\section{Excitation of conical points by one genuine resonance} \label{hs:modal-next-section} This section proves a local bridge that the driven-field construction does not provide: a single analytically continued Maxwell resonance retains a nonzero electric singular coefficient at every member of a finite family of newly inserted conical points. The contrast of the insertions may depend on that finite family; the bounds $1\leq\varepsilon_r\leq\varepsilon_+$ do not change. A resonance here has a complex frequency and satisfies the analytically continued outgoing condition. It is not a real-frequency driven solution. The issue is not merely whether each cone \emph{can} support a singular field. A resonator mode is selected by the global source-free problem, so we cannot tune an independent illumination at each tip. We start with one mode that is nonzero at the proposed insertion points, add small positive-contrast cones, and show that the resulting \emph{same} continued mode excites each tip. This section proves the local excitation step; Section~\ref{ms:section} constructs and normalizes the global mode. \subsection{The angular branch through a homogeneous medium} The two spherical caps below encode opposite conical sectors. As their contrast tends to zero, the angular eigenfunction approaches a linear potential along the cone axis. Its electric field approaches a constant axial vector. At every positive contrast in the family the exponent is less than one, so a nonzero coefficient produces an unbounded field at the sharp tip. The limiting constant field tells us which component of the original mode can excite this singularity. Fix an axis $e\in S^2$, a number $t\in(1/\sqrt3,1)$, and put \begin{equation} a(\widehat x)=\mathbf1_{\{|\widehat x\cdot e|>t\}},\qquad q_\eta(\widehat x)=1+\eta a(\widehat x). \label{hs:modal-angular-material} \end{equation} The physical permittivity near a selected point is $\varepsilon_b q_\eta$, where $1<\varepsilon_b<\varepsilon_+$ is the permittivity of its initially homogeneous host. The contrast parameter is real and positive. We need a convergent real angular family; analytic continuation of its singular coefficient or of its angular eigenfunction is not required. \begin{lemma}[A positive-contrast angular family] \label{hs:modal-angular-branch} Fix $1/2\leq t<1$ and any sequence $\eta_j>0$ tending to zero. There are real, axisymmetric, odd angular eigenfunctions $\phi_j$ and numbers $0<\lambda_j<1$, $\nu_j=\lambda_j(\lambda_j+1)$, satisfying \begin{equation} -\operatorname{div}_{S^2}(q_{\eta_j}\nabla_{S^2}\phi_j) =\nu_j q_{\eta_j}\phi_j, \qquad \int_{S^2}q_{\eta_j}\phi_j\,dS=0. \label{hs:modal-angular-branch-equation} \end{equation} They may be normalized so that $\nu_j\to2$, $\lambda_j\to1$, $\phi_j(\widehat x)\to\widehat x\cdot e$, and, almost everywhere on the sphere, the corresponding angular electric vector $\lambda_j\phi_j\widehat x+\nabla_{S^2}\phi_j$ tends to $e$. The angular bounds needed for dominated convergence are uniform along this family. The construction below uses $t=3/4$. \end{lemma} \begin{proof} Use the regular polar and odd equatorial Legendre solutions of the angular equation, and impose potential continuity and weighted normal flux continuity at the cap boundary. Their matching determinant $D(\eta,t,\nu)$ is continuous for $0\leq\nu\leq2$. It is negative at zero and equals $\eta t>0$ at two, so it has a root $\nu_j\in(0,2)$ for each positive contrast. The associated piecewise profile and its weighted flux satisfy the full weak angular equation; odd symmetry gives the zero weighted mean. At zero contrast the only determinant root in $[0,2]$ is two. Indeed, test its weak angular equation against $\widehat x\cdot e$ and test the weak tangential gradient identity against the constant axial vector. These give, for the same strictly positive profile-axis pairing $I$, the identity $2I=\nu I$. Positivity follows from the polar and odd equatorial profiles for $t\geq1/2$. Every subsequential limit of the roots is a zero-contrast root by continuity of the determinant. Compactness therefore gives $\nu_j\to2$ and $\lambda_j=(-1+\sqrt{1+4\nu_j})/2\to1$. The convergent Legendre series and their first derivatives give the stated normalized profile and electric-vector limits, together with the uniform angular majorants. This supplies the real family used in the annular pairing argument. \end{proof} \subsection{Nonvanishing from an annular pairing} We use only positive real contrasts in the following excitation step. The singular coefficient need not be continued analytically, or even shown to converge, as the contrast tends to zero. The proof tests each mode against a localized angular pattern on an annulus around the tip. The key estimate is uniform as the contrast tends to zero: if the singular coefficient vanished, the pairing would be small compared with the annulus size. The same pairing converges to a nonzero multiple of the original mode's axial field. Choosing the annulus sufficiently close to the tip makes these conclusions incompatible. The uniform estimate depends on cancellation of the constant part of the regular correction; simple $L^2$ convergence alone cannot provide it. \begin{theorem}[Eventual excitation by a convergent mode family] \label{hs:modal-coefficient-continuation} Let $\eta_j>0$ tend to zero and let $E_j$ be actual outgoing volume-response Maxwell fields in a fixed conical ball $B_b(p)$ with material $\varepsilon_bq_{\eta_j}$ as in \eqref{hs:modal-angular-material}. Assume their frequencies are nonzero and bounded, their compactly supported volume sources have uniformly bounded $L^2$ norms in one fixed containing ball, and $E_j\to E_0$ in $L^2(B_b(p))^3$. Suppose $E_0$ has a continuous representative near $p$ and $E_0(p)\cdot e\ne0$. Then the intrinsic conical coefficients of $E_j$ are nonzero for all sufficiently large $j$. \end{theorem} \begin{proof} The real angular eigenpairs in Lemma~\ref{hs:modal-angular-branch} have $\nu_j\to2$, $\lambda_j\to1$, and $\lambda_j<1$ for all sufficiently large $j$. The actual compact-source volume representation supplies the decomposition used in Lemma~\ref{hs:three-coefficient}. Its regular correction $W_j$ has a common bound \begin{equation} |W_j(x)-W_j(p)|\leq M|x-p|^\alpha \qquad (x\in B_b(p)), \label{hs:modal-holder-bound} \end{equation} for some $M<\infty$ and $\alpha>0$. Indeed, on the fixed source ball the difference of two scalar outgoing Green rows has squared $L^2$ norm $O(|x-p|)$: split the source ball into a near part and its complement and use the $1/r$ kernel and its $1/r^2$ derivative. Cauchy--Schwarz and the uniform source and frequency bounds give \eqref{hs:modal-holder-bound}, with $\alpha=1/2$. Let $P_j(\zeta)$ be the electric field's angularly projected pairing with a smooth nonnegative radial test $\zeta$ supported in $(0,b)$. Subtracting the constant $W_j(p)$ before solving the projected radial equation removes the apparent $1/(1-\lambda_j)$ singularity. The angular eigenvalue identity cancels the constant correction exactly. The remaining forcing is $O(r^\alpha)$, and the regular particular solution is $O(r^{1+\alpha})$ uniformly as $\lambda_j\uparrow1$. Finite local energy excludes the singular homogeneous solution. Consequently, if the intrinsic coefficient is zero, there is a constant $C$ independent of $j$ such that \begin{equation} |P_j(\zeta)|\leq CM\int_0^b r^{2+\alpha}|\zeta(r)|\,dr. \label{hs:modal-zero-pair-bound} \end{equation} This is the zero-coefficient consequence of the same radial projection and coefficient formula used in Lemma~\ref{hs:three-coefficient}. For each fixed $\zeta$, the $L^2$ convergence of the actual fields and convergence of the real angular projections give $P_j(\zeta)\to P_0(\zeta)$. The limiting pairing has the form \begin{equation} P_0(\zeta)=c_t\int_0^b r^2\zeta(r)A(r)\,dr, \qquad c_t>0,\quad A(r)\longrightarrow E_0(p)\cdot e \quad(r\downarrow0), \end{equation} where $A$ is the corresponding spherical average of the continuous limiting field. Choose a nonzero nonnegative bump supported in a sufficiently small annulus. Continuity and $A(0)\ne0$ give a strict lower bound on $|P_0(\zeta)|$ proportional to $\int r^2\zeta$, whereas the right side of \eqref{hs:modal-zero-pair-bound} has the extra factor $r^\alpha$. Thus the latter is strictly smaller than $|P_0(\zeta)|$. The convergence of $P_j$ contradicts zero coefficient for every sufficiently large $j$. The conclusion uses the actual fields and their local $L^2$ limit; it does not infer nonvanishing merely from $L^2$ closeness without this quantitative projection estimate. \end{proof} \begin{corollary}[Simultaneous excitation by one mode] \label{hs:modal-all-coefficients} Suppose one family of actual resonant fields satisfies the theorem at each of finitely many disjoint conical balls, with the same positive contrasts $\eta_j\downarrow0$, and that its limiting field has a nonzero axial component at each center. Then one sufficiently large index $j$ gives one resonance whose intrinsic coefficient is nonzero at every selected vertex. The contrast can also be chosen below any prescribed positive ceiling, including that required by $\varepsilon_b(1+\eta_j)\leq\varepsilon_+$. \end{corollary} \begin{proof} Apply the theorem at each center and take the maximum of the finitely many resulting indices. The selected coefficients belong to the same actual eigenfield and the same complex resonance. \end{proof} \subsection{Conditional transfer to rounded modal peaks} The next proposition isolates the global spectral assumptions needed by the local argument: one continuing resonance, a stable nonzero normalization, and control of the rounded fields on collars. It is a conditional transfer statement; the next section verifies its hypotheses for an explicit sphere-based family. \begin{proposition}[Transfer from a simple resonance to rounded modal peaks] \label{hs:modal-transfer} Assume a finite homogeneous dielectric sphere with permittivity $\varepsilon_b\in(1,\varepsilon_+)$ has a simple resonance in the axisymmetric TM sector, whose axial electric component is nonzero on an open interval of its symmetry axis. Assume furthermore: \begin{enumerate} \item This sector's eigenpair has a holomorphic continuation under the finite axisymmetric cone insertions in Corollary~\ref{hs:modal-all-coefficients}. \item An explicitly specified genuine QNM normalization is finite, nonzero and continuous along that branch and under the rounded approximations below. \item Rounding the conical vertices while retaining the same material values gives a nearby resonance and normalized eigenfield converging locally in $L^2$, with uniformly bounded one-sided fields on fixed smooth-interface collars away from the vertices. \end{enumerate} Then, for every finite $N$ and every $M,\ell>0$, one normalized resonance of a finite smooth dielectric has $N$ disjoint three-dimensional peak balls of diameter less than $\ell$, each with one-sided maximum larger than $M$ and a collar maximum smaller than half its peak. Their centers may be placed on a line in a prescribed open plane patch, and their projections may be contained in that patch, provided that the chosen nonzero axial interval is positioned to meet the patch. If the symmetry axis lies in that plane and the rounding remains axisymmetric, a maximizing point in each ball can itself be selected in the plane patch. The surrounding resonator need not project into the patch. A maximum off the symmetry axis may belong to a rotational ring inside its small ball; no nondegeneracy or uniqueness of a maximum is asserted. \end{proposition} \begin{proof} Choose $N$ distinct axial centers in a sufficiently short nonzero subinterval and choose disjoint balls and insertions as small as necessary. Axisymmetric insertions preserve the indicated sector. Choose one small positive contrast from Corollary~\ref{hs:modal-all-coefficients}. Its one normalized sharp resonance is unbounded at every tip. The uniform collar hypothesis gives one bound for all sufficiently late rounded modes on the chosen collars. In each ball, essential unboundedness gives a positive-volume superlevel above twice that common collar bound and the target $M$, with strict margins. Local $L^2$ convergence preserves a positive-volume superlevel above these thresholds at every sufficiently small rounding scale. The eventual collar bound and finiteness of the collection let one choose a common scale. Smooth-interface regularity makes the one-sided maxima on the closed balls finite and attained; each peak lies inside its ball because of the collar inequality. This is the same final compactness step as the driven construction, but it is applied to a homogeneous resonance branch with its stated QNM normalization. For the last placement assertion, an axisymmetric vector field obeys $E(\mathcal R x)=\mathcal R E(x)$ under rotations about its axis, and therefore $|E|$ is invariant. The same is true for its one-sided material traces. A ball centered on that axis is rotation invariant. Rotate any maximizing point into the specified plane through the axis; it stays in the same ball and retains its maximal value. The projection constraint puts this representative in the prescribed patch. \end{proof} This proposition is a proved implication with explicit global hypotheses. Lemmas~\ref{ms:sphere-simple}, \ref{ms:continuation}, \ref{ms:normalization}, and~\ref{ms:rounding} supply those hypotheses for the finite-device construction in Theorem~\ref{ms:modal-packing}. Theorem~\ref{hs:modal-coefficient-continuation} and Corollary~\ref{hs:modal-all-coefficients} establish its local excitation bridge. No resonance is inferred from a driven field, and no radiating mode is normalized by a divergent all-space positive energy integral. A real dilation fixes the real part of the nonzero resonance frequency; its negative imaginary part remains part of the QNM statement.