Volume 5, Issue 4 (2025) – 15 articles
Cover Picture: The cover illustrates the central theme of this Review: machine learning (ML) is closing the long-standing gap between ab initio accuracy and simulation scalability by learning either the inter-atomic potentials (IAPs) or the electronic Hamiltonian (HAM) from materials data.
On the left, the ML-IAPs panel shows an atomic neighborhood graph with rotational symmetry, highlighting symmetry-aware message passing and equivariant representations that keep scalar energies invariant while ensuring vectors transform correctly.
On the right, the ML-HAM panel presents a block-structured Hermitian matrix and band structures, capturing the idea of learning energy operators to recover band structures with improved physical interpretability.
view this paper On the left, the ML-IAPs panel shows an atomic neighborhood graph with rotational symmetry, highlighting symmetry-aware message passing and equivariant representations that keep scalar energies invariant while ensuring vectors transform correctly.
On the right, the ML-HAM panel presents a block-structured Hermitian matrix and band structures, capturing the idea of learning energy operators to recover band structures with improved physical interpretability.
Back Cover Picture: The cover illustrates FIND, a forward and inverse navigation and discovery platform for the intelligent design of solid-state hydrogen storage alloys. At the center, a bright-blue cavity embodies the smart-algorithm engine that executes bidirectional workflows. Two automated pipelines pierce this chamber: the forward design (blue arrow pipeline) conveys assorted hydride precursors into a multi-objective machine learning model that instantly outputs predicted PCT attributes (plateau pressure, capacity, entropy and enthalpy) displayed as translucent holographic curves. The inverse design (yellow arrow pipeline) accepts a user-defined target PCT curve, feeds it to a variational autoencoder, and spawns a family of alloy compositions engineered to meet the specified performance metrics. Both pipelines converge on a solid block that contains over 6000 validated records from the Digital Hydrogen-S database, which underpins the predictive robustness of FIND. By offering these design routes free of charge, FIND furnishes the hydrogen storage community with a high-throughput, data-driven toolkit and establishes a paradigmatic example of materials informatics enabled alloy optimization.
view this paper
