\section{Edge circulation and its physical completion} \label{ct:section} The absolute-bypass condition permits a qualitatively different mechanism when $\delta>\eta$: positive circulation inside a sharp section edge may be accompanied by an equal negative return immediately outside it. This section proves an exact electrostatic cross-flux calculation and a finite-dimensional Maxwell jet-interpolation result for that mechanism. It then states the additional global estimate needed to make the mechanism an admissible counterexample. Theorems~\ref{cl:remainder} and~\ref{rg:physical-limit} discharge that estimate in the following sections, leading to Theorem~\ref{uc:counterexample}. In particular, no impressed dipole is included in the admissible material class and no electrostatic field is presented as an exact solution at the fixed nonzero frequency. The proposed mechanism exploits the difference between signed passage and absolute bypass. A small inclusion close to the edge creates a local pattern with forward flux on one side and return flux on the other. By assigning independent local field data to different edge points, one can make this pattern for every coherent input in a finite-dimensional space. That is a local calculation: the device must still generate those data and keep the error small on the entire plane. The next two sections supply those global ingredients for the particular family used in the final theorem. \subsection{Exact local profile and physical coherent jets} Use the physical fields, the convention $e^{-i\omega t}$, and $Z_0=\sqrt{\mu_0/\eps_0}$. At one straight edge introduce orthonormal coordinates $(x,y,z)$ with the section locally $x<0$, outward normal $\widehat x$, tangent $\widehat y$, and $\widehat x\times\widehat y=\widehat z$. Fix dimensionless numbers $0<\rho0. \end{equation} \begin{lemma}[Electrostatic edge cross-flux] \label{ct:profile} For $s>0$, put a dielectric sphere of radius $\rho s$ at $(0,0,hs)$. Its exact electrostatic exterior response to a uniform normal electric field $A e\widehat z$, where $A>0$ and $e\in\mathbb C$, has dipole moment $\eps_0\alpha s^3 A e\widehat z$. On the plane $z=0$ its scattered tangential electric field is \begin{equation}\label{ct:dipoletrace} E^{\rm stat}_t(s\xi,s\upsilon,0) =-\frac{3\alpha h A e}{4\pi} \frac{(\xi,\upsilon)}{(\xi^2+\upsilon^2+h^2)^{5/2}}. \end{equation} Cross this field with a constant tangential magnetic jet $A b\widehat y$, $b\in\mathbb C$. The resulting real cross-flux profile is exactly \begin{equation}\label{ct:profileformula} f_s(s\xi,s\upsilon) :=\tfrac12\operatorname{Re} (E^{\rm stat}_t\times A\overline b\widehat y)\cdot\widehat z =-\frac{3\alpha h A^2}{8\pi}\operatorname{Re}(e\overline b) \frac{\xi}{(\xi^2+\upsilon^2+h^2)^{5/2}}. \end{equation} If $q=\operatorname{Re}(e\overline b)\geq0$, then \begin{equation}\label{ct:profileintegrals} \int_{x<0}f_s\,dx\,dy =\int_{x>0}|f_s|\,dx\,dy =\frac{s^2 A^2\alpha q}{4\pi h},\qquad \int_{\mathbb R^2}f_s\,dx\,dy=0. \end{equation} The profile is nonnegative inside the edge and nonpositive outside. These are exact statements about the stated electrostatic cross-term. \end{lemma} \begin{proof} Center coordinates at the sphere. Write the exterior potential as $-Ae z+cAe(\rho s)^3z/r^3$ and the interior potential as $-dAe z$. Continuity of potential and normal electric displacement gives $d=1-c$ and $\eps_p(1-c)=1+2c$. Therefore $c=(\eps_p-1)/(\eps_p+2)$, proving the polarizability. The dipole field is $[3(p\cdot\widehat r)\widehat r-p]/(4\pi\eps_0 r^3)$. At the observation plane the center-relative vector is $s(\xi,\upsilon,-h)$, which gives \eqref{ct:dipoletrace}. Taking its cross product gives \eqref{ct:profileformula}. Finally, \[ \int_{\mathbb R}\frac{d\upsilon} { (\xi^2+\upsilon^2+h^2)^{5/2}} =\frac{4}{3(\xi^2+h^2)^2},\qquad \int_{-\infty}^0\frac{-\xi\,d\xi}{(\xi^2+h^2)^2} =\frac1{2h^2}. \] Together with $dx\,dy=s^2d\xi\,d\upsilon$ these prove the half-plane integrals. Oddness in $\xi$ proves the last identity. \end{proof} An electric-wall vacuum Maxwell field satisfies \eqref{cd:electric-wall}; in particular $E_t=H_z=0$ on the plane. Its normal Poynting density is identically zero there, including for every coherent combination of fields in that sector. We next prescribe the nonzero jets $E_z,H_x,H_y$ while retaining this exact cancellation. \begin{lemma}[Finite even-TE jet interpolation] \label{ct:jets} Let $p_1,\ldots,p_N$ be nonzero points of $z=0$, with $p_i\ne p_j$ and $p_i\ne-p_j$ for $i\ne j$. Arbitrary complex values of $(E_z,H_x,H_y)$ at these $N$ points can be realized by a finite linear combination of regular vacuum TE spherical waves with even angular order $l\geq2$ and even azimuthal index $m$. The field is electric-wall and invariant, with the vectors rotated, under a rotation by $\pi$ about the $z$ axis. In particular \begin{equation}\label{ct:partnerjets} E_z(-p_j)=E_z(p_j),\qquad H_t(-p_j)=-H_t(p_j). \end{equation} The interpolating regular waves have finite physical incoming power. No bound uniform in the interpolation points or target jets is asserted. \end{lemma} \begin{proof} Consider a scalar regular Helmholtz solution $u$ that is even in $z$ and invariant under $(x,y)\mapsto(-x,-y)$, and set \begin{equation}\label{ct:debye} E=\nabla\times(xu)=\nabla u\times x, \qquad H=\frac{\nabla\times E}{i\omega\mu_0}. \end{equation} Since $\nabla\cdot E=0$ and $(\Delta+k_0^2)E=0$, these fields satisfy both vacuum Maxwell equations. They have precisely the electric-wall and rotation symmetries just stated. Their scalar spherical-harmonic expansion contains only even $m$ and even $l$; the $l=0$ term gives zero in \eqref{ct:debye}. Thus finite truncations are regular TE fields of the required kind. The full tangential scalar two-jets can first be prescribed at $p_1,\ldots,p_N$. To see this without assuming a Runge theorem, write $r_t=(x,y)$ and use the entire scalar solutions \[ u_\xi(r_t,z)=\cos(\xi\cdot r_t) \cos\bigl(\sqrt{k_0^2-|\xi|^2}\,z\bigr), \qquad |\xi|0$. Formula \eqref{ct:debye} gives \begin{equation}\label{ct:jetmap} E_z=-r u_y,\qquad H_x=\frac{2u_x+r u_{xx}+k_0^2r u}{i\omega\mu_0},\qquad H_y=\frac{2u_y+r u_{xy}}{i\omega\mu_0}. \end{equation} Choose $u_y$ for $E_z$, then $u_{xy}$ for $H_y$, and $u_{xx}$ for $H_x$. This gives a surjective map at each point, hence at all points simultaneously. Each $u_\xi$ is a finite sum of plane waves with real wavevectors of length $k_0$. Replace its point-supported angular measures by smooth, symmetrized densities on the sphere; their fields and any fixed finite set of derivatives converge on compact sets. Thus the jet map of smooth symmetric scalar Herglotz fields is onto: choose finitely many approximately interpolating columns close enough to an invertible interpolation matrix. For each of these smooth fields, truncate the spherical-harmonic expansion, preserving both symmetries. Compact derivative convergence again preserves the invertibility of that finite matrix. Invert it to obtain exact jets using one finite angular band. Its Maxwell fields are finite regular TE sums; their incoming spherical amplitudes have finite power. The parity identities give \eqref{ct:partnerjets}. \end{proof} \begin{lemma}[Balancing the physical input Gram matrix] \label{ct:gram} For any finite collection of target jets covered by Lemma~\ref{ct:jets}, there are interpolants $v_1,\ldots,v_N$ in a common finite even-TE band whose physical incoming Gram matrix is $\lambda I_N$ for some $0<\lambda<\infty$. Consequently $g_j=v_j/\sqrt\lambda$ are orthonormal physical incoming data. All $v_j$ retain their prescribed jets. \end{lemma} \begin{proof} Start with arbitrary interpolants $u_j$ and their positive semidefinite incoming Gram matrix $G$. The space of finite even-TE regular waves has unbounded dimension, whereas the jet map has finite rank. Enlarge its finite band until its jet kernel contains an $N$-dimensional subspace orthogonal, in physical incoming power, to the span of the $u_j$. Choose $\lambda>\|G\|$. In an orthonormal basis of that kernel subspace, a matrix square root of $\lambda I-G$ supplies vectors $w_j$ with Gram matrix $\lambda I-G$. They have zero prescribed jets and are orthogonal to all $u_j$. Hence $v_j=u_j+w_j$ has the same jets as $u_j$ and Gram matrix $G+(\lambda I-G)=\lambda I$. \end{proof} \subsection{Finite-rank diagonal edge circulation} Choose distinct points $p_j=(L/2,y_j,0)$ on the interior of the right edge of the square, with $-L/20$, Theorem~\ref{cd:wall-isolation} can amplify every even TE order in this finite band by the same factor $A\sqrt\lambda$, up to a common phase and arbitrarily small compact derivative errors. Applied to the orthonormal inputs $g_j=v_j/\sqrt\lambda$, its ideal core jets at $p_j$ are \begin{equation}\label{ct:amplifiedjets} E_z(c;p_j)=\frac{A\sqrt{Z_0}}{\ell_0}c_j,\qquad H_t(c;p_j)=\frac{A}{\sqrt{Z_0}\ell_0}c_j\widehat y, \qquad c\in\mathbb C^N. \end{equation} The original radial-device background has zero plane flux exactly, at every finite stage, by reflection symmetry. Compact derivative approximation concerns its illumination only, before adding spheres. For $a>0$ set $s=a\ell_0$ and $A=t/a$, with fixed $t>0$. At each of $p_j$ and $-p_j$, put the electrostatic cross-profile of Lemma~\ref{ct:profile}, with the local outward/tangent directions $(\widehat x,\widehat y)$ on the right and $(-\widehat x,-\widehat y)$ on the left. Use the jets \eqref{ct:amplifiedjets}, and denote the sum of these $2N$ profiles by $F_a(c;x,y)$. Thus $F_a$ is defined on the whole plane by explicit functions; it is quadratic in $c$ and diagonal in the coefficients $|c_j|^2$. \begin{proposition}[Coherent local-model passage and absolute bypass] \label{ct:model} For each fixed finite $N$, the profiles just defined obey \begin{equation}\label{ct:modellimits} \sup_{\|c\|=1}\left|\int_D F_a(c)\,dA-p_0\right|\longrightarrow0, \qquad \sup_{\|c\|=1}\left|\int_{\mathbb R^2\setminus D}|F_a(c)|\,dA-p_0\right| \longrightarrow0, \qquad p_0=\frac{\alpha t^2}{2\pi h}. \end{equation} Also $\int_{\mathbb R^2}F_a(c)\,dA=0$ exactly. The geometric spheres associated with these profiles have radius $\rho a\ell_0$, lie strictly above the plane, are disjoint for sufficiently small $a$, and preserve the rotation by $\pi$ symmetry. Their contrasts remain in the fixed ordinary dielectric interval. \end{proposition} \begin{proof} For one point, the product in Lemma~\ref{ct:profile} is $q=|c_j|^2/\ell_0^2$, so its positive half-plane mass equals $\alpha t^2|c_j|^2/(4\pi h)$. Its partner has the same mass: both the local tangent and $H_t$ reverse at $-p_j$, leaving the scalar product and the orientation of positive interior flux unchanged. Summing gives $p_0\sum_j|c_j|^2$. Choose disjoint fixed disks about the $2N$ edge points, avoiding corners. Within each disk the square is exactly its interior half-plane. The rescaled profile has an integrable envelope bounded by a constant times $(1+\xi^2+\upsilon^2)^{-2}$, so its $L^1$ mass outside its disk tends to zero. More precisely its tail outside scaled radius $R$ is $O(R^{-2})$. On each disk retain only its own profile. The $L^1$ difference between this truncated sum and $F_a$ tends to zero, uniformly for $\sum_j|c_j|^2=1$, since there are finitely many centers. The retained profiles have the required signs on the two sides. The inequality $\bigl||u|-|v|\bigr|\leq|u-v|$ proves both limits, including the absolute-value limit for arbitrary coherent $c$. Oddness gives zero total signed integral for every individual profile. Finally $h>\rho$ leaves a vacuum gap above the plane, and the fixed distinct centers ensure disjointness when $a$ is small. \end{proof} \begin{remark}[Why conditioning of the interpolation is not a bypass loss] \label{ct:conditioning} The number $\lambda$ in Lemma~\ref{ct:gram} can be large and depends on $N$ and the points. It is fixed before $a\downarrow0$ and is absorbed into the common core gain $A\sqrt\lambda$. The incoming data remain orthonormal. Thus the model's passage and bypass matrices are both $p_0 I_N$ in physical input coordinates, rather than matrices whose ratio deteriorates with an interpolation condition number. This is a finite-dimensional statement and does not supply a bound on the perturbed Maxwell Green operator. \end{remark} \subsection{The precise physical estimate still required} For each $a$, the finite-band control theorem permits choosing a finite radial background with vacuum core large enough that its unperturbed fields approximate \eqref{ct:amplifiedjets} and any prescribed finite number of derivatives as accurately as desired on a fixed neighborhood of all the edge points. The target gain $t\sqrt\lambda/a$ is finite for each $a$. This diagonal choice is legitimate for illumination. Add the $2N$ actual dielectric spheres to that background, and let $S_{z,a}(c)$ denote the plane flux of the resulting exact total scattering field for the same orthonormal incoming inputs. A background approximation theorem alone does not estimate this new field. \begin{proposition}[Conditional global completion estimate] \label{ct:completion} Fix $0<\tau<1$ and $0<\eta<\delta$. Choose $t$ so that the constant $p_0$ of \eqref{ct:modellimits} lies strictly between $\eta$ and $\delta$. Suppose that for each fixed $N$ one can choose the finite radial backgrounds and sphere sizes above so that the actual physical scattering fields have legitimate integrable plane fluxes and satisfy \begin{equation}\label{ct:missingestimate} \sup_{\|c\|=1} \|S_{z,a}(c)-F_a(c)\|_{L^1(\mathbb R^2)}\longrightarrow0. \end{equation} Take the selected output map to be the full unitary scattering map of each finite lossless dielectric. Then for sufficiently small $a$ the span of those $N$ physical incoming data is counted under Definition~\ref{pr:count}, with the same fixed $\tau,\eta,\delta$. Consequently \emph{if} \eqref{ct:missingestimate} can be established for every finite $N$, the unrestricted finiteness target is false for this one admissible fixed threshold tuple. \end{proposition} \begin{proof} Orthonormality gives input power $\|c\|^2$; unitarity gives $\|Tc\|^2=\|c\|^2\geq\tau\|c\|^2$. Equations \eqref{ct:modellimits}--\eqref{ct:missingestimate} imply, uniformly for unit $c$, that actual passage tends to $p_0$ and actual absolute bypass tends to $p_0$. Their strict margins from $\eta$ and $\delta$ prove the other two counting inequalities. Homogeneity covers all $c$. For each member the sphere-to-plane gap is positive, so it supplies a source-free vacuum neighborhood of the section; its shrinking size is not a forbidden resource. \end{proof} \begin{remark}[The bridge discharged in the following sections] \label{ct:gap} The exact isolated-sphere electrostatic polarizability and finite-band illumination control alone do not imply \eqref{ct:missingestimate}. Three issues must be controlled together: the small-inclusion expansion in a background whose resonant amplification grows like $a^{-1}$, the source Green response of unselected angular orders, and the absolute flux of all correction fields across the entire plane, including the remote layers. A finite inverse for each separate device would be insufficient unless its dependence on that varying background is controlled. The sphere volume is $O(a^3)$ and the amplified jet is $O(a^{-1})$. Theorem~\ref{cl:remainder} turns the resulting scale $a^3a^{-2}=a$ into an actual bounded-inverse and whole-plane remainder estimate for an explicit parity reaction kernel. Theorem~\ref{rg:physical-limit} identifies that kernel as the source response of distant finite common stacks in the limit, controlling all angular orders and the remote absolute-flux tails. Its inverse stability step transfers the estimate to actual perturbed devices. These two additional proofs, rather than incident gain or signed power conservation alone, discharge \eqref{ct:missingestimate}. The final fixed-threshold construction is Theorem~\ref{uc:counterexample}. \end{remark}