JANUS is a multimodal neural sampler that couples continuous and masked discrete diffusion through an equivariant graph neural network trained directly from energy evaluations, without pre-generated equilibrium data, providing a foundation for thermodynamic sampling, characterization and inverse design of chemically disordered materials.
Abstract
Many problems in disordered materials require sampling beyond fixed composition and volume, where coupled changes in atomic identities and structure create a prohibitively expensive discrete-continuous sampling problem. Here we introduce JANUS, a multimodal neural sampler that couples continuous and masked discrete diffusion through an equivariant graph neural network trained directly from energy evaluations, without pre-generated equilibrium data. In benchmark Ising and isobaric $\Delta\mu NPT$ alloy systems, JANUS reproduces reference Monte Carlo equilibrium observables and recovers free energies and phase behavior with more than three orders of magnitude fewer energy evaluations. In multicomponent alloys, JANUS enables conditional steering toward prescribed chemical short-range order and enhanced bulk modulus and, when coupled to a large language model evolutionary agent, performs efficient inverse design for balanced optical and mechanical properties. In semiconductors like silicon and diamond, JANUS explores vacancies and dopants spanning 15 elements in grand-canonical $\mu VT$ ensembles, recovers established defects including the silicon $E$ centre, and identifies new candidate defect pairs and triplets for quantum engineering, including S-Ti in silicon and B-O-O in diamond, with deep in-gap states validated by hybrid-functional density functional theory. By unifying discrete site identities with continuous structural and volumetric relaxation, JANUS provides a foundation for thermodynamic sampling, characterization and inverse design of chemically disordered materials.
LATAS, an efficient sampler that learns a diffusion process to generate Boltzmann-distributed amorphous structures directly from a target energy function, is introduced, establishing ATLAS as a foundation model for sampling, steering and designing amorphous materials.
Mouyang Cheng, Denis Blessing, Botao Yu et al.· 1 citation
We present a high-dimensional neural network potential (HDNNP) for the martensitic phase of the NiTi shape-memory alloy trained to density functional theory (DFT) data. A central aspect of this work is the systematic validation of the potential with respect to the underlying DFT reference method for key properties governing structural evolution, including equilibrium crystal structures, elastic constants, generalized-stacking fault energies, and vibrational spectra. The HDNNP accurately describes the relative stability of the B19$^\prime$ and B33 phases, including subtle energy differences on the order of meV/atom. The predicted stacking-fault energy landscape is strongly anisotropic and reveals a preferential shear pathway, providing atomistic insight into deformation and twinning mechanisms. Finite-temperature molecular dynamics simulations further enable the investigation of unconstrained structural evolution as a function of temperature. Overall, the developed HDNNP provides a robust basis for atomistic simulations of the complex structural and functional behavior of martensitic NiTi systems containing hundreds of thousands of atoms on nanosecond time scales.
P. Jaroš, Petr Sedlák, Petr Šesták et al.· 0 citations
The prediction of stable alloys forming solid-state solutions across large portions of the composition space is a serious theoretical challenge, since one has to evaluate the Gibbs free energy, including both configurational and vibrational contributions. This requires an energy theory capable of extremely high throughput. By taking the Ni-Pd system as prototype, we construct an efficient Jacobi-Legendre machine-learning potential based on density-functional-theory data, which provides accurate energies and forces across the entire composition space. Based on a cluster expansion up to three-body terms and only 873 trainable parameters, this allows us to compute the partition function by directly integrating all accessible microstates, differing for composition, atomic configuration and thermal agitation. We confirm that Ni and Pd are fully miscible, forming an $fcc$ solid-state solution. This is only metastable at room temperature, while becomes thermodynamically stable at around 600~K, with the stability achieved first at the Pd-rich end of the composition range. Interestingly, entropy and heat capacity analysis reveal a competition between the solid-state solution and two intermetallic phases with long-period L1$_0$ structure for NiPd and NiPd$_3$. All in all, our approach offers a powerful and high-throughput workflow for the study of disordered alloys, an approach that can be extended to multi-component systems such as high-entropy alloys.
Rutchapon Hunkao, U. Patil, S. Sanvito· 0 citations
Gas adsorption in porous materials is naturally modeled in the grand-canonical ensemble. We give a compact, self-contained derivation of Metropolis-Hastings acceptance criteria for insertion, deletion, and displacement moves, including nonuniform insertion proposals based on a mixture of Gaussians (MoG) matched to pore geometry. We clarify where indistinguishability and the de Broglie wavelength enter, and show how observables map to adsorption isotherms, isosteric heats, number-fluctuation compressibilities, and the grand potential Ω. Beyond the formulas, we provide a computationally efficient, differentiable reference implementation in JAX and JAX molecular dynamics (JAX-MD), featuring periodic boundaries, cutoff-shifted Lennard-Jones interactions, and static, mask-based arrays amenable to just-in-time compilation (jit) and vectorized mapping (vmap) calculations. We include pragmatic diagnostics (autocorrelation-aware uncertainties, effective sample sizes, acceptance ratios), per-pore and voxel-resolved estimators, visualizations of pores and configurations. The open-source implementation is available online. The full source code, scripts, RASPA2 input files, CIFs, and raw benchmark JSON outputs are accessible at the link above.
Egor I Tuzharov, Yury Maximov, A. Bochenkova· Journal of Chemical Theory a...· 0 citations
Metal surfaces undergo structural, compositional, and morphological changes in response to their chemical environment. Tuning the surfaces'function and stability for a given application correspondingly necessitates an understanding of how this surface evolution couples to external conditions. Here, we demonstrate the feasibility of nested sampling simulations to obtain this coupling at first-principles predictive quality. By exploring the full configuration space, nested sampling estimates the partition function and gives direct access to desired thermodynamic ensemble averages at any temperature without prior knowledge. Computational feasibility is achieved through machine-learned interatomic potentials, an efficient GPU implementation of the sampling algorithm and bespoke sampling moves. Applied to the early oxidation of Cu(100), the approach successfully predicts the experimentally observed, complex $(2\sqrt{2}\times\sqrt{2})$R45$^\circ$-O missing-row reconstruction. The full access to the partition function enables a detailed characterization of the temperature-dependent surface evolution, mapping the emergence of defect states and the order-disorder transition of the reconstructed surface.
F. Riccius, Karsten Reuter, H. H. Heenen et al.· 0 citations
The vast compositional space of high-entropy alloys (HEA) holds promise for next-generation functional materials, yet its exploration is stifled by a combinatorial bottleneck: conventional first-principles accuracy is computationally prohibitive for complex disordered lattices, while black-box machine learning (ML) lacks physical interpretability. We overcome this by establishing a minimal-supercell principle, demonstrating that the magnetostructural behaviour is an emergent feature of local atomic environments rather than long-range configurations. This enables a physics-informed ML framework that inverts the conventional screening approach, transitioning from limited local identification to global design-space mapping, reducing computational time by 40 – 60% without compromising accuracy. Applying this framework to MM’X family uncovers a fundamental design dichotomy: mean electronic descriptors govern baseline phase stability, while dispersion descriptors, specifically the magnetic-moment dispersion, drive the critical transformation response. This analysis reveals an intrinsic stability-performance trade-off, where the disorder required to maximise the transformation driving force inevitably penalises thermodynamic stability. By quantifying this Pareto frontier, we propose a hierarchical tuning strategy that decouples phase transition temperature from hysteresis, providing a scalable paradigm to transform HEA exploration from serendipitous discovery into rational design.
Zhe Cui, C. Romero-Muñiz, J. Law et al.· Nature Communications· 0 citations