betaFeedback

WorksPhysics and Astronomy

Molecular hydrodynamic theory of the velocity autocorrelation function

Abstract

The velocity autocorrelation function (VACF) encapsulates extensive information about a fluid's molecular-structural and hydrodynamic properties. We address the following fundamental question: How well can a purely hydrodynamic description recover the molecular features of a fluid as exhibited by the VACF? To this end, we formulate a bona fide hydrodynamic theory of the tagged-particle VACF for simple fluids. Our approach is distinguished from previous efforts in two key ways: collective hydrodynamic modes are modeled by linear hydrodynamic equations; the fluid's static kinetic energy spectrum is identified as a necessary initial condition for the momentum current correlation. Our formulation leads to a natural physical interpretation of the hydrodynamic VACF as a superposition of quasinormal hydrodynamic modes weighted commensurately with the static kinetic energy spectrum, which appears to be essential to bridging continuum hydrodynamical behavior and discrete-particle kinetics. Our methodology yields VACF calculations quantitatively on par with existing approaches for liquid noble gases and alkali metals; moreover, our hydrodynamic model for the self-intermediate scattering function extends the applicable domain to low densities where the Schmidt number is of order unity, enabling calculations for gases and supercritical fluids.

Introduction

The velocity autocorrelation function (VACF) is, perhaps, the quintessential time-correlation function, as it holds unique significance in condensed matter physics and chemistry.[1, 2, 3] In particular, the VACF is intrinsically linked to Brownian motion: the self-diffusion coefficient, DsD_s, of a large diffusing particle is the time integral of the VACF, a connection traceable to the famous Einstein relation, Ds=lim⁡t→∞⟨∣x(t)−x(0)∣2⟩/6tD_s = \lim_{t \to\infty} \langle|\mathbf{x}(t) - \mathbf{x}(0)|^2 \rangle/6t.[4] It was therefore unexpected when Alder and Wainwright,[5, 6] using molecular dynamics (MD) simulations, revealed that the VACF of a “tagged” fluid particle—one mechanically identical to all others—decays as t−3/2t^{-3/2} at long times, not exponentially as predicted by Chapman-Enskog-Boltzmann theory.[7, 8, 9, 10, 11] Using simple physical arguments, they recognized that this protracted decay was due to delayed viscous momentum transport—a phenomenon known as hydrodynamic memory.

Zwanzig and Bixon[12] promptly recognized that viscoelastic hydrodynamics could account for both hydrodynamic memory and molecular-scale granularity in the VACF. Although Zwanzig-Bixon theory remains questionable at short times, its success for liquid argon substantiates the hydrodynamic perspective.[1, 13] Indeed, the fluid-particle VACF has broad utility as it probes molecular structure and dynamics across all timescales, yet efforts to obtain physically consistent analytic models spanning all scales and densities have been hampered by subtle challenges.[14, 15, 16, 17, 18, 19, 20, 21] Meanwhile, MD simulation, along with methods based on the projection operator and memory function formalisms,[22, 23, 24] has been a workhorse for probing molecular scales[25, 26, 27, 28, 29, 30, 31] while underpinning a burgeoning interest in multiscale modeling, including generalized Langevin equations (GLEs) and other coarse-graining (CG) techniques.[32, 33, 34, 35, 36]

Nevertheless, complex fluid systems—nanocolloidal suspensions,[33, 36, 37] active fluids and microswimmers,[38] solvated biomolecules,[39] ionic liquids and electrolyte solutions,[40, 41] and others[42]—exhibit a myriad of mesoscopic phenomena, including long-ranged hydrodynamic interactions and (active) Brownian motion, that computationally challenge particle-based simulations. And yet, it is not clear for even simple fluids[3] what general rules delineate the applicable domain of continuum hydrodynamic models.[30, 43] What ingredients are necessary to bridge continuum and discrete-particle behavior at molecular scales? And down to what scale can a hydrodynamic model yield a viable account of the VACF?

In this Communication, we formulate a theory of the VACF from a purely hydrodynamic standpoint that directly confronts these questions. Central to our framework is a general set of physical desiderata (Table I), which constrains a chosen hydrodynamic model so as to correctly reproduce both short-time molecular kinetics and collective dynamics across all timescales while minimizing ad hoc assumptions. Crucially, we find physical consistency requires that the equal-time velocity covariance be an integrable distribution over wavenumber, which we identify as an initial condition corresponding to the fluid’s static kinetic energy spectrum. These considerations lead to a general theory representing a considerably different physical picture than the otherwise formally similar velocity-field approach.[13, 44, 45, 46, 47, 48]

To this end, we develop a 10-moment molecular-hydrodynamic model based on the well-established 13-moment equations,[49, 50, 51, 52, 53, 54, 55, 56] which originate as velocity moments of the linearized BGK-Boltzmann kinetic equation.[57, 58, 59] The resulting moment-equations naturally yield wavenumber-dependent current correlations that reproduce, among other phenomena, molecular-scale aGeneral requirements that constrain the functional forms of correlation functions.

bSpecific requirements that constrain the values of correlation function parameters.

cRequires D⊥,Ds≠0D_{\perp},D_{s}\ne0 and kk-independence of C⊥C_{\perp} and FsF_{s} for large kk. See Appendix § 13.

dSum rule constraint is valid when Einstein frequency, ωE\omega_{E}, is well defined, e.g., for liquids. See Appendix § 11.

eImplies the invariant forms a2νa^{2}\nu and a2γsa^{2}\gamma_{s}. See Eqs. (32) to (33) and Eqs. (59) and (60).

fLiquids: C∥≈0C_{\parallel}\approx0, Fs≈1F_{s}\approx1 yields analytic solution for D⊥D_{\perp}. Gases: C∥≈0C_{\parallel}\approx0 yields constraint for γs\gamma_{s}, D⊥D_{\perp}. See Appendices § 12 and § 13.

viscoelasticity.[60, 61] By contrast, generalized hydrodynamics[1, 3, 62] begins with the Navier-Stokes equations, extending them to molecular scales by phenomenologically promoting transport coefficients to nonlocal quantities with frequency- and wavenumber-dependence. In practice, the velocity-field method has leveraged generalized hydrodynamics with notable success.[21, 41, 62, 63, 64, 65, 66, 67, 68, 69]

The present theory offers important advantages, however. In particular, it implies a natural physical interpretation of the hydrodynamic VACF: a superposition of quasinormal hydrodynamic modes, each mode kk having a wavelength 2π/k2\pi/k and weight proportional to the equilibrium probability density of fluid kinetic energy. Moreover, unlike other approaches, the methodology enables realistic VACF calculations not only for simple liquids, but also gases using the same underlying framework. We apply the methodology to representative fluids—liquid rubidium and argon, and gaseous and supercritical argon—and find remarkable agreement with MD calculations.[28, 31, 34, 70]

THEORY OF THE HYDRODYNAMIC VACF

Let u(x,t)\mathbf{u}(\mathbf{x},t) be the Eulerian velocity field of a fluid derived as the velocity moment of an exact single-particle distribution function, f(x,v,t)f(\mathbf{x},\mathbf{v},t), and xα(t)\mathbf{x}_{\alpha}(t) be the position of constituent particle α\alpha of mass mm. We then require the fluid velocity at xα(t)\mathbf{x}_{\alpha}(t) to be equal to that particle velocity: u[xα(t),t]=vα(t)\mathbf{u}[\mathbf{x}_{\alpha}(t),t]=\mathbf{v}_{\alpha}(t), where vα(t)≡x˙α(t)\mathbf{v}_{\alpha}(t)\equiv\dot{\mathbf{x}}_{\alpha}(t). Expressing the Lagrangian velocity using Fourier components uk(t)\mathbf{u}_{k}(t),

u[xα(t),t]=∫dk uk(t)eik⋅xα(t)(1)\mathbf{u}[\mathbf{x}_{\alpha}(t),t]=\int d\mathbf{k}\,\mathbf{u}_{k}(t)e^{i\mathbf{k}\cdot\mathbf{x}_{\alpha}(t)} \tag*{(1)}

the velocity covariance Q(t)≡⟨vα(0)⋅vα(t)⟩Q(t)\equiv\langle\mathbf{v}_{\alpha}(0)\cdot\mathbf{v}_{\alpha}(t)\rangle becomes

Q(t)=⟨u[xα(0),0]⋅u[xα(t),t]⟩=∫dkdk′⟨uk(0)⋅uk′(t)eik⋅xα(0)+ik′⋅xα(t)⟩≈∫dk⟨uk∗(0)⋅uk(t)⟨eik⋅[xα(t)−xα(0)]⟩⟩(2)\begin{aligned} Q(t)=\left\langle\mathbf{u}[\mathbf{x}_{\alpha}(0),0]\cdot\mathbf{u}[\mathbf{x}_{\alpha}(t),t]\right\rangle\\ &=\int d\mathbf{k}d\mathbf{k}'\left\langle\mathbf{u}_{k}(0)\cdot\mathbf{u}_{k'}(t)e^{i\mathbf{k}\cdot\mathbf{x}_{\alpha}(0)+i\mathbf{k}'\cdot\mathbf{x}_{\alpha}(t)}\right\rangle\\ &\approx\int d\mathbf{k}\left\langle\mathbf{u}_{k}^{*}(0)\cdot\mathbf{u}_{k}(t)\left\langle e^{i\mathbf{k}\cdot[\mathbf{x}_{\alpha}(t)-\mathbf{x}_{\alpha}(0)]}\right\rangle\right\rangle \tag*{(2)} \end{aligned}

where in the last line we assume a spatially homogeneous system and decompose the ensemble average into a product of averages. Importantly, we assume a priori uncorrelated particle displacements and fluid velocities whose underlying correlations will be reproduced by applying appropriate physical constraints (cf. Table I).

Desideratum (req.)ExpressionRemarks
(1) memory equationaC˙+∫0tdτ K(k,t−τ)C(k,τ)=0\dot{C}+\int_{0}^{t}d\tau\,K(k,t-\tau)C(k,\tau)=0Memory kernel K(k,t)K(k,t) must be consistent with desiderata.
(2) normalizabilityaC(k,0)=1C(k,0)=1Static spectrum must be integrable: ∫0∞dk k2q(ka)\int_{0}^{\infty}dk\,k^{2}q(ka) exists.
(3) time reversibilityaC˙(k,0)=0\dot{C}(k,0)=0Process must be time-reversal symmetric.
(4) exponential rangeaC⊥(k,t)∼e−t/τ⊥, Fs(k,t)∼e−t/τsC_{\perp}(k,t)\sim e^{-t/\tau_{\perp}},\ F_{s}(k,t)\sim e^{-t/\tau_{s}}Must decay exponentially for kv0>τ⊥,s−1kv_{0}>\tau_{\perp,s}^{-1} at lower densities.c
(5) ballistic rangeabψ(t)∼N∫0∞dk k2q(ka)e−k2v02t2/2\psi(t)\sim\mathcal{N}\int_{0}^{\infty}dk\,k^{2}q(ka)e^{-k^{2}v_{0}^{2}t^{2}/2}Short-time sum rule constrains aa using ωE\omega_{E} for liquids.d
(6) diffusive rangeabC⊥(k,t)Fs(k,t)∼e−k2(ν+Ds)tC_{\perp}(k,t)F_{s}(k,t)\sim e^{-k^{2}(\nu+D_{s})t}Must preserve long-time asymptotic form of VACF.e
(7) long-time diffusivitybDs=N∫0∞dt∫0∞dk k2q(ka)C⊥(k,t)Fs(k,t)D_{s}=\mathcal{N}\int_{0}^{\infty}dt\int_{0}^{\infty}dk\,k^{2}q(ka)C_{\perp}(k,t)F_{s}(k,t)Zero-frequency sum rule from Green-Kubo relation.f

Table I. Desiderata and constraints for physically admissible correlation functions.

Assuming an isotropic fluid, we decompose the velocity covariance, Q(k,t)≡⟨uk∗(0)⋅uk(t)⟩=q(k)C(k,t)Q(k,t)\equiv\langle\mathbf{u}_{k}^{*}(0)\cdot\mathbf{u}_{k}(t)\rangle=q(k)C(k,t), where q(k)≡Q(k,0)q(k)\equiv Q(k,0) and C(k,t)≡Q(k,t)/q(k)C(k,t)\equiv Q(k,t)/q(k) is the (normalized) current correlation; Fs(k,t)≡⟨exp⁡{ik⋅[xα(t)−xα(0)]}⟩F_{s}(k,t)\equiv\langle\exp\{i\mathbf{k}\cdot[\mathbf{x}_{\alpha}(t)-\mathbf{x}_{\alpha}(0)]\}\rangle is the self-intermediate scattering function (SISF), the characteristic function for tagged-particle displacements. Altogether, this yields a general expression for the VACF

ψ(t)=N∫0∞dk k2q(k)C(k,t)Fs(k,t)(3)\psi(t)=\mathcal{N}\int_{0}^{\infty}dk\,k^{2}q(k)C(k,t)F_{s}(k,t) \tag*{(3)}

where N=[∫0∞dk k2q(k)]−1\mathcal{N}=\left[\int_{0}^{\infty}dk\,k^{2}q(k)\right]^{-1} is the normalization at t=0t=0.

Equilibrium energy partitioning

From Eq. (3), it is apparent that the customary statement of equipartition, q(k)=kBT/mq(k)=k_{B}T/m, leads to a divergent integral. This is unsurprising, as a true white noise spectrum is unphysical. Thus, the fluid static kinetic energy spectrum, or spectral density for short,

q(k)=⟨∣uk(0)∣2⟩(4)q(k)=\left\langle\left|\mathbf{u}_{k}(0)\right|^{2}\right\rangle \tag*{(4)}

must contain a microscopic parameter, aa, that attenuates large-wavenumber contributions. In turbulence theory, a fundamental quantity analogous to q(k)q(k) is the omnidirectional kinetic energy spectrum, E(k)=2πρk2q(k)E(k)=2\pi\rho k^{2}q(k).[71, 72, 73] The inverse Fourier transform of q(k)q(k) is formally identical to the “form factor,” f(r)f(r), in the velocity-field method, wherein aa is a priori treated as an effective molecular radius, which, in practice, is fixed near the mean intermolecular distance (V/N)1/3=n0−1/3(V/N)^{1/3}=n_{0}^{-1/3}. However, by only requiring consistency with the desiderata in Table I—viz. reqs. (2) and (5)—we deduce that aa more generally represents a kinetic correlation length, a scale above which a continuum description applies. Indeed, we find the shape of q(k)q(k), which must depend on the detailed intermolecular potential, strongly influences VACF oscillations, starkly contrasting with the form factor’s largely peripheral role in ensuring normalizability. Note that Eq. (4) implies q(k)≥0q(k)\geq0, whereas f^(k)\hat{f}(k) is negative for a range of wavenumbers; conversely, q(r)=∫0∞dk k2q(k)sin⁡(kr)/krq(r)=\int_{0}^{\infty}dk\,k^{2}q(k)\sin(kr)/kr can oscillate radially about zero, thereby inducing oscillations in the VACF.

The hydrodynamic VACF

For an isotropic fluid, the VACF decomposes into longitudinal and transverse components with respect to wavevector kk as ψ(t)=13ψ∥(t)+23ψ⊥(t)\psi(t)=\frac{1}{3}\psi_{\parallel}(t)+\frac{2}{3}\psi_{\perp}(t); Eq. (3) becomes

ψ(t)=Npa33∫0∞dk k2q(ka)Fs(k,t)[C∥(k,t)+2C⊥(k,t)](5)\psi(t)=\frac{\mathcal{N}_{p}a^{3}}{3}\int_{0}^{\infty}dk\,k^{2}q(ka)F_{s}(k,t)\left[C_{\parallel}(k,t)+2C_{\perp}(k,t)\right] \tag*{(5)}

where, after changing notation from q(k)q(k) to q(ka)q(ka) and setting N=Npa3\mathcal{N}=\mathcal{N}_{p}a^{3}, we have explicitly introduced aa and a parameter pp that controls the shape of q(ka)q(ka).

Equation (5) and the desiderata in Table I constitute a general formulation of the hydrodynamic VACF. To implement the framework, one specifies the static spectrum, q(ka)q(ka), and correlation functions, C∥,⊥(k,t)C_{\parallel,\perp}(k,t) and Fs(k,t)F_{s}(k,t). In what follows, we develop hydrodynamic models satisfying reqs. (1-7) in Table I to obtain solutions for C∥,⊥(k,t)C_{\parallel,\perp}(k,t) and Fs(k,t)F_{s}(k,t). We then introduce representative models for q(ka)q(ka) that yield simple-fluid VACFs in good agreement with MD calculations.

Hydrodynamic model formulation

To describe collective hydrodynamic modes, we develop a regularized variant of the well-known 10-moment equations, a relaxation system of partial differential equations (PDEs) that we call R10R10.[59] When appropriately constrained by the desiderata, R10 reproduces accurate VACFs for simple liquids, as well as realistic VACFs for intermediate-density fluids and gases. Note that R10 is not necessarily the optimal (or unique) hydrodynamic model consistent with the desiderata; in principle, our framework is compatible with other methodologies, including generalized hydrodynamics[1, 62, 74] and the GENERIC formalism.[50, 51]

R10 may be obtained as a generalization of the BGK-Boltzmann equation using the moment method of hydrodynamics, where we consider two collisional relaxation rates, one each for the longitudinal and transverse components of the deviatoric stress tensor. We derive analytic expressions for longitudinal and transverse current correlations, which capture damped molecular-scale sound and elastic shear waves, respectively. In particular, the corresponding memory equations and kernels [req. (1)] are not assumed, as is commonly done, but rather implied by the R10 PDEs.

To describe the self-motion of tagged particle α\alpha, we propose a hydrodynamic model for the self-density, nsn_{s}, motivated by the R10 equations for collective hydrodynamics variables (nn, j\mathbf{j}, and SS). We then derive analytic expressions for the SISF via Fs(k,t)=⟨ns∗(k,0)ns(k,t)⟩F_{s}(k,t)=\langle n_{s}^{*}(k,0)n_{s}(k,t)\rangle.

As there are no known exact solutions for FsF_{s}, approximations have often leveraged the Gaussian assumption, which is exact in the small- and large-kk limits and, conveniently, directly relates to the mean-square displacement (MSD), e.g., Fs(k,t)=exp⁡[−12k2Δ(t)]F_{s}(k,t)=\exp\left[-\frac{1}{2}k^{2}\Delta(t)\right], where Δ(t)≡13⟨xα(t)−xα(0)⟩2\Delta(t)\equiv\frac{1}{3}\langle\mathbf{x}_{\alpha}(t)-\mathbf{x}_{\alpha}(0)\rangle^{2} is the MSD of tagged particle α\alpha. For instance, a cumulant expansion of Fs(k,t)F_{s}(k,t) with a Gaussian leading term directly reveals non-Gaussian effects, which are known to be relatively small for simple liquids.[1, 75] Unlike common approximations, our model for FsF_{s} satisfies reqs. (1–6), which, along with our model for C⊥C_{\perp}, extends the description to low Schmidt number, Sc=ν/Ds∼O(1)\mathrm{Sc}=\nu/D_{s}\sim O(1), and reproduces the exponential decay range of gases—a feature not captured by simple diffusion (i.e., Fick’s law) and other Gaussian models.

Regularized 10-moment model (R10)

Consider the following linear transport equations for an adiabatic equation of state:

∂tρ+∇⋅j=0(6)\partial_{t}\rho+\nabla\cdot\mathbf{j}=0 \tag*{(6)}
∂tj+v∥2∇ρ+∇⋅S=0(7)\partial_{t}\mathbf{j}+v_{\parallel}^{2}\nabla\rho+\nabla\cdot\mathbf{S}=0 \tag*{(7)}
∂tS+2v02ε˙=∫dx′[D(x−x′)∇′2−Γ(x−x′)]⋅S(x′,t)(8)\partial_{t}\mathbf{S}+2v_{0}^{2}\dot{\boldsymbol{\varepsilon}}=\int d\mathbf{x}'\left[\mathcal{D}(\mathbf{x}-\mathbf{x}')\nabla'^{2}-\boldsymbol{\Gamma}(\mathbf{x}-\mathbf{x}')\right]\cdot\mathbf{S}(\mathbf{x}',t) \tag*{(8)}

where ρ(x,t)\rho(\mathbf{x},t) is the mass density, j(x,t)\mathbf{j}(\mathbf{x},t) is the mass current, S(x,t)\mathbf{S}(\mathbf{x},t) is the deviatoric stress tensor, and ε˙≡12[∇j+(∇j)T−23(∇⋅j)1]\dot{\boldsymbol{\varepsilon}}\equiv\frac{1}{2}\left[\nabla\mathbf{j}+(\nabla\mathbf{j})^{T}-\frac{2}{3}(\nabla\cdot\mathbf{j})\mathbf{1}\right] is the rate-of-strain tensor; v0≡kBT/mv_{0}\equiv\sqrt{k_{B}T/m} and v∥v_{\parallel} are thermal and longitudinal phase velocities, respectively. We define the Fourier transform of the collision frequency tensor Γ(x)\boldsymbol{\Gamma}(\mathbf{x}) as

Γk=k^k^γ+(1−k^k^)ν(9)\boldsymbol{\Gamma}_{k}=\hat{\mathbf{k}}\hat{\mathbf{k}}\gamma+(1-\hat{\mathbf{k}}\hat{\mathbf{k}})\nu \tag*{(9)}

where k^≡k/k\hat{\mathbf{k}}\equiv\mathbf{k}/k, 1\mathbf{1} is the unit tensor, and k^k^\hat{\mathbf{k}}\hat{\mathbf{k}} and 1−k^k^1-\hat{\mathbf{k}}\hat{\mathbf{k}} are longitudinal (normal) and transverse (shear) projection operators; γ\gamma and ν\nu are the respective collision frequencies. Similarly, for the stress diffusion tensor, D(x)\mathcal{D}(\mathbf{x}),

Dk≡k^k^D∥+(1−k^k^)D⊥(10)\mathcal{D}_{k}\equiv\hat{\mathbf{k}}\hat{\mathbf{k}}\mathcal{D}_{\parallel}+(1-\hat{\mathbf{k}}\hat{\mathbf{k}})\mathcal{D}_{\perp} \tag*{(10)}

with regularization diffusion coefficients D∥\mathcal{D}_{\parallel} and D⊥\mathcal{D}_{\perp}. Two points should be highlighted. First, the Fourier-transformed collision term, −Υk⋅Sk-\boldsymbol{\Upsilon}_{k}\cdot\mathbf{S}_{k}, is a relaxation approximation related to the BGK collision operator in the BGK-Boltzmann equation that gives rise to viscoelasticity: e.g., ν−1\nu^{-1} is the transverse component’s Maxwell relaxation time. Second, the Fourier-transformed diffusive regularization term, −k2Dk⋅Sk-k^{2}\mathbf{D}_{k}\cdot\mathbf{S}_{k}, extends the description to higher Knudsen (Kn) number[50, 55]—essential for molecular-scale fluid flows where Kn∼O(1)\mathrm{Kn}\sim O(1) and conventional Navier-Stokes fails.[76, 77] Importantly, regularization implements moment closure when constrained by the Green-Kubo relation for self-diffusion [req. (7)], as well as eliminating spatial structure below physically meaningful scales [req. (4)].

Relaxation limit of the stress components

Working in Fourier space, the transverse projection of Eqs. (7) and (8) yields

dj⊥kdt=−ik⋅S⊥k(11)\frac{d\mathbf{j}_{\perp k}}{dt}=-i\mathbf{k}\cdot\mathbf{S}_{\perp k} \tag*{(11)}
dS⊥kdt+(ν+k2D⊥)S⊥k=−iν02kj⊥k(12)\frac{d\mathbf{S}_{\perp k}}{dt}+(\nu+k^{2}D_{\perp})\mathbf{S}_{\perp k}=-i\nu_{0}^{2}\mathbf{k}\mathbf{j}_{\perp k} \tag*{(12)}

while the longitudinal projection of Eqs. (6) to (8) yields

dρkdt=−ik⋅j∥k(13)\frac{d\rho_{k}}{dt}=-i\mathbf{k}\cdot\mathbf{j}_{\parallel k} \tag*{(13)}
dj∥kdt=−iν∥2kρk−ik⋅S∥k(14)\frac{d\mathbf{j}_{\parallel k}}{dt}=-i\nu_{\parallel}^{2}\mathbf{k}\rho_{k}-i\mathbf{k}\cdot\mathbf{S}_{\parallel k} \tag*{(14)}
dS∥kdt+(γ+k2D∥)S∥k=−iν02(13kj∥k+j∥kk)(15)\frac{d\mathbf{S}_{\parallel k}}{dt}+\left(\gamma+k^{2}D_{\parallel}\right)\mathbf{S}_{\parallel k}=-i\nu_{0}^{2}\left(\frac{1}{3}\mathbf{k}\mathbf{j}_{\parallel k}+\mathbf{j}_{\parallel k}\mathbf{k}\right) \tag*{(15)}

We then make the key assumption that the longitudinal stress relaxation rate, γ+k2D∥\gamma+k^{2}D_{\parallel}, is faster than any other timescale, which circumvents solving a third-order differential equation for C∥C_{\parallel} and proves to be a good approximation; Eq. (15) relaxes to

k⋅S∥k=−iμkkk⋅j∥k(16)\mathbf{k}\cdot\mathbf{S}_{\parallel k}=-i\mu_{k}\mathbf{k}\mathbf{k}\cdot\mathbf{j}_{\parallel k} \tag*{(16)}

where μk≡43ν02/(γ+k2D∥)\mu_{k}\equiv\frac{4}{3}\nu_{0}^{2}/(\gamma+k^{2}D_{\parallel}) is a kk-dependent kinematic bulk viscosity. Here, γ\gamma is determined in the small-kk limit from available data for the bulk viscosity, μ≡ν02/γ\mu\equiv\nu_{0}^{2}/\gamma, and D∥D_{\parallel} is treated as a free parameter; the intuitive choice D∥→4μ/3D_{\parallel}\to4\mu/3 yields good results for present calculations.

It is instructive to note that the isotropic relaxation limit (i.e., γ=ν\gamma=\nu, ν≫∂t\nu\gg\partial_{t}, and ν≫k2D⊥\nu\gg k^{2}D_{\perp}) yields the linearized Navier-Stokes equations with steady-state kinematic viscosity ν=ν02/ν\nu=\nu_{0}^{2}/\nu, where the stress equation relaxes to Newton’s law of viscosity, S=−2νε˙\mathbf{S}=-2\nu\dot{\boldsymbol{\varepsilon}}.

Using Eqs. (11) and (12) for the transverse component, and the relaxation approximation for the longitudinal component, Eqs. (13), (14) and (16), we derive ordinary differential equations (ODEs) in time for the modal correlation functions using jk(t)=ρ0uk(t)\mathbf{j}_{k}(t)=\rho_{0}\mathbf{u}_{k}(t), ρ0\rho_{0} is the equilibrium mass density. For transverse current correlations, C⊥C_{\perp}, we combine Eqs. (11) and (12) to obtain a second-order ODE for the transverse current

d2j⊥kdt2+(ν+k2D⊥)dj⊥kdt+k2ν02j⊥k=0(17)\frac{d^{2}\mathbf{j}_{\perp k}}{dt^{2}}+(\nu+k^{2}D_{\perp})\frac{d\mathbf{j}_{\perp k}}{dt}+k^{2}\nu_{0}^{2}\mathbf{j}_{\perp k}=0 \tag*{(17)}

which leads to

C¨⊥(k,t)+νkC˙⊥(k,t)+k2ν02C⊥(k,t)=0(18)\ddot{C}_{\perp}(k,t)+\nu_{k}\dot{C}_{\perp}(k,t)+k^{2}\nu_{0}^{2}C_{\perp}(k,t)=0 \tag*{(18)}

where νk≡ν+k2D⊥\nu_{k}\equiv\nu+k^{2}D_{\perp}. Equation (18) has a memory equation form, eq. (1), with kernel K⊥(k,t)=k2ν02exp⁡(−νkt)K_{\perp}(k,t)=k^{2}\nu_{0}^{2}\exp(-\nu_{k}t). Similarly, substitution of the relaxation limit, Eq. (16), into Eq. (14) eventually yields a second-order ODE for the longitudinal current correlations

C¨∥(k,t)+γkC˙∥(k,t)+k2ν∥2C∥(k,t)=0(19)\ddot{C}_{\parallel}(k,t)+\gamma_{k}\dot{C}_{\parallel}(k,t)+k^{2}\nu_{\parallel}^{2}C_{\parallel}(k,t)=0 \tag*{(19)}

where γk≡k2μk\gamma_{k}\equiv k^{2}\mu_{k}. With the initial conditions C∥,⊥(k,0)=1C_{\parallel,\perp}(k,0)=1 and C˙∥,⊥(k,0)=0\dot{C}_{\parallel,\perp}(k,0)=0 [eqs. (2)–(3)], solutions to Eqs. (18) and (19) are readily found:

C⊥(k,t)=12e−12νkt[(1−νkαk)e−12αkt+(1+νkαk)e12αkt](20)C_{\perp}(k,t)=\frac{1}{2}e^{-\frac{1}{2}\nu_{k}t}\left[\left(1-\frac{\nu_{k}}{\alpha_{k}}\right)e^{-\frac{1}{2}\alpha_{k}t}+\left(1+\frac{\nu_{k}}{\alpha_{k}}\right)e^{\frac{1}{2}\alpha_{k}t}\right] \tag*{(20)}
C∥(k,t)=12e−12γkt[(1−γkβk)e−12βkt+(1+γkβk)e12βkt](21)C_{\parallel}(k,t)=\frac{1}{2}e^{-\frac{1}{2}\gamma_{k}t}\left[\left(1-\frac{\gamma_{k}}{\beta_{k}}\right)e^{-\frac{1}{2}\beta_{k}t}+\left(1+\frac{\gamma_{k}}{\beta_{k}}\right)e^{\frac{1}{2}\beta_{k}t}\right] \tag*{(21)}

where αk=νk2−4k2ν02\alpha_{k}=\sqrt{\nu_{k}^{2}-4k^{2}\nu_{0}^{2}} and βk=γk2−4k2ν∥2\beta_{k}=\sqrt{\gamma_{k}^{2}-4k^{2}\nu_{\parallel}^{2}}. Equations (20) and (21) have the flexibility to satisfy all requirements in Table I; the corresponding dynamic spectral densities are provided in Appendix § 10.

Self-intermediate scattering function

The regularized relaxation models represented by Eqs. (11) and (12) and Eqs. (13) to (15) for the collective hydrodynamic variables suggest the following model

dnsdt=−ik⋅js(22)\frac{dn_{s}}{dt}=-i\mathbf{k}\cdot\mathbf{j}_{s} \tag*{(22)}
djsdt+(γs+k2Ds)js=−ikν02ns(23)\frac{d\mathbf{j}_{s}}{dt}+(\gamma_{s}+k^{2}D_{s})\mathbf{j}_{s}=-i k\nu_{0}^{2}n_{s} \tag*{(23)}

where ns(k,t)n_{s}(k,t) is the (self-)density of a non-interacting collection of tagged particles (i.e., test particles[75]), js(k,t)\mathbf{j}_{s}(k,t) the self-current, γs\gamma_{s} the Brownian collision frequency corresponding to (Stokes) friction, and DsD_{s} the self-diffusion coefficient; here, we take Ds→ν02/γsD_{s}\to\nu_{0}^{2}/\gamma_{s}, which enforces eq. (4) and ensures the positivity of FsF_{s} for all kk and tt. Equations (22) and (23) are a natural generalization of Fick’s Law of self-diffusion,[44, 45] which is recovered in the full relaxation limit (γs→∞\gamma_{s}\to\infty, k→0k\to0) where Eqs. (22)… and (23) reduce to a diffusion equation for ns(k,t)n_s(k,t). The full ODE corresponding to Eqs. (22) and (23) is

F¨s(k,t)+(γs+k2Ds)F˙s(k,t)+k2v02Fs(k,t)=0(24)\ddot{F}_s(k,t)+(\gamma_s+k^2D_s)\dot{F}_s(k,t)+k^2v_0^2F_s(k,t)=0 \tag*{(24)}

the roots of which factor nicely to give

Fs(k,t)=γse−k2Dst−k2Dse−γstγs−k2Ds(25)F_s(k,t)=\frac{\gamma_s e^{-k^2D_st}-k^2D_se^{-\gamma_st}}{\gamma_s-k^2D_s} \tag*{(25)}

Note that density fluctuations decay exponentially for large kk. To see that this model is reasonable, consider Eq. (24) in memory equation form

F˙s(k,t)+k2v02∫0tdt′e−γsk(t−t′)Fs(k,t′)=0(26)\dot{F}_s(k,t)+k^2v_0^2\int_0^t dt' e^{-\gamma_{sk}(t-t')}F_s(k,t')=0 \tag*{(26)}

where γsk=γs+k2Ds\gamma_{sk}=\gamma_s+k^2D_s, γs>0\gamma_s>0. The Markovian solution, obtained by freezing Fs(k,t′)F_s(k,t') at the upper limit tt, is

Fs(k,t)=exp⁡[−k2(v0γsk)2(γskt+e−γskt−1)](27)F_s(k,t)=\exp\left[-k^2\left(\frac{v_0}{\gamma_{sk}}\right)^2\left(\gamma_{sk}t+e^{-\gamma_{sk}t}-1\right)\right] \tag*{(27)}

which, when Ds=0D_s=0, yields the formal result for conventional Langevin dynamics.2 Also note that when Ds→0D_s\to0 in Eq. (25), Fs(k,t)→1F_s(k,t)\to1 and Eq. (27) would not satisfy req. (4). It is unsurprising that Ds>0D_s>0 is necessary for a physically meaningful SISF.

Implementation of the framework

The VACF is calculated by evaluating Eq. (5) using the transverse and longitudinal current correlations, Eq. (20) and Eq. (21), and SISF, Eq. (25), along with the static spectrum, q(ka)q(ka), discussed subsequently; importantly, all VACF calculations use these equations (and parameters). However, determining gas parameters (Appendix § 13) is considerably more complicated since the short-time sum rule [req. (5)] only applies to dense fluids, Fs≈1F_s\approx1 cannot be assumed for req. (7), and empirical data for transport parameters is limited. A suitable kinetic theory, such as Enskog theory (used here) or modifications thereof,[78, 79, 80] can be used with our framework to derive sum rules connecting molecular parameters to macroscopic quantities,[10, 81] as well as estimating transport coefficients in lieu of empirical inputs.

Static spectrum

Pending a first-principles derivation of q(ka)q(ka), we consider representative two-parameter models: a symmetric generalized Gaussian

qG(ka)=NGv02e−12∣ka∣p(28)q_G(ka)=\mathcal{N}^Gv_0^2e^{-\frac{1}{2}|ka|^p} \tag*{(28)}

with p≥1p\ge1, and generalized Lorentzian

qL(ka)=NLv021+∣ka∣p(29)q_L(ka)=\mathcal{N}^L\frac{v_0^2}{1+|ka|^p} \tag*{(29)}
Param.Expr.InputValue
ν\nuη/ρ0\eta/\rho_0TT319.0 K [86]87.0 K [47]
μ\muηb/ρ0\eta_b/\rho_0ρ0\rho_01500.0 kg/m3^3 [86]1430.0 kg/m3^3 [47]
ν0\nu_0v02/νv_0^2/\nuη\eta644.0 μ\muPa⋅\cdots [87]290.0 μ\muPa⋅\cdots [88]
γ0\gamma_0v02/μv_0^2/\muηb\eta_b2020 μ\muPa⋅\cdots [89]90.0 μ\muPa⋅\cdots [90]
γs0\gamma_{s0}v02/Dsv_0^2/D_sDsD_s2.85⋅10−92.85\cdot10^{-9} m2^2/s [91]1.75⋅10−91.75\cdot10^{-9} m2^2/s [92]
aareq. (5)ωE\omega_E6.1 ps−1^{-1} [46]7.7 ps−1^{-1} [28]
ν,γ,γs\nu,\gamma,\gamma_sreq. (6)v∥2v_\parallel^21.8v021.8v_0^2 a2.4v022.4v_0^2 a
D⊥D_\perpreq. (7)pp1414 a22 a
D∥D_\parallel4μ/34\mu/3q(ka)q(ka)qL(ka)q_L(ka) bqG(ka)q_G(ka) b

Table 2. Table II. Key parameters. Left two columns: relations and constraints for molecular parameters. Right three columns: input data for liquid rubidium/argon calculations in Figure 1 and Figure 2.

Figure 2. Liquid argon VACF near the triple point (T≈87T \approx87 K, ρ0≈1.43⋅103\rho_0 \approx1.43 \cdot10^3 kg/m3^3) with longitudinal (orange) and transverse (blue) components, using a Gaussian static spectrum (p=2p = 2).

aFree model parameter

See Eqs. (28) and (29).

with p≥6p\ge6, where the normalizations depend on shape parameter pp via NG=NpGa3\mathcal{N}^G=\mathcal{N}_p^Ga^3 and NL=NpLa3\mathcal{N}^L=\mathcal{N}_p^La^3. Equations (28) and (29) approach the equipartition spectrum for small aa; softer intermolecular potentials correspond to sharper spectral cutoffs (larger pp), which amplify oscillations in the VACF (cf. Figure 1).[82, 83, 84, 85] Note that normalization [req. (2)] yields NpG=p/[81/pΓ(3/p)]\mathcal{N}_p^G=p/[8^{1/p}\Gamma(3/p)], whereas solutions for NpL\mathcal{N}_p^L are unavailable for general pp, so Eq. (28) is preferred for analytic manipulation.

Figure 1. Liquid rubidium VACF (T≈319T \approx319 K, ρ0≈1.51⋅103\rho_0 \approx1.51 \cdot10^3 kg/m3^3) with longitudinal (orange) and transverse (blue) components, using a generalized Lorentzian static spectrum (p=14p = 14).

Parameter determination for liquids

The short- [req. (5)] and long-time [req. (6)] behavior of the VACF uniquely determines a ν\nu and relaxation frequencies ν\nu and γs\gamma_s. To see this, consider the long-time behavior by evaluating Eq. (5) with the asymptotic forms C⊥(k,t)∼exp⁡[−k2(v02/ν)t]C_\perp(k,t)\sim\exp[-k^2(v_0^2/\nu)t] and Fs(k,t)∼exp⁡[−k2(v02/γs)t]F_s(k,t)\sim\exp[-k^2(v_0^2/\gamma_s)t]

ψ(t)∼23Npa3∫0∞dk k2q(ka)e−k2(v02/γs)te−k2(v02/ν)t(30)\psi(t)\sim\frac{2}{3}\mathcal{N}_pa^3\int_0^\infty dk\,k^2q(ka)e^{-k^2(v_0^2/\gamma_s)t}e^{-k^2(v_0^2/\nu)t} \tag*{(30)}
∼πNp6[v02a2(1ν+1γs)t]−3/2(31)\sim\frac{\sqrt{\pi}\mathcal{N}_p}{6}\left[\frac{v_0^2}{a^2}\left(\frac{1}{\nu}+\frac{1}{\gamma_s}\right)t\right]^{-3/2} \tag*{(31)}

along with the theoretical prediction from mode-coupling[44, 45, 93] and kinetic theory[94, 95]

ψ(t)∼23n0a3[14π(ν+Ds)t]3/2(32)\psi(t)\sim\frac{2}{3n_0a^3}\left[\frac{1}{4\pi(\nu+D_s)t}\right]^{3/2} \tag*{(32)}

Naively, one can match Eq. (32) by assuming ν→v02/ν\nu\to v_0^2/\nu, γs→v02/Ds\gamma_s\to v_0^2/D_s, and πNpa3/6→(2/3n0)(4π)−3/2\sqrt{\pi}\mathcal{N}_pa^3/6\to(2/3n_0)(4\pi)^{-3/2}. This satisfies req. (6) and implies a→a0≡(2π2Npn0)−1/3a\to a_0\equiv(2\pi^2\mathcal{N}_pn_0)^{-1/3}. But the short-time sum rule [req. (5)] generally yields a≤a0a\le a_0 (cf. Appendix § 11), implying that the molecular-scale parameters ν≠v02/ν\nu\ne v_0^2/\nu and γs≠v02/Ds\gamma_s\ne v_0^2/D_s. Assigning consistent values for aa, ν\nu, and γs\gamma_s therefore requires a more careful procedure to concomitantly satisfy reqs. (5) and (6). Importantly, treating the products a2νa^2\nu and a2γsa^2\gamma_s in Eq. (31) as invariant quantities preserves the asymptotic form.

These observations suggest the following procedure. First, define the following “base” values: ν0≡vˉ02/ν\nu_0 \equiv\bar{v}_0^2/\nu and γs0≡vˉ02/Ds\gamma_{s0} \equiv\bar{v}_0^2/D_s (cf. Table II). Second, determine aa through the short-time sum rule. E.g., for a generalized Gaussian spectrum, one obtains (cf. Appendix § 11)

a2=41/pΓ(5/p)Γ(3/p)×13ωE2(vˉ∥2+2vˉ02)(33)a^2 = \frac{4^{1/p}\Gamma(5/p)}{\Gamma(3/p)} \times\frac{1}{3\omega_{\mathrm{E}}^2}\left(\bar{v}_{\parallel}^2 + 2\bar{v}_{0}^{2}\right) \tag*{(33)}

Finally, take ν→(a0/a)2ν0\nu\to(a_0/a)^2\nu_0 and γs→(a0/a)2γs0\gamma_s \to(a_0/a)^2\gamma_{s0}, which preserves Eq. (32). Consistency suggests the longitudinal collision frequency be rescaled accordingly: γ→(a0/a)2γ0\gamma\to(a_0/a)^2\gamma_0, where γ0≡vˉ02/μ\gamma_0 \equiv\bar{v}_0^2/\mu.

Parameter determination at lower densities

Given the paucity of numerical and experimental data for dilute monatomic fluids, we determine inputs using Enskog theory supplemented by heat capacity data from the National Institute of Standards and Technology (NIST) Chemistry WebBook.[88] For brevity, procedural details are provided in Appendix § 13.

Numerical Results

VACF calculations for liquid rubidium and argon use values listed in Table II. Calculations for argon-like gaseous and supercritical fluids use expressions for the Enskog viscosity and diffusion coefficients combined with reqs. (4) and (7) to derive reasonable values for the key parameters D⊥D_{\perp}, ν\nu, γs\gamma_s, and aa (cf. Appendix § 13). All calculations in presented in the article and its appendices can be performed with the Mathematica notebook provided in the supplementary material.

Figure 1 shows the VACF for liquid rubidium using a generalized Lorentzian spectrum with p=14p = 14, which should be compared to the velocity-field results (Ref. 46, Fig. 1; Ref. 70, Fig. 2). The oscillatory VACF behavior of liquid alkali metals observed in MD simulations[46, 47, 70, 96] requires a relatively sharp cutoff (large pp), which corresponds to the relatively soft repulsive core of the intermolecular potential as compared to liquid argon.[21, 82, 83]

Figure 2 shows the VACF for liquid argon near the triple point using a Gaussian (p=2p = 2) spectrum (cf. velocity-field result: Ref. 47, Fig. 2). We remark that the plateau often seen in liquid argon(-like) VACFs from MD[25, 26, 27, 28, 92, 97, 98, 99] can be reasonably reproduced in our formulation using, e.g., a generalized Lorentzian spectrum with p≈7.5p \approx7.5, vˉ∥2≈3.0vˉ02\bar{v}_{\parallel}^{2} \approx3.0\bar{v}_{0}^{2}, and adjusting DsD_s upward by ∼10%\sim10\% so as to increase oscillations primarily in the longitudinal (but not transverse) component.

Figure 3 shows VACFs for argon-like gaseous and supercritical fluids. The exponential and diffusive (t−3/2t^{-3/2}) decay ranges are a direct consequence of the explicit handoff in Eq. (25) and Eq. (56). These prominent features were observed in MD calculations for hard-spheres[31] and a supercritical Lennard-Jones (LJ) fluid,[34] as well as immersed-particle fluctuating hydrodynamics simulations at low ScSc[13, 100, 101] and analytic calculations for Basset-Boussinesq-Oseen (BBO) dynamics[102, 103, 104, 105] with general slip boundary conditions.[106] In particular, the dips at early times (Figure 3, t≲10t \lesssim10 ps) come from the longitudinal current, which appears to capture the effect of strongly damped sound waves. Similar features are clearly evident in MD calculations: see Ref. 31, Fig. S1 and also Appendix § 14 for an indirect comparison with Ref. 34, Fig. 1, which shows excellent quantitative agreement. Analytic calculations for “BBO particles”[107] also exhibit qualitative similarities (Ref. 106, Figs. 2 and 3). finite relaxation time of molecular-scale stresses. Regularization, however, is essential—especially at low densities. Importantly, Eq. (25) shows that exponential decay occurs when k2Ds>γsk^{2}D_{\mathrm{s}}>\gamma_{\mathrm{s}}, which implies kλE>1k\lambda_{\mathrm{E}}>1, where λE∼v0/γs\lambda_{\mathrm{E}}\sim v_{0}/\gamma_{\mathrm{s}} is the Enskog mean free path. More precisely, when a<k−1<λEa<k^{-1}<\lambda_{\mathrm{E}}, the expected exponential decay range of a gas emerges from the contributions of large-kk modes. Also, note that the slowest relaxation rate of C⊥(k,t)C_{\perp}(k,t) is given by the smaller of the two exponents in Eq. (20), which, when expanded for large kk, gives −v02/D⊥-v_{0}^{2}/D_{\perp} when D⊥>0D_{\perp}>0 (cf. Appendix § 13). Thus, the regularization coefficients D⊥D_{\perp} and DsD_{\mathrm{s}}, the latter of which arises from our hydrodynamic model for the self-density, are necessary for obtaining the expected exponential decay of the dilute gas VACF.

Hard-sphere VACFs emulating gaseous and supercritical argon

Figure 3. Hard-sphere VACFs emulating gaseous (left) and supercritical (right) argon ranging from low (red) to high (blue) densities. Pure exponential (dotted) and t−3/2t^{-3/2} (dashed) decay are shown as guides to the eye. Note that at 300 K, 41.2 kg/m3^{3} (right, red) argon is gaseous (and not supercritical).

The exponential decay of the SISF (and VACF) at low densities also hints at a deeper connection to VACFs for large Brownian particles, which also exhibit exponential decay.[114] It is worth exploring the connection between our equations of motion for the self-density and GLE models of (single-)particle dynamics—particularly the fluctuating BBO equation, which describes hydrodynamic Brownian motion.[115, 116, 117] Recent developments in memory kernel reconstruction methods may offer MD-driven insights into these connections, and it would be fruitful to leverage these tools to study Brownian motion in colloidal solutions and active matter.[107, 117, 118, 119, 120, 121, 122, 123, 124, 125]

We remark that the non-Gaussian behavior of the VACF at intermediate times, which switches from damped oscillations to pure exponential decay, coincides with the emergent exponential ranges in FsF_{\mathrm{s}} and C⊥C_{\perp}. It is therefore possible that the hydrodynamic VACF formulation can lend dynamical insight into the liquid-vapor phase transition not otherwise available through other methods. Realizing a fully capable VACF theory would, however, require a means (i.e., a sum rule[10, 81]) to determine α\alpha as a continuous function of density (and temperature) through the phase transition without relying solely on the short-time sum rule involving ωE\omega_{\mathrm{E}} [eq. (5)], which is ill-defined at lower densities. Our methodology is nevertheless amenable to the use of different (kinetic) models used to determine model parameters (e.g., our use of Enskog theory for gaseous and supercritical argon calculations), which may be useful for probing fundamental questions pertaining to the behavior of supercritical[126, 127, 128] and supercooled fluids.[20, 129, 130, 131, 132, 133, 134, 135, 136]

Extensions to the present work include deriving q(ka)q(ka) from first principles, as well as (numerically) solving the full zero-frequency constraint for D⊥D_{\perp} [eq. (7)] at all densities and third-order ODE for C∥(k,t)C_{\parallel}(k,t) (i.e., beyond the relaxation limit). However, we expect the present theory to be valuable well beyond VACF calculations. For example, Alder and Wainwright[6] and, more recently, Han et al. 30 and Lesnicki and Vuilleumier,[29] have convincingly demonstrated that collective motions in discrete-particle fluids are well represented by hydrodynamic flow fields down to the single-particle scale. This suggests that stochastically driven R10 equations,[77] which gen-

Discussion

The present theory reflects an eclectic synthesis of ideas dispersed throughout the literature along with several new concepts. We have organized the theoretical formulation and framework in this Communication so as to highlight important results. In summary, we:

  1. Presented a new derivation and interpretation of the hydrodynamic VACF formulation, Eqs. (2) to (5).

  2. Established a core set of physical desiderata whose constraints are sufficient to recover realistic VACFs.

  3. Identified q(ka)q(ka) as the initial condition of the velocity covariance function that characterizes molecular-scale kinetic fluctuations and reproduces the VACF over all timescales by judiciously superimposing each hydrodynamic mode.

  4. Proposed linear PDEs (R10) to model hydrodynamic collective modes, where regularization and the zero-frequency sum rule effect moment closure; yields analytic solutions for C⊥C_{\perp} and C∥C_{\parallel} with sufficiently rich structure to resolve subtle details in the VACF.

  5. Proposed a new hydrodynamic model for the self-density, leading to a viable analytic form of the self-intermediate scattering function, Fs(k,t)F_{\mathrm{s}}(k,t), for all densities; captures exponential decay at low densities.

  6. Described the (re-)scaling of hydrodynamic model parameters that recover the short-time VACF behavior while preserving the correct long-time t−3/2t^{-3/2} decay.

It is worth mentioning that one can derive telegrapher’s equations from R10 (for D⊥D_{\perp} or Ds=0D_{\mathrm{s}}=0). Trachenko [108] derived a telegrapher’s equation as a continuum liquid dynamics model. We remark that telegrapher’s equations describe persistent random walks by accounting for directional correlations in Brownian motion.[109, 110, 111] As pointed out by Khrapak,[112] Zwanzig’s speculative model of molecular self-motion in a liquid[113]—where collective rearrangements correspond to configurational transitions between metastable equilibria—is consistent with the persistent random walk picture. From our hydrodynamic standpoint, this persistence originates from the finite relaxation time of molecular-scale stresses. eralize the Landau-Lifschitz Navier-Stokes equations,[137] would represent a viable molecular-hydrodynamic model capable of reproducing molecular-scale flows.

Indeed, the efficacy of our VACF formulation rests largely on q(ka)q(ka). Even apparently subtle differences in the shape of q(ka)q(ka) substantially alter the character of the VACF, e.g., the oscillatory nature for liquid rubidium or “plateau” for liquid argon (data not shown). The details of the VACF depend on the manner in which the fluid’s distribution of kinetic energy transitions from (approximate) equipartition at long wavelengths to zero below the molecular scale. The kinetic energy distribution must, in turn, depend on the intermolecular potential, which also determines the radial distribution function, g(r)g(r), or static structure factor, S(k)S(k). However, whereas g(r)g(r) or S(k)S(k) characterizes static density correlations (zeroth velocity-moment), q(k)q(k) characterizes static momentum correlations (first velocity-moment). Given its pivotal role in capturing molecular-scale behavior, it is thus our belief that the static spectrum is key to describing how continuum hydrodynamic modes emerge from the molecular scale and bridging the continuum and discrete-particle perspectives.

Acknowledgments

The authors are grateful for valuable discussions with Mark A. Hayes, Dmitry Matyushov, Jason Hamilton, Ralph V. Chamberlin, Paul Campitelli, and Kyle L. Seyler. SLS would like to warmly acknowledge Oliver Beckstein and the Blue Waters Graduate Fellowship program—a part of the Blue Waters sustained-petascale computing project supported by the National Science Foundation (awards OCI-0725070 and ACI-1238993) and the state of Illinois—whose generous support helped nucleate this research; Blue Waters is a joint effort of the University of Illinois at Urbana-Champaign and its National Center for Supercomputing Applications. CES was supported by the National Nuclear Security Administration Stewardship Sciences Academic Programs under Department of Energy Cooperative Agreement No. DE-NA0003764.

Author Declarations

Conflict of Interest

The authors have no conflicts to disclose.

Author Contributions

All authors contributed equally to this work.

Data Availability

The data that support the findings of this study are available within the article and its supplementary material.

Dynamic spectral densities

The dynamic structure factor J(k,ω)J(k,\omega) is the cosine transform of the current correlation function.

J(k,ω)=2∫0∞dt C(k,t)cos⁡(ωt)(34)J(k,\omega)=2\int_{0}^{\infty}dt\,C(k,t)\cos(\omega t) \tag*{(34)}

Recalling that νk=ν+k2D⊥\nu_k=\nu+k^2D_{\perp} and γk=γ+k2D∥\gamma_k=\gamma+k^2D_{\parallel}, we find for the transverse and longitudinal correlations respectively

J⊥(k,ω)=2k2v02νk(ω2−k2v02)2+ω2νk2(35)J_{\perp}(k,\omega)=\frac{2k^2v_0^2\nu_k}{(\omega^2-k^2v_0^2)^2+\omega^2\nu_k^2} \tag*{(35)}
J∥(k,ω)=2k2v∥2γk(ω2−k2v∥2)2+ω2k4D∥2(36)J_{\parallel}(k,\omega)=\frac{2k^2v_{\parallel}^2\gamma_k}{(\omega^2-k^2v_{\parallel}^2)^2+\omega^2k^4D_{\parallel}^2} \tag*{(36)}

Note the following property: as νk,γk→0\nu_k,\gamma_k\to0 and D⊥,D∥→0D_{\perp},D_{\parallel}\to0, the spectral densities approach delta functions in the arguments ω=±kv0\omega=\pm kv_0 and ω=±kv∥\omega=\pm kv_{\parallel}. Compare these results to purely viscous decay that one would obtain from the Navier-Stokes equation

JNS(k,ω)=2k2νω2+k4ν2(37)J_{NS}(k,\omega)=\frac{2k^2\nu}{\omega^2+k^4\nu^2} \tag*{(37)}

Determination of scale aa for liquids [req. (5)]

In sufficiently dense fluids such as a liquid, the Einstein frequency, ωE\omega_E, has physical relevance and aa can be computed from the second-moment condition by considering the short-time expansion of the VACF

ψ(t)=1−12(tτc)2+⋯=1−12ωE2t2+⋯(38)\psi(t)=1-\frac{1}{2}\left(\frac{t}{\tau_c}\right)^2+\cdots=1-\frac{1}{2}\omega_E^2t^2+\cdots \tag*{(38)}

where τc=ωE−1\tau_c=\omega_E^{-1} is the timescale characterizing the initial (de)correlation of the VACF and the Einstein frequency, ωE\omega_E, is formally defined via

ωE2≡4πn03m∫0∞dr r2g(r)(d2ϕdr2+2rdϕdr)(39)\omega_E^2\equiv\frac{4\pi n_0}{3m}\int_0^\infty dr\,r^2g(r)\left(\frac{d^2\phi}{dr^2}+\frac{2}{r}\frac{d\phi}{dr}\right) \tag*{(39)}

where g(r)g(r) is the radial distribution function and ϕ(r)\phi(r) is the intermolecular potential. For large Sc—assumed to be the case for liquids—the short-time expansion of the modal correlation function dominates that from the self-intermediate scattering function. That is, when ν≫Ds\nu\gg D_s, we have v02/ν≫v02/γsv_0^2/\nu\gg v_0^2/\gamma_s and then γs≫ν\gamma_s\gg\nu. Thus, in the liquid state, Fs(k,t)≈1F_s(k,t) \approx1 over the timescale τc\tau_c and the initial decay of the VACF (and the location of the first zero-crossing) is controlled by the modal current correlation functions, C⊥C_{\perp} and C∥C_{\parallel}.

The integrand of the hydrodynamic VACF formula, Eq. (5), can thus be approximated as

Fs(k,t)C(k,t)≈1−12k2(13v∥2+23v02)t2(40)F_s(k,t)C(k,t) \approx1-\frac{1}{2}k^2\left(\frac{1}{3}v_{\parallel}^2+\frac{2}{3}v_0^2\right)t^2 \tag*{(40)}

and so the short-time VACF is approximately

ψ(t)≈Npa33∫0∞dk k2q(ka)[1−12k2(v∥2+2v02)t2](41)\psi(t)\approx\frac{\mathcal{N}_p a^3}{3}\int_0^\infty dk\,k^2q(ka)\left[1-\frac{1}{2}k^2\left(v_{\parallel}^2+2v_0^2\right)t^2\right] \tag*{(41)}

Equating the quadratic terms in Eq. (38) and Eq. (41), we obtain

ωE2=Npa33∫0∞dk k4q(ka)(v∥2+2v02)(42)\omega_E^2=\frac{\mathcal{N}_p a^3}{3}\int_0^\infty dk\,k^4q(ka)\left(v_{\parallel}^2+2v_0^2\right) \tag*{(42)}

which is the second frequency-moment condition (i.e., ff-sum rule).3

To obtain Eq. (33), explicitly evaluate Eq. (42) for the generalized Gaussian spectrum

ψ(t)=pa381/pΓ(3/p)∫0∞dk k2e−12∣ka∣pFs(k,t)C(k,t)\psi(t)=\frac{pa^3}{8^{1/p}\Gamma(3/p)}\int_0^\infty dk\,k^2e^{-\frac{1}{2}|ka|^p}F_s(k,t)C(k,t)

The scale aa can now be deduced from the correlation time τc\tau_c through the second wavenumber moment of the static spectrum, q(k)q(k):

ωE2=pa3va281/pΓ(3/p)∫0∞dk k4e−12∣ka∣p(43)\omega_E^2=\frac{pa^3v_a^2}{8^{1/p}\Gamma(3/p)}\int_0^\infty dk\,k^4e^{-\frac{1}{2}|ka|^p} \tag*{(43)}

where va2≡(v∥2+2v02)/3v_a^2\equiv(v_{\parallel}^2+2v_0^2)/3. Carrying out the integral analytically and rearranging to solve for aa yields

a=41/pΓ(5/p)Γ(3/p)vaωE(44)a=\sqrt{\frac{4^{1/p}\Gamma(5/p)}{\Gamma(3/p)}}\frac{v_a}{\omega_E} \tag*{(44)}

which agrees with Eq. (33). For the specific case of a pure Gaussian static spectrum, a=3va/ωEa=\sqrt{3}v_a/\omega_E. In the liquid state, it is seen that the main decay timescale is controlled by the molecular scale aa via τc=ωE−1∼a/va\tau_c=\omega_E^{-1}\sim a/v_a.

Zero-frequency constraint for liquids

The Green-Kubo relation for self-diffusion is

Ds=v02∫0∞dtψ(t)(45)D_s=v_0^2\int_0^\infty dt\psi(t) \tag*{(45)}

In the liquid regime, it is reasonable to take C∥(k,t)→0C_{\parallel}(k,t)\to0 and Fs(k,t)→1F_s(k,t)\to1, which allows us to evaluate Eq. (45) analytically using Eq. (5) to obtain

Ds=23[D⊥+χ(p)a2v](46)D_s=\frac{2}{3}\left[D_{\perp}+\chi(p)a^2v\right] \tag*{(46)}

where χ(p)\chi(p) is a coefficient that depends on the sharpness of the spectral cutoff. After rearranging, we obtain

D⊥=32Ds−χ(p)a2v(47)D_{\perp}=\frac{3}{2}D_s-\chi(p)a^2v \tag*{(47)}

which, as shown below, yields analytic constraints for D⊥D_{\perp} when DsD_s is known (e.g., experimentally) and the static spectrum is represented by either the generalized Gaussian or Lorentzian forms (Eqs. (28) and (29)).

For the transverse current, Eq. (45) becomes

Ds=23v02N∫0∞dk k2q(ka)∫0∞dtC⊥(k,t)(48)D_s=\frac{2}{3}v_0^2\mathcal{N}\int_0^\infty dk\,k^2q(ka)\int_0^\infty dt C_{\perp}(k,t) \tag*{(48)}

Carrying out the time integration first yields

v02∫0∞dtC⊥(k,t)=D⊥+vk2(49)v_0^2\int_0^\infty dtC_{\perp}(k,t)=D_{\perp}+\frac{v}{k^2} \tag*{(49)}

so that the kk-integral that remains to be evaluated is

I⊥[q(ka)]=23Npa3∫0∞dk k2q(ka)(D⊥+vk2)(50)I_{\perp}[q(ka)]=\frac{2}{3}\mathcal{N}_p a^3\int_0^\infty dk\,k^2q(ka)\left(D_{\perp}+\frac{v}{k^2}\right) \tag*{(50)}

For a generalized Gaussian spectrum, Eq. (50) becomes

32I⊥[e−12(ka)p]=D⊥+141/pΓ(1/p)Γ(3/p)a2v(51)\frac{3}{2}I_{\perp}\left[e^{-\frac{1}{2}(ka)^p}\right]=D_{\perp}+\frac{1}{4^{1/p}}\frac{\Gamma(1/p)}{\Gamma(3/p)}a^2v \tag*{(51)}

and, for a generalized Lorentzian,

32I⊥[11+(ka)p]=D⊥+csc⁡(π/p)csc⁡(3π/p)a2v(52)\frac{3}{2}I_{\perp}\left[\frac{1}{1+(ka)^p}\right]=D_{\perp}+\frac{\csc(\pi/p)}{\csc(3\pi/p)}a^2v \tag*{(52)}

Inspection of Eqs. (51) and (52) reveals that the transverse current correlation uniquely contributes a pp-dependent coefficient of the a2va^2v term, χ(p)\chi(p), unlike the longitudinal current correlation, as well as depending on the functional form of q(ka)=qp(ka)q(ka)=q_p(ka):

32Ds=32I⊥[qp(ka)]=D⊥+χ(p)a2v(53)\frac{3}{2}D_s=\frac{3}{2}I_{\perp}[q_p(ka)]=D_{\perp}+\chi(p)a^2v \tag*{(53)}

Thus, D⊥D_{\perp} can be readily computed using Eq. (51) or Eq. (52) with Eq. (47). Figure C.1 compares the magnitude of χ(p)\chi(p) for the generalized Gaussian (blue) and generalized Lorentzian (orange) forms for the static spectrum. It is seen, for instance, that a generalized Gaussian with p=4p=4 gives roughly the same contribution to the overall diffusivity as the generalized Lorentzian with p≈6p\approx6.

Parameter determination for dilute fluids

Regularization and the exponential range [req. (4)]. At lower densities, self-diffusion becomes important and Sc\mathrm{Sc} is relatively small as compared to the liquid case. For a dilute gas, where Sc∼O(1)\mathrm{Sc}\sim\mathcal{O}(1), it would be reasonable to assume the exponential range [req. (4)] arises only from the product Fs(k,t)C⊥(k,t)F_s(k,t)C_\perp(k,t). Then, one might define an effective decay rate, Ω0\Omega_0, to be the sum of the SISF and transverse current collision rates, i.e., Ω0=γs+υ\Omega_0=\gamma_s+\upsilon. It will be shown below that this assumption is not unreasonable and can be formally justified by enforcing req. (7) using large-kk approximations for the correlation functions. Specifically, the exponential range should arise for values of kk where kv0>τkv_0>\tau and Ω0=τ−1\Omega_0=\tau^{-1} [req. (4)]. Indeed, unlike in the liquid case, the SISF cannot be treated as approximately constant (i.e., Fs≠1F_s\ne1) over the time interval with the dominant contribution to self-diffusion, which precludes an exact analytic integration over kk. However, for dilute monatomic gases, the dominant contribution to DsD_s generally comes from the large-kk forms of C⊥(k,t)C_\perp(k,t) and Fs(k,t)F_s(k,t), in the Green–Kubo relation allowing an approximate kk integration.

Coefficient $\chi(p)$ for generalized Gaussian and generalized Lorentzian static spectrum functions

Figure C.1. Coefficient χ(p)\chi(p) in Eq. (53) from the transverse component’s contribution to the long-time diffusivity for generalized Gaussian, Eq. (28), and Lorentzian, Eq. (29), static spectrum functions. The dashed line indicates values of pp for which the second-moment kk-integral [req. (5a)] is divergent for the generalized Lorentzian spectrum (i.e., p<6p < 6).

In this scenario, regularization plays a critical role, as seen by considering the slowest relaxation rate of C⊥(k,t)C_\perp(k,t)—the smaller of the two exponents in Eq. (20), which, for large kk, is

−12υk+12υk2−4k2v02∼−k2v02υ+k2D⊥∼−v02D⊥(54)-\frac{1}{2}\upsilon_k+\frac{1}{2}\sqrt{\upsilon_k^2-4k^2v_0^2}\sim-\frac{k^2v_0^2}{\upsilon+k^2D_\perp}\sim-\frac{v_0^2}{D_\perp} \tag*{(54)}

This relaxation rate must be finite as k→∞k\to\infty [req. (4)], which, evidently, necessitates the inclusion of the regularization diffusion coefficient, D⊥D_\perp. A similar argument holds for the exponential range of Fs(k,t)F_s(k,t), where DsD_s is the corresponding regularization coefficient.

Similarly, the longitudinal relaxation rate for large-kk is ∼v02/μ\sim v_0^2/\mu; however, the bulk viscosity of a dilute monatomic gas is typically very small compared to its shear viscosity,[138, 139, 140] making its contribution to the diffusivity small as well. To see this analytically, consider the VACF with a Gaussian static spectrum (p=2p=2) and bulk viscosity set to zero: the longitudinal contribution is the rapidly decaying form ψ∥(t)∼(1−v02t2/a2)exp⁡(−v02t2/2a2)\psi_\parallel(t)\sim(1-v_0^2t^2/a^2)\exp(-v_0^2t^2/2a^2), which integrates exactly to zero. Thus, while C∥C_\parallel can affect the short-time VACF structure, only C⊥C_\perp and FsF_s dictate the exponential range and overall diffusivity under dilute conditions.

Zero-frequency constraint [req. (7)]

Excluding the longitudinal component, the Green–Kubo relation for large kk becomes [req. (7)]

Ds=23v02(γs+v02D⊥)−1(55)D_s=\frac{2}{3}v_0^2\left(\gamma_s+\frac{v_0^2}{D_\perp}\right)^{-1} \tag*{(55)}

Note the appearance of the additional parameter γs\gamma_s due to the treatment of finite ScSc. To uniquely determine the parameters, we now require a separate relation between D⊥D_\perp and υ\upsilon or another way to determine γs\gamma_s. At present, it seems reasonable to assume D⊥→v02/υD_\perp\to v_0^2/\upsilon, which is consistent with the assumption that Ds→v02/γsD_s\to v_0^2/\gamma_s, which was used in Eqs. (22) to (25)—the equations for the self-density and SISF. We remark that when D⊥=v02/υD_\perp=v_0^2/\upsilon, the roots of Eq. (18) factor nicely to give

C⊥(k,t)=υe−k2D⊥t−k2D⊥e−υtυ−k2D⊥(56)C_\perp(k,t)=\frac{\upsilon e^{-k^2D_\perp t}-k^2D_\perp e^{-\upsilon t}}{\upsilon-k^2D_\perp} \tag*{(56)}

matching the neat form of Eq. (25) and preserving the positivity of the VACF in accordance with what is expected for a dilute gas.

Setting D⊥=v02/υD_\perp=v_0^2/\upsilon in Eq. (55), we now have

Ds=23v02γs+υ(57)D_s=\frac{2}{3}\frac{v_0^2}{\gamma_s+\upsilon} \tag*{(57)}

Clearly, we cannot directly set Ds=v02/γsD_s=v_0^2/\gamma_s on the left-hand side of Eq. (57), as there would be no positive solution for υ\upsilon. We instead assume, as in the liquid case, that γs\gamma_s represents the rescaled (self-)collision frequency, while γs0\gamma_{s0} is its corresponding base value determined by Ds=v02/γs0D_s=v_0^2/\gamma_{s0}. With this assumption, Eq. (57) becomes

γs0=32(γs+υ)(58)\gamma_{s0}=\frac{3}{2}(\gamma_s+\upsilon) \tag*{(58)}

Parameter rescaling [req. (6)]

Finally, we require that the Schmidt number be invariant after rescaling: Sc≡υ/Ds=γs0/v0=γs/υSc\equiv\upsilon/D_s=\gamma_{s0}/v_0=\gamma_s/\upsilon, which fixes the ratio of the rescaled collision frequencies to the ratio of their base values. Substituting υ=γs/Sc\upsilon=\gamma_s/Sc into Eq. (58) and solving for γs\gamma_s leads to the condition

γs=23ScSc+1γs0=(a0a)2γs0(59)\gamma_s=\frac{2}{3}\frac{Sc}{Sc+1}\gamma_{s0}=\left(\frac{a_0}{a}\right)^2\gamma_{s0} \tag*{(59)}

where the second equality is obtained by recalling that aa is determined by preserving the long-time asymptotic form of the VACF [req. (6)], which implies

a2=32(1+Sc−1)a02(60)a^2=\frac{3}{2}\left(1+Sc^{-1}\right)a_0^2 \tag*{(60)}

Thus, γs\gamma_s, υ\upsilon, and γ\gamma may be determined, respectively, via γs0=v02/Ds\gamma_{s0}=v_0^2/D_s, υ0=v02/ν\upsilon_0=v_0^2/\nu, and γ0=v02/μ\gamma_0=v_0^2/\mu given known inputs for the transport coefficients (DsD_s, ν\nu, and μ\mu).

Parameters for gaseous and supercritical argon. Given the sparsity of numerical and experimental data for transport coefficients of dilute monatomic gases, the VACFs in Figure 3 were produced using transport coefficients calculated from Enskog theory,[11, 141, 142] supplemented by isothermal data for argon (at 130 K and 300 K) from the NIST Chemistry WebBook. For the transport coefficients, we took Ds→DE=v02/γs0D_s \to D_E = v_0^2/\gamma_{s0}, ν→νE=v02/ν0\nu\to\nu_E = v_0^2/\nu_0, and μ→νbE=v02/γ0\mu\to\nu_{bE} = v_0^2/\gamma_0, where

DE=1.01896D0g(σ)(61)D_E = 1.01896\frac{D_0}{g(\sigma)} \tag*{(61)}
νE=1.016ν0[1g(σ)+0.8(bn0)+0.7615(bn0)2g(σ)](62)\nu_E = 1.016\nu_0\left[\frac{1}{g(\sigma)}+0.8(bn_0)+0.7615(bn_0)^2g(\sigma)\right] \tag*{(62)}
νbE=165πν0(bn0)2g(σ)(63)\nu_{bE} = \frac{16}{5\pi}\nu_0(bn_0)^2g(\sigma) \tag*{(63)}

are the Enskog diffusion coefficient, (kinematic) shear viscosity, and (kinematic) bulk viscosity, respectively.[142] In Eqs. (61) to (63), n0n_0 is the equilibrium number density, σ\sigma the hard-sphere diameter, and b≡(2/3)πσ3b \equiv(2/3)\pi\sigma^3 is the second virial coefficient for a hard-sphere fluid. D0D_0 and ν0\nu_0 are the values of the self-diffusion and shear viscosity coefficients in the limit of zero density,

D0=3ν08πn0σ2(64)D_0 = \frac{3\nu_0}{8\sqrt{\pi}n_0\sigma^2} \tag*{(64)}
ν0=5ν016πn0σ2(65)\nu_0 = \frac{5\nu_0}{16\sqrt{\pi}n_0\sigma^2} \tag*{(65)}

and g(σ)g(\sigma) is the radial distribution function evaluated at the hard-sphere point of contact. Analytic approximations for g(σ)g(\sigma) can be obtained from the Percus–Yevick equation, scaled particle theory,[143, 144, 145] or the Carnahan–Starling equation of state;[146, 147, 148] we used the Carnahan–Starling approximation for the 3D pair distribution:

g(σ)≈1−ϕ/2(1−ϕ)3(66)g(\sigma) \approx\frac{1-\phi/2}{(1-\phi)^3} \tag*{(66)}

where the packing fraction ϕ≡(π/6)n0σ3\phi\equiv(\pi/6)n_0\sigma^3. For all calculations in Figure 3, we used a Gaussian static spectrum (p=2p = 2) primarily as proof-of-principle, though one should expect higher values of pp to apply only at the highest densities; e.g., in the liquid state, where propagating longitudinal and shear wave modes may be significant for softer intermolecular potentials. We also took σ=3.405\sigma= 3.405 Å (argon Lennard-Jones diameter), m=40m = 40 Da for the (atomic) mass, and set D⊥=v02/νD_\perp= v_0^2/\nu (as discussed above) and D∥=4μ/3D_\parallel= 4\mu/3 (as in the liquid case). We determined v∥2v_\parallel^2 from the adiabatic index, i.e., v∥2/v02=CP/CVv_\parallel^2/v_0^2 = C_P/C_V, using heat capacity values obtained from the NIST Chemistry WebBook for each combination of temperature and density, TT and ρ0\rho_0, where ρ0=mn0\rho_0 = mn_0.

Figure E.1. VACFs corresponding to a dense supercritical Lennard-Jones fluid studied in Ref. [34] (cf. Figure 1). The VACF from the hybrid procedure described in this appendix (blue) is depicted with VACFs produced for two different (larger) values of aa: setting a→3/2(1+Sc−1)a0a \to\sqrt{3/2}(1+\mathrm{Sc}^{-1})a_0 via Eq. (60) (orange) and explicitly fixing a→a0a \to a_0 (green).

Comparison with MD calculations for a supercritical Lennard-Jones fluid

In Ref. 34, MD simulations are performed for a supercritical Lennard-Jones (LJ) fluid at a reduced density n0∗≡n0σ3=0.5n_0^*\equiv n_0\sigma^3=0.5 and reduced temperature T∗≡kBT/ϵ=1.5T^*\equiv k_BT/\epsilon=1.5, where σ\sigma and ϵ\epsilon are the LJ diameter and energy, respectively. To compare with the results of Ref. 34, we took m=40m=40 Da, σ=3.405\sigma=3.405 Å, and ϵ=120\epsilon=120 K, so that the (dimensional) mass density and temperature are, respectively, ρ=mn0=841.3 kg/m3\rho=mn_0=841.3\,\mathrm{kg}/\mathrm{m}^3 and T=180T=180 K; note that, for this temperature-density combination, P≈19.2P\approx19.2 MPa. For ν∥\nu_\parallel, we again use the adiabatic index computed from heat capacities obtained from the NIST Chemistry WebBook, SRD 69 for isothermal properties of argon (i.e., ν∥2/ν02=CP/CV\nu_\parallel^2/\nu_0^2=C_P/C_V). To obtain base values for the collision frequencies, i.e., ν0\nu_0, γ0\gamma_0, etc., we used the same Enskog diffusion and viscosity coefficient formulas given in Appendix D. However, we chose to treat this comparatively dense supercritical (SC) state slightly differently, so that the packing fraction is ϕ≈0.262\phi\approx0.262—more than half of argon’s triple point (TP) density (ϕ≈0.445\phi\approx0.445). In particular, we employed a hybrid treatment (detailed below) wherein the molecular scale aa was computed via the second-moment condition (5) and Eq. (44) using simple physical arguments to obtain an estimated Einstein frequency of ωE=4.75 ps−1\omega_E=4.75\ \mathrm{ps}^{-1}. The resulting VACF—Fig. E.1, blue—is in striking agreement with the MD calculations shown in Figure 1 of Ref. 34. Our estimate for the Einstein frequency is based on the following calculations and physical reasoning.

First, at ϕ≈0.262\phi\approx0.262, the mean interparticle separation (center-to-center) is d0SC∼n0−1/3=4.29d_0^{\mathrm{SC}}\sim n_0^{-1/3}=4.29 Å, so the contact (surface-to-surface) distance, dSCd^{\mathrm{SC}}, may be estimated as dSC≈d0SC−σ≈0.885d^{\mathrm{SC}}\approx d_0^{\mathrm{SC}}-\sigma\approx0.885 Å; likewise, dTP≈0.190d^{\mathrm{TP}}\approx0.190 Å at the TP. It is within reason to assume the Einstein frequency, ωE\omega_E is physically relevant for SC argon at the given density because the surface-to-surface spacing is (still) substantially smaller than the particle diameter; i.e., we assume that particles cannot easily “break through the cage” formed by its neighbors at this (packing) density. Second, to arrive at an estimate for ωE\omega_E, we assume that the larger interparticle spacing in the SC case leads to an increased oscillation period, Δτ\Delta\tau, over that of TP argon, where Δτ=τSC−τTP\Delta\tau= \tau^{\mathrm{SC}} - \tau^{\mathrm{TP}}, and τSC≡2π/ωESC\tau^{\mathrm{SC}} \equiv2\pi/\omega_E^{\mathrm{SC}} and τTP≡2π/ωETP\tau^{\mathrm{TP}} \equiv2\pi/\omega_E^{\mathrm{TP}} are the respective oscillation periods. Finally, to estimate the increase in oscillation period, we make the simple physical assumption that Δτ\Delta\tau is an intervening ballistic interval arising from the increased surface-to-surface distance, an additional separation of dSC−dTP≈0.695d^{\mathrm{SC}} - d^{\mathrm{TP}} \approx0.695 Å over the TP case. This leads to Δτ≈2(dSC−dTP)/vb\Delta\tau\approx2(d^{\mathrm{SC}} - d^{\mathrm{TP}})/v_b, where the factor of 2 accounts for two ballistic traversals per cage oscillation and we take the ballistic speed to vb≈2kBT/m=2v0v_b \approx\sqrt{2k_{\mathrm{B}}T/m} = \sqrt{2}v_0 (i.e., the most probable speed of a Maxwell-Boltzmann distribution).

Putting everything together, we arrive at a rough estimate for the oscillation period for the dense SC fluid:

τSC=2πωETP+2(dSC−dTP)2v0≈2π7.7 ps−1+2(0.695 A˚)2.74 A˚/ps≈1.32 ps\tau^{\mathrm{SC}} = \frac{2\pi}{\omega_E^{\mathrm{TP}}} + \frac{2(d^{\mathrm{SC}} - d^{\mathrm{TP}})}{\sqrt{2}v_0} \approx\frac{2\pi}{7.7\ \mathrm{ps}^{-1}} + \frac{2(0.695\ \text{Å})}{2.74\ \text{Å}/\mathrm{ps}} \approx1.32\ \mathrm{ps}

which yields for the Einstein frequency

ωESC=2πτSC≈4.75 ps−1(67)\omega_E^{\mathrm{SC}} = \frac{2\pi}{\tau^{\mathrm{SC}}} \approx4.75\ \mathrm{ps}^{-1} \tag*{(67)}

Finally, we use a generalized Gaussian spectrum with p≈1.5p \approx1.5 for which Eq. (44) gives a≈2.28a \approx2.28 Å and yields excellent agreement (Fig. E.1, blue) throughout the timescales sampled by MD. Note that time is expressed in LJ reduced units, t∗≡t/τLJt^* \equiv t/\tau_{\mathrm{LJ}}, where τLJ≡mσ2/ϵ\tau_{\mathrm{LJ}} \equiv\sqrt{m\sigma^2/\epsilon} is the characteristic LJ timescale; for argon, τLJ≈2.156\tau_{\mathrm{LJ}} \approx2.156 ps. Figure E.1 also shows VACFs computed using two different (larger) values of aa: computing aa from Eq. (60) for dilute fluids (orange) and explicitly setting a→a0a \to a_0 (green). Note that the blue curve reproduces the rather subtle “plateau” region (t∗≈0.15t^* \approx0.15–0.60.6) due to the sound-like mode originating from the longitudinal component, as well as the nuanced transition from exponential-like to diffusive decay for t∗≳1t^* \gtrsim1.

References

  1. [1]J. P. Boon and S. Yip, Molecular Hydrodynamics (Dover, New York, 1991).
  2. [2]D. A. McQuarrie, Statistical Mechanics, 1st ed. (University Science Books, Sausalito, 2000).
  3. [3]J.-P. Hansen and I. R. McDonald, Theory of Simple Liquids: with Applications to Soft Matter, 4th ed. (Academic Press, Oxford, 2013).
  4. [4]A. Einstein, Ann. Phys. 322, 549 (1905).
  5. [5]B. J. Alder and T. E. Wainwright, Phys. Rev. Lett. 18, 988 (1967).
  6. [6]B. J. Alder and T. E. Wainwright, Phys. Rev. A 1, 18 (1970).
  7. [7]D. Enskog, Uppsala: Almquist & Wiksells Boktryckeri (1917).
  8. [8]D. Enskog, Vetenskaps Akad. Handl. 63 (1922).
  9. [9]G. Brush, Kinetic Theory: The Chapman-Enskog solution of the transport equation for moderately dense gases, Vol. 3 (Pergamon Press, Oxford, 1972).
  10. [10]M. V. Reisbois and M. D. Leener, Classical kinetic theory of fluids, 1st ed. (Wiley, New York, 1977).
  11. [11]S. Chapman and T. G. Cowling, The Mathematical Theory of Non-uniform Gases: An Account of the Kinetic Theory of Viscosity, Thermal Conduction and Diffusion in Gases, 3rd ed. (Cambridge University Press, Cambridge, 1990).
  12. [12]R. Zwanzig and M. Bixon, Phys. Rev. A 2, 2005 (1970).
  13. [13]C. Schofield, in Statistical Mechanics, SPR - Statistical Mechanics, Vol. 2, edited by K. Singer (The Royal Society of Chemistry, 1975) pp. 1–54.
  14. [14]P. C. Martin and S. Yip, Phys. Rev. 170, 151 (1968).
  15. [15]D. Forster, P. C. Martin, and S. Yip, Phys. Rev. 170, 155 (1968).
  16. [16]K. Kim and M. Nelkin, Phys. Rev. A 4, 2065 (1971).
  17. [17]Y. Pomeau and P. Résibois, Phys. Rep. 19, 63 (1975).
  18. [18]L. Sjogren and A. Sjolander, J. Phys. C: Solid State Phys. 12, 4369 (1979).
  19. [19]T. Gaskell, U. Balucani, and R. Vallauri, Phys. Chem. Liq. 19, 193 (1989).
  20. [20]U. Balucani, R. Vallauri, T. Gaskell, and S. F. Duffy, J. Phys. Condens. Matter 2, 5015 (1990).
  21. [21]N. Anento, J. A. Padró, and M. Canales, J. Chem. Phys. 111, 10210 (1999).
  22. [22]R. W. Zwanzig, Lectures in Theoretical Physics, edited by W. E. Brittin, B. W. Downs, and J. Downs, Lectures Delivered at the Summer Institute for Theoretical Physics, Vol. 3 (Wiley Interscience, New York, 1961) pp. 106–141.
  23. [23]H. Mori, Progr. Theoret. Phys. 33, 423 (1965).
  24. [24]H. Mori, Progr. Theoret. Phys. 34, 399 (1965).
  25. [25]D. Levesque and L. Verlet, Phys. Rev. A 2, 2514 (1970).
  26. [26]D. Levesque, L. Verlet, and J. Kürkijarvi, Phys. Rev. A Gen. Phys. 7, 1690 (1973).
  27. [27]J. Kushick and B. J. Berne, J. Chem. Phys. 59, 3732 (1973).
  28. [28]C. Schofield, Comput. Phys. Commun. 5, 17 (1973); I. M. de Schepper, J. C. van Rijs, A. A. van Well, P. Verkkerk, L. A. de Graaf, and C. Bruin, Phys. Rev. A 29, 1602 (1984); A. McDonough, S. P. Russo, and I. K. Snook, Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 63, 026109 (2001); R. F. A. A. F. Ould-Kaddour, and D. Levesque, Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 74, 011202 (2006); T. Sanghi, R. Bhadauira, and N. R. Aluru, J. Chem. Phys. 145, 134108 (2016).
  29. [29]D. Lesnicki and R. Vuilleumier, J. Chem. Phys. 147, 094502 (2017).
  30. [30]K. H. Han, C. Kim, P. Talkner, G. E. Karniadakis, and E. K. Lee, J. Chem. Phys. 148, 024506 (2018); K. Mizuta, Y. Ishii, K. Kim, and N. Matubayasi, Soft Matter 15, 4380 (2019).
  31. [31]H. Zhao and H. Zhao, Phys Rev E 103, L030102 (2021).
  32. [32]C. Hijón, P. Español, E. Vanden-Eijnden, and R. Delgado-Buscalioni, Faraday Discuss. 144, 301 (2009); P. Español, J. G. Anero, and I. Zúñiga, J. Chem. Phys. 131, 244117 (2009); P. Español and H. Löwen, J. Chem. Phys. 131, 244101 (2009); J. A. Morrone, T. E. Markland, M. Ceriotti, and B. J. Berne, J. Chem. Phys. 134, 014103 (2011); S. Izvekov, J. Chem. Phys. 138, 134106 (2013); A. Carof, R. Vuilleumier, and B. Rotenberg, J. Chem. Phys. 140, 124103 (2014); Z. Li, X. Bian, X. Li, and G. E. Karniadakis, J. Chem. Phys. 143, 243128 (2015).
  33. [33]P. Español and A. Donev, J. Chem. Phys. 143, 234104 (2015).
  34. [34]D. Lesnicki, R. Vuilleumier, A. Carof, and B. Rotenberg, Phys. Rev. Lett. 116, 147804 (2016).
  35. [35]G. Jung and F. Schmid, J. Chem. Phys. 144, 204104 (2016); S. Izvekov, J. Chem. Phys. 146, 124109 (2017); G. Jung, M. Hanke, and F. Schmid, J. Chem. Theory Comput. 13, 2481 (2017).
  36. [36]G. Jung, M. Hanke, and F. Schmid, Soft Matter 14, 9368 (2018); M. Rossi, V. Kapil, and M. Ceriotti, J. Chem. Phys. 148, 102301 (2018); S. Izvekov, J. Chem. Phys. 151, 104109 (2019); D. Duque-Zumajo, J. A. de la Torre, and P. Español, J. Chem. Phys. 152, 174108 (2020); V. Klippenstein, M. Tripathy, G. Jung, F. Schmid, and N. F. A. van der Vegt, J. Phys. Chem. B 10.1021/acs.jpcb.1c01120 (2021).
  37. [37]A. Malevanets and R. Kapral, J. Chem. Phys. 112, 7260 (2000); J. T. Padding and A. A. Louis, Phys. Rev. Lett. 93, 220601 (2004); Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 74, 031402 (2006); V. Dahirel, M. Jartad, J.-F. Dufrêche, and P. Turq, J. Chem. Phys. 126, 114108 (2007); L. Wang and M. Quintard, in Advances in Transport Phenomena: 2009, edited by L. Wang (Springer Science & Business Media, Heidelberg, 2009), pp. 179–243; F. Bonaccorso, A. Montessori, A. Tiribocchi, G. Amati, M. Bernaschi, M. Lauricella, and S. Succi, Comput. Phys. Commun. 256, 107455 (2020); F. Moore, J. Russo, T. B. Liverpool, and C. P. Royall, J. Chem. Phys. 158, 104907 (2023).
  38. [38]D. L. Koch and G. Subramanian, Annu. Rev. Fluid Mech. 43, 637 (2011); S. Wang and A. M. Ardekani, J. Fluid Mech. 702, 286 (2012); J. Elgeti, R. G. Winkler, and G. Gompper, Rep. Prog. Phys. 78, 056601 (2015); P. K. Ghosh, Y. Li, G. Marchegiani, and F. Marchesoni, J. Chem. Phys. 143, 211101 (2015); A. Sharma and J. M. Brader, J. Chem. Phys. 145, 161101 (2016); G. Szamel, J. Chem. Phys. 150, 124901 (2019); B. Chakrabarti, M. J. Shelley, and S. Fürthauer, Phys. Rev. Lett. 130, 128202 (2023); L. Caprini and H. Löwen, Phys. Rev. Lett. 130, 148202 (2023).
  39. [39]J. García De La Torre, M. L. Huertas, and B. Carrasco, Biophys. J. 78, 719 (2000); M. X. Fernandes and J. G. de la Torre, Biophys. J. 83, 3039 (2002); C. P. Brangwynne, G. H. Koenderink, F. C. MacKintosh, and D. A. Weitz, J. Cell Biol. 183, 583 (2008); R. Kapral and A. S. Mikhailov, Physica D 318–319, 100 (2016).
  40. [40]T.-W. Nee and R. Zwanzig, J. Chem. Phys. 52, 6353 (1970); F. Lantelmе, P. Turq, and P. Schofield, J. Chem. Phys. 71, 2507 (1979).
  41. [41]T. Gaskell and M. S. Woolfson, J. Phys. C: Solid State Phys. 15, 6339 (1982); R. Malik, D. Burch, M. Bazant, and G. Ceder, Nano Lett. 10, 4123 (2010); M. A. Gebbie, M. Valtiner, X. Banquy, E. T. Fox, W. A. Henderson, and J. N. Israelachvili, Proc. Natl. Acad. Sci. U. S. A. 110, 9674 (2013); D. M. Wilkins, D. E. Manolopoulos, S. Roke, and M. Ceriotti, J. Chem. Phys. 146, 181103 (2017); J. P. de Souza and M. Z. Bazant, J. Phys. Chem. C 124, 11414 (2020); P. K. Ghorai and D. V. Matyushov, J. Phys. Chem. B 124, 3754 (2020); S. M. Sarhangi and D. V. Matyushov, J. Phys. Chem. Lett. 10137 (2020); T. Samanta and D. V. Matyushov, Phys. Rev. Research 3, 023025 (2021); S. Kournopoulos, A. J. Haslam, G. Jackson, A. Galindo, and M. Schoen, J. Chem. Phys. 156, 154111 (2022).
  42. [42]H. Tang and Schweizer,Kenneth S., J. Chem. Phys. 105, 779 (1996); V. Lisy, J. Tothova, and A. V. Zatovsky, J. Chem. Phys. 121, 10699 (2004); J. R. Howse, R. A. L. Jones, A. J. Ryan, T. Gough, R. Vafabakhsh, and R. Golestanian, Phys. Rev. Lett. 99, 048102 (2007); M. Bernabei, A. J. Moreno, E. Zaccarelli, F. Sciortino, and J. Colmenero, J. Chem. Phys. 134, 024523 (2011); M.-J. Huang, J. Schofield, P. Gaspard, and R. Kapral, J. Chem. Phys. 149, 024904 (2018); P. Gaspard and R. Kapral, J. Chem. Phys. 148, 134104 (2018); G. Szamel, J. Chem. Phys. 156, 191102 (2022).
  43. [43]F. B. Usabiaga, X. Xie, R. Delgado-Buscalioni, and A. Donev, J. Chem. Phys. 139, 214113 (2013).
  44. [44]M. H. Ernst, E. H. Hauge, and J. M. J. van Leeuwen, Phys. Rev. Lett. 25, 1254 (1970).
  45. [45]M. H. Ernst, E. H. Hauge, and J. M. J. van Leeuwen, Phys. Rev. A 4, 2055 (1971).
  46. [46]T. Gaskell and S. Miller, J. phys. 11, 3749 (1978).
  47. [47]T. Gaskell and S. Miller, J. phys. 11, 4839 (1978).
  48. [48]T. Gaskell and S. Miller, J. phys. 12, 2705 (1979).
  49. [49]H. Grad, Communications on pure and applied mathematics 2, 331 (1949).
  50. [50]H. Struchtrup and M. Torrilhon, Phys. Fluids 15, 2668 (2003).
  51. [51]H. C. Öttinger, Beyond Equilibrium Thermodynamics, edited by H. C. Öttinger (John Wiley & Sons, 2005).DOI
  52. [52]H. Öttinger and H. Struchtrup, Multiscale Model. Simul. 6, 53 (2007).
  53. [53]D. Jou, J. Casas-Vázquez, and M. Criado-Sancho, Thermodynamics of Fluids Under Flow (Springer Science & Business Media, 2010).
  54. [54]H. C. Öttinger, Phys. Rev. Lett. 104, 120601 (2010).
  55. [55]H. Struchtrup and M. Torrilhon, Phys. Fluids 25, 052001 (2013).
  56. [56]M. Torrilhon, Commun. Comput. Phys. 18, 529 (2015).
  57. [57]P. L. Bhatnagar, E. P. Gross, and M. Krook, Phys. Rev. 94, 511 (1954).
  58. [58]R. Zwanzig, Nonequilibrium Statistical Mechanics (Oxford University Press, Oxford, 2001).DOI
  59. [59]Y. L. Klimontovich, Statistical Theory of Open Systems: Volume 1: A Unified Approach to Kinetic Description of Processes in Active Systems, edited by A. Van der Merwe, Fundamental Theories of Physics, Vol. 67 (Kluwer Academic Publishers, Dordrecht, 2012).
  60. [60]R. Zwanzig and R. D. Mountain, J. Chem. Phys. 43, 4464 (1965).
  61. [61]D. Schofield, Proc. Phys. Soc. London 88, 149 (1966).
  62. [62]B. J. Alder and W. E. Alley, Phys. Today 37, 56 (1984).
  63. [63]C.-H. Chung and S. Yip, Phys. Rev. 182, 323 (1969).
  64. [64]U. Balucani, R. Vallauri, T. Gaskell, and M. Gori, J. Phys. C: Solid State Phys. 18, 3133 (1985).
  65. [65]U. Balucani, R. Vallauri, and T. Gaskell, Phys. Rev. A Gen. Phys. 35, 4263 (1987).
  66. [66]U. Balucani, R. Vallauri, and T. Gaskell, Ber. Bunsenges. Phys. Chem. 94, 261 (1990).
  67. [67]A. Verdaguer and J. A. Padró, Phys. Rev. E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 62, 532 (2000).
  68. [68]M. Colangeli, M. Kröger, and H. C. Öttinger, Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 80, 051202 (2009).
  69. [69]G. Garbaroglio, R. Vallauri, and U. Bafile, J. Chem. Phys. 148, 174501 (2018).
  70. [70]U. Balucani, R. Vallauri, T. Gaskell, and M. Gori, Phys. Lett. A 102, 109 (1984).
  71. [71]D. C. Leslie, Rep. Prog. Phys. 36, 1365 (1973).
  72. [72]V. S. L’vov, Scale invariant theory of fully developed hydrodynamic turbulence-Hamiltonian approach (1991).DOI
  73. [73]Y. Zhou, Phys. Rep. 935, 1 (2021).
  74. [74]A. Z. Akcasu and E. Daniels, Phys. Rev. A Gen. Phys. 2, 962 (1970).
  75. [75]S.-H. Chen and A. Rahman, Mol. Phys. 34, 1247 (1977).
  76. [76]G. Karniadakis, A. Beskok, and N. Aluru, Microflows and Nanoflows: Fundamentals and Simulation, edited by S. S. Antman, J. E. Marsden, and L. Sirovich, Interdisciplinary Applied Mathematics, Vol. 29 (Springer, New York, 2005).
  77. [77]S. L. Seyler, Computational Approaches to Simulation and Analysis of Large Conformational Transitions in Proteins, Ph.D. thesis, Arizona State University (2017).
  78. [78]H. Van Beijeren and M. H. Ernst, Physica 68, 437 (1973).
  79. [79]H. Van Beijeren and M. H. Ernst, Physica 70, 225 (1973).
  80. [80]J. Karkheck and G. Stell, J. Chem. Phys. 75, 1475 (1981).
  81. [81]D. Henderson, Fundamentals of Inhomogeneous Fluids (CRC Press, New York, 1992).DOI
  82. [82]D. Schiff, Phys. Rev. 186, 151 (1969).
  83. [83]T. Geszti, J. Phys. C: Solid State Phys. 9, L263 (1976).
  84. [84]M. Canales and J. A. Padró, J. Phys. Condens. Matter 9, 11009 (1997).
  85. [85]M. Canales and J. A. Padró, Phys. Rev. E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 60, 551 (1999).
  86. [86]A. Rahman, Phys. Rev. A 9, 1667 (1974).
  87. [87]N. D. C. Andrade and E. R. Dobbs, Proc. R. Soc. Lond. A Math. Phys. Sci. 211, 12 (1952).
  88. [88]E. W. Lemmon and R. T. Jacobsen, Int. J. Thermophys. 25, 21 (2004).
  89. [89]A. H. M. Zaheri, S. Srivastava, and K. Tankeshwar, J. Phys. Condens. Matter 15, 6683 (2003).
  90. [90]R. S. Chatwell and J. Vrabej, J. Chem. Phys. 152, 094503 (2020).
  91. [91]J. J. Van Loef, Physica 75, 115 (1974).
  92. [92]D. Fincham and D. M. Heyes, Chem. Phys. 78, 425 (1983).
  93. [93]M. H. Ernst, E. H. Hauge, and J. M. J. van Leeuwen, Phys. Lett. A 34, 419 (1971).
  94. [94]J. R. Dorfman and E. G. D. Cohen, Phys. Rev. Lett. 25, 1257 (1970).
  95. [95]J. R. Dorfman and E. G. D. Cohen, Phys. Rev. A 12, 292 (1975).
  96. [96]G. Balucani, A. Torcini, and R. Vallauri, Phys. Rev. A 46, 2159 (1992).
  97. [97]A. Rahman, Phys. Rev. 136, A405 (1964).
  98. [98]K. Meier, A. Laesecke, and S. Kabelac, J. Chem. Phys. 121, 9526 (2004).
  99. [99]C. Kim, O. Borodin, and G. E. Karniadakis, J. Comput. Phys. 302, 485 (2015).
  100. [100]F. B. Usabiaga, I. Pagonabarraga, and R. Delgado-Buscalioni, J. Comput. Phys. 235, 701 (2013).
  101. [101]F. Balboa Usabiaga, R. Delgado-Buscalioni, B. E. Griffith, and A. Donev, Comput. Methods Appl. Mech. Eng. 269, 139 (2014).
  102. [102]J. V. Boussinesq, C. R. Acad. Sci. Paris 100, 935 (1885).
  103. [103]A. B. Basset, Proc. Lond. Math. Soc. s1-19, 46 (1887).
  104. [104]C. W. Oseen, Neuere methoden und ergebnisze in der hydrodynamik (1927).
  105. [105]M. R. Maxey and J. J. Riley, Phys. Fluids 26, 883 (1983).
  106. [106]P. Gaspard, Physica A: Statistical Mechanics and its Applications, 121823 (2019).
  107. [107]S. L. Seyler and Pressé, Steve, Phys. Rev. Research 1, 032003 (2019).
  108. [108]K. Trachenko, Phys Rev E 96, 062134 (2017).
  109. [109]R. Fürth, Ann. Phys. 53, 177 (1917).
  110. [110]G. I. Taylor, Proc. Lond. Math. Soc. 2, 196 (1922).
  111. [111]S. Goldstein, Quart. J. Mech. Appl. Math. 4, 129 (1951).
  112. [112]S. A. Khrapak, Molecules 26, 103390/ molecules26247499 (2021).
  113. [113]R. Zwanzig, J. Chem. Phys. 79, 4507 (1983).
  114. [114]U. Balucani and M. Zoppi, Dynamics of the Liquid State, edited by S. W. Lovesey and E. W. J. Mitchell, Oxford Series on Neutron Scattering in Condensed Matter, Vol. 10 (Clarendon Press, Oxford, 1994).DOI
  115. [115]T. S. Chow and J. J. Hermans, J. Chem. Phys. 56, 3150 (1972).
  116. [116]M. M. Nelkin, The Physics of Fluids 15, 1685 (1972).
  117. [117]S. L. Seyler and Pressé, J. Chem. Phys. 153, 041102 (2020).
  118. [118]D. Chakraborty, Eur. Phys. J. B 83, 375 (2011).
  119. [119]G. Jung and F. Schmid, Phys. Fluids 29, 126101 (2017).
  120. [120]J. Lee, S. L. Seyler, and Pressé, J. Chem. Phys. 151, 094108 (2019).
  121. [121]I. Goychuk, Phys. Rev. Lett. 123, 180603 (2019).
  122. [122]I. Goychuk and T. Pöschel, Phys. Rev. E 102, 012139 (2020).
  123. [123]M. V. Díaz, Eur. Phys. J. E Soft Matter 44, 141 (2021).
  124. [124]B. J. Cherayil, J. Phys. Chem. B 10.1021/acs.jpcb.2c03273 (2022).
  125. [125]J. Spiechowicz, I. G. Marchenko, P. Häנגgi, and J. Luczka, Entropy 25, 10.3390/e25010042 (2022).
  126. [126]U. R. Pedersen, N. P. Bailey, T. B. Schrøder, and J. C. Dyre, Phys. Rev. Lett. 100, 015701 (2008).
  127. [127]N. Ohtori, S. Miyamoto, and Y. Ishii, Phys Rev E 95, 052122 (2017).
  128. [128]Y. D. Fomin, V. N. Ryzhov, E. N. Tsiok, J. E. Proctor, C. Prescher, V. B. Prakapenka, K. Trachenko, and V. V. Brazhkin, J. Phys. Condens. Matter 30, 134003 (2018).
  129. [129]G. Pastore, B. Bernu, J. P. Hansen, and Y. Hiwatari, Phys. Rev. A Gen. Phys. 38, 454 (1988).
  130. [130]J.-L. Barrat, J.-N. Roux, and J.-P. Hansen, Chem. Phys. 149, 197 (1990).
  131. [131]U. Balucani, R. Vallauri, and T. Gaskell, Nuovo Cimento C 12, 511 (1990).
  132. [132]A. M. Puertas, C. De Michele, F. Sciortino, P. Tartaglia, and E. Zaccarelli, J. Chem. Phys. 127, 144906 (2007).
  133. [133]M. Baity-Jesi and D. R. Reichman, J. Chem. Phys. 151, 084503 (2019).
  134. [134]G. Ren and Y. Wang, Phys. Chem. Chem. Phys. 23, 24541 (2021).
  135. [135]V. A. Levashov, J. R. Morris, and T. Egami, J. Chem. Phys. 138, 044507 (2013).
  136. [136]M. Ozawa and G. Biroli, Phys. Rev. Lett. 130, 138201 (2023).
  137. [137]L. D. Landau and E. M. Lifshitz, Fluid Mechanics, 3rd ed., Vol. 6 (Pergamon Press, Oxford, 1966).
  138. [138]K. Meier, A. Laesecke, and S. Kabelac, J. Chem. Phys. 122, 14513 (2005).
  139. [139]F. Jaeger, O. K. Matar, and E. A. Müller, J. Chem. Phys. 148, 174504 (2018).
  140. [140]B. Sharma, S. Pareek, and R. Kumar, Eur. J. Mech. B. Fluids 98, 32 (2023).
  141. [141]J. J. Erpenbeck and W. W. Wood, Phys. Rev. A 43, 4254 (1991).
  142. [142]D. M. Heyes, S. Pieprzyk, and A. C. Brańka, J. Chem. Phys. 157, 114502 (2022).
  143. [143]H. Reiss, H. L. Frisch, and J. L. Lebowitz, J. Chem. Phys. 31, 369 (1959).
  144. [144]H. Reiss, H. L. Frisch, E. Helfand, and J. L. Lebowitz, J. Chem. Phys. 32, 119 (1960).
  145. [145]E. Helfand, H. Reiss, H. L. Frisch, and J. L. Lebowitz, J. Chem. Phys. 33, 1379 (1960).
  146. [146]N. F. Carnahan and K. E. Starling, J. Chem. Phys. 51, 635 (1969).
  147. [147]Y. Song, E. A. Mason, and R. M. Statt, J. Phys. Chem. 93, 6916 (1989).
  148. [148]H. Sigurgeirsson and D. M. Heyes, Mol. Phys. 101, 469 (2003).

Paper details

vlnc:W_xkxnzyvm
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents