The quality, consistency, and information content of training data is often what determines the practical value of machine-learning models for atomistic simulations. Yet, many widely used electronic-structure databases are assembled having materials screening as primary goal rather than robust force-field learning, are limited in their scope to a specific class of chemical compounds, and/or employ inconsistent DFT functionals and settings. Here we introduce MAD-1.5, a highly curated dataset designed explicitly for training broadly applicable atomistic models across the periodic table at high levels of theory. MAD-1.5 extends the MAD dataset with targeted enrichment strategies that improve the coverage of chemical space to 102 elements while keeping the total number of configurations compact. All structures are computed with a single, standardized all-electron DFT workflow using the r2SCAN meta-GGA functional and consistent convergence settings, ensuring uniformity across chemically heterogeneous systems. The dataset encompasses molecules, clusters, bulk crystals, surfaces, and low-dimensional structures, and its quality and consistency are further enhanced by outlier removal using uncertainty quantification. We demonstrate the high accuracy that can be achieved with the proposed dataset by training PET-MAD-1.5, a generally applicable r2SCAN interatomic potential that covers 102 elements in the periodic table and achieves exceptional levels of benchmark accuracy and stability in challenging simulation protocols.

Two-dimensional (2D) materials are among the most promising candidates for beyond silicon electronic and optoelectronic applications. Recently, their recognized importance, sparked a race to discover and characterize new 2D materials. Within few years the number of experimentally exfoliated or synthesized 2D materials went from a couple of dozens to few hundreds while the number theoretically predicted compounds reached a few thousands. In 2018 we first contributed to this effort with the identification of 1825 compounds that are either easily (1036) or potentially (789) exfoliable from experimentally known 3D compounds. In the present work we report on the new materials recently added to the 2D-portfolio thanks to the extension of the screening to an additional experimental database (MPDS) as well as the most up-to-date versions of the two databases (ICSD and COD) used in our previous work. This expansion led to the discovery of an additional 1252 unique monolayers bringing the total to 3077 compounds and, notably, almost doubling the number of easily exfoliable materials (2004). Moreover, we optimized the structural properties of all the materials (regardless of their binding energy or number of atoms in the unit cell) as isolated mono-layer and explored their electronic band structure. This archive entry contains the database of 2D materials in particular it contains the structural parameters for all the 3077 structures of the global Material Cloud 2D database as extracted from their bulk 3D parent, 2710 optimized 2D structures and 2345 electronic band structure together with the provenance of all data and calculations as stored by AiiDA.

Density functional theory (DFT) is a widely used method to compute properties of materials, which are often collected in databases and serve as valuable starting points for further studies. In this article, we present the Materials Cloud Three-Dimensional Structure Database (MC3D), an online database of computed 3D inorganic crystal structures. Close to a million experimentally reported structures were imported from the COD, ICSD and MPDS databases; these were parsed and filtered to yield a collection of 72589 unique and stoichiometric structures, of which 95% are, to date, classified as experimentally known. The geometries of structures with up to 64 atoms were then optimized using DFT with automated workflows and curated input protocols. The procedure was repeated for different functionals (and computational protocols), with the latest version (MC3D PBEsol-v2) comprising 32013 unique structures. All versions of the MC3D are made available on the Materials Cloud portal, which provides a graphical interface to explore and download the data. The database includes the full provenance graph of all the calculations driven by the automated workflows, thus establishing full reproducibility of the results and more-than-FAIR procedures.

This data entry includes the data with full provenance for all three versions: PBE-v1, PBEsol-v1 and PBEsol-v2.

The Materials Cloud three-dimensional database is a curated set of relaxed three-dimensional crystal structures based on raw CIF data taken from the external experimental databases MPDS, COD and ICSD. The raw CIF data have been imported, cleaned and parsed into a crystal structure; their ground-state has been computed using the SIRIUS-enabled pw.x code of the Quantum ESPRESSO distribution, and tight tolerance criteria for the calculations using the SSSP protocols.

This entire procedure is encoded into an AiiDA workflow which automates the process while keeping full data provenance. Here, since the original source data of the ICSD and MPDS databases are copyrighted, only the provenance of the final SCF calculation on the relaxed structures can be made publicly available.

The MC3D ID numbers come from a list of unique "parent" stoichiometric structures that has been created and curated from a collection of these experimental databases. Once a parent structure has been optimized using density-functional theory, it is made public and added to the online Discover section of the Materials Cloud (as mentioned, copyright might prevent publishing the original parent). Note that since not all structures have been calculated, some ID numbers are missing from the public version of the database. The full ID of each structure also contains as an appended modifier the functional that was used in the calculations. Since the ID number points to the same unique parent, mc3d-1234/pbe and mc3d-1234/pbesol have the same starting point, but have been then relaxed according to their respective functionals.

In the past decade we have witnessed the appearance of large databases of calculated material properties. These are most often obtained with the Perdew-Burke-Ernzerhof (PBE) functional of density-functional theory, a well established and reliable technique that is by now the standard in materials science. However, there have been recent theoretical developments that allow for an increased accuracy in the calculations. Here, we present the updated alexandria dataset of calculations for more than 415k solid-state materials obtained with two improved functionals: PBE for solids (that yields consistently better geometries than the PBE) and SCAN (probably the best all-around functional at the moment). Our results provide an accurate overview of the landscape of stable (and nearly stable) materials, and as such can be used for more reliable predictions of novel compounds. They can also be used for training machine learning models, or even for the comparison and benchmark of PBE, PBE for solids, and SCAN.

This work presents an application of the strain-fluctuation method, exploiting the fluctuations of the strain from extensive first-principles molecular dynamics simulations in the isobaric-isothermal ensemble, to the study of the elastic tensors of superionic materials. As the superionic materials for solid-state electrolyte applications usually do not have well-defined ground-state configurations, it is challenging to apply the static methods to calculate the elastic tensors of these materials. Instead, the strain-fluctuation method captures the dynamical nature of the elastic response of these materials and is a promising approach to studying their elastic properties. In this work: a protocol is presented and documented to extract the elastic the elastic moduli and their statistical errors from the molecular dynamics trajectories (open-source code available at https://github.com/materzanini); results for two benchmark superionic materials (Li₁₀GeP₂S₁₂ and Li₁₀GeP₂O₁₂) are given; for these superionic materials, a comparison to static methods is also provided, showing that static methods overestimate the moduli with respect to the correct dynamical treatment by ~25-50%.

The dataset provides records of tensile properties of nuclear structural materials. The focus is on studying the influence of specimen dimensions and geometry on mechanical properties such as yield strength, ultimate tensile strength, uniform elongation, and total elongation. The dataset was created through an extensive literature review of scientific articles and databases. The search inclusion criteria targeted peer-reviewed studies on tensile testing of sub-sized specimens, providing quantitative data on tensile properties relative to specimen size. The extracted data points from the literature review were organized into a tabular format database containing 1,070 tensile testing records with 54 parameters, including material type and composition, manufacturing information, irradiation conditions, specimen size and dimensions, and tensile properties. Materials science experts conducted systematic checks to validate the collected data, ensuring accuracy in the material type, manufacturing processes and treatment methods, and testing conditions, as well as verifying the chemical composition and other pertinent information. Our team performed statistical analyses to identify and address data outliers, ensuring the reliability of the dataset.

Machine Learning (ML) can be conveniently applied to continuum materials simulations, allowing for the investigation of larger systems and longer timescales, pushing the limits of tractable systems. Here we provide a comprehensive dataset of strained Ge films on Si and their corresponding strain states, which can be used to train a ML model capable of such acceleration. Approximately 80k 2D cases are included, reporting the profiles h(x) and the corresponding elastic energy densities and strain fields. The profiles are conveniently sampled using Perlin-noise and pure-sine waves. A 100nm-large computational domain is considered. The mechanical equilibrium problem is solved using Finite Element Method (FEM). Ge is modeled as an isotropic material and an eigenstrain of 3.99% is used, as in Ge/Si(001). The database has been exploited for training a (fully) Convolutional Neural Network (CNN) which maps the free surface profile h(x) to the corresponding energy density. If plugged into the proper time-dependent Partial Differential Equation, this term can be used to accelerate continuum simulations of the morphological evolution of strained films while retaining FEM-level accuracy. Tests of the reliability of such CNN model are also provided in the repository, together with the output of surface morphology minimization procedures and morphological evolution simulations during coarsening and growth. In the latter, evolution by surface diffusion has been considered as an important case, but applications to other mechanisms are possible. Generalization examples to larger computational cells with respect to those in the dataset are also available.

Materials are often represented in machine learning applications by (chemical-)geometric descriptions of their atomic structure. In this work, we propose an alternative framework for representing materials using descriptions of their electronic structure called Spectral Operator Representations (SOREPs). This record contains the code and data used to study carbon nanotubes (CNTs), barium titanate polymorphs, and the accelerated screening of transparent conducting materials with SOREPs. A data set for each application is provided: p_z tight binding band structures for the three CNT configurations studied; the structures, band dispersions, and SOREP features of 127 BaTiO₃ polymorphs; and the SOREP features and ML targets for the MC3D materials considered in the accelerated screening. Additionally, code including patch files for Quantum ESPRESSO, the "sorep" python package, and the set of scripts used to prepare these data, train ML models, and plot results is provided.

Atomistic simulations provide insights into structure-property relations on an atomic size and length scale that are complementary to the macroscopic observables that can be obtained from experiments. Quantitative predictions, however, are usually hindered by the need to strike a balance between the accuracy of the calculation of the interatomic potential and the modelling of realistic thermodynamic conditions. Machine-learning techniques make it possible to efficiently approximate the outcome of accurate electronic-structure calculations that can, therefore, be combined with extensive thermodynamic sampling. We take elemental nickel as a prototypical material, whose alloys have applications from cryogenic temperatures up to close to their melting point and use it to demonstrate how a combination of machine-learning models of electronic properties and statistical sampling methods makes it possible to compute accurate finite-temperature properties at an affordable cost. We demonstrate the calculation of a broad array of bulk, interfacial, and defect properties over a temperature range from 100 to 2500 K, modelling also, when needed, the impact of nuclear quantum fluctuations and electronic entropy. The framework we demonstrate here can be easily generalized to more complex alloys and different classes of materials.

Long-range electrostatic interactions critically affect polar materials. However, state-of-the-art atomistic potentials, such as neural networks or Gaussian approximation potentials employed in large-scale simulations, often neglect the role of these long-range electrostatic interactions.This study introduces a novel model derived from first principles to evaluate the contribution of long-range electrostatic interactions to total energies, forces, and stresses. The model is designed to integrate seamlessly with existing short-range force fields without further first-principles calculations or retraining. The approach relies solely on physical observables, like the dielectric tensor and Born effective charges, that can be consistently calculated from first principles. We demonstrate that the model reproduces critical features, such as the LO-TO splitting and the long-wavelength phonon dispersions of polar materials, with benchmark results on the cubic phase of barium titanate (BaTiO3).

This dataset reports the raw dynamical matrices computed for BaTiO3 using DFPT, the bare short-range GAP potential, and the long-range corrected GAP potential. A more accurate description of the methodology and all the parameters used to reproduce these data are discussed and reported in the referenced paper.

Crystal-graph attention networks have emerged recently as remarkable tools for the prediction of thermodynamic stability and materials properties from unrelaxed crystal structures. Previous networks trained on two million materials exhibited, however, strong biases originating from underrepresented chemical elements and structural prototypes in the available data. We tackled this issue computing additional data to provide better balance across both chemical and crystal-symmetry space. Crystal-graph networks trained with this new data show unprecedented generalization accuracy, and allow for reliable, accelerated exploration of the whole space of inorganic compounds. We applied this universal network to performed machine-learning assisted high-throughput materials searches including 2500 binary and ternary prototypes and spanning about 1 billion compounds. After validation using density-functional theory, we uncover in total 19512 additional materials on the convex hull of thermodynamic stability and around 150000 compounds with a distance of less than 50 meV/atom from the hull. Here we include the DCGAT-1, DCGAT-2, and DCGAT-3 datasets used in this work.

Two-dimensional (2D) materials have emerged as promising candidates for next-generation electronic and optoelectronic applications. Yet, only a few dozens of 2D materials have been successfully synthesized or exfoliated. Here, we search for novel 2D materials that can be easily exfoliated from their parent compounds. Starting from 108423 unique, experimentally known three-dimensional compounds we identify a subset of 5619 that appear layered according to robust geometric and bonding criteria. High-throughput calculations using van-der-Waals density-functional theory, validated against experimental structural data and calculated random-phase-approximation binding energies, allow to identify 1825 compounds that are either easily or potentially exfoliable. In particular, the subset of 1036 easily exfoliable cases provides novel structural prototypes and simple ternary compounds as well as a large portfolio of materials to search from for optimal properties. For a subset of 258 compounds we explore vibrational, electronic, magnetic, and topological properties, identifying 56 ferromagnetic and antiferromagnetic systems, including half-metals and half-semiconductors. This archive entry contains the database of 2D materials (structural parameters, band structures, binding energies, phonons for the subset of the 258 easily exfoliable materials with less than 6 atoms, structures and binding energies for the remaining 1567 materials) together with the provenance of all data and calculations as stored by AiiDA.

Functionalizing Silicon Nanowires (SiNWs) through covalent attachment of organic molecules offers diverse advantages, including surface passivation, introduction of new functionalities, and enhanced material performance in applications like electronic devices and biosensors. Given the wide range of available functional molecules, systematic large-scale screening is crucial. Therefore, we developed an automated computational workflow using Python scripts in conjunction with the AiiDa framework to explore structural configurations of functional molecules adsorbed onto silicon surfaces. This workflow generates multiple adhesion configurations corresponding to different binding orientations using surface and functional molecule structures as inputs.
This dataset contains data related to the structural optimization of molecules with single, double, and triple carbon-carbon bonds attached to the nanowire surface in various adhesion configurations. We describe the chemisorption on SiNWs using the slab models for the Si facets since our reference are samples with diameters of SiNWs around 50 nm, while the quantum confinement effects are important for diameters below 10 nm. For each configuration, structural characterization was conducted by calculating quantities including the bond distance between the two carbons closest to the surface and their respective bond angle relative to the z-axis, the carbon-silicon bond distance and its respective bond angle relative to the z-axis, along with the molecule's rotation angle in the xy plane. The values obtained are summarized in the main folder. The version v1 of dataset contains data related to the Si(111) surface and alkanes, alkenes, and alkynes with lengths from C2 to C10. The version v2 extends the dataset to moieties from C12 to C18. This version (v3) extends the dataset with new configurations for moieties from C2 to C18. The dataset will be extended to characterize the Si(110) surface of the nanowire. For each system the most stable configuration will be identified, and the analysis of the electronic properties will be conducted.

Transition-metal-based kagome materials at van Hove filling are a rich frontier for the investigation of novel topological electronic states and correlated phenomena. To date, in the idealized two-dimensional kagome lattice, topologically Dirac surface states (TDSSs) have not been unambiguously observed, and the manipulation of TDSSs and van Hove singularities (VHSs) remains largely unexplored. Here, we reveal TDSSs originating from a Z₂ bulk topology and identify multiple VHSs near the Fermi level (E_F) in magnetic kagome material GdV₆Sn₆. Using in situ surface potassium deposition, we successfully realize manipulation of the TDSSs and VHSs. The Dirac point of the TDSSs can be tuned from above to below E_F, which reverses the chirality of the spin texture at the Fermi surface. These results establish GdV₆Sn₆ as a fascinating platform for studying the nontrivial topology, magnetism, and correlation effects native to kagome lattices. They also suggest potential application of spintronic devices based on kagome materials.

Two-dimensional (2D) layered materials offer a materials platform with potential applications from energy to information processing devices. Although some single- and few-layer forms of materials such as graphene and transition metal dichalcogenides have been realized and thoroughly studied, the space of arbitrarily layered assemblies is still mostly unexplored. The main goal of this work is to demonstrate precise control of layered materials' electronic properties through careful choice of the constituent layers, their stacking, and relative orientation. Physics-based and AI-driven approaches for the automated planning, execution, and analysis of electronic structure calculations are applied to layered assemblies based on prototype one-dimensional (1D) materials and realistic 2D materials. We find it is possible to routinely generate moiré band structures in 1D with desired electronic characteristics such as a band gap of any value within a large range, even with few layers and materials (here, four and six, respectively). We argue that this tunability extends to 2D materials by showing the essential physical ingredients are already evident in calculations of two-layer MoS2 and multi-layer graphene moiré assemblies.

The design rules for materials are clear for applications with a single objective. For most applications, however, there are often multiple, sometimes competing objectives where there is no single best material, and the design rules change to finding the set of Pareto optimal materials. In this work, we introduce an active learning algorithm that directly uses the Pareto dominance relation to compute the set of Pareto optimal materials with desirable accuracy. We apply our algorithm to de novo polymer design with a prohibitively large search space. Using molecular simulations, we compute key descriptors for dispersant applications and reduce the number of materials that need to be evaluated to reconstruct the Pareto front with a desired confidence by over 98% compared to random search. This work showcases how simulation and machine learning techniques can be coupled to discover materials within a design space that would be intractable using conventional screening approaches.

Graph neural networks have enjoyed great success in the prediction of material properties for both molecules and crystals. These networks typically use the atomic positions (usually expanded in a Gaussian basis) and the atomic species as input. Unfortunately, this information is in general not available when predicting new materials, for which the precise geometrical information is unknown. In this work, we circumvent this problem by predicting the thermodynamic stability of crystal structures without using the knowledge of the precise bond distances. We replace this information with embeddings of graph distances, allowing our networks to be used directly in high-throughput studies based on both composition and crystal structure prototype. Using these embeddings, we combine the newest developments in graph neural networks and apply them to the prediction of the distances to the convex hull. To train these networks, we curate a dataset of over 2 million density-functional calculations of crystals with consistent calculation parameters from various sources. The new dataset allows for the creation of a high quality convex hull and a large scale transfer learning approach. We apply the resulting model to the high-throughput search of 15 million tetragonal perovskites of composition ABCD2. As a result, we identify several thousand potentially stable compounds and demonstrate that transfer learning from the newly curated dataset reduces the required training data by 50%.

Please consider citing the following paper:

@misc{carbon2020data,
 doi = {10.24435/MATERIALSCLOUD:2020.0026/V1},
 url = {https://archive.materialscloud.org/record/2020.0026/v1},
 author = {Pickard, Chris J.},
 keywords = {DFT, ab initio random structure searching, carbon},
 language = {en},
 title = {AIRSS data for carbon at 10GPa and the C+N+H+O system at 1GPa},
 publisher = {Materials Cloud},
 year = {2020},
 copyright = {info:eu-repo/semantics/openAccess}
}

Semiconducting alloys, in particular SiGe, have been employed for several decades as high- temperature thermoelectric materials. Devising strategies to reduce their thermal conductivity may provide a substantial improvement in their thermoelectric performance also at lower temper- atures. We have carried out an ab initio investigation of the thermal conductivity of SiGe alloys with random and spatially correlated compositional disorder employing the Quasi-Harmonic Green- Kubo (QHGK) theory with force constants computed from density functional theory. Leveraging QHGK and the hydrodynamic extrapolation to achieve size convergence, we obtained a detailed understanding of lattice heat conduction in SiGe and demonstrated that colored disorder suppresses thermal transport across the acoustic vibrational spectrum, leading to up to a 4-fold enhancement in the intrinsic thermoelectric figure of merit.

This record contains input and analysis scripts to reproduce the findings of this article.