\section{The drag theorem and the double-pulsar bound} \label{sec:drag} This section makes the argument of Section~\ref{sec:idea} precise for a bath with a zero-frequency floor. Theorem~\ref{thm:drag} is the Einstein relation, the classical content of the fluctuation--dissipation theorem~\cite{CallenWelton1951,Kubo1966} and of the link between decoherence and friction in quantum Brownian motion~\cite{Caldeira1983}, applied to Di\'{o}si--Penrose noise. What is new is the bound from binary neutron stars and the identification of $\Teff/C$ as the parameter that the orbital decay bounds. \subsection{The drag theorem: friction from the noise and one temperature} Let $H_{\rm int}=\int\rho\,\Phi\,d^3x$ with $\Phi$ a bath operator, and let $\Phi_{\mathbf{k}}$ be its spatial Fourier transform. Per unit volume define the symmetrized noise and the dissipative response \begin{equation} S(\mathbf{k},\omega)=\frac12\int\!dt\,e^{i\omega t}\bigl\langle\{\Phi_{\mathbf{k}}(t),\Phi_{-\mathbf{k}}\}\bigr\rangle,\qquad \chi''(\mathbf{k},\omega)=\frac{1}{2\hbar}\int\!dt\,e^{i\omega t}\bigl\langle[\Phi_{\mathbf{k}}(t),\Phi_{-\mathbf{k}}]\bigr\rangle . \label{eq:S-chi} \end{equation} Fix a wavenumber $\kstar$ and a speed $v_{\max}$, and assume: \begin{list}{}{\setlength{\labelwidth}{2.6em}\setlength{\labelsep}{0.5em}% \setlength{\leftmargin}{3.1em}} \item[(H)] \emph{Reciprocity:} the operator $\rho$ couples to is the one that carries the noise, through a Hamiltonian (or Schwinger--Keldysh) coupling, so $S$ and $\chi''$ come from the same correlator. \item[(S)] \emph{Stationarity:} the bath state is stationary, homogeneous and isotropic in a rest frame, at scales $\geq1/\kstar$. \item[(DP)] \emph{Di\'{o}si--Penrose noise:} $S(k,0)=4\pi C\hbar G/k^2$ for $k\leq\kstar$. \item[(W)] \emph{Whiteness over the drift band:} $\chi''(k,\omega)/\omega$ equals its $\omega\to0$ value for $|\omega|\leq\kstar v_{\max}$, $k\leq\kstar$. \item[(F)] \emph{Finite positive temperature:} $\Teff(k)\equiv S(k,0)/\bigl(2\kB\,\partial_\omega\chi''(k,0)\bigr)\in(0,\infty)$ for $k\leq\kstar$, with $\Teff\equiv\sup_{k<\kstar}\Teff(k)<\infty$; a KMS state at temperature $T$ has $\Teff\equiv T$. \item[(P)] \emph{Passivity:} $\omega\chi''(k,\omega)\geq0$ at every $k$ and $\omega$. This holds in every KMS state at $T>0$; for other states it is a separate assumption. \item[(L)] \emph{Linear response:} the bath responds linearly to the body's density, so the friction is linear in the velocity for all $|\mathbf{v}|\leq v_{\max}$. \end{list} $\Teff$ is a fluctuation--dissipation parameter, not the thermodynamic temperature of any identified degrees of freedom. The theorem bounds the friction on a moving body from below, using only the noise at wavenumbers below $\kstar$. \begin{theorem}[Drag] \label{thm:drag} Under (H), (S), (DP), (W), (F), (P) and (L), a body moving with velocity $\mathbf{v}$, $|\mathbf{v}|\leq v_{\max}$, relative to the rest frame feels $\mathbf{F}=-M\gamma\mathbf{v}$ with \begin{equation} \gamma\;\geq\;\frac{2\pi}{3}\,\frac{C\hbar G}{\kB\Teff}\;\rho_{\kstar},\qquad \rho_{\kstar}\equiv\frac1M\int_{k<\kstar}\frac{d^3k}{(2\pi)^3}\,|\rho_{\mathbf{k}}|^2 . \label{eq:drag} \end{equation} \end{theorem} \begin{proof} Uniform motion gives $\rho_{\mathbf{k}}(t)=\rho_{\mathbf{k}}e^{-i\mathbf{k}\cdot\mathbf{v}t}$, so the body drives mode $\mathbf{k}$ at $\omega=\mathbf{k}\cdot\mathbf{v}$. By (H), (S) and (L), $\langle\Phi_{\mathbf{k}}\rangle=-\chi(\mathbf{k},\mathbf{k}\cdot\mathbf{v}) \rho_{\mathbf{k}}$, and the self-force is \begin{equation} \mathbf{F}=-\int\rho\,\nabla\langle\Phi\rangle\,d^3x =-\int\frac{d^3k}{(2\pi)^3}\,\mathbf{k}\,\chi''(\mathbf{k},\mathbf{k}\cdot\mathbf{v})\,|\rho_{\mathbf{k}}|^2 , \label{eq:selfforce} \end{equation} because $\chi'$ is even in $\omega$ and drops out against the odd factor $\mathbf{k}$. The power dissipated in mode $\mathbf{k}$, $(\mathbf{k}\cdot\mathbf{v})\chi''(\mathbf{k},\mathbf{k}\cdot\mathbf{v})|\rho_{\mathbf{k}}|^2$, is non-negative by (P), so restricting the integral to $k<\kstar$ can only lower the drag. There, (W) gives $\chi''(\mathbf{k},\mathbf{k}\cdot\mathbf{v})=(\mathbf{k}\cdot\mathbf{v})\, \partial_\omega\chi''(k,0)$; isotropy replaces $k_ik_j$ by $\delta_{ij}k^2/3$; and (F) with (DP) gives \begin{equation} \partial_\omega\chi''(k,0)=\frac{S(k,0)}{2\kB\Teff(k)}\geq\frac{2\pi C\hbar G}{\kB\Teff\,k^2}. \label{eq:slope} \end{equation} The factor $k^2$ cancels, and Eq.~\eqref{eq:drag} follows. \end{proof} \noindent The theorem says that the friction rate is the noise strength divided by the temperature, times the density of the body within the band $k<\kstar$. No other property of the bath enters. \begin{remark}[Reading the bound] \label{rem:drag} (i) In classical form $\gamma=(4\pi/3)G\rho_{\kstar}/\kappa_{\rm FDR}$ with the fluctuation--dissipation rate $\kappa_{\rm FDR}\equiv2\kB\Teff/(C\hbar)$, in which $\hbar$ cancels. The laboratory rate fixes $C$; the drag bounds $\Teff/C$. If $\Teff(k)$ is unbounded on $k<\kstar$, (F) fails and the bound degenerates; this is escape (E4) of Section~\ref{sec:gap}. (ii) Writing the momentum diffusion as $D_p=\frac12\int k_ik_j\,S(k,0)|\rho_{\mathbf{k}}|^2$, Eq.~\eqref{eq:drag} is the Einstein relation $\gamma=D_p/(M\kB\Teff)$. A Markovian model whose centre-of-mass momentum relaxes to a Maxwellian at $\Teff$ is covered, whether or not a bath is written down, provided its friction is linear in the velocity up to orbital speeds and its noise satisfies (DP), (W) and (P) at $k\leq\kstar$. (iii) At $\kstar=10/R$, $\rho_{\kstar}/\bar\rho=0.909$ for a uniform sphere of radius $R$ and mean density $\bar\rho$, $1.42$ for $\rho\propto1-r^2$ and $1.64$ for an $n=1$ polytrope. The uniform sphere is the conservative case. (iv) Treating a body as rigid with linear friction requires its centre-of-mass motion not to be thermal in the bath, $Mv^2\gg\kB\Teff$, so that the fluctuating force is negligible against the drag. Many-body versions of dissipative collapse can differ: collective pair friction in a Fock-space formulation gives a stationary per-particle temperature of order $2T_\beta/N$ at collapse temperature $T_\beta$~\cite{Gomes2026}. Under the stated condition such effects only lower the centre-of-mass temperature, and the bound tightens. (v) By Kramers--Kronig, (P), (W) and (F) also imply a static response $\chi'(k,0)\geq(2/\pi)\int_0^{\kstar v_{\max}}\chi''\,\omega^{-1}d\omega$, a renormalization of Newton's constant with $\delta G/G\geq C\hbar\kstar v_{\max}/(\pi\kB\Teff)$, which is $5\times10^{-26}$ at the bound below: the bath does not otherwise modify Newton's law observably. \end{remark} The next lemma converts a friction rate into the observable of pulsar timing, the rate of change $\dot P_b$ of the orbital period $P_b$. \begin{lemma}[Orbital decay] \label{lem:orbit} Two bodies of masses $M_1,M_2$ with drags $\gamma_1,\gamma_2$ in a bound orbit lose orbital energy at the relative rate $\Gamma_{\rm orb}=(\gamma_1M_2+\gamma_2M_1)/(M_1+M_2)\geq\min\gamma_i$, and $\dot P_b/P_b=-3\Gamma_{\rm orb}$. \end{lemma} \begin{proof} Write $M=M_1+M_2$, $\mu=M_1M_2/M$ and $\mathbf{v}$ for the relative velocity. In the centre-of-mass frame $\mathbf{v}_1=(M_2/M)\mathbf{v}$ and $\mathbf{v}_2=-(M_1/M)\mathbf{v}$, so the drag on the relative velocity is $-\Gamma_{\rm orb}\mathbf{v}$. The velocity $\mathbf{w}$ of the centre of mass relative to the rest frame enters only as the constant differential acceleration $-(\gamma_1-\gamma_2)\mathbf{w}$, which does no secular work on the orbital energy at linear order. With $\dot E=-\mu\Gamma_{\rm orb}v^2$ and Kepler's law, $\dot a=-2\Gamma_{\rm orb}a$ for the semi-major axis, and $P_b\propto a^{3/2}$ gives $\dot P_b/P_b=-3\Gamma_{\rm orb}$, independent of the eccentricity to this order. \end{proof} \subsection{The double-pulsar bound on \texorpdfstring{$\Teff/C$}{T\_eff/C}} By the lemma, a limit on any unexplained shortening of the period is a limit on the friction. PSR J0737$-$3039A/B has $P_b=0.1022515593$~d and observed $\dot P_b=-1.247920(78)\times10^{-12}$; after subtracting the external terms its orbital decay agrees with the quadrupole formula to $1.3\times10^{-4}$ at $95\%$ confidence~\cite{Kramer2021}. Any additional decay therefore obeys \begin{equation} \Gamma_{\rm orb}<\gamma_{\max}=\frac{1.3\times10^{-4}\,|\dot P_b|}{3P_b}=6.1\times10^{-21}~\mathrm{s}^{-1}. \label{eq:gammamax} \end{equation} Both bodies are neutron stars, of measured masses $1.338$ and $1.249\,\Msun$~\cite{Kramer2021}. We model each as a uniform sphere of $M=1.2\,\Msun$ and $\RNS=14$~km, lighter and larger than the real stars and therefore of lower density, $\bar\rho=2.1\times10^{17}$~kg\,m$^{-3}$. Each star moves about the centre of mass at $300$--$325$~km\,s$^{-1}$, so $v_{\max}=10^3$~km\,s$^{-1}$ also covers a centre-of-mass velocity of several hundred km\,s$^{-1}$ relative to the bath. With $\kstar=10/\RNS$, (W) and (L) are needed only up to $\kstar v_{\max}=714~\mathrm{s}^{-1}$ at kilometre wavelengths. \begin{corollary}[Pulsar bound] \label{cor:pulsar} For the Newtonian uniform-sphere model of the two stars described above, every bath obeying the hypotheses of Theorem~\ref{thm:drag} at $\kstar=10/\RNS$ and $v_{\max}=10^3$~km\,s$^{-1}$ satisfies \begin{equation} \frac{\Teff}{C}>\frac{2\pi\hbar G\,(0.909\,\bar\rho)}{3\kB\gamma_{\max}} =3.3\times10^{16}~\mathrm{K}, \label{eq:pulsar-bound} \end{equation} equivalently $\kB\Teff/C>2.8$~TeV or $\kappa_{\rm FDR}>8.6\times10^{27}~\mathrm{s}^{-1}$. \end{corollary} \noindent Hypothesis (DP) is a statement at $k\leq10/\RNS=7\times10^{-4}$~m$^{-1}$. A laboratory rate does not test it there: for $d\geq2a$ the small-$k$ part of $\EG$ is suppressed by $|\Delta\tilde\rho|^2\propto k^2$, so the rate fixes the kernel only at $k\gtrsim1/a$. The corollary therefore bounds models whose kernel and single temperature are postulated at every $k$, such as the linear-friction dissipative Di\'{o}si--Penrose model of Section~\ref{sec:models}, and not every reciprocal medium; baths that differ between the two scales are the escapes of Section~\ref{sec:gap}. The bound is an order-of-magnitude statement. Strong-field corrections (compactness $\approx0.2$) are of order one; a realistic stellar profile raises $\rho_{\kstar}$ (Remark~\ref{rem:drag}) and strengthens it. With the measured masses and a radius of $12$~km the bound is $5.6\times10^{16}$~K. A drag only adds decay, so a one-sided limit from the measured ratio $\dot P_b^{\rm GW}/\dot P_{b,\rm GR}^{\rm GW}=0.999963(63)$~\cite{Kramer2021} is about twice as strong; we quote the two-sided value. Free fields at $T>0$ are not constrained: their spectral weight sits on $|\omega|=ck$, so $S(k,0)=0$ and they fail (DP). Binary-pulsar timing has been used as a probe of friction from a medium for dark matter~\cite{Pani2015}, and the finite-temperature question for gravitational decoherence was raised for the classical-channel models~\cite{DiBartolomeo2021}. \paragraph{The granular part.} If the bath couples to the full density with Gaussian smearing $R_0$, the incoherent sum over nucleons adds to $\rho_{\kstar}$ a piece $m_n/(8\pi^{3/2}R_0^3)$, with $m_n$ the nucleon mass, with its own whiteness requirement up to $\omega\sim v/R_0$. It does not depend on the stellar structure or on any infrared cut. At $R_0=1$~fm it is $0.18\,\bar\rho$, at $R_0=4.5$~\AA{} it is $0.41$~kg\,m$^{-3}$. It only strengthens the bound, and it is the part that survives an infrared cut (escape E2 of Section~\ref{sec:gap}). \subsection{Which published models the bound reaches} \label{sec:models} \paragraph{Dissipative Di\'{o}si--Penrose.} The Di\'{o}si--Penrose model needs a short-distance cutoff to keep its energy increase finite~\cite{GhirardiGrassiRimini1990}. Bahrami, Smirne and Bassi introduced dissipation to tame that increase, and found that it does not reconcile the model with data unless the cutoff is kept artificially large or the model is applied only to massive systems~\cite{Bahrami2014}; the analogous dissipative and coloured extensions of continuous spontaneous localization are bounded in Refs.~\cite{SmirneBassi2015,TorosGasbarriBassi2017}. The linear-friction model of Ref.~\cite{DiBartolomeo2023} adds friction at a collapse temperature $T_\beta$ to the kernel $D_k=4\pi\hbar G\,e^{-R_0^2k^2}/k^2$. Its centre-of-mass damping $\Gamma=\hbar^2\eta/(2M\kB T_\beta)$, with $\eta$ the $k^2$-weighted Di\'{o}si--Penrose diffusion~\cite{DiBartolomeoCarlesso2024}, is Eq.~\eqref{eq:drag} at $C=1$, $\Teff=T_\beta$. Corollary~\ref{cor:pulsar} gives $T_\beta>3.3\times10^{16}$~K for $R_0\ll\RNS$. We have made this identification only for the linear-friction model; another dissipative construction is bounded if it meets the conditions of Remark~\ref{rem:drag}(ii). Levitated-mechanics data exclude $T_\beta<10^{-13}$~K for $R_0<1~\mu$m and $T_\beta<6\times10^{-12}$~K for $R_0<10^{-8}$~m~\cite{DiBartolomeoCarlesso2024}; the pulsar bound is stronger by $3\times10^{29}$ and $5.5\times10^{27}$, respectively. The second range contains the X-ray floor of Section~\ref{sec:survivor} and the point $R_0=10^{-15}$~m discussed next, so the improvement is more than $5\times10^{27}$. An earlier levitated-magnet test bounds the related collisional model~\cite{Vinante2020}, not $T_\beta$ of this one. The pulsar bound excludes the parameter point $R_0=10^{-15}$~m, $T_\beta\approx3$~K (the temperature of the cosmic microwave background) discussed in Ref.~\cite{DiBartolomeoCarlesso2024} by $10^{16}$. At $T_\beta>3\times10^{16}$~K the dissipative model is plain Di\'{o}si--Penrose for every laboratory observable, so the X-ray bounds of Section~\ref{sec:survivor} apply to it unchanged. \paragraph{Classical-channel and classical--quantum gravity.} The models of Kafri, Taylor and Milburn~\cite{KafriTaylorMilburn2014} and of Tilloy and Di\'{o}si~\cite{TilloyDiosi2016} treat the Newtonian field as classical and sourced by a measurement record; the feedback formulation of Ref.~\cite{GaonaReyes2021} is of the same kind. Without friction the response of the potential to the signal is instantaneous and real, $\chi''=0$, and there is no drag. Dissipative generalizations that thermalize to a finite effective temperature exist~\cite{DiBartolomeo2021}, but Corollary~\ref{cor:pulsar} does not bound them as published. In the dissipative Tilloy--Di\'{o}si model the friction acts on the deviation of each body's momentum from its classical trajectory, so the bath in effect moves with each body and the mean orbit is not damped; this is the co-moving escape at the end of Section~\ref{sec:gap}. The dissipative Kafri--Taylor--Milburn model is formulated for two trapped oscillators and does not describe an orbit. A variant whose friction acted linearly on the full momentum would have Di\'{o}si--Penrose noise at $C=1$, would fall under Theorem~\ref{thm:drag} through the Einstein relation, and would require $\Teff>3.3\times10^{16}$~K; we know of no published model of this kind. The absence of drag also holds for the Newtonian limit of classical--quantum gravity~\cite{Oppenheim2023PRX,Layton2023}, where the potential is classical and diffusive and friction terms are possible but not included; a friction term would set a temperature scale and break Lorentz invariance~\cite{Oppenheim2023}. A frictionless classical--quantum model sits at $\Teff=\infty$: the drag vanishes, and the trade-off of that framework between decoherence and diffusion~\cite{Oppenheim2023} bounds the diffusion of the potential, not a temperature. \subsection{The escapes: no back-action, or a scale-dependent bath} \label{sec:gap} Corollary~\ref{cor:pulsar} leaves open the limit $\Teff\to\infty$ at fixed noise. In that limit $\chi''\to0$ while $S$ stays finite, so the drag vanishes. It is noise without back-action: a classical potential field acting on matter, whose reduced dynamics is the random-unitary Di\'{o}si--Penrose master equation with Hermitian Lindblad operators~\cite{Diosi1989}; in the classical-channel construction it is the floor set by the principle of least decoherence~\cite{TilloyDiosi2017}. For a bath with a finite-dimensional or type-I algebra the limit is its tracial state; a quantum field algebra of type III has no tracial state, and only the limit, or a classical noise field, realizes it. Classical stochastic gravitational fields acting on matter belong to this class~\cite{Asprea2021}. If $\Teff$ were the thermodynamic temperature of ordinary degrees of freedom, a bath at the bound would carry an energy density of order $a_{\rm rad}T^4\sim10^{51}$~J\,m$^{-3}$, with $a_{\rm rad}$ the radiation constant; the survivor is therefore a non-thermodynamic noise field. Its non-reciprocal noise heats each pulsar's centre-of-mass motion at $3\gamma\kB\Teff=2\pi C\hbar G\rho_{\kstar}=8\times10^{-27}$~W at $C=1$, independent of $\Teff$, so the double pulsar does not constrain it. Noise with and without back-action are thus not two kinds of model but the two ends of one family, parametrized by $\Teff$, and the pulsars allow only $\Teff/C>3.3\times10^{16}$~K. This end of the family, with every bath above that temperature, is escape (E1). It is not the only escape. A bath at ordinary temperature can still hide where its kilometre-scale behaviour differs from its laboratory-scale behaviour: \begin{enumerate} \item[(E2)] \emph{an infrared cut:} the kernel departs from $1/k^2$ below some $k_{\rm IR}\gg1/\RNS$. The coherent drag then comes only from the star's density structure above $k_{\rm IR}$. For a sharp-edged uniform sphere that fraction is $0.955/(k_{\rm IR}\RNS)$, $6.8\times10^{-7}$ at $k_{\rm IR}=100$~m$^{-1}$; a smooth envelope lowers it further, and we have not modelled the crust. The granular part remains, because it lives at $k\sim1/R_0$, where the laboratory bounds test the kernel: at the X-ray floor $R_0=4.5$~\AA$\,C^{1/3}$ of Section~\ref{sec:survivor} it gives $\Teff>0.07$~K, independent of $C$ and weaker at larger $R_0$, if the bath is white up to $v/R_0\approx2\times10^{15}$~s$^{-1}$; \item[(E3)] \emph{a slow bath:} a correlation time longer than $\RNS/(10v_{\max})=1.4$~ms at kilometre wavelengths violates (W) and suppresses the drag, while the bath can still be white at laboratory $k$; \item[(E4)] \emph{a scale-dependent temperature:} $\Teff(k)$ finite at laboratory $k$ but above $3.3\times10^{16}\,C$~K at $k\lesssim10/\RNS$; \item[(E5)] \emph{active modes:} a non-KMS state with $\omega\chi''<0$ at $k>\kstar$ violates (P) and can cancel the drag from $k<\kstar$. \end{enumerate} A bath that co-moves with each body (non-linear entrainment, violating (L) and (S) locally) also escapes, at the price of having no rest frame. These escapes are a class, not a corner: any bath whose correlation length or temperature differs between $1~\mu$m and $1$~km passes.