A generative machine learning model for designing metal hydrides applied to hydrogen storage

TL;DR

This work tackles the challenge of identifying safe, efficient hydrogen-storage metal hydrides amid limited datasets. It combines causal discovery to pinpoint key features with a lightweight Crystal Diffusion Variational Autoencoder (CDVAE) to generate novel formulas and crystal structures, followed by refinement with M3GNet and validation via density functional theory (DFT). Using a 450-sample MP-derived dataset, the pipeline generates 1,000 candidates and identifies six new alloy hydrides, with four showing favorable stability and hydrogen storage prospects by DFT. The framework demonstrates scalable, cost-effective materials discovery that can expand hydrogen-storage databases and accelerate experimental exploration under realistic operating conditions.

Abstract

Developing new metal hydrides is a critical step toward efficient hydrogen storage in carbon-neutral energy systems. However, existing materials databases, such as the Materials Project, contain a limited number of well-characterized hydrides, which constrains the discovery of optimal candidates. This work presents a framework that integrates causal discovery with a lightweight generative machine learning model to generate novel metal hydride candidates that may not exist in current databases. Using a dataset of 450 samples (270 training, 90 validation, and 90 testing), the model generates 1,000 candidates. After ranking and filtering, six previously unreported chemical formulas and crystal structures are identified, four of which are validated by density functional theory simulations and show strong potential for future experimental investigation. Overall, the proposed framework provides a scalable and time-efficient approach for expanding hydrogen storage datasets and accelerating materials discovery.

Figures (6)

• Figure 1: Workflow of the proposed method for discovering novel hydrogen storage materials.
• Figure 2: Formation energy-based weight factor comparison. The original function is shown in blue, and the modified version is shown in orange.
• Figure 3: Illustration of how the PC algorithm works.
• Figure 4: Illustration of how the VAE algorithm works.
• Figure 5: Causal graph generated by the FCI algorithm using the Hydrogen Storage Database. Edges represent statistical dependencies identified through Chi-squared independence tests at a significance level of $\alpha = 0.05$. The focus is on the hydrogen storage score node, highlighting its relationship with key features such as hydrogen weight fraction ($W_{H_2}$), formation energy ($E_{form}$), Band Gap, and crystal structure (f Character). Bidirectional edges indicate correlation rather than causation.