Recycling steel at scale is hindered by tramp elements such as Cu and Sn, which degrade material properties. Atomistic simulations using foundational machine-learned interatomic potentials (MLIPs) trained on large databases, such as Materials Project, Alexandria, and OMAT, offer a promising approach to study the effects of these impurities. However, fine-tuning these models to specific systems can lead to catastrophic forgetting–the loss of general chemical knowledge acquired during pretraining. Here, we evaluate forgetting in three foundational MLIPs: CHGNet, SevenNet-O, and MACE, by fine-tuning on a data set of bcc-based structures, with Fe atoms only. When evaluated on a subset of the Materials Project data set with a learning rate of 0.0001, the fine-tuned MLIPs of CHGNet and SevenNet-O exhibited only a minor increase in RMSE of 0.047 and 0.022 eV/atom, respectively, indicating markedly minor forgetting. In contrast, fine-tuned MACE exhibited catastrophic forgetting, despite a range of additional strategies such as layer freezing and data set replay. We attribute the catastrophic forgetting to architectural sensitivity. These results highlight the importance of fine-tuning hyperparameters, model architecture, and data set design, with fine-tuned models of CHGnet and SevenNet-O showing some potential for efficient and transferable modeling of recycled steels.

Universal machine-learning interatomic potentials (uMLIPs) are emerging as foundation models for atomistic simulation, offering near-ab initio accuracy at far lower cost. Their safe, broad deployment is limited by the absence of reliable, general uncertainty estimates. We present a unified, scalable uncertainty metric, U, built from a heterogeneous ensemble that reuses existing pretrained MLIPs. Across diverse chemistries and structures, U strongly tracks true prediction errors and robustly ranks configuration-level risk. Using U, we perform uncertainty-aware distillation to train system-specific potentials with far fewer labels: for tungsten, we match full density-functional-theory (DFT) training using 4% of the DFT data; for MoNbTaW, a dataset distilled by U supports high-accuracy potential training. By filtering numerical label noise, the distilled models can in some cases exceed the accuracy of the MLIPs trained on DFT data. This framework provides a practical reliability monitor and guides data selection and fine-tuning, enabling cost-efficient, accurate, and safer deployment of foundation models.

Cold-drawn pearlite wire is widely used in industry due to its exceptional high strength. Understanding the deformation mechanisms during the cold-drawing process of pearlite, particularly the deformation and decomposition of cementite, is of great significance. In this study, a bicrystal model tailored to lamellar structures is developed to calculate the elastic properties and stress concentration of pearlite. By analyzing slip activation in both ferrite and cementite, along with the yield strength, we reveal the significant influence of loading direction on pearlite deformability. Notably, the yield strength varies from 9.5 GPa to 17.0 GPa. Under specific loading conditions, plastic deformation is observed to initiate in cementite, challenging the conventional assumption that slip bands always originate in ferrite. Furthermore, factors that influence the plastic deformation of pearlite are discussed. A successive strengthening mechanism is proposed to explain the excellent deformability and high strength of pearlite after extensive deformation. This work introduces a novel method for directional loading of lamellar structures. The surprising finding that plastic deformation, without fracture, can initiate in cementite, might offer directions for developing other structural materials with extreme tensile strength and deformability.

In 1970, Hillert and Staffansson published a paper entitled “The Regular Solution Model for Stoichiometric Phases and Ionic Melts”. It was the beginning of the sublattice model that has been a key component in the development of Computational Thermodynamics. This formalism, now often called the Compound Energy Formalism (CEF), has been used to describe a great variety of phases driven by the need for accurate descriptions of thermodynamic phase stability in a wide range of materials involving many elements. The purpose of this paper is to describe the formalism, the physical meaning of its various parameters and the way they can be assessed using experimental and theoretical data. Furthermore, new developments derived from the CEF, such as the Effective Bond Energy Formalism, and other ideas for further development are presented.

Current deep neural networks (DNNs) are overparameterized and use most of their neuronal connections during inference for each task. The human brain, however, developed specialized regions for different tasks and performs inference with a small fraction of its neuronal connections. We propose an iterative pruning strategy introducing a simple importance-score metric that deactivates unimportant connections, tackling overparameterization in DNNs and modulating the firing patterns. The aim is to find the smallest number of connections that is still capable of solving a given task with comparable accuracy, i.e. a simpler subnetwork. We achieve comparable performance for LeNet architectures on MNIST, and significantly higher parameter compression than state-of-the-art algorithms for VGG and ResNet architectures on CIFAR-10/100 and Tiny-ImageNet. Our approach also performs well for the two different optimizers considered—Adam and SGD. The algorithm is not designed to minimize FLOPs when considering current hardware and software implementations, although it performs reasonably when compared to the state of the art.

Rapid developments in the field of hydrogen energy have prompted the need for safe and efficient hydrogen transportation and storage. Steels form the backbone of the current energy infrastructure and thus offer a fast and cost-effective solution. Their excellent mechanical properties are attributed to the underlying microstructure which comprises of finely dispersed nano-precipitates. However, one major factor restricting their application is their susceptibility to Hydrogen Embrittlement (HE). In the past decade, experimental and theoretical works have been carried out to understand if the nano-sized carbides can aid in reducing the susceptibility to HE along with providing strengthening. Within this ab-inito study, we investigated the effectiveness of fully coherent nano-carbides (i.e. TiC, VC and NbC) to limit the diffusible hydrogen content in bcc Fe. Our study revealed that the interplay between hydrogen and carbon vacancies, local atomic environment at interface as well as elastic strain fields at the interface can lead to significantly increased hydrogen solubilities. While in TiC, the deepest traps were found to be in the bulk of carbides, in VC and NbC, the elastic strain fields around the interface led to the strongest trapping. Further, the formation of a two-hydrogen-vacancy complex was found to be favourable in VC. Finally, the migration barriers for hydrogen trapping in bulk TiC as well as across the Fe/TiC coherent interface indicate that these deep traps in the form of carbon vacancies are fairly accessible.

In a material under stress, grain boundaries may give rise to stress discontinuities. The stress state at grain boundaries strongly affects microscopic processes, such as diffusion and segregation, as well as failure initiation, such as fatigue, creep, and corrosion. Here the general condition of incompatibility stress at grain boundaries is studied with a bicrystal model for linear elastic materials. In materials with cubic crystal structures, it is proven that hydrostatic stress does not lead to a stress discontinuity at grain boundaries. For bicrystals with inclined grain boundaries under uniaxial stress, the extreme values of the incompatibility stress as a function of the inclination angle are obtained by a simulated annealing method. A simple criterion is proposed to classify cubic materials into three groups. For cubic crystals with at most moderate anisotropy, the highest incompatibility stress occurs when the grain boundary plane is perpendicular to the uniaxial stress. For highly anisotropic materials, such as alkali metals and polymorphic high-temperature phases, the highest incompatibility stress occurs on grain boundaries with an inclination of about 47^o.

Carbide nano-precipitates are commonly used to improve mechanical properties of steel. It has been experimentally observed that TiC, NbC, and VC carbide precipitates initially form as ‘plate-like’ particles oriented in the {1 0 0} planes of the ferrite lattice. These platelets share similarities with Guinier-Preston zones in Al-Cu alloys. The clustering of group IV and V transition metal atoms (M = Ti, Zr, Hf, V, Nb, Ta) in ferrite is studied using density functional theory. It is deduced that the transition metal carbides all form in a similar way. Furthermore, the transition from an initial M–C cluster to a NaCl-structured platelet to a NaCl-structured precipitate is examined through atomistic simulations using Modified Embedded Atom Method potentials. A route is established along which transition metal carbides form and transform into precipitates that possess the Baker-Nutting orientation relation with the ferrite matrix.

Physics-informed neural networks (PINNs) have recently become a powerful tool for solving partial differential equations (PDEs). However, finding a set of neural network parameters that fulfill a PDE at the boundary and within the domain of interest can be challenging and non-unique due to the complexity of the loss landscape that needs to be traversed. Although a variety of multi-task learning and transfer learning approaches have been proposed to overcome these issues, no incremental training procedure has been proposed for PINNs. As demonstrated herein, by developing incremental PINNs (iPINNs) we can effectively mitigate such training challenges and learn multiple tasks (equations) sequentially without additional parameters for new tasks. Interestingly, we show that this also improves performance for every equation in the sequence. Our approach learns multiple PDEs starting from the simplest one by creating its own subnetwork for each PDE and allowing each subnetwork to overlap with previously learned subnetworks. We demonstrate that previous subnetworks are a good initialization for a new equation if PDEs share similarities. We also show that iPINNs achieve lower prediction error than regular PINNs for two different scenarios: (1) learning a family of equations (e.g., 1-D convection PDE); and (2) learning PDEs resulting from a combination of processes (e.g., 1-D reaction–diffusion PDE). The ability to learn all problems with a single network together with learning more complex PDEs with better generalization than regular PINNs will open new avenues in this field.

The human brain is capable of learning tasks sequentially mostly without forgetting. However, deep neural networks (DNNs) suffer from catastrophic forgetting when learning one task after another. We address this challenge considering a class-incremental learning scenario where the DNN sees test data without knowing the task from which this data originates. During training, Continual Prune-and-Select (CP&S) finds a subnetwork within the DNN that is responsible for solving a given task. Then, during inference, CP&S selects the correct subnetwork to make predictions for that task. A new task is learned by training available neuronal connections of the DNN (previously untrained) to create a new subnetwork by pruning, which can include previously trained connections belonging to other subnetwork(s) because it does not update shared connections. This enables to eliminate catastrophic forgetting by creating specialized regions in the DNN that do not conflict with each other while still allowing knowledge transfer across them. The CP&S strategy is implemented with different subnetwork selection strategies, revealing superior performance to state-of-the-art continual learning methods tested on various datasets (CIFAR-100, CUB-200-2011, ImageNet-100 and ImageNet-1000). In particular, CP&S is capable of sequentially learning 10 tasks from ImageNet-1000 keeping an accuracy around 94% with negligible forgetting, a first-of-its-kind result in class-incremental learning. To the best of the authors’ knowledge, this represents an improvement in accuracy above 10% when compared to the best alternative method.

In a material under stress, grain boundaries may give rise to stress discontinuities. Stress localization is crucial to materials' behavior such as segregation, precipitation, and void nucleation. Here, the stress state at a grain boundary perpendicular to a uniaxial external stress is studied systematically. The grain boundary with the most extreme stress discontinuity is determined for cubic materials within the elastic limit for a bicrystal model. Additionally, grain boundaries with negligible stress discontinuity are identified. The influence of the elastic tensor components, C₁₁, C₁₂, and C₄₄, and grain orientation is studied quantitatively.

In this study possible routes from dissolved M and C atoms to a M-C (M = Ti, Nb) cluster are studied. Using atomistic modelling to perform relaxation simulations and molecular dynamics (MD) simulations for the Fe-M-C ternary system, the formation of clusters is studied for M. Additionally the stability of M-C clusters is assessed. The clustering of M and C atoms as observed in experiments is also found in simulations. The initial clusters found in this work have a (Fe,M)C composition with a large Fe fraction. Moreover, structurally relaxed clusters reveal that there are growth pathways with a monotone decrease in Gibbs energy, suggesting that the highest energy barrier in the formation of M-C clusters is the diffusion barrier for the atoms forming the cluster. The development of M-C clusters as found in this study suggests a formation mechanism for nano-precipitation of carbides consisting of several steps; first a C cluster forms, then M atoms attach to the C cluster forming a (Fe,M)C cluster, and in the final step the (Fe,M)C cluster transforms to a NaCl-structured carbide.

A reference-free modified embedded atom method (RF-MEAM) potential for iron has been constructed. The new potential is made to predict both bcc and fcc (α-Fe and γ-Fe) lattice properties, with a special interest in modelling in the 800-1300 K temperature range. This is the range in which transformations and key processes in steel occur. RF-MEAM potentials can be used directly in commonly used molecular dynamics simulation software (e.g. LAMMPS). The new potential is compared to several other (M)EAM potentials which are commonly used. It is demonstrated that the new potential combines good characteristics for point defect energies with free surface and stacking fault energies. Also the Nishiyama-Wassermann and Kurdjumov-Sachs orientation relation ratios and interface energies are reproduced, allowing for simulations of α-Fe and γ-Fe interphases.

γ-M₂₃C₆ carbide often forms at grain boundaries in creep-resistant steels and plays a crucial role in creep resistance by blocking microstructural changes at elevated temperatures. Given that the dislocations and stacking faults (SFs) in carbides may affect their stability, the observation of SF formation in γ-M₂₃C₆ using atomic-scale microscopy has implications for Cr-rich creep-resistant steels. Our analysis of SF energies (SFEs) derived from density functional theory calculations reveals that the SFEs are high, which suggests that the SF is not induced by external shear stress deformation but by lattice misfit between conjoined subgrains nucleated from the same austenite grain.

A new generation of radiation detectors relies on the crystalline Si and amorphous B (c-Si/a-B) junctions that are prepared through chemical vapor deposition of diborane (B₂H₆) on Si at low temperature (∼400 C). The Si wafer surface is dominated by the Si{0 0 1}3 1 domains that consist of two different Si species at low temperature. Here we investigate the geometry, stability and electronic properties of the hydrogen passivated Si{0 0 1}3 1 surfaces with deposited BH_n (n = 0 to 3) radicals using parameter-free first-principles approaches. Ab initio molecular dynamics simulations using the density functional theory (DFT) including van der Waals interaction reveal that in the initial stage the BH₃ molecules/radicals deposit on the Si(-H), forming (-Si)BH₄ radicals which then decompose into (-Si)BH₂ with release of H₂ molecules. Structural optimizations provide strong local relaxation and reconstructions at the deposited Si surface. Electronic structure calculations reveal the formation of various defect states in the forbidden gap. This indicates limitations of the presently used rigid electron-counting and band-filling models. The attained information enhances our understanding of the initial stage of the PureB process and the electric properties of the products.

Single-crystal copper films on sapphire have recently been reported upon in relation to graphene growth on these films. In the present paper the kinetics of the formation of single crystal copper films is investigated. We demonstrate the importance of heating the sapphire substrate in 1000 hPa oxygen, followed by a fast cooling prior to depositing the copper film. The importance of this treatment is tentatively explained by the dissolution of oxygen in sapphire and subsequent out-diffusion during recrystallization of the copper film to form a copper-oxide interface layer. Also, the importance of avoiding oxygen incorporation in the sputter deposited film is demonstrated.