\section{Coherent bypass and limits of local constructions} \label{ca:section} Theorem~\ref{uc:counterexample} gives a physical construction for $\delta>\eta$. Here we examine the implications of coherent passage and absolute bypass separately from that construction. These conditions imply a global feasibility restriction, but they do not bound the dimension of local Maxwell solutions normalized by total plane flux, even when bypass is controlled on the entire transverse plane. The local fields constructed below do not by themselves have the required far-field accessibility. \subsection{A global power-budget restriction} \begin{proposition}[A necessary threshold condition] \label{ca:powerbudget} Consider the compact-scatterer vacuum-exterior class of Assumptions~\ref{pr:class}--\ref{pr:scattering}, with a legitimate integrable flux on the full plane $\Pi=\{z=0\}$. Use the usual radiation normalization: the angular incoming and outgoing far-field amplitudes $a_\infty,b_\infty$ have squared norms equal to their powers, with $\|b_\infty\|^2\leq\|a_\infty\|^2=\|a\|^2$. Then \begin{equation} -\|a\|^2\leq P_\Pi(a)\leq\|a\|^2, \qquad |P_D(a)|\leq(1+\delta)\|a\|^2 \label{ca:powerbudgeteq} \end{equation} whenever the absolute bypass is at most $\delta\|a\|^2$. Consequently a nonzero counted subspace requires \begin{equation} \eta\tau\leq1+\delta. \label{ca:thresholdfeasibility} \end{equation} In particular, its dimension is zero if $\eta\tau>1+\delta$. \end{proposition} \begin{proof} Apply the real Poynting conservation law in a large upper half-ball. Its flat face has outward normal $-\hat z$. On passing to infinity, the hemispherical outward flux is the outgoing upper-hemisphere power minus the incoming upper-hemisphere power. Thus \begin{equation} P_\Pi(a)=\|\mathbf1_{\Omega_+}b_\infty\|^2 -\|\mathbf1_{\Omega_+}a_\infty\|^2. \label{ca:hemisphere} \end{equation} Here the far-field assertion can be checked directly: the magnetic amplitudes of outgoing and incoming waves are respectively $Z_0^{-1}\hat r\times E_{\rm out}$ and $-Z_0^{-1}\hat r\times E_{\rm in}$. The incoming--outgoing cross terms in the real radial Poynting flux cancel, leaving their squared-power difference. The definition of the radiation power amplitudes includes this integrated far-field limit. Integrability of the plane flux permits the disk integral to converge to $P_\Pi$. Each term on the right of \eqref{ca:hemisphere} is nonnegative, and no larger than $\|a\|^2$, proving the first bound. Also $|P_D-P_\Pi|\leq\int_{\Pi\setminus D}|S_z|\leq\delta\|a\|^2$. Finally the counted inequalities give $\eta\tau\|a\|^2\leq P_D(a)\leq(1+\delta)\|a\|^2$. \end{proof} \begin{remark}[Loewner bounds do not count dimension] \label{ca:loewner} On any finite counted input space let $Q_D$ be its Hermitian section-flux matrix. Polarization and Proposition~\ref{ca:powerbudget} give \begin{equation} \eta T^*T\preceq Q_D\preceq(1+\delta)I, \qquad \tau I\preceq T^*T\preceq I. \label{ca:loewnereq} \end{equation} For $0<\eta\leq1$, the choice $T=I_N$, $Q_D=I_N$, with zero signed and absolute bypass, satisfies these algebraic inequalities for every $N$. This is not a physical construction. It establishes precisely that the coherent Loewner inequalities and global power budget alone supply no rank bound. A spatial Maxwell-to-input accessibility restriction is necessary. \end{remark} \subsection{Arbitrarily many local fields with whole-plane coherent bypass control} The next theorem strengthens the local evanescent example of Proposition~\ref{cx:evanescent}: it certifies absolute bypass on all of $\Pi\setminus D$, rather than only vanishing flux on the sides of a finite collar. Its normalization is total \emph{local plane flux}, not power supplied by asymptotic incoming channels. \begin{theorem}[Coherent local passage with arbitrarily small bypass] \label{ca:localpackets} Fix $k_0>0$, $L>0$, $h>0$, an integer $N\geq1$, and $0<\delta<1$. Let $D=(-L/2,L/2)^2$ and $\Omega_h=\mathbb R^2\times(-h,h)$. There are $N$ linearly independent, smooth, source-free vacuum Maxwell fields on $\Omega_h$, with finite weighted field norm there, such that every coefficient vector $a\in\mathbb C^N$ has \begin{equation} S_z(a;x,y,0)\geq0,\quad P_\Pi(a)=\|a\|^2,\quad P_D(a)\geq(1-\delta)\|a\|^2,\quad \int_{\Pi\setminus D}|S_z(a)|\leq\delta\|a\|^2. \label{ca:packetflux} \end{equation} All tangential Fourier supports can be chosen above any prescribed transverse wavenumber. Thus local source-free Maxwell equations and all-superposition passage and whole-plane bypass alone do not bound the dimension, even at fixed section, frequency, and homogeneous permittivity. \end{theorem} \begin{proof} Use the unitary Fourier transform on each transverse coordinate. For $b>0$, define \begin{equation} \widehat\phi_b(\xi)=(2b)^{-1/2}\mathbf1_{[-b,b]}(\xi),\qquad \phi_b(x)=\frac{\sin(bx)}{\sqrt{\pi b}\,x},\qquad g_b(x,y)=\phi_b(x)\phi_b(y), \label{ca:sinc} \end{equation} with the continuous value at zero. Plancherel gives $\|\phi_b\|_2=\|g_b\|_2=1$. Since $|\sin(bx)|\leq1$, \begin{equation} \int_{|x|\geq L/2}|\phi_b(x)|^2\,dx\leq\frac4{\pi bL}, \qquad \int_{\mathbb R^2\setminus D}|g_b|^2\leq\frac8{\pi bL}. \label{ca:sincescape} \end{equation} Choose $b\geq8N/(\pi L\delta)$, and choose $Q>k_0$ as large as desired. For $j=1,\ldots,N$, put \begin{equation} K_j=Q+3bj,\qquad f_j(x,y)=e^{iK_jx}g_b(x,y),\qquad f_a=\sum_{j=1}^Na_jf_j. \label{ca:packets} \end{equation} Their Fourier supports are disjoint rectangles $[K_j-b,K_j+b]\times[-b,b]$. Thus the $f_j$ are orthonormal and $\|f_a\|_2^2=\|a\|^2$. Pointwise Cauchy--Schwarz gives $|f_a|^2\leq N|g_b|^2\|a\|^2$, hence \begin{equation} \int_{\mathbb R^2\setminus D}|f_a|^2\leq\delta\|a\|^2. \label{ca:packettail} \end{equation} Every support lies in $\xi_x\geq Q+2b>k_0$, entirely in the evanescent sector, but its union is bounded for each finite $N,Q,b$. Use the normalized fields $e=\sqrt{\varepsilon_0}E$, $h=\sqrt{\mu_0}H$ of \eqref{ch:normalization}, and set $C=\sqrt{2/c_0}$. Prescribe the tangential data at $z=0$ by \begin{equation} e_t=(0,Cf_a),\qquad h_t=(-Cf_a,0),\qquad e_z=\frac{iC}{k_0}\partial_yf_a,\qquad h_z=-\frac{iC}{k_0}\partial_xf_a. \label{ca:cauchydata} \end{equation} The normal components are determined by the two normal curl equations. For completeness, the full extension is constructed, not assumed. Put $J(v_x,v_y)=(-v_y,v_x)$ and evolve the tangential components by \begin{equation} \begin{aligned} e_z&=\frac{i}{k_0}(\partial_xh_y-\partial_yh_x),& \partial_z e_t&=\nabla_te_z-ik_0J h_t,\\ h_z&=-\frac{i}{k_0}(\partial_xe_y-\partial_ye_x),& \partial_z h_t&=\nabla_th_z+ik_0J e_t. \end{aligned} \label{ca:cauchyevolution} \end{equation} After transverse Fourier transformation this is a four-component constant coefficient ordinary differential equation for each $\xi$. Its matrix entries are polynomials of degree at most two in $\xi$, with fixed coefficients depending on $k_0$. On the bounded Fourier support the matrix exponential and every fixed number of its derivatives are bounded for $|z|\leq h$. Inverse Fourier transformation therefore supplies smooth fields, with every transverse derivative and every $z$ derivative in $L^2(\Omega_h)$. In particular their weighted field norm is finite. The normal definitions and the tangential evolution are exactly all six curl equations, so $\operatorname{curl}e=ik_0h$ and $\operatorname{curl}h=-ik_0e$. This also proves the divergence conditions. At the measurement plane, $S_z=(c_0/2)\operatorname{Re}(e\times\overline h)\cdot\hat z =(c_0C^2/2)|f_a|^2=|f_a|^2$. Orthogonality and \eqref{ca:packettail} give all four assertions in \eqref{ca:packetflux}. They hold for every vector, not merely for the basis fields or a sample of random inputs. Nonzero $f_a$ also proves linear independence of the fields. \end{proof} \begin{proposition}[The flux normalization permits unbounded field resource] \label{ca:packetresource} For the fields in Theorem~\ref{ca:localpackets}, set $q_{\min}=Q+2b$. Their local slab resource satisfies \begin{equation} \int_{\Omega_h}(|e_a|^2+|h_a|^2)\,dx\,dy\,dz \geq\frac{4h}{c_0k_0^2}q_{\min}^2\|a\|^2. \label{ca:packetresourceeq} \end{equation} Thus the same coherent flux and bypass bounds coexist with arbitrarily large resource at fixed $N,D,h,k_0,\delta$ in this local class. \end{proposition} \begin{proof} The vacuum Maxwell equations imply $(\Delta+k_0^2)h_z=0$. For each Fourier frequency in the support, put $\kappa(\xi)=\sqrt{|\xi|^2-k_0^2}>0$. The component has the form $u(z)=u(0)\cosh(\kappa z)+u'(0)\sinh(\kappa z)/\kappa$. When $|u|^2$ is integrated on $(-h,h)$ its cross term is odd, and both remaining terms are nonnegative. Hence $\int_{-h}^h|u(z)|^2dz\geq2h|u(0)|^2$. Plancherel and \eqref{ca:cauchydata} yield \begin{equation} \int_{\Omega_h}|h_z|^2\geq \frac{2hC^2}{k_0^2}\|\partial_xf_a\|_2^2 \geq\frac{4h}{c_0k_0^2}q_{\min}^2\|a\|^2, \label{ca:packetresproof} \end{equation} which is bounded above by the total resource. \end{proof} \subsection{Why the local theorem is not the requested counterexample} \begin{proposition}[Failure of a vacuum outgoing completion] \label{ca:outgoingfailure} No nonzero field of Theorem~\ref{ca:localpackets} can be continued into an entire homogeneous vacuum upper half-space with the outgoing condition that every evanescent Fourier component decays as $z\to+\infty$. \end{proposition} \begin{proof} The transverse Fourier support is preserved by vacuum evolution and contains only $|\xi|>k_0$, with a positive distance from the light circle. An outgoing continuation therefore has exponentially decaying $L^2$ electric and magnetic slices as $z\to+\infty$, implying $P_\Pi(z)\to0$ by Cauchy--Schwarz. On the other hand, the full-plane integrated flux of a source-free vacuum solution with these $L^2$ slices is independent of $z$. One way to justify this is to integrate the divergence-free real Poynting vector over a transverse cutoff of radius $R$. Its lateral error is bounded by $C/R$ times the $L^1$ norm of the transverse Poynting field integrated over the finite $z$ interval. That norm is finite by Cauchy--Schwarz, so the error tends to zero. Conservation would give $P_\Pi(z)=P_\Pi(0)=\|a\|^2>0$, a contradiction. \end{proof} The example therefore cannot acquire a physical incoming-power normalization by simply declaring its coefficient space to be an incident channel space. Formally appending $T=I_N$ would satisfy throughput and, for $\eta\leq1-\delta$, passage, but it would not construct the global field-to-port map required in Assumption~\ref{pr:scattering}. \begin{proposition}[No exact realization of these entire slab data by a compact exterior device] \label{ca:compactcompletion} Suppose a proposed dielectric perturbation is compactly supported outside $\Omega_h$ and the exterior convention consists of propagating incoming radiation and outgoing scattered radiation. It cannot have a nonzero field of Theorem~\ref{ca:localpackets} as its exact total field on all of $\Omega_h$. \end{proposition} \begin{proof} The Fourier Cauchy construction defines a homogeneous Maxwell solution $F^0=(e^0,h^0)$ on all of $\mathbb R^3$: for every finite $z$ its bounded Fourier support makes the matrix exponential well defined. Let $F$ be the proposed scattering field. The difference $F-F^0$ is a homogeneous vacuum Maxwell field in the unbounded connected component of the complement of the material support, and is zero on the entire open slab. Each component satisfies the constant-coefficient Helmholtz equation, so analytic continuation implies equality throughout that connected component, in particular everywhere outside a sufficiently large ball. Choose $z_0$ above that ball. The entire half-space $z>z_0$ is vacuum, and $F=F^0$ there. Its transverse spectrum is the same bounded set $|\xi|>k_0$ as before. In a homogeneous exterior half-space the stated radiation convention permits incoming propagating components, outgoing propagating components, and evanescent components decaying toward infinity; it excludes growing evanescent components. This follows as well by transverse Fourier transformation of the outgoing vacuum Green function and of the incident propagating angular spectrum: the longitudinal root for $|\xi|>k_0$ has positive imaginary part. There are no propagating components in the present support. The field therefore decays toward $+\infty$ in $L^2$ slices, contradicting its conserved flux $\|a\|^2>0$, exactly as in Proposition~\ref{ca:outgoingfailure}. \end{proof} Thus even the exact local packet data are not a ready-made physical counterexample. A resonant construction would need a quantitatively controlled \emph{approximation} to suitable concentrated data, or different accessible fields, with input power and every coherent error tracked. Proposition~\ref{ca:compactcompletion} does not exclude that possibility, nor does it exclude arbitrary external resonant matching. \subsection{Matching theory and the global realization problem} The reflectionless-scattering analysis of Stone and coauthors~\cite{stone2021} gives a Maxwell formulation of reflection zeros and structural tuning to a real-frequency reflectionless state. Its multichannel state is a selected coherent input vector. Such a state does not assert an $N$-dimensional subspace with uniformly high transmission at the same frequency, nor does it certify passage and absolute bypass through a fixed small section. Applying that framework therefore still requires the simultaneous synthesis and error control identified in Remark~\ref{cx:missing-bridge}. \begin{remark}[Global bridge required beyond these local packets] \label{ca:decisivebridge} The local analysis identifies two substantive global tasks. A proof must restrict the radiatively accessible trace subspaces of all admitted bounded-index devices strongly enough to give the uniform finite-rank estimate of Lemma~\ref{ch:ranklemma}, despite arbitrary external support and feedback. A disproof must realize an unbounded sequence of positive-flux trace subspaces with power-normalized incident and delivered channels and controlled absolute bypass. Theorem \ref{ca:localpackets} shows that local Maxwell equations and even a whole-plane coherent bypass certificate cannot replace that global step. Proposition~\ref{ca:powerbudget} only closes the empty parameter regime $\eta\tau>1+\delta$; it does not resolve the meaningful positive-throughput regime. These packet arguments do not give a physical counterexample. The subsequent edge-sphere construction in Theorem~\ref{uc:counterexample} does supply one for $\delta>\eta$, without completing these exact packets. \end{remark} \begin{remark}[Formalization boundary] \label{ca:leanboundary} Finite-dimensional Loewner consequences, norm perturbation certificates, and dimension lemmas can be formalized independently. Such Lean statements do not formalize the existence of a Maxwell scattering realization. The local packet argument requires Fourier analysis, the stated Maxwell evolution, and flux integration; finite-dimensional certificates alone cannot supply a global realization. The later edge-sphere construction in Theorem~\ref{uc:counterexample} supplies that realization through its actual material volume equation and whole-plane error estimates. Section~\ref{ver:lean} identifies its complete Lean existence endpoint, including the physical incoming Hilbert space and the same-device passivity identity. The global bridge is proved within this chain, rather than introduced as a mathematical axiom. \end{remark}