Materials discovery, especially for applications that require extreme operating conditions, requires extensive testing that naturally limits the ability to inquire the wealth of possible ...compositions. Machine Learning (ML) has nowadays a well-established role in facilitating this effort in systematic ways. The increasing amount of available accurate Density Functional Theory (DFT) data represents a solid basis upon which new ML models can be trained and tested. While conventional models rely on static descriptors, generally suitable for a limited class of systems, the flexibility of Graph Neural Networks (GNNs) allows for direct learning representations on graphs, such as the ones formed by crystals. We utilize crystal graph neural networks (CGNNs) known to predict crystal properties with DFT level accuracy through graphs by encoding the atomic (node/vertex), bond (edge), and global state attributes. In this work, we aim at testing the ability of the CGNN MegNet framework in predicting a number of properties of systems previously unseen in the model, which are obtained by adding a substitutional defect to bulk crystals that are included in the training set. We perform DFT validation to assess the accuracy in the prediction of formation energies and structural features (such as elastic moduli). Using CGNNs, one may identify promising paths in alloy discovery.
Materials design has traditionally evolved through trial-error approaches, mainly due to the non-local relationship between microstructures and properties such as strength and toughness. We propose ...‘alloy informatics’ as a machine learning based prototype predictive approach for alloys and compounds, using electron charge density profiles derived from first-principle calculations. We demonstrate this framework in the case of hydrogen interstitials in face-centred cubic crystals, showing that their differential electron charge density profiles capture crystal properties and defect-crystal interaction properties. Radial Distribution Functions (RDFs) of defect-induced differential charge density perturbations highlight the resulting screening effect, and, together with hydrogen Bader charges, strongly correlate to a large set of atomic properties of the metal species forming the bulk crystal. We observe the spontaneous emergence of classes of charge responses while coarse-graining over crystal compositions. Nudge-Elastic-Band calculations show that RDFs and charge features also connect to hydrogen migration energy barriers between interstitial sites. Unsupervised machine-learning on RDFs supports classification, unveiling compositional and configurational non-localities in the similarities of the perturbed densities. Electron charge density perturbations may be considered as bias-free descriptors for a large variety of defects.
Display omitted
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZRSKP
Abstract
By introducing a suitable range-separation of the Coulomb coupling in analogy to Ambrosetti
et al
(2014
J. Chem. Phys.
140
18A508), here we extend the many-body dispersion approach to ...include beyond-dipole van der Waals (vdW) interactions at a full many-body level, in combination with semi-local density functional theory. A reciprocal-space implementation is further introduced in order to efficiently treat periodic systems. Consistent reliability is found from molecular dimers to large supramolecular complexes and two-dimensional systems. The large weight of both many-body effects and multipolar terms illustrates how a correct description of vdW forces in large-scale systems requires full account of both contributions, beyond standard pairwise dipolar approaches.
Material characterization in nano-mechanical tests may provide information on the potential heterogeneity of mechanical properties. Here, we develop a robust neural-network interatomic potential ...(NNIP), and we provide a test for the example of molecular dynamics (MD) nanoindentation, and the case of body-centered cubic crystalline molybdenum (Mo). We employ a similarity measurement protocol, using standard local environment descriptors, to select ab initio configurations for the training dataset that capture the behavior of the indented sample. We find that it is critical to include generalized stacking fault (GSF) configurations, featuring a dumbbell self-interstitial on the surface, to capture dislocation cores, and also high-temperature configurations with frozen atom layers for the indenter tip contact. We develop a NNIP with distinct dislocation nucleation mechanisms, realistic generalized stacking fault energy (GSFE) curves, and an informative energy landscape for the atoms on the sample surface during nanoindentation. We compare our NNIP results with nanoindentation simulations, performed with three existing potentials – an embedded atom method (EAM) potential, a gaussian approximation potential (GAP), and a tabulated GAP (tabGAP) potential – that predict different dislocation nucleation mechanisms, and display the absence of essential information on the shear stress at the sample surface in the elastic region. Finally, we compared our NNIP nanoindentation results with experiments, showing reliable predictions for reduced Young’s modulus and observable slip traces.
Display omitted
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZRSKP
In this work, we evaluate the role of agronomic factors in the selection for herbicide resistance in Apera spica-venti L. Beauv. (silky windgrass). During a period of three years, populations were ...collected in more than 250 conventional fields across Europe and tested for resistance in the greenhouse. After recording the field history of locations, a geo-referenced database has been developed to map the distribution of herbicide-resistant A. spica-venti populations in Europe. A Logistic Regression Model was used to assess whether and to what extent agricultural and biological factors (crop rotation, soil tillage, sowing date, soil texture and weed density) affect the probability of resistance selection apart from the selection pressure due to herbicide application. Our results revealed that rotation management and soil tillage are the factors that have the greatest influence on the model. In addition, first order interactions between these two variables were highly significant. Under conventional tillage, a percentage of winter crops in the rotation exceeding 75% resulted in a 1280-times higher risk of resistance selection compared to rotations with less than 50% of winter crops. Under conservation tillage, the adoption of >75% of winter crops increased the risk of resistance 13-times compared to rotations with less than 50% of winter crops. Finally, early sowing and high weed density significantly increased the risk of resistance compared to the reference categories (later sowing and low weed density, respectively). Soil texture had no significant influence. The developed model can find application in management programs aimed at preventing the evolution and spread of herbicide resistance in weed populations.
In this study, the results of two-year investigations on herbicide resistance in silky bent grass (Apera spica-venti) populations are presented. Whole-plant bioassays were conducted with different ...herbicides on over 250 A spica-venti populations from Central and Eastern European agricultural fields where herbicides failed to achieve satisfactory control. Results showed that over 60% of the suspected populations could be rated resistant to acetolactate synthase (ALS)-inhibitors, resistance to acetyl-CoA (ACCase)-inhibitors could be observed in only a few cases and no resistance to photosystem II (PSII)-inhibitors was detected. Dose-response experiments conducted in the greenhouse on resistant populations with the herbicides flupyrsulfuron-methyl, mesosulfuron+iodosulfuron and fenoxaprop-P-ethyl revealed resistance factors at ED₅₀ and ED₉₀ ranging respectively from 11 to 142, from 2 to 15 and from 4 to 6, thus confirming the prevalence of resistance to ALS-inhibitors in A spica-venti. In greenhouse experiments, percentage canopy cover after herbicide treatment was determined in susceptible and resistant populations for the herbicides sulfosulfuron and fenoxaprop-P-ethyl by using digital image analysis. A significant effect of herbicide dose on canopy cover was observed in susceptible plants 7 and 15 days after treatment with sulfosulfuron, as well as in all populations when treated with fenoxaprop-P-ethyl. Canopy cover correlated significantly with plant dry weight in all populations, thus indicating that digital image analysis may represent a valid alternative approach to whole-plant bioassays and dose-response analysis for estimating biomass reduction after herbicide treatment. This work provides weed scientists with reliable tools for the verification of herbicide resistance in suspected weed populations.
In this work, the dynamic realms of Materials Science and Computer Science
advancements meet the critical challenge of identifying efficient descriptors
capable of capturing the essential features of ...physical systems. Such task has
remained formidable, with solutions often involving ad-hoc scalar and vectorial
sets of materials properties, making optimization and transferability
challenging. We extract representations directly from ab-initio differential
electron charge density profiles using Neural Networks, highlighting the
pivotal role of transfer learning in such task. Firstly, we demonstrate
significant improvements in regression of a specific defected-materials
property with respect to training a deep network from scratch, both in terms of
predictions and their reproducibilities, by considering various pre-trained
models and selecting the optimal one after fine-tuning. The remarkable
performances obtained confirmed the transferability of the existent pre-trained
Convolutional Neural Networks (CNNs) on physics domain data, very different
from the original training data. Secondly, we demonstrate a saturation in the
regression capabilities of computer vision models towards properties of an
extensive variety of undefected systems, and how it can be overcome with the
help of large language model (LLM) transformers, with as little text
information as composition names. Finally, we prove the insufficiency of
open-models, like GPT-4, in achieving the analogous tasks and performances as
the proposed domain-specific ones. The work offers a promising avenue for
enhancing the effectiveness of descriptor identification in complex physical
systems, shedding light over the power of transfer learning to easily adapt and
combine available models, with different modalities, to the physics domain, at
the same time opening space to a benchmark for LLMs capabilities in such
domain.
Materials design has traditionally evolved through trial-error approaches,
mainly due to the non-local relationship between microstructures and properties
such as strength and toughness. We propose ...'alloy informatics' as a machine
learning based prototype predictive approach for alloys and compounds, using
electron charge density profiles derived from first-principle calculations. We
demonstrate this framework in the case of hydrogen interstitials in
face-centered cubic crystals, showing that their differential electron charge
density profiles capture crystal properties and defect-crystal interaction
properties. Radial Distribution Functions (RDFs) of defect-induced differential
charge density perturbations highlight the resulting screening effect, and,
together with hydrogen Bader charges, strongly correlate to a large set of
atomic properties of the metal species forming the bulk crystal. We observe the
spontaneous emergence of classes of charge responses while coarse-graining over
crystal compositions. Nudge-Elastic-Band calculations show that RDFs and charge
features also connect to hydrogen migration energy barriers between
interstitial sites. Unsupervised machine-learning on RDFs supports
classification, unveiling compositional and configurational non-localities in
the similarities of the perturbed densities. Electron charge density
perturbations may be considered as bias-free descriptors for a large variety of
defects.
Materials discovery, especially for applications that require extreme operating conditions, requires extensive testing that naturally limits the ability to inquire the wealth of possible ...compositions. Machine Learning (ML) has nowadays a well established role in facilitating this effort in systematic ways. The increasing amount of available accurate DFT data represents a solid basis upon which new ML models can be trained and tested. While conventional models rely on static descriptors, generally suitable for a limited class of systems, the flexibility of Graph Neural Networks (GNNs) allows for direct learning representations on graphs, such as the ones formed by crystals. We utilize crystal graph neural networks (CGNN) to predict crystal properties with DFT level accuracy, through graphs with encoding of the atomic (node/vertex), bond (edge), and global state attributes. In this work, we aim at testing the ability of the CGNN MegNet framework in predicting a number of properties of systems previously unseen from the model, obtained by adding a substitutional defect in bulk crystals that are included in the training set. We perform DFT validation to assess the accuracy in the prediction of formation energies and structural features (such as elastic moduli). Using CGNNs, one may identify promising paths in alloy discovery.
Understanding and accurately predicting hydrogen diffusion in materials is challenging due to the complex interactions between hydrogen defects and the crystal lattice. These interactions span large ...length and time scales, making them difficult to address with standard ab initio techniques. This work addresses this challenge by employing accelerated machine learning (ML) molecular dynamics simulations through active learning. We conduct a comparative study of different ML-based interatomic potential schemes, including VASP, MACE, and CHGNet, utilizing various training strategies such as on-the-fly learning, pre-trained universal models, and fine-tuning. We obtain an optimal hydrogen diffusion coefficient value of \(2.1 \cdot 10^{-8}\) m\(^2\)/s at 673 K in MgH\(_{0.06}\), which aligns exceptionally well with experimental results, underlining the efficacy and accuracy of ML-assisted methodologies in the context of diffusive dynamics. Particularly, our procedure significantly reduces the computational effort associated with traditional transition state calculations or ad-hoc designed interatomic potentials. The results highlight the limitations of pre-trained universal solutions for defective materials and how they can be improved by fine-tuning. Specifically, fine-tuning the models on a database produced during on-the-fly training of VASP ML force field allows the retrieving of DFT-level accuracy at a fraction of the computational cost.