\documentclass[final,reqno]{amsart} \usepackage[colorlinks=true,allcolors=blue]{hyperref} \usepackage{amsaddr} \usepackage{amssymb,amsfonts,amsmath,graphicx,amsthm} \usepackage[square,numbers]{natbib} \usepackage{bm} \usepackage{enumerate} \usepackage[dvipsnames]{xcolor} \usepackage{physics} \usepackage{bbold} \usepackage{mathtools} \usepackage[capitalise]{cleveref} \undef\div \DeclareMathOperator{\div}{div} \DeclareMathOperator{\supp}{supp} \DeclareMathOperator{\spec}{spec} \crefname{equation}{}{} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \crefname{lemma}{Lemma}{Lemmas} \Crefname{lemma}{Lemma}{Lemmas} \newtheorem{proposition}[theorem]{Proposition} \crefname{proposition}{Proposition}{Propositions} \Crefname{proposition}{Proposition}{Propositions} \newtheorem*{claim}{Claim} \theoremstyle{remark} \newtheorem{remark}{Remark} \crefname{remark}{Remark}{Remarks} \Crefname{remark}{Remark}{Remarks} \newcommand{\RR}{\mathbb{R}} \newcommand{\CC}{\mathbb{C}} \newcommand{\ZZ}{\mathbb{Z}} \newcommand{\iden}{\mathbb{1}} \renewcommand{\i}{\mathrm{i}} \newcommand{\e}{\mathrm{e}} \newcommand{\dua}[2]{\langle #1, #2 \rangle} \newcommand{\ol}[1]{\overline{#1}} \newcommand{\HC}{\mathcal{H}} \newcommand{\DC}{D} \newcommand{\Cc}{\mathcal{C}} \newcommand{\AF}{\bm{\mathfrak{A}}} \newcommand{\BF}{\bm{\mathfrak{B}}} \newcommand{\Af}{\mathfrak{A}} \newcommand{\Bf}{\mathfrak{B}} \newcommand{\vbf}{\bm{\mathfrak{b}}} \newcommand{\vjp}{\mathbf{j}^{\mathrm{p}}} \newcommand{\vjph}{\mathbf{j}^{\mathrm{ph}}} \newcommand{\vjc}{\mathbf{j}^{\mathrm{c}}} \newcommand{\PF}{\mathrm{PF}} \newcommand{\ph}{\mathrm{ph}} \newcommand{\rad}{\mathrm{rad}} \newcommand{\sol}{\mathrm{sol}} \newcommand{\Fock}{\mathcal{F}} \newcommand{\Hil}{\mathfrak{H}} \let\phi=\varphi \newcommand{\vsigma}{{\bm{\sigma}}} \newcommand{\vzeta}{{\bm{\zeta}}} \newcommand{\vepsilon}{{\bm{\epsilon}}} \newcommand{\vpi}{{\bm{\pi}}} \newcommand{\vOmega}{{\bm{\Omega}}} \renewcommand{\va}{\mathbf{a}} \newcommand{\ve}{\mathbf{e}} \newcommand{\vy}{\mathbf{y}} \newcommand{\vc}{\mathbf{c}} \newcommand{\vk}{\mathbf{k}} \newcommand{\vm}{\mathbf{m}} \newcommand{\vx}{\mathbf{x}} \newcommand{\vj}{\mathbf{j}} \newcommand{\vA}{{\mathbf{A}}} \newcommand{\vB}{{\mathbf{B}}} \newcommand{\vF}{{\mathbf{F}}} \newcommand{\vX}{{\mathbf{X}}} \newcommand{\vD}{{\mathbf{D}}} \undef\vb \newcommand{\vb}{{\mathbf{b}}} \newcommand{\wh}[1]{\widehat{#1}} \DeclareSymbolFont{PFwideaccents}{OMX}{yhex}{m}{n} \DeclareMathAccent{\PFwideaccent}{\mathord}{PFwideaccents}{"62} \newcommand{\widehatfull}[1]{\mathpalette\PFwidehatfull{#1}} \makeatletter \newcommand{\PFwidehatfull}[2]{% \begingroup \sbox0{$\m@th#1#2$}% \sbox2{$\m@th#1\PFwideaccent{\vphantom{#2}\rule{4em}{0pt}}$}% \mathord{\rlap{\resizebox{\wd0}{\ht2}{\usebox2}}\usebox0}% \endgroup} \makeatother \renewcommand{\epsilon}{\varepsilon} \title[Unique continuation for the Pauli--Fierz Hamiltonian]{Unique continuation for the\\ Pauli--Fierz Hamiltonian} \author[M.\,A. Csirik]{Mih\'aly~A.~Csirik} \address{Department of Mathematical Sciences, Chalmers University of Technology and the University of Gothenburg, G\"oteborg, Sweden} \thanks{Email: \href{mailto:csirik@gmail.com}{\texttt{csirik@gmail.com}}} \hyphenation{Carleman} \begin{document} \begin{abstract} We prove a unique continuation property for eigenstates of the Pauli--Fierz Hamiltonian, viewed as functions of the electron configuration with values in the spin--photon space. The vector potential may lie in $L^6_\mathrm{loc}$, with its divergence, curl, and the scalar potential in $L^p_\mathrm{loc}$ for some $p>2$. As an application, we prove a Hohenberg--Kohn theorem at positive coupling strength: the spin-summed ground-state density and combined current density determine the external potentials up to a gauge transformation and an additive constant. \end{abstract} \maketitle \section{Introduction}\label{sec-intro} A differential operator is said to have the unique continuation property if every solution that vanishes on an open set, or to infinite order at a point, vanishes identically. Carleman~\cite{carleman1939probleme} introduced the weighted inequalities now bearing his name to prove such results. For Schr\"odinger operators $-\Delta+V$ on $\RR^n$, Jerison and Kenig~\cite{jerison1985unique} proved unique continuation under the scale-invariant condition $V\in L^{n/2}_\mathrm{loc}$; see~\cite{wolff1993recent} for a review of such sharp results. Singular magnetic fields and nonsmooth coefficients were treated by Kurata~\cite{kurata1997unique} and by Koch and Tataru~\cite{koch2001carleman}. In density-functional theory, the property enters through the Hohenberg--Kohn theorem~\cite{hohenberg1964inhomogeneous}; see~\cite{lewin2023universal} for a review. As Lieb emphasized~\cite{lieb1983density}, for the usual $L^p$ potentials its proof requires that ground states do not vanish on sets of positive measure. For Laplace form-bounded distributional potentials, Corso recently showed that the relevant condition is instead strict positivity of the density quasi-everywhere, and verified it for non-interacting systems~\cite{corso2026maximal}; see also~\cite{corso2025torus,corso2026spinless} in one dimension. For $N$ electrons, the unique continuation theorems of~\cite{jerison1985unique,kurata1997unique,koch2001carleman} must be applied on $\RR^{3N}$, where their conditions deteriorate with $N$, although the physical potentials are sums of one- and two-body terms. For instance, the condition of Jerison and Kenig becomes $V\in L^{3N/2}_\mathrm{loc}$. Garrigue~\cite{garrigue2018unique} removed this dependence. He proved strong unique continuation for many-body Schr\"odinger operators whose one- and two-body potentials lie in $L^p_\mathrm{loc}(\RR^3)$ with $p>2$, for every $N$, and deduced the Hohenberg--Kohn theorem under the same assumptions. In a sequel~\cite{garrigue2020unique} he treated the many-body Pauli Hamiltonian with external magnetic fields. Garrigue's main tool is a Carleman estimate with the singular weight $-\log|\vx|+(-\log|\vx|)^{-\alpha}$ that also controls fractional derivatives, so that each potential is estimated in its own three-dimensional variables. For the Hohenberg--Kohn theorem, unique continuation from sets of positive measure suffices. He obtains it from an argument of de Figueiredo and Gossez~\cite{defigueiredo1992strict}, in which a local energy inequality shows that an eigenfunction vanishes to infinite order at every Lebesgue density point of its zero set.\looseness=-1 Quantum-electrodynamical density-functional theory (QEDFT) extends density-functional theory to electrons coupled to quantized photons. Rajagopal formulated a time-dependent version~\cite{rajagopal1994time}. Ruggenthaler and coworkers developed a time-dependent Kohn--Sham approach to relativistic quantum electrodynamics~\cite{ruggenthaler2011time} and a hierarchy of such theories~\cite{ruggenthaler2014quantum}. In the non-relativistic limit, described by the Pauli--Fierz Hamiltonian, the basic variables are the electronic current density and the expectation value of the quantized vector potential~\cite{ruggenthaler2014quantum}. The ground-state theory~\cite{ruggenthaler2015ground} derives a Hohenberg--Kohn result: the external scalar potential and a transversal external current acting on the photon field are in one-to-one correspondence with the electronic density and the expectation value of the vector potential. As in the original argument~\cite{hohenberg1964inhomogeneous}, the step determining the scalar potential tacitly assumes that the ground state does not vanish on a set of positive measure. In the dipole approximation, Tokatly~\cite{tokatly2013time} used the electronic density, or current density, together with the expectation values of the photon mode coordinates. The time-dependent works~\cite{ruggenthaler2011time,ruggenthaler2014quantum,tokatly2013time} derive the time-dependent analogue of the Hohenberg--Kohn theorem, the Runge--Gross theorem, for their variables. In this paper, we carry out Garrigue's program to derive a unique continuation property for the Pauli--Fierz Hamiltonian. Two new challenges arise: the eigenstates take values in the tensor product of the electron spin space and the photon Fock space, and the quantized vector potential is, at each point, an unbounded operator on the photon Fock space rather than a function. Our Carleman estimate uses the same weight as in~\cite{garrigue2018unique,garrigue2020unique} but keeps the covariant kinetic energy and the radiation Hamiltonian together in the principal part. Under an additional ultraviolet moment condition on the cutoff, \cref{ucp-main-theorem} gives unique continuation for external vector potentials in $L^6_\mathrm{loc}$, with the divergence, curl, and scalar potential in $L^p_\mathrm{loc}$ for some $p>2$. As an application, \cref{thm-hk-nonvanishing} gives a Hohenberg--Kohn theorem at positive coupling. For this theorem we follow~\cite{Q}: unlike in~\cite{ruggenthaler2015ground}, the photon field carries no external current, and the basic variables are the spin-summed density and the combined current density, including the curl of the magnetization density. The proofs are deferred to \cref{sec-proofs}, and the Carleman estimates they rely on are proved in \cref{ucp-carleman}. Counterexamples to the Hohenberg--Kohn theorem at zero coupling, both without and with electron spin, are constructed in \cref{app-zero-coupling}. \medskip \noindent\textbf{Acknowledgements.} The author thanks T.~C. Corso for his useful comments on the manuscript. The author acknowledges funding from the European Research Council through ERC-2021-STG Grant Agreement No.~101041487 \mbox{(REGAL)}. \medskip \noindent\textbf{Lean formalization.} The unique continuation and Hohenberg--Kohn theorems, the Carleman estimates in \cref{ucp-carleman}, and both zero-coupling counterexamples in \cref{app-zero-coupling} are formalized in Lean~4, building on the formalization of~\cite{Q}. The formalization is in the directory \texttt{lean} of the repository of this paper\footnote{\url{https://github.com/xyzw/pf_ucp}}. Standard results from analysis and operator theory enter as common assumptions. \enlargethispage{2\baselineskip} \medskip \noindent\textbf{AI disclosure.} The large language models GPT-6 Astra (OpenAI), and Claude Opus~5.5 and Claude Fable~5.2 (Anthropic), were used throughout the development of this paper and its Lean formalization. The author has reviewed all content and takes full responsibility for the contents of this paper. \section{Setting}\label{sec-setting} We consider $N\geq1$ spin-$1/2$ electrons. We follow the conventions and notations of~\cite{Q}, although that paper only considers spinless electrons. The Hilbert space and spin--photon fiber are \[ \HC_N=\bigwedge^N L^2(\RR^3,\CC^2)\otimes\Fock_\ph, \qquad \mathcal K_N=(\CC^2)^{\otimes N}\otimes\Fock_\ph, \] \[ \Fock_\ph=\Fock_s(\Hil_\ph),\qquad \Hil_\ph=L^2(\RR^3\times\{1,2\}). \] We write $\Omega\in\Fock_\ph$ for the vacuum. The dispersion relation is $\omega(\vk)=|\vk|$, and the radiation Hamiltonian is \[ H_\rad=\sum_{\lambda=1,2}\int_{\RR^3}\omega(\vk)\, a^*(\vk,\lambda)a(\vk,\lambda)\dd\vk, \] where $a^*(\vk,\lambda)$ and $a(\vk,\lambda)$ create and annihilate a photon with momentum $\vk$ and polarization $\lambda$. To define the Hamiltonian, the photon field must be smeared out by an ultraviolet cutoff $\wh\chi$; without it, the quantized vector potential $\AF(\vx)$ is only an operator-valued distribution in $\vx$, not an operator-valued function \cite{lieb2010stability}. Throughout, the cutoff satisfies \begin{equation}\label{cutoff-assumption} \wh\chi(\vk)=\ol{\wh\chi(-\vk)},\qquad \wh\chi\in L^2(\RR^3),\qquad \wh\chi/\sqrt\omega\in L^2(\RR^3). \end{equation} With real transverse polarization vectors $\vepsilon_\lambda$, set \begin{equation}\label{Gx-def} G^\mu_\vx(\vk,\lambda) =\frac{\epsilon_\lambda^\mu(\vk)\wh\chi(-\vk) \e^{-\i\vk\cdot\vx}}{\sqrt2},\qquad \Af^\mu(\vx)=a^*(G^\mu_\vx)+a(G^\mu_\vx). \end{equation} Here, for $f\in\Hil_\ph$, $a^*(f)$ and $a(f)$ are the photon creation and annihilation operators on $\Fock_\ph$, as in~\cite{Q}. The operator $a^*(f)$ is linear and $a(f)$ is antilinear in $f$, they are mutually adjoint, and they satisfy the canonical commutation relations \[ [a(g),a^*(f)]=\dua{g}{f}_{\Hil_\ph},\qquad [a(g),a(f)]=[a^*(g),a^*(f)]=0, \qquad f,g\in\Hil_\ph. \] By \cref{cutoff-assumption}, $G^\mu_\vx\in\Hil_\ph$, so $\Af^\mu(\vx)$ is well-defined. For fixed $e\geq0$, the covariant derivative and external Hamiltonian are \begin{align} \vD_n&=-\i\grad_{\vx_n}-e\AF(\vx_n),\notag\\ H^{v,\vA} &=\sum_{n=1}^N\left[\frac12\bigl(\vsigma_n\cdot(\vD_n+\vA(\vx_n))\bigr)^2 +v(\vx_n)\right] +\sum_{1\leq n0$, \[ \int_{B_r(\vX_0)}\|\Psi(\vX)\|_{\mathcal K_N}^2\dd\vX=\mathcal O(r^m) \qquad(r\downarrow0). \] This notion connects unique continuation with the positivity of densities. The \emph{de Figueiredo--Gossez argument} \cite{defigueiredo1992strict} (\cref{ucp-infinite-order}) uses a local energy estimate to show that an eigenstate vanishes to infinite order at every Lebesgue density point of its zero set. Hence an eigenstate without zeros of infinite order cannot vanish on a set of positive measure. In order to state our result, define the collision-free configuration space via \begin{equation}\label{ucp-regular-configurations} \mathcal U=\{\vX=(\vx_1,\ldots,\vx_N)\in\RR^{3N} :\vx_n\ne\vx_m\text{ if }n\ne m\}. \end{equation} Our first main result shows that a nonzero eigenstate cannot vanish to infinite order at any point of $\mathcal U$. \begin{theorem}[Pauli--Fierz unique continuation]\label[theorem]{ucp-main-theorem} Assume that \begin{equation}\label{ucp-extra-cutoff} \omega\wh\chi\in L^2(\RR^3) \end{equation} and that, for some $p>2$, \[ \begin{gathered} v\in L^p_\mathrm{loc}(\RR^3),\qquad \vA\in L^6_\mathrm{loc}(\RR^3,\RR^3),\\ \div\vA\in L^p_\mathrm{loc}(\RR^3),\qquad \vB\in L^p_\mathrm{loc}(\RR^3,\RR^3). \end{gathered} \] An eigenstate $\Psi$ of $H^{v,\vA}$ that vanishes to infinite order at any point of $\mathcal U$ is identically zero. Moreover, every nonzero eigenstate satisfies \begin{equation*} \|\Psi(\vX)\|_{\mathcal K_N}>0\quad\text{for a.a. }\vX\in\RR^{3N}, \qquad \rho_\Psi(\vx)>0\quad\text{for a.a. }\vx\in\RR^3. \end{equation*} The same conclusions hold without the condition on $\vB$ for the spinless operator $H_{\mathrm{sl}}^{v,\vA}$ on $\HC_N^{\mathrm{sl}}$, with the fiber $\mathcal K_N$ replaced by $\Fock_\ph$. \end{theorem} The proof in \cref{sec-ucp} follows the method of Garrigue~\cite{garrigue2018unique,garrigue2020unique} and relies on the covariant Carleman estimates of \cref{ucp-carleman}. \begin{theorem}[Hohenberg--Kohn] \label[theorem]{thm-hk-nonvanishing} Let $e>0$, and let the cutoff satisfy $\wh\chi\not\equiv0$ and $\omega\wh\chi\in L^2(\RR^3)$. Let $(v_k,\vA_k)$, $k=1,2$, be two potential pairs with $\vB_k=\grad\times\vA_k$, and suppose that, for at least one $k$ and some $p>2$, \begin{equation}\label{ucp-potential-regularity} \begin{gathered} v_k\in L^p_\mathrm{loc}(\RR^3),\qquad \vA_k\in L^6_\mathrm{loc}(\RR^3,\RR^3),\\ \div\vA_k\in L^p_\mathrm{loc}(\RR^3),\qquad \vB_k\in L^p_\mathrm{loc}(\RR^3,\RR^3). \end{gathered} \end{equation} Let $\Psi_k$ be normalized ground states of $H^{v_k,\vA_k}$, with energies $E_k=E(v_k,\vA_k)$, $k=1,2$. Assume that there is a real function $\theta\in W^{1,1}_\mathrm{loc}(\RR^3)$ with $\grad\theta\in L^3(\RR^3,\RR^3)+L^\infty(\RR^3,\RR^3)$ such that \[ \rho_{\Psi_1}=\rho_{\Psi_2}=\rho,\qquad \vjc_{\Psi_2}=\vjc_{\Psi_1}+\rho\grad\theta \quad\text{a.e.} \] Assume moreover that at least one of the following holds: $\wh\chi\ne0$ almost everywhere, or $\rho$ has an exponential moment, that is, for some $\alpha>0$, \begin{equation}\label{density-exponential-moment} \int_{\RR^3}\e^{\alpha|\vx|}\rho(\vx)\dd\vx<\infty. \end{equation} Then \begin{equation}\label{hk-conclusion} \boxed{\vA_2=\vA_1-\grad\theta,\qquad v_1=v_2+c\quad\text{a.e.}, \qquad c=\frac{E_1-E_2}{N}.} \end{equation} The same conclusion holds without the condition on $\vB_k$ for ground states of the spinless operators $H_{\mathrm{sl}}^{v_k,\vA_k}$, with combined current density $\vjc_{\Psi}=\vjp_\Psi+\vjph_\Psi$. \end{theorem} Condition \cref{ucp-potential-regularity} includes molecular Coulomb scalar potentials with locally $L^p$ perturbations for $p>2$. The singularities at the nuclei are allowed, since $|\vx|^{-1}\in L^q_\mathrm{loc}(\RR^3)$ for every $q<3$. In particular, the global class $L^p(\RR^3)+L^\infty(\RR^3)$, $p>2$, satisfies the scalar hypothesis. The hard ultraviolet cutoff $\wh\chi_\Lambda(\vk)=\iden_{\{|\vk|\leq\Lambda\}}/\sqrt{|\vk|}$, $\Lambda>0$, satisfies $\omega\wh\chi_\Lambda\in L^2(\RR^3)$ but vanishes for $|\vk|>\Lambda$, so it is covered only by the exponential-moment alternative \cref{density-exponential-moment}. Exponential decay below an ionization threshold is known for Pauli--Fierz models under appropriate binding hypotheses \cite{griesemer2004exponential}. The vector-potential part of the conclusion, restricted to the support of the density, does not need the local regularity \cref{ucp-potential-regularity}. In fact, for identical combined current densities, \cref{prop-vector-uniqueness} below gives $\rho(\vA_1-\vA_2)=0$ under the remaining hypotheses alone. This is the only step that uses $e>0$, through the photon variations in \cref{prop-cutoff-kernel} below. Note that the unique continuation theorem holds for every $e\geq0$. \begin{remark}[Regularity away from an exceptional set] \label{rem-ucp-open-set} The conclusion of \cref{thm-hk-nonvanishing} remains valid if the conditions in \cref{ucp-potential-regularity} are required only on a connected open set $\mathcal O\subset\RR^3$ with $|\RR^3\setminus\mathcal O|=0$. The divergence and curl conditions are then understood in the sense of distributions on $\mathcal O$. This extension follows from \cref{rem-ucp-domain-extension} by the same proof. It allows singularities that fail the stated local integrability conditions near the excluded set. For example, $v(\vx)=\eta(\vx)|\vx|^{-3/2}$, with a smooth compactly supported cutoff $\eta$ equal to one near the origin, belongs to $L^{3/2}(\RR^3)$ and is locally bounded on $\mathcal O=\RR^3\setminus\{0\}$, but is not in $L^p_\mathrm{loc}(\RR^3)$ for any $p>2$. \end{remark} The positive coupling assumption $e>0$ cannot be dropped from \cref{thm-hk-nonvanishing}, either without or with electron spin. In \cref{app-zero-coupling} we construct, for $N=1$ and $e=0$, families of bounded azimuthal vector potentials $\vA_t$, linear in $t$ and with different magnetic fields for different $t$, together with scalar potentials $v_t$, such that every $H^{v_t,\vA_t}$ has the same nondegenerate ground state, the hydrogen $1s$ orbital times the photon vacuum (\cref{prop-zero-coupling-counterexample,prop-spinful-zero-coupling-counterexample}). Hence the density and combined current density do not depend on $t$, and for a nonzero compact ultraviolet cutoff all hypotheses of \cref{thm-hk-nonvanishing} other than $e>0$ hold. The same examples work at $e>0$ if $\wh\chi=0$. \section{Proofs}\label{sec-proofs} The rest of the paper is devoted to proofs. \subsection{Preliminaries} For the convenience of the reader, we recall that all Hamiltonians $H^{v,\vA}$ with potentials satisfying \cref{potential-spaces} are well defined, with a common form domain. \begin{proposition}[Form domain]\label[proposition]{prop-form-domain} Let $v$ and $\vA$ satisfy \cref{potential-spaces}. The form \begin{align*} q^{v,\vA}[\Phi] &=\frac12\sum_{n=1}^N\|\vsigma_n\cdot\vpi_n\Phi\|^2 +\|H_\rad^{1/2}\Phi\|^2\\ &\quad+\int_{\RR^{3N}}\Bigl(\sum_{n=1}^Nv(\vx_n) +\sum_{1\leq n0$ and $C<\infty$ such that \begin{equation}\label{pauli-form-coercivity} q^{v,\vA}[\Phi]+C\|\Phi\|^2 \geq c\Bigl(\sum_{n=1}^N\|\vpi_n\Phi\|^2 +\|\grad_{\vX}\Phi\|^2+\|H_\rad^{1/2}\Phi\|^2\Bigr), \qquad \Phi\in\mathcal Q. \end{equation} \end{proposition} \begin{proof} For the spinless Hamiltonian, this is the KLMN argument of \cite[Section~3.4]{Q}. The same argument applies with spin. The only additional facts needed are the free Pauli identity $\|\grad_{\vX}\Phi\|^2 =\sum_n\|\vsigma_n\cdot(-\i\grad_{\vx_n})\Phi\|^2$ and $(\vsigma_n\cdot\va)^2=|\va|^2$ for real $\va\in\RR^3$. \end{proof} Next, we derive a consequence of the variational principle: if the ground states of two Hamiltonians share their density and combined current density, then each is also a ground state of the other Hamiltonian. \begin{proposition}[Common ground states]\label[proposition]{prop-common-ground-state} Let $e\geq0$, let $(v_k,\vA_k)$, $k=1,2$, be two potential pairs, and let $\Psi_k$ be normalized ground states of $H^{v_k,\vA_k}$ with energies $E_k=E(v_k,\vA_k)$. Suppose that \[ \rho_{\Psi_1}=\rho_{\Psi_2}=\rho,\qquad \vjc_{\Psi_1}=\vjc_{\Psi_2}=\vjc. \] Then $\Psi_1$ and $\Psi_2$ are ground states of both $H^{v_1,\vA_1}$ and $H^{v_2,\vA_2}$, and \begin{equation*} E_1-E_2=\dua{v_1-v_2+\tfrac12(|\vA_1|^2-|\vA_2|^2)}{\rho}+\dua{\vA_1-\vA_2}{\vjc}. \end{equation*} \end{proposition} This is the analogue of the ``HK1 theorem'' for paramagnetic CDFT \cite[Theorem~1]{penz2023structureII}, with the combined current density in place of the paramagnetic one; \cref{energy-decomposition} supplies the required linearity of the energy in $(v+\frac12|\vA|^2,\vA)$. Note that the ground states may be degenerate. \begin{proof} Let $d=\dua{v_1-v_2+\tfrac12(|\vA_1|^2-|\vA_2|^2)}{\rho}+\dua{\vA_1-\vA_2}{\vjc}$. The variational principle and \cref{energy-decomposition} give \[ E_1\leq\dua{\Psi_2}{H^{v_1,\vA_1}\Psi_2}=E_2+d, \qquad E_2\leq\dua{\Psi_1}{H^{v_2,\vA_2}\Psi_1}=E_1-d. \] The differences $(E_2+d)-E_1$ and $(E_1-d)-E_2$ are nonnegative and sum to zero, so both vanish. Hence $E_1-E_2=d$, $\dua{\Psi_2}{H^{v_1,\vA_1}\Psi_2}=E_1$ and $\dua{\Psi_1}{H^{v_2,\vA_2}\Psi_1}=E_2$. Since $E_k=\inf\spec(H^{v_k,\vA_k})$, the spectral theorem shows that every normalized $\Phi\in\mathcal Q_\PF$ with $\dua{\Phi}{H^{v_k,\vA_k}\Phi}=E_k$ satisfies $H^{v_k,\vA_k}\Phi=E_k\Phi$. This proves both assertions. \end{proof} At positive coupling, we obtain the following restriction on $\rho(\vA_1-\vA_2)$. \begin{proposition}[The ambiguity left by the cutoff] \label[proposition]{prop-cutoff-kernel} Let $e>0$, let $(v_k,\vA_k)$, $k=1,2$, be two potential pairs, and let $\Psi_k$ be normalized ground states of $H^{v_k,\vA_k}$ with energies $E_k=E(v_k,\vA_k)$. Suppose that \[ \rho_{\Psi_1}=\rho_{\Psi_2}=\rho,\qquad \vjc_{\Psi_1}=\vjc_{\Psi_2}=\vjc. \] Then \begin{equation}\label{cutoff-kernel} \boxed{\begin{aligned} &\div_{\vx}\!\left[\rho(\vx) \big(\vA_1(\vx)-\vA_2(\vx)\big)\right]=0,\\ &\wh\chi(-\vk)\, \widehatfull{\Bigl(\rho(\cdot) \bigl(\vA_1(\cdot)-\vA_2(\cdot)\bigr)\Bigr)}(\vk)=0 \quad\text{for a.a. }\vk. \end{aligned}} \end{equation} The first identity is distributional in $\vx$. \end{proposition} \begin{proof} By \cref{prop-common-ground-state}, we may use the same state $\Psi=\Psi_1$ as a ground state of both Hamiltonians. Its density and combined current density are $\rho$ and $\vjc$. Let $\vj_k$ be its physical current density \cref{physical-current-def} for the vector potential $\vA_k$. Then \begin{equation*} \vj_1-\vj_2=\rho(\vA_1-\vA_2). \end{equation*} Both $\vj_k$ belong to $L^1$, since $\rho\in L^1\cap L^{3/2}$ and $\vA_k\in L^3+L^\infty$. For a real function $\theta$ on $\RR^3$, let $U_\theta$ denote multiplication by $\e^{\i\sum_n\theta(\vx_n)}$. For $\theta\in C_c^\infty(\RR^3)$ and $t\in\RR$, the state $U_{t\theta}\Psi$ belongs to $\mathcal Q_\PF$, and its energy expectation in $H^{v_k,\vA_k}$ is \[ \dua{U_{t\theta}\Psi}{H^{v_k,\vA_k}U_{t\theta}\Psi} =E_k+t\int_{\RR^3}\grad\theta\cdot\vj_k +\frac{t^2}{2}\int_{\RR^3}\rho|\grad\theta|^2. \] Stationarity at $t=0$ gives $\div\vj_k=0$. Subtraction proves $\div\big(\rho(\vA_1-\vA_2)\big)=0$. Here the magnetization density is unchanged by the phase, and $\int_{\RR^3}\grad\theta\cdot\grad\times\vm_\Psi=0$. For the photon variation, let $f\in \DC(\omega)\subset\Hil_\ph$, and set \[ W(tf)=\exp\big(t a^*(f)-t a(f)\big),\qquad c_f^\mu(\vx)=2\Re\dua{G^\mu_\vx}{f}. \] The Weyl field shift of \cite[Proposition~A.6(iii)]{Q} gives $W(tf)^*\vpi_n W(tf)=\vpi_n-et\vc_f(\vx_n)$. By \cite[Corollary~A.8]{Q}, $W(tf)$ also preserves the common form domain. Apply \cref{energy-decomposition} to this shift of the external vector potential and add the radiation-energy variation. The Pauli cross term includes $-(et/2)\int_{\RR^3}\vc_f\cdot(\grad\times\vm_\Psi)$, which is already part of the physical current density $\vj_k$. Thus \begin{equation}\label{weyl-energy-polynomial} \begin{split} \dua{W(tf)\Psi}{H^{v_k,\vA_k} W(tf)\Psi} &=E_k+t\left(2\Re\dua{\Psi}{a(\omega f)\Psi} -e\int_{\RR^3}\vc_f\cdot\vj_k\right)\\ &\quad+t^2\left(\|\sqrt\omega f\|^2 +\frac{e^2}{2}\int_{\RR^3}\rho|\vc_f|^2\right). \end{split} \end{equation} All terms are finite. In particular, the annihilation expectation is controlled by $\|\sqrt\omega f\|\|H_\rad^{1/2}\Psi\|$, and $\vc_f$ is bounded. The linear coefficient in \cref{weyl-energy-polynomial} vanishes for both $k$. Since the photon expectation refers to the same $\Psi$, subtraction and $e>0$ give \begin{equation}\label{weyl-orthogonality} \int_{\RR^3}\rho\,\vc_f\cdot(\vA_1-\vA_2)=0\qquad(f\in \DC(\omega)). \end{equation} The Bochner integral \begin{equation}\label{gb-def} g=\sum_{\mu=1}^3\int_{\RR^3}\rho(\vx)\,\big(A_1^\mu(\vx)-A_2^\mu(\vx)\big) G^\mu_\vx\dd\vx\in\Hil_\ph \end{equation} exists, since $\rho(\vA_1-\vA_2)\in L^1$ and $\|G^\mu_\vx\|$ is bounded uniformly in $\vx$. By \cref{weyl-orthogonality}, $2\Re\dua{g}{f}=0$. Using also $\i f$ and density of $\DC(\omega)$ in $\Hil_\ph$ proves $g=0$. Inserting \cref{Gx-def} into \cref{gb-def} and evaluating the Bochner integral pointwise gives \begin{equation*} g(\vk,\lambda) =\frac{(2\pi)^{3/2}}{\sqrt2}\wh\chi(-\vk) \vepsilon_\lambda(\vk)\cdot\widehatfull{\bigl(\rho\,(\vA_1-\vA_2)\bigr)}(\vk). \end{equation*} Thus $\wh\chi(-\vk)P_\sol(\vk)\widehatfull{\bigl(\rho\,(\vA_1-\vA_2)\bigr)}(\vk)=0$ almost everywhere, where $P_\sol(\vk)=\iden-\vk\otimes\vk/|\vk|^2$. On the other hand, $\div\big(\rho(\vA_1-\vA_2)\big)=0$ gives $\vk\cdot\widehatfull{\bigl(\rho\,(\vA_1-\vA_2)\bigr)}(\vk)=0$. The transverse projection is therefore redundant for $\vk\ne0$, proving \cref{cutoff-kernel}. \end{proof} The continuity equation, which gives the first identity in \cref{cutoff-kernel}, and the photon variation, which gives \cref{weyl-orthogonality}, have distinct roles. Write $\vF=\rho(\vA_1-\vA_2)$ and decompose, for $\vk\ne0$, $\wh\vF(\vk)=P_\sol(\vk)\wh\vF(\vk) +(\vk\otimes\vk/|\vk|^2)\wh\vF(\vk)$. The photon variation gives $\wh\chi(-\vk)P_\sol(\vk)\wh\vF(\vk)=0$, so it detects the transverse part only at momenta with $\wh\chi(-\vk)\ne0$. The continuity equation $\div\vF=0$ gives $\vk\cdot\wh\vF(\vk)=0$, which removes the longitudinal part at all momenta. Under an additional condition on either the cutoff or the density, \cref{cutoff-kernel} determines the vector potential on the support of the density. \begin{proposition}[Vector potential on the density support] \label[proposition]{prop-vector-uniqueness} Let $e>0$ and $\wh\chi\not\equiv0$, let $(v_k,\vA_k)$, $k=1,2$, be two potential pairs, and let $\Psi_k$ be normalized ground states of $H^{v_k,\vA_k}$. Suppose that \[ \rho_{\Psi_1}=\rho_{\Psi_2}=\rho,\qquad \vjc_{\Psi_1}=\vjc_{\Psi_2}=\vjc, \] and that at least one of the following conditions holds. \begin{enumerate}[(i)] \item $\wh\chi(\vk)\ne0$ for almost every $\vk\in\RR^3$. \item The density has the exponential moment \cref{density-exponential-moment} for some $\alpha>0$. \end{enumerate} Then $\rho(\vA_1-\vA_2)=0$ almost everywhere. In particular, if $\rho>0$ almost everywhere, then $\vA_1=\vA_2$ almost everywhere. \end{proposition} \begin{proof} By H\"older's inequality, $\rho(\vA_1-\vA_2)\in L^1(\RR^3,\RR^3)$. Under condition (i), \cref{cutoff-kernel} of \cref{prop-cutoff-kernel} gives $\widehatfull{\bigl(\rho\,(\vA_1-\vA_2)\bigr)}=0$ almost everywhere. Fourier uniqueness on $L^1$ then gives $\rho(\vA_1-\vA_2)=0$. Under condition (ii), write $\vA_k=\vA_{k,3}+\vA_{k,\infty}$ with $\vA_{k,3}\in L^3$ and $\vA_{k,\infty}\in L^\infty$ for $k=1,2$. Interpolation gives \begin{equation*} \|\e^{\alpha|\vx|/2}\rho\|_{3/2} \leq\|\e^{\alpha|\vx|}\rho\|_1^{1/2}\|\rho\|_3^{1/2}. \end{equation*} Consequently, \begin{equation}\label{b-exponential-moment} \begin{split} &\int_{\RR^3}\e^{\alpha|\vx|/2}\rho(\vx)|\vA_1(\vx)-\vA_2(\vx)|\dd\vx\\ &\quad\leq\sum_{k=1}^2\left( \|\vA_{k,3}\|_3\|\e^{\alpha|\vx|/2}\rho\|_{3/2} +\|\vA_{k,\infty}\|_\infty\|\e^{\alpha|\vx|/2}\rho\|_1 \right) <\infty. \end{split} \end{equation} Since $|\e^{-\i\vzeta\cdot\vx}|\leq \e^{|\Im\vzeta||\vx|}$, the bound \cref{b-exponential-moment} makes the Fourier integral of each component of $\rho(\vA_1-\vA_2)$ converge absolutely and locally uniformly on $\{\vzeta\in\CC^3:|\Im\vzeta|<\alpha/2\}$. Each component of $\widehatfull{\bigl(\rho\,(\vA_1-\vA_2)\bigr)}$ therefore extends holomorphically to this strip and is real analytic on $\RR^3$. Because $\wh\chi\not\equiv0$ in $L^2$, the set on which $\wh\chi(-\vk)\ne0$ has positive Lebesgue measure. Equation \cref{cutoff-kernel} of \cref{prop-cutoff-kernel} makes every component of $\widehatfull{\bigl(\rho\,(\vA_1-\vA_2)\bigr)}$ zero almost everywhere on this set. A real-analytic function on a connected open set that vanishes on a set of positive measure is identically zero. Applying this fact to real and imaginary parts gives $\widehatfull{\bigl(\rho\,(\vA_1-\vA_2)\bigr)}=0$ on $\RR^3$. Fourier uniqueness gives $\rho(\vA_1-\vA_2)=0$. Division by $\rho$ on $\{\rho>0\}$ proves the last assertion. \end{proof} \begin{remark}[The role of conditions (i) and (ii)]\label{rem-cutoff-kernel} Without condition (i) or (ii) of \cref{prop-vector-uniqueness}, the identities \cref{cutoff-kernel} alone do not force $\rho(\vA_1-\vA_2)=0$. For example, the hard cutoff $\wh\chi_\Lambda$ vanishes for $|\vk|>\Lambda$. Let $\eta\in C_c^\infty(\RR^3)$ be nonzero, real and even, with support in $\{\Lambda+1<|\vk|<\Lambda+2\}$, and let $\vF$ be the vector field with $\wh\vF(\vk)=\i\eta(\vk)\,\vk\times\ve_3$. Then $\vF$ is a nonzero, real, divergence-free Schwartz field with $\wh\chi_\Lambda(-\vk)\wh\vF(\vk)=0$ for all $\vk$, so it satisfies both identities in \cref{cutoff-kernel} in place of $\rho(\vA_1-\vA_2)$. Condition (i) rules this out, since it forces the Fourier transform to vanish almost everywhere. Condition (ii) rules it out as well, since it makes the Fourier transform of $\rho(\vA_1-\vA_2)$ real analytic, whereas $\wh\vF$ is nonzero with compact support. The example concerns only the identities \cref{cutoff-kernel} and is not built from two Pauli--Fierz ground states, so it is not a counterexample at positive coupling. Removing conditions (i) and (ii) would require information beyond \cref{cutoff-kernel}. \end{remark} \subsection{Proof of \texorpdfstring{\cref{ucp-main-theorem}}{Theorem \getrefnumber{ucp-main-theorem}}}\label{sec-ucp} The total scalar potential \[ V(\vX)=\sum_{n=1}^N v(\vx_n)+\sum_{1\leq n0$ and put $u_\epsilon(\vX)=(u_\Phi(\vX)^2+\epsilon^2)^{1/2}-\epsilon$. The map $\xi\mapsto(\|\xi\|_{\Fock_\ph}^2+\epsilon^2)^{1/2}$ is continuously differentiable and Lipschitz on $\Fock_\ph$, so the chain rule for $\Phi\in H^1(\RR^{3N},\Fock_\ph)$ gives $u_\epsilon\in H^1(\RR^{3N})$ with \[ \partial_{x_n^\mu}u_\epsilon(\vX) =\frac{\Re\dua{\Phi(\vX)}{\partial_{x_n^\mu}\Phi(\vX)}_{\Fock_\ph}} {(u_\Phi(\vX)^2+\epsilon^2)^{1/2}}. \] By the definition of $\vpi_n$, \[ \partial_{x_n^\mu}\Phi(\vX)=\i(\pi_n^\mu\Phi)(\vX) -\i\bigl(A^\mu(\vx_n)-e\Af^\mu(\vx_n)\bigr)\Phi(\vX). \] The operator $A^\mu(\vx_n)-e\Af^\mu(\vx_n)$ is symmetric, so its expectation in $\Phi(\vX)$ is real, and therefore \[ \Re\dua{\Phi(\vX)}{\partial_{x_n^\mu}\Phi(\vX)}_{\Fock_\ph} =\Re\dua{\Phi(\vX)}{\i(\pi_n^\mu\Phi)(\vX)}_{\Fock_\ph} \leq u_\Phi(\vX)\|(\pi_n^\mu\Phi)(\vX)\|_{\Fock_\ph}. \] These expectations are defined by \cref{cutoff-assumption}, since $\Phi(\vX)\in\DC(H_\rad^{1/2})$ for almost every $\vX$. Summing over $n$ and $\mu$ gives \[ |\grad_{\vX}u_\epsilon(\vX)|^2 \leq\sum_{n=1}^N\|(\vpi_n\Phi)(\vX)\|_{\Fock_\ph}^2 \qquad\text{for almost every }\vX. \] Since $0\leq u_\Phi(\vX)-u_\epsilon(\vX)\leq\min(u_\Phi(\vX),\epsilon)$, $u_\epsilon\to u_\Phi$ in $L^2$ as $\epsilon\downarrow0$, while $\grad_{\vX}u_\epsilon$ stays bounded in $L^2$. Hence $u_\Phi\in H^1(\RR^{3N})$, and $\grad_{\vX}u_\epsilon$ converges weakly to $\grad_{\vX}u_\Phi$ along a subsequence. The set of $g\in L^2(\RR^{3N},\RR^{3N})$ with $|g(\vX)|^2\leq\sum_n\|(\vpi_n\Phi)(\vX)\|_{\Fock_\ph}^2$ for almost every $\vX$ is convex and closed, hence weakly closed, which proves \cref{ucp-diamagnetic}. \end{proof} The assumption on $v$ and the three-dimensional Sobolev inequality, applied separately to every electronic coordinate and then integrated in the other coordinates, give, with $v_-=\max(-v,0)$ the negative part of $v$, for every $\epsilon>0$, \begin{equation}\label{ucp-negative-potential} \sum_{n=1}^N\int_{\RR^{3N}}v_-(\vx_n)u_\Phi(\vX)^2\dd\vX \leq\epsilon\|\grad_{\vX}u_\Phi\|^2 +C_\epsilon\|\Phi\|^2. \end{equation} To obtain the arbitrarily small coefficient, split the $L^{3/2}$ part of $v_-$ into a bounded function and an $L^{3/2}$ function of arbitrarily small norm. Since the electron--electron interaction and $H_\rad$ are nonnegative, \cref{ucp-diamagnetic,ucp-negative-potential} with $\epsilon=1/4$ imply \begin{equation}\label{ucp-form-lower-bound} \dua{\Phi}{H_{\mathrm{sl}}^{v,\vA}\Phi}\geq\frac14\sum_{n=1}^N\|\vpi_n\Phi\|^2 +\|H_\rad^{1/2}\Phi\|^2-C_0\|\Phi\|^2. \end{equation} Let $\eta$ be as in (ii). Multiplication by $\eta$ preserves $\mathcal Q^{\mathrm{sl}}$. Since $\eta$ is a real scalar function, $\vpi_n(\eta\Psi)=\eta\vpi_n\Psi-\i(\grad_{\vx_n}\eta)\Psi$. This and the eigenvalue equation \cref{ucp-full-weak-equation}, tested with $\eta^2\Psi$, give the IMS localization identity \begin{equation}\label{ucp-ims} \begin{split} \dua{\eta\Psi}{H_{\mathrm{sl}}^{v,\vA}\eta\Psi} &=\Re\dua{\eta^2\Psi}{H_{\mathrm{sl}}^{v,\vA}\Psi} +\frac12\int_{\RR^{3N}}|\grad_{\vX}\eta(\vX)|^2u_\Psi(\vX)^2\dd\vX\\ &=E\|\eta\Psi\|^2 +\frac12\int_{\RR^{3N}}|\grad_{\vX}\eta(\vX)|^2u_\Psi(\vX)^2\dd\vX. \end{split} \end{equation} Combining \cref{ucp-form-lower-bound,ucp-ims} proves \begin{equation}\label{ucp-localized-form-energy} \sum_{n=1}^N\|\vpi_n(\eta\Psi)\|^2 +\|H_\rad^{1/2}\eta\Psi\|^2 +\|\grad_{\vX}(\eta u_\Psi)\|^2 \leq Cr^{-2}\int_{B_{2r}(\vX_0)}u_\Psi(\vX)^2\dd\vX. \end{equation} Since $\eta=1$ and $\grad\eta=0$ on $B_r(\vX_0)$, the product rule above gives $\vpi_n(\eta\Psi)=\vpi_n\Psi$ and $H_\rad^{1/2}(\eta\Psi)=H_\rad^{1/2}\Psi$ on $B_r(\vX_0)$. Restricting the norms on the left of \cref{ucp-localized-form-energy} to $B_r(\vX_0)$ therefore proves \cref{ucp-local-energy}, and dropping the first two terms on the left proves \cref{ucp-local-gradient}. \end{proof} Next, the de Figueiredo--Gossez argument \cite{defigueiredo1992strict}, in the form used by Garrigue \cite{garrigue2020unique}, turns the preceding inequality into infinite-order vanishing: an eigenstate vanishes to infinite order at every Lebesgue density point of its zero set. \begin{proposition}[Zeros of infinite order, cf.\ \cite{defigueiredo1992strict}]\label[proposition]{ucp-infinite-order} Assume \cref{cutoff-assumption,potential-spaces}, let $\Psi\in\mathcal Q_\PF^{\mathrm{sl}}$ be an eigenstate of $H_{\mathrm{sl}}^{v,\vA}$ with eigenvalue $E$, and put \[ \mathcal Z=\{\vX\in\RR^{3N}:\Psi(\vX)=0\}. \] At every Lebesgue density point $\vX_0$ of $\mathcal Z$, $\Psi$ vanishes to infinite order, with the norm of $\Fock_\ph$ in place of that of $\mathcal K_N$. Moreover, for every $m>0$, \begin{equation}\label{ucp-energy-flat} \int_{B_r(\vX_0)}\Bigl(\sum_{n=1}^N\|\vpi_n\Psi(\vX)\|_{\Fock_\ph}^2 +\|H_\rad^{1/2}\Psi(\vX)\|_{\Fock_\ph}^2\Bigr)\dd\vX=\mathcal O(r^m) \qquad(r\downarrow0). \end{equation} In particular, if $|\mathcal Z|>0$, such a point exists. \end{proposition} \begin{proof} Without loss of generality, we can translate $\vX_0$ to the origin and set \[ U(r)=\int_{B_r}u_\Psi(\vX)^2\dd\vX. \] Then $\theta(r):=|B_r\setminus \mathcal Z|/|B_r|\to0$ as $r\downarrow0$. Let $\eta\in C_c^\infty(\RR^{3N},\RR)$ satisfy $0\leq\eta\leq1$, $\eta=1$ on $B_r$, $\supp\eta\subset B_{2r}$, and $|\grad\eta|\leq C_1/r$, as in \cref{ucp-caccioppoli}(ii). H\"older's inequality, the Sobolev inequality on $\RR^{3N}$, and \cref{ucp-local-gradient} of \cref{ucp-caccioppoli} give \begin{align} U(r)&\leq\|\eta u_\Psi\|_2^2\notag\\ &\leq |B_{2r}\setminus \mathcal Z|^{2/(3N)} \|\eta u_\Psi\|_{6N/(3N-2)}^2\notag\\ &\leq C|B_{2r}\setminus \mathcal Z|^{2/(3N)} \|\grad_{\vX}(\eta u_\Psi)\|_2^2\notag\\ &\leq C\theta(2r)^{2/(3N)}U(2r). \label{ucp-density-recurrence} \end{align} Fix $m>0$. Since $\theta(s)\to0$ as $s\downarrow0$, there is $r_m\in(0,1)$ such that $C\theta(s)^{2/(3N)}\leq2^{-m}$ whenever $00$, the set of such points has positive measure, and in particular it is nonempty. \end{proof} The Carleman estimates in \cref{ucp-carleman} are proved for smooth functions. To apply them to a localized eigenstate, we show that the localized state lies in the operator domain of $-\Delta_{\vX}+H_\rad$ and can be approximated there by smooth functions. This step uses the first ultraviolet moment \cref{ucp-extra-cutoff}, which includes the hard ultraviolet cutoff. We repeatedly use the moment inequality: for a nonnegative self-adjoint operator $A$ and $0\leq\theta\leq1$, \begin{equation}\label{moment-inequality} \|A^\theta u\|\leq\|Au\|^\theta\|u\|^{1-\theta}, \qquad u\in\DC(A). \end{equation} It follows from the spectral theorem and H\"older's inequality for the spectral measure of $u$. For $\theta<1$, Young's inequality turns it into $\|A^\theta u\|\leq\epsilon\|Au\|+C_{\epsilon,\theta}\|u\|$ for every $\epsilon>0$. \begin{lemma}[Local operator domain]\label[lemma]{ucp-local-operator-domain} Assume \cref{cutoff-assumption,potential-spaces,ucp-extra-cutoff}. Let $\mathcal O\subset\mathcal U$ be open, with $\mathcal U$ as in \cref{ucp-regular-configurations}, and let $\mathcal O_1\subset\RR^3$ be the union of its electronic coordinate projections. Suppose that, for some $p>2$, \[ v\in L^p_\mathrm{loc}(\mathcal O_1),\qquad \vA\in L^6_\mathrm{loc}(\mathcal O_1,\RR^3),\qquad \div\vA\in L^p_\mathrm{loc}(\mathcal O_1), \] where the divergence is distributional. If $\Psi$ is an eigenstate of $H_{\mathrm{sl}}^{v,\vA}$ with eigenvalue $E$, then for every $\eta\in C_c^\infty(\mathcal O)$, \begin{equation}\label{ucp-local-free-domain} \eta\Psi\in\DC(-\Delta_{\vX})\cap\DC(H_\rad), \end{equation} and \begin{equation}\label{ucp-localized-operator-equation} (H_{\mathrm{sl}}^{v,\vA}-E)(\eta\Psi) =-\i\sum_{n=1}^N(\grad_{\vx_n}\eta)\cdot\vpi_n\Psi -\frac12(\Delta_{\vX}\eta)\Psi. \end{equation} Here $H_{\mathrm{sl}}^{v,\vA}$ acts on $u\in\DC(-\Delta_{\vX})\cap\DC(H_\rad)$ with compact support in $\mathcal O$ by \begin{equation}\label{ucp-local-expression} H_{\mathrm{sl}}^{v,\vA}u =T^0u+\sum_{n=1}^N\Bigl(\vA(\vx_n)\cdot\vD_nu -\frac{\i}{2}(\div\vA)(\vx_n)u+\frac12|\vA(\vx_n)|^2u\Bigr)+Vu, \end{equation} where $T^0=\frac12\sum_{n=1}^N\vD_n^2+H_\rad$, $V$ is the total scalar potential, and each term belongs to $L^2(\RR^{3N},\Fock_\ph)$. \end{lemma} \begin{proof} Fix $\eta\in C_c^\infty(\mathcal O)$. Since $\supp\eta$ is compact, there is an open set $\mathcal V\supset\supp\eta$ with compact closure in $\mathcal O$. Choose $\zeta\in C_c^\infty(\mathcal V)$ equal to one near $\supp\eta$. We replace $H_{\mathrm{sl}}^{v,\vA}$ by an operator $\widetilde H_{\mathrm{sl}}$ on $L^2(\RR^{3N},\Fock_\ph)$ whose potentials are cut off by $\zeta$. It agrees with $H_{\mathrm{sl}}^{v,\vA}$ near $\supp\eta$, and its perturbations are relatively bounded with respect to $-\Delta_{\vX}+H_\rad$. We show in Steps~1--3 that $\DC(\widetilde H_{\mathrm{sl}})=\DC(-\Delta_{\vX}+H_\rad)$, with equivalent graph norms, and in Step~4 that $\eta\Psi\in\DC(\widetilde H_{\mathrm{sl}})$ and \begin{equation}\label{ucp-localized-tilde-equation} \widetilde H_{\mathrm{sl}}(\eta\Psi) =E\eta\Psi-\i\sum_{n=1}^N(\grad_{\vx_n}\eta)\cdot\vpi_n\Psi -\frac12(\Delta_{\vX}\eta)\Psi. \end{equation} Since $\DC(-\Delta_{\vX}+H_\rad)=\DC(-\Delta_{\vX})\cap\DC(H_\rad)$, these two facts prove \cref{ucp-local-free-domain}. By Step~1, $\widetilde H_{\mathrm{sl}}u$ is given by \cref{ucp-local-expression} for $u\in\DC(\widetilde H_{\mathrm{sl}})$ supported where $\zeta=1$, so \cref{ucp-localized-tilde-equation} also proves \cref{ucp-localized-operator-equation}. Since $\eta$ is arbitrary, Steps~1--3, applied with a cutoff equal to one near the support of a given $u$, also show that each term of \cref{ucp-local-expression} belongs to $L^2$. \noindent\textbf{Step 1: The localized Hamiltonian.} Let $K_n\subset\mathcal O_1$ be the compact projection of $\ol{\mathcal V}$ onto the coordinate $\vx_n$, and put $v_n=v\iden_{K_n}\in L^p(\RR^3)$. Since $\ol{\mathcal V}$ is a compact subset of $\mathcal U$, the electron--electron interaction is bounded on $\mathcal V$. Hence the total scalar potential $V$ from the beginning of this subsection satisfies \begin{equation*} V(\vX)=\sum_{n=1}^N v_n(\vx_n)+W(\vX),\qquad W\in L^\infty(\mathcal V), \end{equation*} for $\vX\in\mathcal V$. Let $\widetilde V$ be the extension of $\zeta V$ by zero. It is a real multiplication operator, equal to $V$ near $\supp\eta$. For $n=1,\ldots,N$, let $P_n\subset\mathcal O_1$ be the compact projection of $\supp\zeta$ onto the coordinate $\vx_n$. Since $\vA\in L^6_\mathrm{loc}(\mathcal O_1,\RR^3)$, we have $\iden_{P_n}\vA\in L^6(\RR^3,\RR^3)$. Put \[ \widetilde\vA_n(\vX)=\zeta(\vX)\vA(\vx_n),\qquad n=1,\ldots,N, \] with $\widetilde\vA_n(\vX)=0$ for $\vX\notin\supp\zeta$. Then $\widetilde\vA_n(\vX)=\vA(\vx_n)$ for $\vX$ near $\supp\eta$, and $\widetilde\vA_n(\vX)=\zeta(\vX)(\iden_{P_n}\vA)(\vx_n)$ for all $\vX$. The distributional product rule gives \[ \grad_{\vx_n}\cdot\widetilde\vA_n =\zeta\,(\div\vA)(\vx_n)+\grad_{\vx_n}\zeta\cdot\vA(\vx_n). \] The first term is bounded in modulus by a one-body $L^p$ multiplier, and the second by a one-body $L^6$ multiplier. Also $|\widetilde\vA_n|^2$ is bounded in modulus by a one-body $L^3$ multiplier. The localized Hamiltonian is \begin{equation}\label{ucp-localized-hamiltonian} \widetilde H_{\mathrm{sl}} =\frac12\sum_{n=1}^N\bigl(\vD_n+\widetilde\vA_n\bigr)^2+H_\rad+\widetilde V =T^0+\widetilde R+\widetilde V, \end{equation} with $T^0$ as in \cref{ucp-local-expression}, and \[ \widetilde R =\sum_{n=1}^N\Bigl(\widetilde\vA_n\cdot\vD_n -\frac{\i}{2}\grad_{\vx_n}\cdot\widetilde\vA_n +\frac12|\widetilde\vA_n|^2\Bigr). \] Note that $\widetilde H_{\mathrm{sl}}$ is not of the form $H_{\mathrm{sl}}^{v',\vA'}$, since $\widetilde\vA_n$ and $\widetilde V$ depend on all electronic coordinates. Steps~2 and~3 realize \cref{ucp-localized-hamiltonian} as a self-adjoint operator. \noindent\textbf{Step 2: The free operator $T^0$.} Theorem 6 of \cite{hasler2008selfadjoint} gives $\DC(T^0)=\DC(-\Delta_{\vX}+H_\rad)$ and states that any operator core for $-\Delta_{\vX}+H_\rad$ is an operator core for $T^0$. Theorem 7 there shows that $T^0$, defined through its closed form, acts on this domain by the expanded expression $\frac12\sum_n\vD_n^2+H_\rad$. To check the hypotheses of both theorems, write $\Hil_\omega$ for the one-photon space equipped with the norm \[ \|g\|_{\Hil_\omega}^2=\|g\|^2+\|g/\sqrt\omega\|^2. \] Our coupling functions lie uniformly in $\Hil_\omega$. Since $\sqrt\omega\wh\chi\in L^2$ follows from $\wh\chi,\omega\wh\chi\in L^2$, the extra assumption \cref{ucp-extra-cutoff} puts $\omega G^\mu_\vx$ and all first electronic derivatives of $G^\mu_\vx$ uniformly in this space. Moreover, $\sum_{\mu=1}^3\partial_{x^\mu}G^\mu_\vx=0$, since $\vk\cdot\vepsilon_\lambda(\vk)=0$; that is, $\div\AF=0$. These are Hypothesis (G) of Theorem 6 and the hypotheses of Theorem 7 in that reference. The different positive coefficient of the electronic kinetic energy is accommodated by a linear rescaling of electronic coordinates. The coupling strength is arbitrary. In particular, $T^0$ and $-\Delta_{\vX}+H_\rad$ have equivalent graph norms. Put \begin{equation}\label{ucp-free-comparison-operator} L=-\Delta_{\vX}+H_\rad+1. \end{equation} \noindent\textbf{Step 3: Relative bounds and the domain of $\widetilde H_{\mathrm{sl}}$.} Let $b\in L^q(\RR^3)$ with $q>2$ and put $s=3/q$. Then the three-dimensional Sobolev inequality in the $n$th electronic coordinate gives \begin{equation}\label{ucp-coordinate-multiplier} \begin{split} \|b(\vx_n)\Phi\|_2 &\leq C_q\|b\|_q \|(-\Delta_{\vx_n})^{s/2}\Phi\|_2\\ &\leq C_q\|b\|_q \|(-\Delta_{\vX})^{s/2}\Phi\|_2. \end{split} \end{equation} The Sobolev inequality holds for Hilbert-space-valued functions. We apply it with values in $\Fock_\ph$ at fixed values of the remaining electronic coordinates and then integrate over these coordinates. For $q=6$, apply this estimate to electronic derivatives and to $(H_\rad+1)^{1/2}\Phi$. The electronic and radiation spectral multipliers commute, and \[ \|(-\Delta_{\vx_n})^{1/4}(H_\rad+1)^{1/2}\Phi\|_2 \leq\|L^{3/4}\Phi\|_2. \] Together with the pointwise field bound \cref{ucp-field-form-bound}, this yields, for $\vb\in L^6(\RR^3,\RR^3)$, \begin{equation}\label{ucp-vector-multiplier} \|\vb(\vx_n)\cdot\vD_n\Phi\|_2 \leq C\|\vb\|_6\|L^{3/4}\Phi\|_2. \end{equation} Since $\widetilde\vA_n=\zeta\,(\iden_{P_n}\vA)(\vx_n)$ and $\zeta$ is bounded, \cref{ucp-vector-multiplier} with $\vb=\iden_{P_n}\vA$ bounds $\widetilde\vA_n\cdot\vD_n$. The scalar terms are bounded by $\|L^{s/2}\Phi\|_2$ with $s/2<3/4$, and the first-order term by $\|L^{3/4}\Phi\|_2$. Since these powers of $L$ are below one, the moment inequality \cref{moment-inequality} gives, for every $\epsilon>0$, \[ \|\widetilde R\Phi\|_2+\|\widetilde V\Phi\|_2 \leq\epsilon\|L\Phi\|_2 +C_\epsilon\|\Phi\|_2. \] By Step~2, both perturbations are infinitesimally bounded with respect to $T^0$. The distributional product rule, justified first by smooth approximation and then by these bounds, shows that $\widetilde R$ is symmetric on $\DC(T^0)$ and agrees there with the difference of the kinetic forms. The Kato--Rellich theorem \cite[Theorem~V.4.3]{kato1995perturbation} identifies $T^0+\widetilde R+\widetilde V$ with the form realization of \cref{ucp-localized-hamiltonian} and gives \begin{equation}\label{ucp-truncated-domain} \DC(\widetilde H_{\mathrm{sl}})=\DC(-\Delta_{\vX}+H_\rad), \end{equation} again with equivalent graph norms and the same operator cores. \noindent\textbf{Step 4: The localized eigenvalue equation.} Since $\zeta=1$ near $\supp\eta$, the form of $\widetilde H_{\mathrm{sl}}$ agrees with that of $H_{\mathrm{sl}}^{v,\vA}$ on pairs $(\Phi,\eta\Psi)$ with $\Phi\in\mathcal Q^{\mathrm{sl}}$. The weak eigenvalue equation, extended to full-space tests as in \cref{ucp-full-weak-equation}, is tested with $\eta\Phi$. The product rule $\vpi_n(\eta\Psi)=\eta\vpi_n\Psi-\i(\grad_{\vx_n}\eta)\Psi$ then gives \cref{ucp-localized-tilde-equation} in the form sense. The right-hand side of \cref{ucp-localized-tilde-equation} is in $L^2$, since $\Psi$ is in the form domain. The first representation theorem for closed forms \cite[Theorem~VI.2.1]{kato1995perturbation} puts $\eta\Psi$ in $\DC(\widetilde H_{\mathrm{sl}})$. \end{proof} Let $\Fock_{\mathrm{fin}}\subset\Fock_\ph$ be the span of $\Omega$ and of the vectors $a^*(h_1)\cdots a^*(h_m)\Omega$ with $m\geq1$ and $h_1,\ldots,h_m\in\Hil_\ph$ of bounded support, and let $\mathcal C\subset L^2(\RR^{3N},\Fock_\ph)$ be the span of the functions $\vX\mapsto\phi(\vX)\xi$ with $\phi\in C_c^\infty(\RR^{3N})$ and $\xi\in\Fock_{\mathrm{fin}}$. \begin{lemma}[Approximation with prescribed support]\label[lemma]{ucp-core-lemma} Let $w$ be a compactly supported element of $\DC(-\Delta_{\vX})\cap\DC(H_\rad)$, and let $\mathcal G\subset\RR^{3N}$ be an open neighborhood of $\supp w$. Then there are $w_j\in\mathcal C$ with $\supp w_j\subset\mathcal G$ and \begin{equation}\label{ucp-core-approximation} \|w_j-w\|+\|\Delta_{\vX}(w_j-w)\|+\|H_\rad(w_j-w)\|\to0 \qquad(j\to\infty). \end{equation} The same holds for $w\in L^2(\RR^{3N},\mathcal K_N)$, with $\mathcal C$ replaced by $\mathcal C\otimes(\CC^2)^{\otimes N}$. \end{lemma} \begin{proof} Let $L=-\Delta_{\vX}+H_\rad+1$ as in \cref{ucp-free-comparison-operator}. Since $-\Delta_{\vX}$ and $H_\rad$ are commuting nonnegative self-adjoint operators, $\DC(L)=\DC(-\Delta_{\vX})\cap\DC(H_\rad)$, and the graph norm of $L$ is equivalent to the norm in \cref{ucp-core-approximation}. The set $\mathcal C$ is the algebraic tensor product of the core $C_c^\infty(\RR^{3N})$ of $-\Delta_{\vX}$ and the core $\Fock_{\mathrm{fin}}$ of $H_\rad$, and hence an operator core for $L$. Choose $\Phi_j\in\mathcal C$ with $\Phi_j\to w$ in the graph norm of $L$, and let $\zeta\in C_c^\infty(\mathcal G)$ be equal to one near $\supp w$. Then $w_j=\zeta\Phi_j\in\mathcal C$, $\supp w_j\subset\mathcal G$, and \[ Lw_j=\zeta L\Phi_j-2\grad_{\vX}\zeta\cdot\grad_{\vX}\Phi_j -(\Delta_{\vX}\zeta)\Phi_j. \] Since $\|\grad_{\vX}\Phi\|^2\leq\|\Phi\|\|L\Phi\|$, also $\grad_{\vX}\Phi_j\to\grad_{\vX}w$, and therefore $w_j\to\zeta w=w$ in the graph norm of $L$. This proves \cref{ucp-core-approximation}. The version with spin follows by applying this argument to each spin component, since $-\Delta_{\vX}$ and $H_\rad$ act componentwise on $(\CC^2)^{\otimes N}$. \end{proof} \begin{proof}[Proof of \cref{ucp-main-theorem} in the spinless case] Suppose first that $\Psi$ vanishes to infinite order at $\vX_0\in\mathcal U$. Without loss of generality, we can translate this point to the origin. Fix $\alpha>0$, and let \[ r=|\vX|,\qquad t=-\log r,\qquad f(t)=t+t^{-\alpha},\qquad \beta(t)=\ddot f(t)=\alpha(\alpha+1)t^{-\alpha-2}. \] Choose $R>0$ with $\ol{B_R}\subset\mathcal U$. On $B_R$ the electron--electron interaction is bounded, so $V-E$ is a sum of one-body $L^p$ terms and a bounded term. Since $H_{\mathrm{sl}}^{v,\vA}-E=T^{\vA}+(V-E)$ with $T^{\vA}$ as in \cref{ucp-carleman}, \cref{ucp-scalar-perturbation} applies with $W=V-E$. After decreasing $R$, it gives $\tau_0\geq1$ and $C<\infty$ such that, for $\tau\geq\tau_0$ and every $u$ in the local operator domain of $H_{\mathrm{sl}}^{v,\vA}$ with compact support in $B_R\setminus\{0\}$, \begin{equation}\label{ucp-spinless-carleman} \tau^3\int_{B_R}\beta(t)\e^{2\tau f(t)}r^{-2}\|u(\vX)\|_{\Fock_\ph}^2\dd\vX \leq C\int_{B_R}\e^{2\tau f(t)}r^2 \|(H_{\mathrm{sl}}^{v,\vA}-E)u(\vX)\|_{\Fock_\ph}^2\dd\vX. \end{equation} This is \cref{ucp-scalar-carleman} of \cref{ucp-scalar-perturbation}, keeping only the first term on its left-hand side. By \cref{ucp-caccioppoli}, the covariant derivatives of $\Psi$ also vanish to infinite order at the origin, in the sense that $\int_{B_s}\sum_n\|\vpi_n\Psi(\vX)\|_{\Fock_\ph}^2\dd\vX=\mathcal O(s^m)$ as $s\downarrow0$ for every $m>0$. Choose $\zeta\in C_c^\infty(B_R)$ with $\zeta=1$ on $B_{R/2}$, and choose a radial cutoff $\eta_\epsilon$ equal to zero on $B_\epsilon$ and one outside $B_{2\epsilon}$, with $|\grad\eta_\epsilon|\leq C/\epsilon$ and $|\Delta\eta_\epsilon|\leq C/\epsilon^2$. For $0<\epsilonf(-\log(R/2))$, and $\beta(-\log r)r^{-2}$ has a positive lower bound. Letting $\tau\to\infty$ in \cref{ucp-outer-error} proves $\Psi=0$ on $B_{R/4}$. This local vanishing propagates throughout $\mathcal U$. The set $\mathcal U$ is open, connected and of full measure. Indeed, paths in $\RR^3$ can avoid any finite set of points. Moving the electron positions one at a time, through distinct auxiliary positions if necessary, therefore connects any two configurations in $\mathcal U$. Its complement in $\RR^{3N}$ consists of diagonals of codimension three. To justify the propagation, let $\mathcal Z_\mathrm{open}$ be the set of points having a neighborhood on which $\Psi=0$ almost everywhere. It is nonempty and open. On each sufficiently small ball compactly contained in $\mathcal U$, the preceding argument gives a vanishing radius that is uniform in the center. If a point belongs to the relative closure of $\mathcal Z_\mathrm{open}$, choose a point of $\mathcal Z_\mathrm{open}$ closer than half this uniform radius. Infinite-order vanishing at that interior point then supplies a zero neighborhood of the closure point. Thus $\mathcal Z_\mathrm{open}$ is also relatively closed. Connectedness of $\mathcal U$ implies $\mathcal Z_\mathrm{open}=\mathcal U$. Since its complement has measure zero, $\Psi=0$ almost everywhere on the full configuration space $\RR^{3N}$. Now suppose that the configuration-space zero set of a nonzero $\Psi$ has positive measure. It has a Lebesgue density point in $\mathcal U$, because $\mathcal U$ has full measure. Proposition~\ref{ucp-infinite-order} gives infinite-order vanishing there, contradicting the conclusion just proved. This establishes the nonvanishing assertion of \cref{ucp-main-theorem} for the spinless Hamiltonian. If $\rho_\Psi$ vanished on a set $\Sigma\subset\RR^3$ of positive measure, then by the definition of $\rho_\Psi$ and Tonelli's theorem, $\Psi$ would vanish almost everywhere on $\Sigma\times\RR^{3(N-1)}$, a set of positive measure, contradicting the nonvanishing just proved. This gives the density-positivity assertion. \end{proof} With spin, the local kinetic energy estimates take the following form. \begin{lemma}[Local kinetic energy estimates with spin]\label[lemma]{ucp-caccioppoli-spin} Assume \cref{cutoff-assumption,potential-spaces}, and let $\Psi\in\mathcal Q_\PF$ be an eigenstate of $H^{v,\vA}$ with eigenvalue $E$. There is a constant $C$ such that, for every $\vX_0\in\RR^{3N}$ and $00$. Now let $\Psi$ vanish to infinite order at a point of $\mathcal U$, translated to the origin. Choose $R>0$ with $\ol{B_R}\subset\mathcal U$. On $B_R$, the total scalar potential $V$ is the sum of the one-body terms $v(\vx_n)$ and the bounded electron--electron interaction. Let $H^{V,\vA}=T^{\vA}+V+S_{\mathrm{ext}}+S_{\mathrm{ph}}$. We apply \cref{ucp-pauli-carleman} with $\vX_0=0$, $v_n=v$, $V_b$ the electron--electron interaction and $E$ the eigenvalue, and decrease $R$ so that $R\leq r_0$. Keeping only the first term on the left of \cref{pauli-carleman} of \cref{ucp-pauli-carleman}, we obtain $\tau_0\geq1$ and $C<\infty$ such that, for $\tau\geq\tau_0$ and every $u\in\DC(-\Delta_{\vX})\cap\DC(H_\rad)$ with compact support in $B_R\setminus\{0\}$, \[ \tau^3\int_{B_R}\beta(t)\e^{2\tau f(t)}r^{-2} \|u(\vX)\|_{\mathcal K_N}^2\dd\vX \leq C\int_{B_R}\e^{2\tau f(t)}r^2 \|(H^{V,\vA}-E)u(\vX)\|_{\mathcal K_N}^2\dd\vX. \] Let $\zeta$, $\eta_\epsilon$ and $u_\epsilon=\zeta\eta_\epsilon\Psi$ be as in the spinless case. By \cref{ucp-local-expression} of \cref{ucp-local-operator-domain}, $H^{v,\vA}u=H^{V,\vA}u$ for $u\in\DC(-\Delta_{\vX})\cap\DC(H_\rad)$ with compact support in $B_R$. Thus \cref{ucp-local-operator-domain-spin}, applied with $\mathcal O=B_R$ and $\eta=\zeta\eta_\epsilon$, gives $u_\epsilon\in\DC(-\Delta_{\vX})\cap\DC(H_\rad)$ and \[ (H^{V,\vA}-E)u_\epsilon =-\i\sum_{n=1}^N\grad_{\vx_n}(\zeta\eta_\epsilon)\cdot\vpi_n\Psi -\frac12\Delta_{\vX}(\zeta\eta_\epsilon)\Psi. \] Inserting $u=u_\epsilon$ in the Carleman estimate above gives, for $0<\epsilon0 \quad\text{for a.a. }\vX\in\RR^{3N},\qquad \rho(\vx)>0\quad\text{for a.a. }\vx\in\RR^3. \end{equation} \Cref{prop-vector-uniqueness} now gives $\vA_1=\vA_2$. Subtracting the weak eigenvalue equations cancels the kinetic, interaction, and radiation forms, leaving \begin{equation}\label{scalar-multiplier-equation} M(\vX)\Psi_1(\vX)=0,\qquad M(\vX)=\sum_{n=1}^N\big(v_1(\vx_n)-v_2(\vx_n)\big)-(E_1-E_2). \end{equation} The equation $M\Psi_1=0$ holds in the sense of distributions. The weak eigenvalue equations only provide antisymmetric test functions $\Phi\in\mathcal Q_\PF$. Since the multiplier $M$ is symmetric under permutations, $M\Psi_1$ is antisymmetric, so $\dua{\Phi}{M\Psi_1}=\dua{\Pi_-\Phi}{M\Psi_1}$ for every test function $\Phi$. Hence \cref{scalar-multiplier-equation} holds against all test functions. The products $\big(v_1(\vx_n)-v_2(\vx_n)\big)\Psi_1(\vX)$ are locally Bochner integrable. Indeed, on each bounded configuration-space set $K$, Cauchy--Schwarz gives \[ \begin{aligned} &\int_K |v_1(\vx_n)-v_2(\vx_n)|\|\Psi_1(\vX)\|_{\mathcal K_N}\dd\vX\\ &\quad\leq\left(\int_K|v_1(\vx_n)-v_2(\vx_n)|\|\Psi_1(\vX)\|_{\mathcal K_N}^2\dd\vX\right)^{1/2}\\ &\qquad\times\left(\int_K|v_1(\vx_n)-v_2(\vx_n)|\dd\vX\right)^{1/2}<\infty, \end{aligned} \] by scalar form boundedness and local integrability of $v_1-v_2$. Consequently the multiplier equation holds almost everywhere. Using \cref{ucp-joint-nonvanishing}, it reduces to \[ \sum_{n=1}^N \big(v_1(\vx_n)-v_2(\vx_n)\big)=E_1-E_2 \qquad\text{for a.a. }\vX=(\vx_1,\ldots,\vx_N)\in\RR^{3N}. \] Write $f=v_1-v_2\in L^1_\mathrm{loc}(\RR^3)$ and fix a ball $B\subset\RR^3$. Averaging the identity over $(\vx_2,\ldots,\vx_N)\in B^{N-1}$ and using Fubini's theorem gives \[ f(\vx_1)=E_1-E_2-\frac{N-1}{|B|}\int_B f(\vy)\dd\vy \qquad\text{for a.a. }\vx_1\in\RR^3. \] Thus $f=c$ almost everywhere for a constant $c$, and substituting this into the identity gives $Nc=E_1-E_2$. Now let $\theta$ be any function satisfying the hypotheses of the theorem, so that \[ \rho_{\Psi_1}=\rho_{\Psi_2}=\rho,\qquad \vjc_{\Psi_2}=\vjc_{\Psi_1}+\rho\grad\theta. \] As in the proof of \cref{prop-cutoff-kernel}, $U_\theta$ denotes multiplication by $\e^{\i\sum_n\theta(\vx_n)}$. The weak chain rule and the multiplier bounds for $L^3+L^\infty$ show that $U_\theta$ maps the fermionic form domain onto itself. The covariant derivatives satisfy \[ \bigl(\vD_n+\vA(\vx_n)-\grad\theta(\vx_n)\bigr)U_\theta\Psi =U_\theta\bigl(\vD_n+\vA(\vx_n)\bigr)\Psi. \] The phase also commutes with the Pauli matrices, so this identity holds after contraction with $\vsigma_n$. Since $U_\theta$ commutes with the scalar, interaction, and radiation terms, the closed forms give \[ H^{v,\vA-\grad\theta} =U_\theta H^{v,\vA}U_\theta^{-1}. \] Thus gauge changes preserve ground-state energies and transform the data by $\rho_{U_\theta\Psi}=\rho_\Psi$ and $\vjc_{U_\theta\Psi}=\vjc_\Psi+\rho_\Psi\grad\theta$. The magnetization density is unchanged by the phase. We now reduce to the case $\theta=0$ by gauge transforming one of the two pairs. The pair satisfying \cref{ucp-potential-regularity} is left unchanged, since that condition is not gauge invariant: $\div(\vA_k\pm\grad\theta)=\div\vA_k\pm\Delta\theta$ need not lie in $L^p_\mathrm{loc}$. Suppose first that the first potential pair satisfies \cref{ucp-potential-regularity}. We apply the case $\theta=0$ to the pairs $(v_1,\vA_1)$ and $(v_2,\vA_2+\grad\theta)$, with the states $\Psi_1$ and $U_{-\theta}\Psi_2$. Its hypotheses hold for these data. First, since $\grad\theta\in L^3+L^\infty$, the pair $(v_2,\vA_2+\grad\theta)$ satisfies \cref{potential-spaces}. Second, the conjugation identity above with $-\theta$ in place of $\theta$ gives $H^{v_2,\vA_2+\grad\theta}=U_{-\theta}H^{v_2,\vA_2}U_{-\theta}^{-1}$. Hence $U_{-\theta}\Psi_2$ is a normalized ground state of $H^{v_2,\vA_2+\grad\theta}$, and its energy is $E(v_2,\vA_2+\grad\theta)=E_2$. Third, by the transformation rule above, \[ \rho_{U_{-\theta}\Psi_2}=\rho=\rho_{\Psi_1},\qquad \vjc_{U_{-\theta}\Psi_2}=\vjc_{\Psi_2}-\rho\grad\theta=\vjc_{\Psi_1}, \] so the two states have the same density and combined current density. Fourth, the first pair is unchanged and still satisfies \cref{ucp-potential-regularity}. Finally, $e$, $\wh\chi$, and the common density $\rho$ are unchanged, so whichever cutoff alternative holds for the original data also holds for the new data. The case $\theta=0$ therefore gives $\vA_1=\vA_2+\grad\theta$ and $v_1-v_2=(E_1-E_2)/N$ almost everywhere. If instead the second pair supplies the local regularity, we apply the case $\theta=0$ to the pairs $(v_1,\vA_1-\grad\theta)$ and $(v_2,\vA_2)$, with the states $U_\theta\Psi_1$ and $\Psi_2$. The same checks apply. The pair $(v_1,\vA_1-\grad\theta)$ satisfies \cref{potential-spaces}, and $U_\theta\Psi_1$ is a normalized ground state of $H^{v_1,\vA_1-\grad\theta}=U_\theta H^{v_1,\vA_1}U_\theta^{-1}$ with energy $E_1$. Moreover, \[ \rho_{U_\theta\Psi_1}=\rho=\rho_{\Psi_2},\qquad \vjc_{U_\theta\Psi_1}=\vjc_{\Psi_1}+\rho\grad\theta=\vjc_{\Psi_2}. \] The unchanged second pair satisfies \cref{ucp-potential-regularity}, and the cutoff alternative is preserved. The case $\theta=0$ gives $\vA_1-\grad\theta=\vA_2$ and the same scalar constant. This proves \cref{hk-conclusion}. \appendix \section{Covariant Pauli--Fierz Carleman estimates}\label[appendix]{ucp-carleman} Define the auxiliary spinless operator on $L^2(\RR^{3N},\Fock_\ph)$ by \[ T^{\vA}=\frac12\sum_{n=1}^N\vpi_n^2+H_\rad =\frac12\sum_{n=1}^N\big(\vD_n+\vA(\vx_n)\big)^2+H_\rad. \] We first prove a Carleman estimate for $T^0$, keeping the covariant derivatives and the radiation energy in the principal operator, so that the left-hand side controls $\vD_nu$ rather than $\grad u$. This covariant form is essential for two reasons. First, $\AF(\vx)$ is an unbounded operator on $\Fock_\ph$, so the first-order term $e\AF(\vx_n)\cdot\grad_{\vx_n}$ cannot be treated as a perturbation of a Carleman estimate for the Laplacian. It is controlled only through $H_\rad^{1/2}$. Second, in a local radial gauge the commutators in the Carleman computation involve the quantized field only through its curvature, the quantized magnetic field. This is controlled by the first ultraviolet moment \cref{ucp-extra-cutoff} and is small near the center of the weight. Fractional Carleman estimates in the mixed electronic and radiation norms of $L=-\Delta_{\vX}+H_\rad+1$ then permit the external vector potential in $L^6_\mathrm{loc}$ as a first-order perturbation. For an electronic configuration $\vX_0$ and $\alpha>0$, we use the weights \[ r=|\vX-\vX_0|,\qquad t=-\log r,\qquad f(t)=t+t^{-\alpha},\qquad \beta(t)=\ddot f(t)=\alpha(\alpha+1)t^{-\alpha-2}. \] For $00$ and $u\in\DC(T^0)$ with compact support in $B_{r_0}(\vX_0)\setminus\{\vX_0\}$, put \[ \mathcal T^0(u)=\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|T^0u(\vX)\|_{\Fock_\ph}^2\dd\vX. \] We suppress the dependence of $\mathcal T^0(u)$ on $\vX_0$, $\alpha$, $r_0$ and $\tau$. \begin{lemma}[A Carleman estimate with $\vA=0$] \label[lemma]{lem-pf-free-carleman} Assume \cref{cutoff-assumption} and $\omega\wh\chi\in L^2(\RR^3)$. Fix an electronic configuration $\vX_0$ and $\alpha>0$, with $r$, $t$, $f$ and $\beta$ as above. There are $r_0\in(0,\e^{-2})$, $\tau_0\geq1$ and $C<\infty$ such that, for $\tau\geq\tau_0$ and every $u\in\DC(T^0)$ with compact support in $B_{r_0}(\vX_0)\setminus\{\vX_0\}$, \begin{align} &\tau^3\int_{B_{r_0}(\vX_0)}\beta(t)\e^{2\tau f(t)}r^{-2} \|u(\vX)\|_{\Fock_\ph}^2\dd\vX \nonumber\\ &\quad+\tau\int_{B_{r_0}(\vX_0)}\beta(t)\e^{2\tau f(t)} \sum_{n=1}^N\|\vD_nu(\vX)\|_{\Fock_\ph}^2\dd\vX \nonumber\\ &\quad+\tau\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)} \|H_\rad^{1/2}u(\vX)\|_{\Fock_\ph}^2\dd\vX \leq C\mathcal T^0(u). \label{pf-free-carleman} \end{align} The constants are independent of $\vX_0$, and the same estimate holds on every smaller ball with the same $C$ and $\tau_0$. \end{lemma} \begin{proof} The proof has three steps. We first prove \cref{pf-free-carleman} by a computation in a radial gauge, assuming that $\wh\chi$ has compact support and that $u\in\mathcal C$, with $\mathcal C$ as defined before \cref{ucp-core-lemma}. The resulting constants depend only on the first ultraviolet moment. We then remove the compact support of $\wh\chi$ by an ultraviolet limit, and finally extend the estimate to all of $\DC(T^0)$ by an approximation argument. \noindent\textbf{Step 1: Compactly supported cutoff.} Without loss of generality, we translate $\vX_0$ to the origin. Translating the $n$th electron by $(\vX_0)_n$ multiplies $G^\mu_{\vx_n}$ by the phase $\e^{-\i\vk\cdot(\vX_0)_n}$, which leaves all properties used below unchanged, and we suppress it from the notation. We write $\vx_n=(x_n^1,x_n^2,x_n^3)$ and $\varPhi(h)=a^*(h)+a(h)$, so that $\Af^\mu(\vx)=\varPhi(G^\mu_\vx)$. The cutoff assumptions give uniform bounds for every spatial derivative of $G^\mu_\vx$ in \[ \Hil_\omega =\{h\in\Hil_\ph:h/\sqrt\omega\in\Hil_\ph\}, \qquad \|h\|_\omega^2=\|h\|^2+\|h/\sqrt\omega\|^2, \] and the same holds after multiplying these derivatives by $\omega$. The reality assumption and the real, even transverse polarization projector imply \[ \Im\dua{G^\mu_\vx}{G^\nu_\vy}=0 \qquad(\vx,\vy\in\RR^3). \] Consequently all the fields $\Af^\mu(\vx)$ strongly commute \cite[Proposition~A.6(iii)]{Q}. Their spatial derivatives have the same commutation property. Define the radial line integral and its unitary exponential by \begin{align*} \Theta(\vX)&=-e\sum_{n=1}^N\int_0^1\vx_n\cdot\AF(s\vx_n)\dd s =\varPhi(h_\vX), \qquad U_\vX=\e^{\i\Theta(\vX)},\\ h_\vX&=-e\sum_{n=1}^N\sum_{\mu=1}^3\int_0^1x_n^\mu G^\mu_{s\vx_n}\dd s. \end{align*} These formulas may first be read on vectors with finitely many photons. Strong commutativity, the field bounds and the Weyl relations then give their closed-operator and form interpretations. Write $U$ for the unitary $(Uu)(\vX)=U_\vX u(\vX)$ on $L^2(\RR^{3N},\Fock_\ph)$. Since $U_\vX$ commutes with all the fields, conjugation gives \[ U\vD_nU^*=-\i\grad_{\vx_n}+\vbf_n, \] where \[ \vbf_n(\vX)=-e\AF(\vx_n)-\grad_{\vx_n}\Theta(\vX). \] The operator-valued connection $\vbf=(\vbf_1,\ldots,\vbf_N)$ is $-e\AF$ in radial gauge, that is, it has no radial component: $\vx_n\cdot\vbf_n(\vX)=0$. This follows from the classical radial-gauge identity: for a smooth vector field $\vA$ on $\RR^3$, differentiation under the integral and the fundamental theorem of calculus in $s$ give \[ \vA(\vx)-\grad_\vx\int_0^1\vx\cdot\vA(s\vx)\dd s =-\int_0^1s\,\vx\times(\grad\times\vA)(s\vx)\dd s. \] The identity is real-linear in $\vA$, so it also holds for the $\Hil_\ph$-valued field $\vx\mapsto(G^1_\vx,G^2_\vx,G^3_\vx)$. Applying $\varPhi$, with $\BF=\grad\times\AF$, gives \[ \vbf_n(\vX)=e\int_0^1s\,\vx_n\times\BF(s\vx_n)\dd s. \] Let $\vX=r\vOmega$ with $\vOmega=(\vOmega_1,\ldots,\vOmega_N)\in\mathbb S^{3N-1}$, and recall $t=-\log r$, so that $r=\e^{-t}$ and $\partial_t=-r\partial_r$. Put $t_0=-\log r_0>2$. In $(t,\vOmega)$-coordinates, $B_{r_0}\setminus\{0\}$ is the cylinder $(t_0,\infty)\times\mathbb S^{3N-1}$. For each $t>t_0$, define the angular connection, an operator-valued one-form on the unit sphere $\mathbb S^{3N-1}$ acting on $\Fock_\ph$-valued functions, by \[ \mathfrak a(t,\vOmega)[\vzeta]=r\,\vzeta\cdot\vbf(r\vOmega), \qquad \vzeta=(\vzeta_1,\ldots,\vzeta_N)\in T_\vOmega\mathbb S^{3N-1}. \] Since $\vX\cdot\vbf(\vX)=0$, the vector $\vbf(r\vOmega)$ is tangent to the sphere, so $\mathfrak a$ determines $\vbf$ completely. Differentiating the radial integral in $t$, with $\vOmega$ fixed, gives \begin{equation}\label{pf-angular-derivative} \dot{\mathfrak a}(t,\vOmega)[\vzeta] =er^2\sum_{n=1}^N(\vOmega_n\times\vzeta_n)\cdot\BF(r\vOmega_n). \end{equation} Let $\nabla_{\mathbb S}$ denote the spherical gradient on $\mathbb S^{3N-1}$, acting on $\Fock_\ph$-valued functions. Set \[ \nabla_{\mathbb S,\mathfrak a} =\nabla_{\mathbb S}+\i\mathfrak a, \qquad K(t)=\nabla_{\mathbb S,\mathfrak a}^* \nabla_{\mathbb S,\mathfrak a}, \qquad \Upsilon(t,\vOmega)=2U_{r\vOmega}H_\rad U_{r\vOmega}^*. \] Since $\vX\cdot\vbf(\vX)=0$, the gauge-transformed covariant Laplacian has no connection in the radial direction, so \[ (-\i\grad_{\vX}+\vbf)^2 =-\partial_r^2-\frac{3N-1}{r}\partial_r+\frac{1}{r^2}K(t). \] Since $UT^0U^*=\frac12(-\i\grad_\vX+\vbf)^2+UH_\rad U^*$, multiplying by $2r^2$ and using $\partial_t=-r\partial_r$ gives \begin{equation}\label{pf-polar-form} 2r^2\,UT^0U^*=-\partial_t^2+2\gamma\,\partial_t+K(t)+r^2\Upsilon(t), \qquad \gamma=\frac{3N-2}{2}. \end{equation} Here $\Upsilon(t)=\Upsilon(t,\cdot)$ is a nonnegative multiplication operator on the angular Hilbert space $L^2(\mathbb S^{3N-1},\Fock_\ph)$, with the surface measure $\dd\vOmega$. For $\Xi$ in this space we write \[ \Upsilon(t)[\Xi]=2\int_{\mathbb S^{3N-1}} \|H_\rad^{1/2}U_{r\vOmega}^*\Xi(\vOmega)\|_{\Fock_\ph}^2\dd\vOmega, \qquad K(t)[\Xi]=\int_{\mathbb S^{3N-1}} \|\nabla_{\mathbb S,\mathfrak a}\Xi(\vOmega)\|^2\dd\vOmega \] for the corresponding quadratic forms, where the integrand of $K(t)[\Xi]$ is the norm of a tangent-vector-valued function with values in $\Fock_\ph$. By the Weyl relations and $\sqrt\omega\wh\chi\in L^2$, $U_\vX$ preserves $\DC(H_\rad^{1/2})$, and by the field bound $\mathfrak a$ is bounded by $(H_\rad+1)^{1/2}$. Hence the form domain of $K(t)+\Upsilon(t)$ is \[ \mathcal Q_{\mathbb S} =H^1(\mathbb S^{3N-1},\Fock_\ph)\cap\DC(H_\rad^{1/2}), \] with $H_\rad$ acting pointwise on the sphere, independently of $t>t_0$. We write $\dot K(t)[\Xi]$ for the derivative of $t\mapsto K(t)[\Xi]$ with $\Xi$ fixed. \begin{claim} There is $C<\infty$, independent of $t$ and $\Xi$, such that for $t>t_0$ and $\Xi\in\mathcal Q_{\mathbb S}$, \begin{equation}\label{pf-angular-form-derivative} |\dot K(t)[\Xi]| \leq Cr^2K(t)[\Xi]^{1/2} \big(\Upsilon(t)[\Xi]+\|\Xi\|^2\big)^{1/2}. \end{equation} \end{claim} \begin{proof}[Proof of the claim] In \[ K(t)[\Xi]=\int_{\mathbb S^{3N-1}} \|\nabla_{\mathbb S}\Xi(\vOmega)+\i\mathfrak a(t,\vOmega)\Xi(\vOmega)\|^2\dd\vOmega \] the term $\nabla_{\mathbb S}\Xi$ does not depend on $t$, so only the connection is differentiated, and \[ \dot K(t)[\Xi]=2\Re\dua{\nabla_{\mathbb S,\mathfrak a}\Xi} {\i\dot{\mathfrak a}\Xi}. \] By the Cauchy--Schwarz inequality, $|\dot K(t)[\Xi]|\leq2K(t)[\Xi]^{1/2}\|\dot{\mathfrak a}\Xi\|$, so it suffices to show \begin{equation}\label{pf-angular-connection-derivative} \|\dot{\mathfrak a}\Xi\| \leq Cr^2\big(\Upsilon(t)[\Xi]+\|\Xi\|^2\big)^{1/2}. \end{equation} Let $\vzeta\in T_\vOmega\mathbb S^{3N-1}$ and $\xi\in\Fock_\ph$. By \cref{pf-angular-derivative} and $\sum_{n=1}^N|\vOmega_n||\vzeta_n|\leq|\vOmega||\vzeta|=|\vzeta|$, \[ \|\dot{\mathfrak a}(t,\vOmega)[\vzeta]\xi\| \leq er^2|\vzeta|\max_n\|\BF(r\vOmega_n)\xi\|, \] where $\|\BF(\vx)\xi\|^2=\sum_\lambda\|\Bf^\lambda(\vx)\xi\|^2$. The form factors of $\BF(\vx)$ are first derivatives of $G^\mu_\vx$, which are uniformly bounded in $\Hil_\omega$, so the field bound gives $\|\BF(\vx)\eta\|\leq C(\|H_\rad^{1/2}\eta\|^2+\|\eta\|^2)^{1/2}$ uniformly in $\vx$. Since $\BF(\vx)$ commutes with $U_{r\vOmega}$, applying this to $\eta=U_{r\vOmega}^*\xi$ gives \[ \|\BF(r\vOmega_n)\xi\|=\|\BF(r\vOmega_n)U_{r\vOmega}^*\xi\| \leq C\big(\|H_\rad^{1/2}U_{r\vOmega}^*\xi\|^2+\|\xi\|^2\big)^{1/2}. \] Taking $\xi=\Xi(\vOmega)$, summing over an orthonormal basis of $T_\vOmega\mathbb S^{3N-1}$ and integrating over the sphere gives \cref{pf-angular-connection-derivative}. \end{proof} Similarly, we write $\dot\Upsilon(t)[\Xi]$ for the derivative of $t\mapsto\Upsilon(t)[\Xi]$ with $\Xi$ fixed. \begin{claim} There is $C<\infty$, independent of $t$ and $\Xi$, such that for $t>t_0$ and $\Xi\in\mathcal Q_{\mathbb S}$, \begin{equation}\label{pf-radiation-derivative} |\dot\Upsilon(t)[\Xi]|\leq Cr\,\Upsilon(t)[\Xi]^{1/2}\|\Xi\|. \end{equation} \end{claim} \begin{proof}[Proof of the claim] Write $\varPi(h)=\i(a^*(h)-a(h))$, so that $[\varPhi(h),H_\rad]=\i\varPi(\omega h)$. Since $U_{r\vOmega}=\e^{\i\varPhi(h_{r\vOmega})}$ and $\varPhi(h_{r\vOmega})$ commutes with $\varPhi(\dot h_{r\vOmega})$, \[ \dot\Upsilon(t,\vOmega) =2\i\,U_{r\vOmega}\big[\varPhi(\dot h_{r\vOmega}),H_\rad\big]U_{r\vOmega}^* =-2U_{r\vOmega}\varPi(\omega\,\dot h_{r\vOmega})U_{r\vOmega}^*, \] where \[ \dot h_{r\vOmega} =er\sum_{n=1}^N\sum_{\mu=1}^3\Omega_n^\mu G^\mu_{r\vOmega_n}. \] Since $\sqrt\omega\wh\chi\in L^2$, we have $\|\sqrt\omega\,\dot h_{r\vOmega}\|\leq Cr$. With $\xi=U_{r\vOmega}^*\Xi(\vOmega)$, the bound $|\dua{\xi}{\varPi(g)\xi}|\leq2\|g/\sqrt\omega\|\|H_\rad^{1/2}\xi\|\|\xi\|$ gives \[ |\dua{\Xi(\vOmega)}{\dot\Upsilon(t,\vOmega)\Xi(\vOmega)}| \leq Cr\|H_\rad^{1/2}U_{r\vOmega}^*\Xi(\vOmega)\|\|\Xi(\vOmega)\|. \] Integrating over the sphere and using the Cauchy--Schwarz inequality gives \cref{pf-radiation-derivative}. \end{proof} Thus $\dot\Upsilon$ is controlled by the transformed radiation energy itself, so $H_\rad$ can be kept in the principal part of the operator $Q$ in \cref{pf-cylinder-operator} below. Conjugating \cref{pf-polar-form} by $r^\gamma=\e^{-\gamma t}$ removes the first-order term: \begin{equation}\label{pf-cylinder-operator} 2r^{\gamma+2}UT^0u=Q(r^\gamma Uu), \qquad Q=-\partial_t^2+K(t)+\gamma^2+r^2\Upsilon(t). \end{equation} Set \[ z(t,\vOmega)=\e^{\tau f(t)}r^\gamma U_{r\vOmega}u(r\vOmega),\qquad z(t)=z(t,\cdot)\in L^2(\mathbb S^{3N-1},\Fock_\ph). \] Then $z$ has compact support in the cylinder $(t_0,\infty)\times\mathbb S^{3N-1}$. We write $\|z(t)\|$ for the norm in $L^2(\mathbb S^{3N-1},\Fock_\ph)$ and $\|z\|$ for the norm in $L^2((t_0,\infty)\times\mathbb S^{3N-1},\Fock_\ph)$ with measure $\dd t\dd\vOmega$, so that $\|z\|^2=\int_{t_0}^\infty\|z(t)\|^2\dd t$. The operators $K(t)$, $\gamma^2$ and $r^2\Upsilon(t)$ act at fixed $t$ and commute with multiplication by $\e^{\tau f(t)}$, while \[ \e^{\tau f}(-\partial_t^2)\e^{-\tau f}=-(\partial_t-\tau\dot f)^2 =-\partial_t^2-\tau^2(\dot f)^2+2\tau\dot f\partial_t+\tau\ddot f. \] Hence \[ \e^{\tau f}Q\e^{-\tau f}=S+B, \qquad S=Q-\tau^2(\dot f)^2, \qquad B=2\tau\dot f\partial_t+\tau\ddot f. \] By \cref{pf-cylinder-operator}, $(S+B)z=2\e^{\tau f}r^{\gamma+2}UT^0u$. Since $\wh\chi$ has compact support, all moments $\omega^k\wh\chi$ are finite, and the Weyl relations show that $z$, $Sz$ and $Bz$ are smooth, compactly supported in $(t_0,\infty)\times\mathbb S^{3N-1}$, and take values in $\DC(H_\rad^k)$ for all $k$, together with all field moments. On such vectors, integration by parts in $t$ shows that $S$ is symmetric and $B$ is skew symmetric, so $\|(S+B)z\|^2=\|Sz\|^2+\|Bz\|^2+\dua{z}{[S,B]z}$. Since $B$ involves only $\partial_t$ and functions of $t$, the operators $K$ and $\Upsilon$ enter $[S,B]$ only through $\dot K$ and $\partial_t(r^2\Upsilon)$, which are controlled by the claims above; no commutator of $K$ and $\Upsilon$ is needed. Computing $[S,B]$ term by term, with $r=\e^{-t}$ so that $\partial_t(r^2)=-2r^2$ and with $\beta=\ddot f>0$, gives the identity \begin{align} \|(S+B)z\|^2 ={}&\|Sz\|^2+\|Bz\|^2 +4\tau\int_{t_0}^\infty\beta(t)\|\partial_tz(t)\|^2\dd t \nonumber\\ &+4\tau^3\int_{t_0}^\infty(\dot f(t))^2\beta(t)\|z(t)\|^2\dd t -\tau\int_{t_0}^\infty f^{(4)}(t)\|z(t)\|^2\dd t \nonumber\\ &-2\tau\int_{t_0}^\infty\dot f(t)\dot K(t)[z(t)]\dd t +4\tau\int_{t_0}^\infty\dot f(t)r^2\Upsilon(t)[z(t)]\dd t \nonumber\\ &-2\tau\int_{t_0}^\infty\dot f(t)r^2\dot\Upsilon(t)[z(t)]\dd t. \label{pf-cylinder-commutator} \end{align} It remains to absorb the three terms in \cref{pf-cylinder-commutator} that may be negative, those containing $f^{(4)}$, $\dot K$ and $\dot\Upsilon$, into the nonnegative terms. For $t\geq2$, $\dot f(t)=1-\alpha t^{-\alpha-1}$ lies between $2/(\alpha+2)$ and $1$, because $t^{\alpha+1}\geq2^{\alpha+1}\geq\alpha+2$ by Bernoulli's inequality, and $|f^{(4)}|\leq C\beta$. For any fixed $\epsilon>0$, \cref{pf-angular-form-derivative} gives \begin{align*} \tau\int_{t_0}^\infty|\dot K(t)[z(t)]|\dd t &\leq\epsilon\tau\int_{t_0}^\infty\beta(t)K(t)[z(t)]\dd t\\ &\quad+C_\epsilon\tau\int_{t_0}^\infty\frac{r^4}{\beta(t)} \big(\Upsilon(t)[z(t)]+\|z(t)\|^2\big)\dd t. \end{align*} Similarly, \cref{pf-radiation-derivative} gives \[ \tau\int_{t_0}^\infty r^2|\dot\Upsilon(t)[z(t)]|\dd t \leq\epsilon\tau\int_{t_0}^\infty r^2\Upsilon(t)[z(t)]\dd t +C_\epsilon\tau\int_{t_0}^\infty r^4\|z(t)\|^2\dd t. \] Since $K$ enters \cref{pf-cylinder-commutator} only through $\dot K$, none of the nonnegative terms there can absorb $\epsilon\tau\int_{t_0}^\infty\beta(t)K(t)[z(t)]\dd t$. Instead, such a term is controlled by $\|Sz\|^2$. \begin{claim} There is $C<\infty$, independent of $\tau\geq1$ and $z$, such that \begin{equation}\label{pf-angular-multiplier} \tau\int_{t_0}^\infty\beta(t)K(t)[z(t)]\dd t \leq C\left(\|Sz\|^2+ \tau^3\int_{t_0}^\infty\beta(t)\|z(t)\|^2\dd t\right). \end{equation} \end{claim} \begin{proof}[Proof of the claim] Integrating by parts in $t$, \begin{align*} \Re\dua{Sz}{\beta z} ={}&\int_{t_0}^\infty\beta(t)\big(\|\partial_tz(t)\|^2+K(t)[z(t)] +r^2\Upsilon(t)[z(t)]+\gamma^2\|z(t)\|^2\big)\dd t\\ &-\tau^2\int_{t_0}^\infty(\dot f(t))^2\beta(t)\|z(t)\|^2\dd t -\frac12\int_{t_0}^\infty\ddot\beta(t)\|z(t)\|^2\dd t. \end{align*} Keeping the $K$ term, discarding the other nonnegative terms and multiplying by $\tau$ gives \begin{align*} \tau\int_{t_0}^\infty\beta(t)K(t)[z(t)]\dd t &\leq\tau\Re\dua{Sz}{\beta z} +\tau^3\int_{t_0}^\infty(\dot f(t))^2\beta(t)\|z(t)\|^2\dd t\\ &\quad+\frac\tau2\int_{t_0}^\infty|\ddot\beta(t)|\|z(t)\|^2\dd t. \end{align*} By Young's inequality, \[ \tau\Re\dua{Sz}{\beta z} \leq\frac12\|Sz\|^2 +\frac{\tau^2}2\int_{t_0}^\infty\beta(t)^2\|z(t)\|^2\dd t. \] For $t\geq2$, $\dot f$ and $\beta$ are bounded and $|\ddot\beta|\leq C\beta$. Combining the two displays, we get for $\tau\geq1$ \[ \tau\int_{t_0}^\infty\beta(t)K(t)[z(t)]\dd t \leq\frac12\|Sz\|^2+C\tau^3\int_{t_0}^\infty\beta(t)\|z(t)\|^2\dd t, \] which is \cref{pf-angular-multiplier}. \end{proof} Choose $\epsilon$ small enough that \[ \epsilon\tau\int_{t_0}^\infty\beta(t)K(t)[z(t)]\dd t \qquad\text{and}\qquad \epsilon\tau\int_{t_0}^\infty r^2\Upsilon(t)[z(t)]\dd t \] are absorbed, the first by $\|Sz\|^2$ and the $\tau^3$ term of \cref{pf-cylinder-commutator} using \cref{pf-angular-multiplier}, and the second by the $\Upsilon$ term of \cref{pf-cylinder-commutator}. The remaining errors are \[ C_\epsilon\tau\int_{t_0}^\infty\Big(\frac{r^4}{\beta(t)} \big(\Upsilon(t)[z(t)]+\|z(t)\|^2\big)+r^4\|z(t)\|^2\Big)\dd t. \] Since \[ \sup_{0\Lambda}_\vx=G^\mu_\vx\iden_{\{\omega>\Lambda\}}$, so that $\Af^\mu(\vx)-\Af^{\mu,\le\Lambda}(\vx)=\varPhi(G^{\mu,>\Lambda}_\vx)$. Hence the components of $\vD_n^{\le\Lambda}u-\vD_nu$ are $e\varPhi(G^{\mu,>\Lambda}_{\vx_n})u$. The difference $T^{0,\le\Lambda}u-T^0u$ is a sum of terms $\varPhi(G^{\mu,>\Lambda}_{\vx_n})\partial_{x_n^\mu}u$ and of products of two fields applied to $u$, each with a factor $\varPhi(G^{\mu,>\Lambda}_{\vx_n})$. The creation and annihilation bounds of \cite[Section~3.3]{Q} give $\|\varPhi(h)\xi\|\leq C\|h\|_\omega\|(H_\rad+1)^{1/2}\xi\|$. Together with \cref{pf-two-field-bound}, this gives, with $C$ independent of $\Lambda$, \[ \begin{split} &\|\vD_n^{\le\Lambda}u-\vD_nu\|+\|T^{0,\le\Lambda}u-T^0u\|\\ &\qquad\leq C\sup_{\mu,\vx}\|G^{\mu,>\Lambda}_\vx\|_\omega \bigl(\|(H_\rad+1)u\|+\|(H_\rad+1)^{1/2}\grad_\vX u\|\bigr). \end{split} \] The norms of $u$ on the right are finite, since $\Fock_{\mathrm{fin}}\subset\DC(H_\rad)$. Finally, $\|G^{\mu,>\Lambda}_\vx\|_\omega$ does not depend on $\vx$, and it tends to zero by \cref{cutoff-assumption} and dominated convergence. Thus, for each such $u$, both sides of \cref{pf-free-carleman} for the cutoff $\wh\chi^{\le\Lambda}$ converge to those for $\wh\chi$, so \cref{pf-free-carleman} holds for all such $u$. The gauge transformation $U$ depends on $\Lambda$ but does not appear in \cref{pf-free-carleman}, so it plays no role in this limit. \noindent\textbf{Step 3: Approximation argument.} Step~2 of the proof of \cref{ucp-local-operator-domain} uses only \cref{cutoff-assumption,ucp-extra-cutoff}. It gives $\DC(T^0)=\DC(-\Delta_{\vX})\cap\DC(H_\rad)$ and the equivalence of the graph norms of $T^0$ and $L=-\Delta_{\vX}+H_\rad+1$. Hence, by \cref{ucp-core-lemma}, an annularly supported $u\in\DC(T^0)$ is a limit in the graph norm of $T^0$ of functions in $\mathcal C$ supported in a slightly larger annulus. Graph convergence implies form convergence, and the weights are bounded on this annulus. Consequently all terms in \cref{pf-free-carleman} pass to the limit. After the ultraviolet limit the approximation is performed in the original gauge, so no higher ultraviolet moments are needed for the final cutoff. The bounds on the form factors are uniform in the center. Once the radius and the lower bound on $\tau$ have been fixed, restriction to a smaller ball preserves the estimate with the same constants. \end{proof} We use a dyadic decomposition of the punctured ball. With $r$, $\beta$, $\vX_0$ and $r_0$ as in \cref{lem-pf-free-carleman}, let, for $j\geq0$, \[ R_j=2^{-j}r_0,\qquad \beta_j=\beta(-\log R_j),\qquad \mathcal A_j=\{R_j/22, \] where the divergence is distributional. With the weights of \cref{lem-pf-free-carleman}, there are $r_0>0$, $\tau_0\geq1$ and $C<\infty$ such that, for every $\tau\geq\tau_0$ and every $u$ in the local operator domain of $T^{\vA}$ with compact support in $B_{r_0}(\vX_0)\setminus\{\vX_0\}$, \begin{equation}\label{pf-covariant-carleman} \begin{aligned} \int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}\Bigl( &\tau^3\beta(t)r^{-2}\|u(\vX)\|_{\Fock_\ph}^2 +\tau\beta(t)\sum_{n=1}^N\|\vpi_nu(\vX)\|_{\Fock_\ph}^2\\ &+\tau\|H_\rad^{1/2}u(\vX)\|_{\Fock_\ph}^2\Bigr)\dd\vX \leq C \mathcal T^{\vA}(u). \end{aligned} \end{equation} Moreover, the mixed estimate \cref{ucp-dyadic-fractional} holds with $\mathcal T^{\vA}(u)$ in place of $\mathcal T^0(u)$ on the right. These choices are uniform on compact sets with uniform local $L^6$ bounds on $\vA$ and $L^p$ bounds on $\div\vA$. \end{lemma} \begin{proof} Write $T^{\vA}=T^0+R_\vA$, where $R_\vA$ consists of the first-order term $\sum_{n=1}^N\vA(\vx_n)\cdot\vD_n$ and a zero-order multiplier $q_\vA$. Using the mixed estimate \cref{ucp-dyadic-fractional} of \cref{ucp-fractional-carleman}, we show that \[ \int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2\|R_\vA u(\vX)\|_{\Fock_\ph}^2\dd\vX \leq\tfrac14\mathcal T^0(u) \] for $r_0$ small and $\tau$ large: the first-order part is small when $r_0$ is small, and the zero-order part is small when $\tau$ is large. Since $T^0u=T^{\vA}u-R_\vA u$, this gives $\mathcal T^0(u)\leq4\mathcal T^{\vA}(u)$, so \cref{lem-pf-free-carleman} holds with $\mathcal T^{\vA}(u)$ on the right. Finally, we replace $\vD_n$ by $\vpi_n=\vD_n+\vA(\vx_n)$, using \cref{ucp-dyadic-fractional} of \cref{ucp-fractional-carleman} once more. \noindent\textbf{Step 1: Decomposition of $T^{\vA}$.} Fix neighborhoods of the projected center on which the local norms of $\vA$ and $\div\vA$ are finite, and write \begin{align*} M_A^2&=\sum_{n=1}^N\|\vA\|_{L^6(B_R((\vX_0)_n))}^2,\\ M_d^2&=\sum_{n=1}^N\|\div\vA\|_{L^p(B_R((\vX_0)_n))}^2 \end{align*} for a sufficiently small fixed $R>0$. We will decrease $r_00$, put \[ \sigma_a(r_0)=\sup_{02$, the Sobolev inequality and the commuting spectral operators give \[ \|b(\vx_n)\eta_j\e^{\tau f}u\| \leq C_q\|b\|_q\|(-\Delta_{\vx_n})^{3/(2q)}(\eta_j\e^{\tau f}u)\| \leq C_q\|b\|_q\|L^{3/(2q)}(\eta_j\e^{\tau f}u)\|. \] Applying \cref{ucp-dyadic-fractional} of \cref{ucp-fractional-carleman} with $s=1$ and $s=3/p$ yields directly \begin{align} &\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2\|q_\vA(\vX)u(\vX)\|_{\Fock_\ph}^2\dd\vX\notag\\ &\quad\leq C\left( M_A^4\sigma_2(r_0)\tau^{-1} +M_d^2\sigma_{4-6/p}(r_0)\tau^{-(3-6/p)} \right)\mathcal T^0(u). \label{ucp-endpoint-zero-order} \end{align} Only the moduli of the multipliers enter, so the imaginary divergence term causes no change. \noindent\textbf{Step 4: Absorption.} Since $\|R_\vA u\|^2\leq2\big\|\sum_{n=1}^N\vA(\vx_n)\cdot\vD_nu\big\|^2 +2\|q_\vA u\|^2$, \cref{ucp-endpoint-first-order,ucp-endpoint-zero-order} give \begin{align*} &\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2\|R_\vA u(\vX)\|_{\Fock_\ph}^2\dd\vX\\ &\quad\leq C\left(M_A^2\sigma_1(r_0)+M_A^4\sigma_2(r_0)\tau^{-1} +M_d^2\sigma_{4-6/p}(r_0)\tau^{-(3-6/p)}\right)\mathcal T^0(u). \end{align*} Choose $r_0$ so small that $CM_A^2\sigma_1(r_0)\leq\frac18$. This is possible because $\sigma_1(r_0)\to0$ as $r_0\downarrow0$, while $C$ and $M_A$ do not grow when $r_0$ is decreased: $C$ comes from \cref{lem-pf-free-carleman,ucp-fractional-carleman}, which hold on every smaller ball with the same constants, and $M_A$ is a norm on the fixed balls $B_R((\vX_0)_n)$. Next increase $\tau_0$ so that the remaining two terms are at most $\frac18$, using $3-6/p>0$ and that $\sigma_a(r_0)$ does not increase as $r_0$ decreases. Then \[ \int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2\|R_\vA u(\vX)\|_{\Fock_\ph}^2\dd\vX \leq\tfrac14\mathcal T^0(u). \] Since $T^0u=T^{\vA}u-R_\vA u$, this gives $\mathcal T^0(u)\leq2\mathcal T^{\vA}(u)+\frac12\mathcal T^0(u)$, that is, \begin{equation}\label{ucp-external-source-control} \mathcal T^0(u)\leq4\mathcal T^{\vA}(u). \end{equation} This proves the assertion about the mixed estimate and bounds the left-hand side of \cref{pf-free-carleman} of \cref{lem-pf-free-carleman} by $C \mathcal T^{\vA}(u)$. \noindent\textbf{Step 5: Replacing $\vD_n$ by $\vpi_n$.} To replace $\vD_n$ by $\vpi_n$, use the second estimate in \cref{ucp-endpoint-multipliers} and \cref{ucp-dyadic-fractional} of \cref{ucp-fractional-carleman} with $s=1/2$ to obtain \[ \tau\int_{B_{r_0}(\vX_0)}\beta(t)\e^{2\tau f(t)} \sum_{n=1}^N\|\vA(\vx_n)u(\vX)\|_{\Fock_\ph}^2\dd\vX \leq C M_A^2\frac{r_0}{\tau}\mathcal T^0(u). \] Together with $\|\vpi_nu\|^2\leq2\|\vD_nu\|^2+2\|\vA(\vx_n)u\|^2$ and \cref{ucp-external-source-control}, this gives \cref{pf-covariant-carleman}. The fixed local norm bounds permit all choices uniformly on compact sets. The endpoint $L^6$ assumption is used through smallness of the radius, without requiring smallness of the external field itself. \end{proof} \begin{lemma}[A Carleman estimate with scalar potential] \label[lemma]{ucp-scalar-perturbation} Under the assumptions of \cref{lem-pf-covariant-carleman}, suppose on $B_{r_0}(\vX_0)$ that, with the same $p$, \[ \begin{gathered} W(\vX)=\sum_{n=1}^Nv_n(\vx_n)+W_b(\vX),\\ v_n\in L^p(B_{r_0}((\vX_0)_n)),\qquad W_b\in L^\infty(B_{r_0}(\vX_0)). \end{gathered} \] Then, after increasing $\tau_0$, for every $\tau\geq\tau_0$ and every $u$ in the local operator domain of $T^{\vA}+W$ with compact support in $B_{r_0}(\vX_0)\setminus\{\vX_0\}$, \begin{equation}\label{ucp-scalar-carleman} \begin{aligned} \int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}\Bigl( &\tau^3\beta(t)r^{-2}\|u(\vX)\|_{\Fock_\ph}^2 +\tau\beta(t)\sum_{n=1}^N\|\vpi_nu(\vX)\|_{\Fock_\ph}^2\\ &+\tau\|H_\rad^{1/2}u(\vX)\|_{\Fock_\ph}^2\Bigr)\dd\vX\\ \leq C&\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|T^{\vA}u(\vX)+W(\vX)u(\vX)\|_{\Fock_\ph}^2\dd\vX. \end{aligned} \end{equation} The constants are uniform on compact sets under the local bounds of \cref{lem-pf-covariant-carleman}, the indicated $L^p$ norms, and $\|W_b\|_\infty$. \end{lemma} \begin{proof} Put $s=3/p<3/2$, $M_p^2=\sum_{n=1}^N\|v_n\|_p^2$, and $M_b=\|W_b\|_\infty$. Extend each $v_n$ by zero outside its three-dimensional ball. The Sobolev inequality, as in \cref{ucp-coordinate-multiplier}, gives \begin{align*} \|v_n(\vx_n)\eta_j\e^{\tau f}u\| &\leq C\|v_n\|_p\|(-\Delta_{\vx_n})^{s/2}(\eta_j\e^{\tau f}u)\|\\ &\leq C\|v_n\|_p\|(-\Delta_{\vX})^{s/2}(\eta_j\e^{\tau f}u)\|. \end{align*} The second inequality follows from the corresponding Fourier multipliers. The identity $\sum_{j\geq0}\eta_j^2=1$ and $r\asymp R_j$ on $\supp\eta_j$ give \begin{align*} \mathcal W(u)&:=\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|W(\vX)u(\vX)\|_{\Fock_\ph}^2\dd\vX\\ &\leq C M_p^2\sum_{j\geq0}R_j^2 \|(-\Delta_{\vX})^{s/2}(\eta_j\e^{\tau f}u)\|^2 +C M_b^2\sum_{j\geq0}R_j^2\|\eta_j\e^{\tau f}u\|^2. \end{align*} By the joint spectral calculus, $\|(-\Delta_{\vX})^{s/2}\Phi\|\leq\|L^{s/2}\Phi\|$. Using \cref{ucp-dyadic-fractional} of \cref{ucp-fractional-carleman} with $s=3/p$ and $s=0$, followed by \cref{ucp-external-source-control}, we obtain, with $\sigma_a(r_0)$ as in the proof of \cref{lem-pf-covariant-carleman}, \begin{equation}\label{ucp-scalar-smallness} \mathcal W(u)\leq C\left( M_p^2\sigma_{4-2s}(r_0)\tau^{-(3-2s)} +M_b^2\sigma_4(r_0)\tau^{-3}\right)\mathcal T^{\vA}(u). \end{equation} Both powers of $\tau$ decay because $p>2$. The scalar multiplier is infinitesimally bounded relative to the free electronic operator by \cref{ucp-coordinate-multiplier} and the moment inequality \cref{moment-inequality}. Thus, as in Step~3 of the proof of \cref{ucp-local-operator-domain}, which involves no eigenstate, $T^{\vA}+W$ and $T^{\vA}$ have the same local operator domain. All terms above are consequently defined for the vectors in the statement. Writing $T^{\vA}u=(T^{\vA}u+Wu)-Wu$ gives \[ \mathcal T^{\vA}(u)\leq2\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|T^{\vA}u(\vX)+W(\vX)u(\vX)\|_{\Fock_\ph}^2\dd\vX +2\mathcal W(u). \] Increase $\tau_0$ so that, for $\tau\geq\tau_0$, \[ 2C\left(M_p^2\sigma_{4-2s}(r_0)\tau^{-(3-2s)} +M_b^2\sigma_4(r_0)\tau^{-3}\right)\leq\frac12, \] with $C$ from \cref{ucp-scalar-smallness}. Then $2\mathcal W(u)\leq\frac12\mathcal T^{\vA}(u)$, and since $\mathcal T^{\vA}(u)<\infty$, subtracting this term gives \[ \mathcal T^{\vA}(u)\leq4\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|T^{\vA}u(\vX)+W(\vX)u(\vX)\|_{\Fock_\ph}^2\dd\vX. \] Inserting this bound on the right of \cref{pf-covariant-carleman} of \cref{lem-pf-covariant-carleman} gives \cref{ucp-scalar-carleman}. All estimates use only the stated local norms, which gives uniformity in the center. \end{proof} Finally, we include electron spin. On $L^2(\RR^{3N},(\CC^2)^{\otimes N}\otimes\Fock_\ph)$, let $T^{\vA}$ and $L$ act componentwise and set \[ \begin{split} V(\vX)&=\sum_{n=1}^Nv_n(\vx_n)+V_b(\vX),\\ S_{\mathrm{ext}}(\vX)&=\frac12\sum_{n=1}^N \vsigma_n\cdot\vB(\vx_n),\\ S_{\mathrm{ph}}(\vX)&=-\frac e2\sum_{n=1}^N \vsigma_n\cdot\BF(\vx_n),\\ H^{V,\vA}&=T^{\vA}+V+S_{\mathrm{ext}}+S_{\mathrm{ph}}, \end{split} \] where $\vB=\grad\times\vA$ in the sense of distributions. By the Pauli identity \cref{pauli-expansion}, \[ H^{V,\vA}=\frac12\sum_{n=1}^N(\vsigma_n\cdot\vpi_n)^2+H_\rad+V \] in the sense of quadratic forms. The Zeeman terms $S_{\mathrm{ext}}$ and $S_{\mathrm{ph}}$ are separated because they are treated as perturbations of $T^{\vA}$. \begin{lemma}[A Carleman estimate with spin] \label[lemma]{ucp-pauli-carleman} In addition to the hypotheses of \cref{lem-pf-covariant-carleman}, suppose, with the same $p$, that on the configuration ball under consideration \[ v_n\in L^p(B_{r_0}((\vX_0)_n)),\qquad V_b\in L^\infty(B_{r_0}(\vX_0)),\qquad \vB\in L^p(B_{r_0}((\vX_0)_n),\RR^3) \] for every $n$. Let $E\in\CC$ be fixed. Then, after increasing $\tau_0$, for every $\tau\geq\tau_0$ and every $u$ in the local operator domain of $H^{V,\vA}$ with compact support in $B_{r_0}(\vX_0)\setminus\{\vX_0\}$, \begin{equation}\label{pauli-carleman} \begin{aligned} \int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}\Bigl( &\tau^3\beta(t)r^{-2}\|u(\vX)\|_{\mathcal K_N}^2 +\tau\beta(t)\sum_{n=1}^N\|\vpi_nu(\vX)\|_{\mathcal K_N}^2\\ &+\tau\|H_\rad^{1/2}u(\vX)\|_{\mathcal K_N}^2\Bigr)\dd\vX\\ \leq C&\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|(H^{V,\vA}-E)u(\vX)\|_{\mathcal K_N}^2\dd\vX, \end{aligned} \end{equation} where $\mathcal K_N=(\CC^2)^{\otimes N}\otimes\Fock_\ph$. The constants are uniform on compact sets under the local bounds in \cref{lem-pf-covariant-carleman}, the displayed $L^p$ bounds, and a bound on $\|V_b-E\|_\infty$. \end{lemma} \begin{proof} Expand $u$ in an orthonormal basis of $(\CC^2)^{\otimes N}$. Each component is an $\Fock_\ph$-valued function, and $T^{\vA}$, $\vpi_n$, $H_\rad^{1/2}$ and $L$ act on each component separately. The squared $\mathcal K_N$-norms in \cref{pf-covariant-carleman} of \cref{lem-pf-covariant-carleman} and \cref{ucp-dyadic-fractional} of \cref{ucp-fractional-carleman} are the sums of the squared $\Fock_\ph$-norms of the components. Applying \cref{lem-pf-covariant-carleman} to each component and summing, we obtain \cref{pf-covariant-carleman} of \cref{lem-pf-covariant-carleman} and the mixed estimate \cref{ucp-dyadic-fractional} of \cref{ucp-fractional-carleman}, with $\mathcal T^{\vA}(u)$ on the right, for $\mathcal K_N$-valued $u$, with the same constants. Here $\mathcal T^{\vA}(u)$ is the source integral of \cref{lem-pf-covariant-carleman}, with norms in $\mathcal K_N$. It remains to replace $T^{\vA}$ by $H^{V,\vA}-E$. The difference $V+S_{\mathrm{ext}}+S_{\mathrm{ph}}-E$ is a perturbation of the same kind as in \cref{ucp-scalar-perturbation}. We show that its weighted $L^2$ norm is at most $\delta(\tau)\mathcal T^{\vA}(u)$ with $\delta(\tau)\to0$ as $\tau\to\infty$, and then absorb it. The classical part $V+S_{\mathrm{ext}}-E$ is handled as in \cref{ucp-scalar-perturbation}, using the operator norms of its matrix coefficients. The quantized Zeeman term $S_{\mathrm{ph}}$ is controlled by the radiation and zero-order terms of \cref{pf-covariant-carleman} of \cref{lem-pf-covariant-carleman}. Put \[ \begin{split} s&=3/p<3/2,\qquad M_b=\|V_b-E\|_\infty,\\ M_p^2&=\sum_{n=1}^N\left( \|v_n\|_{L^p(B_{r_0}((\vX_0)_n))}^2 +\tfrac14\|\vB\|_{L^p(B_{r_0}((\vX_0)_n))}^2\right), \end{split} \] with $\sigma_a(r_0)$ as in the proof of \cref{lem-pf-covariant-carleman}. The Pauli identity gives $\|\vsigma_n\cdot\vB(\vx_n)\|_{\mathrm{op}}=|\vB(\vx_n)|$. Thus the Sobolev inequality gives \[ \big\|\big(v_n(\vx_n)+\tfrac12 \vsigma_n\cdot\vB(\vx_n)\big)\eta_j\e^{\tau f}u\big\| \leq C\big(\|v_n\|_p+\tfrac12\|\vB\|_p\big) \|L^{s/2}(\eta_j\e^{\tau f}u)\|. \] Here the three-dimensional coefficients are extended by zero outside their respective balls. The identity $\sum_{j\geq0}\eta_j^2=1$ and the mixed estimates with exponents $s$ and $0$ now give, exactly as in \cref{ucp-scalar-smallness}, \begin{equation}\label{pauli-classical-smallness} \begin{aligned} &\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2\|(V+S_{\mathrm{ext}}-E)u(\vX)\|_{\mathcal K_N}^2\dd\vX\\ &\quad\leq C\left( M_p^2\sigma_{4-2s}(r_0)\tau^{-(3-2s)} +M_b^2\sigma_4(r_0)\tau^{-3}\right)\mathcal T^{\vA}(u). \end{aligned} \end{equation} This argument uses the operator norm of each matrix coefficient and does not require simultaneous diagonalization of the Pauli matrices. For the quantized term, \cref{pauli-magnetic-field-bound} and the boundedness of the Pauli matrices give a constant $C_{\mathrm{ph}}$, depending only on $e$, $N$ and the cutoff norms, such that \[ \|S_{\mathrm{ph}}(\vX)\xi\|^2 \leq C_{\mathrm{ph}}\big(\|H_\rad^{1/2}\xi\|^2+\|\xi\|^2\big) \qquad\big(\xi\in\DC(H_\rad^{1/2})\big), \] uniformly in $\vX$. The radiation term of \cref{pf-covariant-carleman} of \cref{lem-pf-covariant-carleman} has no factor $\beta$. Its radiation and zero-order terms therefore imply \begin{align} &\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2\|S_{\mathrm{ph}}u(\vX)\|_{\mathcal K_N}^2\dd\vX \notag\\ &\quad\leq C_{\mathrm{ph}}r_0^2\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)} \|H_\rad^{1/2}u(\vX)\|_{\mathcal K_N}^2\dd\vX \notag\\ &\qquad+C_{\mathrm{ph}}\sigma_4(r_0)\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)} \beta(t)r^{-2}\|u(\vX)\|_{\mathcal K_N}^2\dd\vX \notag\\ &\quad\leq CC_{\mathrm{ph}}\left(\frac{r_0^2}{\tau} +\frac{\sigma_4(r_0)}{\tau^3}\right)\mathcal T^{\vA}(u). \label{pauli-quantized-smallness} \end{align} In particular, no derivative of $\BF$ occurs in this perturbation argument. It is carried out in the original gauge, after the covariant estimate has been proved, and needs only the existing moment assumption $\omega\wh\chi\in L^2$. As in the proof of \cref{ucp-local-operator-domain-spin}, the local domains used here are $\DC(-\Delta_{\vX})\cap\DC(H_\rad)$. Alternatively, the classical matrix estimate above and the moment inequality \cref{moment-inequality} give infinitesimal operator boundedness relative to $L$. The same conclusion for $S_{\mathrm{ph}}$ follows from its $H_\rad^{1/2}$ bound. Hence all the preceding terms are defined for the vectors in the statement. Let $R_E=V+S_{\mathrm{ext}}+S_{\mathrm{ph}}-E$. Combining \cref{pauli-classical-smallness,pauli-quantized-smallness} yields \[ \int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2\|R_Eu(\vX)\|_{\mathcal K_N}^2\dd\vX \leq\delta(\tau)\mathcal T^{\vA}(u), \] where \[ \delta(\tau)=C\left( M_p^2\sigma_{4-2s}(r_0)\tau^{-(3-2s)} +M_b^2\sigma_4(r_0)\tau^{-3} +C_{\mathrm{ph}}\frac{r_0^2}{\tau} +C_{\mathrm{ph}}\frac{\sigma_4(r_0)}{\tau^3}\right) \longrightarrow0. \] Since $T^{\vA}u=(H^{V,\vA}-E)u-R_Eu$, we obtain \[ \mathcal T^{\vA}(u)\leq2\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|(H^{V,\vA}-E)u(\vX)\|_{\mathcal K_N}^2\dd\vX+2\delta(\tau)\mathcal T^{\vA}(u). \] Choose $\tau_0$ so that $2\delta(\tau)\leq1/2$ for $\tau\geq\tau_0$. Since $\mathcal T^{\vA}(u)<\infty$, subtracting the last term gives \[ \mathcal T^{\vA}(u)\leq4\int_{B_{r_0}(\vX_0)}\e^{2\tau f(t)}r^2 \|(H^{V,\vA}-E)u(\vX)\|_{\mathcal K_N}^2\dd\vX. \] Substitution into the componentwise version of \cref{pf-covariant-carleman} of \cref{lem-pf-covariant-carleman} proves \cref{pauli-carleman}. The construction uses only the stated local norms, which also proves uniformity in the center. \end{proof} \section{Counterexamples at zero coupling}\label[appendix]{app-zero-coupling} We show that the assumption $e>0$ in \cref{thm-hk-nonvanishing} cannot be dropped, either without or with electron spin, even with a nonzero compact ultraviolet cutoff, exponential localization, and nondegenerate ground states. The examples stay within the potential spaces \cref{potential-spaces}. \begin{proposition}[A spinless common ground state for different magnetic fields] \label[proposition]{prop-zero-coupling-counterexample} Let $N=1$, $e=0$, and $Z>0$. The cutoff may be any fixed cutoff satisfying \cref{cutoff-assumption}. Write $r=|\vx|$. For $-10$ if $\wh\chi=0$, when $\AF=0$. \begin{proof}[Proof of \cref{prop-zero-coupling-counterexample}] We have $|\va|\leq r/(1+r^2)\leq1/2$, $\div\va=0$, and $\va\cdot\vx=0$. Thus $\vA_t\in L^\infty$, while $v_t\in L^{3/2}+L^\infty$ by splitting the Coulomb potential at $r=1$. Since $e=0$, expanding the square gives \begin{equation*} H_{\mathrm{sl}}^{v_t,\vA_t}=h_t+H_\rad,\qquad h_t=-\frac12\Delta-\frac Zr+\frac{t}{1+r^2}L_z, \qquad L_z=-\i(x_1\partial_{x_2}-x_2\partial_{x_1}). \end{equation*} The radial function $\phi_Z$ has $L_z\phi_Z=0$. It remains to check that it is the ground state for every $|t|<1$. We expand an electronic wavefunction $\psi$ in its form domain in spherical harmonics, \[ \psi(\vx)=\sum_{l=0}^\infty\sum_{m=-l}^l r^{-1}u_{lm}(r)Y_{lm}(\vx/r),\qquad u_{lm}(r)=r\int_{\mathbb S^2}\ol{Y_{lm}(\vOmega)}\,\psi(r\vOmega)\dd\vOmega. \] Here $Y_{lm}$ are the spherical harmonics. They form an orthonormal basis of $L^2(\mathbb S^2)$ and satisfy $-\Delta_{\mathbb S^2}Y_{lm}=l(l+1)Y_{lm}$ and $L_zY_{lm}=mY_{lm}$. The $u_{lm}$ are the radial coefficients, so that $\|\psi\|^2=\sum_{l,m}\int_0^\infty|u_{lm}(r)|^2\dd r$. Let $w(r)=1/r-Z$ be the logarithmic derivative of the hydrogen $1s$ radial function $r\e^{-Zr}$. For a radial coefficient $u$, completing the square and integrating by parts give \begin{equation}\label{radial-factorization} \begin{split} \mathcal E_Z[u] &:=\frac12\int_0^\infty|u'|^2\dd r -Z\int_0^\infty\frac{|u|^2}{r}\dd r +\frac{Z^2}{2}\int_0^\infty|u|^2\dd r\\ &=\frac12\int_0^\infty |u'-wu|^2\dd r \geq0. \end{split} \end{equation} This follows by integration by parts on a radial core and extension to the form domain. In the $(l,m)$ channel, $-\frac12\Delta$ acts on $u_{lm}$ as $-\frac12\frac{\dd^2}{\dd r^2}+\frac{l(l+1)}{2r^2}$, and $(1+r^2)^{-1}L_z$ as multiplication by $m/(1+r^2)$. Hence \begin{equation}\label{channel-decomposition} \Bigl\langle\psi,\Bigl(h_t+\frac{Z^2}{2}\Bigr)\psi\Bigr\rangle =\sum_{l=0}^\infty\sum_{m=-l}^l\Bigl(\mathcal E_Z[u_{lm}] +\int_0^\infty\Bigl(\frac{l(l+1)}{2r^2}+\frac{tm}{1+r^2}\Bigr) |u_{lm}(r)|^2\dd r\Bigr). \end{equation} The weight in the last integral satisfies \begin{equation}\label{angular-lower-bound} \frac{l(l+1)}{2r^2}+\frac{tm}{1+r^2} \geq(1-|t|)\frac{l(l+1)}{2r^2}\geq0, \end{equation} because $1/(1+r^2)\leq1/r^2$ and $|m|\leq l\leq l(l+1)/2$ for $l\geq1$, while both sides vanish for $l=0$. By \cref{radial-factorization,angular-lower-bound}, every term in \cref{channel-decomposition} is nonnegative, so $h_t\geq-Z^2/2$. For $|t|<1$, the weight in \cref{angular-lower-bound} is positive when $l\geq1$, so equality forces every $l\geq1$ channel to vanish. In the remaining channel, $\mathcal E_Z[u]=0$ forces $u(r)=C r \e^{-Zr}$. The electronic ground-state space is therefore one dimensional and spanned by $\phi_Z$. Since $H_\rad\geq0$ and $\ker H_\rad=\CC\Omega$, the full ground-state space is spanned by $\Psi_Z$. It is not separated from the rest of the spectrum by a positive gap: since $\omega(\vk)=|\vk|$, we have $\spec(H_\rad)=[0,\infty)$, so $[E_t,\infty)\subset\spec(H_{\mathrm{sl}}^{v_t,\vA_t})$. The orbital is real, and at $e=0$ the field-induced current density vanishes, proving \cref{counterexample-data}. Finally, $\grad\times\va$ is clearly not identically zero, so different $t$ give different magnetic fields. \end{proof} To obtain a spinful counterpart, we replace the spherical denominator of $\va$ by a cylindrical one. Its magnetic field then has a fixed direction, and a scalar correction cancels the Zeeman term in the spin-down sector. \begin{proposition}[A spinful common ground state for different magnetic fields] \label[proposition]{prop-spinful-zero-coupling-counterexample} Let $N=1$, $e=0$, and $Z>0$. The cutoff may be any fixed cutoff satisfying \cref{cutoff-assumption}. Write $r=|\vx|$ and $s=(x_1^2+x_2^2)^{1/2}$. For $00$ if $\wh\chi=0$. At $t=0$ the same state is still a ground state, but the ground-state space is two dimensional because of spin. \begin{proof}[Proof of \cref{prop-spinful-zero-coupling-counterexample}] Set $f(s)=(1+s^2)^{-1}$ and $h_Z=-\Delta/2-Z/r$. The vector field $\va_\perp$ is smooth, $|\va_\perp|=s/(1+s^2)\leq1/2$, and $\div\va_\perp=0$. Direct differentiation gives \cref{spinful-counterexample-field}. Hence the vector potential and its curl are bounded, and the scalar potential is Coulomb plus a bounded function. This proves the stated global admissibility and local regularity. The Pauli identity and the scalar correction in \cref{spinful-counterexample-potentials} give the form decomposition \begin{equation*} H^{v_t,\vA_t}=h_t^{\mathrm{spin}}+H_\rad,\qquad h_t^{\mathrm{spin}} =\bigl(h_Z+tf(s)L_z\bigr)\iden_{\CC^2} +2tf(s)^2P_+, \end{equation*} where $P_+=(\iden_{\CC^2}+\sigma^3)/2$ and $L_z=-\i(x_1\partial_{x_2}-x_2\partial_{x_1})$. Since $L_z\phi_Z=0$ and $P_+(0,\phi_Z)^{\mathsf T}=0$, the displayed spinor is an eigenstate with energy $-Z^2/2$. For a scalar $\psi\in H^1(\RR^3)$, define the nonnegative form \begin{equation*} Q_Z[\psi] =\Bigl\langle\psi,\Bigl(h_Z+\frac{Z^2}{2}\Bigr)\psi\Bigr\rangle =\frac12\int_{\RR^3} \left|\grad\psi(\vx)+Z\frac{\vx}{r}\psi(\vx)\right|^2\dd\vx. \end{equation*} The identity follows by integration by parts from $\div(\vx/r)=2/r$ in the sense of distributions, first on a smooth compact core and then on $H^1$ by form continuity. Moreover, $Q_Z[\psi]=0$ precisely when $\psi$ is a scalar multiple of $\phi_Z$, since equality implies $\grad(\e^{Zr}\psi(\vx))=0$ locally. In cylindrical coordinates $(s,\theta,z)$, $L_z=-\i\partial_\theta$. Expansion in the angular Fourier modes $\e^{\i m\theta}$ and $|m|\leq m^2$ for $m\in\ZZ$ yield \begin{equation*} \left|\dua{\psi}{f(s)L_z\psi}\right| \leq\int_{\RR^3}f(s)|\partial_\theta\psi(\vx)|^2\dd\vx \leq\int_{\RR^3}\frac{|\partial_\theta\psi(\vx)|^2}{s^2}\dd\vx \leq2Q_Z[\psi]. \end{equation*} For the last inequality, note that $s^{-1}\partial_\theta\psi(\vx)$ is the azimuthal component of $\grad\psi(\vx)+Z(\vx/r)\psi(\vx)$. These inequalities hold first on the smooth core and extend to $H^1$, since $f(s)L_z=-\i\va_\perp\cdot\grad$ has bounded coefficients. Consequently, for a spinor $\psi=(\psi_+,\psi_-)^{\mathsf T}\in H^1(\RR^3,\CC^2)$, \begin{equation*} \begin{split} \Bigl\langle\psi, \Bigl(h_t^{\mathrm{spin}}+\frac{Z^2}{2}\Bigr)\psi\Bigr\rangle &\geq(1-2t)\bigl(Q_Z[\psi_+]+Q_Z[\psi_-]\bigr) +2t\int_{\RR^3}f(s)^2|\psi_+(\vx)|^2\dd\vx\\ &\geq0. \end{split} \end{equation*} Since $00$, equality forces $\psi_+(\vx)=0$ and $\psi_-(\vx)=C\phi_Z(\vx)$. Thus the electronic ground state is unique up to phase. As $H_\rad\geq0$ and $\ker H_\rad=\CC\Omega$, the same is true of the full ground state \cref{spinful-counterexample-state}. The orbital is real and $e=0$, so $\vjp=\vjph=0$. The spin-down polarization gives $\vm(\vx)=-\rho(\vx)\ve_3$, and $\grad\rho(\vx)=-2Z\rho(\vx)\vx/r$ gives \cref{spinful-counterexample-data}. The exponential moments follow directly from the displayed density. Finally, \cref{spinful-counterexample-field} shows that distinct parameters give different curls, which excludes even a distributional gradient as their difference. \end{proof} \enlargethispage{3\baselineskip} \begin{thebibliography}{99} \bibitem{carleman1939probleme} T. Carleman, \emph{Sur un probl\`eme d'unicit\'e pour les syst\`emes d'\'equations aux d\'eriv\'ees partielles \`a deux variables ind\'ependantes}, Arkiv f\"or Matematik, Astronomi och Fysik \textbf{26B} (1939), no.~17, 1--9. \bibitem{jerison1985unique} D. Jerison and C.~E. Kenig, \emph{Unique continuation and absence of positive eigenvalues for Schr\"odinger operators}, Annals of Mathematics \textbf{121} (1985), 463--494. \bibitem{wolff1993recent} T.~H. Wolff, \emph{Recent work on sharp estimates in second-order elliptic unique continuation problems}, Journal of Geometric Analysis \textbf{3} (1993), 621--650. \bibitem{kurata1997unique} K. Kurata, \emph{A unique continuation theorem for the Schr\"odinger equation with singular magnetic field}, Proceedings of the American Mathematical Society \textbf{125} (1997), 853--860. \bibitem{koch2001carleman} H. Koch and D. Tataru, \emph{Carleman estimates and unique continuation for second-order elliptic equations with nonsmooth coefficients}, Communications on Pure and Applied Mathematics \textbf{54} (2001), 339--360. \bibitem{hohenberg1964inhomogeneous} P. Hohenberg and W. Kohn, \emph{Inhomogeneous electron gas}, Physical Review \textbf{136} (1964), B864--B871. \href{https://doi.org/10.1103/PhysRev.136.B864}{doi\ 10.1103/PhysRev.136.B864}. \bibitem{lieb1983density} E.~H. Lieb, \emph{Density functionals for Coulomb systems}, International Journal of Quantum Chemistry \textbf{24} (1983), 243--277. \bibitem{lewin2023universal} M. Lewin, E.~H. Lieb, and R. Seiringer, \emph{Universal functionals in density functional theory}, in: \emph{Density Functional Theory} (E. Canc\`es and G. Friesecke, eds.), Mathematics and Molecular Modeling, Springer, Cham, 2023, 115--182. \href{https://doi.org/10.1007/978-3-031-22340-2_3}{doi\ 10.1007/978-3-031-22340-2\_3}. \bibitem{corso2026maximal} T.~C. Corso, \emph{A maximal Hohenberg--Kohn theorem for non-interacting systems via potential theory}, preprint (2026). \href{https://arxiv.org/abs/2607.12852}{arXiv:2607.12852}. \bibitem{corso2025torus} T.~C. Corso, \emph{$v$-representability and Hohenberg--Kohn theorem for non-interacting Schr\"odinger operators with distributional potentials on the one-dimensional torus}, Journal of Physics A: Mathematical and Theoretical \textbf{58} (2025), 125203. \href{https://doi.org/10.1088/1751-8121/adc04c}{doi\ 10.1088/1751-8121/adc04c}. \bibitem{corso2026spinless} T.~C. Corso, \emph{A rigorous formulation of density functional theory for spinless fermions in one dimension}, Letters in Mathematical Physics \textbf{116} (2026), 27. \href{https://doi.org/10.1007/s11005-026-02051-1}{doi\ 10.1007/s11005-026-02051-1}. \bibitem{garrigue2018unique} L. Garrigue, \emph{Unique continuation for many-body Schr\"odinger operators and the Hohenberg--Kohn theorem}, Mathematical Physics, Analysis and Geometry \textbf{21} (2018), 27. \href{https://doi.org/10.1007/s11040-018-9287-z}{doi\ 10.1007/s11040-018-9287-z}. \bibitem{garrigue2020unique} L. Garrigue, \emph{Unique continuation for many-body Schr\"odinger operators and the Hohenberg--Kohn theorem. II. The Pauli Hamiltonian}, Documenta Mathematica \textbf{25} (2020), 869--898. \href{https://doi.org/10.4171/DM/765}{doi\ 10.4171/DM/765}. \bibitem{defigueiredo1992strict} D.~G. de Figueiredo and J.-P. Gossez, \emph{Strict monotonicity of eigenvalues and unique continuation}, Communications in Partial Differential Equations \textbf{17} (1992), 339--346. \bibitem{rajagopal1994time} A.~K. Rajagopal, \emph{Time-dependent functional theory of coupled electron and electromagnetic fields in condensed-matter systems}, Physical Review A \textbf{50} (1994), 3759--3765. \href{https://doi.org/10.1103/PhysRevA.50.3759}{doi\ 10.1103/PhysRevA.50.3759}. \bibitem{ruggenthaler2011time} M. Ruggenthaler, F. Mackenroth, and D. Bauer, \emph{Time-dependent Kohn--Sham approach to quantum electrodynamics}, Physical Review A \textbf{84} (2011), 042107. \href{https://doi.org/10.1103/PhysRevA.84.042107}{doi\ 10.1103/PhysRevA.84.042107}. \bibitem{ruggenthaler2014quantum} M. Ruggenthaler, J. Flick, C. Pellegrini, H. Appel, I.~V. Tokatly, and A. Rubio, \emph{Quantum-electrodynamical density-functional theory: Bridging quantum optics and electronic-structure theory}, Physical Review A \textbf{90} (2014), 012508. \href{https://doi.org/10.1103/PhysRevA.90.012508}{doi\ 10.1103/PhysRevA.90.012508}. \bibitem{ruggenthaler2015ground} M. Ruggenthaler, \emph{Ground-state quantum-electrodynamical density-functional theory}, preprint (2015). \href{https://arxiv.org/abs/1509.01417}{arXiv:1509.01417}. \bibitem{tokatly2013time} I.~V. Tokatly, \emph{Time-dependent density functional theory for many-electron systems interacting with cavity photons}, Physical Review Letters \textbf{110} (2013), 233001. \href{https://doi.org/10.1103/PhysRevLett.110.233001}{doi\ 10.1103/PhysRevLett.110.233001}. \bibitem{Q} M.~A. Csirik, A. Laestadius, M. Penz, and E.~I. Tellgren, \emph{Density functionals for the Pauli--Fierz Hamiltonian}, preprint (2026). \href{https://arxiv.org/abs/2611.01208}{arXiv:2611.01208}. \bibitem{lieb2010stability} E.~H. Lieb and R. Seiringer, \emph{The Stability of Matter in Quantum Mechanics}, Cambridge University Press, Cambridge, 2010. \href{https://doi.org/10.1017/CBO9780511819681}{doi\ 10.1017/CBO9780511819681}. \bibitem{kato1995perturbation} T. Kato, \emph{Perturbation Theory for Linear Operators}, Classics in Mathematics, Springer, Berlin, 1995. \href{https://doi.org/10.1007/978-3-642-66282-9}{doi\ 10.1007/978-3-642-66282-9}. \bibitem{hasler2008selfadjoint} D. Hasler and I. Herbst, \emph{On the self-adjointness and domain of Pauli--Fierz type Hamiltonians}, Reviews in Mathematical Physics \textbf{20} (2008), 787--800. \href{https://doi.org/10.1142/S0129055X08003389} {doi\ 10.1142/S0129055X08003389}. \bibitem{griesemer2004exponential} M. Griesemer, \emph{Exponential decay and ionization thresholds in non-relativistic quantum electrodynamics}, Journal of Functional Analysis \textbf{210} (2004), 321--340. \href{https://doi.org/10.1016/j.jfa.2003.06.001}{doi\ 10.1016/j.jfa.2003.06.001}. \bibitem{penz2023structureII} M. Penz, E.~I. Tellgren, M.~A. Csirik, M. Ruggenthaler, and A. Laestadius, \emph{The structure of the density-potential mapping. Part II. Including magnetic fields}, ACS Physical Chemistry Au \textbf{3} (2023), 492--511. \href{https://doi.org/10.1021/acsphyschemau.3c00006}{doi\ 10.1021/acsphyschemau.3c00006}. \end{thebibliography} \end{document}