Despite the simplicity of its molecular unit, water is a challenging system because of its uniquely rich polymorphism and predicted but yet unconfirmed features. Introducing a novel space of generalized coordinates that capture changes in the topology of the interatomic network, we are able to systematically track transitions among liquid, amorphous and crystalline forms throughout the whole phase diagram of water, including the nucleation of crystals above and below the melting point. Our approach, based on molecular dynamics and enhanced sampling / free energy calculation techniques, is not specific to water and could be applied to very different structural phase transitions, paving the way towards the prediction of kinetic routes connecting polymorphic structures in a range of materials.
PACS numbers:Computational structure prediction methods [1, 2] have strongly contributed to the rapid increase of new predicted phases of materials with enhanced properties for applications (see, e.g., Ref. [3]). However, at present, no general approach has been developed for guiding experiments through the pathways connecting stable structures of condensed matter. Moreover, metastable phases are very often involved in phase transitions [4] and sometimes their kinetic stability is very high [5]. Thus, in order to recover the global minimum structure, one needs to find specific routes, by e.g. acting on pressure or temperature, in a way that is not at all trivial to guess [6]. A precise understanding of transition mechanisms and the corresponding kinetics is therefore the key to explain and control the behavior of matter. The case of water is emblematic because several experiments have disclosed connections between stable and metastable phases [7][8][9][10] and recently simulations have highlighted the importance of metastable states in understanding the mechanism of phase transitions and related transformations [11]. A classic example is the connection between the crystalline ice stable at ambient pressure (Ice I), and the low-density amorphous (LDA) and high-density amorphous (HDA) ices: by compressing Ice I up to 10 kbar at ≈ 80 K one obtains HDA instead of Ice VI, [12] , a simulation method that yields the atomic trajectories as a function of time at given thermodynamic conditions, is in principle able to track such transitions. The kinetic barriers are however generally too large to allow an efficient exploration of the configuration space within typical MD timescales. Hence, so far it has been necessary to introduce (i) simplistic interaction models [14] and/or (ii) seeding techniques [15]. Another approach consists in using enhanced sampling techniques that accelerate the occurrence of rare events by focusing on low-dimensional order parameters, also called collective variables (CV) [16]. Yet the CVs available to describe phase transitions are specifically designed for a given type of structural transformation [17][18][19], while no general CV scheme has been proven successful for a wide class of problems, in parti...