J. J.
Williamson
* and
P. D.
Olmsted
Department of Physics, Institute for Soft Matter Synthesis and Metrology, Georgetown University, 37th and O Streets, N.W., Washington, D.C. 20057, USA. E-mail: johnjosephwilliamson@gmail.com; pdo7@georgetown.edu
First published on 28th September 2015
We study the kinetics governing the attainment of inter-leaflet domain symmetry in a phase-separating amphiphilic bilayer. “Indirect” inter-leaflet coupling via hydrophobic mismatch can induce an instability towards a metastable pattern of locally asymmetric domains upon quenching from high temperature. This necessitates a nucleation step to form the conventional symmetric domains, which are favoured by a “direct” inter-leaflet coupling. We model the energetics for a symmetric domain to nucleate from the metastable state, and find that an interplay between hydrophobic mismatch and thickness stretching/compression causes the effective hydrophobic mismatch, and thus line tension, to depend on domain size. This leads to strong departure from classical nucleation theory. We speculate on implications for cell membrane rafts or clusters, whose size may be of similar magnitude to estimated critical radii for domain symmetry.
![]() | ||
Fig. 1 Partial phase diagram showing competing equilibrium R–R (black) and metastable AR–AR (red) phase coexistences.1 Spinodals enclose the regions of local stability. Cartoons of the dominant inter-leaflet arrangement in each bilayer phase are shown. The grey dotted line illustrates R–AR coexistence, which is briefly discussed in Section III A. Other phase coexistences not considered here are omitted.1 Parameters: Δ0 = 2a, κ = 3a−2kBT, V = 0.6kBT, J = 4a−2kBT, B = 0.48a−2kBT. |
![]() | ||
Fig. 2 (a) Mixed bilayer containing S and U (Lo or gel, and Ld-like) model species, illustrating the locally symmetric (R–R) and locally asymmetric (AR–AR) phase coexistences considered here. (b) Microscopic lattice model for coupled leaflets, which can be coarse-grained (CG) (c) to give the mean-field free-energy density f(ϕt, ϕb) as a function of locally-averaged leaflet compositions, and analysed for kinetics of domain formation with the inclusion of gradient costs for domain boundaries.1 The lattice model can also be directly simulated. |
Using such a model, we have shown how coexistence of antiregistered phases can be kinetically preferred due to the effect of hydrophobic mismatch1 so that, before equilibrating, a quenched bilayer must escape a (typically metastable) locally asymmetric state. The understanding of this novel statistical thermodynamics will allow greater control over artificial membranes, and the underlying interactions are expected to play a role in transmembrane organisation of rafts or clusters in vivo, with possible relevance to signalling13 and anaesthetic action.10 The predicted behaviour constitutes an example of Ostwald's “rule of stages”,14 by which a system will pass through available metastable states on its way to equilibrium. Ostwald's rule is familiar (via different origins) in colloids, metallurgy and drug design.
Immediately after quenching a bilayer to a phase-separating region of parameter space, any spinodal instabilities to which the uniform state is subject compete to determine the dominant initial demixing mode. A bilayer with roughly equimolar composition in each leaflet15 can be subject both to an “R mode” with composition perturbations locally symmetric between leaflets, and a perpendicular AR mode with asymmetric perturbations (Fig. 1). If the AR mode is fastest-growing, spinodal decomposition to AR–AR coexistence occurs first, leading to local asymmetry throughout the bilayer. To equilibrate to R–R from a metastable AR–AR state, the bilayer must undergo nucleation of registered bilayer phases. Hence, three classes of kinetics arise: direct separation into equilibrium phases, equilibration via nucleation out of a metastable state, or trapping in a metastable state. For other overall compositions subject to competing instabilities, the competition of symmetric and asymmetric phases is qualitatively similar,2 though more complex.
In this paper, we focus on the nucleation energetics that determine whether equilibrium domain symmetry is reached from a metastable state. First, we introduce the model and discuss the interpretation of bilayer domain symmetry and asymmetry via phase diagrams with a composition axis for each leaflet, with reference to existing experiment and theory. We then identify the three classes of kinetics in simulation, guided by a linear instability analysis, and develop a theory for the nucleation of registered domains, which captures the interplay of bulk free energy with thickness mismatch occurring at the perimeter of a registered domain. Together with the linear stability analysis, the calculated nucleation energetics are consistent with the simulation results. We find that the effective hydrophobic mismatch between a domain and its surroundings is domain size-dependent, which causes strong departure from classical nucleation theory.
The Hamiltonian is
![]() | (1) |
V ≡ V10 − ½(V00 + V11) quantifies purely intra-leaflet interactions, such as those between headgroups. The “direct” coupling B promotes pairwise R between lipids, nominally by penalising tail structure mismatch (which we treat as implicit in tail length mismatch16) across the midplane. The particular mechanisms responsible for the direct coupling are not crucial to our model, however – for comparison with the literature we can simply estimate an effective strength of the conventional inter-leaflet mismatch energy γ,1 which is shown on Fig. 3 as well.4,5,7,17–19 The hydrophobic “indirect” coupling promotes pairwise AR, by penalising mismatch in the bilayer thickness profile. We also define J ≡ 4
, which appears in the mean-field approximation of eqn (1) used to derive the coarse-grained free energy. κ can be related to the area compression modulus κA,1 and penalises variation from species-dependent ideal length. Weaker κ means the species can more easily adapt their tail length and structure to one another's presence. The mismatch parameter Δ0 ≡
S0 −
U0 is cast as a length, but represents both tail length mismatch and structure mismatch; it couples to both the indirect and direct inter-leaflet couplings, J and B. Once fiducial values of the parameters are set, varying J alone approximates changing the mismatch in tail length but not structure (e.g., adding carbons to one species9), while varying B alone approximates varying the mismatch in tail structure, e.g., unsaturation. We arbitrarily choose
S0 >
U0. The reference total thickness d0 ≡
S0 +
U0 is irrelevant in the absence of an external field acting on bilayer thickness.
![]() | ||
Fig. 3 Red/blue colours and dashed/dotted lines: mean-field theory parameter map showing relative growth rates of AR versus R instability modes for a bilayer comprising equimolar mixed leaflets, from linear stability analysis of initial demixing,1 with Δ0 = 2a, κ = 3a−2kBT, V = 0.6kBT. The equilibrium state is R–R in all cases. Below the “AR instability” line AR–AR coexistence is possible but is metastable. Below the “R instability” line the homogeneous state is not unstable to the R mode, although R–R separation is still the equilibrium state. Overlaid dots from simulation (with V = 0.9kBT) show the average degree of registration λ at the end of the simulation time, λ = 0 (black, AR–AR) to λ = 1 (white, R–R). We identify three kinetic classes discussed in the text (circled 1, 2, 3, bold line marks approximate boundaries). Illustrative simulation snapshot sequences for each class are shown (![]() |
The overall leaflet compositions are conserved, since we do not consider flip-flop or exchange with the solvent, and are given by
![]() | (2) |
![]() | (3) |
λ can vary in the range
|Φt + Φb − 1| ≤ λ ≤ 1 − |Φt − Φb|. | (4) |
Here we focus on symmetric, equimolar overall leaflet compositions Φt = Φb = 0.5. By eqn (4) the degree of registration λ can vary from 0 (full pairwise antiregistration) to 1 (full pairwise registration).
We choose a mismatch parameter Δ0 = 2a ∼ 1.6 nm, somewhat larger than the typical length mismatch in phospholipid mixtures, for which a (registered) phase thickness mismatch ≲2 nm23,24 would imply Δ0 ≲ 1 nm. Large Δ0 couples to both the indirect (J) and direct (B) couplings, increasing the energetic driving forces for both antiregistration and registration and making the competing phases clearer to interpret in simulation. The phenomenology is qualitatively similar upon reducing Δ0,2 with the caveat that Δ0 affects the effective value of γ arising from a given B (eqn (8)). In Section V we mention the dependence of the nucleation theory on Δ0.
We use V = 0.6kBT in the mean-field theory, above the mean-field threshold V0 ≡ 0.5kBT for phase separation in the absence of other couplings, and use V = 0.9kBT in simulation, where the corresponding threshold is Vsim.0 = 0.88kBT due to fluctuations.25 Although we cannot expect precise quantitative agreement between the mean-field theory and simulation, in this regime it appears the exact value of V is not crucial to the kinetics; for example, a value V = 0.9kBT in the mean-field theory would yield a similar predicted landscape of relative R/AR growth rate Δω to that calculated in Fig. 3.2
An alternative definition of registration is sometimes used, which we call “colocalised enrichment”.26,27 This describes a bilayer in which the regions of largest top-leaflet composition ϕt (relative to the average in that leaflet) spatially superimpose on the regions of largest bottom-leaflet composition ϕb. Colocalised enrichment is therefore a property of domain morphology over the entire bilayer, not of an individual bilayer phase. It requires a tie-line of finite positive slope in (ϕt, ϕb) space, so that: (i) both leaflets contain domains of larger and smaller than average ϕt(b); and (ii) the domains of large ϕt belong to the same bilayer phase as the domains of large ϕb and are thus spatially colocalised with them.
R–AR coexistence (Fig. 1 dotted line) can be accessed by a bilayer of asymmetric overall leaflet compositions. R–AR tie-lines have positive slope, and thus exhibit colocalised enrichment, although the AR phase is highly asymmetric in composition. Hence, the colocalised enriched domains in each leaflet reported in ref. 27 and 28 (where the leaflets were separately imaged) are consistent with either an R–R or R–AR tie-line. Quantitative composition information would be required to unambiguously determine which was observed.
In the present work, we consider a bilayer of symmetric overall leaflet compositions for which R–R and AR–AR tie-lines compete. In this case, “registration” versus “colocalised enrichment” are practically equivalent. R–R has positive tie-line slope (colocalised enrichment), and both bilayer phases are compositionally symmetric (registered phases, under our definition). Conversely, AR–AR tie-lines are negatively-sloped so that enrichment in one leaflet colocalises with depletion in the other, and both bilayer phases are compositionally asymmetric (antiregistered phases, under our definition).
The literature is ambiguous. For example, full colocalised enrichment (which can reflect either an R–R or R–AR tie-line) is described in ref. 27 as “registration”. In contrast, in ref. 10, “registration” is also defined as the presence of Lo (or Ld) on both sides of the bilayer, which is more like our definition of R phases versus AR phases. This ambiguity is not surprising because often, as here, the two definitions outlined above are similar. The two concepts can clash for asymmetric overall leaflet compositions28 where R–AR tie-lines can play a role.
In asymmetric supported bilayers in ref. 27, the leaflets were imaged with separate fluorophores, revealing colocalised domains in both leaflets. The same was found in highly asymmetric vesicles in ref. 28. On the face of it, these findings differ from some supported bilayers in ref. 24 which exhibited compositionally asymmetric regions interpreted as domains in only one leaflet. These were inferred from variations in total bilayer thickness without imaging the separate leaflets. Height mismatch smaller than a known value for R–R was measured, indicating either R–AR or AR–AR–R coexistence. R–AR was then inferred by detecting that saturated lipids were predominantly in the top leaflet. The system was interpreted as having domains in only the top leaflet.24
However, it is probable that the asymmetric bilayers in both ref. 27 and 28 and ref. 24 represent an R–AR tie-line. Thus, imaging the leaflets in ref. 24 separately could have shown colocalised enrichment just as in ref. 27 and 28. The top-leaflet gel domains in ref. 24 may have apposed regions in the bottom leaflet weakly more “gel-like” than the average in the bottom leaflet. Similarly, the colocalised enrichment in ref. 27 and 28 does not require that bottom27 (or inner28) leaflet domains are truly Lo – for the R–AR tie-line shown in Fig. 1 they will, like the rest of the bottom/inner leaflet, be dominated by unsaturated lipids. It only requires that they are weakly more “Lo-like” than the average in their leaflet, i.e., that the R–AR tie-line is tilted. Hence, the question of whether domains in one leaflet “induce domains in the other” becomes the question of whether the given tie-line is tilted enough for both leaflets to exhibit detectable domain formation when a different fluorophore is used in each. The degree of R–AR tie-line tilt will depend on the direct inter-leaflet coupling B and hence on molecular features.
In Fig. 3, R–R separation is equilibrium (except for a very small region Ba2/kBT ≲ 0.005). Below the “AR instability” line, AR–AR separation is also possible, but metastable. Below the “R instability” line, the initial state is not subject to R instability although R–R separation remains the equilibrium state. The red and blue colours show the difference in growth rates, Δω, between R and AR instability modes. For example, red signifies a faster-growing AR instability mode, so that AR–AR separation dominates initial demixing.
For physical parameter ranges,1 neither the R or AR mode is trivially dominant, so that moderate changes to lipid tail length mismatch (affecting the effective value of J) or tail structure mismatch (affecting B) can determine whether the initial instability after a quench leads to locally symmetric (R) or locally asymmetric (AR) domains. In addition, the long-lived AR–AR states simulated in ref. 9 and 10 provide prima facie evidence that metastable trapping due to failure to nucleate R domains is possible for physical phospholipids. However, due to small simulation sizes in such studies, this could constitute a stable, not metastably trapped, state, an issue which we address in the Discussion.
Finally we note that, contrary to ref. 30, AR–AR coexistence does not require an exactly equimolar (or equal area fractions) mixture in each leaflet. Firstly, an overall composition away from Φt = Φb = 0.5 can lie on one of a set of AR–AR tie-lines running parallel to the central one depicted in Fig. 1.2 Secondly, even if the overall composition is outside any AR–AR tie-line, AR–AR–R coexistence can occur, in which the presence of some R phase is forced by the composition being far from equimolar, but the R phase coexists with two AR phases.2 In that situation, every region of the bilayer except the R phase is, as for AR–AR, locally asymmetric. Like AR–AR, for typical parameters, AR–AR–R is a metastable state, which could become trapped for strong hydrophobic mismatch or stabilised if domain size is limited (e.g., by simulation size).
This simulated “kinetic phase diagram” agrees semi-quantitatively with the theoretical linear stability analysis of initial demixing (red/blue on Fig. 3): inside the blue region, where R–R coexistence should be accessed directly, the simulation exhibits full registration. If instead the AR mode is fastest (red), we find two possibilities – the bilayer may reach R–R coexistence by nucleation out of the AR–AR state, or remain metastably trapped in AR–AR. Fig. 4a shows the successful formation of registered nuclei, which grow from the boundaries of antiregistered domains. In Fig. 4b, a different random quench in the simulation leads to failure to nucleate despite unchanged parameters, illustrating the stochastic nature of the nucleation process.
Having identified the expected three classes of kinetics, we next model the energetics of nucleating registered domains, to clarify the fates of the metastable AR–AR state (the second and third kinetic classes identified in Fig. 3).
Fig. 4a exhibits typical nucleation of registered domains. Our goal is to model the role of hydrophobic mismatch around the edge of R nuclei in determining whether they successfully grow (Fig. 4a) or decay (Fig. 4b). For the reference (zero free energy) state, we take a registered bilayer at equilibrium with no transmidplane mismatch and domains coarse enough for edges to make a vanishing energy contribution. We then assume that a dominant AR instability mode has led to an initially antiregistered state, which incurs everywhere an energy cost, relative to the registered reference state, due to the direct coupling B. We then introduce a circular R nucleus of radius R, thus removing the direct coupling energy within the domain's area but introducing a thickness mismatch around its perimeter (Fig. 5).
![]() | ||
Fig. 5 Schematic structure of a registered SS nucleus introduced to an antiregistered background. The leaflet thicknesses inside and outside the nucleus are calculated from eqn (6). The midplane (dashed) shifts discontinuously at r = R to maintain smooth outer contours (cf.ref. 9 and 30). S (light) and U (dark) lipids illustrating composition are superimposed. Primes indicate that leaflet thicknesses outside the nucleus differ from those inside. |
In the continuum limit of the lattice model, the free energy is
![]() | (5) |
The leaflets' compositions are reflected in their ideal thicknesses t,b0(r). The actual thickness profiles
t,b(r) adopted by the leaflets minimise the free energy Gcont. The actual thickness difference and total thickness profiles are denoted Δ(r) ≡
t(r) −
b(r) and d(r) ≡
t(r) +
b(r). The calculations underlying the following results are detailed in Appendix A.
![]() ![]() | (6a) |
![]() ![]() | (6b) |
![]() | (7) |
![]() | (8) |
dAR = ![]() ![]() | (9) |
![]() ![]() | (10a) |
![]() | (10b) |
![]() | (11) |
If the composition interface were not sharp one would consider the coupling between composition and thickness gradients. Moreover, one could also allow for sliding into antiregistration at a registered domain boundary, which smears out thickness mismatch.2,9,30 These would not qualitatively alter the fact that the registered nucleus' boundary experiences an energy cost from thickness mismatch, but could help reduce it.
Calculating the total thickness profile after the nucleus is introduced yields
![]() | (12) |
![]() | (13) |
For large ξ, the R nucleus does not reach its ideal thickness even at its centre, r/a = 0. This is an important point; for small nuclei R ∼ ξ, the effective thickness mismatch between the nucleus and its surroundings is less than its value Δ0 in the limit of large domain size (Fig. 6b). The thickness at the centre of the nucleus is given by
![]() | (14) |
After introducing the nucleus, the individual thicknesses t,nuc.(r) and
b,nuc.(r) of the top and bottom leaflets are given by
![]() | (15a) |
![]() | (15b) |
![]() | (16) |
Some illustrative nucleation energy curves are shown in Fig. 7. For CNT in 2D, the ratio ΔG/R would be a straight line of negative slope. Deviation from a straight line indicates deviation from CNT. If ξ is small (weak hydrophobic mismatch compared to stiffness), the behaviour is CNT-like over most of R; moreover the critical radius R* (defined as the value where ΔG(R) is maximised) occurs deep into the CNT-like regime. For larger ξ (stronger hydrophobic length mismatch relative to stiffness) the critical radius no longer occurs in the CNT-like regime, so non-CNT effects arising from nonlocal thickness deformation influence nucleation.
![]() | ||
Fig. 7 Energy ΔG for a registered nucleus of radius R, calculated from eqn (16), with Δ0 = 2a and κ = 3a−2kBT. We vary the decay length ξ by varying ![]() ![]() |
The nucleation theory can be related to the simulations by plotting landscapes of the nucleation barrier ΔG(R*) (Fig. 8) and critical radius R* (Fig. 9). A small barrier clearly facilitates nucleation, although the role of the critical radius is more subtle.35–37 One important factor is that for smaller critical radii, regions randomly enriched in registration from the initial quench are more likely to be near or above the critical radius, so can grow more easily.
![]() | ||
Fig. 8 Simulated kinetic phase diagram (notation as in Fig. 3), overlaid on the theoretical nucleation barrier for an R domain nucleating from an AR state (Δ0 = 2a, κ = 3a−2kBT). The arrowed line shows J/B = 8, as used in Fig. 7. |
![]() | ||
Fig. 9 As Fig. 8 but showing the calculated critical radius for R domain nucleation. |
The predicted energetics for nucleation are consistent with the classes of kinetics identified in Fig. 3. Where the R instability mode is fastest, nucleation energetics are irrelevant since the equilibrium phases are formed immediately. If the AR mode is fastest, a successful subsequent transition to equilibrium R–R coexistence generally occurs where the predicted nucleation barrier and critical radius are smaller. As we have used a relatively large value of the mismatch parameter Δ0 = 2a, we briefly mention the dependence on it.38 The critical radii are independent of Δ0, because Δ0 represents both tail length and structure mismatch so affects both the indirect and direct coupling strengths. Thus the only change to Fig. 9 would be in the effective value of γ shown on the secondary axis, which scales as Δ02 (eqn (8)). On Fig. 8, in addition to the change in effective γ, the nucleation barrier ΔG(R*) scales as Δ02.
If the fastest-growing instability mode is R, spinodal decomposition directly into equilibrium R–R coexistence takes place (class 1 Fig. 3). If the AR mode grows fastest then a metastable AR–AR state forms, and nucleation is required to reach equilibrium (class 2 Fig. 3). Unsuccessful nucleation leaves the bilayer metastably trapped (class 3 Fig. 3).
The key parameters for the initial demixing and subsequent nucleation energetics are the indirect inter-leaflet coupling J via hydrophobic mismatch, favouring antiregistration, and the direct coupling B which favours registration. Physically, the hydrophobic tail length mismatch or structure mismatch would affect the effective values of J or B respectively. Both J and B couple to the stiffness κ, which determines how easily mixed lipids can change their tail length and structure to adapt to one another's presence. Here we have focused on an equimolar mixture in both leaflets such that AR–AR coexistence competes with R–R; the kinetic considerations are qualitatively similar for other overall compositions subject to competing instability modes, in which other states including three-phase coexistence (AR–AR–R or R–R–AR) can enter.2
Nucleation implies a critical radius, which a registered domain must exceed so that the penalty of thickness mismatch at the perimeter does not outweigh the bulk free-energy gain. This does not mean domains automatically become registered beyond a certain size, as is sometimes implied.5,10,39 If nucleation is energetically prohibitive, the mere presence of large coarsening antiregistered domains does not guarantee registered domains will ever form, just as assembling a large volume of supercooled water does not guarantee a supercritical ice nucleus will form. That said, it is intriguing to consider whether some hydrodynamic or curvature-mediated mechanism (neither of which are included here) could dynamically assist very large antiregistered domains to overcome the nucleation barrier and become registered.
In the present simulations it appears that a nucleation barrier ΔG(R*) ≳ 5kBT inhibits nucleation on the simulated timescale for this simulation size. In principle the nucleation rate I (per unit area per unit time) is determined by ΔG(R*) via I = I0exp(−ΔG(R*)/kBT), although the unknown kinetic prefactor I0 severely limits the quantitative utility of such a picture. We also expect the critical radius R* to determine how large random regions of inter-leaflet symmetry during the initial quench must be in order for them to grow. The critical radius also captures a size-dependence of potential relevance to cell membrane rafts and clusters, discussed below.
By examining the effect of thickness mismatch at the boundary of a nucleating registered domain, we showed that the process departs strongly from classical nucleation theory due to deformation of the thickness profile, which reduces the effective line tension of small nuclei. For small nuclei, the deformation essentially spreads over the whole nucleus, reducing the effective hydrophobic mismatch. To our knowledge this behaviour has not been noted in previous theories which tended to focus on domains large enough to treat the boundary as a straight line.30,31 The prediction is supported by a recent molecular simulation of coarsening of a quenched bilayer,33 where registered liquid-ordered and liquid-disordered domains became respectively thicker and thinner as they grew through time.
The simulation method does not include hydrodynamics, which are expected to dominate domain coarsening beyond a lengthscale ∼10−6 m such that the purely diffusive dynamics simulated here would no longer apply.40 However, this lengthscale is far beyond both the size simulated here and the predicted critical radii (Fig. 9). Note that hydrodynamics cannot influence the free-energy landscape of the system, and thus cannot change the competing metastable and equilibrium states.
The phase equilibria of “macroscopic” domains (which coarsen until limited by bilayer size) are not influenced by edge energies and, for most reasonable parameters, we predict that R phases are lower in free energy so that R–R is the equilibrium state in bulk,1 as seen in fluorescence studies of large domains.41,42 At the opposite end of the size spectrum, experiments reveal pairwise antiregistration at the single-lipid level,11,43 as also reported in simulation8 and predicted by our mean-field theory.1 In between these regimes, edge energies can influence the equilibrium state; for example, the AR–AR coexistence in ref. 9 and 10 is probably metastable in the limit of large size but could be stabilised if the simulation box is too small to accommodate a supercritical registered domain.
This has important physical consequences, because cell membrane rafts or clusters are not macroscopic. Small domain size in vivo could be due to elastic repulsion,44 hybrid lipids,45 critical fluctuations,46 active recycling47,48 or another mechanism. In either case, it is crucial to recognise that edge energies could influence the thermodynamic preference and even stabilise an AR–AR type of domain formation, in which one leaflet's local enrichment in longer species (relative to the background in that leaflet) colocalises with a relative depletion of such species in the other leaflet.
Although the cell membrane is maintained out of equilibrium, thermodynamic driving forces can be expected to play a role. The potential importance of this size-dependence is underlined by the fact that the estimated critical radii (Fig. 9) can be of the order of putative lipid raft sizes.49 The basic biophysical question is: if a cluster of longer lipids and proteins exists in one leaflet, does it colocalise a similar cluster in the opposite leaflet to maintain transbilayer structural similarity, or does it choose shorter lipids and proteins to maintain uniform thickness? Our work implies that finite-size effects, metastable states and phase-transition kinetics can be key in determining the answer.
κ(Δ(r) − Δ0*(r)) + 2BΔ(r) = 0, | (A1a) |
![]() | (A1b) |
In the AR background the S and U species' lengths are not equal to their ideal values – the direct coupling B encourages equality of tail length (thus degree of ordering) across the bilayer. Eqn (A1a) is solved by
![]() | (A2) |
After introducing the R nucleus, the difference in ideal leaflet thicknesses is zero within the R nucleus (Δ0*(r) = 0 for r ≤ R), which gives eqn (11). The total thickness profile is found by solving eqn (A1b) (which is expressible as a modified Bessel's equation of order zero). We require that the gradient vanishes at the centre of the nucleus (dd(r)/dr|r=0 = 0), and that the total thickness approaches its ideal value away from the domain (d(∞) = S0 +
U0 = d0). The profile d(r) and its gradient are required to be continuous at r = R in order to match the profiles inside and outside the nucleus. With ideal thicknesses given by eqn (10), this yields eqn (12).
To calculate the energy ΔG required to introduce the registered nucleus, we consider the initial energy GARcont from inserting eqn (7) and (9) into eqn (5), and the energy Gnuc.cont when the nucleus is introduced, from inserting eqn (11) and (12) instead. The energy for introducing the nucleus is then
ΔG = Gnuc.cont − GARcont. | (A3) |
![]() | (A4) |
![]() | (A5) |
Note that the negative term in ΔGin yields the estimated inter-leaflet mismatch energy density γ under our microscopic definition (eqn (8)).1 Near the strong-segregation regime this is a close approximation to the difference in bulk free-energy density between registered and antiregistered phases (which can be used to construct an alternative “macroscopic” definition of γ1), becoming exact in the strong-segregation limit assumed in Section V. However, in principle one could replace this term with the actual free-energy difference in cases where the correspondence between microscopic and macroscopic γ breaks down.1
Assuming the original AR–AR boundary is straight and the R nucleus is a circle, the length of AR–AR boundary removed by an R domain of radius R is equal to 2R. The length of R–AR boundary introduced by the nucleus is 2πR. It is extremely difficult to estimate the line tension caused by V, because (i) the separated phases will not be fully pure and (ii) the interface will relax its energy by smearing the composition change over some finite distance, and (iii) this will be coupled to the thickness gradients. However, an upper bound can be obtained by assuming, as in the main text, that the phases are strongly-segregated and that the compositional interface is sharp. With these assumptions, the line tension of the AR–AR boundary is 2V/a, and the contribution from V to an R–AR boundary is V/a. Subtracting the AR–AR energy contribution and adding the R–AR contribution, the correction to eqn (16) would be:
![]() | (B1) |
We reiterate that eqn (B1) must overestimate the contribution V to nucleation energetics, possibly quite severely, so that the argument outlined here provides only a rough upper bound.
This journal is © The Royal Society of Chemistry 2015 |