Polymorph Selection in Charged Colloids in the Second Nucleation Step C. Patrick Royall Abstract We study polymorph selection in a model of charged colloids, with a focus on the higher-order structure prior to and during nucleation. Specifically, we carry out molecular dynamics simulations of a repulsive Yukawa system with a slightly softened (Weeks-Chandler-Andersen) core. We consider the case where the interaction is long-ranged and the BCC crystal is stable, and also intermediate- and short-ranged cases where the FCC crystal is stable. We use two methods for structure identification, the topological cluster classification (TCC) [A. Malins et al., J. Chem. Phys. 139 , 234506 (2013)] and the bond orientational order parameter analysis of W. Lechner and C. Dellago [ J. Chem. Phys. 129 , 114707 (2008)]. Under conditions of high supersaturation, appropriate to experiments with colloids, we find that the system forms a precursor state in which the particles are hexagonally ordered. That is to say, the precursors are indistinguishable from an HCP crystal using the bond orientational order parameters. This ordering occurs at state points both when the body-centred cubic crystal is the stable phase, and also when the face-centred cubic crystal is stable. In all cases, the stable polymorph forms from the precursor phase in a second stage. Although at freezing, the fluid is very much more ordered when the interactions are short-ranged (when FCC is stable), at the supersaturations where nucleation occurs in our simulations, the higher-order structure of the metastable fluids is almost identical for the long-, short-, and intermediate-ranged systems when measured with the TCC. I Introduction Crystallisation is an everyday phenomenon and although it has been studied for many years, it remains a challenging problem. Crystals may form via homogeneous nucleation, which is a rare event that occurs on a microscopic lengthscale and timescale Sear 2012 . It is hard to access such phenomena in conventional materials, but experiments with colloidal dispersions in which the individual particles can be tracked Gasser et al. 2001 ; Taffs et al. 2012 ; Tan, Xu, and Xu 2014 ; Ivlev et al. 2012 ; Royall et al. 2024 and computer simulation Sosso et al. 2016 play an important role in understanding crystal nucleation. Almost all materials have more than one crystal polymorph with differing thermodynamic stability, depending on the state point. Crucially, the thermodynamic stability may not predict the kinetic stability of the material Sear 2012 . This can have spectacular consequences, for example in the well–known case of the anti–AIDS drug Ritonavir which formed a previously unknown polymorph, rendering it unusable with very significant patient care consequences, not to mention the cost, of $250M Buc̆ar, Lancaster, and Bernstein 2015 ; Bauer et al. 2001 ; Mithen, Callison, and Sear 2015 . Other examples include biomineralisation in which the less stable polymorph of calcium carbonate, aragonite, is formed in a biological setting, yet in the laboratory the stable form calcite is usually found, at least at temperatures appropriate to corals and shells Meldrum and Cölfen 2008 ; Zeng et al. 2017 . Understanding of such complex phenomena can be aided by the use of model systems. Here we consider the Yukawa model with a softened core, which has two polymorphs. This model is well approximated by charged colloids, provided that the degree of electrostatic charge is not too high so that the linear Poisson-Boltzmann approximation in the Derjagiun-Landau-Vervey-Overbeek theory can be used Royall, Leunissen, and van Blaaderen 2003 ; Royall et al. 2006 ; Rios de Anda et al. 2015 . Due to the finite size of the colloidal particles, a hard core is often included, which significantly changes the phase behaviour when the Yukawa contribution is weak or short-ranged Royall et al. 2006 ; Taffs et al. 2012 . Often, (when the Yukawa interactions are strong enough and long-ranged enough that the cores do not come into contact) the hard core Yukawa model gives results that are indistinguishable from those in the absence of a core Hynninen and Dijkstra 2003 . As shown in the phase diagram in Fig. 1 , these polymorphs are FCC in the case that the screening is strong (short range) and BCC when the screening is weak (long range). The triple point (fluid-FCC-BCC coexistence) has also been accessed in experiments, where a rather low surface tension between the polymorphs was found Chaudhuri et al. 2017 . Now Alexander and McTague 1978 argued in general mean-field terms that for simple liquids with a spherically symmetric interaction, BCC is typically to be expected as the first polymorph that forms, and the repulsive Yukawa system forms a suitable testbed for such a prediction. Polymorph selection in Yukawa systems has been studied using computer simulation by Desgranges and Delhommelle 2007 , who found that when the interaction was long-ranged, BCC nucleated, yet when the interaction was short-ranged, in addition to the stable FCC phase, BCC was also found, consistent with the ideas of Alexander and McTague 1978 . BCC was also found in the FCC-stable regime by Krazter and Arnold 2015 and also in experiment Xu et al. 2010 . In some model systems, precursors, i.e. ordered regions with structure distinct from that of the nucleus have been found before the nucleus forms. In particular Russo and Tanaka 2012a found a hexagonal ordering ahead of FCC nucleation in hard spheres, Lechner, Dellago, and Bolhuis 2011 found hexagonal ordering in the Gaussian core model in the case that the stable phase was both FCC and BCC. In the same model, Russo and Tanaka 2012b additionally found a BCC precursor when the stable phase was FCC. In an experimental tour-de-force , Tan et al. Tan, Xu, and Xu 2014 showed evidence for a precursor in experiments on charged colloids, ie a hard core Yukawa system. Russo and Tanaka 2016 also found a precursor in simulations of water. More recently, using a range of order parameters de Jager et al. de Jager, Smallenburg, and Filion 2023 , found no evidence for precursors in a hard core Yukawa system. Mithen et al. Mithen, Callison, and Sear 2015 performed large-scale simulations of the Gaussian Core Model, and found (unlike refs. Lechner, Dellago, and Bolhuis 2011 ; Russo and Tanaka 2012b ) that nuclei of mixed BCC, FCC and HCP composition formed. Clearly, the picture that emerges from this selection of model systems is not entirely consistent. Variations in state point and perhaps order parameter used may play a role, a topic to which we return at the end of this work. Due to their mesoscopic size, colloidal particles exhibit dynamics that are very much slower than in molecular systems Russel, Saville, and Schowalter 1989 . In fact, even without rare-event sampling methods often used to study molecular systems, it is possible to directly compare nucleation in colloids with brute force (unbiased) computer simulations Taffs et al. 2012 , and this is the approach we shall employ. Here we make a detailed study of the structure of nuclei and precursors in a Yukawa system using molecular dynamics simulations. simulations, We consider cases where both the BCC and FCC form the stable polymorph, with a long-ranged and short-ranged Yukawa interaction respectively. We also investigate an intermediate case close to the BCC-FCC phase boundary. We use two order parameters to probe the higher-order structure of the nucleating system. Firstly, we investigate the structure of the supersaturated fluid using the topological cluster classification (TCC), which identifies geometric motifs whose bond network is identical to minimum energy clusters of the variable-ranged Morse potential (Fig. 2 ) Malins et al. 2013 . Secondly, we use the bond orientational order (BOOP) parameter method Steinhardt, Nelson, and Ronchetti 1983 . In particular we implement the variant introduced by Lechner and Dellago 2008 in which second-nearest neighbours are also considered. In the case for both BCC and FCC stable state points, the formation of the nucleus is preceded by a hexagonally ordered structure. This then gives way to the stable polymorph in a second nucleation step. Figure 1: Phase diagram of the hard core-Yukawa system for a contact potential \beta\epsilon_{\mathrm{yuk}}=39 . Reproduced with permission from A.-P. Hynninen and M. Dijkstra, Phys. Rev. E. 68 , 021407 (2003). Copyright 2003 American Physical Society. State points shown are the weakest supercooling that crystallised in this work. Arrows indicate the degree of supercooling with respect to the phase boundary. Violet data is the long-ranged case with BCC the stable polymorph. Blue data is the short-ranged case with BCC the stable polymorph. Green data is the intermediate-ranged case where, for the supersaturations studied here, FCC is the stable polymorph. FCC. This work is organised as follows. In the methodology section (Sec. II ), we first outline the molecular dynamics simulations used (Sec. II.1 ) before describing briefly the topological cluster classification (Sec. II.2 ) and then the bond orientational order parameter used (Sec. III.2 ). The results are presented in Sec. III . This is broken up into the results from the TCC analysis of the higher-order structure of the bulk fluid prior to nucleation in Sec. III.1 , followed by the study of the nucleation process using the BOOP (Sec. III.2 ). The structure of the precursors and the surrounding fluid is considered in Sec. III.3 . Finally we discuss our findings in Sec. IV and conclude in Sec. V . II Methods Figure 2: Higher-order structures identified by the topological cluster classification and crystals. Numbers correspond to the number of particles in the cluster. Letters to the range of the Morse potential \rho_{0} for which these are minimum energy clusters (See Eq. A1 in the Appendix) Doye, Wales, and Berry 1995 . Reproduced from [A. Malins et al., J. Chem. Phys. 139 , 234506 (2013)], with the permission of AIP Publishing. II.1 Molecular dynamics simulation As our model, we use a Yukawa system with a slightly softened core, which is representative of charged colloids Taffs et al. 2012 ; Royall, Poon, and Weeks 2013 . \beta u_{\mathrm{hyuk}}(r)=\beta u_{\mathrm{wca}}(r)+\beta u_{\mathrm{yuk}}(r) (1) where \beta is the inverse of the thermal energy k_{B}T . For the core, we take the Weeks-Chandler-Andersen truncation of the Lennard-Jones interaction Weeks, Chandler, and Andersen 1971 . \displaystyle u_{\mathrm{wca}}(r) \displaystyle=\begin{cases}4\varepsilon_{\mathrm{wca}}[(\frac{\sigma^{\prime}}{r})^{12}-(\frac{\sigma^{\prime}}{r})^{6}]+\varepsilon_{\mathrm{wca}}&\ r\leq 2^{\frac{1}{6}}\sigma^{\prime}\\[2.00749pt] 0&\ r>2^{\frac{1}{6}}\sigma^{\prime}\end{cases} (2) where \varepsilon_{\mathrm{wca}}=10k_{B}T is the interaction energy and r is the separation between two particles of diameter \sigma^{\prime} . The Yukawa potential reads \beta u_{\mathrm{yuk}}(r)=\beta\varepsilon_{\mathrm{yuk}}\frac{\exp\left(-\kappa(r-\sigma^{\prime})\right)}{r/\sigma^{\prime}} (3) where \kappa is the inverse of the Debye screening length. \beta\varepsilon_{\mathrm{yuk}} is the potential at contact. To express the potential at contact, we set the unit of length to be the effective hard sphere diameter, defined according to Barker and Henderson Barker and Henderson 1976 . \sigma_{\mathrm{eff}}=\int_{0}^{\infty}1-\exp\left(-\beta u_{\mathrm{wca}}(r)\right)dr (4) Here \sigma_{\mathrm{eff}}\approx 1.0786 . We use \sigma_{\mathrm{eff}} as our unit of length throughout. Henceforth, we drop the subscript x_{\mathrm{eff}} and define the interaction parameters as \kappa\sigma=2.0,4.0 and 10.0 for long, intermediate, and short-ranged cases respectively.short ranged case. Throughout, we set the contact potential \beta\varepsilon_{\mathrm{yuk}}=39.0 . The phase diagram for this model was determined by Hynninen and Dijkstra Hynninen and Dijkstra 2003 and is shown in Fig. 1 where \phi=\pi\rho/(6\sigma^{3}) is the volume fraction and \rho is the number density. Hynninen and Dijkstra Hynninen and Dijkstra 2003 consider an perfectly hard core. However for our parameters, we see no reason to suppose that our system would behave in a significantly different manner. We use molecular dynamics simulations. While colloids of course exhibit overdamped dynamics, in dense fluids dynamical quantities have been shown to exhibit remarkably little dependence on the particular dynamics used Berthier and Kob 2007 and indeed molecular dynamics simulations have been successful directly compared with colloid experiment Royall, Louis, and Tanaka 2007 . For the nucleation simulations presented here, we used the NPT ensemble. We set the system size N=11664 or 65536 for the long-ranged case \kappa\sigma=2.0 , N=10976 or 87808 for the intermediate-ranged case \kappa\sigma=4.0 , and 16384 or 87808 for the short-ranged \kappa\sigma=10.0 case. These system sizes are “magic numbers” in the sense that BCC and FCC crystals can form perfect crystals for the long-ranged ( N=11664,65536 ) or intermediate- and short-ranged ( N=10976,16384,87808 ) cases respectively. We run the system for up to 3\times 10^{6} Lennard Jones time units, which we take as our unit of time throughout. No qualitative difference was observed between the different system sizes. Renderings are presented for N=10976,11664,16384 data. The simulations were carried out using the LAMMPS package Plimpton 1995 and we performed at least 6 runs for each state point. The system was prepared as a set of random coordinates, minimised under Eq. 2 to remove overlaps between particles before the runs were started. We determined the equation of state for the supersaturated fluid, from which we selected the volume fraction at which to launch the nucleation runs. We find that freezing occurs on the simulation timescale for volume fractions \phi=0.356...0.389 and 0.279...0.300 for the long-ranged ( \kappa\sigma=2.0 ), 0.230..0.284 for the intermediate ( \kappa\sigma=4.0 ) and \phi=0.356...0.389 for the short-ranged ( \kappa\sigma=10.0 ) systems respectively. At higher volume fractions, crystallisation occurs rapidly, in a “spinodal-like” manner, as is the case for hard spheres Zaccarelli et al. 2009 ; Royall et al. 2024 . II.2 Topological cluster classification Figure 3: Population of particles in clusters identified by the topological cluster classification at freezing. Shown are data for the long-ranged system (violet), intermediate (green), short-ranged (blue) and WCA (red). The WCA system is mapped to the freezing point of hard spheres. Figure 4: Population of particles in clusters identified by the topological cluster classification for supersaturated fluids prior to nucleation. Shown are data for the long-ranged system (violet) and short-ranged (blue). The state points shown are \phi=0.3597 and 0.2793 for the long-ranged system and short-ranged systems respectively. Figure 5: Bond orientational order parameters for polymorphs considered. (a) q_{4}-q_{6} representation. (b) w_{6}-q_{6} representation. In (a) and (b), the data are coloured as fluid, grey, BCC, violet, HCP, teal and FCC, blue. FCC at melting (FCC m ) is shown in red. Lines in (b) indicate regions identified with the structures indicated in black type. Grey type refers to data from simulations for each state point. (c) Probability distribution of q_{6} for the HCP (teal) and FCC at melting (red) states. Here the teal line is HCP and the red is FCC at melting. Further details of the state points samples are given in Table 1 in the Appendix. Figure 6: Time dependence of population of particles in different environments as characterised by the bond orientational order parameter. (a) Long-ranged case. Here the initial volume fraction is \phi=0.363 and the pressure is p=93.66k_{B}T/\sigma^{3} . (b) Intermediate-ranged case. Here the initial volume fraction is \phi=0.363 and the pressure is p=15.93k_{B}T/\sigma^{3} . (c) Short-ranged case. Here the initial volume fraction is \phi=0.279 and the pressure is p=8.368k_{B}T/\sigma^{3} . Inset shows the same data zoomed in around the nucleation time. Time is expressed in Lennard-Jones time units and the colours correspond to fluid, grey, BCC, violet, HCP teal and FCC, blue. For a complete explanation of the topological cluster classification Malins et al. 2013 should be referred to. The topological cluster classification identifies target clusters by their bond network, these are polyhedra which are associated with a unique bond topology that corresponds to minimum (free) energy clusters of a given interaction potential. The bond network here is defined with a Voronoi decomposition combined with a distance criterion. As the basic building block for clusters, the algorithm constructs all the three, four and five-membered rings which can be constructed along the bond network. In Fig. 2 , we show minimum energy clusters for the Morse potential. Here the grey particles indicate rings and the yellow particles are spindles Fig. 2 and the red are neither spindle nor ring particles. As shown in Eq. A1 in the Appendix, the Morse potential has a variable range, in that for a given number of particles, the topology of the minimum energy cluster may change for different values of the range parameter \rho_{0} . Here we follow the nomenclature of Doye et al Doye, Wales, and Berry 1995 where clusters corresponding to small values of \rho_{0} take letters towards the start of the alphabet (eg 11A) with progressively larger values of \rho_{0} given letters B.., ie 11B, 11C, etc. The 6Z cluster is the minimum energy cluster of the Dzugutov potential Malins et al. 2013 . For consistency with the BOOP, we set the maximum bond length to the second nearest neighbours, although we note that for these dense systems, the Voronoi construction used will typically dominate in the determination of the bond network Malins et al. 2013 . We set the parameter f_{c}=0.82 which controls the choice between three- and four-membered rings Malins et al. 2013 . In our analysis of the bulk liquids (Figs. 3 and 4 ), we have sampled at least 3\times 10^{6} particles for each state point and so the statistical errors are relatively small with at least 3\times 10^{3} sampled even for relatively rare clusters. II.3 Bond Orientational Order Parameters We have found the bond orientational order parameter (BOOP) analysis to be a reliable measure of particles in BCC environments Ingebrigtsen et al. 2019 and proceed to use the same method here. We also use the BOOP to identify particles in FCC and HCP environments, although very similar results are obtained using the TCC. Here we have followed the method of Lechner and Dellago Lechner and Dellago 2008 who obtained a clear distinction between the crystal structures of interest here by including second nearest neighbours. Q_{lm}\equiv\frac{1}{n_{b}}\sum_{\mathrm{bonds}}Y_{lm}(\mathbf{r}) (5) where Y_{lm}(\mathbf{r}) are the spherical harnonics with polar and azimuthal angles of the bond between a particle and its neighbours with respect to a fixed reference frame. The invariant (with respect to the reference frame) Q_{l}\equiv\sqrt{\frac{4\pi}{2l+1}\sum_{\mathrm{m=-l}}^{l}|Q_{lm}(\mathbf{r})|^{2}}. (6) The third-order invariants W^{\prime}_{l}\equiv\sum_{m_{1},m_{2},m_{3}}\begin{Bmatrix}l&l&l\\ m_{1}&m_{2}&m_{3}\end{Bmatrix}Q_{lm_{1}}Q_{lm_{2}}Q_{lm_{3}} (7) and are normalised as W_{l}\equiv\frac{W^{\prime}_{l}}{\sum_{\mathrm{m}}|\bar{Q}_{lm}(\mathbf{r})|^{3/2}}. (8) Here, in order to include contributions from the second nearest neighbours Lechner and Dellago 2008 , we set the bond length to be the second minimum of the radial distribution function g(r) in the (supersaturated) fluid. In the case of crystal phases we take the fluid at coexistence, or, if the crystal is somewhat more compressed, we take the second minimum of the g(r) at coexistence and scale the range according to (\phi/\phi_{\mathrm{coex}})^{1/3} where \phi is the volume fraction of interest and \phi_{\mathrm{coex}} is that at coexistence. To evaluate the BOOP, we use the BOP code of Wang et al Wang, Teitel, and Dellago 2005 . III Results III.1 Higher-Order Fluid Structure We begin our presentation of the results by considering the supercooled liquid prior to crystallisation. We compare the higher-order structure (as elucidated with the TCC) at freezing for the long-ranged (BCC stable) and short- and intermediate-ranged (FCC stable) cases in Fig. 3 . Data for the long-ranged system (inverse Debye screening length ( \kappa\sigma=2 ) are shown in violet, intermediate ( \kappa\sigma=4 ) in green and the short ranged ( \kappa\sigma=10 ) in blue. We see that there is very much more higher-order structure for the short-ranged system than the long-ranged case, with the intermediate-ranged case lying in between. For comparison, we also show a Weeks-Chandler-Anderson (WCA) system at freezing (which we take as an effective hard sphere volume fraction \phi=0.4917 Royall et al. 2024 ) for the same value of the WCA interaction strength \beta\varepsilon_{\mathrm{wca}}=10 . This in fact has somewhat less higher-order structure than the short-ranged system, but rather more than the long-ranged case. We note that similar behaviour, of a “lack of structure” in a long-ranged system has been seen previously albeit for a different interaction potential and under a different mapping Taffs et al. 2010 . In our simulations, we find that spontaneous nucleation occurs at rather different degrees of compression with respect to the phase boundary, as shown in Fig. 1 . That is \phi_{\mathrm{nucl}}/\phi_{f}=1.73 for the long-ranged case while for the short-ranged system it is just \phi_{\mathrm{nucl}}/\phi_{f}=1.18 , and \phi_{\mathrm{nucl}}/\phi_{f}=1.29 for the intermediate-ranged system where \phi_{f} is the volume fraction of the fluid at freezing Hynninen and Dijkstra 2003 . This corresponds to a pressure scaled by that at freezing of p/p_{f}=3.48 , p/p_{f}=1.63 and p/p_{f}=1.67 for the long-, short- and intermediate-ranged cases respectively. Such a large supersaturation of the long-range system before spontaneous nucleation is observed may seem surprising, but it is compatible with the kinetic phase diagram of the Yukawa model Gispen and Dijkstra 2022 . In colloidal systems, the population of TCC clusters typically increases increases with volume fraction Taffs et al. 2012 ; Royall, Williams, and Tanaka 2018 . Here we find the same, in Fig. 4 , where we show the populations of TCC clusters for simulations at the lowest volume fractions where the system crystallised for the three systems. We sample only the metastable fluid and to be confident that these are not influenced by nucleation, we take data only for t\sigma.\\ \end{cases} (A2) One can compare u_{\mathrm{tm}}(r) to the Yukawa potential with the softened core Eq. 1 and select a value of \rho_{0} which best matches Eq. 1 . We arrive at \rho_{0}\approx 2.5 for the long-ranged case \kappa\sigma=2.0 and \rho_{0}\approx 4.0 for the long-ranged case \kappa\sigma=10.0 . For the full Morse potential, \rho_{0}\approx 6.0 is rather close to the Lennard-Jones interaction. State \phi \kappa\sigma fluid 0.2368 10.0 fluid 0.208 2.0 fluid 0.492 WCA FCC m 0.2468 10.0 FCC 0.35 10.0 HCP 0.661 WCA HCP 0.741 WCA BCC 0.21 2.0 BCC 0.315 2.0 Table 1: State points used to parameterize the BOOP in Fig. 5 . The lower volume fraction for BCC and FCC were taken at melting. Figure 12: Time dependence of population of particles in different environments as characterised by the bond orientational order parameter. (a) Long-ranged case. Here the initial volume fraction is \phi=0.377 and the pressure is p=95.64k_{B}T/\sigma^{3} . (b) Short-ranged case. Here the initial volume fraction is \phi=0.2813 . and the pressure is p=9.564k_{B}T/\sigma^{3} . Time is expressed in Lennard-Jones time units and the colours correspond to fluid, grey, BCC, violet, HCP teal and FCC, blue. Figure 13: TCC analysis of particles in HCP precursors and bulk HCP crystal. Shown is the distribution of the number of TCC clusters n a precursor particle is found in. Here we show data for the intermediate-ranged system (a) for \phi=0.230 and a bulk HCP crystal (b) where \phi=0.661 . Figure 14: Rendering showing how many 8A dodecahedra a particle is in. Short ranged system, \phi=0.279 . Grey particles are precursors. Pink shading denotes the number of 8A a particle is in.