Machine Learning Achieves 95% Accuracy in Optimized K-Point Mesh Generation for Quantum ESPRESSO

Accurate and efficient materials modelling relies heavily on density functional theory calculations, but determining optimal computational settings remains a significant challenge, particularly for large-scale studies. Elena Patyukova, Junwen Yin, Susmita Basak, and colleagues at the Scientific Computing Department, Daresbury and Rutherford Appleton Laboratories, now address this problem with a novel machine learning approach.

The team develops a system that automatically generates input files for Quantum Espresso calculations, crucially optimising the k-point mesh, a parameter governing the accuracy and computational cost of the simulation. By training models on a dataset of over 20,000 materials, they achieve reliable prediction of appropriate k-point settings, ensuring converged results for the vast majority of compounds and streamlining materials discovery workflows. This advancement promises to accelerate materials modelling by removing a key bottleneck in computational efficiency and accuracy. This represents a particularly important problem for high-throughput and agentic workflows, where, due to computational cost, any additional convergence studies are preferably avoided. Therefore, there is a need for tools and models which are able to predict Density Functional Theory (DFT) parameters from basic input information, such as a structure.

Machine Learning Predicts DFT Convergence Parameters Scientists have developed a machine learning approach to predict optimal convergence parameters for density functional theory (DFT) calculations, a cornerstone of computational materials science. Selecting appropriate parameters, such as k-point density, often requires time-consuming trial and error. This work aims to automate and optimise this process, building models that reliably and efficiently predict these parameters for materials simulations. A key innovation lies in ensuring confidence in the predictions through the use of conformalised quantile regression, a statistical technique that provides a prediction interval with a guaranteed coverage probability. This allows scientists to assess the reliability of the predictions and make informed decisions.

The team explored a variety of machine learning models, including random forests, gradient boosting machines, graph neural networks, and attention-based networks. These models leverage a range of descriptors to represent materials, encompassing compositional information, crystal structure, and even insights extracted from materials science literature. The focus extends beyond simple prediction accuracy to encompass a robust quantification of uncertainty, crucial for high-throughput computing and efficient materials screening. By making their code and data publicly available, the authors promote reproducibility and collaboration within the scientific community.

Machine Learning Predicts DFT K-Point Convergence Scientists developed a machine learning approach to predict optimal parameters for density functional theory (DFT) calculations, specifically focusing on k-point sampling within the Quantum ESPRESSO software package. This addresses a critical need in high-throughput materials science, where extensive convergence studies to determine appropriate parameters are computationally expensive and time-consuming. Researchers generated a comprehensive training dataset comprising over 20,000 materials, each evaluated with an energy convergence threshold of 1 meV/atom, to facilitate accurate model training and validation.

The team evaluated several machine learning models to predict k-point distances, prioritising models capable of estimating prediction uncertainty. Crucially, the models were designed to ensure that, for at least 85% of compounds, the predicted k-point distance lies within the convergence region, minimising the risk of inaccurate results. Further refinement involved predicting specific quantiles, guaranteeing that a high percentage of predictions are not underestimated, thereby balancing accuracy and computational cost. Experiments demonstrate the successful development of a practical web application that automatically generates input files for single-point calculations. This tool streamlines the DFT workflow, promising to improve the quality of high-throughput datasets and reduce unnecessary computational time, contributing to more efficient and sustainable materials discovery. The research delivers a significant advancement in automating DFT parameter selection, enabling more robust and reliable high-throughput calculations. Predicting K-point Convergence With Machine Learning This research presents a machine learning approach to predict appropriate k-point sampling for density functional theory calculations, addressing a key challenge in high-throughput materials science.

The team developed models trained on a dataset of over 20,000 compounds, enabling prediction of k-point distances needed to achieve energy convergence with a high degree of confidence. By accurately estimating uncertainty, the models ensure reliable predictions for a substantial majority of materials, typically 85-95%. The resulting models represent a significant step towards automating parameter selection in computational workflows, potentially reducing computational cost and improving the efficiency of materials discovery.

The team has made these models publicly available through a web application, facilitating wider adoption and use within the scientific community. While the work focuses on single-point calculations, the authors acknowledge that extending the approach to other types of calculations and properties represents an area for future development. They also note that further research could explore the integration of these predictive models within emerging, AI-driven computational frameworks. This work promises to accelerate materials discovery by reducing the computational burden associated with parameter optimisation and enhancing the reliability of high-throughput calculations. 👉 More information 🗞 Automatic generation of input files with optimised k-point meshes for Quantum Espresso self-consistent field single point total energy calculations 🧠 ArXiv: https://arxiv.org/abs/2512.15303 Tags: