Does Maxwell's hypothesis of air saturation near the surface of evaporating liquid hold at all spatial scales?

Abstract The classical model of evaporation of liquids hinges on Maxwell's assumption that the air near the liquid's surface is saturated. It allows one to find the evaporative flux without considering the interface separating liquid and air. Maxwell's hypothesis is based on an implicit assumption that the vapour-emission capacity of the interface exceeds the throughput of air (i.e. its ability to pass the vapour on to infinity). If this is indeed so, then the air adjacent to the liquid would get quickly saturated, justifying Maxwell's hypothesis. In the present paper, the so-called diffuse-interface model is used to account for the interfacial physics and thus derive a generalised version of Maxwell's boundary condition for the near-interface vapour density. It is then applied to a spherical drop floating in air. It turns out that the vapour-emission capacity of the interface exceeds the throughput of air only if the drop's radius is $r_{d}\gtrsim 10\ \mathrm {\mu } \mathrm {m}$, but for $r_{d}\approx 2\ \mathrm {\mu } {\rm m}$, the two are comparable. For $r_{d} \lesssim 1\ \mathrm {\mu } {\rm m}$, evaporation is interface-driven, and the resulting evaporation rate is noticeably smaller than that predicted by the classical model.


Introduction
Consider a liquid drop floating in the air and evaporating.In its simplest formulation, this problem was examined more than 140 years ago by Maxwell (1877Maxwell ( , 1890)), and its more elaborate versions are still explored now.Several of these are relevant to the present study: Sobac et al. (2015), Cossali & Tonini (2019) and Finneran (2021) argued that the dependence of the fluid's properties on the temperature can affect the evaporation rate strongly; Talbot et al. (2016) examined transient deviations of drop evaporation from the quasi-steady regime; Finneran, Garner & Nadal (2021) showed that the transient effects can cause a noticeable departure from the so-called 'd 2 law' (stating that the time of a drop's evaporation is proportional to its diameter squared); Long, Micci & Wong (1996), Landry et al. (2007), Jakubczyk et al. (2012), Hołyst et al. (2013), Hołyst, Litniewski & Jakubczyk (2017), Rana, Lockerby & Sprittles (2018, 2019) and Zhao & Nadal (2023) showed that the 'd 2 law' can be violated by out-of-equilibrium gas kinetics.A more detailed review of the recent literature on evaporating drops can be found in Sazhin (2017).
Most work on evaporation has been based on Maxwell's hypothesis that the air near the drop's surface -or, more generally, near any liquid/air interface -is close to saturation.This allows one to solve the problem without knowing anything about the interfacial dynamics.The only exceptions are models accounting for the so-called Knudsen layera layer where the out-of-equilibrium kinetics plays a part (e.g.Rana et al. 2018Rana et al. , 2019;;Zhao & Nadal 2023).
The physics behind Maxwell's hypothesis becomes clear if evaporation is viewed as a two-tier process: first, the interface emits molecules of the liquid into the air; second, the air passes them on to infinity.If the former emits more vapour than the latter can pass on, then the vapour accumulates near the drop -bringing the adjacent air layer close to saturation and the interface close to thermodynamic equilibrium.As a result, the interface reduces vapour emission in line with the air throughput.In terms of this mechanism, Maxwell's hypothesis amounts to an assumption that the vapour-emission capacity of the interface exceeds the throughput of air.
The diffuse-interface model (DIM) is based on two assumptions regarding the van der Waals force.
(i) Van der Waals interactions are pairwise -hence the net force affecting a molecule is an algebraic sum of the forces exerted by the other molecules.(ii) The thickness of the interface is much greater than the spatial scale of van der Waals interactions.
Admittedly, assumption (ii) does not hold well at room temperature, in which case the interfacial thickness rarely exceeds the molecular size by an order of magnitude (e.g.Liu, Harder & Berne 2005;Verde, Bolhuis & Campen 2012;Pezzotti, Galimberti & Gaigeot 2017;Pezzotti, Serva & Gaigeot 2018;Dodia et al. 2019;and references therein).Thus the DIM should be viewed as a means to estimate qualitatively the importance of interfacial dynamics for evaporation, whereas quantitative predictions should be left to the Enskog-Vlasov kinetic equation (which can handle thin interfaces).These two models are to each other what the first term of a Taylor series is to the exact value of the function.
In the present paper, the DIM is used to examine evaporation of liquids into air under 'normal conditions', i.e. when the temperature is between 15 • C and 35 • C, and the air pressure is at 1 atm.A generalised Maxwell boundary condition is derived for this case; to facilitate its use without going into the detail of the derivation, this condition is summarised here.
The following notation will be used: ρ 1 is the density of the liquid or its vapour, ρ 2 is that of air.Outside the drop, the air is almost homogeneous, i.e.
be the saturated vapour density at a temperature T, and let ρ (l.sat) 1 (T) be the matching liquid density.Introduce also the vapour-in-the-air diffusivity D, the shear and bulk viscosities of air, μ s and μ b , respectively, and the specific gas constant R i of the liquid or its vapour (i = 1), or the air (i = 2).The interface as a whole is characterised by its curvature C and surface tension σ .
All these parameters are macroscopic, i.e. they characterise the fluid as a continuum, whereas the Korteweg matrix K ij is a microscopic characteristic.It describes the intermolecular (van der Waals) force due to the liquid-liquid (K 11 ), air-air (K 22 ) and liquid-air (K 12 = K 21 ) interactions.Macroscopically, K ij determine the capillary properties of the fluid and thus can be deduced from measurements of its surface tension (see Appendix B).The numerical values of K ij determined this way for water and air are presented later, in (3.12).
Let ρ (+)   1 be the macroscopic vapour density immediately outside the interface, and let (n • ∇ρ 1 ) (+) be its normal derivative (where n is directed towards the air).Then the generalised Maxwell boundary condition has the form where ξ is the relative vapour flux, and the parameter will be referred to as the 'vapour-emission capacity' of the interface.Physically, E 0 is determined by a narrow zone at the outskirts of the interface, driven by the van der Waals force, pressure gradient and viscosity.The location of this zone suggests that it is a DIM analogue of the Knudsen layer, i.e. the Knudsen layer plus the van der Waals force -which is an order-one effect in the DIM, but is neglected by the standard Hertz-Knudsen kinetic model.
The universal non-dimensional function A(ξ ) that appears in (1.1) is also a characteristic of the above-mentioned zone at the outskirts of the interface.It can be approximated by the expression obtained through a combination of asymptotics, numerics and curve fitting (see Appendix C).
The term involving the curvature C in condition (1.1) describes the Kelvin effect, i.e. the shift of the saturated vapour pressure near a curved surface (e.g.Eggers & Pismen 2010;Colinet & Rednikov 2011;Rednikov & Colinet 2013, 2017, 2019;Morris 2014;Janeček et al. 2015;Saxton et al. 2017).If the interface is flat (C = 0) and its vapour-emission capacity exceeds the diffusive flux, then (1.1) reduces to the classical Maxwell boundary condition.This paper has the following structure.In § 2, Maxwell's boundary condition is reviewed briefly in application to evaporation of floating drops.Its generalised version is derived in § 3, and applied to drops in § 4. In § 5, the proposed model is compared to the Hertz-Knudsen kinetic model, and further applications of the former are outlined.

Formulation
Consider a mixture of two fluids with partial densities ρ i , where i = 1, 2. The first fluid will be referred to as either 'liquid' or 'vapour' (depending on the phase it is in), and the second fluid will be referred to as 'air'.
Under the assumption of spherical symmetry, ρ i generally vary with the radial coordinate r and time t.In the classical model, however, one assumes that inside the drop where 1 is a known constant, and the drop's radius r d depends on t.Outside the drop, both densities do vary.The air density is much larger than that of the vapour, but their variations are of the same order (as, physically, the pressure should be spatially uniform).As a result, the variations of ρ 2 are relatively weak, so one assumes where the dry-air density ρ (a) 2 is a known parameter.Note that, physically, ρ (l) 2 are inter-related by the requirement that the pressure difference between the inside and outside of the drop equals the capillary correction due to the curvature of the interface.
The classical model of evaporation also involves the liquid's thermal conductivity κ (l) , air's thermal conductivity κ (a) , saturated vapour density ρ (v.sat) 1 , vaporisation heat Δ, and 971 A20-4 vapour-in-the-air diffusivity D. The dependence of these parameters on the temperature T is supposed to be known.
The classical model is usually combined with the quasi-steady approximation, based on an assumption that the diffusivity and thermal conductivity are sufficiently large (which they are indeed for water under normal conditions).Then the term ∂ρ 1 /∂t in the diffusion equation and the term ∂T/∂t in the heat-conduction equation can both be neglected.The simplest version of the classical model also neglects the Stefan flow (which carries the vapour away from the surface of the liquid), the Soret effects (the mass flux due to the temperature gradient), and the Dufour effect (the heat flux due to the density gradient).
Then, for the spherically symmetric case, one obtains (2.4) At the drop's centre, the usual boundary condition applies: To formulate the boundary conditions at the drop's surface, introduce (2.6a,b) where (+) and (−) mean 'just outside' and 'just inside' the drop, respectively.Then the continuity of the temperature and heat flux imply − κ (a) ∂T ∂r (2.8) where ṙd = dr d dt . (2.9) As discussed in the Introduction, the classical model assumes the Maxwell boundary condition for the vapour density at the drop's surface, i.e. (2.10) and the following conditions should be imposed on the vapour-air mixture far away from the drop: 1 , T → T (a) as r → ∞. (2.11) To ensure that the liquid is evaporating -not condensating or being in equilibrium -let the vapour at infinity be undersaturated: , vapour-in-the-air diffusivity D, thermal conductivity κ (a) of the air, and vaporisation heat Δ (all non-dimensionalised on their reference (a) (25 • C), and (4) Δ(T)/Δ(25 • C).Curves (1)-( 4) correspond to the empirical formulae of Wagner & Pruss (1993), Engineering ToolBox (2018), White (2005) and Lindstrom & Mallard (1997), respectively.

The dependence of ρ (v.sat) 1 on T
One's experience with raindrops sitting on ones's skin suggests that evaporative cooling does not exceed several degrees.This seems to enable one to neglect the dependence of the external parameters on the temperature -say, 'freeze' them at T = T (a) .Such an approximation is indeed applicable to all parameters but one -namely, the saturated vapour density ρ (v.sat) 1 .This claim is illustrated for water in figure 1.One can see that the relative change of ρ (v.sat)   1 across the room temperature range exceeds those of the remaining parameters.In this paper, the dependence ρ (v.sat)   1 (T) is approximated by Antoine's equation (Antoine 1888).If written in terms of the density -as opposed to the pressure, as in the original formulation -it takes the form where, for water, ρ A = 6.6326 × 10 4 kg m −3 , T A = 5292 K. (2.15a,b) These values have been obtained by fitting (2.14) to the high-accuracy IAPWS empirical formula (Wagner & Pruss 1993) for the room temperature range.The relative deviation of the two formulae is less than 0.23 %, which is much better than the accuracy of Antoine's equation with ρ A and T A taken from a thermodynamics handbook.Not only does Antoine's equation provide an accurate approximation, but it also has physical meaning.As shown by Benilov (2020), (2.14) describes the generic van der Waals fluid in the low-temperature limit -i.e. when T is much closer to the freezing point than to the critical point.Observe also that T A is large, which is what makes the dependence of ρ (v.sat) 1 on T strong.Equation (2.14) can be rearranged conveniently using a reference temperature (as done by Sobac et al. 2015).Choosing that to be T (a) , one obtains a) , (2.16) where is the saturated vapour density at infinity.One can further assume (T (a) − T)/T (a)  1 and reduce (2.16) to a) ) . (2.17) Observe that even if T is close to T (a) , the expression in the square brackets can still be order-one (because T A T (a) ).
and (2.13) for ṙd , one obtains where is the relative humidity of air far away from the drop.Ordinary differential equation (2.21) determines r d (t); the terms on its left-hand side are labelled according to their physical meanings.
Equation (2.21) yields the following expression for a drop's time of evaporation: where r d (0) is the drop's initial radius, and x is the solution of the non-dimensional algebraic equation (2.25) Formula (2.23) is often referred to as the 'd 2 law', where d is the drop's diameter.
(2.28) 2.4.Discussion It is interesting to assess the importance of non-isothermal effects by comparing the above model with its isothermal reduction.It can be verified readily that the latter amounts to taking in (2.24) the limit α → 0, so that x = 1 − H. Denoting the isothermal time of evaporation by t e , one obtains It turns out that the isothermal approach noticeably underpredicts the evaporation time.The largest discrepancy is observed at high humidity, in which case (2.24) can be solved asymptotically.Assuming estimate (2.28) for α, one obtains (2.30) The smallest discrepancy, in turn, is observed at low humidity.Solving (2.24) with H = 0 numerically, one obtains t e t e ≈ 2.95 for H = 0. (2.31) Evidently, the isothermal model may yield a significant error in the whole range of H, which agrees broadly with conclusions of Sobac et al. (2015), Cossali & Tonini (2019) and Finneran (2021).
It is also interesting to calculate the time of evaporation of a typical 'cough drop' (through which COVID-19, flu, etc. are spread).The radii of such drops are typically between 0.5 and 16.0 µm (Yang et al. 2007).The largest drops from this range survive the longest -hence are the most dangerous when spreading the disease.The dependence t e on T for this case is illustrated in figure 2.

The generalised Maxwell boundary condition as described by the diffuse-interface model
Unlike the classical model where the drop's boundary has zero thickness, the DIM assumes the liquid and air to be separated by a finitely thin interface where the partial densities ρ i vary smoothly.In this section, the DIM will be used to derive a model of evaporation that takes into account interfacial physics.Mathematically, the interface should be viewed as an inner asymptotic region separating the outer regions of liquid and air.The DIM describes them all, but it can be simplified depending on the region to which it is applied.
First, one can take advantage of the fact that the temperature field associated with evaporative cooling occurs at a macroscopic scale (e.g. the drop's radius).The interface, on the other hand, is microscopichence the temperature change in it is negligible, and it can be examined via the isothermal version of the DIM.
One should still account for the temperature dependence of the saturated vapour density (which was found to be important in the classical model) -but there is no need to use the full non-isothermal equations just for this.Instead, one can use the classical (macroscopic) model to calculate the temperature field and then use it to prescribe the correct value of ρ (v.sat)   1 (T) near the drop.This workaround allows one to avoid cumbersome calculations associated with non-isothermality, yet obtain the same result.
One can also consider first a flat interface.Once the generalised Maxwell boundary condition has been derived for this case, the pressure difference due to the capillary effect can be incorporated readily into it.

The governing equations
Let z be the vertical Cartesian coordinate, and let w be the vertical velocity of the mixture as a whole.Introduce also the pressure p, and keep in mind that it is not an independent unknown: since the DIM assumes the fluid to be compressible, p is a known function of ρ 1 , ρ 2 and T (the equation of state).The shear and bulk viscosities -μ s and μ b , respectivelyare also known functions of the same arguments.
Introduce the partial chemical potentials G 1 and G 2 .Their thermodynamic definition is given in Appendix A.1 -but technically, one can simply treat them as given functions of ρ 1 , ρ 2 and T, such that In the low-density limit, p and G i tend to their ideal gas approximations: where R i are the specific gas constants.For water and air, they are with R 2 calculated as the 70/30 weighted average of the gas constants of nitrogen and oxygen.
The DIM comprises the standard hydrodynamic equations involving an undetermined external force F i , which will be identified later with the van der Waals force.Under the slow-flow (Stokes) approximation, a compressible isothermal binary fluid is governed by where the effective viscosity is the diffusive flux is (e.g.Giovangigli 2021) and D is the diffusion coefficient.
To understand the relationship of D with the diffusivity D from the classical model, assume that the fluid is diluted, so that the van der Waals force is negligible (F i = 0) and the chemical potentials are given by (3.2).As a result, (3.7) reduces to Let the vapour density be much smaller than that of the air (ρ 1 ρ 2 ), but let their variations be comparable (∂ρ 1 /∂z ∼ ∂ρ 2 /∂z) -so that the above expression reduces to A comparison of this expression with that for the classical diffusive flux yields For a dense mixture of fluids with comparable densities, D cannot be deduced from D, but its dependence on (ρ 1 , ρ 2 , T) is supposed to be known from measurements.For a multicomponent flow, the (one-dimensional) DIM approximation of the van der Waals force is (Benilov 2023a) (3.11)where the Korteweg matrix K ij characterises the interaction of molecules of components i and j (Newton's third law implies K ij = K ji ).For the water-air combination, the Korteweg matrix is (see Appendix B) Next, the pressure term in the momentum equation (3.5) can be rewritten conveniently in terms of the chemical potentials.Using identity (3.1), one obtains This equation and (3.7) can be viewed as a set of algebraic equations for F 1 − ∂G 1 /∂z and F 2 − ∂G 2 /∂z.Solving these equations and recalling (3.11) for F i , one obtains Equations (3.4) and (3.14)-(3.15)determine the unknowns ρ 1 , ρ 2 , w and J.

Non-dimensionalisation
In what follows, the vertical coordinate z will be non-dimensionalised on the interfacial thickness z, which implies that distances should be measured from the current height Z(t) of the interface.Mathematically, this corresponds to the change of variables (z, t) → (z new , t new ), where Rewriting (3.4) in terms of the new variables, introducing Ż = dZ/dt, and omitting the subscript new , one obtains whereas (3.14)-(3.15)remain the same as before.
There are three density scales in the problem -the liquid density, the air density and the saturated vapour density.Since they differ from each other by orders of magnitude, no single scaling of ρ i would fit all the asymptotic regions.Still, a 'generic' non-dimensionalisation is needed, so let the density scale be that of air, ρ = ρ respectively.These choices for p and w correspond to an asymptotic regime where the pressure gradient, viscous stress and van der Waals force in the momentum equation (3.5) are of the same order.Let t be the evaporation time scale, and introduce This parameter is responsible for the quasi-steady approximation -hence it is expected to be small.Another important parameter is the advection/diffusion ratio, defined by where D is the characteristic value of the diffusion coefficient D.
The following non-dimensional variables will be used: Rewriting (3.14)-(3.17) in terms of the non-dimensional variables, and omitting the subscript nd , one obtains Evidently, the non-dimensional equations (3.22)-(3.24)can be transformed back to their dimensional versions, (3.14)-(3.17),by setting ε = δ = 1.This shortcut will be used for re-dimensionalising the results obtained.
In what follows, one will need the non-dimensional versions of the low-density expressions (3.2) for p and G i , and (3.10) for D. To obtain these, introduce where R is the characteristic gas constant.Rewriting (3.2) and (3.10) in terms of T nd and D nd , and omitting the subscript nd , one can verify that the dimensional and non-dimensional versions of these expressions look exactly the same.

Non-dimensional parameters
Let the characteristic value of the Korteweg parameter be that of dry air, K = K 22 , given by (3.12).The characteristic values of the rest of the parameters are given by (2.26)-(2.27)and (3.3).
For common fluids, the room temperature range is much closer to the freezing point than the critical point (e.g. for water, the former is 0 • C and the latter is 374 • C).In such cases, the interfacial thickness z is comparable to the liquid's molecular size -i.e.several ångströms -as confirmed by both experiments (e.g.Verde et al. 2012;Pezzotti et al. 2017) and molecular simulations (e.g.Liu et al. 2005;Pezzotti et al. 2017;Dodia et al. 2019;and references therein).
To obtain an estimate of z from within the DIM, one can use the particular case of an equilibrium interface.It is examined briefly in Appendix A.2 for the parameters specific to the water-air combination listed in Appendix B. The resulting profiles of the water and air densities are shown in figure 3, suggesting that the interfacial thickness is z ∼ 5Å. (3.26) This value exceeds the water's molecular size by a factor of 2-3 (depending on how it is defined).
According to (3.19), ε depends on the time scale t -which, in turn, depends on the evaporation rate -hence ε can be estimated only in application to a specific setting.971 A20-13 (3.31)

E.S. Benilov
In addition to ε and δ, the problem involves two 'hidden' small parameters: The presence of four independent small parameters makes the analysis awkward: they arise in various combinations, and each time, one has to check whether these are large, small or order-one.In particular, the following estimates will be needed: Thus all of the above parameter combinations will be assumed small.

The boundary/matching conditions
When applied to the interface (inner region), (3.22)-(3.24)need boundary conditions at large z.These are to be obtained via matching the inner solution to those in the liquid and air (outer regions).The former is trivial: since the liquid is homogeneous and at rest, the interfacial solution should satisfy Unlike the classical model where ρ (l)  i are external parameters, the DIM determines them as functions of the pressure and humidity of air.The DIM also describes dissolution of air in liquid -hence ρ (l) 2 / = 0.This quantity is very small and thus unimportant both dynamically and thermodynamically.
The matching between the interface and air is less trivial: it can be carried out only after examining the large-distance behaviours of the inner solution and choosing the one that matches the outer solution.The properties of the latter are easy to guess: it should be such that ρ 1 ρ 2 (small vapour concentration in the air), and the variations of ρ 1 and ρ 2 should be such that the total pressure is spatially uniform (e.g.Ricci & Rocca 1984;Mills & Coimbra 2016).
To verify that (3.22)-(3.24)admit such an asymptotic solution, let , etc. depend on t (but not on z) and may involve the hidden small parameters (a/l) and (v/a) .It turns out that only ρ (+)   i is independent in these expansions, with the other coefficients being expressible through ρ (+) i .To find how w (+) and J (+) depend on ρ (+) i , observe that to leading order, (3.24) can be integrated.Using boundary conditions (3.35)-(3.36) to fix the constants of integration, 971 A20-15 one obtains where η 0 depends only on T. This property implies that when expressions (3.38) are substituted into (3.22)-(3.23), the linear term in the expansion for w cancels out.The quadratic term does not, but its contribution is much smaller than that of J (0) (due to estimates (3.33)), and to leading order, one obtains the following algebraic equations for ρ where 2 , T) and Given the smallness of the air-to-liquid and vapour-to-air density ratios (3.32), (3.41)-(3.42)can be simplified.Using the diluted-fluid expressions (3.2) and (3.10) for G i and D, respectively, and keeping the leading order only, one obtains Equality (3.44) is the desired matching condition; it can also be viewed as the relationship of the speed Ż of the interface to the vapour flux Dρ (+) , whereas (3.45) is the standard condition of constant pressure for diffusion in diluted fluids.As mentioned before, both conditions are obvious physically, but it is still important to verify that they are intrinsic to the DIM formalism.
Note that Zone 1 is not part of the air region: in the latter, the air density is nearly uniform, 2 , whereas in the former, variations of ρ 2 are comparable to ρ (a) 2 .A schematic illustrating the asymptotic structure of the problem can be found in figure 3 (which is computed for the equilibrium interface, i.e. that where the relative humidity of air is H = 1).Its most counter-intuitive features are the local maximum and minimum of the air density ρ 2 (z) in Zone 2. Note that they both disappear in the limit K 12 → 0 (see figure 11 of Benilov 2023a).One can thus assume that the maximum of ρ 2 is created by the van der Waals force exerted by the bulk of the liquid by attracting a certain amount of air towards the interface.The existence of the minimum of ρ 2 is harder to understand, but it is dynamically unimportant anyway.The air density is extremely small there -even smaller than that of the air dissolved in the liquid -and they both have a minuscule effect on the dynamics of the rest of the fluid.
It turns out that only Zone 1 contributes significantly to the evaporation rate, whereas the contributions of Zones 2-3 are negligible.To understand why, recall that the fluid density in the latter two zones exceeds that of the vapour component of air by an order of magnitude.As a result, Zones 2-3 are close to equilibrium (the fact that the 'outside' air is undersaturated cannot perturb them too much).Thus Zone 1 is the only region out of equilibrium -hence it is the only place where irreversible processes, such as evaporation, can occur.
The calculations associated with Zones 2-3 can be avoided via a certain workaround developed by Benilov (2022bBenilov ( , 2023b) ) for evaporation of pure fluids.The workaround involves the following two steps.
(i) Equations (3.22)-(3.24)can be rearranged in such a way that the velocity Ż of the interface is expressed in terms of the fluid's known parameters (relative humidity, etc.) and a certain integral of the solution.(ii) The contribution of Zone 1 to this integral exceeds the contributions of Zones 2-3 by an order of magnitude.Thus to evaluate this integral asymptotically, it is sufficient to find the solution only in Zone 1.
where Thus the pressure in the liquid coincides with that in the air (as expected physically).Equalities (3.50) and (3.53) relate the (unknown) liquid densities ρ (l)  i and the velocity Ż of the interface to the air parameters (marked with (+) ).In addition, Ż is also related to the air parameters by equality (3.44).If two of these three equalities are used to eliminate ρ (l)  i and Ż, then the remaining one will inter-relate the air parameters and thus turn into the generalised Maxwell boundary condition (GMBC).
To eliminate ρ (l)  i from equalities (3.50) and (3.53), recall that equations of state of liquids at normal conditions are typically such that even a minuscule change of the density gives rise to a huge change of the pressure.This means effectively that ρ (l) (3.54) In the state of saturation, the chemical potentials of the liquid and vapour are equal (see Appendix A.2) -hence one can change in the above equality G (l.sat) This is, essentially, the GMBC.With A calculated (see the next subsubsection), equality (3.55) inter-relates the near-interface vapour density ρ (+) 1 and its gradient ρ (+) , which is the classical Maxwell boundary condition.If, however, ε is small but not zero, then ρ (+) 1 can differ significantly from ρ (v.sat) 1 -because A can actually be large, so that εA ∼ 1.This does not violate the employed asymptotic approach: it requires only ε 1, which makes the time derivatives in (3.24) small and thus justifies the quasi-steady approximation.

Zone 1 of the interfacial asymptotic region
The coefficient A arises in all problems where the DIM is applied to evaporation or condensation -but, so far, it has been calculated only for pure fluids and the case where the relative humidity is close to unity (Benilov 2020(Benilov , 2022a(Benilov ,b, 2023b)).The calculation of A for a binary mixture and arbitrary humidity is more complex technically, but the underlying idea is the same.
It is based on the observation that the main contribution to (3.51) comes from the zone where ρ 1 + ρ 2 is small, but its derivatives are relatively large -i.e.near the air region, but not in it.This means effectively that (3.51) can be calculated asymptotically by considering Zone 1 only (see its definition at the beginning of § 3.5).
Before presenting the scaling of Zone 1, it is instructive to apply the 'obvious' approximations -i.e.take advantage of the diluted-fluid expressions (3.2), (3.10) and (3.40), plus neglect ρ 1 and ρ (l)  2 if they appear next to ρ 2 and ρ (l) 1 , respectively.As a result, (3.47)-(3.48)and (3.51) become (3.58) The most general asymptotic limit is when the three terms in (3.57) are of the same order, which implies the rescaling where z 0 is the 'location' of Zone 1, E 0 is given by (1.3) of the Introduction, and the minus in the expression for Ż is introduced to make ξ positive (recall that evaporation corresponds to Ż < 0).Since the problem under consideration is translationally invariant, the actual value of z 0 does not feature in, and is not needed for, any leading-order results.
One does not need (3.56) either, because now A depends only on ρ 2 .
As shown in Appendix C, boundary condition (3.62) is sufficient to uniquely fix the solution of (3.61), making it unnecessary to examine the other asymptotic zones of the interfacial region.
It can be derived readily from (3.61) that The growth of ρ 2 (z) as z → −∞ in Zone 1 agrees with the fact that this function has a local maximum in Zone 2 (see figure 3).Asymptotics (3.64) also guarantees that (3.63) converges at the lower limit.
The function A(ξ ) was computed numerically using the algorithm described in § C.2; the result is plotted in figure 4. For practical use, however, it is more convenient to use the approximation formula (1.4) (for more details, see § C.3).

Curved interfaces
Interfacial curvature influences evaporation via the so-called Kelvin effect, i.e. by changing the effective saturated vapour pressure near the liquid's (curved) surface.This effect can be incorporated readily into the analysis under the assumption that the curvature radius is much larger than the interfacial thickness.This has been done, and the result has turned out to be physically obvious: the Kelvin effect adds a new term to the pressure condition (3.53), where C is the interfacial curvature, and σ is the surface tension.(The DIM expression for σ is given by (B8) of Appendix B.) The extra term in the pressure equation entails an extra term in the GMBC: instead of (3.66), it becomes (1.1) of the Introduction, whereas definition (3.65) of the relative evaporation rate ξ coincides with (1.2) if one sets ρ (+) 1 = (n • ∇ρ 1 ) (+) .

Evaporation of drops as described by the generalised model
To compare the generalised model to the classical one, the former will be applied to evaporation of floating spherical drops.To do so, one needs to replace the classical boundary condition (2.10) with its generalised version (1.1), and leave the rest of the governing set (2.3)-(2.13) the same as before.The ensuing algebra is also the same as before, and one eventually obtains the following differential equation for r d (t): is the ratio of the evaporative flux to the interfacial emission capacity (the latter given by (1.3)).
Equations (4.1)-(4.2) generalise the corresponding classical result (2.21).The term involving A(ξ ) in (4.1) describes emission of vapour by the interface, and the term involving σ gives the shift of the effective saturation pressure near a curved surface.The remaining two terms on the left-hand side of (4.1) are the same as those in its classical counterpart (2.21).
The classical and modified models -described by (2.21) and (4.1), respectively -are compared in figures 5(a,b).For figure 5(a), both were treated as algebraic equations for the non-dimensional evaporation flux ξ with the drop's radius r d being a parameter -so that the result is the dependence ξ versus r d .Figure 5(b), in turn, shows the dependence of the time of full evaporation on the drop's initial radius.
The following features of these figures should be observed.
(i) The (classical) d 2 law states that t e ∼ r 2 d , which is why curve (M) in figure 5(b) is a straight line with slope 2. In figure 5(a), curve (M) is also a straight line, but its slope is −1 (which agrees with the corresponding result of the classical model).
(ii) The shaded areas in figures 5(a,b) mark the region ξ < 1, where the two models are supposed to yield similar results -and indeed, they do.The value of r d corresponding to ξ = 1 is r d ≈ 2 µm.(iii) For smaller r d , curves (M) and (B) begin to diverge.The difference between the classical and generalised results peaks when the drop's radius is close to, and slightly less than, r d = 10 nm.A similar effect is observed in models resolving or parametrising the Knudsen layer (Long et al. 1996;Haut & Colinet 2005;Landry et al. 2007;Jakubczyk et al. 2012;Hołyst et al. 2013Hołyst et al. , 2017;;Rana et al. 2018Rana et al. , 2019;;Zhao & Nadal 2023).(iv) For drops with r d 10 nm, the Kelvin effect 'kicks in': it increases the effective saturated pressure and thus accelerates the vapour emission by the interface -as a result, curves (M) and (B) re-converge.
To verify this explanation, the four terms of (4.1) have been computed individually and plotted in figure 5(c).Evidently, evaporation of small drops is governed by the balance of interfacial effects and the Kelvin effect.Non-isothermality is less important, whereas the diffusion of vapour in the air is unimportant.(v) A general criterion of importance of the van der Waals force relative to diffusion can be obtained by estimating the characteristic ratio B of the second term in the square brackets in (4.1) to the first term.Recalling (4.2), one obtains where Ā is a characteristic value of the function A(ξ ).The van der Waals force is important when B 1, which amounts to For the parameters of water at 25 • C and with Ā = A(0) ≈ 0.14219, one obtains r d 0.4 µm, which agrees qualitatively with the fact that curves (d) and (i) in figure 5(c) intersect at r d ≈ 1 µm.(vi) Figure 5(c) also suggests that for large drops (say, r d 0.1 µm), the Kelvin effect is negligible, which agrees broadly with the estimate of Sobac et al. (2015) .

Concluding remarks
Since the main result of this paper has been summarised in the Introduction, it only remains to comment on further applications of the proposed model and its connections with other models of evaporation.
(i) Apart from drops, the generalised Maxwell boundary condition can be applied to menisci -and if their radii are small, then the GMBC should produce results more accurate than those of the classical model.This should be important for diffusion of fluids in porous media with small pores.Note also that interfacial effects are particularly strong for concave menisci, in which case the Kelvin effect works with the interfacial effects, not against them.This should make the difference between the predictions of the classical and generalised models larger than that for convex menisci and drops.(ii) For flat and cylindrical interfaces, the results obtained via the two models converge with time.In both cases, the diffusion equation does not have a one-dimensional solution describing a steady flux of vapour towards infinity, which effectively means that the throughput of air is zero.As a result, after a sufficiently long time, the air near the interface gets almost saturated.At finite times, however, the classical and generalised models yield different results even for cylindrical and flat interfaces.(iii) A sessile or pendant drop -especially that with moving contact lines -evaporates differently from floating ones.Its surface is highly curved near the contact lines, changing the local evaporation rate.To account for this effect is not a straightforward task, however, as the flow inside the drop is not one-dimensional in this case.One might hope that the generalised model might explain the problem of abnormally small Navier slip arising for several liquids, including water (Podgorski, Flesselles & Limat 2001;Winkels et al. 2011;Puthenveettil, Senthilkumar & Hopfinger 2013;Benilov & Benilov 2015).(iv) All of the problems listed above are easier to solve using the quasi-static approximation.One should keep in mind, however, that Finneran et al. (2021) compared that with the transient solution of the governing equations (for the classical model) and observed that at high pressure, the quasi-steady approximation overpredicts the droplet lifetime by up to 80 %.One might conjecture that the difference between the transient and quasi-steady approaches is due to the fact that the latter implies that the mass of evaporated liquid is infinite.Indeed, the steady solution of the spherically symmetric diffusion equation for vapour is ∼1/r -hence the integral of this solution over the region between the drop's surface and infinity diverges.
Thus for the quasi-steady solution to establish itself around the drop, a significant part of the drop should have evaporated already.This applies to both classical and generalised models.
(v) If the drop's radius is small, but the Kelvin effect is nevertheless neglected, the generalised model -i.e.(4.1) with σ = 0 -predicts that the time of a drop's evaporation is proportional to its radius.A similar behaviour occurs due to the temperature jump associated with the Knudsen layer (e.g.Rana et al. 2019;Zhao & Nadal 2023), but this seems to be just a coincidence.Indeed, the DIM does not describe temperature jumps (which are a purely kinetic effect (Bond & Struchtrup 2004;Persad & Ward 2016;Rana et al. 2018)), whereas the existing kinetic models do not include the van der Waals force (which is vital for Zone 1 of the present analysis).
Yet the Knudsen layer and Zone 1 are both located at the outskirts of the interface, out of equilibrium and affected by irreversible processes: the former by collisions, and the latter by viscosity (which is the hydrodynamic analogue of collisions).All this suggests that Zone 1 is the DIM equivalent of the Knudsen layer.If this is indeed the case, then the main conclusion of this paper is that the van der Waals force is one of the driving effects in the Knudsen layer and thus should be accounted for.
This conjecture can be verified using the Enskog-Vlasov kinetic equation; it includes the van der Waals force (just like the DIM) and can handle temperature jumps (like the Boltzmann equation).Since both effects work the same way -to slow evaporation down -their joint effect should be stronger than predicted by the DIM and existing models of the Knudsen layer separately.(vi) Despite the fact that the DIM cannot handle temperature jumps, it is still closely related to the Enskog-Vlasov equation, with the former being the hydrodynamic approximation of the latter (Giovangigli 2020(Giovangigli , 2021)).The asymptotic structure of the solution of the two models should be similar, and the results obtained in the present paper should make the analysis of the Enskog-Vlasov equation much easier.(vii) Note that, strictly speaking, the surface tension depends on the drop's curvature (Tolman's effect), which can have a profound influence on, say, cavitation (Menzl et al. 2016;Magaletti, Gallo & Casciola 2021;Aasen et al. 2023;Lamas et al. 2023).
Since cavitating bubbles are created with zero radii, Tolman's length -no matter how small -may affect the whole 'trajectory' of their growth.Evaporation appears to be different, however; since Tolman's length for, say, water under normal conditions is less than 1 Å (e.g.Wilhelmsen, Bedeaux & Reguera 2015; Burian et al. 2017), it is much smaller than typical radii of evaporating drops, and the whole effect is likely to be negligible.
Declaration of interests.The author reports no conflict of interests.
one can use boundary conditions (A6)-(A7) to obtain Observe that identity (3.1) implies that ) This equality helps one to integrate the last term of (A11) and obtain p (l.sat) = p (v.sat) .(A13) Physically, equalities (A8) and (A13) reflect the famous Maxwell construction (Maxwell 1875), which inter-relates the partial densities of the components in vapour and liquid.One should also require p (v.sat) = p A , (A14) where p A is the atmospheric pressure.An example of a numerical solution of boundary-value problem (A5)-(A7) in application to a water/air interface (with the chemical potentials G i described in the next appendix) is shown in figure 3. Θ(ρ 1 ) = R −q (0) ln 1 − 0.99 where ρ tp = 999.79kg m −3 , (B5) q (0) = 0.071894, q (1) = 1.4139, q (2) = 8.1126, (B6a-c) q (3) = −8.3669,q (4) = 4. from that of the liquid-air/vapour-air interface.The resulting values, computed by Benilov (2023a), are listed in equality (3.12).The coefficient K 12 is harder to calculate.One might think that it can be determined by comparing the liquid-water/vapour-water interface with the liquid-water/air interfacebut at normal conditions, their characteristics are almost indistinguishable.Instead, K 12 was deduced from the surface tension of the liquid-water/air interface at high pressure, measured by Hinton & Alvarez (2021).The following approach was used.
Within the framework of the DIM, the surface tension is given by where ρ i (z) is the solution of the boundary-value problem (A5)-(A7).It was solved numerically with different values of K 12 and the rest of the parameters given by (B1)-(B7) and (3.12).For each value of K 12 , problem (A5)-(A7) was solved for various values of the atmospheric pressure p A , and this computation was repeated until the best fit with the experimental dependence σ versus p A is achieved.The results are shown in figure 6.Unfortunately, Hinton & Alvarez (2021) provide data for only six values of p A , and the resulting pairs (p A , σ ) do not sit on a smooth curve.Thus these measurements do not seem to merit the use of formal curve-fitting methods; instead, a value of K 12 was chosen expression into (C8), one obtains q (ρ ) ρ 4 dρ .(C15) One might be tempted now to take the limit ξ → ∞, but the integral in (C15) is actually divergent (due to asymptotics (C13)).One can still use (C13) to express this integral through a convergent one: ln ρ d dρ q ρ 3 dρ .(C16) Now take the limit ξ → ∞, so that (C15) becomes The integral in this expression does not involve ξ , hence it can be evaluated 'once and for all'.This was done by solving boundary-value problem (C12)-(C13) numerically and substituting the resulting solution into (C17), which yields A ≈ where ξ 0 and A 0 are undetermined constants; the former regularises asymptotics (C18) at ξ = 0, and the latter improves the accuracy for large ξ .(Note that a term ∼ξ −7/5 would appear as a correction to the leading-order solution if (C18) were extended to higher orders.) The coefficients ξ 0 and A 0 were fixed by fitting (C19) to the corresponding numerical result for the range ξ ∈ [0, 10 5 ], using the MATLAB function 'fit'.The resulting values are those in (1.4) of the Introduction.The relative error of (1.4) within the range ξ ∈ [0, 10 5 ] is approximately 0.8 %.

Figure 1 .
Figure 1.The temperature dependencies of the saturated vapour density ρ (v.sat) 1

Figure 2 .
Figure 2. The time of full evaporation of a water drop of radius 16 µm versus the temperature.The curves are marked with the corresponding relative humidity (in percentage).
the air viscosity scale η, define the characteristic pressure and velocity by

Figure 3 .
Figure3.A schematic illustrating the asymptotic structure of the solution.The curves ρ 1 (z) and ρ 2 (z) describe a flat equilibrium water/air interface at T = 25 • C, computed using the DIM (see Appendix A.2).The scale of the horizontal axes in both panels is logarithmic.The height of the non-shaded region is used as an estimate for the interfacial thickness z.

Figure 4 .
Figure 4.The function A(ξ ).Note that in the right-hand half of the figure, the scale of ξ is logarithmic.

Figure 5 .
Figure 5. Characteristics of an evaporating spherical drop versus its radius, for T = 25 • C and relative humidity H = 50 %.Curves (M) and (B) are computed using Maxwell's classical model and its generalised version, respectively.The numbers in the ellipses show the relative deviations of curve (M) from curve (B).(a) The evaporative flux normalised by the emission capacity of the interface (where ξ is defined by (4.2)).(b) The time of evaporation.The shaded areas of (a) and (b) correspond to ξ < 1, which is where the classical model is supposed to work.(c) The four terms (effects) in (4.1).

Figure 6 .
Figure 6.The surface tension of the water/air interface versus the pressure, at T = 22.5 • C. The black squares show the experimental data of Hinton & Alvarez (2021), and the solid line shows the theoretical result (B8) computed using the parameter values described in Appendix B and K 12 = 0.84978 × 10 −17 m 7 s −2 kg −1 (chosen as the best fit of the experimental results).The dashed curves marked with '±' are computed for K 12 = (0.84978 ± 0.04810) × 10 −17 m 7 s −2 kg −1 , respectively.