Monday 8:30 AM

February 27, 2017

Room: 11A

Location: San Diego Convention Ctr

The high-throughput approach for computational materials science has led to the generation of huge databases of DFT-based calculations. Direct mining of this data has led to the discovery of new materials and is of considerable utility. But the real potential for these data to impact American competitiveness, as envisioned in the MGI, is in "interpolation"---using the data to discover materials not present in the databases. I will discuss an approach for materials interpolation that combines cluster expansion, the new SOAP (smooth overlap of atomic positions) representation, and machine learning.

We propose an efficient approach for first-principles crystal structure prediction. The new method explores and finds crystal structures by tiling together elementary tetrahedra that are energetically favorable and geometrically matching each other. Our approach has three distinguishing features: a favorable building unit, an efficient calculation of local energy, and a stochastic Monte Carlo simulation of crystal growth. (1) A deformable tetrahedron is naturally an ideal tiling unit. It eliminates common problems associated with geometric mismatches in commonly used genetic algorithms. (2) Local energy density of tetrahedron is determined by recently-developed density functional theory energy-density method. This strategy, combined with interpolation, provides an energy function for any tetrahedron in a highly efficiently manner, waiving the requirement of a large-scale fitting exercise typically associated with a construction of effective interatomic potential energy model. (3) Stochastic Monte Carlo formalism utilizes the energy function to simulate tetrahedron tiling and crystal growth.

The local arrangement of atoms in solid state matter is arguably the most important signature of the materials that influences a broad range of mechanical and functional properties. However, description and quantification of the local structure is non-trivial and the automated analysis of such descriptors for large sets of data is lacking. In this talk we present a point-pattern matching algorithm that can efficiently detect the instances of a template structure in a given set of atom coordinates. In addition, we introduce a metric that can uniquely quantify the amount of distortion in a set of atom coordinates with respect to an underlying template. We will demonstrate how the combination of the proposed algorithm and the distortion metric can be utilized in an unsupervised machine learning framework for identifying distinct local atomic environments (e.g. those corresponding to distinct phases) in atomistic simulations of heterogeneous bulk materials.

Molecular systems such as metal-organic–frameworks and other coordination polymers that respond to external stimuli could enable a whole new class of materials with remarkable properties, such as, materials with externally tunable stiffness, variable porosity, or tunable catalytic properties. Here we present a new approach for discovering candidate molecules for MOFs with tailored mechanical and kinematic properties. In combination with our approach for designing molecules we present a path based scheme for fingerprinting the structure in a way that enables us to compare structural similarity of molecules, identify kinematically active components, and from them learn new structure property relations. We will show how this approach can be used to design pressure switching MOFs, that is, MOFs that exhibit a reversible structural collapse whenever a stress threshold is exceeded. We will also demonstrate how this approach might be expanded to design MOFs with photoisomerizing moieties to realize photoactuating behavior.

Density-functional theory (DFT) is one of the most reliable simulation methodologies used in materials science. Experimental and computational scientists alike use it to interpret their results and to fit larger scale models. While DFT presents an in-principle exact theory, various approximations are required to perform practical simulations. These approximations can be classified as: (i) controlled approximations, whose errors can be made arbitrarily small at the expense of increased computational cost, and (ii) uncontrolled approximations, whose errors are unknown exactly. To this day, a systematic evaluation of the uncertainties related to either kind of approximation is still lacking and it is the scope of this work. We computed quantities like formation energies, lattice constants and elastic properties for single elements in several phases, using various DFT codes. We then investigated correlations in the precision of the different properties obtained as a function of various convergence parameters (basis set and k-points).

Molecular dynamics (MD) is a popular approach for understanding fundamental materials' behavior since it can be extended to larger length and time scales than first principles methods. However, MD simulations are adversely limited by the the interatomic potentials' accuracy. Parameterization and optimization of complex potentials can become a severe bottleneck. Furthermore, optimization can become prohibitively expensive for complex potentials that contain hundreds of fitting parameters, e.g. ReaxFF. We have developed a systematic approach that relies on computational experimental designs to overcome these optimization barriers. To counter the expense associated of optimizing large domain spaces, we perform a sensitivity analysis to efficiently identify ``essential'' parameters. With this reduced domain, experimental designs (using LHS) are used to develop surrogate models. These are then employed in multiobjective optimization schemes, which yield optimized domain sets. We validate the surrogate models to assure high fidelity from the resulting optimized parameterizations.

Atomistic simulations using empirical interatomic potentials play a key role in realistic scientific and industrial applications. The Open Knowledgebase of Interatomic Models project (https://OpenKIM.org) includes an automated user-extendable framework for testing the predictions of potentials for a host of material properties. Visualization tools have been developed to compare potential predictions to help select the most appropriate one for a given application. The potentials in OpenKIM conform to the KIM application programming interface, which means that they can be seamlessly used with several major molecular simulation codes that support the KIM standard. This talk will describe the OpenKIM project and how the testing framework can assist materials researchers.

Advances in accelerating materials discovery in line with the Materials Genome Initiative (MGI) have been characterized by a glut of data arising from high-throughput calculations; which do not always translate into knowledge. Recently, it has been shown by Lookman et al. that a targeted approach using adaptive design strategies is a more efficient and 'intelligent' way to look at the problem of materials design. Inspired by this, an on-the-fly , real-time implementation of the Effective Global Optimization (EGO) algorithm to answer real-world materials questions is presented. Two cases are shown using a test set of 432 MAX phases: i) single objective EGO in which we isolate the material with the maximum Bulk Modulus and ii) multi-objective parallel EGO wherein we isolate the MAX phase with the maximum 'ductility index', i.e. the ratio of Bulk Modulus to Shear Modulus.

When it comes to the exploration of multi-dimensional phase stability spaces, there is a large class of problems that need to be defined in terms of the satisfaction of a set of non-linear constraints. We define an "inverse phase stability problem" as one in which one needs to define the thermodynamic conditions that satisfy specific phase constitution states, as opposed to the general, forward determination of the equilibrium state of a system through Gibbs energy minimization subject to constraints. As such, the "inverse phase stability problem" is essentially a set-based design exercise, that can in turn be formulated as a constraint satisfaction problem. In this work we present a novel formulation for the exploration of multi-component systems under the constraint satisfaction framework and show how this approach can be translated into a materials design task.