Machine-learned free energies of nanoalloys

Mixing two or multiple metals together at the nanoscale results in the production of so-called nanoalloys. At this scale, the tendency for mixing the different elements together is strongly altered by their relative preference for residing at the surface of the nanoparticle. However, successful applications of nanoalloys require a good control of their surface, and it is important to understand the driving mechanisms behind the spatial distribution of the different elements, called chemical ordering.
Theoretically, simple finite lattice approaches built on Ising or Potts models already provide a rough approximation to bi- or multimetallic nanoalloys with a good tendency for mixing such as gold-silver or palladium-platinum. However, these models do not capture local relaxation effects due to lattice mismatch and fail to represent finite temperature and plasticity effects. Atomistic off-lattice models, with continuous descriptions of the many-body effects at play in metals, can correct these flaws but are computationally more involved.
This internship aims to bridge the gap between lattice and off-lattice models, by incorporating the main effects of local lattice deformations into a lattice model. The key idea is to evaluate the free energy of any lattice configuration using a machine-learning representation trained on the more realistic off-lattice model.
In practice, the internship will consist of
-getting familiar with Monte Carlo simulations of the true off-lattice model for a simple bimetallic Ag-Au nanoalloy, varying the temperature
-moving towards a lattice representation by treating the free energy through simple approximations valid for entire temperature ranges
-identifying appropriate descriptors connecting the chemical ordering in the nanoalloy to its free energy
-training a machine-learning model to improve the lattice model so it can describe the vibrational contributions to the free energy
Published on March 18, 2026
Updated on March 18, 2026