Robust crystal structure identification at extreme conditions using a density-independent spectral descriptor and supervised learning

Paul Lafourcade,Jean-Bernard Maillet,Christophe Denoual,Eléonore Duval,Arnaud Allera,Alexandra M. Goryaeva,Mihai-Cosmin Marinica

2023-07-10

Abstract:The increased time- and length-scale of classical molecular dynamics simulations have led to raw data flows surpassing storage capacities, necessitating on-the-fly integration of structural analysis algorithms. As a result, algorithms must be computationally efficient, accurate, and stable at finite temperature to reliably extract the relevant features of the data at simulation time. In this work, we leverage spectral descriptors to encode local atomic environments and build crystal structure classification models. In addition to the classical way spectral descriptors are computed, i.e. over a fixed radius neighborhood sphere around a central atom, we propose an extension to make them independent from the material's density. Models are trained on defect-free crystal structures with moderate thermal noise and elastic deformation, using the linear discriminant analysis (LDA) method for dimensionality reduction and logistic regression (LR) for subsequent classification. The proposed classification model is intentionally designed to be simple, incorporating only a limited number of parameters. This deliberate simplicity enables the model to be trained effectively even when working with small databases. Despite the limited training data, the model still demonstrates inherent transferability, making it applicable to a broader range of scenarios and datasets. The accuracy of our models in extreme conditions is compared to traditional algorithms from the literature, namely adaptive common neighbor analysis (a-CNA), polyhedral template matching (PTM) and diamond structure identification (IDS). Finally, we showcase two applications of our method: tracking a solid-solid BCC-to-HCP phase transformation in Zirconium at high pressure up to high temperature, and visualizing stress-induced dislocation loop expansion in single crystal FCC Aluminum containing a Frank-Read source, at high temperature.

Materials Science

What problem does this paper attempt to address?

The paper aims to address the problem of crystal structure identification under extreme conditions (such as high temperature, high pressure, and large deformation). As the time and length scales of classical molecular dynamics simulations increase, the generated raw data stream exceeds storage capacity, necessitating real-time integration of structural analysis algorithms. These algorithms must efficiently, accurately, and stably extract relevant features of the data at finite temperatures. This paper proposes a crystal structure classification model based on spectral descriptors and supervised learning, which can operate independently of material density and maintain high accuracy under extreme conditions.

Specifically, the paper attempts to address the following key issues:

1. **Data Storage and Processing**: Large-scale molecular dynamics simulations generate enormous amounts of data that exceed storage capacity, requiring real-time processing and analysis during the simulation.
2. **Algorithm Efficiency and Accuracy**: Existing structural analysis methods perform poorly under extreme conditions such as high temperature and high pressure, necessitating the development of new algorithms to improve efficiency and accuracy.
3. **Robustness of Structure Identification**: How to accurately identify crystal structures and detect defects in the presence of thermal noise and elastic deformation.
4. **Generalization Ability of the Model**: How to construct a model that can generalize to different materials and conditions with limited training data.

To achieve these goals, the paper proposes a crystal structure classification model based on spectral descriptors. The model performs dimensionality reduction through Linear Discriminant Analysis (LDA) and then uses Logistic Regression (LR) for classification. Additionally, the paper introduces a method of calculating spectral descriptors with a fixed number of neighbors, making it independent of material density and thereby enhancing robustness under extreme conditions.