Amorphous Mediated Commensurate-Incommensurate Phase Transition in Dense Potassium
Long Zhao,Hongxiang Zong,Xiangdong Ding,Graeme J. Ackland
IF: 3.7
2019-01-01
Physical Review B
Abstract:Solid-solid phase transitions in metals normally procede in an orderly way such that the position of the atoms in the two phases are related by some well-defined transformation path. This leads to particular orientation relationships between the phases. Alkali metals exhibit complex structures under pressure, however, the transformation kinetics and microscopic mechanisms between these are yet to be elucidated. Here, taking potassium as an example, we investigate the phase transformation behavior between close-packed fcc and incommensurate host-guest structures (KIII). We use multi-scale molecular dynamics, with a machine-learned potential which fully reproduces the phase diagram and known phase transitions of potassium. We find that no straightforward, low energy path exists: previously proposed mechanisms having impossible high transformation barriers. The fcc to KIII transition occurs via a complex intermediate state which persists for many picoseconds. Any phase transition could occur via an amorphous intermediate, so the free energy of the amorphous state places an upper bound on the transformation barrier, and any mechanism involving an intermediate state of higher free energy than amorphous can be precluded. Solid–solid transitions are the most common type of structural phase transitions [1, 2], and thus of central interest in metallurgy and crystallography. Initial and final crystal orientations are often known, however, the transformation itself cannot be observed experimentally at the atomiclevel, which has resulted in many controversies over the underlying transformation pathways [35]. The second law of thermodynamics means than a crystal can spontaneously transform to another structure with lower free energy. In the simplest case, there is no barrier and the transformation follows a path of ever-decreasing free energy [6]. In other cases, a transition barrier separates the two structures [7-9], which must be overcome by thermal fluctuations, initiating nucleation and growth. In both cases, a limited number of orientation relationships between the two phases are possible, as determined by the symmetry [10-12]. Transformation pathways typically concern the relationship between neighboring atoms: where the relationship between phases generates long ranged strain this must be relaxed either by twinning or other plasticity [13]. However, there is nothing in thermodynamics to define the rate at which such transformation happens. Typically, only those transformations which do not require large atomic rearrangements, or generate long ranged strains, will occur quickly [14, 15]. Consider an experiment, where the driver for the transformation, e.g. pressure, is increased steadily until a phase A becomes unstable with respect to B. If the A→B transformation is slow, A may persist metastably and pressure may continue to increase until A becomes unstable with respect to third phase C. If the A→C transformation is fast enough, the material may transform first to the metastable C phase, rather than B. Metastable phase C might then be observed, despite being thermodynamically unstable at all temperatures and pressure. This depends on the free energies, but also on the rate of pressure increase. The existence of such intermediate phases could be reproducibly observed in experiments with fast pressure increase through quasi-equilibrium states, such as shock waves or ramp compression [9, 16-18]. Elemental metals which crystallize in the closely-packed atomic arrangements, such as bodycentered-cubic (bcc), face-centered-cubic (fcc) and hexagonal closed-packed (hcp) [10, 19] provide the simplest examples of solid-solid transitions. Their kinetic pathways often follow a displacive mechanism (characterized by atomic displacement within a unit-cell, see e.g. pathway I in Fig. 1(a)) [12] or a diffusionless martensitic transformation with atoms moving in concert [10]. These pathways are strongly related to the symmetries of the parent and the product crystals [20, 21]. In the absence of a group-subgroup relationship [9], the transitions may go through one or more intermediate metastable stages corresponding to common subgroups of the initial and final structures [10, 22]. By contrast solid-amorphous transformations require no strain alleviation or symmetry-constrained orientation relation; however, because the amorphous phase is never the most thermodynamically stable phase, the energy barrier to transformation is high, and the transform proceeds by a slower process of nucleation and growth. Temperature-driven transformations in elements are fairly rare, but under pressure most elements undergo a series of structural phase transitions that show diverse and sometimes unexpectedly complex structures [23-25]. For example, potassium adopts the bcc crystal structure at ambient conditions, and on compression transforms to the fcc structure via a standard martensitic pathway [10]. Compressing further results in the emergence of complex incommensurate host-guest (h-g) structures (named KIII), which consist of two sub-lattices with 1D atomic chains “guest” located in channels within a zeolite-type “host” structure. So far, experimental measurements [26-28] as well as DFT calculations [29, 30] have mainly focused on thermodynamics, crystal and electronic structures of host-guest structures. The microscopic kinetics of their transition pathways and associated mechanisms remain poorly understood, however the tendency to form large crystals of K-III [31-33] suggests that the transformation is slow. Martensitic transitions imply a group-subgroup relationship between two phases, however, the h-g structures are best described using a four dimensional space group, confusing the issue of whether displacive/martensitic [34, 35] are possible, or if a quite different nucleation process is required. In this Letter, we use a combined machine learning technique and atomic simulation approach to demonstrate, for the first time, that the microscopic kinetics of the closely packed fcc incommensurate K-III transition in potassium occurs via a two-step nucleation process involving an intermediate amorphous state in which the atomic displacement during the transition can exceed the unit cell length. This mechanism arises from the high entropy, which makes the pathway more favorable than energy-minimizing displacive mechanisms. The interatomic interactions in potassium were described by a machine learning interatomic potential (MLIP), which is based on high-throughput quantum-mechanical (QM) calculations [36, 37]. In a previous work [38], we have shown that the MLIP can accurately describe all the different phases in the phase diagram, including the fcc and h-g. Additional information about the ML potential, in particular the choice of combined descriptors for atomic environments and learning set, is provided in Supplemental Material [39]. Here we utilized the ML potential to construct the potassium phase diagram. A series of MD simulations on pressure-temperature grids using the micro-canonical ensemble were used to find the stable structures, as shown in Fig. 2. At pressures below 20 GPa, the stable region of bcc and fcc lattices at low temperatures agrees with experiments and previous theoretical calculations [27, 29]. The model also reproduces the melt line with a maximum and minimum by the Z-method [52], at very similar pressures to that seen in experiment, and slightly higher temperatures. Between 20 and 44 GPa the simulations reproduce not only the h-g structures, but also the IIIaIIIb-IIIa chain-ordering transformations, and the chain-melting phenomenon, in line with the experimental situation [28]. To our knowledge, this is the first interatomic potential with only one set of parameters that captures the incommensurately host-guest structures and the pressuretemperature phase diagram of potassium, or indeed any other material with such a range of phases. Since the MD simulations based MLIP reproduce the phase stability, the microscopic kinetics of the fcc h-g transition pathways can be investigated. The isothermal-isobaric (NPT) simulation of K single crystals starting from KIII lattices is first performed. Figure 1(b) shows the change in potential energy plotted as a function of simulation time for a typical MD trajectory at T = 300 K (which is 200 K below the melting temperature) and P = 17.0 GPa. We find the potential energy curve possesses a “double-step” profile. A close examination of the microstructure shows that the steps in the energy profile result from a two-step structural transformation process. The first drop in the energy (~137.3 meV/atom) is due to the formation of a metastable phase with a disordered atomic arrangement. Then, fcc lattices nucleates and grows at the expense of the metastable phase, giving rise to the second decrease of the potential energy (~45.5 meV/atom). The simulation takes up to ~220 ps for the entire KIII→fcc transformation to complete. Here, we note that the simulation runs for about 190 ps before the nucleation of the disordered metastable phase, indicating mechanical stability of the initial metastable KIII structures. Figure 1(c) shows the radial distribution functions (RDFs) g(r) of corresponding structures during the transition. The g(r) at the metastable state is quite different from either fcc or KIII, and it only includes only one sharp peak at R = 3.3 Å. This lack of structure or long-ranged order is characteristic of a liquid or amorphous material. A similar situation has been reported on bismuth experiments, which show a metastable liquid in the pressure-temperature region of supercooled liquid upon decompression from host-guest phases [18]. Therefore, we compared the g(r) of the potassium intermediate state with that of a typical liquid as well as glass at the same pressures. As shown in the inset of Fig. 1(c), the similarity between the metastable phases (red curve) and liquid structure (gree