Force Field-Agnostic Phase Classification of Zeolitic Imidazolate Framework Polymorphs
Emilio Méndez, Léna Triestram, Dune André, François-Xavier Coudert, Rocio Semino · 2026
Read Force Field-Agnostic Phase Classification of Zeolitic Imidazolate Framework Polymorphs by Emilio Méndez, Léna Triestram, Dune André, François-Xavier Coudert, Rocio Semino (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-09084.txt.
Introduction
Méndez and four colleagues build a force-field-free phase sorter for zeolitic imidazolate frameworks. These metal-organic solids link metal centers to imidazole ligands in zeolite-like shapes. Zinc-based versions show many forms, shifting with recipe and heat or pressure. The team finds low-dim markers already sort phases well, while high-dim markers do even better. Training on setups from other force fields removes force-field bias and lifts scores and reach. The sorters reveal details of the ZIF-4-cp to ZIF-4-cp-II change.
Catalog record
| CREATOR | Emilio Méndez, Léna Triestram, Dune André, François-Xavier Coudert, Rocio Semino |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50500 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.09084
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Force Field-Agnostic Phase Classification of Zeolitic Imidazolate Framework Polymorphs Emilio Méndez, Léna Triestram, Dune André, François-Xavier Coudert, Rocio Semino Abstract Zeolitic Imidazolate Frameworks (ZIFs) are a family of metal–organic frameworks that feature metal centers tetrahedrally linked to imidazole-based ligands and adopt zeolite-like topologies. ZIFs formed by Zinc cations and imidazolate linkers exhibit a remarkable degree of polymorphism, which can be modulated by varying synthesis parameters or thermodynamic conditions (i.e., temperature and pressure). Computer simulations provide a unique way of studying these materials and their phase transitions from the microscopic standpoint, revealing their underlying molecular mechanisms. However, studying these mechanisms requires to be able to classify the phase of each molecular entity in an agnostic and automatic fashion, which is particularly challenging when the two phases involved are structurally very similar. In this work, we systematically study neural network classifiers to classify ZIF phases on-the-fly during molecular dynamics simulations. We test a variety of input features, differing both in the dimensionality and nature of the descriptors and in the kind of force field used for building the training/testing database. We reveal that even with low-dimensional descriptors the classification is highly accurate, while the use of high-dimensional descriptors leads to an even better performance. Training the classifier with configurations coming from different force fields we can remove force field bias and enhance the classifier performance and general applicability. Finally, we apply our classifiers to reveal mechanistic details of the ZIF-4-cp → \xrightarrow{} ZIF-4-cp-II phase transition. † † affiliation: Sorbonne Université, CNRS, Physico-chimie des Electrolytes et Nanosystèmes Interfaciaux, PHENIX, F-75005, Paris, France † † affiliation: These authors contributed equally to the work. † † affiliation: Chimie ParisTech, PSL University, CNRS, Institut de Recherche de Chimie Paris, 75005 Paris, France † † affiliation: These authors contributed equally to the work. † † affiliation: Chimie ParisTech, PSL University, CNRS, Institut de Recherche de Chimie Paris, 75005 Paris, France † † affiliation: Chimie ParisTech, PSL University, CNRS, Institut de Recherche de Chimie Paris, 75005 Paris, France † † email: [email protected] † † affiliation: Sorbonne Université, CNRS, Physico-chimie des Electrolytes et Nanosystèmes Interfaciaux, PHENIX, F-75005, Paris, France † † email: [email protected] † † suppinfo: I 1 Introduction Metal–organic frameworks, or MOFs, are hybrid organic–inorganic supramolecular assemblies based on relatively weak interactions that present a high number of intramolecular degrees of freedom. Due to this combination of factors, they present a higher propensity than classical solids for phenomena such as large-scale flexibility, framework dynamics, stimuli-responsive behavior, the presence of defects and disorder. Schneemann et al. 2014 ; Peh et al. 2020 ; Coudert 2025 For the same reasons, many MOFs also display polymorphism, i.e., the ability to exhibit two or more crystalline phases that differ in their arrangement of the organic linkers or their conformation within the crystal lattice. Such MOFs can therefore display several phases with the same chemical composition but with differences in their structure, impacting or not the topology of the framework. As is known in the case of molecular crystals, polymorphism has an important impact on both the physical and chemical properties of MOFs, Aulakh et al. 2015 and has been widely studied both experimentally and computationally. Cheetham et al. 2018 ; Castillo-Blas et al. 2024 Zeolitic Imidazolate Frameworks (ZIFs) are a typical example for the occurrence of polymorphism in MOFs, with dozens of reported Zn(imidazole) 2 phases varying in topology as a function of synthesis conditions, linker functionalization, thermal and mechanical treatment, and other influences during and after their synthesis. Hughes et al. 2012 ; Widmer et al. 2019 This extends even to the existence of disordered phases, including liquids, glasses, gels, and composites thereof. Bumstead et al. 2020 ; Hou et al. 2020 ; Smirnova et al. 2023 However, despite the large amount of experimental evidence gathered on the polymorphism of ZIFs, many important questions remain widely open. The exact pressure–temperature phase diagram of ZIF-4, for example, has not yet been fully solved; although several studies have addressed the question, Widmer et al. 2019 ; Méndez and Semino 2024 these phase diagrams do not include many of the ZIF polymorphs originally found by Park and coworkers Park et al. 2006 nor are they really representative of thermodynamic stability, since it is well-known that the most stable phases are the dense ones, and porous phases occur due to kinetic considerations. Moreover, the nature and structure of some of the high-pressure phases of ZIF-4, ZIF-8 and ZIF-62 are still not fully solved, Vervoorts et al. 2021 ; Song et al. 2022 ; Robertson et al. 2025 the differences in the structures of the known phases can be quite subtle and some of the phases have a very narrow thermodynamic stability range. Bouëssel du Bourg et al. 2014 Computational techniques give us access to molecular-level resolution, which is key in revealing structural and mechanistic information pertaining to ZIF polymorphs and their phase transformations. Castel et al. 2024 ; Castel and Coudert 2022 ; Méndez and Semino 2024 ; Méndez and Semino 2024 However, simulation trajectories contain large volumes of data which require automation and agnostic data analysis methods to reveal all their hidden potential. For these reasons, there is interest in the development of computational methodologies for the identification of phases in ZIF systems. The challenge is even greater for the assignment of phases in a time-resolved and local manner. Being able to determine, during a molecular dynamics simulation, that a specific atomic environment at a given time is closest to a specific phase allows for the study of the dynamics of phase transitions and their mechanisms at the atomistic scale. This is made difficult because differences between structures are subtle, and simple geometrical criteria based on intuition are largely insufficient. Helfrecht et al. 2019 Techniques for local classification of phases have been employed previously: for example, geometric methods such as Polyhedral Template Matching (PTM) Larsen et al. 2016 have been used to classify amorphous phases of silicon. Rosset et al. 2025 In a previous study, a database of imidazolate-based ZIF polymorphs was created and used to train a neural-network classifier to identify ZIF polymorphs. Méndez and Semino 2024 This study only dealt with ambient pressure phases of these materials, thus leaving out the many high-pressure polymorphs. In this work, we present two methodologies for the classification of ZIF phases in a wide range of temperatures and pressures, which we apply to molecular dynamics trajectories in a time- and space-resolved manner. Our identification methods are based on encoding the local atomic geometry into environment descriptors such as Behler-Parrinello symmetry functions (BPSF) Behler 2011 and Smooth Overlap of Atomic Positions (SOAP) Bartók et al. 2013 , which are then fed to neural networks performing a multi-class classification task. We show that our classifiers are not highly force field dependent, and that a classifier trained with a small set of BPSF functions had a similar accuracy to the one trained with high-dimensional SOAP vectors as features. After validating the methods on known phases produced by two different force fields, we demonstrate their power in providing insight into the molecular-level mechanisms of the phase transitions between the high-pressure phases ZIF-4-cp and ZIF-4-cp-II, that belong to the ZIF-4 family of materials. 2 Systems & Methods 2.1 ZIF-4 polymorphs and