betaFeedback

WorksPhysics and Astronomy

Linear stability of ultrathin spherical coatings

Abstract

We unravel the linear stability properties of an otherwise stagnant ultrathin non-wetting liquid film of thickness hoh_o coating a spherical substrate of radius RR. The configuration is known to be unstable due to the competition of the destabilizing van der Waals (vdW) forces and the stabilizing surface tension force. The governing equations of motion written in the Stokes limit of negligible liquid inertia and an accompanying lubrication model are linearised about the zero-velocity base state and decomposed into normal modes in order to obtain the temporal dispersion relation. Discrete unstable modes are identified and tracked as a function of a capillary number CaCa measuring the relative importance of surface tension to vdW forces and the film aspect ratio η=R/hoη= R/h_o. For small enough values of the capillary number only the first polar mode is unstable, and the corresponding maximum growth rate is shown to be a universal function of ηη. Lubrication theory is seen to provide a good quantitative prediction of the film stability properties for η≫1η\gg 1.

Introduction

Thin liquid films are found in a great number of both natural and industrial applications, and they play a central role in many manufacturing processes, physiological and medical processes, biophysics or geophysics. This has drawn the attention of relevant work over the years; for instance, falling films are relevant in coating processes [1, 2], where the ideal scenario is a stable film, spreading gravity currents are important in geological studies [3–6], pulsed-laser-heated thin films are used in patterning and plasmonic applications [7–9], and tear-film dynamics [10, 11, 12, 13] and surfactant replacement therapy [14, 15, 16, 17, 18, 19, 20] are examples where the stability of surfactant-driven films is crucial for a healthy functioning of the eye and lungs. The reader is referred to [21, 22, 23, 24, 25] [9] and references therein for a detailed review of the literature of films.

When the substrate supporting the liquid film is not planar, the classical lubrication description needs to be modified to account for the corresponding curvature distribution [26, 27], which is known to significantly affect both the thin film flow and its stability [28, 29]. The particular case of a thin liquid film coating a spherical surface has been studied in several contexts, ranging from the impact of a drop onto a solid spherical target [30], to the dynamics of the canonical gravity current [31, 32], including the effects sphere rotation [33] and thermocapillarity [34].

If the thickness of the coating is sufficiently small van der Waals (vdW) forces become comparable to the surface tension force, and their influence cannot be neglected in the analysis. Indeed, for instance, Grigor'ev et al. [35] clearly showed that vdW forces profoundly affect the spectrum of unstable disturbances in electrically charged liquid shells coating a rigid spherical core. In the context of soft elastic thin films bonded to curved substrates, Li et al. [36] investigated the critical conditions for the occurrence of vdW-induced wrinkling using linear theory in both spherical and cylindrical configurations. The curvature of the substrate is shown to have a significant influence on the wrinkling behavior of the film. In their analysis, independence of the mode of instability with respect to azimuthal perturbations is also exhibited, which was anticipated by Colin [37].

However, to the best of our knowledge, the influence of vdW forces on the spinodal dewetting process associated to ultrathin, non-wetting liquid films coating a sphere has not been reported before. Thus, our main objective here is to present a detailed analysis of the linear stability properties of these ultrathin spherical liquid shells, using both a Stokes description and lubrication theory.

The paper is structured as follows: the general formulation of the problem is given in section II followed by the linear stability analysis presented in section III. Concluding remarks and future prospects are discussed in section IV.

Formulation of the problem

A clean and initially static thin film of a Newtonian liquid of viscosity μ\mu coating a spherical solid substrate of radius RR is considered. Let the instantaneous film thickness be referred to as hh such that its location may be defined as

r=R+h(θ,φ,t)r = R + h(\theta,\varphi,t), where spherical coordinates (r,θ,φ)(r,\theta,\varphi) are employed (see schematic in Figure 1). The film, of initial thickness hoh_o, is surrounded by passive atmospheric air at constant pressure pap_a. The coefficient of surface tension σ\sigma between the air and the liquid will be taken as a constant. Finally, dispersive long-range van der Waals (vdW) forces are to be modelled through the use of a disjoining pressure ϕ(h)\phi(h) proportional to the effective Hamaker constant AA of the liquid-substrate pair. The initially spherical liquid-air interface may become unstable if the destabilizing vdW forces are no longer balanced by the capillary force.

Figure 1. Schematic of the initial flow configuration, parameters, and spherical coordinates employed in the description.

Dimensionless equations of motion

We make use of

lc=ho,vc=A6πμho2,tc=6πμho3A,pc=A6πho3(1)l_c = h_o,\qquad v_c = \frac{A}{6\pi\mu h_o^2},\qquad t_c = \frac{6\pi\mu h_o^3}{A},\qquad p_c = \frac{A}{6\pi h_o^3} \tag*{(1)}

as relevant scales of length, velocity, time, and pressure to nondimensionalize the governing equations of motion. Since the effects of liquid inertia will be neglected under the assumption of small Laplace numbers, ρσho2/μ≪1\rho\sigma h_o^2/\mu\ll1, the motion of the liquid is effectively governed by the Stokes equations

∇⋅v=0,∇⋅T=0,(2)\nabla\cdot\boldsymbol{v} = 0,\qquad\nabla\cdot\mathbf{T} = 0, \tag*{(2)}

where T=−pI+∇v+∇vT\mathbf{T}=-p\mathbf{I}+\nabla\boldsymbol{v}+\nabla\boldsymbol{v}^{\mathsf{T}} is the liquid stress tensor, pp is the gauge pressure, and I\mathbf{I} is the identity tensor. Eq. (2) needs to integrated with appropriate boundary conditions, namely, no-slip at the substrate r=ηr=\eta, where η=R/ho\eta=R/h_o, and the kinematic and stress balance at the liquid-air interface r=η+hr=\eta+h,

n⋅(x˙s−v)=0,(3)\boldsymbol{n}\cdot(\dot{\boldsymbol{x}}_s-\boldsymbol{v})=0, \tag*{(3)}
−T⋅n=Ca−1n(∇s⋅n)+ϕ(h)n,(4)-\mathbf{T}\cdot\boldsymbol{n}=Ca^{-1}\boldsymbol{n}(\nabla_s\cdot\boldsymbol{n})+\phi(h)\boldsymbol{n}, \tag*{(4)}

where the surface has been parameterised by the position vector xs\boldsymbol{x}_s (dot indicating time derivative) and its outward-pointing unit normal vector n\boldsymbol{n}. The surface divergence operator is defined as ∇s=(I−nn)⋅∇\nabla_s=(\mathbf{I}-\boldsymbol{n}\boldsymbol{n})\cdot\nabla. It is seen that for a given initial geometrical configuration, i.e., for fixed values of η\eta, the capillary number Ca=A/(6πσho2)Ca=A/(6\pi\sigma h_o^2) emerges as the only nondimensional parameter of the problem, measuring the relative importance of vdW and capillary forces. Conversely, one can make use of the molecular length scale a=A/(6πσ)a=\sqrt{A/(6\pi\sigma)} [38, 39] instead of hoh_o in order to make the problem dimensionless, with ho/a=Ca−1/2h_o/a=Ca^{-1/2} the corresponding re-scaling.

Van der Waals potential for a liquid film coating a spherical substrate

Equations (2)–(4) are completed upon specification of the disjoining pressure accounting for the vdW forces. The standard expression for a planar film, ϕplanar=A/(6πh3)\phi_{\mathrm{planar}} = A/(6\pi h^3), would be adequate in the limit η≫1\eta\gg1 of very small initial thickness compared to the substrate radius. However, it is well known that substrate curvature affects the disjoining pressure substantially [40], so that appropriate corrections are needed to describe cases where η≲1\eta\lesssim1. Fortunately, the spherical shell configuration admits a closed-form description that can be determined following [36, 41] to give, in dimensional form,

ϕ(h)=A6πh38R3(h+2R)3,(5)\phi(h) = \frac{A}{6\pi h^3}\frac{8R^3}{(h+2R)^3}, \tag*{(5)}

for a liquid coating of thickness hh resting on top of a spherical substrate of radius RR. Note that AA may be considered either positive (dewetting) or negative (wetting) in the analysis. In the present work, we will restrict ourselves to dewetting scenarios for which A>0A > 0, since the wetting case is trivially stable. Upon making use of (1), the dimensionless version of (5) is

ϕ(h)=8η3h3(h+2η)3,(6)\phi(h) = \frac{8\eta^3}{h^3(h+2\eta)^3}, \tag*{(6)}

to be employed in Eq. (4). The associated small-curvature expansion is given by

ϕ(h)=1h3−32h2η+32hη2+O(η−3),(7)\phi(h) = \frac{1}{h^3} - \frac{3}{2h^2\eta} + \frac{3}{2h\eta^2} + O(\eta^{-3}), \tag*{(7)}

in reciprocal powers of η\eta, which effectively recovers the planar limit ϕplanar=h−3\phi_{\mathrm{planar}} = h^{-3} to within O(η−1)O(\eta^{-1}) relative errors. The derivative of the disjoining pressure at h=1h = 1 is given by

ϕ′(1)=−48η3(1+η)(1+2η)4,(8)\phi'(1) = -\frac{48\eta^3(1+\eta)}{(1+2\eta)^4}, \tag*{(8)}

which will become useful below when deriving the linearized stability equations, and which has a large-η\eta expansion

ϕ′(1)=−3+3η+O(η−2).(9)\phi'(1) = -3 + \frac{3}{\eta} + O(\eta^{-2}). \tag*{(9)}

Lubrication approximation

Following a standard procedure, briefly explained in Appendix A, the lubrication model for the evolution of the film thickness hh is easily deduced, yielding the evolution equation

(η+h)2sin⁡θ∂h∂t+∂∂θ(G(h,η)sin⁡θ∂p∂θ)+1sin⁡θ∂∂φ(G(h,η)∂p∂φ)=0,(10)(\eta+h)^2\sin\theta\frac{\partial h}{\partial t} +\frac{\partial}{\partial\theta}\left(G(h,\eta)\sin\theta\frac{\partial p}{\partial\theta}\right) +\frac{1}{\sin\theta}\frac{\partial}{\partial\varphi}\left(G(h,\eta)\frac{\partial p}{\partial\varphi}\right)=0, \tag*{(10)}

with mobility function GG and static pressure pp computed through

G(h,η)=h3(h+2η)6(h−η),p=CCa+ϕ,(11)G(h,\eta)=\frac{h^3(h+2\eta)}{6(h-\eta)},\qquad p=\frac{\mathcal{C}}{\mathrm{Ca}}+\phi, \tag*{(11)}

where the curvature C\mathcal{C}, whose exact expression is provided as Eq. (29) in Appendix A. The lubrication model given by Eq. (10) was also deduced by Kang et al. [33] (see also [34]). Note that, although Eq. (10) assumes slender flow, the effect of substrate curvature is retained not only through the metric factor sin⁡θ\sin\theta, but also through the functions of η\eta appearing in the first term and in the mobility function G(h,η)G(h,\eta). The classical thin-film approximation [32, 33] is obtained by expanding Eq. (10) in powers of η−1\eta^{-1} for η≫1\eta\gg1 to yield, at leading order,

η2sin⁡θ∂h∂t−∂∂θ(sin⁡θh33∂p∂θ)−1sin⁡θ∂∂φ(h33∂p∂φ)=0,(12)\eta^2\sin\theta\frac{\partial h}{\partial t} -\frac{\partial}{\partial\theta}\left(\sin\theta\frac{h^3}{3}\frac{\partial p}{\partial\theta}\right) -\frac{1}{\sin\theta}\frac{\partial}{\partial\varphi}\left(\frac{h^3}{3}\frac{\partial p}{\partial\varphi}\right)=0, \tag*{(12)}

with pressure now given by p=C/Ca+h−3p=\mathcal{C}/\mathrm{Ca}+h^{-3}, and curvature C\mathcal{C} linearized around the base state according to

C=21+η−1(1+η)2[1sin⁡2θ∂2h∂φ2+1sin⁡θ∂∂θ(sin⁡θ∂h∂θ)+2h].(13)\mathcal{C}=\frac{2}{1+\eta}-\frac{1}{(1+\eta)^2}\left[\frac{1}{\sin^2\theta}\frac{\partial^2h}{\partial\varphi^2}+\frac{1}{\sin\theta}\frac{\partial}{\partial\theta}\left(\sin\theta\frac{\partial h}{\partial\theta}\right)+2h\right]. \tag*{(13)}

Note that the prefactor in Eq. (12) calls for the temporal rescaling t→t/η2t\to t/\eta^2, which would render it parameter-free. However, curvature effects must be retained in the definition of the capillary pressure through the curvature in Eq. (13) so as to produce the cut-off of temporal instability.

Linear stability

Cut-off polar wavenumber

As shown in Appendix B, a Plateau-like free energy argument allows the derivation of the neutral stability condition, which reads

ℓ(ℓ+1)=2−ϕ′(1)Ca(η+1)2,(14)\ell(\ell+1)=2-\phi'(1)Ca(\eta+1)^2, \tag*{(14)}

where ℓ=0,1,2,…\ell=0,1,2,\ldots is the polar wavenumber of the spherical harmonic and ϕ′(1)=−48η3(1+η)/(1+2η)4<0\phi'(1)=-48\eta^3(1+\eta)/(1+2\eta)^4<0 in the non-wetting case considered herein. The cut-off polar wavenumber ℓc\ell_c is then given by

ℓc=⌊12(9−4Ca ϕ′(1)(η+1)2−1)⌋,(15)\ell_c=\left\lfloor\frac{1}{2}\left(\sqrt{9-4Ca\,\phi'(1)(\eta+1)^2}-1\right)\right\rfloor, \tag*{(15)}

where ⌊⋅⌋\lfloor\cdot\rfloor represents integer part, revealing that perturbations with 0<ℓ<ℓc0<\ell<\ell_c and ∣m∣≤ℓ|m|\leq\ell, are linearly unstable. Note that the spherically symmetric mode ℓ=0\ell=0 is always neutrally stable, the mode ℓ=1\ell=1 is unconditionally unstable, and the value of ℓc≥2\ell_c\geq2 increases with CaCa and η\eta. In the limit of very thin films, η≫1\eta\gg1, it proves convenient to define k=ℓ/ηk=\ell/\eta as the equivalent wavenumber of a planar film. Indeed, a large-η\eta expansion of (15) provides kc=ℓc/η=3Ca+O(η−1)k_c=\ell_c/\eta=\sqrt{3Ca}+O(\eta^{-1}), effectively recovering the classical cutoff of a planar film [42, 43].

The dispersion relation

The governing equations (2) and the boundary conditions (4) are decomposed into normal modes by introducing the usual ansatz q(x,t)=Q(x)+ϵq^(x)exp⁡(ωt)q(x,t)=Q(x)+\epsilon\hat{q}(x)\exp(\omega t) where q=(vr,vθ,vϕ,p,h)q=(v_r,v_\theta,v_\phi,p,h) is the vector of state variables, ω\omega is the growth rate, 0<ϵ≪10<\epsilon\ll1 is a small amplitude and the steady-state solution is Q=(0,0,0,po,1)Q=(0,0,0,p_o,1). Note that the dependence of the eigenfunctions on the spatial coordinates has not been yet anticipated. Upon substitution of the ansatz in Eq. (2) one finds

∇⋅v^=0,0=−∇p^+∇2v^,(16)\nabla\cdot\hat{\boldsymbol{v}}=0,\qquad0=-\nabla\hat{p}+\nabla^2\hat{\boldsymbol{v}}, \tag*{(16)}

valid in the spherical shell η<r<η+1\eta<r<\eta+1, 0<θ<π0<\theta<\pi, 0<φ<2π0<\varphi<2\pi. Following the procedure explained in Appendix C, a closed-form solution to the associated dispersion relation for the growth rate ω\omega is found as a rational function of ℓ\ell, η\eta and CaCa, given by

ω=ℓ(ℓ+1)(2ℓ+1)2Ca(η+1)×B1[ℓ(ℓ+1)−2+ϕ′(1)Ca(1+η)2]B2+(B3+B4+B5+B6)η2ℓ(η+1)2ℓ+B7(17)\omega=\frac{\ell(\ell+1)(2\ell+1)}{2Ca(\eta+1)}\times\frac{B_1\left[\ell(\ell+1)-2+\phi'(1)Ca(1+\eta)^2\right]}{B_2+(B_3+B_4+B_5+B_6)\eta^{2\ell}(\eta+1)^{2\ell}+B_7} \tag*{(17)}

where the polynomials BiB_i, i=1…7i=1\ldots7, provided in Appendix D, have been introduced for convenience.

It may be recognised that ω>0\omega>0 for Ca>0Ca>0 and a range of discrete positive values of 0<ℓ≤ℓc0<\ell\leq\ell_c, where the cut-off wavenumber ℓc\ell_c, defined as the largest value of ℓ\ell for which ω>0\omega>0, is given by Eq. (15). The cut-off value can be found by inspection of the linearized curvature term, which appears in Eq. (17) multiplying the polynomial B1B_1, providing the same cut-off as that found above in (14). Since the azimuthal wavenumber does not appear explicitly in the dispersion relation, the flow is unstable for all values of ∣m∣≤ℓ|m|\leq\ell, provided that ℓ≤ℓc\ell\leq\ell_c. Note that a similar result is obtained, for instance, by [37] in the context of stressed composite spherical shells and by [36] for the vdW-induced instability of thin elastic spherical shells.

The limit η≫1\eta\gg1 of a liquid film coating a planar substrate may be effectively recovered by noticing that as η\eta grows, the number of unstable polar modes accommodated must grow accordingly. According to the curvature term in (C12), ℓ∼η\ell\sim\eta as η→∞\eta\to\infty in order to produce a finite wavenumber k=ℓ/ηk=\ell/\eta, which plays the role of the streamwise wavenumber defined in [44]. Indeed, the planar growth rate

ωplanar=−(k2−3Ca)(k−cosh⁡ksinh⁡k)2kCa(k2+cosh⁡2k)(18)\omega_{\mathrm{planar}}=-\frac{(k^2-3Ca)(k-\cosh k\sinh k)}{2kCa(k^2+\cosh^2 k)} \tag*{(18)}

(see for instance [45]) may be obtained by taking the double limit η,ℓ→∞\eta,\ell\to\infty of (17) for finite ℓ/η\ell/\eta and retaining only the leading-order term in (7). The classical result for the cut-off wavenumber kc=3Cak_c=\sqrt{3Ca} is recovered, originally deduced by Scheludko [42] and Vrij [43] for planar films. The required algebra for such simplification, which is not trivial, has been carried out with the help of [46].

In the case of the lubrication models (10) and (12), the associated dispersion relations are given by

ωLUB1=(2η+1)ℓ(ℓ+1)[2−ϕ′(1)Ca(η+1)2−ℓ(ℓ+1)]6Ca(η−1)(η+1)4(19)\omega_{\mathrm{LUB1}} = \frac{(2\eta+ 1)\ell(\ell+ 1)\left[2 - \phi^{\prime}(1)Ca(\eta+ 1)^2 - \ell(\ell+ 1)\right]}{6Ca(\eta- 1)(\eta+ 1)^4} \tag*{(19)}

and

ωLUB2=ℓ(ℓ+1)[2+3Ca(η+1)2−ℓ(ℓ+1)]3Caη2(η+1)2(20)\omega_{\mathrm{LUB2}} = \frac{\ell(\ell+ 1)\left[2 + 3Ca(\eta+ 1)^2 - \ell(\ell+ 1)\right]}{3Ca\eta^2(\eta+ 1)^2} \tag*{(20)}

respectively, where, in the latter case, it has been taken into account that ϕ′(1)=−3\phi^{\prime}(1) = -3 in the thin-film limit η≫1\eta\gg1.

An unexpected feature of the lubrication model with substrate curvature effects is the existence of a singularity at η=1\eta= 1, manifested by the factor η−1\eta- 1 appearing in the denominator of Eq. (19). This failure can be traced back to the solution of the velocity field in spherical coordinates, given by Eqs. (26)–(27), where the factor h−ηh - \eta appears in the denominator as a consequence of the imposition of the leading-order expression for the stress-free boundary condition at the liquid-air interface. This singularity represents a major limitation for the practical use of the leading-order lubrication model for relatively thick films which, in particular, predicts an infinite disturbance growth rate for η=1\eta= 1, and a negative growth rate for η<1\eta< 1. As shown in Appendix E, a simple regularization of the singularity can be achieved simply by retaining the radial velocity of the interface in the stress-free boundary condition. In contrast, such a singularity is absent in the thin-film lubrication model (20) although, as shown below, and not unexpectedly, the latter model also fails catastrophically in the prediction of the stability properties of the film for η≲1\eta\lesssim1.

Figure 2 shows the temporal growth rate ω(ℓ)\omega(\ell) for two different cases, namely (a) Ca=0.1Ca = 0.1, η=10\eta= 10, for which ℓc=5\ell_c = 5, and (b) Ca=0.01Ca = 0.01, η=100\eta= 100, for which ℓc=17\ell_c = 17. Four results are displayed in Figure 2, corresponding to the Stokes dispersion relation (17) (circles), the lubrication model with substrate curvature effects (19) (squares), the lubrication model in the thin-film limit (20) (diamonds) and the planar limit (18) (dashed line). The maximum growth rate is highlighted in red. Both lubrication models overpredict the temporal growth rate considerably, being the model without curvature effects the worst performing one. Moreover, in the case of the relatively thicker film with η=10\eta= 10, both lubrication models fail to account for the optimal polar wavenumber, since they predict ℓM=4\ell_M = 4, while the Stokes dispersion relation yields ℓM=3\ell_M = 3. Interestingly, the equivalent planar dispersion relation (18) is able to reproduce the exact growth rate fairly well even in the case with η=10\eta= 10, since it does not rely on the slenderness of the self-induced flow, but only on the weak curvature approximation. In the limit of very thin films, exemplified by the case with η=100\eta= 100, the three models are seen to correctly capture the exact result with small relative errors within the whole range of unstable wavenumbers, including the prediction of ℓM=12\ell_M = 12.

Image not extracted

Figure 2. (a) Temporal growth rate for (a) Ca=0.1Ca = 0.1, η=10\eta= 10 (ℓc=5\ell_c = 5) and (b) Ca=0.01Ca = 0.01, η=100\eta= 100 (ℓc=17\ell_c = 17). Symbols represent the results obtained using the Stokes dispersion relation (17) (circles), the lubrication model with curvature effects (19) (squares) and the thin-film lubrication model (20) (diamonds). The planar limit given by (18) is also represented for continuous values of ηk\eta k and 0<k<3Ca0 < k < \sqrt{3Ca} (dashed line).

The main result of the present paper is summarized in Figure 3, which shows the maximum growth rate ωM\omega_M (left panels) and the corresponding optimal wavenumber ℓM\ell_M (right panels) as functions of η\eta, for Ca=0.1Ca = 0.1 (upper panels) and Ca=0.01Ca=0.01 (lower panels). The exact result, obtained from the Stokes dispersion relation (black solid lines), clearly shows that the function ωM(η)\omega_M(\eta) has discontinuous derivatives at discrete values of η=ηi(Ca)\eta=\eta_i(Ca), i=1,2,…i=1,2,\ldots, associated with the monotonic increase of the dominant wavenumber ℓM=i\ell_M=i as η\eta increases. For Ca=0.1Ca=0.1, the Stokes dispersion relation provides ηi=(4.38,7.31,10.05,12.89,…)\eta_i=(4.38, 7.31, 10.05, 12.89,\ldots), corresponding to changes of the dominant wavenumber from ℓM=1\ell_M=1 for η<4.38\eta<4.38, to ℓM=2\ell_M=2 for 4.38<η<7.314.38<\eta<7.31, and so on. These results are to be compared with the lubrication results, namely ηi,LUB1=(4.03,6.82,9.51,12.03,…)\eta_{i,\mathrm{LUB1}}=(4.03, 6.82, 9.51, 12.03,\ldots) and ηi,LUB2=(3.46,6.27,8.99,11.54,14.20,…)\eta_{i,\mathrm{LUB2}}=(3.46, 6.27, 8.99, 11.54, 14.20,\ldots), which clearly show the better relative performance of the lubrication model with account taken for the substrate curvature. In the case Ca=0.01Ca=0.01, we obtain ηi=(13.62,22.74,31.26,39.55,…)\eta_i=(13.62, 22.74, 31.26, 39.55,\ldots) using the Stokes dispersion relation, while the lubrication models provide ηi,LUB1=(13.62,22.42,30.83,39.01,…)\eta_{i,\mathrm{LUB1}}=(13.62, 22.42, 30.83, 39.01,\ldots) and ηi,LUB2=(13.07,21.81,30.41,38.470,…)\eta_{i,\mathrm{LUB2}}=(13.07, 21.81, 30.41, 38.470,\ldots) revealing that, as expected, the performance of both lubrication models improves as CaCa decreases and η\eta increases. Another important conclusion drawn from Figure 3 is that lubrication theory must be used with great care in spherical geometries if quantitative precision is desired. Indeed, for Ca=0.1Ca=0.1, the results of the lubrication models are seen to substantially overpredict the maximum growth rate over the whole range of values of η\eta, especially for η≲10\eta\lesssim10. For η→∞\eta\to\infty, both lubrication models reach an asymptotic value that is larger than the exact value provided by the spherical and planar

Image not extracted

Figure 3. Maximum growth rate ωM\omega_M (left panels) and corresponding optimal wavenumber ℓM\ell_M (right panels) as functions of the aspect ratio η\eta for Ca=0.1Ca=0.1 (upper panels) and Ca=0.01Ca=0.01 (lower panels), as predicted by the Stokes dispersion relation (17) (black solid lines), the lubrication model with curvature effects (19) (blue solid lines), the thin-film lubrication model (20) (orange solid lines), and the planar dispersion relation (18) (dashed lines). Red symbols indicate the values of ωM\omega_M and ℓM\ell_M corresponding to Figure 2.

Stokes dispersion relations. For a smaller value of Ca=0.01Ca = 0.01 the characteristic streamwise length of the most unstable mode is larger, and thus the lubrication prediction improves with respect to the case with Ca=0.1Ca = 0.1, especially for η≫1\eta\gg1. Nevertheless, the quantitative agreement deteriorates with decreasing values of η\eta, and is completely lost for η≲3\eta\lesssim3. Finally, it is interesting to note that the equivalent planar dispersion relation (18) correctly reproduces the exact result for η≫1\eta\gg1 and arbitrary values of CaCa, since it does not rely on the slenderness of the flow.

Since the polar mode ℓ=1\ell= 1 is the dominant one for a substantial range of film aspect ratios, it proves convenient to particularize the Stokes dispersion relation (17) to ℓ=1\ell= 1, yielding the universal function

ωℓ=1(η)=16η2(η+1)2[5η(η+1)+2](2η+1)3[5η(η(η(η+3)+6)+6)+3)+3](21)\omega_{\ell=1}(\eta) = \frac{16\eta^2(\eta+1)^2[5\eta(\eta+1)+2]}{(2\eta+1)^3[5\eta(\eta(\eta(\eta+3)+6)+6)+3)+3]} \tag*{(21)}

represented with a dashed line in Figure 4, which, interestingly enough, is independent of CaCa. The function ωℓ=1\omega_{\ell=1} displays a maximum value ω=0.3039\omega= 0.3039 at η=0.736\eta= 0.736, and is seen to correctly capture the linear stability properties of the film for η≲2\eta\lesssim2 and any realistic value of CaCa. Indeed, note that Ca=1Ca = 1 corresponds to a film whose initial thickness hoh_o is equal to the molecular length scale aa. The smaller the value of CaCa, the larger the maximum value of η=ηmax⁡\eta= \eta_{\max} for which Eq. (21) correctly represents the linear stability properties. For instance, ηmax⁡=(4.38,13.62,44.2)\eta_{\max} = (4.38, 13.62, 44.2) for Ca=(0.1,0.01,0.001)Ca = (0.1, 0.01, 0.001), corresponding to the values of η\eta at which the dominant ℓ\ell-mode jumps by one unit. For η≪1\eta\ll1, Eq. (21) has the expansion ωℓ=1=32η2/3−208η3/3+O(η4)\omega_{\ell=1} = 32\eta^{2/3} - 208\eta^{3/3} + O(\eta^4), revealing a strong stabilization effect for very thick films.

Image not extracted

Figure 4. Optimal growth rate for the Stokes dispersion relation for several values of CaCa using the complete expression for ϕ′(1)\phi^{\prime(1)} (solid black lines) and ϕ′(1)≃−3\phi^{\prime(1)} \simeq-3 (solid light grey lines). The envelope function given by Eq. (21) is represented for visual guidance (dashed blue line).

To assess the role of the vdW potential, Figure 4 also includes the result of a calculation where the exact expression for ϕ′(1)\phi^{\prime(1)}, given by Eq. (8), is substituted by its thin-film approximation of -3, corresponding to the first term of the expansion (9). The result, represented with a gray solid line, reveals that a proper description of thin films with aspect ratios η≲10\eta\lesssim10 requires the use of the exact vdW potential, which cannot be substituted by its thin-film approximation if quantitative precision is required.

Conclusions

We have presented a detailed analysis of the linear stability properties of films of a Newtonian liquid coating the external surface of a solid spherical substrate under conditions in which the inertia of the liquid is negligible and van der Waals (vdW) forces are comparable to surface tension forces. The linearized Stokes equations have been solved analytically to yield an exact dispersion relation describing the set of discrete temporal modes of the liquid film in terms of spherical harmonics parameterized by the polar wavenumber ℓ\ell. Two nondimensional parameters govern the flow, namely a capillary number CaCa, which compares the relative strength of the vdW and surface tension forces, and the ratio of the sphere radius to the initial film thickness, η\eta. The first polar mode, ℓ=1\ell= 1, has been shown to be unconditionally unstable, while higher-order modes with ℓ≥2\ell\ge2 are destabilized when either CaCa or η\eta become large enough. The dominant mode ℓM\ell_M has been determined by solving the dispersion relation in wide ranges of the (η,Ca)(\eta, Ca) parameter plane. For a given value of CaCa, the value of ℓM\ell_M has been shown to increase in unit steps at a discrete set of values of η\eta. In the limit of very thin films, η≫1\eta\gg1, the stability properties converge to those of a planar film with an equivalent continuous streamwise wavenumber k=ℓ/ηk = \ell/\eta defined in the double limit η,ℓ→∞\eta, \ell\to\infty. In the limit of thick films, η≲1\eta\lesssim1, the dominant mode is always ℓM=1\ell_M = 1 for realistic values of CaCa, and the maximum growth rate becomes a universal function of η\eta displaying a maximum of 0.3039 at η=0.736\eta= 0.736. We have also shown that the use of the exact expression of the spherical vdW potential is mandatory for films with aspect ratio η≲10\eta\lesssim10, which cannot be described using the standard simplification routinely used in the literature.

The performance of lubrication theory, which is routinely employed to describe the dynamics of the type of liquid films studied herein, has also been assessed using two different approximations: a first model that retains the curvature of the spherical substrate, and a second one valid in the limit η≫1\eta\gg1 of thin films relative to the substrate. The latter model, which is the one used in previous works, has been shown to provide a reasonably good quantitative agreement with the Stokes equations, but only for small enough values of CaCa and large enough values of η\eta. The first model improves the agreement with the full equations of motion for η>1\eta> 1, although it displays an unexpected singularity at η=1\eta= 1 related with the structure of the velocity field in the lubrication approximation when the stress-free boundary condition is imposed at the liquid-air interface. A regularization of this singularity is possible if the radial velocity is retained in the shear-free boundary condition, although the relative errors of the regularized model remain large when applied to relatively thick films with η≲1\eta\lesssim1, which can only be properly described with use of the full equations of motion. The failure of the lubrication models to account for the linear stability properties of relatively thick films rises doubts about their validity for the description of the nonlinear dynamics of the film, which will be reported in a separate contribution.

Acknowledgments

The authors thank Dr. A. Martínez-Calvo for insightful discussions. This research was funded by the Spanish MCIU-Agencia Estatal de Investigación through project PID2020-115655GB-C22, partly financed through FEDER European funds.

Derivation of the lubrication models

We follow the steps of Kang et al. [33] in order to derive a nonlinear equation for the evolution of the film height h(θ,φ,t)h(\theta,\varphi,t). We begin by writing the kinematic condition in the liquid-air interface r=η+hr = \eta+ h,

∂h∂t=vr−vθr∂h∂θ−vφrsin⁡θ∂h∂φ,(22)\frac{\partial h}{\partial t}=v_r-\frac{v_\theta}{r}\frac{\partial h}{\partial\theta}-\frac{v_\varphi}{r\sin\theta}\frac{\partial h}{\partial\varphi}, \tag*{(22)}

which combined with the continuity equation after integrating across the liquid layer results in

(η+h)2sin⁡θ∂h∂t+∂∂θ(sin⁡θ∫ηη+hrvθ dr)+∂∂φ∫ηη+hrvφ dr=0.(23)(\eta+h)^2\sin\theta\frac{\partial h}{\partial t}+\frac{\partial}{\partial\theta}\left(\sin\theta\int_{\eta}^{\eta+h}rv_\theta\,\mathrm{d}r\right)+\frac{\partial}{\partial\varphi}\int_{\eta}^{\eta+h}rv_\varphi\,\mathrm{d}r=0. \tag*{(23)}

The radial-component of the momentum equation yields ∂rp=0\partial_r p = 0 to leading order, therefore p=p(θ,φ,t)p = p(\theta,\varphi,t). Direct integration of the in-plane components of the momentum equation across the liquid layer is used to determine the polar and azimuthal components of the velocity, namely

1r∂p∂θ=1r2∂∂r(r2∂vθ∂r),(24)\frac{1}{r}\frac{\partial p}{\partial\theta}=\frac{1}{r^2}\frac{\partial}{\partial r}\left(r^2\frac{\partial v_\theta}{\partial r}\right), \tag*{(24)}
1rsin⁡θ∂p∂φ=1r2∂∂r(r2∂vφ∂r),(25)\frac{1}{r\sin\theta}\frac{\partial p}{\partial\varphi}=\frac{1}{r^2}\frac{\partial}{\partial r}\left(r^2\frac{\partial v_\varphi}{\partial r}\right), \tag*{(25)}

for which

vθ=∂p/∂θ2rr−ηh−η[h(r+η)−rη+η2],(26)v_\theta= \frac{\partial p/\partial\theta}{2r}\frac{r-\eta}{h-\eta}\left[h(r+\eta)-r\eta+\eta^2\right], \tag*{(26)}
vφ=∂p/∂φ2rsin⁡θr−ηh−η[h(r+η)−rη+η2],(27)v_\varphi= \frac{\partial p/\partial\varphi}{2r\sin\theta}\frac{r-\eta}{h-\eta}\left[h(r+\eta)-r\eta+\eta^2\right], \tag*{(27)}

where the no-slip boundary condition at r=ηr=\eta, vθ=vφ=0v_\theta=v_\varphi=0, and the leading-order zero-tangential-stress boundary condition at r=η+hr=\eta+h, ∂r(vθ/r)=∂r(vφ/r)=0\partial_r(v_\theta/r)=\partial_r(v_\varphi/r)=0, have been imposed. Inserting the expressions for the velocities into (A2) and integrating yields

(η+h)2sin⁡θ∂h∂t+∂∂θ(G(h,η)sin⁡θ∂p∂θ)+1sin⁡θ∂∂φ(G(h,η)∂p∂φ)=0,(28)(\eta+h)^2\sin\theta\frac{\partial h}{\partial t}+\frac{\partial}{\partial\theta}\left(G(h,\eta)\sin\theta\frac{\partial p}{\partial\theta}\right)+\frac{1}{\sin\theta}\frac{\partial}{\partial\varphi}\left(G(h,\eta)\frac{\partial p}{\partial\varphi}\right)=0, \tag*{(28)}

where the auxiliary mobility function GG is defined in (11). This evolution equation for the film height must be complemented with the normal-stress condition at leading order p=C+ϕp=C+\phi, where the complete curvature CC is given by

C=hθ{3hθrs−[hθ2+rs2]cot⁡θ}−rs2hθθ−{hφφ[hθ2+rs2]+hφ2[hθθ+2hθcot⁡θ−3rs]+2hφhθhθφ}csc⁡2θ+2rs3rs[hθ2+csc⁡2θhφ2+rs2]3/2,(29)C=\frac{h_\theta\left\{3h_\theta r_s-\left[h_\theta^2+r_s^2\right]\cot\theta\right\}-r_s^2h_{\theta\theta}-\left\{h_{\varphi\varphi}\left[h_\theta^2+r_s^2\right]+h_\varphi^2\left[h_{\theta\theta}+2h_\theta\cot\theta-3r_s\right]+2h_\varphi h_\theta h_{\theta\varphi}\right\}\csc^2\theta+2r_s^3}{r_s\left[h_\theta^2+\csc^2\theta h_\varphi^2+r_s^2\right]^{3/2}}, \tag*{(29)}

with rs=η+hr_s=\eta+h.

Calculation of the critical conditions

Following the Plateau-like argument given in [42, 43, 47], in which undulations of the free surface with negative extra Gibbs energy tend to grow in time, let us consider the dimensionless excess Gibbs energy ΔG\Delta G for our particular configuration. To that end, let ΔG=ΔGs+ΔGmolec\Delta G=\Delta G_s+\Delta G_{\mathrm{molec}}, where ΔGs=Ca−1ΔS\Delta G_s=\mathrm{Ca}^{-1}\Delta S is the Gibbs energy associated to the extra surface area ΔS\Delta S due to the perturbation and ΔGmolec\Delta G_{\mathrm{molec}} is the extra interaction energy due to the molecular interactions, which take the form

ΔS=∫02π∫0π∣Xθ∧Xφ∣ dθdφ−4π(η+1)2,(30)\Delta S=\int_0^{2\pi}\int_0^\pi\lvert\mathbf{X}_\theta\wedge\mathbf{X}_\varphi\rvert\,d\theta d\varphi-4\pi(\eta+1)^2, \tag*{(30)}
ΔGmolec=∫02π∫0π[ϕ(1)Δh+ϕ′(1)(Δh)2/2+⋯ ](η+1)2sin⁡θ dθdφ,(31)\Delta G_{\mathrm{molec}}=\int_0^{2\pi}\int_0^\pi\left[\phi(1)\Delta h+\phi'(1)(\Delta h)^2/2+\cdots\right](\eta+1)^2\sin\theta\,d\theta d\varphi, \tag*{(31)}

where X=r(θ,φ)(sin⁡θcos⁡φ,sin⁡θsin⁡φ,cos⁡θ)\mathbf{X}=r(\theta,\varphi)(\sin\theta\cos\varphi,\sin\theta\sin\varphi,\cos\theta) is the Cartesian parameterization of the corrugated surface r=r(θ,φ)=η+h(θ,φ)=η+1+ϵYℓm(θ,φ)r=r(\theta,\varphi)=\eta+h(\theta,\varphi)=\eta+1+\epsilon Y_\ell^m(\theta,\varphi), YℓmY_\ell^m are the spherical harmonics of degree ℓ≥0\ell\geq0 and order −ℓ≤m≤ℓ-\ell\leq m\leq\ell (integer numbers), Δh=h−1=ϵYℓm(θ,φ)\Delta h=h-1=\epsilon Y_\ell^m(\theta,\varphi), and V(h)=−1/(2h2)V(h)=-1/(2h^2) is the energy of interaction per unit area among the molecules in the film, assumed to be essentially determined by van der Waals forces, such that V′(h)=ϕ(h)V'(h)=\phi(h). Note that V′′(1)=ϕ′(1)V''(1)=\phi'(1) which for η≫1\eta\gg1 gives ϕ′(1)=−3+O(1/η)\phi'(1)=-3+O(1/\eta), according to Eq. (7). Eqs. (B1)–(B2) must be complemented with the incompressibility condition

ΔV=∫02π∫0π∫ηη+1+ϵYℓmr2sin⁡θ drdθdφ−43π[(η+1)3−η3]=0,(32)\Delta\mathcal{V}=\int_0^{2\pi}\int_0^\pi\int_\eta^{\eta+1+\epsilon Y_\ell^m}r^2\sin\theta\,drd\theta d\varphi-\frac{4}{3}\pi\left[(\eta+1)^3-\eta^3\right]=0, \tag*{(32)}

where V\mathcal{V} is the volume of the liquid shell, which yields the following condition between the first-order and the higher-order contributions, namely,

ϵ(η+1)∫02π∫0πYℓmsin⁡θ dθdφ=−ϵ2∫02π∫0π(Yℓm)2sin⁡θ dθdφ+O(ϵ3)(33)\epsilon(\eta+1)\int_0^{2\pi}\int_0^\pi Y_\ell^m\sin\theta\,d\theta d\varphi=-\epsilon^2\int_0^{2\pi}\int_0^\pi(Y_\ell^m)^2\sin\theta\,d\theta d\varphi+O(\epsilon^3) \tag*{(33)}

Upon introducing (B4) into (B1), the excess surface turns out to be

ΔS=ϵ2∫02π∫0π[−(Yℓm)2+12(1sin⁡θ∂Yℓm∂φ)2+12(∂Yℓm∂θ)2]sin⁡θ dθdφ+O(ϵ3)=ϵ22[ℓ(ℓ+1)−2]+O(ϵ3),(34)\begin{aligned} \Delta S=\epsilon^2\int_0^{2\pi}\int_0^\pi\left[-(Y_\ell^m)^2+\frac{1}{2}\left(\frac{1}{\sin\theta}\frac{\partial Y_\ell^m}{\partial\varphi}\right)^2+\frac{1}{2}\left(\frac{\partial Y_\ell^m}{\partial\theta}\right)^2\right]\sin\theta\,d\theta d\varphi+O(\epsilon^3) \\ &=\frac{\epsilon^2}{2}\left[\ell(\ell+1)-2\right]+O(\epsilon^3), \tag*{(34)} \end{aligned}

where the integrals have been performed leveraging the orthogonality properties of the spherical harmonics and the results found in Barrera et al. [48], namely,

∫Ω(Yℓm)2 dΩ=1,and∫Ω∣r∇Yℓm∣2 dΩ=ℓ(ℓ+1),(35)\int_{\Omega} (Y_{\ell}^{m})^{2}\,\mathrm{d}\Omega= 1,\quad\text{and}\quad\int_{\Omega} \lvert r\nabla Y_{\ell}^{m}\rvert^{2}\,\mathrm{d}\Omega= \ell(\ell+1), \tag*{(35)}

where dΩ=sin⁡θ dθdφ\mathrm{d}\Omega= \sin\theta\,\mathrm{d}\theta\mathrm{d}\varphi is the spherical solid angle element. The O(ϵ2)O(\epsilon^{2})-contribution to the Gibbs energy is

ΔGmolec=ϕ′(1)2(η+1)2ϵ2∫02π∫0π(Yℓm)2sin⁡θ dθdφ=ϕ′(1)2(η+1)2ϵ2.(36)\Delta G_{\mathrm{molec}} = \frac{\phi^{\prime}(1)}{2}(\eta+1)^{2}\epsilon^{2}\int_{0}^{2\pi}\int_{0}^{\pi}(Y_{\ell}^{m})^{2}\sin\theta\,\mathrm{d}\theta\mathrm{d}\varphi= \frac{\phi^{\prime}(1)}{2}(\eta+1)^{2}\epsilon^{2}. \tag*{(36)}

Finally, any perturbation is allowed to grow exponentially if the total excess energy ΔG\Delta G becomes negative. The cutoff is thus found when ΔG=0\Delta G=0 or, equivalently, to second order,

0=Ca−1ΔS+ΔGmolec=ϵ22Ca[ℓ(ℓ+1)−2]+ϕ′(1)2(η+1)2ϵ2=0,(37)0 = Ca^{-1}\Delta S+\Delta G_{\mathrm{molec}} = \frac{\epsilon^{2}}{2Ca}\left[\ell(\ell+1)-2\right]+\frac{\phi^{\prime}(1)}{2}(\eta+1)^{2}\epsilon^{2}=0, \tag*{(37)}

for which the cutoff wavenumber ℓc\ell_{c} is readily obtained as the positive solution (recall that ϕ′(1)<0\phi^{\prime}(1)<0) to the equation

ℓc(ℓc+1)−2+ϕ′(1)Ca(η+1)2=0,(38)\ell_{c}(\ell_{c}+1)-2+\phi^{\prime}(1)Ca(\eta+1)^{2}=0, \tag*{(38)}

which is given by

ℓc=⌊12(9−4Ca ϕ′(1)(η+1)2−1)⌋,(39)\ell_{c}=\left\lfloor\frac{1}{2}\left(\sqrt{9-4Ca\,\phi^{\prime}(1)(\eta+1)^{2}}-1\right)\right\rfloor, \tag*{(39)}

where ⌊⋅⌋\lfloor\cdot\rfloor represents integer part, revealing that perturbations with 0<ℓ<ℓc0<\ell<\ell_{c} and ∣m∣≤ℓ\lvert m\rvert\leq\ell, are linearly unstable.

Solution of the linearised Stokes equations

The perturbed apparent pressure Π^=p^+ϕ′(1)h^\widehat{\Pi}=\widehat{p}+\phi^{\prime}(1)\widehat{h}, where ϕ′(1)=dϕ/dh∣h=1\phi^{\prime}(1)=\left.\mathrm{d}\phi/\mathrm{d}h\right|_{h=1} is specified in the main text, satisfies Laplace’s equation

∂∂r(r2∂Π^∂r)+1sin⁡θ∂∂θ(sin⁡θ∂Π^∂θ)+1sin⁡2θ∂2Π^∂φ2=0,(40)\frac{\partial}{\partial r}\left(r^{2}\frac{\partial\widehat{\Pi}}{\partial r}\right)+\frac{1}{\sin\theta}\frac{\partial}{\partial\theta}\left(\sin\theta\frac{\partial\widehat{\Pi}}{\partial\theta}\right)+\frac{1}{\sin^{2}\theta}\frac{\partial^{2}\widehat{\Pi}}{\partial\varphi^{2}}=0, \tag*{(40)}

and therefore must be of the form

Π^=P(r)Yℓm(θ,φ),withP(r)=A1rℓ+A2r−(ℓ+1),(41)\widehat{\Pi}=P(r)Y_{\ell}^{m}(\theta,\varphi),\quad\text{with}\quad P(r)=A_{1}r^{\ell}+A_{2}r^{-(\ell+1)}, \tag*{(41)}

where A1A_{1} and A2A_{2} are constants of integration. As per the velocity field, suggested by [49] and [50] the problem is rephrased in terms of the perturbed vorticity ω^=∇∧v^\widehat{\boldsymbol{\omega}}=\nabla\wedge\widehat{\mathbf{v}}, which is further decomposed into its poloidal and toroidal components ω^=S+T\widehat{\boldsymbol{\omega}}=\mathbf{S}+\mathbf{T}. By leveraging the fact that ∇∧∇∧ω^=0\nabla\wedge\nabla\wedge\widehat{\boldsymbol{\omega}}=0, as provided by taking the curl of (16), and letting S=∇∧∇∧[S(r)Yℓmer]\mathbf{S}=\nabla\wedge\nabla\wedge[S(r)Y_{\ell}^{m}\mathbf{e}_{r}] and T=∇∧[T(r)Yℓmer]\mathbf{T}=\nabla\wedge[T(r)Y_{\ell}^{m}\mathbf{e}_{r}], where S(r)S(r) and T(r)T(r) are some scalar-valued functions, it can be seen after some algebra that both satisfy the same differential equation, namely,

DℓS=DℓT=0,Dℓ=d2dr2−ℓ(ℓ+1)r2,(42)D_{\ell}S=D_{\ell}T=0,\quad D_{\ell}=\frac{\mathrm{d}^{2}}{\mathrm{d}r^{2}}-\frac{\ell(\ell+1)}{r^{2}}, \tag*{(42)}

which are of the Euler type. The general solutions are S(r)=A3r−ℓ+A4rℓ+1S(r)=A_{3}r^{-\ell}+A_{4}r^{\ell+1} and T(r)=A5r−ℓ+A6rℓ+1T(r)=A_{5}r^{-\ell}+A_{6}r^{\ell+1}, where A3A_{3}, A4A_{4}, A5A_{5} and A6A_{6} are constants of integration. The complete determination of the perturbed velocity field is then available in the form

v^=T(r)Yℓmer+∇∧[S(r)Yℓmer]−∇Φ,(43)\widehat{\mathbf{v}}=T(r)Y_{\ell}^{m}\mathbf{e}_{r}+\nabla\wedge[S(r)Y_{\ell}^{m}\mathbf{e}_{r}]-\nabla\Phi, \tag*{(43)}

up to a scalar gauge Φ\Phi so as to guarantee the solenoidality of v^\widehat{\mathbf{v}}. Such condition yields the Poisson equation

∇2Φ=∇⋅[T(r)Yℓmer],(44)\nabla^{2}\Phi=\nabla\cdot[T(r)Y_{\ell}^{m}\mathbf{e}_{r}], \tag*{(44)}

which can be solved by means of separation of variables by letting Φ=F(r)Yℓm\Phi= F(r)Y_{\ell}^{m}, where the separation function F(r)F(r) satisfies the non-homogeneous equidimensional equation

ddr(r2dFdr)−ℓ(ℓ+1)F=ddr(r2T),(45)\frac{\mathrm{d}}{\mathrm{d}r}\left(r^{2}\frac{\mathrm{d}F}{\mathrm{d}r}\right)-\ell(\ell+1)F=\frac{\mathrm{d}}{\mathrm{d}r}\left(r^{2}T\right), \tag*{(45)}

whose solution is F(r)=A7rℓ+A8r−(ℓ+1)+FpF(r)=A_{7}r^{\ell}+A_{8}r^{-(\ell+1)}+F_{p}, where

Fp=−rℓ∫ηrr^−(ℓ+3)Wddr^(r^2T)dr^+r−(ℓ+1)∫ηrr^2ℓWddr^(r^2T)dr^F_{p}=-r^{\ell}\int_{\eta}^{r}\frac{\hat{r}^{-(\ell+3)}}{W}\frac{\mathrm{d}}{\mathrm{d}\hat{r}}\left(\hat{r}^{2}T\right)\mathrm{d}\hat{r}+r^{-(\ell+1)}\int_{\eta}^{r}\frac{\hat{r}^{2\ell}}{W}\frac{\mathrm{d}}{\mathrm{d}\hat{r}}\left(\hat{r}^{2}T\right)\mathrm{d}\hat{r}

is the particular solution and W=−(1+2ℓ)/r2W=-(1+2\ell)/r^{2} is the Wronskian of the two homogeneous solutions. This, combined with (C4), finally provides the components of the perturbed velocity field v^\hat{\boldsymbol{v}} up to the determination of the shape functions TT, SS, and FF, namely,

v^r=(T−F′)Yℓm,(46)\hat{v}_{r}=(T-F')Y_{\ell}^{m}, \tag*{(46)}
v^θ=Srsin⁡θ∂Yℓm∂φ−Fr∂Yℓm∂θ,(47)\hat{v}_{\theta}=\frac{S}{r\sin\theta}\frac{\partial Y_{\ell}^{m}}{\partial\varphi}-\frac{F}{r}\frac{\partial Y_{\ell}^{m}}{\partial\theta}, \tag*{(47)}
v^φ=−Frsin⁡θ∂Yℓm∂φ−Sr∂Yℓm∂θ,(48)\hat{v}_{\varphi}=-\frac{F}{r\sin\theta}\frac{\partial Y_{\ell}^{m}}{\partial\varphi}-\frac{S}{r}\frac{\partial Y_{\ell}^{m}}{\partial\theta}, \tag*{(48)}

where the primes indicate differentiation with respect to the radial coordinate. The constants of integration A1A_{1} and A2A_{2} can be written in terms of A5A_{5} and A6A_{6} by simply introducing (C4) into the radial component of (16) and matching powers of rr to find A1=−(1+ℓ)A6A_{1}=-(1+\ell)A_{6} and A2=A5ℓA_{2}=A_{5}\ell. The description is then completed by imposing the boundary conditions

v^r=v^θ=v^φ=0,(49)\hat{v}_{r}=\hat{v}_{\theta}=\hat{v}_{\varphi}=0, \tag*{(49)}

as required by non-slippage at the substrate wall r=ηr=\eta and

v^r−ωh^=0,(50)\hat{v}_{r}-\omega\hat{h}=0, \tag*{(50)}
−p^+2∂v^r∂r+ℓ(ℓ+1)−2Ca(η+1)2h^=0,(51)-\hat{p}+2\frac{\partial\hat{v}_{r}}{\partial r}+\frac{\ell(\ell+1)-2}{\mathrm{Ca}(\eta+1)^{2}}\hat{h}=0, \tag*{(51)}
r∂∂rv^θr+1r∂v^r∂θ=0,(52)r\frac{\partial}{\partial r}\frac{\hat{v}_{\theta}}{r}+\frac{1}{r}\frac{\partial\hat{v}_{r}}{\partial\theta}=0, \tag*{(52)}
1rsin⁡θ∂v^r∂φ+r∂∂rv^φr=0,(53)\frac{1}{r\sin\theta}\frac{\partial\hat{v}_{r}}{\partial\varphi}+r\frac{\partial}{\partial r}\frac{\hat{v}_{\varphi}}{r}=0, \tag*{(53)}

at r=η+1r=\eta+1, accounting for kinematic, normal- and tangential-stress equilibrium. Note that by virtue of (C2) and (C7)–(C9), this amounts to imposing boundary conditions on the generating functions PP, TT, SS, and FF. The no-slip condition together with zero tangential stress produce A3=A4=0A_{3}=A_{4}=0 and, consequently, the poloidal part of the velocity perturbation vanishes, i.e., S(r)=0S(r)=0. The problem may be then recast into the system of equations M⋅A=0\mathbf{M}\cdot\mathbf{A}=0 where M\mathbf{M} is a 4×44\times4 matrix and A=(A5,…,A8)T\mathbf{A}=(A_{5},\ldots,A_{8})^{\mathrm{T}} is the vector of constants of integration. Finding nontrivial solutions to this system of equations needs det⁡(M)=0\det(\mathbf{M})=0 thus rendering the dispersion relation in the form D(ω,ℓ,m;η,Ca)=0D(\omega,\ell,m;\eta,\mathrm{Ca})=0 where ω\omega emerges as a temporal eigenvalue for discrete values of ℓ\ell and ∣m∣≤ℓ|m|\le\ell for given η\eta and Ca\mathrm{Ca}.

Functions appearing in the Stokes dispersion relation

The explicit expressions for the functions BiB_{i} in Eq. (17) are given by

B1=[2η2(2η+3)+4η+1](2ℓ+1)η2ℓ(η+1)2ℓ+2(η+1)η4ℓ+3−2η(η+1)4ℓ+3,(54)B_{1}=\left[2\eta^{2}(2\eta+3)+4\eta+1\right](2\ell+1)\eta^{2\ell}(\eta+1)^{2\ell}+2(\eta+1)\eta^{4\ell+3}-2\eta(\eta+1)^{4\ell+3}, \tag*{(54)}
B2=2(η+1)(2ℓ2+1)(ℓ+2)ℓη4ℓ+3,(55)B_{2}=2(\eta+1)(2\ell^{2}+1)(\ell+2)\ell\eta^{4\ell+3}, \tag*{(55)}
B3=3(η+1)4+4(2η+1)2ℓ6+12(2η+1)2ℓ5,(56)B_{3}=3(\eta+1)^{4}+4(2\eta+1)^{2}\ell^{6}+12(2\eta+1)^{2}\ell^{5}, \tag*{(56)}
B4=[4η(η+1)(2η(η+1)+5)+5]ℓ4+2[4η(η+1)(2η(η+1)−5)−5]ℓ3,(57)B_{4}=\left[4\eta(\eta+1)\left(2\eta(\eta+1)+5\right)+5\right]\ell^{4}+2\left[4\eta(\eta+1)\left(2\eta(\eta+1)-5\right)-5\right]\ell^{3}, \tag*{(57)}
B5=ℓ2[4η2(η+1)(η+7)+12η+3],(58)B_{5}=\ell^{2}\left[4\eta^{2}(\eta+1)(\eta+7)+12\eta+3\right], \tag*{(58)}
B6=2ℓη[20−2η(η−6)(η+2)]+10ℓ,(59)B_{6}=2\ell\eta\left[20-2\eta(\eta-6)(\eta+2)\right]+10\ell, \tag*{(59)}
B7=2η(ℓ−1)(ℓ+1)[2(ℓ+2)ℓ+3](η+1)4ℓ+3.(60)B_{7}=2\eta(\ell-1)(\ell+1)\left[2(\ell+2)\ell+3\right](\eta+1)^{4\ell+3}. \tag*{(60)}

A simple regularization of the lubrication model

As noted in [33, 34], the lubrication model (10), and the corresponding dispersion relation (19), become singular in the limit h=ηh = \eta, which is manifested by the in-plane velocities (A5)–(A6) having a factor h−ηh - \eta in their denominators. It is noted therein that this singularity is of purely mathematical origin and does not correspond to any physical mechanism. One way to remedy this is to simply invoke the fact that η≫1\eta\gg1, which ultimately leads to the derivation of the thin-film model (12). The resulting dispersion relation (20) is then regular but is seen to perform quite poorly for order unity values of η\eta, as discussed in the main text.

A particular way of circumventing the singularity of the lubrication dispersion relation (19) and improve on the lack of performance of (20) consists in incorporating information from the kinematic condition (A1) into the boundary conditions for the determination of the in-plane velocities. When integrating Eqs. (A3)–(A4) with the no-slip condition, vθ(η)=vφ(η)=0v_\theta(\eta) = v_\varphi(\eta) = 0, and vanishing tangential stress at r=η+hr = \eta+ h,

∂vθ∂r−vθη+h+1η+h∂2h∂θ∂t=0,(61)\frac{\partial v_\theta}{\partial r} - \frac{v_\theta}{\eta+h} + \frac{1}{\eta+h}\frac{\partial^2 h}{\partial\theta\partial t}=0, \tag*{(61)}
∂vφ∂r−vφη+h+csc⁡θη+h∂2h∂θ∂t=0,(62)\frac{\partial v_\varphi}{\partial r} - \frac{v_\varphi}{\eta+h} + \frac{\csc\theta}{\eta+h}\frac{\partial^2 h}{\partial\theta\partial t}=0, \tag*{(62)}

which is rewritten with advantage taken from the approximation vr≃∂h∂tv_r \simeq\frac{\partial h}{\partial t}, which is exact in linearised form (C11), as dictated by continuity. It is important to emphasize that Eq. (E1) is not asymptotically consistent, since the terms involving the radial velocity are smaller than those involving the polar and azimuthal velocities by a factor on the order of the flow slenderness, which is assumed to be small. The resulting lubrication model becomes

(η+h)2sin⁡θ∂h∂t=∂∂θ[sin⁡θ(G(h,η)∂p∂θ+H(h,η)∂2h∂θ∂t)]+1sin⁡θ∂∂φ(G(h,η)∂p∂φ+H(h,η)∂2h∂φ∂t),(63)(\eta+h)^2\sin\theta\frac{\partial h}{\partial t} = \frac{\partial}{\partial\theta}\left[ \sin\theta\left( G(h,\eta)\frac{\partial p}{\partial\theta} + H(h,\eta)\frac{\partial^2h}{\partial\theta\partial t} \right) \right] + \frac{1}{\sin\theta}\frac{\partial}{\partial\varphi}\left( G(h,\eta)\frac{\partial p}{\partial\varphi} + H(h,\eta)\frac{\partial^2h}{\partial\varphi\partial t} \right), \tag*{(63)}

where

H(η,h)=3h2(h+η)6(h−η)(64)H(\eta,h)=\frac{3h^2(h+\eta)}{6(h-\eta)} \tag*{(64)}

is a new mobility function. Note that Eq. (E3) is still singular when h→ηh \to\eta. However, introduction of normal modes leads to a new dispersion relation given by

ω=(2η+1)ℓ(ℓ+1)[2−Ca ϕ′(1)(1+η)2−ℓ(ℓ+1)]3Ca(1+η)3[2(η2−1)+ℓ(ℓ+1)],(65)\omega= \frac{(2\eta+1)\ell(\ell+1)\left[2-Ca\,\phi'(1)(1+\eta)^2-\ell(\ell+1)\right]} {3Ca(1+\eta)^3\left[2(\eta^2-1)+\ell(\ell+1)\right]}, \tag*{(65)}

which is regular for any value of η\eta. The performance of this dispersion relation is shown in Figure 5 where the optimal growth rate obtained with the regularized model is represented for Ca=0.1Ca = 0.1 as a function of η\eta with a yellow solid line, along with those from the main paper sections. Although the agreement is seen to improve the performance is still quite poor for η≲1\eta\lesssim1, so that relatively thick films must be described with use made of the full equations of motion.

Image not extracted

Figure 5. As in Figure 3 for Ca=0.1Ca = 0.1, but now including the optimal growth rate as predicted by (E5).

References

References

  1. [10]R. J. Braun, Dynamics of the tear film, Annu. Rev. Fluid Mech. 44, 267 (2012).DOI
  2. [11]G. G. Fuller and J. Vermant, Complex fluid-fluid interfaces: rheology and structure, Annu. Rev. Chem. Bio. Eng. 3, 519 (2012).
  3. [12]E. Hermans, M. S. Bhamla, P. Kao, G. G. Fuller, and J. Vermant, Lung surfactants and different contributions to thin film stability, Soft Matt. 11, 8048 (2015).DOI
  4. [13]M. Dey, A. S. V., H. N. Dixit, A. Richhariya, and J. J. Feng, A model of tear-film breakup with continuous mucin concentration and viscosity profiles, J. Fluid Mech. 858, 352 (2019).
  5. [14]O. E. Jensen and J. B. Grotberg, Insoluble surfactant spreading on a thin viscous film: shock evolution and film rupture, J. Fluid Mech. 240, 259 (1992).DOI
  6. [15]D. Halpern and J. B. Grotberg, Dynamics and transport of a localized soluble surfactant on a thin film, J. Fluid Mech. 237, 1 (1992).DOI
  7. [16]O. E. Jensen and J. B. Grotberg, The spreading of heat or soluble surfactant along a thin liquid film, Phys. Fluids A: Fluid Dyn. 5, 58 (1993).DOI
  8. [17]D. Halpern and J. B. Grotberg, Surfactant effects on fluid-elastic instabilities of liquid-lined flexible tubes: a model of airway closure, J. Bio. Eng. 115, 271 (1993).pubmed.ncbi.nlm.nih.gov/8231142
  9. [18]J. B. Grotberg, Pulmonary flow and transport phenomena, Annu. Rev. Fluid Mech. 26, 529 (1994).DOI
  10. [19]D. Halpern, O. E. Jensen, and J. B. Grotberg, A theoretical study of surfactant and liquid delivery into the lung, J. App. Phys. 85, 333 (1998).pubmed.ncbi.nlm.nih.gov/9655794
  11. [20]K. J. Cassidy, D. Halpern, B. G. Ressler, and J. B. Grotberg, Surfactant effects in model airway closure experiments, J. App. Phys. 87, 415 (1999).pubmed.ncbi.nlm.nih.gov/10409603
  12. [21]P.-G. De Gennes, Wetting: statics and dynamics, Rev. Mod. Phys. 57, 827 (1985).
  13. [22]A. Oron, S. H. Davis, and S. G. Bankoff, Long-scale evolution of thin liquid films, Rev. Mod. Phys. 69, 931 (1997).DOI
  14. [23]D. Bonn, J. Eggers, J. Indekeu, J. Meunier, and E. Rolley, Wetting and spreading, Rev. Mod. Phys. 81, 739 (2009).DOI
  15. [24]R. Craster and O. Matar, Dynamics and stability of thin liquid films, Rev. Mod. Phys. 81, 1131 (2009).DOI
  16. [25]R. Blossey, Thin liquid films: dewetting and polymer flow (Springer Science & Business Media, 2012).DOI
  17. [26]L. Schwartz and D. Weidner, Modeling of coating flows on curved surfaces, J. Engng. Math. 29, 91 (1995).DOI
  18. [27]R. V. Roy, A. Roberts, and M. Simpson, A lubrication model of coating flows over a curved substrate in space, J. Fluid Mech. 454, 235 (2002).arxiv.org/abs/patt-sol/9705002
  19. [28]N. Xue and H. A. Stone, Draining and spreading along geometries that cause converging flows: Viscous gravity currents on a downward-pointing cone and a bowl-shaped hemisphere, Phys. Rev. Fluids 6, 043801 (2021).DOI
  20. [29]P. G. Ledda, M. Pezzulla, E. Jambon-Puillet, P.-T. Brun, and F. Gallaire, Gravity-driven coatings on curved substrates: a differential geometry approach, J. Fluid Mech. 949, A38 (2022).
  21. [30]S. Bakhshi, I. V. Roisman, and C. Tropea, Investigations on the impact of a drop onto a small spherical target, Phys. Fluids 19, 032201 (2007).
  22. [31]D. Takagi and H. E. Huppert, Flow and instability of thin films on a cylinder and sphere, J. Fluid Mech. 647, 221 (2010).DOI
  23. [32]G. Balestra, D. Nguyen, and F. Gallaire, Rayleigh-taylor instability under a spherical substrate, Physical Review Fluids 3, 084005 (2018).DOI
  24. [33]D. Kang, A. Nadim, and M. Chugunova, Dynamics and equilibria of thin viscous coating films on a rotating sphere, Journal of Fluid Mech. 791, 495 (2016).DOI
  25. [34]D. Kang, A. Nadim, and M. Chugunova, Marangoni effects on a thin liquid film coating a sphere with axial or radial thermal gradients, Phys. Fluids 29, 072106 (2017).DOI
  26. [35]A. Grigor’ev, M. Munichev, and S. Shiryaeva, Influence of disjoining pressure upon stability in the electric field of a charged liquid film on the surface of a hard core, J. Colloid Interface Sci. 166, 267 (1994).
  27. [36]B. Li, H. Zhao, and X. Feng, Spontaneous instability of soft thin films on curved substrates due to van der waals interaction, J. Mech. Phys. Solids 59, 610 (2011).DOI
  28. [37]J. Colin, Morphological instability of two stressed spherical shells, International journal of solids and structures 44, 3218 (2007).DOI
  29. [38]P. De Gennes, Wetting: statics and dynamics, Reviews of modern physics 57, 827 (1985).
  30. [39]D. Moreno-Boza, A. Martínez-Calvo, and A. Sevilla, Stokes theory of thin-film rupture, Physical Review Fluids 5, 014002 (2020).
  31. [40]I. Napari and A. Laaksonen, Disjoining pressure of thin films on spherical core particles, J Chem. Phys. 119, 10363 (2003).DOI
  32. [41]J. Israelachvili, Intermolecular and surface forces (Academic Press, 1985).
  33. [42]A. Scheludko, Sur certaines particularités des lames mousseuses, ii. stabilité cinétique, épaisseur critique et épaisseur d’équilibre, Proc Koninkl Nederl Akad Wetenschap 65, 87 (1962).
  34. [43]A. Vrij, Possible mechanism for the spontaneous rupture of thin, free liquid films, Discussions of the Faraday Society 42, 23 (1966).DOI
  35. [44]R. Jain and E. Ruckenstein, Stability of stagnant viscous films on a solid surface, J. Colloid Interface Sci. 54, 108 (1976).
  36. [45]E. Ruckenstein and R. K. Jain, Spontaneous rupture of thin liquid films, J. Chem. Soc., Faraday Trans. 2 70, 132 (1974).DOI
  37. [46]W. R. Inc., Mathematica, Version 12.3.1 (2021), champaign, IL.
  38. [47]A. Vrij and J. Overbeek, Rupture of thin liquid films due to spontaneous fluctuations in thickness, Journal of the American Chemical Society 90, 3074 (1968).DOI
  39. [48]R. Barrera, G. Estevez, and J. Giraldo, Vector spherical harmonics and their application to magnetostatics, Eur. J. Phys. 6, 287 (1985).DOI
  40. [49]S. Chandrasekhar, Hydrodynamic and hydromagnetic stability (Clarendon Press, Oxford, 1961).
  41. [50]A. Prosperetti, Viscous effects on perturbed spherical flows, Q. Appl. Math. 34, 339 (1977).DOI

Paper details

vlnc:W_x25vr7fp
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