Soft Matter
View Article Online PAPER View Journal | View Issue
Phase behavior and morphology of multicomponent liquid mixtures† Cite this: Soft Matter, 2019, 15,1297 a ab ac ac Sheng Mao, Derek Kuldinow, Mikko P. Haataja* and Andrej Kosˇmrlj *
Multicomponent systems are ubiquitous in nature and industry. While the physics of few-component liquid mixtures (i.e., binary and ternary ones) is well-understood and routinely taught in undergraduate courses, the thermodynamic and kinetic properties of N-component mixtures with N 4 3 have remained relatively unexplored. An example of such a mixture is provided by the intracellular fluid, in which protein-rich droplets phase separate into distinct membraneless organelles. In this work, we investigate equilibrium phase behavior and morphology of N-component liquid mixtures within the Flory–Huggins theory of regular solutions. In order to determine the number of coexisting phases and their compositions, we developed a new algorithm for constructing complete phase diagrams, based on numerical convexification of the discretized free energy landscape. Together with a Cahn–Hilliard approach for kinetics, we employ this method to study mixtures with N = 4 and 5 components. We report on both Received 8th October 2018, the coarsening behavior of such systems, as well as the resulting morphologies in three spatial dimensions. Accepted 23rd November 2018 We discuss how the number of coexisting phases and their compositions can be extracted with Principal DOI: 10.1039/c8sm02045k Component Analysis (PCA) and K-means clustering algorithms. Finally, we discuss how one can reverse engineer the interaction parameters and volume fractions of components in order to achieve a range of rsc.li/soft-matter-journal desired packing structures, such as nested ‘‘Russian dolls’’ and encapsulated Janus droplets.
1 Introduction a Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ, 08544, USA Phase separation and multi-phase coexistence are commonly b Department of Mechanical Engineering and Materials Science, Yale University, seen in our everyday experience, from simple observations of New Haven, CT 06511, USA the demixing of water and oil to sophisticated liquid extraction
Published on 26 November 2018. Downloaded by Princeton University 2/6/2019 1:01:32 PM. c Princeton Institute of Science and Technology of Materials (PRISM), techniques employed in the chemical engineering industry to Princeton University, Princeton, NJ 08544, USA. E-mail: [email protected], separate certain components of solutions. In non-biological [email protected] † Electronic supplementary information (ESI) available. See DOI: 10.1039/c8sm02045k systems, phase separation has been studied for a long time dating back to Gibbs.1 Very recently, it has been demonstrated that living cells are also mixtures composed of a large number Andrej Kosˇmrlj graduated from the of components, with phase separation behavior reminiscent of University of Ljubljana in 2006. those found in inanimate systems in equilibrium.2–8 This He obtained his PhD in physics process has been shown to drive the formation of membraneless from Massachusetts Institute of organelles in the form of simple droplets,7–13 and even hierarchical Technology in 2011. Before joining nested packing structures.14 Princeton University in 2015, he The physics of binary (N = 2) and ternary (N = 3) mixtures are worked as a postdoctoral fellow at well-understood by now, with binary mixtures comprising Harvard University. He is now an standard material in undergraduate statistical thermodynamics Assistant Professor of Mechanical courses. Given, say, a molar Gibbs free energy of the mixture as and Aerospace Engineering at a function of composition, the presence of coexisting phases Princeton University. His research can be ascertained via the common tangent construction, and interests are at the interface of soft repeating this process at several temperatures, the phase diagram Andrej Kosˇmrlj matter, mechanics and biophysics. can be readily constructed. Similar arguments also hold for He has received the NSF Career ternary15,16 and N 4 3 mixtures, while the construction of phase Award and the Excellence in Teaching Award from the School of diagrams becomes rapidly more challenging, in accordance Engineering and Applied Science at Princeton University. with the Gibbs phase rule,1 which states that the maximum
This journal is © The Royal Society of Chemistry 2019 Soft Matter, 2019, 15, 1297--1311 | 1297 View Article Online
Paper Soft Matter
number of coexisting phases in an N-component mixture is F–H theory and model the spatio-temporal evolution of the N + 2. On the other hand, when N c 1, statistical approaches local compositions. The method to construct local phase indicator for predicting generic properties of phase diagrams become functions is also outlined in this section. Resulting microstructures applicable. for representative 4-component systems are presented in Section 4. In their pioneering work, Sear and Cuesta17 modeled an We also demonstrate how interaction parameters can be tuned to N-component system with N c 1 within a simple theoretical achieve different final packing morphologies of the coexisting approach, which incorporated entropy of mixing terms and phases. In Section 5, we focus on the coarsening kinetics of the interactions between the components at the level of second phase separation process. We examine the validity of the virial coefficients; the virial coefficients were in turn treated as dynamic scaling theory in multicomponent settings and discuss Gaussian random variables with mean b and variance s2. In the the coarsening behavior due to the multiple coexisting phases. special case of an equimolar mixture, their analysis based on In Section 6, we provide guidelines for the design of hierarchical Random Matrix Theory showed that for N1/2b/s o 1, the nested structures, and employ them to design three different system is likely to undergo phase separation via spinodal nested structures in 5-component mixtures. Finally, brief con- decomposition, leading to compositionally distinct phases. cluding remarks can be found in Section 7. On the other hand, for N1/2b/s 4 1, the mixture will likely undergo a condensation transition, which leads to the formation of two phases differing in only density (and not relative compositions). 2 Equilibrium phase behavior These predictions were later confirmed by Jacobs and Frenkel using 2.1 Flory–Huggins theory grand canonical Monte Carlo simulations of a lattice gas model with up to N =16components.18 In subsequent work,19 Jacobs and In this study, we model a dilute solution as a continuum Frenkel argued that multiphase coexistence in biologically-relevant multicomponent incompressible fluid composed of N different mixtures with N c 1 does not result from the presence of a large components, where fi represents the volume fraction of PN number of components, but requires fine tuning of the inter- component i fi ¼ 1 . For simplicity, we only focus on molecular interactions. i¼1 In order to begin to bridge the gap between the well-studied the phase behavior of condensates and solvent is not explicitly binary and ternary systems on one hand, and mixtures with considered in our treatment. Furthermore, we assume that all N c 1 on the other, herein we systematically investigate the components in the condensates have equal density, such that phase behavior and morphology of liquid mixtures with N =4 buoyancy effects can be neglected. and 5 components. We develop a method to construct full First, we briefly review some properties of binary mixtures. phase diagrams of such systems based on free energy convexification According to the Flory–Huggins theory of regular solutions20,21 within the Flory–Huggins theory of regular solutions,20,21 and the free energy density (per volume) is expressed as employ the Cahn–Hilliard22 formalism to study associated domain f (f , f )=c RT[f ln f + f ln f + f f w ], (1) growth and coarsening processes during morphology evolution. FH 1 2 0 1 1 2 2 1 2 12 In order to identify and locate the emerging phases in the where c0 is the total molar concentration of solutes, R is the gas simulations, we employ a combination of principal component
Published on 26 November 2018. Downloaded by Princeton University 2/6/2019 1:01:32 PM. constant, T the absolute temperature, and f1 + f2 = 1 due to 23 24 analysis (PCA) and K-means clustering method to translate local incompressibility. The first two terms in eqn (1) incorporate compositions to phase indicator functions. The phase indicator the entropy of mixing, which favors a homogeneous binary functions are, in turn, employed to quantify the domain growth and mixture. The last term describes the enthalpic part of the free coarsening kinetics. Specifically, characteristic domain sizes for energy. The Flory interaction parameter w12 is related to the each phase are extracted from time-dependent structure factors, pairwise interaction energies oij between components i and j and their behavior is compared against classical theories of as w12 = z(2o12 o11 o22)/(2kBT), where kB denotes the 25–28 coarsening kinetics. Finally, we demonstrate how tuning Boltzmann constant and z is the coordination number.22 When the interfacial energies between phases enables one to engineer w12 o 0, the two different components attract each other and morphologies with a wide range of packing structures, including favor mixing. When w12 4 0, the two components repel each Janus-particle like domains and nested ‘‘Russian doll’’ droplets- other, which can drive the system to demix and form two within-droplets with up to 5 layers. coexisting phases (one enriched with component 1 and one The rest of this paper is organized as follows. In Section 2, enriched with component 2) once the Flory parameter becomes the equilibrium phase behavior of an N-component liquid sufficiently large (w12 4 2), such that enthalpy dominates over mixture is examined within the Flory–Huggins (F–H) theory of the mixing entropy15,16 (see Fig. 1a). regular solutions. An algorithm based on convex hull construc- The Flory–Huggins free energy density in eqn (1) can be tion to compute the phase diagram of the mixture is developed, easily generalized to describe an incompressible liquid mixture and a graph theory based method is employed to determine the with N different components as28 number of coexisting phases corresponding to a given set of "# interaction parameters within the F–H theory and average XN 1 XN composition. In Section 3, the Cahn–Hilliard formalism f ðÞ¼fgf c RT f ln f þ f f w : (2) FH i 0 i i 2 i j ij is employed to both incorporate interfacial effects within the i¼1 i;j¼1
1298 | Soft Matter, 2019, 15, 1297--1311 This journal is © The Royal Society of Chemistry 2019 View Article Online
Soft Matter Paper
Fig. 1 Construction of phase diagrams based on finding the convex hulls of free energy functions for (a) binary and (b and c) ternary mixtures. (a) The original free energy function (black solid line in left) and the convexified one using the common tangent construction (black solid line in right) for a binary
mixture with Flory interaction parameter w12 = 2.31. Red dots and lines correspond to a discrete approximation of the free energy function evaluated on a uniform grid (left) and to the convex hull of the free energy function (right). Red points are projected to the abscissa (composition space). Short projected segments from the convex hull correspond to single phase regions, while long projected segments correspond to two phase ones. (b) Discrete approximation of the free energy function (left) and its convex hull (right). (c) Projected triangles from the original free energy function (left) and from the convex hull (middle). The number of stretched sides for projected triangles corresponds to the number of coexisting phases for the composition points that reside within such triangles. This information is used to construct the ternary phase diagram (right).
The first term describes the mixing entropy and the second many components. Here we describe how a complete phase
term describes the enthalpic part, where wij = z(2oij oii ojj)/ diagram can be obtained via a convex hull construction of the
(2kBT) are the Flory interaction parameters between components i discretized free energy manifold, which is implemented via 33 and j. Note that by definition wii =0. the standard Qhull algorithm. This method was initially The Flory–Huggins theory presented above has been widely introduced by Wolff et al.34 for the analysis of ternary mixtures, used to model mixtures of regular solutions in dilute limit and and here we generalize it to systems with an arbitrary number Published on 26 November 2018. Downloaded by Princeton University 2/6/2019 1:01:32 PM. was also generalized to model polymeric systems.28–32 Now, of components N. according to the Gibbs phase rule,1 there can be as many as N To illustrate the main idea of the algorithm, it is useful to coexisting liquid phases at fixed temperature and pressure, but first recall the phase diagram construction process for binary
the actual number depends on the interaction parameters {wij} mixtures (N = 2). When the Flory parameter is sufficiently large and the average composition {fi}. In the next subsection we (w12 4 2), the free energy density becomes a double well potential a b describe an algorithm for constructing a complete phase diagram with two minima located at f1 and f1 (see Fig. 1a). When the a b for a given set of interaction parameters {wij}, which is based on average composition f1 is between the two minima (f1 o f1 o f1), the convexification of the free energy density in eqn (2). the free energy of the system can be lowered by demixing and a b thus forming two phases a and b with compositions f1 and f1, 2.2 Phase diagram based on the convex hull construction respectively. The volume fractions Za and Zb (Za + Zb =1)ofthe In order to construct a phase diagram, one needs to find the two phases can then be obtained from the lever rule,15,16 such a b convex envelope of the free energy density in eqn (2). For binary that f1 = Zaf1 + Zbf1. mixtures the free energy density depends on a single variable Now, we show how identical information can be obtained
(f1) and the two phase coexistence regions can be identified via via the convex hull construction of the discretized free energy 15,16 the standard common tangent construction. For mixtures landscape. First, we discretize the composition space f1 A [0, 1] with N components, the free energy landscape can be represented as with regular segments and make a discrete approximation of an (N 1)-dimensional manifold embedded in an N-dimensional the free energy function (see Fig. 1a). Then, we construct the space. The regions in composition space that correspond to the P convex hull of the discretized free energy function and we
coexisting phases can in principle be obtained by identifying project it back onto the composition space f1 A [0, 1]. Note common tangent hyperplanes that touch the free energy manifold that the projected segments remain unchanged in the regions a b at P distinct points. This is a very daunting task for mixtures with that correspond to a single phase (i.e. for f1 o f1 and f1 4 f1),
This journal is © The Royal Society of Chemistry 2019 Soft Matter, 2019, 15, 1297--1311 | 1299 View Article Online
Paper Soft Matter
while all the segments between the two free energy minima are Conceptually, it is straightforward to generalize the phase replaced with a single stretched line segment. The stretched diagram construction to mixtures with N 4 3 components. segment of the projected free energy convex hull thus denotes First, we discretize the composition space with regular (N 1)- the two phase coexisting region, where the two ends of the simplexes and make a discrete approximation of the free energy a b segment correspond to the compositions f1 and f1 of the two function. Then, we construct the convex hull of the discretized coexisting phases, respectively. Note that the discretized points free energy function and project it back onto the composition may not exactly coincide with the true free energy minima, but space. The projected (N 1)-simplexes are distorted when they the error can be made arbitrarily small by refining the mesh. correspond to regions with multiple coexisting phases. In Fig. 2 For a ternary mixture we follow the same procedure. First, we display all distinct types of distorted tetrahedra (simplexes) we discretize the composition space with small equilateral for an N = 4 component mixture. Next we demonstrate that triangles and we make a discrete approximation of the free determining the number P of different coexisting phases for energy function (see Fig. 1b). Then, we construct the convex distorted simplexes can be mapped to the problem of counting hull of the discretized free energy function and project it back the number of distinct connected components in a simple graph. onto the composition space. Now, there are in general three To this end, based on our knowledge from ternary mixtures different types of projected triangles: triangles with three short (see Fig. 1c), we make the observation that the two vertices of sides, triangles with two elongated sides, and triangles with simplexes that are connected by a stretched line segment corre- three elongated sides (see Fig. 1c). According to Wolff et al.,34 spond to two distinct phases, while the two vertices that are the three different types of triangles correspond to a single connected by a short line segment correspond to the identical phase regions, 2-phase regions, and 3-phase regions, respec- phase. Now, the vertices of simplexes can be represented as graph tively. For the 3-phase region, the corners of triangles describe vertices. The two simplex vertices i and j are considered con- a the equilibrium compositions {fi } of the phases, where i and a nected (disconnected), i.e. they correspond to the identical phase - - denote the indices of the component and of the phase, respectively. (two distinct phases), when their Euclidian distance 8ri rj8 in For a mixture with average composition {fi}, that resides inside the composition space is smaller (larger) than some threshold D, which we typically set to be slightly larger than the initial mesh such a triangle, the mixture phase separatesP into three phasesP with a size. Note that the threshold needs to be slightly larger, because volume fractions 0 o Za o 1, such that fi ¼ Zafi and Za ¼ 1. a a the convex hull algorithm may return small irregular simplexes in For the 2-phase regions, the two long sides of a projected triangle the 1-phase regions (see Fig. 1c). In practice we find that the are approximations for the tie-lines that connect the two coexisting threshold D needs to be set at about B5 times the initial mesh phases, while the two corners that are connected by the short side size. Thus we define the adjacency matrix Aij for such graph as correspond to the identical phase. For a mixture with an average ( composition that lies inside such a triangle, the mixture phase 1; kk ~ri ~ri D; A separates into the two phases located at the ends of the tie-line. ij ¼ (3) 0; otherwise: Refining this process with arbitrarily small mesh sizes enables 34 one to obtain complete information about ternary phase diagrams The number of distinct phases P for a given simplex is thus (see Fig. 1c). equivalent to determining the number of distinct connected Published on 26 November 2018. Downloaded by Princeton University 2/6/2019 1:01:32 PM.
Fig. 2 Distinct types of stretched tetrahedra, which correspond to regions with different numbers of coexisting phases, resulting from the projection of the free energy convex hull to the composition space, and their respective adjacency matrices A (see text). Vertices with identical colors which are connected with short line segments correspond to the same phase, while vertices with opposite colors that are connected with long line segments correspond to different phases.
1300 | Soft Matter, 2019, 15, 1297--1311 This journal is © The Royal Society of Chemistry 2019 View Article Online
Soft Matter Paper
components for a graph characterized with the adjacency 35 matrix Aij. From graph theory we know that this number is related to the spectrum of the Laplacian of A, which is defined as Lij P= Dij Aij, where Dij is the weight matrix defined as a Dii ¼ Aik and Dij = 0 for i j. The number of distinct k connected components is then equal to the algebraic multi- plicity of the eigenvalue 0. The examples for tetrahedra in N =4 component mixtures are shown in Fig. 2. The locations of vertices also provide approximate values for the compositions a {fi } of each phase a. For a mixture with an average composition {fi} that resides inside a simplex, the mixture phase separates into P coexisting phases with volume fractions 0 o Za o 1, such PP a that fi ¼ Zafi .Thevolumefractions{Za} can be determined by a¼1 36,37 a calculating the pseudo-inverse of the N P matrix F fi as