In the contested terrain of nanoscale chemistry, where a handful of atoms can rearrange themselves into thousands of possible geometries, researchers have unveiled a new computational engine capable of finding the most stable arrangement of gold-copper bimetallic clusters with unprecedented reliability. A team led by Yuxiang Tang, Yuxin Zhang, Hua Shi, Yuheng Chen and Xiaomin Wu, based at Xiamen University of Technology and Xiamen University, reports in the Journal of Nanoparticle Research the development of an improved optimization algorithm, Elite Niching Adaptive Differential Evolution, or ENADE, that outperforms established search methods on some of the most demanding benchmark problems in cluster science. The work, published on 7 July 2026, addresses a stubborn bottleneck in computational materials design: reliably locating the global minimum energy structure of alloy nanoparticles, a task that becomes exponentially harder as the number of atoms grows.
Bimetallic clusters of gold and copper are far more than academic curiosities. Their synergistic electronic interactions and tunable atomic arrangements make them attractive catalysts for industrially significant reactions, including the electrochemical reduction of carbon dioxide and the oxidation of biomass-derived platform molecules such as 5-(hydroxymethyl)furfural. In these applications, catalytic activity and selectivity depend exquisitely on which atoms sit at the surface, which form the core, and how the two species distribute themselves across the particle. Predicting that distribution from first principles is therefore a central problem in nanoscience, and one that has resisted simple solution because the potential energy landscape of even a modest cluster contains a combinatorial explosion of local minima separated by energy barriers.
The mathematical character of the problem is what makes it so punishing. For a cluster of N atoms, each with three Cartesian degrees of freedom, the search space of possible geometries spans 3N dimensions, and the energy surface is riddled with local minima corresponding to distorted icosahedra, decahedra, truncated octahedra, amorphous packings and countless hybrid motifs. When two chemical species are involved, the combinatorics multiply again, because each atomic site can host either a gold or a copper atom. Classical heuristics such as basin-hopping, which repeatedly perturbs a structure and locally relaxes it, and genetic algorithms, which evolve populations of candidate structures through crossover and mutation, have long served as the field’s workhorses. Yet both are prone to premature convergence, in which the search becomes trapped in a deep but suboptimal basin, especially on landscapes with many nearly degenerate competing structures.
ENADE attacks this difficulty by layering several improvements onto the standard differential evolution framework. Differential evolution, a population-based metaheuristic introduced in the 1990s, generates new candidate solutions by combining scaled differences between existing population members with a target vector, then keeping the trial solution only if it improves on the incumbent. Conventional differential evolution, however, was designed for smooth numerical optimization problems and struggles with the rugged, symmetry-rich landscapes of atomic clusters. The Xiamen team’s variant introduces an elite pool, a curated archive of the best structures found so far that guides mutation toward promising regions of configuration space and accelerates convergence. It also deploys a multi-population parallel strategy, in which several subpopulations explore the landscape simultaneously, strengthening the global search and reducing the chance that all candidates collapse into the same deceptive basin.
Two further innovations target the specific physics of cluster optimization. The algorithm employs a combined sphere-cutting crossover operator, which exploits the roughly spherical geometry of compact clusters to exchange structural fragments between parent solutions in a physically meaningful way, together with an atomic-exchange operator that swaps the chemical identities of gold and copper atoms within a given geometry. The latter operator is crucial for bimetallic systems, where finding the right geometry is only half the battle; the chemical ordering pattern, which species occupies which site, must be optimized simultaneously. Finally, ENADE couples a restart mechanism with niching, a technique borrowed from evolutionary computation that maintains diversity by encouraging the population to settle into multiple distinct basins rather than a single one. When the population stagnates, the restart mechanism reinitializes parts of the search while the niching structure preserves the distinct low-energy families already discovered, preventing the premature convergence that plagues simpler algorithms.
The performance benchmarks are striking. Working with the Gupta potential, a semi-empirical tight-binding description of interatomic forces widely used for transition-metal clusters, the team validated ENADE on gold-copper clusters containing 13, 19, 23 and 38 atoms, systems whose global minima are already well characterized in the literature and therefore serve as rigorous tests of any search algorithm. For the 38-atom cluster, a notoriously difficult case corresponding to a complete Mackay icosahedron or, in the alloy case, the expected truncated-octahedral motif, ENADE located the known global minimum in 70 percent of independent runs, a success rate that substantially exceeded those of basin-hopping, genetic-algorithm and conventional differential evolution baselines run under comparable computational budgets. Even more telling is the algorithm’s reproducibility: across 20 independent runs on a 55-atom cluster, the standard deviation of the best energy found was a mere 0.002 electron volts, a margin so narrow that it amounts to near-deterministic recovery of the global minimum in a domain where scatter of tenths of an electron volt is common.
The structures located by ENADE serve, as the authors emphasize, as a consistency check rather than as claims of new structural discoveries. Reproducing the known Cu-core/Au-surface segregation pattern confirms both the algorithm and the underlying physics: because copper atoms are smaller than gold atoms, a compact copper core relieves the size-mismatch strain, while gold atoms, with their more diffuse electron clouds and lower surface energy, preferentially decorate the surface. The algorithm also recovered the expected truncated-octahedral motif for the 38-atom benchmark, and its stability analysis showed that the mixed gold-copper clusters are more stable, in terms of binding energy per atom, than their pure-metal counterparts of the same size. This latter finding quantifies the alloying advantage that experiments on Au-Cu nanoparticles have long suggested, tying the thermodynamic preference for mixing directly to the computed energetics of the most stable configurations.
The technical significance of the work lies less in the specific structures found, which were largely known, than in the robustness of the ENADE framework itself on chemically ordered bimetallic energy landscapes. Global optimization for alloy clusters differs qualitatively from the monometallic case, because the search must navigate both configurational space, the arrangement of atoms in space, and chemical space, the assignment of species to sites. Algorithms tuned for one aspect often neglect the other. By integrating atomic-exchange moves directly into the evolutionary cycle alongside geometric operators, ENADE treats both dimensions of the problem on equal footing. The elite pool ensures that hard-won low-energy configurations are never lost to stochastic drift, while the parallel subpopulations and niching mechanism maintain the population diversity that high-dimensional rugged landscapes demand. The authors suggest this combination could extend naturally to larger clusters, other alloy systems such as Ag-Au and Co-Pt, and eventually to searches driven by more accurate machine-learned interatomic potentials.
The broader context makes the contribution timely. The past few years have seen a surge of interest in automated structure searchers, from machine learning and graph theory assisted universal searchers to deep reinforcement learning navigators and specialized walker methods developed by the nanoalloy community. Each new tool addresses the same underlying imperative: as researchers design catalysts atom by atom, they need algorithms that can be trusted to find the true ground state rather than a convincing impostor. A missed global minimum can lead to incorrect predictions of catalytic activity, melting behavior or optical response. The Xiamen team’s demonstration that a carefully engineered differential evolution scheme can beat the field’s standard methods on canonical benchmarks adds a powerful and computationally inexpensive option to the toolkit, one that requires no gradient information and can be coupled to any energy model, from the Gupta potential used here to density functional theory or neural network potentials.
Support for the research came from the National Natural Science Foundation of China under grant number 62372392. The corresponding author is Xiaomin Wu of the School of Opto-Electronic and Communication Engineering at Xiamen University of Technology. For experimentalists synthesizing Au-Cu nanocatalysts in ionic liquids or probing their behavior in electrochemical cells, the study offers reassurance that the structures they target in the laboratory can now be identified computationally with far greater confidence. And for the computational community, ENADE provides a template: elite guidance, parallel populations, chemistry-aware operators and disciplined diversity control, working in concert, appear to be the ingredients that tame even the most treacherous atomic landscapes.
Subject of Research: Global structural optimization and stability of Au-Cu bimetallic nanoclusters using an improved Elite Niching Adaptive Differential Evolution algorithm with the Gupta potential.
Subject of Research: Technology and Engineering
Article Title: Global structural optimization and stability study of Au-Cu bimetallic clusters based on Elite Niching Adaptive Differential Evolution
Article References: Tang, Y., Zhang, Y., Shi, H., Chen, Y., & Wu, X. (2026). Global structural optimization and stability study of Au-Cu bimetallic clusters based on Elite Niching Adaptive Differential Evolution. Journal of Nanoparticle Research, 28(7), Article 188. https://doi.org/10.1007/s11051-026-06711-0
Image Credits: AI Generated
DOI: 10.1007/s11051-026-06711-0
Keywords: Bimetallic clusters, Differential evolution algorithm, Global optimization, Gupta potential, Au-Cu clusters, Nanostructured catalysts, Modeling and simulation
Cite Scienmag News
APA MLA Chicago
Gavin Prescott. (September 7, 2026). Evolutionary algorithm reveals stable structures of gold-copper clusters. Scienmag. https://scienmag.com/evolutionary-algorithm-reveals-stable-structures-of-gold-copper-clusters/
Gavin Prescott. “Evolutionary algorithm reveals stable structures of gold-copper clusters.” Scienmag, 7 September 2026, https://scienmag.com/evolutionary-algorithm-reveals-stable-structures-of-gold-copper-clusters/. Accessed 7 September 2026.
Gavin Prescott. “Evolutionary algorithm reveals stable structures of gold-copper clusters.” Scienmag. September 7, 2026. https://scienmag.com/evolutionary-algorithm-reveals-stable-structures-of-gold-copper-clusters/
Copy citation Download RIS
Tags: advanced computational methods for nanoparticle stabilityadvanced search algorithms in nanoscienceapplication of evolutionary algorithms in materials scienceatomic arrangement optimization in bimetallic nanoparticlesbimetallic nanoparticle stabilitybiomass oxidation catalystscatalyst surface atom configurationcatalytic properties of alloy nanoparticleschallenges in nanocluster energy landscape explorationcomputational chemistry in nanomaterialscomputational design of alloy clusterselectrochemical reduction catalystselite niching adaptive differential evolution algorithmENADE optimization methodevolutionary algorithms for materials designglobal minimum energy search in nanoscale clustersglobal minimum energy structuresgold-copper bimetallic nanoparticle stabilityGold-copper clustersnanocatalyst design for CO2 reductionnanoscale chemistry and cluster configurationnanoscale cluster geometrystable atomic arrangements in bimetallic clustersstable gold-copper cluster structures
