} \label{cl:section} We now discharge the small-inclusion and whole-plane error estimates for a specific limiting core Green kernel. This is a theorem about that explicit kernel. Its identification as the limit of physical finite radial backgrounds, together with control of the remote annulus, is proved separately in Theorem~\ref{rg:physical-limit}. Neither step is inferred just from finite-band incident amplification; their combination gives Theorem~\ref{uc:counterexample}. The important point is self-consistency. Once an inclusion is introduced, its induced field feeds back on the other inclusions and on the background; simply adding prescribed dipole fields would omit that interaction. We solve the inclusion equation for the explicit limiting kernel and estimate the difference between the resulting plane flux and the desired edge profile in an integral norm over the whole plane. The separate radial Green limit is what connects this calculation to actual finite stacks. \subsection{The exact kernel and elementary plane estimates} Write $k=k_0$, $\Phi_k(x)=e^{ik|x|}/(4\pi|x|)$, and use the dimensionless electric polarization $P=\chi E$, where $\chi=\eps_r-1$. The free electric response is \begin{equation}\label{cl:free} \mathcal K_0P=(k^2I+\nabla\nabla)\Phi_k*P =-\mathcal P P+\mathcal D P. \end{equation} Here $\mathcal P$ is the longitudinal orthogonal projection on $L^2(\mathbb R^3;\mathbb C^3)$, whose Fourier multiplier is $\xi\xi^T/|\xi|^2$. Thus $\mathcal P\geq0$ and $\|\mathcal P\|=1$. The remainder $\mathcal D$ has a kernel bounded by $C/|x|$ on every fixed bounded set. Outside the support of $P$ the static kernel $-\mathcal P$ is bounded by $C/|x|^3$. The magnetic response is \begin{equation}\label{cl:magnetic} \mathcal HP(x)=\frac{\nabla\times\mathcal K_0P(x)}{i\omega\mu_0} =\frac{k^2}{i\omega\mu_0} \int\nabla\Phi_k(x-y)\times P(y)\,dy. \end{equation} Its kernel is bounded by $C(|x|^{-2}+|x|^{-1})$. Let $R_z=\operatorname{diag}(1,1,-1)$ and define the electric-field wall projections \begin{equation}\label{cl:projections} \mathcal W E(x)=-R_zE(R_zx),\qquad \Pi_{\rm e}=\tfrac12(I+\mathcal W),\qquad \Pi_{\rm m}=\tfrac12(I-\mathcal W). \end{equation} The first range has electric-wall parity. The entire real kernel \begin{equation}\label{cl:J} J(x)=(k^2I+\nabla\nabla)\frac{\sin(k|x|)}{4\pi|x|} =\frac{k^3}{16\pi^2}\int_{S^2} (I-\theta\theta^T)e^{ik\theta\cdot x}\,d\theta \end{equation} is the imaginary part of the free electric kernel. Set \[ \mathcal R_{\rm e}P=i\Pi_{\rm e}(J*P),\qquad \mathcal R_{\rm m}P=i\Pi_{\rm m}(J*P). \] For an actual core gain parameter $B>0$, define \begin{equation}\label{cl:kernel} \mathcal K_BP=\mathcal K_0P +(B^2-1)\mathcal R_{\rm e}P +(B^{-2}-1)\mathcal R_{\rm m}P. \end{equation} All derivatives in these definitions are distributional where necessary. Away from the source support the formulas are classical. Both reaction fields are entire vacuum Maxwell electric fields; their magnetic partners are their curls divided by $i\omega\mu_0$. On $z=0$ the electric-wall reaction has zero tangential electric field, and the opposite-wall reaction has zero tangential magnetic field. \begin{lemma}[Uniform local and far-field kernel bounds] \label{cl:kernelbounds} Fix a ball containing all source supports under consideration. On each fixed observation ball, $\mathcal R_{\rm e}P$, $\mathcal R_{\rm m}P$, and their magnetic partners, with any fixed number of derivatives, are bounded by $C\|P\|_1$. For $|x|=r$ outside a larger fixed ball, each has an expansion \begin{equation}\label{cl:transverse} F(x)=\frac{e^{ikr}F_+(\widehat x)+e^{-ikr}F_-(\widehat x)}r +O(C\|P\|_1r^{-2}),\qquad F_\pm(\widehat x)\cdot\widehat x=0, \end{equation} with $\|F_\pm\|_\infty\leq C\|P\|_1$. The free electric and magnetic responses have the same statement with $F_-=0$. For the dynamic electric response $\mathcal D P$ the same leading far-field statement holds. These estimates are uniform over all $L^1$ polarizations in that fixed source ball. If two such fields have amplitude bounds $M_1,M_2$ in this statement, then on the measurement plane outside the fixed larger disk \begin{equation}\label{cl:tailproduct} |(F_1\times\overline{F_2})\cdot\widehat z| \leq C M_1M_2r^{-3},\qquad \int_{r>R}|(F_1\times\overline{F_2})\cdot\widehat z|\,dA \leq C M_1M_2/R. \end{equation} The same assertions apply to any fixed finite collection of regular vacuum waves, with its coefficient norm replacing $\|P\|_1$. \end{lemma} \begin{proof} The integral in \eqref{cl:J} gives all compact derivative bounds directly, including after reflection. The reaction fields are smooth spherical Herglotz fields. Their angular amplitudes and all their derivatives are bounded by $C\|P\|_1$, since $y$ stays in a fixed ball in the factor $e^{-ik\theta\cdot y}$. Stationary phase at the two points $\theta=\pm\widehat x$ gives \eqref{cl:transverse}, uniformly in direction. The factor $I-\theta\theta^T$ makes the electric leading amplitudes transverse. A curl gives the transverse magnetic leading amplitudes and the same remainder order. Equivalently these statements follow by differentiating the explicit sine kernel. For the free responses, expand $|x-y|=r-\widehat x\cdot y+O(r^{-1})$ and the derivatives of $\Phi_k(x-y)$ uniformly for bounded $y$. The electric leading matrix is $k^2(I-\widehat x\widehat x^T)e^{ikr}/(4\pi r)$; the magnetic leading vector is a cross product with $\widehat x$. Both are transverse. The static subtraction is $O(\|P\|_1r^{-3})$, so it does not change the dynamic leading terms. On the plane, $\widehat x\cdot\widehat z=0$. The cross product of two vectors transverse to $\widehat x$ is parallel to $\widehat x$, even with complex coefficients. Consequently every $r^{-2}$ leading contribution to the normal cross product is zero. The remainders give \eqref{cl:tailproduct}, whose radial integral is finite. Finite regular waves have the same transverse incoming and outgoing expansions. \end{proof} \begin{lemma}[Small-source plane estimates] \label{cl:planebounds} Let $\Omega_a$ be the union of the $2N$ radius-$\rho a\ell_0$ balls from Proposition~\ref{ct:model}, where $N$ is fixed and $a\downarrow0$. They lie at distances at least $(h-\rho)a\ell_0$ from the plane, and $|\Omega_a|=O(a^3)$. For $P\in L^2(\Omega_a)$, extended by zero, and a fixed plane disk $K$ containing all their projections, \begin{align} \|\mathcal P P\|_{L^1(\mathbb R^2)} &\leq C a^{-1}\|P\|_1,\label{cl:staticL1}\\ \|\mathcal P P\|_{L^2(\mathbb R^2)} &\leq C a^{-1/2}\|P\|_2,\label{cl:staticL2}\\ \|\mathcal D P\|_{L^1(K)} &\leq C\|P\|_1,\label{cl:dynamicL1}\\ \|\mathcal K_0P\|_{L^2(K)} &\leq C a^{-1/2}\|P\|_2,\qquad \|\mathcal HP\|_{L^2(K)} \leq C a^{1/2}\|P\|_2.\label{cl:nearL2} \end{align} The constants are independent of $a$ and $P$. \end{lemma} \begin{proof} For a source point $y$ at height $d\geq c a$, \[ \int_{\mathbb R^2}|(x,0)-y|^{-3}\,dx=2\pi/d, \qquad \int_{\mathbb R^2}|(x,0)-y|^{-6}\,dx=\pi/(2d^4). \] The first proves \eqref{cl:staticL1}. The second, integrated over $\Omega_a$, bounds the squared Hilbert--Schmidt norm by $C a^3a^{-4}=C a^{-1}$, proving \eqref{cl:staticL2}. The planar integral of $|x-y|^{-1}$ over $K$ is uniformly bounded, which proves \eqref{cl:dynamicL1}. For the magnetic squared kernel, the $|x-y|^{-4}$ planar integral is $\pi/d^2$, and the $|x-y|^{-2}$ integral over $K$ is $O(1+|\log d|)$. Its squared Hilbert--Schmidt norm is thus $O(a)+O(a^3|\log a|)=O(a)$. The dynamic electric squared kernel has Hilbert--Schmidt norm squared $O(a^3|\log a|)$. Combine these with the static bound to obtain \eqref{cl:nearL2}. \end{proof} \subsection{Self-consistent inclusions and the complete plane remainder} Fix the interpolants $v_j$, their number $\lambda$, the physical incoming basis $g_j$, and the geometry from Lemmas~\ref{ct:jets}--\ref{ct:gram}. Let $(E^v(c),H^v(c))$ be the finite regular electric-wall field $\sum_jc_jv_j$. For fixed $t>0$ put \begin{equation}\label{cl:scales} A_a=t/a,\qquad B_a=\sqrt\lambda A_a,\qquad (E^{\rm bg}_a,H^{\rm bg}_a)=A_a(E^v(c),H^v(c)),\qquad \chi_a=(\eps_p-1)\mathbf1_{\Omega_a}. \end{equation} The distinction between $A_a$ and $B_a$ is necessary: $A_a$ is the jet scale, whereas $B_a$ is the actual common core multiplier on the orthonormal physical inputs $g_j$. \begin{theorem}[Explicit-kernel small-inclusion remainder] \label{cl:remainder} For all sufficiently small $a$, the equation \begin{equation}\label{cl:integralequation} P_a=\chi_a\bigl(E^{\rm bg}_a+\mathcal K_{B_a}P_a\bigr) \quad\hbox{in }L^2(\Omega_a) \end{equation} has a unique solution. Define the auxiliary total fields by \[ E_a=E^{\rm bg}_a+\mathcal K_{B_a}P_a, \qquad H_a=\frac{\nabla\times E_a}{i\omega\mu_0}. \] They solve Maxwell's equations with the actual sphere permittivities locally and outside the spheres, with the feedback condition encoded by \eqref{cl:kernel}. If $F_a(c)$ is the explicit cross-profile of Proposition~\ref{ct:model}, then \begin{equation}\label{cl:actualremainder} \left\|\tfrac12\operatorname{Re}(E_a\times\overline{H_a}) \cdot\widehat z-F_a(c)\right\|_{L^1(\mathbb R^2)} \leq C a\|c\|^2. \end{equation} All constants may depend on the fixed $N$, points, interpolants, $\lambda,t,k,\rho,h,\eps_p$, but not on $a$ or $c$. The theorem does not normalize these auxiliary fields by their own free-space incoming amplitude. Their use for physical unit-power scattering requires the separate finite-background limit. \end{theorem} \begin{proof} It suffices to prove estimates for $\|c\|=1$; linearity of the fields then gives the stated quadratic dependence. Write $s_a=\sqrt{\chi_a}$ and $u_a=s_aE_a$, so $P_a=s_au_a$. Equation \eqref{cl:integralequation} becomes \begin{equation}\label{cl:positiveblock} \bigl(I+s_a\mathcal P s_a-s_a\mathcal D s_a -(B_a^2-1)s_a\mathcal R_{\rm e}s_a -(B_a^{-2}-1)s_a\mathcal R_{\rm m}s_a\bigr)u_a =s_aE^{\rm bg}_a. \end{equation} The first two terms form a self-adjoint operator at least $I$, so its inverse has norm at most one. A Schur estimate for the $|x-y|^{-1}$ dynamic kernel on $\Omega_a$ is $O(a^2)$: within one ball integrate $|x-y|^{-1}$ over a ball of radius $O(a)$, and between distinct balls use their fixed separation. The smooth reaction kernels have operator norms $O(a^3)$. Since $B_a=O(a^{-1})$, the norm of the remaining perturbation in \eqref{cl:positiveblock} is $O(a^2)+O(B_a^2a^3)=O(a)$. A Neumann series after the positive inverse proves invertibility, with inverse norm at most two for small $a$. In particular \begin{equation}\label{cl:sourcebounds} \|P_a\|_2\leq C a^{1/2},\qquad \|P_a\|_1\leq C a^2. \end{equation} The equation and all its kernels commute with the rotation by $\pi$ about $z$; the source and solution are in its invariant sector. No unsupported odd-$m$ extension of a core kernel is used. Let $P_a^0$ be constant on each sphere, with the isolated static value \begin{equation}\label{cl:leadingP} P_a^0\big|_{\Omega_{a,j}^{\pm}} =(\eps_p-1)\frac3{\eps_p+2} E^{\rm bg}_a(p_j^{\pm}),\qquad p_j^+=p_j, p_j^-=-p_j. \end{equation} These vectors are normal to the plane. For one isolated sphere they solve the static equation exactly: the interior electric field of its constant polarization is $-P_a^0/3$, so $P_a^0=\chi(E^{\rm bg}_a-P_a^0/3)$. The actual background differs from its value at $p_j^{\pm}$ by $O(A_aa)=O(1)$ throughout the corresponding sphere. Its $L^2$ residual is $O(a^{3/2})$. Static interaction between distinct spheres has operator norm $O(a^3)$, because their mutual distances remain positive. The dynamic and reaction perturbations on $P_a^0$ have norm $O(a)\|P_a^0\|_2=O(a^{3/2})$ by the preceding bounds. Applying the uniformly bounded inverse in \eqref{cl:positiveblock} to this residual proves \begin{equation}\label{cl:sourceerror} \|P_a-P_a^0\|_2\leq C a^{3/2},\qquad \|P_a-P_a^0\|_1\leq C a^3. \end{equation} It remains to estimate the whole plane, not merely a compact set. Abbreviate the free response by $(E^f,H^f)$, and the two full reaction responses, including their coefficients in \eqref{cl:kernel}, by $(E^e,H^e)$ and $(E^m,H^m)$. On the plane the wall identities give exactly \begin{equation}\label{cl:fluxsplit} E_{a,t}=E^f_t+E^m_t,\qquad H_{a,t}=H^{\rm bg}_{a,t}+H^f_t+H^e_t. \end{equation} By Lemma~\ref{cl:kernelbounds} and \eqref{cl:sourcebounds}, the opposite-wall reaction fields have compact and far-field amplitude $O(a^2)$, while the wall reaction fields have amplitude $O(B_a^2a^2)=O(1)$. The background fields have amplitude $O(A_a)$. First take the cross product with $H^{\rm bg}_a$ in \eqref{cl:fluxsplit}. Replacing the static free electric response of $P_a$ by that of $P_a^0$ has plane $L^1$ error at most $C a^{-1}\|P_a-P_a^0\|_1=O(a^2)$ by \eqref{cl:staticL1}. Since the background is globally bounded by $C A_a$, its flux error is $O(a)$. The dynamic free electric response has $L^1$ norm $O(a^2)$ on a fixed disk, by \eqref{cl:dynamicL1}; outside that disk the transverse cancellation \eqref{cl:tailproduct} gives a flux bound $O(A_a\|P_a\|_1)=O(a)$. The same compact bound and tail argument apply to $E^m\times\overline{H^{\rm bg}_a}$. The remaining static field of $P_a^0$ is exactly a sum of normal dipoles, with the polarizabilities from Lemma~\ref{ct:profile}. For each individual dipole, its tangential plane trace satisfies \begin{equation}\label{cl:firstmoment} \int_{\mathbb R^2}|E^{0,j}_t(x)|\,dA\leq C a, \qquad \int_{\mathbb R^2}|x-p_j^{\pm}|\,|E^{0,j}_t(x)|\,dA \leq C a^2. \end{equation} These follow directly by scaling \eqref{ct:dipoletrace}; the second radial integrand at infinity is $O(r^{-2})$ and is integrable. A finite regular background has global Lipschitz constant $C A_a$. Replacing its magnetic field in each such cross product by $H^{\rm bg}_a(p_j^{\pm})$ therefore costs at most $C A_a a^2=O(a)$. The resulting sum is exactly $F_a(c)$, including the coefficients and signs in \eqref{ct:profileformula}. We finally control all cross products with the two correction magnetic fields in \eqref{cl:fluxsplit}. On a fixed disk, \eqref{cl:nearL2} and \eqref{cl:sourcebounds} give \[ \|E^f\|_2\leq C, \qquad \|H^f\|_2\leq Ca, \qquad \|E^m\|_\infty\leq Ca^2, \qquad \|H^e\|_\infty\leq C. \] Thus the free self-cross product and the product $E^m\times\overline{H^f}$ have $L^1$ norm $O(a)$. For products with $H^e$, use $\|\mathcal P P_a\|_{L^1}\leq C a^{-1}\|P_a\|_1=O(a)$, $\|\mathcal D P_a\|_{L^1(K)}=O(a^2)$, and the bounded opposite-wall reaction, obtaining another $O(a)$ bound. Outside the disk, \eqref{cl:tailproduct} applies to these products with amplitude pairs $(a^2,a^2)$ or $(a^2,1)$; both give $O(a)$ or smaller. All normal products are therefore integrable, and their sum proves \eqref{cl:actualremainder}. \end{proof} \begin{remark}[What has been discharged] \label{cl:scope} For the explicit kernel \eqref{cl:kernel}, the sufficient plane remainder in Proposition~\ref{ct:completion} is now a conclusion, not an assumption. The proof uses the uniformly positive static block, includes every coherent input, and estimates the full absolute plane flux. Theorem~\ref{rg:physical-limit} supplies the separate physical sequence: its core source Green kernels converge to \eqref{cl:kernel}, and its remote-annulus and exterior plane-flux contributions tend to zero at each fixed $a$. The order is fixed $a$ first, remote radius second, then a diagonal sequence with $a\downarrow0$. Theorem~\ref{uc:counterexample} combines these independently proved limits into finite ordinary devices with the original physical incoming normalization. \end{remark}