\section{Known results: a bath in its ground state gives no rate} \label{sec:known} The statements in this section are standard results or direct consequences of them; the mathematics is the zero-temperature fluctuation--dissipation theorem and the spectral support of Wightman functions. We state them with their hypotheses because the drag theorem of Section~\ref{sec:drag} addresses what they leave open. Throughout, the bath is kept to second cumulant and the static source enters as an external c-number density. \subsection{Any ground state: the noise is the absorption} Let $H\geq0$ generate a time translation of the laboratory under which the bath state and both branch configurations are stationary, and let $\Omega$ be a ground state, $H\Omega=0$. No spatial symmetry is assumed: the Earth, the apparatus and the source may break every spatial translation. The proposition says that the noise then equals the absorption at each frequency, and that a soft enough absorption bounds the loss of coherence for all time. \begin{proposition}[Ground state; standard] \label{prop:ground} (i) $S_B(\omega)=\hbar|\chi''_B(\omega)|$ for $\omega\neq0$, and $S_B(0)=\hbar\chi''_B(0^+)$ when the limit exists. (ii) The rate vanishes if and only if the Fej\'{e}r mean of $\chi''_B$ at $\omega=0$ vanishes and the spectral measure of $\delta B\,\Omega$ has no atom at $\omega=0$. (iii) If $\Lambda_2\equiv\pi^{-1}\int_0^\infty\chi''_B(\omega)\,\omega^{-2}d\omega$ is finite, then \begin{equation} -\ln|z(t)|\leq\frac{2\Lambda_2}{\hbar}\quad\text{for all }t,\qquad -\ln|z(t)|\xrightarrow{\;t\to\infty\;}\frac{\Lambda_2}{\hbar}. \label{eq:all-times} \end{equation} \end{proposition} \begin{proof} With $H\Omega=0$ and $H\geq0$, the spectral measure of $\delta B\,\Omega$ lives on $E\geq0$, so $\langle\delta B(t)\delta B\rangle$ has its Fourier support at $\omega\geq0$ and $\langle\delta B\,\delta B(t)\rangle$ at $\omega\leq0$; (i) is the zero-temperature fluctuation--dissipation theorem~\cite{CallenWelton1951,Kubo1966}. Then Eq.~\eqref{eq:dephasing} reads $-\ln|z(t)|=(\pi\hbar)^{-1}\int_0^\infty\chi''_B(\omega)(1-\cos\omega t)\,\omega^{-2}d\omega$. Substituting $x=\omega t$ in $-\ln|z|/t$ gives (ii); an atom at $\omega=0$ would add a term growing as $t^2$. Bounding $1-\cos\omega t\leq2$ and applying the Riemann--Lebesgue lemma gives (iii). \end{proof} \noindent In (ii), the Fej\'{e}r mean of $\chi''_B$ at $\omega=0$ is $\lim_{t\to\infty}\int_0^\infty\chi''_B(\omega)\,2(1-\cos\omega t)/(\pi t\,\omega^2)\,d\omega$; it equals $\chi''_B(0^+)$ when that limit exists. The proposition means that a ground-state bath produces a rate only if its absorption stays finite down to zero frequency. Since $\chi''_B$ is odd in $\omega$, that requires a response that is discontinuous at $\omega=0$: a Di\'{o}si--Penrose rate would need $\chi''_B(0^+)=2C\EG$, that is, a response tail $\chi_B(t)\simeq4C\EG/(\pi t)$ of the bath to the which-path variable. This is the loophole that the proposition leaves. For $\chi''_B\propto\omega^s$ at small $\omega$, a rate needs $s\leq0$; Ohmic baths, $s=1$, give $-\ln|z|\propto\ln t$, and only $s>1$ gives a bounded exponent~\cite{Unruh1995,Leggett1987,PalmaSuominenEkert1996}. A finite reorganization energy $\pi^{-1}\int_0^\infty\chi''_B\,\omega^{-1}d\omega$ suffices for (ii), but it does not bound decoherence on laboratory times: a $\chi''_B$ that is flat over a band $[\omega_{\rm IR},\omega_{\rm UV}]$ and normalized to reorganization energy $\EG$ mimics Eq.~\eqref{eq:dp-rate} with $C_{\rm eff}=\pi/[2\ln(\omega_{\rm UV}/\omega_{\rm IR})]$ for $1/\omega_{\rm UV}\ll t\ll1/\omega_{\rm IR}$, which is $0.07$ between $0.1$~ps and $1$~ms. The all-times bound~\eqref{eq:all-times} is the statement that excludes this. \subsection{The free linearized gravitational field} In a thermal equilibrium (Kubo--Martin--Schwinger, KMS) state at inverse temperature $\beta$, the fluctuation--dissipation theorem gives $S_B(\omega)=\hbar\chi''_B(\omega)\coth(\beta\hbar\omega/2)$, so \begin{equation} S_B(0)=\frac{2}{\beta}\lim_{\omega\to0}\frac{\chi''_B(\omega)}{\omega}: \label{eq:fdt-floor} \end{equation} a floor at finite temperature needs an Ohmic response. The vacuum restricted to a Rindler wedge~\cite{BisognanoWichmann1975}, or for a conformal field to a causal diamond~\cite{CasiniHuertaMyers2011}, is KMS for its modular flow; describing the vacuum in modular time adds no excitations. A free massless field obeys $\Box\hat\phi=0$, so its commutator, and hence $\chi''_B$, is a c-number supported on $\omega=\pm c|\mathbf{k}|$, the same in every state. Inhomogeneity of the laboratory changes the propagator only at order $GM_\oplus/(R_\oplus c^2)=7\times10^{-10}$. Weight at $\omega\to0$ comes only from $k\to0$. For two equal spheres with form factor $\tilde\rho(k)$, \begin{equation} \bigl\langle|\Delta\tilde\rho_{\mathbf{k}}|^2\bigr\rangle =2\,\tilde\rho(k)^2\Bigl(1-\frac{\sin kd}{kd}\Bigr) \;\xrightarrow{\;kd\to0\;}\;\tfrac13 M^2k^2d^2 . \label{eq:drho} \end{equation} Take as a scalar proxy a propagating field normalized so that its reorganization energy, $\pi^{-1}\int_0^\infty\chi''_B\,\omega^{-1}d\omega$, equals $\EG$ (Appendix~\ref{app:energy}). Then \begin{equation} \chi''_B(\omega)=\frac{G\omega}{c}\bigl\langle|\Delta\tilde\rho_{\omega/c}|^2\bigr\rangle \;\simeq\;\frac{GM^2d^2}{3c^3}\,\omega^3\qquad(\omega d/c\ll1), \label{eq:J} \end{equation} which is super-Ohmic. Only a difference of branch \emph{energies} makes it Ohmic. That is the channel of the graviton master equations~\cite{Anastopoulos2013,Blencowe2013,OnigaWang2016}, whose rates are first order in $G$ for superpositions of different energies and vanish for the equal-mass spatial superposition considered here. Linearized general relativity has less infrared response than the proxy. Its Newtonian potential is fixed by the constraint, an instantaneous real kernel with $\chi''=0$ that fluctuates only through the matter that sources it, at order $G^2$. Its radiative tensor modes couple through the trace-free quadrupole difference of the branches, $\chi''_{\rm TT}=(G/5c^5)\,|\Delta Q^{\rm TF}|^2\,\omega^5$, which for $|\Delta Q^{\rm TF}|^2=\frac23M^2d^4$ is $\frac25(\omega d/c)^2$ times Eq.~\eqref{eq:J}. The response of the field is therefore too soft at low frequency for a floor, in the vacuum and at any temperature: \begin{proposition}[Field channel] \label{prop:field} For a static inertial source with equal branch masses, $S_B(0)=0$ in the vacuum and in every thermal state of the free linearized field, including states that are KMS for modular flow, and more generally in every stationary Gaussian state whose mode occupation obeys $\lim_{\omega\to0^+}n(\omega)\,\chi''_B(\omega)=0$. The dephasing exponent saturates. \end{proposition} \begin{proof} For a free field $S_B=\hbar\chi''_B\,[1+2n(\omega)]$ with $\chi''_B$ state-independent and $\propto\omega^3$ or softer at small $\omega$. The vacuum, $n=0$, is Proposition~\ref{prop:ground}. A thermal state has $2n\simeq2/(\beta\hbar\omega)$, so $S_B=\order{\omega^2}$. The general case is the stated condition. In each case $S_B(\omega)/\omega^2$ is integrable at $\omega=0$ and Eq.~\eqref{eq:dephasing} converges as $t\to\infty$. \end{proof} \noindent The free field gives a static equal-mass superposition no rate, whatever its temperature. The occupation condition is not automatic. A statistically homogeneous stochastic gravitational-wave background with strain spectrum $S_h\propto\omega^{-4}$ violates it and gives a white floor at first order in $G$, but $\propto M^2d^4$, not of Di\'{o}si--Penrose form. In the tensor channel such a background has an infrared-divergent energy density, so it can only be band-limited. At closure density at $1$~Hz its rate is $3\times10^{-8}$ of Eq.~\eqref{eq:dp-rate} for the benchmark of Section~\ref{sec:prediction} at $d=1$~mm. For the vacuum, Proposition~\ref{prop:ground}(iii) makes the statement quantitative. For $1~\mu$g of silica the proxy~\eqref{eq:J} gives $\Lambda_2/\hbar=1.1\times10^{-3}$ at $d=2a$ and $4.2\times10^{-3}$ at $d=1$~mm, of order $GM^2/(\hbar c)=2.1\times10^{-3}$; the tensor channel gives less. The coherence loss is bounded at all times and is reversible: it is the overlap of the field states dressing the two branches, not a rate. Direct calculations agree. In the graviton vacuum, the decoherence functional of a spatial superposition that is opened and closed again approaches a constant at late times rather than growing linearly~\cite{KannoSodaTokuda2021}. The which-path information radiated by a superposition is set by the time taken to recombine it, not by the time it is held apart~\cite{DSW2022newton}, and in the inertial Minkowski vacuum it does not grow with the holding time~\cite{DSW2023}. \subsection{The covariant case, and what these results leave open} If the state is invariant under spacetime translations and the energy--momentum spectrum lies in the forward cone, the zero follows without computing any response: \begin{proposition}[Spectrum condition; standard] \label{prop:spectrum} Let $\Omega$ be invariant under spacetime translations $U(a)=e^{iP\cdot a}$, let the joint spectrum of $P^\mu$ lie in the closed forward cone $\bar V_+$, let $\Phi(x)=U(x)\Phi(0)U(x)^\dagger$ (local or gravitationally dressed) have a locally finite connected two-point spectral measure $\mu$, and let the source be rigid. Then $S_B(0)=0$. \end{proposition} \begin{proof} With $P|\Omega\rangle=0$, $\langle\Omega|\Phi(x)\Phi(0)|\Omega\rangle_c=\int e^{-ip\cdot x}\,d\mu(p)$ with $\mathrm{supp}\,\mu\subset\bar V_+$. The zero-frequency noise of $B$ is the density at $p^0=0$ of $|\Delta\tilde\rho(\mathbf{p})|^2d\mu(p)$ and its mirror image, $S_B(0)=c_0\lim_{\varepsilon\to0}\varepsilon^{-1}\int_{0\leq p^0\leq\varepsilon} |\Delta\tilde\rho|^2\,d\mu$, with $c_0$ a convention-dependent constant. On the support $|\mathbf{p}|\leq\varepsilon/c$, and a rigid body has $|\Delta\tilde\rho|^2=4|\tilde\rho|^2\sin^2(\mathbf{k}\cdot\mathbf{d}/2)\leq M^2k^2d^2$, so \begin{equation} S_B(0)\leq c_0\lim_{\varepsilon\to0}\frac{M^2d^2}{c^2}\,\varepsilon\, \mu\bigl(\{p^0\leq\varepsilon\}\bigr)=0 \label{eq:nogo2-bound} \end{equation} by local finiteness. \end{proof} \noindent Since the cone implies $P^0\geq0$, Proposition~\ref{prop:spectrum} is the special case of Proposition~\ref{prop:ground} in which the cone and the dipole suppression of an equal-mass source supply $\chi''_B(0^+)=0$. An atom of $\mu$ at $p=0$ is weighted by $|\Delta\tilde\rho(0)|^2=0$, so it needs no separate assumption. This is a standard consequence of the spectral support of Wightman functions~\cite{Wightman1956,Haag1996} combined with the dipole suppression of an equal-mass source. The Lorentz-cone support forces, for any covariant field coupled to a source with vanishing monopole difference, the conclusion that the theory of dephasing reaches for Ohmic and super-Ohmic baths~\cite{Unruh1995,Leggett1987,PalmaSuominenEkert1996}. In asymptotically flat gravity the cone would follow from positivity of the total energy, which we do not derive. The propositions leave open whatever breaks one of their hypotheses. \begin{itemize} \item \emph{The state is not a ground state.} In a thermal, stationary non-equilibrium or infinite-temperature state the generator of time translations that leaves the state invariant is not bounded below; for a thermal state of an infinite system its spectrum is all of $\mathbb{R}$. This is the class of Section~\ref{sec:drag}. Relativistic collapse models escape the same way: a mediating field at each spacetime point~\cite{Bedingham2011}, a scalar field whose energy spectrum takes all real values~\cite{Pearle2015}, a clock field that fixes the foliation~\cite{VaidhyanathanMilburn2025}, or classical noise with a Lorentz-invariant correlation, which as a classical field need not have a positive generator. Why such noise is hard to reconcile with a stable vacuum is the subject of a long literature~\cite{Pearle1999,Pearle2005,Tumulka2006,Myrvold2017,JonesGasbarriBassi2021,Diosi2022GKLS,Diosi2024hybrid}. Gundhi, Di\'{o}si and Carlesso obtain a finite energy rate for Lorentz-invariant, non-Markovian noise~\cite{GundhiDiosiCarlesso2026}; their dynamics still excites the vacuum, as they note, and the finite rate comes from a normal-ordering prescription applied to the evolved observables, which is not a positive map on states. The model therefore supplies no stationary positive-energy bath with a floor, and its kernel is not of Di\'{o}si--Penrose form. \item \emph{Killing horizons} replace the positive generator of laboratory time by a boost generator with spectrum $\mathbb{R}$. They decohere at rates set by horizon data that vanish for an inertial laboratory without a horizon~\cite{DSW2023,DSW2024local,DSW2022,WilsonGerow2024}. \item \emph{Non-inertial laboratories.} The generator of the co-rotating frame is not bounded below. A source fixed on the rotating Earth radiates branch-dependently at a rate of order $10^{-49}$ of Eq.~\eqref{eq:dp-rate} in the scalar proxy, and less in general relativity, where the dipole does not radiate. \item \emph{A quantized source.} Internal dynamics that diffuse the source's trajectory lie outside the c-number treatment~\cite{MoreiraCeleri2026}. So does the source's own Gauss-law sector, where boosts are spontaneously broken; we conjecture that its corrections are of higher order in $GM$ than the leading term, but have not shown it. \item \emph{Beyond second cumulant}: higher cumulants of the bath are not controlled by these propositions. \item \emph{Non-unitary dynamics}: Penrose-type objective reduction~\cite{Penrose1996} and collapse models whose record is a sequence of jumps with a stable vacuum~\cite{Tilloy2018,PiccioneBassi2025} lie outside the bath framework. \end{itemize}