Abstract
Researchers have developed Quantum-inspired Cluster Expansion (QCE), a new framework that maps the search for new materials onto an Ising Hamiltonian—the "native language" of quantum annealers. This approach searches the vast landscape of possible chemical combinations 10–50 times faster than traditional algorithms, leading to the discovery of a novel multi-metal oxide catalyst for hydrogen production that is 8 times more active and significantly more stable than current state-of-the-art materials.
Introduction
The "chemical space"—the total number of possible ways to combine elements into materials—is unimaginably vast. Finding a material with specific properties, such as high conductivity or catalytic efficiency, is a monumental challenge. Traditionally, researchers rely on Density Functional Theory (DFT) for materials simulation, but its computational expense makes large-scale searches prohibitive. While Machine Learning (ML) surrogate models have accelerated the process, the search phase remains a bottleneck. Standard optimization tools like Genetic Algorithms (GA) or Bayesian Optimization (BO) often converge on local minima rather than the global optimum. QCE addresses this by translating complex material properties into a format solvable by quantum-inspired hardware.
Methodology
The QCE framework involves three primary steps:
- Cluster Expansion (CE): Representing material properties (like stability or catalytic activity) as a power series of many-body atomic interactions.
- QUBO Mapping: Transforming these interactions into a Quadratic Unconstrained Binary Optimization (QUBO) form through term reduction and binary variable superposition.
- Digital Annealing: Solving the QUBO using Fujitsu’s Digital Annealer (DA), hardware specifically designed to find the absolute ground state of Ising models.
The study utilized benchmark sets (Cu-Ni-Pd-Ag) and OER catalyst sets (Ru-Ti-W-Sb-Cr-Mn-V-Co and Ru-Zr-Hf-Y-V-Co-Fe-Ce). The model's predictions were experimentally validated by synthesizing and testing 49 new electrocatalyst candidates.
Results and Discussion
The QCE method achieved a 10–50x speedup over GA and BO while consistently finding more stable, lower-energy structures. The search identified a previously unknown catalyst family, with the top performer being a Ru-Cr-Mn-Sb oxide ($Ru_{0.58}Cr_{0.25}Mn_{0.09}Sb_{0.08}O_2$). This material demonstrated a mass activity 8 times higher than the industrial standard ($RuO_2$) and a 10x improvement in stability, maintaining performance for 180 hours in harsh acidic conditions.
The success of this catalyst is attributed to its unique electronic and thermodynamic properties. Post-hoc analysis revealed that the addition of Cr, Mn, and Sb alters the Density of States (DOS) at the Fermi level and improves mixing energy, which serves as a robust proxy for catalytic activity.
As shown in Table 1, the discovered catalyst exhibits a distinct Bader charge and a significantly altered DOS at the Fermi level compared to standard $RuO_2$. These changes facilitate stronger metal-metal bonds and lower the energy barrier for the rate-determining step of the reaction.
Conclusion
QCE provides a powerful new tool for navigating the chemical space, offering a 95x improvement over random sampling for finding high-performing materials. While currently restricted to predefined crystal lattices, future work aims to incorporate "extended cluster expansion" to search across multiple lattice types simultaneously. This quantum-inspired approach paves the way for the rapid discovery of next-generation materials for sustainable energy applications.
Summary updated