Characterising and understanding the microscopic structure of amorphous materials

Thesis

Amorphous materials are ubiquitous and play an important role in various applications. Fully exploiting these materials requires a thorough understanding of their structure, which is challenging due to the lack of long-range order that complicates the direct interpretation of experimental data.

This thesis explores the use of hybrid Reverse Monte Carlo (HRMC) refinement, a method in which atomic coordinates in a model system are refined to match experimental total scattering data while also minimising the system energy, for obtaining realistic amorphous models. The key to the success of this method is the use of accurate force fields to describe the energetics.

I use the HRMC approach to refine a model of amorphous calcium carbonate (ACC). I find that the principal component of structural order in ACC is the distribution of calcium ions. To understand this structure better, I extract an effective Ca· · ·Ca interaction potential. This potential is well described by a Lennard-Jones–Gauss functional form, and the relevant empirical parameters found for ACC are known from theory to promote structural complexity. In this way, I show that the complex structure of ACC arises from geometric frustration.

Next, I explore the relationship between local geometric preferences and network topology in AB₂ metal-organic frameworks (MOFs). I develop an automated coarse-graining routine to reduce complex structures to their fundamental building units. By combining these coarse-grained representations with atom-density-based descriptors and dimensionality reduction algorithms, I demonstrate how to visualise and quantify structural diversity in MOFs. This investigation reveals that building blocks with similar local geometric preferences can favour distinctly different topologies.

With this result in mind, the final two chapters focus on modelling zeolitic imidazolate frameworks (ZIFs), a subset of MOFs known for their ability to amorphise, melt, and vitrify. In Chapter 4, I detail the development of a fast and accurate machine-learned (ML) potential for ZIFs. I show that HRMC refinement can be used to iteratively improve the ML model’s ability to describe the amorphous ZIF structure.

In Chapter 5, I use the ML potential alongside molecular dynamics and HRMC to generate an amorphous ZIF model, ordering insights into the structure and topology of the amorphous MOF. I contextualise the findings by introducing a topological metric based on established crystalline descriptors that is able to discriminate between amorphous networks with different topological features. This enables a discussion of the (dis)similarities between amorphous Si, amorphous SiO₂, and amorphous ZIF, suggesting that the landscape of AB₂ continuous random networks may be more nuanced than previously understood.

Authors: Nicholas, TC

• DOI: 10.5287/ora-ngrgkj6yp
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: machine-learning; total scattering; structural modelling; disorder; Monte-Carlo; inorganic chemistry; metal–organic frameworks
• Subjects: Metal-organic frameworks; Machine learning; Structural chemistry; Monte Carlo method; Chemistry, Inorganic; Materials