\section{Finite three-dimensional driven hotspot packing} \label{hs:three-section} This section constructs finite inclusions with conical \emph{point} tips. Small positive contrast supplies a convenient uniform scattering inverse; it does not remove the point singularity. The material bounds and frequency remain fixed. The modal question is separated at the end. The two-dimensional construction suggests a mechanism, but truncating its invariant direction would introduce new boundaries and radiation. Here each sharp reference point instead has two opposite dielectric cones meeting at their tips. The resulting high-field region is localized in all three spatial directions. We again first prove excitation of the sharp points by one physical illumination and then round the tips. \begin{theorem}[Finite three-dimensional driven point localization] \label{hs:three-packing} Fix $k>0$ and an upper bound $\varepsilon_+>1$. There is a number $\varepsilon_h\in(1,\varepsilon_+]$ that can be kept fixed in the following construction. Given any nonempty open patch $\mathcal P$ in a plane, any finite $N$, and any $\ell,M>0$, there exist a compactly supported binary dielectric $\varepsilon_r\in\{1,\varepsilon_h\}$ with $2N$ disjoint smooth inclusions and one unit-incoming-power Maxwell illumination at frequency $k$ such that there are $N$ pairwise disjoint three-dimensional balls $Q_j$, of diameter less than $\ell$, satisfying \begin{equation} \max_{\overline Q_j,\,\text{one-sided}}|E|>M, \qquad \sup_{\text{a collar of }\partial Q_j,\,\text{one-sided}}|E| <\tfrac12\max_{\overline Q_j,\,\text{one-sided}}|E|. \label{hs:three-peak} \end{equation} The centers of the balls lie in $\mathcal P$, and the orthogonal projections of the balls and of the material support are contained in $\mathcal P$. In particular the selected superlevel components are localized in all three spatial dimensions. The maximizing points after rounding need not lie exactly in the original plane. There is no invariant infinite direction, no internal impressed source, and no conducting boundary in the physical device. \end{theorem} The prescribed patch is therefore a projection constraint, while the localized neighborhoods themselves are three-dimensional. In particular, the theorem concerns finite point neighborhoods rather than invariant ridges. As before, no minimum energy fraction in an individual peak is required. The proof follows the same logic as the cross-sectional case, but each step is a three-dimensional Maxwell statement. A spherical angular problem supplies a conical exponent below one; a coefficient functional detects whether the corresponding electric singularity is present. A scattering inverse and total-field approximation provide an illumination with a nonzero coefficient at every chosen point. Finally, convergence under rounding retains high values inside each ball, and bounds on its collar separate the resulting maximum from the boundary. The small contrast is selected once within the prescribed material bound; the number and size of the peaks may then be chosen as stated in the theorem. \subsection{A genuine conical point exponent} \label{hs:three-angular-subsection} The two caps in \eqref{hs:three-caps} describe the directions occupied by the opposing cones. The angular eigenfunction determines the shape of a potential $r^\lambda\phi$ near their common tip. Because $0<\lambda<1$, its gradient is unbounded at the tip but has finite local energy. This electrostatic calculation identifies the candidate singularity; the following subsection is needed to establish its role in the finite-frequency Maxwell field. \begin{lemma}[Strict positive contrast gives a conical exponent below one] \label{hs:three-angular} Choose $t\in(1/\sqrt3,1)$ and set, on the unit sphere, \begin{equation} q(\widehat x)=1+d\,\mathbf1_{\{|\widehat x\cdot e_3|>t\}}, \qquad d>0. \label{hs:three-caps} \end{equation} There are a real $\phi\in H^1(S^2)$ and $\lambda\in(0,1)$ satisfying \begin{equation} -\operatorname{div}_{S^2}(q\nabla_{S^2}\phi) =\nu q\phi,\qquad \nu=\lambda(\lambda+1),\qquad \int_{S^2}q\phi^2=1,\quad \int_{S^2}q\phi=0. \label{hs:three-angular-eigenproblem} \end{equation} The electrostatic field $E_0=\nabla(r^\lambda\phi)$ is curl free, satisfies $\operatorname{div}(qE_0)=0$ weakly in a ball, and obeys \begin{equation} \int_{B_R}q|E_0|^2\,dx=\lambda R^{2\lambda+1}. \label{hs:three-static-energy} \end{equation} \end{lemma} \begin{proof} The weighted Rayleigh quotient on the subspace orthogonal to constants has a positive minimum by the compact embedding $H^1(S^2)\hookrightarrow L^2(S^2)$ and the sphere Poincare inequality. Use the trial function $z=\widehat x\cdot e_3$, whose weighted mean vanishes by reflection symmetry. Since $|\nabla_{S^2}z|^2=1-z^2$, \begin{equation} \int_{S^2}q(|\nabla_{S^2}z|^2-2z^2) =d\int_{|z|>t}(1-3z^2)<0. \label{hs:three-trial} \end{equation} Explicitly, the trial quotient is \begin{equation} \mathcal Q(z)=\frac{2+d(2-3t+t^3)}{1+d(1-t^3)},\qquad \mathcal Q(z)-2=-\frac{3dt(1-t^2)}{1+d(1-t^3)}<0. \label{hs:three-trial-explicit} \end{equation} Thus the first positive eigenvalue $\nu$ is less than two; its positive indicial root satisfies $0<\lambda<1$. Separation in spherical coordinates proves the weak electrostatic equation. The interface conditions are exactly those in the angular weak problem. The flux through a small sphere is of order $r^{\lambda+1}$, so there is no point source at the origin. Finally $\int q|\nabla_{S^2}\phi|^2=\nu$ gives $(\lambda^2+\nu)\int_0^Rr^{2\lambda}\,dr =\lambda R^{2\lambda+1}$. \end{proof} \subsection{An exact Maxwell coefficient functional at the conical point} \label{hs:three-functional-subsection} At positive frequency the Maxwell field is not simply the electrostatic field just described. The decomposition below separates a gradient part, which can contain the conical singularity, from a more regular remainder. Projecting the gradient part onto the angular eigenfunction then defines a coefficient that can be tested for the actual total field. This is the bridge from the local model to a physical scattering solution. We use one standard regularity theorem in a precise local form. On a ball there is a bounded linear regularized Poincare lifting \begin{equation} \mathcal R:H^1(B)^3\cap\ker\operatorname{div}\longrightarrow H^2(B)^3, \qquad \operatorname{curl}\mathcal R f=f. \label{hs:three-lifting} \end{equation} This is the regularized curl-potential operator of Costabel and McIntosh~\cite{costabelMcIntosh2010}; their homotopy identity and order-minus-one mapping theorem apply to a ball. Only this established local Sobolev mapping is imported. It is not a scattering, corner-excitation, or rounding theorem. Use normalized units $\operatorname{curl}E=ikH$, $\operatorname{curl}H=-ikqE$ in the local arguments; the magnetic field here is the physical magnetic field multiplied by $Z_0$. \begin{lemma}[Local regular decomposition with controlled remainder] \label{hs:three-decomposition} Suppose $E,H$ solve the source-free Maxwell system with $1\leq q\leq\varepsilon_h$ on $B_{3R}$ and have finite local energy. On $B_R$ one can write \begin{equation} E=\nabla v+W,\qquad v\in H^1(B_R),\quad W\in H^2(B_R)^3, \quad \|v\|_{H^1}+\|W\|_{H^2}\leq C\|E\|_{L^2(B_{3R})}, \label{hs:three-decomposition-bound} \end{equation} after fixing the additive constant of $v$. The maps can be chosen linear. The constant depends on the displayed radii, frequency and material bounds, not on derivatives of $q$. \end{lemma} \begin{proof} Testing $\operatorname{curl}\operatorname{curl}E=k^2qE$ against $\chi^2E$, for a compactly supported cutoff, and applying Young's inequality gives $\|\operatorname{curl}E\|_{L^2(B_{5R/2})} \leq C\|E\|_{L^2(B_{3R})}$. Since $\operatorname{div}H=0$ and $\operatorname{curl}H=-ikqE$, the Fourier curl/div identity applied to a cutoff of $H$ gives $\|H\|_{H^1(B_{2R})}\leq C\|E\|_{L^2(B_{3R})}$. This is the same constant-permeability interior mechanism as Lemma~\ref{ch:magneticHone}. Apply \eqref{hs:three-lifting} on $B_{2R}$ to $\operatorname{curl}E=ikH$, which is divergence free, and restrict the resulting $W$ to $B_R$. The curl-free difference $E-W$ is a gradient on the simply connected ball. Its potential with zero mean satisfies the stated bound by the ordinary Poincare inequality. All operations are bounded and linear. \end{proof} The local decomposition separates the singular potential from a bounded remainder. The coefficient below is defined by an angular projection; its independence of the chosen lifting makes it a property of the field, rather than of the decomposition. \begin{lemma}[Intrinsic conical singular coefficient] \label{hs:three-coefficient} In the conical material of Lemma~\ref{hs:three-angular}, every local Maxwell solution as above has a linear coefficient $c(E)$, continuous for the $L^2(B_{3R})$ topology on its solution space. If $c(E)\ne0$, then $|E|$ is unbounded at ordinary points approaching the conical vertex from the homogeneous regions. Its local energy remains finite. \end{lemma} \begin{proof} Use $E=\nabla v+W$ from Lemma~\ref{hs:three-decomposition}. Since $H^2(B_R)\hookrightarrow C^{0,\gamma}(\overline B_R)$ for $0<\gamma<1/2$, $W$ is bounded. Define \begin{equation} U(r)=\int_{S^2}qv(r,\widehat x)\phi,\qquad A(r)=\int_{S^2}qW_r(r,\widehat x)\phi,\qquad B(r)=\int_{S^2}qW_t(r,\widehat x)\cdot\nabla_{S^2}\phi. \label{hs:three-projections} \end{equation} The subscripts denote radial and tangential components. Projecting $\operatorname{div}(q(\nabla v+W))=0$ gives \begin{equation} (r^2U')'-\lambda(\lambda+1)U=-(r^2A)'+rB. \label{hs:three-radial} \end{equation} The exact solution with finite $H^1$ energy is \begin{align} U(r)&=c(E)r^\lambda+p(r),\notag\\ p(r)&=\frac{1}{2\lambda+1}\left[ r^\lambda\int_0^rs^{-\lambda}\{B(s)-(\lambda+1)A(s)\}\,ds \right.\notag\\[-2pt] &\hspace{36mm}\left. -r^{-\lambda-1}\int_0^rs^{\lambda+1}\{B(s)+\lambda A(s)\}\,ds \right]. \label{hs:three-exact} \end{align} To verify this formula, variation of constants for the two homogeneous solutions $r^\lambda,r^{-\lambda-1}$ first gives integrals of $-A'-2A/r+B/r$. Integrating the $A'$ terms by parts cancels their endpoint contributions and gives \eqref{hs:three-exact}. The integrals converge because $A,B$ are bounded and $\lambda<1$. They show \begin{equation} |p(r)|\leq Cr\|W\|_\infty,\qquad |p'(r)|\leq C\|W\|_\infty. \label{hs:three-particular-bound} \end{equation} For the second inequality, differentiation produces the additional term $-A(r)$ and two bounded integral terms. Finite $H^1$ energy excludes the other homogeneous term $r^{-\lambda-1}$. Average $r^{-\lambda}(U(r)-p(r))$ on $R/20$ there are local source-free Maxwell solutions on a sufficiently small ball with $c(E)\ne0$. \end{lemma} \begin{proof} Scale a physical ball to $B_1$ and write $\kappa=kR$. Let $E_0=\nabla(r^\lambda\phi)$ and consider \begin{equation} X_q=\{w\in H_0(\operatorname{curl},B_1): \operatorname{div}(qw)=0\}. \label{hs:three-local-space} \end{equation} There is a finite constant $C_q$ with $\|w\|_2\leq C_q\|\operatorname{curl}w\|_2$ on $X_q$. Here is a reduction to the ordinary ball estimate, which avoids an unproved smoothness assumption on $q$. Decompose $w=w_T+\nabla p$ with $p\in H_0^1$ and $\operatorname{div}w_T=0$, $w_T\in H_0(\operatorname{curl})$. Weighted divergence yields $\int q\nabla p\cdot\nabla\overline\psi =-\int(q-1)w_T\cdot\nabla\overline\psi$, so $\|\nabla p\|_2\leq d\|w_T\|_2$. For a ball the usual Maxwell estimate is $\|w_T\|_2\leq C\|\operatorname{curl}w_T\|_2$. For clarity, $H^1$ membership in this assertion is not assumed. The order-minus-one lifting also maps $L^2$ to $H^1$; lift $\operatorname{curl}w_T$ to $V\in H^1$. The difference $w_T-V=\nabla p_0$ has $\Delta p_0=-\operatorname{div}V$ and tangential boundary gradient $-V_t$. The latter is in $H^{1/2}(S^2)$, so the boundary trace of $p_0$, modulo a constant, is in $H^{3/2}(S^2)$. This follows by the scalar sphere Laplacian estimate applied to its tangential gradient. Dirichlet regularity on the smooth ball gives $p_0\in H^2$, hence $w_T\in H^1$. One can now use the Gaffney identity for tangential-zero fields, \begin{equation} \|\operatorname{curl}w_T\|_2^2+\|\operatorname{div}w_T\|_2^2 =\|\nabla w_T\|_2^2+2\|w_T\cdot n\|_{L^2(S^2)}^2, \label{hs:three-gaffney} \end{equation} initially for smooth fields and then by trace-preserving approximation on the smooth ball. Poincare's inequality with the tangential trace condition excludes constant vectors and bounds the $L^2$ norm. Thus $C_q\leq(1+d)C$ is sufficient. Solve on $X_q$ \begin{equation} \int\operatorname{curl}w\cdot\operatorname{curl}\overline z -\kappa^2q w\cdot\overline z =\kappa^2\int qE_0\cdot\overline z. \label{hs:three-local-variational} \end{equation} For $\kappa^2\varepsilon_h C_q^2<1/2$ this is coercive and gives $\|w\|_{H(\operatorname{curl})}\leq C\kappa^2\|E_0\|_2$. It is also the equation for every test in $H_0(\operatorname{curl})$: the weighted Helmholtz decomposition splits such a test into an $X_q$ part and a gradient from $H_0^1$, and both sides vanish on the gradient part. Hence $E=E_0+w$ satisfies $\operatorname{curl}\operatorname{curl}E=\kappa^2qE$ and $H=(i\kappa)^{-1}\operatorname{curl}E$ completes the exact Maxwell solution. On an interior ball, cutoff curl/div regularity gives \begin{equation} \|\operatorname{curl}w\|_{H^1(B_{1/2})} \leq C\left(\|\operatorname{curl}w\|_2 +\kappa^2\|q(E_0+w)\|_2\right)=O(\kappa^2). \label{hs:three-local-curl-small} \end{equation} Apply the lifting to $\operatorname{curl}w$. Its $W$ has $H^2$ norm $O(\kappa^2)$; the potential in $E=\nabla v+W$ differs from $r^\lambda\phi$, up to a constant, by $O(\kappa^2)$ in $H^1$. The averaged formula in Lemma~\ref{hs:three-coefficient} therefore gives $c(E)=1+O(\kappa^2)$. This proves nontriviality for small positive $\kappa$. The local tangential boundary data used to construct this solution are a mathematical matching device. The physical construction below uses global incoming waves and has no PEC boundary. \end{proof} \subsection{A uniform finite-device Maxwell inverse at fixed small contrast} \label{hs:three-volume-subsection} The following inverse controls the complete response of the finite dielectric, including multiple scattering among the inclusions. Small contrast makes the volume operator invertible uniformly over the family used for rounding. It is a device for controlling the global problem; the local conical exponent remains below one at every positive contrast. \begin{lemma}[Uniform volume-integral inverse] \label{hs:three-volume} Fix a ball $B_L$ and $k>0$. There is $C_{kL}<\infty$ such that the electric volume operator \begin{equation} \mathcal K_k f=(k^2I+\nabla\nabla)\Phi_k*f, \qquad \Phi_k(x)=\frac{e^{ik|x|}}{4\pi|x|}, \label{hs:three-volume-operator} \end{equation} restricted to $B_L$, with $f$ zero extended, has $L^2$ norm at most $C_{kL}$. For $0\max\{M,2B\}$. Each nonzero conical coefficient supplies a positive-volume superlevel $|E|>2S$ inside its observation ball. Local $L^2$ convergence preserves a positive-volume superlevel $|E_\rho|>S$ in every ball for all sufficiently small $\rho$. The continuous one-sided branches attain their joint maximum on the compact ball. Each maximum exceeds $S$ and twice the collar bound. The collar is therefore a strict low barrier for its half-amplitude superlevel set. Its full connected component is compact and stays strictly inside the observation ball, proving \eqref{hs:three-peak}. All choices of illumination precede rounding; the same normalized input is used for every rounded field. \end{proof} \begin{corollary}[Finite-region volume-to-peak ratios in three dimensions] \label{hs:three-regional-volume} Along the rounded sequence for one fixed sharp configuration and one normalized input, electromagnetic energy in a fixed bounded region stays bounded, while the electric maximum in every selected ball tends to infinity. Each resulting finite-region energy-to-peak ratio tends to zero. This is a driven local-energy statement, not an all-space modal normalization. \end{corollary} \begin{proof} The electric $L^2$ norms are uniformly bounded by the volume inverse. The magnetic fields are bounded on fixed bounded regions by the Maxwell equations, the local curl estimate, and exterior vacuum regularity; a slightly larger comparison ball may be used for the cutoffs. For every finite amplitude threshold, the sharp field has a positive-volume superlevel in every selected ball. Local $L^2$ convergence preserves a smaller superlevel for every sufficiently late rounded device. The attained one-sided maxima therefore diverge along the same family. The material factors are bounded above and below. \end{proof} \subsection{What an exact mode can mean in a finite dielectric device} \label{hs:three-modal-subsection} \begin{proposition}[Absence of real-frequency finite-energy modes] \label{hs:three-no-real-mode} Let the real positive scalar permittivity be constant on finitely many material pieces with piecewise smooth interfaces, and equal one outside a bounded set. At a fixed real $\omega>0$, a source-free Maxwell field with finite all-space electromagnetic energy is zero. The same conclusion holds for an outgoing source-free field, without a finite-energy assumption. \end{proposition} \begin{proof} In the exterior vacuum every Cartesian component satisfies the scalar Helmholtz equation. A spherical harmonic expansion of an $L^2$ exterior solution has radial terms proportional to incoming and outgoing spherical Hankel functions. Each nonzero radial combination has nondecaying mean radial $L^2$ mass, because its leading amplitude is a nonzero combination of $e^{ikr}$ and $e^{-ikr}$ divided by $r$. It cannot be square integrable on $(R,\infty)$ with volume weight $r^2$. Orthogonality of spherical harmonics excludes cancellation between different angular terms. Thus a finite-energy field is zero in the exterior. For an outgoing source-free field, the real part of the complex Poynting identity instead sets its nonnegative far-field radiation power to zero, giving the same exterior conclusion. Tangential electric and magnetic continuity then propagates zero across each open smooth material interface by the constant-coefficient extension argument used in Lemma~\ref{hs:three-runge}. Paths can avoid the finite lower-dimensional interface singular sets. Hence the field is zero everywhere. \end{proof} This proposition diagnoses an inappropriate literal interpretation of the modal numerator; it does \emph{not} refute the tutorial's high-$Q$ or quasinormal-mode interpretation. A finite radiating resonator has complex-frequency poles. Their outgoing fields grow at spatial infinity and require a specified regularized mode norm, or a justified high-$Q$ approximation, instead of the ordinary positive all-space energy integral. The tutorial explicitly notes open-cavity generalizations of its mode-volume definition. \begin{remark}[The driven proof and the separate modal construction] \label{hs:three-modal-gap} Theorem~\ref{hs:three-packing} establishes the finite-device \emph{driven} point-localization statement given above. It does not turn the chosen illumination into a resonator mode. A possible modal route starts with a simple isolated dielectric resonance, inserts finitely many small conical inclusions where its electric field is nonzero, and then rounds their tips. To complete that route one must prove that a single continued resonance has nonzero conical coefficients at every selected vertex, and that the chosen resonance normalization stays finite and nonzero under insertion and rounding. These are not consequences of Runge approximation: Runge varies external illumination, whereas a pole's eigenfield is fixed by its homogeneous outgoing eigenproblem. Strong local field convergence without a coefficient topology also does not prove nonzero singular coefficients. Sections~\ref{hs:modal-next-section} and~\ref{ms:section} establish these conditions for the particular modal family considered there. Eventual nonzero excitation is proved by a strict annular pairing in Theorem~\ref{hs:modal-coefficient-continuation}, without requiring continuation of the intrinsic coefficient through zero contrast; a simple pole, its genuine reciprocal normalization and its rounded continuation are proved in Lemmas~\ref{ms:sphere-simple}--\ref{ms:rounding}. The resulting closed modal statement is Theorem~\ref{ms:modal-packing}. It does not rely on varying a driven illumination or on a divergent all-space positive energy integral. \end{remark}