Abstract:Stable or metastable crystal structures of assembled atoms can be predicted by finding the global or local minima of the energy surface within a broad space of atomic configurations. Generally, this requires repeated first-principles energy calculations, which is often impractical for large crystalline systems. Here, we present significant progress toward solving the crystal structure prediction problem: we performed noniterative, single-shot screening using a large library of virtually created crystal structures with a machine-learning energy predictor. This shotgun method (ShotgunCSP) has two key technical components: transfer learning for accurate energy prediction of pre-relaxed crystalline states, and two generative models based on element substitution and symmetry-restricted structure generation to produce promising and diverse crystal structures. First-principles calculations were performed only to generate the training samples and to refine a few selected pre-relaxed crystal structures. The ShotunCSP method is computationally less intensive than conventional methods and exhibits exceptional prediction accuracy, reaching 93.3% in benchmark tests with 90 different crystal structures.

What problem does this paper attempt to address?

The paper aims to address the problem of Crystal Structure Prediction (CSP), which involves predicting stable or metastable crystal structures from given chemical compositions. Traditional methods often require a large number of first-principles energy calculations, which are usually impractical for large-scale crystal systems. The paper proposes a new method called ShotgunCSP, which achieves efficient and accurate crystal structure prediction through the following key techniques: 1. **Machine Learning Energy Predictor**: First, a crystal graph convolutional neural network (CGCNN) is trained using data from the Materials Project database. Then, transfer learning is used to fine-tune the pre-trained model to accurately predict the formation energies of specific systems. This approach significantly improves the accuracy of energy predictions and is applicable to low-energy states. 2. **Generation Models**: To generate candidate crystal structures, the paper proposes two generation models: - **Element Substitution Method (ShotgunCSP-GT)**: Generates new virtual crystal structures by substituting elements in known crystal templates. - **Wyckoff Position Generator (ShotgunCSP-GW)**: When no suitable template is available, it randomly generates atomic coordinates with constraints based on the given space group and Wyckoff positions. 3. **High-Throughput Virtual Screening**: The machine learning energy predictor is used to screen a large number of generated virtual crystal structures, thereby narrowing down the range of candidate structures. Finally, only a few screened candidate structures undergo first-principles optimization calculations, greatly reducing computational costs. Through these techniques, the ShotgunCSP method not only far exceeds traditional methods in computational efficiency but also performs excellently in benchmark tests, achieving an accuracy of 93.3%, validating its effectiveness in various crystal structure prediction tasks.

Shotgun crystal structure prediction using machine-learned formation energies

Crystal structure prediction with machine learning-based element substitution

Machine-Learned Fragment-Based Energies for Crystal Structure Prediction

Accelerated Organic Crystal Structure Prediction with Genetic Algorithms and Machine Learning

Predictive crystallography at scale: mapping, validating, and learning from 1,000 crystal energy landscapes

A Robust Crystal Structure Prediction Method to Support Small Molecule Drug Development with Large Scale Validation and Prospective Studies

Reliable and Practical Computational Prediction of Molecular Crystal Polymorphs

Data-efficient machine learning for molecular crystal structure prediction

Latent Conservative Objective Models for Data-Driven Crystal Structure Prediction

End-to-End Crystal Structure Prediction from Powder X-Ray Diffraction

How Accurate Can Crystal Structure Predictions Be for High-Energy Molecular Crystals?

CrySPY: a crystal structure prediction tool accelerated by machine learning

Accelerating crystal structure prediction by machine-learning interatomic potentials with active learning

AlphaCrystal-II: Distance matrix based crystal structure prediction using deep learning

Machine learning based energy-free structure predictions of molecules, transition states, and solids

A machine learning potential-based generative algorithm for on-lattice crystal structure prediction

Optimality guarantees for crystal structure prediction

AlphaCrystal: Contact map based crystal structure prediction using deep learning

Concurrent learning scheme for crystal structure prediction

A Machine-Learning-Assisted Crystalline Structure Prediction Framework To Accelerate Materials Discovery