WorksPhysics and Astronomy
Flux-induced condensation in a radial geometry
Abstract
Stirring and shear usually suppress phase separation, but transport through a boundary can induce it. We study a binary mixture that is stable everywhere at rest and fills a two-dimensional annulus. The inner boundary is semipermeable and exchanges material with a permeate held at a fixed chemical potential, so that a steady flux of one component crosses the annulus. Although this driven system carries a steady current and produces entropy, within Model B dynamics its steady state is an equilibrium profile in an effective external field that is logarithmic in the distance from the center. Its strength is not imposed from outside but set by the permeate and the permeability of the boundary. Above a threshold driving flux, a condensed annulus forms around the inner boundary. It is bounded by a stationary front whose radius depends exponentially on the ratio of the bulk driving force to the flux. The front is pinned by the effective field and does not coarsen, and the threshold flux decreases only as the inverse logarithm of the system size. The driving flux shifts the phase lines but leaves the critical temperature unchanged. Interfacial tension gives rise to a critical nucleus and to a fold, at which the condensed state ceases to exist, with a threshold that we obtain in closed form. Numerical solutions confirm these predictions and reveal a hysteresis loop bounded by two saddle-node bifurcations. By symmetry, the theory also describes cavitation around a sink.
Introduction
Flow and phase separation have been studied together for several decades, and in most cases the flow acts against demixing. In simple fluids near a critical point, shear suppresses composition fluctuations and shifts the coexistence curve to lower temperatures, so that a sheared mixture can remain homogeneous below its quiescent critical temperature[1]. Stirring arrests the coarsening of domains and, at sufficient amplitude, restores a homogeneous mixture that would otherwise separate[2]. Flow can also induce demixing, as in semidilute polymer solutions and in the shear banding of complex fluids, but it does so through a coupling of the stress to the composition or to an internal structure[1, 3, 4]. In shear banding, the coexistence of two flowing states shows analogies with phase transitions induced by external fields in equilibrium, and an effective nonequilibrium potential has been proposed to account for them[4].
Transport through a boundary can have the opposite effect. When a boundary passes one component of a mixture and retains another, the retained component accumulates next to it. In membrane filtration, this concentration polarization raises the solute concentration at the membrane surface above its bulk value[5], and fouling sets in above a critical permeate flux[6]. The critical flux has been interpreted as a flux-induced transition from a dispersed polarization layer to a condensed phase at the membrane, and its irreversibility appears as hysteresis when the flux is raised and then lowered[7]. Near production wells in gas–condensate reservoirs, a bank of liquid accumulates where the pressure falls below the dew point, and it impairs production[8]. In drying films and during membrane formation, the loss of solvent through a free surface drives demixing near that surface. Either the mixture is quenched into the two-phase region from the outset, so that the flux shapes the morphology rather than creating it[9], or an initially homogeneous film crosses the binodal first at the surface, and an ordering front then propagates until the whole film has separated[10]. In neither case is there a steady state.
A close analogue exists for heat transport. In liquid near the superfluid transition, a steady heat current leads to a stationary interface between the normal and the superfluid phases. In a finite system, the onset of the two-phase state is first order, with a range of bistability[11]. The interface induced by heat flow has been compared with the one induced by gravity[12], and the heat current also depresses the transition temperature[13]. There, however, the transported quantity is energy rather than the order parameter, and the underlying transition is continuous.
These examples raise a question that can be posed in the simplest conserved description of a mixture, Model B[14], with a square-gradient free energy[15]. Consider a binary mixture that is stable everywhere at rest, occupies a two-dimensional annulus, and exchanges material through its inner boundary with a permeate held at a fixed chemical potential, so that a steady flux of the order parameter crosses the annulus. Can this flux, with no chemical reaction, no coupling to the stress, and no applied field, create a stationary phase boundary? In ordinary phase separation, domains coarsen without limit and no length is selected[16]. A front held at a fixed position is therefore qualitatively different.
We show that the answer is yes, and that the problem has a familiar structure. A single integration of the steady-state equation turns the flux into a spatially varying bias of the chemical potential, which acts as an external field conjugate to the composition. The problem therefore belongs to the class of phase transitions in nonuniform external fields, which includes fluids in gravity[17, 18, 19], mixtures in a centrifuge[20, 21, 22, 23], and liquids in electric field gradients[24, 25, 26, 27, 28]. A localized pump is likewise known to influence a critical fluid over long distances[29]. A general mean-field treatment of such problems, in which the external term in the free energy is the product of a function of the order parameter and a function of position, was given in Ref. 30, where the present flux-driven case appeared briefly as an example. Here, however, the field is generated by transport rather than applied from outside. Its strength is not a control parameter but is fixed by the chemical potential of the permeate and the permeability of the boundary, and the steady state carries a current and produces entropy, even though its composition profile coincides with an equilibrium profile in the effective field. In a two-dimensional radial geometry the field is, moreover, logarithmic in the distance, and this marginal form has consequences that we develop below.
The system is thus driven out of equilibrium, in the manner of boundary-driven diffusive systems, in which reservoirs at different chemical potentials maintain a steady current[31, 32]. For some such models, the stationary state is a Gibbs measure with a linearly varying chemical potential[32]. Below the critical temperature, boundary-driven systems in planar geometry sustain a stationary interface between phases imposed by the two reservoirs; macroscopically, the profile solves a stationary Stefan problem[33]. The current there can even flow up the density gradient[34, 35]. Here, by contrast, the mixture is homogeneous and stable at rest, the condensed phase is created by the flux, and the radial geometry selects a unique front radius.
The resulting picture differs in several respects from ordinary phase separation. Above a threshold flux, a condensed annulus forms around the inner boundary, separated from the surrounding mixture by a stationary front whose radius is selected by the flux. The front is pinned by the field and does not coarsen. The threshold flux decreases only slowly as the system grows, so that the transition retains a memory of the system size. The critical temperature is unchanged by the flux, whereas the curvature of the front gives rise to a critical nucleus and limits the range over which the condensed state exists. The condensed and the front-free states both exist over a range of fluxes bounded by two saddle-node bifurcations, so that the transition is hysteretic. By symmetry, the same theory describes a dilute region that forms around a sink in a dense mixture.
The paper is organized as follows. Section II formulates the model for a partially permeable inner boundary in general, with advection. It then focuses on the case without flow, where the effective field is exactly logarithmic. Section III describes the large-scale composition profiles and the position of the front (Sec. III A), the pressure (Sec. III B), the domain wall and the curvature correction to the front (Sec. III C), numerical solutions (Sec. III D), the shifted binodal lines (Sec. III E), the critical nucleus and the fold (Sec. III F), and metastability and hysteresis (Sec. III G). Section IV discusses the results and their relation to other systems. The case of advection through a rejecting boundary, which describes filtration, will be treated in a separate paper.
General model: radial advection, diffusion, and a partially permeable inner boundary
Geometry and fields
We consider a binary mixture or a liquid–vapor system confined within a two-dimensional annular geometry of depth in the direction, bounded by an inner radius and an outer radius . Fluid is withdrawn through the inner boundary at a fixed volumetric rate . We use for the dimensionless volume fraction of the “heavy” (condensable) component. At the outer boundary, the composition is held at a bath value corresponding to a stable, homogeneous, phase.
For an incompressible mixture, the radial volume-averaged velocity is fixed by conservation,
so that corresponds to inward suction. The composition obeys
where the Onsager transport coefficient (mobility) is in general composition dependent. Model B is recovered for and a constant .
The thermodynamics is described by the Ginzburg–Landau functional
where is a molecular volume, is Boltzmann’s constant, and is the absolute temperature. The bulk free energy density is assumed to be written as a symmetric Landau expansion in , where is the critical composition:
The parameter measures the distance from the critical temperature , and ensures stability. The square-gradient coefficient sets the surface tension and the domain-wall width. Throughout, denotes the chemical potential per molecular volume in units of ,
so that , , and the pressure below all have the same units, and the prefactor of Eq. (4) never appears explicitly.
We take the temperature to be uniform, so that is the same everywhere. This is an approximation. In a steady state, the composition current crosses any phase boundary continuously, so that heat of mixing is absorbed or released there, and the current also dissipates energy. The resulting temperature variations are small when heat conduction is fast compared with mass diffusion. This is the case in liquid mixtures, where the thermal diffusivity () typically exceeds the mass diffusivity () by about two orders of magnitude.
The inner boundary
The inner boundary is in general partially permeable: the withdrawn stream is depleted in the “heavy” component relative to the local mixture. We encode this in a rejection coefficient [36], defined so that the permeate carries the heavy component at composition , where . The total flux there is
Here, describes a fully transparent boundary that passes the mixture at its local composition, and a perfectly rejecting boundary that passes only the light component. The left-hand side of Eq. (7) is the total flux, diffusive contribution included; the right-hand side is a constitutive statement about the composition of the withdrawn stream.
A real membrane also leaks diffusively. In linear response, this adds a flux , where is the chemical potential at the wall, that of the permeate on the inner side of the membrane, and an interfacial transfer coefficient obtained by homogenizing a layer of thickness and mobility . The generalized boundary condition is thus
This is the Kedem–Katchalsky form of the boundary condition[37], with the composition on the feed side in place of the mean concentration on the two sides of the membrane, and the difference in chemical potential in place of the difference in osmotic pressure. Here, and appear as the two independent parameters of the interface. The limit pins , a Dirichlet condition describing a boundary that has no resistance. The remaining conditions are the natural (neutral-wall) ones together with the bath value,
Steady state: the general relation
In steady state with axisymmetry, , where the constant is given by Eq. (8). Equations (3) and (1) then yield
where we defined . The structure is geometric: it follows from in two dimensions and is independent of the constitutive laws. The chemical potential is biased logarithmically in whenever the brace is constant, and this is the origin of exponential laws derived below.
The brace contains two physically distinct drivings: advective delivery through and boundary leakage through . Written in this grouping, Eq. (10) is regular as , and the two mechanisms behave quite differently.
In this paper, we treat the case of no flow, , in which only boundary leakage drives the system and the logarithmic field is exact (Sec. (III)). The opposite case, a perfectly rejecting boundary without leakage (, ) under inward suction (), gives . There, the total flux vanishes, the diffusive flux exactly cancelling the advective delivery, and the heavy component accumulates at the inner boundary: this is concentration polarization. The field is then not logarithmic, since the numerator varies with through ; it reduces to the form treated here only in the weak-polarization limit . The advective problem will be treated elsewhere.
No flow,
With no volume flow, the inner boundary is a semipermeable window onto a permeate held at , and only leakage drives the system. We non-dimensionalize the spatial coordinates by scaling all lengths relative to the inner radius , i.e., , , and drop the primes throughout. For a constant mobility and in axisymmetric polar coordinates, Eq. (10) is written as
and we have used . This equation is subject to the boundary conditions and .
(11) integrates to
where is an integration constant. The forced transport thus acts exactly as a spatially inhomogeneous external field conjugate to . The profiles increase monotonically from as decreases from to : the driving shifts the local chemical potential logarithmically, pushing the system toward liquid–vapor coexistence near the inner radius.
What is not known in advance is the value of the driving . Equation (12) involves , which is part of the solution, so is an eigenvalue determined together with the profile.
Closure. The eigenvalue can be eliminated in closed form to leading order in . Since , the only gradient contribution to at the outer boundary is . Hence and , and, since , Eq. (12) yields a relation that is linear in ,
This error affects the value of and not the form of the solution, which remains logarithmic. If greater accuracy is wanted, may be read off from the computed profile and Eq. (12) iterated.
There are two limits to Eq. (14). For an impermeable boundary, , the limit is : there is no leak and no driving. In the opposite extreme, , one obtains , which means that the mixture at the wall is in equilibrium with the permeate behind the wall.
Two cases follow:
Condensation (liquid injection): and .
In this case, liquid is injected at . The permeate’s chemical potential is positive and high; it decreases across the membrane and continues to decrease with . The solute diffuses outward (). As we show in Sec. III A, for a sufficiently large system, there is a finite distance where , and coexistence with gas occurs. At larger radii, the fluid is gas.
Cavitation (gas suction): and .
Here, the “wellbore” contains gas at a low density and with a negative . The chemical potential increases with . The flux is inward (), with material leaving through the membrane at . The fluid at the outer boundary is liquid. If the system is large enough, coexistence with the liquid bath is possible at a distance where .
Note that cavitation is obtained from condensation by the operation and .
Flux-induced coexistence: bulk domains and interface location
In the case considered here, the logarithmic field, Eq. (13), is exact, and the mixture is in a spatially inhomogeneous external field. As for other such fields[30], the bulk critical point becomes a line of second-order phase transitions that terminates at a surface critical point. Below this point, a line of first-order transitions separates the stable and unstable regions of the phase diagram. The flux term in Eq. (13) is independent of , and therefore the temperature of the surface critical point equals the bulk critical temperature.
Bulk domains. (6) and (13) give
Because scales with molecular dimensions and is macroscopic, the gradient parameter is small, and we work in the strongly segregated limit. The domain then decomposes into large-scale bulk regions separated by thin boundary layers. Far from any interface, the gradient terms are negligible. Since the outer boundary condition makes the gradient term vanish there as well, the large-scale composition obeys the algebraic cubic equation
(16) states that the composition adapts locally and monotonically to the logarithmic field. At low
enough temperature (sufficiently negative ), the argument enters the coexistence region, and the cubic equation for has three real roots. The outer two are the locally stable liquid () and gas () branches; the middle root is unstable. The system follows the branch whose chemical potential lies on the appropriate side of the coexistence value. For the symmetric free energy (5) and for shifted chemical potentials that do not depend on , the coexistence chemical potential is , so the liquid branch is selected where and the gas branch where . Figure 1 shows for increasing . For the two smallest fluxes, the profile remains on the gas branch throughout; above the threshold derived below, it jumps to the liquid branch at a radius that moves outward as grows.
Figure 1. Analytical large-scale composition profiles obtained by solving the cubic equation (16) for the values of in the legend. At large distances, , which corresponds to the gas phase. As decreases, increases, and at the profile jumps to the liquid branch; for the liquid composition increases monotonically. The horizontal dashed lines are the binodal (black, ) and spinodal (red, ) compositions. The two smallest fluxes lie below the threshold of Eq. (19) and show no transition. We used , , , , and .
The first-order transition therefore occurs at the radius where the shifted chemical potential obeys ,
In the bulk gas phase, , and hence . The solution for is
The front radius depends exponentially on the ratio of the bulk thermodynamic chemical potential to the imposed flux. Since is required for a condensed region to exist at all, the minimal flux for condensation is
The threshold vanishes only logarithmically as : the transition is not a local bulk property but retains a memory of the system size and is a consequence of the marginal, logarithmic form of the effective potential in two dimensions. Figure 2 shows this minimal flux in the composition–temperature plane . Close to the bulk binodal line, is small, and it increases with the distance from it.
Figure 2. The minimal flux for condensation at , Eq. (19), in the plane of vapor composition and temperature . The critical point is at ; the binodal is magenta, and the two-phase region, where the mixture separates without any driving, is shaded. Parameter values are the same as in Figure 1.
We emphasize that the phase separation here is pinned at a finite radius and, unlike classical phase separation, does not coarsen. This is a consequence of the dependence of the effective driving on .
Pressure profiles and mechanical balance
The pressure tensor associated with the functional of Eq. (4) is
where the bulk pressure is
With Eq. (5), is expressed in terms of as
In azimuthal symmetry, the derivatives vanish, and the pressure tensor has two components: and , expressed using as
The pressure is spatially nonuniform due to the axial symmetry and the surface-tension forces (the terms proportional to ). The drive by the external flux contributes to the pressure implicitly through the boundary conditions on .
Unlike the equilibrium case, the stress tensor here is not divergence free. We use together with Eq. (15) and obtain
The driving therefore exerts a distributed body force on the fluid, largest where the composition is largest, i.e., in the condensed region.
Domain wall, surface tension, and curvature correction
Near , the composition varies over a distance much shorter than , so the curvature term in Eq. (15) is smaller than by a factor and may be dropped at leading order. Since , Eq. (15) reduces locally to the planar problem , whose solution joining the liquid branch at to the gas branch at is
The wall width is . Far from the wall, the profile relaxes onto the branch values exponentially, at the rate set by the curvature of on the branches, . Equation (17) treats the interface as flat. To use the Gibbs–Thomson shift of the chemical potential, , with the composition jump , we need to calculate the surface tension using the stress components introduced above,
As a crude approximation valid for a cylindrical liquid region of sufficiently large , we use the surface tension of the planar wall, Eq. (25), which yields
Since lengths are measured in units of , Eq. (27) gives in units of . In physical units, , which is independent of : the apparent dependence on through is only that of the length unit.
The corrected criterion replacing Eq. (17) is
with given by Eq. (27). Because increases inward, requiring a positive rather than zero value of at the front moves it to smaller ; far above threshold the shift is
Having obtained the corrected value of , we can now construct the profile valid uniformly across the annulus by matching the inner solution Eq. (25) to the large-scale solution Eq. (16):
where the last term removes the common limit of the two. The same must be used for the branch
switch in and for the center of the tanh; otherwise the composite carries a spurious jump of between the two radii.
Figure 1 depicts the algebraic solutions for , where the phase transition manifests as a mathematical discontinuity. Figure 3 shows the numerical integration of the full differential equation with finite . The analytical interface (circles) lies extremely close to the numerical one (solid line). For the parameters of Figure 3, the numerical front is at , compared with 15.06 from Eq. (18) and 14.52 from the curvature-corrected criterion, Eq. (28); the composite analytical solution, Eq. (30), deviates from the numerical profile by at most 0.046, or 7% of the composition jump .
Figure 3. Composition profiles obtained by solving Eq. (11) including the square-gradient term. The full blue line is the numerical solution, the dashed black curve is the large-scale solution, and the composite approximation, Eq. (30), centered on the curvature-corrected radius of Eq. (28), appears as symbols. As , the interface location converges to the sharp-interface radius of Eq. (18). The horizontal lines are the bulk spinodal (, red) and binodal (, black) values at the given temperature. We used and other parameters as in Figure 1.
Numerical solutions
The steady profiles are obtained by solving Eq. (11) as a two-point boundary-value problem with the three conditions stated there, using a collocation solver (MATLAB bvp5c) on a uniform grid of 350 points spanning . The solution at each flux is used as the initial guess for the next. With and , the interfacial width is , resolved by about seven grid points.
Figure 4(a) shows the emergence of a sharp liquid–vapor interface as the flux is increased. At small flux, the profile is a smooth, monotonically increasing gas density with no interface. At larger flux, a condensed annulus at is separated from the vapor at by a wall of width ; as the flux grows further, the front moves outward, and the composition at the inner wall continues to rise above the binodal value because the local chemical potential keeps increasing.
Figure 4. (a) Composition profiles obtained by numerical integration of Eq. (11) for increasing flux. (b) The corresponding radial pressure from Eq. (23). We used the same parameter values as in Figure 1.
The sequence in Figure 4 was computed by stepping upward in from a front-free initial state, and this matters. The profiles for and show no interface, although both exceed the sharp-interface threshold and the second exceeds the curvature-corrected threshold of Sec. III F as well. Along an upward sweep, the solver follows the front-free branch, which remains a solution while it is metastable, and an interface appears only once that branch ceases to exist. A downward sweep from a condensed state instead retains the front to . This is the hysteresis discussed in Secs. III F and III G, shown explicitly in Figure 5(b): both branches end at saddle-node bifurcations, and the two ends bound the hysteresis loop. These ends are definite properties of the deterministic steady-state problem. They are not, however, the fluxes at which a real system condenses or evaporates: the calculation contains no fluctuations, so it cannot nucleate the stable phase, and in practice the transition occurs somewhere inside the loop.

Figure 5. (a) Front location versus flux. Circles: from the numerical solutions of Eq. (11), obtained in a downward sweep of . Solid line: the sharp-interface prediction, Eq. (18). Dashed: the stable (large) root of Eq. (28). Dotted: the critical-nucleus radius , the small, unstable root. The vertical dash-dotted line marks the fold, Eq. (35), where the two roots merge. (b) Hysteresis. The wall composition for a downward sweep (open circles), which follows the condensed branch until it ends at the fold, and for an upward sweep from a front-free state (filled circles), which follows the front-free branch until it ends at . Dotted vertical segments mark the jumps between the branches. The vertical dotted line is the sharp-kink spinodal estimate, Eq. (39), and the horizontal dotted line the spinodal composition . We used , , and other parameters as in Figure 1.
Figure 4(b) shows the corresponding . Outside the front, the pressure is essentially the uniform bulk value; inside, it rises steeply toward , following Eq. (24). At , departs from its plateau value.
Shifted binodal lines
The forced transport renormalizes the local thermodynamic landscape and enlarges the region of the phase diagram in which a transition can occur. Condensation first appears at the inner boundary. One can read Eq. (19) as setting the temperature for condensation at a given flux . Solving for the temperature variable, we obtain
and therefore, relative to the equilibrium binodal ,
The lines lie above the binodal: condensation is induced at temperatures higher, and vapor compositions smaller, than the equilibrium coexistence. The shift grows linearly with and logarithmically with system size.
Critical nucleus and the fold
Equation (28) is different from Eq. (17) because has a single maximum at
and therefore possesses either two roots or none. When two exist, the larger is the stable front and the smaller, , is the radius of the critical nucleus: the unstable stationary state separating the homogeneous gas from the condensed annulus. Two thresholds follow.
a. (i) Threshold flux. The two roots merge when , a saddle-node bifurcation. With , this condition reads
which can be solved in closed form,
with the Lambert function. This is the threshold flux for condensation in the presence of surface tension. Comparison with the sharp-interface result, Eq. (19), shows that takes the place of . To see that , so that surface tension raises the threshold, note that at the fold, so Eq. (34) reads , i.e.,
The fold lies at a physical radius, , only if , which implies . If instead , the fold would lie inside the bore; then decreases monotonically on , the critical nucleus is always smaller than the bore, and the threshold is given by Eq. (37) below.
b. (ii) Barrier-free condensation. When the critical nucleus is smaller than the inner hole, , the wall itself is a supercritical seed: a liquid film that forms at the wall grows without a curvature barrier. We set in Eq. (28) and get
This is therefore the spontaneous threshold of the curvature mechanism, not the existence threshold; reading it as “the flux at which the front reaches the wall” is incorrect once curvature is retained.
Figure 5(a) compares the front location obtained numerically from Eq. (11), in a downward sweep of , with the analytical predictions. The numerical front always lies inside the sharp-interface radius of Eq. (18), by an amount that grows as the flux decreases, in agreement with Eq. (29). The stable root of the curvature-corrected criterion in Eq. (28) passes through the numerical points over the whole range. The threshold is reproduced as well: the sharp-interface estimate, Eq. (19), gives 0.0050, roughly half the observed value, whereas the fold, Eq. (35), gives 0.0094, and the last numerical front appears at . The agreement is better than might be expected, since near the fold is only about three wall widths and the sharp-interface expansion is marginal there.
Metastability and hysteresis
Equations (19) and (31) describe binodal crossing. The vapor at becomes metastable there, not unstable, and the appearance of the condensed annulus predicted by Eq. (18) presupposes that an interface exists. If the system starts from a homogeneous gas, condensation at this threshold therefore requires a nucleation event, most plausibly heterogeneous at the inner wall.
Spontaneous condensation, instead, requires the local state to reach the spinodal, , at which the gas branch of Eq. (16) ceases to exist. The corresponding chemical potential is
giving a second, higher threshold
and a second family of curves in the – plane, defined implicitly by
Between the two families, the driven gas is metastable: condensation is thermodynamically favored but kinetically blocked. The transition is then hysteretic, since the flux required to nucleate the annulus exceeds the flux below which an existing annulus disappears.
The spinodal family appears as dashed curves in Figure 6. Both families terminate at the same point on the axis, since binodal and spinodal merge at . For the small fluxes shown, much of each dashed curve lies inside the equilibrium two-phase region, so the metastable band outside that region is narrow; it widens as increases.
Figure 6. Stability diagram in the – plane, with the critical point at . Blue: the equilibrium binodal , below which the mixture separates with no driving (shaded). The other curves are the condensation lines, Eq. (31), in the presence of a flux. Below these curves, liquid condenses at small radii, gas occupies large radii, and an interface exists at a finite radius ; above them, the gas remains single phase, though with a nonuniform density profile. Dashed: the spinodal family, Eq. (40); between a solid curve and the dashed curve of the same color, the gas at the inner boundary is metastable (Sec. III G`). Horizontal segments at : the second-order lines joining each surface critical point (circles) to the bulk critical point (square). We used , , and , , and .
Interfacial tension produces a second metastable window. When , a stable annulus exists but is separated from the homogeneous gas by a barrier, since (Sec. III F), and the transition is again hysteretic. This window is the curvature counterpart of the bulk binodal/spinodal pair. The two mechanisms are distinct: the bulk one arises from the loss of local convexity of , the curvature one from interfacial energy, and both operate.
Equation (39) is a sharp-kink estimate of the upper edge of the metastable range, and the full solution shows that it is a lower bound. For the parameters of Fig. 5, Eq. (39) gives , whereas the front-free branch computed in an upward sweep persists to [Figure 5(b)]. The difference comes from the inner boundary layer: the condition keeps the wall composition below its large-scale value, so the wall reaches the end of the front-free branch only at larger flux. At that end, , beyond the spinodal composition ; the thin near-wall region that has crossed the spinodal is stabilized by the gradient term. That the branch ends at a genuine saddle-node, and not through a failure of the numerics, is confirmed by the approach of to its end value as .
For the parameters of Figure 5, the thresholds are thus ordered as , where the last value is the end of the front-free branch. The hysteresis loop extends from the fold to that end, and both barrier-free estimates lie inside it. Above , a liquid film at the wall would grow, but forming it still requires the wall composition to leave the gas branch, which the bulk barrier prevents until that branch ends.
Discussion and outlook
We have shown that, in a binary mixture that is stable everywhere at rest, a steady flux of the order parameter through the inner boundary of an annulus drives the system out of equilibrium and creates a condensed region bounded by a stationary front. With diffusion only and no flow, the steady state is an equilibrium profile in an effective external field that is logarithmic in the distance from the center, and the analysis of Sec. III follows from this mapping.
The logarithmic form of the bias is a property of the geometry rather than of the constitutive laws. In steady state, conservation alone fixes , and the bias inherits the logarithm of the two-dimensional Green’s function (Sec. II). In a planar slab, the same argument gives a bias that is linear in the distance, and the problem reduces to that of a fluid in gravity[17, 18]. In the classification of Ref. 30, in which the external term in the free energy has the form , the present case has and . In the other examples considered there, the field decays away from its source, whereas here its amplitude at the inner boundary grows as , so that the outer boundary cannot be sent to infinity. As a result, the threshold flux, Eq. (19), vanishes only as , and the front radius, Eq. (18), depends exponentially on the ratio of the bulk driving force to the flux.
The same geometric argument applies to viscous flow through a porous medium. If the radial flow obeys Darcy’s law, , the pressure is logarithmic as well,
and a first-order transition that is crossed on lowering the pressure occurs at , where is the transition pressure. This expression has the same form as Eq. (18), with in place of and in place of . Near production wells in gas–condensate reservoirs, the falling pressure produces a bank of condensate that reduces the productivity of the well[8]. This is retrograde condensation, in which liquid appears when the pressure is lowered. The correspondence with the present model is structural only: decompression in a liquid mixture is outside of the current model.
The field changes how the interface displaces. In ordinary phase separation, translating an interface costs no free energy, so domains coarsen and no length is selected[16]. Here, the chemical potential departs from its coexistence value on either side of the front, so a displacement of the front costs free energy, and the front is pinned at a uniquely selected radius (Sec. III A). Arrested coarsening also occurs in active emulsions, where chemical reactions suppress Ostwald ripening[38], and condensates there can be positioned when sources and sinks maintain a concentration gradient of a third, regulating component[39]. In the present system, the same role is played by the transport of the phase-separating component itself, with neither a reaction nor a third component.
The comparison with superfluid helium under heat flow is instructive. There, a heat current sustains a stationary interface between the normal and the superfluid phases[11, 12], but it also depresses the transition temperature[13] and makes the onset of the two-phase state first order, even though the transition itself is continuous[11]. A drive in the bulk can likewise shift the critical temperature: in a driven lattice gas, it rises with the driving field[40]. In the present case, the flux changes neither the critical temperature nor the order of the transition. The effective field couples linearly to the order parameter, so that the second-order line lies at the bulk critical temperature and ends at a surface critical point whose composition, but not whose temperature, is shifted by the flux (Sec. III E). A shift of the critical temperature would require a coupling that is nonlinear in the order parameter, as for dielectric mixtures in nonuniform electric fields[30].
The case of advection through a rejecting boundary, which describes filtration and drying, differs in an essential respect. The diffusive flux must then balance an advective delivery proportional to the local composition (Sec. II), so the bias depends on the composition itself, and the steady state is no longer an equilibrium profile in a fixed external field. In that case, the position of the front must be selected by a criterion that does not rely on a free energy, for example the existence of a stationary interface between two homogeneous states, as for flow-induced coexistence in complex fluids[3]. In membrane science, the critical flux has already been interpreted as a transition to a condensed phase, with hysteresis when the flux is raised and then lowered[7]. In models of colloidal filtration, this transition is placed at the maximum of the osmotic pressure of the dispersion, that is, at its spinodal[41]. In the present diffusive problem, by contrast, the spinodal bounds the hysteresis loop only from above, while the fold, shifted from the binodal by curvature, sets its lower edge. In drying films, a dense skin or a demixing front forms at the evaporating surface[9, 10]. Whether two saddle-nodes bound an analogous hysteresis loop in the advective case will be addressed in a future study.
Interfacial tension generates the nucleation structure of the problem without further assumptions. The curvature-corrected criterion, Eq. (28), has two roots, a stable front and a critical nucleus, which merge at a fold, Eq. (35). Below the fold, no condensed annulus exists. Above the barrier-free threshold, Eq. (37), the critical nucleus is smaller than the inner boundary, which then seeds condensation without a curvature barrier. A similar structure arises around chemically active cores in active emulsions: a core that produces droplet material guarantees nucleation above a critical activity, which increases with the surface tension[38]. There, the source is a reaction, whereas here it is a flux across the boundary.
The fold and the end of the front-free branch bound a hysteresis loop, which, for the parameters of Figure 5(b), spans nearly a factor of two in flux. The loop suggests a direct experimental test. In an annulus whose inner wall is in contact with a permeate of controlled chemical potential, condensation and evaporation should occur at measurably different permeate compositions when the permeate is first enriched and then depleted. Because the calculation contains no fluctuations, the two ends of the loop are limits of metastability; in a real system, condensation will occur inside the loop, at a flux set by the kinetics of nucleation.
The model omits the hydrodynamic consequences of the driving. In the steady state, the stress tensor is not divergence free, Eq. (24), so the flux exerts a body force on the fluid that Model B leaves unbalanced. In the applications discussed above, the condensed phase also impedes the transport that created it: a condensate bank reduces the productivity of a well[8], and a deposit on a membrane reduces the permeate flux[7]. Inclusion of this feedback requires the use of composition-dependent mobility , or coupling the composition to a Darcy or a Stokes equation.
Two further questions are left for future work. The first is the stability of the circular front against azimuthal undulations. The pinning field resists displacements of the front, but a coupling to flow, together with a viscosity contrast between the phases, may produce a fingering instability. The second regards a non-circular inner boundary. Since the chemical potential is harmonic outside the front, the front around such a boundary is, at leading order, a level set of the corresponding harmonic function, and we expect the threshold to be governed by the logarithmic capacity of the cross section.
Acknowledgments
Rudi Podgornik was a great scientist. I heard about him and met him years ago, while I was still a young student. Over the years, we had discussions on various scientific topics. His knowledge was as wide as it was deep. His concise style often clarified seemingly complex issues. Rudi was living proof that one can do excellent, rigorous science while remaining kind, generous, and modest. His support for us, even in times of real danger, will not be forgotten. I am grateful for financial support from the Israel Science Foundation Grant No. 332/24, and thank H. Diamant for insightful discussions and Juny Tsori for many endearing and enjoyable moments.
References
- [1]A. Onuki, “Phase transitions of fluids in shear flow,” J. Phys.: Condens. Matter 9, 6119–6157 (1997).
- [2]L. Ó Náraigh and J.-L. Thiffeault, “Bubbles and filaments: Stirring a Cahn-Hilliard fluid,” Phys. Rev. E 75, 016216 (2007).
- [3]P. D. Olmsted and C.-Y. D. Lu, “Coexistence and phase separation in sheared complex fluids,” Phys. Rev. E 56, R55–R58 (1997).DOI
- [4]G. Porte, J.-F. Berret, and J. L. Harden, “Inhomogeneous flows of complex fluids: Mechanical instability versus non-equilibrium phase transition,” J. Phys. II France 7, 459–472 (1997).DOI
- [5]L. Song and M. Elimelech, “Theory of concentration polarization in crossflow filtration,” J. Chem. Soc. Faraday Trans. 91, 3389–3398 (1995).DOI
- [6]R. W. Field, D. Wu, J. A. Howell, and B. B. Gupta, “Critical flux concept for microfiltration fouling,” J. Membr. Sci. 100, 259–272 (1995).DOI
- [7]P. Bacchin, P. Aimar, and R. W. Field, “Critical and sustainable fluxes: Theory, experiments and applications,” J. Membr. Sci. 281, 42–69 (2006).DOI
- [8]M. A. Sayed and G. A. Al-Muntasheri, “Mitigation of the effects of condensate banking: A critical review,” SPE Prod. & Oper. 31, 85–102 (2016), paper SPE-168153-PA.DOI
- [9]H. Manzanarez, J. P. Mericq, P. Guenoun, and D. Bouyer, “Modeling the interplay between solvent evaporation and phase separation dynamics during membrane,” J. Membr. Sci. 620, 118941 (2021).DOI
- [10]G. A. Buxton and N. Clarke, “Ordering polymer blend morphologies via solvent evaporation,” EPL (Europhysics Letters) 78, 56006 (2007).DOI
- [11]A. Onuki, “Theory of helium under heat flow near the λ point. I. interface of He I and He II,” J. Low Temp. Phys. 50, 433–454 (1983).DOI
- [12]A. Onuki, “Nonequilibrium phase transitions in extreme conditions: effects of shear flow and heat flow,” J. Phys.: Condens. Matter 10, 11473–11490 (1998).DOI
- [13]R. V. Duncan, G. Ahlers, and V. Steinberg, “Depression of the superfluid transition temperature in 4He by a heat current,” Phys. Rev. Lett. 60, 1522–1525 (1988).pubmed.ncbi.nlm.nih.gov/10038062
- [14]P. C. Hohenberg and B. I. Halperin, “Theory of dynamic critical phenomena,” Rev. Mod. Phys. 49, 435–479 (1977).DOI
- [15]J. W. Cahn and J. E. Hilliard, “Free energy of a nonuniform system. I. interfacial free energy,” J. Chem. Phys. 28, 258–267 (1958).
- [16]A. J. Bray, “Theory of phase-ordering kinetics,” Adv. Phys. 43, 357–459 (1994).
- [17]P. Hohenberg and M. Barmatz, “Gravity effects near the gas-liquid critical point,” Phys. Rev. A 6, 289 (1972).DOI
- [18]M. R. Moldover, J. V. Sengers, R. W. Gammon, and R. J. Hocken, “Gravity effects in fluids near gas-liquid critical point,” Rev. Mod. Phys. 51, 79–99 (1979).DOI
- [19]E. Dickinson, C. M. Knobler, V. N. Schumaker, and R. L. Scott, “Gravitationally induced density gradients near the liquid-liquid critical solution point,” Phys. Rev. Lett. 34, 180 (1975).DOI
- [20]W. R. Rossen, H. T. Davis, and L. E. Scriven, “Sedimentation of molecular solutions in the ultracentrifuge. 1. equilibrium phase-behavior,” J. Colloid. Interface Sci. 113, 248–268 (1986).
- [21]Y. Tsori and L. Leibler, “Phase-separation of miscible liquids in a centrifuge,” C. R. Phys. 8, 955–960 (2007).DOI
- [22]T. Zemb, R. Rosenberg, S. Marčelja, D. Haffke, J.-F. Dufrêche, W. Kunz, D. Horinek, and H. Cölfen, “Phase separation of binary mixtures induced by soft centrifugal fields,” Phys. Chem. Chem. Phys. 23, 8261–8272 (2021).DOI
- [23]H. Cölfen, R. Rosenberg, D. Haffke, S. Stemplinger, T. Zemb, and D. Horinek, “Two types of liquid phase separation induced by soft centrifugation in aqueous ethyl acetate using ethanol as cosolvent,” Research 6, 0026 (2023).DOI
- [24]Y. Tsori, F. Turnbullach, and L. Leibler, “Demixing in simple fluids induced by electric field gradients,” Nature (London) 430, 544–547 (2004).pubmed.ncbi.nlm.nih.gov/15282601
- [25]Y. Tsori and L. Leibler, “Phase-separation in ion-containing mixtures in electric fields,” Proc. Natl. Acad. Sci. U.S.A. 104, 7348–7350 (2007).pubmed.ncbi.nlm.nih.gov/17442755
- [26]Y. Tsori, “Colloquium: Phase transitions in polymers and liquids in electric fields,” Rev. Mod. Phys. 81, 1471 (2009).DOI
- [27]S. Samin and Y. Tsori, “Vapor - liquid equilibrium in electric field gradients,” J. Phys. Chem. B 115, 75–83 (2011).DOI
- [28]R. Kroll and Y. Tsori, “Liquid nucleation around charged particles in the vapor phase,” J. Chem. Phys. 155, 174101 (2021).DOI
- [29]Y. Ben Dor, Y. Kafri, D. Mukamel, and A. M. Turner, “Long-range influence of a pump on a critical fluid,” Phys. Rev. Lett. 128, 154501 (2022).DOI
- [30]R. Kroll and Y. Tsori, “Phase lines in mean-field models with nonuniform external forces,” J. Chem. Phys. 157, 231103 (2022).DOI
- [31]B. Schmittmann and R. K. P. Zia, “Driven diffusive systems. an introduction and recent developments,” Phys. Rep. 301, 45–64 (1998).DOI
- [32]G. Eyink, J. L. Lebowitz, and H. Spohn, “Hydrodynamics of stationary non-equilibrium states for some stochastic lattice gas models,” Commun. Math. Phys. 132, 253–283 (1990).DOI
- [33]A. De Masi, E. Presutti, and D. Tsagkarogiannis, “Fourier law, phase transitions and the stationary stefan problem,” Arch. Rational Mech. Anal. 201, 681–725 (2011).DOI
- [34]M. Colangeli, A. De Masi, and E. Presutti, “Microscopic models for uphill diffusion,” J. Phys. A: Math. Theor. 50, 435002 (2017).DOI
- [35]M. Colangeli, C. Giardinà, C. Giberti, and C. Vernia, “Nonequilibrium two-dimensional Ising model with stationary uphill diffusion,” Phys. Rev. E 97, 030103 (2018).DOI
- [36]A. J. Staverman, “The theory of measurement of osmotic pressure,” Recl. Trav. Chim. Pays-Bas 70, 344–352 (1951).DOI
- [37]O. Kedem and A. Katchalsky, “Thermodynamic analysis of the permeability of biological membranes to non-electrolytes,” Biochim. Biophys. Acta 27, 229–246 (1958).pubmed.ncbi.nlm.nih.gov/13522722
- [38]D. Zwicker, A. A. Hyman, and F. Jülicher, “Suppression of Ostwald ripening in active emulsions,” Phys. Rev. E 92, 012317 (2015).pubmed.ncbi.nlm.nih.gov/26274171
- [39]C. A. Weber, D. Zwicker, F. Jülicher, and C. F. Lee, “Physics of active emulsions,” Rep. Prog. Phys. 82, 064601 (2019).DOI
- [40]S. Katz, J. L. Lebowitz, and H. Spohn, “Nonequilibrium steady states of stochastic lattice gas models of fast ionic conductors,” J. Stat. Phys. 34, 497–537 (1984).
- [41]P. Bacchin, D. Si-Hassen, V. Starov, M. J. Clifton, and P. Aimar, “A unifying model for concentration polarization, gel-layer formation and particle deposition in cross-flow membrane filtration of colloidal suspensions,” Chem. Eng. Sci. 57, 77–91 (2002).DOI




