\section{Simultaneous driven hotspots in an invariant Maxwell geometry} \label{hs:section} Can one finite-power illumination produce many strong, sharply localized electric-field peaks at once? We first answer a cross-sectional version of this question. The dielectric pattern and the fields are invariant along one direction, so each peak in the picture represents an infinitely long ridge. The three-dimensional construction in Section~\ref{hs:three-section} removes that invariant direction by a separate argument. The mechanism begins with two dielectric regions meeting at a sharp corner. At an excited corner, the field can grow without bound even though its local energy remains finite. A physical device in our construction has rounded, smooth interfaces: rounding cuts off this growth, leaving a finite peak. The proof chooses the rounding small enough that this peak is high while the field on a surrounding collar remains bounded. Figure~\ref{hs:fig-corner} shows the local geometry; the full construction places finitely many such neighborhoods in one device and uses one incident field to excite them all. \input{figures/hotspot-schematic} \noindent\textbf{Scope.} Theorem~\ref{hs:packing} proves a driven, two-dimensional cross-sectional version of hotspot packing, with physical incoming power per unit invariant length. It uses ordinary binary dielectrics at one fixed positive frequency, one incident field, and smooth rounded interfaces. The proof includes local finite-frequency excitation, global illumination, and the rounding limit. It does not prove packing of finite-three-dimensional point hotspots or packing by a normalized resonator mode. Those distinctions are essential when comparing with the modal question in \cite{millerMonticone2026}. \subsection{Physical reduction, normalization, and the exact target} \label{hs:setup} Fix $k>0$, $\omega=ck$, and $\varepsilon_+>1$. Set $b=\varepsilon_+^{-1}\in(0,1)$ and $Z_0=\sqrt{\mu_0/\varepsilon_0}$. Every coefficient below is independent of $z$. Put $a=1/\varepsilon_r$, so that $b\leq a\leq1$, and consider \begin{equation} H=(0,0,u),\qquad E=\frac{ia}{\omega\varepsilon_0}(\partial_yu,-\partial_xu,0), \qquad L_a u:=\nabla\cdot(a\nabla u)+k^2u=0. \label{hs:tm} \end{equation} These are the full three-dimensional Maxwell equations restricted to $z$-invariant fields, rather than an electrostatic approximation. Indeed $\nabla\times H=-i\omega\varepsilon_0\varepsilon_rE$, and $\nabla\times E=i\omega\mu_0H$ is exactly $L_au=0$. The weak scalar transmission conditions are $[u]=[a\partial_nu]=0$. They give continuity of tangential $H$ and $E$, and of normal displacement. The local electric energy is finite precisely when $\nabla u$ is locally square integrable, since \begin{equation} \int_Q\varepsilon_0\varepsilon_r|E|^2\,dx\,dy =\frac{1}{\omega^2\varepsilon_0}\int_Q a|\nabla u|^2\,dx\,dy. \label{hs:energy} \end{equation} The incident field is a smooth Herglotz wave \begin{equation} u_g^{\rm in}(x)=\left(\frac{k}{\pi Z_0}\right)^{1/2} \int_{S^1}g(d)e^{ikd\cdot x}\,ds(d),\qquad g\in C^\infty(S^1). \label{hs:herglotz} \end{equation} The scattered field satisfies the two-dimensional outgoing Sommerfeld condition. All material lies in a bounded cross section, and the exterior is vacuum. This convention specifies the incoming physical channels; no freely chosen homogeneous outgoing solution is added. Stationary phase, for $x=r\widehat x$, gives \begin{equation} u_g^{\rm in}(r\widehat x)= \left(\frac{2}{Z_0r}\right)^{1/2} \left[e^{i(kr-\pi/4)}g(\widehat x) +e^{-i(kr-\pi/4)}g(-\widehat x)\right]+O(r^{-3/2}). \label{hs:asymptotic-input} \end{equation} An incoming radial wave has inward flux $Z_0|u|^2/2$; an outgoing one has outward flux $Z_0|u|^2/2$. Therefore \begin{equation} P_{\rm in}\text{ per unit }z\text{ length}=\|g\|_{L^2(S^1)}^2. \label{hs:power} \end{equation} This is incoming, rather than net, power. The latter vanishes for a lossless scatterer after the outgoing power is included. A plane wave alone would not have this finite incoming-power normalization. \begin{definition}[Localized peaks used here] \label{hs:peak-definition} For a smooth-interface, piecewise homogeneous configuration, let $E_*(x)$ equal $|E(x)|$ off interfaces and the maximum of the finitely many one-sided values at an interface. A collection of $N$ localized peaks at length target $\ell>0$ and amplitude target $M>0$ means $N$ pairwise disjoint closed disks $Q_j$, all of diameter below $\ell$, such that \begin{equation} P_j:=\max_{\overline Q_j,\,\text{one-sided}}|E|>M, \qquad \max_{\partial Q_j,\,\text{one-sided}}|E|P_j/2\}$ containing a maximizing point stays in $Q_j$. Its diameter is below $\ell$. The corresponding intensity half-maximum component is contained in it, since $P_j/\sqrt2>P_j/2$. No positive lower bound on the energy in each peak is stipulated. \end{definition} \begin{theorem}[Simultaneous rounded driven hotspot packing] \label{hs:packing} For every fixed $k>0$ and $\varepsilon_+>1$, every nonempty open patch $\mathcal P\subset\mathbb R^2$, every finite $N\geq1$, and every $\ell,M>0$, there exist $2N$ bounded dielectric inclusions with pairwise disjoint closures and $C^\infty$ boundaries, all contained in $\mathcal P$, and one $g\in C^\infty(S^1)$ with $\|g\|_2=1$, such that: \begin{enumerate} \item $\varepsilon_r=\varepsilon_+$ inside the inclusions and $\varepsilon_r=1$ outside them; \item the unique outgoing scattering solution of \eqref{hs:tm} driven by \eqref{hs:herglotz} has $N$ localized peaks in $\mathcal P$ satisfying Definition~\ref{hs:peak-definition}; \item the same one incident field excites every peak, at the same frequency and with unit incoming power per unit $z$ length. \end{enumerate} There is no uniform lower bound on the rounding radius, on any peak's integrated energy, or on any singular coefficient as $N$ or the target parameters change. The physical construction is infinitely extruded in $z$ and is not a finite-volume three-dimensional device. \end{theorem} The proof has four steps. First, a sharp checkerboard corner admits a finite-energy field with an electric singularity. Second, an outgoing scattering and approximation argument shows that an incident field can excite each prescribed corner. Linearity then lets us choose one field that excites all finitely many corners simultaneously. Third, the total fields converge as the corners are rounded. Finally, the sharp field provides large values near each corner, while a uniform collar bound keeps the boundary of each observation disk low. This last comparison gives actual localized maxima in the smooth devices, not merely large values in an ideal singular geometry. The strict contrast assumption matters: for $\varepsilon_+=1$ the material is vacuum. Cauchy--Schwarz in \eqref{hs:herglotz} then gives a finite pointwise electric-field bound at fixed unit incoming power and fixed $k$, so the arbitrary-amplitude conclusion cannot include that endpoint. \subsection{A local singular projection at positive frequency} \label{hs:local} Corner and edge singularities of positive-coefficient Maxwell transmission problems are established in \cite{costabel1999}. Here we give the needed two-dimensional construction directly. In particular, we do not transfer a bounded-domain boundary condition from that paper into an unproved scattering claim. The angular eigenvalue below controls the radial behavior near the corner. Its exponent lies between zero and one: the corresponding singular model term tends to zero like $r^\alpha$, but its gradient, and hence the electric field, can grow like $r^{\alpha-1}$. The exponent also makes the squared gradient locally integrable. A nonzero coefficient of this angular pattern is essential; the existence of the pattern alone does not say that a particular illumination excites it. \begin{lemma}[A binary checkerboard angular exponent] \label{hs:angular} Let $a(\theta)$ alternate between $1$ and $b$ in successive sectors of angle $\pi/2$. There is a nonzero real periodic $\phi\in H^1(S^1)$ such that \begin{equation} -(a\phi')'=\alpha^2a\phi, \qquad \alpha=\frac4\pi\arctan\sqrt b\in(0,1), \qquad \int_0^{2\pi}a\phi^2\,d\theta=1. \label{hs:angular-eigenproblem} \end{equation} Moreover $\int a\phi=0$ and $\int a|\phi'|^2=\alpha^2$. \end{lemma} \begin{proof} In a sector with constant coefficient $a_0$, the state $(\phi,a_0\phi'/\alpha)^T$ propagates through angle $L$ by \begin{equation} M_{a_0}(L)= \begin{pmatrix} \cos\alpha L & a_0^{-1}\sin\alpha L\\ -a_0\sin\alpha L & \cos\alpha L \end{pmatrix}. \label{hs:transfer} \end{equation} Both state components are continuous at an interface. The half-turn matrix $B=M_b(\pi/2)M_1(\pi/2)$ has determinant one and trace $2\cos^2t-(b+b^{-1})\sin^2t$, where $t=\alpha\pi/2$. With $t=2\arctan\sqrt b$ its trace is $-2$. Choose a nonzero eigenvector of $B$ with eigenvalue $-1$. The resulting angular function is antiperiodic after a half turn and periodic after a full turn. It satisfies the stated weak eigenproblem. Integrating that equation and then testing it against $\phi$ proves the two identities. \end{proof} The coefficient in the next lemma identifies the part of the physical finite-frequency field associated with this angular profile. Its continuity in the local $L^2$ topology is needed later: an approximation of a locally excited field must preserve a nonzero singular coefficient. \begin{lemma}[Continuous finite-frequency singular coefficient] \label{hs:coefficient} Suppose the checkerboard coefficient of Lemma~\ref{hs:angular} holds in $B_R(0)$ and $u\in H^1(B_R)$ weakly satisfies $L_au=0$ there. For almost every $r$, put \begin{equation} U(r)=\int_0^{2\pi}a(\theta)u(r,\theta)\phi(\theta)\,d\theta, \qquad F(r)=\int_0^{2\pi}u(r,\theta)\phi(\theta)\,d\theta. \label{hs:projections} \end{equation} There is a complex number $c(u)$, linear in $u$, such that \begin{align} U(r)&=c(u)r^\alpha+p(r),\notag\\ p(r)&=-\frac{k^2}{2\alpha}\left[ r^\alpha\int_0^r s^{1-\alpha}F(s)\,ds -r^{-\alpha}\int_0^r s^{1+\alpha}F(s)\,ds\right]. \label{hs:exact-projection} \end{align} In particular, \begin{equation} |p(r)|\leq C_{b,\phi,\alpha}k^2r\|u\|_{L^2(B_r)}, \qquad |c(u)|\leq C_{R,k,b,\phi}\|u\|_{L^2(B_R)}. \label{hs:projection-estimates} \end{equation} Thus $c$ is continuous even for the local $L^2$ topology on the space of solutions. If $c(u)\ne0$, ordinary points strictly inside the homogeneous sectors have arbitrarily large $|E|$ as they approach the origin. \end{lemma} \begin{proof} Testing the weak equation with a product of a radial test function and $\phi$ gives the distributional ordinary differential equation \begin{equation} (rU')'-\frac{\alpha^2}{r}U=-k^2rF. \label{hs:radial-ode} \end{equation} Variation of constants gives the particular solution in \eqref{hs:exact-projection}. The two integrals converge because $F\in L^2((0,R),r\,dr)$ and $\alpha<1$. For example, \begin{equation} \int_0^r s^{1-\alpha}|F(s)|\,ds \leq\left(\int_0^r s|F(s)|^2\,ds\right)^{1/2} \frac{r^{1-\alpha}}{\sqrt{2-2\alpha}}. \label{hs:weighted-integral} \end{equation} The second integral is bounded in the same way, with exponent $1+\alpha$ and denominator $\sqrt{2+2\alpha}$. These estimates prove the first bound in \eqref{hs:projection-estimates}. With $I_\mp(r)=\int_0^r s^{1\mp\alpha}F(s)\,ds$, differentiation gives \begin{equation} p'(r)=-\frac{k^2}{2} \left[r^{\alpha-1}I_-(r)+r^{-\alpha-1}I_+(r)\right], \qquad |p'(r)|\leq Ck^2\|u\|_{L^2(B_r)}. \label{hs:particular-derivative} \end{equation} Thus $p'$ is locally square integrable with radial measure. The general remaining homogeneous term is $c r^\alpha+d r^{-\alpha}$. Since $u\in H^1$, $U'$ is square integrable with measure $r\,dr$, which excludes $d\ne0$. To avoid any assumption about point evaluation of an $H^1$ function, solve \eqref{hs:exact-projection} for $c$ and average over $r\in(R/2,R)$ with the probability measure $2\,dr/R$. The resulting formula is \begin{equation} c(u)=\frac2R\int_{R/2}^R \left\{r^{-\alpha}U(r)+\frac{k^2}{2\alpha} \left[\int_0^r s^{1-\alpha}F(s)\,ds -r^{-2\alpha}\int_0^r s^{1+\alpha}F(s)\,ds\right]\right\}\,dr. \label{hs:coefficient-functional} \end{equation} Cauchy--Schwarz and \eqref{hs:weighted-integral} bound this expression by $C\|u\|_{L^2(B_R)}$. This proves continuity without invoking a weighted corner-expansion theorem. For almost every radius, angular integration by parts and \eqref{hs:angular-eigenproblem} yield \begin{equation} \alpha|U(r)|\leq \left(\int_0^{2\pi}a|\partial_\theta u(r,\theta)|^2\,d\theta\right)^{1/2}. \label{hs:gradient-lower-bound} \end{equation} Since $p(r)=o(r)$ and $\alpha<1$, $c\ne0$ implies $|U(r)|\geq |c|r^\alpha/2$ for sufficiently small radii. Consequently the essential supremum of $|\nabla u|$ on each such circle is at least $\alpha|c|r^{\alpha-1}/(2\sqrt{2\pi})$. Here $a\leq1$ was used. Interior analyticity in each constant-coefficient sector permits the large values to be selected at ordinary points, off the interfaces. Formula~\eqref{hs:tm} gives $|E|\geq b|\nabla u|/(\omega\varepsilon_0)$ there. \end{proof} \begin{remark}[Energy and what has actually been expanded] \label{hs:integrability} The static singular scalar field $u_0=r^\alpha\phi$ has \begin{equation} \int_{B_R}a|\nabla u_0|^2\,dx\,dy=\alpha R^{2\alpha}. \label{hs:static-energy} \end{equation} Its electric-field exponent is $\alpha-1$, and its energy is integrable. The finite-frequency fields above have finite local energy by membership in $H^1$. Formula~\eqref{hs:exact-projection} is an exact singular \emph{angular projection}; no unproved pointwise expansion of all vector components is needed for the lower-bound argument. In particular, finite local energy does not bound the peak electric field. \end{remark} \begin{lemma}[A local finite-frequency solution with $c\ne0$] \label{hs:local-excitation} For every fixed $k>0$ there is $R_*>0$, depending on $k$ and $b$, such that for $0\max\{M,2B\}$. By Lemma~\ref{hs:coefficient}, for every $j$ one can choose an ordinary point $x_j\in Q_j$, strictly off the interfaces, with \begin{equation} |E(x_j)|>4S. \label{hs:selected-values} \end{equation} All these choices precede rounding. Round every vertex of every material square \emph{inwards} within a distance $\rho$, leaving all edge portions outside the vertex balls unchanged. A smooth curve joining the shortened straight edges with all endpoint derivatives matched gives $C^\infty$ boundaries. For $\rho>0$ the two opposite inclusions have disjoint closures. The resulting binary coefficient $a_\rho$ differs from $a$ only in the union of finitely many shrinking vertex balls, so $a_\rho\to a$ almost everywhere and retains exactly the original material bounds and exterior background. Choose one sufficiently small $\rho$ so that no changed vertex ball meets any selected point or collar. Lemma~\ref{hs:rounding-convergence} and finiteness of $N$ then permit a further decrease of this same $\rho$ until simultaneously \begin{equation} |E_\rho(x_j)|>3S\quad(1\leq j\leq N), \qquad \sup_{\text{all collars, one-sided}}|E_\rho|3S>M$. The collar values are below $S