betaFeedback

WorksPhysics and Astronomy

Heat transfer problem of a dense gas described by the Enskog equation with a modification of the Enskog factor

Abstract

Heat transfer in a dense gas between two parallel plates is studied for assessing the quantitative difference of two variants of the Enskog equation, i.e., the original Enskog equation and the Enskog equation with a modified Enskog factor recently proposed in Phys. Rev. E 111, 065108 (2025). The main advantage of the latter is that the H theorem, which the former lacks, has been established. The influence of the modification is assessed by comparing the macroscopic quantities in detail. It is found that the impact of this modification is quite limited and the two variants reasonably agree as a whole.

Introduction

Recently a study of the dense gas behavior through the kinetic theory attracts attentions, coming from the increasing demands for applications to micro-, submicro-, and even smaller systems. As the mean free path of molecules and the system size simultaneously decrease, the molecular size and the volume fraction of molecules become not necessarily negligible, and accordingly the Enskog equation [1, 2] comes into play in place of the Boltzmann equation. Although the Enskog equation with its original form (original Enskog equation, OEE for short) has been successfully applied to fundamental physical problems, thanks to the developments of its numerical methods [3, 4, 5], it does not satisfy the Boltzmann’s H theorem, unfortunately. In the meantime, we have recently proposed [6] a variant of the Enskog equation that slightly modifies the Enskog factor in the collision integral (EESM for short), for which the H theorem has been established. This new variant is currently in the initial stages of quantitative evaluation for application to physical problems.

In the present work, we revisit the time evolutional heat transfer problem between two parallel plates [7, 8]. On the basis of both OEE and EESM, we numerically study the gas behavior triggered by a sudden temperature change of one plate and evaluate the influence of modifying the Enskog factor.

Problem and Formulation

Consider a dense gas between two parallel plates at rest located at X1=±L/2X_1=\pm L/2. The gas is in thermal equilibrium with the plates kept at a common temperature T−T_-. At time t=0t=0, the temperature of the plate at X1=L/2X_1=L/2 is instantaneously changed from T−T_- to T+T_+, see Figure 1. We will investigate the time-dependent behavior of the gas triggered by this change under the following assumptions. (i) the gas molecules are hard spheres of a common diameter σ\sigma and a mass mm, (ii) the behavior of the gas is described by the Enskog equation, (iii) the gas obeys the Carnahan–Starling equation of state (EoS) [9] at a uniform equilibrium state, (iv) the state of the gas does not change in the X2X_2- and X3X_3-directions, and (v) the molecules are diffusely reflected on the plates.

Figure 1. Problem setting

Let D={X1∣−(L−σ)/2<X1<(L−σ)/2}D=\{X_1\mid-(L-\sigma)/2<X_1<(L-\sigma)/2\}, where the center of gas molecules can be located. This DD is different from D#D^\# = {X1∣−L/2<X1<L/2}\{X_1\mid-L/2<X_1<L/2\}, which denotes the spatial domain between the plates. Let tt and ξ\boldsymbol{\xi} be a time and a molecular velocity, respectively. Then, denoting the one-particle velocity distribution function (VDF) of gas molecules by f(t,X1,ξ)f(t,X_1,\boldsymbol{\xi}), the Enskog equation for the present spatially one-dimensional problem is written as

∂f∂t+ξ1∂f∂X1=σ2m∫S2×R3[g(X1+,X1)f∗′(X1+)f′(X1)−g(X1−,X1)f∗(X1−)f(X1)]Vαθ(Vα) dΩ(α) dξ∗(1)\frac{\partial f}{\partial t}+\xi_1\frac{\partial f}{\partial X_1}=\frac{\sigma^2}{m}\int_{\mathbb{S}^2\times\mathbb{R}^3}\left[g(X_1^+,X_1)f_*'(X_1^+)f'(X_1)-g(X_1^-,X_1)f_*(X_1^-)f(X_1)\right]V_\alpha\theta(V_\alpha)\,d\Omega(\alpha)\,d\boldsymbol{\xi}_* \tag*{(1)}

where X1∈DX_1\in D, X1±=X1±σα1X_1^\pm=X_1\pm\sigma\alpha_1, α=(α1,α2,α3)\boldsymbol{\alpha}=(\alpha_1,\alpha_2,\alpha_3) is a unit vector, dΩ(α)d\Omega(\boldsymbol{\alpha}) is a solid angle element in the direction of α\boldsymbol{\alpha}, θ\theta is the Heaviside function, i.e., θ(x)=1\theta(x)=1 for x≥0x\geq0 and 00 otherwise, and the following notation convention has been used:

f(⋅)=f(⋅,ξ),f′(⋅)=f(⋅,ξ′),f∗(⋅)=f(⋅,ξ∗),f∗′(⋅)=f(⋅,ξ∗′),(2)f(\cdot)=f(\cdot,\boldsymbol{\xi}),\qquad f'(\cdot)=f(\cdot,\boldsymbol{\xi}'),\qquad f_*(\cdot)=f(\cdot,\boldsymbol{\xi}_*),\qquad f_*'(\cdot)=f(\cdot,\boldsymbol{\xi}_*'), \tag*{(2)}
ξ′=ξ+Vαα,ξ∗′=ξ∗−Vαα,Vα=V⋅α,V=ξ∗−ξ.(3)\boldsymbol{\xi}'=\boldsymbol{\xi}+V_\alpha\boldsymbol{\alpha},\qquad\boldsymbol{\xi}_*'=\boldsymbol{\xi}_*-V_\alpha\boldsymbol{\alpha},\qquad V_\alpha=\mathbf{V}\cdot\boldsymbol{\alpha},\qquad\mathbf{V}=\boldsymbol{\xi}_*-\boldsymbol{\xi}. \tag*{(3)}

Here and in what follows, the argument tt is often suppressed, unless confusion is anticipated. The convention (2) will be applied only to the quantities that depend on molecular velocity. The factor gg requires detailed explanation, which will be provided later in this section.

The diffuse reflection condition [10] in the present problem is written as

f(±L−σ2)=12π(RT±)(∫ξ∗1≷0∣ξ∗1∣f∗(±L−σ2) dξ∗)exp⁡(−ξ22RT±),ξ1≶0,t>0,(4)f\left(\pm\frac{L-\sigma}{2}\right)=\frac{1}{2\pi(RT_\pm)}\left(\int_{\xi_{*1}\gtrless0}|\xi_{*1}|f_*\left(\pm\frac{L-\sigma}{2}\right)\,d\boldsymbol{\xi}_*\right)\exp\left(-\frac{\boldsymbol{\xi}^2}{2RT_\pm}\right),\qquad\xi_1\lessgtr0,\quad t>0, \tag*{(4)}

where RR is the specific gas constant.

It should be noted that (1) makes sense only when the pair of positions (X1,X1±)(X_1,X_1^\pm) is in DD. Hence, gg should be considered as a function such that

g(X1,Y1)=g˚(X1,Y1)χD(X1)χD(Y1),(5)g(X_1,Y_1)=\mathring{g}(X_1,Y_1)\chi_D(X_1)\chi_D(Y_1), \tag*{(5)}

where χD(x)\chi_D(x) is the indicator function, i.e., χD(x)=1\chi_D(x)=1 for x∈Dx\in D and 0 otherwise, gg is positive and symmetric with respect to the exchange of positions: g(X1,Y1)=g(Y1,X1)g(X_1,Y_1)=g(Y_1,X_1). We call gg the Enskog factor in the sequel. Although some variants of gg have been proposed in the literature (e.g.,[1, 11, 12, 13]), we focus on two variants in the present paper: the original form by Enskog [1]

g(X1,Y1)=2S(4πσ33mρ(X1+Y12)),(6)g(X_1,Y_1)=2\mathcal{S}\left(\frac{4\pi\sigma^3}{3m}\rho\left(\frac{X_1+Y_1}{2}\right)\right), \tag*{(6)}

and the form proposed in [6]

g(X1,Y1)=S(R(X1))+S(R(Y1)),(7)g(X_1,Y_1)=\mathcal{S}(\mathcal{R}(X_1))+\mathcal{S}(\mathcal{R}(Y_1)), \tag*{(7)}

where

R(X1)=2πm∫D∫0∞ρ(Z1)θ(σ−r2+(Z1−X1)2)r dr dZ1,(8)\mathcal{R}(X_1)=\frac{2\pi}{m}\int_D\int_0^\infty\rho(Z_1)\theta\left(\sigma-\sqrt{r^2+(Z_1-X_1)^2}\right)r\,dr\,dZ_1, \tag*{(8)}

and ρ\rho is the mass density defined as

ρ=⟨f⟩,⟨∙⟩≡∫R3∙ dξ.(9)\rho=\langle f\rangle,\qquad\langle\mathord{\bullet}\rangle\equiv\int_{\mathbb{R}^3}\mathord{\bullet}\,d\xi. \tag*{(9)}

What we call OEE is the Enskog equation equipped with (6a), while what we call EESM is that equipped with (6b). Moreover, the form of S\mathcal{S} is related to the EoS under consideration as

p=ρRT(1+8ηS(8η)),η=πσ36mρ,(10)p=\rho RT\left(1+\frac{8}{\eta}\mathcal{S}(8\eta)\right),\qquad\eta=\frac{\pi\sigma^3}{6m}\rho, \tag*{(10)}

in the uniform equilibrium state, where pp is the thermodynamic static pressure and TT is the temperature defined by

T=13Rρ⟨(ξ−v)2f⟩,v=1ρ⟨ξf⟩.(11)T=\frac{1}{3R\rho}\left\langle(\xi-v)^2f\right\rangle,\qquad v=\frac{1}{\rho}\langle\xi f\rangle. \tag*{(11)}

In the above, η\eta is the volume fraction of molecules [2] and v=(v1,v2,v3)v=(v_1,v_2,v_3) is the flow velocity. Since the Carnahan–Starling EoS [9] is written as

pρRT=1+η+η2−η3(1−η)3=1+2η(2−η)(1−η)3,(12)\frac{p}{\rho RT}=\frac{1+\eta+\eta^2-\eta^3}{(1-\eta)^3}=1+\frac{2\eta(2-\eta)}{(1-\eta)^3}, \tag*{(12)}

the corresponding S\mathcal{S} is given as

S(x)=16(16−x)(8−x)3.(13)\mathcal{S}(x)=\frac{16(16-x)}{(8-x)^3}. \tag*{(13)}

In closing this section, we give the definition of the stress tensor pijp_{ij} and the heat flow vector qiq_i for later convenience. They are expressed as a sum of the kinetic and the collisional contribution:

pij=⟨cicjf⟩+σ22m∫S2×R3×R3∫0σαiαjVα2θ(Vα)G(X1,s,α1;ξ∗,ξ) ds dΩ(α) dξ∗ dξ,(14)p_{ij}=\langle c_i c_j f\rangle+\frac{\sigma^2}{2m}\int_{\mathbb{S}^2\times\mathbb{R}^3\times\mathbb{R}^3}\int_0^\sigma\alpha_i\alpha_j V_\alpha^2\theta(V_\alpha)G(X_1,s,\alpha_1;\xi_*,\xi)\,ds\,d\Omega(\alpha)\,d\xi_*\,d\xi, \tag*{(14)}
qi=12⟨cic2f⟩+σ24m∫S2×R3×R3∫0σαi(c+c∗)⋅αV2θ(Vα)×G(X1,s,α1;ξ∗,ξ) ds dΩ(α) dξ∗ dξ,(15)q_i = \frac{1}{2}\langle c_i c^2 f\rangle+ \frac{\sigma^2}{4m}\int_{\mathbb{S}^2 \times\mathbb{R}^3 \times\mathbb{R}^3}\int_0^\sigma\alpha_i(\mathbf{c}+\mathbf{c}_*)\cdot\boldsymbol{\alpha}V^2\theta(V\alpha) \times G(X_1,s,\alpha_1;\boldsymbol{\xi}_*,\boldsymbol{\xi})\,ds\,d\Omega(\boldsymbol{\alpha})\,d\boldsymbol{\xi}_*\,d\boldsymbol{\xi}, \tag*{(15)}

where the outer integration with respect to α\boldsymbol{\alpha}, ξ∗\boldsymbol{\xi}_*, and ξ\boldsymbol{\xi} is over their whole domain, i.e., S2×R3×R3\mathbb{S}^2 \times\mathbb{R}^3 \times\mathbb{R}^3, c=ξ−v\mathbf{c}=\boldsymbol{\xi}-\mathbf{v}, c∗=ξ∗−v\mathbf{c}_*=\boldsymbol{\xi}_*-\mathbf{v}, and

G(X1,s,α1;ξ∗,ξ)=g(X1+sα1,X1+(s−σ)α1)f∗(X1+(s−σ)α1)f(X1+sα1).(16)G(X_1,s,\alpha_1;\boldsymbol{\xi}_*,\boldsymbol{\xi})=g(X_1+s\alpha_1,X_1+(s-\sigma)\alpha_1)f_*(X_1+(s-\sigma)\alpha_1)f(X_1+s\alpha_1). \tag*{(16)}

Numerical analysis

Numerical method and related data overview

Numerical computations have been carried out for the dimensionless version of the problem (1)–(4) supplemented by the initial condition stated in the problem setting. Because of the sudden change of the plate temperature, there is a mismatch of the initial condition and the boundary condition at X1=(L−σ)/2X_1=(L-\sigma)/2 and it propagates into the gas as a discontinuity of the velocity distribution function. However, we have not made a special care of it in the present work, since the present work primarily aim at observing the quantitative difference of macroscopic quantities between OEE and EESM.

The numerical method adopted here is basically a combination of the second-order upwind (backward) finite-difference method for a spatial (time) derivative1 and the fast Fourier spectral method for the collision integral. The code is essentially the same as that was used in [14] improved from the original code for OEE developed in [8]. For further details, the reader is referred to these references. The idea of applying the fast Fourier spectral method to the collision integral can be found in [15] and the references therein. Its application to the Enskog equation was probably first done in [5]. Although numerical convergence has been checked with several patterns of grid system, the results to be presented below are all carried out with 480 nonuniform grid intervals (varying from 1.1×10−51.1 \times10^{-5} to 3.3×10−33.3 \times10^{-3} multiplied by 1−σ/L1-\sigma/L) in X1/LX_1/L with finer grid intervals near the boundaries, 512 uniform grid intervals in ξ1/2RT−\xi_1/\sqrt{2RT_-}, and 32 uniform grid intervals both in ξ2/2RT−\xi_2/\sqrt{2RT_-} and ξ3/2RT−\xi_3/\sqrt{2RT_-} directions in the finite cubic domain [−8,8]3[-8,8]^3. The time discretization is uniform with the interval Δt=1.0×10−3t0\Delta t=1.0 \times10^{-3}t_0 with t0=L/2RT−t_0=L/\sqrt{2RT_-}.

Results and discussions

The present problem is characterized by three parameters: T+/T−T_+/T_-, σ/L\sigma/L, and η0=(πσ3/6)(ρ0/m)\eta_0=(\pi\sigma^3/6)(\rho_0/m), where ρ0=∫Dρ dX1/(L−σ)\rho_0=\int_D\rho\,dX_1/(L-\sigma). The Knudsen number Kn is not an independent parameter, since it is expressed in terms of σ/L\sigma/L and η0\eta_0 as

Kn⁡≡ℓ0L=σ/L122η0S(8η0),ℓ0=122πσ2(ρ0/m)S(8η0),(17)\operatorname{Kn} \equiv\frac{\ell_0}{L} = \frac{\sigma/L}{12\sqrt{2}\eta_0 S(8\eta_0)}, \qquad\ell_0 = \frac{1}{2\sqrt{2}\pi\sigma^2(\rho_0/m)S(8\eta_0)}, \tag*{(17)}

where ℓ0\ell_0 is the reference mean free path of a molecule.

The time evolution of the temperature TT and the density ρ\rho at the final steady state are shown in the left column of Figure 2 for three different pairs of η0\eta_0 and σ/L\sigma/L in the case of T+/T−=1.1T_+/T_- = 1.1. As is seen, the results based on OEE and EESM are almost identical over the time evolution process. The same applies to the time evolution of the heat flow q1q_1, which is also shown in the right column of Figure 2. These agreements between OEE and EESM are observed even for higher volume fraction in our computations.

Figure 2. Time evolution of the temperature TT and the density ρ\rho at the final steady state (left column) and of the heat flow q1q_1 (right column) in the case of T+/T−=1.1T_+/T_- = 1.1. (a) σ/L=0.05\sigma/L = 0.05 and η0=0.01\eta_0 = 0.01 (Kn = 0.5746). (b) σ/L=0.05\sigma/L = 0.05, η0=0.2\eta_0 = 0.2 (Kn = 0.0168). (c) σ/L=0.1\sigma/L = 0.1, η0=0.2\eta_0 = 0.2 (Kn = 0.0335). Solid lines: EESM. Dashed lines: OEE.

However, this is not necessarily the case for the stress tensor p11p_{11} as shown in Figure 3. A discernible difference can be observed in the time evolution for highly confined cases with large volume fraction, as shown in Figure 3(c). In reality, the variation of p11p_{11} is much smaller than the variations of its kinetic and collisional parts near the boundary. The large variations of these two parts mostly cancel each other out, resulting in much smaller variation of p11p_{11}. Accordingly, the discernible difference observed in p11p_{11} is not truly significant relative to the scale of variation of those respective parts. The dependence of this difference on Kn and σ/L\sigma/L (or η0\eta_{0}) is evaluated based on the constant value achieved at the final steady state, and the results are shown in Figure 4. The difference is proportional to Kn−2\mathrm{Kn}^{-2} (solid lines in Figure 4) and increases with an increase in σ/L\sigma/L and/or η0\eta_{0}. It is a little curious that the difference also appears in the regime where both σ/L\sigma/L and Kn are small, since in this regime EESM and OEE are expected to give the identical results up to the Navier–Stokes level [16] according to the Chapman–Enskog (CE) expansion. This is likely due to the deviation near the boundary, where the CE expansion does not apply, shifting the entire profile between the plates. The difference is, in general, larger for larger σ/L\sigma/L and larger η0\eta_{0}. However, it is also observed that, if η0\eta_{0} is fixed, the difference in p11p_{11} decreases as Kn decreases, see the dashed lines in Figure 4. This implies that the deviation near the boundary also decreases if Kn and σ/L\sigma/L simultaneously decrease with their ratio fixed, namely under the parameter setting consistent with the CE expansion (see [16]). The decreasing rate is, however, a little worse than O(Kn)O(\mathrm{Kn}), say O(Kn0.7)−O(Kn0.95)O(\mathrm{Kn}^{0.7})-O(\mathrm{Kn}^{0.95}).

Figure 3. Time evolution of the stress tensor p11p_{11} in the case of T+/T−=1.1T_{+}/T_{-}=1.1. (a) σ/L=0.05\sigma/L=0.05, η0=0.01\eta_{0}=0.01 (Kn = 0.5746). (b) σ/L=0.05\sigma/L=0.05, η0=0.05\eta_{0}=0.05 (Kn = 0.1036). (c) σ/L=0.05\sigma/L=0.05, η0=0.15\eta_{0}=0.15 (Kn = 0.0261). (d) σ/L=0.03\sigma/L=0.03, η0=0.15\eta_{0}=0.15 (Kn = 0.0156). Solid lines: EESM. Dashed lines: OEE.

Figure 4. The difference of stress tensor Δp11=p11OEE−p11EESM\Delta p_{11}=p_{11}^{\mathrm{OEE}}-p_{11}^{\mathrm{EESM}} between OEE and EESM at the steady state in the case of T+/T−=1.1T_{+}/T_{-}=1.1.

We have also made comparisons with preliminary molecular dynamics results [17] at the steady states. As a whole, both the EESM and OEE results agree reasonably with the MD results. For example, the relative error in p11p_{11} (q1q_{1}) between EESM and MD is less than 1.3% (0.7%) in the same cases as in Figure 5 that represent rather hard physical settings. The profile with dimples in density near the boundary is closer between EESM and MD than between OEE and MD. The details are, however, omitted here.

Figure 5. Temperature and density profiles at the final steady state in the case of T+/T−=2T_{+}/T_{-}=2 and η0=0.2\eta_{0}=0.2. (a) σ/L=0.05\sigma/L=0.05 (Kn =0.0168=0.0168). (b) σ/L=0.1\sigma/L=0.1 (Kn =0.0335=0.0335). Solid lines: EESM. Dashed lines: OEE.

In closing, the density and temperature at the final steady state for T+/T−=2T_{+}/T_{-}=2 and η0=0.2\eta_{0}=0.2 are shown in Figure 5 as an example of comparisons for extremely severe situations. It is seen that, in contrast to the rarefied gas, the density profile does not change inversely with temperature, which is considered a characteristic of a non-ideal gas. However, it should be also noted that the density at the boundary is inversely proportional to the temperature at the boundary. This is a manifestation of the ideal gas behavior there, since the collisional contribution to the pressure is suppressed at the boundary. These features also hold true for the results already shown in Figure 2.

Conclusion

In the present work, we have numerically studied the behavior of a gas between two parallel plates, initially in thermal equilibrium with the plates at a common temperature, induced by a sudden temperature change of one plate. The analyses have been carried out on the basis of both the OEE and the EESM, with a special interest in the influence of modifying the Enskog factor. Overall, the two variants show reasonable agreement, except for rather extreme choices of parameters corresponding to a high volume fraction and high confinement. The primary difference lies in the depth of the first few density troughs away from the boundary.

Acknowledgments. This work is partly supported by Kyoto University Foundation, by JSPS Grant-in-Aid for Scientific Research(B) (No. 26K00870), and by HPCI system Research Projects (Nos. hp250004 and hp260026). A part of computations was conducted on SQUID at D3 Center, The University of Osaka.

References

  1. [1]Enskog, D.: in Kinetic Theory, Vol. 3, S. G. Brush ed., Pergamon Press, Oxford, Part 2, 1972, pp. 226–259.
  2. [2]Soto, R.: Kinetic Theory and Transport Phenomena, OUP, New York, 2016.DOI
  3. [3]Frezzotti, A.: Phys. Fluids 9, 1329–1335 (1997).
  4. [4]Montanero, J. M. and Santos, A.: Phys. Fluids 9, 2057–2060 (1997).
  5. [5]Wu, L., Zhang, Y., and Reese, J. M.: J. Comput. Phys. 303, 66–79 (2015).
  6. [6]Takata, S. and Takahashi, A.: Phys. Rev. E 111, 065108 (2025).
  7. [7]Frezzotti, A.: Eur. J. Mech. B/Fluids 18, 103–119 (1999).
  8. [8]Hattori, M., Tanaka, S., and Takata, S.: AIP Advances 12, 055220 (2022).
  9. [9]Carnahan, N. F. and Starling, K. E.: J. Chem. Phys. 51, 635–636 (1969).
  10. [10]Sone, Y.: Molecular Gas Dynamics, Birkhäuser, Boston, 2007; Supplemental notes and errata are available from http://hdl.handle.net/2433/66098.
  11. [11]van Beijeren, H. and Ernst, M. H.: Physica 68, 437–456 (1973).
  12. [12]Benilov, E. S. and Benilov, M. S.: Phys. Rev. E 97, 062115 (2018).
  13. [13]Benilov, E. S. and Benilov, M. S.: J. Stat. Mech. 2019, 103205 (2019).
  14. [14]Takata, S., Sakata, S., Takahashi, A. and Hattori, M.: arXiv:2603.23839.
  15. [15]Filbet, F., Mouhot, C., and Pareschi, L.: SIAM J. Sci. Comput. 28, 1029–1053 (2006).
  16. [16]Takahashi, A. and Takata, S.: Bull. Inst. Math., Acad. Sin. (N.S.) 21, 217–253 (2026).
  17. [17]Kubota, N.: private communication.

Paper details

vlnc:W_ufmfxqdr
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