\section{Spectral construction and normalization of the modal branch} \label{ms:section} This section supplies the global hypotheses used by Proposition~\ref{hs:modal-transfer}. Frequencies in this section are complex vacuum wavenumbers. An outgoing resonance means the analytic continuation of the vacuum outgoing Green function from $\operatorname{Im}k>0$; no decay at real spatial infinity is imposed when $\operatorname{Im}k<0$. The material is still real, positive, scalar, nondispersive, and equal to vacuum outside a bounded set. The normalization below is a reciprocal Green-function residue norm. It is not a positive all-space energy integral. The construction starts with a dielectric sphere whose selected electric resonance has a nonzero axial field. A sequence of radial dipole resonances provides arbitrarily high quality factor. Small conical insertions preserve one simple resonance in a specified symmetry sector; the preceding section shows that the continued field excites all their tips. Rounding then yields a smooth finite device with large, localized peaks in that same mode. Finally, a spatial dilation restores the exact target real frequency without changing the quality factor. There are two distinct reasons for the spectral machinery. The pole must survive changes of the material and remain a genuine outgoing Maxwell solution; separately, the amplitude of its eigenfield must be fixed by a nonzero normalization that also survives those changes. The following subsections establish those properties before combining them in Theorem~\ref{ms:modal-packing}. \subsection{A compact analytic pencil and continuity under rounding} An eigenfield can be multiplied by any nonzero constant, so convergence of modes requires a choice of scale. The operator pencil below first locates resonances as values of $k$ at which the source-free volume equation has a nonzero solution. Splitting off the static part leaves a compact analytic problem. This form also controls changes in the material that converge almost everywhere, as happens when tips are rounded; pointwise material convergence alone would not generally give operator-norm convergence of the original multiplication operator. Let $B$ be a ball containing all material changes, let $X=L^2(B;\C^3)$, and extend its elements by zero when applying whole-space operators. Write $\chi=\varepsilon_r-1$. With $\Phi_k(x)=e^{ik|x|}/(4\pi|x|)$, put \begin{equation} K(k)f=\left.(k^2I+\nabla\nabla)\Phi_k*f\right|_B, \qquad L_\chi(k)=I-K(k)\chi. \label{ms:pencil} \end{equation} Multiplication by $\chi$ is included in the last expression. The static longitudinal Fourier projection, compressed to $B$, is denoted by $P$. It satisfies $P=P^*$ and $0\leq P\leq I$. The distributional identity \begin{equation} K(k)=-P+R(k),\qquad R(k)=k^2\Phi_k I+\nabla\nabla(\Phi_k-\Phi_0),\qquad R(0)=0 \label{ms:compact-split} \end{equation} holds as an identity of bounded operators on $X$. \begin{lemma}[Analytic Fredholm reduction with a fixed material bound] \label{ms:fredholm} For every measurable $0\leq\chi\leq C$ supported in $B$, \begin{equation} L_\chi(k)=A_\chi F_\chi(k),\quad A_\chi=I+P\chi,\quad F_\chi(k)=I-A_\chi^{-1}R(k)\chi \label{ms:factorization} \end{equation} has $A_\chi$ invertible with $\|A_\chi^{-1}\|\leq1+C$. The operator $R(k)$ is an entire Hilbert--Schmidt family, and $F_\chi$ is an analytic identity-minus-compact pencil with $F_\chi(0)=I$. Its noninvertible values are isolated, of finite algebraic multiplicity. At $k\ne0$, its kernel is exactly the restriction to $B$ of the finite local-energy Maxwell resonances for this material, with exterior field given by \eqref{ms:pencil}. If $0\leq\chi_j\leq C$ and $\chi_j\to\chi$ almost everywhere in this same ball, then $F_{\chi_j}(k)\to F_\chi(k)$ in operator norm, locally uniformly in $k$, as do all $k$ derivatives. \end{lemma} \begin{proof} The Fourier symbol of $\nabla\nabla\Phi_0$ is $-\xi\xi^T/|\xi|^2$, including its distributional local term. Taylor expansion of $e^{ikr}$ shows that each entry of the remaining kernel in \eqref{ms:compact-split} is $O(1/r)$ as $r\downarrow0$, uniformly for $k$ in compact subsets of $\C$. The constant term $ik/(4\pi)$ in $\Phi_k-\Phi_0$ has zero Hessian. The squared kernel is locally integrable in three dimensions; on bounded $B\times B$ it is integrable. Its parameter power series and derivatives have the same local majorants on compact parameter sets. This proves the entire Hilbert--Schmidt assertion. Put $s=\sqrt\chi$. The positive selfadjoint operator $I+sPs$ has inverse of norm at most one, and direct multiplication gives \begin{equation} A_\chi^{-1}=I-Ps(I+sPs)^{-1}s. \label{ms:static-inverse} \end{equation} This proves the stated bound and factorization. Analytic Fredholm theory applies, with the invertible value at zero fixing its nontrivial alternative. This application needs no small contrast, smooth boundary, or uniform feature size. For completeness, the volume representation is an exact Maxwell equation. In distributions, $(\operatorname{curl}\operatorname{curl}-k^2)K(k)f=k^2f$. Consequently a kernel vector extended by $E=K(k)(\chi E)$ satisfies $\operatorname{curl}\operatorname{curl}E =k^2(1+\chi)E$. Its curl is locally square integrable, by the first-derivative Newton-potential estimate; equivalently this follows from local cutoff testing. Set $H=(ik)^{-1}\operatorname{curl}E$ in units with vacuum impedance one. This gives the two first-order Maxwell equations and the outgoing continued representation. Conversely, subtracting this volume field from any such resonance leaves an entire vacuum Maxwell field regular at the origin and outgoing outside a ball. In each vector spherical harmonic its regular radial function and outgoing radial function have nonzero Wronskian for $k\ne0$, so all coefficients vanish. Thus the restriction is a kernel vector. A zero restriction gives the zero global field. Almost-everywhere convergence and the common bound give strong convergence of the multiplication operators. The resolvent identity and \eqref{ms:static-inverse} give strong convergence of $A_{\chi_j}^{-1}$ with a uniform bound. Dominated convergence of the integral kernels gives $R(k)\chi_j\to R(k)\chi$ in Hilbert--Schmidt norm. A uniformly bounded strongly convergent operator sequence converges in norm after composition on the right with a fixed compact operator: approximate that compact operator by finite-rank operators. Apply this observation to $A_{\chi_j}^{-1}R(k)\chi$ and add the Hilbert--Schmidt difference. Compactness in the parameter $k$ makes the convergence locally uniform; Cauchy's formula gives the same claim for derivatives. \end{proof} All these arguments also hold on a fixed closed symmetry sector preserved by $P$, $R(k)$, and the material multiplication. We use the axisymmetric electric sector: $E$ has only meridional components and $H$ only the azimuthal component in cylindrical coordinates. Rotationally invariant scalar material preserves this sector. It is closed in $L^2$ and orthogonal to the axisymmetric magnetic sector, both for the Hermitian pairing and for the unconjugated bilinear pairing. At the coordinate axis these statements are understood in the distributional and almost-everywhere sense. \begin{lemma}[Persistence of a simple resonance] \label{ms:continuation} Suppose the restriction of $F_\chi$ to such a fixed sector has an isolated characteristic value $k_*$ of algebraic multiplicity one. An analytic $L^\infty$ material perturbation has a locally holomorphic continued wavenumber $k_\eta$ and nonzero eigenfield $E_\eta\in X$. Their values at zero are $k_*$ and a prescribed nonzero eigenfield. If instead bounded nonnegative materials $\chi_j$ converge almost everywhere to $\chi$, and preserve this sector, they have unique nearby simple characteristic values $k_j\to k_*$ and eigenfields $E_j\to E$ in $X$, after one fixed nonzero linear functional fixes their scale. \end{lemma} \begin{proof} Choose closed complements of the one-dimensional kernel and cokernel of $F_\chi(k_*)$. The block between those complements is invertible, and remains so in a small operator-norm neighborhood. Block elimination reduces the kernel equation to a single scalar analytic function $f(k)$, multiplied by invertible analytic block operators. Algebraic multiplicity one is exactly $f(k_*)=0$, $f'(k_*)\ne0$. For an analytic material perturbation, $A_\chi^{-1}$ continues analytically by a Neumann series near the base material, so every block is jointly analytic. The scalar analytic implicit-function theorem supplies $k_\eta$; back substitution supplies $E_\eta$. For almost-everywhere material convergence, Lemma~\ref{ms:fredholm} gives uniform convergence of the blocks and scalar functions on a small closed disk. Rouch\'e's theorem gives exactly one zero, counting multiplicity. It tends to $k_*$ and is simple. Back substitution and a linear functional nonzero on $E$ give the stated field convergence. This reduction must use $F$ for the almost-everywhere convergence claim: the leading multiplication in $L$ need not converge in operator norm. Multiplication by $A$ is analytic and invertible and does not alter a characteristic value's multiplicity or its right kernel. \end{proof} \subsection{A simple sphere resonance without a genericity assumption} We use a sphere because its modes can be separated into angular and radial factors and its resonance condition can be written explicitly. The selected dipole-like mode has nonzero axial electric field near the center, providing places for the cone tips. Simplicity is required only within the invariant sector used for the construction; the familiar degeneracy among other azimuthal directions is not asserted to disappear. Choose an algebraic real refractive index $11$. In a sphere of radius $a$ and permittivity $n^2$, write $z=ka$ and $\psi_l(z)=zj_l(z)$, $\xi_l(z)=zh_l^{(1)}(z)$. Matching the tangential electric and magnetic fields of the regular interior electric multipole and the outgoing exterior multipole gives, for $l\geq1$, the determinant \begin{equation} D_l(z)=n\psi_l(nz)\xi_l'(z)-\xi_l(z)\psi_l'(nz). \label{ms:mie-denominator} \end{equation} Primes mean derivatives with respect to the displayed argument. For example the interior tangential electric component is proportional to $\psi_l'(nz)/(nz)$, and its tangential magnetic component to $n\psi_l(nz)/(nz)$; their exterior analogues are $\xi_l'(z)/z$ and $\xi_l(z)/z$. Eliminating their two amplitudes gives \eqref{ms:mie-denominator}. \begin{lemma}[Simplicity and separation of electric sphere poles] \label{ms:sphere-simple} For algebraic $n>1$, every nonzero zero of $D_l$ is simple, and $D_l$ and $D_j$ have no common nonzero zero if $l\ne j$. In the axisymmetric electric sector each such zero is therefore a simple Maxwell characteristic value. There are $l=1$ poles \begin{equation} z_m=\frac{\pi(m+1/2)}{n} -\frac{i}{2n}\log\frac{n+1}{n-1}+O(m^{-1}), \qquad m\longrightarrow\infty. \label{ms:pole-asymptotic} \end{equation} Their real parts are positive, imaginary parts negative, and $\operatorname{Re}z_m/(-2\operatorname{Im}z_m)\to\infty$. The $l=1,m_{\rm az}=0$ interior electric field has a nonzero axial component at the center and on a sufficiently short axial interval. \end{lemma} \begin{proof} The exact finite spherical-Hankel expansions\cite{dlmfExplicit} are \begin{align} \xi_l(z)&=(-i)^{l+1}e^{iz}p_l(z),& z h_l^{(2)}(z)&=i^{l+1}e^{-iz}\widetilde p_l(z),\\ p_l(z)&=\sum_{s=0}^l\frac{i^s(l+s)!}{2^s s!(l-s)!z^s},& \widetilde p_l(z)&=\sum_{s=0}^l\frac{(-i)^s(l+s)!}{2^s s!(l-s)!z^s}. \label{ms:hankel-polynomials} \end{align} Using $\psi_l=(\xi_l+zh_l^{(2)})/2$, the determinant, up to a nonzero factor independent of its zero condition, is \begin{equation} C_{l,+}(z)e^{inz}+C_{l,-}(z)e^{-inz}, \label{ms:exp-polynomial} \end{equation} where, with $p=p_l$, $\widetilde p=\widetilde p_l$, \begin{align} C_{l,+}&=(-i)^{l+1}\{i(n-1)p(z)p(nz) +np(nz)p'(z)-p(z)p'(nz)\},\\ C_{l,-}&=i^{l+1}\{i(n+1)p(z)\widetilde p(nz) +n\widetilde p(nz)p'(z)-p(z)\widetilde p'(nz)\}. \label{ms:rational-coefficients} \end{align} These are rational functions with algebraic coefficients. They cannot vanish together at $z\ne0$: the two corresponding equations would annihilate both independent interior Riccati--Hankel solutions. Their nonzero Wronskian would then force both $\xi_l(z)$ and $\xi_l'(z)$ to vanish, contradicting uniqueness for their second-order radial equation. Thus at any zero of $D_l$ both coefficients are nonzero and the equation is equivalently \begin{equation} e^{2inz}=R_l(z):=-C_{l,-}(z)/C_{l,+}(z). \label{ms:rational-pole} \end{equation} The expansion $p_l(z)=1+i l(l+1)/(2z)+O(z^{-2})$ gives \begin{equation} R_l(z)=(-1)^l\frac{n+1}{n-1} \left(1-\frac{i l(l+1)}{nz}+O(z^{-2})\right). \label{ms:rational-asymptotic} \end{equation} For different parity the leading constants differ; for the same parity and distinct $l$ the coefficients of $1/z$ differ. Hence $R_l-R_j$ is never identically zero when $l\ne j$. The Lindemann--Weierstrass theorem implies that $e^\alpha$ is transcendental for every nonzero algebraic complex $\alpha$\cite{bezivinRobba1989}. If $D_l$ and $D_j$ had a common nonzero zero $z$, the nontrivial rational equation $R_l(z)=R_j(z)$ would make $z$ algebraic. Equation \eqref{ms:rational-pole} would make $e^{2inz}$ algebraic, a contradiction. If $D_l$ had a multiple nonzero zero, differentiation of \eqref{ms:rational-pole} would give $R_l'(z)=2inR_l(z)$. This too is a nontrivial rational equation: $R_l'/R_l\to0$ at infinity, whereas $2in\ne0$. It again makes $z$ algebraic and yields the same contradiction. No division by a zero coefficient was used at a pole. Separation of variables decomposes the axisymmetric electric sector into the degrees $l=1,2,\ldots$, with one azimuthal mode at each degree. Regularity at the origin and the outgoing condition leave one radial amplitude on each side of the sphere. Locally in $k\ne0$, solving the inhomogeneous radial equations and matching them reduces each block, by analytic invertible elimination, to its two-by-two tangential matching matrix. Its determinant is \eqref{ms:mie-denominator} times a nonzero analytic factor. Thus a simple zero is a one-dimensional simple characteristic value of that block. Lemma~\ref{ms:fredholm} excludes accumulation at a finite nonzero $k$; the other angular blocks are invertible near the selected zero. The proved separation across $l$ gives algebraic multiplicity one in the whole selected sector. The familiar degeneracy across the other azimuthal indices in the full sphere is irrelevant to this invariant-sector statement. For $l=1$ let $r_\infty=-(n+1)/(n-1)$ and choose the centers $z_m^0$ equal to the two leading terms in \eqref{ms:pole-asymptotic}. They obey $e^{2inz_m^0}=r_\infty$. On the circle $|z-z_m^0|=C/m$, the modulus of $e^{2inz}-r_\infty$ is at least $cC/m$ for large $m$, with $c>0$ independent of $m$. The rational correction $R_1(z)-r_\infty$ is at most $C_0/m$ there. Taking $C$ large enough, Rouch\'e's theorem gives exactly one zero in this disk. This proves \eqref{ms:pole-asymptotic}. Finally $\psi_1(u)=u^2/3+O(u^4)$. The regular electric dipole vector spherical wave therefore tends to a nonzero constant axial vector at the center, for a nonzero radial amplitude. The amplitude cannot vanish without making the whole matched eigenfield zero. Analyticity in the homogeneous core preserves the nonzero component on a short axial interval. \end{proof} \subsection{Reciprocal normalization of the resonant mode} An outgoing mode at a complex resonance frequency grows at real spatial infinity, so its ordinary positive all-space energy is not a finite normalization. Instead, we use the coefficient of the Green operator's pole. A simple pole has a nonzero derivative pairing, and this pairing sets the scale of the mode. The definition below is bilinear and may be complex; setting it to one is a choice of mode amplitude, not a claim that it equals positive stored energy. The bilinear pairing in the remainder of this section has no complex conjugation. Reciprocity gives $K(k)^T=K(k)$ on $X$. For a simple characteristic value in the selected sector, define \begin{equation} d(E,k,\chi)=-\int_B \chi E\cdot K'(k)(\chi E)\,dx, \qquad \mathcal N(E)=-\frac{k}{2}d(E,k,\chi). \label{ms:residue-norm} \end{equation} The derivative kernel is a bounded compact operator; these are finite integrals of locally square-integrable fields. Their values are independent of which containing ball is used. \begin{lemma}[Nonzero residue norm and its stability] \label{ms:normalization} At a simple nonzero Maxwell pole in the selected sector, $\mathcal N(E)\ne0$. Scaling $E$ by $a\in\C$ scales $\mathcal N$ by $a^2$. The reciprocal electric Green operator for $\operatorname{curl}\operatorname{curl}-k^2\varepsilon_r$, with both the source and output restricted to this sector, has residue \begin{equation} \operatorname*{Res}_{k=k_*}G(k;x,y) =-\frac{E(x)\otimes E(y)}{2k_*\mathcal N(E)}. \label{ms:green-residue} \end{equation} The norm is continuous along the almost-everywhere material and eigenfield convergence of Lemma~\ref{ms:continuation}, and holomorphic along its analytic branches. A local nonzero analytic choice of scale makes $\mathcal N=1$. \end{lemma} \begin{proof} The functional $\ell_E(v)=\int_B\chi E\cdot v$ is nonzero: if $\chi E=0$, the volume equation makes $E=0$. It lies in the left kernel of $L_\chi(k_*)$, since $\ell_E L_\chi=\langle\chi E,\cdot\rangle- \langle K(k_*)\chi E,\chi\,\cdot\rangle=0$. It is a nonzero functional also on the chosen sector, because that sector is closed under complex conjugation and multiplication by the real rotationally invariant $\chi$. Its derivative pairing $\ell_E L_\chi'(k_*)E$ is exactly $d$. The scalar Schur reduction in Lemma~\ref{ms:continuation} shows that this pairing is nonzero at a simple characteristic value; otherwise the scalar zero would have multiplicity at least two. The residue of $L_\chi^{-1}$ is therefore $E\otimes\ell_E/d$. The vacuum electric resolvent is $G_0(k)=k^{-2}K(k)$, with its natural restriction when the source is supported in $B$. The full electric field for a compact source is $L_\chi(k)^{-1}G_0(k)f$ in $B$. At the pole, reciprocity and $K(k_*)\chi E=E$ give \begin{equation} \ell_E G_0(k_*)f=k_*^{-2}\int E(y)\cdot f(y)\,dy. \end{equation} For an arbitrary compact source, first project it onto the selected sector. The same identity for sources outside $B$ follows from the global volume representation and the symmetric kernel. Since $E$ lies in the sector, the projection can be removed from its bilinear pairing with $E$; the Maxwell source itself remains the projected source. Multiplying the two residue factors yields $E\otimes E/(k_*^2d)$, which is \eqref{ms:green-residue}. This is a genuine pole normalization of a quasinormal mode, consistent with reciprocal Green-residue normalizations\cite{muljarov2016}; the sign reflects the displayed choice of Maxwell operator. For a degenerate full-space pole the formula describes the selected symmetry block, not a claim that the entire Green residue has rank one. Under rounding, $\chi_j E_j\to\chi E$ in $L^2$: split the difference into the uniformly bounded multiplier times $E_j-E$ and dominated convergence applied to $(\chi_j-\chi)E$. Operator-norm continuity of $K'(k)$ proves convergence of $d$ and $\mathcal N$. Holomorphy and quadratic scaling follow from the formulas. A nonzero analytic square root of $\mathcal N$ on a sufficiently small parameter disk normalizes the field. \end{proof} \begin{lemma}[Normalized resonance convergence and fixed-collar bounds] \label{ms:rounding} Fix the finite axisymmetric conical construction and a simple sector pole with $\mathcal N=1$. Round all its vertices axisymmetrically within balls of radius $\rho$, leaving the material values and all other interfaces fixed. For sufficiently small $\rho$, there is a simple sector pole $k_\rho\to k_*$ and eigenfield with $\mathcal N(E_\rho)=1$ converging to $E$ locally in $L^2$. On the fixed collars around the selected tips, contained in their conical neighborhoods and separated from the vertices, the one-sided fields have a common finite bound independent of the sufficiently small rounding radius. \end{lemma} \begin{proof} The rounded coefficients converge almost everywhere with fixed support and the same positive material bound. Apply Lemmas~\ref{ms:fredholm}--\ref{ms:normalization} and choose the square-root scale converging to the sharp field's scale. The volume representation supplies local $L^2$ convergence also on any larger fixed ball, by the same static and compact estimates. For the collar estimate, the convergent normalized volume unknowns give uniformly bounded compact $L^2$ sources and frequencies. The scalar Green representation writes the actual electric field as $E_\rho=\nabla v_\rho+W_\rho$, with $W_\rho=k_\rho^2\Phi_{k_\rho}*(\chi_\rho E_\rho)$. On a fixed containing ball the scalar Green operator, its gradient, and the dynamic-minus-static Hessian have uniform operator bounds. The static Hessian is controlled by the Fourier Newton projection. After multiplication by a fixed smooth cutoff, $W_\rho$ therefore has uniformly bounded first and second weak derivatives. The localized scalar potential and its first gradient are uniformly bounded in $L^2$ as well. Choose a smooth reference rounding before the final family index. Every sufficiently late rounded body agrees exactly with it on the fixed collars. Cover those collars by finitely many fixed signed $C^4$ graph charts, including constant-material charts at points away from the interfaces. The scalar Gauss equation becomes a positive binary transmission equation in each flattened chart. A regional difference-quotient estimate supplies the mixed scalar derivative needed in the outer-cutoff forcing. Product rules then bound both tangential derivatives of that forcing from the original data and fixed cutoff constants. Thus the subsequent localized iterations do not assume extra scalar regularity on the outer cutoff annulus. Two tangential iterations bound the tangential gradient components. The divergence equation gives the corresponding bounds for the normal flux. The local Sobolev estimates for these derivative arrays yield continuous bounded representatives. On the inner cutoff plateau they are the original tangential gradient and physical flux; the positive material values reconstruct the actual one-sided electric components. Taking a finite cover gives a single collar bound chosen before the rounding index and the peak threshold. \end{proof} \subsection{Modal hotspot theorem} The theorem assembles the local excitation and global continuation steps. All peaks belong to the same mode; the construction does not require a separate mode for each observation region. The target frequency is the real part of the complex resonance frequency, and the norm is the reciprocal residue normalization defined above. The theorem allows the support and fine feature scales to vary with the requested number and strength of peaks. \begin{theorem}[Arbitrarily many peaks in one normalized finite-device mode] \label{ms:modal-packing} Fix $\varepsilon_+>1$, a real target wavenumber $k_0>0$, a nonempty open patch $\mathcal P$ of a plane, a positive integer $N$, and numbers $\ell,M,Q_0>0$. There is a finite smooth piecewise-constant dielectric with vacuum exterior and $1\leq\varepsilon_r\leq\varepsilon_+$, and one nonzero outgoing Maxwell quasinormal mode $(E,H,k_*)$, such that \begin{equation} \operatorname{Re}k_*=k_0,\qquad \operatorname{Im}k_*<0,\qquad \frac{\operatorname{Re}k_*}{-2\operatorname{Im}k_*}>Q_0, \qquad \mathcal N(E)=1. \label{ms:modal-parameters} \end{equation} There are $N$ disjoint three-dimensional observation balls of diameter less than $\ell$, centered in $\mathcal P$ and with orthogonal projections contained in $\mathcal P$. In each ball the one-sided electric-field maximum exceeds $M$ and is more than twice a uniform bound for both material-side fields on an open collar of its boundary. A point attaining each maximum can be chosen in $\mathcal P$. Thus each peak's half-amplitude connected region stays in its small ball. No uniform support-size bound, minimum feature size, nondegeneracy of the maximum, or positive energy fraction per peak is asserted. The insertion contrast may depend on $N$; the common material bounds do not. \end{theorem} \begin{proof} Take an algebraic index $n$ strictly between one and $\sqrt{\varepsilon_+}$. Lemma~\ref{ms:sphere-simple} gives a simple axisymmetric electric dipole pole with quality factor strictly greater than $Q_0$. Scale the sphere initially to make the real part of its wavenumber $k_0$. Put its symmetry axis in the target plane and its center in the open patch. Choose $N$ distinct points on a sufficiently short axial interval where the background axial component is nonzero, with ample clearance from the patch boundary. Choose disjoint arbitrarily small conical neighborhoods and close their material insertions smoothly away from their vertices, inside the homogeneous sphere. Axisymmetry preserves the electric sector. Lemma~\ref{ms:continuation} gives one holomorphic pole and eigenfield under simultaneous positive cone contrast $\eta$. Theorem~\ref{hs:modal-coefficient-continuation} and Corollary~\ref{hs:modal-all-coefficients} give nonzero singular coefficients at all $N$ vertices for one sufficiently small $\eta>0$. Choose it also so that $n^2(1+\eta)<\varepsilon_+$, the pole remains simple, its imaginary part remains negative, and its quality factor remains strictly above $Q_0$. Lemma~\ref{ms:normalization} fixes its genuine norm to one. Lemma~\ref{ms:rounding} supplies every rounding and normalization hypothesis of Proposition~\ref{hs:modal-transfer}. Its high-point and collar argument now proves all the simultaneous strict peak inequalities for sufficiently small rounding radius. The same axisymmetric rotation argument puts an actual maximizing point of each ball in the specified plane. There is one final harmless adjustment because the real pole frequency has shifted. About a point in the interior of the patch, dilate space by $s=\operatorname{Re}k_*/k_0>0$. Maxwell scaling sends the wavenumber to $k_*/s$, keeps the quality factor and material values unchanged, and tends to the identity as the insertion contrast tends to zero. A normalized field scales as $E_s(x)=s^{-3/2}E(x/s)$: indeed \eqref{ms:residue-norm} and the scaled kernel give $\mathcal N(E(x/s))=s^3\mathcal N(E)$. Choose all observation-ball diameters with strict margins below $\ell$, their projections with strict interior clearance, and their pre-scaling peak targets larger than $s^{3/2}M$ for every $s$ in a fixed small neighborhood of one. These choices can be made before the final rounding. The dilation therefore preserves every desired inequality and sets the exact real frequency. \end{proof} \begin{corollary}[Vanishing residue-normalized localization ratios] \label{ms:modal-volume} For one fixed sharp modal construction with $N$ tips, its rounded normalized modes, with the frequency-adjusting dilations and corresponding observation balls, have reciprocal norm one, and the maximum in each selected ball tends to infinity. Consequently \begin{equation} \frac{|\mathcal N(E_\rho)|} {\displaystyle\max_{x\text{ in ball }j} \varepsilon_\rho(x)|E_\rho(x)|^2}\longrightarrow0 \quad (1\leq j\leq N). \end{equation} These are positive intensity ratios formed from the genuine reciprocal norm. The displayed maximum is over both continuous material-side traces on the closed observation ball; at an interface, each trace is weighted by the permittivity of that same side. The same representatives are defined continuously on the phase closures in a strictly larger ball and agree almost everywhere there with the actual field. Thus their values on the observation ball are intrinsic, including at tangential contact between its sphere and an interface. A directional complex mode volume uses $\mathcal N/[\varepsilon(u\cdot E)^2]$ instead; selecting a Cartesian component of size at least $|E|/\sqrt3$ gives the same vanishing magnitude along a subsequence with a fixed component. No divergent all-space positive energy is substituted for $\mathcal N$. \end{corollary} \begin{proof} For every finite amplitude threshold, essential unboundedness of the fixed sharp eigenfield gives a positive-volume superlevel in each selected ball. Local $L^2$ convergence preserves a positive-volume superlevel above any smaller threshold for every sufficiently late rounded field. Its attained one-sided maximum is at least that threshold. This proves divergence of every maximum along the fixed rounded family. For the maximum selection, first construct the two continuous phase representatives on a larger closed ball and then restrict these same functions to the smaller observation ball. Two such representatives agree on the open phase by continuity and almost-everywhere equality; taking phase limits proves equality on its closure in the smaller ball. This argument also handles tangencies, where continuity and almost-everywhere equality on the smaller ball alone would not suffice. The normalized numerator is one, and the positive permittivities have a common lower bound. There are only three Cartesian components, so the indicated subsequence exists for a directional volume. The finite-frequency dilation has a finite nonzero limit and leaves the vanishing-ratio conclusion unchanged. \end{proof}