White banner promoting the 2nd EPICURE meets Slovenia event
2nd EPICURE meets Slovenia
06/01/2026

Research supported by EPICURE contributes to advances in machine learning for quantum chemistry

A study published in Nature Communications explained a real-space machine learning approach designed to improve density functional theory (DFT), a fundamental methodology in quantum chemistry and materials science.

 

Developed by Elias Polak, Heng Zhao, and Stefan Vuckovic, the research introduces a framework in which machine-learning models are trained directly on correlation energy densities, learned point by point in real space. This method enables a more accurate description of electronic correlations, addressing long-standing limitations in the development of next-generation density functionals.

 

The work relies on advanced high-performance computing (HPC) resources to build, test, and validate the computational framework. Application support provided through the EPICURE project contributed to the development of the infrastructure used in the study.

 

 

EPICURE’s role in computational support

 

The article in Nature Communications mentions EPICURE’s contribution to developing the Python-based framework used in the study. With EPICURE support in code optimisation, memory efficiency, and scaling for large computational tasks, the team accelerated complex training workflows essential to real-space machine learning of density functionals.

 

EPICURE enabled improved performance and memory efficiency for the framework, generating energy density training data, emphasising how targeted HPC services can directly accelerate scientific discovery.

 

This collaboration highlights the impact of EPICURE’s European HPC support services on next-generation computational science. It emphasises the role beyond standard user help-desk support, including Level 2 and 3 application support, encompassing application porting, optimisation, and execution of key applications.

 

For researchers seeking powerful computational frameworks and expert support to use European HPC systems, EPICURE is a solution. Through coordinated user support and technical guidance, EPICURE helps unlock the full potential of EuroHPC resources.

 

Read the full Nature Communications publication: Polak, E., Zhao, H. & Vuckovic, S. Real-space machine learning of correlation density functionals. Nat Commun 16, 11306 (2025)

Discover our latest news

More information soon available!