A Set of Moment Tensor Potentials for Zirconium with Increasing Complexity.

Machine learning force fields (MLFFs) are an increasingly popular choice for atomistic simulations due to their high fidelity and improvable nature. Here we propose a hybrid small-cell approach that combines attributes of both offline and active learning to systematically expand a quantum-mechanical (QM) database while constructing MLFFs with increasing model complexity. Our MLFFs employ the moment tensor potential formalism. During this process, we quantitatively assessed the structural properties, elastic properties, dimer potential energies, melting temperatures, phase stability, point defect formation energies, point defect migration energies, free surface energies, and generalized stacking fault (GSF) energies of Zr as predicted by our MLFFs. Unsurprisingly, the model complexity has a positive correlation with prediction accuracy. We also find that the MLFFs were able to predict the properties of out-of-sample configurations without directly including these specific configurations in the training dataset. Additionally, we generated 100 MLFFs of high complexity (1513 parameters each) that reached different local optima during training. Their predictions cluster around the benchmark DFT values, but subtle physical features such as the location of local minima on the GSF energy surface are washed out by statistical noise.

Tensor-Reduced Atomic Density Representations.

Density-based representations of atomic environments that are invariant under Euclidean symmetries have become a widely used tool in the machine learning of interatomic potentials, broader data-driven atomistic modeling, and the visualization and analysis of material datasets. The standard mechanism used to incorporate chemical element information is to create separate densities for each element and form tensor products between them. This leads to a steep scaling in the size of the representation as the number of elements increases. Graph neural networks, which do not explicitly use density representations, escape this scaling by mapping the chemical element information into a fixed dimensional space in a learnable way. By exploiting symmetry, we recast this approach as tensor factorization of the standard neighbour-density-based descriptors and, using a new notation, identify connections to existing compression algorithms. In doing so, we form compact tensor-reduced representation of the local atomic environment whose size does not depend on the number of chemical elements, is systematically convergable, and therefore remains applicable to a wide range of data analysis and regression tasks.

First principles thermodynamical modeling of the binodal and spinodal curves in lead chalcogenides.

High-throughput ab initio calculations, cluster expansion techniques, and thermodynamic modeling have been synergistically combined to characterize the binodal and the spinodal decompositions features in the pseudo-binary lead chalcogenides PbSe-PbTe, PbS-PbTe, and PbS-PbSe. While our results agree with the available experimental data, our consolute temperatures substantially improve with respect to previous computational modeling. The computed phase diagrams corroborate that in ad hoc synthesis conditions the formation of nanostructure may occur justifying the low thermal conductivities in these alloys. The presented approach, making a rational use of online quantum repositories, can be extended to study thermodynamical and kinetic properties of materials of technological interest.

The high-throughput highway to computational materials design.

High-throughput computational materials design is an emerging area of materials science. By combining advanced thermodynamic and electronic-structure methods with intelligent data mining and database construction, and exploiting the power of current supercomputer architectures, scientists generate, manage and analyse enormous data repositories for the discovery of novel materials. In this Review we provide a current snapshot of this rapidly evolving field, and highlight the challenges and opportunities that lie ahead.

Ordered structures in rhenium binary alloys from first-principles calculations.

Rhenium is an important alloying agent in catalytic materials and superalloys, but the experimental and computational data on its binary alloys are sparse. Only 6 out of 28 Re transition-metal systems are reported as compound-forming. Fifteen are reported as phase-separating, and seven have high-temperature disordered σ or χ phases. Comprehensive high-throughput first-principles calculations predict stable ordered structures in 20 of those 28 systems. In the known compound-forming systems, they reproduce all the known compounds and predict a few unreported ones. These results indicate the need for an extensive revision of our current understanding of Re alloys through a combination of theoretical predictions and experimental validations. The following systems are investigated: AgRe(★), AuRe(★), CdRe(★), CoRe, CrRe(★), CuRe(★), FeRe, HfRe, HgRe(★), IrRe, MnRe, MoRe, NbRe, NiRe, OsRe, PdRe, PtRe, ReRh, ReRu, ReSc, ReTa, ReTc, ReTi, ReV, ReW(★), ReY, ReZn(★), and ReZr ((★) = systems in which the ab initio method predicts that no compounds are stable).

Predictions of the Pt(8)Ti phase in unexpected systems.

The binary A(8)B phase (prototype Pt(8)Ti) has been experimentally observed in 11 systems. A high-throughput search over all the binary transition intermetallics, however, reveals 59 occurrences of the A(8)B phase: Au(8)Zn(dagger), Cd(8)Sc(dagger), Cu(8)Ni(dagger), Cu(8)Zn(dagger), Hg(8)La, Ir(8)Os(dagger), Ir(8)Re, Ir(8)Ru(dagger), Ir(8)Tc, Ir(8)W(dagger), Nb(8)Os(dagger), Nb(8)Rh(dagger), Nb(8)Ru(dagger), Nb(8)Ta(dagger), Ni(8)Fe, Ni(8)Mo(dagger)*, Ni(8)Nb(dagger)*, Ni(8)Ta*, Ni(8)V*, Ni(8)W, Pd(8)Al(dagger), Pd(8)Fe, Pd(8)Hf, Pd(8)Mn, Pd(8)Mo*, Pd(8)Nb, Pd(8)Sc, Pd(8)Ta, Pd(8)Ti, Pd(8)V*, Pd(8)W*, Pd(8)Zn, Pd(8)Zr, Pt(8)Al(dagger), Pt(8)Cr*, Pt(8)Hf, Pt(8)Mn, Pt(8)Mo, Pt(8)Nb, Pt(8)Rh(dagger), Pt(8)Sc, Pt(8)Ta, Pt(8)Ti*, Pt(8)V*, Pt(8)W, Pt(8)Zr*, Rh(8)Mo, Rh(8)W, Ta(8)Pd, Ta(8)Pt, Ta(8)Rh, V(8)Cr(dagger), V(8)Fe(dagger), V(8)Ir(dagger), V(8)Ni(dagger), V(8)Pd, V(8)Pt, V(8)Rh, and V(8)Ru(dagger) ((dagger) = metastable, * = experimentally observed). This is surprising for the wealth of new occurrences that are predicted, especially in well-characterized systems (e.g., Cu-Zn). By verifying all experimental results while offering additional predictions, our study serves as a striking demonstration of the power of the high-throughput approach. The practicality of the method is demonstrated in the Rh-W system. A cluster-expansion-based Monte Carlo model reveals a relatively high order-disorder transition temperature.

Uncovering compounds by synergy of cluster expansion and high-throughput methods.

Predicting from first-principles calculations whether mixed metallic elements phase-separate or form ordered structures is a major challenge of current materials research. It can be partially addressed in cases where experiments suggest the underlying lattice is conserved, using cluster expansion (CE) and a variety of exhaustive evaluation or genetic search algorithms. Evolutionary algorithms have been recently introduced to search for stable off-lattice structures at fixed mixture compositions. The general off-lattice problem is still unsolved. We present an integrated approach of CE and high-throughput ab initio calculations (HT) applicable to the full range of compositions in binary systems where the constituent elements or the intermediate ordered structures have different lattice types. The HT method replaces the search algorithms by direct calculation of a moderate number of naturally occurring prototypes representing all crystal systems and guides CE calculations of derivative structures. This synergy achieves the precision of the CE and the guiding strengths of the HT. Its application to poorly characterized binary Hf systems, believed to be phase-separating, defines three classes of alloys where CE and HT complement each other to uncover new ordered structures.

The new face of rhodium alloys: revealing ordered structures from first principles.

The experimental and computational data on rhodium binary alloys is sparse despite its importance in numerous applications, especially as an alloying agent in catalytic materials. Half of the Rh-transition metal systems (14 out of 28) are reported to be phase separating or are lacking experimental data. Comprehensive high-throughput first-principles calculations predict stable ordered structures in 9 of those 14 binary systems. They also predict a few unreported compounds in the known compound-forming systems. These results indicate the need for an extensive revision of our current understanding of Rh alloys through a combination of theoretical predictions and experimental validations.

Where are nature's missing structures?

Our society's environmental and economic progress depends on the development of high-performance materials such as lightweight alloys, high-energy-density battery materials, recyclable motor vehicle and building components, and energy-efficient lighting. Meeting these needs requires us to understand the central role of crystal structure in a material's properties. Despite more than 50 years of progress in first-principles calculations, it is still impossible in most materials to infer ground-state properties purely from a knowledge of their atomic components--a situation described as 'scandalous' in the well-known essay by Maddox. Many methods attempt to predict crystal structures and compound stability, but here I take a different tack--to infer the existence of structures on the basis of combinatorics and geometric simplicity. The method identifies 'least random' structures, for which the energy is an extremum (maximum or minimum). Although the key to the generic nature of the approach is energy minimization, the extrema are found in a chemistry-independent way.

Evolutionary approach for determining first-principles hamiltonians.

Modern condensed-matter theory from first principles is highly successful when applied to materials of given structure-type or restricted unit-cell size. But this approach is limited where large cells or searches over millions of structure types become necessary. To treat these with first-principles accuracy, one 'coarse-grains' the many-particle Schrodinger equation into 'model hamiltonians' whose variables are configurational order parameters (atomic positions, spin and so on), connected by a few 'interaction parameters' obtained from a microscopic theory. But to construct a truly quantitative model hamiltonian, one must know just which types of interaction parameters to use, from possibly 10(6)-10(8) alternative selections. Here we show how genetic algorithms, mimicking biological evolution ('survival of the fittest'), can be used to distil reliable model hamiltonian parameters from a database of first-principles calculations. We demonstrate this for a classic dilemma in solid-state physics, structural inorganic chemistry and metallurgy: how to predict the stable crystal structure of a compound given only its composition. The selection of leading parameters based on a genetic algorithm is general and easily applied to construct any other type of complex model hamiltonian from direct quantum-mechanical results.