Non-Gaussian diffusive fluctuations in Dirac fluids Authors: Sarang Gopalakrishnan, Ewan McCulloch, Romain Vasseur Published: 2024 ABSTRACT SignificanceDynamical universality, the principle that microscopically diverse systems display the same hydrodynamic behavior and universal scaling laws, is a central concept in nonequilibrium physics. For mean transport, this principle is embodied by dynamical exponents. Whether universal scaling laws also apply to fluctuations remains an important open question. We show that in Dirac fluids—the hydrodynamic description of Dirac semimetals such as graphene—fluctuations are highly non-Gaussian at neutrality, in contrast to “standard metals” where fluctuations follow central limit scaling. We predict an asymptotically exact form for the full counting statistics (FCS) of charge transfer in quasi-1D geometries, generalizing previous results for interacting integrable systems, and propose charge FCS as an unambiguous signature of Dirac fluids in systems with graphene dispersion. Hydrodynamics is an effective theory for the coarse-grained nonequilibrium dynamics of interacting systems (1, 2). Hydrodynamics assumes that all but a few degrees of freedom relax fast. Symmetries and conservation laws determine the remaining slow modes, their equations of motion, and consequently, e.g., the dynamical exponent that governs how perturbations spread. Hydrodynamics has usually been interpreted as a theory of expectation values; however, it can also be extended to a framework (“fluctuating hydrodynamics”) that describes the spatiotemporal fluctuations of conserved densities (3–5). Such density fluctuations are accessible in present-day experiments with cold atoms and superconducting qubit arrays (6–21), and are related to nonlinear response coefficients that can be probed using pump–probe spectroscopy (22–24). A central question in fluctuating hydrodynamics is whether density fluctuations obey universal scaling laws, and if so, what determines their universality classes. In the present work, we explore this question for “Dirac fluids,” i.e., Lorentz-invariant systems with particle–hole symmetry. Lorentz-invariant fluids (25) have recently been considered in many contexts including heavy-ion collisions (26, 27) and, most prominently, the dynamics of electrons in semimetals such as graphene (28–34). In solid-state systems, the crystal lattice breaks Lorentz invariance through umklapp scattering, but this scattering rate is negligible in the cleanest available samples (35–38). Nevertheless, finding unambiguous signatures of hydrodynamic behavior has been a challenge: While heat transport in Dirac fluids is ballistic as the energy current is proportional to momentum density, the charge current can decay because of particle–hole symmetry. The hydrodynamic theory thus predicts that charge transport is simply diffusive, so experimental efforts have had to study heat transport (which strongly violates the Wiedemann–Franz law) and thermoelectric effects (31, 39–42), which are challenging to detect. We find that charge noise in the Dirac fluid is anomalous, both in and out of equilibrium. For concreteness, we characterize the noise through the full counting statistics (FCS) of charge transport in quasi-one-dimensional geometries (43–46), but the mechanism we consider extends to general fluctuation phenomena in finite-temperature states (or nonequilibrium states at finite entropy density). This mechanism is closely related to the one recently identified in integrable systems (47–52) (see also refs. 53, 54, 55, 56). In both the integrable and relativistic cases, diffusion is due to the interaction of slow collective charged modes with ballistically propagating neutral modes (57–60). Because the neutral modes are long-lived, the noise they generate has strong spatiotemporal correlations, so the charge dynamics is that of a few giant diffusing objects rather than many independently moving diffusive particles, and the noise is strongly enhanced (Fig. 1). Out of equilibrium, charge fluctuations across a cut are parametrically enhanced in the Dirac fluid relative to a conventional diffusive system. In equilibrium, their variance is set by the fluctuation–dissipation theorem; however, the higher moments of noise remain anomalous—in the sense that they violate expectations from large deviations and central limit theorem arguments. We show that these features are a general consequence of a simple three-mode fluctuating hydrodynamics that emerges in both the relativistic and integrable contexts. Diffusion from convection. Schematic depiction of the convective mechanism for charge diffusion. Thermal energy fluctuations in the initial state (shades of blue) propagate as sound waves that impart fluctuating velocities to representative charge elements (black). Each charge element undergoes a random walk, but the random walks of different charge elements are correlated: All of them encounter the same sound waves. We will primarily be interested in the FCS of charge in a macroscopic subsystem. The FCS can be defined and measured in many equivalent ways; for conceptual simplicity, we define it through the following protocol. On a given run of the experiment, the subsystem of interest, S, is initialized in a state of definite charge Q0. It is then suddenly connected to the rest of the system and the total charge in S is measured after some time t, giving a random outcome Q from a distribution Pt(Q). We are interested in Pt(ΔQ=Q−Q0). The theory of FCS in generic diffusive systems is “macroscopic fluctuation theory” (61), which posits that this entire distribution is determined by the dependence of the diffusion constant on the conserved densities in the system. In standard metals, this dependence is smooth, giving rise to conventional asymptotically Gaussian FCS. In relativistic hydrodynamics, however, this dependence is singular (33) (as the diffusion constant is finite at charge neutrality, but diverges away from it), so the FCS cannot be conventional. Three-Mode Hydrodynamics In d dimensions, the hydrodynamics we will consider has 2+d conserved densities: charge n, energy e, and energy current ϕi (i runs over the spatial indices, and we implicitly sum over repeated indices). ϕi might be conserved as a consequence of Lorentz invariance or for any other reason. The first hydrodynamic equation is essentially a definition of ϕi being the energy current:[1]∂te+∂iϕi=0. Next, we write a continuity equation for ϕi. In what follows, we will consider n, e, and ϕi to be small deviations from an equilibrium background, and write down linearized hydrodynamic equations, keeping nonlinearities only when important (see below for justification). Since ϕi is already a vector, its current is a second-rank tensor qij, which is even under time-reversal and inversion and neutral under the conserved charge. The tensor eδij satisfies the necessary symmetries to appear in qij, up to an order one constant that we set to unity for simplicity. (In the equation of state for a Dirac fluid, pressure is indeed directly proportional to energy density (33)—a consequence of both Lorentz invariance and the masslessness of the excitations.) Ignoring any other potential conserved quantities we arrive at the equation[2]∂tϕi+∂ie+⋯=0, where ⋯ consists of terms beyond Euler scale that will not be relevant. In the relativistic context, the energy current ϕi is also momentum density (by symmetry of the stress energy tensor Tμν), and the current of energy current is pressure, which is indeed directly proportional to energy density from the equation of state—see e.g. (33, 62–64). Eqs. 1 and 2 imply that both e and ϕi propagate ballistically,[3]∂t2e−∂i∂ie=0, Divergenceless patterns of the velocity field ∂iϕi=0 are frozen in place at the Euler scale, but this will not play any role in the quasi-one-dimensional geometries considered in this letter. It remains to write down a continuity equation for n. At charge neutrality, the equations have a particle–hole symmetry n→−n, which prevents the charge current from overlapping with energy current fluctuations. Within linearized hydrodynamics, charge transport is purely diffusive at this point, with a diffusion constant D and a corresponding (conserving) noise term ∂iξi. Away from charge neutrality—i.e., at net charge n0—charge and energy fluctuations mix; the charge current has a term ∼n0ϕi, and charge transport is ballistic. Even at charge neutrality, the charge has local fluctuations n, which couple to energy current, giving a nonlinear coupling ∼nϕi, again up to an irrelevant proportionality constant that we set to unity. Including both the diffusive and nonlinear terms we arrive at the equation[4]∂tn+∂i(ϕin)=D∂i∂in+∂iξi. Eqs. 1, 2, and 4 define a closed three-mode nonlinear fluctuating hydrodynamic system, which coincides in the relativistic case with the equations derived in ref. 33 in the linear case. As discussed above, the equations for e and ϕi can be combined into a wave equation. Energy and charge fluctuations travel at different speeds and decorrelate. At each time, an energy (charge) fluctuation sees a totally different charge (energy) environment. Since we are interested in charge transport, this decoupling justifies leaving Eqs. 1 and 2 at Euler scale. Nonlinear Terms and Fluctuations Above, we wrote a very simple set of hydrodynamic equations, ignoring various nonlinearities that are known to qualitatively affect low-dimensional hydrodynamics, giving rise to long-time tails and frequency-dependent hydrodynamic coefficients. We now briefly justify this neglect. First, the current of energy current Eq. 2 contains nonlinear terms of the form ϕiϕj by symmetry. Upon including higher-order gradient corrections, the ballistic “sound waves” carried by ϕi will be broadened, either diffusively in dimensions d>2, or with anomalous (KPZ) scaling in d=1 due to these nonlinear contributions (65–70). This broadening is subleading to the Euler-scale propagation of sound waves and can therefore be neglected for our purposes. Second, the charge current Eq. 4 contains a nonlinear coupling to the energy current—whose effects on the FCS we will keep. One might worry however that this nonlinear coupling will make the charge spread superdiffusively: For example, in one-dimensional Galilean fluids the “heat mode” is diffusive in linearized hydrodynamics but spreads as t3/5 once nonlinearities are included (68–70). In the present case, the “diagonal” nonlinearities ε2,ϕ2,n2 that could appear in jn—and which are responsible for KPZ and t3/5 scaling within mode-coupling theory (65, 66, 68–70)—are precluded by the particle–hole symmetry, thus preventing either a KPZ or t3/5 scaling of the charge mode. Finally, shear modes that the charged mode could couple to do not exist in one dimension. Therefore, in one-dimensional fluids with particle–hole symmetry, diffusive hydrodynamics is stable at the level of equilibrium correlators and the leading nontrivial effect is on the FCS, as we now discuss. Charge Diffusion by Convection Eq. 4 describes two qualitatively different diffusion mechanisms. The first is standard diffusion, described by the right hand side. The second is charge diffusion due to energy current fluctuations: Because of hydrodynamic decoupling, ϕi is effectively a random (from thermal fluctuations in the initial state), ballistically correlated velocity field for the charge—a particle evolving with a random velocity undergoes diffusion. This nonlinear diffusion mechanism has recently been explored extensively in the integrability literature (47–52, 57–60, 71, 72), and is the source of the anomalous fluctuations discussed here. Toy Model For simplicity, we will first ignore the “standard” diffusion term, and reinstate it later; and also consider the case of one spatial dimension d=1. Charge transport is then described by the equation[5]∂tn+∂x(ϕn)=0, where ϕ effectively decouples and becomes a source of noise for charge n. It remains however a specific type of noise, as it propagates ballistically ϕ(x,t)=ϕR(x−vt)+ϕL(x+vt), with v the sound velocity, up to [KPZ, in 1d (65–70)] broadening corrections. As we now discuss, charges moving with noisy but ballistically correlated velocities undergo some diffusive motion but with anomalous fluctuations and full counting statistics. To see this, we write the charge density as n(x,t)=∫dx0n0(x0)δ(x−X(x0,t)) so that X(x0,t) is the trajectory of the parcel of charge initially at position x0. Inserting this into Eq. 5 yields the equation of motion for the trajectories, or “characteristics,” X(x0,t),[6]∂tX=ϕ(X(x0,t),t). Due to the relative velocity of the ballistic noise and the charge at each point along the trajectory, the charge spends a time ∝(v+ϕ)−1 ((v−ϕ)−1) at each position of the noise ϕL(x) (ϕR(x)). For simplicity, we will consider the large v limit (v/|ϕ|→∞), where the charge spends an equal time receiving each velocity kick from ϕR/L as they are propagated through. For the generalization to finite (but still large) v, see Materials and Methods. The noisy velocity field ϕ will then lead to diffusive trajectories, X(x0,t)=x0+O(t), so that ϕR,L are translated a distance ±vt+O(t) relative to the charge over a time t. By neglecting the diffusive part of these shifts, the total displacement X is found by summing over every velocity kicks at x0, X(x0,t)≈x0+∫0tdτϕ(x0,τ). The error from discarding O(t) random terms is O(t1/4) and is therefore subleading. By the same reasoning, we can evaluate the noise at the endpoint of a trajectory rather than its initial position x0 up to a O(t1/4) error. This justifies the replacement δ(x−X(x0,t))→δ(x−x0−∫dτϕ(x,τ)) in the definition of the charge density in terms of trajectories X(x0,t). Using this, we find that at long times[7]n(x,t)≈n0x−∫0tdτϕ(x,τ). Let us now use this result to compute the FCS of charge of the right-half x≥0 of the system, Q(t)=∫0∞dxn(x,t). Using the continuity Eq. 5, we have Q˙=ϕ(x=0,t)n(x=0,t). Integrating that equation and using Eq. 7, we find:[8]ΔQ(t)=∫0X(t)dXtn0−Xt=X(t)n[−X(t),0], where X(t)=∫0tdτϕ(0,τ) undergoes a random walk with a diffusion constant set by the variance of ϕ(0,τ). For the second equality, we have defined n[−X(t),0] to be the charge density in the interval between the origin and −X(t) in the initial state. Unlike in ordinary diffusive systems, where charge transfer occurs as a result of many independent random processes, the statistics of the charge transfer in Eq. 8 depends on only a single random walk X(t) with variance 2DXt. Because of this, the statistics is highly anomalous. To see this more clearly, we consider the following two cases: i) a domain wall initial charge configuration, and ii) a thermal equilibrium initial state: i)We take the charge distribution to be a sharp domain wall n0(x)=−n0 if x>0, and n0(x)=+n0 for x<0. Then from Eq. 8, ΔQ(t) is directly proportional to |X(t)|. The distribution Pt(ΔQ) is that of a random walk on the half-line and is therefore highly skewed. While the mean charge transfer scales diffusively as expected ⟨ΔQ(t)⟩∼t, its variance scales as ⟨ΔQ(t)2⟩c∼t, in sharp contrast with the typical behavior expected of generic diffusive systems where all cumulants scale as t (consistent with expectations from central limiting behavior and large deviations).ii)In thermal equilibrium at the charge neutral point, n[0,x] is instead a normally distributed variable with zero mean and variance Tχ(T)/|x|, where χ=dn/dμμ=0 is the charge susceptibility—in what follows, we will set Tχ=1 for simplicity. This gives the equilibrium FCS as a function of time [9]Pt(ΔQ)=∫0∞dXexp−X24DXt−ΔQ22X2π2DXtX. This formula was recently obtained in the context of spin transport in integrable XXZ chains (48) (for which energy current is also conserved as a consequence of integrability), integrable classical spin-chains (52), and of some integrable cellular automata (49, 51). This distribution characterizes charge transfer at the scale ΔQ∼t1/4 and is highly non-Gaussian and nonanalytic at ΔQ=0. i)We take the charge distribution to be a sharp domain wall n0(x)=−n0 if x>0, and n0(x)=+n0 for x<0. Then from Eq. 8, ΔQ(t) is directly proportional to |X(t)|. The distribution Pt(ΔQ) is that of a random walk on the half-line and is therefore highly skewed. While the mean charge transfer scales diffusively as expected ⟨ΔQ(t)⟩∼t, its variance scales as ⟨ΔQ(t)2⟩c∼t, in sharp contrast with the typical behavior expected of generic diffusive systems where all cumulants scale as t (consistent with expectations from central limiting behavior and large deviations). We take the charge distribution to be a sharp domain wall n0(x)=−n0 if x>0, and n0(x)=+n0 for x<0. Then from Eq. 8, ΔQ(t) is directly proportional to |X(t)|. The distribution Pt(ΔQ) is that of a random walk on the half-line and is therefore highly skewed. While the mean charge transfer scales diffusively as expected ⟨ΔQ(t)⟩∼t, its variance scales as ⟨ΔQ(t)2⟩c∼t, in sharp contrast with the typical behavior expected of generic diffusive systems where all cumulants scale as t (consistent with expectations from central limiting behavior and large deviations). ii)In thermal equilibrium at the charge neutral point, n[0,x] is instead a normally distributed variable with zero mean and variance Tχ(T)/|x|, where χ=dn/dμμ=0 is the charge susceptibility—in what follows, we will set Tχ=1 for simplicity. This gives the equilibrium FCS as a function of time [9]Pt(ΔQ)=∫0∞dXexp−X24DXt−ΔQ22X2π2DXtX. This formula was recently obtained in the context of spin transport in integrable XXZ chains (48) (for which energy current is also conserved as a consequence of integrability), integrable classical spin-chains (52), and of some integrable cellular automata (49, 51). This distribution characterizes charge transfer at the scale ΔQ∼t1/4 and is highly non-Gaussian and nonanalytic at ΔQ=0. In thermal equilibrium at the charge neutral point, n[0,x] is instead a normally distributed variable with zero mean and variance Tχ(T)/|x|, where χ=dn/dμμ=0 is the charge susceptibility—in what follows, we will set Tχ=1 for simplicity. This gives the equilibrium FCS as a function of time [9]Pt(ΔQ)=∫0∞dXexp−X24DXt−ΔQ22X2π2DXtX. This formula was recently obtained in the context of spin transport in integrable XXZ chains (48) (for which energy current is also conserved as a consequence of integrability), integrable classical spin-chains (52), and of some integrable cellular automata (49, 51). This distribution characterizes charge transfer at the scale ΔQ∼t1/4 and is highly non-Gaussian and nonanalytic at ΔQ=0. When the system is biased slightly out of equilibrium, with a domain wall of contrast ϵ, the growth of variance crosses over from equilibrium to the nonequilibrium behavior on a timescale t⋆(ϵ)∼1/ϵ4. Restoring Ordinary Diffusion We now consider the effects of ordinary diffusion on the charge transfer by restoring the Fick’s law term in Eq. 4. First, we note that we can neglect the associated noise term ξ in the case of the domain-wall distribution. Over a time t, the convective term moves the domain wall a distance t1/2; meanwhile, the deterministic part of the Fick’s-law diffusion smears the domain wall out over a distance t1/2, and the associated noise contributes t1/4 to the charge transfer SD. This last contribution is subleading for the domain-wall initial condition and we will neglect it. Proceeding without the noise term, 4 can be solved by convolving a solution n~ of the toy model with the diffusion kernel G for ordinary diffusion, n(x,t)=(n~∗G)(x,t). Charge transfer now has two contributions, one from the shared kicks from ϕ, and another from Fick’s law: ΔQ(t)=∫0tdτ∫dxn0(x−X(τ))(ϕ(τ)+D∂x)G(x,τ). For a domain-wall initial, this is once again given in terms of a random walk Xt,[10]ΔQ(t)=n0∫0tdτerfX(τ)2DτdXdτ+2n0D∫0tdτG(X(τ),τ). By sampling over random walks Xt (with DX=1), we numerically evaluate this integral for various diffusivities D and find the probability distribution Pt(ΔQ) (shown in Fig. 2B). The inclusion of ordinary diffusion smooths the nonanalyticity at ΔQ=0, but the distribution remains non-Gaussian and is highly skewed at both small and large diffusivity. The distributions obey a scaling collapse with the rescaling ΔQ/t, and all cumulants scale anomalously as Cn∼tn/2 (Fig. 2C). Anomalous full counting statistics. Charge transfer distributions P(ΔQ(t)) in one dimension: (A) in equilibrium and (B) for a domain wall initial state. Charge transfer is calculated approximately, by evaluating the charge transfer integral (Eq. 10 in the domain wall case) in the absence of the noise ξ (A and B) for various diffusivities D, and directly, by numerically solving Eq. 4 (InsetsA and B). Ordinary diffusion smooths the nonanalyticities at ΔQ=0, but the distributions remain highly non-Gaussian. A scaling collapse of Pt(ΔQ) is observed under the rescaling ΔQ/t1/4 in equilibrium and ΔQ/t1/2 for a domain wall; the scaled distributions are shown for two times t=400 (dashed) and t=800 (solid) in the main panels and for multiple times t=400,600,800 (multiple shades) in the Insets. The cumulants of charge transfer in the domain wall case asymptotically scale as Cn(t)∼tn/2 (shown in panel C for D=0.5), indicating anomalous transport statistics. In equilibrium, the fluctuations associated with ordinary diffusion are a relevant correction to the charge transfer statistics and neglecting them will yield quantitatively inaccurate statistics. Nevertheless, by restoring deterministic diffusion, the cusp at ΔQ=0 is removed while the distribution remains non-Gaussian and is qualitatively similar to the distribution obtained by solving Eq. 4 numerically (Fig. 2 A, Inset). Experimental Relevance So far, we have focused on quasi-one-dimensional geometries, for which it makes sense to define FCS in a segment of the system. These geometries can be much wider than the atomic spacing (e.g., wide enough that it is appropriate to use the bulk dispersion), provided that the segment in which FCS is evaluated is even larger than the transverse width w. In such geometries, at Euler scale, the only truly slow modes are those carrying momentum along the infinitely long direction. By the central limit theorem (CLT), the strength of the fluctuations discussed here scales down as 1/w, so narrower channels are better. To use the bulk dispersion in a graphene nanoribbon, the bandgap opened by the finite width, Egap∼ħvF/w (73), should be much smaller than the thermal fluctuations, Egap≪kBT. We can estimate the effects of having multiple independent charge transfer events (from wider ribbons) by viewing the carriers as particles with diameter equal to the thermal de Broglie wavelength, λth=π2/3ħvF/kBT. Since this defines the same length scale as the density of excited electrons, ρ∼1/λth2, the appropriate length scale with which to compare the ribbon width is λth. This means there is a direct trade-off between the validity of the graphene dispersion, and the CLT suppression of fluctuations, both of which are controlled by the dimensionless ratio w/λth. At charge neutrality, transport in graphene is dominated by phonons above T≈100K (40, 41), and by the presence of charge puddles below T≈50K (29, 31, 74). Therefore, in the operational window for Dirac fluid behavior, λth∼100nm is many times smaller than the typical dimensions for ultraclean devices (31, 75). Quasi-1D geometries can be realized both in synthetic matter—e.g., interacting particles in optical lattices, where honeycomb dispersion have already been synthesized (76–80)—and in solid-state settings using gate-defined channels to confine a 2D electron gas (81–83). In cold-atom systems, imaging with single-site resolution is possible, so the measurement of FCS is feasible and has been demonstrated (18–21). In the solid-state setting, direct FCS measurements are harder, although they have been accomplished using quantum point contacts (84). Moreover, recent experimental advances [e.g., noise spectroscopy using nitrogen-vacancy centers in diamond (85)] in principle allow for more robust measurements of bulk FCS involving larger subsystems. Discussion The key result of this letter is the non-Gaussian nature of diffusive charge fluctuations in relativistic hydrodynamics. Our predictions rely on the following ingredients: 1) the existence of a conserved energy current, 2) particle–hole symmetry, 3) thermal fluctuations in the initial state, and 4) the absence of other conserved quantities that would overlap with the charge current [e.g., a conserved charge that is parity odd and transforms nontrivially under the particle-hole symmetry (PHS), as occurs in the gapless Heisenberg quantum spin chain (86)], so charge transport is diffusive. Those conditions are satisfied in Dirac fluids (in particular, 1) follows from Lorentz invariance), but also in the gapped phase of the Heisenberg spin chain (48) and in some integrable cellular automata (49, 51)—in those integrable examples, the ordinary diffusion contribution is equal to zero. A natural question is whether our results would carry over to energy transport in Galilean invariant systems, where charge current (momentum) is conserved (instead of energy current in Lorentz invariant systems). However, the analog of particle–hole symmetry in that case e→−e only holds at infinite temperature, ruling out continuum models: In general, energy in Galilean-invariant systems propagates ballistically through sound waves. Due to fundamental limitations of simulation in many-body quantum systems, direct numerical checks on quasi-1D Dirac semimetals are not possible. However, our results also apply to classical Galilean invariant fluids with a conserved U(1) charge and PHS, and numerical simulations of a hard-sphere gas (with added charge degrees of freedom) confirm our predictions—charge transfer statistics is anomalous, with cumulants scaling as Cn(t)∼tn/4 in equilibrium, and as Cn(t)∼tn/2 for a domain-wall charge initial (Materials and Methods). In higher dimensions, FCS across macroscopic bipartitions is asymptotically Gaussian because it is a sum over many independent regions. However, the basic physics we have outlined—of ballistic energy fluctuations creating correlated charge noise—can be seen more generally using nonlinear spectroscopy. Nonlinear charge response involves correlation functions like ⟨n2(x,t)n2(0,0)⟩, which spread ballistically because n2 has the same symmetries as e and can therefore mix with it. Thus the nonlinear charge response in the Dirac fluid is radically different from the response that was recently predicted for standard diffusive systems (24). Devising specific near-term experimental probes of this effect is an important task for future work. Finally, whether or not a qualitatively different “strong coupling” regimes exist which share the same hydrodynamics equations that we have studied here, but which have a strong coupling between the ballistic and charge mode, should be investigated further. In particular, whether a mechanism in which charge is “pulled” along with ballistic modes, endowing it with ballistic correlations, is stable, and whether physical examples can be found, are questions we leave to future work. Materials and Methods Method of Characteristics. In this section, we explicitly show that the ansatz n(x,t)=∫dx0n0(x0)δ(x−X(x0,t)), along with Eq. 6 solve Eq. 5 for the charge density in the toy model approximation. Taking a derivative of the ansatz with respect to t yields,[11]∂tn=−∫dx0n0(x0)δ′(x−X(x0,t))∂tX(x0,t), where δ′(x)≡∂xδ(x). Using Eq. 6, this becomes[12]∂tn=−∂x∫dx0n0(x0)δ(x−X(x0,t))ϕ(X(x0,t),t). Finally, using the delta function to replace X(x0,t)→x in ϕ and using the definition of the charge density, we recover Eq. 5 for the charge density, ∂tn+∂xϕn=0. Toy Model with v < ∞. In this section, we consider Eq. 6 away from the v=∞ limit. We will remain in the regime where the noise is fast moving compared to its coupling to the charge mode. Specifically, we will take the noisy velocity field to be bounded with |ϕ|